4 - workaround for gcc 3.2, which has macro-argument issues (850)
5 - lmc/lconf syntax change for OST UUIDs
7 2003-02-11 Phil Schwan <phil@clusterfs.com>
10 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
11 - Fix fsfilt_extN_readpage() to read a full page of directory
12 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
13 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
14 - Fix imbalanced LOV object allocation and out-of-bound access (469)
15 - Most intent operations were removed, in favour of a new RPC mode
16 that does a single RPC to the server and bypasses most of the VFS
17 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
18 - Aggressively cancel local locks on DLM servers
19 - mds_reint_unlink sends EA to the client if it's the last nlink.
20 client uses that EA to unlink OST objects.
21 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
22 - recursive symlinks were fixed (439)
23 - fixed NULL deref in DEBUG_REQ
24 - filter_update_lastobjid no longer calls sync, which annoyed extN
25 - fixed multi-client small-writes to a single file problem (445)
26 - fixed mtime updates during file writes (607)
27 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
28 - fixed bug in obd_brw_read/write() (under guise of testing 367)
29 - fixed Linux OST size reporting problem (444, 656)
30 - OST now updates object mtime with writes or setattr (607, 619)
31 - client verifies file size before zeroing page past EOF (445)
32 - OST now writes last allocated objid to disk with allocation (108)
33 - LOV on echo now works (409)
35 - mds_reint_unlink sends a new buffer, with the EA included. this
36 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
37 set if a regular file was being unlinked, and it was the last link
38 - use PtlGet from the target for bulk writes (315)
39 - OST now updates object mtime with writes or setattr (607, 619)
40 - LDLM now has a grant-time callback to revalidate locked items, if
42 - Many MDS operations were reorganized to combat race conditions
44 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
45 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
47 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
50 - Fully reactivate OST imports after reconnection (512, others)
51 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
52 - More graceful error handling for truncating on dead OST (515)
53 - Don't error out unless we're actually accessing dead stripes (474)
54 - Fix garbage sizes when stripes are missing (410)
55 - LRU counters were broken, causing constant lock purge (433, 432)
56 - garbage on read from stripes with failed OSTs (441)
57 - mark OSCs as active before reconnecting during recovery (438)
58 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
59 - lfind did not preserve OST order in output (443)
60 - symlinks cause hung clients, incorrect data (439)
61 - stop dereferencing request after dropping refcount (457)
62 - don't LASSERT(spin_is_locked) on non-SMP (455)
63 - fixes for many rename() bugs
64 - fstat didn't correctly synchronize attributes (399)
65 - server must handle lock cancellation during blocking AST prep (487)
66 - bulk descriptors were free()d too soon (511)
67 - fix paths in lconf, which would load incorrect modules (451, 507)
68 - fix confusing lconf 'host not found' error message (386)
69 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
70 - fix race condition in mdc_blocking_ast() for inode access (526)
71 - fix lov_unpackmd() unpacking wrong number of stripes (537)
72 - fix lov_set_osc_active() marking wrong OSC inactive (440)
73 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
74 - fix multiple writes of stripe MD to MDS (358, maybe 519)
75 - fix lstripe in several ways (kernel side) (527)
76 - fix request leak in ldlm_cli_enqueue (262)
77 - incorrect OSC was marked inactive after OST failure
78 - call mds_fs_cleanup before unmounting filesystem (524)
79 - fix races between taking ns_lock and ldlm_lock_change_resource
80 - fix races updating LOV export open file list
81 - fix lov_enqueue error path, avoid decref-ing bad lock handle
82 - fix recovery NULL deref in ldlm_cli_cancel_unused
83 - fix some DLM races by using new hash table for lock handles (419)
84 - permit the client to specify desired inodes, at replay
85 - duplicate requests when we queue them for replay reintegration
86 - fix last_rcvd offset calculation
87 - sync after each recovered transaction, so we always make progress
88 - never, not always, ERESTART requests without transnos
89 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
91 - skip replay if the MDS says that the client is already connected
92 - don't check for a recovery-enabled export to match lctl's UUID
93 - don't INC_USE_COUNT for phantom exports
94 - don't crash when cleaning up phantom exports (567)
95 - don't double-finish or set replay data for errored mdc_open requests
96 - abort requests when they time out, so we don't get old replies
97 - send/receive replies for AST messages again
98 - if the client says that it doesn't have the lock, cancel it on the
100 - if we timeout during I/O, don't try to cancel an in-use lock; instead
101 mark it as destroyed, it will all work out when decref is called
102 - fix module use counts (22, 581)
104 - ASTs now expect a reply (server cancels lock on error reply)
106 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
109 - fix many simultaneous client startup (392)
110 - fix dentry->d_it clobbering
111 - credentials weren't being shipped for readdir/getattr operations
112 - remove invalid assertions triggered during some concurrent MD
114 - proper Lustre versions added (336, 389)
115 - fix memory leak for create error case (398)
116 - fix LOV locking bug that would get cli/srv out of sync
117 - fix echo client over LOV (409)
118 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
119 - fix double-O_EXCL intent crash (424)
120 - avoid sending multiple lock CANCELs (352)
122 - MDS can do multi-client recovery (modulo bugs in new code)
124 - many updates, edits, cleanups
126 2002-11-18 Phil Schwan <phil@clusterfs.com>
129 - fix null d_it dereference (346)
130 - fix full OST/dbench hang (333)
131 - fix permission problem with file removal (286)
132 - fix removal of OSCs from LOV when they fail
133 - fix NULL deref during bulk timeout (214)
134 - fix problems related to multiple filesystems on one MDS (241)
135 - fixed serious subtle metadata locking bugs
136 - free locks on clients when inodes are removed due to memory
138 - fix inode pointer in lock data (285)
139 - partial support for multiple MDS on a single host (241)
140 - data locks weren't cancelled at clear_inode time (290, 311)
141 - intent locks could lead to unbounded lock growth (205)
142 - added a maximum lock count, an LRU list, and a flusher
143 - fix multiple rename (365)
144 - properly abstracted the echo client
145 - OSC locked 1 byte too many; fixed
146 - rewrote brw callback code:
147 - fixed recovery bugs related to LOVs (306)
148 - fixed too-many-pages-in-one-write crash (191)
149 - fixed (again) crash in sync_io_timeout (214)
150 - probably fixed callback-related race (385)
152 - Add capability to MDS protocol
153 - LDLM cancellations and callbacks on different portals
155 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
158 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
159 - defer open object destruction to close time (601981, 138)
160 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
161 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
162 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
163 - don't call obd_setattr in ll_file_release for destroyed objects
164 * protocol change to lustre_msg: move |version| and add |flags|
165 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
166 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
167 * added replay of create, unlink, link and rename operations during
168 MDS failover; recovery should be much more robust now
169 * remove failed OSCs from LOVs (only lov_create uses this so far)
170 * the lustre-HOWTO was brought (more) up to date (582544)
172 2002-10-23 Phil Schwan <phil@clusterfs.com>
175 - in-use dentries weren't being reused properly (617851)
176 - prevent multiple LDLM setup (599178)
177 - fix LOV size calculations for truncate (617853)
178 - fix client handling of MDS intent errors (POSIX)
179 - fix permission bug in lovstripe.c test (624321)
180 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
181 - truncate past end of file could corrupt data
182 - proper cleanup after timeouts, crashes, etc (592524, 550815)
183 - a race in recovery could return ETIMEDOUT to apps (623947)
184 - building outside the source directory was fixed
185 * the lustre-HOWTO was brought (more) up to date (582544)
186 * major progress was made on recovery functionality
188 2002-10-10 Phil Schwan <phil@clusterfs.com>
191 - recovery deadlock fix
192 - rm -rf causes LBUG fix (617817)
193 - file open by multiple tasks fix (618962)
194 - directory permissions bugs (602707 and 620007)
195 - journal_stop fixed with locking (611313)
196 - O_APPEND failures resolved (618273, perhaps 614459)
197 - lconf PATH fix (619770)
198 - IA64 build fix (621450)
199 - RPC buffer sizes scale with amount of memory
201 2002-10-01 Phil Schwan <phil@clusterfs.com>
204 - locks would be cancelled without throwing away data pages,
205 resulting in inconsistent data (605627)
206 - inode attributes were not always being refreshed (605627, 612449)
207 - lconf now continues to cleanup after lctl reports an error
208 - MDS now enforces user permissions (602707)
209 - lprocfs cleanup fixed, but not yet enabled (614157)
210 - fixed infinite server hang, should a client not respond to an AST
211 - avoid going into recovery if user calls readlink() with a buffer
212 that's too small (613941)
213 - AST RPCs no longer require replies (614867) -- this may be changed
214 - don't crash server if client sends an IOV that's too big (611336)
215 - fixed lock conversion deadlock (611892)
216 - fixed the following of symlinks (614622)
217 * recovery: the server can remove locks from a client that dies, other
218 clients can make progress
219 * more extN patch fixes
220 * compile-time configurable ptlrpc buffer allocations
222 - collaborative read cache document
223 - Lustre Lite Performance CDR document-in-progress
225 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
228 - fix typo in patch-2.4.18
230 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
233 - clear ptlrpc request each time in handle_incoming_request()
234 - unlink of files now destroys the object on the OST
236 2002-09-19 Peter Braam <braam@clusterfs.com>
238 * add hard link support
239 * change obdfile creation method
240 * kernel patch changed
242 2002-09-19 Peter Braam <braam@clusterfs.com>
245 - stack overflow bug in extN fixed
247 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
249 * documentation updates
250 - add man pages for config tools
251 - update tests/README to describe testing with new config tools
252 - finish metadata API descriptions
253 * bug fixes and cleanups
254 - statfs workaround for 16TB limit
255 - LOV stripe allocation improved, can stripe on subset of OSTs
256 - LOV file size/IO offset was wrong for files > 4GB in size
257 - object EA data was being dropped, caused files to be unreadable
258 - memory overflow with non-LOV OST caused memory corruption
259 - fixed regression tests to work with new config tools, obdfilter
260 - fixed bug when directory size became larger than 1 block
261 - fixed bug (for single client case) when PWD was deleted
262 - invalidate local directory pages when doing intent-based ops
263 - avoid LDLM oops when lock callback contained bad data
265 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
267 * documentation updates
268 * bug fixes and cleanups
269 - configuration tools
272 - 64-bit compile warnings
273 - 64-bit internal statfs data
275 * test_brw on persistent OST devices
279 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
281 * documentation updates
282 * bug fixes and cleanups
283 * configuration tools
285 2002-08-30 Peter J. Braam <braam@clusterfs.com>
288 * many small fixes to 0_5_4
289 * io/network handling
290 * thinkos in MDS operations
292 2002-08-24 Peter J. Braam <braam@clusterfs.com>
295 * crucial basic fixes to 0.5.3
296 * IOR, Iozone work over Elan
299 2002-08-07 Phil Schwan <phil@clusterfs.com>
300 * version 0_5_3, our first alpha
301 * we use the new Portals iovs
302 * documentation updates
303 * bug fixes and cleanups
304 * small changes in the DLM wire protocol
306 2002-07-25 Peter J. Braam <braam@clusterfs.com>
307 * version 0_5_1 with some initial stability,
308 * locking on MD and file I/O.
309 * documentation updates
310 * several bug fixes since 0.5.0
311 * small changes in wire protocol
313 2002-07-18 Phil Schwan <phil@clusterfs.com>
315 * delivered as Lustre Light Alpha
316 * fixed a crash after handling invalid MDS requests
317 * fixed directory pages for architectures with non-4k pages sizes
319 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
320 * release version v0_4_4
321 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
323 * Fixes a number of interruption problems with OST operations.
324 * Update documentation for portals header changes
325 * Move all wire protocol structs/defines to lustre_idl.h
326 * Fixes symlink length bug.
327 * Add tcpdump to repository.
329 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
330 * release version v0_4_3
331 * Fixes statfs for inodes on extN.
332 * Fixes bug in runtests which would delete /etc/hosts.
333 * Use 64-bit object IDs wherever possible (not into VFS though)
334 Remove ost_get_info, which is unused by lustre, and out of date.
336 2002-07-03 Peter Braam <braam@clusterfs.com>
337 * release version v0_4_2 Fixes a lookup error (type not passed)
338 * move forward to head of Portals
339 * move forward to latest Lustre kernel
341 2002-06-25 Peter Braam <braam@clusterfs.com>
342 * release version v0_4_1. Hopefully stable on single node use.