1 /*
2  * pass1b.c --- Pass #1b of e2fsck
3  *
4  * This file contains pass1B, pass1C, and pass1D of e2fsck.  They are
5  * only invoked if pass 1 discovered blocks which are in use by more
6  * than one inode.
7  *
8  * Pass1B scans the data blocks of all the inodes again, generating a
9  * complete list of duplicate blocks and which inodes have claimed
10  * them.
11  *
12  * Pass1C does a tree-traversal of the filesystem, to determine the
13  * parent directories of these inodes.  This step is necessary so that
14  * e2fsck can print out the pathnames of affected inodes.
15  *
16  * Pass1D is a reconciliation pass.  For each inode with duplicate
17  * blocks, the user is prompted if s/he would like to clone the file
18  * (so that the file gets a fresh copy of the duplicated blocks) or
19  * simply to delete the file.
20  *
21  * Copyright (C) 1993, 1994, 1995, 1996, 1997 Theodore Ts'o.
22  *
23  * %Begin-Header%
24  * This file may be redistributed under the terms of the GNU Public
25  * License.
26  * %End-Header%
27  *
28  */
29 
30 #include "config.h"
31 #include <time.h>
32 #ifdef HAVE_ERRNO_H
33 #include <errno.h>
34 #endif
35 
36 #ifdef HAVE_INTTYPES_H
37 #include <inttypes.h>
38 #endif
39 
40 #ifndef HAVE_INTPTR_T
41 typedef long intptr_t;
42 #endif
43 
44 /* Needed for architectures where sizeof(int) != sizeof(void *) */
45 #define INT_TO_VOIDPTR(val)  ((void *)(intptr_t)(val))
46 #define VOIDPTR_TO_INT(ptr)  ((int)(intptr_t)(ptr))
47 
48 #include <et/com_err.h>
49 #include "e2fsck.h"
50 
51 #include "problem.h"
52 #include "support/dict.h"
53 
54 /* Define an extension to the ext2 library's block count information */
55 #define BLOCK_COUNT_EXTATTR	(-5)
56 
57 struct cluster_el {
58 	blk64_t	cluster;
59 	struct cluster_el *next;
60 };
61 
62 struct inode_el {
63 	ext2_ino_t	inode;
64 	struct inode_el *next;
65 };
66 
67 struct dup_cluster {
68 	int		num_bad;
69 	struct inode_el *inode_list;
70 };
71 
72 /*
73  * This structure stores information about a particular inode which
74  * is sharing blocks with other inodes.  This information is collected
75  * to display to the user, so that the user knows what files he or she
76  * is dealing with, when trying to decide how to resolve the conflict
77  * of multiply-claimed blocks.
78  */
79 struct dup_inode {
80 	ext2_ino_t		dir;
81 	int			num_dupblocks;
82 	struct ext2_inode_large	inode;
83 	struct cluster_el	*cluster_list;
84 };
85 
86 static int process_pass1b_block(ext2_filsys fs, blk64_t	*blocknr,
87 				e2_blkcnt_t blockcnt, blk64_t ref_blk,
88 				int ref_offset, void *priv_data);
89 static void delete_file(e2fsck_t ctx, ext2_ino_t ino,
90 			struct dup_inode *dp, char *block_buf);
91 static errcode_t clone_file(e2fsck_t ctx, ext2_ino_t ino,
92 			    struct dup_inode *dp, char* block_buf);
93 static int check_if_fs_block(e2fsck_t ctx, blk64_t test_block);
94 static int check_if_fs_cluster(e2fsck_t ctx, blk64_t cluster);
95 
96 static void pass1b(e2fsck_t ctx, char *block_buf);
97 static void pass1c(e2fsck_t ctx, char *block_buf);
98 static void pass1d(e2fsck_t ctx, char *block_buf);
99 
100 static int dup_inode_count = 0;
101 static int dup_inode_founddir = 0;
102 
103 static dict_t clstr_dict, ino_dict;
104 
105 static ext2fs_inode_bitmap inode_dup_map;
106 
dict_int_cmp(const void * a,const void * b)107 static int dict_int_cmp(const void *a, const void *b)
108 {
109 	intptr_t	ia, ib;
110 
111 	ia = (intptr_t)a;
112 	ib = (intptr_t)b;
113 
114 	return (ia-ib);
115 }
116 
117 /*
118  * Add a duplicate block record
119  */
add_dupe(e2fsck_t ctx,ext2_ino_t ino,blk64_t cluster,struct ext2_inode_large * inode)120 static void add_dupe(e2fsck_t ctx, ext2_ino_t ino, blk64_t cluster,
121 		     struct ext2_inode_large *inode)
122 {
123 	dnode_t	*n;
124 	struct dup_cluster	*db;
125 	struct dup_inode	*di;
126 	struct cluster_el	*cluster_el;
127 	struct inode_el 	*ino_el;
128 
129 	n = dict_lookup(&clstr_dict, INT_TO_VOIDPTR(cluster));
130 	if (n)
131 		db = (struct dup_cluster *) dnode_get(n);
132 	else {
133 		db = (struct dup_cluster *) e2fsck_allocate_memory(ctx,
134 			sizeof(struct dup_cluster), "duplicate cluster header");
135 		db->num_bad = 0;
136 		db->inode_list = 0;
137 		dict_alloc_insert(&clstr_dict, INT_TO_VOIDPTR(cluster), db);
138 	}
139 	ino_el = (struct inode_el *) e2fsck_allocate_memory(ctx,
140 			 sizeof(struct inode_el), "inode element");
141 	ino_el->inode = ino;
142 	ino_el->next = db->inode_list;
143 	db->inode_list = ino_el;
144 	db->num_bad++;
145 
146 	n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(ino));
147 	if (n)
148 		di = (struct dup_inode *) dnode_get(n);
149 	else {
150 		di = (struct dup_inode *) e2fsck_allocate_memory(ctx,
151 			 sizeof(struct dup_inode), "duplicate inode header");
152 		if (ino == EXT2_ROOT_INO) {
153 			di->dir = EXT2_ROOT_INO;
154 			dup_inode_founddir++;
155 		} else
156 			di->dir = 0;
157 
158 		di->num_dupblocks = 0;
159 		di->cluster_list = 0;
160 		di->inode = *inode;
161 		dict_alloc_insert(&ino_dict, INT_TO_VOIDPTR(ino), di);
162 	}
163 	cluster_el = (struct cluster_el *) e2fsck_allocate_memory(ctx,
164 			 sizeof(struct cluster_el), "cluster element");
165 	cluster_el->cluster = cluster;
166 	cluster_el->next = di->cluster_list;
167 	di->cluster_list = cluster_el;
168 	di->num_dupblocks++;
169 }
170 
171 /*
172  * Free a duplicate inode record
173  */
inode_dnode_free(dnode_t * node,void * context EXT2FS_ATTR ((unused)))174 static void inode_dnode_free(dnode_t *node,
175 			     void *context EXT2FS_ATTR((unused)))
176 {
177 	struct dup_inode	*di;
178 	struct cluster_el		*p, *next;
179 
180 	di = (struct dup_inode *) dnode_get(node);
181 	for (p = di->cluster_list; p; p = next) {
182 		next = p->next;
183 		free(p);
184 	}
185 	free(di);
186 	free(node);
187 }
188 
189 /*
190  * Free a duplicate cluster record
191  */
cluster_dnode_free(dnode_t * node,void * context EXT2FS_ATTR ((unused)))192 static void cluster_dnode_free(dnode_t *node,
193 			       void *context EXT2FS_ATTR((unused)))
194 {
195 	struct dup_cluster	*dc;
196 	struct inode_el		*p, *next;
197 
198 	dc = (struct dup_cluster *) dnode_get(node);
199 	for (p = dc->inode_list; p; p = next) {
200 		next = p->next;
201 		free(p);
202 	}
203 	free(dc);
204 	free(node);
205 }
206 
207 
208 /*
209  * Main procedure for handling duplicate blocks
210  */
e2fsck_pass1_dupblocks(e2fsck_t ctx,char * block_buf)211 void e2fsck_pass1_dupblocks(e2fsck_t ctx, char *block_buf)
212 {
213 	ext2_filsys 		fs = ctx->fs;
214 	struct problem_context	pctx;
215 #ifdef RESOURCE_TRACK
216 	struct resource_track	rtrack;
217 #endif
218 
219 	clear_problem_context(&pctx);
220 
221 	pctx.errcode = e2fsck_allocate_inode_bitmap(fs,
222 			_("multiply claimed inode map"),
223 			EXT2FS_BMAP64_RBTREE, "inode_dup_map",
224 			&inode_dup_map);
225 	if (pctx.errcode) {
226 		fix_problem(ctx, PR_1B_ALLOCATE_IBITMAP_ERROR, &pctx);
227 		ctx->flags |= E2F_FLAG_ABORT;
228 		return;
229 	}
230 
231 	dict_init(&ino_dict, DICTCOUNT_T_MAX, dict_int_cmp);
232 	dict_init(&clstr_dict, DICTCOUNT_T_MAX, dict_int_cmp);
233 	dict_set_allocator(&ino_dict, NULL, inode_dnode_free, NULL);
234 	dict_set_allocator(&clstr_dict, NULL, cluster_dnode_free, NULL);
235 
236 	init_resource_track(&rtrack, ctx->fs->io);
237 	pass1b(ctx, block_buf);
238 	print_resource_track(ctx, "Pass 1b", &rtrack, ctx->fs->io);
239 
240 	init_resource_track(&rtrack, ctx->fs->io);
241 	pass1c(ctx, block_buf);
242 	print_resource_track(ctx, "Pass 1c", &rtrack, ctx->fs->io);
243 
244 	init_resource_track(&rtrack, ctx->fs->io);
245 	pass1d(ctx, block_buf);
246 	print_resource_track(ctx, "Pass 1d", &rtrack, ctx->fs->io);
247 
248 	if (ext2fs_has_feature_shared_blocks(ctx->fs->super) &&
249 	    (ctx->options & E2F_OPT_UNSHARE_BLOCKS)) {
250 		/*
251 		 * If we successfully managed to unshare all blocks, unset the
252 		 * shared block feature.
253 		 */
254 		blk64_t next;
255 		int result = ext2fs_find_first_set_block_bitmap2(
256 			ctx->block_dup_map,
257 			ctx->fs->super->s_first_data_block,
258 			ext2fs_blocks_count(ctx->fs->super) - 1,
259 			&next);
260 		if (result == ENOENT && !(ctx->options & E2F_OPT_NO)) {
261 			ext2fs_clear_feature_shared_blocks(ctx->fs->super);
262 			ext2fs_mark_super_dirty(ctx->fs);
263 		}
264 	}
265 
266 	/*
267 	 * Time to free all of the accumulated data structures that we
268 	 * don't need anymore.
269 	 */
270 	dict_free_nodes(&ino_dict);
271 	dict_free_nodes(&clstr_dict);
272 	ext2fs_free_inode_bitmap(inode_dup_map);
273 }
274 
275 /*
276  * Scan the inodes looking for inodes that contain duplicate blocks.
277  */
278 struct process_block_struct {
279 	e2fsck_t	ctx;
280 	ext2_ino_t	ino;
281 	int		dup_blocks;
282 	blk64_t		cur_cluster, phys_cluster;
283 	blk64_t		last_blk;
284 	struct ext2_inode_large *inode;
285 	struct problem_context *pctx;
286 };
287 
pass1b(e2fsck_t ctx,char * block_buf)288 static void pass1b(e2fsck_t ctx, char *block_buf)
289 {
290 	ext2_filsys fs = ctx->fs;
291 	ext2_ino_t ino = 0;
292 	struct ext2_inode_large inode;
293 	ext2_inode_scan	scan;
294 	struct process_block_struct pb;
295 	struct problem_context pctx;
296 	problem_t op;
297 
298 	clear_problem_context(&pctx);
299 
300 	if (!(ctx->options & E2F_OPT_PREEN))
301 		fix_problem(ctx, PR_1B_PASS_HEADER, &pctx);
302 	pctx.errcode = ext2fs_open_inode_scan(fs, ctx->inode_buffer_blocks,
303 					      &scan);
304 	if (pctx.errcode) {
305 		fix_problem(ctx, PR_1B_ISCAN_ERROR, &pctx);
306 		ctx->flags |= E2F_FLAG_ABORT;
307 		return;
308 	}
309 	ctx->stashed_inode = EXT2_INODE(&inode);
310 	pb.ctx = ctx;
311 	pb.pctx = &pctx;
312 	pctx.str = "pass1b";
313 	while (1) {
314 		if (ino % (fs->super->s_inodes_per_group * 4) == 1) {
315 			if (e2fsck_mmp_update(fs))
316 				fatal_error(ctx, 0);
317 		}
318 		pctx.errcode = ext2fs_get_next_inode_full(scan, &ino,
319 				EXT2_INODE(&inode), sizeof(inode));
320 		if (pctx.errcode == EXT2_ET_BAD_BLOCK_IN_INODE_TABLE)
321 			continue;
322 		if (pctx.errcode) {
323 			pctx.ino = ino;
324 			fix_problem(ctx, PR_1B_ISCAN_ERROR, &pctx);
325 			ctx->flags |= E2F_FLAG_ABORT;
326 			return;
327 		}
328 		if (!ino)
329 			break;
330 		pctx.ino = ctx->stashed_ino = ino;
331 		if ((ino != EXT2_BAD_INO) &&
332 		    !ext2fs_test_inode_bitmap2(ctx->inode_used_map, ino))
333 			continue;
334 
335 		pb.ino = ino;
336 		pb.dup_blocks = 0;
337 		pb.inode = &inode;
338 		pb.cur_cluster = ~0;
339 		pb.phys_cluster = ~0;
340 		pb.last_blk = 0;
341 		pb.pctx->blk = pb.pctx->blk2 = 0;
342 
343 		if (ext2fs_inode_has_valid_blocks2(fs, EXT2_INODE(&inode)) ||
344 		    (ino == EXT2_BAD_INO))
345 			pctx.errcode = ext2fs_block_iterate3(fs, ino,
346 					     BLOCK_FLAG_READ_ONLY, block_buf,
347 					     process_pass1b_block, &pb);
348 		/* If the feature is not set, attrs will be cleared later anyway */
349 		if (ext2fs_has_feature_xattr(fs->super) &&
350 		    ext2fs_file_acl_block(fs, EXT2_INODE(&inode))) {
351 			blk64_t blk = ext2fs_file_acl_block(fs, EXT2_INODE(&inode));
352 			process_pass1b_block(fs, &blk,
353 					     BLOCK_COUNT_EXTATTR, 0, 0, &pb);
354 			ext2fs_file_acl_block_set(fs, EXT2_INODE(&inode), blk);
355 		}
356 		if (pb.dup_blocks) {
357 			if (ino != EXT2_BAD_INO) {
358 				op = pctx.blk == pctx.blk2 ?
359 					PR_1B_DUP_BLOCK : PR_1B_DUP_RANGE;
360 				fix_problem(ctx, op, pb.pctx);
361 			}
362 			end_problem_latch(ctx, PR_LATCH_DBLOCK);
363 			if (ino >= EXT2_FIRST_INODE(fs->super) ||
364 			    ino == EXT2_ROOT_INO)
365 				dup_inode_count++;
366 		}
367 		if (pctx.errcode)
368 			fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
369 	}
370 	ext2fs_close_inode_scan(scan);
371 	e2fsck_use_inode_shortcuts(ctx, 0);
372 }
373 
process_pass1b_block(ext2_filsys fs EXT2FS_ATTR ((unused)),blk64_t * block_nr,e2_blkcnt_t blockcnt,blk64_t ref_blk EXT2FS_ATTR ((unused)),int ref_offset EXT2FS_ATTR ((unused)),void * priv_data)374 static int process_pass1b_block(ext2_filsys fs EXT2FS_ATTR((unused)),
375 				blk64_t	*block_nr,
376 				e2_blkcnt_t blockcnt,
377 				blk64_t ref_blk EXT2FS_ATTR((unused)),
378 				int ref_offset EXT2FS_ATTR((unused)),
379 				void *priv_data)
380 {
381 	struct process_block_struct *p;
382 	e2fsck_t ctx;
383 	blk64_t	lc, pc;
384 	problem_t op;
385 
386 	if (*block_nr == 0)
387 		return 0;
388 	p = (struct process_block_struct *) priv_data;
389 	ctx = p->ctx;
390 	lc = EXT2FS_B2C(fs, blockcnt);
391 	pc = EXT2FS_B2C(fs, *block_nr);
392 
393 	if (!ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr))
394 		goto finish;
395 
396 	/* OK, this is a duplicate block */
397 	if (p->ino != EXT2_BAD_INO) {
398 		if (p->last_blk + 1 != *block_nr) {
399 			if (p->last_blk) {
400 				op = p->pctx->blk == p->pctx->blk2 ?
401 						PR_1B_DUP_BLOCK :
402 						PR_1B_DUP_RANGE;
403 				fix_problem(ctx, op, p->pctx);
404 			}
405 			p->pctx->blk = *block_nr;
406 		}
407 		p->pctx->blk2 = *block_nr;
408 		p->last_blk = *block_nr;
409 	}
410 	p->dup_blocks++;
411 	ext2fs_mark_inode_bitmap2(inode_dup_map, p->ino);
412 
413 	/*
414 	 * Qualifications for submitting a block for duplicate processing:
415 	 * It's an extent/indirect block (and has a negative logical offset);
416 	 * we've crossed a logical cluster boundary; or the physical cluster
417 	 * suddenly changed, which indicates that blocks in a logical cluster
418 	 * are mapped to multiple physical clusters.
419 	 */
420 	if (blockcnt < 0 || lc != p->cur_cluster || pc != p->phys_cluster)
421 		add_dupe(ctx, p->ino, EXT2FS_B2C(fs, *block_nr), p->inode);
422 
423 finish:
424 	p->cur_cluster = lc;
425 	p->phys_cluster = pc;
426 	return 0;
427 }
428 
429 /*
430  * Pass 1c: Scan directories for inodes with duplicate blocks.  This
431  * is used so that we can print pathnames when prompting the user for
432  * what to do.
433  */
434 struct search_dir_struct {
435 	int		count;
436 	ext2_ino_t	first_inode;
437 	ext2_ino_t	max_inode;
438 };
439 
search_dirent_proc(ext2_ino_t dir,int entry,struct ext2_dir_entry * dirent,int offset EXT2FS_ATTR ((unused)),int blocksize EXT2FS_ATTR ((unused)),char * buf EXT2FS_ATTR ((unused)),void * priv_data)440 static int search_dirent_proc(ext2_ino_t dir, int entry,
441 			      struct ext2_dir_entry *dirent,
442 			      int offset EXT2FS_ATTR((unused)),
443 			      int blocksize EXT2FS_ATTR((unused)),
444 			      char *buf EXT2FS_ATTR((unused)),
445 			      void *priv_data)
446 {
447 	struct search_dir_struct *sd;
448 	struct dup_inode	*p;
449 	dnode_t			*n;
450 
451 	sd = (struct search_dir_struct *) priv_data;
452 
453 	if (dirent->inode > sd->max_inode)
454 		/* Should abort this inode, but not everything */
455 		return 0;
456 
457 	if ((dirent->inode < sd->first_inode) || (entry < DIRENT_OTHER_FILE) ||
458 	    !ext2fs_test_inode_bitmap2(inode_dup_map, dirent->inode))
459 		return 0;
460 
461 	n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(dirent->inode));
462 	if (!n)
463 		return 0;
464 	p = (struct dup_inode *) dnode_get(n);
465 	if (!p->dir) {
466 		p->dir = dir;
467 		sd->count--;
468 	}
469 
470 	return(sd->count ? 0 : DIRENT_ABORT);
471 }
472 
473 
pass1c(e2fsck_t ctx,char * block_buf)474 static void pass1c(e2fsck_t ctx, char *block_buf)
475 {
476 	ext2_filsys fs = ctx->fs;
477 	struct search_dir_struct sd;
478 	struct problem_context pctx;
479 
480 	clear_problem_context(&pctx);
481 
482 	if (!(ctx->options & E2F_OPT_PREEN))
483 		fix_problem(ctx, PR_1C_PASS_HEADER, &pctx);
484 
485 	/*
486 	 * Search through all directories to translate inodes to names
487 	 * (by searching for the containing directory for that inode.)
488 	 */
489 	sd.count = dup_inode_count - dup_inode_founddir;
490 	sd.first_inode = EXT2_FIRST_INODE(fs->super);
491 	sd.max_inode = fs->super->s_inodes_count;
492 	ext2fs_dblist_dir_iterate(fs->dblist, 0, block_buf,
493 				  search_dirent_proc, &sd);
494 }
495 
pass1d(e2fsck_t ctx,char * block_buf)496 static void pass1d(e2fsck_t ctx, char *block_buf)
497 {
498 	ext2_filsys fs = ctx->fs;
499 	struct dup_inode	*p, *t;
500 	struct dup_cluster	*q;
501 	ext2_ino_t		*shared, ino;
502 	int	shared_len;
503 	int	i;
504 	int	file_ok;
505 	int	meta_data = 0;
506 	struct problem_context pctx;
507 	dnode_t	*n, *m;
508 	struct cluster_el	*s;
509 	struct inode_el *r;
510 
511 	clear_problem_context(&pctx);
512 
513 	if (!(ctx->options & E2F_OPT_PREEN))
514 		fix_problem(ctx, PR_1D_PASS_HEADER, &pctx);
515 	e2fsck_read_bitmaps(ctx);
516 
517 	pctx.num = dup_inode_count; /* dict_count(&ino_dict); */
518 	fix_problem(ctx, PR_1D_NUM_DUP_INODES, &pctx);
519 	shared = (ext2_ino_t *) e2fsck_allocate_memory(ctx,
520 				sizeof(ext2_ino_t) * dict_count(&ino_dict),
521 				"Shared inode list");
522 	for (n = dict_first(&ino_dict); n; n = dict_next(&ino_dict, n)) {
523 		p = (struct dup_inode *) dnode_get(n);
524 		shared_len = 0;
525 		file_ok = 1;
526 		ino = (ext2_ino_t)VOIDPTR_TO_INT(dnode_getkey(n));
527 		if (ino == EXT2_BAD_INO || ino == EXT2_RESIZE_INO)
528 			continue;
529 
530 		/*
531 		 * Find all of the inodes which share blocks with this
532 		 * one.  First we find all of the duplicate blocks
533 		 * belonging to this inode, and then search each block
534 		 * get the list of inodes, and merge them together.
535 		 */
536 		for (s = p->cluster_list; s; s = s->next) {
537 			m = dict_lookup(&clstr_dict,
538 					INT_TO_VOIDPTR(s->cluster));
539 			if (!m)
540 				continue; /* Should never happen... */
541 			q = (struct dup_cluster *) dnode_get(m);
542 			if (q->num_bad > 1)
543 				file_ok = 0;
544 			if (check_if_fs_cluster(ctx, s->cluster)) {
545 				file_ok = 0;
546 				meta_data = 1;
547 			}
548 
549 			/*
550 			 * Add all inodes used by this block to the
551 			 * shared[] --- which is a unique list, so
552 			 * if an inode is already in shared[], don't
553 			 * add it again.
554 			 */
555 			for (r = q->inode_list; r; r = r->next) {
556 				if (r->inode == ino)
557 					continue;
558 				for (i = 0; i < shared_len; i++)
559 					if (shared[i] == r->inode)
560 						break;
561 				if (i == shared_len) {
562 					shared[shared_len++] = r->inode;
563 				}
564 			}
565 		}
566 
567 		/*
568 		 * Report the inode that we are working on
569 		 */
570 		pctx.inode = EXT2_INODE(&p->inode);
571 		pctx.ino = ino;
572 		pctx.dir = p->dir;
573 		pctx.blkcount = p->num_dupblocks;
574 		pctx.num = meta_data ? shared_len+1 : shared_len;
575 		fix_problem(ctx, PR_1D_DUP_FILE, &pctx);
576 		pctx.blkcount = 0;
577 		pctx.num = 0;
578 
579 		if (meta_data)
580 			fix_problem(ctx, PR_1D_SHARE_METADATA, &pctx);
581 
582 		for (i = 0; i < shared_len; i++) {
583 			m = dict_lookup(&ino_dict, INT_TO_VOIDPTR(shared[i]));
584 			if (!m)
585 				continue; /* should never happen */
586 			t = (struct dup_inode *) dnode_get(m);
587 			/*
588 			 * Report the inode that we are sharing with
589 			 */
590 			pctx.inode = EXT2_INODE(&t->inode);
591 			pctx.ino = shared[i];
592 			pctx.dir = t->dir;
593 			fix_problem(ctx, PR_1D_DUP_FILE_LIST, &pctx);
594 		}
595 		/*
596 		 * Even if the file shares blocks with itself, we still need to
597 		 * clone the blocks.
598 		 */
599 		if (file_ok && (meta_data ? shared_len+1 : shared_len) != 0) {
600 			fix_problem(ctx, PR_1D_DUP_BLOCKS_DEALT, &pctx);
601 			continue;
602 		}
603 		if ((ctx->options & E2F_OPT_UNSHARE_BLOCKS) ||
604                     fix_problem(ctx, PR_1D_CLONE_QUESTION, &pctx)) {
605 			pctx.errcode = clone_file(ctx, ino, p, block_buf);
606 			if (pctx.errcode)
607 				fix_problem(ctx, PR_1D_CLONE_ERROR, &pctx);
608 			else
609 				continue;
610 		}
611 		/*
612 		 * Note: When unsharing blocks, we don't prompt to delete
613 		 * files. If the clone operation fails than the unshare
614 		 * operation should fail too.
615 		 */
616 		if (!(ctx->options & E2F_OPT_UNSHARE_BLOCKS) &&
617                     fix_problem(ctx, PR_1D_DELETE_QUESTION, &pctx))
618 			delete_file(ctx, ino, p, block_buf);
619 		else
620 			ext2fs_unmark_valid(fs);
621 	}
622 	ext2fs_free_mem(&shared);
623 }
624 
625 /*
626  * Drop the refcount on the dup_block structure, and clear the entry
627  * in the block_dup_map if appropriate.
628  */
decrement_badcount(e2fsck_t ctx,blk64_t block,struct dup_cluster * p)629 static void decrement_badcount(e2fsck_t ctx, blk64_t block,
630 			       struct dup_cluster *p)
631 {
632 	p->num_bad--;
633 	if (p->num_bad <= 0 ||
634 	    (p->num_bad == 1 && !check_if_fs_block(ctx, block))) {
635 		if (check_if_fs_cluster(ctx, EXT2FS_B2C(ctx->fs, block)))
636 			return;
637 		ext2fs_unmark_block_bitmap2(ctx->block_dup_map, block);
638 	}
639 }
640 
delete_file_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)641 static int delete_file_block(ext2_filsys fs,
642 			     blk64_t	*block_nr,
643 			     e2_blkcnt_t blockcnt,
644 			     blk64_t ref_block EXT2FS_ATTR((unused)),
645 			     int ref_offset EXT2FS_ATTR((unused)),
646 			     void *priv_data)
647 {
648 	struct process_block_struct *pb;
649 	struct dup_cluster *p;
650 	dnode_t	*n;
651 	e2fsck_t ctx;
652 	blk64_t c, lc;
653 
654 	pb = (struct process_block_struct *) priv_data;
655 	ctx = pb->ctx;
656 
657 	if (*block_nr == 0)
658 		return 0;
659 
660 	c = EXT2FS_B2C(fs, *block_nr);
661 	lc = EXT2FS_B2C(fs, blockcnt);
662 	if (ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr)) {
663 		n = dict_lookup(&clstr_dict, INT_TO_VOIDPTR(c));
664 		if (n) {
665 			if (lc != pb->cur_cluster) {
666 				p = (struct dup_cluster *) dnode_get(n);
667 				decrement_badcount(ctx, *block_nr, p);
668 				pb->dup_blocks++;
669 			}
670 		} else
671 			com_err("delete_file_block", 0,
672 			    _("internal error: can't find dup_blk for %llu\n"),
673 				*block_nr);
674 	} else {
675 		if ((*block_nr % EXT2FS_CLUSTER_RATIO(ctx->fs)) == 0)
676 			ext2fs_block_alloc_stats2(fs, *block_nr, -1);
677 		pb->dup_blocks++;
678 	}
679 	pb->cur_cluster = lc;
680 
681 	return 0;
682 }
683 
delete_file(e2fsck_t ctx,ext2_ino_t ino,struct dup_inode * dp,char * block_buf)684 static void delete_file(e2fsck_t ctx, ext2_ino_t ino,
685 			struct dup_inode *dp, char* block_buf)
686 {
687 	ext2_filsys fs = ctx->fs;
688 	struct process_block_struct pb;
689 	struct problem_context	pctx;
690 	unsigned int		count;
691 
692 	clear_problem_context(&pctx);
693 	pctx.ino = pb.ino = ino;
694 	pb.dup_blocks = 0;
695 	pb.ctx = ctx;
696 	pctx.str = "delete_file";
697 	pb.cur_cluster = ~0;
698 
699 	if (ext2fs_inode_has_valid_blocks2(fs, EXT2_INODE(&dp->inode)))
700 		pctx.errcode = ext2fs_block_iterate3(fs, ino,
701 						     BLOCK_FLAG_READ_ONLY,
702 						     block_buf,
703 						     delete_file_block, &pb);
704 	if (pctx.errcode)
705 		fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
706 	if (ctx->inode_bad_map)
707 		ext2fs_unmark_inode_bitmap2(ctx->inode_bad_map, ino);
708 	ext2fs_inode_alloc_stats2(fs, ino, -1, LINUX_S_ISDIR(dp->inode.i_mode));
709 	quota_data_sub(ctx->qctx, &dp->inode, ino,
710 		       pb.dup_blocks * fs->blocksize);
711 	quota_data_inodes(ctx->qctx, &dp->inode, ino, -1);
712 
713 	/* Inode may have changed by block_iterate, so reread it */
714 	e2fsck_read_inode_full(ctx, ino, EXT2_INODE(&dp->inode),
715 			       sizeof(dp->inode), "delete_file");
716 	e2fsck_clear_inode(ctx, ino, EXT2_INODE(&dp->inode), 0, "delete_file");
717 	if (ext2fs_file_acl_block(fs, EXT2_INODE(&dp->inode)) &&
718 	    ext2fs_has_feature_xattr(fs->super)) {
719 		blk64_t file_acl_block = ext2fs_file_acl_block(fs,
720 						EXT2_INODE(&dp->inode));
721 
722 		count = 1;
723 		pctx.errcode = ext2fs_adjust_ea_refcount3(fs, file_acl_block,
724 					block_buf, -1, &count, ino);
725 		if (pctx.errcode == EXT2_ET_BAD_EA_BLOCK_NUM) {
726 			pctx.errcode = 0;
727 			count = 1;
728 		}
729 		if (pctx.errcode) {
730 			pctx.blk = file_acl_block;
731 			fix_problem(ctx, PR_1B_ADJ_EA_REFCOUNT, &pctx);
732 		}
733 		/*
734 		 * If the count is zero, then arrange to have the
735 		 * block deleted.  If the block is in the block_dup_map,
736 		 * also call delete_file_block since it will take care
737 		 * of keeping the accounting straight.
738 		 */
739 		if ((count == 0) ||
740 		    ext2fs_test_block_bitmap2(ctx->block_dup_map,
741 					      file_acl_block)) {
742 			delete_file_block(fs, &file_acl_block,
743 					  BLOCK_COUNT_EXTATTR, 0, 0, &pb);
744 			ext2fs_file_acl_block_set(fs, EXT2_INODE(&dp->inode),
745 						  file_acl_block);
746 			quota_data_sub(ctx->qctx, &dp->inode, ino,
747 				       fs->blocksize);
748 		}
749 	}
750 }
751 
752 struct clone_struct {
753 	errcode_t	errcode;
754 	blk64_t		dup_cluster;
755 	blk64_t		alloc_block;
756 	ext2_ino_t	dir, ino;
757 	char	*buf;
758 	e2fsck_t ctx;
759 	struct ext2_inode_large	*inode;
760 
761 	struct dup_cluster *save_dup_cluster;
762 	blk64_t save_blocknr;
763 };
764 
765 /*
766  * Decrement the bad count *after* we've shown that (a) we can allocate a
767  * replacement block and (b) remap the file blocks.  Unfortunately, there's no
768  * way to find out if the remap succeeded until either the next
769  * clone_file_block() call (an error when remapping the block after returning
770  * BLOCK_CHANGED will halt the iteration) or after block_iterate() returns.
771  * Otherwise, it's possible that we decrease the badcount once in preparation
772  * to remap, then the remap fails (either we can't find a replacement block or
773  * we have to split the extent tree and can't find a new extent block), so we
774  * delete the file, which decreases the badcount again.
775  */
deferred_dec_badcount(struct clone_struct * cs)776 static void deferred_dec_badcount(struct clone_struct *cs)
777 {
778 	if (!cs->save_dup_cluster)
779 		return;
780 	decrement_badcount(cs->ctx, cs->save_blocknr, cs->save_dup_cluster);
781 	cs->save_dup_cluster = NULL;
782 }
783 
clone_file_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)784 static int clone_file_block(ext2_filsys fs,
785 			    blk64_t	*block_nr,
786 			    e2_blkcnt_t blockcnt,
787 			    blk64_t ref_block EXT2FS_ATTR((unused)),
788 			    int ref_offset EXT2FS_ATTR((unused)),
789 			    void *priv_data)
790 {
791 	struct dup_cluster *p = NULL;
792 	blk64_t	new_block;
793 	errcode_t	retval;
794 	struct clone_struct *cs = (struct clone_struct *) priv_data;
795 	dnode_t *n;
796 	e2fsck_t ctx;
797 	blk64_t c;
798 	int is_meta = 0;
799 	int should_write = 1;
800 
801 	ctx = cs->ctx;
802 	deferred_dec_badcount(cs);
803 
804 	if (*block_nr == 0)
805 		return 0;
806 
807 	if (ext2fs_has_feature_shared_blocks(ctx->fs->super) &&
808 	    (ctx->options & E2F_OPT_UNSHARE_BLOCKS) &&
809 	    (ctx->options & E2F_OPT_NO))
810 		should_write = 0;
811 
812 	c = EXT2FS_B2C(fs, blockcnt);
813 	if (check_if_fs_cluster(ctx, EXT2FS_B2C(fs, *block_nr)))
814 		is_meta = 1;
815 
816 	if (c == cs->dup_cluster && cs->alloc_block) {
817 		new_block = cs->alloc_block;
818 		goto got_block;
819 	}
820 
821 	if (ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr)) {
822 		n = dict_lookup(&clstr_dict,
823 				INT_TO_VOIDPTR(EXT2FS_B2C(fs, *block_nr)));
824 		if (!n) {
825 			com_err("clone_file_block", 0,
826 			    _("internal error: can't find dup_blk for %llu\n"),
827 				*block_nr);
828 			return 0;
829 		}
830 
831 		p = (struct dup_cluster *) dnode_get(n);
832 
833 		cs->dup_cluster = c;
834 		/*
835 		 * Let's try an implied cluster allocation.  If we get the same
836 		 * cluster back, then we need to find a new block; otherwise,
837 		 * we're merely fixing the problem of one logical cluster being
838 		 * mapped to multiple physical clusters.
839 		 */
840 		new_block = 0;
841 		retval = ext2fs_map_cluster_block(fs, cs->ino,
842 						  EXT2_INODE(cs->inode),
843 						  blockcnt, &new_block);
844 		if (retval == 0 && new_block != 0 &&
845 		    EXT2FS_B2C(ctx->fs, new_block) !=
846 		    EXT2FS_B2C(ctx->fs, *block_nr))
847 			goto cluster_alloc_ok;
848 		retval = ext2fs_new_block2(fs, 0, ctx->block_found_map,
849 					   &new_block);
850 		if (retval) {
851 			cs->errcode = retval;
852 			return BLOCK_ABORT;
853 		}
854 		if (ext2fs_has_feature_shared_blocks(fs->super)) {
855 			/*
856 			 * Update the block stats so we don't get a prompt to fix block
857 			 * counts in the final pass.
858 			 */
859 			ext2fs_block_alloc_stats2(fs, new_block, +1);
860 		}
861 cluster_alloc_ok:
862 		cs->alloc_block = new_block;
863 
864 	got_block:
865 		new_block &= ~EXT2FS_CLUSTER_MASK(fs);
866 		new_block += EXT2FS_CLUSTER_MASK(fs) & blockcnt;
867 		if (cs->dir && (blockcnt >= 0)) {
868 			retval = ext2fs_set_dir_block2(fs->dblist,
869 					cs->dir, new_block, blockcnt);
870 			if (retval) {
871 				cs->errcode = retval;
872 				return BLOCK_ABORT;
873 			}
874 		}
875 #if 0
876  		printf("Cloning block #%lld from %llu to %llu\n",
877 		       blockcnt, *block_nr, new_block);
878 #endif
879 		retval = io_channel_read_blk64(fs->io, *block_nr, 1, cs->buf);
880 		if (retval) {
881 			cs->errcode = retval;
882 			return BLOCK_ABORT;
883 		}
884 		if (should_write) {
885 			retval = io_channel_write_blk64(fs->io, new_block, 1, cs->buf);
886 			if (retval) {
887 				cs->errcode = retval;
888 				return BLOCK_ABORT;
889 			}
890 		}
891 		cs->save_dup_cluster = (is_meta ? NULL : p);
892 		cs->save_blocknr = *block_nr;
893 		*block_nr = new_block;
894 		ext2fs_mark_block_bitmap2(ctx->block_found_map, new_block);
895 		ext2fs_mark_block_bitmap2(fs->block_map, new_block);
896 
897 		if (!should_write) {
898 			/* Don't try to change extent information; we want e2fsck to
899 			 * return success.
900 			 */
901 			return 0;
902 		}
903 		return BLOCK_CHANGED;
904 	}
905 	return 0;
906 }
907 
clone_file(e2fsck_t ctx,ext2_ino_t ino,struct dup_inode * dp,char * block_buf)908 static errcode_t clone_file(e2fsck_t ctx, ext2_ino_t ino,
909 			    struct dup_inode *dp, char* block_buf)
910 {
911 	ext2_filsys fs = ctx->fs;
912 	errcode_t	retval;
913 	struct clone_struct cs;
914 	struct problem_context	pctx;
915 	blk64_t		blk, new_blk;
916 	dnode_t		*n;
917 	struct inode_el	*ino_el;
918 	struct dup_cluster	*dc;
919 	struct dup_inode	*di;
920 
921 	clear_problem_context(&pctx);
922 	cs.errcode = 0;
923 	cs.dir = 0;
924 	cs.dup_cluster = ~0;
925 	cs.alloc_block = 0;
926 	cs.ctx = ctx;
927 	cs.ino = ino;
928 	cs.inode = &dp->inode;
929 	cs.save_dup_cluster = NULL;
930 	cs.save_blocknr = 0;
931 	retval = ext2fs_get_mem(fs->blocksize, &cs.buf);
932 	if (retval)
933 		return retval;
934 
935 	if (ext2fs_test_inode_bitmap2(ctx->inode_dir_map, ino))
936 		cs.dir = ino;
937 
938 	pctx.ino = ino;
939 	pctx.str = "clone_file";
940 	if (ext2fs_inode_has_valid_blocks2(fs, EXT2_INODE(&dp->inode)))
941 		pctx.errcode = ext2fs_block_iterate3(fs, ino, 0, block_buf,
942 						     clone_file_block, &cs);
943 	deferred_dec_badcount(&cs);
944 	ext2fs_mark_bb_dirty(fs);
945 	if (pctx.errcode) {
946 		fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
947 		retval = pctx.errcode;
948 		goto errout;
949 	}
950 	if (cs.errcode) {
951 		com_err("clone_file", cs.errcode, "%s",
952 			_("returned from clone_file_block"));
953 		retval = cs.errcode;
954 		goto errout;
955 	}
956 	/* The inode may have changed on disk, so we have to re-read it */
957 	e2fsck_read_inode_full(ctx, ino, EXT2_INODE(&dp->inode),
958 			       sizeof(dp->inode), "clone file EA");
959 	blk = ext2fs_file_acl_block(fs, EXT2_INODE(&dp->inode));
960 	new_blk = blk;
961 	if (blk && (clone_file_block(fs, &new_blk,
962 				     BLOCK_COUNT_EXTATTR, 0, 0, &cs) ==
963 		    BLOCK_CHANGED)) {
964 		ext2fs_file_acl_block_set(fs, EXT2_INODE(&dp->inode), new_blk);
965 		e2fsck_write_inode_full(ctx, ino, EXT2_INODE(&dp->inode),
966 					sizeof(dp->inode), "clone file EA");
967 		/*
968 		 * If we cloned the EA block, find all other inodes
969 		 * which referred to that EA block, and modify
970 		 * them to point to the new EA block.
971 		 */
972 		n = dict_lookup(&clstr_dict,
973 				INT_TO_VOIDPTR(EXT2FS_B2C(fs, blk)));
974 		if (!n) {
975 			com_err("clone_file", 0,
976 				_("internal error: couldn't lookup EA "
977 				  "block record for %llu"), blk);
978 			retval = 0; /* OK to stumble on... */
979 			goto errout;
980 		}
981 		dc = (struct dup_cluster *) dnode_get(n);
982 		for (ino_el = dc->inode_list; ino_el; ino_el = ino_el->next) {
983 			if (ino_el->inode == ino)
984 				continue;
985 			n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(ino_el->inode));
986 			if (!n) {
987 				com_err("clone_file", 0,
988 					_("internal error: couldn't lookup EA "
989 					  "inode record for %u"),
990 					ino_el->inode);
991 				retval = 0; /* OK to stumble on... */
992 				goto errout;
993 			}
994 			di = (struct dup_inode *) dnode_get(n);
995 			if (ext2fs_file_acl_block(fs,
996 					EXT2_INODE(&di->inode)) == blk) {
997 				ext2fs_file_acl_block_set(fs,
998 					EXT2_INODE(&di->inode),
999 					ext2fs_file_acl_block(fs, EXT2_INODE(&dp->inode)));
1000 				e2fsck_write_inode_full(ctx, ino_el->inode,
1001 					EXT2_INODE(&di->inode),
1002 					sizeof(di->inode), "clone file EA");
1003 				decrement_badcount(ctx, blk, dc);
1004 			}
1005 		}
1006 	}
1007 	retval = 0;
1008 errout:
1009 	ext2fs_free_mem(&cs.buf);
1010 	return retval;
1011 }
1012 
1013 /*
1014  * This routine returns 1 if a block overlaps with one of the superblocks,
1015  * group descriptors, inode bitmaps, or block bitmaps.
1016  */
check_if_fs_block(e2fsck_t ctx,blk64_t test_block)1017 static int check_if_fs_block(e2fsck_t ctx, blk64_t test_block)
1018 {
1019 	ext2_filsys fs = ctx->fs;
1020 	blk64_t	first_block;
1021 	dgrp_t	i;
1022 
1023 	first_block = fs->super->s_first_data_block;
1024 	for (i = 0; i < fs->group_desc_count; i++) {
1025 
1026 		/* Check superblocks/block group descriptors */
1027 		if (ext2fs_bg_has_super(fs, i)) {
1028 			if (test_block >= first_block &&
1029 			    (test_block <= first_block + fs->desc_blocks))
1030 				return 1;
1031 		}
1032 
1033 		/* Check the inode table */
1034 		if ((ext2fs_inode_table_loc(fs, i)) &&
1035 		    (test_block >= ext2fs_inode_table_loc(fs, i)) &&
1036 		    (test_block < (ext2fs_inode_table_loc(fs, i) +
1037 				   fs->inode_blocks_per_group)))
1038 			return 1;
1039 
1040 		/* Check the bitmap blocks */
1041 		if ((test_block == ext2fs_block_bitmap_loc(fs, i)) ||
1042 		    (test_block == ext2fs_inode_bitmap_loc(fs, i)))
1043 			return 1;
1044 
1045 		first_block += fs->super->s_blocks_per_group;
1046 	}
1047 	return 0;
1048 }
1049 
1050 /*
1051  * This routine returns 1 if a cluster overlaps with one of the superblocks,
1052  * group descriptors, inode bitmaps, or block bitmaps.
1053  */
check_if_fs_cluster(e2fsck_t ctx,blk64_t cluster)1054 static int check_if_fs_cluster(e2fsck_t ctx, blk64_t cluster)
1055 {
1056 	ext2_filsys fs = ctx->fs;
1057 	blk64_t	first_block;
1058 	dgrp_t	i;
1059 
1060 	first_block = fs->super->s_first_data_block;
1061 	for (i = 0; i < fs->group_desc_count; i++) {
1062 
1063 		/* Check superblocks/block group descriptors */
1064 		if (ext2fs_bg_has_super(fs, i)) {
1065 			if (cluster >= EXT2FS_B2C(fs, first_block) &&
1066 			    (cluster <= EXT2FS_B2C(fs, first_block +
1067 						   fs->desc_blocks)))
1068 				return 1;
1069 		}
1070 
1071 		/* Check the inode table */
1072 		if ((ext2fs_inode_table_loc(fs, i)) &&
1073 		    (cluster >= EXT2FS_B2C(fs,
1074 					   ext2fs_inode_table_loc(fs, i))) &&
1075 		    (cluster <= EXT2FS_B2C(fs,
1076 					   ext2fs_inode_table_loc(fs, i) +
1077 					   fs->inode_blocks_per_group - 1)))
1078 			return 1;
1079 
1080 		/* Check the bitmap blocks */
1081 		if ((cluster == EXT2FS_B2C(fs,
1082 					   ext2fs_block_bitmap_loc(fs, i))) ||
1083 		    (cluster == EXT2FS_B2C(fs,
1084 					   ext2fs_inode_bitmap_loc(fs, i))))
1085 			return 1;
1086 
1087 		first_block += fs->super->s_blocks_per_group;
1088 	}
1089 	return 0;
1090 }
1091