]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) International Business Machines Corp., 2000-2005 | |
3 | * Portions Copyright (C) Christoph Hellwig, 2001-2002 | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See | |
13 | * the GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 | */ | |
19 | ||
20 | /* | |
21 | * jfs_txnmgr.c: transaction manager | |
22 | * | |
23 | * notes: | |
24 | * transaction starts with txBegin() and ends with txCommit() | |
25 | * or txAbort(). | |
26 | * | |
27 | * tlock is acquired at the time of update; | |
28 | * (obviate scan at commit time for xtree and dtree) | |
29 | * tlock and mp points to each other; | |
30 | * (no hashlist for mp -> tlock). | |
31 | * | |
32 | * special cases: | |
33 | * tlock on in-memory inode: | |
34 | * in-place tlock in the in-memory inode itself; | |
35 | * converted to page lock by iWrite() at commit time. | |
36 | * | |
37 | * tlock during write()/mmap() under anonymous transaction (tid = 0): | |
38 | * transferred (?) to transaction at commit time. | |
39 | * | |
40 | * use the page itself to update allocation maps | |
41 | * (obviate intermediate replication of allocation/deallocation data) | |
42 | * hold on to mp+lock thru update of maps | |
43 | */ | |
44 | ||
45 | ||
46 | #include <linux/fs.h> | |
47 | #include <linux/vmalloc.h> | |
48 | #include <linux/smp_lock.h> | |
49 | #include <linux/completion.h> | |
50 | #include <linux/suspend.h> | |
51 | #include <linux/module.h> | |
52 | #include <linux/moduleparam.h> | |
53 | #include "jfs_incore.h" | |
54 | #include "jfs_filsys.h" | |
55 | #include "jfs_metapage.h" | |
56 | #include "jfs_dinode.h" | |
57 | #include "jfs_imap.h" | |
58 | #include "jfs_dmap.h" | |
59 | #include "jfs_superblock.h" | |
60 | #include "jfs_debug.h" | |
61 | ||
62 | /* | |
63 | * transaction management structures | |
64 | */ | |
65 | static struct { | |
66 | int freetid; /* index of a free tid structure */ | |
67 | int freelock; /* index first free lock word */ | |
68 | wait_queue_head_t freewait; /* eventlist of free tblock */ | |
69 | wait_queue_head_t freelockwait; /* eventlist of free tlock */ | |
70 | wait_queue_head_t lowlockwait; /* eventlist of ample tlocks */ | |
71 | int tlocksInUse; /* Number of tlocks in use */ | |
72 | spinlock_t LazyLock; /* synchronize sync_queue & unlock_queue */ | |
73 | /* struct tblock *sync_queue; * Transactions waiting for data sync */ | |
74 | struct list_head unlock_queue; /* Txns waiting to be released */ | |
75 | struct list_head anon_list; /* inodes having anonymous txns */ | |
76 | struct list_head anon_list2; /* inodes having anonymous txns | |
77 | that couldn't be sync'ed */ | |
78 | } TxAnchor; | |
79 | ||
80 | int jfs_tlocks_low; /* Indicates low number of available tlocks */ | |
81 | ||
82 | #ifdef CONFIG_JFS_STATISTICS | |
83 | static struct { | |
84 | uint txBegin; | |
85 | uint txBegin_barrier; | |
86 | uint txBegin_lockslow; | |
87 | uint txBegin_freetid; | |
88 | uint txBeginAnon; | |
89 | uint txBeginAnon_barrier; | |
90 | uint txBeginAnon_lockslow; | |
91 | uint txLockAlloc; | |
92 | uint txLockAlloc_freelock; | |
93 | } TxStat; | |
94 | #endif | |
95 | ||
96 | static int nTxBlock = -1; /* number of transaction blocks */ | |
97 | module_param(nTxBlock, int, 0); | |
98 | MODULE_PARM_DESC(nTxBlock, | |
99 | "Number of transaction blocks (max:65536)"); | |
100 | ||
101 | static int nTxLock = -1; /* number of transaction locks */ | |
102 | module_param(nTxLock, int, 0); | |
103 | MODULE_PARM_DESC(nTxLock, | |
104 | "Number of transaction locks (max:65536)"); | |
105 | ||
106 | struct tblock *TxBlock; /* transaction block table */ | |
107 | static int TxLockLWM; /* Low water mark for number of txLocks used */ | |
108 | static int TxLockHWM; /* High water mark for number of txLocks used */ | |
109 | static int TxLockVHWM; /* Very High water mark */ | |
110 | struct tlock *TxLock; /* transaction lock table */ | |
111 | ||
112 | ||
113 | /* | |
114 | * transaction management lock | |
115 | */ | |
116 | static DEFINE_SPINLOCK(jfsTxnLock); | |
117 | ||
118 | #define TXN_LOCK() spin_lock(&jfsTxnLock) | |
119 | #define TXN_UNLOCK() spin_unlock(&jfsTxnLock) | |
120 | ||
121 | #define LAZY_LOCK_INIT() spin_lock_init(&TxAnchor.LazyLock); | |
122 | #define LAZY_LOCK(flags) spin_lock_irqsave(&TxAnchor.LazyLock, flags) | |
123 | #define LAZY_UNLOCK(flags) spin_unlock_irqrestore(&TxAnchor.LazyLock, flags) | |
124 | ||
125 | DECLARE_WAIT_QUEUE_HEAD(jfs_sync_thread_wait); | |
126 | DECLARE_WAIT_QUEUE_HEAD(jfs_commit_thread_wait); | |
127 | static int jfs_commit_thread_waking; | |
128 | ||
129 | /* | |
130 | * Retry logic exist outside these macros to protect from spurrious wakeups. | |
131 | */ | |
132 | static inline void TXN_SLEEP_DROP_LOCK(wait_queue_head_t * event) | |
133 | { | |
134 | DECLARE_WAITQUEUE(wait, current); | |
135 | ||
136 | add_wait_queue(event, &wait); | |
137 | set_current_state(TASK_UNINTERRUPTIBLE); | |
138 | TXN_UNLOCK(); | |
139 | schedule(); | |
140 | current->state = TASK_RUNNING; | |
141 | remove_wait_queue(event, &wait); | |
142 | } | |
143 | ||
144 | #define TXN_SLEEP(event)\ | |
145 | {\ | |
146 | TXN_SLEEP_DROP_LOCK(event);\ | |
147 | TXN_LOCK();\ | |
148 | } | |
149 | ||
150 | #define TXN_WAKEUP(event) wake_up_all(event) | |
151 | ||
152 | ||
153 | /* | |
154 | * statistics | |
155 | */ | |
156 | static struct { | |
157 | tid_t maxtid; /* 4: biggest tid ever used */ | |
158 | lid_t maxlid; /* 4: biggest lid ever used */ | |
159 | int ntid; /* 4: # of transactions performed */ | |
160 | int nlid; /* 4: # of tlocks acquired */ | |
161 | int waitlock; /* 4: # of tlock wait */ | |
162 | } stattx; | |
163 | ||
164 | ||
165 | /* | |
166 | * external references | |
167 | */ | |
168 | extern int lmGroupCommit(struct jfs_log *, struct tblock *); | |
169 | extern int jfs_commit_inode(struct inode *, int); | |
170 | extern int jfs_stop_threads; | |
171 | ||
172 | extern struct completion jfsIOwait; | |
173 | ||
174 | /* | |
175 | * forward references | |
176 | */ | |
177 | static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
178 | struct tlock * tlck, struct commit * cd); | |
179 | static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
180 | struct tlock * tlck); | |
181 | static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
182 | struct tlock * tlck); | |
183 | static void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
184 | struct tlock * tlck); | |
185 | static void txAllocPMap(struct inode *ip, struct maplock * maplock, | |
186 | struct tblock * tblk); | |
187 | static void txForce(struct tblock * tblk); | |
188 | static int txLog(struct jfs_log * log, struct tblock * tblk, | |
189 | struct commit * cd); | |
190 | static void txUpdateMap(struct tblock * tblk); | |
191 | static void txRelease(struct tblock * tblk); | |
192 | static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
193 | struct tlock * tlck); | |
194 | static void LogSyncRelease(struct metapage * mp); | |
195 | ||
196 | /* | |
197 | * transaction block/lock management | |
198 | * --------------------------------- | |
199 | */ | |
200 | ||
201 | /* | |
202 | * Get a transaction lock from the free list. If the number in use is | |
203 | * greater than the high water mark, wake up the sync daemon. This should | |
204 | * free some anonymous transaction locks. (TXN_LOCK must be held.) | |
205 | */ | |
206 | static lid_t txLockAlloc(void) | |
207 | { | |
208 | lid_t lid; | |
209 | ||
210 | INCREMENT(TxStat.txLockAlloc); | |
211 | if (!TxAnchor.freelock) { | |
212 | INCREMENT(TxStat.txLockAlloc_freelock); | |
213 | } | |
214 | ||
215 | while (!(lid = TxAnchor.freelock)) | |
216 | TXN_SLEEP(&TxAnchor.freelockwait); | |
217 | TxAnchor.freelock = TxLock[lid].next; | |
218 | HIGHWATERMARK(stattx.maxlid, lid); | |
219 | if ((++TxAnchor.tlocksInUse > TxLockHWM) && (jfs_tlocks_low == 0)) { | |
220 | jfs_info("txLockAlloc tlocks low"); | |
221 | jfs_tlocks_low = 1; | |
222 | wake_up(&jfs_sync_thread_wait); | |
223 | } | |
224 | ||
225 | return lid; | |
226 | } | |
227 | ||
228 | static void txLockFree(lid_t lid) | |
229 | { | |
7fab479b | 230 | TxLock[lid].tid = 0; |
1da177e4 LT |
231 | TxLock[lid].next = TxAnchor.freelock; |
232 | TxAnchor.freelock = lid; | |
233 | TxAnchor.tlocksInUse--; | |
234 | if (jfs_tlocks_low && (TxAnchor.tlocksInUse < TxLockLWM)) { | |
235 | jfs_info("txLockFree jfs_tlocks_low no more"); | |
236 | jfs_tlocks_low = 0; | |
237 | TXN_WAKEUP(&TxAnchor.lowlockwait); | |
238 | } | |
239 | TXN_WAKEUP(&TxAnchor.freelockwait); | |
240 | } | |
241 | ||
242 | /* | |
243 | * NAME: txInit() | |
244 | * | |
245 | * FUNCTION: initialize transaction management structures | |
246 | * | |
247 | * RETURN: | |
248 | * | |
249 | * serialization: single thread at jfs_init() | |
250 | */ | |
251 | int txInit(void) | |
252 | { | |
253 | int k, size; | |
254 | struct sysinfo si; | |
255 | ||
256 | /* Set defaults for nTxLock and nTxBlock if unset */ | |
257 | ||
258 | if (nTxLock == -1) { | |
259 | if (nTxBlock == -1) { | |
260 | /* Base default on memory size */ | |
261 | si_meminfo(&si); | |
262 | if (si.totalram > (256 * 1024)) /* 1 GB */ | |
263 | nTxLock = 64 * 1024; | |
264 | else | |
265 | nTxLock = si.totalram >> 2; | |
266 | } else if (nTxBlock > (8 * 1024)) | |
267 | nTxLock = 64 * 1024; | |
268 | else | |
269 | nTxLock = nTxBlock << 3; | |
270 | } | |
271 | if (nTxBlock == -1) | |
272 | nTxBlock = nTxLock >> 3; | |
273 | ||
274 | /* Verify tunable parameters */ | |
275 | if (nTxBlock < 16) | |
276 | nTxBlock = 16; /* No one should set it this low */ | |
277 | if (nTxBlock > 65536) | |
278 | nTxBlock = 65536; | |
279 | if (nTxLock < 256) | |
280 | nTxLock = 256; /* No one should set it this low */ | |
281 | if (nTxLock > 65536) | |
282 | nTxLock = 65536; | |
283 | ||
284 | printk(KERN_INFO "JFS: nTxBlock = %d, nTxLock = %d\n", | |
285 | nTxBlock, nTxLock); | |
286 | /* | |
287 | * initialize transaction block (tblock) table | |
288 | * | |
289 | * transaction id (tid) = tblock index | |
290 | * tid = 0 is reserved. | |
291 | */ | |
292 | TxLockLWM = (nTxLock * 4) / 10; | |
293 | TxLockHWM = (nTxLock * 7) / 10; | |
294 | TxLockVHWM = (nTxLock * 8) / 10; | |
295 | ||
296 | size = sizeof(struct tblock) * nTxBlock; | |
297 | TxBlock = (struct tblock *) vmalloc(size); | |
298 | if (TxBlock == NULL) | |
299 | return -ENOMEM; | |
300 | ||
301 | for (k = 1; k < nTxBlock - 1; k++) { | |
302 | TxBlock[k].next = k + 1; | |
303 | init_waitqueue_head(&TxBlock[k].gcwait); | |
304 | init_waitqueue_head(&TxBlock[k].waitor); | |
305 | } | |
306 | TxBlock[k].next = 0; | |
307 | init_waitqueue_head(&TxBlock[k].gcwait); | |
308 | init_waitqueue_head(&TxBlock[k].waitor); | |
309 | ||
310 | TxAnchor.freetid = 1; | |
311 | init_waitqueue_head(&TxAnchor.freewait); | |
312 | ||
313 | stattx.maxtid = 1; /* statistics */ | |
314 | ||
315 | /* | |
316 | * initialize transaction lock (tlock) table | |
317 | * | |
318 | * transaction lock id = tlock index | |
319 | * tlock id = 0 is reserved. | |
320 | */ | |
321 | size = sizeof(struct tlock) * nTxLock; | |
322 | TxLock = (struct tlock *) vmalloc(size); | |
323 | if (TxLock == NULL) { | |
324 | vfree(TxBlock); | |
325 | return -ENOMEM; | |
326 | } | |
327 | ||
328 | /* initialize tlock table */ | |
329 | for (k = 1; k < nTxLock - 1; k++) | |
330 | TxLock[k].next = k + 1; | |
331 | TxLock[k].next = 0; | |
332 | init_waitqueue_head(&TxAnchor.freelockwait); | |
333 | init_waitqueue_head(&TxAnchor.lowlockwait); | |
334 | ||
335 | TxAnchor.freelock = 1; | |
336 | TxAnchor.tlocksInUse = 0; | |
337 | INIT_LIST_HEAD(&TxAnchor.anon_list); | |
338 | INIT_LIST_HEAD(&TxAnchor.anon_list2); | |
339 | ||
340 | LAZY_LOCK_INIT(); | |
341 | INIT_LIST_HEAD(&TxAnchor.unlock_queue); | |
342 | ||
343 | stattx.maxlid = 1; /* statistics */ | |
344 | ||
345 | return 0; | |
346 | } | |
347 | ||
348 | /* | |
349 | * NAME: txExit() | |
350 | * | |
351 | * FUNCTION: clean up when module is unloaded | |
352 | */ | |
353 | void txExit(void) | |
354 | { | |
355 | vfree(TxLock); | |
356 | TxLock = NULL; | |
357 | vfree(TxBlock); | |
358 | TxBlock = NULL; | |
359 | } | |
360 | ||
361 | ||
362 | /* | |
363 | * NAME: txBegin() | |
364 | * | |
365 | * FUNCTION: start a transaction. | |
366 | * | |
367 | * PARAMETER: sb - superblock | |
368 | * flag - force for nested tx; | |
369 | * | |
370 | * RETURN: tid - transaction id | |
371 | * | |
372 | * note: flag force allows to start tx for nested tx | |
373 | * to prevent deadlock on logsync barrier; | |
374 | */ | |
375 | tid_t txBegin(struct super_block *sb, int flag) | |
376 | { | |
377 | tid_t t; | |
378 | struct tblock *tblk; | |
379 | struct jfs_log *log; | |
380 | ||
381 | jfs_info("txBegin: flag = 0x%x", flag); | |
382 | log = JFS_SBI(sb)->log; | |
383 | ||
384 | TXN_LOCK(); | |
385 | ||
386 | INCREMENT(TxStat.txBegin); | |
387 | ||
388 | retry: | |
389 | if (!(flag & COMMIT_FORCE)) { | |
390 | /* | |
391 | * synchronize with logsync barrier | |
392 | */ | |
393 | if (test_bit(log_SYNCBARRIER, &log->flag) || | |
394 | test_bit(log_QUIESCE, &log->flag)) { | |
395 | INCREMENT(TxStat.txBegin_barrier); | |
396 | TXN_SLEEP(&log->syncwait); | |
397 | goto retry; | |
398 | } | |
399 | } | |
400 | if (flag == 0) { | |
401 | /* | |
402 | * Don't begin transaction if we're getting starved for tlocks | |
403 | * unless COMMIT_FORCE or COMMIT_INODE (which may ultimately | |
404 | * free tlocks) | |
405 | */ | |
406 | if (TxAnchor.tlocksInUse > TxLockVHWM) { | |
407 | INCREMENT(TxStat.txBegin_lockslow); | |
408 | TXN_SLEEP(&TxAnchor.lowlockwait); | |
409 | goto retry; | |
410 | } | |
411 | } | |
412 | ||
413 | /* | |
414 | * allocate transaction id/block | |
415 | */ | |
416 | if ((t = TxAnchor.freetid) == 0) { | |
417 | jfs_info("txBegin: waiting for free tid"); | |
418 | INCREMENT(TxStat.txBegin_freetid); | |
419 | TXN_SLEEP(&TxAnchor.freewait); | |
420 | goto retry; | |
421 | } | |
422 | ||
423 | tblk = tid_to_tblock(t); | |
424 | ||
425 | if ((tblk->next == 0) && !(flag & COMMIT_FORCE)) { | |
426 | /* Don't let a non-forced transaction take the last tblk */ | |
427 | jfs_info("txBegin: waiting for free tid"); | |
428 | INCREMENT(TxStat.txBegin_freetid); | |
429 | TXN_SLEEP(&TxAnchor.freewait); | |
430 | goto retry; | |
431 | } | |
432 | ||
433 | TxAnchor.freetid = tblk->next; | |
434 | ||
435 | /* | |
436 | * initialize transaction | |
437 | */ | |
438 | ||
439 | /* | |
440 | * We can't zero the whole thing or we screw up another thread being | |
441 | * awakened after sleeping on tblk->waitor | |
442 | * | |
443 | * memset(tblk, 0, sizeof(struct tblock)); | |
444 | */ | |
445 | tblk->next = tblk->last = tblk->xflag = tblk->flag = tblk->lsn = 0; | |
446 | ||
447 | tblk->sb = sb; | |
448 | ++log->logtid; | |
449 | tblk->logtid = log->logtid; | |
450 | ||
451 | ++log->active; | |
452 | ||
453 | HIGHWATERMARK(stattx.maxtid, t); /* statistics */ | |
454 | INCREMENT(stattx.ntid); /* statistics */ | |
455 | ||
456 | TXN_UNLOCK(); | |
457 | ||
458 | jfs_info("txBegin: returning tid = %d", t); | |
459 | ||
460 | return t; | |
461 | } | |
462 | ||
463 | ||
464 | /* | |
465 | * NAME: txBeginAnon() | |
466 | * | |
467 | * FUNCTION: start an anonymous transaction. | |
468 | * Blocks if logsync or available tlocks are low to prevent | |
469 | * anonymous tlocks from depleting supply. | |
470 | * | |
471 | * PARAMETER: sb - superblock | |
472 | * | |
473 | * RETURN: none | |
474 | */ | |
475 | void txBeginAnon(struct super_block *sb) | |
476 | { | |
477 | struct jfs_log *log; | |
478 | ||
479 | log = JFS_SBI(sb)->log; | |
480 | ||
481 | TXN_LOCK(); | |
482 | INCREMENT(TxStat.txBeginAnon); | |
483 | ||
484 | retry: | |
485 | /* | |
486 | * synchronize with logsync barrier | |
487 | */ | |
488 | if (test_bit(log_SYNCBARRIER, &log->flag) || | |
489 | test_bit(log_QUIESCE, &log->flag)) { | |
490 | INCREMENT(TxStat.txBeginAnon_barrier); | |
491 | TXN_SLEEP(&log->syncwait); | |
492 | goto retry; | |
493 | } | |
494 | ||
495 | /* | |
496 | * Don't begin transaction if we're getting starved for tlocks | |
497 | */ | |
498 | if (TxAnchor.tlocksInUse > TxLockVHWM) { | |
499 | INCREMENT(TxStat.txBeginAnon_lockslow); | |
500 | TXN_SLEEP(&TxAnchor.lowlockwait); | |
501 | goto retry; | |
502 | } | |
503 | TXN_UNLOCK(); | |
504 | } | |
505 | ||
506 | ||
507 | /* | |
508 | * txEnd() | |
509 | * | |
510 | * function: free specified transaction block. | |
511 | * | |
512 | * logsync barrier processing: | |
513 | * | |
514 | * serialization: | |
515 | */ | |
516 | void txEnd(tid_t tid) | |
517 | { | |
518 | struct tblock *tblk = tid_to_tblock(tid); | |
519 | struct jfs_log *log; | |
520 | ||
521 | jfs_info("txEnd: tid = %d", tid); | |
522 | TXN_LOCK(); | |
523 | ||
524 | /* | |
525 | * wakeup transactions waiting on the page locked | |
526 | * by the current transaction | |
527 | */ | |
528 | TXN_WAKEUP(&tblk->waitor); | |
529 | ||
530 | log = JFS_SBI(tblk->sb)->log; | |
531 | ||
532 | /* | |
533 | * Lazy commit thread can't free this guy until we mark it UNLOCKED, | |
534 | * otherwise, we would be left with a transaction that may have been | |
535 | * reused. | |
536 | * | |
537 | * Lazy commit thread will turn off tblkGC_LAZY before calling this | |
538 | * routine. | |
539 | */ | |
540 | if (tblk->flag & tblkGC_LAZY) { | |
541 | jfs_info("txEnd called w/lazy tid: %d, tblk = 0x%p", tid, tblk); | |
542 | TXN_UNLOCK(); | |
543 | ||
544 | spin_lock_irq(&log->gclock); // LOGGC_LOCK | |
545 | tblk->flag |= tblkGC_UNLOCKED; | |
546 | spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK | |
547 | return; | |
548 | } | |
549 | ||
550 | jfs_info("txEnd: tid: %d, tblk = 0x%p", tid, tblk); | |
551 | ||
552 | assert(tblk->next == 0); | |
553 | ||
554 | /* | |
555 | * insert tblock back on freelist | |
556 | */ | |
557 | tblk->next = TxAnchor.freetid; | |
558 | TxAnchor.freetid = tid; | |
559 | ||
560 | /* | |
561 | * mark the tblock not active | |
562 | */ | |
563 | if (--log->active == 0) { | |
564 | clear_bit(log_FLUSH, &log->flag); | |
565 | ||
566 | /* | |
567 | * synchronize with logsync barrier | |
568 | */ | |
569 | if (test_bit(log_SYNCBARRIER, &log->flag)) { | |
570 | /* forward log syncpt */ | |
571 | /* lmSync(log); */ | |
572 | ||
573 | jfs_info("log barrier off: 0x%x", log->lsn); | |
574 | ||
575 | /* enable new transactions start */ | |
576 | clear_bit(log_SYNCBARRIER, &log->flag); | |
577 | ||
578 | /* wakeup all waitors for logsync barrier */ | |
579 | TXN_WAKEUP(&log->syncwait); | |
580 | } | |
581 | } | |
582 | ||
583 | /* | |
584 | * wakeup all waitors for a free tblock | |
585 | */ | |
586 | TXN_WAKEUP(&TxAnchor.freewait); | |
587 | ||
588 | TXN_UNLOCK(); | |
589 | } | |
590 | ||
591 | ||
592 | /* | |
593 | * txLock() | |
594 | * | |
595 | * function: acquire a transaction lock on the specified <mp> | |
596 | * | |
597 | * parameter: | |
598 | * | |
599 | * return: transaction lock id | |
600 | * | |
601 | * serialization: | |
602 | */ | |
603 | struct tlock *txLock(tid_t tid, struct inode *ip, struct metapage * mp, | |
604 | int type) | |
605 | { | |
606 | struct jfs_inode_info *jfs_ip = JFS_IP(ip); | |
607 | int dir_xtree = 0; | |
608 | lid_t lid; | |
609 | tid_t xtid; | |
610 | struct tlock *tlck; | |
611 | struct xtlock *xtlck; | |
612 | struct linelock *linelock; | |
613 | xtpage_t *p; | |
614 | struct tblock *tblk; | |
615 | ||
616 | TXN_LOCK(); | |
617 | ||
618 | if (S_ISDIR(ip->i_mode) && (type & tlckXTREE) && | |
619 | !(mp->xflag & COMMIT_PAGE)) { | |
620 | /* | |
621 | * Directory inode is special. It can have both an xtree tlock | |
622 | * and a dtree tlock associated with it. | |
623 | */ | |
624 | dir_xtree = 1; | |
625 | lid = jfs_ip->xtlid; | |
626 | } else | |
627 | lid = mp->lid; | |
628 | ||
629 | /* is page not locked by a transaction ? */ | |
630 | if (lid == 0) | |
631 | goto allocateLock; | |
632 | ||
633 | jfs_info("txLock: tid:%d ip:0x%p mp:0x%p lid:%d", tid, ip, mp, lid); | |
634 | ||
635 | /* is page locked by the requester transaction ? */ | |
636 | tlck = lid_to_tlock(lid); | |
7fab479b DK |
637 | if ((xtid = tlck->tid) == tid) { |
638 | TXN_UNLOCK(); | |
1da177e4 | 639 | goto grantLock; |
7fab479b | 640 | } |
1da177e4 LT |
641 | |
642 | /* | |
643 | * is page locked by anonymous transaction/lock ? | |
644 | * | |
645 | * (page update without transaction (i.e., file write) is | |
646 | * locked under anonymous transaction tid = 0: | |
647 | * anonymous tlocks maintained on anonymous tlock list of | |
648 | * the inode of the page and available to all anonymous | |
649 | * transactions until txCommit() time at which point | |
650 | * they are transferred to the transaction tlock list of | |
651 | * the commiting transaction of the inode) | |
652 | */ | |
653 | if (xtid == 0) { | |
654 | tlck->tid = tid; | |
7fab479b | 655 | TXN_UNLOCK(); |
1da177e4 LT |
656 | tblk = tid_to_tblock(tid); |
657 | /* | |
658 | * The order of the tlocks in the transaction is important | |
659 | * (during truncate, child xtree pages must be freed before | |
660 | * parent's tlocks change the working map). | |
661 | * Take tlock off anonymous list and add to tail of | |
662 | * transaction list | |
663 | * | |
664 | * Note: We really need to get rid of the tid & lid and | |
665 | * use list_head's. This code is getting UGLY! | |
666 | */ | |
667 | if (jfs_ip->atlhead == lid) { | |
668 | if (jfs_ip->atltail == lid) { | |
669 | /* only anonymous txn. | |
670 | * Remove from anon_list | |
671 | */ | |
672 | list_del_init(&jfs_ip->anon_inode_list); | |
673 | } | |
674 | jfs_ip->atlhead = tlck->next; | |
675 | } else { | |
676 | lid_t last; | |
677 | for (last = jfs_ip->atlhead; | |
678 | lid_to_tlock(last)->next != lid; | |
679 | last = lid_to_tlock(last)->next) { | |
680 | assert(last); | |
681 | } | |
682 | lid_to_tlock(last)->next = tlck->next; | |
683 | if (jfs_ip->atltail == lid) | |
684 | jfs_ip->atltail = last; | |
685 | } | |
686 | ||
687 | /* insert the tlock at tail of transaction tlock list */ | |
688 | ||
689 | if (tblk->next) | |
690 | lid_to_tlock(tblk->last)->next = lid; | |
691 | else | |
692 | tblk->next = lid; | |
693 | tlck->next = 0; | |
694 | tblk->last = lid; | |
695 | ||
696 | goto grantLock; | |
697 | } | |
698 | ||
699 | goto waitLock; | |
700 | ||
701 | /* | |
702 | * allocate a tlock | |
703 | */ | |
704 | allocateLock: | |
705 | lid = txLockAlloc(); | |
706 | tlck = lid_to_tlock(lid); | |
707 | ||
708 | /* | |
709 | * initialize tlock | |
710 | */ | |
711 | tlck->tid = tid; | |
712 | ||
7fab479b DK |
713 | TXN_UNLOCK(); |
714 | ||
1da177e4 LT |
715 | /* mark tlock for meta-data page */ |
716 | if (mp->xflag & COMMIT_PAGE) { | |
717 | ||
718 | tlck->flag = tlckPAGELOCK; | |
719 | ||
720 | /* mark the page dirty and nohomeok */ | |
7fab479b | 721 | metapage_nohomeok(mp); |
1da177e4 LT |
722 | |
723 | jfs_info("locking mp = 0x%p, nohomeok = %d tid = %d tlck = 0x%p", | |
7fab479b | 724 | mp, mp->nohomeok, tid, tlck); |
1da177e4 LT |
725 | |
726 | /* if anonymous transaction, and buffer is on the group | |
727 | * commit synclist, mark inode to show this. This will | |
728 | * prevent the buffer from being marked nohomeok for too | |
729 | * long a time. | |
730 | */ | |
731 | if ((tid == 0) && mp->lsn) | |
732 | set_cflag(COMMIT_Synclist, ip); | |
733 | } | |
734 | /* mark tlock for in-memory inode */ | |
735 | else | |
736 | tlck->flag = tlckINODELOCK; | |
737 | ||
738 | tlck->type = 0; | |
739 | ||
740 | /* bind the tlock and the page */ | |
741 | tlck->ip = ip; | |
742 | tlck->mp = mp; | |
743 | if (dir_xtree) | |
744 | jfs_ip->xtlid = lid; | |
745 | else | |
746 | mp->lid = lid; | |
747 | ||
748 | /* | |
749 | * enqueue transaction lock to transaction/inode | |
750 | */ | |
751 | /* insert the tlock at tail of transaction tlock list */ | |
752 | if (tid) { | |
753 | tblk = tid_to_tblock(tid); | |
754 | if (tblk->next) | |
755 | lid_to_tlock(tblk->last)->next = lid; | |
756 | else | |
757 | tblk->next = lid; | |
758 | tlck->next = 0; | |
759 | tblk->last = lid; | |
760 | } | |
761 | /* anonymous transaction: | |
762 | * insert the tlock at head of inode anonymous tlock list | |
763 | */ | |
764 | else { | |
765 | tlck->next = jfs_ip->atlhead; | |
766 | jfs_ip->atlhead = lid; | |
767 | if (tlck->next == 0) { | |
768 | /* This inode's first anonymous transaction */ | |
769 | jfs_ip->atltail = lid; | |
7fab479b | 770 | TXN_LOCK(); |
1da177e4 LT |
771 | list_add_tail(&jfs_ip->anon_inode_list, |
772 | &TxAnchor.anon_list); | |
7fab479b | 773 | TXN_UNLOCK(); |
1da177e4 LT |
774 | } |
775 | } | |
776 | ||
777 | /* initialize type dependent area for linelock */ | |
778 | linelock = (struct linelock *) & tlck->lock; | |
779 | linelock->next = 0; | |
780 | linelock->flag = tlckLINELOCK; | |
781 | linelock->maxcnt = TLOCKSHORT; | |
782 | linelock->index = 0; | |
783 | ||
784 | switch (type & tlckTYPE) { | |
785 | case tlckDTREE: | |
786 | linelock->l2linesize = L2DTSLOTSIZE; | |
787 | break; | |
788 | ||
789 | case tlckXTREE: | |
790 | linelock->l2linesize = L2XTSLOTSIZE; | |
791 | ||
792 | xtlck = (struct xtlock *) linelock; | |
793 | xtlck->header.offset = 0; | |
794 | xtlck->header.length = 2; | |
795 | ||
796 | if (type & tlckNEW) { | |
797 | xtlck->lwm.offset = XTENTRYSTART; | |
798 | } else { | |
799 | if (mp->xflag & COMMIT_PAGE) | |
800 | p = (xtpage_t *) mp->data; | |
801 | else | |
802 | p = &jfs_ip->i_xtroot; | |
803 | xtlck->lwm.offset = | |
804 | le16_to_cpu(p->header.nextindex); | |
805 | } | |
806 | xtlck->lwm.length = 0; /* ! */ | |
807 | xtlck->twm.offset = 0; | |
808 | xtlck->hwm.offset = 0; | |
809 | ||
810 | xtlck->index = 2; | |
811 | break; | |
812 | ||
813 | case tlckINODE: | |
814 | linelock->l2linesize = L2INODESLOTSIZE; | |
815 | break; | |
816 | ||
817 | case tlckDATA: | |
818 | linelock->l2linesize = L2DATASLOTSIZE; | |
819 | break; | |
820 | ||
821 | default: | |
822 | jfs_err("UFO tlock:0x%p", tlck); | |
823 | } | |
824 | ||
825 | /* | |
826 | * update tlock vector | |
827 | */ | |
828 | grantLock: | |
829 | tlck->type |= type; | |
830 | ||
1da177e4 LT |
831 | return tlck; |
832 | ||
833 | /* | |
834 | * page is being locked by another transaction: | |
835 | */ | |
836 | waitLock: | |
837 | /* Only locks on ipimap or ipaimap should reach here */ | |
838 | /* assert(jfs_ip->fileset == AGGREGATE_I); */ | |
839 | if (jfs_ip->fileset != AGGREGATE_I) { | |
840 | jfs_err("txLock: trying to lock locked page!"); | |
841 | dump_mem("ip", ip, sizeof(struct inode)); | |
842 | dump_mem("mp", mp, sizeof(struct metapage)); | |
843 | dump_mem("Locker's tblk", tid_to_tblock(tid), | |
844 | sizeof(struct tblock)); | |
845 | dump_mem("Tlock", tlck, sizeof(struct tlock)); | |
846 | BUG(); | |
847 | } | |
848 | INCREMENT(stattx.waitlock); /* statistics */ | |
7fab479b | 849 | TXN_UNLOCK(); |
1da177e4 | 850 | release_metapage(mp); |
7fab479b DK |
851 | TXN_LOCK(); |
852 | xtid = tlck->tid; /* reaquire after dropping TXN_LOCK */ | |
1da177e4 LT |
853 | |
854 | jfs_info("txLock: in waitLock, tid = %d, xtid = %d, lid = %d", | |
855 | tid, xtid, lid); | |
7fab479b DK |
856 | |
857 | /* Recheck everything since dropping TXN_LOCK */ | |
858 | if (xtid && (tlck->mp == mp) && (mp->lid == lid)) | |
859 | TXN_SLEEP_DROP_LOCK(&tid_to_tblock(xtid)->waitor); | |
860 | else | |
861 | TXN_UNLOCK(); | |
1da177e4 LT |
862 | jfs_info("txLock: awakened tid = %d, lid = %d", tid, lid); |
863 | ||
864 | return NULL; | |
865 | } | |
866 | ||
867 | ||
868 | /* | |
869 | * NAME: txRelease() | |
870 | * | |
871 | * FUNCTION: Release buffers associated with transaction locks, but don't | |
872 | * mark homeok yet. The allows other transactions to modify | |
873 | * buffers, but won't let them go to disk until commit record | |
874 | * actually gets written. | |
875 | * | |
876 | * PARAMETER: | |
877 | * tblk - | |
878 | * | |
879 | * RETURN: Errors from subroutines. | |
880 | */ | |
881 | static void txRelease(struct tblock * tblk) | |
882 | { | |
883 | struct metapage *mp; | |
884 | lid_t lid; | |
885 | struct tlock *tlck; | |
886 | ||
887 | TXN_LOCK(); | |
888 | ||
889 | for (lid = tblk->next; lid; lid = tlck->next) { | |
890 | tlck = lid_to_tlock(lid); | |
891 | if ((mp = tlck->mp) != NULL && | |
892 | (tlck->type & tlckBTROOT) == 0) { | |
893 | assert(mp->xflag & COMMIT_PAGE); | |
894 | mp->lid = 0; | |
895 | } | |
896 | } | |
897 | ||
898 | /* | |
899 | * wakeup transactions waiting on a page locked | |
900 | * by the current transaction | |
901 | */ | |
902 | TXN_WAKEUP(&tblk->waitor); | |
903 | ||
904 | TXN_UNLOCK(); | |
905 | } | |
906 | ||
907 | ||
908 | /* | |
909 | * NAME: txUnlock() | |
910 | * | |
911 | * FUNCTION: Initiates pageout of pages modified by tid in journalled | |
912 | * objects and frees their lockwords. | |
913 | */ | |
914 | static void txUnlock(struct tblock * tblk) | |
915 | { | |
916 | struct tlock *tlck; | |
917 | struct linelock *linelock; | |
918 | lid_t lid, next, llid, k; | |
919 | struct metapage *mp; | |
920 | struct jfs_log *log; | |
921 | int difft, diffp; | |
7fab479b | 922 | unsigned long flags; |
1da177e4 LT |
923 | |
924 | jfs_info("txUnlock: tblk = 0x%p", tblk); | |
925 | log = JFS_SBI(tblk->sb)->log; | |
926 | ||
927 | /* | |
928 | * mark page under tlock homeok (its log has been written): | |
929 | */ | |
930 | for (lid = tblk->next; lid; lid = next) { | |
931 | tlck = lid_to_tlock(lid); | |
932 | next = tlck->next; | |
933 | ||
934 | jfs_info("unlocking lid = %d, tlck = 0x%p", lid, tlck); | |
935 | ||
936 | /* unbind page from tlock */ | |
937 | if ((mp = tlck->mp) != NULL && | |
938 | (tlck->type & tlckBTROOT) == 0) { | |
939 | assert(mp->xflag & COMMIT_PAGE); | |
940 | ||
941 | /* hold buffer | |
1da177e4 | 942 | */ |
7fab479b | 943 | hold_metapage(mp); |
1da177e4 | 944 | |
7fab479b DK |
945 | assert(mp->nohomeok > 0); |
946 | _metapage_homeok(mp); | |
1da177e4 LT |
947 | |
948 | /* inherit younger/larger clsn */ | |
7fab479b | 949 | LOGSYNC_LOCK(log, flags); |
1da177e4 LT |
950 | if (mp->clsn) { |
951 | logdiff(difft, tblk->clsn, log); | |
952 | logdiff(diffp, mp->clsn, log); | |
953 | if (difft > diffp) | |
954 | mp->clsn = tblk->clsn; | |
955 | } else | |
956 | mp->clsn = tblk->clsn; | |
7fab479b | 957 | LOGSYNC_UNLOCK(log, flags); |
1da177e4 LT |
958 | |
959 | assert(!(tlck->flag & tlckFREEPAGE)); | |
960 | ||
7fab479b | 961 | put_metapage(mp); |
1da177e4 LT |
962 | } |
963 | ||
964 | /* insert tlock, and linelock(s) of the tlock if any, | |
965 | * at head of freelist | |
966 | */ | |
967 | TXN_LOCK(); | |
968 | ||
969 | llid = ((struct linelock *) & tlck->lock)->next; | |
970 | while (llid) { | |
971 | linelock = (struct linelock *) lid_to_tlock(llid); | |
972 | k = linelock->next; | |
973 | txLockFree(llid); | |
974 | llid = k; | |
975 | } | |
976 | txLockFree(lid); | |
977 | ||
978 | TXN_UNLOCK(); | |
979 | } | |
980 | tblk->next = tblk->last = 0; | |
981 | ||
982 | /* | |
983 | * remove tblock from logsynclist | |
984 | * (allocation map pages inherited lsn of tblk and | |
985 | * has been inserted in logsync list at txUpdateMap()) | |
986 | */ | |
987 | if (tblk->lsn) { | |
7fab479b | 988 | LOGSYNC_LOCK(log, flags); |
1da177e4 LT |
989 | log->count--; |
990 | list_del(&tblk->synclist); | |
7fab479b | 991 | LOGSYNC_UNLOCK(log, flags); |
1da177e4 LT |
992 | } |
993 | } | |
994 | ||
995 | ||
996 | /* | |
997 | * txMaplock() | |
998 | * | |
999 | * function: allocate a transaction lock for freed page/entry; | |
1000 | * for freed page, maplock is used as xtlock/dtlock type; | |
1001 | */ | |
1002 | struct tlock *txMaplock(tid_t tid, struct inode *ip, int type) | |
1003 | { | |
1004 | struct jfs_inode_info *jfs_ip = JFS_IP(ip); | |
1005 | lid_t lid; | |
1006 | struct tblock *tblk; | |
1007 | struct tlock *tlck; | |
1008 | struct maplock *maplock; | |
1009 | ||
1010 | TXN_LOCK(); | |
1011 | ||
1012 | /* | |
1013 | * allocate a tlock | |
1014 | */ | |
1015 | lid = txLockAlloc(); | |
1016 | tlck = lid_to_tlock(lid); | |
1017 | ||
1018 | /* | |
1019 | * initialize tlock | |
1020 | */ | |
1021 | tlck->tid = tid; | |
1022 | ||
1023 | /* bind the tlock and the object */ | |
1024 | tlck->flag = tlckINODELOCK; | |
1025 | tlck->ip = ip; | |
1026 | tlck->mp = NULL; | |
1027 | ||
1028 | tlck->type = type; | |
1029 | ||
1030 | /* | |
1031 | * enqueue transaction lock to transaction/inode | |
1032 | */ | |
1033 | /* insert the tlock at tail of transaction tlock list */ | |
1034 | if (tid) { | |
1035 | tblk = tid_to_tblock(tid); | |
1036 | if (tblk->next) | |
1037 | lid_to_tlock(tblk->last)->next = lid; | |
1038 | else | |
1039 | tblk->next = lid; | |
1040 | tlck->next = 0; | |
1041 | tblk->last = lid; | |
1042 | } | |
1043 | /* anonymous transaction: | |
1044 | * insert the tlock at head of inode anonymous tlock list | |
1045 | */ | |
1046 | else { | |
1047 | tlck->next = jfs_ip->atlhead; | |
1048 | jfs_ip->atlhead = lid; | |
1049 | if (tlck->next == 0) { | |
1050 | /* This inode's first anonymous transaction */ | |
1051 | jfs_ip->atltail = lid; | |
1052 | list_add_tail(&jfs_ip->anon_inode_list, | |
1053 | &TxAnchor.anon_list); | |
1054 | } | |
1055 | } | |
1056 | ||
1057 | TXN_UNLOCK(); | |
1058 | ||
1059 | /* initialize type dependent area for maplock */ | |
1060 | maplock = (struct maplock *) & tlck->lock; | |
1061 | maplock->next = 0; | |
1062 | maplock->maxcnt = 0; | |
1063 | maplock->index = 0; | |
1064 | ||
1065 | return tlck; | |
1066 | } | |
1067 | ||
1068 | ||
1069 | /* | |
1070 | * txLinelock() | |
1071 | * | |
1072 | * function: allocate a transaction lock for log vector list | |
1073 | */ | |
1074 | struct linelock *txLinelock(struct linelock * tlock) | |
1075 | { | |
1076 | lid_t lid; | |
1077 | struct tlock *tlck; | |
1078 | struct linelock *linelock; | |
1079 | ||
1080 | TXN_LOCK(); | |
1081 | ||
1082 | /* allocate a TxLock structure */ | |
1083 | lid = txLockAlloc(); | |
1084 | tlck = lid_to_tlock(lid); | |
1085 | ||
1086 | TXN_UNLOCK(); | |
1087 | ||
1088 | /* initialize linelock */ | |
1089 | linelock = (struct linelock *) tlck; | |
1090 | linelock->next = 0; | |
1091 | linelock->flag = tlckLINELOCK; | |
1092 | linelock->maxcnt = TLOCKLONG; | |
1093 | linelock->index = 0; | |
1094 | ||
1095 | /* append linelock after tlock */ | |
1096 | linelock->next = tlock->next; | |
1097 | tlock->next = lid; | |
1098 | ||
1099 | return linelock; | |
1100 | } | |
1101 | ||
1102 | ||
1103 | ||
1104 | /* | |
1105 | * transaction commit management | |
1106 | * ----------------------------- | |
1107 | */ | |
1108 | ||
1109 | /* | |
1110 | * NAME: txCommit() | |
1111 | * | |
1112 | * FUNCTION: commit the changes to the objects specified in | |
1113 | * clist. For journalled segments only the | |
1114 | * changes of the caller are committed, ie by tid. | |
1115 | * for non-journalled segments the data are flushed to | |
1116 | * disk and then the change to the disk inode and indirect | |
1117 | * blocks committed (so blocks newly allocated to the | |
1118 | * segment will be made a part of the segment atomically). | |
1119 | * | |
1120 | * all of the segments specified in clist must be in | |
1121 | * one file system. no more than 6 segments are needed | |
1122 | * to handle all unix svcs. | |
1123 | * | |
1124 | * if the i_nlink field (i.e. disk inode link count) | |
1125 | * is zero, and the type of inode is a regular file or | |
1126 | * directory, or symbolic link , the inode is truncated | |
1127 | * to zero length. the truncation is committed but the | |
1128 | * VM resources are unaffected until it is closed (see | |
1129 | * iput and iclose). | |
1130 | * | |
1131 | * PARAMETER: | |
1132 | * | |
1133 | * RETURN: | |
1134 | * | |
1135 | * serialization: | |
1136 | * on entry the inode lock on each segment is assumed | |
1137 | * to be held. | |
1138 | * | |
1139 | * i/o error: | |
1140 | */ | |
1141 | int txCommit(tid_t tid, /* transaction identifier */ | |
1142 | int nip, /* number of inodes to commit */ | |
1143 | struct inode **iplist, /* list of inode to commit */ | |
1144 | int flag) | |
1145 | { | |
1146 | int rc = 0; | |
1147 | struct commit cd; | |
1148 | struct jfs_log *log; | |
1149 | struct tblock *tblk; | |
1150 | struct lrd *lrd; | |
1151 | int lsn; | |
1152 | struct inode *ip; | |
1153 | struct jfs_inode_info *jfs_ip; | |
1154 | int k, n; | |
1155 | ino_t top; | |
1156 | struct super_block *sb; | |
1157 | ||
1158 | jfs_info("txCommit, tid = %d, flag = %d", tid, flag); | |
1159 | /* is read-only file system ? */ | |
1160 | if (isReadOnly(iplist[0])) { | |
1161 | rc = -EROFS; | |
1162 | goto TheEnd; | |
1163 | } | |
1164 | ||
1165 | sb = cd.sb = iplist[0]->i_sb; | |
1166 | cd.tid = tid; | |
1167 | ||
1168 | if (tid == 0) | |
1169 | tid = txBegin(sb, 0); | |
1170 | tblk = tid_to_tblock(tid); | |
1171 | ||
1172 | /* | |
1173 | * initialize commit structure | |
1174 | */ | |
1175 | log = JFS_SBI(sb)->log; | |
1176 | cd.log = log; | |
1177 | ||
1178 | /* initialize log record descriptor in commit */ | |
1179 | lrd = &cd.lrd; | |
1180 | lrd->logtid = cpu_to_le32(tblk->logtid); | |
1181 | lrd->backchain = 0; | |
1182 | ||
1183 | tblk->xflag |= flag; | |
1184 | ||
1185 | if ((flag & (COMMIT_FORCE | COMMIT_SYNC)) == 0) | |
1186 | tblk->xflag |= COMMIT_LAZY; | |
1187 | /* | |
1188 | * prepare non-journaled objects for commit | |
1189 | * | |
1190 | * flush data pages of non-journaled file | |
1191 | * to prevent the file getting non-initialized disk blocks | |
1192 | * in case of crash. | |
1193 | * (new blocks - ) | |
1194 | */ | |
1195 | cd.iplist = iplist; | |
1196 | cd.nip = nip; | |
1197 | ||
1198 | /* | |
1199 | * acquire transaction lock on (on-disk) inodes | |
1200 | * | |
1201 | * update on-disk inode from in-memory inode | |
1202 | * acquiring transaction locks for AFTER records | |
1203 | * on the on-disk inode of file object | |
1204 | * | |
1205 | * sort the inodes array by inode number in descending order | |
1206 | * to prevent deadlock when acquiring transaction lock | |
1207 | * of on-disk inodes on multiple on-disk inode pages by | |
1208 | * multiple concurrent transactions | |
1209 | */ | |
1210 | for (k = 0; k < cd.nip; k++) { | |
1211 | top = (cd.iplist[k])->i_ino; | |
1212 | for (n = k + 1; n < cd.nip; n++) { | |
1213 | ip = cd.iplist[n]; | |
1214 | if (ip->i_ino > top) { | |
1215 | top = ip->i_ino; | |
1216 | cd.iplist[n] = cd.iplist[k]; | |
1217 | cd.iplist[k] = ip; | |
1218 | } | |
1219 | } | |
1220 | ||
1221 | ip = cd.iplist[k]; | |
1222 | jfs_ip = JFS_IP(ip); | |
1223 | ||
1224 | /* | |
1225 | * BUGBUG - This code has temporarily been removed. The | |
1226 | * intent is to ensure that any file data is written before | |
1227 | * the metadata is committed to the journal. This prevents | |
1228 | * uninitialized data from appearing in a file after the | |
1229 | * journal has been replayed. (The uninitialized data | |
1230 | * could be sensitive data removed by another user.) | |
1231 | * | |
1232 | * The problem now is that we are holding the IWRITELOCK | |
1233 | * on the inode, and calling filemap_fdatawrite on an | |
1234 | * unmapped page will cause a deadlock in jfs_get_block. | |
1235 | * | |
1236 | * The long term solution is to pare down the use of | |
1237 | * IWRITELOCK. We are currently holding it too long. | |
1238 | * We could also be smarter about which data pages need | |
1239 | * to be written before the transaction is committed and | |
1240 | * when we don't need to worry about it at all. | |
1241 | * | |
1242 | * if ((!S_ISDIR(ip->i_mode)) | |
1243 | * && (tblk->flag & COMMIT_DELETE) == 0) { | |
1244 | * filemap_fdatawrite(ip->i_mapping); | |
1245 | * filemap_fdatawait(ip->i_mapping); | |
1246 | * } | |
1247 | */ | |
1248 | ||
1249 | /* | |
1250 | * Mark inode as not dirty. It will still be on the dirty | |
1251 | * inode list, but we'll know not to commit it again unless | |
1252 | * it gets marked dirty again | |
1253 | */ | |
1254 | clear_cflag(COMMIT_Dirty, ip); | |
1255 | ||
1256 | /* inherit anonymous tlock(s) of inode */ | |
1257 | if (jfs_ip->atlhead) { | |
1258 | lid_to_tlock(jfs_ip->atltail)->next = tblk->next; | |
1259 | tblk->next = jfs_ip->atlhead; | |
1260 | if (!tblk->last) | |
1261 | tblk->last = jfs_ip->atltail; | |
1262 | jfs_ip->atlhead = jfs_ip->atltail = 0; | |
1263 | TXN_LOCK(); | |
1264 | list_del_init(&jfs_ip->anon_inode_list); | |
1265 | TXN_UNLOCK(); | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * acquire transaction lock on on-disk inode page | |
1270 | * (become first tlock of the tblk's tlock list) | |
1271 | */ | |
1272 | if (((rc = diWrite(tid, ip)))) | |
1273 | goto out; | |
1274 | } | |
1275 | ||
1276 | /* | |
1277 | * write log records from transaction locks | |
1278 | * | |
1279 | * txUpdateMap() resets XAD_NEW in XAD. | |
1280 | */ | |
1281 | if ((rc = txLog(log, tblk, &cd))) | |
1282 | goto TheEnd; | |
1283 | ||
1284 | /* | |
1285 | * Ensure that inode isn't reused before | |
1286 | * lazy commit thread finishes processing | |
1287 | */ | |
1288 | if (tblk->xflag & COMMIT_DELETE) { | |
1289 | atomic_inc(&tblk->u.ip->i_count); | |
1290 | /* | |
1291 | * Avoid a rare deadlock | |
1292 | * | |
1293 | * If the inode is locked, we may be blocked in | |
1294 | * jfs_commit_inode. If so, we don't want the | |
1295 | * lazy_commit thread doing the last iput() on the inode | |
1296 | * since that may block on the locked inode. Instead, | |
1297 | * commit the transaction synchronously, so the last iput | |
1298 | * will be done by the calling thread (or later) | |
1299 | */ | |
1300 | if (tblk->u.ip->i_state & I_LOCK) | |
1301 | tblk->xflag &= ~COMMIT_LAZY; | |
1302 | } | |
1303 | ||
1304 | ASSERT((!(tblk->xflag & COMMIT_DELETE)) || | |
1305 | ((tblk->u.ip->i_nlink == 0) && | |
1306 | !test_cflag(COMMIT_Nolink, tblk->u.ip))); | |
1307 | ||
1308 | /* | |
1309 | * write COMMIT log record | |
1310 | */ | |
1311 | lrd->type = cpu_to_le16(LOG_COMMIT); | |
1312 | lrd->length = 0; | |
1313 | lsn = lmLog(log, tblk, lrd, NULL); | |
1314 | ||
1315 | lmGroupCommit(log, tblk); | |
1316 | ||
1317 | /* | |
1318 | * - transaction is now committed - | |
1319 | */ | |
1320 | ||
1321 | /* | |
1322 | * force pages in careful update | |
1323 | * (imap addressing structure update) | |
1324 | */ | |
1325 | if (flag & COMMIT_FORCE) | |
1326 | txForce(tblk); | |
1327 | ||
1328 | /* | |
1329 | * update allocation map. | |
1330 | * | |
1331 | * update inode allocation map and inode: | |
1332 | * free pager lock on memory object of inode if any. | |
1333 | * update block allocation map. | |
1334 | * | |
1335 | * txUpdateMap() resets XAD_NEW in XAD. | |
1336 | */ | |
1337 | if (tblk->xflag & COMMIT_FORCE) | |
1338 | txUpdateMap(tblk); | |
1339 | ||
1340 | /* | |
1341 | * free transaction locks and pageout/free pages | |
1342 | */ | |
1343 | txRelease(tblk); | |
1344 | ||
1345 | if ((tblk->flag & tblkGC_LAZY) == 0) | |
1346 | txUnlock(tblk); | |
1347 | ||
1348 | ||
1349 | /* | |
1350 | * reset in-memory object state | |
1351 | */ | |
1352 | for (k = 0; k < cd.nip; k++) { | |
1353 | ip = cd.iplist[k]; | |
1354 | jfs_ip = JFS_IP(ip); | |
1355 | ||
1356 | /* | |
1357 | * reset in-memory inode state | |
1358 | */ | |
1359 | jfs_ip->bxflag = 0; | |
1360 | jfs_ip->blid = 0; | |
1361 | } | |
1362 | ||
1363 | out: | |
1364 | if (rc != 0) | |
1365 | txAbort(tid, 1); | |
1366 | ||
1367 | TheEnd: | |
1368 | jfs_info("txCommit: tid = %d, returning %d", tid, rc); | |
1369 | return rc; | |
1370 | } | |
1371 | ||
1372 | ||
1373 | /* | |
1374 | * NAME: txLog() | |
1375 | * | |
1376 | * FUNCTION: Writes AFTER log records for all lines modified | |
1377 | * by tid for segments specified by inodes in comdata. | |
1378 | * Code assumes only WRITELOCKS are recorded in lockwords. | |
1379 | * | |
1380 | * PARAMETERS: | |
1381 | * | |
1382 | * RETURN : | |
1383 | */ | |
1384 | static int txLog(struct jfs_log * log, struct tblock * tblk, struct commit * cd) | |
1385 | { | |
1386 | int rc = 0; | |
1387 | struct inode *ip; | |
1388 | lid_t lid; | |
1389 | struct tlock *tlck; | |
1390 | struct lrd *lrd = &cd->lrd; | |
1391 | ||
1392 | /* | |
1393 | * write log record(s) for each tlock of transaction, | |
1394 | */ | |
1395 | for (lid = tblk->next; lid; lid = tlck->next) { | |
1396 | tlck = lid_to_tlock(lid); | |
1397 | ||
1398 | tlck->flag |= tlckLOG; | |
1399 | ||
1400 | /* initialize lrd common */ | |
1401 | ip = tlck->ip; | |
1402 | lrd->aggregate = cpu_to_le32(JFS_SBI(ip->i_sb)->aggregate); | |
1403 | lrd->log.redopage.fileset = cpu_to_le32(JFS_IP(ip)->fileset); | |
1404 | lrd->log.redopage.inode = cpu_to_le32(ip->i_ino); | |
1405 | ||
1406 | /* write log record of page from the tlock */ | |
1407 | switch (tlck->type & tlckTYPE) { | |
1408 | case tlckXTREE: | |
1409 | xtLog(log, tblk, lrd, tlck); | |
1410 | break; | |
1411 | ||
1412 | case tlckDTREE: | |
1413 | dtLog(log, tblk, lrd, tlck); | |
1414 | break; | |
1415 | ||
1416 | case tlckINODE: | |
1417 | diLog(log, tblk, lrd, tlck, cd); | |
1418 | break; | |
1419 | ||
1420 | case tlckMAP: | |
1421 | mapLog(log, tblk, lrd, tlck); | |
1422 | break; | |
1423 | ||
1424 | case tlckDATA: | |
1425 | dataLog(log, tblk, lrd, tlck); | |
1426 | break; | |
1427 | ||
1428 | default: | |
1429 | jfs_err("UFO tlock:0x%p", tlck); | |
1430 | } | |
1431 | } | |
1432 | ||
1433 | return rc; | |
1434 | } | |
1435 | ||
1436 | ||
1437 | /* | |
1438 | * diLog() | |
1439 | * | |
1440 | * function: log inode tlock and format maplock to update bmap; | |
1441 | */ | |
1442 | static int diLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
1443 | struct tlock * tlck, struct commit * cd) | |
1444 | { | |
1445 | int rc = 0; | |
1446 | struct metapage *mp; | |
1447 | pxd_t *pxd; | |
1448 | struct pxd_lock *pxdlock; | |
1449 | ||
1450 | mp = tlck->mp; | |
1451 | ||
1452 | /* initialize as REDOPAGE record format */ | |
1453 | lrd->log.redopage.type = cpu_to_le16(LOG_INODE); | |
1454 | lrd->log.redopage.l2linesize = cpu_to_le16(L2INODESLOTSIZE); | |
1455 | ||
1456 | pxd = &lrd->log.redopage.pxd; | |
1457 | ||
1458 | /* | |
1459 | * inode after image | |
1460 | */ | |
1461 | if (tlck->type & tlckENTRY) { | |
1462 | /* log after-image for logredo(): */ | |
1463 | lrd->type = cpu_to_le16(LOG_REDOPAGE); | |
1464 | // *pxd = mp->cm_pxd; | |
1465 | PXDaddress(pxd, mp->index); | |
1466 | PXDlength(pxd, | |
1467 | mp->logical_size >> tblk->sb->s_blocksize_bits); | |
1468 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); | |
1469 | ||
1470 | /* mark page as homeward bound */ | |
1471 | tlck->flag |= tlckWRITEPAGE; | |
1472 | } else if (tlck->type & tlckFREE) { | |
1473 | /* | |
1474 | * free inode extent | |
1475 | * | |
1476 | * (pages of the freed inode extent have been invalidated and | |
1477 | * a maplock for free of the extent has been formatted at | |
1478 | * txLock() time); | |
1479 | * | |
1480 | * the tlock had been acquired on the inode allocation map page | |
1481 | * (iag) that specifies the freed extent, even though the map | |
1482 | * page is not itself logged, to prevent pageout of the map | |
1483 | * page before the log; | |
1484 | */ | |
1485 | ||
1486 | /* log LOG_NOREDOINOEXT of the freed inode extent for | |
1487 | * logredo() to start NoRedoPage filters, and to update | |
1488 | * imap and bmap for free of the extent; | |
1489 | */ | |
1490 | lrd->type = cpu_to_le16(LOG_NOREDOINOEXT); | |
1491 | /* | |
1492 | * For the LOG_NOREDOINOEXT record, we need | |
1493 | * to pass the IAG number and inode extent | |
1494 | * index (within that IAG) from which the | |
1495 | * the extent being released. These have been | |
1496 | * passed to us in the iplist[1] and iplist[2]. | |
1497 | */ | |
1498 | lrd->log.noredoinoext.iagnum = | |
1499 | cpu_to_le32((u32) (size_t) cd->iplist[1]); | |
1500 | lrd->log.noredoinoext.inoext_idx = | |
1501 | cpu_to_le32((u32) (size_t) cd->iplist[2]); | |
1502 | ||
1503 | pxdlock = (struct pxd_lock *) & tlck->lock; | |
1504 | *pxd = pxdlock->pxd; | |
1505 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL)); | |
1506 | ||
1507 | /* update bmap */ | |
1508 | tlck->flag |= tlckUPDATEMAP; | |
1509 | ||
1510 | /* mark page as homeward bound */ | |
1511 | tlck->flag |= tlckWRITEPAGE; | |
1512 | } else | |
1513 | jfs_err("diLog: UFO type tlck:0x%p", tlck); | |
1514 | #ifdef _JFS_WIP | |
1515 | /* | |
1516 | * alloc/free external EA extent | |
1517 | * | |
1518 | * a maplock for txUpdateMap() to update bPWMAP for alloc/free | |
1519 | * of the extent has been formatted at txLock() time; | |
1520 | */ | |
1521 | else { | |
1522 | assert(tlck->type & tlckEA); | |
1523 | ||
1524 | /* log LOG_UPDATEMAP for logredo() to update bmap for | |
1525 | * alloc of new (and free of old) external EA extent; | |
1526 | */ | |
1527 | lrd->type = cpu_to_le16(LOG_UPDATEMAP); | |
1528 | pxdlock = (struct pxd_lock *) & tlck->lock; | |
1529 | nlock = pxdlock->index; | |
1530 | for (i = 0; i < nlock; i++, pxdlock++) { | |
1531 | if (pxdlock->flag & mlckALLOCPXD) | |
1532 | lrd->log.updatemap.type = | |
1533 | cpu_to_le16(LOG_ALLOCPXD); | |
1534 | else | |
1535 | lrd->log.updatemap.type = | |
1536 | cpu_to_le16(LOG_FREEPXD); | |
1537 | lrd->log.updatemap.nxd = cpu_to_le16(1); | |
1538 | lrd->log.updatemap.pxd = pxdlock->pxd; | |
1539 | lrd->backchain = | |
1540 | cpu_to_le32(lmLog(log, tblk, lrd, NULL)); | |
1541 | } | |
1542 | ||
1543 | /* update bmap */ | |
1544 | tlck->flag |= tlckUPDATEMAP; | |
1545 | } | |
1546 | #endif /* _JFS_WIP */ | |
1547 | ||
1548 | return rc; | |
1549 | } | |
1550 | ||
1551 | ||
1552 | /* | |
1553 | * dataLog() | |
1554 | * | |
1555 | * function: log data tlock | |
1556 | */ | |
1557 | static int dataLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
1558 | struct tlock * tlck) | |
1559 | { | |
1560 | struct metapage *mp; | |
1561 | pxd_t *pxd; | |
1562 | ||
1563 | mp = tlck->mp; | |
1564 | ||
1565 | /* initialize as REDOPAGE record format */ | |
1566 | lrd->log.redopage.type = cpu_to_le16(LOG_DATA); | |
1567 | lrd->log.redopage.l2linesize = cpu_to_le16(L2DATASLOTSIZE); | |
1568 | ||
1569 | pxd = &lrd->log.redopage.pxd; | |
1570 | ||
1571 | /* log after-image for logredo(): */ | |
1572 | lrd->type = cpu_to_le16(LOG_REDOPAGE); | |
1573 | ||
1574 | if (jfs_dirtable_inline(tlck->ip)) { | |
1575 | /* | |
1576 | * The table has been truncated, we've must have deleted | |
1577 | * the last entry, so don't bother logging this | |
1578 | */ | |
1579 | mp->lid = 0; | |
7fab479b DK |
1580 | grab_metapage(mp); |
1581 | metapage_homeok(mp); | |
1da177e4 LT |
1582 | discard_metapage(mp); |
1583 | tlck->mp = NULL; | |
1584 | return 0; | |
1585 | } | |
1586 | ||
1587 | PXDaddress(pxd, mp->index); | |
1588 | PXDlength(pxd, mp->logical_size >> tblk->sb->s_blocksize_bits); | |
1589 | ||
1590 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); | |
1591 | ||
1592 | /* mark page as homeward bound */ | |
1593 | tlck->flag |= tlckWRITEPAGE; | |
1594 | ||
1595 | return 0; | |
1596 | } | |
1597 | ||
1598 | ||
1599 | /* | |
1600 | * dtLog() | |
1601 | * | |
1602 | * function: log dtree tlock and format maplock to update bmap; | |
1603 | */ | |
1604 | static void dtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
1605 | struct tlock * tlck) | |
1606 | { | |
1607 | struct metapage *mp; | |
1608 | struct pxd_lock *pxdlock; | |
1609 | pxd_t *pxd; | |
1610 | ||
1611 | mp = tlck->mp; | |
1612 | ||
1613 | /* initialize as REDOPAGE/NOREDOPAGE record format */ | |
1614 | lrd->log.redopage.type = cpu_to_le16(LOG_DTREE); | |
1615 | lrd->log.redopage.l2linesize = cpu_to_le16(L2DTSLOTSIZE); | |
1616 | ||
1617 | pxd = &lrd->log.redopage.pxd; | |
1618 | ||
1619 | if (tlck->type & tlckBTROOT) | |
1620 | lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT); | |
1621 | ||
1622 | /* | |
1623 | * page extension via relocation: entry insertion; | |
1624 | * page extension in-place: entry insertion; | |
1625 | * new right page from page split, reinitialized in-line | |
1626 | * root from root page split: entry insertion; | |
1627 | */ | |
1628 | if (tlck->type & (tlckNEW | tlckEXTEND)) { | |
1629 | /* log after-image of the new page for logredo(): | |
1630 | * mark log (LOG_NEW) for logredo() to initialize | |
1631 | * freelist and update bmap for alloc of the new page; | |
1632 | */ | |
1633 | lrd->type = cpu_to_le16(LOG_REDOPAGE); | |
1634 | if (tlck->type & tlckEXTEND) | |
1635 | lrd->log.redopage.type |= cpu_to_le16(LOG_EXTEND); | |
1636 | else | |
1637 | lrd->log.redopage.type |= cpu_to_le16(LOG_NEW); | |
1638 | // *pxd = mp->cm_pxd; | |
1639 | PXDaddress(pxd, mp->index); | |
1640 | PXDlength(pxd, | |
1641 | mp->logical_size >> tblk->sb->s_blocksize_bits); | |
1642 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); | |
1643 | ||
1644 | /* format a maplock for txUpdateMap() to update bPMAP for | |
1645 | * alloc of the new page; | |
1646 | */ | |
1647 | if (tlck->type & tlckBTROOT) | |
1648 | return; | |
1649 | tlck->flag |= tlckUPDATEMAP; | |
1650 | pxdlock = (struct pxd_lock *) & tlck->lock; | |
1651 | pxdlock->flag = mlckALLOCPXD; | |
1652 | pxdlock->pxd = *pxd; | |
1653 | ||
1654 | pxdlock->index = 1; | |
1655 | ||
1656 | /* mark page as homeward bound */ | |
1657 | tlck->flag |= tlckWRITEPAGE; | |
1658 | return; | |
1659 | } | |
1660 | ||
1661 | /* | |
1662 | * entry insertion/deletion, | |
1663 | * sibling page link update (old right page before split); | |
1664 | */ | |
1665 | if (tlck->type & (tlckENTRY | tlckRELINK)) { | |
1666 | /* log after-image for logredo(): */ | |
1667 | lrd->type = cpu_to_le16(LOG_REDOPAGE); | |
1668 | PXDaddress(pxd, mp->index); | |
1669 | PXDlength(pxd, | |
1670 | mp->logical_size >> tblk->sb->s_blocksize_bits); | |
1671 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); | |
1672 | ||
1673 | /* mark page as homeward bound */ | |
1674 | tlck->flag |= tlckWRITEPAGE; | |
1675 | return; | |
1676 | } | |
1677 | ||
1678 | /* | |
1679 | * page deletion: page has been invalidated | |
1680 | * page relocation: source extent | |
1681 | * | |
1682 | * a maplock for free of the page has been formatted | |
1683 | * at txLock() time); | |
1684 | */ | |
1685 | if (tlck->type & (tlckFREE | tlckRELOCATE)) { | |
1686 | /* log LOG_NOREDOPAGE of the deleted page for logredo() | |
1687 | * to start NoRedoPage filter and to update bmap for free | |
1688 | * of the deletd page | |
1689 | */ | |
1690 | lrd->type = cpu_to_le16(LOG_NOREDOPAGE); | |
1691 | pxdlock = (struct pxd_lock *) & tlck->lock; | |
1692 | *pxd = pxdlock->pxd; | |
1693 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL)); | |
1694 | ||
1695 | /* a maplock for txUpdateMap() for free of the page | |
1696 | * has been formatted at txLock() time; | |
1697 | */ | |
1698 | tlck->flag |= tlckUPDATEMAP; | |
1699 | } | |
1700 | return; | |
1701 | } | |
1702 | ||
1703 | ||
1704 | /* | |
1705 | * xtLog() | |
1706 | * | |
1707 | * function: log xtree tlock and format maplock to update bmap; | |
1708 | */ | |
1709 | static void xtLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
1710 | struct tlock * tlck) | |
1711 | { | |
1712 | struct inode *ip; | |
1713 | struct metapage *mp; | |
1714 | xtpage_t *p; | |
1715 | struct xtlock *xtlck; | |
1716 | struct maplock *maplock; | |
1717 | struct xdlistlock *xadlock; | |
1718 | struct pxd_lock *pxdlock; | |
66f3131f | 1719 | pxd_t *page_pxd; |
1da177e4 LT |
1720 | int next, lwm, hwm; |
1721 | ||
1722 | ip = tlck->ip; | |
1723 | mp = tlck->mp; | |
1724 | ||
1725 | /* initialize as REDOPAGE/NOREDOPAGE record format */ | |
1726 | lrd->log.redopage.type = cpu_to_le16(LOG_XTREE); | |
1727 | lrd->log.redopage.l2linesize = cpu_to_le16(L2XTSLOTSIZE); | |
1728 | ||
66f3131f | 1729 | page_pxd = &lrd->log.redopage.pxd; |
1da177e4 LT |
1730 | |
1731 | if (tlck->type & tlckBTROOT) { | |
1732 | lrd->log.redopage.type |= cpu_to_le16(LOG_BTROOT); | |
1733 | p = &JFS_IP(ip)->i_xtroot; | |
1734 | if (S_ISDIR(ip->i_mode)) | |
1735 | lrd->log.redopage.type |= | |
1736 | cpu_to_le16(LOG_DIR_XTREE); | |
1737 | } else | |
1738 | p = (xtpage_t *) mp->data; | |
1739 | next = le16_to_cpu(p->header.nextindex); | |
1740 | ||
1741 | xtlck = (struct xtlock *) & tlck->lock; | |
1742 | ||
1743 | maplock = (struct maplock *) & tlck->lock; | |
1744 | xadlock = (struct xdlistlock *) maplock; | |
1745 | ||
1746 | /* | |
1747 | * entry insertion/extension; | |
1748 | * sibling page link update (old right page before split); | |
1749 | */ | |
1750 | if (tlck->type & (tlckNEW | tlckGROW | tlckRELINK)) { | |
1751 | /* log after-image for logredo(): | |
1752 | * logredo() will update bmap for alloc of new/extended | |
1753 | * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from | |
1754 | * after-image of XADlist; | |
1755 | * logredo() resets (XAD_NEW|XAD_EXTEND) flag when | |
1756 | * applying the after-image to the meta-data page. | |
1757 | */ | |
1758 | lrd->type = cpu_to_le16(LOG_REDOPAGE); | |
66f3131f DK |
1759 | // *page_pxd = mp->cm_pxd; |
1760 | PXDaddress(page_pxd, mp->index); | |
1761 | PXDlength(page_pxd, | |
1da177e4 LT |
1762 | mp->logical_size >> tblk->sb->s_blocksize_bits); |
1763 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); | |
1764 | ||
1765 | /* format a maplock for txUpdateMap() to update bPMAP | |
1766 | * for alloc of new/extended extents of XAD[lwm:next) | |
1767 | * from the page itself; | |
1768 | * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag. | |
1769 | */ | |
1770 | lwm = xtlck->lwm.offset; | |
1771 | if (lwm == 0) | |
1772 | lwm = XTPAGEMAXSLOT; | |
1773 | ||
1774 | if (lwm == next) | |
1775 | goto out; | |
1776 | if (lwm > next) { | |
1777 | jfs_err("xtLog: lwm > next\n"); | |
1778 | goto out; | |
1779 | } | |
1780 | tlck->flag |= tlckUPDATEMAP; | |
1781 | xadlock->flag = mlckALLOCXADLIST; | |
1782 | xadlock->count = next - lwm; | |
66f3131f | 1783 | if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) { |
1da177e4 | 1784 | int i; |
66f3131f | 1785 | pxd_t *pxd; |
1da177e4 LT |
1786 | /* |
1787 | * Lazy commit may allow xtree to be modified before | |
1788 | * txUpdateMap runs. Copy xad into linelock to | |
1789 | * preserve correct data. | |
66f3131f DK |
1790 | * |
1791 | * We can fit twice as may pxd's as xads in the lock | |
1da177e4 | 1792 | */ |
66f3131f DK |
1793 | xadlock->flag = mlckALLOCPXDLIST; |
1794 | pxd = xadlock->xdlist = &xtlck->pxdlock; | |
1795 | for (i = 0; i < xadlock->count; i++) { | |
1796 | PXDaddress(pxd, addressXAD(&p->xad[lwm + i])); | |
1797 | PXDlength(pxd, lengthXAD(&p->xad[lwm + i])); | |
1da177e4 LT |
1798 | p->xad[lwm + i].flag &= |
1799 | ~(XAD_NEW | XAD_EXTENDED); | |
66f3131f DK |
1800 | pxd++; |
1801 | } | |
1da177e4 LT |
1802 | } else { |
1803 | /* | |
1804 | * xdlist will point to into inode's xtree, ensure | |
1805 | * that transaction is not committed lazily. | |
1806 | */ | |
66f3131f | 1807 | xadlock->flag = mlckALLOCXADLIST; |
1da177e4 LT |
1808 | xadlock->xdlist = &p->xad[lwm]; |
1809 | tblk->xflag &= ~COMMIT_LAZY; | |
1810 | } | |
1811 | jfs_info("xtLog: alloc ip:0x%p mp:0x%p tlck:0x%p lwm:%d " | |
1812 | "count:%d", tlck->ip, mp, tlck, lwm, xadlock->count); | |
1813 | ||
1814 | maplock->index = 1; | |
1815 | ||
1816 | out: | |
1817 | /* mark page as homeward bound */ | |
1818 | tlck->flag |= tlckWRITEPAGE; | |
1819 | ||
1820 | return; | |
1821 | } | |
1822 | ||
1823 | /* | |
1824 | * page deletion: file deletion/truncation (ref. xtTruncate()) | |
1825 | * | |
1826 | * (page will be invalidated after log is written and bmap | |
1827 | * is updated from the page); | |
1828 | */ | |
1829 | if (tlck->type & tlckFREE) { | |
1830 | /* LOG_NOREDOPAGE log for NoRedoPage filter: | |
1831 | * if page free from file delete, NoRedoFile filter from | |
1832 | * inode image of zero link count will subsume NoRedoPage | |
1833 | * filters for each page; | |
1834 | * if page free from file truncattion, write NoRedoPage | |
1835 | * filter; | |
1836 | * | |
1837 | * upadte of block allocation map for the page itself: | |
1838 | * if page free from deletion and truncation, LOG_UPDATEMAP | |
1839 | * log for the page itself is generated from processing | |
1840 | * its parent page xad entries; | |
1841 | */ | |
1842 | /* if page free from file truncation, log LOG_NOREDOPAGE | |
1843 | * of the deleted page for logredo() to start NoRedoPage | |
1844 | * filter for the page; | |
1845 | */ | |
1846 | if (tblk->xflag & COMMIT_TRUNCATE) { | |
1847 | /* write NOREDOPAGE for the page */ | |
1848 | lrd->type = cpu_to_le16(LOG_NOREDOPAGE); | |
66f3131f DK |
1849 | PXDaddress(page_pxd, mp->index); |
1850 | PXDlength(page_pxd, | |
1da177e4 LT |
1851 | mp->logical_size >> tblk->sb-> |
1852 | s_blocksize_bits); | |
1853 | lrd->backchain = | |
1854 | cpu_to_le32(lmLog(log, tblk, lrd, NULL)); | |
1855 | ||
1856 | if (tlck->type & tlckBTROOT) { | |
1857 | /* Empty xtree must be logged */ | |
1858 | lrd->type = cpu_to_le16(LOG_REDOPAGE); | |
1859 | lrd->backchain = | |
1860 | cpu_to_le32(lmLog(log, tblk, lrd, tlck)); | |
1861 | } | |
1862 | } | |
1863 | ||
1864 | /* init LOG_UPDATEMAP of the freed extents | |
1865 | * XAD[XTENTRYSTART:hwm) from the deleted page itself | |
1866 | * for logredo() to update bmap; | |
1867 | */ | |
1868 | lrd->type = cpu_to_le16(LOG_UPDATEMAP); | |
1869 | lrd->log.updatemap.type = cpu_to_le16(LOG_FREEXADLIST); | |
1870 | xtlck = (struct xtlock *) & tlck->lock; | |
1871 | hwm = xtlck->hwm.offset; | |
1872 | lrd->log.updatemap.nxd = | |
1873 | cpu_to_le16(hwm - XTENTRYSTART + 1); | |
1874 | /* reformat linelock for lmLog() */ | |
1875 | xtlck->header.offset = XTENTRYSTART; | |
1876 | xtlck->header.length = hwm - XTENTRYSTART + 1; | |
1877 | xtlck->index = 1; | |
1878 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); | |
1879 | ||
1880 | /* format a maplock for txUpdateMap() to update bmap | |
1881 | * to free extents of XAD[XTENTRYSTART:hwm) from the | |
1882 | * deleted page itself; | |
1883 | */ | |
1884 | tlck->flag |= tlckUPDATEMAP; | |
1da177e4 | 1885 | xadlock->count = hwm - XTENTRYSTART + 1; |
66f3131f DK |
1886 | if ((xadlock->count <= 4) && (tblk->xflag & COMMIT_LAZY)) { |
1887 | int i; | |
1888 | pxd_t *pxd; | |
1da177e4 LT |
1889 | /* |
1890 | * Lazy commit may allow xtree to be modified before | |
1891 | * txUpdateMap runs. Copy xad into linelock to | |
1892 | * preserve correct data. | |
66f3131f DK |
1893 | * |
1894 | * We can fit twice as may pxd's as xads in the lock | |
1da177e4 | 1895 | */ |
66f3131f DK |
1896 | xadlock->flag = mlckFREEPXDLIST; |
1897 | pxd = xadlock->xdlist = &xtlck->pxdlock; | |
1898 | for (i = 0; i < xadlock->count; i++) { | |
1899 | PXDaddress(pxd, | |
1900 | addressXAD(&p->xad[XTENTRYSTART + i])); | |
1901 | PXDlength(pxd, | |
1902 | lengthXAD(&p->xad[XTENTRYSTART + i])); | |
1903 | pxd++; | |
1904 | } | |
1da177e4 LT |
1905 | } else { |
1906 | /* | |
1907 | * xdlist will point to into inode's xtree, ensure | |
1908 | * that transaction is not committed lazily. | |
1909 | */ | |
66f3131f | 1910 | xadlock->flag = mlckFREEXADLIST; |
1da177e4 LT |
1911 | xadlock->xdlist = &p->xad[XTENTRYSTART]; |
1912 | tblk->xflag &= ~COMMIT_LAZY; | |
1913 | } | |
1914 | jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d lwm:2", | |
1915 | tlck->ip, mp, xadlock->count); | |
1916 | ||
1917 | maplock->index = 1; | |
1918 | ||
1919 | /* mark page as invalid */ | |
1920 | if (((tblk->xflag & COMMIT_PWMAP) || S_ISDIR(ip->i_mode)) | |
1921 | && !(tlck->type & tlckBTROOT)) | |
1922 | tlck->flag |= tlckFREEPAGE; | |
1923 | /* | |
1924 | else (tblk->xflag & COMMIT_PMAP) | |
1925 | ? release the page; | |
1926 | */ | |
1927 | return; | |
1928 | } | |
1929 | ||
1930 | /* | |
1931 | * page/entry truncation: file truncation (ref. xtTruncate()) | |
1932 | * | |
1933 | * |----------+------+------+---------------| | |
1934 | * | | | | |
1935 | * | | hwm - hwm before truncation | |
1936 | * | next - truncation point | |
1937 | * lwm - lwm before truncation | |
1938 | * header ? | |
1939 | */ | |
1940 | if (tlck->type & tlckTRUNCATE) { | |
66f3131f | 1941 | pxd_t pxd; /* truncated extent of xad */ |
1da177e4 LT |
1942 | int twm; |
1943 | ||
1944 | /* | |
1945 | * For truncation the entire linelock may be used, so it would | |
1946 | * be difficult to store xad list in linelock itself. | |
1947 | * Therefore, we'll just force transaction to be committed | |
1948 | * synchronously, so that xtree pages won't be changed before | |
1949 | * txUpdateMap runs. | |
1950 | */ | |
1951 | tblk->xflag &= ~COMMIT_LAZY; | |
1952 | lwm = xtlck->lwm.offset; | |
1953 | if (lwm == 0) | |
1954 | lwm = XTPAGEMAXSLOT; | |
1955 | hwm = xtlck->hwm.offset; | |
1956 | twm = xtlck->twm.offset; | |
1957 | ||
1958 | /* | |
1959 | * write log records | |
1960 | */ | |
1961 | /* log after-image for logredo(): | |
1962 | * | |
1963 | * logredo() will update bmap for alloc of new/extended | |
1964 | * extents (XAD_NEW|XAD_EXTEND) of XAD[lwm:next) from | |
1965 | * after-image of XADlist; | |
1966 | * logredo() resets (XAD_NEW|XAD_EXTEND) flag when | |
1967 | * applying the after-image to the meta-data page. | |
1968 | */ | |
1969 | lrd->type = cpu_to_le16(LOG_REDOPAGE); | |
66f3131f DK |
1970 | PXDaddress(page_pxd, mp->index); |
1971 | PXDlength(page_pxd, | |
1972 | mp->logical_size >> tblk->sb->s_blocksize_bits); | |
1da177e4 LT |
1973 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, tlck)); |
1974 | ||
1975 | /* | |
1976 | * truncate entry XAD[twm == next - 1]: | |
1977 | */ | |
1978 | if (twm == next - 1) { | |
1979 | /* init LOG_UPDATEMAP for logredo() to update bmap for | |
1980 | * free of truncated delta extent of the truncated | |
1981 | * entry XAD[next - 1]: | |
1982 | * (xtlck->pxdlock = truncated delta extent); | |
1983 | */ | |
1984 | pxdlock = (struct pxd_lock *) & xtlck->pxdlock; | |
1985 | /* assert(pxdlock->type & tlckTRUNCATE); */ | |
1986 | lrd->type = cpu_to_le16(LOG_UPDATEMAP); | |
1987 | lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD); | |
1988 | lrd->log.updatemap.nxd = cpu_to_le16(1); | |
1989 | lrd->log.updatemap.pxd = pxdlock->pxd; | |
66f3131f | 1990 | pxd = pxdlock->pxd; /* save to format maplock */ |
1da177e4 LT |
1991 | lrd->backchain = |
1992 | cpu_to_le32(lmLog(log, tblk, lrd, NULL)); | |
1993 | } | |
1994 | ||
1995 | /* | |
1996 | * free entries XAD[next:hwm]: | |
1997 | */ | |
1998 | if (hwm >= next) { | |
1999 | /* init LOG_UPDATEMAP of the freed extents | |
2000 | * XAD[next:hwm] from the deleted page itself | |
2001 | * for logredo() to update bmap; | |
2002 | */ | |
2003 | lrd->type = cpu_to_le16(LOG_UPDATEMAP); | |
2004 | lrd->log.updatemap.type = | |
2005 | cpu_to_le16(LOG_FREEXADLIST); | |
2006 | xtlck = (struct xtlock *) & tlck->lock; | |
2007 | hwm = xtlck->hwm.offset; | |
2008 | lrd->log.updatemap.nxd = | |
2009 | cpu_to_le16(hwm - next + 1); | |
2010 | /* reformat linelock for lmLog() */ | |
2011 | xtlck->header.offset = next; | |
2012 | xtlck->header.length = hwm - next + 1; | |
2013 | xtlck->index = 1; | |
2014 | lrd->backchain = | |
2015 | cpu_to_le32(lmLog(log, tblk, lrd, tlck)); | |
2016 | } | |
2017 | ||
2018 | /* | |
2019 | * format maplock(s) for txUpdateMap() to update bmap | |
2020 | */ | |
2021 | maplock->index = 0; | |
2022 | ||
2023 | /* | |
2024 | * allocate entries XAD[lwm:next): | |
2025 | */ | |
2026 | if (lwm < next) { | |
2027 | /* format a maplock for txUpdateMap() to update bPMAP | |
2028 | * for alloc of new/extended extents of XAD[lwm:next) | |
2029 | * from the page itself; | |
2030 | * txUpdateMap() resets (XAD_NEW|XAD_EXTEND) flag. | |
2031 | */ | |
2032 | tlck->flag |= tlckUPDATEMAP; | |
2033 | xadlock->flag = mlckALLOCXADLIST; | |
2034 | xadlock->count = next - lwm; | |
2035 | xadlock->xdlist = &p->xad[lwm]; | |
2036 | ||
2037 | jfs_info("xtLog: alloc ip:0x%p mp:0x%p count:%d " | |
2038 | "lwm:%d next:%d", | |
2039 | tlck->ip, mp, xadlock->count, lwm, next); | |
2040 | maplock->index++; | |
2041 | xadlock++; | |
2042 | } | |
2043 | ||
2044 | /* | |
2045 | * truncate entry XAD[twm == next - 1]: | |
2046 | */ | |
2047 | if (twm == next - 1) { | |
2048 | struct pxd_lock *pxdlock; | |
2049 | ||
2050 | /* format a maplock for txUpdateMap() to update bmap | |
2051 | * to free truncated delta extent of the truncated | |
2052 | * entry XAD[next - 1]; | |
2053 | * (xtlck->pxdlock = truncated delta extent); | |
2054 | */ | |
2055 | tlck->flag |= tlckUPDATEMAP; | |
2056 | pxdlock = (struct pxd_lock *) xadlock; | |
2057 | pxdlock->flag = mlckFREEPXD; | |
2058 | pxdlock->count = 1; | |
66f3131f | 2059 | pxdlock->pxd = pxd; |
1da177e4 LT |
2060 | |
2061 | jfs_info("xtLog: truncate ip:0x%p mp:0x%p count:%d " | |
2062 | "hwm:%d", ip, mp, pxdlock->count, hwm); | |
2063 | maplock->index++; | |
2064 | xadlock++; | |
2065 | } | |
2066 | ||
2067 | /* | |
2068 | * free entries XAD[next:hwm]: | |
2069 | */ | |
2070 | if (hwm >= next) { | |
2071 | /* format a maplock for txUpdateMap() to update bmap | |
2072 | * to free extents of XAD[next:hwm] from thedeleted | |
2073 | * page itself; | |
2074 | */ | |
2075 | tlck->flag |= tlckUPDATEMAP; | |
2076 | xadlock->flag = mlckFREEXADLIST; | |
2077 | xadlock->count = hwm - next + 1; | |
2078 | xadlock->xdlist = &p->xad[next]; | |
2079 | ||
2080 | jfs_info("xtLog: free ip:0x%p mp:0x%p count:%d " | |
2081 | "next:%d hwm:%d", | |
2082 | tlck->ip, mp, xadlock->count, next, hwm); | |
2083 | maplock->index++; | |
2084 | } | |
2085 | ||
2086 | /* mark page as homeward bound */ | |
2087 | tlck->flag |= tlckWRITEPAGE; | |
2088 | } | |
2089 | return; | |
2090 | } | |
2091 | ||
2092 | ||
2093 | /* | |
2094 | * mapLog() | |
2095 | * | |
2096 | * function: log from maplock of freed data extents; | |
2097 | */ | |
2098 | void mapLog(struct jfs_log * log, struct tblock * tblk, struct lrd * lrd, | |
2099 | struct tlock * tlck) | |
2100 | { | |
2101 | struct pxd_lock *pxdlock; | |
2102 | int i, nlock; | |
2103 | pxd_t *pxd; | |
2104 | ||
2105 | /* | |
2106 | * page relocation: free the source page extent | |
2107 | * | |
2108 | * a maplock for txUpdateMap() for free of the page | |
2109 | * has been formatted at txLock() time saving the src | |
2110 | * relocated page address; | |
2111 | */ | |
2112 | if (tlck->type & tlckRELOCATE) { | |
2113 | /* log LOG_NOREDOPAGE of the old relocated page | |
2114 | * for logredo() to start NoRedoPage filter; | |
2115 | */ | |
2116 | lrd->type = cpu_to_le16(LOG_NOREDOPAGE); | |
2117 | pxdlock = (struct pxd_lock *) & tlck->lock; | |
2118 | pxd = &lrd->log.redopage.pxd; | |
2119 | *pxd = pxdlock->pxd; | |
2120 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL)); | |
2121 | ||
2122 | /* (N.B. currently, logredo() does NOT update bmap | |
2123 | * for free of the page itself for (LOG_XTREE|LOG_NOREDOPAGE); | |
2124 | * if page free from relocation, LOG_UPDATEMAP log is | |
2125 | * specifically generated now for logredo() | |
2126 | * to update bmap for free of src relocated page; | |
2127 | * (new flag LOG_RELOCATE may be introduced which will | |
2128 | * inform logredo() to start NORedoPage filter and also | |
2129 | * update block allocation map at the same time, thus | |
2130 | * avoiding an extra log write); | |
2131 | */ | |
2132 | lrd->type = cpu_to_le16(LOG_UPDATEMAP); | |
2133 | lrd->log.updatemap.type = cpu_to_le16(LOG_FREEPXD); | |
2134 | lrd->log.updatemap.nxd = cpu_to_le16(1); | |
2135 | lrd->log.updatemap.pxd = pxdlock->pxd; | |
2136 | lrd->backchain = cpu_to_le32(lmLog(log, tblk, lrd, NULL)); | |
2137 | ||
2138 | /* a maplock for txUpdateMap() for free of the page | |
2139 | * has been formatted at txLock() time; | |
2140 | */ | |
2141 | tlck->flag |= tlckUPDATEMAP; | |
2142 | return; | |
2143 | } | |
2144 | /* | |
2145 | ||
2146 | * Otherwise it's not a relocate request | |
2147 | * | |
2148 | */ | |
2149 | else { | |
2150 | /* log LOG_UPDATEMAP for logredo() to update bmap for | |
2151 | * free of truncated/relocated delta extent of the data; | |
2152 | * e.g.: external EA extent, relocated/truncated extent | |
2153 | * from xtTailgate(); | |
2154 | */ | |
2155 | lrd->type = cpu_to_le16(LOG_UPDATEMAP); | |
2156 | pxdlock = (struct pxd_lock *) & tlck->lock; | |
2157 | nlock = pxdlock->index; | |
2158 | for (i = 0; i < nlock; i++, pxdlock++) { | |
2159 | if (pxdlock->flag & mlckALLOCPXD) | |
2160 | lrd->log.updatemap.type = | |
2161 | cpu_to_le16(LOG_ALLOCPXD); | |
2162 | else | |
2163 | lrd->log.updatemap.type = | |
2164 | cpu_to_le16(LOG_FREEPXD); | |
2165 | lrd->log.updatemap.nxd = cpu_to_le16(1); | |
2166 | lrd->log.updatemap.pxd = pxdlock->pxd; | |
2167 | lrd->backchain = | |
2168 | cpu_to_le32(lmLog(log, tblk, lrd, NULL)); | |
2169 | jfs_info("mapLog: xaddr:0x%lx xlen:0x%x", | |
2170 | (ulong) addressPXD(&pxdlock->pxd), | |
2171 | lengthPXD(&pxdlock->pxd)); | |
2172 | } | |
2173 | ||
2174 | /* update bmap */ | |
2175 | tlck->flag |= tlckUPDATEMAP; | |
2176 | } | |
2177 | } | |
2178 | ||
2179 | ||
2180 | /* | |
2181 | * txEA() | |
2182 | * | |
2183 | * function: acquire maplock for EA/ACL extents or | |
2184 | * set COMMIT_INLINE flag; | |
2185 | */ | |
2186 | void txEA(tid_t tid, struct inode *ip, dxd_t * oldea, dxd_t * newea) | |
2187 | { | |
2188 | struct tlock *tlck = NULL; | |
2189 | struct pxd_lock *maplock = NULL, *pxdlock = NULL; | |
2190 | ||
2191 | /* | |
2192 | * format maplock for alloc of new EA extent | |
2193 | */ | |
2194 | if (newea) { | |
2195 | /* Since the newea could be a completely zeroed entry we need to | |
2196 | * check for the two flags which indicate we should actually | |
2197 | * commit new EA data | |
2198 | */ | |
2199 | if (newea->flag & DXD_EXTENT) { | |
2200 | tlck = txMaplock(tid, ip, tlckMAP); | |
2201 | maplock = (struct pxd_lock *) & tlck->lock; | |
2202 | pxdlock = (struct pxd_lock *) maplock; | |
2203 | pxdlock->flag = mlckALLOCPXD; | |
2204 | PXDaddress(&pxdlock->pxd, addressDXD(newea)); | |
2205 | PXDlength(&pxdlock->pxd, lengthDXD(newea)); | |
2206 | pxdlock++; | |
2207 | maplock->index = 1; | |
2208 | } else if (newea->flag & DXD_INLINE) { | |
2209 | tlck = NULL; | |
2210 | ||
2211 | set_cflag(COMMIT_Inlineea, ip); | |
2212 | } | |
2213 | } | |
2214 | ||
2215 | /* | |
2216 | * format maplock for free of old EA extent | |
2217 | */ | |
2218 | if (!test_cflag(COMMIT_Nolink, ip) && oldea->flag & DXD_EXTENT) { | |
2219 | if (tlck == NULL) { | |
2220 | tlck = txMaplock(tid, ip, tlckMAP); | |
2221 | maplock = (struct pxd_lock *) & tlck->lock; | |
2222 | pxdlock = (struct pxd_lock *) maplock; | |
2223 | maplock->index = 0; | |
2224 | } | |
2225 | pxdlock->flag = mlckFREEPXD; | |
2226 | PXDaddress(&pxdlock->pxd, addressDXD(oldea)); | |
2227 | PXDlength(&pxdlock->pxd, lengthDXD(oldea)); | |
2228 | maplock->index++; | |
2229 | } | |
2230 | } | |
2231 | ||
2232 | ||
2233 | /* | |
2234 | * txForce() | |
2235 | * | |
2236 | * function: synchronously write pages locked by transaction | |
2237 | * after txLog() but before txUpdateMap(); | |
2238 | */ | |
2239 | void txForce(struct tblock * tblk) | |
2240 | { | |
2241 | struct tlock *tlck; | |
2242 | lid_t lid, next; | |
2243 | struct metapage *mp; | |
2244 | ||
2245 | /* | |
2246 | * reverse the order of transaction tlocks in | |
2247 | * careful update order of address index pages | |
2248 | * (right to left, bottom up) | |
2249 | */ | |
2250 | tlck = lid_to_tlock(tblk->next); | |
2251 | lid = tlck->next; | |
2252 | tlck->next = 0; | |
2253 | while (lid) { | |
2254 | tlck = lid_to_tlock(lid); | |
2255 | next = tlck->next; | |
2256 | tlck->next = tblk->next; | |
2257 | tblk->next = lid; | |
2258 | lid = next; | |
2259 | } | |
2260 | ||
2261 | /* | |
2262 | * synchronously write the page, and | |
2263 | * hold the page for txUpdateMap(); | |
2264 | */ | |
2265 | for (lid = tblk->next; lid; lid = next) { | |
2266 | tlck = lid_to_tlock(lid); | |
2267 | next = tlck->next; | |
2268 | ||
2269 | if ((mp = tlck->mp) != NULL && | |
2270 | (tlck->type & tlckBTROOT) == 0) { | |
2271 | assert(mp->xflag & COMMIT_PAGE); | |
2272 | ||
2273 | if (tlck->flag & tlckWRITEPAGE) { | |
2274 | tlck->flag &= ~tlckWRITEPAGE; | |
2275 | ||
2276 | /* do not release page to freelist */ | |
7fab479b DK |
2277 | force_metapage(mp); |
2278 | #if 0 | |
1da177e4 LT |
2279 | /* |
2280 | * The "right" thing to do here is to | |
2281 | * synchronously write the metadata. | |
2282 | * With the current implementation this | |
2283 | * is hard since write_metapage requires | |
2284 | * us to kunmap & remap the page. If we | |
2285 | * have tlocks pointing into the metadata | |
2286 | * pages, we don't want to do this. I think | |
2287 | * we can get by with synchronously writing | |
2288 | * the pages when they are released. | |
2289 | */ | |
7fab479b | 2290 | assert(mp->nohomeok); |
1da177e4 LT |
2291 | set_bit(META_dirty, &mp->flag); |
2292 | set_bit(META_sync, &mp->flag); | |
7fab479b | 2293 | #endif |
1da177e4 LT |
2294 | } |
2295 | } | |
2296 | } | |
2297 | } | |
2298 | ||
2299 | ||
2300 | /* | |
2301 | * txUpdateMap() | |
2302 | * | |
2303 | * function: update persistent allocation map (and working map | |
2304 | * if appropriate); | |
2305 | * | |
2306 | * parameter: | |
2307 | */ | |
2308 | static void txUpdateMap(struct tblock * tblk) | |
2309 | { | |
2310 | struct inode *ip; | |
2311 | struct inode *ipimap; | |
2312 | lid_t lid; | |
2313 | struct tlock *tlck; | |
2314 | struct maplock *maplock; | |
2315 | struct pxd_lock pxdlock; | |
2316 | int maptype; | |
2317 | int k, nlock; | |
2318 | struct metapage *mp = NULL; | |
2319 | ||
2320 | ipimap = JFS_SBI(tblk->sb)->ipimap; | |
2321 | ||
2322 | maptype = (tblk->xflag & COMMIT_PMAP) ? COMMIT_PMAP : COMMIT_PWMAP; | |
2323 | ||
2324 | ||
2325 | /* | |
2326 | * update block allocation map | |
2327 | * | |
2328 | * update allocation state in pmap (and wmap) and | |
2329 | * update lsn of the pmap page; | |
2330 | */ | |
2331 | /* | |
2332 | * scan each tlock/page of transaction for block allocation/free: | |
2333 | * | |
2334 | * for each tlock/page of transaction, update map. | |
2335 | * ? are there tlock for pmap and pwmap at the same time ? | |
2336 | */ | |
2337 | for (lid = tblk->next; lid; lid = tlck->next) { | |
2338 | tlck = lid_to_tlock(lid); | |
2339 | ||
2340 | if ((tlck->flag & tlckUPDATEMAP) == 0) | |
2341 | continue; | |
2342 | ||
2343 | if (tlck->flag & tlckFREEPAGE) { | |
2344 | /* | |
2345 | * Another thread may attempt to reuse freed space | |
2346 | * immediately, so we want to get rid of the metapage | |
2347 | * before anyone else has a chance to get it. | |
2348 | * Lock metapage, update maps, then invalidate | |
2349 | * the metapage. | |
2350 | */ | |
2351 | mp = tlck->mp; | |
2352 | ASSERT(mp->xflag & COMMIT_PAGE); | |
7fab479b | 2353 | grab_metapage(mp); |
1da177e4 LT |
2354 | } |
2355 | ||
2356 | /* | |
2357 | * extent list: | |
2358 | * . in-line PXD list: | |
2359 | * . out-of-line XAD list: | |
2360 | */ | |
2361 | maplock = (struct maplock *) & tlck->lock; | |
2362 | nlock = maplock->index; | |
2363 | ||
2364 | for (k = 0; k < nlock; k++, maplock++) { | |
2365 | /* | |
2366 | * allocate blocks in persistent map: | |
2367 | * | |
2368 | * blocks have been allocated from wmap at alloc time; | |
2369 | */ | |
2370 | if (maplock->flag & mlckALLOC) { | |
2371 | txAllocPMap(ipimap, maplock, tblk); | |
2372 | } | |
2373 | /* | |
2374 | * free blocks in persistent and working map: | |
2375 | * blocks will be freed in pmap and then in wmap; | |
2376 | * | |
2377 | * ? tblock specifies the PMAP/PWMAP based upon | |
2378 | * transaction | |
2379 | * | |
2380 | * free blocks in persistent map: | |
2381 | * blocks will be freed from wmap at last reference | |
2382 | * release of the object for regular files; | |
2383 | * | |
2384 | * Alway free blocks from both persistent & working | |
2385 | * maps for directories | |
2386 | */ | |
2387 | else { /* (maplock->flag & mlckFREE) */ | |
2388 | ||
2389 | if (S_ISDIR(tlck->ip->i_mode)) | |
2390 | txFreeMap(ipimap, maplock, | |
2391 | tblk, COMMIT_PWMAP); | |
2392 | else | |
2393 | txFreeMap(ipimap, maplock, | |
2394 | tblk, maptype); | |
2395 | } | |
2396 | } | |
2397 | if (tlck->flag & tlckFREEPAGE) { | |
2398 | if (!(tblk->flag & tblkGC_LAZY)) { | |
2399 | /* This is equivalent to txRelease */ | |
2400 | ASSERT(mp->lid == lid); | |
2401 | tlck->mp->lid = 0; | |
2402 | } | |
7fab479b DK |
2403 | assert(mp->nohomeok == 1); |
2404 | metapage_homeok(mp); | |
1da177e4 LT |
2405 | discard_metapage(mp); |
2406 | tlck->mp = NULL; | |
2407 | } | |
2408 | } | |
2409 | /* | |
2410 | * update inode allocation map | |
2411 | * | |
2412 | * update allocation state in pmap and | |
2413 | * update lsn of the pmap page; | |
2414 | * update in-memory inode flag/state | |
2415 | * | |
2416 | * unlock mapper/write lock | |
2417 | */ | |
2418 | if (tblk->xflag & COMMIT_CREATE) { | |
2419 | diUpdatePMap(ipimap, tblk->ino, FALSE, tblk); | |
2420 | ipimap->i_state |= I_DIRTY; | |
2421 | /* update persistent block allocation map | |
2422 | * for the allocation of inode extent; | |
2423 | */ | |
2424 | pxdlock.flag = mlckALLOCPXD; | |
2425 | pxdlock.pxd = tblk->u.ixpxd; | |
2426 | pxdlock.index = 1; | |
2427 | txAllocPMap(ipimap, (struct maplock *) & pxdlock, tblk); | |
2428 | } else if (tblk->xflag & COMMIT_DELETE) { | |
2429 | ip = tblk->u.ip; | |
2430 | diUpdatePMap(ipimap, ip->i_ino, TRUE, tblk); | |
2431 | ipimap->i_state |= I_DIRTY; | |
2432 | iput(ip); | |
2433 | } | |
2434 | } | |
2435 | ||
2436 | ||
2437 | /* | |
2438 | * txAllocPMap() | |
2439 | * | |
2440 | * function: allocate from persistent map; | |
2441 | * | |
2442 | * parameter: | |
2443 | * ipbmap - | |
2444 | * malock - | |
2445 | * xad list: | |
2446 | * pxd: | |
2447 | * | |
2448 | * maptype - | |
2449 | * allocate from persistent map; | |
2450 | * free from persistent map; | |
2451 | * (e.g., tmp file - free from working map at releae | |
2452 | * of last reference); | |
2453 | * free from persistent and working map; | |
2454 | * | |
2455 | * lsn - log sequence number; | |
2456 | */ | |
2457 | static void txAllocPMap(struct inode *ip, struct maplock * maplock, | |
2458 | struct tblock * tblk) | |
2459 | { | |
2460 | struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; | |
2461 | struct xdlistlock *xadlistlock; | |
2462 | xad_t *xad; | |
2463 | s64 xaddr; | |
2464 | int xlen; | |
2465 | struct pxd_lock *pxdlock; | |
2466 | struct xdlistlock *pxdlistlock; | |
2467 | pxd_t *pxd; | |
2468 | int n; | |
2469 | ||
2470 | /* | |
2471 | * allocate from persistent map; | |
2472 | */ | |
2473 | if (maplock->flag & mlckALLOCXADLIST) { | |
2474 | xadlistlock = (struct xdlistlock *) maplock; | |
2475 | xad = xadlistlock->xdlist; | |
2476 | for (n = 0; n < xadlistlock->count; n++, xad++) { | |
2477 | if (xad->flag & (XAD_NEW | XAD_EXTENDED)) { | |
2478 | xaddr = addressXAD(xad); | |
2479 | xlen = lengthXAD(xad); | |
2480 | dbUpdatePMap(ipbmap, FALSE, xaddr, | |
2481 | (s64) xlen, tblk); | |
2482 | xad->flag &= ~(XAD_NEW | XAD_EXTENDED); | |
2483 | jfs_info("allocPMap: xaddr:0x%lx xlen:%d", | |
2484 | (ulong) xaddr, xlen); | |
2485 | } | |
2486 | } | |
2487 | } else if (maplock->flag & mlckALLOCPXD) { | |
2488 | pxdlock = (struct pxd_lock *) maplock; | |
2489 | xaddr = addressPXD(&pxdlock->pxd); | |
2490 | xlen = lengthPXD(&pxdlock->pxd); | |
2491 | dbUpdatePMap(ipbmap, FALSE, xaddr, (s64) xlen, tblk); | |
2492 | jfs_info("allocPMap: xaddr:0x%lx xlen:%d", (ulong) xaddr, xlen); | |
2493 | } else { /* (maplock->flag & mlckALLOCPXDLIST) */ | |
2494 | ||
2495 | pxdlistlock = (struct xdlistlock *) maplock; | |
2496 | pxd = pxdlistlock->xdlist; | |
2497 | for (n = 0; n < pxdlistlock->count; n++, pxd++) { | |
2498 | xaddr = addressPXD(pxd); | |
2499 | xlen = lengthPXD(pxd); | |
2500 | dbUpdatePMap(ipbmap, FALSE, xaddr, (s64) xlen, | |
2501 | tblk); | |
2502 | jfs_info("allocPMap: xaddr:0x%lx xlen:%d", | |
2503 | (ulong) xaddr, xlen); | |
2504 | } | |
2505 | } | |
2506 | } | |
2507 | ||
2508 | ||
2509 | /* | |
2510 | * txFreeMap() | |
2511 | * | |
2512 | * function: free from persistent and/or working map; | |
2513 | * | |
2514 | * todo: optimization | |
2515 | */ | |
2516 | void txFreeMap(struct inode *ip, | |
2517 | struct maplock * maplock, struct tblock * tblk, int maptype) | |
2518 | { | |
2519 | struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap; | |
2520 | struct xdlistlock *xadlistlock; | |
2521 | xad_t *xad; | |
2522 | s64 xaddr; | |
2523 | int xlen; | |
2524 | struct pxd_lock *pxdlock; | |
2525 | struct xdlistlock *pxdlistlock; | |
2526 | pxd_t *pxd; | |
2527 | int n; | |
2528 | ||
2529 | jfs_info("txFreeMap: tblk:0x%p maplock:0x%p maptype:0x%x", | |
2530 | tblk, maplock, maptype); | |
2531 | ||
2532 | /* | |
2533 | * free from persistent map; | |
2534 | */ | |
2535 | if (maptype == COMMIT_PMAP || maptype == COMMIT_PWMAP) { | |
2536 | if (maplock->flag & mlckFREEXADLIST) { | |
2537 | xadlistlock = (struct xdlistlock *) maplock; | |
2538 | xad = xadlistlock->xdlist; | |
2539 | for (n = 0; n < xadlistlock->count; n++, xad++) { | |
2540 | if (!(xad->flag & XAD_NEW)) { | |
2541 | xaddr = addressXAD(xad); | |
2542 | xlen = lengthXAD(xad); | |
2543 | dbUpdatePMap(ipbmap, TRUE, xaddr, | |
2544 | (s64) xlen, tblk); | |
2545 | jfs_info("freePMap: xaddr:0x%lx " | |
2546 | "xlen:%d", | |
2547 | (ulong) xaddr, xlen); | |
2548 | } | |
2549 | } | |
2550 | } else if (maplock->flag & mlckFREEPXD) { | |
2551 | pxdlock = (struct pxd_lock *) maplock; | |
2552 | xaddr = addressPXD(&pxdlock->pxd); | |
2553 | xlen = lengthPXD(&pxdlock->pxd); | |
2554 | dbUpdatePMap(ipbmap, TRUE, xaddr, (s64) xlen, | |
2555 | tblk); | |
2556 | jfs_info("freePMap: xaddr:0x%lx xlen:%d", | |
2557 | (ulong) xaddr, xlen); | |
2558 | } else { /* (maplock->flag & mlckALLOCPXDLIST) */ | |
2559 | ||
2560 | pxdlistlock = (struct xdlistlock *) maplock; | |
2561 | pxd = pxdlistlock->xdlist; | |
2562 | for (n = 0; n < pxdlistlock->count; n++, pxd++) { | |
2563 | xaddr = addressPXD(pxd); | |
2564 | xlen = lengthPXD(pxd); | |
2565 | dbUpdatePMap(ipbmap, TRUE, xaddr, | |
2566 | (s64) xlen, tblk); | |
2567 | jfs_info("freePMap: xaddr:0x%lx xlen:%d", | |
2568 | (ulong) xaddr, xlen); | |
2569 | } | |
2570 | } | |
2571 | } | |
2572 | ||
2573 | /* | |
2574 | * free from working map; | |
2575 | */ | |
2576 | if (maptype == COMMIT_PWMAP || maptype == COMMIT_WMAP) { | |
2577 | if (maplock->flag & mlckFREEXADLIST) { | |
2578 | xadlistlock = (struct xdlistlock *) maplock; | |
2579 | xad = xadlistlock->xdlist; | |
2580 | for (n = 0; n < xadlistlock->count; n++, xad++) { | |
2581 | xaddr = addressXAD(xad); | |
2582 | xlen = lengthXAD(xad); | |
2583 | dbFree(ip, xaddr, (s64) xlen); | |
2584 | xad->flag = 0; | |
2585 | jfs_info("freeWMap: xaddr:0x%lx xlen:%d", | |
2586 | (ulong) xaddr, xlen); | |
2587 | } | |
2588 | } else if (maplock->flag & mlckFREEPXD) { | |
2589 | pxdlock = (struct pxd_lock *) maplock; | |
2590 | xaddr = addressPXD(&pxdlock->pxd); | |
2591 | xlen = lengthPXD(&pxdlock->pxd); | |
2592 | dbFree(ip, xaddr, (s64) xlen); | |
2593 | jfs_info("freeWMap: xaddr:0x%lx xlen:%d", | |
2594 | (ulong) xaddr, xlen); | |
2595 | } else { /* (maplock->flag & mlckFREEPXDLIST) */ | |
2596 | ||
2597 | pxdlistlock = (struct xdlistlock *) maplock; | |
2598 | pxd = pxdlistlock->xdlist; | |
2599 | for (n = 0; n < pxdlistlock->count; n++, pxd++) { | |
2600 | xaddr = addressPXD(pxd); | |
2601 | xlen = lengthPXD(pxd); | |
2602 | dbFree(ip, xaddr, (s64) xlen); | |
2603 | jfs_info("freeWMap: xaddr:0x%lx xlen:%d", | |
2604 | (ulong) xaddr, xlen); | |
2605 | } | |
2606 | } | |
2607 | } | |
2608 | } | |
2609 | ||
2610 | ||
2611 | /* | |
2612 | * txFreelock() | |
2613 | * | |
2614 | * function: remove tlock from inode anonymous locklist | |
2615 | */ | |
2616 | void txFreelock(struct inode *ip) | |
2617 | { | |
2618 | struct jfs_inode_info *jfs_ip = JFS_IP(ip); | |
2619 | struct tlock *xtlck, *tlck; | |
2620 | lid_t xlid = 0, lid; | |
2621 | ||
2622 | if (!jfs_ip->atlhead) | |
2623 | return; | |
2624 | ||
2625 | TXN_LOCK(); | |
2626 | xtlck = (struct tlock *) &jfs_ip->atlhead; | |
2627 | ||
2628 | while ((lid = xtlck->next) != 0) { | |
2629 | tlck = lid_to_tlock(lid); | |
2630 | if (tlck->flag & tlckFREELOCK) { | |
2631 | xtlck->next = tlck->next; | |
2632 | txLockFree(lid); | |
2633 | } else { | |
2634 | xtlck = tlck; | |
2635 | xlid = lid; | |
2636 | } | |
2637 | } | |
2638 | ||
2639 | if (jfs_ip->atlhead) | |
2640 | jfs_ip->atltail = xlid; | |
2641 | else { | |
2642 | jfs_ip->atltail = 0; | |
2643 | /* | |
2644 | * If inode was on anon_list, remove it | |
2645 | */ | |
2646 | list_del_init(&jfs_ip->anon_inode_list); | |
2647 | } | |
2648 | TXN_UNLOCK(); | |
2649 | } | |
2650 | ||
2651 | ||
2652 | /* | |
2653 | * txAbort() | |
2654 | * | |
2655 | * function: abort tx before commit; | |
2656 | * | |
2657 | * frees line-locks and segment locks for all | |
2658 | * segments in comdata structure. | |
2659 | * Optionally sets state of file-system to FM_DIRTY in super-block. | |
2660 | * log age of page-frames in memory for which caller has | |
2661 | * are reset to 0 (to avoid logwarap). | |
2662 | */ | |
2663 | void txAbort(tid_t tid, int dirty) | |
2664 | { | |
2665 | lid_t lid, next; | |
2666 | struct metapage *mp; | |
2667 | struct tblock *tblk = tid_to_tblock(tid); | |
2668 | struct tlock *tlck; | |
2669 | ||
2670 | /* | |
2671 | * free tlocks of the transaction | |
2672 | */ | |
2673 | for (lid = tblk->next; lid; lid = next) { | |
2674 | tlck = lid_to_tlock(lid); | |
2675 | next = tlck->next; | |
2676 | mp = tlck->mp; | |
2677 | JFS_IP(tlck->ip)->xtlid = 0; | |
2678 | ||
2679 | if (mp) { | |
2680 | mp->lid = 0; | |
2681 | ||
2682 | /* | |
2683 | * reset lsn of page to avoid logwarap: | |
2684 | * | |
2685 | * (page may have been previously committed by another | |
2686 | * transaction(s) but has not been paged, i.e., | |
2687 | * it may be on logsync list even though it has not | |
2688 | * been logged for the current tx.) | |
2689 | */ | |
2690 | if (mp->xflag & COMMIT_PAGE && mp->lsn) | |
2691 | LogSyncRelease(mp); | |
2692 | } | |
2693 | /* insert tlock at head of freelist */ | |
2694 | TXN_LOCK(); | |
2695 | txLockFree(lid); | |
2696 | TXN_UNLOCK(); | |
2697 | } | |
2698 | ||
2699 | /* caller will free the transaction block */ | |
2700 | ||
2701 | tblk->next = tblk->last = 0; | |
2702 | ||
2703 | /* | |
2704 | * mark filesystem dirty | |
2705 | */ | |
2706 | if (dirty) | |
2707 | jfs_error(tblk->sb, "txAbort"); | |
2708 | ||
2709 | return; | |
2710 | } | |
2711 | ||
2712 | /* | |
2713 | * txLazyCommit(void) | |
2714 | * | |
2715 | * All transactions except those changing ipimap (COMMIT_FORCE) are | |
2716 | * processed by this routine. This insures that the inode and block | |
2717 | * allocation maps are updated in order. For synchronous transactions, | |
2718 | * let the user thread finish processing after txUpdateMap() is called. | |
2719 | */ | |
2720 | static void txLazyCommit(struct tblock * tblk) | |
2721 | { | |
2722 | struct jfs_log *log; | |
2723 | ||
2724 | while (((tblk->flag & tblkGC_READY) == 0) && | |
2725 | ((tblk->flag & tblkGC_UNLOCKED) == 0)) { | |
2726 | /* We must have gotten ahead of the user thread | |
2727 | */ | |
2728 | jfs_info("jfs_lazycommit: tblk 0x%p not unlocked", tblk); | |
2729 | yield(); | |
2730 | } | |
2731 | ||
2732 | jfs_info("txLazyCommit: processing tblk 0x%p", tblk); | |
2733 | ||
2734 | txUpdateMap(tblk); | |
2735 | ||
2736 | log = (struct jfs_log *) JFS_SBI(tblk->sb)->log; | |
2737 | ||
2738 | spin_lock_irq(&log->gclock); // LOGGC_LOCK | |
2739 | ||
2740 | tblk->flag |= tblkGC_COMMITTED; | |
2741 | ||
2742 | if (tblk->flag & tblkGC_READY) | |
2743 | log->gcrtc--; | |
2744 | ||
2745 | wake_up_all(&tblk->gcwait); // LOGGC_WAKEUP | |
2746 | ||
2747 | /* | |
2748 | * Can't release log->gclock until we've tested tblk->flag | |
2749 | */ | |
2750 | if (tblk->flag & tblkGC_LAZY) { | |
2751 | spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK | |
2752 | txUnlock(tblk); | |
2753 | tblk->flag &= ~tblkGC_LAZY; | |
2754 | txEnd(tblk - TxBlock); /* Convert back to tid */ | |
2755 | } else | |
2756 | spin_unlock_irq(&log->gclock); // LOGGC_UNLOCK | |
2757 | ||
2758 | jfs_info("txLazyCommit: done: tblk = 0x%p", tblk); | |
2759 | } | |
2760 | ||
2761 | /* | |
2762 | * jfs_lazycommit(void) | |
2763 | * | |
2764 | * To be run as a kernel daemon. If lbmIODone is called in an interrupt | |
2765 | * context, or where blocking is not wanted, this routine will process | |
2766 | * committed transactions from the unlock queue. | |
2767 | */ | |
2768 | int jfs_lazycommit(void *arg) | |
2769 | { | |
2770 | int WorkDone; | |
2771 | struct tblock *tblk; | |
2772 | unsigned long flags; | |
2773 | struct jfs_sb_info *sbi; | |
2774 | ||
2775 | daemonize("jfsCommit"); | |
2776 | ||
2777 | complete(&jfsIOwait); | |
2778 | ||
2779 | do { | |
2780 | LAZY_LOCK(flags); | |
2781 | jfs_commit_thread_waking = 0; /* OK to wake another thread */ | |
2782 | while (!list_empty(&TxAnchor.unlock_queue)) { | |
2783 | WorkDone = 0; | |
2784 | list_for_each_entry(tblk, &TxAnchor.unlock_queue, | |
2785 | cqueue) { | |
2786 | ||
2787 | sbi = JFS_SBI(tblk->sb); | |
2788 | /* | |
2789 | * For each volume, the transactions must be | |
2790 | * handled in order. If another commit thread | |
2791 | * is handling a tblk for this superblock, | |
2792 | * skip it | |
2793 | */ | |
2794 | if (sbi->commit_state & IN_LAZYCOMMIT) | |
2795 | continue; | |
2796 | ||
2797 | sbi->commit_state |= IN_LAZYCOMMIT; | |
2798 | WorkDone = 1; | |
2799 | ||
2800 | /* | |
2801 | * Remove transaction from queue | |
2802 | */ | |
2803 | list_del(&tblk->cqueue); | |
2804 | ||
2805 | LAZY_UNLOCK(flags); | |
2806 | txLazyCommit(tblk); | |
2807 | LAZY_LOCK(flags); | |
2808 | ||
2809 | sbi->commit_state &= ~IN_LAZYCOMMIT; | |
2810 | /* | |
2811 | * Don't continue in the for loop. (We can't | |
2812 | * anyway, it's unsafe!) We want to go back to | |
2813 | * the beginning of the list. | |
2814 | */ | |
2815 | break; | |
2816 | } | |
2817 | ||
2818 | /* If there was nothing to do, don't continue */ | |
2819 | if (!WorkDone) | |
2820 | break; | |
2821 | } | |
2822 | /* In case a wakeup came while all threads were active */ | |
2823 | jfs_commit_thread_waking = 0; | |
2824 | ||
2825 | if (current->flags & PF_FREEZE) { | |
2826 | LAZY_UNLOCK(flags); | |
2827 | refrigerator(PF_FREEZE); | |
2828 | } else { | |
2829 | DECLARE_WAITQUEUE(wq, current); | |
2830 | ||
2831 | add_wait_queue(&jfs_commit_thread_wait, &wq); | |
2832 | set_current_state(TASK_INTERRUPTIBLE); | |
2833 | LAZY_UNLOCK(flags); | |
2834 | schedule(); | |
2835 | current->state = TASK_RUNNING; | |
2836 | remove_wait_queue(&jfs_commit_thread_wait, &wq); | |
2837 | } | |
2838 | } while (!jfs_stop_threads); | |
2839 | ||
2840 | if (!list_empty(&TxAnchor.unlock_queue)) | |
2841 | jfs_err("jfs_lazycommit being killed w/pending transactions!"); | |
2842 | else | |
2843 | jfs_info("jfs_lazycommit being killed\n"); | |
2844 | complete_and_exit(&jfsIOwait, 0); | |
2845 | } | |
2846 | ||
2847 | void txLazyUnlock(struct tblock * tblk) | |
2848 | { | |
2849 | unsigned long flags; | |
2850 | ||
2851 | LAZY_LOCK(flags); | |
2852 | ||
2853 | list_add_tail(&tblk->cqueue, &TxAnchor.unlock_queue); | |
2854 | /* | |
2855 | * Don't wake up a commit thread if there is already one servicing | |
2856 | * this superblock, or if the last one we woke up hasn't started yet. | |
2857 | */ | |
2858 | if (!(JFS_SBI(tblk->sb)->commit_state & IN_LAZYCOMMIT) && | |
2859 | !jfs_commit_thread_waking) { | |
2860 | jfs_commit_thread_waking = 1; | |
2861 | wake_up(&jfs_commit_thread_wait); | |
2862 | } | |
2863 | LAZY_UNLOCK(flags); | |
2864 | } | |
2865 | ||
2866 | static void LogSyncRelease(struct metapage * mp) | |
2867 | { | |
2868 | struct jfs_log *log = mp->log; | |
2869 | ||
7fab479b | 2870 | assert(mp->nohomeok); |
1da177e4 | 2871 | assert(log); |
7fab479b | 2872 | metapage_homeok(mp); |
1da177e4 LT |
2873 | } |
2874 | ||
2875 | /* | |
2876 | * txQuiesce | |
2877 | * | |
2878 | * Block all new transactions and push anonymous transactions to | |
2879 | * completion | |
2880 | * | |
2881 | * This does almost the same thing as jfs_sync below. We don't | |
2882 | * worry about deadlocking when jfs_tlocks_low is set, since we would | |
2883 | * expect jfs_sync to get us out of that jam. | |
2884 | */ | |
2885 | void txQuiesce(struct super_block *sb) | |
2886 | { | |
2887 | struct inode *ip; | |
2888 | struct jfs_inode_info *jfs_ip; | |
2889 | struct jfs_log *log = JFS_SBI(sb)->log; | |
2890 | tid_t tid; | |
2891 | ||
2892 | set_bit(log_QUIESCE, &log->flag); | |
2893 | ||
2894 | TXN_LOCK(); | |
2895 | restart: | |
2896 | while (!list_empty(&TxAnchor.anon_list)) { | |
2897 | jfs_ip = list_entry(TxAnchor.anon_list.next, | |
2898 | struct jfs_inode_info, | |
2899 | anon_inode_list); | |
2900 | ip = &jfs_ip->vfs_inode; | |
2901 | ||
2902 | /* | |
2903 | * inode will be removed from anonymous list | |
2904 | * when it is committed | |
2905 | */ | |
2906 | TXN_UNLOCK(); | |
2907 | tid = txBegin(ip->i_sb, COMMIT_INODE | COMMIT_FORCE); | |
2908 | down(&jfs_ip->commit_sem); | |
2909 | txCommit(tid, 1, &ip, 0); | |
2910 | txEnd(tid); | |
2911 | up(&jfs_ip->commit_sem); | |
2912 | /* | |
2913 | * Just to be safe. I don't know how | |
2914 | * long we can run without blocking | |
2915 | */ | |
2916 | cond_resched(); | |
2917 | TXN_LOCK(); | |
2918 | } | |
2919 | ||
2920 | /* | |
2921 | * If jfs_sync is running in parallel, there could be some inodes | |
2922 | * on anon_list2. Let's check. | |
2923 | */ | |
2924 | if (!list_empty(&TxAnchor.anon_list2)) { | |
2925 | list_splice(&TxAnchor.anon_list2, &TxAnchor.anon_list); | |
2926 | INIT_LIST_HEAD(&TxAnchor.anon_list2); | |
2927 | goto restart; | |
2928 | } | |
2929 | TXN_UNLOCK(); | |
2930 | ||
2931 | /* | |
2932 | * We may need to kick off the group commit | |
2933 | */ | |
2934 | jfs_flush_journal(log, 0); | |
2935 | } | |
2936 | ||
2937 | /* | |
2938 | * txResume() | |
2939 | * | |
2940 | * Allows transactions to start again following txQuiesce | |
2941 | */ | |
2942 | void txResume(struct super_block *sb) | |
2943 | { | |
2944 | struct jfs_log *log = JFS_SBI(sb)->log; | |
2945 | ||
2946 | clear_bit(log_QUIESCE, &log->flag); | |
2947 | TXN_WAKEUP(&log->syncwait); | |
2948 | } | |
2949 | ||
2950 | /* | |
2951 | * jfs_sync(void) | |
2952 | * | |
2953 | * To be run as a kernel daemon. This is awakened when tlocks run low. | |
2954 | * We write any inodes that have anonymous tlocks so they will become | |
2955 | * available. | |
2956 | */ | |
2957 | int jfs_sync(void *arg) | |
2958 | { | |
2959 | struct inode *ip; | |
2960 | struct jfs_inode_info *jfs_ip; | |
2961 | int rc; | |
2962 | tid_t tid; | |
2963 | ||
2964 | daemonize("jfsSync"); | |
2965 | ||
2966 | complete(&jfsIOwait); | |
2967 | ||
2968 | do { | |
2969 | /* | |
2970 | * write each inode on the anonymous inode list | |
2971 | */ | |
2972 | TXN_LOCK(); | |
2973 | while (jfs_tlocks_low && !list_empty(&TxAnchor.anon_list)) { | |
2974 | jfs_ip = list_entry(TxAnchor.anon_list.next, | |
2975 | struct jfs_inode_info, | |
2976 | anon_inode_list); | |
2977 | ip = &jfs_ip->vfs_inode; | |
2978 | ||
2979 | if (! igrab(ip)) { | |
2980 | /* | |
2981 | * Inode is being freed | |
2982 | */ | |
2983 | list_del_init(&jfs_ip->anon_inode_list); | |
2984 | } else if (! down_trylock(&jfs_ip->commit_sem)) { | |
2985 | /* | |
2986 | * inode will be removed from anonymous list | |
2987 | * when it is committed | |
2988 | */ | |
2989 | TXN_UNLOCK(); | |
2990 | tid = txBegin(ip->i_sb, COMMIT_INODE); | |
2991 | rc = txCommit(tid, 1, &ip, 0); | |
2992 | txEnd(tid); | |
2993 | up(&jfs_ip->commit_sem); | |
2994 | ||
2995 | iput(ip); | |
2996 | /* | |
2997 | * Just to be safe. I don't know how | |
2998 | * long we can run without blocking | |
2999 | */ | |
3000 | cond_resched(); | |
3001 | TXN_LOCK(); | |
3002 | } else { | |
3003 | /* We can't get the commit semaphore. It may | |
3004 | * be held by a thread waiting for tlock's | |
3005 | * so let's not block here. Save it to | |
3006 | * put back on the anon_list. | |
3007 | */ | |
3008 | ||
3009 | /* Take off anon_list */ | |
3010 | list_del(&jfs_ip->anon_inode_list); | |
3011 | ||
3012 | /* Put on anon_list2 */ | |
3013 | list_add(&jfs_ip->anon_inode_list, | |
3014 | &TxAnchor.anon_list2); | |
3015 | ||
3016 | TXN_UNLOCK(); | |
3017 | iput(ip); | |
3018 | TXN_LOCK(); | |
3019 | } | |
3020 | } | |
3021 | /* Add anon_list2 back to anon_list */ | |
3022 | list_splice_init(&TxAnchor.anon_list2, &TxAnchor.anon_list); | |
3023 | ||
3024 | if (current->flags & PF_FREEZE) { | |
3025 | TXN_UNLOCK(); | |
3026 | refrigerator(PF_FREEZE); | |
3027 | } else { | |
3028 | DECLARE_WAITQUEUE(wq, current); | |
3029 | ||
3030 | add_wait_queue(&jfs_sync_thread_wait, &wq); | |
3031 | set_current_state(TASK_INTERRUPTIBLE); | |
3032 | TXN_UNLOCK(); | |
3033 | schedule(); | |
3034 | current->state = TASK_RUNNING; | |
3035 | remove_wait_queue(&jfs_sync_thread_wait, &wq); | |
3036 | } | |
3037 | } while (!jfs_stop_threads); | |
3038 | ||
3039 | jfs_info("jfs_sync being killed"); | |
3040 | complete_and_exit(&jfsIOwait, 0); | |
3041 | } | |
3042 | ||
3043 | #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_DEBUG) | |
3044 | int jfs_txanchor_read(char *buffer, char **start, off_t offset, int length, | |
3045 | int *eof, void *data) | |
3046 | { | |
3047 | int len = 0; | |
3048 | off_t begin; | |
3049 | char *freewait; | |
3050 | char *freelockwait; | |
3051 | char *lowlockwait; | |
3052 | ||
3053 | freewait = | |
3054 | waitqueue_active(&TxAnchor.freewait) ? "active" : "empty"; | |
3055 | freelockwait = | |
3056 | waitqueue_active(&TxAnchor.freelockwait) ? "active" : "empty"; | |
3057 | lowlockwait = | |
3058 | waitqueue_active(&TxAnchor.lowlockwait) ? "active" : "empty"; | |
3059 | ||
3060 | len += sprintf(buffer, | |
3061 | "JFS TxAnchor\n" | |
3062 | "============\n" | |
3063 | "freetid = %d\n" | |
3064 | "freewait = %s\n" | |
3065 | "freelock = %d\n" | |
3066 | "freelockwait = %s\n" | |
3067 | "lowlockwait = %s\n" | |
3068 | "tlocksInUse = %d\n" | |
3069 | "jfs_tlocks_low = %d\n" | |
3070 | "unlock_queue is %sempty\n", | |
3071 | TxAnchor.freetid, | |
3072 | freewait, | |
3073 | TxAnchor.freelock, | |
3074 | freelockwait, | |
3075 | lowlockwait, | |
3076 | TxAnchor.tlocksInUse, | |
3077 | jfs_tlocks_low, | |
3078 | list_empty(&TxAnchor.unlock_queue) ? "" : "not "); | |
3079 | ||
3080 | begin = offset; | |
3081 | *start = buffer + begin; | |
3082 | len -= begin; | |
3083 | ||
3084 | if (len > length) | |
3085 | len = length; | |
3086 | else | |
3087 | *eof = 1; | |
3088 | ||
3089 | if (len < 0) | |
3090 | len = 0; | |
3091 | ||
3092 | return len; | |
3093 | } | |
3094 | #endif | |
3095 | ||
3096 | #if defined(CONFIG_PROC_FS) && defined(CONFIG_JFS_STATISTICS) | |
3097 | int jfs_txstats_read(char *buffer, char **start, off_t offset, int length, | |
3098 | int *eof, void *data) | |
3099 | { | |
3100 | int len = 0; | |
3101 | off_t begin; | |
3102 | ||
3103 | len += sprintf(buffer, | |
3104 | "JFS TxStats\n" | |
3105 | "===========\n" | |
3106 | "calls to txBegin = %d\n" | |
3107 | "txBegin blocked by sync barrier = %d\n" | |
3108 | "txBegin blocked by tlocks low = %d\n" | |
3109 | "txBegin blocked by no free tid = %d\n" | |
3110 | "calls to txBeginAnon = %d\n" | |
3111 | "txBeginAnon blocked by sync barrier = %d\n" | |
3112 | "txBeginAnon blocked by tlocks low = %d\n" | |
3113 | "calls to txLockAlloc = %d\n" | |
3114 | "tLockAlloc blocked by no free lock = %d\n", | |
3115 | TxStat.txBegin, | |
3116 | TxStat.txBegin_barrier, | |
3117 | TxStat.txBegin_lockslow, | |
3118 | TxStat.txBegin_freetid, | |
3119 | TxStat.txBeginAnon, | |
3120 | TxStat.txBeginAnon_barrier, | |
3121 | TxStat.txBeginAnon_lockslow, | |
3122 | TxStat.txLockAlloc, | |
3123 | TxStat.txLockAlloc_freelock); | |
3124 | ||
3125 | begin = offset; | |
3126 | *start = buffer + begin; | |
3127 | len -= begin; | |
3128 | ||
3129 | if (len > length) | |
3130 | len = length; | |
3131 | else | |
3132 | *eof = 1; | |
3133 | ||
3134 | if (len < 0) | |
3135 | len = 0; | |
3136 | ||
3137 | return len; | |
3138 | } | |
3139 | #endif |