xref: /openbmc/linux/fs/ubifs/tnc_commit.c (revision 545e4006)
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: Adrian Hunter
20  *          Artem Bityutskiy (Битюцкий Артём)
21  */
22 
23 /* This file implements TNC functions for committing */
24 
25 #include "ubifs.h"
26 
27 /**
28  * make_idx_node - make an index node for fill-the-gaps method of TNC commit.
29  * @c: UBIFS file-system description object
30  * @idx: buffer in which to place new index node
31  * @znode: znode from which to make new index node
32  * @lnum: LEB number where new index node will be written
33  * @offs: offset where new index node will be written
34  * @len: length of new index node
35  */
36 static int make_idx_node(struct ubifs_info *c, struct ubifs_idx_node *idx,
37 			 struct ubifs_znode *znode, int lnum, int offs, int len)
38 {
39 	struct ubifs_znode *zp;
40 	int i, err;
41 
42 	/* Make index node */
43 	idx->ch.node_type = UBIFS_IDX_NODE;
44 	idx->child_cnt = cpu_to_le16(znode->child_cnt);
45 	idx->level = cpu_to_le16(znode->level);
46 	for (i = 0; i < znode->child_cnt; i++) {
47 		struct ubifs_branch *br = ubifs_idx_branch(c, idx, i);
48 		struct ubifs_zbranch *zbr = &znode->zbranch[i];
49 
50 		key_write_idx(c, &zbr->key, &br->key);
51 		br->lnum = cpu_to_le32(zbr->lnum);
52 		br->offs = cpu_to_le32(zbr->offs);
53 		br->len = cpu_to_le32(zbr->len);
54 		if (!zbr->lnum || !zbr->len) {
55 			ubifs_err("bad ref in znode");
56 			dbg_dump_znode(c, znode);
57 			if (zbr->znode)
58 				dbg_dump_znode(c, zbr->znode);
59 		}
60 	}
61 	ubifs_prepare_node(c, idx, len, 0);
62 
63 #ifdef CONFIG_UBIFS_FS_DEBUG
64 	znode->lnum = lnum;
65 	znode->offs = offs;
66 	znode->len = len;
67 #endif
68 
69 	err = insert_old_idx_znode(c, znode);
70 
71 	/* Update the parent */
72 	zp = znode->parent;
73 	if (zp) {
74 		struct ubifs_zbranch *zbr;
75 
76 		zbr = &zp->zbranch[znode->iip];
77 		zbr->lnum = lnum;
78 		zbr->offs = offs;
79 		zbr->len = len;
80 	} else {
81 		c->zroot.lnum = lnum;
82 		c->zroot.offs = offs;
83 		c->zroot.len = len;
84 	}
85 	c->calc_idx_sz += ALIGN(len, 8);
86 
87 	atomic_long_dec(&c->dirty_zn_cnt);
88 
89 	ubifs_assert(ubifs_zn_dirty(znode));
90 	ubifs_assert(test_bit(COW_ZNODE, &znode->flags));
91 
92 	__clear_bit(DIRTY_ZNODE, &znode->flags);
93 	__clear_bit(COW_ZNODE, &znode->flags);
94 
95 	return err;
96 }
97 
98 /**
99  * fill_gap - make index nodes in gaps in dirty index LEBs.
100  * @c: UBIFS file-system description object
101  * @lnum: LEB number that gap appears in
102  * @gap_start: offset of start of gap
103  * @gap_end: offset of end of gap
104  * @dirt: adds dirty space to this
105  *
106  * This function returns the number of index nodes written into the gap.
107  */
108 static int fill_gap(struct ubifs_info *c, int lnum, int gap_start, int gap_end,
109 		    int *dirt)
110 {
111 	int len, gap_remains, gap_pos, written, pad_len;
112 
113 	ubifs_assert((gap_start & 7) == 0);
114 	ubifs_assert((gap_end & 7) == 0);
115 	ubifs_assert(gap_end >= gap_start);
116 
117 	gap_remains = gap_end - gap_start;
118 	if (!gap_remains)
119 		return 0;
120 	gap_pos = gap_start;
121 	written = 0;
122 	while (c->enext) {
123 		len = ubifs_idx_node_sz(c, c->enext->child_cnt);
124 		if (len < gap_remains) {
125 			struct ubifs_znode *znode = c->enext;
126 			const int alen = ALIGN(len, 8);
127 			int err;
128 
129 			ubifs_assert(alen <= gap_remains);
130 			err = make_idx_node(c, c->ileb_buf + gap_pos, znode,
131 					    lnum, gap_pos, len);
132 			if (err)
133 				return err;
134 			gap_remains -= alen;
135 			gap_pos += alen;
136 			c->enext = znode->cnext;
137 			if (c->enext == c->cnext)
138 				c->enext = NULL;
139 			written += 1;
140 		} else
141 			break;
142 	}
143 	if (gap_end == c->leb_size) {
144 		c->ileb_len = ALIGN(gap_pos, c->min_io_size);
145 		/* Pad to end of min_io_size */
146 		pad_len = c->ileb_len - gap_pos;
147 	} else
148 		/* Pad to end of gap */
149 		pad_len = gap_remains;
150 	dbg_gc("LEB %d:%d to %d len %d nodes written %d wasted bytes %d",
151 	       lnum, gap_start, gap_end, gap_end - gap_start, written, pad_len);
152 	ubifs_pad(c, c->ileb_buf + gap_pos, pad_len);
153 	*dirt += pad_len;
154 	return written;
155 }
156 
157 /**
158  * find_old_idx - find an index node obsoleted since the last commit start.
159  * @c: UBIFS file-system description object
160  * @lnum: LEB number of obsoleted index node
161  * @offs: offset of obsoleted index node
162  *
163  * Returns %1 if found and %0 otherwise.
164  */
165 static int find_old_idx(struct ubifs_info *c, int lnum, int offs)
166 {
167 	struct ubifs_old_idx *o;
168 	struct rb_node *p;
169 
170 	p = c->old_idx.rb_node;
171 	while (p) {
172 		o = rb_entry(p, struct ubifs_old_idx, rb);
173 		if (lnum < o->lnum)
174 			p = p->rb_left;
175 		else if (lnum > o->lnum)
176 			p = p->rb_right;
177 		else if (offs < o->offs)
178 			p = p->rb_left;
179 		else if (offs > o->offs)
180 			p = p->rb_right;
181 		else
182 			return 1;
183 	}
184 	return 0;
185 }
186 
187 /**
188  * is_idx_node_in_use - determine if an index node can be overwritten.
189  * @c: UBIFS file-system description object
190  * @key: key of index node
191  * @level: index node level
192  * @lnum: LEB number of index node
193  * @offs: offset of index node
194  *
195  * If @key / @lnum / @offs identify an index node that was not part of the old
196  * index, then this function returns %0 (obsolete).  Else if the index node was
197  * part of the old index but is now dirty %1 is returned, else if it is clean %2
198  * is returned. A negative error code is returned on failure.
199  */
200 static int is_idx_node_in_use(struct ubifs_info *c, union ubifs_key *key,
201 			      int level, int lnum, int offs)
202 {
203 	int ret;
204 
205 	ret = is_idx_node_in_tnc(c, key, level, lnum, offs);
206 	if (ret < 0)
207 		return ret; /* Error code */
208 	if (ret == 0)
209 		if (find_old_idx(c, lnum, offs))
210 			return 1;
211 	return ret;
212 }
213 
214 /**
215  * layout_leb_in_gaps - layout index nodes using in-the-gaps method.
216  * @c: UBIFS file-system description object
217  * @p: return LEB number here
218  *
219  * This function lays out new index nodes for dirty znodes using in-the-gaps
220  * method of TNC commit.
221  * This function merely puts the next znode into the next gap, making no attempt
222  * to try to maximise the number of znodes that fit.
223  * This function returns the number of index nodes written into the gaps, or a
224  * negative error code on failure.
225  */
226 static int layout_leb_in_gaps(struct ubifs_info *c, int *p)
227 {
228 	struct ubifs_scan_leb *sleb;
229 	struct ubifs_scan_node *snod;
230 	int lnum, dirt = 0, gap_start, gap_end, err, written, tot_written;
231 
232 	tot_written = 0;
233 	/* Get an index LEB with lots of obsolete index nodes */
234 	lnum = ubifs_find_dirty_idx_leb(c);
235 	if (lnum < 0)
236 		/*
237 		 * There also may be dirt in the index head that could be
238 		 * filled, however we do not check there at present.
239 		 */
240 		return lnum; /* Error code */
241 	*p = lnum;
242 	dbg_gc("LEB %d", lnum);
243 	/*
244 	 * Scan the index LEB.  We use the generic scan for this even though
245 	 * it is more comprehensive and less efficient than is needed for this
246 	 * purpose.
247 	 */
248 	sleb = ubifs_scan(c, lnum, 0, c->ileb_buf);
249 	c->ileb_len = 0;
250 	if (IS_ERR(sleb))
251 		return PTR_ERR(sleb);
252 	gap_start = 0;
253 	list_for_each_entry(snod, &sleb->nodes, list) {
254 		struct ubifs_idx_node *idx;
255 		int in_use, level;
256 
257 		ubifs_assert(snod->type == UBIFS_IDX_NODE);
258 		idx = snod->node;
259 		key_read(c, ubifs_idx_key(c, idx), &snod->key);
260 		level = le16_to_cpu(idx->level);
261 		/* Determine if the index node is in use (not obsolete) */
262 		in_use = is_idx_node_in_use(c, &snod->key, level, lnum,
263 					    snod->offs);
264 		if (in_use < 0) {
265 			ubifs_scan_destroy(sleb);
266 			return in_use; /* Error code */
267 		}
268 		if (in_use) {
269 			if (in_use == 1)
270 				dirt += ALIGN(snod->len, 8);
271 			/*
272 			 * The obsolete index nodes form gaps that can be
273 			 * overwritten.  This gap has ended because we have
274 			 * found an index node that is still in use
275 			 * i.e. not obsolete
276 			 */
277 			gap_end = snod->offs;
278 			/* Try to fill gap */
279 			written = fill_gap(c, lnum, gap_start, gap_end, &dirt);
280 			if (written < 0) {
281 				ubifs_scan_destroy(sleb);
282 				return written; /* Error code */
283 			}
284 			tot_written += written;
285 			gap_start = ALIGN(snod->offs + snod->len, 8);
286 		}
287 	}
288 	ubifs_scan_destroy(sleb);
289 	c->ileb_len = c->leb_size;
290 	gap_end = c->leb_size;
291 	/* Try to fill gap */
292 	written = fill_gap(c, lnum, gap_start, gap_end, &dirt);
293 	if (written < 0)
294 		return written; /* Error code */
295 	tot_written += written;
296 	if (tot_written == 0) {
297 		struct ubifs_lprops lp;
298 
299 		dbg_gc("LEB %d wrote %d index nodes", lnum, tot_written);
300 		err = ubifs_read_one_lp(c, lnum, &lp);
301 		if (err)
302 			return err;
303 		if (lp.free == c->leb_size) {
304 			/*
305 			 * We must have snatched this LEB from the idx_gc list
306 			 * so we need to correct the free and dirty space.
307 			 */
308 			err = ubifs_change_one_lp(c, lnum,
309 						  c->leb_size - c->ileb_len,
310 						  dirt, 0, 0, 0);
311 			if (err)
312 				return err;
313 		}
314 		return 0;
315 	}
316 	err = ubifs_change_one_lp(c, lnum, c->leb_size - c->ileb_len, dirt,
317 				  0, 0, 0);
318 	if (err)
319 		return err;
320 	err = ubifs_leb_change(c, lnum, c->ileb_buf, c->ileb_len,
321 			       UBI_SHORTTERM);
322 	if (err)
323 		return err;
324 	dbg_gc("LEB %d wrote %d index nodes", lnum, tot_written);
325 	return tot_written;
326 }
327 
328 /**
329  * get_leb_cnt - calculate the number of empty LEBs needed to commit.
330  * @c: UBIFS file-system description object
331  * @cnt: number of znodes to commit
332  *
333  * This function returns the number of empty LEBs needed to commit @cnt znodes
334  * to the current index head.  The number is not exact and may be more than
335  * needed.
336  */
337 static int get_leb_cnt(struct ubifs_info *c, int cnt)
338 {
339 	int d;
340 
341 	/* Assume maximum index node size (i.e. overestimate space needed) */
342 	cnt -= (c->leb_size - c->ihead_offs) / c->max_idx_node_sz;
343 	if (cnt < 0)
344 		cnt = 0;
345 	d = c->leb_size / c->max_idx_node_sz;
346 	return DIV_ROUND_UP(cnt, d);
347 }
348 
349 /**
350  * layout_in_gaps - in-the-gaps method of committing TNC.
351  * @c: UBIFS file-system description object
352  * @cnt: number of dirty znodes to commit.
353  *
354  * This function lays out new index nodes for dirty znodes using in-the-gaps
355  * method of TNC commit.
356  *
357  * This function returns %0 on success and a negative error code on failure.
358  */
359 static int layout_in_gaps(struct ubifs_info *c, int cnt)
360 {
361 	int err, leb_needed_cnt, written, *p;
362 
363 	dbg_gc("%d znodes to write", cnt);
364 
365 	c->gap_lebs = kmalloc(sizeof(int) * (c->lst.idx_lebs + 1), GFP_NOFS);
366 	if (!c->gap_lebs)
367 		return -ENOMEM;
368 
369 	p = c->gap_lebs;
370 	do {
371 		ubifs_assert(p < c->gap_lebs + sizeof(int) * c->lst.idx_lebs);
372 		written = layout_leb_in_gaps(c, p);
373 		if (written < 0) {
374 			err = written;
375 			if (err == -ENOSPC) {
376 				if (!dbg_force_in_the_gaps_enabled) {
377 					/*
378 					 * Do not print scary warnings if the
379 					 * debugging option which forces
380 					 * in-the-gaps is enabled.
381 					 */
382 					ubifs_err("out of space");
383 					spin_lock(&c->space_lock);
384 					dbg_dump_budg(c);
385 					spin_unlock(&c->space_lock);
386 					dbg_dump_lprops(c);
387 				}
388 				/* Try to commit anyway */
389 				err = 0;
390 				break;
391 			}
392 			kfree(c->gap_lebs);
393 			c->gap_lebs = NULL;
394 			return err;
395 		}
396 		p++;
397 		cnt -= written;
398 		leb_needed_cnt = get_leb_cnt(c, cnt);
399 		dbg_gc("%d znodes remaining, need %d LEBs, have %d", cnt,
400 		       leb_needed_cnt, c->ileb_cnt);
401 	} while (leb_needed_cnt > c->ileb_cnt);
402 
403 	*p = -1;
404 	return 0;
405 }
406 
407 /**
408  * layout_in_empty_space - layout index nodes in empty space.
409  * @c: UBIFS file-system description object
410  *
411  * This function lays out new index nodes for dirty znodes using empty LEBs.
412  *
413  * This function returns %0 on success and a negative error code on failure.
414  */
415 static int layout_in_empty_space(struct ubifs_info *c)
416 {
417 	struct ubifs_znode *znode, *cnext, *zp;
418 	int lnum, offs, len, next_len, buf_len, buf_offs, used, avail;
419 	int wlen, blen, err;
420 
421 	cnext = c->enext;
422 	if (!cnext)
423 		return 0;
424 
425 	lnum = c->ihead_lnum;
426 	buf_offs = c->ihead_offs;
427 
428 	buf_len = ubifs_idx_node_sz(c, c->fanout);
429 	buf_len = ALIGN(buf_len, c->min_io_size);
430 	used = 0;
431 	avail = buf_len;
432 
433 	/* Ensure there is enough room for first write */
434 	next_len = ubifs_idx_node_sz(c, cnext->child_cnt);
435 	if (buf_offs + next_len > c->leb_size)
436 		lnum = -1;
437 
438 	while (1) {
439 		znode = cnext;
440 
441 		len = ubifs_idx_node_sz(c, znode->child_cnt);
442 
443 		/* Determine the index node position */
444 		if (lnum == -1) {
445 			if (c->ileb_nxt >= c->ileb_cnt) {
446 				ubifs_err("out of space");
447 				return -ENOSPC;
448 			}
449 			lnum = c->ilebs[c->ileb_nxt++];
450 			buf_offs = 0;
451 			used = 0;
452 			avail = buf_len;
453 		}
454 
455 		offs = buf_offs + used;
456 
457 #ifdef CONFIG_UBIFS_FS_DEBUG
458 		znode->lnum = lnum;
459 		znode->offs = offs;
460 		znode->len = len;
461 #endif
462 
463 		/* Update the parent */
464 		zp = znode->parent;
465 		if (zp) {
466 			struct ubifs_zbranch *zbr;
467 			int i;
468 
469 			i = znode->iip;
470 			zbr = &zp->zbranch[i];
471 			zbr->lnum = lnum;
472 			zbr->offs = offs;
473 			zbr->len = len;
474 		} else {
475 			c->zroot.lnum = lnum;
476 			c->zroot.offs = offs;
477 			c->zroot.len = len;
478 		}
479 		c->calc_idx_sz += ALIGN(len, 8);
480 
481 		/*
482 		 * Once lprops is updated, we can decrease the dirty znode count
483 		 * but it is easier to just do it here.
484 		 */
485 		atomic_long_dec(&c->dirty_zn_cnt);
486 
487 		/*
488 		 * Calculate the next index node length to see if there is
489 		 * enough room for it
490 		 */
491 		cnext = znode->cnext;
492 		if (cnext == c->cnext)
493 			next_len = 0;
494 		else
495 			next_len = ubifs_idx_node_sz(c, cnext->child_cnt);
496 
497 		if (c->min_io_size == 1) {
498 			buf_offs += ALIGN(len, 8);
499 			if (next_len) {
500 				if (buf_offs + next_len <= c->leb_size)
501 					continue;
502 				err = ubifs_update_one_lp(c, lnum, 0,
503 						c->leb_size - buf_offs, 0, 0);
504 				if (err)
505 					return err;
506 				lnum = -1;
507 				continue;
508 			}
509 			err = ubifs_update_one_lp(c, lnum,
510 					c->leb_size - buf_offs, 0, 0, 0);
511 			if (err)
512 				return err;
513 			break;
514 		}
515 
516 		/* Update buffer positions */
517 		wlen = used + len;
518 		used += ALIGN(len, 8);
519 		avail -= ALIGN(len, 8);
520 
521 		if (next_len != 0 &&
522 		    buf_offs + used + next_len <= c->leb_size &&
523 		    avail > 0)
524 			continue;
525 
526 		if (avail <= 0 && next_len &&
527 		    buf_offs + used + next_len <= c->leb_size)
528 			blen = buf_len;
529 		else
530 			blen = ALIGN(wlen, c->min_io_size);
531 
532 		/* The buffer is full or there are no more znodes to do */
533 		buf_offs += blen;
534 		if (next_len) {
535 			if (buf_offs + next_len > c->leb_size) {
536 				err = ubifs_update_one_lp(c, lnum,
537 					c->leb_size - buf_offs, blen - used,
538 					0, 0);
539 				if (err)
540 					return err;
541 				lnum = -1;
542 			}
543 			used -= blen;
544 			if (used < 0)
545 				used = 0;
546 			avail = buf_len - used;
547 			continue;
548 		}
549 		err = ubifs_update_one_lp(c, lnum, c->leb_size - buf_offs,
550 					  blen - used, 0, 0);
551 		if (err)
552 			return err;
553 		break;
554 	}
555 
556 #ifdef CONFIG_UBIFS_FS_DEBUG
557 	c->new_ihead_lnum = lnum;
558 	c->new_ihead_offs = buf_offs;
559 #endif
560 
561 	return 0;
562 }
563 
564 /**
565  * layout_commit - determine positions of index nodes to commit.
566  * @c: UBIFS file-system description object
567  * @no_space: indicates that insufficient empty LEBs were allocated
568  * @cnt: number of znodes to commit
569  *
570  * Calculate and update the positions of index nodes to commit.  If there were
571  * an insufficient number of empty LEBs allocated, then index nodes are placed
572  * into the gaps created by obsolete index nodes in non-empty index LEBs.  For
573  * this purpose, an obsolete index node is one that was not in the index as at
574  * the end of the last commit.  To write "in-the-gaps" requires that those index
575  * LEBs are updated atomically in-place.
576  */
577 static int layout_commit(struct ubifs_info *c, int no_space, int cnt)
578 {
579 	int err;
580 
581 	if (no_space) {
582 		err = layout_in_gaps(c, cnt);
583 		if (err)
584 			return err;
585 	}
586 	err = layout_in_empty_space(c);
587 	return err;
588 }
589 
590 /**
591  * find_first_dirty - find first dirty znode.
592  * @znode: znode to begin searching from
593  */
594 static struct ubifs_znode *find_first_dirty(struct ubifs_znode *znode)
595 {
596 	int i, cont;
597 
598 	if (!znode)
599 		return NULL;
600 
601 	while (1) {
602 		if (znode->level == 0) {
603 			if (ubifs_zn_dirty(znode))
604 				return znode;
605 			return NULL;
606 		}
607 		cont = 0;
608 		for (i = 0; i < znode->child_cnt; i++) {
609 			struct ubifs_zbranch *zbr = &znode->zbranch[i];
610 
611 			if (zbr->znode && ubifs_zn_dirty(zbr->znode)) {
612 				znode = zbr->znode;
613 				cont = 1;
614 				break;
615 			}
616 		}
617 		if (!cont) {
618 			if (ubifs_zn_dirty(znode))
619 				return znode;
620 			return NULL;
621 		}
622 	}
623 }
624 
625 /**
626  * find_next_dirty - find next dirty znode.
627  * @znode: znode to begin searching from
628  */
629 static struct ubifs_znode *find_next_dirty(struct ubifs_znode *znode)
630 {
631 	int n = znode->iip + 1;
632 
633 	znode = znode->parent;
634 	if (!znode)
635 		return NULL;
636 	for (; n < znode->child_cnt; n++) {
637 		struct ubifs_zbranch *zbr = &znode->zbranch[n];
638 
639 		if (zbr->znode && ubifs_zn_dirty(zbr->znode))
640 			return find_first_dirty(zbr->znode);
641 	}
642 	return znode;
643 }
644 
645 /**
646  * get_znodes_to_commit - create list of dirty znodes to commit.
647  * @c: UBIFS file-system description object
648  *
649  * This function returns the number of znodes to commit.
650  */
651 static int get_znodes_to_commit(struct ubifs_info *c)
652 {
653 	struct ubifs_znode *znode, *cnext;
654 	int cnt = 0;
655 
656 	c->cnext = find_first_dirty(c->zroot.znode);
657 	znode = c->enext = c->cnext;
658 	if (!znode) {
659 		dbg_cmt("no znodes to commit");
660 		return 0;
661 	}
662 	cnt += 1;
663 	while (1) {
664 		ubifs_assert(!test_bit(COW_ZNODE, &znode->flags));
665 		__set_bit(COW_ZNODE, &znode->flags);
666 		znode->alt = 0;
667 		cnext = find_next_dirty(znode);
668 		if (!cnext) {
669 			znode->cnext = c->cnext;
670 			break;
671 		}
672 		znode->cnext = cnext;
673 		znode = cnext;
674 		cnt += 1;
675 	}
676 	dbg_cmt("committing %d znodes", cnt);
677 	ubifs_assert(cnt == atomic_long_read(&c->dirty_zn_cnt));
678 	return cnt;
679 }
680 
681 /**
682  * alloc_idx_lebs - allocate empty LEBs to be used to commit.
683  * @c: UBIFS file-system description object
684  * @cnt: number of znodes to commit
685  *
686  * This function returns %-ENOSPC if it cannot allocate a sufficient number of
687  * empty LEBs.  %0 is returned on success, otherwise a negative error code
688  * is returned.
689  */
690 static int alloc_idx_lebs(struct ubifs_info *c, int cnt)
691 {
692 	int i, leb_cnt, lnum;
693 
694 	c->ileb_cnt = 0;
695 	c->ileb_nxt = 0;
696 	leb_cnt = get_leb_cnt(c, cnt);
697 	dbg_cmt("need about %d empty LEBS for TNC commit", leb_cnt);
698 	if (!leb_cnt)
699 		return 0;
700 	c->ilebs = kmalloc(leb_cnt * sizeof(int), GFP_NOFS);
701 	if (!c->ilebs)
702 		return -ENOMEM;
703 	for (i = 0; i < leb_cnt; i++) {
704 		lnum = ubifs_find_free_leb_for_idx(c);
705 		if (lnum < 0)
706 			return lnum;
707 		c->ilebs[c->ileb_cnt++] = lnum;
708 		dbg_cmt("LEB %d", lnum);
709 	}
710 	if (dbg_force_in_the_gaps())
711 		return -ENOSPC;
712 	return 0;
713 }
714 
715 /**
716  * free_unused_idx_lebs - free unused LEBs that were allocated for the commit.
717  * @c: UBIFS file-system description object
718  *
719  * It is possible that we allocate more empty LEBs for the commit than we need.
720  * This functions frees the surplus.
721  *
722  * This function returns %0 on success and a negative error code on failure.
723  */
724 static int free_unused_idx_lebs(struct ubifs_info *c)
725 {
726 	int i, err = 0, lnum, er;
727 
728 	for (i = c->ileb_nxt; i < c->ileb_cnt; i++) {
729 		lnum = c->ilebs[i];
730 		dbg_cmt("LEB %d", lnum);
731 		er = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
732 					 LPROPS_INDEX | LPROPS_TAKEN, 0);
733 		if (!err)
734 			err = er;
735 	}
736 	return err;
737 }
738 
739 /**
740  * free_idx_lebs - free unused LEBs after commit end.
741  * @c: UBIFS file-system description object
742  *
743  * This function returns %0 on success and a negative error code on failure.
744  */
745 static int free_idx_lebs(struct ubifs_info *c)
746 {
747 	int err;
748 
749 	err = free_unused_idx_lebs(c);
750 	kfree(c->ilebs);
751 	c->ilebs = NULL;
752 	return err;
753 }
754 
755 /**
756  * ubifs_tnc_start_commit - start TNC commit.
757  * @c: UBIFS file-system description object
758  * @zroot: new index root position is returned here
759  *
760  * This function prepares the list of indexing nodes to commit and lays out
761  * their positions on flash. If there is not enough free space it uses the
762  * in-gap commit method. Returns zero in case of success and a negative error
763  * code in case of failure.
764  */
765 int ubifs_tnc_start_commit(struct ubifs_info *c, struct ubifs_zbranch *zroot)
766 {
767 	int err = 0, cnt;
768 
769 	mutex_lock(&c->tnc_mutex);
770 	err = dbg_check_tnc(c, 1);
771 	if (err)
772 		goto out;
773 	cnt = get_znodes_to_commit(c);
774 	if (cnt != 0) {
775 		int no_space = 0;
776 
777 		err = alloc_idx_lebs(c, cnt);
778 		if (err == -ENOSPC)
779 			no_space = 1;
780 		else if (err)
781 			goto out_free;
782 		err = layout_commit(c, no_space, cnt);
783 		if (err)
784 			goto out_free;
785 		ubifs_assert(atomic_long_read(&c->dirty_zn_cnt) == 0);
786 		err = free_unused_idx_lebs(c);
787 		if (err)
788 			goto out;
789 	}
790 	destroy_old_idx(c);
791 	memcpy(zroot, &c->zroot, sizeof(struct ubifs_zbranch));
792 
793 	err = ubifs_save_dirty_idx_lnums(c);
794 	if (err)
795 		goto out;
796 
797 	spin_lock(&c->space_lock);
798 	/*
799 	 * Although we have not finished committing yet, update size of the
800 	 * committed index ('c->old_idx_sz') and zero out the index growth
801 	 * budget. It is OK to do this now, because we've reserved all the
802 	 * space which is needed to commit the index, and it is save for the
803 	 * budgeting subsystem to assume the index is already committed,
804 	 * even though it is not.
805 	 */
806 	c->old_idx_sz = c->calc_idx_sz;
807 	c->budg_uncommitted_idx = 0;
808 	spin_unlock(&c->space_lock);
809 	mutex_unlock(&c->tnc_mutex);
810 
811 	dbg_cmt("number of index LEBs %d", c->lst.idx_lebs);
812 	dbg_cmt("size of index %llu", c->calc_idx_sz);
813 	return err;
814 
815 out_free:
816 	free_idx_lebs(c);
817 out:
818 	mutex_unlock(&c->tnc_mutex);
819 	return err;
820 }
821 
822 /**
823  * write_index - write index nodes.
824  * @c: UBIFS file-system description object
825  *
826  * This function writes the index nodes whose positions were laid out in the
827  * layout_in_empty_space function.
828  */
829 static int write_index(struct ubifs_info *c)
830 {
831 	struct ubifs_idx_node *idx;
832 	struct ubifs_znode *znode, *cnext;
833 	int i, lnum, offs, len, next_len, buf_len, buf_offs, used;
834 	int avail, wlen, err, lnum_pos = 0;
835 
836 	cnext = c->enext;
837 	if (!cnext)
838 		return 0;
839 
840 	/*
841 	 * Always write index nodes to the index head so that index nodes and
842 	 * other types of nodes are never mixed in the same erase block.
843 	 */
844 	lnum = c->ihead_lnum;
845 	buf_offs = c->ihead_offs;
846 
847 	/* Allocate commit buffer */
848 	buf_len = ALIGN(c->max_idx_node_sz, c->min_io_size);
849 	used = 0;
850 	avail = buf_len;
851 
852 	/* Ensure there is enough room for first write */
853 	next_len = ubifs_idx_node_sz(c, cnext->child_cnt);
854 	if (buf_offs + next_len > c->leb_size) {
855 		err = ubifs_update_one_lp(c, lnum, LPROPS_NC, 0, 0,
856 					  LPROPS_TAKEN);
857 		if (err)
858 			return err;
859 		lnum = -1;
860 	}
861 
862 	while (1) {
863 		cond_resched();
864 
865 		znode = cnext;
866 		idx = c->cbuf + used;
867 
868 		/* Make index node */
869 		idx->ch.node_type = UBIFS_IDX_NODE;
870 		idx->child_cnt = cpu_to_le16(znode->child_cnt);
871 		idx->level = cpu_to_le16(znode->level);
872 		for (i = 0; i < znode->child_cnt; i++) {
873 			struct ubifs_branch *br = ubifs_idx_branch(c, idx, i);
874 			struct ubifs_zbranch *zbr = &znode->zbranch[i];
875 
876 			key_write_idx(c, &zbr->key, &br->key);
877 			br->lnum = cpu_to_le32(zbr->lnum);
878 			br->offs = cpu_to_le32(zbr->offs);
879 			br->len = cpu_to_le32(zbr->len);
880 			if (!zbr->lnum || !zbr->len) {
881 				ubifs_err("bad ref in znode");
882 				dbg_dump_znode(c, znode);
883 				if (zbr->znode)
884 					dbg_dump_znode(c, zbr->znode);
885 			}
886 		}
887 		len = ubifs_idx_node_sz(c, znode->child_cnt);
888 		ubifs_prepare_node(c, idx, len, 0);
889 
890 		/* Determine the index node position */
891 		if (lnum == -1) {
892 			lnum = c->ilebs[lnum_pos++];
893 			buf_offs = 0;
894 			used = 0;
895 			avail = buf_len;
896 		}
897 		offs = buf_offs + used;
898 
899 #ifdef CONFIG_UBIFS_FS_DEBUG
900 		if (lnum != znode->lnum || offs != znode->offs ||
901 		    len != znode->len) {
902 			ubifs_err("inconsistent znode posn");
903 			return -EINVAL;
904 		}
905 #endif
906 
907 		/* Grab some stuff from znode while we still can */
908 		cnext = znode->cnext;
909 
910 		ubifs_assert(ubifs_zn_dirty(znode));
911 		ubifs_assert(test_bit(COW_ZNODE, &znode->flags));
912 
913 		/*
914 		 * It is important that other threads should see %DIRTY_ZNODE
915 		 * flag cleared before %COW_ZNODE. Specifically, it matters in
916 		 * the 'dirty_cow_znode()' function. This is the reason for the
917 		 * first barrier. Also, we want the bit changes to be seen to
918 		 * other threads ASAP, to avoid unnecesarry copying, which is
919 		 * the reason for the second barrier.
920 		 */
921 		clear_bit(DIRTY_ZNODE, &znode->flags);
922 		smp_mb__before_clear_bit();
923 		clear_bit(COW_ZNODE, &znode->flags);
924 		smp_mb__after_clear_bit();
925 
926 		/* Do not access znode from this point on */
927 
928 		/* Update buffer positions */
929 		wlen = used + len;
930 		used += ALIGN(len, 8);
931 		avail -= ALIGN(len, 8);
932 
933 		/*
934 		 * Calculate the next index node length to see if there is
935 		 * enough room for it
936 		 */
937 		if (cnext == c->cnext)
938 			next_len = 0;
939 		else
940 			next_len = ubifs_idx_node_sz(c, cnext->child_cnt);
941 
942 		if (c->min_io_size == 1) {
943 			/*
944 			 * Write the prepared index node immediately if there is
945 			 * no minimum IO size
946 			 */
947 			err = ubifs_leb_write(c, lnum, c->cbuf, buf_offs,
948 					      wlen, UBI_SHORTTERM);
949 			if (err)
950 				return err;
951 			buf_offs += ALIGN(wlen, 8);
952 			if (next_len) {
953 				used = 0;
954 				avail = buf_len;
955 				if (buf_offs + next_len > c->leb_size) {
956 					err = ubifs_update_one_lp(c, lnum,
957 						LPROPS_NC, 0, 0, LPROPS_TAKEN);
958 					if (err)
959 						return err;
960 					lnum = -1;
961 				}
962 				continue;
963 			}
964 		} else {
965 			int blen, nxt_offs = buf_offs + used + next_len;
966 
967 			if (next_len && nxt_offs <= c->leb_size) {
968 				if (avail > 0)
969 					continue;
970 				else
971 					blen = buf_len;
972 			} else {
973 				wlen = ALIGN(wlen, 8);
974 				blen = ALIGN(wlen, c->min_io_size);
975 				ubifs_pad(c, c->cbuf + wlen, blen - wlen);
976 			}
977 			/*
978 			 * The buffer is full or there are no more znodes
979 			 * to do
980 			 */
981 			err = ubifs_leb_write(c, lnum, c->cbuf, buf_offs,
982 					      blen, UBI_SHORTTERM);
983 			if (err)
984 				return err;
985 			buf_offs += blen;
986 			if (next_len) {
987 				if (nxt_offs > c->leb_size) {
988 					err = ubifs_update_one_lp(c, lnum,
989 						LPROPS_NC, 0, 0, LPROPS_TAKEN);
990 					if (err)
991 						return err;
992 					lnum = -1;
993 				}
994 				used -= blen;
995 				if (used < 0)
996 					used = 0;
997 				avail = buf_len - used;
998 				memmove(c->cbuf, c->cbuf + blen, used);
999 				continue;
1000 			}
1001 		}
1002 		break;
1003 	}
1004 
1005 #ifdef CONFIG_UBIFS_FS_DEBUG
1006 	if (lnum != c->new_ihead_lnum || buf_offs != c->new_ihead_offs) {
1007 		ubifs_err("inconsistent ihead");
1008 		return -EINVAL;
1009 	}
1010 #endif
1011 
1012 	c->ihead_lnum = lnum;
1013 	c->ihead_offs = buf_offs;
1014 
1015 	return 0;
1016 }
1017 
1018 /**
1019  * free_obsolete_znodes - free obsolete znodes.
1020  * @c: UBIFS file-system description object
1021  *
1022  * At the end of commit end, obsolete znodes are freed.
1023  */
1024 static void free_obsolete_znodes(struct ubifs_info *c)
1025 {
1026 	struct ubifs_znode *znode, *cnext;
1027 
1028 	cnext = c->cnext;
1029 	do {
1030 		znode = cnext;
1031 		cnext = znode->cnext;
1032 		if (test_bit(OBSOLETE_ZNODE, &znode->flags))
1033 			kfree(znode);
1034 		else {
1035 			znode->cnext = NULL;
1036 			atomic_long_inc(&c->clean_zn_cnt);
1037 			atomic_long_inc(&ubifs_clean_zn_cnt);
1038 		}
1039 	} while (cnext != c->cnext);
1040 }
1041 
1042 /**
1043  * return_gap_lebs - return LEBs used by the in-gap commit method.
1044  * @c: UBIFS file-system description object
1045  *
1046  * This function clears the "taken" flag for the LEBs which were used by the
1047  * "commit in-the-gaps" method.
1048  */
1049 static int return_gap_lebs(struct ubifs_info *c)
1050 {
1051 	int *p, err;
1052 
1053 	if (!c->gap_lebs)
1054 		return 0;
1055 
1056 	dbg_cmt("");
1057 	for (p = c->gap_lebs; *p != -1; p++) {
1058 		err = ubifs_change_one_lp(c, *p, LPROPS_NC, LPROPS_NC, 0,
1059 					  LPROPS_TAKEN, 0);
1060 		if (err)
1061 			return err;
1062 	}
1063 
1064 	kfree(c->gap_lebs);
1065 	c->gap_lebs = NULL;
1066 	return 0;
1067 }
1068 
1069 /**
1070  * ubifs_tnc_end_commit - update the TNC for commit end.
1071  * @c: UBIFS file-system description object
1072  *
1073  * Write the dirty znodes.
1074  */
1075 int ubifs_tnc_end_commit(struct ubifs_info *c)
1076 {
1077 	int err;
1078 
1079 	if (!c->cnext)
1080 		return 0;
1081 
1082 	err = return_gap_lebs(c);
1083 	if (err)
1084 		return err;
1085 
1086 	err = write_index(c);
1087 	if (err)
1088 		return err;
1089 
1090 	mutex_lock(&c->tnc_mutex);
1091 
1092 	dbg_cmt("TNC height is %d", c->zroot.znode->level + 1);
1093 
1094 	free_obsolete_znodes(c);
1095 
1096 	c->cnext = NULL;
1097 	kfree(c->ilebs);
1098 	c->ilebs = NULL;
1099 
1100 	mutex_unlock(&c->tnc_mutex);
1101 
1102 	return 0;
1103 }
1104