xref: /openbmc/linux/fs/ubifs/gc.c (revision 35267cea)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * This file is part of UBIFS.
4  *
5  * Copyright (C) 2006-2008 Nokia Corporation.
6  *
7  * Authors: Adrian Hunter
8  *          Artem Bityutskiy (Битюцкий Артём)
9  */
10 
11 /*
12  * This file implements garbage collection. The procedure for garbage collection
13  * is different depending on whether a LEB as an index LEB (contains index
14  * nodes) or not. For non-index LEBs, garbage collection finds a LEB which
15  * contains a lot of dirty space (obsolete nodes), and copies the non-obsolete
16  * nodes to the journal, at which point the garbage-collected LEB is free to be
17  * reused. For index LEBs, garbage collection marks the non-obsolete index nodes
18  * dirty in the TNC, and after the next commit, the garbage-collected LEB is
19  * to be reused. Garbage collection will cause the number of dirty index nodes
20  * to grow, however sufficient space is reserved for the index to ensure the
21  * commit will never run out of space.
22  *
23  * Notes about dead watermark. At current UBIFS implementation we assume that
24  * LEBs which have less than @c->dead_wm bytes of free + dirty space are full
25  * and not worth garbage-collecting. The dead watermark is one min. I/O unit
26  * size, or min. UBIFS node size, depending on what is greater. Indeed, UBIFS
27  * Garbage Collector has to synchronize the GC head's write buffer before
28  * returning, so this is about wasting one min. I/O unit. However, UBIFS GC can
29  * actually reclaim even very small pieces of dirty space by garbage collecting
30  * enough dirty LEBs, but we do not bother doing this at this implementation.
31  *
32  * Notes about dark watermark. The results of GC work depends on how big are
33  * the UBIFS nodes GC deals with. Large nodes make GC waste more space. Indeed,
34  * if GC move data from LEB A to LEB B and nodes in LEB A are large, GC would
35  * have to waste large pieces of free space at the end of LEB B, because nodes
36  * from LEB A would not fit. And the worst situation is when all nodes are of
37  * maximum size. So dark watermark is the amount of free + dirty space in LEB
38  * which are guaranteed to be reclaimable. If LEB has less space, the GC might
39  * be unable to reclaim it. So, LEBs with free + dirty greater than dark
40  * watermark are "good" LEBs from GC's point of view. The other LEBs are not so
41  * good, and GC takes extra care when moving them.
42  */
43 
44 #include <linux/slab.h>
45 #include <linux/pagemap.h>
46 #include <linux/list_sort.h>
47 #include "ubifs.h"
48 
49 /*
50  * GC may need to move more than one LEB to make progress. The below constants
51  * define "soft" and "hard" limits on the number of LEBs the garbage collector
52  * may move.
53  */
54 #define SOFT_LEBS_LIMIT 4
55 #define HARD_LEBS_LIMIT 32
56 
57 /**
58  * switch_gc_head - switch the garbage collection journal head.
59  * @c: UBIFS file-system description object
60  *
61  * This function switch the GC head to the next LEB which is reserved in
62  * @c->gc_lnum. Returns %0 in case of success, %-EAGAIN if commit is required,
63  * and other negative error code in case of failures.
64  */
65 static int switch_gc_head(struct ubifs_info *c)
66 {
67 	int err, gc_lnum = c->gc_lnum;
68 	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
69 
70 	ubifs_assert(c, gc_lnum != -1);
71 	dbg_gc("switch GC head from LEB %d:%d to LEB %d (waste %d bytes)",
72 	       wbuf->lnum, wbuf->offs + wbuf->used, gc_lnum,
73 	       c->leb_size - wbuf->offs - wbuf->used);
74 
75 	err = ubifs_wbuf_sync_nolock(wbuf);
76 	if (err)
77 		return err;
78 
79 	/*
80 	 * The GC write-buffer was synchronized, we may safely unmap
81 	 * 'c->gc_lnum'.
82 	 */
83 	err = ubifs_leb_unmap(c, gc_lnum);
84 	if (err)
85 		return err;
86 
87 	err = ubifs_add_bud_to_log(c, GCHD, gc_lnum, 0);
88 	if (err)
89 		return err;
90 
91 	c->gc_lnum = -1;
92 	err = ubifs_wbuf_seek_nolock(wbuf, gc_lnum, 0);
93 	return err;
94 }
95 
96 /**
97  * data_nodes_cmp - compare 2 data nodes.
98  * @priv: UBIFS file-system description object
99  * @a: first data node
100  * @b: second data node
101  *
102  * This function compares data nodes @a and @b. Returns %1 if @a has greater
103  * inode or block number, and %-1 otherwise.
104  */
105 static int data_nodes_cmp(void *priv, const struct list_head *a,
106 			  const struct list_head *b)
107 {
108 	ino_t inuma, inumb;
109 	struct ubifs_info *c = priv;
110 	struct ubifs_scan_node *sa, *sb;
111 
112 	cond_resched();
113 	if (a == b)
114 		return 0;
115 
116 	sa = list_entry(a, struct ubifs_scan_node, list);
117 	sb = list_entry(b, struct ubifs_scan_node, list);
118 
119 	ubifs_assert(c, key_type(c, &sa->key) == UBIFS_DATA_KEY);
120 	ubifs_assert(c, key_type(c, &sb->key) == UBIFS_DATA_KEY);
121 	ubifs_assert(c, sa->type == UBIFS_DATA_NODE);
122 	ubifs_assert(c, sb->type == UBIFS_DATA_NODE);
123 
124 	inuma = key_inum(c, &sa->key);
125 	inumb = key_inum(c, &sb->key);
126 
127 	if (inuma == inumb) {
128 		unsigned int blka = key_block(c, &sa->key);
129 		unsigned int blkb = key_block(c, &sb->key);
130 
131 		if (blka <= blkb)
132 			return -1;
133 	} else if (inuma <= inumb)
134 		return -1;
135 
136 	return 1;
137 }
138 
139 /*
140  * nondata_nodes_cmp - compare 2 non-data nodes.
141  * @priv: UBIFS file-system description object
142  * @a: first node
143  * @a: second node
144  *
145  * This function compares nodes @a and @b. It makes sure that inode nodes go
146  * first and sorted by length in descending order. Directory entry nodes go
147  * after inode nodes and are sorted in ascending hash valuer order.
148  */
149 static int nondata_nodes_cmp(void *priv, const struct list_head *a,
150 			     const struct list_head *b)
151 {
152 	ino_t inuma, inumb;
153 	struct ubifs_info *c = priv;
154 	struct ubifs_scan_node *sa, *sb;
155 
156 	cond_resched();
157 	if (a == b)
158 		return 0;
159 
160 	sa = list_entry(a, struct ubifs_scan_node, list);
161 	sb = list_entry(b, struct ubifs_scan_node, list);
162 
163 	ubifs_assert(c, key_type(c, &sa->key) != UBIFS_DATA_KEY &&
164 		     key_type(c, &sb->key) != UBIFS_DATA_KEY);
165 	ubifs_assert(c, sa->type != UBIFS_DATA_NODE &&
166 		     sb->type != UBIFS_DATA_NODE);
167 
168 	/* Inodes go before directory entries */
169 	if (sa->type == UBIFS_INO_NODE) {
170 		if (sb->type == UBIFS_INO_NODE)
171 			return sb->len - sa->len;
172 		return -1;
173 	}
174 	if (sb->type == UBIFS_INO_NODE)
175 		return 1;
176 
177 	ubifs_assert(c, key_type(c, &sa->key) == UBIFS_DENT_KEY ||
178 		     key_type(c, &sa->key) == UBIFS_XENT_KEY);
179 	ubifs_assert(c, key_type(c, &sb->key) == UBIFS_DENT_KEY ||
180 		     key_type(c, &sb->key) == UBIFS_XENT_KEY);
181 	ubifs_assert(c, sa->type == UBIFS_DENT_NODE ||
182 		     sa->type == UBIFS_XENT_NODE);
183 	ubifs_assert(c, sb->type == UBIFS_DENT_NODE ||
184 		     sb->type == UBIFS_XENT_NODE);
185 
186 	inuma = key_inum(c, &sa->key);
187 	inumb = key_inum(c, &sb->key);
188 
189 	if (inuma == inumb) {
190 		uint32_t hasha = key_hash(c, &sa->key);
191 		uint32_t hashb = key_hash(c, &sb->key);
192 
193 		if (hasha <= hashb)
194 			return -1;
195 	} else if (inuma <= inumb)
196 		return -1;
197 
198 	return 1;
199 }
200 
201 /**
202  * sort_nodes - sort nodes for GC.
203  * @c: UBIFS file-system description object
204  * @sleb: describes nodes to sort and contains the result on exit
205  * @nondata: contains non-data nodes on exit
206  * @min: minimum node size is returned here
207  *
208  * This function sorts the list of inodes to garbage collect. First of all, it
209  * kills obsolete nodes and separates data and non-data nodes to the
210  * @sleb->nodes and @nondata lists correspondingly.
211  *
212  * Data nodes are then sorted in block number order - this is important for
213  * bulk-read; data nodes with lower inode number go before data nodes with
214  * higher inode number, and data nodes with lower block number go before data
215  * nodes with higher block number;
216  *
217  * Non-data nodes are sorted as follows.
218  *   o First go inode nodes - they are sorted in descending length order.
219  *   o Then go directory entry nodes - they are sorted in hash order, which
220  *     should supposedly optimize 'readdir()'. Direntry nodes with lower parent
221  *     inode number go before direntry nodes with higher parent inode number,
222  *     and direntry nodes with lower name hash values go before direntry nodes
223  *     with higher name hash values.
224  *
225  * This function returns zero in case of success and a negative error code in
226  * case of failure.
227  */
228 static int sort_nodes(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
229 		      struct list_head *nondata, int *min)
230 {
231 	int err;
232 	struct ubifs_scan_node *snod, *tmp;
233 
234 	*min = INT_MAX;
235 
236 	/* Separate data nodes and non-data nodes */
237 	list_for_each_entry_safe(snod, tmp, &sleb->nodes, list) {
238 		ubifs_assert(c, snod->type == UBIFS_INO_NODE  ||
239 			     snod->type == UBIFS_DATA_NODE ||
240 			     snod->type == UBIFS_DENT_NODE ||
241 			     snod->type == UBIFS_XENT_NODE ||
242 			     snod->type == UBIFS_TRUN_NODE ||
243 			     snod->type == UBIFS_AUTH_NODE);
244 
245 		if (snod->type != UBIFS_INO_NODE  &&
246 		    snod->type != UBIFS_DATA_NODE &&
247 		    snod->type != UBIFS_DENT_NODE &&
248 		    snod->type != UBIFS_XENT_NODE) {
249 			/* Probably truncation node, zap it */
250 			list_del(&snod->list);
251 			kfree(snod);
252 			continue;
253 		}
254 
255 		ubifs_assert(c, key_type(c, &snod->key) == UBIFS_DATA_KEY ||
256 			     key_type(c, &snod->key) == UBIFS_INO_KEY  ||
257 			     key_type(c, &snod->key) == UBIFS_DENT_KEY ||
258 			     key_type(c, &snod->key) == UBIFS_XENT_KEY);
259 
260 		err = ubifs_tnc_has_node(c, &snod->key, 0, sleb->lnum,
261 					 snod->offs, 0);
262 		if (err < 0)
263 			return err;
264 
265 		if (!err) {
266 			/* The node is obsolete, remove it from the list */
267 			list_del(&snod->list);
268 			kfree(snod);
269 			continue;
270 		}
271 
272 		if (snod->len < *min)
273 			*min = snod->len;
274 
275 		if (key_type(c, &snod->key) != UBIFS_DATA_KEY)
276 			list_move_tail(&snod->list, nondata);
277 	}
278 
279 	/* Sort data and non-data nodes */
280 	list_sort(c, &sleb->nodes, &data_nodes_cmp);
281 	list_sort(c, nondata, &nondata_nodes_cmp);
282 
283 	err = dbg_check_data_nodes_order(c, &sleb->nodes);
284 	if (err)
285 		return err;
286 	err = dbg_check_nondata_nodes_order(c, nondata);
287 	if (err)
288 		return err;
289 	return 0;
290 }
291 
292 /**
293  * move_node - move a node.
294  * @c: UBIFS file-system description object
295  * @sleb: describes the LEB to move nodes from
296  * @snod: the mode to move
297  * @wbuf: write-buffer to move node to
298  *
299  * This function moves node @snod to @wbuf, changes TNC correspondingly, and
300  * destroys @snod. Returns zero in case of success and a negative error code in
301  * case of failure.
302  */
303 static int move_node(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
304 		     struct ubifs_scan_node *snod, struct ubifs_wbuf *wbuf)
305 {
306 	int err, new_lnum = wbuf->lnum, new_offs = wbuf->offs + wbuf->used;
307 
308 	cond_resched();
309 	err = ubifs_wbuf_write_nolock(wbuf, snod->node, snod->len);
310 	if (err)
311 		return err;
312 
313 	err = ubifs_tnc_replace(c, &snod->key, sleb->lnum,
314 				snod->offs, new_lnum, new_offs,
315 				snod->len);
316 	list_del(&snod->list);
317 	kfree(snod);
318 	return err;
319 }
320 
321 /**
322  * move_nodes - move nodes.
323  * @c: UBIFS file-system description object
324  * @sleb: describes the LEB to move nodes from
325  *
326  * This function moves valid nodes from data LEB described by @sleb to the GC
327  * journal head. This function returns zero in case of success, %-EAGAIN if
328  * commit is required, and other negative error codes in case of other
329  * failures.
330  */
331 static int move_nodes(struct ubifs_info *c, struct ubifs_scan_leb *sleb)
332 {
333 	int err, min;
334 	LIST_HEAD(nondata);
335 	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
336 
337 	if (wbuf->lnum == -1) {
338 		/*
339 		 * The GC journal head is not set, because it is the first GC
340 		 * invocation since mount.
341 		 */
342 		err = switch_gc_head(c);
343 		if (err)
344 			return err;
345 	}
346 
347 	err = sort_nodes(c, sleb, &nondata, &min);
348 	if (err)
349 		goto out;
350 
351 	/* Write nodes to their new location. Use the first-fit strategy */
352 	while (1) {
353 		int avail, moved = 0;
354 		struct ubifs_scan_node *snod, *tmp;
355 
356 		/* Move data nodes */
357 		list_for_each_entry_safe(snod, tmp, &sleb->nodes, list) {
358 			avail = c->leb_size - wbuf->offs - wbuf->used -
359 					ubifs_auth_node_sz(c);
360 			if  (snod->len > avail)
361 				/*
362 				 * Do not skip data nodes in order to optimize
363 				 * bulk-read.
364 				 */
365 				break;
366 
367 			err = ubifs_shash_update(c, c->jheads[GCHD].log_hash,
368 						 snod->node, snod->len);
369 			if (err)
370 				goto out;
371 
372 			err = move_node(c, sleb, snod, wbuf);
373 			if (err)
374 				goto out;
375 			moved = 1;
376 		}
377 
378 		/* Move non-data nodes */
379 		list_for_each_entry_safe(snod, tmp, &nondata, list) {
380 			avail = c->leb_size - wbuf->offs - wbuf->used -
381 					ubifs_auth_node_sz(c);
382 			if (avail < min)
383 				break;
384 
385 			if  (snod->len > avail) {
386 				/*
387 				 * Keep going only if this is an inode with
388 				 * some data. Otherwise stop and switch the GC
389 				 * head. IOW, we assume that data-less inode
390 				 * nodes and direntry nodes are roughly of the
391 				 * same size.
392 				 */
393 				if (key_type(c, &snod->key) == UBIFS_DENT_KEY ||
394 				    snod->len == UBIFS_INO_NODE_SZ)
395 					break;
396 				continue;
397 			}
398 
399 			err = ubifs_shash_update(c, c->jheads[GCHD].log_hash,
400 						 snod->node, snod->len);
401 			if (err)
402 				goto out;
403 
404 			err = move_node(c, sleb, snod, wbuf);
405 			if (err)
406 				goto out;
407 			moved = 1;
408 		}
409 
410 		if (ubifs_authenticated(c) && moved) {
411 			struct ubifs_auth_node *auth;
412 
413 			auth = kmalloc(ubifs_auth_node_sz(c), GFP_NOFS);
414 			if (!auth) {
415 				err = -ENOMEM;
416 				goto out;
417 			}
418 
419 			err = ubifs_prepare_auth_node(c, auth,
420 						c->jheads[GCHD].log_hash);
421 			if (err) {
422 				kfree(auth);
423 				goto out;
424 			}
425 
426 			err = ubifs_wbuf_write_nolock(wbuf, auth,
427 						      ubifs_auth_node_sz(c));
428 			if (err) {
429 				kfree(auth);
430 				goto out;
431 			}
432 
433 			ubifs_add_dirt(c, wbuf->lnum, ubifs_auth_node_sz(c));
434 		}
435 
436 		if (list_empty(&sleb->nodes) && list_empty(&nondata))
437 			break;
438 
439 		/*
440 		 * Waste the rest of the space in the LEB and switch to the
441 		 * next LEB.
442 		 */
443 		err = switch_gc_head(c);
444 		if (err)
445 			goto out;
446 	}
447 
448 	return 0;
449 
450 out:
451 	list_splice_tail(&nondata, &sleb->nodes);
452 	return err;
453 }
454 
455 /**
456  * gc_sync_wbufs - sync write-buffers for GC.
457  * @c: UBIFS file-system description object
458  *
459  * We must guarantee that obsoleting nodes are on flash. Unfortunately they may
460  * be in a write-buffer instead. That is, a node could be written to a
461  * write-buffer, obsoleting another node in a LEB that is GC'd. If that LEB is
462  * erased before the write-buffer is sync'd and then there is an unclean
463  * unmount, then an existing node is lost. To avoid this, we sync all
464  * write-buffers.
465  *
466  * This function returns %0 on success or a negative error code on failure.
467  */
468 static int gc_sync_wbufs(struct ubifs_info *c)
469 {
470 	int err, i;
471 
472 	for (i = 0; i < c->jhead_cnt; i++) {
473 		if (i == GCHD)
474 			continue;
475 		err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
476 		if (err)
477 			return err;
478 	}
479 	return 0;
480 }
481 
482 /**
483  * ubifs_garbage_collect_leb - garbage-collect a logical eraseblock.
484  * @c: UBIFS file-system description object
485  * @lp: describes the LEB to garbage collect
486  *
487  * This function garbage-collects an LEB and returns one of the @LEB_FREED,
488  * @LEB_RETAINED, etc positive codes in case of success, %-EAGAIN if commit is
489  * required, and other negative error codes in case of failures.
490  */
491 int ubifs_garbage_collect_leb(struct ubifs_info *c, struct ubifs_lprops *lp)
492 {
493 	struct ubifs_scan_leb *sleb;
494 	struct ubifs_scan_node *snod;
495 	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
496 	int err = 0, lnum = lp->lnum;
497 
498 	ubifs_assert(c, c->gc_lnum != -1 || wbuf->offs + wbuf->used == 0 ||
499 		     c->need_recovery);
500 	ubifs_assert(c, c->gc_lnum != lnum);
501 	ubifs_assert(c, wbuf->lnum != lnum);
502 
503 	if (lp->free + lp->dirty == c->leb_size) {
504 		/* Special case - a free LEB  */
505 		dbg_gc("LEB %d is free, return it", lp->lnum);
506 		ubifs_assert(c, !(lp->flags & LPROPS_INDEX));
507 
508 		if (lp->free != c->leb_size) {
509 			/*
510 			 * Write buffers must be sync'd before unmapping
511 			 * freeable LEBs, because one of them may contain data
512 			 * which obsoletes something in 'lp->lnum'.
513 			 */
514 			err = gc_sync_wbufs(c);
515 			if (err)
516 				return err;
517 			err = ubifs_change_one_lp(c, lp->lnum, c->leb_size,
518 						  0, 0, 0, 0);
519 			if (err)
520 				return err;
521 		}
522 		err = ubifs_leb_unmap(c, lp->lnum);
523 		if (err)
524 			return err;
525 
526 		if (c->gc_lnum == -1) {
527 			c->gc_lnum = lnum;
528 			return LEB_RETAINED;
529 		}
530 
531 		return LEB_FREED;
532 	}
533 
534 	/*
535 	 * We scan the entire LEB even though we only really need to scan up to
536 	 * (c->leb_size - lp->free).
537 	 */
538 	sleb = ubifs_scan(c, lnum, 0, c->sbuf, 0);
539 	if (IS_ERR(sleb))
540 		return PTR_ERR(sleb);
541 
542 	ubifs_assert(c, !list_empty(&sleb->nodes));
543 	snod = list_entry(sleb->nodes.next, struct ubifs_scan_node, list);
544 
545 	if (snod->type == UBIFS_IDX_NODE) {
546 		struct ubifs_gced_idx_leb *idx_gc;
547 
548 		dbg_gc("indexing LEB %d (free %d, dirty %d)",
549 		       lnum, lp->free, lp->dirty);
550 		list_for_each_entry(snod, &sleb->nodes, list) {
551 			struct ubifs_idx_node *idx = snod->node;
552 			int level = le16_to_cpu(idx->level);
553 
554 			ubifs_assert(c, snod->type == UBIFS_IDX_NODE);
555 			key_read(c, ubifs_idx_key(c, idx), &snod->key);
556 			err = ubifs_dirty_idx_node(c, &snod->key, level, lnum,
557 						   snod->offs);
558 			if (err)
559 				goto out;
560 		}
561 
562 		idx_gc = kmalloc(sizeof(struct ubifs_gced_idx_leb), GFP_NOFS);
563 		if (!idx_gc) {
564 			err = -ENOMEM;
565 			goto out;
566 		}
567 
568 		idx_gc->lnum = lnum;
569 		idx_gc->unmap = 0;
570 		list_add(&idx_gc->list, &c->idx_gc);
571 
572 		/*
573 		 * Don't release the LEB until after the next commit, because
574 		 * it may contain data which is needed for recovery. So
575 		 * although we freed this LEB, it will become usable only after
576 		 * the commit.
577 		 */
578 		err = ubifs_change_one_lp(c, lnum, c->leb_size, 0, 0,
579 					  LPROPS_INDEX, 1);
580 		if (err)
581 			goto out;
582 		err = LEB_FREED_IDX;
583 	} else {
584 		dbg_gc("data LEB %d (free %d, dirty %d)",
585 		       lnum, lp->free, lp->dirty);
586 
587 		err = move_nodes(c, sleb);
588 		if (err)
589 			goto out_inc_seq;
590 
591 		err = gc_sync_wbufs(c);
592 		if (err)
593 			goto out_inc_seq;
594 
595 		err = ubifs_change_one_lp(c, lnum, c->leb_size, 0, 0, 0, 0);
596 		if (err)
597 			goto out_inc_seq;
598 
599 		/* Allow for races with TNC */
600 		c->gced_lnum = lnum;
601 		smp_wmb();
602 		c->gc_seq += 1;
603 		smp_wmb();
604 
605 		if (c->gc_lnum == -1) {
606 			c->gc_lnum = lnum;
607 			err = LEB_RETAINED;
608 		} else {
609 			err = ubifs_wbuf_sync_nolock(wbuf);
610 			if (err)
611 				goto out;
612 
613 			err = ubifs_leb_unmap(c, lnum);
614 			if (err)
615 				goto out;
616 
617 			err = LEB_FREED;
618 		}
619 	}
620 
621 out:
622 	ubifs_scan_destroy(sleb);
623 	return err;
624 
625 out_inc_seq:
626 	/* We may have moved at least some nodes so allow for races with TNC */
627 	c->gced_lnum = lnum;
628 	smp_wmb();
629 	c->gc_seq += 1;
630 	smp_wmb();
631 	goto out;
632 }
633 
634 /**
635  * ubifs_garbage_collect - UBIFS garbage collector.
636  * @c: UBIFS file-system description object
637  * @anyway: do GC even if there are free LEBs
638  *
639  * This function does out-of-place garbage collection. The return codes are:
640  *   o positive LEB number if the LEB has been freed and may be used;
641  *   o %-EAGAIN if the caller has to run commit;
642  *   o %-ENOSPC if GC failed to make any progress;
643  *   o other negative error codes in case of other errors.
644  *
645  * Garbage collector writes data to the journal when GC'ing data LEBs, and just
646  * marking indexing nodes dirty when GC'ing indexing LEBs. Thus, at some point
647  * commit may be required. But commit cannot be run from inside GC, because the
648  * caller might be holding the commit lock, so %-EAGAIN is returned instead;
649  * And this error code means that the caller has to run commit, and re-run GC
650  * if there is still no free space.
651  *
652  * There are many reasons why this function may return %-EAGAIN:
653  * o the log is full and there is no space to write an LEB reference for
654  *   @c->gc_lnum;
655  * o the journal is too large and exceeds size limitations;
656  * o GC moved indexing LEBs, but they can be used only after the commit;
657  * o the shrinker fails to find clean znodes to free and requests the commit;
658  * o etc.
659  *
660  * Note, if the file-system is close to be full, this function may return
661  * %-EAGAIN infinitely, so the caller has to limit amount of re-invocations of
662  * the function. E.g., this happens if the limits on the journal size are too
663  * tough and GC writes too much to the journal before an LEB is freed. This
664  * might also mean that the journal is too large, and the TNC becomes to big,
665  * so that the shrinker is constantly called, finds not clean znodes to free,
666  * and requests commit. Well, this may also happen if the journal is all right,
667  * but another kernel process consumes too much memory. Anyway, infinite
668  * %-EAGAIN may happen, but in some extreme/misconfiguration cases.
669  */
670 int ubifs_garbage_collect(struct ubifs_info *c, int anyway)
671 {
672 	int i, err, ret, min_space = c->dead_wm;
673 	struct ubifs_lprops lp;
674 	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
675 
676 	ubifs_assert_cmt_locked(c);
677 	ubifs_assert(c, !c->ro_media && !c->ro_mount);
678 
679 	if (ubifs_gc_should_commit(c))
680 		return -EAGAIN;
681 
682 	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
683 
684 	if (c->ro_error) {
685 		ret = -EROFS;
686 		goto out_unlock;
687 	}
688 
689 	/* We expect the write-buffer to be empty on entry */
690 	ubifs_assert(c, !wbuf->used);
691 
692 	for (i = 0; ; i++) {
693 		int space_before, space_after;
694 
695 		cond_resched();
696 
697 		/* Give the commit an opportunity to run */
698 		if (ubifs_gc_should_commit(c)) {
699 			ret = -EAGAIN;
700 			break;
701 		}
702 
703 		if (i > SOFT_LEBS_LIMIT && !list_empty(&c->idx_gc)) {
704 			/*
705 			 * We've done enough iterations. Indexing LEBs were
706 			 * moved and will be available after the commit.
707 			 */
708 			dbg_gc("soft limit, some index LEBs GC'ed, -EAGAIN");
709 			ubifs_commit_required(c);
710 			ret = -EAGAIN;
711 			break;
712 		}
713 
714 		if (i > HARD_LEBS_LIMIT) {
715 			/*
716 			 * We've moved too many LEBs and have not made
717 			 * progress, give up.
718 			 */
719 			dbg_gc("hard limit, -ENOSPC");
720 			ret = -ENOSPC;
721 			break;
722 		}
723 
724 		/*
725 		 * Empty and freeable LEBs can turn up while we waited for
726 		 * the wbuf lock, or while we have been running GC. In that
727 		 * case, we should just return one of those instead of
728 		 * continuing to GC dirty LEBs. Hence we request
729 		 * 'ubifs_find_dirty_leb()' to return an empty LEB if it can.
730 		 */
731 		ret = ubifs_find_dirty_leb(c, &lp, min_space, anyway ? 0 : 1);
732 		if (ret) {
733 			if (ret == -ENOSPC)
734 				dbg_gc("no more dirty LEBs");
735 			break;
736 		}
737 
738 		dbg_gc("found LEB %d: free %d, dirty %d, sum %d (min. space %d)",
739 		       lp.lnum, lp.free, lp.dirty, lp.free + lp.dirty,
740 		       min_space);
741 
742 		space_before = c->leb_size - wbuf->offs - wbuf->used;
743 		if (wbuf->lnum == -1)
744 			space_before = 0;
745 
746 		ret = ubifs_garbage_collect_leb(c, &lp);
747 		if (ret < 0) {
748 			if (ret == -EAGAIN) {
749 				/*
750 				 * This is not error, so we have to return the
751 				 * LEB to lprops. But if 'ubifs_return_leb()'
752 				 * fails, its failure code is propagated to the
753 				 * caller instead of the original '-EAGAIN'.
754 				 */
755 				err = ubifs_return_leb(c, lp.lnum);
756 				if (err)
757 					ret = err;
758 				break;
759 			}
760 			goto out;
761 		}
762 
763 		if (ret == LEB_FREED) {
764 			/* An LEB has been freed and is ready for use */
765 			dbg_gc("LEB %d freed, return", lp.lnum);
766 			ret = lp.lnum;
767 			break;
768 		}
769 
770 		if (ret == LEB_FREED_IDX) {
771 			/*
772 			 * This was an indexing LEB and it cannot be
773 			 * immediately used. And instead of requesting the
774 			 * commit straight away, we try to garbage collect some
775 			 * more.
776 			 */
777 			dbg_gc("indexing LEB %d freed, continue", lp.lnum);
778 			continue;
779 		}
780 
781 		ubifs_assert(c, ret == LEB_RETAINED);
782 		space_after = c->leb_size - wbuf->offs - wbuf->used;
783 		dbg_gc("LEB %d retained, freed %d bytes", lp.lnum,
784 		       space_after - space_before);
785 
786 		if (space_after > space_before) {
787 			/* GC makes progress, keep working */
788 			min_space >>= 1;
789 			if (min_space < c->dead_wm)
790 				min_space = c->dead_wm;
791 			continue;
792 		}
793 
794 		dbg_gc("did not make progress");
795 
796 		/*
797 		 * GC moved an LEB bud have not done any progress. This means
798 		 * that the previous GC head LEB contained too few free space
799 		 * and the LEB which was GC'ed contained only large nodes which
800 		 * did not fit that space.
801 		 *
802 		 * We can do 2 things:
803 		 * 1. pick another LEB in a hope it'll contain a small node
804 		 *    which will fit the space we have at the end of current GC
805 		 *    head LEB, but there is no guarantee, so we try this out
806 		 *    unless we have already been working for too long;
807 		 * 2. request an LEB with more dirty space, which will force
808 		 *    'ubifs_find_dirty_leb()' to start scanning the lprops
809 		 *    table, instead of just picking one from the heap
810 		 *    (previously it already picked the dirtiest LEB).
811 		 */
812 		if (i < SOFT_LEBS_LIMIT) {
813 			dbg_gc("try again");
814 			continue;
815 		}
816 
817 		min_space <<= 1;
818 		if (min_space > c->dark_wm)
819 			min_space = c->dark_wm;
820 		dbg_gc("set min. space to %d", min_space);
821 	}
822 
823 	if (ret == -ENOSPC && !list_empty(&c->idx_gc)) {
824 		dbg_gc("no space, some index LEBs GC'ed, -EAGAIN");
825 		ubifs_commit_required(c);
826 		ret = -EAGAIN;
827 	}
828 
829 	err = ubifs_wbuf_sync_nolock(wbuf);
830 	if (!err)
831 		err = ubifs_leb_unmap(c, c->gc_lnum);
832 	if (err) {
833 		ret = err;
834 		goto out;
835 	}
836 out_unlock:
837 	mutex_unlock(&wbuf->io_mutex);
838 	return ret;
839 
840 out:
841 	ubifs_assert(c, ret < 0);
842 	ubifs_assert(c, ret != -ENOSPC && ret != -EAGAIN);
843 	ubifs_wbuf_sync_nolock(wbuf);
844 	ubifs_ro_mode(c, ret);
845 	mutex_unlock(&wbuf->io_mutex);
846 	ubifs_return_leb(c, lp.lnum);
847 	return ret;
848 }
849 
850 /**
851  * ubifs_gc_start_commit - garbage collection at start of commit.
852  * @c: UBIFS file-system description object
853  *
854  * If a LEB has only dirty and free space, then we may safely unmap it and make
855  * it free.  Note, we cannot do this with indexing LEBs because dirty space may
856  * correspond index nodes that are required for recovery.  In that case, the
857  * LEB cannot be unmapped until after the next commit.
858  *
859  * This function returns %0 upon success and a negative error code upon failure.
860  */
861 int ubifs_gc_start_commit(struct ubifs_info *c)
862 {
863 	struct ubifs_gced_idx_leb *idx_gc;
864 	const struct ubifs_lprops *lp;
865 	int err = 0, flags;
866 
867 	ubifs_get_lprops(c);
868 
869 	/*
870 	 * Unmap (non-index) freeable LEBs. Note that recovery requires that all
871 	 * wbufs are sync'd before this, which is done in 'do_commit()'.
872 	 */
873 	while (1) {
874 		lp = ubifs_fast_find_freeable(c);
875 		if (!lp)
876 			break;
877 		ubifs_assert(c, !(lp->flags & LPROPS_TAKEN));
878 		ubifs_assert(c, !(lp->flags & LPROPS_INDEX));
879 		err = ubifs_leb_unmap(c, lp->lnum);
880 		if (err)
881 			goto out;
882 		lp = ubifs_change_lp(c, lp, c->leb_size, 0, lp->flags, 0);
883 		if (IS_ERR(lp)) {
884 			err = PTR_ERR(lp);
885 			goto out;
886 		}
887 		ubifs_assert(c, !(lp->flags & LPROPS_TAKEN));
888 		ubifs_assert(c, !(lp->flags & LPROPS_INDEX));
889 	}
890 
891 	/* Mark GC'd index LEBs OK to unmap after this commit finishes */
892 	list_for_each_entry(idx_gc, &c->idx_gc, list)
893 		idx_gc->unmap = 1;
894 
895 	/* Record index freeable LEBs for unmapping after commit */
896 	while (1) {
897 		lp = ubifs_fast_find_frdi_idx(c);
898 		if (IS_ERR(lp)) {
899 			err = PTR_ERR(lp);
900 			goto out;
901 		}
902 		if (!lp)
903 			break;
904 		idx_gc = kmalloc(sizeof(struct ubifs_gced_idx_leb), GFP_NOFS);
905 		if (!idx_gc) {
906 			err = -ENOMEM;
907 			goto out;
908 		}
909 		ubifs_assert(c, !(lp->flags & LPROPS_TAKEN));
910 		ubifs_assert(c, lp->flags & LPROPS_INDEX);
911 		/* Don't release the LEB until after the next commit */
912 		flags = (lp->flags | LPROPS_TAKEN) ^ LPROPS_INDEX;
913 		lp = ubifs_change_lp(c, lp, c->leb_size, 0, flags, 1);
914 		if (IS_ERR(lp)) {
915 			err = PTR_ERR(lp);
916 			kfree(idx_gc);
917 			goto out;
918 		}
919 		ubifs_assert(c, lp->flags & LPROPS_TAKEN);
920 		ubifs_assert(c, !(lp->flags & LPROPS_INDEX));
921 		idx_gc->lnum = lp->lnum;
922 		idx_gc->unmap = 1;
923 		list_add(&idx_gc->list, &c->idx_gc);
924 	}
925 out:
926 	ubifs_release_lprops(c);
927 	return err;
928 }
929 
930 /**
931  * ubifs_gc_end_commit - garbage collection at end of commit.
932  * @c: UBIFS file-system description object
933  *
934  * This function completes out-of-place garbage collection of index LEBs.
935  */
936 int ubifs_gc_end_commit(struct ubifs_info *c)
937 {
938 	struct ubifs_gced_idx_leb *idx_gc, *tmp;
939 	struct ubifs_wbuf *wbuf;
940 	int err = 0;
941 
942 	wbuf = &c->jheads[GCHD].wbuf;
943 	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
944 	list_for_each_entry_safe(idx_gc, tmp, &c->idx_gc, list)
945 		if (idx_gc->unmap) {
946 			dbg_gc("LEB %d", idx_gc->lnum);
947 			err = ubifs_leb_unmap(c, idx_gc->lnum);
948 			if (err)
949 				goto out;
950 			err = ubifs_change_one_lp(c, idx_gc->lnum, LPROPS_NC,
951 					  LPROPS_NC, 0, LPROPS_TAKEN, -1);
952 			if (err)
953 				goto out;
954 			list_del(&idx_gc->list);
955 			kfree(idx_gc);
956 		}
957 out:
958 	mutex_unlock(&wbuf->io_mutex);
959 	return err;
960 }
961 
962 /**
963  * ubifs_destroy_idx_gc - destroy idx_gc list.
964  * @c: UBIFS file-system description object
965  *
966  * This function destroys the @c->idx_gc list. It is called when unmounting
967  * so locks are not needed. Returns zero in case of success and a negative
968  * error code in case of failure.
969  */
970 void ubifs_destroy_idx_gc(struct ubifs_info *c)
971 {
972 	while (!list_empty(&c->idx_gc)) {
973 		struct ubifs_gced_idx_leb *idx_gc;
974 
975 		idx_gc = list_entry(c->idx_gc.next, struct ubifs_gced_idx_leb,
976 				    list);
977 		c->idx_gc_cnt -= 1;
978 		list_del(&idx_gc->list);
979 		kfree(idx_gc);
980 	}
981 }
982 
983 /**
984  * ubifs_get_idx_gc_leb - get a LEB from GC'd index LEB list.
985  * @c: UBIFS file-system description object
986  *
987  * Called during start commit so locks are not needed.
988  */
989 int ubifs_get_idx_gc_leb(struct ubifs_info *c)
990 {
991 	struct ubifs_gced_idx_leb *idx_gc;
992 	int lnum;
993 
994 	if (list_empty(&c->idx_gc))
995 		return -ENOSPC;
996 	idx_gc = list_entry(c->idx_gc.next, struct ubifs_gced_idx_leb, list);
997 	lnum = idx_gc->lnum;
998 	/* c->idx_gc_cnt is updated by the caller when lprops are updated */
999 	list_del(&idx_gc->list);
1000 	kfree(idx_gc);
1001 	return lnum;
1002 }
1003