xref: /linux/fs/ubifs/shrinker.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /*
2  * This file is part of UBIFS.
3  *
4  * Copyright (C) 2006-2008 Nokia Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License version 2 as published by
8  * the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13  * more details.
14  *
15  * You should have received a copy of the GNU General Public License along with
16  * this program; if not, write to the Free Software Foundation, Inc., 51
17  * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18  *
19  * Authors: Artem Bityutskiy (Битюцкий Артём)
20  *          Adrian Hunter
21  */
22 
23 /*
24  * This file implements UBIFS shrinker which evicts clean znodes from the TNC
25  * tree when Linux VM needs more RAM.
26  *
27  * We do not implement any LRU lists to find oldest znodes to free because it
28  * would add additional overhead to the file system fast paths. So the shrinker
29  * just walks the TNC tree when searching for znodes to free.
30  *
31  * If the root of a TNC sub-tree is clean and old enough, then the children are
32  * also clean and old enough. So the shrinker walks the TNC in level order and
33  * dumps entire sub-trees.
34  *
35  * The age of znodes is just the time-stamp when they were last looked at.
36  * The current shrinker first tries to evict old znodes, then young ones.
37  *
38  * Since the shrinker is global, it has to protect against races with FS
39  * un-mounts, which is done by the 'ubifs_infos_lock' and 'c->umount_mutex'.
40  */
41 
42 #include "ubifs.h"
43 
44 /* List of all UBIFS file-system instances */
45 LIST_HEAD(ubifs_infos);
46 
47 /*
48  * We number each shrinker run and record the number on the ubifs_info structure
49  * so that we can easily work out which ubifs_info structures have already been
50  * done by the current run.
51  */
52 static unsigned int shrinker_run_no;
53 
54 /* Protects 'ubifs_infos' list */
55 DEFINE_SPINLOCK(ubifs_infos_lock);
56 
57 /* Global clean znode counter (for all mounted UBIFS instances) */
58 atomic_long_t ubifs_clean_zn_cnt;
59 
60 /**
61  * shrink_tnc - shrink TNC tree.
62  * @c: UBIFS file-system description object
63  * @nr: number of znodes to free
64  * @age: the age of znodes to free
65  * @contention: if any contention, this is set to %1
66  *
67  * This function traverses TNC tree and frees clean znodes. It does not free
68  * clean znodes which younger then @age. Returns number of freed znodes.
69  */
70 static int shrink_tnc(struct ubifs_info *c, int nr, int age, int *contention)
71 {
72 	int total_freed = 0;
73 	struct ubifs_znode *znode, *zprev;
74 	int time = get_seconds();
75 
76 	ubifs_assert(mutex_is_locked(&c->umount_mutex));
77 	ubifs_assert(mutex_is_locked(&c->tnc_mutex));
78 
79 	if (!c->zroot.znode || atomic_long_read(&c->clean_zn_cnt) == 0)
80 		return 0;
81 
82 	/*
83 	 * Traverse the TNC tree in levelorder manner, so that it is possible
84 	 * to destroy large sub-trees. Indeed, if a znode is old, then all its
85 	 * children are older or of the same age.
86 	 *
87 	 * Note, we are holding 'c->tnc_mutex', so we do not have to lock the
88 	 * 'c->space_lock' when _reading_ 'c->clean_zn_cnt', because it is
89 	 * changed only when the 'c->tnc_mutex' is held.
90 	 */
91 	zprev = NULL;
92 	znode = ubifs_tnc_levelorder_next(c->zroot.znode, NULL);
93 	while (znode && total_freed < nr &&
94 	       atomic_long_read(&c->clean_zn_cnt) > 0) {
95 		int freed;
96 
97 		/*
98 		 * If the znode is clean, but it is in the 'c->cnext' list, this
99 		 * means that this znode has just been written to flash as a
100 		 * part of commit and was marked clean. They will be removed
101 		 * from the list at end commit. We cannot change the list,
102 		 * because it is not protected by any mutex (design decision to
103 		 * make commit really independent and parallel to main I/O). So
104 		 * we just skip these znodes.
105 		 *
106 		 * Note, the 'clean_zn_cnt' counters are not updated until
107 		 * after the commit, so the UBIFS shrinker does not report
108 		 * the znodes which are in the 'c->cnext' list as freeable.
109 		 *
110 		 * Also note, if the root of a sub-tree is not in 'c->cnext',
111 		 * then the whole sub-tree is not in 'c->cnext' as well, so it
112 		 * is safe to dump whole sub-tree.
113 		 */
114 
115 		if (znode->cnext) {
116 			/*
117 			 * Very soon these znodes will be removed from the list
118 			 * and become freeable.
119 			 */
120 			*contention = 1;
121 		} else if (!ubifs_zn_dirty(znode) &&
122 			   abs(time - znode->time) >= age) {
123 			if (znode->parent)
124 				znode->parent->zbranch[znode->iip].znode = NULL;
125 			else
126 				c->zroot.znode = NULL;
127 
128 			freed = ubifs_destroy_tnc_subtree(znode);
129 			atomic_long_sub(freed, &ubifs_clean_zn_cnt);
130 			atomic_long_sub(freed, &c->clean_zn_cnt);
131 			total_freed += freed;
132 			znode = zprev;
133 		}
134 
135 		if (unlikely(!c->zroot.znode))
136 			break;
137 
138 		zprev = znode;
139 		znode = ubifs_tnc_levelorder_next(c->zroot.znode, znode);
140 		cond_resched();
141 	}
142 
143 	return total_freed;
144 }
145 
146 /**
147  * shrink_tnc_trees - shrink UBIFS TNC trees.
148  * @nr: number of znodes to free
149  * @age: the age of znodes to free
150  * @contention: if any contention, this is set to %1
151  *
152  * This function walks the list of mounted UBIFS file-systems and frees clean
153  * znodes which are older than @age, until at least @nr znodes are freed.
154  * Returns the number of freed znodes.
155  */
156 static int shrink_tnc_trees(int nr, int age, int *contention)
157 {
158 	struct ubifs_info *c;
159 	struct list_head *p;
160 	unsigned int run_no;
161 	int freed = 0;
162 
163 	spin_lock(&ubifs_infos_lock);
164 	do {
165 		run_no = ++shrinker_run_no;
166 	} while (run_no == 0);
167 	/* Iterate over all mounted UBIFS file-systems and try to shrink them */
168 	p = ubifs_infos.next;
169 	while (p != &ubifs_infos) {
170 		c = list_entry(p, struct ubifs_info, infos_list);
171 		/*
172 		 * We move the ones we do to the end of the list, so we stop
173 		 * when we see one we have already done.
174 		 */
175 		if (c->shrinker_run_no == run_no)
176 			break;
177 		if (!mutex_trylock(&c->umount_mutex)) {
178 			/* Some un-mount is in progress, try next FS */
179 			*contention = 1;
180 			p = p->next;
181 			continue;
182 		}
183 		/*
184 		 * We're holding 'c->umount_mutex', so the file-system won't go
185 		 * away.
186 		 */
187 		if (!mutex_trylock(&c->tnc_mutex)) {
188 			mutex_unlock(&c->umount_mutex);
189 			*contention = 1;
190 			p = p->next;
191 			continue;
192 		}
193 		spin_unlock(&ubifs_infos_lock);
194 		/*
195 		 * OK, now we have TNC locked, the file-system cannot go away -
196 		 * it is safe to reap the cache.
197 		 */
198 		c->shrinker_run_no = run_no;
199 		freed += shrink_tnc(c, nr, age, contention);
200 		mutex_unlock(&c->tnc_mutex);
201 		spin_lock(&ubifs_infos_lock);
202 		/* Get the next list element before we move this one */
203 		p = p->next;
204 		/*
205 		 * Move this one to the end of the list to provide some
206 		 * fairness.
207 		 */
208 		list_move_tail(&c->infos_list, &ubifs_infos);
209 		mutex_unlock(&c->umount_mutex);
210 		if (freed >= nr)
211 			break;
212 	}
213 	spin_unlock(&ubifs_infos_lock);
214 	return freed;
215 }
216 
217 /**
218  * kick_a_thread - kick a background thread to start commit.
219  *
220  * This function kicks a background thread to start background commit. Returns
221  * %-1 if a thread was kicked or there is another reason to assume the memory
222  * will soon be freed or become freeable. If there are no dirty znodes, returns
223  * %0.
224  */
225 static int kick_a_thread(void)
226 {
227 	int i;
228 	struct ubifs_info *c;
229 
230 	/*
231 	 * Iterate over all mounted UBIFS file-systems and find out if there is
232 	 * already an ongoing commit operation there. If no, then iterate for
233 	 * the second time and initiate background commit.
234 	 */
235 	spin_lock(&ubifs_infos_lock);
236 	for (i = 0; i < 2; i++) {
237 		list_for_each_entry(c, &ubifs_infos, infos_list) {
238 			long dirty_zn_cnt;
239 
240 			if (!mutex_trylock(&c->umount_mutex)) {
241 				/*
242 				 * Some un-mount is in progress, it will
243 				 * certainly free memory, so just return.
244 				 */
245 				spin_unlock(&ubifs_infos_lock);
246 				return -1;
247 			}
248 
249 			dirty_zn_cnt = atomic_long_read(&c->dirty_zn_cnt);
250 
251 			if (!dirty_zn_cnt || c->cmt_state == COMMIT_BROKEN ||
252 			    c->ro_mount || c->ro_error) {
253 				mutex_unlock(&c->umount_mutex);
254 				continue;
255 			}
256 
257 			if (c->cmt_state != COMMIT_RESTING) {
258 				spin_unlock(&ubifs_infos_lock);
259 				mutex_unlock(&c->umount_mutex);
260 				return -1;
261 			}
262 
263 			if (i == 1) {
264 				list_move_tail(&c->infos_list, &ubifs_infos);
265 				spin_unlock(&ubifs_infos_lock);
266 
267 				ubifs_request_bg_commit(c);
268 				mutex_unlock(&c->umount_mutex);
269 				return -1;
270 			}
271 			mutex_unlock(&c->umount_mutex);
272 		}
273 	}
274 	spin_unlock(&ubifs_infos_lock);
275 
276 	return 0;
277 }
278 
279 unsigned long ubifs_shrink_count(struct shrinker *shrink,
280 				 struct shrink_control *sc)
281 {
282 	long clean_zn_cnt = atomic_long_read(&ubifs_clean_zn_cnt);
283 
284 	/*
285 	 * Due to the way UBIFS updates the clean znode counter it may
286 	 * temporarily be negative.
287 	 */
288 	return clean_zn_cnt >= 0 ? clean_zn_cnt : 1;
289 }
290 
291 unsigned long ubifs_shrink_scan(struct shrinker *shrink,
292 				struct shrink_control *sc)
293 {
294 	unsigned long nr = sc->nr_to_scan;
295 	int contention = 0;
296 	unsigned long freed;
297 	long clean_zn_cnt = atomic_long_read(&ubifs_clean_zn_cnt);
298 
299 	if (!clean_zn_cnt) {
300 		/*
301 		 * No clean znodes, nothing to reap. All we can do in this case
302 		 * is to kick background threads to start commit, which will
303 		 * probably make clean znodes which, in turn, will be freeable.
304 		 * And we return -1 which means will make VM call us again
305 		 * later.
306 		 */
307 		dbg_tnc("no clean znodes, kick a thread");
308 		return kick_a_thread();
309 	}
310 
311 	freed = shrink_tnc_trees(nr, OLD_ZNODE_AGE, &contention);
312 	if (freed >= nr)
313 		goto out;
314 
315 	dbg_tnc("not enough old znodes, try to free young ones");
316 	freed += shrink_tnc_trees(nr - freed, YOUNG_ZNODE_AGE, &contention);
317 	if (freed >= nr)
318 		goto out;
319 
320 	dbg_tnc("not enough young znodes, free all");
321 	freed += shrink_tnc_trees(nr - freed, 0, &contention);
322 
323 	if (!freed && contention) {
324 		dbg_tnc("freed nothing, but contention");
325 		return SHRINK_STOP;
326 	}
327 
328 out:
329 	dbg_tnc("%lu znodes were freed, requested %lu", freed, nr);
330 	return freed;
331 }
332