1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.19.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.7-sun3
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Description: do not process already freed flock
18 Details : flock can possibly be freed by another thread before it reaches
19 to ldlm_flock_completion_ast.
23 Description: LBUG during stress test
24 Details : Need properly lock accesses the flock deadlock detection list.
28 Description: oops in page fault handler
29 Details : kernel page fault handler can return two special 'pages' in error case, don't
30 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
34 Description: timeout with invalidate import.
35 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
36 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
37 ptlrpc_invalidate_import and as result LASSERT.
39 Severity : enhancement
41 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
45 Description: enable MGS and MDT services start separately
46 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
47 a co-located MGS without starting the MGS, which is a complement
48 to 'nosvc' mount option.
52 Frequency : after recovery
53 Description: precreate to many object's after del orphan.
54 Details : del orphan st in oscc last_id == next_id and this triger growing
55 count of precreated objects. Set flag LOW to skip increase count
56 of precreated objects.
60 Frequency : rare, on clear nid stats
61 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
62 Details : when clean nid stats sometimes try destroy live entry,
63 and this produce panic in free.
67 Description: Stack overflow during MDS log replay
68 ease stack pressure by using a thread dealing llog_process.
72 Description: wait until IO finished before start new when do lock cancel.
73 Details : VM protocol want old IO finished before start new, in this case
74 need wait until PG_writeback is cleared until check dirty flag and
75 call writepages in lock cancel callback.
77 Severity : enhancement
79 Description: using special macro for print time and cleanup in includes.
83 Description: mds_mfd_close() ASSERTION(rc == 0)
84 Details : In mds_mfd_close(), we need protect inode's writecount change
85 within its orphan write semaphore to prevent possible races.
89 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
93 Frequency : rare, on shutdown ost
94 Description: don't hit live lock with umount ost.
95 Details : shrink_dcache_parent can be in long loop with destroy dentries,
96 use shrink_dcache_sb instead.
100 Description: don't panic with use echo client
101 Details : echo client pass NULL as client nid pointer and this produce null
106 Description: fix build on ppc32
107 Details : compile code with -m64 flag produce wrong object file for ppc32.
111 Description: add message levels for liblustreapi
115 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
116 Details : Move the 'good_osts' check before the 'total_bavail' check. This
117 will result in an -EAGAIN and in the exit call path we call
118 alloc_rr() which will with increasing aggressiveness attempt to
119 aquire precreated objects on the minimum number of required OSCs.
123 Description: Use old size assignment to avoid deadlock
124 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
125 the scheduling while holding a spinlock. We do not need locking
126 for size in ll_update_inode() because size is only updated from
127 the MDS for directories or files without objects, so there is no
128 other place to do the update, and concurrent access to such inodes
129 are protected by the inode lock.
133 Description: resolve "_IOWR redefined" build error on SLES10
137 Description: dump the memory debugging after all modules are unloaded to
138 suppress false negative in conf_sanity test 39
140 Severity : enhancement
142 Description: build kernel-ib packages for OFED 1.3 in our release cycle
147 Description: fix SLES kernel versioning
148 Details : the kernel version for our SLES 10 kernel did not include a "-"
149 before the "smp" at the end. while this was not a problem in
150 general, it did mean that software trying to use the kernel
151 version to try to detect a vendor specific kernel would fail.
152 this was most evident by the OFED build scripts.
156 Description: Don't update lov_desc members until making sure they are valid
157 Details : When updating lov_desc members via proc fs, need fix their
158 validities before doing the real update.
162 Description: don't put request into delay list while invalidate in flight.
163 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
164 import in flight. this produce timeout for invalidate and sometimes
165 can cause stale data.
169 Frequency : on ppc only
170 Description: not convert ost objects for directory because it's not exist.
171 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
175 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
177 Severity : enhancement
179 Description: Update to support 2.6.22.14 vanilla kernel.
183 Frequency : rare, on recovery
184 Description: read procfs can produce deadlock in some situation
185 Details : Holding lprocfs lock which send rpc can produce block for destroy
186 obd objects and this also block reconnect with -EALREADY. This isn't
187 fix all lprocfs bugs - but make it rare.
189 Severity : enhancement
191 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
194 Frequency : frequent on X2 node
196 Description: mdc_set_open_replay_data LBUG
197 Details : Set replay data for requests that are eligible for replay.
201 Description: lustre_mgs: operation 101 on unconnected MGS
202 Details : When MGC is disconnected from MGS long enough, MGS will evict the
203 MGC, and late on MGC cannot successfully connect to MGS and a lot
204 of the error messages complaining that MGS is not connected.
208 Frequency : on network error
209 Description: panic with double free request if network error
210 Details : mdc_finish_enqueue is finish request if any network error ocuring,
211 but it's true only for synchronus enqueue, for async enqueue
212 (via ptlrpcd) this incorrect and ptlrpcd want finish request
215 Severity : enhancement
217 Description: client-side metadata stat-ahead during readdir(directory readahead)
218 Details : perform client-side metadata stat-ahead when the client detects
219 readdir and sequential stat of dir entries therein
222 Frequency : on start mds
224 Description: Implement get_info(last_id) in obdfilter.
227 Frequency : occasional
229 Description: Correctly check stale fid, not start epoch if ost not support SOM
230 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
231 overwrited with new generated fid, but mds can anwer with one of these
232 two fids and both is not stale. setattr incorectly start epoch and
233 assume will be called done_writeting, but without SOM done_writing
237 Frequency : rare, depends on device drivers and load
239 Description: MDS or OSS nodes crash due to stack overflow
240 Details : Code changes in 1.8.0 increased the stack usage of some functions.
241 In some cases, in conjunction with device drivers that use a lot
242 of stack the MDS (or possibly OSS) service threads could overflow
243 the stack. One change which was identified to consume additional
244 stack has been reworked to avoid the extra stack usage.
247 Frequency : occasional
249 Description: Do not fail import if osc_interpret_create gets -EAGAIN
250 Details : If osc_interpret_create got -EAGAIN it immediately exits and
251 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
252 oscc_has_objects and see OSC has no objests and call
253 oscc_internal_create to resend create request.
255 Severity : enhancement
257 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
259 Severity : enhancement
261 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
264 Frequency : very rare
266 Description: Fix replay if there is an un-replied request and open
267 Details : In some cases, older replay request will revert the
268 mcd->mcd_last_xid on MDS which is used to record the client's
271 Severity : enhancement
273 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
275 Severity : enhancement
277 Description: Add rhel5 support to HEAD.
279 Serverity : enhancement
281 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
286 Description: Don't allow skipping OSTs if index has been specified.
287 Details : Don't allow skipping OSTs if index has been specified, make locking
288 in internal create lots better.
292 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
293 Details : access to bitfield in structure is always rounded to long
294 and this produce problem with not atomic change any bit.
298 Description: Lustre make rpms failed.
299 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
300 builds Lustre rpms from tarball.
303 Frequency : rare on shutdown ost
305 Description: If llog cancel was not send before clean_exports phase, this can
306 produce deadlock in llog code.
307 Details : If llog thread has last reference to obd and call class_import_put
308 this produce deadlock because llog_cleanup_commit_master wait when
309 last llog_commit_thread exited, but this never success because was
310 called from llog_commit_thread.
314 Description: allow userland application know is lost one of stripes.
315 Details : fill lvb_blocks with error code on ost and return it to
316 application if error flag found.
320 Description: NULL lov_tgts causing MDS oops
321 Details : more safe checks for NULL lov_tgts for avoid oops.
323 Severity : enhancement
325 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
329 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
330 Details : If an OST is down the MDS will hang indefinitely in
331 obd_statfs_rqset() waiting for the statfs data. While for
332 MDS QOS usage of statfs, it should not stuck in waiting.
334 Severity : enhancement
336 Description: remote_acl support
337 Details : Support ACL-based permission check for remote user.
338 Support setfacl/getfacl for remote user with the utils
339 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
340 the system "{s,g}etfacl" utils.
342 Severity : enhancement
344 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
346 Severity : enhancement
348 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
352 Description: make mgs_setparam() handle fsname containing dash
353 Details : fsname containing a dash does not work with lctl conf_param
355 Severity : enhancement
357 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
359 Severity : enhancement
361 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
365 Frequency : rare, at shutdown
366 Description: access already free / zero obd_namespace.
367 Details : if client_disconnect_export was called without force flag set,
368 and exist connect request in flight, this can produce access to
369 NULL pointer (or already free pointer) when connect_interpret
370 store ocd flags in obd_namespace.
374 Frequency : only at startup
375 Description: not alloc memory with spinlock held.
376 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
377 if any spinlock held.
379 Severity : enhancement
381 Description: make lustre randomly fail allocating memory
382 Details : Make lustre randomly failed allocating memory for testing purpose.
384 Severity : enhancement
386 Description: lost problems with lov objid file
387 Details : Fixes some scability and access to not inited memory problems
388 in work with lov objdid file.
393 Description: lfs find does not continue on file error
394 Details : Continue other files processing when a file/dir is absent.
398 Description: Inconsistent usage of lustre_pack_reply()
399 Details : Standardize the usage of lustre_pack_reply() such that it
400 always generate a CERROR on failure.
403 Frequency : occasional
405 Description: MDS hang or stay in waiting lock
406 Details : If client receive lock with CBPENDING flag ldlm need send lock
407 cancel as separate rpc, to avoid situation when cancel request
408 can't processed due all i/o threads stay in wait lock.
412 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
416 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
420 Description: Update to SLES9 kernel-2.6.5-7.287.3.
424 Description: Update to RHEL4 latest kernel.
426 Severity : enhancement
428 Description: Build SLES10 patchless client fails
429 Details : The configure was broken by run ./configure with
430 --with-linux-obj=.... argument for patchless client. When the
431 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
432 can't search header adequately. Use absolute path such as
433 -I($LINUX)/include instead.
437 Description: interrupt oig_wait produce painc on resend.
438 Details : brw_redo_request can be used for resend requests from ptlrpcd and
439 private set, and this produce situation when rq_ptlrpcd_data not
440 copyed to new allocated request and triggered LBUG on assert
441 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
442 ptlrpcd set if request is changed and can be safety changed to use
447 Description: LASSERT_{REQ,REP}SWAB macros are buggy
448 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
449 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
450 these macros and replace them with their difinitions instead.
454 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
455 Details : Update which_patch & target file for SLES10 latest kernel.
458 Description: Remove client patches from SLES 10 kernel.
459 Details : This causes SLES 10 clients to behave as patchless clients
460 even on a Lustre-patched (server) kernel.
462 Severity : enhancement
464 Description: self-adjustable client's lru lists
465 Details : use adaptive algorithm for managing client cached locks lru
466 lists according to current server load, other client's work
467 pattern, memory activities, etc. Both, server and client
468 side namespaces provide number of proc tunables for controlling
471 Severity : enhancement
473 Description: light-weight GSS support
474 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
475 and reduce performance overhead.
477 Severity : enhancement
479 Description: Linux keyring support
480 Details : Support using service of Linux keyring for Lustre GSS internal
481 context refresh/cache mechanism.
485 Description: Fix errors in lfs documentation
486 Details : Fixes man pages
490 Description: don't use GFP_* in generic Lustre code.
491 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
495 Description: obdclass is limited by single OBD_ALLOC(idarray)
496 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
500 Description: warnings with build patchless client with vanila 2.6.19 and up
501 Details : change old ctl_table style and replace ctl_table/ctl_table_header
502 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
506 Frequency : Only for SLES
507 Description: sanity_quota fail test_1
508 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
509 file, which makes TSTID[2] inunique.
513 Frequency : Only for SLES9
514 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
515 Details : Update target/ChangeLog/which_patch .
519 Frequency : Only for test_10 in sanity-quota.sh
520 Description: a bug in quota test code
521 Details : checking a test flag in a wrong place
525 Frequency : Only for test_13 in sanity-quota.sh
526 Description: a bug in quota test script
527 Details : shouldn't check overfull allocated quota
531 Description: jbd statistics
532 Details : Port older jbd statistics patch for sles10
534 Severity : enhancement
536 Description: Add printing inode info into message about error in writepage.
541 Description: reply_lock_interpret crash due to race with it and lock cancel.
542 Details : Do not replay locks that are being cancelled. Do not reference
543 locks by their address during replay, just by their handle.
547 Frequency : When flocks are used.
548 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
549 Details : Flock locks might destroy just granted lock if it could be merged
550 with another existing flock, this is done in completion handler,
551 so teach ldlm_cli_enquque_fini that this is a valid case for
557 Description: Oops in read and write path when failing to allocate lock.
558 Details : Check if lock allocation failed and return error back.
562 Description: lstripe command fails for valid OST index
563 Details : The stripe offset is compared to lov->desc.ld_tgt_count
564 instead of lov->desc.ld_active_tgt_count.
568 Description: sanity.sh failed test 103
569 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
574 Description: df doesn't work properly if diskfs blocksize != 4K
575 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
579 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
580 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
584 Description: obd_health_check_timeout too short
585 Details : set obd_health_check_timeout as 1.5x of obd_timeout
589 Description: llapi_file_create() does not allow some changes
590 Details : add llapi_file_open() that allows specifying the mode and
591 open flags, and also returns an open file handle.
595 Description: merge and cleanup kernel patches.
596 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
600 Description: Add journal checksum support.(Kernel part)
601 Details : The journal checksum feature adds two new flags i.e
602 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
603 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
604 indicates that the commit block contains the checksum for
605 the blocks described by the descriptor blocks. Now commit
606 record can be sent to disk without waiting for descriptor
607 blocks to be written to disk. This behavior is controlled
608 using JBD2_FEATURE_ASYNC_COMMIT flag.
612 Description: OSS needs mutliple precreate threads
613 Details : Add ability to start more than one create thread per OSS.
617 Description: Sanity.sh test_65k failed: setstripe should have succeeded
618 Details : execute lfs setstripe on client
622 Description: mds_obd_create error creating tmp object
623 Details : When the user sets quota on root, llog will be affected and can't
624 create files and write files.
627 Frequency : Always on ia64 patchless client, and possibly others.
629 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
630 Details : This allows the patchless client to be loaded on architectures
635 Description: RedHat Update kernel for RHEL5
636 Details : Add the kernel config file for RHEL5.
640 Description: RedHat Update kernel for RHEL5
641 Details : Modify the kernel config file more closer RHEL5.
645 Description: Build failure against Centos5 (RHEL5)
646 Details : Define PAGE_SIZE when it isn't present.
650 Description: test_7 on sanity-quota.sh can't be used on separate machines
651 Details : add facet to handle it
655 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
656 Details : with NFS, the anon dentry's parent was set to itself in
657 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
658 corresponding dentry other than use rec->ur_name.
660 Severity : enhancement
662 Description: lfs setstripe enhancement
663 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
667 Description: enable data checksumming by default
668 Details : enable checksum by default, allow --disable-checksum
669 configure option and "-o nochecksum" mount option. Checksums
670 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
674 Description: lustre support for RHEL5
675 Details : Add support for RHEL5.
679 Description: Client eviction due to failover config
680 Details : after a connection loss, the lustre client should attempt to
681 reconnect to the last active server first before trying the
682 other potential connections.
686 Description: when mds and osts use different quota unit(32bit and 64bit),
687 quota will be released repeatly.
688 Details : void sending multiple quota reqs to mds, which will keep the
689 status between the reqs.
693 Description: osts not allocated evenly to files
694 Details : change the condition to increase offset_idx
697 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
699 Description: Data corruption for OSTs that are formatted larger than 2TB
701 Details : When generating the bio request for lustre file writes the
702 sector number would overflow a temporary variable before being
703 used for the IO. The data reads correctly from Lustre (which
704 will overflow in a similar manner) but other file data or
705 filesystem metadata may be corrupted in some cases.
709 Description: Tune the kernel for good SCSI performance.
710 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
711 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
716 Description: rewrite ext2-derived code in obdclass/uuid.c
717 Details : rewrite inherited code (uuid parsing code from ext2 utils)
718 from scratch preserving functionality.
722 Description: "lfs find -obd UUID" prints directories
723 Details : "lfs find -obd UUID" will return all directory names instead
724 of just file names. It is incorrect because the directories
725 do not reside on the OSTs.
729 Description: use i_size_read and i_size_write in 2.6 port
730 Details : replace inode->i_size access with i_size_read/write()
733 Frequency : when removing large files
735 Description: scheduling issue during removal of large Lustre files
736 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
737 It causes scheduling issues when removing large files (17TB in the
741 Frequency : only with liblustre clients on XT3
743 Description: evictions taking too long
744 Details : allow llrd to evict clients directly on OSTs
747 Frequency : only on ppc
749 Description: /proc/fs/lustre/devices broken on ppc
750 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
751 We should make sure the type of 'index' is loff_t and then cast
752 explicitly as needed below. Do not assign an explicitly cast
756 Frequency : only for rhel5
758 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
759 Details : Modify the target file & which_kernel.
761 Severity : enhancement
763 Description: omit set fsid for export NFS
764 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
765 as FS_REQUIRES_DEV for avoid problems with generate fsid.
769 Frequency : Always, for kernels after 2.6.16
770 Description: Fix warning idr_remove called for id=.. which is not allocated.
771 Details : Last kernels save old s_dev before kill super and not allow
772 to restore from callback - restore it before call kill_anon_super.
776 Description: buffer overruns could theoretically occur
777 Details : llapi_semantic_traverse() modifies the "path" argument by
778 appending values to the end of the origin string, and a
779 overrun may occur. Adding buffer overrun check in liblustreapi.
783 Description: Fix error on 'ls .' at the top of the Lustre mount.
784 Details : Don't revalidate dentry if it is a root dentry.
788 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
789 Details : Modify vm-tunables-rhel4.patch.
793 Description: Kernel config for 2.6.18-vanilla.
794 Details : Modify targets/2.6-vanilla.target.in.
795 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
796 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
797 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
798 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
802 Description: improve handling recoverable errors
803 Details : if request processig with error which can be recoverable on server
804 request should be resend, otherwise page released from cache and
809 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
810 Details : Modify target file & which_patch.
811 A flaw was found in the IA32 system call emulation provided
812 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
813 value could be stored in the %RAX register, which could trigger an
814 out-of-bounds system call table access. An untrusted local user
815 could exploit this flaw to run code in the kernel
816 (ie a root privilege escalation). (CVE-2007-4573).
820 Description: O_DIRECT bypasses client statistics.
821 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
822 read_ahead_stats were updated. copied the stats section from
823 osc_send_oap_rpc() into async_internal().
827 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
831 Description: change order of libsysio includes
832 Details : '#include sysio.h' should always come before '#include xtio.h'
836 Description: handle "serious error: objid * already exists" more gracefully
837 Details : If LAST_ID value on disk is smaller than the objects existing in
838 the O/0/d* directories, it indicates disk corruption and causes an
839 LBUG(). If the object is 0-length, then we should use the existing
840 object. This will help to avoid a full fsck in most cases.
842 Severity : enhancement
844 Description: adapt the lustre_config script to support the upgrade case
845 Details : Add "-u" option for lustre_config script to support upgrading 1.4
846 server targets to 1.6 in parallel.
850 Description: To avoid grant space > avaible space when the disk is almost
851 full. Without this patch you might see the error "grant XXXX >
852 available" or some LBUG about grant, when the disk is almost
854 Details : In filter_check_grant, for non_grant cache write, we should
855 check the left space by if (*left > ungranted + bytes), instead
856 of (*left > ungranted), because only we are sure the left space
857 is enough for another "bytes", then the ungrant space should be
858 increase. In client, we should update cl_avail_grant only there
859 is OBD_MD_FLGRANT in the reply.
863 Description: Update RHEL 4 kernel to fix local root privilege escalation.
864 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
865 described in CVE-2007-4573. This problem could allow untrusted
866 local users to gain root access.
869 Frequency : when using O_DIRECT and quotas
871 Description: Incorrect file ownership on O_DIRECT output files
872 Details : block usage reported by 'lfs quota' does not take into account
873 files that have been written with O_DIRECT.
878 Description: touch file failed when fs is not full
879 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
880 otherwise we can get ENOSP while fs is not full.
884 Description: parallel lock callbacks
885 Details : Instead of sending blocking and completion callbacks as separated
886 requests, adding them to a set and sending in parallel.
889 Frequency : only for Cray XT3
890 Bugzilla : 12829/13455
891 Description: Changing primary group doesn't change the group lustre assigns to
893 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
894 overridden with the primary group provided by the group upcall,
895 whereas the supplied fsgid can be trusted if it is in the list of
896 supplementary groups returned by the group upcall.
898 Severity : enhancement
900 Description: Allow masking D_WARNING, D_ERROR messages from console
901 Details : Console messages can now be disabled via lnet.printk.
905 Description: User code with malformed file open parameter crashes client node
906 Details : Before packing join_file req, all the related reference should be
907 checked carefully in case some malformed flags cause fake
908 join_file req on client.
912 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
913 Details : ldlm_completion_ast() assumes that a lock is granted when the req
914 mode is equal to the granted mode. However, it should also check
915 that LDLM_FL_CP_REQD is not set.
919 Description: Heavy nfs access might result in deadlocks
920 Details : After ELC code landed, it is now improper to enqueue any mds
921 locks under och_sem, because enqueue might want to decide to
922 cancel open locks for same inode we are holding och_sem for.
926 Description: Client eviction while running blogbench
927 Details : A lot of unlink operations with concurrent I/O can lead to a
928 deadlock causing evictions. To address the problem, the number of
929 oustanding OST_DESTROY requests is now throttled to
930 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
931 callbacks are processed in priority.
935 Description: enable ACLs on MDS by default
936 Details : ACLs must be enabled on MDS by default.
939 Frequency : PPC/PPC64 only
941 Description: conflicts between asm-ppc64/types.h and lustre_types.h
942 Details : fix duplicated definitions between asm-ppc64/types.h and
943 lustre_types.h on PPC.
946 Frequency : PPC/PPC64 only
948 Description: asm-ppc/segment.h does not exist
949 Details : fix compile issue on PPC.
953 Description: better handle error messages in extents code
956 Frequency : RHEL4 only
958 Description: mkfs is very slow on IA64/RHEL4
959 Details : A performance regression has been discovered in the MPT Fusion
960 driver between versions 3.02.73rh and 3.02.99.00rh. As a
961 consequence, we have downgraded the MPT Fusion driver in the RHEL4
962 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
964 Severity : enhancement
966 Description: SNMP support enhancement
967 Details : Adding total number of sampled request for an MDS node in snmp
970 Severity : enhancement
972 Description: Optimize ldlm waiting list processing for PR extent locks
973 Details : When processing waiting list for read extent lock and meeting read
974 lock that is same or wider to it that is not contended, skip
975 processing rest of the list and immediatelly return current
976 status of conflictness, since we are guaranteed there are no
977 conflicting locks in the rest of the list.
981 Description: Time out and refuse to reconnect
982 Details : When the failover node is the primary node, it is possible
983 to have two identical connections in imp_conn_list. We must
984 compare not conn's pointers but NIDs, otherwise we can defeat
985 connection throttling.
989 Description: port llog fixes from b1_6 into HEAD
990 Details : Port llog reference couting and some llog cleanups from b1_6
991 (bug 10800) into HEAD, for protect from panic and access to already
992 free llog structures.
996 Description: Detect stride IO mode in read-ahead
997 Details : When a client does stride read, read-ahead should detect that and
998 read-ahead pages according to the detected stride pattern.
1002 Description: data checksumming impacts single node performance
1003 Details : add support for several checksum algorithm. Currently, only CRC32
1004 and Adler-32 are supported. The checksum type can be changed on
1005 the fly via /proc/fs/lustre/osc/*/checksum_type.
1009 Description: use adler32 for page checksums
1010 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1015 Description: build for x2 fails
1016 Details : fix compile issue on Cray systems.
1020 Description: Properly match for duplicate locks
1021 Details : Due to different lock order from skiplists code, we need to
1022 traverse entire list for now
1025 Frequency : only on PPC/SLES10
1027 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1028 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1029 be defined. Add a hack in mkfs_lustre.c to work around this compile
1034 Description: LASSERT on MDS when client holding flock lock dies
1035 Details : ldlm pool logic depends on number of granted locks equal to
1036 number of released locks which is not true for flock locks, so
1037 just exclude such locks from consideration.
1041 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1042 Details : Use fsfilt_sync() for both the whole filesystem sync and
1043 individual file sync to eliminate dangerous inode locking
1044 with I_LOCK that can lead to a deadlock.
1048 Description: performance in 1.6.3
1049 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1050 sure that 1MB requests are not fragmented by the block layer.
1052 Severity : enhancement
1054 Description: organize the server-side client stats on per-nid basis
1055 Details : Change the structure of stats under obdfilter and mds to
1063 The "uuid"s file would list the uuids of _active_ exports.
1064 And the clear entry is to clear all stats and stale nids.
1066 Severity : enhancement
1068 Description: eliminate client locks in face of contention
1069 Details : file contention detection and lockless i/o implementation
1070 for contended files.
1074 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1077 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1079 Description: MDS may lose file striping (and hence file data) in some cases
1080 Details : If there are additional extended attributes stored on the MDS,
1081 in particular ACLs, SELinux, or user attributes (if user_xattr
1082 is specified for the client mount options) then there is a risk
1083 of attribute loss. Additionally, the Lustre file striping
1084 needs to be larger than default (e.g. striped over all OSTs),
1085 and an additional attribute must be stored initially in the
1086 inode and then increase in size enough to be moved to the
1087 external attribute block (e.g. ACL growing in size) for file
1092 Description: skiplist implementation simplification
1093 Details : skiplists are used to group compatible locks on granted list
1094 that was implemented as tracking first and last lock of each lock group
1095 the patch changes that to using doubly linked lists
1099 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1100 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1101 because LASSERT on both the data supplied by a client, and the data
1102 on disk is dangerous and incorrect.
1104 Severity : enhancement
1106 Description: Slow trucate/writes to huge files at high offsets.
1107 Details : Directly associate cached pages to lock that protect those pages,
1108 this allows us to quickly find what pages to write and remove
1109 once lock callback is received.
1111 --------------------------------------------------------------------------------
1113 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1115 * Support for kernels:
1116 2.6.5-7.283 (SLES 9),
1117 2.6.9-55.EL (RHEL 4),
1118 2.6.16.46-0.14 (SLES 10),
1119 2.6.18.8 vanilla (kernel.org)
1120 * Client support for unpatched kernels:
1121 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1122 2.6.16 - 2.6.22 vanilla (kernel.org)
1123 * Due to recently discovered recovery problems, we do not recommend
1124 using patchless RHEL 4 clients with this or any earlier release.
1125 * Recommended e2fsprogs version: 1.39.cfs8
1126 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1127 * Starting with this release, the ldiskfs backing filesystem required
1128 by Lustre is now in its own package, lustre-ldiskfs. This package
1129 should be installed. It is versioned separately from Lustre and
1130 may be released separately in future.
1134 Description: block reactivating mgc import until all deactivates complete
1135 Details : Fix race when failing back MDT/MGS to itself (testing)
1138 Frequency : at statup only
1140 Description: mds_lov_synchronize race leads to various problems
1141 Details : simultaneous MDT->OST connections at startup can cause the
1142 sync to abort, leaving the OSC in a bad state.
1144 Severity : enhancement
1146 Description: add optional extra BUILD_VERSION info
1147 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1148 to override the lustre version.
1151 Frequency : 2.6.18 servers only
1153 Description: ll_kern_mount() doesn't release the module reference
1154 Details : The ldiskfs module reference count never drops down to 0
1155 because ll_kern_mount() doesn't release the module reference.
1160 Description: server LBUG when using old ost_num_threads parameter
1161 Details : Accept the old ost_num_threads parameter but warn that it
1162 is deprecated, and fix an off-by-one error that caused an LBUG.
1167 Description: Transient SCSI error results in persistent IO issue
1168 Details : iobuf->dr_error is not reinitialized to 0 between two
1172 Frequency : sometimes when underlying device returns I/O errors
1174 Description: OSTs not going read-only during write failures
1175 Details : OSTs are not remounted read-only when the journal commit threads
1176 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1177 instead of the ext3 wrappers.
1181 Description: poor connect scaling with increasing client count
1182 Details : Don't run filter_grant_sanity_check for more than 100 exports
1183 to improve scaling for large numbers of clients.
1186 Frequency : SLES10 only
1188 Description: sanity-quota.sh quotacheck failed: rc = -22
1189 Details : Quotas cannot be enabled on SLES10.
1192 Frequency : liblustre clients only
1194 Description: getdirentries does not give error when run on compute nodes
1195 Details : getdirentries does not fail when the size specified as an argument
1196 is too small to contain at least one entry
1198 Severity : enhancement
1200 Description: Add LNET router traceability for debug purposes
1201 Details : If a checksum failure occurs with a router as part of the
1202 IO path, the NID of the last router that forwarded the bulk data
1203 is printed so it can be identified.
1208 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1209 Details : Due to a race condition, liblustre clients were occasionally
1210 evicted incorrectly.
1212 Severity : enhancement
1214 Description: lfs setstripe use optional parameters instead of postional
1217 Severity : enhancement
1219 Description: Nanosecond timestamp support for ldiskfs
1220 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1221 resolution timestamps. There is not yet support for this at
1222 the Lustre filesystem level.
1225 Frequency : during server recovery
1227 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1228 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1232 Frequency : nfs export on patchless client
1234 Description: connectathon hang when test nfs export over patchless client
1235 Details : Disconnected dentry cannot be found with lookup, so we do not need
1236 to unhash it or make it invalid
1239 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1240 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1241 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1246 Description: open req refcounting wrong on reconnect
1247 Details : If reconnect happened between getting open reply from server and
1248 call to mdc_set_replay_data in ll_file_open, we will schedule
1249 replay for unreferenced request that we are about to free.
1250 Subsequent close will crash in variety of ways.
1251 Check that request is still eligible for replay in
1252 mdc_set_replay_data().
1257 Description: disable writes to filesystem when reading health_check file
1258 Details : the default for reading the health_check proc file has changed
1259 to NOT do a journal transaction and write to disk, because this
1260 can cause reads of the /proc file to hang and block HA state
1261 checking on a healthy but otherwise heavily loaded system. It
1262 is possible to return to the previous behaviour during configure
1263 with --enable-health-write.
1265 Severity : enhancement
1267 Description: 64-bit inode version
1268 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1269 to the inode. This will be required for version-based recovery.
1274 Description: MDS fails to start if a duplicate client export is detected
1275 Details : in some rare cases it was possible for a client to connect to
1276 an MDS multiple times. Upon recovery the MDS would detect this
1277 and fail during startup. Handle this more gracefully.
1279 Severity : enhancement
1281 Description: Add -o localflock option to simulate old noflock
1283 Details : This will achieve local-only flock/fcntl locks
1289 Description: log_commit_thread vs filter_destroy race leads to crash
1290 Details : Take import reference before releasing llog record semaphore
1295 Description: Wrong request locking in request set processing
1296 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1297 imp_delayed_list, in this place should be used imp_lock.
1300 Frequency : when reconnection
1302 Description: Grant Leak when osc reconnect to OST
1303 Details : When osc reconnect ost, OST(filter) should check whether it
1304 should grant more space to client by comparing fed_grant and
1305 cl_avail_grant, and return the granted space to client instead
1306 of "new granted" space, because client will call osc_init_grant
1307 to update the client grant space info.
1310 Frequency : when client reconnect to OST
1312 Description: Grant Leak when osc do resend and replay bulk write
1313 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1314 bulk write request, because the grant info while be sync between
1315 OSC and OST when reconnect, and we should ignore the grant info
1316 these of resend/replay write req.
1321 Description: Grant space more than avaiable left space sometimes.
1322 Details : When then OST is about to be full, if two bulk writing from
1323 different clients came to OST. Accord the avaliable space of the
1324 OST, the first req should be permitted, and the second one
1325 should be denied by ENOSPC. But if the seconde arrived before
1326 the first one is commited. The OST might wrongly permit second
1327 writing, which will cause grant space > avaiable space.
1330 Frequency : when client is evicted
1332 Description: Grant might be wrongly erased when osc is evicted by OST
1333 Details : when the import is evicted by server, it will fork another
1334 thread ptlrpc_invalidate_import_thread to invalidate the
1335 import, where the grant will be set to 0. While the original
1336 thread will update the grant it got when connecting. So if
1337 the former happened latter, the grant will be wrongly errased
1338 because of this race.
1343 Description: Checking Stale with correct fid
1344 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1345 to check whether it is stale, because sometimes, we want the
1346 enqueue happened anyway, and op_data.fid2 will not be initialized.
1348 Severity : enhancement
1350 Description: update patchless client
1351 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1354 Frequency : only with 2.4 kernel
1356 Description: random memory corruption
1357 Details : size of struct ll_inode_info is to big for union inode.u and this
1358 can be cause of random memory corruption.
1363 Description: Memory leak in recovery
1364 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1365 It should also check obd_fail before fsfilt_start, otherwise if
1366 fsfilt_start return -EROFS,(failover mds during mds recovery).
1367 then the req will return with repmsg->transno = 0 and rc = EROFS.
1368 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1369 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1370 be freed no matter whether fsfilt_commit success or not.
1373 Frequency : only with huge count clients
1375 Description: Prevents from taking the superblock lock in llap_from_page for
1377 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1378 for llap_from_page call in ll_removepage prevents from taking the
1379 superblock lock for a soon died page.
1384 Description: Not check open intent error before release open handle
1385 Details : in some rare cases, the open intent error is not checked before
1386 release open handle, which may cause
1387 ASSERTION(open_req->rq_transno != 0), because it tries to release
1388 the failed open handle.
1393 Description: Set cat log bitmap only after create log success.
1394 Details : in some rare cases, the cat log bitmap is set too early. and it
1395 should be set only after create log success.
1399 Description: Accessing a block bevice can re-enable I/O when Lustre is
1400 tearing down a device.
1401 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1406 Description: service threads may hog cpus when there are a lot of requests
1407 Details : Insert cond_resched to give other threads a chance to use some CPU
1412 Description: the cat log was not initialized in recovery
1413 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1414 unlink log on mds will not be initialized until mds post
1415 recovery. And also in mds post recovery, the unlink log will
1416 initialization will be done asynchronausly, so there will be race
1417 between add unlink log and unlink log initialization.
1421 Description: brw_stats were being printed incorrectly
1422 Details : brw_stats were being printed as log2 but all of them were not
1423 recorded as log2. Also remove some code duplication arising from
1424 filter_tally_{read,write}.
1428 Frequency : rare, only in recovery.
1429 Description: ASSERTION(req->rq_type != LI_POISON) failed
1430 Details : imp_lock should be held while iterating over imp_sending_list for
1431 prevent destroy request after get timeout in ptlrpc_queue_wait.
1435 Description: replay-single.sh test 52 fails
1436 Details : A lock's skiplist need to be cleanup when it being unlinked
1437 from its resource list.
1441 Description: Short directio read returns full requested size rather than
1443 Details : Direct I/O operations should return actual amount of bytes
1444 transferred rather than requested size.
1446 Severity : enhancement
1448 Description: metadata RPC reduction (e.g. for rm performance)
1449 Details : decrease the amount of synchronous RPC between clients and servers
1450 by canceling conflicing lock before the operation on the client
1451 and packing thier handles into the main operation RPC to server.
1453 Severity : enhancement
1455 Description: Async OSC create to avoid the blocking unnecessarily.
1456 Details : If a OST has no remain object, system will block on the creating
1457 when need to create a new object on this OST. Now, ways use
1458 pre-created objects when available, instead of blocking on an
1459 empty osc while others are not empty. If we must block, we block
1460 for the shortest possible period of time.
1462 Severity : enhancement
1464 Description: refine locking for avoid write wrong info into lov_objid file
1465 Details : fix possible races with add new target and write/update data in
1468 --------------------------------------------------------------------------------
1470 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1475 Frequency : on some architectures
1477 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1478 Details : Uninitialized flags sometimes cause configuration commands to
1482 Frequency : patchless clients only
1484 Description: missing __iget() symbol export
1485 Details : The __iget() symbol export is missing. To avoid the need for
1486 this on patchless clients the deathrow inode reaper is turned
1487 off, and we depend on the VM to clean up old inodes. This
1488 dependency was during via the fix for bug 12181.
1490 --------------------------------------------------------------------------------
1492 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1494 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1495 INTEROPERATE with older versions automatically. In many cases a
1496 special upgrade step is needed. Please read the
1497 user documentation before upgrading any part of a live system.
1498 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1499 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1500 * WARNING: Lustre configuration and startup changes are required with
1501 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1507 Severity : enhancement
1509 Description: MountConf
1510 Details : Lustre configuration is now managed via mkfs and mount
1511 commands instead of lmc and lconf. New obd types (MGS, MGC)
1512 are added for dynamic configuration management. See
1513 https://mail.clusterfs.com/wikis/lustre/MountConf for
1516 Severity : enhancement
1518 Description: dynamic OST addition
1519 Details : OSTs can now be added to a live filesystem
1521 Severity : enhancement
1523 Description: startup order invariance
1524 Details : MDTs and OSTs can be started in any order. Clients only
1525 require the MDT to complete startup.
1527 Severity : enhancement
1529 Description: parallel, asynchronous orphan cleanup
1530 Details : orphan cleanup is now performed in separate threads for each
1531 OST, allowing parallel non-blocking operation.
1533 Severity : enhancement
1535 Description: optimized stripe assignment
1536 Details : stripe assignments are now made based on ost space available,
1537 ost previous usage, and OSS previous usage, in order to try
1538 to optimize storage space and networking resources.
1540 Severity : enhancement
1542 Description: Permanently set tunables
1543 Details : All writable /proc/fs/lustre tunables can now be permanently
1544 set on a per-server basis, at mkfs time or on a live system.
1546 Severity : enhancement
1548 Description: Lustre message v2
1549 Details : Add lustre message format v2.
1551 Severity : enhancement
1553 Description: client OST exclusion list
1554 Details : Clients can be started with a list of OSTs that should be
1555 declared "inactive" for known non-responsive OSTs.
1559 Description: SPEC SFS validation failure on NFS v2 over lustre.
1560 Details : Changes the blocksize for regular files to be 2x RPC size,
1561 and not depend on stripe size.
1563 Severity : enhancement
1565 Description: Multiple MD RPCs in flight.
1566 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1567 To avoid overly-overloading MDS, introduce a limit on number of
1568 MDS RPCs in flight for a single client and add /proc controls
1569 to adjust this limit.
1571 Severity : enhancement
1573 Description: client read/write statistics
1574 Details : Add client read/write call usage stats for performance
1575 analysis of user processes.
1576 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1577 file access. extents_stats shows chunk size distribution.
1578 extents_stats_per_process show chunk size distribution per
1581 Severity : enhancement
1583 Description: mds statistics
1584 Details : Add detailed mds operations statistics in
1585 /proc/fs/lustre/mds/*/stats.
1589 Description: Failure of copying files with lustre special EAs.
1590 Details : Client side always return success for setxattr call for lustre
1591 special xattr (currently only "trusted.lov").
1596 Description: Refcount LNET uuids
1597 Details : The global LNET uuid list grew linearly with every startup;
1598 refcount repeated list entries instead of always adding to
1601 Severity : enhancement
1603 Description: Dynamic service threads
1604 Details : Within a small range, start extra service threads
1605 automatically when the request queue builds up.
1608 Frequency : mixed-endian client/server environments
1610 Description: mixed-endian crashes
1611 Details : The new msg_v2 system had some failures in mixed-endian
1614 Severity : enhancement
1616 Description: Easy OST removal
1617 Details : OSTs can be permanently deactivated with e.g. 'lctl
1618 conf_param lustre-OST0001.osc.active=0'
1620 Severity : enhancement
1622 Description: MGS proc entries
1623 Details : Added basic proc entries for the MGS showing what filesystems
1626 Severity : enhancement
1628 Description: provide MGS failover
1629 Details : Added config lock reacquisition after MGS server failover.
1631 Severity : enhancement
1633 Description: add Linux 2.4 support
1634 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1638 Description: a large application tries to do I/O to the same resource and dies
1639 in the middle of it.
1640 Details : Check the req->rq_arrival time after the call to
1641 ost_brw_lock_get(), but before we do anything about
1642 processing it & sending the BULK transfer request. This
1643 should help move old stale pending locks off the queue as
1644 quickly as obd_timeout.
1647 Frequency : when an incorrect nid is specified during startup
1649 Description: ptlrpc connect to non-existant node causes kernel crash
1650 Details : LNET can't be re-entered from an event callback, which
1651 happened when we expire a message after the export has been
1652 cleaned up. Instead, hand the zombie cleanup off to another
1655 Severity : enhancement
1657 Description: plain/inodebits lock performance improvement
1658 Details : Grouping plain/inodebits in granted list by their request modes
1659 and bits policy, thus improving the performance of search through
1663 Frequency : only if OST filesystem is corrupted
1665 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1666 Details : for a short time RPCs with bulk IO are in the replay list,
1667 but replay of bulk IOs is unimplemented. If the OST filesystem
1668 is corrupted due to disk cache incoherency and then replay is
1669 started it is possible to trip an assertion. Avoid putting
1670 committed RPCs into the replay list at all to avoid this issue.
1673 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1675 Description: System hang on startup
1676 Details : This bug allowed the liblustre (e.g. catamount) client to
1677 return to the app before handling all startup RPCs. This
1678 could leave the node unresponsive to lustre network traffic
1679 and manifested as a server ptllnd timeout.
1681 Severity : enhancement
1683 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1684 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1685 boolean to control whether to print peer debug info when a
1686 client's RPC times out.
1689 Frequency : only for kernels with patches from Lustre below 1.4.3
1691 Description: Remove old rdonly API
1692 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1695 Frequency : only for devices with external journals
1697 Description: Set external device read-only also
1698 Details : During a commanded failover stop, we set the disk device
1699 read-only while the server shuts down. We now also set any
1700 external journal device read-only at the same time.
1703 Frequency : when upgrading from 1.4 while trying to change parameters
1705 Description: The wrong (new) MDC name was used when setting parameters for
1706 upgraded MDT's. Also allows changing of OSC (and MDC)
1707 parameters if --writeconf is specified at tunefs upgrade time.
1710 Frequency : when setting specific ost indicies
1712 Description: QOS code breaks on skipped indicies
1713 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1714 created with --index need not be sequential.
1716 Severity : enhancement
1718 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1719 Details : The uninit_groups feature works in conjunction with the kernel
1720 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1721 up the pass1 processing of e2fsck. This is a read-only feature
1722 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1723 mount filesystems that have had this feature enabled.
1725 Severity : enhancement
1727 Description: Improve multi-block allocation algorithm to avoid fragmentation
1728 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1729 allocation locality and avoid filesystem fragmentation.
1731 ------------------------------------------------------------------------------
1733 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1735 * Support for kernels:
1736 2.4.21-47.0.1.EL (RHEL 3)
1737 2.6.5-7.283 (SLES 9)
1738 2.6.9-42.0.10.EL (RHEL 4)
1739 2.6.12.6 vanilla (kernel.org)
1740 2.6.16.27-0.9 (SLES 10)
1741 * Recommended e2fsprogs version: 1.39.cfs5
1743 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1747 Frequency : occasional, depends on client load and configuration
1748 Bugzilla : 12181, 12203
1749 Description: data loss for recently-modified files
1751 Details : In some cases it is possible that recently written or created
1752 files may not be written to disk in a timely manner (this should
1753 normally be within 30s unless client IO load is very high).
1754 The problem appears as zero-length files or files that are a
1755 multiple of 1MB in size after a client crash or client eviction
1756 that are missing data at the end of the file.
1758 This problem is more likely to be hit on clients where files are
1759 repeatedly created and unlinked in the same directory, clients
1760 have a large amount of RAM, have many CPUs, the filesystem has
1761 many OSTs, the clients are rebooted frequently, and/or the files
1762 are not accessed by other nodes after being written.
1764 The presence of the problem can be detected by looking at
1765 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1766 smaller than the second (nr_unused) then dirty files will not
1767 be flushed automatically to disk. "sync; sleep 10" should be
1768 run several times on the node before unmounting it to update
1769 Lustre (this is also safe to run on nodes without this problem).
1771 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1772 that can cause this same problem, so customers using that kernel
1773 also need to update the kernel in addition to Lustre. In order
1774 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1776 It is normal that files written just before a client crash (less
1777 than 30s) may not yet have been flushed to disk, even for local
1781 Frequency : frequent on thin XT3 nodes
1783 Description: UUID collision on thin XT3 Linux nodes
1784 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1785 randomly, since we relied on an insufficiently-seeded PRNG.
1790 Description: OSS hangs after "All ost request buffers busy"
1791 Details : A deadlock between quota and journal operations caused OSS
1792 hangs after printing "All ost request buffers busy."
1795 Frequency : always on liblustre builds
1797 Description: Cleanup compiler warnings on liblustre
1800 Frequency : always on liblustre builds on XT3
1802 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1806 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1807 > 32000 subdirectories
1808 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1809 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1810 subdirectory count crosses 32000. This will aid e2fsck to
1811 correctly handle more than 32000 subdirectories.
1814 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1816 Description: System hang on startup
1817 Details : This bug allowed the liblustre (e.g. catamount) client to
1818 return to the app before handling all startup RPCs. This
1819 could leave the node unresponsive to lustre network traffic
1820 and manifested as a server ptllnd timeout.
1822 Severity : enhancement
1824 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1825 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1826 boolean to control whether to print peer debug info when a
1827 client's RPC times out.
1832 Description: make O_SYNC working on 2.6 kernels
1833 Details : 2.6 kernels use different method for mark pages for write,
1834 so need add a code to lustre for O_SYNC work.
1839 Description: Failure to close file and release space on NFS
1840 Details : Put inode details into lock acquired in ll_intent_file_open.
1841 Use mdc_intent_lock in ll_intent_open to properly
1842 detect all kind of errors unhandled by mdc_enqueue.
1847 Description: proc file read during shutdown sometimes raced obd removal,
1849 Details : Add lock to prevent obd access after proc file removal.
1852 Frequency : Only for files larger than 4GB on 32-bit clients.
1854 Description: improperly doing page alignment of locks
1855 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1856 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1861 Description: RPCs being resent when they shouldn't be
1862 Details : Some RPCs that should not be resent are being resent. This
1863 can cause inconsistencies in the RPC state machine. Do not
1864 resend such requests.
1867 Frequency : rare, only with NFS export
1869 Description: Crash on NFS re-export node
1870 Details : under very unusual load conditions an assertion is hit in
1871 ll_intent_file_open()
1874 Frequency : only if OST filesystem is corrupted
1876 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1877 Details : for a short time RPCs with bulk IO are in the replay list,
1878 but replay of bulk IOs is unimplemented. If the OST filesystem
1879 is corrupted due to disk cache incoherency and then replay is
1880 started it is possible to trip an assertion. Avoid putting
1881 committed RPCs into the replay list at all to avoid this issue.
1886 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1887 Details : Large single O_DIRECT read and write calls can fail to allocate
1888 a sufficiently large buffer to process the request. In case of
1889 allocation failure the allocation is retried with a smaller
1890 buffer and broken into smaller requests.
1892 Severity : enhancement
1894 Description: Add -o localflock option to simulate old noflock behaviour.
1895 Details : This will achieve local-only flock/fcntl locks coherentness.
1900 Description: versioning check is incomplete
1901 Details : Checking the version difference of client vs. server, report
1902 error if the gap is too big.
1907 Description: add support PG_writeback bit
1908 Details : add support for PG_writeback bit for Lustre, for more carefull
1909 work with page cache in 2.6 kernel. This also fix some deadlocks
1910 and remove hack for work O_SYNC with 2.6 kernel.
1912 Severity : enhancement
1914 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1915 Details : The uninit_groups feature works in conjunction with the kernel
1916 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1917 up the pass1 processing of e2fsck. This is a read-only feature
1918 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1919 mount filesystems that have had this feature enabled.
1921 Severity : enhancement
1923 Description: Improve multi-block allocation algorithm to avoid fragmentation
1924 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1925 allocation locality and avoid filesystem fragmentation.
1927 ------------------------------------------------------------------------------
1929 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1931 * Support for kernels:
1932 2.6.9-42.0.3.EL (RHEL 4)
1933 2.6.5-7.276 (SLES 9)
1934 2.4.21-47.0.1.EL (RHEL 3)
1935 2.6.12.6 vanilla (kernel.org)
1936 2.6.16.21-0.8 (SLES10)
1937 * Recommended e2fsprogs version: 1.39.cfs2-0
1939 * The backwards-compatible /proc/sys/portals symlink has been removed
1940 in this release. Before upgrading, please ensure that you change
1941 any configuration scripts or /etc/sysctl.conf files that access
1942 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1943 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1944 in advance of the upgrade on any system running Lustre 1.4.6 or
1945 newer, since /proc/sys/lnet was added in that version.
1946 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1950 Frequency : only when quota is used
1952 Description: avoid scanning export list for quota master
1953 Details : Change the algorithms to avoid scanning export list in order
1954 to improve the efficiency.
1957 Frequency : MDS failover only, very rarely
1959 Description: "went back in time" messages on mds failover
1960 Details : The greatest transno may be lost when the current operation
1961 finishes with an error (transno==0) and the client's last_rcvd
1962 record is over-written. Save the greatest transno in the
1963 mds_last_transno for this case.
1966 Frequency : always for specific kernels and striping counts
1968 Description: client may get "Matching packet too big" without ACL support
1969 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1970 when trying to access files in certain configurations. The
1971 clients should in fact be denied when mounting because they do
1972 not understand ACLs.
1975 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1977 Description: many clients connecting with IO in progress causes connect timeouts
1978 Details : Avoid synchronous journal commits to avoid delays caused by many
1979 clients connecting/disconnecting when bulk IO is in progress.
1980 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1981 OST_IO_PORTAL to avoid delays behind potentially many pending
1985 Frequency : occasionally with multiple writers to a single file
1987 Description: shared writes to file may result in wrong size reported by stat()
1988 Details : Allow growing of kms when extent lock is cancelled
1991 Frequency : always with random mmap IO to multi-striped file
1993 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1994 Details : Only if the hole is at the end of OST object so that kms is too
1995 small. Fix is to increase kms accordingly in ll_nopage.
1998 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2000 Description: writes to a missing object would leak memory on the OST
2001 Details : If there is an inconsistency between the MDS and OST filesystems,
2002 such that the MDS references an object that doesn't exist, writes
2003 to that object will leak memory due to incorrect cleanup in the
2004 error handling path, eventually running out of memory on the OST.
2009 Description: Creating too long symlink causes lustre errors
2010 Details : Check symlink and name lengths before sending requests to MDS.
2013 Frequency : only if flock is enabled (not on by default)
2015 Description: posix locks not released on fd closure on 2.6.9+
2016 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2017 kernels, this caused such locks not to be released on fd close and
2018 then assertions on fs unmount about still used locks.
2021 Frequency : MDS failover only, very rarely
2023 Description: clients may get ASSERTION(granted_lock != NULL)
2024 Details : When request was taking a long time, and a client was resending
2025 a getattr by name lock request. The were multiple lock requests
2026 with the same client lock handle and
2027 mds_getattr_name->fixup_handle_for_resent_request found one of the
2028 lock handles but later failed with ASSERTION(granted_lock != NULL).
2033 Description: handle->h_buffer_credits > 0, assertion failure
2034 Details : h_buffer_credits is zero after truncate, causing assertion
2035 failure. This patch extends the transaction or creates a new
2039 Frequency : NFS re-export or patchless client
2040 Bugzilla : 11179, 10796
2041 Description: Crash on NFS re-export node (__d_move)
2042 Details : We do not want to hash the dentry if we don't have a lock.
2043 But if this dentry is later used in d_move, we'd hit uninitialised
2044 list head d_hash, so we just do this to init d_hash field but
2045 leave dentry unhashed.
2048 Frequency : NFS re-export or patchless client
2050 Description: NFS exports has problem with symbolic link
2051 Details : lustre client didn't properly install dentry when re-exported
2052 to NFS or running patchless client.
2055 Frequency : NFS re-export or patchless client
2057 Description: Various nfs/patchless fixes.
2058 Details : fixes reuse disconected alias for lookup process - this fixes
2059 warning "find_exported_dentry: npd != pd",
2060 fix permission error with open files at nfs.
2061 fix apply umask when do revalidate.
2064 Frequency : occasional
2066 Description: Crash on NFS re-export node
2067 Details : calling clear_page() on the wrong pointer triggered oops in
2068 generic_mapping_read().
2071 Frequency : rarely, using O_DIRECT IO
2073 Description: unaligned directio crashes client with LASSERT
2074 Details : check for unaligned buffers before trying any requests.
2077 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2079 Description: stale data returned from RAID cache
2080 Details : If only a small amount of IO is done to the RAID device before
2081 reading it again it is possible to get stale data from the RAID
2082 cache instead of reading it from disk.
2085 Frequency : always for sles10 kernel
2087 Description: sles10 support
2088 Details : ll_follow_link: compile fixes and using of nd_set_link
2089 under newer kernels.
2092 Frequency : depends on arch, kernel and compiler version, always on sles10
2095 Description: recursive or deep enough symlinks cause stack overflow
2096 Details : getting rid of large stack-allocated variable in
2100 Frequency : depends on hardware
2102 Description: lustre write performance loss in the SLES10 kernel
2103 Details : the performance loss is caused by using of write barriers in the
2104 ext3 code. The SLES10 kernel turns barrier support on by
2105 default. The fix is to undo that change for ldiskfs.
2107 ------------------------------------------------------------------------------
2109 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2111 * Support for kernels:
2112 2.6.9-42.0.3EL (RHEL 4)
2113 2.6.5-7.276 (SLES 9)
2114 2.4.21-47.0.1.EL (RHEL 3)
2115 2.6.12.6 vanilla (kernel.org)
2119 Frequency : quota enabled and large files being deleted
2121 Description: releasing more than 4GB of quota at once hangs OST
2122 Details : If a user deletes more than 4GB of files on a single OST it
2123 will cause the OST to spin in an infinite loop. Release
2124 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2129 Description: statfs data retrieved from /proc may be stale or zero
2130 Details : When reading per-device statfs data from /proc, in the
2131 {kbytes,files}_{total,free,avail} files, it may appear
2132 as zero or be out of date.
2135 Frequency : systems with MD RAID1 external journal devices
2137 Description: lconf's call to blkid is confused by RAID1 journal devices
2138 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2139 of returning all block devices that match the journal UUID.
2142 Frequency : always, for aggregate stripe size over 4GB
2144 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2145 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2146 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2149 Frequency : always if "lfs find" used on a local file/directory
2151 Description: "lfs find" segfaults if used on a local file/directory
2152 Details : The case where a directory component was not specified wasn't
2153 handled correctly. Handle this properly.
2156 Frequency : always on ppc64
2158 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2159 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2160 perform endianness conversion.
2163 Frequency : rarely (truncate to non-zero file size after write under load)
2164 Bugzilla : 10730, 10687
2165 Description: Files padded with zeros to next 4K multiple
2166 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2167 it is possible that files that are truncated to a non-zero size
2168 immediately after being written are filled with zero bytes beyond
2169 the truncated size. No file data is lost.
2171 Severity : enhancement
2173 Description: Allow recovery/failover for liblustre clients.
2174 Details : liblustre clients were unaware of failover configurations until
2177 Severity : enhancement
2179 Description: user file locks should fail when not mounting with flock option
2180 Details : Set up an error-returning stub in ll_file_operations.lock field
2181 to prevent incorrect behaviour when client is mounted without
2182 flock option. Also, set up properly f_op->flock field for
2186 Frequency : always on ia64
2188 Description: "lfs df" loops on printing out MDS statfs information
2189 Details : The obd_ioctl_data was not initialized and in some systems
2190 this caused a failure during the ioctl that did not return
2191 an error. Initialize the struct and return an error on failure.
2194 Frequency : SLES 9 only
2196 Description: Error of copying files with lustre special EAs as root
2197 Details : Client side always return success for setxattr call for lustre
2198 special xattr (currently only "trusted.lov").
2201 Frequency : rarely on clusters with both ia64+i386 clients
2203 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2204 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2205 clients are doing concurrent writes to the same file it is
2206 possible that the ia64 clients may LASSERT because the OST
2207 extent locks are not PAGE_SIZE aligned. Ensure that grown
2208 locks are always aligned on the request boundary.
2211 Frequency : specific use, occasional
2213 Description: Overwriting in use executable truncates on-disk binary image
2214 Details : If one node attempts to overwrite an executable in use by
2215 another node, we now correctly return ETXTBSY instead of
2216 truncating the file.
2218 Severity : enhancement
2220 Description: Async OSC create to avoid the blocking unnecessarily.
2221 Details : If a OST has no remain object, system will block on the creating
2222 when need to create a new object on this OST. Now, ways use
2223 pre-created objects when available, instead of blocking on an
2224 empty osc while others are not empty. If we must block, we block
2225 for the shortest possible period of time.
2230 Description: chmod on Lustre root is propagated to other clients
2231 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2232 invalid by the follow_mount time.
2235 Frequency : liblustre clients only
2237 Description: Race in 'instant cancel' lock handling could lead to such locks
2238 never to be granted in case of SMP MDS
2239 Details : Do not destroy not yet granted but cbpending locks in
2243 Frequency : replay/resend of open
2245 Description: non null lock assetion failure in mds_intent_policy
2246 Details : Trying to replay/resend lockless open requests resulted in
2247 mds_open() returning 0 with no lock. Now it sets a flag if
2248 a lock is going to be returned.
2250 Severity : enhancement
2252 Description: Checksum enhancements
2253 Details : New checksum enhancements allow for resending RPCs that failed
2256 Severity : enhancement
2258 Description: Tunables on number of dirty pages in cacche
2259 Details : Allow to set limit on number of dirty pages cached.
2264 Description: client crash on unmount - lock still has references
2265 Details : In some error handling cases it was possible to leak a lock
2266 reference on a client while accessing a file. This was not
2267 harmful to the client during operation, but would cause the
2268 client to crash when the filesystem is unmounted.
2271 Frequency : specific case, rare
2273 Description: ETXTBSY on mds though file not in use by client
2274 Details : ETXTBSY is no longer incorrectly returned when attempting to
2275 chmod or chown a directory that the user previously tried to
2276 execute or a currently-executing binary.
2279 Frequency : extremely rare except on liblustre-based clients
2281 Description: Lustre space not freed when files are deleted
2282 Details : Clean up open-unlinked files after client eviction. Previously
2283 the unlink was skipped and the files remained as orphans.
2288 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2289 Details : In some cases it was possible to send a blocking callback to
2290 a client doing a glimpse, even though that client didn't get
2291 a lock granted. When the glimpse lock is cancelled on the OST
2292 the freed lock is left on the waiting list and corrupted the list.
2295 Frequency : all core dumps
2297 Description: Broke core dumps to lustre
2298 Details : Negative dentry may be unhashed if parent does not have UPDATE
2299 lock, but some callers, e.g. do_coredump, expect dentry to be
2300 hashed after successful create, hash it in ll_create_it.
2302 ------------------------------------------------------------------------------
2304 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2306 * Support for kernels:
2307 2.6.9-42.0.2.EL (RHEL 4)
2308 2.6.5-7.276 (SLES 9)
2309 2.4.21-40.EL (RHEL 3)
2310 2.6.12.6 vanilla (kernel.org)
2314 Frequency : always on RHEL 3
2316 Description: Number of open files grows over time
2317 Details : The number of open files grows over time, whether or not
2318 Lustre is started. This was due to a filp leak introduced
2319 by one of our kernel patches.
2321 ------------------------------------------------------------------------------
2323 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2325 * Support for kernels:
2326 2.6.9-42.EL (RHEL 4)
2327 2.6.5-7.276 (SLES 9)
2328 2.4.21-40.EL (RHEL 3)
2329 2.6.12.6 vanilla (kernel.org)
2334 Bugzilla : 5719, 9635, 9792, 9684
2335 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2336 Details : If a server is under heavy load and cannot reply to new
2337 connection requests before the client resends the (re)connect,
2338 the connection handling code can behave badly if two service
2339 threads are concurrently handing separate (re)connections from
2340 the same client. Add better locking to the connection handling
2341 code, and ensure that only a single connection will be processed
2342 for a given client UUID, even if the lock is dropped.
2344 Severity : enhancement
2346 Description: add TCP zero-copy support to kernel
2347 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2348 sends if the hardware supports scatter-gather and checksumming.
2349 This allows socklnd to do client-write and server-read more
2350 efficiently and reduce CPU utilization from skbuf copying.
2353 Frequency : only if NFS exporting from client
2355 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2356 Details : If setting attributes on a file created under NFS that had
2357 never been opened it would be possible to oops the client
2358 if the file had no objects.
2361 Frequency : always for liblustre
2363 Description: liblustre client does MDS+OSTs setattr RPC for each write
2364 Details : When doing a write from a liblustre client, the client
2365 incorrectly issued an RPC to the MDS and each OST the file was
2366 striped over in order to update the timestamps. When writing
2367 with small chunks and many clients this could overwhelm the MDS
2368 with RPCs. In all cases it would slow down the write because
2369 these RPCs are unnecessary.
2371 Severity : enhancement
2373 Description: allow number of MDS service threads to be changed at module load
2374 Details : It is now possible to change the number of MDS service threads
2375 running. Adding "options mds mds_num_threads={N}" to the MDS's
2376 /etc/modprobe.conf will set the number of threads for the next
2377 time Lustre is restarted (assuming the "mds" module is also
2378 reloaded at that time). The default number of threads will
2379 stay the same, 32 for most systems.
2384 Description: OST crash if filesystem is unformatted or corrupt
2385 Details : If an OST is started on a device that has never been formatted
2386 or if the filesystem is corrupt and cannot even mount then the
2387 error handling cleanup routines would dereference a NULL pointer.
2392 Description: NULL pointer deref in llap_from_page.
2393 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2394 incorrect) mapping if the page was truncated/reclaimed while it was
2395 searched for. Check for this condition and skip such pages when
2396 doing readahead. Introduce extra check to llap_from_page() to
2397 verify page->mapping->host is non-NULL (so page is not anonymous).
2400 Frequency : Sometimes when using sys_sendfile
2402 Description: "page not covered by a lock" warnings from ll_readpage
2403 Details : sendfile called ll_readpage without right page locks present.
2404 Now we introduced ll_file_sendfile that does necessary locking
2405 around call to generic_file_sendfile() much like we do in
2409 Frequency : with certain MDS communication failures at client mount time
2411 Description: NULL pointer deref after failed client mount
2412 Details : a client connection request may delayed by the network layer
2413 and not be sent until after the PTLRPC layer has timed out the
2414 request. If the client fails the mount immediately it will try
2415 to clean up before the network times out the request. Add a
2416 reference from the request import to the obd device and delay
2417 the cleanup until the network drops the request.
2420 Frequency : occasionally during client (re)connect
2422 Description: assertion failure during client (re)connect
2423 Details : processing a client connection request may be delayed by the
2424 client or server longer than the client connect timeout. This
2425 causes the client to resend the connection request. If the
2426 original connection request is replied in this interval, the
2427 client may trip an assertion failure in ptlrpc_connect_interpret()
2428 which thought it would be the only running connect process.
2431 Frequency : only with obd_echo servers and clients that are rebooted
2433 Description: kernel BUG accessing uninitialized data structure
2434 Details : When running an obd_echo server it did not start the ping_evictor
2435 thread, and when a client was evicted an uninitialized data
2436 structure was accessed. Start the ping_evictor in the RPC
2437 service startup instead of the OBD startup.
2439 Severity : enhancement
2440 Bugzilla : 10193 (patchless)
2441 Description: Remove dependency on various unexported kernel interfaces.
2442 Details : No longer need reparent_to_init, exit_mm, exit_files,
2443 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2446 Frequency : rare (only users of deprecated and unsupported LDAP config)
2448 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2449 Details : LDAP apparently contains 'lustreName' attributes instead of
2450 'name'. A simple remapping of the name is sufficient.
2453 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2455 Description: waiting_locks_callback trips kernel BUG if client is evicted
2456 Details : Running with the dump_on_timeout debug flag turned on makes
2457 it possible that the waiting_locks_callback() can try to dump
2458 the Lustre kernel debug logs from an interrupt handler. Defer
2459 this log dumping to the expired_lock_main() thread.
2461 Severity : enhancement
2463 Description: Support NFS exporting on 2.6 kernels.
2464 Details : Implement non-rawops metadata methods for NFS server to use without
2465 changing NFS server code.
2468 Frequency : very rare (synthetic metadata workload only)
2470 Description: two racing renames might cause an MDS thread to deadlock
2471 Details : Running the "racer" program may cause one MDS thread to rename
2472 a file from being the source of a rename to being the target of
2473 a rename at exactly the same time that another thread is doing
2474 so, and the second thread has already enqueued these locks after
2475 doing a lookup of the target and is trying to relock them in
2476 order. Ensure that we don't try to re-lock the same resource.
2479 Frequency : only very large systems with liblustre clients
2481 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2482 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2483 all OSTs in parallel. This allows the eviction of stale liblustre
2484 clients to proceed much faster than if they were done in series,
2485 and also offers similar improvements for other set_info RPCs.
2490 Description: excessive CPU usage during initial read phase on client
2491 Details : During the initial read phase on a client, it would agressively
2492 retry readahead on the file, consuming too much CPU and impacting
2493 performance (since 1.4.5.8). Improve the readahead algorithm
2494 to avoid this, and also improve some other common cases (read
2495 of small files in particular, where "small" is files smaller than
2496 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2501 Description: MDS crash when receiving packet with unknown intent.
2502 Details : Do not LBUG in unknown intent case, just return -EFAULT
2504 Severity : enhancement
2505 Bugzilla : 9293, 9385
2506 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2507 Details : Do not serialize getattr (non-intent version) and statfs.
2510 Frequency : occasional, when OST network is overloaded/intermittent
2512 Description: client evicted by OST after bulk IO timeout
2513 Details : If a client sends a bulk IO request (read or write) the OST
2514 may evict the client if it is unresposive to its data GET/PUT
2515 request. This is incorrect if the network is overloaded (takes
2516 too long to transfer the RPC data) or dropped the OST GET/PUT
2517 request. There is no need to evict the client at all, since
2518 the pinger and/or lock callbacks will handle this, and the
2519 client can restart the bulk request.
2522 Frequency : Always when mmapping file with no objects
2524 Description: client crashes when mmapping file with no objects
2525 Details : Check that we actually have objects in a file before doing any
2526 operations on objects in ll_vm_open, ll_vm_close and
2532 Description: Request leak when working with deleted CWD
2533 Details : Introduce advanced request refcount tracking for requests
2534 referenced from lustre intent.
2536 Severity : Enhancement
2538 Description: Cache open file handles on client.
2539 Details : MDS now will return special lock along with openhandle, if
2540 requested and client is allowed to hold openhandle, even if unused,
2541 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2542 closing files for every read/write openration.
2544 Severity : Enhancement
2546 Description: Cache open negative dentries on client when possible.
2547 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2548 negative dentries on lock revocation.
2553 Description: Remounting a client read-only wasn't possible with a zconf mount
2554 Details : It wasn't possible to remount a client read-only with llmount.
2556 Severity : enhancement
2557 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2558 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2559 MPICH is included to add Lustre-specific ADIO interfaces.
2560 This is based closely on the UFS ADIO layer and only differs
2561 in file creation, in order to allow the OST striping to be set.
2562 This is user-contributed code and not supported by CFS.
2567 Description: extended inode attributes (immutable, append-only) work improperly
2568 when 2.4 and 2.6 kernels are used on client/server or vice versa
2569 Details : Introduce kernel-independent values for these flags.
2571 Severity : enhancement
2574 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2575 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2576 granularity. Now, Fractional MB granularity is be supported,
2577 this is very useful for low memory system.
2579 Severity : enhancement
2581 Description: Getattr by fid
2582 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2583 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2584 have locks and avoids getattr rpc requests for it on every stat.
2587 Frequency : Always, for filesystems larger than 2TB
2589 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2590 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2591 filesystems larger than 2TB to be mounted with mballoc enabled.
2594 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2596 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2597 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2598 filesystems larger than 2TB is mounted then the kernel will
2599 silently corrupt the start of the filesystem. CONFIG_LBD is
2600 enabled for all CFS-supported kernels, but the possibility of
2601 this happening with a modified kernel config exists.
2603 Severity : enhancement
2605 Description: add client O_DIRECT support for 2.6 kernels
2606 Details : It is now possible to do O_DIRECT reads and writes to files
2607 in the Lustre client mountpoint on 2.6 kernel clients.
2609 Severity : enhancement
2611 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2612 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2613 parallel, not waiting for response from every OST before sending
2614 a rpc to the next OST.
2619 Description: setattr vs write race when updating file timestamps
2620 Details : Client processes that update a file timestamp into the past
2621 right after writing to the file (e.g. tar) it is possible that
2622 the updated file modification time can be reset to the current
2623 time due to a race between processing the setattr and write RPC.
2625 Severity : enhancement
2627 Description: Bring 'lfs find' closer in line with regular Linux find.
2628 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2629 -print0 options and obtains all the needed info through the lustre
2632 Severity : enhancement
2634 Description: support up to 1024 configured devices on one node
2635 Details : change obd_dev array from statically allocated to dynamically
2636 allocated structs as they are first used to reduce memory usage
2641 Description: Flush dirty partially truncated pages during truncate
2642 Details : Immediatelly flush partially truncated pages in filter_setattr,
2643 this way we completely avoid having any pages in page cache on OST
2644 and can retire ugly workarounds during writes to flush such pages.
2649 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2650 Details : protect inode from truncation within vfs_unlink() context
2651 just take a reference before calling vfs_unlink() and release it
2652 when parent's i_sem is free.
2657 Description: last_id value checked outside lock on OST caused LASSERT failure
2658 Details : If there were multiple MDS->OST object precreate requests in
2659 flight, it was possible that the OST's last object id was checked
2660 outside a lock and incorrectly tripped an assertion. Move checks
2661 inside locks, and discard old precreate requests.
2664 Frequency : always, if extents are used on OSTs
2666 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2667 Details : OSTs using the extents format would not zero the high 16 bits of
2668 the index physical block number. This is not a problem for any
2669 OST filesystems smaller than 16TB, and no kernels support ext3
2670 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2671 new/modified files) and can be fixed for existing filesystems
2672 with e2fsprogs-1.39-cfs1.
2677 Description: import connection selection may be incorrect if timer wraps
2678 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2679 import connections to be ignored if the 32-bit jiffies counter
2680 wraps. Use a 64-bit jiffies counter.
2683 Frequency : very large clusters immediately after boot
2685 Description: LNET request buffers exhausted under heavy short-term load
2686 Details : If a large number of client requests are generated on a service
2687 that has previously never seen so many requests it is possible
2688 that the request buffer growth cannot keep up with the spike in
2689 demand. Instead of dropping incoming requests, they are held in
2690 the LND until the RPC service can accept more requests.
2693 Frequency : Sometimes during replay
2695 Description: Assertion failure in ll_local_open after replay.
2696 Details : If replay happened on an open request reply before we were able
2697 to set replay handler, reply will become not swabbed tripping the
2698 assertion in ll_local_open. Now we set the handler right after
2699 recognising of open request
2702 Frequency : very rare
2704 Description: kernel reports "badness in vsnprintf"
2705 Details : Reading from the "recovery_status" /proc file in small chunks
2706 may cause a negative length in lprocfs_obd_rd_recovery_status()
2707 call to vsnprintf() (which is otherwise harmless). Exit early
2708 if there is no more space in the output buffer.
2710 Severity : enhancement
2712 Description: clear OBD RPC statistics by writing to them
2713 Details : It is now possible to clear the OBD RPC statistics by writing
2714 to the "stats" file.
2719 Description: Client mtime is not the same on different clients after utimes
2720 Details : In some cases, the client was using the utimes() syscall on
2721 a file cached on another node. The clients now validate the
2722 ctime from the MDS + OSTs to determine which one is right.
2727 Description: Inability to activate failout mode
2728 Details : lconf script incorrectly assumed that in pythong string's numeric
2729 value is used in comparisons.
2732 Frequency : always with multiple stripes per file
2734 Description: Inefficient object allocation for mutli-stripe files
2735 Details : When selecting which OSTs to stripe files over, for files with
2736 a stripe count that divides evenly into the number of OSTs,
2737 the MDS is always picking the same starting OST for each file.
2738 Return the OST selection heuristic to the original design.
2743 Description: mount failures may take full timeout to return an error
2744 Details : Under some heavy load conditions it is possible that a
2745 failed mount can wait for the full obd_timeout interval,
2746 possibly several minutes, before reporting an error.
2747 Instead return an error as soon as the status is known.
2749 Frequency : quota enabled and large files being deleted
2751 Description: releasing more than 4GB of quota at once hangs OST
2752 Details : If a user deletes more than 4GB of files on a single OST it
2753 will cause the OST to spin in an infinite loop. Release
2754 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2759 Description: statfs data retrieved from /proc may be stale or zero
2760 Details : When reading per-device statfs data from /proc, in the
2761 {kbytes,files}_{total,free,avail} files, it may appear
2762 as zero or be out of date.
2765 Frequency : systems with MD RAID1 external journal devices
2767 Description: lconf's call to blkid is confused by RAID1 journal devices
2768 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2769 of returning all block devices that match the journal UUID.
2772 Frequency : always, for aggregate stripe size over 4GB
2774 Description: assertion fails when trying to use 4GB stripe size
2775 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2776 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2779 Frequency : always on ppc64
2781 Description: the first write on an ext3 filesystem with mballoc got stuck
2782 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2783 perform endianness conversion.
2785 ------------------------------------------------------------------------------
2787 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2789 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2790 INTEROPERATE with older versions automatically. Please read the
2791 user documentation before upgrading any part of a live system.
2792 * WARNING: Lustre networking configuration changes are required with
2793 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2796 * Support for kernels:
2797 2.6.9-22.0.2.EL (RHEL 4)
2798 2.6.5-7.244 (SLES 9)
2799 2.6.12.6 vanilla (kernel.org)
2802 Severity : enhancement
2803 Bugzilla : 7981/8208
2804 Description: Introduced Lustre Networking (LNET)
2805 Details : LNET is new networking infrastructure for Lustre, it includes
2806 a reorganized network configuration mode (see the user
2807 documentation for full details) as well as support for routing
2808 between different network fabrics. Lustre Networking Devices
2809 (LNDs) for the supported network fabrics have also been
2810 created for this new infrastructure.
2812 Severity : enhancement
2813 Description: Introduced Access control lists
2814 Details : clients can set ACLs on files and directories in order to have
2815 more fine-grained permissions than the standard Unix UGO+RWX.
2816 The MDS must be started with the "-o acl" mount option.
2818 Severity : enhancement
2819 Description: Introduced filesystem quotas
2820 Details : Administrators may now establish per-user quotas on the
2823 Severity : enhancement
2825 Description: Configuration change for the XT3
2826 The PTLLND is now used to run Lustre over Portals on the XT3
2827 The configure option(s) --with-cray-portals are no longer used.
2828 Rather --with-portals=<path-to-portals-includes> is used to
2829 enable building on the XT3. In addition to enable XT3 specific
2830 features the option --enable-cray-xt3 must be used.
2835 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2836 Details : A race between allocating a new llcd and re-getting the llcd_lock
2837 allowed another thread to grab newly-allocated llcd.
2839 Severity : enhancement
2841 Description: 2.6 OST async journal commit and locking fix to improve performance
2842 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2843 2.6 kernels are now async as they already were in 2.4 kernels so
2844 that they can commit concurrently with the network bulk transfer.
2845 For block-allocated files the filter allocation semaphore is held
2846 to avoid filesystem fragmentation during allocation. BKL lock
2847 removed for 2.6 xattr operations where it is no longer needed.
2852 Description: lconf incorrectly determined whether two IP networks could talk
2853 Details : In some more complicated routing and multiple-network
2854 configurations, lconf will avoid trying to make a network
2855 connection to a disjoint part of the IP space. It was doing the
2856 math incorrectly for one set of cases.
2861 Description: Fix for potential infinite loop processing records in an llog.
2862 Details : If an llog record is corrupted/zeroed, it is possible to loop
2863 forever in llog_process(). Validate the llog record length
2864 and skip the remainder of the block on error.
2867 Frequency : occasional (liblustre only)
2869 Description: liblustre could not open files whose last component is a symlink
2870 Details : sysio_path_walk() would incorrectly pass the open intent to
2871 intermediate path components.
2874 Frequency : rare (liblustre only with non-standard tuning)
2875 Bugzilla : 7201 (7350)
2876 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2877 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2878 after it has been released, only do this for referenced locks
2880 Severity : enhancement
2882 Description: specifying an (invalid) directory default stripe_size of -1
2883 would reset the directory default striping
2884 Details : stripe_size -1 was used internally to signal directory stripe
2885 removal, now use "all default" to signal dir stripe removal
2886 as a directory striping of "all default" is not useful
2889 Frequency : common for large clusters running liblustre clients
2891 Description: doing an ls when liblustre clients are running is slow
2892 Details : sending a glimpse AST to a liblustre client waits for every AST
2893 to time out, as liblustre clients will not respond. Since they
2894 cannot cache data we refresh the OST lock LVB from disk instead.
2896 Severity : enhancement
2898 Description: doing an ls at the same time as file IO can be slow
2899 Details : enqueue and other "small" requests can be blocked behind many
2900 large IO requests. Create a new OST IO portal for non-IO
2901 requests so they can be processed faster.
2904 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2906 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2907 created file with mode 000
2908 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2909 allow an ftruncate() as a non-root user to a file with mode 000.
2910 Lustre now respects this flag to disable mode checks when
2911 truncating a file owned by the user
2914 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2917 Description: Revoking locks from clients that went dead or catatonic might take
2919 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2920 cancellation of such locks instant on servers without waiting for
2921 any reply from clients. Clients drops these locks when cancel
2922 notification from server is received without replying.
2925 Frequency : liblustre-only, when liblustre client dies or becomes busy
2927 Description: Doing ls on Linux clients can take a long time with active
2929 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2930 granting such locks to it in the first place if possible. Locks
2931 are taken by proxy on the OST during the read or write and
2932 dropped immediately afterward. Add connect flags handling, do
2933 not grant locks to liblustre clients for glimpse ASTs.
2935 Severity : enhancement
2937 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2938 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2939 sequential reads and behaves badly with applications doing random
2940 reads. Improve it by reading ahead at least read region, and
2941 avoiding excessive large RPC for small reads.
2943 Severity : enhancement
2945 Description: Creating more than 1000 files for a single job may cause a load
2946 imbalance on the OSTs if there are also a large number of OSTs.
2947 Details : qos_prep_create() uses an OST index reseed value that is an
2948 even multiple of the number of available OSTs so that if the
2949 reseed happens in the middle of the object allocation it will
2950 still utilize the OSTs as uniformly as possible.
2955 Description: OST or MDS may oops in ping_evictor_main()
2956 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2957 but doesn't restart at beginning of obd_exports_timed list
2960 Severity : enhancement
2962 Description: improve by-nid export eviction on the MDS and OST
2963 Details : allow multiple exports with the same NID to be evicted at one
2964 time without re-searching the exports list.
2967 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2969 Description: MDS may oops in groups_free()
2970 Details : in rare race conditions a newly allocated group_info struct is
2971 freed again, and this can be NULL. The 2.4 compatibility code
2972 for groups_free() checked for a NULL pointer, but 2.6 did not.
2975 Frequency : common for liblustre clients doing little filesystem IO
2976 Bugzilla : 9352, 7313
2977 Description: server may evict liblustre clients accessing contended locks
2978 Details : if a client is granted a lock or receives a completion AST
2979 with a blocking AST already set it would not reply to the AST
2980 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2983 Frequency : lfs setstripe, only systems with more than 160 OSTs
2985 Description: unable to set striping with a starting offset beyond OST 160
2986 Details : llapi_create_file() incorrectly limited the starting stripe
2987 index to the maximum single-file stripe count.
2990 Frequency : LDAP users only
2992 Description: lconf did not handle in-kernel recovery with LDAP properly
2993 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2995 Severity : enhancement
2997 Description: bind OST threads to NUMA nodes to improve performance
2998 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2999 node and do their allocations there to localize memory access
3001 Severity : enhancement
3003 Description: llmount can determine client NID directly from Myrinet (GM)
3004 Details : the client NID code from gmnalnid was moved directly into
3005 llmount, removing the need to use this or specifying the
3006 client NID explicitly when mounting GM clients with zeroconf
3009 Frequency : if client is started with down MDS
3011 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3012 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3013 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3014 ptlrpc_import_delayed_req() because we didn't check if the
3015 request was interrupted, and we also didn't break out of the
3016 event loop if there was no timeout
3021 Description: data loss during non-page-aligned writes to a single file from
3022 both multiple nodes and multiple threads on one node at same time
3023 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3024 inconsistency led to false short-reads, that were cached and later
3025 used by ->prepare_write() to fill in partially written page,
3026 leading to data loss.
3029 Frequency : always, if lconf --abort_recovery used
3031 Description: lconf --abort_recovery fails with 'Operation not supported'
3032 Details : lconf was attempting to abort recovery on the MDT device and not
3035 Severity : enhancement
3037 Description: remove cleanup logs
3038 Details : replace lconf-generated cleanup logs with lustre internal
3039 cleanup routines. Eliminates the need for client-cleanup and
3042 Severity : enhancement
3044 Description: add support for EAs (user and system) on lustre filesystems
3045 Details : it is now possible to store extended attributes in the Lustre
3046 client filesystem, and with the user_xattr mount option it
3047 is possible to allow users to store EAs on their files also
3049 Severity : enhancement
3051 Description: Add possibility (config option) to show minimal available OST free
3053 Details : When compiled with --enable-mindf configure option, statfs(2)
3054 (and so, df) will return least minimal free space available from
3055 all OSTs as amount of free space on FS, instead of summary of
3056 free spaces of all OSTs.
3058 Severity : enhancement
3060 Description: do not expand extent locks acquired on OST-side
3061 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3062 by server: they are not cached anyway.
3065 Frequency : when mmap is used/binaries executed from Lustre
3067 Description: Unmmap pages before throwing them away from read cache.
3068 Details : llap_shrink cache now attempts to unmap pages before discarding
3069 them (if unmapping failed - do not discard). SLES9 kernel has
3070 extra checks that trigger if this unmapping is not done first.
3075 Description: lconf didn't resolve symlinks before checking to see whether a
3076 given mountpoint was already in use
3079 Frequency : when migrating failover services
3080 Bugzilla : 6395, 9514
3081 Description: When migrating a subset of services from a node (e.g. failback
3082 from a failover service node) the remaining services would
3083 time out and evict clients.
3084 Details : lconf --force (implied by --failover) sets the global obd_timeout
3085 to 5 seconds in order to quickly disconnect, but this caused
3086 other RPCs to time out too quickly. Do not change the global
3087 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3089 Severity : enhancement
3090 Frequency : if MDS is started with down OST
3091 Bugzilla : 9439,5706
3092 Description: Allow startup/shutdown of an MDS without depending on the
3093 availability of the OSTs.
3094 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3095 Add appropriate locking and lov-osc refcounts for safe
3096 cleaning. Add osc abort_inflight calls in case the
3097 synchronize never started.
3100 Frequency : occasional (Cray XT3 only)
3102 Description: root not authorized to access files in CRAY_PORTALS environment
3103 Details : The client process capabilities were not honoured on the MDS in
3104 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3105 been accessed by an authorized user then root was able to access
3106 the file on the local client also. The root user capabilities
3107 are now allowed on the MDS, as this environment has secure UID.
3110 Frequency : occasional
3112 Description: ldiskfs "too long searching" message happens too often
3113 Details : A debugging message (otherwise harmless) prints too often on
3114 the OST console. This has been reduced to only happen when
3115 there are fragmentation problems on the filesystem.
3120 Description: Division by zero in statfs when all OSCs are inactive
3121 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3122 lov_statfs divides by value returned by lov_get_stripecnt().
3126 Bugzilla : 9489, 3273
3127 Description: First write from each client to each OST was only 4kB in size,
3128 to initialize client writeback cache, which caused sub-optimal
3129 RPCs and poor layout on disk for the first writen file.
3130 Details : Clients now request an initial cache grant at (re)connect time
3131 and so that they can start streaming writes to the cache right
3132 away and always do full-sized RPCs if there is enough data.
3133 If the OST is rebooted the client also re-establishes its grant
3134 so that client cached writes will be honoured under the grant.
3139 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3140 Details : Now I/O RPCs go to different portal number and (presumably) fast
3141 lock requests (and glimses) and other RPCs get their own service
3142 threads pool that should be able to service those RPCs
3145 Severity : enhancement
3147 Description: Ability to exchange lustre version between client and servers and
3148 issue warnings at client side if client is too old. Also for
3149 liblustre clients there is ability to refuse connection of too old
3151 Details : New 'version' field is added to connect data structure that is
3152 filled with version info. That info is later checked by server and
3156 Frequency : rare, liblustre only.
3157 Bugzilla : 9296, 9581
3158 Description: Two simultaneous writes from liblustre at offset within same page
3159 might proceed at the same time overwriting eachother with stale
3161 Details : I/O lock withing llu_file_prwv was released too early, before data
3162 actually was hitting the wire. Extended lock-holding time until
3163 server acknowledges receiving data.
3166 Frequency : extremely rare. Never observed in practice.
3168 Description: avoid generating lustre_handle cookie of 0.
3169 Details : class_handle_hash() generates handle cookies by incrementing
3170 global counter, and can hit 0 occasionaly (this is unlikely, but
3171 not impossible, because initial value of cookie counter is
3172 selected randonly). Value of 0 is used as a sentinel meaning
3173 "unassigned handle" --- avoid it. Also coalesce two critical
3174 sections in this function into one.
3176 Severity : enhancement
3178 Description: allow liblustre clients to delegate truncate locking to OST
3179 Details : To avoid overhead of locking, liblustre client instructs OST to
3180 take extent lock in ost_punch() on client's behalf. New connection
3181 flag is added to handle backward compatibility.
3183 Severity : enhancement
3184 Bugzilla : 4928, 7341, 9758
3185 Description: allow number of OST service threads to be specified
3186 Details : a module parameter allows the number of OST service threads
3187 to be specified via "options ost ost_num_threads={N}" in the
3188 OSS's /etc/modules.conf or /etc/modprobe.conf.
3192 Bugzilla : 6146, 9635, 9895
3193 Description: servers crash with bad pointer in target_handle_connect()
3194 Details : In rare cases when a client is reconnecting it was possible that
3195 the connection request was the last reference for that export.
3196 We would temporarily drop the export reference and get a new
3197 one, but this may have been the last reference and the export
3198 was just destroyed. Get new reference before dropping old one.
3200 Severity : enhancement
3201 Frequency : if client is started with failover MDS
3203 Description: Allow multiple MDS hostnames in the mount command
3204 Details : Try to read the configuration from all specified MDS
3205 hostnames during a client mount in case the "primary"
3208 Severity : enhancement
3210 Description: Stop sending data to evicted clients as soon as possible.
3211 Details : Check if the client we are about to send or are sending data to
3212 was evicted already. (Check is done every second of waiting,
3213 for which l_wait_event interface was extended to allow checking
3214 of exit condition at specified intervals).
3217 Frequency : rare, normally only when NFS exporting is done from client
3219 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3221 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3224 Frequency : always, if srand() is called before liblustre initialization
3226 Description: Liblustre uses system PRNG disturbing its usage by user application
3227 Details : Introduce internal to lustre fast and high-quality PRNG for
3228 lustre usage and make liblustre and some other places in generic
3229 lustre code to use it.
3231 Severity : enhancement
3232 Bugzilla : 9477, 9557, 9870
3233 Description: Verify that the MDS configuration logs are updated when xml is
3234 Details : Check if the .xml configuration logs are newer than the config
3235 logs stored on the MDS and report an error if this is the case.
3236 Request --write-conf, or allow starting with --old_conf.
3238 Severity : enhancement
3240 Description: Handle symlinks in the path when checking if Lustre is mounted.
3241 Details : Resolve intermediate symlinks when checking if a client has
3242 mounted a filesystem to avoid duplicate client mounts.
3247 Description: lconf can hit an error exception but still return success.
3248 Details : The lconf command catches the Command error exception at the top
3249 level script context and will exit with the associated exit
3250 status, but doesn't ensure that this exit status is non-zero.
3255 Description: failure of ptlrpc thread startup can cause oops
3256 Details : Starting a ptlrpc service thread can fail if there are a large
3257 number of threads or the server memory is very fragmented.
3258 Handle this without oopsing.
3261 Frequency : always, only if liblustre and non-default acceptor port was used
3263 Description: liblustre cannot connect to servers with non-default acceptor port
3264 Details : tcpnal_set_default_params() was not called and was therefore
3265 ignoring the environment varaible TCPNAL_PORT, as well as other
3266 TCPNAL_ environment variables
3271 Description: two objects could be created on the same OST for a single file
3272 Details : If an OST is down, in some cases it was possible to create two
3273 objects on a single OST for a single file. No problems other
3274 than potential performance impact and spurious error messages.
3278 Bugzilla : 5681, 9562
3279 Description: Client may oops in ll_unhash_aliases
3280 Details : Client dcache may become inconsistent in race condition.
3281 In some cases "getcwd" can fail if the current directory is
3287 Description: Inode refcounting problems in NFS export code
3288 Details : link_raw functions used to call d_instantiate without obtaining
3289 extra inode reference first.
3293 Bugzilla : 9942, 9903
3294 Description: Referencing freed requests leading to crash, memleaks with NFS.
3295 Details : We used to require that call to ll_revalidate_it was always
3296 followed by ll_lookup_it. Also with revalidate_special() it is
3297 possible to call ll_revalidate_it() twice for the same dentry
3298 even if first occurence returned success. This fix changes semantic
3299 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3300 reference on a request referred from the intent.
3301 ll_intent_release() then releases such a request.
3304 Frequency : rare, normally benchmark loads only
3306 Description: unlinked inodes were kept in memory on the client
3307 Details : If a client is repeatedly creating and unlinking files it
3308 can accumulate a lot of stale inodes in the inode slab cache.
3309 If there is no other client load running this can cause the
3310 client node to run out of memory. Instead flush old inodes
3311 from client cache that have the same inode number as a new inode.
3314 Frequency : SLES9 2.6.5 kernel and long filenames only
3315 Bugzilla : 9969, 10379
3316 Description: utime reports stale NFS file handle
3317 Details : SLES9 uses out-of-dentry names in some cases, which confused
3318 the lustre dentry revalidation. Change it to always use the
3322 Frequency : rare, unless heavy write-truncate concurrency is continuous
3323 Bugzilla : 4180, 6984, 7171, 9963, 9331
3324 Description: OST becomes very slow and/or deadlocked during object unlink
3325 Details : filter_destroy() was holding onto the parent directory lock
3326 while truncating+unlinking objects. For very large objects this
3327 may block other threads for a long time and slow overall OST
3328 responsiveness. It may also be possible to get a lock ordering
3329 deadlock in this case, or run out of journal credits because of
3330 the combined truncate+unlink. Solution is to do object truncate
3331 first in one transaction without parent lock, and then do the
3332 final unlink in a new transaction with the parent lock. This
3333 reduces the lock hold time dramatically.
3336 Frequency : rare, 2.4 kernels only
3338 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3339 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3340 before unmounting their filesystem in order to avoid deadlock.
3341 The kernel_locked() function in 2.4 kernels only checks whether
3342 the kernel lock is held, not whether it is this process that is
3343 holding it as 2.6 kernels do.
3348 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3349 Details : The client ptlrpc code may be trying to reconnect to a down
3350 server before a previous connection attempt has timed out.
3351 Increase the reconnect interval to be longer than the connection
3352 timeout interval to avoid sending duplicate connections to
3356 Frequency : echo_client brw_test command
3358 Description: fix echo_client to work with OST preallocated code
3359 Details : OST preallocation code (5137) didn't take echo_client IO path
3360 into account: echo_client calls filter methods outside of any
3361 OST thread and, hence, there is no per-thread preallocated
3362 pages and buffers to use. Solution: hijack pga pages for IO. As
3363 a byproduct, this avoids unnecessary data copying.
3367 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3368 Description: Client can oops in mdc_commit_close() after open replay
3369 Details : It was possible for the MDS to return an open request with no
3370 transaction number in mds_finish_transno() if the client was
3371 evicted, but without actually returning an error. Clients
3372 would later try to replay that open and may trip an assertion
3373 Simplify the client close codepath, and always return an error
3374 from the MDS in case the open is not successful.
3377 Frequency : rare, 2.6 OSTs only
3379 Description: OST may deadlock under high load on fragmented files
3380 Details : If there was a heavy load and highly-fragmented OST filesystems
3381 it was possible to have all the OST threads deadlock waiting on
3382 allocation of biovecs, because the biovecs were not released
3383 until the entire RPC IO was completed. Instead, release biovecs
3384 as soon as they are complete to ensure forward IO progress.
3386 Severity : enhancement
3388 Description: Support for specifying external journal device at mount
3389 Details : If an OST or MDS device is formatted with an external journal
3390 device, this device major/minor is stored in the ext3 superblock
3391 and may not be valid for failover. Allow detecting and
3392 specifying the external journal at mount time.
3397 Description: Mounting an MDS with pending unlinked files may cause oops
3398 Details : target_finish_recovery() calls mds_postrecov() which returned
3399 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3400 considers this an error and immediately begins cleaning up the
3401 lov, just after starting the mds_lov process
3403 Severity : enhancement
3405 Description: Implement 'lfs df' to report actual free space on per-OST basis
3406 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3407 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3408 usage of objects; '-h' to report in human readable format.
3410 ------------------------------------------------------------------------------
3412 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3419 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3420 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3421 at startup during orphan unlinking. Instead, skip buddy update
3422 before setup, as it will be regenerated after recovery is complete.
3427 Description: If an OST is inactive, its locks might reference stale inodes.
3428 Details : lov_change_cbdata() must iterate over all namespaces, even if
3429 they are inactive to clear inode references from the lock.
3431 Severity : enhancement
3432 Frequency : occasional, if non-standard max_dirty_mb used
3434 Description: Client will block write RPCs if not enough grant
3435 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3436 then the client will block writes while waiting for another RPC
3437 to complete instead of consuming its dirty limit. With change
3438 we get improved performance when max_dirty_mb is small.
3440 Severity : enhancement
3441 Bugzilla : 3389, 6253
3442 Description: Add support for supplementary groups on the MDS.
3443 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3444 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3445 lookups for user supplementary groups into a cache.
3449 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3450 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3451 for liblustre O_EXCL is enforced by MDS.
3455 Description: readdir never returns NULL in liblustre.
3456 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3457 dentry in struct dirent.
3461 Frequency : liblustre only, depends on application IO pattern
3462 Description: liblustre clients evicted if not contacting servers
3463 Details : Don't put liblustre clients into the ping_evictor list, so
3464 they will not be evicted by the pinger ever.
3466 Severity : enhancement
3468 Description: Add ability to evict clients by NID from MDS.
3469 Details : By echoing "nid:$NID" string into
3470 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3471 $NID would be instantly evicted from this MDS and from all active
3472 OSTs connected to it.
3476 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3477 Details : lookup_it_finish() used to query file size from OSTs that was not
3482 Description: service threads change working directory to that of init
3483 Details : Starting lustre service threads may pin the working directory
3484 of the parent thread, making that filesystem busy. Threads
3485 now change to the working directory of init to avoid this.
3489 Frequency : during shutdown only
3490 Description: shutdown with a failed MDS or OST can cause unmount to hang
3491 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3496 Frequency : 2.6 only
3497 Description: chmod/chown may include an extra supplementary group
3498 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3499 supplementary group and if none is specified this is used.
3502 Bugzilla : 5479 (6816)
3504 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3505 Details : If lookup is in progress on a file that is unlinked we might try
3506 to revalidate the inode and fail in revalidate after lookup is
3507 complete and ll_file_open() enqueues the open again but
3508 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3511 Frequency : always, if lconf --abort_recovery used
3513 Description: lconf --abort_recovery fails with 'Operation not supported'
3514 Details : lconf was attempting to abort recovery on the MDT device and not
3517 ------------------------------------------------------------------------------
3519 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3524 Frequency : rare (only unsupported configurations with a node running as an
3526 Bugzilla : 6514, 5137
3527 Description: Mounting a Lustre file system on a node running as an OST could
3529 Details : OSTs now preallocates memory needed to write out data at
3530 startup, instead of when needed, to avoid having to
3531 allocate memory in possibly low memory situations.
3532 Specifically, if the file system is mounted on on OST,
3533 memory pressure could force it to try to write out data,
3534 which it needed to allocate memory to do. Due to the low
3535 memory, it would be unable to do so and the node would
3536 become unresponsive.
3538 Severity : enhancement
3540 Description: Addition of lconf --service command line option
3541 Details : lconf now accepts a '--service <arg>' option, which is
3542 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3544 Severity : enhancement
3546 Description: Failover mode is now the default for OSTs.
3547 Details : By default, OSTs will now run in failover mode. To return to
3548 the old behaviour, add '--failout' to the lmc line for OSTs.
3550 Severity : enhancement
3552 Description: Health checks are now provided for MDS and OSTs
3553 Details : Additional detailed health check information on MSD and OSTs
3554 is now provided through the procfs health_check value.
3557 Frequency : occasional, depends on IO load
3559 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3560 after numerous create/delete cycles
3561 Details : The ext3 inode allocation policy would not allocate new inodes
3562 very well on the OSTs because there are no new directories
3563 being created. Instead we look for groups with free space if
3564 the parent directories are nearly full.
3569 Description: Network or server problems during mount may cause partially
3570 mounted clients instead of returning an error.
3571 Details : The config llog parsing code may overwrite the error return
3572 code during mount error handling, returning success instead
3578 Description: MDS can fail to allocate large reply buffers
3579 Details : After long uptimes the MDS can fail to allocate large reply
3580 buffers (e.g. zconf client mount config records) due to memory
3581 fragmentation or consumption by the buffer cache. Preallocate
3582 some large reply buffers so that these replies can be sent even
3583 under memory pressure.
3587 Frequency : rare (liblustre)
3588 Description: fsx running with liblustre complained that using truncate() to
3589 extend the file doesn't work. This patch corrects that issue.
3590 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3591 ATTR_SIZE and lsm use in llu_setattr_raw.
3595 Frequency : rare, only 2.6 kernels
3596 Description: Unusual file access patterns on the MDS may result in inode
3597 data being lost in very rare circumstances.
3598 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3599 code in the 2.6 kernel caused the inode and/or EA data not to
3600 be read from disk, causing single-file corruption.
3604 Frequency : rare, only 2.6 filesystems using extents
3605 Description: Heavy concurrent write and delete load may cause data corruption.
3606 Details : It was possible under high-load situations to have an extent
3607 metadata block in the block device cache from a just-unlinked
3608 file overwrite a newly-allocated data block. We now unmap any
3609 metadata buffers that alias just-allocated data blocks.
3613 Frequency : filesystems with default stripe_count larger than 77
3614 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3615 Details : lconf specifies an inode size of 4096 bytes when the default
3616 stripe_count is larger than 77. This conflicts with the default
3617 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3618 this case to avoid pinning too much memory for large EAs.
3620 ------------------------------------------------------------------------------
3622 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3627 Frequency : rare (extremely heavy IO load with hundreds of clients)
3629 Description: Client is evicted, gets IO error writing to file
3630 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3631 caused clients to be evicted for AST timeouts. The fixes in
3632 bug 5192 mean we no longer need to have such short AST timeouts
3633 so ldlm_timeout has been increased.
3636 Frequency : occasional during --force or --failover shutdown under load
3637 Bugzilla : 5949, 4834
3638 Description: Server oops/LBUG if stopped with --force or --failover under load
3639 Details : a collection of import/export refcount and cleanup ordering
3640 issues fixed for safer force cleanup
3643 Frequency : only filesystems larger than 120 OSTs
3644 Bugzilla : 5990, 6223
3645 Description: lfs getstripe would oops on a very large filesystem
3646 Details : lov_getconfig used kfree on vmalloc'd memory
3649 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3650 Bugzilla : 6242, 6243
3651 Description: reading from files that had been truncated to a non-zero size
3652 but never opened returned no data
3653 Details : ll_file_read() reads zeros from no-object files to EOF
3658 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3659 Details : There's little harm in aborting MDS/OSS recovery and letting it
3660 try again, so I removed the LASSERT and return an error instead.
3662 Severity : enhancement
3664 Description: New debugging infrastructure for tracking down data corruption
3665 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3666 (b) cover more of the client-side code path, and (c) try to narrow
3667 down where problems occurred
3671 Bugzilla : 3819, 4364, 4397, 6313
3672 Description: Racing close and eviction MDS could cause assertion in mds_close
3673 Details : It was possible to get multiple mfd references during close and
3674 client eviction, leading to one thread referencing a freed mfd.
3676 Severity: : enhancement
3677 Bugzilla : 3262, 6359
3678 Description: Attempts to reconnect to servers are now more aggressive.
3679 Details : This builds on the enhanced upcall-less recovery that was added
3680 in 1.4.2. When trying to reconnect to servers, clients will
3681 now try each server in the failover group every 10 seconds. By
3682 default, clients would previously try one server every 25 seconds.
3687 Description: After recovery, certain operations trigger a failed
3688 assertion on a client.
3689 Details : Failing over an mds, using lconf -d --failover, while a
3690 client was doing a readdir() call would cause the client to
3691 LBUG after recovery completed and the readdir() was resent.
3693 Severity : enhancement
3695 Description: Default groups are now added by lconf
3696 Details : You can now run lconf --group <servicename> without having to
3697 manually add groups with lmc.
3700 Frequency : occasional
3702 Description: Nodes with an elan id of 0 trigger a failed assertion
3705 Frequency : always when accessing e.g. tty/console device nodes
3707 Description: tty and some other devices nodes cannot be used on lustre
3708 Details : file's private_data field is used by device data and lustre
3709 values in there got lost. New field was added to struct file to
3710 store fs-specific private data.
3713 Frequency : when exporting Lustre via NFS
3715 Description: NFSD failed occasionally when looking up a path component
3716 Details : NFSD is looking up ".." which was broken in ext3 directories
3717 that had grown large enough to become hashed.
3720 Frequency : Clusters with multiple interfaces not on the same subnet
3722 Description: Nodes will repeatedly try to reconnect to an interface which it
3723 cannot reach and report an error to the log.
3724 Details : Extra peer list entries will be created by lconf with some peers
3725 unreachable. lconf now validates the peer before adding it.
3728 Frequency : Only if a default stripe is set on the filesystem root.
3730 Description: Setting a default stripe on the filesystem root prevented the
3731 filesystem from being remounted.
3732 Details : The client was sending extra request flags in the root getattr
3733 request and did not allocate a reply buffer for the dir EA.
3736 Frequency : occasional, higher if lots of files are accessed by one client
3737 Bugzilla : 6159, 6097
3738 Description: Client trips assertion regarding lsm mismatch/magic
3739 Details : While revalidating inodes the VFS looks up inodes with ifind()
3740 and in rare cases can find an inode that is being freed.
3741 The ll_test_inode() code will free the lsm during ifind()
3742 when it finds an existing inode and then the VFS later attaches
3743 this free lsm to a new inode.
3747 Bugzilla : 6422, 7030
3748 Description: MDS deadlock between mkdir and client eviction
3749 Details : Creating a new file via mkdir or mknod (starting a transaction
3750 and getting the ns lock) can deadlock with client eviction
3751 (gets ns lock and trying to finish a synchronous transaction).
3754 Frequency : occasional
3755 Description: While starting a server, the fsfilt_ext3 module could not be
3757 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3758 kernels. Previously, lconf would still use the ext3 name
3759 when trying to load modules. Now, it will correctly use
3760 ext3 on 2.4 and ldiskfs on 2.6.
3762 Severity : enhancement
3763 Description: The default stripe count has been changed to 1
3764 Details : The interpretation of the default stripe count (0, to lfs
3765 or lmc) has been changed to mean striping across a single
3766 OST, rather than all available. For general usage we have
3767 found a stripe count of 1 or 2 works best.
3769 Severity : enhancement
3770 Description: Add support for compiling against Cray portals.
3771 Details : Conditional compiling for some areas that are different
3775 Frequency : occasional
3776 Bugzilla : 6409, 6834
3777 Description: Creating files with an explicit stripe count may lead to
3778 a failed assertion on the MDS
3779 Details : If some OSTs are full or unavailable, creating files may
3780 trigger a failed assertion on the MDS. Now, Lustre will
3781 try to use other servers or return an error to the
3785 Frequency : occasional
3787 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3788 caused lock timeout during readahead (since 1.4.2).
3789 Details : Processes doing readahead might match a lock that hasn't been
3790 granted yet if there are overlapping and conflicting lock
3791 requests. The readahead process waits on ungranted lock
3792 (original lock is CBPENDING), while OST waits for that process
3793 to cancel CBPENDING read lock and eventually evicts client.
3795 Severity : enhancement
3797 Description: Initial enabling of flock support for clients
3798 Details : Implements fcntl advisory locking and file status functions.
3799 This feature is provided as an optional mount flag (default
3800 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3801 locking are implemented yet, and those that are are not guaranteed
3802 to be completely correct in production environments.
3803 mount -t lustre -o [flock|noflock] ...
3806 Frequency : occasional
3808 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3809 assertion in the ext3 JBD (journaling) layer.
3810 Details : The b_committed_data struct is protected by the big kernel lock
3811 in 2.4 kernels, serializing journal_commit_transaction() and
3812 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3813 access to this struct. In 2.6 kernels there is finer grained
3814 locking to improve SMP performance of the JBD layer.
3818 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3823 Description: merge and cleanup kernel patches.
3826 -----------------------------------------------------------------------------
3828 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3830 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3831 versions of Lustre. Please update all servers and clients to
3832 version 1.4.2 or later at the same time. You must also run
3833 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3834 to update the configuration logs.
3836 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3837 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3838 - swab llog records of type '0' so we get proper header size/idx (5861)
3839 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3840 - fix rename of one directory over another leaking an inode (5953)
3841 - avoid SetPageDirty on 2.6 (5981)
3842 - don't re-add just-being-destroyed locks to the waiting list (5653)
3843 - when creating new directories, inherit the parent's custom
3844 striping settings if present parent (3048)
3845 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3846 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3847 - don't hold client locks on temporary worklist from l_lru (5666)
3848 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3849 - automatically evict dead clients (5921)
3850 - Update file size properly in create+truncate+fstat case (6196)
3851 - Do not unhash mountpoint dentries, do not allow removal of
3853 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3854 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3855 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3856 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3857 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3858 - Correctly handle reads of files with no objects (6243)
3859 - lctl recover will also mark a device active if deactivate used (5933)
3861 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3862 - allow --write-conf on an MDS with different nettype than client (5619)
3863 - don't write config llogs to MDS for mounts not from that MDS (5617)
3864 - lconf should create multiple TCP connections from a client (5201)
3865 - init scripts are now turned off by default; run chkconfig --on
3866 lustre and chkconfig --on lustrefs to use them
3867 - upcalls are no longer needed for clients to recover to failover
3869 - add --abort-recovery option to lconf to abort recovery on device
3871 - add support for an arbitrary number of OSTs (3026)
3872 - Quota support protocol changes.
3873 - forward compatibility changes to wire structs (6007)
3874 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3875 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3876 - improved "lustre status" script
3877 - initialize blocksize for non-regular files (6062)
3878 - added --disable-server and --disable-client configure options (5782)
3879 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3880 - Vanilla 2.4.29 support
3881 - increase maximum number of obd devices to 520 (6242)
3882 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3883 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3884 SLES 9 2.6.5 kernels
3885 - limit stripes per file to 160 (the maximum EA size) (6093)
3887 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3890 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3891 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3892 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3893 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3894 - small ext3 extents cleanups and fixes (5733)
3895 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3896 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3897 - avoid needless client->OST connect, fix handle mismatch (5317)
3898 - fix DLM error path that led to out-of-sync client, long delays (5779)
3899 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3900 - fix several locking issues related to i_size (5492,5624,5654,5672)
3901 - don't move pending lock onto export if it is already evicted (5683)
3902 - fix kernel oops when creating .foo in unlinked directory (5548)
3903 - fix deadlock in obdfilter statistics vs. object create (5811)
3904 - use time_{before,after} to avoid timer jiffies wrap (5882)
3905 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3906 - Do not leak request if server was not able to process it (5154)
3907 - If mds_open unable to find parent dir, make that negative lookup(5154)
3908 - don't create new directories with extent-mapping (5909, 5936)
3910 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3911 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3912 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3913 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3914 - support for RHEL4 kernel on i686 (5773)
3915 - provide error messages when incompatible logs are encountered (5898)
3917 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3918 * version 1.4.0.10 (1.4.1 release candidate 1)
3920 - don't keep a lock reference when lock is not granted (4238)
3921 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3922 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3923 - drop import inflight refcount on signal_completed_replay error (5255)
3924 - unlock page after async write error during send (3677)
3925 - handle missing objects in filter_preprw_read properly (5265)
3926 - no transno return for symlink open, don't save no-trasno open (3440)
3927 - don't try to complete elan receive that already failed (4012)
3928 - free RPC server reply state on error (5406)
3929 - clean up thread from ptlrpc_start_thread() on error (5160)
3930 - readahead could read extra page into cache that wasn't ejected (5388)
3931 - prevent races in class_attach/setup/cleanup/detach (5260)
3932 - don't dereference de->d_inode after l_dput of de (5458)
3933 - use "int" for stripe value returned from lock_to_stripe (5544)
3934 - mballoc allocation and error-checking fixes in 2.6 (5504)
3935 - block device patches to fix I/O request sizes in 2.6 (5482)
3936 - look up hostnames for IB nals (5602)
3937 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3938 - don't start multiple acceptors for the same port (5277)
3939 - fix incorrect LASSERT in mds_getattr_name (5635)
3940 - export a proc file for general "ping" checking (5628)
3941 - fix "lfs check" to not block when the MDS is down (5628)
3943 - service request history (4965)
3944 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3945 - create an "evict_client" /proc entry on OSTs, like the MDS has
3946 - fix mount usage message, return errors per mount(8) (5168)
3947 - change grep [] to grep "[]" in tests so they work in more UMLs
3948 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3949 - remove ancient LOV_MAGIC_V0 EA support (5047)
3950 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3951 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3952 - allow readahead allocations to fail when low on memory (5383)
3953 - mmap locking landed again, after considerable improvement (2828)
3954 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3956 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3959 - send OST transaction number in read/write reply to free req (4966)
3960 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3961 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3962 - the watchdog thread now runs as interruptible (5246)
3963 - flock/lockf fixes (but it's still disabled, pending 5135)
3964 - don't use EXT3 constants in llite code (5094)
3965 - memory shortage at startup could cause assertion (5176)
3967 - reorganization of lov code
3968 - single portals codebase
3970 - add extents/mballoc support (5025)
3971 - direct I/O reads in the obdfilter (4048)
3972 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3974 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3977 - send OST transaction number in read/write reply to free req (4966)
3978 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3979 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3980 - the watchdog thread now runs as interruptible (5246)
3981 - handle missing objects in filter_preprw_read properly (5265)
3982 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3983 - drop import inflight refcount on signal_completed_replay error (5255)
3984 - unlock page after async write error during send (3677)
3985 - return original error code on reconstructed replies (3761)
3986 - no transno return for symlink open, don't save no-trasno open (3440)
3988 - add pid to ldlm debugging output (4922)
3989 - bump the watchdog timeouts -- we can't handle 30sec yet
3990 - extra debugging for orphan dentry/inode bug (5259)
3992 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3995 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3996 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3997 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3998 - fix NULL dereference in /proc/sys/portals/routes (4827)
3999 - allow failed mdc_close() operations to be interrupted (4561)
4000 - stop precreate on OST before MDS would time out on it (4778)
4001 - don't send partial-page writes before EOF from client (4410)
4002 - discard client grant for sub-page writes on large-page clients (4520)
4003 - don't free dentries not owned by NFS code, check generation (4806)
4004 - fix lsm leak if mds_create_objects() fails (4801)
4005 - limit debug_daemon file size, always print CERROR messages (4789)
4006 - use transno after validating reply (3892)
4007 - process timed out requests if import state changes (3754)
4008 - update mtime on OST during writes, return in glimpse (4829)
4009 - add mkfsoptions to LDAP (4679)
4010 - use ->max_readahead method instead of zapping global ra (5039)
4011 - don't interrupt __l_wait_event() during strace
4013 - add software watchdogs to catch hung threads quickly (4941)
4014 - make lustrefs init script start after nfs is mounted
4015 - fix CWARN/ERROR duplication (4930)
4016 - return async write errors to application if possible (2248)
4017 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4018 - print NAL number in %x format (4645)
4019 - update barely-supported suse-2.4.21-171 series (4842)
4020 - support for sles 9 %post scripts
4021 - support for building 2.6 kernel-source packages
4022 - support for sles km_* packages
4024 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4027 - ignore -ENOENT errors in osc_destroy (3639)
4028 - notify osc create thread that OSC is being cleaned up (4600)
4029 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4030 - reconstruct ost_handle() like mds_handle() (4657)
4031 - create a new thread to do import eviction to avoid deadlock (3969)
4032 - let lconf resolve symlinked-to devices (4629)
4033 - don't unlink "objects" from directory with default EA (4554)
4034 - hold socknal file ref over connect in case target is down (4394)
4035 - allow more than 32000 subdirectories in a single directory (3244)
4036 - fix blocks count for O_DIRECT writes (3751)
4037 - OST returns ENOSPC from object create when no space left (4539)
4038 - don't send truncate RPC if file size isn't changing (4410)
4039 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4040 - bind to privileged port in socknal and tcpnal (3689)
4042 - rate limit CERROR/CWARN console message to avoid overload (4519)
4043 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4044 - basic mmap support (3918)
4045 - kernel patch series update from b1_4 (4711)
4047 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4050 - avoid crash during MDS cleanup with OST shut down (2775)
4051 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4052 - don't use bad inodes on the MDS (3744)
4053 - dynamic object preallocation to improve recovery speed (4236)
4054 - don't hold spinlock over lock dumping or change debug flags (4401)
4055 - don't zero obd_dev when it is force cleaned (3651)
4056 - print grants to console if they go negative (4431)
4057 - "lctl deactivate" will stop automatic recovery attempts (3406)
4058 - look for existing locks in ldlm_handle_enqueue() (3764)
4059 - don't resolve lock handle twice in recovery avoiding race (4401)
4060 - revalidate should check working dir is a directory (4134)
4062 - don't always mark "slow" obdfilter messages as errors (4418)
4064 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4067 - don't close LustreDB during write_conf until it is done (3860)
4068 - fix typo in lconf for_each_profile (3821)
4069 - allow dumping logs from multiple threads at one time (3820)
4070 - don't allow multiple threads in OSC recovery (3812)
4071 - fix debug_size parameters (3864)
4072 - fix mds_postrecov to initialize import for llog ctxt (3121)
4073 - replace config semaphore with spinlock (3306)
4074 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4075 - don't allow enqueue to complete on a destroyed export (3822)
4076 - down write_lock before checking llog header bitmap (3825)
4077 - recover from lock replay timeout (3764)
4078 - up llog sem before sending rpc (3652)
4079 - reduce ns lock hold times when setting kms (3267)
4080 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4081 - fix NULL deref and obd_dev leak on setup error (3312)
4082 - replace some LBUG about llog ops with error handling (3841)
4083 - don't match INVALID dentries from d_lookup and spin (3784)
4084 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4085 - fix invalid assertion in ptlrpc_set_wait (3880)
4087 - add libwrap support for the TCP acceptor (3996)
4088 - add /proc/sys/portals/routes for non-root route listing (3994)
4089 - allow setting MDS UUID in .xml (2580)
4090 - print the stack of a process that LBUGs (4228)
4092 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4095 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4096 - make sure to unset replay flag from failed open requests (3440)
4097 - if default stripe count is 0, use OST count for inode size (3636)
4098 - update parent mtime/ctime on client for create/unlink (2611)
4099 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4100 - free recovery state on server during a forced cleanup (3571)
4101 - unregister_reply for resent reqs (3063)
4102 - loop back devices mounting and status check on 2.6 (3563)
4103 - fix resource-creation race that can provoke i_size == 0 (3513)
4104 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4105 - more debugging for page-accounting assertion (3746)
4106 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4107 - avoid deadlock after precreation failure (3758)
4108 - fix race and lock order deadlock in orphan handling (3450, 3750)
4109 - add validity checks when grabbing inodes from l_ast_data (3599)
4111 - add /proc/.../recovery_status to obdfilter (3428)
4112 - lightweight CDEBUG infrastructure, debug daemon (3668)
4113 - change default OSC RPC parameters to be better on small clusters
4114 - turn off OST read cache for files smaller than 32MB
4115 - install man pages and include them in rpms (3100)
4116 - add new init script for (un)mounting lustre filesystems (2593)
4117 - run chkconfig in %post for init scripts (3701)
4118 - drop scimac NAL (unmaintained)
4120 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4123 - clean kiobufs before and after use (3485)
4124 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4125 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4126 - change init script to fail more clearly if not run as root (1528)
4127 - allow clients to reconnect during replay (1742)
4128 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4129 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4130 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4131 - fix cond_resched (3554)
4132 - only evict unfinished clients after recovery (3515)
4133 - allow bulk resend, prevent data loss (3570)
4134 - dynamic ptlrpc request buffer allocation (2102)
4135 - don't allow unlinking open directory if it isn't empty (2904)
4136 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4137 - remove extraneous obd dereference causing LASSERT failure (3334)
4138 - don't use get_cycles() when creating temp. files on the mds (3156)
4139 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4140 - handle EEXIST for set-stripe, set proper directory name (3336)
4142 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4143 - fix ksocknal_fmb_callback() error messages (2918)
4145 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4148 - don't copy lvb into (possibly NULL) reply on error (2983)
4149 - don't deref dentry after dput, don't free lvb on error (2922)
4150 - use the kms to determine writeback rpc length (2947)
4151 - increment oti_logcookies when osc is inactive (2948)
4152 - update client's i_blocks count via lvb messages (2543)
4153 - handle intent open/close of special files properly (1557)
4154 - mount MDS with errors=remount-ro, like obdfilter (2009)
4155 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4156 - don't use cancelling-locks' kms values (2947)
4157 - use highest lock extent for kms, not last one (2925)
4158 - don't dereference ERR_PTR() dentry in error handling path (3107)
4159 - fix thread race in portals_debug_dumplog() (3122)
4160 - create lprocfs device entries at setup instead of at attach (1519)
4161 - common AST error handler, don't evict client on completion race (3145)
4162 - zero nameidata in detach_mnt in 2.6 (3118)
4163 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4164 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4165 - initialize RPC timeout timer earlier for 2.6 (3219)
4166 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4167 - print nal/nid for unknown nid (3258)
4168 - additional checks for oscc recovery before doing precreate (3284)
4169 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4170 - don't crash in mdc_close for bad permissions on open (3285)
4171 - zero i_rdev for non-device files (3147)
4172 - clear page->private before handing to FS, better assertion (3119)
4173 - tune the read pipeline (3236)
4174 - fix incorrect decref of invalidated dentry (2350)
4175 - provide read-ahead stats and refine rpc in flight stats (3328)
4176 - don't hold journal transaction open across create RPC (3313)
4177 - update atime on MDS at close time (3265)
4178 - close LDAP connection when recovering to avoid server load (3315)
4179 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4180 - don't leak open file on MDS after open resend (3325)
4181 - serialize filter_precreate and filter_destroy_precreated (3329)
4182 - loop device shouldn't call sync_dev() for nul device (3092)
4183 - clear page cache after eviction (2766)
4184 - resynchronize MDS->OST in background (2824)
4185 - refuse to mount the same filesystem twice on same mountpoint (3394)
4186 - allow llmount to create routes for mounting behind routers (3320)
4187 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4188 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4189 - fix rare problem with rename on htree directories (3417)
4191 - allow default OST striping configuration per directory (1414)
4192 - fix compilation for qswnal for 2.6 kernels (3125)
4193 - increase maximum number of MDS request buffers for large systems
4194 - change liblustreapi to be useful for external progs like lfsck (3098)
4195 - increase local configuration timeout for slow disks (3353)
4196 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4198 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4201 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4202 - don't overwrite extent policy data in reply if lock was blocked (2901)
4203 - drop filter export grants atomically with removal from device (2663)
4204 - del obd_self_export from work_list in class_disconnect_exports (2908)
4205 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4206 - swab reply message in mdc_close, other PPC fixes (2464)
4207 - fix destroying of named logs (2325)
4208 - overwrite old logs when running lconf --write_conf (2264)
4209 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4210 - fix race in target_handle_connect (2898)
4211 - mds_reint_create() should take same inode create lock (2926)
4212 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4213 - don't close files for self_export to avoid uninitialized obd (2936)
4214 - allow MDS with the same name as client node (2939)
4215 - hold dentry reference for closed log files for unlink (2325)
4216 - reserve space for all logs during transactions (2059)
4217 - don't evict page beyond end of stripe extent (2925)
4218 - don't oops on a deleted current working directory (2399)
4219 - handle hard links to targets without a parent properly (2517)
4220 - don't dereference NULL lock when racing during eviction (2867)
4221 - don't grow lock extents when lots of conflicting locks (2919)
4223 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4226 - account for cache space usage on clients to avoid data loss (974)
4227 - lfsck support in lustre kernel code (2349)
4228 - reduce journal credits needed for BRW writes (2370)
4229 - orphan handling to avoid losing space on client/server crashes
4230 - ptlrpcd can be blocked, stopping ALL progress (2477)
4231 - use lock value blocks to assist in proper KMS, faster stat (1021)
4232 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4233 - recovery for initial connections (2355)
4234 - fixes for mds_cleanup_orphans (1934)
4235 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4236 - block all file creations until orphan recovery completes (1901)
4237 - client remove rq_connection from request struct (2423)
4238 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4239 - recovery timer race (2670)
4240 - mdc_close recovey bug (2532)
4241 - ptlrpc cleanup bug (2710)
4242 - mds timeout on local locks (2588)
4243 - namespace lock held during RPCs (2431)
4244 - handle interrupted sync write properly (2503)
4245 - don't try to handle a message that hasn't been replied to (2699)
4246 - client assert failure during cleanup after abort recovery (2701)
4247 - leak mdc device after failed mount (2712)
4248 - ptlrpc_check_set allows timedout requests to complete (2714)
4249 - wait for inflight reqs when ptlrpcd finishes (2710)
4250 - make sure unregistered services are removed from the srv_list
4251 - reset bulk XID's when resending them (caught by 1138 test)
4252 - unregister_bulk after timeout
4253 - fix lconf error (2694)
4254 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4255 - readahead locks pages, leaves pending causing memory pressure (2673)
4256 - increase OST request buffers to 4096 on large machines (2729)
4257 - fix up permission of existing directories in simple_mkdir (2661)
4258 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4259 - don't assign transno to errored transactions (2742)
4260 - don't delete objects on OST if given a bogus objid from MDS (2751)
4261 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4262 - if rq_no_resend, then timeout request after recovery (2432)
4263 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4264 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4265 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4266 - MDS recovery shouldn't delete objects that it has given out (2730)
4267 - if enqueue arrives after completion, don't clobber LVB (2819)
4268 - don't unlock pages twice when trigger_group_io returns error (2814)
4269 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4270 - don't write pages to disk if there was an error (1450)
4271 - don't ping imports that have recovery disabled (2676)
4272 - take buffered bytes into account when balancing socknal conn (2817)
4273 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4274 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4275 - remove little-endian swabbing of llog records (1987)
4276 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4277 - retry reposting request buffers if they fail (1191)
4278 - grow extent at grant time to avoid granting a revoked lock (2809)
4279 - lock revoke doesn't evict page if covered by a second lock (2765)
4280 - disable VM readahead to avoid reading outside lock extents (2805)
4282 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4283 - updated kernel patches for hp-2.4.20 kernel (2681)
4285 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4288 - fix truncated write corruption (2366)
4289 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4291 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4292 - ensure len(uuid) < 37 in lmc (1171)
4293 - fix ia64 OOPS in llog_test (2255)
4294 - zero end of page at obdfilter for partial page writes (2648)
4295 - don't leave stale dentries around after renames (bug 2428)
4296 - fix timeouts when evicting a client with a single lock held (2642)
4297 - set deadline for the initial HELLO message to drain (2634)
4298 - print out dotted-quad IP addresses in the socknal (2302)
4300 - additional debugging for MDS client eviction problem (2443)
4301 - fix mkfsoptions support for osts (2603, 2604)
4303 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4306 - add series for the vanilla 2.6.0 kernel
4307 - add series for the vanilla 2.4.24 kernel
4308 - add series for a cray x86/64 UL kernel drop
4309 - fix xattr patches for the vanilla 2.4.19 series
4311 - generate true UUIDs in lmc (1171)
4312 - have portals stack dumping break in UML (2466)
4313 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4314 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4315 - wake callback waiting threads on client eviction (2460)
4316 - Add --ptldebug and --subsystem to lmc (1719)
4317 - update assertion to allow safe interrupt allocation
4318 - set rq_no_resend for cancel requests (2432)
4319 - recalculate ptlrpcd timeout after resend (2494)
4320 - call vfs_rmdir when removing pending directories (2368)
4321 - fix renaming a file to itself (2429)
4322 - lmc creates a default one-stripe lov (2454)
4323 - expand procfs space to handle large clusters (2326)
4324 - increase UML stack to avoid overflow
4325 - update lconf's list of debug and subsystem masks
4326 - fix lfs find --obd (2510)
4327 - /proc tunable for disabling filter read caching (2591)
4328 - stop rpm packages from altering slapd.conf (2301)
4329 - disable nagle in the socknal under 0conf (2578)
4330 - choose mds inode size based on stripe count (2572)
4331 - fix kernel-source rpm problems (2516)
4333 - add --disable-doc to avoid pdf generation (2421)
4334 - update documentation, tests, type-os, comments
4335 - avoid format warnings on ia64
4336 - remove the TOE NAL
4337 - tiny code cleanups by removing unused fields
4339 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4342 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4343 - remove the most blatant lies from BUILDING file (2371)
4344 - change default debug level to reasonable production setting
4345 - reduce client side cache size to reduce cache flush time
4346 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4347 - make TCP zerocopy and pinger support enabled by default (2476)
4348 - sync writes completed after process exits caused crashes (2319)
4349 - maintain correct mount count on the MDS (2356)
4350 - backout 1557, because 2316 wasn't really fixed
4351 - better file I/O statistics gathering in /proc
4352 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4353 - another kernel patch to fix zero-copy TCP function export
4354 - don't take duplicate lock when processing re-sent getattr (2420)
4355 - lctl uses obd_self_export instead of creating new conn (2353)
4356 - MDS/OST recovery case which requires object creation asserted (2425)
4357 - move lfs from /usr/sbin to /usr/bin in packages
4358 - fix race between mds_client_add and mds_client_free (2417)
4359 - use kmalloc instead of slabs in portals (2430)
4360 - don't create duplicate records when a failover MDS is present (2442)
4361 - remove unnecessary mount age check (2332)
4362 - don't remove directory inodes from locks prematurely (2451)
4363 - don't break if MDS service name is the same as hostname (2103)
4364 - fix races in client write RPC generation when cache full (2482)
4366 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4369 - remove now-unused request->rq_obd (278)
4370 - if an allocation fails, print out how much memory we've used (1933)
4371 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4372 - add the "configurable stack size" patch to most series files (1256)
4373 - ability to write large log records, for 100+ OST configs (2306)
4374 - fix NULL deref when filter_prep fails (2314)
4375 - fix operator precedence error in filter_sync
4376 - dynamic allocation of socknal TX descriptors (2315)
4377 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4378 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4379 - ensure that CWARN messages in Portals always get to the syslog
4380 - __init/__exit are not for prototype decls (ldlm_init/exit)
4381 - x86-64 compile warning fixes
4382 - fix gateway LMC keyword conflict (2318)
4383 - fix MDS lock inversions in getattr/reint paths (1844)
4384 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4385 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4386 - DLM race condition prevented some lock evictions (2328)
4387 - ENOMEM detection and retry on socknal sends (2230)
4388 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4389 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4390 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4391 - better error messages when a client is rejected during recovery (1505)
4392 - avoid cancelling locks which were never granted, after failure (2330)
4393 - fix i_sem/journal inversion in mds_client_add (2333)
4394 - fix truncate/getattr lock cycle deadlock (2334)
4395 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4396 - fix two rare exit paths which could leak an l_lock() ref (2321)
4397 - fencepost error in MDS/OST orphan recovery (2226)
4398 - make log record alignment 8 bytes (1988)
4399 - lstripe now fails when requested offset > ost_count (2237)
4400 - ensure that all kernel series have a complete list.h (1607)
4401 - fix crashes in special-file operations (2316)
4402 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4404 - allow configurable automake binary, for testing new versions
4405 - small update to the lfs documentation
4407 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4409 * fix negative export reference count in fsfilt_sync (2312)
4411 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4412 * release candidate 0.9.1
4414 - orphans are moved into the PENDING directory for possible recovery
4415 - replayed opens now open by fid for orphan/rename safety (1042)
4416 - last close of an orphan inode generates a transno (683)
4417 - chdir() and mount() now pin the directory entry (1020)
4418 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4419 - discard very old requests without processing them (1502)
4420 - remove some common, well-understood CERRORs (1505)
4421 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4422 - clear "grant" flags in OST replies until OST grant code lands (1644)
4423 - fix read performance by not clobbering i_blksize on client (1598)
4424 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4425 - make LRU size a /proc tunable, clears locks when reduced (707)
4426 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4427 - server locks take references on exports now (1558)
4428 - build fixes for 2.4.20-rh trees (1663)
4429 - return an error from lov_create if all OSCs are inactive (1751)
4430 - fix import levels when a reconnect happens without a timeout (1597)
4431 - exit early from mds_open if we get a lookup error (1749)
4432 - partial page read at EOF wouldn't wait for disk before sending (1642)
4433 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4434 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4435 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4436 - write/truncate lock inversion (1639)
4437 - Don't auto-load obdclass, portals modules during cleanup (1495)
4438 - fix timestamps from jumping to "now" (1763)
4439 - extra journal assertions (1648)
4440 - add an extra multiunlink test (1771)
4441 - fix read_record/write_record API (1776)
4442 - fix leak of offset_extent, possible incorrect i_size later (1772)
4443 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4444 - Debugging for the kqswnal_get_idle_tx problems (1820)
4445 - Allow recovery to be attempted multiple times (1536)
4446 - Write out MDS last_rcvd file after it is first created (1600)
4447 - Fix tx_descriptor leak in failed transmit situations (1827)
4448 - ext3 journaling fixes for assertion failure after IO error (1871)
4449 - class_export_put() on freed export after completion AST error (1896)
4450 - Fix revalidate looping in VFS (1322)
4451 - Don't access a freed export during MDS_REINT timeout (1521)
4452 - Add open-unlink recovery support on the MDS (1673,1764)
4453 - Return an error if no MDS data was read from last_rcvd (1946)
4454 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4455 - Don't setup a disk device that doesn't match exported UUID (317)
4456 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4457 - avoid committing NULL handle in force close
4458 - local.sh is now a one-stripe LOV configuration
4459 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4460 - fix ext3 htree duplicate directory entry corruption (1516)
4461 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4462 - update to new LOV EA format (2097)
4463 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4465 2003-06-15 Phil Schwan <phil@clusterfs.com>
4468 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4469 - per-import/export recovery handling (958, 931, 959)
4470 - multiple last-rcvd slots, for serving multiple FSes (949)
4471 - connections are again shared between multiple imp/exports (963, 964)
4472 - "umount -f" would hang if any requests needed to be sent (393, 978)
4473 - avoid pinning large req buffer by copying for queued messages (989)
4474 - add "uuid" to "lctl device" command to help upcalls (991)
4475 - "open" RPCs with transnos would confuse recovery counters (1037)
4476 - do proper endian conversion of all wire messages (288, 340, 891)
4477 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4478 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4479 - fix ptlrpc_connection leak in target_handle_connect (1174)
4480 - fix import refcounting bug in OST and MDS cleanup (1134)
4481 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4482 - fix very unlikely obd_types race condition (501)
4483 - remove osc_open hack for echo_client (1187)
4484 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4485 - a failure in read_inode2 leads to deadlock (1139)
4486 - cancel ack-locks as soon as transaction is committed (1072)
4487 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4488 - make sure to commitrw after any preprw to avoid deadlock (1162)
4489 - failing to execute a file in a lustre FS would lock inode (1203)
4490 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4491 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4492 - obd_uuid2tgt would walk off the end of the list (1255)
4493 - on IA64 the osc would give portals incorrect bulk size (1258)
4494 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4495 - fix lock inversion caused by new llite matching code (1282)
4496 - limit the number of dirty pages on a client to 10MB (1286)
4497 - timed out locks were not being corrected cancelled (1289)
4498 - fix O_DIRECT above 4GB on IA-32 (1292)
4499 * major user-visible changes
4500 - fail out/fail over policy now controlled by the upcall (993)
4502 - add OBD_PING to check server availability and failure (954)
4503 - lustre messages are now sent in sending host order (288, 340, 891)
4504 - add eadatalen to MDS getattr reply (340)
4505 - OST read replies may contain second buffer, with per-page status (593)
4507 2003-03-11 Phil Schwan <phil@clusterfs.com>
4510 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4511 - failed open()s could cause deadlock; fixed (867, 869)
4512 - stop cancelling OST locks when files are closed (481)
4513 - overlapping XID spaces caused network corruption (851, 853)
4514 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4515 - setattr_raw added, to do single-RPC, server-side setattrs
4516 - lmc/lconf syntax change for OST UUIDs
4517 - fix crashy race condition between ptlrpc_free_req and osc_close
4518 - don't use request in mdc_enqueue if we hit a timeout (889)
4519 - don't set the inode i_size for regular files from the MDS (896)
4520 - handle out of order completion AST (842)
4521 - don't LBUG if a lock request times out after receiving AST (913)
4522 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4523 - if a bad lock AST arrives, send an error instead of dropping entirely
4524 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4525 - fix leak in bulk IO when only partially completed (899, 900, 926)
4526 - fix O_DIRECT for ia64 (55)
4527 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4528 - C-z after timeout could hang a process forever; fixed (977)
4530 - client-side I/O cache (678, 924, 929, 941, 970)
4532 - READPAGE and SETATTRs which don't take server-side locks get
4535 2003-02-11 Phil Schwan <phil@clusterfs.com>
4538 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4539 - Fix fsfilt_extN_readpage() to read a full page of directory
4540 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4541 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4542 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4543 - Most intent operations were removed, in favour of a new RPC mode
4544 that does a single RPC to the server and bypasses most of the VFS
4545 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4546 - Aggressively cancel local locks on DLM servers
4547 - mds_reint_unlink sends EA to the client if it's the last nlink.
4548 client uses that EA to unlink OST objects.
4549 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4550 - recursive symlinks were fixed (439)
4551 - fixed NULL deref in DEBUG_REQ
4552 - filter_update_lastobjid no longer calls sync, which annoyed extN
4553 - fixed multi-client small-writes to a single file problem (445)
4554 - fixed mtime updates during file writes (607)
4555 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4556 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4557 - fixed Linux OST size reporting problem (444, 656)
4558 - OST now updates object mtime with writes or setattr (607, 619)
4559 - client verifies file size before zeroing page past EOF (445)
4560 - OST now writes last allocated objid to disk with allocation (108)
4561 - LOV on echo now works (409)
4563 - mds_reint_unlink sends a new buffer, with the EA included. this
4564 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4565 set if a regular file was being unlinked, and it was the last link
4566 - use PtlGet from the target for bulk writes (315)
4567 - OST now updates object mtime with writes or setattr (607, 619)
4568 - LDLM now has a grant-time callback to revalidate locked items, if
4570 - Many MDS operations were reorganized to combat race conditions
4572 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4573 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4575 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4578 - Fully reactivate OST imports after reconnection (512, others)
4579 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4580 - More graceful error handling for truncating on dead OST (515)
4581 - Don't error out unless we're actually accessing dead stripes (474)
4582 - Fix garbage sizes when stripes are missing (410)
4583 - LRU counters were broken, causing constant lock purge (433, 432)
4584 - garbage on read from stripes with failed OSTs (441)
4585 - mark OSCs as active before reconnecting during recovery (438)
4586 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4587 - lfind did not preserve OST order in output (443)
4588 - symlinks cause hung clients, incorrect data (439)
4589 - stop dereferencing request after dropping refcount (457)
4590 - don't LASSERT(spin_is_locked) on non-SMP (455)
4591 - fixes for many rename() bugs
4592 - fstat didn't correctly synchronize attributes (399)
4593 - server must handle lock cancellation during blocking AST prep (487)
4594 - bulk descriptors were free()d too soon (511)
4595 - fix paths in lconf, which would load incorrect modules (451, 507)
4596 - fix confusing lconf 'host not found' error message (386)
4597 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4598 - fix race condition in mdc_blocking_ast() for inode access (526)
4599 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4600 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4601 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4602 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4603 - fix lstripe in several ways (kernel side) (527)
4604 - fix request leak in ldlm_cli_enqueue (262)
4605 - incorrect OSC was marked inactive after OST failure
4606 - call mds_fs_cleanup before unmounting filesystem (524)
4607 - fix races between taking ns_lock and ldlm_lock_change_resource
4608 - fix races updating LOV export open file list
4609 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4610 - fix recovery NULL deref in ldlm_cli_cancel_unused
4611 - fix some DLM races by using new hash table for lock handles (419)
4612 - permit the client to specify desired inodes, at replay
4613 - duplicate requests when we queue them for replay reintegration
4614 - fix last_rcvd offset calculation
4615 - sync after each recovered transaction, so we always make progress
4616 - never, not always, ERESTART requests without transnos
4617 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4619 - skip replay if the MDS says that the client is already connected
4620 - don't check for a recovery-enabled export to match lctl's UUID
4621 - don't INC_USE_COUNT for phantom exports
4622 - don't crash when cleaning up phantom exports (567)
4623 - don't double-finish or set replay data for errored mdc_open requests
4624 - abort requests when they time out, so we don't get old replies
4625 - send/receive replies for AST messages again
4626 - if the client says that it doesn't have the lock, cancel it on the
4628 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4629 mark it as destroyed, it will all work out when decref is called
4630 - fix module use counts (22, 581)
4632 - ASTs now expect a reply (server cancels lock on error reply)
4634 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4637 - fix many simultaneous client startup (392)
4638 - fix dentry->d_it clobbering
4639 - credentials weren't being shipped for readdir/getattr operations
4640 - remove invalid assertions triggered during some concurrent MD
4642 - proper Lustre versions added (336, 389)
4643 - fix memory leak for create error case (398)
4644 - fix LOV locking bug that would get cli/srv out of sync
4645 - fix echo client over LOV (409)
4646 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4647 - fix double-O_EXCL intent crash (424)
4648 - avoid sending multiple lock CANCELs (352)
4650 - MDS can do multi-client recovery (modulo bugs in new code)
4652 - many updates, edits, cleanups
4654 2002-11-18 Phil Schwan <phil@clusterfs.com>
4657 - fix null d_it dereference (346)
4658 - fix full OST/dbench hang (333)
4659 - fix permission problem with file removal (286)
4660 - fix removal of OSCs from LOV when they fail
4661 - fix NULL deref during bulk timeout (214)
4662 - fix problems related to multiple filesystems on one MDS (241)
4663 - fixed serious subtle metadata locking bugs
4664 - free locks on clients when inodes are removed due to memory
4666 - fix inode pointer in lock data (285)
4667 - partial support for multiple MDS on a single host (241)
4668 - data locks weren't cancelled at clear_inode time (290, 311)
4669 - intent locks could lead to unbounded lock growth (205)
4670 - added a maximum lock count, an LRU list, and a flusher
4671 - fix multiple rename (365)
4672 - properly abstracted the echo client
4673 - OSC locked 1 byte too many; fixed
4674 - rewrote brw callback code:
4675 - fixed recovery bugs related to LOVs (306)
4676 - fixed too-many-pages-in-one-write crash (191)
4677 - fixed (again) crash in sync_io_timeout (214)
4678 - probably fixed callback-related race (385)
4680 - Add capability to MDS protocol
4681 - LDLM cancellations and callbacks on different portals
4683 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4686 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4687 - defer open object destruction to close time (601981, 138)
4688 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4689 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4690 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4691 - don't call obd_setattr in ll_file_release for destroyed objects
4692 * protocol change to lustre_msg: move |version| and add |flags|
4693 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4694 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4695 * added replay of create, unlink, link and rename operations during
4696 MDS failover; recovery should be much more robust now
4697 * remove failed OSCs from LOVs (only lov_create uses this so far)
4698 * the lustre-HOWTO was brought (more) up to date (582544)
4700 2002-10-23 Phil Schwan <phil@clusterfs.com>
4703 - in-use dentries weren't being reused properly (617851)
4704 - prevent multiple LDLM setup (599178)
4705 - fix LOV size calculations for truncate (617853)
4706 - fix client handling of MDS intent errors (POSIX)
4707 - fix permission bug in lovstripe.c test (624321)
4708 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4709 - truncate past end of file could corrupt data
4710 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4711 - a race in recovery could return ETIMEDOUT to apps (623947)
4712 - building outside the source directory was fixed
4713 * the lustre-HOWTO was brought (more) up to date (582544)
4714 * major progress was made on recovery functionality
4716 2002-10-10 Phil Schwan <phil@clusterfs.com>
4719 - recovery deadlock fix
4720 - rm -rf causes LBUG fix (617817)
4721 - file open by multiple tasks fix (618962)
4722 - directory permissions bugs (602707 and 620007)
4723 - journal_stop fixed with locking (611313)
4724 - O_APPEND failures resolved (618273, perhaps 614459)
4725 - lconf PATH fix (619770)
4726 - IA64 build fix (621450)
4727 - RPC buffer sizes scale with amount of memory
4729 2002-10-01 Phil Schwan <phil@clusterfs.com>
4732 - locks would be cancelled without throwing away data pages,
4733 resulting in inconsistent data (605627)
4734 - inode attributes were not always being refreshed (605627, 612449)
4735 - lconf now continues to cleanup after lctl reports an error
4736 - MDS now enforces user permissions (602707)
4737 - lprocfs cleanup fixed, but not yet enabled (614157)
4738 - fixed infinite server hang, should a client not respond to an AST
4739 - avoid going into recovery if user calls readlink() with a buffer
4740 that's too small (613941)
4741 - AST RPCs no longer require replies (614867) -- this may be changed
4742 - don't crash server if client sends an IOV that's too big (611336)
4743 - fixed lock conversion deadlock (611892)
4744 - fixed the following of symlinks (614622)
4745 * recovery: the server can remove locks from a client that dies, other
4746 clients can make progress
4747 * more extN patch fixes
4748 * compile-time configurable ptlrpc buffer allocations
4750 - collaborative read cache document
4751 - Lustre Lite Performance CDR document-in-progress
4753 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4756 - fix typo in patch-2.4.18
4758 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4761 - clear ptlrpc request each time in handle_incoming_request()
4762 - unlink of files now destroys the object on the OST
4764 2002-09-19 Peter Braam <braam@clusterfs.com>
4766 * add hard link support
4767 * change obdfile creation method
4768 * kernel patch changed
4770 2002-09-19 Peter Braam <braam@clusterfs.com>
4773 - stack overflow bug in extN fixed
4775 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4777 * documentation updates
4778 - add man pages for config tools
4779 - update tests/README to describe testing with new config tools
4780 - finish metadata API descriptions
4781 * bug fixes and cleanups
4782 - statfs workaround for 16TB limit
4783 - LOV stripe allocation improved, can stripe on subset of OSTs
4784 - LOV file size/IO offset was wrong for files > 4GB in size
4785 - object EA data was being dropped, caused files to be unreadable
4786 - memory overflow with non-LOV OST caused memory corruption
4787 - fixed regression tests to work with new config tools, obdfilter
4788 - fixed bug when directory size became larger than 1 block
4789 - fixed bug (for single client case) when PWD was deleted
4790 - invalidate local directory pages when doing intent-based ops
4791 - avoid LDLM oops when lock callback contained bad data
4793 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4795 * documentation updates
4796 * bug fixes and cleanups
4797 - configuration tools
4800 - 64-bit compile warnings
4801 - 64-bit internal statfs data
4803 * test_brw on persistent OST devices
4805 * lprocfs (disabled)
4807 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4809 * documentation updates
4810 * bug fixes and cleanups
4811 * configuration tools
4813 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4816 * many small fixes to 0_5_4
4817 * io/network handling
4818 * thinkos in MDS operations
4820 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4823 * crucial basic fixes to 0.5.3
4824 * IOR, Iozone work over Elan
4827 2002-08-07 Phil Schwan <phil@clusterfs.com>
4828 * version 0_5_3, our first alpha
4829 * we use the new Portals iovs
4830 * documentation updates
4831 * bug fixes and cleanups
4832 * small changes in the DLM wire protocol
4834 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4835 * version 0_5_1 with some initial stability,
4836 * locking on MD and file I/O.
4837 * documentation updates
4838 * several bug fixes since 0.5.0
4839 * small changes in wire protocol
4841 2002-07-18 Phil Schwan <phil@clusterfs.com>
4843 * delivered as Lustre Light Alpha
4844 * fixed a crash after handling invalid MDS requests
4845 * fixed directory pages for architectures with non-4k pages sizes
4847 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4848 * release version v0_4_4
4849 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4851 * Fixes a number of interruption problems with OST operations.
4852 * Update documentation for portals header changes
4853 * Move all wire protocol structs/defines to lustre_idl.h
4854 * Fixes symlink length bug.
4855 * Add tcpdump to repository.
4857 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4858 * release version v0_4_3
4859 * Fixes statfs for inodes on extN.
4860 * Fixes bug in runtests which would delete /etc/hosts.
4861 * Use 64-bit object IDs wherever possible (not into VFS though)
4862 Remove ost_get_info, which is unused by lustre, and out of date.
4864 2002-07-03 Peter Braam <braam@clusterfs.com>
4865 * release version v0_4_2 Fixes a lookup error (type not passed)
4866 * move forward to head of Portals
4867 * move forward to latest Lustre kernel
4869 2002-06-25 Peter Braam <braam@clusterfs.com>
4870 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob