Loading...
1/*
2 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_mru_cache.h"
20
21/*
22 * The MRU Cache data structure consists of a data store, an array of lists and
23 * a lock to protect its internal state. At initialisation time, the client
24 * supplies an element lifetime in milliseconds and a group count, as well as a
25 * function pointer to call when deleting elements. A data structure for
26 * queueing up work in the form of timed callbacks is also included.
27 *
28 * The group count controls how many lists are created, and thereby how finely
29 * the elements are grouped in time. When reaping occurs, all the elements in
30 * all the lists whose time has expired are deleted.
31 *
32 * To give an example of how this works in practice, consider a client that
33 * initialises an MRU Cache with a lifetime of ten seconds and a group count of
34 * five. Five internal lists will be created, each representing a two second
35 * period in time. When the first element is added, time zero for the data
36 * structure is initialised to the current time.
37 *
38 * All the elements added in the first two seconds are appended to the first
39 * list. Elements added in the third second go into the second list, and so on.
40 * If an element is accessed at any point, it is removed from its list and
41 * inserted at the head of the current most-recently-used list.
42 *
43 * The reaper function will have nothing to do until at least twelve seconds
44 * have elapsed since the first element was added. The reason for this is that
45 * if it were called at t=11s, there could be elements in the first list that
46 * have only been inactive for nine seconds, so it still does nothing. If it is
47 * called anywhere between t=12 and t=14 seconds, it will delete all the
48 * elements that remain in the first list. It's therefore possible for elements
49 * to remain in the data store even after they've been inactive for up to
50 * (t + t/g) seconds, where t is the inactive element lifetime and g is the
51 * number of groups.
52 *
53 * The above example assumes that the reaper function gets called at least once
54 * every (t/g) seconds. If it is called less frequently, unused elements will
55 * accumulate in the reap list until the reaper function is eventually called.
56 * The current implementation uses work queue callbacks to carefully time the
57 * reaper function calls, so this should happen rarely, if at all.
58 *
59 * From a design perspective, the primary reason for the choice of a list array
60 * representing discrete time intervals is that it's only practical to reap
61 * expired elements in groups of some appreciable size. This automatically
62 * introduces a granularity to element lifetimes, so there's no point storing an
63 * individual timeout with each element that specifies a more precise reap time.
64 * The bonus is a saving of sizeof(long) bytes of memory per element stored.
65 *
66 * The elements could have been stored in just one list, but an array of
67 * counters or pointers would need to be maintained to allow them to be divided
68 * up into discrete time groups. More critically, the process of touching or
69 * removing an element would involve walking large portions of the entire list,
70 * which would have a detrimental effect on performance. The additional memory
71 * requirement for the array of list heads is minimal.
72 *
73 * When an element is touched or deleted, it needs to be removed from its
74 * current list. Doubly linked lists are used to make the list maintenance
75 * portion of these operations O(1). Since reaper timing can be imprecise,
76 * inserts and lookups can occur when there are no free lists available. When
77 * this happens, all the elements on the LRU list need to be migrated to the end
78 * of the reap list. To keep the list maintenance portion of these operations
79 * O(1) also, list tails need to be accessible without walking the entire list.
80 * This is the reason why doubly linked list heads are used.
81 */
82
83/*
84 * An MRU Cache is a dynamic data structure that stores its elements in a way
85 * that allows efficient lookups, but also groups them into discrete time
86 * intervals based on insertion time. This allows elements to be efficiently
87 * and automatically reaped after a fixed period of inactivity.
88 *
89 * When a client data pointer is stored in the MRU Cache it needs to be added to
90 * both the data store and to one of the lists. It must also be possible to
91 * access each of these entries via the other, i.e. to:
92 *
93 * a) Walk a list, removing the corresponding data store entry for each item.
94 * b) Look up a data store entry, then access its list entry directly.
95 *
96 * To achieve both of these goals, each entry must contain both a list entry and
97 * a key, in addition to the user's data pointer. Note that it's not a good
98 * idea to have the client embed one of these structures at the top of their own
99 * data structure, because inserting the same item more than once would most
100 * likely result in a loop in one of the lists. That's a sure-fire recipe for
101 * an infinite loop in the code.
102 */
103struct xfs_mru_cache {
104 struct radix_tree_root store; /* Core storage data structure. */
105 struct list_head *lists; /* Array of lists, one per grp. */
106 struct list_head reap_list; /* Elements overdue for reaping. */
107 spinlock_t lock; /* Lock to protect this struct. */
108 unsigned int grp_count; /* Number of discrete groups. */
109 unsigned int grp_time; /* Time period spanned by grps. */
110 unsigned int lru_grp; /* Group containing time zero. */
111 unsigned long time_zero; /* Time first element was added. */
112 xfs_mru_cache_free_func_t free_func; /* Function pointer for freeing. */
113 struct delayed_work work; /* Workqueue data for reaping. */
114 unsigned int queued; /* work has been queued */
115 void *data;
116};
117
118static struct workqueue_struct *xfs_mru_reap_wq;
119
120/*
121 * When inserting, destroying or reaping, it's first necessary to update the
122 * lists relative to a particular time. In the case of destroying, that time
123 * will be well in the future to ensure that all items are moved to the reap
124 * list. In all other cases though, the time will be the current time.
125 *
126 * This function enters a loop, moving the contents of the LRU list to the reap
127 * list again and again until either a) the lists are all empty, or b) time zero
128 * has been advanced sufficiently to be within the immediate element lifetime.
129 *
130 * Case a) above is detected by counting how many groups are migrated and
131 * stopping when they've all been moved. Case b) is detected by monitoring the
132 * time_zero field, which is updated as each group is migrated.
133 *
134 * The return value is the earliest time that more migration could be needed, or
135 * zero if there's no need to schedule more work because the lists are empty.
136 */
137STATIC unsigned long
138_xfs_mru_cache_migrate(
139 struct xfs_mru_cache *mru,
140 unsigned long now)
141{
142 unsigned int grp;
143 unsigned int migrated = 0;
144 struct list_head *lru_list;
145
146 /* Nothing to do if the data store is empty. */
147 if (!mru->time_zero)
148 return 0;
149
150 /* While time zero is older than the time spanned by all the lists. */
151 while (mru->time_zero <= now - mru->grp_count * mru->grp_time) {
152
153 /*
154 * If the LRU list isn't empty, migrate its elements to the tail
155 * of the reap list.
156 */
157 lru_list = mru->lists + mru->lru_grp;
158 if (!list_empty(lru_list))
159 list_splice_init(lru_list, mru->reap_list.prev);
160
161 /*
162 * Advance the LRU group number, freeing the old LRU list to
163 * become the new MRU list; advance time zero accordingly.
164 */
165 mru->lru_grp = (mru->lru_grp + 1) % mru->grp_count;
166 mru->time_zero += mru->grp_time;
167
168 /*
169 * If reaping is so far behind that all the elements on all the
170 * lists have been migrated to the reap list, it's now empty.
171 */
172 if (++migrated == mru->grp_count) {
173 mru->lru_grp = 0;
174 mru->time_zero = 0;
175 return 0;
176 }
177 }
178
179 /* Find the first non-empty list from the LRU end. */
180 for (grp = 0; grp < mru->grp_count; grp++) {
181
182 /* Check the grp'th list from the LRU end. */
183 lru_list = mru->lists + ((mru->lru_grp + grp) % mru->grp_count);
184 if (!list_empty(lru_list))
185 return mru->time_zero +
186 (mru->grp_count + grp) * mru->grp_time;
187 }
188
189 /* All the lists must be empty. */
190 mru->lru_grp = 0;
191 mru->time_zero = 0;
192 return 0;
193}
194
195/*
196 * When inserting or doing a lookup, an element needs to be inserted into the
197 * MRU list. The lists must be migrated first to ensure that they're
198 * up-to-date, otherwise the new element could be given a shorter lifetime in
199 * the cache than it should.
200 */
201STATIC void
202_xfs_mru_cache_list_insert(
203 struct xfs_mru_cache *mru,
204 struct xfs_mru_cache_elem *elem)
205{
206 unsigned int grp = 0;
207 unsigned long now = jiffies;
208
209 /*
210 * If the data store is empty, initialise time zero, leave grp set to
211 * zero and start the work queue timer if necessary. Otherwise, set grp
212 * to the number of group times that have elapsed since time zero.
213 */
214 if (!_xfs_mru_cache_migrate(mru, now)) {
215 mru->time_zero = now;
216 if (!mru->queued) {
217 mru->queued = 1;
218 queue_delayed_work(xfs_mru_reap_wq, &mru->work,
219 mru->grp_count * mru->grp_time);
220 }
221 } else {
222 grp = (now - mru->time_zero) / mru->grp_time;
223 grp = (mru->lru_grp + grp) % mru->grp_count;
224 }
225
226 /* Insert the element at the tail of the corresponding list. */
227 list_add_tail(&elem->list_node, mru->lists + grp);
228}
229
230/*
231 * When destroying or reaping, all the elements that were migrated to the reap
232 * list need to be deleted. For each element this involves removing it from the
233 * data store, removing it from the reap list, calling the client's free
234 * function and deleting the element from the element zone.
235 *
236 * We get called holding the mru->lock, which we drop and then reacquire.
237 * Sparse need special help with this to tell it we know what we are doing.
238 */
239STATIC void
240_xfs_mru_cache_clear_reap_list(
241 struct xfs_mru_cache *mru)
242 __releases(mru->lock) __acquires(mru->lock)
243{
244 struct xfs_mru_cache_elem *elem, *next;
245 struct list_head tmp;
246
247 INIT_LIST_HEAD(&tmp);
248 list_for_each_entry_safe(elem, next, &mru->reap_list, list_node) {
249
250 /* Remove the element from the data store. */
251 radix_tree_delete(&mru->store, elem->key);
252
253 /*
254 * remove to temp list so it can be freed without
255 * needing to hold the lock
256 */
257 list_move(&elem->list_node, &tmp);
258 }
259 spin_unlock(&mru->lock);
260
261 list_for_each_entry_safe(elem, next, &tmp, list_node) {
262 list_del_init(&elem->list_node);
263 mru->free_func(mru->data, elem);
264 }
265
266 spin_lock(&mru->lock);
267}
268
269/*
270 * We fire the reap timer every group expiry interval so
271 * we always have a reaper ready to run. This makes shutdown
272 * and flushing of the reaper easy to do. Hence we need to
273 * keep when the next reap must occur so we can determine
274 * at each interval whether there is anything we need to do.
275 */
276STATIC void
277_xfs_mru_cache_reap(
278 struct work_struct *work)
279{
280 struct xfs_mru_cache *mru =
281 container_of(work, struct xfs_mru_cache, work.work);
282 unsigned long now, next;
283
284 ASSERT(mru && mru->lists);
285 if (!mru || !mru->lists)
286 return;
287
288 spin_lock(&mru->lock);
289 next = _xfs_mru_cache_migrate(mru, jiffies);
290 _xfs_mru_cache_clear_reap_list(mru);
291
292 mru->queued = next;
293 if ((mru->queued > 0)) {
294 now = jiffies;
295 if (next <= now)
296 next = 0;
297 else
298 next -= now;
299 queue_delayed_work(xfs_mru_reap_wq, &mru->work, next);
300 }
301
302 spin_unlock(&mru->lock);
303}
304
305int
306xfs_mru_cache_init(void)
307{
308 xfs_mru_reap_wq = alloc_workqueue("xfs_mru_cache",
309 WQ_MEM_RECLAIM|WQ_FREEZABLE, 1);
310 if (!xfs_mru_reap_wq)
311 return -ENOMEM;
312 return 0;
313}
314
315void
316xfs_mru_cache_uninit(void)
317{
318 destroy_workqueue(xfs_mru_reap_wq);
319}
320
321/*
322 * To initialise a struct xfs_mru_cache pointer, call xfs_mru_cache_create()
323 * with the address of the pointer, a lifetime value in milliseconds, a group
324 * count and a free function to use when deleting elements. This function
325 * returns 0 if the initialisation was successful.
326 */
327int
328xfs_mru_cache_create(
329 struct xfs_mru_cache **mrup,
330 void *data,
331 unsigned int lifetime_ms,
332 unsigned int grp_count,
333 xfs_mru_cache_free_func_t free_func)
334{
335 struct xfs_mru_cache *mru = NULL;
336 int err = 0, grp;
337 unsigned int grp_time;
338
339 if (mrup)
340 *mrup = NULL;
341
342 if (!mrup || !grp_count || !lifetime_ms || !free_func)
343 return -EINVAL;
344
345 if (!(grp_time = msecs_to_jiffies(lifetime_ms) / grp_count))
346 return -EINVAL;
347
348 if (!(mru = kmem_zalloc(sizeof(*mru), KM_SLEEP)))
349 return -ENOMEM;
350
351 /* An extra list is needed to avoid reaping up to a grp_time early. */
352 mru->grp_count = grp_count + 1;
353 mru->lists = kmem_zalloc(mru->grp_count * sizeof(*mru->lists), KM_SLEEP);
354
355 if (!mru->lists) {
356 err = -ENOMEM;
357 goto exit;
358 }
359
360 for (grp = 0; grp < mru->grp_count; grp++)
361 INIT_LIST_HEAD(mru->lists + grp);
362
363 /*
364 * We use GFP_KERNEL radix tree preload and do inserts under a
365 * spinlock so GFP_ATOMIC is appropriate for the radix tree itself.
366 */
367 INIT_RADIX_TREE(&mru->store, GFP_ATOMIC);
368 INIT_LIST_HEAD(&mru->reap_list);
369 spin_lock_init(&mru->lock);
370 INIT_DELAYED_WORK(&mru->work, _xfs_mru_cache_reap);
371
372 mru->grp_time = grp_time;
373 mru->free_func = free_func;
374 mru->data = data;
375 *mrup = mru;
376
377exit:
378 if (err && mru && mru->lists)
379 kmem_free(mru->lists);
380 if (err && mru)
381 kmem_free(mru);
382
383 return err;
384}
385
386/*
387 * Call xfs_mru_cache_flush() to flush out all cached entries, calling their
388 * free functions as they're deleted. When this function returns, the caller is
389 * guaranteed that all the free functions for all the elements have finished
390 * executing and the reaper is not running.
391 */
392static void
393xfs_mru_cache_flush(
394 struct xfs_mru_cache *mru)
395{
396 if (!mru || !mru->lists)
397 return;
398
399 spin_lock(&mru->lock);
400 if (mru->queued) {
401 spin_unlock(&mru->lock);
402 cancel_delayed_work_sync(&mru->work);
403 spin_lock(&mru->lock);
404 }
405
406 _xfs_mru_cache_migrate(mru, jiffies + mru->grp_count * mru->grp_time);
407 _xfs_mru_cache_clear_reap_list(mru);
408
409 spin_unlock(&mru->lock);
410}
411
412void
413xfs_mru_cache_destroy(
414 struct xfs_mru_cache *mru)
415{
416 if (!mru || !mru->lists)
417 return;
418
419 xfs_mru_cache_flush(mru);
420
421 kmem_free(mru->lists);
422 kmem_free(mru);
423}
424
425/*
426 * To insert an element, call xfs_mru_cache_insert() with the data store, the
427 * element's key and the client data pointer. This function returns 0 on
428 * success or ENOMEM if memory for the data element couldn't be allocated.
429 */
430int
431xfs_mru_cache_insert(
432 struct xfs_mru_cache *mru,
433 unsigned long key,
434 struct xfs_mru_cache_elem *elem)
435{
436 int error;
437
438 ASSERT(mru && mru->lists);
439 if (!mru || !mru->lists)
440 return -EINVAL;
441
442 if (radix_tree_preload(GFP_NOFS))
443 return -ENOMEM;
444
445 INIT_LIST_HEAD(&elem->list_node);
446 elem->key = key;
447
448 spin_lock(&mru->lock);
449 error = radix_tree_insert(&mru->store, key, elem);
450 radix_tree_preload_end();
451 if (!error)
452 _xfs_mru_cache_list_insert(mru, elem);
453 spin_unlock(&mru->lock);
454
455 return error;
456}
457
458/*
459 * To remove an element without calling the free function, call
460 * xfs_mru_cache_remove() with the data store and the element's key. On success
461 * the client data pointer for the removed element is returned, otherwise this
462 * function will return a NULL pointer.
463 */
464struct xfs_mru_cache_elem *
465xfs_mru_cache_remove(
466 struct xfs_mru_cache *mru,
467 unsigned long key)
468{
469 struct xfs_mru_cache_elem *elem;
470
471 ASSERT(mru && mru->lists);
472 if (!mru || !mru->lists)
473 return NULL;
474
475 spin_lock(&mru->lock);
476 elem = radix_tree_delete(&mru->store, key);
477 if (elem)
478 list_del(&elem->list_node);
479 spin_unlock(&mru->lock);
480
481 return elem;
482}
483
484/*
485 * To remove and element and call the free function, call xfs_mru_cache_delete()
486 * with the data store and the element's key.
487 */
488void
489xfs_mru_cache_delete(
490 struct xfs_mru_cache *mru,
491 unsigned long key)
492{
493 struct xfs_mru_cache_elem *elem;
494
495 elem = xfs_mru_cache_remove(mru, key);
496 if (elem)
497 mru->free_func(mru->data, elem);
498}
499
500/*
501 * To look up an element using its key, call xfs_mru_cache_lookup() with the
502 * data store and the element's key. If found, the element will be moved to the
503 * head of the MRU list to indicate that it's been touched.
504 *
505 * The internal data structures are protected by a spinlock that is STILL HELD
506 * when this function returns. Call xfs_mru_cache_done() to release it. Note
507 * that it is not safe to call any function that might sleep in the interim.
508 *
509 * The implementation could have used reference counting to avoid this
510 * restriction, but since most clients simply want to get, set or test a member
511 * of the returned data structure, the extra per-element memory isn't warranted.
512 *
513 * If the element isn't found, this function returns NULL and the spinlock is
514 * released. xfs_mru_cache_done() should NOT be called when this occurs.
515 *
516 * Because sparse isn't smart enough to know about conditional lock return
517 * status, we need to help it get it right by annotating the path that does
518 * not release the lock.
519 */
520struct xfs_mru_cache_elem *
521xfs_mru_cache_lookup(
522 struct xfs_mru_cache *mru,
523 unsigned long key)
524{
525 struct xfs_mru_cache_elem *elem;
526
527 ASSERT(mru && mru->lists);
528 if (!mru || !mru->lists)
529 return NULL;
530
531 spin_lock(&mru->lock);
532 elem = radix_tree_lookup(&mru->store, key);
533 if (elem) {
534 list_del(&elem->list_node);
535 _xfs_mru_cache_list_insert(mru, elem);
536 __release(mru_lock); /* help sparse not be stupid */
537 } else
538 spin_unlock(&mru->lock);
539
540 return elem;
541}
542
543/*
544 * To release the internal data structure spinlock after having performed an
545 * xfs_mru_cache_lookup() or an xfs_mru_cache_peek(), call xfs_mru_cache_done()
546 * with the data store pointer.
547 */
548void
549xfs_mru_cache_done(
550 struct xfs_mru_cache *mru)
551 __releases(mru->lock)
552{
553 spin_unlock(&mru->lock);
554}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2006-2007 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_mru_cache.h"
8
9/*
10 * The MRU Cache data structure consists of a data store, an array of lists and
11 * a lock to protect its internal state. At initialisation time, the client
12 * supplies an element lifetime in milliseconds and a group count, as well as a
13 * function pointer to call when deleting elements. A data structure for
14 * queueing up work in the form of timed callbacks is also included.
15 *
16 * The group count controls how many lists are created, and thereby how finely
17 * the elements are grouped in time. When reaping occurs, all the elements in
18 * all the lists whose time has expired are deleted.
19 *
20 * To give an example of how this works in practice, consider a client that
21 * initialises an MRU Cache with a lifetime of ten seconds and a group count of
22 * five. Five internal lists will be created, each representing a two second
23 * period in time. When the first element is added, time zero for the data
24 * structure is initialised to the current time.
25 *
26 * All the elements added in the first two seconds are appended to the first
27 * list. Elements added in the third second go into the second list, and so on.
28 * If an element is accessed at any point, it is removed from its list and
29 * inserted at the head of the current most-recently-used list.
30 *
31 * The reaper function will have nothing to do until at least twelve seconds
32 * have elapsed since the first element was added. The reason for this is that
33 * if it were called at t=11s, there could be elements in the first list that
34 * have only been inactive for nine seconds, so it still does nothing. If it is
35 * called anywhere between t=12 and t=14 seconds, it will delete all the
36 * elements that remain in the first list. It's therefore possible for elements
37 * to remain in the data store even after they've been inactive for up to
38 * (t + t/g) seconds, where t is the inactive element lifetime and g is the
39 * number of groups.
40 *
41 * The above example assumes that the reaper function gets called at least once
42 * every (t/g) seconds. If it is called less frequently, unused elements will
43 * accumulate in the reap list until the reaper function is eventually called.
44 * The current implementation uses work queue callbacks to carefully time the
45 * reaper function calls, so this should happen rarely, if at all.
46 *
47 * From a design perspective, the primary reason for the choice of a list array
48 * representing discrete time intervals is that it's only practical to reap
49 * expired elements in groups of some appreciable size. This automatically
50 * introduces a granularity to element lifetimes, so there's no point storing an
51 * individual timeout with each element that specifies a more precise reap time.
52 * The bonus is a saving of sizeof(long) bytes of memory per element stored.
53 *
54 * The elements could have been stored in just one list, but an array of
55 * counters or pointers would need to be maintained to allow them to be divided
56 * up into discrete time groups. More critically, the process of touching or
57 * removing an element would involve walking large portions of the entire list,
58 * which would have a detrimental effect on performance. The additional memory
59 * requirement for the array of list heads is minimal.
60 *
61 * When an element is touched or deleted, it needs to be removed from its
62 * current list. Doubly linked lists are used to make the list maintenance
63 * portion of these operations O(1). Since reaper timing can be imprecise,
64 * inserts and lookups can occur when there are no free lists available. When
65 * this happens, all the elements on the LRU list need to be migrated to the end
66 * of the reap list. To keep the list maintenance portion of these operations
67 * O(1) also, list tails need to be accessible without walking the entire list.
68 * This is the reason why doubly linked list heads are used.
69 */
70
71/*
72 * An MRU Cache is a dynamic data structure that stores its elements in a way
73 * that allows efficient lookups, but also groups them into discrete time
74 * intervals based on insertion time. This allows elements to be efficiently
75 * and automatically reaped after a fixed period of inactivity.
76 *
77 * When a client data pointer is stored in the MRU Cache it needs to be added to
78 * both the data store and to one of the lists. It must also be possible to
79 * access each of these entries via the other, i.e. to:
80 *
81 * a) Walk a list, removing the corresponding data store entry for each item.
82 * b) Look up a data store entry, then access its list entry directly.
83 *
84 * To achieve both of these goals, each entry must contain both a list entry and
85 * a key, in addition to the user's data pointer. Note that it's not a good
86 * idea to have the client embed one of these structures at the top of their own
87 * data structure, because inserting the same item more than once would most
88 * likely result in a loop in one of the lists. That's a sure-fire recipe for
89 * an infinite loop in the code.
90 */
91struct xfs_mru_cache {
92 struct radix_tree_root store; /* Core storage data structure. */
93 struct list_head *lists; /* Array of lists, one per grp. */
94 struct list_head reap_list; /* Elements overdue for reaping. */
95 spinlock_t lock; /* Lock to protect this struct. */
96 unsigned int grp_count; /* Number of discrete groups. */
97 unsigned int grp_time; /* Time period spanned by grps. */
98 unsigned int lru_grp; /* Group containing time zero. */
99 unsigned long time_zero; /* Time first element was added. */
100 xfs_mru_cache_free_func_t free_func; /* Function pointer for freeing. */
101 struct delayed_work work; /* Workqueue data for reaping. */
102 unsigned int queued; /* work has been queued */
103 void *data;
104};
105
106static struct workqueue_struct *xfs_mru_reap_wq;
107
108/*
109 * When inserting, destroying or reaping, it's first necessary to update the
110 * lists relative to a particular time. In the case of destroying, that time
111 * will be well in the future to ensure that all items are moved to the reap
112 * list. In all other cases though, the time will be the current time.
113 *
114 * This function enters a loop, moving the contents of the LRU list to the reap
115 * list again and again until either a) the lists are all empty, or b) time zero
116 * has been advanced sufficiently to be within the immediate element lifetime.
117 *
118 * Case a) above is detected by counting how many groups are migrated and
119 * stopping when they've all been moved. Case b) is detected by monitoring the
120 * time_zero field, which is updated as each group is migrated.
121 *
122 * The return value is the earliest time that more migration could be needed, or
123 * zero if there's no need to schedule more work because the lists are empty.
124 */
125STATIC unsigned long
126_xfs_mru_cache_migrate(
127 struct xfs_mru_cache *mru,
128 unsigned long now)
129{
130 unsigned int grp;
131 unsigned int migrated = 0;
132 struct list_head *lru_list;
133
134 /* Nothing to do if the data store is empty. */
135 if (!mru->time_zero)
136 return 0;
137
138 /* While time zero is older than the time spanned by all the lists. */
139 while (mru->time_zero <= now - mru->grp_count * mru->grp_time) {
140
141 /*
142 * If the LRU list isn't empty, migrate its elements to the tail
143 * of the reap list.
144 */
145 lru_list = mru->lists + mru->lru_grp;
146 if (!list_empty(lru_list))
147 list_splice_init(lru_list, mru->reap_list.prev);
148
149 /*
150 * Advance the LRU group number, freeing the old LRU list to
151 * become the new MRU list; advance time zero accordingly.
152 */
153 mru->lru_grp = (mru->lru_grp + 1) % mru->grp_count;
154 mru->time_zero += mru->grp_time;
155
156 /*
157 * If reaping is so far behind that all the elements on all the
158 * lists have been migrated to the reap list, it's now empty.
159 */
160 if (++migrated == mru->grp_count) {
161 mru->lru_grp = 0;
162 mru->time_zero = 0;
163 return 0;
164 }
165 }
166
167 /* Find the first non-empty list from the LRU end. */
168 for (grp = 0; grp < mru->grp_count; grp++) {
169
170 /* Check the grp'th list from the LRU end. */
171 lru_list = mru->lists + ((mru->lru_grp + grp) % mru->grp_count);
172 if (!list_empty(lru_list))
173 return mru->time_zero +
174 (mru->grp_count + grp) * mru->grp_time;
175 }
176
177 /* All the lists must be empty. */
178 mru->lru_grp = 0;
179 mru->time_zero = 0;
180 return 0;
181}
182
183/*
184 * When inserting or doing a lookup, an element needs to be inserted into the
185 * MRU list. The lists must be migrated first to ensure that they're
186 * up-to-date, otherwise the new element could be given a shorter lifetime in
187 * the cache than it should.
188 */
189STATIC void
190_xfs_mru_cache_list_insert(
191 struct xfs_mru_cache *mru,
192 struct xfs_mru_cache_elem *elem)
193{
194 unsigned int grp = 0;
195 unsigned long now = jiffies;
196
197 /*
198 * If the data store is empty, initialise time zero, leave grp set to
199 * zero and start the work queue timer if necessary. Otherwise, set grp
200 * to the number of group times that have elapsed since time zero.
201 */
202 if (!_xfs_mru_cache_migrate(mru, now)) {
203 mru->time_zero = now;
204 if (!mru->queued) {
205 mru->queued = 1;
206 queue_delayed_work(xfs_mru_reap_wq, &mru->work,
207 mru->grp_count * mru->grp_time);
208 }
209 } else {
210 grp = (now - mru->time_zero) / mru->grp_time;
211 grp = (mru->lru_grp + grp) % mru->grp_count;
212 }
213
214 /* Insert the element at the tail of the corresponding list. */
215 list_add_tail(&elem->list_node, mru->lists + grp);
216}
217
218/*
219 * When destroying or reaping, all the elements that were migrated to the reap
220 * list need to be deleted. For each element this involves removing it from the
221 * data store, removing it from the reap list, calling the client's free
222 * function and deleting the element from the element cache.
223 *
224 * We get called holding the mru->lock, which we drop and then reacquire.
225 * Sparse need special help with this to tell it we know what we are doing.
226 */
227STATIC void
228_xfs_mru_cache_clear_reap_list(
229 struct xfs_mru_cache *mru)
230 __releases(mru->lock) __acquires(mru->lock)
231{
232 struct xfs_mru_cache_elem *elem, *next;
233 struct list_head tmp;
234
235 INIT_LIST_HEAD(&tmp);
236 list_for_each_entry_safe(elem, next, &mru->reap_list, list_node) {
237
238 /* Remove the element from the data store. */
239 radix_tree_delete(&mru->store, elem->key);
240
241 /*
242 * remove to temp list so it can be freed without
243 * needing to hold the lock
244 */
245 list_move(&elem->list_node, &tmp);
246 }
247 spin_unlock(&mru->lock);
248
249 list_for_each_entry_safe(elem, next, &tmp, list_node) {
250 list_del_init(&elem->list_node);
251 mru->free_func(mru->data, elem);
252 }
253
254 spin_lock(&mru->lock);
255}
256
257/*
258 * We fire the reap timer every group expiry interval so
259 * we always have a reaper ready to run. This makes shutdown
260 * and flushing of the reaper easy to do. Hence we need to
261 * keep when the next reap must occur so we can determine
262 * at each interval whether there is anything we need to do.
263 */
264STATIC void
265_xfs_mru_cache_reap(
266 struct work_struct *work)
267{
268 struct xfs_mru_cache *mru =
269 container_of(work, struct xfs_mru_cache, work.work);
270 unsigned long now, next;
271
272 ASSERT(mru && mru->lists);
273 if (!mru || !mru->lists)
274 return;
275
276 spin_lock(&mru->lock);
277 next = _xfs_mru_cache_migrate(mru, jiffies);
278 _xfs_mru_cache_clear_reap_list(mru);
279
280 mru->queued = next;
281 if ((mru->queued > 0)) {
282 now = jiffies;
283 if (next <= now)
284 next = 0;
285 else
286 next -= now;
287 queue_delayed_work(xfs_mru_reap_wq, &mru->work, next);
288 }
289
290 spin_unlock(&mru->lock);
291}
292
293int
294xfs_mru_cache_init(void)
295{
296 xfs_mru_reap_wq = alloc_workqueue("xfs_mru_cache",
297 XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 1);
298 if (!xfs_mru_reap_wq)
299 return -ENOMEM;
300 return 0;
301}
302
303void
304xfs_mru_cache_uninit(void)
305{
306 destroy_workqueue(xfs_mru_reap_wq);
307}
308
309/*
310 * To initialise a struct xfs_mru_cache pointer, call xfs_mru_cache_create()
311 * with the address of the pointer, a lifetime value in milliseconds, a group
312 * count and a free function to use when deleting elements. This function
313 * returns 0 if the initialisation was successful.
314 */
315int
316xfs_mru_cache_create(
317 struct xfs_mru_cache **mrup,
318 void *data,
319 unsigned int lifetime_ms,
320 unsigned int grp_count,
321 xfs_mru_cache_free_func_t free_func)
322{
323 struct xfs_mru_cache *mru = NULL;
324 int err = 0, grp;
325 unsigned int grp_time;
326
327 if (mrup)
328 *mrup = NULL;
329
330 if (!mrup || !grp_count || !lifetime_ms || !free_func)
331 return -EINVAL;
332
333 if (!(grp_time = msecs_to_jiffies(lifetime_ms) / grp_count))
334 return -EINVAL;
335
336 mru = kzalloc(sizeof(*mru), GFP_KERNEL | __GFP_NOFAIL);
337 if (!mru)
338 return -ENOMEM;
339
340 /* An extra list is needed to avoid reaping up to a grp_time early. */
341 mru->grp_count = grp_count + 1;
342 mru->lists = kzalloc(mru->grp_count * sizeof(*mru->lists),
343 GFP_KERNEL | __GFP_NOFAIL);
344 if (!mru->lists) {
345 err = -ENOMEM;
346 goto exit;
347 }
348
349 for (grp = 0; grp < mru->grp_count; grp++)
350 INIT_LIST_HEAD(mru->lists + grp);
351
352 /*
353 * We use GFP_KERNEL radix tree preload and do inserts under a
354 * spinlock so GFP_ATOMIC is appropriate for the radix tree itself.
355 */
356 INIT_RADIX_TREE(&mru->store, GFP_ATOMIC);
357 INIT_LIST_HEAD(&mru->reap_list);
358 spin_lock_init(&mru->lock);
359 INIT_DELAYED_WORK(&mru->work, _xfs_mru_cache_reap);
360
361 mru->grp_time = grp_time;
362 mru->free_func = free_func;
363 mru->data = data;
364 *mrup = mru;
365
366exit:
367 if (err && mru && mru->lists)
368 kfree(mru->lists);
369 if (err && mru)
370 kfree(mru);
371
372 return err;
373}
374
375/*
376 * Call xfs_mru_cache_flush() to flush out all cached entries, calling their
377 * free functions as they're deleted. When this function returns, the caller is
378 * guaranteed that all the free functions for all the elements have finished
379 * executing and the reaper is not running.
380 */
381static void
382xfs_mru_cache_flush(
383 struct xfs_mru_cache *mru)
384{
385 if (!mru || !mru->lists)
386 return;
387
388 spin_lock(&mru->lock);
389 if (mru->queued) {
390 spin_unlock(&mru->lock);
391 cancel_delayed_work_sync(&mru->work);
392 spin_lock(&mru->lock);
393 }
394
395 _xfs_mru_cache_migrate(mru, jiffies + mru->grp_count * mru->grp_time);
396 _xfs_mru_cache_clear_reap_list(mru);
397
398 spin_unlock(&mru->lock);
399}
400
401void
402xfs_mru_cache_destroy(
403 struct xfs_mru_cache *mru)
404{
405 if (!mru || !mru->lists)
406 return;
407
408 xfs_mru_cache_flush(mru);
409
410 kfree(mru->lists);
411 kfree(mru);
412}
413
414/*
415 * To insert an element, call xfs_mru_cache_insert() with the data store, the
416 * element's key and the client data pointer. This function returns 0 on
417 * success or ENOMEM if memory for the data element couldn't be allocated.
418 */
419int
420xfs_mru_cache_insert(
421 struct xfs_mru_cache *mru,
422 unsigned long key,
423 struct xfs_mru_cache_elem *elem)
424{
425 int error;
426
427 ASSERT(mru && mru->lists);
428 if (!mru || !mru->lists)
429 return -EINVAL;
430
431 if (radix_tree_preload(GFP_KERNEL))
432 return -ENOMEM;
433
434 INIT_LIST_HEAD(&elem->list_node);
435 elem->key = key;
436
437 spin_lock(&mru->lock);
438 error = radix_tree_insert(&mru->store, key, elem);
439 radix_tree_preload_end();
440 if (!error)
441 _xfs_mru_cache_list_insert(mru, elem);
442 spin_unlock(&mru->lock);
443
444 return error;
445}
446
447/*
448 * To remove an element without calling the free function, call
449 * xfs_mru_cache_remove() with the data store and the element's key. On success
450 * the client data pointer for the removed element is returned, otherwise this
451 * function will return a NULL pointer.
452 */
453struct xfs_mru_cache_elem *
454xfs_mru_cache_remove(
455 struct xfs_mru_cache *mru,
456 unsigned long key)
457{
458 struct xfs_mru_cache_elem *elem;
459
460 ASSERT(mru && mru->lists);
461 if (!mru || !mru->lists)
462 return NULL;
463
464 spin_lock(&mru->lock);
465 elem = radix_tree_delete(&mru->store, key);
466 if (elem)
467 list_del(&elem->list_node);
468 spin_unlock(&mru->lock);
469
470 return elem;
471}
472
473/*
474 * To remove and element and call the free function, call xfs_mru_cache_delete()
475 * with the data store and the element's key.
476 */
477void
478xfs_mru_cache_delete(
479 struct xfs_mru_cache *mru,
480 unsigned long key)
481{
482 struct xfs_mru_cache_elem *elem;
483
484 elem = xfs_mru_cache_remove(mru, key);
485 if (elem)
486 mru->free_func(mru->data, elem);
487}
488
489/*
490 * To look up an element using its key, call xfs_mru_cache_lookup() with the
491 * data store and the element's key. If found, the element will be moved to the
492 * head of the MRU list to indicate that it's been touched.
493 *
494 * The internal data structures are protected by a spinlock that is STILL HELD
495 * when this function returns. Call xfs_mru_cache_done() to release it. Note
496 * that it is not safe to call any function that might sleep in the interim.
497 *
498 * The implementation could have used reference counting to avoid this
499 * restriction, but since most clients simply want to get, set or test a member
500 * of the returned data structure, the extra per-element memory isn't warranted.
501 *
502 * If the element isn't found, this function returns NULL and the spinlock is
503 * released. xfs_mru_cache_done() should NOT be called when this occurs.
504 *
505 * Because sparse isn't smart enough to know about conditional lock return
506 * status, we need to help it get it right by annotating the path that does
507 * not release the lock.
508 */
509struct xfs_mru_cache_elem *
510xfs_mru_cache_lookup(
511 struct xfs_mru_cache *mru,
512 unsigned long key)
513{
514 struct xfs_mru_cache_elem *elem;
515
516 ASSERT(mru && mru->lists);
517 if (!mru || !mru->lists)
518 return NULL;
519
520 spin_lock(&mru->lock);
521 elem = radix_tree_lookup(&mru->store, key);
522 if (elem) {
523 list_del(&elem->list_node);
524 _xfs_mru_cache_list_insert(mru, elem);
525 __release(mru_lock); /* help sparse not be stupid */
526 } else
527 spin_unlock(&mru->lock);
528
529 return elem;
530}
531
532/*
533 * To release the internal data structure spinlock after having performed an
534 * xfs_mru_cache_lookup() or an xfs_mru_cache_peek(), call xfs_mru_cache_done()
535 * with the data store pointer.
536 */
537void
538xfs_mru_cache_done(
539 struct xfs_mru_cache *mru)
540 __releases(mru->lock)
541{
542 spin_unlock(&mru->lock);
543}