blob: 8012bfbc6dc0c7891d9a2e7330cc595306217a6a [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 Chinner0bf6a5b2011-04-08 12:45:07 +100031struct workqueue_struct *xfs_ail_wq; /* AIL workqueue */
32
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#ifdef DEBUG
Dave Chinnercd4a3c52011-04-08 12:45:07 +100034/*
35 * Check that the list is sorted as it should be.
36 */
37STATIC void
38xfs_ail_check(
39 struct xfs_ail *ailp,
40 xfs_log_item_t *lip)
41{
42 xfs_log_item_t *prev_lip;
43
44 if (list_empty(&ailp->xa_ail))
45 return;
46
47 /*
48 * Check the next and previous entries are valid.
49 */
50 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
51 prev_lip = list_entry(lip->li_ail.prev, xfs_log_item_t, li_ail);
52 if (&prev_lip->li_ail != &ailp->xa_ail)
53 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
54
55 prev_lip = list_entry(lip->li_ail.next, xfs_log_item_t, li_ail);
56 if (&prev_lip->li_ail != &ailp->xa_ail)
57 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0);
58
59
60#ifdef XFS_TRANS_DEBUG
61 /*
62 * Walk the list checking lsn ordering, and that every entry has the
63 * XFS_LI_IN_AIL flag set. This is really expensive, so only do it
64 * when specifically debugging the transaction subsystem.
65 */
66 prev_lip = list_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
67 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
68 if (&prev_lip->li_ail != &ailp->xa_ail)
69 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
70 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
71 prev_lip = lip;
72 }
73#endif /* XFS_TRANS_DEBUG */
74}
75#else /* !DEBUG */
David Chinnerde08dbc2008-02-05 12:13:38 +110076#define xfs_ail_check(a,l)
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#endif /* DEBUG */
78
Dave Chinnercd4a3c52011-04-08 12:45:07 +100079/*
80 * Return a pointer to the first item in the AIL. If the AIL is empty, then
81 * return NULL.
82 */
83static xfs_log_item_t *
84xfs_ail_min(
85 struct xfs_ail *ailp)
86{
87 if (list_empty(&ailp->xa_ail))
88 return NULL;
89
90 return list_first_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
91}
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
93/*
Dave Chinnercd4a3c52011-04-08 12:45:07 +100094 * Return a pointer to the item which follows the given item in the AIL. If
95 * the given item is the last item in the list, then return NULL.
96 */
97static xfs_log_item_t *
98xfs_ail_next(
99 struct xfs_ail *ailp,
100 xfs_log_item_t *lip)
101{
102 if (lip->li_ail.next == &ailp->xa_ail)
103 return NULL;
104
105 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
106}
107
108/*
109 * This is called by the log manager code to determine the LSN of the tail of
110 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
111 * is empty, then this function returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 *
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000113 * We need the AIL lock in order to get a coherent read of the lsn of the last
114 * item in the AIL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 */
116xfs_lsn_t
David Chinner5b00f142008-10-30 17:39:00 +1100117xfs_trans_ail_tail(
118 struct xfs_ail *ailp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000120 xfs_lsn_t lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
David Chinnerc7e8f262008-10-30 17:39:23 +1100123 spin_lock(&ailp->xa_lock);
David Chinner5b00f142008-10-30 17:39:00 +1100124 lip = xfs_ail_min(ailp);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000125 if (lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 lsn = lip->li_lsn;
David Chinnerc7e8f262008-10-30 17:39:23 +1100127 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129 return lsn;
130}
131
132/*
David Chinner27d8d5f2008-10-30 17:38:39 +1100133 * AIL traversal cursor initialisation.
134 *
135 * The cursor keeps track of where our current traversal is up
136 * to by tracking the next ƣtem in the list for us. However, for
137 * this to be safe, removing an object from the AIL needs to invalidate
138 * any cursor that points to it. hence the traversal cursor needs to
139 * be linked to the struct xfs_ail so that deletion can search all the
140 * active cursors for invalidation.
141 *
142 * We don't link the push cursor because it is embedded in the struct
143 * xfs_ail and hence easily findable.
144 */
David Chinner5b00f142008-10-30 17:39:00 +1100145STATIC void
David Chinner27d8d5f2008-10-30 17:38:39 +1100146xfs_trans_ail_cursor_init(
147 struct xfs_ail *ailp,
148 struct xfs_ail_cursor *cur)
149{
150 cur->item = NULL;
151 if (cur == &ailp->xa_cursors)
152 return;
153
154 cur->next = ailp->xa_cursors.next;
155 ailp->xa_cursors.next = cur;
156}
157
158/*
159 * Set the cursor to the next item, because when we look
160 * up the cursor the current item may have been freed.
161 */
162STATIC void
163xfs_trans_ail_cursor_set(
164 struct xfs_ail *ailp,
165 struct xfs_ail_cursor *cur,
166 struct xfs_log_item *lip)
167{
168 if (lip)
169 cur->item = xfs_ail_next(ailp, lip);
170}
171
172/*
173 * Get the next item in the traversal and advance the cursor.
174 * If the cursor was invalidated (inidicated by a lip of 1),
175 * restart the traversal.
176 */
David Chinner5b00f142008-10-30 17:39:00 +1100177struct xfs_log_item *
David Chinner27d8d5f2008-10-30 17:38:39 +1100178xfs_trans_ail_cursor_next(
179 struct xfs_ail *ailp,
180 struct xfs_ail_cursor *cur)
181{
182 struct xfs_log_item *lip = cur->item;
183
184 if ((__psint_t)lip & 1)
185 lip = xfs_ail_min(ailp);
186 xfs_trans_ail_cursor_set(ailp, cur, lip);
187 return lip;
188}
189
190/*
David Chinner27d8d5f2008-10-30 17:38:39 +1100191 * Now that the traversal is complete, we need to remove the cursor
192 * from the list of traversing cursors. Avoid removing the embedded
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200193 * push cursor, but use the fact it is always present to make the
David Chinner27d8d5f2008-10-30 17:38:39 +1100194 * list deletion simple.
195 */
196void
197xfs_trans_ail_cursor_done(
198 struct xfs_ail *ailp,
199 struct xfs_ail_cursor *done)
200{
201 struct xfs_ail_cursor *prev = NULL;
202 struct xfs_ail_cursor *cur;
203
204 done->item = NULL;
205 if (done == &ailp->xa_cursors)
206 return;
207 prev = &ailp->xa_cursors;
208 for (cur = prev->next; cur; prev = cur, cur = prev->next) {
209 if (cur == done) {
210 prev->next = cur->next;
211 break;
212 }
213 }
214 ASSERT(cur);
215}
216
217/*
David Chinner5b00f142008-10-30 17:39:00 +1100218 * Invalidate any cursor that is pointing to this item. This is
219 * called when an item is removed from the AIL. Any cursor pointing
220 * to this object is now invalid and the traversal needs to be
221 * terminated so it doesn't reference a freed object. We set the
222 * cursor item to a value of 1 so we can distinguish between an
223 * invalidation and the end of the list when getting the next item
224 * from the cursor.
225 */
226STATIC void
227xfs_trans_ail_cursor_clear(
228 struct xfs_ail *ailp,
229 struct xfs_log_item *lip)
230{
231 struct xfs_ail_cursor *cur;
232
233 /* need to search all cursors */
234 for (cur = &ailp->xa_cursors; cur; cur = cur->next) {
235 if (cur->item == lip)
236 cur->item = (struct xfs_log_item *)
237 ((__psint_t)cur->item | 1);
238 }
239}
240
241/*
David Chinner249a8c12008-02-05 12:13:32 +1100242 * Return the item in the AIL with the current lsn.
243 * Return the current tree generation number for use
244 * in calls to xfs_trans_next_ail().
245 */
David Chinner5b00f142008-10-30 17:39:00 +1100246xfs_log_item_t *
247xfs_trans_ail_cursor_first(
David Chinner27d8d5f2008-10-30 17:38:39 +1100248 struct xfs_ail *ailp,
249 struct xfs_ail_cursor *cur,
250 xfs_lsn_t lsn)
David Chinner249a8c12008-02-05 12:13:32 +1100251{
David Chinner27d8d5f2008-10-30 17:38:39 +1100252 xfs_log_item_t *lip;
David Chinner249a8c12008-02-05 12:13:32 +1100253
David Chinner5b00f142008-10-30 17:39:00 +1100254 xfs_trans_ail_cursor_init(ailp, cur);
David Chinner27d8d5f2008-10-30 17:38:39 +1100255 lip = xfs_ail_min(ailp);
David Chinner249a8c12008-02-05 12:13:32 +1100256 if (lsn == 0)
David Chinner5b00f142008-10-30 17:39:00 +1100257 goto out;
David Chinner249a8c12008-02-05 12:13:32 +1100258
David Chinner27d8d5f2008-10-30 17:38:39 +1100259 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
David Chinner5b00f142008-10-30 17:39:00 +1100260 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
David Chinner7ee49ac2008-10-30 18:26:51 +1100261 goto out;
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100262 }
David Chinner5b00f142008-10-30 17:39:00 +1100263 lip = NULL;
264out:
265 xfs_trans_ail_cursor_set(ailp, cur, lip);
266 return lip;
David Chinner249a8c12008-02-05 12:13:32 +1100267}
268
269/*
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000270 * splice the log item list into the AIL at the given LSN.
271 */
272static void
273xfs_ail_splice(
274 struct xfs_ail *ailp,
275 struct list_head *list,
276 xfs_lsn_t lsn)
277{
278 xfs_log_item_t *next_lip;
279
280 /* If the list is empty, just insert the item. */
281 if (list_empty(&ailp->xa_ail)) {
282 list_splice(list, &ailp->xa_ail);
283 return;
284 }
285
286 list_for_each_entry_reverse(next_lip, &ailp->xa_ail, li_ail) {
287 if (XFS_LSN_CMP(next_lip->li_lsn, lsn) <= 0)
288 break;
289 }
290
291 ASSERT(&next_lip->li_ail == &ailp->xa_ail ||
292 XFS_LSN_CMP(next_lip->li_lsn, lsn) <= 0);
293
294 list_splice_init(list, &next_lip->li_ail);
295}
296
297/*
298 * Delete the given item from the AIL. Return a pointer to the item.
299 */
300static void
301xfs_ail_delete(
302 struct xfs_ail *ailp,
303 xfs_log_item_t *lip)
304{
305 xfs_ail_check(ailp, lip);
306 list_del(&lip->li_ail);
307 xfs_trans_ail_cursor_clear(ailp, lip);
308}
309
310/*
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000311 * xfs_ail_worker does the work of pushing on the AIL. It will requeue itself
312 * to run at a later time if there is more work to do to complete the push.
David Chinner249a8c12008-02-05 12:13:32 +1100313 */
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000314STATIC void
315xfs_ail_worker(
316 struct work_struct *work)
David Chinner249a8c12008-02-05 12:13:32 +1100317{
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000318 struct xfs_ail *ailp = container_of(to_delayed_work(work),
319 struct xfs_ail, xa_work);
320 long tout;
David Chinner82fa9012008-10-30 17:38:26 +1100321 xfs_lsn_t target = ailp->xa_target;
David Chinner249a8c12008-02-05 12:13:32 +1100322 xfs_lsn_t lsn;
323 xfs_log_item_t *lip;
David Chinner249a8c12008-02-05 12:13:32 +1100324 int flush_log, count, stuck;
David Chinner82fa9012008-10-30 17:38:26 +1100325 xfs_mount_t *mp = ailp->xa_mount;
David Chinner27d8d5f2008-10-30 17:38:39 +1100326 struct xfs_ail_cursor *cur = &ailp->xa_cursors;
Dave Chinnerd808f612010-02-02 10:13:42 +1100327 int push_xfsbufd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
David Chinnerc7e8f262008-10-30 17:39:23 +1100329 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +1100330 xfs_trans_ail_cursor_init(ailp, cur);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000331 lip = xfs_trans_ail_cursor_first(ailp, cur, ailp->xa_last_pushed_lsn);
David Chinner249a8c12008-02-05 12:13:32 +1100332 if (!lip || XFS_FORCED_SHUTDOWN(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 /*
David Chinner249a8c12008-02-05 12:13:32 +1100334 * AIL is empty or our push has reached the end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 */
David Chinner27d8d5f2008-10-30 17:38:39 +1100336 xfs_trans_ail_cursor_done(ailp, cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100337 spin_unlock(&ailp->xa_lock);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000338 ailp->xa_last_pushed_lsn = 0;
339 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 }
341
342 XFS_STATS_INC(xs_push_ail);
343
344 /*
345 * While the item we are looking at is below the given threshold
David Chinner249a8c12008-02-05 12:13:32 +1100346 * try to flush it out. We'd like not to stop until we've at least
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 * tried to push on everything in the AIL with an LSN less than
David Chinner249a8c12008-02-05 12:13:32 +1100348 * the given threshold.
349 *
350 * However, we will stop after a certain number of pushes and wait
351 * for a reduced timeout to fire before pushing further. This
352 * prevents use from spinning when we can't do anything or there is
353 * lots of contention on the AIL lists.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 */
David Chinner249a8c12008-02-05 12:13:32 +1100355 lsn = lip->li_lsn;
David Chinner27d8d5f2008-10-30 17:38:39 +1100356 flush_log = stuck = count = 0;
David Chinner249a8c12008-02-05 12:13:32 +1100357 while ((XFS_LSN_CMP(lip->li_lsn, target) < 0)) {
358 int lock_result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 /*
David Chinner249a8c12008-02-05 12:13:32 +1100360 * If we can lock the item without sleeping, unlock the AIL
361 * lock and flush the item. Then re-grab the AIL lock so we
362 * can look for the next item on the AIL. List changes are
363 * handled by the AIL lookup functions internally
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 *
David Chinner249a8c12008-02-05 12:13:32 +1100365 * If we can't lock the item, either its holder will flush it
366 * or it is already being flushed or it is being relogged. In
367 * any of these case it is being taken care of and we can just
368 * skip to the next item in the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 */
370 lock_result = IOP_TRYLOCK(lip);
David Chinnerc7e8f262008-10-30 17:39:23 +1100371 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 switch (lock_result) {
David Chinner249a8c12008-02-05 12:13:32 +1100373 case XFS_ITEM_SUCCESS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 XFS_STATS_INC(xs_push_ail_success);
375 IOP_PUSH(lip);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000376 ailp->xa_last_pushed_lsn = lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 break;
378
David Chinner249a8c12008-02-05 12:13:32 +1100379 case XFS_ITEM_PUSHBUF:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 XFS_STATS_INC(xs_push_ail_pushbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 IOP_PUSHBUF(lip);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000382 ailp->xa_last_pushed_lsn = lsn;
Dave Chinnerd808f612010-02-02 10:13:42 +1100383 push_xfsbufd = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 break;
385
David Chinner249a8c12008-02-05 12:13:32 +1100386 case XFS_ITEM_PINNED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 XFS_STATS_INC(xs_push_ail_pinned);
David Chinner249a8c12008-02-05 12:13:32 +1100388 stuck++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 flush_log = 1;
390 break;
391
David Chinner249a8c12008-02-05 12:13:32 +1100392 case XFS_ITEM_LOCKED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 XFS_STATS_INC(xs_push_ail_locked);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000394 ailp->xa_last_pushed_lsn = lsn;
David Chinner249a8c12008-02-05 12:13:32 +1100395 stuck++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 break;
397
David Chinner249a8c12008-02-05 12:13:32 +1100398 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 ASSERT(0);
400 break;
401 }
402
David Chinnerc7e8f262008-10-30 17:39:23 +1100403 spin_lock(&ailp->xa_lock);
David Chinner249a8c12008-02-05 12:13:32 +1100404 /* should we bother continuing? */
405 if (XFS_FORCED_SHUTDOWN(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 break;
David Chinner249a8c12008-02-05 12:13:32 +1100407 ASSERT(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
David Chinner249a8c12008-02-05 12:13:32 +1100409 count++;
410
411 /*
412 * Are there too many items we can't do anything with?
413 * If we we are skipping too many items because we can't flush
414 * them or they are already being flushed, we back off and
415 * given them time to complete whatever operation is being
416 * done. i.e. remove pressure from the AIL while we can't make
417 * progress so traversals don't slow down further inserts and
418 * removals to/from the AIL.
419 *
420 * The value of 100 is an arbitrary magic number based on
421 * observation.
422 */
423 if (stuck > 100)
424 break;
425
David Chinner27d8d5f2008-10-30 17:38:39 +1100426 lip = xfs_trans_ail_cursor_next(ailp, cur);
David Chinner249a8c12008-02-05 12:13:32 +1100427 if (lip == NULL)
428 break;
David Chinner249a8c12008-02-05 12:13:32 +1100429 lsn = lip->li_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 }
David Chinner27d8d5f2008-10-30 17:38:39 +1100431 xfs_trans_ail_cursor_done(ailp, cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100432 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
434 if (flush_log) {
435 /*
436 * If something we need to push out was pinned, then
437 * push out the log so it will become unpinned and
438 * move forward in the AIL.
439 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 XFS_STATS_INC(xs_push_ail_flush);
Christoph Hellwiga14a3482010-01-19 09:56:46 +0000441 xfs_log_force(mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 }
443
Dave Chinnerd808f612010-02-02 10:13:42 +1100444 if (push_xfsbufd) {
445 /* we've got delayed write buffers to flush */
446 wake_up_process(mp->m_ddev_targp->bt_task);
447 }
448
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000449 /* assume we have more work to do in a short while */
450 tout = 10;
David Chinner92d9cd12008-03-06 13:45:10 +1100451 if (!count) {
452 /* We're past our target or empty, so idle */
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000453 ailp->xa_last_pushed_lsn = 0;
454
455 /*
456 * Check for an updated push target before clearing the
457 * XFS_AIL_PUSHING_BIT. If the target changed, we've got more
458 * work to do. Wait a bit longer before starting that work.
459 */
460 smp_rmb();
461 if (ailp->xa_target == target) {
462 clear_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags);
463 return;
464 }
465 tout = 50;
David Chinner92d9cd12008-03-06 13:45:10 +1100466 } else if (XFS_LSN_CMP(lsn, target) >= 0) {
467 /*
468 * We reached the target so wait a bit longer for I/O to
469 * complete and remove pushed items from the AIL before we
470 * start the next scan from the start of the AIL.
471 */
Dave Chinner453eac82010-01-11 11:49:58 +0000472 tout = 50;
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000473 ailp->xa_last_pushed_lsn = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +1100474 } else if ((stuck * 100) / count > 90) {
David Chinner249a8c12008-02-05 12:13:32 +1100475 /*
476 * Either there is a lot of contention on the AIL or we
477 * are stuck due to operations in progress. "Stuck" in this
478 * case is defined as >90% of the items we tried to push
479 * were stuck.
480 *
481 * Backoff a bit more to allow some I/O to complete before
482 * continuing from where we were.
483 */
Dave Chinner453eac82010-01-11 11:49:58 +0000484 tout = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 }
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000486
487 /* There is more to do, requeue us. */
488 queue_delayed_work(xfs_syncd_wq, &ailp->xa_work,
489 msecs_to_jiffies(tout));
Dave Chinner453eac82010-01-11 11:49:58 +0000490}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000492/*
493 * This routine is called to move the tail of the AIL forward. It does this by
494 * trying to flush items in the AIL whose lsns are below the given
495 * threshold_lsn.
496 *
497 * The push is run asynchronously in a workqueue, which means the caller needs
498 * to handle waiting on the async flush for space to become available.
499 * We don't want to interrupt any push that is in progress, hence we only queue
500 * work if we set the pushing bit approriately.
501 *
502 * We do this unlocked - we only need to know whether there is anything in the
503 * AIL at the time we are called. We don't need to access the contents of
504 * any of the objects, so the lock is not needed.
505 */
506void
507xfs_trans_ail_push(
508 struct xfs_ail *ailp,
509 xfs_lsn_t threshold_lsn)
510{
511 xfs_log_item_t *lip;
512
513 lip = xfs_ail_min(ailp);
514 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
515 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
516 return;
517
518 /*
519 * Ensure that the new target is noticed in push code before it clears
520 * the XFS_AIL_PUSHING_BIT.
521 */
522 smp_wmb();
523 ailp->xa_target = threshold_lsn;
524 if (!test_and_set_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags))
525 queue_delayed_work(xfs_syncd_wq, &ailp->xa_work, 0);
526}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528/*
529 * This is to be called when an item is unlocked that may have
530 * been in the AIL. It will wake up the first member of the AIL
531 * wait list if this item's unlocking might allow it to progress.
532 * If the item is in the AIL, then we need to get the AIL lock
533 * while doing our checking so we don't race with someone going
534 * to sleep waiting for this event in xfs_trans_push_ail().
535 */
536void
537xfs_trans_unlocked_item(
David Chinner783a2f62008-10-30 17:39:58 +1100538 struct xfs_ail *ailp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 xfs_log_item_t *lip)
540{
541 xfs_log_item_t *min_lip;
542
543 /*
544 * If we're forcibly shutting down, we may have
545 * unlocked log items arbitrarily. The last thing
546 * we want to do is to move the tail of the log
547 * over some potentially valid data.
548 */
549 if (!(lip->li_flags & XFS_LI_IN_AIL) ||
David Chinner783a2f62008-10-30 17:39:58 +1100550 XFS_FORCED_SHUTDOWN(ailp->xa_mount)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 return;
552 }
553
554 /*
555 * This is the one case where we can call into xfs_ail_min()
556 * without holding the AIL lock because we only care about the
557 * case where we are at the tail of the AIL. If the object isn't
558 * at the tail, it doesn't matter what result we get back. This
559 * is slightly racy because since we were just unlocked, we could
560 * go to sleep between the call to xfs_ail_min and the call to
561 * xfs_log_move_tail, have someone else lock us, commit to us disk,
562 * move us out of the tail of the AIL, and then we wake up. However,
563 * the call to xfs_log_move_tail() doesn't do anything if there's
564 * not enough free space to wake people up so we're safe calling it.
565 */
David Chinner783a2f62008-10-30 17:39:58 +1100566 min_lip = xfs_ail_min(ailp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
568 if (min_lip == lip)
David Chinner783a2f62008-10-30 17:39:58 +1100569 xfs_log_move_tail(ailp->xa_mount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570} /* xfs_trans_unlocked_item */
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572/*
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100573 * xfs_trans_ail_update - bulk AIL insertion operation.
574 *
575 * @xfs_trans_ail_update takes an array of log items that all need to be
576 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
577 * be added. Otherwise, it will be repositioned by removing it and re-adding
578 * it to the AIL. If we move the first item in the AIL, update the log tail to
579 * match the new minimum LSN in the AIL.
580 *
581 * This function takes the AIL lock once to execute the update operations on
582 * all the items in the array, and as such should not be called with the AIL
583 * lock held. As a result, once we have the AIL lock, we need to check each log
584 * item LSN to confirm it needs to be moved forward in the AIL.
585 *
586 * To optimise the insert operation, we delete all the items from the AIL in
587 * the first pass, moving them into a temporary list, then splice the temporary
588 * list into the correct position in the AIL. This avoids needing to do an
589 * insert operation on every item.
590 *
591 * This function must be called with the AIL lock held. The lock is dropped
592 * before returning.
593 */
594void
595xfs_trans_ail_update_bulk(
596 struct xfs_ail *ailp,
597 struct xfs_log_item **log_items,
598 int nr_items,
599 xfs_lsn_t lsn) __releases(ailp->xa_lock)
600{
601 xfs_log_item_t *mlip;
602 xfs_lsn_t tail_lsn;
603 int mlip_changed = 0;
604 int i;
605 LIST_HEAD(tmp);
606
607 mlip = xfs_ail_min(ailp);
608
609 for (i = 0; i < nr_items; i++) {
610 struct xfs_log_item *lip = log_items[i];
611 if (lip->li_flags & XFS_LI_IN_AIL) {
612 /* check if we really need to move the item */
613 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
614 continue;
615
616 xfs_ail_delete(ailp, lip);
617 if (mlip == lip)
618 mlip_changed = 1;
619 } else {
620 lip->li_flags |= XFS_LI_IN_AIL;
621 }
622 lip->li_lsn = lsn;
623 list_add(&lip->li_ail, &tmp);
624 }
625
626 xfs_ail_splice(ailp, &tmp, lsn);
627
628 if (!mlip_changed) {
629 spin_unlock(&ailp->xa_lock);
630 return;
631 }
632
633 /*
634 * It is not safe to access mlip after the AIL lock is dropped, so we
635 * must get a copy of li_lsn before we do so. This is especially
636 * important on 32-bit platforms where accessing and updating 64-bit
637 * values like li_lsn is not atomic.
638 */
639 mlip = xfs_ail_min(ailp);
640 tail_lsn = mlip->li_lsn;
641 spin_unlock(&ailp->xa_lock);
642 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
643}
644
645/*
Dave Chinner30136832010-12-20 12:03:17 +1100646 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
647 *
648 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
649 * removed from the AIL. The caller is already holding the AIL lock, and done
650 * all the checks necessary to ensure the items passed in via @log_items are
651 * ready for deletion. This includes checking that the items are in the AIL.
652 *
653 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
654 * flag from the item and reset the item's lsn to 0. If we remove the first
655 * item in the AIL, update the log tail to match the new minimum LSN in the
656 * AIL.
657 *
658 * This function will not drop the AIL lock until all items are removed from
659 * the AIL to minimise the amount of lock traffic on the AIL. This does not
660 * greatly increase the AIL hold time, but does significantly reduce the amount
661 * of traffic on the lock, especially during IO completion.
662 *
663 * This function must be called with the AIL lock held. The lock is dropped
664 * before returning.
665 */
666void
667xfs_trans_ail_delete_bulk(
668 struct xfs_ail *ailp,
669 struct xfs_log_item **log_items,
670 int nr_items) __releases(ailp->xa_lock)
671{
672 xfs_log_item_t *mlip;
673 xfs_lsn_t tail_lsn;
674 int mlip_changed = 0;
675 int i;
676
677 mlip = xfs_ail_min(ailp);
678
679 for (i = 0; i < nr_items; i++) {
680 struct xfs_log_item *lip = log_items[i];
681 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
682 struct xfs_mount *mp = ailp->xa_mount;
683
684 spin_unlock(&ailp->xa_lock);
685 if (!XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner6a19d932011-03-07 10:02:35 +1100686 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
Dave Chinner30136832010-12-20 12:03:17 +1100687 "%s: attempting to delete a log item that is not in the AIL",
688 __func__);
689 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
690 }
691 return;
692 }
693
694 xfs_ail_delete(ailp, lip);
695 lip->li_flags &= ~XFS_LI_IN_AIL;
696 lip->li_lsn = 0;
697 if (mlip == lip)
698 mlip_changed = 1;
699 }
700
701 if (!mlip_changed) {
702 spin_unlock(&ailp->xa_lock);
703 return;
704 }
705
706 /*
707 * It is not safe to access mlip after the AIL lock is dropped, so we
708 * must get a copy of li_lsn before we do so. This is especially
709 * important on 32-bit platforms where accessing and updating 64-bit
710 * values like li_lsn is not atomic. It is possible we've emptied the
711 * AIL here, so if that is the case, pass an LSN of 0 to the tail move.
712 */
713 mlip = xfs_ail_min(ailp);
714 tail_lsn = mlip ? mlip->li_lsn : 0;
715 spin_unlock(&ailp->xa_lock);
716 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
717}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 * The active item list (AIL) is a doubly linked list of log
721 * items sorted by ascending lsn. The base of the list is
722 * a forw/back pointer pair embedded in the xfs mount structure.
723 * The base is initialized with both pointers pointing to the
724 * base. This case always needs to be distinguished, because
725 * the base has no lsn to look at. We almost always insert
726 * at the end of the list, so on inserts we search from the
727 * end of the list to find where the new item belongs.
728 */
729
730/*
731 * Initialize the doubly linked list to point only to itself.
732 */
David Chinner249a8c12008-02-05 12:13:32 +1100733int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734xfs_trans_ail_init(
735 xfs_mount_t *mp)
736{
David Chinner82fa9012008-10-30 17:38:26 +1100737 struct xfs_ail *ailp;
738
739 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
740 if (!ailp)
741 return ENOMEM;
742
743 ailp->xa_mount = mp;
744 INIT_LIST_HEAD(&ailp->xa_ail);
David Chinnerc7e8f262008-10-30 17:39:23 +1100745 spin_lock_init(&ailp->xa_lock);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000746 INIT_DELAYED_WORK(&ailp->xa_work, xfs_ail_worker);
David Chinner27d8d5f2008-10-30 17:38:39 +1100747 mp->m_ail = ailp;
748 return 0;
David Chinner249a8c12008-02-05 12:13:32 +1100749}
750
751void
752xfs_trans_ail_destroy(
753 xfs_mount_t *mp)
754{
David Chinner82fa9012008-10-30 17:38:26 +1100755 struct xfs_ail *ailp = mp->m_ail;
756
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000757 cancel_delayed_work_sync(&ailp->xa_work);
David Chinner82fa9012008-10-30 17:38:26 +1100758 kmem_free(ailp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759}