blob: 21c6d7ddbc06b474e102b30cb6a13c7690d1a2a5 [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"
23#include "xfs_trans.h"
24#include "xfs_sb.h"
David Chinnerda353b02007-08-28 14:00:13 +100025#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.h"
27#include "xfs_trans_priv.h"
Christoph Hellwig9e4c1092011-10-11 15:14:11 +000028#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_error.h"
30
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#ifdef DEBUG
Dave Chinnercd4a3c52011-04-08 12:45:07 +100032/*
33 * Check that the list is sorted as it should be.
34 */
35STATIC void
36xfs_ail_check(
37 struct xfs_ail *ailp,
38 xfs_log_item_t *lip)
39{
40 xfs_log_item_t *prev_lip;
41
42 if (list_empty(&ailp->xa_ail))
43 return;
44
45 /*
46 * Check the next and previous entries are valid.
47 */
48 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
49 prev_lip = list_entry(lip->li_ail.prev, xfs_log_item_t, li_ail);
50 if (&prev_lip->li_ail != &ailp->xa_ail)
51 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
52
53 prev_lip = list_entry(lip->li_ail.next, xfs_log_item_t, li_ail);
54 if (&prev_lip->li_ail != &ailp->xa_ail)
55 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0);
56
57
Dave Chinnercd4a3c52011-04-08 12:45:07 +100058}
59#else /* !DEBUG */
David Chinnerde08dbc2008-02-05 12:13:38 +110060#define xfs_ail_check(a,l)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#endif /* DEBUG */
62
Dave Chinnercd4a3c52011-04-08 12:45:07 +100063/*
Dave Chinnerfd074842011-04-08 12:45:07 +100064 * Return a pointer to the last item in the AIL. If the AIL is empty, then
65 * return NULL.
66 */
67static xfs_log_item_t *
68xfs_ail_max(
69 struct xfs_ail *ailp)
70{
71 if (list_empty(&ailp->xa_ail))
72 return NULL;
73
74 return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
75}
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/*
Dave Chinnercd4a3c52011-04-08 12:45:07 +100078 * Return a pointer to the item which follows the given item in the AIL. If
79 * the given item is the last item in the list, then return NULL.
80 */
81static xfs_log_item_t *
82xfs_ail_next(
83 struct xfs_ail *ailp,
84 xfs_log_item_t *lip)
85{
86 if (lip->li_ail.next == &ailp->xa_ail)
87 return NULL;
88
89 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
90}
91
92/*
93 * This is called by the log manager code to determine the LSN of the tail of
94 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
95 * is empty, then this function returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 *
Dave Chinnercd4a3c52011-04-08 12:45:07 +100097 * We need the AIL lock in order to get a coherent read of the lsn of the last
98 * item in the AIL.
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 */
100xfs_lsn_t
Dave Chinnerfd074842011-04-08 12:45:07 +1000101xfs_ail_min_lsn(
David Chinner5b00f142008-10-30 17:39:00 +1100102 struct xfs_ail *ailp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103{
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000104 xfs_lsn_t lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
David Chinnerc7e8f262008-10-30 17:39:23 +1100107 spin_lock(&ailp->xa_lock);
David Chinner5b00f142008-10-30 17:39:00 +1100108 lip = xfs_ail_min(ailp);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000109 if (lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 lsn = lip->li_lsn;
David Chinnerc7e8f262008-10-30 17:39:23 +1100111 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
113 return lsn;
114}
115
116/*
Dave Chinnerfd074842011-04-08 12:45:07 +1000117 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
118 */
119static xfs_lsn_t
120xfs_ail_max_lsn(
121 struct xfs_ail *ailp)
122{
123 xfs_lsn_t lsn = 0;
124 xfs_log_item_t *lip;
125
126 spin_lock(&ailp->xa_lock);
127 lip = xfs_ail_max(ailp);
128 if (lip)
129 lsn = lip->li_lsn;
130 spin_unlock(&ailp->xa_lock);
131
132 return lsn;
133}
134
135/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000136 * The cursor keeps track of where our current traversal is up to by tracking
137 * the next item in the list for us. However, for this to be safe, removing an
138 * object from the AIL needs to invalidate any cursor that points to it. hence
139 * the traversal cursor needs to be linked to the struct xfs_ail so that
140 * deletion can search all the active cursors for invalidation.
David Chinner27d8d5f2008-10-30 17:38:39 +1100141 */
David Chinner5b00f142008-10-30 17:39:00 +1100142STATIC void
David Chinner27d8d5f2008-10-30 17:38:39 +1100143xfs_trans_ail_cursor_init(
144 struct xfs_ail *ailp,
145 struct xfs_ail_cursor *cur)
146{
147 cur->item = NULL;
Dave Chinneraf3e4022011-07-18 03:40:18 +0000148 list_add_tail(&cur->list, &ailp->xa_cursors);
David Chinner27d8d5f2008-10-30 17:38:39 +1100149}
150
151/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000152 * Get the next item in the traversal and advance the cursor. If the cursor
153 * was invalidated (indicated by a lip of 1), restart the traversal.
David Chinner27d8d5f2008-10-30 17:38:39 +1100154 */
David Chinner5b00f142008-10-30 17:39:00 +1100155struct xfs_log_item *
David Chinner27d8d5f2008-10-30 17:38:39 +1100156xfs_trans_ail_cursor_next(
157 struct xfs_ail *ailp,
158 struct xfs_ail_cursor *cur)
159{
160 struct xfs_log_item *lip = cur->item;
161
162 if ((__psint_t)lip & 1)
163 lip = xfs_ail_min(ailp);
Dave Chinner16b59022011-07-18 03:40:17 +0000164 if (lip)
165 cur->item = xfs_ail_next(ailp, lip);
David Chinner27d8d5f2008-10-30 17:38:39 +1100166 return lip;
167}
168
169/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000170 * When the traversal is complete, we need to remove the cursor from the list
171 * of traversing cursors.
David Chinner27d8d5f2008-10-30 17:38:39 +1100172 */
173void
174xfs_trans_ail_cursor_done(
175 struct xfs_ail *ailp,
Dave Chinneraf3e4022011-07-18 03:40:18 +0000176 struct xfs_ail_cursor *cur)
David Chinner27d8d5f2008-10-30 17:38:39 +1100177{
Dave Chinneraf3e4022011-07-18 03:40:18 +0000178 cur->item = NULL;
179 list_del_init(&cur->list);
David Chinner27d8d5f2008-10-30 17:38:39 +1100180}
181
182/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000183 * Invalidate any cursor that is pointing to this item. This is called when an
184 * item is removed from the AIL. Any cursor pointing to this object is now
185 * invalid and the traversal needs to be terminated so it doesn't reference a
186 * freed object. We set the low bit of the cursor item pointer so we can
187 * distinguish between an invalidation and the end of the list when getting the
188 * next item from the cursor.
David Chinner5b00f142008-10-30 17:39:00 +1100189 */
190STATIC void
191xfs_trans_ail_cursor_clear(
192 struct xfs_ail *ailp,
193 struct xfs_log_item *lip)
194{
195 struct xfs_ail_cursor *cur;
196
Dave Chinneraf3e4022011-07-18 03:40:18 +0000197 list_for_each_entry(cur, &ailp->xa_cursors, list) {
David Chinner5b00f142008-10-30 17:39:00 +1100198 if (cur->item == lip)
199 cur->item = (struct xfs_log_item *)
200 ((__psint_t)cur->item | 1);
201 }
202}
203
204/*
Dave Chinner16b59022011-07-18 03:40:17 +0000205 * Find the first item in the AIL with the given @lsn by searching in ascending
206 * LSN order and initialise the cursor to point to the next item for a
207 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
208 * first item in the AIL. Returns NULL if the list is empty.
David Chinner249a8c12008-02-05 12:13:32 +1100209 */
David Chinner5b00f142008-10-30 17:39:00 +1100210xfs_log_item_t *
211xfs_trans_ail_cursor_first(
David Chinner27d8d5f2008-10-30 17:38:39 +1100212 struct xfs_ail *ailp,
213 struct xfs_ail_cursor *cur,
214 xfs_lsn_t lsn)
David Chinner249a8c12008-02-05 12:13:32 +1100215{
David Chinner27d8d5f2008-10-30 17:38:39 +1100216 xfs_log_item_t *lip;
David Chinner249a8c12008-02-05 12:13:32 +1100217
David Chinner5b00f142008-10-30 17:39:00 +1100218 xfs_trans_ail_cursor_init(ailp, cur);
Dave Chinner16b59022011-07-18 03:40:17 +0000219
220 if (lsn == 0) {
221 lip = xfs_ail_min(ailp);
David Chinner5b00f142008-10-30 17:39:00 +1100222 goto out;
Dave Chinner16b59022011-07-18 03:40:17 +0000223 }
David Chinner249a8c12008-02-05 12:13:32 +1100224
David Chinner27d8d5f2008-10-30 17:38:39 +1100225 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
David Chinner5b00f142008-10-30 17:39:00 +1100226 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
David Chinner7ee49ac2008-10-30 18:26:51 +1100227 goto out;
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100228 }
Dave Chinner16b59022011-07-18 03:40:17 +0000229 return NULL;
230
David Chinner5b00f142008-10-30 17:39:00 +1100231out:
Dave Chinner16b59022011-07-18 03:40:17 +0000232 if (lip)
233 cur->item = xfs_ail_next(ailp, lip);
David Chinner5b00f142008-10-30 17:39:00 +1100234 return lip;
David Chinner249a8c12008-02-05 12:13:32 +1100235}
236
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000237static struct xfs_log_item *
238__xfs_trans_ail_cursor_last(
239 struct xfs_ail *ailp,
240 xfs_lsn_t lsn)
241{
242 xfs_log_item_t *lip;
243
244 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
245 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
246 return lip;
247 }
248 return NULL;
249}
250
251/*
Dave Chinner16b59022011-07-18 03:40:17 +0000252 * Find the last item in the AIL with the given @lsn by searching in descending
253 * LSN order and initialise the cursor to point to that item. If there is no
254 * item with the value of @lsn, then it sets the cursor to the last item with an
255 * LSN lower than @lsn. Returns NULL if the list is empty.
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000256 */
257struct xfs_log_item *
258xfs_trans_ail_cursor_last(
259 struct xfs_ail *ailp,
260 struct xfs_ail_cursor *cur,
261 xfs_lsn_t lsn)
262{
263 xfs_trans_ail_cursor_init(ailp, cur);
264 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
265 return cur->item;
266}
267
268/*
Dave Chinner16b59022011-07-18 03:40:17 +0000269 * Splice the log item list into the AIL at the given LSN. We splice to the
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000270 * tail of the given LSN to maintain insert order for push traversals. The
271 * cursor is optional, allowing repeated updates to the same LSN to avoid
Alex Eldere44f4112011-07-22 16:04:41 +0000272 * repeated traversals. This should not be called with an empty list.
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000273 */
274static void
275xfs_ail_splice(
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000276 struct xfs_ail *ailp,
277 struct xfs_ail_cursor *cur,
278 struct list_head *list,
279 xfs_lsn_t lsn)
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000280{
Alex Eldere44f4112011-07-22 16:04:41 +0000281 struct xfs_log_item *lip;
282
283 ASSERT(!list_empty(list));
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000284
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000285 /*
Alex Eldere44f4112011-07-22 16:04:41 +0000286 * Use the cursor to determine the insertion point if one is
287 * provided. If not, or if the one we got is not valid,
288 * find the place in the AIL where the items belong.
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000289 */
Alex Eldere44f4112011-07-22 16:04:41 +0000290 lip = cur ? cur->item : NULL;
291 if (!lip || (__psint_t) lip & 1)
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000292 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
293
Alex Eldere44f4112011-07-22 16:04:41 +0000294 /*
295 * If a cursor is provided, we know we're processing the AIL
296 * in lsn order, and future items to be spliced in will
297 * follow the last one being inserted now. Update the
298 * cursor to point to that last item, now while we have a
299 * reliable pointer to it.
300 */
301 if (cur)
302 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000303
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000304 /*
Alex Eldere44f4112011-07-22 16:04:41 +0000305 * Finally perform the splice. Unless the AIL was empty,
306 * lip points to the item in the AIL _after_ which the new
307 * items should go. If lip is null the AIL was empty, so
308 * the new items go at the head of the AIL.
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000309 */
Alex Eldere44f4112011-07-22 16:04:41 +0000310 if (lip)
311 list_splice(list, &lip->li_ail);
312 else
313 list_splice(list, &ailp->xa_ail);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000314}
315
316/*
317 * Delete the given item from the AIL. Return a pointer to the item.
318 */
319static void
320xfs_ail_delete(
321 struct xfs_ail *ailp,
322 xfs_log_item_t *lip)
323{
324 xfs_ail_check(ailp, lip);
325 list_del(&lip->li_ail);
326 xfs_trans_ail_cursor_clear(ailp, lip);
327}
328
Christoph Hellwig00308072011-10-11 11:14:10 -0400329static long
330xfsaild_push(
331 struct xfs_ail *ailp)
David Chinner249a8c12008-02-05 12:13:32 +1100332{
Dave Chinner9e7004e2011-05-06 02:54:05 +0000333 xfs_mount_t *mp = ailp->xa_mount;
Dave Chinneraf3e4022011-07-18 03:40:18 +0000334 struct xfs_ail_cursor cur;
Dave Chinner9e7004e2011-05-06 02:54:05 +0000335 xfs_log_item_t *lip;
336 xfs_lsn_t lsn;
Dave Chinnerfe0da762011-05-06 02:54:07 +0000337 xfs_lsn_t target;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000338 long tout;
Dave Chinner9e7004e2011-05-06 02:54:05 +0000339 int stuck = 0;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000340 int flushing = 0;
Dave Chinner9e7004e2011-05-06 02:54:05 +0000341 int count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Dave Chinner670ce932011-09-30 04:45:03 +0000343 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000344 * If we encountered pinned items or did not finish writing out all
345 * buffers the last time we ran, force the log first and wait for it
346 * before pushing again.
Dave Chinner670ce932011-09-30 04:45:03 +0000347 */
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000348 if (ailp->xa_log_flush && ailp->xa_last_pushed_lsn == 0 &&
349 (!list_empty_careful(&ailp->xa_buf_list) ||
350 xfs_ail_min_lsn(ailp))) {
Dave Chinner670ce932011-09-30 04:45:03 +0000351 ailp->xa_log_flush = 0;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000352
Dave Chinner670ce932011-09-30 04:45:03 +0000353 XFS_STATS_INC(xs_push_ail_flush);
354 xfs_log_force(mp, XFS_LOG_SYNC);
Dave Chinner670ce932011-09-30 04:45:03 +0000355 }
356
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000357 spin_lock(&ailp->xa_lock);
Brian Foster8375f922012-06-28 06:52:56 -0400358
359 /* barrier matches the xa_target update in xfs_ail_push() */
360 smp_rmb();
361 target = ailp->xa_target;
362 ailp->xa_target_prev = target;
363
Dave Chinneraf3e4022011-07-18 03:40:18 +0000364 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000365 if (!lip) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000367 * If the AIL is empty or our push has reached the end we are
368 * done now.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 */
Dave Chinneraf3e4022011-07-18 03:40:18 +0000370 xfs_trans_ail_cursor_done(ailp, &cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100371 spin_unlock(&ailp->xa_lock);
Dave Chinner9e7004e2011-05-06 02:54:05 +0000372 goto out_done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
374
375 XFS_STATS_INC(xs_push_ail);
376
David Chinner249a8c12008-02-05 12:13:32 +1100377 lsn = lip->li_lsn;
Dave Chinner50e86682011-05-06 02:54:06 +0000378 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
David Chinner249a8c12008-02-05 12:13:32 +1100379 int lock_result;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000380
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 /*
Dave Chinner904c17e2013-08-28 21:12:03 +1000382 * Note that iop_push may unlock and reacquire the AIL lock. We
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000383 * rely on the AIL cursor implementation to be able to deal with
384 * the dropped lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 */
Dave Chinner904c17e2013-08-28 21:12:03 +1000386 lock_result = lip->li_ops->iop_push(lip, &ailp->xa_buf_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 switch (lock_result) {
David Chinner249a8c12008-02-05 12:13:32 +1100388 case XFS_ITEM_SUCCESS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 XFS_STATS_INC(xs_push_ail_success);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000390 trace_xfs_ail_push(lip);
391
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000392 ailp->xa_last_pushed_lsn = lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 break;
394
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000395 case XFS_ITEM_FLUSHING:
396 /*
397 * The item or its backing buffer is already beeing
398 * flushed. The typical reason for that is that an
399 * inode buffer is locked because we already pushed the
400 * updates to it as part of inode clustering.
401 *
402 * We do not want to to stop flushing just because lots
403 * of items are already beeing flushed, but we need to
404 * re-try the flushing relatively soon if most of the
405 * AIL is beeing flushed.
406 */
407 XFS_STATS_INC(xs_push_ail_flushing);
408 trace_xfs_ail_flushing(lip);
Christoph Hellwig17b38472011-10-11 15:14:09 +0000409
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000410 flushing++;
411 ailp->xa_last_pushed_lsn = lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 break;
413
David Chinner249a8c12008-02-05 12:13:32 +1100414 case XFS_ITEM_PINNED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 XFS_STATS_INC(xs_push_ail_pinned);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000416 trace_xfs_ail_pinned(lip);
417
David Chinner249a8c12008-02-05 12:13:32 +1100418 stuck++;
Dave Chinner670ce932011-09-30 04:45:03 +0000419 ailp->xa_log_flush++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 break;
David Chinner249a8c12008-02-05 12:13:32 +1100421 case XFS_ITEM_LOCKED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 XFS_STATS_INC(xs_push_ail_locked);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000423 trace_xfs_ail_locked(lip);
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000424
David Chinner249a8c12008-02-05 12:13:32 +1100425 stuck++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 break;
David Chinner249a8c12008-02-05 12:13:32 +1100427 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 ASSERT(0);
429 break;
430 }
431
David Chinner249a8c12008-02-05 12:13:32 +1100432 count++;
433
434 /*
435 * Are there too many items we can't do anything with?
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000436 *
David Chinner249a8c12008-02-05 12:13:32 +1100437 * If we we are skipping too many items because we can't flush
438 * them or they are already being flushed, we back off and
439 * given them time to complete whatever operation is being
440 * done. i.e. remove pressure from the AIL while we can't make
441 * progress so traversals don't slow down further inserts and
442 * removals to/from the AIL.
443 *
444 * The value of 100 is an arbitrary magic number based on
445 * observation.
446 */
447 if (stuck > 100)
448 break;
449
Dave Chinneraf3e4022011-07-18 03:40:18 +0000450 lip = xfs_trans_ail_cursor_next(ailp, &cur);
David Chinner249a8c12008-02-05 12:13:32 +1100451 if (lip == NULL)
452 break;
David Chinner249a8c12008-02-05 12:13:32 +1100453 lsn = lip->li_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
Dave Chinneraf3e4022011-07-18 03:40:18 +0000455 xfs_trans_ail_cursor_done(ailp, &cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100456 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000458 if (xfs_buf_delwri_submit_nowait(&ailp->xa_buf_list))
459 ailp->xa_log_flush++;
Dave Chinnerd808f612010-02-02 10:13:42 +1100460
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000461 if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
Dave Chinner9e7004e2011-05-06 02:54:05 +0000462out_done:
David Chinner92d9cd12008-03-06 13:45:10 +1100463 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000464 * We reached the target or the AIL is empty, so wait a bit
465 * longer for I/O to complete and remove pushed items from the
466 * AIL before we start the next scan from the start of the AIL.
David Chinner92d9cd12008-03-06 13:45:10 +1100467 */
Dave Chinner453eac82010-01-11 11:49:58 +0000468 tout = 50;
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000469 ailp->xa_last_pushed_lsn = 0;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000470 } else if (((stuck + flushing) * 100) / count > 90) {
David Chinner249a8c12008-02-05 12:13:32 +1100471 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000472 * Either there is a lot of contention on the AIL or we are
473 * stuck due to operations in progress. "Stuck" in this case
474 * is defined as >90% of the items we tried to push were stuck.
David Chinner249a8c12008-02-05 12:13:32 +1100475 *
476 * Backoff a bit more to allow some I/O to complete before
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000477 * restarting from the start of the AIL. This prevents us from
478 * spinning on the same items, and if they are pinned will all
479 * the restart to issue a log force to unpin the stuck items.
David Chinner249a8c12008-02-05 12:13:32 +1100480 */
Dave Chinner453eac82010-01-11 11:49:58 +0000481 tout = 20;
Dave Chinner670ce932011-09-30 04:45:03 +0000482 ailp->xa_last_pushed_lsn = 0;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000483 } else {
484 /*
485 * Assume we have more work to do in a short while.
486 */
487 tout = 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 }
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000489
Christoph Hellwig00308072011-10-11 11:14:10 -0400490 return tout;
491}
492
493static int
494xfsaild(
495 void *data)
496{
497 struct xfs_ail *ailp = data;
498 long tout = 0; /* milliseconds */
499
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000500 current->flags |= PF_MEMALLOC;
501
Christoph Hellwig00308072011-10-11 11:14:10 -0400502 while (!kthread_should_stop()) {
503 if (tout && tout <= 20)
504 __set_current_state(TASK_KILLABLE);
505 else
506 __set_current_state(TASK_INTERRUPTIBLE);
Brian Foster8375f922012-06-28 06:52:56 -0400507
508 spin_lock(&ailp->xa_lock);
509
510 /*
511 * Idle if the AIL is empty and we are not racing with a target
512 * update. We check the AIL after we set the task to a sleep
513 * state to guarantee that we either catch an xa_target update
514 * or that a wake_up resets the state to TASK_RUNNING.
515 * Otherwise, we run the risk of sleeping indefinitely.
516 *
517 * The barrier matches the xa_target update in xfs_ail_push().
518 */
519 smp_rmb();
520 if (!xfs_ail_min(ailp) &&
521 ailp->xa_target == ailp->xa_target_prev) {
522 spin_unlock(&ailp->xa_lock);
523 schedule();
524 tout = 0;
525 continue;
526 }
527 spin_unlock(&ailp->xa_lock);
528
529 if (tout)
530 schedule_timeout(msecs_to_jiffies(tout));
531
532 __set_current_state(TASK_RUNNING);
Christoph Hellwig00308072011-10-11 11:14:10 -0400533
534 try_to_freeze();
535
536 tout = xfsaild_push(ailp);
537 }
538
539 return 0;
Dave Chinner453eac82010-01-11 11:49:58 +0000540}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000542/*
543 * This routine is called to move the tail of the AIL forward. It does this by
544 * trying to flush items in the AIL whose lsns are below the given
545 * threshold_lsn.
546 *
547 * The push is run asynchronously in a workqueue, which means the caller needs
548 * to handle waiting on the async flush for space to become available.
549 * We don't want to interrupt any push that is in progress, hence we only queue
550 * work if we set the pushing bit approriately.
551 *
552 * We do this unlocked - we only need to know whether there is anything in the
553 * AIL at the time we are called. We don't need to access the contents of
554 * any of the objects, so the lock is not needed.
555 */
556void
Dave Chinnerfd074842011-04-08 12:45:07 +1000557xfs_ail_push(
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000558 struct xfs_ail *ailp,
559 xfs_lsn_t threshold_lsn)
560{
561 xfs_log_item_t *lip;
562
563 lip = xfs_ail_min(ailp);
564 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
565 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
566 return;
567
568 /*
569 * Ensure that the new target is noticed in push code before it clears
570 * the XFS_AIL_PUSHING_BIT.
571 */
572 smp_wmb();
Dave Chinnerfe0da762011-05-06 02:54:07 +0000573 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
Christoph Hellwig00308072011-10-11 11:14:10 -0400574 smp_wmb();
575
576 wake_up_process(ailp->xa_task);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000577}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
579/*
Dave Chinnerfd074842011-04-08 12:45:07 +1000580 * Push out all items in the AIL immediately
581 */
582void
583xfs_ail_push_all(
584 struct xfs_ail *ailp)
585{
586 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
587
588 if (threshold_lsn)
589 xfs_ail_push(ailp, threshold_lsn);
590}
591
592/*
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000593 * Push out all items in the AIL immediately and wait until the AIL is empty.
594 */
595void
596xfs_ail_push_all_sync(
597 struct xfs_ail *ailp)
598{
599 struct xfs_log_item *lip;
600 DEFINE_WAIT(wait);
601
602 spin_lock(&ailp->xa_lock);
603 while ((lip = xfs_ail_max(ailp)) != NULL) {
604 prepare_to_wait(&ailp->xa_empty, &wait, TASK_UNINTERRUPTIBLE);
605 ailp->xa_target = lip->li_lsn;
606 wake_up_process(ailp->xa_task);
607 spin_unlock(&ailp->xa_lock);
608 schedule();
609 spin_lock(&ailp->xa_lock);
610 }
611 spin_unlock(&ailp->xa_lock);
612
613 finish_wait(&ailp->xa_empty, &wait);
614}
615
616/*
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100617 * xfs_trans_ail_update - bulk AIL insertion operation.
618 *
619 * @xfs_trans_ail_update takes an array of log items that all need to be
620 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
621 * be added. Otherwise, it will be repositioned by removing it and re-adding
622 * it to the AIL. If we move the first item in the AIL, update the log tail to
623 * match the new minimum LSN in the AIL.
624 *
625 * This function takes the AIL lock once to execute the update operations on
626 * all the items in the array, and as such should not be called with the AIL
627 * lock held. As a result, once we have the AIL lock, we need to check each log
628 * item LSN to confirm it needs to be moved forward in the AIL.
629 *
630 * To optimise the insert operation, we delete all the items from the AIL in
631 * the first pass, moving them into a temporary list, then splice the temporary
632 * list into the correct position in the AIL. This avoids needing to do an
633 * insert operation on every item.
634 *
635 * This function must be called with the AIL lock held. The lock is dropped
636 * before returning.
637 */
638void
639xfs_trans_ail_update_bulk(
640 struct xfs_ail *ailp,
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000641 struct xfs_ail_cursor *cur,
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100642 struct xfs_log_item **log_items,
643 int nr_items,
644 xfs_lsn_t lsn) __releases(ailp->xa_lock)
645{
646 xfs_log_item_t *mlip;
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100647 int mlip_changed = 0;
648 int i;
649 LIST_HEAD(tmp);
650
Alex Eldere44f4112011-07-22 16:04:41 +0000651 ASSERT(nr_items > 0); /* Not required, but true. */
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100652 mlip = xfs_ail_min(ailp);
653
654 for (i = 0; i < nr_items; i++) {
655 struct xfs_log_item *lip = log_items[i];
656 if (lip->li_flags & XFS_LI_IN_AIL) {
657 /* check if we really need to move the item */
658 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
659 continue;
660
661 xfs_ail_delete(ailp, lip);
662 if (mlip == lip)
663 mlip_changed = 1;
664 } else {
665 lip->li_flags |= XFS_LI_IN_AIL;
666 }
667 lip->li_lsn = lsn;
668 list_add(&lip->li_ail, &tmp);
669 }
670
Alex Eldere44f4112011-07-22 16:04:41 +0000671 if (!list_empty(&tmp))
672 xfs_ail_splice(ailp, cur, &tmp, lsn);
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000673
Christoph Hellwig1c304622012-04-23 15:58:33 +1000674 if (mlip_changed) {
675 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
676 xlog_assign_tail_lsn_locked(ailp->xa_mount);
677 spin_unlock(&ailp->xa_lock);
678
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +0000679 xfs_log_space_wake(ailp->xa_mount);
Christoph Hellwig1c304622012-04-23 15:58:33 +1000680 } else {
681 spin_unlock(&ailp->xa_lock);
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000682 }
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100683}
684
685/*
Dave Chinner30136832010-12-20 12:03:17 +1100686 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
687 *
688 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
689 * removed from the AIL. The caller is already holding the AIL lock, and done
690 * all the checks necessary to ensure the items passed in via @log_items are
691 * ready for deletion. This includes checking that the items are in the AIL.
692 *
693 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
694 * flag from the item and reset the item's lsn to 0. If we remove the first
695 * item in the AIL, update the log tail to match the new minimum LSN in the
696 * AIL.
697 *
698 * This function will not drop the AIL lock until all items are removed from
699 * the AIL to minimise the amount of lock traffic on the AIL. This does not
700 * greatly increase the AIL hold time, but does significantly reduce the amount
701 * of traffic on the lock, especially during IO completion.
702 *
703 * This function must be called with the AIL lock held. The lock is dropped
704 * before returning.
705 */
706void
707xfs_trans_ail_delete_bulk(
708 struct xfs_ail *ailp,
709 struct xfs_log_item **log_items,
Dave Chinner04913fd2012-04-23 15:58:41 +1000710 int nr_items,
711 int shutdown_type) __releases(ailp->xa_lock)
Dave Chinner30136832010-12-20 12:03:17 +1100712{
713 xfs_log_item_t *mlip;
Dave Chinner30136832010-12-20 12:03:17 +1100714 int mlip_changed = 0;
715 int i;
716
717 mlip = xfs_ail_min(ailp);
718
719 for (i = 0; i < nr_items; i++) {
720 struct xfs_log_item *lip = log_items[i];
721 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
722 struct xfs_mount *mp = ailp->xa_mount;
723
724 spin_unlock(&ailp->xa_lock);
725 if (!XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner6a19d932011-03-07 10:02:35 +1100726 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
Dave Chinner30136832010-12-20 12:03:17 +1100727 "%s: attempting to delete a log item that is not in the AIL",
728 __func__);
Dave Chinner04913fd2012-04-23 15:58:41 +1000729 xfs_force_shutdown(mp, shutdown_type);
Dave Chinner30136832010-12-20 12:03:17 +1100730 }
731 return;
732 }
733
734 xfs_ail_delete(ailp, lip);
735 lip->li_flags &= ~XFS_LI_IN_AIL;
736 lip->li_lsn = 0;
737 if (mlip == lip)
738 mlip_changed = 1;
739 }
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000740
Christoph Hellwig1c304622012-04-23 15:58:33 +1000741 if (mlip_changed) {
742 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
743 xlog_assign_tail_lsn_locked(ailp->xa_mount);
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000744 if (list_empty(&ailp->xa_ail))
745 wake_up_all(&ailp->xa_empty);
Christoph Hellwig1c304622012-04-23 15:58:33 +1000746 spin_unlock(&ailp->xa_lock);
747
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +0000748 xfs_log_space_wake(ailp->xa_mount);
Christoph Hellwig1c304622012-04-23 15:58:33 +1000749 } else {
750 spin_unlock(&ailp->xa_lock);
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000751 }
Dave Chinner30136832010-12-20 12:03:17 +1100752}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
David Chinner249a8c12008-02-05 12:13:32 +1100754int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755xfs_trans_ail_init(
756 xfs_mount_t *mp)
757{
David Chinner82fa9012008-10-30 17:38:26 +1100758 struct xfs_ail *ailp;
759
760 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
761 if (!ailp)
762 return ENOMEM;
763
764 ailp->xa_mount = mp;
765 INIT_LIST_HEAD(&ailp->xa_ail);
Dave Chinneraf3e4022011-07-18 03:40:18 +0000766 INIT_LIST_HEAD(&ailp->xa_cursors);
David Chinnerc7e8f262008-10-30 17:39:23 +1100767 spin_lock_init(&ailp->xa_lock);
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000768 INIT_LIST_HEAD(&ailp->xa_buf_list);
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000769 init_waitqueue_head(&ailp->xa_empty);
Christoph Hellwig00308072011-10-11 11:14:10 -0400770
771 ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
772 ailp->xa_mount->m_fsname);
773 if (IS_ERR(ailp->xa_task))
774 goto out_free_ailp;
775
David Chinner27d8d5f2008-10-30 17:38:39 +1100776 mp->m_ail = ailp;
777 return 0;
Christoph Hellwig00308072011-10-11 11:14:10 -0400778
779out_free_ailp:
780 kmem_free(ailp);
781 return ENOMEM;
David Chinner249a8c12008-02-05 12:13:32 +1100782}
783
784void
785xfs_trans_ail_destroy(
786 xfs_mount_t *mp)
787{
David Chinner82fa9012008-10-30 17:38:26 +1100788 struct xfs_ail *ailp = mp->m_ail;
789
Christoph Hellwig00308072011-10-11 11:14:10 -0400790 kthread_stop(ailp->xa_task);
David Chinner82fa9012008-10-30 17:38:26 +1100791 kmem_free(ailp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792}