pass1b.c revision 0f5eba7501f467f757792ee449d16c9259b994fd
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 "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 int 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 = ext2fs_allocate_inode_bitmap(fs,
222		      _("multiply claimed inode map"), &inode_dup_map);
223	if (pctx.errcode) {
224		fix_problem(ctx, PR_1B_ALLOCATE_IBITMAP_ERROR, &pctx);
225		ctx->flags |= E2F_FLAG_ABORT;
226		return;
227	}
228
229	dict_init(&ino_dict, DICTCOUNT_T_MAX, dict_int_cmp);
230	dict_init(&clstr_dict, DICTCOUNT_T_MAX, dict_int_cmp);
231	dict_set_allocator(&ino_dict, NULL, inode_dnode_free, NULL);
232	dict_set_allocator(&clstr_dict, NULL, cluster_dnode_free, NULL);
233
234	init_resource_track(&rtrack, ctx->fs->io);
235	pass1b(ctx, block_buf);
236	print_resource_track(ctx, "Pass 1b", &rtrack, ctx->fs->io);
237
238	init_resource_track(&rtrack, ctx->fs->io);
239	pass1c(ctx, block_buf);
240	print_resource_track(ctx, "Pass 1c", &rtrack, ctx->fs->io);
241
242	init_resource_track(&rtrack, ctx->fs->io);
243	pass1d(ctx, block_buf);
244	print_resource_track(ctx, "Pass 1d", &rtrack, ctx->fs->io);
245
246	/*
247	 * Time to free all of the accumulated data structures that we
248	 * don't need anymore.
249	 */
250	dict_free_nodes(&ino_dict);
251	dict_free_nodes(&clstr_dict);
252	ext2fs_free_inode_bitmap(inode_dup_map);
253}
254
255/*
256 * Scan the inodes looking for inodes that contain duplicate blocks.
257 */
258struct process_block_struct {
259	e2fsck_t	ctx;
260	ext2_ino_t	ino;
261	int		dup_blocks;
262	struct ext2_inode *inode;
263	struct problem_context *pctx;
264};
265
266static void pass1b(e2fsck_t ctx, char *block_buf)
267{
268	ext2_filsys fs = ctx->fs;
269	ext2_ino_t ino;
270	struct ext2_inode inode;
271	ext2_inode_scan	scan;
272	struct process_block_struct pb;
273	struct problem_context pctx;
274
275	clear_problem_context(&pctx);
276
277	if (!(ctx->options & E2F_OPT_PREEN))
278		fix_problem(ctx, PR_1B_PASS_HEADER, &pctx);
279	pctx.errcode = ext2fs_open_inode_scan(fs, ctx->inode_buffer_blocks,
280					      &scan);
281	if (pctx.errcode) {
282		fix_problem(ctx, PR_1B_ISCAN_ERROR, &pctx);
283		ctx->flags |= E2F_FLAG_ABORT;
284		return;
285	}
286	ctx->stashed_inode = &inode;
287	pb.ctx = ctx;
288	pb.pctx = &pctx;
289	pctx.str = "pass1b";
290	while (1) {
291		if (ino % (fs->super->s_inodes_per_group * 4) == 1) {
292			if (e2fsck_mmp_update(fs))
293				fatal_error(ctx, 0);
294		}
295		pctx.errcode = ext2fs_get_next_inode(scan, &ino, &inode);
296		if (pctx.errcode == EXT2_ET_BAD_BLOCK_IN_INODE_TABLE)
297			continue;
298		if (pctx.errcode) {
299			fix_problem(ctx, PR_1B_ISCAN_ERROR, &pctx);
300			ctx->flags |= E2F_FLAG_ABORT;
301			return;
302		}
303		if (!ino)
304			break;
305		pctx.ino = ctx->stashed_ino = ino;
306		if ((ino != EXT2_BAD_INO) &&
307		    !ext2fs_test_inode_bitmap2(ctx->inode_used_map, ino))
308			continue;
309
310		pb.ino = ino;
311		pb.dup_blocks = 0;
312		pb.inode = &inode;
313
314		if (ext2fs_inode_has_valid_blocks(&inode) ||
315		    (ino == EXT2_BAD_INO))
316			pctx.errcode = ext2fs_block_iterate3(fs, ino,
317					     BLOCK_FLAG_READ_ONLY, block_buf,
318					     process_pass1b_block, &pb);
319		/* If the feature is not set, attrs will be cleared later anyway */
320		if ((fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_EXT_ATTR) &&
321		    ext2fs_file_acl_block(&inode)) {
322			blk64_t blk = ext2fs_file_acl_block(&inode);
323			process_pass1b_block(fs, &blk,
324					     BLOCK_COUNT_EXTATTR, 0, 0, &pb);
325			ext2fs_file_acl_block_set(&inode, blk);
326		}
327		if (pb.dup_blocks) {
328			end_problem_latch(ctx, PR_LATCH_DBLOCK);
329			if (ino >= EXT2_FIRST_INODE(fs->super) ||
330			    ino == EXT2_ROOT_INO)
331				dup_inode_count++;
332		}
333		if (pctx.errcode)
334			fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
335	}
336	ext2fs_close_inode_scan(scan);
337	e2fsck_use_inode_shortcuts(ctx, 0);
338}
339
340static int process_pass1b_block(ext2_filsys fs EXT2FS_ATTR((unused)),
341				blk64_t	*block_nr,
342				e2_blkcnt_t blockcnt EXT2FS_ATTR((unused)),
343				blk64_t ref_blk EXT2FS_ATTR((unused)),
344				int ref_offset EXT2FS_ATTR((unused)),
345				void *priv_data)
346{
347	struct process_block_struct *p;
348	e2fsck_t ctx;
349
350	if (HOLE_BLKADDR(*block_nr))
351		return 0;
352	p = (struct process_block_struct *) priv_data;
353	ctx = p->ctx;
354
355	if (!ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr))
356		return 0;
357
358	/* OK, this is a duplicate block */
359	if (p->ino != EXT2_BAD_INO) {
360		p->pctx->blk = *block_nr;
361		fix_problem(ctx, PR_1B_DUP_BLOCK, p->pctx);
362	}
363	p->dup_blocks++;
364	ext2fs_mark_inode_bitmap2(inode_dup_map, p->ino);
365
366	add_dupe(ctx, p->ino, EXT2FS_B2C(fs, *block_nr), p->inode);
367
368	return 0;
369}
370
371/*
372 * Pass 1c: Scan directories for inodes with duplicate blocks.  This
373 * is used so that we can print pathnames when prompting the user for
374 * what to do.
375 */
376struct search_dir_struct {
377	int		count;
378	ext2_ino_t	first_inode;
379	ext2_ino_t	max_inode;
380};
381
382static int search_dirent_proc(ext2_ino_t dir, int entry,
383			      struct ext2_dir_entry *dirent,
384			      int offset EXT2FS_ATTR((unused)),
385			      int blocksize EXT2FS_ATTR((unused)),
386			      char *buf EXT2FS_ATTR((unused)),
387			      void *priv_data)
388{
389	struct search_dir_struct *sd;
390	struct dup_inode	*p;
391	dnode_t			*n;
392
393	sd = (struct search_dir_struct *) priv_data;
394
395	if (dirent->inode > sd->max_inode)
396		/* Should abort this inode, but not everything */
397		return 0;
398
399	if ((dirent->inode < sd->first_inode) || (entry < DIRENT_OTHER_FILE) ||
400	    !ext2fs_test_inode_bitmap2(inode_dup_map, dirent->inode))
401		return 0;
402
403	n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(dirent->inode));
404	if (!n)
405		return 0;
406	p = (struct dup_inode *) dnode_get(n);
407	if (!p->dir) {
408		p->dir = dir;
409		sd->count--;
410	}
411
412	return(sd->count ? 0 : DIRENT_ABORT);
413}
414
415
416static void pass1c(e2fsck_t ctx, char *block_buf)
417{
418	ext2_filsys fs = ctx->fs;
419	struct search_dir_struct sd;
420	struct problem_context pctx;
421
422	clear_problem_context(&pctx);
423
424	if (!(ctx->options & E2F_OPT_PREEN))
425		fix_problem(ctx, PR_1C_PASS_HEADER, &pctx);
426
427	/*
428	 * Search through all directories to translate inodes to names
429	 * (by searching for the containing directory for that inode.)
430	 */
431	sd.count = dup_inode_count - dup_inode_founddir;
432	sd.first_inode = EXT2_FIRST_INODE(fs->super);
433	sd.max_inode = fs->super->s_inodes_count;
434	ext2fs_dblist_dir_iterate(fs->dblist, 0, block_buf,
435				  search_dirent_proc, &sd);
436}
437
438static void pass1d(e2fsck_t ctx, char *block_buf)
439{
440	ext2_filsys fs = ctx->fs;
441	struct dup_inode	*p, *t;
442	struct dup_cluster	*q;
443	ext2_ino_t		*shared, ino;
444	int	shared_len;
445	int	i;
446	int	file_ok;
447	int	meta_data = 0;
448	struct problem_context pctx;
449	dnode_t	*n, *m;
450	struct cluster_el	*s;
451	struct inode_el *r;
452
453	clear_problem_context(&pctx);
454
455	if (!(ctx->options & E2F_OPT_PREEN))
456		fix_problem(ctx, PR_1D_PASS_HEADER, &pctx);
457	e2fsck_read_bitmaps(ctx);
458
459	pctx.num = dup_inode_count; /* dict_count(&ino_dict); */
460	fix_problem(ctx, PR_1D_NUM_DUP_INODES, &pctx);
461	shared = (ext2_ino_t *) e2fsck_allocate_memory(ctx,
462				sizeof(ext2_ino_t) * dict_count(&ino_dict),
463				"Shared inode list");
464	for (n = dict_first(&ino_dict); n; n = dict_next(&ino_dict, n)) {
465		p = (struct dup_inode *) dnode_get(n);
466		shared_len = 0;
467		file_ok = 1;
468		ino = (ext2_ino_t)VOIDPTR_TO_INT(dnode_getkey(n));
469		if (ino == EXT2_BAD_INO || ino == EXT2_RESIZE_INO)
470			continue;
471
472		/*
473		 * Find all of the inodes which share blocks with this
474		 * one.  First we find all of the duplicate blocks
475		 * belonging to this inode, and then search each block
476		 * get the list of inodes, and merge them together.
477		 */
478		for (s = p->cluster_list; s; s = s->next) {
479			m = dict_lookup(&clstr_dict,
480					INT_TO_VOIDPTR(s->cluster));
481			if (!m)
482				continue; /* Should never happen... */
483			q = (struct dup_cluster *) dnode_get(m);
484			if (q->num_bad > 1)
485				file_ok = 0;
486			if (check_if_fs_cluster(ctx, s->cluster)) {
487				file_ok = 0;
488				meta_data = 1;
489			}
490
491			/*
492			 * Add all inodes used by this block to the
493			 * shared[] --- which is a unique list, so
494			 * if an inode is already in shared[], don't
495			 * add it again.
496			 */
497			for (r = q->inode_list; r; r = r->next) {
498				if (r->inode == ino)
499					continue;
500				for (i = 0; i < shared_len; i++)
501					if (shared[i] == r->inode)
502						break;
503				if (i == shared_len) {
504					shared[shared_len++] = r->inode;
505				}
506			}
507		}
508
509		/*
510		 * Report the inode that we are working on
511		 */
512		pctx.inode = &p->inode;
513		pctx.ino = ino;
514		pctx.dir = p->dir;
515		pctx.blkcount = p->num_dupblocks;
516		pctx.num = meta_data ? shared_len+1 : shared_len;
517		fix_problem(ctx, PR_1D_DUP_FILE, &pctx);
518		pctx.blkcount = 0;
519		pctx.num = 0;
520
521		if (meta_data)
522			fix_problem(ctx, PR_1D_SHARE_METADATA, &pctx);
523
524		for (i = 0; i < shared_len; i++) {
525			m = dict_lookup(&ino_dict, INT_TO_VOIDPTR(shared[i]));
526			if (!m)
527				continue; /* should never happen */
528			t = (struct dup_inode *) dnode_get(m);
529			/*
530			 * Report the inode that we are sharing with
531			 */
532			pctx.inode = &t->inode;
533			pctx.ino = shared[i];
534			pctx.dir = t->dir;
535			fix_problem(ctx, PR_1D_DUP_FILE_LIST, &pctx);
536		}
537		if (file_ok) {
538			fix_problem(ctx, PR_1D_DUP_BLOCKS_DEALT, &pctx);
539			continue;
540		}
541		if (fix_problem(ctx, PR_1D_CLONE_QUESTION, &pctx)) {
542			pctx.errcode = clone_file(ctx, ino, p, block_buf);
543			if (pctx.errcode)
544				fix_problem(ctx, PR_1D_CLONE_ERROR, &pctx);
545			else
546				continue;
547		}
548		if (fix_problem(ctx, PR_1D_DELETE_QUESTION, &pctx))
549			delete_file(ctx, ino, p, block_buf);
550		else
551			ext2fs_unmark_valid(fs);
552	}
553	ext2fs_free_mem(&shared);
554}
555
556/*
557 * Drop the refcount on the dup_block structure, and clear the entry
558 * in the block_dup_map if appropriate.
559 */
560static void decrement_badcount(e2fsck_t ctx, blk64_t block,
561			       struct dup_cluster *p)
562{
563	p->num_bad--;
564	if (p->num_bad <= 0 ||
565	    (p->num_bad == 1 && !check_if_fs_block(ctx, block))) {
566		if (check_if_fs_cluster(ctx, EXT2FS_B2C(ctx->fs, block)))
567			return;
568		ext2fs_unmark_block_bitmap2(ctx->block_dup_map, block);
569	}
570}
571
572static int delete_file_block(ext2_filsys fs,
573			     blk64_t	*block_nr,
574			     e2_blkcnt_t blockcnt EXT2FS_ATTR((unused)),
575			     blk64_t ref_block EXT2FS_ATTR((unused)),
576			     int ref_offset EXT2FS_ATTR((unused)),
577			     void *priv_data)
578{
579	struct process_block_struct *pb;
580	struct dup_cluster *p;
581	dnode_t	*n;
582	e2fsck_t ctx;
583	blk64_t c;
584
585	pb = (struct process_block_struct *) priv_data;
586	ctx = pb->ctx;
587
588	if (HOLE_BLKADDR(*block_nr))
589		return 0;
590
591	c = EXT2FS_B2C(fs, *block_nr);
592	if (ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr)) {
593		n = dict_lookup(&clstr_dict, INT_TO_VOIDPTR(c));
594		if (n) {
595			p = (struct dup_cluster *) dnode_get(n);
596			decrement_badcount(ctx, *block_nr, p);
597		} else
598			com_err("delete_file_block", 0,
599			    _("internal error: can't find dup_blk for %llu\n"),
600				*block_nr);
601	} else {
602		ext2fs_unmark_block_bitmap2(ctx->block_found_map, *block_nr);
603		ext2fs_block_alloc_stats2(fs, *block_nr, -1);
604		pb->dup_blocks++;
605	}
606
607	return 0;
608}
609
610static void delete_file(e2fsck_t ctx, ext2_ino_t ino,
611			struct dup_inode *dp, char* block_buf)
612{
613	ext2_filsys fs = ctx->fs;
614	struct process_block_struct pb;
615	struct ext2_inode	inode;
616	struct problem_context	pctx;
617	unsigned int		count;
618
619	clear_problem_context(&pctx);
620	pctx.ino = pb.ino = ino;
621	pb.dup_blocks = 0;
622	pb.ctx = ctx;
623	pctx.str = "delete_file";
624
625	e2fsck_read_inode(ctx, ino, &inode, "delete_file");
626	if (ext2fs_inode_has_valid_blocks(&inode))
627		pctx.errcode = ext2fs_block_iterate3(fs, ino, BLOCK_FLAG_READ_ONLY,
628						     block_buf, delete_file_block, &pb);
629	if (pctx.errcode)
630		fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
631	if (ctx->inode_bad_map)
632		ext2fs_unmark_inode_bitmap2(ctx->inode_bad_map, ino);
633	ext2fs_inode_alloc_stats2(fs, ino, -1, LINUX_S_ISDIR(inode.i_mode));
634	quota_data_sub(ctx->qctx, &inode, ino, pb.dup_blocks * fs->blocksize);
635	quota_data_inodes(ctx->qctx, &inode, ino, -1);
636
637	/* Inode may have changed by block_iterate, so reread it */
638	e2fsck_read_inode(ctx, ino, &inode, "delete_file");
639	e2fsck_clear_inode(ctx, ino, &inode, 0, "delete_file");
640	if (ext2fs_file_acl_block(&inode) &&
641	    (fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_EXT_ATTR)) {
642		count = 1;
643		pctx.errcode = ext2fs_adjust_ea_refcount2(fs,
644						   ext2fs_file_acl_block(&inode),
645						   block_buf, -1, &count);
646		if (pctx.errcode == EXT2_ET_BAD_EA_BLOCK_NUM) {
647			pctx.errcode = 0;
648			count = 1;
649		}
650		if (pctx.errcode) {
651			pctx.blk = ext2fs_file_acl_block(&inode);
652			fix_problem(ctx, PR_1B_ADJ_EA_REFCOUNT, &pctx);
653		}
654		/*
655		 * If the count is zero, then arrange to have the
656		 * block deleted.  If the block is in the block_dup_map,
657		 * also call delete_file_block since it will take care
658		 * of keeping the accounting straight.
659		 */
660		if ((count == 0) ||
661		    ext2fs_test_block_bitmap2(ctx->block_dup_map,
662					      ext2fs_file_acl_block(&inode))) {
663			blk64_t blk = ext2fs_file_acl_block(&inode);
664			delete_file_block(fs, &blk,
665					  BLOCK_COUNT_EXTATTR, 0, 0, &pb);
666			ext2fs_file_acl_block_set(&inode, blk);
667			quota_data_sub(ctx->qctx, &inode, ino, fs->blocksize);
668		}
669	}
670}
671
672struct clone_struct {
673	errcode_t	errcode;
674	blk64_t		dup_cluster;
675	blk64_t		alloc_block;
676	ext2_ino_t	dir;
677	char	*buf;
678	e2fsck_t ctx;
679};
680
681static int clone_file_block(ext2_filsys fs,
682			    blk64_t	*block_nr,
683			    e2_blkcnt_t blockcnt,
684			    blk64_t ref_block EXT2FS_ATTR((unused)),
685			    int ref_offset EXT2FS_ATTR((unused)),
686			    void *priv_data)
687{
688	struct dup_cluster *p;
689	blk64_t	new_block;
690	errcode_t	retval;
691	struct clone_struct *cs = (struct clone_struct *) priv_data;
692	dnode_t *n;
693	e2fsck_t ctx;
694	blk64_t c;
695	int is_meta = 0;
696
697	ctx = cs->ctx;
698
699	if (HOLE_BLKADDR(*block_nr))
700		return 0;
701
702	c = EXT2FS_B2C(fs, blockcnt);
703	if (check_if_fs_cluster(ctx, EXT2FS_B2C(fs, *block_nr)))
704		is_meta = 1;
705
706	if (((blockcnt > 0) && c == cs->dup_cluster) ||
707	    ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr)) {
708		n = dict_lookup(&clstr_dict,
709				INT_TO_VOIDPTR(EXT2FS_B2C(fs, *block_nr)));
710		if (!n) {
711			com_err("clone_file_block", 0,
712			    _("internal error: can't find dup_blk for %llu\n"),
713				*block_nr);
714			return 0;
715		}
716
717		p = (struct dup_cluster *) dnode_get(n);
718		if (!is_meta)
719			decrement_badcount(ctx, *block_nr, p);
720
721		if (p->num_bad == 0 && !is_meta) {
722			/*
723			 * Normally num_bad never gets to zero; but in
724			 * the case of bigalloc file systems, we don't
725			 * how many blocks might be in use by a
726			 * particular inode.  So we may end up
727			 * relocating the cluster even though this
728			 * inode is the last user of the cluster.  In
729			 * that case, since we've already moved some
730			 * of the blocks of that cluster, we'll
731			 * complete the relocation and free the
732			 * original cluster here.
733			 */
734			ext2fs_unmark_block_bitmap2(ctx->block_found_map,
735						    *block_nr);
736			ext2fs_block_alloc_stats2(fs, *block_nr, -1);
737		}
738
739		if ((blockcnt > 0) && c == cs->dup_cluster) {
740			new_block = cs->alloc_block;
741			goto got_block;
742		}
743		cs->dup_cluster = c;
744
745		retval = ext2fs_new_block2(fs, 0, ctx->block_found_map,
746					   &new_block);
747		if (retval) {
748			cs->errcode = retval;
749			return BLOCK_ABORT;
750		}
751		cs->alloc_block = new_block;
752
753	got_block:
754		new_block &= ~EXT2FS_CLUSTER_MASK(fs);
755		new_block += EXT2FS_CLUSTER_MASK(fs) & blockcnt;
756		if (cs->dir && (blockcnt >= 0)) {
757			retval = ext2fs_set_dir_block2(fs->dblist,
758					cs->dir, new_block, blockcnt);
759			if (retval) {
760				cs->errcode = retval;
761				return BLOCK_ABORT;
762			}
763		}
764#if 0
765 		printf("Cloning block #%lld from %llu to %llu\n",
766		       blockcnt, *block_nr, new_block);
767#endif
768		retval = io_channel_read_blk64(fs->io, *block_nr, 1, cs->buf);
769		if (retval) {
770			cs->errcode = retval;
771			return BLOCK_ABORT;
772		}
773		retval = io_channel_write_blk64(fs->io, new_block, 1, cs->buf);
774		if (retval) {
775			cs->errcode = retval;
776			return BLOCK_ABORT;
777		}
778		*block_nr = new_block;
779		ext2fs_mark_block_bitmap2(ctx->block_found_map, new_block);
780		ext2fs_mark_block_bitmap2(fs->block_map, new_block);
781		return BLOCK_CHANGED;
782	}
783	return 0;
784}
785
786static int clone_file(e2fsck_t ctx, ext2_ino_t ino,
787		      struct dup_inode *dp, char* block_buf)
788{
789	ext2_filsys fs = ctx->fs;
790	errcode_t	retval;
791	struct clone_struct cs;
792	struct problem_context	pctx;
793	blk64_t		blk, new_blk;
794	dnode_t		*n;
795	struct inode_el	*ino_el;
796	struct dup_cluster	*dc;
797	struct dup_inode	*di;
798
799	clear_problem_context(&pctx);
800	cs.errcode = 0;
801	cs.dir = 0;
802	cs.dup_cluster = 0;
803	cs.alloc_block = 0;
804	cs.ctx = ctx;
805	retval = ext2fs_get_mem(fs->blocksize, &cs.buf);
806	if (retval)
807		return retval;
808
809	if (ext2fs_test_inode_bitmap2(ctx->inode_dir_map, ino))
810		cs.dir = ino;
811
812	pctx.ino = ino;
813	pctx.str = "clone_file";
814	if (ext2fs_inode_has_valid_blocks(&dp->inode))
815		pctx.errcode = ext2fs_block_iterate3(fs, ino, 0, block_buf,
816						     clone_file_block, &cs);
817	ext2fs_mark_bb_dirty(fs);
818	if (pctx.errcode) {
819		fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
820		retval = pctx.errcode;
821		goto errout;
822	}
823	if (cs.errcode) {
824		com_err("clone_file", cs.errcode,
825			_("returned from clone_file_block"));
826		retval = cs.errcode;
827		goto errout;
828	}
829	/* The inode may have changed on disk, so we have to re-read it */
830	e2fsck_read_inode(ctx, ino, &dp->inode, "clone file EA");
831	blk = ext2fs_file_acl_block(&dp->inode);
832	new_blk = blk;
833	if (blk && (clone_file_block(fs, &new_blk,
834				     BLOCK_COUNT_EXTATTR, 0, 0, &cs) ==
835		    BLOCK_CHANGED)) {
836		ext2fs_file_acl_block_set(&dp->inode, new_blk);
837		e2fsck_write_inode(ctx, ino, &dp->inode, "clone file EA");
838		/*
839		 * If we cloned the EA block, find all other inodes
840		 * which refered to that EA block, and modify
841		 * them to point to the new EA block.
842		 */
843		n = dict_lookup(&clstr_dict,
844				INT_TO_VOIDPTR(EXT2FS_B2C(fs, blk)));
845		if (!n) {
846			com_err("clone_file", 0,
847				_("internal error: couldn't lookup EA "
848				  "block record for %llu"), blk);
849			retval = 0; /* OK to stumble on... */
850			goto errout;
851		}
852		dc = (struct dup_cluster *) dnode_get(n);
853		for (ino_el = dc->inode_list; ino_el; ino_el = ino_el->next) {
854			if (ino_el->inode == ino)
855				continue;
856			n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(ino_el->inode));
857			if (!n) {
858				com_err("clone_file", 0,
859					_("internal error: couldn't lookup EA "
860					  "inode record for %u"),
861					ino_el->inode);
862				retval = 0; /* OK to stumble on... */
863				goto errout;
864			}
865			di = (struct dup_inode *) dnode_get(n);
866			if (ext2fs_file_acl_block(&di->inode) == blk) {
867				ext2fs_file_acl_block_set(&di->inode,
868					  ext2fs_file_acl_block(&dp->inode));
869				e2fsck_write_inode(ctx, ino_el->inode,
870					   &di->inode, "clone file EA");
871				decrement_badcount(ctx, blk, dc);
872			}
873		}
874	}
875	retval = 0;
876errout:
877	ext2fs_free_mem(&cs.buf);
878	return retval;
879}
880
881/*
882 * This routine returns 1 if a block overlaps with one of the superblocks,
883 * group descriptors, inode bitmaps, or block bitmaps.
884 */
885static int check_if_fs_block(e2fsck_t ctx, blk64_t test_block)
886{
887	ext2_filsys fs = ctx->fs;
888	blk64_t	first_block;
889	dgrp_t	i;
890
891	first_block = fs->super->s_first_data_block;
892	for (i = 0; i < fs->group_desc_count; i++) {
893
894		/* Check superblocks/block group descriptors */
895		if (ext2fs_bg_has_super(fs, i)) {
896			if (test_block >= first_block &&
897			    (test_block <= first_block + fs->desc_blocks))
898				return 1;
899		}
900
901		/* Check the inode table */
902		if ((ext2fs_inode_table_loc(fs, i)) &&
903		    (test_block >= ext2fs_inode_table_loc(fs, i)) &&
904		    (test_block < (ext2fs_inode_table_loc(fs, i) +
905				   fs->inode_blocks_per_group)))
906			return 1;
907
908		/* Check the bitmap blocks */
909		if ((test_block == ext2fs_block_bitmap_loc(fs, i)) ||
910		    (test_block == ext2fs_inode_bitmap_loc(fs, i)))
911			return 1;
912
913		first_block += fs->super->s_blocks_per_group;
914	}
915	return 0;
916}
917
918/*
919 * This routine returns 1 if a cluster overlaps with one of the superblocks,
920 * group descriptors, inode bitmaps, or block bitmaps.
921 */
922static int check_if_fs_cluster(e2fsck_t ctx, blk64_t cluster)
923{
924	ext2_filsys fs = ctx->fs;
925	blk64_t	first_block;
926	dgrp_t	i;
927
928	first_block = fs->super->s_first_data_block;
929	for (i = 0; i < fs->group_desc_count; i++) {
930
931		/* Check superblocks/block group descriptors */
932		if (ext2fs_bg_has_super(fs, i)) {
933			if (cluster >= EXT2FS_B2C(fs, first_block) &&
934			    (cluster <= EXT2FS_B2C(fs, first_block +
935						   fs->desc_blocks)))
936				return 1;
937		}
938
939		/* Check the inode table */
940		if ((ext2fs_inode_table_loc(fs, i)) &&
941		    (cluster >= EXT2FS_B2C(fs,
942					   ext2fs_inode_table_loc(fs, i))) &&
943		    (cluster <= EXT2FS_B2C(fs,
944					   ext2fs_inode_table_loc(fs, i) +
945					   fs->inode_blocks_per_group - 1)))
946			return 1;
947
948		/* Check the bitmap blocks */
949		if ((cluster == EXT2FS_B2C(fs,
950					   ext2fs_block_bitmap_loc(fs, i))) ||
951		    (cluster == EXT2FS_B2C(fs,
952					   ext2fs_inode_bitmap_loc(fs, i))))
953			return 1;
954
955		first_block += fs->super->s_blocks_per_group;
956	}
957	return 0;
958}
959