1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.19.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: LBUG during stress test
18 Details : Need properly lock accesses the flock deadlock detection list.
22 Description: oops in page fault handler
23 Details : kernel page fault handler can return two special 'pages' in error case, don't
24 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
28 Description: timeout with invalidate import.
29 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
30 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
31 ptlrpc_invalidate_import and as result LASSERT.
33 Severity : enhancement
35 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
39 Description: enable MGS and MDT services start separately
40 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
41 a co-located MGS without starting the MGS, which is a complement
42 to 'nosvc' mount option.
46 Frequency : after recovery
47 Description: precreate to many object's after del orphan.
48 Details : del orphan st in oscc last_id == next_id and this triger growing
49 count of precreated objects. Set flag LOW to skip increase count
50 of precreated objects.
54 Frequency : rare, on clear nid stats
55 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
56 Details : when clean nid stats sometimes try destroy live entry,
57 and this produce panic in free.
61 Description: Stack overflow during MDS log replay
62 ease stack pressure by using a thread dealing llog_process.
66 Description: wait until IO finished before start new when do lock cancel.
67 Details : VM protocol want old IO finished before start new, in this case
68 need wait until PG_writeback is cleared until check dirty flag and
69 call writepages in lock cancel callback.
71 Severity : enhancement
73 Description: using special macro for print time and cleanup in includes.
77 Description: mds_mfd_close() ASSERTION(rc == 0)
78 Details : In mds_mfd_close(), we need protect inode's writecount change
79 within its orphan write semaphore to prevent possible races.
83 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
87 Frequency : rare, on shutdown ost
88 Description: don't hit live lock with umount ost.
89 Details : shrink_dcache_parent can be in long loop with destroy dentries,
90 use shrink_dcache_sb instead.
94 Description: don't panic with use echo client
95 Details : echo client pass NULL as client nid pointer and this produce null
100 Description: fix build on ppc32
101 Details : compile code with -m64 flag produce wrong object file for ppc32.
105 Description: add message levels for liblustreapi
109 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
110 Details : Move the 'good_osts' check before the 'total_bavail' check. This
111 will result in an -EAGAIN and in the exit call path we call
112 alloc_rr() which will with increasing aggressiveness attempt to
113 aquire precreated objects on the minimum number of required OSCs.
117 Description: Use old size assignment to avoid deadlock
118 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
119 the scheduling while holding a spinlock. We do not need locking
120 for size in ll_update_inode() because size is only updated from
121 the MDS for directories or files without objects, so there is no
122 other place to do the update, and concurrent access to such inodes
123 are protected by the inode lock.
127 Description: resolve "_IOWR redefined" build error on SLES10
131 Description: dump the memory debugging after all modules are unloaded to
132 suppress false negative in conf_sanity test 39
134 Severity : enhancement
136 Description: build kernel-ib packages for OFED 1.3 in our release cycle
141 Description: fix SLES kernel versioning
142 Details : the kernel version for our SLES 10 kernel did not include a "-"
143 before the "smp" at the end. while this was not a problem in
144 general, it did mean that software trying to use the kernel
145 version to try to detect a vendor specific kernel would fail.
146 this was most evident by the OFED build scripts.
150 Description: Don't update lov_desc members until making sure they are valid
151 Details : When updating lov_desc members via proc fs, need fix their
152 validities before doing the real update.
156 Description: don't put request into delay list while invalidate in flight.
157 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
158 import in flight. this produce timeout for invalidate and sometimes
159 can cause stale data.
163 Frequency : on ppc only
164 Description: not convert ost objects for directory because it's not exist.
165 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
169 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
171 Severity : enhancement
173 Description: Update to support 2.6.22.14 vanilla kernel.
177 Frequency : rare, on recovery
178 Description: read procfs can produce deadlock in some situation
179 Details : Holding lprocfs lock which send rpc can produce block for destroy
180 obd objects and this also block reconnect with -EALREADY. This isn't
181 fix all lprocfs bugs - but make it rare.
183 Severity : enhancement
185 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
188 Frequency : frequent on X2 node
190 Description: mdc_set_open_replay_data LBUG
191 Details : Set replay data for requests that are eligible for replay.
195 Description: lustre_mgs: operation 101 on unconnected MGS
196 Details : When MGC is disconnected from MGS long enough, MGS will evict the
197 MGC, and late on MGC cannot successfully connect to MGS and a lot
198 of the error messages complaining that MGS is not connected.
202 Frequency : on network error
203 Description: panic with double free request if network error
204 Details : mdc_finish_enqueue is finish request if any network error ocuring,
205 but it's true only for synchronus enqueue, for async enqueue
206 (via ptlrpcd) this incorrect and ptlrpcd want finish request
209 Severity : enhancement
211 Description: client-side metadata stat-ahead during readdir(directory readahead)
212 Details : perform client-side metadata stat-ahead when the client detects
213 readdir and sequential stat of dir entries therein
216 Frequency : on start mds
218 Description: Implement get_info(last_id) in obdfilter.
221 Frequency : occasional
223 Description: Correctly check stale fid, not start epoch if ost not support SOM
224 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
225 overwrited with new generated fid, but mds can anwer with one of these
226 two fids and both is not stale. setattr incorectly start epoch and
227 assume will be called done_writeting, but without SOM done_writing
231 Frequency : rare, depends on device drivers and load
233 Description: MDS or OSS nodes crash due to stack overflow
234 Details : Code changes in 1.8.0 increased the stack usage of some functions.
235 In some cases, in conjunction with device drivers that use a lot
236 of stack the MDS (or possibly OSS) service threads could overflow
237 the stack. One change which was identified to consume additional
238 stack has been reworked to avoid the extra stack usage.
241 Frequency : occasional
243 Description: Do not fail import if osc_interpret_create gets -EAGAIN
244 Details : If osc_interpret_create got -EAGAIN it immediately exits and
245 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
246 oscc_has_objects and see OSC has no objests and call
247 oscc_internal_create to resend create request.
249 Severity : enhancement
251 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
253 Severity : enhancement
255 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
258 Frequency : very rare
260 Description: Fix replay if there is an un-replied request and open
261 Details : In some cases, older replay request will revert the
262 mcd->mcd_last_xid on MDS which is used to record the client's
265 Severity : enhancement
267 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
269 Severity : enhancement
271 Description: Add rhel5 support to HEAD.
273 Serverity : enhancement
275 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
280 Description: Don't allow skipping OSTs if index has been specified.
281 Details : Don't allow skipping OSTs if index has been specified, make locking
282 in internal create lots better.
286 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
287 Details : access to bitfield in structure is always rounded to long
288 and this produce problem with not atomic change any bit.
292 Description: Lustre make rpms failed.
293 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
294 builds Lustre rpms from tarball.
297 Frequency : rare on shutdown ost
299 Description: If llog cancel was not send before clean_exports phase, this can
300 produce deadlock in llog code.
301 Details : If llog thread has last reference to obd and call class_import_put
302 this produce deadlock because llog_cleanup_commit_master wait when
303 last llog_commit_thread exited, but this never success because was
304 called from llog_commit_thread.
308 Description: allow userland application know is lost one of stripes.
309 Details : fill lvb_blocks with error code on ost and return it to
310 application if error flag found.
314 Description: NULL lov_tgts causing MDS oops
315 Details : more safe checks for NULL lov_tgts for avoid oops.
317 Severity : enhancement
319 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
323 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
324 Details : If an OST is down the MDS will hang indefinitely in
325 obd_statfs_rqset() waiting for the statfs data. While for
326 MDS QOS usage of statfs, it should not stuck in waiting.
328 Severity : enhancement
330 Description: remote_acl support
331 Details : Support ACL-based permission check for remote user.
332 Support setfacl/getfacl for remote user with the utils
333 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
334 the system "{s,g}etfacl" utils.
336 Severity : enhancement
338 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
340 Severity : enhancement
342 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
346 Description: make mgs_setparam() handle fsname containing dash
347 Details : fsname containing a dash does not work with lctl conf_param
349 Severity : enhancement
351 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
353 Severity : enhancement
355 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
359 Frequency : rare, at shutdown
360 Description: access already free / zero obd_namespace.
361 Details : if client_disconnect_export was called without force flag set,
362 and exist connect request in flight, this can produce access to
363 NULL pointer (or already free pointer) when connect_interpret
364 store ocd flags in obd_namespace.
368 Frequency : only at startup
369 Description: not alloc memory with spinlock held.
370 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
371 if any spinlock held.
373 Severity : enhancement
375 Description: make lustre randomly fail allocating memory
376 Details : Make lustre randomly failed allocating memory for testing purpose.
378 Severity : enhancement
380 Description: lost problems with lov objid file
381 Details : Fixes some scability and access to not inited memory problems
382 in work with lov objdid file.
387 Description: lfs find does not continue on file error
388 Details : Continue other files processing when a file/dir is absent.
392 Description: Inconsistent usage of lustre_pack_reply()
393 Details : Standardize the usage of lustre_pack_reply() such that it
394 always generate a CERROR on failure.
397 Frequency : occasional
399 Description: MDS hang or stay in waiting lock
400 Details : If client receive lock with CBPENDING flag ldlm need send lock
401 cancel as separate rpc, to avoid situation when cancel request
402 can't processed due all i/o threads stay in wait lock.
406 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
410 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
414 Description: Update to SLES9 kernel-2.6.5-7.287.3.
418 Description: Update to RHEL4 latest kernel.
420 Severity : enhancement
422 Description: Build SLES10 patchless client fails
423 Details : The configure was broken by run ./configure with
424 --with-linux-obj=.... argument for patchless client. When the
425 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
426 can't search header adequately. Use absolute path such as
427 -I($LINUX)/include instead.
431 Description: interrupt oig_wait produce painc on resend.
432 Details : brw_redo_request can be used for resend requests from ptlrpcd and
433 private set, and this produce situation when rq_ptlrpcd_data not
434 copyed to new allocated request and triggered LBUG on assert
435 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
436 ptlrpcd set if request is changed and can be safety changed to use
441 Description: LASSERT_{REQ,REP}SWAB macros are buggy
442 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
443 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
444 these macros and replace them with their difinitions instead.
448 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
449 Details : Update which_patch & target file for SLES10 latest kernel.
452 Description: Remove client patches from SLES 10 kernel.
453 Details : This causes SLES 10 clients to behave as patchless clients
454 even on a Lustre-patched (server) kernel.
456 Severity : enhancement
458 Description: self-adjustable client's lru lists
459 Details : use adaptive algorithm for managing client cached locks lru
460 lists according to current server load, other client's work
461 pattern, memory activities, etc. Both, server and client
462 side namespaces provide number of proc tunables for controlling
465 Severity : enhancement
467 Description: light-weight GSS support
468 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
469 and reduce performance overhead.
471 Severity : enhancement
473 Description: Linux keyring support
474 Details : Support using service of Linux keyring for Lustre GSS internal
475 context refresh/cache mechanism.
479 Description: Fix errors in lfs documentation
480 Details : Fixes man pages
484 Description: don't use GFP_* in generic Lustre code.
485 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
489 Description: obdclass is limited by single OBD_ALLOC(idarray)
490 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
494 Description: warnings with build patchless client with vanila 2.6.19 and up
495 Details : change old ctl_table style and replace ctl_table/ctl_table_header
496 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
500 Frequency : Only for SLES
501 Description: sanity_quota fail test_1
502 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
503 file, which makes TSTID[2] inunique.
507 Frequency : Only for SLES9
508 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
509 Details : Update target/ChangeLog/which_patch .
513 Frequency : Only for test_10 in sanity-quota.sh
514 Description: a bug in quota test code
515 Details : checking a test flag in a wrong place
519 Frequency : Only for test_13 in sanity-quota.sh
520 Description: a bug in quota test script
521 Details : shouldn't check overfull allocated quota
525 Description: jbd statistics
526 Details : Port older jbd statistics patch for sles10
528 Severity : enhancement
530 Description: Add printing inode info into message about error in writepage.
535 Description: reply_lock_interpret crash due to race with it and lock cancel.
536 Details : Do not replay locks that are being cancelled. Do not reference
537 locks by their address during replay, just by their handle.
541 Frequency : When flocks are used.
542 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
543 Details : Flock locks might destroy just granted lock if it could be merged
544 with another existing flock, this is done in completion handler,
545 so teach ldlm_cli_enquque_fini that this is a valid case for
551 Description: Oops in read and write path when failing to allocate lock.
552 Details : Check if lock allocation failed and return error back.
556 Description: lstripe command fails for valid OST index
557 Details : The stripe offset is compared to lov->desc.ld_tgt_count
558 instead of lov->desc.ld_active_tgt_count.
562 Description: sanity.sh failed test 103
563 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
568 Description: df doesn't work properly if diskfs blocksize != 4K
569 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
573 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
574 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
578 Description: obd_health_check_timeout too short
579 Details : set obd_health_check_timeout as 1.5x of obd_timeout
583 Description: llapi_file_create() does not allow some changes
584 Details : add llapi_file_open() that allows specifying the mode and
585 open flags, and also returns an open file handle.
589 Description: merge and cleanup kernel patches.
590 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
594 Description: Add journal checksum support.(Kernel part)
595 Details : The journal checksum feature adds two new flags i.e
596 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
597 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
598 indicates that the commit block contains the checksum for
599 the blocks described by the descriptor blocks. Now commit
600 record can be sent to disk without waiting for descriptor
601 blocks to be written to disk. This behavior is controlled
602 using JBD2_FEATURE_ASYNC_COMMIT flag.
606 Description: OSS needs mutliple precreate threads
607 Details : Add ability to start more than one create thread per OSS.
611 Description: Sanity.sh test_65k failed: setstripe should have succeeded
612 Details : execute lfs setstripe on client
616 Description: mds_obd_create error creating tmp object
617 Details : When the user sets quota on root, llog will be affected and can't
618 create files and write files.
621 Frequency : Always on ia64 patchless client, and possibly others.
623 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
624 Details : This allows the patchless client to be loaded on architectures
629 Description: RedHat Update kernel for RHEL5
630 Details : Add the kernel config file for RHEL5.
634 Description: RedHat Update kernel for RHEL5
635 Details : Modify the kernel config file more closer RHEL5.
639 Description: Build failure against Centos5 (RHEL5)
640 Details : Define PAGE_SIZE when it isn't present.
644 Description: test_7 on sanity-quota.sh can't be used on separate machines
645 Details : add facet to handle it
649 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
650 Details : with NFS, the anon dentry's parent was set to itself in
651 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
652 corresponding dentry other than use rec->ur_name.
654 Severity : enhancement
656 Description: lfs setstripe enhancement
657 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
661 Description: enable data checksumming by default
662 Details : enable checksum by default, allow --disable-checksum
663 configure option and "-o nochecksum" mount option. Checksums
664 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
668 Description: lustre support for RHEL5
669 Details : Add support for RHEL5.
673 Description: Client eviction due to failover config
674 Details : after a connection loss, the lustre client should attempt to
675 reconnect to the last active server first before trying the
676 other potential connections.
680 Description: when mds and osts use different quota unit(32bit and 64bit),
681 quota will be released repeatly.
682 Details : void sending multiple quota reqs to mds, which will keep the
683 status between the reqs.
687 Description: osts not allocated evenly to files
688 Details : change the condition to increase offset_idx
691 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
693 Description: Data corruption for OSTs that are formatted larger than 2TB
695 Details : When generating the bio request for lustre file writes the
696 sector number would overflow a temporary variable before being
697 used for the IO. The data reads correctly from Lustre (which
698 will overflow in a similar manner) but other file data or
699 filesystem metadata may be corrupted in some cases.
703 Description: Tune the kernel for good SCSI performance.
704 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
705 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
710 Description: rewrite ext2-derived code in obdclass/uuid.c
711 Details : rewrite inherited code (uuid parsing code from ext2 utils)
712 from scratch preserving functionality.
716 Description: "lfs find -obd UUID" prints directories
717 Details : "lfs find -obd UUID" will return all directory names instead
718 of just file names. It is incorrect because the directories
719 do not reside on the OSTs.
723 Description: use i_size_read and i_size_write in 2.6 port
724 Details : replace inode->i_size access with i_size_read/write()
727 Frequency : when removing large files
729 Description: scheduling issue during removal of large Lustre files
730 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
731 It causes scheduling issues when removing large files (17TB in the
735 Frequency : only with liblustre clients on XT3
737 Description: evictions taking too long
738 Details : allow llrd to evict clients directly on OSTs
741 Frequency : only on ppc
743 Description: /proc/fs/lustre/devices broken on ppc
744 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
745 We should make sure the type of 'index' is loff_t and then cast
746 explicitly as needed below. Do not assign an explicitly cast
750 Frequency : only for rhel5
752 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
753 Details : Modify the target file & which_kernel.
755 Severity : enhancement
757 Description: omit set fsid for export NFS
758 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
759 as FS_REQUIRES_DEV for avoid problems with generate fsid.
763 Frequency : Always, for kernels after 2.6.16
764 Description: Fix warning idr_remove called for id=.. which is not allocated.
765 Details : Last kernels save old s_dev before kill super and not allow
766 to restore from callback - restore it before call kill_anon_super.
770 Description: buffer overruns could theoretically occur
771 Details : llapi_semantic_traverse() modifies the "path" argument by
772 appending values to the end of the origin string, and a
773 overrun may occur. Adding buffer overrun check in liblustreapi.
777 Description: Fix error on 'ls .' at the top of the Lustre mount.
778 Details : Don't revalidate dentry if it is a root dentry.
782 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
783 Details : Modify vm-tunables-rhel4.patch.
787 Description: Kernel config for 2.6.18-vanilla.
788 Details : Modify targets/2.6-vanilla.target.in.
789 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
790 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
791 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
792 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
796 Description: improve handling recoverable errors
797 Details : if request processig with error which can be recoverable on server
798 request should be resend, otherwise page released from cache and
803 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
804 Details : Modify target file & which_patch.
805 A flaw was found in the IA32 system call emulation provided
806 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
807 value could be stored in the %RAX register, which could trigger an
808 out-of-bounds system call table access. An untrusted local user
809 could exploit this flaw to run code in the kernel
810 (ie a root privilege escalation). (CVE-2007-4573).
814 Description: O_DIRECT bypasses client statistics.
815 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
816 read_ahead_stats were updated. copied the stats section from
817 osc_send_oap_rpc() into async_internal().
821 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
825 Description: change order of libsysio includes
826 Details : '#include sysio.h' should always come before '#include xtio.h'
830 Description: handle "serious error: objid * already exists" more gracefully
831 Details : If LAST_ID value on disk is smaller than the objects existing in
832 the O/0/d* directories, it indicates disk corruption and causes an
833 LBUG(). If the object is 0-length, then we should use the existing
834 object. This will help to avoid a full fsck in most cases.
836 Severity : enhancement
838 Description: adapt the lustre_config script to support the upgrade case
839 Details : Add "-u" option for lustre_config script to support upgrading 1.4
840 server targets to 1.6 in parallel.
844 Description: To avoid grant space > avaible space when the disk is almost
845 full. Without this patch you might see the error "grant XXXX >
846 available" or some LBUG about grant, when the disk is almost
848 Details : In filter_check_grant, for non_grant cache write, we should
849 check the left space by if (*left > ungranted + bytes), instead
850 of (*left > ungranted), because only we are sure the left space
851 is enough for another "bytes", then the ungrant space should be
852 increase. In client, we should update cl_avail_grant only there
853 is OBD_MD_FLGRANT in the reply.
857 Description: Update RHEL 4 kernel to fix local root privilege escalation.
858 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
859 described in CVE-2007-4573. This problem could allow untrusted
860 local users to gain root access.
863 Frequency : when using O_DIRECT and quotas
865 Description: Incorrect file ownership on O_DIRECT output files
866 Details : block usage reported by 'lfs quota' does not take into account
867 files that have been written with O_DIRECT.
872 Description: touch file failed when fs is not full
873 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
874 otherwise we can get ENOSP while fs is not full.
878 Description: parallel lock callbacks
879 Details : Instead of sending blocking and completion callbacks as separated
880 requests, adding them to a set and sending in parallel.
883 Frequency : only for Cray XT3
884 Bugzilla : 12829/13455
885 Description: Changing primary group doesn't change the group lustre assigns to
887 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
888 overridden with the primary group provided by the group upcall,
889 whereas the supplied fsgid can be trusted if it is in the list of
890 supplementary groups returned by the group upcall.
892 Severity : enhancement
894 Description: Allow masking D_WARNING, D_ERROR messages from console
895 Details : Console messages can now be disabled via lnet.printk.
899 Description: User code with malformed file open parameter crashes client node
900 Details : Before packing join_file req, all the related reference should be
901 checked carefully in case some malformed flags cause fake
902 join_file req on client.
906 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
907 Details : ldlm_completion_ast() assumes that a lock is granted when the req
908 mode is equal to the granted mode. However, it should also check
909 that LDLM_FL_CP_REQD is not set.
913 Description: Heavy nfs access might result in deadlocks
914 Details : After ELC code landed, it is now improper to enqueue any mds
915 locks under och_sem, because enqueue might want to decide to
916 cancel open locks for same inode we are holding och_sem for.
920 Description: Client eviction while running blogbench
921 Details : A lot of unlink operations with concurrent I/O can lead to a
922 deadlock causing evictions. To address the problem, the number of
923 oustanding OST_DESTROY requests is now throttled to
924 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
925 callbacks are processed in priority.
929 Description: enable ACLs on MDS by default
930 Details : ACLs must be enabled on MDS by default.
933 Frequency : PPC/PPC64 only
935 Description: conflicts between asm-ppc64/types.h and lustre_types.h
936 Details : fix duplicated definitions between asm-ppc64/types.h and
937 lustre_types.h on PPC.
940 Frequency : PPC/PPC64 only
942 Description: asm-ppc/segment.h does not exist
943 Details : fix compile issue on PPC.
947 Description: better handle error messages in extents code
950 Frequency : RHEL4 only
952 Description: mkfs is very slow on IA64/RHEL4
953 Details : A performance regression has been discovered in the MPT Fusion
954 driver between versions 3.02.73rh and 3.02.99.00rh. As a
955 consequence, we have downgraded the MPT Fusion driver in the RHEL4
956 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
958 Severity : enhancement
960 Description: SNMP support enhancement
961 Details : Adding total number of sampled request for an MDS node in snmp
964 Severity : enhancement
966 Description: Optimize ldlm waiting list processing for PR extent locks
967 Details : When processing waiting list for read extent lock and meeting read
968 lock that is same or wider to it that is not contended, skip
969 processing rest of the list and immediatelly return current
970 status of conflictness, since we are guaranteed there are no
971 conflicting locks in the rest of the list.
975 Description: Time out and refuse to reconnect
976 Details : When the failover node is the primary node, it is possible
977 to have two identical connections in imp_conn_list. We must
978 compare not conn's pointers but NIDs, otherwise we can defeat
979 connection throttling.
983 Description: port llog fixes from b1_6 into HEAD
984 Details : Port llog reference couting and some llog cleanups from b1_6
985 (bug 10800) into HEAD, for protect from panic and access to already
986 free llog structures.
990 Description: Detect stride IO mode in read-ahead
991 Details : When a client does stride read, read-ahead should detect that and
992 read-ahead pages according to the detected stride pattern.
996 Description: data checksumming impacts single node performance
997 Details : add support for several checksum algorithm. Currently, only CRC32
998 and Adler-32 are supported. The checksum type can be changed on
999 the fly via /proc/fs/lustre/osc/*/checksum_type.
1003 Description: use adler32 for page checksums
1004 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1009 Description: build for x2 fails
1010 Details : fix compile issue on Cray systems.
1014 Description: Properly match for duplicate locks
1015 Details : Due to different lock order from skiplists code, we need to
1016 traverse entire list for now
1019 Frequency : only on PPC/SLES10
1021 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1022 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1023 be defined. Add a hack in mkfs_lustre.c to work around this compile
1028 Description: LASSERT on MDS when client holding flock lock dies
1029 Details : ldlm pool logic depends on number of granted locks equal to
1030 number of released locks which is not true for flock locks, so
1031 just exclude such locks from consideration.
1035 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1036 Details : Use fsfilt_sync() for both the whole filesystem sync and
1037 individual file sync to eliminate dangerous inode locking
1038 with I_LOCK that can lead to a deadlock.
1042 Description: performance in 1.6.3
1043 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1044 sure that 1MB requests are not fragmented by the block layer.
1046 Severity : enhancement
1048 Description: organize the server-side client stats on per-nid basis
1049 Details : Change the structure of stats under obdfilter and mds to
1057 The "uuid"s file would list the uuids of _active_ exports.
1058 And the clear entry is to clear all stats and stale nids.
1060 Severity : enhancement
1062 Description: eliminate client locks in face of contention
1063 Details : file contention detection and lockless i/o implementation
1064 for contended files.
1068 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1071 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1073 Description: MDS may lose file striping (and hence file data) in some cases
1074 Details : If there are additional extended attributes stored on the MDS,
1075 in particular ACLs, SELinux, or user attributes (if user_xattr
1076 is specified for the client mount options) then there is a risk
1077 of attribute loss. Additionally, the Lustre file striping
1078 needs to be larger than default (e.g. striped over all OSTs),
1079 and an additional attribute must be stored initially in the
1080 inode and then increase in size enough to be moved to the
1081 external attribute block (e.g. ACL growing in size) for file
1086 Description: skiplist implementation simplification
1087 Details : skiplists are used to group compatible locks on granted list
1088 that was implemented as tracking first and last lock of each lock group
1089 the patch changes that to using doubly linked lists
1093 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1094 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1095 because LASSERT on both the data supplied by a client, and the data
1096 on disk is dangerous and incorrect.
1098 Severity : enhancement
1100 Description: Slow trucate/writes to huge files at high offsets.
1101 Details : Directly associate cached pages to lock that protect those pages,
1102 this allows us to quickly find what pages to write and remove
1103 once lock callback is received.
1105 --------------------------------------------------------------------------------
1107 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1109 * Support for kernels:
1110 2.6.5-7.283 (SLES 9),
1111 2.6.9-55.EL (RHEL 4),
1112 2.6.16.46-0.14 (SLES 10),
1113 2.6.18.8 vanilla (kernel.org)
1114 * Client support for unpatched kernels:
1115 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1116 2.6.16 - 2.6.22 vanilla (kernel.org)
1117 * Due to recently discovered recovery problems, we do not recommend
1118 using patchless RHEL 4 clients with this or any earlier release.
1119 * Recommended e2fsprogs version: 1.39.cfs8
1120 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1121 * Starting with this release, the ldiskfs backing filesystem required
1122 by Lustre is now in its own package, lustre-ldiskfs. This package
1123 should be installed. It is versioned separately from Lustre and
1124 may be released separately in future.
1128 Description: block reactivating mgc import until all deactivates complete
1129 Details : Fix race when failing back MDT/MGS to itself (testing)
1132 Frequency : at statup only
1134 Description: mds_lov_synchronize race leads to various problems
1135 Details : simultaneous MDT->OST connections at startup can cause the
1136 sync to abort, leaving the OSC in a bad state.
1138 Severity : enhancement
1140 Description: add optional extra BUILD_VERSION info
1141 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1142 to override the lustre version.
1145 Frequency : 2.6.18 servers only
1147 Description: ll_kern_mount() doesn't release the module reference
1148 Details : The ldiskfs module reference count never drops down to 0
1149 because ll_kern_mount() doesn't release the module reference.
1154 Description: server LBUG when using old ost_num_threads parameter
1155 Details : Accept the old ost_num_threads parameter but warn that it
1156 is deprecated, and fix an off-by-one error that caused an LBUG.
1161 Description: Transient SCSI error results in persistent IO issue
1162 Details : iobuf->dr_error is not reinitialized to 0 between two
1166 Frequency : sometimes when underlying device returns I/O errors
1168 Description: OSTs not going read-only during write failures
1169 Details : OSTs are not remounted read-only when the journal commit threads
1170 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1171 instead of the ext3 wrappers.
1175 Description: poor connect scaling with increasing client count
1176 Details : Don't run filter_grant_sanity_check for more than 100 exports
1177 to improve scaling for large numbers of clients.
1180 Frequency : SLES10 only
1182 Description: sanity-quota.sh quotacheck failed: rc = -22
1183 Details : Quotas cannot be enabled on SLES10.
1186 Frequency : liblustre clients only
1188 Description: getdirentries does not give error when run on compute nodes
1189 Details : getdirentries does not fail when the size specified as an argument
1190 is too small to contain at least one entry
1192 Severity : enhancement
1194 Description: Add LNET router traceability for debug purposes
1195 Details : If a checksum failure occurs with a router as part of the
1196 IO path, the NID of the last router that forwarded the bulk data
1197 is printed so it can be identified.
1202 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1203 Details : Due to a race condition, liblustre clients were occasionally
1204 evicted incorrectly.
1206 Severity : enhancement
1208 Description: lfs setstripe use optional parameters instead of postional
1211 Severity : enhancement
1213 Description: Nanosecond timestamp support for ldiskfs
1214 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1215 resolution timestamps. There is not yet support for this at
1216 the Lustre filesystem level.
1219 Frequency : during server recovery
1221 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1222 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1226 Frequency : nfs export on patchless client
1228 Description: connectathon hang when test nfs export over patchless client
1229 Details : Disconnected dentry cannot be found with lookup, so we do not need
1230 to unhash it or make it invalid
1233 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1234 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1235 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1240 Description: open req refcounting wrong on reconnect
1241 Details : If reconnect happened between getting open reply from server and
1242 call to mdc_set_replay_data in ll_file_open, we will schedule
1243 replay for unreferenced request that we are about to free.
1244 Subsequent close will crash in variety of ways.
1245 Check that request is still eligible for replay in
1246 mdc_set_replay_data().
1251 Description: disable writes to filesystem when reading health_check file
1252 Details : the default for reading the health_check proc file has changed
1253 to NOT do a journal transaction and write to disk, because this
1254 can cause reads of the /proc file to hang and block HA state
1255 checking on a healthy but otherwise heavily loaded system. It
1256 is possible to return to the previous behaviour during configure
1257 with --enable-health-write.
1259 Severity : enhancement
1261 Description: 64-bit inode version
1262 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1263 to the inode. This will be required for version-based recovery.
1268 Description: MDS fails to start if a duplicate client export is detected
1269 Details : in some rare cases it was possible for a client to connect to
1270 an MDS multiple times. Upon recovery the MDS would detect this
1271 and fail during startup. Handle this more gracefully.
1273 Severity : enhancement
1275 Description: Add -o localflock option to simulate old noflock
1277 Details : This will achieve local-only flock/fcntl locks
1283 Description: log_commit_thread vs filter_destroy race leads to crash
1284 Details : Take import reference before releasing llog record semaphore
1289 Description: Wrong request locking in request set processing
1290 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1291 imp_delayed_list, in this place should be used imp_lock.
1294 Frequency : when reconnection
1296 Description: Grant Leak when osc reconnect to OST
1297 Details : When osc reconnect ost, OST(filter) should check whether it
1298 should grant more space to client by comparing fed_grant and
1299 cl_avail_grant, and return the granted space to client instead
1300 of "new granted" space, because client will call osc_init_grant
1301 to update the client grant space info.
1304 Frequency : when client reconnect to OST
1306 Description: Grant Leak when osc do resend and replay bulk write
1307 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1308 bulk write request, because the grant info while be sync between
1309 OSC and OST when reconnect, and we should ignore the grant info
1310 these of resend/replay write req.
1315 Description: Grant space more than avaiable left space sometimes.
1316 Details : When then OST is about to be full, if two bulk writing from
1317 different clients came to OST. Accord the avaliable space of the
1318 OST, the first req should be permitted, and the second one
1319 should be denied by ENOSPC. But if the seconde arrived before
1320 the first one is commited. The OST might wrongly permit second
1321 writing, which will cause grant space > avaiable space.
1324 Frequency : when client is evicted
1326 Description: Grant might be wrongly erased when osc is evicted by OST
1327 Details : when the import is evicted by server, it will fork another
1328 thread ptlrpc_invalidate_import_thread to invalidate the
1329 import, where the grant will be set to 0. While the original
1330 thread will update the grant it got when connecting. So if
1331 the former happened latter, the grant will be wrongly errased
1332 because of this race.
1337 Description: Checking Stale with correct fid
1338 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1339 to check whether it is stale, because sometimes, we want the
1340 enqueue happened anyway, and op_data.fid2 will not be initialized.
1342 Severity : enhancement
1344 Description: update patchless client
1345 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1348 Frequency : only with 2.4 kernel
1350 Description: random memory corruption
1351 Details : size of struct ll_inode_info is to big for union inode.u and this
1352 can be cause of random memory corruption.
1357 Description: Memory leak in recovery
1358 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1359 It should also check obd_fail before fsfilt_start, otherwise if
1360 fsfilt_start return -EROFS,(failover mds during mds recovery).
1361 then the req will return with repmsg->transno = 0 and rc = EROFS.
1362 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1363 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1364 be freed no matter whether fsfilt_commit success or not.
1367 Frequency : only with huge count clients
1369 Description: Prevents from taking the superblock lock in llap_from_page for
1371 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1372 for llap_from_page call in ll_removepage prevents from taking the
1373 superblock lock for a soon died page.
1378 Description: Not check open intent error before release open handle
1379 Details : in some rare cases, the open intent error is not checked before
1380 release open handle, which may cause
1381 ASSERTION(open_req->rq_transno != 0), because it tries to release
1382 the failed open handle.
1387 Description: Set cat log bitmap only after create log success.
1388 Details : in some rare cases, the cat log bitmap is set too early. and it
1389 should be set only after create log success.
1393 Description: Accessing a block bevice can re-enable I/O when Lustre is
1394 tearing down a device.
1395 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1400 Description: service threads may hog cpus when there are a lot of requests
1401 Details : Insert cond_resched to give other threads a chance to use some CPU
1406 Description: the cat log was not initialized in recovery
1407 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1408 unlink log on mds will not be initialized until mds post
1409 recovery. And also in mds post recovery, the unlink log will
1410 initialization will be done asynchronausly, so there will be race
1411 between add unlink log and unlink log initialization.
1415 Description: brw_stats were being printed incorrectly
1416 Details : brw_stats were being printed as log2 but all of them were not
1417 recorded as log2. Also remove some code duplication arising from
1418 filter_tally_{read,write}.
1422 Frequency : rare, only in recovery.
1423 Description: ASSERTION(req->rq_type != LI_POISON) failed
1424 Details : imp_lock should be held while iterating over imp_sending_list for
1425 prevent destroy request after get timeout in ptlrpc_queue_wait.
1429 Description: replay-single.sh test 52 fails
1430 Details : A lock's skiplist need to be cleanup when it being unlinked
1431 from its resource list.
1435 Description: Short directio read returns full requested size rather than
1437 Details : Direct I/O operations should return actual amount of bytes
1438 transferred rather than requested size.
1440 Severity : enhancement
1442 Description: metadata RPC reduction (e.g. for rm performance)
1443 Details : decrease the amount of synchronous RPC between clients and servers
1444 by canceling conflicing lock before the operation on the client
1445 and packing thier handles into the main operation RPC to server.
1447 Severity : enhancement
1449 Description: Async OSC create to avoid the blocking unnecessarily.
1450 Details : If a OST has no remain object, system will block on the creating
1451 when need to create a new object on this OST. Now, ways use
1452 pre-created objects when available, instead of blocking on an
1453 empty osc while others are not empty. If we must block, we block
1454 for the shortest possible period of time.
1456 Severity : enhancement
1458 Description: refine locking for avoid write wrong info into lov_objid file
1459 Details : fix possible races with add new target and write/update data in
1462 --------------------------------------------------------------------------------
1464 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1469 Frequency : on some architectures
1471 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1472 Details : Uninitialized flags sometimes cause configuration commands to
1476 Frequency : patchless clients only
1478 Description: missing __iget() symbol export
1479 Details : The __iget() symbol export is missing. To avoid the need for
1480 this on patchless clients the deathrow inode reaper is turned
1481 off, and we depend on the VM to clean up old inodes. This
1482 dependency was during via the fix for bug 12181.
1484 --------------------------------------------------------------------------------
1486 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1488 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1489 INTEROPERATE with older versions automatically. In many cases a
1490 special upgrade step is needed. Please read the
1491 user documentation before upgrading any part of a live system.
1492 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1493 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1494 * WARNING: Lustre configuration and startup changes are required with
1495 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1501 Severity : enhancement
1503 Description: MountConf
1504 Details : Lustre configuration is now managed via mkfs and mount
1505 commands instead of lmc and lconf. New obd types (MGS, MGC)
1506 are added for dynamic configuration management. See
1507 https://mail.clusterfs.com/wikis/lustre/MountConf for
1510 Severity : enhancement
1512 Description: dynamic OST addition
1513 Details : OSTs can now be added to a live filesystem
1515 Severity : enhancement
1517 Description: startup order invariance
1518 Details : MDTs and OSTs can be started in any order. Clients only
1519 require the MDT to complete startup.
1521 Severity : enhancement
1523 Description: parallel, asynchronous orphan cleanup
1524 Details : orphan cleanup is now performed in separate threads for each
1525 OST, allowing parallel non-blocking operation.
1527 Severity : enhancement
1529 Description: optimized stripe assignment
1530 Details : stripe assignments are now made based on ost space available,
1531 ost previous usage, and OSS previous usage, in order to try
1532 to optimize storage space and networking resources.
1534 Severity : enhancement
1536 Description: Permanently set tunables
1537 Details : All writable /proc/fs/lustre tunables can now be permanently
1538 set on a per-server basis, at mkfs time or on a live system.
1540 Severity : enhancement
1542 Description: Lustre message v2
1543 Details : Add lustre message format v2.
1545 Severity : enhancement
1547 Description: client OST exclusion list
1548 Details : Clients can be started with a list of OSTs that should be
1549 declared "inactive" for known non-responsive OSTs.
1553 Description: SPEC SFS validation failure on NFS v2 over lustre.
1554 Details : Changes the blocksize for regular files to be 2x RPC size,
1555 and not depend on stripe size.
1557 Severity : enhancement
1559 Description: Multiple MD RPCs in flight.
1560 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1561 To avoid overly-overloading MDS, introduce a limit on number of
1562 MDS RPCs in flight for a single client and add /proc controls
1563 to adjust this limit.
1565 Severity : enhancement
1567 Description: client read/write statistics
1568 Details : Add client read/write call usage stats for performance
1569 analysis of user processes.
1570 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1571 file access. extents_stats shows chunk size distribution.
1572 extents_stats_per_process show chunk size distribution per
1575 Severity : enhancement
1577 Description: mds statistics
1578 Details : Add detailed mds operations statistics in
1579 /proc/fs/lustre/mds/*/stats.
1583 Description: Failure of copying files with lustre special EAs.
1584 Details : Client side always return success for setxattr call for lustre
1585 special xattr (currently only "trusted.lov").
1590 Description: Refcount LNET uuids
1591 Details : The global LNET uuid list grew linearly with every startup;
1592 refcount repeated list entries instead of always adding to
1595 Severity : enhancement
1597 Description: Dynamic service threads
1598 Details : Within a small range, start extra service threads
1599 automatically when the request queue builds up.
1602 Frequency : mixed-endian client/server environments
1604 Description: mixed-endian crashes
1605 Details : The new msg_v2 system had some failures in mixed-endian
1608 Severity : enhancement
1610 Description: Easy OST removal
1611 Details : OSTs can be permanently deactivated with e.g. 'lctl
1612 conf_param lustre-OST0001.osc.active=0'
1614 Severity : enhancement
1616 Description: MGS proc entries
1617 Details : Added basic proc entries for the MGS showing what filesystems
1620 Severity : enhancement
1622 Description: provide MGS failover
1623 Details : Added config lock reacquisition after MGS server failover.
1625 Severity : enhancement
1627 Description: add Linux 2.4 support
1628 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1632 Description: a large application tries to do I/O to the same resource and dies
1633 in the middle of it.
1634 Details : Check the req->rq_arrival time after the call to
1635 ost_brw_lock_get(), but before we do anything about
1636 processing it & sending the BULK transfer request. This
1637 should help move old stale pending locks off the queue as
1638 quickly as obd_timeout.
1641 Frequency : when an incorrect nid is specified during startup
1643 Description: ptlrpc connect to non-existant node causes kernel crash
1644 Details : LNET can't be re-entered from an event callback, which
1645 happened when we expire a message after the export has been
1646 cleaned up. Instead, hand the zombie cleanup off to another
1649 Severity : enhancement
1651 Description: plain/inodebits lock performance improvement
1652 Details : Grouping plain/inodebits in granted list by their request modes
1653 and bits policy, thus improving the performance of search through
1657 Frequency : only if OST filesystem is corrupted
1659 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1660 Details : for a short time RPCs with bulk IO are in the replay list,
1661 but replay of bulk IOs is unimplemented. If the OST filesystem
1662 is corrupted due to disk cache incoherency and then replay is
1663 started it is possible to trip an assertion. Avoid putting
1664 committed RPCs into the replay list at all to avoid this issue.
1667 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1669 Description: System hang on startup
1670 Details : This bug allowed the liblustre (e.g. catamount) client to
1671 return to the app before handling all startup RPCs. This
1672 could leave the node unresponsive to lustre network traffic
1673 and manifested as a server ptllnd timeout.
1675 Severity : enhancement
1677 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1678 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1679 boolean to control whether to print peer debug info when a
1680 client's RPC times out.
1683 Frequency : only for kernels with patches from Lustre below 1.4.3
1685 Description: Remove old rdonly API
1686 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1689 Frequency : only for devices with external journals
1691 Description: Set external device read-only also
1692 Details : During a commanded failover stop, we set the disk device
1693 read-only while the server shuts down. We now also set any
1694 external journal device read-only at the same time.
1697 Frequency : when upgrading from 1.4 while trying to change parameters
1699 Description: The wrong (new) MDC name was used when setting parameters for
1700 upgraded MDT's. Also allows changing of OSC (and MDC)
1701 parameters if --writeconf is specified at tunefs upgrade time.
1704 Frequency : when setting specific ost indicies
1706 Description: QOS code breaks on skipped indicies
1707 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1708 created with --index need not be sequential.
1710 Severity : enhancement
1712 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1713 Details : The uninit_groups feature works in conjunction with the kernel
1714 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1715 up the pass1 processing of e2fsck. This is a read-only feature
1716 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1717 mount filesystems that have had this feature enabled.
1719 Severity : enhancement
1721 Description: Improve multi-block allocation algorithm to avoid fragmentation
1722 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1723 allocation locality and avoid filesystem fragmentation.
1725 ------------------------------------------------------------------------------
1727 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1729 * Support for kernels:
1730 2.4.21-47.0.1.EL (RHEL 3)
1731 2.6.5-7.283 (SLES 9)
1732 2.6.9-42.0.10.EL (RHEL 4)
1733 2.6.12.6 vanilla (kernel.org)
1734 2.6.16.27-0.9 (SLES 10)
1735 * Recommended e2fsprogs version: 1.39.cfs5
1737 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1741 Frequency : occasional, depends on client load and configuration
1742 Bugzilla : 12181, 12203
1743 Description: data loss for recently-modified files
1745 Details : In some cases it is possible that recently written or created
1746 files may not be written to disk in a timely manner (this should
1747 normally be within 30s unless client IO load is very high).
1748 The problem appears as zero-length files or files that are a
1749 multiple of 1MB in size after a client crash or client eviction
1750 that are missing data at the end of the file.
1752 This problem is more likely to be hit on clients where files are
1753 repeatedly created and unlinked in the same directory, clients
1754 have a large amount of RAM, have many CPUs, the filesystem has
1755 many OSTs, the clients are rebooted frequently, and/or the files
1756 are not accessed by other nodes after being written.
1758 The presence of the problem can be detected by looking at
1759 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1760 smaller than the second (nr_unused) then dirty files will not
1761 be flushed automatically to disk. "sync; sleep 10" should be
1762 run several times on the node before unmounting it to update
1763 Lustre (this is also safe to run on nodes without this problem).
1765 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1766 that can cause this same problem, so customers using that kernel
1767 also need to update the kernel in addition to Lustre. In order
1768 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1770 It is normal that files written just before a client crash (less
1771 than 30s) may not yet have been flushed to disk, even for local
1775 Frequency : frequent on thin XT3 nodes
1777 Description: UUID collision on thin XT3 Linux nodes
1778 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1779 randomly, since we relied on an insufficiently-seeded PRNG.
1784 Description: OSS hangs after "All ost request buffers busy"
1785 Details : A deadlock between quota and journal operations caused OSS
1786 hangs after printing "All ost request buffers busy."
1789 Frequency : always on liblustre builds
1791 Description: Cleanup compiler warnings on liblustre
1794 Frequency : always on liblustre builds on XT3
1796 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1800 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1801 > 32000 subdirectories
1802 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1803 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1804 subdirectory count crosses 32000. This will aid e2fsck to
1805 correctly handle more than 32000 subdirectories.
1808 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1810 Description: System hang on startup
1811 Details : This bug allowed the liblustre (e.g. catamount) client to
1812 return to the app before handling all startup RPCs. This
1813 could leave the node unresponsive to lustre network traffic
1814 and manifested as a server ptllnd timeout.
1816 Severity : enhancement
1818 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1819 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1820 boolean to control whether to print peer debug info when a
1821 client's RPC times out.
1826 Description: make O_SYNC working on 2.6 kernels
1827 Details : 2.6 kernels use different method for mark pages for write,
1828 so need add a code to lustre for O_SYNC work.
1833 Description: Failure to close file and release space on NFS
1834 Details : Put inode details into lock acquired in ll_intent_file_open.
1835 Use mdc_intent_lock in ll_intent_open to properly
1836 detect all kind of errors unhandled by mdc_enqueue.
1841 Description: proc file read during shutdown sometimes raced obd removal,
1843 Details : Add lock to prevent obd access after proc file removal.
1846 Frequency : Only for files larger than 4GB on 32-bit clients.
1848 Description: improperly doing page alignment of locks
1849 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1850 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1855 Description: RPCs being resent when they shouldn't be
1856 Details : Some RPCs that should not be resent are being resent. This
1857 can cause inconsistencies in the RPC state machine. Do not
1858 resend such requests.
1861 Frequency : rare, only with NFS export
1863 Description: Crash on NFS re-export node
1864 Details : under very unusual load conditions an assertion is hit in
1865 ll_intent_file_open()
1868 Frequency : only if OST filesystem is corrupted
1870 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1871 Details : for a short time RPCs with bulk IO are in the replay list,
1872 but replay of bulk IOs is unimplemented. If the OST filesystem
1873 is corrupted due to disk cache incoherency and then replay is
1874 started it is possible to trip an assertion. Avoid putting
1875 committed RPCs into the replay list at all to avoid this issue.
1880 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1881 Details : Large single O_DIRECT read and write calls can fail to allocate
1882 a sufficiently large buffer to process the request. In case of
1883 allocation failure the allocation is retried with a smaller
1884 buffer and broken into smaller requests.
1886 Severity : enhancement
1888 Description: Add -o localflock option to simulate old noflock behaviour.
1889 Details : This will achieve local-only flock/fcntl locks coherentness.
1894 Description: versioning check is incomplete
1895 Details : Checking the version difference of client vs. server, report
1896 error if the gap is too big.
1901 Description: add support PG_writeback bit
1902 Details : add support for PG_writeback bit for Lustre, for more carefull
1903 work with page cache in 2.6 kernel. This also fix some deadlocks
1904 and remove hack for work O_SYNC with 2.6 kernel.
1906 Severity : enhancement
1908 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1909 Details : The uninit_groups feature works in conjunction with the kernel
1910 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1911 up the pass1 processing of e2fsck. This is a read-only feature
1912 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1913 mount filesystems that have had this feature enabled.
1915 Severity : enhancement
1917 Description: Improve multi-block allocation algorithm to avoid fragmentation
1918 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1919 allocation locality and avoid filesystem fragmentation.
1921 ------------------------------------------------------------------------------
1923 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1925 * Support for kernels:
1926 2.6.9-42.0.3.EL (RHEL 4)
1927 2.6.5-7.276 (SLES 9)
1928 2.4.21-47.0.1.EL (RHEL 3)
1929 2.6.12.6 vanilla (kernel.org)
1930 2.6.16.21-0.8 (SLES10)
1931 * Recommended e2fsprogs version: 1.39.cfs2-0
1933 * The backwards-compatible /proc/sys/portals symlink has been removed
1934 in this release. Before upgrading, please ensure that you change
1935 any configuration scripts or /etc/sysctl.conf files that access
1936 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1937 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1938 in advance of the upgrade on any system running Lustre 1.4.6 or
1939 newer, since /proc/sys/lnet was added in that version.
1940 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1944 Frequency : only when quota is used
1946 Description: avoid scanning export list for quota master
1947 Details : Change the algorithms to avoid scanning export list in order
1948 to improve the efficiency.
1951 Frequency : MDS failover only, very rarely
1953 Description: "went back in time" messages on mds failover
1954 Details : The greatest transno may be lost when the current operation
1955 finishes with an error (transno==0) and the client's last_rcvd
1956 record is over-written. Save the greatest transno in the
1957 mds_last_transno for this case.
1960 Frequency : always for specific kernels and striping counts
1962 Description: client may get "Matching packet too big" without ACL support
1963 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1964 when trying to access files in certain configurations. The
1965 clients should in fact be denied when mounting because they do
1966 not understand ACLs.
1969 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1971 Description: many clients connecting with IO in progress causes connect timeouts
1972 Details : Avoid synchronous journal commits to avoid delays caused by many
1973 clients connecting/disconnecting when bulk IO is in progress.
1974 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1975 OST_IO_PORTAL to avoid delays behind potentially many pending
1979 Frequency : occasionally with multiple writers to a single file
1981 Description: shared writes to file may result in wrong size reported by stat()
1982 Details : Allow growing of kms when extent lock is cancelled
1985 Frequency : always with random mmap IO to multi-striped file
1987 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1988 Details : Only if the hole is at the end of OST object so that kms is too
1989 small. Fix is to increase kms accordingly in ll_nopage.
1992 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1994 Description: writes to a missing object would leak memory on the OST
1995 Details : If there is an inconsistency between the MDS and OST filesystems,
1996 such that the MDS references an object that doesn't exist, writes
1997 to that object will leak memory due to incorrect cleanup in the
1998 error handling path, eventually running out of memory on the OST.
2003 Description: Creating too long symlink causes lustre errors
2004 Details : Check symlink and name lengths before sending requests to MDS.
2007 Frequency : only if flock is enabled (not on by default)
2009 Description: posix locks not released on fd closure on 2.6.9+
2010 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2011 kernels, this caused such locks not to be released on fd close and
2012 then assertions on fs unmount about still used locks.
2015 Frequency : MDS failover only, very rarely
2017 Description: clients may get ASSERTION(granted_lock != NULL)
2018 Details : When request was taking a long time, and a client was resending
2019 a getattr by name lock request. The were multiple lock requests
2020 with the same client lock handle and
2021 mds_getattr_name->fixup_handle_for_resent_request found one of the
2022 lock handles but later failed with ASSERTION(granted_lock != NULL).
2027 Description: handle->h_buffer_credits > 0, assertion failure
2028 Details : h_buffer_credits is zero after truncate, causing assertion
2029 failure. This patch extends the transaction or creates a new
2033 Frequency : NFS re-export or patchless client
2034 Bugzilla : 11179, 10796
2035 Description: Crash on NFS re-export node (__d_move)
2036 Details : We do not want to hash the dentry if we don't have a lock.
2037 But if this dentry is later used in d_move, we'd hit uninitialised
2038 list head d_hash, so we just do this to init d_hash field but
2039 leave dentry unhashed.
2042 Frequency : NFS re-export or patchless client
2044 Description: NFS exports has problem with symbolic link
2045 Details : lustre client didn't properly install dentry when re-exported
2046 to NFS or running patchless client.
2049 Frequency : NFS re-export or patchless client
2051 Description: Various nfs/patchless fixes.
2052 Details : fixes reuse disconected alias for lookup process - this fixes
2053 warning "find_exported_dentry: npd != pd",
2054 fix permission error with open files at nfs.
2055 fix apply umask when do revalidate.
2058 Frequency : occasional
2060 Description: Crash on NFS re-export node
2061 Details : calling clear_page() on the wrong pointer triggered oops in
2062 generic_mapping_read().
2065 Frequency : rarely, using O_DIRECT IO
2067 Description: unaligned directio crashes client with LASSERT
2068 Details : check for unaligned buffers before trying any requests.
2071 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2073 Description: stale data returned from RAID cache
2074 Details : If only a small amount of IO is done to the RAID device before
2075 reading it again it is possible to get stale data from the RAID
2076 cache instead of reading it from disk.
2079 Frequency : always for sles10 kernel
2081 Description: sles10 support
2082 Details : ll_follow_link: compile fixes and using of nd_set_link
2083 under newer kernels.
2086 Frequency : depends on arch, kernel and compiler version, always on sles10
2089 Description: recursive or deep enough symlinks cause stack overflow
2090 Details : getting rid of large stack-allocated variable in
2094 Frequency : depends on hardware
2096 Description: lustre write performance loss in the SLES10 kernel
2097 Details : the performance loss is caused by using of write barriers in the
2098 ext3 code. The SLES10 kernel turns barrier support on by
2099 default. The fix is to undo that change for ldiskfs.
2101 ------------------------------------------------------------------------------
2103 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2105 * Support for kernels:
2106 2.6.9-42.0.3EL (RHEL 4)
2107 2.6.5-7.276 (SLES 9)
2108 2.4.21-47.0.1.EL (RHEL 3)
2109 2.6.12.6 vanilla (kernel.org)
2113 Frequency : quota enabled and large files being deleted
2115 Description: releasing more than 4GB of quota at once hangs OST
2116 Details : If a user deletes more than 4GB of files on a single OST it
2117 will cause the OST to spin in an infinite loop. Release
2118 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2123 Description: statfs data retrieved from /proc may be stale or zero
2124 Details : When reading per-device statfs data from /proc, in the
2125 {kbytes,files}_{total,free,avail} files, it may appear
2126 as zero or be out of date.
2129 Frequency : systems with MD RAID1 external journal devices
2131 Description: lconf's call to blkid is confused by RAID1 journal devices
2132 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2133 of returning all block devices that match the journal UUID.
2136 Frequency : always, for aggregate stripe size over 4GB
2138 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2139 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2140 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2143 Frequency : always if "lfs find" used on a local file/directory
2145 Description: "lfs find" segfaults if used on a local file/directory
2146 Details : The case where a directory component was not specified wasn't
2147 handled correctly. Handle this properly.
2150 Frequency : always on ppc64
2152 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2153 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2154 perform endianness conversion.
2157 Frequency : rarely (truncate to non-zero file size after write under load)
2158 Bugzilla : 10730, 10687
2159 Description: Files padded with zeros to next 4K multiple
2160 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2161 it is possible that files that are truncated to a non-zero size
2162 immediately after being written are filled with zero bytes beyond
2163 the truncated size. No file data is lost.
2165 Severity : enhancement
2167 Description: Allow recovery/failover for liblustre clients.
2168 Details : liblustre clients were unaware of failover configurations until
2171 Severity : enhancement
2173 Description: user file locks should fail when not mounting with flock option
2174 Details : Set up an error-returning stub in ll_file_operations.lock field
2175 to prevent incorrect behaviour when client is mounted without
2176 flock option. Also, set up properly f_op->flock field for
2180 Frequency : always on ia64
2182 Description: "lfs df" loops on printing out MDS statfs information
2183 Details : The obd_ioctl_data was not initialized and in some systems
2184 this caused a failure during the ioctl that did not return
2185 an error. Initialize the struct and return an error on failure.
2188 Frequency : SLES 9 only
2190 Description: Error of copying files with lustre special EAs as root
2191 Details : Client side always return success for setxattr call for lustre
2192 special xattr (currently only "trusted.lov").
2195 Frequency : rarely on clusters with both ia64+i386 clients
2197 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2198 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2199 clients are doing concurrent writes to the same file it is
2200 possible that the ia64 clients may LASSERT because the OST
2201 extent locks are not PAGE_SIZE aligned. Ensure that grown
2202 locks are always aligned on the request boundary.
2205 Frequency : specific use, occasional
2207 Description: Overwriting in use executable truncates on-disk binary image
2208 Details : If one node attempts to overwrite an executable in use by
2209 another node, we now correctly return ETXTBSY instead of
2210 truncating the file.
2212 Severity : enhancement
2214 Description: Async OSC create to avoid the blocking unnecessarily.
2215 Details : If a OST has no remain object, system will block on the creating
2216 when need to create a new object on this OST. Now, ways use
2217 pre-created objects when available, instead of blocking on an
2218 empty osc while others are not empty. If we must block, we block
2219 for the shortest possible period of time.
2224 Description: chmod on Lustre root is propagated to other clients
2225 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2226 invalid by the follow_mount time.
2229 Frequency : liblustre clients only
2231 Description: Race in 'instant cancel' lock handling could lead to such locks
2232 never to be granted in case of SMP MDS
2233 Details : Do not destroy not yet granted but cbpending locks in
2237 Frequency : replay/resend of open
2239 Description: non null lock assetion failure in mds_intent_policy
2240 Details : Trying to replay/resend lockless open requests resulted in
2241 mds_open() returning 0 with no lock. Now it sets a flag if
2242 a lock is going to be returned.
2244 Severity : enhancement
2246 Description: Checksum enhancements
2247 Details : New checksum enhancements allow for resending RPCs that failed
2250 Severity : enhancement
2252 Description: Tunables on number of dirty pages in cacche
2253 Details : Allow to set limit on number of dirty pages cached.
2258 Description: client crash on unmount - lock still has references
2259 Details : In some error handling cases it was possible to leak a lock
2260 reference on a client while accessing a file. This was not
2261 harmful to the client during operation, but would cause the
2262 client to crash when the filesystem is unmounted.
2265 Frequency : specific case, rare
2267 Description: ETXTBSY on mds though file not in use by client
2268 Details : ETXTBSY is no longer incorrectly returned when attempting to
2269 chmod or chown a directory that the user previously tried to
2270 execute or a currently-executing binary.
2273 Frequency : extremely rare except on liblustre-based clients
2275 Description: Lustre space not freed when files are deleted
2276 Details : Clean up open-unlinked files after client eviction. Previously
2277 the unlink was skipped and the files remained as orphans.
2282 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2283 Details : In some cases it was possible to send a blocking callback to
2284 a client doing a glimpse, even though that client didn't get
2285 a lock granted. When the glimpse lock is cancelled on the OST
2286 the freed lock is left on the waiting list and corrupted the list.
2289 Frequency : all core dumps
2291 Description: Broke core dumps to lustre
2292 Details : Negative dentry may be unhashed if parent does not have UPDATE
2293 lock, but some callers, e.g. do_coredump, expect dentry to be
2294 hashed after successful create, hash it in ll_create_it.
2296 ------------------------------------------------------------------------------
2298 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2300 * Support for kernels:
2301 2.6.9-42.0.2.EL (RHEL 4)
2302 2.6.5-7.276 (SLES 9)
2303 2.4.21-40.EL (RHEL 3)
2304 2.6.12.6 vanilla (kernel.org)
2308 Frequency : always on RHEL 3
2310 Description: Number of open files grows over time
2311 Details : The number of open files grows over time, whether or not
2312 Lustre is started. This was due to a filp leak introduced
2313 by one of our kernel patches.
2315 ------------------------------------------------------------------------------
2317 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2319 * Support for kernels:
2320 2.6.9-42.EL (RHEL 4)
2321 2.6.5-7.276 (SLES 9)
2322 2.4.21-40.EL (RHEL 3)
2323 2.6.12.6 vanilla (kernel.org)
2328 Bugzilla : 5719, 9635, 9792, 9684
2329 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2330 Details : If a server is under heavy load and cannot reply to new
2331 connection requests before the client resends the (re)connect,
2332 the connection handling code can behave badly if two service
2333 threads are concurrently handing separate (re)connections from
2334 the same client. Add better locking to the connection handling
2335 code, and ensure that only a single connection will be processed
2336 for a given client UUID, even if the lock is dropped.
2338 Severity : enhancement
2340 Description: add TCP zero-copy support to kernel
2341 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2342 sends if the hardware supports scatter-gather and checksumming.
2343 This allows socklnd to do client-write and server-read more
2344 efficiently and reduce CPU utilization from skbuf copying.
2347 Frequency : only if NFS exporting from client
2349 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2350 Details : If setting attributes on a file created under NFS that had
2351 never been opened it would be possible to oops the client
2352 if the file had no objects.
2355 Frequency : always for liblustre
2357 Description: liblustre client does MDS+OSTs setattr RPC for each write
2358 Details : When doing a write from a liblustre client, the client
2359 incorrectly issued an RPC to the MDS and each OST the file was
2360 striped over in order to update the timestamps. When writing
2361 with small chunks and many clients this could overwhelm the MDS
2362 with RPCs. In all cases it would slow down the write because
2363 these RPCs are unnecessary.
2365 Severity : enhancement
2367 Description: allow number of MDS service threads to be changed at module load
2368 Details : It is now possible to change the number of MDS service threads
2369 running. Adding "options mds mds_num_threads={N}" to the MDS's
2370 /etc/modprobe.conf will set the number of threads for the next
2371 time Lustre is restarted (assuming the "mds" module is also
2372 reloaded at that time). The default number of threads will
2373 stay the same, 32 for most systems.
2378 Description: OST crash if filesystem is unformatted or corrupt
2379 Details : If an OST is started on a device that has never been formatted
2380 or if the filesystem is corrupt and cannot even mount then the
2381 error handling cleanup routines would dereference a NULL pointer.
2386 Description: NULL pointer deref in llap_from_page.
2387 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2388 incorrect) mapping if the page was truncated/reclaimed while it was
2389 searched for. Check for this condition and skip such pages when
2390 doing readahead. Introduce extra check to llap_from_page() to
2391 verify page->mapping->host is non-NULL (so page is not anonymous).
2394 Frequency : Sometimes when using sys_sendfile
2396 Description: "page not covered by a lock" warnings from ll_readpage
2397 Details : sendfile called ll_readpage without right page locks present.
2398 Now we introduced ll_file_sendfile that does necessary locking
2399 around call to generic_file_sendfile() much like we do in
2403 Frequency : with certain MDS communication failures at client mount time
2405 Description: NULL pointer deref after failed client mount
2406 Details : a client connection request may delayed by the network layer
2407 and not be sent until after the PTLRPC layer has timed out the
2408 request. If the client fails the mount immediately it will try
2409 to clean up before the network times out the request. Add a
2410 reference from the request import to the obd device and delay
2411 the cleanup until the network drops the request.
2414 Frequency : occasionally during client (re)connect
2416 Description: assertion failure during client (re)connect
2417 Details : processing a client connection request may be delayed by the
2418 client or server longer than the client connect timeout. This
2419 causes the client to resend the connection request. If the
2420 original connection request is replied in this interval, the
2421 client may trip an assertion failure in ptlrpc_connect_interpret()
2422 which thought it would be the only running connect process.
2425 Frequency : only with obd_echo servers and clients that are rebooted
2427 Description: kernel BUG accessing uninitialized data structure
2428 Details : When running an obd_echo server it did not start the ping_evictor
2429 thread, and when a client was evicted an uninitialized data
2430 structure was accessed. Start the ping_evictor in the RPC
2431 service startup instead of the OBD startup.
2433 Severity : enhancement
2434 Bugzilla : 10193 (patchless)
2435 Description: Remove dependency on various unexported kernel interfaces.
2436 Details : No longer need reparent_to_init, exit_mm, exit_files,
2437 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2440 Frequency : rare (only users of deprecated and unsupported LDAP config)
2442 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2443 Details : LDAP apparently contains 'lustreName' attributes instead of
2444 'name'. A simple remapping of the name is sufficient.
2447 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2449 Description: waiting_locks_callback trips kernel BUG if client is evicted
2450 Details : Running with the dump_on_timeout debug flag turned on makes
2451 it possible that the waiting_locks_callback() can try to dump
2452 the Lustre kernel debug logs from an interrupt handler. Defer
2453 this log dumping to the expired_lock_main() thread.
2455 Severity : enhancement
2457 Description: Support NFS exporting on 2.6 kernels.
2458 Details : Implement non-rawops metadata methods for NFS server to use without
2459 changing NFS server code.
2462 Frequency : very rare (synthetic metadata workload only)
2464 Description: two racing renames might cause an MDS thread to deadlock
2465 Details : Running the "racer" program may cause one MDS thread to rename
2466 a file from being the source of a rename to being the target of
2467 a rename at exactly the same time that another thread is doing
2468 so, and the second thread has already enqueued these locks after
2469 doing a lookup of the target and is trying to relock them in
2470 order. Ensure that we don't try to re-lock the same resource.
2473 Frequency : only very large systems with liblustre clients
2475 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2476 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2477 all OSTs in parallel. This allows the eviction of stale liblustre
2478 clients to proceed much faster than if they were done in series,
2479 and also offers similar improvements for other set_info RPCs.
2484 Description: excessive CPU usage during initial read phase on client
2485 Details : During the initial read phase on a client, it would agressively
2486 retry readahead on the file, consuming too much CPU and impacting
2487 performance (since 1.4.5.8). Improve the readahead algorithm
2488 to avoid this, and also improve some other common cases (read
2489 of small files in particular, where "small" is files smaller than
2490 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2495 Description: MDS crash when receiving packet with unknown intent.
2496 Details : Do not LBUG in unknown intent case, just return -EFAULT
2498 Severity : enhancement
2499 Bugzilla : 9293, 9385
2500 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2501 Details : Do not serialize getattr (non-intent version) and statfs.
2504 Frequency : occasional, when OST network is overloaded/intermittent
2506 Description: client evicted by OST after bulk IO timeout
2507 Details : If a client sends a bulk IO request (read or write) the OST
2508 may evict the client if it is unresposive to its data GET/PUT
2509 request. This is incorrect if the network is overloaded (takes
2510 too long to transfer the RPC data) or dropped the OST GET/PUT
2511 request. There is no need to evict the client at all, since
2512 the pinger and/or lock callbacks will handle this, and the
2513 client can restart the bulk request.
2516 Frequency : Always when mmapping file with no objects
2518 Description: client crashes when mmapping file with no objects
2519 Details : Check that we actually have objects in a file before doing any
2520 operations on objects in ll_vm_open, ll_vm_close and
2526 Description: Request leak when working with deleted CWD
2527 Details : Introduce advanced request refcount tracking for requests
2528 referenced from lustre intent.
2530 Severity : Enhancement
2532 Description: Cache open file handles on client.
2533 Details : MDS now will return special lock along with openhandle, if
2534 requested and client is allowed to hold openhandle, even if unused,
2535 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2536 closing files for every read/write openration.
2538 Severity : Enhancement
2540 Description: Cache open negative dentries on client when possible.
2541 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2542 negative dentries on lock revocation.
2547 Description: Remounting a client read-only wasn't possible with a zconf mount
2548 Details : It wasn't possible to remount a client read-only with llmount.
2550 Severity : enhancement
2551 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2552 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2553 MPICH is included to add Lustre-specific ADIO interfaces.
2554 This is based closely on the UFS ADIO layer and only differs
2555 in file creation, in order to allow the OST striping to be set.
2556 This is user-contributed code and not supported by CFS.
2561 Description: extended inode attributes (immutable, append-only) work improperly
2562 when 2.4 and 2.6 kernels are used on client/server or vice versa
2563 Details : Introduce kernel-independent values for these flags.
2565 Severity : enhancement
2568 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2569 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2570 granularity. Now, Fractional MB granularity is be supported,
2571 this is very useful for low memory system.
2573 Severity : enhancement
2575 Description: Getattr by fid
2576 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2577 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2578 have locks and avoids getattr rpc requests for it on every stat.
2581 Frequency : Always, for filesystems larger than 2TB
2583 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2584 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2585 filesystems larger than 2TB to be mounted with mballoc enabled.
2588 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2590 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2591 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2592 filesystems larger than 2TB is mounted then the kernel will
2593 silently corrupt the start of the filesystem. CONFIG_LBD is
2594 enabled for all CFS-supported kernels, but the possibility of
2595 this happening with a modified kernel config exists.
2597 Severity : enhancement
2599 Description: add client O_DIRECT support for 2.6 kernels
2600 Details : It is now possible to do O_DIRECT reads and writes to files
2601 in the Lustre client mountpoint on 2.6 kernel clients.
2603 Severity : enhancement
2605 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2606 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2607 parallel, not waiting for response from every OST before sending
2608 a rpc to the next OST.
2613 Description: setattr vs write race when updating file timestamps
2614 Details : Client processes that update a file timestamp into the past
2615 right after writing to the file (e.g. tar) it is possible that
2616 the updated file modification time can be reset to the current
2617 time due to a race between processing the setattr and write RPC.
2619 Severity : enhancement
2621 Description: Bring 'lfs find' closer in line with regular Linux find.
2622 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2623 -print0 options and obtains all the needed info through the lustre
2626 Severity : enhancement
2628 Description: support up to 1024 configured devices on one node
2629 Details : change obd_dev array from statically allocated to dynamically
2630 allocated structs as they are first used to reduce memory usage
2635 Description: Flush dirty partially truncated pages during truncate
2636 Details : Immediatelly flush partially truncated pages in filter_setattr,
2637 this way we completely avoid having any pages in page cache on OST
2638 and can retire ugly workarounds during writes to flush such pages.
2643 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2644 Details : protect inode from truncation within vfs_unlink() context
2645 just take a reference before calling vfs_unlink() and release it
2646 when parent's i_sem is free.
2651 Description: last_id value checked outside lock on OST caused LASSERT failure
2652 Details : If there were multiple MDS->OST object precreate requests in
2653 flight, it was possible that the OST's last object id was checked
2654 outside a lock and incorrectly tripped an assertion. Move checks
2655 inside locks, and discard old precreate requests.
2658 Frequency : always, if extents are used on OSTs
2660 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2661 Details : OSTs using the extents format would not zero the high 16 bits of
2662 the index physical block number. This is not a problem for any
2663 OST filesystems smaller than 16TB, and no kernels support ext3
2664 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2665 new/modified files) and can be fixed for existing filesystems
2666 with e2fsprogs-1.39-cfs1.
2671 Description: import connection selection may be incorrect if timer wraps
2672 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2673 import connections to be ignored if the 32-bit jiffies counter
2674 wraps. Use a 64-bit jiffies counter.
2677 Frequency : very large clusters immediately after boot
2679 Description: LNET request buffers exhausted under heavy short-term load
2680 Details : If a large number of client requests are generated on a service
2681 that has previously never seen so many requests it is possible
2682 that the request buffer growth cannot keep up with the spike in
2683 demand. Instead of dropping incoming requests, they are held in
2684 the LND until the RPC service can accept more requests.
2687 Frequency : Sometimes during replay
2689 Description: Assertion failure in ll_local_open after replay.
2690 Details : If replay happened on an open request reply before we were able
2691 to set replay handler, reply will become not swabbed tripping the
2692 assertion in ll_local_open. Now we set the handler right after
2693 recognising of open request
2696 Frequency : very rare
2698 Description: kernel reports "badness in vsnprintf"
2699 Details : Reading from the "recovery_status" /proc file in small chunks
2700 may cause a negative length in lprocfs_obd_rd_recovery_status()
2701 call to vsnprintf() (which is otherwise harmless). Exit early
2702 if there is no more space in the output buffer.
2704 Severity : enhancement
2706 Description: clear OBD RPC statistics by writing to them
2707 Details : It is now possible to clear the OBD RPC statistics by writing
2708 to the "stats" file.
2713 Description: Client mtime is not the same on different clients after utimes
2714 Details : In some cases, the client was using the utimes() syscall on
2715 a file cached on another node. The clients now validate the
2716 ctime from the MDS + OSTs to determine which one is right.
2721 Description: Inability to activate failout mode
2722 Details : lconf script incorrectly assumed that in pythong string's numeric
2723 value is used in comparisons.
2726 Frequency : always with multiple stripes per file
2728 Description: Inefficient object allocation for mutli-stripe files
2729 Details : When selecting which OSTs to stripe files over, for files with
2730 a stripe count that divides evenly into the number of OSTs,
2731 the MDS is always picking the same starting OST for each file.
2732 Return the OST selection heuristic to the original design.
2737 Description: mount failures may take full timeout to return an error
2738 Details : Under some heavy load conditions it is possible that a
2739 failed mount can wait for the full obd_timeout interval,
2740 possibly several minutes, before reporting an error.
2741 Instead return an error as soon as the status is known.
2743 Frequency : quota enabled and large files being deleted
2745 Description: releasing more than 4GB of quota at once hangs OST
2746 Details : If a user deletes more than 4GB of files on a single OST it
2747 will cause the OST to spin in an infinite loop. Release
2748 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2753 Description: statfs data retrieved from /proc may be stale or zero
2754 Details : When reading per-device statfs data from /proc, in the
2755 {kbytes,files}_{total,free,avail} files, it may appear
2756 as zero or be out of date.
2759 Frequency : systems with MD RAID1 external journal devices
2761 Description: lconf's call to blkid is confused by RAID1 journal devices
2762 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2763 of returning all block devices that match the journal UUID.
2766 Frequency : always, for aggregate stripe size over 4GB
2768 Description: assertion fails when trying to use 4GB stripe size
2769 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2770 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2773 Frequency : always on ppc64
2775 Description: the first write on an ext3 filesystem with mballoc got stuck
2776 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2777 perform endianness conversion.
2779 ------------------------------------------------------------------------------
2781 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2783 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2784 INTEROPERATE with older versions automatically. Please read the
2785 user documentation before upgrading any part of a live system.
2786 * WARNING: Lustre networking configuration changes are required with
2787 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2790 * Support for kernels:
2791 2.6.9-22.0.2.EL (RHEL 4)
2792 2.6.5-7.244 (SLES 9)
2793 2.6.12.6 vanilla (kernel.org)
2796 Severity : enhancement
2797 Bugzilla : 7981/8208
2798 Description: Introduced Lustre Networking (LNET)
2799 Details : LNET is new networking infrastructure for Lustre, it includes
2800 a reorganized network configuration mode (see the user
2801 documentation for full details) as well as support for routing
2802 between different network fabrics. Lustre Networking Devices
2803 (LNDs) for the supported network fabrics have also been
2804 created for this new infrastructure.
2806 Severity : enhancement
2807 Description: Introduced Access control lists
2808 Details : clients can set ACLs on files and directories in order to have
2809 more fine-grained permissions than the standard Unix UGO+RWX.
2810 The MDS must be started with the "-o acl" mount option.
2812 Severity : enhancement
2813 Description: Introduced filesystem quotas
2814 Details : Administrators may now establish per-user quotas on the
2817 Severity : enhancement
2819 Description: Configuration change for the XT3
2820 The PTLLND is now used to run Lustre over Portals on the XT3
2821 The configure option(s) --with-cray-portals are no longer used.
2822 Rather --with-portals=<path-to-portals-includes> is used to
2823 enable building on the XT3. In addition to enable XT3 specific
2824 features the option --enable-cray-xt3 must be used.
2829 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2830 Details : A race between allocating a new llcd and re-getting the llcd_lock
2831 allowed another thread to grab newly-allocated llcd.
2833 Severity : enhancement
2835 Description: 2.6 OST async journal commit and locking fix to improve performance
2836 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2837 2.6 kernels are now async as they already were in 2.4 kernels so
2838 that they can commit concurrently with the network bulk transfer.
2839 For block-allocated files the filter allocation semaphore is held
2840 to avoid filesystem fragmentation during allocation. BKL lock
2841 removed for 2.6 xattr operations where it is no longer needed.
2846 Description: lconf incorrectly determined whether two IP networks could talk
2847 Details : In some more complicated routing and multiple-network
2848 configurations, lconf will avoid trying to make a network
2849 connection to a disjoint part of the IP space. It was doing the
2850 math incorrectly for one set of cases.
2855 Description: Fix for potential infinite loop processing records in an llog.
2856 Details : If an llog record is corrupted/zeroed, it is possible to loop
2857 forever in llog_process(). Validate the llog record length
2858 and skip the remainder of the block on error.
2861 Frequency : occasional (liblustre only)
2863 Description: liblustre could not open files whose last component is a symlink
2864 Details : sysio_path_walk() would incorrectly pass the open intent to
2865 intermediate path components.
2868 Frequency : rare (liblustre only with non-standard tuning)
2869 Bugzilla : 7201 (7350)
2870 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2871 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2872 after it has been released, only do this for referenced locks
2874 Severity : enhancement
2876 Description: specifying an (invalid) directory default stripe_size of -1
2877 would reset the directory default striping
2878 Details : stripe_size -1 was used internally to signal directory stripe
2879 removal, now use "all default" to signal dir stripe removal
2880 as a directory striping of "all default" is not useful
2883 Frequency : common for large clusters running liblustre clients
2885 Description: doing an ls when liblustre clients are running is slow
2886 Details : sending a glimpse AST to a liblustre client waits for every AST
2887 to time out, as liblustre clients will not respond. Since they
2888 cannot cache data we refresh the OST lock LVB from disk instead.
2890 Severity : enhancement
2892 Description: doing an ls at the same time as file IO can be slow
2893 Details : enqueue and other "small" requests can be blocked behind many
2894 large IO requests. Create a new OST IO portal for non-IO
2895 requests so they can be processed faster.
2898 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2900 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2901 created file with mode 000
2902 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2903 allow an ftruncate() as a non-root user to a file with mode 000.
2904 Lustre now respects this flag to disable mode checks when
2905 truncating a file owned by the user
2908 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2911 Description: Revoking locks from clients that went dead or catatonic might take
2913 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2914 cancellation of such locks instant on servers without waiting for
2915 any reply from clients. Clients drops these locks when cancel
2916 notification from server is received without replying.
2919 Frequency : liblustre-only, when liblustre client dies or becomes busy
2921 Description: Doing ls on Linux clients can take a long time with active
2923 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2924 granting such locks to it in the first place if possible. Locks
2925 are taken by proxy on the OST during the read or write and
2926 dropped immediately afterward. Add connect flags handling, do
2927 not grant locks to liblustre clients for glimpse ASTs.
2929 Severity : enhancement
2931 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2932 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2933 sequential reads and behaves badly with applications doing random
2934 reads. Improve it by reading ahead at least read region, and
2935 avoiding excessive large RPC for small reads.
2937 Severity : enhancement
2939 Description: Creating more than 1000 files for a single job may cause a load
2940 imbalance on the OSTs if there are also a large number of OSTs.
2941 Details : qos_prep_create() uses an OST index reseed value that is an
2942 even multiple of the number of available OSTs so that if the
2943 reseed happens in the middle of the object allocation it will
2944 still utilize the OSTs as uniformly as possible.
2949 Description: OST or MDS may oops in ping_evictor_main()
2950 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2951 but doesn't restart at beginning of obd_exports_timed list
2954 Severity : enhancement
2956 Description: improve by-nid export eviction on the MDS and OST
2957 Details : allow multiple exports with the same NID to be evicted at one
2958 time without re-searching the exports list.
2961 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2963 Description: MDS may oops in groups_free()
2964 Details : in rare race conditions a newly allocated group_info struct is
2965 freed again, and this can be NULL. The 2.4 compatibility code
2966 for groups_free() checked for a NULL pointer, but 2.6 did not.
2969 Frequency : common for liblustre clients doing little filesystem IO
2970 Bugzilla : 9352, 7313
2971 Description: server may evict liblustre clients accessing contended locks
2972 Details : if a client is granted a lock or receives a completion AST
2973 with a blocking AST already set it would not reply to the AST
2974 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2977 Frequency : lfs setstripe, only systems with more than 160 OSTs
2979 Description: unable to set striping with a starting offset beyond OST 160
2980 Details : llapi_create_file() incorrectly limited the starting stripe
2981 index to the maximum single-file stripe count.
2984 Frequency : LDAP users only
2986 Description: lconf did not handle in-kernel recovery with LDAP properly
2987 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2989 Severity : enhancement
2991 Description: bind OST threads to NUMA nodes to improve performance
2992 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2993 node and do their allocations there to localize memory access
2995 Severity : enhancement
2997 Description: llmount can determine client NID directly from Myrinet (GM)
2998 Details : the client NID code from gmnalnid was moved directly into
2999 llmount, removing the need to use this or specifying the
3000 client NID explicitly when mounting GM clients with zeroconf
3003 Frequency : if client is started with down MDS
3005 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3006 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3007 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3008 ptlrpc_import_delayed_req() because we didn't check if the
3009 request was interrupted, and we also didn't break out of the
3010 event loop if there was no timeout
3015 Description: data loss during non-page-aligned writes to a single file from
3016 both multiple nodes and multiple threads on one node at same time
3017 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3018 inconsistency led to false short-reads, that were cached and later
3019 used by ->prepare_write() to fill in partially written page,
3020 leading to data loss.
3023 Frequency : always, if lconf --abort_recovery used
3025 Description: lconf --abort_recovery fails with 'Operation not supported'
3026 Details : lconf was attempting to abort recovery on the MDT device and not
3029 Severity : enhancement
3031 Description: remove cleanup logs
3032 Details : replace lconf-generated cleanup logs with lustre internal
3033 cleanup routines. Eliminates the need for client-cleanup and
3036 Severity : enhancement
3038 Description: add support for EAs (user and system) on lustre filesystems
3039 Details : it is now possible to store extended attributes in the Lustre
3040 client filesystem, and with the user_xattr mount option it
3041 is possible to allow users to store EAs on their files also
3043 Severity : enhancement
3045 Description: Add possibility (config option) to show minimal available OST free
3047 Details : When compiled with --enable-mindf configure option, statfs(2)
3048 (and so, df) will return least minimal free space available from
3049 all OSTs as amount of free space on FS, instead of summary of
3050 free spaces of all OSTs.
3052 Severity : enhancement
3054 Description: do not expand extent locks acquired on OST-side
3055 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3056 by server: they are not cached anyway.
3059 Frequency : when mmap is used/binaries executed from Lustre
3061 Description: Unmmap pages before throwing them away from read cache.
3062 Details : llap_shrink cache now attempts to unmap pages before discarding
3063 them (if unmapping failed - do not discard). SLES9 kernel has
3064 extra checks that trigger if this unmapping is not done first.
3069 Description: lconf didn't resolve symlinks before checking to see whether a
3070 given mountpoint was already in use
3073 Frequency : when migrating failover services
3074 Bugzilla : 6395, 9514
3075 Description: When migrating a subset of services from a node (e.g. failback
3076 from a failover service node) the remaining services would
3077 time out and evict clients.
3078 Details : lconf --force (implied by --failover) sets the global obd_timeout
3079 to 5 seconds in order to quickly disconnect, but this caused
3080 other RPCs to time out too quickly. Do not change the global
3081 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3083 Severity : enhancement
3084 Frequency : if MDS is started with down OST
3085 Bugzilla : 9439,5706
3086 Description: Allow startup/shutdown of an MDS without depending on the
3087 availability of the OSTs.
3088 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3089 Add appropriate locking and lov-osc refcounts for safe
3090 cleaning. Add osc abort_inflight calls in case the
3091 synchronize never started.
3094 Frequency : occasional (Cray XT3 only)
3096 Description: root not authorized to access files in CRAY_PORTALS environment
3097 Details : The client process capabilities were not honoured on the MDS in
3098 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3099 been accessed by an authorized user then root was able to access
3100 the file on the local client also. The root user capabilities
3101 are now allowed on the MDS, as this environment has secure UID.
3104 Frequency : occasional
3106 Description: ldiskfs "too long searching" message happens too often
3107 Details : A debugging message (otherwise harmless) prints too often on
3108 the OST console. This has been reduced to only happen when
3109 there are fragmentation problems on the filesystem.
3114 Description: Division by zero in statfs when all OSCs are inactive
3115 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3116 lov_statfs divides by value returned by lov_get_stripecnt().
3120 Bugzilla : 9489, 3273
3121 Description: First write from each client to each OST was only 4kB in size,
3122 to initialize client writeback cache, which caused sub-optimal
3123 RPCs and poor layout on disk for the first writen file.
3124 Details : Clients now request an initial cache grant at (re)connect time
3125 and so that they can start streaming writes to the cache right
3126 away and always do full-sized RPCs if there is enough data.
3127 If the OST is rebooted the client also re-establishes its grant
3128 so that client cached writes will be honoured under the grant.
3133 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3134 Details : Now I/O RPCs go to different portal number and (presumably) fast
3135 lock requests (and glimses) and other RPCs get their own service
3136 threads pool that should be able to service those RPCs
3139 Severity : enhancement
3141 Description: Ability to exchange lustre version between client and servers and
3142 issue warnings at client side if client is too old. Also for
3143 liblustre clients there is ability to refuse connection of too old
3145 Details : New 'version' field is added to connect data structure that is
3146 filled with version info. That info is later checked by server and
3150 Frequency : rare, liblustre only.
3151 Bugzilla : 9296, 9581
3152 Description: Two simultaneous writes from liblustre at offset within same page
3153 might proceed at the same time overwriting eachother with stale
3155 Details : I/O lock withing llu_file_prwv was released too early, before data
3156 actually was hitting the wire. Extended lock-holding time until
3157 server acknowledges receiving data.
3160 Frequency : extremely rare. Never observed in practice.
3162 Description: avoid generating lustre_handle cookie of 0.
3163 Details : class_handle_hash() generates handle cookies by incrementing
3164 global counter, and can hit 0 occasionaly (this is unlikely, but
3165 not impossible, because initial value of cookie counter is
3166 selected randonly). Value of 0 is used as a sentinel meaning
3167 "unassigned handle" --- avoid it. Also coalesce two critical
3168 sections in this function into one.
3170 Severity : enhancement
3172 Description: allow liblustre clients to delegate truncate locking to OST
3173 Details : To avoid overhead of locking, liblustre client instructs OST to
3174 take extent lock in ost_punch() on client's behalf. New connection
3175 flag is added to handle backward compatibility.
3177 Severity : enhancement
3178 Bugzilla : 4928, 7341, 9758
3179 Description: allow number of OST service threads to be specified
3180 Details : a module parameter allows the number of OST service threads
3181 to be specified via "options ost ost_num_threads={N}" in the
3182 OSS's /etc/modules.conf or /etc/modprobe.conf.
3186 Bugzilla : 6146, 9635, 9895
3187 Description: servers crash with bad pointer in target_handle_connect()
3188 Details : In rare cases when a client is reconnecting it was possible that
3189 the connection request was the last reference for that export.
3190 We would temporarily drop the export reference and get a new
3191 one, but this may have been the last reference and the export
3192 was just destroyed. Get new reference before dropping old one.
3194 Severity : enhancement
3195 Frequency : if client is started with failover MDS
3197 Description: Allow multiple MDS hostnames in the mount command
3198 Details : Try to read the configuration from all specified MDS
3199 hostnames during a client mount in case the "primary"
3202 Severity : enhancement
3204 Description: Stop sending data to evicted clients as soon as possible.
3205 Details : Check if the client we are about to send or are sending data to
3206 was evicted already. (Check is done every second of waiting,
3207 for which l_wait_event interface was extended to allow checking
3208 of exit condition at specified intervals).
3211 Frequency : rare, normally only when NFS exporting is done from client
3213 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3215 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3218 Frequency : always, if srand() is called before liblustre initialization
3220 Description: Liblustre uses system PRNG disturbing its usage by user application
3221 Details : Introduce internal to lustre fast and high-quality PRNG for
3222 lustre usage and make liblustre and some other places in generic
3223 lustre code to use it.
3225 Severity : enhancement
3226 Bugzilla : 9477, 9557, 9870
3227 Description: Verify that the MDS configuration logs are updated when xml is
3228 Details : Check if the .xml configuration logs are newer than the config
3229 logs stored on the MDS and report an error if this is the case.
3230 Request --write-conf, or allow starting with --old_conf.
3232 Severity : enhancement
3234 Description: Handle symlinks in the path when checking if Lustre is mounted.
3235 Details : Resolve intermediate symlinks when checking if a client has
3236 mounted a filesystem to avoid duplicate client mounts.
3241 Description: lconf can hit an error exception but still return success.
3242 Details : The lconf command catches the Command error exception at the top
3243 level script context and will exit with the associated exit
3244 status, but doesn't ensure that this exit status is non-zero.
3249 Description: failure of ptlrpc thread startup can cause oops
3250 Details : Starting a ptlrpc service thread can fail if there are a large
3251 number of threads or the server memory is very fragmented.
3252 Handle this without oopsing.
3255 Frequency : always, only if liblustre and non-default acceptor port was used
3257 Description: liblustre cannot connect to servers with non-default acceptor port
3258 Details : tcpnal_set_default_params() was not called and was therefore
3259 ignoring the environment varaible TCPNAL_PORT, as well as other
3260 TCPNAL_ environment variables
3265 Description: two objects could be created on the same OST for a single file
3266 Details : If an OST is down, in some cases it was possible to create two
3267 objects on a single OST for a single file. No problems other
3268 than potential performance impact and spurious error messages.
3272 Bugzilla : 5681, 9562
3273 Description: Client may oops in ll_unhash_aliases
3274 Details : Client dcache may become inconsistent in race condition.
3275 In some cases "getcwd" can fail if the current directory is
3281 Description: Inode refcounting problems in NFS export code
3282 Details : link_raw functions used to call d_instantiate without obtaining
3283 extra inode reference first.
3287 Bugzilla : 9942, 9903
3288 Description: Referencing freed requests leading to crash, memleaks with NFS.
3289 Details : We used to require that call to ll_revalidate_it was always
3290 followed by ll_lookup_it. Also with revalidate_special() it is
3291 possible to call ll_revalidate_it() twice for the same dentry
3292 even if first occurence returned success. This fix changes semantic
3293 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3294 reference on a request referred from the intent.
3295 ll_intent_release() then releases such a request.
3298 Frequency : rare, normally benchmark loads only
3300 Description: unlinked inodes were kept in memory on the client
3301 Details : If a client is repeatedly creating and unlinking files it
3302 can accumulate a lot of stale inodes in the inode slab cache.
3303 If there is no other client load running this can cause the
3304 client node to run out of memory. Instead flush old inodes
3305 from client cache that have the same inode number as a new inode.
3308 Frequency : SLES9 2.6.5 kernel and long filenames only
3309 Bugzilla : 9969, 10379
3310 Description: utime reports stale NFS file handle
3311 Details : SLES9 uses out-of-dentry names in some cases, which confused
3312 the lustre dentry revalidation. Change it to always use the
3316 Frequency : rare, unless heavy write-truncate concurrency is continuous
3317 Bugzilla : 4180, 6984, 7171, 9963, 9331
3318 Description: OST becomes very slow and/or deadlocked during object unlink
3319 Details : filter_destroy() was holding onto the parent directory lock
3320 while truncating+unlinking objects. For very large objects this
3321 may block other threads for a long time and slow overall OST
3322 responsiveness. It may also be possible to get a lock ordering
3323 deadlock in this case, or run out of journal credits because of
3324 the combined truncate+unlink. Solution is to do object truncate
3325 first in one transaction without parent lock, and then do the
3326 final unlink in a new transaction with the parent lock. This
3327 reduces the lock hold time dramatically.
3330 Frequency : rare, 2.4 kernels only
3332 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3333 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3334 before unmounting their filesystem in order to avoid deadlock.
3335 The kernel_locked() function in 2.4 kernels only checks whether
3336 the kernel lock is held, not whether it is this process that is
3337 holding it as 2.6 kernels do.
3342 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3343 Details : The client ptlrpc code may be trying to reconnect to a down
3344 server before a previous connection attempt has timed out.
3345 Increase the reconnect interval to be longer than the connection
3346 timeout interval to avoid sending duplicate connections to
3350 Frequency : echo_client brw_test command
3352 Description: fix echo_client to work with OST preallocated code
3353 Details : OST preallocation code (5137) didn't take echo_client IO path
3354 into account: echo_client calls filter methods outside of any
3355 OST thread and, hence, there is no per-thread preallocated
3356 pages and buffers to use. Solution: hijack pga pages for IO. As
3357 a byproduct, this avoids unnecessary data copying.
3361 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3362 Description: Client can oops in mdc_commit_close() after open replay
3363 Details : It was possible for the MDS to return an open request with no
3364 transaction number in mds_finish_transno() if the client was
3365 evicted, but without actually returning an error. Clients
3366 would later try to replay that open and may trip an assertion
3367 Simplify the client close codepath, and always return an error
3368 from the MDS in case the open is not successful.
3371 Frequency : rare, 2.6 OSTs only
3373 Description: OST may deadlock under high load on fragmented files
3374 Details : If there was a heavy load and highly-fragmented OST filesystems
3375 it was possible to have all the OST threads deadlock waiting on
3376 allocation of biovecs, because the biovecs were not released
3377 until the entire RPC IO was completed. Instead, release biovecs
3378 as soon as they are complete to ensure forward IO progress.
3380 Severity : enhancement
3382 Description: Support for specifying external journal device at mount
3383 Details : If an OST or MDS device is formatted with an external journal
3384 device, this device major/minor is stored in the ext3 superblock
3385 and may not be valid for failover. Allow detecting and
3386 specifying the external journal at mount time.
3391 Description: Mounting an MDS with pending unlinked files may cause oops
3392 Details : target_finish_recovery() calls mds_postrecov() which returned
3393 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3394 considers this an error and immediately begins cleaning up the
3395 lov, just after starting the mds_lov process
3397 Severity : enhancement
3399 Description: Implement 'lfs df' to report actual free space on per-OST basis
3400 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3401 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3402 usage of objects; '-h' to report in human readable format.
3404 ------------------------------------------------------------------------------
3406 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3413 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3414 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3415 at startup during orphan unlinking. Instead, skip buddy update
3416 before setup, as it will be regenerated after recovery is complete.
3421 Description: If an OST is inactive, its locks might reference stale inodes.
3422 Details : lov_change_cbdata() must iterate over all namespaces, even if
3423 they are inactive to clear inode references from the lock.
3425 Severity : enhancement
3426 Frequency : occasional, if non-standard max_dirty_mb used
3428 Description: Client will block write RPCs if not enough grant
3429 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3430 then the client will block writes while waiting for another RPC
3431 to complete instead of consuming its dirty limit. With change
3432 we get improved performance when max_dirty_mb is small.
3434 Severity : enhancement
3435 Bugzilla : 3389, 6253
3436 Description: Add support for supplementary groups on the MDS.
3437 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3438 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3439 lookups for user supplementary groups into a cache.
3443 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3444 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3445 for liblustre O_EXCL is enforced by MDS.
3449 Description: readdir never returns NULL in liblustre.
3450 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3451 dentry in struct dirent.
3455 Frequency : liblustre only, depends on application IO pattern
3456 Description: liblustre clients evicted if not contacting servers
3457 Details : Don't put liblustre clients into the ping_evictor list, so
3458 they will not be evicted by the pinger ever.
3460 Severity : enhancement
3462 Description: Add ability to evict clients by NID from MDS.
3463 Details : By echoing "nid:$NID" string into
3464 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3465 $NID would be instantly evicted from this MDS and from all active
3466 OSTs connected to it.
3470 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3471 Details : lookup_it_finish() used to query file size from OSTs that was not
3476 Description: service threads change working directory to that of init
3477 Details : Starting lustre service threads may pin the working directory
3478 of the parent thread, making that filesystem busy. Threads
3479 now change to the working directory of init to avoid this.
3483 Frequency : during shutdown only
3484 Description: shutdown with a failed MDS or OST can cause unmount to hang
3485 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3490 Frequency : 2.6 only
3491 Description: chmod/chown may include an extra supplementary group
3492 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3493 supplementary group and if none is specified this is used.
3496 Bugzilla : 5479 (6816)
3498 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3499 Details : If lookup is in progress on a file that is unlinked we might try
3500 to revalidate the inode and fail in revalidate after lookup is
3501 complete and ll_file_open() enqueues the open again but
3502 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3505 Frequency : always, if lconf --abort_recovery used
3507 Description: lconf --abort_recovery fails with 'Operation not supported'
3508 Details : lconf was attempting to abort recovery on the MDT device and not
3511 ------------------------------------------------------------------------------
3513 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3518 Frequency : rare (only unsupported configurations with a node running as an
3520 Bugzilla : 6514, 5137
3521 Description: Mounting a Lustre file system on a node running as an OST could
3523 Details : OSTs now preallocates memory needed to write out data at
3524 startup, instead of when needed, to avoid having to
3525 allocate memory in possibly low memory situations.
3526 Specifically, if the file system is mounted on on OST,
3527 memory pressure could force it to try to write out data,
3528 which it needed to allocate memory to do. Due to the low
3529 memory, it would be unable to do so and the node would
3530 become unresponsive.
3532 Severity : enhancement
3534 Description: Addition of lconf --service command line option
3535 Details : lconf now accepts a '--service <arg>' option, which is
3536 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3538 Severity : enhancement
3540 Description: Failover mode is now the default for OSTs.
3541 Details : By default, OSTs will now run in failover mode. To return to
3542 the old behaviour, add '--failout' to the lmc line for OSTs.
3544 Severity : enhancement
3546 Description: Health checks are now provided for MDS and OSTs
3547 Details : Additional detailed health check information on MSD and OSTs
3548 is now provided through the procfs health_check value.
3551 Frequency : occasional, depends on IO load
3553 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3554 after numerous create/delete cycles
3555 Details : The ext3 inode allocation policy would not allocate new inodes
3556 very well on the OSTs because there are no new directories
3557 being created. Instead we look for groups with free space if
3558 the parent directories are nearly full.
3563 Description: Network or server problems during mount may cause partially
3564 mounted clients instead of returning an error.
3565 Details : The config llog parsing code may overwrite the error return
3566 code during mount error handling, returning success instead
3572 Description: MDS can fail to allocate large reply buffers
3573 Details : After long uptimes the MDS can fail to allocate large reply
3574 buffers (e.g. zconf client mount config records) due to memory
3575 fragmentation or consumption by the buffer cache. Preallocate
3576 some large reply buffers so that these replies can be sent even
3577 under memory pressure.
3581 Frequency : rare (liblustre)
3582 Description: fsx running with liblustre complained that using truncate() to
3583 extend the file doesn't work. This patch corrects that issue.
3584 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3585 ATTR_SIZE and lsm use in llu_setattr_raw.
3589 Frequency : rare, only 2.6 kernels
3590 Description: Unusual file access patterns on the MDS may result in inode
3591 data being lost in very rare circumstances.
3592 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3593 code in the 2.6 kernel caused the inode and/or EA data not to
3594 be read from disk, causing single-file corruption.
3598 Frequency : rare, only 2.6 filesystems using extents
3599 Description: Heavy concurrent write and delete load may cause data corruption.
3600 Details : It was possible under high-load situations to have an extent
3601 metadata block in the block device cache from a just-unlinked
3602 file overwrite a newly-allocated data block. We now unmap any
3603 metadata buffers that alias just-allocated data blocks.
3607 Frequency : filesystems with default stripe_count larger than 77
3608 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3609 Details : lconf specifies an inode size of 4096 bytes when the default
3610 stripe_count is larger than 77. This conflicts with the default
3611 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3612 this case to avoid pinning too much memory for large EAs.
3614 ------------------------------------------------------------------------------
3616 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3621 Frequency : rare (extremely heavy IO load with hundreds of clients)
3623 Description: Client is evicted, gets IO error writing to file
3624 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3625 caused clients to be evicted for AST timeouts. The fixes in
3626 bug 5192 mean we no longer need to have such short AST timeouts
3627 so ldlm_timeout has been increased.
3630 Frequency : occasional during --force or --failover shutdown under load
3631 Bugzilla : 5949, 4834
3632 Description: Server oops/LBUG if stopped with --force or --failover under load
3633 Details : a collection of import/export refcount and cleanup ordering
3634 issues fixed for safer force cleanup
3637 Frequency : only filesystems larger than 120 OSTs
3638 Bugzilla : 5990, 6223
3639 Description: lfs getstripe would oops on a very large filesystem
3640 Details : lov_getconfig used kfree on vmalloc'd memory
3643 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3644 Bugzilla : 6242, 6243
3645 Description: reading from files that had been truncated to a non-zero size
3646 but never opened returned no data
3647 Details : ll_file_read() reads zeros from no-object files to EOF
3652 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3653 Details : There's little harm in aborting MDS/OSS recovery and letting it
3654 try again, so I removed the LASSERT and return an error instead.
3656 Severity : enhancement
3658 Description: New debugging infrastructure for tracking down data corruption
3659 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3660 (b) cover more of the client-side code path, and (c) try to narrow
3661 down where problems occurred
3665 Bugzilla : 3819, 4364, 4397, 6313
3666 Description: Racing close and eviction MDS could cause assertion in mds_close
3667 Details : It was possible to get multiple mfd references during close and
3668 client eviction, leading to one thread referencing a freed mfd.
3670 Severity: : enhancement
3671 Bugzilla : 3262, 6359
3672 Description: Attempts to reconnect to servers are now more aggressive.
3673 Details : This builds on the enhanced upcall-less recovery that was added
3674 in 1.4.2. When trying to reconnect to servers, clients will
3675 now try each server in the failover group every 10 seconds. By
3676 default, clients would previously try one server every 25 seconds.
3681 Description: After recovery, certain operations trigger a failed
3682 assertion on a client.
3683 Details : Failing over an mds, using lconf -d --failover, while a
3684 client was doing a readdir() call would cause the client to
3685 LBUG after recovery completed and the readdir() was resent.
3687 Severity : enhancement
3689 Description: Default groups are now added by lconf
3690 Details : You can now run lconf --group <servicename> without having to
3691 manually add groups with lmc.
3694 Frequency : occasional
3696 Description: Nodes with an elan id of 0 trigger a failed assertion
3699 Frequency : always when accessing e.g. tty/console device nodes
3701 Description: tty and some other devices nodes cannot be used on lustre
3702 Details : file's private_data field is used by device data and lustre
3703 values in there got lost. New field was added to struct file to
3704 store fs-specific private data.
3707 Frequency : when exporting Lustre via NFS
3709 Description: NFSD failed occasionally when looking up a path component
3710 Details : NFSD is looking up ".." which was broken in ext3 directories
3711 that had grown large enough to become hashed.
3714 Frequency : Clusters with multiple interfaces not on the same subnet
3716 Description: Nodes will repeatedly try to reconnect to an interface which it
3717 cannot reach and report an error to the log.
3718 Details : Extra peer list entries will be created by lconf with some peers
3719 unreachable. lconf now validates the peer before adding it.
3722 Frequency : Only if a default stripe is set on the filesystem root.
3724 Description: Setting a default stripe on the filesystem root prevented the
3725 filesystem from being remounted.
3726 Details : The client was sending extra request flags in the root getattr
3727 request and did not allocate a reply buffer for the dir EA.
3730 Frequency : occasional, higher if lots of files are accessed by one client
3731 Bugzilla : 6159, 6097
3732 Description: Client trips assertion regarding lsm mismatch/magic
3733 Details : While revalidating inodes the VFS looks up inodes with ifind()
3734 and in rare cases can find an inode that is being freed.
3735 The ll_test_inode() code will free the lsm during ifind()
3736 when it finds an existing inode and then the VFS later attaches
3737 this free lsm to a new inode.
3741 Bugzilla : 6422, 7030
3742 Description: MDS deadlock between mkdir and client eviction
3743 Details : Creating a new file via mkdir or mknod (starting a transaction
3744 and getting the ns lock) can deadlock with client eviction
3745 (gets ns lock and trying to finish a synchronous transaction).
3748 Frequency : occasional
3749 Description: While starting a server, the fsfilt_ext3 module could not be
3751 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3752 kernels. Previously, lconf would still use the ext3 name
3753 when trying to load modules. Now, it will correctly use
3754 ext3 on 2.4 and ldiskfs on 2.6.
3756 Severity : enhancement
3757 Description: The default stripe count has been changed to 1
3758 Details : The interpretation of the default stripe count (0, to lfs
3759 or lmc) has been changed to mean striping across a single
3760 OST, rather than all available. For general usage we have
3761 found a stripe count of 1 or 2 works best.
3763 Severity : enhancement
3764 Description: Add support for compiling against Cray portals.
3765 Details : Conditional compiling for some areas that are different
3769 Frequency : occasional
3770 Bugzilla : 6409, 6834
3771 Description: Creating files with an explicit stripe count may lead to
3772 a failed assertion on the MDS
3773 Details : If some OSTs are full or unavailable, creating files may
3774 trigger a failed assertion on the MDS. Now, Lustre will
3775 try to use other servers or return an error to the
3779 Frequency : occasional
3781 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3782 caused lock timeout during readahead (since 1.4.2).
3783 Details : Processes doing readahead might match a lock that hasn't been
3784 granted yet if there are overlapping and conflicting lock
3785 requests. The readahead process waits on ungranted lock
3786 (original lock is CBPENDING), while OST waits for that process
3787 to cancel CBPENDING read lock and eventually evicts client.
3789 Severity : enhancement
3791 Description: Initial enabling of flock support for clients
3792 Details : Implements fcntl advisory locking and file status functions.
3793 This feature is provided as an optional mount flag (default
3794 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3795 locking are implemented yet, and those that are are not guaranteed
3796 to be completely correct in production environments.
3797 mount -t lustre -o [flock|noflock] ...
3800 Frequency : occasional
3802 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3803 assertion in the ext3 JBD (journaling) layer.
3804 Details : The b_committed_data struct is protected by the big kernel lock
3805 in 2.4 kernels, serializing journal_commit_transaction() and
3806 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3807 access to this struct. In 2.6 kernels there is finer grained
3808 locking to improve SMP performance of the JBD layer.
3812 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3817 Description: merge and cleanup kernel patches.
3820 -----------------------------------------------------------------------------
3822 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3824 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3825 versions of Lustre. Please update all servers and clients to
3826 version 1.4.2 or later at the same time. You must also run
3827 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3828 to update the configuration logs.
3830 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3831 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3832 - swab llog records of type '0' so we get proper header size/idx (5861)
3833 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3834 - fix rename of one directory over another leaking an inode (5953)
3835 - avoid SetPageDirty on 2.6 (5981)
3836 - don't re-add just-being-destroyed locks to the waiting list (5653)
3837 - when creating new directories, inherit the parent's custom
3838 striping settings if present parent (3048)
3839 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3840 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3841 - don't hold client locks on temporary worklist from l_lru (5666)
3842 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3843 - automatically evict dead clients (5921)
3844 - Update file size properly in create+truncate+fstat case (6196)
3845 - Do not unhash mountpoint dentries, do not allow removal of
3847 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3848 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3849 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3850 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3851 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3852 - Correctly handle reads of files with no objects (6243)
3853 - lctl recover will also mark a device active if deactivate used (5933)
3855 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3856 - allow --write-conf on an MDS with different nettype than client (5619)
3857 - don't write config llogs to MDS for mounts not from that MDS (5617)
3858 - lconf should create multiple TCP connections from a client (5201)
3859 - init scripts are now turned off by default; run chkconfig --on
3860 lustre and chkconfig --on lustrefs to use them
3861 - upcalls are no longer needed for clients to recover to failover
3863 - add --abort-recovery option to lconf to abort recovery on device
3865 - add support for an arbitrary number of OSTs (3026)
3866 - Quota support protocol changes.
3867 - forward compatibility changes to wire structs (6007)
3868 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3869 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3870 - improved "lustre status" script
3871 - initialize blocksize for non-regular files (6062)
3872 - added --disable-server and --disable-client configure options (5782)
3873 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3874 - Vanilla 2.4.29 support
3875 - increase maximum number of obd devices to 520 (6242)
3876 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3877 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3878 SLES 9 2.6.5 kernels
3879 - limit stripes per file to 160 (the maximum EA size) (6093)
3881 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3884 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3885 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3886 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3887 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3888 - small ext3 extents cleanups and fixes (5733)
3889 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3890 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3891 - avoid needless client->OST connect, fix handle mismatch (5317)
3892 - fix DLM error path that led to out-of-sync client, long delays (5779)
3893 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3894 - fix several locking issues related to i_size (5492,5624,5654,5672)
3895 - don't move pending lock onto export if it is already evicted (5683)
3896 - fix kernel oops when creating .foo in unlinked directory (5548)
3897 - fix deadlock in obdfilter statistics vs. object create (5811)
3898 - use time_{before,after} to avoid timer jiffies wrap (5882)
3899 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3900 - Do not leak request if server was not able to process it (5154)
3901 - If mds_open unable to find parent dir, make that negative lookup(5154)
3902 - don't create new directories with extent-mapping (5909, 5936)
3904 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3905 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3906 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3907 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3908 - support for RHEL4 kernel on i686 (5773)
3909 - provide error messages when incompatible logs are encountered (5898)
3911 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3912 * version 1.4.0.10 (1.4.1 release candidate 1)
3914 - don't keep a lock reference when lock is not granted (4238)
3915 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3916 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3917 - drop import inflight refcount on signal_completed_replay error (5255)
3918 - unlock page after async write error during send (3677)
3919 - handle missing objects in filter_preprw_read properly (5265)
3920 - no transno return for symlink open, don't save no-trasno open (3440)
3921 - don't try to complete elan receive that already failed (4012)
3922 - free RPC server reply state on error (5406)
3923 - clean up thread from ptlrpc_start_thread() on error (5160)
3924 - readahead could read extra page into cache that wasn't ejected (5388)
3925 - prevent races in class_attach/setup/cleanup/detach (5260)
3926 - don't dereference de->d_inode after l_dput of de (5458)
3927 - use "int" for stripe value returned from lock_to_stripe (5544)
3928 - mballoc allocation and error-checking fixes in 2.6 (5504)
3929 - block device patches to fix I/O request sizes in 2.6 (5482)
3930 - look up hostnames for IB nals (5602)
3931 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3932 - don't start multiple acceptors for the same port (5277)
3933 - fix incorrect LASSERT in mds_getattr_name (5635)
3934 - export a proc file for general "ping" checking (5628)
3935 - fix "lfs check" to not block when the MDS is down (5628)
3937 - service request history (4965)
3938 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3939 - create an "evict_client" /proc entry on OSTs, like the MDS has
3940 - fix mount usage message, return errors per mount(8) (5168)
3941 - change grep [] to grep "[]" in tests so they work in more UMLs
3942 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3943 - remove ancient LOV_MAGIC_V0 EA support (5047)
3944 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3945 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3946 - allow readahead allocations to fail when low on memory (5383)
3947 - mmap locking landed again, after considerable improvement (2828)
3948 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3950 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3953 - send OST transaction number in read/write reply to free req (4966)
3954 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3955 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3956 - the watchdog thread now runs as interruptible (5246)
3957 - flock/lockf fixes (but it's still disabled, pending 5135)
3958 - don't use EXT3 constants in llite code (5094)
3959 - memory shortage at startup could cause assertion (5176)
3961 - reorganization of lov code
3962 - single portals codebase
3964 - add extents/mballoc support (5025)
3965 - direct I/O reads in the obdfilter (4048)
3966 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3968 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3971 - send OST transaction number in read/write reply to free req (4966)
3972 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3973 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3974 - the watchdog thread now runs as interruptible (5246)
3975 - handle missing objects in filter_preprw_read properly (5265)
3976 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3977 - drop import inflight refcount on signal_completed_replay error (5255)
3978 - unlock page after async write error during send (3677)
3979 - return original error code on reconstructed replies (3761)
3980 - no transno return for symlink open, don't save no-trasno open (3440)
3982 - add pid to ldlm debugging output (4922)
3983 - bump the watchdog timeouts -- we can't handle 30sec yet
3984 - extra debugging for orphan dentry/inode bug (5259)
3986 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3989 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3990 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3991 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3992 - fix NULL dereference in /proc/sys/portals/routes (4827)
3993 - allow failed mdc_close() operations to be interrupted (4561)
3994 - stop precreate on OST before MDS would time out on it (4778)
3995 - don't send partial-page writes before EOF from client (4410)
3996 - discard client grant for sub-page writes on large-page clients (4520)
3997 - don't free dentries not owned by NFS code, check generation (4806)
3998 - fix lsm leak if mds_create_objects() fails (4801)
3999 - limit debug_daemon file size, always print CERROR messages (4789)
4000 - use transno after validating reply (3892)
4001 - process timed out requests if import state changes (3754)
4002 - update mtime on OST during writes, return in glimpse (4829)
4003 - add mkfsoptions to LDAP (4679)
4004 - use ->max_readahead method instead of zapping global ra (5039)
4005 - don't interrupt __l_wait_event() during strace
4007 - add software watchdogs to catch hung threads quickly (4941)
4008 - make lustrefs init script start after nfs is mounted
4009 - fix CWARN/ERROR duplication (4930)
4010 - return async write errors to application if possible (2248)
4011 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4012 - print NAL number in %x format (4645)
4013 - update barely-supported suse-2.4.21-171 series (4842)
4014 - support for sles 9 %post scripts
4015 - support for building 2.6 kernel-source packages
4016 - support for sles km_* packages
4018 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4021 - ignore -ENOENT errors in osc_destroy (3639)
4022 - notify osc create thread that OSC is being cleaned up (4600)
4023 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4024 - reconstruct ost_handle() like mds_handle() (4657)
4025 - create a new thread to do import eviction to avoid deadlock (3969)
4026 - let lconf resolve symlinked-to devices (4629)
4027 - don't unlink "objects" from directory with default EA (4554)
4028 - hold socknal file ref over connect in case target is down (4394)
4029 - allow more than 32000 subdirectories in a single directory (3244)
4030 - fix blocks count for O_DIRECT writes (3751)
4031 - OST returns ENOSPC from object create when no space left (4539)
4032 - don't send truncate RPC if file size isn't changing (4410)
4033 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4034 - bind to privileged port in socknal and tcpnal (3689)
4036 - rate limit CERROR/CWARN console message to avoid overload (4519)
4037 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4038 - basic mmap support (3918)
4039 - kernel patch series update from b1_4 (4711)
4041 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4044 - avoid crash during MDS cleanup with OST shut down (2775)
4045 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4046 - don't use bad inodes on the MDS (3744)
4047 - dynamic object preallocation to improve recovery speed (4236)
4048 - don't hold spinlock over lock dumping or change debug flags (4401)
4049 - don't zero obd_dev when it is force cleaned (3651)
4050 - print grants to console if they go negative (4431)
4051 - "lctl deactivate" will stop automatic recovery attempts (3406)
4052 - look for existing locks in ldlm_handle_enqueue() (3764)
4053 - don't resolve lock handle twice in recovery avoiding race (4401)
4054 - revalidate should check working dir is a directory (4134)
4056 - don't always mark "slow" obdfilter messages as errors (4418)
4058 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4061 - don't close LustreDB during write_conf until it is done (3860)
4062 - fix typo in lconf for_each_profile (3821)
4063 - allow dumping logs from multiple threads at one time (3820)
4064 - don't allow multiple threads in OSC recovery (3812)
4065 - fix debug_size parameters (3864)
4066 - fix mds_postrecov to initialize import for llog ctxt (3121)
4067 - replace config semaphore with spinlock (3306)
4068 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4069 - don't allow enqueue to complete on a destroyed export (3822)
4070 - down write_lock before checking llog header bitmap (3825)
4071 - recover from lock replay timeout (3764)
4072 - up llog sem before sending rpc (3652)
4073 - reduce ns lock hold times when setting kms (3267)
4074 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4075 - fix NULL deref and obd_dev leak on setup error (3312)
4076 - replace some LBUG about llog ops with error handling (3841)
4077 - don't match INVALID dentries from d_lookup and spin (3784)
4078 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4079 - fix invalid assertion in ptlrpc_set_wait (3880)
4081 - add libwrap support for the TCP acceptor (3996)
4082 - add /proc/sys/portals/routes for non-root route listing (3994)
4083 - allow setting MDS UUID in .xml (2580)
4084 - print the stack of a process that LBUGs (4228)
4086 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4089 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4090 - make sure to unset replay flag from failed open requests (3440)
4091 - if default stripe count is 0, use OST count for inode size (3636)
4092 - update parent mtime/ctime on client for create/unlink (2611)
4093 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4094 - free recovery state on server during a forced cleanup (3571)
4095 - unregister_reply for resent reqs (3063)
4096 - loop back devices mounting and status check on 2.6 (3563)
4097 - fix resource-creation race that can provoke i_size == 0 (3513)
4098 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4099 - more debugging for page-accounting assertion (3746)
4100 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4101 - avoid deadlock after precreation failure (3758)
4102 - fix race and lock order deadlock in orphan handling (3450, 3750)
4103 - add validity checks when grabbing inodes from l_ast_data (3599)
4105 - add /proc/.../recovery_status to obdfilter (3428)
4106 - lightweight CDEBUG infrastructure, debug daemon (3668)
4107 - change default OSC RPC parameters to be better on small clusters
4108 - turn off OST read cache for files smaller than 32MB
4109 - install man pages and include them in rpms (3100)
4110 - add new init script for (un)mounting lustre filesystems (2593)
4111 - run chkconfig in %post for init scripts (3701)
4112 - drop scimac NAL (unmaintained)
4114 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4117 - clean kiobufs before and after use (3485)
4118 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4119 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4120 - change init script to fail more clearly if not run as root (1528)
4121 - allow clients to reconnect during replay (1742)
4122 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4123 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4124 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4125 - fix cond_resched (3554)
4126 - only evict unfinished clients after recovery (3515)
4127 - allow bulk resend, prevent data loss (3570)
4128 - dynamic ptlrpc request buffer allocation (2102)
4129 - don't allow unlinking open directory if it isn't empty (2904)
4130 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4131 - remove extraneous obd dereference causing LASSERT failure (3334)
4132 - don't use get_cycles() when creating temp. files on the mds (3156)
4133 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4134 - handle EEXIST for set-stripe, set proper directory name (3336)
4136 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4137 - fix ksocknal_fmb_callback() error messages (2918)
4139 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4142 - don't copy lvb into (possibly NULL) reply on error (2983)
4143 - don't deref dentry after dput, don't free lvb on error (2922)
4144 - use the kms to determine writeback rpc length (2947)
4145 - increment oti_logcookies when osc is inactive (2948)
4146 - update client's i_blocks count via lvb messages (2543)
4147 - handle intent open/close of special files properly (1557)
4148 - mount MDS with errors=remount-ro, like obdfilter (2009)
4149 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4150 - don't use cancelling-locks' kms values (2947)
4151 - use highest lock extent for kms, not last one (2925)
4152 - don't dereference ERR_PTR() dentry in error handling path (3107)
4153 - fix thread race in portals_debug_dumplog() (3122)
4154 - create lprocfs device entries at setup instead of at attach (1519)
4155 - common AST error handler, don't evict client on completion race (3145)
4156 - zero nameidata in detach_mnt in 2.6 (3118)
4157 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4158 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4159 - initialize RPC timeout timer earlier for 2.6 (3219)
4160 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4161 - print nal/nid for unknown nid (3258)
4162 - additional checks for oscc recovery before doing precreate (3284)
4163 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4164 - don't crash in mdc_close for bad permissions on open (3285)
4165 - zero i_rdev for non-device files (3147)
4166 - clear page->private before handing to FS, better assertion (3119)
4167 - tune the read pipeline (3236)
4168 - fix incorrect decref of invalidated dentry (2350)
4169 - provide read-ahead stats and refine rpc in flight stats (3328)
4170 - don't hold journal transaction open across create RPC (3313)
4171 - update atime on MDS at close time (3265)
4172 - close LDAP connection when recovering to avoid server load (3315)
4173 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4174 - don't leak open file on MDS after open resend (3325)
4175 - serialize filter_precreate and filter_destroy_precreated (3329)
4176 - loop device shouldn't call sync_dev() for nul device (3092)
4177 - clear page cache after eviction (2766)
4178 - resynchronize MDS->OST in background (2824)
4179 - refuse to mount the same filesystem twice on same mountpoint (3394)
4180 - allow llmount to create routes for mounting behind routers (3320)
4181 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4182 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4183 - fix rare problem with rename on htree directories (3417)
4185 - allow default OST striping configuration per directory (1414)
4186 - fix compilation for qswnal for 2.6 kernels (3125)
4187 - increase maximum number of MDS request buffers for large systems
4188 - change liblustreapi to be useful for external progs like lfsck (3098)
4189 - increase local configuration timeout for slow disks (3353)
4190 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4192 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4195 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4196 - don't overwrite extent policy data in reply if lock was blocked (2901)
4197 - drop filter export grants atomically with removal from device (2663)
4198 - del obd_self_export from work_list in class_disconnect_exports (2908)
4199 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4200 - swab reply message in mdc_close, other PPC fixes (2464)
4201 - fix destroying of named logs (2325)
4202 - overwrite old logs when running lconf --write_conf (2264)
4203 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4204 - fix race in target_handle_connect (2898)
4205 - mds_reint_create() should take same inode create lock (2926)
4206 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4207 - don't close files for self_export to avoid uninitialized obd (2936)
4208 - allow MDS with the same name as client node (2939)
4209 - hold dentry reference for closed log files for unlink (2325)
4210 - reserve space for all logs during transactions (2059)
4211 - don't evict page beyond end of stripe extent (2925)
4212 - don't oops on a deleted current working directory (2399)
4213 - handle hard links to targets without a parent properly (2517)
4214 - don't dereference NULL lock when racing during eviction (2867)
4215 - don't grow lock extents when lots of conflicting locks (2919)
4217 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4220 - account for cache space usage on clients to avoid data loss (974)
4221 - lfsck support in lustre kernel code (2349)
4222 - reduce journal credits needed for BRW writes (2370)
4223 - orphan handling to avoid losing space on client/server crashes
4224 - ptlrpcd can be blocked, stopping ALL progress (2477)
4225 - use lock value blocks to assist in proper KMS, faster stat (1021)
4226 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4227 - recovery for initial connections (2355)
4228 - fixes for mds_cleanup_orphans (1934)
4229 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4230 - block all file creations until orphan recovery completes (1901)
4231 - client remove rq_connection from request struct (2423)
4232 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4233 - recovery timer race (2670)
4234 - mdc_close recovey bug (2532)
4235 - ptlrpc cleanup bug (2710)
4236 - mds timeout on local locks (2588)
4237 - namespace lock held during RPCs (2431)
4238 - handle interrupted sync write properly (2503)
4239 - don't try to handle a message that hasn't been replied to (2699)
4240 - client assert failure during cleanup after abort recovery (2701)
4241 - leak mdc device after failed mount (2712)
4242 - ptlrpc_check_set allows timedout requests to complete (2714)
4243 - wait for inflight reqs when ptlrpcd finishes (2710)
4244 - make sure unregistered services are removed from the srv_list
4245 - reset bulk XID's when resending them (caught by 1138 test)
4246 - unregister_bulk after timeout
4247 - fix lconf error (2694)
4248 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4249 - readahead locks pages, leaves pending causing memory pressure (2673)
4250 - increase OST request buffers to 4096 on large machines (2729)
4251 - fix up permission of existing directories in simple_mkdir (2661)
4252 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4253 - don't assign transno to errored transactions (2742)
4254 - don't delete objects on OST if given a bogus objid from MDS (2751)
4255 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4256 - if rq_no_resend, then timeout request after recovery (2432)
4257 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4258 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4259 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4260 - MDS recovery shouldn't delete objects that it has given out (2730)
4261 - if enqueue arrives after completion, don't clobber LVB (2819)
4262 - don't unlock pages twice when trigger_group_io returns error (2814)
4263 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4264 - don't write pages to disk if there was an error (1450)
4265 - don't ping imports that have recovery disabled (2676)
4266 - take buffered bytes into account when balancing socknal conn (2817)
4267 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4268 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4269 - remove little-endian swabbing of llog records (1987)
4270 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4271 - retry reposting request buffers if they fail (1191)
4272 - grow extent at grant time to avoid granting a revoked lock (2809)
4273 - lock revoke doesn't evict page if covered by a second lock (2765)
4274 - disable VM readahead to avoid reading outside lock extents (2805)
4276 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4277 - updated kernel patches for hp-2.4.20 kernel (2681)
4279 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4282 - fix truncated write corruption (2366)
4283 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4285 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4286 - ensure len(uuid) < 37 in lmc (1171)
4287 - fix ia64 OOPS in llog_test (2255)
4288 - zero end of page at obdfilter for partial page writes (2648)
4289 - don't leave stale dentries around after renames (bug 2428)
4290 - fix timeouts when evicting a client with a single lock held (2642)
4291 - set deadline for the initial HELLO message to drain (2634)
4292 - print out dotted-quad IP addresses in the socknal (2302)
4294 - additional debugging for MDS client eviction problem (2443)
4295 - fix mkfsoptions support for osts (2603, 2604)
4297 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4300 - add series for the vanilla 2.6.0 kernel
4301 - add series for the vanilla 2.4.24 kernel
4302 - add series for a cray x86/64 UL kernel drop
4303 - fix xattr patches for the vanilla 2.4.19 series
4305 - generate true UUIDs in lmc (1171)
4306 - have portals stack dumping break in UML (2466)
4307 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4308 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4309 - wake callback waiting threads on client eviction (2460)
4310 - Add --ptldebug and --subsystem to lmc (1719)
4311 - update assertion to allow safe interrupt allocation
4312 - set rq_no_resend for cancel requests (2432)
4313 - recalculate ptlrpcd timeout after resend (2494)
4314 - call vfs_rmdir when removing pending directories (2368)
4315 - fix renaming a file to itself (2429)
4316 - lmc creates a default one-stripe lov (2454)
4317 - expand procfs space to handle large clusters (2326)
4318 - increase UML stack to avoid overflow
4319 - update lconf's list of debug and subsystem masks
4320 - fix lfs find --obd (2510)
4321 - /proc tunable for disabling filter read caching (2591)
4322 - stop rpm packages from altering slapd.conf (2301)
4323 - disable nagle in the socknal under 0conf (2578)
4324 - choose mds inode size based on stripe count (2572)
4325 - fix kernel-source rpm problems (2516)
4327 - add --disable-doc to avoid pdf generation (2421)
4328 - update documentation, tests, type-os, comments
4329 - avoid format warnings on ia64
4330 - remove the TOE NAL
4331 - tiny code cleanups by removing unused fields
4333 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4336 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4337 - remove the most blatant lies from BUILDING file (2371)
4338 - change default debug level to reasonable production setting
4339 - reduce client side cache size to reduce cache flush time
4340 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4341 - make TCP zerocopy and pinger support enabled by default (2476)
4342 - sync writes completed after process exits caused crashes (2319)
4343 - maintain correct mount count on the MDS (2356)
4344 - backout 1557, because 2316 wasn't really fixed
4345 - better file I/O statistics gathering in /proc
4346 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4347 - another kernel patch to fix zero-copy TCP function export
4348 - don't take duplicate lock when processing re-sent getattr (2420)
4349 - lctl uses obd_self_export instead of creating new conn (2353)
4350 - MDS/OST recovery case which requires object creation asserted (2425)
4351 - move lfs from /usr/sbin to /usr/bin in packages
4352 - fix race between mds_client_add and mds_client_free (2417)
4353 - use kmalloc instead of slabs in portals (2430)
4354 - don't create duplicate records when a failover MDS is present (2442)
4355 - remove unnecessary mount age check (2332)
4356 - don't remove directory inodes from locks prematurely (2451)
4357 - don't break if MDS service name is the same as hostname (2103)
4358 - fix races in client write RPC generation when cache full (2482)
4360 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4363 - remove now-unused request->rq_obd (278)
4364 - if an allocation fails, print out how much memory we've used (1933)
4365 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4366 - add the "configurable stack size" patch to most series files (1256)
4367 - ability to write large log records, for 100+ OST configs (2306)
4368 - fix NULL deref when filter_prep fails (2314)
4369 - fix operator precedence error in filter_sync
4370 - dynamic allocation of socknal TX descriptors (2315)
4371 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4372 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4373 - ensure that CWARN messages in Portals always get to the syslog
4374 - __init/__exit are not for prototype decls (ldlm_init/exit)
4375 - x86-64 compile warning fixes
4376 - fix gateway LMC keyword conflict (2318)
4377 - fix MDS lock inversions in getattr/reint paths (1844)
4378 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4379 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4380 - DLM race condition prevented some lock evictions (2328)
4381 - ENOMEM detection and retry on socknal sends (2230)
4382 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4383 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4384 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4385 - better error messages when a client is rejected during recovery (1505)
4386 - avoid cancelling locks which were never granted, after failure (2330)
4387 - fix i_sem/journal inversion in mds_client_add (2333)
4388 - fix truncate/getattr lock cycle deadlock (2334)
4389 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4390 - fix two rare exit paths which could leak an l_lock() ref (2321)
4391 - fencepost error in MDS/OST orphan recovery (2226)
4392 - make log record alignment 8 bytes (1988)
4393 - lstripe now fails when requested offset > ost_count (2237)
4394 - ensure that all kernel series have a complete list.h (1607)
4395 - fix crashes in special-file operations (2316)
4396 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4398 - allow configurable automake binary, for testing new versions
4399 - small update to the lfs documentation
4401 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4403 * fix negative export reference count in fsfilt_sync (2312)
4405 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4406 * release candidate 0.9.1
4408 - orphans are moved into the PENDING directory for possible recovery
4409 - replayed opens now open by fid for orphan/rename safety (1042)
4410 - last close of an orphan inode generates a transno (683)
4411 - chdir() and mount() now pin the directory entry (1020)
4412 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4413 - discard very old requests without processing them (1502)
4414 - remove some common, well-understood CERRORs (1505)
4415 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4416 - clear "grant" flags in OST replies until OST grant code lands (1644)
4417 - fix read performance by not clobbering i_blksize on client (1598)
4418 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4419 - make LRU size a /proc tunable, clears locks when reduced (707)
4420 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4421 - server locks take references on exports now (1558)
4422 - build fixes for 2.4.20-rh trees (1663)
4423 - return an error from lov_create if all OSCs are inactive (1751)
4424 - fix import levels when a reconnect happens without a timeout (1597)
4425 - exit early from mds_open if we get a lookup error (1749)
4426 - partial page read at EOF wouldn't wait for disk before sending (1642)
4427 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4428 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4429 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4430 - write/truncate lock inversion (1639)
4431 - Don't auto-load obdclass, portals modules during cleanup (1495)
4432 - fix timestamps from jumping to "now" (1763)
4433 - extra journal assertions (1648)
4434 - add an extra multiunlink test (1771)
4435 - fix read_record/write_record API (1776)
4436 - fix leak of offset_extent, possible incorrect i_size later (1772)
4437 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4438 - Debugging for the kqswnal_get_idle_tx problems (1820)
4439 - Allow recovery to be attempted multiple times (1536)
4440 - Write out MDS last_rcvd file after it is first created (1600)
4441 - Fix tx_descriptor leak in failed transmit situations (1827)
4442 - ext3 journaling fixes for assertion failure after IO error (1871)
4443 - class_export_put() on freed export after completion AST error (1896)
4444 - Fix revalidate looping in VFS (1322)
4445 - Don't access a freed export during MDS_REINT timeout (1521)
4446 - Add open-unlink recovery support on the MDS (1673,1764)
4447 - Return an error if no MDS data was read from last_rcvd (1946)
4448 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4449 - Don't setup a disk device that doesn't match exported UUID (317)
4450 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4451 - avoid committing NULL handle in force close
4452 - local.sh is now a one-stripe LOV configuration
4453 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4454 - fix ext3 htree duplicate directory entry corruption (1516)
4455 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4456 - update to new LOV EA format (2097)
4457 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4459 2003-06-15 Phil Schwan <phil@clusterfs.com>
4462 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4463 - per-import/export recovery handling (958, 931, 959)
4464 - multiple last-rcvd slots, for serving multiple FSes (949)
4465 - connections are again shared between multiple imp/exports (963, 964)
4466 - "umount -f" would hang if any requests needed to be sent (393, 978)
4467 - avoid pinning large req buffer by copying for queued messages (989)
4468 - add "uuid" to "lctl device" command to help upcalls (991)
4469 - "open" RPCs with transnos would confuse recovery counters (1037)
4470 - do proper endian conversion of all wire messages (288, 340, 891)
4471 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4472 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4473 - fix ptlrpc_connection leak in target_handle_connect (1174)
4474 - fix import refcounting bug in OST and MDS cleanup (1134)
4475 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4476 - fix very unlikely obd_types race condition (501)
4477 - remove osc_open hack for echo_client (1187)
4478 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4479 - a failure in read_inode2 leads to deadlock (1139)
4480 - cancel ack-locks as soon as transaction is committed (1072)
4481 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4482 - make sure to commitrw after any preprw to avoid deadlock (1162)
4483 - failing to execute a file in a lustre FS would lock inode (1203)
4484 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4485 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4486 - obd_uuid2tgt would walk off the end of the list (1255)
4487 - on IA64 the osc would give portals incorrect bulk size (1258)
4488 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4489 - fix lock inversion caused by new llite matching code (1282)
4490 - limit the number of dirty pages on a client to 10MB (1286)
4491 - timed out locks were not being corrected cancelled (1289)
4492 - fix O_DIRECT above 4GB on IA-32 (1292)
4493 * major user-visible changes
4494 - fail out/fail over policy now controlled by the upcall (993)
4496 - add OBD_PING to check server availability and failure (954)
4497 - lustre messages are now sent in sending host order (288, 340, 891)
4498 - add eadatalen to MDS getattr reply (340)
4499 - OST read replies may contain second buffer, with per-page status (593)
4501 2003-03-11 Phil Schwan <phil@clusterfs.com>
4504 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4505 - failed open()s could cause deadlock; fixed (867, 869)
4506 - stop cancelling OST locks when files are closed (481)
4507 - overlapping XID spaces caused network corruption (851, 853)
4508 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4509 - setattr_raw added, to do single-RPC, server-side setattrs
4510 - lmc/lconf syntax change for OST UUIDs
4511 - fix crashy race condition between ptlrpc_free_req and osc_close
4512 - don't use request in mdc_enqueue if we hit a timeout (889)
4513 - don't set the inode i_size for regular files from the MDS (896)
4514 - handle out of order completion AST (842)
4515 - don't LBUG if a lock request times out after receiving AST (913)
4516 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4517 - if a bad lock AST arrives, send an error instead of dropping entirely
4518 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4519 - fix leak in bulk IO when only partially completed (899, 900, 926)
4520 - fix O_DIRECT for ia64 (55)
4521 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4522 - C-z after timeout could hang a process forever; fixed (977)
4524 - client-side I/O cache (678, 924, 929, 941, 970)
4526 - READPAGE and SETATTRs which don't take server-side locks get
4529 2003-02-11 Phil Schwan <phil@clusterfs.com>
4532 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4533 - Fix fsfilt_extN_readpage() to read a full page of directory
4534 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4535 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4536 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4537 - Most intent operations were removed, in favour of a new RPC mode
4538 that does a single RPC to the server and bypasses most of the VFS
4539 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4540 - Aggressively cancel local locks on DLM servers
4541 - mds_reint_unlink sends EA to the client if it's the last nlink.
4542 client uses that EA to unlink OST objects.
4543 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4544 - recursive symlinks were fixed (439)
4545 - fixed NULL deref in DEBUG_REQ
4546 - filter_update_lastobjid no longer calls sync, which annoyed extN
4547 - fixed multi-client small-writes to a single file problem (445)
4548 - fixed mtime updates during file writes (607)
4549 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4550 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4551 - fixed Linux OST size reporting problem (444, 656)
4552 - OST now updates object mtime with writes or setattr (607, 619)
4553 - client verifies file size before zeroing page past EOF (445)
4554 - OST now writes last allocated objid to disk with allocation (108)
4555 - LOV on echo now works (409)
4557 - mds_reint_unlink sends a new buffer, with the EA included. this
4558 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4559 set if a regular file was being unlinked, and it was the last link
4560 - use PtlGet from the target for bulk writes (315)
4561 - OST now updates object mtime with writes or setattr (607, 619)
4562 - LDLM now has a grant-time callback to revalidate locked items, if
4564 - Many MDS operations were reorganized to combat race conditions
4566 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4567 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4569 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4572 - Fully reactivate OST imports after reconnection (512, others)
4573 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4574 - More graceful error handling for truncating on dead OST (515)
4575 - Don't error out unless we're actually accessing dead stripes (474)
4576 - Fix garbage sizes when stripes are missing (410)
4577 - LRU counters were broken, causing constant lock purge (433, 432)
4578 - garbage on read from stripes with failed OSTs (441)
4579 - mark OSCs as active before reconnecting during recovery (438)
4580 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4581 - lfind did not preserve OST order in output (443)
4582 - symlinks cause hung clients, incorrect data (439)
4583 - stop dereferencing request after dropping refcount (457)
4584 - don't LASSERT(spin_is_locked) on non-SMP (455)
4585 - fixes for many rename() bugs
4586 - fstat didn't correctly synchronize attributes (399)
4587 - server must handle lock cancellation during blocking AST prep (487)
4588 - bulk descriptors were free()d too soon (511)
4589 - fix paths in lconf, which would load incorrect modules (451, 507)
4590 - fix confusing lconf 'host not found' error message (386)
4591 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4592 - fix race condition in mdc_blocking_ast() for inode access (526)
4593 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4594 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4595 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4596 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4597 - fix lstripe in several ways (kernel side) (527)
4598 - fix request leak in ldlm_cli_enqueue (262)
4599 - incorrect OSC was marked inactive after OST failure
4600 - call mds_fs_cleanup before unmounting filesystem (524)
4601 - fix races between taking ns_lock and ldlm_lock_change_resource
4602 - fix races updating LOV export open file list
4603 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4604 - fix recovery NULL deref in ldlm_cli_cancel_unused
4605 - fix some DLM races by using new hash table for lock handles (419)
4606 - permit the client to specify desired inodes, at replay
4607 - duplicate requests when we queue them for replay reintegration
4608 - fix last_rcvd offset calculation
4609 - sync after each recovered transaction, so we always make progress
4610 - never, not always, ERESTART requests without transnos
4611 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4613 - skip replay if the MDS says that the client is already connected
4614 - don't check for a recovery-enabled export to match lctl's UUID
4615 - don't INC_USE_COUNT for phantom exports
4616 - don't crash when cleaning up phantom exports (567)
4617 - don't double-finish or set replay data for errored mdc_open requests
4618 - abort requests when they time out, so we don't get old replies
4619 - send/receive replies for AST messages again
4620 - if the client says that it doesn't have the lock, cancel it on the
4622 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4623 mark it as destroyed, it will all work out when decref is called
4624 - fix module use counts (22, 581)
4626 - ASTs now expect a reply (server cancels lock on error reply)
4628 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4631 - fix many simultaneous client startup (392)
4632 - fix dentry->d_it clobbering
4633 - credentials weren't being shipped for readdir/getattr operations
4634 - remove invalid assertions triggered during some concurrent MD
4636 - proper Lustre versions added (336, 389)
4637 - fix memory leak for create error case (398)
4638 - fix LOV locking bug that would get cli/srv out of sync
4639 - fix echo client over LOV (409)
4640 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4641 - fix double-O_EXCL intent crash (424)
4642 - avoid sending multiple lock CANCELs (352)
4644 - MDS can do multi-client recovery (modulo bugs in new code)
4646 - many updates, edits, cleanups
4648 2002-11-18 Phil Schwan <phil@clusterfs.com>
4651 - fix null d_it dereference (346)
4652 - fix full OST/dbench hang (333)
4653 - fix permission problem with file removal (286)
4654 - fix removal of OSCs from LOV when they fail
4655 - fix NULL deref during bulk timeout (214)
4656 - fix problems related to multiple filesystems on one MDS (241)
4657 - fixed serious subtle metadata locking bugs
4658 - free locks on clients when inodes are removed due to memory
4660 - fix inode pointer in lock data (285)
4661 - partial support for multiple MDS on a single host (241)
4662 - data locks weren't cancelled at clear_inode time (290, 311)
4663 - intent locks could lead to unbounded lock growth (205)
4664 - added a maximum lock count, an LRU list, and a flusher
4665 - fix multiple rename (365)
4666 - properly abstracted the echo client
4667 - OSC locked 1 byte too many; fixed
4668 - rewrote brw callback code:
4669 - fixed recovery bugs related to LOVs (306)
4670 - fixed too-many-pages-in-one-write crash (191)
4671 - fixed (again) crash in sync_io_timeout (214)
4672 - probably fixed callback-related race (385)
4674 - Add capability to MDS protocol
4675 - LDLM cancellations and callbacks on different portals
4677 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4680 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4681 - defer open object destruction to close time (601981, 138)
4682 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4683 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4684 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4685 - don't call obd_setattr in ll_file_release for destroyed objects
4686 * protocol change to lustre_msg: move |version| and add |flags|
4687 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4688 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4689 * added replay of create, unlink, link and rename operations during
4690 MDS failover; recovery should be much more robust now
4691 * remove failed OSCs from LOVs (only lov_create uses this so far)
4692 * the lustre-HOWTO was brought (more) up to date (582544)
4694 2002-10-23 Phil Schwan <phil@clusterfs.com>
4697 - in-use dentries weren't being reused properly (617851)
4698 - prevent multiple LDLM setup (599178)
4699 - fix LOV size calculations for truncate (617853)
4700 - fix client handling of MDS intent errors (POSIX)
4701 - fix permission bug in lovstripe.c test (624321)
4702 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4703 - truncate past end of file could corrupt data
4704 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4705 - a race in recovery could return ETIMEDOUT to apps (623947)
4706 - building outside the source directory was fixed
4707 * the lustre-HOWTO was brought (more) up to date (582544)
4708 * major progress was made on recovery functionality
4710 2002-10-10 Phil Schwan <phil@clusterfs.com>
4713 - recovery deadlock fix
4714 - rm -rf causes LBUG fix (617817)
4715 - file open by multiple tasks fix (618962)
4716 - directory permissions bugs (602707 and 620007)
4717 - journal_stop fixed with locking (611313)
4718 - O_APPEND failures resolved (618273, perhaps 614459)
4719 - lconf PATH fix (619770)
4720 - IA64 build fix (621450)
4721 - RPC buffer sizes scale with amount of memory
4723 2002-10-01 Phil Schwan <phil@clusterfs.com>
4726 - locks would be cancelled without throwing away data pages,
4727 resulting in inconsistent data (605627)
4728 - inode attributes were not always being refreshed (605627, 612449)
4729 - lconf now continues to cleanup after lctl reports an error
4730 - MDS now enforces user permissions (602707)
4731 - lprocfs cleanup fixed, but not yet enabled (614157)
4732 - fixed infinite server hang, should a client not respond to an AST
4733 - avoid going into recovery if user calls readlink() with a buffer
4734 that's too small (613941)
4735 - AST RPCs no longer require replies (614867) -- this may be changed
4736 - don't crash server if client sends an IOV that's too big (611336)
4737 - fixed lock conversion deadlock (611892)
4738 - fixed the following of symlinks (614622)
4739 * recovery: the server can remove locks from a client that dies, other
4740 clients can make progress
4741 * more extN patch fixes
4742 * compile-time configurable ptlrpc buffer allocations
4744 - collaborative read cache document
4745 - Lustre Lite Performance CDR document-in-progress
4747 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4750 - fix typo in patch-2.4.18
4752 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4755 - clear ptlrpc request each time in handle_incoming_request()
4756 - unlink of files now destroys the object on the OST
4758 2002-09-19 Peter Braam <braam@clusterfs.com>
4760 * add hard link support
4761 * change obdfile creation method
4762 * kernel patch changed
4764 2002-09-19 Peter Braam <braam@clusterfs.com>
4767 - stack overflow bug in extN fixed
4769 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4771 * documentation updates
4772 - add man pages for config tools
4773 - update tests/README to describe testing with new config tools
4774 - finish metadata API descriptions
4775 * bug fixes and cleanups
4776 - statfs workaround for 16TB limit
4777 - LOV stripe allocation improved, can stripe on subset of OSTs
4778 - LOV file size/IO offset was wrong for files > 4GB in size
4779 - object EA data was being dropped, caused files to be unreadable
4780 - memory overflow with non-LOV OST caused memory corruption
4781 - fixed regression tests to work with new config tools, obdfilter
4782 - fixed bug when directory size became larger than 1 block
4783 - fixed bug (for single client case) when PWD was deleted
4784 - invalidate local directory pages when doing intent-based ops
4785 - avoid LDLM oops when lock callback contained bad data
4787 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4789 * documentation updates
4790 * bug fixes and cleanups
4791 - configuration tools
4794 - 64-bit compile warnings
4795 - 64-bit internal statfs data
4797 * test_brw on persistent OST devices
4799 * lprocfs (disabled)
4801 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4803 * documentation updates
4804 * bug fixes and cleanups
4805 * configuration tools
4807 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4810 * many small fixes to 0_5_4
4811 * io/network handling
4812 * thinkos in MDS operations
4814 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4817 * crucial basic fixes to 0.5.3
4818 * IOR, Iozone work over Elan
4821 2002-08-07 Phil Schwan <phil@clusterfs.com>
4822 * version 0_5_3, our first alpha
4823 * we use the new Portals iovs
4824 * documentation updates
4825 * bug fixes and cleanups
4826 * small changes in the DLM wire protocol
4828 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4829 * version 0_5_1 with some initial stability,
4830 * locking on MD and file I/O.
4831 * documentation updates
4832 * several bug fixes since 0.5.0
4833 * small changes in wire protocol
4835 2002-07-18 Phil Schwan <phil@clusterfs.com>
4837 * delivered as Lustre Light Alpha
4838 * fixed a crash after handling invalid MDS requests
4839 * fixed directory pages for architectures with non-4k pages sizes
4841 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4842 * release version v0_4_4
4843 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4845 * Fixes a number of interruption problems with OST operations.
4846 * Update documentation for portals header changes
4847 * Move all wire protocol structs/defines to lustre_idl.h
4848 * Fixes symlink length bug.
4849 * Add tcpdump to repository.
4851 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4852 * release version v0_4_3
4853 * Fixes statfs for inodes on extN.
4854 * Fixes bug in runtests which would delete /etc/hosts.
4855 * Use 64-bit object IDs wherever possible (not into VFS though)
4856 Remove ost_get_info, which is unused by lustre, and out of date.
4858 2002-07-03 Peter Braam <braam@clusterfs.com>
4859 * release version v0_4_2 Fixes a lookup error (type not passed)
4860 * move forward to head of Portals
4861 * move forward to latest Lustre kernel
4863 2002-06-25 Peter Braam <braam@clusterfs.com>
4864 * release version v0_4_1. Hopefully stable on single node use.