Home | History | Annotate | Download | only in e2fsck
      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 
    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  */
    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  */
    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  */
    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  */
    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 	/*
    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  */
    260 struct 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_large *inode;
    267 	struct problem_context *pctx;
    268 };
    269 
    270 static 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_large 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 = EXT2_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_full(scan, &ino,
    301 				EXT2_INODE(&inode), sizeof(inode));
    302 		if (pctx.errcode == EXT2_ET_BAD_BLOCK_IN_INODE_TABLE)
    303 			continue;
    304 		if (pctx.errcode) {
    305 			pctx.ino = ino;
    306 			fix_problem(ctx, PR_1B_ISCAN_ERROR, &pctx);
    307 			ctx->flags |= E2F_FLAG_ABORT;
    308 			return;
    309 		}
    310 		if (!ino)
    311 			break;
    312 		pctx.ino = ctx->stashed_ino = ino;
    313 		if ((ino != EXT2_BAD_INO) &&
    314 		    !ext2fs_test_inode_bitmap2(ctx->inode_used_map, ino))
    315 			continue;
    316 
    317 		pb.ino = ino;
    318 		pb.dup_blocks = 0;
    319 		pb.inode = &inode;
    320 		pb.cur_cluster = ~0;
    321 		pb.phys_cluster = ~0;
    322 		pb.last_blk = 0;
    323 		pb.pctx->blk = pb.pctx->blk2 = 0;
    324 
    325 		if (ext2fs_inode_has_valid_blocks2(fs, EXT2_INODE(&inode)) ||
    326 		    (ino == EXT2_BAD_INO))
    327 			pctx.errcode = ext2fs_block_iterate3(fs, ino,
    328 					     BLOCK_FLAG_READ_ONLY, block_buf,
    329 					     process_pass1b_block, &pb);
    330 		/* If the feature is not set, attrs will be cleared later anyway */
    331 		if (ext2fs_has_feature_xattr(fs->super) &&
    332 		    ext2fs_file_acl_block(fs, EXT2_INODE(&inode))) {
    333 			blk64_t blk = ext2fs_file_acl_block(fs, EXT2_INODE(&inode));
    334 			process_pass1b_block(fs, &blk,
    335 					     BLOCK_COUNT_EXTATTR, 0, 0, &pb);
    336 			ext2fs_file_acl_block_set(fs, EXT2_INODE(&inode), blk);
    337 		}
    338 		if (pb.dup_blocks) {
    339 			if (ino != EXT2_BAD_INO) {
    340 				op = pctx.blk == pctx.blk2 ?
    341 					PR_1B_DUP_BLOCK : PR_1B_DUP_RANGE;
    342 				fix_problem(ctx, op, pb.pctx);
    343 			}
    344 			end_problem_latch(ctx, PR_LATCH_DBLOCK);
    345 			if (ino >= EXT2_FIRST_INODE(fs->super) ||
    346 			    ino == EXT2_ROOT_INO)
    347 				dup_inode_count++;
    348 		}
    349 		if (pctx.errcode)
    350 			fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
    351 	}
    352 	ext2fs_close_inode_scan(scan);
    353 	e2fsck_use_inode_shortcuts(ctx, 0);
    354 }
    355 
    356 static int process_pass1b_block(ext2_filsys fs EXT2FS_ATTR((unused)),
    357 				blk64_t	*block_nr,
    358 				e2_blkcnt_t blockcnt,
    359 				blk64_t ref_blk EXT2FS_ATTR((unused)),
    360 				int ref_offset EXT2FS_ATTR((unused)),
    361 				void *priv_data)
    362 {
    363 	struct process_block_struct *p;
    364 	e2fsck_t ctx;
    365 	blk64_t	lc, pc;
    366 	problem_t op;
    367 
    368 	if (*block_nr == 0)
    369 		return 0;
    370 	p = (struct process_block_struct *) priv_data;
    371 	ctx = p->ctx;
    372 	lc = EXT2FS_B2C(fs, blockcnt);
    373 	pc = EXT2FS_B2C(fs, *block_nr);
    374 
    375 	if (!ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr))
    376 		goto finish;
    377 
    378 	/* OK, this is a duplicate block */
    379 	if (p->ino != EXT2_BAD_INO) {
    380 		if (p->last_blk + 1 != *block_nr) {
    381 			if (p->last_blk) {
    382 				op = p->pctx->blk == p->pctx->blk2 ?
    383 						PR_1B_DUP_BLOCK :
    384 						PR_1B_DUP_RANGE;
    385 				fix_problem(ctx, op, p->pctx);
    386 			}
    387 			p->pctx->blk = *block_nr;
    388 		}
    389 		p->pctx->blk2 = *block_nr;
    390 		p->last_blk = *block_nr;
    391 	}
    392 	p->dup_blocks++;
    393 	ext2fs_mark_inode_bitmap2(inode_dup_map, p->ino);
    394 
    395 	/*
    396 	 * Qualifications for submitting a block for duplicate processing:
    397 	 * It's an extent/indirect block (and has a negative logical offset);
    398 	 * we've crossed a logical cluster boundary; or the physical cluster
    399 	 * suddenly changed, which indicates that blocks in a logical cluster
    400 	 * are mapped to multiple physical clusters.
    401 	 */
    402 	if (blockcnt < 0 || lc != p->cur_cluster || pc != p->phys_cluster)
    403 		add_dupe(ctx, p->ino, EXT2FS_B2C(fs, *block_nr), p->inode);
    404 
    405 finish:
    406 	p->cur_cluster = lc;
    407 	p->phys_cluster = pc;
    408 	return 0;
    409 }
    410 
    411 /*
    412  * Pass 1c: Scan directories for inodes with duplicate blocks.  This
    413  * is used so that we can print pathnames when prompting the user for
    414  * what to do.
    415  */
    416 struct search_dir_struct {
    417 	int		count;
    418 	ext2_ino_t	first_inode;
    419 	ext2_ino_t	max_inode;
    420 };
    421 
    422 static int search_dirent_proc(ext2_ino_t dir, int entry,
    423 			      struct ext2_dir_entry *dirent,
    424 			      int offset EXT2FS_ATTR((unused)),
    425 			      int blocksize EXT2FS_ATTR((unused)),
    426 			      char *buf EXT2FS_ATTR((unused)),
    427 			      void *priv_data)
    428 {
    429 	struct search_dir_struct *sd;
    430 	struct dup_inode	*p;
    431 	dnode_t			*n;
    432 
    433 	sd = (struct search_dir_struct *) priv_data;
    434 
    435 	if (dirent->inode > sd->max_inode)
    436 		/* Should abort this inode, but not everything */
    437 		return 0;
    438 
    439 	if ((dirent->inode < sd->first_inode) || (entry < DIRENT_OTHER_FILE) ||
    440 	    !ext2fs_test_inode_bitmap2(inode_dup_map, dirent->inode))
    441 		return 0;
    442 
    443 	n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(dirent->inode));
    444 	if (!n)
    445 		return 0;
    446 	p = (struct dup_inode *) dnode_get(n);
    447 	if (!p->dir) {
    448 		p->dir = dir;
    449 		sd->count--;
    450 	}
    451 
    452 	return(sd->count ? 0 : DIRENT_ABORT);
    453 }
    454 
    455 
    456 static void pass1c(e2fsck_t ctx, char *block_buf)
    457 {
    458 	ext2_filsys fs = ctx->fs;
    459 	struct search_dir_struct sd;
    460 	struct problem_context pctx;
    461 
    462 	clear_problem_context(&pctx);
    463 
    464 	if (!(ctx->options & E2F_OPT_PREEN))
    465 		fix_problem(ctx, PR_1C_PASS_HEADER, &pctx);
    466 
    467 	/*
    468 	 * Search through all directories to translate inodes to names
    469 	 * (by searching for the containing directory for that inode.)
    470 	 */
    471 	sd.count = dup_inode_count - dup_inode_founddir;
    472 	sd.first_inode = EXT2_FIRST_INODE(fs->super);
    473 	sd.max_inode = fs->super->s_inodes_count;
    474 	ext2fs_dblist_dir_iterate(fs->dblist, 0, block_buf,
    475 				  search_dirent_proc, &sd);
    476 }
    477 
    478 static void pass1d(e2fsck_t ctx, char *block_buf)
    479 {
    480 	ext2_filsys fs = ctx->fs;
    481 	struct dup_inode	*p, *t;
    482 	struct dup_cluster	*q;
    483 	ext2_ino_t		*shared, ino;
    484 	int	shared_len;
    485 	int	i;
    486 	int	file_ok;
    487 	int	meta_data = 0;
    488 	struct problem_context pctx;
    489 	dnode_t	*n, *m;
    490 	struct cluster_el	*s;
    491 	struct inode_el *r;
    492 
    493 	clear_problem_context(&pctx);
    494 
    495 	if (!(ctx->options & E2F_OPT_PREEN))
    496 		fix_problem(ctx, PR_1D_PASS_HEADER, &pctx);
    497 	e2fsck_read_bitmaps(ctx);
    498 
    499 	pctx.num = dup_inode_count; /* dict_count(&ino_dict); */
    500 	fix_problem(ctx, PR_1D_NUM_DUP_INODES, &pctx);
    501 	shared = (ext2_ino_t *) e2fsck_allocate_memory(ctx,
    502 				sizeof(ext2_ino_t) * dict_count(&ino_dict),
    503 				"Shared inode list");
    504 	for (n = dict_first(&ino_dict); n; n = dict_next(&ino_dict, n)) {
    505 		p = (struct dup_inode *) dnode_get(n);
    506 		shared_len = 0;
    507 		file_ok = 1;
    508 		ino = (ext2_ino_t)VOIDPTR_TO_INT(dnode_getkey(n));
    509 		if (ino == EXT2_BAD_INO || ino == EXT2_RESIZE_INO)
    510 			continue;
    511 
    512 		/*
    513 		 * Find all of the inodes which share blocks with this
    514 		 * one.  First we find all of the duplicate blocks
    515 		 * belonging to this inode, and then search each block
    516 		 * get the list of inodes, and merge them together.
    517 		 */
    518 		for (s = p->cluster_list; s; s = s->next) {
    519 			m = dict_lookup(&clstr_dict,
    520 					INT_TO_VOIDPTR(s->cluster));
    521 			if (!m)
    522 				continue; /* Should never happen... */
    523 			q = (struct dup_cluster *) dnode_get(m);
    524 			if (q->num_bad > 1)
    525 				file_ok = 0;
    526 			if (check_if_fs_cluster(ctx, s->cluster)) {
    527 				file_ok = 0;
    528 				meta_data = 1;
    529 			}
    530 
    531 			/*
    532 			 * Add all inodes used by this block to the
    533 			 * shared[] --- which is a unique list, so
    534 			 * if an inode is already in shared[], don't
    535 			 * add it again.
    536 			 */
    537 			for (r = q->inode_list; r; r = r->next) {
    538 				if (r->inode == ino)
    539 					continue;
    540 				for (i = 0; i < shared_len; i++)
    541 					if (shared[i] == r->inode)
    542 						break;
    543 				if (i == shared_len) {
    544 					shared[shared_len++] = r->inode;
    545 				}
    546 			}
    547 		}
    548 
    549 		/*
    550 		 * Report the inode that we are working on
    551 		 */
    552 		pctx.inode = EXT2_INODE(&p->inode);
    553 		pctx.ino = ino;
    554 		pctx.dir = p->dir;
    555 		pctx.blkcount = p->num_dupblocks;
    556 		pctx.num = meta_data ? shared_len+1 : shared_len;
    557 		fix_problem(ctx, PR_1D_DUP_FILE, &pctx);
    558 		pctx.blkcount = 0;
    559 		pctx.num = 0;
    560 
    561 		if (meta_data)
    562 			fix_problem(ctx, PR_1D_SHARE_METADATA, &pctx);
    563 
    564 		for (i = 0; i < shared_len; i++) {
    565 			m = dict_lookup(&ino_dict, INT_TO_VOIDPTR(shared[i]));
    566 			if (!m)
    567 				continue; /* should never happen */
    568 			t = (struct dup_inode *) dnode_get(m);
    569 			/*
    570 			 * Report the inode that we are sharing with
    571 			 */
    572 			pctx.inode = EXT2_INODE(&t->inode);
    573 			pctx.ino = shared[i];
    574 			pctx.dir = t->dir;
    575 			fix_problem(ctx, PR_1D_DUP_FILE_LIST, &pctx);
    576 		}
    577 		/*
    578 		 * Even if the file shares blocks with itself, we still need to
    579 		 * clone the blocks.
    580 		 */
    581 		if (file_ok && (meta_data ? shared_len+1 : shared_len) != 0) {
    582 			fix_problem(ctx, PR_1D_DUP_BLOCKS_DEALT, &pctx);
    583 			continue;
    584 		}
    585 		if (fix_problem(ctx, PR_1D_CLONE_QUESTION, &pctx)) {
    586 			pctx.errcode = clone_file(ctx, ino, p, block_buf);
    587 			if (pctx.errcode)
    588 				fix_problem(ctx, PR_1D_CLONE_ERROR, &pctx);
    589 			else
    590 				continue;
    591 		}
    592 		if (fix_problem(ctx, PR_1D_DELETE_QUESTION, &pctx))
    593 			delete_file(ctx, ino, p, block_buf);
    594 		else
    595 			ext2fs_unmark_valid(fs);
    596 	}
    597 	ext2fs_free_mem(&shared);
    598 }
    599 
    600 /*
    601  * Drop the refcount on the dup_block structure, and clear the entry
    602  * in the block_dup_map if appropriate.
    603  */
    604 static void decrement_badcount(e2fsck_t ctx, blk64_t block,
    605 			       struct dup_cluster *p)
    606 {
    607 	p->num_bad--;
    608 	if (p->num_bad <= 0 ||
    609 	    (p->num_bad == 1 && !check_if_fs_block(ctx, block))) {
    610 		if (check_if_fs_cluster(ctx, EXT2FS_B2C(ctx->fs, block)))
    611 			return;
    612 		ext2fs_unmark_block_bitmap2(ctx->block_dup_map, block);
    613 	}
    614 }
    615 
    616 static int delete_file_block(ext2_filsys fs,
    617 			     blk64_t	*block_nr,
    618 			     e2_blkcnt_t blockcnt,
    619 			     blk64_t ref_block EXT2FS_ATTR((unused)),
    620 			     int ref_offset EXT2FS_ATTR((unused)),
    621 			     void *priv_data)
    622 {
    623 	struct process_block_struct *pb;
    624 	struct dup_cluster *p;
    625 	dnode_t	*n;
    626 	e2fsck_t ctx;
    627 	blk64_t c, lc;
    628 
    629 	pb = (struct process_block_struct *) priv_data;
    630 	ctx = pb->ctx;
    631 
    632 	if (*block_nr == 0)
    633 		return 0;
    634 
    635 	c = EXT2FS_B2C(fs, *block_nr);
    636 	lc = EXT2FS_B2C(fs, blockcnt);
    637 	if (ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr)) {
    638 		n = dict_lookup(&clstr_dict, INT_TO_VOIDPTR(c));
    639 		if (n) {
    640 			p = (struct dup_cluster *) dnode_get(n);
    641 			if (lc != pb->cur_cluster)
    642 				decrement_badcount(ctx, *block_nr, p);
    643 		} else
    644 			com_err("delete_file_block", 0,
    645 			    _("internal error: can't find dup_blk for %llu\n"),
    646 				*block_nr);
    647 	} else {
    648 		if ((*block_nr % EXT2FS_CLUSTER_RATIO(ctx->fs)) == 0)
    649 			ext2fs_block_alloc_stats2(fs, *block_nr, -1);
    650 		pb->dup_blocks++;
    651 	}
    652 	pb->cur_cluster = lc;
    653 
    654 	return 0;
    655 }
    656 
    657 static void delete_file(e2fsck_t ctx, ext2_ino_t ino,
    658 			struct dup_inode *dp, char* block_buf)
    659 {
    660 	ext2_filsys fs = ctx->fs;
    661 	struct process_block_struct pb;
    662 	struct problem_context	pctx;
    663 	unsigned int		count;
    664 
    665 	clear_problem_context(&pctx);
    666 	pctx.ino = pb.ino = ino;
    667 	pb.dup_blocks = 0;
    668 	pb.ctx = ctx;
    669 	pctx.str = "delete_file";
    670 	pb.cur_cluster = ~0;
    671 
    672 	if (ext2fs_inode_has_valid_blocks2(fs, EXT2_INODE(&dp->inode)))
    673 		pctx.errcode = ext2fs_block_iterate3(fs, ino,
    674 						     BLOCK_FLAG_READ_ONLY,
    675 						     block_buf,
    676 						     delete_file_block, &pb);
    677 	if (pctx.errcode)
    678 		fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
    679 	if (ctx->inode_bad_map)
    680 		ext2fs_unmark_inode_bitmap2(ctx->inode_bad_map, ino);
    681 	ext2fs_inode_alloc_stats2(fs, ino, -1, LINUX_S_ISDIR(dp->inode.i_mode));
    682 	quota_data_sub(ctx->qctx, &dp->inode, ino,
    683 		       pb.dup_blocks * fs->blocksize);
    684 	quota_data_inodes(ctx->qctx, &dp->inode, ino, -1);
    685 
    686 	/* Inode may have changed by block_iterate, so reread it */
    687 	e2fsck_read_inode_full(ctx, ino, EXT2_INODE(&dp->inode),
    688 			       sizeof(dp->inode), "delete_file");
    689 	e2fsck_clear_inode(ctx, ino, EXT2_INODE(&dp->inode), 0, "delete_file");
    690 	if (ext2fs_file_acl_block(fs, EXT2_INODE(&dp->inode)) &&
    691 	    ext2fs_has_feature_xattr(fs->super)) {
    692 		blk64_t file_acl_block = ext2fs_file_acl_block(fs,
    693 						EXT2_INODE(&dp->inode));
    694 
    695 		count = 1;
    696 		pctx.errcode = ext2fs_adjust_ea_refcount3(fs, file_acl_block,
    697 					block_buf, -1, &count, ino);
    698 		if (pctx.errcode == EXT2_ET_BAD_EA_BLOCK_NUM) {
    699 			pctx.errcode = 0;
    700 			count = 1;
    701 		}
    702 		if (pctx.errcode) {
    703 			pctx.blk = file_acl_block;
    704 			fix_problem(ctx, PR_1B_ADJ_EA_REFCOUNT, &pctx);
    705 		}
    706 		/*
    707 		 * If the count is zero, then arrange to have the
    708 		 * block deleted.  If the block is in the block_dup_map,
    709 		 * also call delete_file_block since it will take care
    710 		 * of keeping the accounting straight.
    711 		 */
    712 		if ((count == 0) ||
    713 		    ext2fs_test_block_bitmap2(ctx->block_dup_map,
    714 					      file_acl_block)) {
    715 			delete_file_block(fs, &file_acl_block,
    716 					  BLOCK_COUNT_EXTATTR, 0, 0, &pb);
    717 			ext2fs_file_acl_block_set(fs, EXT2_INODE(&dp->inode),
    718 						  file_acl_block);
    719 			quota_data_sub(ctx->qctx, &dp->inode, ino,
    720 				       fs->blocksize);
    721 		}
    722 	}
    723 }
    724 
    725 struct clone_struct {
    726 	errcode_t	errcode;
    727 	blk64_t		dup_cluster;
    728 	blk64_t		alloc_block;
    729 	ext2_ino_t	dir, ino;
    730 	char	*buf;
    731 	e2fsck_t ctx;
    732 	struct ext2_inode_large	*inode;
    733 
    734 	struct dup_cluster *save_dup_cluster;
    735 	blk64_t save_blocknr;
    736 };
    737 
    738 /*
    739  * Decrement the bad count *after* we've shown that (a) we can allocate a
    740  * replacement block and (b) remap the file blocks.  Unfortunately, there's no
    741  * way to find out if the remap succeeded until either the next
    742  * clone_file_block() call (an error when remapping the block after returning
    743  * BLOCK_CHANGED will halt the iteration) or after block_iterate() returns.
    744  * Otherwise, it's possible that we decrease the badcount once in preparation
    745  * to remap, then the remap fails (either we can't find a replacement block or
    746  * we have to split the extent tree and can't find a new extent block), so we
    747  * delete the file, which decreases the badcount again.
    748  */
    749 static void deferred_dec_badcount(struct clone_struct *cs)
    750 {
    751 	if (!cs->save_dup_cluster)
    752 		return;
    753 	decrement_badcount(cs->ctx, cs->save_blocknr, cs->save_dup_cluster);
    754 	cs->save_dup_cluster = NULL;
    755 }
    756 
    757 static int clone_file_block(ext2_filsys fs,
    758 			    blk64_t	*block_nr,
    759 			    e2_blkcnt_t blockcnt,
    760 			    blk64_t ref_block EXT2FS_ATTR((unused)),
    761 			    int ref_offset EXT2FS_ATTR((unused)),
    762 			    void *priv_data)
    763 {
    764 	struct dup_cluster *p = NULL;
    765 	blk64_t	new_block;
    766 	errcode_t	retval;
    767 	struct clone_struct *cs = (struct clone_struct *) priv_data;
    768 	dnode_t *n;
    769 	e2fsck_t ctx;
    770 	blk64_t c;
    771 	int is_meta = 0;
    772 
    773 	ctx = cs->ctx;
    774 	deferred_dec_badcount(cs);
    775 
    776 	if (*block_nr == 0)
    777 		return 0;
    778 
    779 	c = EXT2FS_B2C(fs, blockcnt);
    780 	if (check_if_fs_cluster(ctx, EXT2FS_B2C(fs, *block_nr)))
    781 		is_meta = 1;
    782 
    783 	if (c == cs->dup_cluster && cs->alloc_block) {
    784 		new_block = cs->alloc_block;
    785 		goto got_block;
    786 	}
    787 
    788 	if (ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr)) {
    789 		n = dict_lookup(&clstr_dict,
    790 				INT_TO_VOIDPTR(EXT2FS_B2C(fs, *block_nr)));
    791 		if (!n) {
    792 			com_err("clone_file_block", 0,
    793 			    _("internal error: can't find dup_blk for %llu\n"),
    794 				*block_nr);
    795 			return 0;
    796 		}
    797 
    798 		p = (struct dup_cluster *) dnode_get(n);
    799 
    800 		cs->dup_cluster = c;
    801 		/*
    802 		 * Let's try an implied cluster allocation.  If we get the same
    803 		 * cluster back, then we need to find a new block; otherwise,
    804 		 * we're merely fixing the problem of one logical cluster being
    805 		 * mapped to multiple physical clusters.
    806 		 */
    807 		new_block = 0;
    808 		retval = ext2fs_map_cluster_block(fs, cs->ino,
    809 						  EXT2_INODE(cs->inode),
    810 						  blockcnt, &new_block);
    811 		if (retval == 0 && new_block != 0 &&
    812 		    EXT2FS_B2C(ctx->fs, new_block) !=
    813 		    EXT2FS_B2C(ctx->fs, *block_nr))
    814 			goto cluster_alloc_ok;
    815 		retval = ext2fs_new_block2(fs, 0, ctx->block_found_map,
    816 					   &new_block);
    817 		if (retval) {
    818 			cs->errcode = retval;
    819 			return BLOCK_ABORT;
    820 		}
    821 cluster_alloc_ok:
    822 		cs->alloc_block = new_block;
    823 
    824 	got_block:
    825 		new_block &= ~EXT2FS_CLUSTER_MASK(fs);
    826 		new_block += EXT2FS_CLUSTER_MASK(fs) & blockcnt;
    827 		if (cs->dir && (blockcnt >= 0)) {
    828 			retval = ext2fs_set_dir_block2(fs->dblist,
    829 					cs->dir, new_block, blockcnt);
    830 			if (retval) {
    831 				cs->errcode = retval;
    832 				return BLOCK_ABORT;
    833 			}
    834 		}
    835 #if 0
    836  		printf("Cloning block #%lld from %llu to %llu\n",
    837 		       blockcnt, *block_nr, new_block);
    838 #endif
    839 		retval = io_channel_read_blk64(fs->io, *block_nr, 1, cs->buf);
    840 		if (retval) {
    841 			cs->errcode = retval;
    842 			return BLOCK_ABORT;
    843 		}
    844 		retval = io_channel_write_blk64(fs->io, new_block, 1, cs->buf);
    845 		if (retval) {
    846 			cs->errcode = retval;
    847 			return BLOCK_ABORT;
    848 		}
    849 		cs->save_dup_cluster = (is_meta ? NULL : p);
    850 		cs->save_blocknr = *block_nr;
    851 		*block_nr = new_block;
    852 		ext2fs_mark_block_bitmap2(ctx->block_found_map, new_block);
    853 		ext2fs_mark_block_bitmap2(fs->block_map, new_block);
    854 		return BLOCK_CHANGED;
    855 	}
    856 	return 0;
    857 }
    858 
    859 static errcode_t clone_file(e2fsck_t ctx, ext2_ino_t ino,
    860 			    struct dup_inode *dp, char* block_buf)
    861 {
    862 	ext2_filsys fs = ctx->fs;
    863 	errcode_t	retval;
    864 	struct clone_struct cs;
    865 	struct problem_context	pctx;
    866 	blk64_t		blk, new_blk;
    867 	dnode_t		*n;
    868 	struct inode_el	*ino_el;
    869 	struct dup_cluster	*dc;
    870 	struct dup_inode	*di;
    871 
    872 	clear_problem_context(&pctx);
    873 	cs.errcode = 0;
    874 	cs.dir = 0;
    875 	cs.dup_cluster = ~0;
    876 	cs.alloc_block = 0;
    877 	cs.ctx = ctx;
    878 	cs.ino = ino;
    879 	cs.inode = &dp->inode;
    880 	cs.save_dup_cluster = NULL;
    881 	cs.save_blocknr = 0;
    882 	retval = ext2fs_get_mem(fs->blocksize, &cs.buf);
    883 	if (retval)
    884 		return retval;
    885 
    886 	if (ext2fs_test_inode_bitmap2(ctx->inode_dir_map, ino))
    887 		cs.dir = ino;
    888 
    889 	pctx.ino = ino;
    890 	pctx.str = "clone_file";
    891 	if (ext2fs_inode_has_valid_blocks2(fs, EXT2_INODE(&dp->inode)))
    892 		pctx.errcode = ext2fs_block_iterate3(fs, ino, 0, block_buf,
    893 						     clone_file_block, &cs);
    894 	deferred_dec_badcount(&cs);
    895 	ext2fs_mark_bb_dirty(fs);
    896 	if (pctx.errcode) {
    897 		fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
    898 		retval = pctx.errcode;
    899 		goto errout;
    900 	}
    901 	if (cs.errcode) {
    902 		com_err("clone_file", cs.errcode, "%s",
    903 			_("returned from clone_file_block"));
    904 		retval = cs.errcode;
    905 		goto errout;
    906 	}
    907 	/* The inode may have changed on disk, so we have to re-read it */
    908 	e2fsck_read_inode_full(ctx, ino, EXT2_INODE(&dp->inode),
    909 			       sizeof(dp->inode), "clone file EA");
    910 	blk = ext2fs_file_acl_block(fs, EXT2_INODE(&dp->inode));
    911 	new_blk = blk;
    912 	if (blk && (clone_file_block(fs, &new_blk,
    913 				     BLOCK_COUNT_EXTATTR, 0, 0, &cs) ==
    914 		    BLOCK_CHANGED)) {
    915 		ext2fs_file_acl_block_set(fs, EXT2_INODE(&dp->inode), new_blk);
    916 		e2fsck_write_inode_full(ctx, ino, EXT2_INODE(&dp->inode),
    917 					sizeof(dp->inode), "clone file EA");
    918 		/*
    919 		 * If we cloned the EA block, find all other inodes
    920 		 * which refered to that EA block, and modify
    921 		 * them to point to the new EA block.
    922 		 */
    923 		n = dict_lookup(&clstr_dict,
    924 				INT_TO_VOIDPTR(EXT2FS_B2C(fs, blk)));
    925 		if (!n) {
    926 			com_err("clone_file", 0,
    927 				_("internal error: couldn't lookup EA "
    928 				  "block record for %llu"), blk);
    929 			retval = 0; /* OK to stumble on... */
    930 			goto errout;
    931 		}
    932 		dc = (struct dup_cluster *) dnode_get(n);
    933 		for (ino_el = dc->inode_list; ino_el; ino_el = ino_el->next) {
    934 			if (ino_el->inode == ino)
    935 				continue;
    936 			n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(ino_el->inode));
    937 			if (!n) {
    938 				com_err("clone_file", 0,
    939 					_("internal error: couldn't lookup EA "
    940 					  "inode record for %u"),
    941 					ino_el->inode);
    942 				retval = 0; /* OK to stumble on... */
    943 				goto errout;
    944 			}
    945 			di = (struct dup_inode *) dnode_get(n);
    946 			if (ext2fs_file_acl_block(fs,
    947 					EXT2_INODE(&di->inode)) == blk) {
    948 				ext2fs_file_acl_block_set(fs,
    949 					EXT2_INODE(&di->inode),
    950 					ext2fs_file_acl_block(fs, EXT2_INODE(&dp->inode)));
    951 				e2fsck_write_inode_full(ctx, ino_el->inode,
    952 					EXT2_INODE(&di->inode),
    953 					sizeof(di->inode), "clone file EA");
    954 				decrement_badcount(ctx, blk, dc);
    955 			}
    956 		}
    957 	}
    958 	retval = 0;
    959 errout:
    960 	ext2fs_free_mem(&cs.buf);
    961 	return retval;
    962 }
    963 
    964 /*
    965  * This routine returns 1 if a block overlaps with one of the superblocks,
    966  * group descriptors, inode bitmaps, or block bitmaps.
    967  */
    968 static int check_if_fs_block(e2fsck_t ctx, blk64_t test_block)
    969 {
    970 	ext2_filsys fs = ctx->fs;
    971 	blk64_t	first_block;
    972 	dgrp_t	i;
    973 
    974 	first_block = fs->super->s_first_data_block;
    975 	for (i = 0; i < fs->group_desc_count; i++) {
    976 
    977 		/* Check superblocks/block group descriptors */
    978 		if (ext2fs_bg_has_super(fs, i)) {
    979 			if (test_block >= first_block &&
    980 			    (test_block <= first_block + fs->desc_blocks))
    981 				return 1;
    982 		}
    983 
    984 		/* Check the inode table */
    985 		if ((ext2fs_inode_table_loc(fs, i)) &&
    986 		    (test_block >= ext2fs_inode_table_loc(fs, i)) &&
    987 		    (test_block < (ext2fs_inode_table_loc(fs, i) +
    988 				   fs->inode_blocks_per_group)))
    989 			return 1;
    990 
    991 		/* Check the bitmap blocks */
    992 		if ((test_block == ext2fs_block_bitmap_loc(fs, i)) ||
    993 		    (test_block == ext2fs_inode_bitmap_loc(fs, i)))
    994 			return 1;
    995 
    996 		first_block += fs->super->s_blocks_per_group;
    997 	}
    998 	return 0;
    999 }
   1000 
   1001 /*
   1002  * This routine returns 1 if a cluster overlaps with one of the superblocks,
   1003  * group descriptors, inode bitmaps, or block bitmaps.
   1004  */
   1005 static int check_if_fs_cluster(e2fsck_t ctx, blk64_t cluster)
   1006 {
   1007 	ext2_filsys fs = ctx->fs;
   1008 	blk64_t	first_block;
   1009 	dgrp_t	i;
   1010 
   1011 	first_block = fs->super->s_first_data_block;
   1012 	for (i = 0; i < fs->group_desc_count; i++) {
   1013 
   1014 		/* Check superblocks/block group descriptors */
   1015 		if (ext2fs_bg_has_super(fs, i)) {
   1016 			if (cluster >= EXT2FS_B2C(fs, first_block) &&
   1017 			    (cluster <= EXT2FS_B2C(fs, first_block +
   1018 						   fs->desc_blocks)))
   1019 				return 1;
   1020 		}
   1021 
   1022 		/* Check the inode table */
   1023 		if ((ext2fs_inode_table_loc(fs, i)) &&
   1024 		    (cluster >= EXT2FS_B2C(fs,
   1025 					   ext2fs_inode_table_loc(fs, i))) &&
   1026 		    (cluster <= EXT2FS_B2C(fs,
   1027 					   ext2fs_inode_table_loc(fs, i) +
   1028 					   fs->inode_blocks_per_group - 1)))
   1029 			return 1;
   1030 
   1031 		/* Check the bitmap blocks */
   1032 		if ((cluster == EXT2FS_B2C(fs,
   1033 					   ext2fs_block_bitmap_loc(fs, i))) ||
   1034 		    (cluster == EXT2FS_B2C(fs,
   1035 					   ext2fs_inode_bitmap_loc(fs, i))))
   1036 			return 1;
   1037 
   1038 		first_block += fs->super->s_blocks_per_group;
   1039 	}
   1040 	return 0;
   1041 }
   1042