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 | 165 | /* |
af3e4022 DC |
166 | * The cursor keeps track of where our current traversal is up to by tracking |
167 | * the next item in the list for us. However, for this to be safe, removing an | |
168 | * object from the AIL needs to invalidate any cursor that points to it. hence | |
169 | * the traversal cursor needs to be linked to the struct xfs_ail so that | |
170 | * deletion can search all the active cursors for invalidation. | |
27d8d5fe | 171 | */ |
5b00f14f | 172 | STATIC void |
27d8d5fe DC |
173 | xfs_trans_ail_cursor_init( |
174 | struct xfs_ail *ailp, | |
175 | struct xfs_ail_cursor *cur) | |
176 | { | |
177 | cur->item = NULL; | |
af3e4022 | 178 | list_add_tail(&cur->list, &ailp->xa_cursors); |
27d8d5fe DC |
179 | } |
180 | ||
27d8d5fe | 181 | /* |
af3e4022 DC |
182 | * Get the next item in the traversal and advance the cursor. If the cursor |
183 | * was invalidated (indicated by a lip of 1), restart the traversal. | |
27d8d5fe | 184 | */ |
5b00f14f | 185 | struct xfs_log_item * |
27d8d5fe DC |
186 | xfs_trans_ail_cursor_next( |
187 | struct xfs_ail *ailp, | |
188 | struct xfs_ail_cursor *cur) | |
189 | { | |
190 | struct xfs_log_item *lip = cur->item; | |
191 | ||
192 | if ((__psint_t)lip & 1) | |
193 | lip = xfs_ail_min(ailp); | |
16b59029 DC |
194 | if (lip) |
195 | cur->item = xfs_ail_next(ailp, lip); | |
27d8d5fe DC |
196 | return lip; |
197 | } | |
198 | ||
27d8d5fe | 199 | /* |
af3e4022 DC |
200 | * When the traversal is complete, we need to remove the cursor from the list |
201 | * of traversing cursors. | |
27d8d5fe DC |
202 | */ |
203 | void | |
204 | xfs_trans_ail_cursor_done( | |
205 | struct xfs_ail *ailp, | |
af3e4022 | 206 | struct xfs_ail_cursor *cur) |
27d8d5fe | 207 | { |
af3e4022 DC |
208 | cur->item = NULL; |
209 | list_del_init(&cur->list); | |
27d8d5fe DC |
210 | } |
211 | ||
5b00f14f | 212 | /* |
af3e4022 DC |
213 | * Invalidate any cursor that is pointing to this item. This is called when an |
214 | * item is removed from the AIL. Any cursor pointing to this object is now | |
215 | * invalid and the traversal needs to be terminated so it doesn't reference a | |
216 | * freed object. We set the low bit of the cursor item pointer so we can | |
217 | * distinguish between an invalidation and the end of the list when getting the | |
218 | * next item from the cursor. | |
5b00f14f DC |
219 | */ |
220 | STATIC void | |
221 | xfs_trans_ail_cursor_clear( | |
222 | struct xfs_ail *ailp, | |
223 | struct xfs_log_item *lip) | |
224 | { | |
225 | struct xfs_ail_cursor *cur; | |
226 | ||
af3e4022 | 227 | list_for_each_entry(cur, &ailp->xa_cursors, list) { |
5b00f14f DC |
228 | if (cur->item == lip) |
229 | cur->item = (struct xfs_log_item *) | |
230 | ((__psint_t)cur->item | 1); | |
231 | } | |
232 | } | |
233 | ||
249a8c11 | 234 | /* |
16b59029 DC |
235 | * Find the first item in the AIL with the given @lsn by searching in ascending |
236 | * LSN order and initialise the cursor to point to the next item for a | |
237 | * ascending traversal. Pass a @lsn of zero to initialise the cursor to the | |
238 | * first item in the AIL. Returns NULL if the list is empty. | |
249a8c11 | 239 | */ |
5b00f14f DC |
240 | xfs_log_item_t * |
241 | xfs_trans_ail_cursor_first( | |
27d8d5fe DC |
242 | struct xfs_ail *ailp, |
243 | struct xfs_ail_cursor *cur, | |
244 | xfs_lsn_t lsn) | |
249a8c11 | 245 | { |
27d8d5fe | 246 | xfs_log_item_t *lip; |
249a8c11 | 247 | |
5b00f14f | 248 | xfs_trans_ail_cursor_init(ailp, cur); |
16b59029 DC |
249 | |
250 | if (lsn == 0) { | |
251 | lip = xfs_ail_min(ailp); | |
5b00f14f | 252 | goto out; |
16b59029 | 253 | } |
249a8c11 | 254 | |
27d8d5fe | 255 | list_for_each_entry(lip, &ailp->xa_ail, li_ail) { |
5b00f14f | 256 | if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0) |
7ee49acf | 257 | goto out; |
535f6b37 | 258 | } |
16b59029 DC |
259 | return NULL; |
260 | ||
5b00f14f | 261 | out: |
16b59029 DC |
262 | if (lip) |
263 | cur->item = xfs_ail_next(ailp, lip); | |
5b00f14f | 264 | return lip; |
249a8c11 DC |
265 | } |
266 | ||
1d8c95a3 DC |
267 | static struct xfs_log_item * |
268 | __xfs_trans_ail_cursor_last( | |
269 | struct xfs_ail *ailp, | |
270 | xfs_lsn_t lsn) | |
271 | { | |
272 | xfs_log_item_t *lip; | |
273 | ||
274 | list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) { | |
275 | if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0) | |
276 | return lip; | |
277 | } | |
278 | return NULL; | |
279 | } | |
280 | ||
281 | /* | |
16b59029 DC |
282 | * Find the last item in the AIL with the given @lsn by searching in descending |
283 | * LSN order and initialise the cursor to point to that item. If there is no | |
284 | * item with the value of @lsn, then it sets the cursor to the last item with an | |
285 | * LSN lower than @lsn. Returns NULL if the list is empty. | |
1d8c95a3 DC |
286 | */ |
287 | struct xfs_log_item * | |
288 | xfs_trans_ail_cursor_last( | |
289 | struct xfs_ail *ailp, | |
290 | struct xfs_ail_cursor *cur, | |
291 | xfs_lsn_t lsn) | |
292 | { | |
293 | xfs_trans_ail_cursor_init(ailp, cur); | |
294 | cur->item = __xfs_trans_ail_cursor_last(ailp, lsn); | |
295 | return cur->item; | |
296 | } | |
297 | ||
298 | /* | |
16b59029 | 299 | * Splice the log item list into the AIL at the given LSN. We splice to the |
1d8c95a3 DC |
300 | * tail of the given LSN to maintain insert order for push traversals. The |
301 | * cursor is optional, allowing repeated updates to the same LSN to avoid | |
302 | * repeated traversals. | |
cd4a3c50 DC |
303 | */ |
304 | static void | |
305 | xfs_ail_splice( | |
1d8c95a3 DC |
306 | struct xfs_ail *ailp, |
307 | struct xfs_ail_cursor *cur, | |
308 | struct list_head *list, | |
309 | xfs_lsn_t lsn) | |
cd4a3c50 | 310 | { |
1d8c95a3 DC |
311 | struct xfs_log_item *lip = cur ? cur->item : NULL; |
312 | struct xfs_log_item *next_lip; | |
cd4a3c50 | 313 | |
1d8c95a3 DC |
314 | /* |
315 | * Get a new cursor if we don't have a placeholder or the existing one | |
316 | * has been invalidated. | |
317 | */ | |
318 | if (!lip || (__psint_t)lip & 1) { | |
319 | lip = __xfs_trans_ail_cursor_last(ailp, lsn); | |
320 | ||
321 | if (!lip) { | |
322 | /* The list is empty, so just splice and return. */ | |
323 | if (cur) | |
324 | cur->item = NULL; | |
325 | list_splice(list, &ailp->xa_ail); | |
326 | return; | |
327 | } | |
cd4a3c50 DC |
328 | } |
329 | ||
1d8c95a3 DC |
330 | /* |
331 | * Our cursor points to the item we want to insert _after_, so we have | |
332 | * to update the cursor to point to the end of the list we are splicing | |
333 | * in so that it points to the correct location for the next splice. | |
334 | * i.e. before the splice | |
335 | * | |
336 | * lsn -> lsn -> lsn + x -> lsn + x ... | |
337 | * ^ | |
338 | * | cursor points here | |
339 | * | |
340 | * After the splice we have: | |
341 | * | |
342 | * lsn -> lsn -> lsn -> lsn -> .... -> lsn -> lsn + x -> lsn + x ... | |
343 | * ^ ^ | |
344 | * | cursor points here | needs to move here | |
345 | * | |
346 | * So we set the cursor to the last item in the list to be spliced | |
347 | * before we execute the splice, resulting in the cursor pointing to | |
348 | * the correct item after the splice occurs. | |
349 | */ | |
350 | if (cur) { | |
351 | next_lip = list_entry(list->prev, struct xfs_log_item, li_ail); | |
352 | cur->item = next_lip; | |
cd4a3c50 | 353 | } |
1d8c95a3 | 354 | list_splice(list, &lip->li_ail); |
cd4a3c50 DC |
355 | } |
356 | ||
357 | /* | |
358 | * Delete the given item from the AIL. Return a pointer to the item. | |
359 | */ | |
360 | static void | |
361 | xfs_ail_delete( | |
362 | struct xfs_ail *ailp, | |
363 | xfs_log_item_t *lip) | |
364 | { | |
365 | xfs_ail_check(ailp, lip); | |
366 | list_del(&lip->li_ail); | |
367 | xfs_trans_ail_cursor_clear(ailp, lip); | |
368 | } | |
369 | ||
249a8c11 | 370 | /* |
0bf6a5bd DC |
371 | * xfs_ail_worker does the work of pushing on the AIL. It will requeue itself |
372 | * to run at a later time if there is more work to do to complete the push. | |
249a8c11 | 373 | */ |
0bf6a5bd DC |
374 | STATIC void |
375 | xfs_ail_worker( | |
9e7004e7 | 376 | struct work_struct *work) |
249a8c11 | 377 | { |
9e7004e7 | 378 | struct xfs_ail *ailp = container_of(to_delayed_work(work), |
0bf6a5bd | 379 | struct xfs_ail, xa_work); |
9e7004e7 | 380 | xfs_mount_t *mp = ailp->xa_mount; |
af3e4022 | 381 | struct xfs_ail_cursor cur; |
9e7004e7 DC |
382 | xfs_log_item_t *lip; |
383 | xfs_lsn_t lsn; | |
fe0da767 | 384 | xfs_lsn_t target; |
9e7004e7 DC |
385 | long tout = 10; |
386 | int flush_log = 0; | |
387 | int stuck = 0; | |
388 | int count = 0; | |
389 | int push_xfsbufd = 0; | |
1da177e4 | 390 | |
c7e8f268 | 391 | spin_lock(&ailp->xa_lock); |
fe0da767 | 392 | target = ailp->xa_target; |
af3e4022 | 393 | lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn); |
249a8c11 | 394 | if (!lip || XFS_FORCED_SHUTDOWN(mp)) { |
1da177e4 | 395 | /* |
249a8c11 | 396 | * AIL is empty or our push has reached the end. |
1da177e4 | 397 | */ |
af3e4022 | 398 | xfs_trans_ail_cursor_done(ailp, &cur); |
c7e8f268 | 399 | spin_unlock(&ailp->xa_lock); |
9e7004e7 | 400 | goto out_done; |
1da177e4 LT |
401 | } |
402 | ||
403 | XFS_STATS_INC(xs_push_ail); | |
404 | ||
405 | /* | |
406 | * While the item we are looking at is below the given threshold | |
249a8c11 | 407 | * try to flush it out. We'd like not to stop until we've at least |
1da177e4 | 408 | * tried to push on everything in the AIL with an LSN less than |
249a8c11 DC |
409 | * the given threshold. |
410 | * | |
411 | * However, we will stop after a certain number of pushes and wait | |
412 | * for a reduced timeout to fire before pushing further. This | |
413 | * prevents use from spinning when we can't do anything or there is | |
414 | * lots of contention on the AIL lists. | |
1da177e4 | 415 | */ |
249a8c11 | 416 | lsn = lip->li_lsn; |
50e86686 | 417 | while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) { |
249a8c11 | 418 | int lock_result; |
1da177e4 | 419 | /* |
249a8c11 DC |
420 | * If we can lock the item without sleeping, unlock the AIL |
421 | * lock and flush the item. Then re-grab the AIL lock so we | |
422 | * can look for the next item on the AIL. List changes are | |
423 | * handled by the AIL lookup functions internally | |
1da177e4 | 424 | * |
249a8c11 DC |
425 | * If we can't lock the item, either its holder will flush it |
426 | * or it is already being flushed or it is being relogged. In | |
427 | * any of these case it is being taken care of and we can just | |
428 | * skip to the next item in the list. | |
1da177e4 LT |
429 | */ |
430 | lock_result = IOP_TRYLOCK(lip); | |
c7e8f268 | 431 | spin_unlock(&ailp->xa_lock); |
1da177e4 | 432 | switch (lock_result) { |
249a8c11 | 433 | case XFS_ITEM_SUCCESS: |
1da177e4 LT |
434 | XFS_STATS_INC(xs_push_ail_success); |
435 | IOP_PUSH(lip); | |
0bf6a5bd | 436 | ailp->xa_last_pushed_lsn = lsn; |
1da177e4 LT |
437 | break; |
438 | ||
249a8c11 | 439 | case XFS_ITEM_PUSHBUF: |
1da177e4 | 440 | XFS_STATS_INC(xs_push_ail_pushbuf); |
1da177e4 | 441 | IOP_PUSHBUF(lip); |
0bf6a5bd | 442 | ailp->xa_last_pushed_lsn = lsn; |
d808f617 | 443 | push_xfsbufd = 1; |
1da177e4 LT |
444 | break; |
445 | ||
249a8c11 | 446 | case XFS_ITEM_PINNED: |
1da177e4 | 447 | XFS_STATS_INC(xs_push_ail_pinned); |
249a8c11 | 448 | stuck++; |
1da177e4 LT |
449 | flush_log = 1; |
450 | break; | |
451 | ||
249a8c11 | 452 | case XFS_ITEM_LOCKED: |
1da177e4 | 453 | XFS_STATS_INC(xs_push_ail_locked); |
0bf6a5bd | 454 | ailp->xa_last_pushed_lsn = lsn; |
249a8c11 | 455 | stuck++; |
1da177e4 LT |
456 | break; |
457 | ||
249a8c11 | 458 | default: |
1da177e4 LT |
459 | ASSERT(0); |
460 | break; | |
461 | } | |
462 | ||
c7e8f268 | 463 | spin_lock(&ailp->xa_lock); |
249a8c11 DC |
464 | /* should we bother continuing? */ |
465 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1da177e4 | 466 | break; |
249a8c11 DC |
467 | ASSERT(mp->m_log); |
468 | ||
469 | count++; | |
1da177e4 | 470 | |
249a8c11 DC |
471 | /* |
472 | * Are there too many items we can't do anything with? | |
473 | * If we we are skipping too many items because we can't flush | |
474 | * them or they are already being flushed, we back off and | |
475 | * given them time to complete whatever operation is being | |
476 | * done. i.e. remove pressure from the AIL while we can't make | |
477 | * progress so traversals don't slow down further inserts and | |
478 | * removals to/from the AIL. | |
479 | * | |
480 | * The value of 100 is an arbitrary magic number based on | |
481 | * observation. | |
482 | */ | |
483 | if (stuck > 100) | |
484 | break; | |
485 | ||
af3e4022 | 486 | lip = xfs_trans_ail_cursor_next(ailp, &cur); |
249a8c11 DC |
487 | if (lip == NULL) |
488 | break; | |
249a8c11 | 489 | lsn = lip->li_lsn; |
1da177e4 | 490 | } |
af3e4022 | 491 | xfs_trans_ail_cursor_done(ailp, &cur); |
c7e8f268 | 492 | spin_unlock(&ailp->xa_lock); |
1da177e4 LT |
493 | |
494 | if (flush_log) { | |
495 | /* | |
496 | * If something we need to push out was pinned, then | |
497 | * push out the log so it will become unpinned and | |
498 | * move forward in the AIL. | |
499 | */ | |
1da177e4 | 500 | XFS_STATS_INC(xs_push_ail_flush); |
a14a348b | 501 | xfs_log_force(mp, 0); |
1da177e4 LT |
502 | } |
503 | ||
d808f617 DC |
504 | if (push_xfsbufd) { |
505 | /* we've got delayed write buffers to flush */ | |
506 | wake_up_process(mp->m_ddev_targp->bt_task); | |
507 | } | |
508 | ||
0bf6a5bd | 509 | /* assume we have more work to do in a short while */ |
9e7004e7 | 510 | out_done: |
92d9cd10 DC |
511 | if (!count) { |
512 | /* We're past our target or empty, so idle */ | |
0bf6a5bd DC |
513 | ailp->xa_last_pushed_lsn = 0; |
514 | ||
515 | /* | |
7ac95657 DC |
516 | * We clear the XFS_AIL_PUSHING_BIT first before checking |
517 | * whether the target has changed. If the target has changed, | |
518 | * this pushes the requeue race directly onto the result of the | |
519 | * atomic test/set bit, so we are guaranteed that either the | |
520 | * the pusher that changed the target or ourselves will requeue | |
521 | * the work (but not both). | |
0bf6a5bd | 522 | */ |
7ac95657 | 523 | clear_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags); |
0bf6a5bd | 524 | smp_rmb(); |
7ac95657 DC |
525 | if (XFS_LSN_CMP(ailp->xa_target, target) == 0 || |
526 | test_and_set_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags)) | |
0bf6a5bd | 527 | return; |
7ac95657 | 528 | |
0bf6a5bd | 529 | tout = 50; |
92d9cd10 DC |
530 | } else if (XFS_LSN_CMP(lsn, target) >= 0) { |
531 | /* | |
532 | * We reached the target so wait a bit longer for I/O to | |
533 | * complete and remove pushed items from the AIL before we | |
534 | * start the next scan from the start of the AIL. | |
535 | */ | |
453eac8a | 536 | tout = 50; |
0bf6a5bd | 537 | ailp->xa_last_pushed_lsn = 0; |
27d8d5fe | 538 | } else if ((stuck * 100) / count > 90) { |
249a8c11 DC |
539 | /* |
540 | * Either there is a lot of contention on the AIL or we | |
541 | * are stuck due to operations in progress. "Stuck" in this | |
542 | * case is defined as >90% of the items we tried to push | |
543 | * were stuck. | |
544 | * | |
545 | * Backoff a bit more to allow some I/O to complete before | |
546 | * continuing from where we were. | |
547 | */ | |
453eac8a | 548 | tout = 20; |
1da177e4 | 549 | } |
0bf6a5bd DC |
550 | |
551 | /* There is more to do, requeue us. */ | |
552 | queue_delayed_work(xfs_syncd_wq, &ailp->xa_work, | |
553 | msecs_to_jiffies(tout)); | |
453eac8a | 554 | } |
1da177e4 | 555 | |
0bf6a5bd DC |
556 | /* |
557 | * This routine is called to move the tail of the AIL forward. It does this by | |
558 | * trying to flush items in the AIL whose lsns are below the given | |
559 | * threshold_lsn. | |
560 | * | |
561 | * The push is run asynchronously in a workqueue, which means the caller needs | |
562 | * to handle waiting on the async flush for space to become available. | |
563 | * We don't want to interrupt any push that is in progress, hence we only queue | |
564 | * work if we set the pushing bit approriately. | |
565 | * | |
566 | * We do this unlocked - we only need to know whether there is anything in the | |
567 | * AIL at the time we are called. We don't need to access the contents of | |
568 | * any of the objects, so the lock is not needed. | |
569 | */ | |
570 | void | |
fd074841 | 571 | xfs_ail_push( |
0bf6a5bd DC |
572 | struct xfs_ail *ailp, |
573 | xfs_lsn_t threshold_lsn) | |
574 | { | |
575 | xfs_log_item_t *lip; | |
576 | ||
577 | lip = xfs_ail_min(ailp); | |
578 | if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) || | |
579 | XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0) | |
580 | return; | |
581 | ||
582 | /* | |
583 | * Ensure that the new target is noticed in push code before it clears | |
584 | * the XFS_AIL_PUSHING_BIT. | |
585 | */ | |
586 | smp_wmb(); | |
fe0da767 | 587 | xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn); |
0bf6a5bd DC |
588 | if (!test_and_set_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags)) |
589 | queue_delayed_work(xfs_syncd_wq, &ailp->xa_work, 0); | |
590 | } | |
1da177e4 | 591 | |
fd074841 DC |
592 | /* |
593 | * Push out all items in the AIL immediately | |
594 | */ | |
595 | void | |
596 | xfs_ail_push_all( | |
597 | struct xfs_ail *ailp) | |
598 | { | |
599 | xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp); | |
600 | ||
601 | if (threshold_lsn) | |
602 | xfs_ail_push(ailp, threshold_lsn); | |
603 | } | |
604 | ||
1da177e4 LT |
605 | /* |
606 | * This is to be called when an item is unlocked that may have | |
607 | * been in the AIL. It will wake up the first member of the AIL | |
608 | * wait list if this item's unlocking might allow it to progress. | |
609 | * If the item is in the AIL, then we need to get the AIL lock | |
610 | * while doing our checking so we don't race with someone going | |
611 | * to sleep waiting for this event in xfs_trans_push_ail(). | |
612 | */ | |
613 | void | |
614 | xfs_trans_unlocked_item( | |
783a2f65 | 615 | struct xfs_ail *ailp, |
1da177e4 LT |
616 | xfs_log_item_t *lip) |
617 | { | |
618 | xfs_log_item_t *min_lip; | |
619 | ||
620 | /* | |
621 | * If we're forcibly shutting down, we may have | |
622 | * unlocked log items arbitrarily. The last thing | |
623 | * we want to do is to move the tail of the log | |
624 | * over some potentially valid data. | |
625 | */ | |
626 | if (!(lip->li_flags & XFS_LI_IN_AIL) || | |
783a2f65 | 627 | XFS_FORCED_SHUTDOWN(ailp->xa_mount)) { |
1da177e4 LT |
628 | return; |
629 | } | |
630 | ||
631 | /* | |
632 | * This is the one case where we can call into xfs_ail_min() | |
633 | * without holding the AIL lock because we only care about the | |
634 | * case where we are at the tail of the AIL. If the object isn't | |
635 | * at the tail, it doesn't matter what result we get back. This | |
636 | * is slightly racy because since we were just unlocked, we could | |
637 | * go to sleep between the call to xfs_ail_min and the call to | |
638 | * xfs_log_move_tail, have someone else lock us, commit to us disk, | |
639 | * move us out of the tail of the AIL, and then we wake up. However, | |
640 | * the call to xfs_log_move_tail() doesn't do anything if there's | |
641 | * not enough free space to wake people up so we're safe calling it. | |
642 | */ | |
783a2f65 | 643 | min_lip = xfs_ail_min(ailp); |
1da177e4 LT |
644 | |
645 | if (min_lip == lip) | |
783a2f65 | 646 | xfs_log_move_tail(ailp->xa_mount, 1); |
1da177e4 LT |
647 | } /* xfs_trans_unlocked_item */ |
648 | ||
0e57f6a3 DC |
649 | /* |
650 | * xfs_trans_ail_update - bulk AIL insertion operation. | |
651 | * | |
652 | * @xfs_trans_ail_update takes an array of log items that all need to be | |
653 | * positioned at the same LSN in the AIL. If an item is not in the AIL, it will | |
654 | * be added. Otherwise, it will be repositioned by removing it and re-adding | |
655 | * it to the AIL. If we move the first item in the AIL, update the log tail to | |
656 | * match the new minimum LSN in the AIL. | |
657 | * | |
658 | * This function takes the AIL lock once to execute the update operations on | |
659 | * all the items in the array, and as such should not be called with the AIL | |
660 | * lock held. As a result, once we have the AIL lock, we need to check each log | |
661 | * item LSN to confirm it needs to be moved forward in the AIL. | |
662 | * | |
663 | * To optimise the insert operation, we delete all the items from the AIL in | |
664 | * the first pass, moving them into a temporary list, then splice the temporary | |
665 | * list into the correct position in the AIL. This avoids needing to do an | |
666 | * insert operation on every item. | |
667 | * | |
668 | * This function must be called with the AIL lock held. The lock is dropped | |
669 | * before returning. | |
670 | */ | |
671 | void | |
672 | xfs_trans_ail_update_bulk( | |
673 | struct xfs_ail *ailp, | |
1d8c95a3 | 674 | struct xfs_ail_cursor *cur, |
0e57f6a3 DC |
675 | struct xfs_log_item **log_items, |
676 | int nr_items, | |
677 | xfs_lsn_t lsn) __releases(ailp->xa_lock) | |
678 | { | |
679 | xfs_log_item_t *mlip; | |
680 | xfs_lsn_t tail_lsn; | |
681 | int mlip_changed = 0; | |
682 | int i; | |
683 | LIST_HEAD(tmp); | |
684 | ||
685 | mlip = xfs_ail_min(ailp); | |
686 | ||
687 | for (i = 0; i < nr_items; i++) { | |
688 | struct xfs_log_item *lip = log_items[i]; | |
689 | if (lip->li_flags & XFS_LI_IN_AIL) { | |
690 | /* check if we really need to move the item */ | |
691 | if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0) | |
692 | continue; | |
693 | ||
694 | xfs_ail_delete(ailp, lip); | |
695 | if (mlip == lip) | |
696 | mlip_changed = 1; | |
697 | } else { | |
698 | lip->li_flags |= XFS_LI_IN_AIL; | |
699 | } | |
700 | lip->li_lsn = lsn; | |
701 | list_add(&lip->li_ail, &tmp); | |
702 | } | |
703 | ||
1d8c95a3 | 704 | xfs_ail_splice(ailp, cur, &tmp, lsn); |
0e57f6a3 DC |
705 | |
706 | if (!mlip_changed) { | |
707 | spin_unlock(&ailp->xa_lock); | |
708 | return; | |
709 | } | |
710 | ||
711 | /* | |
712 | * It is not safe to access mlip after the AIL lock is dropped, so we | |
713 | * must get a copy of li_lsn before we do so. This is especially | |
714 | * important on 32-bit platforms where accessing and updating 64-bit | |
715 | * values like li_lsn is not atomic. | |
716 | */ | |
717 | mlip = xfs_ail_min(ailp); | |
718 | tail_lsn = mlip->li_lsn; | |
719 | spin_unlock(&ailp->xa_lock); | |
720 | xfs_log_move_tail(ailp->xa_mount, tail_lsn); | |
721 | } | |
722 | ||
30136832 DC |
723 | /* |
724 | * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL | |
725 | * | |
726 | * @xfs_trans_ail_delete_bulk takes an array of log items that all need to | |
727 | * removed from the AIL. The caller is already holding the AIL lock, and done | |
728 | * all the checks necessary to ensure the items passed in via @log_items are | |
729 | * ready for deletion. This includes checking that the items are in the AIL. | |
730 | * | |
731 | * For each log item to be removed, unlink it from the AIL, clear the IN_AIL | |
732 | * flag from the item and reset the item's lsn to 0. If we remove the first | |
733 | * item in the AIL, update the log tail to match the new minimum LSN in the | |
734 | * AIL. | |
735 | * | |
736 | * This function will not drop the AIL lock until all items are removed from | |
737 | * the AIL to minimise the amount of lock traffic on the AIL. This does not | |
738 | * greatly increase the AIL hold time, but does significantly reduce the amount | |
739 | * of traffic on the lock, especially during IO completion. | |
740 | * | |
741 | * This function must be called with the AIL lock held. The lock is dropped | |
742 | * before returning. | |
743 | */ | |
744 | void | |
745 | xfs_trans_ail_delete_bulk( | |
746 | struct xfs_ail *ailp, | |
747 | struct xfs_log_item **log_items, | |
748 | int nr_items) __releases(ailp->xa_lock) | |
749 | { | |
750 | xfs_log_item_t *mlip; | |
751 | xfs_lsn_t tail_lsn; | |
752 | int mlip_changed = 0; | |
753 | int i; | |
754 | ||
755 | mlip = xfs_ail_min(ailp); | |
756 | ||
757 | for (i = 0; i < nr_items; i++) { | |
758 | struct xfs_log_item *lip = log_items[i]; | |
759 | if (!(lip->li_flags & XFS_LI_IN_AIL)) { | |
760 | struct xfs_mount *mp = ailp->xa_mount; | |
761 | ||
762 | spin_unlock(&ailp->xa_lock); | |
763 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
6a19d939 | 764 | xfs_alert_tag(mp, XFS_PTAG_AILDELETE, |
30136832 DC |
765 | "%s: attempting to delete a log item that is not in the AIL", |
766 | __func__); | |
767 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
768 | } | |
769 | return; | |
770 | } | |
771 | ||
772 | xfs_ail_delete(ailp, lip); | |
773 | lip->li_flags &= ~XFS_LI_IN_AIL; | |
774 | lip->li_lsn = 0; | |
775 | if (mlip == lip) | |
776 | mlip_changed = 1; | |
777 | } | |
778 | ||
779 | if (!mlip_changed) { | |
780 | spin_unlock(&ailp->xa_lock); | |
781 | return; | |
782 | } | |
783 | ||
784 | /* | |
785 | * It is not safe to access mlip after the AIL lock is dropped, so we | |
786 | * must get a copy of li_lsn before we do so. This is especially | |
787 | * important on 32-bit platforms where accessing and updating 64-bit | |
788 | * values like li_lsn is not atomic. It is possible we've emptied the | |
789 | * AIL here, so if that is the case, pass an LSN of 0 to the tail move. | |
790 | */ | |
791 | mlip = xfs_ail_min(ailp); | |
792 | tail_lsn = mlip ? mlip->li_lsn : 0; | |
793 | spin_unlock(&ailp->xa_lock); | |
794 | xfs_log_move_tail(ailp->xa_mount, tail_lsn); | |
795 | } | |
1da177e4 | 796 | |
1da177e4 LT |
797 | /* |
798 | * The active item list (AIL) is a doubly linked list of log | |
799 | * items sorted by ascending lsn. The base of the list is | |
800 | * a forw/back pointer pair embedded in the xfs mount structure. | |
801 | * The base is initialized with both pointers pointing to the | |
802 | * base. This case always needs to be distinguished, because | |
803 | * the base has no lsn to look at. We almost always insert | |
804 | * at the end of the list, so on inserts we search from the | |
805 | * end of the list to find where the new item belongs. | |
806 | */ | |
807 | ||
808 | /* | |
809 | * Initialize the doubly linked list to point only to itself. | |
810 | */ | |
249a8c11 | 811 | int |
1da177e4 LT |
812 | xfs_trans_ail_init( |
813 | xfs_mount_t *mp) | |
814 | { | |
82fa9012 DC |
815 | struct xfs_ail *ailp; |
816 | ||
817 | ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL); | |
818 | if (!ailp) | |
819 | return ENOMEM; | |
820 | ||
821 | ailp->xa_mount = mp; | |
822 | INIT_LIST_HEAD(&ailp->xa_ail); | |
af3e4022 | 823 | INIT_LIST_HEAD(&ailp->xa_cursors); |
c7e8f268 | 824 | spin_lock_init(&ailp->xa_lock); |
0bf6a5bd | 825 | INIT_DELAYED_WORK(&ailp->xa_work, xfs_ail_worker); |
27d8d5fe DC |
826 | mp->m_ail = ailp; |
827 | return 0; | |
249a8c11 DC |
828 | } |
829 | ||
830 | void | |
831 | xfs_trans_ail_destroy( | |
832 | xfs_mount_t *mp) | |
833 | { | |
82fa9012 DC |
834 | struct xfs_ail *ailp = mp->m_ail; |
835 | ||
0bf6a5bd | 836 | cancel_delayed_work_sync(&ailp->xa_work); |
82fa9012 | 837 | kmem_free(ailp); |
1da177e4 | 838 | } |