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_dmapi.h"
26#include "xfs_mount.h"
27#include "xfs_trans_priv.h"
28#include "xfs_error.h"
29
30STATIC void xfs_ail_insert(xfs_ail_entry_t *, xfs_log_item_t *);
31STATIC xfs_log_item_t * xfs_ail_delete(xfs_ail_entry_t *, xfs_log_item_t *);
32STATIC xfs_log_item_t * xfs_ail_min(xfs_ail_entry_t *);
33STATIC xfs_log_item_t * xfs_ail_next(xfs_ail_entry_t *, xfs_log_item_t *);
34
35#ifdef DEBUG
36STATIC void xfs_ail_check(xfs_ail_entry_t *);
37#else
38#define	xfs_ail_check(a)
39#endif /* DEBUG */
40
41
42/*
43 * This is called by the log manager code to determine the LSN
44 * of the tail of the log.  This is exactly the LSN of the first
45 * item in the AIL.  If the AIL is empty, then this function
46 * returns 0.
47 *
48 * We need the AIL lock in order to get a coherent read of the
49 * lsn of the last item in the AIL.
50 */
51xfs_lsn_t
52xfs_trans_tail_ail(
53	xfs_mount_t	*mp)
54{
55	xfs_lsn_t	lsn;
56	xfs_log_item_t	*lip;
57	SPLDECL(s);
58
59	AIL_LOCK(mp,s);
60	lip = xfs_ail_min(&(mp->m_ail));
61	if (lip == NULL) {
62		lsn = (xfs_lsn_t)0;
63	} else {
64		lsn = lip->li_lsn;
65	}
66	AIL_UNLOCK(mp, s);
67
68	return lsn;
69}
70
71/*
72 * xfs_trans_push_ail
73 *
74 * This routine is called to move the tail of the AIL
75 * forward.  It does this by trying to flush items in the AIL
76 * whose lsns are below the given threshold_lsn.
77 *
78 * The routine returns the lsn of the tail of the log.
79 */
80xfs_lsn_t
81xfs_trans_push_ail(
82	xfs_mount_t		*mp,
83	xfs_lsn_t		threshold_lsn)
84{
85	xfs_lsn_t		lsn;
86	xfs_log_item_t		*lip;
87	int			gen;
88	int			restarts;
89	int			lock_result;
90	int			flush_log;
91	SPLDECL(s);
92
93#define	XFS_TRANS_PUSH_AIL_RESTARTS	1000
94
95	AIL_LOCK(mp,s);
96	lip = xfs_trans_first_ail(mp, &gen);
97	if (lip == NULL || XFS_FORCED_SHUTDOWN(mp)) {
98		/*
99		 * Just return if the AIL is empty.
100		 */
101		AIL_UNLOCK(mp, s);
102		return (xfs_lsn_t)0;
103	}
104
105	XFS_STATS_INC(xs_push_ail);
106
107	/*
108	 * While the item we are looking at is below the given threshold
109	 * try to flush it out.  Make sure to limit the number of times
110	 * we allow xfs_trans_next_ail() to restart scanning from the
111	 * beginning of the list.  We'd like not to stop until we've at least
112	 * tried to push on everything in the AIL with an LSN less than
113	 * the given threshold. However, we may give up before that if
114	 * we realize that we've been holding the AIL_LOCK for 'too long',
115	 * blocking interrupts. Currently, too long is < 500us roughly.
116	 */
117	flush_log = 0;
118	restarts = 0;
119	while (((restarts < XFS_TRANS_PUSH_AIL_RESTARTS) &&
120		(XFS_LSN_CMP(lip->li_lsn, threshold_lsn) < 0))) {
121		/*
122		 * If we can lock the item without sleeping, unlock
123		 * the AIL lock and flush the item.  Then re-grab the
124		 * AIL lock so we can look for the next item on the
125		 * AIL.  Since we unlock the AIL while we flush the
126		 * item, the next routine may start over again at the
127		 * the beginning of the list if anything has changed.
128		 * That is what the generation count is for.
129		 *
130		 * If we can't lock the item, either its holder will flush
131		 * it or it is already being flushed or it is being relogged.
132		 * In any of these case it is being taken care of and we
133		 * can just skip to the next item in the list.
134		 */
135		lock_result = IOP_TRYLOCK(lip);
136		switch (lock_result) {
137		      case XFS_ITEM_SUCCESS:
138			AIL_UNLOCK(mp, s);
139			XFS_STATS_INC(xs_push_ail_success);
140			IOP_PUSH(lip);
141			AIL_LOCK(mp,s);
142			break;
143
144		      case XFS_ITEM_PUSHBUF:
145			AIL_UNLOCK(mp, s);
146			XFS_STATS_INC(xs_push_ail_pushbuf);
147#ifdef XFSRACEDEBUG
148			delay_for_intr();
149			delay(300);
150#endif
151			ASSERT(lip->li_ops->iop_pushbuf);
152			ASSERT(lip);
153			IOP_PUSHBUF(lip);
154			AIL_LOCK(mp,s);
155			break;
156
157		      case XFS_ITEM_PINNED:
158			XFS_STATS_INC(xs_push_ail_pinned);
159			flush_log = 1;
160			break;
161
162		      case XFS_ITEM_LOCKED:
163			XFS_STATS_INC(xs_push_ail_locked);
164			break;
165
166		      case XFS_ITEM_FLUSHING:
167			XFS_STATS_INC(xs_push_ail_flushing);
168			break;
169
170		      default:
171			ASSERT(0);
172			break;
173		}
174
175		lip = xfs_trans_next_ail(mp, lip, &gen, &restarts);
176		if (lip == NULL) {
177			break;
178		}
179		if (XFS_FORCED_SHUTDOWN(mp)) {
180			/*
181			 * Just return if we shut down during the last try.
182			 */
183			AIL_UNLOCK(mp, s);
184			return (xfs_lsn_t)0;
185		}
186
187	}
188
189	if (flush_log) {
190		/*
191		 * If something we need to push out was pinned, then
192		 * push out the log so it will become unpinned and
193		 * move forward in the AIL.
194		 */
195		AIL_UNLOCK(mp, s);
196		XFS_STATS_INC(xs_push_ail_flush);
197		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
198		AIL_LOCK(mp, s);
199	}
200
201	lip = xfs_ail_min(&(mp->m_ail));
202	if (lip == NULL) {
203		lsn = (xfs_lsn_t)0;
204	} else {
205		lsn = lip->li_lsn;
206	}
207
208	AIL_UNLOCK(mp, s);
209	return lsn;
210}	/* xfs_trans_push_ail */
211
212
213/*
214 * This is to be called when an item is unlocked that may have
215 * been in the AIL.  It will wake up the first member of the AIL
216 * wait list if this item's unlocking might allow it to progress.
217 * If the item is in the AIL, then we need to get the AIL lock
218 * while doing our checking so we don't race with someone going
219 * to sleep waiting for this event in xfs_trans_push_ail().
220 */
221void
222xfs_trans_unlocked_item(
223	xfs_mount_t	*mp,
224	xfs_log_item_t	*lip)
225{
226	xfs_log_item_t	*min_lip;
227
228	/*
229	 * If we're forcibly shutting down, we may have
230	 * unlocked log items arbitrarily. The last thing
231	 * we want to do is to move the tail of the log
232	 * over some potentially valid data.
233	 */
234	if (!(lip->li_flags & XFS_LI_IN_AIL) ||
235	    XFS_FORCED_SHUTDOWN(mp)) {
236		return;
237	}
238
239	/*
240	 * This is the one case where we can call into xfs_ail_min()
241	 * without holding the AIL lock because we only care about the
242	 * case where we are at the tail of the AIL.  If the object isn't
243	 * at the tail, it doesn't matter what result we get back.  This
244	 * is slightly racy because since we were just unlocked, we could
245	 * go to sleep between the call to xfs_ail_min and the call to
246	 * xfs_log_move_tail, have someone else lock us, commit to us disk,
247	 * move us out of the tail of the AIL, and then we wake up.  However,
248	 * the call to xfs_log_move_tail() doesn't do anything if there's
249	 * not enough free space to wake people up so we're safe calling it.
250	 */
251	min_lip = xfs_ail_min(&mp->m_ail);
252
253	if (min_lip == lip)
254		xfs_log_move_tail(mp, 1);
255}	/* xfs_trans_unlocked_item */
256
257
258/*
259 * Update the position of the item in the AIL with the new
260 * lsn.  If it is not yet in the AIL, add it.  Otherwise, move
261 * it to its new position by removing it and re-adding it.
262 *
263 * Wakeup anyone with an lsn less than the item's lsn.  If the item
264 * we move in the AIL is the minimum one, update the tail lsn in the
265 * log manager.
266 *
267 * Increment the AIL's generation count to indicate that the tree
268 * has changed.
269 *
270 * This function must be called with the AIL lock held.  The lock
271 * is dropped before returning, so the caller must pass in the
272 * cookie returned by AIL_LOCK.
273 */
274void
275xfs_trans_update_ail(
276	xfs_mount_t	*mp,
277	xfs_log_item_t	*lip,
278	xfs_lsn_t	lsn,
279	unsigned long	s) __releases(mp->m_ail_lock)
280{
281	xfs_ail_entry_t		*ailp;
282	xfs_log_item_t		*dlip=NULL;
283	xfs_log_item_t		*mlip;	/* ptr to minimum lip */
284
285	ailp = &(mp->m_ail);
286	mlip = xfs_ail_min(ailp);
287
288	if (lip->li_flags & XFS_LI_IN_AIL) {
289		dlip = xfs_ail_delete(ailp, lip);
290		ASSERT(dlip == lip);
291	} else {
292		lip->li_flags |= XFS_LI_IN_AIL;
293	}
294
295	lip->li_lsn = lsn;
296
297	xfs_ail_insert(ailp, lip);
298	mp->m_ail_gen++;
299
300	if (mlip == dlip) {
301		mlip = xfs_ail_min(&(mp->m_ail));
302		AIL_UNLOCK(mp, s);
303		xfs_log_move_tail(mp, mlip->li_lsn);
304	} else {
305		AIL_UNLOCK(mp, s);
306	}
307
308
309}	/* xfs_trans_update_ail */
310
311/*
312 * Delete the given item from the AIL.  It must already be in
313 * the AIL.
314 *
315 * Wakeup anyone with an lsn less than item's lsn.    If the item
316 * we delete in the AIL is the minimum one, update the tail lsn in the
317 * log manager.
318 *
319 * Clear the IN_AIL flag from the item, reset its lsn to 0, and
320 * bump the AIL's generation count to indicate that the tree
321 * has changed.
322 *
323 * This function must be called with the AIL lock held.  The lock
324 * is dropped before returning, so the caller must pass in the
325 * cookie returned by AIL_LOCK.
326 */
327void
328xfs_trans_delete_ail(
329	xfs_mount_t	*mp,
330	xfs_log_item_t	*lip,
331	unsigned long	s) __releases(mp->m_ail_lock)
332{
333	xfs_ail_entry_t		*ailp;
334	xfs_log_item_t		*dlip;
335	xfs_log_item_t		*mlip;
336
337	if (lip->li_flags & XFS_LI_IN_AIL) {
338		ailp = &(mp->m_ail);
339		mlip = xfs_ail_min(ailp);
340		dlip = xfs_ail_delete(ailp, lip);
341		ASSERT(dlip == lip);
342
343
344		lip->li_flags &= ~XFS_LI_IN_AIL;
345		lip->li_lsn = 0;
346		mp->m_ail_gen++;
347
348		if (mlip == dlip) {
349			mlip = xfs_ail_min(&(mp->m_ail));
350			AIL_UNLOCK(mp, s);
351			xfs_log_move_tail(mp, (mlip ? mlip->li_lsn : 0));
352		} else {
353			AIL_UNLOCK(mp, s);
354		}
355	}
356	else {
357		/*
358		 * If the file system is not being shutdown, we are in
359		 * serious trouble if we get to this stage.
360		 */
361		if (XFS_FORCED_SHUTDOWN(mp))
362			AIL_UNLOCK(mp, s);
363		else {
364			xfs_cmn_err(XFS_PTAG_AILDELETE, CE_ALERT, mp,
365		"%s: attempting to delete a log item that is not in the AIL",
366					__FUNCTION__);
367			AIL_UNLOCK(mp, s);
368			xfs_force_shutdown(mp, SHUTDOWN_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