xref: /openbmc/linux/fs/ubifs/recovery.c (revision db181ce0)
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 /*
24  * This file implements functions needed to recover from unclean un-mounts.
25  * When UBIFS is mounted, it checks a flag on the master node to determine if
26  * an un-mount was completed successfully. If not, the process of mounting
27  * incorporates additional checking and fixing of on-flash data structures.
28  * UBIFS always cleans away all remnants of an unclean un-mount, so that
29  * errors do not accumulate. However UBIFS defers recovery if it is mounted
30  * read-only, and the flash is not modified in that case.
31  *
32  * The general UBIFS approach to the recovery is that it recovers from
33  * corruptions which could be caused by power cuts, but it refuses to recover
34  * from corruption caused by other reasons. And UBIFS tries to distinguish
35  * between these 2 reasons of corruptions and silently recover in the former
36  * case and loudly complain in the latter case.
37  *
38  * UBIFS writes only to erased LEBs, so it writes only to the flash space
39  * containing only 0xFFs. UBIFS also always writes strictly from the beginning
40  * of the LEB to the end. And UBIFS assumes that the underlying flash media
41  * writes in @c->max_write_size bytes at a time.
42  *
43  * Hence, if UBIFS finds a corrupted node at offset X, it expects only the min.
44  * I/O unit corresponding to offset X to contain corrupted data, all the
45  * following min. I/O units have to contain empty space (all 0xFFs). If this is
46  * not true, the corruption cannot be the result of a power cut, and UBIFS
47  * refuses to mount.
48  */
49 
50 #include <linux/crc32.h>
51 #include <linux/slab.h>
52 #include "ubifs.h"
53 
54 /**
55  * is_empty - determine whether a buffer is empty (contains all 0xff).
56  * @buf: buffer to clean
57  * @len: length of buffer
58  *
59  * This function returns %1 if the buffer is empty (contains all 0xff) otherwise
60  * %0 is returned.
61  */
62 static int is_empty(void *buf, int len)
63 {
64 	uint8_t *p = buf;
65 	int i;
66 
67 	for (i = 0; i < len; i++)
68 		if (*p++ != 0xff)
69 			return 0;
70 	return 1;
71 }
72 
73 /**
74  * first_non_ff - find offset of the first non-0xff byte.
75  * @buf: buffer to search in
76  * @len: length of buffer
77  *
78  * This function returns offset of the first non-0xff byte in @buf or %-1 if
79  * the buffer contains only 0xff bytes.
80  */
81 static int first_non_ff(void *buf, int len)
82 {
83 	uint8_t *p = buf;
84 	int i;
85 
86 	for (i = 0; i < len; i++)
87 		if (*p++ != 0xff)
88 			return i;
89 	return -1;
90 }
91 
92 /**
93  * get_master_node - get the last valid master node allowing for corruption.
94  * @c: UBIFS file-system description object
95  * @lnum: LEB number
96  * @pbuf: buffer containing the LEB read, is returned here
97  * @mst: master node, if found, is returned here
98  * @cor: corruption, if found, is returned here
99  *
100  * This function allocates a buffer, reads the LEB into it, and finds and
101  * returns the last valid master node allowing for one area of corruption.
102  * The corrupt area, if there is one, must be consistent with the assumption
103  * that it is the result of an unclean unmount while the master node was being
104  * written. Under those circumstances, it is valid to use the previously written
105  * master node.
106  *
107  * This function returns %0 on success and a negative error code on failure.
108  */
109 static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf,
110 			   struct ubifs_mst_node **mst, void **cor)
111 {
112 	const int sz = c->mst_node_alsz;
113 	int err, offs, len;
114 	void *sbuf, *buf;
115 
116 	sbuf = vmalloc(c->leb_size);
117 	if (!sbuf)
118 		return -ENOMEM;
119 
120 	err = ubifs_leb_read(c, lnum, sbuf, 0, c->leb_size, 0);
121 	if (err && err != -EBADMSG)
122 		goto out_free;
123 
124 	/* Find the first position that is definitely not a node */
125 	offs = 0;
126 	buf = sbuf;
127 	len = c->leb_size;
128 	while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) {
129 		struct ubifs_ch *ch = buf;
130 
131 		if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
132 			break;
133 		offs += sz;
134 		buf  += sz;
135 		len  -= sz;
136 	}
137 	/* See if there was a valid master node before that */
138 	if (offs) {
139 		int ret;
140 
141 		offs -= sz;
142 		buf  -= sz;
143 		len  += sz;
144 		ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
145 		if (ret != SCANNED_A_NODE && offs) {
146 			/* Could have been corruption so check one place back */
147 			offs -= sz;
148 			buf  -= sz;
149 			len  += sz;
150 			ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
151 			if (ret != SCANNED_A_NODE)
152 				/*
153 				 * We accept only one area of corruption because
154 				 * we are assuming that it was caused while
155 				 * trying to write a master node.
156 				 */
157 				goto out_err;
158 		}
159 		if (ret == SCANNED_A_NODE) {
160 			struct ubifs_ch *ch = buf;
161 
162 			if (ch->node_type != UBIFS_MST_NODE)
163 				goto out_err;
164 			dbg_rcvry("found a master node at %d:%d", lnum, offs);
165 			*mst = buf;
166 			offs += sz;
167 			buf  += sz;
168 			len  -= sz;
169 		}
170 	}
171 	/* Check for corruption */
172 	if (offs < c->leb_size) {
173 		if (!is_empty(buf, min_t(int, len, sz))) {
174 			*cor = buf;
175 			dbg_rcvry("found corruption at %d:%d", lnum, offs);
176 		}
177 		offs += sz;
178 		buf  += sz;
179 		len  -= sz;
180 	}
181 	/* Check remaining empty space */
182 	if (offs < c->leb_size)
183 		if (!is_empty(buf, len))
184 			goto out_err;
185 	*pbuf = sbuf;
186 	return 0;
187 
188 out_err:
189 	err = -EINVAL;
190 out_free:
191 	vfree(sbuf);
192 	*mst = NULL;
193 	*cor = NULL;
194 	return err;
195 }
196 
197 /**
198  * write_rcvrd_mst_node - write recovered master node.
199  * @c: UBIFS file-system description object
200  * @mst: master node
201  *
202  * This function returns %0 on success and a negative error code on failure.
203  */
204 static int write_rcvrd_mst_node(struct ubifs_info *c,
205 				struct ubifs_mst_node *mst)
206 {
207 	int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz;
208 	__le32 save_flags;
209 
210 	dbg_rcvry("recovery");
211 
212 	save_flags = mst->flags;
213 	mst->flags |= cpu_to_le32(UBIFS_MST_RCVRY);
214 
215 	ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1);
216 	err = ubifs_leb_change(c, lnum, mst, sz);
217 	if (err)
218 		goto out;
219 	err = ubifs_leb_change(c, lnum + 1, mst, sz);
220 	if (err)
221 		goto out;
222 out:
223 	mst->flags = save_flags;
224 	return err;
225 }
226 
227 /**
228  * ubifs_recover_master_node - recover the master node.
229  * @c: UBIFS file-system description object
230  *
231  * This function recovers the master node from corruption that may occur due to
232  * an unclean unmount.
233  *
234  * This function returns %0 on success and a negative error code on failure.
235  */
236 int ubifs_recover_master_node(struct ubifs_info *c)
237 {
238 	void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL;
239 	struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst;
240 	const int sz = c->mst_node_alsz;
241 	int err, offs1, offs2;
242 
243 	dbg_rcvry("recovery");
244 
245 	err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1);
246 	if (err)
247 		goto out_free;
248 
249 	err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2);
250 	if (err)
251 		goto out_free;
252 
253 	if (mst1) {
254 		offs1 = (void *)mst1 - buf1;
255 		if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) &&
256 		    (offs1 == 0 && !cor1)) {
257 			/*
258 			 * mst1 was written by recovery at offset 0 with no
259 			 * corruption.
260 			 */
261 			dbg_rcvry("recovery recovery");
262 			mst = mst1;
263 		} else if (mst2) {
264 			offs2 = (void *)mst2 - buf2;
265 			if (offs1 == offs2) {
266 				/* Same offset, so must be the same */
267 				if (memcmp((void *)mst1 + UBIFS_CH_SZ,
268 					   (void *)mst2 + UBIFS_CH_SZ,
269 					   UBIFS_MST_NODE_SZ - UBIFS_CH_SZ))
270 					goto out_err;
271 				mst = mst1;
272 			} else if (offs2 + sz == offs1) {
273 				/* 1st LEB was written, 2nd was not */
274 				if (cor1)
275 					goto out_err;
276 				mst = mst1;
277 			} else if (offs1 == 0 &&
278 				   c->leb_size - offs2 - sz < sz) {
279 				/* 1st LEB was unmapped and written, 2nd not */
280 				if (cor1)
281 					goto out_err;
282 				mst = mst1;
283 			} else
284 				goto out_err;
285 		} else {
286 			/*
287 			 * 2nd LEB was unmapped and about to be written, so
288 			 * there must be only one master node in the first LEB
289 			 * and no corruption.
290 			 */
291 			if (offs1 != 0 || cor1)
292 				goto out_err;
293 			mst = mst1;
294 		}
295 	} else {
296 		if (!mst2)
297 			goto out_err;
298 		/*
299 		 * 1st LEB was unmapped and about to be written, so there must
300 		 * be no room left in 2nd LEB.
301 		 */
302 		offs2 = (void *)mst2 - buf2;
303 		if (offs2 + sz + sz <= c->leb_size)
304 			goto out_err;
305 		mst = mst2;
306 	}
307 
308 	ubifs_msg("recovered master node from LEB %d",
309 		  (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1));
310 
311 	memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ);
312 
313 	if (c->ro_mount) {
314 		/* Read-only mode. Keep a copy for switching to rw mode */
315 		c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL);
316 		if (!c->rcvrd_mst_node) {
317 			err = -ENOMEM;
318 			goto out_free;
319 		}
320 		memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ);
321 
322 		/*
323 		 * We had to recover the master node, which means there was an
324 		 * unclean reboot. However, it is possible that the master node
325 		 * is clean at this point, i.e., %UBIFS_MST_DIRTY is not set.
326 		 * E.g., consider the following chain of events:
327 		 *
328 		 * 1. UBIFS was cleanly unmounted, so the master node is clean
329 		 * 2. UBIFS is being mounted R/W and starts changing the master
330 		 *    node in the first (%UBIFS_MST_LNUM). A power cut happens,
331 		 *    so this LEB ends up with some amount of garbage at the
332 		 *    end.
333 		 * 3. UBIFS is being mounted R/O. We reach this place and
334 		 *    recover the master node from the second LEB
335 		 *    (%UBIFS_MST_LNUM + 1). But we cannot update the media
336 		 *    because we are being mounted R/O. We have to defer the
337 		 *    operation.
338 		 * 4. However, this master node (@c->mst_node) is marked as
339 		 *    clean (since the step 1). And if we just return, the
340 		 *    mount code will be confused and won't recover the master
341 		 *    node when it is re-mounter R/W later.
342 		 *
343 		 *    Thus, to force the recovery by marking the master node as
344 		 *    dirty.
345 		 */
346 		c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
347 	} else {
348 		/* Write the recovered master node */
349 		c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1;
350 		err = write_rcvrd_mst_node(c, c->mst_node);
351 		if (err)
352 			goto out_free;
353 	}
354 
355 	vfree(buf2);
356 	vfree(buf1);
357 
358 	return 0;
359 
360 out_err:
361 	err = -EINVAL;
362 out_free:
363 	ubifs_err("failed to recover master node");
364 	if (mst1) {
365 		ubifs_err("dumping first master node");
366 		ubifs_dump_node(c, mst1);
367 	}
368 	if (mst2) {
369 		ubifs_err("dumping second master node");
370 		ubifs_dump_node(c, mst2);
371 	}
372 	vfree(buf2);
373 	vfree(buf1);
374 	return err;
375 }
376 
377 /**
378  * ubifs_write_rcvrd_mst_node - write the recovered master node.
379  * @c: UBIFS file-system description object
380  *
381  * This function writes the master node that was recovered during mounting in
382  * read-only mode and must now be written because we are remounting rw.
383  *
384  * This function returns %0 on success and a negative error code on failure.
385  */
386 int ubifs_write_rcvrd_mst_node(struct ubifs_info *c)
387 {
388 	int err;
389 
390 	if (!c->rcvrd_mst_node)
391 		return 0;
392 	c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
393 	c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
394 	err = write_rcvrd_mst_node(c, c->rcvrd_mst_node);
395 	if (err)
396 		return err;
397 	kfree(c->rcvrd_mst_node);
398 	c->rcvrd_mst_node = NULL;
399 	return 0;
400 }
401 
402 /**
403  * is_last_write - determine if an offset was in the last write to a LEB.
404  * @c: UBIFS file-system description object
405  * @buf: buffer to check
406  * @offs: offset to check
407  *
408  * This function returns %1 if @offs was in the last write to the LEB whose data
409  * is in @buf, otherwise %0 is returned. The determination is made by checking
410  * for subsequent empty space starting from the next @c->max_write_size
411  * boundary.
412  */
413 static int is_last_write(const struct ubifs_info *c, void *buf, int offs)
414 {
415 	int empty_offs, check_len;
416 	uint8_t *p;
417 
418 	/*
419 	 * Round up to the next @c->max_write_size boundary i.e. @offs is in
420 	 * the last wbuf written. After that should be empty space.
421 	 */
422 	empty_offs = ALIGN(offs + 1, c->max_write_size);
423 	check_len = c->leb_size - empty_offs;
424 	p = buf + empty_offs - offs;
425 	return is_empty(p, check_len);
426 }
427 
428 /**
429  * clean_buf - clean the data from an LEB sitting in a buffer.
430  * @c: UBIFS file-system description object
431  * @buf: buffer to clean
432  * @lnum: LEB number to clean
433  * @offs: offset from which to clean
434  * @len: length of buffer
435  *
436  * This function pads up to the next min_io_size boundary (if there is one) and
437  * sets empty space to all 0xff. @buf, @offs and @len are updated to the next
438  * @c->min_io_size boundary.
439  */
440 static void clean_buf(const struct ubifs_info *c, void **buf, int lnum,
441 		      int *offs, int *len)
442 {
443 	int empty_offs, pad_len;
444 
445 	lnum = lnum;
446 	dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs);
447 
448 	ubifs_assert(!(*offs & 7));
449 	empty_offs = ALIGN(*offs, c->min_io_size);
450 	pad_len = empty_offs - *offs;
451 	ubifs_pad(c, *buf, pad_len);
452 	*offs += pad_len;
453 	*buf += pad_len;
454 	*len -= pad_len;
455 	memset(*buf, 0xff, c->leb_size - empty_offs);
456 }
457 
458 /**
459  * no_more_nodes - determine if there are no more nodes in a buffer.
460  * @c: UBIFS file-system description object
461  * @buf: buffer to check
462  * @len: length of buffer
463  * @lnum: LEB number of the LEB from which @buf was read
464  * @offs: offset from which @buf was read
465  *
466  * This function ensures that the corrupted node at @offs is the last thing
467  * written to a LEB. This function returns %1 if more data is not found and
468  * %0 if more data is found.
469  */
470 static int no_more_nodes(const struct ubifs_info *c, void *buf, int len,
471 			int lnum, int offs)
472 {
473 	struct ubifs_ch *ch = buf;
474 	int skip, dlen = le32_to_cpu(ch->len);
475 
476 	/* Check for empty space after the corrupt node's common header */
477 	skip = ALIGN(offs + UBIFS_CH_SZ, c->max_write_size) - offs;
478 	if (is_empty(buf + skip, len - skip))
479 		return 1;
480 	/*
481 	 * The area after the common header size is not empty, so the common
482 	 * header must be intact. Check it.
483 	 */
484 	if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) {
485 		dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs);
486 		return 0;
487 	}
488 	/* Now we know the corrupt node's length we can skip over it */
489 	skip = ALIGN(offs + dlen, c->max_write_size) - offs;
490 	/* After which there should be empty space */
491 	if (is_empty(buf + skip, len - skip))
492 		return 1;
493 	dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip);
494 	return 0;
495 }
496 
497 /**
498  * fix_unclean_leb - fix an unclean LEB.
499  * @c: UBIFS file-system description object
500  * @sleb: scanned LEB information
501  * @start: offset where scan started
502  */
503 static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
504 			   int start)
505 {
506 	int lnum = sleb->lnum, endpt = start;
507 
508 	/* Get the end offset of the last node we are keeping */
509 	if (!list_empty(&sleb->nodes)) {
510 		struct ubifs_scan_node *snod;
511 
512 		snod = list_entry(sleb->nodes.prev,
513 				  struct ubifs_scan_node, list);
514 		endpt = snod->offs + snod->len;
515 	}
516 
517 	if (c->ro_mount && !c->remounting_rw) {
518 		/* Add to recovery list */
519 		struct ubifs_unclean_leb *ucleb;
520 
521 		dbg_rcvry("need to fix LEB %d start %d endpt %d",
522 			  lnum, start, sleb->endpt);
523 		ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS);
524 		if (!ucleb)
525 			return -ENOMEM;
526 		ucleb->lnum = lnum;
527 		ucleb->endpt = endpt;
528 		list_add_tail(&ucleb->list, &c->unclean_leb_list);
529 	} else {
530 		/* Write the fixed LEB back to flash */
531 		int err;
532 
533 		dbg_rcvry("fixing LEB %d start %d endpt %d",
534 			  lnum, start, sleb->endpt);
535 		if (endpt == 0) {
536 			err = ubifs_leb_unmap(c, lnum);
537 			if (err)
538 				return err;
539 		} else {
540 			int len = ALIGN(endpt, c->min_io_size);
541 
542 			if (start) {
543 				err = ubifs_leb_read(c, lnum, sleb->buf, 0,
544 						     start, 1);
545 				if (err)
546 					return err;
547 			}
548 			/* Pad to min_io_size */
549 			if (len > endpt) {
550 				int pad_len = len - ALIGN(endpt, 8);
551 
552 				if (pad_len > 0) {
553 					void *buf = sleb->buf + len - pad_len;
554 
555 					ubifs_pad(c, buf, pad_len);
556 				}
557 			}
558 			err = ubifs_leb_change(c, lnum, sleb->buf, len);
559 			if (err)
560 				return err;
561 		}
562 	}
563 	return 0;
564 }
565 
566 /**
567  * drop_last_group - drop the last group of nodes.
568  * @sleb: scanned LEB information
569  * @offs: offset of dropped nodes is returned here
570  *
571  * This is a helper function for 'ubifs_recover_leb()' which drops the last
572  * group of nodes of the scanned LEB.
573  */
574 static void drop_last_group(struct ubifs_scan_leb *sleb, int *offs)
575 {
576 	while (!list_empty(&sleb->nodes)) {
577 		struct ubifs_scan_node *snod;
578 		struct ubifs_ch *ch;
579 
580 		snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
581 				  list);
582 		ch = snod->node;
583 		if (ch->group_type != UBIFS_IN_NODE_GROUP)
584 			break;
585 
586 		dbg_rcvry("dropping grouped node at %d:%d",
587 			  sleb->lnum, snod->offs);
588 		*offs = snod->offs;
589 		list_del(&snod->list);
590 		kfree(snod);
591 		sleb->nodes_cnt -= 1;
592 	}
593 }
594 
595 /**
596  * drop_last_node - drop the last node.
597  * @sleb: scanned LEB information
598  * @offs: offset of dropped nodes is returned here
599  * @grouped: non-zero if whole group of nodes have to be dropped
600  *
601  * This is a helper function for 'ubifs_recover_leb()' which drops the last
602  * node of the scanned LEB.
603  */
604 static void drop_last_node(struct ubifs_scan_leb *sleb, int *offs)
605 {
606 	struct ubifs_scan_node *snod;
607 
608 	if (!list_empty(&sleb->nodes)) {
609 		snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
610 				  list);
611 
612 		dbg_rcvry("dropping last node at %d:%d",
613 			  sleb->lnum, snod->offs);
614 		*offs = snod->offs;
615 		list_del(&snod->list);
616 		kfree(snod);
617 		sleb->nodes_cnt -= 1;
618 	}
619 }
620 
621 /**
622  * ubifs_recover_leb - scan and recover a LEB.
623  * @c: UBIFS file-system description object
624  * @lnum: LEB number
625  * @offs: offset
626  * @sbuf: LEB-sized buffer to use
627  * @jhead: journal head number this LEB belongs to (%-1 if the LEB does not
628  *         belong to any journal head)
629  *
630  * This function does a scan of a LEB, but caters for errors that might have
631  * been caused by the unclean unmount from which we are attempting to recover.
632  * Returns %0 in case of success, %-EUCLEAN if an unrecoverable corruption is
633  * found, and a negative error code in case of failure.
634  */
635 struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum,
636 					 int offs, void *sbuf, int jhead)
637 {
638 	int ret = 0, err, len = c->leb_size - offs, start = offs, min_io_unit;
639 	int grouped = jhead == -1 ? 0 : c->jheads[jhead].grouped;
640 	struct ubifs_scan_leb *sleb;
641 	void *buf = sbuf + offs;
642 
643 	dbg_rcvry("%d:%d, jhead %d, grouped %d", lnum, offs, jhead, grouped);
644 
645 	sleb = ubifs_start_scan(c, lnum, offs, sbuf);
646 	if (IS_ERR(sleb))
647 		return sleb;
648 
649 	ubifs_assert(len >= 8);
650 	while (len >= 8) {
651 		dbg_scan("look at LEB %d:%d (%d bytes left)",
652 			 lnum, offs, len);
653 
654 		cond_resched();
655 
656 		/*
657 		 * Scan quietly until there is an error from which we cannot
658 		 * recover
659 		 */
660 		ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
661 		if (ret == SCANNED_A_NODE) {
662 			/* A valid node, and not a padding node */
663 			struct ubifs_ch *ch = buf;
664 			int node_len;
665 
666 			err = ubifs_add_snod(c, sleb, buf, offs);
667 			if (err)
668 				goto error;
669 			node_len = ALIGN(le32_to_cpu(ch->len), 8);
670 			offs += node_len;
671 			buf += node_len;
672 			len -= node_len;
673 		} else if (ret > 0) {
674 			/* Padding bytes or a valid padding node */
675 			offs += ret;
676 			buf += ret;
677 			len -= ret;
678 		} else if (ret == SCANNED_EMPTY_SPACE ||
679 			   ret == SCANNED_GARBAGE     ||
680 			   ret == SCANNED_A_BAD_PAD_NODE ||
681 			   ret == SCANNED_A_CORRUPT_NODE) {
682 			dbg_rcvry("found corruption (%d) at %d:%d",
683 				  ret, lnum, offs);
684 			break;
685 		} else {
686 			ubifs_err("unexpected return value %d", ret);
687 			err = -EINVAL;
688 			goto error;
689 		}
690 	}
691 
692 	if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE) {
693 		if (!is_last_write(c, buf, offs))
694 			goto corrupted_rescan;
695 	} else if (ret == SCANNED_A_CORRUPT_NODE) {
696 		if (!no_more_nodes(c, buf, len, lnum, offs))
697 			goto corrupted_rescan;
698 	} else if (!is_empty(buf, len)) {
699 		if (!is_last_write(c, buf, offs)) {
700 			int corruption = first_non_ff(buf, len);
701 
702 			/*
703 			 * See header comment for this file for more
704 			 * explanations about the reasons we have this check.
705 			 */
706 			ubifs_err("corrupt empty space LEB %d:%d, corruption starts at %d",
707 				  lnum, offs, corruption);
708 			/* Make sure we dump interesting non-0xFF data */
709 			offs += corruption;
710 			buf += corruption;
711 			goto corrupted;
712 		}
713 	}
714 
715 	min_io_unit = round_down(offs, c->min_io_size);
716 	if (grouped)
717 		/*
718 		 * If nodes are grouped, always drop the incomplete group at
719 		 * the end.
720 		 */
721 		drop_last_group(sleb, &offs);
722 
723 	if (jhead == GCHD) {
724 		/*
725 		 * If this LEB belongs to the GC head then while we are in the
726 		 * middle of the same min. I/O unit keep dropping nodes. So
727 		 * basically, what we want is to make sure that the last min.
728 		 * I/O unit where we saw the corruption is dropped completely
729 		 * with all the uncorrupted nodes which may possibly sit there.
730 		 *
731 		 * In other words, let's name the min. I/O unit where the
732 		 * corruption starts B, and the previous min. I/O unit A. The
733 		 * below code tries to deal with a situation when half of B
734 		 * contains valid nodes or the end of a valid node, and the
735 		 * second half of B contains corrupted data or garbage. This
736 		 * means that UBIFS had been writing to B just before the power
737 		 * cut happened. I do not know how realistic is this scenario
738 		 * that half of the min. I/O unit had been written successfully
739 		 * and the other half not, but this is possible in our 'failure
740 		 * mode emulation' infrastructure at least.
741 		 *
742 		 * So what is the problem, why we need to drop those nodes? Why
743 		 * can't we just clean-up the second half of B by putting a
744 		 * padding node there? We can, and this works fine with one
745 		 * exception which was reproduced with power cut emulation
746 		 * testing and happens extremely rarely.
747 		 *
748 		 * Imagine the file-system is full, we run GC which starts
749 		 * moving valid nodes from LEB X to LEB Y (obviously, LEB Y is
750 		 * the current GC head LEB). The @c->gc_lnum is -1, which means
751 		 * that GC will retain LEB X and will try to continue. Imagine
752 		 * that LEB X is currently the dirtiest LEB, and the amount of
753 		 * used space in LEB Y is exactly the same as amount of free
754 		 * space in LEB X.
755 		 *
756 		 * And a power cut happens when nodes are moved from LEB X to
757 		 * LEB Y. We are here trying to recover LEB Y which is the GC
758 		 * head LEB. We find the min. I/O unit B as described above.
759 		 * Then we clean-up LEB Y by padding min. I/O unit. And later
760 		 * 'ubifs_rcvry_gc_commit()' function fails, because it cannot
761 		 * find a dirty LEB which could be GC'd into LEB Y! Even LEB X
762 		 * does not match because the amount of valid nodes there does
763 		 * not fit the free space in LEB Y any more! And this is
764 		 * because of the padding node which we added to LEB Y. The
765 		 * user-visible effect of this which I once observed and
766 		 * analysed is that we cannot mount the file-system with
767 		 * -ENOSPC error.
768 		 *
769 		 * So obviously, to make sure that situation does not happen we
770 		 * should free min. I/O unit B in LEB Y completely and the last
771 		 * used min. I/O unit in LEB Y should be A. This is basically
772 		 * what the below code tries to do.
773 		 */
774 		while (offs > min_io_unit)
775 			drop_last_node(sleb, &offs);
776 	}
777 
778 	buf = sbuf + offs;
779 	len = c->leb_size - offs;
780 
781 	clean_buf(c, &buf, lnum, &offs, &len);
782 	ubifs_end_scan(c, sleb, lnum, offs);
783 
784 	err = fix_unclean_leb(c, sleb, start);
785 	if (err)
786 		goto error;
787 
788 	return sleb;
789 
790 corrupted_rescan:
791 	/* Re-scan the corrupted data with verbose messages */
792 	ubifs_err("corruption %d", ret);
793 	ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
794 corrupted:
795 	ubifs_scanned_corruption(c, lnum, offs, buf);
796 	err = -EUCLEAN;
797 error:
798 	ubifs_err("LEB %d scanning failed", lnum);
799 	ubifs_scan_destroy(sleb);
800 	return ERR_PTR(err);
801 }
802 
803 /**
804  * get_cs_sqnum - get commit start sequence number.
805  * @c: UBIFS file-system description object
806  * @lnum: LEB number of commit start node
807  * @offs: offset of commit start node
808  * @cs_sqnum: commit start sequence number is returned here
809  *
810  * This function returns %0 on success and a negative error code on failure.
811  */
812 static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs,
813 			unsigned long long *cs_sqnum)
814 {
815 	struct ubifs_cs_node *cs_node = NULL;
816 	int err, ret;
817 
818 	dbg_rcvry("at %d:%d", lnum, offs);
819 	cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL);
820 	if (!cs_node)
821 		return -ENOMEM;
822 	if (c->leb_size - offs < UBIFS_CS_NODE_SZ)
823 		goto out_err;
824 	err = ubifs_leb_read(c, lnum, (void *)cs_node, offs,
825 			     UBIFS_CS_NODE_SZ, 0);
826 	if (err && err != -EBADMSG)
827 		goto out_free;
828 	ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0);
829 	if (ret != SCANNED_A_NODE) {
830 		ubifs_err("Not a valid node");
831 		goto out_err;
832 	}
833 	if (cs_node->ch.node_type != UBIFS_CS_NODE) {
834 		ubifs_err("Node a CS node, type is %d", cs_node->ch.node_type);
835 		goto out_err;
836 	}
837 	if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) {
838 		ubifs_err("CS node cmt_no %llu != current cmt_no %llu",
839 			  (unsigned long long)le64_to_cpu(cs_node->cmt_no),
840 			  c->cmt_no);
841 		goto out_err;
842 	}
843 	*cs_sqnum = le64_to_cpu(cs_node->ch.sqnum);
844 	dbg_rcvry("commit start sqnum %llu", *cs_sqnum);
845 	kfree(cs_node);
846 	return 0;
847 
848 out_err:
849 	err = -EINVAL;
850 out_free:
851 	ubifs_err("failed to get CS sqnum");
852 	kfree(cs_node);
853 	return err;
854 }
855 
856 /**
857  * ubifs_recover_log_leb - scan and recover a log LEB.
858  * @c: UBIFS file-system description object
859  * @lnum: LEB number
860  * @offs: offset
861  * @sbuf: LEB-sized buffer to use
862  *
863  * This function does a scan of a LEB, but caters for errors that might have
864  * been caused by unclean reboots from which we are attempting to recover
865  * (assume that only the last log LEB can be corrupted by an unclean reboot).
866  *
867  * This function returns %0 on success and a negative error code on failure.
868  */
869 struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum,
870 					     int offs, void *sbuf)
871 {
872 	struct ubifs_scan_leb *sleb;
873 	int next_lnum;
874 
875 	dbg_rcvry("LEB %d", lnum);
876 	next_lnum = lnum + 1;
877 	if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs)
878 		next_lnum = UBIFS_LOG_LNUM;
879 	if (next_lnum != c->ltail_lnum) {
880 		/*
881 		 * We can only recover at the end of the log, so check that the
882 		 * next log LEB is empty or out of date.
883 		 */
884 		sleb = ubifs_scan(c, next_lnum, 0, sbuf, 0);
885 		if (IS_ERR(sleb))
886 			return sleb;
887 		if (sleb->nodes_cnt) {
888 			struct ubifs_scan_node *snod;
889 			unsigned long long cs_sqnum = c->cs_sqnum;
890 
891 			snod = list_entry(sleb->nodes.next,
892 					  struct ubifs_scan_node, list);
893 			if (cs_sqnum == 0) {
894 				int err;
895 
896 				err = get_cs_sqnum(c, lnum, offs, &cs_sqnum);
897 				if (err) {
898 					ubifs_scan_destroy(sleb);
899 					return ERR_PTR(err);
900 				}
901 			}
902 			if (snod->sqnum > cs_sqnum) {
903 				ubifs_err("unrecoverable log corruption in LEB %d",
904 					  lnum);
905 				ubifs_scan_destroy(sleb);
906 				return ERR_PTR(-EUCLEAN);
907 			}
908 		}
909 		ubifs_scan_destroy(sleb);
910 	}
911 	return ubifs_recover_leb(c, lnum, offs, sbuf, -1);
912 }
913 
914 /**
915  * recover_head - recover a head.
916  * @c: UBIFS file-system description object
917  * @lnum: LEB number of head to recover
918  * @offs: offset of head to recover
919  * @sbuf: LEB-sized buffer to use
920  *
921  * This function ensures that there is no data on the flash at a head location.
922  *
923  * This function returns %0 on success and a negative error code on failure.
924  */
925 static int recover_head(struct ubifs_info *c, int lnum, int offs, void *sbuf)
926 {
927 	int len = c->max_write_size, err;
928 
929 	if (offs + len > c->leb_size)
930 		len = c->leb_size - offs;
931 
932 	if (!len)
933 		return 0;
934 
935 	/* Read at the head location and check it is empty flash */
936 	err = ubifs_leb_read(c, lnum, sbuf, offs, len, 1);
937 	if (err || !is_empty(sbuf, len)) {
938 		dbg_rcvry("cleaning head at %d:%d", lnum, offs);
939 		if (offs == 0)
940 			return ubifs_leb_unmap(c, lnum);
941 		err = ubifs_leb_read(c, lnum, sbuf, 0, offs, 1);
942 		if (err)
943 			return err;
944 		return ubifs_leb_change(c, lnum, sbuf, offs);
945 	}
946 
947 	return 0;
948 }
949 
950 /**
951  * ubifs_recover_inl_heads - recover index and LPT heads.
952  * @c: UBIFS file-system description object
953  * @sbuf: LEB-sized buffer to use
954  *
955  * This function ensures that there is no data on the flash at the index and
956  * LPT head locations.
957  *
958  * This deals with the recovery of a half-completed journal commit. UBIFS is
959  * careful never to overwrite the last version of the index or the LPT. Because
960  * the index and LPT are wandering trees, data from a half-completed commit will
961  * not be referenced anywhere in UBIFS. The data will be either in LEBs that are
962  * assumed to be empty and will be unmapped anyway before use, or in the index
963  * and LPT heads.
964  *
965  * This function returns %0 on success and a negative error code on failure.
966  */
967 int ubifs_recover_inl_heads(struct ubifs_info *c, void *sbuf)
968 {
969 	int err;
970 
971 	ubifs_assert(!c->ro_mount || c->remounting_rw);
972 
973 	dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs);
974 	err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf);
975 	if (err)
976 		return err;
977 
978 	dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs);
979 	err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf);
980 	if (err)
981 		return err;
982 
983 	return 0;
984 }
985 
986 /**
987  * clean_an_unclean_leb - read and write a LEB to remove corruption.
988  * @c: UBIFS file-system description object
989  * @ucleb: unclean LEB information
990  * @sbuf: LEB-sized buffer to use
991  *
992  * This function reads a LEB up to a point pre-determined by the mount recovery,
993  * checks the nodes, and writes the result back to the flash, thereby cleaning
994  * off any following corruption, or non-fatal ECC errors.
995  *
996  * This function returns %0 on success and a negative error code on failure.
997  */
998 static int clean_an_unclean_leb(struct ubifs_info *c,
999 				struct ubifs_unclean_leb *ucleb, void *sbuf)
1000 {
1001 	int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1;
1002 	void *buf = sbuf;
1003 
1004 	dbg_rcvry("LEB %d len %d", lnum, len);
1005 
1006 	if (len == 0) {
1007 		/* Nothing to read, just unmap it */
1008 		err = ubifs_leb_unmap(c, lnum);
1009 		if (err)
1010 			return err;
1011 		return 0;
1012 	}
1013 
1014 	err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1015 	if (err && err != -EBADMSG)
1016 		return err;
1017 
1018 	while (len >= 8) {
1019 		int ret;
1020 
1021 		cond_resched();
1022 
1023 		/* Scan quietly until there is an error */
1024 		ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
1025 
1026 		if (ret == SCANNED_A_NODE) {
1027 			/* A valid node, and not a padding node */
1028 			struct ubifs_ch *ch = buf;
1029 			int node_len;
1030 
1031 			node_len = ALIGN(le32_to_cpu(ch->len), 8);
1032 			offs += node_len;
1033 			buf += node_len;
1034 			len -= node_len;
1035 			continue;
1036 		}
1037 
1038 		if (ret > 0) {
1039 			/* Padding bytes or a valid padding node */
1040 			offs += ret;
1041 			buf += ret;
1042 			len -= ret;
1043 			continue;
1044 		}
1045 
1046 		if (ret == SCANNED_EMPTY_SPACE) {
1047 			ubifs_err("unexpected empty space at %d:%d",
1048 				  lnum, offs);
1049 			return -EUCLEAN;
1050 		}
1051 
1052 		if (quiet) {
1053 			/* Redo the last scan but noisily */
1054 			quiet = 0;
1055 			continue;
1056 		}
1057 
1058 		ubifs_scanned_corruption(c, lnum, offs, buf);
1059 		return -EUCLEAN;
1060 	}
1061 
1062 	/* Pad to min_io_size */
1063 	len = ALIGN(ucleb->endpt, c->min_io_size);
1064 	if (len > ucleb->endpt) {
1065 		int pad_len = len - ALIGN(ucleb->endpt, 8);
1066 
1067 		if (pad_len > 0) {
1068 			buf = c->sbuf + len - pad_len;
1069 			ubifs_pad(c, buf, pad_len);
1070 		}
1071 	}
1072 
1073 	/* Write back the LEB atomically */
1074 	err = ubifs_leb_change(c, lnum, sbuf, len);
1075 	if (err)
1076 		return err;
1077 
1078 	dbg_rcvry("cleaned LEB %d", lnum);
1079 
1080 	return 0;
1081 }
1082 
1083 /**
1084  * ubifs_clean_lebs - clean LEBs recovered during read-only mount.
1085  * @c: UBIFS file-system description object
1086  * @sbuf: LEB-sized buffer to use
1087  *
1088  * This function cleans a LEB identified during recovery that needs to be
1089  * written but was not because UBIFS was mounted read-only. This happens when
1090  * remounting to read-write mode.
1091  *
1092  * This function returns %0 on success and a negative error code on failure.
1093  */
1094 int ubifs_clean_lebs(struct ubifs_info *c, void *sbuf)
1095 {
1096 	dbg_rcvry("recovery");
1097 	while (!list_empty(&c->unclean_leb_list)) {
1098 		struct ubifs_unclean_leb *ucleb;
1099 		int err;
1100 
1101 		ucleb = list_entry(c->unclean_leb_list.next,
1102 				   struct ubifs_unclean_leb, list);
1103 		err = clean_an_unclean_leb(c, ucleb, sbuf);
1104 		if (err)
1105 			return err;
1106 		list_del(&ucleb->list);
1107 		kfree(ucleb);
1108 	}
1109 	return 0;
1110 }
1111 
1112 /**
1113  * grab_empty_leb - grab an empty LEB to use as GC LEB and run commit.
1114  * @c: UBIFS file-system description object
1115  *
1116  * This is a helper function for 'ubifs_rcvry_gc_commit()' which grabs an empty
1117  * LEB to be used as GC LEB (@c->gc_lnum), and then runs the commit. Returns
1118  * zero in case of success and a negative error code in case of failure.
1119  */
1120 static int grab_empty_leb(struct ubifs_info *c)
1121 {
1122 	int lnum, err;
1123 
1124 	/*
1125 	 * Note, it is very important to first search for an empty LEB and then
1126 	 * run the commit, not vice-versa. The reason is that there might be
1127 	 * only one empty LEB at the moment, the one which has been the
1128 	 * @c->gc_lnum just before the power cut happened. During the regular
1129 	 * UBIFS operation (not now) @c->gc_lnum is marked as "taken", so no
1130 	 * one but GC can grab it. But at this moment this single empty LEB is
1131 	 * not marked as taken, so if we run commit - what happens? Right, the
1132 	 * commit will grab it and write the index there. Remember that the
1133 	 * index always expands as long as there is free space, and it only
1134 	 * starts consolidating when we run out of space.
1135 	 *
1136 	 * IOW, if we run commit now, we might not be able to find a free LEB
1137 	 * after this.
1138 	 */
1139 	lnum = ubifs_find_free_leb_for_idx(c);
1140 	if (lnum < 0) {
1141 		ubifs_err("could not find an empty LEB");
1142 		ubifs_dump_lprops(c);
1143 		ubifs_dump_budg(c, &c->bi);
1144 		return lnum;
1145 	}
1146 
1147 	/* Reset the index flag */
1148 	err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
1149 				  LPROPS_INDEX, 0);
1150 	if (err)
1151 		return err;
1152 
1153 	c->gc_lnum = lnum;
1154 	dbg_rcvry("found empty LEB %d, run commit", lnum);
1155 
1156 	return ubifs_run_commit(c);
1157 }
1158 
1159 /**
1160  * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit.
1161  * @c: UBIFS file-system description object
1162  *
1163  * Out-of-place garbage collection requires always one empty LEB with which to
1164  * start garbage collection. The LEB number is recorded in c->gc_lnum and is
1165  * written to the master node on unmounting. In the case of an unclean unmount
1166  * the value of gc_lnum recorded in the master node is out of date and cannot
1167  * be used. Instead, recovery must allocate an empty LEB for this purpose.
1168  * However, there may not be enough empty space, in which case it must be
1169  * possible to GC the dirtiest LEB into the GC head LEB.
1170  *
1171  * This function also runs the commit which causes the TNC updates from
1172  * size-recovery and orphans to be written to the flash. That is important to
1173  * ensure correct replay order for subsequent mounts.
1174  *
1175  * This function returns %0 on success and a negative error code on failure.
1176  */
1177 int ubifs_rcvry_gc_commit(struct ubifs_info *c)
1178 {
1179 	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
1180 	struct ubifs_lprops lp;
1181 	int err;
1182 
1183 	dbg_rcvry("GC head LEB %d, offs %d", wbuf->lnum, wbuf->offs);
1184 
1185 	c->gc_lnum = -1;
1186 	if (wbuf->lnum == -1 || wbuf->offs == c->leb_size)
1187 		return grab_empty_leb(c);
1188 
1189 	err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2);
1190 	if (err) {
1191 		if (err != -ENOSPC)
1192 			return err;
1193 
1194 		dbg_rcvry("could not find a dirty LEB");
1195 		return grab_empty_leb(c);
1196 	}
1197 
1198 	ubifs_assert(!(lp.flags & LPROPS_INDEX));
1199 	ubifs_assert(lp.free + lp.dirty >= wbuf->offs);
1200 
1201 	/*
1202 	 * We run the commit before garbage collection otherwise subsequent
1203 	 * mounts will see the GC and orphan deletion in a different order.
1204 	 */
1205 	dbg_rcvry("committing");
1206 	err = ubifs_run_commit(c);
1207 	if (err)
1208 		return err;
1209 
1210 	dbg_rcvry("GC'ing LEB %d", lp.lnum);
1211 	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1212 	err = ubifs_garbage_collect_leb(c, &lp);
1213 	if (err >= 0) {
1214 		int err2 = ubifs_wbuf_sync_nolock(wbuf);
1215 
1216 		if (err2)
1217 			err = err2;
1218 	}
1219 	mutex_unlock(&wbuf->io_mutex);
1220 	if (err < 0) {
1221 		ubifs_err("GC failed, error %d", err);
1222 		if (err == -EAGAIN)
1223 			err = -EINVAL;
1224 		return err;
1225 	}
1226 
1227 	ubifs_assert(err == LEB_RETAINED);
1228 	if (err != LEB_RETAINED)
1229 		return -EINVAL;
1230 
1231 	err = ubifs_leb_unmap(c, c->gc_lnum);
1232 	if (err)
1233 		return err;
1234 
1235 	dbg_rcvry("allocated LEB %d for GC", lp.lnum);
1236 	return 0;
1237 }
1238 
1239 /**
1240  * struct size_entry - inode size information for recovery.
1241  * @rb: link in the RB-tree of sizes
1242  * @inum: inode number
1243  * @i_size: size on inode
1244  * @d_size: maximum size based on data nodes
1245  * @exists: indicates whether the inode exists
1246  * @inode: inode if pinned in memory awaiting rw mode to fix it
1247  */
1248 struct size_entry {
1249 	struct rb_node rb;
1250 	ino_t inum;
1251 	loff_t i_size;
1252 	loff_t d_size;
1253 	int exists;
1254 	struct inode *inode;
1255 };
1256 
1257 /**
1258  * add_ino - add an entry to the size tree.
1259  * @c: UBIFS file-system description object
1260  * @inum: inode number
1261  * @i_size: size on inode
1262  * @d_size: maximum size based on data nodes
1263  * @exists: indicates whether the inode exists
1264  */
1265 static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size,
1266 		   loff_t d_size, int exists)
1267 {
1268 	struct rb_node **p = &c->size_tree.rb_node, *parent = NULL;
1269 	struct size_entry *e;
1270 
1271 	while (*p) {
1272 		parent = *p;
1273 		e = rb_entry(parent, struct size_entry, rb);
1274 		if (inum < e->inum)
1275 			p = &(*p)->rb_left;
1276 		else
1277 			p = &(*p)->rb_right;
1278 	}
1279 
1280 	e = kzalloc(sizeof(struct size_entry), GFP_KERNEL);
1281 	if (!e)
1282 		return -ENOMEM;
1283 
1284 	e->inum = inum;
1285 	e->i_size = i_size;
1286 	e->d_size = d_size;
1287 	e->exists = exists;
1288 
1289 	rb_link_node(&e->rb, parent, p);
1290 	rb_insert_color(&e->rb, &c->size_tree);
1291 
1292 	return 0;
1293 }
1294 
1295 /**
1296  * find_ino - find an entry on the size tree.
1297  * @c: UBIFS file-system description object
1298  * @inum: inode number
1299  */
1300 static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum)
1301 {
1302 	struct rb_node *p = c->size_tree.rb_node;
1303 	struct size_entry *e;
1304 
1305 	while (p) {
1306 		e = rb_entry(p, struct size_entry, rb);
1307 		if (inum < e->inum)
1308 			p = p->rb_left;
1309 		else if (inum > e->inum)
1310 			p = p->rb_right;
1311 		else
1312 			return e;
1313 	}
1314 	return NULL;
1315 }
1316 
1317 /**
1318  * remove_ino - remove an entry from the size tree.
1319  * @c: UBIFS file-system description object
1320  * @inum: inode number
1321  */
1322 static void remove_ino(struct ubifs_info *c, ino_t inum)
1323 {
1324 	struct size_entry *e = find_ino(c, inum);
1325 
1326 	if (!e)
1327 		return;
1328 	rb_erase(&e->rb, &c->size_tree);
1329 	kfree(e);
1330 }
1331 
1332 /**
1333  * ubifs_destroy_size_tree - free resources related to the size tree.
1334  * @c: UBIFS file-system description object
1335  */
1336 void ubifs_destroy_size_tree(struct ubifs_info *c)
1337 {
1338 	struct size_entry *e, *n;
1339 
1340 	rbtree_postorder_for_each_entry_safe(e, n, &c->size_tree, rb) {
1341 		if (e->inode)
1342 			iput(e->inode);
1343 		kfree(e);
1344 	}
1345 
1346 	c->size_tree = RB_ROOT;
1347 }
1348 
1349 /**
1350  * ubifs_recover_size_accum - accumulate inode sizes for recovery.
1351  * @c: UBIFS file-system description object
1352  * @key: node key
1353  * @deletion: node is for a deletion
1354  * @new_size: inode size
1355  *
1356  * This function has two purposes:
1357  *     1) to ensure there are no data nodes that fall outside the inode size
1358  *     2) to ensure there are no data nodes for inodes that do not exist
1359  * To accomplish those purposes, a rb-tree is constructed containing an entry
1360  * for each inode number in the journal that has not been deleted, and recording
1361  * the size from the inode node, the maximum size of any data node (also altered
1362  * by truncations) and a flag indicating a inode number for which no inode node
1363  * was present in the journal.
1364  *
1365  * Note that there is still the possibility that there are data nodes that have
1366  * been committed that are beyond the inode size, however the only way to find
1367  * them would be to scan the entire index. Alternatively, some provision could
1368  * be made to record the size of inodes at the start of commit, which would seem
1369  * very cumbersome for a scenario that is quite unlikely and the only negative
1370  * consequence of which is wasted space.
1371  *
1372  * This functions returns %0 on success and a negative error code on failure.
1373  */
1374 int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key,
1375 			     int deletion, loff_t new_size)
1376 {
1377 	ino_t inum = key_inum(c, key);
1378 	struct size_entry *e;
1379 	int err;
1380 
1381 	switch (key_type(c, key)) {
1382 	case UBIFS_INO_KEY:
1383 		if (deletion)
1384 			remove_ino(c, inum);
1385 		else {
1386 			e = find_ino(c, inum);
1387 			if (e) {
1388 				e->i_size = new_size;
1389 				e->exists = 1;
1390 			} else {
1391 				err = add_ino(c, inum, new_size, 0, 1);
1392 				if (err)
1393 					return err;
1394 			}
1395 		}
1396 		break;
1397 	case UBIFS_DATA_KEY:
1398 		e = find_ino(c, inum);
1399 		if (e) {
1400 			if (new_size > e->d_size)
1401 				e->d_size = new_size;
1402 		} else {
1403 			err = add_ino(c, inum, 0, new_size, 0);
1404 			if (err)
1405 				return err;
1406 		}
1407 		break;
1408 	case UBIFS_TRUN_KEY:
1409 		e = find_ino(c, inum);
1410 		if (e)
1411 			e->d_size = new_size;
1412 		break;
1413 	}
1414 	return 0;
1415 }
1416 
1417 /**
1418  * fix_size_in_place - fix inode size in place on flash.
1419  * @c: UBIFS file-system description object
1420  * @e: inode size information for recovery
1421  */
1422 static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e)
1423 {
1424 	struct ubifs_ino_node *ino = c->sbuf;
1425 	unsigned char *p;
1426 	union ubifs_key key;
1427 	int err, lnum, offs, len;
1428 	loff_t i_size;
1429 	uint32_t crc;
1430 
1431 	/* Locate the inode node LEB number and offset */
1432 	ino_key_init(c, &key, e->inum);
1433 	err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs);
1434 	if (err)
1435 		goto out;
1436 	/*
1437 	 * If the size recorded on the inode node is greater than the size that
1438 	 * was calculated from nodes in the journal then don't change the inode.
1439 	 */
1440 	i_size = le64_to_cpu(ino->size);
1441 	if (i_size >= e->d_size)
1442 		return 0;
1443 	/* Read the LEB */
1444 	err = ubifs_leb_read(c, lnum, c->sbuf, 0, c->leb_size, 1);
1445 	if (err)
1446 		goto out;
1447 	/* Change the size field and recalculate the CRC */
1448 	ino = c->sbuf + offs;
1449 	ino->size = cpu_to_le64(e->d_size);
1450 	len = le32_to_cpu(ino->ch.len);
1451 	crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8);
1452 	ino->ch.crc = cpu_to_le32(crc);
1453 	/* Work out where data in the LEB ends and free space begins */
1454 	p = c->sbuf;
1455 	len = c->leb_size - 1;
1456 	while (p[len] == 0xff)
1457 		len -= 1;
1458 	len = ALIGN(len + 1, c->min_io_size);
1459 	/* Atomically write the fixed LEB back again */
1460 	err = ubifs_leb_change(c, lnum, c->sbuf, len);
1461 	if (err)
1462 		goto out;
1463 	dbg_rcvry("inode %lu at %d:%d size %lld -> %lld",
1464 		  (unsigned long)e->inum, lnum, offs, i_size, e->d_size);
1465 	return 0;
1466 
1467 out:
1468 	ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d",
1469 		   (unsigned long)e->inum, e->i_size, e->d_size, err);
1470 	return err;
1471 }
1472 
1473 /**
1474  * ubifs_recover_size - recover inode size.
1475  * @c: UBIFS file-system description object
1476  *
1477  * This function attempts to fix inode size discrepancies identified by the
1478  * 'ubifs_recover_size_accum()' function.
1479  *
1480  * This functions returns %0 on success and a negative error code on failure.
1481  */
1482 int ubifs_recover_size(struct ubifs_info *c)
1483 {
1484 	struct rb_node *this = rb_first(&c->size_tree);
1485 
1486 	while (this) {
1487 		struct size_entry *e;
1488 		int err;
1489 
1490 		e = rb_entry(this, struct size_entry, rb);
1491 		if (!e->exists) {
1492 			union ubifs_key key;
1493 
1494 			ino_key_init(c, &key, e->inum);
1495 			err = ubifs_tnc_lookup(c, &key, c->sbuf);
1496 			if (err && err != -ENOENT)
1497 				return err;
1498 			if (err == -ENOENT) {
1499 				/* Remove data nodes that have no inode */
1500 				dbg_rcvry("removing ino %lu",
1501 					  (unsigned long)e->inum);
1502 				err = ubifs_tnc_remove_ino(c, e->inum);
1503 				if (err)
1504 					return err;
1505 			} else {
1506 				struct ubifs_ino_node *ino = c->sbuf;
1507 
1508 				e->exists = 1;
1509 				e->i_size = le64_to_cpu(ino->size);
1510 			}
1511 		}
1512 
1513 		if (e->exists && e->i_size < e->d_size) {
1514 			if (c->ro_mount) {
1515 				/* Fix the inode size and pin it in memory */
1516 				struct inode *inode;
1517 				struct ubifs_inode *ui;
1518 
1519 				ubifs_assert(!e->inode);
1520 
1521 				inode = ubifs_iget(c->vfs_sb, e->inum);
1522 				if (IS_ERR(inode))
1523 					return PTR_ERR(inode);
1524 
1525 				ui = ubifs_inode(inode);
1526 				if (inode->i_size < e->d_size) {
1527 					dbg_rcvry("ino %lu size %lld -> %lld",
1528 						  (unsigned long)e->inum,
1529 						  inode->i_size, e->d_size);
1530 					inode->i_size = e->d_size;
1531 					ui->ui_size = e->d_size;
1532 					ui->synced_i_size = e->d_size;
1533 					e->inode = inode;
1534 					this = rb_next(this);
1535 					continue;
1536 				}
1537 				iput(inode);
1538 			} else {
1539 				/* Fix the size in place */
1540 				err = fix_size_in_place(c, e);
1541 				if (err)
1542 					return err;
1543 				if (e->inode)
1544 					iput(e->inode);
1545 			}
1546 		}
1547 
1548 		this = rb_next(this);
1549 		rb_erase(&e->rb, &c->size_tree);
1550 		kfree(e);
1551 	}
1552 
1553 	return 0;
1554 }
1555