pass1b.c revision 86f3b6cf98a72c6dad0738e3af2512ddcbd49be9
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
41typedef 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
57struct cluster_el {
58	blk64_t	cluster;
59	struct cluster_el *next;
60};
61
62struct inode_el {
63	ext2_ino_t	inode;
64	struct inode_el *next;
65};
66
67struct 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 */
79struct dup_inode {
80	ext2_ino_t		dir;
81	int			num_dupblocks;
82	struct ext2_inode	inode;
83	struct cluster_el	*cluster_list;
84};
85
86static 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);
89static void delete_file(e2fsck_t ctx, ext2_ino_t ino,
90			struct dup_inode *dp, char *block_buf);
91static errcode_t clone_file(e2fsck_t ctx, ext2_ino_t ino,
92			    struct dup_inode *dp, char* block_buf);
93static int check_if_fs_block(e2fsck_t ctx, blk64_t test_block);
94static int check_if_fs_cluster(e2fsck_t ctx, blk64_t cluster);
95
96static void pass1b(e2fsck_t ctx, char *block_buf);
97static void pass1c(e2fsck_t ctx, char *block_buf);
98static void pass1d(e2fsck_t ctx, char *block_buf);
99
100static int dup_inode_count = 0;
101static int dup_inode_founddir = 0;
102
103static dict_t clstr_dict, ino_dict;
104
105static ext2fs_inode_bitmap inode_dup_map;
106
107static 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 */
120static void add_dupe(e2fsck_t ctx, ext2_ino_t ino, blk64_t cluster,
121		     struct ext2_inode *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 */
174static 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 */
192static 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 */
211void 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	/*
249	 * Time to free all of the accumulated data structures that we
250	 * don't need anymore.
251	 */
252	dict_free_nodes(&ino_dict);
253	dict_free_nodes(&clstr_dict);
254	ext2fs_free_inode_bitmap(inode_dup_map);
255}
256
257/*
258 * Scan the inodes looking for inodes that contain duplicate blocks.
259 */
260struct process_block_struct {
261	e2fsck_t	ctx;
262	ext2_ino_t	ino;
263	int		dup_blocks;
264	blk64_t		cur_cluster, phys_cluster;
265	blk64_t		last_blk;
266	struct ext2_inode *inode;
267	struct problem_context *pctx;
268};
269
270static void pass1b(e2fsck_t ctx, char *block_buf)
271{
272	ext2_filsys fs = ctx->fs;
273	ext2_ino_t ino = 0;
274	struct ext2_inode inode;
275	ext2_inode_scan	scan;
276	struct process_block_struct pb;
277	struct problem_context pctx;
278	problem_t op;
279
280	clear_problem_context(&pctx);
281
282	if (!(ctx->options & E2F_OPT_PREEN))
283		fix_problem(ctx, PR_1B_PASS_HEADER, &pctx);
284	pctx.errcode = ext2fs_open_inode_scan(fs, ctx->inode_buffer_blocks,
285					      &scan);
286	if (pctx.errcode) {
287		fix_problem(ctx, PR_1B_ISCAN_ERROR, &pctx);
288		ctx->flags |= E2F_FLAG_ABORT;
289		return;
290	}
291	ctx->stashed_inode = &inode;
292	pb.ctx = ctx;
293	pb.pctx = &pctx;
294	pctx.str = "pass1b";
295	while (1) {
296		if (ino % (fs->super->s_inodes_per_group * 4) == 1) {
297			if (e2fsck_mmp_update(fs))
298				fatal_error(ctx, 0);
299		}
300		pctx.errcode = ext2fs_get_next_inode(scan, &ino, &inode);
301		if (pctx.errcode == EXT2_ET_BAD_BLOCK_IN_INODE_TABLE)
302			continue;
303		if (pctx.errcode) {
304			pctx.ino = ino;
305			fix_problem(ctx, PR_1B_ISCAN_ERROR, &pctx);
306			ctx->flags |= E2F_FLAG_ABORT;
307			return;
308		}
309		if (!ino)
310			break;
311		pctx.ino = ctx->stashed_ino = ino;
312		if ((ino != EXT2_BAD_INO) &&
313		    !ext2fs_test_inode_bitmap2(ctx->inode_used_map, ino))
314			continue;
315
316		pb.ino = ino;
317		pb.dup_blocks = 0;
318		pb.inode = &inode;
319		pb.cur_cluster = ~0;
320		pb.phys_cluster = ~0;
321		pb.last_blk = 0;
322		pb.pctx->blk = pb.pctx->blk2 = 0;
323
324		if (ext2fs_inode_has_valid_blocks2(fs, &inode) ||
325		    (ino == EXT2_BAD_INO))
326			pctx.errcode = ext2fs_block_iterate3(fs, ino,
327					     BLOCK_FLAG_READ_ONLY, block_buf,
328					     process_pass1b_block, &pb);
329		/* If the feature is not set, attrs will be cleared later anyway */
330		if (ext2fs_has_feature_xattr(fs->super) &&
331		    ext2fs_file_acl_block(fs, &inode)) {
332			blk64_t blk = ext2fs_file_acl_block(fs, &inode);
333			process_pass1b_block(fs, &blk,
334					     BLOCK_COUNT_EXTATTR, 0, 0, &pb);
335			ext2fs_file_acl_block_set(fs, &inode, blk);
336		}
337		if (pb.dup_blocks) {
338			if (ino != EXT2_BAD_INO) {
339				op = pctx.blk == pctx.blk2 ?
340					PR_1B_DUP_BLOCK : PR_1B_DUP_RANGE;
341				fix_problem(ctx, op, pb.pctx);
342			}
343			end_problem_latch(ctx, PR_LATCH_DBLOCK);
344			if (ino >= EXT2_FIRST_INODE(fs->super) ||
345			    ino == EXT2_ROOT_INO)
346				dup_inode_count++;
347		}
348		if (pctx.errcode)
349			fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
350	}
351	ext2fs_close_inode_scan(scan);
352	e2fsck_use_inode_shortcuts(ctx, 0);
353}
354
355static int process_pass1b_block(ext2_filsys fs EXT2FS_ATTR((unused)),
356				blk64_t	*block_nr,
357				e2_blkcnt_t blockcnt,
358				blk64_t ref_blk EXT2FS_ATTR((unused)),
359				int ref_offset EXT2FS_ATTR((unused)),
360				void *priv_data)
361{
362	struct process_block_struct *p;
363	e2fsck_t ctx;
364	blk64_t	lc, pc;
365	problem_t op;
366
367	if (*block_nr == 0)
368		return 0;
369	p = (struct process_block_struct *) priv_data;
370	ctx = p->ctx;
371	lc = EXT2FS_B2C(fs, blockcnt);
372	pc = EXT2FS_B2C(fs, *block_nr);
373
374	if (!ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr))
375		goto finish;
376
377	/* OK, this is a duplicate block */
378	if (p->ino != EXT2_BAD_INO) {
379		if (p->last_blk + 1 != *block_nr) {
380			if (p->last_blk) {
381				op = p->pctx->blk == p->pctx->blk2 ?
382						PR_1B_DUP_BLOCK :
383						PR_1B_DUP_RANGE;
384				fix_problem(ctx, op, p->pctx);
385			}
386			p->pctx->blk = *block_nr;
387		}
388		p->pctx->blk2 = *block_nr;
389		p->last_blk = *block_nr;
390	}
391	p->dup_blocks++;
392	ext2fs_mark_inode_bitmap2(inode_dup_map, p->ino);
393
394	/*
395	 * Qualifications for submitting a block for duplicate processing:
396	 * It's an extent/indirect block (and has a negative logical offset);
397	 * we've crossed a logical cluster boundary; or the physical cluster
398	 * suddenly changed, which indicates that blocks in a logical cluster
399	 * are mapped to multiple physical clusters.
400	 */
401	if (blockcnt < 0 || lc != p->cur_cluster || pc != p->phys_cluster)
402		add_dupe(ctx, p->ino, EXT2FS_B2C(fs, *block_nr), p->inode);
403
404finish:
405	p->cur_cluster = lc;
406	p->phys_cluster = pc;
407	return 0;
408}
409
410/*
411 * Pass 1c: Scan directories for inodes with duplicate blocks.  This
412 * is used so that we can print pathnames when prompting the user for
413 * what to do.
414 */
415struct search_dir_struct {
416	int		count;
417	ext2_ino_t	first_inode;
418	ext2_ino_t	max_inode;
419};
420
421static int search_dirent_proc(ext2_ino_t dir, int entry,
422			      struct ext2_dir_entry *dirent,
423			      int offset EXT2FS_ATTR((unused)),
424			      int blocksize EXT2FS_ATTR((unused)),
425			      char *buf EXT2FS_ATTR((unused)),
426			      void *priv_data)
427{
428	struct search_dir_struct *sd;
429	struct dup_inode	*p;
430	dnode_t			*n;
431
432	sd = (struct search_dir_struct *) priv_data;
433
434	if (dirent->inode > sd->max_inode)
435		/* Should abort this inode, but not everything */
436		return 0;
437
438	if ((dirent->inode < sd->first_inode) || (entry < DIRENT_OTHER_FILE) ||
439	    !ext2fs_test_inode_bitmap2(inode_dup_map, dirent->inode))
440		return 0;
441
442	n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(dirent->inode));
443	if (!n)
444		return 0;
445	p = (struct dup_inode *) dnode_get(n);
446	if (!p->dir) {
447		p->dir = dir;
448		sd->count--;
449	}
450
451	return(sd->count ? 0 : DIRENT_ABORT);
452}
453
454
455static void pass1c(e2fsck_t ctx, char *block_buf)
456{
457	ext2_filsys fs = ctx->fs;
458	struct search_dir_struct sd;
459	struct problem_context pctx;
460
461	clear_problem_context(&pctx);
462
463	if (!(ctx->options & E2F_OPT_PREEN))
464		fix_problem(ctx, PR_1C_PASS_HEADER, &pctx);
465
466	/*
467	 * Search through all directories to translate inodes to names
468	 * (by searching for the containing directory for that inode.)
469	 */
470	sd.count = dup_inode_count - dup_inode_founddir;
471	sd.first_inode = EXT2_FIRST_INODE(fs->super);
472	sd.max_inode = fs->super->s_inodes_count;
473	ext2fs_dblist_dir_iterate(fs->dblist, 0, block_buf,
474				  search_dirent_proc, &sd);
475}
476
477static void pass1d(e2fsck_t ctx, char *block_buf)
478{
479	ext2_filsys fs = ctx->fs;
480	struct dup_inode	*p, *t;
481	struct dup_cluster	*q;
482	ext2_ino_t		*shared, ino;
483	int	shared_len;
484	int	i;
485	int	file_ok;
486	int	meta_data = 0;
487	struct problem_context pctx;
488	dnode_t	*n, *m;
489	struct cluster_el	*s;
490	struct inode_el *r;
491
492	clear_problem_context(&pctx);
493
494	if (!(ctx->options & E2F_OPT_PREEN))
495		fix_problem(ctx, PR_1D_PASS_HEADER, &pctx);
496	e2fsck_read_bitmaps(ctx);
497
498	pctx.num = dup_inode_count; /* dict_count(&ino_dict); */
499	fix_problem(ctx, PR_1D_NUM_DUP_INODES, &pctx);
500	shared = (ext2_ino_t *) e2fsck_allocate_memory(ctx,
501				sizeof(ext2_ino_t) * dict_count(&ino_dict),
502				"Shared inode list");
503	for (n = dict_first(&ino_dict); n; n = dict_next(&ino_dict, n)) {
504		p = (struct dup_inode *) dnode_get(n);
505		shared_len = 0;
506		file_ok = 1;
507		ino = (ext2_ino_t)VOIDPTR_TO_INT(dnode_getkey(n));
508		if (ino == EXT2_BAD_INO || ino == EXT2_RESIZE_INO)
509			continue;
510
511		/*
512		 * Find all of the inodes which share blocks with this
513		 * one.  First we find all of the duplicate blocks
514		 * belonging to this inode, and then search each block
515		 * get the list of inodes, and merge them together.
516		 */
517		for (s = p->cluster_list; s; s = s->next) {
518			m = dict_lookup(&clstr_dict,
519					INT_TO_VOIDPTR(s->cluster));
520			if (!m)
521				continue; /* Should never happen... */
522			q = (struct dup_cluster *) dnode_get(m);
523			if (q->num_bad > 1)
524				file_ok = 0;
525			if (check_if_fs_cluster(ctx, s->cluster)) {
526				file_ok = 0;
527				meta_data = 1;
528			}
529
530			/*
531			 * Add all inodes used by this block to the
532			 * shared[] --- which is a unique list, so
533			 * if an inode is already in shared[], don't
534			 * add it again.
535			 */
536			for (r = q->inode_list; r; r = r->next) {
537				if (r->inode == ino)
538					continue;
539				for (i = 0; i < shared_len; i++)
540					if (shared[i] == r->inode)
541						break;
542				if (i == shared_len) {
543					shared[shared_len++] = r->inode;
544				}
545			}
546		}
547
548		/*
549		 * Report the inode that we are working on
550		 */
551		pctx.inode = &p->inode;
552		pctx.ino = ino;
553		pctx.dir = p->dir;
554		pctx.blkcount = p->num_dupblocks;
555		pctx.num = meta_data ? shared_len+1 : shared_len;
556		fix_problem(ctx, PR_1D_DUP_FILE, &pctx);
557		pctx.blkcount = 0;
558		pctx.num = 0;
559
560		if (meta_data)
561			fix_problem(ctx, PR_1D_SHARE_METADATA, &pctx);
562
563		for (i = 0; i < shared_len; i++) {
564			m = dict_lookup(&ino_dict, INT_TO_VOIDPTR(shared[i]));
565			if (!m)
566				continue; /* should never happen */
567			t = (struct dup_inode *) dnode_get(m);
568			/*
569			 * Report the inode that we are sharing with
570			 */
571			pctx.inode = &t->inode;
572			pctx.ino = shared[i];
573			pctx.dir = t->dir;
574			fix_problem(ctx, PR_1D_DUP_FILE_LIST, &pctx);
575		}
576		/*
577		 * Even if the file shares blocks with itself, we still need to
578		 * clone the blocks.
579		 */
580		if (file_ok && (meta_data ? shared_len+1 : shared_len) != 0) {
581			fix_problem(ctx, PR_1D_DUP_BLOCKS_DEALT, &pctx);
582			continue;
583		}
584		if (fix_problem(ctx, PR_1D_CLONE_QUESTION, &pctx)) {
585			pctx.errcode = clone_file(ctx, ino, p, block_buf);
586			if (pctx.errcode)
587				fix_problem(ctx, PR_1D_CLONE_ERROR, &pctx);
588			else
589				continue;
590		}
591		if (fix_problem(ctx, PR_1D_DELETE_QUESTION, &pctx))
592			delete_file(ctx, ino, p, block_buf);
593		else
594			ext2fs_unmark_valid(fs);
595	}
596	ext2fs_free_mem(&shared);
597}
598
599/*
600 * Drop the refcount on the dup_block structure, and clear the entry
601 * in the block_dup_map if appropriate.
602 */
603static void decrement_badcount(e2fsck_t ctx, blk64_t block,
604			       struct dup_cluster *p)
605{
606	p->num_bad--;
607	if (p->num_bad <= 0 ||
608	    (p->num_bad == 1 && !check_if_fs_block(ctx, block))) {
609		if (check_if_fs_cluster(ctx, EXT2FS_B2C(ctx->fs, block)))
610			return;
611		ext2fs_unmark_block_bitmap2(ctx->block_dup_map, block);
612	}
613}
614
615static int delete_file_block(ext2_filsys fs,
616			     blk64_t	*block_nr,
617			     e2_blkcnt_t blockcnt,
618			     blk64_t ref_block EXT2FS_ATTR((unused)),
619			     int ref_offset EXT2FS_ATTR((unused)),
620			     void *priv_data)
621{
622	struct process_block_struct *pb;
623	struct dup_cluster *p;
624	dnode_t	*n;
625	e2fsck_t ctx;
626	blk64_t c, lc;
627
628	pb = (struct process_block_struct *) priv_data;
629	ctx = pb->ctx;
630
631	if (*block_nr == 0)
632		return 0;
633
634	c = EXT2FS_B2C(fs, *block_nr);
635	lc = EXT2FS_B2C(fs, blockcnt);
636	if (ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr)) {
637		n = dict_lookup(&clstr_dict, INT_TO_VOIDPTR(c));
638		if (n) {
639			p = (struct dup_cluster *) dnode_get(n);
640			if (lc != pb->cur_cluster)
641				decrement_badcount(ctx, *block_nr, p);
642		} else
643			com_err("delete_file_block", 0,
644			    _("internal error: can't find dup_blk for %llu\n"),
645				*block_nr);
646	} else {
647		if ((*block_nr % EXT2FS_CLUSTER_RATIO(ctx->fs)) == 0)
648			ext2fs_block_alloc_stats2(fs, *block_nr, -1);
649		pb->dup_blocks++;
650	}
651	pb->cur_cluster = lc;
652
653	return 0;
654}
655
656static void delete_file(e2fsck_t ctx, ext2_ino_t ino,
657			struct dup_inode *dp, char* block_buf)
658{
659	ext2_filsys fs = ctx->fs;
660	struct process_block_struct pb;
661	struct problem_context	pctx;
662	unsigned int		count;
663
664	clear_problem_context(&pctx);
665	pctx.ino = pb.ino = ino;
666	pb.dup_blocks = 0;
667	pb.ctx = ctx;
668	pctx.str = "delete_file";
669	pb.cur_cluster = ~0;
670
671	if (ext2fs_inode_has_valid_blocks2(fs, &dp->inode))
672		pctx.errcode = ext2fs_block_iterate3(fs, ino,
673						     BLOCK_FLAG_READ_ONLY,
674						     block_buf,
675						     delete_file_block, &pb);
676	if (pctx.errcode)
677		fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
678	if (ctx->inode_bad_map)
679		ext2fs_unmark_inode_bitmap2(ctx->inode_bad_map, ino);
680	ext2fs_inode_alloc_stats2(fs, ino, -1, LINUX_S_ISDIR(dp->inode.i_mode));
681	quota_data_sub(ctx->qctx, &dp->inode, ino,
682		       pb.dup_blocks * fs->blocksize);
683	quota_data_inodes(ctx->qctx, &dp->inode, ino, -1);
684
685	/* Inode may have changed by block_iterate, so reread it */
686	e2fsck_read_inode(ctx, ino, &dp->inode, "delete_file");
687	e2fsck_clear_inode(ctx, ino, &dp->inode, 0, "delete_file");
688	if (ext2fs_file_acl_block(fs, &dp->inode) &&
689	    ext2fs_has_feature_xattr(fs->super)) {
690		count = 1;
691		pctx.errcode = ext2fs_adjust_ea_refcount3(fs,
692					ext2fs_file_acl_block(fs, &dp->inode),
693					block_buf, -1, &count, ino);
694		if (pctx.errcode == EXT2_ET_BAD_EA_BLOCK_NUM) {
695			pctx.errcode = 0;
696			count = 1;
697		}
698		if (pctx.errcode) {
699			pctx.blk = ext2fs_file_acl_block(fs, &dp->inode);
700			fix_problem(ctx, PR_1B_ADJ_EA_REFCOUNT, &pctx);
701		}
702		/*
703		 * If the count is zero, then arrange to have the
704		 * block deleted.  If the block is in the block_dup_map,
705		 * also call delete_file_block since it will take care
706		 * of keeping the accounting straight.
707		 */
708		if ((count == 0) ||
709		    ext2fs_test_block_bitmap2(ctx->block_dup_map,
710					ext2fs_file_acl_block(fs, &dp->inode))) {
711			blk64_t blk = ext2fs_file_acl_block(fs, &dp->inode);
712			delete_file_block(fs, &blk,
713					  BLOCK_COUNT_EXTATTR, 0, 0, &pb);
714			ext2fs_file_acl_block_set(fs, &dp->inode, blk);
715			quota_data_sub(ctx->qctx, &dp->inode, ino, fs->blocksize);
716		}
717	}
718}
719
720struct clone_struct {
721	errcode_t	errcode;
722	blk64_t		dup_cluster;
723	blk64_t		alloc_block;
724	ext2_ino_t	dir, ino;
725	char	*buf;
726	e2fsck_t ctx;
727	struct ext2_inode	*inode;
728
729	struct dup_cluster *save_dup_cluster;
730	blk64_t save_blocknr;
731};
732
733/*
734 * Decrement the bad count *after* we've shown that (a) we can allocate a
735 * replacement block and (b) remap the file blocks.  Unfortunately, there's no
736 * way to find out if the remap succeeded until either the next
737 * clone_file_block() call (an error when remapping the block after returning
738 * BLOCK_CHANGED will halt the iteration) or after block_iterate() returns.
739 * Otherwise, it's possible that we decrease the badcount once in preparation
740 * to remap, then the remap fails (either we can't find a replacement block or
741 * we have to split the extent tree and can't find a new extent block), so we
742 * delete the file, which decreases the badcount again.
743 */
744static void deferred_dec_badcount(struct clone_struct *cs)
745{
746	if (!cs->save_dup_cluster)
747		return;
748	decrement_badcount(cs->ctx, cs->save_blocknr, cs->save_dup_cluster);
749	cs->save_dup_cluster = NULL;
750}
751
752static int clone_file_block(ext2_filsys fs,
753			    blk64_t	*block_nr,
754			    e2_blkcnt_t blockcnt,
755			    blk64_t ref_block EXT2FS_ATTR((unused)),
756			    int ref_offset EXT2FS_ATTR((unused)),
757			    void *priv_data)
758{
759	struct dup_cluster *p = NULL;
760	blk64_t	new_block;
761	errcode_t	retval;
762	struct clone_struct *cs = (struct clone_struct *) priv_data;
763	dnode_t *n;
764	e2fsck_t ctx;
765	blk64_t c;
766	int is_meta = 0;
767
768	ctx = cs->ctx;
769	deferred_dec_badcount(cs);
770
771	if (*block_nr == 0)
772		return 0;
773
774	c = EXT2FS_B2C(fs, blockcnt);
775	if (check_if_fs_cluster(ctx, EXT2FS_B2C(fs, *block_nr)))
776		is_meta = 1;
777
778	if (c == cs->dup_cluster && cs->alloc_block) {
779		new_block = cs->alloc_block;
780		goto got_block;
781	}
782
783	if (ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr)) {
784		n = dict_lookup(&clstr_dict,
785				INT_TO_VOIDPTR(EXT2FS_B2C(fs, *block_nr)));
786		if (!n) {
787			com_err("clone_file_block", 0,
788			    _("internal error: can't find dup_blk for %llu\n"),
789				*block_nr);
790			return 0;
791		}
792
793		p = (struct dup_cluster *) dnode_get(n);
794
795		cs->dup_cluster = c;
796		/*
797		 * Let's try an implied cluster allocation.  If we get the same
798		 * cluster back, then we need to find a new block; otherwise,
799		 * we're merely fixing the problem of one logical cluster being
800		 * mapped to multiple physical clusters.
801		 */
802		new_block = 0;
803		retval = ext2fs_map_cluster_block(fs, cs->ino, cs->inode,
804						  blockcnt, &new_block);
805		if (retval == 0 && new_block != 0 &&
806		    EXT2FS_B2C(ctx->fs, new_block) !=
807		    EXT2FS_B2C(ctx->fs, *block_nr))
808			goto cluster_alloc_ok;
809		retval = ext2fs_new_block2(fs, 0, ctx->block_found_map,
810					   &new_block);
811		if (retval) {
812			cs->errcode = retval;
813			return BLOCK_ABORT;
814		}
815cluster_alloc_ok:
816		cs->alloc_block = new_block;
817
818	got_block:
819		new_block &= ~EXT2FS_CLUSTER_MASK(fs);
820		new_block += EXT2FS_CLUSTER_MASK(fs) & blockcnt;
821		if (cs->dir && (blockcnt >= 0)) {
822			retval = ext2fs_set_dir_block2(fs->dblist,
823					cs->dir, new_block, blockcnt);
824			if (retval) {
825				cs->errcode = retval;
826				return BLOCK_ABORT;
827			}
828		}
829#if 0
830 		printf("Cloning block #%lld from %llu to %llu\n",
831		       blockcnt, *block_nr, new_block);
832#endif
833		retval = io_channel_read_blk64(fs->io, *block_nr, 1, cs->buf);
834		if (retval) {
835			cs->errcode = retval;
836			return BLOCK_ABORT;
837		}
838		retval = io_channel_write_blk64(fs->io, new_block, 1, cs->buf);
839		if (retval) {
840			cs->errcode = retval;
841			return BLOCK_ABORT;
842		}
843		cs->save_dup_cluster = (is_meta ? NULL : p);
844		cs->save_blocknr = *block_nr;
845		*block_nr = new_block;
846		ext2fs_mark_block_bitmap2(ctx->block_found_map, new_block);
847		ext2fs_mark_block_bitmap2(fs->block_map, new_block);
848		return BLOCK_CHANGED;
849	}
850	return 0;
851}
852
853static errcode_t clone_file(e2fsck_t ctx, ext2_ino_t ino,
854			    struct dup_inode *dp, char* block_buf)
855{
856	ext2_filsys fs = ctx->fs;
857	errcode_t	retval;
858	struct clone_struct cs;
859	struct problem_context	pctx;
860	blk64_t		blk, new_blk;
861	dnode_t		*n;
862	struct inode_el	*ino_el;
863	struct dup_cluster	*dc;
864	struct dup_inode	*di;
865
866	clear_problem_context(&pctx);
867	cs.errcode = 0;
868	cs.dir = 0;
869	cs.dup_cluster = ~0;
870	cs.alloc_block = 0;
871	cs.ctx = ctx;
872	cs.ino = ino;
873	cs.inode = &dp->inode;
874	cs.save_dup_cluster = NULL;
875	cs.save_blocknr = 0;
876	retval = ext2fs_get_mem(fs->blocksize, &cs.buf);
877	if (retval)
878		return retval;
879
880	if (ext2fs_test_inode_bitmap2(ctx->inode_dir_map, ino))
881		cs.dir = ino;
882
883	pctx.ino = ino;
884	pctx.str = "clone_file";
885	if (ext2fs_inode_has_valid_blocks2(fs, &dp->inode))
886		pctx.errcode = ext2fs_block_iterate3(fs, ino, 0, block_buf,
887						     clone_file_block, &cs);
888	deferred_dec_badcount(&cs);
889	ext2fs_mark_bb_dirty(fs);
890	if (pctx.errcode) {
891		fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
892		retval = pctx.errcode;
893		goto errout;
894	}
895	if (cs.errcode) {
896		com_err("clone_file", cs.errcode, "%s",
897			_("returned from clone_file_block"));
898		retval = cs.errcode;
899		goto errout;
900	}
901	/* The inode may have changed on disk, so we have to re-read it */
902	e2fsck_read_inode(ctx, ino, &dp->inode, "clone file EA");
903	blk = ext2fs_file_acl_block(fs, &dp->inode);
904	new_blk = blk;
905	if (blk && (clone_file_block(fs, &new_blk,
906				     BLOCK_COUNT_EXTATTR, 0, 0, &cs) ==
907		    BLOCK_CHANGED)) {
908		ext2fs_file_acl_block_set(fs, &dp->inode, new_blk);
909		e2fsck_write_inode(ctx, ino, &dp->inode, "clone file EA");
910		/*
911		 * If we cloned the EA block, find all other inodes
912		 * which refered to that EA block, and modify
913		 * them to point to the new EA block.
914		 */
915		n = dict_lookup(&clstr_dict,
916				INT_TO_VOIDPTR(EXT2FS_B2C(fs, blk)));
917		if (!n) {
918			com_err("clone_file", 0,
919				_("internal error: couldn't lookup EA "
920				  "block record for %llu"), blk);
921			retval = 0; /* OK to stumble on... */
922			goto errout;
923		}
924		dc = (struct dup_cluster *) dnode_get(n);
925		for (ino_el = dc->inode_list; ino_el; ino_el = ino_el->next) {
926			if (ino_el->inode == ino)
927				continue;
928			n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(ino_el->inode));
929			if (!n) {
930				com_err("clone_file", 0,
931					_("internal error: couldn't lookup EA "
932					  "inode record for %u"),
933					ino_el->inode);
934				retval = 0; /* OK to stumble on... */
935				goto errout;
936			}
937			di = (struct dup_inode *) dnode_get(n);
938			if (ext2fs_file_acl_block(fs, &di->inode) == blk) {
939				ext2fs_file_acl_block_set(fs, &di->inode,
940					ext2fs_file_acl_block(fs, &dp->inode));
941				e2fsck_write_inode(ctx, ino_el->inode,
942					   &di->inode, "clone file EA");
943				decrement_badcount(ctx, blk, dc);
944			}
945		}
946	}
947	retval = 0;
948errout:
949	ext2fs_free_mem(&cs.buf);
950	return retval;
951}
952
953/*
954 * This routine returns 1 if a block overlaps with one of the superblocks,
955 * group descriptors, inode bitmaps, or block bitmaps.
956 */
957static int check_if_fs_block(e2fsck_t ctx, blk64_t test_block)
958{
959	ext2_filsys fs = ctx->fs;
960	blk64_t	first_block;
961	dgrp_t	i;
962
963	first_block = fs->super->s_first_data_block;
964	for (i = 0; i < fs->group_desc_count; i++) {
965
966		/* Check superblocks/block group descriptors */
967		if (ext2fs_bg_has_super(fs, i)) {
968			if (test_block >= first_block &&
969			    (test_block <= first_block + fs->desc_blocks))
970				return 1;
971		}
972
973		/* Check the inode table */
974		if ((ext2fs_inode_table_loc(fs, i)) &&
975		    (test_block >= ext2fs_inode_table_loc(fs, i)) &&
976		    (test_block < (ext2fs_inode_table_loc(fs, i) +
977				   fs->inode_blocks_per_group)))
978			return 1;
979
980		/* Check the bitmap blocks */
981		if ((test_block == ext2fs_block_bitmap_loc(fs, i)) ||
982		    (test_block == ext2fs_inode_bitmap_loc(fs, i)))
983			return 1;
984
985		first_block += fs->super->s_blocks_per_group;
986	}
987	return 0;
988}
989
990/*
991 * This routine returns 1 if a cluster overlaps with one of the superblocks,
992 * group descriptors, inode bitmaps, or block bitmaps.
993 */
994static int check_if_fs_cluster(e2fsck_t ctx, blk64_t cluster)
995{
996	ext2_filsys fs = ctx->fs;
997	blk64_t	first_block;
998	dgrp_t	i;
999
1000	first_block = fs->super->s_first_data_block;
1001	for (i = 0; i < fs->group_desc_count; i++) {
1002
1003		/* Check superblocks/block group descriptors */
1004		if (ext2fs_bg_has_super(fs, i)) {
1005			if (cluster >= EXT2FS_B2C(fs, first_block) &&
1006			    (cluster <= EXT2FS_B2C(fs, first_block +
1007						   fs->desc_blocks)))
1008				return 1;
1009		}
1010
1011		/* Check the inode table */
1012		if ((ext2fs_inode_table_loc(fs, i)) &&
1013		    (cluster >= EXT2FS_B2C(fs,
1014					   ext2fs_inode_table_loc(fs, i))) &&
1015		    (cluster <= EXT2FS_B2C(fs,
1016					   ext2fs_inode_table_loc(fs, i) +
1017					   fs->inode_blocks_per_group - 1)))
1018			return 1;
1019
1020		/* Check the bitmap blocks */
1021		if ((cluster == EXT2FS_B2C(fs,
1022					   ext2fs_block_bitmap_loc(fs, i))) ||
1023		    (cluster == EXT2FS_B2C(fs,
1024					   ext2fs_inode_bitmap_loc(fs, i))))
1025			return 1;
1026
1027		first_block += fs->super->s_blocks_per_group;
1028	}
1029	return 0;
1030}
1031