1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.19.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.7-sun3
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Description: timeout with invalidate import.
18 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
19 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
20 ptlrpc_invalidate_import and as result LASSERT.
22 Severity : enhancement
24 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
28 Description: enable MGS and MDT services start separately
29 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
30 a co-located MGS without starting the MGS, which is a complement
31 to 'nosvc' mount option.
34 Frequency : always with o2ib 1.3 and sles10
36 Description: fix build with SLES10 and o2ib v3.
37 Details : sles10 uses diffrent name for Module.symver file but configure
38 assume this file has same name on RHEL/SLES/vanila kernels.
42 Frequency : after recovery
43 Description: precreate to many object's after del orphan.
44 Details : del orphan st in oscc last_id == next_id and this triger growing
45 count of precreated objects. Set flag LOW to skip increase count
46 of precreated objects.
50 Frequency : rare, on clear nid stats
51 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
52 Details : when clean nid stats sometimes try destroy live entry,
53 and this produce panic in free.
57 Description: Stack overflow during MDS log replay
58 ease stack pressure by using a thread dealing llog_process.
62 Description: wait until IO finished before start new when do lock cancel.
63 Details : VM protocol want old IO finished before start new, in this case
64 need wait until PG_writeback is cleared until check dirty flag and
65 call writepages in lock cancel callback.
67 Severity : enhancement
69 Description: using special macro for print time and cleanup in includes.
73 Description: mds_mfd_close() ASSERTION(rc == 0)
74 Details : In mds_mfd_close(), we need protect inode's writecount change
75 within its orphan write semaphore to prevent possible races.
79 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
83 Frequency : rare, on shutdown ost
84 Description: don't hit live lock with umount ost.
85 Details : shrink_dcache_parent can be in long loop with destroy dentries,
86 use shrink_dcache_sb instead.
90 Description: don't panic with use echo client
91 Details : echo client pass NULL as client nid pointer and this produce null
96 Description: fix build on ppc32
97 Details : compile code with -m64 flag produce wrong object file for ppc32.
101 Description: add message levels for liblustreapi
105 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
106 Details : Move the 'good_osts' check before the 'total_bavail' check. This
107 will result in an -EAGAIN and in the exit call path we call
108 alloc_rr() which will with increasing aggressiveness attempt to
109 aquire precreated objects on the minimum number of required OSCs.
113 Description: Use old size assignment to avoid deadlock
114 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
115 the scheduling while holding a spinlock. We do not need locking
116 for size in ll_update_inode() because size is only updated from
117 the MDS for directories or files without objects, so there is no
118 other place to do the update, and concurrent access to such inodes
119 are protected by the inode lock.
123 Description: resolve "_IOWR redefined" build error on SLES10
127 Description: dump the memory debugging after all modules are unloaded to
128 suppress false negative in conf_sanity test 39
130 Severity : enhancement
132 Description: build kernel-ib packages for OFED 1.3 in our release cycle
137 Description: fix SLES kernel versioning
138 Details : the kernel version for our SLES 10 kernel did not include a "-"
139 before the "smp" at the end. while this was not a problem in
140 general, it did mean that software trying to use the kernel
141 version to try to detect a vendor specific kernel would fail.
142 this was most evident by the OFED build scripts.
146 Description: Don't update lov_desc members until making sure they are valid
147 Details : When updating lov_desc members via proc fs, need fix their
148 validities before doing the real update.
152 Description: don't put request into delay list while invalidate in flight.
153 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
154 import in flight. this produce timeout for invalidate and sometimes
155 can cause stale data.
159 Frequency : on ppc only
160 Description: not convert ost objects for directory because it's not exist.
161 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
165 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
167 Severity : enhancement
169 Description: Update to support 2.6.22.14 vanilla kernel.
173 Frequency : rare, on recovery
174 Description: read procfs can produce deadlock in some situation
175 Details : Holding lprocfs lock which send rpc can produce block for destroy
176 obd objects and this also block reconnect with -EALREADY. This isn't
177 fix all lprocfs bugs - but make it rare.
179 Severity : enhancement
181 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
184 Frequency : frequent on X2 node
186 Description: mdc_set_open_replay_data LBUG
187 Details : Set replay data for requests that are eligible for replay.
191 Description: lustre_mgs: operation 101 on unconnected MGS
192 Details : When MGC is disconnected from MGS long enough, MGS will evict the
193 MGC, and late on MGC cannot successfully connect to MGS and a lot
194 of the error messages complaining that MGS is not connected.
198 Frequency : on network error
199 Description: panic with double free request if network error
200 Details : mdc_finish_enqueue is finish request if any network error ocuring,
201 but it's true only for synchronus enqueue, for async enqueue
202 (via ptlrpcd) this incorrect and ptlrpcd want finish request
205 Severity : enhancement
207 Description: client-side metadata stat-ahead during readdir(directory readahead)
208 Details : perform client-side metadata stat-ahead when the client detects
209 readdir and sequential stat of dir entries therein
212 Frequency : on start mds
214 Description: Implement get_info(last_id) in obdfilter.
217 Frequency : occasional
219 Description: Correctly check stale fid, not start epoch if ost not support SOM
220 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
221 overwrited with new generated fid, but mds can anwer with one of these
222 two fids and both is not stale. setattr incorectly start epoch and
223 assume will be called done_writeting, but without SOM done_writing
227 Frequency : rare, depends on device drivers and load
229 Description: MDS or OSS nodes crash due to stack overflow
230 Details : Code changes in 1.8.0 increased the stack usage of some functions.
231 In some cases, in conjunction with device drivers that use a lot
232 of stack the MDS (or possibly OSS) service threads could overflow
233 the stack. One change which was identified to consume additional
234 stack has been reworked to avoid the extra stack usage.
237 Frequency : occasional
239 Description: Do not fail import if osc_interpret_create gets -EAGAIN
240 Details : If osc_interpret_create got -EAGAIN it immediately exits and
241 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
242 oscc_has_objects and see OSC has no objests and call
243 oscc_internal_create to resend create request.
245 Severity : enhancement
247 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
249 Severity : enhancement
251 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
254 Frequency : very rare
256 Description: Fix replay if there is an un-replied request and open
257 Details : In some cases, older replay request will revert the
258 mcd->mcd_last_xid on MDS which is used to record the client's
261 Severity : enhancement
263 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
265 Severity : enhancement
267 Description: Add rhel5 support to HEAD.
269 Serverity : enhancement
271 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
276 Description: Don't allow skipping OSTs if index has been specified.
277 Details : Don't allow skipping OSTs if index has been specified, make locking
278 in internal create lots better.
282 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
283 Details : access to bitfield in structure is always rounded to long
284 and this produce problem with not atomic change any bit.
288 Description: Lustre make rpms failed.
289 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
290 builds Lustre rpms from tarball.
293 Frequency : rare on shutdown ost
295 Description: If llog cancel was not send before clean_exports phase, this can
296 produce deadlock in llog code.
297 Details : If llog thread has last reference to obd and call class_import_put
298 this produce deadlock because llog_cleanup_commit_master wait when
299 last llog_commit_thread exited, but this never success because was
300 called from llog_commit_thread.
304 Description: allow userland application know is lost one of stripes.
305 Details : fill lvb_blocks with error code on ost and return it to
306 application if error flag found.
310 Description: NULL lov_tgts causing MDS oops
311 Details : more safe checks for NULL lov_tgts for avoid oops.
313 Severity : enhancement
315 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
319 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
320 Details : If an OST is down the MDS will hang indefinitely in
321 obd_statfs_rqset() waiting for the statfs data. While for
322 MDS QOS usage of statfs, it should not stuck in waiting.
324 Severity : enhancement
326 Description: remote_acl support
327 Details : Support ACL-based permission check for remote user.
328 Support setfacl/getfacl for remote user with the utils
329 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
330 the system "{s,g}etfacl" utils.
332 Severity : enhancement
334 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
336 Severity : enhancement
338 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
342 Description: make mgs_setparam() handle fsname containing dash
343 Details : fsname containing a dash does not work with lctl conf_param
345 Severity : enhancement
347 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
349 Severity : enhancement
351 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
355 Frequency : rare, at shutdown
356 Description: access already free / zero obd_namespace.
357 Details : if client_disconnect_export was called without force flag set,
358 and exist connect request in flight, this can produce access to
359 NULL pointer (or already free pointer) when connect_interpret
360 store ocd flags in obd_namespace.
364 Frequency : only at startup
365 Description: not alloc memory with spinlock held.
366 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
367 if any spinlock held.
369 Severity : enhancement
371 Description: make lustre randomly fail allocating memory
372 Details : Make lustre randomly failed allocating memory for testing purpose.
374 Severity : enhancement
376 Description: lost problems with lov objid file
377 Details : Fixes some scability and access to not inited memory problems
378 in work with lov objdid file.
383 Description: lfs find does not continue on file error
384 Details : Continue other files processing when a file/dir is absent.
388 Description: Inconsistent usage of lustre_pack_reply()
389 Details : Standardize the usage of lustre_pack_reply() such that it
390 always generate a CERROR on failure.
393 Frequency : occasional
395 Description: MDS hang or stay in waiting lock
396 Details : If client receive lock with CBPENDING flag ldlm need send lock
397 cancel as separate rpc, to avoid situation when cancel request
398 can't processed due all i/o threads stay in wait lock.
402 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
406 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
410 Description: Update to SLES9 kernel-2.6.5-7.287.3.
414 Description: Update to RHEL4 latest kernel.
416 Severity : enhancement
418 Description: Build SLES10 patchless client fails
419 Details : The configure was broken by run ./configure with
420 --with-linux-obj=.... argument for patchless client. When the
421 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
422 can't search header adequately. Use absolute path such as
423 -I($LINUX)/include instead.
427 Description: interrupt oig_wait produce painc on resend.
428 Details : brw_redo_request can be used for resend requests from ptlrpcd and
429 private set, and this produce situation when rq_ptlrpcd_data not
430 copyed to new allocated request and triggered LBUG on assert
431 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
432 ptlrpcd set if request is changed and can be safety changed to use
437 Description: LASSERT_{REQ,REP}SWAB macros are buggy
438 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
439 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
440 these macros and replace them with their difinitions instead.
444 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
445 Details : Update which_patch & target file for SLES10 latest kernel.
448 Description: Remove client patches from SLES 10 kernel.
449 Details : This causes SLES 10 clients to behave as patchless clients
450 even on a Lustre-patched (server) kernel.
452 Severity : enhancement
454 Description: self-adjustable client's lru lists
455 Details : use adaptive algorithm for managing client cached locks lru
456 lists according to current server load, other client's work
457 pattern, memory activities, etc. Both, server and client
458 side namespaces provide number of proc tunables for controlling
461 Severity : enhancement
463 Description: light-weight GSS support
464 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
465 and reduce performance overhead.
467 Severity : enhancement
469 Description: Linux keyring support
470 Details : Support using service of Linux keyring for Lustre GSS internal
471 context refresh/cache mechanism.
475 Description: Fix errors in lfs documentation
476 Details : Fixes man pages
480 Description: don't use GFP_* in generic Lustre code.
481 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
485 Description: obdclass is limited by single OBD_ALLOC(idarray)
486 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
490 Description: warnings with build patchless client with vanila 2.6.19 and up
491 Details : change old ctl_table style and replace ctl_table/ctl_table_header
492 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
496 Frequency : Only for SLES
497 Description: sanity_quota fail test_1
498 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
499 file, which makes TSTID[2] inunique.
503 Frequency : Only for SLES9
504 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
505 Details : Update target/ChangeLog/which_patch .
509 Frequency : Only for test_10 in sanity-quota.sh
510 Description: a bug in quota test code
511 Details : checking a test flag in a wrong place
515 Frequency : Only for test_13 in sanity-quota.sh
516 Description: a bug in quota test script
517 Details : shouldn't check overfull allocated quota
521 Description: jbd statistics
522 Details : Port older jbd statistics patch for sles10
524 Severity : enhancement
526 Description: Add printing inode info into message about error in writepage.
531 Description: reply_lock_interpret crash due to race with it and lock cancel.
532 Details : Do not replay locks that are being cancelled. Do not reference
533 locks by their address during replay, just by their handle.
537 Frequency : When flocks are used.
538 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
539 Details : Flock locks might destroy just granted lock if it could be merged
540 with another existing flock, this is done in completion handler,
541 so teach ldlm_cli_enquque_fini that this is a valid case for
547 Description: Oops in read and write path when failing to allocate lock.
548 Details : Check if lock allocation failed and return error back.
552 Description: lstripe command fails for valid OST index
553 Details : The stripe offset is compared to lov->desc.ld_tgt_count
554 instead of lov->desc.ld_active_tgt_count.
558 Description: sanity.sh failed test 103
559 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
564 Description: df doesn't work properly if diskfs blocksize != 4K
565 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
569 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
570 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
574 Description: obd_health_check_timeout too short
575 Details : set obd_health_check_timeout as 1.5x of obd_timeout
579 Description: llapi_file_create() does not allow some changes
580 Details : add llapi_file_open() that allows specifying the mode and
581 open flags, and also returns an open file handle.
585 Description: merge and cleanup kernel patches.
586 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
590 Description: Add journal checksum support.(Kernel part)
591 Details : The journal checksum feature adds two new flags i.e
592 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
593 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
594 indicates that the commit block contains the checksum for
595 the blocks described by the descriptor blocks. Now commit
596 record can be sent to disk without waiting for descriptor
597 blocks to be written to disk. This behavior is controlled
598 using JBD2_FEATURE_ASYNC_COMMIT flag.
602 Description: OSS needs mutliple precreate threads
603 Details : Add ability to start more than one create thread per OSS.
607 Description: Sanity.sh test_65k failed: setstripe should have succeeded
608 Details : execute lfs setstripe on client
612 Description: mds_obd_create error creating tmp object
613 Details : When the user sets quota on root, llog will be affected and can't
614 create files and write files.
617 Frequency : Always on ia64 patchless client, and possibly others.
619 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
620 Details : This allows the patchless client to be loaded on architectures
625 Description: RedHat Update kernel for RHEL5
626 Details : Add the kernel config file for RHEL5.
630 Description: RedHat Update kernel for RHEL5
631 Details : Modify the kernel config file more closer RHEL5.
635 Description: Build failure against Centos5 (RHEL5)
636 Details : Define PAGE_SIZE when it isn't present.
640 Description: test_7 on sanity-quota.sh can't be used on separate machines
641 Details : add facet to handle it
645 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
646 Details : with NFS, the anon dentry's parent was set to itself in
647 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
648 corresponding dentry other than use rec->ur_name.
650 Severity : enhancement
652 Description: lfs setstripe enhancement
653 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
657 Description: enable data checksumming by default
658 Details : enable checksum by default, allow --disable-checksum
659 configure option and "-o nochecksum" mount option. Checksums
660 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
664 Description: lustre support for RHEL5
665 Details : Add support for RHEL5.
669 Description: Client eviction due to failover config
670 Details : after a connection loss, the lustre client should attempt to
671 reconnect to the last active server first before trying the
672 other potential connections.
676 Description: when mds and osts use different quota unit(32bit and 64bit),
677 quota will be released repeatly.
678 Details : void sending multiple quota reqs to mds, which will keep the
679 status between the reqs.
683 Description: osts not allocated evenly to files
684 Details : change the condition to increase offset_idx
687 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
689 Description: Data corruption for OSTs that are formatted larger than 2TB
691 Details : When generating the bio request for lustre file writes the
692 sector number would overflow a temporary variable before being
693 used for the IO. The data reads correctly from Lustre (which
694 will overflow in a similar manner) but other file data or
695 filesystem metadata may be corrupted in some cases.
699 Description: Tune the kernel for good SCSI performance.
700 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
701 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
706 Description: rewrite ext2-derived code in obdclass/uuid.c
707 Details : rewrite inherited code (uuid parsing code from ext2 utils)
708 from scratch preserving functionality.
712 Description: "lfs find -obd UUID" prints directories
713 Details : "lfs find -obd UUID" will return all directory names instead
714 of just file names. It is incorrect because the directories
715 do not reside on the OSTs.
719 Description: use i_size_read and i_size_write in 2.6 port
720 Details : replace inode->i_size access with i_size_read/write()
723 Frequency : when removing large files
725 Description: scheduling issue during removal of large Lustre files
726 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
727 It causes scheduling issues when removing large files (17TB in the
731 Frequency : only with liblustre clients on XT3
733 Description: evictions taking too long
734 Details : allow llrd to evict clients directly on OSTs
737 Frequency : only on ppc
739 Description: /proc/fs/lustre/devices broken on ppc
740 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
741 We should make sure the type of 'index' is loff_t and then cast
742 explicitly as needed below. Do not assign an explicitly cast
746 Frequency : only for rhel5
748 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
749 Details : Modify the target file & which_kernel.
751 Severity : enhancement
753 Description: omit set fsid for export NFS
754 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
755 as FS_REQUIRES_DEV for avoid problems with generate fsid.
759 Frequency : Always, for kernels after 2.6.16
760 Description: Fix warning idr_remove called for id=.. which is not allocated.
761 Details : Last kernels save old s_dev before kill super and not allow
762 to restore from callback - restore it before call kill_anon_super.
766 Description: buffer overruns could theoretically occur
767 Details : llapi_semantic_traverse() modifies the "path" argument by
768 appending values to the end of the origin string, and a
769 overrun may occur. Adding buffer overrun check in liblustreapi.
773 Description: Fix error on 'ls .' at the top of the Lustre mount.
774 Details : Don't revalidate dentry if it is a root dentry.
778 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
779 Details : Modify vm-tunables-rhel4.patch.
783 Description: Kernel config for 2.6.18-vanilla.
784 Details : Modify targets/2.6-vanilla.target.in.
785 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
786 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
787 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
788 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
792 Description: improve handling recoverable errors
793 Details : if request processig with error which can be recoverable on server
794 request should be resend, otherwise page released from cache and
799 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
800 Details : Modify target file & which_patch.
801 A flaw was found in the IA32 system call emulation provided
802 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
803 value could be stored in the %RAX register, which could trigger an
804 out-of-bounds system call table access. An untrusted local user
805 could exploit this flaw to run code in the kernel
806 (ie a root privilege escalation). (CVE-2007-4573).
810 Description: O_DIRECT bypasses client statistics.
811 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
812 read_ahead_stats were updated. copied the stats section from
813 osc_send_oap_rpc() into async_internal().
817 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
821 Description: change order of libsysio includes
822 Details : '#include sysio.h' should always come before '#include xtio.h'
826 Description: handle "serious error: objid * already exists" more gracefully
827 Details : If LAST_ID value on disk is smaller than the objects existing in
828 the O/0/d* directories, it indicates disk corruption and causes an
829 LBUG(). If the object is 0-length, then we should use the existing
830 object. This will help to avoid a full fsck in most cases.
832 Severity : enhancement
834 Description: adapt the lustre_config script to support the upgrade case
835 Details : Add "-u" option for lustre_config script to support upgrading 1.4
836 server targets to 1.6 in parallel.
840 Description: To avoid grant space > avaible space when the disk is almost
841 full. Without this patch you might see the error "grant XXXX >
842 available" or some LBUG about grant, when the disk is almost
844 Details : In filter_check_grant, for non_grant cache write, we should
845 check the left space by if (*left > ungranted + bytes), instead
846 of (*left > ungranted), because only we are sure the left space
847 is enough for another "bytes", then the ungrant space should be
848 increase. In client, we should update cl_avail_grant only there
849 is OBD_MD_FLGRANT in the reply.
853 Description: Update RHEL 4 kernel to fix local root privilege escalation.
854 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
855 described in CVE-2007-4573. This problem could allow untrusted
856 local users to gain root access.
859 Frequency : when using O_DIRECT and quotas
861 Description: Incorrect file ownership on O_DIRECT output files
862 Details : block usage reported by 'lfs quota' does not take into account
863 files that have been written with O_DIRECT.
868 Description: touch file failed when fs is not full
869 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
870 otherwise we can get ENOSP while fs is not full.
874 Description: parallel lock callbacks
875 Details : Instead of sending blocking and completion callbacks as separated
876 requests, adding them to a set and sending in parallel.
879 Frequency : only for Cray XT3
880 Bugzilla : 12829/13455
881 Description: Changing primary group doesn't change the group lustre assigns to
883 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
884 overridden with the primary group provided by the group upcall,
885 whereas the supplied fsgid can be trusted if it is in the list of
886 supplementary groups returned by the group upcall.
888 Severity : enhancement
890 Description: Allow masking D_WARNING, D_ERROR messages from console
891 Details : Console messages can now be disabled via lnet.printk.
895 Description: User code with malformed file open parameter crashes client node
896 Details : Before packing join_file req, all the related reference should be
897 checked carefully in case some malformed flags cause fake
898 join_file req on client.
902 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
903 Details : ldlm_completion_ast() assumes that a lock is granted when the req
904 mode is equal to the granted mode. However, it should also check
905 that LDLM_FL_CP_REQD is not set.
909 Description: Heavy nfs access might result in deadlocks
910 Details : After ELC code landed, it is now improper to enqueue any mds
911 locks under och_sem, because enqueue might want to decide to
912 cancel open locks for same inode we are holding och_sem for.
916 Description: Client eviction while running blogbench
917 Details : A lot of unlink operations with concurrent I/O can lead to a
918 deadlock causing evictions. To address the problem, the number of
919 oustanding OST_DESTROY requests is now throttled to
920 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
921 callbacks are processed in priority.
925 Description: enable ACLs on MDS by default
926 Details : ACLs must be enabled on MDS by default.
929 Frequency : PPC/PPC64 only
931 Description: conflicts between asm-ppc64/types.h and lustre_types.h
932 Details : fix duplicated definitions between asm-ppc64/types.h and
933 lustre_types.h on PPC.
936 Frequency : PPC/PPC64 only
938 Description: asm-ppc/segment.h does not exist
939 Details : fix compile issue on PPC.
943 Description: better handle error messages in extents code
946 Frequency : RHEL4 only
948 Description: mkfs is very slow on IA64/RHEL4
949 Details : A performance regression has been discovered in the MPT Fusion
950 driver between versions 3.02.73rh and 3.02.99.00rh. As a
951 consequence, we have downgraded the MPT Fusion driver in the RHEL4
952 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
954 Severity : enhancement
956 Description: SNMP support enhancement
957 Details : Adding total number of sampled request for an MDS node in snmp
960 Severity : enhancement
962 Description: Optimize ldlm waiting list processing for PR extent locks
963 Details : When processing waiting list for read extent lock and meeting read
964 lock that is same or wider to it that is not contended, skip
965 processing rest of the list and immediatelly return current
966 status of conflictness, since we are guaranteed there are no
967 conflicting locks in the rest of the list.
971 Description: Time out and refuse to reconnect
972 Details : When the failover node is the primary node, it is possible
973 to have two identical connections in imp_conn_list. We must
974 compare not conn's pointers but NIDs, otherwise we can defeat
975 connection throttling.
979 Description: port llog fixes from b1_6 into HEAD
980 Details : Port llog reference couting and some llog cleanups from b1_6
981 (bug 10800) into HEAD, for protect from panic and access to already
982 free llog structures.
986 Description: Detect stride IO mode in read-ahead
987 Details : When a client does stride read, read-ahead should detect that and
988 read-ahead pages according to the detected stride pattern.
992 Description: data checksumming impacts single node performance
993 Details : add support for several checksum algorithm. Currently, only CRC32
994 and Adler-32 are supported. The checksum type can be changed on
995 the fly via /proc/fs/lustre/osc/*/checksum_type.
999 Description: use adler32 for page checksums
1000 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1005 Description: build for x2 fails
1006 Details : fix compile issue on Cray systems.
1010 Description: Properly match for duplicate locks
1011 Details : Due to different lock order from skiplists code, we need to
1012 traverse entire list for now
1015 Frequency : only on PPC/SLES10
1017 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1018 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1019 be defined. Add a hack in mkfs_lustre.c to work around this compile
1024 Description: LASSERT on MDS when client holding flock lock dies
1025 Details : ldlm pool logic depends on number of granted locks equal to
1026 number of released locks which is not true for flock locks, so
1027 just exclude such locks from consideration.
1031 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1032 Details : Use fsfilt_sync() for both the whole filesystem sync and
1033 individual file sync to eliminate dangerous inode locking
1034 with I_LOCK that can lead to a deadlock.
1038 Description: performance in 1.6.3
1039 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1040 sure that 1MB requests are not fragmented by the block layer.
1042 Severity : enhancement
1044 Description: organize the server-side client stats on per-nid basis
1045 Details : Change the structure of stats under obdfilter and mds to
1053 The "uuid"s file would list the uuids of _active_ exports.
1054 And the clear entry is to clear all stats and stale nids.
1056 Severity : enhancement
1058 Description: eliminate client locks in face of contention
1059 Details : file contention detection and lockless i/o implementation
1060 for contended files.
1064 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1067 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1069 Description: MDS may lose file striping (and hence file data) in some cases
1070 Details : If there are additional extended attributes stored on the MDS,
1071 in particular ACLs, SELinux, or user attributes (if user_xattr
1072 is specified for the client mount options) then there is a risk
1073 of attribute loss. Additionally, the Lustre file striping
1074 needs to be larger than default (e.g. striped over all OSTs),
1075 and an additional attribute must be stored initially in the
1076 inode and then increase in size enough to be moved to the
1077 external attribute block (e.g. ACL growing in size) for file
1082 Description: skiplist implementation simplification
1083 Details : skiplists are used to group compatible locks on granted list
1084 that was implemented as tracking first and last lock of each lock group
1085 the patch changes that to using doubly linked lists
1089 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1090 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1091 because LASSERT on both the data supplied by a client, and the data
1092 on disk is dangerous and incorrect.
1094 Severity : enhancement
1096 Description: Slow trucate/writes to huge files at high offsets.
1097 Details : Directly associate cached pages to lock that protect those pages,
1098 this allows us to quickly find what pages to write and remove
1099 once lock callback is received.
1101 --------------------------------------------------------------------------------
1103 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1105 * Support for kernels:
1106 2.6.5-7.283 (SLES 9),
1107 2.6.9-55.EL (RHEL 4),
1108 2.6.16.46-0.14 (SLES 10),
1109 2.6.18.8 vanilla (kernel.org)
1110 * Client support for unpatched kernels:
1111 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1112 2.6.16 - 2.6.22 vanilla (kernel.org)
1113 * Due to recently discovered recovery problems, we do not recommend
1114 using patchless RHEL 4 clients with this or any earlier release.
1115 * Recommended e2fsprogs version: 1.39.cfs8
1116 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1117 * Starting with this release, the ldiskfs backing filesystem required
1118 by Lustre is now in its own package, lustre-ldiskfs. This package
1119 should be installed. It is versioned separately from Lustre and
1120 may be released separately in future.
1124 Description: block reactivating mgc import until all deactivates complete
1125 Details : Fix race when failing back MDT/MGS to itself (testing)
1128 Frequency : at statup only
1130 Description: mds_lov_synchronize race leads to various problems
1131 Details : simultaneous MDT->OST connections at startup can cause the
1132 sync to abort, leaving the OSC in a bad state.
1134 Severity : enhancement
1136 Description: add optional extra BUILD_VERSION info
1137 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1138 to override the lustre version.
1141 Frequency : 2.6.18 servers only
1143 Description: ll_kern_mount() doesn't release the module reference
1144 Details : The ldiskfs module reference count never drops down to 0
1145 because ll_kern_mount() doesn't release the module reference.
1150 Description: server LBUG when using old ost_num_threads parameter
1151 Details : Accept the old ost_num_threads parameter but warn that it
1152 is deprecated, and fix an off-by-one error that caused an LBUG.
1157 Description: Transient SCSI error results in persistent IO issue
1158 Details : iobuf->dr_error is not reinitialized to 0 between two
1162 Frequency : sometimes when underlying device returns I/O errors
1164 Description: OSTs not going read-only during write failures
1165 Details : OSTs are not remounted read-only when the journal commit threads
1166 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1167 instead of the ext3 wrappers.
1171 Description: poor connect scaling with increasing client count
1172 Details : Don't run filter_grant_sanity_check for more than 100 exports
1173 to improve scaling for large numbers of clients.
1176 Frequency : SLES10 only
1178 Description: sanity-quota.sh quotacheck failed: rc = -22
1179 Details : Quotas cannot be enabled on SLES10.
1182 Frequency : liblustre clients only
1184 Description: getdirentries does not give error when run on compute nodes
1185 Details : getdirentries does not fail when the size specified as an argument
1186 is too small to contain at least one entry
1188 Severity : enhancement
1190 Description: Add LNET router traceability for debug purposes
1191 Details : If a checksum failure occurs with a router as part of the
1192 IO path, the NID of the last router that forwarded the bulk data
1193 is printed so it can be identified.
1198 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1199 Details : Due to a race condition, liblustre clients were occasionally
1200 evicted incorrectly.
1202 Severity : enhancement
1204 Description: lfs setstripe use optional parameters instead of postional
1207 Severity : enhancement
1209 Description: Nanosecond timestamp support for ldiskfs
1210 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1211 resolution timestamps. There is not yet support for this at
1212 the Lustre filesystem level.
1215 Frequency : during server recovery
1217 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1218 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1222 Frequency : nfs export on patchless client
1224 Description: connectathon hang when test nfs export over patchless client
1225 Details : Disconnected dentry cannot be found with lookup, so we do not need
1226 to unhash it or make it invalid
1229 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1230 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1231 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1236 Description: open req refcounting wrong on reconnect
1237 Details : If reconnect happened between getting open reply from server and
1238 call to mdc_set_replay_data in ll_file_open, we will schedule
1239 replay for unreferenced request that we are about to free.
1240 Subsequent close will crash in variety of ways.
1241 Check that request is still eligible for replay in
1242 mdc_set_replay_data().
1247 Description: disable writes to filesystem when reading health_check file
1248 Details : the default for reading the health_check proc file has changed
1249 to NOT do a journal transaction and write to disk, because this
1250 can cause reads of the /proc file to hang and block HA state
1251 checking on a healthy but otherwise heavily loaded system. It
1252 is possible to return to the previous behaviour during configure
1253 with --enable-health-write.
1255 Severity : enhancement
1257 Description: 64-bit inode version
1258 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1259 to the inode. This will be required for version-based recovery.
1264 Description: MDS fails to start if a duplicate client export is detected
1265 Details : in some rare cases it was possible for a client to connect to
1266 an MDS multiple times. Upon recovery the MDS would detect this
1267 and fail during startup. Handle this more gracefully.
1269 Severity : enhancement
1271 Description: Add -o localflock option to simulate old noflock
1273 Details : This will achieve local-only flock/fcntl locks
1279 Description: log_commit_thread vs filter_destroy race leads to crash
1280 Details : Take import reference before releasing llog record semaphore
1285 Description: Wrong request locking in request set processing
1286 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1287 imp_delayed_list, in this place should be used imp_lock.
1290 Frequency : when reconnection
1292 Description: Grant Leak when osc reconnect to OST
1293 Details : When osc reconnect ost, OST(filter) should check whether it
1294 should grant more space to client by comparing fed_grant and
1295 cl_avail_grant, and return the granted space to client instead
1296 of "new granted" space, because client will call osc_init_grant
1297 to update the client grant space info.
1300 Frequency : when client reconnect to OST
1302 Description: Grant Leak when osc do resend and replay bulk write
1303 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1304 bulk write request, because the grant info while be sync between
1305 OSC and OST when reconnect, and we should ignore the grant info
1306 these of resend/replay write req.
1311 Description: Grant space more than avaiable left space sometimes.
1312 Details : When then OST is about to be full, if two bulk writing from
1313 different clients came to OST. Accord the avaliable space of the
1314 OST, the first req should be permitted, and the second one
1315 should be denied by ENOSPC. But if the seconde arrived before
1316 the first one is commited. The OST might wrongly permit second
1317 writing, which will cause grant space > avaiable space.
1320 Frequency : when client is evicted
1322 Description: Grant might be wrongly erased when osc is evicted by OST
1323 Details : when the import is evicted by server, it will fork another
1324 thread ptlrpc_invalidate_import_thread to invalidate the
1325 import, where the grant will be set to 0. While the original
1326 thread will update the grant it got when connecting. So if
1327 the former happened latter, the grant will be wrongly errased
1328 because of this race.
1333 Description: Checking Stale with correct fid
1334 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1335 to check whether it is stale, because sometimes, we want the
1336 enqueue happened anyway, and op_data.fid2 will not be initialized.
1338 Severity : enhancement
1340 Description: update patchless client
1341 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1344 Frequency : only with 2.4 kernel
1346 Description: random memory corruption
1347 Details : size of struct ll_inode_info is to big for union inode.u and this
1348 can be cause of random memory corruption.
1353 Description: Memory leak in recovery
1354 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1355 It should also check obd_fail before fsfilt_start, otherwise if
1356 fsfilt_start return -EROFS,(failover mds during mds recovery).
1357 then the req will return with repmsg->transno = 0 and rc = EROFS.
1358 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1359 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1360 be freed no matter whether fsfilt_commit success or not.
1363 Frequency : only with huge count clients
1365 Description: Prevents from taking the superblock lock in llap_from_page for
1367 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1368 for llap_from_page call in ll_removepage prevents from taking the
1369 superblock lock for a soon died page.
1374 Description: Not check open intent error before release open handle
1375 Details : in some rare cases, the open intent error is not checked before
1376 release open handle, which may cause
1377 ASSERTION(open_req->rq_transno != 0), because it tries to release
1378 the failed open handle.
1383 Description: Set cat log bitmap only after create log success.
1384 Details : in some rare cases, the cat log bitmap is set too early. and it
1385 should be set only after create log success.
1389 Description: Accessing a block bevice can re-enable I/O when Lustre is
1390 tearing down a device.
1391 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1396 Description: service threads may hog cpus when there are a lot of requests
1397 Details : Insert cond_resched to give other threads a chance to use some CPU
1402 Description: the cat log was not initialized in recovery
1403 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1404 unlink log on mds will not be initialized until mds post
1405 recovery. And also in mds post recovery, the unlink log will
1406 initialization will be done asynchronausly, so there will be race
1407 between add unlink log and unlink log initialization.
1411 Description: brw_stats were being printed incorrectly
1412 Details : brw_stats were being printed as log2 but all of them were not
1413 recorded as log2. Also remove some code duplication arising from
1414 filter_tally_{read,write}.
1418 Frequency : rare, only in recovery.
1419 Description: ASSERTION(req->rq_type != LI_POISON) failed
1420 Details : imp_lock should be held while iterating over imp_sending_list for
1421 prevent destroy request after get timeout in ptlrpc_queue_wait.
1425 Description: replay-single.sh test 52 fails
1426 Details : A lock's skiplist need to be cleanup when it being unlinked
1427 from its resource list.
1431 Description: Short directio read returns full requested size rather than
1433 Details : Direct I/O operations should return actual amount of bytes
1434 transferred rather than requested size.
1436 Severity : enhancement
1438 Description: metadata RPC reduction (e.g. for rm performance)
1439 Details : decrease the amount of synchronous RPC between clients and servers
1440 by canceling conflicing lock before the operation on the client
1441 and packing thier handles into the main operation RPC to server.
1443 Severity : enhancement
1445 Description: Async OSC create to avoid the blocking unnecessarily.
1446 Details : If a OST has no remain object, system will block on the creating
1447 when need to create a new object on this OST. Now, ways use
1448 pre-created objects when available, instead of blocking on an
1449 empty osc while others are not empty. If we must block, we block
1450 for the shortest possible period of time.
1452 Severity : enhancement
1454 Description: refine locking for avoid write wrong info into lov_objid file
1455 Details : fix possible races with add new target and write/update data in
1458 --------------------------------------------------------------------------------
1460 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1465 Frequency : on some architectures
1467 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1468 Details : Uninitialized flags sometimes cause configuration commands to
1472 Frequency : patchless clients only
1474 Description: missing __iget() symbol export
1475 Details : The __iget() symbol export is missing. To avoid the need for
1476 this on patchless clients the deathrow inode reaper is turned
1477 off, and we depend on the VM to clean up old inodes. This
1478 dependency was during via the fix for bug 12181.
1480 --------------------------------------------------------------------------------
1482 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1484 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1485 INTEROPERATE with older versions automatically. In many cases a
1486 special upgrade step is needed. Please read the
1487 user documentation before upgrading any part of a live system.
1488 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1489 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1490 * WARNING: Lustre configuration and startup changes are required with
1491 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1497 Severity : enhancement
1499 Description: MountConf
1500 Details : Lustre configuration is now managed via mkfs and mount
1501 commands instead of lmc and lconf. New obd types (MGS, MGC)
1502 are added for dynamic configuration management. See
1503 https://mail.clusterfs.com/wikis/lustre/MountConf for
1506 Severity : enhancement
1508 Description: dynamic OST addition
1509 Details : OSTs can now be added to a live filesystem
1511 Severity : enhancement
1513 Description: startup order invariance
1514 Details : MDTs and OSTs can be started in any order. Clients only
1515 require the MDT to complete startup.
1517 Severity : enhancement
1519 Description: parallel, asynchronous orphan cleanup
1520 Details : orphan cleanup is now performed in separate threads for each
1521 OST, allowing parallel non-blocking operation.
1523 Severity : enhancement
1525 Description: optimized stripe assignment
1526 Details : stripe assignments are now made based on ost space available,
1527 ost previous usage, and OSS previous usage, in order to try
1528 to optimize storage space and networking resources.
1530 Severity : enhancement
1532 Description: Permanently set tunables
1533 Details : All writable /proc/fs/lustre tunables can now be permanently
1534 set on a per-server basis, at mkfs time or on a live system.
1536 Severity : enhancement
1538 Description: Lustre message v2
1539 Details : Add lustre message format v2.
1541 Severity : enhancement
1543 Description: client OST exclusion list
1544 Details : Clients can be started with a list of OSTs that should be
1545 declared "inactive" for known non-responsive OSTs.
1549 Description: SPEC SFS validation failure on NFS v2 over lustre.
1550 Details : Changes the blocksize for regular files to be 2x RPC size,
1551 and not depend on stripe size.
1553 Severity : enhancement
1555 Description: Multiple MD RPCs in flight.
1556 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1557 To avoid overly-overloading MDS, introduce a limit on number of
1558 MDS RPCs in flight for a single client and add /proc controls
1559 to adjust this limit.
1561 Severity : enhancement
1563 Description: client read/write statistics
1564 Details : Add client read/write call usage stats for performance
1565 analysis of user processes.
1566 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1567 file access. extents_stats shows chunk size distribution.
1568 extents_stats_per_process show chunk size distribution per
1571 Severity : enhancement
1573 Description: mds statistics
1574 Details : Add detailed mds operations statistics in
1575 /proc/fs/lustre/mds/*/stats.
1579 Description: Failure of copying files with lustre special EAs.
1580 Details : Client side always return success for setxattr call for lustre
1581 special xattr (currently only "trusted.lov").
1586 Description: Refcount LNET uuids
1587 Details : The global LNET uuid list grew linearly with every startup;
1588 refcount repeated list entries instead of always adding to
1591 Severity : enhancement
1593 Description: Dynamic service threads
1594 Details : Within a small range, start extra service threads
1595 automatically when the request queue builds up.
1598 Frequency : mixed-endian client/server environments
1600 Description: mixed-endian crashes
1601 Details : The new msg_v2 system had some failures in mixed-endian
1604 Severity : enhancement
1606 Description: Easy OST removal
1607 Details : OSTs can be permanently deactivated with e.g. 'lctl
1608 conf_param lustre-OST0001.osc.active=0'
1610 Severity : enhancement
1612 Description: MGS proc entries
1613 Details : Added basic proc entries for the MGS showing what filesystems
1616 Severity : enhancement
1618 Description: provide MGS failover
1619 Details : Added config lock reacquisition after MGS server failover.
1621 Severity : enhancement
1623 Description: add Linux 2.4 support
1624 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1628 Description: a large application tries to do I/O to the same resource and dies
1629 in the middle of it.
1630 Details : Check the req->rq_arrival time after the call to
1631 ost_brw_lock_get(), but before we do anything about
1632 processing it & sending the BULK transfer request. This
1633 should help move old stale pending locks off the queue as
1634 quickly as obd_timeout.
1637 Frequency : when an incorrect nid is specified during startup
1639 Description: ptlrpc connect to non-existant node causes kernel crash
1640 Details : LNET can't be re-entered from an event callback, which
1641 happened when we expire a message after the export has been
1642 cleaned up. Instead, hand the zombie cleanup off to another
1645 Severity : enhancement
1647 Description: plain/inodebits lock performance improvement
1648 Details : Grouping plain/inodebits in granted list by their request modes
1649 and bits policy, thus improving the performance of search through
1653 Frequency : only if OST filesystem is corrupted
1655 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1656 Details : for a short time RPCs with bulk IO are in the replay list,
1657 but replay of bulk IOs is unimplemented. If the OST filesystem
1658 is corrupted due to disk cache incoherency and then replay is
1659 started it is possible to trip an assertion. Avoid putting
1660 committed RPCs into the replay list at all to avoid this issue.
1663 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1665 Description: System hang on startup
1666 Details : This bug allowed the liblustre (e.g. catamount) client to
1667 return to the app before handling all startup RPCs. This
1668 could leave the node unresponsive to lustre network traffic
1669 and manifested as a server ptllnd timeout.
1671 Severity : enhancement
1673 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1674 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1675 boolean to control whether to print peer debug info when a
1676 client's RPC times out.
1679 Frequency : only for kernels with patches from Lustre below 1.4.3
1681 Description: Remove old rdonly API
1682 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1685 Frequency : only for devices with external journals
1687 Description: Set external device read-only also
1688 Details : During a commanded failover stop, we set the disk device
1689 read-only while the server shuts down. We now also set any
1690 external journal device read-only at the same time.
1693 Frequency : when upgrading from 1.4 while trying to change parameters
1695 Description: The wrong (new) MDC name was used when setting parameters for
1696 upgraded MDT's. Also allows changing of OSC (and MDC)
1697 parameters if --writeconf is specified at tunefs upgrade time.
1700 Frequency : when setting specific ost indicies
1702 Description: QOS code breaks on skipped indicies
1703 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1704 created with --index need not be sequential.
1706 Severity : enhancement
1708 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1709 Details : The uninit_groups feature works in conjunction with the kernel
1710 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1711 up the pass1 processing of e2fsck. This is a read-only feature
1712 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1713 mount filesystems that have had this feature enabled.
1715 Severity : enhancement
1717 Description: Improve multi-block allocation algorithm to avoid fragmentation
1718 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1719 allocation locality and avoid filesystem fragmentation.
1721 ------------------------------------------------------------------------------
1723 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1725 * Support for kernels:
1726 2.4.21-47.0.1.EL (RHEL 3)
1727 2.6.5-7.283 (SLES 9)
1728 2.6.9-42.0.10.EL (RHEL 4)
1729 2.6.12.6 vanilla (kernel.org)
1730 2.6.16.27-0.9 (SLES 10)
1731 * Recommended e2fsprogs version: 1.39.cfs5
1733 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1737 Frequency : occasional, depends on client load and configuration
1738 Bugzilla : 12181, 12203
1739 Description: data loss for recently-modified files
1741 Details : In some cases it is possible that recently written or created
1742 files may not be written to disk in a timely manner (this should
1743 normally be within 30s unless client IO load is very high).
1744 The problem appears as zero-length files or files that are a
1745 multiple of 1MB in size after a client crash or client eviction
1746 that are missing data at the end of the file.
1748 This problem is more likely to be hit on clients where files are
1749 repeatedly created and unlinked in the same directory, clients
1750 have a large amount of RAM, have many CPUs, the filesystem has
1751 many OSTs, the clients are rebooted frequently, and/or the files
1752 are not accessed by other nodes after being written.
1754 The presence of the problem can be detected by looking at
1755 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1756 smaller than the second (nr_unused) then dirty files will not
1757 be flushed automatically to disk. "sync; sleep 10" should be
1758 run several times on the node before unmounting it to update
1759 Lustre (this is also safe to run on nodes without this problem).
1761 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1762 that can cause this same problem, so customers using that kernel
1763 also need to update the kernel in addition to Lustre. In order
1764 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1766 It is normal that files written just before a client crash (less
1767 than 30s) may not yet have been flushed to disk, even for local
1771 Frequency : frequent on thin XT3 nodes
1773 Description: UUID collision on thin XT3 Linux nodes
1774 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1775 randomly, since we relied on an insufficiently-seeded PRNG.
1780 Description: OSS hangs after "All ost request buffers busy"
1781 Details : A deadlock between quota and journal operations caused OSS
1782 hangs after printing "All ost request buffers busy."
1785 Frequency : always on liblustre builds
1787 Description: Cleanup compiler warnings on liblustre
1790 Frequency : always on liblustre builds on XT3
1792 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1796 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1797 > 32000 subdirectories
1798 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1799 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1800 subdirectory count crosses 32000. This will aid e2fsck to
1801 correctly handle more than 32000 subdirectories.
1804 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1806 Description: System hang on startup
1807 Details : This bug allowed the liblustre (e.g. catamount) client to
1808 return to the app before handling all startup RPCs. This
1809 could leave the node unresponsive to lustre network traffic
1810 and manifested as a server ptllnd timeout.
1812 Severity : enhancement
1814 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1815 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1816 boolean to control whether to print peer debug info when a
1817 client's RPC times out.
1822 Description: make O_SYNC working on 2.6 kernels
1823 Details : 2.6 kernels use different method for mark pages for write,
1824 so need add a code to lustre for O_SYNC work.
1829 Description: Failure to close file and release space on NFS
1830 Details : Put inode details into lock acquired in ll_intent_file_open.
1831 Use mdc_intent_lock in ll_intent_open to properly
1832 detect all kind of errors unhandled by mdc_enqueue.
1837 Description: proc file read during shutdown sometimes raced obd removal,
1839 Details : Add lock to prevent obd access after proc file removal.
1842 Frequency : Only for files larger than 4GB on 32-bit clients.
1844 Description: improperly doing page alignment of locks
1845 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1846 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1851 Description: RPCs being resent when they shouldn't be
1852 Details : Some RPCs that should not be resent are being resent. This
1853 can cause inconsistencies in the RPC state machine. Do not
1854 resend such requests.
1857 Frequency : rare, only with NFS export
1859 Description: Crash on NFS re-export node
1860 Details : under very unusual load conditions an assertion is hit in
1861 ll_intent_file_open()
1864 Frequency : only if OST filesystem is corrupted
1866 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1867 Details : for a short time RPCs with bulk IO are in the replay list,
1868 but replay of bulk IOs is unimplemented. If the OST filesystem
1869 is corrupted due to disk cache incoherency and then replay is
1870 started it is possible to trip an assertion. Avoid putting
1871 committed RPCs into the replay list at all to avoid this issue.
1876 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1877 Details : Large single O_DIRECT read and write calls can fail to allocate
1878 a sufficiently large buffer to process the request. In case of
1879 allocation failure the allocation is retried with a smaller
1880 buffer and broken into smaller requests.
1882 Severity : enhancement
1884 Description: Add -o localflock option to simulate old noflock behaviour.
1885 Details : This will achieve local-only flock/fcntl locks coherentness.
1890 Description: versioning check is incomplete
1891 Details : Checking the version difference of client vs. server, report
1892 error if the gap is too big.
1897 Description: add support PG_writeback bit
1898 Details : add support for PG_writeback bit for Lustre, for more carefull
1899 work with page cache in 2.6 kernel. This also fix some deadlocks
1900 and remove hack for work O_SYNC with 2.6 kernel.
1902 Severity : enhancement
1904 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1905 Details : The uninit_groups feature works in conjunction with the kernel
1906 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1907 up the pass1 processing of e2fsck. This is a read-only feature
1908 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1909 mount filesystems that have had this feature enabled.
1911 Severity : enhancement
1913 Description: Improve multi-block allocation algorithm to avoid fragmentation
1914 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1915 allocation locality and avoid filesystem fragmentation.
1917 ------------------------------------------------------------------------------
1919 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1921 * Support for kernels:
1922 2.6.9-42.0.3.EL (RHEL 4)
1923 2.6.5-7.276 (SLES 9)
1924 2.4.21-47.0.1.EL (RHEL 3)
1925 2.6.12.6 vanilla (kernel.org)
1926 2.6.16.21-0.8 (SLES10)
1927 * Recommended e2fsprogs version: 1.39.cfs2-0
1929 * The backwards-compatible /proc/sys/portals symlink has been removed
1930 in this release. Before upgrading, please ensure that you change
1931 any configuration scripts or /etc/sysctl.conf files that access
1932 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1933 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1934 in advance of the upgrade on any system running Lustre 1.4.6 or
1935 newer, since /proc/sys/lnet was added in that version.
1936 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1940 Frequency : only when quota is used
1942 Description: avoid scanning export list for quota master
1943 Details : Change the algorithms to avoid scanning export list in order
1944 to improve the efficiency.
1947 Frequency : MDS failover only, very rarely
1949 Description: "went back in time" messages on mds failover
1950 Details : The greatest transno may be lost when the current operation
1951 finishes with an error (transno==0) and the client's last_rcvd
1952 record is over-written. Save the greatest transno in the
1953 mds_last_transno for this case.
1956 Frequency : always for specific kernels and striping counts
1958 Description: client may get "Matching packet too big" without ACL support
1959 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1960 when trying to access files in certain configurations. The
1961 clients should in fact be denied when mounting because they do
1962 not understand ACLs.
1965 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1967 Description: many clients connecting with IO in progress causes connect timeouts
1968 Details : Avoid synchronous journal commits to avoid delays caused by many
1969 clients connecting/disconnecting when bulk IO is in progress.
1970 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1971 OST_IO_PORTAL to avoid delays behind potentially many pending
1975 Frequency : occasionally with multiple writers to a single file
1977 Description: shared writes to file may result in wrong size reported by stat()
1978 Details : Allow growing of kms when extent lock is cancelled
1981 Frequency : always with random mmap IO to multi-striped file
1983 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1984 Details : Only if the hole is at the end of OST object so that kms is too
1985 small. Fix is to increase kms accordingly in ll_nopage.
1988 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1990 Description: writes to a missing object would leak memory on the OST
1991 Details : If there is an inconsistency between the MDS and OST filesystems,
1992 such that the MDS references an object that doesn't exist, writes
1993 to that object will leak memory due to incorrect cleanup in the
1994 error handling path, eventually running out of memory on the OST.
1999 Description: Creating too long symlink causes lustre errors
2000 Details : Check symlink and name lengths before sending requests to MDS.
2003 Frequency : only if flock is enabled (not on by default)
2005 Description: posix locks not released on fd closure on 2.6.9+
2006 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2007 kernels, this caused such locks not to be released on fd close and
2008 then assertions on fs unmount about still used locks.
2011 Frequency : MDS failover only, very rarely
2013 Description: clients may get ASSERTION(granted_lock != NULL)
2014 Details : When request was taking a long time, and a client was resending
2015 a getattr by name lock request. The were multiple lock requests
2016 with the same client lock handle and
2017 mds_getattr_name->fixup_handle_for_resent_request found one of the
2018 lock handles but later failed with ASSERTION(granted_lock != NULL).
2023 Description: handle->h_buffer_credits > 0, assertion failure
2024 Details : h_buffer_credits is zero after truncate, causing assertion
2025 failure. This patch extends the transaction or creates a new
2029 Frequency : NFS re-export or patchless client
2030 Bugzilla : 11179, 10796
2031 Description: Crash on NFS re-export node (__d_move)
2032 Details : We do not want to hash the dentry if we don't have a lock.
2033 But if this dentry is later used in d_move, we'd hit uninitialised
2034 list head d_hash, so we just do this to init d_hash field but
2035 leave dentry unhashed.
2038 Frequency : NFS re-export or patchless client
2040 Description: NFS exports has problem with symbolic link
2041 Details : lustre client didn't properly install dentry when re-exported
2042 to NFS or running patchless client.
2045 Frequency : NFS re-export or patchless client
2047 Description: Various nfs/patchless fixes.
2048 Details : fixes reuse disconected alias for lookup process - this fixes
2049 warning "find_exported_dentry: npd != pd",
2050 fix permission error with open files at nfs.
2051 fix apply umask when do revalidate.
2054 Frequency : occasional
2056 Description: Crash on NFS re-export node
2057 Details : calling clear_page() on the wrong pointer triggered oops in
2058 generic_mapping_read().
2061 Frequency : rarely, using O_DIRECT IO
2063 Description: unaligned directio crashes client with LASSERT
2064 Details : check for unaligned buffers before trying any requests.
2067 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2069 Description: stale data returned from RAID cache
2070 Details : If only a small amount of IO is done to the RAID device before
2071 reading it again it is possible to get stale data from the RAID
2072 cache instead of reading it from disk.
2075 Frequency : always for sles10 kernel
2077 Description: sles10 support
2078 Details : ll_follow_link: compile fixes and using of nd_set_link
2079 under newer kernels.
2082 Frequency : depends on arch, kernel and compiler version, always on sles10
2085 Description: recursive or deep enough symlinks cause stack overflow
2086 Details : getting rid of large stack-allocated variable in
2090 Frequency : depends on hardware
2092 Description: lustre write performance loss in the SLES10 kernel
2093 Details : the performance loss is caused by using of write barriers in the
2094 ext3 code. The SLES10 kernel turns barrier support on by
2095 default. The fix is to undo that change for ldiskfs.
2097 ------------------------------------------------------------------------------
2099 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2101 * Support for kernels:
2102 2.6.9-42.0.3EL (RHEL 4)
2103 2.6.5-7.276 (SLES 9)
2104 2.4.21-47.0.1.EL (RHEL 3)
2105 2.6.12.6 vanilla (kernel.org)
2109 Frequency : quota enabled and large files being deleted
2111 Description: releasing more than 4GB of quota at once hangs OST
2112 Details : If a user deletes more than 4GB of files on a single OST it
2113 will cause the OST to spin in an infinite loop. Release
2114 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2119 Description: statfs data retrieved from /proc may be stale or zero
2120 Details : When reading per-device statfs data from /proc, in the
2121 {kbytes,files}_{total,free,avail} files, it may appear
2122 as zero or be out of date.
2125 Frequency : systems with MD RAID1 external journal devices
2127 Description: lconf's call to blkid is confused by RAID1 journal devices
2128 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2129 of returning all block devices that match the journal UUID.
2132 Frequency : always, for aggregate stripe size over 4GB
2134 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2135 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2136 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2139 Frequency : always if "lfs find" used on a local file/directory
2141 Description: "lfs find" segfaults if used on a local file/directory
2142 Details : The case where a directory component was not specified wasn't
2143 handled correctly. Handle this properly.
2146 Frequency : always on ppc64
2148 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2149 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2150 perform endianness conversion.
2153 Frequency : rarely (truncate to non-zero file size after write under load)
2154 Bugzilla : 10730, 10687
2155 Description: Files padded with zeros to next 4K multiple
2156 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2157 it is possible that files that are truncated to a non-zero size
2158 immediately after being written are filled with zero bytes beyond
2159 the truncated size. No file data is lost.
2161 Severity : enhancement
2163 Description: Allow recovery/failover for liblustre clients.
2164 Details : liblustre clients were unaware of failover configurations until
2167 Severity : enhancement
2169 Description: user file locks should fail when not mounting with flock option
2170 Details : Set up an error-returning stub in ll_file_operations.lock field
2171 to prevent incorrect behaviour when client is mounted without
2172 flock option. Also, set up properly f_op->flock field for
2176 Frequency : always on ia64
2178 Description: "lfs df" loops on printing out MDS statfs information
2179 Details : The obd_ioctl_data was not initialized and in some systems
2180 this caused a failure during the ioctl that did not return
2181 an error. Initialize the struct and return an error on failure.
2184 Frequency : SLES 9 only
2186 Description: Error of copying files with lustre special EAs as root
2187 Details : Client side always return success for setxattr call for lustre
2188 special xattr (currently only "trusted.lov").
2191 Frequency : rarely on clusters with both ia64+i386 clients
2193 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2194 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2195 clients are doing concurrent writes to the same file it is
2196 possible that the ia64 clients may LASSERT because the OST
2197 extent locks are not PAGE_SIZE aligned. Ensure that grown
2198 locks are always aligned on the request boundary.
2201 Frequency : specific use, occasional
2203 Description: Overwriting in use executable truncates on-disk binary image
2204 Details : If one node attempts to overwrite an executable in use by
2205 another node, we now correctly return ETXTBSY instead of
2206 truncating the file.
2208 Severity : enhancement
2210 Description: Async OSC create to avoid the blocking unnecessarily.
2211 Details : If a OST has no remain object, system will block on the creating
2212 when need to create a new object on this OST. Now, ways use
2213 pre-created objects when available, instead of blocking on an
2214 empty osc while others are not empty. If we must block, we block
2215 for the shortest possible period of time.
2220 Description: chmod on Lustre root is propagated to other clients
2221 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2222 invalid by the follow_mount time.
2225 Frequency : liblustre clients only
2227 Description: Race in 'instant cancel' lock handling could lead to such locks
2228 never to be granted in case of SMP MDS
2229 Details : Do not destroy not yet granted but cbpending locks in
2233 Frequency : replay/resend of open
2235 Description: non null lock assetion failure in mds_intent_policy
2236 Details : Trying to replay/resend lockless open requests resulted in
2237 mds_open() returning 0 with no lock. Now it sets a flag if
2238 a lock is going to be returned.
2240 Severity : enhancement
2242 Description: Checksum enhancements
2243 Details : New checksum enhancements allow for resending RPCs that failed
2246 Severity : enhancement
2248 Description: Tunables on number of dirty pages in cacche
2249 Details : Allow to set limit on number of dirty pages cached.
2254 Description: client crash on unmount - lock still has references
2255 Details : In some error handling cases it was possible to leak a lock
2256 reference on a client while accessing a file. This was not
2257 harmful to the client during operation, but would cause the
2258 client to crash when the filesystem is unmounted.
2261 Frequency : specific case, rare
2263 Description: ETXTBSY on mds though file not in use by client
2264 Details : ETXTBSY is no longer incorrectly returned when attempting to
2265 chmod or chown a directory that the user previously tried to
2266 execute or a currently-executing binary.
2269 Frequency : extremely rare except on liblustre-based clients
2271 Description: Lustre space not freed when files are deleted
2272 Details : Clean up open-unlinked files after client eviction. Previously
2273 the unlink was skipped and the files remained as orphans.
2278 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2279 Details : In some cases it was possible to send a blocking callback to
2280 a client doing a glimpse, even though that client didn't get
2281 a lock granted. When the glimpse lock is cancelled on the OST
2282 the freed lock is left on the waiting list and corrupted the list.
2285 Frequency : all core dumps
2287 Description: Broke core dumps to lustre
2288 Details : Negative dentry may be unhashed if parent does not have UPDATE
2289 lock, but some callers, e.g. do_coredump, expect dentry to be
2290 hashed after successful create, hash it in ll_create_it.
2292 ------------------------------------------------------------------------------
2294 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2296 * Support for kernels:
2297 2.6.9-42.0.2.EL (RHEL 4)
2298 2.6.5-7.276 (SLES 9)
2299 2.4.21-40.EL (RHEL 3)
2300 2.6.12.6 vanilla (kernel.org)
2304 Frequency : always on RHEL 3
2306 Description: Number of open files grows over time
2307 Details : The number of open files grows over time, whether or not
2308 Lustre is started. This was due to a filp leak introduced
2309 by one of our kernel patches.
2311 ------------------------------------------------------------------------------
2313 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2315 * Support for kernels:
2316 2.6.9-42.EL (RHEL 4)
2317 2.6.5-7.276 (SLES 9)
2318 2.4.21-40.EL (RHEL 3)
2319 2.6.12.6 vanilla (kernel.org)
2324 Bugzilla : 5719, 9635, 9792, 9684
2325 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2326 Details : If a server is under heavy load and cannot reply to new
2327 connection requests before the client resends the (re)connect,
2328 the connection handling code can behave badly if two service
2329 threads are concurrently handing separate (re)connections from
2330 the same client. Add better locking to the connection handling
2331 code, and ensure that only a single connection will be processed
2332 for a given client UUID, even if the lock is dropped.
2334 Severity : enhancement
2336 Description: add TCP zero-copy support to kernel
2337 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2338 sends if the hardware supports scatter-gather and checksumming.
2339 This allows socklnd to do client-write and server-read more
2340 efficiently and reduce CPU utilization from skbuf copying.
2343 Frequency : only if NFS exporting from client
2345 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2346 Details : If setting attributes on a file created under NFS that had
2347 never been opened it would be possible to oops the client
2348 if the file had no objects.
2351 Frequency : always for liblustre
2353 Description: liblustre client does MDS+OSTs setattr RPC for each write
2354 Details : When doing a write from a liblustre client, the client
2355 incorrectly issued an RPC to the MDS and each OST the file was
2356 striped over in order to update the timestamps. When writing
2357 with small chunks and many clients this could overwhelm the MDS
2358 with RPCs. In all cases it would slow down the write because
2359 these RPCs are unnecessary.
2361 Severity : enhancement
2363 Description: allow number of MDS service threads to be changed at module load
2364 Details : It is now possible to change the number of MDS service threads
2365 running. Adding "options mds mds_num_threads={N}" to the MDS's
2366 /etc/modprobe.conf will set the number of threads for the next
2367 time Lustre is restarted (assuming the "mds" module is also
2368 reloaded at that time). The default number of threads will
2369 stay the same, 32 for most systems.
2374 Description: OST crash if filesystem is unformatted or corrupt
2375 Details : If an OST is started on a device that has never been formatted
2376 or if the filesystem is corrupt and cannot even mount then the
2377 error handling cleanup routines would dereference a NULL pointer.
2382 Description: NULL pointer deref in llap_from_page.
2383 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2384 incorrect) mapping if the page was truncated/reclaimed while it was
2385 searched for. Check for this condition and skip such pages when
2386 doing readahead. Introduce extra check to llap_from_page() to
2387 verify page->mapping->host is non-NULL (so page is not anonymous).
2390 Frequency : Sometimes when using sys_sendfile
2392 Description: "page not covered by a lock" warnings from ll_readpage
2393 Details : sendfile called ll_readpage without right page locks present.
2394 Now we introduced ll_file_sendfile that does necessary locking
2395 around call to generic_file_sendfile() much like we do in
2399 Frequency : with certain MDS communication failures at client mount time
2401 Description: NULL pointer deref after failed client mount
2402 Details : a client connection request may delayed by the network layer
2403 and not be sent until after the PTLRPC layer has timed out the
2404 request. If the client fails the mount immediately it will try
2405 to clean up before the network times out the request. Add a
2406 reference from the request import to the obd device and delay
2407 the cleanup until the network drops the request.
2410 Frequency : occasionally during client (re)connect
2412 Description: assertion failure during client (re)connect
2413 Details : processing a client connection request may be delayed by the
2414 client or server longer than the client connect timeout. This
2415 causes the client to resend the connection request. If the
2416 original connection request is replied in this interval, the
2417 client may trip an assertion failure in ptlrpc_connect_interpret()
2418 which thought it would be the only running connect process.
2421 Frequency : only with obd_echo servers and clients that are rebooted
2423 Description: kernel BUG accessing uninitialized data structure
2424 Details : When running an obd_echo server it did not start the ping_evictor
2425 thread, and when a client was evicted an uninitialized data
2426 structure was accessed. Start the ping_evictor in the RPC
2427 service startup instead of the OBD startup.
2429 Severity : enhancement
2430 Bugzilla : 10193 (patchless)
2431 Description: Remove dependency on various unexported kernel interfaces.
2432 Details : No longer need reparent_to_init, exit_mm, exit_files,
2433 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2436 Frequency : rare (only users of deprecated and unsupported LDAP config)
2438 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2439 Details : LDAP apparently contains 'lustreName' attributes instead of
2440 'name'. A simple remapping of the name is sufficient.
2443 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2445 Description: waiting_locks_callback trips kernel BUG if client is evicted
2446 Details : Running with the dump_on_timeout debug flag turned on makes
2447 it possible that the waiting_locks_callback() can try to dump
2448 the Lustre kernel debug logs from an interrupt handler. Defer
2449 this log dumping to the expired_lock_main() thread.
2451 Severity : enhancement
2453 Description: Support NFS exporting on 2.6 kernels.
2454 Details : Implement non-rawops metadata methods for NFS server to use without
2455 changing NFS server code.
2458 Frequency : very rare (synthetic metadata workload only)
2460 Description: two racing renames might cause an MDS thread to deadlock
2461 Details : Running the "racer" program may cause one MDS thread to rename
2462 a file from being the source of a rename to being the target of
2463 a rename at exactly the same time that another thread is doing
2464 so, and the second thread has already enqueued these locks after
2465 doing a lookup of the target and is trying to relock them in
2466 order. Ensure that we don't try to re-lock the same resource.
2469 Frequency : only very large systems with liblustre clients
2471 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2472 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2473 all OSTs in parallel. This allows the eviction of stale liblustre
2474 clients to proceed much faster than if they were done in series,
2475 and also offers similar improvements for other set_info RPCs.
2480 Description: excessive CPU usage during initial read phase on client
2481 Details : During the initial read phase on a client, it would agressively
2482 retry readahead on the file, consuming too much CPU and impacting
2483 performance (since 1.4.5.8). Improve the readahead algorithm
2484 to avoid this, and also improve some other common cases (read
2485 of small files in particular, where "small" is files smaller than
2486 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2491 Description: MDS crash when receiving packet with unknown intent.
2492 Details : Do not LBUG in unknown intent case, just return -EFAULT
2494 Severity : enhancement
2495 Bugzilla : 9293, 9385
2496 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2497 Details : Do not serialize getattr (non-intent version) and statfs.
2500 Frequency : occasional, when OST network is overloaded/intermittent
2502 Description: client evicted by OST after bulk IO timeout
2503 Details : If a client sends a bulk IO request (read or write) the OST
2504 may evict the client if it is unresposive to its data GET/PUT
2505 request. This is incorrect if the network is overloaded (takes
2506 too long to transfer the RPC data) or dropped the OST GET/PUT
2507 request. There is no need to evict the client at all, since
2508 the pinger and/or lock callbacks will handle this, and the
2509 client can restart the bulk request.
2512 Frequency : Always when mmapping file with no objects
2514 Description: client crashes when mmapping file with no objects
2515 Details : Check that we actually have objects in a file before doing any
2516 operations on objects in ll_vm_open, ll_vm_close and
2522 Description: Request leak when working with deleted CWD
2523 Details : Introduce advanced request refcount tracking for requests
2524 referenced from lustre intent.
2526 Severity : Enhancement
2528 Description: Cache open file handles on client.
2529 Details : MDS now will return special lock along with openhandle, if
2530 requested and client is allowed to hold openhandle, even if unused,
2531 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2532 closing files for every read/write openration.
2534 Severity : Enhancement
2536 Description: Cache open negative dentries on client when possible.
2537 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2538 negative dentries on lock revocation.
2543 Description: Remounting a client read-only wasn't possible with a zconf mount
2544 Details : It wasn't possible to remount a client read-only with llmount.
2546 Severity : enhancement
2547 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2548 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2549 MPICH is included to add Lustre-specific ADIO interfaces.
2550 This is based closely on the UFS ADIO layer and only differs
2551 in file creation, in order to allow the OST striping to be set.
2552 This is user-contributed code and not supported by CFS.
2557 Description: extended inode attributes (immutable, append-only) work improperly
2558 when 2.4 and 2.6 kernels are used on client/server or vice versa
2559 Details : Introduce kernel-independent values for these flags.
2561 Severity : enhancement
2564 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2565 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2566 granularity. Now, Fractional MB granularity is be supported,
2567 this is very useful for low memory system.
2569 Severity : enhancement
2571 Description: Getattr by fid
2572 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2573 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2574 have locks and avoids getattr rpc requests for it on every stat.
2577 Frequency : Always, for filesystems larger than 2TB
2579 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2580 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2581 filesystems larger than 2TB to be mounted with mballoc enabled.
2584 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2586 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2587 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2588 filesystems larger than 2TB is mounted then the kernel will
2589 silently corrupt the start of the filesystem. CONFIG_LBD is
2590 enabled for all CFS-supported kernels, but the possibility of
2591 this happening with a modified kernel config exists.
2593 Severity : enhancement
2595 Description: add client O_DIRECT support for 2.6 kernels
2596 Details : It is now possible to do O_DIRECT reads and writes to files
2597 in the Lustre client mountpoint on 2.6 kernel clients.
2599 Severity : enhancement
2601 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2602 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2603 parallel, not waiting for response from every OST before sending
2604 a rpc to the next OST.
2609 Description: setattr vs write race when updating file timestamps
2610 Details : Client processes that update a file timestamp into the past
2611 right after writing to the file (e.g. tar) it is possible that
2612 the updated file modification time can be reset to the current
2613 time due to a race between processing the setattr and write RPC.
2615 Severity : enhancement
2617 Description: Bring 'lfs find' closer in line with regular Linux find.
2618 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2619 -print0 options and obtains all the needed info through the lustre
2622 Severity : enhancement
2624 Description: support up to 1024 configured devices on one node
2625 Details : change obd_dev array from statically allocated to dynamically
2626 allocated structs as they are first used to reduce memory usage
2631 Description: Flush dirty partially truncated pages during truncate
2632 Details : Immediatelly flush partially truncated pages in filter_setattr,
2633 this way we completely avoid having any pages in page cache on OST
2634 and can retire ugly workarounds during writes to flush such pages.
2639 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2640 Details : protect inode from truncation within vfs_unlink() context
2641 just take a reference before calling vfs_unlink() and release it
2642 when parent's i_sem is free.
2647 Description: last_id value checked outside lock on OST caused LASSERT failure
2648 Details : If there were multiple MDS->OST object precreate requests in
2649 flight, it was possible that the OST's last object id was checked
2650 outside a lock and incorrectly tripped an assertion. Move checks
2651 inside locks, and discard old precreate requests.
2654 Frequency : always, if extents are used on OSTs
2656 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2657 Details : OSTs using the extents format would not zero the high 16 bits of
2658 the index physical block number. This is not a problem for any
2659 OST filesystems smaller than 16TB, and no kernels support ext3
2660 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2661 new/modified files) and can be fixed for existing filesystems
2662 with e2fsprogs-1.39-cfs1.
2667 Description: import connection selection may be incorrect if timer wraps
2668 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2669 import connections to be ignored if the 32-bit jiffies counter
2670 wraps. Use a 64-bit jiffies counter.
2673 Frequency : very large clusters immediately after boot
2675 Description: LNET request buffers exhausted under heavy short-term load
2676 Details : If a large number of client requests are generated on a service
2677 that has previously never seen so many requests it is possible
2678 that the request buffer growth cannot keep up with the spike in
2679 demand. Instead of dropping incoming requests, they are held in
2680 the LND until the RPC service can accept more requests.
2683 Frequency : Sometimes during replay
2685 Description: Assertion failure in ll_local_open after replay.
2686 Details : If replay happened on an open request reply before we were able
2687 to set replay handler, reply will become not swabbed tripping the
2688 assertion in ll_local_open. Now we set the handler right after
2689 recognising of open request
2692 Frequency : very rare
2694 Description: kernel reports "badness in vsnprintf"
2695 Details : Reading from the "recovery_status" /proc file in small chunks
2696 may cause a negative length in lprocfs_obd_rd_recovery_status()
2697 call to vsnprintf() (which is otherwise harmless). Exit early
2698 if there is no more space in the output buffer.
2700 Severity : enhancement
2702 Description: clear OBD RPC statistics by writing to them
2703 Details : It is now possible to clear the OBD RPC statistics by writing
2704 to the "stats" file.
2709 Description: Client mtime is not the same on different clients after utimes
2710 Details : In some cases, the client was using the utimes() syscall on
2711 a file cached on another node. The clients now validate the
2712 ctime from the MDS + OSTs to determine which one is right.
2717 Description: Inability to activate failout mode
2718 Details : lconf script incorrectly assumed that in pythong string's numeric
2719 value is used in comparisons.
2722 Frequency : always with multiple stripes per file
2724 Description: Inefficient object allocation for mutli-stripe files
2725 Details : When selecting which OSTs to stripe files over, for files with
2726 a stripe count that divides evenly into the number of OSTs,
2727 the MDS is always picking the same starting OST for each file.
2728 Return the OST selection heuristic to the original design.
2733 Description: mount failures may take full timeout to return an error
2734 Details : Under some heavy load conditions it is possible that a
2735 failed mount can wait for the full obd_timeout interval,
2736 possibly several minutes, before reporting an error.
2737 Instead return an error as soon as the status is known.
2739 Frequency : quota enabled and large files being deleted
2741 Description: releasing more than 4GB of quota at once hangs OST
2742 Details : If a user deletes more than 4GB of files on a single OST it
2743 will cause the OST to spin in an infinite loop. Release
2744 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2749 Description: statfs data retrieved from /proc may be stale or zero
2750 Details : When reading per-device statfs data from /proc, in the
2751 {kbytes,files}_{total,free,avail} files, it may appear
2752 as zero or be out of date.
2755 Frequency : systems with MD RAID1 external journal devices
2757 Description: lconf's call to blkid is confused by RAID1 journal devices
2758 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2759 of returning all block devices that match the journal UUID.
2762 Frequency : always, for aggregate stripe size over 4GB
2764 Description: assertion fails when trying to use 4GB stripe size
2765 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2766 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2769 Frequency : always on ppc64
2771 Description: the first write on an ext3 filesystem with mballoc got stuck
2772 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2773 perform endianness conversion.
2775 ------------------------------------------------------------------------------
2777 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2779 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2780 INTEROPERATE with older versions automatically. Please read the
2781 user documentation before upgrading any part of a live system.
2782 * WARNING: Lustre networking configuration changes are required with
2783 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2786 * Support for kernels:
2787 2.6.9-22.0.2.EL (RHEL 4)
2788 2.6.5-7.244 (SLES 9)
2789 2.6.12.6 vanilla (kernel.org)
2792 Severity : enhancement
2793 Bugzilla : 7981/8208
2794 Description: Introduced Lustre Networking (LNET)
2795 Details : LNET is new networking infrastructure for Lustre, it includes
2796 a reorganized network configuration mode (see the user
2797 documentation for full details) as well as support for routing
2798 between different network fabrics. Lustre Networking Devices
2799 (LNDs) for the supported network fabrics have also been
2800 created for this new infrastructure.
2802 Severity : enhancement
2803 Description: Introduced Access control lists
2804 Details : clients can set ACLs on files and directories in order to have
2805 more fine-grained permissions than the standard Unix UGO+RWX.
2806 The MDS must be started with the "-o acl" mount option.
2808 Severity : enhancement
2809 Description: Introduced filesystem quotas
2810 Details : Administrators may now establish per-user quotas on the
2813 Severity : enhancement
2815 Description: Configuration change for the XT3
2816 The PTLLND is now used to run Lustre over Portals on the XT3
2817 The configure option(s) --with-cray-portals are no longer used.
2818 Rather --with-portals=<path-to-portals-includes> is used to
2819 enable building on the XT3. In addition to enable XT3 specific
2820 features the option --enable-cray-xt3 must be used.
2825 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2826 Details : A race between allocating a new llcd and re-getting the llcd_lock
2827 allowed another thread to grab newly-allocated llcd.
2829 Severity : enhancement
2831 Description: 2.6 OST async journal commit and locking fix to improve performance
2832 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2833 2.6 kernels are now async as they already were in 2.4 kernels so
2834 that they can commit concurrently with the network bulk transfer.
2835 For block-allocated files the filter allocation semaphore is held
2836 to avoid filesystem fragmentation during allocation. BKL lock
2837 removed for 2.6 xattr operations where it is no longer needed.
2842 Description: lconf incorrectly determined whether two IP networks could talk
2843 Details : In some more complicated routing and multiple-network
2844 configurations, lconf will avoid trying to make a network
2845 connection to a disjoint part of the IP space. It was doing the
2846 math incorrectly for one set of cases.
2851 Description: Fix for potential infinite loop processing records in an llog.
2852 Details : If an llog record is corrupted/zeroed, it is possible to loop
2853 forever in llog_process(). Validate the llog record length
2854 and skip the remainder of the block on error.
2857 Frequency : occasional (liblustre only)
2859 Description: liblustre could not open files whose last component is a symlink
2860 Details : sysio_path_walk() would incorrectly pass the open intent to
2861 intermediate path components.
2864 Frequency : rare (liblustre only with non-standard tuning)
2865 Bugzilla : 7201 (7350)
2866 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2867 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2868 after it has been released, only do this for referenced locks
2870 Severity : enhancement
2872 Description: specifying an (invalid) directory default stripe_size of -1
2873 would reset the directory default striping
2874 Details : stripe_size -1 was used internally to signal directory stripe
2875 removal, now use "all default" to signal dir stripe removal
2876 as a directory striping of "all default" is not useful
2879 Frequency : common for large clusters running liblustre clients
2881 Description: doing an ls when liblustre clients are running is slow
2882 Details : sending a glimpse AST to a liblustre client waits for every AST
2883 to time out, as liblustre clients will not respond. Since they
2884 cannot cache data we refresh the OST lock LVB from disk instead.
2886 Severity : enhancement
2888 Description: doing an ls at the same time as file IO can be slow
2889 Details : enqueue and other "small" requests can be blocked behind many
2890 large IO requests. Create a new OST IO portal for non-IO
2891 requests so they can be processed faster.
2894 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2896 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2897 created file with mode 000
2898 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2899 allow an ftruncate() as a non-root user to a file with mode 000.
2900 Lustre now respects this flag to disable mode checks when
2901 truncating a file owned by the user
2904 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2907 Description: Revoking locks from clients that went dead or catatonic might take
2909 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2910 cancellation of such locks instant on servers without waiting for
2911 any reply from clients. Clients drops these locks when cancel
2912 notification from server is received without replying.
2915 Frequency : liblustre-only, when liblustre client dies or becomes busy
2917 Description: Doing ls on Linux clients can take a long time with active
2919 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2920 granting such locks to it in the first place if possible. Locks
2921 are taken by proxy on the OST during the read or write and
2922 dropped immediately afterward. Add connect flags handling, do
2923 not grant locks to liblustre clients for glimpse ASTs.
2925 Severity : enhancement
2927 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2928 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2929 sequential reads and behaves badly with applications doing random
2930 reads. Improve it by reading ahead at least read region, and
2931 avoiding excessive large RPC for small reads.
2933 Severity : enhancement
2935 Description: Creating more than 1000 files for a single job may cause a load
2936 imbalance on the OSTs if there are also a large number of OSTs.
2937 Details : qos_prep_create() uses an OST index reseed value that is an
2938 even multiple of the number of available OSTs so that if the
2939 reseed happens in the middle of the object allocation it will
2940 still utilize the OSTs as uniformly as possible.
2945 Description: OST or MDS may oops in ping_evictor_main()
2946 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2947 but doesn't restart at beginning of obd_exports_timed list
2950 Severity : enhancement
2952 Description: improve by-nid export eviction on the MDS and OST
2953 Details : allow multiple exports with the same NID to be evicted at one
2954 time without re-searching the exports list.
2957 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2959 Description: MDS may oops in groups_free()
2960 Details : in rare race conditions a newly allocated group_info struct is
2961 freed again, and this can be NULL. The 2.4 compatibility code
2962 for groups_free() checked for a NULL pointer, but 2.6 did not.
2965 Frequency : common for liblustre clients doing little filesystem IO
2966 Bugzilla : 9352, 7313
2967 Description: server may evict liblustre clients accessing contended locks
2968 Details : if a client is granted a lock or receives a completion AST
2969 with a blocking AST already set it would not reply to the AST
2970 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2973 Frequency : lfs setstripe, only systems with more than 160 OSTs
2975 Description: unable to set striping with a starting offset beyond OST 160
2976 Details : llapi_create_file() incorrectly limited the starting stripe
2977 index to the maximum single-file stripe count.
2980 Frequency : LDAP users only
2982 Description: lconf did not handle in-kernel recovery with LDAP properly
2983 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2985 Severity : enhancement
2987 Description: bind OST threads to NUMA nodes to improve performance
2988 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2989 node and do their allocations there to localize memory access
2991 Severity : enhancement
2993 Description: llmount can determine client NID directly from Myrinet (GM)
2994 Details : the client NID code from gmnalnid was moved directly into
2995 llmount, removing the need to use this or specifying the
2996 client NID explicitly when mounting GM clients with zeroconf
2999 Frequency : if client is started with down MDS
3001 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3002 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3003 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3004 ptlrpc_import_delayed_req() because we didn't check if the
3005 request was interrupted, and we also didn't break out of the
3006 event loop if there was no timeout
3011 Description: data loss during non-page-aligned writes to a single file from
3012 both multiple nodes and multiple threads on one node at same time
3013 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3014 inconsistency led to false short-reads, that were cached and later
3015 used by ->prepare_write() to fill in partially written page,
3016 leading to data loss.
3019 Frequency : always, if lconf --abort_recovery used
3021 Description: lconf --abort_recovery fails with 'Operation not supported'
3022 Details : lconf was attempting to abort recovery on the MDT device and not
3025 Severity : enhancement
3027 Description: remove cleanup logs
3028 Details : replace lconf-generated cleanup logs with lustre internal
3029 cleanup routines. Eliminates the need for client-cleanup and
3032 Severity : enhancement
3034 Description: add support for EAs (user and system) on lustre filesystems
3035 Details : it is now possible to store extended attributes in the Lustre
3036 client filesystem, and with the user_xattr mount option it
3037 is possible to allow users to store EAs on their files also
3039 Severity : enhancement
3041 Description: Add possibility (config option) to show minimal available OST free
3043 Details : When compiled with --enable-mindf configure option, statfs(2)
3044 (and so, df) will return least minimal free space available from
3045 all OSTs as amount of free space on FS, instead of summary of
3046 free spaces of all OSTs.
3048 Severity : enhancement
3050 Description: do not expand extent locks acquired on OST-side
3051 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3052 by server: they are not cached anyway.
3055 Frequency : when mmap is used/binaries executed from Lustre
3057 Description: Unmmap pages before throwing them away from read cache.
3058 Details : llap_shrink cache now attempts to unmap pages before discarding
3059 them (if unmapping failed - do not discard). SLES9 kernel has
3060 extra checks that trigger if this unmapping is not done first.
3065 Description: lconf didn't resolve symlinks before checking to see whether a
3066 given mountpoint was already in use
3069 Frequency : when migrating failover services
3070 Bugzilla : 6395, 9514
3071 Description: When migrating a subset of services from a node (e.g. failback
3072 from a failover service node) the remaining services would
3073 time out and evict clients.
3074 Details : lconf --force (implied by --failover) sets the global obd_timeout
3075 to 5 seconds in order to quickly disconnect, but this caused
3076 other RPCs to time out too quickly. Do not change the global
3077 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3079 Severity : enhancement
3080 Frequency : if MDS is started with down OST
3081 Bugzilla : 9439,5706
3082 Description: Allow startup/shutdown of an MDS without depending on the
3083 availability of the OSTs.
3084 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3085 Add appropriate locking and lov-osc refcounts for safe
3086 cleaning. Add osc abort_inflight calls in case the
3087 synchronize never started.
3090 Frequency : occasional (Cray XT3 only)
3092 Description: root not authorized to access files in CRAY_PORTALS environment
3093 Details : The client process capabilities were not honoured on the MDS in
3094 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3095 been accessed by an authorized user then root was able to access
3096 the file on the local client also. The root user capabilities
3097 are now allowed on the MDS, as this environment has secure UID.
3100 Frequency : occasional
3102 Description: ldiskfs "too long searching" message happens too often
3103 Details : A debugging message (otherwise harmless) prints too often on
3104 the OST console. This has been reduced to only happen when
3105 there are fragmentation problems on the filesystem.
3110 Description: Division by zero in statfs when all OSCs are inactive
3111 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3112 lov_statfs divides by value returned by lov_get_stripecnt().
3116 Bugzilla : 9489, 3273
3117 Description: First write from each client to each OST was only 4kB in size,
3118 to initialize client writeback cache, which caused sub-optimal
3119 RPCs and poor layout on disk for the first writen file.
3120 Details : Clients now request an initial cache grant at (re)connect time
3121 and so that they can start streaming writes to the cache right
3122 away and always do full-sized RPCs if there is enough data.
3123 If the OST is rebooted the client also re-establishes its grant
3124 so that client cached writes will be honoured under the grant.
3129 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3130 Details : Now I/O RPCs go to different portal number and (presumably) fast
3131 lock requests (and glimses) and other RPCs get their own service
3132 threads pool that should be able to service those RPCs
3135 Severity : enhancement
3137 Description: Ability to exchange lustre version between client and servers and
3138 issue warnings at client side if client is too old. Also for
3139 liblustre clients there is ability to refuse connection of too old
3141 Details : New 'version' field is added to connect data structure that is
3142 filled with version info. That info is later checked by server and
3146 Frequency : rare, liblustre only.
3147 Bugzilla : 9296, 9581
3148 Description: Two simultaneous writes from liblustre at offset within same page
3149 might proceed at the same time overwriting eachother with stale
3151 Details : I/O lock withing llu_file_prwv was released too early, before data
3152 actually was hitting the wire. Extended lock-holding time until
3153 server acknowledges receiving data.
3156 Frequency : extremely rare. Never observed in practice.
3158 Description: avoid generating lustre_handle cookie of 0.
3159 Details : class_handle_hash() generates handle cookies by incrementing
3160 global counter, and can hit 0 occasionaly (this is unlikely, but
3161 not impossible, because initial value of cookie counter is
3162 selected randonly). Value of 0 is used as a sentinel meaning
3163 "unassigned handle" --- avoid it. Also coalesce two critical
3164 sections in this function into one.
3166 Severity : enhancement
3168 Description: allow liblustre clients to delegate truncate locking to OST
3169 Details : To avoid overhead of locking, liblustre client instructs OST to
3170 take extent lock in ost_punch() on client's behalf. New connection
3171 flag is added to handle backward compatibility.
3173 Severity : enhancement
3174 Bugzilla : 4928, 7341, 9758
3175 Description: allow number of OST service threads to be specified
3176 Details : a module parameter allows the number of OST service threads
3177 to be specified via "options ost ost_num_threads={N}" in the
3178 OSS's /etc/modules.conf or /etc/modprobe.conf.
3182 Bugzilla : 6146, 9635, 9895
3183 Description: servers crash with bad pointer in target_handle_connect()
3184 Details : In rare cases when a client is reconnecting it was possible that
3185 the connection request was the last reference for that export.
3186 We would temporarily drop the export reference and get a new
3187 one, but this may have been the last reference and the export
3188 was just destroyed. Get new reference before dropping old one.
3190 Severity : enhancement
3191 Frequency : if client is started with failover MDS
3193 Description: Allow multiple MDS hostnames in the mount command
3194 Details : Try to read the configuration from all specified MDS
3195 hostnames during a client mount in case the "primary"
3198 Severity : enhancement
3200 Description: Stop sending data to evicted clients as soon as possible.
3201 Details : Check if the client we are about to send or are sending data to
3202 was evicted already. (Check is done every second of waiting,
3203 for which l_wait_event interface was extended to allow checking
3204 of exit condition at specified intervals).
3207 Frequency : rare, normally only when NFS exporting is done from client
3209 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3211 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3214 Frequency : always, if srand() is called before liblustre initialization
3216 Description: Liblustre uses system PRNG disturbing its usage by user application
3217 Details : Introduce internal to lustre fast and high-quality PRNG for
3218 lustre usage and make liblustre and some other places in generic
3219 lustre code to use it.
3221 Severity : enhancement
3222 Bugzilla : 9477, 9557, 9870
3223 Description: Verify that the MDS configuration logs are updated when xml is
3224 Details : Check if the .xml configuration logs are newer than the config
3225 logs stored on the MDS and report an error if this is the case.
3226 Request --write-conf, or allow starting with --old_conf.
3228 Severity : enhancement
3230 Description: Handle symlinks in the path when checking if Lustre is mounted.
3231 Details : Resolve intermediate symlinks when checking if a client has
3232 mounted a filesystem to avoid duplicate client mounts.
3237 Description: lconf can hit an error exception but still return success.
3238 Details : The lconf command catches the Command error exception at the top
3239 level script context and will exit with the associated exit
3240 status, but doesn't ensure that this exit status is non-zero.
3245 Description: failure of ptlrpc thread startup can cause oops
3246 Details : Starting a ptlrpc service thread can fail if there are a large
3247 number of threads or the server memory is very fragmented.
3248 Handle this without oopsing.
3251 Frequency : always, only if liblustre and non-default acceptor port was used
3253 Description: liblustre cannot connect to servers with non-default acceptor port
3254 Details : tcpnal_set_default_params() was not called and was therefore
3255 ignoring the environment varaible TCPNAL_PORT, as well as other
3256 TCPNAL_ environment variables
3261 Description: two objects could be created on the same OST for a single file
3262 Details : If an OST is down, in some cases it was possible to create two
3263 objects on a single OST for a single file. No problems other
3264 than potential performance impact and spurious error messages.
3268 Bugzilla : 5681, 9562
3269 Description: Client may oops in ll_unhash_aliases
3270 Details : Client dcache may become inconsistent in race condition.
3271 In some cases "getcwd" can fail if the current directory is
3277 Description: Inode refcounting problems in NFS export code
3278 Details : link_raw functions used to call d_instantiate without obtaining
3279 extra inode reference first.
3283 Bugzilla : 9942, 9903
3284 Description: Referencing freed requests leading to crash, memleaks with NFS.
3285 Details : We used to require that call to ll_revalidate_it was always
3286 followed by ll_lookup_it. Also with revalidate_special() it is
3287 possible to call ll_revalidate_it() twice for the same dentry
3288 even if first occurence returned success. This fix changes semantic
3289 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3290 reference on a request referred from the intent.
3291 ll_intent_release() then releases such a request.
3294 Frequency : rare, normally benchmark loads only
3296 Description: unlinked inodes were kept in memory on the client
3297 Details : If a client is repeatedly creating and unlinking files it
3298 can accumulate a lot of stale inodes in the inode slab cache.
3299 If there is no other client load running this can cause the
3300 client node to run out of memory. Instead flush old inodes
3301 from client cache that have the same inode number as a new inode.
3304 Frequency : SLES9 2.6.5 kernel and long filenames only
3305 Bugzilla : 9969, 10379
3306 Description: utime reports stale NFS file handle
3307 Details : SLES9 uses out-of-dentry names in some cases, which confused
3308 the lustre dentry revalidation. Change it to always use the
3312 Frequency : rare, unless heavy write-truncate concurrency is continuous
3313 Bugzilla : 4180, 6984, 7171, 9963, 9331
3314 Description: OST becomes very slow and/or deadlocked during object unlink
3315 Details : filter_destroy() was holding onto the parent directory lock
3316 while truncating+unlinking objects. For very large objects this
3317 may block other threads for a long time and slow overall OST
3318 responsiveness. It may also be possible to get a lock ordering
3319 deadlock in this case, or run out of journal credits because of
3320 the combined truncate+unlink. Solution is to do object truncate
3321 first in one transaction without parent lock, and then do the
3322 final unlink in a new transaction with the parent lock. This
3323 reduces the lock hold time dramatically.
3326 Frequency : rare, 2.4 kernels only
3328 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3329 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3330 before unmounting their filesystem in order to avoid deadlock.
3331 The kernel_locked() function in 2.4 kernels only checks whether
3332 the kernel lock is held, not whether it is this process that is
3333 holding it as 2.6 kernels do.
3338 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3339 Details : The client ptlrpc code may be trying to reconnect to a down
3340 server before a previous connection attempt has timed out.
3341 Increase the reconnect interval to be longer than the connection
3342 timeout interval to avoid sending duplicate connections to
3346 Frequency : echo_client brw_test command
3348 Description: fix echo_client to work with OST preallocated code
3349 Details : OST preallocation code (5137) didn't take echo_client IO path
3350 into account: echo_client calls filter methods outside of any
3351 OST thread and, hence, there is no per-thread preallocated
3352 pages and buffers to use. Solution: hijack pga pages for IO. As
3353 a byproduct, this avoids unnecessary data copying.
3357 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3358 Description: Client can oops in mdc_commit_close() after open replay
3359 Details : It was possible for the MDS to return an open request with no
3360 transaction number in mds_finish_transno() if the client was
3361 evicted, but without actually returning an error. Clients
3362 would later try to replay that open and may trip an assertion
3363 Simplify the client close codepath, and always return an error
3364 from the MDS in case the open is not successful.
3367 Frequency : rare, 2.6 OSTs only
3369 Description: OST may deadlock under high load on fragmented files
3370 Details : If there was a heavy load and highly-fragmented OST filesystems
3371 it was possible to have all the OST threads deadlock waiting on
3372 allocation of biovecs, because the biovecs were not released
3373 until the entire RPC IO was completed. Instead, release biovecs
3374 as soon as they are complete to ensure forward IO progress.
3376 Severity : enhancement
3378 Description: Support for specifying external journal device at mount
3379 Details : If an OST or MDS device is formatted with an external journal
3380 device, this device major/minor is stored in the ext3 superblock
3381 and may not be valid for failover. Allow detecting and
3382 specifying the external journal at mount time.
3387 Description: Mounting an MDS with pending unlinked files may cause oops
3388 Details : target_finish_recovery() calls mds_postrecov() which returned
3389 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3390 considers this an error and immediately begins cleaning up the
3391 lov, just after starting the mds_lov process
3393 Severity : enhancement
3395 Description: Implement 'lfs df' to report actual free space on per-OST basis
3396 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3397 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3398 usage of objects; '-h' to report in human readable format.
3400 ------------------------------------------------------------------------------
3402 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3409 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3410 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3411 at startup during orphan unlinking. Instead, skip buddy update
3412 before setup, as it will be regenerated after recovery is complete.
3417 Description: If an OST is inactive, its locks might reference stale inodes.
3418 Details : lov_change_cbdata() must iterate over all namespaces, even if
3419 they are inactive to clear inode references from the lock.
3421 Severity : enhancement
3422 Frequency : occasional, if non-standard max_dirty_mb used
3424 Description: Client will block write RPCs if not enough grant
3425 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3426 then the client will block writes while waiting for another RPC
3427 to complete instead of consuming its dirty limit. With change
3428 we get improved performance when max_dirty_mb is small.
3430 Severity : enhancement
3431 Bugzilla : 3389, 6253
3432 Description: Add support for supplementary groups on the MDS.
3433 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3434 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3435 lookups for user supplementary groups into a cache.
3439 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3440 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3441 for liblustre O_EXCL is enforced by MDS.
3445 Description: readdir never returns NULL in liblustre.
3446 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3447 dentry in struct dirent.
3451 Frequency : liblustre only, depends on application IO pattern
3452 Description: liblustre clients evicted if not contacting servers
3453 Details : Don't put liblustre clients into the ping_evictor list, so
3454 they will not be evicted by the pinger ever.
3456 Severity : enhancement
3458 Description: Add ability to evict clients by NID from MDS.
3459 Details : By echoing "nid:$NID" string into
3460 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3461 $NID would be instantly evicted from this MDS and from all active
3462 OSTs connected to it.
3466 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3467 Details : lookup_it_finish() used to query file size from OSTs that was not
3472 Description: service threads change working directory to that of init
3473 Details : Starting lustre service threads may pin the working directory
3474 of the parent thread, making that filesystem busy. Threads
3475 now change to the working directory of init to avoid this.
3479 Frequency : during shutdown only
3480 Description: shutdown with a failed MDS or OST can cause unmount to hang
3481 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3486 Frequency : 2.6 only
3487 Description: chmod/chown may include an extra supplementary group
3488 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3489 supplementary group and if none is specified this is used.
3492 Bugzilla : 5479 (6816)
3494 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3495 Details : If lookup is in progress on a file that is unlinked we might try
3496 to revalidate the inode and fail in revalidate after lookup is
3497 complete and ll_file_open() enqueues the open again but
3498 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3501 Frequency : always, if lconf --abort_recovery used
3503 Description: lconf --abort_recovery fails with 'Operation not supported'
3504 Details : lconf was attempting to abort recovery on the MDT device and not
3507 ------------------------------------------------------------------------------
3509 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3514 Frequency : rare (only unsupported configurations with a node running as an
3516 Bugzilla : 6514, 5137
3517 Description: Mounting a Lustre file system on a node running as an OST could
3519 Details : OSTs now preallocates memory needed to write out data at
3520 startup, instead of when needed, to avoid having to
3521 allocate memory in possibly low memory situations.
3522 Specifically, if the file system is mounted on on OST,
3523 memory pressure could force it to try to write out data,
3524 which it needed to allocate memory to do. Due to the low
3525 memory, it would be unable to do so and the node would
3526 become unresponsive.
3528 Severity : enhancement
3530 Description: Addition of lconf --service command line option
3531 Details : lconf now accepts a '--service <arg>' option, which is
3532 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3534 Severity : enhancement
3536 Description: Failover mode is now the default for OSTs.
3537 Details : By default, OSTs will now run in failover mode. To return to
3538 the old behaviour, add '--failout' to the lmc line for OSTs.
3540 Severity : enhancement
3542 Description: Health checks are now provided for MDS and OSTs
3543 Details : Additional detailed health check information on MSD and OSTs
3544 is now provided through the procfs health_check value.
3547 Frequency : occasional, depends on IO load
3549 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3550 after numerous create/delete cycles
3551 Details : The ext3 inode allocation policy would not allocate new inodes
3552 very well on the OSTs because there are no new directories
3553 being created. Instead we look for groups with free space if
3554 the parent directories are nearly full.
3559 Description: Network or server problems during mount may cause partially
3560 mounted clients instead of returning an error.
3561 Details : The config llog parsing code may overwrite the error return
3562 code during mount error handling, returning success instead
3568 Description: MDS can fail to allocate large reply buffers
3569 Details : After long uptimes the MDS can fail to allocate large reply
3570 buffers (e.g. zconf client mount config records) due to memory
3571 fragmentation or consumption by the buffer cache. Preallocate
3572 some large reply buffers so that these replies can be sent even
3573 under memory pressure.
3577 Frequency : rare (liblustre)
3578 Description: fsx running with liblustre complained that using truncate() to
3579 extend the file doesn't work. This patch corrects that issue.
3580 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3581 ATTR_SIZE and lsm use in llu_setattr_raw.
3585 Frequency : rare, only 2.6 kernels
3586 Description: Unusual file access patterns on the MDS may result in inode
3587 data being lost in very rare circumstances.
3588 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3589 code in the 2.6 kernel caused the inode and/or EA data not to
3590 be read from disk, causing single-file corruption.
3594 Frequency : rare, only 2.6 filesystems using extents
3595 Description: Heavy concurrent write and delete load may cause data corruption.
3596 Details : It was possible under high-load situations to have an extent
3597 metadata block in the block device cache from a just-unlinked
3598 file overwrite a newly-allocated data block. We now unmap any
3599 metadata buffers that alias just-allocated data blocks.
3603 Frequency : filesystems with default stripe_count larger than 77
3604 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3605 Details : lconf specifies an inode size of 4096 bytes when the default
3606 stripe_count is larger than 77. This conflicts with the default
3607 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3608 this case to avoid pinning too much memory for large EAs.
3610 ------------------------------------------------------------------------------
3612 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3617 Frequency : rare (extremely heavy IO load with hundreds of clients)
3619 Description: Client is evicted, gets IO error writing to file
3620 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3621 caused clients to be evicted for AST timeouts. The fixes in
3622 bug 5192 mean we no longer need to have such short AST timeouts
3623 so ldlm_timeout has been increased.
3626 Frequency : occasional during --force or --failover shutdown under load
3627 Bugzilla : 5949, 4834
3628 Description: Server oops/LBUG if stopped with --force or --failover under load
3629 Details : a collection of import/export refcount and cleanup ordering
3630 issues fixed for safer force cleanup
3633 Frequency : only filesystems larger than 120 OSTs
3634 Bugzilla : 5990, 6223
3635 Description: lfs getstripe would oops on a very large filesystem
3636 Details : lov_getconfig used kfree on vmalloc'd memory
3639 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3640 Bugzilla : 6242, 6243
3641 Description: reading from files that had been truncated to a non-zero size
3642 but never opened returned no data
3643 Details : ll_file_read() reads zeros from no-object files to EOF
3648 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3649 Details : There's little harm in aborting MDS/OSS recovery and letting it
3650 try again, so I removed the LASSERT and return an error instead.
3652 Severity : enhancement
3654 Description: New debugging infrastructure for tracking down data corruption
3655 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3656 (b) cover more of the client-side code path, and (c) try to narrow
3657 down where problems occurred
3661 Bugzilla : 3819, 4364, 4397, 6313
3662 Description: Racing close and eviction MDS could cause assertion in mds_close
3663 Details : It was possible to get multiple mfd references during close and
3664 client eviction, leading to one thread referencing a freed mfd.
3666 Severity: : enhancement
3667 Bugzilla : 3262, 6359
3668 Description: Attempts to reconnect to servers are now more aggressive.
3669 Details : This builds on the enhanced upcall-less recovery that was added
3670 in 1.4.2. When trying to reconnect to servers, clients will
3671 now try each server in the failover group every 10 seconds. By
3672 default, clients would previously try one server every 25 seconds.
3677 Description: After recovery, certain operations trigger a failed
3678 assertion on a client.
3679 Details : Failing over an mds, using lconf -d --failover, while a
3680 client was doing a readdir() call would cause the client to
3681 LBUG after recovery completed and the readdir() was resent.
3683 Severity : enhancement
3685 Description: Default groups are now added by lconf
3686 Details : You can now run lconf --group <servicename> without having to
3687 manually add groups with lmc.
3690 Frequency : occasional
3692 Description: Nodes with an elan id of 0 trigger a failed assertion
3695 Frequency : always when accessing e.g. tty/console device nodes
3697 Description: tty and some other devices nodes cannot be used on lustre
3698 Details : file's private_data field is used by device data and lustre
3699 values in there got lost. New field was added to struct file to
3700 store fs-specific private data.
3703 Frequency : when exporting Lustre via NFS
3705 Description: NFSD failed occasionally when looking up a path component
3706 Details : NFSD is looking up ".." which was broken in ext3 directories
3707 that had grown large enough to become hashed.
3710 Frequency : Clusters with multiple interfaces not on the same subnet
3712 Description: Nodes will repeatedly try to reconnect to an interface which it
3713 cannot reach and report an error to the log.
3714 Details : Extra peer list entries will be created by lconf with some peers
3715 unreachable. lconf now validates the peer before adding it.
3718 Frequency : Only if a default stripe is set on the filesystem root.
3720 Description: Setting a default stripe on the filesystem root prevented the
3721 filesystem from being remounted.
3722 Details : The client was sending extra request flags in the root getattr
3723 request and did not allocate a reply buffer for the dir EA.
3726 Frequency : occasional, higher if lots of files are accessed by one client
3727 Bugzilla : 6159, 6097
3728 Description: Client trips assertion regarding lsm mismatch/magic
3729 Details : While revalidating inodes the VFS looks up inodes with ifind()
3730 and in rare cases can find an inode that is being freed.
3731 The ll_test_inode() code will free the lsm during ifind()
3732 when it finds an existing inode and then the VFS later attaches
3733 this free lsm to a new inode.
3737 Bugzilla : 6422, 7030
3738 Description: MDS deadlock between mkdir and client eviction
3739 Details : Creating a new file via mkdir or mknod (starting a transaction
3740 and getting the ns lock) can deadlock with client eviction
3741 (gets ns lock and trying to finish a synchronous transaction).
3744 Frequency : occasional
3745 Description: While starting a server, the fsfilt_ext3 module could not be
3747 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3748 kernels. Previously, lconf would still use the ext3 name
3749 when trying to load modules. Now, it will correctly use
3750 ext3 on 2.4 and ldiskfs on 2.6.
3752 Severity : enhancement
3753 Description: The default stripe count has been changed to 1
3754 Details : The interpretation of the default stripe count (0, to lfs
3755 or lmc) has been changed to mean striping across a single
3756 OST, rather than all available. For general usage we have
3757 found a stripe count of 1 or 2 works best.
3759 Severity : enhancement
3760 Description: Add support for compiling against Cray portals.
3761 Details : Conditional compiling for some areas that are different
3765 Frequency : occasional
3766 Bugzilla : 6409, 6834
3767 Description: Creating files with an explicit stripe count may lead to
3768 a failed assertion on the MDS
3769 Details : If some OSTs are full or unavailable, creating files may
3770 trigger a failed assertion on the MDS. Now, Lustre will
3771 try to use other servers or return an error to the
3775 Frequency : occasional
3777 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3778 caused lock timeout during readahead (since 1.4.2).
3779 Details : Processes doing readahead might match a lock that hasn't been
3780 granted yet if there are overlapping and conflicting lock
3781 requests. The readahead process waits on ungranted lock
3782 (original lock is CBPENDING), while OST waits for that process
3783 to cancel CBPENDING read lock and eventually evicts client.
3785 Severity : enhancement
3787 Description: Initial enabling of flock support for clients
3788 Details : Implements fcntl advisory locking and file status functions.
3789 This feature is provided as an optional mount flag (default
3790 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3791 locking are implemented yet, and those that are are not guaranteed
3792 to be completely correct in production environments.
3793 mount -t lustre -o [flock|noflock] ...
3796 Frequency : occasional
3798 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3799 assertion in the ext3 JBD (journaling) layer.
3800 Details : The b_committed_data struct is protected by the big kernel lock
3801 in 2.4 kernels, serializing journal_commit_transaction() and
3802 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3803 access to this struct. In 2.6 kernels there is finer grained
3804 locking to improve SMP performance of the JBD layer.
3808 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3813 Description: merge and cleanup kernel patches.
3816 -----------------------------------------------------------------------------
3818 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3820 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3821 versions of Lustre. Please update all servers and clients to
3822 version 1.4.2 or later at the same time. You must also run
3823 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3824 to update the configuration logs.
3826 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3827 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3828 - swab llog records of type '0' so we get proper header size/idx (5861)
3829 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3830 - fix rename of one directory over another leaking an inode (5953)
3831 - avoid SetPageDirty on 2.6 (5981)
3832 - don't re-add just-being-destroyed locks to the waiting list (5653)
3833 - when creating new directories, inherit the parent's custom
3834 striping settings if present parent (3048)
3835 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3836 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3837 - don't hold client locks on temporary worklist from l_lru (5666)
3838 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3839 - automatically evict dead clients (5921)
3840 - Update file size properly in create+truncate+fstat case (6196)
3841 - Do not unhash mountpoint dentries, do not allow removal of
3843 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3844 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3845 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3846 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3847 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3848 - Correctly handle reads of files with no objects (6243)
3849 - lctl recover will also mark a device active if deactivate used (5933)
3851 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3852 - allow --write-conf on an MDS with different nettype than client (5619)
3853 - don't write config llogs to MDS for mounts not from that MDS (5617)
3854 - lconf should create multiple TCP connections from a client (5201)
3855 - init scripts are now turned off by default; run chkconfig --on
3856 lustre and chkconfig --on lustrefs to use them
3857 - upcalls are no longer needed for clients to recover to failover
3859 - add --abort-recovery option to lconf to abort recovery on device
3861 - add support for an arbitrary number of OSTs (3026)
3862 - Quota support protocol changes.
3863 - forward compatibility changes to wire structs (6007)
3864 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3865 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3866 - improved "lustre status" script
3867 - initialize blocksize for non-regular files (6062)
3868 - added --disable-server and --disable-client configure options (5782)
3869 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3870 - Vanilla 2.4.29 support
3871 - increase maximum number of obd devices to 520 (6242)
3872 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3873 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3874 SLES 9 2.6.5 kernels
3875 - limit stripes per file to 160 (the maximum EA size) (6093)
3877 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3880 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3881 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3882 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3883 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3884 - small ext3 extents cleanups and fixes (5733)
3885 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3886 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3887 - avoid needless client->OST connect, fix handle mismatch (5317)
3888 - fix DLM error path that led to out-of-sync client, long delays (5779)
3889 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3890 - fix several locking issues related to i_size (5492,5624,5654,5672)
3891 - don't move pending lock onto export if it is already evicted (5683)
3892 - fix kernel oops when creating .foo in unlinked directory (5548)
3893 - fix deadlock in obdfilter statistics vs. object create (5811)
3894 - use time_{before,after} to avoid timer jiffies wrap (5882)
3895 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3896 - Do not leak request if server was not able to process it (5154)
3897 - If mds_open unable to find parent dir, make that negative lookup(5154)
3898 - don't create new directories with extent-mapping (5909, 5936)
3900 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3901 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3902 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3903 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3904 - support for RHEL4 kernel on i686 (5773)
3905 - provide error messages when incompatible logs are encountered (5898)
3907 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3908 * version 1.4.0.10 (1.4.1 release candidate 1)
3910 - don't keep a lock reference when lock is not granted (4238)
3911 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3912 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3913 - drop import inflight refcount on signal_completed_replay error (5255)
3914 - unlock page after async write error during send (3677)
3915 - handle missing objects in filter_preprw_read properly (5265)
3916 - no transno return for symlink open, don't save no-trasno open (3440)
3917 - don't try to complete elan receive that already failed (4012)
3918 - free RPC server reply state on error (5406)
3919 - clean up thread from ptlrpc_start_thread() on error (5160)
3920 - readahead could read extra page into cache that wasn't ejected (5388)
3921 - prevent races in class_attach/setup/cleanup/detach (5260)
3922 - don't dereference de->d_inode after l_dput of de (5458)
3923 - use "int" for stripe value returned from lock_to_stripe (5544)
3924 - mballoc allocation and error-checking fixes in 2.6 (5504)
3925 - block device patches to fix I/O request sizes in 2.6 (5482)
3926 - look up hostnames for IB nals (5602)
3927 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3928 - don't start multiple acceptors for the same port (5277)
3929 - fix incorrect LASSERT in mds_getattr_name (5635)
3930 - export a proc file for general "ping" checking (5628)
3931 - fix "lfs check" to not block when the MDS is down (5628)
3933 - service request history (4965)
3934 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3935 - create an "evict_client" /proc entry on OSTs, like the MDS has
3936 - fix mount usage message, return errors per mount(8) (5168)
3937 - change grep [] to grep "[]" in tests so they work in more UMLs
3938 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3939 - remove ancient LOV_MAGIC_V0 EA support (5047)
3940 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3941 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3942 - allow readahead allocations to fail when low on memory (5383)
3943 - mmap locking landed again, after considerable improvement (2828)
3944 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3946 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3949 - send OST transaction number in read/write reply to free req (4966)
3950 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3951 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3952 - the watchdog thread now runs as interruptible (5246)
3953 - flock/lockf fixes (but it's still disabled, pending 5135)
3954 - don't use EXT3 constants in llite code (5094)
3955 - memory shortage at startup could cause assertion (5176)
3957 - reorganization of lov code
3958 - single portals codebase
3960 - add extents/mballoc support (5025)
3961 - direct I/O reads in the obdfilter (4048)
3962 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3964 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3967 - send OST transaction number in read/write reply to free req (4966)
3968 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3969 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3970 - the watchdog thread now runs as interruptible (5246)
3971 - handle missing objects in filter_preprw_read properly (5265)
3972 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3973 - drop import inflight refcount on signal_completed_replay error (5255)
3974 - unlock page after async write error during send (3677)
3975 - return original error code on reconstructed replies (3761)
3976 - no transno return for symlink open, don't save no-trasno open (3440)
3978 - add pid to ldlm debugging output (4922)
3979 - bump the watchdog timeouts -- we can't handle 30sec yet
3980 - extra debugging for orphan dentry/inode bug (5259)
3982 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3985 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3986 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3987 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3988 - fix NULL dereference in /proc/sys/portals/routes (4827)
3989 - allow failed mdc_close() operations to be interrupted (4561)
3990 - stop precreate on OST before MDS would time out on it (4778)
3991 - don't send partial-page writes before EOF from client (4410)
3992 - discard client grant for sub-page writes on large-page clients (4520)
3993 - don't free dentries not owned by NFS code, check generation (4806)
3994 - fix lsm leak if mds_create_objects() fails (4801)
3995 - limit debug_daemon file size, always print CERROR messages (4789)
3996 - use transno after validating reply (3892)
3997 - process timed out requests if import state changes (3754)
3998 - update mtime on OST during writes, return in glimpse (4829)
3999 - add mkfsoptions to LDAP (4679)
4000 - use ->max_readahead method instead of zapping global ra (5039)
4001 - don't interrupt __l_wait_event() during strace
4003 - add software watchdogs to catch hung threads quickly (4941)
4004 - make lustrefs init script start after nfs is mounted
4005 - fix CWARN/ERROR duplication (4930)
4006 - return async write errors to application if possible (2248)
4007 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4008 - print NAL number in %x format (4645)
4009 - update barely-supported suse-2.4.21-171 series (4842)
4010 - support for sles 9 %post scripts
4011 - support for building 2.6 kernel-source packages
4012 - support for sles km_* packages
4014 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4017 - ignore -ENOENT errors in osc_destroy (3639)
4018 - notify osc create thread that OSC is being cleaned up (4600)
4019 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4020 - reconstruct ost_handle() like mds_handle() (4657)
4021 - create a new thread to do import eviction to avoid deadlock (3969)
4022 - let lconf resolve symlinked-to devices (4629)
4023 - don't unlink "objects" from directory with default EA (4554)
4024 - hold socknal file ref over connect in case target is down (4394)
4025 - allow more than 32000 subdirectories in a single directory (3244)
4026 - fix blocks count for O_DIRECT writes (3751)
4027 - OST returns ENOSPC from object create when no space left (4539)
4028 - don't send truncate RPC if file size isn't changing (4410)
4029 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4030 - bind to privileged port in socknal and tcpnal (3689)
4032 - rate limit CERROR/CWARN console message to avoid overload (4519)
4033 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4034 - basic mmap support (3918)
4035 - kernel patch series update from b1_4 (4711)
4037 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4040 - avoid crash during MDS cleanup with OST shut down (2775)
4041 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4042 - don't use bad inodes on the MDS (3744)
4043 - dynamic object preallocation to improve recovery speed (4236)
4044 - don't hold spinlock over lock dumping or change debug flags (4401)
4045 - don't zero obd_dev when it is force cleaned (3651)
4046 - print grants to console if they go negative (4431)
4047 - "lctl deactivate" will stop automatic recovery attempts (3406)
4048 - look for existing locks in ldlm_handle_enqueue() (3764)
4049 - don't resolve lock handle twice in recovery avoiding race (4401)
4050 - revalidate should check working dir is a directory (4134)
4052 - don't always mark "slow" obdfilter messages as errors (4418)
4054 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4057 - don't close LustreDB during write_conf until it is done (3860)
4058 - fix typo in lconf for_each_profile (3821)
4059 - allow dumping logs from multiple threads at one time (3820)
4060 - don't allow multiple threads in OSC recovery (3812)
4061 - fix debug_size parameters (3864)
4062 - fix mds_postrecov to initialize import for llog ctxt (3121)
4063 - replace config semaphore with spinlock (3306)
4064 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4065 - don't allow enqueue to complete on a destroyed export (3822)
4066 - down write_lock before checking llog header bitmap (3825)
4067 - recover from lock replay timeout (3764)
4068 - up llog sem before sending rpc (3652)
4069 - reduce ns lock hold times when setting kms (3267)
4070 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4071 - fix NULL deref and obd_dev leak on setup error (3312)
4072 - replace some LBUG about llog ops with error handling (3841)
4073 - don't match INVALID dentries from d_lookup and spin (3784)
4074 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4075 - fix invalid assertion in ptlrpc_set_wait (3880)
4077 - add libwrap support for the TCP acceptor (3996)
4078 - add /proc/sys/portals/routes for non-root route listing (3994)
4079 - allow setting MDS UUID in .xml (2580)
4080 - print the stack of a process that LBUGs (4228)
4082 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4085 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4086 - make sure to unset replay flag from failed open requests (3440)
4087 - if default stripe count is 0, use OST count for inode size (3636)
4088 - update parent mtime/ctime on client for create/unlink (2611)
4089 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4090 - free recovery state on server during a forced cleanup (3571)
4091 - unregister_reply for resent reqs (3063)
4092 - loop back devices mounting and status check on 2.6 (3563)
4093 - fix resource-creation race that can provoke i_size == 0 (3513)
4094 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4095 - more debugging for page-accounting assertion (3746)
4096 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4097 - avoid deadlock after precreation failure (3758)
4098 - fix race and lock order deadlock in orphan handling (3450, 3750)
4099 - add validity checks when grabbing inodes from l_ast_data (3599)
4101 - add /proc/.../recovery_status to obdfilter (3428)
4102 - lightweight CDEBUG infrastructure, debug daemon (3668)
4103 - change default OSC RPC parameters to be better on small clusters
4104 - turn off OST read cache for files smaller than 32MB
4105 - install man pages and include them in rpms (3100)
4106 - add new init script for (un)mounting lustre filesystems (2593)
4107 - run chkconfig in %post for init scripts (3701)
4108 - drop scimac NAL (unmaintained)
4110 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4113 - clean kiobufs before and after use (3485)
4114 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4115 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4116 - change init script to fail more clearly if not run as root (1528)
4117 - allow clients to reconnect during replay (1742)
4118 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4119 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4120 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4121 - fix cond_resched (3554)
4122 - only evict unfinished clients after recovery (3515)
4123 - allow bulk resend, prevent data loss (3570)
4124 - dynamic ptlrpc request buffer allocation (2102)
4125 - don't allow unlinking open directory if it isn't empty (2904)
4126 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4127 - remove extraneous obd dereference causing LASSERT failure (3334)
4128 - don't use get_cycles() when creating temp. files on the mds (3156)
4129 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4130 - handle EEXIST for set-stripe, set proper directory name (3336)
4132 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4133 - fix ksocknal_fmb_callback() error messages (2918)
4135 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4138 - don't copy lvb into (possibly NULL) reply on error (2983)
4139 - don't deref dentry after dput, don't free lvb on error (2922)
4140 - use the kms to determine writeback rpc length (2947)
4141 - increment oti_logcookies when osc is inactive (2948)
4142 - update client's i_blocks count via lvb messages (2543)
4143 - handle intent open/close of special files properly (1557)
4144 - mount MDS with errors=remount-ro, like obdfilter (2009)
4145 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4146 - don't use cancelling-locks' kms values (2947)
4147 - use highest lock extent for kms, not last one (2925)
4148 - don't dereference ERR_PTR() dentry in error handling path (3107)
4149 - fix thread race in portals_debug_dumplog() (3122)
4150 - create lprocfs device entries at setup instead of at attach (1519)
4151 - common AST error handler, don't evict client on completion race (3145)
4152 - zero nameidata in detach_mnt in 2.6 (3118)
4153 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4154 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4155 - initialize RPC timeout timer earlier for 2.6 (3219)
4156 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4157 - print nal/nid for unknown nid (3258)
4158 - additional checks for oscc recovery before doing precreate (3284)
4159 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4160 - don't crash in mdc_close for bad permissions on open (3285)
4161 - zero i_rdev for non-device files (3147)
4162 - clear page->private before handing to FS, better assertion (3119)
4163 - tune the read pipeline (3236)
4164 - fix incorrect decref of invalidated dentry (2350)
4165 - provide read-ahead stats and refine rpc in flight stats (3328)
4166 - don't hold journal transaction open across create RPC (3313)
4167 - update atime on MDS at close time (3265)
4168 - close LDAP connection when recovering to avoid server load (3315)
4169 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4170 - don't leak open file on MDS after open resend (3325)
4171 - serialize filter_precreate and filter_destroy_precreated (3329)
4172 - loop device shouldn't call sync_dev() for nul device (3092)
4173 - clear page cache after eviction (2766)
4174 - resynchronize MDS->OST in background (2824)
4175 - refuse to mount the same filesystem twice on same mountpoint (3394)
4176 - allow llmount to create routes for mounting behind routers (3320)
4177 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4178 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4179 - fix rare problem with rename on htree directories (3417)
4181 - allow default OST striping configuration per directory (1414)
4182 - fix compilation for qswnal for 2.6 kernels (3125)
4183 - increase maximum number of MDS request buffers for large systems
4184 - change liblustreapi to be useful for external progs like lfsck (3098)
4185 - increase local configuration timeout for slow disks (3353)
4186 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4188 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4191 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4192 - don't overwrite extent policy data in reply if lock was blocked (2901)
4193 - drop filter export grants atomically with removal from device (2663)
4194 - del obd_self_export from work_list in class_disconnect_exports (2908)
4195 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4196 - swab reply message in mdc_close, other PPC fixes (2464)
4197 - fix destroying of named logs (2325)
4198 - overwrite old logs when running lconf --write_conf (2264)
4199 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4200 - fix race in target_handle_connect (2898)
4201 - mds_reint_create() should take same inode create lock (2926)
4202 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4203 - don't close files for self_export to avoid uninitialized obd (2936)
4204 - allow MDS with the same name as client node (2939)
4205 - hold dentry reference for closed log files for unlink (2325)
4206 - reserve space for all logs during transactions (2059)
4207 - don't evict page beyond end of stripe extent (2925)
4208 - don't oops on a deleted current working directory (2399)
4209 - handle hard links to targets without a parent properly (2517)
4210 - don't dereference NULL lock when racing during eviction (2867)
4211 - don't grow lock extents when lots of conflicting locks (2919)
4213 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4216 - account for cache space usage on clients to avoid data loss (974)
4217 - lfsck support in lustre kernel code (2349)
4218 - reduce journal credits needed for BRW writes (2370)
4219 - orphan handling to avoid losing space on client/server crashes
4220 - ptlrpcd can be blocked, stopping ALL progress (2477)
4221 - use lock value blocks to assist in proper KMS, faster stat (1021)
4222 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4223 - recovery for initial connections (2355)
4224 - fixes for mds_cleanup_orphans (1934)
4225 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4226 - block all file creations until orphan recovery completes (1901)
4227 - client remove rq_connection from request struct (2423)
4228 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4229 - recovery timer race (2670)
4230 - mdc_close recovey bug (2532)
4231 - ptlrpc cleanup bug (2710)
4232 - mds timeout on local locks (2588)
4233 - namespace lock held during RPCs (2431)
4234 - handle interrupted sync write properly (2503)
4235 - don't try to handle a message that hasn't been replied to (2699)
4236 - client assert failure during cleanup after abort recovery (2701)
4237 - leak mdc device after failed mount (2712)
4238 - ptlrpc_check_set allows timedout requests to complete (2714)
4239 - wait for inflight reqs when ptlrpcd finishes (2710)
4240 - make sure unregistered services are removed from the srv_list
4241 - reset bulk XID's when resending them (caught by 1138 test)
4242 - unregister_bulk after timeout
4243 - fix lconf error (2694)
4244 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4245 - readahead locks pages, leaves pending causing memory pressure (2673)
4246 - increase OST request buffers to 4096 on large machines (2729)
4247 - fix up permission of existing directories in simple_mkdir (2661)
4248 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4249 - don't assign transno to errored transactions (2742)
4250 - don't delete objects on OST if given a bogus objid from MDS (2751)
4251 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4252 - if rq_no_resend, then timeout request after recovery (2432)
4253 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4254 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4255 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4256 - MDS recovery shouldn't delete objects that it has given out (2730)
4257 - if enqueue arrives after completion, don't clobber LVB (2819)
4258 - don't unlock pages twice when trigger_group_io returns error (2814)
4259 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4260 - don't write pages to disk if there was an error (1450)
4261 - don't ping imports that have recovery disabled (2676)
4262 - take buffered bytes into account when balancing socknal conn (2817)
4263 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4264 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4265 - remove little-endian swabbing of llog records (1987)
4266 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4267 - retry reposting request buffers if they fail (1191)
4268 - grow extent at grant time to avoid granting a revoked lock (2809)
4269 - lock revoke doesn't evict page if covered by a second lock (2765)
4270 - disable VM readahead to avoid reading outside lock extents (2805)
4272 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4273 - updated kernel patches for hp-2.4.20 kernel (2681)
4275 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4278 - fix truncated write corruption (2366)
4279 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4281 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4282 - ensure len(uuid) < 37 in lmc (1171)
4283 - fix ia64 OOPS in llog_test (2255)
4284 - zero end of page at obdfilter for partial page writes (2648)
4285 - don't leave stale dentries around after renames (bug 2428)
4286 - fix timeouts when evicting a client with a single lock held (2642)
4287 - set deadline for the initial HELLO message to drain (2634)
4288 - print out dotted-quad IP addresses in the socknal (2302)
4290 - additional debugging for MDS client eviction problem (2443)
4291 - fix mkfsoptions support for osts (2603, 2604)
4293 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4296 - add series for the vanilla 2.6.0 kernel
4297 - add series for the vanilla 2.4.24 kernel
4298 - add series for a cray x86/64 UL kernel drop
4299 - fix xattr patches for the vanilla 2.4.19 series
4301 - generate true UUIDs in lmc (1171)
4302 - have portals stack dumping break in UML (2466)
4303 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4304 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4305 - wake callback waiting threads on client eviction (2460)
4306 - Add --ptldebug and --subsystem to lmc (1719)
4307 - update assertion to allow safe interrupt allocation
4308 - set rq_no_resend for cancel requests (2432)
4309 - recalculate ptlrpcd timeout after resend (2494)
4310 - call vfs_rmdir when removing pending directories (2368)
4311 - fix renaming a file to itself (2429)
4312 - lmc creates a default one-stripe lov (2454)
4313 - expand procfs space to handle large clusters (2326)
4314 - increase UML stack to avoid overflow
4315 - update lconf's list of debug and subsystem masks
4316 - fix lfs find --obd (2510)
4317 - /proc tunable for disabling filter read caching (2591)
4318 - stop rpm packages from altering slapd.conf (2301)
4319 - disable nagle in the socknal under 0conf (2578)
4320 - choose mds inode size based on stripe count (2572)
4321 - fix kernel-source rpm problems (2516)
4323 - add --disable-doc to avoid pdf generation (2421)
4324 - update documentation, tests, type-os, comments
4325 - avoid format warnings on ia64
4326 - remove the TOE NAL
4327 - tiny code cleanups by removing unused fields
4329 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4332 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4333 - remove the most blatant lies from BUILDING file (2371)
4334 - change default debug level to reasonable production setting
4335 - reduce client side cache size to reduce cache flush time
4336 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4337 - make TCP zerocopy and pinger support enabled by default (2476)
4338 - sync writes completed after process exits caused crashes (2319)
4339 - maintain correct mount count on the MDS (2356)
4340 - backout 1557, because 2316 wasn't really fixed
4341 - better file I/O statistics gathering in /proc
4342 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4343 - another kernel patch to fix zero-copy TCP function export
4344 - don't take duplicate lock when processing re-sent getattr (2420)
4345 - lctl uses obd_self_export instead of creating new conn (2353)
4346 - MDS/OST recovery case which requires object creation asserted (2425)
4347 - move lfs from /usr/sbin to /usr/bin in packages
4348 - fix race between mds_client_add and mds_client_free (2417)
4349 - use kmalloc instead of slabs in portals (2430)
4350 - don't create duplicate records when a failover MDS is present (2442)
4351 - remove unnecessary mount age check (2332)
4352 - don't remove directory inodes from locks prematurely (2451)
4353 - don't break if MDS service name is the same as hostname (2103)
4354 - fix races in client write RPC generation when cache full (2482)
4356 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4359 - remove now-unused request->rq_obd (278)
4360 - if an allocation fails, print out how much memory we've used (1933)
4361 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4362 - add the "configurable stack size" patch to most series files (1256)
4363 - ability to write large log records, for 100+ OST configs (2306)
4364 - fix NULL deref when filter_prep fails (2314)
4365 - fix operator precedence error in filter_sync
4366 - dynamic allocation of socknal TX descriptors (2315)
4367 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4368 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4369 - ensure that CWARN messages in Portals always get to the syslog
4370 - __init/__exit are not for prototype decls (ldlm_init/exit)
4371 - x86-64 compile warning fixes
4372 - fix gateway LMC keyword conflict (2318)
4373 - fix MDS lock inversions in getattr/reint paths (1844)
4374 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4375 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4376 - DLM race condition prevented some lock evictions (2328)
4377 - ENOMEM detection and retry on socknal sends (2230)
4378 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4379 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4380 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4381 - better error messages when a client is rejected during recovery (1505)
4382 - avoid cancelling locks which were never granted, after failure (2330)
4383 - fix i_sem/journal inversion in mds_client_add (2333)
4384 - fix truncate/getattr lock cycle deadlock (2334)
4385 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4386 - fix two rare exit paths which could leak an l_lock() ref (2321)
4387 - fencepost error in MDS/OST orphan recovery (2226)
4388 - make log record alignment 8 bytes (1988)
4389 - lstripe now fails when requested offset > ost_count (2237)
4390 - ensure that all kernel series have a complete list.h (1607)
4391 - fix crashes in special-file operations (2316)
4392 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4394 - allow configurable automake binary, for testing new versions
4395 - small update to the lfs documentation
4397 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4399 * fix negative export reference count in fsfilt_sync (2312)
4401 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4402 * release candidate 0.9.1
4404 - orphans are moved into the PENDING directory for possible recovery
4405 - replayed opens now open by fid for orphan/rename safety (1042)
4406 - last close of an orphan inode generates a transno (683)
4407 - chdir() and mount() now pin the directory entry (1020)
4408 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4409 - discard very old requests without processing them (1502)
4410 - remove some common, well-understood CERRORs (1505)
4411 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4412 - clear "grant" flags in OST replies until OST grant code lands (1644)
4413 - fix read performance by not clobbering i_blksize on client (1598)
4414 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4415 - make LRU size a /proc tunable, clears locks when reduced (707)
4416 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4417 - server locks take references on exports now (1558)
4418 - build fixes for 2.4.20-rh trees (1663)
4419 - return an error from lov_create if all OSCs are inactive (1751)
4420 - fix import levels when a reconnect happens without a timeout (1597)
4421 - exit early from mds_open if we get a lookup error (1749)
4422 - partial page read at EOF wouldn't wait for disk before sending (1642)
4423 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4424 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4425 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4426 - write/truncate lock inversion (1639)
4427 - Don't auto-load obdclass, portals modules during cleanup (1495)
4428 - fix timestamps from jumping to "now" (1763)
4429 - extra journal assertions (1648)
4430 - add an extra multiunlink test (1771)
4431 - fix read_record/write_record API (1776)
4432 - fix leak of offset_extent, possible incorrect i_size later (1772)
4433 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4434 - Debugging for the kqswnal_get_idle_tx problems (1820)
4435 - Allow recovery to be attempted multiple times (1536)
4436 - Write out MDS last_rcvd file after it is first created (1600)
4437 - Fix tx_descriptor leak in failed transmit situations (1827)
4438 - ext3 journaling fixes for assertion failure after IO error (1871)
4439 - class_export_put() on freed export after completion AST error (1896)
4440 - Fix revalidate looping in VFS (1322)
4441 - Don't access a freed export during MDS_REINT timeout (1521)
4442 - Add open-unlink recovery support on the MDS (1673,1764)
4443 - Return an error if no MDS data was read from last_rcvd (1946)
4444 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4445 - Don't setup a disk device that doesn't match exported UUID (317)
4446 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4447 - avoid committing NULL handle in force close
4448 - local.sh is now a one-stripe LOV configuration
4449 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4450 - fix ext3 htree duplicate directory entry corruption (1516)
4451 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4452 - update to new LOV EA format (2097)
4453 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4455 2003-06-15 Phil Schwan <phil@clusterfs.com>
4458 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4459 - per-import/export recovery handling (958, 931, 959)
4460 - multiple last-rcvd slots, for serving multiple FSes (949)
4461 - connections are again shared between multiple imp/exports (963, 964)
4462 - "umount -f" would hang if any requests needed to be sent (393, 978)
4463 - avoid pinning large req buffer by copying for queued messages (989)
4464 - add "uuid" to "lctl device" command to help upcalls (991)
4465 - "open" RPCs with transnos would confuse recovery counters (1037)
4466 - do proper endian conversion of all wire messages (288, 340, 891)
4467 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4468 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4469 - fix ptlrpc_connection leak in target_handle_connect (1174)
4470 - fix import refcounting bug in OST and MDS cleanup (1134)
4471 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4472 - fix very unlikely obd_types race condition (501)
4473 - remove osc_open hack for echo_client (1187)
4474 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4475 - a failure in read_inode2 leads to deadlock (1139)
4476 - cancel ack-locks as soon as transaction is committed (1072)
4477 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4478 - make sure to commitrw after any preprw to avoid deadlock (1162)
4479 - failing to execute a file in a lustre FS would lock inode (1203)
4480 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4481 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4482 - obd_uuid2tgt would walk off the end of the list (1255)
4483 - on IA64 the osc would give portals incorrect bulk size (1258)
4484 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4485 - fix lock inversion caused by new llite matching code (1282)
4486 - limit the number of dirty pages on a client to 10MB (1286)
4487 - timed out locks were not being corrected cancelled (1289)
4488 - fix O_DIRECT above 4GB on IA-32 (1292)
4489 * major user-visible changes
4490 - fail out/fail over policy now controlled by the upcall (993)
4492 - add OBD_PING to check server availability and failure (954)
4493 - lustre messages are now sent in sending host order (288, 340, 891)
4494 - add eadatalen to MDS getattr reply (340)
4495 - OST read replies may contain second buffer, with per-page status (593)
4497 2003-03-11 Phil Schwan <phil@clusterfs.com>
4500 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4501 - failed open()s could cause deadlock; fixed (867, 869)
4502 - stop cancelling OST locks when files are closed (481)
4503 - overlapping XID spaces caused network corruption (851, 853)
4504 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4505 - setattr_raw added, to do single-RPC, server-side setattrs
4506 - lmc/lconf syntax change for OST UUIDs
4507 - fix crashy race condition between ptlrpc_free_req and osc_close
4508 - don't use request in mdc_enqueue if we hit a timeout (889)
4509 - don't set the inode i_size for regular files from the MDS (896)
4510 - handle out of order completion AST (842)
4511 - don't LBUG if a lock request times out after receiving AST (913)
4512 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4513 - if a bad lock AST arrives, send an error instead of dropping entirely
4514 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4515 - fix leak in bulk IO when only partially completed (899, 900, 926)
4516 - fix O_DIRECT for ia64 (55)
4517 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4518 - C-z after timeout could hang a process forever; fixed (977)
4520 - client-side I/O cache (678, 924, 929, 941, 970)
4522 - READPAGE and SETATTRs which don't take server-side locks get
4525 2003-02-11 Phil Schwan <phil@clusterfs.com>
4528 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4529 - Fix fsfilt_extN_readpage() to read a full page of directory
4530 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4531 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4532 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4533 - Most intent operations were removed, in favour of a new RPC mode
4534 that does a single RPC to the server and bypasses most of the VFS
4535 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4536 - Aggressively cancel local locks on DLM servers
4537 - mds_reint_unlink sends EA to the client if it's the last nlink.
4538 client uses that EA to unlink OST objects.
4539 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4540 - recursive symlinks were fixed (439)
4541 - fixed NULL deref in DEBUG_REQ
4542 - filter_update_lastobjid no longer calls sync, which annoyed extN
4543 - fixed multi-client small-writes to a single file problem (445)
4544 - fixed mtime updates during file writes (607)
4545 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4546 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4547 - fixed Linux OST size reporting problem (444, 656)
4548 - OST now updates object mtime with writes or setattr (607, 619)
4549 - client verifies file size before zeroing page past EOF (445)
4550 - OST now writes last allocated objid to disk with allocation (108)
4551 - LOV on echo now works (409)
4553 - mds_reint_unlink sends a new buffer, with the EA included. this
4554 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4555 set if a regular file was being unlinked, and it was the last link
4556 - use PtlGet from the target for bulk writes (315)
4557 - OST now updates object mtime with writes or setattr (607, 619)
4558 - LDLM now has a grant-time callback to revalidate locked items, if
4560 - Many MDS operations were reorganized to combat race conditions
4562 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4563 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4565 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4568 - Fully reactivate OST imports after reconnection (512, others)
4569 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4570 - More graceful error handling for truncating on dead OST (515)
4571 - Don't error out unless we're actually accessing dead stripes (474)
4572 - Fix garbage sizes when stripes are missing (410)
4573 - LRU counters were broken, causing constant lock purge (433, 432)
4574 - garbage on read from stripes with failed OSTs (441)
4575 - mark OSCs as active before reconnecting during recovery (438)
4576 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4577 - lfind did not preserve OST order in output (443)
4578 - symlinks cause hung clients, incorrect data (439)
4579 - stop dereferencing request after dropping refcount (457)
4580 - don't LASSERT(spin_is_locked) on non-SMP (455)
4581 - fixes for many rename() bugs
4582 - fstat didn't correctly synchronize attributes (399)
4583 - server must handle lock cancellation during blocking AST prep (487)
4584 - bulk descriptors were free()d too soon (511)
4585 - fix paths in lconf, which would load incorrect modules (451, 507)
4586 - fix confusing lconf 'host not found' error message (386)
4587 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4588 - fix race condition in mdc_blocking_ast() for inode access (526)
4589 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4590 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4591 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4592 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4593 - fix lstripe in several ways (kernel side) (527)
4594 - fix request leak in ldlm_cli_enqueue (262)
4595 - incorrect OSC was marked inactive after OST failure
4596 - call mds_fs_cleanup before unmounting filesystem (524)
4597 - fix races between taking ns_lock and ldlm_lock_change_resource
4598 - fix races updating LOV export open file list
4599 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4600 - fix recovery NULL deref in ldlm_cli_cancel_unused
4601 - fix some DLM races by using new hash table for lock handles (419)
4602 - permit the client to specify desired inodes, at replay
4603 - duplicate requests when we queue them for replay reintegration
4604 - fix last_rcvd offset calculation
4605 - sync after each recovered transaction, so we always make progress
4606 - never, not always, ERESTART requests without transnos
4607 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4609 - skip replay if the MDS says that the client is already connected
4610 - don't check for a recovery-enabled export to match lctl's UUID
4611 - don't INC_USE_COUNT for phantom exports
4612 - don't crash when cleaning up phantom exports (567)
4613 - don't double-finish or set replay data for errored mdc_open requests
4614 - abort requests when they time out, so we don't get old replies
4615 - send/receive replies for AST messages again
4616 - if the client says that it doesn't have the lock, cancel it on the
4618 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4619 mark it as destroyed, it will all work out when decref is called
4620 - fix module use counts (22, 581)
4622 - ASTs now expect a reply (server cancels lock on error reply)
4624 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4627 - fix many simultaneous client startup (392)
4628 - fix dentry->d_it clobbering
4629 - credentials weren't being shipped for readdir/getattr operations
4630 - remove invalid assertions triggered during some concurrent MD
4632 - proper Lustre versions added (336, 389)
4633 - fix memory leak for create error case (398)
4634 - fix LOV locking bug that would get cli/srv out of sync
4635 - fix echo client over LOV (409)
4636 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4637 - fix double-O_EXCL intent crash (424)
4638 - avoid sending multiple lock CANCELs (352)
4640 - MDS can do multi-client recovery (modulo bugs in new code)
4642 - many updates, edits, cleanups
4644 2002-11-18 Phil Schwan <phil@clusterfs.com>
4647 - fix null d_it dereference (346)
4648 - fix full OST/dbench hang (333)
4649 - fix permission problem with file removal (286)
4650 - fix removal of OSCs from LOV when they fail
4651 - fix NULL deref during bulk timeout (214)
4652 - fix problems related to multiple filesystems on one MDS (241)
4653 - fixed serious subtle metadata locking bugs
4654 - free locks on clients when inodes are removed due to memory
4656 - fix inode pointer in lock data (285)
4657 - partial support for multiple MDS on a single host (241)
4658 - data locks weren't cancelled at clear_inode time (290, 311)
4659 - intent locks could lead to unbounded lock growth (205)
4660 - added a maximum lock count, an LRU list, and a flusher
4661 - fix multiple rename (365)
4662 - properly abstracted the echo client
4663 - OSC locked 1 byte too many; fixed
4664 - rewrote brw callback code:
4665 - fixed recovery bugs related to LOVs (306)
4666 - fixed too-many-pages-in-one-write crash (191)
4667 - fixed (again) crash in sync_io_timeout (214)
4668 - probably fixed callback-related race (385)
4670 - Add capability to MDS protocol
4671 - LDLM cancellations and callbacks on different portals
4673 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4676 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4677 - defer open object destruction to close time (601981, 138)
4678 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4679 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4680 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4681 - don't call obd_setattr in ll_file_release for destroyed objects
4682 * protocol change to lustre_msg: move |version| and add |flags|
4683 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4684 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4685 * added replay of create, unlink, link and rename operations during
4686 MDS failover; recovery should be much more robust now
4687 * remove failed OSCs from LOVs (only lov_create uses this so far)
4688 * the lustre-HOWTO was brought (more) up to date (582544)
4690 2002-10-23 Phil Schwan <phil@clusterfs.com>
4693 - in-use dentries weren't being reused properly (617851)
4694 - prevent multiple LDLM setup (599178)
4695 - fix LOV size calculations for truncate (617853)
4696 - fix client handling of MDS intent errors (POSIX)
4697 - fix permission bug in lovstripe.c test (624321)
4698 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4699 - truncate past end of file could corrupt data
4700 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4701 - a race in recovery could return ETIMEDOUT to apps (623947)
4702 - building outside the source directory was fixed
4703 * the lustre-HOWTO was brought (more) up to date (582544)
4704 * major progress was made on recovery functionality
4706 2002-10-10 Phil Schwan <phil@clusterfs.com>
4709 - recovery deadlock fix
4710 - rm -rf causes LBUG fix (617817)
4711 - file open by multiple tasks fix (618962)
4712 - directory permissions bugs (602707 and 620007)
4713 - journal_stop fixed with locking (611313)
4714 - O_APPEND failures resolved (618273, perhaps 614459)
4715 - lconf PATH fix (619770)
4716 - IA64 build fix (621450)
4717 - RPC buffer sizes scale with amount of memory
4719 2002-10-01 Phil Schwan <phil@clusterfs.com>
4722 - locks would be cancelled without throwing away data pages,
4723 resulting in inconsistent data (605627)
4724 - inode attributes were not always being refreshed (605627, 612449)
4725 - lconf now continues to cleanup after lctl reports an error
4726 - MDS now enforces user permissions (602707)
4727 - lprocfs cleanup fixed, but not yet enabled (614157)
4728 - fixed infinite server hang, should a client not respond to an AST
4729 - avoid going into recovery if user calls readlink() with a buffer
4730 that's too small (613941)
4731 - AST RPCs no longer require replies (614867) -- this may be changed
4732 - don't crash server if client sends an IOV that's too big (611336)
4733 - fixed lock conversion deadlock (611892)
4734 - fixed the following of symlinks (614622)
4735 * recovery: the server can remove locks from a client that dies, other
4736 clients can make progress
4737 * more extN patch fixes
4738 * compile-time configurable ptlrpc buffer allocations
4740 - collaborative read cache document
4741 - Lustre Lite Performance CDR document-in-progress
4743 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4746 - fix typo in patch-2.4.18
4748 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4751 - clear ptlrpc request each time in handle_incoming_request()
4752 - unlink of files now destroys the object on the OST
4754 2002-09-19 Peter Braam <braam@clusterfs.com>
4756 * add hard link support
4757 * change obdfile creation method
4758 * kernel patch changed
4760 2002-09-19 Peter Braam <braam@clusterfs.com>
4763 - stack overflow bug in extN fixed
4765 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4767 * documentation updates
4768 - add man pages for config tools
4769 - update tests/README to describe testing with new config tools
4770 - finish metadata API descriptions
4771 * bug fixes and cleanups
4772 - statfs workaround for 16TB limit
4773 - LOV stripe allocation improved, can stripe on subset of OSTs
4774 - LOV file size/IO offset was wrong for files > 4GB in size
4775 - object EA data was being dropped, caused files to be unreadable
4776 - memory overflow with non-LOV OST caused memory corruption
4777 - fixed regression tests to work with new config tools, obdfilter
4778 - fixed bug when directory size became larger than 1 block
4779 - fixed bug (for single client case) when PWD was deleted
4780 - invalidate local directory pages when doing intent-based ops
4781 - avoid LDLM oops when lock callback contained bad data
4783 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4785 * documentation updates
4786 * bug fixes and cleanups
4787 - configuration tools
4790 - 64-bit compile warnings
4791 - 64-bit internal statfs data
4793 * test_brw on persistent OST devices
4795 * lprocfs (disabled)
4797 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4799 * documentation updates
4800 * bug fixes and cleanups
4801 * configuration tools
4803 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4806 * many small fixes to 0_5_4
4807 * io/network handling
4808 * thinkos in MDS operations
4810 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4813 * crucial basic fixes to 0.5.3
4814 * IOR, Iozone work over Elan
4817 2002-08-07 Phil Schwan <phil@clusterfs.com>
4818 * version 0_5_3, our first alpha
4819 * we use the new Portals iovs
4820 * documentation updates
4821 * bug fixes and cleanups
4822 * small changes in the DLM wire protocol
4824 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4825 * version 0_5_1 with some initial stability,
4826 * locking on MD and file I/O.
4827 * documentation updates
4828 * several bug fixes since 0.5.0
4829 * small changes in wire protocol
4831 2002-07-18 Phil Schwan <phil@clusterfs.com>
4833 * delivered as Lustre Light Alpha
4834 * fixed a crash after handling invalid MDS requests
4835 * fixed directory pages for architectures with non-4k pages sizes
4837 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4838 * release version v0_4_4
4839 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4841 * Fixes a number of interruption problems with OST operations.
4842 * Update documentation for portals header changes
4843 * Move all wire protocol structs/defines to lustre_idl.h
4844 * Fixes symlink length bug.
4845 * Add tcpdump to repository.
4847 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4848 * release version v0_4_3
4849 * Fixes statfs for inodes on extN.
4850 * Fixes bug in runtests which would delete /etc/hosts.
4851 * Use 64-bit object IDs wherever possible (not into VFS though)
4852 Remove ost_get_info, which is unused by lustre, and out of date.
4854 2002-07-03 Peter Braam <braam@clusterfs.com>
4855 * release version v0_4_2 Fixes a lookup error (type not passed)
4856 * move forward to head of Portals
4857 * move forward to latest Lustre kernel
4859 2002-06-25 Peter Braam <braam@clusterfs.com>
4860 * release version v0_4_1. Hopefully stable on single node use.