1/*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_types.h"
21#include "xfs_log.h"
22#include "xfs_inum.h"
23#include "xfs_trans.h"
24#include "xfs_sb.h"
25#include "xfs_dir.h"
26#include "xfs_dmapi.h"
27#include "xfs_mount.h"
28#include "xfs_trans_priv.h"
29#include "xfs_error.h"
30
31STATIC void xfs_ail_insert(xfs_ail_entry_t *, xfs_log_item_t *);
32STATIC xfs_log_item_t * xfs_ail_delete(xfs_ail_entry_t *, xfs_log_item_t *);
33STATIC xfs_log_item_t * xfs_ail_min(xfs_ail_entry_t *);
34STATIC xfs_log_item_t * xfs_ail_next(xfs_ail_entry_t *, xfs_log_item_t *);
35
36#ifdef DEBUG
37STATIC void xfs_ail_check(xfs_ail_entry_t *);
38#else
39#define	xfs_ail_check(a)
40#endif /* DEBUG */
41
42
43/*
44 * This is called by the log manager code to determine the LSN
45 * of the tail of the log.  This is exactly the LSN of the first
46 * item in the AIL.  If the AIL is empty, then this function
47 * returns 0.
48 *
49 * We need the AIL lock in order to get a coherent read of the
50 * lsn of the last item in the AIL.
51 */
52xfs_lsn_t
53xfs_trans_tail_ail(
54	xfs_mount_t	*mp)
55{
56	xfs_lsn_t	lsn;
57	xfs_log_item_t	*lip;
58	SPLDECL(s);
59
60	AIL_LOCK(mp,s);
61	lip = xfs_ail_min(&(mp->m_ail));
62	if (lip == NULL) {
63		lsn = (xfs_lsn_t)0;
64	} else {
65		lsn = lip->li_lsn;
66	}
67	AIL_UNLOCK(mp, s);
68
69	return lsn;
70}
71
72/*
73 * xfs_trans_push_ail
74 *
75 * This routine is called to move the tail of the AIL
76 * forward.  It does this by trying to flush items in the AIL
77 * whose lsns are below the given threshold_lsn.
78 *
79 * The routine returns the lsn of the tail of the log.
80 */
81xfs_lsn_t
82xfs_trans_push_ail(
83	xfs_mount_t		*mp,
84	xfs_lsn_t		threshold_lsn)
85{
86	xfs_lsn_t		lsn;
87	xfs_log_item_t		*lip;
88	int			gen;
89	int			restarts;
90	int			lock_result;
91	int			flush_log;
92	SPLDECL(s);
93
94#define	XFS_TRANS_PUSH_AIL_RESTARTS	10
95
96	AIL_LOCK(mp,s);
97	lip = xfs_trans_first_ail(mp, &gen);
98	if (lip == NULL || XFS_FORCED_SHUTDOWN(mp)) {
99		/*
100		 * Just return if the AIL is empty.
101		 */
102		AIL_UNLOCK(mp, s);
103		return (xfs_lsn_t)0;
104	}
105
106	XFS_STATS_INC(xs_push_ail);
107
108	/*
109	 * While the item we are looking at is below the given threshold
110	 * try to flush it out.  Make sure to limit the number of times
111	 * we allow xfs_trans_next_ail() to restart scanning from the
112	 * beginning of the list.  We'd like not to stop until we've at least
113	 * tried to push on everything in the AIL with an LSN less than
114	 * the given threshold. However, we may give up before that if
115	 * we realize that we've been holding the AIL_LOCK for 'too long',
116	 * blocking interrupts. Currently, too long is < 500us roughly.
117	 */
118	flush_log = 0;
119	restarts = 0;
120	while (((restarts < XFS_TRANS_PUSH_AIL_RESTARTS) &&
121		(XFS_LSN_CMP(lip->li_lsn, threshold_lsn) < 0))) {
122		/*
123		 * If we can lock the item without sleeping, unlock
124		 * the AIL lock and flush the item.  Then re-grab the
125		 * AIL lock so we can look for the next item on the
126		 * AIL.  Since we unlock the AIL while we flush the
127		 * item, the next routine may start over again at the
128		 * the beginning of the list if anything has changed.
129		 * That is what the generation count is for.
130		 *
131		 * If we can't lock the item, either its holder will flush
132		 * it or it is already being flushed or it is being relogged.
133		 * In any of these case it is being taken care of and we
134		 * can just skip to the next item in the list.
135		 */
136		lock_result = IOP_TRYLOCK(lip);
137		switch (lock_result) {
138		      case XFS_ITEM_SUCCESS:
139			AIL_UNLOCK(mp, s);
140			XFS_STATS_INC(xs_push_ail_success);
141			IOP_PUSH(lip);
142			AIL_LOCK(mp,s);
143			break;
144
145		      case XFS_ITEM_PUSHBUF:
146			AIL_UNLOCK(mp, s);
147			XFS_STATS_INC(xs_push_ail_pushbuf);
148#ifdef XFSRACEDEBUG
149			delay_for_intr();
150			delay(300);
151#endif
152			ASSERT(lip->li_ops->iop_pushbuf);
153			ASSERT(lip);
154			IOP_PUSHBUF(lip);
155			AIL_LOCK(mp,s);
156			break;
157
158		      case XFS_ITEM_PINNED:
159			XFS_STATS_INC(xs_push_ail_pinned);
160			flush_log = 1;
161			break;
162
163		      case XFS_ITEM_LOCKED:
164			XFS_STATS_INC(xs_push_ail_locked);
165			break;
166
167		      case XFS_ITEM_FLUSHING:
168			XFS_STATS_INC(xs_push_ail_flushing);
169			break;
170
171		      default:
172			ASSERT(0);
173			break;
174		}
175
176		lip = xfs_trans_next_ail(mp, lip, &gen, &restarts);
177		if (lip == NULL) {
178			break;
179		}
180		if (XFS_FORCED_SHUTDOWN(mp)) {
181			/*
182			 * Just return if we shut down during the last try.
183			 */
184			AIL_UNLOCK(mp, s);
185			return (xfs_lsn_t)0;
186		}
187
188	}
189
190	if (flush_log) {
191		/*
192		 * If something we need to push out was pinned, then
193		 * push out the log so it will become unpinned and
194		 * move forward in the AIL.
195		 */
196		AIL_UNLOCK(mp, s);
197		XFS_STATS_INC(xs_push_ail_flush);
198		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
199		AIL_LOCK(mp, s);
200	}
201
202	lip = xfs_ail_min(&(mp->m_ail));
203	if (lip == NULL) {
204		lsn = (xfs_lsn_t)0;
205	} else {
206		lsn = lip->li_lsn;
207	}
208
209	AIL_UNLOCK(mp, s);
210	return lsn;
211}	/* xfs_trans_push_ail */
212
213
214/*
215 * This is to be called when an item is unlocked that may have
216 * been in the AIL.  It will wake up the first member of the AIL
217 * wait list if this item's unlocking might allow it to progress.
218 * If the item is in the AIL, then we need to get the AIL lock
219 * while doing our checking so we don't race with someone going
220 * to sleep waiting for this event in xfs_trans_push_ail().
221 */
222void
223xfs_trans_unlocked_item(
224	xfs_mount_t	*mp,
225	xfs_log_item_t	*lip)
226{
227	xfs_log_item_t	*min_lip;
228
229	/*
230	 * If we're forcibly shutting down, we may have
231	 * unlocked log items arbitrarily. The last thing
232	 * we want to do is to move the tail of the log
233	 * over some potentially valid data.
234	 */
235	if (!(lip->li_flags & XFS_LI_IN_AIL) ||
236	    XFS_FORCED_SHUTDOWN(mp)) {
237		return;
238	}
239
240	/*
241	 * This is the one case where we can call into xfs_ail_min()
242	 * without holding the AIL lock because we only care about the
243	 * case where we are at the tail of the AIL.  If the object isn't
244	 * at the tail, it doesn't matter what result we get back.  This
245	 * is slightly racy because since we were just unlocked, we could
246	 * go to sleep between the call to xfs_ail_min and the call to
247	 * xfs_log_move_tail, have someone else lock us, commit to us disk,
248	 * move us out of the tail of the AIL, and then we wake up.  However,
249	 * the call to xfs_log_move_tail() doesn't do anything if there's
250	 * not enough free space to wake people up so we're safe calling it.
251	 */
252	min_lip = xfs_ail_min(&mp->m_ail);
253
254	if (min_lip == lip)
255		xfs_log_move_tail(mp, 1);
256}	/* xfs_trans_unlocked_item */
257
258
259/*
260 * Update the position of the item in the AIL with the new
261 * lsn.  If it is not yet in the AIL, add it.  Otherwise, move
262 * it to its new position by removing it and re-adding it.
263 *
264 * Wakeup anyone with an lsn less than the item's lsn.  If the item
265 * we move in the AIL is the minimum one, update the tail lsn in the
266 * log manager.
267 *
268 * Increment the AIL's generation count to indicate that the tree
269 * has changed.
270 *
271 * This function must be called with the AIL lock held.  The lock
272 * is dropped before returning, so the caller must pass in the
273 * cookie returned by AIL_LOCK.
274 */
275void
276xfs_trans_update_ail(
277	xfs_mount_t	*mp,
278	xfs_log_item_t	*lip,
279	xfs_lsn_t	lsn,
280	unsigned long	s)
281{
282	xfs_ail_entry_t		*ailp;
283	xfs_log_item_t		*dlip=NULL;
284	xfs_log_item_t		*mlip;	/* ptr to minimum lip */
285
286	ailp = &(mp->m_ail);
287	mlip = xfs_ail_min(ailp);
288
289	if (lip->li_flags & XFS_LI_IN_AIL) {
290		dlip = xfs_ail_delete(ailp, lip);
291		ASSERT(dlip == lip);
292	} else {
293		lip->li_flags |= XFS_LI_IN_AIL;
294	}
295
296	lip->li_lsn = lsn;
297
298	xfs_ail_insert(ailp, lip);
299	mp->m_ail_gen++;
300
301	if (mlip == dlip) {
302		mlip = xfs_ail_min(&(mp->m_ail));
303		AIL_UNLOCK(mp, s);
304		xfs_log_move_tail(mp, mlip->li_lsn);
305	} else {
306		AIL_UNLOCK(mp, s);
307	}
308
309
310}	/* xfs_trans_update_ail */
311
312/*
313 * Delete the given item from the AIL.  It must already be in
314 * the AIL.
315 *
316 * Wakeup anyone with an lsn less than item's lsn.    If the item
317 * we delete in the AIL is the minimum one, update the tail lsn in the
318 * log manager.
319 *
320 * Clear the IN_AIL flag from the item, reset its lsn to 0, and
321 * bump the AIL's generation count to indicate that the tree
322 * has changed.
323 *
324 * This function must be called with the AIL lock held.  The lock
325 * is dropped before returning, so the caller must pass in the
326 * cookie returned by AIL_LOCK.
327 */
328void
329xfs_trans_delete_ail(
330	xfs_mount_t	*mp,
331	xfs_log_item_t	*lip,
332	unsigned long	s)
333{
334	xfs_ail_entry_t		*ailp;
335	xfs_log_item_t		*dlip;
336	xfs_log_item_t		*mlip;
337
338	if (lip->li_flags & XFS_LI_IN_AIL) {
339		ailp = &(mp->m_ail);
340		mlip = xfs_ail_min(ailp);
341		dlip = xfs_ail_delete(ailp, lip);
342		ASSERT(dlip == lip);
343
344
345		lip->li_flags &= ~XFS_LI_IN_AIL;
346		lip->li_lsn = 0;
347		mp->m_ail_gen++;
348
349		if (mlip == dlip) {
350			mlip = xfs_ail_min(&(mp->m_ail));
351			AIL_UNLOCK(mp, s);
352			xfs_log_move_tail(mp, (mlip ? mlip->li_lsn : 0));
353		} else {
354			AIL_UNLOCK(mp, s);
355		}
356	}
357	else {
358		/*
359		 * If the file system is not being shutdown, we are in
360		 * serious trouble if we get to this stage.
361		 */
362		if (XFS_FORCED_SHUTDOWN(mp))
363			AIL_UNLOCK(mp, s);
364		else {
365			xfs_cmn_err(XFS_PTAG_AILDELETE, CE_ALERT, mp,
366				"xfs_trans_delete_ail: attempting to delete a log item that is not in the AIL");
367			AIL_UNLOCK(mp, s);
368			xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
369		}
370	}
371}
372
373
374
375/*
376 * Return the item in the AIL with the smallest lsn.
377 * Return the current tree generation number for use
378 * in calls to xfs_trans_next_ail().
379 */
380xfs_log_item_t *
381xfs_trans_first_ail(
382	xfs_mount_t	*mp,
383	int		*gen)
384{
385	xfs_log_item_t	*lip;
386
387	lip = xfs_ail_min(&(mp->m_ail));
388	*gen = (int)mp->m_ail_gen;
389
390	return (lip);
391}
392
393/*
394 * If the generation count of the tree has not changed since the
395 * caller last took something from the AIL, then return the elmt
396 * in the tree which follows the one given.  If the count has changed,
397 * then return the minimum elmt of the AIL and bump the restarts counter
398 * if one is given.
399 */
400xfs_log_item_t *
401xfs_trans_next_ail(
402	xfs_mount_t	*mp,
403	xfs_log_item_t	*lip,
404	int		*gen,
405	int		*restarts)
406{
407	xfs_log_item_t	*nlip;
408
409	ASSERT(mp && lip && gen);
410	if (mp->m_ail_gen == *gen) {
411		nlip = xfs_ail_next(&(mp->m_ail), lip);
412	} else {
413		nlip = xfs_ail_min(&(mp->m_ail));
414		*gen = (int)mp->m_ail_gen;
415		if (restarts != NULL) {
416			XFS_STATS_INC(xs_push_ail_restarts);
417			(*restarts)++;
418		}
419	}
420
421	return (nlip);
422}
423
424
425/*
426 * The active item list (AIL) is a doubly linked list of log
427 * items sorted by ascending lsn.  The base of the list is
428 * a forw/back pointer pair embedded in the xfs mount structure.
429 * The base is initialized with both pointers pointing to the
430 * base.  This case always needs to be distinguished, because
431 * the base has no lsn to look at.  We almost always insert
432 * at the end of the list, so on inserts we search from the
433 * end of the list to find where the new item belongs.
434 */
435
436/*
437 * Initialize the doubly linked list to point only to itself.
438 */
439void
440xfs_trans_ail_init(
441	xfs_mount_t	*mp)
442{
443	mp->m_ail.ail_forw = (xfs_log_item_t*)&(mp->m_ail);
444	mp->m_ail.ail_back = (xfs_log_item_t*)&(mp->m_ail);
445}
446
447/*
448 * Insert the given log item into the AIL.
449 * We almost always insert at the end of the list, so on inserts
450 * we search from the end of the list to find where the
451 * new item belongs.
452 */
453STATIC void
454xfs_ail_insert(
455	xfs_ail_entry_t	*base,
456	xfs_log_item_t	*lip)
457/* ARGSUSED */
458{
459	xfs_log_item_t	*next_lip;
460
461	/*
462	 * If the list is empty, just insert the item.
463	 */
464	if (base->ail_back == (xfs_log_item_t*)base) {
465		base->ail_forw = lip;
466		base->ail_back = lip;
467		lip->li_ail.ail_forw = (xfs_log_item_t*)base;
468		lip->li_ail.ail_back = (xfs_log_item_t*)base;
469		return;
470	}
471
472	next_lip = base->ail_back;
473	while ((next_lip != (xfs_log_item_t*)base) &&
474	       (XFS_LSN_CMP(next_lip->li_lsn, lip->li_lsn) > 0)) {
475		next_lip = next_lip->li_ail.ail_back;
476	}
477	ASSERT((next_lip == (xfs_log_item_t*)base) ||
478	       (XFS_LSN_CMP(next_lip->li_lsn, lip->li_lsn) <= 0));
479	lip->li_ail.ail_forw = next_lip->li_ail.ail_forw;
480	lip->li_ail.ail_back = next_lip;
481	next_lip->li_ail.ail_forw = lip;
482	lip->li_ail.ail_forw->li_ail.ail_back = lip;
483
484	xfs_ail_check(base);
485	return;
486}
487
488/*
489 * Delete the given item from the AIL.  Return a pointer to the item.
490 */
491/*ARGSUSED*/
492STATIC xfs_log_item_t *
493xfs_ail_delete(
494	xfs_ail_entry_t	*base,
495	xfs_log_item_t	*lip)
496/* ARGSUSED */
497{
498	lip->li_ail.ail_forw->li_ail.ail_back = lip->li_ail.ail_back;
499	lip->li_ail.ail_back->li_ail.ail_forw = lip->li_ail.ail_forw;
500	lip->li_ail.ail_forw = NULL;
501	lip->li_ail.ail_back = NULL;
502
503	xfs_ail_check(base);
504	return lip;
505}
506
507/*
508 * Return a pointer to the first item in the AIL.
509 * If the AIL is empty, then return NULL.
510 */
511STATIC xfs_log_item_t *
512xfs_ail_min(
513	xfs_ail_entry_t	*base)
514/* ARGSUSED */
515{
516	register xfs_log_item_t *forw = base->ail_forw;
517	if (forw == (xfs_log_item_t*)base) {
518		return NULL;
519	}
520	return forw;
521}
522
523/*
524 * Return a pointer to the item which follows
525 * the given item in the AIL.  If the given item
526 * is the last item in the list, then return NULL.
527 */
528STATIC xfs_log_item_t *
529xfs_ail_next(
530	xfs_ail_entry_t	*base,
531	xfs_log_item_t	*lip)
532/* ARGSUSED */
533{
534	if (lip->li_ail.ail_forw == (xfs_log_item_t*)base) {
535		return NULL;
536	}
537	return lip->li_ail.ail_forw;
538
539}
540
541#ifdef DEBUG
542/*
543 * Check that the list is sorted as it should be.
544 */
545STATIC void
546xfs_ail_check(
547	xfs_ail_entry_t *base)
548{
549	xfs_log_item_t	*lip;
550	xfs_log_item_t	*prev_lip;
551
552	lip = base->ail_forw;
553	if (lip == (xfs_log_item_t*)base) {
554		/*
555		 * Make sure the pointers are correct when the list
556		 * is empty.
557		 */
558		ASSERT(base->ail_back == (xfs_log_item_t*)base);
559		return;
560	}
561
562	/*
563	 * Walk the list checking forward and backward pointers,
564	 * lsn ordering, and that every entry has the XFS_LI_IN_AIL
565	 * flag set.
566	 */
567	prev_lip = (xfs_log_item_t*)base;
568	while (lip != (xfs_log_item_t*)base) {
569		if (prev_lip != (xfs_log_item_t*)base) {
570			ASSERT(prev_lip->li_ail.ail_forw == lip);
571			ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
572		}
573		ASSERT(lip->li_ail.ail_back == prev_lip);
574		ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
575		prev_lip = lip;
576		lip = lip->li_ail.ail_forw;
577	}
578	ASSERT(lip == (xfs_log_item_t*)base);
579	ASSERT(base->ail_back == prev_lip);
580}
581#endif /* DEBUG */
582