1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.14.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.7-sun3
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
16 Frequency : always with o2ib 1.3 and sles10
18 Description: fix build with SLES10 and o2ib v3.
19 Details : sles10 uses diffrent name for Module.symver file but configure
20 assume this file has same name on RHEL/SLES/vanila kernels.
24 Frequency : after recovery
25 Description: precreate to many object's after del orphan.
26 Details : del orphan st in oscc last_id == next_id and this triger growing
27 count of precreated objects. Set flag LOW to skip increase count
28 of precreated objects.
32 Frequency : rare, on clear nid stats
33 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
34 Details : when clean nid stats sometimes try destroy live entry,
35 and this produce panic in free.
39 Description: Stack overflow during MDS log replay
40 ease stack pressure by using a thread dealing llog_process.
44 Description: wait until IO finished before start new when do lock cancel.
45 Details : VM protocol want old IO finished before start new, in this case
46 need wait until PG_writeback is cleared until check dirty flag and
47 call writepages in lock cancel callback.
49 Severity : enhancement
51 Description: using special macro for print time and cleanup in includes.
55 Description: mds_mfd_close() ASSERTION(rc == 0)
56 Details : In mds_mfd_close(), we need protect inode's writecount change
57 within its orphan write semaphore to prevent possible races.
61 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
66 Description: don't hit live lock with umount ost.
67 Details : shrink_dcache_parent can be in long loop with destroy dentries,
68 use shrink_dcache_sb instead.
72 Description: don't panic with use echo client
73 Details : echo client pass NULL as client nid pointer and this produce null
78 Description: fix build on ppc32
79 Details : compile code with -m64 flag produce wrong object file for ppc32.
83 Description: add message levels for liblustreapi
87 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
88 Details : Move the 'good_osts' check before the 'total_bavail' check. This
89 will result in an -EAGAIN and in the exit call path we call
90 alloc_rr() which will with increasing aggressiveness attempt to
91 aquire precreated objects on the minimum number of required OSCs.
95 Description: Use old size assignment to avoid deadlock
96 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
97 the scheduling while holding a spinlock. We do not need locking
98 for size in ll_update_inode() because size is only updated from
99 the MDS for directories or files without objects, so there is no
100 other place to do the update, and concurrent access to such inodes
101 are protected by the inode lock.
105 Description: resolve "_IOWR redefined" build error on SLES10
109 Description: dump the memory debugging after all modules are unloaded to
110 suppress false negative in conf_sanity test 39
112 Severity : enhancement
114 Description: build kernel-ib packages for OFED 1.3 in our release cycle
119 Description: fix SLES kernel versioning
120 Details : the kernel version for our SLES 10 kernel did not include a "-"
121 before the "smp" at the end. while this was not a problem in
122 general, it did mean that software trying to use the kernel
123 version to try to detect a vendor specific kernel would fail.
124 this was most evident by the OFED build scripts.
128 Description: Don't update lov_desc members until making sure they are valid
129 Details : When updating lov_desc members via proc fs, need fix their
130 validities before doing the real update.
134 Description: don't put request into delay list while invalidate in flight.
135 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
136 import in flight. this produce timeout for invalidate and sometimes
137 can cause stale data.
141 Frequency : on ppc only
142 Description: not convert ost objects for directory because it's not exist.
143 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
147 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
149 Severity : enhancement
151 Description: Update to support 2.6.22.14 vanilla kernel.
155 Frequency : rare, on recovery
156 Description: read procfs can produce deadlock in some situation
157 Details : Holding lprocfs lock which send rpc can produce block for destroy
158 obd objects and this also block reconnect with -EALREADY. This isn't
159 fix all lprocfs bugs - but make it rare.
161 Severity : enhancement
163 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
166 Frequency : frequent on X2 node
168 Description: mdc_set_open_replay_data LBUG
169 Details : Set replay data for requests that are eligible for replay.
173 Description: lustre_mgs: operation 101 on unconnected MGS
174 Details : When MGC is disconnected from MGS long enough, MGS will evict the
175 MGC, and late on MGC cannot successfully connect to MGS and a lot
176 of the error messages complaining that MGS is not connected.
180 Frequency : on network error
181 Description: panic with double free request if network error
182 Details : mdc_finish_enqueue is finish request if any network error ocuring,
183 but it's true only for synchronus enqueue, for async enqueue
184 (via ptlrpcd) this incorrect and ptlrpcd want finish request
187 Severity : enhancement
189 Description: client-side metadata stat-ahead during readdir(directory readahead)
190 Details : perform client-side metadata stat-ahead when the client detects
191 readdir and sequential stat of dir entries therein
194 Frequency : on start mds
196 Description: Implement get_info(last_id) in obdfilter.
199 Frequency : occasional
201 Description: Correctly check stale fid, not start epoch if ost not support SOM
202 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
203 overwrited with new generated fid, but mds can anwer with one of these
204 two fids and both is not stale. setattr incorectly start epoch and
205 assume will be called done_writeting, but without SOM done_writing
209 Frequency : rare, depends on device drivers and load
211 Description: MDS or OSS nodes crash due to stack overflow
212 Details : Code changes in 1.8.0 increased the stack usage of some functions.
213 In some cases, in conjunction with device drivers that use a lot
214 of stack the MDS (or possibly OSS) service threads could overflow
215 the stack. One change which was identified to consume additional
216 stack has been reworked to avoid the extra stack usage.
219 Frequency : occasional
221 Description: Do not fail import if osc_interpret_create gets -EAGAIN
222 Details : If osc_interpret_create got -EAGAIN it immediately exits and
223 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
224 oscc_has_objects and see OSC has no objests and call
225 oscc_internal_create to resend create request.
227 Severity : enhancement
229 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
231 Severity : enhancement
233 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
236 Frequency : very rare
238 Description: Fix replay if there is an un-replied request and open
239 Details : In some cases, older replay request will revert the
240 mcd->mcd_last_xid on MDS which is used to record the client's
243 Severity : enhancement
245 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
247 Severity : enhancement
249 Description: Add rhel5 support to HEAD.
251 Serverity : enhancement
253 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
258 Description: Don't allow skipping OSTs if index has been specified.
259 Details : Don't allow skipping OSTs if index has been specified, make locking
260 in internal create lots better.
264 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
265 Details : access to bitfield in structure is always rounded to long
266 and this produce problem with not atomic change any bit.
270 Description: Lustre make rpms failed.
271 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
272 builds Lustre rpms from tarball.
275 Frequency : rare on shutdown ost
277 Description: If llog cancel was not send before clean_exports phase, this can
278 produce deadlock in llog code.
279 Details : If llog thread has last reference to obd and call class_import_put
280 this produce deadlock because llog_cleanup_commit_master wait when
281 last llog_commit_thread exited, but this never success because was
282 called from llog_commit_thread.
286 Description: allow userland application know is lost one of stripes.
287 Details : fill lvb_blocks with error code on ost and return it to
288 application if error flag found.
292 Description: NULL lov_tgts causing MDS oops
293 Details : more safe checks for NULL lov_tgts for avoid oops.
295 Severity : enhancement
297 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
301 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
302 Details : If an OST is down the MDS will hang indefinitely in
303 obd_statfs_rqset() waiting for the statfs data. While for
304 MDS QOS usage of statfs, it should not stuck in waiting.
306 Severity : enhancement
308 Description: remote_acl support
309 Details : Support ACL-based permission check for remote user.
310 Support setfacl/getfacl for remote user with the utils
311 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
312 the system "{s,g}etfacl" utils.
314 Severity : enhancement
316 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
318 Severity : enhancement
320 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
324 Description: make mgs_setparam() handle fsname containing dash
325 Details : fsname containing a dash does not work with lctl conf_param
327 Severity : enhancement
329 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
331 Severity : enhancement
333 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
337 Frequency : rare, at shutdown
338 Description: access already free / zero obd_namespace.
339 Details : if client_disconnect_export was called without force flag set,
340 and exist connect request in flight, this can produce access to
341 NULL pointer (or already free pointer) when connect_interpret
342 store ocd flags in obd_namespace.
346 Frequency : only at startup
347 Description: not alloc memory with spinlock held.
348 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
349 if any spinlock held.
351 Severity : enhancement
353 Description: make lustre randomly fail allocating memory
354 Details : Make lustre randomly failed allocating memory for testing purpose.
356 Severity : enhancement
358 Description: lost problems with lov objid file
359 Details : Fixes some scability and access to not inited memory problems
360 in work with lov objdid file.
365 Description: lfs find does not continue on file error
366 Details : Continue other files processing when a file/dir is absent.
370 Description: Inconsistent usage of lustre_pack_reply()
371 Details : Standardize the usage of lustre_pack_reply() such that it
372 always generate a CERROR on failure.
375 Frequency : occasional
377 Description: MDS hang or stay in waiting lock
378 Details : If client receive lock with CBPENDING flag ldlm need send lock
379 cancel as separate rpc, to avoid situation when cancel request
380 can't processed due all i/o threads stay in wait lock.
384 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
388 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
392 Description: Update to SLES9 kernel-2.6.5-7.287.3.
396 Description: Update to RHEL4 latest kernel.
398 Severity : enhancement
400 Description: Build SLES10 patchless client fails
401 Details : The configure was broken by run ./configure with
402 --with-linux-obj=.... argument for patchless client. When the
403 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
404 can't search header adequately. Use absolute path such as
405 -I($LINUX)/include instead.
409 Description: interrupt oig_wait produce painc on resend.
410 Details : brw_redo_request can be used for resend requests from ptlrpcd and
411 private set, and this produce situation when rq_ptlrpcd_data not
412 copyed to new allocated request and triggered LBUG on assert
413 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
414 ptlrpcd set if request is changed and can be safety changed to use
419 Description: LASSERT_{REQ,REP}SWAB macros are buggy
420 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
421 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
422 these macros and replace them with their difinitions instead.
426 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
427 Details : Update which_patch & target file for SLES10 latest kernel.
430 Description: Remove client patches from SLES 10 kernel.
431 Details : This causes SLES 10 clients to behave as patchless clients
432 even on a Lustre-patched (server) kernel.
434 Severity : enhancement
436 Description: self-adjustable client's lru lists
437 Details : use adaptive algorithm for managing client cached locks lru
438 lists according to current server load, other client's work
439 pattern, memory activities, etc. Both, server and client
440 side namespaces provide number of proc tunables for controlling
443 Severity : enhancement
445 Description: light-weight GSS support
446 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
447 and reduce performance overhead.
449 Severity : enhancement
451 Description: Linux keyring support
452 Details : Support using service of Linux keyring for Lustre GSS internal
453 context refresh/cache mechanism.
457 Description: Fix errors in lfs documentation
458 Details : Fixes man pages
462 Description: don't use GFP_* in generic Lustre code.
463 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
467 Description: obdclass is limited by single OBD_ALLOC(idarray)
468 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
472 Description: warnings with build patchless client with vanila 2.6.19 and up
473 Details : change old ctl_table style and replace ctl_table/ctl_table_header
474 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
478 Frequency : Only for SLES
479 Description: sanity_quota fail test_1
480 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
481 file, which makes TSTID[2] inunique.
485 Frequency : Only for SLES9
486 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
487 Details : Update target/ChangeLog/which_patch .
491 Frequency : Only for test_10 in sanity-quota.sh
492 Description: a bug in quota test code
493 Details : checking a test flag in a wrong place
497 Frequency : Only for test_13 in sanity-quota.sh
498 Description: a bug in quota test script
499 Details : shouldn't check overfull allocated quota
503 Description: jbd statistics
504 Details : Port older jbd statistics patch for sles10
506 Severity : enhancement
508 Description: Add printing inode info into message about error in writepage.
513 Description: reply_lock_interpret crash due to race with it and lock cancel.
514 Details : Do not replay locks that are being cancelled. Do not reference
515 locks by their address during replay, just by their handle.
519 Frequency : When flocks are used.
520 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
521 Details : Flock locks might destroy just granted lock if it could be merged
522 with another existing flock, this is done in completion handler,
523 so teach ldlm_cli_enquque_fini that this is a valid case for
529 Description: Oops in read and write path when failing to allocate lock.
530 Details : Check if lock allocation failed and return error back.
534 Description: lstripe command fails for valid OST index
535 Details : The stripe offset is compared to lov->desc.ld_tgt_count
536 instead of lov->desc.ld_active_tgt_count.
540 Description: sanity.sh failed test 103
541 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
546 Description: df doesn't work properly if diskfs blocksize != 4K
547 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
551 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
552 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
556 Description: obd_health_check_timeout too short
557 Details : set obd_health_check_timeout as 1.5x of obd_timeout
561 Description: llapi_file_create() does not allow some changes
562 Details : add llapi_file_open() that allows specifying the mode and
563 open flags, and also returns an open file handle.
567 Description: merge and cleanup kernel patches.
568 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
572 Description: Add journal checksum support.(Kernel part)
573 Details : The journal checksum feature adds two new flags i.e
574 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
575 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
576 indicates that the commit block contains the checksum for
577 the blocks described by the descriptor blocks. Now commit
578 record can be sent to disk without waiting for descriptor
579 blocks to be written to disk. This behavior is controlled
580 using JBD2_FEATURE_ASYNC_COMMIT flag.
584 Description: OSS needs mutliple precreate threads
585 Details : Add ability to start more than one create thread per OSS.
589 Description: Sanity.sh test_65k failed: setstripe should have succeeded
590 Details : execute lfs setstripe on client
594 Description: mds_obd_create error creating tmp object
595 Details : When the user sets quota on root, llog will be affected and can't
596 create files and write files.
599 Frequency : Always on ia64 patchless client, and possibly others.
601 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
602 Details : This allows the patchless client to be loaded on architectures
607 Description: RedHat Update kernel for RHEL5
608 Details : Add the kernel config file for RHEL5.
612 Description: RedHat Update kernel for RHEL5
613 Details : Modify the kernel config file more closer RHEL5.
617 Description: Build failure against Centos5 (RHEL5)
618 Details : Define PAGE_SIZE when it isn't present.
622 Description: test_7 on sanity-quota.sh can't be used on separate machines
623 Details : add facet to handle it
627 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
628 Details : with NFS, the anon dentry's parent was set to itself in
629 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
630 corresponding dentry other than use rec->ur_name.
632 Severity : enhancement
634 Description: lfs setstripe enhancement
635 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
639 Description: enable data checksumming by default
640 Details : enable checksum by default, allow --disable-checksum
641 configure option and "-o nochecksum" mount option. Checksums
642 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
646 Description: lustre support for RHEL5
647 Details : Add support for RHEL5.
651 Description: Client eviction due to failover config
652 Details : after a connection loss, the lustre client should attempt to
653 reconnect to the last active server first before trying the
654 other potential connections.
658 Description: when mds and osts use different quota unit(32bit and 64bit),
659 quota will be released repeatly.
660 Details : void sending multiple quota reqs to mds, which will keep the
661 status between the reqs.
665 Description: osts not allocated evenly to files
666 Details : change the condition to increase offset_idx
669 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
671 Description: Data corruption for OSTs that are formatted larger than 2TB
673 Details : When generating the bio request for lustre file writes the
674 sector number would overflow a temporary variable before being
675 used for the IO. The data reads correctly from Lustre (which
676 will overflow in a similar manner) but other file data or
677 filesystem metadata may be corrupted in some cases.
681 Description: Tune the kernel for good SCSI performance.
682 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
683 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
688 Description: rewrite ext2-derived code in obdclass/uuid.c
689 Details : rewrite inherited code (uuid parsing code from ext2 utils)
690 from scratch preserving functionality.
694 Description: "lfs find -obd UUID" prints directories
695 Details : "lfs find -obd UUID" will return all directory names instead
696 of just file names. It is incorrect because the directories
697 do not reside on the OSTs.
701 Description: use i_size_read and i_size_write in 2.6 port
702 Details : replace inode->i_size access with i_size_read/write()
705 Frequency : when removing large files
707 Description: scheduling issue during removal of large Lustre files
708 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
709 It causes scheduling issues when removing large files (17TB in the
713 Frequency : only with liblustre clients on XT3
715 Description: evictions taking too long
716 Details : allow llrd to evict clients directly on OSTs
719 Frequency : only on ppc
721 Description: /proc/fs/lustre/devices broken on ppc
722 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
723 We should make sure the type of 'index' is loff_t and then cast
724 explicitly as needed below. Do not assign an explicitly cast
728 Frequency : only for rhel5
730 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
731 Details : Modify the target file & which_kernel.
733 Severity : enhancement
735 Description: omit set fsid for export NFS
736 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
737 as FS_REQUIRES_DEV for avoid problems with generate fsid.
741 Frequency : Always, for kernels after 2.6.16
742 Description: Fix warning idr_remove called for id=.. which is not allocated.
743 Details : Last kernels save old s_dev before kill super and not allow
744 to restore from callback - restore it before call kill_anon_super.
748 Description: buffer overruns could theoretically occur
749 Details : llapi_semantic_traverse() modifies the "path" argument by
750 appending values to the end of the origin string, and a
751 overrun may occur. Adding buffer overrun check in liblustreapi.
755 Description: Fix error on 'ls .' at the top of the Lustre mount.
756 Details : Don't revalidate dentry if it is a root dentry.
760 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
761 Details : Modify vm-tunables-rhel4.patch.
765 Description: Kernel config for 2.6.18-vanilla.
766 Details : Modify targets/2.6-vanilla.target.in.
767 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
768 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
769 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
770 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
774 Description: improve handling recoverable errors
775 Details : if request processig with error which can be recoverable on server
776 request should be resend, otherwise page released from cache and
781 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
782 Details : Modify target file & which_patch.
783 A flaw was found in the IA32 system call emulation provided
784 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
785 value could be stored in the %RAX register, which could trigger an
786 out-of-bounds system call table access. An untrusted local user
787 could exploit this flaw to run code in the kernel
788 (ie a root privilege escalation). (CVE-2007-4573).
792 Description: O_DIRECT bypasses client statistics.
793 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
794 read_ahead_stats were updated. copied the stats section from
795 osc_send_oap_rpc() into async_internal().
799 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
803 Description: change order of libsysio includes
804 Details : '#include sysio.h' should always come before '#include xtio.h'
808 Description: handle "serious error: objid * already exists" more gracefully
809 Details : If LAST_ID value on disk is smaller than the objects existing in
810 the O/0/d* directories, it indicates disk corruption and causes an
811 LBUG(). If the object is 0-length, then we should use the existing
812 object. This will help to avoid a full fsck in most cases.
814 Severity : enhancement
816 Description: adapt the lustre_config script to support the upgrade case
817 Details : Add "-u" option for lustre_config script to support upgrading 1.4
818 server targets to 1.6 in parallel.
822 Description: To avoid grant space > avaible space when the disk is almost
823 full. Without this patch you might see the error "grant XXXX >
824 available" or some LBUG about grant, when the disk is almost
826 Details : In filter_check_grant, for non_grant cache write, we should
827 check the left space by if (*left > ungranted + bytes), instead
828 of (*left > ungranted), because only we are sure the left space
829 is enough for another "bytes", then the ungrant space should be
830 increase. In client, we should update cl_avail_grant only there
831 is OBD_MD_FLGRANT in the reply.
835 Description: Update RHEL 4 kernel to fix local root privilege escalation.
836 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
837 described in CVE-2007-4573. This problem could allow untrusted
838 local users to gain root access.
841 Frequency : when using O_DIRECT and quotas
843 Description: Incorrect file ownership on O_DIRECT output files
844 Details : block usage reported by 'lfs quota' does not take into account
845 files that have been written with O_DIRECT.
850 Description: touch file failed when fs is not full
851 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
852 otherwise we can get ENOSP while fs is not full.
856 Description: parallel lock callbacks
857 Details : Instead of sending blocking and completion callbacks as separated
858 requests, adding them to a set and sending in parallel.
861 Frequency : only for Cray XT3
862 Bugzilla : 12829/13455
863 Description: Changing primary group doesn't change the group lustre assigns to
865 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
866 overridden with the primary group provided by the group upcall,
867 whereas the supplied fsgid can be trusted if it is in the list of
868 supplementary groups returned by the group upcall.
870 Severity : enhancement
872 Description: Allow masking D_WARNING, D_ERROR messages from console
873 Details : Console messages can now be disabled via lnet.printk.
877 Description: User code with malformed file open parameter crashes client node
878 Details : Before packing join_file req, all the related reference should be
879 checked carefully in case some malformed flags cause fake
880 join_file req on client.
884 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
885 Details : ldlm_completion_ast() assumes that a lock is granted when the req
886 mode is equal to the granted mode. However, it should also check
887 that LDLM_FL_CP_REQD is not set.
891 Description: Heavy nfs access might result in deadlocks
892 Details : After ELC code landed, it is now improper to enqueue any mds
893 locks under och_sem, because enqueue might want to decide to
894 cancel open locks for same inode we are holding och_sem for.
898 Description: Client eviction while running blogbench
899 Details : A lot of unlink operations with concurrent I/O can lead to a
900 deadlock causing evictions. To address the problem, the number of
901 oustanding OST_DESTROY requests is now throttled to
902 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
903 callbacks are processed in priority.
907 Description: enable ACLs on MDS by default
908 Details : ACLs must be enabled on MDS by default.
911 Frequency : PPC/PPC64 only
913 Description: conflicts between asm-ppc64/types.h and lustre_types.h
914 Details : fix duplicated definitions between asm-ppc64/types.h and
915 lustre_types.h on PPC.
918 Frequency : PPC/PPC64 only
920 Description: asm-ppc/segment.h does not exist
921 Details : fix compile issue on PPC.
925 Description: better handle error messages in extents code
928 Frequency : RHEL4 only
930 Description: mkfs is very slow on IA64/RHEL4
931 Details : A performance regression has been discovered in the MPT Fusion
932 driver between versions 3.02.73rh and 3.02.99.00rh. As a
933 consequence, we have downgraded the MPT Fusion driver in the RHEL4
934 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
936 Severity : enhancement
938 Description: SNMP support enhancement
939 Details : Adding total number of sampled request for an MDS node in snmp
942 Severity : enhancement
944 Description: Optimize ldlm waiting list processing for PR extent locks
945 Details : When processing waiting list for read extent lock and meeting read
946 lock that is same or wider to it that is not contended, skip
947 processing rest of the list and immediatelly return current
948 status of conflictness, since we are guaranteed there are no
949 conflicting locks in the rest of the list.
953 Description: Time out and refuse to reconnect
954 Details : When the failover node is the primary node, it is possible
955 to have two identical connections in imp_conn_list. We must
956 compare not conn's pointers but NIDs, otherwise we can defeat
957 connection throttling.
961 Description: port llog fixes from b1_6 into HEAD
962 Details : Port llog reference couting and some llog cleanups from b1_6
963 (bug 10800) into HEAD, for protect from panic and access to already
964 free llog structures.
968 Description: Detect stride IO mode in read-ahead
969 Details : When a client does stride read, read-ahead should detect that and
970 read-ahead pages according to the detected stride pattern.
974 Description: data checksumming impacts single node performance
975 Details : add support for several checksum algorithm. Currently, only CRC32
976 and Adler-32 are supported. The checksum type can be changed on
977 the fly via /proc/fs/lustre/osc/*/checksum_type.
981 Description: use adler32 for page checksums
982 Details : when available, use the Adler-32 algorithm instead of CRC32 for
987 Description: build for x2 fails
988 Details : fix compile issue on Cray systems.
992 Description: Properly match for duplicate locks
993 Details : Due to different lock order from skiplists code, we need to
994 traverse entire list for now
997 Frequency : only on PPC/SLES10
999 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1000 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1001 be defined. Add a hack in mkfs_lustre.c to work around this compile
1006 Description: LASSERT on MDS when client holding flock lock dies
1007 Details : ldlm pool logic depends on number of granted locks equal to
1008 number of released locks which is not true for flock locks, so
1009 just exclude such locks from consideration.
1013 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1014 Details : Use fsfilt_sync() for both the whole filesystem sync and
1015 individual file sync to eliminate dangerous inode locking
1016 with I_LOCK that can lead to a deadlock.
1020 Description: performance in 1.6.3
1021 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1022 sure that 1MB requests are not fragmented by the block layer.
1024 Severity : enhancement
1026 Description: organize the server-side client stats on per-nid basis
1027 Details : Change the structure of stats under obdfilter and mds to
1035 The "uuid"s file would list the uuids of _active_ exports.
1036 And the clear entry is to clear all stats and stale nids.
1038 Severity : enhancement
1040 Description: eliminate client locks in face of contention
1041 Details : file contention detection and lockless i/o implementation
1042 for contended files.
1046 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1049 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1051 Description: MDS may lose file striping (and hence file data) in some cases
1052 Details : If there are additional extended attributes stored on the MDS,
1053 in particular ACLs, SELinux, or user attributes (if user_xattr
1054 is specified for the client mount options) then there is a risk
1055 of attribute loss. Additionally, the Lustre file striping
1056 needs to be larger than default (e.g. striped over all OSTs),
1057 and an additional attribute must be stored initially in the
1058 inode and then increase in size enough to be moved to the
1059 external attribute block (e.g. ACL growing in size) for file
1064 Description: skiplist implementation simplification
1065 Details : skiplists are used to group compatible locks on granted list
1066 that was implemented as tracking first and last lock of each lock group
1067 the patch changes that to using doubly linked lists
1069 --------------------------------------------------------------------------------
1071 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1073 * Support for kernels:
1074 2.6.5-7.283 (SLES 9),
1075 2.6.9-55.EL (RHEL 4),
1076 2.6.16.46-0.14 (SLES 10),
1077 2.6.18.8 vanilla (kernel.org)
1078 * Client support for unpatched kernels:
1079 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1080 2.6.16 - 2.6.22 vanilla (kernel.org)
1081 * Due to recently discovered recovery problems, we do not recommend
1082 using patchless RHEL 4 clients with this or any earlier release.
1083 * Recommended e2fsprogs version: 1.39.cfs8
1084 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1085 * Starting with this release, the ldiskfs backing filesystem required
1086 by Lustre is now in its own package, lustre-ldiskfs. This package
1087 should be installed. It is versioned separately from Lustre and
1088 may be released separately in future.
1092 Description: block reactivating mgc import until all deactivates complete
1093 Details : Fix race when failing back MDT/MGS to itself (testing)
1096 Frequency : at statup only
1098 Description: mds_lov_synchronize race leads to various problems
1099 Details : simultaneous MDT->OST connections at startup can cause the
1100 sync to abort, leaving the OSC in a bad state.
1102 Severity : enhancement
1104 Description: add optional extra BUILD_VERSION info
1105 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1106 to override the lustre version.
1109 Frequency : 2.6.18 servers only
1111 Description: ll_kern_mount() doesn't release the module reference
1112 Details : The ldiskfs module reference count never drops down to 0
1113 because ll_kern_mount() doesn't release the module reference.
1118 Description: server LBUG when using old ost_num_threads parameter
1119 Details : Accept the old ost_num_threads parameter but warn that it
1120 is deprecated, and fix an off-by-one error that caused an LBUG.
1125 Description: Transient SCSI error results in persistent IO issue
1126 Details : iobuf->dr_error is not reinitialized to 0 between two
1130 Frequency : sometimes when underlying device returns I/O errors
1132 Description: OSTs not going read-only during write failures
1133 Details : OSTs are not remounted read-only when the journal commit threads
1134 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1135 instead of the ext3 wrappers.
1139 Description: poor connect scaling with increasing client count
1140 Details : Don't run filter_grant_sanity_check for more than 100 exports
1141 to improve scaling for large numbers of clients.
1144 Frequency : SLES10 only
1146 Description: sanity-quota.sh quotacheck failed: rc = -22
1147 Details : Quotas cannot be enabled on SLES10.
1150 Frequency : liblustre clients only
1152 Description: getdirentries does not give error when run on compute nodes
1153 Details : getdirentries does not fail when the size specified as an argument
1154 is too small to contain at least one entry
1156 Severity : enhancement
1158 Description: Add LNET router traceability for debug purposes
1159 Details : If a checksum failure occurs with a router as part of the
1160 IO path, the NID of the last router that forwarded the bulk data
1161 is printed so it can be identified.
1166 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1167 Details : Due to a race condition, liblustre clients were occasionally
1168 evicted incorrectly.
1170 Severity : enhancement
1172 Description: lfs setstripe use optional parameters instead of postional
1175 Severity : enhancement
1177 Description: Nanosecond timestamp support for ldiskfs
1178 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1179 resolution timestamps. There is not yet support for this at
1180 the Lustre filesystem level.
1183 Frequency : during server recovery
1185 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1186 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1190 Frequency : nfs export on patchless client
1192 Description: connectathon hang when test nfs export over patchless client
1193 Details : Disconnected dentry cannot be found with lookup, so we do not need
1194 to unhash it or make it invalid
1197 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1198 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1199 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1204 Description: open req refcounting wrong on reconnect
1205 Details : If reconnect happened between getting open reply from server and
1206 call to mdc_set_replay_data in ll_file_open, we will schedule
1207 replay for unreferenced request that we are about to free.
1208 Subsequent close will crash in variety of ways.
1209 Check that request is still eligible for replay in
1210 mdc_set_replay_data().
1215 Description: disable writes to filesystem when reading health_check file
1216 Details : the default for reading the health_check proc file has changed
1217 to NOT do a journal transaction and write to disk, because this
1218 can cause reads of the /proc file to hang and block HA state
1219 checking on a healthy but otherwise heavily loaded system. It
1220 is possible to return to the previous behaviour during configure
1221 with --enable-health-write.
1223 Severity : enhancement
1225 Description: 64-bit inode version
1226 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1227 to the inode. This will be required for version-based recovery.
1232 Description: MDS fails to start if a duplicate client export is detected
1233 Details : in some rare cases it was possible for a client to connect to
1234 an MDS multiple times. Upon recovery the MDS would detect this
1235 and fail during startup. Handle this more gracefully.
1237 Severity : enhancement
1239 Description: Add -o localflock option to simulate old noflock
1241 Details : This will achieve local-only flock/fcntl locks
1247 Description: log_commit_thread vs filter_destroy race leads to crash
1248 Details : Take import reference before releasing llog record semaphore
1253 Description: Wrong request locking in request set processing
1254 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1255 imp_delayed_list, in this place should be used imp_lock.
1258 Frequency : when reconnection
1260 Description: Grant Leak when osc reconnect to OST
1261 Details : When osc reconnect ost, OST(filter) should check whether it
1262 should grant more space to client by comparing fed_grant and
1263 cl_avail_grant, and return the granted space to client instead
1264 of "new granted" space, because client will call osc_init_grant
1265 to update the client grant space info.
1268 Frequency : when client reconnect to OST
1270 Description: Grant Leak when osc do resend and replay bulk write
1271 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1272 bulk write request, because the grant info while be sync between
1273 OSC and OST when reconnect, and we should ignore the grant info
1274 these of resend/replay write req.
1279 Description: Grant space more than avaiable left space sometimes.
1280 Details : When then OST is about to be full, if two bulk writing from
1281 different clients came to OST. Accord the avaliable space of the
1282 OST, the first req should be permitted, and the second one
1283 should be denied by ENOSPC. But if the seconde arrived before
1284 the first one is commited. The OST might wrongly permit second
1285 writing, which will cause grant space > avaiable space.
1288 Frequency : when client is evicted
1290 Description: Grant might be wrongly erased when osc is evicted by OST
1291 Details : when the import is evicted by server, it will fork another
1292 thread ptlrpc_invalidate_import_thread to invalidate the
1293 import, where the grant will be set to 0. While the original
1294 thread will update the grant it got when connecting. So if
1295 the former happened latter, the grant will be wrongly errased
1296 because of this race.
1301 Description: Checking Stale with correct fid
1302 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1303 to check whether it is stale, because sometimes, we want the
1304 enqueue happened anyway, and op_data.fid2 will not be initialized.
1306 Severity : enhancement
1308 Description: update patchless client
1309 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1312 Frequency : only with 2.4 kernel
1314 Description: random memory corruption
1315 Details : size of struct ll_inode_info is to big for union inode.u and this
1316 can be cause of random memory corruption.
1321 Description: Memory leak in recovery
1322 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1323 It should also check obd_fail before fsfilt_start, otherwise if
1324 fsfilt_start return -EROFS,(failover mds during mds recovery).
1325 then the req will return with repmsg->transno = 0 and rc = EROFS.
1326 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1327 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1328 be freed no matter whether fsfilt_commit success or not.
1331 Frequency : only with huge count clients
1333 Description: Prevents from taking the superblock lock in llap_from_page for
1335 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1336 for llap_from_page call in ll_removepage prevents from taking the
1337 superblock lock for a soon died page.
1342 Description: Not check open intent error before release open handle
1343 Details : in some rare cases, the open intent error is not checked before
1344 release open handle, which may cause
1345 ASSERTION(open_req->rq_transno != 0), because it tries to release
1346 the failed open handle.
1351 Description: Set cat log bitmap only after create log success.
1352 Details : in some rare cases, the cat log bitmap is set too early. and it
1353 should be set only after create log success.
1357 Description: Accessing a block bevice can re-enable I/O when Lustre is
1358 tearing down a device.
1359 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1364 Description: service threads may hog cpus when there are a lot of requests
1365 Details : Insert cond_resched to give other threads a chance to use some CPU
1370 Description: the cat log was not initialized in recovery
1371 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1372 unlink log on mds will not be initialized until mds post
1373 recovery. And also in mds post recovery, the unlink log will
1374 initialization will be done asynchronausly, so there will be race
1375 between add unlink log and unlink log initialization.
1379 Description: brw_stats were being printed incorrectly
1380 Details : brw_stats were being printed as log2 but all of them were not
1381 recorded as log2. Also remove some code duplication arising from
1382 filter_tally_{read,write}.
1386 Frequency : rare, only in recovery.
1387 Description: ASSERTION(req->rq_type != LI_POISON) failed
1388 Details : imp_lock should be held while iterating over imp_sending_list for
1389 prevent destroy request after get timeout in ptlrpc_queue_wait.
1393 Description: replay-single.sh test 52 fails
1394 Details : A lock's skiplist need to be cleanup when it being unlinked
1395 from its resource list.
1399 Description: Short directio read returns full requested size rather than
1401 Details : Direct I/O operations should return actual amount of bytes
1402 transferred rather than requested size.
1404 Severity : enhancement
1406 Description: metadata RPC reduction (e.g. for rm performance)
1407 Details : decrease the amount of synchronous RPC between clients and servers
1408 by canceling conflicing lock before the operation on the client
1409 and packing thier handles into the main operation RPC to server.
1411 Severity : enhancement
1413 Description: Async OSC create to avoid the blocking unnecessarily.
1414 Details : If a OST has no remain object, system will block on the creating
1415 when need to create a new object on this OST. Now, ways use
1416 pre-created objects when available, instead of blocking on an
1417 empty osc while others are not empty. If we must block, we block
1418 for the shortest possible period of time.
1420 Severity : enhancement
1422 Description: refine locking for avoid write wrong info into lov_objid file
1423 Details : fix possible races with add new target and write/update data in
1426 --------------------------------------------------------------------------------
1428 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1433 Frequency : on some architectures
1435 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1436 Details : Uninitialized flags sometimes cause configuration commands to
1440 Frequency : patchless clients only
1442 Description: missing __iget() symbol export
1443 Details : The __iget() symbol export is missing. To avoid the need for
1444 this on patchless clients the deathrow inode reaper is turned
1445 off, and we depend on the VM to clean up old inodes. This
1446 dependency was during via the fix for bug 12181.
1448 --------------------------------------------------------------------------------
1450 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1452 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1453 INTEROPERATE with older versions automatically. In many cases a
1454 special upgrade step is needed. Please read the
1455 user documentation before upgrading any part of a live system.
1456 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1457 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1458 * WARNING: Lustre configuration and startup changes are required with
1459 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1465 Severity : enhancement
1467 Description: MountConf
1468 Details : Lustre configuration is now managed via mkfs and mount
1469 commands instead of lmc and lconf. New obd types (MGS, MGC)
1470 are added for dynamic configuration management. See
1471 https://mail.clusterfs.com/wikis/lustre/MountConf for
1474 Severity : enhancement
1476 Description: dynamic OST addition
1477 Details : OSTs can now be added to a live filesystem
1479 Severity : enhancement
1481 Description: startup order invariance
1482 Details : MDTs and OSTs can be started in any order. Clients only
1483 require the MDT to complete startup.
1485 Severity : enhancement
1487 Description: parallel, asynchronous orphan cleanup
1488 Details : orphan cleanup is now performed in separate threads for each
1489 OST, allowing parallel non-blocking operation.
1491 Severity : enhancement
1493 Description: optimized stripe assignment
1494 Details : stripe assignments are now made based on ost space available,
1495 ost previous usage, and OSS previous usage, in order to try
1496 to optimize storage space and networking resources.
1498 Severity : enhancement
1500 Description: Permanently set tunables
1501 Details : All writable /proc/fs/lustre tunables can now be permanently
1502 set on a per-server basis, at mkfs time or on a live system.
1504 Severity : enhancement
1506 Description: Lustre message v2
1507 Details : Add lustre message format v2.
1509 Severity : enhancement
1511 Description: client OST exclusion list
1512 Details : Clients can be started with a list of OSTs that should be
1513 declared "inactive" for known non-responsive OSTs.
1517 Description: SPEC SFS validation failure on NFS v2 over lustre.
1518 Details : Changes the blocksize for regular files to be 2x RPC size,
1519 and not depend on stripe size.
1521 Severity : enhancement
1523 Description: Multiple MD RPCs in flight.
1524 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1525 To avoid overly-overloading MDS, introduce a limit on number of
1526 MDS RPCs in flight for a single client and add /proc controls
1527 to adjust this limit.
1529 Severity : enhancement
1531 Description: client read/write statistics
1532 Details : Add client read/write call usage stats for performance
1533 analysis of user processes.
1534 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1535 file access. extents_stats shows chunk size distribution.
1536 extents_stats_per_process show chunk size distribution per
1539 Severity : enhancement
1541 Description: mds statistics
1542 Details : Add detailed mds operations statistics in
1543 /proc/fs/lustre/mds/*/stats.
1547 Description: Failure of copying files with lustre special EAs.
1548 Details : Client side always return success for setxattr call for lustre
1549 special xattr (currently only "trusted.lov").
1554 Description: Refcount LNET uuids
1555 Details : The global LNET uuid list grew linearly with every startup;
1556 refcount repeated list entries instead of always adding to
1559 Severity : enhancement
1561 Description: Dynamic service threads
1562 Details : Within a small range, start extra service threads
1563 automatically when the request queue builds up.
1566 Frequency : mixed-endian client/server environments
1568 Description: mixed-endian crashes
1569 Details : The new msg_v2 system had some failures in mixed-endian
1572 Severity : enhancement
1574 Description: Easy OST removal
1575 Details : OSTs can be permanently deactivated with e.g. 'lctl
1576 conf_param lustre-OST0001.osc.active=0'
1578 Severity : enhancement
1580 Description: MGS proc entries
1581 Details : Added basic proc entries for the MGS showing what filesystems
1584 Severity : enhancement
1586 Description: provide MGS failover
1587 Details : Added config lock reacquisition after MGS server failover.
1589 Severity : enhancement
1591 Description: add Linux 2.4 support
1592 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1596 Description: a large application tries to do I/O to the same resource and dies
1597 in the middle of it.
1598 Details : Check the req->rq_arrival time after the call to
1599 ost_brw_lock_get(), but before we do anything about
1600 processing it & sending the BULK transfer request. This
1601 should help move old stale pending locks off the queue as
1602 quickly as obd_timeout.
1605 Frequency : when an incorrect nid is specified during startup
1607 Description: ptlrpc connect to non-existant node causes kernel crash
1608 Details : LNET can't be re-entered from an event callback, which
1609 happened when we expire a message after the export has been
1610 cleaned up. Instead, hand the zombie cleanup off to another
1613 Severity : enhancement
1615 Description: plain/inodebits lock performance improvement
1616 Details : Grouping plain/inodebits in granted list by their request modes
1617 and bits policy, thus improving the performance of search through
1621 Frequency : only if OST filesystem is corrupted
1623 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1624 Details : for a short time RPCs with bulk IO are in the replay list,
1625 but replay of bulk IOs is unimplemented. If the OST filesystem
1626 is corrupted due to disk cache incoherency and then replay is
1627 started it is possible to trip an assertion. Avoid putting
1628 committed RPCs into the replay list at all to avoid this issue.
1631 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1633 Description: System hang on startup
1634 Details : This bug allowed the liblustre (e.g. catamount) client to
1635 return to the app before handling all startup RPCs. This
1636 could leave the node unresponsive to lustre network traffic
1637 and manifested as a server ptllnd timeout.
1639 Severity : enhancement
1641 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1642 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1643 boolean to control whether to print peer debug info when a
1644 client's RPC times out.
1647 Frequency : only for kernels with patches from Lustre below 1.4.3
1649 Description: Remove old rdonly API
1650 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1653 Frequency : only for devices with external journals
1655 Description: Set external device read-only also
1656 Details : During a commanded failover stop, we set the disk device
1657 read-only while the server shuts down. We now also set any
1658 external journal device read-only at the same time.
1661 Frequency : when upgrading from 1.4 while trying to change parameters
1663 Description: The wrong (new) MDC name was used when setting parameters for
1664 upgraded MDT's. Also allows changing of OSC (and MDC)
1665 parameters if --writeconf is specified at tunefs upgrade time.
1668 Frequency : when setting specific ost indicies
1670 Description: QOS code breaks on skipped indicies
1671 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1672 created with --index need not be sequential.
1674 Severity : enhancement
1676 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1677 Details : The uninit_groups feature works in conjunction with the kernel
1678 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1679 up the pass1 processing of e2fsck. This is a read-only feature
1680 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1681 mount filesystems that have had this feature enabled.
1683 Severity : enhancement
1685 Description: Improve multi-block allocation algorithm to avoid fragmentation
1686 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1687 allocation locality and avoid filesystem fragmentation.
1689 ------------------------------------------------------------------------------
1691 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1693 * Support for kernels:
1694 2.4.21-47.0.1.EL (RHEL 3)
1695 2.6.5-7.283 (SLES 9)
1696 2.6.9-42.0.10.EL (RHEL 4)
1697 2.6.12.6 vanilla (kernel.org)
1698 2.6.16.27-0.9 (SLES 10)
1699 * Recommended e2fsprogs version: 1.39.cfs5
1701 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1705 Frequency : occasional, depends on client load and configuration
1706 Bugzilla : 12181, 12203
1707 Description: data loss for recently-modified files
1709 Details : In some cases it is possible that recently written or created
1710 files may not be written to disk in a timely manner (this should
1711 normally be within 30s unless client IO load is very high).
1712 The problem appears as zero-length files or files that are a
1713 multiple of 1MB in size after a client crash or client eviction
1714 that are missing data at the end of the file.
1716 This problem is more likely to be hit on clients where files are
1717 repeatedly created and unlinked in the same directory, clients
1718 have a large amount of RAM, have many CPUs, the filesystem has
1719 many OSTs, the clients are rebooted frequently, and/or the files
1720 are not accessed by other nodes after being written.
1722 The presence of the problem can be detected by looking at
1723 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1724 smaller than the second (nr_unused) then dirty files will not
1725 be flushed automatically to disk. "sync; sleep 10" should be
1726 run several times on the node before unmounting it to update
1727 Lustre (this is also safe to run on nodes without this problem).
1729 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1730 that can cause this same problem, so customers using that kernel
1731 also need to update the kernel in addition to Lustre. In order
1732 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1734 It is normal that files written just before a client crash (less
1735 than 30s) may not yet have been flushed to disk, even for local
1739 Frequency : frequent on thin XT3 nodes
1741 Description: UUID collision on thin XT3 Linux nodes
1742 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1743 randomly, since we relied on an insufficiently-seeded PRNG.
1748 Description: OSS hangs after "All ost request buffers busy"
1749 Details : A deadlock between quota and journal operations caused OSS
1750 hangs after printing "All ost request buffers busy."
1753 Frequency : always on liblustre builds
1755 Description: Cleanup compiler warnings on liblustre
1758 Frequency : always on liblustre builds on XT3
1760 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1764 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1765 > 32000 subdirectories
1766 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1767 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1768 subdirectory count crosses 32000. This will aid e2fsck to
1769 correctly handle more than 32000 subdirectories.
1772 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1774 Description: System hang on startup
1775 Details : This bug allowed the liblustre (e.g. catamount) client to
1776 return to the app before handling all startup RPCs. This
1777 could leave the node unresponsive to lustre network traffic
1778 and manifested as a server ptllnd timeout.
1780 Severity : enhancement
1782 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1783 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1784 boolean to control whether to print peer debug info when a
1785 client's RPC times out.
1790 Description: make O_SYNC working on 2.6 kernels
1791 Details : 2.6 kernels use different method for mark pages for write,
1792 so need add a code to lustre for O_SYNC work.
1797 Description: Failure to close file and release space on NFS
1798 Details : Put inode details into lock acquired in ll_intent_file_open.
1799 Use mdc_intent_lock in ll_intent_open to properly
1800 detect all kind of errors unhandled by mdc_enqueue.
1805 Description: proc file read during shutdown sometimes raced obd removal,
1807 Details : Add lock to prevent obd access after proc file removal.
1810 Frequency : Only for files larger than 4GB on 32-bit clients.
1812 Description: improperly doing page alignment of locks
1813 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1814 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1819 Description: RPCs being resent when they shouldn't be
1820 Details : Some RPCs that should not be resent are being resent. This
1821 can cause inconsistencies in the RPC state machine. Do not
1822 resend such requests.
1825 Frequency : rare, only with NFS export
1827 Description: Crash on NFS re-export node
1828 Details : under very unusual load conditions an assertion is hit in
1829 ll_intent_file_open()
1832 Frequency : only if OST filesystem is corrupted
1834 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1835 Details : for a short time RPCs with bulk IO are in the replay list,
1836 but replay of bulk IOs is unimplemented. If the OST filesystem
1837 is corrupted due to disk cache incoherency and then replay is
1838 started it is possible to trip an assertion. Avoid putting
1839 committed RPCs into the replay list at all to avoid this issue.
1844 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1845 Details : Large single O_DIRECT read and write calls can fail to allocate
1846 a sufficiently large buffer to process the request. In case of
1847 allocation failure the allocation is retried with a smaller
1848 buffer and broken into smaller requests.
1850 Severity : enhancement
1852 Description: Add -o localflock option to simulate old noflock behaviour.
1853 Details : This will achieve local-only flock/fcntl locks coherentness.
1858 Description: versioning check is incomplete
1859 Details : Checking the version difference of client vs. server, report
1860 error if the gap is too big.
1865 Description: add support PG_writeback bit
1866 Details : add support for PG_writeback bit for Lustre, for more carefull
1867 work with page cache in 2.6 kernel. This also fix some deadlocks
1868 and remove hack for work O_SYNC with 2.6 kernel.
1870 Severity : enhancement
1872 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1873 Details : The uninit_groups feature works in conjunction with the kernel
1874 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1875 up the pass1 processing of e2fsck. This is a read-only feature
1876 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1877 mount filesystems that have had this feature enabled.
1879 Severity : enhancement
1881 Description: Improve multi-block allocation algorithm to avoid fragmentation
1882 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1883 allocation locality and avoid filesystem fragmentation.
1885 ------------------------------------------------------------------------------
1887 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1889 * Support for kernels:
1890 2.6.9-42.0.3.EL (RHEL 4)
1891 2.6.5-7.276 (SLES 9)
1892 2.4.21-47.0.1.EL (RHEL 3)
1893 2.6.12.6 vanilla (kernel.org)
1894 2.6.16.21-0.8 (SLES10)
1895 * Recommended e2fsprogs version: 1.39.cfs2-0
1897 * The backwards-compatible /proc/sys/portals symlink has been removed
1898 in this release. Before upgrading, please ensure that you change
1899 any configuration scripts or /etc/sysctl.conf files that access
1900 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1901 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1902 in advance of the upgrade on any system running Lustre 1.4.6 or
1903 newer, since /proc/sys/lnet was added in that version.
1904 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1908 Frequency : only when quota is used
1910 Description: avoid scanning export list for quota master
1911 Details : Change the algorithms to avoid scanning export list in order
1912 to improve the efficiency.
1915 Frequency : MDS failover only, very rarely
1917 Description: "went back in time" messages on mds failover
1918 Details : The greatest transno may be lost when the current operation
1919 finishes with an error (transno==0) and the client's last_rcvd
1920 record is over-written. Save the greatest transno in the
1921 mds_last_transno for this case.
1924 Frequency : always for specific kernels and striping counts
1926 Description: client may get "Matching packet too big" without ACL support
1927 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1928 when trying to access files in certain configurations. The
1929 clients should in fact be denied when mounting because they do
1930 not understand ACLs.
1933 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1935 Description: many clients connecting with IO in progress causes connect timeouts
1936 Details : Avoid synchronous journal commits to avoid delays caused by many
1937 clients connecting/disconnecting when bulk IO is in progress.
1938 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1939 OST_IO_PORTAL to avoid delays behind potentially many pending
1943 Frequency : occasionally with multiple writers to a single file
1945 Description: shared writes to file may result in wrong size reported by stat()
1946 Details : Allow growing of kms when extent lock is cancelled
1949 Frequency : always with random mmap IO to multi-striped file
1951 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1952 Details : Only if the hole is at the end of OST object so that kms is too
1953 small. Fix is to increase kms accordingly in ll_nopage.
1956 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1958 Description: writes to a missing object would leak memory on the OST
1959 Details : If there is an inconsistency between the MDS and OST filesystems,
1960 such that the MDS references an object that doesn't exist, writes
1961 to that object will leak memory due to incorrect cleanup in the
1962 error handling path, eventually running out of memory on the OST.
1967 Description: Creating too long symlink causes lustre errors
1968 Details : Check symlink and name lengths before sending requests to MDS.
1971 Frequency : only if flock is enabled (not on by default)
1973 Description: posix locks not released on fd closure on 2.6.9+
1974 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1975 kernels, this caused such locks not to be released on fd close and
1976 then assertions on fs unmount about still used locks.
1979 Frequency : MDS failover only, very rarely
1981 Description: clients may get ASSERTION(granted_lock != NULL)
1982 Details : When request was taking a long time, and a client was resending
1983 a getattr by name lock request. The were multiple lock requests
1984 with the same client lock handle and
1985 mds_getattr_name->fixup_handle_for_resent_request found one of the
1986 lock handles but later failed with ASSERTION(granted_lock != NULL).
1991 Description: handle->h_buffer_credits > 0, assertion failure
1992 Details : h_buffer_credits is zero after truncate, causing assertion
1993 failure. This patch extends the transaction or creates a new
1997 Frequency : NFS re-export or patchless client
1998 Bugzilla : 11179, 10796
1999 Description: Crash on NFS re-export node (__d_move)
2000 Details : We do not want to hash the dentry if we don't have a lock.
2001 But if this dentry is later used in d_move, we'd hit uninitialised
2002 list head d_hash, so we just do this to init d_hash field but
2003 leave dentry unhashed.
2006 Frequency : NFS re-export or patchless client
2008 Description: NFS exports has problem with symbolic link
2009 Details : lustre client didn't properly install dentry when re-exported
2010 to NFS or running patchless client.
2013 Frequency : NFS re-export or patchless client
2015 Description: Various nfs/patchless fixes.
2016 Details : fixes reuse disconected alias for lookup process - this fixes
2017 warning "find_exported_dentry: npd != pd",
2018 fix permission error with open files at nfs.
2019 fix apply umask when do revalidate.
2022 Frequency : occasional
2024 Description: Crash on NFS re-export node
2025 Details : calling clear_page() on the wrong pointer triggered oops in
2026 generic_mapping_read().
2029 Frequency : rarely, using O_DIRECT IO
2031 Description: unaligned directio crashes client with LASSERT
2032 Details : check for unaligned buffers before trying any requests.
2035 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2037 Description: stale data returned from RAID cache
2038 Details : If only a small amount of IO is done to the RAID device before
2039 reading it again it is possible to get stale data from the RAID
2040 cache instead of reading it from disk.
2043 Frequency : always for sles10 kernel
2045 Description: sles10 support
2046 Details : ll_follow_link: compile fixes and using of nd_set_link
2047 under newer kernels.
2050 Frequency : depends on arch, kernel and compiler version, always on sles10
2053 Description: recursive or deep enough symlinks cause stack overflow
2054 Details : getting rid of large stack-allocated variable in
2058 Frequency : depends on hardware
2060 Description: lustre write performance loss in the SLES10 kernel
2061 Details : the performance loss is caused by using of write barriers in the
2062 ext3 code. The SLES10 kernel turns barrier support on by
2063 default. The fix is to undo that change for ldiskfs.
2065 ------------------------------------------------------------------------------
2067 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2069 * Support for kernels:
2070 2.6.9-42.0.3EL (RHEL 4)
2071 2.6.5-7.276 (SLES 9)
2072 2.4.21-47.0.1.EL (RHEL 3)
2073 2.6.12.6 vanilla (kernel.org)
2077 Frequency : quota enabled and large files being deleted
2079 Description: releasing more than 4GB of quota at once hangs OST
2080 Details : If a user deletes more than 4GB of files on a single OST it
2081 will cause the OST to spin in an infinite loop. Release
2082 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2087 Description: statfs data retrieved from /proc may be stale or zero
2088 Details : When reading per-device statfs data from /proc, in the
2089 {kbytes,files}_{total,free,avail} files, it may appear
2090 as zero or be out of date.
2093 Frequency : systems with MD RAID1 external journal devices
2095 Description: lconf's call to blkid is confused by RAID1 journal devices
2096 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2097 of returning all block devices that match the journal UUID.
2100 Frequency : always, for aggregate stripe size over 4GB
2102 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2103 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2104 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2107 Frequency : always if "lfs find" used on a local file/directory
2109 Description: "lfs find" segfaults if used on a local file/directory
2110 Details : The case where a directory component was not specified wasn't
2111 handled correctly. Handle this properly.
2114 Frequency : always on ppc64
2116 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2117 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2118 perform endianness conversion.
2121 Frequency : rarely (truncate to non-zero file size after write under load)
2122 Bugzilla : 10730, 10687
2123 Description: Files padded with zeros to next 4K multiple
2124 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2125 it is possible that files that are truncated to a non-zero size
2126 immediately after being written are filled with zero bytes beyond
2127 the truncated size. No file data is lost.
2129 Severity : enhancement
2131 Description: Allow recovery/failover for liblustre clients.
2132 Details : liblustre clients were unaware of failover configurations until
2135 Severity : enhancement
2137 Description: user file locks should fail when not mounting with flock option
2138 Details : Set up an error-returning stub in ll_file_operations.lock field
2139 to prevent incorrect behaviour when client is mounted without
2140 flock option. Also, set up properly f_op->flock field for
2144 Frequency : always on ia64
2146 Description: "lfs df" loops on printing out MDS statfs information
2147 Details : The obd_ioctl_data was not initialized and in some systems
2148 this caused a failure during the ioctl that did not return
2149 an error. Initialize the struct and return an error on failure.
2152 Frequency : SLES 9 only
2154 Description: Error of copying files with lustre special EAs as root
2155 Details : Client side always return success for setxattr call for lustre
2156 special xattr (currently only "trusted.lov").
2159 Frequency : rarely on clusters with both ia64+i386 clients
2161 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2162 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2163 clients are doing concurrent writes to the same file it is
2164 possible that the ia64 clients may LASSERT because the OST
2165 extent locks are not PAGE_SIZE aligned. Ensure that grown
2166 locks are always aligned on the request boundary.
2169 Frequency : specific use, occasional
2171 Description: Overwriting in use executable truncates on-disk binary image
2172 Details : If one node attempts to overwrite an executable in use by
2173 another node, we now correctly return ETXTBSY instead of
2174 truncating the file.
2176 Severity : enhancement
2178 Description: Async OSC create to avoid the blocking unnecessarily.
2179 Details : If a OST has no remain object, system will block on the creating
2180 when need to create a new object on this OST. Now, ways use
2181 pre-created objects when available, instead of blocking on an
2182 empty osc while others are not empty. If we must block, we block
2183 for the shortest possible period of time.
2188 Description: chmod on Lustre root is propagated to other clients
2189 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2190 invalid by the follow_mount time.
2193 Frequency : liblustre clients only
2195 Description: Race in 'instant cancel' lock handling could lead to such locks
2196 never to be granted in case of SMP MDS
2197 Details : Do not destroy not yet granted but cbpending locks in
2201 Frequency : replay/resend of open
2203 Description: non null lock assetion failure in mds_intent_policy
2204 Details : Trying to replay/resend lockless open requests resulted in
2205 mds_open() returning 0 with no lock. Now it sets a flag if
2206 a lock is going to be returned.
2208 Severity : enhancement
2210 Description: Checksum enhancements
2211 Details : New checksum enhancements allow for resending RPCs that failed
2214 Severity : enhancement
2216 Description: Tunables on number of dirty pages in cacche
2217 Details : Allow to set limit on number of dirty pages cached.
2222 Description: client crash on unmount - lock still has references
2223 Details : In some error handling cases it was possible to leak a lock
2224 reference on a client while accessing a file. This was not
2225 harmful to the client during operation, but would cause the
2226 client to crash when the filesystem is unmounted.
2229 Frequency : specific case, rare
2231 Description: ETXTBSY on mds though file not in use by client
2232 Details : ETXTBSY is no longer incorrectly returned when attempting to
2233 chmod or chown a directory that the user previously tried to
2234 execute or a currently-executing binary.
2237 Frequency : extremely rare except on liblustre-based clients
2239 Description: Lustre space not freed when files are deleted
2240 Details : Clean up open-unlinked files after client eviction. Previously
2241 the unlink was skipped and the files remained as orphans.
2246 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2247 Details : In some cases it was possible to send a blocking callback to
2248 a client doing a glimpse, even though that client didn't get
2249 a lock granted. When the glimpse lock is cancelled on the OST
2250 the freed lock is left on the waiting list and corrupted the list.
2253 Frequency : all core dumps
2255 Description: Broke core dumps to lustre
2256 Details : Negative dentry may be unhashed if parent does not have UPDATE
2257 lock, but some callers, e.g. do_coredump, expect dentry to be
2258 hashed after successful create, hash it in ll_create_it.
2260 ------------------------------------------------------------------------------
2262 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2264 * Support for kernels:
2265 2.6.9-42.0.2.EL (RHEL 4)
2266 2.6.5-7.276 (SLES 9)
2267 2.4.21-40.EL (RHEL 3)
2268 2.6.12.6 vanilla (kernel.org)
2272 Frequency : always on RHEL 3
2274 Description: Number of open files grows over time
2275 Details : The number of open files grows over time, whether or not
2276 Lustre is started. This was due to a filp leak introduced
2277 by one of our kernel patches.
2279 ------------------------------------------------------------------------------
2281 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2283 * Support for kernels:
2284 2.6.9-42.EL (RHEL 4)
2285 2.6.5-7.276 (SLES 9)
2286 2.4.21-40.EL (RHEL 3)
2287 2.6.12.6 vanilla (kernel.org)
2292 Bugzilla : 5719, 9635, 9792, 9684
2293 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2294 Details : If a server is under heavy load and cannot reply to new
2295 connection requests before the client resends the (re)connect,
2296 the connection handling code can behave badly if two service
2297 threads are concurrently handing separate (re)connections from
2298 the same client. Add better locking to the connection handling
2299 code, and ensure that only a single connection will be processed
2300 for a given client UUID, even if the lock is dropped.
2302 Severity : enhancement
2304 Description: add TCP zero-copy support to kernel
2305 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2306 sends if the hardware supports scatter-gather and checksumming.
2307 This allows socklnd to do client-write and server-read more
2308 efficiently and reduce CPU utilization from skbuf copying.
2311 Frequency : only if NFS exporting from client
2313 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2314 Details : If setting attributes on a file created under NFS that had
2315 never been opened it would be possible to oops the client
2316 if the file had no objects.
2319 Frequency : always for liblustre
2321 Description: liblustre client does MDS+OSTs setattr RPC for each write
2322 Details : When doing a write from a liblustre client, the client
2323 incorrectly issued an RPC to the MDS and each OST the file was
2324 striped over in order to update the timestamps. When writing
2325 with small chunks and many clients this could overwhelm the MDS
2326 with RPCs. In all cases it would slow down the write because
2327 these RPCs are unnecessary.
2329 Severity : enhancement
2331 Description: allow number of MDS service threads to be changed at module load
2332 Details : It is now possible to change the number of MDS service threads
2333 running. Adding "options mds mds_num_threads={N}" to the MDS's
2334 /etc/modprobe.conf will set the number of threads for the next
2335 time Lustre is restarted (assuming the "mds" module is also
2336 reloaded at that time). The default number of threads will
2337 stay the same, 32 for most systems.
2342 Description: OST crash if filesystem is unformatted or corrupt
2343 Details : If an OST is started on a device that has never been formatted
2344 or if the filesystem is corrupt and cannot even mount then the
2345 error handling cleanup routines would dereference a NULL pointer.
2350 Description: NULL pointer deref in llap_from_page.
2351 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2352 incorrect) mapping if the page was truncated/reclaimed while it was
2353 searched for. Check for this condition and skip such pages when
2354 doing readahead. Introduce extra check to llap_from_page() to
2355 verify page->mapping->host is non-NULL (so page is not anonymous).
2358 Frequency : Sometimes when using sys_sendfile
2360 Description: "page not covered by a lock" warnings from ll_readpage
2361 Details : sendfile called ll_readpage without right page locks present.
2362 Now we introduced ll_file_sendfile that does necessary locking
2363 around call to generic_file_sendfile() much like we do in
2367 Frequency : with certain MDS communication failures at client mount time
2369 Description: NULL pointer deref after failed client mount
2370 Details : a client connection request may delayed by the network layer
2371 and not be sent until after the PTLRPC layer has timed out the
2372 request. If the client fails the mount immediately it will try
2373 to clean up before the network times out the request. Add a
2374 reference from the request import to the obd device and delay
2375 the cleanup until the network drops the request.
2378 Frequency : occasionally during client (re)connect
2380 Description: assertion failure during client (re)connect
2381 Details : processing a client connection request may be delayed by the
2382 client or server longer than the client connect timeout. This
2383 causes the client to resend the connection request. If the
2384 original connection request is replied in this interval, the
2385 client may trip an assertion failure in ptlrpc_connect_interpret()
2386 which thought it would be the only running connect process.
2389 Frequency : only with obd_echo servers and clients that are rebooted
2391 Description: kernel BUG accessing uninitialized data structure
2392 Details : When running an obd_echo server it did not start the ping_evictor
2393 thread, and when a client was evicted an uninitialized data
2394 structure was accessed. Start the ping_evictor in the RPC
2395 service startup instead of the OBD startup.
2397 Severity : enhancement
2398 Bugzilla : 10193 (patchless)
2399 Description: Remove dependency on various unexported kernel interfaces.
2400 Details : No longer need reparent_to_init, exit_mm, exit_files,
2401 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2404 Frequency : rare (only users of deprecated and unsupported LDAP config)
2406 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2407 Details : LDAP apparently contains 'lustreName' attributes instead of
2408 'name'. A simple remapping of the name is sufficient.
2411 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2413 Description: waiting_locks_callback trips kernel BUG if client is evicted
2414 Details : Running with the dump_on_timeout debug flag turned on makes
2415 it possible that the waiting_locks_callback() can try to dump
2416 the Lustre kernel debug logs from an interrupt handler. Defer
2417 this log dumping to the expired_lock_main() thread.
2419 Severity : enhancement
2421 Description: Support NFS exporting on 2.6 kernels.
2422 Details : Implement non-rawops metadata methods for NFS server to use without
2423 changing NFS server code.
2426 Frequency : very rare (synthetic metadata workload only)
2428 Description: two racing renames might cause an MDS thread to deadlock
2429 Details : Running the "racer" program may cause one MDS thread to rename
2430 a file from being the source of a rename to being the target of
2431 a rename at exactly the same time that another thread is doing
2432 so, and the second thread has already enqueued these locks after
2433 doing a lookup of the target and is trying to relock them in
2434 order. Ensure that we don't try to re-lock the same resource.
2437 Frequency : only very large systems with liblustre clients
2439 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2440 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2441 all OSTs in parallel. This allows the eviction of stale liblustre
2442 clients to proceed much faster than if they were done in series,
2443 and also offers similar improvements for other set_info RPCs.
2448 Description: excessive CPU usage during initial read phase on client
2449 Details : During the initial read phase on a client, it would agressively
2450 retry readahead on the file, consuming too much CPU and impacting
2451 performance (since 1.4.5.8). Improve the readahead algorithm
2452 to avoid this, and also improve some other common cases (read
2453 of small files in particular, where "small" is files smaller than
2454 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2459 Description: MDS crash when receiving packet with unknown intent.
2460 Details : Do not LBUG in unknown intent case, just return -EFAULT
2462 Severity : enhancement
2463 Bugzilla : 9293, 9385
2464 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2465 Details : Do not serialize getattr (non-intent version) and statfs.
2468 Frequency : occasional, when OST network is overloaded/intermittent
2470 Description: client evicted by OST after bulk IO timeout
2471 Details : If a client sends a bulk IO request (read or write) the OST
2472 may evict the client if it is unresposive to its data GET/PUT
2473 request. This is incorrect if the network is overloaded (takes
2474 too long to transfer the RPC data) or dropped the OST GET/PUT
2475 request. There is no need to evict the client at all, since
2476 the pinger and/or lock callbacks will handle this, and the
2477 client can restart the bulk request.
2480 Frequency : Always when mmapping file with no objects
2482 Description: client crashes when mmapping file with no objects
2483 Details : Check that we actually have objects in a file before doing any
2484 operations on objects in ll_vm_open, ll_vm_close and
2490 Description: Request leak when working with deleted CWD
2491 Details : Introduce advanced request refcount tracking for requests
2492 referenced from lustre intent.
2494 Severity : Enhancement
2496 Description: Cache open file handles on client.
2497 Details : MDS now will return special lock along with openhandle, if
2498 requested and client is allowed to hold openhandle, even if unused,
2499 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2500 closing files for every read/write openration.
2502 Severity : Enhancement
2504 Description: Cache open negative dentries on client when possible.
2505 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2506 negative dentries on lock revocation.
2511 Description: Remounting a client read-only wasn't possible with a zconf mount
2512 Details : It wasn't possible to remount a client read-only with llmount.
2514 Severity : enhancement
2515 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2516 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2517 MPICH is included to add Lustre-specific ADIO interfaces.
2518 This is based closely on the UFS ADIO layer and only differs
2519 in file creation, in order to allow the OST striping to be set.
2520 This is user-contributed code and not supported by CFS.
2525 Description: extended inode attributes (immutable, append-only) work improperly
2526 when 2.4 and 2.6 kernels are used on client/server or vice versa
2527 Details : Introduce kernel-independent values for these flags.
2529 Severity : enhancement
2532 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2533 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2534 granularity. Now, Fractional MB granularity is be supported,
2535 this is very useful for low memory system.
2537 Severity : enhancement
2539 Description: Getattr by fid
2540 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2541 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2542 have locks and avoids getattr rpc requests for it on every stat.
2545 Frequency : Always, for filesystems larger than 2TB
2547 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2548 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2549 filesystems larger than 2TB to be mounted with mballoc enabled.
2552 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2554 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2555 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2556 filesystems larger than 2TB is mounted then the kernel will
2557 silently corrupt the start of the filesystem. CONFIG_LBD is
2558 enabled for all CFS-supported kernels, but the possibility of
2559 this happening with a modified kernel config exists.
2561 Severity : enhancement
2563 Description: add client O_DIRECT support for 2.6 kernels
2564 Details : It is now possible to do O_DIRECT reads and writes to files
2565 in the Lustre client mountpoint on 2.6 kernel clients.
2567 Severity : enhancement
2569 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2570 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2571 parallel, not waiting for response from every OST before sending
2572 a rpc to the next OST.
2577 Description: setattr vs write race when updating file timestamps
2578 Details : Client processes that update a file timestamp into the past
2579 right after writing to the file (e.g. tar) it is possible that
2580 the updated file modification time can be reset to the current
2581 time due to a race between processing the setattr and write RPC.
2583 Severity : enhancement
2585 Description: Bring 'lfs find' closer in line with regular Linux find.
2586 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2587 -print0 options and obtains all the needed info through the lustre
2590 Severity : enhancement
2592 Description: support up to 1024 configured devices on one node
2593 Details : change obd_dev array from statically allocated to dynamically
2594 allocated structs as they are first used to reduce memory usage
2599 Description: Flush dirty partially truncated pages during truncate
2600 Details : Immediatelly flush partially truncated pages in filter_setattr,
2601 this way we completely avoid having any pages in page cache on OST
2602 and can retire ugly workarounds during writes to flush such pages.
2607 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2608 Details : protect inode from truncation within vfs_unlink() context
2609 just take a reference before calling vfs_unlink() and release it
2610 when parent's i_sem is free.
2615 Description: last_id value checked outside lock on OST caused LASSERT failure
2616 Details : If there were multiple MDS->OST object precreate requests in
2617 flight, it was possible that the OST's last object id was checked
2618 outside a lock and incorrectly tripped an assertion. Move checks
2619 inside locks, and discard old precreate requests.
2622 Frequency : always, if extents are used on OSTs
2624 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2625 Details : OSTs using the extents format would not zero the high 16 bits of
2626 the index physical block number. This is not a problem for any
2627 OST filesystems smaller than 16TB, and no kernels support ext3
2628 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2629 new/modified files) and can be fixed for existing filesystems
2630 with e2fsprogs-1.39-cfs1.
2635 Description: import connection selection may be incorrect if timer wraps
2636 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2637 import connections to be ignored if the 32-bit jiffies counter
2638 wraps. Use a 64-bit jiffies counter.
2641 Frequency : very large clusters immediately after boot
2643 Description: LNET request buffers exhausted under heavy short-term load
2644 Details : If a large number of client requests are generated on a service
2645 that has previously never seen so many requests it is possible
2646 that the request buffer growth cannot keep up with the spike in
2647 demand. Instead of dropping incoming requests, they are held in
2648 the LND until the RPC service can accept more requests.
2651 Frequency : Sometimes during replay
2653 Description: Assertion failure in ll_local_open after replay.
2654 Details : If replay happened on an open request reply before we were able
2655 to set replay handler, reply will become not swabbed tripping the
2656 assertion in ll_local_open. Now we set the handler right after
2657 recognising of open request
2660 Frequency : very rare
2662 Description: kernel reports "badness in vsnprintf"
2663 Details : Reading from the "recovery_status" /proc file in small chunks
2664 may cause a negative length in lprocfs_obd_rd_recovery_status()
2665 call to vsnprintf() (which is otherwise harmless). Exit early
2666 if there is no more space in the output buffer.
2668 Severity : enhancement
2670 Description: clear OBD RPC statistics by writing to them
2671 Details : It is now possible to clear the OBD RPC statistics by writing
2672 to the "stats" file.
2677 Description: Client mtime is not the same on different clients after utimes
2678 Details : In some cases, the client was using the utimes() syscall on
2679 a file cached on another node. The clients now validate the
2680 ctime from the MDS + OSTs to determine which one is right.
2685 Description: Inability to activate failout mode
2686 Details : lconf script incorrectly assumed that in pythong string's numeric
2687 value is used in comparisons.
2690 Frequency : always with multiple stripes per file
2692 Description: Inefficient object allocation for mutli-stripe files
2693 Details : When selecting which OSTs to stripe files over, for files with
2694 a stripe count that divides evenly into the number of OSTs,
2695 the MDS is always picking the same starting OST for each file.
2696 Return the OST selection heuristic to the original design.
2701 Description: mount failures may take full timeout to return an error
2702 Details : Under some heavy load conditions it is possible that a
2703 failed mount can wait for the full obd_timeout interval,
2704 possibly several minutes, before reporting an error.
2705 Instead return an error as soon as the status is known.
2707 Frequency : quota enabled and large files being deleted
2709 Description: releasing more than 4GB of quota at once hangs OST
2710 Details : If a user deletes more than 4GB of files on a single OST it
2711 will cause the OST to spin in an infinite loop. Release
2712 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2717 Description: statfs data retrieved from /proc may be stale or zero
2718 Details : When reading per-device statfs data from /proc, in the
2719 {kbytes,files}_{total,free,avail} files, it may appear
2720 as zero or be out of date.
2723 Frequency : systems with MD RAID1 external journal devices
2725 Description: lconf's call to blkid is confused by RAID1 journal devices
2726 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2727 of returning all block devices that match the journal UUID.
2730 Frequency : always, for aggregate stripe size over 4GB
2732 Description: assertion fails when trying to use 4GB stripe size
2733 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2734 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2737 Frequency : always on ppc64
2739 Description: the first write on an ext3 filesystem with mballoc got stuck
2740 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2741 perform endianness conversion.
2743 ------------------------------------------------------------------------------
2745 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2747 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2748 INTEROPERATE with older versions automatically. Please read the
2749 user documentation before upgrading any part of a live system.
2750 * WARNING: Lustre networking configuration changes are required with
2751 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2754 * Support for kernels:
2755 2.6.9-22.0.2.EL (RHEL 4)
2756 2.6.5-7.244 (SLES 9)
2757 2.6.12.6 vanilla (kernel.org)
2760 Severity : enhancement
2761 Bugzilla : 7981/8208
2762 Description: Introduced Lustre Networking (LNET)
2763 Details : LNET is new networking infrastructure for Lustre, it includes
2764 a reorganized network configuration mode (see the user
2765 documentation for full details) as well as support for routing
2766 between different network fabrics. Lustre Networking Devices
2767 (LNDs) for the supported network fabrics have also been
2768 created for this new infrastructure.
2770 Severity : enhancement
2771 Description: Introduced Access control lists
2772 Details : clients can set ACLs on files and directories in order to have
2773 more fine-grained permissions than the standard Unix UGO+RWX.
2774 The MDS must be started with the "-o acl" mount option.
2776 Severity : enhancement
2777 Description: Introduced filesystem quotas
2778 Details : Administrators may now establish per-user quotas on the
2781 Severity : enhancement
2783 Description: Configuration change for the XT3
2784 The PTLLND is now used to run Lustre over Portals on the XT3
2785 The configure option(s) --with-cray-portals are no longer used.
2786 Rather --with-portals=<path-to-portals-includes> is used to
2787 enable building on the XT3. In addition to enable XT3 specific
2788 features the option --enable-cray-xt3 must be used.
2793 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2794 Details : A race between allocating a new llcd and re-getting the llcd_lock
2795 allowed another thread to grab newly-allocated llcd.
2797 Severity : enhancement
2799 Description: 2.6 OST async journal commit and locking fix to improve performance
2800 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2801 2.6 kernels are now async as they already were in 2.4 kernels so
2802 that they can commit concurrently with the network bulk transfer.
2803 For block-allocated files the filter allocation semaphore is held
2804 to avoid filesystem fragmentation during allocation. BKL lock
2805 removed for 2.6 xattr operations where it is no longer needed.
2810 Description: lconf incorrectly determined whether two IP networks could talk
2811 Details : In some more complicated routing and multiple-network
2812 configurations, lconf will avoid trying to make a network
2813 connection to a disjoint part of the IP space. It was doing the
2814 math incorrectly for one set of cases.
2819 Description: Fix for potential infinite loop processing records in an llog.
2820 Details : If an llog record is corrupted/zeroed, it is possible to loop
2821 forever in llog_process(). Validate the llog record length
2822 and skip the remainder of the block on error.
2825 Frequency : occasional (liblustre only)
2827 Description: liblustre could not open files whose last component is a symlink
2828 Details : sysio_path_walk() would incorrectly pass the open intent to
2829 intermediate path components.
2832 Frequency : rare (liblustre only with non-standard tuning)
2833 Bugzilla : 7201 (7350)
2834 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2835 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2836 after it has been released, only do this for referenced locks
2838 Severity : enhancement
2840 Description: specifying an (invalid) directory default stripe_size of -1
2841 would reset the directory default striping
2842 Details : stripe_size -1 was used internally to signal directory stripe
2843 removal, now use "all default" to signal dir stripe removal
2844 as a directory striping of "all default" is not useful
2847 Frequency : common for large clusters running liblustre clients
2849 Description: doing an ls when liblustre clients are running is slow
2850 Details : sending a glimpse AST to a liblustre client waits for every AST
2851 to time out, as liblustre clients will not respond. Since they
2852 cannot cache data we refresh the OST lock LVB from disk instead.
2854 Severity : enhancement
2856 Description: doing an ls at the same time as file IO can be slow
2857 Details : enqueue and other "small" requests can be blocked behind many
2858 large IO requests. Create a new OST IO portal for non-IO
2859 requests so they can be processed faster.
2862 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2864 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2865 created file with mode 000
2866 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2867 allow an ftruncate() as a non-root user to a file with mode 000.
2868 Lustre now respects this flag to disable mode checks when
2869 truncating a file owned by the user
2872 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2875 Description: Revoking locks from clients that went dead or catatonic might take
2877 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2878 cancellation of such locks instant on servers without waiting for
2879 any reply from clients. Clients drops these locks when cancel
2880 notification from server is received without replying.
2883 Frequency : liblustre-only, when liblustre client dies or becomes busy
2885 Description: Doing ls on Linux clients can take a long time with active
2887 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2888 granting such locks to it in the first place if possible. Locks
2889 are taken by proxy on the OST during the read or write and
2890 dropped immediately afterward. Add connect flags handling, do
2891 not grant locks to liblustre clients for glimpse ASTs.
2893 Severity : enhancement
2895 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2896 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2897 sequential reads and behaves badly with applications doing random
2898 reads. Improve it by reading ahead at least read region, and
2899 avoiding excessive large RPC for small reads.
2901 Severity : enhancement
2903 Description: Creating more than 1000 files for a single job may cause a load
2904 imbalance on the OSTs if there are also a large number of OSTs.
2905 Details : qos_prep_create() uses an OST index reseed value that is an
2906 even multiple of the number of available OSTs so that if the
2907 reseed happens in the middle of the object allocation it will
2908 still utilize the OSTs as uniformly as possible.
2913 Description: OST or MDS may oops in ping_evictor_main()
2914 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2915 but doesn't restart at beginning of obd_exports_timed list
2918 Severity : enhancement
2920 Description: improve by-nid export eviction on the MDS and OST
2921 Details : allow multiple exports with the same NID to be evicted at one
2922 time without re-searching the exports list.
2925 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2927 Description: MDS may oops in groups_free()
2928 Details : in rare race conditions a newly allocated group_info struct is
2929 freed again, and this can be NULL. The 2.4 compatibility code
2930 for groups_free() checked for a NULL pointer, but 2.6 did not.
2933 Frequency : common for liblustre clients doing little filesystem IO
2934 Bugzilla : 9352, 7313
2935 Description: server may evict liblustre clients accessing contended locks
2936 Details : if a client is granted a lock or receives a completion AST
2937 with a blocking AST already set it would not reply to the AST
2938 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2941 Frequency : lfs setstripe, only systems with more than 160 OSTs
2943 Description: unable to set striping with a starting offset beyond OST 160
2944 Details : llapi_create_file() incorrectly limited the starting stripe
2945 index to the maximum single-file stripe count.
2948 Frequency : LDAP users only
2950 Description: lconf did not handle in-kernel recovery with LDAP properly
2951 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2953 Severity : enhancement
2955 Description: bind OST threads to NUMA nodes to improve performance
2956 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2957 node and do their allocations there to localize memory access
2959 Severity : enhancement
2961 Description: llmount can determine client NID directly from Myrinet (GM)
2962 Details : the client NID code from gmnalnid was moved directly into
2963 llmount, removing the need to use this or specifying the
2964 client NID explicitly when mounting GM clients with zeroconf
2967 Frequency : if client is started with down MDS
2969 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2970 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2971 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2972 ptlrpc_import_delayed_req() because we didn't check if the
2973 request was interrupted, and we also didn't break out of the
2974 event loop if there was no timeout
2979 Description: data loss during non-page-aligned writes to a single file from
2980 both multiple nodes and multiple threads on one node at same time
2981 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2982 inconsistency led to false short-reads, that were cached and later
2983 used by ->prepare_write() to fill in partially written page,
2984 leading to data loss.
2987 Frequency : always, if lconf --abort_recovery used
2989 Description: lconf --abort_recovery fails with 'Operation not supported'
2990 Details : lconf was attempting to abort recovery on the MDT device and not
2993 Severity : enhancement
2995 Description: remove cleanup logs
2996 Details : replace lconf-generated cleanup logs with lustre internal
2997 cleanup routines. Eliminates the need for client-cleanup and
3000 Severity : enhancement
3002 Description: add support for EAs (user and system) on lustre filesystems
3003 Details : it is now possible to store extended attributes in the Lustre
3004 client filesystem, and with the user_xattr mount option it
3005 is possible to allow users to store EAs on their files also
3007 Severity : enhancement
3009 Description: Add possibility (config option) to show minimal available OST free
3011 Details : When compiled with --enable-mindf configure option, statfs(2)
3012 (and so, df) will return least minimal free space available from
3013 all OSTs as amount of free space on FS, instead of summary of
3014 free spaces of all OSTs.
3016 Severity : enhancement
3018 Description: do not expand extent locks acquired on OST-side
3019 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3020 by server: they are not cached anyway.
3023 Frequency : when mmap is used/binaries executed from Lustre
3025 Description: Unmmap pages before throwing them away from read cache.
3026 Details : llap_shrink cache now attempts to unmap pages before discarding
3027 them (if unmapping failed - do not discard). SLES9 kernel has
3028 extra checks that trigger if this unmapping is not done first.
3033 Description: lconf didn't resolve symlinks before checking to see whether a
3034 given mountpoint was already in use
3037 Frequency : when migrating failover services
3038 Bugzilla : 6395, 9514
3039 Description: When migrating a subset of services from a node (e.g. failback
3040 from a failover service node) the remaining services would
3041 time out and evict clients.
3042 Details : lconf --force (implied by --failover) sets the global obd_timeout
3043 to 5 seconds in order to quickly disconnect, but this caused
3044 other RPCs to time out too quickly. Do not change the global
3045 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3047 Severity : enhancement
3048 Frequency : if MDS is started with down OST
3049 Bugzilla : 9439,5706
3050 Description: Allow startup/shutdown of an MDS without depending on the
3051 availability of the OSTs.
3052 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3053 Add appropriate locking and lov-osc refcounts for safe
3054 cleaning. Add osc abort_inflight calls in case the
3055 synchronize never started.
3058 Frequency : occasional (Cray XT3 only)
3060 Description: root not authorized to access files in CRAY_PORTALS environment
3061 Details : The client process capabilities were not honoured on the MDS in
3062 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3063 been accessed by an authorized user then root was able to access
3064 the file on the local client also. The root user capabilities
3065 are now allowed on the MDS, as this environment has secure UID.
3068 Frequency : occasional
3070 Description: ldiskfs "too long searching" message happens too often
3071 Details : A debugging message (otherwise harmless) prints too often on
3072 the OST console. This has been reduced to only happen when
3073 there are fragmentation problems on the filesystem.
3078 Description: Division by zero in statfs when all OSCs are inactive
3079 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3080 lov_statfs divides by value returned by lov_get_stripecnt().
3084 Bugzilla : 9489, 3273
3085 Description: First write from each client to each OST was only 4kB in size,
3086 to initialize client writeback cache, which caused sub-optimal
3087 RPCs and poor layout on disk for the first writen file.
3088 Details : Clients now request an initial cache grant at (re)connect time
3089 and so that they can start streaming writes to the cache right
3090 away and always do full-sized RPCs if there is enough data.
3091 If the OST is rebooted the client also re-establishes its grant
3092 so that client cached writes will be honoured under the grant.
3097 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3098 Details : Now I/O RPCs go to different portal number and (presumably) fast
3099 lock requests (and glimses) and other RPCs get their own service
3100 threads pool that should be able to service those RPCs
3103 Severity : enhancement
3105 Description: Ability to exchange lustre version between client and servers and
3106 issue warnings at client side if client is too old. Also for
3107 liblustre clients there is ability to refuse connection of too old
3109 Details : New 'version' field is added to connect data structure that is
3110 filled with version info. That info is later checked by server and
3114 Frequency : rare, liblustre only.
3115 Bugzilla : 9296, 9581
3116 Description: Two simultaneous writes from liblustre at offset within same page
3117 might proceed at the same time overwriting eachother with stale
3119 Details : I/O lock withing llu_file_prwv was released too early, before data
3120 actually was hitting the wire. Extended lock-holding time until
3121 server acknowledges receiving data.
3124 Frequency : extremely rare. Never observed in practice.
3126 Description: avoid generating lustre_handle cookie of 0.
3127 Details : class_handle_hash() generates handle cookies by incrementing
3128 global counter, and can hit 0 occasionaly (this is unlikely, but
3129 not impossible, because initial value of cookie counter is
3130 selected randonly). Value of 0 is used as a sentinel meaning
3131 "unassigned handle" --- avoid it. Also coalesce two critical
3132 sections in this function into one.
3134 Severity : enhancement
3136 Description: allow liblustre clients to delegate truncate locking to OST
3137 Details : To avoid overhead of locking, liblustre client instructs OST to
3138 take extent lock in ost_punch() on client's behalf. New connection
3139 flag is added to handle backward compatibility.
3141 Severity : enhancement
3142 Bugzilla : 4928, 7341, 9758
3143 Description: allow number of OST service threads to be specified
3144 Details : a module parameter allows the number of OST service threads
3145 to be specified via "options ost ost_num_threads={N}" in the
3146 OSS's /etc/modules.conf or /etc/modprobe.conf.
3150 Bugzilla : 6146, 9635, 9895
3151 Description: servers crash with bad pointer in target_handle_connect()
3152 Details : In rare cases when a client is reconnecting it was possible that
3153 the connection request was the last reference for that export.
3154 We would temporarily drop the export reference and get a new
3155 one, but this may have been the last reference and the export
3156 was just destroyed. Get new reference before dropping old one.
3158 Severity : enhancement
3159 Frequency : if client is started with failover MDS
3161 Description: Allow multiple MDS hostnames in the mount command
3162 Details : Try to read the configuration from all specified MDS
3163 hostnames during a client mount in case the "primary"
3166 Severity : enhancement
3168 Description: Stop sending data to evicted clients as soon as possible.
3169 Details : Check if the client we are about to send or are sending data to
3170 was evicted already. (Check is done every second of waiting,
3171 for which l_wait_event interface was extended to allow checking
3172 of exit condition at specified intervals).
3175 Frequency : rare, normally only when NFS exporting is done from client
3177 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3179 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3182 Frequency : always, if srand() is called before liblustre initialization
3184 Description: Liblustre uses system PRNG disturbing its usage by user application
3185 Details : Introduce internal to lustre fast and high-quality PRNG for
3186 lustre usage and make liblustre and some other places in generic
3187 lustre code to use it.
3189 Severity : enhancement
3190 Bugzilla : 9477, 9557, 9870
3191 Description: Verify that the MDS configuration logs are updated when xml is
3192 Details : Check if the .xml configuration logs are newer than the config
3193 logs stored on the MDS and report an error if this is the case.
3194 Request --write-conf, or allow starting with --old_conf.
3196 Severity : enhancement
3198 Description: Handle symlinks in the path when checking if Lustre is mounted.
3199 Details : Resolve intermediate symlinks when checking if a client has
3200 mounted a filesystem to avoid duplicate client mounts.
3205 Description: lconf can hit an error exception but still return success.
3206 Details : The lconf command catches the Command error exception at the top
3207 level script context and will exit with the associated exit
3208 status, but doesn't ensure that this exit status is non-zero.
3213 Description: failure of ptlrpc thread startup can cause oops
3214 Details : Starting a ptlrpc service thread can fail if there are a large
3215 number of threads or the server memory is very fragmented.
3216 Handle this without oopsing.
3219 Frequency : always, only if liblustre and non-default acceptor port was used
3221 Description: liblustre cannot connect to servers with non-default acceptor port
3222 Details : tcpnal_set_default_params() was not called and was therefore
3223 ignoring the environment varaible TCPNAL_PORT, as well as other
3224 TCPNAL_ environment variables
3229 Description: two objects could be created on the same OST for a single file
3230 Details : If an OST is down, in some cases it was possible to create two
3231 objects on a single OST for a single file. No problems other
3232 than potential performance impact and spurious error messages.
3236 Bugzilla : 5681, 9562
3237 Description: Client may oops in ll_unhash_aliases
3238 Details : Client dcache may become inconsistent in race condition.
3239 In some cases "getcwd" can fail if the current directory is
3245 Description: Inode refcounting problems in NFS export code
3246 Details : link_raw functions used to call d_instantiate without obtaining
3247 extra inode reference first.
3251 Bugzilla : 9942, 9903
3252 Description: Referencing freed requests leading to crash, memleaks with NFS.
3253 Details : We used to require that call to ll_revalidate_it was always
3254 followed by ll_lookup_it. Also with revalidate_special() it is
3255 possible to call ll_revalidate_it() twice for the same dentry
3256 even if first occurence returned success. This fix changes semantic
3257 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3258 reference on a request referred from the intent.
3259 ll_intent_release() then releases such a request.
3262 Frequency : rare, normally benchmark loads only
3264 Description: unlinked inodes were kept in memory on the client
3265 Details : If a client is repeatedly creating and unlinking files it
3266 can accumulate a lot of stale inodes in the inode slab cache.
3267 If there is no other client load running this can cause the
3268 client node to run out of memory. Instead flush old inodes
3269 from client cache that have the same inode number as a new inode.
3272 Frequency : SLES9 2.6.5 kernel and long filenames only
3273 Bugzilla : 9969, 10379
3274 Description: utime reports stale NFS file handle
3275 Details : SLES9 uses out-of-dentry names in some cases, which confused
3276 the lustre dentry revalidation. Change it to always use the
3280 Frequency : rare, unless heavy write-truncate concurrency is continuous
3281 Bugzilla : 4180, 6984, 7171, 9963, 9331
3282 Description: OST becomes very slow and/or deadlocked during object unlink
3283 Details : filter_destroy() was holding onto the parent directory lock
3284 while truncating+unlinking objects. For very large objects this
3285 may block other threads for a long time and slow overall OST
3286 responsiveness. It may also be possible to get a lock ordering
3287 deadlock in this case, or run out of journal credits because of
3288 the combined truncate+unlink. Solution is to do object truncate
3289 first in one transaction without parent lock, and then do the
3290 final unlink in a new transaction with the parent lock. This
3291 reduces the lock hold time dramatically.
3294 Frequency : rare, 2.4 kernels only
3296 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3297 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3298 before unmounting their filesystem in order to avoid deadlock.
3299 The kernel_locked() function in 2.4 kernels only checks whether
3300 the kernel lock is held, not whether it is this process that is
3301 holding it as 2.6 kernels do.
3306 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3307 Details : The client ptlrpc code may be trying to reconnect to a down
3308 server before a previous connection attempt has timed out.
3309 Increase the reconnect interval to be longer than the connection
3310 timeout interval to avoid sending duplicate connections to
3314 Frequency : echo_client brw_test command
3316 Description: fix echo_client to work with OST preallocated code
3317 Details : OST preallocation code (5137) didn't take echo_client IO path
3318 into account: echo_client calls filter methods outside of any
3319 OST thread and, hence, there is no per-thread preallocated
3320 pages and buffers to use. Solution: hijack pga pages for IO. As
3321 a byproduct, this avoids unnecessary data copying.
3325 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3326 Description: Client can oops in mdc_commit_close() after open replay
3327 Details : It was possible for the MDS to return an open request with no
3328 transaction number in mds_finish_transno() if the client was
3329 evicted, but without actually returning an error. Clients
3330 would later try to replay that open and may trip an assertion
3331 Simplify the client close codepath, and always return an error
3332 from the MDS in case the open is not successful.
3335 Frequency : rare, 2.6 OSTs only
3337 Description: OST may deadlock under high load on fragmented files
3338 Details : If there was a heavy load and highly-fragmented OST filesystems
3339 it was possible to have all the OST threads deadlock waiting on
3340 allocation of biovecs, because the biovecs were not released
3341 until the entire RPC IO was completed. Instead, release biovecs
3342 as soon as they are complete to ensure forward IO progress.
3344 Severity : enhancement
3346 Description: Support for specifying external journal device at mount
3347 Details : If an OST or MDS device is formatted with an external journal
3348 device, this device major/minor is stored in the ext3 superblock
3349 and may not be valid for failover. Allow detecting and
3350 specifying the external journal at mount time.
3355 Description: Mounting an MDS with pending unlinked files may cause oops
3356 Details : target_finish_recovery() calls mds_postrecov() which returned
3357 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3358 considers this an error and immediately begins cleaning up the
3359 lov, just after starting the mds_lov process
3361 Severity : enhancement
3363 Description: Implement 'lfs df' to report actual free space on per-OST basis
3364 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3365 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3366 usage of objects; '-h' to report in human readable format.
3368 ------------------------------------------------------------------------------
3370 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3377 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3378 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3379 at startup during orphan unlinking. Instead, skip buddy update
3380 before setup, as it will be regenerated after recovery is complete.
3385 Description: If an OST is inactive, its locks might reference stale inodes.
3386 Details : lov_change_cbdata() must iterate over all namespaces, even if
3387 they are inactive to clear inode references from the lock.
3389 Severity : enhancement
3390 Frequency : occasional, if non-standard max_dirty_mb used
3392 Description: Client will block write RPCs if not enough grant
3393 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3394 then the client will block writes while waiting for another RPC
3395 to complete instead of consuming its dirty limit. With change
3396 we get improved performance when max_dirty_mb is small.
3398 Severity : enhancement
3399 Bugzilla : 3389, 6253
3400 Description: Add support for supplementary groups on the MDS.
3401 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3402 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3403 lookups for user supplementary groups into a cache.
3407 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3408 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3409 for liblustre O_EXCL is enforced by MDS.
3413 Description: readdir never returns NULL in liblustre.
3414 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3415 dentry in struct dirent.
3419 Frequency : liblustre only, depends on application IO pattern
3420 Description: liblustre clients evicted if not contacting servers
3421 Details : Don't put liblustre clients into the ping_evictor list, so
3422 they will not be evicted by the pinger ever.
3424 Severity : enhancement
3426 Description: Add ability to evict clients by NID from MDS.
3427 Details : By echoing "nid:$NID" string into
3428 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3429 $NID would be instantly evicted from this MDS and from all active
3430 OSTs connected to it.
3434 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3435 Details : lookup_it_finish() used to query file size from OSTs that was not
3440 Description: service threads change working directory to that of init
3441 Details : Starting lustre service threads may pin the working directory
3442 of the parent thread, making that filesystem busy. Threads
3443 now change to the working directory of init to avoid this.
3447 Frequency : during shutdown only
3448 Description: shutdown with a failed MDS or OST can cause unmount to hang
3449 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3454 Frequency : 2.6 only
3455 Description: chmod/chown may include an extra supplementary group
3456 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3457 supplementary group and if none is specified this is used.
3460 Bugzilla : 5479 (6816)
3462 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3463 Details : If lookup is in progress on a file that is unlinked we might try
3464 to revalidate the inode and fail in revalidate after lookup is
3465 complete and ll_file_open() enqueues the open again but
3466 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3469 Frequency : always, if lconf --abort_recovery used
3471 Description: lconf --abort_recovery fails with 'Operation not supported'
3472 Details : lconf was attempting to abort recovery on the MDT device and not
3475 ------------------------------------------------------------------------------
3477 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3482 Frequency : rare (only unsupported configurations with a node running as an
3484 Bugzilla : 6514, 5137
3485 Description: Mounting a Lustre file system on a node running as an OST could
3487 Details : OSTs now preallocates memory needed to write out data at
3488 startup, instead of when needed, to avoid having to
3489 allocate memory in possibly low memory situations.
3490 Specifically, if the file system is mounted on on OST,
3491 memory pressure could force it to try to write out data,
3492 which it needed to allocate memory to do. Due to the low
3493 memory, it would be unable to do so and the node would
3494 become unresponsive.
3496 Severity : enhancement
3498 Description: Addition of lconf --service command line option
3499 Details : lconf now accepts a '--service <arg>' option, which is
3500 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3502 Severity : enhancement
3504 Description: Failover mode is now the default for OSTs.
3505 Details : By default, OSTs will now run in failover mode. To return to
3506 the old behaviour, add '--failout' to the lmc line for OSTs.
3508 Severity : enhancement
3510 Description: Health checks are now provided for MDS and OSTs
3511 Details : Additional detailed health check information on MSD and OSTs
3512 is now provided through the procfs health_check value.
3515 Frequency : occasional, depends on IO load
3517 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3518 after numerous create/delete cycles
3519 Details : The ext3 inode allocation policy would not allocate new inodes
3520 very well on the OSTs because there are no new directories
3521 being created. Instead we look for groups with free space if
3522 the parent directories are nearly full.
3527 Description: Network or server problems during mount may cause partially
3528 mounted clients instead of returning an error.
3529 Details : The config llog parsing code may overwrite the error return
3530 code during mount error handling, returning success instead
3536 Description: MDS can fail to allocate large reply buffers
3537 Details : After long uptimes the MDS can fail to allocate large reply
3538 buffers (e.g. zconf client mount config records) due to memory
3539 fragmentation or consumption by the buffer cache. Preallocate
3540 some large reply buffers so that these replies can be sent even
3541 under memory pressure.
3545 Frequency : rare (liblustre)
3546 Description: fsx running with liblustre complained that using truncate() to
3547 extend the file doesn't work. This patch corrects that issue.
3548 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3549 ATTR_SIZE and lsm use in llu_setattr_raw.
3553 Frequency : rare, only 2.6 kernels
3554 Description: Unusual file access patterns on the MDS may result in inode
3555 data being lost in very rare circumstances.
3556 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3557 code in the 2.6 kernel caused the inode and/or EA data not to
3558 be read from disk, causing single-file corruption.
3562 Frequency : rare, only 2.6 filesystems using extents
3563 Description: Heavy concurrent write and delete load may cause data corruption.
3564 Details : It was possible under high-load situations to have an extent
3565 metadata block in the block device cache from a just-unlinked
3566 file overwrite a newly-allocated data block. We now unmap any
3567 metadata buffers that alias just-allocated data blocks.
3571 Frequency : filesystems with default stripe_count larger than 77
3572 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3573 Details : lconf specifies an inode size of 4096 bytes when the default
3574 stripe_count is larger than 77. This conflicts with the default
3575 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3576 this case to avoid pinning too much memory for large EAs.
3578 ------------------------------------------------------------------------------
3580 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3585 Frequency : rare (extremely heavy IO load with hundreds of clients)
3587 Description: Client is evicted, gets IO error writing to file
3588 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3589 caused clients to be evicted for AST timeouts. The fixes in
3590 bug 5192 mean we no longer need to have such short AST timeouts
3591 so ldlm_timeout has been increased.
3594 Frequency : occasional during --force or --failover shutdown under load
3595 Bugzilla : 5949, 4834
3596 Description: Server oops/LBUG if stopped with --force or --failover under load
3597 Details : a collection of import/export refcount and cleanup ordering
3598 issues fixed for safer force cleanup
3601 Frequency : only filesystems larger than 120 OSTs
3602 Bugzilla : 5990, 6223
3603 Description: lfs getstripe would oops on a very large filesystem
3604 Details : lov_getconfig used kfree on vmalloc'd memory
3607 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3608 Bugzilla : 6242, 6243
3609 Description: reading from files that had been truncated to a non-zero size
3610 but never opened returned no data
3611 Details : ll_file_read() reads zeros from no-object files to EOF
3616 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3617 Details : There's little harm in aborting MDS/OSS recovery and letting it
3618 try again, so I removed the LASSERT and return an error instead.
3620 Severity : enhancement
3622 Description: New debugging infrastructure for tracking down data corruption
3623 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3624 (b) cover more of the client-side code path, and (c) try to narrow
3625 down where problems occurred
3629 Bugzilla : 3819, 4364, 4397, 6313
3630 Description: Racing close and eviction MDS could cause assertion in mds_close
3631 Details : It was possible to get multiple mfd references during close and
3632 client eviction, leading to one thread referencing a freed mfd.
3634 Severity: : enhancement
3635 Bugzilla : 3262, 6359
3636 Description: Attempts to reconnect to servers are now more aggressive.
3637 Details : This builds on the enhanced upcall-less recovery that was added
3638 in 1.4.2. When trying to reconnect to servers, clients will
3639 now try each server in the failover group every 10 seconds. By
3640 default, clients would previously try one server every 25 seconds.
3645 Description: After recovery, certain operations trigger a failed
3646 assertion on a client.
3647 Details : Failing over an mds, using lconf -d --failover, while a
3648 client was doing a readdir() call would cause the client to
3649 LBUG after recovery completed and the readdir() was resent.
3651 Severity : enhancement
3653 Description: Default groups are now added by lconf
3654 Details : You can now run lconf --group <servicename> without having to
3655 manually add groups with lmc.
3658 Frequency : occasional
3660 Description: Nodes with an elan id of 0 trigger a failed assertion
3663 Frequency : always when accessing e.g. tty/console device nodes
3665 Description: tty and some other devices nodes cannot be used on lustre
3666 Details : file's private_data field is used by device data and lustre
3667 values in there got lost. New field was added to struct file to
3668 store fs-specific private data.
3671 Frequency : when exporting Lustre via NFS
3673 Description: NFSD failed occasionally when looking up a path component
3674 Details : NFSD is looking up ".." which was broken in ext3 directories
3675 that had grown large enough to become hashed.
3678 Frequency : Clusters with multiple interfaces not on the same subnet
3680 Description: Nodes will repeatedly try to reconnect to an interface which it
3681 cannot reach and report an error to the log.
3682 Details : Extra peer list entries will be created by lconf with some peers
3683 unreachable. lconf now validates the peer before adding it.
3686 Frequency : Only if a default stripe is set on the filesystem root.
3688 Description: Setting a default stripe on the filesystem root prevented the
3689 filesystem from being remounted.
3690 Details : The client was sending extra request flags in the root getattr
3691 request and did not allocate a reply buffer for the dir EA.
3694 Frequency : occasional, higher if lots of files are accessed by one client
3695 Bugzilla : 6159, 6097
3696 Description: Client trips assertion regarding lsm mismatch/magic
3697 Details : While revalidating inodes the VFS looks up inodes with ifind()
3698 and in rare cases can find an inode that is being freed.
3699 The ll_test_inode() code will free the lsm during ifind()
3700 when it finds an existing inode and then the VFS later attaches
3701 this free lsm to a new inode.
3705 Bugzilla : 6422, 7030
3706 Description: MDS deadlock between mkdir and client eviction
3707 Details : Creating a new file via mkdir or mknod (starting a transaction
3708 and getting the ns lock) can deadlock with client eviction
3709 (gets ns lock and trying to finish a synchronous transaction).
3712 Frequency : occasional
3713 Description: While starting a server, the fsfilt_ext3 module could not be
3715 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3716 kernels. Previously, lconf would still use the ext3 name
3717 when trying to load modules. Now, it will correctly use
3718 ext3 on 2.4 and ldiskfs on 2.6.
3720 Severity : enhancement
3721 Description: The default stripe count has been changed to 1
3722 Details : The interpretation of the default stripe count (0, to lfs
3723 or lmc) has been changed to mean striping across a single
3724 OST, rather than all available. For general usage we have
3725 found a stripe count of 1 or 2 works best.
3727 Severity : enhancement
3728 Description: Add support for compiling against Cray portals.
3729 Details : Conditional compiling for some areas that are different
3733 Frequency : occasional
3734 Bugzilla : 6409, 6834
3735 Description: Creating files with an explicit stripe count may lead to
3736 a failed assertion on the MDS
3737 Details : If some OSTs are full or unavailable, creating files may
3738 trigger a failed assertion on the MDS. Now, Lustre will
3739 try to use other servers or return an error to the
3743 Frequency : occasional
3745 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3746 caused lock timeout during readahead (since 1.4.2).
3747 Details : Processes doing readahead might match a lock that hasn't been
3748 granted yet if there are overlapping and conflicting lock
3749 requests. The readahead process waits on ungranted lock
3750 (original lock is CBPENDING), while OST waits for that process
3751 to cancel CBPENDING read lock and eventually evicts client.
3753 Severity : enhancement
3755 Description: Initial enabling of flock support for clients
3756 Details : Implements fcntl advisory locking and file status functions.
3757 This feature is provided as an optional mount flag (default
3758 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3759 locking are implemented yet, and those that are are not guaranteed
3760 to be completely correct in production environments.
3761 mount -t lustre -o [flock|noflock] ...
3764 Frequency : occasional
3766 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3767 assertion in the ext3 JBD (journaling) layer.
3768 Details : The b_committed_data struct is protected by the big kernel lock
3769 in 2.4 kernels, serializing journal_commit_transaction() and
3770 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3771 access to this struct. In 2.6 kernels there is finer grained
3772 locking to improve SMP performance of the JBD layer.
3776 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3781 Description: merge and cleanup kernel patches.
3784 -----------------------------------------------------------------------------
3786 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3788 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3789 versions of Lustre. Please update all servers and clients to
3790 version 1.4.2 or later at the same time. You must also run
3791 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3792 to update the configuration logs.
3794 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3795 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3796 - swab llog records of type '0' so we get proper header size/idx (5861)
3797 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3798 - fix rename of one directory over another leaking an inode (5953)
3799 - avoid SetPageDirty on 2.6 (5981)
3800 - don't re-add just-being-destroyed locks to the waiting list (5653)
3801 - when creating new directories, inherit the parent's custom
3802 striping settings if present parent (3048)
3803 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3804 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3805 - don't hold client locks on temporary worklist from l_lru (5666)
3806 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3807 - automatically evict dead clients (5921)
3808 - Update file size properly in create+truncate+fstat case (6196)
3809 - Do not unhash mountpoint dentries, do not allow removal of
3811 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3812 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3813 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3814 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3815 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3816 - Correctly handle reads of files with no objects (6243)
3817 - lctl recover will also mark a device active if deactivate used (5933)
3819 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3820 - allow --write-conf on an MDS with different nettype than client (5619)
3821 - don't write config llogs to MDS for mounts not from that MDS (5617)
3822 - lconf should create multiple TCP connections from a client (5201)
3823 - init scripts are now turned off by default; run chkconfig --on
3824 lustre and chkconfig --on lustrefs to use them
3825 - upcalls are no longer needed for clients to recover to failover
3827 - add --abort-recovery option to lconf to abort recovery on device
3829 - add support for an arbitrary number of OSTs (3026)
3830 - Quota support protocol changes.
3831 - forward compatibility changes to wire structs (6007)
3832 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3833 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3834 - improved "lustre status" script
3835 - initialize blocksize for non-regular files (6062)
3836 - added --disable-server and --disable-client configure options (5782)
3837 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3838 - Vanilla 2.4.29 support
3839 - increase maximum number of obd devices to 520 (6242)
3840 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3841 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3842 SLES 9 2.6.5 kernels
3843 - limit stripes per file to 160 (the maximum EA size) (6093)
3845 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3848 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3849 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3850 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3851 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3852 - small ext3 extents cleanups and fixes (5733)
3853 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3854 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3855 - avoid needless client->OST connect, fix handle mismatch (5317)
3856 - fix DLM error path that led to out-of-sync client, long delays (5779)
3857 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3858 - fix several locking issues related to i_size (5492,5624,5654,5672)
3859 - don't move pending lock onto export if it is already evicted (5683)
3860 - fix kernel oops when creating .foo in unlinked directory (5548)
3861 - fix deadlock in obdfilter statistics vs. object create (5811)
3862 - use time_{before,after} to avoid timer jiffies wrap (5882)
3863 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3864 - Do not leak request if server was not able to process it (5154)
3865 - If mds_open unable to find parent dir, make that negative lookup(5154)
3866 - don't create new directories with extent-mapping (5909, 5936)
3868 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3869 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3870 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3871 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3872 - support for RHEL4 kernel on i686 (5773)
3873 - provide error messages when incompatible logs are encountered (5898)
3875 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3876 * version 1.4.0.10 (1.4.1 release candidate 1)
3878 - don't keep a lock reference when lock is not granted (4238)
3879 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3880 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3881 - drop import inflight refcount on signal_completed_replay error (5255)
3882 - unlock page after async write error during send (3677)
3883 - handle missing objects in filter_preprw_read properly (5265)
3884 - no transno return for symlink open, don't save no-trasno open (3440)
3885 - don't try to complete elan receive that already failed (4012)
3886 - free RPC server reply state on error (5406)
3887 - clean up thread from ptlrpc_start_thread() on error (5160)
3888 - readahead could read extra page into cache that wasn't ejected (5388)
3889 - prevent races in class_attach/setup/cleanup/detach (5260)
3890 - don't dereference de->d_inode after l_dput of de (5458)
3891 - use "int" for stripe value returned from lock_to_stripe (5544)
3892 - mballoc allocation and error-checking fixes in 2.6 (5504)
3893 - block device patches to fix I/O request sizes in 2.6 (5482)
3894 - look up hostnames for IB nals (5602)
3895 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3896 - don't start multiple acceptors for the same port (5277)
3897 - fix incorrect LASSERT in mds_getattr_name (5635)
3898 - export a proc file for general "ping" checking (5628)
3899 - fix "lfs check" to not block when the MDS is down (5628)
3901 - service request history (4965)
3902 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3903 - create an "evict_client" /proc entry on OSTs, like the MDS has
3904 - fix mount usage message, return errors per mount(8) (5168)
3905 - change grep [] to grep "[]" in tests so they work in more UMLs
3906 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3907 - remove ancient LOV_MAGIC_V0 EA support (5047)
3908 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3909 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3910 - allow readahead allocations to fail when low on memory (5383)
3911 - mmap locking landed again, after considerable improvement (2828)
3912 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3914 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3917 - send OST transaction number in read/write reply to free req (4966)
3918 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3919 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3920 - the watchdog thread now runs as interruptible (5246)
3921 - flock/lockf fixes (but it's still disabled, pending 5135)
3922 - don't use EXT3 constants in llite code (5094)
3923 - memory shortage at startup could cause assertion (5176)
3925 - reorganization of lov code
3926 - single portals codebase
3928 - add extents/mballoc support (5025)
3929 - direct I/O reads in the obdfilter (4048)
3930 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3932 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3935 - send OST transaction number in read/write reply to free req (4966)
3936 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3937 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3938 - the watchdog thread now runs as interruptible (5246)
3939 - handle missing objects in filter_preprw_read properly (5265)
3940 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3941 - drop import inflight refcount on signal_completed_replay error (5255)
3942 - unlock page after async write error during send (3677)
3943 - return original error code on reconstructed replies (3761)
3944 - no transno return for symlink open, don't save no-trasno open (3440)
3946 - add pid to ldlm debugging output (4922)
3947 - bump the watchdog timeouts -- we can't handle 30sec yet
3948 - extra debugging for orphan dentry/inode bug (5259)
3950 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3953 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3954 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3955 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3956 - fix NULL dereference in /proc/sys/portals/routes (4827)
3957 - allow failed mdc_close() operations to be interrupted (4561)
3958 - stop precreate on OST before MDS would time out on it (4778)
3959 - don't send partial-page writes before EOF from client (4410)
3960 - discard client grant for sub-page writes on large-page clients (4520)
3961 - don't free dentries not owned by NFS code, check generation (4806)
3962 - fix lsm leak if mds_create_objects() fails (4801)
3963 - limit debug_daemon file size, always print CERROR messages (4789)
3964 - use transno after validating reply (3892)
3965 - process timed out requests if import state changes (3754)
3966 - update mtime on OST during writes, return in glimpse (4829)
3967 - add mkfsoptions to LDAP (4679)
3968 - use ->max_readahead method instead of zapping global ra (5039)
3969 - don't interrupt __l_wait_event() during strace
3971 - add software watchdogs to catch hung threads quickly (4941)
3972 - make lustrefs init script start after nfs is mounted
3973 - fix CWARN/ERROR duplication (4930)
3974 - return async write errors to application if possible (2248)
3975 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3976 - print NAL number in %x format (4645)
3977 - update barely-supported suse-2.4.21-171 series (4842)
3978 - support for sles 9 %post scripts
3979 - support for building 2.6 kernel-source packages
3980 - support for sles km_* packages
3982 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3985 - ignore -ENOENT errors in osc_destroy (3639)
3986 - notify osc create thread that OSC is being cleaned up (4600)
3987 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3988 - reconstruct ost_handle() like mds_handle() (4657)
3989 - create a new thread to do import eviction to avoid deadlock (3969)
3990 - let lconf resolve symlinked-to devices (4629)
3991 - don't unlink "objects" from directory with default EA (4554)
3992 - hold socknal file ref over connect in case target is down (4394)
3993 - allow more than 32000 subdirectories in a single directory (3244)
3994 - fix blocks count for O_DIRECT writes (3751)
3995 - OST returns ENOSPC from object create when no space left (4539)
3996 - don't send truncate RPC if file size isn't changing (4410)
3997 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3998 - bind to privileged port in socknal and tcpnal (3689)
4000 - rate limit CERROR/CWARN console message to avoid overload (4519)
4001 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4002 - basic mmap support (3918)
4003 - kernel patch series update from b1_4 (4711)
4005 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4008 - avoid crash during MDS cleanup with OST shut down (2775)
4009 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4010 - don't use bad inodes on the MDS (3744)
4011 - dynamic object preallocation to improve recovery speed (4236)
4012 - don't hold spinlock over lock dumping or change debug flags (4401)
4013 - don't zero obd_dev when it is force cleaned (3651)
4014 - print grants to console if they go negative (4431)
4015 - "lctl deactivate" will stop automatic recovery attempts (3406)
4016 - look for existing locks in ldlm_handle_enqueue() (3764)
4017 - don't resolve lock handle twice in recovery avoiding race (4401)
4018 - revalidate should check working dir is a directory (4134)
4020 - don't always mark "slow" obdfilter messages as errors (4418)
4022 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4025 - don't close LustreDB during write_conf until it is done (3860)
4026 - fix typo in lconf for_each_profile (3821)
4027 - allow dumping logs from multiple threads at one time (3820)
4028 - don't allow multiple threads in OSC recovery (3812)
4029 - fix debug_size parameters (3864)
4030 - fix mds_postrecov to initialize import for llog ctxt (3121)
4031 - replace config semaphore with spinlock (3306)
4032 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4033 - don't allow enqueue to complete on a destroyed export (3822)
4034 - down write_lock before checking llog header bitmap (3825)
4035 - recover from lock replay timeout (3764)
4036 - up llog sem before sending rpc (3652)
4037 - reduce ns lock hold times when setting kms (3267)
4038 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4039 - fix NULL deref and obd_dev leak on setup error (3312)
4040 - replace some LBUG about llog ops with error handling (3841)
4041 - don't match INVALID dentries from d_lookup and spin (3784)
4042 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4043 - fix invalid assertion in ptlrpc_set_wait (3880)
4045 - add libwrap support for the TCP acceptor (3996)
4046 - add /proc/sys/portals/routes for non-root route listing (3994)
4047 - allow setting MDS UUID in .xml (2580)
4048 - print the stack of a process that LBUGs (4228)
4050 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4053 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4054 - make sure to unset replay flag from failed open requests (3440)
4055 - if default stripe count is 0, use OST count for inode size (3636)
4056 - update parent mtime/ctime on client for create/unlink (2611)
4057 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4058 - free recovery state on server during a forced cleanup (3571)
4059 - unregister_reply for resent reqs (3063)
4060 - loop back devices mounting and status check on 2.6 (3563)
4061 - fix resource-creation race that can provoke i_size == 0 (3513)
4062 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4063 - more debugging for page-accounting assertion (3746)
4064 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4065 - avoid deadlock after precreation failure (3758)
4066 - fix race and lock order deadlock in orphan handling (3450, 3750)
4067 - add validity checks when grabbing inodes from l_ast_data (3599)
4069 - add /proc/.../recovery_status to obdfilter (3428)
4070 - lightweight CDEBUG infrastructure, debug daemon (3668)
4071 - change default OSC RPC parameters to be better on small clusters
4072 - turn off OST read cache for files smaller than 32MB
4073 - install man pages and include them in rpms (3100)
4074 - add new init script for (un)mounting lustre filesystems (2593)
4075 - run chkconfig in %post for init scripts (3701)
4076 - drop scimac NAL (unmaintained)
4078 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4081 - clean kiobufs before and after use (3485)
4082 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4083 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4084 - change init script to fail more clearly if not run as root (1528)
4085 - allow clients to reconnect during replay (1742)
4086 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4087 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4088 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4089 - fix cond_resched (3554)
4090 - only evict unfinished clients after recovery (3515)
4091 - allow bulk resend, prevent data loss (3570)
4092 - dynamic ptlrpc request buffer allocation (2102)
4093 - don't allow unlinking open directory if it isn't empty (2904)
4094 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4095 - remove extraneous obd dereference causing LASSERT failure (3334)
4096 - don't use get_cycles() when creating temp. files on the mds (3156)
4097 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4098 - handle EEXIST for set-stripe, set proper directory name (3336)
4100 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4101 - fix ksocknal_fmb_callback() error messages (2918)
4103 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4106 - don't copy lvb into (possibly NULL) reply on error (2983)
4107 - don't deref dentry after dput, don't free lvb on error (2922)
4108 - use the kms to determine writeback rpc length (2947)
4109 - increment oti_logcookies when osc is inactive (2948)
4110 - update client's i_blocks count via lvb messages (2543)
4111 - handle intent open/close of special files properly (1557)
4112 - mount MDS with errors=remount-ro, like obdfilter (2009)
4113 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4114 - don't use cancelling-locks' kms values (2947)
4115 - use highest lock extent for kms, not last one (2925)
4116 - don't dereference ERR_PTR() dentry in error handling path (3107)
4117 - fix thread race in portals_debug_dumplog() (3122)
4118 - create lprocfs device entries at setup instead of at attach (1519)
4119 - common AST error handler, don't evict client on completion race (3145)
4120 - zero nameidata in detach_mnt in 2.6 (3118)
4121 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4122 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4123 - initialize RPC timeout timer earlier for 2.6 (3219)
4124 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4125 - print nal/nid for unknown nid (3258)
4126 - additional checks for oscc recovery before doing precreate (3284)
4127 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4128 - don't crash in mdc_close for bad permissions on open (3285)
4129 - zero i_rdev for non-device files (3147)
4130 - clear page->private before handing to FS, better assertion (3119)
4131 - tune the read pipeline (3236)
4132 - fix incorrect decref of invalidated dentry (2350)
4133 - provide read-ahead stats and refine rpc in flight stats (3328)
4134 - don't hold journal transaction open across create RPC (3313)
4135 - update atime on MDS at close time (3265)
4136 - close LDAP connection when recovering to avoid server load (3315)
4137 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4138 - don't leak open file on MDS after open resend (3325)
4139 - serialize filter_precreate and filter_destroy_precreated (3329)
4140 - loop device shouldn't call sync_dev() for nul device (3092)
4141 - clear page cache after eviction (2766)
4142 - resynchronize MDS->OST in background (2824)
4143 - refuse to mount the same filesystem twice on same mountpoint (3394)
4144 - allow llmount to create routes for mounting behind routers (3320)
4145 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4146 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4147 - fix rare problem with rename on htree directories (3417)
4149 - allow default OST striping configuration per directory (1414)
4150 - fix compilation for qswnal for 2.6 kernels (3125)
4151 - increase maximum number of MDS request buffers for large systems
4152 - change liblustreapi to be useful for external progs like lfsck (3098)
4153 - increase local configuration timeout for slow disks (3353)
4154 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4156 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4159 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4160 - don't overwrite extent policy data in reply if lock was blocked (2901)
4161 - drop filter export grants atomically with removal from device (2663)
4162 - del obd_self_export from work_list in class_disconnect_exports (2908)
4163 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4164 - swab reply message in mdc_close, other PPC fixes (2464)
4165 - fix destroying of named logs (2325)
4166 - overwrite old logs when running lconf --write_conf (2264)
4167 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4168 - fix race in target_handle_connect (2898)
4169 - mds_reint_create() should take same inode create lock (2926)
4170 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4171 - don't close files for self_export to avoid uninitialized obd (2936)
4172 - allow MDS with the same name as client node (2939)
4173 - hold dentry reference for closed log files for unlink (2325)
4174 - reserve space for all logs during transactions (2059)
4175 - don't evict page beyond end of stripe extent (2925)
4176 - don't oops on a deleted current working directory (2399)
4177 - handle hard links to targets without a parent properly (2517)
4178 - don't dereference NULL lock when racing during eviction (2867)
4179 - don't grow lock extents when lots of conflicting locks (2919)
4181 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4184 - account for cache space usage on clients to avoid data loss (974)
4185 - lfsck support in lustre kernel code (2349)
4186 - reduce journal credits needed for BRW writes (2370)
4187 - orphan handling to avoid losing space on client/server crashes
4188 - ptlrpcd can be blocked, stopping ALL progress (2477)
4189 - use lock value blocks to assist in proper KMS, faster stat (1021)
4190 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4191 - recovery for initial connections (2355)
4192 - fixes for mds_cleanup_orphans (1934)
4193 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4194 - block all file creations until orphan recovery completes (1901)
4195 - client remove rq_connection from request struct (2423)
4196 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4197 - recovery timer race (2670)
4198 - mdc_close recovey bug (2532)
4199 - ptlrpc cleanup bug (2710)
4200 - mds timeout on local locks (2588)
4201 - namespace lock held during RPCs (2431)
4202 - handle interrupted sync write properly (2503)
4203 - don't try to handle a message that hasn't been replied to (2699)
4204 - client assert failure during cleanup after abort recovery (2701)
4205 - leak mdc device after failed mount (2712)
4206 - ptlrpc_check_set allows timedout requests to complete (2714)
4207 - wait for inflight reqs when ptlrpcd finishes (2710)
4208 - make sure unregistered services are removed from the srv_list
4209 - reset bulk XID's when resending them (caught by 1138 test)
4210 - unregister_bulk after timeout
4211 - fix lconf error (2694)
4212 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4213 - readahead locks pages, leaves pending causing memory pressure (2673)
4214 - increase OST request buffers to 4096 on large machines (2729)
4215 - fix up permission of existing directories in simple_mkdir (2661)
4216 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4217 - don't assign transno to errored transactions (2742)
4218 - don't delete objects on OST if given a bogus objid from MDS (2751)
4219 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4220 - if rq_no_resend, then timeout request after recovery (2432)
4221 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4222 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4223 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4224 - MDS recovery shouldn't delete objects that it has given out (2730)
4225 - if enqueue arrives after completion, don't clobber LVB (2819)
4226 - don't unlock pages twice when trigger_group_io returns error (2814)
4227 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4228 - don't write pages to disk if there was an error (1450)
4229 - don't ping imports that have recovery disabled (2676)
4230 - take buffered bytes into account when balancing socknal conn (2817)
4231 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4232 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4233 - remove little-endian swabbing of llog records (1987)
4234 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4235 - retry reposting request buffers if they fail (1191)
4236 - grow extent at grant time to avoid granting a revoked lock (2809)
4237 - lock revoke doesn't evict page if covered by a second lock (2765)
4238 - disable VM readahead to avoid reading outside lock extents (2805)
4240 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4241 - updated kernel patches for hp-2.4.20 kernel (2681)
4243 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4246 - fix truncated write corruption (2366)
4247 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4249 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4250 - ensure len(uuid) < 37 in lmc (1171)
4251 - fix ia64 OOPS in llog_test (2255)
4252 - zero end of page at obdfilter for partial page writes (2648)
4253 - don't leave stale dentries around after renames (bug 2428)
4254 - fix timeouts when evicting a client with a single lock held (2642)
4255 - set deadline for the initial HELLO message to drain (2634)
4256 - print out dotted-quad IP addresses in the socknal (2302)
4258 - additional debugging for MDS client eviction problem (2443)
4259 - fix mkfsoptions support for osts (2603, 2604)
4261 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4264 - add series for the vanilla 2.6.0 kernel
4265 - add series for the vanilla 2.4.24 kernel
4266 - add series for a cray x86/64 UL kernel drop
4267 - fix xattr patches for the vanilla 2.4.19 series
4269 - generate true UUIDs in lmc (1171)
4270 - have portals stack dumping break in UML (2466)
4271 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4272 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4273 - wake callback waiting threads on client eviction (2460)
4274 - Add --ptldebug and --subsystem to lmc (1719)
4275 - update assertion to allow safe interrupt allocation
4276 - set rq_no_resend for cancel requests (2432)
4277 - recalculate ptlrpcd timeout after resend (2494)
4278 - call vfs_rmdir when removing pending directories (2368)
4279 - fix renaming a file to itself (2429)
4280 - lmc creates a default one-stripe lov (2454)
4281 - expand procfs space to handle large clusters (2326)
4282 - increase UML stack to avoid overflow
4283 - update lconf's list of debug and subsystem masks
4284 - fix lfs find --obd (2510)
4285 - /proc tunable for disabling filter read caching (2591)
4286 - stop rpm packages from altering slapd.conf (2301)
4287 - disable nagle in the socknal under 0conf (2578)
4288 - choose mds inode size based on stripe count (2572)
4289 - fix kernel-source rpm problems (2516)
4291 - add --disable-doc to avoid pdf generation (2421)
4292 - update documentation, tests, type-os, comments
4293 - avoid format warnings on ia64
4294 - remove the TOE NAL
4295 - tiny code cleanups by removing unused fields
4297 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4300 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4301 - remove the most blatant lies from BUILDING file (2371)
4302 - change default debug level to reasonable production setting
4303 - reduce client side cache size to reduce cache flush time
4304 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4305 - make TCP zerocopy and pinger support enabled by default (2476)
4306 - sync writes completed after process exits caused crashes (2319)
4307 - maintain correct mount count on the MDS (2356)
4308 - backout 1557, because 2316 wasn't really fixed
4309 - better file I/O statistics gathering in /proc
4310 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4311 - another kernel patch to fix zero-copy TCP function export
4312 - don't take duplicate lock when processing re-sent getattr (2420)
4313 - lctl uses obd_self_export instead of creating new conn (2353)
4314 - MDS/OST recovery case which requires object creation asserted (2425)
4315 - move lfs from /usr/sbin to /usr/bin in packages
4316 - fix race between mds_client_add and mds_client_free (2417)
4317 - use kmalloc instead of slabs in portals (2430)
4318 - don't create duplicate records when a failover MDS is present (2442)
4319 - remove unnecessary mount age check (2332)
4320 - don't remove directory inodes from locks prematurely (2451)
4321 - don't break if MDS service name is the same as hostname (2103)
4322 - fix races in client write RPC generation when cache full (2482)
4324 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4327 - remove now-unused request->rq_obd (278)
4328 - if an allocation fails, print out how much memory we've used (1933)
4329 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4330 - add the "configurable stack size" patch to most series files (1256)
4331 - ability to write large log records, for 100+ OST configs (2306)
4332 - fix NULL deref when filter_prep fails (2314)
4333 - fix operator precedence error in filter_sync
4334 - dynamic allocation of socknal TX descriptors (2315)
4335 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4336 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4337 - ensure that CWARN messages in Portals always get to the syslog
4338 - __init/__exit are not for prototype decls (ldlm_init/exit)
4339 - x86-64 compile warning fixes
4340 - fix gateway LMC keyword conflict (2318)
4341 - fix MDS lock inversions in getattr/reint paths (1844)
4342 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4343 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4344 - DLM race condition prevented some lock evictions (2328)
4345 - ENOMEM detection and retry on socknal sends (2230)
4346 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4347 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4348 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4349 - better error messages when a client is rejected during recovery (1505)
4350 - avoid cancelling locks which were never granted, after failure (2330)
4351 - fix i_sem/journal inversion in mds_client_add (2333)
4352 - fix truncate/getattr lock cycle deadlock (2334)
4353 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4354 - fix two rare exit paths which could leak an l_lock() ref (2321)
4355 - fencepost error in MDS/OST orphan recovery (2226)
4356 - make log record alignment 8 bytes (1988)
4357 - lstripe now fails when requested offset > ost_count (2237)
4358 - ensure that all kernel series have a complete list.h (1607)
4359 - fix crashes in special-file operations (2316)
4360 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4362 - allow configurable automake binary, for testing new versions
4363 - small update to the lfs documentation
4365 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4367 * fix negative export reference count in fsfilt_sync (2312)
4369 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4370 * release candidate 0.9.1
4372 - orphans are moved into the PENDING directory for possible recovery
4373 - replayed opens now open by fid for orphan/rename safety (1042)
4374 - last close of an orphan inode generates a transno (683)
4375 - chdir() and mount() now pin the directory entry (1020)
4376 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4377 - discard very old requests without processing them (1502)
4378 - remove some common, well-understood CERRORs (1505)
4379 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4380 - clear "grant" flags in OST replies until OST grant code lands (1644)
4381 - fix read performance by not clobbering i_blksize on client (1598)
4382 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4383 - make LRU size a /proc tunable, clears locks when reduced (707)
4384 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4385 - server locks take references on exports now (1558)
4386 - build fixes for 2.4.20-rh trees (1663)
4387 - return an error from lov_create if all OSCs are inactive (1751)
4388 - fix import levels when a reconnect happens without a timeout (1597)
4389 - exit early from mds_open if we get a lookup error (1749)
4390 - partial page read at EOF wouldn't wait for disk before sending (1642)
4391 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4392 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4393 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4394 - write/truncate lock inversion (1639)
4395 - Don't auto-load obdclass, portals modules during cleanup (1495)
4396 - fix timestamps from jumping to "now" (1763)
4397 - extra journal assertions (1648)
4398 - add an extra multiunlink test (1771)
4399 - fix read_record/write_record API (1776)
4400 - fix leak of offset_extent, possible incorrect i_size later (1772)
4401 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4402 - Debugging for the kqswnal_get_idle_tx problems (1820)
4403 - Allow recovery to be attempted multiple times (1536)
4404 - Write out MDS last_rcvd file after it is first created (1600)
4405 - Fix tx_descriptor leak in failed transmit situations (1827)
4406 - ext3 journaling fixes for assertion failure after IO error (1871)
4407 - class_export_put() on freed export after completion AST error (1896)
4408 - Fix revalidate looping in VFS (1322)
4409 - Don't access a freed export during MDS_REINT timeout (1521)
4410 - Add open-unlink recovery support on the MDS (1673,1764)
4411 - Return an error if no MDS data was read from last_rcvd (1946)
4412 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4413 - Don't setup a disk device that doesn't match exported UUID (317)
4414 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4415 - avoid committing NULL handle in force close
4416 - local.sh is now a one-stripe LOV configuration
4417 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4418 - fix ext3 htree duplicate directory entry corruption (1516)
4419 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4420 - update to new LOV EA format (2097)
4421 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4423 2003-06-15 Phil Schwan <phil@clusterfs.com>
4426 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4427 - per-import/export recovery handling (958, 931, 959)
4428 - multiple last-rcvd slots, for serving multiple FSes (949)
4429 - connections are again shared between multiple imp/exports (963, 964)
4430 - "umount -f" would hang if any requests needed to be sent (393, 978)
4431 - avoid pinning large req buffer by copying for queued messages (989)
4432 - add "uuid" to "lctl device" command to help upcalls (991)
4433 - "open" RPCs with transnos would confuse recovery counters (1037)
4434 - do proper endian conversion of all wire messages (288, 340, 891)
4435 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4436 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4437 - fix ptlrpc_connection leak in target_handle_connect (1174)
4438 - fix import refcounting bug in OST and MDS cleanup (1134)
4439 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4440 - fix very unlikely obd_types race condition (501)
4441 - remove osc_open hack for echo_client (1187)
4442 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4443 - a failure in read_inode2 leads to deadlock (1139)
4444 - cancel ack-locks as soon as transaction is committed (1072)
4445 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4446 - make sure to commitrw after any preprw to avoid deadlock (1162)
4447 - failing to execute a file in a lustre FS would lock inode (1203)
4448 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4449 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4450 - obd_uuid2tgt would walk off the end of the list (1255)
4451 - on IA64 the osc would give portals incorrect bulk size (1258)
4452 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4453 - fix lock inversion caused by new llite matching code (1282)
4454 - limit the number of dirty pages on a client to 10MB (1286)
4455 - timed out locks were not being corrected cancelled (1289)
4456 - fix O_DIRECT above 4GB on IA-32 (1292)
4457 * major user-visible changes
4458 - fail out/fail over policy now controlled by the upcall (993)
4460 - add OBD_PING to check server availability and failure (954)
4461 - lustre messages are now sent in sending host order (288, 340, 891)
4462 - add eadatalen to MDS getattr reply (340)
4463 - OST read replies may contain second buffer, with per-page status (593)
4465 2003-03-11 Phil Schwan <phil@clusterfs.com>
4468 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4469 - failed open()s could cause deadlock; fixed (867, 869)
4470 - stop cancelling OST locks when files are closed (481)
4471 - overlapping XID spaces caused network corruption (851, 853)
4472 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4473 - setattr_raw added, to do single-RPC, server-side setattrs
4474 - lmc/lconf syntax change for OST UUIDs
4475 - fix crashy race condition between ptlrpc_free_req and osc_close
4476 - don't use request in mdc_enqueue if we hit a timeout (889)
4477 - don't set the inode i_size for regular files from the MDS (896)
4478 - handle out of order completion AST (842)
4479 - don't LBUG if a lock request times out after receiving AST (913)
4480 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4481 - if a bad lock AST arrives, send an error instead of dropping entirely
4482 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4483 - fix leak in bulk IO when only partially completed (899, 900, 926)
4484 - fix O_DIRECT for ia64 (55)
4485 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4486 - C-z after timeout could hang a process forever; fixed (977)
4488 - client-side I/O cache (678, 924, 929, 941, 970)
4490 - READPAGE and SETATTRs which don't take server-side locks get
4493 2003-02-11 Phil Schwan <phil@clusterfs.com>
4496 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4497 - Fix fsfilt_extN_readpage() to read a full page of directory
4498 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4499 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4500 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4501 - Most intent operations were removed, in favour of a new RPC mode
4502 that does a single RPC to the server and bypasses most of the VFS
4503 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4504 - Aggressively cancel local locks on DLM servers
4505 - mds_reint_unlink sends EA to the client if it's the last nlink.
4506 client uses that EA to unlink OST objects.
4507 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4508 - recursive symlinks were fixed (439)
4509 - fixed NULL deref in DEBUG_REQ
4510 - filter_update_lastobjid no longer calls sync, which annoyed extN
4511 - fixed multi-client small-writes to a single file problem (445)
4512 - fixed mtime updates during file writes (607)
4513 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4514 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4515 - fixed Linux OST size reporting problem (444, 656)
4516 - OST now updates object mtime with writes or setattr (607, 619)
4517 - client verifies file size before zeroing page past EOF (445)
4518 - OST now writes last allocated objid to disk with allocation (108)
4519 - LOV on echo now works (409)
4521 - mds_reint_unlink sends a new buffer, with the EA included. this
4522 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4523 set if a regular file was being unlinked, and it was the last link
4524 - use PtlGet from the target for bulk writes (315)
4525 - OST now updates object mtime with writes or setattr (607, 619)
4526 - LDLM now has a grant-time callback to revalidate locked items, if
4528 - Many MDS operations were reorganized to combat race conditions
4530 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4531 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4533 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4536 - Fully reactivate OST imports after reconnection (512, others)
4537 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4538 - More graceful error handling for truncating on dead OST (515)
4539 - Don't error out unless we're actually accessing dead stripes (474)
4540 - Fix garbage sizes when stripes are missing (410)
4541 - LRU counters were broken, causing constant lock purge (433, 432)
4542 - garbage on read from stripes with failed OSTs (441)
4543 - mark OSCs as active before reconnecting during recovery (438)
4544 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4545 - lfind did not preserve OST order in output (443)
4546 - symlinks cause hung clients, incorrect data (439)
4547 - stop dereferencing request after dropping refcount (457)
4548 - don't LASSERT(spin_is_locked) on non-SMP (455)
4549 - fixes for many rename() bugs
4550 - fstat didn't correctly synchronize attributes (399)
4551 - server must handle lock cancellation during blocking AST prep (487)
4552 - bulk descriptors were free()d too soon (511)
4553 - fix paths in lconf, which would load incorrect modules (451, 507)
4554 - fix confusing lconf 'host not found' error message (386)
4555 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4556 - fix race condition in mdc_blocking_ast() for inode access (526)
4557 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4558 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4559 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4560 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4561 - fix lstripe in several ways (kernel side) (527)
4562 - fix request leak in ldlm_cli_enqueue (262)
4563 - incorrect OSC was marked inactive after OST failure
4564 - call mds_fs_cleanup before unmounting filesystem (524)
4565 - fix races between taking ns_lock and ldlm_lock_change_resource
4566 - fix races updating LOV export open file list
4567 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4568 - fix recovery NULL deref in ldlm_cli_cancel_unused
4569 - fix some DLM races by using new hash table for lock handles (419)
4570 - permit the client to specify desired inodes, at replay
4571 - duplicate requests when we queue them for replay reintegration
4572 - fix last_rcvd offset calculation
4573 - sync after each recovered transaction, so we always make progress
4574 - never, not always, ERESTART requests without transnos
4575 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4577 - skip replay if the MDS says that the client is already connected
4578 - don't check for a recovery-enabled export to match lctl's UUID
4579 - don't INC_USE_COUNT for phantom exports
4580 - don't crash when cleaning up phantom exports (567)
4581 - don't double-finish or set replay data for errored mdc_open requests
4582 - abort requests when they time out, so we don't get old replies
4583 - send/receive replies for AST messages again
4584 - if the client says that it doesn't have the lock, cancel it on the
4586 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4587 mark it as destroyed, it will all work out when decref is called
4588 - fix module use counts (22, 581)
4590 - ASTs now expect a reply (server cancels lock on error reply)
4592 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4595 - fix many simultaneous client startup (392)
4596 - fix dentry->d_it clobbering
4597 - credentials weren't being shipped for readdir/getattr operations
4598 - remove invalid assertions triggered during some concurrent MD
4600 - proper Lustre versions added (336, 389)
4601 - fix memory leak for create error case (398)
4602 - fix LOV locking bug that would get cli/srv out of sync
4603 - fix echo client over LOV (409)
4604 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4605 - fix double-O_EXCL intent crash (424)
4606 - avoid sending multiple lock CANCELs (352)
4608 - MDS can do multi-client recovery (modulo bugs in new code)
4610 - many updates, edits, cleanups
4612 2002-11-18 Phil Schwan <phil@clusterfs.com>
4615 - fix null d_it dereference (346)
4616 - fix full OST/dbench hang (333)
4617 - fix permission problem with file removal (286)
4618 - fix removal of OSCs from LOV when they fail
4619 - fix NULL deref during bulk timeout (214)
4620 - fix problems related to multiple filesystems on one MDS (241)
4621 - fixed serious subtle metadata locking bugs
4622 - free locks on clients when inodes are removed due to memory
4624 - fix inode pointer in lock data (285)
4625 - partial support for multiple MDS on a single host (241)
4626 - data locks weren't cancelled at clear_inode time (290, 311)
4627 - intent locks could lead to unbounded lock growth (205)
4628 - added a maximum lock count, an LRU list, and a flusher
4629 - fix multiple rename (365)
4630 - properly abstracted the echo client
4631 - OSC locked 1 byte too many; fixed
4632 - rewrote brw callback code:
4633 - fixed recovery bugs related to LOVs (306)
4634 - fixed too-many-pages-in-one-write crash (191)
4635 - fixed (again) crash in sync_io_timeout (214)
4636 - probably fixed callback-related race (385)
4638 - Add capability to MDS protocol
4639 - LDLM cancellations and callbacks on different portals
4641 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4644 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4645 - defer open object destruction to close time (601981, 138)
4646 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4647 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4648 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4649 - don't call obd_setattr in ll_file_release for destroyed objects
4650 * protocol change to lustre_msg: move |version| and add |flags|
4651 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4652 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4653 * added replay of create, unlink, link and rename operations during
4654 MDS failover; recovery should be much more robust now
4655 * remove failed OSCs from LOVs (only lov_create uses this so far)
4656 * the lustre-HOWTO was brought (more) up to date (582544)
4658 2002-10-23 Phil Schwan <phil@clusterfs.com>
4661 - in-use dentries weren't being reused properly (617851)
4662 - prevent multiple LDLM setup (599178)
4663 - fix LOV size calculations for truncate (617853)
4664 - fix client handling of MDS intent errors (POSIX)
4665 - fix permission bug in lovstripe.c test (624321)
4666 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4667 - truncate past end of file could corrupt data
4668 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4669 - a race in recovery could return ETIMEDOUT to apps (623947)
4670 - building outside the source directory was fixed
4671 * the lustre-HOWTO was brought (more) up to date (582544)
4672 * major progress was made on recovery functionality
4674 2002-10-10 Phil Schwan <phil@clusterfs.com>
4677 - recovery deadlock fix
4678 - rm -rf causes LBUG fix (617817)
4679 - file open by multiple tasks fix (618962)
4680 - directory permissions bugs (602707 and 620007)
4681 - journal_stop fixed with locking (611313)
4682 - O_APPEND failures resolved (618273, perhaps 614459)
4683 - lconf PATH fix (619770)
4684 - IA64 build fix (621450)
4685 - RPC buffer sizes scale with amount of memory
4687 2002-10-01 Phil Schwan <phil@clusterfs.com>
4690 - locks would be cancelled without throwing away data pages,
4691 resulting in inconsistent data (605627)
4692 - inode attributes were not always being refreshed (605627, 612449)
4693 - lconf now continues to cleanup after lctl reports an error
4694 - MDS now enforces user permissions (602707)
4695 - lprocfs cleanup fixed, but not yet enabled (614157)
4696 - fixed infinite server hang, should a client not respond to an AST
4697 - avoid going into recovery if user calls readlink() with a buffer
4698 that's too small (613941)
4699 - AST RPCs no longer require replies (614867) -- this may be changed
4700 - don't crash server if client sends an IOV that's too big (611336)
4701 - fixed lock conversion deadlock (611892)
4702 - fixed the following of symlinks (614622)
4703 * recovery: the server can remove locks from a client that dies, other
4704 clients can make progress
4705 * more extN patch fixes
4706 * compile-time configurable ptlrpc buffer allocations
4708 - collaborative read cache document
4709 - Lustre Lite Performance CDR document-in-progress
4711 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4714 - fix typo in patch-2.4.18
4716 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4719 - clear ptlrpc request each time in handle_incoming_request()
4720 - unlink of files now destroys the object on the OST
4722 2002-09-19 Peter Braam <braam@clusterfs.com>
4724 * add hard link support
4725 * change obdfile creation method
4726 * kernel patch changed
4728 2002-09-19 Peter Braam <braam@clusterfs.com>
4731 - stack overflow bug in extN fixed
4733 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4735 * documentation updates
4736 - add man pages for config tools
4737 - update tests/README to describe testing with new config tools
4738 - finish metadata API descriptions
4739 * bug fixes and cleanups
4740 - statfs workaround for 16TB limit
4741 - LOV stripe allocation improved, can stripe on subset of OSTs
4742 - LOV file size/IO offset was wrong for files > 4GB in size
4743 - object EA data was being dropped, caused files to be unreadable
4744 - memory overflow with non-LOV OST caused memory corruption
4745 - fixed regression tests to work with new config tools, obdfilter
4746 - fixed bug when directory size became larger than 1 block
4747 - fixed bug (for single client case) when PWD was deleted
4748 - invalidate local directory pages when doing intent-based ops
4749 - avoid LDLM oops when lock callback contained bad data
4751 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4753 * documentation updates
4754 * bug fixes and cleanups
4755 - configuration tools
4758 - 64-bit compile warnings
4759 - 64-bit internal statfs data
4761 * test_brw on persistent OST devices
4763 * lprocfs (disabled)
4765 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4767 * documentation updates
4768 * bug fixes and cleanups
4769 * configuration tools
4771 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4774 * many small fixes to 0_5_4
4775 * io/network handling
4776 * thinkos in MDS operations
4778 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4781 * crucial basic fixes to 0.5.3
4782 * IOR, Iozone work over Elan
4785 2002-08-07 Phil Schwan <phil@clusterfs.com>
4786 * version 0_5_3, our first alpha
4787 * we use the new Portals iovs
4788 * documentation updates
4789 * bug fixes and cleanups
4790 * small changes in the DLM wire protocol
4792 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4793 * version 0_5_1 with some initial stability,
4794 * locking on MD and file I/O.
4795 * documentation updates
4796 * several bug fixes since 0.5.0
4797 * small changes in wire protocol
4799 2002-07-18 Phil Schwan <phil@clusterfs.com>
4801 * delivered as Lustre Light Alpha
4802 * fixed a crash after handling invalid MDS requests
4803 * fixed directory pages for architectures with non-4k pages sizes
4805 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4806 * release version v0_4_4
4807 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4809 * Fixes a number of interruption problems with OST operations.
4810 * Update documentation for portals header changes
4811 * Move all wire protocol structs/defines to lustre_idl.h
4812 * Fixes symlink length bug.
4813 * Add tcpdump to repository.
4815 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4816 * release version v0_4_3
4817 * Fixes statfs for inodes on extN.
4818 * Fixes bug in runtests which would delete /etc/hosts.
4819 * Use 64-bit object IDs wherever possible (not into VFS though)
4820 Remove ost_get_info, which is unused by lustre, and out of date.
4822 2002-07-03 Peter Braam <braam@clusterfs.com>
4823 * release version v0_4_2 Fixes a lookup error (type not passed)
4824 * move forward to head of Portals
4825 * move forward to latest Lustre kernel
4827 2002-06-25 Peter Braam <braam@clusterfs.com>
4828 * release version v0_4_1. Hopefully stable on single node use.