blob: 4b47cfebd25b8ad34f353469281484d4a8df1295 [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"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_sb.h"
David Chinnerda353b02007-08-28 14:00:13 +100024#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_mount.h"
Dave Chinner239880e2013-10-23 10:50:10 +110026#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#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"
Dave Chinner239880e2013-10-23 10:50:10 +110030#include "xfs_log.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#ifdef DEBUG
Dave Chinnercd4a3c52011-04-08 12:45:07 +100033/*
34 * Check that the list is sorted as it should be.
35 */
36STATIC void
37xfs_ail_check(
38 struct xfs_ail *ailp,
39 xfs_log_item_t *lip)
40{
41 xfs_log_item_t *prev_lip;
42
43 if (list_empty(&ailp->xa_ail))
44 return;
45
46 /*
47 * Check the next and previous entries are valid.
48 */
49 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
50 prev_lip = list_entry(lip->li_ail.prev, xfs_log_item_t, li_ail);
51 if (&prev_lip->li_ail != &ailp->xa_ail)
52 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
53
54 prev_lip = list_entry(lip->li_ail.next, xfs_log_item_t, li_ail);
55 if (&prev_lip->li_ail != &ailp->xa_ail)
56 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0);
57
58
Dave Chinnercd4a3c52011-04-08 12:45:07 +100059}
60#else /* !DEBUG */
David Chinnerde08dbc2008-02-05 12:13:38 +110061#define xfs_ail_check(a,l)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062#endif /* DEBUG */
63
Dave Chinnercd4a3c52011-04-08 12:45:07 +100064/*
Dave Chinnerfd074842011-04-08 12:45:07 +100065 * Return a pointer to the last item in the AIL. If the AIL is empty, then
66 * return NULL.
67 */
68static xfs_log_item_t *
69xfs_ail_max(
70 struct xfs_ail *ailp)
71{
72 if (list_empty(&ailp->xa_ail))
73 return NULL;
74
75 return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
76}
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/*
Dave Chinnercd4a3c52011-04-08 12:45:07 +100079 * Return a pointer to the item which follows the given item in the AIL. If
80 * the given item is the last item in the list, then return NULL.
81 */
82static xfs_log_item_t *
83xfs_ail_next(
84 struct xfs_ail *ailp,
85 xfs_log_item_t *lip)
86{
87 if (lip->li_ail.next == &ailp->xa_ail)
88 return NULL;
89
90 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
91}
92
93/*
94 * This is called by the log manager code to determine the LSN of the tail of
95 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
96 * is empty, then this function returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 *
Dave Chinnercd4a3c52011-04-08 12:45:07 +100098 * We need the AIL lock in order to get a coherent read of the lsn of the last
99 * item in the AIL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 */
101xfs_lsn_t
Dave Chinnerfd074842011-04-08 12:45:07 +1000102xfs_ail_min_lsn(
David Chinner5b00f142008-10-30 17:39:00 +1100103 struct xfs_ail *ailp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000105 xfs_lsn_t lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
David Chinnerc7e8f262008-10-30 17:39:23 +1100108 spin_lock(&ailp->xa_lock);
David Chinner5b00f142008-10-30 17:39:00 +1100109 lip = xfs_ail_min(ailp);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000110 if (lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 lsn = lip->li_lsn;
David Chinnerc7e8f262008-10-30 17:39:23 +1100112 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
114 return lsn;
115}
116
117/*
Dave Chinnerfd074842011-04-08 12:45:07 +1000118 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
119 */
120static xfs_lsn_t
121xfs_ail_max_lsn(
122 struct xfs_ail *ailp)
123{
124 xfs_lsn_t lsn = 0;
125 xfs_log_item_t *lip;
126
127 spin_lock(&ailp->xa_lock);
128 lip = xfs_ail_max(ailp);
129 if (lip)
130 lsn = lip->li_lsn;
131 spin_unlock(&ailp->xa_lock);
132
133 return lsn;
134}
135
136/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000137 * The cursor keeps track of where our current traversal is up to by tracking
138 * the next item in the list for us. However, for this to be safe, removing an
139 * object from the AIL needs to invalidate any cursor that points to it. hence
140 * the traversal cursor needs to be linked to the struct xfs_ail so that
141 * deletion can search all the active cursors for invalidation.
David Chinner27d8d5f2008-10-30 17:38:39 +1100142 */
David Chinner5b00f142008-10-30 17:39:00 +1100143STATIC void
David Chinner27d8d5f2008-10-30 17:38:39 +1100144xfs_trans_ail_cursor_init(
145 struct xfs_ail *ailp,
146 struct xfs_ail_cursor *cur)
147{
148 cur->item = NULL;
Dave Chinneraf3e4022011-07-18 03:40:18 +0000149 list_add_tail(&cur->list, &ailp->xa_cursors);
David Chinner27d8d5f2008-10-30 17:38:39 +1100150}
151
152/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000153 * Get the next item in the traversal and advance the cursor. If the cursor
154 * was invalidated (indicated by a lip of 1), restart the traversal.
David Chinner27d8d5f2008-10-30 17:38:39 +1100155 */
David Chinner5b00f142008-10-30 17:39:00 +1100156struct xfs_log_item *
David Chinner27d8d5f2008-10-30 17:38:39 +1100157xfs_trans_ail_cursor_next(
158 struct xfs_ail *ailp,
159 struct xfs_ail_cursor *cur)
160{
161 struct xfs_log_item *lip = cur->item;
162
163 if ((__psint_t)lip & 1)
164 lip = xfs_ail_min(ailp);
Dave Chinner16b59022011-07-18 03:40:17 +0000165 if (lip)
166 cur->item = xfs_ail_next(ailp, lip);
David Chinner27d8d5f2008-10-30 17:38:39 +1100167 return lip;
168}
169
170/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000171 * When the traversal is complete, we need to remove the cursor from the list
172 * of traversing cursors.
David Chinner27d8d5f2008-10-30 17:38:39 +1100173 */
174void
175xfs_trans_ail_cursor_done(
176 struct xfs_ail *ailp,
Dave Chinneraf3e4022011-07-18 03:40:18 +0000177 struct xfs_ail_cursor *cur)
David Chinner27d8d5f2008-10-30 17:38:39 +1100178{
Dave Chinneraf3e4022011-07-18 03:40:18 +0000179 cur->item = NULL;
180 list_del_init(&cur->list);
David Chinner27d8d5f2008-10-30 17:38:39 +1100181}
182
183/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000184 * Invalidate any cursor that is pointing to this item. This is called when an
185 * item is removed from the AIL. Any cursor pointing to this object is now
186 * invalid and the traversal needs to be terminated so it doesn't reference a
187 * freed object. We set the low bit of the cursor item pointer so we can
188 * distinguish between an invalidation and the end of the list when getting the
189 * next item from the cursor.
David Chinner5b00f142008-10-30 17:39:00 +1100190 */
191STATIC void
192xfs_trans_ail_cursor_clear(
193 struct xfs_ail *ailp,
194 struct xfs_log_item *lip)
195{
196 struct xfs_ail_cursor *cur;
197
Dave Chinneraf3e4022011-07-18 03:40:18 +0000198 list_for_each_entry(cur, &ailp->xa_cursors, list) {
David Chinner5b00f142008-10-30 17:39:00 +1100199 if (cur->item == lip)
200 cur->item = (struct xfs_log_item *)
201 ((__psint_t)cur->item | 1);
202 }
203}
204
205/*
Dave Chinner16b59022011-07-18 03:40:17 +0000206 * Find the first item in the AIL with the given @lsn by searching in ascending
207 * LSN order and initialise the cursor to point to the next item for a
208 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
209 * first item in the AIL. Returns NULL if the list is empty.
David Chinner249a8c12008-02-05 12:13:32 +1100210 */
David Chinner5b00f142008-10-30 17:39:00 +1100211xfs_log_item_t *
212xfs_trans_ail_cursor_first(
David Chinner27d8d5f2008-10-30 17:38:39 +1100213 struct xfs_ail *ailp,
214 struct xfs_ail_cursor *cur,
215 xfs_lsn_t lsn)
David Chinner249a8c12008-02-05 12:13:32 +1100216{
David Chinner27d8d5f2008-10-30 17:38:39 +1100217 xfs_log_item_t *lip;
David Chinner249a8c12008-02-05 12:13:32 +1100218
David Chinner5b00f142008-10-30 17:39:00 +1100219 xfs_trans_ail_cursor_init(ailp, cur);
Dave Chinner16b59022011-07-18 03:40:17 +0000220
221 if (lsn == 0) {
222 lip = xfs_ail_min(ailp);
David Chinner5b00f142008-10-30 17:39:00 +1100223 goto out;
Dave Chinner16b59022011-07-18 03:40:17 +0000224 }
David Chinner249a8c12008-02-05 12:13:32 +1100225
David Chinner27d8d5f2008-10-30 17:38:39 +1100226 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
David Chinner5b00f142008-10-30 17:39:00 +1100227 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
David Chinner7ee49ac2008-10-30 18:26:51 +1100228 goto out;
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100229 }
Dave Chinner16b59022011-07-18 03:40:17 +0000230 return NULL;
231
David Chinner5b00f142008-10-30 17:39:00 +1100232out:
Dave Chinner16b59022011-07-18 03:40:17 +0000233 if (lip)
234 cur->item = xfs_ail_next(ailp, lip);
David Chinner5b00f142008-10-30 17:39:00 +1100235 return lip;
David Chinner249a8c12008-02-05 12:13:32 +1100236}
237
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000238static struct xfs_log_item *
239__xfs_trans_ail_cursor_last(
240 struct xfs_ail *ailp,
241 xfs_lsn_t lsn)
242{
243 xfs_log_item_t *lip;
244
245 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
246 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
247 return lip;
248 }
249 return NULL;
250}
251
252/*
Dave Chinner16b59022011-07-18 03:40:17 +0000253 * Find the last item in the AIL with the given @lsn by searching in descending
254 * LSN order and initialise the cursor to point to that item. If there is no
255 * item with the value of @lsn, then it sets the cursor to the last item with an
256 * LSN lower than @lsn. Returns NULL if the list is empty.
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000257 */
258struct xfs_log_item *
259xfs_trans_ail_cursor_last(
260 struct xfs_ail *ailp,
261 struct xfs_ail_cursor *cur,
262 xfs_lsn_t lsn)
263{
264 xfs_trans_ail_cursor_init(ailp, cur);
265 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
266 return cur->item;
267}
268
269/*
Dave Chinner16b59022011-07-18 03:40:17 +0000270 * Splice the log item list into the AIL at the given LSN. We splice to the
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000271 * tail of the given LSN to maintain insert order for push traversals. The
272 * cursor is optional, allowing repeated updates to the same LSN to avoid
Alex Eldere44f4112011-07-22 16:04:41 +0000273 * repeated traversals. This should not be called with an empty list.
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000274 */
275static void
276xfs_ail_splice(
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000277 struct xfs_ail *ailp,
278 struct xfs_ail_cursor *cur,
279 struct list_head *list,
280 xfs_lsn_t lsn)
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000281{
Alex Eldere44f4112011-07-22 16:04:41 +0000282 struct xfs_log_item *lip;
283
284 ASSERT(!list_empty(list));
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000285
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000286 /*
Alex Eldere44f4112011-07-22 16:04:41 +0000287 * Use the cursor to determine the insertion point if one is
288 * provided. If not, or if the one we got is not valid,
289 * find the place in the AIL where the items belong.
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000290 */
Alex Eldere44f4112011-07-22 16:04:41 +0000291 lip = cur ? cur->item : NULL;
292 if (!lip || (__psint_t) lip & 1)
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000293 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
294
Alex Eldere44f4112011-07-22 16:04:41 +0000295 /*
296 * If a cursor is provided, we know we're processing the AIL
297 * in lsn order, and future items to be spliced in will
298 * follow the last one being inserted now. Update the
299 * cursor to point to that last item, now while we have a
300 * reliable pointer to it.
301 */
302 if (cur)
303 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000304
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000305 /*
Alex Eldere44f4112011-07-22 16:04:41 +0000306 * Finally perform the splice. Unless the AIL was empty,
307 * lip points to the item in the AIL _after_ which the new
308 * items should go. If lip is null the AIL was empty, so
309 * the new items go at the head of the AIL.
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000310 */
Alex Eldere44f4112011-07-22 16:04:41 +0000311 if (lip)
312 list_splice(list, &lip->li_ail);
313 else
314 list_splice(list, &ailp->xa_ail);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000315}
316
317/*
318 * Delete the given item from the AIL. Return a pointer to the item.
319 */
320static void
321xfs_ail_delete(
322 struct xfs_ail *ailp,
323 xfs_log_item_t *lip)
324{
325 xfs_ail_check(ailp, lip);
326 list_del(&lip->li_ail);
327 xfs_trans_ail_cursor_clear(ailp, lip);
328}
329
Christoph Hellwig00308072011-10-11 11:14:10 -0400330static long
331xfsaild_push(
332 struct xfs_ail *ailp)
David Chinner249a8c12008-02-05 12:13:32 +1100333{
Dave Chinner9e7004e2011-05-06 02:54:05 +0000334 xfs_mount_t *mp = ailp->xa_mount;
Dave Chinneraf3e4022011-07-18 03:40:18 +0000335 struct xfs_ail_cursor cur;
Dave Chinner9e7004e2011-05-06 02:54:05 +0000336 xfs_log_item_t *lip;
337 xfs_lsn_t lsn;
Dave Chinnerfe0da762011-05-06 02:54:07 +0000338 xfs_lsn_t target;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000339 long tout;
Dave Chinner9e7004e2011-05-06 02:54:05 +0000340 int stuck = 0;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000341 int flushing = 0;
Dave Chinner9e7004e2011-05-06 02:54:05 +0000342 int count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Dave Chinner670ce932011-09-30 04:45:03 +0000344 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000345 * If we encountered pinned items or did not finish writing out all
346 * buffers the last time we ran, force the log first and wait for it
347 * before pushing again.
Dave Chinner670ce932011-09-30 04:45:03 +0000348 */
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000349 if (ailp->xa_log_flush && ailp->xa_last_pushed_lsn == 0 &&
350 (!list_empty_careful(&ailp->xa_buf_list) ||
351 xfs_ail_min_lsn(ailp))) {
Dave Chinner670ce932011-09-30 04:45:03 +0000352 ailp->xa_log_flush = 0;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000353
Dave Chinner670ce932011-09-30 04:45:03 +0000354 XFS_STATS_INC(xs_push_ail_flush);
355 xfs_log_force(mp, XFS_LOG_SYNC);
Dave Chinner670ce932011-09-30 04:45:03 +0000356 }
357
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000358 spin_lock(&ailp->xa_lock);
Brian Foster8375f922012-06-28 06:52:56 -0400359
360 /* barrier matches the xa_target update in xfs_ail_push() */
361 smp_rmb();
362 target = ailp->xa_target;
363 ailp->xa_target_prev = target;
364
Dave Chinneraf3e4022011-07-18 03:40:18 +0000365 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000366 if (!lip) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000368 * If the AIL is empty or our push has reached the end we are
369 * done now.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 */
Dave Chinneraf3e4022011-07-18 03:40:18 +0000371 xfs_trans_ail_cursor_done(ailp, &cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100372 spin_unlock(&ailp->xa_lock);
Dave Chinner9e7004e2011-05-06 02:54:05 +0000373 goto out_done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 }
375
376 XFS_STATS_INC(xs_push_ail);
377
David Chinner249a8c12008-02-05 12:13:32 +1100378 lsn = lip->li_lsn;
Dave Chinner50e86682011-05-06 02:54:06 +0000379 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
David Chinner249a8c12008-02-05 12:13:32 +1100380 int lock_result;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 /*
Dave Chinner904c17e2013-08-28 21:12:03 +1000383 * Note that iop_push may unlock and reacquire the AIL lock. We
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000384 * rely on the AIL cursor implementation to be able to deal with
385 * the dropped lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 */
Dave Chinner904c17e2013-08-28 21:12:03 +1000387 lock_result = lip->li_ops->iop_push(lip, &ailp->xa_buf_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 switch (lock_result) {
David Chinner249a8c12008-02-05 12:13:32 +1100389 case XFS_ITEM_SUCCESS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 XFS_STATS_INC(xs_push_ail_success);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000391 trace_xfs_ail_push(lip);
392
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000393 ailp->xa_last_pushed_lsn = lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 break;
395
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000396 case XFS_ITEM_FLUSHING:
397 /*
398 * The item or its backing buffer is already beeing
399 * flushed. The typical reason for that is that an
400 * inode buffer is locked because we already pushed the
401 * updates to it as part of inode clustering.
402 *
403 * We do not want to to stop flushing just because lots
404 * of items are already beeing flushed, but we need to
405 * re-try the flushing relatively soon if most of the
406 * AIL is beeing flushed.
407 */
408 XFS_STATS_INC(xs_push_ail_flushing);
409 trace_xfs_ail_flushing(lip);
Christoph Hellwig17b38472011-10-11 15:14:09 +0000410
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000411 flushing++;
412 ailp->xa_last_pushed_lsn = lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 break;
414
David Chinner249a8c12008-02-05 12:13:32 +1100415 case XFS_ITEM_PINNED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 XFS_STATS_INC(xs_push_ail_pinned);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000417 trace_xfs_ail_pinned(lip);
418
David Chinner249a8c12008-02-05 12:13:32 +1100419 stuck++;
Dave Chinner670ce932011-09-30 04:45:03 +0000420 ailp->xa_log_flush++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 break;
David Chinner249a8c12008-02-05 12:13:32 +1100422 case XFS_ITEM_LOCKED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 XFS_STATS_INC(xs_push_ail_locked);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000424 trace_xfs_ail_locked(lip);
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000425
David Chinner249a8c12008-02-05 12:13:32 +1100426 stuck++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 break;
David Chinner249a8c12008-02-05 12:13:32 +1100428 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 ASSERT(0);
430 break;
431 }
432
David Chinner249a8c12008-02-05 12:13:32 +1100433 count++;
434
435 /*
436 * Are there too many items we can't do anything with?
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000437 *
David Chinner249a8c12008-02-05 12:13:32 +1100438 * If we we are skipping too many items because we can't flush
439 * them or they are already being flushed, we back off and
440 * given them time to complete whatever operation is being
441 * done. i.e. remove pressure from the AIL while we can't make
442 * progress so traversals don't slow down further inserts and
443 * removals to/from the AIL.
444 *
445 * The value of 100 is an arbitrary magic number based on
446 * observation.
447 */
448 if (stuck > 100)
449 break;
450
Dave Chinneraf3e4022011-07-18 03:40:18 +0000451 lip = xfs_trans_ail_cursor_next(ailp, &cur);
David Chinner249a8c12008-02-05 12:13:32 +1100452 if (lip == NULL)
453 break;
David Chinner249a8c12008-02-05 12:13:32 +1100454 lsn = lip->li_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
Dave Chinneraf3e4022011-07-18 03:40:18 +0000456 xfs_trans_ail_cursor_done(ailp, &cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100457 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000459 if (xfs_buf_delwri_submit_nowait(&ailp->xa_buf_list))
460 ailp->xa_log_flush++;
Dave Chinnerd808f612010-02-02 10:13:42 +1100461
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000462 if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
Dave Chinner9e7004e2011-05-06 02:54:05 +0000463out_done:
David Chinner92d9cd12008-03-06 13:45:10 +1100464 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000465 * We reached the target or the AIL is empty, so wait a bit
466 * longer for I/O to complete and remove pushed items from the
467 * AIL before we start the next scan from the start of the AIL.
David Chinner92d9cd12008-03-06 13:45:10 +1100468 */
Dave Chinner453eac82010-01-11 11:49:58 +0000469 tout = 50;
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000470 ailp->xa_last_pushed_lsn = 0;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000471 } else if (((stuck + flushing) * 100) / count > 90) {
David Chinner249a8c12008-02-05 12:13:32 +1100472 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000473 * Either there is a lot of contention on the AIL or we are
474 * stuck due to operations in progress. "Stuck" in this case
475 * is defined as >90% of the items we tried to push were stuck.
David Chinner249a8c12008-02-05 12:13:32 +1100476 *
477 * Backoff a bit more to allow some I/O to complete before
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000478 * restarting from the start of the AIL. This prevents us from
479 * spinning on the same items, and if they are pinned will all
480 * the restart to issue a log force to unpin the stuck items.
David Chinner249a8c12008-02-05 12:13:32 +1100481 */
Dave Chinner453eac82010-01-11 11:49:58 +0000482 tout = 20;
Dave Chinner670ce932011-09-30 04:45:03 +0000483 ailp->xa_last_pushed_lsn = 0;
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000484 } else {
485 /*
486 * Assume we have more work to do in a short while.
487 */
488 tout = 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 }
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000490
Christoph Hellwig00308072011-10-11 11:14:10 -0400491 return tout;
492}
493
494static int
495xfsaild(
496 void *data)
497{
498 struct xfs_ail *ailp = data;
499 long tout = 0; /* milliseconds */
500
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000501 current->flags |= PF_MEMALLOC;
502
Christoph Hellwig00308072011-10-11 11:14:10 -0400503 while (!kthread_should_stop()) {
504 if (tout && tout <= 20)
505 __set_current_state(TASK_KILLABLE);
506 else
507 __set_current_state(TASK_INTERRUPTIBLE);
Brian Foster8375f922012-06-28 06:52:56 -0400508
509 spin_lock(&ailp->xa_lock);
510
511 /*
512 * Idle if the AIL is empty and we are not racing with a target
513 * update. We check the AIL after we set the task to a sleep
514 * state to guarantee that we either catch an xa_target update
515 * or that a wake_up resets the state to TASK_RUNNING.
516 * Otherwise, we run the risk of sleeping indefinitely.
517 *
518 * The barrier matches the xa_target update in xfs_ail_push().
519 */
520 smp_rmb();
521 if (!xfs_ail_min(ailp) &&
522 ailp->xa_target == ailp->xa_target_prev) {
523 spin_unlock(&ailp->xa_lock);
524 schedule();
525 tout = 0;
526 continue;
527 }
528 spin_unlock(&ailp->xa_lock);
529
530 if (tout)
531 schedule_timeout(msecs_to_jiffies(tout));
532
533 __set_current_state(TASK_RUNNING);
Christoph Hellwig00308072011-10-11 11:14:10 -0400534
535 try_to_freeze();
536
537 tout = xfsaild_push(ailp);
538 }
539
540 return 0;
Dave Chinner453eac82010-01-11 11:49:58 +0000541}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000543/*
544 * This routine is called to move the tail of the AIL forward. It does this by
545 * trying to flush items in the AIL whose lsns are below the given
546 * threshold_lsn.
547 *
548 * The push is run asynchronously in a workqueue, which means the caller needs
549 * to handle waiting on the async flush for space to become available.
550 * We don't want to interrupt any push that is in progress, hence we only queue
551 * work if we set the pushing bit approriately.
552 *
553 * We do this unlocked - we only need to know whether there is anything in the
554 * AIL at the time we are called. We don't need to access the contents of
555 * any of the objects, so the lock is not needed.
556 */
557void
Dave Chinnerfd074842011-04-08 12:45:07 +1000558xfs_ail_push(
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000559 struct xfs_ail *ailp,
560 xfs_lsn_t threshold_lsn)
561{
562 xfs_log_item_t *lip;
563
564 lip = xfs_ail_min(ailp);
565 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
566 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
567 return;
568
569 /*
570 * Ensure that the new target is noticed in push code before it clears
571 * the XFS_AIL_PUSHING_BIT.
572 */
573 smp_wmb();
Dave Chinnerfe0da762011-05-06 02:54:07 +0000574 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
Christoph Hellwig00308072011-10-11 11:14:10 -0400575 smp_wmb();
576
577 wake_up_process(ailp->xa_task);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000578}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580/*
Dave Chinnerfd074842011-04-08 12:45:07 +1000581 * Push out all items in the AIL immediately
582 */
583void
584xfs_ail_push_all(
585 struct xfs_ail *ailp)
586{
587 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
588
589 if (threshold_lsn)
590 xfs_ail_push(ailp, threshold_lsn);
591}
592
593/*
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000594 * Push out all items in the AIL immediately and wait until the AIL is empty.
595 */
596void
597xfs_ail_push_all_sync(
598 struct xfs_ail *ailp)
599{
600 struct xfs_log_item *lip;
601 DEFINE_WAIT(wait);
602
603 spin_lock(&ailp->xa_lock);
604 while ((lip = xfs_ail_max(ailp)) != NULL) {
605 prepare_to_wait(&ailp->xa_empty, &wait, TASK_UNINTERRUPTIBLE);
606 ailp->xa_target = lip->li_lsn;
607 wake_up_process(ailp->xa_task);
608 spin_unlock(&ailp->xa_lock);
609 schedule();
610 spin_lock(&ailp->xa_lock);
611 }
612 spin_unlock(&ailp->xa_lock);
613
614 finish_wait(&ailp->xa_empty, &wait);
615}
616
617/*
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100618 * xfs_trans_ail_update - bulk AIL insertion operation.
619 *
620 * @xfs_trans_ail_update takes an array of log items that all need to be
621 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
622 * be added. Otherwise, it will be repositioned by removing it and re-adding
623 * it to the AIL. If we move the first item in the AIL, update the log tail to
624 * match the new minimum LSN in the AIL.
625 *
626 * This function takes the AIL lock once to execute the update operations on
627 * all the items in the array, and as such should not be called with the AIL
628 * lock held. As a result, once we have the AIL lock, we need to check each log
629 * item LSN to confirm it needs to be moved forward in the AIL.
630 *
631 * To optimise the insert operation, we delete all the items from the AIL in
632 * the first pass, moving them into a temporary list, then splice the temporary
633 * list into the correct position in the AIL. This avoids needing to do an
634 * insert operation on every item.
635 *
636 * This function must be called with the AIL lock held. The lock is dropped
637 * before returning.
638 */
639void
640xfs_trans_ail_update_bulk(
641 struct xfs_ail *ailp,
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000642 struct xfs_ail_cursor *cur,
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100643 struct xfs_log_item **log_items,
644 int nr_items,
645 xfs_lsn_t lsn) __releases(ailp->xa_lock)
646{
647 xfs_log_item_t *mlip;
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100648 int mlip_changed = 0;
649 int i;
650 LIST_HEAD(tmp);
651
Alex Eldere44f4112011-07-22 16:04:41 +0000652 ASSERT(nr_items > 0); /* Not required, but true. */
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100653 mlip = xfs_ail_min(ailp);
654
655 for (i = 0; i < nr_items; i++) {
656 struct xfs_log_item *lip = log_items[i];
657 if (lip->li_flags & XFS_LI_IN_AIL) {
658 /* check if we really need to move the item */
659 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
660 continue;
661
662 xfs_ail_delete(ailp, lip);
663 if (mlip == lip)
664 mlip_changed = 1;
665 } else {
666 lip->li_flags |= XFS_LI_IN_AIL;
667 }
668 lip->li_lsn = lsn;
669 list_add(&lip->li_ail, &tmp);
670 }
671
Alex Eldere44f4112011-07-22 16:04:41 +0000672 if (!list_empty(&tmp))
673 xfs_ail_splice(ailp, cur, &tmp, lsn);
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000674
Christoph Hellwig1c304622012-04-23 15:58:33 +1000675 if (mlip_changed) {
676 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
677 xlog_assign_tail_lsn_locked(ailp->xa_mount);
678 spin_unlock(&ailp->xa_lock);
679
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +0000680 xfs_log_space_wake(ailp->xa_mount);
Christoph Hellwig1c304622012-04-23 15:58:33 +1000681 } else {
682 spin_unlock(&ailp->xa_lock);
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000683 }
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100684}
685
686/*
Dave Chinner30136832010-12-20 12:03:17 +1100687 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
688 *
689 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
690 * removed from the AIL. The caller is already holding the AIL lock, and done
691 * all the checks necessary to ensure the items passed in via @log_items are
692 * ready for deletion. This includes checking that the items are in the AIL.
693 *
694 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
695 * flag from the item and reset the item's lsn to 0. If we remove the first
696 * item in the AIL, update the log tail to match the new minimum LSN in the
697 * AIL.
698 *
699 * This function will not drop the AIL lock until all items are removed from
700 * the AIL to minimise the amount of lock traffic on the AIL. This does not
701 * greatly increase the AIL hold time, but does significantly reduce the amount
702 * of traffic on the lock, especially during IO completion.
703 *
704 * This function must be called with the AIL lock held. The lock is dropped
705 * before returning.
706 */
707void
708xfs_trans_ail_delete_bulk(
709 struct xfs_ail *ailp,
710 struct xfs_log_item **log_items,
Dave Chinner04913fd2012-04-23 15:58:41 +1000711 int nr_items,
712 int shutdown_type) __releases(ailp->xa_lock)
Dave Chinner30136832010-12-20 12:03:17 +1100713{
714 xfs_log_item_t *mlip;
Dave Chinner30136832010-12-20 12:03:17 +1100715 int mlip_changed = 0;
716 int i;
717
718 mlip = xfs_ail_min(ailp);
719
720 for (i = 0; i < nr_items; i++) {
721 struct xfs_log_item *lip = log_items[i];
722 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
723 struct xfs_mount *mp = ailp->xa_mount;
724
725 spin_unlock(&ailp->xa_lock);
726 if (!XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner6a19d932011-03-07 10:02:35 +1100727 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
Dave Chinner30136832010-12-20 12:03:17 +1100728 "%s: attempting to delete a log item that is not in the AIL",
729 __func__);
Dave Chinner04913fd2012-04-23 15:58:41 +1000730 xfs_force_shutdown(mp, shutdown_type);
Dave Chinner30136832010-12-20 12:03:17 +1100731 }
732 return;
733 }
734
735 xfs_ail_delete(ailp, lip);
736 lip->li_flags &= ~XFS_LI_IN_AIL;
737 lip->li_lsn = 0;
738 if (mlip == lip)
739 mlip_changed = 1;
740 }
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000741
Christoph Hellwig1c304622012-04-23 15:58:33 +1000742 if (mlip_changed) {
743 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
744 xlog_assign_tail_lsn_locked(ailp->xa_mount);
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000745 if (list_empty(&ailp->xa_ail))
746 wake_up_all(&ailp->xa_empty);
Christoph Hellwig1c304622012-04-23 15:58:33 +1000747 spin_unlock(&ailp->xa_lock);
748
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +0000749 xfs_log_space_wake(ailp->xa_mount);
Christoph Hellwig1c304622012-04-23 15:58:33 +1000750 } else {
751 spin_unlock(&ailp->xa_lock);
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000752 }
Dave Chinner30136832010-12-20 12:03:17 +1100753}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
David Chinner249a8c12008-02-05 12:13:32 +1100755int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756xfs_trans_ail_init(
757 xfs_mount_t *mp)
758{
David Chinner82fa9012008-10-30 17:38:26 +1100759 struct xfs_ail *ailp;
760
761 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
762 if (!ailp)
763 return ENOMEM;
764
765 ailp->xa_mount = mp;
766 INIT_LIST_HEAD(&ailp->xa_ail);
Dave Chinneraf3e4022011-07-18 03:40:18 +0000767 INIT_LIST_HEAD(&ailp->xa_cursors);
David Chinnerc7e8f262008-10-30 17:39:23 +1100768 spin_lock_init(&ailp->xa_lock);
Christoph Hellwig43ff2122012-04-23 15:58:39 +1000769 INIT_LIST_HEAD(&ailp->xa_buf_list);
Christoph Hellwig211e4d42012-04-23 15:58:34 +1000770 init_waitqueue_head(&ailp->xa_empty);
Christoph Hellwig00308072011-10-11 11:14:10 -0400771
772 ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
773 ailp->xa_mount->m_fsname);
774 if (IS_ERR(ailp->xa_task))
775 goto out_free_ailp;
776
David Chinner27d8d5f2008-10-30 17:38:39 +1100777 mp->m_ail = ailp;
778 return 0;
Christoph Hellwig00308072011-10-11 11:14:10 -0400779
780out_free_ailp:
781 kmem_free(ailp);
782 return ENOMEM;
David Chinner249a8c12008-02-05 12:13:32 +1100783}
784
785void
786xfs_trans_ail_destroy(
787 xfs_mount_t *mp)
788{
David Chinner82fa9012008-10-30 17:38:26 +1100789 struct xfs_ail *ailp = mp->m_ail;
790
Christoph Hellwig00308072011-10-11 11:14:10 -0400791 kthread_stop(ailp->xa_task);
David Chinner82fa9012008-10-30 17:38:26 +1100792 kmem_free(ailp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}