1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.14.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.7-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: enable MGS and MDT services start separately
18 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
19 a co-located MGS without starting the MGS, which is a complement
20 to 'nosvc' mount option.
23 Frequency : always with o2ib 1.3 and sles10
25 Description: fix build with SLES10 and o2ib v3.
26 Details : sles10 uses diffrent name for Module.symver file but configure
27 assume this file has same name on RHEL/SLES/vanila kernels.
31 Frequency : after recovery
32 Description: precreate to many object's after del orphan.
33 Details : del orphan st in oscc last_id == next_id and this triger growing
34 count of precreated objects. Set flag LOW to skip increase count
35 of precreated objects.
39 Frequency : rare, on clear nid stats
40 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
41 Details : when clean nid stats sometimes try destroy live entry,
42 and this produce panic in free.
46 Description: Stack overflow during MDS log replay
47 ease stack pressure by using a thread dealing llog_process.
51 Description: wait until IO finished before start new when do lock cancel.
52 Details : VM protocol want old IO finished before start new, in this case
53 need wait until PG_writeback is cleared until check dirty flag and
54 call writepages in lock cancel callback.
56 Severity : enhancement
58 Description: using special macro for print time and cleanup in includes.
62 Description: mds_mfd_close() ASSERTION(rc == 0)
63 Details : In mds_mfd_close(), we need protect inode's writecount change
64 within its orphan write semaphore to prevent possible races.
68 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
73 Description: don't hit live lock with umount ost.
74 Details : shrink_dcache_parent can be in long loop with destroy dentries,
75 use shrink_dcache_sb instead.
79 Description: don't panic with use echo client
80 Details : echo client pass NULL as client nid pointer and this produce null
85 Description: fix build on ppc32
86 Details : compile code with -m64 flag produce wrong object file for ppc32.
90 Description: add message levels for liblustreapi
94 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
95 Details : Move the 'good_osts' check before the 'total_bavail' check. This
96 will result in an -EAGAIN and in the exit call path we call
97 alloc_rr() which will with increasing aggressiveness attempt to
98 aquire precreated objects on the minimum number of required OSCs.
102 Description: Use old size assignment to avoid deadlock
103 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
104 the scheduling while holding a spinlock. We do not need locking
105 for size in ll_update_inode() because size is only updated from
106 the MDS for directories or files without objects, so there is no
107 other place to do the update, and concurrent access to such inodes
108 are protected by the inode lock.
112 Description: resolve "_IOWR redefined" build error on SLES10
116 Description: dump the memory debugging after all modules are unloaded to
117 suppress false negative in conf_sanity test 39
119 Severity : enhancement
121 Description: build kernel-ib packages for OFED 1.3 in our release cycle
126 Description: fix SLES kernel versioning
127 Details : the kernel version for our SLES 10 kernel did not include a "-"
128 before the "smp" at the end. while this was not a problem in
129 general, it did mean that software trying to use the kernel
130 version to try to detect a vendor specific kernel would fail.
131 this was most evident by the OFED build scripts.
135 Description: Don't update lov_desc members until making sure they are valid
136 Details : When updating lov_desc members via proc fs, need fix their
137 validities before doing the real update.
141 Description: don't put request into delay list while invalidate in flight.
142 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
143 import in flight. this produce timeout for invalidate and sometimes
144 can cause stale data.
148 Frequency : on ppc only
149 Description: not convert ost objects for directory because it's not exist.
150 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
154 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
156 Severity : enhancement
158 Description: Update to support 2.6.22.14 vanilla kernel.
162 Frequency : rare, on recovery
163 Description: read procfs can produce deadlock in some situation
164 Details : Holding lprocfs lock which send rpc can produce block for destroy
165 obd objects and this also block reconnect with -EALREADY. This isn't
166 fix all lprocfs bugs - but make it rare.
168 Severity : enhancement
170 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
173 Frequency : frequent on X2 node
175 Description: mdc_set_open_replay_data LBUG
176 Details : Set replay data for requests that are eligible for replay.
180 Description: lustre_mgs: operation 101 on unconnected MGS
181 Details : When MGC is disconnected from MGS long enough, MGS will evict the
182 MGC, and late on MGC cannot successfully connect to MGS and a lot
183 of the error messages complaining that MGS is not connected.
187 Frequency : on network error
188 Description: panic with double free request if network error
189 Details : mdc_finish_enqueue is finish request if any network error ocuring,
190 but it's true only for synchronus enqueue, for async enqueue
191 (via ptlrpcd) this incorrect and ptlrpcd want finish request
194 Severity : enhancement
196 Description: client-side metadata stat-ahead during readdir(directory readahead)
197 Details : perform client-side metadata stat-ahead when the client detects
198 readdir and sequential stat of dir entries therein
201 Frequency : on start mds
203 Description: Implement get_info(last_id) in obdfilter.
206 Frequency : occasional
208 Description: Correctly check stale fid, not start epoch if ost not support SOM
209 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
210 overwrited with new generated fid, but mds can anwer with one of these
211 two fids and both is not stale. setattr incorectly start epoch and
212 assume will be called done_writeting, but without SOM done_writing
216 Frequency : rare, depends on device drivers and load
218 Description: MDS or OSS nodes crash due to stack overflow
219 Details : Code changes in 1.8.0 increased the stack usage of some functions.
220 In some cases, in conjunction with device drivers that use a lot
221 of stack the MDS (or possibly OSS) service threads could overflow
222 the stack. One change which was identified to consume additional
223 stack has been reworked to avoid the extra stack usage.
226 Frequency : occasional
228 Description: Do not fail import if osc_interpret_create gets -EAGAIN
229 Details : If osc_interpret_create got -EAGAIN it immediately exits and
230 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
231 oscc_has_objects and see OSC has no objests and call
232 oscc_internal_create to resend create request.
234 Severity : enhancement
236 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
238 Severity : enhancement
240 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
243 Frequency : very rare
245 Description: Fix replay if there is an un-replied request and open
246 Details : In some cases, older replay request will revert the
247 mcd->mcd_last_xid on MDS which is used to record the client's
250 Severity : enhancement
252 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
254 Severity : enhancement
256 Description: Add rhel5 support to HEAD.
258 Serverity : enhancement
260 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
265 Description: Don't allow skipping OSTs if index has been specified.
266 Details : Don't allow skipping OSTs if index has been specified, make locking
267 in internal create lots better.
271 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
272 Details : access to bitfield in structure is always rounded to long
273 and this produce problem with not atomic change any bit.
277 Description: Lustre make rpms failed.
278 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
279 builds Lustre rpms from tarball.
282 Frequency : rare on shutdown ost
284 Description: If llog cancel was not send before clean_exports phase, this can
285 produce deadlock in llog code.
286 Details : If llog thread has last reference to obd and call class_import_put
287 this produce deadlock because llog_cleanup_commit_master wait when
288 last llog_commit_thread exited, but this never success because was
289 called from llog_commit_thread.
293 Description: allow userland application know is lost one of stripes.
294 Details : fill lvb_blocks with error code on ost and return it to
295 application if error flag found.
299 Description: NULL lov_tgts causing MDS oops
300 Details : more safe checks for NULL lov_tgts for avoid oops.
302 Severity : enhancement
304 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
308 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
309 Details : If an OST is down the MDS will hang indefinitely in
310 obd_statfs_rqset() waiting for the statfs data. While for
311 MDS QOS usage of statfs, it should not stuck in waiting.
313 Severity : enhancement
315 Description: remote_acl support
316 Details : Support ACL-based permission check for remote user.
317 Support setfacl/getfacl for remote user with the utils
318 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
319 the system "{s,g}etfacl" utils.
321 Severity : enhancement
323 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
325 Severity : enhancement
327 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
331 Description: make mgs_setparam() handle fsname containing dash
332 Details : fsname containing a dash does not work with lctl conf_param
334 Severity : enhancement
336 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
338 Severity : enhancement
340 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
344 Frequency : rare, at shutdown
345 Description: access already free / zero obd_namespace.
346 Details : if client_disconnect_export was called without force flag set,
347 and exist connect request in flight, this can produce access to
348 NULL pointer (or already free pointer) when connect_interpret
349 store ocd flags in obd_namespace.
353 Frequency : only at startup
354 Description: not alloc memory with spinlock held.
355 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
356 if any spinlock held.
358 Severity : enhancement
360 Description: make lustre randomly fail allocating memory
361 Details : Make lustre randomly failed allocating memory for testing purpose.
363 Severity : enhancement
365 Description: lost problems with lov objid file
366 Details : Fixes some scability and access to not inited memory problems
367 in work with lov objdid file.
372 Description: lfs find does not continue on file error
373 Details : Continue other files processing when a file/dir is absent.
377 Description: Inconsistent usage of lustre_pack_reply()
378 Details : Standardize the usage of lustre_pack_reply() such that it
379 always generate a CERROR on failure.
382 Frequency : occasional
384 Description: MDS hang or stay in waiting lock
385 Details : If client receive lock with CBPENDING flag ldlm need send lock
386 cancel as separate rpc, to avoid situation when cancel request
387 can't processed due all i/o threads stay in wait lock.
391 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
395 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
399 Description: Update to SLES9 kernel-2.6.5-7.287.3.
403 Description: Update to RHEL4 latest kernel.
405 Severity : enhancement
407 Description: Build SLES10 patchless client fails
408 Details : The configure was broken by run ./configure with
409 --with-linux-obj=.... argument for patchless client. When the
410 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
411 can't search header adequately. Use absolute path such as
412 -I($LINUX)/include instead.
416 Description: interrupt oig_wait produce painc on resend.
417 Details : brw_redo_request can be used for resend requests from ptlrpcd and
418 private set, and this produce situation when rq_ptlrpcd_data not
419 copyed to new allocated request and triggered LBUG on assert
420 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
421 ptlrpcd set if request is changed and can be safety changed to use
426 Description: LASSERT_{REQ,REP}SWAB macros are buggy
427 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
428 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
429 these macros and replace them with their difinitions instead.
433 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
434 Details : Update which_patch & target file for SLES10 latest kernel.
437 Description: Remove client patches from SLES 10 kernel.
438 Details : This causes SLES 10 clients to behave as patchless clients
439 even on a Lustre-patched (server) kernel.
441 Severity : enhancement
443 Description: self-adjustable client's lru lists
444 Details : use adaptive algorithm for managing client cached locks lru
445 lists according to current server load, other client's work
446 pattern, memory activities, etc. Both, server and client
447 side namespaces provide number of proc tunables for controlling
450 Severity : enhancement
452 Description: light-weight GSS support
453 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
454 and reduce performance overhead.
456 Severity : enhancement
458 Description: Linux keyring support
459 Details : Support using service of Linux keyring for Lustre GSS internal
460 context refresh/cache mechanism.
464 Description: Fix errors in lfs documentation
465 Details : Fixes man pages
469 Description: don't use GFP_* in generic Lustre code.
470 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
474 Description: obdclass is limited by single OBD_ALLOC(idarray)
475 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
479 Description: warnings with build patchless client with vanila 2.6.19 and up
480 Details : change old ctl_table style and replace ctl_table/ctl_table_header
481 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
485 Frequency : Only for SLES
486 Description: sanity_quota fail test_1
487 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
488 file, which makes TSTID[2] inunique.
492 Frequency : Only for SLES9
493 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
494 Details : Update target/ChangeLog/which_patch .
498 Frequency : Only for test_10 in sanity-quota.sh
499 Description: a bug in quota test code
500 Details : checking a test flag in a wrong place
504 Frequency : Only for test_13 in sanity-quota.sh
505 Description: a bug in quota test script
506 Details : shouldn't check overfull allocated quota
510 Description: jbd statistics
511 Details : Port older jbd statistics patch for sles10
513 Severity : enhancement
515 Description: Add printing inode info into message about error in writepage.
520 Description: reply_lock_interpret crash due to race with it and lock cancel.
521 Details : Do not replay locks that are being cancelled. Do not reference
522 locks by their address during replay, just by their handle.
526 Frequency : When flocks are used.
527 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
528 Details : Flock locks might destroy just granted lock if it could be merged
529 with another existing flock, this is done in completion handler,
530 so teach ldlm_cli_enquque_fini that this is a valid case for
536 Description: Oops in read and write path when failing to allocate lock.
537 Details : Check if lock allocation failed and return error back.
541 Description: lstripe command fails for valid OST index
542 Details : The stripe offset is compared to lov->desc.ld_tgt_count
543 instead of lov->desc.ld_active_tgt_count.
547 Description: sanity.sh failed test 103
548 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
553 Description: df doesn't work properly if diskfs blocksize != 4K
554 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
558 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
559 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
563 Description: obd_health_check_timeout too short
564 Details : set obd_health_check_timeout as 1.5x of obd_timeout
568 Description: llapi_file_create() does not allow some changes
569 Details : add llapi_file_open() that allows specifying the mode and
570 open flags, and also returns an open file handle.
574 Description: merge and cleanup kernel patches.
575 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
579 Description: Add journal checksum support.(Kernel part)
580 Details : The journal checksum feature adds two new flags i.e
581 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
582 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
583 indicates that the commit block contains the checksum for
584 the blocks described by the descriptor blocks. Now commit
585 record can be sent to disk without waiting for descriptor
586 blocks to be written to disk. This behavior is controlled
587 using JBD2_FEATURE_ASYNC_COMMIT flag.
591 Description: OSS needs mutliple precreate threads
592 Details : Add ability to start more than one create thread per OSS.
596 Description: Sanity.sh test_65k failed: setstripe should have succeeded
597 Details : execute lfs setstripe on client
601 Description: mds_obd_create error creating tmp object
602 Details : When the user sets quota on root, llog will be affected and can't
603 create files and write files.
606 Frequency : Always on ia64 patchless client, and possibly others.
608 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
609 Details : This allows the patchless client to be loaded on architectures
614 Description: RedHat Update kernel for RHEL5
615 Details : Add the kernel config file for RHEL5.
619 Description: RedHat Update kernel for RHEL5
620 Details : Modify the kernel config file more closer RHEL5.
624 Description: Build failure against Centos5 (RHEL5)
625 Details : Define PAGE_SIZE when it isn't present.
629 Description: test_7 on sanity-quota.sh can't be used on separate machines
630 Details : add facet to handle it
634 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
635 Details : with NFS, the anon dentry's parent was set to itself in
636 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
637 corresponding dentry other than use rec->ur_name.
639 Severity : enhancement
641 Description: lfs setstripe enhancement
642 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
646 Description: enable data checksumming by default
647 Details : enable checksum by default, allow --disable-checksum
648 configure option and "-o nochecksum" mount option. Checksums
649 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
653 Description: lustre support for RHEL5
654 Details : Add support for RHEL5.
658 Description: Client eviction due to failover config
659 Details : after a connection loss, the lustre client should attempt to
660 reconnect to the last active server first before trying the
661 other potential connections.
665 Description: when mds and osts use different quota unit(32bit and 64bit),
666 quota will be released repeatly.
667 Details : void sending multiple quota reqs to mds, which will keep the
668 status between the reqs.
672 Description: osts not allocated evenly to files
673 Details : change the condition to increase offset_idx
676 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
678 Description: Data corruption for OSTs that are formatted larger than 2TB
680 Details : When generating the bio request for lustre file writes the
681 sector number would overflow a temporary variable before being
682 used for the IO. The data reads correctly from Lustre (which
683 will overflow in a similar manner) but other file data or
684 filesystem metadata may be corrupted in some cases.
688 Description: Tune the kernel for good SCSI performance.
689 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
690 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
695 Description: rewrite ext2-derived code in obdclass/uuid.c
696 Details : rewrite inherited code (uuid parsing code from ext2 utils)
697 from scratch preserving functionality.
701 Description: "lfs find -obd UUID" prints directories
702 Details : "lfs find -obd UUID" will return all directory names instead
703 of just file names. It is incorrect because the directories
704 do not reside on the OSTs.
708 Description: use i_size_read and i_size_write in 2.6 port
709 Details : replace inode->i_size access with i_size_read/write()
712 Frequency : when removing large files
714 Description: scheduling issue during removal of large Lustre files
715 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
716 It causes scheduling issues when removing large files (17TB in the
720 Frequency : only with liblustre clients on XT3
722 Description: evictions taking too long
723 Details : allow llrd to evict clients directly on OSTs
726 Frequency : only on ppc
728 Description: /proc/fs/lustre/devices broken on ppc
729 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
730 We should make sure the type of 'index' is loff_t and then cast
731 explicitly as needed below. Do not assign an explicitly cast
735 Frequency : only for rhel5
737 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
738 Details : Modify the target file & which_kernel.
740 Severity : enhancement
742 Description: omit set fsid for export NFS
743 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
744 as FS_REQUIRES_DEV for avoid problems with generate fsid.
748 Frequency : Always, for kernels after 2.6.16
749 Description: Fix warning idr_remove called for id=.. which is not allocated.
750 Details : Last kernels save old s_dev before kill super and not allow
751 to restore from callback - restore it before call kill_anon_super.
755 Description: buffer overruns could theoretically occur
756 Details : llapi_semantic_traverse() modifies the "path" argument by
757 appending values to the end of the origin string, and a
758 overrun may occur. Adding buffer overrun check in liblustreapi.
762 Description: Fix error on 'ls .' at the top of the Lustre mount.
763 Details : Don't revalidate dentry if it is a root dentry.
767 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
768 Details : Modify vm-tunables-rhel4.patch.
772 Description: Kernel config for 2.6.18-vanilla.
773 Details : Modify targets/2.6-vanilla.target.in.
774 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
775 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
776 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
777 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
781 Description: improve handling recoverable errors
782 Details : if request processig with error which can be recoverable on server
783 request should be resend, otherwise page released from cache and
788 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
789 Details : Modify target file & which_patch.
790 A flaw was found in the IA32 system call emulation provided
791 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
792 value could be stored in the %RAX register, which could trigger an
793 out-of-bounds system call table access. An untrusted local user
794 could exploit this flaw to run code in the kernel
795 (ie a root privilege escalation). (CVE-2007-4573).
799 Description: O_DIRECT bypasses client statistics.
800 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
801 read_ahead_stats were updated. copied the stats section from
802 osc_send_oap_rpc() into async_internal().
806 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
810 Description: change order of libsysio includes
811 Details : '#include sysio.h' should always come before '#include xtio.h'
815 Description: handle "serious error: objid * already exists" more gracefully
816 Details : If LAST_ID value on disk is smaller than the objects existing in
817 the O/0/d* directories, it indicates disk corruption and causes an
818 LBUG(). If the object is 0-length, then we should use the existing
819 object. This will help to avoid a full fsck in most cases.
821 Severity : enhancement
823 Description: adapt the lustre_config script to support the upgrade case
824 Details : Add "-u" option for lustre_config script to support upgrading 1.4
825 server targets to 1.6 in parallel.
829 Description: To avoid grant space > avaible space when the disk is almost
830 full. Without this patch you might see the error "grant XXXX >
831 available" or some LBUG about grant, when the disk is almost
833 Details : In filter_check_grant, for non_grant cache write, we should
834 check the left space by if (*left > ungranted + bytes), instead
835 of (*left > ungranted), because only we are sure the left space
836 is enough for another "bytes", then the ungrant space should be
837 increase. In client, we should update cl_avail_grant only there
838 is OBD_MD_FLGRANT in the reply.
842 Description: Update RHEL 4 kernel to fix local root privilege escalation.
843 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
844 described in CVE-2007-4573. This problem could allow untrusted
845 local users to gain root access.
848 Frequency : when using O_DIRECT and quotas
850 Description: Incorrect file ownership on O_DIRECT output files
851 Details : block usage reported by 'lfs quota' does not take into account
852 files that have been written with O_DIRECT.
857 Description: touch file failed when fs is not full
858 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
859 otherwise we can get ENOSP while fs is not full.
863 Description: parallel lock callbacks
864 Details : Instead of sending blocking and completion callbacks as separated
865 requests, adding them to a set and sending in parallel.
868 Frequency : only for Cray XT3
869 Bugzilla : 12829/13455
870 Description: Changing primary group doesn't change the group lustre assigns to
872 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
873 overridden with the primary group provided by the group upcall,
874 whereas the supplied fsgid can be trusted if it is in the list of
875 supplementary groups returned by the group upcall.
877 Severity : enhancement
879 Description: Allow masking D_WARNING, D_ERROR messages from console
880 Details : Console messages can now be disabled via lnet.printk.
884 Description: User code with malformed file open parameter crashes client node
885 Details : Before packing join_file req, all the related reference should be
886 checked carefully in case some malformed flags cause fake
887 join_file req on client.
891 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
892 Details : ldlm_completion_ast() assumes that a lock is granted when the req
893 mode is equal to the granted mode. However, it should also check
894 that LDLM_FL_CP_REQD is not set.
898 Description: Heavy nfs access might result in deadlocks
899 Details : After ELC code landed, it is now improper to enqueue any mds
900 locks under och_sem, because enqueue might want to decide to
901 cancel open locks for same inode we are holding och_sem for.
905 Description: Client eviction while running blogbench
906 Details : A lot of unlink operations with concurrent I/O can lead to a
907 deadlock causing evictions. To address the problem, the number of
908 oustanding OST_DESTROY requests is now throttled to
909 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
910 callbacks are processed in priority.
914 Description: enable ACLs on MDS by default
915 Details : ACLs must be enabled on MDS by default.
918 Frequency : PPC/PPC64 only
920 Description: conflicts between asm-ppc64/types.h and lustre_types.h
921 Details : fix duplicated definitions between asm-ppc64/types.h and
922 lustre_types.h on PPC.
925 Frequency : PPC/PPC64 only
927 Description: asm-ppc/segment.h does not exist
928 Details : fix compile issue on PPC.
932 Description: better handle error messages in extents code
935 Frequency : RHEL4 only
937 Description: mkfs is very slow on IA64/RHEL4
938 Details : A performance regression has been discovered in the MPT Fusion
939 driver between versions 3.02.73rh and 3.02.99.00rh. As a
940 consequence, we have downgraded the MPT Fusion driver in the RHEL4
941 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
943 Severity : enhancement
945 Description: SNMP support enhancement
946 Details : Adding total number of sampled request for an MDS node in snmp
949 Severity : enhancement
951 Description: Optimize ldlm waiting list processing for PR extent locks
952 Details : When processing waiting list for read extent lock and meeting read
953 lock that is same or wider to it that is not contended, skip
954 processing rest of the list and immediatelly return current
955 status of conflictness, since we are guaranteed there are no
956 conflicting locks in the rest of the list.
960 Description: Time out and refuse to reconnect
961 Details : When the failover node is the primary node, it is possible
962 to have two identical connections in imp_conn_list. We must
963 compare not conn's pointers but NIDs, otherwise we can defeat
964 connection throttling.
968 Description: port llog fixes from b1_6 into HEAD
969 Details : Port llog reference couting and some llog cleanups from b1_6
970 (bug 10800) into HEAD, for protect from panic and access to already
971 free llog structures.
975 Description: Detect stride IO mode in read-ahead
976 Details : When a client does stride read, read-ahead should detect that and
977 read-ahead pages according to the detected stride pattern.
981 Description: data checksumming impacts single node performance
982 Details : add support for several checksum algorithm. Currently, only CRC32
983 and Adler-32 are supported. The checksum type can be changed on
984 the fly via /proc/fs/lustre/osc/*/checksum_type.
988 Description: use adler32 for page checksums
989 Details : when available, use the Adler-32 algorithm instead of CRC32 for
994 Description: build for x2 fails
995 Details : fix compile issue on Cray systems.
999 Description: Properly match for duplicate locks
1000 Details : Due to different lock order from skiplists code, we need to
1001 traverse entire list for now
1004 Frequency : only on PPC/SLES10
1006 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1007 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1008 be defined. Add a hack in mkfs_lustre.c to work around this compile
1013 Description: LASSERT on MDS when client holding flock lock dies
1014 Details : ldlm pool logic depends on number of granted locks equal to
1015 number of released locks which is not true for flock locks, so
1016 just exclude such locks from consideration.
1020 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1021 Details : Use fsfilt_sync() for both the whole filesystem sync and
1022 individual file sync to eliminate dangerous inode locking
1023 with I_LOCK that can lead to a deadlock.
1027 Description: performance in 1.6.3
1028 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1029 sure that 1MB requests are not fragmented by the block layer.
1031 Severity : enhancement
1033 Description: organize the server-side client stats on per-nid basis
1034 Details : Change the structure of stats under obdfilter and mds to
1042 The "uuid"s file would list the uuids of _active_ exports.
1043 And the clear entry is to clear all stats and stale nids.
1045 Severity : enhancement
1047 Description: eliminate client locks in face of contention
1048 Details : file contention detection and lockless i/o implementation
1049 for contended files.
1053 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1056 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1058 Description: MDS may lose file striping (and hence file data) in some cases
1059 Details : If there are additional extended attributes stored on the MDS,
1060 in particular ACLs, SELinux, or user attributes (if user_xattr
1061 is specified for the client mount options) then there is a risk
1062 of attribute loss. Additionally, the Lustre file striping
1063 needs to be larger than default (e.g. striped over all OSTs),
1064 and an additional attribute must be stored initially in the
1065 inode and then increase in size enough to be moved to the
1066 external attribute block (e.g. ACL growing in size) for file
1071 Description: skiplist implementation simplification
1072 Details : skiplists are used to group compatible locks on granted list
1073 that was implemented as tracking first and last lock of each lock group
1074 the patch changes that to using doubly linked lists
1076 --------------------------------------------------------------------------------
1078 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1080 * Support for kernels:
1081 2.6.5-7.283 (SLES 9),
1082 2.6.9-55.EL (RHEL 4),
1083 2.6.16.46-0.14 (SLES 10),
1084 2.6.18.8 vanilla (kernel.org)
1085 * Client support for unpatched kernels:
1086 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1087 2.6.16 - 2.6.22 vanilla (kernel.org)
1088 * Due to recently discovered recovery problems, we do not recommend
1089 using patchless RHEL 4 clients with this or any earlier release.
1090 * Recommended e2fsprogs version: 1.39.cfs8
1091 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1092 * Starting with this release, the ldiskfs backing filesystem required
1093 by Lustre is now in its own package, lustre-ldiskfs. This package
1094 should be installed. It is versioned separately from Lustre and
1095 may be released separately in future.
1099 Description: block reactivating mgc import until all deactivates complete
1100 Details : Fix race when failing back MDT/MGS to itself (testing)
1103 Frequency : at statup only
1105 Description: mds_lov_synchronize race leads to various problems
1106 Details : simultaneous MDT->OST connections at startup can cause the
1107 sync to abort, leaving the OSC in a bad state.
1109 Severity : enhancement
1111 Description: add optional extra BUILD_VERSION info
1112 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1113 to override the lustre version.
1116 Frequency : 2.6.18 servers only
1118 Description: ll_kern_mount() doesn't release the module reference
1119 Details : The ldiskfs module reference count never drops down to 0
1120 because ll_kern_mount() doesn't release the module reference.
1125 Description: server LBUG when using old ost_num_threads parameter
1126 Details : Accept the old ost_num_threads parameter but warn that it
1127 is deprecated, and fix an off-by-one error that caused an LBUG.
1132 Description: Transient SCSI error results in persistent IO issue
1133 Details : iobuf->dr_error is not reinitialized to 0 between two
1137 Frequency : sometimes when underlying device returns I/O errors
1139 Description: OSTs not going read-only during write failures
1140 Details : OSTs are not remounted read-only when the journal commit threads
1141 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1142 instead of the ext3 wrappers.
1146 Description: poor connect scaling with increasing client count
1147 Details : Don't run filter_grant_sanity_check for more than 100 exports
1148 to improve scaling for large numbers of clients.
1151 Frequency : SLES10 only
1153 Description: sanity-quota.sh quotacheck failed: rc = -22
1154 Details : Quotas cannot be enabled on SLES10.
1157 Frequency : liblustre clients only
1159 Description: getdirentries does not give error when run on compute nodes
1160 Details : getdirentries does not fail when the size specified as an argument
1161 is too small to contain at least one entry
1163 Severity : enhancement
1165 Description: Add LNET router traceability for debug purposes
1166 Details : If a checksum failure occurs with a router as part of the
1167 IO path, the NID of the last router that forwarded the bulk data
1168 is printed so it can be identified.
1173 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1174 Details : Due to a race condition, liblustre clients were occasionally
1175 evicted incorrectly.
1177 Severity : enhancement
1179 Description: lfs setstripe use optional parameters instead of postional
1182 Severity : enhancement
1184 Description: Nanosecond timestamp support for ldiskfs
1185 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1186 resolution timestamps. There is not yet support for this at
1187 the Lustre filesystem level.
1190 Frequency : during server recovery
1192 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1193 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1197 Frequency : nfs export on patchless client
1199 Description: connectathon hang when test nfs export over patchless client
1200 Details : Disconnected dentry cannot be found with lookup, so we do not need
1201 to unhash it or make it invalid
1204 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1205 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1206 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1211 Description: open req refcounting wrong on reconnect
1212 Details : If reconnect happened between getting open reply from server and
1213 call to mdc_set_replay_data in ll_file_open, we will schedule
1214 replay for unreferenced request that we are about to free.
1215 Subsequent close will crash in variety of ways.
1216 Check that request is still eligible for replay in
1217 mdc_set_replay_data().
1222 Description: disable writes to filesystem when reading health_check file
1223 Details : the default for reading the health_check proc file has changed
1224 to NOT do a journal transaction and write to disk, because this
1225 can cause reads of the /proc file to hang and block HA state
1226 checking on a healthy but otherwise heavily loaded system. It
1227 is possible to return to the previous behaviour during configure
1228 with --enable-health-write.
1230 Severity : enhancement
1232 Description: 64-bit inode version
1233 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1234 to the inode. This will be required for version-based recovery.
1239 Description: MDS fails to start if a duplicate client export is detected
1240 Details : in some rare cases it was possible for a client to connect to
1241 an MDS multiple times. Upon recovery the MDS would detect this
1242 and fail during startup. Handle this more gracefully.
1244 Severity : enhancement
1246 Description: Add -o localflock option to simulate old noflock
1248 Details : This will achieve local-only flock/fcntl locks
1254 Description: log_commit_thread vs filter_destroy race leads to crash
1255 Details : Take import reference before releasing llog record semaphore
1260 Description: Wrong request locking in request set processing
1261 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1262 imp_delayed_list, in this place should be used imp_lock.
1265 Frequency : when reconnection
1267 Description: Grant Leak when osc reconnect to OST
1268 Details : When osc reconnect ost, OST(filter) should check whether it
1269 should grant more space to client by comparing fed_grant and
1270 cl_avail_grant, and return the granted space to client instead
1271 of "new granted" space, because client will call osc_init_grant
1272 to update the client grant space info.
1275 Frequency : when client reconnect to OST
1277 Description: Grant Leak when osc do resend and replay bulk write
1278 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1279 bulk write request, because the grant info while be sync between
1280 OSC and OST when reconnect, and we should ignore the grant info
1281 these of resend/replay write req.
1286 Description: Grant space more than avaiable left space sometimes.
1287 Details : When then OST is about to be full, if two bulk writing from
1288 different clients came to OST. Accord the avaliable space of the
1289 OST, the first req should be permitted, and the second one
1290 should be denied by ENOSPC. But if the seconde arrived before
1291 the first one is commited. The OST might wrongly permit second
1292 writing, which will cause grant space > avaiable space.
1295 Frequency : when client is evicted
1297 Description: Grant might be wrongly erased when osc is evicted by OST
1298 Details : when the import is evicted by server, it will fork another
1299 thread ptlrpc_invalidate_import_thread to invalidate the
1300 import, where the grant will be set to 0. While the original
1301 thread will update the grant it got when connecting. So if
1302 the former happened latter, the grant will be wrongly errased
1303 because of this race.
1308 Description: Checking Stale with correct fid
1309 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1310 to check whether it is stale, because sometimes, we want the
1311 enqueue happened anyway, and op_data.fid2 will not be initialized.
1313 Severity : enhancement
1315 Description: update patchless client
1316 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1319 Frequency : only with 2.4 kernel
1321 Description: random memory corruption
1322 Details : size of struct ll_inode_info is to big for union inode.u and this
1323 can be cause of random memory corruption.
1328 Description: Memory leak in recovery
1329 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1330 It should also check obd_fail before fsfilt_start, otherwise if
1331 fsfilt_start return -EROFS,(failover mds during mds recovery).
1332 then the req will return with repmsg->transno = 0 and rc = EROFS.
1333 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1334 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1335 be freed no matter whether fsfilt_commit success or not.
1338 Frequency : only with huge count clients
1340 Description: Prevents from taking the superblock lock in llap_from_page for
1342 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1343 for llap_from_page call in ll_removepage prevents from taking the
1344 superblock lock for a soon died page.
1349 Description: Not check open intent error before release open handle
1350 Details : in some rare cases, the open intent error is not checked before
1351 release open handle, which may cause
1352 ASSERTION(open_req->rq_transno != 0), because it tries to release
1353 the failed open handle.
1358 Description: Set cat log bitmap only after create log success.
1359 Details : in some rare cases, the cat log bitmap is set too early. and it
1360 should be set only after create log success.
1364 Description: Accessing a block bevice can re-enable I/O when Lustre is
1365 tearing down a device.
1366 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1371 Description: service threads may hog cpus when there are a lot of requests
1372 Details : Insert cond_resched to give other threads a chance to use some CPU
1377 Description: the cat log was not initialized in recovery
1378 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1379 unlink log on mds will not be initialized until mds post
1380 recovery. And also in mds post recovery, the unlink log will
1381 initialization will be done asynchronausly, so there will be race
1382 between add unlink log and unlink log initialization.
1386 Description: brw_stats were being printed incorrectly
1387 Details : brw_stats were being printed as log2 but all of them were not
1388 recorded as log2. Also remove some code duplication arising from
1389 filter_tally_{read,write}.
1393 Frequency : rare, only in recovery.
1394 Description: ASSERTION(req->rq_type != LI_POISON) failed
1395 Details : imp_lock should be held while iterating over imp_sending_list for
1396 prevent destroy request after get timeout in ptlrpc_queue_wait.
1400 Description: replay-single.sh test 52 fails
1401 Details : A lock's skiplist need to be cleanup when it being unlinked
1402 from its resource list.
1406 Description: Short directio read returns full requested size rather than
1408 Details : Direct I/O operations should return actual amount of bytes
1409 transferred rather than requested size.
1411 Severity : enhancement
1413 Description: metadata RPC reduction (e.g. for rm performance)
1414 Details : decrease the amount of synchronous RPC between clients and servers
1415 by canceling conflicing lock before the operation on the client
1416 and packing thier handles into the main operation RPC to server.
1418 Severity : enhancement
1420 Description: Async OSC create to avoid the blocking unnecessarily.
1421 Details : If a OST has no remain object, system will block on the creating
1422 when need to create a new object on this OST. Now, ways use
1423 pre-created objects when available, instead of blocking on an
1424 empty osc while others are not empty. If we must block, we block
1425 for the shortest possible period of time.
1427 Severity : enhancement
1429 Description: refine locking for avoid write wrong info into lov_objid file
1430 Details : fix possible races with add new target and write/update data in
1433 --------------------------------------------------------------------------------
1435 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1440 Frequency : on some architectures
1442 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1443 Details : Uninitialized flags sometimes cause configuration commands to
1447 Frequency : patchless clients only
1449 Description: missing __iget() symbol export
1450 Details : The __iget() symbol export is missing. To avoid the need for
1451 this on patchless clients the deathrow inode reaper is turned
1452 off, and we depend on the VM to clean up old inodes. This
1453 dependency was during via the fix for bug 12181.
1455 --------------------------------------------------------------------------------
1457 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1459 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1460 INTEROPERATE with older versions automatically. In many cases a
1461 special upgrade step is needed. Please read the
1462 user documentation before upgrading any part of a live system.
1463 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1464 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1465 * WARNING: Lustre configuration and startup changes are required with
1466 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1472 Severity : enhancement
1474 Description: MountConf
1475 Details : Lustre configuration is now managed via mkfs and mount
1476 commands instead of lmc and lconf. New obd types (MGS, MGC)
1477 are added for dynamic configuration management. See
1478 https://mail.clusterfs.com/wikis/lustre/MountConf for
1481 Severity : enhancement
1483 Description: dynamic OST addition
1484 Details : OSTs can now be added to a live filesystem
1486 Severity : enhancement
1488 Description: startup order invariance
1489 Details : MDTs and OSTs can be started in any order. Clients only
1490 require the MDT to complete startup.
1492 Severity : enhancement
1494 Description: parallel, asynchronous orphan cleanup
1495 Details : orphan cleanup is now performed in separate threads for each
1496 OST, allowing parallel non-blocking operation.
1498 Severity : enhancement
1500 Description: optimized stripe assignment
1501 Details : stripe assignments are now made based on ost space available,
1502 ost previous usage, and OSS previous usage, in order to try
1503 to optimize storage space and networking resources.
1505 Severity : enhancement
1507 Description: Permanently set tunables
1508 Details : All writable /proc/fs/lustre tunables can now be permanently
1509 set on a per-server basis, at mkfs time or on a live system.
1511 Severity : enhancement
1513 Description: Lustre message v2
1514 Details : Add lustre message format v2.
1516 Severity : enhancement
1518 Description: client OST exclusion list
1519 Details : Clients can be started with a list of OSTs that should be
1520 declared "inactive" for known non-responsive OSTs.
1524 Description: SPEC SFS validation failure on NFS v2 over lustre.
1525 Details : Changes the blocksize for regular files to be 2x RPC size,
1526 and not depend on stripe size.
1528 Severity : enhancement
1530 Description: Multiple MD RPCs in flight.
1531 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1532 To avoid overly-overloading MDS, introduce a limit on number of
1533 MDS RPCs in flight for a single client and add /proc controls
1534 to adjust this limit.
1536 Severity : enhancement
1538 Description: client read/write statistics
1539 Details : Add client read/write call usage stats for performance
1540 analysis of user processes.
1541 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1542 file access. extents_stats shows chunk size distribution.
1543 extents_stats_per_process show chunk size distribution per
1546 Severity : enhancement
1548 Description: mds statistics
1549 Details : Add detailed mds operations statistics in
1550 /proc/fs/lustre/mds/*/stats.
1554 Description: Failure of copying files with lustre special EAs.
1555 Details : Client side always return success for setxattr call for lustre
1556 special xattr (currently only "trusted.lov").
1561 Description: Refcount LNET uuids
1562 Details : The global LNET uuid list grew linearly with every startup;
1563 refcount repeated list entries instead of always adding to
1566 Severity : enhancement
1568 Description: Dynamic service threads
1569 Details : Within a small range, start extra service threads
1570 automatically when the request queue builds up.
1573 Frequency : mixed-endian client/server environments
1575 Description: mixed-endian crashes
1576 Details : The new msg_v2 system had some failures in mixed-endian
1579 Severity : enhancement
1581 Description: Easy OST removal
1582 Details : OSTs can be permanently deactivated with e.g. 'lctl
1583 conf_param lustre-OST0001.osc.active=0'
1585 Severity : enhancement
1587 Description: MGS proc entries
1588 Details : Added basic proc entries for the MGS showing what filesystems
1591 Severity : enhancement
1593 Description: provide MGS failover
1594 Details : Added config lock reacquisition after MGS server failover.
1596 Severity : enhancement
1598 Description: add Linux 2.4 support
1599 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1603 Description: a large application tries to do I/O to the same resource and dies
1604 in the middle of it.
1605 Details : Check the req->rq_arrival time after the call to
1606 ost_brw_lock_get(), but before we do anything about
1607 processing it & sending the BULK transfer request. This
1608 should help move old stale pending locks off the queue as
1609 quickly as obd_timeout.
1612 Frequency : when an incorrect nid is specified during startup
1614 Description: ptlrpc connect to non-existant node causes kernel crash
1615 Details : LNET can't be re-entered from an event callback, which
1616 happened when we expire a message after the export has been
1617 cleaned up. Instead, hand the zombie cleanup off to another
1620 Severity : enhancement
1622 Description: plain/inodebits lock performance improvement
1623 Details : Grouping plain/inodebits in granted list by their request modes
1624 and bits policy, thus improving the performance of search through
1628 Frequency : only if OST filesystem is corrupted
1630 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1631 Details : for a short time RPCs with bulk IO are in the replay list,
1632 but replay of bulk IOs is unimplemented. If the OST filesystem
1633 is corrupted due to disk cache incoherency and then replay is
1634 started it is possible to trip an assertion. Avoid putting
1635 committed RPCs into the replay list at all to avoid this issue.
1638 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1640 Description: System hang on startup
1641 Details : This bug allowed the liblustre (e.g. catamount) client to
1642 return to the app before handling all startup RPCs. This
1643 could leave the node unresponsive to lustre network traffic
1644 and manifested as a server ptllnd timeout.
1646 Severity : enhancement
1648 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1649 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1650 boolean to control whether to print peer debug info when a
1651 client's RPC times out.
1654 Frequency : only for kernels with patches from Lustre below 1.4.3
1656 Description: Remove old rdonly API
1657 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1660 Frequency : only for devices with external journals
1662 Description: Set external device read-only also
1663 Details : During a commanded failover stop, we set the disk device
1664 read-only while the server shuts down. We now also set any
1665 external journal device read-only at the same time.
1668 Frequency : when upgrading from 1.4 while trying to change parameters
1670 Description: The wrong (new) MDC name was used when setting parameters for
1671 upgraded MDT's. Also allows changing of OSC (and MDC)
1672 parameters if --writeconf is specified at tunefs upgrade time.
1675 Frequency : when setting specific ost indicies
1677 Description: QOS code breaks on skipped indicies
1678 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1679 created with --index need not be sequential.
1681 Severity : enhancement
1683 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1684 Details : The uninit_groups feature works in conjunction with the kernel
1685 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1686 up the pass1 processing of e2fsck. This is a read-only feature
1687 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1688 mount filesystems that have had this feature enabled.
1690 Severity : enhancement
1692 Description: Improve multi-block allocation algorithm to avoid fragmentation
1693 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1694 allocation locality and avoid filesystem fragmentation.
1696 ------------------------------------------------------------------------------
1698 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1700 * Support for kernels:
1701 2.4.21-47.0.1.EL (RHEL 3)
1702 2.6.5-7.283 (SLES 9)
1703 2.6.9-42.0.10.EL (RHEL 4)
1704 2.6.12.6 vanilla (kernel.org)
1705 2.6.16.27-0.9 (SLES 10)
1706 * Recommended e2fsprogs version: 1.39.cfs5
1708 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1712 Frequency : occasional, depends on client load and configuration
1713 Bugzilla : 12181, 12203
1714 Description: data loss for recently-modified files
1716 Details : In some cases it is possible that recently written or created
1717 files may not be written to disk in a timely manner (this should
1718 normally be within 30s unless client IO load is very high).
1719 The problem appears as zero-length files or files that are a
1720 multiple of 1MB in size after a client crash or client eviction
1721 that are missing data at the end of the file.
1723 This problem is more likely to be hit on clients where files are
1724 repeatedly created and unlinked in the same directory, clients
1725 have a large amount of RAM, have many CPUs, the filesystem has
1726 many OSTs, the clients are rebooted frequently, and/or the files
1727 are not accessed by other nodes after being written.
1729 The presence of the problem can be detected by looking at
1730 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1731 smaller than the second (nr_unused) then dirty files will not
1732 be flushed automatically to disk. "sync; sleep 10" should be
1733 run several times on the node before unmounting it to update
1734 Lustre (this is also safe to run on nodes without this problem).
1736 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1737 that can cause this same problem, so customers using that kernel
1738 also need to update the kernel in addition to Lustre. In order
1739 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1741 It is normal that files written just before a client crash (less
1742 than 30s) may not yet have been flushed to disk, even for local
1746 Frequency : frequent on thin XT3 nodes
1748 Description: UUID collision on thin XT3 Linux nodes
1749 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1750 randomly, since we relied on an insufficiently-seeded PRNG.
1755 Description: OSS hangs after "All ost request buffers busy"
1756 Details : A deadlock between quota and journal operations caused OSS
1757 hangs after printing "All ost request buffers busy."
1760 Frequency : always on liblustre builds
1762 Description: Cleanup compiler warnings on liblustre
1765 Frequency : always on liblustre builds on XT3
1767 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1771 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1772 > 32000 subdirectories
1773 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1774 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1775 subdirectory count crosses 32000. This will aid e2fsck to
1776 correctly handle more than 32000 subdirectories.
1779 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1781 Description: System hang on startup
1782 Details : This bug allowed the liblustre (e.g. catamount) client to
1783 return to the app before handling all startup RPCs. This
1784 could leave the node unresponsive to lustre network traffic
1785 and manifested as a server ptllnd timeout.
1787 Severity : enhancement
1789 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1790 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1791 boolean to control whether to print peer debug info when a
1792 client's RPC times out.
1797 Description: make O_SYNC working on 2.6 kernels
1798 Details : 2.6 kernels use different method for mark pages for write,
1799 so need add a code to lustre for O_SYNC work.
1804 Description: Failure to close file and release space on NFS
1805 Details : Put inode details into lock acquired in ll_intent_file_open.
1806 Use mdc_intent_lock in ll_intent_open to properly
1807 detect all kind of errors unhandled by mdc_enqueue.
1812 Description: proc file read during shutdown sometimes raced obd removal,
1814 Details : Add lock to prevent obd access after proc file removal.
1817 Frequency : Only for files larger than 4GB on 32-bit clients.
1819 Description: improperly doing page alignment of locks
1820 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1821 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1826 Description: RPCs being resent when they shouldn't be
1827 Details : Some RPCs that should not be resent are being resent. This
1828 can cause inconsistencies in the RPC state machine. Do not
1829 resend such requests.
1832 Frequency : rare, only with NFS export
1834 Description: Crash on NFS re-export node
1835 Details : under very unusual load conditions an assertion is hit in
1836 ll_intent_file_open()
1839 Frequency : only if OST filesystem is corrupted
1841 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1842 Details : for a short time RPCs with bulk IO are in the replay list,
1843 but replay of bulk IOs is unimplemented. If the OST filesystem
1844 is corrupted due to disk cache incoherency and then replay is
1845 started it is possible to trip an assertion. Avoid putting
1846 committed RPCs into the replay list at all to avoid this issue.
1851 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1852 Details : Large single O_DIRECT read and write calls can fail to allocate
1853 a sufficiently large buffer to process the request. In case of
1854 allocation failure the allocation is retried with a smaller
1855 buffer and broken into smaller requests.
1857 Severity : enhancement
1859 Description: Add -o localflock option to simulate old noflock behaviour.
1860 Details : This will achieve local-only flock/fcntl locks coherentness.
1865 Description: versioning check is incomplete
1866 Details : Checking the version difference of client vs. server, report
1867 error if the gap is too big.
1872 Description: add support PG_writeback bit
1873 Details : add support for PG_writeback bit for Lustre, for more carefull
1874 work with page cache in 2.6 kernel. This also fix some deadlocks
1875 and remove hack for work O_SYNC with 2.6 kernel.
1877 Severity : enhancement
1879 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1880 Details : The uninit_groups feature works in conjunction with the kernel
1881 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1882 up the pass1 processing of e2fsck. This is a read-only feature
1883 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1884 mount filesystems that have had this feature enabled.
1886 Severity : enhancement
1888 Description: Improve multi-block allocation algorithm to avoid fragmentation
1889 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1890 allocation locality and avoid filesystem fragmentation.
1892 ------------------------------------------------------------------------------
1894 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1896 * Support for kernels:
1897 2.6.9-42.0.3.EL (RHEL 4)
1898 2.6.5-7.276 (SLES 9)
1899 2.4.21-47.0.1.EL (RHEL 3)
1900 2.6.12.6 vanilla (kernel.org)
1901 2.6.16.21-0.8 (SLES10)
1902 * Recommended e2fsprogs version: 1.39.cfs2-0
1904 * The backwards-compatible /proc/sys/portals symlink has been removed
1905 in this release. Before upgrading, please ensure that you change
1906 any configuration scripts or /etc/sysctl.conf files that access
1907 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1908 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1909 in advance of the upgrade on any system running Lustre 1.4.6 or
1910 newer, since /proc/sys/lnet was added in that version.
1911 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1915 Frequency : only when quota is used
1917 Description: avoid scanning export list for quota master
1918 Details : Change the algorithms to avoid scanning export list in order
1919 to improve the efficiency.
1922 Frequency : MDS failover only, very rarely
1924 Description: "went back in time" messages on mds failover
1925 Details : The greatest transno may be lost when the current operation
1926 finishes with an error (transno==0) and the client's last_rcvd
1927 record is over-written. Save the greatest transno in the
1928 mds_last_transno for this case.
1931 Frequency : always for specific kernels and striping counts
1933 Description: client may get "Matching packet too big" without ACL support
1934 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1935 when trying to access files in certain configurations. The
1936 clients should in fact be denied when mounting because they do
1937 not understand ACLs.
1940 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1942 Description: many clients connecting with IO in progress causes connect timeouts
1943 Details : Avoid synchronous journal commits to avoid delays caused by many
1944 clients connecting/disconnecting when bulk IO is in progress.
1945 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1946 OST_IO_PORTAL to avoid delays behind potentially many pending
1950 Frequency : occasionally with multiple writers to a single file
1952 Description: shared writes to file may result in wrong size reported by stat()
1953 Details : Allow growing of kms when extent lock is cancelled
1956 Frequency : always with random mmap IO to multi-striped file
1958 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1959 Details : Only if the hole is at the end of OST object so that kms is too
1960 small. Fix is to increase kms accordingly in ll_nopage.
1963 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1965 Description: writes to a missing object would leak memory on the OST
1966 Details : If there is an inconsistency between the MDS and OST filesystems,
1967 such that the MDS references an object that doesn't exist, writes
1968 to that object will leak memory due to incorrect cleanup in the
1969 error handling path, eventually running out of memory on the OST.
1974 Description: Creating too long symlink causes lustre errors
1975 Details : Check symlink and name lengths before sending requests to MDS.
1978 Frequency : only if flock is enabled (not on by default)
1980 Description: posix locks not released on fd closure on 2.6.9+
1981 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1982 kernels, this caused such locks not to be released on fd close and
1983 then assertions on fs unmount about still used locks.
1986 Frequency : MDS failover only, very rarely
1988 Description: clients may get ASSERTION(granted_lock != NULL)
1989 Details : When request was taking a long time, and a client was resending
1990 a getattr by name lock request. The were multiple lock requests
1991 with the same client lock handle and
1992 mds_getattr_name->fixup_handle_for_resent_request found one of the
1993 lock handles but later failed with ASSERTION(granted_lock != NULL).
1998 Description: handle->h_buffer_credits > 0, assertion failure
1999 Details : h_buffer_credits is zero after truncate, causing assertion
2000 failure. This patch extends the transaction or creates a new
2004 Frequency : NFS re-export or patchless client
2005 Bugzilla : 11179, 10796
2006 Description: Crash on NFS re-export node (__d_move)
2007 Details : We do not want to hash the dentry if we don't have a lock.
2008 But if this dentry is later used in d_move, we'd hit uninitialised
2009 list head d_hash, so we just do this to init d_hash field but
2010 leave dentry unhashed.
2013 Frequency : NFS re-export or patchless client
2015 Description: NFS exports has problem with symbolic link
2016 Details : lustre client didn't properly install dentry when re-exported
2017 to NFS or running patchless client.
2020 Frequency : NFS re-export or patchless client
2022 Description: Various nfs/patchless fixes.
2023 Details : fixes reuse disconected alias for lookup process - this fixes
2024 warning "find_exported_dentry: npd != pd",
2025 fix permission error with open files at nfs.
2026 fix apply umask when do revalidate.
2029 Frequency : occasional
2031 Description: Crash on NFS re-export node
2032 Details : calling clear_page() on the wrong pointer triggered oops in
2033 generic_mapping_read().
2036 Frequency : rarely, using O_DIRECT IO
2038 Description: unaligned directio crashes client with LASSERT
2039 Details : check for unaligned buffers before trying any requests.
2042 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2044 Description: stale data returned from RAID cache
2045 Details : If only a small amount of IO is done to the RAID device before
2046 reading it again it is possible to get stale data from the RAID
2047 cache instead of reading it from disk.
2050 Frequency : always for sles10 kernel
2052 Description: sles10 support
2053 Details : ll_follow_link: compile fixes and using of nd_set_link
2054 under newer kernels.
2057 Frequency : depends on arch, kernel and compiler version, always on sles10
2060 Description: recursive or deep enough symlinks cause stack overflow
2061 Details : getting rid of large stack-allocated variable in
2065 Frequency : depends on hardware
2067 Description: lustre write performance loss in the SLES10 kernel
2068 Details : the performance loss is caused by using of write barriers in the
2069 ext3 code. The SLES10 kernel turns barrier support on by
2070 default. The fix is to undo that change for ldiskfs.
2072 ------------------------------------------------------------------------------
2074 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2076 * Support for kernels:
2077 2.6.9-42.0.3EL (RHEL 4)
2078 2.6.5-7.276 (SLES 9)
2079 2.4.21-47.0.1.EL (RHEL 3)
2080 2.6.12.6 vanilla (kernel.org)
2084 Frequency : quota enabled and large files being deleted
2086 Description: releasing more than 4GB of quota at once hangs OST
2087 Details : If a user deletes more than 4GB of files on a single OST it
2088 will cause the OST to spin in an infinite loop. Release
2089 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2094 Description: statfs data retrieved from /proc may be stale or zero
2095 Details : When reading per-device statfs data from /proc, in the
2096 {kbytes,files}_{total,free,avail} files, it may appear
2097 as zero or be out of date.
2100 Frequency : systems with MD RAID1 external journal devices
2102 Description: lconf's call to blkid is confused by RAID1 journal devices
2103 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2104 of returning all block devices that match the journal UUID.
2107 Frequency : always, for aggregate stripe size over 4GB
2109 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2110 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2111 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2114 Frequency : always if "lfs find" used on a local file/directory
2116 Description: "lfs find" segfaults if used on a local file/directory
2117 Details : The case where a directory component was not specified wasn't
2118 handled correctly. Handle this properly.
2121 Frequency : always on ppc64
2123 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2124 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2125 perform endianness conversion.
2128 Frequency : rarely (truncate to non-zero file size after write under load)
2129 Bugzilla : 10730, 10687
2130 Description: Files padded with zeros to next 4K multiple
2131 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2132 it is possible that files that are truncated to a non-zero size
2133 immediately after being written are filled with zero bytes beyond
2134 the truncated size. No file data is lost.
2136 Severity : enhancement
2138 Description: Allow recovery/failover for liblustre clients.
2139 Details : liblustre clients were unaware of failover configurations until
2142 Severity : enhancement
2144 Description: user file locks should fail when not mounting with flock option
2145 Details : Set up an error-returning stub in ll_file_operations.lock field
2146 to prevent incorrect behaviour when client is mounted without
2147 flock option. Also, set up properly f_op->flock field for
2151 Frequency : always on ia64
2153 Description: "lfs df" loops on printing out MDS statfs information
2154 Details : The obd_ioctl_data was not initialized and in some systems
2155 this caused a failure during the ioctl that did not return
2156 an error. Initialize the struct and return an error on failure.
2159 Frequency : SLES 9 only
2161 Description: Error of copying files with lustre special EAs as root
2162 Details : Client side always return success for setxattr call for lustre
2163 special xattr (currently only "trusted.lov").
2166 Frequency : rarely on clusters with both ia64+i386 clients
2168 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2169 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2170 clients are doing concurrent writes to the same file it is
2171 possible that the ia64 clients may LASSERT because the OST
2172 extent locks are not PAGE_SIZE aligned. Ensure that grown
2173 locks are always aligned on the request boundary.
2176 Frequency : specific use, occasional
2178 Description: Overwriting in use executable truncates on-disk binary image
2179 Details : If one node attempts to overwrite an executable in use by
2180 another node, we now correctly return ETXTBSY instead of
2181 truncating the file.
2183 Severity : enhancement
2185 Description: Async OSC create to avoid the blocking unnecessarily.
2186 Details : If a OST has no remain object, system will block on the creating
2187 when need to create a new object on this OST. Now, ways use
2188 pre-created objects when available, instead of blocking on an
2189 empty osc while others are not empty. If we must block, we block
2190 for the shortest possible period of time.
2195 Description: chmod on Lustre root is propagated to other clients
2196 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2197 invalid by the follow_mount time.
2200 Frequency : liblustre clients only
2202 Description: Race in 'instant cancel' lock handling could lead to such locks
2203 never to be granted in case of SMP MDS
2204 Details : Do not destroy not yet granted but cbpending locks in
2208 Frequency : replay/resend of open
2210 Description: non null lock assetion failure in mds_intent_policy
2211 Details : Trying to replay/resend lockless open requests resulted in
2212 mds_open() returning 0 with no lock. Now it sets a flag if
2213 a lock is going to be returned.
2215 Severity : enhancement
2217 Description: Checksum enhancements
2218 Details : New checksum enhancements allow for resending RPCs that failed
2221 Severity : enhancement
2223 Description: Tunables on number of dirty pages in cacche
2224 Details : Allow to set limit on number of dirty pages cached.
2229 Description: client crash on unmount - lock still has references
2230 Details : In some error handling cases it was possible to leak a lock
2231 reference on a client while accessing a file. This was not
2232 harmful to the client during operation, but would cause the
2233 client to crash when the filesystem is unmounted.
2236 Frequency : specific case, rare
2238 Description: ETXTBSY on mds though file not in use by client
2239 Details : ETXTBSY is no longer incorrectly returned when attempting to
2240 chmod or chown a directory that the user previously tried to
2241 execute or a currently-executing binary.
2244 Frequency : extremely rare except on liblustre-based clients
2246 Description: Lustre space not freed when files are deleted
2247 Details : Clean up open-unlinked files after client eviction. Previously
2248 the unlink was skipped and the files remained as orphans.
2253 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2254 Details : In some cases it was possible to send a blocking callback to
2255 a client doing a glimpse, even though that client didn't get
2256 a lock granted. When the glimpse lock is cancelled on the OST
2257 the freed lock is left on the waiting list and corrupted the list.
2260 Frequency : all core dumps
2262 Description: Broke core dumps to lustre
2263 Details : Negative dentry may be unhashed if parent does not have UPDATE
2264 lock, but some callers, e.g. do_coredump, expect dentry to be
2265 hashed after successful create, hash it in ll_create_it.
2267 ------------------------------------------------------------------------------
2269 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2271 * Support for kernels:
2272 2.6.9-42.0.2.EL (RHEL 4)
2273 2.6.5-7.276 (SLES 9)
2274 2.4.21-40.EL (RHEL 3)
2275 2.6.12.6 vanilla (kernel.org)
2279 Frequency : always on RHEL 3
2281 Description: Number of open files grows over time
2282 Details : The number of open files grows over time, whether or not
2283 Lustre is started. This was due to a filp leak introduced
2284 by one of our kernel patches.
2286 ------------------------------------------------------------------------------
2288 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2290 * Support for kernels:
2291 2.6.9-42.EL (RHEL 4)
2292 2.6.5-7.276 (SLES 9)
2293 2.4.21-40.EL (RHEL 3)
2294 2.6.12.6 vanilla (kernel.org)
2299 Bugzilla : 5719, 9635, 9792, 9684
2300 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2301 Details : If a server is under heavy load and cannot reply to new
2302 connection requests before the client resends the (re)connect,
2303 the connection handling code can behave badly if two service
2304 threads are concurrently handing separate (re)connections from
2305 the same client. Add better locking to the connection handling
2306 code, and ensure that only a single connection will be processed
2307 for a given client UUID, even if the lock is dropped.
2309 Severity : enhancement
2311 Description: add TCP zero-copy support to kernel
2312 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2313 sends if the hardware supports scatter-gather and checksumming.
2314 This allows socklnd to do client-write and server-read more
2315 efficiently and reduce CPU utilization from skbuf copying.
2318 Frequency : only if NFS exporting from client
2320 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2321 Details : If setting attributes on a file created under NFS that had
2322 never been opened it would be possible to oops the client
2323 if the file had no objects.
2326 Frequency : always for liblustre
2328 Description: liblustre client does MDS+OSTs setattr RPC for each write
2329 Details : When doing a write from a liblustre client, the client
2330 incorrectly issued an RPC to the MDS and each OST the file was
2331 striped over in order to update the timestamps. When writing
2332 with small chunks and many clients this could overwhelm the MDS
2333 with RPCs. In all cases it would slow down the write because
2334 these RPCs are unnecessary.
2336 Severity : enhancement
2338 Description: allow number of MDS service threads to be changed at module load
2339 Details : It is now possible to change the number of MDS service threads
2340 running. Adding "options mds mds_num_threads={N}" to the MDS's
2341 /etc/modprobe.conf will set the number of threads for the next
2342 time Lustre is restarted (assuming the "mds" module is also
2343 reloaded at that time). The default number of threads will
2344 stay the same, 32 for most systems.
2349 Description: OST crash if filesystem is unformatted or corrupt
2350 Details : If an OST is started on a device that has never been formatted
2351 or if the filesystem is corrupt and cannot even mount then the
2352 error handling cleanup routines would dereference a NULL pointer.
2357 Description: NULL pointer deref in llap_from_page.
2358 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2359 incorrect) mapping if the page was truncated/reclaimed while it was
2360 searched for. Check for this condition and skip such pages when
2361 doing readahead. Introduce extra check to llap_from_page() to
2362 verify page->mapping->host is non-NULL (so page is not anonymous).
2365 Frequency : Sometimes when using sys_sendfile
2367 Description: "page not covered by a lock" warnings from ll_readpage
2368 Details : sendfile called ll_readpage without right page locks present.
2369 Now we introduced ll_file_sendfile that does necessary locking
2370 around call to generic_file_sendfile() much like we do in
2374 Frequency : with certain MDS communication failures at client mount time
2376 Description: NULL pointer deref after failed client mount
2377 Details : a client connection request may delayed by the network layer
2378 and not be sent until after the PTLRPC layer has timed out the
2379 request. If the client fails the mount immediately it will try
2380 to clean up before the network times out the request. Add a
2381 reference from the request import to the obd device and delay
2382 the cleanup until the network drops the request.
2385 Frequency : occasionally during client (re)connect
2387 Description: assertion failure during client (re)connect
2388 Details : processing a client connection request may be delayed by the
2389 client or server longer than the client connect timeout. This
2390 causes the client to resend the connection request. If the
2391 original connection request is replied in this interval, the
2392 client may trip an assertion failure in ptlrpc_connect_interpret()
2393 which thought it would be the only running connect process.
2396 Frequency : only with obd_echo servers and clients that are rebooted
2398 Description: kernel BUG accessing uninitialized data structure
2399 Details : When running an obd_echo server it did not start the ping_evictor
2400 thread, and when a client was evicted an uninitialized data
2401 structure was accessed. Start the ping_evictor in the RPC
2402 service startup instead of the OBD startup.
2404 Severity : enhancement
2405 Bugzilla : 10193 (patchless)
2406 Description: Remove dependency on various unexported kernel interfaces.
2407 Details : No longer need reparent_to_init, exit_mm, exit_files,
2408 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2411 Frequency : rare (only users of deprecated and unsupported LDAP config)
2413 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2414 Details : LDAP apparently contains 'lustreName' attributes instead of
2415 'name'. A simple remapping of the name is sufficient.
2418 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2420 Description: waiting_locks_callback trips kernel BUG if client is evicted
2421 Details : Running with the dump_on_timeout debug flag turned on makes
2422 it possible that the waiting_locks_callback() can try to dump
2423 the Lustre kernel debug logs from an interrupt handler. Defer
2424 this log dumping to the expired_lock_main() thread.
2426 Severity : enhancement
2428 Description: Support NFS exporting on 2.6 kernels.
2429 Details : Implement non-rawops metadata methods for NFS server to use without
2430 changing NFS server code.
2433 Frequency : very rare (synthetic metadata workload only)
2435 Description: two racing renames might cause an MDS thread to deadlock
2436 Details : Running the "racer" program may cause one MDS thread to rename
2437 a file from being the source of a rename to being the target of
2438 a rename at exactly the same time that another thread is doing
2439 so, and the second thread has already enqueued these locks after
2440 doing a lookup of the target and is trying to relock them in
2441 order. Ensure that we don't try to re-lock the same resource.
2444 Frequency : only very large systems with liblustre clients
2446 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2447 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2448 all OSTs in parallel. This allows the eviction of stale liblustre
2449 clients to proceed much faster than if they were done in series,
2450 and also offers similar improvements for other set_info RPCs.
2455 Description: excessive CPU usage during initial read phase on client
2456 Details : During the initial read phase on a client, it would agressively
2457 retry readahead on the file, consuming too much CPU and impacting
2458 performance (since 1.4.5.8). Improve the readahead algorithm
2459 to avoid this, and also improve some other common cases (read
2460 of small files in particular, where "small" is files smaller than
2461 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2466 Description: MDS crash when receiving packet with unknown intent.
2467 Details : Do not LBUG in unknown intent case, just return -EFAULT
2469 Severity : enhancement
2470 Bugzilla : 9293, 9385
2471 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2472 Details : Do not serialize getattr (non-intent version) and statfs.
2475 Frequency : occasional, when OST network is overloaded/intermittent
2477 Description: client evicted by OST after bulk IO timeout
2478 Details : If a client sends a bulk IO request (read or write) the OST
2479 may evict the client if it is unresposive to its data GET/PUT
2480 request. This is incorrect if the network is overloaded (takes
2481 too long to transfer the RPC data) or dropped the OST GET/PUT
2482 request. There is no need to evict the client at all, since
2483 the pinger and/or lock callbacks will handle this, and the
2484 client can restart the bulk request.
2487 Frequency : Always when mmapping file with no objects
2489 Description: client crashes when mmapping file with no objects
2490 Details : Check that we actually have objects in a file before doing any
2491 operations on objects in ll_vm_open, ll_vm_close and
2497 Description: Request leak when working with deleted CWD
2498 Details : Introduce advanced request refcount tracking for requests
2499 referenced from lustre intent.
2501 Severity : Enhancement
2503 Description: Cache open file handles on client.
2504 Details : MDS now will return special lock along with openhandle, if
2505 requested and client is allowed to hold openhandle, even if unused,
2506 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2507 closing files for every read/write openration.
2509 Severity : Enhancement
2511 Description: Cache open negative dentries on client when possible.
2512 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2513 negative dentries on lock revocation.
2518 Description: Remounting a client read-only wasn't possible with a zconf mount
2519 Details : It wasn't possible to remount a client read-only with llmount.
2521 Severity : enhancement
2522 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2523 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2524 MPICH is included to add Lustre-specific ADIO interfaces.
2525 This is based closely on the UFS ADIO layer and only differs
2526 in file creation, in order to allow the OST striping to be set.
2527 This is user-contributed code and not supported by CFS.
2532 Description: extended inode attributes (immutable, append-only) work improperly
2533 when 2.4 and 2.6 kernels are used on client/server or vice versa
2534 Details : Introduce kernel-independent values for these flags.
2536 Severity : enhancement
2539 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2540 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2541 granularity. Now, Fractional MB granularity is be supported,
2542 this is very useful for low memory system.
2544 Severity : enhancement
2546 Description: Getattr by fid
2547 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2548 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2549 have locks and avoids getattr rpc requests for it on every stat.
2552 Frequency : Always, for filesystems larger than 2TB
2554 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2555 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2556 filesystems larger than 2TB to be mounted with mballoc enabled.
2559 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2561 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2562 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2563 filesystems larger than 2TB is mounted then the kernel will
2564 silently corrupt the start of the filesystem. CONFIG_LBD is
2565 enabled for all CFS-supported kernels, but the possibility of
2566 this happening with a modified kernel config exists.
2568 Severity : enhancement
2570 Description: add client O_DIRECT support for 2.6 kernels
2571 Details : It is now possible to do O_DIRECT reads and writes to files
2572 in the Lustre client mountpoint on 2.6 kernel clients.
2574 Severity : enhancement
2576 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2577 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2578 parallel, not waiting for response from every OST before sending
2579 a rpc to the next OST.
2584 Description: setattr vs write race when updating file timestamps
2585 Details : Client processes that update a file timestamp into the past
2586 right after writing to the file (e.g. tar) it is possible that
2587 the updated file modification time can be reset to the current
2588 time due to a race between processing the setattr and write RPC.
2590 Severity : enhancement
2592 Description: Bring 'lfs find' closer in line with regular Linux find.
2593 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2594 -print0 options and obtains all the needed info through the lustre
2597 Severity : enhancement
2599 Description: support up to 1024 configured devices on one node
2600 Details : change obd_dev array from statically allocated to dynamically
2601 allocated structs as they are first used to reduce memory usage
2606 Description: Flush dirty partially truncated pages during truncate
2607 Details : Immediatelly flush partially truncated pages in filter_setattr,
2608 this way we completely avoid having any pages in page cache on OST
2609 and can retire ugly workarounds during writes to flush such pages.
2614 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2615 Details : protect inode from truncation within vfs_unlink() context
2616 just take a reference before calling vfs_unlink() and release it
2617 when parent's i_sem is free.
2622 Description: last_id value checked outside lock on OST caused LASSERT failure
2623 Details : If there were multiple MDS->OST object precreate requests in
2624 flight, it was possible that the OST's last object id was checked
2625 outside a lock and incorrectly tripped an assertion. Move checks
2626 inside locks, and discard old precreate requests.
2629 Frequency : always, if extents are used on OSTs
2631 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2632 Details : OSTs using the extents format would not zero the high 16 bits of
2633 the index physical block number. This is not a problem for any
2634 OST filesystems smaller than 16TB, and no kernels support ext3
2635 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2636 new/modified files) and can be fixed for existing filesystems
2637 with e2fsprogs-1.39-cfs1.
2642 Description: import connection selection may be incorrect if timer wraps
2643 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2644 import connections to be ignored if the 32-bit jiffies counter
2645 wraps. Use a 64-bit jiffies counter.
2648 Frequency : very large clusters immediately after boot
2650 Description: LNET request buffers exhausted under heavy short-term load
2651 Details : If a large number of client requests are generated on a service
2652 that has previously never seen so many requests it is possible
2653 that the request buffer growth cannot keep up with the spike in
2654 demand. Instead of dropping incoming requests, they are held in
2655 the LND until the RPC service can accept more requests.
2658 Frequency : Sometimes during replay
2660 Description: Assertion failure in ll_local_open after replay.
2661 Details : If replay happened on an open request reply before we were able
2662 to set replay handler, reply will become not swabbed tripping the
2663 assertion in ll_local_open. Now we set the handler right after
2664 recognising of open request
2667 Frequency : very rare
2669 Description: kernel reports "badness in vsnprintf"
2670 Details : Reading from the "recovery_status" /proc file in small chunks
2671 may cause a negative length in lprocfs_obd_rd_recovery_status()
2672 call to vsnprintf() (which is otherwise harmless). Exit early
2673 if there is no more space in the output buffer.
2675 Severity : enhancement
2677 Description: clear OBD RPC statistics by writing to them
2678 Details : It is now possible to clear the OBD RPC statistics by writing
2679 to the "stats" file.
2684 Description: Client mtime is not the same on different clients after utimes
2685 Details : In some cases, the client was using the utimes() syscall on
2686 a file cached on another node. The clients now validate the
2687 ctime from the MDS + OSTs to determine which one is right.
2692 Description: Inability to activate failout mode
2693 Details : lconf script incorrectly assumed that in pythong string's numeric
2694 value is used in comparisons.
2697 Frequency : always with multiple stripes per file
2699 Description: Inefficient object allocation for mutli-stripe files
2700 Details : When selecting which OSTs to stripe files over, for files with
2701 a stripe count that divides evenly into the number of OSTs,
2702 the MDS is always picking the same starting OST for each file.
2703 Return the OST selection heuristic to the original design.
2708 Description: mount failures may take full timeout to return an error
2709 Details : Under some heavy load conditions it is possible that a
2710 failed mount can wait for the full obd_timeout interval,
2711 possibly several minutes, before reporting an error.
2712 Instead return an error as soon as the status is known.
2714 Frequency : quota enabled and large files being deleted
2716 Description: releasing more than 4GB of quota at once hangs OST
2717 Details : If a user deletes more than 4GB of files on a single OST it
2718 will cause the OST to spin in an infinite loop. Release
2719 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2724 Description: statfs data retrieved from /proc may be stale or zero
2725 Details : When reading per-device statfs data from /proc, in the
2726 {kbytes,files}_{total,free,avail} files, it may appear
2727 as zero or be out of date.
2730 Frequency : systems with MD RAID1 external journal devices
2732 Description: lconf's call to blkid is confused by RAID1 journal devices
2733 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2734 of returning all block devices that match the journal UUID.
2737 Frequency : always, for aggregate stripe size over 4GB
2739 Description: assertion fails when trying to use 4GB stripe size
2740 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2741 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2744 Frequency : always on ppc64
2746 Description: the first write on an ext3 filesystem with mballoc got stuck
2747 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2748 perform endianness conversion.
2750 ------------------------------------------------------------------------------
2752 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2754 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2755 INTEROPERATE with older versions automatically. Please read the
2756 user documentation before upgrading any part of a live system.
2757 * WARNING: Lustre networking configuration changes are required with
2758 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2761 * Support for kernels:
2762 2.6.9-22.0.2.EL (RHEL 4)
2763 2.6.5-7.244 (SLES 9)
2764 2.6.12.6 vanilla (kernel.org)
2767 Severity : enhancement
2768 Bugzilla : 7981/8208
2769 Description: Introduced Lustre Networking (LNET)
2770 Details : LNET is new networking infrastructure for Lustre, it includes
2771 a reorganized network configuration mode (see the user
2772 documentation for full details) as well as support for routing
2773 between different network fabrics. Lustre Networking Devices
2774 (LNDs) for the supported network fabrics have also been
2775 created for this new infrastructure.
2777 Severity : enhancement
2778 Description: Introduced Access control lists
2779 Details : clients can set ACLs on files and directories in order to have
2780 more fine-grained permissions than the standard Unix UGO+RWX.
2781 The MDS must be started with the "-o acl" mount option.
2783 Severity : enhancement
2784 Description: Introduced filesystem quotas
2785 Details : Administrators may now establish per-user quotas on the
2788 Severity : enhancement
2790 Description: Configuration change for the XT3
2791 The PTLLND is now used to run Lustre over Portals on the XT3
2792 The configure option(s) --with-cray-portals are no longer used.
2793 Rather --with-portals=<path-to-portals-includes> is used to
2794 enable building on the XT3. In addition to enable XT3 specific
2795 features the option --enable-cray-xt3 must be used.
2800 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2801 Details : A race between allocating a new llcd and re-getting the llcd_lock
2802 allowed another thread to grab newly-allocated llcd.
2804 Severity : enhancement
2806 Description: 2.6 OST async journal commit and locking fix to improve performance
2807 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2808 2.6 kernels are now async as they already were in 2.4 kernels so
2809 that they can commit concurrently with the network bulk transfer.
2810 For block-allocated files the filter allocation semaphore is held
2811 to avoid filesystem fragmentation during allocation. BKL lock
2812 removed for 2.6 xattr operations where it is no longer needed.
2817 Description: lconf incorrectly determined whether two IP networks could talk
2818 Details : In some more complicated routing and multiple-network
2819 configurations, lconf will avoid trying to make a network
2820 connection to a disjoint part of the IP space. It was doing the
2821 math incorrectly for one set of cases.
2826 Description: Fix for potential infinite loop processing records in an llog.
2827 Details : If an llog record is corrupted/zeroed, it is possible to loop
2828 forever in llog_process(). Validate the llog record length
2829 and skip the remainder of the block on error.
2832 Frequency : occasional (liblustre only)
2834 Description: liblustre could not open files whose last component is a symlink
2835 Details : sysio_path_walk() would incorrectly pass the open intent to
2836 intermediate path components.
2839 Frequency : rare (liblustre only with non-standard tuning)
2840 Bugzilla : 7201 (7350)
2841 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2842 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2843 after it has been released, only do this for referenced locks
2845 Severity : enhancement
2847 Description: specifying an (invalid) directory default stripe_size of -1
2848 would reset the directory default striping
2849 Details : stripe_size -1 was used internally to signal directory stripe
2850 removal, now use "all default" to signal dir stripe removal
2851 as a directory striping of "all default" is not useful
2854 Frequency : common for large clusters running liblustre clients
2856 Description: doing an ls when liblustre clients are running is slow
2857 Details : sending a glimpse AST to a liblustre client waits for every AST
2858 to time out, as liblustre clients will not respond. Since they
2859 cannot cache data we refresh the OST lock LVB from disk instead.
2861 Severity : enhancement
2863 Description: doing an ls at the same time as file IO can be slow
2864 Details : enqueue and other "small" requests can be blocked behind many
2865 large IO requests. Create a new OST IO portal for non-IO
2866 requests so they can be processed faster.
2869 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2871 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2872 created file with mode 000
2873 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2874 allow an ftruncate() as a non-root user to a file with mode 000.
2875 Lustre now respects this flag to disable mode checks when
2876 truncating a file owned by the user
2879 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2882 Description: Revoking locks from clients that went dead or catatonic might take
2884 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2885 cancellation of such locks instant on servers without waiting for
2886 any reply from clients. Clients drops these locks when cancel
2887 notification from server is received without replying.
2890 Frequency : liblustre-only, when liblustre client dies or becomes busy
2892 Description: Doing ls on Linux clients can take a long time with active
2894 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2895 granting such locks to it in the first place if possible. Locks
2896 are taken by proxy on the OST during the read or write and
2897 dropped immediately afterward. Add connect flags handling, do
2898 not grant locks to liblustre clients for glimpse ASTs.
2900 Severity : enhancement
2902 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2903 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2904 sequential reads and behaves badly with applications doing random
2905 reads. Improve it by reading ahead at least read region, and
2906 avoiding excessive large RPC for small reads.
2908 Severity : enhancement
2910 Description: Creating more than 1000 files for a single job may cause a load
2911 imbalance on the OSTs if there are also a large number of OSTs.
2912 Details : qos_prep_create() uses an OST index reseed value that is an
2913 even multiple of the number of available OSTs so that if the
2914 reseed happens in the middle of the object allocation it will
2915 still utilize the OSTs as uniformly as possible.
2920 Description: OST or MDS may oops in ping_evictor_main()
2921 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2922 but doesn't restart at beginning of obd_exports_timed list
2925 Severity : enhancement
2927 Description: improve by-nid export eviction on the MDS and OST
2928 Details : allow multiple exports with the same NID to be evicted at one
2929 time without re-searching the exports list.
2932 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2934 Description: MDS may oops in groups_free()
2935 Details : in rare race conditions a newly allocated group_info struct is
2936 freed again, and this can be NULL. The 2.4 compatibility code
2937 for groups_free() checked for a NULL pointer, but 2.6 did not.
2940 Frequency : common for liblustre clients doing little filesystem IO
2941 Bugzilla : 9352, 7313
2942 Description: server may evict liblustre clients accessing contended locks
2943 Details : if a client is granted a lock or receives a completion AST
2944 with a blocking AST already set it would not reply to the AST
2945 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2948 Frequency : lfs setstripe, only systems with more than 160 OSTs
2950 Description: unable to set striping with a starting offset beyond OST 160
2951 Details : llapi_create_file() incorrectly limited the starting stripe
2952 index to the maximum single-file stripe count.
2955 Frequency : LDAP users only
2957 Description: lconf did not handle in-kernel recovery with LDAP properly
2958 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2960 Severity : enhancement
2962 Description: bind OST threads to NUMA nodes to improve performance
2963 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2964 node and do their allocations there to localize memory access
2966 Severity : enhancement
2968 Description: llmount can determine client NID directly from Myrinet (GM)
2969 Details : the client NID code from gmnalnid was moved directly into
2970 llmount, removing the need to use this or specifying the
2971 client NID explicitly when mounting GM clients with zeroconf
2974 Frequency : if client is started with down MDS
2976 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2977 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2978 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2979 ptlrpc_import_delayed_req() because we didn't check if the
2980 request was interrupted, and we also didn't break out of the
2981 event loop if there was no timeout
2986 Description: data loss during non-page-aligned writes to a single file from
2987 both multiple nodes and multiple threads on one node at same time
2988 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2989 inconsistency led to false short-reads, that were cached and later
2990 used by ->prepare_write() to fill in partially written page,
2991 leading to data loss.
2994 Frequency : always, if lconf --abort_recovery used
2996 Description: lconf --abort_recovery fails with 'Operation not supported'
2997 Details : lconf was attempting to abort recovery on the MDT device and not
3000 Severity : enhancement
3002 Description: remove cleanup logs
3003 Details : replace lconf-generated cleanup logs with lustre internal
3004 cleanup routines. Eliminates the need for client-cleanup and
3007 Severity : enhancement
3009 Description: add support for EAs (user and system) on lustre filesystems
3010 Details : it is now possible to store extended attributes in the Lustre
3011 client filesystem, and with the user_xattr mount option it
3012 is possible to allow users to store EAs on their files also
3014 Severity : enhancement
3016 Description: Add possibility (config option) to show minimal available OST free
3018 Details : When compiled with --enable-mindf configure option, statfs(2)
3019 (and so, df) will return least minimal free space available from
3020 all OSTs as amount of free space on FS, instead of summary of
3021 free spaces of all OSTs.
3023 Severity : enhancement
3025 Description: do not expand extent locks acquired on OST-side
3026 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3027 by server: they are not cached anyway.
3030 Frequency : when mmap is used/binaries executed from Lustre
3032 Description: Unmmap pages before throwing them away from read cache.
3033 Details : llap_shrink cache now attempts to unmap pages before discarding
3034 them (if unmapping failed - do not discard). SLES9 kernel has
3035 extra checks that trigger if this unmapping is not done first.
3040 Description: lconf didn't resolve symlinks before checking to see whether a
3041 given mountpoint was already in use
3044 Frequency : when migrating failover services
3045 Bugzilla : 6395, 9514
3046 Description: When migrating a subset of services from a node (e.g. failback
3047 from a failover service node) the remaining services would
3048 time out and evict clients.
3049 Details : lconf --force (implied by --failover) sets the global obd_timeout
3050 to 5 seconds in order to quickly disconnect, but this caused
3051 other RPCs to time out too quickly. Do not change the global
3052 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3054 Severity : enhancement
3055 Frequency : if MDS is started with down OST
3056 Bugzilla : 9439,5706
3057 Description: Allow startup/shutdown of an MDS without depending on the
3058 availability of the OSTs.
3059 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3060 Add appropriate locking and lov-osc refcounts for safe
3061 cleaning. Add osc abort_inflight calls in case the
3062 synchronize never started.
3065 Frequency : occasional (Cray XT3 only)
3067 Description: root not authorized to access files in CRAY_PORTALS environment
3068 Details : The client process capabilities were not honoured on the MDS in
3069 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3070 been accessed by an authorized user then root was able to access
3071 the file on the local client also. The root user capabilities
3072 are now allowed on the MDS, as this environment has secure UID.
3075 Frequency : occasional
3077 Description: ldiskfs "too long searching" message happens too often
3078 Details : A debugging message (otherwise harmless) prints too often on
3079 the OST console. This has been reduced to only happen when
3080 there are fragmentation problems on the filesystem.
3085 Description: Division by zero in statfs when all OSCs are inactive
3086 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3087 lov_statfs divides by value returned by lov_get_stripecnt().
3091 Bugzilla : 9489, 3273
3092 Description: First write from each client to each OST was only 4kB in size,
3093 to initialize client writeback cache, which caused sub-optimal
3094 RPCs and poor layout on disk for the first writen file.
3095 Details : Clients now request an initial cache grant at (re)connect time
3096 and so that they can start streaming writes to the cache right
3097 away and always do full-sized RPCs if there is enough data.
3098 If the OST is rebooted the client also re-establishes its grant
3099 so that client cached writes will be honoured under the grant.
3104 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3105 Details : Now I/O RPCs go to different portal number and (presumably) fast
3106 lock requests (and glimses) and other RPCs get their own service
3107 threads pool that should be able to service those RPCs
3110 Severity : enhancement
3112 Description: Ability to exchange lustre version between client and servers and
3113 issue warnings at client side if client is too old. Also for
3114 liblustre clients there is ability to refuse connection of too old
3116 Details : New 'version' field is added to connect data structure that is
3117 filled with version info. That info is later checked by server and
3121 Frequency : rare, liblustre only.
3122 Bugzilla : 9296, 9581
3123 Description: Two simultaneous writes from liblustre at offset within same page
3124 might proceed at the same time overwriting eachother with stale
3126 Details : I/O lock withing llu_file_prwv was released too early, before data
3127 actually was hitting the wire. Extended lock-holding time until
3128 server acknowledges receiving data.
3131 Frequency : extremely rare. Never observed in practice.
3133 Description: avoid generating lustre_handle cookie of 0.
3134 Details : class_handle_hash() generates handle cookies by incrementing
3135 global counter, and can hit 0 occasionaly (this is unlikely, but
3136 not impossible, because initial value of cookie counter is
3137 selected randonly). Value of 0 is used as a sentinel meaning
3138 "unassigned handle" --- avoid it. Also coalesce two critical
3139 sections in this function into one.
3141 Severity : enhancement
3143 Description: allow liblustre clients to delegate truncate locking to OST
3144 Details : To avoid overhead of locking, liblustre client instructs OST to
3145 take extent lock in ost_punch() on client's behalf. New connection
3146 flag is added to handle backward compatibility.
3148 Severity : enhancement
3149 Bugzilla : 4928, 7341, 9758
3150 Description: allow number of OST service threads to be specified
3151 Details : a module parameter allows the number of OST service threads
3152 to be specified via "options ost ost_num_threads={N}" in the
3153 OSS's /etc/modules.conf or /etc/modprobe.conf.
3157 Bugzilla : 6146, 9635, 9895
3158 Description: servers crash with bad pointer in target_handle_connect()
3159 Details : In rare cases when a client is reconnecting it was possible that
3160 the connection request was the last reference for that export.
3161 We would temporarily drop the export reference and get a new
3162 one, but this may have been the last reference and the export
3163 was just destroyed. Get new reference before dropping old one.
3165 Severity : enhancement
3166 Frequency : if client is started with failover MDS
3168 Description: Allow multiple MDS hostnames in the mount command
3169 Details : Try to read the configuration from all specified MDS
3170 hostnames during a client mount in case the "primary"
3173 Severity : enhancement
3175 Description: Stop sending data to evicted clients as soon as possible.
3176 Details : Check if the client we are about to send or are sending data to
3177 was evicted already. (Check is done every second of waiting,
3178 for which l_wait_event interface was extended to allow checking
3179 of exit condition at specified intervals).
3182 Frequency : rare, normally only when NFS exporting is done from client
3184 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3186 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3189 Frequency : always, if srand() is called before liblustre initialization
3191 Description: Liblustre uses system PRNG disturbing its usage by user application
3192 Details : Introduce internal to lustre fast and high-quality PRNG for
3193 lustre usage and make liblustre and some other places in generic
3194 lustre code to use it.
3196 Severity : enhancement
3197 Bugzilla : 9477, 9557, 9870
3198 Description: Verify that the MDS configuration logs are updated when xml is
3199 Details : Check if the .xml configuration logs are newer than the config
3200 logs stored on the MDS and report an error if this is the case.
3201 Request --write-conf, or allow starting with --old_conf.
3203 Severity : enhancement
3205 Description: Handle symlinks in the path when checking if Lustre is mounted.
3206 Details : Resolve intermediate symlinks when checking if a client has
3207 mounted a filesystem to avoid duplicate client mounts.
3212 Description: lconf can hit an error exception but still return success.
3213 Details : The lconf command catches the Command error exception at the top
3214 level script context and will exit with the associated exit
3215 status, but doesn't ensure that this exit status is non-zero.
3220 Description: failure of ptlrpc thread startup can cause oops
3221 Details : Starting a ptlrpc service thread can fail if there are a large
3222 number of threads or the server memory is very fragmented.
3223 Handle this without oopsing.
3226 Frequency : always, only if liblustre and non-default acceptor port was used
3228 Description: liblustre cannot connect to servers with non-default acceptor port
3229 Details : tcpnal_set_default_params() was not called and was therefore
3230 ignoring the environment varaible TCPNAL_PORT, as well as other
3231 TCPNAL_ environment variables
3236 Description: two objects could be created on the same OST for a single file
3237 Details : If an OST is down, in some cases it was possible to create two
3238 objects on a single OST for a single file. No problems other
3239 than potential performance impact and spurious error messages.
3243 Bugzilla : 5681, 9562
3244 Description: Client may oops in ll_unhash_aliases
3245 Details : Client dcache may become inconsistent in race condition.
3246 In some cases "getcwd" can fail if the current directory is
3252 Description: Inode refcounting problems in NFS export code
3253 Details : link_raw functions used to call d_instantiate without obtaining
3254 extra inode reference first.
3258 Bugzilla : 9942, 9903
3259 Description: Referencing freed requests leading to crash, memleaks with NFS.
3260 Details : We used to require that call to ll_revalidate_it was always
3261 followed by ll_lookup_it. Also with revalidate_special() it is
3262 possible to call ll_revalidate_it() twice for the same dentry
3263 even if first occurence returned success. This fix changes semantic
3264 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3265 reference on a request referred from the intent.
3266 ll_intent_release() then releases such a request.
3269 Frequency : rare, normally benchmark loads only
3271 Description: unlinked inodes were kept in memory on the client
3272 Details : If a client is repeatedly creating and unlinking files it
3273 can accumulate a lot of stale inodes in the inode slab cache.
3274 If there is no other client load running this can cause the
3275 client node to run out of memory. Instead flush old inodes
3276 from client cache that have the same inode number as a new inode.
3279 Frequency : SLES9 2.6.5 kernel and long filenames only
3280 Bugzilla : 9969, 10379
3281 Description: utime reports stale NFS file handle
3282 Details : SLES9 uses out-of-dentry names in some cases, which confused
3283 the lustre dentry revalidation. Change it to always use the
3287 Frequency : rare, unless heavy write-truncate concurrency is continuous
3288 Bugzilla : 4180, 6984, 7171, 9963, 9331
3289 Description: OST becomes very slow and/or deadlocked during object unlink
3290 Details : filter_destroy() was holding onto the parent directory lock
3291 while truncating+unlinking objects. For very large objects this
3292 may block other threads for a long time and slow overall OST
3293 responsiveness. It may also be possible to get a lock ordering
3294 deadlock in this case, or run out of journal credits because of
3295 the combined truncate+unlink. Solution is to do object truncate
3296 first in one transaction without parent lock, and then do the
3297 final unlink in a new transaction with the parent lock. This
3298 reduces the lock hold time dramatically.
3301 Frequency : rare, 2.4 kernels only
3303 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3304 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3305 before unmounting their filesystem in order to avoid deadlock.
3306 The kernel_locked() function in 2.4 kernels only checks whether
3307 the kernel lock is held, not whether it is this process that is
3308 holding it as 2.6 kernels do.
3313 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3314 Details : The client ptlrpc code may be trying to reconnect to a down
3315 server before a previous connection attempt has timed out.
3316 Increase the reconnect interval to be longer than the connection
3317 timeout interval to avoid sending duplicate connections to
3321 Frequency : echo_client brw_test command
3323 Description: fix echo_client to work with OST preallocated code
3324 Details : OST preallocation code (5137) didn't take echo_client IO path
3325 into account: echo_client calls filter methods outside of any
3326 OST thread and, hence, there is no per-thread preallocated
3327 pages and buffers to use. Solution: hijack pga pages for IO. As
3328 a byproduct, this avoids unnecessary data copying.
3332 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3333 Description: Client can oops in mdc_commit_close() after open replay
3334 Details : It was possible for the MDS to return an open request with no
3335 transaction number in mds_finish_transno() if the client was
3336 evicted, but without actually returning an error. Clients
3337 would later try to replay that open and may trip an assertion
3338 Simplify the client close codepath, and always return an error
3339 from the MDS in case the open is not successful.
3342 Frequency : rare, 2.6 OSTs only
3344 Description: OST may deadlock under high load on fragmented files
3345 Details : If there was a heavy load and highly-fragmented OST filesystems
3346 it was possible to have all the OST threads deadlock waiting on
3347 allocation of biovecs, because the biovecs were not released
3348 until the entire RPC IO was completed. Instead, release biovecs
3349 as soon as they are complete to ensure forward IO progress.
3351 Severity : enhancement
3353 Description: Support for specifying external journal device at mount
3354 Details : If an OST or MDS device is formatted with an external journal
3355 device, this device major/minor is stored in the ext3 superblock
3356 and may not be valid for failover. Allow detecting and
3357 specifying the external journal at mount time.
3362 Description: Mounting an MDS with pending unlinked files may cause oops
3363 Details : target_finish_recovery() calls mds_postrecov() which returned
3364 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3365 considers this an error and immediately begins cleaning up the
3366 lov, just after starting the mds_lov process
3368 Severity : enhancement
3370 Description: Implement 'lfs df' to report actual free space on per-OST basis
3371 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3372 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3373 usage of objects; '-h' to report in human readable format.
3375 ------------------------------------------------------------------------------
3377 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3384 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3385 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3386 at startup during orphan unlinking. Instead, skip buddy update
3387 before setup, as it will be regenerated after recovery is complete.
3392 Description: If an OST is inactive, its locks might reference stale inodes.
3393 Details : lov_change_cbdata() must iterate over all namespaces, even if
3394 they are inactive to clear inode references from the lock.
3396 Severity : enhancement
3397 Frequency : occasional, if non-standard max_dirty_mb used
3399 Description: Client will block write RPCs if not enough grant
3400 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3401 then the client will block writes while waiting for another RPC
3402 to complete instead of consuming its dirty limit. With change
3403 we get improved performance when max_dirty_mb is small.
3405 Severity : enhancement
3406 Bugzilla : 3389, 6253
3407 Description: Add support for supplementary groups on the MDS.
3408 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3409 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3410 lookups for user supplementary groups into a cache.
3414 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3415 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3416 for liblustre O_EXCL is enforced by MDS.
3420 Description: readdir never returns NULL in liblustre.
3421 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3422 dentry in struct dirent.
3426 Frequency : liblustre only, depends on application IO pattern
3427 Description: liblustre clients evicted if not contacting servers
3428 Details : Don't put liblustre clients into the ping_evictor list, so
3429 they will not be evicted by the pinger ever.
3431 Severity : enhancement
3433 Description: Add ability to evict clients by NID from MDS.
3434 Details : By echoing "nid:$NID" string into
3435 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3436 $NID would be instantly evicted from this MDS and from all active
3437 OSTs connected to it.
3441 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3442 Details : lookup_it_finish() used to query file size from OSTs that was not
3447 Description: service threads change working directory to that of init
3448 Details : Starting lustre service threads may pin the working directory
3449 of the parent thread, making that filesystem busy. Threads
3450 now change to the working directory of init to avoid this.
3454 Frequency : during shutdown only
3455 Description: shutdown with a failed MDS or OST can cause unmount to hang
3456 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3461 Frequency : 2.6 only
3462 Description: chmod/chown may include an extra supplementary group
3463 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3464 supplementary group and if none is specified this is used.
3467 Bugzilla : 5479 (6816)
3469 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3470 Details : If lookup is in progress on a file that is unlinked we might try
3471 to revalidate the inode and fail in revalidate after lookup is
3472 complete and ll_file_open() enqueues the open again but
3473 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3476 Frequency : always, if lconf --abort_recovery used
3478 Description: lconf --abort_recovery fails with 'Operation not supported'
3479 Details : lconf was attempting to abort recovery on the MDT device and not
3482 ------------------------------------------------------------------------------
3484 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3489 Frequency : rare (only unsupported configurations with a node running as an
3491 Bugzilla : 6514, 5137
3492 Description: Mounting a Lustre file system on a node running as an OST could
3494 Details : OSTs now preallocates memory needed to write out data at
3495 startup, instead of when needed, to avoid having to
3496 allocate memory in possibly low memory situations.
3497 Specifically, if the file system is mounted on on OST,
3498 memory pressure could force it to try to write out data,
3499 which it needed to allocate memory to do. Due to the low
3500 memory, it would be unable to do so and the node would
3501 become unresponsive.
3503 Severity : enhancement
3505 Description: Addition of lconf --service command line option
3506 Details : lconf now accepts a '--service <arg>' option, which is
3507 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3509 Severity : enhancement
3511 Description: Failover mode is now the default for OSTs.
3512 Details : By default, OSTs will now run in failover mode. To return to
3513 the old behaviour, add '--failout' to the lmc line for OSTs.
3515 Severity : enhancement
3517 Description: Health checks are now provided for MDS and OSTs
3518 Details : Additional detailed health check information on MSD and OSTs
3519 is now provided through the procfs health_check value.
3522 Frequency : occasional, depends on IO load
3524 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3525 after numerous create/delete cycles
3526 Details : The ext3 inode allocation policy would not allocate new inodes
3527 very well on the OSTs because there are no new directories
3528 being created. Instead we look for groups with free space if
3529 the parent directories are nearly full.
3534 Description: Network or server problems during mount may cause partially
3535 mounted clients instead of returning an error.
3536 Details : The config llog parsing code may overwrite the error return
3537 code during mount error handling, returning success instead
3543 Description: MDS can fail to allocate large reply buffers
3544 Details : After long uptimes the MDS can fail to allocate large reply
3545 buffers (e.g. zconf client mount config records) due to memory
3546 fragmentation or consumption by the buffer cache. Preallocate
3547 some large reply buffers so that these replies can be sent even
3548 under memory pressure.
3552 Frequency : rare (liblustre)
3553 Description: fsx running with liblustre complained that using truncate() to
3554 extend the file doesn't work. This patch corrects that issue.
3555 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3556 ATTR_SIZE and lsm use in llu_setattr_raw.
3560 Frequency : rare, only 2.6 kernels
3561 Description: Unusual file access patterns on the MDS may result in inode
3562 data being lost in very rare circumstances.
3563 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3564 code in the 2.6 kernel caused the inode and/or EA data not to
3565 be read from disk, causing single-file corruption.
3569 Frequency : rare, only 2.6 filesystems using extents
3570 Description: Heavy concurrent write and delete load may cause data corruption.
3571 Details : It was possible under high-load situations to have an extent
3572 metadata block in the block device cache from a just-unlinked
3573 file overwrite a newly-allocated data block. We now unmap any
3574 metadata buffers that alias just-allocated data blocks.
3578 Frequency : filesystems with default stripe_count larger than 77
3579 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3580 Details : lconf specifies an inode size of 4096 bytes when the default
3581 stripe_count is larger than 77. This conflicts with the default
3582 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3583 this case to avoid pinning too much memory for large EAs.
3585 ------------------------------------------------------------------------------
3587 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3592 Frequency : rare (extremely heavy IO load with hundreds of clients)
3594 Description: Client is evicted, gets IO error writing to file
3595 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3596 caused clients to be evicted for AST timeouts. The fixes in
3597 bug 5192 mean we no longer need to have such short AST timeouts
3598 so ldlm_timeout has been increased.
3601 Frequency : occasional during --force or --failover shutdown under load
3602 Bugzilla : 5949, 4834
3603 Description: Server oops/LBUG if stopped with --force or --failover under load
3604 Details : a collection of import/export refcount and cleanup ordering
3605 issues fixed for safer force cleanup
3608 Frequency : only filesystems larger than 120 OSTs
3609 Bugzilla : 5990, 6223
3610 Description: lfs getstripe would oops on a very large filesystem
3611 Details : lov_getconfig used kfree on vmalloc'd memory
3614 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3615 Bugzilla : 6242, 6243
3616 Description: reading from files that had been truncated to a non-zero size
3617 but never opened returned no data
3618 Details : ll_file_read() reads zeros from no-object files to EOF
3623 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3624 Details : There's little harm in aborting MDS/OSS recovery and letting it
3625 try again, so I removed the LASSERT and return an error instead.
3627 Severity : enhancement
3629 Description: New debugging infrastructure for tracking down data corruption
3630 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3631 (b) cover more of the client-side code path, and (c) try to narrow
3632 down where problems occurred
3636 Bugzilla : 3819, 4364, 4397, 6313
3637 Description: Racing close and eviction MDS could cause assertion in mds_close
3638 Details : It was possible to get multiple mfd references during close and
3639 client eviction, leading to one thread referencing a freed mfd.
3641 Severity: : enhancement
3642 Bugzilla : 3262, 6359
3643 Description: Attempts to reconnect to servers are now more aggressive.
3644 Details : This builds on the enhanced upcall-less recovery that was added
3645 in 1.4.2. When trying to reconnect to servers, clients will
3646 now try each server in the failover group every 10 seconds. By
3647 default, clients would previously try one server every 25 seconds.
3652 Description: After recovery, certain operations trigger a failed
3653 assertion on a client.
3654 Details : Failing over an mds, using lconf -d --failover, while a
3655 client was doing a readdir() call would cause the client to
3656 LBUG after recovery completed and the readdir() was resent.
3658 Severity : enhancement
3660 Description: Default groups are now added by lconf
3661 Details : You can now run lconf --group <servicename> without having to
3662 manually add groups with lmc.
3665 Frequency : occasional
3667 Description: Nodes with an elan id of 0 trigger a failed assertion
3670 Frequency : always when accessing e.g. tty/console device nodes
3672 Description: tty and some other devices nodes cannot be used on lustre
3673 Details : file's private_data field is used by device data and lustre
3674 values in there got lost. New field was added to struct file to
3675 store fs-specific private data.
3678 Frequency : when exporting Lustre via NFS
3680 Description: NFSD failed occasionally when looking up a path component
3681 Details : NFSD is looking up ".." which was broken in ext3 directories
3682 that had grown large enough to become hashed.
3685 Frequency : Clusters with multiple interfaces not on the same subnet
3687 Description: Nodes will repeatedly try to reconnect to an interface which it
3688 cannot reach and report an error to the log.
3689 Details : Extra peer list entries will be created by lconf with some peers
3690 unreachable. lconf now validates the peer before adding it.
3693 Frequency : Only if a default stripe is set on the filesystem root.
3695 Description: Setting a default stripe on the filesystem root prevented the
3696 filesystem from being remounted.
3697 Details : The client was sending extra request flags in the root getattr
3698 request and did not allocate a reply buffer for the dir EA.
3701 Frequency : occasional, higher if lots of files are accessed by one client
3702 Bugzilla : 6159, 6097
3703 Description: Client trips assertion regarding lsm mismatch/magic
3704 Details : While revalidating inodes the VFS looks up inodes with ifind()
3705 and in rare cases can find an inode that is being freed.
3706 The ll_test_inode() code will free the lsm during ifind()
3707 when it finds an existing inode and then the VFS later attaches
3708 this free lsm to a new inode.
3712 Bugzilla : 6422, 7030
3713 Description: MDS deadlock between mkdir and client eviction
3714 Details : Creating a new file via mkdir or mknod (starting a transaction
3715 and getting the ns lock) can deadlock with client eviction
3716 (gets ns lock and trying to finish a synchronous transaction).
3719 Frequency : occasional
3720 Description: While starting a server, the fsfilt_ext3 module could not be
3722 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3723 kernels. Previously, lconf would still use the ext3 name
3724 when trying to load modules. Now, it will correctly use
3725 ext3 on 2.4 and ldiskfs on 2.6.
3727 Severity : enhancement
3728 Description: The default stripe count has been changed to 1
3729 Details : The interpretation of the default stripe count (0, to lfs
3730 or lmc) has been changed to mean striping across a single
3731 OST, rather than all available. For general usage we have
3732 found a stripe count of 1 or 2 works best.
3734 Severity : enhancement
3735 Description: Add support for compiling against Cray portals.
3736 Details : Conditional compiling for some areas that are different
3740 Frequency : occasional
3741 Bugzilla : 6409, 6834
3742 Description: Creating files with an explicit stripe count may lead to
3743 a failed assertion on the MDS
3744 Details : If some OSTs are full or unavailable, creating files may
3745 trigger a failed assertion on the MDS. Now, Lustre will
3746 try to use other servers or return an error to the
3750 Frequency : occasional
3752 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3753 caused lock timeout during readahead (since 1.4.2).
3754 Details : Processes doing readahead might match a lock that hasn't been
3755 granted yet if there are overlapping and conflicting lock
3756 requests. The readahead process waits on ungranted lock
3757 (original lock is CBPENDING), while OST waits for that process
3758 to cancel CBPENDING read lock and eventually evicts client.
3760 Severity : enhancement
3762 Description: Initial enabling of flock support for clients
3763 Details : Implements fcntl advisory locking and file status functions.
3764 This feature is provided as an optional mount flag (default
3765 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3766 locking are implemented yet, and those that are are not guaranteed
3767 to be completely correct in production environments.
3768 mount -t lustre -o [flock|noflock] ...
3771 Frequency : occasional
3773 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3774 assertion in the ext3 JBD (journaling) layer.
3775 Details : The b_committed_data struct is protected by the big kernel lock
3776 in 2.4 kernels, serializing journal_commit_transaction() and
3777 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3778 access to this struct. In 2.6 kernels there is finer grained
3779 locking to improve SMP performance of the JBD layer.
3783 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3788 Description: merge and cleanup kernel patches.
3791 -----------------------------------------------------------------------------
3793 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3795 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3796 versions of Lustre. Please update all servers and clients to
3797 version 1.4.2 or later at the same time. You must also run
3798 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3799 to update the configuration logs.
3801 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3802 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3803 - swab llog records of type '0' so we get proper header size/idx (5861)
3804 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3805 - fix rename of one directory over another leaking an inode (5953)
3806 - avoid SetPageDirty on 2.6 (5981)
3807 - don't re-add just-being-destroyed locks to the waiting list (5653)
3808 - when creating new directories, inherit the parent's custom
3809 striping settings if present parent (3048)
3810 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3811 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3812 - don't hold client locks on temporary worklist from l_lru (5666)
3813 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3814 - automatically evict dead clients (5921)
3815 - Update file size properly in create+truncate+fstat case (6196)
3816 - Do not unhash mountpoint dentries, do not allow removal of
3818 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3819 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3820 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3821 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3822 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3823 - Correctly handle reads of files with no objects (6243)
3824 - lctl recover will also mark a device active if deactivate used (5933)
3826 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3827 - allow --write-conf on an MDS with different nettype than client (5619)
3828 - don't write config llogs to MDS for mounts not from that MDS (5617)
3829 - lconf should create multiple TCP connections from a client (5201)
3830 - init scripts are now turned off by default; run chkconfig --on
3831 lustre and chkconfig --on lustrefs to use them
3832 - upcalls are no longer needed for clients to recover to failover
3834 - add --abort-recovery option to lconf to abort recovery on device
3836 - add support for an arbitrary number of OSTs (3026)
3837 - Quota support protocol changes.
3838 - forward compatibility changes to wire structs (6007)
3839 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3840 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3841 - improved "lustre status" script
3842 - initialize blocksize for non-regular files (6062)
3843 - added --disable-server and --disable-client configure options (5782)
3844 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3845 - Vanilla 2.4.29 support
3846 - increase maximum number of obd devices to 520 (6242)
3847 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3848 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3849 SLES 9 2.6.5 kernels
3850 - limit stripes per file to 160 (the maximum EA size) (6093)
3852 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3855 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3856 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3857 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3858 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3859 - small ext3 extents cleanups and fixes (5733)
3860 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3861 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3862 - avoid needless client->OST connect, fix handle mismatch (5317)
3863 - fix DLM error path that led to out-of-sync client, long delays (5779)
3864 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3865 - fix several locking issues related to i_size (5492,5624,5654,5672)
3866 - don't move pending lock onto export if it is already evicted (5683)
3867 - fix kernel oops when creating .foo in unlinked directory (5548)
3868 - fix deadlock in obdfilter statistics vs. object create (5811)
3869 - use time_{before,after} to avoid timer jiffies wrap (5882)
3870 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3871 - Do not leak request if server was not able to process it (5154)
3872 - If mds_open unable to find parent dir, make that negative lookup(5154)
3873 - don't create new directories with extent-mapping (5909, 5936)
3875 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3876 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3877 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3878 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3879 - support for RHEL4 kernel on i686 (5773)
3880 - provide error messages when incompatible logs are encountered (5898)
3882 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3883 * version 1.4.0.10 (1.4.1 release candidate 1)
3885 - don't keep a lock reference when lock is not granted (4238)
3886 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3887 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3888 - drop import inflight refcount on signal_completed_replay error (5255)
3889 - unlock page after async write error during send (3677)
3890 - handle missing objects in filter_preprw_read properly (5265)
3891 - no transno return for symlink open, don't save no-trasno open (3440)
3892 - don't try to complete elan receive that already failed (4012)
3893 - free RPC server reply state on error (5406)
3894 - clean up thread from ptlrpc_start_thread() on error (5160)
3895 - readahead could read extra page into cache that wasn't ejected (5388)
3896 - prevent races in class_attach/setup/cleanup/detach (5260)
3897 - don't dereference de->d_inode after l_dput of de (5458)
3898 - use "int" for stripe value returned from lock_to_stripe (5544)
3899 - mballoc allocation and error-checking fixes in 2.6 (5504)
3900 - block device patches to fix I/O request sizes in 2.6 (5482)
3901 - look up hostnames for IB nals (5602)
3902 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3903 - don't start multiple acceptors for the same port (5277)
3904 - fix incorrect LASSERT in mds_getattr_name (5635)
3905 - export a proc file for general "ping" checking (5628)
3906 - fix "lfs check" to not block when the MDS is down (5628)
3908 - service request history (4965)
3909 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3910 - create an "evict_client" /proc entry on OSTs, like the MDS has
3911 - fix mount usage message, return errors per mount(8) (5168)
3912 - change grep [] to grep "[]" in tests so they work in more UMLs
3913 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3914 - remove ancient LOV_MAGIC_V0 EA support (5047)
3915 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3916 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3917 - allow readahead allocations to fail when low on memory (5383)
3918 - mmap locking landed again, after considerable improvement (2828)
3919 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3921 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3924 - send OST transaction number in read/write reply to free req (4966)
3925 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3926 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3927 - the watchdog thread now runs as interruptible (5246)
3928 - flock/lockf fixes (but it's still disabled, pending 5135)
3929 - don't use EXT3 constants in llite code (5094)
3930 - memory shortage at startup could cause assertion (5176)
3932 - reorganization of lov code
3933 - single portals codebase
3935 - add extents/mballoc support (5025)
3936 - direct I/O reads in the obdfilter (4048)
3937 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3939 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3942 - send OST transaction number in read/write reply to free req (4966)
3943 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3944 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3945 - the watchdog thread now runs as interruptible (5246)
3946 - handle missing objects in filter_preprw_read properly (5265)
3947 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3948 - drop import inflight refcount on signal_completed_replay error (5255)
3949 - unlock page after async write error during send (3677)
3950 - return original error code on reconstructed replies (3761)
3951 - no transno return for symlink open, don't save no-trasno open (3440)
3953 - add pid to ldlm debugging output (4922)
3954 - bump the watchdog timeouts -- we can't handle 30sec yet
3955 - extra debugging for orphan dentry/inode bug (5259)
3957 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3960 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3961 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3962 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3963 - fix NULL dereference in /proc/sys/portals/routes (4827)
3964 - allow failed mdc_close() operations to be interrupted (4561)
3965 - stop precreate on OST before MDS would time out on it (4778)
3966 - don't send partial-page writes before EOF from client (4410)
3967 - discard client grant for sub-page writes on large-page clients (4520)
3968 - don't free dentries not owned by NFS code, check generation (4806)
3969 - fix lsm leak if mds_create_objects() fails (4801)
3970 - limit debug_daemon file size, always print CERROR messages (4789)
3971 - use transno after validating reply (3892)
3972 - process timed out requests if import state changes (3754)
3973 - update mtime on OST during writes, return in glimpse (4829)
3974 - add mkfsoptions to LDAP (4679)
3975 - use ->max_readahead method instead of zapping global ra (5039)
3976 - don't interrupt __l_wait_event() during strace
3978 - add software watchdogs to catch hung threads quickly (4941)
3979 - make lustrefs init script start after nfs is mounted
3980 - fix CWARN/ERROR duplication (4930)
3981 - return async write errors to application if possible (2248)
3982 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3983 - print NAL number in %x format (4645)
3984 - update barely-supported suse-2.4.21-171 series (4842)
3985 - support for sles 9 %post scripts
3986 - support for building 2.6 kernel-source packages
3987 - support for sles km_* packages
3989 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3992 - ignore -ENOENT errors in osc_destroy (3639)
3993 - notify osc create thread that OSC is being cleaned up (4600)
3994 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3995 - reconstruct ost_handle() like mds_handle() (4657)
3996 - create a new thread to do import eviction to avoid deadlock (3969)
3997 - let lconf resolve symlinked-to devices (4629)
3998 - don't unlink "objects" from directory with default EA (4554)
3999 - hold socknal file ref over connect in case target is down (4394)
4000 - allow more than 32000 subdirectories in a single directory (3244)
4001 - fix blocks count for O_DIRECT writes (3751)
4002 - OST returns ENOSPC from object create when no space left (4539)
4003 - don't send truncate RPC if file size isn't changing (4410)
4004 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4005 - bind to privileged port in socknal and tcpnal (3689)
4007 - rate limit CERROR/CWARN console message to avoid overload (4519)
4008 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4009 - basic mmap support (3918)
4010 - kernel patch series update from b1_4 (4711)
4012 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4015 - avoid crash during MDS cleanup with OST shut down (2775)
4016 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4017 - don't use bad inodes on the MDS (3744)
4018 - dynamic object preallocation to improve recovery speed (4236)
4019 - don't hold spinlock over lock dumping or change debug flags (4401)
4020 - don't zero obd_dev when it is force cleaned (3651)
4021 - print grants to console if they go negative (4431)
4022 - "lctl deactivate" will stop automatic recovery attempts (3406)
4023 - look for existing locks in ldlm_handle_enqueue() (3764)
4024 - don't resolve lock handle twice in recovery avoiding race (4401)
4025 - revalidate should check working dir is a directory (4134)
4027 - don't always mark "slow" obdfilter messages as errors (4418)
4029 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4032 - don't close LustreDB during write_conf until it is done (3860)
4033 - fix typo in lconf for_each_profile (3821)
4034 - allow dumping logs from multiple threads at one time (3820)
4035 - don't allow multiple threads in OSC recovery (3812)
4036 - fix debug_size parameters (3864)
4037 - fix mds_postrecov to initialize import for llog ctxt (3121)
4038 - replace config semaphore with spinlock (3306)
4039 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4040 - don't allow enqueue to complete on a destroyed export (3822)
4041 - down write_lock before checking llog header bitmap (3825)
4042 - recover from lock replay timeout (3764)
4043 - up llog sem before sending rpc (3652)
4044 - reduce ns lock hold times when setting kms (3267)
4045 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4046 - fix NULL deref and obd_dev leak on setup error (3312)
4047 - replace some LBUG about llog ops with error handling (3841)
4048 - don't match INVALID dentries from d_lookup and spin (3784)
4049 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4050 - fix invalid assertion in ptlrpc_set_wait (3880)
4052 - add libwrap support for the TCP acceptor (3996)
4053 - add /proc/sys/portals/routes for non-root route listing (3994)
4054 - allow setting MDS UUID in .xml (2580)
4055 - print the stack of a process that LBUGs (4228)
4057 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4060 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4061 - make sure to unset replay flag from failed open requests (3440)
4062 - if default stripe count is 0, use OST count for inode size (3636)
4063 - update parent mtime/ctime on client for create/unlink (2611)
4064 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4065 - free recovery state on server during a forced cleanup (3571)
4066 - unregister_reply for resent reqs (3063)
4067 - loop back devices mounting and status check on 2.6 (3563)
4068 - fix resource-creation race that can provoke i_size == 0 (3513)
4069 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4070 - more debugging for page-accounting assertion (3746)
4071 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4072 - avoid deadlock after precreation failure (3758)
4073 - fix race and lock order deadlock in orphan handling (3450, 3750)
4074 - add validity checks when grabbing inodes from l_ast_data (3599)
4076 - add /proc/.../recovery_status to obdfilter (3428)
4077 - lightweight CDEBUG infrastructure, debug daemon (3668)
4078 - change default OSC RPC parameters to be better on small clusters
4079 - turn off OST read cache for files smaller than 32MB
4080 - install man pages and include them in rpms (3100)
4081 - add new init script for (un)mounting lustre filesystems (2593)
4082 - run chkconfig in %post for init scripts (3701)
4083 - drop scimac NAL (unmaintained)
4085 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4088 - clean kiobufs before and after use (3485)
4089 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4090 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4091 - change init script to fail more clearly if not run as root (1528)
4092 - allow clients to reconnect during replay (1742)
4093 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4094 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4095 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4096 - fix cond_resched (3554)
4097 - only evict unfinished clients after recovery (3515)
4098 - allow bulk resend, prevent data loss (3570)
4099 - dynamic ptlrpc request buffer allocation (2102)
4100 - don't allow unlinking open directory if it isn't empty (2904)
4101 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4102 - remove extraneous obd dereference causing LASSERT failure (3334)
4103 - don't use get_cycles() when creating temp. files on the mds (3156)
4104 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4105 - handle EEXIST for set-stripe, set proper directory name (3336)
4107 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4108 - fix ksocknal_fmb_callback() error messages (2918)
4110 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4113 - don't copy lvb into (possibly NULL) reply on error (2983)
4114 - don't deref dentry after dput, don't free lvb on error (2922)
4115 - use the kms to determine writeback rpc length (2947)
4116 - increment oti_logcookies when osc is inactive (2948)
4117 - update client's i_blocks count via lvb messages (2543)
4118 - handle intent open/close of special files properly (1557)
4119 - mount MDS with errors=remount-ro, like obdfilter (2009)
4120 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4121 - don't use cancelling-locks' kms values (2947)
4122 - use highest lock extent for kms, not last one (2925)
4123 - don't dereference ERR_PTR() dentry in error handling path (3107)
4124 - fix thread race in portals_debug_dumplog() (3122)
4125 - create lprocfs device entries at setup instead of at attach (1519)
4126 - common AST error handler, don't evict client on completion race (3145)
4127 - zero nameidata in detach_mnt in 2.6 (3118)
4128 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4129 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4130 - initialize RPC timeout timer earlier for 2.6 (3219)
4131 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4132 - print nal/nid for unknown nid (3258)
4133 - additional checks for oscc recovery before doing precreate (3284)
4134 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4135 - don't crash in mdc_close for bad permissions on open (3285)
4136 - zero i_rdev for non-device files (3147)
4137 - clear page->private before handing to FS, better assertion (3119)
4138 - tune the read pipeline (3236)
4139 - fix incorrect decref of invalidated dentry (2350)
4140 - provide read-ahead stats and refine rpc in flight stats (3328)
4141 - don't hold journal transaction open across create RPC (3313)
4142 - update atime on MDS at close time (3265)
4143 - close LDAP connection when recovering to avoid server load (3315)
4144 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4145 - don't leak open file on MDS after open resend (3325)
4146 - serialize filter_precreate and filter_destroy_precreated (3329)
4147 - loop device shouldn't call sync_dev() for nul device (3092)
4148 - clear page cache after eviction (2766)
4149 - resynchronize MDS->OST in background (2824)
4150 - refuse to mount the same filesystem twice on same mountpoint (3394)
4151 - allow llmount to create routes for mounting behind routers (3320)
4152 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4153 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4154 - fix rare problem with rename on htree directories (3417)
4156 - allow default OST striping configuration per directory (1414)
4157 - fix compilation for qswnal for 2.6 kernels (3125)
4158 - increase maximum number of MDS request buffers for large systems
4159 - change liblustreapi to be useful for external progs like lfsck (3098)
4160 - increase local configuration timeout for slow disks (3353)
4161 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4163 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4166 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4167 - don't overwrite extent policy data in reply if lock was blocked (2901)
4168 - drop filter export grants atomically with removal from device (2663)
4169 - del obd_self_export from work_list in class_disconnect_exports (2908)
4170 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4171 - swab reply message in mdc_close, other PPC fixes (2464)
4172 - fix destroying of named logs (2325)
4173 - overwrite old logs when running lconf --write_conf (2264)
4174 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4175 - fix race in target_handle_connect (2898)
4176 - mds_reint_create() should take same inode create lock (2926)
4177 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4178 - don't close files for self_export to avoid uninitialized obd (2936)
4179 - allow MDS with the same name as client node (2939)
4180 - hold dentry reference for closed log files for unlink (2325)
4181 - reserve space for all logs during transactions (2059)
4182 - don't evict page beyond end of stripe extent (2925)
4183 - don't oops on a deleted current working directory (2399)
4184 - handle hard links to targets without a parent properly (2517)
4185 - don't dereference NULL lock when racing during eviction (2867)
4186 - don't grow lock extents when lots of conflicting locks (2919)
4188 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4191 - account for cache space usage on clients to avoid data loss (974)
4192 - lfsck support in lustre kernel code (2349)
4193 - reduce journal credits needed for BRW writes (2370)
4194 - orphan handling to avoid losing space on client/server crashes
4195 - ptlrpcd can be blocked, stopping ALL progress (2477)
4196 - use lock value blocks to assist in proper KMS, faster stat (1021)
4197 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4198 - recovery for initial connections (2355)
4199 - fixes for mds_cleanup_orphans (1934)
4200 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4201 - block all file creations until orphan recovery completes (1901)
4202 - client remove rq_connection from request struct (2423)
4203 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4204 - recovery timer race (2670)
4205 - mdc_close recovey bug (2532)
4206 - ptlrpc cleanup bug (2710)
4207 - mds timeout on local locks (2588)
4208 - namespace lock held during RPCs (2431)
4209 - handle interrupted sync write properly (2503)
4210 - don't try to handle a message that hasn't been replied to (2699)
4211 - client assert failure during cleanup after abort recovery (2701)
4212 - leak mdc device after failed mount (2712)
4213 - ptlrpc_check_set allows timedout requests to complete (2714)
4214 - wait for inflight reqs when ptlrpcd finishes (2710)
4215 - make sure unregistered services are removed from the srv_list
4216 - reset bulk XID's when resending them (caught by 1138 test)
4217 - unregister_bulk after timeout
4218 - fix lconf error (2694)
4219 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4220 - readahead locks pages, leaves pending causing memory pressure (2673)
4221 - increase OST request buffers to 4096 on large machines (2729)
4222 - fix up permission of existing directories in simple_mkdir (2661)
4223 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4224 - don't assign transno to errored transactions (2742)
4225 - don't delete objects on OST if given a bogus objid from MDS (2751)
4226 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4227 - if rq_no_resend, then timeout request after recovery (2432)
4228 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4229 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4230 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4231 - MDS recovery shouldn't delete objects that it has given out (2730)
4232 - if enqueue arrives after completion, don't clobber LVB (2819)
4233 - don't unlock pages twice when trigger_group_io returns error (2814)
4234 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4235 - don't write pages to disk if there was an error (1450)
4236 - don't ping imports that have recovery disabled (2676)
4237 - take buffered bytes into account when balancing socknal conn (2817)
4238 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4239 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4240 - remove little-endian swabbing of llog records (1987)
4241 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4242 - retry reposting request buffers if they fail (1191)
4243 - grow extent at grant time to avoid granting a revoked lock (2809)
4244 - lock revoke doesn't evict page if covered by a second lock (2765)
4245 - disable VM readahead to avoid reading outside lock extents (2805)
4247 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4248 - updated kernel patches for hp-2.4.20 kernel (2681)
4250 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4253 - fix truncated write corruption (2366)
4254 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4256 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4257 - ensure len(uuid) < 37 in lmc (1171)
4258 - fix ia64 OOPS in llog_test (2255)
4259 - zero end of page at obdfilter for partial page writes (2648)
4260 - don't leave stale dentries around after renames (bug 2428)
4261 - fix timeouts when evicting a client with a single lock held (2642)
4262 - set deadline for the initial HELLO message to drain (2634)
4263 - print out dotted-quad IP addresses in the socknal (2302)
4265 - additional debugging for MDS client eviction problem (2443)
4266 - fix mkfsoptions support for osts (2603, 2604)
4268 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4271 - add series for the vanilla 2.6.0 kernel
4272 - add series for the vanilla 2.4.24 kernel
4273 - add series for a cray x86/64 UL kernel drop
4274 - fix xattr patches for the vanilla 2.4.19 series
4276 - generate true UUIDs in lmc (1171)
4277 - have portals stack dumping break in UML (2466)
4278 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4279 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4280 - wake callback waiting threads on client eviction (2460)
4281 - Add --ptldebug and --subsystem to lmc (1719)
4282 - update assertion to allow safe interrupt allocation
4283 - set rq_no_resend for cancel requests (2432)
4284 - recalculate ptlrpcd timeout after resend (2494)
4285 - call vfs_rmdir when removing pending directories (2368)
4286 - fix renaming a file to itself (2429)
4287 - lmc creates a default one-stripe lov (2454)
4288 - expand procfs space to handle large clusters (2326)
4289 - increase UML stack to avoid overflow
4290 - update lconf's list of debug and subsystem masks
4291 - fix lfs find --obd (2510)
4292 - /proc tunable for disabling filter read caching (2591)
4293 - stop rpm packages from altering slapd.conf (2301)
4294 - disable nagle in the socknal under 0conf (2578)
4295 - choose mds inode size based on stripe count (2572)
4296 - fix kernel-source rpm problems (2516)
4298 - add --disable-doc to avoid pdf generation (2421)
4299 - update documentation, tests, type-os, comments
4300 - avoid format warnings on ia64
4301 - remove the TOE NAL
4302 - tiny code cleanups by removing unused fields
4304 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4307 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4308 - remove the most blatant lies from BUILDING file (2371)
4309 - change default debug level to reasonable production setting
4310 - reduce client side cache size to reduce cache flush time
4311 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4312 - make TCP zerocopy and pinger support enabled by default (2476)
4313 - sync writes completed after process exits caused crashes (2319)
4314 - maintain correct mount count on the MDS (2356)
4315 - backout 1557, because 2316 wasn't really fixed
4316 - better file I/O statistics gathering in /proc
4317 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4318 - another kernel patch to fix zero-copy TCP function export
4319 - don't take duplicate lock when processing re-sent getattr (2420)
4320 - lctl uses obd_self_export instead of creating new conn (2353)
4321 - MDS/OST recovery case which requires object creation asserted (2425)
4322 - move lfs from /usr/sbin to /usr/bin in packages
4323 - fix race between mds_client_add and mds_client_free (2417)
4324 - use kmalloc instead of slabs in portals (2430)
4325 - don't create duplicate records when a failover MDS is present (2442)
4326 - remove unnecessary mount age check (2332)
4327 - don't remove directory inodes from locks prematurely (2451)
4328 - don't break if MDS service name is the same as hostname (2103)
4329 - fix races in client write RPC generation when cache full (2482)
4331 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4334 - remove now-unused request->rq_obd (278)
4335 - if an allocation fails, print out how much memory we've used (1933)
4336 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4337 - add the "configurable stack size" patch to most series files (1256)
4338 - ability to write large log records, for 100+ OST configs (2306)
4339 - fix NULL deref when filter_prep fails (2314)
4340 - fix operator precedence error in filter_sync
4341 - dynamic allocation of socknal TX descriptors (2315)
4342 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4343 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4344 - ensure that CWARN messages in Portals always get to the syslog
4345 - __init/__exit are not for prototype decls (ldlm_init/exit)
4346 - x86-64 compile warning fixes
4347 - fix gateway LMC keyword conflict (2318)
4348 - fix MDS lock inversions in getattr/reint paths (1844)
4349 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4350 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4351 - DLM race condition prevented some lock evictions (2328)
4352 - ENOMEM detection and retry on socknal sends (2230)
4353 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4354 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4355 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4356 - better error messages when a client is rejected during recovery (1505)
4357 - avoid cancelling locks which were never granted, after failure (2330)
4358 - fix i_sem/journal inversion in mds_client_add (2333)
4359 - fix truncate/getattr lock cycle deadlock (2334)
4360 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4361 - fix two rare exit paths which could leak an l_lock() ref (2321)
4362 - fencepost error in MDS/OST orphan recovery (2226)
4363 - make log record alignment 8 bytes (1988)
4364 - lstripe now fails when requested offset > ost_count (2237)
4365 - ensure that all kernel series have a complete list.h (1607)
4366 - fix crashes in special-file operations (2316)
4367 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4369 - allow configurable automake binary, for testing new versions
4370 - small update to the lfs documentation
4372 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4374 * fix negative export reference count in fsfilt_sync (2312)
4376 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4377 * release candidate 0.9.1
4379 - orphans are moved into the PENDING directory for possible recovery
4380 - replayed opens now open by fid for orphan/rename safety (1042)
4381 - last close of an orphan inode generates a transno (683)
4382 - chdir() and mount() now pin the directory entry (1020)
4383 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4384 - discard very old requests without processing them (1502)
4385 - remove some common, well-understood CERRORs (1505)
4386 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4387 - clear "grant" flags in OST replies until OST grant code lands (1644)
4388 - fix read performance by not clobbering i_blksize on client (1598)
4389 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4390 - make LRU size a /proc tunable, clears locks when reduced (707)
4391 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4392 - server locks take references on exports now (1558)
4393 - build fixes for 2.4.20-rh trees (1663)
4394 - return an error from lov_create if all OSCs are inactive (1751)
4395 - fix import levels when a reconnect happens without a timeout (1597)
4396 - exit early from mds_open if we get a lookup error (1749)
4397 - partial page read at EOF wouldn't wait for disk before sending (1642)
4398 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4399 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4400 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4401 - write/truncate lock inversion (1639)
4402 - Don't auto-load obdclass, portals modules during cleanup (1495)
4403 - fix timestamps from jumping to "now" (1763)
4404 - extra journal assertions (1648)
4405 - add an extra multiunlink test (1771)
4406 - fix read_record/write_record API (1776)
4407 - fix leak of offset_extent, possible incorrect i_size later (1772)
4408 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4409 - Debugging for the kqswnal_get_idle_tx problems (1820)
4410 - Allow recovery to be attempted multiple times (1536)
4411 - Write out MDS last_rcvd file after it is first created (1600)
4412 - Fix tx_descriptor leak in failed transmit situations (1827)
4413 - ext3 journaling fixes for assertion failure after IO error (1871)
4414 - class_export_put() on freed export after completion AST error (1896)
4415 - Fix revalidate looping in VFS (1322)
4416 - Don't access a freed export during MDS_REINT timeout (1521)
4417 - Add open-unlink recovery support on the MDS (1673,1764)
4418 - Return an error if no MDS data was read from last_rcvd (1946)
4419 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4420 - Don't setup a disk device that doesn't match exported UUID (317)
4421 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4422 - avoid committing NULL handle in force close
4423 - local.sh is now a one-stripe LOV configuration
4424 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4425 - fix ext3 htree duplicate directory entry corruption (1516)
4426 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4427 - update to new LOV EA format (2097)
4428 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4430 2003-06-15 Phil Schwan <phil@clusterfs.com>
4433 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4434 - per-import/export recovery handling (958, 931, 959)
4435 - multiple last-rcvd slots, for serving multiple FSes (949)
4436 - connections are again shared between multiple imp/exports (963, 964)
4437 - "umount -f" would hang if any requests needed to be sent (393, 978)
4438 - avoid pinning large req buffer by copying for queued messages (989)
4439 - add "uuid" to "lctl device" command to help upcalls (991)
4440 - "open" RPCs with transnos would confuse recovery counters (1037)
4441 - do proper endian conversion of all wire messages (288, 340, 891)
4442 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4443 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4444 - fix ptlrpc_connection leak in target_handle_connect (1174)
4445 - fix import refcounting bug in OST and MDS cleanup (1134)
4446 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4447 - fix very unlikely obd_types race condition (501)
4448 - remove osc_open hack for echo_client (1187)
4449 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4450 - a failure in read_inode2 leads to deadlock (1139)
4451 - cancel ack-locks as soon as transaction is committed (1072)
4452 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4453 - make sure to commitrw after any preprw to avoid deadlock (1162)
4454 - failing to execute a file in a lustre FS would lock inode (1203)
4455 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4456 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4457 - obd_uuid2tgt would walk off the end of the list (1255)
4458 - on IA64 the osc would give portals incorrect bulk size (1258)
4459 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4460 - fix lock inversion caused by new llite matching code (1282)
4461 - limit the number of dirty pages on a client to 10MB (1286)
4462 - timed out locks were not being corrected cancelled (1289)
4463 - fix O_DIRECT above 4GB on IA-32 (1292)
4464 * major user-visible changes
4465 - fail out/fail over policy now controlled by the upcall (993)
4467 - add OBD_PING to check server availability and failure (954)
4468 - lustre messages are now sent in sending host order (288, 340, 891)
4469 - add eadatalen to MDS getattr reply (340)
4470 - OST read replies may contain second buffer, with per-page status (593)
4472 2003-03-11 Phil Schwan <phil@clusterfs.com>
4475 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4476 - failed open()s could cause deadlock; fixed (867, 869)
4477 - stop cancelling OST locks when files are closed (481)
4478 - overlapping XID spaces caused network corruption (851, 853)
4479 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4480 - setattr_raw added, to do single-RPC, server-side setattrs
4481 - lmc/lconf syntax change for OST UUIDs
4482 - fix crashy race condition between ptlrpc_free_req and osc_close
4483 - don't use request in mdc_enqueue if we hit a timeout (889)
4484 - don't set the inode i_size for regular files from the MDS (896)
4485 - handle out of order completion AST (842)
4486 - don't LBUG if a lock request times out after receiving AST (913)
4487 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4488 - if a bad lock AST arrives, send an error instead of dropping entirely
4489 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4490 - fix leak in bulk IO when only partially completed (899, 900, 926)
4491 - fix O_DIRECT for ia64 (55)
4492 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4493 - C-z after timeout could hang a process forever; fixed (977)
4495 - client-side I/O cache (678, 924, 929, 941, 970)
4497 - READPAGE and SETATTRs which don't take server-side locks get
4500 2003-02-11 Phil Schwan <phil@clusterfs.com>
4503 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4504 - Fix fsfilt_extN_readpage() to read a full page of directory
4505 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4506 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4507 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4508 - Most intent operations were removed, in favour of a new RPC mode
4509 that does a single RPC to the server and bypasses most of the VFS
4510 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4511 - Aggressively cancel local locks on DLM servers
4512 - mds_reint_unlink sends EA to the client if it's the last nlink.
4513 client uses that EA to unlink OST objects.
4514 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4515 - recursive symlinks were fixed (439)
4516 - fixed NULL deref in DEBUG_REQ
4517 - filter_update_lastobjid no longer calls sync, which annoyed extN
4518 - fixed multi-client small-writes to a single file problem (445)
4519 - fixed mtime updates during file writes (607)
4520 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4521 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4522 - fixed Linux OST size reporting problem (444, 656)
4523 - OST now updates object mtime with writes or setattr (607, 619)
4524 - client verifies file size before zeroing page past EOF (445)
4525 - OST now writes last allocated objid to disk with allocation (108)
4526 - LOV on echo now works (409)
4528 - mds_reint_unlink sends a new buffer, with the EA included. this
4529 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4530 set if a regular file was being unlinked, and it was the last link
4531 - use PtlGet from the target for bulk writes (315)
4532 - OST now updates object mtime with writes or setattr (607, 619)
4533 - LDLM now has a grant-time callback to revalidate locked items, if
4535 - Many MDS operations were reorganized to combat race conditions
4537 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4538 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4540 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4543 - Fully reactivate OST imports after reconnection (512, others)
4544 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4545 - More graceful error handling for truncating on dead OST (515)
4546 - Don't error out unless we're actually accessing dead stripes (474)
4547 - Fix garbage sizes when stripes are missing (410)
4548 - LRU counters were broken, causing constant lock purge (433, 432)
4549 - garbage on read from stripes with failed OSTs (441)
4550 - mark OSCs as active before reconnecting during recovery (438)
4551 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4552 - lfind did not preserve OST order in output (443)
4553 - symlinks cause hung clients, incorrect data (439)
4554 - stop dereferencing request after dropping refcount (457)
4555 - don't LASSERT(spin_is_locked) on non-SMP (455)
4556 - fixes for many rename() bugs
4557 - fstat didn't correctly synchronize attributes (399)
4558 - server must handle lock cancellation during blocking AST prep (487)
4559 - bulk descriptors were free()d too soon (511)
4560 - fix paths in lconf, which would load incorrect modules (451, 507)
4561 - fix confusing lconf 'host not found' error message (386)
4562 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4563 - fix race condition in mdc_blocking_ast() for inode access (526)
4564 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4565 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4566 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4567 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4568 - fix lstripe in several ways (kernel side) (527)
4569 - fix request leak in ldlm_cli_enqueue (262)
4570 - incorrect OSC was marked inactive after OST failure
4571 - call mds_fs_cleanup before unmounting filesystem (524)
4572 - fix races between taking ns_lock and ldlm_lock_change_resource
4573 - fix races updating LOV export open file list
4574 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4575 - fix recovery NULL deref in ldlm_cli_cancel_unused
4576 - fix some DLM races by using new hash table for lock handles (419)
4577 - permit the client to specify desired inodes, at replay
4578 - duplicate requests when we queue them for replay reintegration
4579 - fix last_rcvd offset calculation
4580 - sync after each recovered transaction, so we always make progress
4581 - never, not always, ERESTART requests without transnos
4582 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4584 - skip replay if the MDS says that the client is already connected
4585 - don't check for a recovery-enabled export to match lctl's UUID
4586 - don't INC_USE_COUNT for phantom exports
4587 - don't crash when cleaning up phantom exports (567)
4588 - don't double-finish or set replay data for errored mdc_open requests
4589 - abort requests when they time out, so we don't get old replies
4590 - send/receive replies for AST messages again
4591 - if the client says that it doesn't have the lock, cancel it on the
4593 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4594 mark it as destroyed, it will all work out when decref is called
4595 - fix module use counts (22, 581)
4597 - ASTs now expect a reply (server cancels lock on error reply)
4599 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4602 - fix many simultaneous client startup (392)
4603 - fix dentry->d_it clobbering
4604 - credentials weren't being shipped for readdir/getattr operations
4605 - remove invalid assertions triggered during some concurrent MD
4607 - proper Lustre versions added (336, 389)
4608 - fix memory leak for create error case (398)
4609 - fix LOV locking bug that would get cli/srv out of sync
4610 - fix echo client over LOV (409)
4611 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4612 - fix double-O_EXCL intent crash (424)
4613 - avoid sending multiple lock CANCELs (352)
4615 - MDS can do multi-client recovery (modulo bugs in new code)
4617 - many updates, edits, cleanups
4619 2002-11-18 Phil Schwan <phil@clusterfs.com>
4622 - fix null d_it dereference (346)
4623 - fix full OST/dbench hang (333)
4624 - fix permission problem with file removal (286)
4625 - fix removal of OSCs from LOV when they fail
4626 - fix NULL deref during bulk timeout (214)
4627 - fix problems related to multiple filesystems on one MDS (241)
4628 - fixed serious subtle metadata locking bugs
4629 - free locks on clients when inodes are removed due to memory
4631 - fix inode pointer in lock data (285)
4632 - partial support for multiple MDS on a single host (241)
4633 - data locks weren't cancelled at clear_inode time (290, 311)
4634 - intent locks could lead to unbounded lock growth (205)
4635 - added a maximum lock count, an LRU list, and a flusher
4636 - fix multiple rename (365)
4637 - properly abstracted the echo client
4638 - OSC locked 1 byte too many; fixed
4639 - rewrote brw callback code:
4640 - fixed recovery bugs related to LOVs (306)
4641 - fixed too-many-pages-in-one-write crash (191)
4642 - fixed (again) crash in sync_io_timeout (214)
4643 - probably fixed callback-related race (385)
4645 - Add capability to MDS protocol
4646 - LDLM cancellations and callbacks on different portals
4648 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4651 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4652 - defer open object destruction to close time (601981, 138)
4653 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4654 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4655 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4656 - don't call obd_setattr in ll_file_release for destroyed objects
4657 * protocol change to lustre_msg: move |version| and add |flags|
4658 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4659 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4660 * added replay of create, unlink, link and rename operations during
4661 MDS failover; recovery should be much more robust now
4662 * remove failed OSCs from LOVs (only lov_create uses this so far)
4663 * the lustre-HOWTO was brought (more) up to date (582544)
4665 2002-10-23 Phil Schwan <phil@clusterfs.com>
4668 - in-use dentries weren't being reused properly (617851)
4669 - prevent multiple LDLM setup (599178)
4670 - fix LOV size calculations for truncate (617853)
4671 - fix client handling of MDS intent errors (POSIX)
4672 - fix permission bug in lovstripe.c test (624321)
4673 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4674 - truncate past end of file could corrupt data
4675 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4676 - a race in recovery could return ETIMEDOUT to apps (623947)
4677 - building outside the source directory was fixed
4678 * the lustre-HOWTO was brought (more) up to date (582544)
4679 * major progress was made on recovery functionality
4681 2002-10-10 Phil Schwan <phil@clusterfs.com>
4684 - recovery deadlock fix
4685 - rm -rf causes LBUG fix (617817)
4686 - file open by multiple tasks fix (618962)
4687 - directory permissions bugs (602707 and 620007)
4688 - journal_stop fixed with locking (611313)
4689 - O_APPEND failures resolved (618273, perhaps 614459)
4690 - lconf PATH fix (619770)
4691 - IA64 build fix (621450)
4692 - RPC buffer sizes scale with amount of memory
4694 2002-10-01 Phil Schwan <phil@clusterfs.com>
4697 - locks would be cancelled without throwing away data pages,
4698 resulting in inconsistent data (605627)
4699 - inode attributes were not always being refreshed (605627, 612449)
4700 - lconf now continues to cleanup after lctl reports an error
4701 - MDS now enforces user permissions (602707)
4702 - lprocfs cleanup fixed, but not yet enabled (614157)
4703 - fixed infinite server hang, should a client not respond to an AST
4704 - avoid going into recovery if user calls readlink() with a buffer
4705 that's too small (613941)
4706 - AST RPCs no longer require replies (614867) -- this may be changed
4707 - don't crash server if client sends an IOV that's too big (611336)
4708 - fixed lock conversion deadlock (611892)
4709 - fixed the following of symlinks (614622)
4710 * recovery: the server can remove locks from a client that dies, other
4711 clients can make progress
4712 * more extN patch fixes
4713 * compile-time configurable ptlrpc buffer allocations
4715 - collaborative read cache document
4716 - Lustre Lite Performance CDR document-in-progress
4718 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4721 - fix typo in patch-2.4.18
4723 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4726 - clear ptlrpc request each time in handle_incoming_request()
4727 - unlink of files now destroys the object on the OST
4729 2002-09-19 Peter Braam <braam@clusterfs.com>
4731 * add hard link support
4732 * change obdfile creation method
4733 * kernel patch changed
4735 2002-09-19 Peter Braam <braam@clusterfs.com>
4738 - stack overflow bug in extN fixed
4740 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4742 * documentation updates
4743 - add man pages for config tools
4744 - update tests/README to describe testing with new config tools
4745 - finish metadata API descriptions
4746 * bug fixes and cleanups
4747 - statfs workaround for 16TB limit
4748 - LOV stripe allocation improved, can stripe on subset of OSTs
4749 - LOV file size/IO offset was wrong for files > 4GB in size
4750 - object EA data was being dropped, caused files to be unreadable
4751 - memory overflow with non-LOV OST caused memory corruption
4752 - fixed regression tests to work with new config tools, obdfilter
4753 - fixed bug when directory size became larger than 1 block
4754 - fixed bug (for single client case) when PWD was deleted
4755 - invalidate local directory pages when doing intent-based ops
4756 - avoid LDLM oops when lock callback contained bad data
4758 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4760 * documentation updates
4761 * bug fixes and cleanups
4762 - configuration tools
4765 - 64-bit compile warnings
4766 - 64-bit internal statfs data
4768 * test_brw on persistent OST devices
4770 * lprocfs (disabled)
4772 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4774 * documentation updates
4775 * bug fixes and cleanups
4776 * configuration tools
4778 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4781 * many small fixes to 0_5_4
4782 * io/network handling
4783 * thinkos in MDS operations
4785 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4788 * crucial basic fixes to 0.5.3
4789 * IOR, Iozone work over Elan
4792 2002-08-07 Phil Schwan <phil@clusterfs.com>
4793 * version 0_5_3, our first alpha
4794 * we use the new Portals iovs
4795 * documentation updates
4796 * bug fixes and cleanups
4797 * small changes in the DLM wire protocol
4799 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4800 * version 0_5_1 with some initial stability,
4801 * locking on MD and file I/O.
4802 * documentation updates
4803 * several bug fixes since 0.5.0
4804 * small changes in wire protocol
4806 2002-07-18 Phil Schwan <phil@clusterfs.com>
4808 * delivered as Lustre Light Alpha
4809 * fixed a crash after handling invalid MDS requests
4810 * fixed directory pages for architectures with non-4k pages sizes
4812 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4813 * release version v0_4_4
4814 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4816 * Fixes a number of interruption problems with OST operations.
4817 * Update documentation for portals header changes
4818 * Move all wire protocol structs/defines to lustre_idl.h
4819 * Fixes symlink length bug.
4820 * Add tcpdump to repository.
4822 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4823 * release version v0_4_3
4824 * Fixes statfs for inodes on extN.
4825 * Fixes bug in runtests which would delete /etc/hosts.
4826 * Use 64-bit object IDs wherever possible (not into VFS though)
4827 Remove ost_get_info, which is unused by lustre, and out of date.
4829 2002-07-03 Peter Braam <braam@clusterfs.com>
4830 * release version v0_4_2 Fixes a lookup error (type not passed)
4831 * move forward to head of Portals
4832 * move forward to latest Lustre kernel
4834 2002-06-25 Peter Braam <braam@clusterfs.com>
4835 * release version v0_4_1. Hopefully stable on single node use.