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  */
is_empty(void * buf,int len)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  */
first_non_ff(void * buf,int len)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  */
get_master_node(const struct ubifs_info * c,int lnum,void ** pbuf,struct ubifs_mst_node ** mst,void ** cor)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  */
write_rcvrd_mst_node(struct ubifs_info * c,struct ubifs_mst_node * mst)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  */
ubifs_recover_master_node(struct ubifs_info * c)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(c, "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(c, "failed to recover master node");
364 	if (mst1) {
365 		ubifs_err(c, "dumping first master node");
366 		ubifs_dump_node(c, mst1);
367 	}
368 	if (mst2) {
369 		ubifs_err(c, "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  */
ubifs_write_rcvrd_mst_node(struct ubifs_info * c)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  */
is_last_write(const struct ubifs_info * c,void * buf,int offs)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  */
clean_buf(const struct ubifs_info * c,void ** buf,int lnum,int * offs,int * len)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 	dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs);
446 
447 	ubifs_assert(c, !(*offs & 7));
448 	empty_offs = ALIGN(*offs, c->min_io_size);
449 	pad_len = empty_offs - *offs;
450 	ubifs_pad(c, *buf, pad_len);
451 	*offs += pad_len;
452 	*buf += pad_len;
453 	*len -= pad_len;
454 	memset(*buf, 0xff, c->leb_size - empty_offs);
455 }
456 
457 /**
458  * no_more_nodes - determine if there are no more nodes in a buffer.
459  * @c: UBIFS file-system description object
460  * @buf: buffer to check
461  * @len: length of buffer
462  * @lnum: LEB number of the LEB from which @buf was read
463  * @offs: offset from which @buf was read
464  *
465  * This function ensures that the corrupted node at @offs is the last thing
466  * written to a LEB. This function returns %1 if more data is not found and
467  * %0 if more data is found.
468  */
no_more_nodes(const struct ubifs_info * c,void * buf,int len,int lnum,int offs)469 static int no_more_nodes(const struct ubifs_info *c, void *buf, int len,
470 			int lnum, int offs)
471 {
472 	struct ubifs_ch *ch = buf;
473 	int skip, dlen = le32_to_cpu(ch->len);
474 
475 	/* Check for empty space after the corrupt node's common header */
476 	skip = ALIGN(offs + UBIFS_CH_SZ, c->max_write_size) - offs;
477 	if (is_empty(buf + skip, len - skip))
478 		return 1;
479 	/*
480 	 * The area after the common header size is not empty, so the common
481 	 * header must be intact. Check it.
482 	 */
483 	if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) {
484 		dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs);
485 		return 0;
486 	}
487 	/* Now we know the corrupt node's length we can skip over it */
488 	skip = ALIGN(offs + dlen, c->max_write_size) - offs;
489 	/* After which there should be empty space */
490 	if (is_empty(buf + skip, len - skip))
491 		return 1;
492 	dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip);
493 	return 0;
494 }
495 
496 /**
497  * fix_unclean_leb - fix an unclean LEB.
498  * @c: UBIFS file-system description object
499  * @sleb: scanned LEB information
500  * @start: offset where scan started
501  */
fix_unclean_leb(struct ubifs_info * c,struct ubifs_scan_leb * sleb,int start)502 static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
503 			   int start)
504 {
505 	int lnum = sleb->lnum, endpt = start;
506 
507 	/* Get the end offset of the last node we are keeping */
508 	if (!list_empty(&sleb->nodes)) {
509 		struct ubifs_scan_node *snod;
510 
511 		snod = list_entry(sleb->nodes.prev,
512 				  struct ubifs_scan_node, list);
513 		endpt = snod->offs + snod->len;
514 	}
515 
516 	if (c->ro_mount && !c->remounting_rw) {
517 		/* Add to recovery list */
518 		struct ubifs_unclean_leb *ucleb;
519 
520 		dbg_rcvry("need to fix LEB %d start %d endpt %d",
521 			  lnum, start, sleb->endpt);
522 		ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS);
523 		if (!ucleb)
524 			return -ENOMEM;
525 		ucleb->lnum = lnum;
526 		ucleb->endpt = endpt;
527 		list_add_tail(&ucleb->list, &c->unclean_leb_list);
528 	} else {
529 		/* Write the fixed LEB back to flash */
530 		int err;
531 
532 		dbg_rcvry("fixing LEB %d start %d endpt %d",
533 			  lnum, start, sleb->endpt);
534 		if (endpt == 0) {
535 			err = ubifs_leb_unmap(c, lnum);
536 			if (err)
537 				return err;
538 		} else {
539 			int len = ALIGN(endpt, c->min_io_size);
540 
541 			if (start) {
542 				err = ubifs_leb_read(c, lnum, sleb->buf, 0,
543 						     start, 1);
544 				if (err)
545 					return err;
546 			}
547 			/* Pad to min_io_size */
548 			if (len > endpt) {
549 				int pad_len = len - ALIGN(endpt, 8);
550 
551 				if (pad_len > 0) {
552 					void *buf = sleb->buf + len - pad_len;
553 
554 					ubifs_pad(c, buf, pad_len);
555 				}
556 			}
557 			err = ubifs_leb_change(c, lnum, sleb->buf, len);
558 			if (err)
559 				return err;
560 		}
561 	}
562 	return 0;
563 }
564 
565 /**
566  * drop_last_group - drop the last group of nodes.
567  * @sleb: scanned LEB information
568  * @offs: offset of dropped nodes is returned here
569  *
570  * This is a helper function for 'ubifs_recover_leb()' which drops the last
571  * group of nodes of the scanned LEB.
572  */
drop_last_group(struct ubifs_scan_leb * sleb,int * offs)573 static void drop_last_group(struct ubifs_scan_leb *sleb, int *offs)
574 {
575 	while (!list_empty(&sleb->nodes)) {
576 		struct ubifs_scan_node *snod;
577 		struct ubifs_ch *ch;
578 
579 		snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
580 				  list);
581 		ch = snod->node;
582 		if (ch->group_type != UBIFS_IN_NODE_GROUP)
583 			break;
584 
585 		dbg_rcvry("dropping grouped node at %d:%d",
586 			  sleb->lnum, snod->offs);
587 		*offs = snod->offs;
588 		list_del(&snod->list);
589 		kfree(snod);
590 		sleb->nodes_cnt -= 1;
591 	}
592 }
593 
594 /**
595  * drop_last_node - drop the last node.
596  * @sleb: scanned LEB information
597  * @offs: offset of dropped nodes is returned here
598  *
599  * This is a helper function for 'ubifs_recover_leb()' which drops the last
600  * node of the scanned LEB.
601  */
drop_last_node(struct ubifs_scan_leb * sleb,int * offs)602 static void drop_last_node(struct ubifs_scan_leb *sleb, int *offs)
603 {
604 	struct ubifs_scan_node *snod;
605 
606 	if (!list_empty(&sleb->nodes)) {
607 		snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
608 				  list);
609 
610 		dbg_rcvry("dropping last node at %d:%d",
611 			  sleb->lnum, snod->offs);
612 		*offs = snod->offs;
613 		list_del(&snod->list);
614 		kfree(snod);
615 		sleb->nodes_cnt -= 1;
616 	}
617 }
618 
619 /**
620  * ubifs_recover_leb - scan and recover a LEB.
621  * @c: UBIFS file-system description object
622  * @lnum: LEB number
623  * @offs: offset
624  * @sbuf: LEB-sized buffer to use
625  * @jhead: journal head number this LEB belongs to (%-1 if the LEB does not
626  *         belong to any journal head)
627  *
628  * This function does a scan of a LEB, but caters for errors that might have
629  * been caused by the unclean unmount from which we are attempting to recover.
630  * Returns the scanned information on success and a negative error code on
631  * failure.
632  */
ubifs_recover_leb(struct ubifs_info * c,int lnum,int offs,void * sbuf,int jhead)633 struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum,
634 					 int offs, void *sbuf, int jhead)
635 {
636 	int ret = 0, err, len = c->leb_size - offs, start = offs, min_io_unit;
637 	int grouped = jhead == -1 ? 0 : c->jheads[jhead].grouped;
638 	struct ubifs_scan_leb *sleb;
639 	void *buf = sbuf + offs;
640 
641 	dbg_rcvry("%d:%d, jhead %d, grouped %d", lnum, offs, jhead, grouped);
642 
643 	sleb = ubifs_start_scan(c, lnum, offs, sbuf);
644 	if (IS_ERR(sleb))
645 		return sleb;
646 
647 	ubifs_assert(c, len >= 8);
648 	while (len >= 8) {
649 		dbg_scan("look at LEB %d:%d (%d bytes left)",
650 			 lnum, offs, len);
651 
652 		cond_resched();
653 
654 		/*
655 		 * Scan quietly until there is an error from which we cannot
656 		 * recover
657 		 */
658 		ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
659 		if (ret == SCANNED_A_NODE) {
660 			/* A valid node, and not a padding node */
661 			struct ubifs_ch *ch = buf;
662 			int node_len;
663 
664 			err = ubifs_add_snod(c, sleb, buf, offs);
665 			if (err)
666 				goto error;
667 			node_len = ALIGN(le32_to_cpu(ch->len), 8);
668 			offs += node_len;
669 			buf += node_len;
670 			len -= node_len;
671 		} else if (ret > 0) {
672 			/* Padding bytes or a valid padding node */
673 			offs += ret;
674 			buf += ret;
675 			len -= ret;
676 		} else if (ret == SCANNED_EMPTY_SPACE ||
677 			   ret == SCANNED_GARBAGE     ||
678 			   ret == SCANNED_A_BAD_PAD_NODE ||
679 			   ret == SCANNED_A_CORRUPT_NODE) {
680 			dbg_rcvry("found corruption (%d) at %d:%d",
681 				  ret, lnum, offs);
682 			break;
683 		} else {
684 			ubifs_err(c, "unexpected return value %d", ret);
685 			err = -EINVAL;
686 			goto error;
687 		}
688 	}
689 
690 	if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE) {
691 		if (!is_last_write(c, buf, offs))
692 			goto corrupted_rescan;
693 	} else if (ret == SCANNED_A_CORRUPT_NODE) {
694 		if (!no_more_nodes(c, buf, len, lnum, offs))
695 			goto corrupted_rescan;
696 	} else if (!is_empty(buf, len)) {
697 		if (!is_last_write(c, buf, offs)) {
698 			int corruption = first_non_ff(buf, len);
699 
700 			/*
701 			 * See header comment for this file for more
702 			 * explanations about the reasons we have this check.
703 			 */
704 			ubifs_err(c, "corrupt empty space LEB %d:%d, corruption starts at %d",
705 				  lnum, offs, corruption);
706 			/* Make sure we dump interesting non-0xFF data */
707 			offs += corruption;
708 			buf += corruption;
709 			goto corrupted;
710 		}
711 	}
712 
713 	min_io_unit = round_down(offs, c->min_io_size);
714 	if (grouped)
715 		/*
716 		 * If nodes are grouped, always drop the incomplete group at
717 		 * the end.
718 		 */
719 		drop_last_group(sleb, &offs);
720 
721 	if (jhead == GCHD) {
722 		/*
723 		 * If this LEB belongs to the GC head then while we are in the
724 		 * middle of the same min. I/O unit keep dropping nodes. So
725 		 * basically, what we want is to make sure that the last min.
726 		 * I/O unit where we saw the corruption is dropped completely
727 		 * with all the uncorrupted nodes which may possibly sit there.
728 		 *
729 		 * In other words, let's name the min. I/O unit where the
730 		 * corruption starts B, and the previous min. I/O unit A. The
731 		 * below code tries to deal with a situation when half of B
732 		 * contains valid nodes or the end of a valid node, and the
733 		 * second half of B contains corrupted data or garbage. This
734 		 * means that UBIFS had been writing to B just before the power
735 		 * cut happened. I do not know how realistic is this scenario
736 		 * that half of the min. I/O unit had been written successfully
737 		 * and the other half not, but this is possible in our 'failure
738 		 * mode emulation' infrastructure at least.
739 		 *
740 		 * So what is the problem, why we need to drop those nodes? Why
741 		 * can't we just clean-up the second half of B by putting a
742 		 * padding node there? We can, and this works fine with one
743 		 * exception which was reproduced with power cut emulation
744 		 * testing and happens extremely rarely.
745 		 *
746 		 * Imagine the file-system is full, we run GC which starts
747 		 * moving valid nodes from LEB X to LEB Y (obviously, LEB Y is
748 		 * the current GC head LEB). The @c->gc_lnum is -1, which means
749 		 * that GC will retain LEB X and will try to continue. Imagine
750 		 * that LEB X is currently the dirtiest LEB, and the amount of
751 		 * used space in LEB Y is exactly the same as amount of free
752 		 * space in LEB X.
753 		 *
754 		 * And a power cut happens when nodes are moved from LEB X to
755 		 * LEB Y. We are here trying to recover LEB Y which is the GC
756 		 * head LEB. We find the min. I/O unit B as described above.
757 		 * Then we clean-up LEB Y by padding min. I/O unit. And later
758 		 * 'ubifs_rcvry_gc_commit()' function fails, because it cannot
759 		 * find a dirty LEB which could be GC'd into LEB Y! Even LEB X
760 		 * does not match because the amount of valid nodes there does
761 		 * not fit the free space in LEB Y any more! And this is
762 		 * because of the padding node which we added to LEB Y. The
763 		 * user-visible effect of this which I once observed and
764 		 * analysed is that we cannot mount the file-system with
765 		 * -ENOSPC error.
766 		 *
767 		 * So obviously, to make sure that situation does not happen we
768 		 * should free min. I/O unit B in LEB Y completely and the last
769 		 * used min. I/O unit in LEB Y should be A. This is basically
770 		 * what the below code tries to do.
771 		 */
772 		while (offs > min_io_unit)
773 			drop_last_node(sleb, &offs);
774 	}
775 
776 	buf = sbuf + offs;
777 	len = c->leb_size - offs;
778 
779 	clean_buf(c, &buf, lnum, &offs, &len);
780 	ubifs_end_scan(c, sleb, lnum, offs);
781 
782 	err = fix_unclean_leb(c, sleb, start);
783 	if (err)
784 		goto error;
785 
786 	return sleb;
787 
788 corrupted_rescan:
789 	/* Re-scan the corrupted data with verbose messages */
790 	ubifs_err(c, "corruption %d", ret);
791 	ubifs_scan_a_node(c, buf, len, lnum, offs, 0);
792 corrupted:
793 	ubifs_scanned_corruption(c, lnum, offs, buf);
794 	err = -EUCLEAN;
795 error:
796 	ubifs_err(c, "LEB %d scanning failed", lnum);
797 	ubifs_scan_destroy(sleb);
798 	return ERR_PTR(err);
799 }
800 
801 /**
802  * get_cs_sqnum - get commit start sequence number.
803  * @c: UBIFS file-system description object
804  * @lnum: LEB number of commit start node
805  * @offs: offset of commit start node
806  * @cs_sqnum: commit start sequence number is returned here
807  *
808  * This function returns %0 on success and a negative error code on failure.
809  */
get_cs_sqnum(struct ubifs_info * c,int lnum,int offs,unsigned long long * cs_sqnum)810 static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs,
811 			unsigned long long *cs_sqnum)
812 {
813 	struct ubifs_cs_node *cs_node = NULL;
814 	int err, ret;
815 
816 	dbg_rcvry("at %d:%d", lnum, offs);
817 	cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL);
818 	if (!cs_node)
819 		return -ENOMEM;
820 	if (c->leb_size - offs < UBIFS_CS_NODE_SZ)
821 		goto out_err;
822 	err = ubifs_leb_read(c, lnum, (void *)cs_node, offs,
823 			     UBIFS_CS_NODE_SZ, 0);
824 	if (err && err != -EBADMSG)
825 		goto out_free;
826 	ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0);
827 	if (ret != SCANNED_A_NODE) {
828 		ubifs_err(c, "Not a valid node");
829 		goto out_err;
830 	}
831 	if (cs_node->ch.node_type != UBIFS_CS_NODE) {
832 		ubifs_err(c, "Node a CS node, type is %d", cs_node->ch.node_type);
833 		goto out_err;
834 	}
835 	if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) {
836 		ubifs_err(c, "CS node cmt_no %llu != current cmt_no %llu",
837 			  (unsigned long long)le64_to_cpu(cs_node->cmt_no),
838 			  c->cmt_no);
839 		goto out_err;
840 	}
841 	*cs_sqnum = le64_to_cpu(cs_node->ch.sqnum);
842 	dbg_rcvry("commit start sqnum %llu", *cs_sqnum);
843 	kfree(cs_node);
844 	return 0;
845 
846 out_err:
847 	err = -EINVAL;
848 out_free:
849 	ubifs_err(c, "failed to get CS sqnum");
850 	kfree(cs_node);
851 	return err;
852 }
853 
854 /**
855  * ubifs_recover_log_leb - scan and recover a log LEB.
856  * @c: UBIFS file-system description object
857  * @lnum: LEB number
858  * @offs: offset
859  * @sbuf: LEB-sized buffer to use
860  *
861  * This function does a scan of a LEB, but caters for errors that might have
862  * been caused by unclean reboots from which we are attempting to recover
863  * (assume that only the last log LEB can be corrupted by an unclean reboot).
864  *
865  * This function returns %0 on success and a negative error code on failure.
866  */
ubifs_recover_log_leb(struct ubifs_info * c,int lnum,int offs,void * sbuf)867 struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum,
868 					     int offs, void *sbuf)
869 {
870 	struct ubifs_scan_leb *sleb;
871 	int next_lnum;
872 
873 	dbg_rcvry("LEB %d", lnum);
874 	next_lnum = lnum + 1;
875 	if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs)
876 		next_lnum = UBIFS_LOG_LNUM;
877 	if (next_lnum != c->ltail_lnum) {
878 		/*
879 		 * We can only recover at the end of the log, so check that the
880 		 * next log LEB is empty or out of date.
881 		 */
882 		sleb = ubifs_scan(c, next_lnum, 0, sbuf, 0);
883 		if (IS_ERR(sleb))
884 			return sleb;
885 		if (sleb->nodes_cnt) {
886 			struct ubifs_scan_node *snod;
887 			unsigned long long cs_sqnum = c->cs_sqnum;
888 
889 			snod = list_entry(sleb->nodes.next,
890 					  struct ubifs_scan_node, list);
891 			if (cs_sqnum == 0) {
892 				int err;
893 
894 				err = get_cs_sqnum(c, lnum, offs, &cs_sqnum);
895 				if (err) {
896 					ubifs_scan_destroy(sleb);
897 					return ERR_PTR(err);
898 				}
899 			}
900 			if (snod->sqnum > cs_sqnum) {
901 				ubifs_err(c, "unrecoverable log corruption in LEB %d",
902 					  lnum);
903 				ubifs_scan_destroy(sleb);
904 				return ERR_PTR(-EUCLEAN);
905 			}
906 		}
907 		ubifs_scan_destroy(sleb);
908 	}
909 	return ubifs_recover_leb(c, lnum, offs, sbuf, -1);
910 }
911 
912 /**
913  * recover_head - recover a head.
914  * @c: UBIFS file-system description object
915  * @lnum: LEB number of head to recover
916  * @offs: offset of head to recover
917  * @sbuf: LEB-sized buffer to use
918  *
919  * This function ensures that there is no data on the flash at a head location.
920  *
921  * This function returns %0 on success and a negative error code on failure.
922  */
recover_head(struct ubifs_info * c,int lnum,int offs,void * sbuf)923 static int recover_head(struct ubifs_info *c, int lnum, int offs, void *sbuf)
924 {
925 	int len = c->max_write_size, err;
926 
927 	if (offs + len > c->leb_size)
928 		len = c->leb_size - offs;
929 
930 	if (!len)
931 		return 0;
932 
933 	/* Read at the head location and check it is empty flash */
934 	err = ubifs_leb_read(c, lnum, sbuf, offs, len, 1);
935 	if (err || !is_empty(sbuf, len)) {
936 		dbg_rcvry("cleaning head at %d:%d", lnum, offs);
937 		if (offs == 0)
938 			return ubifs_leb_unmap(c, lnum);
939 		err = ubifs_leb_read(c, lnum, sbuf, 0, offs, 1);
940 		if (err)
941 			return err;
942 		return ubifs_leb_change(c, lnum, sbuf, offs);
943 	}
944 
945 	return 0;
946 }
947 
948 /**
949  * ubifs_recover_inl_heads - recover index and LPT heads.
950  * @c: UBIFS file-system description object
951  * @sbuf: LEB-sized buffer to use
952  *
953  * This function ensures that there is no data on the flash at the index and
954  * LPT head locations.
955  *
956  * This deals with the recovery of a half-completed journal commit. UBIFS is
957  * careful never to overwrite the last version of the index or the LPT. Because
958  * the index and LPT are wandering trees, data from a half-completed commit will
959  * not be referenced anywhere in UBIFS. The data will be either in LEBs that are
960  * assumed to be empty and will be unmapped anyway before use, or in the index
961  * and LPT heads.
962  *
963  * This function returns %0 on success and a negative error code on failure.
964  */
ubifs_recover_inl_heads(struct ubifs_info * c,void * sbuf)965 int ubifs_recover_inl_heads(struct ubifs_info *c, void *sbuf)
966 {
967 	int err;
968 
969 	ubifs_assert(c, !c->ro_mount || c->remounting_rw);
970 
971 	dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs);
972 	err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf);
973 	if (err)
974 		return err;
975 
976 	dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs);
977 
978 	return recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf);
979 }
980 
981 /**
982  * clean_an_unclean_leb - read and write a LEB to remove corruption.
983  * @c: UBIFS file-system description object
984  * @ucleb: unclean LEB information
985  * @sbuf: LEB-sized buffer to use
986  *
987  * This function reads a LEB up to a point pre-determined by the mount recovery,
988  * checks the nodes, and writes the result back to the flash, thereby cleaning
989  * off any following corruption, or non-fatal ECC errors.
990  *
991  * This function returns %0 on success and a negative error code on failure.
992  */
clean_an_unclean_leb(struct ubifs_info * c,struct ubifs_unclean_leb * ucleb,void * sbuf)993 static int clean_an_unclean_leb(struct ubifs_info *c,
994 				struct ubifs_unclean_leb *ucleb, void *sbuf)
995 {
996 	int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1;
997 	void *buf = sbuf;
998 
999 	dbg_rcvry("LEB %d len %d", lnum, len);
1000 
1001 	if (len == 0) {
1002 		/* Nothing to read, just unmap it */
1003 		return ubifs_leb_unmap(c, lnum);
1004 	}
1005 
1006 	err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
1007 	if (err && err != -EBADMSG)
1008 		return err;
1009 
1010 	while (len >= 8) {
1011 		int ret;
1012 
1013 		cond_resched();
1014 
1015 		/* Scan quietly until there is an error */
1016 		ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
1017 
1018 		if (ret == SCANNED_A_NODE) {
1019 			/* A valid node, and not a padding node */
1020 			struct ubifs_ch *ch = buf;
1021 			int node_len;
1022 
1023 			node_len = ALIGN(le32_to_cpu(ch->len), 8);
1024 			offs += node_len;
1025 			buf += node_len;
1026 			len -= node_len;
1027 			continue;
1028 		}
1029 
1030 		if (ret > 0) {
1031 			/* Padding bytes or a valid padding node */
1032 			offs += ret;
1033 			buf += ret;
1034 			len -= ret;
1035 			continue;
1036 		}
1037 
1038 		if (ret == SCANNED_EMPTY_SPACE) {
1039 			ubifs_err(c, "unexpected empty space at %d:%d",
1040 				  lnum, offs);
1041 			return -EUCLEAN;
1042 		}
1043 
1044 		if (quiet) {
1045 			/* Redo the last scan but noisily */
1046 			quiet = 0;
1047 			continue;
1048 		}
1049 
1050 		ubifs_scanned_corruption(c, lnum, offs, buf);
1051 		return -EUCLEAN;
1052 	}
1053 
1054 	/* Pad to min_io_size */
1055 	len = ALIGN(ucleb->endpt, c->min_io_size);
1056 	if (len > ucleb->endpt) {
1057 		int pad_len = len - ALIGN(ucleb->endpt, 8);
1058 
1059 		if (pad_len > 0) {
1060 			buf = c->sbuf + len - pad_len;
1061 			ubifs_pad(c, buf, pad_len);
1062 		}
1063 	}
1064 
1065 	/* Write back the LEB atomically */
1066 	err = ubifs_leb_change(c, lnum, sbuf, len);
1067 	if (err)
1068 		return err;
1069 
1070 	dbg_rcvry("cleaned LEB %d", lnum);
1071 
1072 	return 0;
1073 }
1074 
1075 /**
1076  * ubifs_clean_lebs - clean LEBs recovered during read-only mount.
1077  * @c: UBIFS file-system description object
1078  * @sbuf: LEB-sized buffer to use
1079  *
1080  * This function cleans a LEB identified during recovery that needs to be
1081  * written but was not because UBIFS was mounted read-only. This happens when
1082  * remounting to read-write mode.
1083  *
1084  * This function returns %0 on success and a negative error code on failure.
1085  */
ubifs_clean_lebs(struct ubifs_info * c,void * sbuf)1086 int ubifs_clean_lebs(struct ubifs_info *c, void *sbuf)
1087 {
1088 	dbg_rcvry("recovery");
1089 	while (!list_empty(&c->unclean_leb_list)) {
1090 		struct ubifs_unclean_leb *ucleb;
1091 		int err;
1092 
1093 		ucleb = list_entry(c->unclean_leb_list.next,
1094 				   struct ubifs_unclean_leb, list);
1095 		err = clean_an_unclean_leb(c, ucleb, sbuf);
1096 		if (err)
1097 			return err;
1098 		list_del(&ucleb->list);
1099 		kfree(ucleb);
1100 	}
1101 	return 0;
1102 }
1103 
1104 /**
1105  * grab_empty_leb - grab an empty LEB to use as GC LEB and run commit.
1106  * @c: UBIFS file-system description object
1107  *
1108  * This is a helper function for 'ubifs_rcvry_gc_commit()' which grabs an empty
1109  * LEB to be used as GC LEB (@c->gc_lnum), and then runs the commit. Returns
1110  * zero in case of success and a negative error code in case of failure.
1111  */
grab_empty_leb(struct ubifs_info * c)1112 static int grab_empty_leb(struct ubifs_info *c)
1113 {
1114 	int lnum, err;
1115 
1116 	/*
1117 	 * Note, it is very important to first search for an empty LEB and then
1118 	 * run the commit, not vice-versa. The reason is that there might be
1119 	 * only one empty LEB at the moment, the one which has been the
1120 	 * @c->gc_lnum just before the power cut happened. During the regular
1121 	 * UBIFS operation (not now) @c->gc_lnum is marked as "taken", so no
1122 	 * one but GC can grab it. But at this moment this single empty LEB is
1123 	 * not marked as taken, so if we run commit - what happens? Right, the
1124 	 * commit will grab it and write the index there. Remember that the
1125 	 * index always expands as long as there is free space, and it only
1126 	 * starts consolidating when we run out of space.
1127 	 *
1128 	 * IOW, if we run commit now, we might not be able to find a free LEB
1129 	 * after this.
1130 	 */
1131 	lnum = ubifs_find_free_leb_for_idx(c);
1132 	if (lnum < 0) {
1133 		ubifs_err(c, "could not find an empty LEB");
1134 		ubifs_dump_lprops(c);
1135 		ubifs_dump_budg(c, &c->bi);
1136 		return lnum;
1137 	}
1138 
1139 	/* Reset the index flag */
1140 	err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
1141 				  LPROPS_INDEX, 0);
1142 	if (err)
1143 		return err;
1144 
1145 	c->gc_lnum = lnum;
1146 	dbg_rcvry("found empty LEB %d, run commit", lnum);
1147 
1148 	return ubifs_run_commit(c);
1149 }
1150 
1151 /**
1152  * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit.
1153  * @c: UBIFS file-system description object
1154  *
1155  * Out-of-place garbage collection requires always one empty LEB with which to
1156  * start garbage collection. The LEB number is recorded in c->gc_lnum and is
1157  * written to the master node on unmounting. In the case of an unclean unmount
1158  * the value of gc_lnum recorded in the master node is out of date and cannot
1159  * be used. Instead, recovery must allocate an empty LEB for this purpose.
1160  * However, there may not be enough empty space, in which case it must be
1161  * possible to GC the dirtiest LEB into the GC head LEB.
1162  *
1163  * This function also runs the commit which causes the TNC updates from
1164  * size-recovery and orphans to be written to the flash. That is important to
1165  * ensure correct replay order for subsequent mounts.
1166  *
1167  * This function returns %0 on success and a negative error code on failure.
1168  */
ubifs_rcvry_gc_commit(struct ubifs_info * c)1169 int ubifs_rcvry_gc_commit(struct ubifs_info *c)
1170 {
1171 	struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
1172 	struct ubifs_lprops lp;
1173 	int err;
1174 
1175 	dbg_rcvry("GC head LEB %d, offs %d", wbuf->lnum, wbuf->offs);
1176 
1177 	c->gc_lnum = -1;
1178 	if (wbuf->lnum == -1 || wbuf->offs == c->leb_size)
1179 		return grab_empty_leb(c);
1180 
1181 	err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2);
1182 	if (err) {
1183 		if (err != -ENOSPC)
1184 			return err;
1185 
1186 		dbg_rcvry("could not find a dirty LEB");
1187 		return grab_empty_leb(c);
1188 	}
1189 
1190 	ubifs_assert(c, !(lp.flags & LPROPS_INDEX));
1191 	ubifs_assert(c, lp.free + lp.dirty >= wbuf->offs);
1192 
1193 	/*
1194 	 * We run the commit before garbage collection otherwise subsequent
1195 	 * mounts will see the GC and orphan deletion in a different order.
1196 	 */
1197 	dbg_rcvry("committing");
1198 	err = ubifs_run_commit(c);
1199 	if (err)
1200 		return err;
1201 
1202 	dbg_rcvry("GC'ing LEB %d", lp.lnum);
1203 	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
1204 	err = ubifs_garbage_collect_leb(c, &lp);
1205 	if (err >= 0) {
1206 		int err2 = ubifs_wbuf_sync_nolock(wbuf);
1207 
1208 		if (err2)
1209 			err = err2;
1210 	}
1211 	mutex_unlock(&wbuf->io_mutex);
1212 	if (err < 0) {
1213 		ubifs_err(c, "GC failed, error %d", err);
1214 		if (err == -EAGAIN)
1215 			err = -EINVAL;
1216 		return err;
1217 	}
1218 
1219 	ubifs_assert(c, err == LEB_RETAINED);
1220 	if (err != LEB_RETAINED)
1221 		return -EINVAL;
1222 
1223 	err = ubifs_leb_unmap(c, c->gc_lnum);
1224 	if (err)
1225 		return err;
1226 
1227 	dbg_rcvry("allocated LEB %d for GC", lp.lnum);
1228 	return 0;
1229 }
1230 
1231 /**
1232  * struct size_entry - inode size information for recovery.
1233  * @rb: link in the RB-tree of sizes
1234  * @inum: inode number
1235  * @i_size: size on inode
1236  * @d_size: maximum size based on data nodes
1237  * @exists: indicates whether the inode exists
1238  * @inode: inode if pinned in memory awaiting rw mode to fix it
1239  */
1240 struct size_entry {
1241 	struct rb_node rb;
1242 	ino_t inum;
1243 	loff_t i_size;
1244 	loff_t d_size;
1245 	int exists;
1246 	struct inode *inode;
1247 };
1248 
1249 /**
1250  * add_ino - add an entry to the size tree.
1251  * @c: UBIFS file-system description object
1252  * @inum: inode number
1253  * @i_size: size on inode
1254  * @d_size: maximum size based on data nodes
1255  * @exists: indicates whether the inode exists
1256  */
add_ino(struct ubifs_info * c,ino_t inum,loff_t i_size,loff_t d_size,int exists)1257 static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size,
1258 		   loff_t d_size, int exists)
1259 {
1260 	struct rb_node **p = &c->size_tree.rb_node, *parent = NULL;
1261 	struct size_entry *e;
1262 
1263 	while (*p) {
1264 		parent = *p;
1265 		e = rb_entry(parent, struct size_entry, rb);
1266 		if (inum < e->inum)
1267 			p = &(*p)->rb_left;
1268 		else
1269 			p = &(*p)->rb_right;
1270 	}
1271 
1272 	e = kzalloc(sizeof(struct size_entry), GFP_KERNEL);
1273 	if (!e)
1274 		return -ENOMEM;
1275 
1276 	e->inum = inum;
1277 	e->i_size = i_size;
1278 	e->d_size = d_size;
1279 	e->exists = exists;
1280 
1281 	rb_link_node(&e->rb, parent, p);
1282 	rb_insert_color(&e->rb, &c->size_tree);
1283 
1284 	return 0;
1285 }
1286 
1287 /**
1288  * find_ino - find an entry on the size tree.
1289  * @c: UBIFS file-system description object
1290  * @inum: inode number
1291  */
find_ino(struct ubifs_info * c,ino_t inum)1292 static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum)
1293 {
1294 	struct rb_node *p = c->size_tree.rb_node;
1295 	struct size_entry *e;
1296 
1297 	while (p) {
1298 		e = rb_entry(p, struct size_entry, rb);
1299 		if (inum < e->inum)
1300 			p = p->rb_left;
1301 		else if (inum > e->inum)
1302 			p = p->rb_right;
1303 		else
1304 			return e;
1305 	}
1306 	return NULL;
1307 }
1308 
1309 /**
1310  * remove_ino - remove an entry from the size tree.
1311  * @c: UBIFS file-system description object
1312  * @inum: inode number
1313  */
remove_ino(struct ubifs_info * c,ino_t inum)1314 static void remove_ino(struct ubifs_info *c, ino_t inum)
1315 {
1316 	struct size_entry *e = find_ino(c, inum);
1317 
1318 	if (!e)
1319 		return;
1320 	rb_erase(&e->rb, &c->size_tree);
1321 	kfree(e);
1322 }
1323 
1324 /**
1325  * ubifs_destroy_size_tree - free resources related to the size tree.
1326  * @c: UBIFS file-system description object
1327  */
ubifs_destroy_size_tree(struct ubifs_info * c)1328 void ubifs_destroy_size_tree(struct ubifs_info *c)
1329 {
1330 	struct size_entry *e, *n;
1331 
1332 	rbtree_postorder_for_each_entry_safe(e, n, &c->size_tree, rb) {
1333 		iput(e->inode);
1334 		kfree(e);
1335 	}
1336 
1337 	c->size_tree = RB_ROOT;
1338 }
1339 
1340 /**
1341  * ubifs_recover_size_accum - accumulate inode sizes for recovery.
1342  * @c: UBIFS file-system description object
1343  * @key: node key
1344  * @deletion: node is for a deletion
1345  * @new_size: inode size
1346  *
1347  * This function has two purposes:
1348  *     1) to ensure there are no data nodes that fall outside the inode size
1349  *     2) to ensure there are no data nodes for inodes that do not exist
1350  * To accomplish those purposes, a rb-tree is constructed containing an entry
1351  * for each inode number in the journal that has not been deleted, and recording
1352  * the size from the inode node, the maximum size of any data node (also altered
1353  * by truncations) and a flag indicating a inode number for which no inode node
1354  * was present in the journal.
1355  *
1356  * Note that there is still the possibility that there are data nodes that have
1357  * been committed that are beyond the inode size, however the only way to find
1358  * them would be to scan the entire index. Alternatively, some provision could
1359  * be made to record the size of inodes at the start of commit, which would seem
1360  * very cumbersome for a scenario that is quite unlikely and the only negative
1361  * consequence of which is wasted space.
1362  *
1363  * This functions returns %0 on success and a negative error code on failure.
1364  */
ubifs_recover_size_accum(struct ubifs_info * c,union ubifs_key * key,int deletion,loff_t new_size)1365 int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key,
1366 			     int deletion, loff_t new_size)
1367 {
1368 	ino_t inum = key_inum(c, key);
1369 	struct size_entry *e;
1370 	int err;
1371 
1372 	switch (key_type(c, key)) {
1373 	case UBIFS_INO_KEY:
1374 		if (deletion)
1375 			remove_ino(c, inum);
1376 		else {
1377 			e = find_ino(c, inum);
1378 			if (e) {
1379 				e->i_size = new_size;
1380 				e->exists = 1;
1381 			} else {
1382 				err = add_ino(c, inum, new_size, 0, 1);
1383 				if (err)
1384 					return err;
1385 			}
1386 		}
1387 		break;
1388 	case UBIFS_DATA_KEY:
1389 		e = find_ino(c, inum);
1390 		if (e) {
1391 			if (new_size > e->d_size)
1392 				e->d_size = new_size;
1393 		} else {
1394 			err = add_ino(c, inum, 0, new_size, 0);
1395 			if (err)
1396 				return err;
1397 		}
1398 		break;
1399 	case UBIFS_TRUN_KEY:
1400 		e = find_ino(c, inum);
1401 		if (e)
1402 			e->d_size = new_size;
1403 		break;
1404 	}
1405 	return 0;
1406 }
1407 
1408 /**
1409  * fix_size_in_place - fix inode size in place on flash.
1410  * @c: UBIFS file-system description object
1411  * @e: inode size information for recovery
1412  */
fix_size_in_place(struct ubifs_info * c,struct size_entry * e)1413 static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e)
1414 {
1415 	struct ubifs_ino_node *ino = c->sbuf;
1416 	unsigned char *p;
1417 	union ubifs_key key;
1418 	int err, lnum, offs, len;
1419 	loff_t i_size;
1420 	uint32_t crc;
1421 
1422 	/* Locate the inode node LEB number and offset */
1423 	ino_key_init(c, &key, e->inum);
1424 	err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs);
1425 	if (err)
1426 		goto out;
1427 	/*
1428 	 * If the size recorded on the inode node is greater than the size that
1429 	 * was calculated from nodes in the journal then don't change the inode.
1430 	 */
1431 	i_size = le64_to_cpu(ino->size);
1432 	if (i_size >= e->d_size)
1433 		return 0;
1434 	/* Read the LEB */
1435 	err = ubifs_leb_read(c, lnum, c->sbuf, 0, c->leb_size, 1);
1436 	if (err)
1437 		goto out;
1438 	/* Change the size field and recalculate the CRC */
1439 	ino = c->sbuf + offs;
1440 	ino->size = cpu_to_le64(e->d_size);
1441 	len = le32_to_cpu(ino->ch.len);
1442 	crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8);
1443 	ino->ch.crc = cpu_to_le32(crc);
1444 	/* Work out where data in the LEB ends and free space begins */
1445 	p = c->sbuf;
1446 	len = c->leb_size - 1;
1447 	while (p[len] == 0xff)
1448 		len -= 1;
1449 	len = ALIGN(len + 1, c->min_io_size);
1450 	/* Atomically write the fixed LEB back again */
1451 	err = ubifs_leb_change(c, lnum, c->sbuf, len);
1452 	if (err)
1453 		goto out;
1454 	dbg_rcvry("inode %lu at %d:%d size %lld -> %lld",
1455 		  (unsigned long)e->inum, lnum, offs, i_size, e->d_size);
1456 	return 0;
1457 
1458 out:
1459 	ubifs_warn(c, "inode %lu failed to fix size %lld -> %lld error %d",
1460 		   (unsigned long)e->inum, e->i_size, e->d_size, err);
1461 	return err;
1462 }
1463 
1464 /**
1465  * ubifs_recover_size - recover inode size.
1466  * @c: UBIFS file-system description object
1467  *
1468  * This function attempts to fix inode size discrepancies identified by the
1469  * 'ubifs_recover_size_accum()' function.
1470  *
1471  * This functions returns %0 on success and a negative error code on failure.
1472  */
ubifs_recover_size(struct ubifs_info * c)1473 int ubifs_recover_size(struct ubifs_info *c)
1474 {
1475 	struct rb_node *this = rb_first(&c->size_tree);
1476 
1477 	while (this) {
1478 		struct size_entry *e;
1479 		int err;
1480 
1481 		e = rb_entry(this, struct size_entry, rb);
1482 		if (!e->exists) {
1483 			union ubifs_key key;
1484 
1485 			ino_key_init(c, &key, e->inum);
1486 			err = ubifs_tnc_lookup(c, &key, c->sbuf);
1487 			if (err && err != -ENOENT)
1488 				return err;
1489 			if (err == -ENOENT) {
1490 				/* Remove data nodes that have no inode */
1491 				dbg_rcvry("removing ino %lu",
1492 					  (unsigned long)e->inum);
1493 				err = ubifs_tnc_remove_ino(c, e->inum);
1494 				if (err)
1495 					return err;
1496 			} else {
1497 				struct ubifs_ino_node *ino = c->sbuf;
1498 
1499 				e->exists = 1;
1500 				e->i_size = le64_to_cpu(ino->size);
1501 			}
1502 		}
1503 
1504 		if (e->exists && e->i_size < e->d_size) {
1505 			if (c->ro_mount) {
1506 				/* Fix the inode size and pin it in memory */
1507 				struct inode *inode;
1508 				struct ubifs_inode *ui;
1509 
1510 				ubifs_assert(c, !e->inode);
1511 
1512 				inode = ubifs_iget(c->vfs_sb, e->inum);
1513 				if (IS_ERR(inode))
1514 					return PTR_ERR(inode);
1515 
1516 				ui = ubifs_inode(inode);
1517 				if (inode->i_size < e->d_size) {
1518 					dbg_rcvry("ino %lu size %lld -> %lld",
1519 						  (unsigned long)e->inum,
1520 						  inode->i_size, e->d_size);
1521 					inode->i_size = e->d_size;
1522 					ui->ui_size = e->d_size;
1523 					ui->synced_i_size = e->d_size;
1524 					e->inode = inode;
1525 					this = rb_next(this);
1526 					continue;
1527 				}
1528 				iput(inode);
1529 			} else {
1530 				/* Fix the size in place */
1531 				err = fix_size_in_place(c, e);
1532 				if (err)
1533 					return err;
1534 				iput(e->inode);
1535 			}
1536 		}
1537 
1538 		this = rb_next(this);
1539 		rb_erase(&e->rb, &c->size_tree);
1540 		kfree(e);
1541 	}
1542 
1543 	return 0;
1544 }
1545