blob: 12aff9584e299a2d026a84f1871c669f0fdb9a4c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
David Chinnerc7e8f262008-10-30 17:39:23 +11003 * Copyright (c) 2008 Dave Chinner
Nathan Scott7b718762005-11-02 14:58:39 +11004 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
Nathan Scott7b718762005-11-02 14:58:39 +11006 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * published by the Free Software Foundation.
9 *
Nathan Scott7b718762005-11-02 14:58:39 +110010 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 *
Nathan Scott7b718762005-11-02 14:58:39 +110015 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070018 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_types.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
David Chinnerda353b02007-08-28 14:00:13 +100026#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
28#include "xfs_trans_priv.h"
29#include "xfs_error.h"
30
Dave Chinner0e57f6a2010-12-20 12:02:19 +110031STATIC void xfs_ail_splice(struct xfs_ail *, struct list_head *, xfs_lsn_t);
Dave Chinnereb3efa12010-12-03 16:42:57 +110032STATIC void xfs_ail_delete(struct xfs_ail *, xfs_log_item_t *);
David Chinner82fa9012008-10-30 17:38:26 +110033STATIC xfs_log_item_t * xfs_ail_min(struct xfs_ail *);
34STATIC xfs_log_item_t * xfs_ail_next(struct xfs_ail *, xfs_log_item_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36#ifdef DEBUG
David Chinner82fa9012008-10-30 17:38:26 +110037STATIC void xfs_ail_check(struct xfs_ail *, xfs_log_item_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#else
David Chinnerde08dbc2008-02-05 12:13:38 +110039#define xfs_ail_check(a,l)
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#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
David Chinner5b00f142008-10-30 17:39:00 +110053xfs_trans_ail_tail(
54 struct xfs_ail *ailp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070055{
56 xfs_lsn_t lsn;
57 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
David Chinnerc7e8f262008-10-30 17:39:23 +110059 spin_lock(&ailp->xa_lock);
David Chinner5b00f142008-10-30 17:39:00 +110060 lip = xfs_ail_min(ailp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 if (lip == NULL) {
62 lsn = (xfs_lsn_t)0;
63 } else {
64 lsn = lip->li_lsn;
65 }
David Chinnerc7e8f262008-10-30 17:39:23 +110066 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
68 return lsn;
69}
70
71/*
72 * xfs_trans_push_ail
73 *
David Chinner249a8c12008-02-05 12:13:32 +110074 * 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
76 * threshold_lsn.
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 *
David Chinner249a8c12008-02-05 12:13:32 +110078 * 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
Malcolm Parsons9da096f2009-03-29 09:55:42 +020081 * do its work.
David Chinner249a8c12008-02-05 12:13:32 +110082 *
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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 */
David Chinner249a8c12008-02-05 12:13:32 +110087void
David Chinner783a2f62008-10-30 17:39:58 +110088xfs_trans_ail_push(
89 struct xfs_ail *ailp,
90 xfs_lsn_t threshold_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
David Chinner783a2f62008-10-30 17:39:58 +110092 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
David Chinner783a2f62008-10-30 17:39:58 +110094 lip = xfs_ail_min(ailp);
95 if (lip && !XFS_FORCED_SHUTDOWN(ailp->xa_mount)) {
96 if (XFS_LSN_CMP(threshold_lsn, ailp->xa_target) > 0)
97 xfsaild_wakeup(ailp, threshold_lsn);
David Chinner249a8c12008-02-05 12:13:32 +110098 }
99}
100
101/*
David Chinner27d8d5f2008-10-30 17:38:39 +1100102 * AIL traversal cursor initialisation.
103 *
104 * The cursor keeps track of where our current traversal is up
105 * to by tracking the next ƣtem in the list for us. However, for
106 * this to be safe, removing an object from the AIL needs to invalidate
107 * any cursor that points to it. hence the traversal cursor needs to
108 * be linked to the struct xfs_ail so that deletion can search all the
109 * active cursors for invalidation.
110 *
111 * We don't link the push cursor because it is embedded in the struct
112 * xfs_ail and hence easily findable.
113 */
David Chinner5b00f142008-10-30 17:39:00 +1100114STATIC void
David Chinner27d8d5f2008-10-30 17:38:39 +1100115xfs_trans_ail_cursor_init(
116 struct xfs_ail *ailp,
117 struct xfs_ail_cursor *cur)
118{
119 cur->item = NULL;
120 if (cur == &ailp->xa_cursors)
121 return;
122
123 cur->next = ailp->xa_cursors.next;
124 ailp->xa_cursors.next = cur;
125}
126
127/*
128 * Set the cursor to the next item, because when we look
129 * up the cursor the current item may have been freed.
130 */
131STATIC void
132xfs_trans_ail_cursor_set(
133 struct xfs_ail *ailp,
134 struct xfs_ail_cursor *cur,
135 struct xfs_log_item *lip)
136{
137 if (lip)
138 cur->item = xfs_ail_next(ailp, lip);
139}
140
141/*
142 * Get the next item in the traversal and advance the cursor.
143 * If the cursor was invalidated (inidicated by a lip of 1),
144 * restart the traversal.
145 */
David Chinner5b00f142008-10-30 17:39:00 +1100146struct xfs_log_item *
David Chinner27d8d5f2008-10-30 17:38:39 +1100147xfs_trans_ail_cursor_next(
148 struct xfs_ail *ailp,
149 struct xfs_ail_cursor *cur)
150{
151 struct xfs_log_item *lip = cur->item;
152
153 if ((__psint_t)lip & 1)
154 lip = xfs_ail_min(ailp);
155 xfs_trans_ail_cursor_set(ailp, cur, lip);
156 return lip;
157}
158
159/*
David Chinner27d8d5f2008-10-30 17:38:39 +1100160 * Now that the traversal is complete, we need to remove the cursor
161 * from the list of traversing cursors. Avoid removing the embedded
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200162 * push cursor, but use the fact it is always present to make the
David Chinner27d8d5f2008-10-30 17:38:39 +1100163 * list deletion simple.
164 */
165void
166xfs_trans_ail_cursor_done(
167 struct xfs_ail *ailp,
168 struct xfs_ail_cursor *done)
169{
170 struct xfs_ail_cursor *prev = NULL;
171 struct xfs_ail_cursor *cur;
172
173 done->item = NULL;
174 if (done == &ailp->xa_cursors)
175 return;
176 prev = &ailp->xa_cursors;
177 for (cur = prev->next; cur; prev = cur, cur = prev->next) {
178 if (cur == done) {
179 prev->next = cur->next;
180 break;
181 }
182 }
183 ASSERT(cur);
184}
185
186/*
David Chinner5b00f142008-10-30 17:39:00 +1100187 * Invalidate any cursor that is pointing to this item. This is
188 * called when an item is removed from the AIL. Any cursor pointing
189 * to this object is now invalid and the traversal needs to be
190 * terminated so it doesn't reference a freed object. We set the
191 * cursor item to a value of 1 so we can distinguish between an
192 * invalidation and the end of the list when getting the next item
193 * from the cursor.
194 */
195STATIC void
196xfs_trans_ail_cursor_clear(
197 struct xfs_ail *ailp,
198 struct xfs_log_item *lip)
199{
200 struct xfs_ail_cursor *cur;
201
202 /* need to search all cursors */
203 for (cur = &ailp->xa_cursors; cur; cur = cur->next) {
204 if (cur->item == lip)
205 cur->item = (struct xfs_log_item *)
206 ((__psint_t)cur->item | 1);
207 }
208}
209
210/*
David Chinner249a8c12008-02-05 12:13:32 +1100211 * Return the item in the AIL with the current lsn.
212 * Return the current tree generation number for use
213 * in calls to xfs_trans_next_ail().
214 */
David Chinner5b00f142008-10-30 17:39:00 +1100215xfs_log_item_t *
216xfs_trans_ail_cursor_first(
David Chinner27d8d5f2008-10-30 17:38:39 +1100217 struct xfs_ail *ailp,
218 struct xfs_ail_cursor *cur,
219 xfs_lsn_t lsn)
David Chinner249a8c12008-02-05 12:13:32 +1100220{
David Chinner27d8d5f2008-10-30 17:38:39 +1100221 xfs_log_item_t *lip;
David Chinner249a8c12008-02-05 12:13:32 +1100222
David Chinner5b00f142008-10-30 17:39:00 +1100223 xfs_trans_ail_cursor_init(ailp, cur);
David Chinner27d8d5f2008-10-30 17:38:39 +1100224 lip = xfs_ail_min(ailp);
David Chinner249a8c12008-02-05 12:13:32 +1100225 if (lsn == 0)
David Chinner5b00f142008-10-30 17:39:00 +1100226 goto out;
David Chinner249a8c12008-02-05 12:13:32 +1100227
David Chinner27d8d5f2008-10-30 17:38:39 +1100228 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
David Chinner5b00f142008-10-30 17:39:00 +1100229 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
David Chinner7ee49ac2008-10-30 18:26:51 +1100230 goto out;
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100231 }
David Chinner5b00f142008-10-30 17:39:00 +1100232 lip = NULL;
233out:
234 xfs_trans_ail_cursor_set(ailp, cur, lip);
235 return lip;
David Chinner249a8c12008-02-05 12:13:32 +1100236}
237
238/*
Dave Chinner453eac82010-01-11 11:49:58 +0000239 * xfsaild_push does the work of pushing on the AIL. Returning a timeout of
240 * zero indicates that the caller should sleep until woken.
David Chinner249a8c12008-02-05 12:13:32 +1100241 */
242long
243xfsaild_push(
David Chinner82fa9012008-10-30 17:38:26 +1100244 struct xfs_ail *ailp,
David Chinner249a8c12008-02-05 12:13:32 +1100245 xfs_lsn_t *last_lsn)
246{
Dave Chinner453eac82010-01-11 11:49:58 +0000247 long tout = 0;
David Chinner249a8c12008-02-05 12:13:32 +1100248 xfs_lsn_t last_pushed_lsn = *last_lsn;
David Chinner82fa9012008-10-30 17:38:26 +1100249 xfs_lsn_t target = ailp->xa_target;
David Chinner249a8c12008-02-05 12:13:32 +1100250 xfs_lsn_t lsn;
251 xfs_log_item_t *lip;
David Chinner249a8c12008-02-05 12:13:32 +1100252 int flush_log, count, stuck;
David Chinner82fa9012008-10-30 17:38:26 +1100253 xfs_mount_t *mp = ailp->xa_mount;
David Chinner27d8d5f2008-10-30 17:38:39 +1100254 struct xfs_ail_cursor *cur = &ailp->xa_cursors;
Dave Chinnerd808f612010-02-02 10:13:42 +1100255 int push_xfsbufd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
David Chinnerc7e8f262008-10-30 17:39:23 +1100257 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +1100258 xfs_trans_ail_cursor_init(ailp, cur);
David Chinner5b00f142008-10-30 17:39:00 +1100259 lip = xfs_trans_ail_cursor_first(ailp, cur, *last_lsn);
David Chinner249a8c12008-02-05 12:13:32 +1100260 if (!lip || XFS_FORCED_SHUTDOWN(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 /*
David Chinner249a8c12008-02-05 12:13:32 +1100262 * AIL is empty or our push has reached the end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
David Chinner27d8d5f2008-10-30 17:38:39 +1100264 xfs_trans_ail_cursor_done(ailp, cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100265 spin_unlock(&ailp->xa_lock);
Dave Chinner453eac82010-01-11 11:49:58 +0000266 *last_lsn = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +1100267 return tout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 }
269
270 XFS_STATS_INC(xs_push_ail);
271
272 /*
273 * While the item we are looking at is below the given threshold
David Chinner249a8c12008-02-05 12:13:32 +1100274 * try to flush it out. We'd like not to stop until we've at least
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 * tried to push on everything in the AIL with an LSN less than
David Chinner249a8c12008-02-05 12:13:32 +1100276 * the given threshold.
277 *
278 * However, we will stop after a certain number of pushes and wait
279 * for a reduced timeout to fire before pushing further. This
280 * prevents use from spinning when we can't do anything or there is
281 * lots of contention on the AIL lists.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 */
David Chinner249a8c12008-02-05 12:13:32 +1100283 lsn = lip->li_lsn;
David Chinner27d8d5f2008-10-30 17:38:39 +1100284 flush_log = stuck = count = 0;
David Chinner249a8c12008-02-05 12:13:32 +1100285 while ((XFS_LSN_CMP(lip->li_lsn, target) < 0)) {
286 int lock_result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 /*
David Chinner249a8c12008-02-05 12:13:32 +1100288 * If we can lock the item without sleeping, unlock the AIL
289 * lock and flush the item. Then re-grab the AIL lock so we
290 * can look for the next item on the AIL. List changes are
291 * handled by the AIL lookup functions internally
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 *
David Chinner249a8c12008-02-05 12:13:32 +1100293 * If we can't lock the item, either its holder will flush it
294 * or it is already being flushed or it is being relogged. In
295 * any of these case it is being taken care of and we can just
296 * skip to the next item in the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 */
298 lock_result = IOP_TRYLOCK(lip);
David Chinnerc7e8f262008-10-30 17:39:23 +1100299 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 switch (lock_result) {
David Chinner249a8c12008-02-05 12:13:32 +1100301 case XFS_ITEM_SUCCESS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 XFS_STATS_INC(xs_push_ail_success);
303 IOP_PUSH(lip);
David Chinner249a8c12008-02-05 12:13:32 +1100304 last_pushed_lsn = lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 break;
306
David Chinner249a8c12008-02-05 12:13:32 +1100307 case XFS_ITEM_PUSHBUF:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 XFS_STATS_INC(xs_push_ail_pushbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 IOP_PUSHBUF(lip);
David Chinner249a8c12008-02-05 12:13:32 +1100310 last_pushed_lsn = lsn;
Dave Chinnerd808f612010-02-02 10:13:42 +1100311 push_xfsbufd = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 break;
313
David Chinner249a8c12008-02-05 12:13:32 +1100314 case XFS_ITEM_PINNED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 XFS_STATS_INC(xs_push_ail_pinned);
David Chinner249a8c12008-02-05 12:13:32 +1100316 stuck++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 flush_log = 1;
318 break;
319
David Chinner249a8c12008-02-05 12:13:32 +1100320 case XFS_ITEM_LOCKED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 XFS_STATS_INC(xs_push_ail_locked);
David Chinner249a8c12008-02-05 12:13:32 +1100322 last_pushed_lsn = lsn;
323 stuck++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 break;
325
David Chinner249a8c12008-02-05 12:13:32 +1100326 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 ASSERT(0);
328 break;
329 }
330
David Chinnerc7e8f262008-10-30 17:39:23 +1100331 spin_lock(&ailp->xa_lock);
David Chinner249a8c12008-02-05 12:13:32 +1100332 /* should we bother continuing? */
333 if (XFS_FORCED_SHUTDOWN(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 break;
David Chinner249a8c12008-02-05 12:13:32 +1100335 ASSERT(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
David Chinner249a8c12008-02-05 12:13:32 +1100337 count++;
338
339 /*
340 * Are there too many items we can't do anything with?
341 * If we we are skipping too many items because we can't flush
342 * them or they are already being flushed, we back off and
343 * given them time to complete whatever operation is being
344 * done. i.e. remove pressure from the AIL while we can't make
345 * progress so traversals don't slow down further inserts and
346 * removals to/from the AIL.
347 *
348 * The value of 100 is an arbitrary magic number based on
349 * observation.
350 */
351 if (stuck > 100)
352 break;
353
David Chinner27d8d5f2008-10-30 17:38:39 +1100354 lip = xfs_trans_ail_cursor_next(ailp, cur);
David Chinner249a8c12008-02-05 12:13:32 +1100355 if (lip == NULL)
356 break;
David Chinner249a8c12008-02-05 12:13:32 +1100357 lsn = lip->li_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 }
David Chinner27d8d5f2008-10-30 17:38:39 +1100359 xfs_trans_ail_cursor_done(ailp, cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100360 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 if (flush_log) {
363 /*
364 * If something we need to push out was pinned, then
365 * push out the log so it will become unpinned and
366 * move forward in the AIL.
367 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 XFS_STATS_INC(xs_push_ail_flush);
Christoph Hellwiga14a3482010-01-19 09:56:46 +0000369 xfs_log_force(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 }
371
Dave Chinnerd808f612010-02-02 10:13:42 +1100372 if (push_xfsbufd) {
373 /* we've got delayed write buffers to flush */
374 wake_up_process(mp->m_ddev_targp->bt_task);
375 }
376
David Chinner92d9cd12008-03-06 13:45:10 +1100377 if (!count) {
378 /* We're past our target or empty, so idle */
Dave Chinner453eac82010-01-11 11:49:58 +0000379 last_pushed_lsn = 0;
David Chinner92d9cd12008-03-06 13:45:10 +1100380 } else if (XFS_LSN_CMP(lsn, target) >= 0) {
381 /*
382 * We reached the target so wait a bit longer for I/O to
383 * complete and remove pushed items from the AIL before we
384 * start the next scan from the start of the AIL.
385 */
Dave Chinner453eac82010-01-11 11:49:58 +0000386 tout = 50;
David Chinner249a8c12008-02-05 12:13:32 +1100387 last_pushed_lsn = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +1100388 } else if ((stuck * 100) / count > 90) {
David Chinner249a8c12008-02-05 12:13:32 +1100389 /*
390 * Either there is a lot of contention on the AIL or we
391 * are stuck due to operations in progress. "Stuck" in this
392 * case is defined as >90% of the items we tried to push
393 * were stuck.
394 *
395 * Backoff a bit more to allow some I/O to complete before
396 * continuing from where we were.
397 */
Dave Chinner453eac82010-01-11 11:49:58 +0000398 tout = 20;
399 } else {
400 /* more to do, but wait a short while before continuing */
401 tout = 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 }
David Chinner249a8c12008-02-05 12:13:32 +1100403 *last_lsn = last_pushed_lsn;
404 return tout;
Dave Chinner453eac82010-01-11 11:49:58 +0000405}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407
408/*
409 * This is to be called when an item is unlocked that may have
410 * been in the AIL. It will wake up the first member of the AIL
411 * wait list if this item's unlocking might allow it to progress.
412 * If the item is in the AIL, then we need to get the AIL lock
413 * while doing our checking so we don't race with someone going
414 * to sleep waiting for this event in xfs_trans_push_ail().
415 */
416void
417xfs_trans_unlocked_item(
David Chinner783a2f62008-10-30 17:39:58 +1100418 struct xfs_ail *ailp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 xfs_log_item_t *lip)
420{
421 xfs_log_item_t *min_lip;
422
423 /*
424 * If we're forcibly shutting down, we may have
425 * unlocked log items arbitrarily. The last thing
426 * we want to do is to move the tail of the log
427 * over some potentially valid data.
428 */
429 if (!(lip->li_flags & XFS_LI_IN_AIL) ||
David Chinner783a2f62008-10-30 17:39:58 +1100430 XFS_FORCED_SHUTDOWN(ailp->xa_mount)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 return;
432 }
433
434 /*
435 * This is the one case where we can call into xfs_ail_min()
436 * without holding the AIL lock because we only care about the
437 * case where we are at the tail of the AIL. If the object isn't
438 * at the tail, it doesn't matter what result we get back. This
439 * is slightly racy because since we were just unlocked, we could
440 * go to sleep between the call to xfs_ail_min and the call to
441 * xfs_log_move_tail, have someone else lock us, commit to us disk,
442 * move us out of the tail of the AIL, and then we wake up. However,
443 * the call to xfs_log_move_tail() doesn't do anything if there's
444 * not enough free space to wake people up so we're safe calling it.
445 */
David Chinner783a2f62008-10-30 17:39:58 +1100446 min_lip = xfs_ail_min(ailp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
448 if (min_lip == lip)
David Chinner783a2f62008-10-30 17:39:58 +1100449 xfs_log_move_tail(ailp->xa_mount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450} /* xfs_trans_unlocked_item */
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452/*
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100453 * xfs_trans_ail_update - bulk AIL insertion operation.
454 *
455 * @xfs_trans_ail_update takes an array of log items that all need to be
456 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
457 * be added. Otherwise, it will be repositioned by removing it and re-adding
458 * it to the AIL. If we move the first item in the AIL, update the log tail to
459 * match the new minimum LSN in the AIL.
460 *
461 * This function takes the AIL lock once to execute the update operations on
462 * all the items in the array, and as such should not be called with the AIL
463 * lock held. As a result, once we have the AIL lock, we need to check each log
464 * item LSN to confirm it needs to be moved forward in the AIL.
465 *
466 * To optimise the insert operation, we delete all the items from the AIL in
467 * the first pass, moving them into a temporary list, then splice the temporary
468 * list into the correct position in the AIL. This avoids needing to do an
469 * insert operation on every item.
470 *
471 * This function must be called with the AIL lock held. The lock is dropped
472 * before returning.
473 */
474void
475xfs_trans_ail_update_bulk(
476 struct xfs_ail *ailp,
477 struct xfs_log_item **log_items,
478 int nr_items,
479 xfs_lsn_t lsn) __releases(ailp->xa_lock)
480{
481 xfs_log_item_t *mlip;
482 xfs_lsn_t tail_lsn;
483 int mlip_changed = 0;
484 int i;
485 LIST_HEAD(tmp);
486
487 mlip = xfs_ail_min(ailp);
488
489 for (i = 0; i < nr_items; i++) {
490 struct xfs_log_item *lip = log_items[i];
491 if (lip->li_flags & XFS_LI_IN_AIL) {
492 /* check if we really need to move the item */
493 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
494 continue;
495
496 xfs_ail_delete(ailp, lip);
497 if (mlip == lip)
498 mlip_changed = 1;
499 } else {
500 lip->li_flags |= XFS_LI_IN_AIL;
501 }
502 lip->li_lsn = lsn;
503 list_add(&lip->li_ail, &tmp);
504 }
505
506 xfs_ail_splice(ailp, &tmp, lsn);
507
508 if (!mlip_changed) {
509 spin_unlock(&ailp->xa_lock);
510 return;
511 }
512
513 /*
514 * It is not safe to access mlip after the AIL lock is dropped, so we
515 * must get a copy of li_lsn before we do so. This is especially
516 * important on 32-bit platforms where accessing and updating 64-bit
517 * values like li_lsn is not atomic.
518 */
519 mlip = xfs_ail_min(ailp);
520 tail_lsn = mlip->li_lsn;
521 spin_unlock(&ailp->xa_lock);
522 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
523}
524
525/*
Dave Chinner30136832010-12-20 12:03:17 +1100526 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
527 *
528 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
529 * removed from the AIL. The caller is already holding the AIL lock, and done
530 * all the checks necessary to ensure the items passed in via @log_items are
531 * ready for deletion. This includes checking that the items are in the AIL.
532 *
533 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
534 * flag from the item and reset the item's lsn to 0. If we remove the first
535 * item in the AIL, update the log tail to match the new minimum LSN in the
536 * AIL.
537 *
538 * This function will not drop the AIL lock until all items are removed from
539 * the AIL to minimise the amount of lock traffic on the AIL. This does not
540 * greatly increase the AIL hold time, but does significantly reduce the amount
541 * of traffic on the lock, especially during IO completion.
542 *
543 * This function must be called with the AIL lock held. The lock is dropped
544 * before returning.
545 */
546void
547xfs_trans_ail_delete_bulk(
548 struct xfs_ail *ailp,
549 struct xfs_log_item **log_items,
550 int nr_items) __releases(ailp->xa_lock)
551{
552 xfs_log_item_t *mlip;
553 xfs_lsn_t tail_lsn;
554 int mlip_changed = 0;
555 int i;
556
557 mlip = xfs_ail_min(ailp);
558
559 for (i = 0; i < nr_items; i++) {
560 struct xfs_log_item *lip = log_items[i];
561 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
562 struct xfs_mount *mp = ailp->xa_mount;
563
564 spin_unlock(&ailp->xa_lock);
565 if (!XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner6a19d932011-03-07 10:02:35 +1100566 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
Dave Chinner30136832010-12-20 12:03:17 +1100567 "%s: attempting to delete a log item that is not in the AIL",
568 __func__);
569 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
570 }
571 return;
572 }
573
574 xfs_ail_delete(ailp, lip);
575 lip->li_flags &= ~XFS_LI_IN_AIL;
576 lip->li_lsn = 0;
577 if (mlip == lip)
578 mlip_changed = 1;
579 }
580
581 if (!mlip_changed) {
582 spin_unlock(&ailp->xa_lock);
583 return;
584 }
585
586 /*
587 * It is not safe to access mlip after the AIL lock is dropped, so we
588 * must get a copy of li_lsn before we do so. This is especially
589 * important on 32-bit platforms where accessing and updating 64-bit
590 * values like li_lsn is not atomic. It is possible we've emptied the
591 * AIL here, so if that is the case, pass an LSN of 0 to the tail move.
592 */
593 mlip = xfs_ail_min(ailp);
594 tail_lsn = mlip ? mlip->li_lsn : 0;
595 spin_unlock(&ailp->xa_lock);
596 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
597}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 * The active item list (AIL) is a doubly linked list of log
601 * items sorted by ascending lsn. The base of the list is
602 * a forw/back pointer pair embedded in the xfs mount structure.
603 * The base is initialized with both pointers pointing to the
604 * base. This case always needs to be distinguished, because
605 * the base has no lsn to look at. We almost always insert
606 * at the end of the list, so on inserts we search from the
607 * end of the list to find where the new item belongs.
608 */
609
610/*
611 * Initialize the doubly linked list to point only to itself.
612 */
David Chinner249a8c12008-02-05 12:13:32 +1100613int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614xfs_trans_ail_init(
615 xfs_mount_t *mp)
616{
David Chinner82fa9012008-10-30 17:38:26 +1100617 struct xfs_ail *ailp;
David Chinner27d8d5f2008-10-30 17:38:39 +1100618 int error;
David Chinner82fa9012008-10-30 17:38:26 +1100619
620 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
621 if (!ailp)
622 return ENOMEM;
623
624 ailp->xa_mount = mp;
625 INIT_LIST_HEAD(&ailp->xa_ail);
David Chinnerc7e8f262008-10-30 17:39:23 +1100626 spin_lock_init(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +1100627 error = xfsaild_start(ailp);
628 if (error)
629 goto out_free_ailp;
630 mp->m_ail = ailp;
631 return 0;
632
633out_free_ailp:
634 kmem_free(ailp);
635 return error;
David Chinner249a8c12008-02-05 12:13:32 +1100636}
637
638void
639xfs_trans_ail_destroy(
640 xfs_mount_t *mp)
641{
David Chinner82fa9012008-10-30 17:38:26 +1100642 struct xfs_ail *ailp = mp->m_ail;
643
644 xfsaild_stop(ailp);
645 kmem_free(ailp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
648/*
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100649 * splice the log item list into the AIL at the given LSN.
650 */
651STATIC void
652xfs_ail_splice(
653 struct xfs_ail *ailp,
654 struct list_head *list,
655 xfs_lsn_t lsn)
656{
657 xfs_log_item_t *next_lip;
658
659 /*
660 * If the list is empty, just insert the item.
661 */
662 if (list_empty(&ailp->xa_ail)) {
663 list_splice(list, &ailp->xa_ail);
664 return;
665 }
666
667 list_for_each_entry_reverse(next_lip, &ailp->xa_ail, li_ail) {
668 if (XFS_LSN_CMP(next_lip->li_lsn, lsn) <= 0)
669 break;
670 }
671
672 ASSERT((&next_lip->li_ail == &ailp->xa_ail) ||
673 (XFS_LSN_CMP(next_lip->li_lsn, lsn) <= 0));
674
675 list_splice_init(list, &next_lip->li_ail);
676 return;
677}
678
679/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 * Delete the given item from the AIL. Return a pointer to the item.
681 */
Dave Chinnereb3efa12010-12-03 16:42:57 +1100682STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683xfs_ail_delete(
David Chinner82fa9012008-10-30 17:38:26 +1100684 struct xfs_ail *ailp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 xfs_log_item_t *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100687 xfs_ail_check(ailp, lip);
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100688 list_del(&lip->li_ail);
Dave Chinnereb3efa12010-12-03 16:42:57 +1100689 xfs_trans_ail_cursor_clear(ailp, lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
692/*
693 * Return a pointer to the first item in the AIL.
694 * If the AIL is empty, then return NULL.
695 */
696STATIC xfs_log_item_t *
697xfs_ail_min(
David Chinner82fa9012008-10-30 17:38:26 +1100698 struct xfs_ail *ailp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100700 if (list_empty(&ailp->xa_ail))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 return NULL;
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100702
703 return list_first_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
705
706/*
707 * Return a pointer to the item which follows
708 * the given item in the AIL. If the given item
709 * is the last item in the list, then return NULL.
710 */
711STATIC xfs_log_item_t *
712xfs_ail_next(
David Chinner82fa9012008-10-30 17:38:26 +1100713 struct xfs_ail *ailp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 xfs_log_item_t *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100716 if (lip->li_ail.next == &ailp->xa_ail)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100719 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720}
721
722#ifdef DEBUG
723/*
724 * Check that the list is sorted as it should be.
725 */
726STATIC void
727xfs_ail_check(
David Chinner82fa9012008-10-30 17:38:26 +1100728 struct xfs_ail *ailp,
David Chinnerde08dbc2008-02-05 12:13:38 +1100729 xfs_log_item_t *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 xfs_log_item_t *prev_lip;
732
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100733 if (list_empty(&ailp->xa_ail))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
736 /*
David Chinnerde08dbc2008-02-05 12:13:38 +1100737 * Check the next and previous entries are valid.
738 */
739 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100740 prev_lip = list_entry(lip->li_ail.prev, xfs_log_item_t, li_ail);
741 if (&prev_lip->li_ail != &ailp->xa_ail)
David Chinnerde08dbc2008-02-05 12:13:38 +1100742 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100743
744 prev_lip = list_entry(lip->li_ail.next, xfs_log_item_t, li_ail);
745 if (&prev_lip->li_ail != &ailp->xa_ail)
David Chinnerde08dbc2008-02-05 12:13:38 +1100746 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0);
David Chinnerde08dbc2008-02-05 12:13:38 +1100747
748
749#ifdef XFS_TRANS_DEBUG
750 /*
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100751 * Walk the list checking lsn ordering, and that every entry has the
752 * XFS_LI_IN_AIL flag set. This is really expensive, so only do it
753 * when specifically debugging the transaction subsystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 */
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100755 prev_lip = list_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
756 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
757 if (&prev_lip->li_ail != &ailp->xa_ail)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
760 prev_lip = lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 }
David Chinnerde08dbc2008-02-05 12:13:38 +1100762#endif /* XFS_TRANS_DEBUG */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763}
764#endif /* DEBUG */