• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * journal.c --- code for handling the "ext3" journal
3  *
4  * Copyright (C) 2000 Andreas Dilger
5  * Copyright (C) 2000 Theodore Ts'o
6  *
7  * Parts of the code are based on fs/jfs/journal.c by Stephen C. Tweedie
8  * Copyright (C) 1999 Red Hat Software
9  *
10  * This file may be redistributed under the terms of the
11  * GNU General Public License version 2 or at your discretion
12  * any later version.
13  */
14 
15 #include "config.h"
16 #ifdef HAVE_SYS_MOUNT_H
17 #include <sys/param.h>
18 #include <sys/mount.h>
19 #define MNT_FL (MS_MGC_VAL | MS_RDONLY)
20 #endif
21 #ifdef HAVE_SYS_STAT_H
22 #include <sys/stat.h>
23 #endif
24 
25 #define E2FSCK_INCLUDE_INLINE_FUNCS
26 #include "jfs_user.h"
27 #include "problem.h"
28 #include "uuid/uuid.h"
29 
30 static int bh_count = 0;
31 
32 /*
33  * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
34  * This creates a larger static binary, and a smaller binary using
35  * shared libraries.  It's also probably slightly less CPU-efficient,
36  * which is why it's not on by default.  But, it's a good way of
37  * testing the functions in inode_io.c and fileio.c.
38  */
39 #undef USE_INODE_IO
40 
41 /* Checksumming functions */
e2fsck_journal_verify_csum_type(journal_t * j,journal_superblock_t * jsb)42 static int e2fsck_journal_verify_csum_type(journal_t *j,
43 					   journal_superblock_t *jsb)
44 {
45 	if (!jbd2_journal_has_csum_v2or3(j))
46 		return 1;
47 
48 	return jsb->s_checksum_type == JBD2_CRC32C_CHKSUM;
49 }
50 
e2fsck_journal_sb_csum(journal_superblock_t * jsb)51 static __u32 e2fsck_journal_sb_csum(journal_superblock_t *jsb)
52 {
53 	__u32 crc, old_crc;
54 
55 	old_crc = jsb->s_checksum;
56 	jsb->s_checksum = 0;
57 	crc = ext2fs_crc32c_le(~0, (unsigned char *)jsb,
58 			       sizeof(journal_superblock_t));
59 	jsb->s_checksum = old_crc;
60 
61 	return crc;
62 }
63 
e2fsck_journal_sb_csum_verify(journal_t * j,journal_superblock_t * jsb)64 static int e2fsck_journal_sb_csum_verify(journal_t *j,
65 					 journal_superblock_t *jsb)
66 {
67 	__u32 provided, calculated;
68 
69 	if (!jbd2_journal_has_csum_v2or3(j))
70 		return 1;
71 
72 	provided = ext2fs_be32_to_cpu(jsb->s_checksum);
73 	calculated = e2fsck_journal_sb_csum(jsb);
74 
75 	return provided == calculated;
76 }
77 
e2fsck_journal_sb_csum_set(journal_t * j,journal_superblock_t * jsb)78 static errcode_t e2fsck_journal_sb_csum_set(journal_t *j,
79 					    journal_superblock_t *jsb)
80 {
81 	__u32 crc;
82 
83 	if (!jbd2_journal_has_csum_v2or3(j))
84 		return 0;
85 
86 	crc = e2fsck_journal_sb_csum(jsb);
87 	jsb->s_checksum = ext2fs_cpu_to_be32(crc);
88 	return 0;
89 }
90 
91 /* Kernel compatibility functions for handling the journal.  These allow us
92  * to use the recovery.c file virtually unchanged from the kernel, so we
93  * don't have to do much to keep kernel and user recovery in sync.
94  */
jbd2_journal_bmap(journal_t * journal,unsigned long block,unsigned long long * phys)95 int jbd2_journal_bmap(journal_t *journal, unsigned long block,
96 		      unsigned long long *phys)
97 {
98 #ifdef USE_INODE_IO
99 	*phys = block;
100 	return 0;
101 #else
102 	struct inode 	*inode = journal->j_inode;
103 	errcode_t	retval;
104 	blk64_t		pblk;
105 
106 	if (!inode) {
107 		*phys = block;
108 		return 0;
109 	}
110 
111 	retval= ext2fs_bmap2(inode->i_ctx->fs, inode->i_ino,
112 			     &inode->i_ext2, NULL, 0, (blk64_t) block,
113 			     0, &pblk);
114 	*phys = pblk;
115 	return -1 * ((int) retval);
116 #endif
117 }
118 
getblk(kdev_t kdev,unsigned long long blocknr,int blocksize)119 struct buffer_head *getblk(kdev_t kdev, unsigned long long blocknr,
120 			   int blocksize)
121 {
122 	struct buffer_head *bh;
123 	int bufsize = sizeof(*bh) + kdev->k_ctx->fs->blocksize -
124 		sizeof(bh->b_data);
125 
126 	bh = e2fsck_allocate_memory(kdev->k_ctx, bufsize, "block buffer");
127 	if (!bh)
128 		return NULL;
129 
130 	if (journal_enable_debug >= 3)
131 		bh_count++;
132 	jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
133 		  blocknr, blocksize, bh_count);
134 
135 	bh->b_ctx = kdev->k_ctx;
136 	if (kdev->k_dev == K_DEV_FS)
137 		bh->b_io = kdev->k_ctx->fs->io;
138 	else
139 		bh->b_io = kdev->k_ctx->journal_io;
140 	bh->b_size = blocksize;
141 	bh->b_blocknr = blocknr;
142 
143 	return bh;
144 }
145 
sync_blockdev(kdev_t kdev)146 int sync_blockdev(kdev_t kdev)
147 {
148 	io_channel	io;
149 
150 	if (kdev->k_dev == K_DEV_FS)
151 		io = kdev->k_ctx->fs->io;
152 	else
153 		io = kdev->k_ctx->journal_io;
154 
155 	return io_channel_flush(io) ? -EIO : 0;
156 }
157 
ll_rw_block(int rw,int op_flags EXT2FS_ATTR ((unused)),int nr,struct buffer_head * bhp[])158 void ll_rw_block(int rw, int op_flags EXT2FS_ATTR((unused)), int nr,
159 		 struct buffer_head *bhp[])
160 {
161 	errcode_t retval;
162 	struct buffer_head *bh;
163 
164 	for (; nr > 0; --nr) {
165 		bh = *bhp++;
166 		if (rw == REQ_OP_READ && !bh->b_uptodate) {
167 			jfs_debug(3, "reading block %llu/%p\n",
168 				  bh->b_blocknr, (void *) bh);
169 			retval = io_channel_read_blk64(bh->b_io,
170 						     bh->b_blocknr,
171 						     1, bh->b_data);
172 			if (retval) {
173 				com_err(bh->b_ctx->device_name, retval,
174 					"while reading block %llu\n",
175 					bh->b_blocknr);
176 				bh->b_err = (int) retval;
177 				continue;
178 			}
179 			bh->b_uptodate = 1;
180 		} else if (rw == REQ_OP_WRITE && bh->b_dirty) {
181 			jfs_debug(3, "writing block %llu/%p\n",
182 				  bh->b_blocknr,
183 				  (void *) bh);
184 			retval = io_channel_write_blk64(bh->b_io,
185 						      bh->b_blocknr,
186 						      1, bh->b_data);
187 			if (retval) {
188 				com_err(bh->b_ctx->device_name, retval,
189 					"while writing block %llu\n",
190 					bh->b_blocknr);
191 				bh->b_err = (int) retval;
192 				continue;
193 			}
194 			bh->b_dirty = 0;
195 			bh->b_uptodate = 1;
196 		} else {
197 			jfs_debug(3, "no-op %s for block %llu\n",
198 				  rw == REQ_OP_READ ? "read" : "write",
199 				  bh->b_blocknr);
200 		}
201 	}
202 }
203 
mark_buffer_dirty(struct buffer_head * bh)204 void mark_buffer_dirty(struct buffer_head *bh)
205 {
206 	bh->b_dirty = 1;
207 }
208 
mark_buffer_clean(struct buffer_head * bh)209 static void mark_buffer_clean(struct buffer_head * bh)
210 {
211 	bh->b_dirty = 0;
212 }
213 
brelse(struct buffer_head * bh)214 void brelse(struct buffer_head *bh)
215 {
216 	if (bh->b_dirty)
217 		ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
218 	jfs_debug(3, "freeing block %llu/%p (total %d)\n",
219 		  bh->b_blocknr, (void *) bh, --bh_count);
220 	ext2fs_free_mem(&bh);
221 }
222 
buffer_uptodate(struct buffer_head * bh)223 int buffer_uptodate(struct buffer_head *bh)
224 {
225 	return bh->b_uptodate;
226 }
227 
mark_buffer_uptodate(struct buffer_head * bh,int val)228 void mark_buffer_uptodate(struct buffer_head *bh, int val)
229 {
230 	bh->b_uptodate = val;
231 }
232 
wait_on_buffer(struct buffer_head * bh)233 void wait_on_buffer(struct buffer_head *bh)
234 {
235 	if (!bh->b_uptodate)
236 		ll_rw_block(REQ_OP_READ, 0, 1, &bh);
237 }
238 
239 
e2fsck_clear_recover(e2fsck_t ctx,int error)240 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
241 {
242 	ext2fs_clear_feature_journal_needs_recovery(ctx->fs->super);
243 
244 	/* if we had an error doing journal recovery, we need a full fsck */
245 	if (error)
246 		ctx->fs->super->s_state &= ~EXT2_VALID_FS;
247 	ext2fs_mark_super_dirty(ctx->fs);
248 }
249 
250 /*
251  * This is a helper function to check the validity of the journal.
252  */
253 struct process_block_struct {
254 	e2_blkcnt_t	last_block;
255 };
256 
process_journal_block(ext2_filsys fs,blk64_t * block_nr,e2_blkcnt_t blockcnt,blk64_t ref_block EXT2FS_ATTR ((unused)),int ref_offset EXT2FS_ATTR ((unused)),void * priv_data)257 static int process_journal_block(ext2_filsys fs,
258 				 blk64_t	*block_nr,
259 				 e2_blkcnt_t blockcnt,
260 				 blk64_t ref_block EXT2FS_ATTR((unused)),
261 				 int ref_offset EXT2FS_ATTR((unused)),
262 				 void *priv_data)
263 {
264 	struct process_block_struct *p;
265 	blk64_t	blk = *block_nr;
266 
267 	p = (struct process_block_struct *) priv_data;
268 
269 	if (!blk || blk < fs->super->s_first_data_block ||
270 	    blk >= ext2fs_blocks_count(fs->super))
271 		return BLOCK_ABORT;
272 
273 	if (blockcnt >= 0)
274 		p->last_block = blockcnt;
275 	return 0;
276 }
277 
ext4_fc_replay_scan(journal_t * j,struct buffer_head * bh,int off,tid_t expected_tid)278 static int ext4_fc_replay_scan(journal_t *j, struct buffer_head *bh,
279 				int off, tid_t expected_tid)
280 {
281 	e2fsck_t ctx = j->j_fs_dev->k_ctx;
282 	struct e2fsck_fc_replay_state *state;
283 	int ret = JBD2_FC_REPLAY_CONTINUE;
284 	struct ext4_fc_add_range *ext;
285 	struct ext4_fc_tl *tl;
286 	struct ext4_fc_tail *tail;
287 	__u8 *start, *end;
288 	struct ext4_fc_head *head;
289 	struct ext2fs_extent ext2fs_ex = {0};
290 
291 	state = &ctx->fc_replay_state;
292 
293 	start = (__u8 *)bh->b_data;
294 	end = (__u8 *)bh->b_data + j->j_blocksize - 1;
295 
296 	jbd_debug(1, "Scan phase starting, expected %d", expected_tid);
297 	if (state->fc_replay_expected_off == 0) {
298 		memset(state, 0, sizeof(*state));
299 		/* Check if we can stop early */
300 		if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
301 			!= EXT4_FC_TAG_HEAD) {
302 			jbd_debug(1, "Ending early!, not a head tag");
303 			return 0;
304 		}
305 	}
306 
307 	if (off != state->fc_replay_expected_off) {
308 		ret = -EFSCORRUPTED;
309 		goto out_err;
310 	}
311 
312 	state->fc_replay_expected_off++;
313 	fc_for_each_tl(start, end, tl) {
314 		jbd_debug(3, "Scan phase, tag:%s, blk %lld\n",
315 			  tag2str(le16_to_cpu(tl->fc_tag)), bh->b_blocknr);
316 		switch (le16_to_cpu(tl->fc_tag)) {
317 		case EXT4_FC_TAG_ADD_RANGE:
318 			ext = (struct ext4_fc_add_range *)ext4_fc_tag_val(tl);
319 			ret = ext2fs_decode_extent(&ext2fs_ex, (void *)&ext->fc_ex,
320 						   sizeof(ext->fc_ex));
321 			if (ret)
322 				ret = JBD2_FC_REPLAY_STOP;
323 			else
324 				ret = JBD2_FC_REPLAY_CONTINUE;
325 			/* fallthrough */
326 		case EXT4_FC_TAG_DEL_RANGE:
327 		case EXT4_FC_TAG_LINK:
328 		case EXT4_FC_TAG_UNLINK:
329 		case EXT4_FC_TAG_CREAT:
330 		case EXT4_FC_TAG_INODE:
331 		case EXT4_FC_TAG_PAD:
332 			state->fc_cur_tag++;
333 			state->fc_crc = jbd2_chksum(j, state->fc_crc, tl,
334 					sizeof(*tl) + ext4_fc_tag_len(tl));
335 			break;
336 		case EXT4_FC_TAG_TAIL:
337 			state->fc_cur_tag++;
338 			tail = (struct ext4_fc_tail *)ext4_fc_tag_val(tl);
339 			state->fc_crc = jbd2_chksum(j, state->fc_crc, tl,
340 						sizeof(*tl) +
341 						offsetof(struct ext4_fc_tail,
342 						fc_crc));
343 			jbd_debug(1, "tail tid %d, expected %d\n",
344 					le32_to_cpu(tail->fc_tid),
345 					expected_tid);
346 			if (le32_to_cpu(tail->fc_tid) == expected_tid &&
347 				le32_to_cpu(tail->fc_crc) == state->fc_crc) {
348 				state->fc_replay_num_tags = state->fc_cur_tag;
349 			} else {
350 				ret = state->fc_replay_num_tags ?
351 					JBD2_FC_REPLAY_STOP : -EFSBADCRC;
352 			}
353 			state->fc_crc = 0;
354 			break;
355 		case EXT4_FC_TAG_HEAD:
356 			head = (struct ext4_fc_head *)ext4_fc_tag_val(tl);
357 			if (le32_to_cpu(head->fc_features) &
358 				~EXT4_FC_SUPPORTED_FEATURES) {
359 				ret = -EOPNOTSUPP;
360 				break;
361 			}
362 			if (le32_to_cpu(head->fc_tid) != expected_tid) {
363 				ret = -EINVAL;
364 				break;
365 			}
366 			state->fc_cur_tag++;
367 			state->fc_crc = jbd2_chksum(j, state->fc_crc, tl,
368 					sizeof(*tl) + ext4_fc_tag_len(tl));
369 			break;
370 		default:
371 			ret = state->fc_replay_num_tags ?
372 				JBD2_FC_REPLAY_STOP : -ECANCELED;
373 		}
374 		if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
375 			break;
376 	}
377 
378 out_err:
379 	return ret;
380 }
381 
__errcode_to_errno(errcode_t err,const char * func,int line)382 static int __errcode_to_errno(errcode_t err, const char *func, int line)
383 {
384 	if (err == 0)
385 		return 0;
386 	fprintf(stderr, "Error \"%s\" encountered in function %s at line %d\n",
387 		error_message(err), func, line);
388 	if (err <= 256)
389 		return -err;
390 	return -EFAULT;
391 }
392 
393 #define errcode_to_errno(err)	__errcode_to_errno(err, __func__, __LINE__)
394 
395 #define ex_end(__ex) ((__ex)->e_lblk + (__ex)->e_len - 1)
396 #define ex_pend(__ex) ((__ex)->e_pblk + (__ex)->e_len - 1)
397 
make_room(struct extent_list * list,int i)398 static int make_room(struct extent_list *list, int i)
399 {
400 	int ret;
401 
402 	if (list->count == list->size) {
403 		unsigned int new_size = (list->size + 341) *
404 					sizeof(struct ext2fs_extent);
405 		ret = errcode_to_errno(ext2fs_resize_mem(0, new_size, &list->extents));
406 		if (ret)
407 			return ret;
408 		list->size += 341;
409 	}
410 
411 	memmove(&list->extents[i + 1], &list->extents[i],
412 			sizeof(list->extents[0]) * (list->count - i));
413 	list->count++;
414 	return 0;
415 }
416 
ex_compar(const void * arg1,const void * arg2)417 static int ex_compar(const void *arg1, const void *arg2)
418 {
419 	struct ext2fs_extent *ex1 = (struct ext2fs_extent *)arg1;
420 	struct ext2fs_extent *ex2 = (struct ext2fs_extent *)arg2;
421 
422 	if (ex1->e_lblk < ex2->e_lblk)
423 		return -1;
424 	if (ex1->e_lblk > ex2->e_lblk)
425 		return 1;
426 	return ex1->e_len - ex2->e_len;
427 }
428 
ex_len_compar(const void * arg1,const void * arg2)429 static int ex_len_compar(const void *arg1, const void *arg2)
430 {
431 	struct ext2fs_extent *ex1 = (struct ext2fs_extent *)arg1;
432 	struct ext2fs_extent *ex2 = (struct ext2fs_extent *)arg2;
433 
434 	if (ex1->e_len < ex2->e_len)
435 		return 1;
436 
437 	if (ex1->e_lblk > ex2->e_lblk)
438 		return -1;
439 
440 	return 0;
441 }
442 
ex_sort_and_merge(struct extent_list * list)443 static void ex_sort_and_merge(struct extent_list *list)
444 {
445 	int i, j;
446 
447 	if (list->count < 2)
448 		return;
449 
450 	/*
451 	 * Reverse sort by length, that way we strip off all the 0 length
452 	 * extents
453 	 */
454 	qsort(list->extents, list->count, sizeof(struct ext2fs_extent),
455 		ex_len_compar);
456 
457 	for (i = 0; i < list->count; i++) {
458 		if (list->extents[i].e_len == 0) {
459 			list->count = i;
460 			break;
461 		}
462 	}
463 
464 	/* Now sort by logical offset */
465 	qsort(list->extents, list->count, sizeof(list->extents[0]),
466 		ex_compar);
467 
468 	/* Merge adjacent extents if they are logically and physically contiguous */
469 	i = 0;
470 	while (i < list->count - 1) {
471 		if (ex_end(&list->extents[i]) + 1 != list->extents[i + 1].e_lblk ||
472 			ex_pend(&list->extents[i]) + 1 != list->extents[i + 1].e_pblk ||
473 			(list->extents[i].e_flags & EXT2_EXTENT_FLAGS_UNINIT) !=
474 				(list->extents[i + 1].e_flags & EXT2_EXTENT_FLAGS_UNINIT)) {
475 			i++;
476 			continue;
477 		}
478 
479 		list->extents[i].e_len += list->extents[i + 1].e_len;
480 		for (j = i + 1; j < list->count - 1; j++)
481 			list->extents[j] = list->extents[j + 1];
482 		list->count--;
483 	}
484 }
485 
486 /* must free blocks that are released */
ext4_modify_extent_list(e2fsck_t ctx,struct extent_list * list,struct ext2fs_extent * ex,int del)487 static int ext4_modify_extent_list(e2fsck_t ctx, struct extent_list *list,
488 					struct ext2fs_extent *ex, int del)
489 {
490 	int ret;
491 	int i, offset;
492 	struct ext2fs_extent add_ex = *ex;
493 
494 	/* First let's create a hole from ex->e_lblk of length ex->e_len */
495 	for (i = 0; i < list->count; i++) {
496 		if (ex_end(&list->extents[i]) < add_ex.e_lblk)
497 			continue;
498 
499 		/* Case 1: No overlap */
500 		if (list->extents[i].e_lblk > ex_end(&add_ex))
501 			break;
502 		/*
503 		 * Unmark all the blocks in bb now. All the blocks get marked
504 		 * before we exit this function.
505 		 */
506 		ext2fs_unmark_block_bitmap_range2(ctx->fs->block_map,
507 			list->extents[i].e_pblk, list->extents[i].e_len);
508 		/* Case 2: Split */
509 		if (list->extents[i].e_lblk < add_ex.e_lblk &&
510 			ex_end(&list->extents[i]) > ex_end(&add_ex)) {
511 			ret = make_room(list, i + 1);
512 			if (ret)
513 				return ret;
514 			list->extents[i + 1] = list->extents[i];
515 			offset = ex_end(&add_ex) + 1 - list->extents[i].e_lblk;
516 			list->extents[i + 1].e_lblk += offset;
517 			list->extents[i + 1].e_pblk += offset;
518 			list->extents[i + 1].e_len -= offset;
519 			list->extents[i].e_len =
520 				add_ex.e_lblk - list->extents[i].e_lblk;
521 			break;
522 		}
523 
524 		/* Case 3: Exact overlap */
525 		if (add_ex.e_lblk <= list->extents[i].e_lblk  &&
526 			ex_end(&list->extents[i]) <= ex_end(&add_ex)) {
527 
528 			list->extents[i].e_len = 0;
529 			continue;
530 		}
531 
532 		/* Case 4: Partial overlap */
533 		if (ex_end(&list->extents[i]) > ex_end(&add_ex)) {
534 			offset = ex_end(&add_ex) + 1 - list->extents[i].e_lblk;
535 			list->extents[i].e_lblk += offset;
536 			list->extents[i].e_pblk += offset;
537 			list->extents[i].e_len -= offset;
538 			break;
539 		}
540 
541 		if (ex_end(&add_ex) >= ex_end(&list->extents[i]))
542 			list->extents[i].e_len =
543 				add_ex.e_lblk > list->extents[i].e_lblk ?
544 				add_ex.e_lblk - list->extents[i].e_lblk : 0;
545 	}
546 
547 	if (add_ex.e_len && !del) {
548 		make_room(list, list->count);
549 		list->extents[list->count - 1] = add_ex;
550 	}
551 
552 	ex_sort_and_merge(list);
553 
554 	/* Mark all occupied blocks allocated */
555 	for (i = 0; i < list->count; i++)
556 		ext2fs_mark_block_bitmap_range2(ctx->fs->block_map,
557 			list->extents[i].e_pblk, list->extents[i].e_len);
558 	ext2fs_mark_bb_dirty(ctx->fs);
559 
560 	return 0;
561 }
562 
ext4_add_extent_to_list(e2fsck_t ctx,struct extent_list * list,struct ext2fs_extent * ex)563 static int ext4_add_extent_to_list(e2fsck_t ctx, struct extent_list *list,
564 					struct ext2fs_extent *ex)
565 {
566 	return ext4_modify_extent_list(ctx, list, ex, 0 /* add */);
567 }
568 
ext4_del_extent_from_list(e2fsck_t ctx,struct extent_list * list,struct ext2fs_extent * ex)569 static int ext4_del_extent_from_list(e2fsck_t ctx, struct extent_list *list,
570 					struct ext2fs_extent *ex)
571 {
572 	return ext4_modify_extent_list(ctx, list, ex, 1 /* delete */);
573 }
574 
ext4_fc_read_extents(e2fsck_t ctx,int ino)575 static int ext4_fc_read_extents(e2fsck_t ctx, int ino)
576 {
577 	struct extent_list *extent_list = &ctx->fc_replay_state.fc_extent_list;
578 
579 	if (extent_list->ino == ino)
580 		return 0;
581 
582 	extent_list->ino = ino;
583 	return errcode_to_errno(e2fsck_read_extents(ctx, extent_list));
584 }
585 
586 /*
587  * Flush extents in replay state on disk. @ino is the inode that is going
588  * to be processed next. So, we hold back flushing of the extent list
589  * if the next inode that's going to be processed is same as the one with
590  * cached extents in our replay state. That allows us to gather multiple extents
591  * for the inode so that we can flush all of them at once and it also saves us
592  * from continuously growing and shrinking the extent tree.
593  */
ext4_fc_flush_extents(e2fsck_t ctx,int ino)594 static void ext4_fc_flush_extents(e2fsck_t ctx, int ino)
595 {
596 	struct extent_list *extent_list = &ctx->fc_replay_state.fc_extent_list;
597 
598 	if (extent_list->ino == ino || extent_list->ino == 0)
599 		return;
600 	e2fsck_rewrite_extent_tree(ctx, extent_list);
601 	ext2fs_free_mem(&extent_list->extents);
602 	memset(extent_list, 0, sizeof(*extent_list));
603 }
604 
605 /* Helper struct for dentry replay routines */
606 struct dentry_info_args {
607 	int parent_ino, dname_len, ino, inode_len;
608 	char *dname;
609 };
610 
tl_to_darg(struct dentry_info_args * darg,struct ext4_fc_tl * tl)611 static inline void tl_to_darg(struct dentry_info_args *darg,
612 				struct  ext4_fc_tl *tl)
613 {
614 	struct ext4_fc_dentry_info *fcd;
615 	int tag = le16_to_cpu(tl->fc_tag);
616 
617 	fcd = (struct ext4_fc_dentry_info *)ext4_fc_tag_val(tl);
618 
619 	darg->parent_ino = le32_to_cpu(fcd->fc_parent_ino);
620 	darg->ino = le32_to_cpu(fcd->fc_ino);
621 	darg->dname = (char *) fcd->fc_dname;
622 	darg->dname_len = ext4_fc_tag_len(tl) -
623 			sizeof(struct ext4_fc_dentry_info);
624 	darg->dname = malloc(darg->dname_len + 1);
625 	memcpy(darg->dname, fcd->fc_dname, darg->dname_len);
626 	darg->dname[darg->dname_len] = 0;
627 	jbd_debug(1, "%s: %s, ino %d, parent %d\n",
628 		tag == EXT4_FC_TAG_CREAT ? "create" :
629 		(tag == EXT4_FC_TAG_LINK ? "link" :
630 		(tag == EXT4_FC_TAG_UNLINK ? "unlink" : "error")),
631 		darg->dname, darg->ino, darg->parent_ino);
632 }
633 
ext4_fc_handle_unlink(e2fsck_t ctx,struct ext4_fc_tl * tl)634 static int ext4_fc_handle_unlink(e2fsck_t ctx, struct ext4_fc_tl *tl)
635 {
636 	struct dentry_info_args darg;
637 	int ret;
638 
639 	tl_to_darg(&darg, tl);
640 	ext4_fc_flush_extents(ctx, darg.ino);
641 	ret = errcode_to_errno(
642 		       ext2fs_unlink(ctx->fs, darg.parent_ino,
643 				     darg.dname, darg.ino, 0));
644 	/* It's okay if the above call fails */
645 	free(darg.dname);
646 	return ret;
647 }
648 
ext4_fc_handle_link_and_create(e2fsck_t ctx,struct ext4_fc_tl * tl)649 static int ext4_fc_handle_link_and_create(e2fsck_t ctx, struct ext4_fc_tl *tl)
650 {
651 	struct dentry_info_args darg;
652 	ext2_filsys fs = ctx->fs;
653 	struct ext2_inode_large inode_large;
654 	int ret, filetype, mode;
655 
656 	tl_to_darg(&darg, tl);
657 	ext4_fc_flush_extents(ctx, 0);
658 	ret = errcode_to_errno(ext2fs_read_inode(fs, darg.ino,
659 						 (struct ext2_inode *)&inode_large));
660 	if (ret)
661 		goto out;
662 
663 	mode = inode_large.i_mode;
664 
665 	if (LINUX_S_ISREG(mode))
666 		filetype = EXT2_FT_REG_FILE;
667 	else if (LINUX_S_ISDIR(mode))
668 		filetype = EXT2_FT_DIR;
669 	else if (LINUX_S_ISCHR(mode))
670 		filetype = EXT2_FT_CHRDEV;
671 	else if (LINUX_S_ISBLK(mode))
672 		filetype = EXT2_FT_BLKDEV;
673 	else if (LINUX_S_ISLNK(mode))
674 		return EXT2_FT_SYMLINK;
675 	else if (LINUX_S_ISFIFO(mode))
676 		filetype = EXT2_FT_FIFO;
677 	else if (LINUX_S_ISSOCK(mode))
678 		filetype = EXT2_FT_SOCK;
679 	else {
680 		ret = -EINVAL;
681 		goto out;
682 	}
683 
684 	/*
685 	 * Forcefully unlink if the same name is present and ignore the error
686 	 * if any, since this dirent might not exist
687 	 */
688 	ext2fs_unlink(fs, darg.parent_ino, darg.dname, darg.ino,
689 			EXT2FS_UNLINK_FORCE);
690 
691 	ret = errcode_to_errno(
692 		       ext2fs_link(fs, darg.parent_ino, darg.dname, darg.ino,
693 				   filetype));
694 out:
695 	free(darg.dname);
696 	return ret;
697 
698 }
699 
700 /* This function fixes the i_blocks field in the replayed indoe */
ext4_fc_replay_fixup_iblocks(struct ext2_inode_large * ondisk_inode,struct ext2_inode_large * fc_inode)701 static void ext4_fc_replay_fixup_iblocks(struct ext2_inode_large *ondisk_inode,
702 	struct ext2_inode_large *fc_inode)
703 {
704 	if (ondisk_inode->i_flags & EXT4_EXTENTS_FL) {
705 		struct ext3_extent_header *eh;
706 
707 		eh = (struct ext3_extent_header *)(&ondisk_inode->i_block[0]);
708 		if (le16_to_cpu(eh->eh_magic) != EXT3_EXT_MAGIC) {
709 			memset(eh, 0, sizeof(*eh));
710 			eh->eh_magic = cpu_to_le16(EXT3_EXT_MAGIC);
711 			eh->eh_max = cpu_to_le16(
712 				(sizeof(ondisk_inode->i_block) -
713 					sizeof(struct ext3_extent_header)) /
714 				sizeof(struct ext3_extent));
715 		}
716 	} else if (ondisk_inode->i_flags & EXT4_INLINE_DATA_FL) {
717 		memcpy(ondisk_inode->i_block, fc_inode->i_block,
718 			sizeof(fc_inode->i_block));
719 	}
720 }
721 
ext4_fc_handle_inode(e2fsck_t ctx,struct ext4_fc_tl * tl)722 static int ext4_fc_handle_inode(e2fsck_t ctx, struct ext4_fc_tl *tl)
723 {
724 	int ino, inode_len = EXT2_GOOD_OLD_INODE_SIZE;
725 	struct ext2_inode_large *inode = NULL, *fc_inode = NULL;
726 	struct ext4_fc_inode *fc_inode_val;
727 	errcode_t err;
728 	blk64_t blks;
729 
730 	fc_inode_val = (struct ext4_fc_inode *)ext4_fc_tag_val(tl);
731 	ino = le32_to_cpu(fc_inode_val->fc_ino);
732 
733 	if (EXT2_INODE_SIZE(ctx->fs->super) > EXT2_GOOD_OLD_INODE_SIZE)
734 		inode_len += ext2fs_le16_to_cpu(
735 			((struct ext2_inode_large *)fc_inode_val->fc_raw_inode)
736 				->i_extra_isize);
737 	err = ext2fs_get_mem(inode_len, &inode);
738 	if (err)
739 		goto out;
740 	err = ext2fs_get_mem(inode_len, &fc_inode);
741 	if (err)
742 		goto out;
743 	ext4_fc_flush_extents(ctx, ino);
744 
745 	err = ext2fs_read_inode_full(ctx->fs, ino, (struct ext2_inode *)inode,
746 					inode_len);
747 	if (err)
748 		goto out;
749 #ifdef WORDS_BIGENDIAN
750 	ext2fs_swap_inode_full(ctx->fs, fc_inode,
751 			       (struct ext2_inode_large *)fc_inode_val->fc_raw_inode,
752 			       0, sizeof(*inode));
753 #else
754 	memcpy(fc_inode, fc_inode_val->fc_raw_inode, inode_len);
755 #endif
756 	memcpy(inode, fc_inode, offsetof(struct ext2_inode_large, i_block));
757 	memcpy(&inode->i_generation, &fc_inode->i_generation,
758 		inode_len - offsetof(struct ext2_inode_large, i_generation));
759 	ext4_fc_replay_fixup_iblocks(inode, fc_inode);
760 	err = ext2fs_count_blocks(ctx->fs, ino, EXT2_INODE(inode), &blks);
761 	if (err)
762 		goto out;
763 	ext2fs_iblk_set(ctx->fs, EXT2_INODE(inode), blks);
764 	ext2fs_inode_csum_set(ctx->fs, ino, inode);
765 
766 	err = ext2fs_write_inode_full(ctx->fs, ino, (struct ext2_inode *)inode,
767 					inode_len);
768 	if (err)
769 		goto out;
770 	if (inode->i_links_count)
771 		ext2fs_mark_inode_bitmap2(ctx->fs->inode_map, ino);
772 	else
773 		ext2fs_unmark_inode_bitmap2(ctx->fs->inode_map, ino);
774 	ext2fs_mark_ib_dirty(ctx->fs);
775 
776 out:
777 	ext2fs_free_mem(&inode);
778 	ext2fs_free_mem(&fc_inode);
779 	return errcode_to_errno(err);
780 }
781 
782 /*
783  * Handle add extent replay tag.
784  */
ext4_fc_handle_add_extent(e2fsck_t ctx,struct ext4_fc_tl * tl)785 static int ext4_fc_handle_add_extent(e2fsck_t ctx, struct ext4_fc_tl *tl)
786 {
787 	struct ext2fs_extent extent;
788 	struct ext4_fc_add_range *add_range;
789 	int ret = 0, ino;
790 
791 	add_range = (struct ext4_fc_add_range *)ext4_fc_tag_val(tl);
792 	ino = le32_to_cpu(add_range->fc_ino);
793 	ext4_fc_flush_extents(ctx, ino);
794 
795 	ret = ext4_fc_read_extents(ctx, ino);
796 	if (ret)
797 		return ret;
798 	memset(&extent, 0, sizeof(extent));
799 	ret = errcode_to_errno(ext2fs_decode_extent(
800 			&extent, (void *)(add_range->fc_ex),
801 			sizeof(add_range->fc_ex)));
802 	if (ret)
803 		return ret;
804 	return ext4_add_extent_to_list(ctx,
805 		&ctx->fc_replay_state.fc_extent_list, &extent);
806 }
807 
808 /*
809  * Handle delete logical range replay tag.
810  */
ext4_fc_handle_del_range(e2fsck_t ctx,struct ext4_fc_tl * tl)811 static int ext4_fc_handle_del_range(e2fsck_t ctx, struct ext4_fc_tl *tl)
812 {
813 	struct ext2fs_extent extent;
814 	struct ext4_fc_del_range *del_range;
815 	int ret, ino;
816 
817 	del_range = (struct ext4_fc_del_range *)ext4_fc_tag_val(tl);
818 	ino = le32_to_cpu(del_range->fc_ino);
819 	ext4_fc_flush_extents(ctx, ino);
820 
821 	memset(&extent, 0, sizeof(extent));
822 	extent.e_lblk = ext2fs_le32_to_cpu(del_range->fc_lblk);
823 	extent.e_len = ext2fs_le32_to_cpu(del_range->fc_len);
824 	ret = ext4_fc_read_extents(ctx, ino);
825 	if (ret)
826 		return ret;
827 	return ext4_del_extent_from_list(ctx,
828 		&ctx->fc_replay_state.fc_extent_list, &extent);
829 }
830 
831 /*
832  * Main recovery path entry point. This function returns JBD2_FC_REPLAY_CONTINUE
833  * to indicate that it is expecting more fast commit blocks. It returns
834  * JBD2_FC_REPLAY_STOP to indicate that replay is done.
835  */
ext4_fc_replay(journal_t * journal,struct buffer_head * bh,enum passtype pass,int off,tid_t expected_tid)836 static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
837 				enum passtype pass, int off, tid_t expected_tid)
838 {
839 	e2fsck_t ctx = journal->j_fs_dev->k_ctx;
840 	struct e2fsck_fc_replay_state *state = &ctx->fc_replay_state;
841 	int ret = JBD2_FC_REPLAY_CONTINUE;
842 	struct ext4_fc_tl *tl;
843 	__u8 *start, *end;
844 
845 	if (pass == PASS_SCAN) {
846 		state->fc_current_pass = PASS_SCAN;
847 		return ext4_fc_replay_scan(journal, bh, off, expected_tid);
848 	}
849 
850 	if (state->fc_replay_num_tags == 0)
851 		goto replay_done;
852 
853 	if (state->fc_current_pass != pass) {
854 		/* Starting replay phase */
855 		state->fc_current_pass = pass;
856 		/* We will reset checksums */
857 		ctx->fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
858 		ret = errcode_to_errno(ext2fs_read_bitmaps(ctx->fs));
859 		if (ret) {
860 			jbd_debug(1, "Error %d while reading bitmaps\n", ret);
861 			return ret;
862 		}
863 		state->fc_super_state = ctx->fs->super->s_state;
864 		/*
865 		 * Mark the file system to indicate it contains errors. That's
866 		 * because the updates performed by fast commit replay code are
867 		 * not atomic and may result in incosistent file system if it
868 		 * crashes before the replay is complete.
869 		 */
870 		ctx->fs->super->s_state |= EXT2_ERROR_FS;
871 		ctx->fs->super->s_state |= EXT4_FC_REPLAY;
872 		ext2fs_mark_super_dirty(ctx->fs);
873 		ext2fs_flush(ctx->fs);
874 	}
875 
876 	start = (__u8 *)bh->b_data;
877 	end = (__u8 *)bh->b_data + journal->j_blocksize - 1;
878 
879 	fc_for_each_tl(start, end, tl) {
880 		if (state->fc_replay_num_tags == 0)
881 			goto replay_done;
882 		jbd_debug(3, "Replay phase processing %s tag\n",
883 				tag2str(le16_to_cpu(tl->fc_tag)));
884 		state->fc_replay_num_tags--;
885 		switch (le16_to_cpu(tl->fc_tag)) {
886 		case EXT4_FC_TAG_CREAT:
887 		case EXT4_FC_TAG_LINK:
888 			ret = ext4_fc_handle_link_and_create(ctx, tl);
889 			break;
890 		case EXT4_FC_TAG_UNLINK:
891 			ret = ext4_fc_handle_unlink(ctx, tl);
892 			break;
893 		case EXT4_FC_TAG_ADD_RANGE:
894 			ret = ext4_fc_handle_add_extent(ctx, tl);
895 			break;
896 		case EXT4_FC_TAG_DEL_RANGE:
897 			ret = ext4_fc_handle_del_range(ctx, tl);
898 			break;
899 		case EXT4_FC_TAG_INODE:
900 			ret = ext4_fc_handle_inode(ctx, tl);
901 			break;
902 		case EXT4_FC_TAG_TAIL:
903 			ext4_fc_flush_extents(ctx, 0);
904 		case EXT4_FC_TAG_PAD:
905 		case EXT4_FC_TAG_HEAD:
906 			break;
907 		default:
908 			ret = -ECANCELED;
909 			break;
910 		}
911 		if (ret < 0)
912 			break;
913 		ret = JBD2_FC_REPLAY_CONTINUE;
914 	}
915 	return ret;
916 replay_done:
917 	jbd_debug(1, "End of fast commit replay\n");
918 	if (state->fc_current_pass != pass)
919 		return JBD2_FC_REPLAY_STOP;
920 
921 	ext2fs_calculate_summary_stats(ctx->fs, 0 /* update bg also */);
922 	ext2fs_write_block_bitmap(ctx->fs);
923 	ext2fs_write_inode_bitmap(ctx->fs);
924 	ext2fs_mark_super_dirty(ctx->fs);
925 	ext2fs_set_gdt_csum(ctx->fs);
926 	ctx->fs->super->s_state = state->fc_super_state;
927 	ext2fs_flush(ctx->fs);
928 
929 	return JBD2_FC_REPLAY_STOP;
930 }
931 
e2fsck_get_journal(e2fsck_t ctx,journal_t ** ret_journal)932 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
933 {
934 	struct process_block_struct pb;
935 	struct ext2_super_block *sb = ctx->fs->super;
936 	struct ext2_super_block jsuper;
937 	struct problem_context	pctx;
938 	struct buffer_head 	*bh;
939 	struct inode		*j_inode = NULL;
940 	struct kdev_s		*dev_fs = NULL, *dev_journal;
941 	const char		*journal_name = 0;
942 	journal_t		*journal = NULL;
943 	errcode_t		retval = 0;
944 	io_manager		io_ptr = 0;
945 	unsigned long long	start = 0;
946 	int			ret;
947 	int			ext_journal = 0;
948 	int			tried_backup_jnl = 0;
949 
950 	clear_problem_context(&pctx);
951 
952 	journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
953 	if (!journal) {
954 		return EXT2_ET_NO_MEMORY;
955 	}
956 
957 	dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
958 	if (!dev_fs) {
959 		retval = EXT2_ET_NO_MEMORY;
960 		goto errout;
961 	}
962 	dev_journal = dev_fs+1;
963 
964 	dev_fs->k_ctx = dev_journal->k_ctx = ctx;
965 	dev_fs->k_dev = K_DEV_FS;
966 	dev_journal->k_dev = K_DEV_JOURNAL;
967 
968 	journal->j_dev = dev_journal;
969 	journal->j_fs_dev = dev_fs;
970 	journal->j_inode = NULL;
971 	journal->j_blocksize = ctx->fs->blocksize;
972 
973 	if (uuid_is_null(sb->s_journal_uuid)) {
974 		if (!sb->s_journal_inum) {
975 			retval = EXT2_ET_BAD_INODE_NUM;
976 			goto errout;
977 		}
978 		j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
979 						 "journal inode");
980 		if (!j_inode) {
981 			retval = EXT2_ET_NO_MEMORY;
982 			goto errout;
983 		}
984 
985 		j_inode->i_ctx = ctx;
986 		j_inode->i_ino = sb->s_journal_inum;
987 
988 		if ((retval = ext2fs_read_inode(ctx->fs,
989 						sb->s_journal_inum,
990 						&j_inode->i_ext2))) {
991 		try_backup_journal:
992 			if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
993 			    tried_backup_jnl)
994 				goto errout;
995 			memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
996 			memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
997 			       EXT2_N_BLOCKS*4);
998 			j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
999 			j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
1000 			j_inode->i_ext2.i_links_count = 1;
1001 			j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
1002 			e2fsck_use_inode_shortcuts(ctx, 1);
1003 			ctx->stashed_ino = j_inode->i_ino;
1004 			ctx->stashed_inode = &j_inode->i_ext2;
1005 			tried_backup_jnl++;
1006 		}
1007 		if (!j_inode->i_ext2.i_links_count ||
1008 		    !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
1009 			retval = EXT2_ET_NO_JOURNAL;
1010 			goto try_backup_journal;
1011 		}
1012 		if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
1013 		    JBD2_MIN_JOURNAL_BLOCKS) {
1014 			retval = EXT2_ET_JOURNAL_TOO_SMALL;
1015 			goto try_backup_journal;
1016 		}
1017 		pb.last_block = -1;
1018 		retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
1019 					       BLOCK_FLAG_HOLE, 0,
1020 					       process_journal_block, &pb);
1021 		if ((pb.last_block + 1) * ctx->fs->blocksize <
1022 		    (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
1023 			retval = EXT2_ET_JOURNAL_TOO_SMALL;
1024 			goto try_backup_journal;
1025 		}
1026 		if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
1027 			retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
1028 						    &j_inode->i_ext2);
1029 			if (retval)
1030 				goto errout;
1031 		}
1032 
1033 		journal->j_total_len = EXT2_I_SIZE(&j_inode->i_ext2) /
1034 			journal->j_blocksize;
1035 
1036 #ifdef USE_INODE_IO
1037 		retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
1038 						 &j_inode->i_ext2,
1039 						 &journal_name);
1040 		if (retval)
1041 			goto errout;
1042 
1043 		io_ptr = inode_io_manager;
1044 #else
1045 		journal->j_inode = j_inode;
1046 		ctx->journal_io = ctx->fs->io;
1047 		if ((ret = jbd2_journal_bmap(journal, 0, &start)) != 0) {
1048 			retval = (errcode_t) (-1 * ret);
1049 			goto errout;
1050 		}
1051 #endif
1052 	} else {
1053 		ext_journal = 1;
1054 		if (!ctx->journal_name) {
1055 			char uuid[37];
1056 
1057 			uuid_unparse(sb->s_journal_uuid, uuid);
1058 			ctx->journal_name = blkid_get_devname(ctx->blkid,
1059 							      "UUID", uuid);
1060 			if (!ctx->journal_name)
1061 				ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
1062 		}
1063 		journal_name = ctx->journal_name;
1064 
1065 		if (!journal_name) {
1066 			fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
1067 			retval = EXT2_ET_LOAD_EXT_JOURNAL;
1068 			goto errout;
1069 		}
1070 
1071 		jfs_debug(1, "Using journal file %s\n", journal_name);
1072 		io_ptr = unix_io_manager;
1073 	}
1074 
1075 #if 0
1076 	test_io_backing_manager = io_ptr;
1077 	io_ptr = test_io_manager;
1078 #endif
1079 #ifndef USE_INODE_IO
1080 	if (ext_journal)
1081 #endif
1082 	{
1083 		int flags = IO_FLAG_RW;
1084 		if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
1085 		      ctx->mount_flags & EXT2_MF_READONLY))
1086 			flags |= IO_FLAG_EXCLUSIVE;
1087 		if ((ctx->mount_flags & EXT2_MF_READONLY) &&
1088 		    (ctx->options & E2F_OPT_FORCE))
1089 			flags &= ~IO_FLAG_EXCLUSIVE;
1090 
1091 
1092 		retval = io_ptr->open(journal_name, flags,
1093 				      &ctx->journal_io);
1094 	}
1095 	if (retval)
1096 		goto errout;
1097 
1098 	io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
1099 
1100 	if (ext_journal) {
1101 		blk64_t maxlen;
1102 
1103 		start = ext2fs_journal_sb_start(ctx->fs->blocksize) - 1;
1104 		bh = getblk(dev_journal, start, ctx->fs->blocksize);
1105 		if (!bh) {
1106 			retval = EXT2_ET_NO_MEMORY;
1107 			goto errout;
1108 		}
1109 		ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1110 		if ((retval = bh->b_err) != 0) {
1111 			brelse(bh);
1112 			goto errout;
1113 		}
1114 		memcpy(&jsuper, start ? bh->b_data :  bh->b_data + SUPERBLOCK_OFFSET,
1115 		       sizeof(jsuper));
1116 #ifdef WORDS_BIGENDIAN
1117 		if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
1118 			ext2fs_swap_super(&jsuper);
1119 #endif
1120 		if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
1121 		    !ext2fs_has_feature_journal_dev(&jsuper)) {
1122 			fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
1123 			retval = EXT2_ET_LOAD_EXT_JOURNAL;
1124 			brelse(bh);
1125 			goto errout;
1126 		}
1127 		/* Make sure the journal UUID is correct */
1128 		if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
1129 			   sizeof(jsuper.s_uuid))) {
1130 			fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
1131 			retval = EXT2_ET_LOAD_EXT_JOURNAL;
1132 			brelse(bh);
1133 			goto errout;
1134 		}
1135 
1136 		/* Check the superblock checksum */
1137 		if (ext2fs_has_feature_metadata_csum(&jsuper)) {
1138 			struct struct_ext2_filsys fsx;
1139 			struct ext2_super_block	superx;
1140 			void *p;
1141 
1142 			p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
1143 			memcpy(&fsx, ctx->fs, sizeof(fsx));
1144 			memcpy(&superx, ctx->fs->super, sizeof(superx));
1145 			fsx.super = &superx;
1146 			ext2fs_set_feature_metadata_csum(fsx.super);
1147 			if (!ext2fs_superblock_csum_verify(&fsx, p) &&
1148 			    fix_problem(ctx, PR_0_EXT_JOURNAL_SUPER_CSUM_INVALID,
1149 					&pctx)) {
1150 				ext2fs_superblock_csum_set(&fsx, p);
1151 				mark_buffer_dirty(bh);
1152 			}
1153 		}
1154 		brelse(bh);
1155 
1156 		maxlen = ext2fs_blocks_count(&jsuper);
1157 		journal->j_total_len = (maxlen < 1ULL << 32) ? maxlen : (1ULL << 32) - 1;
1158 		start++;
1159 	}
1160 
1161 	if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
1162 		retval = EXT2_ET_NO_MEMORY;
1163 		goto errout;
1164 	}
1165 
1166 	journal->j_sb_buffer = bh;
1167 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1168 	if (ext2fs_has_feature_fast_commit(ctx->fs->super))
1169 		journal->j_fc_replay_callback = ext4_fc_replay;
1170 	else
1171 		journal->j_fc_replay_callback = NULL;
1172 
1173 #ifdef USE_INODE_IO
1174 	if (j_inode)
1175 		ext2fs_free_mem(&j_inode);
1176 #endif
1177 
1178 	*ret_journal = journal;
1179 	e2fsck_use_inode_shortcuts(ctx, 0);
1180 	return 0;
1181 
1182 errout:
1183 	e2fsck_use_inode_shortcuts(ctx, 0);
1184 	if (dev_fs)
1185 		ext2fs_free_mem(&dev_fs);
1186 	if (j_inode)
1187 		ext2fs_free_mem(&j_inode);
1188 	if (journal)
1189 		ext2fs_free_mem(&journal);
1190 	return retval;
1191 }
1192 
e2fsck_journal_fix_bad_inode(e2fsck_t ctx,struct problem_context * pctx)1193 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
1194 					      struct problem_context *pctx)
1195 {
1196 	struct ext2_super_block *sb = ctx->fs->super;
1197 	int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1198 	int has_journal = ext2fs_has_feature_journal(ctx->fs->super);
1199 
1200 	if (has_journal || sb->s_journal_inum) {
1201 		/* The journal inode is bogus, remove and force full fsck */
1202 		pctx->ino = sb->s_journal_inum;
1203 		if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
1204 			if (has_journal && sb->s_journal_inum)
1205 				printf("*** journal has been deleted ***\n\n");
1206 			ext2fs_clear_feature_journal(sb);
1207 			sb->s_journal_inum = 0;
1208 			memset(sb->s_jnl_blocks, 0, sizeof(sb->s_jnl_blocks));
1209 			ctx->flags |= E2F_FLAG_JOURNAL_INODE;
1210 			ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1211 			e2fsck_clear_recover(ctx, 1);
1212 			return 0;
1213 		}
1214 		return EXT2_ET_CORRUPT_JOURNAL_SB;
1215 	} else if (recover) {
1216 		if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
1217 			e2fsck_clear_recover(ctx, 1);
1218 			return 0;
1219 		}
1220 		return EXT2_ET_UNSUPP_FEATURE;
1221 	}
1222 	return 0;
1223 }
1224 
1225 #define V1_SB_SIZE	0x0024
clear_v2_journal_fields(journal_t * journal)1226 static void clear_v2_journal_fields(journal_t *journal)
1227 {
1228 	e2fsck_t ctx = journal->j_dev->k_ctx;
1229 	struct problem_context pctx;
1230 
1231 	clear_problem_context(&pctx);
1232 
1233 	if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
1234 		return;
1235 
1236 	ctx->flags |= E2F_FLAG_PROBLEMS_FIXED;
1237 	memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
1238 	       ctx->fs->blocksize-V1_SB_SIZE);
1239 	mark_buffer_dirty(journal->j_sb_buffer);
1240 }
1241 
1242 
e2fsck_journal_load(journal_t * journal)1243 static errcode_t e2fsck_journal_load(journal_t *journal)
1244 {
1245 	e2fsck_t ctx = journal->j_dev->k_ctx;
1246 	journal_superblock_t *jsb;
1247 	struct buffer_head *jbh = journal->j_sb_buffer;
1248 	struct problem_context pctx;
1249 
1250 	clear_problem_context(&pctx);
1251 
1252 	ll_rw_block(REQ_OP_READ, 0, 1, &jbh);
1253 	if (jbh->b_err) {
1254 		com_err(ctx->device_name, jbh->b_err, "%s",
1255 			_("reading journal superblock\n"));
1256 		return jbh->b_err;
1257 	}
1258 
1259 	jsb = journal->j_superblock;
1260 	/* If we don't even have JBD2_MAGIC, we probably have a wrong inode */
1261 	if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER))
1262 		return e2fsck_journal_fix_bad_inode(ctx, &pctx);
1263 
1264 	switch (ntohl(jsb->s_header.h_blocktype)) {
1265 	case JBD2_SUPERBLOCK_V1:
1266 		journal->j_format_version = 1;
1267 		if (jsb->s_feature_compat ||
1268 		    jsb->s_feature_incompat ||
1269 		    jsb->s_feature_ro_compat ||
1270 		    jsb->s_nr_users)
1271 			clear_v2_journal_fields(journal);
1272 		break;
1273 
1274 	case JBD2_SUPERBLOCK_V2:
1275 		journal->j_format_version = 2;
1276 		if (ntohl(jsb->s_nr_users) > 1 &&
1277 		    uuid_is_null(ctx->fs->super->s_journal_uuid))
1278 			clear_v2_journal_fields(journal);
1279 		if (ntohl(jsb->s_nr_users) > 1) {
1280 			fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
1281 			return EXT2_ET_JOURNAL_UNSUPP_VERSION;
1282 		}
1283 		break;
1284 
1285 	/*
1286 	 * These should never appear in a journal super block, so if
1287 	 * they do, the journal is badly corrupted.
1288 	 */
1289 	case JBD2_DESCRIPTOR_BLOCK:
1290 	case JBD2_COMMIT_BLOCK:
1291 	case JBD2_REVOKE_BLOCK:
1292 		return EXT2_ET_CORRUPT_JOURNAL_SB;
1293 
1294 	/* If we don't understand the superblock major type, but there
1295 	 * is a magic number, then it is likely to be a new format we
1296 	 * just don't understand, so leave it alone. */
1297 	default:
1298 		return EXT2_ET_JOURNAL_UNSUPP_VERSION;
1299 	}
1300 
1301 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, ~JBD2_KNOWN_INCOMPAT_FEATURES))
1302 		return EXT2_ET_UNSUPP_FEATURE;
1303 
1304 	if (JBD2_HAS_RO_COMPAT_FEATURE(journal, ~JBD2_KNOWN_ROCOMPAT_FEATURES))
1305 		return EXT2_ET_RO_UNSUPP_FEATURE;
1306 
1307 	/* Checksum v1-3 are mutually exclusive features. */
1308 	if (jbd2_has_feature_csum2(journal) && jbd2_has_feature_csum3(journal))
1309 		return EXT2_ET_CORRUPT_JOURNAL_SB;
1310 
1311 	if (jbd2_journal_has_csum_v2or3(journal) &&
1312 	    jbd2_has_feature_checksum(journal))
1313 		return EXT2_ET_CORRUPT_JOURNAL_SB;
1314 
1315 	if (!e2fsck_journal_verify_csum_type(journal, jsb) ||
1316 	    !e2fsck_journal_sb_csum_verify(journal, jsb))
1317 		return EXT2_ET_CORRUPT_JOURNAL_SB;
1318 
1319 	if (jbd2_journal_has_csum_v2or3(journal))
1320 		journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
1321 						   sizeof(jsb->s_uuid));
1322 
1323 	/* We have now checked whether we know enough about the journal
1324 	 * format to be able to proceed safely, so any other checks that
1325 	 * fail we should attempt to recover from. */
1326 	if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
1327 		com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1328 			_("%s: no valid journal superblock found\n"),
1329 			ctx->device_name);
1330 		return EXT2_ET_CORRUPT_JOURNAL_SB;
1331 	}
1332 
1333 	if (ntohl(jsb->s_maxlen) < journal->j_total_len)
1334 		journal->j_total_len = ntohl(jsb->s_maxlen);
1335 	else if (ntohl(jsb->s_maxlen) > journal->j_total_len) {
1336 		com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1337 			_("%s: journal too short\n"),
1338 			ctx->device_name);
1339 		return EXT2_ET_CORRUPT_JOURNAL_SB;
1340 	}
1341 
1342 	journal->j_tail_sequence = ntohl(jsb->s_sequence);
1343 	journal->j_transaction_sequence = journal->j_tail_sequence;
1344 	journal->j_tail = ntohl(jsb->s_start);
1345 	journal->j_first = ntohl(jsb->s_first);
1346 	if (jbd2_has_feature_fast_commit(journal)) {
1347 		if (ntohl(jsb->s_maxlen) - jbd2_journal_get_num_fc_blks(jsb)
1348 			< JBD2_MIN_JOURNAL_BLOCKS) {
1349 			com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
1350 				_("%s: incorrect fast commit blocks\n"),
1351 				ctx->device_name);
1352 			return EXT2_ET_CORRUPT_JOURNAL_SB;
1353 		}
1354 		journal->j_fc_last = ntohl(jsb->s_maxlen);
1355 		journal->j_last = journal->j_fc_last -
1356 					jbd2_journal_get_num_fc_blks(jsb);
1357 		journal->j_fc_first = journal->j_last + 1;
1358 	} else {
1359 		journal->j_last = ntohl(jsb->s_maxlen);
1360 	}
1361 
1362 	return 0;
1363 }
1364 
e2fsck_journal_reset_super(e2fsck_t ctx,journal_superblock_t * jsb,journal_t * journal)1365 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
1366 				       journal_t *journal)
1367 {
1368 	char *p;
1369 	union {
1370 		uuid_t uuid;
1371 		__u32 val[4];
1372 	} u;
1373 	__u32 new_seq = 0;
1374 	int i;
1375 
1376 	/* Leave a valid existing V1 superblock signature alone.
1377 	 * Anything unrecognisable we overwrite with a new V2
1378 	 * signature. */
1379 
1380 	if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER) ||
1381 	    jsb->s_header.h_blocktype != htonl(JBD2_SUPERBLOCK_V1)) {
1382 		jsb->s_header.h_magic = htonl(JBD2_MAGIC_NUMBER);
1383 		jsb->s_header.h_blocktype = htonl(JBD2_SUPERBLOCK_V2);
1384 	}
1385 
1386 	/* Zero out everything else beyond the superblock header */
1387 
1388 	p = ((char *) jsb) + sizeof(journal_header_t);
1389 	memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
1390 
1391 	jsb->s_blocksize = htonl(ctx->fs->blocksize);
1392 	jsb->s_maxlen = htonl(journal->j_total_len);
1393 	jsb->s_first = htonl(1);
1394 
1395 	/* Initialize the journal sequence number so that there is "no"
1396 	 * chance we will find old "valid" transactions in the journal.
1397 	 * This avoids the need to zero the whole journal (slow to do,
1398 	 * and risky when we are just recovering the filesystem).
1399 	 */
1400 	uuid_generate(u.uuid);
1401 	for (i = 0; i < 4; i ++)
1402 		new_seq ^= u.val[i];
1403 	jsb->s_sequence = htonl(new_seq);
1404 	e2fsck_journal_sb_csum_set(journal, jsb);
1405 
1406 	mark_buffer_dirty(journal->j_sb_buffer);
1407 	ll_rw_block(REQ_OP_WRITE, 0, 1, &journal->j_sb_buffer);
1408 }
1409 
e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,journal_t * journal,struct problem_context * pctx)1410 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
1411 						  journal_t *journal,
1412 						  struct problem_context *pctx)
1413 {
1414 	struct ext2_super_block *sb = ctx->fs->super;
1415 	int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1416 
1417 	if (ext2fs_has_feature_journal(sb)) {
1418 		if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
1419 			e2fsck_journal_reset_super(ctx, journal->j_superblock,
1420 						   journal);
1421 			journal->j_transaction_sequence = 1;
1422 			e2fsck_clear_recover(ctx, recover);
1423 			return 0;
1424 		}
1425 		return EXT2_ET_CORRUPT_JOURNAL_SB;
1426 	} else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
1427 		return EXT2_ET_CORRUPT_JOURNAL_SB;
1428 
1429 	return 0;
1430 }
1431 
e2fsck_journal_release(e2fsck_t ctx,journal_t * journal,int reset,int drop)1432 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
1433 				   int reset, int drop)
1434 {
1435 	journal_superblock_t *jsb;
1436 
1437 	if (drop)
1438 		mark_buffer_clean(journal->j_sb_buffer);
1439 	else if (!(ctx->options & E2F_OPT_READONLY)) {
1440 		jsb = journal->j_superblock;
1441 		jsb->s_sequence = htonl(journal->j_tail_sequence);
1442 		if (reset)
1443 			jsb->s_start = 0; /* this marks the journal as empty */
1444 		e2fsck_journal_sb_csum_set(journal, jsb);
1445 		mark_buffer_dirty(journal->j_sb_buffer);
1446 	}
1447 	brelse(journal->j_sb_buffer);
1448 
1449 	if (ctx->journal_io) {
1450 		if (ctx->fs && ctx->fs->io != ctx->journal_io)
1451 			io_channel_close(ctx->journal_io);
1452 		ctx->journal_io = 0;
1453 	}
1454 
1455 #ifndef USE_INODE_IO
1456 	if (journal->j_inode)
1457 		ext2fs_free_mem(&journal->j_inode);
1458 #endif
1459 	if (journal->j_fs_dev)
1460 		ext2fs_free_mem(&journal->j_fs_dev);
1461 	ext2fs_free_mem(&journal);
1462 }
1463 
1464 /*
1465  * This function makes sure that the superblock fields regarding the
1466  * journal are consistent.
1467  */
e2fsck_check_ext3_journal(e2fsck_t ctx)1468 errcode_t e2fsck_check_ext3_journal(e2fsck_t ctx)
1469 {
1470 	struct ext2_super_block *sb = ctx->fs->super;
1471 	journal_t *journal;
1472 	int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
1473 	struct problem_context pctx;
1474 	problem_t problem;
1475 	int reset = 0, force_fsck = 0;
1476 	errcode_t retval;
1477 
1478 	/* If we don't have any journal features, don't do anything more */
1479 	if (!ext2fs_has_feature_journal(sb) &&
1480 	    !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
1481 	    uuid_is_null(sb->s_journal_uuid))
1482  		return 0;
1483 
1484 	clear_problem_context(&pctx);
1485 	pctx.num = sb->s_journal_inum;
1486 
1487 	retval = e2fsck_get_journal(ctx, &journal);
1488 	if (retval) {
1489 		if ((retval == EXT2_ET_BAD_INODE_NUM) ||
1490 		    (retval == EXT2_ET_BAD_BLOCK_NUM) ||
1491 		    (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
1492 		    (retval == EXT2_ET_NO_JOURNAL))
1493 			return e2fsck_journal_fix_bad_inode(ctx, &pctx);
1494 		return retval;
1495 	}
1496 
1497 	retval = e2fsck_journal_load(journal);
1498 	if (retval) {
1499 		if ((retval == EXT2_ET_CORRUPT_JOURNAL_SB) ||
1500 		    ((retval == EXT2_ET_UNSUPP_FEATURE) &&
1501 		    (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
1502 				  &pctx))) ||
1503 		    ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
1504 		    (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
1505 				  &pctx))) ||
1506 		    ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
1507 		    (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
1508 			retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
1509 								  &pctx);
1510 		e2fsck_journal_release(ctx, journal, 0, 1);
1511 		return retval;
1512 	}
1513 
1514 	/*
1515 	 * We want to make the flags consistent here.  We will not leave with
1516 	 * needs_recovery set but has_journal clear.  We can't get in a loop
1517 	 * with -y, -n, or -p, only if a user isn't making up their mind.
1518 	 */
1519 no_has_journal:
1520 	if (!ext2fs_has_feature_journal(sb)) {
1521 		recover = ext2fs_has_feature_journal_needs_recovery(sb);
1522 		if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
1523 			if (recover &&
1524 			    !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
1525 				goto no_has_journal;
1526 			/*
1527 			 * Need a full fsck if we are releasing a
1528 			 * journal stored on a reserved inode.
1529 			 */
1530 			force_fsck = recover ||
1531 				(sb->s_journal_inum < EXT2_FIRST_INODE(sb));
1532 			/* Clear all of the journal fields */
1533 			sb->s_journal_inum = 0;
1534 			sb->s_journal_dev = 0;
1535 			memset(sb->s_journal_uuid, 0,
1536 			       sizeof(sb->s_journal_uuid));
1537 			e2fsck_clear_recover(ctx, force_fsck);
1538 		} else if (!(ctx->options & E2F_OPT_READONLY)) {
1539 			ext2fs_set_feature_journal(sb);
1540 			ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1541 			ext2fs_mark_super_dirty(ctx->fs);
1542 		}
1543 	}
1544 
1545 	if (ext2fs_has_feature_journal(sb) &&
1546 	    !ext2fs_has_feature_journal_needs_recovery(sb) &&
1547 	    journal->j_superblock->s_start != 0) {
1548 		/* Print status information */
1549 		fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
1550 		if (ctx->superblock)
1551 			problem = PR_0_JOURNAL_RUN_DEFAULT;
1552 		else
1553 			problem = PR_0_JOURNAL_RUN;
1554 		if (fix_problem(ctx, problem, &pctx)) {
1555 			ctx->options |= E2F_OPT_FORCE;
1556 			ext2fs_set_feature_journal_needs_recovery(sb);
1557 			ext2fs_mark_super_dirty(ctx->fs);
1558 		} else if (fix_problem(ctx,
1559 				       PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
1560 			reset = 1;
1561 			sb->s_state &= ~EXT2_VALID_FS;
1562 			ext2fs_mark_super_dirty(ctx->fs);
1563 		}
1564 		/*
1565 		 * If the user answers no to the above question, we
1566 		 * ignore the fact that journal apparently has data;
1567 		 * accidentally replaying over valid data would be far
1568 		 * worse than skipping a questionable recovery.
1569 		 *
1570 		 * XXX should we abort with a fatal error here?  What
1571 		 * will the ext3 kernel code do if a filesystem with
1572 		 * !NEEDS_RECOVERY but with a non-zero
1573 		 * journal->j_superblock->s_start is mounted?
1574 		 */
1575 	}
1576 
1577 	/*
1578 	 * If we don't need to do replay the journal, check to see if
1579 	 * the journal's errno is set; if so, we need to mark the file
1580 	 * system as being corrupt and clear the journal's s_errno.
1581 	 */
1582 	if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
1583 	    journal->j_superblock->s_errno) {
1584 		ctx->fs->super->s_state |= EXT2_ERROR_FS;
1585 		ext2fs_mark_super_dirty(ctx->fs);
1586 		journal->j_superblock->s_errno = 0;
1587 		e2fsck_journal_sb_csum_set(journal, journal->j_superblock);
1588 		mark_buffer_dirty(journal->j_sb_buffer);
1589 	}
1590 
1591 	e2fsck_journal_release(ctx, journal, reset, 0);
1592 	return retval;
1593 }
1594 
recover_ext3_journal(e2fsck_t ctx)1595 static errcode_t recover_ext3_journal(e2fsck_t ctx)
1596 {
1597 	struct problem_context	pctx;
1598 	journal_t *journal;
1599 	errcode_t retval;
1600 
1601 	clear_problem_context(&pctx);
1602 
1603 	retval = jbd2_journal_init_revoke_record_cache();
1604 	if (retval)
1605 		return retval;
1606 
1607 	retval = jbd2_journal_init_revoke_table_cache();
1608 	if (retval)
1609 		return retval;
1610 
1611 	retval = e2fsck_get_journal(ctx, &journal);
1612 	if (retval)
1613 		return retval;
1614 
1615 	retval = e2fsck_journal_load(journal);
1616 	if (retval)
1617 		goto errout;
1618 
1619 	retval = jbd2_journal_init_revoke(journal, 1024);
1620 	if (retval)
1621 		goto errout;
1622 
1623 	retval = -jbd2_journal_recover(journal);
1624 	if (retval)
1625 		goto errout;
1626 
1627 	if (journal->j_failed_commit) {
1628 		pctx.ino = journal->j_failed_commit;
1629 		fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
1630 		journal->j_superblock->s_errno = -EINVAL;
1631 		mark_buffer_dirty(journal->j_sb_buffer);
1632 	}
1633 
1634 	journal->j_tail_sequence = journal->j_transaction_sequence;
1635 
1636 errout:
1637 	jbd2_journal_destroy_revoke(journal);
1638 	jbd2_journal_destroy_revoke_record_cache();
1639 	jbd2_journal_destroy_revoke_table_cache();
1640 	e2fsck_journal_release(ctx, journal, 1, 0);
1641 	return retval;
1642 }
1643 
e2fsck_run_ext3_journal(e2fsck_t ctx)1644 errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
1645 {
1646 	io_manager io_ptr = ctx->fs->io->manager;
1647 	int blocksize = ctx->fs->blocksize;
1648 	errcode_t	retval, recover_retval;
1649 	io_stats	stats = 0;
1650 	unsigned long long kbytes_written = 0;
1651 
1652 	printf(_("%s: recovering journal\n"), ctx->device_name);
1653 	if (ctx->options & E2F_OPT_READONLY) {
1654 		printf(_("%s: won't do journal recovery while read-only\n"),
1655 		       ctx->device_name);
1656 		return EXT2_ET_FILE_RO;
1657 	}
1658 
1659 	if (ctx->fs->flags & EXT2_FLAG_DIRTY)
1660 		ext2fs_flush(ctx->fs);	/* Force out any modifications */
1661 
1662 	recover_retval = recover_ext3_journal(ctx);
1663 
1664 	/*
1665 	 * Reload the filesystem context to get up-to-date data from disk
1666 	 * because journal recovery will change the filesystem under us.
1667 	 */
1668 	if (ctx->fs->super->s_kbytes_written &&
1669 	    ctx->fs->io->manager->get_stats)
1670 		ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
1671 	if (stats && stats->bytes_written)
1672 		kbytes_written = stats->bytes_written >> 10;
1673 
1674 	ext2fs_mmp_stop(ctx->fs);
1675 	ext2fs_free(ctx->fs);
1676 	retval = ext2fs_open(ctx->filesystem_name, ctx->openfs_flags,
1677 			     ctx->superblock, blocksize, io_ptr,
1678 			     &ctx->fs);
1679 	if (retval) {
1680 		com_err(ctx->program_name, retval,
1681 			_("while trying to re-open %s"),
1682 			ctx->device_name);
1683 		fatal_error(ctx, 0);
1684 	}
1685 	ctx->fs->priv_data = ctx;
1686 	ctx->fs->now = ctx->now;
1687 	ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
1688 	ctx->fs->super->s_kbytes_written += kbytes_written;
1689 
1690 	/* Set the superblock flags */
1691 	e2fsck_clear_recover(ctx, recover_retval != 0);
1692 
1693 	/*
1694 	 * Do one last sanity check, and propagate journal->s_errno to
1695 	 * the EXT2_ERROR_FS flag in the fs superblock if needed.
1696 	 */
1697 	retval = e2fsck_check_ext3_journal(ctx);
1698 	return retval ? retval : recover_retval;
1699 }
1700 
1701 /*
1702  * This function will move the journal inode from a visible file in
1703  * the filesystem directory hierarchy to the reserved inode if necessary.
1704  */
1705 static const char * const journal_names[] = {
1706 	".journal", "journal", ".journal.dat", "journal.dat", 0 };
1707 
e2fsck_move_ext3_journal(e2fsck_t ctx)1708 void e2fsck_move_ext3_journal(e2fsck_t ctx)
1709 {
1710 	struct ext2_super_block *sb = ctx->fs->super;
1711 	struct problem_context	pctx;
1712 	struct ext2_inode 	inode;
1713 	ext2_filsys		fs = ctx->fs;
1714 	ext2_ino_t		ino;
1715 	errcode_t		retval;
1716 	const char * const *	cpp;
1717 	dgrp_t			group;
1718 	int			mount_flags;
1719 
1720 	clear_problem_context(&pctx);
1721 
1722 	/*
1723 	 * If the filesystem is opened read-only, or there is no
1724 	 * journal, then do nothing.
1725 	 */
1726 	if ((ctx->options & E2F_OPT_READONLY) ||
1727 	    (sb->s_journal_inum == 0) ||
1728 	    !ext2fs_has_feature_journal(sb))
1729 		return;
1730 
1731 	/*
1732 	 * Read in the journal inode
1733 	 */
1734 	if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
1735 		return;
1736 
1737 	/*
1738 	 * If it's necessary to backup the journal inode, do so.
1739 	 */
1740 	if ((sb->s_jnl_backup_type == 0) ||
1741 	    ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
1742 	     memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
1743 		if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
1744 			memcpy(sb->s_jnl_blocks, inode.i_block,
1745 			       EXT2_N_BLOCKS*4);
1746 			sb->s_jnl_blocks[15] = inode.i_size_high;
1747 			sb->s_jnl_blocks[16] = inode.i_size;
1748 			sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
1749 			ext2fs_mark_super_dirty(fs);
1750 			fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1751 		}
1752 	}
1753 
1754 	/*
1755 	 * If the journal is already the hidden inode, then do nothing
1756 	 */
1757 	if (sb->s_journal_inum == EXT2_JOURNAL_INO)
1758 		return;
1759 
1760 	/*
1761 	 * The journal inode had better have only one link and not be readable.
1762 	 */
1763 	if (inode.i_links_count != 1)
1764 		return;
1765 
1766 	/*
1767 	 * If the filesystem is mounted, or we can't tell whether
1768 	 * or not it's mounted, do nothing.
1769 	 */
1770 	retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
1771 	if (retval || (mount_flags & EXT2_MF_MOUNTED))
1772 		return;
1773 
1774 	/*
1775 	 * If we can't find the name of the journal inode, then do
1776 	 * nothing.
1777 	 */
1778 	for (cpp = journal_names; *cpp; cpp++) {
1779 		retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1780 				       strlen(*cpp), 0, &ino);
1781 		if ((retval == 0) && (ino == sb->s_journal_inum))
1782 			break;
1783 	}
1784 	if (*cpp == 0)
1785 		return;
1786 
1787 	/* We need the inode bitmap to be loaded */
1788 	retval = ext2fs_read_bitmaps(fs);
1789 	if (retval)
1790 		return;
1791 
1792 	pctx.str = *cpp;
1793 	if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1794 		return;
1795 
1796 	/*
1797 	 * OK, we've done all the checks, let's actually move the
1798 	 * journal inode.  Errors at this point mean we need to force
1799 	 * an ext2 filesystem check.
1800 	 */
1801 	if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1802 		goto err_out;
1803 	if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1804 		goto err_out;
1805 	sb->s_journal_inum = EXT2_JOURNAL_INO;
1806 	ext2fs_mark_super_dirty(fs);
1807 	fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1808 	inode.i_links_count = 0;
1809 	inode.i_dtime = ctx->now;
1810 	if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1811 		goto err_out;
1812 
1813 	group = ext2fs_group_of_ino(fs, ino);
1814 	ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1815 	ext2fs_mark_ib_dirty(fs);
1816 	ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1817 	ext2fs_group_desc_csum_set(fs, group);
1818 	fs->super->s_free_inodes_count++;
1819 	return;
1820 
1821 err_out:
1822 	pctx.errcode = retval;
1823 	fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1824 	fs->super->s_state &= ~EXT2_VALID_FS;
1825 	ext2fs_mark_super_dirty(fs);
1826 	return;
1827 }
1828 
1829 /*
1830  * This function makes sure the superblock hint for the external
1831  * journal is correct.
1832  */
e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)1833 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1834 {
1835 	struct ext2_super_block *sb = ctx->fs->super;
1836 	struct problem_context pctx;
1837 	char uuid[37], *journal_name;
1838 	struct stat st;
1839 
1840 	if (!ext2fs_has_feature_journal(sb) ||
1841 	    uuid_is_null(sb->s_journal_uuid))
1842  		return 0;
1843 
1844 	uuid_unparse(sb->s_journal_uuid, uuid);
1845 	journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1846 	if (!journal_name)
1847 		return 0;
1848 
1849 	if (stat(journal_name, &st) < 0) {
1850 		free(journal_name);
1851 		return 0;
1852 	}
1853 
1854 	if (st.st_rdev != sb->s_journal_dev) {
1855 		clear_problem_context(&pctx);
1856 		pctx.num = st.st_rdev;
1857 		if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1858 			sb->s_journal_dev = st.st_rdev;
1859 			ext2fs_mark_super_dirty(ctx->fs);
1860 		}
1861 	}
1862 
1863 	free(journal_name);
1864 	return 0;
1865 }
1866