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: interrupt oig_wait produce painc on resend.
18 Details : brw_redo_request can be used for resend requests from ptlrpcd and
19 private set, and this produce situation when rq_ptlrpcd_data not
20 copyed to new allocated request and triggered LBUG on assert
21 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
22 ptlrpcd set if request is changed and can be safety changed to use
27 Description: LASSERT_{REQ,REP}SWAB macros are buggy
28 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
29 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
30 these macros and replace them with their difinitions instead.
34 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
35 Details : Update which_patch & target file for SLES10 latest kernel.
38 Description: Remove client patches from SLES 10 kernel.
39 Details : This causes SLES 10 clients to behave as patchless clients
40 even on a Lustre-patched (server) kernel.
42 Severity : enhancement
44 Description: self-adjustable client's lru lists
45 Details : use adaptive algorithm for managing client cached locks lru
46 lists according to current server load, other client's work
47 pattern, memory activities, etc. Both, server and client
48 side namespaces provide number of proc tunables for controlling
51 Severity : enhancement
53 Description: light-weight GSS support
54 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
55 and reduce performance overhead.
57 Severity : enhancement
59 Description: Linux keyring support
60 Details : Support using service of Linux keyring for Lustre GSS internal
61 context refresh/cache mechanism.
65 Description: Fix errors in lfs documentation
66 Details : Fixes man pages
70 Description: don't use GFP_* in generic Lustre code.
71 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
75 Description: obdclass is limited by single OBD_ALLOC(idarray)
76 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
80 Description: warnings with build patchless client with vanila 2.6.19 and up
81 Details : change old ctl_table style and replace ctl_table/ctl_table_header
82 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
86 Frequency : Only for SLES
87 Description: sanity_quota fail test_1
88 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
89 file, which makes TSTID[2] inunique.
93 Frequency : Only for SLES9
94 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
95 Details : Update target/ChangeLog/which_patch .
99 Frequency : Only for test_10 in sanity-quota.sh
100 Description: a bug in quota test code
101 Details : checking a test flag in a wrong place
105 Frequency : Only for test_13 in sanity-quota.sh
106 Description: a bug in quota test script
107 Details : shouldn't check overfull allocated quota
111 Description: jbd statistics
112 Details : Port older jbd statistics patch for sles10
114 Severity : enhancement
116 Description: Add printing inode info into message about error in writepage.
121 Description: reply_lock_interpret crash due to race with it and lock cancel.
122 Details : Do not replay locks that are being cancelled. Do not reference
123 locks by their address during replay, just by their handle.
127 Frequency : When flocks are used.
128 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
129 Details : Flock locks might destroy just granted lock if it could be merged
130 with another existing flock, this is done in completion handler,
131 so teach ldlm_cli_enquque_fini that this is a valid case for
137 Description: Oops in read and write path when failing to allocate lock.
138 Details : Check if lock allocation failed and return error back.
142 Description: lstripe command fails for valid OST index
143 Details : The stripe offset is compared to lov->desc.ld_tgt_count
144 instead of lov->desc.ld_active_tgt_count.
148 Description: sanity.sh failed test 103
149 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
154 Description: df doesn't work properly if diskfs blocksize != 4K
155 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
159 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
160 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
164 Description: obd_health_check_timeout too short
165 Details : set obd_health_check_timeout as 1.5x of obd_timeout
169 Description: llapi_file_create() does not allow some changes
170 Details : add llapi_file_open() that allows specifying the mode and
171 open flags, and also returns an open file handle.
175 Description: merge and cleanup kernel patches.
176 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
180 Description: Add journal checksum support.(Kernel part)
181 Details : The journal checksum feature adds two new flags i.e
182 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
183 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
184 indicates that the commit block contains the checksum for
185 the blocks described by the descriptor blocks. Now commit
186 record can be sent to disk without waiting for descriptor
187 blocks to be written to disk. This behavior is controlled
188 using JBD2_FEATURE_ASYNC_COMMIT flag.
192 Description: OSS needs mutliple precreate threads
193 Details : Add ability to start more than one create thread per OSS.
197 Description: Sanity.sh test_65k failed: setstripe should have succeeded
198 Details : execute lfs setstripe on client
202 Description: mds_obd_create error creating tmp object
203 Details : When the user sets quota on root, llog will be affected and can't
204 create files and write files.
207 Frequency : Always on ia64 patchless client, and possibly others.
209 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
210 Details : This allows the patchless client to be loaded on architectures
215 Description: RedHat Update kernel for RHEL5
216 Details : Add the kernel config file for RHEL5.
220 Description: RedHat Update kernel for RHEL5
221 Details : Modify the kernel config file more closer RHEL5.
225 Description: Build failure against Centos5 (RHEL5)
226 Details : Define PAGE_SIZE when it isn't present.
230 Description: test_7 on sanity-quota.sh can't be used on separate machines
231 Details : add facet to handle it
235 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
236 Details : with NFS, the anon dentry's parent was set to itself in
237 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
238 corresponding dentry other than use rec->ur_name.
240 Severity : enhancement
242 Description: lfs setstripe enhancement
243 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
247 Description: testing performance impact of enabling checksumming
248 Details : enable checksum by default, allow --disable-checksum
249 configure option and "-o nochecksum" mount option
253 Description: lustre support for RHEL5
254 Details : Add support for RHEL5.
258 Description: Client eviction due to failover config
259 Details : after a connection loss, the lustre client should attempt to
260 reconnect to the last active server first before trying the
261 other potential connections.
265 Description: when mds and osts use different quota unit(32bit and 64bit),
266 quota will be released repeatly.
267 Details : void sending multiple quota reqs to mds, which will keep the
268 status between the reqs.
272 Description: osts not allocated evenly to files
273 Details : change the condition to increase offset_idx
276 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
278 Description: Data corruption for OSTs that are formatted larger than 2TB
280 Details : When generating the bio request for lustre file writes the
281 sector number would overflow a temporary variable before being
282 used for the IO. The data reads correctly from Lustre (which
283 will overflow in a similar manner) but other file data or
284 filesystem metadata may be corrupted in some cases.
288 Description: Tune the kernel for good SCSI performance.
289 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
290 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
295 Description: rewrite ext2-derived code in obdclass/uuid.c
296 Details : rewrite inherited code (uuid parsing code from ext2 utils)
297 from scratch preserving functionality.
301 Description: "lfs find -obd UUID" prints directories
302 Details : "lfs find -obd UUID" will return all directory names instead
303 of just file names. It is incorrect because the directories
304 do not reside on the OSTs.
308 Description: use i_size_read and i_size_write in 2.6 port
309 Details : replace inode->i_size access with i_size_read/write()
312 Frequency : when removing large files
314 Description: scheduling issue during removal of large Lustre files
315 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
316 It causes scheduling issues when removing large files (17TB in the
320 Frequency : only with liblustre clients on XT3
322 Description: evictions taking too long
323 Details : allow llrd to evict clients directly on OSTs
326 Frequency : only on ppc
328 Description: /proc/fs/lustre/devices broken on ppc
329 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
330 We should make sure the type of 'index' is loff_t and then cast
331 explicitly as needed below. Do not assign an explicitly cast
335 Frequency : only for rhel5
337 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
338 Details : Modify the target file & which_kernel.
340 Severity : enhancement
342 Description: omit set fsid for export NFS
343 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
344 as FS_REQUIRES_DEV for avoid problems with generate fsid.
348 Frequency : Always, for kernels after 2.6.16
349 Description: Fix warning idr_remove called for id=.. which is not
351 Details : Last kernels save old s_dev before kill super and not allow
352 to restore from callback - restore it before call kill_anon_super.
356 Description: buffer overruns could theoretically occur
357 Details : llapi_semantic_traverse() modifies the "path" argument by
358 appending values to the end of the origin string, and a
359 overrun may occur. Adding buffer overrun check in liblustreapi.
363 Description: Fix error on 'ls .' at the top of the Lustre mount.
364 Details : Don't revalidate dentry if it is a root dentry.
368 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
369 Details : Modify vm-tunables-rhel4.patch.
373 Description: Kernel config for 2.6.18-vanilla.
374 Details : Modify targets/2.6-vanilla.target.in.
375 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
376 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
377 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
378 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
382 Description: improve handling recoverable errors
383 Details : if request processig with error which can be recoverable on server
384 request should be resend, otherwise page released from cache and
389 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
390 Details : Modify target file & which_patch.
391 A flaw was found in the IA32 system call emulation provided
392 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
393 value could be stored in the %RAX register, which could trigger an
394 out-of-bounds system call table access. An untrusted local user
395 could exploit this flaw to run code in the kernel
396 (ie a root privilege escalation). (CVE-2007-4573).
400 Description: Remove 2.4 kernel checks lustre 1.8
401 Details : Remove 2.5.0 check from quota_check.c, quota_ctl.c,
402 quota_interface.c in quota subsystem.
403 Remove 2.4 kernel check from obd_mount.c, linux/linux-sysctl.c in
405 Remove lock_24kernel from fsfilt_ext3.c fsfilt_reiserfs.c
406 lvfs_linux.c upcall_cache.c in lvfs subsystem.
410 Description: O_DIRECT bypasses client statistics.
411 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
412 read_ahead_stats were updated. copied the stats section from
413 osc_send_oap_rpc() into async_internal().
417 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
421 Description: change order of libsysio includes
422 Details : '#include sysio.h' should always come before '#include xtio.h'
426 Description: handle "serious error: objid * already exists" more gracefully
427 Details : If LAST_ID value on disk is smaller than the objects existing in
428 the O/0/d* directories, it indicates disk corruption and causes an
429 LBUG(). If the object is 0-length, then we should use the existing
430 object. This will help to avoid a full fsck in most cases.
432 Severity : enhancement
434 Description: adapt the lustre_config script to support the upgrade case
435 Details : Add "-u" option for lustre_config script to support upgrading 1.4
436 server targets to 1.6 in parallel.
440 Description: To avoid grant space > avaible space when the disk is almost
441 full. Without this patch you might see the error "grant XXXX >
442 available" or some LBUG about grant, when the disk is almost
444 Details : In filter_check_grant, for non_grant cache write, we should
445 check the left space by if (*left > ungranted + bytes), instead
446 of (*left > ungranted), because only we are sure the left space
447 is enough for another "bytes", then the ungrant space should be
448 increase. In client, we should update cl_avail_grant only there
449 is OBD_MD_FLGRANT in the reply.
453 Description: Update RHEL 4 kernel to fix local root privilege escalation.
454 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
455 described in CVE-2007-4573. This problem could allow untrusted
456 local users to gain root access.
460 Description: Remove 2.4 kernel checks lustre 1.8
461 Details : Remove 2.5.0 check from dcache.c dir.c file.c llite_internal.h
462 llite_lib.c llite_mmap.c lproc_llite.c namei.c rw.c symlink.c
464 Remove 2.5.0 check from handler.c lproc_mds.c mds_fs.c
465 mds_internal.h mds_lib.c mds_open.c mds_reint.c mds_xattr.c
469 Frequency : when using O_DIRECT and quotas
471 Description: Incorrect file ownership on O_DIRECT output files
472 Details : block usage reported by 'lfs quota' does not take into account
473 files that have been written with O_DIRECT.
475 --------------------------------------------------------------------------------
477 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
479 * Support for kernels:
480 2.6.5-7.283 (SLES 9),
481 2.6.9-55.EL (RHEL 4),
482 2.6.16.46-0.14 (SLES 10),
483 2.6.18.8 vanilla (kernel.org)
484 * Client support for unpatched kernels:
485 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
486 2.6.16 - 2.6.22 vanilla (kernel.org)
487 * Due to recently discovered recovery problems, we do not recommend
488 using patchless RHEL 4 clients with this or any earlier release.
489 * Recommended e2fsprogs version: 1.39.cfs8
490 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
491 * Starting with this release, the ldiskfs backing filesystem required
492 by Lustre is now in its own package, lustre-ldiskfs. This package
493 should be installed. It is versioned separately from Lustre and
494 may be released separately in future.
498 Description: block reactivating mgc import until all deactivates complete
499 Details : Fix race when failing back MDT/MGS to itself (testing)
501 Severity : enhancement
503 Description: add optional extra BUILD_VERSION info
504 Details : add a new environment variable (namely LUSTRE_VERS) which allows
505 to override the lustre version.
508 Frequency : 2.6.18 servers only
510 Description: ll_kern_mount() doesn't release the module reference
511 Details : The ldiskfs module reference count never drops down to 0
512 because ll_kern_mount() doesn't release the module reference.
517 Description: server LBUG when using old ost_num_threads parameter
518 Details : Accept the old ost_num_threads parameter but warn that it
519 is deprecated, and fix an off-by-one error that caused an LBUG.
524 Description: Transient SCSI error results in persistent IO issue
525 Details : iobuf->dr_error is not reinitialized to 0 between two
529 Frequency : sometimes when underlying device returns I/O errors
531 Description: OSTs not going read-only during write failures
532 Details : OSTs are not remounted read-only when the journal commit threads
533 get I/O errors because fsfilt_ext3 calls journal_start/stop()
534 instead of the ext3 wrappers.
538 Description: poor connect scaling with increasing client count
539 Details : Don't run filter_grant_sanity_check for more than 100 exports
540 to improve scaling for large numbers of clients.
543 Frequency : SLES10 only
545 Description: sanity-quota.sh quotacheck failed: rc = -22
546 Details : Quotas cannot be enabled on SLES10.
549 Frequency : liblustre clients only
551 Description: getdirentries does not give error when run on compute nodes
552 Details : getdirentries does not fail when the size specified as an argument
553 is too small to contain at least one entry
555 Severity : enhancement
557 Description: Add LNET router traceability for debug purposes
558 Details : If a checksum failure occurs with a router as part of the
559 IO path, the NID of the last router that forwarded the bulk data
560 is printed so it can be identified.
565 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
566 Details : Due to a race condition, liblustre clients were occasionally
569 Severity : enhancement
571 Description: lfs setstripe use optional parameters instead of postional
574 Severity : enhancement
576 Description: Nanosecond timestamp support for ldiskfs
577 Details : The on-disk ldiskfs filesystem has added support for nanosecond
578 resolution timestamps. There is not yet support for this at
579 the Lustre filesystem level.
582 Frequency : during server recovery
584 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
585 Details : request with rq_no_resend flag not awake l_wait_event if they get a
589 Frequency : nfs export on patchless client
591 Description: connectathon hang when test nfs export over patchless client
592 Details : Disconnected dentry cannot be found with lookup, so we do not need
593 to unhash it or make it invalid
596 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
597 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
598 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
603 Description: open req refcounting wrong on reconnect
604 Details : If reconnect happened between getting open reply from server and
605 call to mdc_set_replay_data in ll_file_open, we will schedule
606 replay for unreferenced request that we are about to free.
607 Subsequent close will crash in variety of ways.
608 Check that request is still eligible for replay in
609 mdc_set_replay_data().
614 Description: disable writes to filesystem when reading health_check file
615 Details : the default for reading the health_check proc file has changed
616 to NOT do a journal transaction and write to disk, because this
617 can cause reads of the /proc file to hang and block HA state
618 checking on a healthy but otherwise heavily loaded system. It
619 is possible to return to the previous behaviour during configure
620 with --enable-health-write.
622 Severity : enhancement
624 Description: 64-bit inode version
625 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
626 to the inode. This will be required for version-based recovery.
631 Description: MDS fails to start if a duplicate client export is detected
632 Details : in some rare cases it was possible for a client to connect to
633 an MDS multiple times. Upon recovery the MDS would detect this
634 and fail during startup. Handle this more gracefully.
636 Severity : enhancement
638 Description: Add -o localflock option to simulate old noflock
640 Details : This will achieve local-only flock/fcntl locks
646 Description: log_commit_thread vs filter_destroy race leads to crash
647 Details : Take import reference before releasing llog record semaphore
652 Description: Wrong request locking in request set processing
653 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
654 imp_delayed_list, in this place should be used imp_lock.
657 Frequency : when reconnection
659 Description: Grant Leak when osc reconnect to OST
660 Details : When osc reconnect ost, OST(filter) should check whether it
661 should grant more space to client by comparing fed_grant and
662 cl_avail_grant, and return the granted space to client instead
663 of "new granted" space, because client will call osc_init_grant
664 to update the client grant space info.
667 Frequency : when client reconnect to OST
669 Description: Grant Leak when osc do resend and replay bulk write
670 Details : When osc reconnect to OST, OST(filter)should clear grant info of
671 bulk write request, because the grant info while be sync between
672 OSC and OST when reconnect, and we should ignore the grant info
673 these of resend/replay write req.
678 Description: Grant space more than avaiable left space sometimes.
679 Details : When then OST is about to be full, if two bulk writing from
680 different clients came to OST. Accord the avaliable space of the
681 OST, the first req should be permitted, and the second one
682 should be denied by ENOSPC. But if the seconde arrived before
683 the first one is commited. The OST might wrongly permit second
684 writing, which will cause grant space > avaiable space.
687 Frequency : when client is evicted
689 Description: Grant might be wrongly erased when osc is evicted by OST
690 Details : when the import is evicted by server, it will fork another
691 thread ptlrpc_invalidate_import_thread to invalidate the
692 import, where the grant will be set to 0. While the original
693 thread will update the grant it got when connecting. So if
694 the former happened latter, the grant will be wrongly errased
695 because of this race.
700 Description: Checking Stale with correct fid
701 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
702 to check whether it is stale, because sometimes, we want the
703 enqueue happened anyway, and op_data.fid2 will not be initialized.
705 Severity : enhancement
707 Description: update patchless client
708 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
711 Frequency : only with 2.4 kernel
713 Description: random memory corruption
714 Details : size of struct ll_inode_info is to big for union inode.u and this
715 can be cause of random memory corruption.
720 Description: Memory leak in recovery
721 Details : Lov_mds_md was not free in an error handler in mds_create_object.
722 It should also check obd_fail before fsfilt_start, otherwise if
723 fsfilt_start return -EROFS,(failover mds during mds recovery).
724 then the req will return with repmsg->transno = 0 and rc = EROFS.
725 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
726 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
727 be freed no matter whether fsfilt_commit success or not.
730 Frequency : only with huge count clients
732 Description: Prevents from taking the superblock lock in llap_from_page for
734 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
735 for llap_from_page call in ll_removepage prevents from taking the
736 superblock lock for a soon died page.
741 Description: Not check open intent error before release open handle
742 Details : in some rare cases, the open intent error is not checked before
743 release open handle, which may cause
744 ASSERTION(open_req->rq_transno != 0), because it tries to release
745 the failed open handle.
750 Description: Set cat log bitmap only after create log success.
751 Details : in some rare cases, the cat log bitmap is set too early. and it
752 should be set only after create log success.
756 Description: Accessing a block bevice can re-enable I/O when Lustre is
757 tearing down a device.
758 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
763 Description: service threads may hog cpus when there are a lot of requests
765 Details : Insert cond_resched to give other threads a chance to use some of
771 Description: the cat log was not initialized in recovery
772 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
773 unlink log on mds will not be initialized until mds post
774 recovery. And also in mds post recovery, the unlink log will
775 initialization will be done asynchronausly, so there will be race
776 between add unlink log and unlink log initialization.
780 Description: brw_stats were being printed incorrectly
781 Details : brw_stats were being printed as log2 but all of them were not
782 recorded as log2. Also remove some code duplication arising from
783 filter_tally_{read,write}.
787 Frequency : rare, only in recovery.
788 Description: ASSERTION(req->rq_type != LI_POISON) failed
789 Details : imp_lock should be held while iterating over imp_sending_list for
790 prevent destroy request after get timeout in ptlrpc_queue_wait.
794 Description: replay-single.sh test 52 fails
795 Details : A lock's skiplist need to be cleanup when it being unlinked
796 from its resource list.
800 Description: Short directio read returns full requested size rather than
802 Details : Direct I/O operations should return actual amount of bytes
803 transferred rather than requested size.
805 Severity : enhancement
807 Description: metadata RPC reduction (e.g. for rm performance)
808 Details : decrease the amount of synchronous RPC between clients and servers
809 by canceling conflicing lock before the operation on the client side
810 and packing thier handles into the main operation RPC to server.
812 Severity : enhancement
814 Description: Async OSC create to avoid the blocking unnecessarily.
815 Details : If a OST has no remain object, system will block on the creating
816 when need to create a new object on this OST. Now, ways use
817 pre-created objects when available, instead of blocking on an
818 empty osc while others are not empty. If we must block, we block
819 for the shortest possible period of time.
821 --------------------------------------------------------------------------------
823 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
828 Frequency : on some architectures
830 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
831 Details : Uninitialized flags sometimes cause configuration commands to
835 Frequency : patchless clients only
837 Description: missing __iget() symbol export
838 Details : The __iget() symbol export is missing. To avoid the need for
839 this on patchless clients the deathrow inode reaper is turned
840 off, and we depend on the VM to clean up old inodes. This
841 dependency was during via the fix for bug 12181.
843 --------------------------------------------------------------------------------
845 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
847 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
848 INTEROPERATE with older versions automatically. In many cases a
849 special upgrade step is needed. Please read the
850 user documentation before upgrading any part of a live system.
851 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
852 version will not interoperate with 1.6 betas before beta5 (1.5.95).
853 * WARNING: Lustre configuration and startup changes are required with
854 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
860 Severity : enhancement
862 Description: MountConf
863 Details : Lustre configuration is now managed via mkfs and mount
864 commands instead of lmc and lconf. New obd types (MGS, MGC)
865 are added for dynamic configuration management. See
866 https://mail.clusterfs.com/wikis/lustre/MountConf for
869 Severity : enhancement
871 Description: dynamic OST addition
872 Details : OSTs can now be added to a live filesystem
874 Severity : enhancement
876 Description: startup order invariance
877 Details : MDTs and OSTs can be started in any order. Clients only
878 require the MDT to complete startup.
880 Severity : enhancement
882 Description: parallel, asynchronous orphan cleanup
883 Details : orphan cleanup is now performed in separate threads for each
884 OST, allowing parallel non-blocking operation.
886 Severity : enhancement
888 Description: optimized stripe assignment
889 Details : stripe assignments are now made based on ost space available,
890 ost previous usage, and OSS previous usage, in order to try
891 to optimize storage space and networking resources.
893 Severity : enhancement
895 Description: Permanently set tunables
896 Details : All writable /proc/fs/lustre tunables can now be permanently
897 set on a per-server basis, at mkfs time or on a live
900 Severity : enhancement
902 Description: Lustre message v2
903 Details : Add lustre message format v2.
905 Severity : enhancement
907 Description: client OST exclusion list
908 Details : Clients can be started with a list of OSTs that should be
909 declared "inactive" for known non-responsive OSTs.
913 Description: SPEC SFS validation failure on NFS v2 over lustre.
914 Details : Changes the blocksize for regular files to be 2x RPC size,
915 and not depend on stripe size.
917 Severity : enhancement
919 Description: Multiple MD RPCs in flight.
920 Details : Further unserialise some read-only MDS RPCs - learn about intents.
921 To avoid overly-overloading MDS, introduce a limit on number of
922 MDS RPCs in flight for a single client and add /proc controls
923 to adjust this limit.
925 Severity : enhancement
927 Description: client read/write statistics
928 Details : Add client read/write call usage stats for performance
929 analysis of user processes.
930 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
931 file access. extents_stats shows chunk size distribution.
932 extents_stats_per_process show chunk size distribution per
935 Severity : enhancement
937 Description: mds statistics
938 Details : Add detailed mds operations statistics in
939 /proc/fs/lustre/mds/*/stats.
943 Description: Failure of copying files with lustre special EAs.
944 Details : Client side always return success for setxattr call for lustre
945 special xattr (currently only "trusted.lov").
950 Description: Refcount LNET uuids
951 Details : The global LNET uuid list grew linearly with every startup;
952 refcount repeated list entries instead of always adding to
955 Severity : enhancement
957 Description: Dynamic service threads
958 Details : Within a small range, start extra service threads
959 automatically when the request queue builds up.
962 Frequency : mixed-endian client/server environments
964 Description: mixed-endian crashes
965 Details : The new msg_v2 system had some failures in mixed-endian
968 Severity : enhancement
970 Description: Easy OST removal
971 Details : OSTs can be permanently deactivated with e.g. 'lctl
972 conf_param lustre-OST0001.osc.active=0'
974 Severity : enhancement
976 Description: MGS proc entries
977 Details : Added basic proc entries for the MGS showing what filesystems
980 Severity : enhancement
982 Description: provide MGS failover
983 Details : Added config lock reacquisition after MGS server failover.
985 Severity : enhancement
987 Description: add Linux 2.4 support
988 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
992 Description: a large application tries to do I/O to the same resource and dies
994 Details : Check the req->rq_arrival time after the call to
995 ost_brw_lock_get(), but before we do anything about
996 processing it & sending the BULK transfer request. This
997 should help move old stale pending locks off the queue as
998 quickly as obd_timeout.
1001 Frequency : when an incorrect nid is specified during startup
1003 Description: ptlrpc connect to non-existant node causes kernel crash
1004 Details : LNET can't be re-entered from an event callback, which
1005 happened when we expire a message after the export has been
1006 cleaned up. Instead, hand the zombie cleanup off to another
1009 Severity : enhancement
1011 Description: plain/inodebits lock performance improvement
1012 Details : Grouping plain/inodebits in granted list by their request modes
1013 and bits policy, thus improving the performance of search through
1017 Frequency : only if OST filesystem is corrupted
1019 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1020 Details : for a short time RPCs with bulk IO are in the replay list,
1021 but replay of bulk IOs is unimplemented. If the OST filesystem
1022 is corrupted due to disk cache incoherency and then replay is
1023 started it is possible to trip an assertion. Avoid putting
1024 committed RPCs into the replay list at all to avoid this issue.
1027 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1029 Description: System hang on startup
1030 Details : This bug allowed the liblustre (e.g. catamount) client to
1031 return to the app before handling all startup RPCs. This
1032 could leave the node unresponsive to lustre network traffic
1033 and manifested as a server ptllnd timeout.
1035 Severity : enhancement
1037 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1038 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1039 boolean to control whether to print peer debug info when a
1040 client's RPC times out.
1043 Frequency : only for kernels with patches from Lustre below 1.4.3
1045 Description: Remove old rdonly API
1046 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1049 Frequency : only for devices with external journals
1051 Description: Set external device read-only also
1052 Details : During a commanded failover stop, we set the disk device
1053 read-only while the server shuts down. We now also set any
1054 external journal device read-only at the same time.
1057 Frequency : when upgrading from 1.4 while trying to change parameters
1059 Description: The wrong (new) MDC name was used when setting parameters for
1060 upgraded MDT's. Also allows changing of OSC (and MDC)
1061 parameters if --writeconf is specified at tunefs upgrade time.
1064 Frequency : when setting specific ost indicies
1066 Description: QOS code breaks on skipped indicies
1067 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1068 created with --index need not be sequential.
1070 Severity : enhancement
1072 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1073 Details : The uninit_groups feature works in conjunction with the kernel
1074 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1075 up the pass1 processing of e2fsck. This is a read-only feature
1076 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1077 mount filesystems that have had this feature enabled.
1079 Severity : enhancement
1081 Description: Improve multi-block allocation algorithm to avoid fragmentation
1082 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1083 allocation locality and avoid filesystem fragmentation.
1085 ------------------------------------------------------------------------------
1087 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1089 * Support for kernels:
1090 2.4.21-47.0.1.EL (RHEL 3)
1091 2.6.5-7.283 (SLES 9)
1092 2.6.9-42.0.10.EL (RHEL 4)
1093 2.6.12.6 vanilla (kernel.org)
1094 2.6.16.27-0.9 (SLES 10)
1095 * Recommended e2fsprogs version: 1.39.cfs5
1097 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1101 Frequency : occasional, depends on client load and configuration
1102 Bugzilla : 12181, 12203
1103 Description: data loss for recently-modified files
1105 Details : In some cases it is possible that recently written or created
1106 files may not be written to disk in a timely manner (this should
1107 normally be within 30s unless client IO load is very high).
1108 The problem appears as zero-length files or files that are a
1109 multiple of 1MB in size after a client crash or client eviction
1110 that are missing data at the end of the file.
1112 This problem is more likely to be hit on clients where files are
1113 repeatedly created and unlinked in the same directory, clients
1114 have a large amount of RAM, have many CPUs, the filesystem has
1115 many OSTs, the clients are rebooted frequently, and/or the files
1116 are not accessed by other nodes after being written.
1118 The presence of the problem can be detected by looking at
1119 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1120 smaller than the second (nr_unused) then dirty files will not
1121 be flushed automatically to disk. "sync; sleep 10" should be
1122 run several times on the node before unmounting it to update
1123 Lustre (this is also safe to run on nodes without this problem).
1125 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1126 that can cause this same problem, so customers using that kernel
1127 also need to update the kernel in addition to Lustre. In order
1128 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1130 It is normal that files written just before a client crash (less
1131 than 30s) may not yet have been flushed to disk, even for local
1135 Frequency : frequent on thin XT3 nodes
1137 Description: UUID collision on thin XT3 Linux nodes
1138 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1139 randomly, since we relied on an insufficiently-seeded PRNG.
1144 Description: OSS hangs after "All ost request buffers busy"
1145 Details : A deadlock between quota and journal operations caused OSS
1146 hangs after printing "All ost request buffers busy."
1149 Frequency : always on liblustre builds
1151 Description: Cleanup compiler warnings on liblustre
1154 Frequency : always on liblustre builds on XT3
1156 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1160 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1161 > 32000 subdirectories
1162 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1163 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1164 subdirectory count crosses 32000. This will aid e2fsck to
1165 correctly handle more than 32000 subdirectories.
1168 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1170 Description: System hang on startup
1171 Details : This bug allowed the liblustre (e.g. catamount) client to
1172 return to the app before handling all startup RPCs. This
1173 could leave the node unresponsive to lustre network traffic
1174 and manifested as a server ptllnd timeout.
1176 Severity : enhancement
1178 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1179 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1180 boolean to control whether to print peer debug info when a
1181 client's RPC times out.
1186 Description: make O_SYNC working on 2.6 kernels
1187 Details : 2.6 kernels use different method for mark pages for write,
1188 so need add a code to lustre for O_SYNC work.
1193 Description: Failure to close file and release space on NFS
1194 Details : Put inode details into lock acquired in ll_intent_file_open.
1195 Use mdc_intent_lock in ll_intent_open to properly
1196 detect all kind of errors unhandled by mdc_enqueue.
1201 Description: proc file read during shutdown sometimes raced obd removal,
1203 Details : Add lock to prevent obd access after proc file removal.
1206 Frequency : Only for files larger than 4GB on 32-bit clients.
1208 Description: improperly doing page alignment of locks
1209 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1210 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1215 Description: RPCs being resent when they shouldn't be
1216 Details : Some RPCs that should not be resent are being resent. This
1217 can cause inconsistencies in the RPC state machine. Do not
1218 resend such requests.
1221 Frequency : rare, only with NFS export
1223 Description: Crash on NFS re-export node
1224 Details : under very unusual load conditions an assertion is hit in
1225 ll_intent_file_open()
1228 Frequency : only if OST filesystem is corrupted
1230 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1231 Details : for a short time RPCs with bulk IO are in the replay list,
1232 but replay of bulk IOs is unimplemented. If the OST filesystem
1233 is corrupted due to disk cache incoherency and then replay is
1234 started it is possible to trip an assertion. Avoid putting
1235 committed RPCs into the replay list at all to avoid this issue.
1240 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1241 Details : Large single O_DIRECT read and write calls can fail to allocate
1242 a sufficiently large buffer to process the request. In case of
1243 allocation failure the allocation is retried with a smaller
1244 buffer and broken into smaller requests.
1246 Severity : enhancement
1248 Description: Add -o localflock option to simulate old noflock behaviour.
1249 Details : This will achieve local-only flock/fcntl locks coherentness.
1254 Description: versioning check is incomplete
1255 Details : Checking the version difference of client vs. server, report
1256 error if the gap is too big.
1261 Description: add support PG_writeback bit
1262 Details : add support for PG_writeback bit for Lustre, for more carefull
1263 work with page cache in 2.6 kernel. This also fix some deadlocks
1264 and remove hack for work O_SYNC with 2.6 kernel.
1266 Severity : enhancement
1268 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1269 Details : The uninit_groups feature works in conjunction with the kernel
1270 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1271 up the pass1 processing of e2fsck. This is a read-only feature
1272 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1273 mount filesystems that have had this feature enabled.
1275 Severity : enhancement
1277 Description: Improve multi-block allocation algorithm to avoid fragmentation
1278 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1279 allocation locality and avoid filesystem fragmentation.
1281 ------------------------------------------------------------------------------
1283 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1285 * Support for kernels:
1286 2.6.9-42.0.3.EL (RHEL 4)
1287 2.6.5-7.276 (SLES 9)
1288 2.4.21-47.0.1.EL (RHEL 3)
1289 2.6.12.6 vanilla (kernel.org)
1290 2.6.16.21-0.8 (SLES10)
1291 * Recommended e2fsprogs version: 1.39.cfs2-0
1293 * The backwards-compatible /proc/sys/portals symlink has been removed
1294 in this release. Before upgrading, please ensure that you change
1295 any configuration scripts or /etc/sysctl.conf files that access
1296 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1297 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1298 in advance of the upgrade on any system running Lustre 1.4.6 or
1299 newer, since /proc/sys/lnet was added in that version.
1300 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1304 Frequency : only when quota is used
1306 Description: avoid scanning export list for quota master
1307 Details : Change the algorithms to avoid scanning export list in order
1308 to improve the efficiency.
1311 Frequency : MDS failover only, very rarely
1313 Description: "went back in time" messages on mds failover
1314 Details : The greatest transno may be lost when the current operation
1315 finishes with an error (transno==0) and the client's last_rcvd
1316 record is over-written. Save the greatest transno in the
1317 mds_last_transno for this case.
1320 Frequency : always for specific kernels and striping counts
1322 Description: client may get "Matching packet too big" without ACL support
1323 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1324 when trying to access files in certain configurations. The
1325 clients should in fact be denied when mounting because they do
1326 not understand ACLs.
1329 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1331 Description: many clients connecting with IO in progress causes connect timeouts
1332 Details : Avoid synchronous journal commits to avoid delays caused by many
1333 clients connecting/disconnecting when bulk IO is in progress.
1334 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1335 OST_IO_PORTAL to avoid delays behind potentially many pending
1339 Frequency : occasionally with multiple writers to a single file
1341 Description: shared writes to file may result in wrong size reported by stat()
1342 Details : Allow growing of kms when extent lock is cancelled
1345 Frequency : always with random mmap IO to multi-striped file
1347 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1348 Details : Only if the hole is at the end of OST object so that kms is too
1349 small. Fix is to increase kms accordingly in ll_nopage.
1352 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1354 Description: writes to a missing object would leak memory on the OST
1355 Details : If there is an inconsistency between the MDS and OST filesystems,
1356 such that the MDS references an object that doesn't exist, writes
1357 to that object will leak memory due to incorrect cleanup in the
1358 error handling path, eventually running out of memory on the OST.
1363 Description: Creating too long symlink causes lustre errors
1364 Details : Check symlink and name lengths before sending requests to MDS.
1367 Frequency : only if flock is enabled (not on by default)
1369 Description: posix locks not released on fd closure on 2.6.9+
1370 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1371 kernels, this caused such locks not to be released on fd close and
1372 then assertions on fs unmount about still used locks.
1375 Frequency : MDS failover only, very rarely
1377 Description: clients may get ASSERTION(granted_lock != NULL)
1378 Details : When request was taking a long time, and a client was resending
1379 a getattr by name lock request. The were multiple lock requests
1380 with the same client lock handle and
1381 mds_getattr_name->fixup_handle_for_resent_request found one of the
1382 lock handles but later failed with ASSERTION(granted_lock != NULL).
1387 Description: handle->h_buffer_credits > 0, assertion failure
1388 Details : h_buffer_credits is zero after truncate, causing assertion
1389 failure. This patch extends the transaction or creates a new
1393 Frequency : NFS re-export or patchless client
1394 Bugzilla : 11179, 10796
1395 Description: Crash on NFS re-export node (__d_move)
1396 Details : We do not want to hash the dentry if we don't have a lock.
1397 But if this dentry is later used in d_move, we'd hit uninitialised
1398 list head d_hash, so we just do this to init d_hash field but
1399 leave dentry unhashed.
1402 Frequency : NFS re-export or patchless client
1404 Description: NFS exports has problem with symbolic link
1405 Details : lustre client didn't properly install dentry when re-exported
1406 to NFS or running patchless client.
1409 Frequency : NFS re-export or patchless client
1411 Description: Various nfs/patchless fixes.
1412 Details : fixes reuse disconected alias for lookup process - this fixes
1413 warning "find_exported_dentry: npd != pd",
1414 fix permission error with open files at nfs.
1415 fix apply umask when do revalidate.
1418 Frequency : occasional
1420 Description: Crash on NFS re-export node
1421 Details : calling clear_page() on the wrong pointer triggered oops in
1422 generic_mapping_read().
1425 Frequency : rarely, using O_DIRECT IO
1427 Description: unaligned directio crashes client with LASSERT
1428 Details : check for unaligned buffers before trying any requests.
1431 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1433 Description: stale data returned from RAID cache
1434 Details : If only a small amount of IO is done to the RAID device before
1435 reading it again it is possible to get stale data from the RAID
1436 cache instead of reading it from disk.
1439 Frequency : always for sles10 kernel
1441 Description: sles10 support
1442 Details : ll_follow_link: compile fixes and using of nd_set_link
1443 under newer kernels.
1446 Frequency : depends on arch, kernel and compiler version, always on sles10
1449 Description: recursive or deep enough symlinks cause stack overflow
1450 Details : getting rid of large stack-allocated variable in
1454 Frequency : depends on hardware
1456 Description: lustre write performance loss in the SLES10 kernel
1457 Details : the performance loss is caused by using of write barriers in the
1458 ext3 code. The SLES10 kernel turns barrier support on by
1459 default. The fix is to undo that change for ldiskfs.
1461 ------------------------------------------------------------------------------
1463 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1465 * Support for kernels:
1466 2.6.9-42.0.3EL (RHEL 4)
1467 2.6.5-7.276 (SLES 9)
1468 2.4.21-47.0.1.EL (RHEL 3)
1469 2.6.12.6 vanilla (kernel.org)
1473 Frequency : quota enabled and large files being deleted
1475 Description: releasing more than 4GB of quota at once hangs OST
1476 Details : If a user deletes more than 4GB of files on a single OST it
1477 will cause the OST to spin in an infinite loop. Release
1478 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1483 Description: statfs data retrieved from /proc may be stale or zero
1484 Details : When reading per-device statfs data from /proc, in the
1485 {kbytes,files}_{total,free,avail} files, it may appear
1486 as zero or be out of date.
1489 Frequency : systems with MD RAID1 external journal devices
1491 Description: lconf's call to blkid is confused by RAID1 journal devices
1492 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1493 of returning all block devices that match the journal UUID.
1496 Frequency : always, for aggregate stripe size over 4GB
1498 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1499 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1500 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1503 Frequency : always if "lfs find" used on a local file/directory
1505 Description: "lfs find" segfaults if used on a local file/directory
1506 Details : The case where a directory component was not specified wasn't
1507 handled correctly. Handle this properly.
1510 Frequency : always on ppc64
1512 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1513 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1514 perform endianness conversion.
1517 Frequency : rarely (truncate to non-zero file size after write under load)
1518 Bugzilla : 10730, 10687
1519 Description: Files padded with zeros to next 4K multiple
1520 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1521 it is possible that files that are truncated to a non-zero size
1522 immediately after being written are filled with zero bytes beyond
1523 the truncated size. No file data is lost.
1525 Severity : enhancement
1527 Description: Allow recovery/failover for liblustre clients.
1528 Details : liblustre clients were unaware of failover configurations until
1531 Severity : enhancement
1533 Description: user file locks should fail when not mounting with flock option
1534 Details : Set up an error-returning stub in ll_file_operations.lock field
1535 to prevent incorrect behaviour when client is mounted without
1536 flock option. Also, set up properly f_op->flock field for
1540 Frequency : always on ia64
1542 Description: "lfs df" loops on printing out MDS statfs information
1543 Details : The obd_ioctl_data was not initialized and in some systems
1544 this caused a failure during the ioctl that did not return
1545 an error. Initialize the struct and return an error on failure.
1548 Frequency : SLES 9 only
1550 Description: Error of copying files with lustre special EAs as root
1551 Details : Client side always return success for setxattr call for lustre
1552 special xattr (currently only "trusted.lov").
1555 Frequency : rarely on clusters with both ia64+i386 clients
1557 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1558 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1559 clients are doing concurrent writes to the same file it is
1560 possible that the ia64 clients may LASSERT because the OST
1561 extent locks are not PAGE_SIZE aligned. Ensure that grown
1562 locks are always aligned on the request boundary.
1565 Frequency : specific use, occasional
1567 Description: Overwriting in use executable truncates on-disk binary image
1568 Details : If one node attempts to overwrite an executable in use by
1569 another node, we now correctly return ETXTBSY instead of
1570 truncating the file.
1572 Severity : enhancement
1574 Description: Async OSC create to avoid the blocking unnecessarily.
1575 Details : If a OST has no remain object, system will block on the creating
1576 when need to create a new object on this OST. Now, ways use
1577 pre-created objects when available, instead of blocking on an
1578 empty osc while others are not empty. If we must block, we block
1579 for the shortest possible period of time.
1584 Description: chmod on Lustre root is propagated to other clients
1585 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1586 invalid by the follow_mount time.
1589 Frequency : liblustre clients only
1591 Description: Race in 'instant cancel' lock handling could lead to such locks
1592 never to be granted in case of SMP MDS
1593 Details : Do not destroy not yet granted but cbpending locks in
1597 Frequency : replay/resend of open
1599 Description: non null lock assetion failure in mds_intent_policy
1600 Details : Trying to replay/resend lockless open requests resulted in
1601 mds_open() returning 0 with no lock. Now it sets a flag if
1602 a lock is going to be returned.
1604 Severity : enhancement
1606 Description: Checksum enhancements
1607 Details : New checksum enhancements allow for resending RPCs that failed
1610 Severity : enhancement
1612 Description: Tunables on number of dirty pages in cacche
1613 Details : Allow to set limit on number of dirty pages cached.
1618 Description: client crash on unmount - lock still has references
1619 Details : In some error handling cases it was possible to leak a lock
1620 reference on a client while accessing a file. This was not
1621 harmful to the client during operation, but would cause the
1622 client to crash when the filesystem is unmounted.
1625 Frequency : specific case, rare
1627 Description: ETXTBSY on mds though file not in use by client
1628 Details : ETXTBSY is no longer incorrectly returned when attempting to
1629 chmod or chown a directory that the user previously tried to
1630 execute or a currently-executing binary.
1633 Frequency : extremely rare except on liblustre-based clients
1635 Description: Lustre space not freed when files are deleted
1636 Details : Clean up open-unlinked files after client eviction. Previously
1637 the unlink was skipped and the files remained as orphans.
1642 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1643 Details : In some cases it was possible to send a blocking callback to
1644 a client doing a glimpse, even though that client didn't get
1645 a lock granted. When the glimpse lock is cancelled on the OST
1646 the freed lock is left on the waiting list and corrupted the list.
1649 Frequency : all core dumps
1651 Description: Broke core dumps to lustre
1652 Details : Negative dentry may be unhashed if parent does not have UPDATE
1653 lock, but some callers, e.g. do_coredump, expect dentry to be
1654 hashed after successful create, hash it in ll_create_it.
1656 ------------------------------------------------------------------------------
1658 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1660 * Support for kernels:
1661 2.6.9-42.0.2.EL (RHEL 4)
1662 2.6.5-7.276 (SLES 9)
1663 2.4.21-40.EL (RHEL 3)
1664 2.6.12.6 vanilla (kernel.org)
1668 Frequency : always on RHEL 3
1670 Description: Number of open files grows over time
1671 Details : The number of open files grows over time, whether or not
1672 Lustre is started. This was due to a filp leak introduced
1673 by one of our kernel patches.
1675 ------------------------------------------------------------------------------
1677 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1679 * Support for kernels:
1680 2.6.9-42.EL (RHEL 4)
1681 2.6.5-7.276 (SLES 9)
1682 2.4.21-40.EL (RHEL 3)
1683 2.6.12.6 vanilla (kernel.org)
1688 Bugzilla : 5719, 9635, 9792, 9684
1689 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1690 Details : If a server is under heavy load and cannot reply to new
1691 connection requests before the client resends the (re)connect,
1692 the connection handling code can behave badly if two service
1693 threads are concurrently handing separate (re)connections from
1694 the same client. Add better locking to the connection handling
1695 code, and ensure that only a single connection will be processed
1696 for a given client UUID, even if the lock is dropped.
1698 Severity : enhancement
1700 Description: add TCP zero-copy support to kernel
1701 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1702 sends if the hardware supports scatter-gather and checksumming.
1703 This allows socklnd to do client-write and server-read more
1704 efficiently and reduce CPU utilization from skbuf copying.
1707 Frequency : only if NFS exporting from client
1709 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1710 Details : If setting attributes on a file created under NFS that had
1711 never been opened it would be possible to oops the client
1712 if the file had no objects.
1715 Frequency : always for liblustre
1717 Description: liblustre client does MDS+OSTs setattr RPC for each write
1718 Details : When doing a write from a liblustre client, the client
1719 incorrectly issued an RPC to the MDS and each OST the file was
1720 striped over in order to update the timestamps. When writing
1721 with small chunks and many clients this could overwhelm the MDS
1722 with RPCs. In all cases it would slow down the write because
1723 these RPCs are unnecessary.
1725 Severity : enhancement
1727 Description: allow number of MDS service threads to be changed at module load
1728 Details : It is now possible to change the number of MDS service threads
1729 running. Adding "options mds mds_num_threads={N}" to the MDS's
1730 /etc/modprobe.conf will set the number of threads for the next
1731 time Lustre is restarted (assuming the "mds" module is also
1732 reloaded at that time). The default number of threads will
1733 stay the same, 32 for most systems.
1738 Description: OST crash if filesystem is unformatted or corrupt
1739 Details : If an OST is started on a device that has never been formatted
1740 or if the filesystem is corrupt and cannot even mount then the
1741 error handling cleanup routines would dereference a NULL pointer.
1746 Description: NULL pointer deref in llap_from_page.
1747 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1748 incorrect) mapping if the page was truncated/reclaimed while it was
1749 searched for. Check for this condition and skip such pages when
1750 doing readahead. Introduce extra check to llap_from_page() to
1751 verify page->mapping->host is non-NULL (so page is not anonymous).
1754 Frequency : Sometimes when using sys_sendfile
1756 Description: "page not covered by a lock" warnings from ll_readpage
1757 Details : sendfile called ll_readpage without right page locks present.
1758 Now we introduced ll_file_sendfile that does necessary locking
1759 around call to generic_file_sendfile() much like we do in
1763 Frequency : with certain MDS communication failures at client mount time
1765 Description: NULL pointer deref after failed client mount
1766 Details : a client connection request may delayed by the network layer
1767 and not be sent until after the PTLRPC layer has timed out the
1768 request. If the client fails the mount immediately it will try
1769 to clean up before the network times out the request. Add a
1770 reference from the request import to the obd device and delay
1771 the cleanup until the network drops the request.
1774 Frequency : occasionally during client (re)connect
1776 Description: assertion failure during client (re)connect
1777 Details : processing a client connection request may be delayed by the
1778 client or server longer than the client connect timeout. This
1779 causes the client to resend the connection request. If the
1780 original connection request is replied in this interval, the
1781 client may trip an assertion failure in ptlrpc_connect_interpret()
1782 which thought it would be the only running connect process.
1785 Frequency : only with obd_echo servers and clients that are rebooted
1787 Description: kernel BUG accessing uninitialized data structure
1788 Details : When running an obd_echo server it did not start the ping_evictor
1789 thread, and when a client was evicted an uninitialized data
1790 structure was accessed. Start the ping_evictor in the RPC
1791 service startup instead of the OBD startup.
1793 Severity : enhancement
1794 Bugzilla : 10193 (patchless)
1795 Description: Remove dependency on various unexported kernel interfaces.
1796 Details : No longer need reparent_to_init, exit_mm, exit_files,
1797 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1800 Frequency : rare (only users of deprecated and unsupported LDAP config)
1802 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1803 Details : LDAP apparently contains 'lustreName' attributes instead of
1804 'name'. A simple remapping of the name is sufficient.
1807 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1809 Description: waiting_locks_callback trips kernel BUG if client is evicted
1810 Details : Running with the dump_on_timeout debug flag turned on makes
1811 it possible that the waiting_locks_callback() can try to dump
1812 the Lustre kernel debug logs from an interrupt handler. Defer
1813 this log dumping to the expired_lock_main() thread.
1815 Severity : enhancement
1817 Description: Support NFS exporting on 2.6 kernels.
1818 Details : Implement non-rawops metadata methods for NFS server to use without
1819 changing NFS server code.
1822 Frequency : very rare (synthetic metadata workload only)
1824 Description: two racing renames might cause an MDS thread to deadlock
1825 Details : Running the "racer" program may cause one MDS thread to rename
1826 a file from being the source of a rename to being the target of
1827 a rename at exactly the same time that another thread is doing
1828 so, and the second thread has already enqueued these locks after
1829 doing a lookup of the target and is trying to relock them in
1830 order. Ensure that we don't try to re-lock the same resource.
1833 Frequency : only very large systems with liblustre clients
1835 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1836 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1837 all OSTs in parallel. This allows the eviction of stale liblustre
1838 clients to proceed much faster than if they were done in series,
1839 and also offers similar improvements for other set_info RPCs.
1844 Description: excessive CPU usage during initial read phase on client
1845 Details : During the initial read phase on a client, it would agressively
1846 retry readahead on the file, consuming too much CPU and impacting
1847 performance (since 1.4.5.8). Improve the readahead algorithm
1848 to avoid this, and also improve some other common cases (read
1849 of small files in particular, where "small" is files smaller than
1850 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1855 Description: MDS crash when receiving packet with unknown intent.
1856 Details : Do not LBUG in unknown intent case, just return -EFAULT
1858 Severity : enhancement
1859 Bugzilla : 9293, 9385
1860 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1861 Details : Do not serialize getattr (non-intent version) and statfs.
1864 Frequency : occasional, when OST network is overloaded/intermittent
1866 Description: client evicted by OST after bulk IO timeout
1867 Details : If a client sends a bulk IO request (read or write) the OST
1868 may evict the client if it is unresposive to its data GET/PUT
1869 request. This is incorrect if the network is overloaded (takes
1870 too long to transfer the RPC data) or dropped the OST GET/PUT
1871 request. There is no need to evict the client at all, since
1872 the pinger and/or lock callbacks will handle this, and the
1873 client can restart the bulk request.
1876 Frequency : Always when mmapping file with no objects
1878 Description: client crashes when mmapping file with no objects
1879 Details : Check that we actually have objects in a file before doing any
1880 operations on objects in ll_vm_open, ll_vm_close and
1886 Description: Request leak when working with deleted CWD
1887 Details : Introduce advanced request refcount tracking for requests
1888 referenced from lustre intent.
1890 Severity : Enhancement
1892 Description: Cache open file handles on client.
1893 Details : MDS now will return special lock along with openhandle, if
1894 requested and client is allowed to hold openhandle, even if unused,
1895 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1896 closing files for every read/write openration.
1898 Severity : Enhancement
1900 Description: Cache open negative dentries on client when possible.
1901 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1902 negative dentries on lock revocation.
1907 Description: Remounting a client read-only wasn't possible with a zconf mount
1908 Details : It wasn't possible to remount a client read-only with llmount.
1910 Severity : enhancement
1911 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1912 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1913 MPICH is included to add Lustre-specific ADIO interfaces.
1914 This is based closely on the UFS ADIO layer and only differs
1915 in file creation, in order to allow the OST striping to be set.
1916 This is user-contributed code and not supported by CFS.
1921 Description: extended inode attributes (immutable, append-only) work improperly
1922 when 2.4 and 2.6 kernels are used on client/server or vice versa
1923 Details : Introduce kernel-independent values for these flags.
1925 Severity : enhancement
1928 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
1929 Details : Many of the /proc/ tunables can only be tuned at a megabyte
1930 granularity. Now, Fractional MB granularity is be supported,
1931 this is very useful for low memory system.
1933 Severity : enhancement
1935 Description: Getattr by fid
1936 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
1937 locks, avoids extra getattr rpc requests to MDS, allows '/' to
1938 have locks and avoids getattr rpc requests for it on every stat.
1941 Frequency : Always, for filesystems larger than 2TB
1943 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1944 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
1945 filesystems larger than 2TB to be mounted with mballoc enabled.
1948 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
1950 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
1951 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
1952 filesystems larger than 2TB is mounted then the kernel will
1953 silently corrupt the start of the filesystem. CONFIG_LBD is
1954 enabled for all CFS-supported kernels, but the possibility of
1955 this happening with a modified kernel config exists.
1957 Severity : enhancement
1959 Description: add client O_DIRECT support for 2.6 kernels
1960 Details : It is now possible to do O_DIRECT reads and writes to files
1961 in the Lustre client mountpoint on 2.6 kernel clients.
1963 Severity : enhancement
1965 Description: parallel glimpse, setattr, statfs, punch, destroy requests
1966 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
1967 parallel, not waiting for response from every OST before sending
1968 a rpc to the next OST.
1973 Description: setattr vs write race when updating file timestamps
1974 Details : Client processes that update a file timestamp into the past
1975 right after writing to the file (e.g. tar) it is possible that
1976 the updated file modification time can be reset to the current
1977 time due to a race between processing the setattr and write RPC.
1979 Severity : enhancement
1981 Description: Bring 'lfs find' closer in line with regular Linux find.
1982 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
1983 -print0 options and obtains all the needed info through the lustre
1986 Severity : enhancement
1988 Description: support up to 1024 configured devices on one node
1989 Details : change obd_dev array from statically allocated to dynamically
1990 allocated structs as they are first used to reduce memory usage
1995 Description: Flush dirty partially truncated pages during truncate
1996 Details : Immediatelly flush partially truncated pages in filter_setattr,
1997 this way we completely avoid having any pages in page cache on OST
1998 and can retire ugly workarounds during writes to flush such pages.
2003 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2004 Details : protect inode from truncation within vfs_unlink() context
2005 just take a reference before calling vfs_unlink() and release it
2006 when parent's i_sem is free.
2011 Description: last_id value checked outside lock on OST caused LASSERT failure
2012 Details : If there were multiple MDS->OST object precreate requests in
2013 flight, it was possible that the OST's last object id was checked
2014 outside a lock and incorrectly tripped an assertion. Move checks
2015 inside locks, and discard old precreate requests.
2018 Frequency : always, if extents are used on OSTs
2020 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2021 Details : OSTs using the extents format would not zero the high 16 bits of
2022 the index physical block number. This is not a problem for any
2023 OST filesystems smaller than 16TB, and no kernels support ext3
2024 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2025 new/modified files) and can be fixed for existing filesystems
2026 with e2fsprogs-1.39-cfs1.
2031 Description: import connection selection may be incorrect if timer wraps
2032 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2033 import connections to be ignored if the 32-bit jiffies counter
2034 wraps. Use a 64-bit jiffies counter.
2037 Frequency : very large clusters immediately after boot
2039 Description: LNET request buffers exhausted under heavy short-term load
2040 Details : If a large number of client requests are generated on a service
2041 that has previously never seen so many requests it is possible
2042 that the request buffer growth cannot keep up with the spike in
2043 demand. Instead of dropping incoming requests, they are held in
2044 the LND until the RPC service can accept more requests.
2047 Frequency : Sometimes during replay
2049 Description: Assertion failure in ll_local_open after replay.
2050 Details : If replay happened on an open request reply before we were able
2051 to set replay handler, reply will become not swabbed tripping the
2052 assertion in ll_local_open. Now we set the handler right after
2053 recognising of open request
2056 Frequency : very rare
2058 Description: kernel reports "badness in vsnprintf"
2059 Details : Reading from the "recovery_status" /proc file in small chunks
2060 may cause a negative length in lprocfs_obd_rd_recovery_status()
2061 call to vsnprintf() (which is otherwise harmless). Exit early
2062 if there is no more space in the output buffer.
2064 Severity : enhancement
2066 Description: clear OBD RPC statistics by writing to them
2067 Details : It is now possible to clear the OBD RPC statistics by writing
2068 to the "stats" file.
2073 Description: Client mtime is not the same on different clients after utimes
2074 Details : In some cases, the client was using the utimes() syscall on
2075 a file cached on another node. The clients now validate the
2076 ctime from the MDS + OSTs to determine which one is right.
2081 Description: Inability to activate failout mode
2082 Details : lconf script incorrectly assumed that in pythong string's numeric
2083 value is used in comparisons.
2086 Frequency : always with multiple stripes per file
2088 Description: Inefficient object allocation for mutli-stripe files
2089 Details : When selecting which OSTs to stripe files over, for files with
2090 a stripe count that divides evenly into the number of OSTs,
2091 the MDS is always picking the same starting OST for each file.
2092 Return the OST selection heuristic to the original design.
2097 Description: mount failures may take full timeout to return an error
2098 Details : Under some heavy load conditions it is possible that a
2099 failed mount can wait for the full obd_timeout interval,
2100 possibly several minutes, before reporting an error.
2101 Instead return an error as soon as the status is known.
2103 Frequency : quota enabled and large files being deleted
2105 Description: releasing more than 4GB of quota at once hangs OST
2106 Details : If a user deletes more than 4GB of files on a single OST it
2107 will cause the OST to spin in an infinite loop. Release
2108 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2113 Description: statfs data retrieved from /proc may be stale or zero
2114 Details : When reading per-device statfs data from /proc, in the
2115 {kbytes,files}_{total,free,avail} files, it may appear
2116 as zero or be out of date.
2119 Frequency : systems with MD RAID1 external journal devices
2121 Description: lconf's call to blkid is confused by RAID1 journal devices
2122 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2123 of returning all block devices that match the journal UUID.
2126 Frequency : always, for aggregate stripe size over 4GB
2128 Description: assertion fails when trying to use 4GB stripe size
2129 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2130 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2133 Frequency : always on ppc64
2135 Description: the first write on an ext3 filesystem with mballoc got stuck
2136 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2137 perform endianness conversion.
2139 ------------------------------------------------------------------------------
2141 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2143 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2144 INTEROPERATE with older versions automatically. Please read the
2145 user documentation before upgrading any part of a live system.
2146 * WARNING: Lustre networking configuration changes are required with
2147 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2150 * Support for kernels:
2151 2.6.9-22.0.2.EL (RHEL 4)
2152 2.6.5-7.244 (SLES 9)
2153 2.6.12.6 vanilla (kernel.org)
2156 Severity : enhancement
2157 Bugzilla : 7981/8208
2158 Description: Introduced Lustre Networking (LNET)
2159 Details : LNET is new networking infrastructure for Lustre, it includes
2160 a reorganized network configuration mode (see the user
2161 documentation for full details) as well as support for routing
2162 between different network fabrics. Lustre Networking Devices
2163 (LNDs) for the supported network fabrics have also been
2164 created for this new infrastructure.
2166 Severity : enhancement
2167 Description: Introduced Access control lists
2168 Details : clients can set ACLs on files and directories in order to have
2169 more fine-grained permissions than the standard Unix UGO+RWX.
2170 The MDS must be started with the "-o acl" mount option.
2172 Severity : enhancement
2173 Description: Introduced filesystem quotas
2174 Details : Administrators may now establish per-user quotas on the
2177 Severity : enhancement
2179 Description: Configuration change for the XT3
2180 The PTLLND is now used to run Lustre over Portals on the XT3
2181 The configure option(s) --with-cray-portals are no longer used.
2182 Rather --with-portals=<path-to-portals-includes> is used to
2183 enable building on the XT3. In addition to enable XT3 specific
2184 features the option --enable-cray-xt3 must be used.
2189 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2190 Details : A race between allocating a new llcd and re-getting the llcd_lock
2191 allowed another thread to grab newly-allocated llcd.
2193 Severity : enhancement
2195 Description: 2.6 OST async journal commit and locking fix to improve performance
2196 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2197 2.6 kernels are now async as they already were in 2.4 kernels so
2198 that they can commit concurrently with the network bulk transfer.
2199 For block-allocated files the filter allocation semaphore is held
2200 to avoid filesystem fragmentation during allocation. BKL lock
2201 removed for 2.6 xattr operations where it is no longer needed.
2206 Description: lconf incorrectly determined whether two IP networks could talk
2207 Details : In some more complicated routing and multiple-network
2208 configurations, lconf will avoid trying to make a network
2209 connection to a disjoint part of the IP space. It was doing the
2210 math incorrectly for one set of cases.
2215 Description: Fix for potential infinite loop processing records in an llog.
2216 Details : If an llog record is corrupted/zeroed, it is possible to loop
2217 forever in llog_process(). Validate the llog record length
2218 and skip the remainder of the block on error.
2221 Frequency : occasional (liblustre only)
2223 Description: liblustre could not open files whose last component is a symlink
2224 Details : sysio_path_walk() would incorrectly pass the open intent to
2225 intermediate path components.
2228 Frequency : rare (liblustre only with non-standard tuning)
2229 Bugzilla : 7201 (7350)
2230 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2231 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2232 after it has been released, only do this for referenced locks
2234 Severity : enhancement
2236 Description: specifying an (invalid) directory default stripe_size of -1
2237 would reset the directory default striping
2238 Details : stripe_size -1 was used internally to signal directory stripe
2239 removal, now use "all default" to signal dir stripe removal
2240 as a directory striping of "all default" is not useful
2243 Frequency : common for large clusters running liblustre clients
2245 Description: doing an ls when liblustre clients are running is slow
2246 Details : sending a glimpse AST to a liblustre client waits for every AST
2247 to time out, as liblustre clients will not respond. Since they
2248 cannot cache data we refresh the OST lock LVB from disk instead.
2250 Severity : enhancement
2252 Description: doing an ls at the same time as file IO can be slow
2253 Details : enqueue and other "small" requests can be blocked behind many
2254 large IO requests. Create a new OST IO portal for non-IO
2255 requests so they can be processed faster.
2258 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2260 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2261 created file with mode 000
2262 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2263 allow an ftruncate() as a non-root user to a file with mode 000.
2264 Lustre now respects this flag to disable mode checks when
2265 truncating a file owned by the user
2268 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2271 Description: Revoking locks from clients that went dead or catatonic might take
2273 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2274 cancellation of such locks instant on servers without waiting for
2275 any reply from clients. Clients drops these locks when cancel
2276 notification from server is received without replying.
2279 Frequency : liblustre-only, when liblustre client dies or becomes busy
2281 Description: Doing ls on Linux clients can take a long time with active
2283 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2284 granting such locks to it in the first place if possible. Locks
2285 are taken by proxy on the OST during the read or write and
2286 dropped immediately afterward. Add connect flags handling, do
2287 not grant locks to liblustre clients for glimpse ASTs.
2289 Severity : enhancement
2291 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2292 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2293 sequential reads and behaves badly with applications doing random
2294 reads. Improve it by reading ahead at least read region, and
2295 avoiding excessive large RPC for small reads.
2297 Severity : enhancement
2299 Description: Creating more than 1000 files for a single job may cause a load
2300 imbalance on the OSTs if there are also a large number of OSTs.
2301 Details : qos_prep_create() uses an OST index reseed value that is an
2302 even multiple of the number of available OSTs so that if the
2303 reseed happens in the middle of the object allocation it will
2304 still utilize the OSTs as uniformly as possible.
2309 Description: OST or MDS may oops in ping_evictor_main()
2310 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2311 but doesn't restart at beginning of obd_exports_timed list
2314 Severity : enhancement
2316 Description: improve by-nid export eviction on the MDS and OST
2317 Details : allow multiple exports with the same NID to be evicted at one
2318 time without re-searching the exports list.
2321 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2323 Description: MDS may oops in groups_free()
2324 Details : in rare race conditions a newly allocated group_info struct is
2325 freed again, and this can be NULL. The 2.4 compatibility code
2326 for groups_free() checked for a NULL pointer, but 2.6 did not.
2329 Frequency : common for liblustre clients doing little filesystem IO
2330 Bugzilla : 9352, 7313
2331 Description: server may evict liblustre clients accessing contended locks
2332 Details : if a client is granted a lock or receives a completion AST
2333 with a blocking AST already set it would not reply to the AST
2334 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2337 Frequency : lfs setstripe, only systems with more than 160 OSTs
2339 Description: unable to set striping with a starting offset beyond OST 160
2340 Details : llapi_create_file() incorrectly limited the starting stripe
2341 index to the maximum single-file stripe count.
2344 Frequency : LDAP users only
2346 Description: lconf did not handle in-kernel recovery with LDAP properly
2347 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2349 Severity : enhancement
2351 Description: bind OST threads to NUMA nodes to improve performance
2352 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2353 node and do their allocations there to localize memory access
2355 Severity : enhancement
2357 Description: llmount can determine client NID directly from Myrinet (GM)
2358 Details : the client NID code from gmnalnid was moved directly into
2359 llmount, removing the need to use this or specifying the
2360 client NID explicitly when mounting GM clients with zeroconf
2363 Frequency : if client is started with down MDS
2365 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2366 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2367 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2368 ptlrpc_import_delayed_req() because we didn't check if the
2369 request was interrupted, and we also didn't break out of the
2370 event loop if there was no timeout
2375 Description: data loss during non-page-aligned writes to a single file from
2376 both multiple nodes and multiple threads on one node at same time
2377 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2378 inconsistency led to false short-reads, that were cached and later
2379 used by ->prepare_write() to fill in partially written page,
2380 leading to data loss.
2383 Frequency : always, if lconf --abort_recovery used
2385 Description: lconf --abort_recovery fails with 'Operation not supported'
2386 Details : lconf was attempting to abort recovery on the MDT device and not
2389 Severity : enhancement
2391 Description: remove cleanup logs
2392 Details : replace lconf-generated cleanup logs with lustre internal
2393 cleanup routines. Eliminates the need for client-cleanup and
2396 Severity : enhancement
2398 Description: add support for EAs (user and system) on lustre filesystems
2399 Details : it is now possible to store extended attributes in the Lustre
2400 client filesystem, and with the user_xattr mount option it
2401 is possible to allow users to store EAs on their files also
2403 Severity : enhancement
2405 Description: Add possibility (config option) to show minimal available OST free
2407 Details : When compiled with --enable-mindf configure option, statfs(2)
2408 (and so, df) will return least minimal free space available from
2409 all OSTs as amount of free space on FS, instead of summary of
2410 free spaces of all OSTs.
2412 Severity : enhancement
2414 Description: do not expand extent locks acquired on OST-side
2415 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2416 by server: they are not cached anyway.
2419 Frequency : when mmap is used/binaries executed from Lustre
2421 Description: Unmmap pages before throwing them away from read cache.
2422 Details : llap_shrink cache now attempts to unmap pages before discarding
2423 them (if unmapping failed - do not discard). SLES9 kernel has
2424 extra checks that trigger if this unmapping is not done first.
2429 Description: lconf didn't resolve symlinks before checking to see whether a
2430 given mountpoint was already in use
2433 Frequency : when migrating failover services
2434 Bugzilla : 6395, 9514
2435 Description: When migrating a subset of services from a node (e.g. failback
2436 from a failover service node) the remaining services would
2437 time out and evict clients.
2438 Details : lconf --force (implied by --failover) sets the global obd_timeout
2439 to 5 seconds in order to quickly disconnect, but this caused
2440 other RPCs to time out too quickly. Do not change the global
2441 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2443 Severity : enhancement
2444 Frequency : if MDS is started with down OST
2445 Bugzilla : 9439,5706
2446 Description: Allow startup/shutdown of an MDS without depending on the
2447 availability of the OSTs.
2448 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2449 Add appropriate locking and lov-osc refcounts for safe
2450 cleaning. Add osc abort_inflight calls in case the
2451 synchronize never started.
2454 Frequency : occasional (Cray XT3 only)
2456 Description: root not authorized to access files in CRAY_PORTALS environment
2457 Details : The client process capabilities were not honoured on the MDS in
2458 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2459 been accessed by an authorized user then root was able to access
2460 the file on the local client also. The root user capabilities
2461 are now allowed on the MDS, as this environment has secure UID.
2464 Frequency : occasional
2466 Description: ldiskfs "too long searching" message happens too often
2467 Details : A debugging message (otherwise harmless) prints too often on
2468 the OST console. This has been reduced to only happen when
2469 there are fragmentation problems on the filesystem.
2474 Description: Division by zero in statfs when all OSCs are inactive
2475 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2476 lov_statfs divides by value returned by lov_get_stripecnt().
2480 Bugzilla : 9489, 3273
2481 Description: First write from each client to each OST was only 4kB in size,
2482 to initialize client writeback cache, which caused sub-optimal
2483 RPCs and poor layout on disk for the first writen file.
2484 Details : Clients now request an initial cache grant at (re)connect time
2485 and so that they can start streaming writes to the cache right
2486 away and always do full-sized RPCs if there is enough data.
2487 If the OST is rebooted the client also re-establishes its grant
2488 so that client cached writes will be honoured under the grant.
2493 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2494 Details : Now I/O RPCs go to different portal number and (presumably) fast
2495 lock requests (and glimses) and other RPCs get their own service
2496 threads pool that should be able to service those RPCs
2499 Severity : enhancement
2501 Description: Ability to exchange lustre version between client and servers and
2502 issue warnings at client side if client is too old. Also for
2503 liblustre clients there is ability to refuse connection of too old
2505 Details : New 'version' field is added to connect data structure that is
2506 filled with version info. That info is later checked by server and
2510 Frequency : rare, liblustre only.
2511 Bugzilla : 9296, 9581
2512 Description: Two simultaneous writes from liblustre at offset within same page
2513 might proceed at the same time overwriting eachother with stale
2515 Details : I/O lock withing llu_file_prwv was released too early, before data
2516 actually was hitting the wire. Extended lock-holding time until
2517 server acknowledges receiving data.
2520 Frequency : extremely rare. Never observed in practice.
2522 Description: avoid generating lustre_handle cookie of 0.
2523 Details : class_handle_hash() generates handle cookies by incrementing
2524 global counter, and can hit 0 occasionaly (this is unlikely, but
2525 not impossible, because initial value of cookie counter is
2526 selected randonly). Value of 0 is used as a sentinel meaning
2527 "unassigned handle" --- avoid it. Also coalesce two critical
2528 sections in this function into one.
2530 Severity : enhancement
2532 Description: allow liblustre clients to delegate truncate locking to OST
2533 Details : To avoid overhead of locking, liblustre client instructs OST to
2534 take extent lock in ost_punch() on client's behalf. New connection
2535 flag is added to handle backward compatibility.
2537 Severity : enhancement
2538 Bugzilla : 4928, 7341, 9758
2539 Description: allow number of OST service threads to be specified
2540 Details : a module parameter allows the number of OST service threads
2541 to be specified via "options ost ost_num_threads={N}" in the
2542 OSS's /etc/modules.conf or /etc/modprobe.conf.
2546 Bugzilla : 6146, 9635, 9895
2547 Description: servers crash with bad pointer in target_handle_connect()
2548 Details : In rare cases when a client is reconnecting it was possible that
2549 the connection request was the last reference for that export.
2550 We would temporarily drop the export reference and get a new
2551 one, but this may have been the last reference and the export
2552 was just destroyed. Get new reference before dropping old one.
2554 Severity : enhancement
2555 Frequency : if client is started with failover MDS
2557 Description: Allow multiple MDS hostnames in the mount command
2558 Details : Try to read the configuration from all specified MDS
2559 hostnames during a client mount in case the "primary"
2562 Severity : enhancement
2564 Description: Stop sending data to evicted clients as soon as possible.
2565 Details : Check if the client we are about to send or are sending data to
2566 was evicted already. (Check is done every second of waiting,
2567 for which l_wait_event interface was extended to allow checking
2568 of exit condition at specified intervals).
2571 Frequency : rare, normally only when NFS exporting is done from client
2573 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2575 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2578 Frequency : always, if srand() is called before liblustre initialization
2580 Description: Liblustre uses system PRNG disturbing its usage by user application
2581 Details : Introduce internal to lustre fast and high-quality PRNG for
2582 lustre usage and make liblustre and some other places in generic
2583 lustre code to use it.
2585 Severity : enhancement
2586 Bugzilla : 9477, 9557, 9870
2587 Description: Verify that the MDS configuration logs are updated when xml is
2588 Details : Check if the .xml configuration logs are newer than the config
2589 logs stored on the MDS and report an error if this is the case.
2590 Request --write-conf, or allow starting with --old_conf.
2592 Severity : enhancement
2594 Description: Handle symlinks in the path when checking if Lustre is mounted.
2595 Details : Resolve intermediate symlinks when checking if a client has
2596 mounted a filesystem to avoid duplicate client mounts.
2601 Description: lconf can hit an error exception but still return success.
2602 Details : The lconf command catches the Command error exception at the top
2603 level script context and will exit with the associated exit
2604 status, but doesn't ensure that this exit status is non-zero.
2609 Description: failure of ptlrpc thread startup can cause oops
2610 Details : Starting a ptlrpc service thread can fail if there are a large
2611 number of threads or the server memory is very fragmented.
2612 Handle this without oopsing.
2615 Frequency : always, only if liblustre and non-default acceptor port was used
2617 Description: liblustre cannot connect to servers with non-default acceptor port
2618 Details : tcpnal_set_default_params() was not called and was therefore
2619 ignoring the environment varaible TCPNAL_PORT, as well as other
2620 TCPNAL_ environment variables
2625 Description: two objects could be created on the same OST for a single file
2626 Details : If an OST is down, in some cases it was possible to create two
2627 objects on a single OST for a single file. No problems other
2628 than potential performance impact and spurious error messages.
2632 Bugzilla : 5681, 9562
2633 Description: Client may oops in ll_unhash_aliases
2634 Details : Client dcache may become inconsistent in race condition.
2635 In some cases "getcwd" can fail if the current directory is
2641 Description: Inode refcounting problems in NFS export code
2642 Details : link_raw functions used to call d_instantiate without obtaining
2643 extra inode reference first.
2647 Bugzilla : 9942, 9903
2648 Description: Referencing freed requests leading to crash, memleaks with NFS.
2649 Details : We used to require that call to ll_revalidate_it was always
2650 followed by ll_lookup_it. Also with revalidate_special() it is
2651 possible to call ll_revalidate_it() twice for the same dentry
2652 even if first occurence returned success. This fix changes semantic
2653 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2654 reference on a request referred from the intent.
2655 ll_intent_release() then releases such a request.
2658 Frequency : rare, normally benchmark loads only
2660 Description: unlinked inodes were kept in memory on the client
2661 Details : If a client is repeatedly creating and unlinking files it
2662 can accumulate a lot of stale inodes in the inode slab cache.
2663 If there is no other client load running this can cause the
2664 client node to run out of memory. Instead flush old inodes
2665 from client cache that have the same inode number as a new inode.
2668 Frequency : SLES9 2.6.5 kernel and long filenames only
2669 Bugzilla : 9969, 10379
2670 Description: utime reports stale NFS file handle
2671 Details : SLES9 uses out-of-dentry names in some cases, which confused
2672 the lustre dentry revalidation. Change it to always use the
2676 Frequency : rare, unless heavy write-truncate concurrency is continuous
2677 Bugzilla : 4180, 6984, 7171, 9963, 9331
2678 Description: OST becomes very slow and/or deadlocked during object unlink
2679 Details : filter_destroy() was holding onto the parent directory lock
2680 while truncating+unlinking objects. For very large objects this
2681 may block other threads for a long time and slow overall OST
2682 responsiveness. It may also be possible to get a lock ordering
2683 deadlock in this case, or run out of journal credits because of
2684 the combined truncate+unlink. Solution is to do object truncate
2685 first in one transaction without parent lock, and then do the
2686 final unlink in a new transaction with the parent lock. This
2687 reduces the lock hold time dramatically.
2690 Frequency : rare, 2.4 kernels only
2692 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2693 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2694 before unmounting their filesystem in order to avoid deadlock.
2695 The kernel_locked() function in 2.4 kernels only checks whether
2696 the kernel lock is held, not whether it is this process that is
2697 holding it as 2.6 kernels do.
2702 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2703 Details : The client ptlrpc code may be trying to reconnect to a down
2704 server before a previous connection attempt has timed out.
2705 Increase the reconnect interval to be longer than the connection
2706 timeout interval to avoid sending duplicate connections to
2710 Frequency : echo_client brw_test command
2712 Description: fix echo_client to work with OST preallocated code
2713 Details : OST preallocation code (5137) didn't take echo_client IO path
2714 into account: echo_client calls filter methods outside of any
2715 OST thread and, hence, there is no per-thread preallocated
2716 pages and buffers to use. Solution: hijack pga pages for IO. As
2717 a byproduct, this avoids unnecessary data copying.
2721 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2722 Description: Client can oops in mdc_commit_close() after open replay
2723 Details : It was possible for the MDS to return an open request with no
2724 transaction number in mds_finish_transno() if the client was
2725 evicted, but without actually returning an error. Clients
2726 would later try to replay that open and may trip an assertion
2727 Simplify the client close codepath, and always return an error
2728 from the MDS in case the open is not successful.
2731 Frequency : rare, 2.6 OSTs only
2733 Description: OST may deadlock under high load on fragmented files
2734 Details : If there was a heavy load and highly-fragmented OST filesystems
2735 it was possible to have all the OST threads deadlock waiting on
2736 allocation of biovecs, because the biovecs were not released
2737 until the entire RPC IO was completed. Instead, release biovecs
2738 as soon as they are complete to ensure forward IO progress.
2740 Severity : enhancement
2742 Description: Support for specifying external journal device at mount
2743 Details : If an OST or MDS device is formatted with an external journal
2744 device, this device major/minor is stored in the ext3 superblock
2745 and may not be valid for failover. Allow detecting and
2746 specifying the external journal at mount time.
2751 Description: Mounting an MDS with pending unlinked files may cause oops
2752 Details : target_finish_recovery() calls mds_postrecov() which returned
2753 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2754 considers this an error and immediately begins cleaning up the
2755 lov, just after starting the mds_lov process
2757 Severity : enhancement
2759 Description: Implement 'lfs df' to report actual free space on per-OST basis
2760 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2761 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2762 usage of objects; '-h' to report in human readable format.
2764 ------------------------------------------------------------------------------
2766 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2773 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2774 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2775 at startup during orphan unlinking. Instead, skip buddy update
2776 before setup, as it will be regenerated after recovery is complete.
2781 Description: If an OST is inactive, its locks might reference stale inodes.
2782 Details : lov_change_cbdata() must iterate over all namespaces, even if
2783 they are inactive to clear inode references from the lock.
2785 Severity : enhancement
2786 Frequency : occasional, if non-standard max_dirty_mb used
2788 Description: Client will block write RPCs if not enough grant
2789 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2790 then the client will block writes while waiting for another RPC
2791 to complete instead of consuming its dirty limit. With change
2792 we get improved performance when max_dirty_mb is small.
2794 Severity : enhancement
2795 Bugzilla : 3389, 6253
2796 Description: Add support for supplementary groups on the MDS.
2797 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2798 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2799 lookups for user supplementary groups into a cache.
2803 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2804 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2805 for liblustre O_EXCL is enforced by MDS.
2809 Description: readdir never returns NULL in liblustre.
2810 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2811 dentry in struct dirent.
2815 Frequency : liblustre only, depends on application IO pattern
2816 Description: liblustre clients evicted if not contacting servers
2817 Details : Don't put liblustre clients into the ping_evictor list, so
2818 they will not be evicted by the pinger ever.
2820 Severity : enhancement
2822 Description: Add ability to evict clients by NID from MDS.
2823 Details : By echoing "nid:$NID" string into
2824 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2825 $NID would be instantly evicted from this MDS and from all active
2826 OSTs connected to it.
2830 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2831 Details : lookup_it_finish() used to query file size from OSTs that was not
2836 Description: service threads change working directory to that of init
2837 Details : Starting lustre service threads may pin the working directory
2838 of the parent thread, making that filesystem busy. Threads
2839 now change to the working directory of init to avoid this.
2843 Frequency : during shutdown only
2844 Description: shutdown with a failed MDS or OST can cause unmount to hang
2845 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2850 Frequency : 2.6 only
2851 Description: chmod/chown may include an extra supplementary group
2852 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2853 supplementary group and if none is specified this is used.
2856 Bugzilla : 5479 (6816)
2858 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2859 Details : If lookup is in progress on a file that is unlinked we might try
2860 to revalidate the inode and fail in revalidate after lookup is
2861 complete and ll_file_open() enqueues the open again but
2862 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2865 Frequency : always, if lconf --abort_recovery used
2867 Description: lconf --abort_recovery fails with 'Operation not supported'
2868 Details : lconf was attempting to abort recovery on the MDT device and not
2871 ------------------------------------------------------------------------------
2873 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2878 Frequency : rare (only unsupported configurations with a node running as an
2880 Bugzilla : 6514, 5137
2881 Description: Mounting a Lustre file system on a node running as an OST could
2883 Details : OSTs now preallocates memory needed to write out data at
2884 startup, instead of when needed, to avoid having to
2885 allocate memory in possibly low memory situations.
2886 Specifically, if the file system is mounted on on OST,
2887 memory pressure could force it to try to write out data,
2888 which it needed to allocate memory to do. Due to the low
2889 memory, it would be unable to do so and the node would
2890 become unresponsive.
2892 Severity : enhancement
2894 Description: Addition of lconf --service command line option
2895 Details : lconf now accepts a '--service <arg>' option, which is
2896 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2898 Severity : enhancement
2900 Description: Failover mode is now the default for OSTs.
2901 Details : By default, OSTs will now run in failover mode. To return to
2902 the old behaviour, add '--failout' to the lmc line for OSTs.
2904 Severity : enhancement
2906 Description: Health checks are now provided for MDS and OSTs
2907 Details : Additional detailed health check information on MSD and OSTs
2908 is now provided through the procfs health_check value.
2911 Frequency : occasional, depends on IO load
2913 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2914 after numerous create/delete cycles
2915 Details : The ext3 inode allocation policy would not allocate new inodes
2916 very well on the OSTs because there are no new directories
2917 being created. Instead we look for groups with free space if
2918 the parent directories are nearly full.
2923 Description: Network or server problems during mount may cause partially
2924 mounted clients instead of returning an error.
2925 Details : The config llog parsing code may overwrite the error return
2926 code during mount error handling, returning success instead
2932 Description: MDS can fail to allocate large reply buffers
2933 Details : After long uptimes the MDS can fail to allocate large reply
2934 buffers (e.g. zconf client mount config records) due to memory
2935 fragmentation or consumption by the buffer cache. Preallocate
2936 some large reply buffers so that these replies can be sent even
2937 under memory pressure.
2941 Frequency : rare (liblustre)
2942 Description: fsx running with liblustre complained that using truncate() to
2943 extend the file doesn't work. This patch corrects that issue.
2944 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2945 ATTR_SIZE and lsm use in llu_setattr_raw.
2949 Frequency : rare, only 2.6 kernels
2950 Description: Unusual file access patterns on the MDS may result in inode
2951 data being lost in very rare circumstances.
2952 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2953 code in the 2.6 kernel caused the inode and/or EA data not to
2954 be read from disk, causing single-file corruption.
2958 Frequency : rare, only 2.6 filesystems using extents
2959 Description: Heavy concurrent write and delete load may cause data corruption.
2960 Details : It was possible under high-load situations to have an extent
2961 metadata block in the block device cache from a just-unlinked
2962 file overwrite a newly-allocated data block. We now unmap any
2963 metadata buffers that alias just-allocated data blocks.
2967 Frequency : filesystems with default stripe_count larger than 77
2968 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
2969 Details : lconf specifies an inode size of 4096 bytes when the default
2970 stripe_count is larger than 77. This conflicts with the default
2971 inode density of 1 per 4096 bytes. Allocate smaller inodes in
2972 this case to avoid pinning too much memory for large EAs.
2974 ------------------------------------------------------------------------------
2976 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
2981 Frequency : rare (extremely heavy IO load with hundreds of clients)
2983 Description: Client is evicted, gets IO error writing to file
2984 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
2985 caused clients to be evicted for AST timeouts. The fixes in
2986 bug 5192 mean we no longer need to have such short AST timeouts
2987 so ldlm_timeout has been increased.
2990 Frequency : occasional during --force or --failover shutdown under load
2991 Bugzilla : 5949, 4834
2992 Description: Server oops/LBUG if stopped with --force or --failover under load
2993 Details : a collection of import/export refcount and cleanup ordering
2994 issues fixed for safer force cleanup
2997 Frequency : only filesystems larger than 120 OSTs
2998 Bugzilla : 5990, 6223
2999 Description: lfs getstripe would oops on a very large filesystem
3000 Details : lov_getconfig used kfree on vmalloc'd memory
3003 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3004 Bugzilla : 6242, 6243
3005 Description: reading from files that had been truncated to a non-zero size
3006 but never opened returned no data
3007 Details : ll_file_read() reads zeros from no-object files to EOF
3012 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3013 Details : There's little harm in aborting MDS/OSS recovery and letting it
3014 try again, so I removed the LASSERT and return an error instead.
3016 Severity : enhancement
3018 Description: New debugging infrastructure for tracking down data corruption
3019 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3020 (b) cover more of the client-side code path, and (c) try to narrow
3021 down where problems occurred
3025 Bugzilla : 3819, 4364, 4397, 6313
3026 Description: Racing close and eviction MDS could cause assertion in mds_close
3027 Details : It was possible to get multiple mfd references during close and
3028 client eviction, leading to one thread referencing a freed mfd.
3030 Severity: : enhancement
3031 Bugzilla : 3262, 6359
3032 Description: Attempts to reconnect to servers are now more aggressive.
3033 Details : This builds on the enhanced upcall-less recovery that was added
3034 in 1.4.2. When trying to reconnect to servers, clients will
3035 now try each server in the failover group every 10 seconds. By
3036 default, clients would previously try one server every 25 seconds.
3041 Description: After recovery, certain operations trigger a failed
3042 assertion on a client.
3043 Details : Failing over an mds, using lconf -d --failover, while a
3044 client was doing a readdir() call would cause the client to
3045 LBUG after recovery completed and the readdir() was resent.
3047 Severity : enhancement
3049 Description: Default groups are now added by lconf
3050 Details : You can now run lconf --group <servicename> without having to
3051 manually add groups with lmc.
3054 Frequency : occasional
3056 Description: Nodes with an elan id of 0 trigger a failed assertion
3059 Frequency : always when accessing e.g. tty/console device nodes
3061 Description: tty and some other devices nodes cannot be used on lustre
3062 Details : file's private_data field is used by device data and lustre
3063 values in there got lost. New field was added to struct file to
3064 store fs-specific private data.
3067 Frequency : when exporting Lustre via NFS
3069 Description: NFSD failed occasionally when looking up a path component
3070 Details : NFSD is looking up ".." which was broken in ext3 directories
3071 that had grown large enough to become hashed.
3074 Frequency : Clusters with multiple interfaces not on the same subnet
3076 Description: Nodes will repeatedly try to reconnect to an interface which it
3077 cannot reach and report an error to the log.
3078 Details : Extra peer list entries will be created by lconf with some peers
3079 unreachable. lconf now validates the peer before adding it.
3082 Frequency : Only if a default stripe is set on the filesystem root.
3084 Description: Setting a default stripe on the filesystem root prevented the
3085 filesystem from being remounted.
3086 Details : The client was sending extra request flags in the root getattr
3087 request and did not allocate a reply buffer for the dir EA.
3090 Frequency : occasional, higher if lots of files are accessed by one client
3091 Bugzilla : 6159, 6097
3092 Description: Client trips assertion regarding lsm mismatch/magic
3093 Details : While revalidating inodes the VFS looks up inodes with ifind()
3094 and in rare cases can find an inode that is being freed.
3095 The ll_test_inode() code will free the lsm during ifind()
3096 when it finds an existing inode and then the VFS later attaches
3097 this free lsm to a new inode.
3101 Bugzilla : 6422, 7030
3102 Description: MDS deadlock between mkdir and client eviction
3103 Details : Creating a new file via mkdir or mknod (starting a transaction
3104 and getting the ns lock) can deadlock with client eviction
3105 (gets ns lock and trying to finish a synchronous transaction).
3108 Frequency : occasional
3109 Description: While starting a server, the fsfilt_ext3 module could not be
3111 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3112 kernels. Previously, lconf would still use the ext3 name
3113 when trying to load modules. Now, it will correctly use
3114 ext3 on 2.4 and ldiskfs on 2.6.
3116 Severity : enhancement
3117 Description: The default stripe count has been changed to 1
3118 Details : The interpretation of the default stripe count (0, to lfs
3119 or lmc) has been changed to mean striping across a single
3120 OST, rather than all available. For general usage we have
3121 found a stripe count of 1 or 2 works best.
3123 Severity : enhancement
3124 Description: Add support for compiling against Cray portals.
3125 Details : Conditional compiling for some areas that are different
3129 Frequency : occasional
3130 Bugzilla : 6409, 6834
3131 Description: Creating files with an explicit stripe count may lead to
3132 a failed assertion on the MDS
3133 Details : If some OSTs are full or unavailable, creating files may
3134 trigger a failed assertion on the MDS. Now, Lustre will
3135 try to use other servers or return an error to the
3139 Frequency : occasional
3141 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3142 caused lock timeout during readahead (since 1.4.2).
3143 Details : Processes doing readahead might match a lock that hasn't been
3144 granted yet if there are overlapping and conflicting lock
3145 requests. The readahead process waits on ungranted lock
3146 (original lock is CBPENDING), while OST waits for that process
3147 to cancel CBPENDING read lock and eventually evicts client.
3149 Severity : enhancement
3151 Description: Initial enabling of flock support for clients
3152 Details : Implements fcntl advisory locking and file status functions.
3153 This feature is provided as an optional mount flag (default
3154 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3155 locking are implemented yet, and those that are are not guaranteed
3156 to be completely correct in production environments.
3157 mount -t lustre -o [flock|noflock] ...
3160 Frequency : occasional
3162 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3163 assertion in the ext3 JBD (journaling) layer.
3164 Details : The b_committed_data struct is protected by the big kernel lock
3165 in 2.4 kernels, serializing journal_commit_transaction() and
3166 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3167 access to this struct. In 2.6 kernels there is finer grained
3168 locking to improve SMP performance of the JBD layer.
3172 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3177 Description: merge and cleanup kernel patches.
3180 -----------------------------------------------------------------------------
3182 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3184 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3185 versions of Lustre. Please update all servers and clients to
3186 version 1.4.2 or later at the same time. You must also run
3187 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3188 to update the configuration logs.
3190 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3191 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3192 - swab llog records of type '0' so we get proper header size/idx (5861)
3193 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3194 - fix rename of one directory over another leaking an inode (5953)
3195 - avoid SetPageDirty on 2.6 (5981)
3196 - don't re-add just-being-destroyed locks to the waiting list (5653)
3197 - when creating new directories, inherit the parent's custom
3198 striping settings if present parent (3048)
3199 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3200 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3201 - don't hold client locks on temporary worklist from l_lru (5666)
3202 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3203 - automatically evict dead clients (5921)
3204 - Update file size properly in create+truncate+fstat case (6196)
3205 - Do not unhash mountpoint dentries, do not allow removal of
3207 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3208 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3209 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3210 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3211 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3212 - Correctly handle reads of files with no objects (6243)
3213 - lctl recover will also mark a device active if deactivate used (5933)
3215 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3216 - allow --write-conf on an MDS with different nettype than client (5619)
3217 - don't write config llogs to MDS for mounts not from that MDS (5617)
3218 - lconf should create multiple TCP connections from a client (5201)
3219 - init scripts are now turned off by default; run chkconfig --on
3220 lustre and chkconfig --on lustrefs to use them
3221 - upcalls are no longer needed for clients to recover to failover
3223 - add --abort-recovery option to lconf to abort recovery on device
3225 - add support for an arbitrary number of OSTs (3026)
3226 - Quota support protocol changes.
3227 - forward compatibility changes to wire structs (6007)
3228 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3229 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3230 - improved "lustre status" script
3231 - initialize blocksize for non-regular files (6062)
3232 - added --disable-server and --disable-client configure options (5782)
3233 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3234 - Vanilla 2.4.29 support
3235 - increase maximum number of obd devices to 520 (6242)
3236 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3237 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3238 SLES 9 2.6.5 kernels
3239 - limit stripes per file to 160 (the maximum EA size) (6093)
3241 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3244 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3245 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3246 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3247 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3248 - small ext3 extents cleanups and fixes (5733)
3249 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3250 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3251 - avoid needless client->OST connect, fix handle mismatch (5317)
3252 - fix DLM error path that led to out-of-sync client, long delays (5779)
3253 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3254 - fix several locking issues related to i_size (5492,5624,5654,5672)
3255 - don't move pending lock onto export if it is already evicted (5683)
3256 - fix kernel oops when creating .foo in unlinked directory (5548)
3257 - fix deadlock in obdfilter statistics vs. object create (5811)
3258 - use time_{before,after} to avoid timer jiffies wrap (5882)
3259 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3260 - Do not leak request if server was not able to process it (5154)
3261 - If mds_open unable to find parent dir, make that negative lookup(5154)
3262 - don't create new directories with extent-mapping (5909, 5936)
3264 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3265 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3266 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3267 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3268 - support for RHEL4 kernel on i686 (5773)
3269 - provide error messages when incompatible logs are encountered (5898)
3271 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3272 * version 1.4.0.10 (1.4.1 release candidate 1)
3274 - don't keep a lock reference when lock is not granted (4238)
3275 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3276 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3277 - drop import inflight refcount on signal_completed_replay error (5255)
3278 - unlock page after async write error during send (3677)
3279 - handle missing objects in filter_preprw_read properly (5265)
3280 - no transno return for symlink open, don't save no-trasno open (3440)
3281 - don't try to complete elan receive that already failed (4012)
3282 - free RPC server reply state on error (5406)
3283 - clean up thread from ptlrpc_start_thread() on error (5160)
3284 - readahead could read extra page into cache that wasn't ejected (5388)
3285 - prevent races in class_attach/setup/cleanup/detach (5260)
3286 - don't dereference de->d_inode after l_dput of de (5458)
3287 - use "int" for stripe value returned from lock_to_stripe (5544)
3288 - mballoc allocation and error-checking fixes in 2.6 (5504)
3289 - block device patches to fix I/O request sizes in 2.6 (5482)
3290 - look up hostnames for IB nals (5602)
3291 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3292 - don't start multiple acceptors for the same port (5277)
3293 - fix incorrect LASSERT in mds_getattr_name (5635)
3294 - export a proc file for general "ping" checking (5628)
3295 - fix "lfs check" to not block when the MDS is down (5628)
3297 - service request history (4965)
3298 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3299 - create an "evict_client" /proc entry on OSTs, like the MDS has
3300 - fix mount usage message, return errors per mount(8) (5168)
3301 - change grep [] to grep "[]" in tests so they work in more UMLs
3302 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3303 - remove ancient LOV_MAGIC_V0 EA support (5047)
3304 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3305 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3306 - allow readahead allocations to fail when low on memory (5383)
3307 - mmap locking landed again, after considerable improvement (2828)
3308 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3310 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3313 - send OST transaction number in read/write reply to free req (4966)
3314 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3315 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3316 - the watchdog thread now runs as interruptible (5246)
3317 - flock/lockf fixes (but it's still disabled, pending 5135)
3318 - don't use EXT3 constants in llite code (5094)
3319 - memory shortage at startup could cause assertion (5176)
3321 - reorganization of lov code
3322 - single portals codebase
3324 - add extents/mballoc support (5025)
3325 - direct I/O reads in the obdfilter (4048)
3326 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3328 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3331 - send OST transaction number in read/write reply to free req (4966)
3332 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3333 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3334 - the watchdog thread now runs as interruptible (5246)
3335 - handle missing objects in filter_preprw_read properly (5265)
3336 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3337 - drop import inflight refcount on signal_completed_replay error (5255)
3338 - unlock page after async write error during send (3677)
3339 - return original error code on reconstructed replies (3761)
3340 - no transno return for symlink open, don't save no-trasno open (3440)
3342 - add pid to ldlm debugging output (4922)
3343 - bump the watchdog timeouts -- we can't handle 30sec yet
3344 - extra debugging for orphan dentry/inode bug (5259)
3346 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3349 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3350 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3351 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3352 - fix NULL dereference in /proc/sys/portals/routes (4827)
3353 - allow failed mdc_close() operations to be interrupted (4561)
3354 - stop precreate on OST before MDS would time out on it (4778)
3355 - don't send partial-page writes before EOF from client (4410)
3356 - discard client grant for sub-page writes on large-page clients (4520)
3357 - don't free dentries not owned by NFS code, check generation (4806)
3358 - fix lsm leak if mds_create_objects() fails (4801)
3359 - limit debug_daemon file size, always print CERROR messages (4789)
3360 - use transno after validating reply (3892)
3361 - process timed out requests if import state changes (3754)
3362 - update mtime on OST during writes, return in glimpse (4829)
3363 - add mkfsoptions to LDAP (4679)
3364 - use ->max_readahead method instead of zapping global ra (5039)
3365 - don't interrupt __l_wait_event() during strace
3367 - add software watchdogs to catch hung threads quickly (4941)
3368 - make lustrefs init script start after nfs is mounted
3369 - fix CWARN/ERROR duplication (4930)
3370 - return async write errors to application if possible (2248)
3371 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3372 - print NAL number in %x format (4645)
3373 - update barely-supported suse-2.4.21-171 series (4842)
3374 - support for sles 9 %post scripts
3375 - support for building 2.6 kernel-source packages
3376 - support for sles km_* packages
3378 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3381 - ignore -ENOENT errors in osc_destroy (3639)
3382 - notify osc create thread that OSC is being cleaned up (4600)
3383 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3384 - reconstruct ost_handle() like mds_handle() (4657)
3385 - create a new thread to do import eviction to avoid deadlock (3969)
3386 - let lconf resolve symlinked-to devices (4629)
3387 - don't unlink "objects" from directory with default EA (4554)
3388 - hold socknal file ref over connect in case target is down (4394)
3389 - allow more than 32000 subdirectories in a single directory (3244)
3390 - fix blocks count for O_DIRECT writes (3751)
3391 - OST returns ENOSPC from object create when no space left (4539)
3392 - don't send truncate RPC if file size isn't changing (4410)
3393 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3394 - bind to privileged port in socknal and tcpnal (3689)
3396 - rate limit CERROR/CWARN console message to avoid overload (4519)
3397 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3398 - basic mmap support (3918)
3399 - kernel patch series update from b1_4 (4711)
3401 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3404 - avoid crash during MDS cleanup with OST shut down (2775)
3405 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3406 - don't use bad inodes on the MDS (3744)
3407 - dynamic object preallocation to improve recovery speed (4236)
3408 - don't hold spinlock over lock dumping or change debug flags (4401)
3409 - don't zero obd_dev when it is force cleaned (3651)
3410 - print grants to console if they go negative (4431)
3411 - "lctl deactivate" will stop automatic recovery attempts (3406)
3412 - look for existing locks in ldlm_handle_enqueue() (3764)
3413 - don't resolve lock handle twice in recovery avoiding race (4401)
3414 - revalidate should check working dir is a directory (4134)
3416 - don't always mark "slow" obdfilter messages as errors (4418)
3418 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3421 - don't close LustreDB during write_conf until it is done (3860)
3422 - fix typo in lconf for_each_profile (3821)
3423 - allow dumping logs from multiple threads at one time (3820)
3424 - don't allow multiple threads in OSC recovery (3812)
3425 - fix debug_size parameters (3864)
3426 - fix mds_postrecov to initialize import for llog ctxt (3121)
3427 - replace config semaphore with spinlock (3306)
3428 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3429 - don't allow enqueue to complete on a destroyed export (3822)
3430 - down write_lock before checking llog header bitmap (3825)
3431 - recover from lock replay timeout (3764)
3432 - up llog sem before sending rpc (3652)
3433 - reduce ns lock hold times when setting kms (3267)
3434 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3435 - fix NULL deref and obd_dev leak on setup error (3312)
3436 - replace some LBUG about llog ops with error handling (3841)
3437 - don't match INVALID dentries from d_lookup and spin (3784)
3438 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3439 - fix invalid assertion in ptlrpc_set_wait (3880)
3441 - add libwrap support for the TCP acceptor (3996)
3442 - add /proc/sys/portals/routes for non-root route listing (3994)
3443 - allow setting MDS UUID in .xml (2580)
3444 - print the stack of a process that LBUGs (4228)
3446 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3449 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3450 - make sure to unset replay flag from failed open requests (3440)
3451 - if default stripe count is 0, use OST count for inode size (3636)
3452 - update parent mtime/ctime on client for create/unlink (2611)
3453 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3454 - free recovery state on server during a forced cleanup (3571)
3455 - unregister_reply for resent reqs (3063)
3456 - loop back devices mounting and status check on 2.6 (3563)
3457 - fix resource-creation race that can provoke i_size == 0 (3513)
3458 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3459 - more debugging for page-accounting assertion (3746)
3460 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3461 - avoid deadlock after precreation failure (3758)
3462 - fix race and lock order deadlock in orphan handling (3450, 3750)
3463 - add validity checks when grabbing inodes from l_ast_data (3599)
3465 - add /proc/.../recovery_status to obdfilter (3428)
3466 - lightweight CDEBUG infrastructure, debug daemon (3668)
3467 - change default OSC RPC parameters to be better on small clusters
3468 - turn off OST read cache for files smaller than 32MB
3469 - install man pages and include them in rpms (3100)
3470 - add new init script for (un)mounting lustre filesystems (2593)
3471 - run chkconfig in %post for init scripts (3701)
3472 - drop scimac NAL (unmaintained)
3474 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3477 - clean kiobufs before and after use (3485)
3478 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3479 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3480 - change init script to fail more clearly if not run as root (1528)
3481 - allow clients to reconnect during replay (1742)
3482 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3483 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3484 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3485 - fix cond_resched (3554)
3486 - only evict unfinished clients after recovery (3515)
3487 - allow bulk resend, prevent data loss (3570)
3488 - dynamic ptlrpc request buffer allocation (2102)
3489 - don't allow unlinking open directory if it isn't empty (2904)
3490 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3491 - remove extraneous obd dereference causing LASSERT failure (3334)
3492 - don't use get_cycles() when creating temp. files on the mds (3156)
3493 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3494 - handle EEXIST for set-stripe, set proper directory name (3336)
3496 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3497 - fix ksocknal_fmb_callback() error messages (2918)
3499 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3502 - don't copy lvb into (possibly NULL) reply on error (2983)
3503 - don't deref dentry after dput, don't free lvb on error (2922)
3504 - use the kms to determine writeback rpc length (2947)
3505 - increment oti_logcookies when osc is inactive (2948)
3506 - update client's i_blocks count via lvb messages (2543)
3507 - handle intent open/close of special files properly (1557)
3508 - mount MDS with errors=remount-ro, like obdfilter (2009)
3509 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3510 - don't use cancelling-locks' kms values (2947)
3511 - use highest lock extent for kms, not last one (2925)
3512 - don't dereference ERR_PTR() dentry in error handling path (3107)
3513 - fix thread race in portals_debug_dumplog() (3122)
3514 - create lprocfs device entries at setup instead of at attach (1519)
3515 - common AST error handler, don't evict client on completion race (3145)
3516 - zero nameidata in detach_mnt in 2.6 (3118)
3517 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3518 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3519 - initialize RPC timeout timer earlier for 2.6 (3219)
3520 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3521 - print nal/nid for unknown nid (3258)
3522 - additional checks for oscc recovery before doing precreate (3284)
3523 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3524 - don't crash in mdc_close for bad permissions on open (3285)
3525 - zero i_rdev for non-device files (3147)
3526 - clear page->private before handing to FS, better assertion (3119)
3527 - tune the read pipeline (3236)
3528 - fix incorrect decref of invalidated dentry (2350)
3529 - provide read-ahead stats and refine rpc in flight stats (3328)
3530 - don't hold journal transaction open across create RPC (3313)
3531 - update atime on MDS at close time (3265)
3532 - close LDAP connection when recovering to avoid server load (3315)
3533 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3534 - don't leak open file on MDS after open resend (3325)
3535 - serialize filter_precreate and filter_destroy_precreated (3329)
3536 - loop device shouldn't call sync_dev() for nul device (3092)
3537 - clear page cache after eviction (2766)
3538 - resynchronize MDS->OST in background (2824)
3539 - refuse to mount the same filesystem twice on same mountpoint (3394)
3540 - allow llmount to create routes for mounting behind routers (3320)
3541 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3542 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3543 - fix rare problem with rename on htree directories (3417)
3545 - allow default OST striping configuration per directory (1414)
3546 - fix compilation for qswnal for 2.6 kernels (3125)
3547 - increase maximum number of MDS request buffers for large systems
3548 - change liblustreapi to be useful for external progs like lfsck (3098)
3549 - increase local configuration timeout for slow disks (3353)
3550 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3552 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3555 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3556 - don't overwrite extent policy data in reply if lock was blocked (2901)
3557 - drop filter export grants atomically with removal from device (2663)
3558 - del obd_self_export from work_list in class_disconnect_exports (2908)
3559 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3560 - swab reply message in mdc_close, other PPC fixes (2464)
3561 - fix destroying of named logs (2325)
3562 - overwrite old logs when running lconf --write_conf (2264)
3563 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3564 - fix race in target_handle_connect (2898)
3565 - mds_reint_create() should take same inode create lock (2926)
3566 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3567 - don't close files for self_export to avoid uninitialized obd (2936)
3568 - allow MDS with the same name as client node (2939)
3569 - hold dentry reference for closed log files for unlink (2325)
3570 - reserve space for all logs during transactions (2059)
3571 - don't evict page beyond end of stripe extent (2925)
3572 - don't oops on a deleted current working directory (2399)
3573 - handle hard links to targets without a parent properly (2517)
3574 - don't dereference NULL lock when racing during eviction (2867)
3575 - don't grow lock extents when lots of conflicting locks (2919)
3577 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3580 - account for cache space usage on clients to avoid data loss (974)
3581 - lfsck support in lustre kernel code (2349)
3582 - reduce journal credits needed for BRW writes (2370)
3583 - orphan handling to avoid losing space on client/server crashes
3584 - ptlrpcd can be blocked, stopping ALL progress (2477)
3585 - use lock value blocks to assist in proper KMS, faster stat (1021)
3586 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3587 - recovery for initial connections (2355)
3588 - fixes for mds_cleanup_orphans (1934)
3589 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3590 - block all file creations until orphan recovery completes (1901)
3591 - client remove rq_connection from request struct (2423)
3592 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3593 - recovery timer race (2670)
3594 - mdc_close recovey bug (2532)
3595 - ptlrpc cleanup bug (2710)
3596 - mds timeout on local locks (2588)
3597 - namespace lock held during RPCs (2431)
3598 - handle interrupted sync write properly (2503)
3599 - don't try to handle a message that hasn't been replied to (2699)
3600 - client assert failure during cleanup after abort recovery (2701)
3601 - leak mdc device after failed mount (2712)
3602 - ptlrpc_check_set allows timedout requests to complete (2714)
3603 - wait for inflight reqs when ptlrpcd finishes (2710)
3604 - make sure unregistered services are removed from the srv_list
3605 - reset bulk XID's when resending them (caught by 1138 test)
3606 - unregister_bulk after timeout
3607 - fix lconf error (2694)
3608 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3609 - readahead locks pages, leaves pending causing memory pressure (2673)
3610 - increase OST request buffers to 4096 on large machines (2729)
3611 - fix up permission of existing directories in simple_mkdir (2661)
3612 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3613 - don't assign transno to errored transactions (2742)
3614 - don't delete objects on OST if given a bogus objid from MDS (2751)
3615 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3616 - if rq_no_resend, then timeout request after recovery (2432)
3617 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3618 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3619 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3620 - MDS recovery shouldn't delete objects that it has given out (2730)
3621 - if enqueue arrives after completion, don't clobber LVB (2819)
3622 - don't unlock pages twice when trigger_group_io returns error (2814)
3623 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3624 - don't write pages to disk if there was an error (1450)
3625 - don't ping imports that have recovery disabled (2676)
3626 - take buffered bytes into account when balancing socknal conn (2817)
3627 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3628 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3629 - remove little-endian swabbing of llog records (1987)
3630 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3631 - retry reposting request buffers if they fail (1191)
3632 - grow extent at grant time to avoid granting a revoked lock (2809)
3633 - lock revoke doesn't evict page if covered by a second lock (2765)
3634 - disable VM readahead to avoid reading outside lock extents (2805)
3636 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3637 - updated kernel patches for hp-2.4.20 kernel (2681)
3639 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3642 - fix truncated write corruption (2366)
3643 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3645 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3646 - ensure len(uuid) < 37 in lmc (1171)
3647 - fix ia64 OOPS in llog_test (2255)
3648 - zero end of page at obdfilter for partial page writes (2648)
3649 - don't leave stale dentries around after renames (bug 2428)
3650 - fix timeouts when evicting a client with a single lock held (2642)
3651 - set deadline for the initial HELLO message to drain (2634)
3652 - print out dotted-quad IP addresses in the socknal (2302)
3654 - additional debugging for MDS client eviction problem (2443)
3655 - fix mkfsoptions support for osts (2603, 2604)
3657 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3660 - add series for the vanilla 2.6.0 kernel
3661 - add series for the vanilla 2.4.24 kernel
3662 - add series for a cray x86/64 UL kernel drop
3663 - fix xattr patches for the vanilla 2.4.19 series
3665 - generate true UUIDs in lmc (1171)
3666 - have portals stack dumping break in UML (2466)
3667 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3668 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3669 - wake callback waiting threads on client eviction (2460)
3670 - Add --ptldebug and --subsystem to lmc (1719)
3671 - update assertion to allow safe interrupt allocation
3672 - set rq_no_resend for cancel requests (2432)
3673 - recalculate ptlrpcd timeout after resend (2494)
3674 - call vfs_rmdir when removing pending directories (2368)
3675 - fix renaming a file to itself (2429)
3676 - lmc creates a default one-stripe lov (2454)
3677 - expand procfs space to handle large clusters (2326)
3678 - increase UML stack to avoid overflow
3679 - update lconf's list of debug and subsystem masks
3680 - fix lfs find --obd (2510)
3681 - /proc tunable for disabling filter read caching (2591)
3682 - stop rpm packages from altering slapd.conf (2301)
3683 - disable nagle in the socknal under 0conf (2578)
3684 - choose mds inode size based on stripe count (2572)
3685 - fix kernel-source rpm problems (2516)
3687 - add --disable-doc to avoid pdf generation (2421)
3688 - update documentation, tests, type-os, comments
3689 - avoid format warnings on ia64
3690 - remove the TOE NAL
3691 - tiny code cleanups by removing unused fields
3693 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3696 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3697 - remove the most blatant lies from BUILDING file (2371)
3698 - change default debug level to reasonable production setting
3699 - reduce client side cache size to reduce cache flush time
3700 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3701 - make TCP zerocopy and pinger support enabled by default (2476)
3702 - sync writes completed after process exits caused crashes (2319)
3703 - maintain correct mount count on the MDS (2356)
3704 - backout 1557, because 2316 wasn't really fixed
3705 - better file I/O statistics gathering in /proc
3706 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3707 - another kernel patch to fix zero-copy TCP function export
3708 - don't take duplicate lock when processing re-sent getattr (2420)
3709 - lctl uses obd_self_export instead of creating new conn (2353)
3710 - MDS/OST recovery case which requires object creation asserted (2425)
3711 - move lfs from /usr/sbin to /usr/bin in packages
3712 - fix race between mds_client_add and mds_client_free (2417)
3713 - use kmalloc instead of slabs in portals (2430)
3714 - don't create duplicate records when a failover MDS is present (2442)
3715 - remove unnecessary mount age check (2332)
3716 - don't remove directory inodes from locks prematurely (2451)
3717 - don't break if MDS service name is the same as hostname (2103)
3718 - fix races in client write RPC generation when cache full (2482)
3720 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3723 - remove now-unused request->rq_obd (278)
3724 - if an allocation fails, print out how much memory we've used (1933)
3725 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3726 - add the "configurable stack size" patch to most series files (1256)
3727 - ability to write large log records, for 100+ OST configs (2306)
3728 - fix NULL deref when filter_prep fails (2314)
3729 - fix operator precedence error in filter_sync
3730 - dynamic allocation of socknal TX descriptors (2315)
3731 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3732 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3733 - ensure that CWARN messages in Portals always get to the syslog
3734 - __init/__exit are not for prototype decls (ldlm_init/exit)
3735 - x86-64 compile warning fixes
3736 - fix gateway LMC keyword conflict (2318)
3737 - fix MDS lock inversions in getattr/reint paths (1844)
3738 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3739 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3740 - DLM race condition prevented some lock evictions (2328)
3741 - ENOMEM detection and retry on socknal sends (2230)
3742 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3743 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3744 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3745 - better error messages when a client is rejected during recovery (1505)
3746 - avoid cancelling locks which were never granted, after failure (2330)
3747 - fix i_sem/journal inversion in mds_client_add (2333)
3748 - fix truncate/getattr lock cycle deadlock (2334)
3749 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3750 - fix two rare exit paths which could leak an l_lock() ref (2321)
3751 - fencepost error in MDS/OST orphan recovery (2226)
3752 - make log record alignment 8 bytes (1988)
3753 - lstripe now fails when requested offset > ost_count (2237)
3754 - ensure that all kernel series have a complete list.h (1607)
3755 - fix crashes in special-file operations (2316)
3756 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3758 - allow configurable automake binary, for testing new versions
3759 - small update to the lfs documentation
3761 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3763 * fix negative export reference count in fsfilt_sync (2312)
3765 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3766 * release candidate 0.9.1
3768 - orphans are moved into the PENDING directory for possible recovery
3769 - replayed opens now open by fid for orphan/rename safety (1042)
3770 - last close of an orphan inode generates a transno (683)
3771 - chdir() and mount() now pin the directory entry (1020)
3772 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3773 - discard very old requests without processing them (1502)
3774 - remove some common, well-understood CERRORs (1505)
3775 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3776 - clear "grant" flags in OST replies until OST grant code lands (1644)
3777 - fix read performance by not clobbering i_blksize on client (1598)
3778 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3779 - make LRU size a /proc tunable, clears locks when reduced (707)
3780 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3781 - server locks take references on exports now (1558)
3782 - build fixes for 2.4.20-rh trees (1663)
3783 - return an error from lov_create if all OSCs are inactive (1751)
3784 - fix import levels when a reconnect happens without a timeout (1597)
3785 - exit early from mds_open if we get a lookup error (1749)
3786 - partial page read at EOF wouldn't wait for disk before sending (1642)
3787 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3788 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3789 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3790 - write/truncate lock inversion (1639)
3791 - Don't auto-load obdclass, portals modules during cleanup (1495)
3792 - fix timestamps from jumping to "now" (1763)
3793 - extra journal assertions (1648)
3794 - add an extra multiunlink test (1771)
3795 - fix read_record/write_record API (1776)
3796 - fix leak of offset_extent, possible incorrect i_size later (1772)
3797 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3798 - Debugging for the kqswnal_get_idle_tx problems (1820)
3799 - Allow recovery to be attempted multiple times (1536)
3800 - Write out MDS last_rcvd file after it is first created (1600)
3801 - Fix tx_descriptor leak in failed transmit situations (1827)
3802 - ext3 journaling fixes for assertion failure after IO error (1871)
3803 - class_export_put() on freed export after completion AST error (1896)
3804 - Fix revalidate looping in VFS (1322)
3805 - Don't access a freed export during MDS_REINT timeout (1521)
3806 - Add open-unlink recovery support on the MDS (1673,1764)
3807 - Return an error if no MDS data was read from last_rcvd (1946)
3808 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3809 - Don't setup a disk device that doesn't match exported UUID (317)
3810 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3811 - avoid committing NULL handle in force close
3812 - local.sh is now a one-stripe LOV configuration
3813 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3814 - fix ext3 htree duplicate directory entry corruption (1516)
3815 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3816 - update to new LOV EA format (2097)
3817 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3819 2003-06-15 Phil Schwan <phil@clusterfs.com>
3822 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3823 - per-import/export recovery handling (958, 931, 959)
3824 - multiple last-rcvd slots, for serving multiple FSes (949)
3825 - connections are again shared between multiple imp/exports (963, 964)
3826 - "umount -f" would hang if any requests needed to be sent (393, 978)
3827 - avoid pinning large req buffer by copying for queued messages (989)
3828 - add "uuid" to "lctl device" command to help upcalls (991)
3829 - "open" RPCs with transnos would confuse recovery counters (1037)
3830 - do proper endian conversion of all wire messages (288, 340, 891)
3831 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3832 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3833 - fix ptlrpc_connection leak in target_handle_connect (1174)
3834 - fix import refcounting bug in OST and MDS cleanup (1134)
3835 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3836 - fix very unlikely obd_types race condition (501)
3837 - remove osc_open hack for echo_client (1187)
3838 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3839 - a failure in read_inode2 leads to deadlock (1139)
3840 - cancel ack-locks as soon as transaction is committed (1072)
3841 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3842 - make sure to commitrw after any preprw to avoid deadlock (1162)
3843 - failing to execute a file in a lustre FS would lock inode (1203)
3844 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3845 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3846 - obd_uuid2tgt would walk off the end of the list (1255)
3847 - on IA64 the osc would give portals incorrect bulk size (1258)
3848 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3849 - fix lock inversion caused by new llite matching code (1282)
3850 - limit the number of dirty pages on a client to 10MB (1286)
3851 - timed out locks were not being corrected cancelled (1289)
3852 - fix O_DIRECT above 4GB on IA-32 (1292)
3853 * major user-visible changes
3854 - fail out/fail over policy now controlled by the upcall (993)
3856 - add OBD_PING to check server availability and failure (954)
3857 - lustre messages are now sent in sending host order (288, 340, 891)
3858 - add eadatalen to MDS getattr reply (340)
3859 - OST read replies may contain second buffer, with per-page status (593)
3861 2003-03-11 Phil Schwan <phil@clusterfs.com>
3864 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3865 - failed open()s could cause deadlock; fixed (867, 869)
3866 - stop cancelling OST locks when files are closed (481)
3867 - overlapping XID spaces caused network corruption (851, 853)
3868 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3869 - setattr_raw added, to do single-RPC, server-side setattrs
3870 - lmc/lconf syntax change for OST UUIDs
3871 - fix crashy race condition between ptlrpc_free_req and osc_close
3872 - don't use request in mdc_enqueue if we hit a timeout (889)
3873 - don't set the inode i_size for regular files from the MDS (896)
3874 - handle out of order completion AST (842)
3875 - don't LBUG if a lock request times out after receiving AST (913)
3876 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3877 - if a bad lock AST arrives, send an error instead of dropping entirely
3878 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3879 - fix leak in bulk IO when only partially completed (899, 900, 926)
3880 - fix O_DIRECT for ia64 (55)
3881 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3882 - C-z after timeout could hang a process forever; fixed (977)
3884 - client-side I/O cache (678, 924, 929, 941, 970)
3886 - READPAGE and SETATTRs which don't take server-side locks get
3889 2003-02-11 Phil Schwan <phil@clusterfs.com>
3892 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3893 - Fix fsfilt_extN_readpage() to read a full page of directory
3894 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3895 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3896 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3897 - Most intent operations were removed, in favour of a new RPC mode
3898 that does a single RPC to the server and bypasses most of the VFS
3899 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3900 - Aggressively cancel local locks on DLM servers
3901 - mds_reint_unlink sends EA to the client if it's the last nlink.
3902 client uses that EA to unlink OST objects.
3903 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3904 - recursive symlinks were fixed (439)
3905 - fixed NULL deref in DEBUG_REQ
3906 - filter_update_lastobjid no longer calls sync, which annoyed extN
3907 - fixed multi-client small-writes to a single file problem (445)
3908 - fixed mtime updates during file writes (607)
3909 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3910 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3911 - fixed Linux OST size reporting problem (444, 656)
3912 - OST now updates object mtime with writes or setattr (607, 619)
3913 - client verifies file size before zeroing page past EOF (445)
3914 - OST now writes last allocated objid to disk with allocation (108)
3915 - LOV on echo now works (409)
3917 - mds_reint_unlink sends a new buffer, with the EA included. this
3918 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3919 set if a regular file was being unlinked, and it was the last link
3920 - use PtlGet from the target for bulk writes (315)
3921 - OST now updates object mtime with writes or setattr (607, 619)
3922 - LDLM now has a grant-time callback to revalidate locked items, if
3924 - Many MDS operations were reorganized to combat race conditions
3926 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3927 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3929 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3932 - Fully reactivate OST imports after reconnection (512, others)
3933 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3934 - More graceful error handling for truncating on dead OST (515)
3935 - Don't error out unless we're actually accessing dead stripes (474)
3936 - Fix garbage sizes when stripes are missing (410)
3937 - LRU counters were broken, causing constant lock purge (433, 432)
3938 - garbage on read from stripes with failed OSTs (441)
3939 - mark OSCs as active before reconnecting during recovery (438)
3940 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3941 - lfind did not preserve OST order in output (443)
3942 - symlinks cause hung clients, incorrect data (439)
3943 - stop dereferencing request after dropping refcount (457)
3944 - don't LASSERT(spin_is_locked) on non-SMP (455)
3945 - fixes for many rename() bugs
3946 - fstat didn't correctly synchronize attributes (399)
3947 - server must handle lock cancellation during blocking AST prep (487)
3948 - bulk descriptors were free()d too soon (511)
3949 - fix paths in lconf, which would load incorrect modules (451, 507)
3950 - fix confusing lconf 'host not found' error message (386)
3951 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3952 - fix race condition in mdc_blocking_ast() for inode access (526)
3953 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3954 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3955 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3956 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3957 - fix lstripe in several ways (kernel side) (527)
3958 - fix request leak in ldlm_cli_enqueue (262)
3959 - incorrect OSC was marked inactive after OST failure
3960 - call mds_fs_cleanup before unmounting filesystem (524)
3961 - fix races between taking ns_lock and ldlm_lock_change_resource
3962 - fix races updating LOV export open file list
3963 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3964 - fix recovery NULL deref in ldlm_cli_cancel_unused
3965 - fix some DLM races by using new hash table for lock handles (419)
3966 - permit the client to specify desired inodes, at replay
3967 - duplicate requests when we queue them for replay reintegration
3968 - fix last_rcvd offset calculation
3969 - sync after each recovered transaction, so we always make progress
3970 - never, not always, ERESTART requests without transnos
3971 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
3973 - skip replay if the MDS says that the client is already connected
3974 - don't check for a recovery-enabled export to match lctl's UUID
3975 - don't INC_USE_COUNT for phantom exports
3976 - don't crash when cleaning up phantom exports (567)
3977 - don't double-finish or set replay data for errored mdc_open requests
3978 - abort requests when they time out, so we don't get old replies
3979 - send/receive replies for AST messages again
3980 - if the client says that it doesn't have the lock, cancel it on the
3982 - if we timeout during I/O, don't try to cancel an in-use lock; instead
3983 mark it as destroyed, it will all work out when decref is called
3984 - fix module use counts (22, 581)
3986 - ASTs now expect a reply (server cancels lock on error reply)
3988 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
3991 - fix many simultaneous client startup (392)
3992 - fix dentry->d_it clobbering
3993 - credentials weren't being shipped for readdir/getattr operations
3994 - remove invalid assertions triggered during some concurrent MD
3996 - proper Lustre versions added (336, 389)
3997 - fix memory leak for create error case (398)
3998 - fix LOV locking bug that would get cli/srv out of sync
3999 - fix echo client over LOV (409)
4000 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4001 - fix double-O_EXCL intent crash (424)
4002 - avoid sending multiple lock CANCELs (352)
4004 - MDS can do multi-client recovery (modulo bugs in new code)
4006 - many updates, edits, cleanups
4008 2002-11-18 Phil Schwan <phil@clusterfs.com>
4011 - fix null d_it dereference (346)
4012 - fix full OST/dbench hang (333)
4013 - fix permission problem with file removal (286)
4014 - fix removal of OSCs from LOV when they fail
4015 - fix NULL deref during bulk timeout (214)
4016 - fix problems related to multiple filesystems on one MDS (241)
4017 - fixed serious subtle metadata locking bugs
4018 - free locks on clients when inodes are removed due to memory
4020 - fix inode pointer in lock data (285)
4021 - partial support for multiple MDS on a single host (241)
4022 - data locks weren't cancelled at clear_inode time (290, 311)
4023 - intent locks could lead to unbounded lock growth (205)
4024 - added a maximum lock count, an LRU list, and a flusher
4025 - fix multiple rename (365)
4026 - properly abstracted the echo client
4027 - OSC locked 1 byte too many; fixed
4028 - rewrote brw callback code:
4029 - fixed recovery bugs related to LOVs (306)
4030 - fixed too-many-pages-in-one-write crash (191)
4031 - fixed (again) crash in sync_io_timeout (214)
4032 - probably fixed callback-related race (385)
4034 - Add capability to MDS protocol
4035 - LDLM cancellations and callbacks on different portals
4037 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4040 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4041 - defer open object destruction to close time (601981, 138)
4042 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4043 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4044 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4045 - don't call obd_setattr in ll_file_release for destroyed objects
4046 * protocol change to lustre_msg: move |version| and add |flags|
4047 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4048 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4049 * added replay of create, unlink, link and rename operations during
4050 MDS failover; recovery should be much more robust now
4051 * remove failed OSCs from LOVs (only lov_create uses this so far)
4052 * the lustre-HOWTO was brought (more) up to date (582544)
4054 2002-10-23 Phil Schwan <phil@clusterfs.com>
4057 - in-use dentries weren't being reused properly (617851)
4058 - prevent multiple LDLM setup (599178)
4059 - fix LOV size calculations for truncate (617853)
4060 - fix client handling of MDS intent errors (POSIX)
4061 - fix permission bug in lovstripe.c test (624321)
4062 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4063 - truncate past end of file could corrupt data
4064 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4065 - a race in recovery could return ETIMEDOUT to apps (623947)
4066 - building outside the source directory was fixed
4067 * the lustre-HOWTO was brought (more) up to date (582544)
4068 * major progress was made on recovery functionality
4070 2002-10-10 Phil Schwan <phil@clusterfs.com>
4073 - recovery deadlock fix
4074 - rm -rf causes LBUG fix (617817)
4075 - file open by multiple tasks fix (618962)
4076 - directory permissions bugs (602707 and 620007)
4077 - journal_stop fixed with locking (611313)
4078 - O_APPEND failures resolved (618273, perhaps 614459)
4079 - lconf PATH fix (619770)
4080 - IA64 build fix (621450)
4081 - RPC buffer sizes scale with amount of memory
4083 2002-10-01 Phil Schwan <phil@clusterfs.com>
4086 - locks would be cancelled without throwing away data pages,
4087 resulting in inconsistent data (605627)
4088 - inode attributes were not always being refreshed (605627, 612449)
4089 - lconf now continues to cleanup after lctl reports an error
4090 - MDS now enforces user permissions (602707)
4091 - lprocfs cleanup fixed, but not yet enabled (614157)
4092 - fixed infinite server hang, should a client not respond to an AST
4093 - avoid going into recovery if user calls readlink() with a buffer
4094 that's too small (613941)
4095 - AST RPCs no longer require replies (614867) -- this may be changed
4096 - don't crash server if client sends an IOV that's too big (611336)
4097 - fixed lock conversion deadlock (611892)
4098 - fixed the following of symlinks (614622)
4099 * recovery: the server can remove locks from a client that dies, other
4100 clients can make progress
4101 * more extN patch fixes
4102 * compile-time configurable ptlrpc buffer allocations
4104 - collaborative read cache document
4105 - Lustre Lite Performance CDR document-in-progress
4107 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4110 - fix typo in patch-2.4.18
4112 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4115 - clear ptlrpc request each time in handle_incoming_request()
4116 - unlink of files now destroys the object on the OST
4118 2002-09-19 Peter Braam <braam@clusterfs.com>
4120 * add hard link support
4121 * change obdfile creation method
4122 * kernel patch changed
4124 2002-09-19 Peter Braam <braam@clusterfs.com>
4127 - stack overflow bug in extN fixed
4129 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4131 * documentation updates
4132 - add man pages for config tools
4133 - update tests/README to describe testing with new config tools
4134 - finish metadata API descriptions
4135 * bug fixes and cleanups
4136 - statfs workaround for 16TB limit
4137 - LOV stripe allocation improved, can stripe on subset of OSTs
4138 - LOV file size/IO offset was wrong for files > 4GB in size
4139 - object EA data was being dropped, caused files to be unreadable
4140 - memory overflow with non-LOV OST caused memory corruption
4141 - fixed regression tests to work with new config tools, obdfilter
4142 - fixed bug when directory size became larger than 1 block
4143 - fixed bug (for single client case) when PWD was deleted
4144 - invalidate local directory pages when doing intent-based ops
4145 - avoid LDLM oops when lock callback contained bad data
4147 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4149 * documentation updates
4150 * bug fixes and cleanups
4151 - configuration tools
4154 - 64-bit compile warnings
4155 - 64-bit internal statfs data
4157 * test_brw on persistent OST devices
4159 * lprocfs (disabled)
4161 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4163 * documentation updates
4164 * bug fixes and cleanups
4165 * configuration tools
4167 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4170 * many small fixes to 0_5_4
4171 * io/network handling
4172 * thinkos in MDS operations
4174 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4177 * crucial basic fixes to 0.5.3
4178 * IOR, Iozone work over Elan
4181 2002-08-07 Phil Schwan <phil@clusterfs.com>
4182 * version 0_5_3, our first alpha
4183 * we use the new Portals iovs
4184 * documentation updates
4185 * bug fixes and cleanups
4186 * small changes in the DLM wire protocol
4188 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4189 * version 0_5_1 with some initial stability,
4190 * locking on MD and file I/O.
4191 * documentation updates
4192 * several bug fixes since 0.5.0
4193 * small changes in wire protocol
4195 2002-07-18 Phil Schwan <phil@clusterfs.com>
4197 * delivered as Lustre Light Alpha
4198 * fixed a crash after handling invalid MDS requests
4199 * fixed directory pages for architectures with non-4k pages sizes
4201 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4202 * release version v0_4_4
4203 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4205 * Fixes a number of interruption problems with OST operations.
4206 * Update documentation for portals header changes
4207 * Move all wire protocol structs/defines to lustre_idl.h
4208 * Fixes symlink length bug.
4209 * Add tcpdump to repository.
4211 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4212 * release version v0_4_3
4213 * Fixes statfs for inodes on extN.
4214 * Fixes bug in runtests which would delete /etc/hosts.
4215 * Use 64-bit object IDs wherever possible (not into VFS though)
4216 Remove ost_get_info, which is unused by lustre, and out of date.
4218 2002-07-03 Peter Braam <braam@clusterfs.com>
4219 * release version v0_4_2 Fixes a lookup error (type not passed)
4220 * move forward to head of Portals
4221 * move forward to latest Lustre kernel
4223 2002-06-25 Peter Braam <braam@clusterfs.com>
4224 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob