1 2003-03-11 Phil Schwan <phil@clusterfs.com>
4 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5 - failed open()s could cause deadlock; fixed (867, 869)
6 - stop cancelling OST locks when files are closed (481)
7 - overlapping XID spaces caused network corruption (851, 853)
8 - fix unsafe fsfilt counter arithmetic; change to atomic_t
9 - setattr_raw added, to do single-RPC, server-side setattrs
10 - lmc/lconf syntax change for OST UUIDs
11 - fix crashy race condition between ptlrpc_free_req and osc_close
12 - don't use request in mdc_enqueue if we hit a timeout (889)
13 - don't set the inode i_size for regular files from the MDS (896)
14 - handle out of order completion AST (842)
15 - don't LBUG if a lock request times out after receiving AST (913)
16 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
17 - if a bad lock AST arrives, send an error instead of dropping entirely
18 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
19 - fix leak in bulk IO when only partially completed (899, 900, 926)
20 - fix O_DIRECT for ia64 (55)
21 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
22 - C-z after timeout could hang a process forever; fixed (977)
24 - client-side I/O cache (678, 924, 929, 941, 970)
26 - READPAGE and SETATTRs which don't take server-side locks get
29 2003-02-11 Phil Schwan <phil@clusterfs.com>
32 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
33 - Fix fsfilt_extN_readpage() to read a full page of directory
34 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
35 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
36 - Fix imbalanced LOV object allocation and out-of-bound access (469)
37 - Most intent operations were removed, in favour of a new RPC mode
38 that does a single RPC to the server and bypasses most of the VFS
39 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
40 - Aggressively cancel local locks on DLM servers
41 - mds_reint_unlink sends EA to the client if it's the last nlink.
42 client uses that EA to unlink OST objects.
43 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
44 - recursive symlinks were fixed (439)
45 - fixed NULL deref in DEBUG_REQ
46 - filter_update_lastobjid no longer calls sync, which annoyed extN
47 - fixed multi-client small-writes to a single file problem (445)
48 - fixed mtime updates during file writes (607)
49 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
50 - fixed bug in obd_brw_read/write() (under guise of testing 367)
51 - fixed Linux OST size reporting problem (444, 656)
52 - OST now updates object mtime with writes or setattr (607, 619)
53 - client verifies file size before zeroing page past EOF (445)
54 - OST now writes last allocated objid to disk with allocation (108)
55 - LOV on echo now works (409)
57 - mds_reint_unlink sends a new buffer, with the EA included. this
58 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
59 set if a regular file was being unlinked, and it was the last link
60 - use PtlGet from the target for bulk writes (315)
61 - OST now updates object mtime with writes or setattr (607, 619)
62 - LDLM now has a grant-time callback to revalidate locked items, if
64 - Many MDS operations were reorganized to combat race conditions
66 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
67 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
69 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
72 - Fully reactivate OST imports after reconnection (512, others)
73 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
74 - More graceful error handling for truncating on dead OST (515)
75 - Don't error out unless we're actually accessing dead stripes (474)
76 - Fix garbage sizes when stripes are missing (410)
77 - LRU counters were broken, causing constant lock purge (433, 432)
78 - garbage on read from stripes with failed OSTs (441)
79 - mark OSCs as active before reconnecting during recovery (438)
80 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
81 - lfind did not preserve OST order in output (443)
82 - symlinks cause hung clients, incorrect data (439)
83 - stop dereferencing request after dropping refcount (457)
84 - don't LASSERT(spin_is_locked) on non-SMP (455)
85 - fixes for many rename() bugs
86 - fstat didn't correctly synchronize attributes (399)
87 - server must handle lock cancellation during blocking AST prep (487)
88 - bulk descriptors were free()d too soon (511)
89 - fix paths in lconf, which would load incorrect modules (451, 507)
90 - fix confusing lconf 'host not found' error message (386)
91 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
92 - fix race condition in mdc_blocking_ast() for inode access (526)
93 - fix lov_unpackmd() unpacking wrong number of stripes (537)
94 - fix lov_set_osc_active() marking wrong OSC inactive (440)
95 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
96 - fix multiple writes of stripe MD to MDS (358, maybe 519)
97 - fix lstripe in several ways (kernel side) (527)
98 - fix request leak in ldlm_cli_enqueue (262)
99 - incorrect OSC was marked inactive after OST failure
100 - call mds_fs_cleanup before unmounting filesystem (524)
101 - fix races between taking ns_lock and ldlm_lock_change_resource
102 - fix races updating LOV export open file list
103 - fix lov_enqueue error path, avoid decref-ing bad lock handle
104 - fix recovery NULL deref in ldlm_cli_cancel_unused
105 - fix some DLM races by using new hash table for lock handles (419)
106 - permit the client to specify desired inodes, at replay
107 - duplicate requests when we queue them for replay reintegration
108 - fix last_rcvd offset calculation
109 - sync after each recovered transaction, so we always make progress
110 - never, not always, ERESTART requests without transnos
111 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
113 - skip replay if the MDS says that the client is already connected
114 - don't check for a recovery-enabled export to match lctl's UUID
115 - don't INC_USE_COUNT for phantom exports
116 - don't crash when cleaning up phantom exports (567)
117 - don't double-finish or set replay data for errored mdc_open requests
118 - abort requests when they time out, so we don't get old replies
119 - send/receive replies for AST messages again
120 - if the client says that it doesn't have the lock, cancel it on the
122 - if we timeout during I/O, don't try to cancel an in-use lock; instead
123 mark it as destroyed, it will all work out when decref is called
124 - fix module use counts (22, 581)
126 - ASTs now expect a reply (server cancels lock on error reply)
128 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
131 - fix many simultaneous client startup (392)
132 - fix dentry->d_it clobbering
133 - credentials weren't being shipped for readdir/getattr operations
134 - remove invalid assertions triggered during some concurrent MD
136 - proper Lustre versions added (336, 389)
137 - fix memory leak for create error case (398)
138 - fix LOV locking bug that would get cli/srv out of sync
139 - fix echo client over LOV (409)
140 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
141 - fix double-O_EXCL intent crash (424)
142 - avoid sending multiple lock CANCELs (352)
144 - MDS can do multi-client recovery (modulo bugs in new code)
146 - many updates, edits, cleanups
148 2002-11-18 Phil Schwan <phil@clusterfs.com>
151 - fix null d_it dereference (346)
152 - fix full OST/dbench hang (333)
153 - fix permission problem with file removal (286)
154 - fix removal of OSCs from LOV when they fail
155 - fix NULL deref during bulk timeout (214)
156 - fix problems related to multiple filesystems on one MDS (241)
157 - fixed serious subtle metadata locking bugs
158 - free locks on clients when inodes are removed due to memory
160 - fix inode pointer in lock data (285)
161 - partial support for multiple MDS on a single host (241)
162 - data locks weren't cancelled at clear_inode time (290, 311)
163 - intent locks could lead to unbounded lock growth (205)
164 - added a maximum lock count, an LRU list, and a flusher
165 - fix multiple rename (365)
166 - properly abstracted the echo client
167 - OSC locked 1 byte too many; fixed
168 - rewrote brw callback code:
169 - fixed recovery bugs related to LOVs (306)
170 - fixed too-many-pages-in-one-write crash (191)
171 - fixed (again) crash in sync_io_timeout (214)
172 - probably fixed callback-related race (385)
174 - Add capability to MDS protocol
175 - LDLM cancellations and callbacks on different portals
177 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
180 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
181 - defer open object destruction to close time (601981, 138)
182 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
183 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
184 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
185 - don't call obd_setattr in ll_file_release for destroyed objects
186 * protocol change to lustre_msg: move |version| and add |flags|
187 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
188 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
189 * added replay of create, unlink, link and rename operations during
190 MDS failover; recovery should be much more robust now
191 * remove failed OSCs from LOVs (only lov_create uses this so far)
192 * the lustre-HOWTO was brought (more) up to date (582544)
194 2002-10-23 Phil Schwan <phil@clusterfs.com>
197 - in-use dentries weren't being reused properly (617851)
198 - prevent multiple LDLM setup (599178)
199 - fix LOV size calculations for truncate (617853)
200 - fix client handling of MDS intent errors (POSIX)
201 - fix permission bug in lovstripe.c test (624321)
202 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
203 - truncate past end of file could corrupt data
204 - proper cleanup after timeouts, crashes, etc (592524, 550815)
205 - a race in recovery could return ETIMEDOUT to apps (623947)
206 - building outside the source directory was fixed
207 * the lustre-HOWTO was brought (more) up to date (582544)
208 * major progress was made on recovery functionality
210 2002-10-10 Phil Schwan <phil@clusterfs.com>
213 - recovery deadlock fix
214 - rm -rf causes LBUG fix (617817)
215 - file open by multiple tasks fix (618962)
216 - directory permissions bugs (602707 and 620007)
217 - journal_stop fixed with locking (611313)
218 - O_APPEND failures resolved (618273, perhaps 614459)
219 - lconf PATH fix (619770)
220 - IA64 build fix (621450)
221 - RPC buffer sizes scale with amount of memory
223 2002-10-01 Phil Schwan <phil@clusterfs.com>
226 - locks would be cancelled without throwing away data pages,
227 resulting in inconsistent data (605627)
228 - inode attributes were not always being refreshed (605627, 612449)
229 - lconf now continues to cleanup after lctl reports an error
230 - MDS now enforces user permissions (602707)
231 - lprocfs cleanup fixed, but not yet enabled (614157)
232 - fixed infinite server hang, should a client not respond to an AST
233 - avoid going into recovery if user calls readlink() with a buffer
234 that's too small (613941)
235 - AST RPCs no longer require replies (614867) -- this may be changed
236 - don't crash server if client sends an IOV that's too big (611336)
237 - fixed lock conversion deadlock (611892)
238 - fixed the following of symlinks (614622)
239 * recovery: the server can remove locks from a client that dies, other
240 clients can make progress
241 * more extN patch fixes
242 * compile-time configurable ptlrpc buffer allocations
244 - collaborative read cache document
245 - Lustre Lite Performance CDR document-in-progress
247 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
250 - fix typo in patch-2.4.18
252 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
255 - clear ptlrpc request each time in handle_incoming_request()
256 - unlink of files now destroys the object on the OST
258 2002-09-19 Peter Braam <braam@clusterfs.com>
260 * add hard link support
261 * change obdfile creation method
262 * kernel patch changed
264 2002-09-19 Peter Braam <braam@clusterfs.com>
267 - stack overflow bug in extN fixed
269 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
271 * documentation updates
272 - add man pages for config tools
273 - update tests/README to describe testing with new config tools
274 - finish metadata API descriptions
275 * bug fixes and cleanups
276 - statfs workaround for 16TB limit
277 - LOV stripe allocation improved, can stripe on subset of OSTs
278 - LOV file size/IO offset was wrong for files > 4GB in size
279 - object EA data was being dropped, caused files to be unreadable
280 - memory overflow with non-LOV OST caused memory corruption
281 - fixed regression tests to work with new config tools, obdfilter
282 - fixed bug when directory size became larger than 1 block
283 - fixed bug (for single client case) when PWD was deleted
284 - invalidate local directory pages when doing intent-based ops
285 - avoid LDLM oops when lock callback contained bad data
287 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
289 * documentation updates
290 * bug fixes and cleanups
291 - configuration tools
294 - 64-bit compile warnings
295 - 64-bit internal statfs data
297 * test_brw on persistent OST devices
301 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
303 * documentation updates
304 * bug fixes and cleanups
305 * configuration tools
307 2002-08-30 Peter J. Braam <braam@clusterfs.com>
310 * many small fixes to 0_5_4
311 * io/network handling
312 * thinkos in MDS operations
314 2002-08-24 Peter J. Braam <braam@clusterfs.com>
317 * crucial basic fixes to 0.5.3
318 * IOR, Iozone work over Elan
321 2002-08-07 Phil Schwan <phil@clusterfs.com>
322 * version 0_5_3, our first alpha
323 * we use the new Portals iovs
324 * documentation updates
325 * bug fixes and cleanups
326 * small changes in the DLM wire protocol
328 2002-07-25 Peter J. Braam <braam@clusterfs.com>
329 * version 0_5_1 with some initial stability,
330 * locking on MD and file I/O.
331 * documentation updates
332 * several bug fixes since 0.5.0
333 * small changes in wire protocol
335 2002-07-18 Phil Schwan <phil@clusterfs.com>
337 * delivered as Lustre Light Alpha
338 * fixed a crash after handling invalid MDS requests
339 * fixed directory pages for architectures with non-4k pages sizes
341 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
342 * release version v0_4_4
343 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
345 * Fixes a number of interruption problems with OST operations.
346 * Update documentation for portals header changes
347 * Move all wire protocol structs/defines to lustre_idl.h
348 * Fixes symlink length bug.
349 * Add tcpdump to repository.
351 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
352 * release version v0_4_3
353 * Fixes statfs for inodes on extN.
354 * Fixes bug in runtests which would delete /etc/hosts.
355 * Use 64-bit object IDs wherever possible (not into VFS though)
356 Remove ost_get_info, which is unused by lustre, and out of date.
358 2002-07-03 Peter Braam <braam@clusterfs.com>
359 * release version v0_4_2 Fixes a lookup error (type not passed)
360 * move forward to head of Portals
361 * move forward to latest Lustre kernel
363 2002-06-25 Peter Braam <braam@clusterfs.com>
364 * release version v0_4_1. Hopefully stable on single node use.