1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.14.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
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: 1.40.7-sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Description: wait until IO finished before start new when do lock cancel.
18 Details : VM protocol want old IO finished before start new, in this case
19 need wait until PG_writeback is cleared until check dirty flag and
20 call writepages in lock cancel callback.
22 Severity : enhancement
24 Description: using special macro for print time and cleanup in includes.
28 Description: mds_mfd_close() ASSERTION(rc == 0)
29 Details : In mds_mfd_close(), we need protect inode's writecount change
30 within its orphan write semaphore to prevent possible races.
34 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
38 Frequency : rare, on shutdown ost
39 Description: don't hit live lock with umount ost.
40 Details : shrink_dcache_parent can be in long loop with destroy dentries,
41 use shrink_dcache_sb instead.
45 Description: don't panic with use echo client
46 Details : echo client pass NULL as client nid pointer and this produce null
51 Description: fix build on ppc32
52 Details : compile code with -m64 flag produce wrong object file for ppc32.
56 Description: add message levels for liblustreapi
60 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
61 Details : Move the 'good_osts' check before the 'total_bavail' check. This
62 will result in an -EAGAIN and in the exit call path we call
63 alloc_rr() which will with increasing aggressiveness attempt to
64 aquire precreated objects on the minimum number of required OSCs.
68 Description: Use old size assignment to avoid deadlock
69 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
70 the scheduling while holding a spinlock. We do not need locking
71 for size in ll_update_inode() because size is only updated from
72 the MDS for directories or files without objects, so there is no
73 other place to do the update, and concurrent access to such inodes
74 are protected by the inode lock.
78 Description: resolve "_IOWR redefined" build error on SLES10
82 Description: dump the memory debugging after all modules are unloaded to
83 suppress false negative in conf_sanity test 39
85 Severity : enhancement
87 Description: build kernel-ib packages for OFED 1.3 in our release cycle
92 Description: fix SLES kernel versioning
93 Details : the kernel version for our SLES 10 kernel did not include a "-"
94 before the "smp" at the end. while this was not a problem in
95 general, it did mean that software trying to use the kernel
96 version to try to detect a vendor specific kernel would fail.
97 this was most evident by the OFED build scripts.
101 Description: Don't update lov_desc members until making sure they are valid
102 Details : When updating lov_desc members via proc fs, need fix their
103 validities before doing the real update.
107 Description: don't put request into delay list while invalidate in flight.
108 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
109 import in flight. this produce timeout for invalidate and sometimes
110 can cause stale data.
114 Frequency : on ppc only
115 Description: not convert ost objects for directory because it's not exist.
116 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
120 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
122 Severity : enhancement
124 Description: Update to support 2.6.22.14 vanilla kernel.
128 Frequency : rare, on recovery
129 Description: read procfs can produce deadlock in some situation
130 Details : Holding lprocfs lock which send rpc can produce block for destroy
131 obd objects and this also block reconnect with -EALREADY. This isn't
132 fix all lprocfs bugs - but make it rare.
134 Severity : enhancement
136 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
139 Frequency : frequent on X2 node
141 Description: mdc_set_open_replay_data LBUG
142 Details : Set replay data for requests that are eligible for replay.
146 Description: lustre_mgs: operation 101 on unconnected MGS
147 Details : When MGC is disconnected from MGS long enough, MGS will evict the
148 MGC, and late on MGC cannot successfully connect to MGS and a lot
149 of the error messages complaining that MGS is not connected.
152 Frequency : on start mds
154 Description: Implement get_info(last_id) in obdfilter.
157 Frequency : occasional
159 Description: Correctly check stale fid, not start epoch if ost not support SOM
160 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
161 overwrited with new generated fid, but mds can anwer with one of these
162 two fids and both is not stale. setattr incorectly start epoch and
163 assume will be called done_writeting, but without SOM done_writing
167 Frequency : rare, depends on device drivers and load
169 Description: MDS or OSS nodes crash due to stack overflow
170 Details : Code changes in 1.8.0 increased the stack usage of some functions.
171 In some cases, in conjunction with device drivers that use a lot
172 of stack the MDS (or possibly OSS) service threads could overflow
173 the stack. One change which was identified to consume additional
174 stack has been reworked to avoid the extra stack usage.
177 Frequency : occasional
179 Description: Do not fail import if osc_interpret_create gets -EAGAIN
180 Details : If osc_interpret_create got -EAGAIN it immediately exits and
181 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
182 oscc_has_objects and see OSC has no objests and call
183 oscc_internal_create to resend create request.
185 Severity : enhancement
187 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
189 Severity : enhancement
191 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
194 Frequency : very rare
196 Description: Fix replay if there is an un-replied request and open
197 Details : In some cases, older replay request will revert the
198 mcd->mcd_last_xid on MDS which is used to record the client's
201 Severity : enhancement
203 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
205 Severity : enhancement
207 Description: Add rhel5 support to HEAD.
209 Serverity : enhancement
211 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
216 Description: Don't allow skipping OSTs if index has been specified.
217 Details : Don't allow skipping OSTs if index has been specified, make locking
218 in internal create lots better.
222 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
223 Details : access to bitfield in structure is always rounded to long
224 and this produce problem with not atomic change any bit.
228 Description: Lustre make rpms failed.
229 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
230 builds Lustre rpms from tarball.
233 Frequency : rare on shutdown ost
235 Description: If llog cancel was not send before clean_exports phase, this can
236 produce deadlock in llog code.
237 Details : If llog thread has last reference to obd and call class_import_put
238 this produce deadlock because llog_cleanup_commit_master wait when
239 last llog_commit_thread exited, but this never success because was
240 called from llog_commit_thread.
244 Description: allow userland application know is lost one of stripes.
245 Details : fill lvb_blocks with error code on ost and return it to
246 application if error flag found.
250 Description: NULL lov_tgts causing MDS oops
251 Details : more safe checks for NULL lov_tgts for avoid oops.
253 Severity : enhancement
255 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
259 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
260 Details : If an OST is down the MDS will hang indefinitely in
261 obd_statfs_rqset() waiting for the statfs data. While for
262 MDS QOS usage of statfs, it should not stuck in waiting.
264 Severity : enhancement
266 Description: remote_acl support
267 Details : Support ACL-based permission check for remote user.
268 Support setfacl/getfacl for remote user with the utils
269 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
270 the system "{s,g}etfacl" utils.
272 Severity : enhancement
274 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
276 Severity : enhancement
278 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
282 Description: make mgs_setparam() handle fsname containing dash
283 Details : fsname containing a dash does not work with lctl conf_param
285 Severity : enhancement
287 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
289 Severity : enhancement
291 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
295 Frequency : rare, at shutdown
296 Description: access already free / zero obd_namespace.
297 Details : if client_disconnect_export was called without force flag set,
298 and exist connect request in flight, this can produce access to
299 NULL pointer (or already free pointer) when connect_interpret
300 store ocd flags in obd_namespace.
304 Frequency : only at startup
305 Description: not alloc memory with spinlock held.
306 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
307 if any spinlock held.
309 Severity : enhancement
311 Description: make lustre randomly fail allocating memory
312 Details : Make lustre randomly failed allocating memory for testing purpose.
314 Severity : enhancement
316 Description: lost problems with lov objid file
317 Details : Fixes some scability and access to not inited memory problems
318 in work with lov objdid file.
323 Description: lfs find does not continue on file error
324 Details : Continue other files processing when a file/dir is absent.
328 Description: Inconsistent usage of lustre_pack_reply()
329 Details : Standardize the usage of lustre_pack_reply() such that it
330 always generate a CERROR on failure.
333 Frequency : occasional
335 Description: MDS hang or stay in waiting lock
336 Details : If client receive lock with CBPENDING flag ldlm need send lock
337 cancel as separate rpc, to avoid situation when cancel request
338 can't processed due all i/o threads stay in wait lock.
342 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
346 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
350 Description: Update to SLES9 kernel-2.6.5-7.287.3.
354 Description: Update to RHEL4 latest kernel.
356 Severity : enhancement
358 Description: Build SLES10 patchless client fails
359 Details : The configure was broken by run ./configure with
360 --with-linux-obj=.... argument for patchless client. When the
361 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
362 can't search header adequately. Use absolute path such as
363 -I($LINUX)/include instead.
367 Description: interrupt oig_wait produce painc on resend.
368 Details : brw_redo_request can be used for resend requests from ptlrpcd and
369 private set, and this produce situation when rq_ptlrpcd_data not
370 copyed to new allocated request and triggered LBUG on assert
371 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
372 ptlrpcd set if request is changed and can be safety changed to use
377 Description: LASSERT_{REQ,REP}SWAB macros are buggy
378 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
379 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
380 these macros and replace them with their difinitions instead.
384 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
385 Details : Update which_patch & target file for SLES10 latest kernel.
388 Description: Remove client patches from SLES 10 kernel.
389 Details : This causes SLES 10 clients to behave as patchless clients
390 even on a Lustre-patched (server) kernel.
392 Severity : enhancement
394 Description: self-adjustable client's lru lists
395 Details : use adaptive algorithm for managing client cached locks lru
396 lists according to current server load, other client's work
397 pattern, memory activities, etc. Both, server and client
398 side namespaces provide number of proc tunables for controlling
401 Severity : enhancement
403 Description: light-weight GSS support
404 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
405 and reduce performance overhead.
407 Severity : enhancement
409 Description: Linux keyring support
410 Details : Support using service of Linux keyring for Lustre GSS internal
411 context refresh/cache mechanism.
415 Description: Fix errors in lfs documentation
416 Details : Fixes man pages
420 Description: don't use GFP_* in generic Lustre code.
421 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
425 Description: obdclass is limited by single OBD_ALLOC(idarray)
426 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
430 Description: warnings with build patchless client with vanila 2.6.19 and up
431 Details : change old ctl_table style and replace ctl_table/ctl_table_header
432 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
436 Frequency : Only for SLES
437 Description: sanity_quota fail test_1
438 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
439 file, which makes TSTID[2] inunique.
443 Frequency : Only for SLES9
444 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
445 Details : Update target/ChangeLog/which_patch .
449 Frequency : Only for test_10 in sanity-quota.sh
450 Description: a bug in quota test code
451 Details : checking a test flag in a wrong place
455 Frequency : Only for test_13 in sanity-quota.sh
456 Description: a bug in quota test script
457 Details : shouldn't check overfull allocated quota
461 Description: jbd statistics
462 Details : Port older jbd statistics patch for sles10
464 Severity : enhancement
466 Description: Add printing inode info into message about error in writepage.
471 Description: reply_lock_interpret crash due to race with it and lock cancel.
472 Details : Do not replay locks that are being cancelled. Do not reference
473 locks by their address during replay, just by their handle.
477 Frequency : When flocks are used.
478 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
479 Details : Flock locks might destroy just granted lock if it could be merged
480 with another existing flock, this is done in completion handler,
481 so teach ldlm_cli_enquque_fini that this is a valid case for
487 Description: Oops in read and write path when failing to allocate lock.
488 Details : Check if lock allocation failed and return error back.
492 Description: lstripe command fails for valid OST index
493 Details : The stripe offset is compared to lov->desc.ld_tgt_count
494 instead of lov->desc.ld_active_tgt_count.
498 Description: sanity.sh failed test 103
499 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
504 Description: df doesn't work properly if diskfs blocksize != 4K
505 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
509 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
510 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
514 Description: obd_health_check_timeout too short
515 Details : set obd_health_check_timeout as 1.5x of obd_timeout
519 Description: llapi_file_create() does not allow some changes
520 Details : add llapi_file_open() that allows specifying the mode and
521 open flags, and also returns an open file handle.
525 Description: merge and cleanup kernel patches.
526 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
530 Description: Add journal checksum support.(Kernel part)
531 Details : The journal checksum feature adds two new flags i.e
532 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
533 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
534 indicates that the commit block contains the checksum for
535 the blocks described by the descriptor blocks. Now commit
536 record can be sent to disk without waiting for descriptor
537 blocks to be written to disk. This behavior is controlled
538 using JBD2_FEATURE_ASYNC_COMMIT flag.
542 Description: OSS needs mutliple precreate threads
543 Details : Add ability to start more than one create thread per OSS.
547 Description: Sanity.sh test_65k failed: setstripe should have succeeded
548 Details : execute lfs setstripe on client
552 Description: mds_obd_create error creating tmp object
553 Details : When the user sets quota on root, llog will be affected and can't
554 create files and write files.
557 Frequency : Always on ia64 patchless client, and possibly others.
559 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
560 Details : This allows the patchless client to be loaded on architectures
565 Description: RedHat Update kernel for RHEL5
566 Details : Add the kernel config file for RHEL5.
570 Description: RedHat Update kernel for RHEL5
571 Details : Modify the kernel config file more closer RHEL5.
575 Description: Build failure against Centos5 (RHEL5)
576 Details : Define PAGE_SIZE when it isn't present.
580 Description: test_7 on sanity-quota.sh can't be used on separate machines
581 Details : add facet to handle it
585 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
586 Details : with NFS, the anon dentry's parent was set to itself in
587 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
588 corresponding dentry other than use rec->ur_name.
590 Severity : enhancement
592 Description: lfs setstripe enhancement
593 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
597 Description: enable data checksumming by default
598 Details : enable checksum by default, allow --disable-checksum
599 configure option and "-o nochecksum" mount option. Checksums
600 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
604 Description: lustre support for RHEL5
605 Details : Add support for RHEL5.
609 Description: Client eviction due to failover config
610 Details : after a connection loss, the lustre client should attempt to
611 reconnect to the last active server first before trying the
612 other potential connections.
616 Description: when mds and osts use different quota unit(32bit and 64bit),
617 quota will be released repeatly.
618 Details : void sending multiple quota reqs to mds, which will keep the
619 status between the reqs.
623 Description: osts not allocated evenly to files
624 Details : change the condition to increase offset_idx
627 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
629 Description: Data corruption for OSTs that are formatted larger than 2TB
631 Details : When generating the bio request for lustre file writes the
632 sector number would overflow a temporary variable before being
633 used for the IO. The data reads correctly from Lustre (which
634 will overflow in a similar manner) but other file data or
635 filesystem metadata may be corrupted in some cases.
639 Description: Tune the kernel for good SCSI performance.
640 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
641 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
646 Description: rewrite ext2-derived code in obdclass/uuid.c
647 Details : rewrite inherited code (uuid parsing code from ext2 utils)
648 from scratch preserving functionality.
652 Description: "lfs find -obd UUID" prints directories
653 Details : "lfs find -obd UUID" will return all directory names instead
654 of just file names. It is incorrect because the directories
655 do not reside on the OSTs.
659 Description: use i_size_read and i_size_write in 2.6 port
660 Details : replace inode->i_size access with i_size_read/write()
663 Frequency : when removing large files
665 Description: scheduling issue during removal of large Lustre files
666 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
667 It causes scheduling issues when removing large files (17TB in the
671 Frequency : only with liblustre clients on XT3
673 Description: evictions taking too long
674 Details : allow llrd to evict clients directly on OSTs
677 Frequency : only on ppc
679 Description: /proc/fs/lustre/devices broken on ppc
680 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
681 We should make sure the type of 'index' is loff_t and then cast
682 explicitly as needed below. Do not assign an explicitly cast
686 Frequency : only for rhel5
688 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
689 Details : Modify the target file & which_kernel.
691 Severity : enhancement
693 Description: omit set fsid for export NFS
694 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
695 as FS_REQUIRES_DEV for avoid problems with generate fsid.
699 Frequency : Always, for kernels after 2.6.16
700 Description: Fix warning idr_remove called for id=.. which is not allocated.
701 Details : Last kernels save old s_dev before kill super and not allow
702 to restore from callback - restore it before call kill_anon_super.
706 Description: buffer overruns could theoretically occur
707 Details : llapi_semantic_traverse() modifies the "path" argument by
708 appending values to the end of the origin string, and a
709 overrun may occur. Adding buffer overrun check in liblustreapi.
713 Description: Fix error on 'ls .' at the top of the Lustre mount.
714 Details : Don't revalidate dentry if it is a root dentry.
718 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
719 Details : Modify vm-tunables-rhel4.patch.
723 Description: Kernel config for 2.6.18-vanilla.
724 Details : Modify targets/2.6-vanilla.target.in.
725 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
726 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
727 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
728 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
732 Description: improve handling recoverable errors
733 Details : if request processig with error which can be recoverable on server
734 request should be resend, otherwise page released from cache and
739 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
740 Details : Modify target file & which_patch.
741 A flaw was found in the IA32 system call emulation provided
742 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
743 value could be stored in the %RAX register, which could trigger an
744 out-of-bounds system call table access. An untrusted local user
745 could exploit this flaw to run code in the kernel
746 (ie a root privilege escalation). (CVE-2007-4573).
750 Description: O_DIRECT bypasses client statistics.
751 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
752 read_ahead_stats were updated. copied the stats section from
753 osc_send_oap_rpc() into async_internal().
757 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
761 Description: change order of libsysio includes
762 Details : '#include sysio.h' should always come before '#include xtio.h'
766 Description: handle "serious error: objid * already exists" more gracefully
767 Details : If LAST_ID value on disk is smaller than the objects existing in
768 the O/0/d* directories, it indicates disk corruption and causes an
769 LBUG(). If the object is 0-length, then we should use the existing
770 object. This will help to avoid a full fsck in most cases.
772 Severity : enhancement
774 Description: adapt the lustre_config script to support the upgrade case
775 Details : Add "-u" option for lustre_config script to support upgrading 1.4
776 server targets to 1.6 in parallel.
780 Description: To avoid grant space > avaible space when the disk is almost
781 full. Without this patch you might see the error "grant XXXX >
782 available" or some LBUG about grant, when the disk is almost
784 Details : In filter_check_grant, for non_grant cache write, we should
785 check the left space by if (*left > ungranted + bytes), instead
786 of (*left > ungranted), because only we are sure the left space
787 is enough for another "bytes", then the ungrant space should be
788 increase. In client, we should update cl_avail_grant only there
789 is OBD_MD_FLGRANT in the reply.
793 Description: Update RHEL 4 kernel to fix local root privilege escalation.
794 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
795 described in CVE-2007-4573. This problem could allow untrusted
796 local users to gain root access.
799 Frequency : when using O_DIRECT and quotas
801 Description: Incorrect file ownership on O_DIRECT output files
802 Details : block usage reported by 'lfs quota' does not take into account
803 files that have been written with O_DIRECT.
808 Description: touch file failed when fs is not full
809 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
810 otherwise we can get ENOSP while fs is not full.
814 Description: parallel lock callbacks
815 Details : Instead of sending blocking and completion callbacks as separated
816 requests, adding them to a set and sending in parallel.
819 Frequency : only for Cray XT3
820 Bugzilla : 12829/13455
821 Description: Changing primary group doesn't change the group lustre assigns to
823 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
824 overridden with the primary group provided by the group upcall,
825 whereas the supplied fsgid can be trusted if it is in the list of
826 supplementary groups returned by the group upcall.
828 Severity : enhancement
830 Description: Allow masking D_WARNING, D_ERROR messages from console
831 Details : Console messages can now be disabled via lnet.printk.
835 Description: User code with malformed file open parameter crashes client node
836 Details : Before packing join_file req, all the related reference should be
837 checked carefully in case some malformed flags cause fake
838 join_file req on client.
842 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
843 Details : ldlm_completion_ast() assumes that a lock is granted when the req
844 mode is equal to the granted mode. However, it should also check
845 that LDLM_FL_CP_REQD is not set.
849 Description: Heavy nfs access might result in deadlocks
850 Details : After ELC code landed, it is now improper to enqueue any mds
851 locks under och_sem, because enqueue might want to decide to
852 cancel open locks for same inode we are holding och_sem for.
856 Description: Client eviction while running blogbench
857 Details : A lot of unlink operations with concurrent I/O can lead to a
858 deadlock causing evictions. To address the problem, the number of
859 oustanding OST_DESTROY requests is now throttled to
860 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
861 callbacks are processed in priority.
865 Description: enable ACLs on MDS by default
866 Details : ACLs must be enabled on MDS by default.
869 Frequency : PPC/PPC64 only
871 Description: conflicts between asm-ppc64/types.h and lustre_types.h
872 Details : fix duplicated definitions between asm-ppc64/types.h and
873 lustre_types.h on PPC.
876 Frequency : PPC/PPC64 only
878 Description: asm-ppc/segment.h does not exist
879 Details : fix compile issue on PPC.
883 Description: better handle error messages in extents code
886 Frequency : RHEL4 only
888 Description: mkfs is very slow on IA64/RHEL4
889 Details : A performance regression has been discovered in the MPT Fusion
890 driver between versions 3.02.73rh and 3.02.99.00rh. As a
891 consequence, we have downgraded the MPT Fusion driver in the RHEL4
892 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
894 Severity : enhancement
896 Description: SNMP support enhancement
897 Details : Adding total number of sampled request for an MDS node in snmp
900 Severity : enhancement
902 Description: Optimize ldlm waiting list processing for PR extent locks
903 Details : When processing waiting list for read extent lock and meeting read
904 lock that is same or wider to it that is not contended, skip
905 processing rest of the list and immediatelly return current
906 status of conflictness, since we are guaranteed there are no
907 conflicting locks in the rest of the list.
911 Description: Time out and refuse to reconnect
912 Details : When the failover node is the primary node, it is possible
913 to have two identical connections in imp_conn_list. We must
914 compare not conn's pointers but NIDs, otherwise we can defeat
915 connection throttling.
919 Description: port llog fixes from b1_6 into HEAD
920 Details : Port llog reference couting and some llog cleanups from b1_6
921 (bug 10800) into HEAD, for protect from panic and access to already
922 free llog structures.
926 Description: Detect stride IO mode in read-ahead
927 Details : When a client does stride read, read-ahead should detect that and
928 read-ahead pages according to the detected stride pattern.
932 Description: data checksumming impacts single node performance
933 Details : add support for several checksum algorithm. Currently, only CRC32
934 and Adler-32 are supported. The checksum type can be changed on
935 the fly via /proc/fs/lustre/osc/*/checksum_type.
939 Description: use adler32 for page checksums
940 Details : when available, use the Adler-32 algorithm instead of CRC32 for
945 Description: build for x2 fails
946 Details : fix compile issue on Cray systems.
950 Description: Properly match for duplicate locks
951 Details : Due to different lock order from skiplists code, we need to
952 traverse entire list for now
955 Frequency : only on PPC/SLES10
957 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
958 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
959 be defined. Add a hack in mkfs_lustre.c to work around this compile
964 Description: LASSERT on MDS when client holding flock lock dies
965 Details : ldlm pool logic depends on number of granted locks equal to
966 number of released locks which is not true for flock locks, so
967 just exclude such locks from consideration.
971 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
972 Details : Use fsfilt_sync() for both the whole filesystem sync and
973 individual file sync to eliminate dangerous inode locking
974 with I_LOCK that can lead to a deadlock.
978 Description: performance in 1.6.3
979 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
980 sure that 1MB requests are not fragmented by the block layer.
982 Severity : enhancement
984 Description: organize the server-side client stats on per-nid basis
985 Details : Change the structure of stats under obdfilter and mds to
993 The "uuid"s file would list the uuids of _active_ exports.
994 And the clear entry is to clear all stats and stale nids.
996 Severity : enhancement
998 Description: eliminate client locks in face of contention
999 Details : file contention detection and lockless i/o implementation
1000 for contended files.
1004 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1007 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1009 Description: MDS may lose file striping (and hence file data) in some cases
1010 Details : If there are additional extended attributes stored on the MDS,
1011 in particular ACLs, SELinux, or user attributes (if user_xattr
1012 is specified for the client mount options) then there is a risk
1013 of attribute loss. Additionally, the Lustre file striping
1014 needs to be larger than default (e.g. striped over all OSTs),
1015 and an additional attribute must be stored initially in the
1016 inode and then increase in size enough to be moved to the
1017 external attribute block (e.g. ACL growing in size) for file
1020 --------------------------------------------------------------------------------
1022 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1024 * Support for kernels:
1025 2.6.5-7.283 (SLES 9),
1026 2.6.9-55.EL (RHEL 4),
1027 2.6.16.46-0.14 (SLES 10),
1028 2.6.18.8 vanilla (kernel.org)
1029 * Client support for unpatched kernels:
1030 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1031 2.6.16 - 2.6.22 vanilla (kernel.org)
1032 * Due to recently discovered recovery problems, we do not recommend
1033 using patchless RHEL 4 clients with this or any earlier release.
1034 * Recommended e2fsprogs version: 1.39.cfs8
1035 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1036 * Starting with this release, the ldiskfs backing filesystem required
1037 by Lustre is now in its own package, lustre-ldiskfs. This package
1038 should be installed. It is versioned separately from Lustre and
1039 may be released separately in future.
1043 Description: block reactivating mgc import until all deactivates complete
1044 Details : Fix race when failing back MDT/MGS to itself (testing)
1047 Frequency : at statup only
1049 Description: mds_lov_synchronize race leads to various problems
1050 Details : simultaneous MDT->OST connections at startup can cause the
1051 sync to abort, leaving the OSC in a bad state.
1053 Severity : enhancement
1055 Description: add optional extra BUILD_VERSION info
1056 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1057 to override the lustre version.
1060 Frequency : 2.6.18 servers only
1062 Description: ll_kern_mount() doesn't release the module reference
1063 Details : The ldiskfs module reference count never drops down to 0
1064 because ll_kern_mount() doesn't release the module reference.
1069 Description: server LBUG when using old ost_num_threads parameter
1070 Details : Accept the old ost_num_threads parameter but warn that it
1071 is deprecated, and fix an off-by-one error that caused an LBUG.
1076 Description: Transient SCSI error results in persistent IO issue
1077 Details : iobuf->dr_error is not reinitialized to 0 between two
1081 Frequency : sometimes when underlying device returns I/O errors
1083 Description: OSTs not going read-only during write failures
1084 Details : OSTs are not remounted read-only when the journal commit threads
1085 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1086 instead of the ext3 wrappers.
1090 Description: poor connect scaling with increasing client count
1091 Details : Don't run filter_grant_sanity_check for more than 100 exports
1092 to improve scaling for large numbers of clients.
1095 Frequency : SLES10 only
1097 Description: sanity-quota.sh quotacheck failed: rc = -22
1098 Details : Quotas cannot be enabled on SLES10.
1101 Frequency : liblustre clients only
1103 Description: getdirentries does not give error when run on compute nodes
1104 Details : getdirentries does not fail when the size specified as an argument
1105 is too small to contain at least one entry
1107 Severity : enhancement
1109 Description: Add LNET router traceability for debug purposes
1110 Details : If a checksum failure occurs with a router as part of the
1111 IO path, the NID of the last router that forwarded the bulk data
1112 is printed so it can be identified.
1117 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1118 Details : Due to a race condition, liblustre clients were occasionally
1119 evicted incorrectly.
1121 Severity : enhancement
1123 Description: lfs setstripe use optional parameters instead of postional
1126 Severity : enhancement
1128 Description: Nanosecond timestamp support for ldiskfs
1129 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1130 resolution timestamps. There is not yet support for this at
1131 the Lustre filesystem level.
1134 Frequency : during server recovery
1136 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1137 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1141 Frequency : nfs export on patchless client
1143 Description: connectathon hang when test nfs export over patchless client
1144 Details : Disconnected dentry cannot be found with lookup, so we do not need
1145 to unhash it or make it invalid
1148 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1149 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1150 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1155 Description: open req refcounting wrong on reconnect
1156 Details : If reconnect happened between getting open reply from server and
1157 call to mdc_set_replay_data in ll_file_open, we will schedule
1158 replay for unreferenced request that we are about to free.
1159 Subsequent close will crash in variety of ways.
1160 Check that request is still eligible for replay in
1161 mdc_set_replay_data().
1166 Description: disable writes to filesystem when reading health_check file
1167 Details : the default for reading the health_check proc file has changed
1168 to NOT do a journal transaction and write to disk, because this
1169 can cause reads of the /proc file to hang and block HA state
1170 checking on a healthy but otherwise heavily loaded system. It
1171 is possible to return to the previous behaviour during configure
1172 with --enable-health-write.
1174 Severity : enhancement
1176 Description: 64-bit inode version
1177 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1178 to the inode. This will be required for version-based recovery.
1183 Description: MDS fails to start if a duplicate client export is detected
1184 Details : in some rare cases it was possible for a client to connect to
1185 an MDS multiple times. Upon recovery the MDS would detect this
1186 and fail during startup. Handle this more gracefully.
1188 Severity : enhancement
1190 Description: Add -o localflock option to simulate old noflock
1192 Details : This will achieve local-only flock/fcntl locks
1198 Description: log_commit_thread vs filter_destroy race leads to crash
1199 Details : Take import reference before releasing llog record semaphore
1204 Description: Wrong request locking in request set processing
1205 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1206 imp_delayed_list, in this place should be used imp_lock.
1209 Frequency : when reconnection
1211 Description: Grant Leak when osc reconnect to OST
1212 Details : When osc reconnect ost, OST(filter) should check whether it
1213 should grant more space to client by comparing fed_grant and
1214 cl_avail_grant, and return the granted space to client instead
1215 of "new granted" space, because client will call osc_init_grant
1216 to update the client grant space info.
1219 Frequency : when client reconnect to OST
1221 Description: Grant Leak when osc do resend and replay bulk write
1222 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1223 bulk write request, because the grant info while be sync between
1224 OSC and OST when reconnect, and we should ignore the grant info
1225 these of resend/replay write req.
1230 Description: Grant space more than avaiable left space sometimes.
1231 Details : When then OST is about to be full, if two bulk writing from
1232 different clients came to OST. Accord the avaliable space of the
1233 OST, the first req should be permitted, and the second one
1234 should be denied by ENOSPC. But if the seconde arrived before
1235 the first one is commited. The OST might wrongly permit second
1236 writing, which will cause grant space > avaiable space.
1239 Frequency : when client is evicted
1241 Description: Grant might be wrongly erased when osc is evicted by OST
1242 Details : when the import is evicted by server, it will fork another
1243 thread ptlrpc_invalidate_import_thread to invalidate the
1244 import, where the grant will be set to 0. While the original
1245 thread will update the grant it got when connecting. So if
1246 the former happened latter, the grant will be wrongly errased
1247 because of this race.
1252 Description: Checking Stale with correct fid
1253 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1254 to check whether it is stale, because sometimes, we want the
1255 enqueue happened anyway, and op_data.fid2 will not be initialized.
1257 Severity : enhancement
1259 Description: update patchless client
1260 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1263 Frequency : only with 2.4 kernel
1265 Description: random memory corruption
1266 Details : size of struct ll_inode_info is to big for union inode.u and this
1267 can be cause of random memory corruption.
1272 Description: Memory leak in recovery
1273 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1274 It should also check obd_fail before fsfilt_start, otherwise if
1275 fsfilt_start return -EROFS,(failover mds during mds recovery).
1276 then the req will return with repmsg->transno = 0 and rc = EROFS.
1277 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1278 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1279 be freed no matter whether fsfilt_commit success or not.
1282 Frequency : only with huge count clients
1284 Description: Prevents from taking the superblock lock in llap_from_page for
1286 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1287 for llap_from_page call in ll_removepage prevents from taking the
1288 superblock lock for a soon died page.
1293 Description: Not check open intent error before release open handle
1294 Details : in some rare cases, the open intent error is not checked before
1295 release open handle, which may cause
1296 ASSERTION(open_req->rq_transno != 0), because it tries to release
1297 the failed open handle.
1302 Description: Set cat log bitmap only after create log success.
1303 Details : in some rare cases, the cat log bitmap is set too early. and it
1304 should be set only after create log success.
1308 Description: Accessing a block bevice can re-enable I/O when Lustre is
1309 tearing down a device.
1310 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1315 Description: service threads may hog cpus when there are a lot of requests
1316 Details : Insert cond_resched to give other threads a chance to use some CPU
1321 Description: the cat log was not initialized in recovery
1322 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1323 unlink log on mds will not be initialized until mds post
1324 recovery. And also in mds post recovery, the unlink log will
1325 initialization will be done asynchronausly, so there will be race
1326 between add unlink log and unlink log initialization.
1330 Description: brw_stats were being printed incorrectly
1331 Details : brw_stats were being printed as log2 but all of them were not
1332 recorded as log2. Also remove some code duplication arising from
1333 filter_tally_{read,write}.
1337 Frequency : rare, only in recovery.
1338 Description: ASSERTION(req->rq_type != LI_POISON) failed
1339 Details : imp_lock should be held while iterating over imp_sending_list for
1340 prevent destroy request after get timeout in ptlrpc_queue_wait.
1344 Description: replay-single.sh test 52 fails
1345 Details : A lock's skiplist need to be cleanup when it being unlinked
1346 from its resource list.
1350 Description: Short directio read returns full requested size rather than
1352 Details : Direct I/O operations should return actual amount of bytes
1353 transferred rather than requested size.
1355 Severity : enhancement
1357 Description: metadata RPC reduction (e.g. for rm performance)
1358 Details : decrease the amount of synchronous RPC between clients and servers
1359 by canceling conflicing lock before the operation on the client
1360 and packing thier handles into the main operation RPC to server.
1362 Severity : enhancement
1364 Description: Async OSC create to avoid the blocking unnecessarily.
1365 Details : If a OST has no remain object, system will block on the creating
1366 when need to create a new object on this OST. Now, ways use
1367 pre-created objects when available, instead of blocking on an
1368 empty osc while others are not empty. If we must block, we block
1369 for the shortest possible period of time.
1371 Severity : enhancement
1373 Description: refine locking for avoid write wrong info into lov_objid file
1374 Details : fix possible races with add new target and write/update data in
1377 --------------------------------------------------------------------------------
1379 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1384 Frequency : on some architectures
1386 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1387 Details : Uninitialized flags sometimes cause configuration commands to
1391 Frequency : patchless clients only
1393 Description: missing __iget() symbol export
1394 Details : The __iget() symbol export is missing. To avoid the need for
1395 this on patchless clients the deathrow inode reaper is turned
1396 off, and we depend on the VM to clean up old inodes. This
1397 dependency was during via the fix for bug 12181.
1399 --------------------------------------------------------------------------------
1401 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1403 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1404 INTEROPERATE with older versions automatically. In many cases a
1405 special upgrade step is needed. Please read the
1406 user documentation before upgrading any part of a live system.
1407 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1408 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1409 * WARNING: Lustre configuration and startup changes are required with
1410 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1416 Severity : enhancement
1418 Description: MountConf
1419 Details : Lustre configuration is now managed via mkfs and mount
1420 commands instead of lmc and lconf. New obd types (MGS, MGC)
1421 are added for dynamic configuration management. See
1422 https://mail.clusterfs.com/wikis/lustre/MountConf for
1425 Severity : enhancement
1427 Description: dynamic OST addition
1428 Details : OSTs can now be added to a live filesystem
1430 Severity : enhancement
1432 Description: startup order invariance
1433 Details : MDTs and OSTs can be started in any order. Clients only
1434 require the MDT to complete startup.
1436 Severity : enhancement
1438 Description: parallel, asynchronous orphan cleanup
1439 Details : orphan cleanup is now performed in separate threads for each
1440 OST, allowing parallel non-blocking operation.
1442 Severity : enhancement
1444 Description: optimized stripe assignment
1445 Details : stripe assignments are now made based on ost space available,
1446 ost previous usage, and OSS previous usage, in order to try
1447 to optimize storage space and networking resources.
1449 Severity : enhancement
1451 Description: Permanently set tunables
1452 Details : All writable /proc/fs/lustre tunables can now be permanently
1453 set on a per-server basis, at mkfs time or on a live system.
1455 Severity : enhancement
1457 Description: Lustre message v2
1458 Details : Add lustre message format v2.
1460 Severity : enhancement
1462 Description: client OST exclusion list
1463 Details : Clients can be started with a list of OSTs that should be
1464 declared "inactive" for known non-responsive OSTs.
1468 Description: SPEC SFS validation failure on NFS v2 over lustre.
1469 Details : Changes the blocksize for regular files to be 2x RPC size,
1470 and not depend on stripe size.
1472 Severity : enhancement
1474 Description: Multiple MD RPCs in flight.
1475 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1476 To avoid overly-overloading MDS, introduce a limit on number of
1477 MDS RPCs in flight for a single client and add /proc controls
1478 to adjust this limit.
1480 Severity : enhancement
1482 Description: client read/write statistics
1483 Details : Add client read/write call usage stats for performance
1484 analysis of user processes.
1485 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1486 file access. extents_stats shows chunk size distribution.
1487 extents_stats_per_process show chunk size distribution per
1490 Severity : enhancement
1492 Description: mds statistics
1493 Details : Add detailed mds operations statistics in
1494 /proc/fs/lustre/mds/*/stats.
1498 Description: Failure of copying files with lustre special EAs.
1499 Details : Client side always return success for setxattr call for lustre
1500 special xattr (currently only "trusted.lov").
1505 Description: Refcount LNET uuids
1506 Details : The global LNET uuid list grew linearly with every startup;
1507 refcount repeated list entries instead of always adding to
1510 Severity : enhancement
1512 Description: Dynamic service threads
1513 Details : Within a small range, start extra service threads
1514 automatically when the request queue builds up.
1517 Frequency : mixed-endian client/server environments
1519 Description: mixed-endian crashes
1520 Details : The new msg_v2 system had some failures in mixed-endian
1523 Severity : enhancement
1525 Description: Easy OST removal
1526 Details : OSTs can be permanently deactivated with e.g. 'lctl
1527 conf_param lustre-OST0001.osc.active=0'
1529 Severity : enhancement
1531 Description: MGS proc entries
1532 Details : Added basic proc entries for the MGS showing what filesystems
1535 Severity : enhancement
1537 Description: provide MGS failover
1538 Details : Added config lock reacquisition after MGS server failover.
1540 Severity : enhancement
1542 Description: add Linux 2.4 support
1543 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1547 Description: a large application tries to do I/O to the same resource and dies
1548 in the middle of it.
1549 Details : Check the req->rq_arrival time after the call to
1550 ost_brw_lock_get(), but before we do anything about
1551 processing it & sending the BULK transfer request. This
1552 should help move old stale pending locks off the queue as
1553 quickly as obd_timeout.
1556 Frequency : when an incorrect nid is specified during startup
1558 Description: ptlrpc connect to non-existant node causes kernel crash
1559 Details : LNET can't be re-entered from an event callback, which
1560 happened when we expire a message after the export has been
1561 cleaned up. Instead, hand the zombie cleanup off to another
1564 Severity : enhancement
1566 Description: plain/inodebits lock performance improvement
1567 Details : Grouping plain/inodebits in granted list by their request modes
1568 and bits policy, thus improving the performance of search through
1572 Frequency : only if OST filesystem is corrupted
1574 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1575 Details : for a short time RPCs with bulk IO are in the replay list,
1576 but replay of bulk IOs is unimplemented. If the OST filesystem
1577 is corrupted due to disk cache incoherency and then replay is
1578 started it is possible to trip an assertion. Avoid putting
1579 committed RPCs into the replay list at all to avoid this issue.
1582 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1584 Description: System hang on startup
1585 Details : This bug allowed the liblustre (e.g. catamount) client to
1586 return to the app before handling all startup RPCs. This
1587 could leave the node unresponsive to lustre network traffic
1588 and manifested as a server ptllnd timeout.
1590 Severity : enhancement
1592 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1593 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1594 boolean to control whether to print peer debug info when a
1595 client's RPC times out.
1598 Frequency : only for kernels with patches from Lustre below 1.4.3
1600 Description: Remove old rdonly API
1601 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1604 Frequency : only for devices with external journals
1606 Description: Set external device read-only also
1607 Details : During a commanded failover stop, we set the disk device
1608 read-only while the server shuts down. We now also set any
1609 external journal device read-only at the same time.
1612 Frequency : when upgrading from 1.4 while trying to change parameters
1614 Description: The wrong (new) MDC name was used when setting parameters for
1615 upgraded MDT's. Also allows changing of OSC (and MDC)
1616 parameters if --writeconf is specified at tunefs upgrade time.
1619 Frequency : when setting specific ost indicies
1621 Description: QOS code breaks on skipped indicies
1622 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1623 created with --index need not be sequential.
1625 Severity : enhancement
1627 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1628 Details : The uninit_groups feature works in conjunction with the kernel
1629 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1630 up the pass1 processing of e2fsck. This is a read-only feature
1631 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1632 mount filesystems that have had this feature enabled.
1634 Severity : enhancement
1636 Description: Improve multi-block allocation algorithm to avoid fragmentation
1637 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1638 allocation locality and avoid filesystem fragmentation.
1640 ------------------------------------------------------------------------------
1642 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1644 * Support for kernels:
1645 2.4.21-47.0.1.EL (RHEL 3)
1646 2.6.5-7.283 (SLES 9)
1647 2.6.9-42.0.10.EL (RHEL 4)
1648 2.6.12.6 vanilla (kernel.org)
1649 2.6.16.27-0.9 (SLES 10)
1650 * Recommended e2fsprogs version: 1.39.cfs5
1652 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1656 Frequency : occasional, depends on client load and configuration
1657 Bugzilla : 12181, 12203
1658 Description: data loss for recently-modified files
1660 Details : In some cases it is possible that recently written or created
1661 files may not be written to disk in a timely manner (this should
1662 normally be within 30s unless client IO load is very high).
1663 The problem appears as zero-length files or files that are a
1664 multiple of 1MB in size after a client crash or client eviction
1665 that are missing data at the end of the file.
1667 This problem is more likely to be hit on clients where files are
1668 repeatedly created and unlinked in the same directory, clients
1669 have a large amount of RAM, have many CPUs, the filesystem has
1670 many OSTs, the clients are rebooted frequently, and/or the files
1671 are not accessed by other nodes after being written.
1673 The presence of the problem can be detected by looking at
1674 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1675 smaller than the second (nr_unused) then dirty files will not
1676 be flushed automatically to disk. "sync; sleep 10" should be
1677 run several times on the node before unmounting it to update
1678 Lustre (this is also safe to run on nodes without this problem).
1680 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1681 that can cause this same problem, so customers using that kernel
1682 also need to update the kernel in addition to Lustre. In order
1683 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1685 It is normal that files written just before a client crash (less
1686 than 30s) may not yet have been flushed to disk, even for local
1690 Frequency : frequent on thin XT3 nodes
1692 Description: UUID collision on thin XT3 Linux nodes
1693 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1694 randomly, since we relied on an insufficiently-seeded PRNG.
1699 Description: OSS hangs after "All ost request buffers busy"
1700 Details : A deadlock between quota and journal operations caused OSS
1701 hangs after printing "All ost request buffers busy."
1704 Frequency : always on liblustre builds
1706 Description: Cleanup compiler warnings on liblustre
1709 Frequency : always on liblustre builds on XT3
1711 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1715 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1716 > 32000 subdirectories
1717 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1718 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1719 subdirectory count crosses 32000. This will aid e2fsck to
1720 correctly handle more than 32000 subdirectories.
1723 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1725 Description: System hang on startup
1726 Details : This bug allowed the liblustre (e.g. catamount) client to
1727 return to the app before handling all startup RPCs. This
1728 could leave the node unresponsive to lustre network traffic
1729 and manifested as a server ptllnd timeout.
1731 Severity : enhancement
1733 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1734 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1735 boolean to control whether to print peer debug info when a
1736 client's RPC times out.
1741 Description: make O_SYNC working on 2.6 kernels
1742 Details : 2.6 kernels use different method for mark pages for write,
1743 so need add a code to lustre for O_SYNC work.
1748 Description: Failure to close file and release space on NFS
1749 Details : Put inode details into lock acquired in ll_intent_file_open.
1750 Use mdc_intent_lock in ll_intent_open to properly
1751 detect all kind of errors unhandled by mdc_enqueue.
1756 Description: proc file read during shutdown sometimes raced obd removal,
1758 Details : Add lock to prevent obd access after proc file removal.
1761 Frequency : Only for files larger than 4GB on 32-bit clients.
1763 Description: improperly doing page alignment of locks
1764 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1765 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1770 Description: RPCs being resent when they shouldn't be
1771 Details : Some RPCs that should not be resent are being resent. This
1772 can cause inconsistencies in the RPC state machine. Do not
1773 resend such requests.
1776 Frequency : rare, only with NFS export
1778 Description: Crash on NFS re-export node
1779 Details : under very unusual load conditions an assertion is hit in
1780 ll_intent_file_open()
1783 Frequency : only if OST filesystem is corrupted
1785 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1786 Details : for a short time RPCs with bulk IO are in the replay list,
1787 but replay of bulk IOs is unimplemented. If the OST filesystem
1788 is corrupted due to disk cache incoherency and then replay is
1789 started it is possible to trip an assertion. Avoid putting
1790 committed RPCs into the replay list at all to avoid this issue.
1795 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1796 Details : Large single O_DIRECT read and write calls can fail to allocate
1797 a sufficiently large buffer to process the request. In case of
1798 allocation failure the allocation is retried with a smaller
1799 buffer and broken into smaller requests.
1801 Severity : enhancement
1803 Description: Add -o localflock option to simulate old noflock behaviour.
1804 Details : This will achieve local-only flock/fcntl locks coherentness.
1809 Description: versioning check is incomplete
1810 Details : Checking the version difference of client vs. server, report
1811 error if the gap is too big.
1816 Description: add support PG_writeback bit
1817 Details : add support for PG_writeback bit for Lustre, for more carefull
1818 work with page cache in 2.6 kernel. This also fix some deadlocks
1819 and remove hack for work O_SYNC with 2.6 kernel.
1821 Severity : enhancement
1823 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1824 Details : The uninit_groups feature works in conjunction with the kernel
1825 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1826 up the pass1 processing of e2fsck. This is a read-only feature
1827 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1828 mount filesystems that have had this feature enabled.
1830 Severity : enhancement
1832 Description: Improve multi-block allocation algorithm to avoid fragmentation
1833 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1834 allocation locality and avoid filesystem fragmentation.
1836 ------------------------------------------------------------------------------
1838 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1840 * Support for kernels:
1841 2.6.9-42.0.3.EL (RHEL 4)
1842 2.6.5-7.276 (SLES 9)
1843 2.4.21-47.0.1.EL (RHEL 3)
1844 2.6.12.6 vanilla (kernel.org)
1845 2.6.16.21-0.8 (SLES10)
1846 * Recommended e2fsprogs version: 1.39.cfs2-0
1848 * The backwards-compatible /proc/sys/portals symlink has been removed
1849 in this release. Before upgrading, please ensure that you change
1850 any configuration scripts or /etc/sysctl.conf files that access
1851 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1852 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1853 in advance of the upgrade on any system running Lustre 1.4.6 or
1854 newer, since /proc/sys/lnet was added in that version.
1855 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1859 Frequency : only when quota is used
1861 Description: avoid scanning export list for quota master
1862 Details : Change the algorithms to avoid scanning export list in order
1863 to improve the efficiency.
1866 Frequency : MDS failover only, very rarely
1868 Description: "went back in time" messages on mds failover
1869 Details : The greatest transno may be lost when the current operation
1870 finishes with an error (transno==0) and the client's last_rcvd
1871 record is over-written. Save the greatest transno in the
1872 mds_last_transno for this case.
1875 Frequency : always for specific kernels and striping counts
1877 Description: client may get "Matching packet too big" without ACL support
1878 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1879 when trying to access files in certain configurations. The
1880 clients should in fact be denied when mounting because they do
1881 not understand ACLs.
1884 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1886 Description: many clients connecting with IO in progress causes connect timeouts
1887 Details : Avoid synchronous journal commits to avoid delays caused by many
1888 clients connecting/disconnecting when bulk IO is in progress.
1889 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1890 OST_IO_PORTAL to avoid delays behind potentially many pending
1894 Frequency : occasionally with multiple writers to a single file
1896 Description: shared writes to file may result in wrong size reported by stat()
1897 Details : Allow growing of kms when extent lock is cancelled
1900 Frequency : always with random mmap IO to multi-striped file
1902 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1903 Details : Only if the hole is at the end of OST object so that kms is too
1904 small. Fix is to increase kms accordingly in ll_nopage.
1907 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1909 Description: writes to a missing object would leak memory on the OST
1910 Details : If there is an inconsistency between the MDS and OST filesystems,
1911 such that the MDS references an object that doesn't exist, writes
1912 to that object will leak memory due to incorrect cleanup in the
1913 error handling path, eventually running out of memory on the OST.
1918 Description: Creating too long symlink causes lustre errors
1919 Details : Check symlink and name lengths before sending requests to MDS.
1922 Frequency : only if flock is enabled (not on by default)
1924 Description: posix locks not released on fd closure on 2.6.9+
1925 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1926 kernels, this caused such locks not to be released on fd close and
1927 then assertions on fs unmount about still used locks.
1930 Frequency : MDS failover only, very rarely
1932 Description: clients may get ASSERTION(granted_lock != NULL)
1933 Details : When request was taking a long time, and a client was resending
1934 a getattr by name lock request. The were multiple lock requests
1935 with the same client lock handle and
1936 mds_getattr_name->fixup_handle_for_resent_request found one of the
1937 lock handles but later failed with ASSERTION(granted_lock != NULL).
1942 Description: handle->h_buffer_credits > 0, assertion failure
1943 Details : h_buffer_credits is zero after truncate, causing assertion
1944 failure. This patch extends the transaction or creates a new
1948 Frequency : NFS re-export or patchless client
1949 Bugzilla : 11179, 10796
1950 Description: Crash on NFS re-export node (__d_move)
1951 Details : We do not want to hash the dentry if we don't have a lock.
1952 But if this dentry is later used in d_move, we'd hit uninitialised
1953 list head d_hash, so we just do this to init d_hash field but
1954 leave dentry unhashed.
1957 Frequency : NFS re-export or patchless client
1959 Description: NFS exports has problem with symbolic link
1960 Details : lustre client didn't properly install dentry when re-exported
1961 to NFS or running patchless client.
1964 Frequency : NFS re-export or patchless client
1966 Description: Various nfs/patchless fixes.
1967 Details : fixes reuse disconected alias for lookup process - this fixes
1968 warning "find_exported_dentry: npd != pd",
1969 fix permission error with open files at nfs.
1970 fix apply umask when do revalidate.
1973 Frequency : occasional
1975 Description: Crash on NFS re-export node
1976 Details : calling clear_page() on the wrong pointer triggered oops in
1977 generic_mapping_read().
1980 Frequency : rarely, using O_DIRECT IO
1982 Description: unaligned directio crashes client with LASSERT
1983 Details : check for unaligned buffers before trying any requests.
1986 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1988 Description: stale data returned from RAID cache
1989 Details : If only a small amount of IO is done to the RAID device before
1990 reading it again it is possible to get stale data from the RAID
1991 cache instead of reading it from disk.
1994 Frequency : always for sles10 kernel
1996 Description: sles10 support
1997 Details : ll_follow_link: compile fixes and using of nd_set_link
1998 under newer kernels.
2001 Frequency : depends on arch, kernel and compiler version, always on sles10
2004 Description: recursive or deep enough symlinks cause stack overflow
2005 Details : getting rid of large stack-allocated variable in
2009 Frequency : depends on hardware
2011 Description: lustre write performance loss in the SLES10 kernel
2012 Details : the performance loss is caused by using of write barriers in the
2013 ext3 code. The SLES10 kernel turns barrier support on by
2014 default. The fix is to undo that change for ldiskfs.
2016 ------------------------------------------------------------------------------
2018 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2020 * Support for kernels:
2021 2.6.9-42.0.3EL (RHEL 4)
2022 2.6.5-7.276 (SLES 9)
2023 2.4.21-47.0.1.EL (RHEL 3)
2024 2.6.12.6 vanilla (kernel.org)
2028 Frequency : quota enabled and large files being deleted
2030 Description: releasing more than 4GB of quota at once hangs OST
2031 Details : If a user deletes more than 4GB of files on a single OST it
2032 will cause the OST to spin in an infinite loop. Release
2033 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2038 Description: statfs data retrieved from /proc may be stale or zero
2039 Details : When reading per-device statfs data from /proc, in the
2040 {kbytes,files}_{total,free,avail} files, it may appear
2041 as zero or be out of date.
2044 Frequency : systems with MD RAID1 external journal devices
2046 Description: lconf's call to blkid is confused by RAID1 journal devices
2047 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2048 of returning all block devices that match the journal UUID.
2051 Frequency : always, for aggregate stripe size over 4GB
2053 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2054 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2055 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2058 Frequency : always if "lfs find" used on a local file/directory
2060 Description: "lfs find" segfaults if used on a local file/directory
2061 Details : The case where a directory component was not specified wasn't
2062 handled correctly. Handle this properly.
2065 Frequency : always on ppc64
2067 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2068 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2069 perform endianness conversion.
2072 Frequency : rarely (truncate to non-zero file size after write under load)
2073 Bugzilla : 10730, 10687
2074 Description: Files padded with zeros to next 4K multiple
2075 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2076 it is possible that files that are truncated to a non-zero size
2077 immediately after being written are filled with zero bytes beyond
2078 the truncated size. No file data is lost.
2080 Severity : enhancement
2082 Description: Allow recovery/failover for liblustre clients.
2083 Details : liblustre clients were unaware of failover configurations until
2086 Severity : enhancement
2088 Description: user file locks should fail when not mounting with flock option
2089 Details : Set up an error-returning stub in ll_file_operations.lock field
2090 to prevent incorrect behaviour when client is mounted without
2091 flock option. Also, set up properly f_op->flock field for
2095 Frequency : always on ia64
2097 Description: "lfs df" loops on printing out MDS statfs information
2098 Details : The obd_ioctl_data was not initialized and in some systems
2099 this caused a failure during the ioctl that did not return
2100 an error. Initialize the struct and return an error on failure.
2103 Frequency : SLES 9 only
2105 Description: Error of copying files with lustre special EAs as root
2106 Details : Client side always return success for setxattr call for lustre
2107 special xattr (currently only "trusted.lov").
2110 Frequency : rarely on clusters with both ia64+i386 clients
2112 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2113 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2114 clients are doing concurrent writes to the same file it is
2115 possible that the ia64 clients may LASSERT because the OST
2116 extent locks are not PAGE_SIZE aligned. Ensure that grown
2117 locks are always aligned on the request boundary.
2120 Frequency : specific use, occasional
2122 Description: Overwriting in use executable truncates on-disk binary image
2123 Details : If one node attempts to overwrite an executable in use by
2124 another node, we now correctly return ETXTBSY instead of
2125 truncating the file.
2127 Severity : enhancement
2129 Description: Async OSC create to avoid the blocking unnecessarily.
2130 Details : If a OST has no remain object, system will block on the creating
2131 when need to create a new object on this OST. Now, ways use
2132 pre-created objects when available, instead of blocking on an
2133 empty osc while others are not empty. If we must block, we block
2134 for the shortest possible period of time.
2139 Description: chmod on Lustre root is propagated to other clients
2140 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2141 invalid by the follow_mount time.
2144 Frequency : liblustre clients only
2146 Description: Race in 'instant cancel' lock handling could lead to such locks
2147 never to be granted in case of SMP MDS
2148 Details : Do not destroy not yet granted but cbpending locks in
2152 Frequency : replay/resend of open
2154 Description: non null lock assetion failure in mds_intent_policy
2155 Details : Trying to replay/resend lockless open requests resulted in
2156 mds_open() returning 0 with no lock. Now it sets a flag if
2157 a lock is going to be returned.
2159 Severity : enhancement
2161 Description: Checksum enhancements
2162 Details : New checksum enhancements allow for resending RPCs that failed
2165 Severity : enhancement
2167 Description: Tunables on number of dirty pages in cacche
2168 Details : Allow to set limit on number of dirty pages cached.
2173 Description: client crash on unmount - lock still has references
2174 Details : In some error handling cases it was possible to leak a lock
2175 reference on a client while accessing a file. This was not
2176 harmful to the client during operation, but would cause the
2177 client to crash when the filesystem is unmounted.
2180 Frequency : specific case, rare
2182 Description: ETXTBSY on mds though file not in use by client
2183 Details : ETXTBSY is no longer incorrectly returned when attempting to
2184 chmod or chown a directory that the user previously tried to
2185 execute or a currently-executing binary.
2188 Frequency : extremely rare except on liblustre-based clients
2190 Description: Lustre space not freed when files are deleted
2191 Details : Clean up open-unlinked files after client eviction. Previously
2192 the unlink was skipped and the files remained as orphans.
2197 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2198 Details : In some cases it was possible to send a blocking callback to
2199 a client doing a glimpse, even though that client didn't get
2200 a lock granted. When the glimpse lock is cancelled on the OST
2201 the freed lock is left on the waiting list and corrupted the list.
2204 Frequency : all core dumps
2206 Description: Broke core dumps to lustre
2207 Details : Negative dentry may be unhashed if parent does not have UPDATE
2208 lock, but some callers, e.g. do_coredump, expect dentry to be
2209 hashed after successful create, hash it in ll_create_it.
2211 ------------------------------------------------------------------------------
2213 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2215 * Support for kernels:
2216 2.6.9-42.0.2.EL (RHEL 4)
2217 2.6.5-7.276 (SLES 9)
2218 2.4.21-40.EL (RHEL 3)
2219 2.6.12.6 vanilla (kernel.org)
2223 Frequency : always on RHEL 3
2225 Description: Number of open files grows over time
2226 Details : The number of open files grows over time, whether or not
2227 Lustre is started. This was due to a filp leak introduced
2228 by one of our kernel patches.
2230 ------------------------------------------------------------------------------
2232 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2234 * Support for kernels:
2235 2.6.9-42.EL (RHEL 4)
2236 2.6.5-7.276 (SLES 9)
2237 2.4.21-40.EL (RHEL 3)
2238 2.6.12.6 vanilla (kernel.org)
2243 Bugzilla : 5719, 9635, 9792, 9684
2244 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2245 Details : If a server is under heavy load and cannot reply to new
2246 connection requests before the client resends the (re)connect,
2247 the connection handling code can behave badly if two service
2248 threads are concurrently handing separate (re)connections from
2249 the same client. Add better locking to the connection handling
2250 code, and ensure that only a single connection will be processed
2251 for a given client UUID, even if the lock is dropped.
2253 Severity : enhancement
2255 Description: add TCP zero-copy support to kernel
2256 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2257 sends if the hardware supports scatter-gather and checksumming.
2258 This allows socklnd to do client-write and server-read more
2259 efficiently and reduce CPU utilization from skbuf copying.
2262 Frequency : only if NFS exporting from client
2264 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2265 Details : If setting attributes on a file created under NFS that had
2266 never been opened it would be possible to oops the client
2267 if the file had no objects.
2270 Frequency : always for liblustre
2272 Description: liblustre client does MDS+OSTs setattr RPC for each write
2273 Details : When doing a write from a liblustre client, the client
2274 incorrectly issued an RPC to the MDS and each OST the file was
2275 striped over in order to update the timestamps. When writing
2276 with small chunks and many clients this could overwhelm the MDS
2277 with RPCs. In all cases it would slow down the write because
2278 these RPCs are unnecessary.
2280 Severity : enhancement
2282 Description: allow number of MDS service threads to be changed at module load
2283 Details : It is now possible to change the number of MDS service threads
2284 running. Adding "options mds mds_num_threads={N}" to the MDS's
2285 /etc/modprobe.conf will set the number of threads for the next
2286 time Lustre is restarted (assuming the "mds" module is also
2287 reloaded at that time). The default number of threads will
2288 stay the same, 32 for most systems.
2293 Description: OST crash if filesystem is unformatted or corrupt
2294 Details : If an OST is started on a device that has never been formatted
2295 or if the filesystem is corrupt and cannot even mount then the
2296 error handling cleanup routines would dereference a NULL pointer.
2301 Description: NULL pointer deref in llap_from_page.
2302 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2303 incorrect) mapping if the page was truncated/reclaimed while it was
2304 searched for. Check for this condition and skip such pages when
2305 doing readahead. Introduce extra check to llap_from_page() to
2306 verify page->mapping->host is non-NULL (so page is not anonymous).
2309 Frequency : Sometimes when using sys_sendfile
2311 Description: "page not covered by a lock" warnings from ll_readpage
2312 Details : sendfile called ll_readpage without right page locks present.
2313 Now we introduced ll_file_sendfile that does necessary locking
2314 around call to generic_file_sendfile() much like we do in
2318 Frequency : with certain MDS communication failures at client mount time
2320 Description: NULL pointer deref after failed client mount
2321 Details : a client connection request may delayed by the network layer
2322 and not be sent until after the PTLRPC layer has timed out the
2323 request. If the client fails the mount immediately it will try
2324 to clean up before the network times out the request. Add a
2325 reference from the request import to the obd device and delay
2326 the cleanup until the network drops the request.
2329 Frequency : occasionally during client (re)connect
2331 Description: assertion failure during client (re)connect
2332 Details : processing a client connection request may be delayed by the
2333 client or server longer than the client connect timeout. This
2334 causes the client to resend the connection request. If the
2335 original connection request is replied in this interval, the
2336 client may trip an assertion failure in ptlrpc_connect_interpret()
2337 which thought it would be the only running connect process.
2340 Frequency : only with obd_echo servers and clients that are rebooted
2342 Description: kernel BUG accessing uninitialized data structure
2343 Details : When running an obd_echo server it did not start the ping_evictor
2344 thread, and when a client was evicted an uninitialized data
2345 structure was accessed. Start the ping_evictor in the RPC
2346 service startup instead of the OBD startup.
2348 Severity : enhancement
2349 Bugzilla : 10193 (patchless)
2350 Description: Remove dependency on various unexported kernel interfaces.
2351 Details : No longer need reparent_to_init, exit_mm, exit_files,
2352 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2355 Frequency : rare (only users of deprecated and unsupported LDAP config)
2357 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2358 Details : LDAP apparently contains 'lustreName' attributes instead of
2359 'name'. A simple remapping of the name is sufficient.
2362 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2364 Description: waiting_locks_callback trips kernel BUG if client is evicted
2365 Details : Running with the dump_on_timeout debug flag turned on makes
2366 it possible that the waiting_locks_callback() can try to dump
2367 the Lustre kernel debug logs from an interrupt handler. Defer
2368 this log dumping to the expired_lock_main() thread.
2370 Severity : enhancement
2372 Description: Support NFS exporting on 2.6 kernels.
2373 Details : Implement non-rawops metadata methods for NFS server to use without
2374 changing NFS server code.
2377 Frequency : very rare (synthetic metadata workload only)
2379 Description: two racing renames might cause an MDS thread to deadlock
2380 Details : Running the "racer" program may cause one MDS thread to rename
2381 a file from being the source of a rename to being the target of
2382 a rename at exactly the same time that another thread is doing
2383 so, and the second thread has already enqueued these locks after
2384 doing a lookup of the target and is trying to relock them in
2385 order. Ensure that we don't try to re-lock the same resource.
2388 Frequency : only very large systems with liblustre clients
2390 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2391 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2392 all OSTs in parallel. This allows the eviction of stale liblustre
2393 clients to proceed much faster than if they were done in series,
2394 and also offers similar improvements for other set_info RPCs.
2399 Description: excessive CPU usage during initial read phase on client
2400 Details : During the initial read phase on a client, it would agressively
2401 retry readahead on the file, consuming too much CPU and impacting
2402 performance (since 1.4.5.8). Improve the readahead algorithm
2403 to avoid this, and also improve some other common cases (read
2404 of small files in particular, where "small" is files smaller than
2405 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2410 Description: MDS crash when receiving packet with unknown intent.
2411 Details : Do not LBUG in unknown intent case, just return -EFAULT
2413 Severity : enhancement
2414 Bugzilla : 9293, 9385
2415 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2416 Details : Do not serialize getattr (non-intent version) and statfs.
2419 Frequency : occasional, when OST network is overloaded/intermittent
2421 Description: client evicted by OST after bulk IO timeout
2422 Details : If a client sends a bulk IO request (read or write) the OST
2423 may evict the client if it is unresposive to its data GET/PUT
2424 request. This is incorrect if the network is overloaded (takes
2425 too long to transfer the RPC data) or dropped the OST GET/PUT
2426 request. There is no need to evict the client at all, since
2427 the pinger and/or lock callbacks will handle this, and the
2428 client can restart the bulk request.
2431 Frequency : Always when mmapping file with no objects
2433 Description: client crashes when mmapping file with no objects
2434 Details : Check that we actually have objects in a file before doing any
2435 operations on objects in ll_vm_open, ll_vm_close and
2441 Description: Request leak when working with deleted CWD
2442 Details : Introduce advanced request refcount tracking for requests
2443 referenced from lustre intent.
2445 Severity : Enhancement
2447 Description: Cache open file handles on client.
2448 Details : MDS now will return special lock along with openhandle, if
2449 requested and client is allowed to hold openhandle, even if unused,
2450 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2451 closing files for every read/write openration.
2453 Severity : Enhancement
2455 Description: Cache open negative dentries on client when possible.
2456 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2457 negative dentries on lock revocation.
2462 Description: Remounting a client read-only wasn't possible with a zconf mount
2463 Details : It wasn't possible to remount a client read-only with llmount.
2465 Severity : enhancement
2466 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2467 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2468 MPICH is included to add Lustre-specific ADIO interfaces.
2469 This is based closely on the UFS ADIO layer and only differs
2470 in file creation, in order to allow the OST striping to be set.
2471 This is user-contributed code and not supported by CFS.
2476 Description: extended inode attributes (immutable, append-only) work improperly
2477 when 2.4 and 2.6 kernels are used on client/server or vice versa
2478 Details : Introduce kernel-independent values for these flags.
2480 Severity : enhancement
2483 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2484 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2485 granularity. Now, Fractional MB granularity is be supported,
2486 this is very useful for low memory system.
2488 Severity : enhancement
2490 Description: Getattr by fid
2491 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2492 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2493 have locks and avoids getattr rpc requests for it on every stat.
2496 Frequency : Always, for filesystems larger than 2TB
2498 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2499 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2500 filesystems larger than 2TB to be mounted with mballoc enabled.
2503 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2505 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2506 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2507 filesystems larger than 2TB is mounted then the kernel will
2508 silently corrupt the start of the filesystem. CONFIG_LBD is
2509 enabled for all CFS-supported kernels, but the possibility of
2510 this happening with a modified kernel config exists.
2512 Severity : enhancement
2514 Description: add client O_DIRECT support for 2.6 kernels
2515 Details : It is now possible to do O_DIRECT reads and writes to files
2516 in the Lustre client mountpoint on 2.6 kernel clients.
2518 Severity : enhancement
2520 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2521 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2522 parallel, not waiting for response from every OST before sending
2523 a rpc to the next OST.
2528 Description: setattr vs write race when updating file timestamps
2529 Details : Client processes that update a file timestamp into the past
2530 right after writing to the file (e.g. tar) it is possible that
2531 the updated file modification time can be reset to the current
2532 time due to a race between processing the setattr and write RPC.
2534 Severity : enhancement
2536 Description: Bring 'lfs find' closer in line with regular Linux find.
2537 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2538 -print0 options and obtains all the needed info through the lustre
2541 Severity : enhancement
2543 Description: support up to 1024 configured devices on one node
2544 Details : change obd_dev array from statically allocated to dynamically
2545 allocated structs as they are first used to reduce memory usage
2550 Description: Flush dirty partially truncated pages during truncate
2551 Details : Immediatelly flush partially truncated pages in filter_setattr,
2552 this way we completely avoid having any pages in page cache on OST
2553 and can retire ugly workarounds during writes to flush such pages.
2558 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2559 Details : protect inode from truncation within vfs_unlink() context
2560 just take a reference before calling vfs_unlink() and release it
2561 when parent's i_sem is free.
2566 Description: last_id value checked outside lock on OST caused LASSERT failure
2567 Details : If there were multiple MDS->OST object precreate requests in
2568 flight, it was possible that the OST's last object id was checked
2569 outside a lock and incorrectly tripped an assertion. Move checks
2570 inside locks, and discard old precreate requests.
2573 Frequency : always, if extents are used on OSTs
2575 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2576 Details : OSTs using the extents format would not zero the high 16 bits of
2577 the index physical block number. This is not a problem for any
2578 OST filesystems smaller than 16TB, and no kernels support ext3
2579 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2580 new/modified files) and can be fixed for existing filesystems
2581 with e2fsprogs-1.39-cfs1.
2586 Description: import connection selection may be incorrect if timer wraps
2587 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2588 import connections to be ignored if the 32-bit jiffies counter
2589 wraps. Use a 64-bit jiffies counter.
2592 Frequency : very large clusters immediately after boot
2594 Description: LNET request buffers exhausted under heavy short-term load
2595 Details : If a large number of client requests are generated on a service
2596 that has previously never seen so many requests it is possible
2597 that the request buffer growth cannot keep up with the spike in
2598 demand. Instead of dropping incoming requests, they are held in
2599 the LND until the RPC service can accept more requests.
2602 Frequency : Sometimes during replay
2604 Description: Assertion failure in ll_local_open after replay.
2605 Details : If replay happened on an open request reply before we were able
2606 to set replay handler, reply will become not swabbed tripping the
2607 assertion in ll_local_open. Now we set the handler right after
2608 recognising of open request
2611 Frequency : very rare
2613 Description: kernel reports "badness in vsnprintf"
2614 Details : Reading from the "recovery_status" /proc file in small chunks
2615 may cause a negative length in lprocfs_obd_rd_recovery_status()
2616 call to vsnprintf() (which is otherwise harmless). Exit early
2617 if there is no more space in the output buffer.
2619 Severity : enhancement
2621 Description: clear OBD RPC statistics by writing to them
2622 Details : It is now possible to clear the OBD RPC statistics by writing
2623 to the "stats" file.
2628 Description: Client mtime is not the same on different clients after utimes
2629 Details : In some cases, the client was using the utimes() syscall on
2630 a file cached on another node. The clients now validate the
2631 ctime from the MDS + OSTs to determine which one is right.
2636 Description: Inability to activate failout mode
2637 Details : lconf script incorrectly assumed that in pythong string's numeric
2638 value is used in comparisons.
2641 Frequency : always with multiple stripes per file
2643 Description: Inefficient object allocation for mutli-stripe files
2644 Details : When selecting which OSTs to stripe files over, for files with
2645 a stripe count that divides evenly into the number of OSTs,
2646 the MDS is always picking the same starting OST for each file.
2647 Return the OST selection heuristic to the original design.
2652 Description: mount failures may take full timeout to return an error
2653 Details : Under some heavy load conditions it is possible that a
2654 failed mount can wait for the full obd_timeout interval,
2655 possibly several minutes, before reporting an error.
2656 Instead return an error as soon as the status is known.
2658 Frequency : quota enabled and large files being deleted
2660 Description: releasing more than 4GB of quota at once hangs OST
2661 Details : If a user deletes more than 4GB of files on a single OST it
2662 will cause the OST to spin in an infinite loop. Release
2663 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2668 Description: statfs data retrieved from /proc may be stale or zero
2669 Details : When reading per-device statfs data from /proc, in the
2670 {kbytes,files}_{total,free,avail} files, it may appear
2671 as zero or be out of date.
2674 Frequency : systems with MD RAID1 external journal devices
2676 Description: lconf's call to blkid is confused by RAID1 journal devices
2677 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2678 of returning all block devices that match the journal UUID.
2681 Frequency : always, for aggregate stripe size over 4GB
2683 Description: assertion fails when trying to use 4GB stripe size
2684 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2685 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2688 Frequency : always on ppc64
2690 Description: the first write on an ext3 filesystem with mballoc got stuck
2691 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2692 perform endianness conversion.
2694 ------------------------------------------------------------------------------
2696 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2698 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2699 INTEROPERATE with older versions automatically. Please read the
2700 user documentation before upgrading any part of a live system.
2701 * WARNING: Lustre networking configuration changes are required with
2702 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2705 * Support for kernels:
2706 2.6.9-22.0.2.EL (RHEL 4)
2707 2.6.5-7.244 (SLES 9)
2708 2.6.12.6 vanilla (kernel.org)
2711 Severity : enhancement
2712 Bugzilla : 7981/8208
2713 Description: Introduced Lustre Networking (LNET)
2714 Details : LNET is new networking infrastructure for Lustre, it includes
2715 a reorganized network configuration mode (see the user
2716 documentation for full details) as well as support for routing
2717 between different network fabrics. Lustre Networking Devices
2718 (LNDs) for the supported network fabrics have also been
2719 created for this new infrastructure.
2721 Severity : enhancement
2722 Description: Introduced Access control lists
2723 Details : clients can set ACLs on files and directories in order to have
2724 more fine-grained permissions than the standard Unix UGO+RWX.
2725 The MDS must be started with the "-o acl" mount option.
2727 Severity : enhancement
2728 Description: Introduced filesystem quotas
2729 Details : Administrators may now establish per-user quotas on the
2732 Severity : enhancement
2734 Description: Configuration change for the XT3
2735 The PTLLND is now used to run Lustre over Portals on the XT3
2736 The configure option(s) --with-cray-portals are no longer used.
2737 Rather --with-portals=<path-to-portals-includes> is used to
2738 enable building on the XT3. In addition to enable XT3 specific
2739 features the option --enable-cray-xt3 must be used.
2744 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2745 Details : A race between allocating a new llcd and re-getting the llcd_lock
2746 allowed another thread to grab newly-allocated llcd.
2748 Severity : enhancement
2750 Description: 2.6 OST async journal commit and locking fix to improve performance
2751 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2752 2.6 kernels are now async as they already were in 2.4 kernels so
2753 that they can commit concurrently with the network bulk transfer.
2754 For block-allocated files the filter allocation semaphore is held
2755 to avoid filesystem fragmentation during allocation. BKL lock
2756 removed for 2.6 xattr operations where it is no longer needed.
2761 Description: lconf incorrectly determined whether two IP networks could talk
2762 Details : In some more complicated routing and multiple-network
2763 configurations, lconf will avoid trying to make a network
2764 connection to a disjoint part of the IP space. It was doing the
2765 math incorrectly for one set of cases.
2770 Description: Fix for potential infinite loop processing records in an llog.
2771 Details : If an llog record is corrupted/zeroed, it is possible to loop
2772 forever in llog_process(). Validate the llog record length
2773 and skip the remainder of the block on error.
2776 Frequency : occasional (liblustre only)
2778 Description: liblustre could not open files whose last component is a symlink
2779 Details : sysio_path_walk() would incorrectly pass the open intent to
2780 intermediate path components.
2783 Frequency : rare (liblustre only with non-standard tuning)
2784 Bugzilla : 7201 (7350)
2785 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2786 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2787 after it has been released, only do this for referenced locks
2789 Severity : enhancement
2791 Description: specifying an (invalid) directory default stripe_size of -1
2792 would reset the directory default striping
2793 Details : stripe_size -1 was used internally to signal directory stripe
2794 removal, now use "all default" to signal dir stripe removal
2795 as a directory striping of "all default" is not useful
2798 Frequency : common for large clusters running liblustre clients
2800 Description: doing an ls when liblustre clients are running is slow
2801 Details : sending a glimpse AST to a liblustre client waits for every AST
2802 to time out, as liblustre clients will not respond. Since they
2803 cannot cache data we refresh the OST lock LVB from disk instead.
2805 Severity : enhancement
2807 Description: doing an ls at the same time as file IO can be slow
2808 Details : enqueue and other "small" requests can be blocked behind many
2809 large IO requests. Create a new OST IO portal for non-IO
2810 requests so they can be processed faster.
2813 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2815 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2816 created file with mode 000
2817 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2818 allow an ftruncate() as a non-root user to a file with mode 000.
2819 Lustre now respects this flag to disable mode checks when
2820 truncating a file owned by the user
2823 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2826 Description: Revoking locks from clients that went dead or catatonic might take
2828 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2829 cancellation of such locks instant on servers without waiting for
2830 any reply from clients. Clients drops these locks when cancel
2831 notification from server is received without replying.
2834 Frequency : liblustre-only, when liblustre client dies or becomes busy
2836 Description: Doing ls on Linux clients can take a long time with active
2838 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2839 granting such locks to it in the first place if possible. Locks
2840 are taken by proxy on the OST during the read or write and
2841 dropped immediately afterward. Add connect flags handling, do
2842 not grant locks to liblustre clients for glimpse ASTs.
2844 Severity : enhancement
2846 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2847 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2848 sequential reads and behaves badly with applications doing random
2849 reads. Improve it by reading ahead at least read region, and
2850 avoiding excessive large RPC for small reads.
2852 Severity : enhancement
2854 Description: Creating more than 1000 files for a single job may cause a load
2855 imbalance on the OSTs if there are also a large number of OSTs.
2856 Details : qos_prep_create() uses an OST index reseed value that is an
2857 even multiple of the number of available OSTs so that if the
2858 reseed happens in the middle of the object allocation it will
2859 still utilize the OSTs as uniformly as possible.
2864 Description: OST or MDS may oops in ping_evictor_main()
2865 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2866 but doesn't restart at beginning of obd_exports_timed list
2869 Severity : enhancement
2871 Description: improve by-nid export eviction on the MDS and OST
2872 Details : allow multiple exports with the same NID to be evicted at one
2873 time without re-searching the exports list.
2876 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2878 Description: MDS may oops in groups_free()
2879 Details : in rare race conditions a newly allocated group_info struct is
2880 freed again, and this can be NULL. The 2.4 compatibility code
2881 for groups_free() checked for a NULL pointer, but 2.6 did not.
2884 Frequency : common for liblustre clients doing little filesystem IO
2885 Bugzilla : 9352, 7313
2886 Description: server may evict liblustre clients accessing contended locks
2887 Details : if a client is granted a lock or receives a completion AST
2888 with a blocking AST already set it would not reply to the AST
2889 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2892 Frequency : lfs setstripe, only systems with more than 160 OSTs
2894 Description: unable to set striping with a starting offset beyond OST 160
2895 Details : llapi_create_file() incorrectly limited the starting stripe
2896 index to the maximum single-file stripe count.
2899 Frequency : LDAP users only
2901 Description: lconf did not handle in-kernel recovery with LDAP properly
2902 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2904 Severity : enhancement
2906 Description: bind OST threads to NUMA nodes to improve performance
2907 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2908 node and do their allocations there to localize memory access
2910 Severity : enhancement
2912 Description: llmount can determine client NID directly from Myrinet (GM)
2913 Details : the client NID code from gmnalnid was moved directly into
2914 llmount, removing the need to use this or specifying the
2915 client NID explicitly when mounting GM clients with zeroconf
2918 Frequency : if client is started with down MDS
2920 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2921 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2922 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2923 ptlrpc_import_delayed_req() because we didn't check if the
2924 request was interrupted, and we also didn't break out of the
2925 event loop if there was no timeout
2930 Description: data loss during non-page-aligned writes to a single file from
2931 both multiple nodes and multiple threads on one node at same time
2932 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2933 inconsistency led to false short-reads, that were cached and later
2934 used by ->prepare_write() to fill in partially written page,
2935 leading to data loss.
2938 Frequency : always, if lconf --abort_recovery used
2940 Description: lconf --abort_recovery fails with 'Operation not supported'
2941 Details : lconf was attempting to abort recovery on the MDT device and not
2944 Severity : enhancement
2946 Description: remove cleanup logs
2947 Details : replace lconf-generated cleanup logs with lustre internal
2948 cleanup routines. Eliminates the need for client-cleanup and
2951 Severity : enhancement
2953 Description: add support for EAs (user and system) on lustre filesystems
2954 Details : it is now possible to store extended attributes in the Lustre
2955 client filesystem, and with the user_xattr mount option it
2956 is possible to allow users to store EAs on their files also
2958 Severity : enhancement
2960 Description: Add possibility (config option) to show minimal available OST free
2962 Details : When compiled with --enable-mindf configure option, statfs(2)
2963 (and so, df) will return least minimal free space available from
2964 all OSTs as amount of free space on FS, instead of summary of
2965 free spaces of all OSTs.
2967 Severity : enhancement
2969 Description: do not expand extent locks acquired on OST-side
2970 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2971 by server: they are not cached anyway.
2974 Frequency : when mmap is used/binaries executed from Lustre
2976 Description: Unmmap pages before throwing them away from read cache.
2977 Details : llap_shrink cache now attempts to unmap pages before discarding
2978 them (if unmapping failed - do not discard). SLES9 kernel has
2979 extra checks that trigger if this unmapping is not done first.
2984 Description: lconf didn't resolve symlinks before checking to see whether a
2985 given mountpoint was already in use
2988 Frequency : when migrating failover services
2989 Bugzilla : 6395, 9514
2990 Description: When migrating a subset of services from a node (e.g. failback
2991 from a failover service node) the remaining services would
2992 time out and evict clients.
2993 Details : lconf --force (implied by --failover) sets the global obd_timeout
2994 to 5 seconds in order to quickly disconnect, but this caused
2995 other RPCs to time out too quickly. Do not change the global
2996 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2998 Severity : enhancement
2999 Frequency : if MDS is started with down OST
3000 Bugzilla : 9439,5706
3001 Description: Allow startup/shutdown of an MDS without depending on the
3002 availability of the OSTs.
3003 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3004 Add appropriate locking and lov-osc refcounts for safe
3005 cleaning. Add osc abort_inflight calls in case the
3006 synchronize never started.
3009 Frequency : occasional (Cray XT3 only)
3011 Description: root not authorized to access files in CRAY_PORTALS environment
3012 Details : The client process capabilities were not honoured on the MDS in
3013 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3014 been accessed by an authorized user then root was able to access
3015 the file on the local client also. The root user capabilities
3016 are now allowed on the MDS, as this environment has secure UID.
3019 Frequency : occasional
3021 Description: ldiskfs "too long searching" message happens too often
3022 Details : A debugging message (otherwise harmless) prints too often on
3023 the OST console. This has been reduced to only happen when
3024 there are fragmentation problems on the filesystem.
3029 Description: Division by zero in statfs when all OSCs are inactive
3030 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3031 lov_statfs divides by value returned by lov_get_stripecnt().
3035 Bugzilla : 9489, 3273
3036 Description: First write from each client to each OST was only 4kB in size,
3037 to initialize client writeback cache, which caused sub-optimal
3038 RPCs and poor layout on disk for the first writen file.
3039 Details : Clients now request an initial cache grant at (re)connect time
3040 and so that they can start streaming writes to the cache right
3041 away and always do full-sized RPCs if there is enough data.
3042 If the OST is rebooted the client also re-establishes its grant
3043 so that client cached writes will be honoured under the grant.
3048 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3049 Details : Now I/O RPCs go to different portal number and (presumably) fast
3050 lock requests (and glimses) and other RPCs get their own service
3051 threads pool that should be able to service those RPCs
3054 Severity : enhancement
3056 Description: Ability to exchange lustre version between client and servers and
3057 issue warnings at client side if client is too old. Also for
3058 liblustre clients there is ability to refuse connection of too old
3060 Details : New 'version' field is added to connect data structure that is
3061 filled with version info. That info is later checked by server and
3065 Frequency : rare, liblustre only.
3066 Bugzilla : 9296, 9581
3067 Description: Two simultaneous writes from liblustre at offset within same page
3068 might proceed at the same time overwriting eachother with stale
3070 Details : I/O lock withing llu_file_prwv was released too early, before data
3071 actually was hitting the wire. Extended lock-holding time until
3072 server acknowledges receiving data.
3075 Frequency : extremely rare. Never observed in practice.
3077 Description: avoid generating lustre_handle cookie of 0.
3078 Details : class_handle_hash() generates handle cookies by incrementing
3079 global counter, and can hit 0 occasionaly (this is unlikely, but
3080 not impossible, because initial value of cookie counter is
3081 selected randonly). Value of 0 is used as a sentinel meaning
3082 "unassigned handle" --- avoid it. Also coalesce two critical
3083 sections in this function into one.
3085 Severity : enhancement
3087 Description: allow liblustre clients to delegate truncate locking to OST
3088 Details : To avoid overhead of locking, liblustre client instructs OST to
3089 take extent lock in ost_punch() on client's behalf. New connection
3090 flag is added to handle backward compatibility.
3092 Severity : enhancement
3093 Bugzilla : 4928, 7341, 9758
3094 Description: allow number of OST service threads to be specified
3095 Details : a module parameter allows the number of OST service threads
3096 to be specified via "options ost ost_num_threads={N}" in the
3097 OSS's /etc/modules.conf or /etc/modprobe.conf.
3101 Bugzilla : 6146, 9635, 9895
3102 Description: servers crash with bad pointer in target_handle_connect()
3103 Details : In rare cases when a client is reconnecting it was possible that
3104 the connection request was the last reference for that export.
3105 We would temporarily drop the export reference and get a new
3106 one, but this may have been the last reference and the export
3107 was just destroyed. Get new reference before dropping old one.
3109 Severity : enhancement
3110 Frequency : if client is started with failover MDS
3112 Description: Allow multiple MDS hostnames in the mount command
3113 Details : Try to read the configuration from all specified MDS
3114 hostnames during a client mount in case the "primary"
3117 Severity : enhancement
3119 Description: Stop sending data to evicted clients as soon as possible.
3120 Details : Check if the client we are about to send or are sending data to
3121 was evicted already. (Check is done every second of waiting,
3122 for which l_wait_event interface was extended to allow checking
3123 of exit condition at specified intervals).
3126 Frequency : rare, normally only when NFS exporting is done from client
3128 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3130 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3133 Frequency : always, if srand() is called before liblustre initialization
3135 Description: Liblustre uses system PRNG disturbing its usage by user application
3136 Details : Introduce internal to lustre fast and high-quality PRNG for
3137 lustre usage and make liblustre and some other places in generic
3138 lustre code to use it.
3140 Severity : enhancement
3141 Bugzilla : 9477, 9557, 9870
3142 Description: Verify that the MDS configuration logs are updated when xml is
3143 Details : Check if the .xml configuration logs are newer than the config
3144 logs stored on the MDS and report an error if this is the case.
3145 Request --write-conf, or allow starting with --old_conf.
3147 Severity : enhancement
3149 Description: Handle symlinks in the path when checking if Lustre is mounted.
3150 Details : Resolve intermediate symlinks when checking if a client has
3151 mounted a filesystem to avoid duplicate client mounts.
3156 Description: lconf can hit an error exception but still return success.
3157 Details : The lconf command catches the Command error exception at the top
3158 level script context and will exit with the associated exit
3159 status, but doesn't ensure that this exit status is non-zero.
3164 Description: failure of ptlrpc thread startup can cause oops
3165 Details : Starting a ptlrpc service thread can fail if there are a large
3166 number of threads or the server memory is very fragmented.
3167 Handle this without oopsing.
3170 Frequency : always, only if liblustre and non-default acceptor port was used
3172 Description: liblustre cannot connect to servers with non-default acceptor port
3173 Details : tcpnal_set_default_params() was not called and was therefore
3174 ignoring the environment varaible TCPNAL_PORT, as well as other
3175 TCPNAL_ environment variables
3180 Description: two objects could be created on the same OST for a single file
3181 Details : If an OST is down, in some cases it was possible to create two
3182 objects on a single OST for a single file. No problems other
3183 than potential performance impact and spurious error messages.
3187 Bugzilla : 5681, 9562
3188 Description: Client may oops in ll_unhash_aliases
3189 Details : Client dcache may become inconsistent in race condition.
3190 In some cases "getcwd" can fail if the current directory is
3196 Description: Inode refcounting problems in NFS export code
3197 Details : link_raw functions used to call d_instantiate without obtaining
3198 extra inode reference first.
3202 Bugzilla : 9942, 9903
3203 Description: Referencing freed requests leading to crash, memleaks with NFS.
3204 Details : We used to require that call to ll_revalidate_it was always
3205 followed by ll_lookup_it. Also with revalidate_special() it is
3206 possible to call ll_revalidate_it() twice for the same dentry
3207 even if first occurence returned success. This fix changes semantic
3208 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3209 reference on a request referred from the intent.
3210 ll_intent_release() then releases such a request.
3213 Frequency : rare, normally benchmark loads only
3215 Description: unlinked inodes were kept in memory on the client
3216 Details : If a client is repeatedly creating and unlinking files it
3217 can accumulate a lot of stale inodes in the inode slab cache.
3218 If there is no other client load running this can cause the
3219 client node to run out of memory. Instead flush old inodes
3220 from client cache that have the same inode number as a new inode.
3223 Frequency : SLES9 2.6.5 kernel and long filenames only
3224 Bugzilla : 9969, 10379
3225 Description: utime reports stale NFS file handle
3226 Details : SLES9 uses out-of-dentry names in some cases, which confused
3227 the lustre dentry revalidation. Change it to always use the
3231 Frequency : rare, unless heavy write-truncate concurrency is continuous
3232 Bugzilla : 4180, 6984, 7171, 9963, 9331
3233 Description: OST becomes very slow and/or deadlocked during object unlink
3234 Details : filter_destroy() was holding onto the parent directory lock
3235 while truncating+unlinking objects. For very large objects this
3236 may block other threads for a long time and slow overall OST
3237 responsiveness. It may also be possible to get a lock ordering
3238 deadlock in this case, or run out of journal credits because of
3239 the combined truncate+unlink. Solution is to do object truncate
3240 first in one transaction without parent lock, and then do the
3241 final unlink in a new transaction with the parent lock. This
3242 reduces the lock hold time dramatically.
3245 Frequency : rare, 2.4 kernels only
3247 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3248 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3249 before unmounting their filesystem in order to avoid deadlock.
3250 The kernel_locked() function in 2.4 kernels only checks whether
3251 the kernel lock is held, not whether it is this process that is
3252 holding it as 2.6 kernels do.
3257 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3258 Details : The client ptlrpc code may be trying to reconnect to a down
3259 server before a previous connection attempt has timed out.
3260 Increase the reconnect interval to be longer than the connection
3261 timeout interval to avoid sending duplicate connections to
3265 Frequency : echo_client brw_test command
3267 Description: fix echo_client to work with OST preallocated code
3268 Details : OST preallocation code (5137) didn't take echo_client IO path
3269 into account: echo_client calls filter methods outside of any
3270 OST thread and, hence, there is no per-thread preallocated
3271 pages and buffers to use. Solution: hijack pga pages for IO. As
3272 a byproduct, this avoids unnecessary data copying.
3276 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3277 Description: Client can oops in mdc_commit_close() after open replay
3278 Details : It was possible for the MDS to return an open request with no
3279 transaction number in mds_finish_transno() if the client was
3280 evicted, but without actually returning an error. Clients
3281 would later try to replay that open and may trip an assertion
3282 Simplify the client close codepath, and always return an error
3283 from the MDS in case the open is not successful.
3286 Frequency : rare, 2.6 OSTs only
3288 Description: OST may deadlock under high load on fragmented files
3289 Details : If there was a heavy load and highly-fragmented OST filesystems
3290 it was possible to have all the OST threads deadlock waiting on
3291 allocation of biovecs, because the biovecs were not released
3292 until the entire RPC IO was completed. Instead, release biovecs
3293 as soon as they are complete to ensure forward IO progress.
3295 Severity : enhancement
3297 Description: Support for specifying external journal device at mount
3298 Details : If an OST or MDS device is formatted with an external journal
3299 device, this device major/minor is stored in the ext3 superblock
3300 and may not be valid for failover. Allow detecting and
3301 specifying the external journal at mount time.
3306 Description: Mounting an MDS with pending unlinked files may cause oops
3307 Details : target_finish_recovery() calls mds_postrecov() which returned
3308 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3309 considers this an error and immediately begins cleaning up the
3310 lov, just after starting the mds_lov process
3312 Severity : enhancement
3314 Description: Implement 'lfs df' to report actual free space on per-OST basis
3315 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3316 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3317 usage of objects; '-h' to report in human readable format.
3319 ------------------------------------------------------------------------------
3321 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3328 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3329 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3330 at startup during orphan unlinking. Instead, skip buddy update
3331 before setup, as it will be regenerated after recovery is complete.
3336 Description: If an OST is inactive, its locks might reference stale inodes.
3337 Details : lov_change_cbdata() must iterate over all namespaces, even if
3338 they are inactive to clear inode references from the lock.
3340 Severity : enhancement
3341 Frequency : occasional, if non-standard max_dirty_mb used
3343 Description: Client will block write RPCs if not enough grant
3344 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3345 then the client will block writes while waiting for another RPC
3346 to complete instead of consuming its dirty limit. With change
3347 we get improved performance when max_dirty_mb is small.
3349 Severity : enhancement
3350 Bugzilla : 3389, 6253
3351 Description: Add support for supplementary groups on the MDS.
3352 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3353 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3354 lookups for user supplementary groups into a cache.
3358 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3359 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3360 for liblustre O_EXCL is enforced by MDS.
3364 Description: readdir never returns NULL in liblustre.
3365 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3366 dentry in struct dirent.
3370 Frequency : liblustre only, depends on application IO pattern
3371 Description: liblustre clients evicted if not contacting servers
3372 Details : Don't put liblustre clients into the ping_evictor list, so
3373 they will not be evicted by the pinger ever.
3375 Severity : enhancement
3377 Description: Add ability to evict clients by NID from MDS.
3378 Details : By echoing "nid:$NID" string into
3379 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3380 $NID would be instantly evicted from this MDS and from all active
3381 OSTs connected to it.
3385 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3386 Details : lookup_it_finish() used to query file size from OSTs that was not
3391 Description: service threads change working directory to that of init
3392 Details : Starting lustre service threads may pin the working directory
3393 of the parent thread, making that filesystem busy. Threads
3394 now change to the working directory of init to avoid this.
3398 Frequency : during shutdown only
3399 Description: shutdown with a failed MDS or OST can cause unmount to hang
3400 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3405 Frequency : 2.6 only
3406 Description: chmod/chown may include an extra supplementary group
3407 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3408 supplementary group and if none is specified this is used.
3411 Bugzilla : 5479 (6816)
3413 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3414 Details : If lookup is in progress on a file that is unlinked we might try
3415 to revalidate the inode and fail in revalidate after lookup is
3416 complete and ll_file_open() enqueues the open again but
3417 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3420 Frequency : always, if lconf --abort_recovery used
3422 Description: lconf --abort_recovery fails with 'Operation not supported'
3423 Details : lconf was attempting to abort recovery on the MDT device and not
3426 ------------------------------------------------------------------------------
3428 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3433 Frequency : rare (only unsupported configurations with a node running as an
3435 Bugzilla : 6514, 5137
3436 Description: Mounting a Lustre file system on a node running as an OST could
3438 Details : OSTs now preallocates memory needed to write out data at
3439 startup, instead of when needed, to avoid having to
3440 allocate memory in possibly low memory situations.
3441 Specifically, if the file system is mounted on on OST,
3442 memory pressure could force it to try to write out data,
3443 which it needed to allocate memory to do. Due to the low
3444 memory, it would be unable to do so and the node would
3445 become unresponsive.
3447 Severity : enhancement
3449 Description: Addition of lconf --service command line option
3450 Details : lconf now accepts a '--service <arg>' option, which is
3451 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3453 Severity : enhancement
3455 Description: Failover mode is now the default for OSTs.
3456 Details : By default, OSTs will now run in failover mode. To return to
3457 the old behaviour, add '--failout' to the lmc line for OSTs.
3459 Severity : enhancement
3461 Description: Health checks are now provided for MDS and OSTs
3462 Details : Additional detailed health check information on MSD and OSTs
3463 is now provided through the procfs health_check value.
3466 Frequency : occasional, depends on IO load
3468 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3469 after numerous create/delete cycles
3470 Details : The ext3 inode allocation policy would not allocate new inodes
3471 very well on the OSTs because there are no new directories
3472 being created. Instead we look for groups with free space if
3473 the parent directories are nearly full.
3478 Description: Network or server problems during mount may cause partially
3479 mounted clients instead of returning an error.
3480 Details : The config llog parsing code may overwrite the error return
3481 code during mount error handling, returning success instead
3487 Description: MDS can fail to allocate large reply buffers
3488 Details : After long uptimes the MDS can fail to allocate large reply
3489 buffers (e.g. zconf client mount config records) due to memory
3490 fragmentation or consumption by the buffer cache. Preallocate
3491 some large reply buffers so that these replies can be sent even
3492 under memory pressure.
3496 Frequency : rare (liblustre)
3497 Description: fsx running with liblustre complained that using truncate() to
3498 extend the file doesn't work. This patch corrects that issue.
3499 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3500 ATTR_SIZE and lsm use in llu_setattr_raw.
3504 Frequency : rare, only 2.6 kernels
3505 Description: Unusual file access patterns on the MDS may result in inode
3506 data being lost in very rare circumstances.
3507 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3508 code in the 2.6 kernel caused the inode and/or EA data not to
3509 be read from disk, causing single-file corruption.
3513 Frequency : rare, only 2.6 filesystems using extents
3514 Description: Heavy concurrent write and delete load may cause data corruption.
3515 Details : It was possible under high-load situations to have an extent
3516 metadata block in the block device cache from a just-unlinked
3517 file overwrite a newly-allocated data block. We now unmap any
3518 metadata buffers that alias just-allocated data blocks.
3522 Frequency : filesystems with default stripe_count larger than 77
3523 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3524 Details : lconf specifies an inode size of 4096 bytes when the default
3525 stripe_count is larger than 77. This conflicts with the default
3526 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3527 this case to avoid pinning too much memory for large EAs.
3529 ------------------------------------------------------------------------------
3531 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3536 Frequency : rare (extremely heavy IO load with hundreds of clients)
3538 Description: Client is evicted, gets IO error writing to file
3539 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3540 caused clients to be evicted for AST timeouts. The fixes in
3541 bug 5192 mean we no longer need to have such short AST timeouts
3542 so ldlm_timeout has been increased.
3545 Frequency : occasional during --force or --failover shutdown under load
3546 Bugzilla : 5949, 4834
3547 Description: Server oops/LBUG if stopped with --force or --failover under load
3548 Details : a collection of import/export refcount and cleanup ordering
3549 issues fixed for safer force cleanup
3552 Frequency : only filesystems larger than 120 OSTs
3553 Bugzilla : 5990, 6223
3554 Description: lfs getstripe would oops on a very large filesystem
3555 Details : lov_getconfig used kfree on vmalloc'd memory
3558 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3559 Bugzilla : 6242, 6243
3560 Description: reading from files that had been truncated to a non-zero size
3561 but never opened returned no data
3562 Details : ll_file_read() reads zeros from no-object files to EOF
3567 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3568 Details : There's little harm in aborting MDS/OSS recovery and letting it
3569 try again, so I removed the LASSERT and return an error instead.
3571 Severity : enhancement
3573 Description: New debugging infrastructure for tracking down data corruption
3574 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3575 (b) cover more of the client-side code path, and (c) try to narrow
3576 down where problems occurred
3580 Bugzilla : 3819, 4364, 4397, 6313
3581 Description: Racing close and eviction MDS could cause assertion in mds_close
3582 Details : It was possible to get multiple mfd references during close and
3583 client eviction, leading to one thread referencing a freed mfd.
3585 Severity: : enhancement
3586 Bugzilla : 3262, 6359
3587 Description: Attempts to reconnect to servers are now more aggressive.
3588 Details : This builds on the enhanced upcall-less recovery that was added
3589 in 1.4.2. When trying to reconnect to servers, clients will
3590 now try each server in the failover group every 10 seconds. By
3591 default, clients would previously try one server every 25 seconds.
3596 Description: After recovery, certain operations trigger a failed
3597 assertion on a client.
3598 Details : Failing over an mds, using lconf -d --failover, while a
3599 client was doing a readdir() call would cause the client to
3600 LBUG after recovery completed and the readdir() was resent.
3602 Severity : enhancement
3604 Description: Default groups are now added by lconf
3605 Details : You can now run lconf --group <servicename> without having to
3606 manually add groups with lmc.
3609 Frequency : occasional
3611 Description: Nodes with an elan id of 0 trigger a failed assertion
3614 Frequency : always when accessing e.g. tty/console device nodes
3616 Description: tty and some other devices nodes cannot be used on lustre
3617 Details : file's private_data field is used by device data and lustre
3618 values in there got lost. New field was added to struct file to
3619 store fs-specific private data.
3622 Frequency : when exporting Lustre via NFS
3624 Description: NFSD failed occasionally when looking up a path component
3625 Details : NFSD is looking up ".." which was broken in ext3 directories
3626 that had grown large enough to become hashed.
3629 Frequency : Clusters with multiple interfaces not on the same subnet
3631 Description: Nodes will repeatedly try to reconnect to an interface which it
3632 cannot reach and report an error to the log.
3633 Details : Extra peer list entries will be created by lconf with some peers
3634 unreachable. lconf now validates the peer before adding it.
3637 Frequency : Only if a default stripe is set on the filesystem root.
3639 Description: Setting a default stripe on the filesystem root prevented the
3640 filesystem from being remounted.
3641 Details : The client was sending extra request flags in the root getattr
3642 request and did not allocate a reply buffer for the dir EA.
3645 Frequency : occasional, higher if lots of files are accessed by one client
3646 Bugzilla : 6159, 6097
3647 Description: Client trips assertion regarding lsm mismatch/magic
3648 Details : While revalidating inodes the VFS looks up inodes with ifind()
3649 and in rare cases can find an inode that is being freed.
3650 The ll_test_inode() code will free the lsm during ifind()
3651 when it finds an existing inode and then the VFS later attaches
3652 this free lsm to a new inode.
3656 Bugzilla : 6422, 7030
3657 Description: MDS deadlock between mkdir and client eviction
3658 Details : Creating a new file via mkdir or mknod (starting a transaction
3659 and getting the ns lock) can deadlock with client eviction
3660 (gets ns lock and trying to finish a synchronous transaction).
3663 Frequency : occasional
3664 Description: While starting a server, the fsfilt_ext3 module could not be
3666 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3667 kernels. Previously, lconf would still use the ext3 name
3668 when trying to load modules. Now, it will correctly use
3669 ext3 on 2.4 and ldiskfs on 2.6.
3671 Severity : enhancement
3672 Description: The default stripe count has been changed to 1
3673 Details : The interpretation of the default stripe count (0, to lfs
3674 or lmc) has been changed to mean striping across a single
3675 OST, rather than all available. For general usage we have
3676 found a stripe count of 1 or 2 works best.
3678 Severity : enhancement
3679 Description: Add support for compiling against Cray portals.
3680 Details : Conditional compiling for some areas that are different
3684 Frequency : occasional
3685 Bugzilla : 6409, 6834
3686 Description: Creating files with an explicit stripe count may lead to
3687 a failed assertion on the MDS
3688 Details : If some OSTs are full or unavailable, creating files may
3689 trigger a failed assertion on the MDS. Now, Lustre will
3690 try to use other servers or return an error to the
3694 Frequency : occasional
3696 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3697 caused lock timeout during readahead (since 1.4.2).
3698 Details : Processes doing readahead might match a lock that hasn't been
3699 granted yet if there are overlapping and conflicting lock
3700 requests. The readahead process waits on ungranted lock
3701 (original lock is CBPENDING), while OST waits for that process
3702 to cancel CBPENDING read lock and eventually evicts client.
3704 Severity : enhancement
3706 Description: Initial enabling of flock support for clients
3707 Details : Implements fcntl advisory locking and file status functions.
3708 This feature is provided as an optional mount flag (default
3709 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3710 locking are implemented yet, and those that are are not guaranteed
3711 to be completely correct in production environments.
3712 mount -t lustre -o [flock|noflock] ...
3715 Frequency : occasional
3717 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3718 assertion in the ext3 JBD (journaling) layer.
3719 Details : The b_committed_data struct is protected by the big kernel lock
3720 in 2.4 kernels, serializing journal_commit_transaction() and
3721 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3722 access to this struct. In 2.6 kernels there is finer grained
3723 locking to improve SMP performance of the JBD layer.
3727 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3732 Description: merge and cleanup kernel patches.
3735 -----------------------------------------------------------------------------
3737 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3739 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3740 versions of Lustre. Please update all servers and clients to
3741 version 1.4.2 or later at the same time. You must also run
3742 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3743 to update the configuration logs.
3745 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3746 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3747 - swab llog records of type '0' so we get proper header size/idx (5861)
3748 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3749 - fix rename of one directory over another leaking an inode (5953)
3750 - avoid SetPageDirty on 2.6 (5981)
3751 - don't re-add just-being-destroyed locks to the waiting list (5653)
3752 - when creating new directories, inherit the parent's custom
3753 striping settings if present parent (3048)
3754 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3755 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3756 - don't hold client locks on temporary worklist from l_lru (5666)
3757 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3758 - automatically evict dead clients (5921)
3759 - Update file size properly in create+truncate+fstat case (6196)
3760 - Do not unhash mountpoint dentries, do not allow removal of
3762 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3763 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3764 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3765 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3766 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3767 - Correctly handle reads of files with no objects (6243)
3768 - lctl recover will also mark a device active if deactivate used (5933)
3770 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3771 - allow --write-conf on an MDS with different nettype than client (5619)
3772 - don't write config llogs to MDS for mounts not from that MDS (5617)
3773 - lconf should create multiple TCP connections from a client (5201)
3774 - init scripts are now turned off by default; run chkconfig --on
3775 lustre and chkconfig --on lustrefs to use them
3776 - upcalls are no longer needed for clients to recover to failover
3778 - add --abort-recovery option to lconf to abort recovery on device
3780 - add support for an arbitrary number of OSTs (3026)
3781 - Quota support protocol changes.
3782 - forward compatibility changes to wire structs (6007)
3783 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3784 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3785 - improved "lustre status" script
3786 - initialize blocksize for non-regular files (6062)
3787 - added --disable-server and --disable-client configure options (5782)
3788 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3789 - Vanilla 2.4.29 support
3790 - increase maximum number of obd devices to 520 (6242)
3791 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3792 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3793 SLES 9 2.6.5 kernels
3794 - limit stripes per file to 160 (the maximum EA size) (6093)
3796 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3799 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3800 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3801 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3802 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3803 - small ext3 extents cleanups and fixes (5733)
3804 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3805 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3806 - avoid needless client->OST connect, fix handle mismatch (5317)
3807 - fix DLM error path that led to out-of-sync client, long delays (5779)
3808 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3809 - fix several locking issues related to i_size (5492,5624,5654,5672)
3810 - don't move pending lock onto export if it is already evicted (5683)
3811 - fix kernel oops when creating .foo in unlinked directory (5548)
3812 - fix deadlock in obdfilter statistics vs. object create (5811)
3813 - use time_{before,after} to avoid timer jiffies wrap (5882)
3814 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3815 - Do not leak request if server was not able to process it (5154)
3816 - If mds_open unable to find parent dir, make that negative lookup(5154)
3817 - don't create new directories with extent-mapping (5909, 5936)
3819 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3820 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3821 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3822 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3823 - support for RHEL4 kernel on i686 (5773)
3824 - provide error messages when incompatible logs are encountered (5898)
3826 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3827 * version 1.4.0.10 (1.4.1 release candidate 1)
3829 - don't keep a lock reference when lock is not granted (4238)
3830 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3831 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3832 - drop import inflight refcount on signal_completed_replay error (5255)
3833 - unlock page after async write error during send (3677)
3834 - handle missing objects in filter_preprw_read properly (5265)
3835 - no transno return for symlink open, don't save no-trasno open (3440)
3836 - don't try to complete elan receive that already failed (4012)
3837 - free RPC server reply state on error (5406)
3838 - clean up thread from ptlrpc_start_thread() on error (5160)
3839 - readahead could read extra page into cache that wasn't ejected (5388)
3840 - prevent races in class_attach/setup/cleanup/detach (5260)
3841 - don't dereference de->d_inode after l_dput of de (5458)
3842 - use "int" for stripe value returned from lock_to_stripe (5544)
3843 - mballoc allocation and error-checking fixes in 2.6 (5504)
3844 - block device patches to fix I/O request sizes in 2.6 (5482)
3845 - look up hostnames for IB nals (5602)
3846 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3847 - don't start multiple acceptors for the same port (5277)
3848 - fix incorrect LASSERT in mds_getattr_name (5635)
3849 - export a proc file for general "ping" checking (5628)
3850 - fix "lfs check" to not block when the MDS is down (5628)
3852 - service request history (4965)
3853 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3854 - create an "evict_client" /proc entry on OSTs, like the MDS has
3855 - fix mount usage message, return errors per mount(8) (5168)
3856 - change grep [] to grep "[]" in tests so they work in more UMLs
3857 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3858 - remove ancient LOV_MAGIC_V0 EA support (5047)
3859 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3860 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3861 - allow readahead allocations to fail when low on memory (5383)
3862 - mmap locking landed again, after considerable improvement (2828)
3863 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3865 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3868 - send OST transaction number in read/write reply to free req (4966)
3869 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3870 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3871 - the watchdog thread now runs as interruptible (5246)
3872 - flock/lockf fixes (but it's still disabled, pending 5135)
3873 - don't use EXT3 constants in llite code (5094)
3874 - memory shortage at startup could cause assertion (5176)
3876 - reorganization of lov code
3877 - single portals codebase
3879 - add extents/mballoc support (5025)
3880 - direct I/O reads in the obdfilter (4048)
3881 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3883 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3886 - send OST transaction number in read/write reply to free req (4966)
3887 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3888 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3889 - the watchdog thread now runs as interruptible (5246)
3890 - handle missing objects in filter_preprw_read properly (5265)
3891 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3892 - drop import inflight refcount on signal_completed_replay error (5255)
3893 - unlock page after async write error during send (3677)
3894 - return original error code on reconstructed replies (3761)
3895 - no transno return for symlink open, don't save no-trasno open (3440)
3897 - add pid to ldlm debugging output (4922)
3898 - bump the watchdog timeouts -- we can't handle 30sec yet
3899 - extra debugging for orphan dentry/inode bug (5259)
3901 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3904 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3905 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3906 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3907 - fix NULL dereference in /proc/sys/portals/routes (4827)
3908 - allow failed mdc_close() operations to be interrupted (4561)
3909 - stop precreate on OST before MDS would time out on it (4778)
3910 - don't send partial-page writes before EOF from client (4410)
3911 - discard client grant for sub-page writes on large-page clients (4520)
3912 - don't free dentries not owned by NFS code, check generation (4806)
3913 - fix lsm leak if mds_create_objects() fails (4801)
3914 - limit debug_daemon file size, always print CERROR messages (4789)
3915 - use transno after validating reply (3892)
3916 - process timed out requests if import state changes (3754)
3917 - update mtime on OST during writes, return in glimpse (4829)
3918 - add mkfsoptions to LDAP (4679)
3919 - use ->max_readahead method instead of zapping global ra (5039)
3920 - don't interrupt __l_wait_event() during strace
3922 - add software watchdogs to catch hung threads quickly (4941)
3923 - make lustrefs init script start after nfs is mounted
3924 - fix CWARN/ERROR duplication (4930)
3925 - return async write errors to application if possible (2248)
3926 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3927 - print NAL number in %x format (4645)
3928 - update barely-supported suse-2.4.21-171 series (4842)
3929 - support for sles 9 %post scripts
3930 - support for building 2.6 kernel-source packages
3931 - support for sles km_* packages
3933 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3936 - ignore -ENOENT errors in osc_destroy (3639)
3937 - notify osc create thread that OSC is being cleaned up (4600)
3938 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3939 - reconstruct ost_handle() like mds_handle() (4657)
3940 - create a new thread to do import eviction to avoid deadlock (3969)
3941 - let lconf resolve symlinked-to devices (4629)
3942 - don't unlink "objects" from directory with default EA (4554)
3943 - hold socknal file ref over connect in case target is down (4394)
3944 - allow more than 32000 subdirectories in a single directory (3244)
3945 - fix blocks count for O_DIRECT writes (3751)
3946 - OST returns ENOSPC from object create when no space left (4539)
3947 - don't send truncate RPC if file size isn't changing (4410)
3948 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3949 - bind to privileged port in socknal and tcpnal (3689)
3951 - rate limit CERROR/CWARN console message to avoid overload (4519)
3952 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3953 - basic mmap support (3918)
3954 - kernel patch series update from b1_4 (4711)
3956 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3959 - avoid crash during MDS cleanup with OST shut down (2775)
3960 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3961 - don't use bad inodes on the MDS (3744)
3962 - dynamic object preallocation to improve recovery speed (4236)
3963 - don't hold spinlock over lock dumping or change debug flags (4401)
3964 - don't zero obd_dev when it is force cleaned (3651)
3965 - print grants to console if they go negative (4431)
3966 - "lctl deactivate" will stop automatic recovery attempts (3406)
3967 - look for existing locks in ldlm_handle_enqueue() (3764)
3968 - don't resolve lock handle twice in recovery avoiding race (4401)
3969 - revalidate should check working dir is a directory (4134)
3971 - don't always mark "slow" obdfilter messages as errors (4418)
3973 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3976 - don't close LustreDB during write_conf until it is done (3860)
3977 - fix typo in lconf for_each_profile (3821)
3978 - allow dumping logs from multiple threads at one time (3820)
3979 - don't allow multiple threads in OSC recovery (3812)
3980 - fix debug_size parameters (3864)
3981 - fix mds_postrecov to initialize import for llog ctxt (3121)
3982 - replace config semaphore with spinlock (3306)
3983 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3984 - don't allow enqueue to complete on a destroyed export (3822)
3985 - down write_lock before checking llog header bitmap (3825)
3986 - recover from lock replay timeout (3764)
3987 - up llog sem before sending rpc (3652)
3988 - reduce ns lock hold times when setting kms (3267)
3989 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3990 - fix NULL deref and obd_dev leak on setup error (3312)
3991 - replace some LBUG about llog ops with error handling (3841)
3992 - don't match INVALID dentries from d_lookup and spin (3784)
3993 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3994 - fix invalid assertion in ptlrpc_set_wait (3880)
3996 - add libwrap support for the TCP acceptor (3996)
3997 - add /proc/sys/portals/routes for non-root route listing (3994)
3998 - allow setting MDS UUID in .xml (2580)
3999 - print the stack of a process that LBUGs (4228)
4001 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4004 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4005 - make sure to unset replay flag from failed open requests (3440)
4006 - if default stripe count is 0, use OST count for inode size (3636)
4007 - update parent mtime/ctime on client for create/unlink (2611)
4008 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4009 - free recovery state on server during a forced cleanup (3571)
4010 - unregister_reply for resent reqs (3063)
4011 - loop back devices mounting and status check on 2.6 (3563)
4012 - fix resource-creation race that can provoke i_size == 0 (3513)
4013 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4014 - more debugging for page-accounting assertion (3746)
4015 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4016 - avoid deadlock after precreation failure (3758)
4017 - fix race and lock order deadlock in orphan handling (3450, 3750)
4018 - add validity checks when grabbing inodes from l_ast_data (3599)
4020 - add /proc/.../recovery_status to obdfilter (3428)
4021 - lightweight CDEBUG infrastructure, debug daemon (3668)
4022 - change default OSC RPC parameters to be better on small clusters
4023 - turn off OST read cache for files smaller than 32MB
4024 - install man pages and include them in rpms (3100)
4025 - add new init script for (un)mounting lustre filesystems (2593)
4026 - run chkconfig in %post for init scripts (3701)
4027 - drop scimac NAL (unmaintained)
4029 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4032 - clean kiobufs before and after use (3485)
4033 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4034 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4035 - change init script to fail more clearly if not run as root (1528)
4036 - allow clients to reconnect during replay (1742)
4037 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4038 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4039 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4040 - fix cond_resched (3554)
4041 - only evict unfinished clients after recovery (3515)
4042 - allow bulk resend, prevent data loss (3570)
4043 - dynamic ptlrpc request buffer allocation (2102)
4044 - don't allow unlinking open directory if it isn't empty (2904)
4045 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4046 - remove extraneous obd dereference causing LASSERT failure (3334)
4047 - don't use get_cycles() when creating temp. files on the mds (3156)
4048 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4049 - handle EEXIST for set-stripe, set proper directory name (3336)
4051 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4052 - fix ksocknal_fmb_callback() error messages (2918)
4054 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4057 - don't copy lvb into (possibly NULL) reply on error (2983)
4058 - don't deref dentry after dput, don't free lvb on error (2922)
4059 - use the kms to determine writeback rpc length (2947)
4060 - increment oti_logcookies when osc is inactive (2948)
4061 - update client's i_blocks count via lvb messages (2543)
4062 - handle intent open/close of special files properly (1557)
4063 - mount MDS with errors=remount-ro, like obdfilter (2009)
4064 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4065 - don't use cancelling-locks' kms values (2947)
4066 - use highest lock extent for kms, not last one (2925)
4067 - don't dereference ERR_PTR() dentry in error handling path (3107)
4068 - fix thread race in portals_debug_dumplog() (3122)
4069 - create lprocfs device entries at setup instead of at attach (1519)
4070 - common AST error handler, don't evict client on completion race (3145)
4071 - zero nameidata in detach_mnt in 2.6 (3118)
4072 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4073 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4074 - initialize RPC timeout timer earlier for 2.6 (3219)
4075 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4076 - print nal/nid for unknown nid (3258)
4077 - additional checks for oscc recovery before doing precreate (3284)
4078 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4079 - don't crash in mdc_close for bad permissions on open (3285)
4080 - zero i_rdev for non-device files (3147)
4081 - clear page->private before handing to FS, better assertion (3119)
4082 - tune the read pipeline (3236)
4083 - fix incorrect decref of invalidated dentry (2350)
4084 - provide read-ahead stats and refine rpc in flight stats (3328)
4085 - don't hold journal transaction open across create RPC (3313)
4086 - update atime on MDS at close time (3265)
4087 - close LDAP connection when recovering to avoid server load (3315)
4088 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4089 - don't leak open file on MDS after open resend (3325)
4090 - serialize filter_precreate and filter_destroy_precreated (3329)
4091 - loop device shouldn't call sync_dev() for nul device (3092)
4092 - clear page cache after eviction (2766)
4093 - resynchronize MDS->OST in background (2824)
4094 - refuse to mount the same filesystem twice on same mountpoint (3394)
4095 - allow llmount to create routes for mounting behind routers (3320)
4096 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4097 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4098 - fix rare problem with rename on htree directories (3417)
4100 - allow default OST striping configuration per directory (1414)
4101 - fix compilation for qswnal for 2.6 kernels (3125)
4102 - increase maximum number of MDS request buffers for large systems
4103 - change liblustreapi to be useful for external progs like lfsck (3098)
4104 - increase local configuration timeout for slow disks (3353)
4105 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4107 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4110 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4111 - don't overwrite extent policy data in reply if lock was blocked (2901)
4112 - drop filter export grants atomically with removal from device (2663)
4113 - del obd_self_export from work_list in class_disconnect_exports (2908)
4114 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4115 - swab reply message in mdc_close, other PPC fixes (2464)
4116 - fix destroying of named logs (2325)
4117 - overwrite old logs when running lconf --write_conf (2264)
4118 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4119 - fix race in target_handle_connect (2898)
4120 - mds_reint_create() should take same inode create lock (2926)
4121 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4122 - don't close files for self_export to avoid uninitialized obd (2936)
4123 - allow MDS with the same name as client node (2939)
4124 - hold dentry reference for closed log files for unlink (2325)
4125 - reserve space for all logs during transactions (2059)
4126 - don't evict page beyond end of stripe extent (2925)
4127 - don't oops on a deleted current working directory (2399)
4128 - handle hard links to targets without a parent properly (2517)
4129 - don't dereference NULL lock when racing during eviction (2867)
4130 - don't grow lock extents when lots of conflicting locks (2919)
4132 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4135 - account for cache space usage on clients to avoid data loss (974)
4136 - lfsck support in lustre kernel code (2349)
4137 - reduce journal credits needed for BRW writes (2370)
4138 - orphan handling to avoid losing space on client/server crashes
4139 - ptlrpcd can be blocked, stopping ALL progress (2477)
4140 - use lock value blocks to assist in proper KMS, faster stat (1021)
4141 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4142 - recovery for initial connections (2355)
4143 - fixes for mds_cleanup_orphans (1934)
4144 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4145 - block all file creations until orphan recovery completes (1901)
4146 - client remove rq_connection from request struct (2423)
4147 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4148 - recovery timer race (2670)
4149 - mdc_close recovey bug (2532)
4150 - ptlrpc cleanup bug (2710)
4151 - mds timeout on local locks (2588)
4152 - namespace lock held during RPCs (2431)
4153 - handle interrupted sync write properly (2503)
4154 - don't try to handle a message that hasn't been replied to (2699)
4155 - client assert failure during cleanup after abort recovery (2701)
4156 - leak mdc device after failed mount (2712)
4157 - ptlrpc_check_set allows timedout requests to complete (2714)
4158 - wait for inflight reqs when ptlrpcd finishes (2710)
4159 - make sure unregistered services are removed from the srv_list
4160 - reset bulk XID's when resending them (caught by 1138 test)
4161 - unregister_bulk after timeout
4162 - fix lconf error (2694)
4163 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4164 - readahead locks pages, leaves pending causing memory pressure (2673)
4165 - increase OST request buffers to 4096 on large machines (2729)
4166 - fix up permission of existing directories in simple_mkdir (2661)
4167 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4168 - don't assign transno to errored transactions (2742)
4169 - don't delete objects on OST if given a bogus objid from MDS (2751)
4170 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4171 - if rq_no_resend, then timeout request after recovery (2432)
4172 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4173 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4174 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4175 - MDS recovery shouldn't delete objects that it has given out (2730)
4176 - if enqueue arrives after completion, don't clobber LVB (2819)
4177 - don't unlock pages twice when trigger_group_io returns error (2814)
4178 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4179 - don't write pages to disk if there was an error (1450)
4180 - don't ping imports that have recovery disabled (2676)
4181 - take buffered bytes into account when balancing socknal conn (2817)
4182 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4183 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4184 - remove little-endian swabbing of llog records (1987)
4185 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4186 - retry reposting request buffers if they fail (1191)
4187 - grow extent at grant time to avoid granting a revoked lock (2809)
4188 - lock revoke doesn't evict page if covered by a second lock (2765)
4189 - disable VM readahead to avoid reading outside lock extents (2805)
4191 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4192 - updated kernel patches for hp-2.4.20 kernel (2681)
4194 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4197 - fix truncated write corruption (2366)
4198 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4200 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4201 - ensure len(uuid) < 37 in lmc (1171)
4202 - fix ia64 OOPS in llog_test (2255)
4203 - zero end of page at obdfilter for partial page writes (2648)
4204 - don't leave stale dentries around after renames (bug 2428)
4205 - fix timeouts when evicting a client with a single lock held (2642)
4206 - set deadline for the initial HELLO message to drain (2634)
4207 - print out dotted-quad IP addresses in the socknal (2302)
4209 - additional debugging for MDS client eviction problem (2443)
4210 - fix mkfsoptions support for osts (2603, 2604)
4212 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4215 - add series for the vanilla 2.6.0 kernel
4216 - add series for the vanilla 2.4.24 kernel
4217 - add series for a cray x86/64 UL kernel drop
4218 - fix xattr patches for the vanilla 2.4.19 series
4220 - generate true UUIDs in lmc (1171)
4221 - have portals stack dumping break in UML (2466)
4222 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4223 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4224 - wake callback waiting threads on client eviction (2460)
4225 - Add --ptldebug and --subsystem to lmc (1719)
4226 - update assertion to allow safe interrupt allocation
4227 - set rq_no_resend for cancel requests (2432)
4228 - recalculate ptlrpcd timeout after resend (2494)
4229 - call vfs_rmdir when removing pending directories (2368)
4230 - fix renaming a file to itself (2429)
4231 - lmc creates a default one-stripe lov (2454)
4232 - expand procfs space to handle large clusters (2326)
4233 - increase UML stack to avoid overflow
4234 - update lconf's list of debug and subsystem masks
4235 - fix lfs find --obd (2510)
4236 - /proc tunable for disabling filter read caching (2591)
4237 - stop rpm packages from altering slapd.conf (2301)
4238 - disable nagle in the socknal under 0conf (2578)
4239 - choose mds inode size based on stripe count (2572)
4240 - fix kernel-source rpm problems (2516)
4242 - add --disable-doc to avoid pdf generation (2421)
4243 - update documentation, tests, type-os, comments
4244 - avoid format warnings on ia64
4245 - remove the TOE NAL
4246 - tiny code cleanups by removing unused fields
4248 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4251 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4252 - remove the most blatant lies from BUILDING file (2371)
4253 - change default debug level to reasonable production setting
4254 - reduce client side cache size to reduce cache flush time
4255 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4256 - make TCP zerocopy and pinger support enabled by default (2476)
4257 - sync writes completed after process exits caused crashes (2319)
4258 - maintain correct mount count on the MDS (2356)
4259 - backout 1557, because 2316 wasn't really fixed
4260 - better file I/O statistics gathering in /proc
4261 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4262 - another kernel patch to fix zero-copy TCP function export
4263 - don't take duplicate lock when processing re-sent getattr (2420)
4264 - lctl uses obd_self_export instead of creating new conn (2353)
4265 - MDS/OST recovery case which requires object creation asserted (2425)
4266 - move lfs from /usr/sbin to /usr/bin in packages
4267 - fix race between mds_client_add and mds_client_free (2417)
4268 - use kmalloc instead of slabs in portals (2430)
4269 - don't create duplicate records when a failover MDS is present (2442)
4270 - remove unnecessary mount age check (2332)
4271 - don't remove directory inodes from locks prematurely (2451)
4272 - don't break if MDS service name is the same as hostname (2103)
4273 - fix races in client write RPC generation when cache full (2482)
4275 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4278 - remove now-unused request->rq_obd (278)
4279 - if an allocation fails, print out how much memory we've used (1933)
4280 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4281 - add the "configurable stack size" patch to most series files (1256)
4282 - ability to write large log records, for 100+ OST configs (2306)
4283 - fix NULL deref when filter_prep fails (2314)
4284 - fix operator precedence error in filter_sync
4285 - dynamic allocation of socknal TX descriptors (2315)
4286 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4287 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4288 - ensure that CWARN messages in Portals always get to the syslog
4289 - __init/__exit are not for prototype decls (ldlm_init/exit)
4290 - x86-64 compile warning fixes
4291 - fix gateway LMC keyword conflict (2318)
4292 - fix MDS lock inversions in getattr/reint paths (1844)
4293 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4294 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4295 - DLM race condition prevented some lock evictions (2328)
4296 - ENOMEM detection and retry on socknal sends (2230)
4297 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4298 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4299 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4300 - better error messages when a client is rejected during recovery (1505)
4301 - avoid cancelling locks which were never granted, after failure (2330)
4302 - fix i_sem/journal inversion in mds_client_add (2333)
4303 - fix truncate/getattr lock cycle deadlock (2334)
4304 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4305 - fix two rare exit paths which could leak an l_lock() ref (2321)
4306 - fencepost error in MDS/OST orphan recovery (2226)
4307 - make log record alignment 8 bytes (1988)
4308 - lstripe now fails when requested offset > ost_count (2237)
4309 - ensure that all kernel series have a complete list.h (1607)
4310 - fix crashes in special-file operations (2316)
4311 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4313 - allow configurable automake binary, for testing new versions
4314 - small update to the lfs documentation
4316 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4318 * fix negative export reference count in fsfilt_sync (2312)
4320 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4321 * release candidate 0.9.1
4323 - orphans are moved into the PENDING directory for possible recovery
4324 - replayed opens now open by fid for orphan/rename safety (1042)
4325 - last close of an orphan inode generates a transno (683)
4326 - chdir() and mount() now pin the directory entry (1020)
4327 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4328 - discard very old requests without processing them (1502)
4329 - remove some common, well-understood CERRORs (1505)
4330 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4331 - clear "grant" flags in OST replies until OST grant code lands (1644)
4332 - fix read performance by not clobbering i_blksize on client (1598)
4333 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4334 - make LRU size a /proc tunable, clears locks when reduced (707)
4335 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4336 - server locks take references on exports now (1558)
4337 - build fixes for 2.4.20-rh trees (1663)
4338 - return an error from lov_create if all OSCs are inactive (1751)
4339 - fix import levels when a reconnect happens without a timeout (1597)
4340 - exit early from mds_open if we get a lookup error (1749)
4341 - partial page read at EOF wouldn't wait for disk before sending (1642)
4342 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4343 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4344 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4345 - write/truncate lock inversion (1639)
4346 - Don't auto-load obdclass, portals modules during cleanup (1495)
4347 - fix timestamps from jumping to "now" (1763)
4348 - extra journal assertions (1648)
4349 - add an extra multiunlink test (1771)
4350 - fix read_record/write_record API (1776)
4351 - fix leak of offset_extent, possible incorrect i_size later (1772)
4352 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4353 - Debugging for the kqswnal_get_idle_tx problems (1820)
4354 - Allow recovery to be attempted multiple times (1536)
4355 - Write out MDS last_rcvd file after it is first created (1600)
4356 - Fix tx_descriptor leak in failed transmit situations (1827)
4357 - ext3 journaling fixes for assertion failure after IO error (1871)
4358 - class_export_put() on freed export after completion AST error (1896)
4359 - Fix revalidate looping in VFS (1322)
4360 - Don't access a freed export during MDS_REINT timeout (1521)
4361 - Add open-unlink recovery support on the MDS (1673,1764)
4362 - Return an error if no MDS data was read from last_rcvd (1946)
4363 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4364 - Don't setup a disk device that doesn't match exported UUID (317)
4365 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4366 - avoid committing NULL handle in force close
4367 - local.sh is now a one-stripe LOV configuration
4368 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4369 - fix ext3 htree duplicate directory entry corruption (1516)
4370 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4371 - update to new LOV EA format (2097)
4372 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4374 2003-06-15 Phil Schwan <phil@clusterfs.com>
4377 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4378 - per-import/export recovery handling (958, 931, 959)
4379 - multiple last-rcvd slots, for serving multiple FSes (949)
4380 - connections are again shared between multiple imp/exports (963, 964)
4381 - "umount -f" would hang if any requests needed to be sent (393, 978)
4382 - avoid pinning large req buffer by copying for queued messages (989)
4383 - add "uuid" to "lctl device" command to help upcalls (991)
4384 - "open" RPCs with transnos would confuse recovery counters (1037)
4385 - do proper endian conversion of all wire messages (288, 340, 891)
4386 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4387 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4388 - fix ptlrpc_connection leak in target_handle_connect (1174)
4389 - fix import refcounting bug in OST and MDS cleanup (1134)
4390 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4391 - fix very unlikely obd_types race condition (501)
4392 - remove osc_open hack for echo_client (1187)
4393 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4394 - a failure in read_inode2 leads to deadlock (1139)
4395 - cancel ack-locks as soon as transaction is committed (1072)
4396 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4397 - make sure to commitrw after any preprw to avoid deadlock (1162)
4398 - failing to execute a file in a lustre FS would lock inode (1203)
4399 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4400 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4401 - obd_uuid2tgt would walk off the end of the list (1255)
4402 - on IA64 the osc would give portals incorrect bulk size (1258)
4403 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4404 - fix lock inversion caused by new llite matching code (1282)
4405 - limit the number of dirty pages on a client to 10MB (1286)
4406 - timed out locks were not being corrected cancelled (1289)
4407 - fix O_DIRECT above 4GB on IA-32 (1292)
4408 * major user-visible changes
4409 - fail out/fail over policy now controlled by the upcall (993)
4411 - add OBD_PING to check server availability and failure (954)
4412 - lustre messages are now sent in sending host order (288, 340, 891)
4413 - add eadatalen to MDS getattr reply (340)
4414 - OST read replies may contain second buffer, with per-page status (593)
4416 2003-03-11 Phil Schwan <phil@clusterfs.com>
4419 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4420 - failed open()s could cause deadlock; fixed (867, 869)
4421 - stop cancelling OST locks when files are closed (481)
4422 - overlapping XID spaces caused network corruption (851, 853)
4423 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4424 - setattr_raw added, to do single-RPC, server-side setattrs
4425 - lmc/lconf syntax change for OST UUIDs
4426 - fix crashy race condition between ptlrpc_free_req and osc_close
4427 - don't use request in mdc_enqueue if we hit a timeout (889)
4428 - don't set the inode i_size for regular files from the MDS (896)
4429 - handle out of order completion AST (842)
4430 - don't LBUG if a lock request times out after receiving AST (913)
4431 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4432 - if a bad lock AST arrives, send an error instead of dropping entirely
4433 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4434 - fix leak in bulk IO when only partially completed (899, 900, 926)
4435 - fix O_DIRECT for ia64 (55)
4436 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4437 - C-z after timeout could hang a process forever; fixed (977)
4439 - client-side I/O cache (678, 924, 929, 941, 970)
4441 - READPAGE and SETATTRs which don't take server-side locks get
4444 2003-02-11 Phil Schwan <phil@clusterfs.com>
4447 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4448 - Fix fsfilt_extN_readpage() to read a full page of directory
4449 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4450 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4451 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4452 - Most intent operations were removed, in favour of a new RPC mode
4453 that does a single RPC to the server and bypasses most of the VFS
4454 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4455 - Aggressively cancel local locks on DLM servers
4456 - mds_reint_unlink sends EA to the client if it's the last nlink.
4457 client uses that EA to unlink OST objects.
4458 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4459 - recursive symlinks were fixed (439)
4460 - fixed NULL deref in DEBUG_REQ
4461 - filter_update_lastobjid no longer calls sync, which annoyed extN
4462 - fixed multi-client small-writes to a single file problem (445)
4463 - fixed mtime updates during file writes (607)
4464 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4465 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4466 - fixed Linux OST size reporting problem (444, 656)
4467 - OST now updates object mtime with writes or setattr (607, 619)
4468 - client verifies file size before zeroing page past EOF (445)
4469 - OST now writes last allocated objid to disk with allocation (108)
4470 - LOV on echo now works (409)
4472 - mds_reint_unlink sends a new buffer, with the EA included. this
4473 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4474 set if a regular file was being unlinked, and it was the last link
4475 - use PtlGet from the target for bulk writes (315)
4476 - OST now updates object mtime with writes or setattr (607, 619)
4477 - LDLM now has a grant-time callback to revalidate locked items, if
4479 - Many MDS operations were reorganized to combat race conditions
4481 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4482 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4484 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4487 - Fully reactivate OST imports after reconnection (512, others)
4488 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4489 - More graceful error handling for truncating on dead OST (515)
4490 - Don't error out unless we're actually accessing dead stripes (474)
4491 - Fix garbage sizes when stripes are missing (410)
4492 - LRU counters were broken, causing constant lock purge (433, 432)
4493 - garbage on read from stripes with failed OSTs (441)
4494 - mark OSCs as active before reconnecting during recovery (438)
4495 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4496 - lfind did not preserve OST order in output (443)
4497 - symlinks cause hung clients, incorrect data (439)
4498 - stop dereferencing request after dropping refcount (457)
4499 - don't LASSERT(spin_is_locked) on non-SMP (455)
4500 - fixes for many rename() bugs
4501 - fstat didn't correctly synchronize attributes (399)
4502 - server must handle lock cancellation during blocking AST prep (487)
4503 - bulk descriptors were free()d too soon (511)
4504 - fix paths in lconf, which would load incorrect modules (451, 507)
4505 - fix confusing lconf 'host not found' error message (386)
4506 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4507 - fix race condition in mdc_blocking_ast() for inode access (526)
4508 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4509 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4510 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4511 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4512 - fix lstripe in several ways (kernel side) (527)
4513 - fix request leak in ldlm_cli_enqueue (262)
4514 - incorrect OSC was marked inactive after OST failure
4515 - call mds_fs_cleanup before unmounting filesystem (524)
4516 - fix races between taking ns_lock and ldlm_lock_change_resource
4517 - fix races updating LOV export open file list
4518 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4519 - fix recovery NULL deref in ldlm_cli_cancel_unused
4520 - fix some DLM races by using new hash table for lock handles (419)
4521 - permit the client to specify desired inodes, at replay
4522 - duplicate requests when we queue them for replay reintegration
4523 - fix last_rcvd offset calculation
4524 - sync after each recovered transaction, so we always make progress
4525 - never, not always, ERESTART requests without transnos
4526 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4528 - skip replay if the MDS says that the client is already connected
4529 - don't check for a recovery-enabled export to match lctl's UUID
4530 - don't INC_USE_COUNT for phantom exports
4531 - don't crash when cleaning up phantom exports (567)
4532 - don't double-finish or set replay data for errored mdc_open requests
4533 - abort requests when they time out, so we don't get old replies
4534 - send/receive replies for AST messages again
4535 - if the client says that it doesn't have the lock, cancel it on the
4537 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4538 mark it as destroyed, it will all work out when decref is called
4539 - fix module use counts (22, 581)
4541 - ASTs now expect a reply (server cancels lock on error reply)
4543 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4546 - fix many simultaneous client startup (392)
4547 - fix dentry->d_it clobbering
4548 - credentials weren't being shipped for readdir/getattr operations
4549 - remove invalid assertions triggered during some concurrent MD
4551 - proper Lustre versions added (336, 389)
4552 - fix memory leak for create error case (398)
4553 - fix LOV locking bug that would get cli/srv out of sync
4554 - fix echo client over LOV (409)
4555 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4556 - fix double-O_EXCL intent crash (424)
4557 - avoid sending multiple lock CANCELs (352)
4559 - MDS can do multi-client recovery (modulo bugs in new code)
4561 - many updates, edits, cleanups
4563 2002-11-18 Phil Schwan <phil@clusterfs.com>
4566 - fix null d_it dereference (346)
4567 - fix full OST/dbench hang (333)
4568 - fix permission problem with file removal (286)
4569 - fix removal of OSCs from LOV when they fail
4570 - fix NULL deref during bulk timeout (214)
4571 - fix problems related to multiple filesystems on one MDS (241)
4572 - fixed serious subtle metadata locking bugs
4573 - free locks on clients when inodes are removed due to memory
4575 - fix inode pointer in lock data (285)
4576 - partial support for multiple MDS on a single host (241)
4577 - data locks weren't cancelled at clear_inode time (290, 311)
4578 - intent locks could lead to unbounded lock growth (205)
4579 - added a maximum lock count, an LRU list, and a flusher
4580 - fix multiple rename (365)
4581 - properly abstracted the echo client
4582 - OSC locked 1 byte too many; fixed
4583 - rewrote brw callback code:
4584 - fixed recovery bugs related to LOVs (306)
4585 - fixed too-many-pages-in-one-write crash (191)
4586 - fixed (again) crash in sync_io_timeout (214)
4587 - probably fixed callback-related race (385)
4589 - Add capability to MDS protocol
4590 - LDLM cancellations and callbacks on different portals
4592 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4595 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4596 - defer open object destruction to close time (601981, 138)
4597 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4598 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4599 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4600 - don't call obd_setattr in ll_file_release for destroyed objects
4601 * protocol change to lustre_msg: move |version| and add |flags|
4602 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4603 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4604 * added replay of create, unlink, link and rename operations during
4605 MDS failover; recovery should be much more robust now
4606 * remove failed OSCs from LOVs (only lov_create uses this so far)
4607 * the lustre-HOWTO was brought (more) up to date (582544)
4609 2002-10-23 Phil Schwan <phil@clusterfs.com>
4612 - in-use dentries weren't being reused properly (617851)
4613 - prevent multiple LDLM setup (599178)
4614 - fix LOV size calculations for truncate (617853)
4615 - fix client handling of MDS intent errors (POSIX)
4616 - fix permission bug in lovstripe.c test (624321)
4617 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4618 - truncate past end of file could corrupt data
4619 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4620 - a race in recovery could return ETIMEDOUT to apps (623947)
4621 - building outside the source directory was fixed
4622 * the lustre-HOWTO was brought (more) up to date (582544)
4623 * major progress was made on recovery functionality
4625 2002-10-10 Phil Schwan <phil@clusterfs.com>
4628 - recovery deadlock fix
4629 - rm -rf causes LBUG fix (617817)
4630 - file open by multiple tasks fix (618962)
4631 - directory permissions bugs (602707 and 620007)
4632 - journal_stop fixed with locking (611313)
4633 - O_APPEND failures resolved (618273, perhaps 614459)
4634 - lconf PATH fix (619770)
4635 - IA64 build fix (621450)
4636 - RPC buffer sizes scale with amount of memory
4638 2002-10-01 Phil Schwan <phil@clusterfs.com>
4641 - locks would be cancelled without throwing away data pages,
4642 resulting in inconsistent data (605627)
4643 - inode attributes were not always being refreshed (605627, 612449)
4644 - lconf now continues to cleanup after lctl reports an error
4645 - MDS now enforces user permissions (602707)
4646 - lprocfs cleanup fixed, but not yet enabled (614157)
4647 - fixed infinite server hang, should a client not respond to an AST
4648 - avoid going into recovery if user calls readlink() with a buffer
4649 that's too small (613941)
4650 - AST RPCs no longer require replies (614867) -- this may be changed
4651 - don't crash server if client sends an IOV that's too big (611336)
4652 - fixed lock conversion deadlock (611892)
4653 - fixed the following of symlinks (614622)
4654 * recovery: the server can remove locks from a client that dies, other
4655 clients can make progress
4656 * more extN patch fixes
4657 * compile-time configurable ptlrpc buffer allocations
4659 - collaborative read cache document
4660 - Lustre Lite Performance CDR document-in-progress
4662 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4665 - fix typo in patch-2.4.18
4667 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4670 - clear ptlrpc request each time in handle_incoming_request()
4671 - unlink of files now destroys the object on the OST
4673 2002-09-19 Peter Braam <braam@clusterfs.com>
4675 * add hard link support
4676 * change obdfile creation method
4677 * kernel patch changed
4679 2002-09-19 Peter Braam <braam@clusterfs.com>
4682 - stack overflow bug in extN fixed
4684 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4686 * documentation updates
4687 - add man pages for config tools
4688 - update tests/README to describe testing with new config tools
4689 - finish metadata API descriptions
4690 * bug fixes and cleanups
4691 - statfs workaround for 16TB limit
4692 - LOV stripe allocation improved, can stripe on subset of OSTs
4693 - LOV file size/IO offset was wrong for files > 4GB in size
4694 - object EA data was being dropped, caused files to be unreadable
4695 - memory overflow with non-LOV OST caused memory corruption
4696 - fixed regression tests to work with new config tools, obdfilter
4697 - fixed bug when directory size became larger than 1 block
4698 - fixed bug (for single client case) when PWD was deleted
4699 - invalidate local directory pages when doing intent-based ops
4700 - avoid LDLM oops when lock callback contained bad data
4702 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4704 * documentation updates
4705 * bug fixes and cleanups
4706 - configuration tools
4709 - 64-bit compile warnings
4710 - 64-bit internal statfs data
4712 * test_brw on persistent OST devices
4714 * lprocfs (disabled)
4716 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4718 * documentation updates
4719 * bug fixes and cleanups
4720 * configuration tools
4722 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4725 * many small fixes to 0_5_4
4726 * io/network handling
4727 * thinkos in MDS operations
4729 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4732 * crucial basic fixes to 0.5.3
4733 * IOR, Iozone work over Elan
4736 2002-08-07 Phil Schwan <phil@clusterfs.com>
4737 * version 0_5_3, our first alpha
4738 * we use the new Portals iovs
4739 * documentation updates
4740 * bug fixes and cleanups
4741 * small changes in the DLM wire protocol
4743 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4744 * version 0_5_1 with some initial stability,
4745 * locking on MD and file I/O.
4746 * documentation updates
4747 * several bug fixes since 0.5.0
4748 * small changes in wire protocol
4750 2002-07-18 Phil Schwan <phil@clusterfs.com>
4752 * delivered as Lustre Light Alpha
4753 * fixed a crash after handling invalid MDS requests
4754 * fixed directory pages for architectures with non-4k pages sizes
4756 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4757 * release version v0_4_4
4758 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4760 * Fixes a number of interruption problems with OST operations.
4761 * Update documentation for portals header changes
4762 * Move all wire protocol structs/defines to lustre_idl.h
4763 * Fixes symlink length bug.
4764 * Add tcpdump to repository.
4766 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4767 * release version v0_4_3
4768 * Fixes statfs for inodes on extN.
4769 * Fixes bug in runtests which would delete /etc/hosts.
4770 * Use 64-bit object IDs wherever possible (not into VFS though)
4771 Remove ost_get_info, which is unused by lustre, and out of date.
4773 2002-07-03 Peter Braam <braam@clusterfs.com>
4774 * release version v0_4_2 Fixes a lookup error (type not passed)
4775 * move forward to head of Portals
4776 * move forward to latest Lustre kernel
4778 2002-06-25 Peter Braam <braam@clusterfs.com>
4779 * release version v0_4_1. Hopefully stable on single node use.