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,
37 Remove 2.4 kernel code from lustre/mgc/mgc_request.c.
38 Remove 2.4 kernel check from lustre/mdt/mdt_lproc.c.
39 Remove 2.4 kernel code from lustre/mdc/mdc_request.c.
40 Remove 2.4 kernel check from lustre/lov/lproc_lov.c.
41 Remove 2.4 kernel check from lustre/lmv/lproc_lmv.c.
42 Remove lock_24kernel from lustre/lvfs/fsfilt_ext3.c.
43 Remove 2.4 kernel code from include/liblustre.h,
44 include/lustre_disk.h, include/lustre_quota.h,
45 include/linux/lprocfs_status.h, include/linux/lvfs.h,
46 include/linux/lustre_lite.h, include/linux/lustre_net.h,
47 include/linux/lustre_user.h, include/linux/lvfs_linux.h,
48 include/linux/obd_class.h, include/linux/lustre_compat25.h.
50 Severity : enhancement
52 Description: Build SLES10 patchless client fails
53 Details : The configure was broken by run ./configure with
54 --with-linux-obj=.... argument for patchless client. When the
55 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
56 can't search header adequately. Use absolute path such as
57 -I($LINUX)/include instead.
61 Description: interrupt oig_wait produce painc on resend.
62 Details : brw_redo_request can be used for resend requests from ptlrpcd and
63 private set, and this produce situation when rq_ptlrpcd_data not
64 copyed to new allocated request and triggered LBUG on assert
65 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
66 ptlrpcd set if request is changed and can be safety changed to use
71 Description: LASSERT_{REQ,REP}SWAB macros are buggy
72 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
73 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
74 these macros and replace them with their difinitions instead.
78 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
79 Details : Update which_patch & target file for SLES10 latest kernel.
82 Description: Remove client patches from SLES 10 kernel.
83 Details : This causes SLES 10 clients to behave as patchless clients
84 even on a Lustre-patched (server) kernel.
86 Severity : enhancement
88 Description: self-adjustable client's lru lists
89 Details : use adaptive algorithm for managing client cached locks lru
90 lists according to current server load, other client's work
91 pattern, memory activities, etc. Both, server and client
92 side namespaces provide number of proc tunables for controlling
95 Severity : enhancement
97 Description: light-weight GSS support
98 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
99 and reduce performance overhead.
101 Severity : enhancement
103 Description: Linux keyring support
104 Details : Support using service of Linux keyring for Lustre GSS internal
105 context refresh/cache mechanism.
109 Description: Fix errors in lfs documentation
110 Details : Fixes man pages
114 Description: don't use GFP_* in generic Lustre code.
115 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
119 Description: obdclass is limited by single OBD_ALLOC(idarray)
120 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
124 Description: warnings with build patchless client with vanila 2.6.19 and up
125 Details : change old ctl_table style and replace ctl_table/ctl_table_header
126 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
130 Frequency : Only for SLES
131 Description: sanity_quota fail test_1
132 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
133 file, which makes TSTID[2] inunique.
137 Frequency : Only for SLES9
138 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
139 Details : Update target/ChangeLog/which_patch .
143 Frequency : Only for test_10 in sanity-quota.sh
144 Description: a bug in quota test code
145 Details : checking a test flag in a wrong place
149 Frequency : Only for test_13 in sanity-quota.sh
150 Description: a bug in quota test script
151 Details : shouldn't check overfull allocated quota
155 Description: jbd statistics
156 Details : Port older jbd statistics patch for sles10
158 Severity : enhancement
160 Description: Add printing inode info into message about error in writepage.
165 Description: reply_lock_interpret crash due to race with it and lock cancel.
166 Details : Do not replay locks that are being cancelled. Do not reference
167 locks by their address during replay, just by their handle.
171 Frequency : When flocks are used.
172 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
173 Details : Flock locks might destroy just granted lock if it could be merged
174 with another existing flock, this is done in completion handler,
175 so teach ldlm_cli_enquque_fini that this is a valid case for
181 Description: Oops in read and write path when failing to allocate lock.
182 Details : Check if lock allocation failed and return error back.
186 Description: lstripe command fails for valid OST index
187 Details : The stripe offset is compared to lov->desc.ld_tgt_count
188 instead of lov->desc.ld_active_tgt_count.
192 Description: sanity.sh failed test 103
193 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
198 Description: df doesn't work properly if diskfs blocksize != 4K
199 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
203 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
204 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
208 Description: obd_health_check_timeout too short
209 Details : set obd_health_check_timeout as 1.5x of obd_timeout
213 Description: llapi_file_create() does not allow some changes
214 Details : add llapi_file_open() that allows specifying the mode and
215 open flags, and also returns an open file handle.
219 Description: merge and cleanup kernel patches.
220 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
224 Description: Add journal checksum support.(Kernel part)
225 Details : The journal checksum feature adds two new flags i.e
226 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
227 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
228 indicates that the commit block contains the checksum for
229 the blocks described by the descriptor blocks. Now commit
230 record can be sent to disk without waiting for descriptor
231 blocks to be written to disk. This behavior is controlled
232 using JBD2_FEATURE_ASYNC_COMMIT flag.
236 Description: OSS needs mutliple precreate threads
237 Details : Add ability to start more than one create thread per OSS.
241 Description: Sanity.sh test_65k failed: setstripe should have succeeded
242 Details : execute lfs setstripe on client
246 Description: mds_obd_create error creating tmp object
247 Details : When the user sets quota on root, llog will be affected and can't
248 create files and write files.
251 Frequency : Always on ia64 patchless client, and possibly others.
253 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
254 Details : This allows the patchless client to be loaded on architectures
259 Description: RedHat Update kernel for RHEL5
260 Details : Add the kernel config file for RHEL5.
264 Description: RedHat Update kernel for RHEL5
265 Details : Modify the kernel config file more closer RHEL5.
269 Description: Build failure against Centos5 (RHEL5)
270 Details : Define PAGE_SIZE when it isn't present.
274 Description: test_7 on sanity-quota.sh can't be used on separate machines
275 Details : add facet to handle it
279 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
280 Details : with NFS, the anon dentry's parent was set to itself in
281 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
282 corresponding dentry other than use rec->ur_name.
284 Severity : enhancement
286 Description: lfs setstripe enhancement
287 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
291 Description: testing performance impact of enabling checksumming
292 Details : enable checksum by default, allow --disable-checksum
293 configure option and "-o nochecksum" mount option
297 Description: lustre support for RHEL5
298 Details : Add support for RHEL5.
302 Description: Client eviction due to failover config
303 Details : after a connection loss, the lustre client should attempt to
304 reconnect to the last active server first before trying the
305 other potential connections.
309 Description: when mds and osts use different quota unit(32bit and 64bit),
310 quota will be released repeatly.
311 Details : void sending multiple quota reqs to mds, which will keep the
312 status between the reqs.
316 Description: osts not allocated evenly to files
317 Details : change the condition to increase offset_idx
320 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
322 Description: Data corruption for OSTs that are formatted larger than 2TB
324 Details : When generating the bio request for lustre file writes the
325 sector number would overflow a temporary variable before being
326 used for the IO. The data reads correctly from Lustre (which
327 will overflow in a similar manner) but other file data or
328 filesystem metadata may be corrupted in some cases.
332 Description: Tune the kernel for good SCSI performance.
333 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
334 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
339 Description: rewrite ext2-derived code in obdclass/uuid.c
340 Details : rewrite inherited code (uuid parsing code from ext2 utils)
341 from scratch preserving functionality.
345 Description: "lfs find -obd UUID" prints directories
346 Details : "lfs find -obd UUID" will return all directory names instead
347 of just file names. It is incorrect because the directories
348 do not reside on the OSTs.
352 Description: use i_size_read and i_size_write in 2.6 port
353 Details : replace inode->i_size access with i_size_read/write()
356 Frequency : when removing large files
358 Description: scheduling issue during removal of large Lustre files
359 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
360 It causes scheduling issues when removing large files (17TB in the
364 Frequency : only with liblustre clients on XT3
366 Description: evictions taking too long
367 Details : allow llrd to evict clients directly on OSTs
370 Frequency : only on ppc
372 Description: /proc/fs/lustre/devices broken on ppc
373 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
374 We should make sure the type of 'index' is loff_t and then cast
375 explicitly as needed below. Do not assign an explicitly cast
379 Frequency : only for rhel5
381 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
382 Details : Modify the target file & which_kernel.
384 Severity : enhancement
386 Description: omit set fsid for export NFS
387 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
388 as FS_REQUIRES_DEV for avoid problems with generate fsid.
392 Frequency : Always, for kernels after 2.6.16
393 Description: Fix warning idr_remove called for id=.. which is not
395 Details : Last kernels save old s_dev before kill super and not allow
396 to restore from callback - restore it before call kill_anon_super.
400 Description: buffer overruns could theoretically occur
401 Details : llapi_semantic_traverse() modifies the "path" argument by
402 appending values to the end of the origin string, and a
403 overrun may occur. Adding buffer overrun check in liblustreapi.
407 Description: Fix error on 'ls .' at the top of the Lustre mount.
408 Details : Don't revalidate dentry if it is a root dentry.
412 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
413 Details : Modify vm-tunables-rhel4.patch.
417 Description: Kernel config for 2.6.18-vanilla.
418 Details : Modify targets/2.6-vanilla.target.in.
419 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
420 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
421 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
422 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
426 Description: improve handling recoverable errors
427 Details : if request processig with error which can be recoverable on server
428 request should be resend, otherwise page released from cache and
433 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
434 Details : Modify target file & which_patch.
435 A flaw was found in the IA32 system call emulation provided
436 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
437 value could be stored in the %RAX register, which could trigger an
438 out-of-bounds system call table access. An untrusted local user
439 could exploit this flaw to run code in the kernel
440 (ie a root privilege escalation). (CVE-2007-4573).
444 Description: O_DIRECT bypasses client statistics.
445 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
446 read_ahead_stats were updated. copied the stats section from
447 osc_send_oap_rpc() into async_internal().
451 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
455 Description: change order of libsysio includes
456 Details : '#include sysio.h' should always come before '#include xtio.h'
460 Description: handle "serious error: objid * already exists" more gracefully
461 Details : If LAST_ID value on disk is smaller than the objects existing in
462 the O/0/d* directories, it indicates disk corruption and causes an
463 LBUG(). If the object is 0-length, then we should use the existing
464 object. This will help to avoid a full fsck in most cases.
466 Severity : enhancement
468 Description: adapt the lustre_config script to support the upgrade case
469 Details : Add "-u" option for lustre_config script to support upgrading 1.4
470 server targets to 1.6 in parallel.
474 Description: To avoid grant space > avaible space when the disk is almost
475 full. Without this patch you might see the error "grant XXXX >
476 available" or some LBUG about grant, when the disk is almost
478 Details : In filter_check_grant, for non_grant cache write, we should
479 check the left space by if (*left > ungranted + bytes), instead
480 of (*left > ungranted), because only we are sure the left space
481 is enough for another "bytes", then the ungrant space should be
482 increase. In client, we should update cl_avail_grant only there
483 is OBD_MD_FLGRANT in the reply.
487 Description: Update RHEL 4 kernel to fix local root privilege escalation.
488 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
489 described in CVE-2007-4573. This problem could allow untrusted
490 local users to gain root access.
494 Frequency : when using O_DIRECT and quotas
496 Description: Incorrect file ownership on O_DIRECT output files
497 Details : block usage reported by 'lfs quota' does not take into account
498 files that have been written with O_DIRECT.
500 --------------------------------------------------------------------------------
502 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
504 * Support for kernels:
505 2.6.5-7.283 (SLES 9),
506 2.6.9-55.EL (RHEL 4),
507 2.6.16.46-0.14 (SLES 10),
508 2.6.18.8 vanilla (kernel.org)
509 * Client support for unpatched kernels:
510 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
511 2.6.16 - 2.6.22 vanilla (kernel.org)
512 * Due to recently discovered recovery problems, we do not recommend
513 using patchless RHEL 4 clients with this or any earlier release.
514 * Recommended e2fsprogs version: 1.39.cfs8
515 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
516 * Starting with this release, the ldiskfs backing filesystem required
517 by Lustre is now in its own package, lustre-ldiskfs. This package
518 should be installed. It is versioned separately from Lustre and
519 may be released separately in future.
523 Description: block reactivating mgc import until all deactivates complete
524 Details : Fix race when failing back MDT/MGS to itself (testing)
527 Frequency : at statup only
529 Description: mds_lov_synchronize race leads to various problems
530 Details : simultaneous MDT->OST connections at startup can cause the
531 sync to abort, leaving the OSC in a bad state.
533 Severity : enhancement
535 Description: add optional extra BUILD_VERSION info
536 Details : add a new environment variable (namely LUSTRE_VERS) which allows
537 to override the lustre version.
540 Frequency : 2.6.18 servers only
542 Description: ll_kern_mount() doesn't release the module reference
543 Details : The ldiskfs module reference count never drops down to 0
544 because ll_kern_mount() doesn't release the module reference.
549 Description: server LBUG when using old ost_num_threads parameter
550 Details : Accept the old ost_num_threads parameter but warn that it
551 is deprecated, and fix an off-by-one error that caused an LBUG.
556 Description: Transient SCSI error results in persistent IO issue
557 Details : iobuf->dr_error is not reinitialized to 0 between two
561 Frequency : sometimes when underlying device returns I/O errors
563 Description: OSTs not going read-only during write failures
564 Details : OSTs are not remounted read-only when the journal commit threads
565 get I/O errors because fsfilt_ext3 calls journal_start/stop()
566 instead of the ext3 wrappers.
570 Description: poor connect scaling with increasing client count
571 Details : Don't run filter_grant_sanity_check for more than 100 exports
572 to improve scaling for large numbers of clients.
575 Frequency : SLES10 only
577 Description: sanity-quota.sh quotacheck failed: rc = -22
578 Details : Quotas cannot be enabled on SLES10.
581 Frequency : liblustre clients only
583 Description: getdirentries does not give error when run on compute nodes
584 Details : getdirentries does not fail when the size specified as an argument
585 is too small to contain at least one entry
587 Severity : enhancement
589 Description: Add LNET router traceability for debug purposes
590 Details : If a checksum failure occurs with a router as part of the
591 IO path, the NID of the last router that forwarded the bulk data
592 is printed so it can be identified.
597 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
598 Details : Due to a race condition, liblustre clients were occasionally
601 Severity : enhancement
603 Description: lfs setstripe use optional parameters instead of postional
606 Severity : enhancement
608 Description: Nanosecond timestamp support for ldiskfs
609 Details : The on-disk ldiskfs filesystem has added support for nanosecond
610 resolution timestamps. There is not yet support for this at
611 the Lustre filesystem level.
614 Frequency : during server recovery
616 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
617 Details : request with rq_no_resend flag not awake l_wait_event if they get a
621 Frequency : nfs export on patchless client
623 Description: connectathon hang when test nfs export over patchless client
624 Details : Disconnected dentry cannot be found with lookup, so we do not need
625 to unhash it or make it invalid
628 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
629 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
630 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
635 Description: open req refcounting wrong on reconnect
636 Details : If reconnect happened between getting open reply from server and
637 call to mdc_set_replay_data in ll_file_open, we will schedule
638 replay for unreferenced request that we are about to free.
639 Subsequent close will crash in variety of ways.
640 Check that request is still eligible for replay in
641 mdc_set_replay_data().
646 Description: disable writes to filesystem when reading health_check file
647 Details : the default for reading the health_check proc file has changed
648 to NOT do a journal transaction and write to disk, because this
649 can cause reads of the /proc file to hang and block HA state
650 checking on a healthy but otherwise heavily loaded system. It
651 is possible to return to the previous behaviour during configure
652 with --enable-health-write.
654 Severity : enhancement
656 Description: 64-bit inode version
657 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
658 to the inode. This will be required for version-based recovery.
663 Description: MDS fails to start if a duplicate client export is detected
664 Details : in some rare cases it was possible for a client to connect to
665 an MDS multiple times. Upon recovery the MDS would detect this
666 and fail during startup. Handle this more gracefully.
668 Severity : enhancement
670 Description: Add -o localflock option to simulate old noflock
672 Details : This will achieve local-only flock/fcntl locks
678 Description: log_commit_thread vs filter_destroy race leads to crash
679 Details : Take import reference before releasing llog record semaphore
684 Description: Wrong request locking in request set processing
685 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
686 imp_delayed_list, in this place should be used imp_lock.
689 Frequency : when reconnection
691 Description: Grant Leak when osc reconnect to OST
692 Details : When osc reconnect ost, OST(filter) should check whether it
693 should grant more space to client by comparing fed_grant and
694 cl_avail_grant, and return the granted space to client instead
695 of "new granted" space, because client will call osc_init_grant
696 to update the client grant space info.
699 Frequency : when client reconnect to OST
701 Description: Grant Leak when osc do resend and replay bulk write
702 Details : When osc reconnect to OST, OST(filter)should clear grant info of
703 bulk write request, because the grant info while be sync between
704 OSC and OST when reconnect, and we should ignore the grant info
705 these of resend/replay write req.
710 Description: Grant space more than avaiable left space sometimes.
711 Details : When then OST is about to be full, if two bulk writing from
712 different clients came to OST. Accord the avaliable space of the
713 OST, the first req should be permitted, and the second one
714 should be denied by ENOSPC. But if the seconde arrived before
715 the first one is commited. The OST might wrongly permit second
716 writing, which will cause grant space > avaiable space.
719 Frequency : when client is evicted
721 Description: Grant might be wrongly erased when osc is evicted by OST
722 Details : when the import is evicted by server, it will fork another
723 thread ptlrpc_invalidate_import_thread to invalidate the
724 import, where the grant will be set to 0. While the original
725 thread will update the grant it got when connecting. So if
726 the former happened latter, the grant will be wrongly errased
727 because of this race.
732 Description: Checking Stale with correct fid
733 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
734 to check whether it is stale, because sometimes, we want the
735 enqueue happened anyway, and op_data.fid2 will not be initialized.
737 Severity : enhancement
739 Description: update patchless client
740 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
743 Frequency : only with 2.4 kernel
745 Description: random memory corruption
746 Details : size of struct ll_inode_info is to big for union inode.u and this
747 can be cause of random memory corruption.
752 Description: Memory leak in recovery
753 Details : Lov_mds_md was not free in an error handler in mds_create_object.
754 It should also check obd_fail before fsfilt_start, otherwise if
755 fsfilt_start return -EROFS,(failover mds during mds recovery).
756 then the req will return with repmsg->transno = 0 and rc = EROFS.
757 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
758 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
759 be freed no matter whether fsfilt_commit success or not.
762 Frequency : only with huge count clients
764 Description: Prevents from taking the superblock lock in llap_from_page for
766 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
767 for llap_from_page call in ll_removepage prevents from taking the
768 superblock lock for a soon died page.
773 Description: Not check open intent error before release open handle
774 Details : in some rare cases, the open intent error is not checked before
775 release open handle, which may cause
776 ASSERTION(open_req->rq_transno != 0), because it tries to release
777 the failed open handle.
782 Description: Set cat log bitmap only after create log success.
783 Details : in some rare cases, the cat log bitmap is set too early. and it
784 should be set only after create log success.
788 Description: Accessing a block bevice can re-enable I/O when Lustre is
789 tearing down a device.
790 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
795 Description: service threads may hog cpus when there are a lot of requests
797 Details : Insert cond_resched to give other threads a chance to use some of
803 Description: the cat log was not initialized in recovery
804 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
805 unlink log on mds will not be initialized until mds post
806 recovery. And also in mds post recovery, the unlink log will
807 initialization will be done asynchronausly, so there will be race
808 between add unlink log and unlink log initialization.
812 Description: brw_stats were being printed incorrectly
813 Details : brw_stats were being printed as log2 but all of them were not
814 recorded as log2. Also remove some code duplication arising from
815 filter_tally_{read,write}.
819 Frequency : rare, only in recovery.
820 Description: ASSERTION(req->rq_type != LI_POISON) failed
821 Details : imp_lock should be held while iterating over imp_sending_list for
822 prevent destroy request after get timeout in ptlrpc_queue_wait.
826 Description: replay-single.sh test 52 fails
827 Details : A lock's skiplist need to be cleanup when it being unlinked
828 from its resource list.
832 Description: Short directio read returns full requested size rather than
834 Details : Direct I/O operations should return actual amount of bytes
835 transferred rather than requested size.
837 Severity : enhancement
839 Description: metadata RPC reduction (e.g. for rm performance)
840 Details : decrease the amount of synchronous RPC between clients and servers
841 by canceling conflicing lock before the operation on the client side
842 and packing thier handles into the main operation RPC to server.
844 Severity : enhancement
846 Description: Async OSC create to avoid the blocking unnecessarily.
847 Details : If a OST has no remain object, system will block on the creating
848 when need to create a new object on this OST. Now, ways use
849 pre-created objects when available, instead of blocking on an
850 empty osc while others are not empty. If we must block, we block
851 for the shortest possible period of time.
853 --------------------------------------------------------------------------------
855 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
860 Frequency : on some architectures
862 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
863 Details : Uninitialized flags sometimes cause configuration commands to
867 Frequency : patchless clients only
869 Description: missing __iget() symbol export
870 Details : The __iget() symbol export is missing. To avoid the need for
871 this on patchless clients the deathrow inode reaper is turned
872 off, and we depend on the VM to clean up old inodes. This
873 dependency was during via the fix for bug 12181.
875 --------------------------------------------------------------------------------
877 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
879 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
880 INTEROPERATE with older versions automatically. In many cases a
881 special upgrade step is needed. Please read the
882 user documentation before upgrading any part of a live system.
883 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
884 version will not interoperate with 1.6 betas before beta5 (1.5.95).
885 * WARNING: Lustre configuration and startup changes are required with
886 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
892 Severity : enhancement
894 Description: MountConf
895 Details : Lustre configuration is now managed via mkfs and mount
896 commands instead of lmc and lconf. New obd types (MGS, MGC)
897 are added for dynamic configuration management. See
898 https://mail.clusterfs.com/wikis/lustre/MountConf for
901 Severity : enhancement
903 Description: dynamic OST addition
904 Details : OSTs can now be added to a live filesystem
906 Severity : enhancement
908 Description: startup order invariance
909 Details : MDTs and OSTs can be started in any order. Clients only
910 require the MDT to complete startup.
912 Severity : enhancement
914 Description: parallel, asynchronous orphan cleanup
915 Details : orphan cleanup is now performed in separate threads for each
916 OST, allowing parallel non-blocking operation.
918 Severity : enhancement
920 Description: optimized stripe assignment
921 Details : stripe assignments are now made based on ost space available,
922 ost previous usage, and OSS previous usage, in order to try
923 to optimize storage space and networking resources.
925 Severity : enhancement
927 Description: Permanently set tunables
928 Details : All writable /proc/fs/lustre tunables can now be permanently
929 set on a per-server basis, at mkfs time or on a live
932 Severity : enhancement
934 Description: Lustre message v2
935 Details : Add lustre message format v2.
937 Severity : enhancement
939 Description: client OST exclusion list
940 Details : Clients can be started with a list of OSTs that should be
941 declared "inactive" for known non-responsive OSTs.
945 Description: SPEC SFS validation failure on NFS v2 over lustre.
946 Details : Changes the blocksize for regular files to be 2x RPC size,
947 and not depend on stripe size.
949 Severity : enhancement
951 Description: Multiple MD RPCs in flight.
952 Details : Further unserialise some read-only MDS RPCs - learn about intents.
953 To avoid overly-overloading MDS, introduce a limit on number of
954 MDS RPCs in flight for a single client and add /proc controls
955 to adjust this limit.
957 Severity : enhancement
959 Description: client read/write statistics
960 Details : Add client read/write call usage stats for performance
961 analysis of user processes.
962 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
963 file access. extents_stats shows chunk size distribution.
964 extents_stats_per_process show chunk size distribution per
967 Severity : enhancement
969 Description: mds statistics
970 Details : Add detailed mds operations statistics in
971 /proc/fs/lustre/mds/*/stats.
975 Description: Failure of copying files with lustre special EAs.
976 Details : Client side always return success for setxattr call for lustre
977 special xattr (currently only "trusted.lov").
982 Description: Refcount LNET uuids
983 Details : The global LNET uuid list grew linearly with every startup;
984 refcount repeated list entries instead of always adding to
987 Severity : enhancement
989 Description: Dynamic service threads
990 Details : Within a small range, start extra service threads
991 automatically when the request queue builds up.
994 Frequency : mixed-endian client/server environments
996 Description: mixed-endian crashes
997 Details : The new msg_v2 system had some failures in mixed-endian
1000 Severity : enhancement
1002 Description: Easy OST removal
1003 Details : OSTs can be permanently deactivated with e.g. 'lctl
1004 conf_param lustre-OST0001.osc.active=0'
1006 Severity : enhancement
1008 Description: MGS proc entries
1009 Details : Added basic proc entries for the MGS showing what filesystems
1012 Severity : enhancement
1014 Description: provide MGS failover
1015 Details : Added config lock reacquisition after MGS server failover.
1017 Severity : enhancement
1019 Description: add Linux 2.4 support
1020 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1024 Description: a large application tries to do I/O to the same resource and dies
1025 in the middle of it.
1026 Details : Check the req->rq_arrival time after the call to
1027 ost_brw_lock_get(), but before we do anything about
1028 processing it & sending the BULK transfer request. This
1029 should help move old stale pending locks off the queue as
1030 quickly as obd_timeout.
1033 Frequency : when an incorrect nid is specified during startup
1035 Description: ptlrpc connect to non-existant node causes kernel crash
1036 Details : LNET can't be re-entered from an event callback, which
1037 happened when we expire a message after the export has been
1038 cleaned up. Instead, hand the zombie cleanup off to another
1041 Severity : enhancement
1043 Description: plain/inodebits lock performance improvement
1044 Details : Grouping plain/inodebits in granted list by their request modes
1045 and bits policy, thus improving the performance of search through
1049 Frequency : only if OST filesystem is corrupted
1051 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1052 Details : for a short time RPCs with bulk IO are in the replay list,
1053 but replay of bulk IOs is unimplemented. If the OST filesystem
1054 is corrupted due to disk cache incoherency and then replay is
1055 started it is possible to trip an assertion. Avoid putting
1056 committed RPCs into the replay list at all to avoid this issue.
1059 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1061 Description: System hang on startup
1062 Details : This bug allowed the liblustre (e.g. catamount) client to
1063 return to the app before handling all startup RPCs. This
1064 could leave the node unresponsive to lustre network traffic
1065 and manifested as a server ptllnd timeout.
1067 Severity : enhancement
1069 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1070 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1071 boolean to control whether to print peer debug info when a
1072 client's RPC times out.
1075 Frequency : only for kernels with patches from Lustre below 1.4.3
1077 Description: Remove old rdonly API
1078 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1081 Frequency : only for devices with external journals
1083 Description: Set external device read-only also
1084 Details : During a commanded failover stop, we set the disk device
1085 read-only while the server shuts down. We now also set any
1086 external journal device read-only at the same time.
1089 Frequency : when upgrading from 1.4 while trying to change parameters
1091 Description: The wrong (new) MDC name was used when setting parameters for
1092 upgraded MDT's. Also allows changing of OSC (and MDC)
1093 parameters if --writeconf is specified at tunefs upgrade time.
1096 Frequency : when setting specific ost indicies
1098 Description: QOS code breaks on skipped indicies
1099 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1100 created with --index need not be sequential.
1102 Severity : enhancement
1104 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1105 Details : The uninit_groups feature works in conjunction with the kernel
1106 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1107 up the pass1 processing of e2fsck. This is a read-only feature
1108 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1109 mount filesystems that have had this feature enabled.
1111 Severity : enhancement
1113 Description: Improve multi-block allocation algorithm to avoid fragmentation
1114 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1115 allocation locality and avoid filesystem fragmentation.
1117 ------------------------------------------------------------------------------
1119 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1121 * Support for kernels:
1122 2.4.21-47.0.1.EL (RHEL 3)
1123 2.6.5-7.283 (SLES 9)
1124 2.6.9-42.0.10.EL (RHEL 4)
1125 2.6.12.6 vanilla (kernel.org)
1126 2.6.16.27-0.9 (SLES 10)
1127 * Recommended e2fsprogs version: 1.39.cfs5
1129 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1133 Frequency : occasional, depends on client load and configuration
1134 Bugzilla : 12181, 12203
1135 Description: data loss for recently-modified files
1137 Details : In some cases it is possible that recently written or created
1138 files may not be written to disk in a timely manner (this should
1139 normally be within 30s unless client IO load is very high).
1140 The problem appears as zero-length files or files that are a
1141 multiple of 1MB in size after a client crash or client eviction
1142 that are missing data at the end of the file.
1144 This problem is more likely to be hit on clients where files are
1145 repeatedly created and unlinked in the same directory, clients
1146 have a large amount of RAM, have many CPUs, the filesystem has
1147 many OSTs, the clients are rebooted frequently, and/or the files
1148 are not accessed by other nodes after being written.
1150 The presence of the problem can be detected by looking at
1151 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1152 smaller than the second (nr_unused) then dirty files will not
1153 be flushed automatically to disk. "sync; sleep 10" should be
1154 run several times on the node before unmounting it to update
1155 Lustre (this is also safe to run on nodes without this problem).
1157 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1158 that can cause this same problem, so customers using that kernel
1159 also need to update the kernel in addition to Lustre. In order
1160 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1162 It is normal that files written just before a client crash (less
1163 than 30s) may not yet have been flushed to disk, even for local
1167 Frequency : frequent on thin XT3 nodes
1169 Description: UUID collision on thin XT3 Linux nodes
1170 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1171 randomly, since we relied on an insufficiently-seeded PRNG.
1176 Description: OSS hangs after "All ost request buffers busy"
1177 Details : A deadlock between quota and journal operations caused OSS
1178 hangs after printing "All ost request buffers busy."
1181 Frequency : always on liblustre builds
1183 Description: Cleanup compiler warnings on liblustre
1186 Frequency : always on liblustre builds on XT3
1188 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1192 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1193 > 32000 subdirectories
1194 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1195 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1196 subdirectory count crosses 32000. This will aid e2fsck to
1197 correctly handle more than 32000 subdirectories.
1200 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1202 Description: System hang on startup
1203 Details : This bug allowed the liblustre (e.g. catamount) client to
1204 return to the app before handling all startup RPCs. This
1205 could leave the node unresponsive to lustre network traffic
1206 and manifested as a server ptllnd timeout.
1208 Severity : enhancement
1210 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1211 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1212 boolean to control whether to print peer debug info when a
1213 client's RPC times out.
1218 Description: make O_SYNC working on 2.6 kernels
1219 Details : 2.6 kernels use different method for mark pages for write,
1220 so need add a code to lustre for O_SYNC work.
1225 Description: Failure to close file and release space on NFS
1226 Details : Put inode details into lock acquired in ll_intent_file_open.
1227 Use mdc_intent_lock in ll_intent_open to properly
1228 detect all kind of errors unhandled by mdc_enqueue.
1233 Description: proc file read during shutdown sometimes raced obd removal,
1235 Details : Add lock to prevent obd access after proc file removal.
1238 Frequency : Only for files larger than 4GB on 32-bit clients.
1240 Description: improperly doing page alignment of locks
1241 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1242 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1247 Description: RPCs being resent when they shouldn't be
1248 Details : Some RPCs that should not be resent are being resent. This
1249 can cause inconsistencies in the RPC state machine. Do not
1250 resend such requests.
1253 Frequency : rare, only with NFS export
1255 Description: Crash on NFS re-export node
1256 Details : under very unusual load conditions an assertion is hit in
1257 ll_intent_file_open()
1260 Frequency : only if OST filesystem is corrupted
1262 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1263 Details : for a short time RPCs with bulk IO are in the replay list,
1264 but replay of bulk IOs is unimplemented. If the OST filesystem
1265 is corrupted due to disk cache incoherency and then replay is
1266 started it is possible to trip an assertion. Avoid putting
1267 committed RPCs into the replay list at all to avoid this issue.
1272 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1273 Details : Large single O_DIRECT read and write calls can fail to allocate
1274 a sufficiently large buffer to process the request. In case of
1275 allocation failure the allocation is retried with a smaller
1276 buffer and broken into smaller requests.
1278 Severity : enhancement
1280 Description: Add -o localflock option to simulate old noflock behaviour.
1281 Details : This will achieve local-only flock/fcntl locks coherentness.
1286 Description: versioning check is incomplete
1287 Details : Checking the version difference of client vs. server, report
1288 error if the gap is too big.
1293 Description: add support PG_writeback bit
1294 Details : add support for PG_writeback bit for Lustre, for more carefull
1295 work with page cache in 2.6 kernel. This also fix some deadlocks
1296 and remove hack for work O_SYNC with 2.6 kernel.
1298 Severity : enhancement
1300 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1301 Details : The uninit_groups feature works in conjunction with the kernel
1302 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1303 up the pass1 processing of e2fsck. This is a read-only feature
1304 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1305 mount filesystems that have had this feature enabled.
1307 Severity : enhancement
1309 Description: Improve multi-block allocation algorithm to avoid fragmentation
1310 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1311 allocation locality and avoid filesystem fragmentation.
1313 ------------------------------------------------------------------------------
1315 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1317 * Support for kernels:
1318 2.6.9-42.0.3.EL (RHEL 4)
1319 2.6.5-7.276 (SLES 9)
1320 2.4.21-47.0.1.EL (RHEL 3)
1321 2.6.12.6 vanilla (kernel.org)
1322 2.6.16.21-0.8 (SLES10)
1323 * Recommended e2fsprogs version: 1.39.cfs2-0
1325 * The backwards-compatible /proc/sys/portals symlink has been removed
1326 in this release. Before upgrading, please ensure that you change
1327 any configuration scripts or /etc/sysctl.conf files that access
1328 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1329 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1330 in advance of the upgrade on any system running Lustre 1.4.6 or
1331 newer, since /proc/sys/lnet was added in that version.
1332 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1336 Frequency : only when quota is used
1338 Description: avoid scanning export list for quota master
1339 Details : Change the algorithms to avoid scanning export list in order
1340 to improve the efficiency.
1343 Frequency : MDS failover only, very rarely
1345 Description: "went back in time" messages on mds failover
1346 Details : The greatest transno may be lost when the current operation
1347 finishes with an error (transno==0) and the client's last_rcvd
1348 record is over-written. Save the greatest transno in the
1349 mds_last_transno for this case.
1352 Frequency : always for specific kernels and striping counts
1354 Description: client may get "Matching packet too big" without ACL support
1355 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1356 when trying to access files in certain configurations. The
1357 clients should in fact be denied when mounting because they do
1358 not understand ACLs.
1361 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1363 Description: many clients connecting with IO in progress causes connect timeouts
1364 Details : Avoid synchronous journal commits to avoid delays caused by many
1365 clients connecting/disconnecting when bulk IO is in progress.
1366 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1367 OST_IO_PORTAL to avoid delays behind potentially many pending
1371 Frequency : occasionally with multiple writers to a single file
1373 Description: shared writes to file may result in wrong size reported by stat()
1374 Details : Allow growing of kms when extent lock is cancelled
1377 Frequency : always with random mmap IO to multi-striped file
1379 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1380 Details : Only if the hole is at the end of OST object so that kms is too
1381 small. Fix is to increase kms accordingly in ll_nopage.
1384 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1386 Description: writes to a missing object would leak memory on the OST
1387 Details : If there is an inconsistency between the MDS and OST filesystems,
1388 such that the MDS references an object that doesn't exist, writes
1389 to that object will leak memory due to incorrect cleanup in the
1390 error handling path, eventually running out of memory on the OST.
1395 Description: Creating too long symlink causes lustre errors
1396 Details : Check symlink and name lengths before sending requests to MDS.
1399 Frequency : only if flock is enabled (not on by default)
1401 Description: posix locks not released on fd closure on 2.6.9+
1402 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1403 kernels, this caused such locks not to be released on fd close and
1404 then assertions on fs unmount about still used locks.
1407 Frequency : MDS failover only, very rarely
1409 Description: clients may get ASSERTION(granted_lock != NULL)
1410 Details : When request was taking a long time, and a client was resending
1411 a getattr by name lock request. The were multiple lock requests
1412 with the same client lock handle and
1413 mds_getattr_name->fixup_handle_for_resent_request found one of the
1414 lock handles but later failed with ASSERTION(granted_lock != NULL).
1419 Description: handle->h_buffer_credits > 0, assertion failure
1420 Details : h_buffer_credits is zero after truncate, causing assertion
1421 failure. This patch extends the transaction or creates a new
1425 Frequency : NFS re-export or patchless client
1426 Bugzilla : 11179, 10796
1427 Description: Crash on NFS re-export node (__d_move)
1428 Details : We do not want to hash the dentry if we don't have a lock.
1429 But if this dentry is later used in d_move, we'd hit uninitialised
1430 list head d_hash, so we just do this to init d_hash field but
1431 leave dentry unhashed.
1434 Frequency : NFS re-export or patchless client
1436 Description: NFS exports has problem with symbolic link
1437 Details : lustre client didn't properly install dentry when re-exported
1438 to NFS or running patchless client.
1441 Frequency : NFS re-export or patchless client
1443 Description: Various nfs/patchless fixes.
1444 Details : fixes reuse disconected alias for lookup process - this fixes
1445 warning "find_exported_dentry: npd != pd",
1446 fix permission error with open files at nfs.
1447 fix apply umask when do revalidate.
1450 Frequency : occasional
1452 Description: Crash on NFS re-export node
1453 Details : calling clear_page() on the wrong pointer triggered oops in
1454 generic_mapping_read().
1457 Frequency : rarely, using O_DIRECT IO
1459 Description: unaligned directio crashes client with LASSERT
1460 Details : check for unaligned buffers before trying any requests.
1463 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1465 Description: stale data returned from RAID cache
1466 Details : If only a small amount of IO is done to the RAID device before
1467 reading it again it is possible to get stale data from the RAID
1468 cache instead of reading it from disk.
1471 Frequency : always for sles10 kernel
1473 Description: sles10 support
1474 Details : ll_follow_link: compile fixes and using of nd_set_link
1475 under newer kernels.
1478 Frequency : depends on arch, kernel and compiler version, always on sles10
1481 Description: recursive or deep enough symlinks cause stack overflow
1482 Details : getting rid of large stack-allocated variable in
1486 Frequency : depends on hardware
1488 Description: lustre write performance loss in the SLES10 kernel
1489 Details : the performance loss is caused by using of write barriers in the
1490 ext3 code. The SLES10 kernel turns barrier support on by
1491 default. The fix is to undo that change for ldiskfs.
1493 ------------------------------------------------------------------------------
1495 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1497 * Support for kernels:
1498 2.6.9-42.0.3EL (RHEL 4)
1499 2.6.5-7.276 (SLES 9)
1500 2.4.21-47.0.1.EL (RHEL 3)
1501 2.6.12.6 vanilla (kernel.org)
1505 Frequency : quota enabled and large files being deleted
1507 Description: releasing more than 4GB of quota at once hangs OST
1508 Details : If a user deletes more than 4GB of files on a single OST it
1509 will cause the OST to spin in an infinite loop. Release
1510 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1515 Description: statfs data retrieved from /proc may be stale or zero
1516 Details : When reading per-device statfs data from /proc, in the
1517 {kbytes,files}_{total,free,avail} files, it may appear
1518 as zero or be out of date.
1521 Frequency : systems with MD RAID1 external journal devices
1523 Description: lconf's call to blkid is confused by RAID1 journal devices
1524 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1525 of returning all block devices that match the journal UUID.
1528 Frequency : always, for aggregate stripe size over 4GB
1530 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1531 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1532 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1535 Frequency : always if "lfs find" used on a local file/directory
1537 Description: "lfs find" segfaults if used on a local file/directory
1538 Details : The case where a directory component was not specified wasn't
1539 handled correctly. Handle this properly.
1542 Frequency : always on ppc64
1544 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1545 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1546 perform endianness conversion.
1549 Frequency : rarely (truncate to non-zero file size after write under load)
1550 Bugzilla : 10730, 10687
1551 Description: Files padded with zeros to next 4K multiple
1552 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1553 it is possible that files that are truncated to a non-zero size
1554 immediately after being written are filled with zero bytes beyond
1555 the truncated size. No file data is lost.
1557 Severity : enhancement
1559 Description: Allow recovery/failover for liblustre clients.
1560 Details : liblustre clients were unaware of failover configurations until
1563 Severity : enhancement
1565 Description: user file locks should fail when not mounting with flock option
1566 Details : Set up an error-returning stub in ll_file_operations.lock field
1567 to prevent incorrect behaviour when client is mounted without
1568 flock option. Also, set up properly f_op->flock field for
1572 Frequency : always on ia64
1574 Description: "lfs df" loops on printing out MDS statfs information
1575 Details : The obd_ioctl_data was not initialized and in some systems
1576 this caused a failure during the ioctl that did not return
1577 an error. Initialize the struct and return an error on failure.
1580 Frequency : SLES 9 only
1582 Description: Error of copying files with lustre special EAs as root
1583 Details : Client side always return success for setxattr call for lustre
1584 special xattr (currently only "trusted.lov").
1587 Frequency : rarely on clusters with both ia64+i386 clients
1589 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1590 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1591 clients are doing concurrent writes to the same file it is
1592 possible that the ia64 clients may LASSERT because the OST
1593 extent locks are not PAGE_SIZE aligned. Ensure that grown
1594 locks are always aligned on the request boundary.
1597 Frequency : specific use, occasional
1599 Description: Overwriting in use executable truncates on-disk binary image
1600 Details : If one node attempts to overwrite an executable in use by
1601 another node, we now correctly return ETXTBSY instead of
1602 truncating the file.
1604 Severity : enhancement
1606 Description: Async OSC create to avoid the blocking unnecessarily.
1607 Details : If a OST has no remain object, system will block on the creating
1608 when need to create a new object on this OST. Now, ways use
1609 pre-created objects when available, instead of blocking on an
1610 empty osc while others are not empty. If we must block, we block
1611 for the shortest possible period of time.
1616 Description: chmod on Lustre root is propagated to other clients
1617 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1618 invalid by the follow_mount time.
1621 Frequency : liblustre clients only
1623 Description: Race in 'instant cancel' lock handling could lead to such locks
1624 never to be granted in case of SMP MDS
1625 Details : Do not destroy not yet granted but cbpending locks in
1629 Frequency : replay/resend of open
1631 Description: non null lock assetion failure in mds_intent_policy
1632 Details : Trying to replay/resend lockless open requests resulted in
1633 mds_open() returning 0 with no lock. Now it sets a flag if
1634 a lock is going to be returned.
1636 Severity : enhancement
1638 Description: Checksum enhancements
1639 Details : New checksum enhancements allow for resending RPCs that failed
1642 Severity : enhancement
1644 Description: Tunables on number of dirty pages in cacche
1645 Details : Allow to set limit on number of dirty pages cached.
1650 Description: client crash on unmount - lock still has references
1651 Details : In some error handling cases it was possible to leak a lock
1652 reference on a client while accessing a file. This was not
1653 harmful to the client during operation, but would cause the
1654 client to crash when the filesystem is unmounted.
1657 Frequency : specific case, rare
1659 Description: ETXTBSY on mds though file not in use by client
1660 Details : ETXTBSY is no longer incorrectly returned when attempting to
1661 chmod or chown a directory that the user previously tried to
1662 execute or a currently-executing binary.
1665 Frequency : extremely rare except on liblustre-based clients
1667 Description: Lustre space not freed when files are deleted
1668 Details : Clean up open-unlinked files after client eviction. Previously
1669 the unlink was skipped and the files remained as orphans.
1674 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1675 Details : In some cases it was possible to send a blocking callback to
1676 a client doing a glimpse, even though that client didn't get
1677 a lock granted. When the glimpse lock is cancelled on the OST
1678 the freed lock is left on the waiting list and corrupted the list.
1681 Frequency : all core dumps
1683 Description: Broke core dumps to lustre
1684 Details : Negative dentry may be unhashed if parent does not have UPDATE
1685 lock, but some callers, e.g. do_coredump, expect dentry to be
1686 hashed after successful create, hash it in ll_create_it.
1688 ------------------------------------------------------------------------------
1690 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1692 * Support for kernels:
1693 2.6.9-42.0.2.EL (RHEL 4)
1694 2.6.5-7.276 (SLES 9)
1695 2.4.21-40.EL (RHEL 3)
1696 2.6.12.6 vanilla (kernel.org)
1700 Frequency : always on RHEL 3
1702 Description: Number of open files grows over time
1703 Details : The number of open files grows over time, whether or not
1704 Lustre is started. This was due to a filp leak introduced
1705 by one of our kernel patches.
1707 ------------------------------------------------------------------------------
1709 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1711 * Support for kernels:
1712 2.6.9-42.EL (RHEL 4)
1713 2.6.5-7.276 (SLES 9)
1714 2.4.21-40.EL (RHEL 3)
1715 2.6.12.6 vanilla (kernel.org)
1720 Bugzilla : 5719, 9635, 9792, 9684
1721 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1722 Details : If a server is under heavy load and cannot reply to new
1723 connection requests before the client resends the (re)connect,
1724 the connection handling code can behave badly if two service
1725 threads are concurrently handing separate (re)connections from
1726 the same client. Add better locking to the connection handling
1727 code, and ensure that only a single connection will be processed
1728 for a given client UUID, even if the lock is dropped.
1730 Severity : enhancement
1732 Description: add TCP zero-copy support to kernel
1733 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1734 sends if the hardware supports scatter-gather and checksumming.
1735 This allows socklnd to do client-write and server-read more
1736 efficiently and reduce CPU utilization from skbuf copying.
1739 Frequency : only if NFS exporting from client
1741 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1742 Details : If setting attributes on a file created under NFS that had
1743 never been opened it would be possible to oops the client
1744 if the file had no objects.
1747 Frequency : always for liblustre
1749 Description: liblustre client does MDS+OSTs setattr RPC for each write
1750 Details : When doing a write from a liblustre client, the client
1751 incorrectly issued an RPC to the MDS and each OST the file was
1752 striped over in order to update the timestamps. When writing
1753 with small chunks and many clients this could overwhelm the MDS
1754 with RPCs. In all cases it would slow down the write because
1755 these RPCs are unnecessary.
1757 Severity : enhancement
1759 Description: allow number of MDS service threads to be changed at module load
1760 Details : It is now possible to change the number of MDS service threads
1761 running. Adding "options mds mds_num_threads={N}" to the MDS's
1762 /etc/modprobe.conf will set the number of threads for the next
1763 time Lustre is restarted (assuming the "mds" module is also
1764 reloaded at that time). The default number of threads will
1765 stay the same, 32 for most systems.
1770 Description: OST crash if filesystem is unformatted or corrupt
1771 Details : If an OST is started on a device that has never been formatted
1772 or if the filesystem is corrupt and cannot even mount then the
1773 error handling cleanup routines would dereference a NULL pointer.
1778 Description: NULL pointer deref in llap_from_page.
1779 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1780 incorrect) mapping if the page was truncated/reclaimed while it was
1781 searched for. Check for this condition and skip such pages when
1782 doing readahead. Introduce extra check to llap_from_page() to
1783 verify page->mapping->host is non-NULL (so page is not anonymous).
1786 Frequency : Sometimes when using sys_sendfile
1788 Description: "page not covered by a lock" warnings from ll_readpage
1789 Details : sendfile called ll_readpage without right page locks present.
1790 Now we introduced ll_file_sendfile that does necessary locking
1791 around call to generic_file_sendfile() much like we do in
1795 Frequency : with certain MDS communication failures at client mount time
1797 Description: NULL pointer deref after failed client mount
1798 Details : a client connection request may delayed by the network layer
1799 and not be sent until after the PTLRPC layer has timed out the
1800 request. If the client fails the mount immediately it will try
1801 to clean up before the network times out the request. Add a
1802 reference from the request import to the obd device and delay
1803 the cleanup until the network drops the request.
1806 Frequency : occasionally during client (re)connect
1808 Description: assertion failure during client (re)connect
1809 Details : processing a client connection request may be delayed by the
1810 client or server longer than the client connect timeout. This
1811 causes the client to resend the connection request. If the
1812 original connection request is replied in this interval, the
1813 client may trip an assertion failure in ptlrpc_connect_interpret()
1814 which thought it would be the only running connect process.
1817 Frequency : only with obd_echo servers and clients that are rebooted
1819 Description: kernel BUG accessing uninitialized data structure
1820 Details : When running an obd_echo server it did not start the ping_evictor
1821 thread, and when a client was evicted an uninitialized data
1822 structure was accessed. Start the ping_evictor in the RPC
1823 service startup instead of the OBD startup.
1825 Severity : enhancement
1826 Bugzilla : 10193 (patchless)
1827 Description: Remove dependency on various unexported kernel interfaces.
1828 Details : No longer need reparent_to_init, exit_mm, exit_files,
1829 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1832 Frequency : rare (only users of deprecated and unsupported LDAP config)
1834 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1835 Details : LDAP apparently contains 'lustreName' attributes instead of
1836 'name'. A simple remapping of the name is sufficient.
1839 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1841 Description: waiting_locks_callback trips kernel BUG if client is evicted
1842 Details : Running with the dump_on_timeout debug flag turned on makes
1843 it possible that the waiting_locks_callback() can try to dump
1844 the Lustre kernel debug logs from an interrupt handler. Defer
1845 this log dumping to the expired_lock_main() thread.
1847 Severity : enhancement
1849 Description: Support NFS exporting on 2.6 kernels.
1850 Details : Implement non-rawops metadata methods for NFS server to use without
1851 changing NFS server code.
1854 Frequency : very rare (synthetic metadata workload only)
1856 Description: two racing renames might cause an MDS thread to deadlock
1857 Details : Running the "racer" program may cause one MDS thread to rename
1858 a file from being the source of a rename to being the target of
1859 a rename at exactly the same time that another thread is doing
1860 so, and the second thread has already enqueued these locks after
1861 doing a lookup of the target and is trying to relock them in
1862 order. Ensure that we don't try to re-lock the same resource.
1865 Frequency : only very large systems with liblustre clients
1867 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1868 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1869 all OSTs in parallel. This allows the eviction of stale liblustre
1870 clients to proceed much faster than if they were done in series,
1871 and also offers similar improvements for other set_info RPCs.
1876 Description: excessive CPU usage during initial read phase on client
1877 Details : During the initial read phase on a client, it would agressively
1878 retry readahead on the file, consuming too much CPU and impacting
1879 performance (since 1.4.5.8). Improve the readahead algorithm
1880 to avoid this, and also improve some other common cases (read
1881 of small files in particular, where "small" is files smaller than
1882 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1887 Description: MDS crash when receiving packet with unknown intent.
1888 Details : Do not LBUG in unknown intent case, just return -EFAULT
1890 Severity : enhancement
1891 Bugzilla : 9293, 9385
1892 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1893 Details : Do not serialize getattr (non-intent version) and statfs.
1896 Frequency : occasional, when OST network is overloaded/intermittent
1898 Description: client evicted by OST after bulk IO timeout
1899 Details : If a client sends a bulk IO request (read or write) the OST
1900 may evict the client if it is unresposive to its data GET/PUT
1901 request. This is incorrect if the network is overloaded (takes
1902 too long to transfer the RPC data) or dropped the OST GET/PUT
1903 request. There is no need to evict the client at all, since
1904 the pinger and/or lock callbacks will handle this, and the
1905 client can restart the bulk request.
1908 Frequency : Always when mmapping file with no objects
1910 Description: client crashes when mmapping file with no objects
1911 Details : Check that we actually have objects in a file before doing any
1912 operations on objects in ll_vm_open, ll_vm_close and
1918 Description: Request leak when working with deleted CWD
1919 Details : Introduce advanced request refcount tracking for requests
1920 referenced from lustre intent.
1922 Severity : Enhancement
1924 Description: Cache open file handles on client.
1925 Details : MDS now will return special lock along with openhandle, if
1926 requested and client is allowed to hold openhandle, even if unused,
1927 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1928 closing files for every read/write openration.
1930 Severity : Enhancement
1932 Description: Cache open negative dentries on client when possible.
1933 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1934 negative dentries on lock revocation.
1939 Description: Remounting a client read-only wasn't possible with a zconf mount
1940 Details : It wasn't possible to remount a client read-only with llmount.
1942 Severity : enhancement
1943 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1944 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1945 MPICH is included to add Lustre-specific ADIO interfaces.
1946 This is based closely on the UFS ADIO layer and only differs
1947 in file creation, in order to allow the OST striping to be set.
1948 This is user-contributed code and not supported by CFS.
1953 Description: extended inode attributes (immutable, append-only) work improperly
1954 when 2.4 and 2.6 kernels are used on client/server or vice versa
1955 Details : Introduce kernel-independent values for these flags.
1957 Severity : enhancement
1960 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1961 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1962 granularity. Now, Fractional MB granularity is be supported,
1963 this is very useful for low memory system.
1965 Severity : enhancement
1967 Description: Getattr by fid
1968 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1969 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1970 have locks and avoids getattr rpc requests for it on every stat.
1973 Frequency : Always, for filesystems larger than 2TB
1975 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1976 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1977 filesystems larger than 2TB to be mounted with mballoc enabled.
1980 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1982 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1983 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1984 filesystems larger than 2TB is mounted then the kernel will
1985 silently corrupt the start of the filesystem. CONFIG_LBD is
1986 enabled for all CFS-supported kernels, but the possibility of
1987 this happening with a modified kernel config exists.
1989 Severity : enhancement
1991 Description: add client O_DIRECT support for 2.6 kernels
1992 Details : It is now possible to do O_DIRECT reads and writes to files
1993 in the Lustre client mountpoint on 2.6 kernel clients.
1995 Severity : enhancement
1997 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1998 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1999 parallel, not waiting for response from every OST before sending
2000 a rpc to the next OST.
2005 Description: setattr vs write race when updating file timestamps
2006 Details : Client processes that update a file timestamp into the past
2007 right after writing to the file (e.g. tar) it is possible that
2008 the updated file modification time can be reset to the current
2009 time due to a race between processing the setattr and write RPC.
2011 Severity : enhancement
2013 Description: Bring 'lfs find' closer in line with regular Linux find.
2014 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2015 -print0 options and obtains all the needed info through the lustre
2018 Severity : enhancement
2020 Description: support up to 1024 configured devices on one node
2021 Details : change obd_dev array from statically allocated to dynamically
2022 allocated structs as they are first used to reduce memory usage
2027 Description: Flush dirty partially truncated pages during truncate
2028 Details : Immediatelly flush partially truncated pages in filter_setattr,
2029 this way we completely avoid having any pages in page cache on OST
2030 and can retire ugly workarounds during writes to flush such pages.
2035 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2036 Details : protect inode from truncation within vfs_unlink() context
2037 just take a reference before calling vfs_unlink() and release it
2038 when parent's i_sem is free.
2043 Description: last_id value checked outside lock on OST caused LASSERT failure
2044 Details : If there were multiple MDS->OST object precreate requests in
2045 flight, it was possible that the OST's last object id was checked
2046 outside a lock and incorrectly tripped an assertion. Move checks
2047 inside locks, and discard old precreate requests.
2050 Frequency : always, if extents are used on OSTs
2052 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2053 Details : OSTs using the extents format would not zero the high 16 bits of
2054 the index physical block number. This is not a problem for any
2055 OST filesystems smaller than 16TB, and no kernels support ext3
2056 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2057 new/modified files) and can be fixed for existing filesystems
2058 with e2fsprogs-1.39-cfs1.
2063 Description: import connection selection may be incorrect if timer wraps
2064 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2065 import connections to be ignored if the 32-bit jiffies counter
2066 wraps. Use a 64-bit jiffies counter.
2069 Frequency : very large clusters immediately after boot
2071 Description: LNET request buffers exhausted under heavy short-term load
2072 Details : If a large number of client requests are generated on a service
2073 that has previously never seen so many requests it is possible
2074 that the request buffer growth cannot keep up with the spike in
2075 demand. Instead of dropping incoming requests, they are held in
2076 the LND until the RPC service can accept more requests.
2079 Frequency : Sometimes during replay
2081 Description: Assertion failure in ll_local_open after replay.
2082 Details : If replay happened on an open request reply before we were able
2083 to set replay handler, reply will become not swabbed tripping the
2084 assertion in ll_local_open. Now we set the handler right after
2085 recognising of open request
2088 Frequency : very rare
2090 Description: kernel reports "badness in vsnprintf"
2091 Details : Reading from the "recovery_status" /proc file in small chunks
2092 may cause a negative length in lprocfs_obd_rd_recovery_status()
2093 call to vsnprintf() (which is otherwise harmless). Exit early
2094 if there is no more space in the output buffer.
2096 Severity : enhancement
2098 Description: clear OBD RPC statistics by writing to them
2099 Details : It is now possible to clear the OBD RPC statistics by writing
2100 to the "stats" file.
2105 Description: Client mtime is not the same on different clients after utimes
2106 Details : In some cases, the client was using the utimes() syscall on
2107 a file cached on another node. The clients now validate the
2108 ctime from the MDS + OSTs to determine which one is right.
2113 Description: Inability to activate failout mode
2114 Details : lconf script incorrectly assumed that in pythong string's numeric
2115 value is used in comparisons.
2118 Frequency : always with multiple stripes per file
2120 Description: Inefficient object allocation for mutli-stripe files
2121 Details : When selecting which OSTs to stripe files over, for files with
2122 a stripe count that divides evenly into the number of OSTs,
2123 the MDS is always picking the same starting OST for each file.
2124 Return the OST selection heuristic to the original design.
2129 Description: mount failures may take full timeout to return an error
2130 Details : Under some heavy load conditions it is possible that a
2131 failed mount can wait for the full obd_timeout interval,
2132 possibly several minutes, before reporting an error.
2133 Instead return an error as soon as the status is known.
2135 Frequency : quota enabled and large files being deleted
2137 Description: releasing more than 4GB of quota at once hangs OST
2138 Details : If a user deletes more than 4GB of files on a single OST it
2139 will cause the OST to spin in an infinite loop. Release
2140 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2145 Description: statfs data retrieved from /proc may be stale or zero
2146 Details : When reading per-device statfs data from /proc, in the
2147 {kbytes,files}_{total,free,avail} files, it may appear
2148 as zero or be out of date.
2151 Frequency : systems with MD RAID1 external journal devices
2153 Description: lconf's call to blkid is confused by RAID1 journal devices
2154 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2155 of returning all block devices that match the journal UUID.
2158 Frequency : always, for aggregate stripe size over 4GB
2160 Description: assertion fails when trying to use 4GB stripe size
2161 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2162 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2165 Frequency : always on ppc64
2167 Description: the first write on an ext3 filesystem with mballoc got stuck
2168 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2169 perform endianness conversion.
2171 ------------------------------------------------------------------------------
2173 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2175 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2176 INTEROPERATE with older versions automatically. Please read the
2177 user documentation before upgrading any part of a live system.
2178 * WARNING: Lustre networking configuration changes are required with
2179 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2182 * Support for kernels:
2183 2.6.9-22.0.2.EL (RHEL 4)
2184 2.6.5-7.244 (SLES 9)
2185 2.6.12.6 vanilla (kernel.org)
2188 Severity : enhancement
2189 Bugzilla : 7981/8208
2190 Description: Introduced Lustre Networking (LNET)
2191 Details : LNET is new networking infrastructure for Lustre, it includes
2192 a reorganized network configuration mode (see the user
2193 documentation for full details) as well as support for routing
2194 between different network fabrics. Lustre Networking Devices
2195 (LNDs) for the supported network fabrics have also been
2196 created for this new infrastructure.
2198 Severity : enhancement
2199 Description: Introduced Access control lists
2200 Details : clients can set ACLs on files and directories in order to have
2201 more fine-grained permissions than the standard Unix UGO+RWX.
2202 The MDS must be started with the "-o acl" mount option.
2204 Severity : enhancement
2205 Description: Introduced filesystem quotas
2206 Details : Administrators may now establish per-user quotas on the
2209 Severity : enhancement
2211 Description: Configuration change for the XT3
2212 The PTLLND is now used to run Lustre over Portals on the XT3
2213 The configure option(s) --with-cray-portals are no longer used.
2214 Rather --with-portals=<path-to-portals-includes> is used to
2215 enable building on the XT3. In addition to enable XT3 specific
2216 features the option --enable-cray-xt3 must be used.
2221 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2222 Details : A race between allocating a new llcd and re-getting the llcd_lock
2223 allowed another thread to grab newly-allocated llcd.
2225 Severity : enhancement
2227 Description: 2.6 OST async journal commit and locking fix to improve performance
2228 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2229 2.6 kernels are now async as they already were in 2.4 kernels so
2230 that they can commit concurrently with the network bulk transfer.
2231 For block-allocated files the filter allocation semaphore is held
2232 to avoid filesystem fragmentation during allocation. BKL lock
2233 removed for 2.6 xattr operations where it is no longer needed.
2238 Description: lconf incorrectly determined whether two IP networks could talk
2239 Details : In some more complicated routing and multiple-network
2240 configurations, lconf will avoid trying to make a network
2241 connection to a disjoint part of the IP space. It was doing the
2242 math incorrectly for one set of cases.
2247 Description: Fix for potential infinite loop processing records in an llog.
2248 Details : If an llog record is corrupted/zeroed, it is possible to loop
2249 forever in llog_process(). Validate the llog record length
2250 and skip the remainder of the block on error.
2253 Frequency : occasional (liblustre only)
2255 Description: liblustre could not open files whose last component is a symlink
2256 Details : sysio_path_walk() would incorrectly pass the open intent to
2257 intermediate path components.
2260 Frequency : rare (liblustre only with non-standard tuning)
2261 Bugzilla : 7201 (7350)
2262 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2263 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2264 after it has been released, only do this for referenced locks
2266 Severity : enhancement
2268 Description: specifying an (invalid) directory default stripe_size of -1
2269 would reset the directory default striping
2270 Details : stripe_size -1 was used internally to signal directory stripe
2271 removal, now use "all default" to signal dir stripe removal
2272 as a directory striping of "all default" is not useful
2275 Frequency : common for large clusters running liblustre clients
2277 Description: doing an ls when liblustre clients are running is slow
2278 Details : sending a glimpse AST to a liblustre client waits for every AST
2279 to time out, as liblustre clients will not respond. Since they
2280 cannot cache data we refresh the OST lock LVB from disk instead.
2282 Severity : enhancement
2284 Description: doing an ls at the same time as file IO can be slow
2285 Details : enqueue and other "small" requests can be blocked behind many
2286 large IO requests. Create a new OST IO portal for non-IO
2287 requests so they can be processed faster.
2290 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2292 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2293 created file with mode 000
2294 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2295 allow an ftruncate() as a non-root user to a file with mode 000.
2296 Lustre now respects this flag to disable mode checks when
2297 truncating a file owned by the user
2300 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2303 Description: Revoking locks from clients that went dead or catatonic might take
2305 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2306 cancellation of such locks instant on servers without waiting for
2307 any reply from clients. Clients drops these locks when cancel
2308 notification from server is received without replying.
2311 Frequency : liblustre-only, when liblustre client dies or becomes busy
2313 Description: Doing ls on Linux clients can take a long time with active
2315 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2316 granting such locks to it in the first place if possible. Locks
2317 are taken by proxy on the OST during the read or write and
2318 dropped immediately afterward. Add connect flags handling, do
2319 not grant locks to liblustre clients for glimpse ASTs.
2321 Severity : enhancement
2323 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2324 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2325 sequential reads and behaves badly with applications doing random
2326 reads. Improve it by reading ahead at least read region, and
2327 avoiding excessive large RPC for small reads.
2329 Severity : enhancement
2331 Description: Creating more than 1000 files for a single job may cause a load
2332 imbalance on the OSTs if there are also a large number of OSTs.
2333 Details : qos_prep_create() uses an OST index reseed value that is an
2334 even multiple of the number of available OSTs so that if the
2335 reseed happens in the middle of the object allocation it will
2336 still utilize the OSTs as uniformly as possible.
2341 Description: OST or MDS may oops in ping_evictor_main()
2342 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2343 but doesn't restart at beginning of obd_exports_timed list
2346 Severity : enhancement
2348 Description: improve by-nid export eviction on the MDS and OST
2349 Details : allow multiple exports with the same NID to be evicted at one
2350 time without re-searching the exports list.
2353 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2355 Description: MDS may oops in groups_free()
2356 Details : in rare race conditions a newly allocated group_info struct is
2357 freed again, and this can be NULL. The 2.4 compatibility code
2358 for groups_free() checked for a NULL pointer, but 2.6 did not.
2361 Frequency : common for liblustre clients doing little filesystem IO
2362 Bugzilla : 9352, 7313
2363 Description: server may evict liblustre clients accessing contended locks
2364 Details : if a client is granted a lock or receives a completion AST
2365 with a blocking AST already set it would not reply to the AST
2366 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2369 Frequency : lfs setstripe, only systems with more than 160 OSTs
2371 Description: unable to set striping with a starting offset beyond OST 160
2372 Details : llapi_create_file() incorrectly limited the starting stripe
2373 index to the maximum single-file stripe count.
2376 Frequency : LDAP users only
2378 Description: lconf did not handle in-kernel recovery with LDAP properly
2379 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2381 Severity : enhancement
2383 Description: bind OST threads to NUMA nodes to improve performance
2384 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2385 node and do their allocations there to localize memory access
2387 Severity : enhancement
2389 Description: llmount can determine client NID directly from Myrinet (GM)
2390 Details : the client NID code from gmnalnid was moved directly into
2391 llmount, removing the need to use this or specifying the
2392 client NID explicitly when mounting GM clients with zeroconf
2395 Frequency : if client is started with down MDS
2397 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2398 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2399 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2400 ptlrpc_import_delayed_req() because we didn't check if the
2401 request was interrupted, and we also didn't break out of the
2402 event loop if there was no timeout
2407 Description: data loss during non-page-aligned writes to a single file from
2408 both multiple nodes and multiple threads on one node at same time
2409 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2410 inconsistency led to false short-reads, that were cached and later
2411 used by ->prepare_write() to fill in partially written page,
2412 leading to data loss.
2415 Frequency : always, if lconf --abort_recovery used
2417 Description: lconf --abort_recovery fails with 'Operation not supported'
2418 Details : lconf was attempting to abort recovery on the MDT device and not
2421 Severity : enhancement
2423 Description: remove cleanup logs
2424 Details : replace lconf-generated cleanup logs with lustre internal
2425 cleanup routines. Eliminates the need for client-cleanup and
2428 Severity : enhancement
2430 Description: add support for EAs (user and system) on lustre filesystems
2431 Details : it is now possible to store extended attributes in the Lustre
2432 client filesystem, and with the user_xattr mount option it
2433 is possible to allow users to store EAs on their files also
2435 Severity : enhancement
2437 Description: Add possibility (config option) to show minimal available OST free
2439 Details : When compiled with --enable-mindf configure option, statfs(2)
2440 (and so, df) will return least minimal free space available from
2441 all OSTs as amount of free space on FS, instead of summary of
2442 free spaces of all OSTs.
2444 Severity : enhancement
2446 Description: do not expand extent locks acquired on OST-side
2447 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2448 by server: they are not cached anyway.
2451 Frequency : when mmap is used/binaries executed from Lustre
2453 Description: Unmmap pages before throwing them away from read cache.
2454 Details : llap_shrink cache now attempts to unmap pages before discarding
2455 them (if unmapping failed - do not discard). SLES9 kernel has
2456 extra checks that trigger if this unmapping is not done first.
2461 Description: lconf didn't resolve symlinks before checking to see whether a
2462 given mountpoint was already in use
2465 Frequency : when migrating failover services
2466 Bugzilla : 6395, 9514
2467 Description: When migrating a subset of services from a node (e.g. failback
2468 from a failover service node) the remaining services would
2469 time out and evict clients.
2470 Details : lconf --force (implied by --failover) sets the global obd_timeout
2471 to 5 seconds in order to quickly disconnect, but this caused
2472 other RPCs to time out too quickly. Do not change the global
2473 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2475 Severity : enhancement
2476 Frequency : if MDS is started with down OST
2477 Bugzilla : 9439,5706
2478 Description: Allow startup/shutdown of an MDS without depending on the
2479 availability of the OSTs.
2480 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2481 Add appropriate locking and lov-osc refcounts for safe
2482 cleaning. Add osc abort_inflight calls in case the
2483 synchronize never started.
2486 Frequency : occasional (Cray XT3 only)
2488 Description: root not authorized to access files in CRAY_PORTALS environment
2489 Details : The client process capabilities were not honoured on the MDS in
2490 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2491 been accessed by an authorized user then root was able to access
2492 the file on the local client also. The root user capabilities
2493 are now allowed on the MDS, as this environment has secure UID.
2496 Frequency : occasional
2498 Description: ldiskfs "too long searching" message happens too often
2499 Details : A debugging message (otherwise harmless) prints too often on
2500 the OST console. This has been reduced to only happen when
2501 there are fragmentation problems on the filesystem.
2506 Description: Division by zero in statfs when all OSCs are inactive
2507 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2508 lov_statfs divides by value returned by lov_get_stripecnt().
2512 Bugzilla : 9489, 3273
2513 Description: First write from each client to each OST was only 4kB in size,
2514 to initialize client writeback cache, which caused sub-optimal
2515 RPCs and poor layout on disk for the first writen file.
2516 Details : Clients now request an initial cache grant at (re)connect time
2517 and so that they can start streaming writes to the cache right
2518 away and always do full-sized RPCs if there is enough data.
2519 If the OST is rebooted the client also re-establishes its grant
2520 so that client cached writes will be honoured under the grant.
2525 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2526 Details : Now I/O RPCs go to different portal number and (presumably) fast
2527 lock requests (and glimses) and other RPCs get their own service
2528 threads pool that should be able to service those RPCs
2531 Severity : enhancement
2533 Description: Ability to exchange lustre version between client and servers and
2534 issue warnings at client side if client is too old. Also for
2535 liblustre clients there is ability to refuse connection of too old
2537 Details : New 'version' field is added to connect data structure that is
2538 filled with version info. That info is later checked by server and
2542 Frequency : rare, liblustre only.
2543 Bugzilla : 9296, 9581
2544 Description: Two simultaneous writes from liblustre at offset within same page
2545 might proceed at the same time overwriting eachother with stale
2547 Details : I/O lock withing llu_file_prwv was released too early, before data
2548 actually was hitting the wire. Extended lock-holding time until
2549 server acknowledges receiving data.
2552 Frequency : extremely rare. Never observed in practice.
2554 Description: avoid generating lustre_handle cookie of 0.
2555 Details : class_handle_hash() generates handle cookies by incrementing
2556 global counter, and can hit 0 occasionaly (this is unlikely, but
2557 not impossible, because initial value of cookie counter is
2558 selected randonly). Value of 0 is used as a sentinel meaning
2559 "unassigned handle" --- avoid it. Also coalesce two critical
2560 sections in this function into one.
2562 Severity : enhancement
2564 Description: allow liblustre clients to delegate truncate locking to OST
2565 Details : To avoid overhead of locking, liblustre client instructs OST to
2566 take extent lock in ost_punch() on client's behalf. New connection
2567 flag is added to handle backward compatibility.
2569 Severity : enhancement
2570 Bugzilla : 4928, 7341, 9758
2571 Description: allow number of OST service threads to be specified
2572 Details : a module parameter allows the number of OST service threads
2573 to be specified via "options ost ost_num_threads={N}" in the
2574 OSS's /etc/modules.conf or /etc/modprobe.conf.
2578 Bugzilla : 6146, 9635, 9895
2579 Description: servers crash with bad pointer in target_handle_connect()
2580 Details : In rare cases when a client is reconnecting it was possible that
2581 the connection request was the last reference for that export.
2582 We would temporarily drop the export reference and get a new
2583 one, but this may have been the last reference and the export
2584 was just destroyed. Get new reference before dropping old one.
2586 Severity : enhancement
2587 Frequency : if client is started with failover MDS
2589 Description: Allow multiple MDS hostnames in the mount command
2590 Details : Try to read the configuration from all specified MDS
2591 hostnames during a client mount in case the "primary"
2594 Severity : enhancement
2596 Description: Stop sending data to evicted clients as soon as possible.
2597 Details : Check if the client we are about to send or are sending data to
2598 was evicted already. (Check is done every second of waiting,
2599 for which l_wait_event interface was extended to allow checking
2600 of exit condition at specified intervals).
2603 Frequency : rare, normally only when NFS exporting is done from client
2605 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2607 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2610 Frequency : always, if srand() is called before liblustre initialization
2612 Description: Liblustre uses system PRNG disturbing its usage by user application
2613 Details : Introduce internal to lustre fast and high-quality PRNG for
2614 lustre usage and make liblustre and some other places in generic
2615 lustre code to use it.
2617 Severity : enhancement
2618 Bugzilla : 9477, 9557, 9870
2619 Description: Verify that the MDS configuration logs are updated when xml is
2620 Details : Check if the .xml configuration logs are newer than the config
2621 logs stored on the MDS and report an error if this is the case.
2622 Request --write-conf, or allow starting with --old_conf.
2624 Severity : enhancement
2626 Description: Handle symlinks in the path when checking if Lustre is mounted.
2627 Details : Resolve intermediate symlinks when checking if a client has
2628 mounted a filesystem to avoid duplicate client mounts.
2633 Description: lconf can hit an error exception but still return success.
2634 Details : The lconf command catches the Command error exception at the top
2635 level script context and will exit with the associated exit
2636 status, but doesn't ensure that this exit status is non-zero.
2641 Description: failure of ptlrpc thread startup can cause oops
2642 Details : Starting a ptlrpc service thread can fail if there are a large
2643 number of threads or the server memory is very fragmented.
2644 Handle this without oopsing.
2647 Frequency : always, only if liblustre and non-default acceptor port was used
2649 Description: liblustre cannot connect to servers with non-default acceptor port
2650 Details : tcpnal_set_default_params() was not called and was therefore
2651 ignoring the environment varaible TCPNAL_PORT, as well as other
2652 TCPNAL_ environment variables
2657 Description: two objects could be created on the same OST for a single file
2658 Details : If an OST is down, in some cases it was possible to create two
2659 objects on a single OST for a single file. No problems other
2660 than potential performance impact and spurious error messages.
2664 Bugzilla : 5681, 9562
2665 Description: Client may oops in ll_unhash_aliases
2666 Details : Client dcache may become inconsistent in race condition.
2667 In some cases "getcwd" can fail if the current directory is
2673 Description: Inode refcounting problems in NFS export code
2674 Details : link_raw functions used to call d_instantiate without obtaining
2675 extra inode reference first.
2679 Bugzilla : 9942, 9903
2680 Description: Referencing freed requests leading to crash, memleaks with NFS.
2681 Details : We used to require that call to ll_revalidate_it was always
2682 followed by ll_lookup_it. Also with revalidate_special() it is
2683 possible to call ll_revalidate_it() twice for the same dentry
2684 even if first occurence returned success. This fix changes semantic
2685 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2686 reference on a request referred from the intent.
2687 ll_intent_release() then releases such a request.
2690 Frequency : rare, normally benchmark loads only
2692 Description: unlinked inodes were kept in memory on the client
2693 Details : If a client is repeatedly creating and unlinking files it
2694 can accumulate a lot of stale inodes in the inode slab cache.
2695 If there is no other client load running this can cause the
2696 client node to run out of memory. Instead flush old inodes
2697 from client cache that have the same inode number as a new inode.
2700 Frequency : SLES9 2.6.5 kernel and long filenames only
2701 Bugzilla : 9969, 10379
2702 Description: utime reports stale NFS file handle
2703 Details : SLES9 uses out-of-dentry names in some cases, which confused
2704 the lustre dentry revalidation. Change it to always use the
2708 Frequency : rare, unless heavy write-truncate concurrency is continuous
2709 Bugzilla : 4180, 6984, 7171, 9963, 9331
2710 Description: OST becomes very slow and/or deadlocked during object unlink
2711 Details : filter_destroy() was holding onto the parent directory lock
2712 while truncating+unlinking objects. For very large objects this
2713 may block other threads for a long time and slow overall OST
2714 responsiveness. It may also be possible to get a lock ordering
2715 deadlock in this case, or run out of journal credits because of
2716 the combined truncate+unlink. Solution is to do object truncate
2717 first in one transaction without parent lock, and then do the
2718 final unlink in a new transaction with the parent lock. This
2719 reduces the lock hold time dramatically.
2722 Frequency : rare, 2.4 kernels only
2724 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2725 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2726 before unmounting their filesystem in order to avoid deadlock.
2727 The kernel_locked() function in 2.4 kernels only checks whether
2728 the kernel lock is held, not whether it is this process that is
2729 holding it as 2.6 kernels do.
2734 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2735 Details : The client ptlrpc code may be trying to reconnect to a down
2736 server before a previous connection attempt has timed out.
2737 Increase the reconnect interval to be longer than the connection
2738 timeout interval to avoid sending duplicate connections to
2742 Frequency : echo_client brw_test command
2744 Description: fix echo_client to work with OST preallocated code
2745 Details : OST preallocation code (5137) didn't take echo_client IO path
2746 into account: echo_client calls filter methods outside of any
2747 OST thread and, hence, there is no per-thread preallocated
2748 pages and buffers to use. Solution: hijack pga pages for IO. As
2749 a byproduct, this avoids unnecessary data copying.
2753 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2754 Description: Client can oops in mdc_commit_close() after open replay
2755 Details : It was possible for the MDS to return an open request with no
2756 transaction number in mds_finish_transno() if the client was
2757 evicted, but without actually returning an error. Clients
2758 would later try to replay that open and may trip an assertion
2759 Simplify the client close codepath, and always return an error
2760 from the MDS in case the open is not successful.
2763 Frequency : rare, 2.6 OSTs only
2765 Description: OST may deadlock under high load on fragmented files
2766 Details : If there was a heavy load and highly-fragmented OST filesystems
2767 it was possible to have all the OST threads deadlock waiting on
2768 allocation of biovecs, because the biovecs were not released
2769 until the entire RPC IO was completed. Instead, release biovecs
2770 as soon as they are complete to ensure forward IO progress.
2772 Severity : enhancement
2774 Description: Support for specifying external journal device at mount
2775 Details : If an OST or MDS device is formatted with an external journal
2776 device, this device major/minor is stored in the ext3 superblock
2777 and may not be valid for failover. Allow detecting and
2778 specifying the external journal at mount time.
2783 Description: Mounting an MDS with pending unlinked files may cause oops
2784 Details : target_finish_recovery() calls mds_postrecov() which returned
2785 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2786 considers this an error and immediately begins cleaning up the
2787 lov, just after starting the mds_lov process
2789 Severity : enhancement
2791 Description: Implement 'lfs df' to report actual free space on per-OST basis
2792 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2793 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2794 usage of objects; '-h' to report in human readable format.
2796 ------------------------------------------------------------------------------
2798 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2805 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2806 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2807 at startup during orphan unlinking. Instead, skip buddy update
2808 before setup, as it will be regenerated after recovery is complete.
2813 Description: If an OST is inactive, its locks might reference stale inodes.
2814 Details : lov_change_cbdata() must iterate over all namespaces, even if
2815 they are inactive to clear inode references from the lock.
2817 Severity : enhancement
2818 Frequency : occasional, if non-standard max_dirty_mb used
2820 Description: Client will block write RPCs if not enough grant
2821 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2822 then the client will block writes while waiting for another RPC
2823 to complete instead of consuming its dirty limit. With change
2824 we get improved performance when max_dirty_mb is small.
2826 Severity : enhancement
2827 Bugzilla : 3389, 6253
2828 Description: Add support for supplementary groups on the MDS.
2829 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2830 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2831 lookups for user supplementary groups into a cache.
2835 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2836 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2837 for liblustre O_EXCL is enforced by MDS.
2841 Description: readdir never returns NULL in liblustre.
2842 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2843 dentry in struct dirent.
2847 Frequency : liblustre only, depends on application IO pattern
2848 Description: liblustre clients evicted if not contacting servers
2849 Details : Don't put liblustre clients into the ping_evictor list, so
2850 they will not be evicted by the pinger ever.
2852 Severity : enhancement
2854 Description: Add ability to evict clients by NID from MDS.
2855 Details : By echoing "nid:$NID" string into
2856 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2857 $NID would be instantly evicted from this MDS and from all active
2858 OSTs connected to it.
2862 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2863 Details : lookup_it_finish() used to query file size from OSTs that was not
2868 Description: service threads change working directory to that of init
2869 Details : Starting lustre service threads may pin the working directory
2870 of the parent thread, making that filesystem busy. Threads
2871 now change to the working directory of init to avoid this.
2875 Frequency : during shutdown only
2876 Description: shutdown with a failed MDS or OST can cause unmount to hang
2877 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2882 Frequency : 2.6 only
2883 Description: chmod/chown may include an extra supplementary group
2884 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2885 supplementary group and if none is specified this is used.
2888 Bugzilla : 5479 (6816)
2890 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2891 Details : If lookup is in progress on a file that is unlinked we might try
2892 to revalidate the inode and fail in revalidate after lookup is
2893 complete and ll_file_open() enqueues the open again but
2894 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2897 Frequency : always, if lconf --abort_recovery used
2899 Description: lconf --abort_recovery fails with 'Operation not supported'
2900 Details : lconf was attempting to abort recovery on the MDT device and not
2903 ------------------------------------------------------------------------------
2905 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2910 Frequency : rare (only unsupported configurations with a node running as an
2912 Bugzilla : 6514, 5137
2913 Description: Mounting a Lustre file system on a node running as an OST could
2915 Details : OSTs now preallocates memory needed to write out data at
2916 startup, instead of when needed, to avoid having to
2917 allocate memory in possibly low memory situations.
2918 Specifically, if the file system is mounted on on OST,
2919 memory pressure could force it to try to write out data,
2920 which it needed to allocate memory to do. Due to the low
2921 memory, it would be unable to do so and the node would
2922 become unresponsive.
2924 Severity : enhancement
2926 Description: Addition of lconf --service command line option
2927 Details : lconf now accepts a '--service <arg>' option, which is
2928 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2930 Severity : enhancement
2932 Description: Failover mode is now the default for OSTs.
2933 Details : By default, OSTs will now run in failover mode. To return to
2934 the old behaviour, add '--failout' to the lmc line for OSTs.
2936 Severity : enhancement
2938 Description: Health checks are now provided for MDS and OSTs
2939 Details : Additional detailed health check information on MSD and OSTs
2940 is now provided through the procfs health_check value.
2943 Frequency : occasional, depends on IO load
2945 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2946 after numerous create/delete cycles
2947 Details : The ext3 inode allocation policy would not allocate new inodes
2948 very well on the OSTs because there are no new directories
2949 being created. Instead we look for groups with free space if
2950 the parent directories are nearly full.
2955 Description: Network or server problems during mount may cause partially
2956 mounted clients instead of returning an error.
2957 Details : The config llog parsing code may overwrite the error return
2958 code during mount error handling, returning success instead
2964 Description: MDS can fail to allocate large reply buffers
2965 Details : After long uptimes the MDS can fail to allocate large reply
2966 buffers (e.g. zconf client mount config records) due to memory
2967 fragmentation or consumption by the buffer cache. Preallocate
2968 some large reply buffers so that these replies can be sent even
2969 under memory pressure.
2973 Frequency : rare (liblustre)
2974 Description: fsx running with liblustre complained that using truncate() to
2975 extend the file doesn't work. This patch corrects that issue.
2976 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2977 ATTR_SIZE and lsm use in llu_setattr_raw.
2981 Frequency : rare, only 2.6 kernels
2982 Description: Unusual file access patterns on the MDS may result in inode
2983 data being lost in very rare circumstances.
2984 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2985 code in the 2.6 kernel caused the inode and/or EA data not to
2986 be read from disk, causing single-file corruption.
2990 Frequency : rare, only 2.6 filesystems using extents
2991 Description: Heavy concurrent write and delete load may cause data corruption.
2992 Details : It was possible under high-load situations to have an extent
2993 metadata block in the block device cache from a just-unlinked
2994 file overwrite a newly-allocated data block. We now unmap any
2995 metadata buffers that alias just-allocated data blocks.
2999 Frequency : filesystems with default stripe_count larger than 77
3000 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3001 Details : lconf specifies an inode size of 4096 bytes when the default
3002 stripe_count is larger than 77. This conflicts with the default
3003 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3004 this case to avoid pinning too much memory for large EAs.
3006 ------------------------------------------------------------------------------
3008 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3013 Frequency : rare (extremely heavy IO load with hundreds of clients)
3015 Description: Client is evicted, gets IO error writing to file
3016 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3017 caused clients to be evicted for AST timeouts. The fixes in
3018 bug 5192 mean we no longer need to have such short AST timeouts
3019 so ldlm_timeout has been increased.
3022 Frequency : occasional during --force or --failover shutdown under load
3023 Bugzilla : 5949, 4834
3024 Description: Server oops/LBUG if stopped with --force or --failover under load
3025 Details : a collection of import/export refcount and cleanup ordering
3026 issues fixed for safer force cleanup
3029 Frequency : only filesystems larger than 120 OSTs
3030 Bugzilla : 5990, 6223
3031 Description: lfs getstripe would oops on a very large filesystem
3032 Details : lov_getconfig used kfree on vmalloc'd memory
3035 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3036 Bugzilla : 6242, 6243
3037 Description: reading from files that had been truncated to a non-zero size
3038 but never opened returned no data
3039 Details : ll_file_read() reads zeros from no-object files to EOF
3044 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3045 Details : There's little harm in aborting MDS/OSS recovery and letting it
3046 try again, so I removed the LASSERT and return an error instead.
3048 Severity : enhancement
3050 Description: New debugging infrastructure for tracking down data corruption
3051 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3052 (b) cover more of the client-side code path, and (c) try to narrow
3053 down where problems occurred
3057 Bugzilla : 3819, 4364, 4397, 6313
3058 Description: Racing close and eviction MDS could cause assertion in mds_close
3059 Details : It was possible to get multiple mfd references during close and
3060 client eviction, leading to one thread referencing a freed mfd.
3062 Severity: : enhancement
3063 Bugzilla : 3262, 6359
3064 Description: Attempts to reconnect to servers are now more aggressive.
3065 Details : This builds on the enhanced upcall-less recovery that was added
3066 in 1.4.2. When trying to reconnect to servers, clients will
3067 now try each server in the failover group every 10 seconds. By
3068 default, clients would previously try one server every 25 seconds.
3073 Description: After recovery, certain operations trigger a failed
3074 assertion on a client.
3075 Details : Failing over an mds, using lconf -d --failover, while a
3076 client was doing a readdir() call would cause the client to
3077 LBUG after recovery completed and the readdir() was resent.
3079 Severity : enhancement
3081 Description: Default groups are now added by lconf
3082 Details : You can now run lconf --group <servicename> without having to
3083 manually add groups with lmc.
3086 Frequency : occasional
3088 Description: Nodes with an elan id of 0 trigger a failed assertion
3091 Frequency : always when accessing e.g. tty/console device nodes
3093 Description: tty and some other devices nodes cannot be used on lustre
3094 Details : file's private_data field is used by device data and lustre
3095 values in there got lost. New field was added to struct file to
3096 store fs-specific private data.
3099 Frequency : when exporting Lustre via NFS
3101 Description: NFSD failed occasionally when looking up a path component
3102 Details : NFSD is looking up ".." which was broken in ext3 directories
3103 that had grown large enough to become hashed.
3106 Frequency : Clusters with multiple interfaces not on the same subnet
3108 Description: Nodes will repeatedly try to reconnect to an interface which it
3109 cannot reach and report an error to the log.
3110 Details : Extra peer list entries will be created by lconf with some peers
3111 unreachable. lconf now validates the peer before adding it.
3114 Frequency : Only if a default stripe is set on the filesystem root.
3116 Description: Setting a default stripe on the filesystem root prevented the
3117 filesystem from being remounted.
3118 Details : The client was sending extra request flags in the root getattr
3119 request and did not allocate a reply buffer for the dir EA.
3122 Frequency : occasional, higher if lots of files are accessed by one client
3123 Bugzilla : 6159, 6097
3124 Description: Client trips assertion regarding lsm mismatch/magic
3125 Details : While revalidating inodes the VFS looks up inodes with ifind()
3126 and in rare cases can find an inode that is being freed.
3127 The ll_test_inode() code will free the lsm during ifind()
3128 when it finds an existing inode and then the VFS later attaches
3129 this free lsm to a new inode.
3133 Bugzilla : 6422, 7030
3134 Description: MDS deadlock between mkdir and client eviction
3135 Details : Creating a new file via mkdir or mknod (starting a transaction
3136 and getting the ns lock) can deadlock with client eviction
3137 (gets ns lock and trying to finish a synchronous transaction).
3140 Frequency : occasional
3141 Description: While starting a server, the fsfilt_ext3 module could not be
3143 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3144 kernels. Previously, lconf would still use the ext3 name
3145 when trying to load modules. Now, it will correctly use
3146 ext3 on 2.4 and ldiskfs on 2.6.
3148 Severity : enhancement
3149 Description: The default stripe count has been changed to 1
3150 Details : The interpretation of the default stripe count (0, to lfs
3151 or lmc) has been changed to mean striping across a single
3152 OST, rather than all available. For general usage we have
3153 found a stripe count of 1 or 2 works best.
3155 Severity : enhancement
3156 Description: Add support for compiling against Cray portals.
3157 Details : Conditional compiling for some areas that are different
3161 Frequency : occasional
3162 Bugzilla : 6409, 6834
3163 Description: Creating files with an explicit stripe count may lead to
3164 a failed assertion on the MDS
3165 Details : If some OSTs are full or unavailable, creating files may
3166 trigger a failed assertion on the MDS. Now, Lustre will
3167 try to use other servers or return an error to the
3171 Frequency : occasional
3173 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3174 caused lock timeout during readahead (since 1.4.2).
3175 Details : Processes doing readahead might match a lock that hasn't been
3176 granted yet if there are overlapping and conflicting lock
3177 requests. The readahead process waits on ungranted lock
3178 (original lock is CBPENDING), while OST waits for that process
3179 to cancel CBPENDING read lock and eventually evicts client.
3181 Severity : enhancement
3183 Description: Initial enabling of flock support for clients
3184 Details : Implements fcntl advisory locking and file status functions.
3185 This feature is provided as an optional mount flag (default
3186 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3187 locking are implemented yet, and those that are are not guaranteed
3188 to be completely correct in production environments.
3189 mount -t lustre -o [flock|noflock] ...
3192 Frequency : occasional
3194 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3195 assertion in the ext3 JBD (journaling) layer.
3196 Details : The b_committed_data struct is protected by the big kernel lock
3197 in 2.4 kernels, serializing journal_commit_transaction() and
3198 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3199 access to this struct. In 2.6 kernels there is finer grained
3200 locking to improve SMP performance of the JBD layer.
3204 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3209 Description: merge and cleanup kernel patches.
3212 -----------------------------------------------------------------------------
3214 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3216 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3217 versions of Lustre. Please update all servers and clients to
3218 version 1.4.2 or later at the same time. You must also run
3219 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3220 to update the configuration logs.
3222 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3223 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3224 - swab llog records of type '0' so we get proper header size/idx (5861)
3225 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3226 - fix rename of one directory over another leaking an inode (5953)
3227 - avoid SetPageDirty on 2.6 (5981)
3228 - don't re-add just-being-destroyed locks to the waiting list (5653)
3229 - when creating new directories, inherit the parent's custom
3230 striping settings if present parent (3048)
3231 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3232 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3233 - don't hold client locks on temporary worklist from l_lru (5666)
3234 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3235 - automatically evict dead clients (5921)
3236 - Update file size properly in create+truncate+fstat case (6196)
3237 - Do not unhash mountpoint dentries, do not allow removal of
3239 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3240 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3241 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3242 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3243 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3244 - Correctly handle reads of files with no objects (6243)
3245 - lctl recover will also mark a device active if deactivate used (5933)
3247 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3248 - allow --write-conf on an MDS with different nettype than client (5619)
3249 - don't write config llogs to MDS for mounts not from that MDS (5617)
3250 - lconf should create multiple TCP connections from a client (5201)
3251 - init scripts are now turned off by default; run chkconfig --on
3252 lustre and chkconfig --on lustrefs to use them
3253 - upcalls are no longer needed for clients to recover to failover
3255 - add --abort-recovery option to lconf to abort recovery on device
3257 - add support for an arbitrary number of OSTs (3026)
3258 - Quota support protocol changes.
3259 - forward compatibility changes to wire structs (6007)
3260 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3261 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3262 - improved "lustre status" script
3263 - initialize blocksize for non-regular files (6062)
3264 - added --disable-server and --disable-client configure options (5782)
3265 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3266 - Vanilla 2.4.29 support
3267 - increase maximum number of obd devices to 520 (6242)
3268 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3269 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3270 SLES 9 2.6.5 kernels
3271 - limit stripes per file to 160 (the maximum EA size) (6093)
3273 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3276 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3277 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3278 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3279 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3280 - small ext3 extents cleanups and fixes (5733)
3281 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3282 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3283 - avoid needless client->OST connect, fix handle mismatch (5317)
3284 - fix DLM error path that led to out-of-sync client, long delays (5779)
3285 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3286 - fix several locking issues related to i_size (5492,5624,5654,5672)
3287 - don't move pending lock onto export if it is already evicted (5683)
3288 - fix kernel oops when creating .foo in unlinked directory (5548)
3289 - fix deadlock in obdfilter statistics vs. object create (5811)
3290 - use time_{before,after} to avoid timer jiffies wrap (5882)
3291 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3292 - Do not leak request if server was not able to process it (5154)
3293 - If mds_open unable to find parent dir, make that negative lookup(5154)
3294 - don't create new directories with extent-mapping (5909, 5936)
3296 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3297 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3298 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3299 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3300 - support for RHEL4 kernel on i686 (5773)
3301 - provide error messages when incompatible logs are encountered (5898)
3303 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3304 * version 1.4.0.10 (1.4.1 release candidate 1)
3306 - don't keep a lock reference when lock is not granted (4238)
3307 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3308 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3309 - drop import inflight refcount on signal_completed_replay error (5255)
3310 - unlock page after async write error during send (3677)
3311 - handle missing objects in filter_preprw_read properly (5265)
3312 - no transno return for symlink open, don't save no-trasno open (3440)
3313 - don't try to complete elan receive that already failed (4012)
3314 - free RPC server reply state on error (5406)
3315 - clean up thread from ptlrpc_start_thread() on error (5160)
3316 - readahead could read extra page into cache that wasn't ejected (5388)
3317 - prevent races in class_attach/setup/cleanup/detach (5260)
3318 - don't dereference de->d_inode after l_dput of de (5458)
3319 - use "int" for stripe value returned from lock_to_stripe (5544)
3320 - mballoc allocation and error-checking fixes in 2.6 (5504)
3321 - block device patches to fix I/O request sizes in 2.6 (5482)
3322 - look up hostnames for IB nals (5602)
3323 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3324 - don't start multiple acceptors for the same port (5277)
3325 - fix incorrect LASSERT in mds_getattr_name (5635)
3326 - export a proc file for general "ping" checking (5628)
3327 - fix "lfs check" to not block when the MDS is down (5628)
3329 - service request history (4965)
3330 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3331 - create an "evict_client" /proc entry on OSTs, like the MDS has
3332 - fix mount usage message, return errors per mount(8) (5168)
3333 - change grep [] to grep "[]" in tests so they work in more UMLs
3334 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3335 - remove ancient LOV_MAGIC_V0 EA support (5047)
3336 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3337 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3338 - allow readahead allocations to fail when low on memory (5383)
3339 - mmap locking landed again, after considerable improvement (2828)
3340 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3342 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3345 - send OST transaction number in read/write reply to free req (4966)
3346 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3347 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3348 - the watchdog thread now runs as interruptible (5246)
3349 - flock/lockf fixes (but it's still disabled, pending 5135)
3350 - don't use EXT3 constants in llite code (5094)
3351 - memory shortage at startup could cause assertion (5176)
3353 - reorganization of lov code
3354 - single portals codebase
3356 - add extents/mballoc support (5025)
3357 - direct I/O reads in the obdfilter (4048)
3358 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3360 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3363 - send OST transaction number in read/write reply to free req (4966)
3364 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3365 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3366 - the watchdog thread now runs as interruptible (5246)
3367 - handle missing objects in filter_preprw_read properly (5265)
3368 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3369 - drop import inflight refcount on signal_completed_replay error (5255)
3370 - unlock page after async write error during send (3677)
3371 - return original error code on reconstructed replies (3761)
3372 - no transno return for symlink open, don't save no-trasno open (3440)
3374 - add pid to ldlm debugging output (4922)
3375 - bump the watchdog timeouts -- we can't handle 30sec yet
3376 - extra debugging for orphan dentry/inode bug (5259)
3378 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3381 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3382 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3383 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3384 - fix NULL dereference in /proc/sys/portals/routes (4827)
3385 - allow failed mdc_close() operations to be interrupted (4561)
3386 - stop precreate on OST before MDS would time out on it (4778)
3387 - don't send partial-page writes before EOF from client (4410)
3388 - discard client grant for sub-page writes on large-page clients (4520)
3389 - don't free dentries not owned by NFS code, check generation (4806)
3390 - fix lsm leak if mds_create_objects() fails (4801)
3391 - limit debug_daemon file size, always print CERROR messages (4789)
3392 - use transno after validating reply (3892)
3393 - process timed out requests if import state changes (3754)
3394 - update mtime on OST during writes, return in glimpse (4829)
3395 - add mkfsoptions to LDAP (4679)
3396 - use ->max_readahead method instead of zapping global ra (5039)
3397 - don't interrupt __l_wait_event() during strace
3399 - add software watchdogs to catch hung threads quickly (4941)
3400 - make lustrefs init script start after nfs is mounted
3401 - fix CWARN/ERROR duplication (4930)
3402 - return async write errors to application if possible (2248)
3403 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3404 - print NAL number in %x format (4645)
3405 - update barely-supported suse-2.4.21-171 series (4842)
3406 - support for sles 9 %post scripts
3407 - support for building 2.6 kernel-source packages
3408 - support for sles km_* packages
3410 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3413 - ignore -ENOENT errors in osc_destroy (3639)
3414 - notify osc create thread that OSC is being cleaned up (4600)
3415 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3416 - reconstruct ost_handle() like mds_handle() (4657)
3417 - create a new thread to do import eviction to avoid deadlock (3969)
3418 - let lconf resolve symlinked-to devices (4629)
3419 - don't unlink "objects" from directory with default EA (4554)
3420 - hold socknal file ref over connect in case target is down (4394)
3421 - allow more than 32000 subdirectories in a single directory (3244)
3422 - fix blocks count for O_DIRECT writes (3751)
3423 - OST returns ENOSPC from object create when no space left (4539)
3424 - don't send truncate RPC if file size isn't changing (4410)
3425 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3426 - bind to privileged port in socknal and tcpnal (3689)
3428 - rate limit CERROR/CWARN console message to avoid overload (4519)
3429 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3430 - basic mmap support (3918)
3431 - kernel patch series update from b1_4 (4711)
3433 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3436 - avoid crash during MDS cleanup with OST shut down (2775)
3437 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3438 - don't use bad inodes on the MDS (3744)
3439 - dynamic object preallocation to improve recovery speed (4236)
3440 - don't hold spinlock over lock dumping or change debug flags (4401)
3441 - don't zero obd_dev when it is force cleaned (3651)
3442 - print grants to console if they go negative (4431)
3443 - "lctl deactivate" will stop automatic recovery attempts (3406)
3444 - look for existing locks in ldlm_handle_enqueue() (3764)
3445 - don't resolve lock handle twice in recovery avoiding race (4401)
3446 - revalidate should check working dir is a directory (4134)
3448 - don't always mark "slow" obdfilter messages as errors (4418)
3450 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3453 - don't close LustreDB during write_conf until it is done (3860)
3454 - fix typo in lconf for_each_profile (3821)
3455 - allow dumping logs from multiple threads at one time (3820)
3456 - don't allow multiple threads in OSC recovery (3812)
3457 - fix debug_size parameters (3864)
3458 - fix mds_postrecov to initialize import for llog ctxt (3121)
3459 - replace config semaphore with spinlock (3306)
3460 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3461 - don't allow enqueue to complete on a destroyed export (3822)
3462 - down write_lock before checking llog header bitmap (3825)
3463 - recover from lock replay timeout (3764)
3464 - up llog sem before sending rpc (3652)
3465 - reduce ns lock hold times when setting kms (3267)
3466 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3467 - fix NULL deref and obd_dev leak on setup error (3312)
3468 - replace some LBUG about llog ops with error handling (3841)
3469 - don't match INVALID dentries from d_lookup and spin (3784)
3470 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3471 - fix invalid assertion in ptlrpc_set_wait (3880)
3473 - add libwrap support for the TCP acceptor (3996)
3474 - add /proc/sys/portals/routes for non-root route listing (3994)
3475 - allow setting MDS UUID in .xml (2580)
3476 - print the stack of a process that LBUGs (4228)
3478 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3481 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3482 - make sure to unset replay flag from failed open requests (3440)
3483 - if default stripe count is 0, use OST count for inode size (3636)
3484 - update parent mtime/ctime on client for create/unlink (2611)
3485 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3486 - free recovery state on server during a forced cleanup (3571)
3487 - unregister_reply for resent reqs (3063)
3488 - loop back devices mounting and status check on 2.6 (3563)
3489 - fix resource-creation race that can provoke i_size == 0 (3513)
3490 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3491 - more debugging for page-accounting assertion (3746)
3492 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3493 - avoid deadlock after precreation failure (3758)
3494 - fix race and lock order deadlock in orphan handling (3450, 3750)
3495 - add validity checks when grabbing inodes from l_ast_data (3599)
3497 - add /proc/.../recovery_status to obdfilter (3428)
3498 - lightweight CDEBUG infrastructure, debug daemon (3668)
3499 - change default OSC RPC parameters to be better on small clusters
3500 - turn off OST read cache for files smaller than 32MB
3501 - install man pages and include them in rpms (3100)
3502 - add new init script for (un)mounting lustre filesystems (2593)
3503 - run chkconfig in %post for init scripts (3701)
3504 - drop scimac NAL (unmaintained)
3506 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3509 - clean kiobufs before and after use (3485)
3510 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3511 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3512 - change init script to fail more clearly if not run as root (1528)
3513 - allow clients to reconnect during replay (1742)
3514 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3515 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3516 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3517 - fix cond_resched (3554)
3518 - only evict unfinished clients after recovery (3515)
3519 - allow bulk resend, prevent data loss (3570)
3520 - dynamic ptlrpc request buffer allocation (2102)
3521 - don't allow unlinking open directory if it isn't empty (2904)
3522 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3523 - remove extraneous obd dereference causing LASSERT failure (3334)
3524 - don't use get_cycles() when creating temp. files on the mds (3156)
3525 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3526 - handle EEXIST for set-stripe, set proper directory name (3336)
3528 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3529 - fix ksocknal_fmb_callback() error messages (2918)
3531 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3534 - don't copy lvb into (possibly NULL) reply on error (2983)
3535 - don't deref dentry after dput, don't free lvb on error (2922)
3536 - use the kms to determine writeback rpc length (2947)
3537 - increment oti_logcookies when osc is inactive (2948)
3538 - update client's i_blocks count via lvb messages (2543)
3539 - handle intent open/close of special files properly (1557)
3540 - mount MDS with errors=remount-ro, like obdfilter (2009)
3541 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3542 - don't use cancelling-locks' kms values (2947)
3543 - use highest lock extent for kms, not last one (2925)
3544 - don't dereference ERR_PTR() dentry in error handling path (3107)
3545 - fix thread race in portals_debug_dumplog() (3122)
3546 - create lprocfs device entries at setup instead of at attach (1519)
3547 - common AST error handler, don't evict client on completion race (3145)
3548 - zero nameidata in detach_mnt in 2.6 (3118)
3549 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3550 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3551 - initialize RPC timeout timer earlier for 2.6 (3219)
3552 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3553 - print nal/nid for unknown nid (3258)
3554 - additional checks for oscc recovery before doing precreate (3284)
3555 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3556 - don't crash in mdc_close for bad permissions on open (3285)
3557 - zero i_rdev for non-device files (3147)
3558 - clear page->private before handing to FS, better assertion (3119)
3559 - tune the read pipeline (3236)
3560 - fix incorrect decref of invalidated dentry (2350)
3561 - provide read-ahead stats and refine rpc in flight stats (3328)
3562 - don't hold journal transaction open across create RPC (3313)
3563 - update atime on MDS at close time (3265)
3564 - close LDAP connection when recovering to avoid server load (3315)
3565 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3566 - don't leak open file on MDS after open resend (3325)
3567 - serialize filter_precreate and filter_destroy_precreated (3329)
3568 - loop device shouldn't call sync_dev() for nul device (3092)
3569 - clear page cache after eviction (2766)
3570 - resynchronize MDS->OST in background (2824)
3571 - refuse to mount the same filesystem twice on same mountpoint (3394)
3572 - allow llmount to create routes for mounting behind routers (3320)
3573 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3574 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3575 - fix rare problem with rename on htree directories (3417)
3577 - allow default OST striping configuration per directory (1414)
3578 - fix compilation for qswnal for 2.6 kernels (3125)
3579 - increase maximum number of MDS request buffers for large systems
3580 - change liblustreapi to be useful for external progs like lfsck (3098)
3581 - increase local configuration timeout for slow disks (3353)
3582 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3584 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3587 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3588 - don't overwrite extent policy data in reply if lock was blocked (2901)
3589 - drop filter export grants atomically with removal from device (2663)
3590 - del obd_self_export from work_list in class_disconnect_exports (2908)
3591 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3592 - swab reply message in mdc_close, other PPC fixes (2464)
3593 - fix destroying of named logs (2325)
3594 - overwrite old logs when running lconf --write_conf (2264)
3595 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3596 - fix race in target_handle_connect (2898)
3597 - mds_reint_create() should take same inode create lock (2926)
3598 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3599 - don't close files for self_export to avoid uninitialized obd (2936)
3600 - allow MDS with the same name as client node (2939)
3601 - hold dentry reference for closed log files for unlink (2325)
3602 - reserve space for all logs during transactions (2059)
3603 - don't evict page beyond end of stripe extent (2925)
3604 - don't oops on a deleted current working directory (2399)
3605 - handle hard links to targets without a parent properly (2517)
3606 - don't dereference NULL lock when racing during eviction (2867)
3607 - don't grow lock extents when lots of conflicting locks (2919)
3609 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3612 - account for cache space usage on clients to avoid data loss (974)
3613 - lfsck support in lustre kernel code (2349)
3614 - reduce journal credits needed for BRW writes (2370)
3615 - orphan handling to avoid losing space on client/server crashes
3616 - ptlrpcd can be blocked, stopping ALL progress (2477)
3617 - use lock value blocks to assist in proper KMS, faster stat (1021)
3618 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3619 - recovery for initial connections (2355)
3620 - fixes for mds_cleanup_orphans (1934)
3621 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3622 - block all file creations until orphan recovery completes (1901)
3623 - client remove rq_connection from request struct (2423)
3624 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3625 - recovery timer race (2670)
3626 - mdc_close recovey bug (2532)
3627 - ptlrpc cleanup bug (2710)
3628 - mds timeout on local locks (2588)
3629 - namespace lock held during RPCs (2431)
3630 - handle interrupted sync write properly (2503)
3631 - don't try to handle a message that hasn't been replied to (2699)
3632 - client assert failure during cleanup after abort recovery (2701)
3633 - leak mdc device after failed mount (2712)
3634 - ptlrpc_check_set allows timedout requests to complete (2714)
3635 - wait for inflight reqs when ptlrpcd finishes (2710)
3636 - make sure unregistered services are removed from the srv_list
3637 - reset bulk XID's when resending them (caught by 1138 test)
3638 - unregister_bulk after timeout
3639 - fix lconf error (2694)
3640 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3641 - readahead locks pages, leaves pending causing memory pressure (2673)
3642 - increase OST request buffers to 4096 on large machines (2729)
3643 - fix up permission of existing directories in simple_mkdir (2661)
3644 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3645 - don't assign transno to errored transactions (2742)
3646 - don't delete objects on OST if given a bogus objid from MDS (2751)
3647 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3648 - if rq_no_resend, then timeout request after recovery (2432)
3649 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3650 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3651 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3652 - MDS recovery shouldn't delete objects that it has given out (2730)
3653 - if enqueue arrives after completion, don't clobber LVB (2819)
3654 - don't unlock pages twice when trigger_group_io returns error (2814)
3655 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3656 - don't write pages to disk if there was an error (1450)
3657 - don't ping imports that have recovery disabled (2676)
3658 - take buffered bytes into account when balancing socknal conn (2817)
3659 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3660 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3661 - remove little-endian swabbing of llog records (1987)
3662 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3663 - retry reposting request buffers if they fail (1191)
3664 - grow extent at grant time to avoid granting a revoked lock (2809)
3665 - lock revoke doesn't evict page if covered by a second lock (2765)
3666 - disable VM readahead to avoid reading outside lock extents (2805)
3668 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3669 - updated kernel patches for hp-2.4.20 kernel (2681)
3671 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3674 - fix truncated write corruption (2366)
3675 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3677 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3678 - ensure len(uuid) < 37 in lmc (1171)
3679 - fix ia64 OOPS in llog_test (2255)
3680 - zero end of page at obdfilter for partial page writes (2648)
3681 - don't leave stale dentries around after renames (bug 2428)
3682 - fix timeouts when evicting a client with a single lock held (2642)
3683 - set deadline for the initial HELLO message to drain (2634)
3684 - print out dotted-quad IP addresses in the socknal (2302)
3686 - additional debugging for MDS client eviction problem (2443)
3687 - fix mkfsoptions support for osts (2603, 2604)
3689 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3692 - add series for the vanilla 2.6.0 kernel
3693 - add series for the vanilla 2.4.24 kernel
3694 - add series for a cray x86/64 UL kernel drop
3695 - fix xattr patches for the vanilla 2.4.19 series
3697 - generate true UUIDs in lmc (1171)
3698 - have portals stack dumping break in UML (2466)
3699 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3700 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3701 - wake callback waiting threads on client eviction (2460)
3702 - Add --ptldebug and --subsystem to lmc (1719)
3703 - update assertion to allow safe interrupt allocation
3704 - set rq_no_resend for cancel requests (2432)
3705 - recalculate ptlrpcd timeout after resend (2494)
3706 - call vfs_rmdir when removing pending directories (2368)
3707 - fix renaming a file to itself (2429)
3708 - lmc creates a default one-stripe lov (2454)
3709 - expand procfs space to handle large clusters (2326)
3710 - increase UML stack to avoid overflow
3711 - update lconf's list of debug and subsystem masks
3712 - fix lfs find --obd (2510)
3713 - /proc tunable for disabling filter read caching (2591)
3714 - stop rpm packages from altering slapd.conf (2301)
3715 - disable nagle in the socknal under 0conf (2578)
3716 - choose mds inode size based on stripe count (2572)
3717 - fix kernel-source rpm problems (2516)
3719 - add --disable-doc to avoid pdf generation (2421)
3720 - update documentation, tests, type-os, comments
3721 - avoid format warnings on ia64
3722 - remove the TOE NAL
3723 - tiny code cleanups by removing unused fields
3725 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3728 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3729 - remove the most blatant lies from BUILDING file (2371)
3730 - change default debug level to reasonable production setting
3731 - reduce client side cache size to reduce cache flush time
3732 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3733 - make TCP zerocopy and pinger support enabled by default (2476)
3734 - sync writes completed after process exits caused crashes (2319)
3735 - maintain correct mount count on the MDS (2356)
3736 - backout 1557, because 2316 wasn't really fixed
3737 - better file I/O statistics gathering in /proc
3738 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3739 - another kernel patch to fix zero-copy TCP function export
3740 - don't take duplicate lock when processing re-sent getattr (2420)
3741 - lctl uses obd_self_export instead of creating new conn (2353)
3742 - MDS/OST recovery case which requires object creation asserted (2425)
3743 - move lfs from /usr/sbin to /usr/bin in packages
3744 - fix race between mds_client_add and mds_client_free (2417)
3745 - use kmalloc instead of slabs in portals (2430)
3746 - don't create duplicate records when a failover MDS is present (2442)
3747 - remove unnecessary mount age check (2332)
3748 - don't remove directory inodes from locks prematurely (2451)
3749 - don't break if MDS service name is the same as hostname (2103)
3750 - fix races in client write RPC generation when cache full (2482)
3752 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3755 - remove now-unused request->rq_obd (278)
3756 - if an allocation fails, print out how much memory we've used (1933)
3757 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3758 - add the "configurable stack size" patch to most series files (1256)
3759 - ability to write large log records, for 100+ OST configs (2306)
3760 - fix NULL deref when filter_prep fails (2314)
3761 - fix operator precedence error in filter_sync
3762 - dynamic allocation of socknal TX descriptors (2315)
3763 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3764 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3765 - ensure that CWARN messages in Portals always get to the syslog
3766 - __init/__exit are not for prototype decls (ldlm_init/exit)
3767 - x86-64 compile warning fixes
3768 - fix gateway LMC keyword conflict (2318)
3769 - fix MDS lock inversions in getattr/reint paths (1844)
3770 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3771 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3772 - DLM race condition prevented some lock evictions (2328)
3773 - ENOMEM detection and retry on socknal sends (2230)
3774 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3775 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3776 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3777 - better error messages when a client is rejected during recovery (1505)
3778 - avoid cancelling locks which were never granted, after failure (2330)
3779 - fix i_sem/journal inversion in mds_client_add (2333)
3780 - fix truncate/getattr lock cycle deadlock (2334)
3781 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3782 - fix two rare exit paths which could leak an l_lock() ref (2321)
3783 - fencepost error in MDS/OST orphan recovery (2226)
3784 - make log record alignment 8 bytes (1988)
3785 - lstripe now fails when requested offset > ost_count (2237)
3786 - ensure that all kernel series have a complete list.h (1607)
3787 - fix crashes in special-file operations (2316)
3788 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3790 - allow configurable automake binary, for testing new versions
3791 - small update to the lfs documentation
3793 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3795 * fix negative export reference count in fsfilt_sync (2312)
3797 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3798 * release candidate 0.9.1
3800 - orphans are moved into the PENDING directory for possible recovery
3801 - replayed opens now open by fid for orphan/rename safety (1042)
3802 - last close of an orphan inode generates a transno (683)
3803 - chdir() and mount() now pin the directory entry (1020)
3804 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3805 - discard very old requests without processing them (1502)
3806 - remove some common, well-understood CERRORs (1505)
3807 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3808 - clear "grant" flags in OST replies until OST grant code lands (1644)
3809 - fix read performance by not clobbering i_blksize on client (1598)
3810 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3811 - make LRU size a /proc tunable, clears locks when reduced (707)
3812 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3813 - server locks take references on exports now (1558)
3814 - build fixes for 2.4.20-rh trees (1663)
3815 - return an error from lov_create if all OSCs are inactive (1751)
3816 - fix import levels when a reconnect happens without a timeout (1597)
3817 - exit early from mds_open if we get a lookup error (1749)
3818 - partial page read at EOF wouldn't wait for disk before sending (1642)
3819 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3820 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3821 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3822 - write/truncate lock inversion (1639)
3823 - Don't auto-load obdclass, portals modules during cleanup (1495)
3824 - fix timestamps from jumping to "now" (1763)
3825 - extra journal assertions (1648)
3826 - add an extra multiunlink test (1771)
3827 - fix read_record/write_record API (1776)
3828 - fix leak of offset_extent, possible incorrect i_size later (1772)
3829 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3830 - Debugging for the kqswnal_get_idle_tx problems (1820)
3831 - Allow recovery to be attempted multiple times (1536)
3832 - Write out MDS last_rcvd file after it is first created (1600)
3833 - Fix tx_descriptor leak in failed transmit situations (1827)
3834 - ext3 journaling fixes for assertion failure after IO error (1871)
3835 - class_export_put() on freed export after completion AST error (1896)
3836 - Fix revalidate looping in VFS (1322)
3837 - Don't access a freed export during MDS_REINT timeout (1521)
3838 - Add open-unlink recovery support on the MDS (1673,1764)
3839 - Return an error if no MDS data was read from last_rcvd (1946)
3840 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3841 - Don't setup a disk device that doesn't match exported UUID (317)
3842 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3843 - avoid committing NULL handle in force close
3844 - local.sh is now a one-stripe LOV configuration
3845 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3846 - fix ext3 htree duplicate directory entry corruption (1516)
3847 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3848 - update to new LOV EA format (2097)
3849 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3851 2003-06-15 Phil Schwan <phil@clusterfs.com>
3854 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3855 - per-import/export recovery handling (958, 931, 959)
3856 - multiple last-rcvd slots, for serving multiple FSes (949)
3857 - connections are again shared between multiple imp/exports (963, 964)
3858 - "umount -f" would hang if any requests needed to be sent (393, 978)
3859 - avoid pinning large req buffer by copying for queued messages (989)
3860 - add "uuid" to "lctl device" command to help upcalls (991)
3861 - "open" RPCs with transnos would confuse recovery counters (1037)
3862 - do proper endian conversion of all wire messages (288, 340, 891)
3863 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3864 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3865 - fix ptlrpc_connection leak in target_handle_connect (1174)
3866 - fix import refcounting bug in OST and MDS cleanup (1134)
3867 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3868 - fix very unlikely obd_types race condition (501)
3869 - remove osc_open hack for echo_client (1187)
3870 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3871 - a failure in read_inode2 leads to deadlock (1139)
3872 - cancel ack-locks as soon as transaction is committed (1072)
3873 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3874 - make sure to commitrw after any preprw to avoid deadlock (1162)
3875 - failing to execute a file in a lustre FS would lock inode (1203)
3876 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3877 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3878 - obd_uuid2tgt would walk off the end of the list (1255)
3879 - on IA64 the osc would give portals incorrect bulk size (1258)
3880 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3881 - fix lock inversion caused by new llite matching code (1282)
3882 - limit the number of dirty pages on a client to 10MB (1286)
3883 - timed out locks were not being corrected cancelled (1289)
3884 - fix O_DIRECT above 4GB on IA-32 (1292)
3885 * major user-visible changes
3886 - fail out/fail over policy now controlled by the upcall (993)
3888 - add OBD_PING to check server availability and failure (954)
3889 - lustre messages are now sent in sending host order (288, 340, 891)
3890 - add eadatalen to MDS getattr reply (340)
3891 - OST read replies may contain second buffer, with per-page status (593)
3893 2003-03-11 Phil Schwan <phil@clusterfs.com>
3896 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3897 - failed open()s could cause deadlock; fixed (867, 869)
3898 - stop cancelling OST locks when files are closed (481)
3899 - overlapping XID spaces caused network corruption (851, 853)
3900 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3901 - setattr_raw added, to do single-RPC, server-side setattrs
3902 - lmc/lconf syntax change for OST UUIDs
3903 - fix crashy race condition between ptlrpc_free_req and osc_close
3904 - don't use request in mdc_enqueue if we hit a timeout (889)
3905 - don't set the inode i_size for regular files from the MDS (896)
3906 - handle out of order completion AST (842)
3907 - don't LBUG if a lock request times out after receiving AST (913)
3908 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3909 - if a bad lock AST arrives, send an error instead of dropping entirely
3910 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3911 - fix leak in bulk IO when only partially completed (899, 900, 926)
3912 - fix O_DIRECT for ia64 (55)
3913 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3914 - C-z after timeout could hang a process forever; fixed (977)
3916 - client-side I/O cache (678, 924, 929, 941, 970)
3918 - READPAGE and SETATTRs which don't take server-side locks get
3921 2003-02-11 Phil Schwan <phil@clusterfs.com>
3924 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3925 - Fix fsfilt_extN_readpage() to read a full page of directory
3926 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3927 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3928 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3929 - Most intent operations were removed, in favour of a new RPC mode
3930 that does a single RPC to the server and bypasses most of the VFS
3931 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3932 - Aggressively cancel local locks on DLM servers
3933 - mds_reint_unlink sends EA to the client if it's the last nlink.
3934 client uses that EA to unlink OST objects.
3935 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3936 - recursive symlinks were fixed (439)
3937 - fixed NULL deref in DEBUG_REQ
3938 - filter_update_lastobjid no longer calls sync, which annoyed extN
3939 - fixed multi-client small-writes to a single file problem (445)
3940 - fixed mtime updates during file writes (607)
3941 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3942 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3943 - fixed Linux OST size reporting problem (444, 656)
3944 - OST now updates object mtime with writes or setattr (607, 619)
3945 - client verifies file size before zeroing page past EOF (445)
3946 - OST now writes last allocated objid to disk with allocation (108)
3947 - LOV on echo now works (409)
3949 - mds_reint_unlink sends a new buffer, with the EA included. this
3950 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3951 set if a regular file was being unlinked, and it was the last link
3952 - use PtlGet from the target for bulk writes (315)
3953 - OST now updates object mtime with writes or setattr (607, 619)
3954 - LDLM now has a grant-time callback to revalidate locked items, if
3956 - Many MDS operations were reorganized to combat race conditions
3958 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3959 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3961 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3964 - Fully reactivate OST imports after reconnection (512, others)
3965 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3966 - More graceful error handling for truncating on dead OST (515)
3967 - Don't error out unless we're actually accessing dead stripes (474)
3968 - Fix garbage sizes when stripes are missing (410)
3969 - LRU counters were broken, causing constant lock purge (433, 432)
3970 - garbage on read from stripes with failed OSTs (441)
3971 - mark OSCs as active before reconnecting during recovery (438)
3972 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3973 - lfind did not preserve OST order in output (443)
3974 - symlinks cause hung clients, incorrect data (439)
3975 - stop dereferencing request after dropping refcount (457)
3976 - don't LASSERT(spin_is_locked) on non-SMP (455)
3977 - fixes for many rename() bugs
3978 - fstat didn't correctly synchronize attributes (399)
3979 - server must handle lock cancellation during blocking AST prep (487)
3980 - bulk descriptors were free()d too soon (511)
3981 - fix paths in lconf, which would load incorrect modules (451, 507)
3982 - fix confusing lconf 'host not found' error message (386)
3983 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3984 - fix race condition in mdc_blocking_ast() for inode access (526)
3985 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3986 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3987 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3988 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3989 - fix lstripe in several ways (kernel side) (527)
3990 - fix request leak in ldlm_cli_enqueue (262)
3991 - incorrect OSC was marked inactive after OST failure
3992 - call mds_fs_cleanup before unmounting filesystem (524)
3993 - fix races between taking ns_lock and ldlm_lock_change_resource
3994 - fix races updating LOV export open file list
3995 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3996 - fix recovery NULL deref in ldlm_cli_cancel_unused
3997 - fix some DLM races by using new hash table for lock handles (419)
3998 - permit the client to specify desired inodes, at replay
3999 - duplicate requests when we queue them for replay reintegration
4000 - fix last_rcvd offset calculation
4001 - sync after each recovered transaction, so we always make progress
4002 - never, not always, ERESTART requests without transnos
4003 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4005 - skip replay if the MDS says that the client is already connected
4006 - don't check for a recovery-enabled export to match lctl's UUID
4007 - don't INC_USE_COUNT for phantom exports
4008 - don't crash when cleaning up phantom exports (567)
4009 - don't double-finish or set replay data for errored mdc_open requests
4010 - abort requests when they time out, so we don't get old replies
4011 - send/receive replies for AST messages again
4012 - if the client says that it doesn't have the lock, cancel it on the
4014 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4015 mark it as destroyed, it will all work out when decref is called
4016 - fix module use counts (22, 581)
4018 - ASTs now expect a reply (server cancels lock on error reply)
4020 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4023 - fix many simultaneous client startup (392)
4024 - fix dentry->d_it clobbering
4025 - credentials weren't being shipped for readdir/getattr operations
4026 - remove invalid assertions triggered during some concurrent MD
4028 - proper Lustre versions added (336, 389)
4029 - fix memory leak for create error case (398)
4030 - fix LOV locking bug that would get cli/srv out of sync
4031 - fix echo client over LOV (409)
4032 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4033 - fix double-O_EXCL intent crash (424)
4034 - avoid sending multiple lock CANCELs (352)
4036 - MDS can do multi-client recovery (modulo bugs in new code)
4038 - many updates, edits, cleanups
4040 2002-11-18 Phil Schwan <phil@clusterfs.com>
4043 - fix null d_it dereference (346)
4044 - fix full OST/dbench hang (333)
4045 - fix permission problem with file removal (286)
4046 - fix removal of OSCs from LOV when they fail
4047 - fix NULL deref during bulk timeout (214)
4048 - fix problems related to multiple filesystems on one MDS (241)
4049 - fixed serious subtle metadata locking bugs
4050 - free locks on clients when inodes are removed due to memory
4052 - fix inode pointer in lock data (285)
4053 - partial support for multiple MDS on a single host (241)
4054 - data locks weren't cancelled at clear_inode time (290, 311)
4055 - intent locks could lead to unbounded lock growth (205)
4056 - added a maximum lock count, an LRU list, and a flusher
4057 - fix multiple rename (365)
4058 - properly abstracted the echo client
4059 - OSC locked 1 byte too many; fixed
4060 - rewrote brw callback code:
4061 - fixed recovery bugs related to LOVs (306)
4062 - fixed too-many-pages-in-one-write crash (191)
4063 - fixed (again) crash in sync_io_timeout (214)
4064 - probably fixed callback-related race (385)
4066 - Add capability to MDS protocol
4067 - LDLM cancellations and callbacks on different portals
4069 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4072 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4073 - defer open object destruction to close time (601981, 138)
4074 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4075 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4076 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4077 - don't call obd_setattr in ll_file_release for destroyed objects
4078 * protocol change to lustre_msg: move |version| and add |flags|
4079 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4080 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4081 * added replay of create, unlink, link and rename operations during
4082 MDS failover; recovery should be much more robust now
4083 * remove failed OSCs from LOVs (only lov_create uses this so far)
4084 * the lustre-HOWTO was brought (more) up to date (582544)
4086 2002-10-23 Phil Schwan <phil@clusterfs.com>
4089 - in-use dentries weren't being reused properly (617851)
4090 - prevent multiple LDLM setup (599178)
4091 - fix LOV size calculations for truncate (617853)
4092 - fix client handling of MDS intent errors (POSIX)
4093 - fix permission bug in lovstripe.c test (624321)
4094 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4095 - truncate past end of file could corrupt data
4096 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4097 - a race in recovery could return ETIMEDOUT to apps (623947)
4098 - building outside the source directory was fixed
4099 * the lustre-HOWTO was brought (more) up to date (582544)
4100 * major progress was made on recovery functionality
4102 2002-10-10 Phil Schwan <phil@clusterfs.com>
4105 - recovery deadlock fix
4106 - rm -rf causes LBUG fix (617817)
4107 - file open by multiple tasks fix (618962)
4108 - directory permissions bugs (602707 and 620007)
4109 - journal_stop fixed with locking (611313)
4110 - O_APPEND failures resolved (618273, perhaps 614459)
4111 - lconf PATH fix (619770)
4112 - IA64 build fix (621450)
4113 - RPC buffer sizes scale with amount of memory
4115 2002-10-01 Phil Schwan <phil@clusterfs.com>
4118 - locks would be cancelled without throwing away data pages,
4119 resulting in inconsistent data (605627)
4120 - inode attributes were not always being refreshed (605627, 612449)
4121 - lconf now continues to cleanup after lctl reports an error
4122 - MDS now enforces user permissions (602707)
4123 - lprocfs cleanup fixed, but not yet enabled (614157)
4124 - fixed infinite server hang, should a client not respond to an AST
4125 - avoid going into recovery if user calls readlink() with a buffer
4126 that's too small (613941)
4127 - AST RPCs no longer require replies (614867) -- this may be changed
4128 - don't crash server if client sends an IOV that's too big (611336)
4129 - fixed lock conversion deadlock (611892)
4130 - fixed the following of symlinks (614622)
4131 * recovery: the server can remove locks from a client that dies, other
4132 clients can make progress
4133 * more extN patch fixes
4134 * compile-time configurable ptlrpc buffer allocations
4136 - collaborative read cache document
4137 - Lustre Lite Performance CDR document-in-progress
4139 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4142 - fix typo in patch-2.4.18
4144 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4147 - clear ptlrpc request each time in handle_incoming_request()
4148 - unlink of files now destroys the object on the OST
4150 2002-09-19 Peter Braam <braam@clusterfs.com>
4152 * add hard link support
4153 * change obdfile creation method
4154 * kernel patch changed
4156 2002-09-19 Peter Braam <braam@clusterfs.com>
4159 - stack overflow bug in extN fixed
4161 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4163 * documentation updates
4164 - add man pages for config tools
4165 - update tests/README to describe testing with new config tools
4166 - finish metadata API descriptions
4167 * bug fixes and cleanups
4168 - statfs workaround for 16TB limit
4169 - LOV stripe allocation improved, can stripe on subset of OSTs
4170 - LOV file size/IO offset was wrong for files > 4GB in size
4171 - object EA data was being dropped, caused files to be unreadable
4172 - memory overflow with non-LOV OST caused memory corruption
4173 - fixed regression tests to work with new config tools, obdfilter
4174 - fixed bug when directory size became larger than 1 block
4175 - fixed bug (for single client case) when PWD was deleted
4176 - invalidate local directory pages when doing intent-based ops
4177 - avoid LDLM oops when lock callback contained bad data
4179 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4181 * documentation updates
4182 * bug fixes and cleanups
4183 - configuration tools
4186 - 64-bit compile warnings
4187 - 64-bit internal statfs data
4189 * test_brw on persistent OST devices
4191 * lprocfs (disabled)
4193 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4195 * documentation updates
4196 * bug fixes and cleanups
4197 * configuration tools
4199 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4202 * many small fixes to 0_5_4
4203 * io/network handling
4204 * thinkos in MDS operations
4206 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4209 * crucial basic fixes to 0.5.3
4210 * IOR, Iozone work over Elan
4213 2002-08-07 Phil Schwan <phil@clusterfs.com>
4214 * version 0_5_3, our first alpha
4215 * we use the new Portals iovs
4216 * documentation updates
4217 * bug fixes and cleanups
4218 * small changes in the DLM wire protocol
4220 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4221 * version 0_5_1 with some initial stability,
4222 * locking on MD and file I/O.
4223 * documentation updates
4224 * several bug fixes since 0.5.0
4225 * small changes in wire protocol
4227 2002-07-18 Phil Schwan <phil@clusterfs.com>
4229 * delivered as Lustre Light Alpha
4230 * fixed a crash after handling invalid MDS requests
4231 * fixed directory pages for architectures with non-4k pages sizes
4233 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4234 * release version v0_4_4
4235 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4237 * Fixes a number of interruption problems with OST operations.
4238 * Update documentation for portals header changes
4239 * Move all wire protocol structs/defines to lustre_idl.h
4240 * Fixes symlink length bug.
4241 * Add tcpdump to repository.
4243 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4244 * release version v0_4_3
4245 * Fixes statfs for inodes on extN.
4246 * Fixes bug in runtests which would delete /etc/hosts.
4247 * Use 64-bit object IDs wherever possible (not into VFS though)
4248 Remove ost_get_info, which is unused by lustre, and out of date.
4250 2002-07-03 Peter Braam <braam@clusterfs.com>
4251 * release version v0_4_2 Fixes a lookup error (type not passed)
4252 * move forward to head of Portals
4253 * move forward to latest Lustre kernel
4255 2002-06-25 Peter Braam <braam@clusterfs.com>
4256 * release version v0_4_1. Hopefully stable on single node use.