1 tbd Sun Microsystems, Inc.
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.el5 (RHEL 5).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.2-cfs5
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
15 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
16 Details : If an OST is down the MDS will hang indefinitely in
17 obd_statfs_rqset() waiting for the statfs data. While for
18 MDS QOS usage of statfs, it should not stuck in waiting.
20 Severity : enhancement
22 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
24 Severity : enhancement
26 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
30 Description: make mgs_setparam() handle fsname containing dash
31 Details : fsname containing a dash does not work with lctl conf_param
33 Severity : enhancement
35 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
37 Severity : enhancement
39 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
43 Frequency : rare, at shutdown
44 Description: access already free / zero obd_namespace.
45 Details : if client_disconnect_export was called without force flag set,
46 and exist connect request in flight, this can produce access to
47 NULL pointer (or already free pointer) when connect_interpret
48 store ocd flags in obd_namespace.
52 Frequency : only at startup
53 Description: not alloc memory with spinlock held.
54 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
57 Severity : enhancement
59 Description: make lustre randomly fail allocating memory
60 Details : Make lustre randomly failed allocating memory for testing purpose.
62 Severity : enhancement
64 Description: lost problems with lov objid file
65 Details : Fixes some scability and access to not inited memory problems
66 in work with lov objdid file.
71 Description: lfs find does not continue on file error
72 Details : Continue other files processing when a file/dir is absent.
76 Description: Inconsistent usage of lustre_pack_reply()
77 Details : Standardize the usage of lustre_pack_reply() such that it
78 always generate a CERROR on failure.
81 Frequency : occasional
83 Description: MDS hang or stay in waiting lock
84 Details : If client receive lock with CBPENDING flag ldlm need send lock
85 cancel as separate rpc, to avoid situation when cancel request
86 can't processed due all i/o threads stay in wait lock.
90 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
94 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
98 Description: Update to SLES9 kernel-2.6.5-7.287.3.
102 Description: Update to RHEL4 latest kernel.
104 Severity : enhancement
106 Description: Build SLES10 patchless client fails
107 Details : The configure was broken by run ./configure with
108 --with-linux-obj=.... argument for patchless client. When the
109 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
110 can't search header adequately. Use absolute path such as
111 -I($LINUX)/include instead.
115 Description: interrupt oig_wait produce painc on resend.
116 Details : brw_redo_request can be used for resend requests from ptlrpcd and
117 private set, and this produce situation when rq_ptlrpcd_data not
118 copyed to new allocated request and triggered LBUG on assert
119 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
120 ptlrpcd set if request is changed and can be safety changed to use
125 Description: LASSERT_{REQ,REP}SWAB macros are buggy
126 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
127 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
128 these macros and replace them with their difinitions instead.
132 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
133 Details : Update which_patch & target file for SLES10 latest kernel.
136 Description: Remove client patches from SLES 10 kernel.
137 Details : This causes SLES 10 clients to behave as patchless clients
138 even on a Lustre-patched (server) kernel.
140 Severity : enhancement
142 Description: self-adjustable client's lru lists
143 Details : use adaptive algorithm for managing client cached locks lru
144 lists according to current server load, other client's work
145 pattern, memory activities, etc. Both, server and client
146 side namespaces provide number of proc tunables for controlling
149 Severity : enhancement
151 Description: light-weight GSS support
152 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
153 and reduce performance overhead.
155 Severity : enhancement
157 Description: Linux keyring support
158 Details : Support using service of Linux keyring for Lustre GSS internal
159 context refresh/cache mechanism.
163 Description: Fix errors in lfs documentation
164 Details : Fixes man pages
168 Description: don't use GFP_* in generic Lustre code.
169 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
173 Description: obdclass is limited by single OBD_ALLOC(idarray)
174 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
178 Description: warnings with build patchless client with vanila 2.6.19 and up
179 Details : change old ctl_table style and replace ctl_table/ctl_table_header
180 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
184 Frequency : Only for SLES
185 Description: sanity_quota fail test_1
186 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
187 file, which makes TSTID[2] inunique.
191 Frequency : Only for SLES9
192 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
193 Details : Update target/ChangeLog/which_patch .
197 Frequency : Only for test_10 in sanity-quota.sh
198 Description: a bug in quota test code
199 Details : checking a test flag in a wrong place
203 Frequency : Only for test_13 in sanity-quota.sh
204 Description: a bug in quota test script
205 Details : shouldn't check overfull allocated quota
209 Description: jbd statistics
210 Details : Port older jbd statistics patch for sles10
212 Severity : enhancement
214 Description: Add printing inode info into message about error in writepage.
219 Description: reply_lock_interpret crash due to race with it and lock cancel.
220 Details : Do not replay locks that are being cancelled. Do not reference
221 locks by their address during replay, just by their handle.
225 Frequency : When flocks are used.
226 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
227 Details : Flock locks might destroy just granted lock if it could be merged
228 with another existing flock, this is done in completion handler,
229 so teach ldlm_cli_enquque_fini that this is a valid case for
235 Description: Oops in read and write path when failing to allocate lock.
236 Details : Check if lock allocation failed and return error back.
240 Description: lstripe command fails for valid OST index
241 Details : The stripe offset is compared to lov->desc.ld_tgt_count
242 instead of lov->desc.ld_active_tgt_count.
246 Description: sanity.sh failed test 103
247 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
252 Description: df doesn't work properly if diskfs blocksize != 4K
253 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
257 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
258 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
262 Description: obd_health_check_timeout too short
263 Details : set obd_health_check_timeout as 1.5x of obd_timeout
267 Description: llapi_file_create() does not allow some changes
268 Details : add llapi_file_open() that allows specifying the mode and
269 open flags, and also returns an open file handle.
273 Description: merge and cleanup kernel patches.
274 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
278 Description: Add journal checksum support.(Kernel part)
279 Details : The journal checksum feature adds two new flags i.e
280 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
281 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
282 indicates that the commit block contains the checksum for
283 the blocks described by the descriptor blocks. Now commit
284 record can be sent to disk without waiting for descriptor
285 blocks to be written to disk. This behavior is controlled
286 using JBD2_FEATURE_ASYNC_COMMIT flag.
290 Description: OSS needs mutliple precreate threads
291 Details : Add ability to start more than one create thread per OSS.
295 Description: Sanity.sh test_65k failed: setstripe should have succeeded
296 Details : execute lfs setstripe on client
300 Description: mds_obd_create error creating tmp object
301 Details : When the user sets quota on root, llog will be affected and can't
302 create files and write files.
305 Frequency : Always on ia64 patchless client, and possibly others.
307 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
308 Details : This allows the patchless client to be loaded on architectures
313 Description: RedHat Update kernel for RHEL5
314 Details : Add the kernel config file for RHEL5.
318 Description: RedHat Update kernel for RHEL5
319 Details : Modify the kernel config file more closer RHEL5.
323 Description: Build failure against Centos5 (RHEL5)
324 Details : Define PAGE_SIZE when it isn't present.
328 Description: test_7 on sanity-quota.sh can't be used on separate machines
329 Details : add facet to handle it
333 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
334 Details : with NFS, the anon dentry's parent was set to itself in
335 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
336 corresponding dentry other than use rec->ur_name.
338 Severity : enhancement
340 Description: lfs setstripe enhancement
341 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
345 Description: enable data checksumming by default
346 Details : enable checksum by default, allow --disable-checksum
347 configure option and "-o nochecksum" mount option. Checksums
348 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
352 Description: lustre support for RHEL5
353 Details : Add support for RHEL5.
357 Description: Client eviction due to failover config
358 Details : after a connection loss, the lustre client should attempt to
359 reconnect to the last active server first before trying the
360 other potential connections.
364 Description: when mds and osts use different quota unit(32bit and 64bit),
365 quota will be released repeatly.
366 Details : void sending multiple quota reqs to mds, which will keep the
367 status between the reqs.
371 Description: osts not allocated evenly to files
372 Details : change the condition to increase offset_idx
375 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
377 Description: Data corruption for OSTs that are formatted larger than 2TB
379 Details : When generating the bio request for lustre file writes the
380 sector number would overflow a temporary variable before being
381 used for the IO. The data reads correctly from Lustre (which
382 will overflow in a similar manner) but other file data or
383 filesystem metadata may be corrupted in some cases.
387 Description: Tune the kernel for good SCSI performance.
388 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
389 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
394 Description: rewrite ext2-derived code in obdclass/uuid.c
395 Details : rewrite inherited code (uuid parsing code from ext2 utils)
396 from scratch preserving functionality.
400 Description: "lfs find -obd UUID" prints directories
401 Details : "lfs find -obd UUID" will return all directory names instead
402 of just file names. It is incorrect because the directories
403 do not reside on the OSTs.
407 Description: use i_size_read and i_size_write in 2.6 port
408 Details : replace inode->i_size access with i_size_read/write()
411 Frequency : when removing large files
413 Description: scheduling issue during removal of large Lustre files
414 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
415 It causes scheduling issues when removing large files (17TB in the
419 Frequency : only with liblustre clients on XT3
421 Description: evictions taking too long
422 Details : allow llrd to evict clients directly on OSTs
425 Frequency : only on ppc
427 Description: /proc/fs/lustre/devices broken on ppc
428 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
429 We should make sure the type of 'index' is loff_t and then cast
430 explicitly as needed below. Do not assign an explicitly cast
434 Frequency : only for rhel5
436 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
437 Details : Modify the target file & which_kernel.
439 Severity : enhancement
441 Description: omit set fsid for export NFS
442 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
443 as FS_REQUIRES_DEV for avoid problems with generate fsid.
447 Frequency : Always, for kernels after 2.6.16
448 Description: Fix warning idr_remove called for id=.. which is not allocated.
449 Details : Last kernels save old s_dev before kill super and not allow
450 to restore from callback - restore it before call kill_anon_super.
454 Description: buffer overruns could theoretically occur
455 Details : llapi_semantic_traverse() modifies the "path" argument by
456 appending values to the end of the origin string, and a
457 overrun may occur. Adding buffer overrun check in liblustreapi.
461 Description: Fix error on 'ls .' at the top of the Lustre mount.
462 Details : Don't revalidate dentry if it is a root dentry.
466 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
467 Details : Modify vm-tunables-rhel4.patch.
471 Description: Kernel config for 2.6.18-vanilla.
472 Details : Modify targets/2.6-vanilla.target.in.
473 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
474 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
475 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
476 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
480 Description: improve handling recoverable errors
481 Details : if request processig with error which can be recoverable on server
482 request should be resend, otherwise page released from cache and
487 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
488 Details : Modify target file & which_patch.
489 A flaw was found in the IA32 system call emulation provided
490 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
491 value could be stored in the %RAX register, which could trigger an
492 out-of-bounds system call table access. An untrusted local user
493 could exploit this flaw to run code in the kernel
494 (ie a root privilege escalation). (CVE-2007-4573).
498 Description: O_DIRECT bypasses client statistics.
499 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
500 read_ahead_stats were updated. copied the stats section from
501 osc_send_oap_rpc() into async_internal().
505 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
509 Description: change order of libsysio includes
510 Details : '#include sysio.h' should always come before '#include xtio.h'
514 Description: handle "serious error: objid * already exists" more gracefully
515 Details : If LAST_ID value on disk is smaller than the objects existing in
516 the O/0/d* directories, it indicates disk corruption and causes an
517 LBUG(). If the object is 0-length, then we should use the existing
518 object. This will help to avoid a full fsck in most cases.
520 Severity : enhancement
522 Description: adapt the lustre_config script to support the upgrade case
523 Details : Add "-u" option for lustre_config script to support upgrading 1.4
524 server targets to 1.6 in parallel.
528 Description: To avoid grant space > avaible space when the disk is almost
529 full. Without this patch you might see the error "grant XXXX >
530 available" or some LBUG about grant, when the disk is almost
532 Details : In filter_check_grant, for non_grant cache write, we should
533 check the left space by if (*left > ungranted + bytes), instead
534 of (*left > ungranted), because only we are sure the left space
535 is enough for another "bytes", then the ungrant space should be
536 increase. In client, we should update cl_avail_grant only there
537 is OBD_MD_FLGRANT in the reply.
541 Description: Update RHEL 4 kernel to fix local root privilege escalation.
542 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
543 described in CVE-2007-4573. This problem could allow untrusted
544 local users to gain root access.
547 Frequency : when using O_DIRECT and quotas
549 Description: Incorrect file ownership on O_DIRECT output files
550 Details : block usage reported by 'lfs quota' does not take into account
551 files that have been written with O_DIRECT.
556 Description: touch file failed when fs is not full
557 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
558 otherwise we can get ENOSP while fs is not full.
562 Description: parallel lock callbacks
563 Details : Instead of sending blocking and completion callbacks as separated
564 requests, adding them to a set and sending in parallel.
567 Frequency : only for Cray XT3
568 Bugzilla : 12829/13455
569 Description: Changing primary group doesn't change the group lustre assigns to
571 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
572 overridden with the primary group provided by the group upcall,
573 whereas the supplied fsgid can be trusted if it is in the list of
574 supplementary groups returned by the group upcall.
576 Severity : enhancement
578 Description: Allow masking D_WARNING, D_ERROR messages from console
579 Details : Console messages can now be disabled via lnet.printk.
581 --------------------------------------------------------------------------------
583 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
585 * Support for kernels:
586 2.6.5-7.283 (SLES 9),
587 2.6.9-55.EL (RHEL 4),
588 2.6.16.46-0.14 (SLES 10),
589 2.6.18.8 vanilla (kernel.org)
590 * Client support for unpatched kernels:
591 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
592 2.6.16 - 2.6.22 vanilla (kernel.org)
593 * Due to recently discovered recovery problems, we do not recommend
594 using patchless RHEL 4 clients with this or any earlier release.
595 * Recommended e2fsprogs version: 1.39.cfs8
596 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
597 * Starting with this release, the ldiskfs backing filesystem required
598 by Lustre is now in its own package, lustre-ldiskfs. This package
599 should be installed. It is versioned separately from Lustre and
600 may be released separately in future.
604 Description: block reactivating mgc import until all deactivates complete
605 Details : Fix race when failing back MDT/MGS to itself (testing)
608 Frequency : at statup only
610 Description: mds_lov_synchronize race leads to various problems
611 Details : simultaneous MDT->OST connections at startup can cause the
612 sync to abort, leaving the OSC in a bad state.
614 Severity : enhancement
616 Description: add optional extra BUILD_VERSION info
617 Details : add a new environment variable (namely LUSTRE_VERS) which allows
618 to override the lustre version.
621 Frequency : 2.6.18 servers only
623 Description: ll_kern_mount() doesn't release the module reference
624 Details : The ldiskfs module reference count never drops down to 0
625 because ll_kern_mount() doesn't release the module reference.
630 Description: server LBUG when using old ost_num_threads parameter
631 Details : Accept the old ost_num_threads parameter but warn that it
632 is deprecated, and fix an off-by-one error that caused an LBUG.
637 Description: Transient SCSI error results in persistent IO issue
638 Details : iobuf->dr_error is not reinitialized to 0 between two
642 Frequency : sometimes when underlying device returns I/O errors
644 Description: OSTs not going read-only during write failures
645 Details : OSTs are not remounted read-only when the journal commit threads
646 get I/O errors because fsfilt_ext3 calls journal_start/stop()
647 instead of the ext3 wrappers.
651 Description: poor connect scaling with increasing client count
652 Details : Don't run filter_grant_sanity_check for more than 100 exports
653 to improve scaling for large numbers of clients.
656 Frequency : SLES10 only
658 Description: sanity-quota.sh quotacheck failed: rc = -22
659 Details : Quotas cannot be enabled on SLES10.
662 Frequency : liblustre clients only
664 Description: getdirentries does not give error when run on compute nodes
665 Details : getdirentries does not fail when the size specified as an argument
666 is too small to contain at least one entry
668 Severity : enhancement
670 Description: Add LNET router traceability for debug purposes
671 Details : If a checksum failure occurs with a router as part of the
672 IO path, the NID of the last router that forwarded the bulk data
673 is printed so it can be identified.
678 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
679 Details : Due to a race condition, liblustre clients were occasionally
682 Severity : enhancement
684 Description: lfs setstripe use optional parameters instead of postional
687 Severity : enhancement
689 Description: Nanosecond timestamp support for ldiskfs
690 Details : The on-disk ldiskfs filesystem has added support for nanosecond
691 resolution timestamps. There is not yet support for this at
692 the Lustre filesystem level.
695 Frequency : during server recovery
697 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
698 Details : request with rq_no_resend flag not awake l_wait_event if they get a
702 Frequency : nfs export on patchless client
704 Description: connectathon hang when test nfs export over patchless client
705 Details : Disconnected dentry cannot be found with lookup, so we do not need
706 to unhash it or make it invalid
709 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
710 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
711 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
716 Description: open req refcounting wrong on reconnect
717 Details : If reconnect happened between getting open reply from server and
718 call to mdc_set_replay_data in ll_file_open, we will schedule
719 replay for unreferenced request that we are about to free.
720 Subsequent close will crash in variety of ways.
721 Check that request is still eligible for replay in
722 mdc_set_replay_data().
727 Description: disable writes to filesystem when reading health_check file
728 Details : the default for reading the health_check proc file has changed
729 to NOT do a journal transaction and write to disk, because this
730 can cause reads of the /proc file to hang and block HA state
731 checking on a healthy but otherwise heavily loaded system. It
732 is possible to return to the previous behaviour during configure
733 with --enable-health-write.
735 Severity : enhancement
737 Description: 64-bit inode version
738 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
739 to the inode. This will be required for version-based recovery.
744 Description: MDS fails to start if a duplicate client export is detected
745 Details : in some rare cases it was possible for a client to connect to
746 an MDS multiple times. Upon recovery the MDS would detect this
747 and fail during startup. Handle this more gracefully.
749 Severity : enhancement
751 Description: Add -o localflock option to simulate old noflock
753 Details : This will achieve local-only flock/fcntl locks
759 Description: log_commit_thread vs filter_destroy race leads to crash
760 Details : Take import reference before releasing llog record semaphore
765 Description: Wrong request locking in request set processing
766 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
767 imp_delayed_list, in this place should be used imp_lock.
770 Frequency : when reconnection
772 Description: Grant Leak when osc reconnect to OST
773 Details : When osc reconnect ost, OST(filter) should check whether it
774 should grant more space to client by comparing fed_grant and
775 cl_avail_grant, and return the granted space to client instead
776 of "new granted" space, because client will call osc_init_grant
777 to update the client grant space info.
780 Frequency : when client reconnect to OST
782 Description: Grant Leak when osc do resend and replay bulk write
783 Details : When osc reconnect to OST, OST(filter)should clear grant info of
784 bulk write request, because the grant info while be sync between
785 OSC and OST when reconnect, and we should ignore the grant info
786 these of resend/replay write req.
791 Description: Grant space more than avaiable left space sometimes.
792 Details : When then OST is about to be full, if two bulk writing from
793 different clients came to OST. Accord the avaliable space of the
794 OST, the first req should be permitted, and the second one
795 should be denied by ENOSPC. But if the seconde arrived before
796 the first one is commited. The OST might wrongly permit second
797 writing, which will cause grant space > avaiable space.
800 Frequency : when client is evicted
802 Description: Grant might be wrongly erased when osc is evicted by OST
803 Details : when the import is evicted by server, it will fork another
804 thread ptlrpc_invalidate_import_thread to invalidate the
805 import, where the grant will be set to 0. While the original
806 thread will update the grant it got when connecting. So if
807 the former happened latter, the grant will be wrongly errased
808 because of this race.
813 Description: Checking Stale with correct fid
814 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
815 to check whether it is stale, because sometimes, we want the
816 enqueue happened anyway, and op_data.fid2 will not be initialized.
818 Severity : enhancement
820 Description: update patchless client
821 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
824 Frequency : only with 2.4 kernel
826 Description: random memory corruption
827 Details : size of struct ll_inode_info is to big for union inode.u and this
828 can be cause of random memory corruption.
833 Description: Memory leak in recovery
834 Details : Lov_mds_md was not free in an error handler in mds_create_object.
835 It should also check obd_fail before fsfilt_start, otherwise if
836 fsfilt_start return -EROFS,(failover mds during mds recovery).
837 then the req will return with repmsg->transno = 0 and rc = EROFS.
838 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
839 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
840 be freed no matter whether fsfilt_commit success or not.
843 Frequency : only with huge count clients
845 Description: Prevents from taking the superblock lock in llap_from_page for
847 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
848 for llap_from_page call in ll_removepage prevents from taking the
849 superblock lock for a soon died page.
854 Description: Not check open intent error before release open handle
855 Details : in some rare cases, the open intent error is not checked before
856 release open handle, which may cause
857 ASSERTION(open_req->rq_transno != 0), because it tries to release
858 the failed open handle.
863 Description: Set cat log bitmap only after create log success.
864 Details : in some rare cases, the cat log bitmap is set too early. and it
865 should be set only after create log success.
869 Description: Accessing a block bevice can re-enable I/O when Lustre is
870 tearing down a device.
871 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
876 Description: service threads may hog cpus when there are a lot of requests
877 Details : Insert cond_resched to give other threads a chance to use some CPU
882 Description: the cat log was not initialized in recovery
883 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
884 unlink log on mds will not be initialized until mds post
885 recovery. And also in mds post recovery, the unlink log will
886 initialization will be done asynchronausly, so there will be race
887 between add unlink log and unlink log initialization.
891 Description: brw_stats were being printed incorrectly
892 Details : brw_stats were being printed as log2 but all of them were not
893 recorded as log2. Also remove some code duplication arising from
894 filter_tally_{read,write}.
898 Frequency : rare, only in recovery.
899 Description: ASSERTION(req->rq_type != LI_POISON) failed
900 Details : imp_lock should be held while iterating over imp_sending_list for
901 prevent destroy request after get timeout in ptlrpc_queue_wait.
905 Description: replay-single.sh test 52 fails
906 Details : A lock's skiplist need to be cleanup when it being unlinked
907 from its resource list.
911 Description: Short directio read returns full requested size rather than
913 Details : Direct I/O operations should return actual amount of bytes
914 transferred rather than requested size.
916 Severity : enhancement
918 Description: metadata RPC reduction (e.g. for rm performance)
919 Details : decrease the amount of synchronous RPC between clients and servers
920 by canceling conflicing lock before the operation on the client
921 and packing thier handles into the main operation RPC to server.
923 Severity : enhancement
925 Description: Async OSC create to avoid the blocking unnecessarily.
926 Details : If a OST has no remain object, system will block on the creating
927 when need to create a new object on this OST. Now, ways use
928 pre-created objects when available, instead of blocking on an
929 empty osc while others are not empty. If we must block, we block
930 for the shortest possible period of time.
932 Severity : enhancement
934 Description: refine locking for avoid write wrong info into lov_objid file
935 Details : fix possible races with add new target and write/update data in
938 --------------------------------------------------------------------------------
940 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
945 Frequency : on some architectures
947 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
948 Details : Uninitialized flags sometimes cause configuration commands to
952 Frequency : patchless clients only
954 Description: missing __iget() symbol export
955 Details : The __iget() symbol export is missing. To avoid the need for
956 this on patchless clients the deathrow inode reaper is turned
957 off, and we depend on the VM to clean up old inodes. This
958 dependency was during via the fix for bug 12181.
960 --------------------------------------------------------------------------------
962 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
964 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
965 INTEROPERATE with older versions automatically. In many cases a
966 special upgrade step is needed. Please read the
967 user documentation before upgrading any part of a live system.
968 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
969 version will not interoperate with 1.6 betas before beta5 (1.5.95).
970 * WARNING: Lustre configuration and startup changes are required with
971 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
977 Severity : enhancement
979 Description: MountConf
980 Details : Lustre configuration is now managed via mkfs and mount
981 commands instead of lmc and lconf. New obd types (MGS, MGC)
982 are added for dynamic configuration management. See
983 https://mail.clusterfs.com/wikis/lustre/MountConf for
986 Severity : enhancement
988 Description: dynamic OST addition
989 Details : OSTs can now be added to a live filesystem
991 Severity : enhancement
993 Description: startup order invariance
994 Details : MDTs and OSTs can be started in any order. Clients only
995 require the MDT to complete startup.
997 Severity : enhancement
999 Description: parallel, asynchronous orphan cleanup
1000 Details : orphan cleanup is now performed in separate threads for each
1001 OST, allowing parallel non-blocking operation.
1003 Severity : enhancement
1005 Description: optimized stripe assignment
1006 Details : stripe assignments are now made based on ost space available,
1007 ost previous usage, and OSS previous usage, in order to try
1008 to optimize storage space and networking resources.
1010 Severity : enhancement
1012 Description: Permanently set tunables
1013 Details : All writable /proc/fs/lustre tunables can now be permanently
1014 set on a per-server basis, at mkfs time or on a live system.
1016 Severity : enhancement
1018 Description: Lustre message v2
1019 Details : Add lustre message format v2.
1021 Severity : enhancement
1023 Description: client OST exclusion list
1024 Details : Clients can be started with a list of OSTs that should be
1025 declared "inactive" for known non-responsive OSTs.
1029 Description: SPEC SFS validation failure on NFS v2 over lustre.
1030 Details : Changes the blocksize for regular files to be 2x RPC size,
1031 and not depend on stripe size.
1033 Severity : enhancement
1035 Description: Multiple MD RPCs in flight.
1036 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1037 To avoid overly-overloading MDS, introduce a limit on number of
1038 MDS RPCs in flight for a single client and add /proc controls
1039 to adjust this limit.
1041 Severity : enhancement
1043 Description: client read/write statistics
1044 Details : Add client read/write call usage stats for performance
1045 analysis of user processes.
1046 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1047 file access. extents_stats shows chunk size distribution.
1048 extents_stats_per_process show chunk size distribution per
1051 Severity : enhancement
1053 Description: mds statistics
1054 Details : Add detailed mds operations statistics in
1055 /proc/fs/lustre/mds/*/stats.
1059 Description: Failure of copying files with lustre special EAs.
1060 Details : Client side always return success for setxattr call for lustre
1061 special xattr (currently only "trusted.lov").
1066 Description: Refcount LNET uuids
1067 Details : The global LNET uuid list grew linearly with every startup;
1068 refcount repeated list entries instead of always adding to
1071 Severity : enhancement
1073 Description: Dynamic service threads
1074 Details : Within a small range, start extra service threads
1075 automatically when the request queue builds up.
1078 Frequency : mixed-endian client/server environments
1080 Description: mixed-endian crashes
1081 Details : The new msg_v2 system had some failures in mixed-endian
1084 Severity : enhancement
1086 Description: Easy OST removal
1087 Details : OSTs can be permanently deactivated with e.g. 'lctl
1088 conf_param lustre-OST0001.osc.active=0'
1090 Severity : enhancement
1092 Description: MGS proc entries
1093 Details : Added basic proc entries for the MGS showing what filesystems
1096 Severity : enhancement
1098 Description: provide MGS failover
1099 Details : Added config lock reacquisition after MGS server failover.
1101 Severity : enhancement
1103 Description: add Linux 2.4 support
1104 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1108 Description: a large application tries to do I/O to the same resource and dies
1109 in the middle of it.
1110 Details : Check the req->rq_arrival time after the call to
1111 ost_brw_lock_get(), but before we do anything about
1112 processing it & sending the BULK transfer request. This
1113 should help move old stale pending locks off the queue as
1114 quickly as obd_timeout.
1117 Frequency : when an incorrect nid is specified during startup
1119 Description: ptlrpc connect to non-existant node causes kernel crash
1120 Details : LNET can't be re-entered from an event callback, which
1121 happened when we expire a message after the export has been
1122 cleaned up. Instead, hand the zombie cleanup off to another
1125 Severity : enhancement
1127 Description: plain/inodebits lock performance improvement
1128 Details : Grouping plain/inodebits in granted list by their request modes
1129 and bits policy, thus improving the performance of search through
1133 Frequency : only if OST filesystem is corrupted
1135 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1136 Details : for a short time RPCs with bulk IO are in the replay list,
1137 but replay of bulk IOs is unimplemented. If the OST filesystem
1138 is corrupted due to disk cache incoherency and then replay is
1139 started it is possible to trip an assertion. Avoid putting
1140 committed RPCs into the replay list at all to avoid this issue.
1143 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1145 Description: System hang on startup
1146 Details : This bug allowed the liblustre (e.g. catamount) client to
1147 return to the app before handling all startup RPCs. This
1148 could leave the node unresponsive to lustre network traffic
1149 and manifested as a server ptllnd timeout.
1151 Severity : enhancement
1153 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1154 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1155 boolean to control whether to print peer debug info when a
1156 client's RPC times out.
1159 Frequency : only for kernels with patches from Lustre below 1.4.3
1161 Description: Remove old rdonly API
1162 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1165 Frequency : only for devices with external journals
1167 Description: Set external device read-only also
1168 Details : During a commanded failover stop, we set the disk device
1169 read-only while the server shuts down. We now also set any
1170 external journal device read-only at the same time.
1173 Frequency : when upgrading from 1.4 while trying to change parameters
1175 Description: The wrong (new) MDC name was used when setting parameters for
1176 upgraded MDT's. Also allows changing of OSC (and MDC)
1177 parameters if --writeconf is specified at tunefs upgrade time.
1180 Frequency : when setting specific ost indicies
1182 Description: QOS code breaks on skipped indicies
1183 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1184 created with --index need not be sequential.
1186 Severity : enhancement
1188 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1189 Details : The uninit_groups feature works in conjunction with the kernel
1190 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1191 up the pass1 processing of e2fsck. This is a read-only feature
1192 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1193 mount filesystems that have had this feature enabled.
1195 Severity : enhancement
1197 Description: Improve multi-block allocation algorithm to avoid fragmentation
1198 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1199 allocation locality and avoid filesystem fragmentation.
1201 ------------------------------------------------------------------------------
1203 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1205 * Support for kernels:
1206 2.4.21-47.0.1.EL (RHEL 3)
1207 2.6.5-7.283 (SLES 9)
1208 2.6.9-42.0.10.EL (RHEL 4)
1209 2.6.12.6 vanilla (kernel.org)
1210 2.6.16.27-0.9 (SLES 10)
1211 * Recommended e2fsprogs version: 1.39.cfs5
1213 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1217 Frequency : occasional, depends on client load and configuration
1218 Bugzilla : 12181, 12203
1219 Description: data loss for recently-modified files
1221 Details : In some cases it is possible that recently written or created
1222 files may not be written to disk in a timely manner (this should
1223 normally be within 30s unless client IO load is very high).
1224 The problem appears as zero-length files or files that are a
1225 multiple of 1MB in size after a client crash or client eviction
1226 that are missing data at the end of the file.
1228 This problem is more likely to be hit on clients where files are
1229 repeatedly created and unlinked in the same directory, clients
1230 have a large amount of RAM, have many CPUs, the filesystem has
1231 many OSTs, the clients are rebooted frequently, and/or the files
1232 are not accessed by other nodes after being written.
1234 The presence of the problem can be detected by looking at
1235 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1236 smaller than the second (nr_unused) then dirty files will not
1237 be flushed automatically to disk. "sync; sleep 10" should be
1238 run several times on the node before unmounting it to update
1239 Lustre (this is also safe to run on nodes without this problem).
1241 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1242 that can cause this same problem, so customers using that kernel
1243 also need to update the kernel in addition to Lustre. In order
1244 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1246 It is normal that files written just before a client crash (less
1247 than 30s) may not yet have been flushed to disk, even for local
1251 Frequency : frequent on thin XT3 nodes
1253 Description: UUID collision on thin XT3 Linux nodes
1254 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1255 randomly, since we relied on an insufficiently-seeded PRNG.
1260 Description: OSS hangs after "All ost request buffers busy"
1261 Details : A deadlock between quota and journal operations caused OSS
1262 hangs after printing "All ost request buffers busy."
1265 Frequency : always on liblustre builds
1267 Description: Cleanup compiler warnings on liblustre
1270 Frequency : always on liblustre builds on XT3
1272 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1276 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1277 > 32000 subdirectories
1278 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1279 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1280 subdirectory count crosses 32000. This will aid e2fsck to
1281 correctly handle more than 32000 subdirectories.
1284 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1286 Description: System hang on startup
1287 Details : This bug allowed the liblustre (e.g. catamount) client to
1288 return to the app before handling all startup RPCs. This
1289 could leave the node unresponsive to lustre network traffic
1290 and manifested as a server ptllnd timeout.
1292 Severity : enhancement
1294 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1295 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1296 boolean to control whether to print peer debug info when a
1297 client's RPC times out.
1302 Description: make O_SYNC working on 2.6 kernels
1303 Details : 2.6 kernels use different method for mark pages for write,
1304 so need add a code to lustre for O_SYNC work.
1309 Description: Failure to close file and release space on NFS
1310 Details : Put inode details into lock acquired in ll_intent_file_open.
1311 Use mdc_intent_lock in ll_intent_open to properly
1312 detect all kind of errors unhandled by mdc_enqueue.
1317 Description: proc file read during shutdown sometimes raced obd removal,
1319 Details : Add lock to prevent obd access after proc file removal.
1322 Frequency : Only for files larger than 4GB on 32-bit clients.
1324 Description: improperly doing page alignment of locks
1325 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1326 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1331 Description: RPCs being resent when they shouldn't be
1332 Details : Some RPCs that should not be resent are being resent. This
1333 can cause inconsistencies in the RPC state machine. Do not
1334 resend such requests.
1337 Frequency : rare, only with NFS export
1339 Description: Crash on NFS re-export node
1340 Details : under very unusual load conditions an assertion is hit in
1341 ll_intent_file_open()
1344 Frequency : only if OST filesystem is corrupted
1346 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1347 Details : for a short time RPCs with bulk IO are in the replay list,
1348 but replay of bulk IOs is unimplemented. If the OST filesystem
1349 is corrupted due to disk cache incoherency and then replay is
1350 started it is possible to trip an assertion. Avoid putting
1351 committed RPCs into the replay list at all to avoid this issue.
1356 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1357 Details : Large single O_DIRECT read and write calls can fail to allocate
1358 a sufficiently large buffer to process the request. In case of
1359 allocation failure the allocation is retried with a smaller
1360 buffer and broken into smaller requests.
1362 Severity : enhancement
1364 Description: Add -o localflock option to simulate old noflock behaviour.
1365 Details : This will achieve local-only flock/fcntl locks coherentness.
1370 Description: versioning check is incomplete
1371 Details : Checking the version difference of client vs. server, report
1372 error if the gap is too big.
1377 Description: add support PG_writeback bit
1378 Details : add support for PG_writeback bit for Lustre, for more carefull
1379 work with page cache in 2.6 kernel. This also fix some deadlocks
1380 and remove hack for work O_SYNC with 2.6 kernel.
1382 Severity : enhancement
1384 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1385 Details : The uninit_groups feature works in conjunction with the kernel
1386 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1387 up the pass1 processing of e2fsck. This is a read-only feature
1388 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1389 mount filesystems that have had this feature enabled.
1391 Severity : enhancement
1393 Description: Improve multi-block allocation algorithm to avoid fragmentation
1394 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1395 allocation locality and avoid filesystem fragmentation.
1397 ------------------------------------------------------------------------------
1399 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1401 * Support for kernels:
1402 2.6.9-42.0.3.EL (RHEL 4)
1403 2.6.5-7.276 (SLES 9)
1404 2.4.21-47.0.1.EL (RHEL 3)
1405 2.6.12.6 vanilla (kernel.org)
1406 2.6.16.21-0.8 (SLES10)
1407 * Recommended e2fsprogs version: 1.39.cfs2-0
1409 * The backwards-compatible /proc/sys/portals symlink has been removed
1410 in this release. Before upgrading, please ensure that you change
1411 any configuration scripts or /etc/sysctl.conf files that access
1412 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1413 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1414 in advance of the upgrade on any system running Lustre 1.4.6 or
1415 newer, since /proc/sys/lnet was added in that version.
1416 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1420 Frequency : only when quota is used
1422 Description: avoid scanning export list for quota master
1423 Details : Change the algorithms to avoid scanning export list in order
1424 to improve the efficiency.
1427 Frequency : MDS failover only, very rarely
1429 Description: "went back in time" messages on mds failover
1430 Details : The greatest transno may be lost when the current operation
1431 finishes with an error (transno==0) and the client's last_rcvd
1432 record is over-written. Save the greatest transno in the
1433 mds_last_transno for this case.
1436 Frequency : always for specific kernels and striping counts
1438 Description: client may get "Matching packet too big" without ACL support
1439 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1440 when trying to access files in certain configurations. The
1441 clients should in fact be denied when mounting because they do
1442 not understand ACLs.
1445 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1447 Description: many clients connecting with IO in progress causes connect timeouts
1448 Details : Avoid synchronous journal commits to avoid delays caused by many
1449 clients connecting/disconnecting when bulk IO is in progress.
1450 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1451 OST_IO_PORTAL to avoid delays behind potentially many pending
1455 Frequency : occasionally with multiple writers to a single file
1457 Description: shared writes to file may result in wrong size reported by stat()
1458 Details : Allow growing of kms when extent lock is cancelled
1461 Frequency : always with random mmap IO to multi-striped file
1463 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1464 Details : Only if the hole is at the end of OST object so that kms is too
1465 small. Fix is to increase kms accordingly in ll_nopage.
1468 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1470 Description: writes to a missing object would leak memory on the OST
1471 Details : If there is an inconsistency between the MDS and OST filesystems,
1472 such that the MDS references an object that doesn't exist, writes
1473 to that object will leak memory due to incorrect cleanup in the
1474 error handling path, eventually running out of memory on the OST.
1479 Description: Creating too long symlink causes lustre errors
1480 Details : Check symlink and name lengths before sending requests to MDS.
1483 Frequency : only if flock is enabled (not on by default)
1485 Description: posix locks not released on fd closure on 2.6.9+
1486 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1487 kernels, this caused such locks not to be released on fd close and
1488 then assertions on fs unmount about still used locks.
1491 Frequency : MDS failover only, very rarely
1493 Description: clients may get ASSERTION(granted_lock != NULL)
1494 Details : When request was taking a long time, and a client was resending
1495 a getattr by name lock request. The were multiple lock requests
1496 with the same client lock handle and
1497 mds_getattr_name->fixup_handle_for_resent_request found one of the
1498 lock handles but later failed with ASSERTION(granted_lock != NULL).
1503 Description: handle->h_buffer_credits > 0, assertion failure
1504 Details : h_buffer_credits is zero after truncate, causing assertion
1505 failure. This patch extends the transaction or creates a new
1509 Frequency : NFS re-export or patchless client
1510 Bugzilla : 11179, 10796
1511 Description: Crash on NFS re-export node (__d_move)
1512 Details : We do not want to hash the dentry if we don't have a lock.
1513 But if this dentry is later used in d_move, we'd hit uninitialised
1514 list head d_hash, so we just do this to init d_hash field but
1515 leave dentry unhashed.
1518 Frequency : NFS re-export or patchless client
1520 Description: NFS exports has problem with symbolic link
1521 Details : lustre client didn't properly install dentry when re-exported
1522 to NFS or running patchless client.
1525 Frequency : NFS re-export or patchless client
1527 Description: Various nfs/patchless fixes.
1528 Details : fixes reuse disconected alias for lookup process - this fixes
1529 warning "find_exported_dentry: npd != pd",
1530 fix permission error with open files at nfs.
1531 fix apply umask when do revalidate.
1534 Frequency : occasional
1536 Description: Crash on NFS re-export node
1537 Details : calling clear_page() on the wrong pointer triggered oops in
1538 generic_mapping_read().
1541 Frequency : rarely, using O_DIRECT IO
1543 Description: unaligned directio crashes client with LASSERT
1544 Details : check for unaligned buffers before trying any requests.
1547 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1549 Description: stale data returned from RAID cache
1550 Details : If only a small amount of IO is done to the RAID device before
1551 reading it again it is possible to get stale data from the RAID
1552 cache instead of reading it from disk.
1555 Frequency : always for sles10 kernel
1557 Description: sles10 support
1558 Details : ll_follow_link: compile fixes and using of nd_set_link
1559 under newer kernels.
1562 Frequency : depends on arch, kernel and compiler version, always on sles10
1565 Description: recursive or deep enough symlinks cause stack overflow
1566 Details : getting rid of large stack-allocated variable in
1570 Frequency : depends on hardware
1572 Description: lustre write performance loss in the SLES10 kernel
1573 Details : the performance loss is caused by using of write barriers in the
1574 ext3 code. The SLES10 kernel turns barrier support on by
1575 default. The fix is to undo that change for ldiskfs.
1577 ------------------------------------------------------------------------------
1579 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1581 * Support for kernels:
1582 2.6.9-42.0.3EL (RHEL 4)
1583 2.6.5-7.276 (SLES 9)
1584 2.4.21-47.0.1.EL (RHEL 3)
1585 2.6.12.6 vanilla (kernel.org)
1589 Frequency : quota enabled and large files being deleted
1591 Description: releasing more than 4GB of quota at once hangs OST
1592 Details : If a user deletes more than 4GB of files on a single OST it
1593 will cause the OST to spin in an infinite loop. Release
1594 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1599 Description: statfs data retrieved from /proc may be stale or zero
1600 Details : When reading per-device statfs data from /proc, in the
1601 {kbytes,files}_{total,free,avail} files, it may appear
1602 as zero or be out of date.
1605 Frequency : systems with MD RAID1 external journal devices
1607 Description: lconf's call to blkid is confused by RAID1 journal devices
1608 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1609 of returning all block devices that match the journal UUID.
1612 Frequency : always, for aggregate stripe size over 4GB
1614 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1615 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1616 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1619 Frequency : always if "lfs find" used on a local file/directory
1621 Description: "lfs find" segfaults if used on a local file/directory
1622 Details : The case where a directory component was not specified wasn't
1623 handled correctly. Handle this properly.
1626 Frequency : always on ppc64
1628 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1629 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1630 perform endianness conversion.
1633 Frequency : rarely (truncate to non-zero file size after write under load)
1634 Bugzilla : 10730, 10687
1635 Description: Files padded with zeros to next 4K multiple
1636 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1637 it is possible that files that are truncated to a non-zero size
1638 immediately after being written are filled with zero bytes beyond
1639 the truncated size. No file data is lost.
1641 Severity : enhancement
1643 Description: Allow recovery/failover for liblustre clients.
1644 Details : liblustre clients were unaware of failover configurations until
1647 Severity : enhancement
1649 Description: user file locks should fail when not mounting with flock option
1650 Details : Set up an error-returning stub in ll_file_operations.lock field
1651 to prevent incorrect behaviour when client is mounted without
1652 flock option. Also, set up properly f_op->flock field for
1656 Frequency : always on ia64
1658 Description: "lfs df" loops on printing out MDS statfs information
1659 Details : The obd_ioctl_data was not initialized and in some systems
1660 this caused a failure during the ioctl that did not return
1661 an error. Initialize the struct and return an error on failure.
1664 Frequency : SLES 9 only
1666 Description: Error of copying files with lustre special EAs as root
1667 Details : Client side always return success for setxattr call for lustre
1668 special xattr (currently only "trusted.lov").
1671 Frequency : rarely on clusters with both ia64+i386 clients
1673 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1674 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1675 clients are doing concurrent writes to the same file it is
1676 possible that the ia64 clients may LASSERT because the OST
1677 extent locks are not PAGE_SIZE aligned. Ensure that grown
1678 locks are always aligned on the request boundary.
1681 Frequency : specific use, occasional
1683 Description: Overwriting in use executable truncates on-disk binary image
1684 Details : If one node attempts to overwrite an executable in use by
1685 another node, we now correctly return ETXTBSY instead of
1686 truncating the file.
1688 Severity : enhancement
1690 Description: Async OSC create to avoid the blocking unnecessarily.
1691 Details : If a OST has no remain object, system will block on the creating
1692 when need to create a new object on this OST. Now, ways use
1693 pre-created objects when available, instead of blocking on an
1694 empty osc while others are not empty. If we must block, we block
1695 for the shortest possible period of time.
1700 Description: chmod on Lustre root is propagated to other clients
1701 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1702 invalid by the follow_mount time.
1705 Frequency : liblustre clients only
1707 Description: Race in 'instant cancel' lock handling could lead to such locks
1708 never to be granted in case of SMP MDS
1709 Details : Do not destroy not yet granted but cbpending locks in
1713 Frequency : replay/resend of open
1715 Description: non null lock assetion failure in mds_intent_policy
1716 Details : Trying to replay/resend lockless open requests resulted in
1717 mds_open() returning 0 with no lock. Now it sets a flag if
1718 a lock is going to be returned.
1720 Severity : enhancement
1722 Description: Checksum enhancements
1723 Details : New checksum enhancements allow for resending RPCs that failed
1726 Severity : enhancement
1728 Description: Tunables on number of dirty pages in cacche
1729 Details : Allow to set limit on number of dirty pages cached.
1734 Description: client crash on unmount - lock still has references
1735 Details : In some error handling cases it was possible to leak a lock
1736 reference on a client while accessing a file. This was not
1737 harmful to the client during operation, but would cause the
1738 client to crash when the filesystem is unmounted.
1741 Frequency : specific case, rare
1743 Description: ETXTBSY on mds though file not in use by client
1744 Details : ETXTBSY is no longer incorrectly returned when attempting to
1745 chmod or chown a directory that the user previously tried to
1746 execute or a currently-executing binary.
1749 Frequency : extremely rare except on liblustre-based clients
1751 Description: Lustre space not freed when files are deleted
1752 Details : Clean up open-unlinked files after client eviction. Previously
1753 the unlink was skipped and the files remained as orphans.
1758 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1759 Details : In some cases it was possible to send a blocking callback to
1760 a client doing a glimpse, even though that client didn't get
1761 a lock granted. When the glimpse lock is cancelled on the OST
1762 the freed lock is left on the waiting list and corrupted the list.
1765 Frequency : all core dumps
1767 Description: Broke core dumps to lustre
1768 Details : Negative dentry may be unhashed if parent does not have UPDATE
1769 lock, but some callers, e.g. do_coredump, expect dentry to be
1770 hashed after successful create, hash it in ll_create_it.
1772 ------------------------------------------------------------------------------
1774 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1776 * Support for kernels:
1777 2.6.9-42.0.2.EL (RHEL 4)
1778 2.6.5-7.276 (SLES 9)
1779 2.4.21-40.EL (RHEL 3)
1780 2.6.12.6 vanilla (kernel.org)
1784 Frequency : always on RHEL 3
1786 Description: Number of open files grows over time
1787 Details : The number of open files grows over time, whether or not
1788 Lustre is started. This was due to a filp leak introduced
1789 by one of our kernel patches.
1791 ------------------------------------------------------------------------------
1793 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1795 * Support for kernels:
1796 2.6.9-42.EL (RHEL 4)
1797 2.6.5-7.276 (SLES 9)
1798 2.4.21-40.EL (RHEL 3)
1799 2.6.12.6 vanilla (kernel.org)
1804 Bugzilla : 5719, 9635, 9792, 9684
1805 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1806 Details : If a server is under heavy load and cannot reply to new
1807 connection requests before the client resends the (re)connect,
1808 the connection handling code can behave badly if two service
1809 threads are concurrently handing separate (re)connections from
1810 the same client. Add better locking to the connection handling
1811 code, and ensure that only a single connection will be processed
1812 for a given client UUID, even if the lock is dropped.
1814 Severity : enhancement
1816 Description: add TCP zero-copy support to kernel
1817 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1818 sends if the hardware supports scatter-gather and checksumming.
1819 This allows socklnd to do client-write and server-read more
1820 efficiently and reduce CPU utilization from skbuf copying.
1823 Frequency : only if NFS exporting from client
1825 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1826 Details : If setting attributes on a file created under NFS that had
1827 never been opened it would be possible to oops the client
1828 if the file had no objects.
1831 Frequency : always for liblustre
1833 Description: liblustre client does MDS+OSTs setattr RPC for each write
1834 Details : When doing a write from a liblustre client, the client
1835 incorrectly issued an RPC to the MDS and each OST the file was
1836 striped over in order to update the timestamps. When writing
1837 with small chunks and many clients this could overwhelm the MDS
1838 with RPCs. In all cases it would slow down the write because
1839 these RPCs are unnecessary.
1841 Severity : enhancement
1843 Description: allow number of MDS service threads to be changed at module load
1844 Details : It is now possible to change the number of MDS service threads
1845 running. Adding "options mds mds_num_threads={N}" to the MDS's
1846 /etc/modprobe.conf will set the number of threads for the next
1847 time Lustre is restarted (assuming the "mds" module is also
1848 reloaded at that time). The default number of threads will
1849 stay the same, 32 for most systems.
1854 Description: OST crash if filesystem is unformatted or corrupt
1855 Details : If an OST is started on a device that has never been formatted
1856 or if the filesystem is corrupt and cannot even mount then the
1857 error handling cleanup routines would dereference a NULL pointer.
1862 Description: NULL pointer deref in llap_from_page.
1863 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1864 incorrect) mapping if the page was truncated/reclaimed while it was
1865 searched for. Check for this condition and skip such pages when
1866 doing readahead. Introduce extra check to llap_from_page() to
1867 verify page->mapping->host is non-NULL (so page is not anonymous).
1870 Frequency : Sometimes when using sys_sendfile
1872 Description: "page not covered by a lock" warnings from ll_readpage
1873 Details : sendfile called ll_readpage without right page locks present.
1874 Now we introduced ll_file_sendfile that does necessary locking
1875 around call to generic_file_sendfile() much like we do in
1879 Frequency : with certain MDS communication failures at client mount time
1881 Description: NULL pointer deref after failed client mount
1882 Details : a client connection request may delayed by the network layer
1883 and not be sent until after the PTLRPC layer has timed out the
1884 request. If the client fails the mount immediately it will try
1885 to clean up before the network times out the request. Add a
1886 reference from the request import to the obd device and delay
1887 the cleanup until the network drops the request.
1890 Frequency : occasionally during client (re)connect
1892 Description: assertion failure during client (re)connect
1893 Details : processing a client connection request may be delayed by the
1894 client or server longer than the client connect timeout. This
1895 causes the client to resend the connection request. If the
1896 original connection request is replied in this interval, the
1897 client may trip an assertion failure in ptlrpc_connect_interpret()
1898 which thought it would be the only running connect process.
1901 Frequency : only with obd_echo servers and clients that are rebooted
1903 Description: kernel BUG accessing uninitialized data structure
1904 Details : When running an obd_echo server it did not start the ping_evictor
1905 thread, and when a client was evicted an uninitialized data
1906 structure was accessed. Start the ping_evictor in the RPC
1907 service startup instead of the OBD startup.
1909 Severity : enhancement
1910 Bugzilla : 10193 (patchless)
1911 Description: Remove dependency on various unexported kernel interfaces.
1912 Details : No longer need reparent_to_init, exit_mm, exit_files,
1913 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1916 Frequency : rare (only users of deprecated and unsupported LDAP config)
1918 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1919 Details : LDAP apparently contains 'lustreName' attributes instead of
1920 'name'. A simple remapping of the name is sufficient.
1923 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1925 Description: waiting_locks_callback trips kernel BUG if client is evicted
1926 Details : Running with the dump_on_timeout debug flag turned on makes
1927 it possible that the waiting_locks_callback() can try to dump
1928 the Lustre kernel debug logs from an interrupt handler. Defer
1929 this log dumping to the expired_lock_main() thread.
1931 Severity : enhancement
1933 Description: Support NFS exporting on 2.6 kernels.
1934 Details : Implement non-rawops metadata methods for NFS server to use without
1935 changing NFS server code.
1938 Frequency : very rare (synthetic metadata workload only)
1940 Description: two racing renames might cause an MDS thread to deadlock
1941 Details : Running the "racer" program may cause one MDS thread to rename
1942 a file from being the source of a rename to being the target of
1943 a rename at exactly the same time that another thread is doing
1944 so, and the second thread has already enqueued these locks after
1945 doing a lookup of the target and is trying to relock them in
1946 order. Ensure that we don't try to re-lock the same resource.
1949 Frequency : only very large systems with liblustre clients
1951 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1952 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1953 all OSTs in parallel. This allows the eviction of stale liblustre
1954 clients to proceed much faster than if they were done in series,
1955 and also offers similar improvements for other set_info RPCs.
1960 Description: excessive CPU usage during initial read phase on client
1961 Details : During the initial read phase on a client, it would agressively
1962 retry readahead on the file, consuming too much CPU and impacting
1963 performance (since 1.4.5.8). Improve the readahead algorithm
1964 to avoid this, and also improve some other common cases (read
1965 of small files in particular, where "small" is files smaller than
1966 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1971 Description: MDS crash when receiving packet with unknown intent.
1972 Details : Do not LBUG in unknown intent case, just return -EFAULT
1974 Severity : enhancement
1975 Bugzilla : 9293, 9385
1976 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1977 Details : Do not serialize getattr (non-intent version) and statfs.
1980 Frequency : occasional, when OST network is overloaded/intermittent
1982 Description: client evicted by OST after bulk IO timeout
1983 Details : If a client sends a bulk IO request (read or write) the OST
1984 may evict the client if it is unresposive to its data GET/PUT
1985 request. This is incorrect if the network is overloaded (takes
1986 too long to transfer the RPC data) or dropped the OST GET/PUT
1987 request. There is no need to evict the client at all, since
1988 the pinger and/or lock callbacks will handle this, and the
1989 client can restart the bulk request.
1992 Frequency : Always when mmapping file with no objects
1994 Description: client crashes when mmapping file with no objects
1995 Details : Check that we actually have objects in a file before doing any
1996 operations on objects in ll_vm_open, ll_vm_close and
2002 Description: Request leak when working with deleted CWD
2003 Details : Introduce advanced request refcount tracking for requests
2004 referenced from lustre intent.
2006 Severity : Enhancement
2008 Description: Cache open file handles on client.
2009 Details : MDS now will return special lock along with openhandle, if
2010 requested and client is allowed to hold openhandle, even if unused,
2011 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2012 closing files for every read/write openration.
2014 Severity : Enhancement
2016 Description: Cache open negative dentries on client when possible.
2017 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2018 negative dentries on lock revocation.
2023 Description: Remounting a client read-only wasn't possible with a zconf mount
2024 Details : It wasn't possible to remount a client read-only with llmount.
2026 Severity : enhancement
2027 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2028 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2029 MPICH is included to add Lustre-specific ADIO interfaces.
2030 This is based closely on the UFS ADIO layer and only differs
2031 in file creation, in order to allow the OST striping to be set.
2032 This is user-contributed code and not supported by CFS.
2037 Description: extended inode attributes (immutable, append-only) work improperly
2038 when 2.4 and 2.6 kernels are used on client/server or vice versa
2039 Details : Introduce kernel-independent values for these flags.
2041 Severity : enhancement
2044 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2045 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2046 granularity. Now, Fractional MB granularity is be supported,
2047 this is very useful for low memory system.
2049 Severity : enhancement
2051 Description: Getattr by fid
2052 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2053 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2054 have locks and avoids getattr rpc requests for it on every stat.
2057 Frequency : Always, for filesystems larger than 2TB
2059 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2060 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2061 filesystems larger than 2TB to be mounted with mballoc enabled.
2064 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2066 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2067 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2068 filesystems larger than 2TB is mounted then the kernel will
2069 silently corrupt the start of the filesystem. CONFIG_LBD is
2070 enabled for all CFS-supported kernels, but the possibility of
2071 this happening with a modified kernel config exists.
2073 Severity : enhancement
2075 Description: add client O_DIRECT support for 2.6 kernels
2076 Details : It is now possible to do O_DIRECT reads and writes to files
2077 in the Lustre client mountpoint on 2.6 kernel clients.
2079 Severity : enhancement
2081 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2082 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2083 parallel, not waiting for response from every OST before sending
2084 a rpc to the next OST.
2089 Description: setattr vs write race when updating file timestamps
2090 Details : Client processes that update a file timestamp into the past
2091 right after writing to the file (e.g. tar) it is possible that
2092 the updated file modification time can be reset to the current
2093 time due to a race between processing the setattr and write RPC.
2095 Severity : enhancement
2097 Description: Bring 'lfs find' closer in line with regular Linux find.
2098 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2099 -print0 options and obtains all the needed info through the lustre
2102 Severity : enhancement
2104 Description: support up to 1024 configured devices on one node
2105 Details : change obd_dev array from statically allocated to dynamically
2106 allocated structs as they are first used to reduce memory usage
2111 Description: Flush dirty partially truncated pages during truncate
2112 Details : Immediatelly flush partially truncated pages in filter_setattr,
2113 this way we completely avoid having any pages in page cache on OST
2114 and can retire ugly workarounds during writes to flush such pages.
2119 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2120 Details : protect inode from truncation within vfs_unlink() context
2121 just take a reference before calling vfs_unlink() and release it
2122 when parent's i_sem is free.
2127 Description: last_id value checked outside lock on OST caused LASSERT failure
2128 Details : If there were multiple MDS->OST object precreate requests in
2129 flight, it was possible that the OST's last object id was checked
2130 outside a lock and incorrectly tripped an assertion. Move checks
2131 inside locks, and discard old precreate requests.
2134 Frequency : always, if extents are used on OSTs
2136 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2137 Details : OSTs using the extents format would not zero the high 16 bits of
2138 the index physical block number. This is not a problem for any
2139 OST filesystems smaller than 16TB, and no kernels support ext3
2140 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2141 new/modified files) and can be fixed for existing filesystems
2142 with e2fsprogs-1.39-cfs1.
2147 Description: import connection selection may be incorrect if timer wraps
2148 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2149 import connections to be ignored if the 32-bit jiffies counter
2150 wraps. Use a 64-bit jiffies counter.
2153 Frequency : very large clusters immediately after boot
2155 Description: LNET request buffers exhausted under heavy short-term load
2156 Details : If a large number of client requests are generated on a service
2157 that has previously never seen so many requests it is possible
2158 that the request buffer growth cannot keep up with the spike in
2159 demand. Instead of dropping incoming requests, they are held in
2160 the LND until the RPC service can accept more requests.
2163 Frequency : Sometimes during replay
2165 Description: Assertion failure in ll_local_open after replay.
2166 Details : If replay happened on an open request reply before we were able
2167 to set replay handler, reply will become not swabbed tripping the
2168 assertion in ll_local_open. Now we set the handler right after
2169 recognising of open request
2172 Frequency : very rare
2174 Description: kernel reports "badness in vsnprintf"
2175 Details : Reading from the "recovery_status" /proc file in small chunks
2176 may cause a negative length in lprocfs_obd_rd_recovery_status()
2177 call to vsnprintf() (which is otherwise harmless). Exit early
2178 if there is no more space in the output buffer.
2180 Severity : enhancement
2182 Description: clear OBD RPC statistics by writing to them
2183 Details : It is now possible to clear the OBD RPC statistics by writing
2184 to the "stats" file.
2189 Description: Client mtime is not the same on different clients after utimes
2190 Details : In some cases, the client was using the utimes() syscall on
2191 a file cached on another node. The clients now validate the
2192 ctime from the MDS + OSTs to determine which one is right.
2197 Description: Inability to activate failout mode
2198 Details : lconf script incorrectly assumed that in pythong string's numeric
2199 value is used in comparisons.
2202 Frequency : always with multiple stripes per file
2204 Description: Inefficient object allocation for mutli-stripe files
2205 Details : When selecting which OSTs to stripe files over, for files with
2206 a stripe count that divides evenly into the number of OSTs,
2207 the MDS is always picking the same starting OST for each file.
2208 Return the OST selection heuristic to the original design.
2213 Description: mount failures may take full timeout to return an error
2214 Details : Under some heavy load conditions it is possible that a
2215 failed mount can wait for the full obd_timeout interval,
2216 possibly several minutes, before reporting an error.
2217 Instead return an error as soon as the status is known.
2219 Frequency : quota enabled and large files being deleted
2221 Description: releasing more than 4GB of quota at once hangs OST
2222 Details : If a user deletes more than 4GB of files on a single OST it
2223 will cause the OST to spin in an infinite loop. Release
2224 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2229 Description: statfs data retrieved from /proc may be stale or zero
2230 Details : When reading per-device statfs data from /proc, in the
2231 {kbytes,files}_{total,free,avail} files, it may appear
2232 as zero or be out of date.
2235 Frequency : systems with MD RAID1 external journal devices
2237 Description: lconf's call to blkid is confused by RAID1 journal devices
2238 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2239 of returning all block devices that match the journal UUID.
2242 Frequency : always, for aggregate stripe size over 4GB
2244 Description: assertion fails when trying to use 4GB stripe size
2245 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2246 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2249 Frequency : always on ppc64
2251 Description: the first write on an ext3 filesystem with mballoc got stuck
2252 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2253 perform endianness conversion.
2255 ------------------------------------------------------------------------------
2257 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2259 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2260 INTEROPERATE with older versions automatically. Please read the
2261 user documentation before upgrading any part of a live system.
2262 * WARNING: Lustre networking configuration changes are required with
2263 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2266 * Support for kernels:
2267 2.6.9-22.0.2.EL (RHEL 4)
2268 2.6.5-7.244 (SLES 9)
2269 2.6.12.6 vanilla (kernel.org)
2272 Severity : enhancement
2273 Bugzilla : 7981/8208
2274 Description: Introduced Lustre Networking (LNET)
2275 Details : LNET is new networking infrastructure for Lustre, it includes
2276 a reorganized network configuration mode (see the user
2277 documentation for full details) as well as support for routing
2278 between different network fabrics. Lustre Networking Devices
2279 (LNDs) for the supported network fabrics have also been
2280 created for this new infrastructure.
2282 Severity : enhancement
2283 Description: Introduced Access control lists
2284 Details : clients can set ACLs on files and directories in order to have
2285 more fine-grained permissions than the standard Unix UGO+RWX.
2286 The MDS must be started with the "-o acl" mount option.
2288 Severity : enhancement
2289 Description: Introduced filesystem quotas
2290 Details : Administrators may now establish per-user quotas on the
2293 Severity : enhancement
2295 Description: Configuration change for the XT3
2296 The PTLLND is now used to run Lustre over Portals on the XT3
2297 The configure option(s) --with-cray-portals are no longer used.
2298 Rather --with-portals=<path-to-portals-includes> is used to
2299 enable building on the XT3. In addition to enable XT3 specific
2300 features the option --enable-cray-xt3 must be used.
2305 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2306 Details : A race between allocating a new llcd and re-getting the llcd_lock
2307 allowed another thread to grab newly-allocated llcd.
2309 Severity : enhancement
2311 Description: 2.6 OST async journal commit and locking fix to improve performance
2312 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2313 2.6 kernels are now async as they already were in 2.4 kernels so
2314 that they can commit concurrently with the network bulk transfer.
2315 For block-allocated files the filter allocation semaphore is held
2316 to avoid filesystem fragmentation during allocation. BKL lock
2317 removed for 2.6 xattr operations where it is no longer needed.
2322 Description: lconf incorrectly determined whether two IP networks could talk
2323 Details : In some more complicated routing and multiple-network
2324 configurations, lconf will avoid trying to make a network
2325 connection to a disjoint part of the IP space. It was doing the
2326 math incorrectly for one set of cases.
2331 Description: Fix for potential infinite loop processing records in an llog.
2332 Details : If an llog record is corrupted/zeroed, it is possible to loop
2333 forever in llog_process(). Validate the llog record length
2334 and skip the remainder of the block on error.
2337 Frequency : occasional (liblustre only)
2339 Description: liblustre could not open files whose last component is a symlink
2340 Details : sysio_path_walk() would incorrectly pass the open intent to
2341 intermediate path components.
2344 Frequency : rare (liblustre only with non-standard tuning)
2345 Bugzilla : 7201 (7350)
2346 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2347 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2348 after it has been released, only do this for referenced locks
2350 Severity : enhancement
2352 Description: specifying an (invalid) directory default stripe_size of -1
2353 would reset the directory default striping
2354 Details : stripe_size -1 was used internally to signal directory stripe
2355 removal, now use "all default" to signal dir stripe removal
2356 as a directory striping of "all default" is not useful
2359 Frequency : common for large clusters running liblustre clients
2361 Description: doing an ls when liblustre clients are running is slow
2362 Details : sending a glimpse AST to a liblustre client waits for every AST
2363 to time out, as liblustre clients will not respond. Since they
2364 cannot cache data we refresh the OST lock LVB from disk instead.
2366 Severity : enhancement
2368 Description: doing an ls at the same time as file IO can be slow
2369 Details : enqueue and other "small" requests can be blocked behind many
2370 large IO requests. Create a new OST IO portal for non-IO
2371 requests so they can be processed faster.
2374 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2376 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2377 created file with mode 000
2378 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2379 allow an ftruncate() as a non-root user to a file with mode 000.
2380 Lustre now respects this flag to disable mode checks when
2381 truncating a file owned by the user
2384 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2387 Description: Revoking locks from clients that went dead or catatonic might take
2389 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2390 cancellation of such locks instant on servers without waiting for
2391 any reply from clients. Clients drops these locks when cancel
2392 notification from server is received without replying.
2395 Frequency : liblustre-only, when liblustre client dies or becomes busy
2397 Description: Doing ls on Linux clients can take a long time with active
2399 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2400 granting such locks to it in the first place if possible. Locks
2401 are taken by proxy on the OST during the read or write and
2402 dropped immediately afterward. Add connect flags handling, do
2403 not grant locks to liblustre clients for glimpse ASTs.
2405 Severity : enhancement
2407 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2408 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2409 sequential reads and behaves badly with applications doing random
2410 reads. Improve it by reading ahead at least read region, and
2411 avoiding excessive large RPC for small reads.
2413 Severity : enhancement
2415 Description: Creating more than 1000 files for a single job may cause a load
2416 imbalance on the OSTs if there are also a large number of OSTs.
2417 Details : qos_prep_create() uses an OST index reseed value that is an
2418 even multiple of the number of available OSTs so that if the
2419 reseed happens in the middle of the object allocation it will
2420 still utilize the OSTs as uniformly as possible.
2425 Description: OST or MDS may oops in ping_evictor_main()
2426 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2427 but doesn't restart at beginning of obd_exports_timed list
2430 Severity : enhancement
2432 Description: improve by-nid export eviction on the MDS and OST
2433 Details : allow multiple exports with the same NID to be evicted at one
2434 time without re-searching the exports list.
2437 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2439 Description: MDS may oops in groups_free()
2440 Details : in rare race conditions a newly allocated group_info struct is
2441 freed again, and this can be NULL. The 2.4 compatibility code
2442 for groups_free() checked for a NULL pointer, but 2.6 did not.
2445 Frequency : common for liblustre clients doing little filesystem IO
2446 Bugzilla : 9352, 7313
2447 Description: server may evict liblustre clients accessing contended locks
2448 Details : if a client is granted a lock or receives a completion AST
2449 with a blocking AST already set it would not reply to the AST
2450 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2453 Frequency : lfs setstripe, only systems with more than 160 OSTs
2455 Description: unable to set striping with a starting offset beyond OST 160
2456 Details : llapi_create_file() incorrectly limited the starting stripe
2457 index to the maximum single-file stripe count.
2460 Frequency : LDAP users only
2462 Description: lconf did not handle in-kernel recovery with LDAP properly
2463 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2465 Severity : enhancement
2467 Description: bind OST threads to NUMA nodes to improve performance
2468 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2469 node and do their allocations there to localize memory access
2471 Severity : enhancement
2473 Description: llmount can determine client NID directly from Myrinet (GM)
2474 Details : the client NID code from gmnalnid was moved directly into
2475 llmount, removing the need to use this or specifying the
2476 client NID explicitly when mounting GM clients with zeroconf
2479 Frequency : if client is started with down MDS
2481 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2482 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2483 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2484 ptlrpc_import_delayed_req() because we didn't check if the
2485 request was interrupted, and we also didn't break out of the
2486 event loop if there was no timeout
2491 Description: data loss during non-page-aligned writes to a single file from
2492 both multiple nodes and multiple threads on one node at same time
2493 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2494 inconsistency led to false short-reads, that were cached and later
2495 used by ->prepare_write() to fill in partially written page,
2496 leading to data loss.
2499 Frequency : always, if lconf --abort_recovery used
2501 Description: lconf --abort_recovery fails with 'Operation not supported'
2502 Details : lconf was attempting to abort recovery on the MDT device and not
2505 Severity : enhancement
2507 Description: remove cleanup logs
2508 Details : replace lconf-generated cleanup logs with lustre internal
2509 cleanup routines. Eliminates the need for client-cleanup and
2512 Severity : enhancement
2514 Description: add support for EAs (user and system) on lustre filesystems
2515 Details : it is now possible to store extended attributes in the Lustre
2516 client filesystem, and with the user_xattr mount option it
2517 is possible to allow users to store EAs on their files also
2519 Severity : enhancement
2521 Description: Add possibility (config option) to show minimal available OST free
2523 Details : When compiled with --enable-mindf configure option, statfs(2)
2524 (and so, df) will return least minimal free space available from
2525 all OSTs as amount of free space on FS, instead of summary of
2526 free spaces of all OSTs.
2528 Severity : enhancement
2530 Description: do not expand extent locks acquired on OST-side
2531 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2532 by server: they are not cached anyway.
2535 Frequency : when mmap is used/binaries executed from Lustre
2537 Description: Unmmap pages before throwing them away from read cache.
2538 Details : llap_shrink cache now attempts to unmap pages before discarding
2539 them (if unmapping failed - do not discard). SLES9 kernel has
2540 extra checks that trigger if this unmapping is not done first.
2545 Description: lconf didn't resolve symlinks before checking to see whether a
2546 given mountpoint was already in use
2549 Frequency : when migrating failover services
2550 Bugzilla : 6395, 9514
2551 Description: When migrating a subset of services from a node (e.g. failback
2552 from a failover service node) the remaining services would
2553 time out and evict clients.
2554 Details : lconf --force (implied by --failover) sets the global obd_timeout
2555 to 5 seconds in order to quickly disconnect, but this caused
2556 other RPCs to time out too quickly. Do not change the global
2557 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2559 Severity : enhancement
2560 Frequency : if MDS is started with down OST
2561 Bugzilla : 9439,5706
2562 Description: Allow startup/shutdown of an MDS without depending on the
2563 availability of the OSTs.
2564 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2565 Add appropriate locking and lov-osc refcounts for safe
2566 cleaning. Add osc abort_inflight calls in case the
2567 synchronize never started.
2570 Frequency : occasional (Cray XT3 only)
2572 Description: root not authorized to access files in CRAY_PORTALS environment
2573 Details : The client process capabilities were not honoured on the MDS in
2574 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2575 been accessed by an authorized user then root was able to access
2576 the file on the local client also. The root user capabilities
2577 are now allowed on the MDS, as this environment has secure UID.
2580 Frequency : occasional
2582 Description: ldiskfs "too long searching" message happens too often
2583 Details : A debugging message (otherwise harmless) prints too often on
2584 the OST console. This has been reduced to only happen when
2585 there are fragmentation problems on the filesystem.
2590 Description: Division by zero in statfs when all OSCs are inactive
2591 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2592 lov_statfs divides by value returned by lov_get_stripecnt().
2596 Bugzilla : 9489, 3273
2597 Description: First write from each client to each OST was only 4kB in size,
2598 to initialize client writeback cache, which caused sub-optimal
2599 RPCs and poor layout on disk for the first writen file.
2600 Details : Clients now request an initial cache grant at (re)connect time
2601 and so that they can start streaming writes to the cache right
2602 away and always do full-sized RPCs if there is enough data.
2603 If the OST is rebooted the client also re-establishes its grant
2604 so that client cached writes will be honoured under the grant.
2609 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2610 Details : Now I/O RPCs go to different portal number and (presumably) fast
2611 lock requests (and glimses) and other RPCs get their own service
2612 threads pool that should be able to service those RPCs
2615 Severity : enhancement
2617 Description: Ability to exchange lustre version between client and servers and
2618 issue warnings at client side if client is too old. Also for
2619 liblustre clients there is ability to refuse connection of too old
2621 Details : New 'version' field is added to connect data structure that is
2622 filled with version info. That info is later checked by server and
2626 Frequency : rare, liblustre only.
2627 Bugzilla : 9296, 9581
2628 Description: Two simultaneous writes from liblustre at offset within same page
2629 might proceed at the same time overwriting eachother with stale
2631 Details : I/O lock withing llu_file_prwv was released too early, before data
2632 actually was hitting the wire. Extended lock-holding time until
2633 server acknowledges receiving data.
2636 Frequency : extremely rare. Never observed in practice.
2638 Description: avoid generating lustre_handle cookie of 0.
2639 Details : class_handle_hash() generates handle cookies by incrementing
2640 global counter, and can hit 0 occasionaly (this is unlikely, but
2641 not impossible, because initial value of cookie counter is
2642 selected randonly). Value of 0 is used as a sentinel meaning
2643 "unassigned handle" --- avoid it. Also coalesce two critical
2644 sections in this function into one.
2646 Severity : enhancement
2648 Description: allow liblustre clients to delegate truncate locking to OST
2649 Details : To avoid overhead of locking, liblustre client instructs OST to
2650 take extent lock in ost_punch() on client's behalf. New connection
2651 flag is added to handle backward compatibility.
2653 Severity : enhancement
2654 Bugzilla : 4928, 7341, 9758
2655 Description: allow number of OST service threads to be specified
2656 Details : a module parameter allows the number of OST service threads
2657 to be specified via "options ost ost_num_threads={N}" in the
2658 OSS's /etc/modules.conf or /etc/modprobe.conf.
2662 Bugzilla : 6146, 9635, 9895
2663 Description: servers crash with bad pointer in target_handle_connect()
2664 Details : In rare cases when a client is reconnecting it was possible that
2665 the connection request was the last reference for that export.
2666 We would temporarily drop the export reference and get a new
2667 one, but this may have been the last reference and the export
2668 was just destroyed. Get new reference before dropping old one.
2670 Severity : enhancement
2671 Frequency : if client is started with failover MDS
2673 Description: Allow multiple MDS hostnames in the mount command
2674 Details : Try to read the configuration from all specified MDS
2675 hostnames during a client mount in case the "primary"
2678 Severity : enhancement
2680 Description: Stop sending data to evicted clients as soon as possible.
2681 Details : Check if the client we are about to send or are sending data to
2682 was evicted already. (Check is done every second of waiting,
2683 for which l_wait_event interface was extended to allow checking
2684 of exit condition at specified intervals).
2687 Frequency : rare, normally only when NFS exporting is done from client
2689 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2691 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2694 Frequency : always, if srand() is called before liblustre initialization
2696 Description: Liblustre uses system PRNG disturbing its usage by user application
2697 Details : Introduce internal to lustre fast and high-quality PRNG for
2698 lustre usage and make liblustre and some other places in generic
2699 lustre code to use it.
2701 Severity : enhancement
2702 Bugzilla : 9477, 9557, 9870
2703 Description: Verify that the MDS configuration logs are updated when xml is
2704 Details : Check if the .xml configuration logs are newer than the config
2705 logs stored on the MDS and report an error if this is the case.
2706 Request --write-conf, or allow starting with --old_conf.
2708 Severity : enhancement
2710 Description: Handle symlinks in the path when checking if Lustre is mounted.
2711 Details : Resolve intermediate symlinks when checking if a client has
2712 mounted a filesystem to avoid duplicate client mounts.
2717 Description: lconf can hit an error exception but still return success.
2718 Details : The lconf command catches the Command error exception at the top
2719 level script context and will exit with the associated exit
2720 status, but doesn't ensure that this exit status is non-zero.
2725 Description: failure of ptlrpc thread startup can cause oops
2726 Details : Starting a ptlrpc service thread can fail if there are a large
2727 number of threads or the server memory is very fragmented.
2728 Handle this without oopsing.
2731 Frequency : always, only if liblustre and non-default acceptor port was used
2733 Description: liblustre cannot connect to servers with non-default acceptor port
2734 Details : tcpnal_set_default_params() was not called and was therefore
2735 ignoring the environment varaible TCPNAL_PORT, as well as other
2736 TCPNAL_ environment variables
2741 Description: two objects could be created on the same OST for a single file
2742 Details : If an OST is down, in some cases it was possible to create two
2743 objects on a single OST for a single file. No problems other
2744 than potential performance impact and spurious error messages.
2748 Bugzilla : 5681, 9562
2749 Description: Client may oops in ll_unhash_aliases
2750 Details : Client dcache may become inconsistent in race condition.
2751 In some cases "getcwd" can fail if the current directory is
2757 Description: Inode refcounting problems in NFS export code
2758 Details : link_raw functions used to call d_instantiate without obtaining
2759 extra inode reference first.
2763 Bugzilla : 9942, 9903
2764 Description: Referencing freed requests leading to crash, memleaks with NFS.
2765 Details : We used to require that call to ll_revalidate_it was always
2766 followed by ll_lookup_it. Also with revalidate_special() it is
2767 possible to call ll_revalidate_it() twice for the same dentry
2768 even if first occurence returned success. This fix changes semantic
2769 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2770 reference on a request referred from the intent.
2771 ll_intent_release() then releases such a request.
2774 Frequency : rare, normally benchmark loads only
2776 Description: unlinked inodes were kept in memory on the client
2777 Details : If a client is repeatedly creating and unlinking files it
2778 can accumulate a lot of stale inodes in the inode slab cache.
2779 If there is no other client load running this can cause the
2780 client node to run out of memory. Instead flush old inodes
2781 from client cache that have the same inode number as a new inode.
2784 Frequency : SLES9 2.6.5 kernel and long filenames only
2785 Bugzilla : 9969, 10379
2786 Description: utime reports stale NFS file handle
2787 Details : SLES9 uses out-of-dentry names in some cases, which confused
2788 the lustre dentry revalidation. Change it to always use the
2792 Frequency : rare, unless heavy write-truncate concurrency is continuous
2793 Bugzilla : 4180, 6984, 7171, 9963, 9331
2794 Description: OST becomes very slow and/or deadlocked during object unlink
2795 Details : filter_destroy() was holding onto the parent directory lock
2796 while truncating+unlinking objects. For very large objects this
2797 may block other threads for a long time and slow overall OST
2798 responsiveness. It may also be possible to get a lock ordering
2799 deadlock in this case, or run out of journal credits because of
2800 the combined truncate+unlink. Solution is to do object truncate
2801 first in one transaction without parent lock, and then do the
2802 final unlink in a new transaction with the parent lock. This
2803 reduces the lock hold time dramatically.
2806 Frequency : rare, 2.4 kernels only
2808 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2809 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2810 before unmounting their filesystem in order to avoid deadlock.
2811 The kernel_locked() function in 2.4 kernels only checks whether
2812 the kernel lock is held, not whether it is this process that is
2813 holding it as 2.6 kernels do.
2818 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2819 Details : The client ptlrpc code may be trying to reconnect to a down
2820 server before a previous connection attempt has timed out.
2821 Increase the reconnect interval to be longer than the connection
2822 timeout interval to avoid sending duplicate connections to
2826 Frequency : echo_client brw_test command
2828 Description: fix echo_client to work with OST preallocated code
2829 Details : OST preallocation code (5137) didn't take echo_client IO path
2830 into account: echo_client calls filter methods outside of any
2831 OST thread and, hence, there is no per-thread preallocated
2832 pages and buffers to use. Solution: hijack pga pages for IO. As
2833 a byproduct, this avoids unnecessary data copying.
2837 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2838 Description: Client can oops in mdc_commit_close() after open replay
2839 Details : It was possible for the MDS to return an open request with no
2840 transaction number in mds_finish_transno() if the client was
2841 evicted, but without actually returning an error. Clients
2842 would later try to replay that open and may trip an assertion
2843 Simplify the client close codepath, and always return an error
2844 from the MDS in case the open is not successful.
2847 Frequency : rare, 2.6 OSTs only
2849 Description: OST may deadlock under high load on fragmented files
2850 Details : If there was a heavy load and highly-fragmented OST filesystems
2851 it was possible to have all the OST threads deadlock waiting on
2852 allocation of biovecs, because the biovecs were not released
2853 until the entire RPC IO was completed. Instead, release biovecs
2854 as soon as they are complete to ensure forward IO progress.
2856 Severity : enhancement
2858 Description: Support for specifying external journal device at mount
2859 Details : If an OST or MDS device is formatted with an external journal
2860 device, this device major/minor is stored in the ext3 superblock
2861 and may not be valid for failover. Allow detecting and
2862 specifying the external journal at mount time.
2867 Description: Mounting an MDS with pending unlinked files may cause oops
2868 Details : target_finish_recovery() calls mds_postrecov() which returned
2869 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2870 considers this an error and immediately begins cleaning up the
2871 lov, just after starting the mds_lov process
2873 Severity : enhancement
2875 Description: Implement 'lfs df' to report actual free space on per-OST basis
2876 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2877 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2878 usage of objects; '-h' to report in human readable format.
2880 ------------------------------------------------------------------------------
2882 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2889 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2890 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2891 at startup during orphan unlinking. Instead, skip buddy update
2892 before setup, as it will be regenerated after recovery is complete.
2897 Description: If an OST is inactive, its locks might reference stale inodes.
2898 Details : lov_change_cbdata() must iterate over all namespaces, even if
2899 they are inactive to clear inode references from the lock.
2901 Severity : enhancement
2902 Frequency : occasional, if non-standard max_dirty_mb used
2904 Description: Client will block write RPCs if not enough grant
2905 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2906 then the client will block writes while waiting for another RPC
2907 to complete instead of consuming its dirty limit. With change
2908 we get improved performance when max_dirty_mb is small.
2910 Severity : enhancement
2911 Bugzilla : 3389, 6253
2912 Description: Add support for supplementary groups on the MDS.
2913 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2914 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2915 lookups for user supplementary groups into a cache.
2919 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2920 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2921 for liblustre O_EXCL is enforced by MDS.
2925 Description: readdir never returns NULL in liblustre.
2926 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2927 dentry in struct dirent.
2931 Frequency : liblustre only, depends on application IO pattern
2932 Description: liblustre clients evicted if not contacting servers
2933 Details : Don't put liblustre clients into the ping_evictor list, so
2934 they will not be evicted by the pinger ever.
2936 Severity : enhancement
2938 Description: Add ability to evict clients by NID from MDS.
2939 Details : By echoing "nid:$NID" string into
2940 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2941 $NID would be instantly evicted from this MDS and from all active
2942 OSTs connected to it.
2946 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2947 Details : lookup_it_finish() used to query file size from OSTs that was not
2952 Description: service threads change working directory to that of init
2953 Details : Starting lustre service threads may pin the working directory
2954 of the parent thread, making that filesystem busy. Threads
2955 now change to the working directory of init to avoid this.
2959 Frequency : during shutdown only
2960 Description: shutdown with a failed MDS or OST can cause unmount to hang
2961 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2966 Frequency : 2.6 only
2967 Description: chmod/chown may include an extra supplementary group
2968 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2969 supplementary group and if none is specified this is used.
2972 Bugzilla : 5479 (6816)
2974 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2975 Details : If lookup is in progress on a file that is unlinked we might try
2976 to revalidate the inode and fail in revalidate after lookup is
2977 complete and ll_file_open() enqueues the open again but
2978 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2981 Frequency : always, if lconf --abort_recovery used
2983 Description: lconf --abort_recovery fails with 'Operation not supported'
2984 Details : lconf was attempting to abort recovery on the MDT device and not
2987 ------------------------------------------------------------------------------
2989 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2994 Frequency : rare (only unsupported configurations with a node running as an
2996 Bugzilla : 6514, 5137
2997 Description: Mounting a Lustre file system on a node running as an OST could
2999 Details : OSTs now preallocates memory needed to write out data at
3000 startup, instead of when needed, to avoid having to
3001 allocate memory in possibly low memory situations.
3002 Specifically, if the file system is mounted on on OST,
3003 memory pressure could force it to try to write out data,
3004 which it needed to allocate memory to do. Due to the low
3005 memory, it would be unable to do so and the node would
3006 become unresponsive.
3008 Severity : enhancement
3010 Description: Addition of lconf --service command line option
3011 Details : lconf now accepts a '--service <arg>' option, which is
3012 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3014 Severity : enhancement
3016 Description: Failover mode is now the default for OSTs.
3017 Details : By default, OSTs will now run in failover mode. To return to
3018 the old behaviour, add '--failout' to the lmc line for OSTs.
3020 Severity : enhancement
3022 Description: Health checks are now provided for MDS and OSTs
3023 Details : Additional detailed health check information on MSD and OSTs
3024 is now provided through the procfs health_check value.
3027 Frequency : occasional, depends on IO load
3029 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3030 after numerous create/delete cycles
3031 Details : The ext3 inode allocation policy would not allocate new inodes
3032 very well on the OSTs because there are no new directories
3033 being created. Instead we look for groups with free space if
3034 the parent directories are nearly full.
3039 Description: Network or server problems during mount may cause partially
3040 mounted clients instead of returning an error.
3041 Details : The config llog parsing code may overwrite the error return
3042 code during mount error handling, returning success instead
3048 Description: MDS can fail to allocate large reply buffers
3049 Details : After long uptimes the MDS can fail to allocate large reply
3050 buffers (e.g. zconf client mount config records) due to memory
3051 fragmentation or consumption by the buffer cache. Preallocate
3052 some large reply buffers so that these replies can be sent even
3053 under memory pressure.
3057 Frequency : rare (liblustre)
3058 Description: fsx running with liblustre complained that using truncate() to
3059 extend the file doesn't work. This patch corrects that issue.
3060 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3061 ATTR_SIZE and lsm use in llu_setattr_raw.
3065 Frequency : rare, only 2.6 kernels
3066 Description: Unusual file access patterns on the MDS may result in inode
3067 data being lost in very rare circumstances.
3068 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3069 code in the 2.6 kernel caused the inode and/or EA data not to
3070 be read from disk, causing single-file corruption.
3074 Frequency : rare, only 2.6 filesystems using extents
3075 Description: Heavy concurrent write and delete load may cause data corruption.
3076 Details : It was possible under high-load situations to have an extent
3077 metadata block in the block device cache from a just-unlinked
3078 file overwrite a newly-allocated data block. We now unmap any
3079 metadata buffers that alias just-allocated data blocks.
3083 Frequency : filesystems with default stripe_count larger than 77
3084 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3085 Details : lconf specifies an inode size of 4096 bytes when the default
3086 stripe_count is larger than 77. This conflicts with the default
3087 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3088 this case to avoid pinning too much memory for large EAs.
3090 ------------------------------------------------------------------------------
3092 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3097 Frequency : rare (extremely heavy IO load with hundreds of clients)
3099 Description: Client is evicted, gets IO error writing to file
3100 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3101 caused clients to be evicted for AST timeouts. The fixes in
3102 bug 5192 mean we no longer need to have such short AST timeouts
3103 so ldlm_timeout has been increased.
3106 Frequency : occasional during --force or --failover shutdown under load
3107 Bugzilla : 5949, 4834
3108 Description: Server oops/LBUG if stopped with --force or --failover under load
3109 Details : a collection of import/export refcount and cleanup ordering
3110 issues fixed for safer force cleanup
3113 Frequency : only filesystems larger than 120 OSTs
3114 Bugzilla : 5990, 6223
3115 Description: lfs getstripe would oops on a very large filesystem
3116 Details : lov_getconfig used kfree on vmalloc'd memory
3119 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3120 Bugzilla : 6242, 6243
3121 Description: reading from files that had been truncated to a non-zero size
3122 but never opened returned no data
3123 Details : ll_file_read() reads zeros from no-object files to EOF
3128 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3129 Details : There's little harm in aborting MDS/OSS recovery and letting it
3130 try again, so I removed the LASSERT and return an error instead.
3132 Severity : enhancement
3134 Description: New debugging infrastructure for tracking down data corruption
3135 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3136 (b) cover more of the client-side code path, and (c) try to narrow
3137 down where problems occurred
3141 Bugzilla : 3819, 4364, 4397, 6313
3142 Description: Racing close and eviction MDS could cause assertion in mds_close
3143 Details : It was possible to get multiple mfd references during close and
3144 client eviction, leading to one thread referencing a freed mfd.
3146 Severity: : enhancement
3147 Bugzilla : 3262, 6359
3148 Description: Attempts to reconnect to servers are now more aggressive.
3149 Details : This builds on the enhanced upcall-less recovery that was added
3150 in 1.4.2. When trying to reconnect to servers, clients will
3151 now try each server in the failover group every 10 seconds. By
3152 default, clients would previously try one server every 25 seconds.
3157 Description: After recovery, certain operations trigger a failed
3158 assertion on a client.
3159 Details : Failing over an mds, using lconf -d --failover, while a
3160 client was doing a readdir() call would cause the client to
3161 LBUG after recovery completed and the readdir() was resent.
3163 Severity : enhancement
3165 Description: Default groups are now added by lconf
3166 Details : You can now run lconf --group <servicename> without having to
3167 manually add groups with lmc.
3170 Frequency : occasional
3172 Description: Nodes with an elan id of 0 trigger a failed assertion
3175 Frequency : always when accessing e.g. tty/console device nodes
3177 Description: tty and some other devices nodes cannot be used on lustre
3178 Details : file's private_data field is used by device data and lustre
3179 values in there got lost. New field was added to struct file to
3180 store fs-specific private data.
3183 Frequency : when exporting Lustre via NFS
3185 Description: NFSD failed occasionally when looking up a path component
3186 Details : NFSD is looking up ".." which was broken in ext3 directories
3187 that had grown large enough to become hashed.
3190 Frequency : Clusters with multiple interfaces not on the same subnet
3192 Description: Nodes will repeatedly try to reconnect to an interface which it
3193 cannot reach and report an error to the log.
3194 Details : Extra peer list entries will be created by lconf with some peers
3195 unreachable. lconf now validates the peer before adding it.
3198 Frequency : Only if a default stripe is set on the filesystem root.
3200 Description: Setting a default stripe on the filesystem root prevented the
3201 filesystem from being remounted.
3202 Details : The client was sending extra request flags in the root getattr
3203 request and did not allocate a reply buffer for the dir EA.
3206 Frequency : occasional, higher if lots of files are accessed by one client
3207 Bugzilla : 6159, 6097
3208 Description: Client trips assertion regarding lsm mismatch/magic
3209 Details : While revalidating inodes the VFS looks up inodes with ifind()
3210 and in rare cases can find an inode that is being freed.
3211 The ll_test_inode() code will free the lsm during ifind()
3212 when it finds an existing inode and then the VFS later attaches
3213 this free lsm to a new inode.
3217 Bugzilla : 6422, 7030
3218 Description: MDS deadlock between mkdir and client eviction
3219 Details : Creating a new file via mkdir or mknod (starting a transaction
3220 and getting the ns lock) can deadlock with client eviction
3221 (gets ns lock and trying to finish a synchronous transaction).
3224 Frequency : occasional
3225 Description: While starting a server, the fsfilt_ext3 module could not be
3227 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3228 kernels. Previously, lconf would still use the ext3 name
3229 when trying to load modules. Now, it will correctly use
3230 ext3 on 2.4 and ldiskfs on 2.6.
3232 Severity : enhancement
3233 Description: The default stripe count has been changed to 1
3234 Details : The interpretation of the default stripe count (0, to lfs
3235 or lmc) has been changed to mean striping across a single
3236 OST, rather than all available. For general usage we have
3237 found a stripe count of 1 or 2 works best.
3239 Severity : enhancement
3240 Description: Add support for compiling against Cray portals.
3241 Details : Conditional compiling for some areas that are different
3245 Frequency : occasional
3246 Bugzilla : 6409, 6834
3247 Description: Creating files with an explicit stripe count may lead to
3248 a failed assertion on the MDS
3249 Details : If some OSTs are full or unavailable, creating files may
3250 trigger a failed assertion on the MDS. Now, Lustre will
3251 try to use other servers or return an error to the
3255 Frequency : occasional
3257 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3258 caused lock timeout during readahead (since 1.4.2).
3259 Details : Processes doing readahead might match a lock that hasn't been
3260 granted yet if there are overlapping and conflicting lock
3261 requests. The readahead process waits on ungranted lock
3262 (original lock is CBPENDING), while OST waits for that process
3263 to cancel CBPENDING read lock and eventually evicts client.
3265 Severity : enhancement
3267 Description: Initial enabling of flock support for clients
3268 Details : Implements fcntl advisory locking and file status functions.
3269 This feature is provided as an optional mount flag (default
3270 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3271 locking are implemented yet, and those that are are not guaranteed
3272 to be completely correct in production environments.
3273 mount -t lustre -o [flock|noflock] ...
3276 Frequency : occasional
3278 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3279 assertion in the ext3 JBD (journaling) layer.
3280 Details : The b_committed_data struct is protected by the big kernel lock
3281 in 2.4 kernels, serializing journal_commit_transaction() and
3282 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3283 access to this struct. In 2.6 kernels there is finer grained
3284 locking to improve SMP performance of the JBD layer.
3288 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3293 Description: merge and cleanup kernel patches.
3296 -----------------------------------------------------------------------------
3298 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3300 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3301 versions of Lustre. Please update all servers and clients to
3302 version 1.4.2 or later at the same time. You must also run
3303 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3304 to update the configuration logs.
3306 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3307 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3308 - swab llog records of type '0' so we get proper header size/idx (5861)
3309 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3310 - fix rename of one directory over another leaking an inode (5953)
3311 - avoid SetPageDirty on 2.6 (5981)
3312 - don't re-add just-being-destroyed locks to the waiting list (5653)
3313 - when creating new directories, inherit the parent's custom
3314 striping settings if present parent (3048)
3315 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3316 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3317 - don't hold client locks on temporary worklist from l_lru (5666)
3318 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3319 - automatically evict dead clients (5921)
3320 - Update file size properly in create+truncate+fstat case (6196)
3321 - Do not unhash mountpoint dentries, do not allow removal of
3323 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3324 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3325 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3326 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3327 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3328 - Correctly handle reads of files with no objects (6243)
3329 - lctl recover will also mark a device active if deactivate used (5933)
3331 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3332 - allow --write-conf on an MDS with different nettype than client (5619)
3333 - don't write config llogs to MDS for mounts not from that MDS (5617)
3334 - lconf should create multiple TCP connections from a client (5201)
3335 - init scripts are now turned off by default; run chkconfig --on
3336 lustre and chkconfig --on lustrefs to use them
3337 - upcalls are no longer needed for clients to recover to failover
3339 - add --abort-recovery option to lconf to abort recovery on device
3341 - add support for an arbitrary number of OSTs (3026)
3342 - Quota support protocol changes.
3343 - forward compatibility changes to wire structs (6007)
3344 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3345 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3346 - improved "lustre status" script
3347 - initialize blocksize for non-regular files (6062)
3348 - added --disable-server and --disable-client configure options (5782)
3349 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3350 - Vanilla 2.4.29 support
3351 - increase maximum number of obd devices to 520 (6242)
3352 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3353 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3354 SLES 9 2.6.5 kernels
3355 - limit stripes per file to 160 (the maximum EA size) (6093)
3357 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3360 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3361 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3362 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3363 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3364 - small ext3 extents cleanups and fixes (5733)
3365 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3366 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3367 - avoid needless client->OST connect, fix handle mismatch (5317)
3368 - fix DLM error path that led to out-of-sync client, long delays (5779)
3369 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3370 - fix several locking issues related to i_size (5492,5624,5654,5672)
3371 - don't move pending lock onto export if it is already evicted (5683)
3372 - fix kernel oops when creating .foo in unlinked directory (5548)
3373 - fix deadlock in obdfilter statistics vs. object create (5811)
3374 - use time_{before,after} to avoid timer jiffies wrap (5882)
3375 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3376 - Do not leak request if server was not able to process it (5154)
3377 - If mds_open unable to find parent dir, make that negative lookup(5154)
3378 - don't create new directories with extent-mapping (5909, 5936)
3380 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3381 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3382 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3383 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3384 - support for RHEL4 kernel on i686 (5773)
3385 - provide error messages when incompatible logs are encountered (5898)
3387 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3388 * version 1.4.0.10 (1.4.1 release candidate 1)
3390 - don't keep a lock reference when lock is not granted (4238)
3391 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3392 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3393 - drop import inflight refcount on signal_completed_replay error (5255)
3394 - unlock page after async write error during send (3677)
3395 - handle missing objects in filter_preprw_read properly (5265)
3396 - no transno return for symlink open, don't save no-trasno open (3440)
3397 - don't try to complete elan receive that already failed (4012)
3398 - free RPC server reply state on error (5406)
3399 - clean up thread from ptlrpc_start_thread() on error (5160)
3400 - readahead could read extra page into cache that wasn't ejected (5388)
3401 - prevent races in class_attach/setup/cleanup/detach (5260)
3402 - don't dereference de->d_inode after l_dput of de (5458)
3403 - use "int" for stripe value returned from lock_to_stripe (5544)
3404 - mballoc allocation and error-checking fixes in 2.6 (5504)
3405 - block device patches to fix I/O request sizes in 2.6 (5482)
3406 - look up hostnames for IB nals (5602)
3407 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3408 - don't start multiple acceptors for the same port (5277)
3409 - fix incorrect LASSERT in mds_getattr_name (5635)
3410 - export a proc file for general "ping" checking (5628)
3411 - fix "lfs check" to not block when the MDS is down (5628)
3413 - service request history (4965)
3414 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3415 - create an "evict_client" /proc entry on OSTs, like the MDS has
3416 - fix mount usage message, return errors per mount(8) (5168)
3417 - change grep [] to grep "[]" in tests so they work in more UMLs
3418 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3419 - remove ancient LOV_MAGIC_V0 EA support (5047)
3420 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3421 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3422 - allow readahead allocations to fail when low on memory (5383)
3423 - mmap locking landed again, after considerable improvement (2828)
3424 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3426 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3429 - send OST transaction number in read/write reply to free req (4966)
3430 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3431 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3432 - the watchdog thread now runs as interruptible (5246)
3433 - flock/lockf fixes (but it's still disabled, pending 5135)
3434 - don't use EXT3 constants in llite code (5094)
3435 - memory shortage at startup could cause assertion (5176)
3437 - reorganization of lov code
3438 - single portals codebase
3440 - add extents/mballoc support (5025)
3441 - direct I/O reads in the obdfilter (4048)
3442 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3444 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3447 - send OST transaction number in read/write reply to free req (4966)
3448 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3449 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3450 - the watchdog thread now runs as interruptible (5246)
3451 - handle missing objects in filter_preprw_read properly (5265)
3452 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3453 - drop import inflight refcount on signal_completed_replay error (5255)
3454 - unlock page after async write error during send (3677)
3455 - return original error code on reconstructed replies (3761)
3456 - no transno return for symlink open, don't save no-trasno open (3440)
3458 - add pid to ldlm debugging output (4922)
3459 - bump the watchdog timeouts -- we can't handle 30sec yet
3460 - extra debugging for orphan dentry/inode bug (5259)
3462 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3465 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3466 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3467 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3468 - fix NULL dereference in /proc/sys/portals/routes (4827)
3469 - allow failed mdc_close() operations to be interrupted (4561)
3470 - stop precreate on OST before MDS would time out on it (4778)
3471 - don't send partial-page writes before EOF from client (4410)
3472 - discard client grant for sub-page writes on large-page clients (4520)
3473 - don't free dentries not owned by NFS code, check generation (4806)
3474 - fix lsm leak if mds_create_objects() fails (4801)
3475 - limit debug_daemon file size, always print CERROR messages (4789)
3476 - use transno after validating reply (3892)
3477 - process timed out requests if import state changes (3754)
3478 - update mtime on OST during writes, return in glimpse (4829)
3479 - add mkfsoptions to LDAP (4679)
3480 - use ->max_readahead method instead of zapping global ra (5039)
3481 - don't interrupt __l_wait_event() during strace
3483 - add software watchdogs to catch hung threads quickly (4941)
3484 - make lustrefs init script start after nfs is mounted
3485 - fix CWARN/ERROR duplication (4930)
3486 - return async write errors to application if possible (2248)
3487 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3488 - print NAL number in %x format (4645)
3489 - update barely-supported suse-2.4.21-171 series (4842)
3490 - support for sles 9 %post scripts
3491 - support for building 2.6 kernel-source packages
3492 - support for sles km_* packages
3494 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3497 - ignore -ENOENT errors in osc_destroy (3639)
3498 - notify osc create thread that OSC is being cleaned up (4600)
3499 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3500 - reconstruct ost_handle() like mds_handle() (4657)
3501 - create a new thread to do import eviction to avoid deadlock (3969)
3502 - let lconf resolve symlinked-to devices (4629)
3503 - don't unlink "objects" from directory with default EA (4554)
3504 - hold socknal file ref over connect in case target is down (4394)
3505 - allow more than 32000 subdirectories in a single directory (3244)
3506 - fix blocks count for O_DIRECT writes (3751)
3507 - OST returns ENOSPC from object create when no space left (4539)
3508 - don't send truncate RPC if file size isn't changing (4410)
3509 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3510 - bind to privileged port in socknal and tcpnal (3689)
3512 - rate limit CERROR/CWARN console message to avoid overload (4519)
3513 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3514 - basic mmap support (3918)
3515 - kernel patch series update from b1_4 (4711)
3517 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3520 - avoid crash during MDS cleanup with OST shut down (2775)
3521 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3522 - don't use bad inodes on the MDS (3744)
3523 - dynamic object preallocation to improve recovery speed (4236)
3524 - don't hold spinlock over lock dumping or change debug flags (4401)
3525 - don't zero obd_dev when it is force cleaned (3651)
3526 - print grants to console if they go negative (4431)
3527 - "lctl deactivate" will stop automatic recovery attempts (3406)
3528 - look for existing locks in ldlm_handle_enqueue() (3764)
3529 - don't resolve lock handle twice in recovery avoiding race (4401)
3530 - revalidate should check working dir is a directory (4134)
3532 - don't always mark "slow" obdfilter messages as errors (4418)
3534 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3537 - don't close LustreDB during write_conf until it is done (3860)
3538 - fix typo in lconf for_each_profile (3821)
3539 - allow dumping logs from multiple threads at one time (3820)
3540 - don't allow multiple threads in OSC recovery (3812)
3541 - fix debug_size parameters (3864)
3542 - fix mds_postrecov to initialize import for llog ctxt (3121)
3543 - replace config semaphore with spinlock (3306)
3544 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3545 - don't allow enqueue to complete on a destroyed export (3822)
3546 - down write_lock before checking llog header bitmap (3825)
3547 - recover from lock replay timeout (3764)
3548 - up llog sem before sending rpc (3652)
3549 - reduce ns lock hold times when setting kms (3267)
3550 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3551 - fix NULL deref and obd_dev leak on setup error (3312)
3552 - replace some LBUG about llog ops with error handling (3841)
3553 - don't match INVALID dentries from d_lookup and spin (3784)
3554 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3555 - fix invalid assertion in ptlrpc_set_wait (3880)
3557 - add libwrap support for the TCP acceptor (3996)
3558 - add /proc/sys/portals/routes for non-root route listing (3994)
3559 - allow setting MDS UUID in .xml (2580)
3560 - print the stack of a process that LBUGs (4228)
3562 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3565 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3566 - make sure to unset replay flag from failed open requests (3440)
3567 - if default stripe count is 0, use OST count for inode size (3636)
3568 - update parent mtime/ctime on client for create/unlink (2611)
3569 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3570 - free recovery state on server during a forced cleanup (3571)
3571 - unregister_reply for resent reqs (3063)
3572 - loop back devices mounting and status check on 2.6 (3563)
3573 - fix resource-creation race that can provoke i_size == 0 (3513)
3574 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3575 - more debugging for page-accounting assertion (3746)
3576 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3577 - avoid deadlock after precreation failure (3758)
3578 - fix race and lock order deadlock in orphan handling (3450, 3750)
3579 - add validity checks when grabbing inodes from l_ast_data (3599)
3581 - add /proc/.../recovery_status to obdfilter (3428)
3582 - lightweight CDEBUG infrastructure, debug daemon (3668)
3583 - change default OSC RPC parameters to be better on small clusters
3584 - turn off OST read cache for files smaller than 32MB
3585 - install man pages and include them in rpms (3100)
3586 - add new init script for (un)mounting lustre filesystems (2593)
3587 - run chkconfig in %post for init scripts (3701)
3588 - drop scimac NAL (unmaintained)
3590 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3593 - clean kiobufs before and after use (3485)
3594 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3595 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3596 - change init script to fail more clearly if not run as root (1528)
3597 - allow clients to reconnect during replay (1742)
3598 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3599 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3600 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3601 - fix cond_resched (3554)
3602 - only evict unfinished clients after recovery (3515)
3603 - allow bulk resend, prevent data loss (3570)
3604 - dynamic ptlrpc request buffer allocation (2102)
3605 - don't allow unlinking open directory if it isn't empty (2904)
3606 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3607 - remove extraneous obd dereference causing LASSERT failure (3334)
3608 - don't use get_cycles() when creating temp. files on the mds (3156)
3609 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3610 - handle EEXIST for set-stripe, set proper directory name (3336)
3612 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3613 - fix ksocknal_fmb_callback() error messages (2918)
3615 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3618 - don't copy lvb into (possibly NULL) reply on error (2983)
3619 - don't deref dentry after dput, don't free lvb on error (2922)
3620 - use the kms to determine writeback rpc length (2947)
3621 - increment oti_logcookies when osc is inactive (2948)
3622 - update client's i_blocks count via lvb messages (2543)
3623 - handle intent open/close of special files properly (1557)
3624 - mount MDS with errors=remount-ro, like obdfilter (2009)
3625 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3626 - don't use cancelling-locks' kms values (2947)
3627 - use highest lock extent for kms, not last one (2925)
3628 - don't dereference ERR_PTR() dentry in error handling path (3107)
3629 - fix thread race in portals_debug_dumplog() (3122)
3630 - create lprocfs device entries at setup instead of at attach (1519)
3631 - common AST error handler, don't evict client on completion race (3145)
3632 - zero nameidata in detach_mnt in 2.6 (3118)
3633 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3634 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3635 - initialize RPC timeout timer earlier for 2.6 (3219)
3636 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3637 - print nal/nid for unknown nid (3258)
3638 - additional checks for oscc recovery before doing precreate (3284)
3639 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3640 - don't crash in mdc_close for bad permissions on open (3285)
3641 - zero i_rdev for non-device files (3147)
3642 - clear page->private before handing to FS, better assertion (3119)
3643 - tune the read pipeline (3236)
3644 - fix incorrect decref of invalidated dentry (2350)
3645 - provide read-ahead stats and refine rpc in flight stats (3328)
3646 - don't hold journal transaction open across create RPC (3313)
3647 - update atime on MDS at close time (3265)
3648 - close LDAP connection when recovering to avoid server load (3315)
3649 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3650 - don't leak open file on MDS after open resend (3325)
3651 - serialize filter_precreate and filter_destroy_precreated (3329)
3652 - loop device shouldn't call sync_dev() for nul device (3092)
3653 - clear page cache after eviction (2766)
3654 - resynchronize MDS->OST in background (2824)
3655 - refuse to mount the same filesystem twice on same mountpoint (3394)
3656 - allow llmount to create routes for mounting behind routers (3320)
3657 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3658 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3659 - fix rare problem with rename on htree directories (3417)
3661 - allow default OST striping configuration per directory (1414)
3662 - fix compilation for qswnal for 2.6 kernels (3125)
3663 - increase maximum number of MDS request buffers for large systems
3664 - change liblustreapi to be useful for external progs like lfsck (3098)
3665 - increase local configuration timeout for slow disks (3353)
3666 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3668 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3671 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3672 - don't overwrite extent policy data in reply if lock was blocked (2901)
3673 - drop filter export grants atomically with removal from device (2663)
3674 - del obd_self_export from work_list in class_disconnect_exports (2908)
3675 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3676 - swab reply message in mdc_close, other PPC fixes (2464)
3677 - fix destroying of named logs (2325)
3678 - overwrite old logs when running lconf --write_conf (2264)
3679 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3680 - fix race in target_handle_connect (2898)
3681 - mds_reint_create() should take same inode create lock (2926)
3682 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3683 - don't close files for self_export to avoid uninitialized obd (2936)
3684 - allow MDS with the same name as client node (2939)
3685 - hold dentry reference for closed log files for unlink (2325)
3686 - reserve space for all logs during transactions (2059)
3687 - don't evict page beyond end of stripe extent (2925)
3688 - don't oops on a deleted current working directory (2399)
3689 - handle hard links to targets without a parent properly (2517)
3690 - don't dereference NULL lock when racing during eviction (2867)
3691 - don't grow lock extents when lots of conflicting locks (2919)
3693 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3696 - account for cache space usage on clients to avoid data loss (974)
3697 - lfsck support in lustre kernel code (2349)
3698 - reduce journal credits needed for BRW writes (2370)
3699 - orphan handling to avoid losing space on client/server crashes
3700 - ptlrpcd can be blocked, stopping ALL progress (2477)
3701 - use lock value blocks to assist in proper KMS, faster stat (1021)
3702 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3703 - recovery for initial connections (2355)
3704 - fixes for mds_cleanup_orphans (1934)
3705 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3706 - block all file creations until orphan recovery completes (1901)
3707 - client remove rq_connection from request struct (2423)
3708 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3709 - recovery timer race (2670)
3710 - mdc_close recovey bug (2532)
3711 - ptlrpc cleanup bug (2710)
3712 - mds timeout on local locks (2588)
3713 - namespace lock held during RPCs (2431)
3714 - handle interrupted sync write properly (2503)
3715 - don't try to handle a message that hasn't been replied to (2699)
3716 - client assert failure during cleanup after abort recovery (2701)
3717 - leak mdc device after failed mount (2712)
3718 - ptlrpc_check_set allows timedout requests to complete (2714)
3719 - wait for inflight reqs when ptlrpcd finishes (2710)
3720 - make sure unregistered services are removed from the srv_list
3721 - reset bulk XID's when resending them (caught by 1138 test)
3722 - unregister_bulk after timeout
3723 - fix lconf error (2694)
3724 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3725 - readahead locks pages, leaves pending causing memory pressure (2673)
3726 - increase OST request buffers to 4096 on large machines (2729)
3727 - fix up permission of existing directories in simple_mkdir (2661)
3728 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3729 - don't assign transno to errored transactions (2742)
3730 - don't delete objects on OST if given a bogus objid from MDS (2751)
3731 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3732 - if rq_no_resend, then timeout request after recovery (2432)
3733 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3734 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3735 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3736 - MDS recovery shouldn't delete objects that it has given out (2730)
3737 - if enqueue arrives after completion, don't clobber LVB (2819)
3738 - don't unlock pages twice when trigger_group_io returns error (2814)
3739 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3740 - don't write pages to disk if there was an error (1450)
3741 - don't ping imports that have recovery disabled (2676)
3742 - take buffered bytes into account when balancing socknal conn (2817)
3743 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3744 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3745 - remove little-endian swabbing of llog records (1987)
3746 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3747 - retry reposting request buffers if they fail (1191)
3748 - grow extent at grant time to avoid granting a revoked lock (2809)
3749 - lock revoke doesn't evict page if covered by a second lock (2765)
3750 - disable VM readahead to avoid reading outside lock extents (2805)
3752 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3753 - updated kernel patches for hp-2.4.20 kernel (2681)
3755 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3758 - fix truncated write corruption (2366)
3759 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3761 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3762 - ensure len(uuid) < 37 in lmc (1171)
3763 - fix ia64 OOPS in llog_test (2255)
3764 - zero end of page at obdfilter for partial page writes (2648)
3765 - don't leave stale dentries around after renames (bug 2428)
3766 - fix timeouts when evicting a client with a single lock held (2642)
3767 - set deadline for the initial HELLO message to drain (2634)
3768 - print out dotted-quad IP addresses in the socknal (2302)
3770 - additional debugging for MDS client eviction problem (2443)
3771 - fix mkfsoptions support for osts (2603, 2604)
3773 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3776 - add series for the vanilla 2.6.0 kernel
3777 - add series for the vanilla 2.4.24 kernel
3778 - add series for a cray x86/64 UL kernel drop
3779 - fix xattr patches for the vanilla 2.4.19 series
3781 - generate true UUIDs in lmc (1171)
3782 - have portals stack dumping break in UML (2466)
3783 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3784 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3785 - wake callback waiting threads on client eviction (2460)
3786 - Add --ptldebug and --subsystem to lmc (1719)
3787 - update assertion to allow safe interrupt allocation
3788 - set rq_no_resend for cancel requests (2432)
3789 - recalculate ptlrpcd timeout after resend (2494)
3790 - call vfs_rmdir when removing pending directories (2368)
3791 - fix renaming a file to itself (2429)
3792 - lmc creates a default one-stripe lov (2454)
3793 - expand procfs space to handle large clusters (2326)
3794 - increase UML stack to avoid overflow
3795 - update lconf's list of debug and subsystem masks
3796 - fix lfs find --obd (2510)
3797 - /proc tunable for disabling filter read caching (2591)
3798 - stop rpm packages from altering slapd.conf (2301)
3799 - disable nagle in the socknal under 0conf (2578)
3800 - choose mds inode size based on stripe count (2572)
3801 - fix kernel-source rpm problems (2516)
3803 - add --disable-doc to avoid pdf generation (2421)
3804 - update documentation, tests, type-os, comments
3805 - avoid format warnings on ia64
3806 - remove the TOE NAL
3807 - tiny code cleanups by removing unused fields
3809 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3812 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3813 - remove the most blatant lies from BUILDING file (2371)
3814 - change default debug level to reasonable production setting
3815 - reduce client side cache size to reduce cache flush time
3816 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3817 - make TCP zerocopy and pinger support enabled by default (2476)
3818 - sync writes completed after process exits caused crashes (2319)
3819 - maintain correct mount count on the MDS (2356)
3820 - backout 1557, because 2316 wasn't really fixed
3821 - better file I/O statistics gathering in /proc
3822 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3823 - another kernel patch to fix zero-copy TCP function export
3824 - don't take duplicate lock when processing re-sent getattr (2420)
3825 - lctl uses obd_self_export instead of creating new conn (2353)
3826 - MDS/OST recovery case which requires object creation asserted (2425)
3827 - move lfs from /usr/sbin to /usr/bin in packages
3828 - fix race between mds_client_add and mds_client_free (2417)
3829 - use kmalloc instead of slabs in portals (2430)
3830 - don't create duplicate records when a failover MDS is present (2442)
3831 - remove unnecessary mount age check (2332)
3832 - don't remove directory inodes from locks prematurely (2451)
3833 - don't break if MDS service name is the same as hostname (2103)
3834 - fix races in client write RPC generation when cache full (2482)
3836 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3839 - remove now-unused request->rq_obd (278)
3840 - if an allocation fails, print out how much memory we've used (1933)
3841 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3842 - add the "configurable stack size" patch to most series files (1256)
3843 - ability to write large log records, for 100+ OST configs (2306)
3844 - fix NULL deref when filter_prep fails (2314)
3845 - fix operator precedence error in filter_sync
3846 - dynamic allocation of socknal TX descriptors (2315)
3847 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3848 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3849 - ensure that CWARN messages in Portals always get to the syslog
3850 - __init/__exit are not for prototype decls (ldlm_init/exit)
3851 - x86-64 compile warning fixes
3852 - fix gateway LMC keyword conflict (2318)
3853 - fix MDS lock inversions in getattr/reint paths (1844)
3854 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3855 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3856 - DLM race condition prevented some lock evictions (2328)
3857 - ENOMEM detection and retry on socknal sends (2230)
3858 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3859 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3860 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3861 - better error messages when a client is rejected during recovery (1505)
3862 - avoid cancelling locks which were never granted, after failure (2330)
3863 - fix i_sem/journal inversion in mds_client_add (2333)
3864 - fix truncate/getattr lock cycle deadlock (2334)
3865 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3866 - fix two rare exit paths which could leak an l_lock() ref (2321)
3867 - fencepost error in MDS/OST orphan recovery (2226)
3868 - make log record alignment 8 bytes (1988)
3869 - lstripe now fails when requested offset > ost_count (2237)
3870 - ensure that all kernel series have a complete list.h (1607)
3871 - fix crashes in special-file operations (2316)
3872 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3874 - allow configurable automake binary, for testing new versions
3875 - small update to the lfs documentation
3877 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3879 * fix negative export reference count in fsfilt_sync (2312)
3881 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3882 * release candidate 0.9.1
3884 - orphans are moved into the PENDING directory for possible recovery
3885 - replayed opens now open by fid for orphan/rename safety (1042)
3886 - last close of an orphan inode generates a transno (683)
3887 - chdir() and mount() now pin the directory entry (1020)
3888 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3889 - discard very old requests without processing them (1502)
3890 - remove some common, well-understood CERRORs (1505)
3891 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3892 - clear "grant" flags in OST replies until OST grant code lands (1644)
3893 - fix read performance by not clobbering i_blksize on client (1598)
3894 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3895 - make LRU size a /proc tunable, clears locks when reduced (707)
3896 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3897 - server locks take references on exports now (1558)
3898 - build fixes for 2.4.20-rh trees (1663)
3899 - return an error from lov_create if all OSCs are inactive (1751)
3900 - fix import levels when a reconnect happens without a timeout (1597)
3901 - exit early from mds_open if we get a lookup error (1749)
3902 - partial page read at EOF wouldn't wait for disk before sending (1642)
3903 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3904 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3905 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3906 - write/truncate lock inversion (1639)
3907 - Don't auto-load obdclass, portals modules during cleanup (1495)
3908 - fix timestamps from jumping to "now" (1763)
3909 - extra journal assertions (1648)
3910 - add an extra multiunlink test (1771)
3911 - fix read_record/write_record API (1776)
3912 - fix leak of offset_extent, possible incorrect i_size later (1772)
3913 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3914 - Debugging for the kqswnal_get_idle_tx problems (1820)
3915 - Allow recovery to be attempted multiple times (1536)
3916 - Write out MDS last_rcvd file after it is first created (1600)
3917 - Fix tx_descriptor leak in failed transmit situations (1827)
3918 - ext3 journaling fixes for assertion failure after IO error (1871)
3919 - class_export_put() on freed export after completion AST error (1896)
3920 - Fix revalidate looping in VFS (1322)
3921 - Don't access a freed export during MDS_REINT timeout (1521)
3922 - Add open-unlink recovery support on the MDS (1673,1764)
3923 - Return an error if no MDS data was read from last_rcvd (1946)
3924 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3925 - Don't setup a disk device that doesn't match exported UUID (317)
3926 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3927 - avoid committing NULL handle in force close
3928 - local.sh is now a one-stripe LOV configuration
3929 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3930 - fix ext3 htree duplicate directory entry corruption (1516)
3931 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3932 - update to new LOV EA format (2097)
3933 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3935 2003-06-15 Phil Schwan <phil@clusterfs.com>
3938 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3939 - per-import/export recovery handling (958, 931, 959)
3940 - multiple last-rcvd slots, for serving multiple FSes (949)
3941 - connections are again shared between multiple imp/exports (963, 964)
3942 - "umount -f" would hang if any requests needed to be sent (393, 978)
3943 - avoid pinning large req buffer by copying for queued messages (989)
3944 - add "uuid" to "lctl device" command to help upcalls (991)
3945 - "open" RPCs with transnos would confuse recovery counters (1037)
3946 - do proper endian conversion of all wire messages (288, 340, 891)
3947 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3948 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3949 - fix ptlrpc_connection leak in target_handle_connect (1174)
3950 - fix import refcounting bug in OST and MDS cleanup (1134)
3951 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3952 - fix very unlikely obd_types race condition (501)
3953 - remove osc_open hack for echo_client (1187)
3954 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3955 - a failure in read_inode2 leads to deadlock (1139)
3956 - cancel ack-locks as soon as transaction is committed (1072)
3957 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3958 - make sure to commitrw after any preprw to avoid deadlock (1162)
3959 - failing to execute a file in a lustre FS would lock inode (1203)
3960 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3961 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3962 - obd_uuid2tgt would walk off the end of the list (1255)
3963 - on IA64 the osc would give portals incorrect bulk size (1258)
3964 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3965 - fix lock inversion caused by new llite matching code (1282)
3966 - limit the number of dirty pages on a client to 10MB (1286)
3967 - timed out locks were not being corrected cancelled (1289)
3968 - fix O_DIRECT above 4GB on IA-32 (1292)
3969 * major user-visible changes
3970 - fail out/fail over policy now controlled by the upcall (993)
3972 - add OBD_PING to check server availability and failure (954)
3973 - lustre messages are now sent in sending host order (288, 340, 891)
3974 - add eadatalen to MDS getattr reply (340)
3975 - OST read replies may contain second buffer, with per-page status (593)
3977 2003-03-11 Phil Schwan <phil@clusterfs.com>
3980 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3981 - failed open()s could cause deadlock; fixed (867, 869)
3982 - stop cancelling OST locks when files are closed (481)
3983 - overlapping XID spaces caused network corruption (851, 853)
3984 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3985 - setattr_raw added, to do single-RPC, server-side setattrs
3986 - lmc/lconf syntax change for OST UUIDs
3987 - fix crashy race condition between ptlrpc_free_req and osc_close
3988 - don't use request in mdc_enqueue if we hit a timeout (889)
3989 - don't set the inode i_size for regular files from the MDS (896)
3990 - handle out of order completion AST (842)
3991 - don't LBUG if a lock request times out after receiving AST (913)
3992 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3993 - if a bad lock AST arrives, send an error instead of dropping entirely
3994 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3995 - fix leak in bulk IO when only partially completed (899, 900, 926)
3996 - fix O_DIRECT for ia64 (55)
3997 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3998 - C-z after timeout could hang a process forever; fixed (977)
4000 - client-side I/O cache (678, 924, 929, 941, 970)
4002 - READPAGE and SETATTRs which don't take server-side locks get
4005 2003-02-11 Phil Schwan <phil@clusterfs.com>
4008 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4009 - Fix fsfilt_extN_readpage() to read a full page of directory
4010 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4011 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4012 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4013 - Most intent operations were removed, in favour of a new RPC mode
4014 that does a single RPC to the server and bypasses most of the VFS
4015 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4016 - Aggressively cancel local locks on DLM servers
4017 - mds_reint_unlink sends EA to the client if it's the last nlink.
4018 client uses that EA to unlink OST objects.
4019 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4020 - recursive symlinks were fixed (439)
4021 - fixed NULL deref in DEBUG_REQ
4022 - filter_update_lastobjid no longer calls sync, which annoyed extN
4023 - fixed multi-client small-writes to a single file problem (445)
4024 - fixed mtime updates during file writes (607)
4025 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4026 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4027 - fixed Linux OST size reporting problem (444, 656)
4028 - OST now updates object mtime with writes or setattr (607, 619)
4029 - client verifies file size before zeroing page past EOF (445)
4030 - OST now writes last allocated objid to disk with allocation (108)
4031 - LOV on echo now works (409)
4033 - mds_reint_unlink sends a new buffer, with the EA included. this
4034 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4035 set if a regular file was being unlinked, and it was the last link
4036 - use PtlGet from the target for bulk writes (315)
4037 - OST now updates object mtime with writes or setattr (607, 619)
4038 - LDLM now has a grant-time callback to revalidate locked items, if
4040 - Many MDS operations were reorganized to combat race conditions
4042 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4043 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4045 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4048 - Fully reactivate OST imports after reconnection (512, others)
4049 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4050 - More graceful error handling for truncating on dead OST (515)
4051 - Don't error out unless we're actually accessing dead stripes (474)
4052 - Fix garbage sizes when stripes are missing (410)
4053 - LRU counters were broken, causing constant lock purge (433, 432)
4054 - garbage on read from stripes with failed OSTs (441)
4055 - mark OSCs as active before reconnecting during recovery (438)
4056 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4057 - lfind did not preserve OST order in output (443)
4058 - symlinks cause hung clients, incorrect data (439)
4059 - stop dereferencing request after dropping refcount (457)
4060 - don't LASSERT(spin_is_locked) on non-SMP (455)
4061 - fixes for many rename() bugs
4062 - fstat didn't correctly synchronize attributes (399)
4063 - server must handle lock cancellation during blocking AST prep (487)
4064 - bulk descriptors were free()d too soon (511)
4065 - fix paths in lconf, which would load incorrect modules (451, 507)
4066 - fix confusing lconf 'host not found' error message (386)
4067 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4068 - fix race condition in mdc_blocking_ast() for inode access (526)
4069 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4070 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4071 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4072 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4073 - fix lstripe in several ways (kernel side) (527)
4074 - fix request leak in ldlm_cli_enqueue (262)
4075 - incorrect OSC was marked inactive after OST failure
4076 - call mds_fs_cleanup before unmounting filesystem (524)
4077 - fix races between taking ns_lock and ldlm_lock_change_resource
4078 - fix races updating LOV export open file list
4079 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4080 - fix recovery NULL deref in ldlm_cli_cancel_unused
4081 - fix some DLM races by using new hash table for lock handles (419)
4082 - permit the client to specify desired inodes, at replay
4083 - duplicate requests when we queue them for replay reintegration
4084 - fix last_rcvd offset calculation
4085 - sync after each recovered transaction, so we always make progress
4086 - never, not always, ERESTART requests without transnos
4087 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4089 - skip replay if the MDS says that the client is already connected
4090 - don't check for a recovery-enabled export to match lctl's UUID
4091 - don't INC_USE_COUNT for phantom exports
4092 - don't crash when cleaning up phantom exports (567)
4093 - don't double-finish or set replay data for errored mdc_open requests
4094 - abort requests when they time out, so we don't get old replies
4095 - send/receive replies for AST messages again
4096 - if the client says that it doesn't have the lock, cancel it on the
4098 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4099 mark it as destroyed, it will all work out when decref is called
4100 - fix module use counts (22, 581)
4102 - ASTs now expect a reply (server cancels lock on error reply)
4104 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4107 - fix many simultaneous client startup (392)
4108 - fix dentry->d_it clobbering
4109 - credentials weren't being shipped for readdir/getattr operations
4110 - remove invalid assertions triggered during some concurrent MD
4112 - proper Lustre versions added (336, 389)
4113 - fix memory leak for create error case (398)
4114 - fix LOV locking bug that would get cli/srv out of sync
4115 - fix echo client over LOV (409)
4116 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4117 - fix double-O_EXCL intent crash (424)
4118 - avoid sending multiple lock CANCELs (352)
4120 - MDS can do multi-client recovery (modulo bugs in new code)
4122 - many updates, edits, cleanups
4124 2002-11-18 Phil Schwan <phil@clusterfs.com>
4127 - fix null d_it dereference (346)
4128 - fix full OST/dbench hang (333)
4129 - fix permission problem with file removal (286)
4130 - fix removal of OSCs from LOV when they fail
4131 - fix NULL deref during bulk timeout (214)
4132 - fix problems related to multiple filesystems on one MDS (241)
4133 - fixed serious subtle metadata locking bugs
4134 - free locks on clients when inodes are removed due to memory
4136 - fix inode pointer in lock data (285)
4137 - partial support for multiple MDS on a single host (241)
4138 - data locks weren't cancelled at clear_inode time (290, 311)
4139 - intent locks could lead to unbounded lock growth (205)
4140 - added a maximum lock count, an LRU list, and a flusher
4141 - fix multiple rename (365)
4142 - properly abstracted the echo client
4143 - OSC locked 1 byte too many; fixed
4144 - rewrote brw callback code:
4145 - fixed recovery bugs related to LOVs (306)
4146 - fixed too-many-pages-in-one-write crash (191)
4147 - fixed (again) crash in sync_io_timeout (214)
4148 - probably fixed callback-related race (385)
4150 - Add capability to MDS protocol
4151 - LDLM cancellations and callbacks on different portals
4153 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4156 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4157 - defer open object destruction to close time (601981, 138)
4158 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4159 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4160 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4161 - don't call obd_setattr in ll_file_release for destroyed objects
4162 * protocol change to lustre_msg: move |version| and add |flags|
4163 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4164 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4165 * added replay of create, unlink, link and rename operations during
4166 MDS failover; recovery should be much more robust now
4167 * remove failed OSCs from LOVs (only lov_create uses this so far)
4168 * the lustre-HOWTO was brought (more) up to date (582544)
4170 2002-10-23 Phil Schwan <phil@clusterfs.com>
4173 - in-use dentries weren't being reused properly (617851)
4174 - prevent multiple LDLM setup (599178)
4175 - fix LOV size calculations for truncate (617853)
4176 - fix client handling of MDS intent errors (POSIX)
4177 - fix permission bug in lovstripe.c test (624321)
4178 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4179 - truncate past end of file could corrupt data
4180 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4181 - a race in recovery could return ETIMEDOUT to apps (623947)
4182 - building outside the source directory was fixed
4183 * the lustre-HOWTO was brought (more) up to date (582544)
4184 * major progress was made on recovery functionality
4186 2002-10-10 Phil Schwan <phil@clusterfs.com>
4189 - recovery deadlock fix
4190 - rm -rf causes LBUG fix (617817)
4191 - file open by multiple tasks fix (618962)
4192 - directory permissions bugs (602707 and 620007)
4193 - journal_stop fixed with locking (611313)
4194 - O_APPEND failures resolved (618273, perhaps 614459)
4195 - lconf PATH fix (619770)
4196 - IA64 build fix (621450)
4197 - RPC buffer sizes scale with amount of memory
4199 2002-10-01 Phil Schwan <phil@clusterfs.com>
4202 - locks would be cancelled without throwing away data pages,
4203 resulting in inconsistent data (605627)
4204 - inode attributes were not always being refreshed (605627, 612449)
4205 - lconf now continues to cleanup after lctl reports an error
4206 - MDS now enforces user permissions (602707)
4207 - lprocfs cleanup fixed, but not yet enabled (614157)
4208 - fixed infinite server hang, should a client not respond to an AST
4209 - avoid going into recovery if user calls readlink() with a buffer
4210 that's too small (613941)
4211 - AST RPCs no longer require replies (614867) -- this may be changed
4212 - don't crash server if client sends an IOV that's too big (611336)
4213 - fixed lock conversion deadlock (611892)
4214 - fixed the following of symlinks (614622)
4215 * recovery: the server can remove locks from a client that dies, other
4216 clients can make progress
4217 * more extN patch fixes
4218 * compile-time configurable ptlrpc buffer allocations
4220 - collaborative read cache document
4221 - Lustre Lite Performance CDR document-in-progress
4223 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4226 - fix typo in patch-2.4.18
4228 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4231 - clear ptlrpc request each time in handle_incoming_request()
4232 - unlink of files now destroys the object on the OST
4234 2002-09-19 Peter Braam <braam@clusterfs.com>
4236 * add hard link support
4237 * change obdfile creation method
4238 * kernel patch changed
4240 2002-09-19 Peter Braam <braam@clusterfs.com>
4243 - stack overflow bug in extN fixed
4245 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4247 * documentation updates
4248 - add man pages for config tools
4249 - update tests/README to describe testing with new config tools
4250 - finish metadata API descriptions
4251 * bug fixes and cleanups
4252 - statfs workaround for 16TB limit
4253 - LOV stripe allocation improved, can stripe on subset of OSTs
4254 - LOV file size/IO offset was wrong for files > 4GB in size
4255 - object EA data was being dropped, caused files to be unreadable
4256 - memory overflow with non-LOV OST caused memory corruption
4257 - fixed regression tests to work with new config tools, obdfilter
4258 - fixed bug when directory size became larger than 1 block
4259 - fixed bug (for single client case) when PWD was deleted
4260 - invalidate local directory pages when doing intent-based ops
4261 - avoid LDLM oops when lock callback contained bad data
4263 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4265 * documentation updates
4266 * bug fixes and cleanups
4267 - configuration tools
4270 - 64-bit compile warnings
4271 - 64-bit internal statfs data
4273 * test_brw on persistent OST devices
4275 * lprocfs (disabled)
4277 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4279 * documentation updates
4280 * bug fixes and cleanups
4281 * configuration tools
4283 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4286 * many small fixes to 0_5_4
4287 * io/network handling
4288 * thinkos in MDS operations
4290 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4293 * crucial basic fixes to 0.5.3
4294 * IOR, Iozone work over Elan
4297 2002-08-07 Phil Schwan <phil@clusterfs.com>
4298 * version 0_5_3, our first alpha
4299 * we use the new Portals iovs
4300 * documentation updates
4301 * bug fixes and cleanups
4302 * small changes in the DLM wire protocol
4304 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4305 * version 0_5_1 with some initial stability,
4306 * locking on MD and file I/O.
4307 * documentation updates
4308 * several bug fixes since 0.5.0
4309 * small changes in wire protocol
4311 2002-07-18 Phil Schwan <phil@clusterfs.com>
4313 * delivered as Lustre Light Alpha
4314 * fixed a crash after handling invalid MDS requests
4315 * fixed directory pages for architectures with non-4k pages sizes
4317 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4318 * release version v0_4_4
4319 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4321 * Fixes a number of interruption problems with OST operations.
4322 * Update documentation for portals header changes
4323 * Move all wire protocol structs/defines to lustre_idl.h
4324 * Fixes symlink length bug.
4325 * Add tcpdump to repository.
4327 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4328 * release version v0_4_3
4329 * Fixes statfs for inodes on extN.
4330 * Fixes bug in runtests which would delete /etc/hosts.
4331 * Use 64-bit object IDs wherever possible (not into VFS though)
4332 Remove ost_get_info, which is unused by lustre, and out of date.
4334 2002-07-03 Peter Braam <braam@clusterfs.com>
4335 * release version v0_4_2 Fixes a lookup error (type not passed)
4336 * move forward to head of Portals
4337 * move forward to latest Lustre kernel
4339 2002-06-25 Peter Braam <braam@clusterfs.com>
4340 * release version v0_4_1. Hopefully stable on single node use.