1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.21.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-sun3
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: sanity test 65a fails if stripecount of -1 is set
18 Details : handle -1 striping on filesystem in ll_dirstripe_verify
23 Description: ASSERTION(CheckWriteback(page,cmd)) failed
24 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
27 Severity : enhancement
29 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
33 Description: do not process already freed flock
34 Details : flock can possibly be freed by another thread before it reaches
35 to ldlm_flock_completion_ast.
39 Description: LBUG during stress test
40 Details : Need properly lock accesses the flock deadlock detection list.
44 Description: oops in page fault handler
45 Details : kernel page fault handler can return two special 'pages' in error case, don't
46 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
50 Description: timeout with invalidate import.
51 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
52 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
53 ptlrpc_invalidate_import and as result LASSERT.
55 Severity : enhancement
57 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
61 Description: enable MGS and MDT services start separately
62 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
63 a co-located MGS without starting the MGS, which is a complement
64 to 'nosvc' mount option.
68 Frequency : after recovery
69 Description: precreate to many object's after del orphan.
70 Details : del orphan st in oscc last_id == next_id and this triger growing
71 count of precreated objects. Set flag LOW to skip increase count
72 of precreated objects.
76 Frequency : rare, on clear nid stats
77 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
78 Details : when clean nid stats sometimes try destroy live entry,
79 and this produce panic in free.
83 Description: Stack overflow during MDS log replay
84 ease stack pressure by using a thread dealing llog_process.
88 Description: wait until IO finished before start new when do lock cancel.
89 Details : VM protocol want old IO finished before start new, in this case
90 need wait until PG_writeback is cleared until check dirty flag and
91 call writepages in lock cancel callback.
93 Severity : enhancement
95 Description: using special macro for print time and cleanup in includes.
99 Description: mds_mfd_close() ASSERTION(rc == 0)
100 Details : In mds_mfd_close(), we need protect inode's writecount change
101 within its orphan write semaphore to prevent possible races.
105 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
109 Frequency : rare, on shutdown ost
110 Description: don't hit live lock with umount ost.
111 Details : shrink_dcache_parent can be in long loop with destroy dentries,
112 use shrink_dcache_sb instead.
116 Description: don't panic with use echo client
117 Details : echo client pass NULL as client nid pointer and this produce null
122 Description: fix build on ppc32
123 Details : compile code with -m64 flag produce wrong object file for ppc32.
127 Description: add message levels for liblustreapi
131 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
132 Details : Move the 'good_osts' check before the 'total_bavail' check. This
133 will result in an -EAGAIN and in the exit call path we call
134 alloc_rr() which will with increasing aggressiveness attempt to
135 aquire precreated objects on the minimum number of required OSCs.
139 Description: Use old size assignment to avoid deadlock
140 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
141 the scheduling while holding a spinlock. We do not need locking
142 for size in ll_update_inode() because size is only updated from
143 the MDS for directories or files without objects, so there is no
144 other place to do the update, and concurrent access to such inodes
145 are protected by the inode lock.
149 Description: resolve "_IOWR redefined" build error on SLES10
153 Description: dump the memory debugging after all modules are unloaded to
154 suppress false negative in conf_sanity test 39
156 Severity : enhancement
158 Description: build kernel-ib packages for OFED 1.3 in our release cycle
163 Description: fix SLES kernel versioning
164 Details : the kernel version for our SLES 10 kernel did not include a "-"
165 before the "smp" at the end. while this was not a problem in
166 general, it did mean that software trying to use the kernel
167 version to try to detect a vendor specific kernel would fail.
168 this was most evident by the OFED build scripts.
172 Description: Don't update lov_desc members until making sure they are valid
173 Details : When updating lov_desc members via proc fs, need fix their
174 validities before doing the real update.
178 Description: don't put request into delay list while invalidate in flight.
179 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
180 import in flight. this produce timeout for invalidate and sometimes
181 can cause stale data.
185 Frequency : on ppc only
186 Description: not convert ost objects for directory because it's not exist.
187 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
191 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
193 Severity : enhancement
195 Description: Update to support 2.6.22.14 vanilla kernel.
199 Frequency : rare, on recovery
200 Description: read procfs can produce deadlock in some situation
201 Details : Holding lprocfs lock which send rpc can produce block for destroy
202 obd objects and this also block reconnect with -EALREADY. This isn't
203 fix all lprocfs bugs - but make it rare.
205 Severity : enhancement
207 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
210 Frequency : frequent on X2 node
212 Description: mdc_set_open_replay_data LBUG
213 Details : Set replay data for requests that are eligible for replay.
217 Description: lustre_mgs: operation 101 on unconnected MGS
218 Details : When MGC is disconnected from MGS long enough, MGS will evict the
219 MGC, and late on MGC cannot successfully connect to MGS and a lot
220 of the error messages complaining that MGS is not connected.
224 Frequency : on network error
225 Description: panic with double free request if network error
226 Details : mdc_finish_enqueue is finish request if any network error ocuring,
227 but it's true only for synchronus enqueue, for async enqueue
228 (via ptlrpcd) this incorrect and ptlrpcd want finish request
231 Severity : enhancement
233 Description: client-side metadata stat-ahead during readdir(directory readahead)
234 Details : perform client-side metadata stat-ahead when the client detects
235 readdir and sequential stat of dir entries therein
238 Frequency : on start mds
240 Description: Implement get_info(last_id) in obdfilter.
243 Frequency : occasional
245 Description: Correctly check stale fid, not start epoch if ost not support SOM
246 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
247 overwrited with new generated fid, but mds can anwer with one of these
248 two fids and both is not stale. setattr incorectly start epoch and
249 assume will be called done_writeting, but without SOM done_writing
253 Frequency : rare, depends on device drivers and load
255 Description: MDS or OSS nodes crash due to stack overflow
256 Details : Code changes in 1.8.0 increased the stack usage of some functions.
257 In some cases, in conjunction with device drivers that use a lot
258 of stack the MDS (or possibly OSS) service threads could overflow
259 the stack. One change which was identified to consume additional
260 stack has been reworked to avoid the extra stack usage.
263 Frequency : occasional
265 Description: Do not fail import if osc_interpret_create gets -EAGAIN
266 Details : If osc_interpret_create got -EAGAIN it immediately exits and
267 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
268 oscc_has_objects and see OSC has no objests and call
269 oscc_internal_create to resend create request.
271 Severity : enhancement
273 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
275 Severity : enhancement
277 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
280 Frequency : very rare
282 Description: Fix replay if there is an un-replied request and open
283 Details : In some cases, older replay request will revert the
284 mcd->mcd_last_xid on MDS which is used to record the client's
287 Severity : enhancement
289 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
291 Severity : enhancement
293 Description: Add rhel5 support to HEAD.
295 Serverity : enhancement
297 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
302 Description: Don't allow skipping OSTs if index has been specified.
303 Details : Don't allow skipping OSTs if index has been specified, make locking
304 in internal create lots better.
308 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
309 Details : access to bitfield in structure is always rounded to long
310 and this produce problem with not atomic change any bit.
314 Description: Lustre make rpms failed.
315 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
316 builds Lustre rpms from tarball.
319 Frequency : rare on shutdown ost
321 Description: If llog cancel was not send before clean_exports phase, this can
322 produce deadlock in llog code.
323 Details : If llog thread has last reference to obd and call class_import_put
324 this produce deadlock because llog_cleanup_commit_master wait when
325 last llog_commit_thread exited, but this never success because was
326 called from llog_commit_thread.
330 Description: allow userland application know is lost one of stripes.
331 Details : fill lvb_blocks with error code on ost and return it to
332 application if error flag found.
336 Description: NULL lov_tgts causing MDS oops
337 Details : more safe checks for NULL lov_tgts for avoid oops.
339 Severity : enhancement
341 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
345 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
346 Details : If an OST is down the MDS will hang indefinitely in
347 obd_statfs_rqset() waiting for the statfs data. While for
348 MDS QOS usage of statfs, it should not stuck in waiting.
350 Severity : enhancement
352 Description: remote_acl support
353 Details : Support ACL-based permission check for remote user.
354 Support setfacl/getfacl for remote user with the utils
355 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
356 the system "{s,g}etfacl" utils.
358 Severity : enhancement
360 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
362 Severity : enhancement
364 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
368 Description: make mgs_setparam() handle fsname containing dash
369 Details : fsname containing a dash does not work with lctl conf_param
371 Severity : enhancement
373 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
375 Severity : enhancement
377 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
381 Frequency : rare, at shutdown
382 Description: access already free / zero obd_namespace.
383 Details : if client_disconnect_export was called without force flag set,
384 and exist connect request in flight, this can produce access to
385 NULL pointer (or already free pointer) when connect_interpret
386 store ocd flags in obd_namespace.
390 Frequency : only at startup
391 Description: not alloc memory with spinlock held.
392 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
393 if any spinlock held.
395 Severity : enhancement
397 Description: make lustre randomly fail allocating memory
398 Details : Make lustre randomly failed allocating memory for testing purpose.
400 Severity : enhancement
402 Description: lost problems with lov objid file
403 Details : Fixes some scability and access to not inited memory problems
404 in work with lov objdid file.
409 Description: lfs find does not continue on file error
410 Details : Continue other files processing when a file/dir is absent.
414 Description: Inconsistent usage of lustre_pack_reply()
415 Details : Standardize the usage of lustre_pack_reply() such that it
416 always generate a CERROR on failure.
419 Frequency : occasional
421 Description: MDS hang or stay in waiting lock
422 Details : If client receive lock with CBPENDING flag ldlm need send lock
423 cancel as separate rpc, to avoid situation when cancel request
424 can't processed due all i/o threads stay in wait lock.
428 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
432 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
436 Description: Update to SLES9 kernel-2.6.5-7.287.3.
440 Description: Update to RHEL4 latest kernel.
442 Severity : enhancement
444 Description: Build SLES10 patchless client fails
445 Details : The configure was broken by run ./configure with
446 --with-linux-obj=.... argument for patchless client. When the
447 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
448 can't search header adequately. Use absolute path such as
449 -I($LINUX)/include instead.
453 Description: interrupt oig_wait produce painc on resend.
454 Details : brw_redo_request can be used for resend requests from ptlrpcd and
455 private set, and this produce situation when rq_ptlrpcd_data not
456 copyed to new allocated request and triggered LBUG on assert
457 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
458 ptlrpcd set if request is changed and can be safety changed to use
463 Description: LASSERT_{REQ,REP}SWAB macros are buggy
464 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
465 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
466 these macros and replace them with their difinitions instead.
470 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
471 Details : Update which_patch & target file for SLES10 latest kernel.
474 Description: Remove client patches from SLES 10 kernel.
475 Details : This causes SLES 10 clients to behave as patchless clients
476 even on a Lustre-patched (server) kernel.
478 Severity : enhancement
480 Description: self-adjustable client's lru lists
481 Details : use adaptive algorithm for managing client cached locks lru
482 lists according to current server load, other client's work
483 pattern, memory activities, etc. Both, server and client
484 side namespaces provide number of proc tunables for controlling
487 Severity : enhancement
489 Description: light-weight GSS support
490 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
491 and reduce performance overhead.
493 Severity : enhancement
495 Description: Linux keyring support
496 Details : Support using service of Linux keyring for Lustre GSS internal
497 context refresh/cache mechanism.
501 Description: Fix errors in lfs documentation
502 Details : Fixes man pages
506 Description: don't use GFP_* in generic Lustre code.
507 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
511 Description: obdclass is limited by single OBD_ALLOC(idarray)
512 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
516 Description: warnings with build patchless client with vanila 2.6.19 and up
517 Details : change old ctl_table style and replace ctl_table/ctl_table_header
518 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
522 Frequency : Only for SLES
523 Description: sanity_quota fail test_1
524 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
525 file, which makes TSTID[2] inunique.
529 Frequency : Only for SLES9
530 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
531 Details : Update target/ChangeLog/which_patch .
535 Frequency : Only for test_10 in sanity-quota.sh
536 Description: a bug in quota test code
537 Details : checking a test flag in a wrong place
541 Frequency : Only for test_13 in sanity-quota.sh
542 Description: a bug in quota test script
543 Details : shouldn't check overfull allocated quota
547 Description: jbd statistics
548 Details : Port older jbd statistics patch for sles10
550 Severity : enhancement
552 Description: Add printing inode info into message about error in writepage.
557 Description: reply_lock_interpret crash due to race with it and lock cancel.
558 Details : Do not replay locks that are being cancelled. Do not reference
559 locks by their address during replay, just by their handle.
563 Frequency : When flocks are used.
564 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
565 Details : Flock locks might destroy just granted lock if it could be merged
566 with another existing flock, this is done in completion handler,
567 so teach ldlm_cli_enquque_fini that this is a valid case for
573 Description: Oops in read and write path when failing to allocate lock.
574 Details : Check if lock allocation failed and return error back.
578 Description: lstripe command fails for valid OST index
579 Details : The stripe offset is compared to lov->desc.ld_tgt_count
580 instead of lov->desc.ld_active_tgt_count.
584 Description: sanity.sh failed test 103
585 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
590 Description: df doesn't work properly if diskfs blocksize != 4K
591 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
595 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
596 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
600 Description: obd_health_check_timeout too short
601 Details : set obd_health_check_timeout as 1.5x of obd_timeout
605 Description: llapi_file_create() does not allow some changes
606 Details : add llapi_file_open() that allows specifying the mode and
607 open flags, and also returns an open file handle.
611 Description: merge and cleanup kernel patches.
612 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
616 Description: Add journal checksum support.(Kernel part)
617 Details : The journal checksum feature adds two new flags i.e
618 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
619 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
620 indicates that the commit block contains the checksum for
621 the blocks described by the descriptor blocks. Now commit
622 record can be sent to disk without waiting for descriptor
623 blocks to be written to disk. This behavior is controlled
624 using JBD2_FEATURE_ASYNC_COMMIT flag.
628 Description: OSS needs mutliple precreate threads
629 Details : Add ability to start more than one create thread per OSS.
633 Description: Sanity.sh test_65k failed: setstripe should have succeeded
634 Details : execute lfs setstripe on client
638 Description: mds_obd_create error creating tmp object
639 Details : When the user sets quota on root, llog will be affected and can't
640 create files and write files.
643 Frequency : Always on ia64 patchless client, and possibly others.
645 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
646 Details : This allows the patchless client to be loaded on architectures
651 Description: RedHat Update kernel for RHEL5
652 Details : Add the kernel config file for RHEL5.
656 Description: RedHat Update kernel for RHEL5
657 Details : Modify the kernel config file more closer RHEL5.
661 Description: Build failure against Centos5 (RHEL5)
662 Details : Define PAGE_SIZE when it isn't present.
666 Description: test_7 on sanity-quota.sh can't be used on separate machines
667 Details : add facet to handle it
671 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
672 Details : with NFS, the anon dentry's parent was set to itself in
673 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
674 corresponding dentry other than use rec->ur_name.
676 Severity : enhancement
678 Description: lfs setstripe enhancement
679 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
683 Description: enable data checksumming by default
684 Details : enable checksum by default, allow --disable-checksum
685 configure option and "-o nochecksum" mount option. Checksums
686 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
690 Description: lustre support for RHEL5
691 Details : Add support for RHEL5.
695 Description: Client eviction due to failover config
696 Details : after a connection loss, the lustre client should attempt to
697 reconnect to the last active server first before trying the
698 other potential connections.
702 Description: when mds and osts use different quota unit(32bit and 64bit),
703 quota will be released repeatly.
704 Details : void sending multiple quota reqs to mds, which will keep the
705 status between the reqs.
709 Description: osts not allocated evenly to files
710 Details : change the condition to increase offset_idx
713 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
715 Description: Data corruption for OSTs that are formatted larger than 2TB
717 Details : When generating the bio request for lustre file writes the
718 sector number would overflow a temporary variable before being
719 used for the IO. The data reads correctly from Lustre (which
720 will overflow in a similar manner) but other file data or
721 filesystem metadata may be corrupted in some cases.
725 Description: Tune the kernel for good SCSI performance.
726 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
727 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
732 Description: rewrite ext2-derived code in obdclass/uuid.c
733 Details : rewrite inherited code (uuid parsing code from ext2 utils)
734 from scratch preserving functionality.
738 Description: "lfs find -obd UUID" prints directories
739 Details : "lfs find -obd UUID" will return all directory names instead
740 of just file names. It is incorrect because the directories
741 do not reside on the OSTs.
745 Description: use i_size_read and i_size_write in 2.6 port
746 Details : replace inode->i_size access with i_size_read/write()
749 Frequency : when removing large files
751 Description: scheduling issue during removal of large Lustre files
752 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
753 It causes scheduling issues when removing large files (17TB in the
757 Frequency : only with liblustre clients on XT3
759 Description: evictions taking too long
760 Details : allow llrd to evict clients directly on OSTs
763 Frequency : only on ppc
765 Description: /proc/fs/lustre/devices broken on ppc
766 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
767 We should make sure the type of 'index' is loff_t and then cast
768 explicitly as needed below. Do not assign an explicitly cast
772 Frequency : only for rhel5
774 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
775 Details : Modify the target file & which_kernel.
777 Severity : enhancement
779 Description: omit set fsid for export NFS
780 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
781 as FS_REQUIRES_DEV for avoid problems with generate fsid.
785 Frequency : Always, for kernels after 2.6.16
786 Description: Fix warning idr_remove called for id=.. which is not allocated.
787 Details : Last kernels save old s_dev before kill super and not allow
788 to restore from callback - restore it before call kill_anon_super.
792 Description: buffer overruns could theoretically occur
793 Details : llapi_semantic_traverse() modifies the "path" argument by
794 appending values to the end of the origin string, and a
795 overrun may occur. Adding buffer overrun check in liblustreapi.
799 Description: Fix error on 'ls .' at the top of the Lustre mount.
800 Details : Don't revalidate dentry if it is a root dentry.
804 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
805 Details : Modify vm-tunables-rhel4.patch.
809 Description: Kernel config for 2.6.18-vanilla.
810 Details : Modify targets/2.6-vanilla.target.in.
811 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
812 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
813 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
814 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
818 Description: improve handling recoverable errors
819 Details : if request processig with error which can be recoverable on server
820 request should be resend, otherwise page released from cache and
825 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
826 Details : Modify target file & which_patch.
827 A flaw was found in the IA32 system call emulation provided
828 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
829 value could be stored in the %RAX register, which could trigger an
830 out-of-bounds system call table access. An untrusted local user
831 could exploit this flaw to run code in the kernel
832 (ie a root privilege escalation). (CVE-2007-4573).
836 Description: O_DIRECT bypasses client statistics.
837 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
838 read_ahead_stats were updated. copied the stats section from
839 osc_send_oap_rpc() into async_internal().
843 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
847 Description: change order of libsysio includes
848 Details : '#include sysio.h' should always come before '#include xtio.h'
852 Description: handle "serious error: objid * already exists" more gracefully
853 Details : If LAST_ID value on disk is smaller than the objects existing in
854 the O/0/d* directories, it indicates disk corruption and causes an
855 LBUG(). If the object is 0-length, then we should use the existing
856 object. This will help to avoid a full fsck in most cases.
858 Severity : enhancement
860 Description: adapt the lustre_config script to support the upgrade case
861 Details : Add "-u" option for lustre_config script to support upgrading 1.4
862 server targets to 1.6 in parallel.
866 Description: To avoid grant space > avaible space when the disk is almost
867 full. Without this patch you might see the error "grant XXXX >
868 available" or some LBUG about grant, when the disk is almost
870 Details : In filter_check_grant, for non_grant cache write, we should
871 check the left space by if (*left > ungranted + bytes), instead
872 of (*left > ungranted), because only we are sure the left space
873 is enough for another "bytes", then the ungrant space should be
874 increase. In client, we should update cl_avail_grant only there
875 is OBD_MD_FLGRANT in the reply.
879 Description: Update RHEL 4 kernel to fix local root privilege escalation.
880 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
881 described in CVE-2007-4573. This problem could allow untrusted
882 local users to gain root access.
885 Frequency : when using O_DIRECT and quotas
887 Description: Incorrect file ownership on O_DIRECT output files
888 Details : block usage reported by 'lfs quota' does not take into account
889 files that have been written with O_DIRECT.
894 Description: touch file failed when fs is not full
895 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
896 otherwise we can get ENOSP while fs is not full.
900 Description: parallel lock callbacks
901 Details : Instead of sending blocking and completion callbacks as separated
902 requests, adding them to a set and sending in parallel.
905 Frequency : only for Cray XT3
906 Bugzilla : 12829/13455
907 Description: Changing primary group doesn't change the group lustre assigns to
909 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
910 overridden with the primary group provided by the group upcall,
911 whereas the supplied fsgid can be trusted if it is in the list of
912 supplementary groups returned by the group upcall.
914 Severity : enhancement
916 Description: Allow masking D_WARNING, D_ERROR messages from console
917 Details : Console messages can now be disabled via lnet.printk.
921 Description: User code with malformed file open parameter crashes client node
922 Details : Before packing join_file req, all the related reference should be
923 checked carefully in case some malformed flags cause fake
924 join_file req on client.
928 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
929 Details : ldlm_completion_ast() assumes that a lock is granted when the req
930 mode is equal to the granted mode. However, it should also check
931 that LDLM_FL_CP_REQD is not set.
935 Description: Heavy nfs access might result in deadlocks
936 Details : After ELC code landed, it is now improper to enqueue any mds
937 locks under och_sem, because enqueue might want to decide to
938 cancel open locks for same inode we are holding och_sem for.
942 Description: Client eviction while running blogbench
943 Details : A lot of unlink operations with concurrent I/O can lead to a
944 deadlock causing evictions. To address the problem, the number of
945 oustanding OST_DESTROY requests is now throttled to
946 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
947 callbacks are processed in priority.
951 Description: enable ACLs on MDS by default
952 Details : ACLs must be enabled on MDS by default.
955 Frequency : PPC/PPC64 only
957 Description: conflicts between asm-ppc64/types.h and lustre_types.h
958 Details : fix duplicated definitions between asm-ppc64/types.h and
959 lustre_types.h on PPC.
962 Frequency : PPC/PPC64 only
964 Description: asm-ppc/segment.h does not exist
965 Details : fix compile issue on PPC.
969 Description: better handle error messages in extents code
972 Frequency : RHEL4 only
974 Description: mkfs is very slow on IA64/RHEL4
975 Details : A performance regression has been discovered in the MPT Fusion
976 driver between versions 3.02.73rh and 3.02.99.00rh. As a
977 consequence, we have downgraded the MPT Fusion driver in the RHEL4
978 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
980 Severity : enhancement
982 Description: SNMP support enhancement
983 Details : Adding total number of sampled request for an MDS node in snmp
986 Severity : enhancement
988 Description: Optimize ldlm waiting list processing for PR extent locks
989 Details : When processing waiting list for read extent lock and meeting read
990 lock that is same or wider to it that is not contended, skip
991 processing rest of the list and immediatelly return current
992 status of conflictness, since we are guaranteed there are no
993 conflicting locks in the rest of the list.
997 Description: Time out and refuse to reconnect
998 Details : When the failover node is the primary node, it is possible
999 to have two identical connections in imp_conn_list. We must
1000 compare not conn's pointers but NIDs, otherwise we can defeat
1001 connection throttling.
1005 Description: port llog fixes from b1_6 into HEAD
1006 Details : Port llog reference couting and some llog cleanups from b1_6
1007 (bug 10800) into HEAD, for protect from panic and access to already
1008 free llog structures.
1012 Description: Detect stride IO mode in read-ahead
1013 Details : When a client does stride read, read-ahead should detect that and
1014 read-ahead pages according to the detected stride pattern.
1018 Description: data checksumming impacts single node performance
1019 Details : add support for several checksum algorithm. Currently, only CRC32
1020 and Adler-32 are supported. The checksum type can be changed on
1021 the fly via /proc/fs/lustre/osc/*/checksum_type.
1025 Description: use adler32 for page checksums
1026 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1031 Description: build for x2 fails
1032 Details : fix compile issue on Cray systems.
1036 Description: Properly match for duplicate locks
1037 Details : Due to different lock order from skiplists code, we need to
1038 traverse entire list for now
1041 Frequency : only on PPC/SLES10
1043 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1044 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1045 be defined. Add a hack in mkfs_lustre.c to work around this compile
1050 Description: LASSERT on MDS when client holding flock lock dies
1051 Details : ldlm pool logic depends on number of granted locks equal to
1052 number of released locks which is not true for flock locks, so
1053 just exclude such locks from consideration.
1057 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1058 Details : Use fsfilt_sync() for both the whole filesystem sync and
1059 individual file sync to eliminate dangerous inode locking
1060 with I_LOCK that can lead to a deadlock.
1064 Description: performance in 1.6.3
1065 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1066 sure that 1MB requests are not fragmented by the block layer.
1068 Severity : enhancement
1070 Description: organize the server-side client stats on per-nid basis
1071 Details : Change the structure of stats under obdfilter and mds to
1079 The "uuid"s file would list the uuids of _active_ exports.
1080 And the clear entry is to clear all stats and stale nids.
1082 Severity : enhancement
1084 Description: eliminate client locks in face of contention
1085 Details : file contention detection and lockless i/o implementation
1086 for contended files.
1090 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1093 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1095 Description: MDS may lose file striping (and hence file data) in some cases
1096 Details : If there are additional extended attributes stored on the MDS,
1097 in particular ACLs, SELinux, or user attributes (if user_xattr
1098 is specified for the client mount options) then there is a risk
1099 of attribute loss. Additionally, the Lustre file striping
1100 needs to be larger than default (e.g. striped over all OSTs),
1101 and an additional attribute must be stored initially in the
1102 inode and then increase in size enough to be moved to the
1103 external attribute block (e.g. ACL growing in size) for file
1108 Description: skiplist implementation simplification
1109 Details : skiplists are used to group compatible locks on granted list
1110 that was implemented as tracking first and last lock of each lock group
1111 the patch changes that to using doubly linked lists
1115 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1116 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1117 because LASSERT on both the data supplied by a client, and the data
1118 on disk is dangerous and incorrect.
1120 Severity : enhancement
1122 Description: Slow truncate/writes to huge files at high offsets.
1123 Details : Directly associate cached pages to lock that protect those pages,
1124 this allows us to quickly find what pages to write and remove
1125 once lock callback is received.
1129 Description: more ldlm soft lockups
1130 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1131 starve other threads from the resource lock for a long time in
1132 case of long waiting queue, so change the debug level from
1133 D_OTHER to the less frequently used D_INFO.
1135 Severity : enhancement
1137 Description: add -gid, -group, -uid, -user options to lfs find
1141 Description: Hung threads in invalidate_inode_pages2_range
1142 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1143 one is completed. As a result, some RPCs are stuck in the queue
1148 Description: filter threads hungs on waiting journal commit
1149 Details : Cleanup filter group llog code, then only filter group llog will
1150 be only created in the MDS/OST syncing process.
1154 Description: Procfs and llog threads access destoryed import sometimes.
1155 Details : Sync the import destoryed process with procfs and llog threads by
1156 the import refcount and semaphore.
1158 --------------------------------------------------------------------------------
1160 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1162 * Support for kernels:
1163 2.6.5-7.283 (SLES 9),
1164 2.6.9-55.EL (RHEL 4),
1165 2.6.16.46-0.14 (SLES 10),
1166 2.6.18.8 vanilla (kernel.org)
1167 * Client support for unpatched kernels:
1168 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1169 2.6.16 - 2.6.22 vanilla (kernel.org)
1170 * Due to recently discovered recovery problems, we do not recommend
1171 using patchless RHEL 4 clients with this or any earlier release.
1172 * Recommended e2fsprogs version: 1.39.cfs8
1173 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1174 * Starting with this release, the ldiskfs backing filesystem required
1175 by Lustre is now in its own package, lustre-ldiskfs. This package
1176 should be installed. It is versioned separately from Lustre and
1177 may be released separately in future.
1181 Description: block reactivating mgc import until all deactivates complete
1182 Details : Fix race when failing back MDT/MGS to itself (testing)
1185 Frequency : at statup only
1187 Description: mds_lov_synchronize race leads to various problems
1188 Details : simultaneous MDT->OST connections at startup can cause the
1189 sync to abort, leaving the OSC in a bad state.
1191 Severity : enhancement
1193 Description: add optional extra BUILD_VERSION info
1194 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1195 to override the lustre version.
1198 Frequency : 2.6.18 servers only
1200 Description: ll_kern_mount() doesn't release the module reference
1201 Details : The ldiskfs module reference count never drops down to 0
1202 because ll_kern_mount() doesn't release the module reference.
1207 Description: server LBUG when using old ost_num_threads parameter
1208 Details : Accept the old ost_num_threads parameter but warn that it
1209 is deprecated, and fix an off-by-one error that caused an LBUG.
1214 Description: Transient SCSI error results in persistent IO issue
1215 Details : iobuf->dr_error is not reinitialized to 0 between two
1219 Frequency : sometimes when underlying device returns I/O errors
1221 Description: OSTs not going read-only during write failures
1222 Details : OSTs are not remounted read-only when the journal commit threads
1223 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1224 instead of the ext3 wrappers.
1228 Description: poor connect scaling with increasing client count
1229 Details : Don't run filter_grant_sanity_check for more than 100 exports
1230 to improve scaling for large numbers of clients.
1233 Frequency : SLES10 only
1235 Description: sanity-quota.sh quotacheck failed: rc = -22
1236 Details : Quotas cannot be enabled on SLES10.
1239 Frequency : liblustre clients only
1241 Description: getdirentries does not give error when run on compute nodes
1242 Details : getdirentries does not fail when the size specified as an argument
1243 is too small to contain at least one entry
1245 Severity : enhancement
1247 Description: Add LNET router traceability for debug purposes
1248 Details : If a checksum failure occurs with a router as part of the
1249 IO path, the NID of the last router that forwarded the bulk data
1250 is printed so it can be identified.
1255 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1256 Details : Due to a race condition, liblustre clients were occasionally
1257 evicted incorrectly.
1259 Severity : enhancement
1261 Description: lfs setstripe use optional parameters instead of postional
1264 Severity : enhancement
1266 Description: Nanosecond timestamp support for ldiskfs
1267 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1268 resolution timestamps. There is not yet support for this at
1269 the Lustre filesystem level.
1272 Frequency : during server recovery
1274 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1275 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1279 Frequency : nfs export on patchless client
1281 Description: connectathon hang when test nfs export over patchless client
1282 Details : Disconnected dentry cannot be found with lookup, so we do not need
1283 to unhash it or make it invalid
1286 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1287 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1288 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1293 Description: open req refcounting wrong on reconnect
1294 Details : If reconnect happened between getting open reply from server and
1295 call to mdc_set_replay_data in ll_file_open, we will schedule
1296 replay for unreferenced request that we are about to free.
1297 Subsequent close will crash in variety of ways.
1298 Check that request is still eligible for replay in
1299 mdc_set_replay_data().
1304 Description: disable writes to filesystem when reading health_check file
1305 Details : the default for reading the health_check proc file has changed
1306 to NOT do a journal transaction and write to disk, because this
1307 can cause reads of the /proc file to hang and block HA state
1308 checking on a healthy but otherwise heavily loaded system. It
1309 is possible to return to the previous behaviour during configure
1310 with --enable-health-write.
1312 Severity : enhancement
1314 Description: 64-bit inode version
1315 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1316 to the inode. This will be required for version-based recovery.
1321 Description: MDS fails to start if a duplicate client export is detected
1322 Details : in some rare cases it was possible for a client to connect to
1323 an MDS multiple times. Upon recovery the MDS would detect this
1324 and fail during startup. Handle this more gracefully.
1326 Severity : enhancement
1328 Description: Add -o localflock option to simulate old noflock
1330 Details : This will achieve local-only flock/fcntl locks
1336 Description: log_commit_thread vs filter_destroy race leads to crash
1337 Details : Take import reference before releasing llog record semaphore
1342 Description: Wrong request locking in request set processing
1343 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1344 imp_delayed_list, in this place should be used imp_lock.
1347 Frequency : when reconnection
1349 Description: Grant Leak when osc reconnect to OST
1350 Details : When osc reconnect ost, OST(filter) should check whether it
1351 should grant more space to client by comparing fed_grant and
1352 cl_avail_grant, and return the granted space to client instead
1353 of "new granted" space, because client will call osc_init_grant
1354 to update the client grant space info.
1357 Frequency : when client reconnect to OST
1359 Description: Grant Leak when osc do resend and replay bulk write
1360 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1361 bulk write request, because the grant info while be sync between
1362 OSC and OST when reconnect, and we should ignore the grant info
1363 these of resend/replay write req.
1368 Description: Grant space more than avaiable left space sometimes.
1369 Details : When then OST is about to be full, if two bulk writing from
1370 different clients came to OST. Accord the avaliable space of the
1371 OST, the first req should be permitted, and the second one
1372 should be denied by ENOSPC. But if the seconde arrived before
1373 the first one is commited. The OST might wrongly permit second
1374 writing, which will cause grant space > avaiable space.
1377 Frequency : when client is evicted
1379 Description: Grant might be wrongly erased when osc is evicted by OST
1380 Details : when the import is evicted by server, it will fork another
1381 thread ptlrpc_invalidate_import_thread to invalidate the
1382 import, where the grant will be set to 0. While the original
1383 thread will update the grant it got when connecting. So if
1384 the former happened latter, the grant will be wrongly errased
1385 because of this race.
1390 Description: Checking Stale with correct fid
1391 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1392 to check whether it is stale, because sometimes, we want the
1393 enqueue happened anyway, and op_data.fid2 will not be initialized.
1395 Severity : enhancement
1397 Description: update patchless client
1398 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1401 Frequency : only with 2.4 kernel
1403 Description: random memory corruption
1404 Details : size of struct ll_inode_info is to big for union inode.u and this
1405 can be cause of random memory corruption.
1410 Description: Memory leak in recovery
1411 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1412 It should also check obd_fail before fsfilt_start, otherwise if
1413 fsfilt_start return -EROFS,(failover mds during mds recovery).
1414 then the req will return with repmsg->transno = 0 and rc = EROFS.
1415 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1416 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1417 be freed no matter whether fsfilt_commit success or not.
1420 Frequency : only with huge count clients
1422 Description: Prevents from taking the superblock lock in llap_from_page for
1424 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1425 for llap_from_page call in ll_removepage prevents from taking the
1426 superblock lock for a soon died page.
1431 Description: Not check open intent error before release open handle
1432 Details : in some rare cases, the open intent error is not checked before
1433 release open handle, which may cause
1434 ASSERTION(open_req->rq_transno != 0), because it tries to release
1435 the failed open handle.
1440 Description: Set cat log bitmap only after create log success.
1441 Details : in some rare cases, the cat log bitmap is set too early. and it
1442 should be set only after create log success.
1446 Description: Accessing a block bevice can re-enable I/O when Lustre is
1447 tearing down a device.
1448 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1453 Description: service threads may hog cpus when there are a lot of requests
1454 Details : Insert cond_resched to give other threads a chance to use some CPU
1459 Description: the cat log was not initialized in recovery
1460 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1461 unlink log on mds will not be initialized until mds post
1462 recovery. And also in mds post recovery, the unlink log will
1463 initialization will be done asynchronausly, so there will be race
1464 between add unlink log and unlink log initialization.
1468 Description: brw_stats were being printed incorrectly
1469 Details : brw_stats were being printed as log2 but all of them were not
1470 recorded as log2. Also remove some code duplication arising from
1471 filter_tally_{read,write}.
1475 Frequency : rare, only in recovery.
1476 Description: ASSERTION(req->rq_type != LI_POISON) failed
1477 Details : imp_lock should be held while iterating over imp_sending_list for
1478 prevent destroy request after get timeout in ptlrpc_queue_wait.
1482 Description: replay-single.sh test 52 fails
1483 Details : A lock's skiplist need to be cleanup when it being unlinked
1484 from its resource list.
1488 Description: Short directio read returns full requested size rather than
1490 Details : Direct I/O operations should return actual amount of bytes
1491 transferred rather than requested size.
1493 Severity : enhancement
1495 Description: metadata RPC reduction (e.g. for rm performance)
1496 Details : decrease the amount of synchronous RPC between clients and servers
1497 by canceling conflicing lock before the operation on the client
1498 and packing thier handles into the main operation RPC to server.
1500 Severity : enhancement
1502 Description: Async OSC create to avoid the blocking unnecessarily.
1503 Details : If a OST has no remain object, system will block on the creating
1504 when need to create a new object on this OST. Now, ways use
1505 pre-created objects when available, instead of blocking on an
1506 empty osc while others are not empty. If we must block, we block
1507 for the shortest possible period of time.
1509 Severity : enhancement
1511 Description: refine locking for avoid write wrong info into lov_objid file
1512 Details : fix possible races with add new target and write/update data in
1515 --------------------------------------------------------------------------------
1517 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1522 Frequency : on some architectures
1524 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1525 Details : Uninitialized flags sometimes cause configuration commands to
1529 Frequency : patchless clients only
1531 Description: missing __iget() symbol export
1532 Details : The __iget() symbol export is missing. To avoid the need for
1533 this on patchless clients the deathrow inode reaper is turned
1534 off, and we depend on the VM to clean up old inodes. This
1535 dependency was during via the fix for bug 12181.
1537 --------------------------------------------------------------------------------
1539 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1541 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1542 INTEROPERATE with older versions automatically. In many cases a
1543 special upgrade step is needed. Please read the
1544 user documentation before upgrading any part of a live system.
1545 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1546 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1547 * WARNING: Lustre configuration and startup changes are required with
1548 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1554 Severity : enhancement
1556 Description: MountConf
1557 Details : Lustre configuration is now managed via mkfs and mount
1558 commands instead of lmc and lconf. New obd types (MGS, MGC)
1559 are added for dynamic configuration management. See
1560 https://mail.clusterfs.com/wikis/lustre/MountConf for
1563 Severity : enhancement
1565 Description: dynamic OST addition
1566 Details : OSTs can now be added to a live filesystem
1568 Severity : enhancement
1570 Description: startup order invariance
1571 Details : MDTs and OSTs can be started in any order. Clients only
1572 require the MDT to complete startup.
1574 Severity : enhancement
1576 Description: parallel, asynchronous orphan cleanup
1577 Details : orphan cleanup is now performed in separate threads for each
1578 OST, allowing parallel non-blocking operation.
1580 Severity : enhancement
1582 Description: optimized stripe assignment
1583 Details : stripe assignments are now made based on ost space available,
1584 ost previous usage, and OSS previous usage, in order to try
1585 to optimize storage space and networking resources.
1587 Severity : enhancement
1589 Description: Permanently set tunables
1590 Details : All writable /proc/fs/lustre tunables can now be permanently
1591 set on a per-server basis, at mkfs time or on a live system.
1593 Severity : enhancement
1595 Description: Lustre message v2
1596 Details : Add lustre message format v2.
1598 Severity : enhancement
1600 Description: client OST exclusion list
1601 Details : Clients can be started with a list of OSTs that should be
1602 declared "inactive" for known non-responsive OSTs.
1606 Description: SPEC SFS validation failure on NFS v2 over lustre.
1607 Details : Changes the blocksize for regular files to be 2x RPC size,
1608 and not depend on stripe size.
1610 Severity : enhancement
1612 Description: Multiple MD RPCs in flight.
1613 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1614 To avoid overly-overloading MDS, introduce a limit on number of
1615 MDS RPCs in flight for a single client and add /proc controls
1616 to adjust this limit.
1618 Severity : enhancement
1620 Description: client read/write statistics
1621 Details : Add client read/write call usage stats for performance
1622 analysis of user processes.
1623 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1624 file access. extents_stats shows chunk size distribution.
1625 extents_stats_per_process show chunk size distribution per
1628 Severity : enhancement
1630 Description: mds statistics
1631 Details : Add detailed mds operations statistics in
1632 /proc/fs/lustre/mds/*/stats.
1636 Description: Failure of copying files with lustre special EAs.
1637 Details : Client side always return success for setxattr call for lustre
1638 special xattr (currently only "trusted.lov").
1643 Description: Refcount LNET uuids
1644 Details : The global LNET uuid list grew linearly with every startup;
1645 refcount repeated list entries instead of always adding to
1648 Severity : enhancement
1650 Description: Dynamic service threads
1651 Details : Within a small range, start extra service threads
1652 automatically when the request queue builds up.
1655 Frequency : mixed-endian client/server environments
1657 Description: mixed-endian crashes
1658 Details : The new msg_v2 system had some failures in mixed-endian
1661 Severity : enhancement
1663 Description: Easy OST removal
1664 Details : OSTs can be permanently deactivated with e.g. 'lctl
1665 conf_param lustre-OST0001.osc.active=0'
1667 Severity : enhancement
1669 Description: MGS proc entries
1670 Details : Added basic proc entries for the MGS showing what filesystems
1673 Severity : enhancement
1675 Description: provide MGS failover
1676 Details : Added config lock reacquisition after MGS server failover.
1678 Severity : enhancement
1680 Description: add Linux 2.4 support
1681 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1685 Description: a large application tries to do I/O to the same resource and dies
1686 in the middle of it.
1687 Details : Check the req->rq_arrival time after the call to
1688 ost_brw_lock_get(), but before we do anything about
1689 processing it & sending the BULK transfer request. This
1690 should help move old stale pending locks off the queue as
1691 quickly as obd_timeout.
1694 Frequency : when an incorrect nid is specified during startup
1696 Description: ptlrpc connect to non-existant node causes kernel crash
1697 Details : LNET can't be re-entered from an event callback, which
1698 happened when we expire a message after the export has been
1699 cleaned up. Instead, hand the zombie cleanup off to another
1702 Severity : enhancement
1704 Description: plain/inodebits lock performance improvement
1705 Details : Grouping plain/inodebits in granted list by their request modes
1706 and bits policy, thus improving the performance of search through
1710 Frequency : only if OST filesystem is corrupted
1712 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1713 Details : for a short time RPCs with bulk IO are in the replay list,
1714 but replay of bulk IOs is unimplemented. If the OST filesystem
1715 is corrupted due to disk cache incoherency and then replay is
1716 started it is possible to trip an assertion. Avoid putting
1717 committed RPCs into the replay list at all to avoid this issue.
1720 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1722 Description: System hang on startup
1723 Details : This bug allowed the liblustre (e.g. catamount) client to
1724 return to the app before handling all startup RPCs. This
1725 could leave the node unresponsive to lustre network traffic
1726 and manifested as a server ptllnd timeout.
1728 Severity : enhancement
1730 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1731 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1732 boolean to control whether to print peer debug info when a
1733 client's RPC times out.
1736 Frequency : only for kernels with patches from Lustre below 1.4.3
1738 Description: Remove old rdonly API
1739 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1742 Frequency : only for devices with external journals
1744 Description: Set external device read-only also
1745 Details : During a commanded failover stop, we set the disk device
1746 read-only while the server shuts down. We now also set any
1747 external journal device read-only at the same time.
1750 Frequency : when upgrading from 1.4 while trying to change parameters
1752 Description: The wrong (new) MDC name was used when setting parameters for
1753 upgraded MDT's. Also allows changing of OSC (and MDC)
1754 parameters if --writeconf is specified at tunefs upgrade time.
1757 Frequency : when setting specific ost indicies
1759 Description: QOS code breaks on skipped indicies
1760 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1761 created with --index need not be sequential.
1763 Severity : enhancement
1765 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1766 Details : The uninit_groups feature works in conjunction with the kernel
1767 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1768 up the pass1 processing of e2fsck. This is a read-only feature
1769 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1770 mount filesystems that have had this feature enabled.
1772 Severity : enhancement
1774 Description: Improve multi-block allocation algorithm to avoid fragmentation
1775 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1776 allocation locality and avoid filesystem fragmentation.
1778 ------------------------------------------------------------------------------
1780 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1782 * Support for kernels:
1783 2.4.21-47.0.1.EL (RHEL 3)
1784 2.6.5-7.283 (SLES 9)
1785 2.6.9-42.0.10.EL (RHEL 4)
1786 2.6.12.6 vanilla (kernel.org)
1787 2.6.16.27-0.9 (SLES 10)
1788 * Recommended e2fsprogs version: 1.39.cfs5
1790 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1794 Frequency : occasional, depends on client load and configuration
1795 Bugzilla : 12181, 12203
1796 Description: data loss for recently-modified files
1798 Details : In some cases it is possible that recently written or created
1799 files may not be written to disk in a timely manner (this should
1800 normally be within 30s unless client IO load is very high).
1801 The problem appears as zero-length files or files that are a
1802 multiple of 1MB in size after a client crash or client eviction
1803 that are missing data at the end of the file.
1805 This problem is more likely to be hit on clients where files are
1806 repeatedly created and unlinked in the same directory, clients
1807 have a large amount of RAM, have many CPUs, the filesystem has
1808 many OSTs, the clients are rebooted frequently, and/or the files
1809 are not accessed by other nodes after being written.
1811 The presence of the problem can be detected by looking at
1812 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1813 smaller than the second (nr_unused) then dirty files will not
1814 be flushed automatically to disk. "sync; sleep 10" should be
1815 run several times on the node before unmounting it to update
1816 Lustre (this is also safe to run on nodes without this problem).
1818 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1819 that can cause this same problem, so customers using that kernel
1820 also need to update the kernel in addition to Lustre. In order
1821 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1823 It is normal that files written just before a client crash (less
1824 than 30s) may not yet have been flushed to disk, even for local
1828 Frequency : frequent on thin XT3 nodes
1830 Description: UUID collision on thin XT3 Linux nodes
1831 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1832 randomly, since we relied on an insufficiently-seeded PRNG.
1837 Description: OSS hangs after "All ost request buffers busy"
1838 Details : A deadlock between quota and journal operations caused OSS
1839 hangs after printing "All ost request buffers busy."
1842 Frequency : always on liblustre builds
1844 Description: Cleanup compiler warnings on liblustre
1847 Frequency : always on liblustre builds on XT3
1849 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1853 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1854 > 32000 subdirectories
1855 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1856 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1857 subdirectory count crosses 32000. This will aid e2fsck to
1858 correctly handle more than 32000 subdirectories.
1861 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1863 Description: System hang on startup
1864 Details : This bug allowed the liblustre (e.g. catamount) client to
1865 return to the app before handling all startup RPCs. This
1866 could leave the node unresponsive to lustre network traffic
1867 and manifested as a server ptllnd timeout.
1869 Severity : enhancement
1871 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1872 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1873 boolean to control whether to print peer debug info when a
1874 client's RPC times out.
1879 Description: make O_SYNC working on 2.6 kernels
1880 Details : 2.6 kernels use different method for mark pages for write,
1881 so need add a code to lustre for O_SYNC work.
1886 Description: Failure to close file and release space on NFS
1887 Details : Put inode details into lock acquired in ll_intent_file_open.
1888 Use mdc_intent_lock in ll_intent_open to properly
1889 detect all kind of errors unhandled by mdc_enqueue.
1894 Description: proc file read during shutdown sometimes raced obd removal,
1896 Details : Add lock to prevent obd access after proc file removal.
1899 Frequency : Only for files larger than 4GB on 32-bit clients.
1901 Description: improperly doing page alignment of locks
1902 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1903 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1908 Description: RPCs being resent when they shouldn't be
1909 Details : Some RPCs that should not be resent are being resent. This
1910 can cause inconsistencies in the RPC state machine. Do not
1911 resend such requests.
1914 Frequency : rare, only with NFS export
1916 Description: Crash on NFS re-export node
1917 Details : under very unusual load conditions an assertion is hit in
1918 ll_intent_file_open()
1921 Frequency : only if OST filesystem is corrupted
1923 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1924 Details : for a short time RPCs with bulk IO are in the replay list,
1925 but replay of bulk IOs is unimplemented. If the OST filesystem
1926 is corrupted due to disk cache incoherency and then replay is
1927 started it is possible to trip an assertion. Avoid putting
1928 committed RPCs into the replay list at all to avoid this issue.
1933 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1934 Details : Large single O_DIRECT read and write calls can fail to allocate
1935 a sufficiently large buffer to process the request. In case of
1936 allocation failure the allocation is retried with a smaller
1937 buffer and broken into smaller requests.
1939 Severity : enhancement
1941 Description: Add -o localflock option to simulate old noflock behaviour.
1942 Details : This will achieve local-only flock/fcntl locks coherentness.
1947 Description: versioning check is incomplete
1948 Details : Checking the version difference of client vs. server, report
1949 error if the gap is too big.
1954 Description: add support PG_writeback bit
1955 Details : add support for PG_writeback bit for Lustre, for more carefull
1956 work with page cache in 2.6 kernel. This also fix some deadlocks
1957 and remove hack for work O_SYNC with 2.6 kernel.
1959 Severity : enhancement
1961 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1962 Details : The uninit_groups feature works in conjunction with the kernel
1963 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1964 up the pass1 processing of e2fsck. This is a read-only feature
1965 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1966 mount filesystems that have had this feature enabled.
1968 Severity : enhancement
1970 Description: Improve multi-block allocation algorithm to avoid fragmentation
1971 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1972 allocation locality and avoid filesystem fragmentation.
1974 ------------------------------------------------------------------------------
1976 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1978 * Support for kernels:
1979 2.6.9-42.0.3.EL (RHEL 4)
1980 2.6.5-7.276 (SLES 9)
1981 2.4.21-47.0.1.EL (RHEL 3)
1982 2.6.12.6 vanilla (kernel.org)
1983 2.6.16.21-0.8 (SLES10)
1984 * Recommended e2fsprogs version: 1.39.cfs2-0
1986 * The backwards-compatible /proc/sys/portals symlink has been removed
1987 in this release. Before upgrading, please ensure that you change
1988 any configuration scripts or /etc/sysctl.conf files that access
1989 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1990 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1991 in advance of the upgrade on any system running Lustre 1.4.6 or
1992 newer, since /proc/sys/lnet was added in that version.
1993 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1997 Frequency : only when quota is used
1999 Description: avoid scanning export list for quota master
2000 Details : Change the algorithms to avoid scanning export list in order
2001 to improve the efficiency.
2004 Frequency : MDS failover only, very rarely
2006 Description: "went back in time" messages on mds failover
2007 Details : The greatest transno may be lost when the current operation
2008 finishes with an error (transno==0) and the client's last_rcvd
2009 record is over-written. Save the greatest transno in the
2010 mds_last_transno for this case.
2013 Frequency : always for specific kernels and striping counts
2015 Description: client may get "Matching packet too big" without ACL support
2016 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2017 when trying to access files in certain configurations. The
2018 clients should in fact be denied when mounting because they do
2019 not understand ACLs.
2022 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2024 Description: many clients connecting with IO in progress causes connect timeouts
2025 Details : Avoid synchronous journal commits to avoid delays caused by many
2026 clients connecting/disconnecting when bulk IO is in progress.
2027 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2028 OST_IO_PORTAL to avoid delays behind potentially many pending
2032 Frequency : occasionally with multiple writers to a single file
2034 Description: shared writes to file may result in wrong size reported by stat()
2035 Details : Allow growing of kms when extent lock is cancelled
2038 Frequency : always with random mmap IO to multi-striped file
2040 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2041 Details : Only if the hole is at the end of OST object so that kms is too
2042 small. Fix is to increase kms accordingly in ll_nopage.
2045 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2047 Description: writes to a missing object would leak memory on the OST
2048 Details : If there is an inconsistency between the MDS and OST filesystems,
2049 such that the MDS references an object that doesn't exist, writes
2050 to that object will leak memory due to incorrect cleanup in the
2051 error handling path, eventually running out of memory on the OST.
2056 Description: Creating too long symlink causes lustre errors
2057 Details : Check symlink and name lengths before sending requests to MDS.
2060 Frequency : only if flock is enabled (not on by default)
2062 Description: posix locks not released on fd closure on 2.6.9+
2063 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2064 kernels, this caused such locks not to be released on fd close and
2065 then assertions on fs unmount about still used locks.
2068 Frequency : MDS failover only, very rarely
2070 Description: clients may get ASSERTION(granted_lock != NULL)
2071 Details : When request was taking a long time, and a client was resending
2072 a getattr by name lock request. The were multiple lock requests
2073 with the same client lock handle and
2074 mds_getattr_name->fixup_handle_for_resent_request found one of the
2075 lock handles but later failed with ASSERTION(granted_lock != NULL).
2080 Description: handle->h_buffer_credits > 0, assertion failure
2081 Details : h_buffer_credits is zero after truncate, causing assertion
2082 failure. This patch extends the transaction or creates a new
2086 Frequency : NFS re-export or patchless client
2087 Bugzilla : 11179, 10796
2088 Description: Crash on NFS re-export node (__d_move)
2089 Details : We do not want to hash the dentry if we don't have a lock.
2090 But if this dentry is later used in d_move, we'd hit uninitialised
2091 list head d_hash, so we just do this to init d_hash field but
2092 leave dentry unhashed.
2095 Frequency : NFS re-export or patchless client
2097 Description: NFS exports has problem with symbolic link
2098 Details : lustre client didn't properly install dentry when re-exported
2099 to NFS or running patchless client.
2102 Frequency : NFS re-export or patchless client
2104 Description: Various nfs/patchless fixes.
2105 Details : fixes reuse disconected alias for lookup process - this fixes
2106 warning "find_exported_dentry: npd != pd",
2107 fix permission error with open files at nfs.
2108 fix apply umask when do revalidate.
2111 Frequency : occasional
2113 Description: Crash on NFS re-export node
2114 Details : calling clear_page() on the wrong pointer triggered oops in
2115 generic_mapping_read().
2118 Frequency : rarely, using O_DIRECT IO
2120 Description: unaligned directio crashes client with LASSERT
2121 Details : check for unaligned buffers before trying any requests.
2124 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2126 Description: stale data returned from RAID cache
2127 Details : If only a small amount of IO is done to the RAID device before
2128 reading it again it is possible to get stale data from the RAID
2129 cache instead of reading it from disk.
2132 Frequency : always for sles10 kernel
2134 Description: sles10 support
2135 Details : ll_follow_link: compile fixes and using of nd_set_link
2136 under newer kernels.
2139 Frequency : depends on arch, kernel and compiler version, always on sles10
2142 Description: recursive or deep enough symlinks cause stack overflow
2143 Details : getting rid of large stack-allocated variable in
2147 Frequency : depends on hardware
2149 Description: lustre write performance loss in the SLES10 kernel
2150 Details : the performance loss is caused by using of write barriers in the
2151 ext3 code. The SLES10 kernel turns barrier support on by
2152 default. The fix is to undo that change for ldiskfs.
2154 ------------------------------------------------------------------------------
2156 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2158 * Support for kernels:
2159 2.6.9-42.0.3EL (RHEL 4)
2160 2.6.5-7.276 (SLES 9)
2161 2.4.21-47.0.1.EL (RHEL 3)
2162 2.6.12.6 vanilla (kernel.org)
2166 Frequency : quota enabled and large files being deleted
2168 Description: releasing more than 4GB of quota at once hangs OST
2169 Details : If a user deletes more than 4GB of files on a single OST it
2170 will cause the OST to spin in an infinite loop. Release
2171 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2176 Description: statfs data retrieved from /proc may be stale or zero
2177 Details : When reading per-device statfs data from /proc, in the
2178 {kbytes,files}_{total,free,avail} files, it may appear
2179 as zero or be out of date.
2182 Frequency : systems with MD RAID1 external journal devices
2184 Description: lconf's call to blkid is confused by RAID1 journal devices
2185 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2186 of returning all block devices that match the journal UUID.
2189 Frequency : always, for aggregate stripe size over 4GB
2191 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2192 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2193 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2196 Frequency : always if "lfs find" used on a local file/directory
2198 Description: "lfs find" segfaults if used on a local file/directory
2199 Details : The case where a directory component was not specified wasn't
2200 handled correctly. Handle this properly.
2203 Frequency : always on ppc64
2205 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2206 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2207 perform endianness conversion.
2210 Frequency : rarely (truncate to non-zero file size after write under load)
2211 Bugzilla : 10730, 10687
2212 Description: Files padded with zeros to next 4K multiple
2213 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2214 it is possible that files that are truncated to a non-zero size
2215 immediately after being written are filled with zero bytes beyond
2216 the truncated size. No file data is lost.
2218 Severity : enhancement
2220 Description: Allow recovery/failover for liblustre clients.
2221 Details : liblustre clients were unaware of failover configurations until
2224 Severity : enhancement
2226 Description: user file locks should fail when not mounting with flock option
2227 Details : Set up an error-returning stub in ll_file_operations.lock field
2228 to prevent incorrect behaviour when client is mounted without
2229 flock option. Also, set up properly f_op->flock field for
2233 Frequency : always on ia64
2235 Description: "lfs df" loops on printing out MDS statfs information
2236 Details : The obd_ioctl_data was not initialized and in some systems
2237 this caused a failure during the ioctl that did not return
2238 an error. Initialize the struct and return an error on failure.
2241 Frequency : SLES 9 only
2243 Description: Error of copying files with lustre special EAs as root
2244 Details : Client side always return success for setxattr call for lustre
2245 special xattr (currently only "trusted.lov").
2248 Frequency : rarely on clusters with both ia64+i386 clients
2250 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2251 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2252 clients are doing concurrent writes to the same file it is
2253 possible that the ia64 clients may LASSERT because the OST
2254 extent locks are not PAGE_SIZE aligned. Ensure that grown
2255 locks are always aligned on the request boundary.
2258 Frequency : specific use, occasional
2260 Description: Overwriting in use executable truncates on-disk binary image
2261 Details : If one node attempts to overwrite an executable in use by
2262 another node, we now correctly return ETXTBSY instead of
2263 truncating the file.
2265 Severity : enhancement
2267 Description: Async OSC create to avoid the blocking unnecessarily.
2268 Details : If a OST has no remain object, system will block on the creating
2269 when need to create a new object on this OST. Now, ways use
2270 pre-created objects when available, instead of blocking on an
2271 empty osc while others are not empty. If we must block, we block
2272 for the shortest possible period of time.
2277 Description: chmod on Lustre root is propagated to other clients
2278 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2279 invalid by the follow_mount time.
2282 Frequency : liblustre clients only
2284 Description: Race in 'instant cancel' lock handling could lead to such locks
2285 never to be granted in case of SMP MDS
2286 Details : Do not destroy not yet granted but cbpending locks in
2290 Frequency : replay/resend of open
2292 Description: non null lock assetion failure in mds_intent_policy
2293 Details : Trying to replay/resend lockless open requests resulted in
2294 mds_open() returning 0 with no lock. Now it sets a flag if
2295 a lock is going to be returned.
2297 Severity : enhancement
2299 Description: Checksum enhancements
2300 Details : New checksum enhancements allow for resending RPCs that failed
2303 Severity : enhancement
2305 Description: Tunables on number of dirty pages in cacche
2306 Details : Allow to set limit on number of dirty pages cached.
2311 Description: client crash on unmount - lock still has references
2312 Details : In some error handling cases it was possible to leak a lock
2313 reference on a client while accessing a file. This was not
2314 harmful to the client during operation, but would cause the
2315 client to crash when the filesystem is unmounted.
2318 Frequency : specific case, rare
2320 Description: ETXTBSY on mds though file not in use by client
2321 Details : ETXTBSY is no longer incorrectly returned when attempting to
2322 chmod or chown a directory that the user previously tried to
2323 execute or a currently-executing binary.
2326 Frequency : extremely rare except on liblustre-based clients
2328 Description: Lustre space not freed when files are deleted
2329 Details : Clean up open-unlinked files after client eviction. Previously
2330 the unlink was skipped and the files remained as orphans.
2335 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2336 Details : In some cases it was possible to send a blocking callback to
2337 a client doing a glimpse, even though that client didn't get
2338 a lock granted. When the glimpse lock is cancelled on the OST
2339 the freed lock is left on the waiting list and corrupted the list.
2342 Frequency : all core dumps
2344 Description: Broke core dumps to lustre
2345 Details : Negative dentry may be unhashed if parent does not have UPDATE
2346 lock, but some callers, e.g. do_coredump, expect dentry to be
2347 hashed after successful create, hash it in ll_create_it.
2349 ------------------------------------------------------------------------------
2351 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2353 * Support for kernels:
2354 2.6.9-42.0.2.EL (RHEL 4)
2355 2.6.5-7.276 (SLES 9)
2356 2.4.21-40.EL (RHEL 3)
2357 2.6.12.6 vanilla (kernel.org)
2361 Frequency : always on RHEL 3
2363 Description: Number of open files grows over time
2364 Details : The number of open files grows over time, whether or not
2365 Lustre is started. This was due to a filp leak introduced
2366 by one of our kernel patches.
2368 ------------------------------------------------------------------------------
2370 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2372 * Support for kernels:
2373 2.6.9-42.EL (RHEL 4)
2374 2.6.5-7.276 (SLES 9)
2375 2.4.21-40.EL (RHEL 3)
2376 2.6.12.6 vanilla (kernel.org)
2381 Bugzilla : 5719, 9635, 9792, 9684
2382 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2383 Details : If a server is under heavy load and cannot reply to new
2384 connection requests before the client resends the (re)connect,
2385 the connection handling code can behave badly if two service
2386 threads are concurrently handing separate (re)connections from
2387 the same client. Add better locking to the connection handling
2388 code, and ensure that only a single connection will be processed
2389 for a given client UUID, even if the lock is dropped.
2391 Severity : enhancement
2393 Description: add TCP zero-copy support to kernel
2394 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2395 sends if the hardware supports scatter-gather and checksumming.
2396 This allows socklnd to do client-write and server-read more
2397 efficiently and reduce CPU utilization from skbuf copying.
2400 Frequency : only if NFS exporting from client
2402 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2403 Details : If setting attributes on a file created under NFS that had
2404 never been opened it would be possible to oops the client
2405 if the file had no objects.
2408 Frequency : always for liblustre
2410 Description: liblustre client does MDS+OSTs setattr RPC for each write
2411 Details : When doing a write from a liblustre client, the client
2412 incorrectly issued an RPC to the MDS and each OST the file was
2413 striped over in order to update the timestamps. When writing
2414 with small chunks and many clients this could overwhelm the MDS
2415 with RPCs. In all cases it would slow down the write because
2416 these RPCs are unnecessary.
2418 Severity : enhancement
2420 Description: allow number of MDS service threads to be changed at module load
2421 Details : It is now possible to change the number of MDS service threads
2422 running. Adding "options mds mds_num_threads={N}" to the MDS's
2423 /etc/modprobe.conf will set the number of threads for the next
2424 time Lustre is restarted (assuming the "mds" module is also
2425 reloaded at that time). The default number of threads will
2426 stay the same, 32 for most systems.
2431 Description: OST crash if filesystem is unformatted or corrupt
2432 Details : If an OST is started on a device that has never been formatted
2433 or if the filesystem is corrupt and cannot even mount then the
2434 error handling cleanup routines would dereference a NULL pointer.
2439 Description: NULL pointer deref in llap_from_page.
2440 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2441 incorrect) mapping if the page was truncated/reclaimed while it was
2442 searched for. Check for this condition and skip such pages when
2443 doing readahead. Introduce extra check to llap_from_page() to
2444 verify page->mapping->host is non-NULL (so page is not anonymous).
2447 Frequency : Sometimes when using sys_sendfile
2449 Description: "page not covered by a lock" warnings from ll_readpage
2450 Details : sendfile called ll_readpage without right page locks present.
2451 Now we introduced ll_file_sendfile that does necessary locking
2452 around call to generic_file_sendfile() much like we do in
2456 Frequency : with certain MDS communication failures at client mount time
2458 Description: NULL pointer deref after failed client mount
2459 Details : a client connection request may delayed by the network layer
2460 and not be sent until after the PTLRPC layer has timed out the
2461 request. If the client fails the mount immediately it will try
2462 to clean up before the network times out the request. Add a
2463 reference from the request import to the obd device and delay
2464 the cleanup until the network drops the request.
2467 Frequency : occasionally during client (re)connect
2469 Description: assertion failure during client (re)connect
2470 Details : processing a client connection request may be delayed by the
2471 client or server longer than the client connect timeout. This
2472 causes the client to resend the connection request. If the
2473 original connection request is replied in this interval, the
2474 client may trip an assertion failure in ptlrpc_connect_interpret()
2475 which thought it would be the only running connect process.
2478 Frequency : only with obd_echo servers and clients that are rebooted
2480 Description: kernel BUG accessing uninitialized data structure
2481 Details : When running an obd_echo server it did not start the ping_evictor
2482 thread, and when a client was evicted an uninitialized data
2483 structure was accessed. Start the ping_evictor in the RPC
2484 service startup instead of the OBD startup.
2486 Severity : enhancement
2487 Bugzilla : 10193 (patchless)
2488 Description: Remove dependency on various unexported kernel interfaces.
2489 Details : No longer need reparent_to_init, exit_mm, exit_files,
2490 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2493 Frequency : rare (only users of deprecated and unsupported LDAP config)
2495 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2496 Details : LDAP apparently contains 'lustreName' attributes instead of
2497 'name'. A simple remapping of the name is sufficient.
2500 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2502 Description: waiting_locks_callback trips kernel BUG if client is evicted
2503 Details : Running with the dump_on_timeout debug flag turned on makes
2504 it possible that the waiting_locks_callback() can try to dump
2505 the Lustre kernel debug logs from an interrupt handler. Defer
2506 this log dumping to the expired_lock_main() thread.
2508 Severity : enhancement
2510 Description: Support NFS exporting on 2.6 kernels.
2511 Details : Implement non-rawops metadata methods for NFS server to use without
2512 changing NFS server code.
2515 Frequency : very rare (synthetic metadata workload only)
2517 Description: two racing renames might cause an MDS thread to deadlock
2518 Details : Running the "racer" program may cause one MDS thread to rename
2519 a file from being the source of a rename to being the target of
2520 a rename at exactly the same time that another thread is doing
2521 so, and the second thread has already enqueued these locks after
2522 doing a lookup of the target and is trying to relock them in
2523 order. Ensure that we don't try to re-lock the same resource.
2526 Frequency : only very large systems with liblustre clients
2528 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2529 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2530 all OSTs in parallel. This allows the eviction of stale liblustre
2531 clients to proceed much faster than if they were done in series,
2532 and also offers similar improvements for other set_info RPCs.
2537 Description: excessive CPU usage during initial read phase on client
2538 Details : During the initial read phase on a client, it would agressively
2539 retry readahead on the file, consuming too much CPU and impacting
2540 performance (since 1.4.5.8). Improve the readahead algorithm
2541 to avoid this, and also improve some other common cases (read
2542 of small files in particular, where "small" is files smaller than
2543 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2548 Description: MDS crash when receiving packet with unknown intent.
2549 Details : Do not LBUG in unknown intent case, just return -EFAULT
2551 Severity : enhancement
2552 Bugzilla : 9293, 9385
2553 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2554 Details : Do not serialize getattr (non-intent version) and statfs.
2557 Frequency : occasional, when OST network is overloaded/intermittent
2559 Description: client evicted by OST after bulk IO timeout
2560 Details : If a client sends a bulk IO request (read or write) the OST
2561 may evict the client if it is unresposive to its data GET/PUT
2562 request. This is incorrect if the network is overloaded (takes
2563 too long to transfer the RPC data) or dropped the OST GET/PUT
2564 request. There is no need to evict the client at all, since
2565 the pinger and/or lock callbacks will handle this, and the
2566 client can restart the bulk request.
2569 Frequency : Always when mmapping file with no objects
2571 Description: client crashes when mmapping file with no objects
2572 Details : Check that we actually have objects in a file before doing any
2573 operations on objects in ll_vm_open, ll_vm_close and
2579 Description: Request leak when working with deleted CWD
2580 Details : Introduce advanced request refcount tracking for requests
2581 referenced from lustre intent.
2583 Severity : Enhancement
2585 Description: Cache open file handles on client.
2586 Details : MDS now will return special lock along with openhandle, if
2587 requested and client is allowed to hold openhandle, even if unused,
2588 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2589 closing files for every read/write openration.
2591 Severity : Enhancement
2593 Description: Cache open negative dentries on client when possible.
2594 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2595 negative dentries on lock revocation.
2600 Description: Remounting a client read-only wasn't possible with a zconf mount
2601 Details : It wasn't possible to remount a client read-only with llmount.
2603 Severity : enhancement
2604 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2605 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2606 MPICH is included to add Lustre-specific ADIO interfaces.
2607 This is based closely on the UFS ADIO layer and only differs
2608 in file creation, in order to allow the OST striping to be set.
2609 This is user-contributed code and not supported by CFS.
2614 Description: extended inode attributes (immutable, append-only) work improperly
2615 when 2.4 and 2.6 kernels are used on client/server or vice versa
2616 Details : Introduce kernel-independent values for these flags.
2618 Severity : enhancement
2621 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2622 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2623 granularity. Now, Fractional MB granularity is be supported,
2624 this is very useful for low memory system.
2626 Severity : enhancement
2628 Description: Getattr by fid
2629 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2630 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2631 have locks and avoids getattr rpc requests for it on every stat.
2634 Frequency : Always, for filesystems larger than 2TB
2636 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2637 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2638 filesystems larger than 2TB to be mounted with mballoc enabled.
2641 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2643 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2644 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2645 filesystems larger than 2TB is mounted then the kernel will
2646 silently corrupt the start of the filesystem. CONFIG_LBD is
2647 enabled for all CFS-supported kernels, but the possibility of
2648 this happening with a modified kernel config exists.
2650 Severity : enhancement
2652 Description: add client O_DIRECT support for 2.6 kernels
2653 Details : It is now possible to do O_DIRECT reads and writes to files
2654 in the Lustre client mountpoint on 2.6 kernel clients.
2656 Severity : enhancement
2658 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2659 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2660 parallel, not waiting for response from every OST before sending
2661 a rpc to the next OST.
2666 Description: setattr vs write race when updating file timestamps
2667 Details : Client processes that update a file timestamp into the past
2668 right after writing to the file (e.g. tar) it is possible that
2669 the updated file modification time can be reset to the current
2670 time due to a race between processing the setattr and write RPC.
2672 Severity : enhancement
2674 Description: Bring 'lfs find' closer in line with regular Linux find.
2675 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2676 -print0 options and obtains all the needed info through the lustre
2679 Severity : enhancement
2681 Description: support up to 1024 configured devices on one node
2682 Details : change obd_dev array from statically allocated to dynamically
2683 allocated structs as they are first used to reduce memory usage
2688 Description: Flush dirty partially truncated pages during truncate
2689 Details : Immediatelly flush partially truncated pages in filter_setattr,
2690 this way we completely avoid having any pages in page cache on OST
2691 and can retire ugly workarounds during writes to flush such pages.
2696 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2697 Details : protect inode from truncation within vfs_unlink() context
2698 just take a reference before calling vfs_unlink() and release it
2699 when parent's i_sem is free.
2704 Description: last_id value checked outside lock on OST caused LASSERT failure
2705 Details : If there were multiple MDS->OST object precreate requests in
2706 flight, it was possible that the OST's last object id was checked
2707 outside a lock and incorrectly tripped an assertion. Move checks
2708 inside locks, and discard old precreate requests.
2711 Frequency : always, if extents are used on OSTs
2713 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2714 Details : OSTs using the extents format would not zero the high 16 bits of
2715 the index physical block number. This is not a problem for any
2716 OST filesystems smaller than 16TB, and no kernels support ext3
2717 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2718 new/modified files) and can be fixed for existing filesystems
2719 with e2fsprogs-1.39-cfs1.
2724 Description: import connection selection may be incorrect if timer wraps
2725 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2726 import connections to be ignored if the 32-bit jiffies counter
2727 wraps. Use a 64-bit jiffies counter.
2730 Frequency : very large clusters immediately after boot
2732 Description: LNET request buffers exhausted under heavy short-term load
2733 Details : If a large number of client requests are generated on a service
2734 that has previously never seen so many requests it is possible
2735 that the request buffer growth cannot keep up with the spike in
2736 demand. Instead of dropping incoming requests, they are held in
2737 the LND until the RPC service can accept more requests.
2740 Frequency : Sometimes during replay
2742 Description: Assertion failure in ll_local_open after replay.
2743 Details : If replay happened on an open request reply before we were able
2744 to set replay handler, reply will become not swabbed tripping the
2745 assertion in ll_local_open. Now we set the handler right after
2746 recognising of open request
2749 Frequency : very rare
2751 Description: kernel reports "badness in vsnprintf"
2752 Details : Reading from the "recovery_status" /proc file in small chunks
2753 may cause a negative length in lprocfs_obd_rd_recovery_status()
2754 call to vsnprintf() (which is otherwise harmless). Exit early
2755 if there is no more space in the output buffer.
2757 Severity : enhancement
2759 Description: clear OBD RPC statistics by writing to them
2760 Details : It is now possible to clear the OBD RPC statistics by writing
2761 to the "stats" file.
2766 Description: Client mtime is not the same on different clients after utimes
2767 Details : In some cases, the client was using the utimes() syscall on
2768 a file cached on another node. The clients now validate the
2769 ctime from the MDS + OSTs to determine which one is right.
2774 Description: Inability to activate failout mode
2775 Details : lconf script incorrectly assumed that in pythong string's numeric
2776 value is used in comparisons.
2779 Frequency : always with multiple stripes per file
2781 Description: Inefficient object allocation for mutli-stripe files
2782 Details : When selecting which OSTs to stripe files over, for files with
2783 a stripe count that divides evenly into the number of OSTs,
2784 the MDS is always picking the same starting OST for each file.
2785 Return the OST selection heuristic to the original design.
2790 Description: mount failures may take full timeout to return an error
2791 Details : Under some heavy load conditions it is possible that a
2792 failed mount can wait for the full obd_timeout interval,
2793 possibly several minutes, before reporting an error.
2794 Instead return an error as soon as the status is known.
2796 Frequency : quota enabled and large files being deleted
2798 Description: releasing more than 4GB of quota at once hangs OST
2799 Details : If a user deletes more than 4GB of files on a single OST it
2800 will cause the OST to spin in an infinite loop. Release
2801 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2806 Description: statfs data retrieved from /proc may be stale or zero
2807 Details : When reading per-device statfs data from /proc, in the
2808 {kbytes,files}_{total,free,avail} files, it may appear
2809 as zero or be out of date.
2812 Frequency : systems with MD RAID1 external journal devices
2814 Description: lconf's call to blkid is confused by RAID1 journal devices
2815 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2816 of returning all block devices that match the journal UUID.
2819 Frequency : always, for aggregate stripe size over 4GB
2821 Description: assertion fails when trying to use 4GB stripe size
2822 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2823 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2826 Frequency : always on ppc64
2828 Description: the first write on an ext3 filesystem with mballoc got stuck
2829 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2830 perform endianness conversion.
2832 ------------------------------------------------------------------------------
2834 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2836 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2837 INTEROPERATE with older versions automatically. Please read the
2838 user documentation before upgrading any part of a live system.
2839 * WARNING: Lustre networking configuration changes are required with
2840 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2843 * Support for kernels:
2844 2.6.9-22.0.2.EL (RHEL 4)
2845 2.6.5-7.244 (SLES 9)
2846 2.6.12.6 vanilla (kernel.org)
2849 Severity : enhancement
2850 Bugzilla : 7981/8208
2851 Description: Introduced Lustre Networking (LNET)
2852 Details : LNET is new networking infrastructure for Lustre, it includes
2853 a reorganized network configuration mode (see the user
2854 documentation for full details) as well as support for routing
2855 between different network fabrics. Lustre Networking Devices
2856 (LNDs) for the supported network fabrics have also been
2857 created for this new infrastructure.
2859 Severity : enhancement
2860 Description: Introduced Access control lists
2861 Details : clients can set ACLs on files and directories in order to have
2862 more fine-grained permissions than the standard Unix UGO+RWX.
2863 The MDS must be started with the "-o acl" mount option.
2865 Severity : enhancement
2866 Description: Introduced filesystem quotas
2867 Details : Administrators may now establish per-user quotas on the
2870 Severity : enhancement
2872 Description: Configuration change for the XT3
2873 The PTLLND is now used to run Lustre over Portals on the XT3
2874 The configure option(s) --with-cray-portals are no longer used.
2875 Rather --with-portals=<path-to-portals-includes> is used to
2876 enable building on the XT3. In addition to enable XT3 specific
2877 features the option --enable-cray-xt3 must be used.
2882 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2883 Details : A race between allocating a new llcd and re-getting the llcd_lock
2884 allowed another thread to grab newly-allocated llcd.
2886 Severity : enhancement
2888 Description: 2.6 OST async journal commit and locking fix to improve performance
2889 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2890 2.6 kernels are now async as they already were in 2.4 kernels so
2891 that they can commit concurrently with the network bulk transfer.
2892 For block-allocated files the filter allocation semaphore is held
2893 to avoid filesystem fragmentation during allocation. BKL lock
2894 removed for 2.6 xattr operations where it is no longer needed.
2899 Description: lconf incorrectly determined whether two IP networks could talk
2900 Details : In some more complicated routing and multiple-network
2901 configurations, lconf will avoid trying to make a network
2902 connection to a disjoint part of the IP space. It was doing the
2903 math incorrectly for one set of cases.
2908 Description: Fix for potential infinite loop processing records in an llog.
2909 Details : If an llog record is corrupted/zeroed, it is possible to loop
2910 forever in llog_process(). Validate the llog record length
2911 and skip the remainder of the block on error.
2914 Frequency : occasional (liblustre only)
2916 Description: liblustre could not open files whose last component is a symlink
2917 Details : sysio_path_walk() would incorrectly pass the open intent to
2918 intermediate path components.
2921 Frequency : rare (liblustre only with non-standard tuning)
2922 Bugzilla : 7201 (7350)
2923 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2924 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2925 after it has been released, only do this for referenced locks
2927 Severity : enhancement
2929 Description: specifying an (invalid) directory default stripe_size of -1
2930 would reset the directory default striping
2931 Details : stripe_size -1 was used internally to signal directory stripe
2932 removal, now use "all default" to signal dir stripe removal
2933 as a directory striping of "all default" is not useful
2936 Frequency : common for large clusters running liblustre clients
2938 Description: doing an ls when liblustre clients are running is slow
2939 Details : sending a glimpse AST to a liblustre client waits for every AST
2940 to time out, as liblustre clients will not respond. Since they
2941 cannot cache data we refresh the OST lock LVB from disk instead.
2943 Severity : enhancement
2945 Description: doing an ls at the same time as file IO can be slow
2946 Details : enqueue and other "small" requests can be blocked behind many
2947 large IO requests. Create a new OST IO portal for non-IO
2948 requests so they can be processed faster.
2951 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2953 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2954 created file with mode 000
2955 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2956 allow an ftruncate() as a non-root user to a file with mode 000.
2957 Lustre now respects this flag to disable mode checks when
2958 truncating a file owned by the user
2961 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2964 Description: Revoking locks from clients that went dead or catatonic might take
2966 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2967 cancellation of such locks instant on servers without waiting for
2968 any reply from clients. Clients drops these locks when cancel
2969 notification from server is received without replying.
2972 Frequency : liblustre-only, when liblustre client dies or becomes busy
2974 Description: Doing ls on Linux clients can take a long time with active
2976 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2977 granting such locks to it in the first place if possible. Locks
2978 are taken by proxy on the OST during the read or write and
2979 dropped immediately afterward. Add connect flags handling, do
2980 not grant locks to liblustre clients for glimpse ASTs.
2982 Severity : enhancement
2984 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2985 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2986 sequential reads and behaves badly with applications doing random
2987 reads. Improve it by reading ahead at least read region, and
2988 avoiding excessive large RPC for small reads.
2990 Severity : enhancement
2992 Description: Creating more than 1000 files for a single job may cause a load
2993 imbalance on the OSTs if there are also a large number of OSTs.
2994 Details : qos_prep_create() uses an OST index reseed value that is an
2995 even multiple of the number of available OSTs so that if the
2996 reseed happens in the middle of the object allocation it will
2997 still utilize the OSTs as uniformly as possible.
3002 Description: OST or MDS may oops in ping_evictor_main()
3003 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3004 but doesn't restart at beginning of obd_exports_timed list
3007 Severity : enhancement
3009 Description: improve by-nid export eviction on the MDS and OST
3010 Details : allow multiple exports with the same NID to be evicted at one
3011 time without re-searching the exports list.
3014 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3016 Description: MDS may oops in groups_free()
3017 Details : in rare race conditions a newly allocated group_info struct is
3018 freed again, and this can be NULL. The 2.4 compatibility code
3019 for groups_free() checked for a NULL pointer, but 2.6 did not.
3022 Frequency : common for liblustre clients doing little filesystem IO
3023 Bugzilla : 9352, 7313
3024 Description: server may evict liblustre clients accessing contended locks
3025 Details : if a client is granted a lock or receives a completion AST
3026 with a blocking AST already set it would not reply to the AST
3027 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3030 Frequency : lfs setstripe, only systems with more than 160 OSTs
3032 Description: unable to set striping with a starting offset beyond OST 160
3033 Details : llapi_create_file() incorrectly limited the starting stripe
3034 index to the maximum single-file stripe count.
3037 Frequency : LDAP users only
3039 Description: lconf did not handle in-kernel recovery with LDAP properly
3040 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3042 Severity : enhancement
3044 Description: bind OST threads to NUMA nodes to improve performance
3045 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3046 node and do their allocations there to localize memory access
3048 Severity : enhancement
3050 Description: llmount can determine client NID directly from Myrinet (GM)
3051 Details : the client NID code from gmnalnid was moved directly into
3052 llmount, removing the need to use this or specifying the
3053 client NID explicitly when mounting GM clients with zeroconf
3056 Frequency : if client is started with down MDS
3058 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3059 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3060 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3061 ptlrpc_import_delayed_req() because we didn't check if the
3062 request was interrupted, and we also didn't break out of the
3063 event loop if there was no timeout
3068 Description: data loss during non-page-aligned writes to a single file from
3069 both multiple nodes and multiple threads on one node at same time
3070 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3071 inconsistency led to false short-reads, that were cached and later
3072 used by ->prepare_write() to fill in partially written page,
3073 leading to data loss.
3076 Frequency : always, if lconf --abort_recovery used
3078 Description: lconf --abort_recovery fails with 'Operation not supported'
3079 Details : lconf was attempting to abort recovery on the MDT device and not
3082 Severity : enhancement
3084 Description: remove cleanup logs
3085 Details : replace lconf-generated cleanup logs with lustre internal
3086 cleanup routines. Eliminates the need for client-cleanup and
3089 Severity : enhancement
3091 Description: add support for EAs (user and system) on lustre filesystems
3092 Details : it is now possible to store extended attributes in the Lustre
3093 client filesystem, and with the user_xattr mount option it
3094 is possible to allow users to store EAs on their files also
3096 Severity : enhancement
3098 Description: Add possibility (config option) to show minimal available OST free
3100 Details : When compiled with --enable-mindf configure option, statfs(2)
3101 (and so, df) will return least minimal free space available from
3102 all OSTs as amount of free space on FS, instead of summary of
3103 free spaces of all OSTs.
3105 Severity : enhancement
3107 Description: do not expand extent locks acquired on OST-side
3108 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3109 by server: they are not cached anyway.
3112 Frequency : when mmap is used/binaries executed from Lustre
3114 Description: Unmmap pages before throwing them away from read cache.
3115 Details : llap_shrink cache now attempts to unmap pages before discarding
3116 them (if unmapping failed - do not discard). SLES9 kernel has
3117 extra checks that trigger if this unmapping is not done first.
3122 Description: lconf didn't resolve symlinks before checking to see whether a
3123 given mountpoint was already in use
3126 Frequency : when migrating failover services
3127 Bugzilla : 6395, 9514
3128 Description: When migrating a subset of services from a node (e.g. failback
3129 from a failover service node) the remaining services would
3130 time out and evict clients.
3131 Details : lconf --force (implied by --failover) sets the global obd_timeout
3132 to 5 seconds in order to quickly disconnect, but this caused
3133 other RPCs to time out too quickly. Do not change the global
3134 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3136 Severity : enhancement
3137 Frequency : if MDS is started with down OST
3138 Bugzilla : 9439,5706
3139 Description: Allow startup/shutdown of an MDS without depending on the
3140 availability of the OSTs.
3141 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3142 Add appropriate locking and lov-osc refcounts for safe
3143 cleaning. Add osc abort_inflight calls in case the
3144 synchronize never started.
3147 Frequency : occasional (Cray XT3 only)
3149 Description: root not authorized to access files in CRAY_PORTALS environment
3150 Details : The client process capabilities were not honoured on the MDS in
3151 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3152 been accessed by an authorized user then root was able to access
3153 the file on the local client also. The root user capabilities
3154 are now allowed on the MDS, as this environment has secure UID.
3157 Frequency : occasional
3159 Description: ldiskfs "too long searching" message happens too often
3160 Details : A debugging message (otherwise harmless) prints too often on
3161 the OST console. This has been reduced to only happen when
3162 there are fragmentation problems on the filesystem.
3167 Description: Division by zero in statfs when all OSCs are inactive
3168 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3169 lov_statfs divides by value returned by lov_get_stripecnt().
3173 Bugzilla : 9489, 3273
3174 Description: First write from each client to each OST was only 4kB in size,
3175 to initialize client writeback cache, which caused sub-optimal
3176 RPCs and poor layout on disk for the first writen file.
3177 Details : Clients now request an initial cache grant at (re)connect time
3178 and so that they can start streaming writes to the cache right
3179 away and always do full-sized RPCs if there is enough data.
3180 If the OST is rebooted the client also re-establishes its grant
3181 so that client cached writes will be honoured under the grant.
3186 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3187 Details : Now I/O RPCs go to different portal number and (presumably) fast
3188 lock requests (and glimses) and other RPCs get their own service
3189 threads pool that should be able to service those RPCs
3192 Severity : enhancement
3194 Description: Ability to exchange lustre version between client and servers and
3195 issue warnings at client side if client is too old. Also for
3196 liblustre clients there is ability to refuse connection of too old
3198 Details : New 'version' field is added to connect data structure that is
3199 filled with version info. That info is later checked by server and
3203 Frequency : rare, liblustre only.
3204 Bugzilla : 9296, 9581
3205 Description: Two simultaneous writes from liblustre at offset within same page
3206 might proceed at the same time overwriting eachother with stale
3208 Details : I/O lock withing llu_file_prwv was released too early, before data
3209 actually was hitting the wire. Extended lock-holding time until
3210 server acknowledges receiving data.
3213 Frequency : extremely rare. Never observed in practice.
3215 Description: avoid generating lustre_handle cookie of 0.
3216 Details : class_handle_hash() generates handle cookies by incrementing
3217 global counter, and can hit 0 occasionaly (this is unlikely, but
3218 not impossible, because initial value of cookie counter is
3219 selected randonly). Value of 0 is used as a sentinel meaning
3220 "unassigned handle" --- avoid it. Also coalesce two critical
3221 sections in this function into one.
3223 Severity : enhancement
3225 Description: allow liblustre clients to delegate truncate locking to OST
3226 Details : To avoid overhead of locking, liblustre client instructs OST to
3227 take extent lock in ost_punch() on client's behalf. New connection
3228 flag is added to handle backward compatibility.
3230 Severity : enhancement
3231 Bugzilla : 4928, 7341, 9758
3232 Description: allow number of OST service threads to be specified
3233 Details : a module parameter allows the number of OST service threads
3234 to be specified via "options ost ost_num_threads={N}" in the
3235 OSS's /etc/modules.conf or /etc/modprobe.conf.
3239 Bugzilla : 6146, 9635, 9895
3240 Description: servers crash with bad pointer in target_handle_connect()
3241 Details : In rare cases when a client is reconnecting it was possible that
3242 the connection request was the last reference for that export.
3243 We would temporarily drop the export reference and get a new
3244 one, but this may have been the last reference and the export
3245 was just destroyed. Get new reference before dropping old one.
3247 Severity : enhancement
3248 Frequency : if client is started with failover MDS
3250 Description: Allow multiple MDS hostnames in the mount command
3251 Details : Try to read the configuration from all specified MDS
3252 hostnames during a client mount in case the "primary"
3255 Severity : enhancement
3257 Description: Stop sending data to evicted clients as soon as possible.
3258 Details : Check if the client we are about to send or are sending data to
3259 was evicted already. (Check is done every second of waiting,
3260 for which l_wait_event interface was extended to allow checking
3261 of exit condition at specified intervals).
3264 Frequency : rare, normally only when NFS exporting is done from client
3266 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3268 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3271 Frequency : always, if srand() is called before liblustre initialization
3273 Description: Liblustre uses system PRNG disturbing its usage by user application
3274 Details : Introduce internal to lustre fast and high-quality PRNG for
3275 lustre usage and make liblustre and some other places in generic
3276 lustre code to use it.
3278 Severity : enhancement
3279 Bugzilla : 9477, 9557, 9870
3280 Description: Verify that the MDS configuration logs are updated when xml is
3281 Details : Check if the .xml configuration logs are newer than the config
3282 logs stored on the MDS and report an error if this is the case.
3283 Request --write-conf, or allow starting with --old_conf.
3285 Severity : enhancement
3287 Description: Handle symlinks in the path when checking if Lustre is mounted.
3288 Details : Resolve intermediate symlinks when checking if a client has
3289 mounted a filesystem to avoid duplicate client mounts.
3294 Description: lconf can hit an error exception but still return success.
3295 Details : The lconf command catches the Command error exception at the top
3296 level script context and will exit with the associated exit
3297 status, but doesn't ensure that this exit status is non-zero.
3302 Description: failure of ptlrpc thread startup can cause oops
3303 Details : Starting a ptlrpc service thread can fail if there are a large
3304 number of threads or the server memory is very fragmented.
3305 Handle this without oopsing.
3308 Frequency : always, only if liblustre and non-default acceptor port was used
3310 Description: liblustre cannot connect to servers with non-default acceptor port
3311 Details : tcpnal_set_default_params() was not called and was therefore
3312 ignoring the environment varaible TCPNAL_PORT, as well as other
3313 TCPNAL_ environment variables
3318 Description: two objects could be created on the same OST for a single file
3319 Details : If an OST is down, in some cases it was possible to create two
3320 objects on a single OST for a single file. No problems other
3321 than potential performance impact and spurious error messages.
3325 Bugzilla : 5681, 9562
3326 Description: Client may oops in ll_unhash_aliases
3327 Details : Client dcache may become inconsistent in race condition.
3328 In some cases "getcwd" can fail if the current directory is
3334 Description: Inode refcounting problems in NFS export code
3335 Details : link_raw functions used to call d_instantiate without obtaining
3336 extra inode reference first.
3340 Bugzilla : 9942, 9903
3341 Description: Referencing freed requests leading to crash, memleaks with NFS.
3342 Details : We used to require that call to ll_revalidate_it was always
3343 followed by ll_lookup_it. Also with revalidate_special() it is
3344 possible to call ll_revalidate_it() twice for the same dentry
3345 even if first occurence returned success. This fix changes semantic
3346 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3347 reference on a request referred from the intent.
3348 ll_intent_release() then releases such a request.
3351 Frequency : rare, normally benchmark loads only
3353 Description: unlinked inodes were kept in memory on the client
3354 Details : If a client is repeatedly creating and unlinking files it
3355 can accumulate a lot of stale inodes in the inode slab cache.
3356 If there is no other client load running this can cause the
3357 client node to run out of memory. Instead flush old inodes
3358 from client cache that have the same inode number as a new inode.
3361 Frequency : SLES9 2.6.5 kernel and long filenames only
3362 Bugzilla : 9969, 10379
3363 Description: utime reports stale NFS file handle
3364 Details : SLES9 uses out-of-dentry names in some cases, which confused
3365 the lustre dentry revalidation. Change it to always use the
3369 Frequency : rare, unless heavy write-truncate concurrency is continuous
3370 Bugzilla : 4180, 6984, 7171, 9963, 9331
3371 Description: OST becomes very slow and/or deadlocked during object unlink
3372 Details : filter_destroy() was holding onto the parent directory lock
3373 while truncating+unlinking objects. For very large objects this
3374 may block other threads for a long time and slow overall OST
3375 responsiveness. It may also be possible to get a lock ordering
3376 deadlock in this case, or run out of journal credits because of
3377 the combined truncate+unlink. Solution is to do object truncate
3378 first in one transaction without parent lock, and then do the
3379 final unlink in a new transaction with the parent lock. This
3380 reduces the lock hold time dramatically.
3383 Frequency : rare, 2.4 kernels only
3385 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3386 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3387 before unmounting their filesystem in order to avoid deadlock.
3388 The kernel_locked() function in 2.4 kernels only checks whether
3389 the kernel lock is held, not whether it is this process that is
3390 holding it as 2.6 kernels do.
3395 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3396 Details : The client ptlrpc code may be trying to reconnect to a down
3397 server before a previous connection attempt has timed out.
3398 Increase the reconnect interval to be longer than the connection
3399 timeout interval to avoid sending duplicate connections to
3403 Frequency : echo_client brw_test command
3405 Description: fix echo_client to work with OST preallocated code
3406 Details : OST preallocation code (5137) didn't take echo_client IO path
3407 into account: echo_client calls filter methods outside of any
3408 OST thread and, hence, there is no per-thread preallocated
3409 pages and buffers to use. Solution: hijack pga pages for IO. As
3410 a byproduct, this avoids unnecessary data copying.
3414 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3415 Description: Client can oops in mdc_commit_close() after open replay
3416 Details : It was possible for the MDS to return an open request with no
3417 transaction number in mds_finish_transno() if the client was
3418 evicted, but without actually returning an error. Clients
3419 would later try to replay that open and may trip an assertion
3420 Simplify the client close codepath, and always return an error
3421 from the MDS in case the open is not successful.
3424 Frequency : rare, 2.6 OSTs only
3426 Description: OST may deadlock under high load on fragmented files
3427 Details : If there was a heavy load and highly-fragmented OST filesystems
3428 it was possible to have all the OST threads deadlock waiting on
3429 allocation of biovecs, because the biovecs were not released
3430 until the entire RPC IO was completed. Instead, release biovecs
3431 as soon as they are complete to ensure forward IO progress.
3433 Severity : enhancement
3435 Description: Support for specifying external journal device at mount
3436 Details : If an OST or MDS device is formatted with an external journal
3437 device, this device major/minor is stored in the ext3 superblock
3438 and may not be valid for failover. Allow detecting and
3439 specifying the external journal at mount time.
3444 Description: Mounting an MDS with pending unlinked files may cause oops
3445 Details : target_finish_recovery() calls mds_postrecov() which returned
3446 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3447 considers this an error and immediately begins cleaning up the
3448 lov, just after starting the mds_lov process
3450 Severity : enhancement
3452 Description: Implement 'lfs df' to report actual free space on per-OST basis
3453 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3454 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3455 usage of objects; '-h' to report in human readable format.
3457 ------------------------------------------------------------------------------
3459 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3466 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3467 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3468 at startup during orphan unlinking. Instead, skip buddy update
3469 before setup, as it will be regenerated after recovery is complete.
3474 Description: If an OST is inactive, its locks might reference stale inodes.
3475 Details : lov_change_cbdata() must iterate over all namespaces, even if
3476 they are inactive to clear inode references from the lock.
3478 Severity : enhancement
3479 Frequency : occasional, if non-standard max_dirty_mb used
3481 Description: Client will block write RPCs if not enough grant
3482 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3483 then the client will block writes while waiting for another RPC
3484 to complete instead of consuming its dirty limit. With change
3485 we get improved performance when max_dirty_mb is small.
3487 Severity : enhancement
3488 Bugzilla : 3389, 6253
3489 Description: Add support for supplementary groups on the MDS.
3490 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3491 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3492 lookups for user supplementary groups into a cache.
3496 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3497 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3498 for liblustre O_EXCL is enforced by MDS.
3502 Description: readdir never returns NULL in liblustre.
3503 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3504 dentry in struct dirent.
3508 Frequency : liblustre only, depends on application IO pattern
3509 Description: liblustre clients evicted if not contacting servers
3510 Details : Don't put liblustre clients into the ping_evictor list, so
3511 they will not be evicted by the pinger ever.
3513 Severity : enhancement
3515 Description: Add ability to evict clients by NID from MDS.
3516 Details : By echoing "nid:$NID" string into
3517 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3518 $NID would be instantly evicted from this MDS and from all active
3519 OSTs connected to it.
3523 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3524 Details : lookup_it_finish() used to query file size from OSTs that was not
3529 Description: service threads change working directory to that of init
3530 Details : Starting lustre service threads may pin the working directory
3531 of the parent thread, making that filesystem busy. Threads
3532 now change to the working directory of init to avoid this.
3536 Frequency : during shutdown only
3537 Description: shutdown with a failed MDS or OST can cause unmount to hang
3538 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3543 Frequency : 2.6 only
3544 Description: chmod/chown may include an extra supplementary group
3545 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3546 supplementary group and if none is specified this is used.
3549 Bugzilla : 5479 (6816)
3551 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3552 Details : If lookup is in progress on a file that is unlinked we might try
3553 to revalidate the inode and fail in revalidate after lookup is
3554 complete and ll_file_open() enqueues the open again but
3555 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3558 Frequency : always, if lconf --abort_recovery used
3560 Description: lconf --abort_recovery fails with 'Operation not supported'
3561 Details : lconf was attempting to abort recovery on the MDT device and not
3564 ------------------------------------------------------------------------------
3566 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3571 Frequency : rare (only unsupported configurations with a node running as an
3573 Bugzilla : 6514, 5137
3574 Description: Mounting a Lustre file system on a node running as an OST could
3576 Details : OSTs now preallocates memory needed to write out data at
3577 startup, instead of when needed, to avoid having to
3578 allocate memory in possibly low memory situations.
3579 Specifically, if the file system is mounted on on OST,
3580 memory pressure could force it to try to write out data,
3581 which it needed to allocate memory to do. Due to the low
3582 memory, it would be unable to do so and the node would
3583 become unresponsive.
3585 Severity : enhancement
3587 Description: Addition of lconf --service command line option
3588 Details : lconf now accepts a '--service <arg>' option, which is
3589 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3591 Severity : enhancement
3593 Description: Failover mode is now the default for OSTs.
3594 Details : By default, OSTs will now run in failover mode. To return to
3595 the old behaviour, add '--failout' to the lmc line for OSTs.
3597 Severity : enhancement
3599 Description: Health checks are now provided for MDS and OSTs
3600 Details : Additional detailed health check information on MSD and OSTs
3601 is now provided through the procfs health_check value.
3604 Frequency : occasional, depends on IO load
3606 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3607 after numerous create/delete cycles
3608 Details : The ext3 inode allocation policy would not allocate new inodes
3609 very well on the OSTs because there are no new directories
3610 being created. Instead we look for groups with free space if
3611 the parent directories are nearly full.
3616 Description: Network or server problems during mount may cause partially
3617 mounted clients instead of returning an error.
3618 Details : The config llog parsing code may overwrite the error return
3619 code during mount error handling, returning success instead
3625 Description: MDS can fail to allocate large reply buffers
3626 Details : After long uptimes the MDS can fail to allocate large reply
3627 buffers (e.g. zconf client mount config records) due to memory
3628 fragmentation or consumption by the buffer cache. Preallocate
3629 some large reply buffers so that these replies can be sent even
3630 under memory pressure.
3634 Frequency : rare (liblustre)
3635 Description: fsx running with liblustre complained that using truncate() to
3636 extend the file doesn't work. This patch corrects that issue.
3637 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3638 ATTR_SIZE and lsm use in llu_setattr_raw.
3642 Frequency : rare, only 2.6 kernels
3643 Description: Unusual file access patterns on the MDS may result in inode
3644 data being lost in very rare circumstances.
3645 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3646 code in the 2.6 kernel caused the inode and/or EA data not to
3647 be read from disk, causing single-file corruption.
3651 Frequency : rare, only 2.6 filesystems using extents
3652 Description: Heavy concurrent write and delete load may cause data corruption.
3653 Details : It was possible under high-load situations to have an extent
3654 metadata block in the block device cache from a just-unlinked
3655 file overwrite a newly-allocated data block. We now unmap any
3656 metadata buffers that alias just-allocated data blocks.
3660 Frequency : filesystems with default stripe_count larger than 77
3661 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3662 Details : lconf specifies an inode size of 4096 bytes when the default
3663 stripe_count is larger than 77. This conflicts with the default
3664 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3665 this case to avoid pinning too much memory for large EAs.
3667 ------------------------------------------------------------------------------
3669 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3674 Frequency : rare (extremely heavy IO load with hundreds of clients)
3676 Description: Client is evicted, gets IO error writing to file
3677 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3678 caused clients to be evicted for AST timeouts. The fixes in
3679 bug 5192 mean we no longer need to have such short AST timeouts
3680 so ldlm_timeout has been increased.
3683 Frequency : occasional during --force or --failover shutdown under load
3684 Bugzilla : 5949, 4834
3685 Description: Server oops/LBUG if stopped with --force or --failover under load
3686 Details : a collection of import/export refcount and cleanup ordering
3687 issues fixed for safer force cleanup
3690 Frequency : only filesystems larger than 120 OSTs
3691 Bugzilla : 5990, 6223
3692 Description: lfs getstripe would oops on a very large filesystem
3693 Details : lov_getconfig used kfree on vmalloc'd memory
3696 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3697 Bugzilla : 6242, 6243
3698 Description: reading from files that had been truncated to a non-zero size
3699 but never opened returned no data
3700 Details : ll_file_read() reads zeros from no-object files to EOF
3705 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3706 Details : There's little harm in aborting MDS/OSS recovery and letting it
3707 try again, so I removed the LASSERT and return an error instead.
3709 Severity : enhancement
3711 Description: New debugging infrastructure for tracking down data corruption
3712 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3713 (b) cover more of the client-side code path, and (c) try to narrow
3714 down where problems occurred
3718 Bugzilla : 3819, 4364, 4397, 6313
3719 Description: Racing close and eviction MDS could cause assertion in mds_close
3720 Details : It was possible to get multiple mfd references during close and
3721 client eviction, leading to one thread referencing a freed mfd.
3723 Severity: : enhancement
3724 Bugzilla : 3262, 6359
3725 Description: Attempts to reconnect to servers are now more aggressive.
3726 Details : This builds on the enhanced upcall-less recovery that was added
3727 in 1.4.2. When trying to reconnect to servers, clients will
3728 now try each server in the failover group every 10 seconds. By
3729 default, clients would previously try one server every 25 seconds.
3734 Description: After recovery, certain operations trigger a failed
3735 assertion on a client.
3736 Details : Failing over an mds, using lconf -d --failover, while a
3737 client was doing a readdir() call would cause the client to
3738 LBUG after recovery completed and the readdir() was resent.
3740 Severity : enhancement
3742 Description: Default groups are now added by lconf
3743 Details : You can now run lconf --group <servicename> without having to
3744 manually add groups with lmc.
3747 Frequency : occasional
3749 Description: Nodes with an elan id of 0 trigger a failed assertion
3752 Frequency : always when accessing e.g. tty/console device nodes
3754 Description: tty and some other devices nodes cannot be used on lustre
3755 Details : file's private_data field is used by device data and lustre
3756 values in there got lost. New field was added to struct file to
3757 store fs-specific private data.
3760 Frequency : when exporting Lustre via NFS
3762 Description: NFSD failed occasionally when looking up a path component
3763 Details : NFSD is looking up ".." which was broken in ext3 directories
3764 that had grown large enough to become hashed.
3767 Frequency : Clusters with multiple interfaces not on the same subnet
3769 Description: Nodes will repeatedly try to reconnect to an interface which it
3770 cannot reach and report an error to the log.
3771 Details : Extra peer list entries will be created by lconf with some peers
3772 unreachable. lconf now validates the peer before adding it.
3775 Frequency : Only if a default stripe is set on the filesystem root.
3777 Description: Setting a default stripe on the filesystem root prevented the
3778 filesystem from being remounted.
3779 Details : The client was sending extra request flags in the root getattr
3780 request and did not allocate a reply buffer for the dir EA.
3783 Frequency : occasional, higher if lots of files are accessed by one client
3784 Bugzilla : 6159, 6097
3785 Description: Client trips assertion regarding lsm mismatch/magic
3786 Details : While revalidating inodes the VFS looks up inodes with ifind()
3787 and in rare cases can find an inode that is being freed.
3788 The ll_test_inode() code will free the lsm during ifind()
3789 when it finds an existing inode and then the VFS later attaches
3790 this free lsm to a new inode.
3794 Bugzilla : 6422, 7030
3795 Description: MDS deadlock between mkdir and client eviction
3796 Details : Creating a new file via mkdir or mknod (starting a transaction
3797 and getting the ns lock) can deadlock with client eviction
3798 (gets ns lock and trying to finish a synchronous transaction).
3801 Frequency : occasional
3802 Description: While starting a server, the fsfilt_ext3 module could not be
3804 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3805 kernels. Previously, lconf would still use the ext3 name
3806 when trying to load modules. Now, it will correctly use
3807 ext3 on 2.4 and ldiskfs on 2.6.
3809 Severity : enhancement
3810 Description: The default stripe count has been changed to 1
3811 Details : The interpretation of the default stripe count (0, to lfs
3812 or lmc) has been changed to mean striping across a single
3813 OST, rather than all available. For general usage we have
3814 found a stripe count of 1 or 2 works best.
3816 Severity : enhancement
3817 Description: Add support for compiling against Cray portals.
3818 Details : Conditional compiling for some areas that are different
3822 Frequency : occasional
3823 Bugzilla : 6409, 6834
3824 Description: Creating files with an explicit stripe count may lead to
3825 a failed assertion on the MDS
3826 Details : If some OSTs are full or unavailable, creating files may
3827 trigger a failed assertion on the MDS. Now, Lustre will
3828 try to use other servers or return an error to the
3832 Frequency : occasional
3834 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3835 caused lock timeout during readahead (since 1.4.2).
3836 Details : Processes doing readahead might match a lock that hasn't been
3837 granted yet if there are overlapping and conflicting lock
3838 requests. The readahead process waits on ungranted lock
3839 (original lock is CBPENDING), while OST waits for that process
3840 to cancel CBPENDING read lock and eventually evicts client.
3842 Severity : enhancement
3844 Description: Initial enabling of flock support for clients
3845 Details : Implements fcntl advisory locking and file status functions.
3846 This feature is provided as an optional mount flag (default
3847 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3848 locking are implemented yet, and those that are are not guaranteed
3849 to be completely correct in production environments.
3850 mount -t lustre -o [flock|noflock] ...
3853 Frequency : occasional
3855 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3856 assertion in the ext3 JBD (journaling) layer.
3857 Details : The b_committed_data struct is protected by the big kernel lock
3858 in 2.4 kernels, serializing journal_commit_transaction() and
3859 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3860 access to this struct. In 2.6 kernels there is finer grained
3861 locking to improve SMP performance of the JBD layer.
3865 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3870 Description: merge and cleanup kernel patches.
3873 -----------------------------------------------------------------------------
3875 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3877 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3878 versions of Lustre. Please update all servers and clients to
3879 version 1.4.2 or later at the same time. You must also run
3880 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3881 to update the configuration logs.
3883 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3884 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3885 - swab llog records of type '0' so we get proper header size/idx (5861)
3886 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3887 - fix rename of one directory over another leaking an inode (5953)
3888 - avoid SetPageDirty on 2.6 (5981)
3889 - don't re-add just-being-destroyed locks to the waiting list (5653)
3890 - when creating new directories, inherit the parent's custom
3891 striping settings if present parent (3048)
3892 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3893 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3894 - don't hold client locks on temporary worklist from l_lru (5666)
3895 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3896 - automatically evict dead clients (5921)
3897 - Update file size properly in create+truncate+fstat case (6196)
3898 - Do not unhash mountpoint dentries, do not allow removal of
3900 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3901 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3902 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3903 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3904 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3905 - Correctly handle reads of files with no objects (6243)
3906 - lctl recover will also mark a device active if deactivate used (5933)
3908 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3909 - allow --write-conf on an MDS with different nettype than client (5619)
3910 - don't write config llogs to MDS for mounts not from that MDS (5617)
3911 - lconf should create multiple TCP connections from a client (5201)
3912 - init scripts are now turned off by default; run chkconfig --on
3913 lustre and chkconfig --on lustrefs to use them
3914 - upcalls are no longer needed for clients to recover to failover
3916 - add --abort-recovery option to lconf to abort recovery on device
3918 - add support for an arbitrary number of OSTs (3026)
3919 - Quota support protocol changes.
3920 - forward compatibility changes to wire structs (6007)
3921 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3922 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3923 - improved "lustre status" script
3924 - initialize blocksize for non-regular files (6062)
3925 - added --disable-server and --disable-client configure options (5782)
3926 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3927 - Vanilla 2.4.29 support
3928 - increase maximum number of obd devices to 520 (6242)
3929 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3930 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3931 SLES 9 2.6.5 kernels
3932 - limit stripes per file to 160 (the maximum EA size) (6093)
3934 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3937 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3938 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3939 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3940 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3941 - small ext3 extents cleanups and fixes (5733)
3942 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3943 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3944 - avoid needless client->OST connect, fix handle mismatch (5317)
3945 - fix DLM error path that led to out-of-sync client, long delays (5779)
3946 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3947 - fix several locking issues related to i_size (5492,5624,5654,5672)
3948 - don't move pending lock onto export if it is already evicted (5683)
3949 - fix kernel oops when creating .foo in unlinked directory (5548)
3950 - fix deadlock in obdfilter statistics vs. object create (5811)
3951 - use time_{before,after} to avoid timer jiffies wrap (5882)
3952 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3953 - Do not leak request if server was not able to process it (5154)
3954 - If mds_open unable to find parent dir, make that negative lookup(5154)
3955 - don't create new directories with extent-mapping (5909, 5936)
3957 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3958 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3959 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3960 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3961 - support for RHEL4 kernel on i686 (5773)
3962 - provide error messages when incompatible logs are encountered (5898)
3964 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3965 * version 1.4.0.10 (1.4.1 release candidate 1)
3967 - don't keep a lock reference when lock is not granted (4238)
3968 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3969 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3970 - drop import inflight refcount on signal_completed_replay error (5255)
3971 - unlock page after async write error during send (3677)
3972 - handle missing objects in filter_preprw_read properly (5265)
3973 - no transno return for symlink open, don't save no-trasno open (3440)
3974 - don't try to complete elan receive that already failed (4012)
3975 - free RPC server reply state on error (5406)
3976 - clean up thread from ptlrpc_start_thread() on error (5160)
3977 - readahead could read extra page into cache that wasn't ejected (5388)
3978 - prevent races in class_attach/setup/cleanup/detach (5260)
3979 - don't dereference de->d_inode after l_dput of de (5458)
3980 - use "int" for stripe value returned from lock_to_stripe (5544)
3981 - mballoc allocation and error-checking fixes in 2.6 (5504)
3982 - block device patches to fix I/O request sizes in 2.6 (5482)
3983 - look up hostnames for IB nals (5602)
3984 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3985 - don't start multiple acceptors for the same port (5277)
3986 - fix incorrect LASSERT in mds_getattr_name (5635)
3987 - export a proc file for general "ping" checking (5628)
3988 - fix "lfs check" to not block when the MDS is down (5628)
3990 - service request history (4965)
3991 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3992 - create an "evict_client" /proc entry on OSTs, like the MDS has
3993 - fix mount usage message, return errors per mount(8) (5168)
3994 - change grep [] to grep "[]" in tests so they work in more UMLs
3995 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3996 - remove ancient LOV_MAGIC_V0 EA support (5047)
3997 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3998 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3999 - allow readahead allocations to fail when low on memory (5383)
4000 - mmap locking landed again, after considerable improvement (2828)
4001 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4003 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4006 - send OST transaction number in read/write reply to free req (4966)
4007 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4008 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4009 - the watchdog thread now runs as interruptible (5246)
4010 - flock/lockf fixes (but it's still disabled, pending 5135)
4011 - don't use EXT3 constants in llite code (5094)
4012 - memory shortage at startup could cause assertion (5176)
4014 - reorganization of lov code
4015 - single portals codebase
4017 - add extents/mballoc support (5025)
4018 - direct I/O reads in the obdfilter (4048)
4019 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4021 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4024 - send OST transaction number in read/write reply to free req (4966)
4025 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4026 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4027 - the watchdog thread now runs as interruptible (5246)
4028 - handle missing objects in filter_preprw_read properly (5265)
4029 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4030 - drop import inflight refcount on signal_completed_replay error (5255)
4031 - unlock page after async write error during send (3677)
4032 - return original error code on reconstructed replies (3761)
4033 - no transno return for symlink open, don't save no-trasno open (3440)
4035 - add pid to ldlm debugging output (4922)
4036 - bump the watchdog timeouts -- we can't handle 30sec yet
4037 - extra debugging for orphan dentry/inode bug (5259)
4039 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4042 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4043 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4044 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4045 - fix NULL dereference in /proc/sys/portals/routes (4827)
4046 - allow failed mdc_close() operations to be interrupted (4561)
4047 - stop precreate on OST before MDS would time out on it (4778)
4048 - don't send partial-page writes before EOF from client (4410)
4049 - discard client grant for sub-page writes on large-page clients (4520)
4050 - don't free dentries not owned by NFS code, check generation (4806)
4051 - fix lsm leak if mds_create_objects() fails (4801)
4052 - limit debug_daemon file size, always print CERROR messages (4789)
4053 - use transno after validating reply (3892)
4054 - process timed out requests if import state changes (3754)
4055 - update mtime on OST during writes, return in glimpse (4829)
4056 - add mkfsoptions to LDAP (4679)
4057 - use ->max_readahead method instead of zapping global ra (5039)
4058 - don't interrupt __l_wait_event() during strace
4060 - add software watchdogs to catch hung threads quickly (4941)
4061 - make lustrefs init script start after nfs is mounted
4062 - fix CWARN/ERROR duplication (4930)
4063 - return async write errors to application if possible (2248)
4064 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4065 - print NAL number in %x format (4645)
4066 - update barely-supported suse-2.4.21-171 series (4842)
4067 - support for sles 9 %post scripts
4068 - support for building 2.6 kernel-source packages
4069 - support for sles km_* packages
4071 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4074 - ignore -ENOENT errors in osc_destroy (3639)
4075 - notify osc create thread that OSC is being cleaned up (4600)
4076 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4077 - reconstruct ost_handle() like mds_handle() (4657)
4078 - create a new thread to do import eviction to avoid deadlock (3969)
4079 - let lconf resolve symlinked-to devices (4629)
4080 - don't unlink "objects" from directory with default EA (4554)
4081 - hold socknal file ref over connect in case target is down (4394)
4082 - allow more than 32000 subdirectories in a single directory (3244)
4083 - fix blocks count for O_DIRECT writes (3751)
4084 - OST returns ENOSPC from object create when no space left (4539)
4085 - don't send truncate RPC if file size isn't changing (4410)
4086 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4087 - bind to privileged port in socknal and tcpnal (3689)
4089 - rate limit CERROR/CWARN console message to avoid overload (4519)
4090 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4091 - basic mmap support (3918)
4092 - kernel patch series update from b1_4 (4711)
4094 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4097 - avoid crash during MDS cleanup with OST shut down (2775)
4098 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4099 - don't use bad inodes on the MDS (3744)
4100 - dynamic object preallocation to improve recovery speed (4236)
4101 - don't hold spinlock over lock dumping or change debug flags (4401)
4102 - don't zero obd_dev when it is force cleaned (3651)
4103 - print grants to console if they go negative (4431)
4104 - "lctl deactivate" will stop automatic recovery attempts (3406)
4105 - look for existing locks in ldlm_handle_enqueue() (3764)
4106 - don't resolve lock handle twice in recovery avoiding race (4401)
4107 - revalidate should check working dir is a directory (4134)
4109 - don't always mark "slow" obdfilter messages as errors (4418)
4111 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4114 - don't close LustreDB during write_conf until it is done (3860)
4115 - fix typo in lconf for_each_profile (3821)
4116 - allow dumping logs from multiple threads at one time (3820)
4117 - don't allow multiple threads in OSC recovery (3812)
4118 - fix debug_size parameters (3864)
4119 - fix mds_postrecov to initialize import for llog ctxt (3121)
4120 - replace config semaphore with spinlock (3306)
4121 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4122 - don't allow enqueue to complete on a destroyed export (3822)
4123 - down write_lock before checking llog header bitmap (3825)
4124 - recover from lock replay timeout (3764)
4125 - up llog sem before sending rpc (3652)
4126 - reduce ns lock hold times when setting kms (3267)
4127 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4128 - fix NULL deref and obd_dev leak on setup error (3312)
4129 - replace some LBUG about llog ops with error handling (3841)
4130 - don't match INVALID dentries from d_lookup and spin (3784)
4131 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4132 - fix invalid assertion in ptlrpc_set_wait (3880)
4134 - add libwrap support for the TCP acceptor (3996)
4135 - add /proc/sys/portals/routes for non-root route listing (3994)
4136 - allow setting MDS UUID in .xml (2580)
4137 - print the stack of a process that LBUGs (4228)
4139 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4142 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4143 - make sure to unset replay flag from failed open requests (3440)
4144 - if default stripe count is 0, use OST count for inode size (3636)
4145 - update parent mtime/ctime on client for create/unlink (2611)
4146 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4147 - free recovery state on server during a forced cleanup (3571)
4148 - unregister_reply for resent reqs (3063)
4149 - loop back devices mounting and status check on 2.6 (3563)
4150 - fix resource-creation race that can provoke i_size == 0 (3513)
4151 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4152 - more debugging for page-accounting assertion (3746)
4153 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4154 - avoid deadlock after precreation failure (3758)
4155 - fix race and lock order deadlock in orphan handling (3450, 3750)
4156 - add validity checks when grabbing inodes from l_ast_data (3599)
4158 - add /proc/.../recovery_status to obdfilter (3428)
4159 - lightweight CDEBUG infrastructure, debug daemon (3668)
4160 - change default OSC RPC parameters to be better on small clusters
4161 - turn off OST read cache for files smaller than 32MB
4162 - install man pages and include them in rpms (3100)
4163 - add new init script for (un)mounting lustre filesystems (2593)
4164 - run chkconfig in %post for init scripts (3701)
4165 - drop scimac NAL (unmaintained)
4167 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4170 - clean kiobufs before and after use (3485)
4171 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4172 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4173 - change init script to fail more clearly if not run as root (1528)
4174 - allow clients to reconnect during replay (1742)
4175 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4176 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4177 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4178 - fix cond_resched (3554)
4179 - only evict unfinished clients after recovery (3515)
4180 - allow bulk resend, prevent data loss (3570)
4181 - dynamic ptlrpc request buffer allocation (2102)
4182 - don't allow unlinking open directory if it isn't empty (2904)
4183 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4184 - remove extraneous obd dereference causing LASSERT failure (3334)
4185 - don't use get_cycles() when creating temp. files on the mds (3156)
4186 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4187 - handle EEXIST for set-stripe, set proper directory name (3336)
4189 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4190 - fix ksocknal_fmb_callback() error messages (2918)
4192 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4195 - don't copy lvb into (possibly NULL) reply on error (2983)
4196 - don't deref dentry after dput, don't free lvb on error (2922)
4197 - use the kms to determine writeback rpc length (2947)
4198 - increment oti_logcookies when osc is inactive (2948)
4199 - update client's i_blocks count via lvb messages (2543)
4200 - handle intent open/close of special files properly (1557)
4201 - mount MDS with errors=remount-ro, like obdfilter (2009)
4202 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4203 - don't use cancelling-locks' kms values (2947)
4204 - use highest lock extent for kms, not last one (2925)
4205 - don't dereference ERR_PTR() dentry in error handling path (3107)
4206 - fix thread race in portals_debug_dumplog() (3122)
4207 - create lprocfs device entries at setup instead of at attach (1519)
4208 - common AST error handler, don't evict client on completion race (3145)
4209 - zero nameidata in detach_mnt in 2.6 (3118)
4210 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4211 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4212 - initialize RPC timeout timer earlier for 2.6 (3219)
4213 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4214 - print nal/nid for unknown nid (3258)
4215 - additional checks for oscc recovery before doing precreate (3284)
4216 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4217 - don't crash in mdc_close for bad permissions on open (3285)
4218 - zero i_rdev for non-device files (3147)
4219 - clear page->private before handing to FS, better assertion (3119)
4220 - tune the read pipeline (3236)
4221 - fix incorrect decref of invalidated dentry (2350)
4222 - provide read-ahead stats and refine rpc in flight stats (3328)
4223 - don't hold journal transaction open across create RPC (3313)
4224 - update atime on MDS at close time (3265)
4225 - close LDAP connection when recovering to avoid server load (3315)
4226 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4227 - don't leak open file on MDS after open resend (3325)
4228 - serialize filter_precreate and filter_destroy_precreated (3329)
4229 - loop device shouldn't call sync_dev() for nul device (3092)
4230 - clear page cache after eviction (2766)
4231 - resynchronize MDS->OST in background (2824)
4232 - refuse to mount the same filesystem twice on same mountpoint (3394)
4233 - allow llmount to create routes for mounting behind routers (3320)
4234 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4235 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4236 - fix rare problem with rename on htree directories (3417)
4238 - allow default OST striping configuration per directory (1414)
4239 - fix compilation for qswnal for 2.6 kernels (3125)
4240 - increase maximum number of MDS request buffers for large systems
4241 - change liblustreapi to be useful for external progs like lfsck (3098)
4242 - increase local configuration timeout for slow disks (3353)
4243 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4245 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4248 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4249 - don't overwrite extent policy data in reply if lock was blocked (2901)
4250 - drop filter export grants atomically with removal from device (2663)
4251 - del obd_self_export from work_list in class_disconnect_exports (2908)
4252 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4253 - swab reply message in mdc_close, other PPC fixes (2464)
4254 - fix destroying of named logs (2325)
4255 - overwrite old logs when running lconf --write_conf (2264)
4256 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4257 - fix race in target_handle_connect (2898)
4258 - mds_reint_create() should take same inode create lock (2926)
4259 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4260 - don't close files for self_export to avoid uninitialized obd (2936)
4261 - allow MDS with the same name as client node (2939)
4262 - hold dentry reference for closed log files for unlink (2325)
4263 - reserve space for all logs during transactions (2059)
4264 - don't evict page beyond end of stripe extent (2925)
4265 - don't oops on a deleted current working directory (2399)
4266 - handle hard links to targets without a parent properly (2517)
4267 - don't dereference NULL lock when racing during eviction (2867)
4268 - don't grow lock extents when lots of conflicting locks (2919)
4270 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4273 - account for cache space usage on clients to avoid data loss (974)
4274 - lfsck support in lustre kernel code (2349)
4275 - reduce journal credits needed for BRW writes (2370)
4276 - orphan handling to avoid losing space on client/server crashes
4277 - ptlrpcd can be blocked, stopping ALL progress (2477)
4278 - use lock value blocks to assist in proper KMS, faster stat (1021)
4279 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4280 - recovery for initial connections (2355)
4281 - fixes for mds_cleanup_orphans (1934)
4282 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4283 - block all file creations until orphan recovery completes (1901)
4284 - client remove rq_connection from request struct (2423)
4285 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4286 - recovery timer race (2670)
4287 - mdc_close recovey bug (2532)
4288 - ptlrpc cleanup bug (2710)
4289 - mds timeout on local locks (2588)
4290 - namespace lock held during RPCs (2431)
4291 - handle interrupted sync write properly (2503)
4292 - don't try to handle a message that hasn't been replied to (2699)
4293 - client assert failure during cleanup after abort recovery (2701)
4294 - leak mdc device after failed mount (2712)
4295 - ptlrpc_check_set allows timedout requests to complete (2714)
4296 - wait for inflight reqs when ptlrpcd finishes (2710)
4297 - make sure unregistered services are removed from the srv_list
4298 - reset bulk XID's when resending them (caught by 1138 test)
4299 - unregister_bulk after timeout
4300 - fix lconf error (2694)
4301 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4302 - readahead locks pages, leaves pending causing memory pressure (2673)
4303 - increase OST request buffers to 4096 on large machines (2729)
4304 - fix up permission of existing directories in simple_mkdir (2661)
4305 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4306 - don't assign transno to errored transactions (2742)
4307 - don't delete objects on OST if given a bogus objid from MDS (2751)
4308 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4309 - if rq_no_resend, then timeout request after recovery (2432)
4310 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4311 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4312 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4313 - MDS recovery shouldn't delete objects that it has given out (2730)
4314 - if enqueue arrives after completion, don't clobber LVB (2819)
4315 - don't unlock pages twice when trigger_group_io returns error (2814)
4316 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4317 - don't write pages to disk if there was an error (1450)
4318 - don't ping imports that have recovery disabled (2676)
4319 - take buffered bytes into account when balancing socknal conn (2817)
4320 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4321 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4322 - remove little-endian swabbing of llog records (1987)
4323 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4324 - retry reposting request buffers if they fail (1191)
4325 - grow extent at grant time to avoid granting a revoked lock (2809)
4326 - lock revoke doesn't evict page if covered by a second lock (2765)
4327 - disable VM readahead to avoid reading outside lock extents (2805)
4329 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4330 - updated kernel patches for hp-2.4.20 kernel (2681)
4332 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4335 - fix truncated write corruption (2366)
4336 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4338 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4339 - ensure len(uuid) < 37 in lmc (1171)
4340 - fix ia64 OOPS in llog_test (2255)
4341 - zero end of page at obdfilter for partial page writes (2648)
4342 - don't leave stale dentries around after renames (bug 2428)
4343 - fix timeouts when evicting a client with a single lock held (2642)
4344 - set deadline for the initial HELLO message to drain (2634)
4345 - print out dotted-quad IP addresses in the socknal (2302)
4347 - additional debugging for MDS client eviction problem (2443)
4348 - fix mkfsoptions support for osts (2603, 2604)
4350 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4353 - add series for the vanilla 2.6.0 kernel
4354 - add series for the vanilla 2.4.24 kernel
4355 - add series for a cray x86/64 UL kernel drop
4356 - fix xattr patches for the vanilla 2.4.19 series
4358 - generate true UUIDs in lmc (1171)
4359 - have portals stack dumping break in UML (2466)
4360 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4361 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4362 - wake callback waiting threads on client eviction (2460)
4363 - Add --ptldebug and --subsystem to lmc (1719)
4364 - update assertion to allow safe interrupt allocation
4365 - set rq_no_resend for cancel requests (2432)
4366 - recalculate ptlrpcd timeout after resend (2494)
4367 - call vfs_rmdir when removing pending directories (2368)
4368 - fix renaming a file to itself (2429)
4369 - lmc creates a default one-stripe lov (2454)
4370 - expand procfs space to handle large clusters (2326)
4371 - increase UML stack to avoid overflow
4372 - update lconf's list of debug and subsystem masks
4373 - fix lfs find --obd (2510)
4374 - /proc tunable for disabling filter read caching (2591)
4375 - stop rpm packages from altering slapd.conf (2301)
4376 - disable nagle in the socknal under 0conf (2578)
4377 - choose mds inode size based on stripe count (2572)
4378 - fix kernel-source rpm problems (2516)
4380 - add --disable-doc to avoid pdf generation (2421)
4381 - update documentation, tests, type-os, comments
4382 - avoid format warnings on ia64
4383 - remove the TOE NAL
4384 - tiny code cleanups by removing unused fields
4386 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4389 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4390 - remove the most blatant lies from BUILDING file (2371)
4391 - change default debug level to reasonable production setting
4392 - reduce client side cache size to reduce cache flush time
4393 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4394 - make TCP zerocopy and pinger support enabled by default (2476)
4395 - sync writes completed after process exits caused crashes (2319)
4396 - maintain correct mount count on the MDS (2356)
4397 - backout 1557, because 2316 wasn't really fixed
4398 - better file I/O statistics gathering in /proc
4399 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4400 - another kernel patch to fix zero-copy TCP function export
4401 - don't take duplicate lock when processing re-sent getattr (2420)
4402 - lctl uses obd_self_export instead of creating new conn (2353)
4403 - MDS/OST recovery case which requires object creation asserted (2425)
4404 - move lfs from /usr/sbin to /usr/bin in packages
4405 - fix race between mds_client_add and mds_client_free (2417)
4406 - use kmalloc instead of slabs in portals (2430)
4407 - don't create duplicate records when a failover MDS is present (2442)
4408 - remove unnecessary mount age check (2332)
4409 - don't remove directory inodes from locks prematurely (2451)
4410 - don't break if MDS service name is the same as hostname (2103)
4411 - fix races in client write RPC generation when cache full (2482)
4413 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4416 - remove now-unused request->rq_obd (278)
4417 - if an allocation fails, print out how much memory we've used (1933)
4418 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4419 - add the "configurable stack size" patch to most series files (1256)
4420 - ability to write large log records, for 100+ OST configs (2306)
4421 - fix NULL deref when filter_prep fails (2314)
4422 - fix operator precedence error in filter_sync
4423 - dynamic allocation of socknal TX descriptors (2315)
4424 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4425 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4426 - ensure that CWARN messages in Portals always get to the syslog
4427 - __init/__exit are not for prototype decls (ldlm_init/exit)
4428 - x86-64 compile warning fixes
4429 - fix gateway LMC keyword conflict (2318)
4430 - fix MDS lock inversions in getattr/reint paths (1844)
4431 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4432 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4433 - DLM race condition prevented some lock evictions (2328)
4434 - ENOMEM detection and retry on socknal sends (2230)
4435 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4436 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4437 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4438 - better error messages when a client is rejected during recovery (1505)
4439 - avoid cancelling locks which were never granted, after failure (2330)
4440 - fix i_sem/journal inversion in mds_client_add (2333)
4441 - fix truncate/getattr lock cycle deadlock (2334)
4442 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4443 - fix two rare exit paths which could leak an l_lock() ref (2321)
4444 - fencepost error in MDS/OST orphan recovery (2226)
4445 - make log record alignment 8 bytes (1988)
4446 - lstripe now fails when requested offset > ost_count (2237)
4447 - ensure that all kernel series have a complete list.h (1607)
4448 - fix crashes in special-file operations (2316)
4449 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4451 - allow configurable automake binary, for testing new versions
4452 - small update to the lfs documentation
4454 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4456 * fix negative export reference count in fsfilt_sync (2312)
4458 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4459 * release candidate 0.9.1
4461 - orphans are moved into the PENDING directory for possible recovery
4462 - replayed opens now open by fid for orphan/rename safety (1042)
4463 - last close of an orphan inode generates a transno (683)
4464 - chdir() and mount() now pin the directory entry (1020)
4465 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4466 - discard very old requests without processing them (1502)
4467 - remove some common, well-understood CERRORs (1505)
4468 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4469 - clear "grant" flags in OST replies until OST grant code lands (1644)
4470 - fix read performance by not clobbering i_blksize on client (1598)
4471 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4472 - make LRU size a /proc tunable, clears locks when reduced (707)
4473 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4474 - server locks take references on exports now (1558)
4475 - build fixes for 2.4.20-rh trees (1663)
4476 - return an error from lov_create if all OSCs are inactive (1751)
4477 - fix import levels when a reconnect happens without a timeout (1597)
4478 - exit early from mds_open if we get a lookup error (1749)
4479 - partial page read at EOF wouldn't wait for disk before sending (1642)
4480 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4481 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4482 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4483 - write/truncate lock inversion (1639)
4484 - Don't auto-load obdclass, portals modules during cleanup (1495)
4485 - fix timestamps from jumping to "now" (1763)
4486 - extra journal assertions (1648)
4487 - add an extra multiunlink test (1771)
4488 - fix read_record/write_record API (1776)
4489 - fix leak of offset_extent, possible incorrect i_size later (1772)
4490 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4491 - Debugging for the kqswnal_get_idle_tx problems (1820)
4492 - Allow recovery to be attempted multiple times (1536)
4493 - Write out MDS last_rcvd file after it is first created (1600)
4494 - Fix tx_descriptor leak in failed transmit situations (1827)
4495 - ext3 journaling fixes for assertion failure after IO error (1871)
4496 - class_export_put() on freed export after completion AST error (1896)
4497 - Fix revalidate looping in VFS (1322)
4498 - Don't access a freed export during MDS_REINT timeout (1521)
4499 - Add open-unlink recovery support on the MDS (1673,1764)
4500 - Return an error if no MDS data was read from last_rcvd (1946)
4501 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4502 - Don't setup a disk device that doesn't match exported UUID (317)
4503 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4504 - avoid committing NULL handle in force close
4505 - local.sh is now a one-stripe LOV configuration
4506 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4507 - fix ext3 htree duplicate directory entry corruption (1516)
4508 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4509 - update to new LOV EA format (2097)
4510 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4512 2003-06-15 Phil Schwan <phil@clusterfs.com>
4515 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4516 - per-import/export recovery handling (958, 931, 959)
4517 - multiple last-rcvd slots, for serving multiple FSes (949)
4518 - connections are again shared between multiple imp/exports (963, 964)
4519 - "umount -f" would hang if any requests needed to be sent (393, 978)
4520 - avoid pinning large req buffer by copying for queued messages (989)
4521 - add "uuid" to "lctl device" command to help upcalls (991)
4522 - "open" RPCs with transnos would confuse recovery counters (1037)
4523 - do proper endian conversion of all wire messages (288, 340, 891)
4524 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4525 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4526 - fix ptlrpc_connection leak in target_handle_connect (1174)
4527 - fix import refcounting bug in OST and MDS cleanup (1134)
4528 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4529 - fix very unlikely obd_types race condition (501)
4530 - remove osc_open hack for echo_client (1187)
4531 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4532 - a failure in read_inode2 leads to deadlock (1139)
4533 - cancel ack-locks as soon as transaction is committed (1072)
4534 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4535 - make sure to commitrw after any preprw to avoid deadlock (1162)
4536 - failing to execute a file in a lustre FS would lock inode (1203)
4537 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4538 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4539 - obd_uuid2tgt would walk off the end of the list (1255)
4540 - on IA64 the osc would give portals incorrect bulk size (1258)
4541 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4542 - fix lock inversion caused by new llite matching code (1282)
4543 - limit the number of dirty pages on a client to 10MB (1286)
4544 - timed out locks were not being corrected cancelled (1289)
4545 - fix O_DIRECT above 4GB on IA-32 (1292)
4546 * major user-visible changes
4547 - fail out/fail over policy now controlled by the upcall (993)
4549 - add OBD_PING to check server availability and failure (954)
4550 - lustre messages are now sent in sending host order (288, 340, 891)
4551 - add eadatalen to MDS getattr reply (340)
4552 - OST read replies may contain second buffer, with per-page status (593)
4554 2003-03-11 Phil Schwan <phil@clusterfs.com>
4557 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4558 - failed open()s could cause deadlock; fixed (867, 869)
4559 - stop cancelling OST locks when files are closed (481)
4560 - overlapping XID spaces caused network corruption (851, 853)
4561 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4562 - setattr_raw added, to do single-RPC, server-side setattrs
4563 - lmc/lconf syntax change for OST UUIDs
4564 - fix crashy race condition between ptlrpc_free_req and osc_close
4565 - don't use request in mdc_enqueue if we hit a timeout (889)
4566 - don't set the inode i_size for regular files from the MDS (896)
4567 - handle out of order completion AST (842)
4568 - don't LBUG if a lock request times out after receiving AST (913)
4569 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4570 - if a bad lock AST arrives, send an error instead of dropping entirely
4571 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4572 - fix leak in bulk IO when only partially completed (899, 900, 926)
4573 - fix O_DIRECT for ia64 (55)
4574 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4575 - C-z after timeout could hang a process forever; fixed (977)
4577 - client-side I/O cache (678, 924, 929, 941, 970)
4579 - READPAGE and SETATTRs which don't take server-side locks get
4582 2003-02-11 Phil Schwan <phil@clusterfs.com>
4585 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4586 - Fix fsfilt_extN_readpage() to read a full page of directory
4587 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4588 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4589 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4590 - Most intent operations were removed, in favour of a new RPC mode
4591 that does a single RPC to the server and bypasses most of the VFS
4592 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4593 - Aggressively cancel local locks on DLM servers
4594 - mds_reint_unlink sends EA to the client if it's the last nlink.
4595 client uses that EA to unlink OST objects.
4596 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4597 - recursive symlinks were fixed (439)
4598 - fixed NULL deref in DEBUG_REQ
4599 - filter_update_lastobjid no longer calls sync, which annoyed extN
4600 - fixed multi-client small-writes to a single file problem (445)
4601 - fixed mtime updates during file writes (607)
4602 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4603 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4604 - fixed Linux OST size reporting problem (444, 656)
4605 - OST now updates object mtime with writes or setattr (607, 619)
4606 - client verifies file size before zeroing page past EOF (445)
4607 - OST now writes last allocated objid to disk with allocation (108)
4608 - LOV on echo now works (409)
4610 - mds_reint_unlink sends a new buffer, with the EA included. this
4611 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4612 set if a regular file was being unlinked, and it was the last link
4613 - use PtlGet from the target for bulk writes (315)
4614 - OST now updates object mtime with writes or setattr (607, 619)
4615 - LDLM now has a grant-time callback to revalidate locked items, if
4617 - Many MDS operations were reorganized to combat race conditions
4619 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4620 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4622 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4625 - Fully reactivate OST imports after reconnection (512, others)
4626 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4627 - More graceful error handling for truncating on dead OST (515)
4628 - Don't error out unless we're actually accessing dead stripes (474)
4629 - Fix garbage sizes when stripes are missing (410)
4630 - LRU counters were broken, causing constant lock purge (433, 432)
4631 - garbage on read from stripes with failed OSTs (441)
4632 - mark OSCs as active before reconnecting during recovery (438)
4633 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4634 - lfind did not preserve OST order in output (443)
4635 - symlinks cause hung clients, incorrect data (439)
4636 - stop dereferencing request after dropping refcount (457)
4637 - don't LASSERT(spin_is_locked) on non-SMP (455)
4638 - fixes for many rename() bugs
4639 - fstat didn't correctly synchronize attributes (399)
4640 - server must handle lock cancellation during blocking AST prep (487)
4641 - bulk descriptors were free()d too soon (511)
4642 - fix paths in lconf, which would load incorrect modules (451, 507)
4643 - fix confusing lconf 'host not found' error message (386)
4644 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4645 - fix race condition in mdc_blocking_ast() for inode access (526)
4646 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4647 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4648 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4649 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4650 - fix lstripe in several ways (kernel side) (527)
4651 - fix request leak in ldlm_cli_enqueue (262)
4652 - incorrect OSC was marked inactive after OST failure
4653 - call mds_fs_cleanup before unmounting filesystem (524)
4654 - fix races between taking ns_lock and ldlm_lock_change_resource
4655 - fix races updating LOV export open file list
4656 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4657 - fix recovery NULL deref in ldlm_cli_cancel_unused
4658 - fix some DLM races by using new hash table for lock handles (419)
4659 - permit the client to specify desired inodes, at replay
4660 - duplicate requests when we queue them for replay reintegration
4661 - fix last_rcvd offset calculation
4662 - sync after each recovered transaction, so we always make progress
4663 - never, not always, ERESTART requests without transnos
4664 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4666 - skip replay if the MDS says that the client is already connected
4667 - don't check for a recovery-enabled export to match lctl's UUID
4668 - don't INC_USE_COUNT for phantom exports
4669 - don't crash when cleaning up phantom exports (567)
4670 - don't double-finish or set replay data for errored mdc_open requests
4671 - abort requests when they time out, so we don't get old replies
4672 - send/receive replies for AST messages again
4673 - if the client says that it doesn't have the lock, cancel it on the
4675 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4676 mark it as destroyed, it will all work out when decref is called
4677 - fix module use counts (22, 581)
4679 - ASTs now expect a reply (server cancels lock on error reply)
4681 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4684 - fix many simultaneous client startup (392)
4685 - fix dentry->d_it clobbering
4686 - credentials weren't being shipped for readdir/getattr operations
4687 - remove invalid assertions triggered during some concurrent MD
4689 - proper Lustre versions added (336, 389)
4690 - fix memory leak for create error case (398)
4691 - fix LOV locking bug that would get cli/srv out of sync
4692 - fix echo client over LOV (409)
4693 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4694 - fix double-O_EXCL intent crash (424)
4695 - avoid sending multiple lock CANCELs (352)
4697 - MDS can do multi-client recovery (modulo bugs in new code)
4699 - many updates, edits, cleanups
4701 2002-11-18 Phil Schwan <phil@clusterfs.com>
4704 - fix null d_it dereference (346)
4705 - fix full OST/dbench hang (333)
4706 - fix permission problem with file removal (286)
4707 - fix removal of OSCs from LOV when they fail
4708 - fix NULL deref during bulk timeout (214)
4709 - fix problems related to multiple filesystems on one MDS (241)
4710 - fixed serious subtle metadata locking bugs
4711 - free locks on clients when inodes are removed due to memory
4713 - fix inode pointer in lock data (285)
4714 - partial support for multiple MDS on a single host (241)
4715 - data locks weren't cancelled at clear_inode time (290, 311)
4716 - intent locks could lead to unbounded lock growth (205)
4717 - added a maximum lock count, an LRU list, and a flusher
4718 - fix multiple rename (365)
4719 - properly abstracted the echo client
4720 - OSC locked 1 byte too many; fixed
4721 - rewrote brw callback code:
4722 - fixed recovery bugs related to LOVs (306)
4723 - fixed too-many-pages-in-one-write crash (191)
4724 - fixed (again) crash in sync_io_timeout (214)
4725 - probably fixed callback-related race (385)
4727 - Add capability to MDS protocol
4728 - LDLM cancellations and callbacks on different portals
4730 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4733 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4734 - defer open object destruction to close time (601981, 138)
4735 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4736 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4737 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4738 - don't call obd_setattr in ll_file_release for destroyed objects
4739 * protocol change to lustre_msg: move |version| and add |flags|
4740 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4741 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4742 * added replay of create, unlink, link and rename operations during
4743 MDS failover; recovery should be much more robust now
4744 * remove failed OSCs from LOVs (only lov_create uses this so far)
4745 * the lustre-HOWTO was brought (more) up to date (582544)
4747 2002-10-23 Phil Schwan <phil@clusterfs.com>
4750 - in-use dentries weren't being reused properly (617851)
4751 - prevent multiple LDLM setup (599178)
4752 - fix LOV size calculations for truncate (617853)
4753 - fix client handling of MDS intent errors (POSIX)
4754 - fix permission bug in lovstripe.c test (624321)
4755 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4756 - truncate past end of file could corrupt data
4757 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4758 - a race in recovery could return ETIMEDOUT to apps (623947)
4759 - building outside the source directory was fixed
4760 * the lustre-HOWTO was brought (more) up to date (582544)
4761 * major progress was made on recovery functionality
4763 2002-10-10 Phil Schwan <phil@clusterfs.com>
4766 - recovery deadlock fix
4767 - rm -rf causes LBUG fix (617817)
4768 - file open by multiple tasks fix (618962)
4769 - directory permissions bugs (602707 and 620007)
4770 - journal_stop fixed with locking (611313)
4771 - O_APPEND failures resolved (618273, perhaps 614459)
4772 - lconf PATH fix (619770)
4773 - IA64 build fix (621450)
4774 - RPC buffer sizes scale with amount of memory
4776 2002-10-01 Phil Schwan <phil@clusterfs.com>
4779 - locks would be cancelled without throwing away data pages,
4780 resulting in inconsistent data (605627)
4781 - inode attributes were not always being refreshed (605627, 612449)
4782 - lconf now continues to cleanup after lctl reports an error
4783 - MDS now enforces user permissions (602707)
4784 - lprocfs cleanup fixed, but not yet enabled (614157)
4785 - fixed infinite server hang, should a client not respond to an AST
4786 - avoid going into recovery if user calls readlink() with a buffer
4787 that's too small (613941)
4788 - AST RPCs no longer require replies (614867) -- this may be changed
4789 - don't crash server if client sends an IOV that's too big (611336)
4790 - fixed lock conversion deadlock (611892)
4791 - fixed the following of symlinks (614622)
4792 * recovery: the server can remove locks from a client that dies, other
4793 clients can make progress
4794 * more extN patch fixes
4795 * compile-time configurable ptlrpc buffer allocations
4797 - collaborative read cache document
4798 - Lustre Lite Performance CDR document-in-progress
4800 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4803 - fix typo in patch-2.4.18
4805 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4808 - clear ptlrpc request each time in handle_incoming_request()
4809 - unlink of files now destroys the object on the OST
4811 2002-09-19 Peter Braam <braam@clusterfs.com>
4813 * add hard link support
4814 * change obdfile creation method
4815 * kernel patch changed
4817 2002-09-19 Peter Braam <braam@clusterfs.com>
4820 - stack overflow bug in extN fixed
4822 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4824 * documentation updates
4825 - add man pages for config tools
4826 - update tests/README to describe testing with new config tools
4827 - finish metadata API descriptions
4828 * bug fixes and cleanups
4829 - statfs workaround for 16TB limit
4830 - LOV stripe allocation improved, can stripe on subset of OSTs
4831 - LOV file size/IO offset was wrong for files > 4GB in size
4832 - object EA data was being dropped, caused files to be unreadable
4833 - memory overflow with non-LOV OST caused memory corruption
4834 - fixed regression tests to work with new config tools, obdfilter
4835 - fixed bug when directory size became larger than 1 block
4836 - fixed bug (for single client case) when PWD was deleted
4837 - invalidate local directory pages when doing intent-based ops
4838 - avoid LDLM oops when lock callback contained bad data
4840 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4842 * documentation updates
4843 * bug fixes and cleanups
4844 - configuration tools
4847 - 64-bit compile warnings
4848 - 64-bit internal statfs data
4850 * test_brw on persistent OST devices
4852 * lprocfs (disabled)
4854 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4856 * documentation updates
4857 * bug fixes and cleanups
4858 * configuration tools
4860 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4863 * many small fixes to 0_5_4
4864 * io/network handling
4865 * thinkos in MDS operations
4867 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4870 * crucial basic fixes to 0.5.3
4871 * IOR, Iozone work over Elan
4874 2002-08-07 Phil Schwan <phil@clusterfs.com>
4875 * version 0_5_3, our first alpha
4876 * we use the new Portals iovs
4877 * documentation updates
4878 * bug fixes and cleanups
4879 * small changes in the DLM wire protocol
4881 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4882 * version 0_5_1 with some initial stability,
4883 * locking on MD and file I/O.
4884 * documentation updates
4885 * several bug fixes since 0.5.0
4886 * small changes in wire protocol
4888 2002-07-18 Phil Schwan <phil@clusterfs.com>
4890 * delivered as Lustre Light Alpha
4891 * fixed a crash after handling invalid MDS requests
4892 * fixed directory pages for architectures with non-4k pages sizes
4894 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4895 * release version v0_4_4
4896 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4898 * Fixes a number of interruption problems with OST operations.
4899 * Update documentation for portals header changes
4900 * Move all wire protocol structs/defines to lustre_idl.h
4901 * Fixes symlink length bug.
4902 * Add tcpdump to repository.
4904 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4905 * release version v0_4_3
4906 * Fixes statfs for inodes on extN.
4907 * Fixes bug in runtests which would delete /etc/hosts.
4908 * Use 64-bit object IDs wherever possible (not into VFS though)
4909 Remove ost_get_info, which is unused by lustre, and out of date.
4911 2002-07-03 Peter Braam <braam@clusterfs.com>
4912 * release version v0_4_2 Fixes a lookup error (type not passed)
4913 * move forward to head of Portals
4914 * move forward to latest Lustre kernel
4916 2002-06-25 Peter Braam <braam@clusterfs.com>
4917 * release version v0_4_1. Hopefully stable on single node use.