blob: ed9252bcdac9c351a79020451cf76eb4287992f7 [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"
Christoph Hellwig9e4c1092011-10-11 15:14:11 +000029#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_error.h"
31
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
59#ifdef XFS_TRANS_DEBUG
60 /*
61 * Walk the list checking lsn ordering, and that every entry has the
62 * XFS_LI_IN_AIL flag set. This is really expensive, so only do it
63 * when specifically debugging the transaction subsystem.
64 */
65 prev_lip = list_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
66 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
67 if (&prev_lip->li_ail != &ailp->xa_ail)
68 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
69 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
70 prev_lip = lip;
71 }
72#endif /* XFS_TRANS_DEBUG */
73}
74#else /* !DEBUG */
David Chinnerde08dbc2008-02-05 12:13:38 +110075#define xfs_ail_check(a,l)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#endif /* DEBUG */
77
Dave Chinnercd4a3c52011-04-08 12:45:07 +100078/*
79 * Return a pointer to the first item in the AIL. If the AIL is empty, then
80 * return NULL.
81 */
82static xfs_log_item_t *
83xfs_ail_min(
84 struct xfs_ail *ailp)
85{
86 if (list_empty(&ailp->xa_ail))
87 return NULL;
88
89 return list_first_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
90}
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Dave Chinnerfd074842011-04-08 12:45:07 +100092 /*
93 * Return a pointer to the last item in the AIL. If the AIL is empty, then
94 * return NULL.
95 */
96static xfs_log_item_t *
97xfs_ail_max(
98 struct xfs_ail *ailp)
99{
100 if (list_empty(&ailp->xa_ail))
101 return NULL;
102
103 return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
104}
105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106/*
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000107 * Return a pointer to the item which follows the given item in the AIL. If
108 * the given item is the last item in the list, then return NULL.
109 */
110static xfs_log_item_t *
111xfs_ail_next(
112 struct xfs_ail *ailp,
113 xfs_log_item_t *lip)
114{
115 if (lip->li_ail.next == &ailp->xa_ail)
116 return NULL;
117
118 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
119}
120
121/*
122 * This is called by the log manager code to determine the LSN of the tail of
123 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
124 * is empty, then this function returns 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 *
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000126 * We need the AIL lock in order to get a coherent read of the lsn of the last
127 * item in the AIL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 */
129xfs_lsn_t
Dave Chinnerfd074842011-04-08 12:45:07 +1000130xfs_ail_min_lsn(
David Chinner5b00f142008-10-30 17:39:00 +1100131 struct xfs_ail *ailp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000133 xfs_lsn_t lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
David Chinnerc7e8f262008-10-30 17:39:23 +1100136 spin_lock(&ailp->xa_lock);
David Chinner5b00f142008-10-30 17:39:00 +1100137 lip = xfs_ail_min(ailp);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000138 if (lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 lsn = lip->li_lsn;
David Chinnerc7e8f262008-10-30 17:39:23 +1100140 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
142 return lsn;
143}
144
145/*
Dave Chinnerfd074842011-04-08 12:45:07 +1000146 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
147 */
148static xfs_lsn_t
149xfs_ail_max_lsn(
150 struct xfs_ail *ailp)
151{
152 xfs_lsn_t lsn = 0;
153 xfs_log_item_t *lip;
154
155 spin_lock(&ailp->xa_lock);
156 lip = xfs_ail_max(ailp);
157 if (lip)
158 lsn = lip->li_lsn;
159 spin_unlock(&ailp->xa_lock);
160
161 return lsn;
162}
163
164/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000165 * The cursor keeps track of where our current traversal is up to by tracking
166 * the next item in the list for us. However, for this to be safe, removing an
167 * object from the AIL needs to invalidate any cursor that points to it. hence
168 * the traversal cursor needs to be linked to the struct xfs_ail so that
169 * deletion can search all the active cursors for invalidation.
David Chinner27d8d5f2008-10-30 17:38:39 +1100170 */
David Chinner5b00f142008-10-30 17:39:00 +1100171STATIC void
David Chinner27d8d5f2008-10-30 17:38:39 +1100172xfs_trans_ail_cursor_init(
173 struct xfs_ail *ailp,
174 struct xfs_ail_cursor *cur)
175{
176 cur->item = NULL;
Dave Chinneraf3e4022011-07-18 03:40:18 +0000177 list_add_tail(&cur->list, &ailp->xa_cursors);
David Chinner27d8d5f2008-10-30 17:38:39 +1100178}
179
180/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000181 * Get the next item in the traversal and advance the cursor. If the cursor
182 * was invalidated (indicated by a lip of 1), restart the traversal.
David Chinner27d8d5f2008-10-30 17:38:39 +1100183 */
David Chinner5b00f142008-10-30 17:39:00 +1100184struct xfs_log_item *
David Chinner27d8d5f2008-10-30 17:38:39 +1100185xfs_trans_ail_cursor_next(
186 struct xfs_ail *ailp,
187 struct xfs_ail_cursor *cur)
188{
189 struct xfs_log_item *lip = cur->item;
190
191 if ((__psint_t)lip & 1)
192 lip = xfs_ail_min(ailp);
Dave Chinner16b59022011-07-18 03:40:17 +0000193 if (lip)
194 cur->item = xfs_ail_next(ailp, lip);
David Chinner27d8d5f2008-10-30 17:38:39 +1100195 return lip;
196}
197
198/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000199 * When the traversal is complete, we need to remove the cursor from the list
200 * of traversing cursors.
David Chinner27d8d5f2008-10-30 17:38:39 +1100201 */
202void
203xfs_trans_ail_cursor_done(
204 struct xfs_ail *ailp,
Dave Chinneraf3e4022011-07-18 03:40:18 +0000205 struct xfs_ail_cursor *cur)
David Chinner27d8d5f2008-10-30 17:38:39 +1100206{
Dave Chinneraf3e4022011-07-18 03:40:18 +0000207 cur->item = NULL;
208 list_del_init(&cur->list);
David Chinner27d8d5f2008-10-30 17:38:39 +1100209}
210
211/*
Dave Chinneraf3e4022011-07-18 03:40:18 +0000212 * Invalidate any cursor that is pointing to this item. This is called when an
213 * item is removed from the AIL. Any cursor pointing to this object is now
214 * invalid and the traversal needs to be terminated so it doesn't reference a
215 * freed object. We set the low bit of the cursor item pointer so we can
216 * distinguish between an invalidation and the end of the list when getting the
217 * next item from the cursor.
David Chinner5b00f142008-10-30 17:39:00 +1100218 */
219STATIC void
220xfs_trans_ail_cursor_clear(
221 struct xfs_ail *ailp,
222 struct xfs_log_item *lip)
223{
224 struct xfs_ail_cursor *cur;
225
Dave Chinneraf3e4022011-07-18 03:40:18 +0000226 list_for_each_entry(cur, &ailp->xa_cursors, list) {
David Chinner5b00f142008-10-30 17:39:00 +1100227 if (cur->item == lip)
228 cur->item = (struct xfs_log_item *)
229 ((__psint_t)cur->item | 1);
230 }
231}
232
233/*
Dave Chinner16b59022011-07-18 03:40:17 +0000234 * Find the first item in the AIL with the given @lsn by searching in ascending
235 * LSN order and initialise the cursor to point to the next item for a
236 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
237 * first item in the AIL. Returns NULL if the list is empty.
David Chinner249a8c12008-02-05 12:13:32 +1100238 */
David Chinner5b00f142008-10-30 17:39:00 +1100239xfs_log_item_t *
240xfs_trans_ail_cursor_first(
David Chinner27d8d5f2008-10-30 17:38:39 +1100241 struct xfs_ail *ailp,
242 struct xfs_ail_cursor *cur,
243 xfs_lsn_t lsn)
David Chinner249a8c12008-02-05 12:13:32 +1100244{
David Chinner27d8d5f2008-10-30 17:38:39 +1100245 xfs_log_item_t *lip;
David Chinner249a8c12008-02-05 12:13:32 +1100246
David Chinner5b00f142008-10-30 17:39:00 +1100247 xfs_trans_ail_cursor_init(ailp, cur);
Dave Chinner16b59022011-07-18 03:40:17 +0000248
249 if (lsn == 0) {
250 lip = xfs_ail_min(ailp);
David Chinner5b00f142008-10-30 17:39:00 +1100251 goto out;
Dave Chinner16b59022011-07-18 03:40:17 +0000252 }
David Chinner249a8c12008-02-05 12:13:32 +1100253
David Chinner27d8d5f2008-10-30 17:38:39 +1100254 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
David Chinner5b00f142008-10-30 17:39:00 +1100255 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
David Chinner7ee49ac2008-10-30 18:26:51 +1100256 goto out;
Josef 'Jeff' Sipek535f6b32008-03-27 17:58:27 +1100257 }
Dave Chinner16b59022011-07-18 03:40:17 +0000258 return NULL;
259
David Chinner5b00f142008-10-30 17:39:00 +1100260out:
Dave Chinner16b59022011-07-18 03:40:17 +0000261 if (lip)
262 cur->item = xfs_ail_next(ailp, lip);
David Chinner5b00f142008-10-30 17:39:00 +1100263 return lip;
David Chinner249a8c12008-02-05 12:13:32 +1100264}
265
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000266static struct xfs_log_item *
267__xfs_trans_ail_cursor_last(
268 struct xfs_ail *ailp,
269 xfs_lsn_t lsn)
270{
271 xfs_log_item_t *lip;
272
273 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
274 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
275 return lip;
276 }
277 return NULL;
278}
279
280/*
Dave Chinner16b59022011-07-18 03:40:17 +0000281 * Find the last item in the AIL with the given @lsn by searching in descending
282 * LSN order and initialise the cursor to point to that item. If there is no
283 * item with the value of @lsn, then it sets the cursor to the last item with an
284 * LSN lower than @lsn. Returns NULL if the list is empty.
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000285 */
286struct xfs_log_item *
287xfs_trans_ail_cursor_last(
288 struct xfs_ail *ailp,
289 struct xfs_ail_cursor *cur,
290 xfs_lsn_t lsn)
291{
292 xfs_trans_ail_cursor_init(ailp, cur);
293 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
294 return cur->item;
295}
296
297/*
Dave Chinner16b59022011-07-18 03:40:17 +0000298 * Splice the log item list into the AIL at the given LSN. We splice to the
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000299 * tail of the given LSN to maintain insert order for push traversals. The
300 * cursor is optional, allowing repeated updates to the same LSN to avoid
Alex Eldere44f4112011-07-22 16:04:41 +0000301 * repeated traversals. This should not be called with an empty list.
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000302 */
303static void
304xfs_ail_splice(
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000305 struct xfs_ail *ailp,
306 struct xfs_ail_cursor *cur,
307 struct list_head *list,
308 xfs_lsn_t lsn)
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000309{
Alex Eldere44f4112011-07-22 16:04:41 +0000310 struct xfs_log_item *lip;
311
312 ASSERT(!list_empty(list));
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000313
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000314 /*
Alex Eldere44f4112011-07-22 16:04:41 +0000315 * Use the cursor to determine the insertion point if one is
316 * provided. If not, or if the one we got is not valid,
317 * find the place in the AIL where the items belong.
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000318 */
Alex Eldere44f4112011-07-22 16:04:41 +0000319 lip = cur ? cur->item : NULL;
320 if (!lip || (__psint_t) lip & 1)
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000321 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
322
Alex Eldere44f4112011-07-22 16:04:41 +0000323 /*
324 * If a cursor is provided, we know we're processing the AIL
325 * in lsn order, and future items to be spliced in will
326 * follow the last one being inserted now. Update the
327 * cursor to point to that last item, now while we have a
328 * reliable pointer to it.
329 */
330 if (cur)
331 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000332
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000333 /*
Alex Eldere44f4112011-07-22 16:04:41 +0000334 * Finally perform the splice. Unless the AIL was empty,
335 * lip points to the item in the AIL _after_ which the new
336 * items should go. If lip is null the AIL was empty, so
337 * the new items go at the head of the AIL.
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000338 */
Alex Eldere44f4112011-07-22 16:04:41 +0000339 if (lip)
340 list_splice(list, &lip->li_ail);
341 else
342 list_splice(list, &ailp->xa_ail);
Dave Chinnercd4a3c52011-04-08 12:45:07 +1000343}
344
345/*
346 * Delete the given item from the AIL. Return a pointer to the item.
347 */
348static void
349xfs_ail_delete(
350 struct xfs_ail *ailp,
351 xfs_log_item_t *lip)
352{
353 xfs_ail_check(ailp, lip);
354 list_del(&lip->li_ail);
355 xfs_trans_ail_cursor_clear(ailp, lip);
356}
357
Christoph Hellwig00308072011-10-11 11:14:10 -0400358static long
359xfsaild_push(
360 struct xfs_ail *ailp)
David Chinner249a8c12008-02-05 12:13:32 +1100361{
Dave Chinner9e7004e2011-05-06 02:54:05 +0000362 xfs_mount_t *mp = ailp->xa_mount;
Dave Chinneraf3e4022011-07-18 03:40:18 +0000363 struct xfs_ail_cursor cur;
Dave Chinner9e7004e2011-05-06 02:54:05 +0000364 xfs_log_item_t *lip;
365 xfs_lsn_t lsn;
Dave Chinnerfe0da762011-05-06 02:54:07 +0000366 xfs_lsn_t target;
Dave Chinner9e7004e2011-05-06 02:54:05 +0000367 long tout = 10;
Dave Chinner9e7004e2011-05-06 02:54:05 +0000368 int stuck = 0;
369 int count = 0;
370 int push_xfsbufd = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Dave Chinner670ce932011-09-30 04:45:03 +0000372 /*
373 * If last time we ran we encountered pinned items, force the log first
374 * and wait for it before pushing again.
375 */
David Chinnerc7e8f262008-10-30 17:39:23 +1100376 spin_lock(&ailp->xa_lock);
Dave Chinner670ce932011-09-30 04:45:03 +0000377 if (ailp->xa_last_pushed_lsn == 0 && ailp->xa_log_flush &&
378 !list_empty(&ailp->xa_ail)) {
379 ailp->xa_log_flush = 0;
380 spin_unlock(&ailp->xa_lock);
381 XFS_STATS_INC(xs_push_ail_flush);
382 xfs_log_force(mp, XFS_LOG_SYNC);
383 spin_lock(&ailp->xa_lock);
384 }
385
Dave Chinnerfe0da762011-05-06 02:54:07 +0000386 target = ailp->xa_target;
Dave Chinneraf3e4022011-07-18 03:40:18 +0000387 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
David Chinner249a8c12008-02-05 12:13:32 +1100388 if (!lip || XFS_FORCED_SHUTDOWN(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 /*
David Chinner249a8c12008-02-05 12:13:32 +1100390 * AIL is empty or our push has reached the end.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 */
Dave Chinneraf3e4022011-07-18 03:40:18 +0000392 xfs_trans_ail_cursor_done(ailp, &cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100393 spin_unlock(&ailp->xa_lock);
Dave Chinner9e7004e2011-05-06 02:54:05 +0000394 goto out_done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 }
396
397 XFS_STATS_INC(xs_push_ail);
398
399 /*
400 * While the item we are looking at is below the given threshold
David Chinner249a8c12008-02-05 12:13:32 +1100401 * try to flush it out. We'd like not to stop until we've at least
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 * tried to push on everything in the AIL with an LSN less than
David Chinner249a8c12008-02-05 12:13:32 +1100403 * the given threshold.
404 *
405 * However, we will stop after a certain number of pushes and wait
406 * for a reduced timeout to fire before pushing further. This
407 * prevents use from spinning when we can't do anything or there is
408 * lots of contention on the AIL lists.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 */
David Chinner249a8c12008-02-05 12:13:32 +1100410 lsn = lip->li_lsn;
Dave Chinner50e86682011-05-06 02:54:06 +0000411 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
David Chinner249a8c12008-02-05 12:13:32 +1100412 int lock_result;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 /*
David Chinner249a8c12008-02-05 12:13:32 +1100414 * If we can lock the item without sleeping, unlock the AIL
415 * lock and flush the item. Then re-grab the AIL lock so we
416 * can look for the next item on the AIL. List changes are
417 * handled by the AIL lookup functions internally
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 *
David Chinner249a8c12008-02-05 12:13:32 +1100419 * If we can't lock the item, either its holder will flush it
420 * or it is already being flushed or it is being relogged. In
421 * any of these case it is being taken care of and we can just
422 * skip to the next item in the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 */
424 lock_result = IOP_TRYLOCK(lip);
David Chinnerc7e8f262008-10-30 17:39:23 +1100425 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 switch (lock_result) {
David Chinner249a8c12008-02-05 12:13:32 +1100427 case XFS_ITEM_SUCCESS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 XFS_STATS_INC(xs_push_ail_success);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000429 trace_xfs_ail_push(lip);
430
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 IOP_PUSH(lip);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000432 ailp->xa_last_pushed_lsn = lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 break;
434
David Chinner249a8c12008-02-05 12:13:32 +1100435 case XFS_ITEM_PUSHBUF:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 XFS_STATS_INC(xs_push_ail_pushbuf);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000437 trace_xfs_ail_pushbuf(lip);
Christoph Hellwig17b38472011-10-11 15:14:09 +0000438
439 if (!IOP_PUSHBUF(lip)) {
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000440 trace_xfs_ail_pushbuf_pinned(lip);
Christoph Hellwig17b38472011-10-11 15:14:09 +0000441 stuck++;
Alex Elder2900b332011-10-18 20:00:14 +0000442 ailp->xa_log_flush++;
Christoph Hellwig17b38472011-10-11 15:14:09 +0000443 } else {
444 ailp->xa_last_pushed_lsn = lsn;
445 }
Dave Chinnerd808f612010-02-02 10:13:42 +1100446 push_xfsbufd = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 break;
448
David Chinner249a8c12008-02-05 12:13:32 +1100449 case XFS_ITEM_PINNED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 XFS_STATS_INC(xs_push_ail_pinned);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000451 trace_xfs_ail_pinned(lip);
452
David Chinner249a8c12008-02-05 12:13:32 +1100453 stuck++;
Dave Chinner670ce932011-09-30 04:45:03 +0000454 ailp->xa_log_flush++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 break;
456
David Chinner249a8c12008-02-05 12:13:32 +1100457 case XFS_ITEM_LOCKED:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 XFS_STATS_INC(xs_push_ail_locked);
Christoph Hellwig9e4c1092011-10-11 15:14:11 +0000459 trace_xfs_ail_locked(lip);
David Chinner249a8c12008-02-05 12:13:32 +1100460 stuck++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 break;
462
David Chinner249a8c12008-02-05 12:13:32 +1100463 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 ASSERT(0);
465 break;
466 }
467
David Chinnerc7e8f262008-10-30 17:39:23 +1100468 spin_lock(&ailp->xa_lock);
David Chinner249a8c12008-02-05 12:13:32 +1100469 /* should we bother continuing? */
470 if (XFS_FORCED_SHUTDOWN(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 break;
David Chinner249a8c12008-02-05 12:13:32 +1100472 ASSERT(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
David Chinner249a8c12008-02-05 12:13:32 +1100474 count++;
475
476 /*
477 * Are there too many items we can't do anything with?
478 * If we we are skipping too many items because we can't flush
479 * them or they are already being flushed, we back off and
480 * given them time to complete whatever operation is being
481 * done. i.e. remove pressure from the AIL while we can't make
482 * progress so traversals don't slow down further inserts and
483 * removals to/from the AIL.
484 *
485 * The value of 100 is an arbitrary magic number based on
486 * observation.
487 */
488 if (stuck > 100)
489 break;
490
Dave Chinneraf3e4022011-07-18 03:40:18 +0000491 lip = xfs_trans_ail_cursor_next(ailp, &cur);
David Chinner249a8c12008-02-05 12:13:32 +1100492 if (lip == NULL)
493 break;
David Chinner249a8c12008-02-05 12:13:32 +1100494 lsn = lip->li_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
Dave Chinneraf3e4022011-07-18 03:40:18 +0000496 xfs_trans_ail_cursor_done(ailp, &cur);
David Chinnerc7e8f262008-10-30 17:39:23 +1100497 spin_unlock(&ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Dave Chinnerd808f612010-02-02 10:13:42 +1100499 if (push_xfsbufd) {
500 /* we've got delayed write buffers to flush */
501 wake_up_process(mp->m_ddev_targp->bt_task);
502 }
503
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000504 /* assume we have more work to do in a short while */
Dave Chinner9e7004e2011-05-06 02:54:05 +0000505out_done:
David Chinner92d9cd12008-03-06 13:45:10 +1100506 if (!count) {
507 /* We're past our target or empty, so idle */
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000508 ailp->xa_last_pushed_lsn = 0;
Dave Chinner670ce932011-09-30 04:45:03 +0000509 ailp->xa_log_flush = 0;
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000510
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000511 tout = 50;
David Chinner92d9cd12008-03-06 13:45:10 +1100512 } else if (XFS_LSN_CMP(lsn, target) >= 0) {
513 /*
514 * We reached the target so wait a bit longer for I/O to
515 * complete and remove pushed items from the AIL before we
516 * start the next scan from the start of the AIL.
517 */
Dave Chinner453eac82010-01-11 11:49:58 +0000518 tout = 50;
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000519 ailp->xa_last_pushed_lsn = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +1100520 } else if ((stuck * 100) / count > 90) {
David Chinner249a8c12008-02-05 12:13:32 +1100521 /*
522 * Either there is a lot of contention on the AIL or we
523 * are stuck due to operations in progress. "Stuck" in this
524 * case is defined as >90% of the items we tried to push
525 * were stuck.
526 *
527 * Backoff a bit more to allow some I/O to complete before
Dave Chinner670ce932011-09-30 04:45:03 +0000528 * restarting from the start of the AIL. This prevents us
529 * from spinning on the same items, and if they are pinned will
530 * all the restart to issue a log force to unpin the stuck
531 * items.
David Chinner249a8c12008-02-05 12:13:32 +1100532 */
Dave Chinner453eac82010-01-11 11:49:58 +0000533 tout = 20;
Dave Chinner670ce932011-09-30 04:45:03 +0000534 ailp->xa_last_pushed_lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000536
Christoph Hellwig00308072011-10-11 11:14:10 -0400537 return tout;
538}
539
540static int
541xfsaild(
542 void *data)
543{
544 struct xfs_ail *ailp = data;
545 long tout = 0; /* milliseconds */
546
547 while (!kthread_should_stop()) {
548 if (tout && tout <= 20)
549 __set_current_state(TASK_KILLABLE);
550 else
551 __set_current_state(TASK_INTERRUPTIBLE);
552 schedule_timeout(tout ?
553 msecs_to_jiffies(tout) : MAX_SCHEDULE_TIMEOUT);
554
555 try_to_freeze();
556
557 tout = xfsaild_push(ailp);
558 }
559
560 return 0;
Dave Chinner453eac82010-01-11 11:49:58 +0000561}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000563/*
564 * This routine is called to move the tail of the AIL forward. It does this by
565 * trying to flush items in the AIL whose lsns are below the given
566 * threshold_lsn.
567 *
568 * The push is run asynchronously in a workqueue, which means the caller needs
569 * to handle waiting on the async flush for space to become available.
570 * We don't want to interrupt any push that is in progress, hence we only queue
571 * work if we set the pushing bit approriately.
572 *
573 * We do this unlocked - we only need to know whether there is anything in the
574 * AIL at the time we are called. We don't need to access the contents of
575 * any of the objects, so the lock is not needed.
576 */
577void
Dave Chinnerfd074842011-04-08 12:45:07 +1000578xfs_ail_push(
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000579 struct xfs_ail *ailp,
580 xfs_lsn_t threshold_lsn)
581{
582 xfs_log_item_t *lip;
583
584 lip = xfs_ail_min(ailp);
585 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
586 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
587 return;
588
589 /*
590 * Ensure that the new target is noticed in push code before it clears
591 * the XFS_AIL_PUSHING_BIT.
592 */
593 smp_wmb();
Dave Chinnerfe0da762011-05-06 02:54:07 +0000594 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
Christoph Hellwig00308072011-10-11 11:14:10 -0400595 smp_wmb();
596
597 wake_up_process(ailp->xa_task);
Dave Chinner0bf6a5b2011-04-08 12:45:07 +1000598}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600/*
Dave Chinnerfd074842011-04-08 12:45:07 +1000601 * Push out all items in the AIL immediately
602 */
603void
604xfs_ail_push_all(
605 struct xfs_ail *ailp)
606{
607 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
608
609 if (threshold_lsn)
610 xfs_ail_push(ailp, threshold_lsn);
611}
612
613/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 * This is to be called when an item is unlocked that may have
615 * been in the AIL. It will wake up the first member of the AIL
616 * wait list if this item's unlocking might allow it to progress.
617 * If the item is in the AIL, then we need to get the AIL lock
618 * while doing our checking so we don't race with someone going
619 * to sleep waiting for this event in xfs_trans_push_ail().
620 */
621void
622xfs_trans_unlocked_item(
David Chinner783a2f62008-10-30 17:39:58 +1100623 struct xfs_ail *ailp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 xfs_log_item_t *lip)
625{
626 xfs_log_item_t *min_lip;
627
628 /*
629 * If we're forcibly shutting down, we may have
630 * unlocked log items arbitrarily. The last thing
631 * we want to do is to move the tail of the log
632 * over some potentially valid data.
633 */
634 if (!(lip->li_flags & XFS_LI_IN_AIL) ||
David Chinner783a2f62008-10-30 17:39:58 +1100635 XFS_FORCED_SHUTDOWN(ailp->xa_mount)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 return;
637 }
638
639 /*
640 * This is the one case where we can call into xfs_ail_min()
641 * without holding the AIL lock because we only care about the
642 * case where we are at the tail of the AIL. If the object isn't
643 * at the tail, it doesn't matter what result we get back. This
644 * is slightly racy because since we were just unlocked, we could
645 * go to sleep between the call to xfs_ail_min and the call to
646 * xfs_log_move_tail, have someone else lock us, commit to us disk,
647 * move us out of the tail of the AIL, and then we wake up. However,
648 * the call to xfs_log_move_tail() doesn't do anything if there's
649 * not enough free space to wake people up so we're safe calling it.
650 */
David Chinner783a2f62008-10-30 17:39:58 +1100651 min_lip = xfs_ail_min(ailp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
653 if (min_lip == lip)
David Chinner783a2f62008-10-30 17:39:58 +1100654 xfs_log_move_tail(ailp->xa_mount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655} /* xfs_trans_unlocked_item */
656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657/*
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100658 * xfs_trans_ail_update - bulk AIL insertion operation.
659 *
660 * @xfs_trans_ail_update takes an array of log items that all need to be
661 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
662 * be added. Otherwise, it will be repositioned by removing it and re-adding
663 * it to the AIL. If we move the first item in the AIL, update the log tail to
664 * match the new minimum LSN in the AIL.
665 *
666 * This function takes the AIL lock once to execute the update operations on
667 * all the items in the array, and as such should not be called with the AIL
668 * lock held. As a result, once we have the AIL lock, we need to check each log
669 * item LSN to confirm it needs to be moved forward in the AIL.
670 *
671 * To optimise the insert operation, we delete all the items from the AIL in
672 * the first pass, moving them into a temporary list, then splice the temporary
673 * list into the correct position in the AIL. This avoids needing to do an
674 * insert operation on every item.
675 *
676 * This function must be called with the AIL lock held. The lock is dropped
677 * before returning.
678 */
679void
680xfs_trans_ail_update_bulk(
681 struct xfs_ail *ailp,
Dave Chinner1d8c95a2011-07-18 03:40:16 +0000682 struct xfs_ail_cursor *cur,
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100683 struct xfs_log_item **log_items,
684 int nr_items,
685 xfs_lsn_t lsn) __releases(ailp->xa_lock)
686{
687 xfs_log_item_t *mlip;
688 xfs_lsn_t tail_lsn;
689 int mlip_changed = 0;
690 int i;
691 LIST_HEAD(tmp);
692
Alex Eldere44f4112011-07-22 16:04:41 +0000693 ASSERT(nr_items > 0); /* Not required, but true. */
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100694 mlip = xfs_ail_min(ailp);
695
696 for (i = 0; i < nr_items; i++) {
697 struct xfs_log_item *lip = log_items[i];
698 if (lip->li_flags & XFS_LI_IN_AIL) {
699 /* check if we really need to move the item */
700 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
701 continue;
702
703 xfs_ail_delete(ailp, lip);
704 if (mlip == lip)
705 mlip_changed = 1;
706 } else {
707 lip->li_flags |= XFS_LI_IN_AIL;
708 }
709 lip->li_lsn = lsn;
710 list_add(&lip->li_ail, &tmp);
711 }
712
Alex Eldere44f4112011-07-22 16:04:41 +0000713 if (!list_empty(&tmp))
714 xfs_ail_splice(ailp, cur, &tmp, lsn);
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100715
716 if (!mlip_changed) {
717 spin_unlock(&ailp->xa_lock);
718 return;
719 }
720
721 /*
722 * It is not safe to access mlip after the AIL lock is dropped, so we
723 * must get a copy of li_lsn before we do so. This is especially
724 * important on 32-bit platforms where accessing and updating 64-bit
725 * values like li_lsn is not atomic.
726 */
727 mlip = xfs_ail_min(ailp);
728 tail_lsn = mlip->li_lsn;
729 spin_unlock(&ailp->xa_lock);
730 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
731}
732
733/*
Dave Chinner30136832010-12-20 12:03:17 +1100734 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
735 *
736 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
737 * removed from the AIL. The caller is already holding the AIL lock, and done
738 * all the checks necessary to ensure the items passed in via @log_items are
739 * ready for deletion. This includes checking that the items are in the AIL.
740 *
741 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
742 * flag from the item and reset the item's lsn to 0. If we remove the first
743 * item in the AIL, update the log tail to match the new minimum LSN in the
744 * AIL.
745 *
746 * This function will not drop the AIL lock until all items are removed from
747 * the AIL to minimise the amount of lock traffic on the AIL. This does not
748 * greatly increase the AIL hold time, but does significantly reduce the amount
749 * of traffic on the lock, especially during IO completion.
750 *
751 * This function must be called with the AIL lock held. The lock is dropped
752 * before returning.
753 */
754void
755xfs_trans_ail_delete_bulk(
756 struct xfs_ail *ailp,
757 struct xfs_log_item **log_items,
758 int nr_items) __releases(ailp->xa_lock)
759{
760 xfs_log_item_t *mlip;
761 xfs_lsn_t tail_lsn;
762 int mlip_changed = 0;
763 int i;
764
765 mlip = xfs_ail_min(ailp);
766
767 for (i = 0; i < nr_items; i++) {
768 struct xfs_log_item *lip = log_items[i];
769 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
770 struct xfs_mount *mp = ailp->xa_mount;
771
772 spin_unlock(&ailp->xa_lock);
773 if (!XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner6a19d932011-03-07 10:02:35 +1100774 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
Dave Chinner30136832010-12-20 12:03:17 +1100775 "%s: attempting to delete a log item that is not in the AIL",
776 __func__);
777 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
778 }
779 return;
780 }
781
782 xfs_ail_delete(ailp, lip);
783 lip->li_flags &= ~XFS_LI_IN_AIL;
784 lip->li_lsn = 0;
785 if (mlip == lip)
786 mlip_changed = 1;
787 }
788
789 if (!mlip_changed) {
790 spin_unlock(&ailp->xa_lock);
791 return;
792 }
793
794 /*
795 * It is not safe to access mlip after the AIL lock is dropped, so we
796 * must get a copy of li_lsn before we do so. This is especially
797 * important on 32-bit platforms where accessing and updating 64-bit
798 * values like li_lsn is not atomic. It is possible we've emptied the
799 * AIL here, so if that is the case, pass an LSN of 0 to the tail move.
800 */
801 mlip = xfs_ail_min(ailp);
802 tail_lsn = mlip ? mlip->li_lsn : 0;
803 spin_unlock(&ailp->xa_lock);
804 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
805}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 * The active item list (AIL) is a doubly linked list of log
809 * items sorted by ascending lsn. The base of the list is
810 * a forw/back pointer pair embedded in the xfs mount structure.
811 * The base is initialized with both pointers pointing to the
812 * base. This case always needs to be distinguished, because
813 * the base has no lsn to look at. We almost always insert
814 * at the end of the list, so on inserts we search from the
815 * end of the list to find where the new item belongs.
816 */
817
818/*
819 * Initialize the doubly linked list to point only to itself.
820 */
David Chinner249a8c12008-02-05 12:13:32 +1100821int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822xfs_trans_ail_init(
823 xfs_mount_t *mp)
824{
David Chinner82fa9012008-10-30 17:38:26 +1100825 struct xfs_ail *ailp;
826
827 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
828 if (!ailp)
829 return ENOMEM;
830
831 ailp->xa_mount = mp;
832 INIT_LIST_HEAD(&ailp->xa_ail);
Dave Chinneraf3e4022011-07-18 03:40:18 +0000833 INIT_LIST_HEAD(&ailp->xa_cursors);
David Chinnerc7e8f262008-10-30 17:39:23 +1100834 spin_lock_init(&ailp->xa_lock);
Christoph Hellwig00308072011-10-11 11:14:10 -0400835
836 ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
837 ailp->xa_mount->m_fsname);
838 if (IS_ERR(ailp->xa_task))
839 goto out_free_ailp;
840
David Chinner27d8d5f2008-10-30 17:38:39 +1100841 mp->m_ail = ailp;
842 return 0;
Christoph Hellwig00308072011-10-11 11:14:10 -0400843
844out_free_ailp:
845 kmem_free(ailp);
846 return ENOMEM;
David Chinner249a8c12008-02-05 12:13:32 +1100847}
848
849void
850xfs_trans_ail_destroy(
851 xfs_mount_t *mp)
852{
David Chinner82fa9012008-10-30 17:38:26 +1100853 struct xfs_ail *ailp = mp->m_ail;
854
Christoph Hellwig00308072011-10-11 11:14:10 -0400855 kthread_stop(ailp->xa_task);
David Chinner82fa9012008-10-30 17:38:26 +1100856 kmem_free(ailp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857}