1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.21.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.7-sun3
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
15 Severity : enhancement
17 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
21 Description: do not process already freed flock
22 Details : flock can possibly be freed by another thread before it reaches
23 to ldlm_flock_completion_ast.
27 Description: LBUG during stress test
28 Details : Need properly lock accesses the flock deadlock detection list.
32 Description: oops in page fault handler
33 Details : kernel page fault handler can return two special 'pages' in error case, don't
34 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
38 Description: timeout with invalidate import.
39 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
40 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
41 ptlrpc_invalidate_import and as result LASSERT.
43 Severity : enhancement
45 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
49 Description: enable MGS and MDT services start separately
50 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
51 a co-located MGS without starting the MGS, which is a complement
52 to 'nosvc' mount option.
56 Frequency : after recovery
57 Description: precreate to many object's after del orphan.
58 Details : del orphan st in oscc last_id == next_id and this triger growing
59 count of precreated objects. Set flag LOW to skip increase count
60 of precreated objects.
64 Frequency : rare, on clear nid stats
65 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
66 Details : when clean nid stats sometimes try destroy live entry,
67 and this produce panic in free.
71 Description: Stack overflow during MDS log replay
72 ease stack pressure by using a thread dealing llog_process.
76 Description: wait until IO finished before start new when do lock cancel.
77 Details : VM protocol want old IO finished before start new, in this case
78 need wait until PG_writeback is cleared until check dirty flag and
79 call writepages in lock cancel callback.
81 Severity : enhancement
83 Description: using special macro for print time and cleanup in includes.
87 Description: mds_mfd_close() ASSERTION(rc == 0)
88 Details : In mds_mfd_close(), we need protect inode's writecount change
89 within its orphan write semaphore to prevent possible races.
93 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
97 Frequency : rare, on shutdown ost
98 Description: don't hit live lock with umount ost.
99 Details : shrink_dcache_parent can be in long loop with destroy dentries,
100 use shrink_dcache_sb instead.
104 Description: don't panic with use echo client
105 Details : echo client pass NULL as client nid pointer and this produce null
110 Description: fix build on ppc32
111 Details : compile code with -m64 flag produce wrong object file for ppc32.
115 Description: add message levels for liblustreapi
119 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
120 Details : Move the 'good_osts' check before the 'total_bavail' check. This
121 will result in an -EAGAIN and in the exit call path we call
122 alloc_rr() which will with increasing aggressiveness attempt to
123 aquire precreated objects on the minimum number of required OSCs.
127 Description: Use old size assignment to avoid deadlock
128 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
129 the scheduling while holding a spinlock. We do not need locking
130 for size in ll_update_inode() because size is only updated from
131 the MDS for directories or files without objects, so there is no
132 other place to do the update, and concurrent access to such inodes
133 are protected by the inode lock.
137 Description: resolve "_IOWR redefined" build error on SLES10
141 Description: dump the memory debugging after all modules are unloaded to
142 suppress false negative in conf_sanity test 39
144 Severity : enhancement
146 Description: build kernel-ib packages for OFED 1.3 in our release cycle
151 Description: fix SLES kernel versioning
152 Details : the kernel version for our SLES 10 kernel did not include a "-"
153 before the "smp" at the end. while this was not a problem in
154 general, it did mean that software trying to use the kernel
155 version to try to detect a vendor specific kernel would fail.
156 this was most evident by the OFED build scripts.
160 Description: Don't update lov_desc members until making sure they are valid
161 Details : When updating lov_desc members via proc fs, need fix their
162 validities before doing the real update.
166 Description: don't put request into delay list while invalidate in flight.
167 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
168 import in flight. this produce timeout for invalidate and sometimes
169 can cause stale data.
173 Frequency : on ppc only
174 Description: not convert ost objects for directory because it's not exist.
175 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
179 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
181 Severity : enhancement
183 Description: Update to support 2.6.22.14 vanilla kernel.
187 Frequency : rare, on recovery
188 Description: read procfs can produce deadlock in some situation
189 Details : Holding lprocfs lock which send rpc can produce block for destroy
190 obd objects and this also block reconnect with -EALREADY. This isn't
191 fix all lprocfs bugs - but make it rare.
193 Severity : enhancement
195 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
198 Frequency : frequent on X2 node
200 Description: mdc_set_open_replay_data LBUG
201 Details : Set replay data for requests that are eligible for replay.
205 Description: lustre_mgs: operation 101 on unconnected MGS
206 Details : When MGC is disconnected from MGS long enough, MGS will evict the
207 MGC, and late on MGC cannot successfully connect to MGS and a lot
208 of the error messages complaining that MGS is not connected.
212 Frequency : on network error
213 Description: panic with double free request if network error
214 Details : mdc_finish_enqueue is finish request if any network error ocuring,
215 but it's true only for synchronus enqueue, for async enqueue
216 (via ptlrpcd) this incorrect and ptlrpcd want finish request
219 Severity : enhancement
221 Description: client-side metadata stat-ahead during readdir(directory readahead)
222 Details : perform client-side metadata stat-ahead when the client detects
223 readdir and sequential stat of dir entries therein
226 Frequency : on start mds
228 Description: Implement get_info(last_id) in obdfilter.
231 Frequency : occasional
233 Description: Correctly check stale fid, not start epoch if ost not support SOM
234 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
235 overwrited with new generated fid, but mds can anwer with one of these
236 two fids and both is not stale. setattr incorectly start epoch and
237 assume will be called done_writeting, but without SOM done_writing
241 Frequency : rare, depends on device drivers and load
243 Description: MDS or OSS nodes crash due to stack overflow
244 Details : Code changes in 1.8.0 increased the stack usage of some functions.
245 In some cases, in conjunction with device drivers that use a lot
246 of stack the MDS (or possibly OSS) service threads could overflow
247 the stack. One change which was identified to consume additional
248 stack has been reworked to avoid the extra stack usage.
251 Frequency : occasional
253 Description: Do not fail import if osc_interpret_create gets -EAGAIN
254 Details : If osc_interpret_create got -EAGAIN it immediately exits and
255 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
256 oscc_has_objects and see OSC has no objests and call
257 oscc_internal_create to resend create request.
259 Severity : enhancement
261 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
263 Severity : enhancement
265 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
268 Frequency : very rare
270 Description: Fix replay if there is an un-replied request and open
271 Details : In some cases, older replay request will revert the
272 mcd->mcd_last_xid on MDS which is used to record the client's
275 Severity : enhancement
277 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
279 Severity : enhancement
281 Description: Add rhel5 support to HEAD.
283 Serverity : enhancement
285 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
290 Description: Don't allow skipping OSTs if index has been specified.
291 Details : Don't allow skipping OSTs if index has been specified, make locking
292 in internal create lots better.
296 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
297 Details : access to bitfield in structure is always rounded to long
298 and this produce problem with not atomic change any bit.
302 Description: Lustre make rpms failed.
303 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
304 builds Lustre rpms from tarball.
307 Frequency : rare on shutdown ost
309 Description: If llog cancel was not send before clean_exports phase, this can
310 produce deadlock in llog code.
311 Details : If llog thread has last reference to obd and call class_import_put
312 this produce deadlock because llog_cleanup_commit_master wait when
313 last llog_commit_thread exited, but this never success because was
314 called from llog_commit_thread.
318 Description: allow userland application know is lost one of stripes.
319 Details : fill lvb_blocks with error code on ost and return it to
320 application if error flag found.
324 Description: NULL lov_tgts causing MDS oops
325 Details : more safe checks for NULL lov_tgts for avoid oops.
327 Severity : enhancement
329 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
333 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
334 Details : If an OST is down the MDS will hang indefinitely in
335 obd_statfs_rqset() waiting for the statfs data. While for
336 MDS QOS usage of statfs, it should not stuck in waiting.
338 Severity : enhancement
340 Description: remote_acl support
341 Details : Support ACL-based permission check for remote user.
342 Support setfacl/getfacl for remote user with the utils
343 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
344 the system "{s,g}etfacl" utils.
346 Severity : enhancement
348 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
350 Severity : enhancement
352 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
356 Description: make mgs_setparam() handle fsname containing dash
357 Details : fsname containing a dash does not work with lctl conf_param
359 Severity : enhancement
361 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
363 Severity : enhancement
365 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
369 Frequency : rare, at shutdown
370 Description: access already free / zero obd_namespace.
371 Details : if client_disconnect_export was called without force flag set,
372 and exist connect request in flight, this can produce access to
373 NULL pointer (or already free pointer) when connect_interpret
374 store ocd flags in obd_namespace.
378 Frequency : only at startup
379 Description: not alloc memory with spinlock held.
380 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
381 if any spinlock held.
383 Severity : enhancement
385 Description: make lustre randomly fail allocating memory
386 Details : Make lustre randomly failed allocating memory for testing purpose.
388 Severity : enhancement
390 Description: lost problems with lov objid file
391 Details : Fixes some scability and access to not inited memory problems
392 in work with lov objdid file.
397 Description: lfs find does not continue on file error
398 Details : Continue other files processing when a file/dir is absent.
402 Description: Inconsistent usage of lustre_pack_reply()
403 Details : Standardize the usage of lustre_pack_reply() such that it
404 always generate a CERROR on failure.
407 Frequency : occasional
409 Description: MDS hang or stay in waiting lock
410 Details : If client receive lock with CBPENDING flag ldlm need send lock
411 cancel as separate rpc, to avoid situation when cancel request
412 can't processed due all i/o threads stay in wait lock.
416 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
420 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
424 Description: Update to SLES9 kernel-2.6.5-7.287.3.
428 Description: Update to RHEL4 latest kernel.
430 Severity : enhancement
432 Description: Build SLES10 patchless client fails
433 Details : The configure was broken by run ./configure with
434 --with-linux-obj=.... argument for patchless client. When the
435 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
436 can't search header adequately. Use absolute path such as
437 -I($LINUX)/include instead.
441 Description: interrupt oig_wait produce painc on resend.
442 Details : brw_redo_request can be used for resend requests from ptlrpcd and
443 private set, and this produce situation when rq_ptlrpcd_data not
444 copyed to new allocated request and triggered LBUG on assert
445 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
446 ptlrpcd set if request is changed and can be safety changed to use
451 Description: LASSERT_{REQ,REP}SWAB macros are buggy
452 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
453 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
454 these macros and replace them with their difinitions instead.
458 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
459 Details : Update which_patch & target file for SLES10 latest kernel.
462 Description: Remove client patches from SLES 10 kernel.
463 Details : This causes SLES 10 clients to behave as patchless clients
464 even on a Lustre-patched (server) kernel.
466 Severity : enhancement
468 Description: self-adjustable client's lru lists
469 Details : use adaptive algorithm for managing client cached locks lru
470 lists according to current server load, other client's work
471 pattern, memory activities, etc. Both, server and client
472 side namespaces provide number of proc tunables for controlling
475 Severity : enhancement
477 Description: light-weight GSS support
478 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
479 and reduce performance overhead.
481 Severity : enhancement
483 Description: Linux keyring support
484 Details : Support using service of Linux keyring for Lustre GSS internal
485 context refresh/cache mechanism.
489 Description: Fix errors in lfs documentation
490 Details : Fixes man pages
494 Description: don't use GFP_* in generic Lustre code.
495 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
499 Description: obdclass is limited by single OBD_ALLOC(idarray)
500 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
504 Description: warnings with build patchless client with vanila 2.6.19 and up
505 Details : change old ctl_table style and replace ctl_table/ctl_table_header
506 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
510 Frequency : Only for SLES
511 Description: sanity_quota fail test_1
512 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
513 file, which makes TSTID[2] inunique.
517 Frequency : Only for SLES9
518 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
519 Details : Update target/ChangeLog/which_patch .
523 Frequency : Only for test_10 in sanity-quota.sh
524 Description: a bug in quota test code
525 Details : checking a test flag in a wrong place
529 Frequency : Only for test_13 in sanity-quota.sh
530 Description: a bug in quota test script
531 Details : shouldn't check overfull allocated quota
535 Description: jbd statistics
536 Details : Port older jbd statistics patch for sles10
538 Severity : enhancement
540 Description: Add printing inode info into message about error in writepage.
545 Description: reply_lock_interpret crash due to race with it and lock cancel.
546 Details : Do not replay locks that are being cancelled. Do not reference
547 locks by their address during replay, just by their handle.
551 Frequency : When flocks are used.
552 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
553 Details : Flock locks might destroy just granted lock if it could be merged
554 with another existing flock, this is done in completion handler,
555 so teach ldlm_cli_enquque_fini that this is a valid case for
561 Description: Oops in read and write path when failing to allocate lock.
562 Details : Check if lock allocation failed and return error back.
566 Description: lstripe command fails for valid OST index
567 Details : The stripe offset is compared to lov->desc.ld_tgt_count
568 instead of lov->desc.ld_active_tgt_count.
572 Description: sanity.sh failed test 103
573 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
578 Description: df doesn't work properly if diskfs blocksize != 4K
579 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
583 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
584 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
588 Description: obd_health_check_timeout too short
589 Details : set obd_health_check_timeout as 1.5x of obd_timeout
593 Description: llapi_file_create() does not allow some changes
594 Details : add llapi_file_open() that allows specifying the mode and
595 open flags, and also returns an open file handle.
599 Description: merge and cleanup kernel patches.
600 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
604 Description: Add journal checksum support.(Kernel part)
605 Details : The journal checksum feature adds two new flags i.e
606 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
607 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
608 indicates that the commit block contains the checksum for
609 the blocks described by the descriptor blocks. Now commit
610 record can be sent to disk without waiting for descriptor
611 blocks to be written to disk. This behavior is controlled
612 using JBD2_FEATURE_ASYNC_COMMIT flag.
616 Description: OSS needs mutliple precreate threads
617 Details : Add ability to start more than one create thread per OSS.
621 Description: Sanity.sh test_65k failed: setstripe should have succeeded
622 Details : execute lfs setstripe on client
626 Description: mds_obd_create error creating tmp object
627 Details : When the user sets quota on root, llog will be affected and can't
628 create files and write files.
631 Frequency : Always on ia64 patchless client, and possibly others.
633 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
634 Details : This allows the patchless client to be loaded on architectures
639 Description: RedHat Update kernel for RHEL5
640 Details : Add the kernel config file for RHEL5.
644 Description: RedHat Update kernel for RHEL5
645 Details : Modify the kernel config file more closer RHEL5.
649 Description: Build failure against Centos5 (RHEL5)
650 Details : Define PAGE_SIZE when it isn't present.
654 Description: test_7 on sanity-quota.sh can't be used on separate machines
655 Details : add facet to handle it
659 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
660 Details : with NFS, the anon dentry's parent was set to itself in
661 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
662 corresponding dentry other than use rec->ur_name.
664 Severity : enhancement
666 Description: lfs setstripe enhancement
667 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
671 Description: enable data checksumming by default
672 Details : enable checksum by default, allow --disable-checksum
673 configure option and "-o nochecksum" mount option. Checksums
674 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
678 Description: lustre support for RHEL5
679 Details : Add support for RHEL5.
683 Description: Client eviction due to failover config
684 Details : after a connection loss, the lustre client should attempt to
685 reconnect to the last active server first before trying the
686 other potential connections.
690 Description: when mds and osts use different quota unit(32bit and 64bit),
691 quota will be released repeatly.
692 Details : void sending multiple quota reqs to mds, which will keep the
693 status between the reqs.
697 Description: osts not allocated evenly to files
698 Details : change the condition to increase offset_idx
701 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
703 Description: Data corruption for OSTs that are formatted larger than 2TB
705 Details : When generating the bio request for lustre file writes the
706 sector number would overflow a temporary variable before being
707 used for the IO. The data reads correctly from Lustre (which
708 will overflow in a similar manner) but other file data or
709 filesystem metadata may be corrupted in some cases.
713 Description: Tune the kernel for good SCSI performance.
714 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
715 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
720 Description: rewrite ext2-derived code in obdclass/uuid.c
721 Details : rewrite inherited code (uuid parsing code from ext2 utils)
722 from scratch preserving functionality.
726 Description: "lfs find -obd UUID" prints directories
727 Details : "lfs find -obd UUID" will return all directory names instead
728 of just file names. It is incorrect because the directories
729 do not reside on the OSTs.
733 Description: use i_size_read and i_size_write in 2.6 port
734 Details : replace inode->i_size access with i_size_read/write()
737 Frequency : when removing large files
739 Description: scheduling issue during removal of large Lustre files
740 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
741 It causes scheduling issues when removing large files (17TB in the
745 Frequency : only with liblustre clients on XT3
747 Description: evictions taking too long
748 Details : allow llrd to evict clients directly on OSTs
751 Frequency : only on ppc
753 Description: /proc/fs/lustre/devices broken on ppc
754 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
755 We should make sure the type of 'index' is loff_t and then cast
756 explicitly as needed below. Do not assign an explicitly cast
760 Frequency : only for rhel5
762 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
763 Details : Modify the target file & which_kernel.
765 Severity : enhancement
767 Description: omit set fsid for export NFS
768 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
769 as FS_REQUIRES_DEV for avoid problems with generate fsid.
773 Frequency : Always, for kernels after 2.6.16
774 Description: Fix warning idr_remove called for id=.. which is not allocated.
775 Details : Last kernels save old s_dev before kill super and not allow
776 to restore from callback - restore it before call kill_anon_super.
780 Description: buffer overruns could theoretically occur
781 Details : llapi_semantic_traverse() modifies the "path" argument by
782 appending values to the end of the origin string, and a
783 overrun may occur. Adding buffer overrun check in liblustreapi.
787 Description: Fix error on 'ls .' at the top of the Lustre mount.
788 Details : Don't revalidate dentry if it is a root dentry.
792 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
793 Details : Modify vm-tunables-rhel4.patch.
797 Description: Kernel config for 2.6.18-vanilla.
798 Details : Modify targets/2.6-vanilla.target.in.
799 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
800 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
801 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
802 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
806 Description: improve handling recoverable errors
807 Details : if request processig with error which can be recoverable on server
808 request should be resend, otherwise page released from cache and
813 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
814 Details : Modify target file & which_patch.
815 A flaw was found in the IA32 system call emulation provided
816 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
817 value could be stored in the %RAX register, which could trigger an
818 out-of-bounds system call table access. An untrusted local user
819 could exploit this flaw to run code in the kernel
820 (ie a root privilege escalation). (CVE-2007-4573).
824 Description: O_DIRECT bypasses client statistics.
825 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
826 read_ahead_stats were updated. copied the stats section from
827 osc_send_oap_rpc() into async_internal().
831 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
835 Description: change order of libsysio includes
836 Details : '#include sysio.h' should always come before '#include xtio.h'
840 Description: handle "serious error: objid * already exists" more gracefully
841 Details : If LAST_ID value on disk is smaller than the objects existing in
842 the O/0/d* directories, it indicates disk corruption and causes an
843 LBUG(). If the object is 0-length, then we should use the existing
844 object. This will help to avoid a full fsck in most cases.
846 Severity : enhancement
848 Description: adapt the lustre_config script to support the upgrade case
849 Details : Add "-u" option for lustre_config script to support upgrading 1.4
850 server targets to 1.6 in parallel.
854 Description: To avoid grant space > avaible space when the disk is almost
855 full. Without this patch you might see the error "grant XXXX >
856 available" or some LBUG about grant, when the disk is almost
858 Details : In filter_check_grant, for non_grant cache write, we should
859 check the left space by if (*left > ungranted + bytes), instead
860 of (*left > ungranted), because only we are sure the left space
861 is enough for another "bytes", then the ungrant space should be
862 increase. In client, we should update cl_avail_grant only there
863 is OBD_MD_FLGRANT in the reply.
867 Description: Update RHEL 4 kernel to fix local root privilege escalation.
868 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
869 described in CVE-2007-4573. This problem could allow untrusted
870 local users to gain root access.
873 Frequency : when using O_DIRECT and quotas
875 Description: Incorrect file ownership on O_DIRECT output files
876 Details : block usage reported by 'lfs quota' does not take into account
877 files that have been written with O_DIRECT.
882 Description: touch file failed when fs is not full
883 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
884 otherwise we can get ENOSP while fs is not full.
888 Description: parallel lock callbacks
889 Details : Instead of sending blocking and completion callbacks as separated
890 requests, adding them to a set and sending in parallel.
893 Frequency : only for Cray XT3
894 Bugzilla : 12829/13455
895 Description: Changing primary group doesn't change the group lustre assigns to
897 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
898 overridden with the primary group provided by the group upcall,
899 whereas the supplied fsgid can be trusted if it is in the list of
900 supplementary groups returned by the group upcall.
902 Severity : enhancement
904 Description: Allow masking D_WARNING, D_ERROR messages from console
905 Details : Console messages can now be disabled via lnet.printk.
909 Description: User code with malformed file open parameter crashes client node
910 Details : Before packing join_file req, all the related reference should be
911 checked carefully in case some malformed flags cause fake
912 join_file req on client.
916 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
917 Details : ldlm_completion_ast() assumes that a lock is granted when the req
918 mode is equal to the granted mode. However, it should also check
919 that LDLM_FL_CP_REQD is not set.
923 Description: Heavy nfs access might result in deadlocks
924 Details : After ELC code landed, it is now improper to enqueue any mds
925 locks under och_sem, because enqueue might want to decide to
926 cancel open locks for same inode we are holding och_sem for.
930 Description: Client eviction while running blogbench
931 Details : A lot of unlink operations with concurrent I/O can lead to a
932 deadlock causing evictions. To address the problem, the number of
933 oustanding OST_DESTROY requests is now throttled to
934 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
935 callbacks are processed in priority.
939 Description: enable ACLs on MDS by default
940 Details : ACLs must be enabled on MDS by default.
943 Frequency : PPC/PPC64 only
945 Description: conflicts between asm-ppc64/types.h and lustre_types.h
946 Details : fix duplicated definitions between asm-ppc64/types.h and
947 lustre_types.h on PPC.
950 Frequency : PPC/PPC64 only
952 Description: asm-ppc/segment.h does not exist
953 Details : fix compile issue on PPC.
957 Description: better handle error messages in extents code
960 Frequency : RHEL4 only
962 Description: mkfs is very slow on IA64/RHEL4
963 Details : A performance regression has been discovered in the MPT Fusion
964 driver between versions 3.02.73rh and 3.02.99.00rh. As a
965 consequence, we have downgraded the MPT Fusion driver in the RHEL4
966 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
968 Severity : enhancement
970 Description: SNMP support enhancement
971 Details : Adding total number of sampled request for an MDS node in snmp
974 Severity : enhancement
976 Description: Optimize ldlm waiting list processing for PR extent locks
977 Details : When processing waiting list for read extent lock and meeting read
978 lock that is same or wider to it that is not contended, skip
979 processing rest of the list and immediatelly return current
980 status of conflictness, since we are guaranteed there are no
981 conflicting locks in the rest of the list.
985 Description: Time out and refuse to reconnect
986 Details : When the failover node is the primary node, it is possible
987 to have two identical connections in imp_conn_list. We must
988 compare not conn's pointers but NIDs, otherwise we can defeat
989 connection throttling.
993 Description: port llog fixes from b1_6 into HEAD
994 Details : Port llog reference couting and some llog cleanups from b1_6
995 (bug 10800) into HEAD, for protect from panic and access to already
996 free llog structures.
1000 Description: Detect stride IO mode in read-ahead
1001 Details : When a client does stride read, read-ahead should detect that and
1002 read-ahead pages according to the detected stride pattern.
1006 Description: data checksumming impacts single node performance
1007 Details : add support for several checksum algorithm. Currently, only CRC32
1008 and Adler-32 are supported. The checksum type can be changed on
1009 the fly via /proc/fs/lustre/osc/*/checksum_type.
1013 Description: use adler32 for page checksums
1014 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1019 Description: build for x2 fails
1020 Details : fix compile issue on Cray systems.
1024 Description: Properly match for duplicate locks
1025 Details : Due to different lock order from skiplists code, we need to
1026 traverse entire list for now
1029 Frequency : only on PPC/SLES10
1031 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1032 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1033 be defined. Add a hack in mkfs_lustre.c to work around this compile
1038 Description: LASSERT on MDS when client holding flock lock dies
1039 Details : ldlm pool logic depends on number of granted locks equal to
1040 number of released locks which is not true for flock locks, so
1041 just exclude such locks from consideration.
1045 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1046 Details : Use fsfilt_sync() for both the whole filesystem sync and
1047 individual file sync to eliminate dangerous inode locking
1048 with I_LOCK that can lead to a deadlock.
1052 Description: performance in 1.6.3
1053 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1054 sure that 1MB requests are not fragmented by the block layer.
1056 Severity : enhancement
1058 Description: organize the server-side client stats on per-nid basis
1059 Details : Change the structure of stats under obdfilter and mds to
1067 The "uuid"s file would list the uuids of _active_ exports.
1068 And the clear entry is to clear all stats and stale nids.
1070 Severity : enhancement
1072 Description: eliminate client locks in face of contention
1073 Details : file contention detection and lockless i/o implementation
1074 for contended files.
1078 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1081 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1083 Description: MDS may lose file striping (and hence file data) in some cases
1084 Details : If there are additional extended attributes stored on the MDS,
1085 in particular ACLs, SELinux, or user attributes (if user_xattr
1086 is specified for the client mount options) then there is a risk
1087 of attribute loss. Additionally, the Lustre file striping
1088 needs to be larger than default (e.g. striped over all OSTs),
1089 and an additional attribute must be stored initially in the
1090 inode and then increase in size enough to be moved to the
1091 external attribute block (e.g. ACL growing in size) for file
1096 Description: skiplist implementation simplification
1097 Details : skiplists are used to group compatible locks on granted list
1098 that was implemented as tracking first and last lock of each lock group
1099 the patch changes that to using doubly linked lists
1103 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1104 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1105 because LASSERT on both the data supplied by a client, and the data
1106 on disk is dangerous and incorrect.
1108 Severity : enhancement
1110 Description: Slow trucate/writes to huge files at high offsets.
1111 Details : Directly associate cached pages to lock that protect those pages,
1112 this allows us to quickly find what pages to write and remove
1113 once lock callback is received.
1115 --------------------------------------------------------------------------------
1117 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1119 * Support for kernels:
1120 2.6.5-7.283 (SLES 9),
1121 2.6.9-55.EL (RHEL 4),
1122 2.6.16.46-0.14 (SLES 10),
1123 2.6.18.8 vanilla (kernel.org)
1124 * Client support for unpatched kernels:
1125 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1126 2.6.16 - 2.6.22 vanilla (kernel.org)
1127 * Due to recently discovered recovery problems, we do not recommend
1128 using patchless RHEL 4 clients with this or any earlier release.
1129 * Recommended e2fsprogs version: 1.39.cfs8
1130 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1131 * Starting with this release, the ldiskfs backing filesystem required
1132 by Lustre is now in its own package, lustre-ldiskfs. This package
1133 should be installed. It is versioned separately from Lustre and
1134 may be released separately in future.
1138 Description: block reactivating mgc import until all deactivates complete
1139 Details : Fix race when failing back MDT/MGS to itself (testing)
1142 Frequency : at statup only
1144 Description: mds_lov_synchronize race leads to various problems
1145 Details : simultaneous MDT->OST connections at startup can cause the
1146 sync to abort, leaving the OSC in a bad state.
1148 Severity : enhancement
1150 Description: add optional extra BUILD_VERSION info
1151 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1152 to override the lustre version.
1155 Frequency : 2.6.18 servers only
1157 Description: ll_kern_mount() doesn't release the module reference
1158 Details : The ldiskfs module reference count never drops down to 0
1159 because ll_kern_mount() doesn't release the module reference.
1164 Description: server LBUG when using old ost_num_threads parameter
1165 Details : Accept the old ost_num_threads parameter but warn that it
1166 is deprecated, and fix an off-by-one error that caused an LBUG.
1171 Description: Transient SCSI error results in persistent IO issue
1172 Details : iobuf->dr_error is not reinitialized to 0 between two
1176 Frequency : sometimes when underlying device returns I/O errors
1178 Description: OSTs not going read-only during write failures
1179 Details : OSTs are not remounted read-only when the journal commit threads
1180 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1181 instead of the ext3 wrappers.
1185 Description: poor connect scaling with increasing client count
1186 Details : Don't run filter_grant_sanity_check for more than 100 exports
1187 to improve scaling for large numbers of clients.
1190 Frequency : SLES10 only
1192 Description: sanity-quota.sh quotacheck failed: rc = -22
1193 Details : Quotas cannot be enabled on SLES10.
1196 Frequency : liblustre clients only
1198 Description: getdirentries does not give error when run on compute nodes
1199 Details : getdirentries does not fail when the size specified as an argument
1200 is too small to contain at least one entry
1202 Severity : enhancement
1204 Description: Add LNET router traceability for debug purposes
1205 Details : If a checksum failure occurs with a router as part of the
1206 IO path, the NID of the last router that forwarded the bulk data
1207 is printed so it can be identified.
1212 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1213 Details : Due to a race condition, liblustre clients were occasionally
1214 evicted incorrectly.
1216 Severity : enhancement
1218 Description: lfs setstripe use optional parameters instead of postional
1221 Severity : enhancement
1223 Description: Nanosecond timestamp support for ldiskfs
1224 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1225 resolution timestamps. There is not yet support for this at
1226 the Lustre filesystem level.
1229 Frequency : during server recovery
1231 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1232 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1236 Frequency : nfs export on patchless client
1238 Description: connectathon hang when test nfs export over patchless client
1239 Details : Disconnected dentry cannot be found with lookup, so we do not need
1240 to unhash it or make it invalid
1243 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1244 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1245 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1250 Description: open req refcounting wrong on reconnect
1251 Details : If reconnect happened between getting open reply from server and
1252 call to mdc_set_replay_data in ll_file_open, we will schedule
1253 replay for unreferenced request that we are about to free.
1254 Subsequent close will crash in variety of ways.
1255 Check that request is still eligible for replay in
1256 mdc_set_replay_data().
1261 Description: disable writes to filesystem when reading health_check file
1262 Details : the default for reading the health_check proc file has changed
1263 to NOT do a journal transaction and write to disk, because this
1264 can cause reads of the /proc file to hang and block HA state
1265 checking on a healthy but otherwise heavily loaded system. It
1266 is possible to return to the previous behaviour during configure
1267 with --enable-health-write.
1269 Severity : enhancement
1271 Description: 64-bit inode version
1272 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1273 to the inode. This will be required for version-based recovery.
1278 Description: MDS fails to start if a duplicate client export is detected
1279 Details : in some rare cases it was possible for a client to connect to
1280 an MDS multiple times. Upon recovery the MDS would detect this
1281 and fail during startup. Handle this more gracefully.
1283 Severity : enhancement
1285 Description: Add -o localflock option to simulate old noflock
1287 Details : This will achieve local-only flock/fcntl locks
1293 Description: log_commit_thread vs filter_destroy race leads to crash
1294 Details : Take import reference before releasing llog record semaphore
1299 Description: Wrong request locking in request set processing
1300 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1301 imp_delayed_list, in this place should be used imp_lock.
1304 Frequency : when reconnection
1306 Description: Grant Leak when osc reconnect to OST
1307 Details : When osc reconnect ost, OST(filter) should check whether it
1308 should grant more space to client by comparing fed_grant and
1309 cl_avail_grant, and return the granted space to client instead
1310 of "new granted" space, because client will call osc_init_grant
1311 to update the client grant space info.
1314 Frequency : when client reconnect to OST
1316 Description: Grant Leak when osc do resend and replay bulk write
1317 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1318 bulk write request, because the grant info while be sync between
1319 OSC and OST when reconnect, and we should ignore the grant info
1320 these of resend/replay write req.
1325 Description: Grant space more than avaiable left space sometimes.
1326 Details : When then OST is about to be full, if two bulk writing from
1327 different clients came to OST. Accord the avaliable space of the
1328 OST, the first req should be permitted, and the second one
1329 should be denied by ENOSPC. But if the seconde arrived before
1330 the first one is commited. The OST might wrongly permit second
1331 writing, which will cause grant space > avaiable space.
1334 Frequency : when client is evicted
1336 Description: Grant might be wrongly erased when osc is evicted by OST
1337 Details : when the import is evicted by server, it will fork another
1338 thread ptlrpc_invalidate_import_thread to invalidate the
1339 import, where the grant will be set to 0. While the original
1340 thread will update the grant it got when connecting. So if
1341 the former happened latter, the grant will be wrongly errased
1342 because of this race.
1347 Description: Checking Stale with correct fid
1348 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1349 to check whether it is stale, because sometimes, we want the
1350 enqueue happened anyway, and op_data.fid2 will not be initialized.
1352 Severity : enhancement
1354 Description: update patchless client
1355 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1358 Frequency : only with 2.4 kernel
1360 Description: random memory corruption
1361 Details : size of struct ll_inode_info is to big for union inode.u and this
1362 can be cause of random memory corruption.
1367 Description: Memory leak in recovery
1368 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1369 It should also check obd_fail before fsfilt_start, otherwise if
1370 fsfilt_start return -EROFS,(failover mds during mds recovery).
1371 then the req will return with repmsg->transno = 0 and rc = EROFS.
1372 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1373 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1374 be freed no matter whether fsfilt_commit success or not.
1377 Frequency : only with huge count clients
1379 Description: Prevents from taking the superblock lock in llap_from_page for
1381 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1382 for llap_from_page call in ll_removepage prevents from taking the
1383 superblock lock for a soon died page.
1388 Description: Not check open intent error before release open handle
1389 Details : in some rare cases, the open intent error is not checked before
1390 release open handle, which may cause
1391 ASSERTION(open_req->rq_transno != 0), because it tries to release
1392 the failed open handle.
1397 Description: Set cat log bitmap only after create log success.
1398 Details : in some rare cases, the cat log bitmap is set too early. and it
1399 should be set only after create log success.
1403 Description: Accessing a block bevice can re-enable I/O when Lustre is
1404 tearing down a device.
1405 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1410 Description: service threads may hog cpus when there are a lot of requests
1411 Details : Insert cond_resched to give other threads a chance to use some CPU
1416 Description: the cat log was not initialized in recovery
1417 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1418 unlink log on mds will not be initialized until mds post
1419 recovery. And also in mds post recovery, the unlink log will
1420 initialization will be done asynchronausly, so there will be race
1421 between add unlink log and unlink log initialization.
1425 Description: brw_stats were being printed incorrectly
1426 Details : brw_stats were being printed as log2 but all of them were not
1427 recorded as log2. Also remove some code duplication arising from
1428 filter_tally_{read,write}.
1432 Frequency : rare, only in recovery.
1433 Description: ASSERTION(req->rq_type != LI_POISON) failed
1434 Details : imp_lock should be held while iterating over imp_sending_list for
1435 prevent destroy request after get timeout in ptlrpc_queue_wait.
1439 Description: replay-single.sh test 52 fails
1440 Details : A lock's skiplist need to be cleanup when it being unlinked
1441 from its resource list.
1445 Description: Short directio read returns full requested size rather than
1447 Details : Direct I/O operations should return actual amount of bytes
1448 transferred rather than requested size.
1450 Severity : enhancement
1452 Description: metadata RPC reduction (e.g. for rm performance)
1453 Details : decrease the amount of synchronous RPC between clients and servers
1454 by canceling conflicing lock before the operation on the client
1455 and packing thier handles into the main operation RPC to server.
1457 Severity : enhancement
1459 Description: Async OSC create to avoid the blocking unnecessarily.
1460 Details : If a OST has no remain object, system will block on the creating
1461 when need to create a new object on this OST. Now, ways use
1462 pre-created objects when available, instead of blocking on an
1463 empty osc while others are not empty. If we must block, we block
1464 for the shortest possible period of time.
1466 Severity : enhancement
1468 Description: refine locking for avoid write wrong info into lov_objid file
1469 Details : fix possible races with add new target and write/update data in
1472 --------------------------------------------------------------------------------
1474 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1479 Frequency : on some architectures
1481 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1482 Details : Uninitialized flags sometimes cause configuration commands to
1486 Frequency : patchless clients only
1488 Description: missing __iget() symbol export
1489 Details : The __iget() symbol export is missing. To avoid the need for
1490 this on patchless clients the deathrow inode reaper is turned
1491 off, and we depend on the VM to clean up old inodes. This
1492 dependency was during via the fix for bug 12181.
1494 --------------------------------------------------------------------------------
1496 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1498 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1499 INTEROPERATE with older versions automatically. In many cases a
1500 special upgrade step is needed. Please read the
1501 user documentation before upgrading any part of a live system.
1502 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1503 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1504 * WARNING: Lustre configuration and startup changes are required with
1505 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1511 Severity : enhancement
1513 Description: MountConf
1514 Details : Lustre configuration is now managed via mkfs and mount
1515 commands instead of lmc and lconf. New obd types (MGS, MGC)
1516 are added for dynamic configuration management. See
1517 https://mail.clusterfs.com/wikis/lustre/MountConf for
1520 Severity : enhancement
1522 Description: dynamic OST addition
1523 Details : OSTs can now be added to a live filesystem
1525 Severity : enhancement
1527 Description: startup order invariance
1528 Details : MDTs and OSTs can be started in any order. Clients only
1529 require the MDT to complete startup.
1531 Severity : enhancement
1533 Description: parallel, asynchronous orphan cleanup
1534 Details : orphan cleanup is now performed in separate threads for each
1535 OST, allowing parallel non-blocking operation.
1537 Severity : enhancement
1539 Description: optimized stripe assignment
1540 Details : stripe assignments are now made based on ost space available,
1541 ost previous usage, and OSS previous usage, in order to try
1542 to optimize storage space and networking resources.
1544 Severity : enhancement
1546 Description: Permanently set tunables
1547 Details : All writable /proc/fs/lustre tunables can now be permanently
1548 set on a per-server basis, at mkfs time or on a live system.
1550 Severity : enhancement
1552 Description: Lustre message v2
1553 Details : Add lustre message format v2.
1555 Severity : enhancement
1557 Description: client OST exclusion list
1558 Details : Clients can be started with a list of OSTs that should be
1559 declared "inactive" for known non-responsive OSTs.
1563 Description: SPEC SFS validation failure on NFS v2 over lustre.
1564 Details : Changes the blocksize for regular files to be 2x RPC size,
1565 and not depend on stripe size.
1567 Severity : enhancement
1569 Description: Multiple MD RPCs in flight.
1570 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1571 To avoid overly-overloading MDS, introduce a limit on number of
1572 MDS RPCs in flight for a single client and add /proc controls
1573 to adjust this limit.
1575 Severity : enhancement
1577 Description: client read/write statistics
1578 Details : Add client read/write call usage stats for performance
1579 analysis of user processes.
1580 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1581 file access. extents_stats shows chunk size distribution.
1582 extents_stats_per_process show chunk size distribution per
1585 Severity : enhancement
1587 Description: mds statistics
1588 Details : Add detailed mds operations statistics in
1589 /proc/fs/lustre/mds/*/stats.
1593 Description: Failure of copying files with lustre special EAs.
1594 Details : Client side always return success for setxattr call for lustre
1595 special xattr (currently only "trusted.lov").
1600 Description: Refcount LNET uuids
1601 Details : The global LNET uuid list grew linearly with every startup;
1602 refcount repeated list entries instead of always adding to
1605 Severity : enhancement
1607 Description: Dynamic service threads
1608 Details : Within a small range, start extra service threads
1609 automatically when the request queue builds up.
1612 Frequency : mixed-endian client/server environments
1614 Description: mixed-endian crashes
1615 Details : The new msg_v2 system had some failures in mixed-endian
1618 Severity : enhancement
1620 Description: Easy OST removal
1621 Details : OSTs can be permanently deactivated with e.g. 'lctl
1622 conf_param lustre-OST0001.osc.active=0'
1624 Severity : enhancement
1626 Description: MGS proc entries
1627 Details : Added basic proc entries for the MGS showing what filesystems
1630 Severity : enhancement
1632 Description: provide MGS failover
1633 Details : Added config lock reacquisition after MGS server failover.
1635 Severity : enhancement
1637 Description: add Linux 2.4 support
1638 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1642 Description: a large application tries to do I/O to the same resource and dies
1643 in the middle of it.
1644 Details : Check the req->rq_arrival time after the call to
1645 ost_brw_lock_get(), but before we do anything about
1646 processing it & sending the BULK transfer request. This
1647 should help move old stale pending locks off the queue as
1648 quickly as obd_timeout.
1651 Frequency : when an incorrect nid is specified during startup
1653 Description: ptlrpc connect to non-existant node causes kernel crash
1654 Details : LNET can't be re-entered from an event callback, which
1655 happened when we expire a message after the export has been
1656 cleaned up. Instead, hand the zombie cleanup off to another
1659 Severity : enhancement
1661 Description: plain/inodebits lock performance improvement
1662 Details : Grouping plain/inodebits in granted list by their request modes
1663 and bits policy, thus improving the performance of search through
1667 Frequency : only if OST filesystem is corrupted
1669 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1670 Details : for a short time RPCs with bulk IO are in the replay list,
1671 but replay of bulk IOs is unimplemented. If the OST filesystem
1672 is corrupted due to disk cache incoherency and then replay is
1673 started it is possible to trip an assertion. Avoid putting
1674 committed RPCs into the replay list at all to avoid this issue.
1677 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1679 Description: System hang on startup
1680 Details : This bug allowed the liblustre (e.g. catamount) client to
1681 return to the app before handling all startup RPCs. This
1682 could leave the node unresponsive to lustre network traffic
1683 and manifested as a server ptllnd timeout.
1685 Severity : enhancement
1687 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1688 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1689 boolean to control whether to print peer debug info when a
1690 client's RPC times out.
1693 Frequency : only for kernels with patches from Lustre below 1.4.3
1695 Description: Remove old rdonly API
1696 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1699 Frequency : only for devices with external journals
1701 Description: Set external device read-only also
1702 Details : During a commanded failover stop, we set the disk device
1703 read-only while the server shuts down. We now also set any
1704 external journal device read-only at the same time.
1707 Frequency : when upgrading from 1.4 while trying to change parameters
1709 Description: The wrong (new) MDC name was used when setting parameters for
1710 upgraded MDT's. Also allows changing of OSC (and MDC)
1711 parameters if --writeconf is specified at tunefs upgrade time.
1714 Frequency : when setting specific ost indicies
1716 Description: QOS code breaks on skipped indicies
1717 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1718 created with --index need not be sequential.
1720 Severity : enhancement
1722 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1723 Details : The uninit_groups feature works in conjunction with the kernel
1724 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1725 up the pass1 processing of e2fsck. This is a read-only feature
1726 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1727 mount filesystems that have had this feature enabled.
1729 Severity : enhancement
1731 Description: Improve multi-block allocation algorithm to avoid fragmentation
1732 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1733 allocation locality and avoid filesystem fragmentation.
1735 ------------------------------------------------------------------------------
1737 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1739 * Support for kernels:
1740 2.4.21-47.0.1.EL (RHEL 3)
1741 2.6.5-7.283 (SLES 9)
1742 2.6.9-42.0.10.EL (RHEL 4)
1743 2.6.12.6 vanilla (kernel.org)
1744 2.6.16.27-0.9 (SLES 10)
1745 * Recommended e2fsprogs version: 1.39.cfs5
1747 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1751 Frequency : occasional, depends on client load and configuration
1752 Bugzilla : 12181, 12203
1753 Description: data loss for recently-modified files
1755 Details : In some cases it is possible that recently written or created
1756 files may not be written to disk in a timely manner (this should
1757 normally be within 30s unless client IO load is very high).
1758 The problem appears as zero-length files or files that are a
1759 multiple of 1MB in size after a client crash or client eviction
1760 that are missing data at the end of the file.
1762 This problem is more likely to be hit on clients where files are
1763 repeatedly created and unlinked in the same directory, clients
1764 have a large amount of RAM, have many CPUs, the filesystem has
1765 many OSTs, the clients are rebooted frequently, and/or the files
1766 are not accessed by other nodes after being written.
1768 The presence of the problem can be detected by looking at
1769 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1770 smaller than the second (nr_unused) then dirty files will not
1771 be flushed automatically to disk. "sync; sleep 10" should be
1772 run several times on the node before unmounting it to update
1773 Lustre (this is also safe to run on nodes without this problem).
1775 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1776 that can cause this same problem, so customers using that kernel
1777 also need to update the kernel in addition to Lustre. In order
1778 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1780 It is normal that files written just before a client crash (less
1781 than 30s) may not yet have been flushed to disk, even for local
1785 Frequency : frequent on thin XT3 nodes
1787 Description: UUID collision on thin XT3 Linux nodes
1788 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1789 randomly, since we relied on an insufficiently-seeded PRNG.
1794 Description: OSS hangs after "All ost request buffers busy"
1795 Details : A deadlock between quota and journal operations caused OSS
1796 hangs after printing "All ost request buffers busy."
1799 Frequency : always on liblustre builds
1801 Description: Cleanup compiler warnings on liblustre
1804 Frequency : always on liblustre builds on XT3
1806 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1810 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1811 > 32000 subdirectories
1812 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1813 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1814 subdirectory count crosses 32000. This will aid e2fsck to
1815 correctly handle more than 32000 subdirectories.
1818 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1820 Description: System hang on startup
1821 Details : This bug allowed the liblustre (e.g. catamount) client to
1822 return to the app before handling all startup RPCs. This
1823 could leave the node unresponsive to lustre network traffic
1824 and manifested as a server ptllnd timeout.
1826 Severity : enhancement
1828 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1829 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1830 boolean to control whether to print peer debug info when a
1831 client's RPC times out.
1836 Description: make O_SYNC working on 2.6 kernels
1837 Details : 2.6 kernels use different method for mark pages for write,
1838 so need add a code to lustre for O_SYNC work.
1843 Description: Failure to close file and release space on NFS
1844 Details : Put inode details into lock acquired in ll_intent_file_open.
1845 Use mdc_intent_lock in ll_intent_open to properly
1846 detect all kind of errors unhandled by mdc_enqueue.
1851 Description: proc file read during shutdown sometimes raced obd removal,
1853 Details : Add lock to prevent obd access after proc file removal.
1856 Frequency : Only for files larger than 4GB on 32-bit clients.
1858 Description: improperly doing page alignment of locks
1859 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1860 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1865 Description: RPCs being resent when they shouldn't be
1866 Details : Some RPCs that should not be resent are being resent. This
1867 can cause inconsistencies in the RPC state machine. Do not
1868 resend such requests.
1871 Frequency : rare, only with NFS export
1873 Description: Crash on NFS re-export node
1874 Details : under very unusual load conditions an assertion is hit in
1875 ll_intent_file_open()
1878 Frequency : only if OST filesystem is corrupted
1880 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1881 Details : for a short time RPCs with bulk IO are in the replay list,
1882 but replay of bulk IOs is unimplemented. If the OST filesystem
1883 is corrupted due to disk cache incoherency and then replay is
1884 started it is possible to trip an assertion. Avoid putting
1885 committed RPCs into the replay list at all to avoid this issue.
1890 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1891 Details : Large single O_DIRECT read and write calls can fail to allocate
1892 a sufficiently large buffer to process the request. In case of
1893 allocation failure the allocation is retried with a smaller
1894 buffer and broken into smaller requests.
1896 Severity : enhancement
1898 Description: Add -o localflock option to simulate old noflock behaviour.
1899 Details : This will achieve local-only flock/fcntl locks coherentness.
1904 Description: versioning check is incomplete
1905 Details : Checking the version difference of client vs. server, report
1906 error if the gap is too big.
1911 Description: add support PG_writeback bit
1912 Details : add support for PG_writeback bit for Lustre, for more carefull
1913 work with page cache in 2.6 kernel. This also fix some deadlocks
1914 and remove hack for work O_SYNC with 2.6 kernel.
1916 Severity : enhancement
1918 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1919 Details : The uninit_groups feature works in conjunction with the kernel
1920 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1921 up the pass1 processing of e2fsck. This is a read-only feature
1922 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1923 mount filesystems that have had this feature enabled.
1925 Severity : enhancement
1927 Description: Improve multi-block allocation algorithm to avoid fragmentation
1928 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1929 allocation locality and avoid filesystem fragmentation.
1931 ------------------------------------------------------------------------------
1933 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1935 * Support for kernels:
1936 2.6.9-42.0.3.EL (RHEL 4)
1937 2.6.5-7.276 (SLES 9)
1938 2.4.21-47.0.1.EL (RHEL 3)
1939 2.6.12.6 vanilla (kernel.org)
1940 2.6.16.21-0.8 (SLES10)
1941 * Recommended e2fsprogs version: 1.39.cfs2-0
1943 * The backwards-compatible /proc/sys/portals symlink has been removed
1944 in this release. Before upgrading, please ensure that you change
1945 any configuration scripts or /etc/sysctl.conf files that access
1946 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1947 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1948 in advance of the upgrade on any system running Lustre 1.4.6 or
1949 newer, since /proc/sys/lnet was added in that version.
1950 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1954 Frequency : only when quota is used
1956 Description: avoid scanning export list for quota master
1957 Details : Change the algorithms to avoid scanning export list in order
1958 to improve the efficiency.
1961 Frequency : MDS failover only, very rarely
1963 Description: "went back in time" messages on mds failover
1964 Details : The greatest transno may be lost when the current operation
1965 finishes with an error (transno==0) and the client's last_rcvd
1966 record is over-written. Save the greatest transno in the
1967 mds_last_transno for this case.
1970 Frequency : always for specific kernels and striping counts
1972 Description: client may get "Matching packet too big" without ACL support
1973 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1974 when trying to access files in certain configurations. The
1975 clients should in fact be denied when mounting because they do
1976 not understand ACLs.
1979 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1981 Description: many clients connecting with IO in progress causes connect timeouts
1982 Details : Avoid synchronous journal commits to avoid delays caused by many
1983 clients connecting/disconnecting when bulk IO is in progress.
1984 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1985 OST_IO_PORTAL to avoid delays behind potentially many pending
1989 Frequency : occasionally with multiple writers to a single file
1991 Description: shared writes to file may result in wrong size reported by stat()
1992 Details : Allow growing of kms when extent lock is cancelled
1995 Frequency : always with random mmap IO to multi-striped file
1997 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1998 Details : Only if the hole is at the end of OST object so that kms is too
1999 small. Fix is to increase kms accordingly in ll_nopage.
2002 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2004 Description: writes to a missing object would leak memory on the OST
2005 Details : If there is an inconsistency between the MDS and OST filesystems,
2006 such that the MDS references an object that doesn't exist, writes
2007 to that object will leak memory due to incorrect cleanup in the
2008 error handling path, eventually running out of memory on the OST.
2013 Description: Creating too long symlink causes lustre errors
2014 Details : Check symlink and name lengths before sending requests to MDS.
2017 Frequency : only if flock is enabled (not on by default)
2019 Description: posix locks not released on fd closure on 2.6.9+
2020 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2021 kernels, this caused such locks not to be released on fd close and
2022 then assertions on fs unmount about still used locks.
2025 Frequency : MDS failover only, very rarely
2027 Description: clients may get ASSERTION(granted_lock != NULL)
2028 Details : When request was taking a long time, and a client was resending
2029 a getattr by name lock request. The were multiple lock requests
2030 with the same client lock handle and
2031 mds_getattr_name->fixup_handle_for_resent_request found one of the
2032 lock handles but later failed with ASSERTION(granted_lock != NULL).
2037 Description: handle->h_buffer_credits > 0, assertion failure
2038 Details : h_buffer_credits is zero after truncate, causing assertion
2039 failure. This patch extends the transaction or creates a new
2043 Frequency : NFS re-export or patchless client
2044 Bugzilla : 11179, 10796
2045 Description: Crash on NFS re-export node (__d_move)
2046 Details : We do not want to hash the dentry if we don't have a lock.
2047 But if this dentry is later used in d_move, we'd hit uninitialised
2048 list head d_hash, so we just do this to init d_hash field but
2049 leave dentry unhashed.
2052 Frequency : NFS re-export or patchless client
2054 Description: NFS exports has problem with symbolic link
2055 Details : lustre client didn't properly install dentry when re-exported
2056 to NFS or running patchless client.
2059 Frequency : NFS re-export or patchless client
2061 Description: Various nfs/patchless fixes.
2062 Details : fixes reuse disconected alias for lookup process - this fixes
2063 warning "find_exported_dentry: npd != pd",
2064 fix permission error with open files at nfs.
2065 fix apply umask when do revalidate.
2068 Frequency : occasional
2070 Description: Crash on NFS re-export node
2071 Details : calling clear_page() on the wrong pointer triggered oops in
2072 generic_mapping_read().
2075 Frequency : rarely, using O_DIRECT IO
2077 Description: unaligned directio crashes client with LASSERT
2078 Details : check for unaligned buffers before trying any requests.
2081 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2083 Description: stale data returned from RAID cache
2084 Details : If only a small amount of IO is done to the RAID device before
2085 reading it again it is possible to get stale data from the RAID
2086 cache instead of reading it from disk.
2089 Frequency : always for sles10 kernel
2091 Description: sles10 support
2092 Details : ll_follow_link: compile fixes and using of nd_set_link
2093 under newer kernels.
2096 Frequency : depends on arch, kernel and compiler version, always on sles10
2099 Description: recursive or deep enough symlinks cause stack overflow
2100 Details : getting rid of large stack-allocated variable in
2104 Frequency : depends on hardware
2106 Description: lustre write performance loss in the SLES10 kernel
2107 Details : the performance loss is caused by using of write barriers in the
2108 ext3 code. The SLES10 kernel turns barrier support on by
2109 default. The fix is to undo that change for ldiskfs.
2111 ------------------------------------------------------------------------------
2113 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2115 * Support for kernels:
2116 2.6.9-42.0.3EL (RHEL 4)
2117 2.6.5-7.276 (SLES 9)
2118 2.4.21-47.0.1.EL (RHEL 3)
2119 2.6.12.6 vanilla (kernel.org)
2123 Frequency : quota enabled and large files being deleted
2125 Description: releasing more than 4GB of quota at once hangs OST
2126 Details : If a user deletes more than 4GB of files on a single OST it
2127 will cause the OST to spin in an infinite loop. Release
2128 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2133 Description: statfs data retrieved from /proc may be stale or zero
2134 Details : When reading per-device statfs data from /proc, in the
2135 {kbytes,files}_{total,free,avail} files, it may appear
2136 as zero or be out of date.
2139 Frequency : systems with MD RAID1 external journal devices
2141 Description: lconf's call to blkid is confused by RAID1 journal devices
2142 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2143 of returning all block devices that match the journal UUID.
2146 Frequency : always, for aggregate stripe size over 4GB
2148 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2149 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2150 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2153 Frequency : always if "lfs find" used on a local file/directory
2155 Description: "lfs find" segfaults if used on a local file/directory
2156 Details : The case where a directory component was not specified wasn't
2157 handled correctly. Handle this properly.
2160 Frequency : always on ppc64
2162 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2163 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2164 perform endianness conversion.
2167 Frequency : rarely (truncate to non-zero file size after write under load)
2168 Bugzilla : 10730, 10687
2169 Description: Files padded with zeros to next 4K multiple
2170 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2171 it is possible that files that are truncated to a non-zero size
2172 immediately after being written are filled with zero bytes beyond
2173 the truncated size. No file data is lost.
2175 Severity : enhancement
2177 Description: Allow recovery/failover for liblustre clients.
2178 Details : liblustre clients were unaware of failover configurations until
2181 Severity : enhancement
2183 Description: user file locks should fail when not mounting with flock option
2184 Details : Set up an error-returning stub in ll_file_operations.lock field
2185 to prevent incorrect behaviour when client is mounted without
2186 flock option. Also, set up properly f_op->flock field for
2190 Frequency : always on ia64
2192 Description: "lfs df" loops on printing out MDS statfs information
2193 Details : The obd_ioctl_data was not initialized and in some systems
2194 this caused a failure during the ioctl that did not return
2195 an error. Initialize the struct and return an error on failure.
2198 Frequency : SLES 9 only
2200 Description: Error of copying files with lustre special EAs as root
2201 Details : Client side always return success for setxattr call for lustre
2202 special xattr (currently only "trusted.lov").
2205 Frequency : rarely on clusters with both ia64+i386 clients
2207 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2208 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2209 clients are doing concurrent writes to the same file it is
2210 possible that the ia64 clients may LASSERT because the OST
2211 extent locks are not PAGE_SIZE aligned. Ensure that grown
2212 locks are always aligned on the request boundary.
2215 Frequency : specific use, occasional
2217 Description: Overwriting in use executable truncates on-disk binary image
2218 Details : If one node attempts to overwrite an executable in use by
2219 another node, we now correctly return ETXTBSY instead of
2220 truncating the file.
2222 Severity : enhancement
2224 Description: Async OSC create to avoid the blocking unnecessarily.
2225 Details : If a OST has no remain object, system will block on the creating
2226 when need to create a new object on this OST. Now, ways use
2227 pre-created objects when available, instead of blocking on an
2228 empty osc while others are not empty. If we must block, we block
2229 for the shortest possible period of time.
2234 Description: chmod on Lustre root is propagated to other clients
2235 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2236 invalid by the follow_mount time.
2239 Frequency : liblustre clients only
2241 Description: Race in 'instant cancel' lock handling could lead to such locks
2242 never to be granted in case of SMP MDS
2243 Details : Do not destroy not yet granted but cbpending locks in
2247 Frequency : replay/resend of open
2249 Description: non null lock assetion failure in mds_intent_policy
2250 Details : Trying to replay/resend lockless open requests resulted in
2251 mds_open() returning 0 with no lock. Now it sets a flag if
2252 a lock is going to be returned.
2254 Severity : enhancement
2256 Description: Checksum enhancements
2257 Details : New checksum enhancements allow for resending RPCs that failed
2260 Severity : enhancement
2262 Description: Tunables on number of dirty pages in cacche
2263 Details : Allow to set limit on number of dirty pages cached.
2268 Description: client crash on unmount - lock still has references
2269 Details : In some error handling cases it was possible to leak a lock
2270 reference on a client while accessing a file. This was not
2271 harmful to the client during operation, but would cause the
2272 client to crash when the filesystem is unmounted.
2275 Frequency : specific case, rare
2277 Description: ETXTBSY on mds though file not in use by client
2278 Details : ETXTBSY is no longer incorrectly returned when attempting to
2279 chmod or chown a directory that the user previously tried to
2280 execute or a currently-executing binary.
2283 Frequency : extremely rare except on liblustre-based clients
2285 Description: Lustre space not freed when files are deleted
2286 Details : Clean up open-unlinked files after client eviction. Previously
2287 the unlink was skipped and the files remained as orphans.
2292 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2293 Details : In some cases it was possible to send a blocking callback to
2294 a client doing a glimpse, even though that client didn't get
2295 a lock granted. When the glimpse lock is cancelled on the OST
2296 the freed lock is left on the waiting list and corrupted the list.
2299 Frequency : all core dumps
2301 Description: Broke core dumps to lustre
2302 Details : Negative dentry may be unhashed if parent does not have UPDATE
2303 lock, but some callers, e.g. do_coredump, expect dentry to be
2304 hashed after successful create, hash it in ll_create_it.
2306 ------------------------------------------------------------------------------
2308 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2310 * Support for kernels:
2311 2.6.9-42.0.2.EL (RHEL 4)
2312 2.6.5-7.276 (SLES 9)
2313 2.4.21-40.EL (RHEL 3)
2314 2.6.12.6 vanilla (kernel.org)
2318 Frequency : always on RHEL 3
2320 Description: Number of open files grows over time
2321 Details : The number of open files grows over time, whether or not
2322 Lustre is started. This was due to a filp leak introduced
2323 by one of our kernel patches.
2325 ------------------------------------------------------------------------------
2327 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2329 * Support for kernels:
2330 2.6.9-42.EL (RHEL 4)
2331 2.6.5-7.276 (SLES 9)
2332 2.4.21-40.EL (RHEL 3)
2333 2.6.12.6 vanilla (kernel.org)
2338 Bugzilla : 5719, 9635, 9792, 9684
2339 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2340 Details : If a server is under heavy load and cannot reply to new
2341 connection requests before the client resends the (re)connect,
2342 the connection handling code can behave badly if two service
2343 threads are concurrently handing separate (re)connections from
2344 the same client. Add better locking to the connection handling
2345 code, and ensure that only a single connection will be processed
2346 for a given client UUID, even if the lock is dropped.
2348 Severity : enhancement
2350 Description: add TCP zero-copy support to kernel
2351 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2352 sends if the hardware supports scatter-gather and checksumming.
2353 This allows socklnd to do client-write and server-read more
2354 efficiently and reduce CPU utilization from skbuf copying.
2357 Frequency : only if NFS exporting from client
2359 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2360 Details : If setting attributes on a file created under NFS that had
2361 never been opened it would be possible to oops the client
2362 if the file had no objects.
2365 Frequency : always for liblustre
2367 Description: liblustre client does MDS+OSTs setattr RPC for each write
2368 Details : When doing a write from a liblustre client, the client
2369 incorrectly issued an RPC to the MDS and each OST the file was
2370 striped over in order to update the timestamps. When writing
2371 with small chunks and many clients this could overwhelm the MDS
2372 with RPCs. In all cases it would slow down the write because
2373 these RPCs are unnecessary.
2375 Severity : enhancement
2377 Description: allow number of MDS service threads to be changed at module load
2378 Details : It is now possible to change the number of MDS service threads
2379 running. Adding "options mds mds_num_threads={N}" to the MDS's
2380 /etc/modprobe.conf will set the number of threads for the next
2381 time Lustre is restarted (assuming the "mds" module is also
2382 reloaded at that time). The default number of threads will
2383 stay the same, 32 for most systems.
2388 Description: OST crash if filesystem is unformatted or corrupt
2389 Details : If an OST is started on a device that has never been formatted
2390 or if the filesystem is corrupt and cannot even mount then the
2391 error handling cleanup routines would dereference a NULL pointer.
2396 Description: NULL pointer deref in llap_from_page.
2397 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2398 incorrect) mapping if the page was truncated/reclaimed while it was
2399 searched for. Check for this condition and skip such pages when
2400 doing readahead. Introduce extra check to llap_from_page() to
2401 verify page->mapping->host is non-NULL (so page is not anonymous).
2404 Frequency : Sometimes when using sys_sendfile
2406 Description: "page not covered by a lock" warnings from ll_readpage
2407 Details : sendfile called ll_readpage without right page locks present.
2408 Now we introduced ll_file_sendfile that does necessary locking
2409 around call to generic_file_sendfile() much like we do in
2413 Frequency : with certain MDS communication failures at client mount time
2415 Description: NULL pointer deref after failed client mount
2416 Details : a client connection request may delayed by the network layer
2417 and not be sent until after the PTLRPC layer has timed out the
2418 request. If the client fails the mount immediately it will try
2419 to clean up before the network times out the request. Add a
2420 reference from the request import to the obd device and delay
2421 the cleanup until the network drops the request.
2424 Frequency : occasionally during client (re)connect
2426 Description: assertion failure during client (re)connect
2427 Details : processing a client connection request may be delayed by the
2428 client or server longer than the client connect timeout. This
2429 causes the client to resend the connection request. If the
2430 original connection request is replied in this interval, the
2431 client may trip an assertion failure in ptlrpc_connect_interpret()
2432 which thought it would be the only running connect process.
2435 Frequency : only with obd_echo servers and clients that are rebooted
2437 Description: kernel BUG accessing uninitialized data structure
2438 Details : When running an obd_echo server it did not start the ping_evictor
2439 thread, and when a client was evicted an uninitialized data
2440 structure was accessed. Start the ping_evictor in the RPC
2441 service startup instead of the OBD startup.
2443 Severity : enhancement
2444 Bugzilla : 10193 (patchless)
2445 Description: Remove dependency on various unexported kernel interfaces.
2446 Details : No longer need reparent_to_init, exit_mm, exit_files,
2447 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2450 Frequency : rare (only users of deprecated and unsupported LDAP config)
2452 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2453 Details : LDAP apparently contains 'lustreName' attributes instead of
2454 'name'. A simple remapping of the name is sufficient.
2457 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2459 Description: waiting_locks_callback trips kernel BUG if client is evicted
2460 Details : Running with the dump_on_timeout debug flag turned on makes
2461 it possible that the waiting_locks_callback() can try to dump
2462 the Lustre kernel debug logs from an interrupt handler. Defer
2463 this log dumping to the expired_lock_main() thread.
2465 Severity : enhancement
2467 Description: Support NFS exporting on 2.6 kernels.
2468 Details : Implement non-rawops metadata methods for NFS server to use without
2469 changing NFS server code.
2472 Frequency : very rare (synthetic metadata workload only)
2474 Description: two racing renames might cause an MDS thread to deadlock
2475 Details : Running the "racer" program may cause one MDS thread to rename
2476 a file from being the source of a rename to being the target of
2477 a rename at exactly the same time that another thread is doing
2478 so, and the second thread has already enqueued these locks after
2479 doing a lookup of the target and is trying to relock them in
2480 order. Ensure that we don't try to re-lock the same resource.
2483 Frequency : only very large systems with liblustre clients
2485 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2486 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2487 all OSTs in parallel. This allows the eviction of stale liblustre
2488 clients to proceed much faster than if they were done in series,
2489 and also offers similar improvements for other set_info RPCs.
2494 Description: excessive CPU usage during initial read phase on client
2495 Details : During the initial read phase on a client, it would agressively
2496 retry readahead on the file, consuming too much CPU and impacting
2497 performance (since 1.4.5.8). Improve the readahead algorithm
2498 to avoid this, and also improve some other common cases (read
2499 of small files in particular, where "small" is files smaller than
2500 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2505 Description: MDS crash when receiving packet with unknown intent.
2506 Details : Do not LBUG in unknown intent case, just return -EFAULT
2508 Severity : enhancement
2509 Bugzilla : 9293, 9385
2510 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2511 Details : Do not serialize getattr (non-intent version) and statfs.
2514 Frequency : occasional, when OST network is overloaded/intermittent
2516 Description: client evicted by OST after bulk IO timeout
2517 Details : If a client sends a bulk IO request (read or write) the OST
2518 may evict the client if it is unresposive to its data GET/PUT
2519 request. This is incorrect if the network is overloaded (takes
2520 too long to transfer the RPC data) or dropped the OST GET/PUT
2521 request. There is no need to evict the client at all, since
2522 the pinger and/or lock callbacks will handle this, and the
2523 client can restart the bulk request.
2526 Frequency : Always when mmapping file with no objects
2528 Description: client crashes when mmapping file with no objects
2529 Details : Check that we actually have objects in a file before doing any
2530 operations on objects in ll_vm_open, ll_vm_close and
2536 Description: Request leak when working with deleted CWD
2537 Details : Introduce advanced request refcount tracking for requests
2538 referenced from lustre intent.
2540 Severity : Enhancement
2542 Description: Cache open file handles on client.
2543 Details : MDS now will return special lock along with openhandle, if
2544 requested and client is allowed to hold openhandle, even if unused,
2545 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2546 closing files for every read/write openration.
2548 Severity : Enhancement
2550 Description: Cache open negative dentries on client when possible.
2551 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2552 negative dentries on lock revocation.
2557 Description: Remounting a client read-only wasn't possible with a zconf mount
2558 Details : It wasn't possible to remount a client read-only with llmount.
2560 Severity : enhancement
2561 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2562 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2563 MPICH is included to add Lustre-specific ADIO interfaces.
2564 This is based closely on the UFS ADIO layer and only differs
2565 in file creation, in order to allow the OST striping to be set.
2566 This is user-contributed code and not supported by CFS.
2571 Description: extended inode attributes (immutable, append-only) work improperly
2572 when 2.4 and 2.6 kernels are used on client/server or vice versa
2573 Details : Introduce kernel-independent values for these flags.
2575 Severity : enhancement
2578 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2579 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2580 granularity. Now, Fractional MB granularity is be supported,
2581 this is very useful for low memory system.
2583 Severity : enhancement
2585 Description: Getattr by fid
2586 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2587 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2588 have locks and avoids getattr rpc requests for it on every stat.
2591 Frequency : Always, for filesystems larger than 2TB
2593 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2594 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2595 filesystems larger than 2TB to be mounted with mballoc enabled.
2598 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2600 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2601 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2602 filesystems larger than 2TB is mounted then the kernel will
2603 silently corrupt the start of the filesystem. CONFIG_LBD is
2604 enabled for all CFS-supported kernels, but the possibility of
2605 this happening with a modified kernel config exists.
2607 Severity : enhancement
2609 Description: add client O_DIRECT support for 2.6 kernels
2610 Details : It is now possible to do O_DIRECT reads and writes to files
2611 in the Lustre client mountpoint on 2.6 kernel clients.
2613 Severity : enhancement
2615 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2616 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2617 parallel, not waiting for response from every OST before sending
2618 a rpc to the next OST.
2623 Description: setattr vs write race when updating file timestamps
2624 Details : Client processes that update a file timestamp into the past
2625 right after writing to the file (e.g. tar) it is possible that
2626 the updated file modification time can be reset to the current
2627 time due to a race between processing the setattr and write RPC.
2629 Severity : enhancement
2631 Description: Bring 'lfs find' closer in line with regular Linux find.
2632 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2633 -print0 options and obtains all the needed info through the lustre
2636 Severity : enhancement
2638 Description: support up to 1024 configured devices on one node
2639 Details : change obd_dev array from statically allocated to dynamically
2640 allocated structs as they are first used to reduce memory usage
2645 Description: Flush dirty partially truncated pages during truncate
2646 Details : Immediatelly flush partially truncated pages in filter_setattr,
2647 this way we completely avoid having any pages in page cache on OST
2648 and can retire ugly workarounds during writes to flush such pages.
2653 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2654 Details : protect inode from truncation within vfs_unlink() context
2655 just take a reference before calling vfs_unlink() and release it
2656 when parent's i_sem is free.
2661 Description: last_id value checked outside lock on OST caused LASSERT failure
2662 Details : If there were multiple MDS->OST object precreate requests in
2663 flight, it was possible that the OST's last object id was checked
2664 outside a lock and incorrectly tripped an assertion. Move checks
2665 inside locks, and discard old precreate requests.
2668 Frequency : always, if extents are used on OSTs
2670 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2671 Details : OSTs using the extents format would not zero the high 16 bits of
2672 the index physical block number. This is not a problem for any
2673 OST filesystems smaller than 16TB, and no kernels support ext3
2674 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2675 new/modified files) and can be fixed for existing filesystems
2676 with e2fsprogs-1.39-cfs1.
2681 Description: import connection selection may be incorrect if timer wraps
2682 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2683 import connections to be ignored if the 32-bit jiffies counter
2684 wraps. Use a 64-bit jiffies counter.
2687 Frequency : very large clusters immediately after boot
2689 Description: LNET request buffers exhausted under heavy short-term load
2690 Details : If a large number of client requests are generated on a service
2691 that has previously never seen so many requests it is possible
2692 that the request buffer growth cannot keep up with the spike in
2693 demand. Instead of dropping incoming requests, they are held in
2694 the LND until the RPC service can accept more requests.
2697 Frequency : Sometimes during replay
2699 Description: Assertion failure in ll_local_open after replay.
2700 Details : If replay happened on an open request reply before we were able
2701 to set replay handler, reply will become not swabbed tripping the
2702 assertion in ll_local_open. Now we set the handler right after
2703 recognising of open request
2706 Frequency : very rare
2708 Description: kernel reports "badness in vsnprintf"
2709 Details : Reading from the "recovery_status" /proc file in small chunks
2710 may cause a negative length in lprocfs_obd_rd_recovery_status()
2711 call to vsnprintf() (which is otherwise harmless). Exit early
2712 if there is no more space in the output buffer.
2714 Severity : enhancement
2716 Description: clear OBD RPC statistics by writing to them
2717 Details : It is now possible to clear the OBD RPC statistics by writing
2718 to the "stats" file.
2723 Description: Client mtime is not the same on different clients after utimes
2724 Details : In some cases, the client was using the utimes() syscall on
2725 a file cached on another node. The clients now validate the
2726 ctime from the MDS + OSTs to determine which one is right.
2731 Description: Inability to activate failout mode
2732 Details : lconf script incorrectly assumed that in pythong string's numeric
2733 value is used in comparisons.
2736 Frequency : always with multiple stripes per file
2738 Description: Inefficient object allocation for mutli-stripe files
2739 Details : When selecting which OSTs to stripe files over, for files with
2740 a stripe count that divides evenly into the number of OSTs,
2741 the MDS is always picking the same starting OST for each file.
2742 Return the OST selection heuristic to the original design.
2747 Description: mount failures may take full timeout to return an error
2748 Details : Under some heavy load conditions it is possible that a
2749 failed mount can wait for the full obd_timeout interval,
2750 possibly several minutes, before reporting an error.
2751 Instead return an error as soon as the status is known.
2753 Frequency : quota enabled and large files being deleted
2755 Description: releasing more than 4GB of quota at once hangs OST
2756 Details : If a user deletes more than 4GB of files on a single OST it
2757 will cause the OST to spin in an infinite loop. Release
2758 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2763 Description: statfs data retrieved from /proc may be stale or zero
2764 Details : When reading per-device statfs data from /proc, in the
2765 {kbytes,files}_{total,free,avail} files, it may appear
2766 as zero or be out of date.
2769 Frequency : systems with MD RAID1 external journal devices
2771 Description: lconf's call to blkid is confused by RAID1 journal devices
2772 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2773 of returning all block devices that match the journal UUID.
2776 Frequency : always, for aggregate stripe size over 4GB
2778 Description: assertion fails when trying to use 4GB stripe size
2779 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2780 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2783 Frequency : always on ppc64
2785 Description: the first write on an ext3 filesystem with mballoc got stuck
2786 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2787 perform endianness conversion.
2789 ------------------------------------------------------------------------------
2791 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2793 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2794 INTEROPERATE with older versions automatically. Please read the
2795 user documentation before upgrading any part of a live system.
2796 * WARNING: Lustre networking configuration changes are required with
2797 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2800 * Support for kernels:
2801 2.6.9-22.0.2.EL (RHEL 4)
2802 2.6.5-7.244 (SLES 9)
2803 2.6.12.6 vanilla (kernel.org)
2806 Severity : enhancement
2807 Bugzilla : 7981/8208
2808 Description: Introduced Lustre Networking (LNET)
2809 Details : LNET is new networking infrastructure for Lustre, it includes
2810 a reorganized network configuration mode (see the user
2811 documentation for full details) as well as support for routing
2812 between different network fabrics. Lustre Networking Devices
2813 (LNDs) for the supported network fabrics have also been
2814 created for this new infrastructure.
2816 Severity : enhancement
2817 Description: Introduced Access control lists
2818 Details : clients can set ACLs on files and directories in order to have
2819 more fine-grained permissions than the standard Unix UGO+RWX.
2820 The MDS must be started with the "-o acl" mount option.
2822 Severity : enhancement
2823 Description: Introduced filesystem quotas
2824 Details : Administrators may now establish per-user quotas on the
2827 Severity : enhancement
2829 Description: Configuration change for the XT3
2830 The PTLLND is now used to run Lustre over Portals on the XT3
2831 The configure option(s) --with-cray-portals are no longer used.
2832 Rather --with-portals=<path-to-portals-includes> is used to
2833 enable building on the XT3. In addition to enable XT3 specific
2834 features the option --enable-cray-xt3 must be used.
2839 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2840 Details : A race between allocating a new llcd and re-getting the llcd_lock
2841 allowed another thread to grab newly-allocated llcd.
2843 Severity : enhancement
2845 Description: 2.6 OST async journal commit and locking fix to improve performance
2846 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2847 2.6 kernels are now async as they already were in 2.4 kernels so
2848 that they can commit concurrently with the network bulk transfer.
2849 For block-allocated files the filter allocation semaphore is held
2850 to avoid filesystem fragmentation during allocation. BKL lock
2851 removed for 2.6 xattr operations where it is no longer needed.
2856 Description: lconf incorrectly determined whether two IP networks could talk
2857 Details : In some more complicated routing and multiple-network
2858 configurations, lconf will avoid trying to make a network
2859 connection to a disjoint part of the IP space. It was doing the
2860 math incorrectly for one set of cases.
2865 Description: Fix for potential infinite loop processing records in an llog.
2866 Details : If an llog record is corrupted/zeroed, it is possible to loop
2867 forever in llog_process(). Validate the llog record length
2868 and skip the remainder of the block on error.
2871 Frequency : occasional (liblustre only)
2873 Description: liblustre could not open files whose last component is a symlink
2874 Details : sysio_path_walk() would incorrectly pass the open intent to
2875 intermediate path components.
2878 Frequency : rare (liblustre only with non-standard tuning)
2879 Bugzilla : 7201 (7350)
2880 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2881 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2882 after it has been released, only do this for referenced locks
2884 Severity : enhancement
2886 Description: specifying an (invalid) directory default stripe_size of -1
2887 would reset the directory default striping
2888 Details : stripe_size -1 was used internally to signal directory stripe
2889 removal, now use "all default" to signal dir stripe removal
2890 as a directory striping of "all default" is not useful
2893 Frequency : common for large clusters running liblustre clients
2895 Description: doing an ls when liblustre clients are running is slow
2896 Details : sending a glimpse AST to a liblustre client waits for every AST
2897 to time out, as liblustre clients will not respond. Since they
2898 cannot cache data we refresh the OST lock LVB from disk instead.
2900 Severity : enhancement
2902 Description: doing an ls at the same time as file IO can be slow
2903 Details : enqueue and other "small" requests can be blocked behind many
2904 large IO requests. Create a new OST IO portal for non-IO
2905 requests so they can be processed faster.
2908 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2910 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2911 created file with mode 000
2912 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2913 allow an ftruncate() as a non-root user to a file with mode 000.
2914 Lustre now respects this flag to disable mode checks when
2915 truncating a file owned by the user
2918 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2921 Description: Revoking locks from clients that went dead or catatonic might take
2923 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2924 cancellation of such locks instant on servers without waiting for
2925 any reply from clients. Clients drops these locks when cancel
2926 notification from server is received without replying.
2929 Frequency : liblustre-only, when liblustre client dies or becomes busy
2931 Description: Doing ls on Linux clients can take a long time with active
2933 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2934 granting such locks to it in the first place if possible. Locks
2935 are taken by proxy on the OST during the read or write and
2936 dropped immediately afterward. Add connect flags handling, do
2937 not grant locks to liblustre clients for glimpse ASTs.
2939 Severity : enhancement
2941 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2942 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2943 sequential reads and behaves badly with applications doing random
2944 reads. Improve it by reading ahead at least read region, and
2945 avoiding excessive large RPC for small reads.
2947 Severity : enhancement
2949 Description: Creating more than 1000 files for a single job may cause a load
2950 imbalance on the OSTs if there are also a large number of OSTs.
2951 Details : qos_prep_create() uses an OST index reseed value that is an
2952 even multiple of the number of available OSTs so that if the
2953 reseed happens in the middle of the object allocation it will
2954 still utilize the OSTs as uniformly as possible.
2959 Description: OST or MDS may oops in ping_evictor_main()
2960 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2961 but doesn't restart at beginning of obd_exports_timed list
2964 Severity : enhancement
2966 Description: improve by-nid export eviction on the MDS and OST
2967 Details : allow multiple exports with the same NID to be evicted at one
2968 time without re-searching the exports list.
2971 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2973 Description: MDS may oops in groups_free()
2974 Details : in rare race conditions a newly allocated group_info struct is
2975 freed again, and this can be NULL. The 2.4 compatibility code
2976 for groups_free() checked for a NULL pointer, but 2.6 did not.
2979 Frequency : common for liblustre clients doing little filesystem IO
2980 Bugzilla : 9352, 7313
2981 Description: server may evict liblustre clients accessing contended locks
2982 Details : if a client is granted a lock or receives a completion AST
2983 with a blocking AST already set it would not reply to the AST
2984 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2987 Frequency : lfs setstripe, only systems with more than 160 OSTs
2989 Description: unable to set striping with a starting offset beyond OST 160
2990 Details : llapi_create_file() incorrectly limited the starting stripe
2991 index to the maximum single-file stripe count.
2994 Frequency : LDAP users only
2996 Description: lconf did not handle in-kernel recovery with LDAP properly
2997 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2999 Severity : enhancement
3001 Description: bind OST threads to NUMA nodes to improve performance
3002 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3003 node and do their allocations there to localize memory access
3005 Severity : enhancement
3007 Description: llmount can determine client NID directly from Myrinet (GM)
3008 Details : the client NID code from gmnalnid was moved directly into
3009 llmount, removing the need to use this or specifying the
3010 client NID explicitly when mounting GM clients with zeroconf
3013 Frequency : if client is started with down MDS
3015 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3016 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3017 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3018 ptlrpc_import_delayed_req() because we didn't check if the
3019 request was interrupted, and we also didn't break out of the
3020 event loop if there was no timeout
3025 Description: data loss during non-page-aligned writes to a single file from
3026 both multiple nodes and multiple threads on one node at same time
3027 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3028 inconsistency led to false short-reads, that were cached and later
3029 used by ->prepare_write() to fill in partially written page,
3030 leading to data loss.
3033 Frequency : always, if lconf --abort_recovery used
3035 Description: lconf --abort_recovery fails with 'Operation not supported'
3036 Details : lconf was attempting to abort recovery on the MDT device and not
3039 Severity : enhancement
3041 Description: remove cleanup logs
3042 Details : replace lconf-generated cleanup logs with lustre internal
3043 cleanup routines. Eliminates the need for client-cleanup and
3046 Severity : enhancement
3048 Description: add support for EAs (user and system) on lustre filesystems
3049 Details : it is now possible to store extended attributes in the Lustre
3050 client filesystem, and with the user_xattr mount option it
3051 is possible to allow users to store EAs on their files also
3053 Severity : enhancement
3055 Description: Add possibility (config option) to show minimal available OST free
3057 Details : When compiled with --enable-mindf configure option, statfs(2)
3058 (and so, df) will return least minimal free space available from
3059 all OSTs as amount of free space on FS, instead of summary of
3060 free spaces of all OSTs.
3062 Severity : enhancement
3064 Description: do not expand extent locks acquired on OST-side
3065 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3066 by server: they are not cached anyway.
3069 Frequency : when mmap is used/binaries executed from Lustre
3071 Description: Unmmap pages before throwing them away from read cache.
3072 Details : llap_shrink cache now attempts to unmap pages before discarding
3073 them (if unmapping failed - do not discard). SLES9 kernel has
3074 extra checks that trigger if this unmapping is not done first.
3079 Description: lconf didn't resolve symlinks before checking to see whether a
3080 given mountpoint was already in use
3083 Frequency : when migrating failover services
3084 Bugzilla : 6395, 9514
3085 Description: When migrating a subset of services from a node (e.g. failback
3086 from a failover service node) the remaining services would
3087 time out and evict clients.
3088 Details : lconf --force (implied by --failover) sets the global obd_timeout
3089 to 5 seconds in order to quickly disconnect, but this caused
3090 other RPCs to time out too quickly. Do not change the global
3091 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3093 Severity : enhancement
3094 Frequency : if MDS is started with down OST
3095 Bugzilla : 9439,5706
3096 Description: Allow startup/shutdown of an MDS without depending on the
3097 availability of the OSTs.
3098 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3099 Add appropriate locking and lov-osc refcounts for safe
3100 cleaning. Add osc abort_inflight calls in case the
3101 synchronize never started.
3104 Frequency : occasional (Cray XT3 only)
3106 Description: root not authorized to access files in CRAY_PORTALS environment
3107 Details : The client process capabilities were not honoured on the MDS in
3108 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3109 been accessed by an authorized user then root was able to access
3110 the file on the local client also. The root user capabilities
3111 are now allowed on the MDS, as this environment has secure UID.
3114 Frequency : occasional
3116 Description: ldiskfs "too long searching" message happens too often
3117 Details : A debugging message (otherwise harmless) prints too often on
3118 the OST console. This has been reduced to only happen when
3119 there are fragmentation problems on the filesystem.
3124 Description: Division by zero in statfs when all OSCs are inactive
3125 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3126 lov_statfs divides by value returned by lov_get_stripecnt().
3130 Bugzilla : 9489, 3273
3131 Description: First write from each client to each OST was only 4kB in size,
3132 to initialize client writeback cache, which caused sub-optimal
3133 RPCs and poor layout on disk for the first writen file.
3134 Details : Clients now request an initial cache grant at (re)connect time
3135 and so that they can start streaming writes to the cache right
3136 away and always do full-sized RPCs if there is enough data.
3137 If the OST is rebooted the client also re-establishes its grant
3138 so that client cached writes will be honoured under the grant.
3143 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3144 Details : Now I/O RPCs go to different portal number and (presumably) fast
3145 lock requests (and glimses) and other RPCs get their own service
3146 threads pool that should be able to service those RPCs
3149 Severity : enhancement
3151 Description: Ability to exchange lustre version between client and servers and
3152 issue warnings at client side if client is too old. Also for
3153 liblustre clients there is ability to refuse connection of too old
3155 Details : New 'version' field is added to connect data structure that is
3156 filled with version info. That info is later checked by server and
3160 Frequency : rare, liblustre only.
3161 Bugzilla : 9296, 9581
3162 Description: Two simultaneous writes from liblustre at offset within same page
3163 might proceed at the same time overwriting eachother with stale
3165 Details : I/O lock withing llu_file_prwv was released too early, before data
3166 actually was hitting the wire. Extended lock-holding time until
3167 server acknowledges receiving data.
3170 Frequency : extremely rare. Never observed in practice.
3172 Description: avoid generating lustre_handle cookie of 0.
3173 Details : class_handle_hash() generates handle cookies by incrementing
3174 global counter, and can hit 0 occasionaly (this is unlikely, but
3175 not impossible, because initial value of cookie counter is
3176 selected randonly). Value of 0 is used as a sentinel meaning
3177 "unassigned handle" --- avoid it. Also coalesce two critical
3178 sections in this function into one.
3180 Severity : enhancement
3182 Description: allow liblustre clients to delegate truncate locking to OST
3183 Details : To avoid overhead of locking, liblustre client instructs OST to
3184 take extent lock in ost_punch() on client's behalf. New connection
3185 flag is added to handle backward compatibility.
3187 Severity : enhancement
3188 Bugzilla : 4928, 7341, 9758
3189 Description: allow number of OST service threads to be specified
3190 Details : a module parameter allows the number of OST service threads
3191 to be specified via "options ost ost_num_threads={N}" in the
3192 OSS's /etc/modules.conf or /etc/modprobe.conf.
3196 Bugzilla : 6146, 9635, 9895
3197 Description: servers crash with bad pointer in target_handle_connect()
3198 Details : In rare cases when a client is reconnecting it was possible that
3199 the connection request was the last reference for that export.
3200 We would temporarily drop the export reference and get a new
3201 one, but this may have been the last reference and the export
3202 was just destroyed. Get new reference before dropping old one.
3204 Severity : enhancement
3205 Frequency : if client is started with failover MDS
3207 Description: Allow multiple MDS hostnames in the mount command
3208 Details : Try to read the configuration from all specified MDS
3209 hostnames during a client mount in case the "primary"
3212 Severity : enhancement
3214 Description: Stop sending data to evicted clients as soon as possible.
3215 Details : Check if the client we are about to send or are sending data to
3216 was evicted already. (Check is done every second of waiting,
3217 for which l_wait_event interface was extended to allow checking
3218 of exit condition at specified intervals).
3221 Frequency : rare, normally only when NFS exporting is done from client
3223 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3225 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3228 Frequency : always, if srand() is called before liblustre initialization
3230 Description: Liblustre uses system PRNG disturbing its usage by user application
3231 Details : Introduce internal to lustre fast and high-quality PRNG for
3232 lustre usage and make liblustre and some other places in generic
3233 lustre code to use it.
3235 Severity : enhancement
3236 Bugzilla : 9477, 9557, 9870
3237 Description: Verify that the MDS configuration logs are updated when xml is
3238 Details : Check if the .xml configuration logs are newer than the config
3239 logs stored on the MDS and report an error if this is the case.
3240 Request --write-conf, or allow starting with --old_conf.
3242 Severity : enhancement
3244 Description: Handle symlinks in the path when checking if Lustre is mounted.
3245 Details : Resolve intermediate symlinks when checking if a client has
3246 mounted a filesystem to avoid duplicate client mounts.
3251 Description: lconf can hit an error exception but still return success.
3252 Details : The lconf command catches the Command error exception at the top
3253 level script context and will exit with the associated exit
3254 status, but doesn't ensure that this exit status is non-zero.
3259 Description: failure of ptlrpc thread startup can cause oops
3260 Details : Starting a ptlrpc service thread can fail if there are a large
3261 number of threads or the server memory is very fragmented.
3262 Handle this without oopsing.
3265 Frequency : always, only if liblustre and non-default acceptor port was used
3267 Description: liblustre cannot connect to servers with non-default acceptor port
3268 Details : tcpnal_set_default_params() was not called and was therefore
3269 ignoring the environment varaible TCPNAL_PORT, as well as other
3270 TCPNAL_ environment variables
3275 Description: two objects could be created on the same OST for a single file
3276 Details : If an OST is down, in some cases it was possible to create two
3277 objects on a single OST for a single file. No problems other
3278 than potential performance impact and spurious error messages.
3282 Bugzilla : 5681, 9562
3283 Description: Client may oops in ll_unhash_aliases
3284 Details : Client dcache may become inconsistent in race condition.
3285 In some cases "getcwd" can fail if the current directory is
3291 Description: Inode refcounting problems in NFS export code
3292 Details : link_raw functions used to call d_instantiate without obtaining
3293 extra inode reference first.
3297 Bugzilla : 9942, 9903
3298 Description: Referencing freed requests leading to crash, memleaks with NFS.
3299 Details : We used to require that call to ll_revalidate_it was always
3300 followed by ll_lookup_it. Also with revalidate_special() it is
3301 possible to call ll_revalidate_it() twice for the same dentry
3302 even if first occurence returned success. This fix changes semantic
3303 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3304 reference on a request referred from the intent.
3305 ll_intent_release() then releases such a request.
3308 Frequency : rare, normally benchmark loads only
3310 Description: unlinked inodes were kept in memory on the client
3311 Details : If a client is repeatedly creating and unlinking files it
3312 can accumulate a lot of stale inodes in the inode slab cache.
3313 If there is no other client load running this can cause the
3314 client node to run out of memory. Instead flush old inodes
3315 from client cache that have the same inode number as a new inode.
3318 Frequency : SLES9 2.6.5 kernel and long filenames only
3319 Bugzilla : 9969, 10379
3320 Description: utime reports stale NFS file handle
3321 Details : SLES9 uses out-of-dentry names in some cases, which confused
3322 the lustre dentry revalidation. Change it to always use the
3326 Frequency : rare, unless heavy write-truncate concurrency is continuous
3327 Bugzilla : 4180, 6984, 7171, 9963, 9331
3328 Description: OST becomes very slow and/or deadlocked during object unlink
3329 Details : filter_destroy() was holding onto the parent directory lock
3330 while truncating+unlinking objects. For very large objects this
3331 may block other threads for a long time and slow overall OST
3332 responsiveness. It may also be possible to get a lock ordering
3333 deadlock in this case, or run out of journal credits because of
3334 the combined truncate+unlink. Solution is to do object truncate
3335 first in one transaction without parent lock, and then do the
3336 final unlink in a new transaction with the parent lock. This
3337 reduces the lock hold time dramatically.
3340 Frequency : rare, 2.4 kernels only
3342 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3343 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3344 before unmounting their filesystem in order to avoid deadlock.
3345 The kernel_locked() function in 2.4 kernels only checks whether
3346 the kernel lock is held, not whether it is this process that is
3347 holding it as 2.6 kernels do.
3352 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3353 Details : The client ptlrpc code may be trying to reconnect to a down
3354 server before a previous connection attempt has timed out.
3355 Increase the reconnect interval to be longer than the connection
3356 timeout interval to avoid sending duplicate connections to
3360 Frequency : echo_client brw_test command
3362 Description: fix echo_client to work with OST preallocated code
3363 Details : OST preallocation code (5137) didn't take echo_client IO path
3364 into account: echo_client calls filter methods outside of any
3365 OST thread and, hence, there is no per-thread preallocated
3366 pages and buffers to use. Solution: hijack pga pages for IO. As
3367 a byproduct, this avoids unnecessary data copying.
3371 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3372 Description: Client can oops in mdc_commit_close() after open replay
3373 Details : It was possible for the MDS to return an open request with no
3374 transaction number in mds_finish_transno() if the client was
3375 evicted, but without actually returning an error. Clients
3376 would later try to replay that open and may trip an assertion
3377 Simplify the client close codepath, and always return an error
3378 from the MDS in case the open is not successful.
3381 Frequency : rare, 2.6 OSTs only
3383 Description: OST may deadlock under high load on fragmented files
3384 Details : If there was a heavy load and highly-fragmented OST filesystems
3385 it was possible to have all the OST threads deadlock waiting on
3386 allocation of biovecs, because the biovecs were not released
3387 until the entire RPC IO was completed. Instead, release biovecs
3388 as soon as they are complete to ensure forward IO progress.
3390 Severity : enhancement
3392 Description: Support for specifying external journal device at mount
3393 Details : If an OST or MDS device is formatted with an external journal
3394 device, this device major/minor is stored in the ext3 superblock
3395 and may not be valid for failover. Allow detecting and
3396 specifying the external journal at mount time.
3401 Description: Mounting an MDS with pending unlinked files may cause oops
3402 Details : target_finish_recovery() calls mds_postrecov() which returned
3403 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3404 considers this an error and immediately begins cleaning up the
3405 lov, just after starting the mds_lov process
3407 Severity : enhancement
3409 Description: Implement 'lfs df' to report actual free space on per-OST basis
3410 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3411 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3412 usage of objects; '-h' to report in human readable format.
3414 ------------------------------------------------------------------------------
3416 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3423 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3424 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3425 at startup during orphan unlinking. Instead, skip buddy update
3426 before setup, as it will be regenerated after recovery is complete.
3431 Description: If an OST is inactive, its locks might reference stale inodes.
3432 Details : lov_change_cbdata() must iterate over all namespaces, even if
3433 they are inactive to clear inode references from the lock.
3435 Severity : enhancement
3436 Frequency : occasional, if non-standard max_dirty_mb used
3438 Description: Client will block write RPCs if not enough grant
3439 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3440 then the client will block writes while waiting for another RPC
3441 to complete instead of consuming its dirty limit. With change
3442 we get improved performance when max_dirty_mb is small.
3444 Severity : enhancement
3445 Bugzilla : 3389, 6253
3446 Description: Add support for supplementary groups on the MDS.
3447 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3448 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3449 lookups for user supplementary groups into a cache.
3453 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3454 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3455 for liblustre O_EXCL is enforced by MDS.
3459 Description: readdir never returns NULL in liblustre.
3460 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3461 dentry in struct dirent.
3465 Frequency : liblustre only, depends on application IO pattern
3466 Description: liblustre clients evicted if not contacting servers
3467 Details : Don't put liblustre clients into the ping_evictor list, so
3468 they will not be evicted by the pinger ever.
3470 Severity : enhancement
3472 Description: Add ability to evict clients by NID from MDS.
3473 Details : By echoing "nid:$NID" string into
3474 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3475 $NID would be instantly evicted from this MDS and from all active
3476 OSTs connected to it.
3480 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3481 Details : lookup_it_finish() used to query file size from OSTs that was not
3486 Description: service threads change working directory to that of init
3487 Details : Starting lustre service threads may pin the working directory
3488 of the parent thread, making that filesystem busy. Threads
3489 now change to the working directory of init to avoid this.
3493 Frequency : during shutdown only
3494 Description: shutdown with a failed MDS or OST can cause unmount to hang
3495 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3500 Frequency : 2.6 only
3501 Description: chmod/chown may include an extra supplementary group
3502 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3503 supplementary group and if none is specified this is used.
3506 Bugzilla : 5479 (6816)
3508 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3509 Details : If lookup is in progress on a file that is unlinked we might try
3510 to revalidate the inode and fail in revalidate after lookup is
3511 complete and ll_file_open() enqueues the open again but
3512 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3515 Frequency : always, if lconf --abort_recovery used
3517 Description: lconf --abort_recovery fails with 'Operation not supported'
3518 Details : lconf was attempting to abort recovery on the MDT device and not
3521 ------------------------------------------------------------------------------
3523 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3528 Frequency : rare (only unsupported configurations with a node running as an
3530 Bugzilla : 6514, 5137
3531 Description: Mounting a Lustre file system on a node running as an OST could
3533 Details : OSTs now preallocates memory needed to write out data at
3534 startup, instead of when needed, to avoid having to
3535 allocate memory in possibly low memory situations.
3536 Specifically, if the file system is mounted on on OST,
3537 memory pressure could force it to try to write out data,
3538 which it needed to allocate memory to do. Due to the low
3539 memory, it would be unable to do so and the node would
3540 become unresponsive.
3542 Severity : enhancement
3544 Description: Addition of lconf --service command line option
3545 Details : lconf now accepts a '--service <arg>' option, which is
3546 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3548 Severity : enhancement
3550 Description: Failover mode is now the default for OSTs.
3551 Details : By default, OSTs will now run in failover mode. To return to
3552 the old behaviour, add '--failout' to the lmc line for OSTs.
3554 Severity : enhancement
3556 Description: Health checks are now provided for MDS and OSTs
3557 Details : Additional detailed health check information on MSD and OSTs
3558 is now provided through the procfs health_check value.
3561 Frequency : occasional, depends on IO load
3563 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3564 after numerous create/delete cycles
3565 Details : The ext3 inode allocation policy would not allocate new inodes
3566 very well on the OSTs because there are no new directories
3567 being created. Instead we look for groups with free space if
3568 the parent directories are nearly full.
3573 Description: Network or server problems during mount may cause partially
3574 mounted clients instead of returning an error.
3575 Details : The config llog parsing code may overwrite the error return
3576 code during mount error handling, returning success instead
3582 Description: MDS can fail to allocate large reply buffers
3583 Details : After long uptimes the MDS can fail to allocate large reply
3584 buffers (e.g. zconf client mount config records) due to memory
3585 fragmentation or consumption by the buffer cache. Preallocate
3586 some large reply buffers so that these replies can be sent even
3587 under memory pressure.
3591 Frequency : rare (liblustre)
3592 Description: fsx running with liblustre complained that using truncate() to
3593 extend the file doesn't work. This patch corrects that issue.
3594 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3595 ATTR_SIZE and lsm use in llu_setattr_raw.
3599 Frequency : rare, only 2.6 kernels
3600 Description: Unusual file access patterns on the MDS may result in inode
3601 data being lost in very rare circumstances.
3602 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3603 code in the 2.6 kernel caused the inode and/or EA data not to
3604 be read from disk, causing single-file corruption.
3608 Frequency : rare, only 2.6 filesystems using extents
3609 Description: Heavy concurrent write and delete load may cause data corruption.
3610 Details : It was possible under high-load situations to have an extent
3611 metadata block in the block device cache from a just-unlinked
3612 file overwrite a newly-allocated data block. We now unmap any
3613 metadata buffers that alias just-allocated data blocks.
3617 Frequency : filesystems with default stripe_count larger than 77
3618 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3619 Details : lconf specifies an inode size of 4096 bytes when the default
3620 stripe_count is larger than 77. This conflicts with the default
3621 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3622 this case to avoid pinning too much memory for large EAs.
3624 ------------------------------------------------------------------------------
3626 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3631 Frequency : rare (extremely heavy IO load with hundreds of clients)
3633 Description: Client is evicted, gets IO error writing to file
3634 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3635 caused clients to be evicted for AST timeouts. The fixes in
3636 bug 5192 mean we no longer need to have such short AST timeouts
3637 so ldlm_timeout has been increased.
3640 Frequency : occasional during --force or --failover shutdown under load
3641 Bugzilla : 5949, 4834
3642 Description: Server oops/LBUG if stopped with --force or --failover under load
3643 Details : a collection of import/export refcount and cleanup ordering
3644 issues fixed for safer force cleanup
3647 Frequency : only filesystems larger than 120 OSTs
3648 Bugzilla : 5990, 6223
3649 Description: lfs getstripe would oops on a very large filesystem
3650 Details : lov_getconfig used kfree on vmalloc'd memory
3653 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3654 Bugzilla : 6242, 6243
3655 Description: reading from files that had been truncated to a non-zero size
3656 but never opened returned no data
3657 Details : ll_file_read() reads zeros from no-object files to EOF
3662 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3663 Details : There's little harm in aborting MDS/OSS recovery and letting it
3664 try again, so I removed the LASSERT and return an error instead.
3666 Severity : enhancement
3668 Description: New debugging infrastructure for tracking down data corruption
3669 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3670 (b) cover more of the client-side code path, and (c) try to narrow
3671 down where problems occurred
3675 Bugzilla : 3819, 4364, 4397, 6313
3676 Description: Racing close and eviction MDS could cause assertion in mds_close
3677 Details : It was possible to get multiple mfd references during close and
3678 client eviction, leading to one thread referencing a freed mfd.
3680 Severity: : enhancement
3681 Bugzilla : 3262, 6359
3682 Description: Attempts to reconnect to servers are now more aggressive.
3683 Details : This builds on the enhanced upcall-less recovery that was added
3684 in 1.4.2. When trying to reconnect to servers, clients will
3685 now try each server in the failover group every 10 seconds. By
3686 default, clients would previously try one server every 25 seconds.
3691 Description: After recovery, certain operations trigger a failed
3692 assertion on a client.
3693 Details : Failing over an mds, using lconf -d --failover, while a
3694 client was doing a readdir() call would cause the client to
3695 LBUG after recovery completed and the readdir() was resent.
3697 Severity : enhancement
3699 Description: Default groups are now added by lconf
3700 Details : You can now run lconf --group <servicename> without having to
3701 manually add groups with lmc.
3704 Frequency : occasional
3706 Description: Nodes with an elan id of 0 trigger a failed assertion
3709 Frequency : always when accessing e.g. tty/console device nodes
3711 Description: tty and some other devices nodes cannot be used on lustre
3712 Details : file's private_data field is used by device data and lustre
3713 values in there got lost. New field was added to struct file to
3714 store fs-specific private data.
3717 Frequency : when exporting Lustre via NFS
3719 Description: NFSD failed occasionally when looking up a path component
3720 Details : NFSD is looking up ".." which was broken in ext3 directories
3721 that had grown large enough to become hashed.
3724 Frequency : Clusters with multiple interfaces not on the same subnet
3726 Description: Nodes will repeatedly try to reconnect to an interface which it
3727 cannot reach and report an error to the log.
3728 Details : Extra peer list entries will be created by lconf with some peers
3729 unreachable. lconf now validates the peer before adding it.
3732 Frequency : Only if a default stripe is set on the filesystem root.
3734 Description: Setting a default stripe on the filesystem root prevented the
3735 filesystem from being remounted.
3736 Details : The client was sending extra request flags in the root getattr
3737 request and did not allocate a reply buffer for the dir EA.
3740 Frequency : occasional, higher if lots of files are accessed by one client
3741 Bugzilla : 6159, 6097
3742 Description: Client trips assertion regarding lsm mismatch/magic
3743 Details : While revalidating inodes the VFS looks up inodes with ifind()
3744 and in rare cases can find an inode that is being freed.
3745 The ll_test_inode() code will free the lsm during ifind()
3746 when it finds an existing inode and then the VFS later attaches
3747 this free lsm to a new inode.
3751 Bugzilla : 6422, 7030
3752 Description: MDS deadlock between mkdir and client eviction
3753 Details : Creating a new file via mkdir or mknod (starting a transaction
3754 and getting the ns lock) can deadlock with client eviction
3755 (gets ns lock and trying to finish a synchronous transaction).
3758 Frequency : occasional
3759 Description: While starting a server, the fsfilt_ext3 module could not be
3761 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3762 kernels. Previously, lconf would still use the ext3 name
3763 when trying to load modules. Now, it will correctly use
3764 ext3 on 2.4 and ldiskfs on 2.6.
3766 Severity : enhancement
3767 Description: The default stripe count has been changed to 1
3768 Details : The interpretation of the default stripe count (0, to lfs
3769 or lmc) has been changed to mean striping across a single
3770 OST, rather than all available. For general usage we have
3771 found a stripe count of 1 or 2 works best.
3773 Severity : enhancement
3774 Description: Add support for compiling against Cray portals.
3775 Details : Conditional compiling for some areas that are different
3779 Frequency : occasional
3780 Bugzilla : 6409, 6834
3781 Description: Creating files with an explicit stripe count may lead to
3782 a failed assertion on the MDS
3783 Details : If some OSTs are full or unavailable, creating files may
3784 trigger a failed assertion on the MDS. Now, Lustre will
3785 try to use other servers or return an error to the
3789 Frequency : occasional
3791 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3792 caused lock timeout during readahead (since 1.4.2).
3793 Details : Processes doing readahead might match a lock that hasn't been
3794 granted yet if there are overlapping and conflicting lock
3795 requests. The readahead process waits on ungranted lock
3796 (original lock is CBPENDING), while OST waits for that process
3797 to cancel CBPENDING read lock and eventually evicts client.
3799 Severity : enhancement
3801 Description: Initial enabling of flock support for clients
3802 Details : Implements fcntl advisory locking and file status functions.
3803 This feature is provided as an optional mount flag (default
3804 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3805 locking are implemented yet, and those that are are not guaranteed
3806 to be completely correct in production environments.
3807 mount -t lustre -o [flock|noflock] ...
3810 Frequency : occasional
3812 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3813 assertion in the ext3 JBD (journaling) layer.
3814 Details : The b_committed_data struct is protected by the big kernel lock
3815 in 2.4 kernels, serializing journal_commit_transaction() and
3816 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3817 access to this struct. In 2.6 kernels there is finer grained
3818 locking to improve SMP performance of the JBD layer.
3822 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3827 Description: merge and cleanup kernel patches.
3830 -----------------------------------------------------------------------------
3832 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3834 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3835 versions of Lustre. Please update all servers and clients to
3836 version 1.4.2 or later at the same time. You must also run
3837 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3838 to update the configuration logs.
3840 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3841 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3842 - swab llog records of type '0' so we get proper header size/idx (5861)
3843 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3844 - fix rename of one directory over another leaking an inode (5953)
3845 - avoid SetPageDirty on 2.6 (5981)
3846 - don't re-add just-being-destroyed locks to the waiting list (5653)
3847 - when creating new directories, inherit the parent's custom
3848 striping settings if present parent (3048)
3849 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3850 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3851 - don't hold client locks on temporary worklist from l_lru (5666)
3852 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3853 - automatically evict dead clients (5921)
3854 - Update file size properly in create+truncate+fstat case (6196)
3855 - Do not unhash mountpoint dentries, do not allow removal of
3857 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3858 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3859 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3860 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3861 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3862 - Correctly handle reads of files with no objects (6243)
3863 - lctl recover will also mark a device active if deactivate used (5933)
3865 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3866 - allow --write-conf on an MDS with different nettype than client (5619)
3867 - don't write config llogs to MDS for mounts not from that MDS (5617)
3868 - lconf should create multiple TCP connections from a client (5201)
3869 - init scripts are now turned off by default; run chkconfig --on
3870 lustre and chkconfig --on lustrefs to use them
3871 - upcalls are no longer needed for clients to recover to failover
3873 - add --abort-recovery option to lconf to abort recovery on device
3875 - add support for an arbitrary number of OSTs (3026)
3876 - Quota support protocol changes.
3877 - forward compatibility changes to wire structs (6007)
3878 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3879 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3880 - improved "lustre status" script
3881 - initialize blocksize for non-regular files (6062)
3882 - added --disable-server and --disable-client configure options (5782)
3883 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3884 - Vanilla 2.4.29 support
3885 - increase maximum number of obd devices to 520 (6242)
3886 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3887 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3888 SLES 9 2.6.5 kernels
3889 - limit stripes per file to 160 (the maximum EA size) (6093)
3891 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3894 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3895 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3896 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3897 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3898 - small ext3 extents cleanups and fixes (5733)
3899 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3900 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3901 - avoid needless client->OST connect, fix handle mismatch (5317)
3902 - fix DLM error path that led to out-of-sync client, long delays (5779)
3903 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3904 - fix several locking issues related to i_size (5492,5624,5654,5672)
3905 - don't move pending lock onto export if it is already evicted (5683)
3906 - fix kernel oops when creating .foo in unlinked directory (5548)
3907 - fix deadlock in obdfilter statistics vs. object create (5811)
3908 - use time_{before,after} to avoid timer jiffies wrap (5882)
3909 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3910 - Do not leak request if server was not able to process it (5154)
3911 - If mds_open unable to find parent dir, make that negative lookup(5154)
3912 - don't create new directories with extent-mapping (5909, 5936)
3914 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3915 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3916 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3917 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3918 - support for RHEL4 kernel on i686 (5773)
3919 - provide error messages when incompatible logs are encountered (5898)
3921 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3922 * version 1.4.0.10 (1.4.1 release candidate 1)
3924 - don't keep a lock reference when lock is not granted (4238)
3925 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3926 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3927 - drop import inflight refcount on signal_completed_replay error (5255)
3928 - unlock page after async write error during send (3677)
3929 - handle missing objects in filter_preprw_read properly (5265)
3930 - no transno return for symlink open, don't save no-trasno open (3440)
3931 - don't try to complete elan receive that already failed (4012)
3932 - free RPC server reply state on error (5406)
3933 - clean up thread from ptlrpc_start_thread() on error (5160)
3934 - readahead could read extra page into cache that wasn't ejected (5388)
3935 - prevent races in class_attach/setup/cleanup/detach (5260)
3936 - don't dereference de->d_inode after l_dput of de (5458)
3937 - use "int" for stripe value returned from lock_to_stripe (5544)
3938 - mballoc allocation and error-checking fixes in 2.6 (5504)
3939 - block device patches to fix I/O request sizes in 2.6 (5482)
3940 - look up hostnames for IB nals (5602)
3941 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3942 - don't start multiple acceptors for the same port (5277)
3943 - fix incorrect LASSERT in mds_getattr_name (5635)
3944 - export a proc file for general "ping" checking (5628)
3945 - fix "lfs check" to not block when the MDS is down (5628)
3947 - service request history (4965)
3948 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3949 - create an "evict_client" /proc entry on OSTs, like the MDS has
3950 - fix mount usage message, return errors per mount(8) (5168)
3951 - change grep [] to grep "[]" in tests so they work in more UMLs
3952 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3953 - remove ancient LOV_MAGIC_V0 EA support (5047)
3954 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3955 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3956 - allow readahead allocations to fail when low on memory (5383)
3957 - mmap locking landed again, after considerable improvement (2828)
3958 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3960 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3963 - send OST transaction number in read/write reply to free req (4966)
3964 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3965 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3966 - the watchdog thread now runs as interruptible (5246)
3967 - flock/lockf fixes (but it's still disabled, pending 5135)
3968 - don't use EXT3 constants in llite code (5094)
3969 - memory shortage at startup could cause assertion (5176)
3971 - reorganization of lov code
3972 - single portals codebase
3974 - add extents/mballoc support (5025)
3975 - direct I/O reads in the obdfilter (4048)
3976 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3978 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3981 - send OST transaction number in read/write reply to free req (4966)
3982 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3983 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3984 - the watchdog thread now runs as interruptible (5246)
3985 - handle missing objects in filter_preprw_read properly (5265)
3986 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3987 - drop import inflight refcount on signal_completed_replay error (5255)
3988 - unlock page after async write error during send (3677)
3989 - return original error code on reconstructed replies (3761)
3990 - no transno return for symlink open, don't save no-trasno open (3440)
3992 - add pid to ldlm debugging output (4922)
3993 - bump the watchdog timeouts -- we can't handle 30sec yet
3994 - extra debugging for orphan dentry/inode bug (5259)
3996 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3999 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4000 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4001 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4002 - fix NULL dereference in /proc/sys/portals/routes (4827)
4003 - allow failed mdc_close() operations to be interrupted (4561)
4004 - stop precreate on OST before MDS would time out on it (4778)
4005 - don't send partial-page writes before EOF from client (4410)
4006 - discard client grant for sub-page writes on large-page clients (4520)
4007 - don't free dentries not owned by NFS code, check generation (4806)
4008 - fix lsm leak if mds_create_objects() fails (4801)
4009 - limit debug_daemon file size, always print CERROR messages (4789)
4010 - use transno after validating reply (3892)
4011 - process timed out requests if import state changes (3754)
4012 - update mtime on OST during writes, return in glimpse (4829)
4013 - add mkfsoptions to LDAP (4679)
4014 - use ->max_readahead method instead of zapping global ra (5039)
4015 - don't interrupt __l_wait_event() during strace
4017 - add software watchdogs to catch hung threads quickly (4941)
4018 - make lustrefs init script start after nfs is mounted
4019 - fix CWARN/ERROR duplication (4930)
4020 - return async write errors to application if possible (2248)
4021 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4022 - print NAL number in %x format (4645)
4023 - update barely-supported suse-2.4.21-171 series (4842)
4024 - support for sles 9 %post scripts
4025 - support for building 2.6 kernel-source packages
4026 - support for sles km_* packages
4028 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4031 - ignore -ENOENT errors in osc_destroy (3639)
4032 - notify osc create thread that OSC is being cleaned up (4600)
4033 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4034 - reconstruct ost_handle() like mds_handle() (4657)
4035 - create a new thread to do import eviction to avoid deadlock (3969)
4036 - let lconf resolve symlinked-to devices (4629)
4037 - don't unlink "objects" from directory with default EA (4554)
4038 - hold socknal file ref over connect in case target is down (4394)
4039 - allow more than 32000 subdirectories in a single directory (3244)
4040 - fix blocks count for O_DIRECT writes (3751)
4041 - OST returns ENOSPC from object create when no space left (4539)
4042 - don't send truncate RPC if file size isn't changing (4410)
4043 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4044 - bind to privileged port in socknal and tcpnal (3689)
4046 - rate limit CERROR/CWARN console message to avoid overload (4519)
4047 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4048 - basic mmap support (3918)
4049 - kernel patch series update from b1_4 (4711)
4051 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4054 - avoid crash during MDS cleanup with OST shut down (2775)
4055 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4056 - don't use bad inodes on the MDS (3744)
4057 - dynamic object preallocation to improve recovery speed (4236)
4058 - don't hold spinlock over lock dumping or change debug flags (4401)
4059 - don't zero obd_dev when it is force cleaned (3651)
4060 - print grants to console if they go negative (4431)
4061 - "lctl deactivate" will stop automatic recovery attempts (3406)
4062 - look for existing locks in ldlm_handle_enqueue() (3764)
4063 - don't resolve lock handle twice in recovery avoiding race (4401)
4064 - revalidate should check working dir is a directory (4134)
4066 - don't always mark "slow" obdfilter messages as errors (4418)
4068 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4071 - don't close LustreDB during write_conf until it is done (3860)
4072 - fix typo in lconf for_each_profile (3821)
4073 - allow dumping logs from multiple threads at one time (3820)
4074 - don't allow multiple threads in OSC recovery (3812)
4075 - fix debug_size parameters (3864)
4076 - fix mds_postrecov to initialize import for llog ctxt (3121)
4077 - replace config semaphore with spinlock (3306)
4078 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4079 - don't allow enqueue to complete on a destroyed export (3822)
4080 - down write_lock before checking llog header bitmap (3825)
4081 - recover from lock replay timeout (3764)
4082 - up llog sem before sending rpc (3652)
4083 - reduce ns lock hold times when setting kms (3267)
4084 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4085 - fix NULL deref and obd_dev leak on setup error (3312)
4086 - replace some LBUG about llog ops with error handling (3841)
4087 - don't match INVALID dentries from d_lookup and spin (3784)
4088 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4089 - fix invalid assertion in ptlrpc_set_wait (3880)
4091 - add libwrap support for the TCP acceptor (3996)
4092 - add /proc/sys/portals/routes for non-root route listing (3994)
4093 - allow setting MDS UUID in .xml (2580)
4094 - print the stack of a process that LBUGs (4228)
4096 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4099 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4100 - make sure to unset replay flag from failed open requests (3440)
4101 - if default stripe count is 0, use OST count for inode size (3636)
4102 - update parent mtime/ctime on client for create/unlink (2611)
4103 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4104 - free recovery state on server during a forced cleanup (3571)
4105 - unregister_reply for resent reqs (3063)
4106 - loop back devices mounting and status check on 2.6 (3563)
4107 - fix resource-creation race that can provoke i_size == 0 (3513)
4108 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4109 - more debugging for page-accounting assertion (3746)
4110 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4111 - avoid deadlock after precreation failure (3758)
4112 - fix race and lock order deadlock in orphan handling (3450, 3750)
4113 - add validity checks when grabbing inodes from l_ast_data (3599)
4115 - add /proc/.../recovery_status to obdfilter (3428)
4116 - lightweight CDEBUG infrastructure, debug daemon (3668)
4117 - change default OSC RPC parameters to be better on small clusters
4118 - turn off OST read cache for files smaller than 32MB
4119 - install man pages and include them in rpms (3100)
4120 - add new init script for (un)mounting lustre filesystems (2593)
4121 - run chkconfig in %post for init scripts (3701)
4122 - drop scimac NAL (unmaintained)
4124 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4127 - clean kiobufs before and after use (3485)
4128 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4129 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4130 - change init script to fail more clearly if not run as root (1528)
4131 - allow clients to reconnect during replay (1742)
4132 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4133 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4134 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4135 - fix cond_resched (3554)
4136 - only evict unfinished clients after recovery (3515)
4137 - allow bulk resend, prevent data loss (3570)
4138 - dynamic ptlrpc request buffer allocation (2102)
4139 - don't allow unlinking open directory if it isn't empty (2904)
4140 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4141 - remove extraneous obd dereference causing LASSERT failure (3334)
4142 - don't use get_cycles() when creating temp. files on the mds (3156)
4143 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4144 - handle EEXIST for set-stripe, set proper directory name (3336)
4146 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4147 - fix ksocknal_fmb_callback() error messages (2918)
4149 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4152 - don't copy lvb into (possibly NULL) reply on error (2983)
4153 - don't deref dentry after dput, don't free lvb on error (2922)
4154 - use the kms to determine writeback rpc length (2947)
4155 - increment oti_logcookies when osc is inactive (2948)
4156 - update client's i_blocks count via lvb messages (2543)
4157 - handle intent open/close of special files properly (1557)
4158 - mount MDS with errors=remount-ro, like obdfilter (2009)
4159 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4160 - don't use cancelling-locks' kms values (2947)
4161 - use highest lock extent for kms, not last one (2925)
4162 - don't dereference ERR_PTR() dentry in error handling path (3107)
4163 - fix thread race in portals_debug_dumplog() (3122)
4164 - create lprocfs device entries at setup instead of at attach (1519)
4165 - common AST error handler, don't evict client on completion race (3145)
4166 - zero nameidata in detach_mnt in 2.6 (3118)
4167 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4168 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4169 - initialize RPC timeout timer earlier for 2.6 (3219)
4170 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4171 - print nal/nid for unknown nid (3258)
4172 - additional checks for oscc recovery before doing precreate (3284)
4173 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4174 - don't crash in mdc_close for bad permissions on open (3285)
4175 - zero i_rdev for non-device files (3147)
4176 - clear page->private before handing to FS, better assertion (3119)
4177 - tune the read pipeline (3236)
4178 - fix incorrect decref of invalidated dentry (2350)
4179 - provide read-ahead stats and refine rpc in flight stats (3328)
4180 - don't hold journal transaction open across create RPC (3313)
4181 - update atime on MDS at close time (3265)
4182 - close LDAP connection when recovering to avoid server load (3315)
4183 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4184 - don't leak open file on MDS after open resend (3325)
4185 - serialize filter_precreate and filter_destroy_precreated (3329)
4186 - loop device shouldn't call sync_dev() for nul device (3092)
4187 - clear page cache after eviction (2766)
4188 - resynchronize MDS->OST in background (2824)
4189 - refuse to mount the same filesystem twice on same mountpoint (3394)
4190 - allow llmount to create routes for mounting behind routers (3320)
4191 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4192 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4193 - fix rare problem with rename on htree directories (3417)
4195 - allow default OST striping configuration per directory (1414)
4196 - fix compilation for qswnal for 2.6 kernels (3125)
4197 - increase maximum number of MDS request buffers for large systems
4198 - change liblustreapi to be useful for external progs like lfsck (3098)
4199 - increase local configuration timeout for slow disks (3353)
4200 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4202 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4205 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4206 - don't overwrite extent policy data in reply if lock was blocked (2901)
4207 - drop filter export grants atomically with removal from device (2663)
4208 - del obd_self_export from work_list in class_disconnect_exports (2908)
4209 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4210 - swab reply message in mdc_close, other PPC fixes (2464)
4211 - fix destroying of named logs (2325)
4212 - overwrite old logs when running lconf --write_conf (2264)
4213 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4214 - fix race in target_handle_connect (2898)
4215 - mds_reint_create() should take same inode create lock (2926)
4216 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4217 - don't close files for self_export to avoid uninitialized obd (2936)
4218 - allow MDS with the same name as client node (2939)
4219 - hold dentry reference for closed log files for unlink (2325)
4220 - reserve space for all logs during transactions (2059)
4221 - don't evict page beyond end of stripe extent (2925)
4222 - don't oops on a deleted current working directory (2399)
4223 - handle hard links to targets without a parent properly (2517)
4224 - don't dereference NULL lock when racing during eviction (2867)
4225 - don't grow lock extents when lots of conflicting locks (2919)
4227 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4230 - account for cache space usage on clients to avoid data loss (974)
4231 - lfsck support in lustre kernel code (2349)
4232 - reduce journal credits needed for BRW writes (2370)
4233 - orphan handling to avoid losing space on client/server crashes
4234 - ptlrpcd can be blocked, stopping ALL progress (2477)
4235 - use lock value blocks to assist in proper KMS, faster stat (1021)
4236 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4237 - recovery for initial connections (2355)
4238 - fixes for mds_cleanup_orphans (1934)
4239 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4240 - block all file creations until orphan recovery completes (1901)
4241 - client remove rq_connection from request struct (2423)
4242 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4243 - recovery timer race (2670)
4244 - mdc_close recovey bug (2532)
4245 - ptlrpc cleanup bug (2710)
4246 - mds timeout on local locks (2588)
4247 - namespace lock held during RPCs (2431)
4248 - handle interrupted sync write properly (2503)
4249 - don't try to handle a message that hasn't been replied to (2699)
4250 - client assert failure during cleanup after abort recovery (2701)
4251 - leak mdc device after failed mount (2712)
4252 - ptlrpc_check_set allows timedout requests to complete (2714)
4253 - wait for inflight reqs when ptlrpcd finishes (2710)
4254 - make sure unregistered services are removed from the srv_list
4255 - reset bulk XID's when resending them (caught by 1138 test)
4256 - unregister_bulk after timeout
4257 - fix lconf error (2694)
4258 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4259 - readahead locks pages, leaves pending causing memory pressure (2673)
4260 - increase OST request buffers to 4096 on large machines (2729)
4261 - fix up permission of existing directories in simple_mkdir (2661)
4262 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4263 - don't assign transno to errored transactions (2742)
4264 - don't delete objects on OST if given a bogus objid from MDS (2751)
4265 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4266 - if rq_no_resend, then timeout request after recovery (2432)
4267 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4268 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4269 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4270 - MDS recovery shouldn't delete objects that it has given out (2730)
4271 - if enqueue arrives after completion, don't clobber LVB (2819)
4272 - don't unlock pages twice when trigger_group_io returns error (2814)
4273 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4274 - don't write pages to disk if there was an error (1450)
4275 - don't ping imports that have recovery disabled (2676)
4276 - take buffered bytes into account when balancing socknal conn (2817)
4277 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4278 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4279 - remove little-endian swabbing of llog records (1987)
4280 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4281 - retry reposting request buffers if they fail (1191)
4282 - grow extent at grant time to avoid granting a revoked lock (2809)
4283 - lock revoke doesn't evict page if covered by a second lock (2765)
4284 - disable VM readahead to avoid reading outside lock extents (2805)
4286 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4287 - updated kernel patches for hp-2.4.20 kernel (2681)
4289 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4292 - fix truncated write corruption (2366)
4293 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4295 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4296 - ensure len(uuid) < 37 in lmc (1171)
4297 - fix ia64 OOPS in llog_test (2255)
4298 - zero end of page at obdfilter for partial page writes (2648)
4299 - don't leave stale dentries around after renames (bug 2428)
4300 - fix timeouts when evicting a client with a single lock held (2642)
4301 - set deadline for the initial HELLO message to drain (2634)
4302 - print out dotted-quad IP addresses in the socknal (2302)
4304 - additional debugging for MDS client eviction problem (2443)
4305 - fix mkfsoptions support for osts (2603, 2604)
4307 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4310 - add series for the vanilla 2.6.0 kernel
4311 - add series for the vanilla 2.4.24 kernel
4312 - add series for a cray x86/64 UL kernel drop
4313 - fix xattr patches for the vanilla 2.4.19 series
4315 - generate true UUIDs in lmc (1171)
4316 - have portals stack dumping break in UML (2466)
4317 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4318 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4319 - wake callback waiting threads on client eviction (2460)
4320 - Add --ptldebug and --subsystem to lmc (1719)
4321 - update assertion to allow safe interrupt allocation
4322 - set rq_no_resend for cancel requests (2432)
4323 - recalculate ptlrpcd timeout after resend (2494)
4324 - call vfs_rmdir when removing pending directories (2368)
4325 - fix renaming a file to itself (2429)
4326 - lmc creates a default one-stripe lov (2454)
4327 - expand procfs space to handle large clusters (2326)
4328 - increase UML stack to avoid overflow
4329 - update lconf's list of debug and subsystem masks
4330 - fix lfs find --obd (2510)
4331 - /proc tunable for disabling filter read caching (2591)
4332 - stop rpm packages from altering slapd.conf (2301)
4333 - disable nagle in the socknal under 0conf (2578)
4334 - choose mds inode size based on stripe count (2572)
4335 - fix kernel-source rpm problems (2516)
4337 - add --disable-doc to avoid pdf generation (2421)
4338 - update documentation, tests, type-os, comments
4339 - avoid format warnings on ia64
4340 - remove the TOE NAL
4341 - tiny code cleanups by removing unused fields
4343 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4346 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4347 - remove the most blatant lies from BUILDING file (2371)
4348 - change default debug level to reasonable production setting
4349 - reduce client side cache size to reduce cache flush time
4350 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4351 - make TCP zerocopy and pinger support enabled by default (2476)
4352 - sync writes completed after process exits caused crashes (2319)
4353 - maintain correct mount count on the MDS (2356)
4354 - backout 1557, because 2316 wasn't really fixed
4355 - better file I/O statistics gathering in /proc
4356 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4357 - another kernel patch to fix zero-copy TCP function export
4358 - don't take duplicate lock when processing re-sent getattr (2420)
4359 - lctl uses obd_self_export instead of creating new conn (2353)
4360 - MDS/OST recovery case which requires object creation asserted (2425)
4361 - move lfs from /usr/sbin to /usr/bin in packages
4362 - fix race between mds_client_add and mds_client_free (2417)
4363 - use kmalloc instead of slabs in portals (2430)
4364 - don't create duplicate records when a failover MDS is present (2442)
4365 - remove unnecessary mount age check (2332)
4366 - don't remove directory inodes from locks prematurely (2451)
4367 - don't break if MDS service name is the same as hostname (2103)
4368 - fix races in client write RPC generation when cache full (2482)
4370 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4373 - remove now-unused request->rq_obd (278)
4374 - if an allocation fails, print out how much memory we've used (1933)
4375 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4376 - add the "configurable stack size" patch to most series files (1256)
4377 - ability to write large log records, for 100+ OST configs (2306)
4378 - fix NULL deref when filter_prep fails (2314)
4379 - fix operator precedence error in filter_sync
4380 - dynamic allocation of socknal TX descriptors (2315)
4381 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4382 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4383 - ensure that CWARN messages in Portals always get to the syslog
4384 - __init/__exit are not for prototype decls (ldlm_init/exit)
4385 - x86-64 compile warning fixes
4386 - fix gateway LMC keyword conflict (2318)
4387 - fix MDS lock inversions in getattr/reint paths (1844)
4388 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4389 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4390 - DLM race condition prevented some lock evictions (2328)
4391 - ENOMEM detection and retry on socknal sends (2230)
4392 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4393 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4394 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4395 - better error messages when a client is rejected during recovery (1505)
4396 - avoid cancelling locks which were never granted, after failure (2330)
4397 - fix i_sem/journal inversion in mds_client_add (2333)
4398 - fix truncate/getattr lock cycle deadlock (2334)
4399 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4400 - fix two rare exit paths which could leak an l_lock() ref (2321)
4401 - fencepost error in MDS/OST orphan recovery (2226)
4402 - make log record alignment 8 bytes (1988)
4403 - lstripe now fails when requested offset > ost_count (2237)
4404 - ensure that all kernel series have a complete list.h (1607)
4405 - fix crashes in special-file operations (2316)
4406 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4408 - allow configurable automake binary, for testing new versions
4409 - small update to the lfs documentation
4411 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4413 * fix negative export reference count in fsfilt_sync (2312)
4415 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4416 * release candidate 0.9.1
4418 - orphans are moved into the PENDING directory for possible recovery
4419 - replayed opens now open by fid for orphan/rename safety (1042)
4420 - last close of an orphan inode generates a transno (683)
4421 - chdir() and mount() now pin the directory entry (1020)
4422 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4423 - discard very old requests without processing them (1502)
4424 - remove some common, well-understood CERRORs (1505)
4425 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4426 - clear "grant" flags in OST replies until OST grant code lands (1644)
4427 - fix read performance by not clobbering i_blksize on client (1598)
4428 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4429 - make LRU size a /proc tunable, clears locks when reduced (707)
4430 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4431 - server locks take references on exports now (1558)
4432 - build fixes for 2.4.20-rh trees (1663)
4433 - return an error from lov_create if all OSCs are inactive (1751)
4434 - fix import levels when a reconnect happens without a timeout (1597)
4435 - exit early from mds_open if we get a lookup error (1749)
4436 - partial page read at EOF wouldn't wait for disk before sending (1642)
4437 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4438 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4439 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4440 - write/truncate lock inversion (1639)
4441 - Don't auto-load obdclass, portals modules during cleanup (1495)
4442 - fix timestamps from jumping to "now" (1763)
4443 - extra journal assertions (1648)
4444 - add an extra multiunlink test (1771)
4445 - fix read_record/write_record API (1776)
4446 - fix leak of offset_extent, possible incorrect i_size later (1772)
4447 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4448 - Debugging for the kqswnal_get_idle_tx problems (1820)
4449 - Allow recovery to be attempted multiple times (1536)
4450 - Write out MDS last_rcvd file after it is first created (1600)
4451 - Fix tx_descriptor leak in failed transmit situations (1827)
4452 - ext3 journaling fixes for assertion failure after IO error (1871)
4453 - class_export_put() on freed export after completion AST error (1896)
4454 - Fix revalidate looping in VFS (1322)
4455 - Don't access a freed export during MDS_REINT timeout (1521)
4456 - Add open-unlink recovery support on the MDS (1673,1764)
4457 - Return an error if no MDS data was read from last_rcvd (1946)
4458 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4459 - Don't setup a disk device that doesn't match exported UUID (317)
4460 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4461 - avoid committing NULL handle in force close
4462 - local.sh is now a one-stripe LOV configuration
4463 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4464 - fix ext3 htree duplicate directory entry corruption (1516)
4465 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4466 - update to new LOV EA format (2097)
4467 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4469 2003-06-15 Phil Schwan <phil@clusterfs.com>
4472 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4473 - per-import/export recovery handling (958, 931, 959)
4474 - multiple last-rcvd slots, for serving multiple FSes (949)
4475 - connections are again shared between multiple imp/exports (963, 964)
4476 - "umount -f" would hang if any requests needed to be sent (393, 978)
4477 - avoid pinning large req buffer by copying for queued messages (989)
4478 - add "uuid" to "lctl device" command to help upcalls (991)
4479 - "open" RPCs with transnos would confuse recovery counters (1037)
4480 - do proper endian conversion of all wire messages (288, 340, 891)
4481 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4482 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4483 - fix ptlrpc_connection leak in target_handle_connect (1174)
4484 - fix import refcounting bug in OST and MDS cleanup (1134)
4485 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4486 - fix very unlikely obd_types race condition (501)
4487 - remove osc_open hack for echo_client (1187)
4488 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4489 - a failure in read_inode2 leads to deadlock (1139)
4490 - cancel ack-locks as soon as transaction is committed (1072)
4491 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4492 - make sure to commitrw after any preprw to avoid deadlock (1162)
4493 - failing to execute a file in a lustre FS would lock inode (1203)
4494 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4495 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4496 - obd_uuid2tgt would walk off the end of the list (1255)
4497 - on IA64 the osc would give portals incorrect bulk size (1258)
4498 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4499 - fix lock inversion caused by new llite matching code (1282)
4500 - limit the number of dirty pages on a client to 10MB (1286)
4501 - timed out locks were not being corrected cancelled (1289)
4502 - fix O_DIRECT above 4GB on IA-32 (1292)
4503 * major user-visible changes
4504 - fail out/fail over policy now controlled by the upcall (993)
4506 - add OBD_PING to check server availability and failure (954)
4507 - lustre messages are now sent in sending host order (288, 340, 891)
4508 - add eadatalen to MDS getattr reply (340)
4509 - OST read replies may contain second buffer, with per-page status (593)
4511 2003-03-11 Phil Schwan <phil@clusterfs.com>
4514 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4515 - failed open()s could cause deadlock; fixed (867, 869)
4516 - stop cancelling OST locks when files are closed (481)
4517 - overlapping XID spaces caused network corruption (851, 853)
4518 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4519 - setattr_raw added, to do single-RPC, server-side setattrs
4520 - lmc/lconf syntax change for OST UUIDs
4521 - fix crashy race condition between ptlrpc_free_req and osc_close
4522 - don't use request in mdc_enqueue if we hit a timeout (889)
4523 - don't set the inode i_size for regular files from the MDS (896)
4524 - handle out of order completion AST (842)
4525 - don't LBUG if a lock request times out after receiving AST (913)
4526 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4527 - if a bad lock AST arrives, send an error instead of dropping entirely
4528 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4529 - fix leak in bulk IO when only partially completed (899, 900, 926)
4530 - fix O_DIRECT for ia64 (55)
4531 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4532 - C-z after timeout could hang a process forever; fixed (977)
4534 - client-side I/O cache (678, 924, 929, 941, 970)
4536 - READPAGE and SETATTRs which don't take server-side locks get
4539 2003-02-11 Phil Schwan <phil@clusterfs.com>
4542 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4543 - Fix fsfilt_extN_readpage() to read a full page of directory
4544 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4545 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4546 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4547 - Most intent operations were removed, in favour of a new RPC mode
4548 that does a single RPC to the server and bypasses most of the VFS
4549 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4550 - Aggressively cancel local locks on DLM servers
4551 - mds_reint_unlink sends EA to the client if it's the last nlink.
4552 client uses that EA to unlink OST objects.
4553 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4554 - recursive symlinks were fixed (439)
4555 - fixed NULL deref in DEBUG_REQ
4556 - filter_update_lastobjid no longer calls sync, which annoyed extN
4557 - fixed multi-client small-writes to a single file problem (445)
4558 - fixed mtime updates during file writes (607)
4559 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4560 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4561 - fixed Linux OST size reporting problem (444, 656)
4562 - OST now updates object mtime with writes or setattr (607, 619)
4563 - client verifies file size before zeroing page past EOF (445)
4564 - OST now writes last allocated objid to disk with allocation (108)
4565 - LOV on echo now works (409)
4567 - mds_reint_unlink sends a new buffer, with the EA included. this
4568 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4569 set if a regular file was being unlinked, and it was the last link
4570 - use PtlGet from the target for bulk writes (315)
4571 - OST now updates object mtime with writes or setattr (607, 619)
4572 - LDLM now has a grant-time callback to revalidate locked items, if
4574 - Many MDS operations were reorganized to combat race conditions
4576 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4577 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4579 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4582 - Fully reactivate OST imports after reconnection (512, others)
4583 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4584 - More graceful error handling for truncating on dead OST (515)
4585 - Don't error out unless we're actually accessing dead stripes (474)
4586 - Fix garbage sizes when stripes are missing (410)
4587 - LRU counters were broken, causing constant lock purge (433, 432)
4588 - garbage on read from stripes with failed OSTs (441)
4589 - mark OSCs as active before reconnecting during recovery (438)
4590 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4591 - lfind did not preserve OST order in output (443)
4592 - symlinks cause hung clients, incorrect data (439)
4593 - stop dereferencing request after dropping refcount (457)
4594 - don't LASSERT(spin_is_locked) on non-SMP (455)
4595 - fixes for many rename() bugs
4596 - fstat didn't correctly synchronize attributes (399)
4597 - server must handle lock cancellation during blocking AST prep (487)
4598 - bulk descriptors were free()d too soon (511)
4599 - fix paths in lconf, which would load incorrect modules (451, 507)
4600 - fix confusing lconf 'host not found' error message (386)
4601 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4602 - fix race condition in mdc_blocking_ast() for inode access (526)
4603 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4604 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4605 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4606 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4607 - fix lstripe in several ways (kernel side) (527)
4608 - fix request leak in ldlm_cli_enqueue (262)
4609 - incorrect OSC was marked inactive after OST failure
4610 - call mds_fs_cleanup before unmounting filesystem (524)
4611 - fix races between taking ns_lock and ldlm_lock_change_resource
4612 - fix races updating LOV export open file list
4613 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4614 - fix recovery NULL deref in ldlm_cli_cancel_unused
4615 - fix some DLM races by using new hash table for lock handles (419)
4616 - permit the client to specify desired inodes, at replay
4617 - duplicate requests when we queue them for replay reintegration
4618 - fix last_rcvd offset calculation
4619 - sync after each recovered transaction, so we always make progress
4620 - never, not always, ERESTART requests without transnos
4621 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4623 - skip replay if the MDS says that the client is already connected
4624 - don't check for a recovery-enabled export to match lctl's UUID
4625 - don't INC_USE_COUNT for phantom exports
4626 - don't crash when cleaning up phantom exports (567)
4627 - don't double-finish or set replay data for errored mdc_open requests
4628 - abort requests when they time out, so we don't get old replies
4629 - send/receive replies for AST messages again
4630 - if the client says that it doesn't have the lock, cancel it on the
4632 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4633 mark it as destroyed, it will all work out when decref is called
4634 - fix module use counts (22, 581)
4636 - ASTs now expect a reply (server cancels lock on error reply)
4638 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4641 - fix many simultaneous client startup (392)
4642 - fix dentry->d_it clobbering
4643 - credentials weren't being shipped for readdir/getattr operations
4644 - remove invalid assertions triggered during some concurrent MD
4646 - proper Lustre versions added (336, 389)
4647 - fix memory leak for create error case (398)
4648 - fix LOV locking bug that would get cli/srv out of sync
4649 - fix echo client over LOV (409)
4650 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4651 - fix double-O_EXCL intent crash (424)
4652 - avoid sending multiple lock CANCELs (352)
4654 - MDS can do multi-client recovery (modulo bugs in new code)
4656 - many updates, edits, cleanups
4658 2002-11-18 Phil Schwan <phil@clusterfs.com>
4661 - fix null d_it dereference (346)
4662 - fix full OST/dbench hang (333)
4663 - fix permission problem with file removal (286)
4664 - fix removal of OSCs from LOV when they fail
4665 - fix NULL deref during bulk timeout (214)
4666 - fix problems related to multiple filesystems on one MDS (241)
4667 - fixed serious subtle metadata locking bugs
4668 - free locks on clients when inodes are removed due to memory
4670 - fix inode pointer in lock data (285)
4671 - partial support for multiple MDS on a single host (241)
4672 - data locks weren't cancelled at clear_inode time (290, 311)
4673 - intent locks could lead to unbounded lock growth (205)
4674 - added a maximum lock count, an LRU list, and a flusher
4675 - fix multiple rename (365)
4676 - properly abstracted the echo client
4677 - OSC locked 1 byte too many; fixed
4678 - rewrote brw callback code:
4679 - fixed recovery bugs related to LOVs (306)
4680 - fixed too-many-pages-in-one-write crash (191)
4681 - fixed (again) crash in sync_io_timeout (214)
4682 - probably fixed callback-related race (385)
4684 - Add capability to MDS protocol
4685 - LDLM cancellations and callbacks on different portals
4687 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4690 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4691 - defer open object destruction to close time (601981, 138)
4692 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4693 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4694 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4695 - don't call obd_setattr in ll_file_release for destroyed objects
4696 * protocol change to lustre_msg: move |version| and add |flags|
4697 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4698 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4699 * added replay of create, unlink, link and rename operations during
4700 MDS failover; recovery should be much more robust now
4701 * remove failed OSCs from LOVs (only lov_create uses this so far)
4702 * the lustre-HOWTO was brought (more) up to date (582544)
4704 2002-10-23 Phil Schwan <phil@clusterfs.com>
4707 - in-use dentries weren't being reused properly (617851)
4708 - prevent multiple LDLM setup (599178)
4709 - fix LOV size calculations for truncate (617853)
4710 - fix client handling of MDS intent errors (POSIX)
4711 - fix permission bug in lovstripe.c test (624321)
4712 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4713 - truncate past end of file could corrupt data
4714 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4715 - a race in recovery could return ETIMEDOUT to apps (623947)
4716 - building outside the source directory was fixed
4717 * the lustre-HOWTO was brought (more) up to date (582544)
4718 * major progress was made on recovery functionality
4720 2002-10-10 Phil Schwan <phil@clusterfs.com>
4723 - recovery deadlock fix
4724 - rm -rf causes LBUG fix (617817)
4725 - file open by multiple tasks fix (618962)
4726 - directory permissions bugs (602707 and 620007)
4727 - journal_stop fixed with locking (611313)
4728 - O_APPEND failures resolved (618273, perhaps 614459)
4729 - lconf PATH fix (619770)
4730 - IA64 build fix (621450)
4731 - RPC buffer sizes scale with amount of memory
4733 2002-10-01 Phil Schwan <phil@clusterfs.com>
4736 - locks would be cancelled without throwing away data pages,
4737 resulting in inconsistent data (605627)
4738 - inode attributes were not always being refreshed (605627, 612449)
4739 - lconf now continues to cleanup after lctl reports an error
4740 - MDS now enforces user permissions (602707)
4741 - lprocfs cleanup fixed, but not yet enabled (614157)
4742 - fixed infinite server hang, should a client not respond to an AST
4743 - avoid going into recovery if user calls readlink() with a buffer
4744 that's too small (613941)
4745 - AST RPCs no longer require replies (614867) -- this may be changed
4746 - don't crash server if client sends an IOV that's too big (611336)
4747 - fixed lock conversion deadlock (611892)
4748 - fixed the following of symlinks (614622)
4749 * recovery: the server can remove locks from a client that dies, other
4750 clients can make progress
4751 * more extN patch fixes
4752 * compile-time configurable ptlrpc buffer allocations
4754 - collaborative read cache document
4755 - Lustre Lite Performance CDR document-in-progress
4757 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4760 - fix typo in patch-2.4.18
4762 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4765 - clear ptlrpc request each time in handle_incoming_request()
4766 - unlink of files now destroys the object on the OST
4768 2002-09-19 Peter Braam <braam@clusterfs.com>
4770 * add hard link support
4771 * change obdfile creation method
4772 * kernel patch changed
4774 2002-09-19 Peter Braam <braam@clusterfs.com>
4777 - stack overflow bug in extN fixed
4779 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4781 * documentation updates
4782 - add man pages for config tools
4783 - update tests/README to describe testing with new config tools
4784 - finish metadata API descriptions
4785 * bug fixes and cleanups
4786 - statfs workaround for 16TB limit
4787 - LOV stripe allocation improved, can stripe on subset of OSTs
4788 - LOV file size/IO offset was wrong for files > 4GB in size
4789 - object EA data was being dropped, caused files to be unreadable
4790 - memory overflow with non-LOV OST caused memory corruption
4791 - fixed regression tests to work with new config tools, obdfilter
4792 - fixed bug when directory size became larger than 1 block
4793 - fixed bug (for single client case) when PWD was deleted
4794 - invalidate local directory pages when doing intent-based ops
4795 - avoid LDLM oops when lock callback contained bad data
4797 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4799 * documentation updates
4800 * bug fixes and cleanups
4801 - configuration tools
4804 - 64-bit compile warnings
4805 - 64-bit internal statfs data
4807 * test_brw on persistent OST devices
4809 * lprocfs (disabled)
4811 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4813 * documentation updates
4814 * bug fixes and cleanups
4815 * configuration tools
4817 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4820 * many small fixes to 0_5_4
4821 * io/network handling
4822 * thinkos in MDS operations
4824 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4827 * crucial basic fixes to 0.5.3
4828 * IOR, Iozone work over Elan
4831 2002-08-07 Phil Schwan <phil@clusterfs.com>
4832 * version 0_5_3, our first alpha
4833 * we use the new Portals iovs
4834 * documentation updates
4835 * bug fixes and cleanups
4836 * small changes in the DLM wire protocol
4838 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4839 * version 0_5_1 with some initial stability,
4840 * locking on MD and file I/O.
4841 * documentation updates
4842 * several bug fixes since 0.5.0
4843 * small changes in wire protocol
4845 2002-07-18 Phil Schwan <phil@clusterfs.com>
4847 * delivered as Lustre Light Alpha
4848 * fixed a crash after handling invalid MDS requests
4849 * fixed directory pages for architectures with non-4k pages sizes
4851 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4852 * release version v0_4_4
4853 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4855 * Fixes a number of interruption problems with OST operations.
4856 * Update documentation for portals header changes
4857 * Move all wire protocol structs/defines to lustre_idl.h
4858 * Fixes symlink length bug.
4859 * Add tcpdump to repository.
4861 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4862 * release version v0_4_3
4863 * Fixes statfs for inodes on extN.
4864 * Fixes bug in runtests which would delete /etc/hosts.
4865 * Use 64-bit object IDs wherever possible (not into VFS though)
4866 Remove ost_get_info, which is unused by lustre, and out of date.
4868 2002-07-03 Peter Braam <braam@clusterfs.com>
4869 * release version v0_4_2 Fixes a lookup error (type not passed)
4870 * move forward to head of Portals
4871 * move forward to latest Lustre kernel
4873 2002-06-25 Peter Braam <braam@clusterfs.com>
4874 * release version v0_4_1. Hopefully stable on single node use.