1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.14.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.7-sun1
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 Frequency : rare, on shutdown ost
18 Description: don't hit live lock with umount ost.
19 Description: shrink_dcache_parent can be in long loop with destroy dentries,
20 use shrink_dcache_sb instead.
24 Description: don't panic with use echo client
25 Details : echo client pass NULL as client nid pointer and this produce null
30 Description: Stack overflow during MDS log replay
31 Details : ease stack pressure by using a thread dealing llog_process.
35 Description: fix build on ppc32
36 Details : compile code with -m64 flag produce wrong object file for ppc32.
40 Description: add message levels for liblustreapi
44 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
45 Details : Move the 'good_osts' check before the 'total_bavail' check. This
46 will result in an -EAGAIN and in the exit call path we call
47 alloc_rr() which will with increasing aggressiveness attempt to
48 aquire precreated objects on the minimum number of required OSCs.
52 Description: Use old size assignment to avoid deadlock
53 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
54 the scheduling while holding a spinlock. We do not need locking
55 for size in ll_update_inode() because size is only updated from
56 the MDS for directories or files without objects, so there is no
57 other place to do the update, and concurrent access to such inodes
58 are protected by the inode lock.
62 Description: resolve "_IOWR redefined" build error on SLES10
66 Description: dump the memory debugging after all modules are unloaded to
67 suppress false negative in conf_sanity test 39
69 Severity : enhancement
71 Description: build kernel-ib packages for OFED 1.3 in our release cycle
76 Description: fix SLES kernel versioning
77 Details : the kernel version for our SLES 10 kernel did not include a "-"
78 before the "smp" at the end. while this was not a problem in
79 general, it did mean that software trying to use the kernel
80 version to try to detect a vendor specific kernel would fail.
81 this was most evident by the OFED build scripts.
85 Description: Don't update lov_desc members until making sure they are valid
86 Details : When updating lov_desc members via proc fs, need fix their
87 validities before doing the real update.
91 Description: don't put request into delay list while invalidate in flight.
92 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
93 import in flight. this produce timeout for invalidate and sometimes
98 Frequency : on ppc only
99 Description: not convert ost objects for directory because it's not exist.
100 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
104 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
106 Severity : enhancement
108 Description: Update to support 2.6.22.14 vanilla kernel.
112 Frequency : rare, on recovery
113 Description: read procfs can produce deadlock in some situation
114 Details : Holding lprocfs lock which send rpc can produce block for destroy
115 obd objects and this also block reconnect with -EALREADY. This isn't
116 fix all lprocfs bugs - but make it rare.
118 Severity : enhancement
120 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
123 Frequency : frequent on X2 node
125 Description: mdc_set_open_replay_data LBUG
126 Details : Set replay data for requests that are eligible for replay.
130 Description: lustre_mgs: operation 101 on unconnected MGS
131 Details : When MGC is disconnected from MGS long enough, MGS will evict the
132 MGC, and late on MGC cannot successfully connect to MGS and a lot
133 of the error messages complaining that MGS is not connected.
136 Frequency : on start mds
138 Description: Implement get_info(last_id) in obdfilter.
141 Frequency : occasional
143 Description: Correctly check stale fid, not start epoch if ost not support SOM
144 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
145 overwrited with new generated fid, but mds can anwer with one of these
146 two fids and both is not stale. setattr incorectly start epoch and
147 assume will be called done_writeting, but without SOM done_writing
151 Frequency : rare, depends on device drivers and load
153 Description: MDS or OSS nodes crash due to stack overflow
154 Details : Code changes in 1.8.0 increased the stack usage of some functions.
155 In some cases, in conjunction with device drivers that use a lot
156 of stack the MDS (or possibly OSS) service threads could overflow
157 the stack. One change which was identified to consume additional
158 stack has been reworked to avoid the extra stack usage.
161 Frequency : occasional
163 Description: Do not fail import if osc_interpret_create gets -EAGAIN
164 Details : If osc_interpret_create got -EAGAIN it immediately exits and
165 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
166 oscc_has_objects and see OSC has no objests and call
167 oscc_internal_create to resend create request.
169 Severity : enhancement
171 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
173 Severity : enhancement
175 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
178 Frequency : very rare
180 Description: Fix replay if there is an un-replied request and open
181 Details : In some cases, older replay request will revert the
182 mcd->mcd_last_xid on MDS which is used to record the client's
185 Severity : enhancement
187 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
189 Severity : enhancement
191 Description: Add rhel5 support to HEAD.
193 Serverity : enhancement
195 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
200 Description: Don't allow skipping OSTs if index has been specified.
201 Details : Don't allow skipping OSTs if index has been specified, make locking
202 in internal create lots better.
206 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
207 Details : access to bitfield in structure is always rounded to long
208 and this produce problem with not atomic change any bit.
212 Description: Lustre make rpms failed.
213 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
214 builds Lustre rpms from tarball.
217 Frequency : rare on shutdown ost
219 Description: If llog cancel was not send before clean_exports phase, this can
220 produce deadlock in llog code.
221 Details : If llog thread has last reference to obd and call class_import_put
222 this produce deadlock because llog_cleanup_commit_master wait when
223 last llog_commit_thread exited, but this never success because was
224 called from llog_commit_thread.
228 Description: allow userland application know is lost one of stripes.
229 Details : fill lvb_blocks with error code on ost and return it to
230 application if error flag found.
234 Description: NULL lov_tgts causing MDS oops
235 Details : more safe checks for NULL lov_tgts for avoid oops.
237 Severity : enhancement
239 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
243 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
244 Details : If an OST is down the MDS will hang indefinitely in
245 obd_statfs_rqset() waiting for the statfs data. While for
246 MDS QOS usage of statfs, it should not stuck in waiting.
248 Severity : enhancement
250 Description: remote_acl support
251 Details : Support ACL-based permission check for remote user.
252 Support setfacl/getfacl for remote user with the utils
253 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
254 the system "{s,g}etfacl" utils.
256 Severity : enhancement
258 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
260 Severity : enhancement
262 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
266 Description: make mgs_setparam() handle fsname containing dash
267 Details : fsname containing a dash does not work with lctl conf_param
269 Severity : enhancement
271 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
273 Severity : enhancement
275 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
279 Frequency : rare, at shutdown
280 Description: access already free / zero obd_namespace.
281 Details : if client_disconnect_export was called without force flag set,
282 and exist connect request in flight, this can produce access to
283 NULL pointer (or already free pointer) when connect_interpret
284 store ocd flags in obd_namespace.
288 Frequency : only at startup
289 Description: not alloc memory with spinlock held.
290 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
291 if any spinlock held.
293 Severity : enhancement
295 Description: make lustre randomly fail allocating memory
296 Details : Make lustre randomly failed allocating memory for testing purpose.
298 Severity : enhancement
300 Description: lost problems with lov objid file
301 Details : Fixes some scability and access to not inited memory problems
302 in work with lov objdid file.
307 Description: lfs find does not continue on file error
308 Details : Continue other files processing when a file/dir is absent.
312 Description: Inconsistent usage of lustre_pack_reply()
313 Details : Standardize the usage of lustre_pack_reply() such that it
314 always generate a CERROR on failure.
317 Frequency : occasional
319 Description: MDS hang or stay in waiting lock
320 Details : If client receive lock with CBPENDING flag ldlm need send lock
321 cancel as separate rpc, to avoid situation when cancel request
322 can't processed due all i/o threads stay in wait lock.
326 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
330 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
334 Description: Update to SLES9 kernel-2.6.5-7.287.3.
338 Description: Update to RHEL4 latest kernel.
340 Severity : enhancement
342 Description: Build SLES10 patchless client fails
343 Details : The configure was broken by run ./configure with
344 --with-linux-obj=.... argument for patchless client. When the
345 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
346 can't search header adequately. Use absolute path such as
347 -I($LINUX)/include instead.
351 Description: interrupt oig_wait produce painc on resend.
352 Details : brw_redo_request can be used for resend requests from ptlrpcd and
353 private set, and this produce situation when rq_ptlrpcd_data not
354 copyed to new allocated request and triggered LBUG on assert
355 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
356 ptlrpcd set if request is changed and can be safety changed to use
361 Description: LASSERT_{REQ,REP}SWAB macros are buggy
362 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
363 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
364 these macros and replace them with their difinitions instead.
368 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
369 Details : Update which_patch & target file for SLES10 latest kernel.
372 Description: Remove client patches from SLES 10 kernel.
373 Details : This causes SLES 10 clients to behave as patchless clients
374 even on a Lustre-patched (server) kernel.
376 Severity : enhancement
378 Description: self-adjustable client's lru lists
379 Details : use adaptive algorithm for managing client cached locks lru
380 lists according to current server load, other client's work
381 pattern, memory activities, etc. Both, server and client
382 side namespaces provide number of proc tunables for controlling
385 Severity : enhancement
387 Description: light-weight GSS support
388 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
389 and reduce performance overhead.
391 Severity : enhancement
393 Description: Linux keyring support
394 Details : Support using service of Linux keyring for Lustre GSS internal
395 context refresh/cache mechanism.
399 Description: Fix errors in lfs documentation
400 Details : Fixes man pages
404 Description: don't use GFP_* in generic Lustre code.
405 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
409 Description: obdclass is limited by single OBD_ALLOC(idarray)
410 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
414 Description: warnings with build patchless client with vanila 2.6.19 and up
415 Details : change old ctl_table style and replace ctl_table/ctl_table_header
416 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
420 Frequency : Only for SLES
421 Description: sanity_quota fail test_1
422 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
423 file, which makes TSTID[2] inunique.
427 Frequency : Only for SLES9
428 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
429 Details : Update target/ChangeLog/which_patch .
433 Frequency : Only for test_10 in sanity-quota.sh
434 Description: a bug in quota test code
435 Details : checking a test flag in a wrong place
439 Frequency : Only for test_13 in sanity-quota.sh
440 Description: a bug in quota test script
441 Details : shouldn't check overfull allocated quota
445 Description: jbd statistics
446 Details : Port older jbd statistics patch for sles10
448 Severity : enhancement
450 Description: Add printing inode info into message about error in writepage.
455 Description: reply_lock_interpret crash due to race with it and lock cancel.
456 Details : Do not replay locks that are being cancelled. Do not reference
457 locks by their address during replay, just by their handle.
461 Frequency : When flocks are used.
462 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
463 Details : Flock locks might destroy just granted lock if it could be merged
464 with another existing flock, this is done in completion handler,
465 so teach ldlm_cli_enquque_fini that this is a valid case for
471 Description: Oops in read and write path when failing to allocate lock.
472 Details : Check if lock allocation failed and return error back.
476 Description: lstripe command fails for valid OST index
477 Details : The stripe offset is compared to lov->desc.ld_tgt_count
478 instead of lov->desc.ld_active_tgt_count.
482 Description: sanity.sh failed test 103
483 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
488 Description: df doesn't work properly if diskfs blocksize != 4K
489 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
493 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
494 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
498 Description: obd_health_check_timeout too short
499 Details : set obd_health_check_timeout as 1.5x of obd_timeout
503 Description: llapi_file_create() does not allow some changes
504 Details : add llapi_file_open() that allows specifying the mode and
505 open flags, and also returns an open file handle.
509 Description: merge and cleanup kernel patches.
510 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
514 Description: Add journal checksum support.(Kernel part)
515 Details : The journal checksum feature adds two new flags i.e
516 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
517 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
518 indicates that the commit block contains the checksum for
519 the blocks described by the descriptor blocks. Now commit
520 record can be sent to disk without waiting for descriptor
521 blocks to be written to disk. This behavior is controlled
522 using JBD2_FEATURE_ASYNC_COMMIT flag.
526 Description: OSS needs mutliple precreate threads
527 Details : Add ability to start more than one create thread per OSS.
531 Description: Sanity.sh test_65k failed: setstripe should have succeeded
532 Details : execute lfs setstripe on client
536 Description: mds_obd_create error creating tmp object
537 Details : When the user sets quota on root, llog will be affected and can't
538 create files and write files.
541 Frequency : Always on ia64 patchless client, and possibly others.
543 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
544 Details : This allows the patchless client to be loaded on architectures
549 Description: RedHat Update kernel for RHEL5
550 Details : Add the kernel config file for RHEL5.
554 Description: RedHat Update kernel for RHEL5
555 Details : Modify the kernel config file more closer RHEL5.
559 Description: Build failure against Centos5 (RHEL5)
560 Details : Define PAGE_SIZE when it isn't present.
564 Description: test_7 on sanity-quota.sh can't be used on separate machines
565 Details : add facet to handle it
569 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
570 Details : with NFS, the anon dentry's parent was set to itself in
571 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
572 corresponding dentry other than use rec->ur_name.
574 Severity : enhancement
576 Description: lfs setstripe enhancement
577 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
581 Description: enable data checksumming by default
582 Details : enable checksum by default, allow --disable-checksum
583 configure option and "-o nochecksum" mount option. Checksums
584 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
588 Description: lustre support for RHEL5
589 Details : Add support for RHEL5.
593 Description: Client eviction due to failover config
594 Details : after a connection loss, the lustre client should attempt to
595 reconnect to the last active server first before trying the
596 other potential connections.
600 Description: when mds and osts use different quota unit(32bit and 64bit),
601 quota will be released repeatly.
602 Details : void sending multiple quota reqs to mds, which will keep the
603 status between the reqs.
607 Description: osts not allocated evenly to files
608 Details : change the condition to increase offset_idx
611 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
613 Description: Data corruption for OSTs that are formatted larger than 2TB
615 Details : When generating the bio request for lustre file writes the
616 sector number would overflow a temporary variable before being
617 used for the IO. The data reads correctly from Lustre (which
618 will overflow in a similar manner) but other file data or
619 filesystem metadata may be corrupted in some cases.
623 Description: Tune the kernel for good SCSI performance.
624 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
625 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
630 Description: rewrite ext2-derived code in obdclass/uuid.c
631 Details : rewrite inherited code (uuid parsing code from ext2 utils)
632 from scratch preserving functionality.
636 Description: "lfs find -obd UUID" prints directories
637 Details : "lfs find -obd UUID" will return all directory names instead
638 of just file names. It is incorrect because the directories
639 do not reside on the OSTs.
643 Description: use i_size_read and i_size_write in 2.6 port
644 Details : replace inode->i_size access with i_size_read/write()
647 Frequency : when removing large files
649 Description: scheduling issue during removal of large Lustre files
650 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
651 It causes scheduling issues when removing large files (17TB in the
655 Frequency : only with liblustre clients on XT3
657 Description: evictions taking too long
658 Details : allow llrd to evict clients directly on OSTs
661 Frequency : only on ppc
663 Description: /proc/fs/lustre/devices broken on ppc
664 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
665 We should make sure the type of 'index' is loff_t and then cast
666 explicitly as needed below. Do not assign an explicitly cast
670 Frequency : only for rhel5
672 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
673 Details : Modify the target file & which_kernel.
675 Severity : enhancement
677 Description: omit set fsid for export NFS
678 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
679 as FS_REQUIRES_DEV for avoid problems with generate fsid.
683 Frequency : Always, for kernels after 2.6.16
684 Description: Fix warning idr_remove called for id=.. which is not allocated.
685 Details : Last kernels save old s_dev before kill super and not allow
686 to restore from callback - restore it before call kill_anon_super.
690 Description: buffer overruns could theoretically occur
691 Details : llapi_semantic_traverse() modifies the "path" argument by
692 appending values to the end of the origin string, and a
693 overrun may occur. Adding buffer overrun check in liblustreapi.
697 Description: Fix error on 'ls .' at the top of the Lustre mount.
698 Details : Don't revalidate dentry if it is a root dentry.
702 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
703 Details : Modify vm-tunables-rhel4.patch.
707 Description: Kernel config for 2.6.18-vanilla.
708 Details : Modify targets/2.6-vanilla.target.in.
709 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
710 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
711 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
712 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
716 Description: improve handling recoverable errors
717 Details : if request processig with error which can be recoverable on server
718 request should be resend, otherwise page released from cache and
723 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
724 Details : Modify target file & which_patch.
725 A flaw was found in the IA32 system call emulation provided
726 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
727 value could be stored in the %RAX register, which could trigger an
728 out-of-bounds system call table access. An untrusted local user
729 could exploit this flaw to run code in the kernel
730 (ie a root privilege escalation). (CVE-2007-4573).
734 Description: O_DIRECT bypasses client statistics.
735 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
736 read_ahead_stats were updated. copied the stats section from
737 osc_send_oap_rpc() into async_internal().
741 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
745 Description: change order of libsysio includes
746 Details : '#include sysio.h' should always come before '#include xtio.h'
750 Description: handle "serious error: objid * already exists" more gracefully
751 Details : If LAST_ID value on disk is smaller than the objects existing in
752 the O/0/d* directories, it indicates disk corruption and causes an
753 LBUG(). If the object is 0-length, then we should use the existing
754 object. This will help to avoid a full fsck in most cases.
756 Severity : enhancement
758 Description: adapt the lustre_config script to support the upgrade case
759 Details : Add "-u" option for lustre_config script to support upgrading 1.4
760 server targets to 1.6 in parallel.
764 Description: To avoid grant space > avaible space when the disk is almost
765 full. Without this patch you might see the error "grant XXXX >
766 available" or some LBUG about grant, when the disk is almost
768 Details : In filter_check_grant, for non_grant cache write, we should
769 check the left space by if (*left > ungranted + bytes), instead
770 of (*left > ungranted), because only we are sure the left space
771 is enough for another "bytes", then the ungrant space should be
772 increase. In client, we should update cl_avail_grant only there
773 is OBD_MD_FLGRANT in the reply.
777 Description: Update RHEL 4 kernel to fix local root privilege escalation.
778 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
779 described in CVE-2007-4573. This problem could allow untrusted
780 local users to gain root access.
783 Frequency : when using O_DIRECT and quotas
785 Description: Incorrect file ownership on O_DIRECT output files
786 Details : block usage reported by 'lfs quota' does not take into account
787 files that have been written with O_DIRECT.
792 Description: touch file failed when fs is not full
793 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
794 otherwise we can get ENOSP while fs is not full.
798 Description: parallel lock callbacks
799 Details : Instead of sending blocking and completion callbacks as separated
800 requests, adding them to a set and sending in parallel.
803 Frequency : only for Cray XT3
804 Bugzilla : 12829/13455
805 Description: Changing primary group doesn't change the group lustre assigns to
807 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
808 overridden with the primary group provided by the group upcall,
809 whereas the supplied fsgid can be trusted if it is in the list of
810 supplementary groups returned by the group upcall.
812 Severity : enhancement
814 Description: Allow masking D_WARNING, D_ERROR messages from console
815 Details : Console messages can now be disabled via lnet.printk.
819 Description: User code with malformed file open parameter crashes client node
820 Details : Before packing join_file req, all the related reference should be
821 checked carefully in case some malformed flags cause fake
822 join_file req on client.
826 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
827 Details : ldlm_completion_ast() assumes that a lock is granted when the req
828 mode is equal to the granted mode. However, it should also check
829 that LDLM_FL_CP_REQD is not set.
833 Description: Heavy nfs access might result in deadlocks
834 Details : After ELC code landed, it is now improper to enqueue any mds
835 locks under och_sem, because enqueue might want to decide to
836 cancel open locks for same inode we are holding och_sem for.
840 Description: Client eviction while running blogbench
841 Details : A lot of unlink operations with concurrent I/O can lead to a
842 deadlock causing evictions. To address the problem, the number of
843 oustanding OST_DESTROY requests is now throttled to
844 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
845 callbacks are processed in priority.
849 Description: enable ACLs on MDS by default
850 Details : ACLs must be enabled on MDS by default.
853 Frequency : PPC/PPC64 only
855 Description: conflicts between asm-ppc64/types.h and lustre_types.h
856 Details : fix duplicated definitions between asm-ppc64/types.h and
857 lustre_types.h on PPC.
860 Frequency : PPC/PPC64 only
862 Description: asm-ppc/segment.h does not exist
863 Details : fix compile issue on PPC.
867 Description: better handle error messages in extents code
870 Frequency : RHEL4 only
872 Description: mkfs is very slow on IA64/RHEL4
873 Details : A performance regression has been discovered in the MPT Fusion
874 driver between versions 3.02.73rh and 3.02.99.00rh. As a
875 consequence, we have downgraded the MPT Fusion driver in the RHEL4
876 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
878 Severity : enhancement
880 Description: SNMP support enhancement
881 Details : Adding total number of sampled request for an MDS node in snmp
884 Severity : enhancement
886 Description: Optimize ldlm waiting list processing for PR extent locks
887 Details : When processing waiting list for read extent lock and meeting read
888 lock that is same or wider to it that is not contended, skip
889 processing rest of the list and immediatelly return current
890 status of conflictness, since we are guaranteed there are no
891 conflicting locks in the rest of the list.
895 Description: Time out and refuse to reconnect
896 Details : When the failover node is the primary node, it is possible
897 to have two identical connections in imp_conn_list. We must
898 compare not conn's pointers but NIDs, otherwise we can defeat
899 connection throttling.
903 Description: port llog fixes from b1_6 into HEAD
904 Details : Port llog reference couting and some llog cleanups from b1_6
905 (bug 10800) into HEAD, for protect from panic and access to already
906 free llog structures.
910 Description: Detect stride IO mode in read-ahead
911 Details : When a client does stride read, read-ahead should detect that and
912 read-ahead pages according to the detected stride pattern.
916 Description: data checksumming impacts single node performance
917 Details : add support for several checksum algorithm. Currently, only CRC32
918 and Adler-32 are supported. The checksum type can be changed on
919 the fly via /proc/fs/lustre/osc/*/checksum_type.
923 Description: use adler32 for page checksums
924 Details : when available, use the Adler-32 algorithm instead of CRC32 for
929 Description: build for x2 fails
930 Details : fix compile issue on Cray systems.
934 Description: Properly match for duplicate locks
935 Details : Due to different lock order from skiplists code, we need to
936 traverse entire list for now
939 Frequency : only on PPC/SLES10
941 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
942 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
943 be defined. Add a hack in mkfs_lustre.c to work around this compile
948 Description: LASSERT on MDS when client holding flock lock dies
949 Details : ldlm pool logic depends on number of granted locks equal to
950 number of released locks which is not true for flock locks, so
951 just exclude such locks from consideration.
955 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
956 Details : Use fsfilt_sync() for both the whole filesystem sync and
957 individual file sync to eliminate dangerous inode locking
958 with I_LOCK that can lead to a deadlock.
962 Description: performance in 1.6.3
963 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
964 sure that 1MB requests are not fragmented by the block layer.
966 Severity : enhancement
968 Description: organize the server-side client stats on per-nid basis
969 Details : Change the structure of stats under obdfilter and mds to
977 The "uuid"s file would list the uuids of _active_ exports.
978 And the clear entry is to clear all stats and stale nids.
980 Severity : enhancement
982 Description: eliminate client locks in face of contention
983 Details : file contention detection and lockless i/o implementation
986 --------------------------------------------------------------------------------
988 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
990 * Support for kernels:
991 2.6.5-7.283 (SLES 9),
992 2.6.9-55.EL (RHEL 4),
993 2.6.16.46-0.14 (SLES 10),
994 2.6.18.8 vanilla (kernel.org)
995 * Client support for unpatched kernels:
996 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
997 2.6.16 - 2.6.22 vanilla (kernel.org)
998 * Due to recently discovered recovery problems, we do not recommend
999 using patchless RHEL 4 clients with this or any earlier release.
1000 * Recommended e2fsprogs version: 1.39.cfs8
1001 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1002 * Starting with this release, the ldiskfs backing filesystem required
1003 by Lustre is now in its own package, lustre-ldiskfs. This package
1004 should be installed. It is versioned separately from Lustre and
1005 may be released separately in future.
1009 Description: block reactivating mgc import until all deactivates complete
1010 Details : Fix race when failing back MDT/MGS to itself (testing)
1013 Frequency : at statup only
1015 Description: mds_lov_synchronize race leads to various problems
1016 Details : simultaneous MDT->OST connections at startup can cause the
1017 sync to abort, leaving the OSC in a bad state.
1019 Severity : enhancement
1021 Description: add optional extra BUILD_VERSION info
1022 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1023 to override the lustre version.
1026 Frequency : 2.6.18 servers only
1028 Description: ll_kern_mount() doesn't release the module reference
1029 Details : The ldiskfs module reference count never drops down to 0
1030 because ll_kern_mount() doesn't release the module reference.
1035 Description: server LBUG when using old ost_num_threads parameter
1036 Details : Accept the old ost_num_threads parameter but warn that it
1037 is deprecated, and fix an off-by-one error that caused an LBUG.
1042 Description: Transient SCSI error results in persistent IO issue
1043 Details : iobuf->dr_error is not reinitialized to 0 between two
1047 Frequency : sometimes when underlying device returns I/O errors
1049 Description: OSTs not going read-only during write failures
1050 Details : OSTs are not remounted read-only when the journal commit threads
1051 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1052 instead of the ext3 wrappers.
1056 Description: poor connect scaling with increasing client count
1057 Details : Don't run filter_grant_sanity_check for more than 100 exports
1058 to improve scaling for large numbers of clients.
1061 Frequency : SLES10 only
1063 Description: sanity-quota.sh quotacheck failed: rc = -22
1064 Details : Quotas cannot be enabled on SLES10.
1067 Frequency : liblustre clients only
1069 Description: getdirentries does not give error when run on compute nodes
1070 Details : getdirentries does not fail when the size specified as an argument
1071 is too small to contain at least one entry
1073 Severity : enhancement
1075 Description: Add LNET router traceability for debug purposes
1076 Details : If a checksum failure occurs with a router as part of the
1077 IO path, the NID of the last router that forwarded the bulk data
1078 is printed so it can be identified.
1083 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1084 Details : Due to a race condition, liblustre clients were occasionally
1085 evicted incorrectly.
1087 Severity : enhancement
1089 Description: lfs setstripe use optional parameters instead of postional
1092 Severity : enhancement
1094 Description: Nanosecond timestamp support for ldiskfs
1095 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1096 resolution timestamps. There is not yet support for this at
1097 the Lustre filesystem level.
1100 Frequency : during server recovery
1102 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1103 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1107 Frequency : nfs export on patchless client
1109 Description: connectathon hang when test nfs export over patchless client
1110 Details : Disconnected dentry cannot be found with lookup, so we do not need
1111 to unhash it or make it invalid
1114 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1115 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1116 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1121 Description: open req refcounting wrong on reconnect
1122 Details : If reconnect happened between getting open reply from server and
1123 call to mdc_set_replay_data in ll_file_open, we will schedule
1124 replay for unreferenced request that we are about to free.
1125 Subsequent close will crash in variety of ways.
1126 Check that request is still eligible for replay in
1127 mdc_set_replay_data().
1132 Description: disable writes to filesystem when reading health_check file
1133 Details : the default for reading the health_check proc file has changed
1134 to NOT do a journal transaction and write to disk, because this
1135 can cause reads of the /proc file to hang and block HA state
1136 checking on a healthy but otherwise heavily loaded system. It
1137 is possible to return to the previous behaviour during configure
1138 with --enable-health-write.
1140 Severity : enhancement
1142 Description: 64-bit inode version
1143 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1144 to the inode. This will be required for version-based recovery.
1149 Description: MDS fails to start if a duplicate client export is detected
1150 Details : in some rare cases it was possible for a client to connect to
1151 an MDS multiple times. Upon recovery the MDS would detect this
1152 and fail during startup. Handle this more gracefully.
1154 Severity : enhancement
1156 Description: Add -o localflock option to simulate old noflock
1158 Details : This will achieve local-only flock/fcntl locks
1164 Description: log_commit_thread vs filter_destroy race leads to crash
1165 Details : Take import reference before releasing llog record semaphore
1170 Description: Wrong request locking in request set processing
1171 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1172 imp_delayed_list, in this place should be used imp_lock.
1175 Frequency : when reconnection
1177 Description: Grant Leak when osc reconnect to OST
1178 Details : When osc reconnect ost, OST(filter) should check whether it
1179 should grant more space to client by comparing fed_grant and
1180 cl_avail_grant, and return the granted space to client instead
1181 of "new granted" space, because client will call osc_init_grant
1182 to update the client grant space info.
1185 Frequency : when client reconnect to OST
1187 Description: Grant Leak when osc do resend and replay bulk write
1188 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1189 bulk write request, because the grant info while be sync between
1190 OSC and OST when reconnect, and we should ignore the grant info
1191 these of resend/replay write req.
1196 Description: Grant space more than avaiable left space sometimes.
1197 Details : When then OST is about to be full, if two bulk writing from
1198 different clients came to OST. Accord the avaliable space of the
1199 OST, the first req should be permitted, and the second one
1200 should be denied by ENOSPC. But if the seconde arrived before
1201 the first one is commited. The OST might wrongly permit second
1202 writing, which will cause grant space > avaiable space.
1205 Frequency : when client is evicted
1207 Description: Grant might be wrongly erased when osc is evicted by OST
1208 Details : when the import is evicted by server, it will fork another
1209 thread ptlrpc_invalidate_import_thread to invalidate the
1210 import, where the grant will be set to 0. While the original
1211 thread will update the grant it got when connecting. So if
1212 the former happened latter, the grant will be wrongly errased
1213 because of this race.
1218 Description: Checking Stale with correct fid
1219 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1220 to check whether it is stale, because sometimes, we want the
1221 enqueue happened anyway, and op_data.fid2 will not be initialized.
1223 Severity : enhancement
1225 Description: update patchless client
1226 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1229 Frequency : only with 2.4 kernel
1231 Description: random memory corruption
1232 Details : size of struct ll_inode_info is to big for union inode.u and this
1233 can be cause of random memory corruption.
1238 Description: Memory leak in recovery
1239 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1240 It should also check obd_fail before fsfilt_start, otherwise if
1241 fsfilt_start return -EROFS,(failover mds during mds recovery).
1242 then the req will return with repmsg->transno = 0 and rc = EROFS.
1243 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1244 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1245 be freed no matter whether fsfilt_commit success or not.
1248 Frequency : only with huge count clients
1250 Description: Prevents from taking the superblock lock in llap_from_page for
1252 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1253 for llap_from_page call in ll_removepage prevents from taking the
1254 superblock lock for a soon died page.
1259 Description: Not check open intent error before release open handle
1260 Details : in some rare cases, the open intent error is not checked before
1261 release open handle, which may cause
1262 ASSERTION(open_req->rq_transno != 0), because it tries to release
1263 the failed open handle.
1268 Description: Set cat log bitmap only after create log success.
1269 Details : in some rare cases, the cat log bitmap is set too early. and it
1270 should be set only after create log success.
1274 Description: Accessing a block bevice can re-enable I/O when Lustre is
1275 tearing down a device.
1276 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1281 Description: service threads may hog cpus when there are a lot of requests
1282 Details : Insert cond_resched to give other threads a chance to use some CPU
1287 Description: the cat log was not initialized in recovery
1288 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1289 unlink log on mds will not be initialized until mds post
1290 recovery. And also in mds post recovery, the unlink log will
1291 initialization will be done asynchronausly, so there will be race
1292 between add unlink log and unlink log initialization.
1296 Description: brw_stats were being printed incorrectly
1297 Details : brw_stats were being printed as log2 but all of them were not
1298 recorded as log2. Also remove some code duplication arising from
1299 filter_tally_{read,write}.
1303 Frequency : rare, only in recovery.
1304 Description: ASSERTION(req->rq_type != LI_POISON) failed
1305 Details : imp_lock should be held while iterating over imp_sending_list for
1306 prevent destroy request after get timeout in ptlrpc_queue_wait.
1310 Description: replay-single.sh test 52 fails
1311 Details : A lock's skiplist need to be cleanup when it being unlinked
1312 from its resource list.
1316 Description: Short directio read returns full requested size rather than
1318 Details : Direct I/O operations should return actual amount of bytes
1319 transferred rather than requested size.
1321 Severity : enhancement
1323 Description: metadata RPC reduction (e.g. for rm performance)
1324 Details : decrease the amount of synchronous RPC between clients and servers
1325 by canceling conflicing lock before the operation on the client
1326 and packing thier handles into the main operation RPC to server.
1328 Severity : enhancement
1330 Description: Async OSC create to avoid the blocking unnecessarily.
1331 Details : If a OST has no remain object, system will block on the creating
1332 when need to create a new object on this OST. Now, ways use
1333 pre-created objects when available, instead of blocking on an
1334 empty osc while others are not empty. If we must block, we block
1335 for the shortest possible period of time.
1337 Severity : enhancement
1339 Description: refine locking for avoid write wrong info into lov_objid file
1340 Details : fix possible races with add new target and write/update data in
1343 --------------------------------------------------------------------------------
1345 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1350 Frequency : on some architectures
1352 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1353 Details : Uninitialized flags sometimes cause configuration commands to
1357 Frequency : patchless clients only
1359 Description: missing __iget() symbol export
1360 Details : The __iget() symbol export is missing. To avoid the need for
1361 this on patchless clients the deathrow inode reaper is turned
1362 off, and we depend on the VM to clean up old inodes. This
1363 dependency was during via the fix for bug 12181.
1365 --------------------------------------------------------------------------------
1367 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1369 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1370 INTEROPERATE with older versions automatically. In many cases a
1371 special upgrade step is needed. Please read the
1372 user documentation before upgrading any part of a live system.
1373 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1374 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1375 * WARNING: Lustre configuration and startup changes are required with
1376 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1382 Severity : enhancement
1384 Description: MountConf
1385 Details : Lustre configuration is now managed via mkfs and mount
1386 commands instead of lmc and lconf. New obd types (MGS, MGC)
1387 are added for dynamic configuration management. See
1388 https://mail.clusterfs.com/wikis/lustre/MountConf for
1391 Severity : enhancement
1393 Description: dynamic OST addition
1394 Details : OSTs can now be added to a live filesystem
1396 Severity : enhancement
1398 Description: startup order invariance
1399 Details : MDTs and OSTs can be started in any order. Clients only
1400 require the MDT to complete startup.
1402 Severity : enhancement
1404 Description: parallel, asynchronous orphan cleanup
1405 Details : orphan cleanup is now performed in separate threads for each
1406 OST, allowing parallel non-blocking operation.
1408 Severity : enhancement
1410 Description: optimized stripe assignment
1411 Details : stripe assignments are now made based on ost space available,
1412 ost previous usage, and OSS previous usage, in order to try
1413 to optimize storage space and networking resources.
1415 Severity : enhancement
1417 Description: Permanently set tunables
1418 Details : All writable /proc/fs/lustre tunables can now be permanently
1419 set on a per-server basis, at mkfs time or on a live system.
1421 Severity : enhancement
1423 Description: Lustre message v2
1424 Details : Add lustre message format v2.
1426 Severity : enhancement
1428 Description: client OST exclusion list
1429 Details : Clients can be started with a list of OSTs that should be
1430 declared "inactive" for known non-responsive OSTs.
1434 Description: SPEC SFS validation failure on NFS v2 over lustre.
1435 Details : Changes the blocksize for regular files to be 2x RPC size,
1436 and not depend on stripe size.
1438 Severity : enhancement
1440 Description: Multiple MD RPCs in flight.
1441 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1442 To avoid overly-overloading MDS, introduce a limit on number of
1443 MDS RPCs in flight for a single client and add /proc controls
1444 to adjust this limit.
1446 Severity : enhancement
1448 Description: client read/write statistics
1449 Details : Add client read/write call usage stats for performance
1450 analysis of user processes.
1451 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1452 file access. extents_stats shows chunk size distribution.
1453 extents_stats_per_process show chunk size distribution per
1456 Severity : enhancement
1458 Description: mds statistics
1459 Details : Add detailed mds operations statistics in
1460 /proc/fs/lustre/mds/*/stats.
1464 Description: Failure of copying files with lustre special EAs.
1465 Details : Client side always return success for setxattr call for lustre
1466 special xattr (currently only "trusted.lov").
1471 Description: Refcount LNET uuids
1472 Details : The global LNET uuid list grew linearly with every startup;
1473 refcount repeated list entries instead of always adding to
1476 Severity : enhancement
1478 Description: Dynamic service threads
1479 Details : Within a small range, start extra service threads
1480 automatically when the request queue builds up.
1483 Frequency : mixed-endian client/server environments
1485 Description: mixed-endian crashes
1486 Details : The new msg_v2 system had some failures in mixed-endian
1489 Severity : enhancement
1491 Description: Easy OST removal
1492 Details : OSTs can be permanently deactivated with e.g. 'lctl
1493 conf_param lustre-OST0001.osc.active=0'
1495 Severity : enhancement
1497 Description: MGS proc entries
1498 Details : Added basic proc entries for the MGS showing what filesystems
1501 Severity : enhancement
1503 Description: provide MGS failover
1504 Details : Added config lock reacquisition after MGS server failover.
1506 Severity : enhancement
1508 Description: add Linux 2.4 support
1509 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1513 Description: a large application tries to do I/O to the same resource and dies
1514 in the middle of it.
1515 Details : Check the req->rq_arrival time after the call to
1516 ost_brw_lock_get(), but before we do anything about
1517 processing it & sending the BULK transfer request. This
1518 should help move old stale pending locks off the queue as
1519 quickly as obd_timeout.
1522 Frequency : when an incorrect nid is specified during startup
1524 Description: ptlrpc connect to non-existant node causes kernel crash
1525 Details : LNET can't be re-entered from an event callback, which
1526 happened when we expire a message after the export has been
1527 cleaned up. Instead, hand the zombie cleanup off to another
1530 Severity : enhancement
1532 Description: plain/inodebits lock performance improvement
1533 Details : Grouping plain/inodebits in granted list by their request modes
1534 and bits policy, thus improving the performance of search through
1538 Frequency : only if OST filesystem is corrupted
1540 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1541 Details : for a short time RPCs with bulk IO are in the replay list,
1542 but replay of bulk IOs is unimplemented. If the OST filesystem
1543 is corrupted due to disk cache incoherency and then replay is
1544 started it is possible to trip an assertion. Avoid putting
1545 committed RPCs into the replay list at all to avoid this issue.
1548 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1550 Description: System hang on startup
1551 Details : This bug allowed the liblustre (e.g. catamount) client to
1552 return to the app before handling all startup RPCs. This
1553 could leave the node unresponsive to lustre network traffic
1554 and manifested as a server ptllnd timeout.
1556 Severity : enhancement
1558 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1559 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1560 boolean to control whether to print peer debug info when a
1561 client's RPC times out.
1564 Frequency : only for kernels with patches from Lustre below 1.4.3
1566 Description: Remove old rdonly API
1567 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1570 Frequency : only for devices with external journals
1572 Description: Set external device read-only also
1573 Details : During a commanded failover stop, we set the disk device
1574 read-only while the server shuts down. We now also set any
1575 external journal device read-only at the same time.
1578 Frequency : when upgrading from 1.4 while trying to change parameters
1580 Description: The wrong (new) MDC name was used when setting parameters for
1581 upgraded MDT's. Also allows changing of OSC (and MDC)
1582 parameters if --writeconf is specified at tunefs upgrade time.
1585 Frequency : when setting specific ost indicies
1587 Description: QOS code breaks on skipped indicies
1588 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1589 created with --index need not be sequential.
1591 Severity : enhancement
1593 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1594 Details : The uninit_groups feature works in conjunction with the kernel
1595 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1596 up the pass1 processing of e2fsck. This is a read-only feature
1597 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1598 mount filesystems that have had this feature enabled.
1600 Severity : enhancement
1602 Description: Improve multi-block allocation algorithm to avoid fragmentation
1603 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1604 allocation locality and avoid filesystem fragmentation.
1606 ------------------------------------------------------------------------------
1608 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1610 * Support for kernels:
1611 2.4.21-47.0.1.EL (RHEL 3)
1612 2.6.5-7.283 (SLES 9)
1613 2.6.9-42.0.10.EL (RHEL 4)
1614 2.6.12.6 vanilla (kernel.org)
1615 2.6.16.27-0.9 (SLES 10)
1616 * Recommended e2fsprogs version: 1.39.cfs5
1618 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1622 Frequency : occasional, depends on client load and configuration
1623 Bugzilla : 12181, 12203
1624 Description: data loss for recently-modified files
1626 Details : In some cases it is possible that recently written or created
1627 files may not be written to disk in a timely manner (this should
1628 normally be within 30s unless client IO load is very high).
1629 The problem appears as zero-length files or files that are a
1630 multiple of 1MB in size after a client crash or client eviction
1631 that are missing data at the end of the file.
1633 This problem is more likely to be hit on clients where files are
1634 repeatedly created and unlinked in the same directory, clients
1635 have a large amount of RAM, have many CPUs, the filesystem has
1636 many OSTs, the clients are rebooted frequently, and/or the files
1637 are not accessed by other nodes after being written.
1639 The presence of the problem can be detected by looking at
1640 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1641 smaller than the second (nr_unused) then dirty files will not
1642 be flushed automatically to disk. "sync; sleep 10" should be
1643 run several times on the node before unmounting it to update
1644 Lustre (this is also safe to run on nodes without this problem).
1646 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1647 that can cause this same problem, so customers using that kernel
1648 also need to update the kernel in addition to Lustre. In order
1649 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1651 It is normal that files written just before a client crash (less
1652 than 30s) may not yet have been flushed to disk, even for local
1656 Frequency : frequent on thin XT3 nodes
1658 Description: UUID collision on thin XT3 Linux nodes
1659 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1660 randomly, since we relied on an insufficiently-seeded PRNG.
1665 Description: OSS hangs after "All ost request buffers busy"
1666 Details : A deadlock between quota and journal operations caused OSS
1667 hangs after printing "All ost request buffers busy."
1670 Frequency : always on liblustre builds
1672 Description: Cleanup compiler warnings on liblustre
1675 Frequency : always on liblustre builds on XT3
1677 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1681 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1682 > 32000 subdirectories
1683 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1684 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1685 subdirectory count crosses 32000. This will aid e2fsck to
1686 correctly handle more than 32000 subdirectories.
1689 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1691 Description: System hang on startup
1692 Details : This bug allowed the liblustre (e.g. catamount) client to
1693 return to the app before handling all startup RPCs. This
1694 could leave the node unresponsive to lustre network traffic
1695 and manifested as a server ptllnd timeout.
1697 Severity : enhancement
1699 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1700 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1701 boolean to control whether to print peer debug info when a
1702 client's RPC times out.
1707 Description: make O_SYNC working on 2.6 kernels
1708 Details : 2.6 kernels use different method for mark pages for write,
1709 so need add a code to lustre for O_SYNC work.
1714 Description: Failure to close file and release space on NFS
1715 Details : Put inode details into lock acquired in ll_intent_file_open.
1716 Use mdc_intent_lock in ll_intent_open to properly
1717 detect all kind of errors unhandled by mdc_enqueue.
1722 Description: proc file read during shutdown sometimes raced obd removal,
1724 Details : Add lock to prevent obd access after proc file removal.
1727 Frequency : Only for files larger than 4GB on 32-bit clients.
1729 Description: improperly doing page alignment of locks
1730 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1731 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1736 Description: RPCs being resent when they shouldn't be
1737 Details : Some RPCs that should not be resent are being resent. This
1738 can cause inconsistencies in the RPC state machine. Do not
1739 resend such requests.
1742 Frequency : rare, only with NFS export
1744 Description: Crash on NFS re-export node
1745 Details : under very unusual load conditions an assertion is hit in
1746 ll_intent_file_open()
1749 Frequency : only if OST filesystem is corrupted
1751 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1752 Details : for a short time RPCs with bulk IO are in the replay list,
1753 but replay of bulk IOs is unimplemented. If the OST filesystem
1754 is corrupted due to disk cache incoherency and then replay is
1755 started it is possible to trip an assertion. Avoid putting
1756 committed RPCs into the replay list at all to avoid this issue.
1761 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1762 Details : Large single O_DIRECT read and write calls can fail to allocate
1763 a sufficiently large buffer to process the request. In case of
1764 allocation failure the allocation is retried with a smaller
1765 buffer and broken into smaller requests.
1767 Severity : enhancement
1769 Description: Add -o localflock option to simulate old noflock behaviour.
1770 Details : This will achieve local-only flock/fcntl locks coherentness.
1775 Description: versioning check is incomplete
1776 Details : Checking the version difference of client vs. server, report
1777 error if the gap is too big.
1782 Description: add support PG_writeback bit
1783 Details : add support for PG_writeback bit for Lustre, for more carefull
1784 work with page cache in 2.6 kernel. This also fix some deadlocks
1785 and remove hack for work O_SYNC with 2.6 kernel.
1787 Severity : enhancement
1789 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1790 Details : The uninit_groups feature works in conjunction with the kernel
1791 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1792 up the pass1 processing of e2fsck. This is a read-only feature
1793 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1794 mount filesystems that have had this feature enabled.
1796 Severity : enhancement
1798 Description: Improve multi-block allocation algorithm to avoid fragmentation
1799 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1800 allocation locality and avoid filesystem fragmentation.
1802 ------------------------------------------------------------------------------
1804 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1806 * Support for kernels:
1807 2.6.9-42.0.3.EL (RHEL 4)
1808 2.6.5-7.276 (SLES 9)
1809 2.4.21-47.0.1.EL (RHEL 3)
1810 2.6.12.6 vanilla (kernel.org)
1811 2.6.16.21-0.8 (SLES10)
1812 * Recommended e2fsprogs version: 1.39.cfs2-0
1814 * The backwards-compatible /proc/sys/portals symlink has been removed
1815 in this release. Before upgrading, please ensure that you change
1816 any configuration scripts or /etc/sysctl.conf files that access
1817 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1818 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1819 in advance of the upgrade on any system running Lustre 1.4.6 or
1820 newer, since /proc/sys/lnet was added in that version.
1821 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1825 Frequency : only when quota is used
1827 Description: avoid scanning export list for quota master
1828 Details : Change the algorithms to avoid scanning export list in order
1829 to improve the efficiency.
1832 Frequency : MDS failover only, very rarely
1834 Description: "went back in time" messages on mds failover
1835 Details : The greatest transno may be lost when the current operation
1836 finishes with an error (transno==0) and the client's last_rcvd
1837 record is over-written. Save the greatest transno in the
1838 mds_last_transno for this case.
1841 Frequency : always for specific kernels and striping counts
1843 Description: client may get "Matching packet too big" without ACL support
1844 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1845 when trying to access files in certain configurations. The
1846 clients should in fact be denied when mounting because they do
1847 not understand ACLs.
1850 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1852 Description: many clients connecting with IO in progress causes connect timeouts
1853 Details : Avoid synchronous journal commits to avoid delays caused by many
1854 clients connecting/disconnecting when bulk IO is in progress.
1855 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1856 OST_IO_PORTAL to avoid delays behind potentially many pending
1860 Frequency : occasionally with multiple writers to a single file
1862 Description: shared writes to file may result in wrong size reported by stat()
1863 Details : Allow growing of kms when extent lock is cancelled
1866 Frequency : always with random mmap IO to multi-striped file
1868 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1869 Details : Only if the hole is at the end of OST object so that kms is too
1870 small. Fix is to increase kms accordingly in ll_nopage.
1873 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1875 Description: writes to a missing object would leak memory on the OST
1876 Details : If there is an inconsistency between the MDS and OST filesystems,
1877 such that the MDS references an object that doesn't exist, writes
1878 to that object will leak memory due to incorrect cleanup in the
1879 error handling path, eventually running out of memory on the OST.
1884 Description: Creating too long symlink causes lustre errors
1885 Details : Check symlink and name lengths before sending requests to MDS.
1888 Frequency : only if flock is enabled (not on by default)
1890 Description: posix locks not released on fd closure on 2.6.9+
1891 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1892 kernels, this caused such locks not to be released on fd close and
1893 then assertions on fs unmount about still used locks.
1896 Frequency : MDS failover only, very rarely
1898 Description: clients may get ASSERTION(granted_lock != NULL)
1899 Details : When request was taking a long time, and a client was resending
1900 a getattr by name lock request. The were multiple lock requests
1901 with the same client lock handle and
1902 mds_getattr_name->fixup_handle_for_resent_request found one of the
1903 lock handles but later failed with ASSERTION(granted_lock != NULL).
1908 Description: handle->h_buffer_credits > 0, assertion failure
1909 Details : h_buffer_credits is zero after truncate, causing assertion
1910 failure. This patch extends the transaction or creates a new
1914 Frequency : NFS re-export or patchless client
1915 Bugzilla : 11179, 10796
1916 Description: Crash on NFS re-export node (__d_move)
1917 Details : We do not want to hash the dentry if we don't have a lock.
1918 But if this dentry is later used in d_move, we'd hit uninitialised
1919 list head d_hash, so we just do this to init d_hash field but
1920 leave dentry unhashed.
1923 Frequency : NFS re-export or patchless client
1925 Description: NFS exports has problem with symbolic link
1926 Details : lustre client didn't properly install dentry when re-exported
1927 to NFS or running patchless client.
1930 Frequency : NFS re-export or patchless client
1932 Description: Various nfs/patchless fixes.
1933 Details : fixes reuse disconected alias for lookup process - this fixes
1934 warning "find_exported_dentry: npd != pd",
1935 fix permission error with open files at nfs.
1936 fix apply umask when do revalidate.
1939 Frequency : occasional
1941 Description: Crash on NFS re-export node
1942 Details : calling clear_page() on the wrong pointer triggered oops in
1943 generic_mapping_read().
1946 Frequency : rarely, using O_DIRECT IO
1948 Description: unaligned directio crashes client with LASSERT
1949 Details : check for unaligned buffers before trying any requests.
1952 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1954 Description: stale data returned from RAID cache
1955 Details : If only a small amount of IO is done to the RAID device before
1956 reading it again it is possible to get stale data from the RAID
1957 cache instead of reading it from disk.
1960 Frequency : always for sles10 kernel
1962 Description: sles10 support
1963 Details : ll_follow_link: compile fixes and using of nd_set_link
1964 under newer kernels.
1967 Frequency : depends on arch, kernel and compiler version, always on sles10
1970 Description: recursive or deep enough symlinks cause stack overflow
1971 Details : getting rid of large stack-allocated variable in
1975 Frequency : depends on hardware
1977 Description: lustre write performance loss in the SLES10 kernel
1978 Details : the performance loss is caused by using of write barriers in the
1979 ext3 code. The SLES10 kernel turns barrier support on by
1980 default. The fix is to undo that change for ldiskfs.
1982 ------------------------------------------------------------------------------
1984 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1986 * Support for kernels:
1987 2.6.9-42.0.3EL (RHEL 4)
1988 2.6.5-7.276 (SLES 9)
1989 2.4.21-47.0.1.EL (RHEL 3)
1990 2.6.12.6 vanilla (kernel.org)
1994 Frequency : quota enabled and large files being deleted
1996 Description: releasing more than 4GB of quota at once hangs OST
1997 Details : If a user deletes more than 4GB of files on a single OST it
1998 will cause the OST to spin in an infinite loop. Release
1999 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2004 Description: statfs data retrieved from /proc may be stale or zero
2005 Details : When reading per-device statfs data from /proc, in the
2006 {kbytes,files}_{total,free,avail} files, it may appear
2007 as zero or be out of date.
2010 Frequency : systems with MD RAID1 external journal devices
2012 Description: lconf's call to blkid is confused by RAID1 journal devices
2013 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2014 of returning all block devices that match the journal UUID.
2017 Frequency : always, for aggregate stripe size over 4GB
2019 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2020 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2021 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2024 Frequency : always if "lfs find" used on a local file/directory
2026 Description: "lfs find" segfaults if used on a local file/directory
2027 Details : The case where a directory component was not specified wasn't
2028 handled correctly. Handle this properly.
2031 Frequency : always on ppc64
2033 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2034 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2035 perform endianness conversion.
2038 Frequency : rarely (truncate to non-zero file size after write under load)
2039 Bugzilla : 10730, 10687
2040 Description: Files padded with zeros to next 4K multiple
2041 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2042 it is possible that files that are truncated to a non-zero size
2043 immediately after being written are filled with zero bytes beyond
2044 the truncated size. No file data is lost.
2046 Severity : enhancement
2048 Description: Allow recovery/failover for liblustre clients.
2049 Details : liblustre clients were unaware of failover configurations until
2052 Severity : enhancement
2054 Description: user file locks should fail when not mounting with flock option
2055 Details : Set up an error-returning stub in ll_file_operations.lock field
2056 to prevent incorrect behaviour when client is mounted without
2057 flock option. Also, set up properly f_op->flock field for
2061 Frequency : always on ia64
2063 Description: "lfs df" loops on printing out MDS statfs information
2064 Details : The obd_ioctl_data was not initialized and in some systems
2065 this caused a failure during the ioctl that did not return
2066 an error. Initialize the struct and return an error on failure.
2069 Frequency : SLES 9 only
2071 Description: Error of copying files with lustre special EAs as root
2072 Details : Client side always return success for setxattr call for lustre
2073 special xattr (currently only "trusted.lov").
2076 Frequency : rarely on clusters with both ia64+i386 clients
2078 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2079 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2080 clients are doing concurrent writes to the same file it is
2081 possible that the ia64 clients may LASSERT because the OST
2082 extent locks are not PAGE_SIZE aligned. Ensure that grown
2083 locks are always aligned on the request boundary.
2086 Frequency : specific use, occasional
2088 Description: Overwriting in use executable truncates on-disk binary image
2089 Details : If one node attempts to overwrite an executable in use by
2090 another node, we now correctly return ETXTBSY instead of
2091 truncating the file.
2093 Severity : enhancement
2095 Description: Async OSC create to avoid the blocking unnecessarily.
2096 Details : If a OST has no remain object, system will block on the creating
2097 when need to create a new object on this OST. Now, ways use
2098 pre-created objects when available, instead of blocking on an
2099 empty osc while others are not empty. If we must block, we block
2100 for the shortest possible period of time.
2105 Description: chmod on Lustre root is propagated to other clients
2106 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2107 invalid by the follow_mount time.
2110 Frequency : liblustre clients only
2112 Description: Race in 'instant cancel' lock handling could lead to such locks
2113 never to be granted in case of SMP MDS
2114 Details : Do not destroy not yet granted but cbpending locks in
2118 Frequency : replay/resend of open
2120 Description: non null lock assetion failure in mds_intent_policy
2121 Details : Trying to replay/resend lockless open requests resulted in
2122 mds_open() returning 0 with no lock. Now it sets a flag if
2123 a lock is going to be returned.
2125 Severity : enhancement
2127 Description: Checksum enhancements
2128 Details : New checksum enhancements allow for resending RPCs that failed
2131 Severity : enhancement
2133 Description: Tunables on number of dirty pages in cacche
2134 Details : Allow to set limit on number of dirty pages cached.
2139 Description: client crash on unmount - lock still has references
2140 Details : In some error handling cases it was possible to leak a lock
2141 reference on a client while accessing a file. This was not
2142 harmful to the client during operation, but would cause the
2143 client to crash when the filesystem is unmounted.
2146 Frequency : specific case, rare
2148 Description: ETXTBSY on mds though file not in use by client
2149 Details : ETXTBSY is no longer incorrectly returned when attempting to
2150 chmod or chown a directory that the user previously tried to
2151 execute or a currently-executing binary.
2154 Frequency : extremely rare except on liblustre-based clients
2156 Description: Lustre space not freed when files are deleted
2157 Details : Clean up open-unlinked files after client eviction. Previously
2158 the unlink was skipped and the files remained as orphans.
2163 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2164 Details : In some cases it was possible to send a blocking callback to
2165 a client doing a glimpse, even though that client didn't get
2166 a lock granted. When the glimpse lock is cancelled on the OST
2167 the freed lock is left on the waiting list and corrupted the list.
2170 Frequency : all core dumps
2172 Description: Broke core dumps to lustre
2173 Details : Negative dentry may be unhashed if parent does not have UPDATE
2174 lock, but some callers, e.g. do_coredump, expect dentry to be
2175 hashed after successful create, hash it in ll_create_it.
2177 ------------------------------------------------------------------------------
2179 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2181 * Support for kernels:
2182 2.6.9-42.0.2.EL (RHEL 4)
2183 2.6.5-7.276 (SLES 9)
2184 2.4.21-40.EL (RHEL 3)
2185 2.6.12.6 vanilla (kernel.org)
2189 Frequency : always on RHEL 3
2191 Description: Number of open files grows over time
2192 Details : The number of open files grows over time, whether or not
2193 Lustre is started. This was due to a filp leak introduced
2194 by one of our kernel patches.
2196 ------------------------------------------------------------------------------
2198 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2200 * Support for kernels:
2201 2.6.9-42.EL (RHEL 4)
2202 2.6.5-7.276 (SLES 9)
2203 2.4.21-40.EL (RHEL 3)
2204 2.6.12.6 vanilla (kernel.org)
2209 Bugzilla : 5719, 9635, 9792, 9684
2210 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2211 Details : If a server is under heavy load and cannot reply to new
2212 connection requests before the client resends the (re)connect,
2213 the connection handling code can behave badly if two service
2214 threads are concurrently handing separate (re)connections from
2215 the same client. Add better locking to the connection handling
2216 code, and ensure that only a single connection will be processed
2217 for a given client UUID, even if the lock is dropped.
2219 Severity : enhancement
2221 Description: add TCP zero-copy support to kernel
2222 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2223 sends if the hardware supports scatter-gather and checksumming.
2224 This allows socklnd to do client-write and server-read more
2225 efficiently and reduce CPU utilization from skbuf copying.
2228 Frequency : only if NFS exporting from client
2230 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2231 Details : If setting attributes on a file created under NFS that had
2232 never been opened it would be possible to oops the client
2233 if the file had no objects.
2236 Frequency : always for liblustre
2238 Description: liblustre client does MDS+OSTs setattr RPC for each write
2239 Details : When doing a write from a liblustre client, the client
2240 incorrectly issued an RPC to the MDS and each OST the file was
2241 striped over in order to update the timestamps. When writing
2242 with small chunks and many clients this could overwhelm the MDS
2243 with RPCs. In all cases it would slow down the write because
2244 these RPCs are unnecessary.
2246 Severity : enhancement
2248 Description: allow number of MDS service threads to be changed at module load
2249 Details : It is now possible to change the number of MDS service threads
2250 running. Adding "options mds mds_num_threads={N}" to the MDS's
2251 /etc/modprobe.conf will set the number of threads for the next
2252 time Lustre is restarted (assuming the "mds" module is also
2253 reloaded at that time). The default number of threads will
2254 stay the same, 32 for most systems.
2259 Description: OST crash if filesystem is unformatted or corrupt
2260 Details : If an OST is started on a device that has never been formatted
2261 or if the filesystem is corrupt and cannot even mount then the
2262 error handling cleanup routines would dereference a NULL pointer.
2267 Description: NULL pointer deref in llap_from_page.
2268 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2269 incorrect) mapping if the page was truncated/reclaimed while it was
2270 searched for. Check for this condition and skip such pages when
2271 doing readahead. Introduce extra check to llap_from_page() to
2272 verify page->mapping->host is non-NULL (so page is not anonymous).
2275 Frequency : Sometimes when using sys_sendfile
2277 Description: "page not covered by a lock" warnings from ll_readpage
2278 Details : sendfile called ll_readpage without right page locks present.
2279 Now we introduced ll_file_sendfile that does necessary locking
2280 around call to generic_file_sendfile() much like we do in
2284 Frequency : with certain MDS communication failures at client mount time
2286 Description: NULL pointer deref after failed client mount
2287 Details : a client connection request may delayed by the network layer
2288 and not be sent until after the PTLRPC layer has timed out the
2289 request. If the client fails the mount immediately it will try
2290 to clean up before the network times out the request. Add a
2291 reference from the request import to the obd device and delay
2292 the cleanup until the network drops the request.
2295 Frequency : occasionally during client (re)connect
2297 Description: assertion failure during client (re)connect
2298 Details : processing a client connection request may be delayed by the
2299 client or server longer than the client connect timeout. This
2300 causes the client to resend the connection request. If the
2301 original connection request is replied in this interval, the
2302 client may trip an assertion failure in ptlrpc_connect_interpret()
2303 which thought it would be the only running connect process.
2306 Frequency : only with obd_echo servers and clients that are rebooted
2308 Description: kernel BUG accessing uninitialized data structure
2309 Details : When running an obd_echo server it did not start the ping_evictor
2310 thread, and when a client was evicted an uninitialized data
2311 structure was accessed. Start the ping_evictor in the RPC
2312 service startup instead of the OBD startup.
2314 Severity : enhancement
2315 Bugzilla : 10193 (patchless)
2316 Description: Remove dependency on various unexported kernel interfaces.
2317 Details : No longer need reparent_to_init, exit_mm, exit_files,
2318 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2321 Frequency : rare (only users of deprecated and unsupported LDAP config)
2323 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2324 Details : LDAP apparently contains 'lustreName' attributes instead of
2325 'name'. A simple remapping of the name is sufficient.
2328 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2330 Description: waiting_locks_callback trips kernel BUG if client is evicted
2331 Details : Running with the dump_on_timeout debug flag turned on makes
2332 it possible that the waiting_locks_callback() can try to dump
2333 the Lustre kernel debug logs from an interrupt handler. Defer
2334 this log dumping to the expired_lock_main() thread.
2336 Severity : enhancement
2338 Description: Support NFS exporting on 2.6 kernels.
2339 Details : Implement non-rawops metadata methods for NFS server to use without
2340 changing NFS server code.
2343 Frequency : very rare (synthetic metadata workload only)
2345 Description: two racing renames might cause an MDS thread to deadlock
2346 Details : Running the "racer" program may cause one MDS thread to rename
2347 a file from being the source of a rename to being the target of
2348 a rename at exactly the same time that another thread is doing
2349 so, and the second thread has already enqueued these locks after
2350 doing a lookup of the target and is trying to relock them in
2351 order. Ensure that we don't try to re-lock the same resource.
2354 Frequency : only very large systems with liblustre clients
2356 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2357 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2358 all OSTs in parallel. This allows the eviction of stale liblustre
2359 clients to proceed much faster than if they were done in series,
2360 and also offers similar improvements for other set_info RPCs.
2365 Description: excessive CPU usage during initial read phase on client
2366 Details : During the initial read phase on a client, it would agressively
2367 retry readahead on the file, consuming too much CPU and impacting
2368 performance (since 1.4.5.8). Improve the readahead algorithm
2369 to avoid this, and also improve some other common cases (read
2370 of small files in particular, where "small" is files smaller than
2371 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2376 Description: MDS crash when receiving packet with unknown intent.
2377 Details : Do not LBUG in unknown intent case, just return -EFAULT
2379 Severity : enhancement
2380 Bugzilla : 9293, 9385
2381 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2382 Details : Do not serialize getattr (non-intent version) and statfs.
2385 Frequency : occasional, when OST network is overloaded/intermittent
2387 Description: client evicted by OST after bulk IO timeout
2388 Details : If a client sends a bulk IO request (read or write) the OST
2389 may evict the client if it is unresposive to its data GET/PUT
2390 request. This is incorrect if the network is overloaded (takes
2391 too long to transfer the RPC data) or dropped the OST GET/PUT
2392 request. There is no need to evict the client at all, since
2393 the pinger and/or lock callbacks will handle this, and the
2394 client can restart the bulk request.
2397 Frequency : Always when mmapping file with no objects
2399 Description: client crashes when mmapping file with no objects
2400 Details : Check that we actually have objects in a file before doing any
2401 operations on objects in ll_vm_open, ll_vm_close and
2407 Description: Request leak when working with deleted CWD
2408 Details : Introduce advanced request refcount tracking for requests
2409 referenced from lustre intent.
2411 Severity : Enhancement
2413 Description: Cache open file handles on client.
2414 Details : MDS now will return special lock along with openhandle, if
2415 requested and client is allowed to hold openhandle, even if unused,
2416 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2417 closing files for every read/write openration.
2419 Severity : Enhancement
2421 Description: Cache open negative dentries on client when possible.
2422 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2423 negative dentries on lock revocation.
2428 Description: Remounting a client read-only wasn't possible with a zconf mount
2429 Details : It wasn't possible to remount a client read-only with llmount.
2431 Severity : enhancement
2432 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2433 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2434 MPICH is included to add Lustre-specific ADIO interfaces.
2435 This is based closely on the UFS ADIO layer and only differs
2436 in file creation, in order to allow the OST striping to be set.
2437 This is user-contributed code and not supported by CFS.
2442 Description: extended inode attributes (immutable, append-only) work improperly
2443 when 2.4 and 2.6 kernels are used on client/server or vice versa
2444 Details : Introduce kernel-independent values for these flags.
2446 Severity : enhancement
2449 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2450 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2451 granularity. Now, Fractional MB granularity is be supported,
2452 this is very useful for low memory system.
2454 Severity : enhancement
2456 Description: Getattr by fid
2457 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2458 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2459 have locks and avoids getattr rpc requests for it on every stat.
2462 Frequency : Always, for filesystems larger than 2TB
2464 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2465 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2466 filesystems larger than 2TB to be mounted with mballoc enabled.
2469 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2471 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2472 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2473 filesystems larger than 2TB is mounted then the kernel will
2474 silently corrupt the start of the filesystem. CONFIG_LBD is
2475 enabled for all CFS-supported kernels, but the possibility of
2476 this happening with a modified kernel config exists.
2478 Severity : enhancement
2480 Description: add client O_DIRECT support for 2.6 kernels
2481 Details : It is now possible to do O_DIRECT reads and writes to files
2482 in the Lustre client mountpoint on 2.6 kernel clients.
2484 Severity : enhancement
2486 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2487 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2488 parallel, not waiting for response from every OST before sending
2489 a rpc to the next OST.
2494 Description: setattr vs write race when updating file timestamps
2495 Details : Client processes that update a file timestamp into the past
2496 right after writing to the file (e.g. tar) it is possible that
2497 the updated file modification time can be reset to the current
2498 time due to a race between processing the setattr and write RPC.
2500 Severity : enhancement
2502 Description: Bring 'lfs find' closer in line with regular Linux find.
2503 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2504 -print0 options and obtains all the needed info through the lustre
2507 Severity : enhancement
2509 Description: support up to 1024 configured devices on one node
2510 Details : change obd_dev array from statically allocated to dynamically
2511 allocated structs as they are first used to reduce memory usage
2516 Description: Flush dirty partially truncated pages during truncate
2517 Details : Immediatelly flush partially truncated pages in filter_setattr,
2518 this way we completely avoid having any pages in page cache on OST
2519 and can retire ugly workarounds during writes to flush such pages.
2524 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2525 Details : protect inode from truncation within vfs_unlink() context
2526 just take a reference before calling vfs_unlink() and release it
2527 when parent's i_sem is free.
2532 Description: last_id value checked outside lock on OST caused LASSERT failure
2533 Details : If there were multiple MDS->OST object precreate requests in
2534 flight, it was possible that the OST's last object id was checked
2535 outside a lock and incorrectly tripped an assertion. Move checks
2536 inside locks, and discard old precreate requests.
2539 Frequency : always, if extents are used on OSTs
2541 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2542 Details : OSTs using the extents format would not zero the high 16 bits of
2543 the index physical block number. This is not a problem for any
2544 OST filesystems smaller than 16TB, and no kernels support ext3
2545 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2546 new/modified files) and can be fixed for existing filesystems
2547 with e2fsprogs-1.39-cfs1.
2552 Description: import connection selection may be incorrect if timer wraps
2553 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2554 import connections to be ignored if the 32-bit jiffies counter
2555 wraps. Use a 64-bit jiffies counter.
2558 Frequency : very large clusters immediately after boot
2560 Description: LNET request buffers exhausted under heavy short-term load
2561 Details : If a large number of client requests are generated on a service
2562 that has previously never seen so many requests it is possible
2563 that the request buffer growth cannot keep up with the spike in
2564 demand. Instead of dropping incoming requests, they are held in
2565 the LND until the RPC service can accept more requests.
2568 Frequency : Sometimes during replay
2570 Description: Assertion failure in ll_local_open after replay.
2571 Details : If replay happened on an open request reply before we were able
2572 to set replay handler, reply will become not swabbed tripping the
2573 assertion in ll_local_open. Now we set the handler right after
2574 recognising of open request
2577 Frequency : very rare
2579 Description: kernel reports "badness in vsnprintf"
2580 Details : Reading from the "recovery_status" /proc file in small chunks
2581 may cause a negative length in lprocfs_obd_rd_recovery_status()
2582 call to vsnprintf() (which is otherwise harmless). Exit early
2583 if there is no more space in the output buffer.
2585 Severity : enhancement
2587 Description: clear OBD RPC statistics by writing to them
2588 Details : It is now possible to clear the OBD RPC statistics by writing
2589 to the "stats" file.
2594 Description: Client mtime is not the same on different clients after utimes
2595 Details : In some cases, the client was using the utimes() syscall on
2596 a file cached on another node. The clients now validate the
2597 ctime from the MDS + OSTs to determine which one is right.
2602 Description: Inability to activate failout mode
2603 Details : lconf script incorrectly assumed that in pythong string's numeric
2604 value is used in comparisons.
2607 Frequency : always with multiple stripes per file
2609 Description: Inefficient object allocation for mutli-stripe files
2610 Details : When selecting which OSTs to stripe files over, for files with
2611 a stripe count that divides evenly into the number of OSTs,
2612 the MDS is always picking the same starting OST for each file.
2613 Return the OST selection heuristic to the original design.
2618 Description: mount failures may take full timeout to return an error
2619 Details : Under some heavy load conditions it is possible that a
2620 failed mount can wait for the full obd_timeout interval,
2621 possibly several minutes, before reporting an error.
2622 Instead return an error as soon as the status is known.
2624 Frequency : quota enabled and large files being deleted
2626 Description: releasing more than 4GB of quota at once hangs OST
2627 Details : If a user deletes more than 4GB of files on a single OST it
2628 will cause the OST to spin in an infinite loop. Release
2629 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2634 Description: statfs data retrieved from /proc may be stale or zero
2635 Details : When reading per-device statfs data from /proc, in the
2636 {kbytes,files}_{total,free,avail} files, it may appear
2637 as zero or be out of date.
2640 Frequency : systems with MD RAID1 external journal devices
2642 Description: lconf's call to blkid is confused by RAID1 journal devices
2643 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2644 of returning all block devices that match the journal UUID.
2647 Frequency : always, for aggregate stripe size over 4GB
2649 Description: assertion fails when trying to use 4GB stripe size
2650 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2651 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2654 Frequency : always on ppc64
2656 Description: the first write on an ext3 filesystem with mballoc got stuck
2657 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2658 perform endianness conversion.
2660 ------------------------------------------------------------------------------
2662 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2664 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2665 INTEROPERATE with older versions automatically. Please read the
2666 user documentation before upgrading any part of a live system.
2667 * WARNING: Lustre networking configuration changes are required with
2668 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2671 * Support for kernels:
2672 2.6.9-22.0.2.EL (RHEL 4)
2673 2.6.5-7.244 (SLES 9)
2674 2.6.12.6 vanilla (kernel.org)
2677 Severity : enhancement
2678 Bugzilla : 7981/8208
2679 Description: Introduced Lustre Networking (LNET)
2680 Details : LNET is new networking infrastructure for Lustre, it includes
2681 a reorganized network configuration mode (see the user
2682 documentation for full details) as well as support for routing
2683 between different network fabrics. Lustre Networking Devices
2684 (LNDs) for the supported network fabrics have also been
2685 created for this new infrastructure.
2687 Severity : enhancement
2688 Description: Introduced Access control lists
2689 Details : clients can set ACLs on files and directories in order to have
2690 more fine-grained permissions than the standard Unix UGO+RWX.
2691 The MDS must be started with the "-o acl" mount option.
2693 Severity : enhancement
2694 Description: Introduced filesystem quotas
2695 Details : Administrators may now establish per-user quotas on the
2698 Severity : enhancement
2700 Description: Configuration change for the XT3
2701 The PTLLND is now used to run Lustre over Portals on the XT3
2702 The configure option(s) --with-cray-portals are no longer used.
2703 Rather --with-portals=<path-to-portals-includes> is used to
2704 enable building on the XT3. In addition to enable XT3 specific
2705 features the option --enable-cray-xt3 must be used.
2710 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2711 Details : A race between allocating a new llcd and re-getting the llcd_lock
2712 allowed another thread to grab newly-allocated llcd.
2714 Severity : enhancement
2716 Description: 2.6 OST async journal commit and locking fix to improve performance
2717 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2718 2.6 kernels are now async as they already were in 2.4 kernels so
2719 that they can commit concurrently with the network bulk transfer.
2720 For block-allocated files the filter allocation semaphore is held
2721 to avoid filesystem fragmentation during allocation. BKL lock
2722 removed for 2.6 xattr operations where it is no longer needed.
2727 Description: lconf incorrectly determined whether two IP networks could talk
2728 Details : In some more complicated routing and multiple-network
2729 configurations, lconf will avoid trying to make a network
2730 connection to a disjoint part of the IP space. It was doing the
2731 math incorrectly for one set of cases.
2736 Description: Fix for potential infinite loop processing records in an llog.
2737 Details : If an llog record is corrupted/zeroed, it is possible to loop
2738 forever in llog_process(). Validate the llog record length
2739 and skip the remainder of the block on error.
2742 Frequency : occasional (liblustre only)
2744 Description: liblustre could not open files whose last component is a symlink
2745 Details : sysio_path_walk() would incorrectly pass the open intent to
2746 intermediate path components.
2749 Frequency : rare (liblustre only with non-standard tuning)
2750 Bugzilla : 7201 (7350)
2751 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2752 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2753 after it has been released, only do this for referenced locks
2755 Severity : enhancement
2757 Description: specifying an (invalid) directory default stripe_size of -1
2758 would reset the directory default striping
2759 Details : stripe_size -1 was used internally to signal directory stripe
2760 removal, now use "all default" to signal dir stripe removal
2761 as a directory striping of "all default" is not useful
2764 Frequency : common for large clusters running liblustre clients
2766 Description: doing an ls when liblustre clients are running is slow
2767 Details : sending a glimpse AST to a liblustre client waits for every AST
2768 to time out, as liblustre clients will not respond. Since they
2769 cannot cache data we refresh the OST lock LVB from disk instead.
2771 Severity : enhancement
2773 Description: doing an ls at the same time as file IO can be slow
2774 Details : enqueue and other "small" requests can be blocked behind many
2775 large IO requests. Create a new OST IO portal for non-IO
2776 requests so they can be processed faster.
2779 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2781 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2782 created file with mode 000
2783 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2784 allow an ftruncate() as a non-root user to a file with mode 000.
2785 Lustre now respects this flag to disable mode checks when
2786 truncating a file owned by the user
2789 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2792 Description: Revoking locks from clients that went dead or catatonic might take
2794 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2795 cancellation of such locks instant on servers without waiting for
2796 any reply from clients. Clients drops these locks when cancel
2797 notification from server is received without replying.
2800 Frequency : liblustre-only, when liblustre client dies or becomes busy
2802 Description: Doing ls on Linux clients can take a long time with active
2804 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2805 granting such locks to it in the first place if possible. Locks
2806 are taken by proxy on the OST during the read or write and
2807 dropped immediately afterward. Add connect flags handling, do
2808 not grant locks to liblustre clients for glimpse ASTs.
2810 Severity : enhancement
2812 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2813 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2814 sequential reads and behaves badly with applications doing random
2815 reads. Improve it by reading ahead at least read region, and
2816 avoiding excessive large RPC for small reads.
2818 Severity : enhancement
2820 Description: Creating more than 1000 files for a single job may cause a load
2821 imbalance on the OSTs if there are also a large number of OSTs.
2822 Details : qos_prep_create() uses an OST index reseed value that is an
2823 even multiple of the number of available OSTs so that if the
2824 reseed happens in the middle of the object allocation it will
2825 still utilize the OSTs as uniformly as possible.
2830 Description: OST or MDS may oops in ping_evictor_main()
2831 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2832 but doesn't restart at beginning of obd_exports_timed list
2835 Severity : enhancement
2837 Description: improve by-nid export eviction on the MDS and OST
2838 Details : allow multiple exports with the same NID to be evicted at one
2839 time without re-searching the exports list.
2842 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2844 Description: MDS may oops in groups_free()
2845 Details : in rare race conditions a newly allocated group_info struct is
2846 freed again, and this can be NULL. The 2.4 compatibility code
2847 for groups_free() checked for a NULL pointer, but 2.6 did not.
2850 Frequency : common for liblustre clients doing little filesystem IO
2851 Bugzilla : 9352, 7313
2852 Description: server may evict liblustre clients accessing contended locks
2853 Details : if a client is granted a lock or receives a completion AST
2854 with a blocking AST already set it would not reply to the AST
2855 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2858 Frequency : lfs setstripe, only systems with more than 160 OSTs
2860 Description: unable to set striping with a starting offset beyond OST 160
2861 Details : llapi_create_file() incorrectly limited the starting stripe
2862 index to the maximum single-file stripe count.
2865 Frequency : LDAP users only
2867 Description: lconf did not handle in-kernel recovery with LDAP properly
2868 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2870 Severity : enhancement
2872 Description: bind OST threads to NUMA nodes to improve performance
2873 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2874 node and do their allocations there to localize memory access
2876 Severity : enhancement
2878 Description: llmount can determine client NID directly from Myrinet (GM)
2879 Details : the client NID code from gmnalnid was moved directly into
2880 llmount, removing the need to use this or specifying the
2881 client NID explicitly when mounting GM clients with zeroconf
2884 Frequency : if client is started with down MDS
2886 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2887 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2888 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2889 ptlrpc_import_delayed_req() because we didn't check if the
2890 request was interrupted, and we also didn't break out of the
2891 event loop if there was no timeout
2896 Description: data loss during non-page-aligned writes to a single file from
2897 both multiple nodes and multiple threads on one node at same time
2898 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2899 inconsistency led to false short-reads, that were cached and later
2900 used by ->prepare_write() to fill in partially written page,
2901 leading to data loss.
2904 Frequency : always, if lconf --abort_recovery used
2906 Description: lconf --abort_recovery fails with 'Operation not supported'
2907 Details : lconf was attempting to abort recovery on the MDT device and not
2910 Severity : enhancement
2912 Description: remove cleanup logs
2913 Details : replace lconf-generated cleanup logs with lustre internal
2914 cleanup routines. Eliminates the need for client-cleanup and
2917 Severity : enhancement
2919 Description: add support for EAs (user and system) on lustre filesystems
2920 Details : it is now possible to store extended attributes in the Lustre
2921 client filesystem, and with the user_xattr mount option it
2922 is possible to allow users to store EAs on their files also
2924 Severity : enhancement
2926 Description: Add possibility (config option) to show minimal available OST free
2928 Details : When compiled with --enable-mindf configure option, statfs(2)
2929 (and so, df) will return least minimal free space available from
2930 all OSTs as amount of free space on FS, instead of summary of
2931 free spaces of all OSTs.
2933 Severity : enhancement
2935 Description: do not expand extent locks acquired on OST-side
2936 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2937 by server: they are not cached anyway.
2940 Frequency : when mmap is used/binaries executed from Lustre
2942 Description: Unmmap pages before throwing them away from read cache.
2943 Details : llap_shrink cache now attempts to unmap pages before discarding
2944 them (if unmapping failed - do not discard). SLES9 kernel has
2945 extra checks that trigger if this unmapping is not done first.
2950 Description: lconf didn't resolve symlinks before checking to see whether a
2951 given mountpoint was already in use
2954 Frequency : when migrating failover services
2955 Bugzilla : 6395, 9514
2956 Description: When migrating a subset of services from a node (e.g. failback
2957 from a failover service node) the remaining services would
2958 time out and evict clients.
2959 Details : lconf --force (implied by --failover) sets the global obd_timeout
2960 to 5 seconds in order to quickly disconnect, but this caused
2961 other RPCs to time out too quickly. Do not change the global
2962 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2964 Severity : enhancement
2965 Frequency : if MDS is started with down OST
2966 Bugzilla : 9439,5706
2967 Description: Allow startup/shutdown of an MDS without depending on the
2968 availability of the OSTs.
2969 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2970 Add appropriate locking and lov-osc refcounts for safe
2971 cleaning. Add osc abort_inflight calls in case the
2972 synchronize never started.
2975 Frequency : occasional (Cray XT3 only)
2977 Description: root not authorized to access files in CRAY_PORTALS environment
2978 Details : The client process capabilities were not honoured on the MDS in
2979 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2980 been accessed by an authorized user then root was able to access
2981 the file on the local client also. The root user capabilities
2982 are now allowed on the MDS, as this environment has secure UID.
2985 Frequency : occasional
2987 Description: ldiskfs "too long searching" message happens too often
2988 Details : A debugging message (otherwise harmless) prints too often on
2989 the OST console. This has been reduced to only happen when
2990 there are fragmentation problems on the filesystem.
2995 Description: Division by zero in statfs when all OSCs are inactive
2996 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2997 lov_statfs divides by value returned by lov_get_stripecnt().
3001 Bugzilla : 9489, 3273
3002 Description: First write from each client to each OST was only 4kB in size,
3003 to initialize client writeback cache, which caused sub-optimal
3004 RPCs and poor layout on disk for the first writen file.
3005 Details : Clients now request an initial cache grant at (re)connect time
3006 and so that they can start streaming writes to the cache right
3007 away and always do full-sized RPCs if there is enough data.
3008 If the OST is rebooted the client also re-establishes its grant
3009 so that client cached writes will be honoured under the grant.
3014 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3015 Details : Now I/O RPCs go to different portal number and (presumably) fast
3016 lock requests (and glimses) and other RPCs get their own service
3017 threads pool that should be able to service those RPCs
3020 Severity : enhancement
3022 Description: Ability to exchange lustre version between client and servers and
3023 issue warnings at client side if client is too old. Also for
3024 liblustre clients there is ability to refuse connection of too old
3026 Details : New 'version' field is added to connect data structure that is
3027 filled with version info. That info is later checked by server and
3031 Frequency : rare, liblustre only.
3032 Bugzilla : 9296, 9581
3033 Description: Two simultaneous writes from liblustre at offset within same page
3034 might proceed at the same time overwriting eachother with stale
3036 Details : I/O lock withing llu_file_prwv was released too early, before data
3037 actually was hitting the wire. Extended lock-holding time until
3038 server acknowledges receiving data.
3041 Frequency : extremely rare. Never observed in practice.
3043 Description: avoid generating lustre_handle cookie of 0.
3044 Details : class_handle_hash() generates handle cookies by incrementing
3045 global counter, and can hit 0 occasionaly (this is unlikely, but
3046 not impossible, because initial value of cookie counter is
3047 selected randonly). Value of 0 is used as a sentinel meaning
3048 "unassigned handle" --- avoid it. Also coalesce two critical
3049 sections in this function into one.
3051 Severity : enhancement
3053 Description: allow liblustre clients to delegate truncate locking to OST
3054 Details : To avoid overhead of locking, liblustre client instructs OST to
3055 take extent lock in ost_punch() on client's behalf. New connection
3056 flag is added to handle backward compatibility.
3058 Severity : enhancement
3059 Bugzilla : 4928, 7341, 9758
3060 Description: allow number of OST service threads to be specified
3061 Details : a module parameter allows the number of OST service threads
3062 to be specified via "options ost ost_num_threads={N}" in the
3063 OSS's /etc/modules.conf or /etc/modprobe.conf.
3067 Bugzilla : 6146, 9635, 9895
3068 Description: servers crash with bad pointer in target_handle_connect()
3069 Details : In rare cases when a client is reconnecting it was possible that
3070 the connection request was the last reference for that export.
3071 We would temporarily drop the export reference and get a new
3072 one, but this may have been the last reference and the export
3073 was just destroyed. Get new reference before dropping old one.
3075 Severity : enhancement
3076 Frequency : if client is started with failover MDS
3078 Description: Allow multiple MDS hostnames in the mount command
3079 Details : Try to read the configuration from all specified MDS
3080 hostnames during a client mount in case the "primary"
3083 Severity : enhancement
3085 Description: Stop sending data to evicted clients as soon as possible.
3086 Details : Check if the client we are about to send or are sending data to
3087 was evicted already. (Check is done every second of waiting,
3088 for which l_wait_event interface was extended to allow checking
3089 of exit condition at specified intervals).
3092 Frequency : rare, normally only when NFS exporting is done from client
3094 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3096 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3099 Frequency : always, if srand() is called before liblustre initialization
3101 Description: Liblustre uses system PRNG disturbing its usage by user application
3102 Details : Introduce internal to lustre fast and high-quality PRNG for
3103 lustre usage and make liblustre and some other places in generic
3104 lustre code to use it.
3106 Severity : enhancement
3107 Bugzilla : 9477, 9557, 9870
3108 Description: Verify that the MDS configuration logs are updated when xml is
3109 Details : Check if the .xml configuration logs are newer than the config
3110 logs stored on the MDS and report an error if this is the case.
3111 Request --write-conf, or allow starting with --old_conf.
3113 Severity : enhancement
3115 Description: Handle symlinks in the path when checking if Lustre is mounted.
3116 Details : Resolve intermediate symlinks when checking if a client has
3117 mounted a filesystem to avoid duplicate client mounts.
3122 Description: lconf can hit an error exception but still return success.
3123 Details : The lconf command catches the Command error exception at the top
3124 level script context and will exit with the associated exit
3125 status, but doesn't ensure that this exit status is non-zero.
3130 Description: failure of ptlrpc thread startup can cause oops
3131 Details : Starting a ptlrpc service thread can fail if there are a large
3132 number of threads or the server memory is very fragmented.
3133 Handle this without oopsing.
3136 Frequency : always, only if liblustre and non-default acceptor port was used
3138 Description: liblustre cannot connect to servers with non-default acceptor port
3139 Details : tcpnal_set_default_params() was not called and was therefore
3140 ignoring the environment varaible TCPNAL_PORT, as well as other
3141 TCPNAL_ environment variables
3146 Description: two objects could be created on the same OST for a single file
3147 Details : If an OST is down, in some cases it was possible to create two
3148 objects on a single OST for a single file. No problems other
3149 than potential performance impact and spurious error messages.
3153 Bugzilla : 5681, 9562
3154 Description: Client may oops in ll_unhash_aliases
3155 Details : Client dcache may become inconsistent in race condition.
3156 In some cases "getcwd" can fail if the current directory is
3162 Description: Inode refcounting problems in NFS export code
3163 Details : link_raw functions used to call d_instantiate without obtaining
3164 extra inode reference first.
3168 Bugzilla : 9942, 9903
3169 Description: Referencing freed requests leading to crash, memleaks with NFS.
3170 Details : We used to require that call to ll_revalidate_it was always
3171 followed by ll_lookup_it. Also with revalidate_special() it is
3172 possible to call ll_revalidate_it() twice for the same dentry
3173 even if first occurence returned success. This fix changes semantic
3174 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3175 reference on a request referred from the intent.
3176 ll_intent_release() then releases such a request.
3179 Frequency : rare, normally benchmark loads only
3181 Description: unlinked inodes were kept in memory on the client
3182 Details : If a client is repeatedly creating and unlinking files it
3183 can accumulate a lot of stale inodes in the inode slab cache.
3184 If there is no other client load running this can cause the
3185 client node to run out of memory. Instead flush old inodes
3186 from client cache that have the same inode number as a new inode.
3189 Frequency : SLES9 2.6.5 kernel and long filenames only
3190 Bugzilla : 9969, 10379
3191 Description: utime reports stale NFS file handle
3192 Details : SLES9 uses out-of-dentry names in some cases, which confused
3193 the lustre dentry revalidation. Change it to always use the
3197 Frequency : rare, unless heavy write-truncate concurrency is continuous
3198 Bugzilla : 4180, 6984, 7171, 9963, 9331
3199 Description: OST becomes very slow and/or deadlocked during object unlink
3200 Details : filter_destroy() was holding onto the parent directory lock
3201 while truncating+unlinking objects. For very large objects this
3202 may block other threads for a long time and slow overall OST
3203 responsiveness. It may also be possible to get a lock ordering
3204 deadlock in this case, or run out of journal credits because of
3205 the combined truncate+unlink. Solution is to do object truncate
3206 first in one transaction without parent lock, and then do the
3207 final unlink in a new transaction with the parent lock. This
3208 reduces the lock hold time dramatically.
3211 Frequency : rare, 2.4 kernels only
3213 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3214 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3215 before unmounting their filesystem in order to avoid deadlock.
3216 The kernel_locked() function in 2.4 kernels only checks whether
3217 the kernel lock is held, not whether it is this process that is
3218 holding it as 2.6 kernels do.
3223 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3224 Details : The client ptlrpc code may be trying to reconnect to a down
3225 server before a previous connection attempt has timed out.
3226 Increase the reconnect interval to be longer than the connection
3227 timeout interval to avoid sending duplicate connections to
3231 Frequency : echo_client brw_test command
3233 Description: fix echo_client to work with OST preallocated code
3234 Details : OST preallocation code (5137) didn't take echo_client IO path
3235 into account: echo_client calls filter methods outside of any
3236 OST thread and, hence, there is no per-thread preallocated
3237 pages and buffers to use. Solution: hijack pga pages for IO. As
3238 a byproduct, this avoids unnecessary data copying.
3242 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3243 Description: Client can oops in mdc_commit_close() after open replay
3244 Details : It was possible for the MDS to return an open request with no
3245 transaction number in mds_finish_transno() if the client was
3246 evicted, but without actually returning an error. Clients
3247 would later try to replay that open and may trip an assertion
3248 Simplify the client close codepath, and always return an error
3249 from the MDS in case the open is not successful.
3252 Frequency : rare, 2.6 OSTs only
3254 Description: OST may deadlock under high load on fragmented files
3255 Details : If there was a heavy load and highly-fragmented OST filesystems
3256 it was possible to have all the OST threads deadlock waiting on
3257 allocation of biovecs, because the biovecs were not released
3258 until the entire RPC IO was completed. Instead, release biovecs
3259 as soon as they are complete to ensure forward IO progress.
3261 Severity : enhancement
3263 Description: Support for specifying external journal device at mount
3264 Details : If an OST or MDS device is formatted with an external journal
3265 device, this device major/minor is stored in the ext3 superblock
3266 and may not be valid for failover. Allow detecting and
3267 specifying the external journal at mount time.
3272 Description: Mounting an MDS with pending unlinked files may cause oops
3273 Details : target_finish_recovery() calls mds_postrecov() which returned
3274 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3275 considers this an error and immediately begins cleaning up the
3276 lov, just after starting the mds_lov process
3278 Severity : enhancement
3280 Description: Implement 'lfs df' to report actual free space on per-OST basis
3281 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3282 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3283 usage of objects; '-h' to report in human readable format.
3285 ------------------------------------------------------------------------------
3287 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3294 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3295 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3296 at startup during orphan unlinking. Instead, skip buddy update
3297 before setup, as it will be regenerated after recovery is complete.
3302 Description: If an OST is inactive, its locks might reference stale inodes.
3303 Details : lov_change_cbdata() must iterate over all namespaces, even if
3304 they are inactive to clear inode references from the lock.
3306 Severity : enhancement
3307 Frequency : occasional, if non-standard max_dirty_mb used
3309 Description: Client will block write RPCs if not enough grant
3310 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3311 then the client will block writes while waiting for another RPC
3312 to complete instead of consuming its dirty limit. With change
3313 we get improved performance when max_dirty_mb is small.
3315 Severity : enhancement
3316 Bugzilla : 3389, 6253
3317 Description: Add support for supplementary groups on the MDS.
3318 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3319 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3320 lookups for user supplementary groups into a cache.
3324 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3325 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3326 for liblustre O_EXCL is enforced by MDS.
3330 Description: readdir never returns NULL in liblustre.
3331 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3332 dentry in struct dirent.
3336 Frequency : liblustre only, depends on application IO pattern
3337 Description: liblustre clients evicted if not contacting servers
3338 Details : Don't put liblustre clients into the ping_evictor list, so
3339 they will not be evicted by the pinger ever.
3341 Severity : enhancement
3343 Description: Add ability to evict clients by NID from MDS.
3344 Details : By echoing "nid:$NID" string into
3345 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3346 $NID would be instantly evicted from this MDS and from all active
3347 OSTs connected to it.
3351 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3352 Details : lookup_it_finish() used to query file size from OSTs that was not
3357 Description: service threads change working directory to that of init
3358 Details : Starting lustre service threads may pin the working directory
3359 of the parent thread, making that filesystem busy. Threads
3360 now change to the working directory of init to avoid this.
3364 Frequency : during shutdown only
3365 Description: shutdown with a failed MDS or OST can cause unmount to hang
3366 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3371 Frequency : 2.6 only
3372 Description: chmod/chown may include an extra supplementary group
3373 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3374 supplementary group and if none is specified this is used.
3377 Bugzilla : 5479 (6816)
3379 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3380 Details : If lookup is in progress on a file that is unlinked we might try
3381 to revalidate the inode and fail in revalidate after lookup is
3382 complete and ll_file_open() enqueues the open again but
3383 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3386 Frequency : always, if lconf --abort_recovery used
3388 Description: lconf --abort_recovery fails with 'Operation not supported'
3389 Details : lconf was attempting to abort recovery on the MDT device and not
3392 ------------------------------------------------------------------------------
3394 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3399 Frequency : rare (only unsupported configurations with a node running as an
3401 Bugzilla : 6514, 5137
3402 Description: Mounting a Lustre file system on a node running as an OST could
3404 Details : OSTs now preallocates memory needed to write out data at
3405 startup, instead of when needed, to avoid having to
3406 allocate memory in possibly low memory situations.
3407 Specifically, if the file system is mounted on on OST,
3408 memory pressure could force it to try to write out data,
3409 which it needed to allocate memory to do. Due to the low
3410 memory, it would be unable to do so and the node would
3411 become unresponsive.
3413 Severity : enhancement
3415 Description: Addition of lconf --service command line option
3416 Details : lconf now accepts a '--service <arg>' option, which is
3417 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3419 Severity : enhancement
3421 Description: Failover mode is now the default for OSTs.
3422 Details : By default, OSTs will now run in failover mode. To return to
3423 the old behaviour, add '--failout' to the lmc line for OSTs.
3425 Severity : enhancement
3427 Description: Health checks are now provided for MDS and OSTs
3428 Details : Additional detailed health check information on MSD and OSTs
3429 is now provided through the procfs health_check value.
3432 Frequency : occasional, depends on IO load
3434 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3435 after numerous create/delete cycles
3436 Details : The ext3 inode allocation policy would not allocate new inodes
3437 very well on the OSTs because there are no new directories
3438 being created. Instead we look for groups with free space if
3439 the parent directories are nearly full.
3444 Description: Network or server problems during mount may cause partially
3445 mounted clients instead of returning an error.
3446 Details : The config llog parsing code may overwrite the error return
3447 code during mount error handling, returning success instead
3453 Description: MDS can fail to allocate large reply buffers
3454 Details : After long uptimes the MDS can fail to allocate large reply
3455 buffers (e.g. zconf client mount config records) due to memory
3456 fragmentation or consumption by the buffer cache. Preallocate
3457 some large reply buffers so that these replies can be sent even
3458 under memory pressure.
3462 Frequency : rare (liblustre)
3463 Description: fsx running with liblustre complained that using truncate() to
3464 extend the file doesn't work. This patch corrects that issue.
3465 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3466 ATTR_SIZE and lsm use in llu_setattr_raw.
3470 Frequency : rare, only 2.6 kernels
3471 Description: Unusual file access patterns on the MDS may result in inode
3472 data being lost in very rare circumstances.
3473 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3474 code in the 2.6 kernel caused the inode and/or EA data not to
3475 be read from disk, causing single-file corruption.
3479 Frequency : rare, only 2.6 filesystems using extents
3480 Description: Heavy concurrent write and delete load may cause data corruption.
3481 Details : It was possible under high-load situations to have an extent
3482 metadata block in the block device cache from a just-unlinked
3483 file overwrite a newly-allocated data block. We now unmap any
3484 metadata buffers that alias just-allocated data blocks.
3488 Frequency : filesystems with default stripe_count larger than 77
3489 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3490 Details : lconf specifies an inode size of 4096 bytes when the default
3491 stripe_count is larger than 77. This conflicts with the default
3492 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3493 this case to avoid pinning too much memory for large EAs.
3495 ------------------------------------------------------------------------------
3497 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3502 Frequency : rare (extremely heavy IO load with hundreds of clients)
3504 Description: Client is evicted, gets IO error writing to file
3505 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3506 caused clients to be evicted for AST timeouts. The fixes in
3507 bug 5192 mean we no longer need to have such short AST timeouts
3508 so ldlm_timeout has been increased.
3511 Frequency : occasional during --force or --failover shutdown under load
3512 Bugzilla : 5949, 4834
3513 Description: Server oops/LBUG if stopped with --force or --failover under load
3514 Details : a collection of import/export refcount and cleanup ordering
3515 issues fixed for safer force cleanup
3518 Frequency : only filesystems larger than 120 OSTs
3519 Bugzilla : 5990, 6223
3520 Description: lfs getstripe would oops on a very large filesystem
3521 Details : lov_getconfig used kfree on vmalloc'd memory
3524 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3525 Bugzilla : 6242, 6243
3526 Description: reading from files that had been truncated to a non-zero size
3527 but never opened returned no data
3528 Details : ll_file_read() reads zeros from no-object files to EOF
3533 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3534 Details : There's little harm in aborting MDS/OSS recovery and letting it
3535 try again, so I removed the LASSERT and return an error instead.
3537 Severity : enhancement
3539 Description: New debugging infrastructure for tracking down data corruption
3540 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3541 (b) cover more of the client-side code path, and (c) try to narrow
3542 down where problems occurred
3546 Bugzilla : 3819, 4364, 4397, 6313
3547 Description: Racing close and eviction MDS could cause assertion in mds_close
3548 Details : It was possible to get multiple mfd references during close and
3549 client eviction, leading to one thread referencing a freed mfd.
3551 Severity: : enhancement
3552 Bugzilla : 3262, 6359
3553 Description: Attempts to reconnect to servers are now more aggressive.
3554 Details : This builds on the enhanced upcall-less recovery that was added
3555 in 1.4.2. When trying to reconnect to servers, clients will
3556 now try each server in the failover group every 10 seconds. By
3557 default, clients would previously try one server every 25 seconds.
3562 Description: After recovery, certain operations trigger a failed
3563 assertion on a client.
3564 Details : Failing over an mds, using lconf -d --failover, while a
3565 client was doing a readdir() call would cause the client to
3566 LBUG after recovery completed and the readdir() was resent.
3568 Severity : enhancement
3570 Description: Default groups are now added by lconf
3571 Details : You can now run lconf --group <servicename> without having to
3572 manually add groups with lmc.
3575 Frequency : occasional
3577 Description: Nodes with an elan id of 0 trigger a failed assertion
3580 Frequency : always when accessing e.g. tty/console device nodes
3582 Description: tty and some other devices nodes cannot be used on lustre
3583 Details : file's private_data field is used by device data and lustre
3584 values in there got lost. New field was added to struct file to
3585 store fs-specific private data.
3588 Frequency : when exporting Lustre via NFS
3590 Description: NFSD failed occasionally when looking up a path component
3591 Details : NFSD is looking up ".." which was broken in ext3 directories
3592 that had grown large enough to become hashed.
3595 Frequency : Clusters with multiple interfaces not on the same subnet
3597 Description: Nodes will repeatedly try to reconnect to an interface which it
3598 cannot reach and report an error to the log.
3599 Details : Extra peer list entries will be created by lconf with some peers
3600 unreachable. lconf now validates the peer before adding it.
3603 Frequency : Only if a default stripe is set on the filesystem root.
3605 Description: Setting a default stripe on the filesystem root prevented the
3606 filesystem from being remounted.
3607 Details : The client was sending extra request flags in the root getattr
3608 request and did not allocate a reply buffer for the dir EA.
3611 Frequency : occasional, higher if lots of files are accessed by one client
3612 Bugzilla : 6159, 6097
3613 Description: Client trips assertion regarding lsm mismatch/magic
3614 Details : While revalidating inodes the VFS looks up inodes with ifind()
3615 and in rare cases can find an inode that is being freed.
3616 The ll_test_inode() code will free the lsm during ifind()
3617 when it finds an existing inode and then the VFS later attaches
3618 this free lsm to a new inode.
3622 Bugzilla : 6422, 7030
3623 Description: MDS deadlock between mkdir and client eviction
3624 Details : Creating a new file via mkdir or mknod (starting a transaction
3625 and getting the ns lock) can deadlock with client eviction
3626 (gets ns lock and trying to finish a synchronous transaction).
3629 Frequency : occasional
3630 Description: While starting a server, the fsfilt_ext3 module could not be
3632 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3633 kernels. Previously, lconf would still use the ext3 name
3634 when trying to load modules. Now, it will correctly use
3635 ext3 on 2.4 and ldiskfs on 2.6.
3637 Severity : enhancement
3638 Description: The default stripe count has been changed to 1
3639 Details : The interpretation of the default stripe count (0, to lfs
3640 or lmc) has been changed to mean striping across a single
3641 OST, rather than all available. For general usage we have
3642 found a stripe count of 1 or 2 works best.
3644 Severity : enhancement
3645 Description: Add support for compiling against Cray portals.
3646 Details : Conditional compiling for some areas that are different
3650 Frequency : occasional
3651 Bugzilla : 6409, 6834
3652 Description: Creating files with an explicit stripe count may lead to
3653 a failed assertion on the MDS
3654 Details : If some OSTs are full or unavailable, creating files may
3655 trigger a failed assertion on the MDS. Now, Lustre will
3656 try to use other servers or return an error to the
3660 Frequency : occasional
3662 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3663 caused lock timeout during readahead (since 1.4.2).
3664 Details : Processes doing readahead might match a lock that hasn't been
3665 granted yet if there are overlapping and conflicting lock
3666 requests. The readahead process waits on ungranted lock
3667 (original lock is CBPENDING), while OST waits for that process
3668 to cancel CBPENDING read lock and eventually evicts client.
3670 Severity : enhancement
3672 Description: Initial enabling of flock support for clients
3673 Details : Implements fcntl advisory locking and file status functions.
3674 This feature is provided as an optional mount flag (default
3675 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3676 locking are implemented yet, and those that are are not guaranteed
3677 to be completely correct in production environments.
3678 mount -t lustre -o [flock|noflock] ...
3681 Frequency : occasional
3683 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3684 assertion in the ext3 JBD (journaling) layer.
3685 Details : The b_committed_data struct is protected by the big kernel lock
3686 in 2.4 kernels, serializing journal_commit_transaction() and
3687 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3688 access to this struct. In 2.6 kernels there is finer grained
3689 locking to improve SMP performance of the JBD layer.
3693 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3698 Description: merge and cleanup kernel patches.
3701 -----------------------------------------------------------------------------
3703 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3705 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3706 versions of Lustre. Please update all servers and clients to
3707 version 1.4.2 or later at the same time. You must also run
3708 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3709 to update the configuration logs.
3711 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3712 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3713 - swab llog records of type '0' so we get proper header size/idx (5861)
3714 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3715 - fix rename of one directory over another leaking an inode (5953)
3716 - avoid SetPageDirty on 2.6 (5981)
3717 - don't re-add just-being-destroyed locks to the waiting list (5653)
3718 - when creating new directories, inherit the parent's custom
3719 striping settings if present parent (3048)
3720 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3721 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3722 - don't hold client locks on temporary worklist from l_lru (5666)
3723 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3724 - automatically evict dead clients (5921)
3725 - Update file size properly in create+truncate+fstat case (6196)
3726 - Do not unhash mountpoint dentries, do not allow removal of
3728 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3729 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3730 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3731 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3732 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3733 - Correctly handle reads of files with no objects (6243)
3734 - lctl recover will also mark a device active if deactivate used (5933)
3736 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3737 - allow --write-conf on an MDS with different nettype than client (5619)
3738 - don't write config llogs to MDS for mounts not from that MDS (5617)
3739 - lconf should create multiple TCP connections from a client (5201)
3740 - init scripts are now turned off by default; run chkconfig --on
3741 lustre and chkconfig --on lustrefs to use them
3742 - upcalls are no longer needed for clients to recover to failover
3744 - add --abort-recovery option to lconf to abort recovery on device
3746 - add support for an arbitrary number of OSTs (3026)
3747 - Quota support protocol changes.
3748 - forward compatibility changes to wire structs (6007)
3749 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3750 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3751 - improved "lustre status" script
3752 - initialize blocksize for non-regular files (6062)
3753 - added --disable-server and --disable-client configure options (5782)
3754 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3755 - Vanilla 2.4.29 support
3756 - increase maximum number of obd devices to 520 (6242)
3757 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3758 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3759 SLES 9 2.6.5 kernels
3760 - limit stripes per file to 160 (the maximum EA size) (6093)
3762 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3765 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3766 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3767 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3768 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3769 - small ext3 extents cleanups and fixes (5733)
3770 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3771 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3772 - avoid needless client->OST connect, fix handle mismatch (5317)
3773 - fix DLM error path that led to out-of-sync client, long delays (5779)
3774 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3775 - fix several locking issues related to i_size (5492,5624,5654,5672)
3776 - don't move pending lock onto export if it is already evicted (5683)
3777 - fix kernel oops when creating .foo in unlinked directory (5548)
3778 - fix deadlock in obdfilter statistics vs. object create (5811)
3779 - use time_{before,after} to avoid timer jiffies wrap (5882)
3780 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3781 - Do not leak request if server was not able to process it (5154)
3782 - If mds_open unable to find parent dir, make that negative lookup(5154)
3783 - don't create new directories with extent-mapping (5909, 5936)
3785 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3786 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3787 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3788 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3789 - support for RHEL4 kernel on i686 (5773)
3790 - provide error messages when incompatible logs are encountered (5898)
3792 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3793 * version 1.4.0.10 (1.4.1 release candidate 1)
3795 - don't keep a lock reference when lock is not granted (4238)
3796 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3797 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3798 - drop import inflight refcount on signal_completed_replay error (5255)
3799 - unlock page after async write error during send (3677)
3800 - handle missing objects in filter_preprw_read properly (5265)
3801 - no transno return for symlink open, don't save no-trasno open (3440)
3802 - don't try to complete elan receive that already failed (4012)
3803 - free RPC server reply state on error (5406)
3804 - clean up thread from ptlrpc_start_thread() on error (5160)
3805 - readahead could read extra page into cache that wasn't ejected (5388)
3806 - prevent races in class_attach/setup/cleanup/detach (5260)
3807 - don't dereference de->d_inode after l_dput of de (5458)
3808 - use "int" for stripe value returned from lock_to_stripe (5544)
3809 - mballoc allocation and error-checking fixes in 2.6 (5504)
3810 - block device patches to fix I/O request sizes in 2.6 (5482)
3811 - look up hostnames for IB nals (5602)
3812 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3813 - don't start multiple acceptors for the same port (5277)
3814 - fix incorrect LASSERT in mds_getattr_name (5635)
3815 - export a proc file for general "ping" checking (5628)
3816 - fix "lfs check" to not block when the MDS is down (5628)
3818 - service request history (4965)
3819 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3820 - create an "evict_client" /proc entry on OSTs, like the MDS has
3821 - fix mount usage message, return errors per mount(8) (5168)
3822 - change grep [] to grep "[]" in tests so they work in more UMLs
3823 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3824 - remove ancient LOV_MAGIC_V0 EA support (5047)
3825 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3826 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3827 - allow readahead allocations to fail when low on memory (5383)
3828 - mmap locking landed again, after considerable improvement (2828)
3829 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3831 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3834 - send OST transaction number in read/write reply to free req (4966)
3835 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3836 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3837 - the watchdog thread now runs as interruptible (5246)
3838 - flock/lockf fixes (but it's still disabled, pending 5135)
3839 - don't use EXT3 constants in llite code (5094)
3840 - memory shortage at startup could cause assertion (5176)
3842 - reorganization of lov code
3843 - single portals codebase
3845 - add extents/mballoc support (5025)
3846 - direct I/O reads in the obdfilter (4048)
3847 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3849 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3852 - send OST transaction number in read/write reply to free req (4966)
3853 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3854 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3855 - the watchdog thread now runs as interruptible (5246)
3856 - handle missing objects in filter_preprw_read properly (5265)
3857 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3858 - drop import inflight refcount on signal_completed_replay error (5255)
3859 - unlock page after async write error during send (3677)
3860 - return original error code on reconstructed replies (3761)
3861 - no transno return for symlink open, don't save no-trasno open (3440)
3863 - add pid to ldlm debugging output (4922)
3864 - bump the watchdog timeouts -- we can't handle 30sec yet
3865 - extra debugging for orphan dentry/inode bug (5259)
3867 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3870 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3871 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3872 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3873 - fix NULL dereference in /proc/sys/portals/routes (4827)
3874 - allow failed mdc_close() operations to be interrupted (4561)
3875 - stop precreate on OST before MDS would time out on it (4778)
3876 - don't send partial-page writes before EOF from client (4410)
3877 - discard client grant for sub-page writes on large-page clients (4520)
3878 - don't free dentries not owned by NFS code, check generation (4806)
3879 - fix lsm leak if mds_create_objects() fails (4801)
3880 - limit debug_daemon file size, always print CERROR messages (4789)
3881 - use transno after validating reply (3892)
3882 - process timed out requests if import state changes (3754)
3883 - update mtime on OST during writes, return in glimpse (4829)
3884 - add mkfsoptions to LDAP (4679)
3885 - use ->max_readahead method instead of zapping global ra (5039)
3886 - don't interrupt __l_wait_event() during strace
3888 - add software watchdogs to catch hung threads quickly (4941)
3889 - make lustrefs init script start after nfs is mounted
3890 - fix CWARN/ERROR duplication (4930)
3891 - return async write errors to application if possible (2248)
3892 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3893 - print NAL number in %x format (4645)
3894 - update barely-supported suse-2.4.21-171 series (4842)
3895 - support for sles 9 %post scripts
3896 - support for building 2.6 kernel-source packages
3897 - support for sles km_* packages
3899 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3902 - ignore -ENOENT errors in osc_destroy (3639)
3903 - notify osc create thread that OSC is being cleaned up (4600)
3904 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3905 - reconstruct ost_handle() like mds_handle() (4657)
3906 - create a new thread to do import eviction to avoid deadlock (3969)
3907 - let lconf resolve symlinked-to devices (4629)
3908 - don't unlink "objects" from directory with default EA (4554)
3909 - hold socknal file ref over connect in case target is down (4394)
3910 - allow more than 32000 subdirectories in a single directory (3244)
3911 - fix blocks count for O_DIRECT writes (3751)
3912 - OST returns ENOSPC from object create when no space left (4539)
3913 - don't send truncate RPC if file size isn't changing (4410)
3914 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3915 - bind to privileged port in socknal and tcpnal (3689)
3917 - rate limit CERROR/CWARN console message to avoid overload (4519)
3918 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3919 - basic mmap support (3918)
3920 - kernel patch series update from b1_4 (4711)
3922 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3925 - avoid crash during MDS cleanup with OST shut down (2775)
3926 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3927 - don't use bad inodes on the MDS (3744)
3928 - dynamic object preallocation to improve recovery speed (4236)
3929 - don't hold spinlock over lock dumping or change debug flags (4401)
3930 - don't zero obd_dev when it is force cleaned (3651)
3931 - print grants to console if they go negative (4431)
3932 - "lctl deactivate" will stop automatic recovery attempts (3406)
3933 - look for existing locks in ldlm_handle_enqueue() (3764)
3934 - don't resolve lock handle twice in recovery avoiding race (4401)
3935 - revalidate should check working dir is a directory (4134)
3937 - don't always mark "slow" obdfilter messages as errors (4418)
3939 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3942 - don't close LustreDB during write_conf until it is done (3860)
3943 - fix typo in lconf for_each_profile (3821)
3944 - allow dumping logs from multiple threads at one time (3820)
3945 - don't allow multiple threads in OSC recovery (3812)
3946 - fix debug_size parameters (3864)
3947 - fix mds_postrecov to initialize import for llog ctxt (3121)
3948 - replace config semaphore with spinlock (3306)
3949 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3950 - don't allow enqueue to complete on a destroyed export (3822)
3951 - down write_lock before checking llog header bitmap (3825)
3952 - recover from lock replay timeout (3764)
3953 - up llog sem before sending rpc (3652)
3954 - reduce ns lock hold times when setting kms (3267)
3955 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3956 - fix NULL deref and obd_dev leak on setup error (3312)
3957 - replace some LBUG about llog ops with error handling (3841)
3958 - don't match INVALID dentries from d_lookup and spin (3784)
3959 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3960 - fix invalid assertion in ptlrpc_set_wait (3880)
3962 - add libwrap support for the TCP acceptor (3996)
3963 - add /proc/sys/portals/routes for non-root route listing (3994)
3964 - allow setting MDS UUID in .xml (2580)
3965 - print the stack of a process that LBUGs (4228)
3967 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3970 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3971 - make sure to unset replay flag from failed open requests (3440)
3972 - if default stripe count is 0, use OST count for inode size (3636)
3973 - update parent mtime/ctime on client for create/unlink (2611)
3974 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3975 - free recovery state on server during a forced cleanup (3571)
3976 - unregister_reply for resent reqs (3063)
3977 - loop back devices mounting and status check on 2.6 (3563)
3978 - fix resource-creation race that can provoke i_size == 0 (3513)
3979 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3980 - more debugging for page-accounting assertion (3746)
3981 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3982 - avoid deadlock after precreation failure (3758)
3983 - fix race and lock order deadlock in orphan handling (3450, 3750)
3984 - add validity checks when grabbing inodes from l_ast_data (3599)
3986 - add /proc/.../recovery_status to obdfilter (3428)
3987 - lightweight CDEBUG infrastructure, debug daemon (3668)
3988 - change default OSC RPC parameters to be better on small clusters
3989 - turn off OST read cache for files smaller than 32MB
3990 - install man pages and include them in rpms (3100)
3991 - add new init script for (un)mounting lustre filesystems (2593)
3992 - run chkconfig in %post for init scripts (3701)
3993 - drop scimac NAL (unmaintained)
3995 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3998 - clean kiobufs before and after use (3485)
3999 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4000 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4001 - change init script to fail more clearly if not run as root (1528)
4002 - allow clients to reconnect during replay (1742)
4003 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4004 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4005 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4006 - fix cond_resched (3554)
4007 - only evict unfinished clients after recovery (3515)
4008 - allow bulk resend, prevent data loss (3570)
4009 - dynamic ptlrpc request buffer allocation (2102)
4010 - don't allow unlinking open directory if it isn't empty (2904)
4011 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4012 - remove extraneous obd dereference causing LASSERT failure (3334)
4013 - don't use get_cycles() when creating temp. files on the mds (3156)
4014 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4015 - handle EEXIST for set-stripe, set proper directory name (3336)
4017 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4018 - fix ksocknal_fmb_callback() error messages (2918)
4020 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4023 - don't copy lvb into (possibly NULL) reply on error (2983)
4024 - don't deref dentry after dput, don't free lvb on error (2922)
4025 - use the kms to determine writeback rpc length (2947)
4026 - increment oti_logcookies when osc is inactive (2948)
4027 - update client's i_blocks count via lvb messages (2543)
4028 - handle intent open/close of special files properly (1557)
4029 - mount MDS with errors=remount-ro, like obdfilter (2009)
4030 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4031 - don't use cancelling-locks' kms values (2947)
4032 - use highest lock extent for kms, not last one (2925)
4033 - don't dereference ERR_PTR() dentry in error handling path (3107)
4034 - fix thread race in portals_debug_dumplog() (3122)
4035 - create lprocfs device entries at setup instead of at attach (1519)
4036 - common AST error handler, don't evict client on completion race (3145)
4037 - zero nameidata in detach_mnt in 2.6 (3118)
4038 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4039 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4040 - initialize RPC timeout timer earlier for 2.6 (3219)
4041 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4042 - print nal/nid for unknown nid (3258)
4043 - additional checks for oscc recovery before doing precreate (3284)
4044 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4045 - don't crash in mdc_close for bad permissions on open (3285)
4046 - zero i_rdev for non-device files (3147)
4047 - clear page->private before handing to FS, better assertion (3119)
4048 - tune the read pipeline (3236)
4049 - fix incorrect decref of invalidated dentry (2350)
4050 - provide read-ahead stats and refine rpc in flight stats (3328)
4051 - don't hold journal transaction open across create RPC (3313)
4052 - update atime on MDS at close time (3265)
4053 - close LDAP connection when recovering to avoid server load (3315)
4054 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4055 - don't leak open file on MDS after open resend (3325)
4056 - serialize filter_precreate and filter_destroy_precreated (3329)
4057 - loop device shouldn't call sync_dev() for nul device (3092)
4058 - clear page cache after eviction (2766)
4059 - resynchronize MDS->OST in background (2824)
4060 - refuse to mount the same filesystem twice on same mountpoint (3394)
4061 - allow llmount to create routes for mounting behind routers (3320)
4062 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4063 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4064 - fix rare problem with rename on htree directories (3417)
4066 - allow default OST striping configuration per directory (1414)
4067 - fix compilation for qswnal for 2.6 kernels (3125)
4068 - increase maximum number of MDS request buffers for large systems
4069 - change liblustreapi to be useful for external progs like lfsck (3098)
4070 - increase local configuration timeout for slow disks (3353)
4071 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4073 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4076 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4077 - don't overwrite extent policy data in reply if lock was blocked (2901)
4078 - drop filter export grants atomically with removal from device (2663)
4079 - del obd_self_export from work_list in class_disconnect_exports (2908)
4080 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4081 - swab reply message in mdc_close, other PPC fixes (2464)
4082 - fix destroying of named logs (2325)
4083 - overwrite old logs when running lconf --write_conf (2264)
4084 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4085 - fix race in target_handle_connect (2898)
4086 - mds_reint_create() should take same inode create lock (2926)
4087 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4088 - don't close files for self_export to avoid uninitialized obd (2936)
4089 - allow MDS with the same name as client node (2939)
4090 - hold dentry reference for closed log files for unlink (2325)
4091 - reserve space for all logs during transactions (2059)
4092 - don't evict page beyond end of stripe extent (2925)
4093 - don't oops on a deleted current working directory (2399)
4094 - handle hard links to targets without a parent properly (2517)
4095 - don't dereference NULL lock when racing during eviction (2867)
4096 - don't grow lock extents when lots of conflicting locks (2919)
4098 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4101 - account for cache space usage on clients to avoid data loss (974)
4102 - lfsck support in lustre kernel code (2349)
4103 - reduce journal credits needed for BRW writes (2370)
4104 - orphan handling to avoid losing space on client/server crashes
4105 - ptlrpcd can be blocked, stopping ALL progress (2477)
4106 - use lock value blocks to assist in proper KMS, faster stat (1021)
4107 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4108 - recovery for initial connections (2355)
4109 - fixes for mds_cleanup_orphans (1934)
4110 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4111 - block all file creations until orphan recovery completes (1901)
4112 - client remove rq_connection from request struct (2423)
4113 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4114 - recovery timer race (2670)
4115 - mdc_close recovey bug (2532)
4116 - ptlrpc cleanup bug (2710)
4117 - mds timeout on local locks (2588)
4118 - namespace lock held during RPCs (2431)
4119 - handle interrupted sync write properly (2503)
4120 - don't try to handle a message that hasn't been replied to (2699)
4121 - client assert failure during cleanup after abort recovery (2701)
4122 - leak mdc device after failed mount (2712)
4123 - ptlrpc_check_set allows timedout requests to complete (2714)
4124 - wait for inflight reqs when ptlrpcd finishes (2710)
4125 - make sure unregistered services are removed from the srv_list
4126 - reset bulk XID's when resending them (caught by 1138 test)
4127 - unregister_bulk after timeout
4128 - fix lconf error (2694)
4129 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4130 - readahead locks pages, leaves pending causing memory pressure (2673)
4131 - increase OST request buffers to 4096 on large machines (2729)
4132 - fix up permission of existing directories in simple_mkdir (2661)
4133 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4134 - don't assign transno to errored transactions (2742)
4135 - don't delete objects on OST if given a bogus objid from MDS (2751)
4136 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4137 - if rq_no_resend, then timeout request after recovery (2432)
4138 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4139 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4140 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4141 - MDS recovery shouldn't delete objects that it has given out (2730)
4142 - if enqueue arrives after completion, don't clobber LVB (2819)
4143 - don't unlock pages twice when trigger_group_io returns error (2814)
4144 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4145 - don't write pages to disk if there was an error (1450)
4146 - don't ping imports that have recovery disabled (2676)
4147 - take buffered bytes into account when balancing socknal conn (2817)
4148 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4149 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4150 - remove little-endian swabbing of llog records (1987)
4151 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4152 - retry reposting request buffers if they fail (1191)
4153 - grow extent at grant time to avoid granting a revoked lock (2809)
4154 - lock revoke doesn't evict page if covered by a second lock (2765)
4155 - disable VM readahead to avoid reading outside lock extents (2805)
4157 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4158 - updated kernel patches for hp-2.4.20 kernel (2681)
4160 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4163 - fix truncated write corruption (2366)
4164 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4166 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4167 - ensure len(uuid) < 37 in lmc (1171)
4168 - fix ia64 OOPS in llog_test (2255)
4169 - zero end of page at obdfilter for partial page writes (2648)
4170 - don't leave stale dentries around after renames (bug 2428)
4171 - fix timeouts when evicting a client with a single lock held (2642)
4172 - set deadline for the initial HELLO message to drain (2634)
4173 - print out dotted-quad IP addresses in the socknal (2302)
4175 - additional debugging for MDS client eviction problem (2443)
4176 - fix mkfsoptions support for osts (2603, 2604)
4178 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4181 - add series for the vanilla 2.6.0 kernel
4182 - add series for the vanilla 2.4.24 kernel
4183 - add series for a cray x86/64 UL kernel drop
4184 - fix xattr patches for the vanilla 2.4.19 series
4186 - generate true UUIDs in lmc (1171)
4187 - have portals stack dumping break in UML (2466)
4188 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4189 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4190 - wake callback waiting threads on client eviction (2460)
4191 - Add --ptldebug and --subsystem to lmc (1719)
4192 - update assertion to allow safe interrupt allocation
4193 - set rq_no_resend for cancel requests (2432)
4194 - recalculate ptlrpcd timeout after resend (2494)
4195 - call vfs_rmdir when removing pending directories (2368)
4196 - fix renaming a file to itself (2429)
4197 - lmc creates a default one-stripe lov (2454)
4198 - expand procfs space to handle large clusters (2326)
4199 - increase UML stack to avoid overflow
4200 - update lconf's list of debug and subsystem masks
4201 - fix lfs find --obd (2510)
4202 - /proc tunable for disabling filter read caching (2591)
4203 - stop rpm packages from altering slapd.conf (2301)
4204 - disable nagle in the socknal under 0conf (2578)
4205 - choose mds inode size based on stripe count (2572)
4206 - fix kernel-source rpm problems (2516)
4208 - add --disable-doc to avoid pdf generation (2421)
4209 - update documentation, tests, type-os, comments
4210 - avoid format warnings on ia64
4211 - remove the TOE NAL
4212 - tiny code cleanups by removing unused fields
4214 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4217 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4218 - remove the most blatant lies from BUILDING file (2371)
4219 - change default debug level to reasonable production setting
4220 - reduce client side cache size to reduce cache flush time
4221 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4222 - make TCP zerocopy and pinger support enabled by default (2476)
4223 - sync writes completed after process exits caused crashes (2319)
4224 - maintain correct mount count on the MDS (2356)
4225 - backout 1557, because 2316 wasn't really fixed
4226 - better file I/O statistics gathering in /proc
4227 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4228 - another kernel patch to fix zero-copy TCP function export
4229 - don't take duplicate lock when processing re-sent getattr (2420)
4230 - lctl uses obd_self_export instead of creating new conn (2353)
4231 - MDS/OST recovery case which requires object creation asserted (2425)
4232 - move lfs from /usr/sbin to /usr/bin in packages
4233 - fix race between mds_client_add and mds_client_free (2417)
4234 - use kmalloc instead of slabs in portals (2430)
4235 - don't create duplicate records when a failover MDS is present (2442)
4236 - remove unnecessary mount age check (2332)
4237 - don't remove directory inodes from locks prematurely (2451)
4238 - don't break if MDS service name is the same as hostname (2103)
4239 - fix races in client write RPC generation when cache full (2482)
4241 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4244 - remove now-unused request->rq_obd (278)
4245 - if an allocation fails, print out how much memory we've used (1933)
4246 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4247 - add the "configurable stack size" patch to most series files (1256)
4248 - ability to write large log records, for 100+ OST configs (2306)
4249 - fix NULL deref when filter_prep fails (2314)
4250 - fix operator precedence error in filter_sync
4251 - dynamic allocation of socknal TX descriptors (2315)
4252 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4253 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4254 - ensure that CWARN messages in Portals always get to the syslog
4255 - __init/__exit are not for prototype decls (ldlm_init/exit)
4256 - x86-64 compile warning fixes
4257 - fix gateway LMC keyword conflict (2318)
4258 - fix MDS lock inversions in getattr/reint paths (1844)
4259 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4260 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4261 - DLM race condition prevented some lock evictions (2328)
4262 - ENOMEM detection and retry on socknal sends (2230)
4263 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4264 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4265 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4266 - better error messages when a client is rejected during recovery (1505)
4267 - avoid cancelling locks which were never granted, after failure (2330)
4268 - fix i_sem/journal inversion in mds_client_add (2333)
4269 - fix truncate/getattr lock cycle deadlock (2334)
4270 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4271 - fix two rare exit paths which could leak an l_lock() ref (2321)
4272 - fencepost error in MDS/OST orphan recovery (2226)
4273 - make log record alignment 8 bytes (1988)
4274 - lstripe now fails when requested offset > ost_count (2237)
4275 - ensure that all kernel series have a complete list.h (1607)
4276 - fix crashes in special-file operations (2316)
4277 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4279 - allow configurable automake binary, for testing new versions
4280 - small update to the lfs documentation
4282 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4284 * fix negative export reference count in fsfilt_sync (2312)
4286 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4287 * release candidate 0.9.1
4289 - orphans are moved into the PENDING directory for possible recovery
4290 - replayed opens now open by fid for orphan/rename safety (1042)
4291 - last close of an orphan inode generates a transno (683)
4292 - chdir() and mount() now pin the directory entry (1020)
4293 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4294 - discard very old requests without processing them (1502)
4295 - remove some common, well-understood CERRORs (1505)
4296 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4297 - clear "grant" flags in OST replies until OST grant code lands (1644)
4298 - fix read performance by not clobbering i_blksize on client (1598)
4299 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4300 - make LRU size a /proc tunable, clears locks when reduced (707)
4301 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4302 - server locks take references on exports now (1558)
4303 - build fixes for 2.4.20-rh trees (1663)
4304 - return an error from lov_create if all OSCs are inactive (1751)
4305 - fix import levels when a reconnect happens without a timeout (1597)
4306 - exit early from mds_open if we get a lookup error (1749)
4307 - partial page read at EOF wouldn't wait for disk before sending (1642)
4308 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4309 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4310 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4311 - write/truncate lock inversion (1639)
4312 - Don't auto-load obdclass, portals modules during cleanup (1495)
4313 - fix timestamps from jumping to "now" (1763)
4314 - extra journal assertions (1648)
4315 - add an extra multiunlink test (1771)
4316 - fix read_record/write_record API (1776)
4317 - fix leak of offset_extent, possible incorrect i_size later (1772)
4318 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4319 - Debugging for the kqswnal_get_idle_tx problems (1820)
4320 - Allow recovery to be attempted multiple times (1536)
4321 - Write out MDS last_rcvd file after it is first created (1600)
4322 - Fix tx_descriptor leak in failed transmit situations (1827)
4323 - ext3 journaling fixes for assertion failure after IO error (1871)
4324 - class_export_put() on freed export after completion AST error (1896)
4325 - Fix revalidate looping in VFS (1322)
4326 - Don't access a freed export during MDS_REINT timeout (1521)
4327 - Add open-unlink recovery support on the MDS (1673,1764)
4328 - Return an error if no MDS data was read from last_rcvd (1946)
4329 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4330 - Don't setup a disk device that doesn't match exported UUID (317)
4331 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4332 - avoid committing NULL handle in force close
4333 - local.sh is now a one-stripe LOV configuration
4334 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4335 - fix ext3 htree duplicate directory entry corruption (1516)
4336 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4337 - update to new LOV EA format (2097)
4338 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4340 2003-06-15 Phil Schwan <phil@clusterfs.com>
4343 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4344 - per-import/export recovery handling (958, 931, 959)
4345 - multiple last-rcvd slots, for serving multiple FSes (949)
4346 - connections are again shared between multiple imp/exports (963, 964)
4347 - "umount -f" would hang if any requests needed to be sent (393, 978)
4348 - avoid pinning large req buffer by copying for queued messages (989)
4349 - add "uuid" to "lctl device" command to help upcalls (991)
4350 - "open" RPCs with transnos would confuse recovery counters (1037)
4351 - do proper endian conversion of all wire messages (288, 340, 891)
4352 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4353 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4354 - fix ptlrpc_connection leak in target_handle_connect (1174)
4355 - fix import refcounting bug in OST and MDS cleanup (1134)
4356 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4357 - fix very unlikely obd_types race condition (501)
4358 - remove osc_open hack for echo_client (1187)
4359 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4360 - a failure in read_inode2 leads to deadlock (1139)
4361 - cancel ack-locks as soon as transaction is committed (1072)
4362 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4363 - make sure to commitrw after any preprw to avoid deadlock (1162)
4364 - failing to execute a file in a lustre FS would lock inode (1203)
4365 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4366 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4367 - obd_uuid2tgt would walk off the end of the list (1255)
4368 - on IA64 the osc would give portals incorrect bulk size (1258)
4369 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4370 - fix lock inversion caused by new llite matching code (1282)
4371 - limit the number of dirty pages on a client to 10MB (1286)
4372 - timed out locks were not being corrected cancelled (1289)
4373 - fix O_DIRECT above 4GB on IA-32 (1292)
4374 * major user-visible changes
4375 - fail out/fail over policy now controlled by the upcall (993)
4377 - add OBD_PING to check server availability and failure (954)
4378 - lustre messages are now sent in sending host order (288, 340, 891)
4379 - add eadatalen to MDS getattr reply (340)
4380 - OST read replies may contain second buffer, with per-page status (593)
4382 2003-03-11 Phil Schwan <phil@clusterfs.com>
4385 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4386 - failed open()s could cause deadlock; fixed (867, 869)
4387 - stop cancelling OST locks when files are closed (481)
4388 - overlapping XID spaces caused network corruption (851, 853)
4389 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4390 - setattr_raw added, to do single-RPC, server-side setattrs
4391 - lmc/lconf syntax change for OST UUIDs
4392 - fix crashy race condition between ptlrpc_free_req and osc_close
4393 - don't use request in mdc_enqueue if we hit a timeout (889)
4394 - don't set the inode i_size for regular files from the MDS (896)
4395 - handle out of order completion AST (842)
4396 - don't LBUG if a lock request times out after receiving AST (913)
4397 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4398 - if a bad lock AST arrives, send an error instead of dropping entirely
4399 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4400 - fix leak in bulk IO when only partially completed (899, 900, 926)
4401 - fix O_DIRECT for ia64 (55)
4402 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4403 - C-z after timeout could hang a process forever; fixed (977)
4405 - client-side I/O cache (678, 924, 929, 941, 970)
4407 - READPAGE and SETATTRs which don't take server-side locks get
4410 2003-02-11 Phil Schwan <phil@clusterfs.com>
4413 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4414 - Fix fsfilt_extN_readpage() to read a full page of directory
4415 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4416 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4417 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4418 - Most intent operations were removed, in favour of a new RPC mode
4419 that does a single RPC to the server and bypasses most of the VFS
4420 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4421 - Aggressively cancel local locks on DLM servers
4422 - mds_reint_unlink sends EA to the client if it's the last nlink.
4423 client uses that EA to unlink OST objects.
4424 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4425 - recursive symlinks were fixed (439)
4426 - fixed NULL deref in DEBUG_REQ
4427 - filter_update_lastobjid no longer calls sync, which annoyed extN
4428 - fixed multi-client small-writes to a single file problem (445)
4429 - fixed mtime updates during file writes (607)
4430 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4431 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4432 - fixed Linux OST size reporting problem (444, 656)
4433 - OST now updates object mtime with writes or setattr (607, 619)
4434 - client verifies file size before zeroing page past EOF (445)
4435 - OST now writes last allocated objid to disk with allocation (108)
4436 - LOV on echo now works (409)
4438 - mds_reint_unlink sends a new buffer, with the EA included. this
4439 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4440 set if a regular file was being unlinked, and it was the last link
4441 - use PtlGet from the target for bulk writes (315)
4442 - OST now updates object mtime with writes or setattr (607, 619)
4443 - LDLM now has a grant-time callback to revalidate locked items, if
4445 - Many MDS operations were reorganized to combat race conditions
4447 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4448 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4450 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4453 - Fully reactivate OST imports after reconnection (512, others)
4454 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4455 - More graceful error handling for truncating on dead OST (515)
4456 - Don't error out unless we're actually accessing dead stripes (474)
4457 - Fix garbage sizes when stripes are missing (410)
4458 - LRU counters were broken, causing constant lock purge (433, 432)
4459 - garbage on read from stripes with failed OSTs (441)
4460 - mark OSCs as active before reconnecting during recovery (438)
4461 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4462 - lfind did not preserve OST order in output (443)
4463 - symlinks cause hung clients, incorrect data (439)
4464 - stop dereferencing request after dropping refcount (457)
4465 - don't LASSERT(spin_is_locked) on non-SMP (455)
4466 - fixes for many rename() bugs
4467 - fstat didn't correctly synchronize attributes (399)
4468 - server must handle lock cancellation during blocking AST prep (487)
4469 - bulk descriptors were free()d too soon (511)
4470 - fix paths in lconf, which would load incorrect modules (451, 507)
4471 - fix confusing lconf 'host not found' error message (386)
4472 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4473 - fix race condition in mdc_blocking_ast() for inode access (526)
4474 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4475 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4476 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4477 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4478 - fix lstripe in several ways (kernel side) (527)
4479 - fix request leak in ldlm_cli_enqueue (262)
4480 - incorrect OSC was marked inactive after OST failure
4481 - call mds_fs_cleanup before unmounting filesystem (524)
4482 - fix races between taking ns_lock and ldlm_lock_change_resource
4483 - fix races updating LOV export open file list
4484 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4485 - fix recovery NULL deref in ldlm_cli_cancel_unused
4486 - fix some DLM races by using new hash table for lock handles (419)
4487 - permit the client to specify desired inodes, at replay
4488 - duplicate requests when we queue them for replay reintegration
4489 - fix last_rcvd offset calculation
4490 - sync after each recovered transaction, so we always make progress
4491 - never, not always, ERESTART requests without transnos
4492 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4494 - skip replay if the MDS says that the client is already connected
4495 - don't check for a recovery-enabled export to match lctl's UUID
4496 - don't INC_USE_COUNT for phantom exports
4497 - don't crash when cleaning up phantom exports (567)
4498 - don't double-finish or set replay data for errored mdc_open requests
4499 - abort requests when they time out, so we don't get old replies
4500 - send/receive replies for AST messages again
4501 - if the client says that it doesn't have the lock, cancel it on the
4503 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4504 mark it as destroyed, it will all work out when decref is called
4505 - fix module use counts (22, 581)
4507 - ASTs now expect a reply (server cancels lock on error reply)
4509 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4512 - fix many simultaneous client startup (392)
4513 - fix dentry->d_it clobbering
4514 - credentials weren't being shipped for readdir/getattr operations
4515 - remove invalid assertions triggered during some concurrent MD
4517 - proper Lustre versions added (336, 389)
4518 - fix memory leak for create error case (398)
4519 - fix LOV locking bug that would get cli/srv out of sync
4520 - fix echo client over LOV (409)
4521 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4522 - fix double-O_EXCL intent crash (424)
4523 - avoid sending multiple lock CANCELs (352)
4525 - MDS can do multi-client recovery (modulo bugs in new code)
4527 - many updates, edits, cleanups
4529 2002-11-18 Phil Schwan <phil@clusterfs.com>
4532 - fix null d_it dereference (346)
4533 - fix full OST/dbench hang (333)
4534 - fix permission problem with file removal (286)
4535 - fix removal of OSCs from LOV when they fail
4536 - fix NULL deref during bulk timeout (214)
4537 - fix problems related to multiple filesystems on one MDS (241)
4538 - fixed serious subtle metadata locking bugs
4539 - free locks on clients when inodes are removed due to memory
4541 - fix inode pointer in lock data (285)
4542 - partial support for multiple MDS on a single host (241)
4543 - data locks weren't cancelled at clear_inode time (290, 311)
4544 - intent locks could lead to unbounded lock growth (205)
4545 - added a maximum lock count, an LRU list, and a flusher
4546 - fix multiple rename (365)
4547 - properly abstracted the echo client
4548 - OSC locked 1 byte too many; fixed
4549 - rewrote brw callback code:
4550 - fixed recovery bugs related to LOVs (306)
4551 - fixed too-many-pages-in-one-write crash (191)
4552 - fixed (again) crash in sync_io_timeout (214)
4553 - probably fixed callback-related race (385)
4555 - Add capability to MDS protocol
4556 - LDLM cancellations and callbacks on different portals
4558 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4561 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4562 - defer open object destruction to close time (601981, 138)
4563 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4564 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4565 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4566 - don't call obd_setattr in ll_file_release for destroyed objects
4567 * protocol change to lustre_msg: move |version| and add |flags|
4568 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4569 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4570 * added replay of create, unlink, link and rename operations during
4571 MDS failover; recovery should be much more robust now
4572 * remove failed OSCs from LOVs (only lov_create uses this so far)
4573 * the lustre-HOWTO was brought (more) up to date (582544)
4575 2002-10-23 Phil Schwan <phil@clusterfs.com>
4578 - in-use dentries weren't being reused properly (617851)
4579 - prevent multiple LDLM setup (599178)
4580 - fix LOV size calculations for truncate (617853)
4581 - fix client handling of MDS intent errors (POSIX)
4582 - fix permission bug in lovstripe.c test (624321)
4583 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4584 - truncate past end of file could corrupt data
4585 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4586 - a race in recovery could return ETIMEDOUT to apps (623947)
4587 - building outside the source directory was fixed
4588 * the lustre-HOWTO was brought (more) up to date (582544)
4589 * major progress was made on recovery functionality
4591 2002-10-10 Phil Schwan <phil@clusterfs.com>
4594 - recovery deadlock fix
4595 - rm -rf causes LBUG fix (617817)
4596 - file open by multiple tasks fix (618962)
4597 - directory permissions bugs (602707 and 620007)
4598 - journal_stop fixed with locking (611313)
4599 - O_APPEND failures resolved (618273, perhaps 614459)
4600 - lconf PATH fix (619770)
4601 - IA64 build fix (621450)
4602 - RPC buffer sizes scale with amount of memory
4604 2002-10-01 Phil Schwan <phil@clusterfs.com>
4607 - locks would be cancelled without throwing away data pages,
4608 resulting in inconsistent data (605627)
4609 - inode attributes were not always being refreshed (605627, 612449)
4610 - lconf now continues to cleanup after lctl reports an error
4611 - MDS now enforces user permissions (602707)
4612 - lprocfs cleanup fixed, but not yet enabled (614157)
4613 - fixed infinite server hang, should a client not respond to an AST
4614 - avoid going into recovery if user calls readlink() with a buffer
4615 that's too small (613941)
4616 - AST RPCs no longer require replies (614867) -- this may be changed
4617 - don't crash server if client sends an IOV that's too big (611336)
4618 - fixed lock conversion deadlock (611892)
4619 - fixed the following of symlinks (614622)
4620 * recovery: the server can remove locks from a client that dies, other
4621 clients can make progress
4622 * more extN patch fixes
4623 * compile-time configurable ptlrpc buffer allocations
4625 - collaborative read cache document
4626 - Lustre Lite Performance CDR document-in-progress
4628 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4631 - fix typo in patch-2.4.18
4633 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4636 - clear ptlrpc request each time in handle_incoming_request()
4637 - unlink of files now destroys the object on the OST
4639 2002-09-19 Peter Braam <braam@clusterfs.com>
4641 * add hard link support
4642 * change obdfile creation method
4643 * kernel patch changed
4645 2002-09-19 Peter Braam <braam@clusterfs.com>
4648 - stack overflow bug in extN fixed
4650 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4652 * documentation updates
4653 - add man pages for config tools
4654 - update tests/README to describe testing with new config tools
4655 - finish metadata API descriptions
4656 * bug fixes and cleanups
4657 - statfs workaround for 16TB limit
4658 - LOV stripe allocation improved, can stripe on subset of OSTs
4659 - LOV file size/IO offset was wrong for files > 4GB in size
4660 - object EA data was being dropped, caused files to be unreadable
4661 - memory overflow with non-LOV OST caused memory corruption
4662 - fixed regression tests to work with new config tools, obdfilter
4663 - fixed bug when directory size became larger than 1 block
4664 - fixed bug (for single client case) when PWD was deleted
4665 - invalidate local directory pages when doing intent-based ops
4666 - avoid LDLM oops when lock callback contained bad data
4668 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4670 * documentation updates
4671 * bug fixes and cleanups
4672 - configuration tools
4675 - 64-bit compile warnings
4676 - 64-bit internal statfs data
4678 * test_brw on persistent OST devices
4680 * lprocfs (disabled)
4682 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4684 * documentation updates
4685 * bug fixes and cleanups
4686 * configuration tools
4688 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4691 * many small fixes to 0_5_4
4692 * io/network handling
4693 * thinkos in MDS operations
4695 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4698 * crucial basic fixes to 0.5.3
4699 * IOR, Iozone work over Elan
4702 2002-08-07 Phil Schwan <phil@clusterfs.com>
4703 * version 0_5_3, our first alpha
4704 * we use the new Portals iovs
4705 * documentation updates
4706 * bug fixes and cleanups
4707 * small changes in the DLM wire protocol
4709 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4710 * version 0_5_1 with some initial stability,
4711 * locking on MD and file I/O.
4712 * documentation updates
4713 * several bug fixes since 0.5.0
4714 * small changes in wire protocol
4716 2002-07-18 Phil Schwan <phil@clusterfs.com>
4718 * delivered as Lustre Light Alpha
4719 * fixed a crash after handling invalid MDS requests
4720 * fixed directory pages for architectures with non-4k pages sizes
4722 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4723 * release version v0_4_4
4724 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4726 * Fixes a number of interruption problems with OST operations.
4727 * Update documentation for portals header changes
4728 * Move all wire protocol structs/defines to lustre_idl.h
4729 * Fixes symlink length bug.
4730 * Add tcpdump to repository.
4732 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4733 * release version v0_4_3
4734 * Fixes statfs for inodes on extN.
4735 * Fixes bug in runtests which would delete /etc/hosts.
4736 * Use 64-bit object IDs wherever possible (not into VFS though)
4737 Remove ost_get_info, which is unused by lustre, and out of date.
4739 2002-07-03 Peter Braam <braam@clusterfs.com>
4740 * release version v0_4_2 Fixes a lookup error (type not passed)
4741 * move forward to head of Portals
4742 * move forward to latest Lustre kernel
4744 2002-06-25 Peter Braam <braam@clusterfs.com>
4745 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob