Whamcloud - gitweb
e2fsck: release clusters only once in release_inode_blocks
[tools/e2fsprogs.git] / e2fsck / pass1b.c
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                         if (lc != pb->cur_cluster) {
641                                 p = (struct dup_cluster *) dnode_get(n);
642                                 decrement_badcount(ctx, *block_nr, p);
643                                 pb->dup_blocks++;
644                         }
645                 } else
646                         com_err("delete_file_block", 0,
647                             _("internal error: can't find dup_blk for %llu\n"),
648                                 *block_nr);
649         } else {
650                 if ((*block_nr % EXT2FS_CLUSTER_RATIO(ctx->fs)) == 0)
651                         ext2fs_block_alloc_stats2(fs, *block_nr, -1);
652                 pb->dup_blocks++;
653         }
654         pb->cur_cluster = lc;
655
656         return 0;
657 }
658
659 static void delete_file(e2fsck_t ctx, ext2_ino_t ino,
660                         struct dup_inode *dp, char* block_buf)
661 {
662         ext2_filsys fs = ctx->fs;
663         struct process_block_struct pb;
664         struct problem_context  pctx;
665         unsigned int            count;
666
667         clear_problem_context(&pctx);
668         pctx.ino = pb.ino = ino;
669         pb.dup_blocks = 0;
670         pb.ctx = ctx;
671         pctx.str = "delete_file";
672         pb.cur_cluster = ~0;
673
674         if (ext2fs_inode_has_valid_blocks2(fs, EXT2_INODE(&dp->inode)))
675                 pctx.errcode = ext2fs_block_iterate3(fs, ino,
676                                                      BLOCK_FLAG_READ_ONLY,
677                                                      block_buf,
678                                                      delete_file_block, &pb);
679         if (pctx.errcode)
680                 fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
681         if (ctx->inode_bad_map)
682                 ext2fs_unmark_inode_bitmap2(ctx->inode_bad_map, ino);
683         ext2fs_inode_alloc_stats2(fs, ino, -1, LINUX_S_ISDIR(dp->inode.i_mode));
684         quota_data_sub(ctx->qctx, &dp->inode, ino,
685                        pb.dup_blocks * fs->blocksize);
686         quota_data_inodes(ctx->qctx, &dp->inode, ino, -1);
687
688         /* Inode may have changed by block_iterate, so reread it */
689         e2fsck_read_inode_full(ctx, ino, EXT2_INODE(&dp->inode),
690                                sizeof(dp->inode), "delete_file");
691         e2fsck_clear_inode(ctx, ino, EXT2_INODE(&dp->inode), 0, "delete_file");
692         if (ext2fs_file_acl_block(fs, EXT2_INODE(&dp->inode)) &&
693             ext2fs_has_feature_xattr(fs->super)) {
694                 blk64_t file_acl_block = ext2fs_file_acl_block(fs,
695                                                 EXT2_INODE(&dp->inode));
696
697                 count = 1;
698                 pctx.errcode = ext2fs_adjust_ea_refcount3(fs, file_acl_block,
699                                         block_buf, -1, &count, ino);
700                 if (pctx.errcode == EXT2_ET_BAD_EA_BLOCK_NUM) {
701                         pctx.errcode = 0;
702                         count = 1;
703                 }
704                 if (pctx.errcode) {
705                         pctx.blk = file_acl_block;
706                         fix_problem(ctx, PR_1B_ADJ_EA_REFCOUNT, &pctx);
707                 }
708                 /*
709                  * If the count is zero, then arrange to have the
710                  * block deleted.  If the block is in the block_dup_map,
711                  * also call delete_file_block since it will take care
712                  * of keeping the accounting straight.
713                  */
714                 if ((count == 0) ||
715                     ext2fs_test_block_bitmap2(ctx->block_dup_map,
716                                               file_acl_block)) {
717                         delete_file_block(fs, &file_acl_block,
718                                           BLOCK_COUNT_EXTATTR, 0, 0, &pb);
719                         ext2fs_file_acl_block_set(fs, EXT2_INODE(&dp->inode),
720                                                   file_acl_block);
721                         quota_data_sub(ctx->qctx, &dp->inode, ino,
722                                        fs->blocksize);
723                 }
724         }
725 }
726
727 struct clone_struct {
728         errcode_t       errcode;
729         blk64_t         dup_cluster;
730         blk64_t         alloc_block;
731         ext2_ino_t      dir, ino;
732         char    *buf;
733         e2fsck_t ctx;
734         struct ext2_inode_large *inode;
735
736         struct dup_cluster *save_dup_cluster;
737         blk64_t save_blocknr;
738 };
739
740 /*
741  * Decrement the bad count *after* we've shown that (a) we can allocate a
742  * replacement block and (b) remap the file blocks.  Unfortunately, there's no
743  * way to find out if the remap succeeded until either the next
744  * clone_file_block() call (an error when remapping the block after returning
745  * BLOCK_CHANGED will halt the iteration) or after block_iterate() returns.
746  * Otherwise, it's possible that we decrease the badcount once in preparation
747  * to remap, then the remap fails (either we can't find a replacement block or
748  * we have to split the extent tree and can't find a new extent block), so we
749  * delete the file, which decreases the badcount again.
750  */
751 static void deferred_dec_badcount(struct clone_struct *cs)
752 {
753         if (!cs->save_dup_cluster)
754                 return;
755         decrement_badcount(cs->ctx, cs->save_blocknr, cs->save_dup_cluster);
756         cs->save_dup_cluster = NULL;
757 }
758
759 static int clone_file_block(ext2_filsys fs,
760                             blk64_t     *block_nr,
761                             e2_blkcnt_t blockcnt,
762                             blk64_t ref_block EXT2FS_ATTR((unused)),
763                             int ref_offset EXT2FS_ATTR((unused)),
764                             void *priv_data)
765 {
766         struct dup_cluster *p = NULL;
767         blk64_t new_block;
768         errcode_t       retval;
769         struct clone_struct *cs = (struct clone_struct *) priv_data;
770         dnode_t *n;
771         e2fsck_t ctx;
772         blk64_t c;
773         int is_meta = 0;
774
775         ctx = cs->ctx;
776         deferred_dec_badcount(cs);
777
778         if (*block_nr == 0)
779                 return 0;
780
781         c = EXT2FS_B2C(fs, blockcnt);
782         if (check_if_fs_cluster(ctx, EXT2FS_B2C(fs, *block_nr)))
783                 is_meta = 1;
784
785         if (c == cs->dup_cluster && cs->alloc_block) {
786                 new_block = cs->alloc_block;
787                 goto got_block;
788         }
789
790         if (ext2fs_test_block_bitmap2(ctx->block_dup_map, *block_nr)) {
791                 n = dict_lookup(&clstr_dict,
792                                 INT_TO_VOIDPTR(EXT2FS_B2C(fs, *block_nr)));
793                 if (!n) {
794                         com_err("clone_file_block", 0,
795                             _("internal error: can't find dup_blk for %llu\n"),
796                                 *block_nr);
797                         return 0;
798                 }
799
800                 p = (struct dup_cluster *) dnode_get(n);
801
802                 cs->dup_cluster = c;
803                 /*
804                  * Let's try an implied cluster allocation.  If we get the same
805                  * cluster back, then we need to find a new block; otherwise,
806                  * we're merely fixing the problem of one logical cluster being
807                  * mapped to multiple physical clusters.
808                  */
809                 new_block = 0;
810                 retval = ext2fs_map_cluster_block(fs, cs->ino,
811                                                   EXT2_INODE(cs->inode),
812                                                   blockcnt, &new_block);
813                 if (retval == 0 && new_block != 0 &&
814                     EXT2FS_B2C(ctx->fs, new_block) !=
815                     EXT2FS_B2C(ctx->fs, *block_nr))
816                         goto cluster_alloc_ok;
817                 retval = ext2fs_new_block2(fs, 0, ctx->block_found_map,
818                                            &new_block);
819                 if (retval) {
820                         cs->errcode = retval;
821                         return BLOCK_ABORT;
822                 }
823 cluster_alloc_ok:
824                 cs->alloc_block = new_block;
825
826         got_block:
827                 new_block &= ~EXT2FS_CLUSTER_MASK(fs);
828                 new_block += EXT2FS_CLUSTER_MASK(fs) & blockcnt;
829                 if (cs->dir && (blockcnt >= 0)) {
830                         retval = ext2fs_set_dir_block2(fs->dblist,
831                                         cs->dir, new_block, blockcnt);
832                         if (retval) {
833                                 cs->errcode = retval;
834                                 return BLOCK_ABORT;
835                         }
836                 }
837 #if 0
838                 printf("Cloning block #%lld from %llu to %llu\n",
839                        blockcnt, *block_nr, new_block);
840 #endif
841                 retval = io_channel_read_blk64(fs->io, *block_nr, 1, cs->buf);
842                 if (retval) {
843                         cs->errcode = retval;
844                         return BLOCK_ABORT;
845                 }
846                 retval = io_channel_write_blk64(fs->io, new_block, 1, cs->buf);
847                 if (retval) {
848                         cs->errcode = retval;
849                         return BLOCK_ABORT;
850                 }
851                 cs->save_dup_cluster = (is_meta ? NULL : p);
852                 cs->save_blocknr = *block_nr;
853                 *block_nr = new_block;
854                 ext2fs_mark_block_bitmap2(ctx->block_found_map, new_block);
855                 ext2fs_mark_block_bitmap2(fs->block_map, new_block);
856                 return BLOCK_CHANGED;
857         }
858         return 0;
859 }
860
861 static errcode_t clone_file(e2fsck_t ctx, ext2_ino_t ino,
862                             struct dup_inode *dp, char* block_buf)
863 {
864         ext2_filsys fs = ctx->fs;
865         errcode_t       retval;
866         struct clone_struct cs;
867         struct problem_context  pctx;
868         blk64_t         blk, new_blk;
869         dnode_t         *n;
870         struct inode_el *ino_el;
871         struct dup_cluster      *dc;
872         struct dup_inode        *di;
873
874         clear_problem_context(&pctx);
875         cs.errcode = 0;
876         cs.dir = 0;
877         cs.dup_cluster = ~0;
878         cs.alloc_block = 0;
879         cs.ctx = ctx;
880         cs.ino = ino;
881         cs.inode = &dp->inode;
882         cs.save_dup_cluster = NULL;
883         cs.save_blocknr = 0;
884         retval = ext2fs_get_mem(fs->blocksize, &cs.buf);
885         if (retval)
886                 return retval;
887
888         if (ext2fs_test_inode_bitmap2(ctx->inode_dir_map, ino))
889                 cs.dir = ino;
890
891         pctx.ino = ino;
892         pctx.str = "clone_file";
893         if (ext2fs_inode_has_valid_blocks2(fs, EXT2_INODE(&dp->inode)))
894                 pctx.errcode = ext2fs_block_iterate3(fs, ino, 0, block_buf,
895                                                      clone_file_block, &cs);
896         deferred_dec_badcount(&cs);
897         ext2fs_mark_bb_dirty(fs);
898         if (pctx.errcode) {
899                 fix_problem(ctx, PR_1B_BLOCK_ITERATE, &pctx);
900                 retval = pctx.errcode;
901                 goto errout;
902         }
903         if (cs.errcode) {
904                 com_err("clone_file", cs.errcode, "%s",
905                         _("returned from clone_file_block"));
906                 retval = cs.errcode;
907                 goto errout;
908         }
909         /* The inode may have changed on disk, so we have to re-read it */
910         e2fsck_read_inode_full(ctx, ino, EXT2_INODE(&dp->inode),
911                                sizeof(dp->inode), "clone file EA");
912         blk = ext2fs_file_acl_block(fs, EXT2_INODE(&dp->inode));
913         new_blk = blk;
914         if (blk && (clone_file_block(fs, &new_blk,
915                                      BLOCK_COUNT_EXTATTR, 0, 0, &cs) ==
916                     BLOCK_CHANGED)) {
917                 ext2fs_file_acl_block_set(fs, EXT2_INODE(&dp->inode), new_blk);
918                 e2fsck_write_inode_full(ctx, ino, EXT2_INODE(&dp->inode),
919                                         sizeof(dp->inode), "clone file EA");
920                 /*
921                  * If we cloned the EA block, find all other inodes
922                  * which referred to that EA block, and modify
923                  * them to point to the new EA block.
924                  */
925                 n = dict_lookup(&clstr_dict,
926                                 INT_TO_VOIDPTR(EXT2FS_B2C(fs, blk)));
927                 if (!n) {
928                         com_err("clone_file", 0,
929                                 _("internal error: couldn't lookup EA "
930                                   "block record for %llu"), blk);
931                         retval = 0; /* OK to stumble on... */
932                         goto errout;
933                 }
934                 dc = (struct dup_cluster *) dnode_get(n);
935                 for (ino_el = dc->inode_list; ino_el; ino_el = ino_el->next) {
936                         if (ino_el->inode == ino)
937                                 continue;
938                         n = dict_lookup(&ino_dict, INT_TO_VOIDPTR(ino_el->inode));
939                         if (!n) {
940                                 com_err("clone_file", 0,
941                                         _("internal error: couldn't lookup EA "
942                                           "inode record for %u"),
943                                         ino_el->inode);
944                                 retval = 0; /* OK to stumble on... */
945                                 goto errout;
946                         }
947                         di = (struct dup_inode *) dnode_get(n);
948                         if (ext2fs_file_acl_block(fs,
949                                         EXT2_INODE(&di->inode)) == blk) {
950                                 ext2fs_file_acl_block_set(fs,
951                                         EXT2_INODE(&di->inode),
952                                         ext2fs_file_acl_block(fs, EXT2_INODE(&dp->inode)));
953                                 e2fsck_write_inode_full(ctx, ino_el->inode,
954                                         EXT2_INODE(&di->inode),
955                                         sizeof(di->inode), "clone file EA");
956                                 decrement_badcount(ctx, blk, dc);
957                         }
958                 }
959         }
960         retval = 0;
961 errout:
962         ext2fs_free_mem(&cs.buf);
963         return retval;
964 }
965
966 /*
967  * This routine returns 1 if a block overlaps with one of the superblocks,
968  * group descriptors, inode bitmaps, or block bitmaps.
969  */
970 static int check_if_fs_block(e2fsck_t ctx, blk64_t test_block)
971 {
972         ext2_filsys fs = ctx->fs;
973         blk64_t first_block;
974         dgrp_t  i;
975
976         first_block = fs->super->s_first_data_block;
977         for (i = 0; i < fs->group_desc_count; i++) {
978
979                 /* Check superblocks/block group descriptors */
980                 if (ext2fs_bg_has_super(fs, i)) {
981                         if (test_block >= first_block &&
982                             (test_block <= first_block + fs->desc_blocks))
983                                 return 1;
984                 }
985
986                 /* Check the inode table */
987                 if ((ext2fs_inode_table_loc(fs, i)) &&
988                     (test_block >= ext2fs_inode_table_loc(fs, i)) &&
989                     (test_block < (ext2fs_inode_table_loc(fs, i) +
990                                    fs->inode_blocks_per_group)))
991                         return 1;
992
993                 /* Check the bitmap blocks */
994                 if ((test_block == ext2fs_block_bitmap_loc(fs, i)) ||
995                     (test_block == ext2fs_inode_bitmap_loc(fs, i)))
996                         return 1;
997
998                 first_block += fs->super->s_blocks_per_group;
999         }
1000         return 0;
1001 }
1002
1003 /*
1004  * This routine returns 1 if a cluster overlaps with one of the superblocks,
1005  * group descriptors, inode bitmaps, or block bitmaps.
1006  */
1007 static int check_if_fs_cluster(e2fsck_t ctx, blk64_t cluster)
1008 {
1009         ext2_filsys fs = ctx->fs;
1010         blk64_t first_block;
1011         dgrp_t  i;
1012
1013         first_block = fs->super->s_first_data_block;
1014         for (i = 0; i < fs->group_desc_count; i++) {
1015
1016                 /* Check superblocks/block group descriptors */
1017                 if (ext2fs_bg_has_super(fs, i)) {
1018                         if (cluster >= EXT2FS_B2C(fs, first_block) &&
1019                             (cluster <= EXT2FS_B2C(fs, first_block +
1020                                                    fs->desc_blocks)))
1021                                 return 1;
1022                 }
1023
1024                 /* Check the inode table */
1025                 if ((ext2fs_inode_table_loc(fs, i)) &&
1026                     (cluster >= EXT2FS_B2C(fs,
1027                                            ext2fs_inode_table_loc(fs, i))) &&
1028                     (cluster <= EXT2FS_B2C(fs,
1029                                            ext2fs_inode_table_loc(fs, i) +
1030                                            fs->inode_blocks_per_group - 1)))
1031                         return 1;
1032
1033                 /* Check the bitmap blocks */
1034                 if ((cluster == EXT2FS_B2C(fs,
1035                                            ext2fs_block_bitmap_loc(fs, i))) ||
1036                     (cluster == EXT2FS_B2C(fs,
1037                                            ext2fs_inode_bitmap_loc(fs, i))))
1038                         return 1;
1039
1040                 first_block += fs->super->s_blocks_per_group;
1041         }
1042         return 0;
1043 }