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 reached the target so wait a bit longer for I/O to complete and
266 * remove pushed items from the AIL before we start the next scan from
267 * the start of the AIL.
269 if ((XFS_LSN_CMP(lsn, target) >= 0)) {
272 } else if ((restarts > XFS_TRANS_PUSH_AIL_RESTARTS) ||
273 (count && ((stuck * 100) / count > 90))) {
275 * Either there is a lot of contention on the AIL or we
276 * are stuck due to operations in progress. "Stuck" in this
277 * case is defined as >90% of the items we tried to push
280 * Backoff a bit more to allow some I/O to complete before
281 * continuing from where we were.
286 *last_lsn = last_pushed_lsn;
292 * This is to be called when an item is unlocked that may have
293 * been in the AIL. It will wake up the first member of the AIL
294 * wait list if this item's unlocking might allow it to progress.
295 * If the item is in the AIL, then we need to get the AIL lock
296 * while doing our checking so we don't race with someone going
297 * to sleep waiting for this event in xfs_trans_push_ail().
300 xfs_trans_unlocked_item(
304 xfs_log_item_t *min_lip;
307 * If we're forcibly shutting down, we may have
308 * unlocked log items arbitrarily. The last thing
309 * we want to do is to move the tail of the log
310 * over some potentially valid data.
312 if (!(lip->li_flags & XFS_LI_IN_AIL) ||
313 XFS_FORCED_SHUTDOWN(mp)) {
318 * This is the one case where we can call into xfs_ail_min()
319 * without holding the AIL lock because we only care about the
320 * case where we are at the tail of the AIL. If the object isn't
321 * at the tail, it doesn't matter what result we get back. This
322 * is slightly racy because since we were just unlocked, we could
323 * go to sleep between the call to xfs_ail_min and the call to
324 * xfs_log_move_tail, have someone else lock us, commit to us disk,
325 * move us out of the tail of the AIL, and then we wake up. However,
326 * the call to xfs_log_move_tail() doesn't do anything if there's
327 * not enough free space to wake people up so we're safe calling it.
329 min_lip = xfs_ail_min(&mp->m_ail.xa_ail);
332 xfs_log_move_tail(mp, 1);
333 } /* xfs_trans_unlocked_item */
337 * Update the position of the item in the AIL with the new
338 * lsn. If it is not yet in the AIL, add it. Otherwise, move
339 * it to its new position by removing it and re-adding it.
341 * Wakeup anyone with an lsn less than the item's lsn. If the item
342 * we move in the AIL is the minimum one, update the tail lsn in the
345 * Increment the AIL's generation count to indicate that the tree
348 * This function must be called with the AIL lock held. The lock
349 * is dropped before returning.
352 xfs_trans_update_ail(
355 xfs_lsn_t lsn) __releases(mp->m_ail_lock)
357 xfs_ail_entry_t *ailp;
358 xfs_log_item_t *dlip=NULL;
359 xfs_log_item_t *mlip; /* ptr to minimum lip */
361 ailp = &(mp->m_ail.xa_ail);
362 mlip = xfs_ail_min(ailp);
364 if (lip->li_flags & XFS_LI_IN_AIL) {
365 dlip = xfs_ail_delete(ailp, lip);
368 lip->li_flags |= XFS_LI_IN_AIL;
373 xfs_ail_insert(ailp, lip);
377 mlip = xfs_ail_min(&(mp->m_ail.xa_ail));
378 spin_unlock(&mp->m_ail_lock);
379 xfs_log_move_tail(mp, mlip->li_lsn);
381 spin_unlock(&mp->m_ail_lock);
385 } /* xfs_trans_update_ail */
388 * Delete the given item from the AIL. It must already be in
391 * Wakeup anyone with an lsn less than item's lsn. If the item
392 * we delete in the AIL is the minimum one, update the tail lsn in the
395 * Clear the IN_AIL flag from the item, reset its lsn to 0, and
396 * bump the AIL's generation count to indicate that the tree
399 * This function must be called with the AIL lock held. The lock
400 * is dropped before returning.
403 xfs_trans_delete_ail(
405 xfs_log_item_t *lip) __releases(mp->m_ail_lock)
407 xfs_ail_entry_t *ailp;
408 xfs_log_item_t *dlip;
409 xfs_log_item_t *mlip;
411 if (lip->li_flags & XFS_LI_IN_AIL) {
412 ailp = &(mp->m_ail.xa_ail);
413 mlip = xfs_ail_min(ailp);
414 dlip = xfs_ail_delete(ailp, lip);
418 lip->li_flags &= ~XFS_LI_IN_AIL;
423 mlip = xfs_ail_min(&(mp->m_ail.xa_ail));
424 spin_unlock(&mp->m_ail_lock);
425 xfs_log_move_tail(mp, (mlip ? mlip->li_lsn : 0));
427 spin_unlock(&mp->m_ail_lock);
432 * If the file system is not being shutdown, we are in
433 * serious trouble if we get to this stage.
435 if (XFS_FORCED_SHUTDOWN(mp))
436 spin_unlock(&mp->m_ail_lock);
438 xfs_cmn_err(XFS_PTAG_AILDELETE, CE_ALERT, mp,
439 "%s: attempting to delete a log item that is not in the AIL",
441 spin_unlock(&mp->m_ail_lock);
442 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
450 * Return the item in the AIL with the smallest lsn.
451 * Return the current tree generation number for use
452 * in calls to xfs_trans_next_ail().
461 lip = xfs_ail_min(&(mp->m_ail.xa_ail));
462 *gen = (int)mp->m_ail.xa_gen;
468 * If the generation count of the tree has not changed since the
469 * caller last took something from the AIL, then return the elmt
470 * in the tree which follows the one given. If the count has changed,
471 * then return the minimum elmt of the AIL and bump the restarts counter
481 xfs_log_item_t *nlip;
483 ASSERT(mp && lip && gen);
484 if (mp->m_ail.xa_gen == *gen) {
485 nlip = xfs_ail_next(&(mp->m_ail.xa_ail), lip);
487 nlip = xfs_ail_min(&(mp->m_ail).xa_ail);
488 *gen = (int)mp->m_ail.xa_gen;
489 if (restarts != NULL) {
490 XFS_STATS_INC(xs_push_ail_restarts);
500 * The active item list (AIL) is a doubly linked list of log
501 * items sorted by ascending lsn. The base of the list is
502 * a forw/back pointer pair embedded in the xfs mount structure.
503 * The base is initialized with both pointers pointing to the
504 * base. This case always needs to be distinguished, because
505 * the base has no lsn to look at. We almost always insert
506 * at the end of the list, so on inserts we search from the
507 * end of the list to find where the new item belongs.
511 * Initialize the doubly linked list to point only to itself.
517 mp->m_ail.xa_ail.ail_forw = (xfs_log_item_t*)&mp->m_ail.xa_ail;
518 mp->m_ail.xa_ail.ail_back = (xfs_log_item_t*)&mp->m_ail.xa_ail;
519 return xfsaild_start(mp);
523 xfs_trans_ail_destroy(
530 * Insert the given log item into the AIL.
531 * We almost always insert at the end of the list, so on inserts
532 * we search from the end of the list to find where the
537 xfs_ail_entry_t *base,
541 xfs_log_item_t *next_lip;
544 * If the list is empty, just insert the item.
546 if (base->ail_back == (xfs_log_item_t*)base) {
547 base->ail_forw = lip;
548 base->ail_back = lip;
549 lip->li_ail.ail_forw = (xfs_log_item_t*)base;
550 lip->li_ail.ail_back = (xfs_log_item_t*)base;
554 next_lip = base->ail_back;
555 while ((next_lip != (xfs_log_item_t*)base) &&
556 (XFS_LSN_CMP(next_lip->li_lsn, lip->li_lsn) > 0)) {
557 next_lip = next_lip->li_ail.ail_back;
559 ASSERT((next_lip == (xfs_log_item_t*)base) ||
560 (XFS_LSN_CMP(next_lip->li_lsn, lip->li_lsn) <= 0));
561 lip->li_ail.ail_forw = next_lip->li_ail.ail_forw;
562 lip->li_ail.ail_back = next_lip;
563 next_lip->li_ail.ail_forw = lip;
564 lip->li_ail.ail_forw->li_ail.ail_back = lip;
566 xfs_ail_check(base, lip);
571 * Delete the given item from the AIL. Return a pointer to the item.
574 STATIC xfs_log_item_t *
576 xfs_ail_entry_t *base,
580 xfs_ail_check(base, lip);
581 lip->li_ail.ail_forw->li_ail.ail_back = lip->li_ail.ail_back;
582 lip->li_ail.ail_back->li_ail.ail_forw = lip->li_ail.ail_forw;
583 lip->li_ail.ail_forw = NULL;
584 lip->li_ail.ail_back = NULL;
590 * Return a pointer to the first item in the AIL.
591 * If the AIL is empty, then return NULL.
593 STATIC xfs_log_item_t *
595 xfs_ail_entry_t *base)
598 register xfs_log_item_t *forw = base->ail_forw;
599 if (forw == (xfs_log_item_t*)base) {
606 * Return a pointer to the item which follows
607 * the given item in the AIL. If the given item
608 * is the last item in the list, then return NULL.
610 STATIC xfs_log_item_t *
612 xfs_ail_entry_t *base,
616 if (lip->li_ail.ail_forw == (xfs_log_item_t*)base) {
619 return lip->li_ail.ail_forw;
625 * Check that the list is sorted as it should be.
629 xfs_ail_entry_t *base,
632 xfs_log_item_t *prev_lip;
634 prev_lip = base->ail_forw;
635 if (prev_lip == (xfs_log_item_t*)base) {
637 * Make sure the pointers are correct when the list
640 ASSERT(base->ail_back == (xfs_log_item_t*)base);
645 * Check the next and previous entries are valid.
647 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
648 prev_lip = lip->li_ail.ail_back;
649 if (prev_lip != (xfs_log_item_t*)base) {
650 ASSERT(prev_lip->li_ail.ail_forw == lip);
651 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
653 prev_lip = lip->li_ail.ail_forw;
654 if (prev_lip != (xfs_log_item_t*)base) {
655 ASSERT(prev_lip->li_ail.ail_back == lip);
656 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0);
660 #ifdef XFS_TRANS_DEBUG
662 * Walk the list checking forward and backward pointers,
663 * lsn ordering, and that every entry has the XFS_LI_IN_AIL
664 * flag set. This is really expensive, so only do it when
665 * specifically debugging the transaction subsystem.
667 prev_lip = (xfs_log_item_t*)base;
668 while (lip != (xfs_log_item_t*)base) {
669 if (prev_lip != (xfs_log_item_t*)base) {
670 ASSERT(prev_lip->li_ail.ail_forw == lip);
671 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
673 ASSERT(lip->li_ail.ail_back == prev_lip);
674 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
676 lip = lip->li_ail.ail_forw;
678 ASSERT(lip == (xfs_log_item_t*)base);
679 ASSERT(base->ail_back == prev_lip);
680 #endif /* XFS_TRANS_DEBUG */