2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
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.
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.
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
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_error.h"
31 STATIC
void xfs_ail_insert(xfs_ail_entry_t
*, xfs_log_item_t
*);
32 STATIC xfs_log_item_t
* xfs_ail_delete(xfs_ail_entry_t
*, xfs_log_item_t
*);
33 STATIC xfs_log_item_t
* xfs_ail_min(xfs_ail_entry_t
*);
34 STATIC xfs_log_item_t
* xfs_ail_next(xfs_ail_entry_t
*, xfs_log_item_t
*);
37 STATIC
void xfs_ail_check(xfs_ail_entry_t
*, xfs_log_item_t
*);
39 #define xfs_ail_check(a,l)
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
49 * We need the AIL lock in order to get a coherent read of the
50 * lsn of the last item in the AIL.
59 spin_lock(&mp
->m_ail_lock
);
60 lip
= xfs_ail_min(&(mp
->m_ail
.xa_ail
));
66 spin_unlock(&mp
->m_ail_lock
);
74 * This routine is called to move the tail of the AIL forward. It does this by
75 * trying to flush items in the AIL whose lsns are below the given
78 * the push is run asynchronously in a separate thread, so we return the tail
79 * of the log right now instead of the tail after the push. This means we will
80 * either continue right away, or we will sleep waiting on the async thread to
83 * We do this unlocked - we only need to know whether there is anything in the
84 * AIL at the time we are called. We don't need to access the contents of
85 * any of the objects, so the lock is not needed.
90 xfs_lsn_t threshold_lsn
)
94 lip
= xfs_ail_min(&mp
->m_ail
.xa_ail
);
95 if (lip
&& !XFS_FORCED_SHUTDOWN(mp
)) {
96 if (XFS_LSN_CMP(threshold_lsn
, mp
->m_ail
.xa_target
) > 0)
97 xfsaild_wakeup(mp
, threshold_lsn
);
102 * Return the item in the AIL with the current lsn.
103 * Return the current tree generation number for use
104 * in calls to xfs_trans_next_ail().
106 STATIC xfs_log_item_t
*
107 xfs_trans_first_push_ail(
114 lip
= xfs_ail_min(&(mp
->m_ail
.xa_ail
));
115 *gen
= (int)mp
->m_ail
.xa_gen
;
119 while (lip
&& (XFS_LSN_CMP(lip
->li_lsn
, lsn
) < 0))
120 lip
= lip
->li_ail
.ail_forw
;
126 * Function that does the work of pushing on the AIL
133 long tout
= 1000; /* milliseconds */
134 xfs_lsn_t last_pushed_lsn
= *last_lsn
;
135 xfs_lsn_t target
= mp
->m_ail
.xa_target
;
140 int flush_log
, count
, stuck
;
142 #define XFS_TRANS_PUSH_AIL_RESTARTS 10
144 spin_lock(&mp
->m_ail_lock
);
145 lip
= xfs_trans_first_push_ail(mp
, &gen
, *last_lsn
);
146 if (!lip
|| XFS_FORCED_SHUTDOWN(mp
)) {
148 * AIL is empty or our push has reached the end.
150 spin_unlock(&mp
->m_ail_lock
);
155 XFS_STATS_INC(xs_push_ail
);
158 * While the item we are looking at is below the given threshold
159 * try to flush it out. We'd like not to stop until we've at least
160 * tried to push on everything in the AIL with an LSN less than
161 * the given threshold.
163 * However, we will stop after a certain number of pushes and wait
164 * for a reduced timeout to fire before pushing further. This
165 * prevents use from spinning when we can't do anything or there is
166 * lots of contention on the AIL lists.
170 flush_log
= stuck
= count
= restarts
= 0;
171 while ((XFS_LSN_CMP(lip
->li_lsn
, target
) < 0)) {
174 * If we can lock the item without sleeping, unlock the AIL
175 * lock and flush the item. Then re-grab the AIL lock so we
176 * can look for the next item on the AIL. List changes are
177 * handled by the AIL lookup functions internally
179 * If we can't lock the item, either its holder will flush it
180 * or it is already being flushed or it is being relogged. In
181 * any of these case it is being taken care of and we can just
182 * skip to the next item in the list.
184 lock_result
= IOP_TRYLOCK(lip
);
185 spin_unlock(&mp
->m_ail_lock
);
186 switch (lock_result
) {
187 case XFS_ITEM_SUCCESS
:
188 XFS_STATS_INC(xs_push_ail_success
);
190 last_pushed_lsn
= lsn
;
193 case XFS_ITEM_PUSHBUF
:
194 XFS_STATS_INC(xs_push_ail_pushbuf
);
196 last_pushed_lsn
= lsn
;
199 case XFS_ITEM_PINNED
:
200 XFS_STATS_INC(xs_push_ail_pinned
);
205 case XFS_ITEM_LOCKED
:
206 XFS_STATS_INC(xs_push_ail_locked
);
207 last_pushed_lsn
= lsn
;
211 case XFS_ITEM_FLUSHING
:
212 XFS_STATS_INC(xs_push_ail_flushing
);
213 last_pushed_lsn
= lsn
;
222 spin_lock(&mp
->m_ail_lock
);
223 /* should we bother continuing? */
224 if (XFS_FORCED_SHUTDOWN(mp
))
231 * Are there too many items we can't do anything with?
232 * If we we are skipping too many items because we can't flush
233 * them or they are already being flushed, we back off and
234 * given them time to complete whatever operation is being
235 * done. i.e. remove pressure from the AIL while we can't make
236 * progress so traversals don't slow down further inserts and
237 * removals to/from the AIL.
239 * The value of 100 is an arbitrary magic number based on
245 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, &restarts
);
248 if (restarts
> XFS_TRANS_PUSH_AIL_RESTARTS
)
252 spin_unlock(&mp
->m_ail_lock
);
256 * If something we need to push out was pinned, then
257 * push out the log so it will become unpinned and
258 * move forward in the AIL.
260 XFS_STATS_INC(xs_push_ail_flush
);
261 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
265 /* We're past our target or empty, so idle */
267 } else if (XFS_LSN_CMP(lsn
, target
) >= 0) {
269 * We reached the target so wait a bit longer for I/O to
270 * complete and remove pushed items from the AIL before we
271 * start the next scan from the start of the AIL.
275 } else if ((restarts
> XFS_TRANS_PUSH_AIL_RESTARTS
) ||
276 ((stuck
* 100) / count
> 90)) {
278 * Either there is a lot of contention on the AIL or we
279 * are stuck due to operations in progress. "Stuck" in this
280 * case is defined as >90% of the items we tried to push
283 * Backoff a bit more to allow some I/O to complete before
284 * continuing from where we were.
289 *last_lsn
= last_pushed_lsn
;
295 * This is to be called when an item is unlocked that may have
296 * been in the AIL. It will wake up the first member of the AIL
297 * wait list if this item's unlocking might allow it to progress.
298 * If the item is in the AIL, then we need to get the AIL lock
299 * while doing our checking so we don't race with someone going
300 * to sleep waiting for this event in xfs_trans_push_ail().
303 xfs_trans_unlocked_item(
307 xfs_log_item_t
*min_lip
;
310 * If we're forcibly shutting down, we may have
311 * unlocked log items arbitrarily. The last thing
312 * we want to do is to move the tail of the log
313 * over some potentially valid data.
315 if (!(lip
->li_flags
& XFS_LI_IN_AIL
) ||
316 XFS_FORCED_SHUTDOWN(mp
)) {
321 * This is the one case where we can call into xfs_ail_min()
322 * without holding the AIL lock because we only care about the
323 * case where we are at the tail of the AIL. If the object isn't
324 * at the tail, it doesn't matter what result we get back. This
325 * is slightly racy because since we were just unlocked, we could
326 * go to sleep between the call to xfs_ail_min and the call to
327 * xfs_log_move_tail, have someone else lock us, commit to us disk,
328 * move us out of the tail of the AIL, and then we wake up. However,
329 * the call to xfs_log_move_tail() doesn't do anything if there's
330 * not enough free space to wake people up so we're safe calling it.
332 min_lip
= xfs_ail_min(&mp
->m_ail
.xa_ail
);
335 xfs_log_move_tail(mp
, 1);
336 } /* xfs_trans_unlocked_item */
340 * Update the position of the item in the AIL with the new
341 * lsn. If it is not yet in the AIL, add it. Otherwise, move
342 * it to its new position by removing it and re-adding it.
344 * Wakeup anyone with an lsn less than the item's lsn. If the item
345 * we move in the AIL is the minimum one, update the tail lsn in the
348 * Increment the AIL's generation count to indicate that the tree
351 * This function must be called with the AIL lock held. The lock
352 * is dropped before returning.
355 xfs_trans_update_ail(
358 xfs_lsn_t lsn
) __releases(mp
->m_ail_lock
)
360 xfs_ail_entry_t
*ailp
;
361 xfs_log_item_t
*dlip
=NULL
;
362 xfs_log_item_t
*mlip
; /* ptr to minimum lip */
364 ailp
= &(mp
->m_ail
.xa_ail
);
365 mlip
= xfs_ail_min(ailp
);
367 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
368 dlip
= xfs_ail_delete(ailp
, lip
);
371 lip
->li_flags
|= XFS_LI_IN_AIL
;
376 xfs_ail_insert(ailp
, lip
);
380 mlip
= xfs_ail_min(&(mp
->m_ail
.xa_ail
));
381 spin_unlock(&mp
->m_ail_lock
);
382 xfs_log_move_tail(mp
, mlip
->li_lsn
);
384 spin_unlock(&mp
->m_ail_lock
);
388 } /* xfs_trans_update_ail */
391 * Delete the given item from the AIL. It must already be in
394 * Wakeup anyone with an lsn less than item's lsn. If the item
395 * we delete in the AIL is the minimum one, update the tail lsn in the
398 * Clear the IN_AIL flag from the item, reset its lsn to 0, and
399 * bump the AIL's generation count to indicate that the tree
402 * This function must be called with the AIL lock held. The lock
403 * is dropped before returning.
406 xfs_trans_delete_ail(
408 xfs_log_item_t
*lip
) __releases(mp
->m_ail_lock
)
410 xfs_ail_entry_t
*ailp
;
411 xfs_log_item_t
*dlip
;
412 xfs_log_item_t
*mlip
;
414 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
415 ailp
= &(mp
->m_ail
.xa_ail
);
416 mlip
= xfs_ail_min(ailp
);
417 dlip
= xfs_ail_delete(ailp
, lip
);
421 lip
->li_flags
&= ~XFS_LI_IN_AIL
;
426 mlip
= xfs_ail_min(&(mp
->m_ail
.xa_ail
));
427 spin_unlock(&mp
->m_ail_lock
);
428 xfs_log_move_tail(mp
, (mlip
? mlip
->li_lsn
: 0));
430 spin_unlock(&mp
->m_ail_lock
);
435 * If the file system is not being shutdown, we are in
436 * serious trouble if we get to this stage.
438 if (XFS_FORCED_SHUTDOWN(mp
))
439 spin_unlock(&mp
->m_ail_lock
);
441 xfs_cmn_err(XFS_PTAG_AILDELETE
, CE_ALERT
, mp
,
442 "%s: attempting to delete a log item that is not in the AIL",
444 spin_unlock(&mp
->m_ail_lock
);
445 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
453 * Return the item in the AIL with the smallest lsn.
454 * Return the current tree generation number for use
455 * in calls to xfs_trans_next_ail().
464 lip
= xfs_ail_min(&(mp
->m_ail
.xa_ail
));
465 *gen
= (int)mp
->m_ail
.xa_gen
;
471 * If the generation count of the tree has not changed since the
472 * caller last took something from the AIL, then return the elmt
473 * in the tree which follows the one given. If the count has changed,
474 * then return the minimum elmt of the AIL and bump the restarts counter
484 xfs_log_item_t
*nlip
;
486 ASSERT(mp
&& lip
&& gen
);
487 if (mp
->m_ail
.xa_gen
== *gen
) {
488 nlip
= xfs_ail_next(&(mp
->m_ail
.xa_ail
), lip
);
490 nlip
= xfs_ail_min(&(mp
->m_ail
).xa_ail
);
491 *gen
= (int)mp
->m_ail
.xa_gen
;
492 if (restarts
!= NULL
) {
493 XFS_STATS_INC(xs_push_ail_restarts
);
503 * The active item list (AIL) is a doubly linked list of log
504 * items sorted by ascending lsn. The base of the list is
505 * a forw/back pointer pair embedded in the xfs mount structure.
506 * The base is initialized with both pointers pointing to the
507 * base. This case always needs to be distinguished, because
508 * the base has no lsn to look at. We almost always insert
509 * at the end of the list, so on inserts we search from the
510 * end of the list to find where the new item belongs.
514 * Initialize the doubly linked list to point only to itself.
520 mp
->m_ail
.xa_ail
.ail_forw
= (xfs_log_item_t
*)&mp
->m_ail
.xa_ail
;
521 mp
->m_ail
.xa_ail
.ail_back
= (xfs_log_item_t
*)&mp
->m_ail
.xa_ail
;
522 return xfsaild_start(mp
);
526 xfs_trans_ail_destroy(
533 * Insert the given log item into the AIL.
534 * We almost always insert at the end of the list, so on inserts
535 * we search from the end of the list to find where the
540 xfs_ail_entry_t
*base
,
544 xfs_log_item_t
*next_lip
;
547 * If the list is empty, just insert the item.
549 if (base
->ail_back
== (xfs_log_item_t
*)base
) {
550 base
->ail_forw
= lip
;
551 base
->ail_back
= lip
;
552 lip
->li_ail
.ail_forw
= (xfs_log_item_t
*)base
;
553 lip
->li_ail
.ail_back
= (xfs_log_item_t
*)base
;
557 next_lip
= base
->ail_back
;
558 while ((next_lip
!= (xfs_log_item_t
*)base
) &&
559 (XFS_LSN_CMP(next_lip
->li_lsn
, lip
->li_lsn
) > 0)) {
560 next_lip
= next_lip
->li_ail
.ail_back
;
562 ASSERT((next_lip
== (xfs_log_item_t
*)base
) ||
563 (XFS_LSN_CMP(next_lip
->li_lsn
, lip
->li_lsn
) <= 0));
564 lip
->li_ail
.ail_forw
= next_lip
->li_ail
.ail_forw
;
565 lip
->li_ail
.ail_back
= next_lip
;
566 next_lip
->li_ail
.ail_forw
= lip
;
567 lip
->li_ail
.ail_forw
->li_ail
.ail_back
= lip
;
569 xfs_ail_check(base
, lip
);
574 * Delete the given item from the AIL. Return a pointer to the item.
577 STATIC xfs_log_item_t
*
579 xfs_ail_entry_t
*base
,
583 xfs_ail_check(base
, lip
);
584 lip
->li_ail
.ail_forw
->li_ail
.ail_back
= lip
->li_ail
.ail_back
;
585 lip
->li_ail
.ail_back
->li_ail
.ail_forw
= lip
->li_ail
.ail_forw
;
586 lip
->li_ail
.ail_forw
= NULL
;
587 lip
->li_ail
.ail_back
= NULL
;
593 * Return a pointer to the first item in the AIL.
594 * If the AIL is empty, then return NULL.
596 STATIC xfs_log_item_t
*
598 xfs_ail_entry_t
*base
)
601 register xfs_log_item_t
*forw
= base
->ail_forw
;
602 if (forw
== (xfs_log_item_t
*)base
) {
609 * Return a pointer to the item which follows
610 * the given item in the AIL. If the given item
611 * is the last item in the list, then return NULL.
613 STATIC xfs_log_item_t
*
615 xfs_ail_entry_t
*base
,
619 if (lip
->li_ail
.ail_forw
== (xfs_log_item_t
*)base
) {
622 return lip
->li_ail
.ail_forw
;
628 * Check that the list is sorted as it should be.
632 xfs_ail_entry_t
*base
,
635 xfs_log_item_t
*prev_lip
;
637 prev_lip
= base
->ail_forw
;
638 if (prev_lip
== (xfs_log_item_t
*)base
) {
640 * Make sure the pointers are correct when the list
643 ASSERT(base
->ail_back
== (xfs_log_item_t
*)base
);
648 * Check the next and previous entries are valid.
650 ASSERT((lip
->li_flags
& XFS_LI_IN_AIL
) != 0);
651 prev_lip
= lip
->li_ail
.ail_back
;
652 if (prev_lip
!= (xfs_log_item_t
*)base
) {
653 ASSERT(prev_lip
->li_ail
.ail_forw
== lip
);
654 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) <= 0);
656 prev_lip
= lip
->li_ail
.ail_forw
;
657 if (prev_lip
!= (xfs_log_item_t
*)base
) {
658 ASSERT(prev_lip
->li_ail
.ail_back
== lip
);
659 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) >= 0);
663 #ifdef XFS_TRANS_DEBUG
665 * Walk the list checking forward and backward pointers,
666 * lsn ordering, and that every entry has the XFS_LI_IN_AIL
667 * flag set. This is really expensive, so only do it when
668 * specifically debugging the transaction subsystem.
670 prev_lip
= (xfs_log_item_t
*)base
;
671 while (lip
!= (xfs_log_item_t
*)base
) {
672 if (prev_lip
!= (xfs_log_item_t
*)base
) {
673 ASSERT(prev_lip
->li_ail
.ail_forw
== lip
);
674 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) <= 0);
676 ASSERT(lip
->li_ail
.ail_back
== prev_lip
);
677 ASSERT((lip
->li_flags
& XFS_LI_IN_AIL
) != 0);
679 lip
= lip
->li_ail
.ail_forw
;
681 ASSERT(lip
== (xfs_log_item_t
*)base
);
682 ASSERT(base
->ail_back
== prev_lip
);
683 #endif /* XFS_TRANS_DEBUG */