Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 | 2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
c7e8f268 | 3 | * Copyright (c) 2008 Dave Chinner |
7b718769 | 4 | * All Rights Reserved. |
1da177e4 | 5 | * |
7b718769 NS |
6 | * This program is free software; you can redistribute it and/or |
7 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
8 | * published by the Free Software Foundation. |
9 | * | |
7b718769 NS |
10 | * 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. | |
1da177e4 | 14 | * |
7b718769 NS |
15 | * 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 | |
1da177e4 | 18 | */ |
1da177e4 | 19 | #include "xfs.h" |
a844f451 | 20 | #include "xfs_fs.h" |
1da177e4 | 21 | #include "xfs_types.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 LT |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" | |
da353b0d | 26 | #include "xfs_ag.h" |
1da177e4 LT |
27 | #include "xfs_mount.h" |
28 | #include "xfs_trans_priv.h" | |
29 | #include "xfs_error.h" | |
30 | ||
0bf6a5bd DC |
31 | struct workqueue_struct *xfs_ail_wq; /* AIL workqueue */ |
32 | ||
1da177e4 | 33 | #ifdef DEBUG |
cd4a3c50 DC |
34 | /* |
35 | * Check that the list is sorted as it should be. | |
36 | */ | |
37 | STATIC void | |
38 | xfs_ail_check( | |
39 | struct xfs_ail *ailp, | |
40 | xfs_log_item_t *lip) | |
41 | { | |
42 | xfs_log_item_t *prev_lip; | |
43 | ||
44 | if (list_empty(&ailp->xa_ail)) | |
45 | return; | |
46 | ||
47 | /* | |
48 | * Check the next and previous entries are valid. | |
49 | */ | |
50 | ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0); | |
51 | prev_lip = list_entry(lip->li_ail.prev, xfs_log_item_t, li_ail); | |
52 | if (&prev_lip->li_ail != &ailp->xa_ail) | |
53 | ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0); | |
54 | ||
55 | prev_lip = list_entry(lip->li_ail.next, xfs_log_item_t, li_ail); | |
56 | if (&prev_lip->li_ail != &ailp->xa_ail) | |
57 | ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0); | |
58 | ||
59 | ||
60 | #ifdef XFS_TRANS_DEBUG | |
61 | /* | |
62 | * Walk the list checking lsn ordering, and that every entry has the | |
63 | * XFS_LI_IN_AIL flag set. This is really expensive, so only do it | |
64 | * when specifically debugging the transaction subsystem. | |
65 | */ | |
66 | prev_lip = list_entry(&ailp->xa_ail, xfs_log_item_t, li_ail); | |
67 | list_for_each_entry(lip, &ailp->xa_ail, li_ail) { | |
68 | if (&prev_lip->li_ail != &ailp->xa_ail) | |
69 | ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0); | |
70 | ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0); | |
71 | prev_lip = lip; | |
72 | } | |
73 | #endif /* XFS_TRANS_DEBUG */ | |
74 | } | |
75 | #else /* !DEBUG */ | |
de08dbc1 | 76 | #define xfs_ail_check(a,l) |
1da177e4 LT |
77 | #endif /* DEBUG */ |
78 | ||
cd4a3c50 DC |
79 | /* |
80 | * Return a pointer to the first item in the AIL. If the AIL is empty, then | |
81 | * return NULL. | |
82 | */ | |
83 | static xfs_log_item_t * | |
84 | xfs_ail_min( | |
85 | struct xfs_ail *ailp) | |
86 | { | |
87 | if (list_empty(&ailp->xa_ail)) | |
88 | return NULL; | |
89 | ||
90 | return list_first_entry(&ailp->xa_ail, xfs_log_item_t, li_ail); | |
91 | } | |
92 | ||
fd074841 DC |
93 | /* |
94 | * Return a pointer to the last item in the AIL. If the AIL is empty, then | |
95 | * return NULL. | |
96 | */ | |
97 | static xfs_log_item_t * | |
98 | xfs_ail_max( | |
99 | struct xfs_ail *ailp) | |
100 | { | |
101 | if (list_empty(&ailp->xa_ail)) | |
102 | return NULL; | |
103 | ||
104 | return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail); | |
105 | } | |
106 | ||
cd4a3c50 DC |
107 | /* |
108 | * Return a pointer to the item which follows the given item in the AIL. If | |
109 | * the given item is the last item in the list, then return NULL. | |
110 | */ | |
111 | static xfs_log_item_t * | |
112 | xfs_ail_next( | |
113 | struct xfs_ail *ailp, | |
114 | xfs_log_item_t *lip) | |
115 | { | |
116 | if (lip->li_ail.next == &ailp->xa_ail) | |
117 | return NULL; | |
118 | ||
119 | return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail); | |
120 | } | |
1da177e4 LT |
121 | |
122 | /* | |
cd4a3c50 DC |
123 | * This is called by the log manager code to determine the LSN of the tail of |
124 | * the log. This is exactly the LSN of the first item in the AIL. If the AIL | |
125 | * is empty, then this function returns 0. | |
1da177e4 | 126 | * |
cd4a3c50 DC |
127 | * We need the AIL lock in order to get a coherent read of the lsn of the last |
128 | * item in the AIL. | |
1da177e4 LT |
129 | */ |
130 | xfs_lsn_t | |
fd074841 | 131 | xfs_ail_min_lsn( |
5b00f14f | 132 | struct xfs_ail *ailp) |
1da177e4 | 133 | { |
cd4a3c50 | 134 | xfs_lsn_t lsn = 0; |
1da177e4 | 135 | xfs_log_item_t *lip; |
1da177e4 | 136 | |
c7e8f268 | 137 | spin_lock(&ailp->xa_lock); |
5b00f14f | 138 | lip = xfs_ail_min(ailp); |
cd4a3c50 | 139 | if (lip) |
1da177e4 | 140 | lsn = lip->li_lsn; |
c7e8f268 | 141 | spin_unlock(&ailp->xa_lock); |
1da177e4 LT |
142 | |
143 | return lsn; | |
144 | } | |
145 | ||
fd074841 DC |
146 | /* |
147 | * Return the maximum lsn held in the AIL, or zero if the AIL is empty. | |
148 | */ | |
149 | static xfs_lsn_t | |
150 | xfs_ail_max_lsn( | |
151 | struct xfs_ail *ailp) | |
152 | { | |
153 | xfs_lsn_t lsn = 0; | |
154 | xfs_log_item_t *lip; | |
155 | ||
156 | spin_lock(&ailp->xa_lock); | |
157 | lip = xfs_ail_max(ailp); | |
158 | if (lip) | |
159 | lsn = lip->li_lsn; | |
160 | spin_unlock(&ailp->xa_lock); | |
161 | ||
162 | return lsn; | |
163 | } | |
164 | ||
27d8d5fe DC |
165 | /* |
166 | * AIL traversal cursor initialisation. | |
167 | * | |
168 | * The cursor keeps track of where our current traversal is up | |
169 | * to by tracking the next ƣtem in the list for us. However, for | |
170 | * this to be safe, removing an object from the AIL needs to invalidate | |
171 | * any cursor that points to it. hence the traversal cursor needs to | |
172 | * be linked to the struct xfs_ail so that deletion can search all the | |
173 | * active cursors for invalidation. | |
174 | * | |
175 | * We don't link the push cursor because it is embedded in the struct | |
176 | * xfs_ail and hence easily findable. | |
177 | */ | |
5b00f14f | 178 | STATIC void |
27d8d5fe DC |
179 | xfs_trans_ail_cursor_init( |
180 | struct xfs_ail *ailp, | |
181 | struct xfs_ail_cursor *cur) | |
182 | { | |
183 | cur->item = NULL; | |
184 | if (cur == &ailp->xa_cursors) | |
185 | return; | |
186 | ||
187 | cur->next = ailp->xa_cursors.next; | |
188 | ailp->xa_cursors.next = cur; | |
189 | } | |
190 | ||
191 | /* | |
192 | * Set the cursor to the next item, because when we look | |
193 | * up the cursor the current item may have been freed. | |
194 | */ | |
195 | STATIC void | |
196 | xfs_trans_ail_cursor_set( | |
197 | struct xfs_ail *ailp, | |
198 | struct xfs_ail_cursor *cur, | |
199 | struct xfs_log_item *lip) | |
200 | { | |
201 | if (lip) | |
202 | cur->item = xfs_ail_next(ailp, lip); | |
203 | } | |
204 | ||
205 | /* | |
206 | * Get the next item in the traversal and advance the cursor. | |
207 | * If the cursor was invalidated (inidicated by a lip of 1), | |
208 | * restart the traversal. | |
209 | */ | |
5b00f14f | 210 | struct xfs_log_item * |
27d8d5fe DC |
211 | xfs_trans_ail_cursor_next( |
212 | struct xfs_ail *ailp, | |
213 | struct xfs_ail_cursor *cur) | |
214 | { | |
215 | struct xfs_log_item *lip = cur->item; | |
216 | ||
217 | if ((__psint_t)lip & 1) | |
218 | lip = xfs_ail_min(ailp); | |
219 | xfs_trans_ail_cursor_set(ailp, cur, lip); | |
220 | return lip; | |
221 | } | |
222 | ||
27d8d5fe DC |
223 | /* |
224 | * Now that the traversal is complete, we need to remove the cursor | |
225 | * from the list of traversing cursors. Avoid removing the embedded | |
9da096fd | 226 | * push cursor, but use the fact it is always present to make the |
27d8d5fe DC |
227 | * list deletion simple. |
228 | */ | |
229 | void | |
230 | xfs_trans_ail_cursor_done( | |
231 | struct xfs_ail *ailp, | |
232 | struct xfs_ail_cursor *done) | |
233 | { | |
234 | struct xfs_ail_cursor *prev = NULL; | |
235 | struct xfs_ail_cursor *cur; | |
236 | ||
237 | done->item = NULL; | |
238 | if (done == &ailp->xa_cursors) | |
239 | return; | |
240 | prev = &ailp->xa_cursors; | |
241 | for (cur = prev->next; cur; prev = cur, cur = prev->next) { | |
242 | if (cur == done) { | |
243 | prev->next = cur->next; | |
244 | break; | |
245 | } | |
246 | } | |
247 | ASSERT(cur); | |
248 | } | |
249 | ||
5b00f14f DC |
250 | /* |
251 | * Invalidate any cursor that is pointing to this item. This is | |
252 | * called when an item is removed from the AIL. Any cursor pointing | |
253 | * to this object is now invalid and the traversal needs to be | |
254 | * terminated so it doesn't reference a freed object. We set the | |
255 | * cursor item to a value of 1 so we can distinguish between an | |
256 | * invalidation and the end of the list when getting the next item | |
257 | * from the cursor. | |
258 | */ | |
259 | STATIC void | |
260 | xfs_trans_ail_cursor_clear( | |
261 | struct xfs_ail *ailp, | |
262 | struct xfs_log_item *lip) | |
263 | { | |
264 | struct xfs_ail_cursor *cur; | |
265 | ||
266 | /* need to search all cursors */ | |
267 | for (cur = &ailp->xa_cursors; cur; cur = cur->next) { | |
268 | if (cur->item == lip) | |
269 | cur->item = (struct xfs_log_item *) | |
270 | ((__psint_t)cur->item | 1); | |
271 | } | |
272 | } | |
273 | ||
249a8c11 DC |
274 | /* |
275 | * Return the item in the AIL with the current lsn. | |
276 | * Return the current tree generation number for use | |
277 | * in calls to xfs_trans_next_ail(). | |
278 | */ | |
5b00f14f DC |
279 | xfs_log_item_t * |
280 | xfs_trans_ail_cursor_first( | |
27d8d5fe DC |
281 | struct xfs_ail *ailp, |
282 | struct xfs_ail_cursor *cur, | |
283 | xfs_lsn_t lsn) | |
249a8c11 | 284 | { |
27d8d5fe | 285 | xfs_log_item_t *lip; |
249a8c11 | 286 | |
5b00f14f | 287 | xfs_trans_ail_cursor_init(ailp, cur); |
27d8d5fe | 288 | lip = xfs_ail_min(ailp); |
249a8c11 | 289 | if (lsn == 0) |
5b00f14f | 290 | goto out; |
249a8c11 | 291 | |
27d8d5fe | 292 | list_for_each_entry(lip, &ailp->xa_ail, li_ail) { |
5b00f14f | 293 | if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0) |
7ee49acf | 294 | goto out; |
535f6b37 | 295 | } |
5b00f14f DC |
296 | lip = NULL; |
297 | out: | |
298 | xfs_trans_ail_cursor_set(ailp, cur, lip); | |
299 | return lip; | |
249a8c11 DC |
300 | } |
301 | ||
cd4a3c50 DC |
302 | /* |
303 | * splice the log item list into the AIL at the given LSN. | |
304 | */ | |
305 | static void | |
306 | xfs_ail_splice( | |
307 | struct xfs_ail *ailp, | |
308 | struct list_head *list, | |
309 | xfs_lsn_t lsn) | |
310 | { | |
311 | xfs_log_item_t *next_lip; | |
312 | ||
313 | /* If the list is empty, just insert the item. */ | |
314 | if (list_empty(&ailp->xa_ail)) { | |
315 | list_splice(list, &ailp->xa_ail); | |
316 | return; | |
317 | } | |
318 | ||
319 | list_for_each_entry_reverse(next_lip, &ailp->xa_ail, li_ail) { | |
320 | if (XFS_LSN_CMP(next_lip->li_lsn, lsn) <= 0) | |
321 | break; | |
322 | } | |
323 | ||
324 | ASSERT(&next_lip->li_ail == &ailp->xa_ail || | |
325 | XFS_LSN_CMP(next_lip->li_lsn, lsn) <= 0); | |
326 | ||
327 | list_splice_init(list, &next_lip->li_ail); | |
328 | } | |
329 | ||
330 | /* | |
331 | * Delete the given item from the AIL. Return a pointer to the item. | |
332 | */ | |
333 | static void | |
334 | xfs_ail_delete( | |
335 | struct xfs_ail *ailp, | |
336 | xfs_log_item_t *lip) | |
337 | { | |
338 | xfs_ail_check(ailp, lip); | |
339 | list_del(&lip->li_ail); | |
340 | xfs_trans_ail_cursor_clear(ailp, lip); | |
341 | } | |
342 | ||
249a8c11 | 343 | /* |
0bf6a5bd DC |
344 | * xfs_ail_worker does the work of pushing on the AIL. It will requeue itself |
345 | * to run at a later time if there is more work to do to complete the push. | |
249a8c11 | 346 | */ |
0bf6a5bd DC |
347 | STATIC void |
348 | xfs_ail_worker( | |
349 | struct work_struct *work) | |
249a8c11 | 350 | { |
0bf6a5bd DC |
351 | struct xfs_ail *ailp = container_of(to_delayed_work(work), |
352 | struct xfs_ail, xa_work); | |
353 | long tout; | |
82fa9012 | 354 | xfs_lsn_t target = ailp->xa_target; |
249a8c11 DC |
355 | xfs_lsn_t lsn; |
356 | xfs_log_item_t *lip; | |
249a8c11 | 357 | int flush_log, count, stuck; |
82fa9012 | 358 | xfs_mount_t *mp = ailp->xa_mount; |
27d8d5fe | 359 | struct xfs_ail_cursor *cur = &ailp->xa_cursors; |
d808f617 | 360 | int push_xfsbufd = 0; |
1da177e4 | 361 | |
c7e8f268 | 362 | spin_lock(&ailp->xa_lock); |
27d8d5fe | 363 | xfs_trans_ail_cursor_init(ailp, cur); |
0bf6a5bd | 364 | lip = xfs_trans_ail_cursor_first(ailp, cur, ailp->xa_last_pushed_lsn); |
249a8c11 | 365 | if (!lip || XFS_FORCED_SHUTDOWN(mp)) { |
1da177e4 | 366 | /* |
249a8c11 | 367 | * AIL is empty or our push has reached the end. |
1da177e4 | 368 | */ |
27d8d5fe | 369 | xfs_trans_ail_cursor_done(ailp, cur); |
c7e8f268 | 370 | spin_unlock(&ailp->xa_lock); |
0bf6a5bd DC |
371 | ailp->xa_last_pushed_lsn = 0; |
372 | return; | |
1da177e4 LT |
373 | } |
374 | ||
375 | XFS_STATS_INC(xs_push_ail); | |
376 | ||
377 | /* | |
378 | * While the item we are looking at is below the given threshold | |
249a8c11 | 379 | * try to flush it out. We'd like not to stop until we've at least |
1da177e4 | 380 | * tried to push on everything in the AIL with an LSN less than |
249a8c11 DC |
381 | * the given threshold. |
382 | * | |
383 | * However, we will stop after a certain number of pushes and wait | |
384 | * for a reduced timeout to fire before pushing further. This | |
385 | * prevents use from spinning when we can't do anything or there is | |
386 | * lots of contention on the AIL lists. | |
1da177e4 | 387 | */ |
249a8c11 | 388 | lsn = lip->li_lsn; |
27d8d5fe | 389 | flush_log = stuck = count = 0; |
249a8c11 DC |
390 | while ((XFS_LSN_CMP(lip->li_lsn, target) < 0)) { |
391 | int lock_result; | |
1da177e4 | 392 | /* |
249a8c11 DC |
393 | * If we can lock the item without sleeping, unlock the AIL |
394 | * lock and flush the item. Then re-grab the AIL lock so we | |
395 | * can look for the next item on the AIL. List changes are | |
396 | * handled by the AIL lookup functions internally | |
1da177e4 | 397 | * |
249a8c11 DC |
398 | * If we can't lock the item, either its holder will flush it |
399 | * or it is already being flushed or it is being relogged. In | |
400 | * any of these case it is being taken care of and we can just | |
401 | * skip to the next item in the list. | |
1da177e4 LT |
402 | */ |
403 | lock_result = IOP_TRYLOCK(lip); | |
c7e8f268 | 404 | spin_unlock(&ailp->xa_lock); |
1da177e4 | 405 | switch (lock_result) { |
249a8c11 | 406 | case XFS_ITEM_SUCCESS: |
1da177e4 LT |
407 | XFS_STATS_INC(xs_push_ail_success); |
408 | IOP_PUSH(lip); | |
0bf6a5bd | 409 | ailp->xa_last_pushed_lsn = lsn; |
1da177e4 LT |
410 | break; |
411 | ||
249a8c11 | 412 | case XFS_ITEM_PUSHBUF: |
1da177e4 | 413 | XFS_STATS_INC(xs_push_ail_pushbuf); |
1da177e4 | 414 | IOP_PUSHBUF(lip); |
0bf6a5bd | 415 | ailp->xa_last_pushed_lsn = lsn; |
d808f617 | 416 | push_xfsbufd = 1; |
1da177e4 LT |
417 | break; |
418 | ||
249a8c11 | 419 | case XFS_ITEM_PINNED: |
1da177e4 | 420 | XFS_STATS_INC(xs_push_ail_pinned); |
249a8c11 | 421 | stuck++; |
1da177e4 LT |
422 | flush_log = 1; |
423 | break; | |
424 | ||
249a8c11 | 425 | case XFS_ITEM_LOCKED: |
1da177e4 | 426 | XFS_STATS_INC(xs_push_ail_locked); |
0bf6a5bd | 427 | ailp->xa_last_pushed_lsn = lsn; |
249a8c11 | 428 | stuck++; |
1da177e4 LT |
429 | break; |
430 | ||
249a8c11 | 431 | default: |
1da177e4 LT |
432 | ASSERT(0); |
433 | break; | |
434 | } | |
435 | ||
c7e8f268 | 436 | spin_lock(&ailp->xa_lock); |
249a8c11 DC |
437 | /* should we bother continuing? */ |
438 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1da177e4 | 439 | break; |
249a8c11 DC |
440 | ASSERT(mp->m_log); |
441 | ||
442 | count++; | |
1da177e4 | 443 | |
249a8c11 DC |
444 | /* |
445 | * Are there too many items we can't do anything with? | |
446 | * If we we are skipping too many items because we can't flush | |
447 | * them or they are already being flushed, we back off and | |
448 | * given them time to complete whatever operation is being | |
449 | * done. i.e. remove pressure from the AIL while we can't make | |
450 | * progress so traversals don't slow down further inserts and | |
451 | * removals to/from the AIL. | |
452 | * | |
453 | * The value of 100 is an arbitrary magic number based on | |
454 | * observation. | |
455 | */ | |
456 | if (stuck > 100) | |
457 | break; | |
458 | ||
27d8d5fe | 459 | lip = xfs_trans_ail_cursor_next(ailp, cur); |
249a8c11 DC |
460 | if (lip == NULL) |
461 | break; | |
249a8c11 | 462 | lsn = lip->li_lsn; |
1da177e4 | 463 | } |
27d8d5fe | 464 | xfs_trans_ail_cursor_done(ailp, cur); |
c7e8f268 | 465 | spin_unlock(&ailp->xa_lock); |
1da177e4 LT |
466 | |
467 | if (flush_log) { | |
468 | /* | |
469 | * If something we need to push out was pinned, then | |
470 | * push out the log so it will become unpinned and | |
471 | * move forward in the AIL. | |
472 | */ | |
1da177e4 | 473 | XFS_STATS_INC(xs_push_ail_flush); |
a14a348b | 474 | xfs_log_force(mp, 0); |
1da177e4 LT |
475 | } |
476 | ||
d808f617 DC |
477 | if (push_xfsbufd) { |
478 | /* we've got delayed write buffers to flush */ | |
479 | wake_up_process(mp->m_ddev_targp->bt_task); | |
480 | } | |
481 | ||
0bf6a5bd DC |
482 | /* assume we have more work to do in a short while */ |
483 | tout = 10; | |
92d9cd10 DC |
484 | if (!count) { |
485 | /* We're past our target or empty, so idle */ | |
0bf6a5bd DC |
486 | ailp->xa_last_pushed_lsn = 0; |
487 | ||
488 | /* | |
489 | * Check for an updated push target before clearing the | |
490 | * XFS_AIL_PUSHING_BIT. If the target changed, we've got more | |
491 | * work to do. Wait a bit longer before starting that work. | |
492 | */ | |
493 | smp_rmb(); | |
494 | if (ailp->xa_target == target) { | |
495 | clear_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags); | |
496 | return; | |
497 | } | |
498 | tout = 50; | |
92d9cd10 DC |
499 | } else if (XFS_LSN_CMP(lsn, target) >= 0) { |
500 | /* | |
501 | * We reached the target so wait a bit longer for I/O to | |
502 | * complete and remove pushed items from the AIL before we | |
503 | * start the next scan from the start of the AIL. | |
504 | */ | |
453eac8a | 505 | tout = 50; |
0bf6a5bd | 506 | ailp->xa_last_pushed_lsn = 0; |
27d8d5fe | 507 | } else if ((stuck * 100) / count > 90) { |
249a8c11 DC |
508 | /* |
509 | * Either there is a lot of contention on the AIL or we | |
510 | * are stuck due to operations in progress. "Stuck" in this | |
511 | * case is defined as >90% of the items we tried to push | |
512 | * were stuck. | |
513 | * | |
514 | * Backoff a bit more to allow some I/O to complete before | |
515 | * continuing from where we were. | |
516 | */ | |
453eac8a | 517 | tout = 20; |
1da177e4 | 518 | } |
0bf6a5bd DC |
519 | |
520 | /* There is more to do, requeue us. */ | |
521 | queue_delayed_work(xfs_syncd_wq, &ailp->xa_work, | |
522 | msecs_to_jiffies(tout)); | |
453eac8a | 523 | } |
1da177e4 | 524 | |
0bf6a5bd DC |
525 | /* |
526 | * This routine is called to move the tail of the AIL forward. It does this by | |
527 | * trying to flush items in the AIL whose lsns are below the given | |
528 | * threshold_lsn. | |
529 | * | |
530 | * The push is run asynchronously in a workqueue, which means the caller needs | |
531 | * to handle waiting on the async flush for space to become available. | |
532 | * We don't want to interrupt any push that is in progress, hence we only queue | |
533 | * work if we set the pushing bit approriately. | |
534 | * | |
535 | * We do this unlocked - we only need to know whether there is anything in the | |
536 | * AIL at the time we are called. We don't need to access the contents of | |
537 | * any of the objects, so the lock is not needed. | |
538 | */ | |
539 | void | |
fd074841 | 540 | xfs_ail_push( |
0bf6a5bd DC |
541 | struct xfs_ail *ailp, |
542 | xfs_lsn_t threshold_lsn) | |
543 | { | |
544 | xfs_log_item_t *lip; | |
545 | ||
546 | lip = xfs_ail_min(ailp); | |
547 | if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) || | |
548 | XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0) | |
549 | return; | |
550 | ||
551 | /* | |
552 | * Ensure that the new target is noticed in push code before it clears | |
553 | * the XFS_AIL_PUSHING_BIT. | |
554 | */ | |
555 | smp_wmb(); | |
556 | ailp->xa_target = threshold_lsn; | |
557 | if (!test_and_set_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags)) | |
558 | queue_delayed_work(xfs_syncd_wq, &ailp->xa_work, 0); | |
559 | } | |
1da177e4 | 560 | |
fd074841 DC |
561 | /* |
562 | * Push out all items in the AIL immediately | |
563 | */ | |
564 | void | |
565 | xfs_ail_push_all( | |
566 | struct xfs_ail *ailp) | |
567 | { | |
568 | xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp); | |
569 | ||
570 | if (threshold_lsn) | |
571 | xfs_ail_push(ailp, threshold_lsn); | |
572 | } | |
573 | ||
1da177e4 LT |
574 | /* |
575 | * This is to be called when an item is unlocked that may have | |
576 | * been in the AIL. It will wake up the first member of the AIL | |
577 | * wait list if this item's unlocking might allow it to progress. | |
578 | * If the item is in the AIL, then we need to get the AIL lock | |
579 | * while doing our checking so we don't race with someone going | |
580 | * to sleep waiting for this event in xfs_trans_push_ail(). | |
581 | */ | |
582 | void | |
583 | xfs_trans_unlocked_item( | |
783a2f65 | 584 | struct xfs_ail *ailp, |
1da177e4 LT |
585 | xfs_log_item_t *lip) |
586 | { | |
587 | xfs_log_item_t *min_lip; | |
588 | ||
589 | /* | |
590 | * If we're forcibly shutting down, we may have | |
591 | * unlocked log items arbitrarily. The last thing | |
592 | * we want to do is to move the tail of the log | |
593 | * over some potentially valid data. | |
594 | */ | |
595 | if (!(lip->li_flags & XFS_LI_IN_AIL) || | |
783a2f65 | 596 | XFS_FORCED_SHUTDOWN(ailp->xa_mount)) { |
1da177e4 LT |
597 | return; |
598 | } | |
599 | ||
600 | /* | |
601 | * This is the one case where we can call into xfs_ail_min() | |
602 | * without holding the AIL lock because we only care about the | |
603 | * case where we are at the tail of the AIL. If the object isn't | |
604 | * at the tail, it doesn't matter what result we get back. This | |
605 | * is slightly racy because since we were just unlocked, we could | |
606 | * go to sleep between the call to xfs_ail_min and the call to | |
607 | * xfs_log_move_tail, have someone else lock us, commit to us disk, | |
608 | * move us out of the tail of the AIL, and then we wake up. However, | |
609 | * the call to xfs_log_move_tail() doesn't do anything if there's | |
610 | * not enough free space to wake people up so we're safe calling it. | |
611 | */ | |
783a2f65 | 612 | min_lip = xfs_ail_min(ailp); |
1da177e4 LT |
613 | |
614 | if (min_lip == lip) | |
783a2f65 | 615 | xfs_log_move_tail(ailp->xa_mount, 1); |
1da177e4 LT |
616 | } /* xfs_trans_unlocked_item */ |
617 | ||
0e57f6a3 DC |
618 | /* |
619 | * xfs_trans_ail_update - bulk AIL insertion operation. | |
620 | * | |
621 | * @xfs_trans_ail_update takes an array of log items that all need to be | |
622 | * positioned at the same LSN in the AIL. If an item is not in the AIL, it will | |
623 | * be added. Otherwise, it will be repositioned by removing it and re-adding | |
624 | * it to the AIL. If we move the first item in the AIL, update the log tail to | |
625 | * match the new minimum LSN in the AIL. | |
626 | * | |
627 | * This function takes the AIL lock once to execute the update operations on | |
628 | * all the items in the array, and as such should not be called with the AIL | |
629 | * lock held. As a result, once we have the AIL lock, we need to check each log | |
630 | * item LSN to confirm it needs to be moved forward in the AIL. | |
631 | * | |
632 | * To optimise the insert operation, we delete all the items from the AIL in | |
633 | * the first pass, moving them into a temporary list, then splice the temporary | |
634 | * list into the correct position in the AIL. This avoids needing to do an | |
635 | * insert operation on every item. | |
636 | * | |
637 | * This function must be called with the AIL lock held. The lock is dropped | |
638 | * before returning. | |
639 | */ | |
640 | void | |
641 | xfs_trans_ail_update_bulk( | |
642 | struct xfs_ail *ailp, | |
643 | 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; | |
648 | xfs_lsn_t tail_lsn; | |
649 | int mlip_changed = 0; | |
650 | int i; | |
651 | LIST_HEAD(tmp); | |
652 | ||
653 | 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 | ||
672 | xfs_ail_splice(ailp, &tmp, lsn); | |
673 | ||
674 | if (!mlip_changed) { | |
675 | spin_unlock(&ailp->xa_lock); | |
676 | return; | |
677 | } | |
678 | ||
679 | /* | |
680 | * It is not safe to access mlip after the AIL lock is dropped, so we | |
681 | * must get a copy of li_lsn before we do so. This is especially | |
682 | * important on 32-bit platforms where accessing and updating 64-bit | |
683 | * values like li_lsn is not atomic. | |
684 | */ | |
685 | mlip = xfs_ail_min(ailp); | |
686 | tail_lsn = mlip->li_lsn; | |
687 | spin_unlock(&ailp->xa_lock); | |
688 | xfs_log_move_tail(ailp->xa_mount, tail_lsn); | |
689 | } | |
690 | ||
30136832 DC |
691 | /* |
692 | * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL | |
693 | * | |
694 | * @xfs_trans_ail_delete_bulk takes an array of log items that all need to | |
695 | * removed from the AIL. The caller is already holding the AIL lock, and done | |
696 | * all the checks necessary to ensure the items passed in via @log_items are | |
697 | * ready for deletion. This includes checking that the items are in the AIL. | |
698 | * | |
699 | * For each log item to be removed, unlink it from the AIL, clear the IN_AIL | |
700 | * flag from the item and reset the item's lsn to 0. If we remove the first | |
701 | * item in the AIL, update the log tail to match the new minimum LSN in the | |
702 | * AIL. | |
703 | * | |
704 | * This function will not drop the AIL lock until all items are removed from | |
705 | * the AIL to minimise the amount of lock traffic on the AIL. This does not | |
706 | * greatly increase the AIL hold time, but does significantly reduce the amount | |
707 | * of traffic on the lock, especially during IO completion. | |
708 | * | |
709 | * This function must be called with the AIL lock held. The lock is dropped | |
710 | * before returning. | |
711 | */ | |
712 | void | |
713 | xfs_trans_ail_delete_bulk( | |
714 | struct xfs_ail *ailp, | |
715 | struct xfs_log_item **log_items, | |
716 | int nr_items) __releases(ailp->xa_lock) | |
717 | { | |
718 | xfs_log_item_t *mlip; | |
719 | xfs_lsn_t tail_lsn; | |
720 | int mlip_changed = 0; | |
721 | int i; | |
722 | ||
723 | mlip = xfs_ail_min(ailp); | |
724 | ||
725 | for (i = 0; i < nr_items; i++) { | |
726 | struct xfs_log_item *lip = log_items[i]; | |
727 | if (!(lip->li_flags & XFS_LI_IN_AIL)) { | |
728 | struct xfs_mount *mp = ailp->xa_mount; | |
729 | ||
730 | spin_unlock(&ailp->xa_lock); | |
731 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
6a19d939 | 732 | xfs_alert_tag(mp, XFS_PTAG_AILDELETE, |
30136832 DC |
733 | "%s: attempting to delete a log item that is not in the AIL", |
734 | __func__); | |
735 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
736 | } | |
737 | return; | |
738 | } | |
739 | ||
740 | xfs_ail_delete(ailp, lip); | |
741 | lip->li_flags &= ~XFS_LI_IN_AIL; | |
742 | lip->li_lsn = 0; | |
743 | if (mlip == lip) | |
744 | mlip_changed = 1; | |
745 | } | |
746 | ||
747 | if (!mlip_changed) { | |
748 | spin_unlock(&ailp->xa_lock); | |
749 | return; | |
750 | } | |
751 | ||
752 | /* | |
753 | * It is not safe to access mlip after the AIL lock is dropped, so we | |
754 | * must get a copy of li_lsn before we do so. This is especially | |
755 | * important on 32-bit platforms where accessing and updating 64-bit | |
756 | * values like li_lsn is not atomic. It is possible we've emptied the | |
757 | * AIL here, so if that is the case, pass an LSN of 0 to the tail move. | |
758 | */ | |
759 | mlip = xfs_ail_min(ailp); | |
760 | tail_lsn = mlip ? mlip->li_lsn : 0; | |
761 | spin_unlock(&ailp->xa_lock); | |
762 | xfs_log_move_tail(ailp->xa_mount, tail_lsn); | |
763 | } | |
1da177e4 | 764 | |
1da177e4 LT |
765 | /* |
766 | * The active item list (AIL) is a doubly linked list of log | |
767 | * items sorted by ascending lsn. The base of the list is | |
768 | * a forw/back pointer pair embedded in the xfs mount structure. | |
769 | * The base is initialized with both pointers pointing to the | |
770 | * base. This case always needs to be distinguished, because | |
771 | * the base has no lsn to look at. We almost always insert | |
772 | * at the end of the list, so on inserts we search from the | |
773 | * end of the list to find where the new item belongs. | |
774 | */ | |
775 | ||
776 | /* | |
777 | * Initialize the doubly linked list to point only to itself. | |
778 | */ | |
249a8c11 | 779 | int |
1da177e4 LT |
780 | xfs_trans_ail_init( |
781 | xfs_mount_t *mp) | |
782 | { | |
82fa9012 DC |
783 | struct xfs_ail *ailp; |
784 | ||
785 | ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL); | |
786 | if (!ailp) | |
787 | return ENOMEM; | |
788 | ||
789 | ailp->xa_mount = mp; | |
790 | INIT_LIST_HEAD(&ailp->xa_ail); | |
c7e8f268 | 791 | spin_lock_init(&ailp->xa_lock); |
0bf6a5bd | 792 | INIT_DELAYED_WORK(&ailp->xa_work, xfs_ail_worker); |
27d8d5fe DC |
793 | mp->m_ail = ailp; |
794 | return 0; | |
249a8c11 DC |
795 | } |
796 | ||
797 | void | |
798 | xfs_trans_ail_destroy( | |
799 | xfs_mount_t *mp) | |
800 | { | |
82fa9012 DC |
801 | struct xfs_ail *ailp = mp->m_ail; |
802 | ||
0bf6a5bd | 803 | cancel_delayed_work_sync(&ailp->xa_work); |
82fa9012 | 804 | kmem_free(ailp); |
1da177e4 | 805 | } |