xref: /dragonfly/sbin/hammer/cmd_recover.c (revision 7ff0fc30)
1 /*
2  * Copyright (c) 2010 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #include "hammer.h"
36 
37 #include <sys/tree.h>
38 
39 struct recover_dict {
40 	struct recover_dict *next;
41 	struct recover_dict *parent;
42 	int64_t	obj_id;
43 	uint8_t obj_type;
44 	uint8_t flags;
45 	uint16_t pfs_id;
46 	int64_t	size;
47 	char	*name;
48 };
49 
50 #define DICTF_MADEDIR	0x01
51 #define DICTF_MADEFILE	0x02
52 #define DICTF_PARENT	0x04	/* parent attached for real */
53 #define DICTF_TRAVERSED	0x80
54 
55 typedef struct bigblock {
56 	RB_ENTRY(bigblock) entry;
57 	hammer_off_t phys_offset; /* zone-2 */
58 	struct hammer_blockmap_layer1 layer1;
59 	struct hammer_blockmap_layer2 layer2;
60 } *bigblock_t;
61 
62 static void recover_top(char *ptr, hammer_off_t offset);
63 static void recover_elm(hammer_btree_leaf_elm_t leaf);
64 static struct recover_dict *get_dict(int64_t obj_id, uint16_t pfs_id);
65 static char *recover_path(struct recover_dict *dict);
66 static void sanitize_string(char *str);
67 static hammer_off_t scan_raw_limit(void);
68 static void scan_bigblocks(int target_zone);
69 static void free_bigblocks(void);
70 static void add_bigblock_entry(hammer_off_t offset,
71 	hammer_blockmap_layer1_t layer1, hammer_blockmap_layer2_t layer2);
72 static bigblock_t get_bigblock_entry(hammer_off_t offset);
73 
74 static const char *TargetDir;
75 static int CachedFd = -1;
76 static char *CachedPath;
77 
78 static int
79 bigblock_cmp(bigblock_t b1, bigblock_t b2)
80 {
81 	if (b1->phys_offset < b2->phys_offset)
82 		return(-1);
83 	if (b1->phys_offset > b2->phys_offset)
84 		return(1);
85 	return(0);
86 }
87 
88 static RB_HEAD(bigblock_rb_tree, bigblock) ZoneTree =
89 					RB_INITIALIZER(&ZoneTree);
90 RB_PROTOTYPE2(bigblock_rb_tree, bigblock, entry, bigblock_cmp, hammer_off_t);
91 RB_GENERATE2(bigblock_rb_tree, bigblock, entry, bigblock_cmp, hammer_off_t,
92 	phys_offset);
93 
94 /*
95  * There was a hidden bug here while iterating zone-2 offset as
96  * shown in an example below.
97  *
98  * If a volume was once used as HAMMER filesystem which consists of
99  * multiple volumes whose usage has reached beyond the first volume,
100  * and then later re-formatted only using 1 volume, hammer recover is
101  * likely to hit assertion in get_buffer() due to having access to
102  * invalid volume (vol1,2,...) from old filesystem data.
103  *
104  * To avoid this, now the command only scans upto the last big-block
105  * that's actually used for filesystem data or meta-data at the moment,
106  * if all layer1/2 entries have correct CRC values. This also avoids
107  * recovery of irrelevant files from old filesystem.
108  *
109  * It also doesn't scan beyond append offset of big-blocks in B-Tree
110  * zone to avoid recovery of irrelevant files from old filesystem,
111  * if layer1/2 entries for those big-blocks have correct CRC values.
112  *
113  * |-----vol0-----|-----vol1-----|-----vol2-----| old filesystem
114  * <-----------------------> used by old filesystem
115  *
116  * |-----vol0-----| new filesystem
117  * <-----> used by new filesystem
118  *        <-------> unused, invalid data from old filesystem
119  *              <-> B-Tree nodes likely to point to vol1
120  */
121 
122 void
123 hammer_cmd_recover(char **av, int ac)
124 {
125 	buffer_info_t data_buffer;
126 	volume_info_t volume;
127 	bigblock_t b = NULL;
128 	hammer_off_t off;
129 	hammer_off_t off_end;
130 	hammer_off_t off_blk;
131 	hammer_off_t raw_limit = 0;
132 	hammer_off_t zone_limit = 0;
133 	char *ptr;
134 	int i;
135 	int target_zone = HAMMER_ZONE_BTREE_INDEX;
136 	int full = 0;
137 	int quick = 0;
138 
139 	if (ac < 1) {
140 		errx(1, "hammer recover <target_dir> [full|quick]");
141 		/* not reached */
142 	}
143 
144 	TargetDir = av[0];
145 	if (ac > 1) {
146 		if (!strcmp(av[1], "full"))
147 			full = 1;
148 		if (!strcmp(av[1], "quick"))
149 			quick = 1;
150 	}
151 	assert(!full || !quick);
152 
153 	if (mkdir(TargetDir, 0777) == -1) {
154 		if (errno != EEXIST) {
155 			err(1, "mkdir");
156 			/* not reached */
157 		}
158 	}
159 
160 	printf("Running %sraw scan of HAMMER image, recovering to %s\n",
161 		full ? "full " : quick ? "quick " : "",
162 		TargetDir);
163 
164 	if (!full) {
165 		scan_bigblocks(target_zone);
166 		raw_limit = scan_raw_limit();
167 		if (raw_limit) {
168 			raw_limit += HAMMER_BIGBLOCK_SIZE;
169 			assert(hammer_is_zone_raw_buffer(raw_limit));
170 		}
171 	}
172 
173 	if (quick) {
174 		assert(!full);
175 		if (!RB_EMPTY(&ZoneTree)) {
176 			printf("Found zone-%d big-blocks at\n", target_zone);
177 			RB_FOREACH(b, bigblock_rb_tree, &ZoneTree)
178 				printf("%016jx\n", b->phys_offset);
179 
180 			b = RB_MAX(bigblock_rb_tree, &ZoneTree);
181 			zone_limit = b->phys_offset + HAMMER_BIGBLOCK_SIZE;
182 			assert(hammer_is_zone_raw_buffer(zone_limit));
183 		}
184 	}
185 
186 	if (raw_limit || zone_limit) {
187 #define _fmt "Scanning zone-%d big-blocks till %016jx"
188 		if (!raw_limit) /* unlikely */
189 			printf(_fmt" ???", target_zone, zone_limit);
190 		else if (!zone_limit)
191 			printf(_fmt, HAMMER_ZONE_RAW_BUFFER_INDEX, raw_limit);
192 		else if (raw_limit >= zone_limit)
193 			printf(_fmt, target_zone, zone_limit);
194 		else /* unlikely */
195 			printf(_fmt" ???", HAMMER_ZONE_RAW_BUFFER_INDEX, raw_limit);
196 		printf("\n");
197 	}
198 
199 	data_buffer = NULL;
200 	for (i = 0; i < HAMMER_MAX_VOLUMES; i++) {
201 		volume = get_volume(i);
202 		if (volume == NULL)
203 			continue;
204 
205 		printf("Scanning volume %d size %s\n",
206 			volume->vol_no, sizetostr(volume->size));
207 		off = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no, 0);
208 		off_end = off + HAMMER_VOL_BUF_SIZE(volume->ondisk);
209 
210 		while (off < off_end) {
211 			off_blk = off & HAMMER_BIGBLOCK_MASK64;
212 			if (off_blk == 0)
213 				b = get_bigblock_entry(off);
214 
215 			if (raw_limit) {
216 				if (off >= raw_limit) {
217 					printf("Done %016jx\n", (uintmax_t)off);
218 					goto end;
219 				}
220 			}
221 			if (zone_limit) {
222 				if (off >= zone_limit) {
223 					printf("Done %016jx\n", (uintmax_t)off);
224 					goto end;
225 				}
226 				if (b == NULL) {
227 					off = HAMMER_ZONE_LAYER2_NEXT_OFFSET(off);
228 					continue;
229 				}
230 			}
231 
232 			if (b) {
233 				if (hammer_crc_test_layer1(HammerVersion,
234 							   &b->layer1) &&
235 				    hammer_crc_test_layer2(HammerVersion,
236 							   &b->layer2) &&
237 				    off_blk >= b->layer2.append_off) {
238 					off = HAMMER_ZONE_LAYER2_NEXT_OFFSET(off);
239 					continue;
240 				}
241 			}
242 
243 			ptr = get_buffer_data(off, &data_buffer, 0);
244 			if (ptr)
245 				recover_top(ptr, off);
246 			off += HAMMER_BUFSIZE;
247 		}
248 	}
249 end:
250 	rel_buffer(data_buffer);
251 	free_bigblocks();
252 
253 	if (CachedPath) {
254 		free(CachedPath);
255 		close(CachedFd);
256 		CachedPath = NULL;
257 		CachedFd = -1;
258 	}
259 }
260 
261 static __inline
262 void
263 print_node(hammer_node_ondisk_t node, hammer_off_t offset)
264 {
265 	char buf[HAMMER_BTREE_LEAF_ELMS + 1];
266 	int maxcount = hammer_node_max_elements(node->type);
267 	int i;
268 
269 	for (i = 0; i < node->count && i < maxcount; ++i)
270 		buf[i] = hammer_elm_btype(&node->elms[i]);
271 	buf[i] = '\0';
272 
273 	printf("%016jx %c %d %s\n", offset, node->type, node->count, buf);
274 }
275 
276 /*
277  * Top level recovery processor.  Assume the data is a B-Tree node.
278  * If the CRC is good we attempt to process the node, building the
279  * object space and creating the dictionary as we go.
280  */
281 static
282 void
283 recover_top(char *ptr, hammer_off_t offset)
284 {
285 	hammer_node_ondisk_t node;
286 	hammer_btree_elm_t elm;
287 	int maxcount;
288 	int i;
289 	int isnode;
290 
291 	for (node = (void *)ptr; (char *)node < ptr + HAMMER_BUFSIZE; ++node) {
292 		isnode = hammer_crc_test_btree(HammerVersion, node);
293 		maxcount = hammer_node_max_elements(node->type);
294 
295 		if (DebugOpt) {
296 			if (isnode)
297 				print_node(node, offset);
298 			else if (DebugOpt > 1)
299 				printf("%016jx -\n", offset);
300 		}
301 		offset += sizeof(*node);
302 
303 		if (isnode && node->type == HAMMER_BTREE_TYPE_LEAF) {
304 			for (i = 0; i < node->count && i < maxcount; ++i) {
305 				elm = &node->elms[i];
306 				if (elm->base.btype == HAMMER_BTREE_TYPE_RECORD)
307 					recover_elm(&elm->leaf);
308 			}
309 		}
310 	}
311 }
312 
313 static
314 void
315 recover_elm(hammer_btree_leaf_elm_t leaf)
316 {
317 	buffer_info_t data_buffer = NULL;
318 	struct recover_dict *dict;
319 	struct recover_dict *dict2;
320 	hammer_data_ondisk_t ondisk;
321 	hammer_off_t data_offset;
322 	struct stat st;
323 	int chunk;
324 	int len;
325 	int zfill;
326 	int64_t file_offset;
327 	uint16_t pfs_id;
328 	size_t nlen;
329 	int fd;
330 	char *name;
331 	char *path1;
332 	char *path2;
333 
334 	/*
335 	 * Ignore deleted records
336 	 */
337 	if (leaf->delete_ts)
338 		return;
339 
340 	/*
341 	 * If we're running full scan, it's possible that data_offset
342 	 * refers to old filesystem data that we can't physically access.
343 	 */
344 	data_offset = leaf->data_offset;
345 	if (get_volume(HAMMER_VOL_DECODE(data_offset)) == NULL)
346 		return;
347 
348 	if (data_offset != 0)
349 		ondisk = get_buffer_data(data_offset, &data_buffer, 0);
350 	else
351 		ondisk = NULL;
352 	if (ondisk == NULL)
353 		goto done;
354 
355 	len = leaf->data_len;
356 	chunk = HAMMER_BUFSIZE - ((int)data_offset & HAMMER_BUFMASK);
357 	if (chunk > len)
358 		chunk = len;
359 
360 	if (len < 0 || len > HAMMER_XBUFSIZE || len > chunk)
361 		goto done;
362 
363 	pfs_id = lo_to_pfs(leaf->base.localization);
364 
365 	/*
366 	 * Note that meaning of leaf->base.obj_id differs depending
367 	 * on record type.  For a direntry, leaf->base.obj_id points
368 	 * to its parent inode that this entry is a part of, but not
369 	 * its corresponding inode.
370 	 */
371 	dict = get_dict(leaf->base.obj_id, pfs_id);
372 
373 	switch(leaf->base.rec_type) {
374 	case HAMMER_RECTYPE_INODE:
375 		/*
376 		 * We found an inode which also tells us where the file
377 		 * or directory is in the directory hierarchy.
378 		 */
379 		if (VerboseOpt) {
380 			printf("inode %016jx:%05d found\n",
381 				(uintmax_t)leaf->base.obj_id, pfs_id);
382 		}
383 		path1 = recover_path(dict);
384 
385 		/*
386 		 * Attach the inode to its parent.  This isn't strictly
387 		 * necessary because the information is also in the
388 		 * directory entries, but if we do not find the directory
389 		 * entry this ensures that the files will still be
390 		 * reasonably well organized in their proper directories.
391 		 */
392 		if ((dict->flags & DICTF_PARENT) == 0 &&
393 		    dict->obj_id != HAMMER_OBJID_ROOT &&
394 		    ondisk->inode.parent_obj_id != 0) {
395 			dict->flags |= DICTF_PARENT;
396 			dict->parent = get_dict(ondisk->inode.parent_obj_id,
397 						pfs_id);
398 			if (dict->parent &&
399 			    (dict->parent->flags & DICTF_MADEDIR) == 0) {
400 				dict->parent->flags |= DICTF_MADEDIR;
401 				path2 = recover_path(dict->parent);
402 				printf("mkdir %s\n", path2);
403 				mkdir(path2, 0777);
404 				free(path2);
405 				path2 = NULL;
406 			}
407 		}
408 		if (dict->obj_type == 0)
409 			dict->obj_type = ondisk->inode.obj_type;
410 		dict->size = ondisk->inode.size;
411 		path2 = recover_path(dict);
412 
413 		if (lstat(path1, &st) == 0) {
414 			if (ondisk->inode.obj_type == HAMMER_OBJTYPE_REGFILE) {
415 				truncate(path1, dict->size);
416 				/* chmod(path1, 0666); */
417 			}
418 			if (strcmp(path1, path2)) {
419 				printf("Rename (inode) %s -> %s\n", path1, path2);
420 				rename(path1, path2);
421 			}
422 		} else if (ondisk->inode.obj_type == HAMMER_OBJTYPE_REGFILE) {
423 			printf("mkinode (file) %s\n", path2);
424 			fd = open(path2, O_RDWR|O_CREAT, 0666);
425 			if (fd > 0)
426 				close(fd);
427 		} else if (ondisk->inode.obj_type == HAMMER_OBJTYPE_DIRECTORY) {
428 			printf("mkinode (dir) %s\n", path2);
429 			mkdir(path2, 0777);
430 			dict->flags |= DICTF_MADEDIR;
431 		}
432 		free(path1);
433 		free(path2);
434 		break;
435 	case HAMMER_RECTYPE_DATA:
436 		/*
437 		 * File record data
438 		 */
439 		if (leaf->base.obj_id == 0)
440 			break;
441 		if (VerboseOpt) {
442 			printf("inode %016jx:%05d data %016jx,%d\n",
443 				(uintmax_t)leaf->base.obj_id,
444 				pfs_id,
445 				(uintmax_t)leaf->base.key - len,
446 				len);
447 		}
448 
449 		/*
450 		 * Update the dictionary entry
451 		 */
452 		if (dict->obj_type == 0)
453 			dict->obj_type = HAMMER_OBJTYPE_REGFILE;
454 
455 		/*
456 		 * If the parent directory has not been created we
457 		 * have to create it (typically a PFS%05d)
458 		 */
459 		if (dict->parent &&
460 		    (dict->parent->flags & DICTF_MADEDIR) == 0) {
461 			dict->parent->flags |= DICTF_MADEDIR;
462 			path2 = recover_path(dict->parent);
463 			printf("mkdir %s\n", path2);
464 			mkdir(path2, 0777);
465 			free(path2);
466 			path2 = NULL;
467 		}
468 
469 		/*
470 		 * Create the file if necessary, report file creations
471 		 */
472 		path1 = recover_path(dict);
473 		if (CachedPath && strcmp(CachedPath, path1) == 0)
474 			fd = CachedFd;
475 		else
476 			fd = open(path1, O_CREAT|O_RDWR, 0666);
477 		if (fd < 0) {
478 			printf("Unable to create %s: %s\n",
479 				path1, strerror(errno));
480 			free(path1);
481 			break;
482 		}
483 		if ((dict->flags & DICTF_MADEFILE) == 0) {
484 			dict->flags |= DICTF_MADEFILE;
485 			printf("mkfile %s\n", path1);
486 		}
487 
488 		/*
489 		 * And write the record.  A HAMMER data block is aligned
490 		 * and may contain trailing zeros after the file EOF.  The
491 		 * inode record is required to get the actual file size.
492 		 *
493 		 * However, when the inode record is not available
494 		 * we can do a sparse write and that will get it right
495 		 * most of the time even if the inode record is never
496 		 * found.
497 		 */
498 		file_offset = (int64_t)leaf->base.key - len;
499 		lseek(fd, (off_t)file_offset, SEEK_SET);
500 		while (len) {
501 			if (dict->size == -1) {
502 				for (zfill = chunk - 1; zfill >= 0; --zfill) {
503 					if (((char *)ondisk)[zfill])
504 						break;
505 				}
506 				++zfill;
507 			} else {
508 				zfill = chunk;
509 			}
510 
511 			if (zfill)
512 				write(fd, ondisk, zfill);
513 			if (zfill < chunk)
514 				lseek(fd, chunk - zfill, SEEK_CUR);
515 
516 			len -= chunk;
517 			data_offset += chunk;
518 			file_offset += chunk;
519 			ondisk = get_buffer_data(data_offset, &data_buffer, 0);
520 			if (ondisk == NULL)
521 				break;
522 			chunk = HAMMER_BUFSIZE -
523 				((int)data_offset & HAMMER_BUFMASK);
524 			if (chunk > len)
525 				chunk = len;
526 		}
527 		if (dict->size >= 0 && file_offset > dict->size) {
528 			ftruncate(fd, dict->size);
529 			/* fchmod(fd, 0666); */
530 		}
531 
532 		if (fd == CachedFd) {
533 			free(path1);
534 		} else if (CachedPath) {
535 			free(CachedPath);
536 			close(CachedFd);
537 			CachedPath = path1;
538 			CachedFd = fd;
539 		} else {
540 			CachedPath = path1;
541 			CachedFd = fd;
542 		}
543 		break;
544 	case HAMMER_RECTYPE_DIRENTRY:
545 		nlen = len - HAMMER_ENTRY_NAME_OFF;
546 		if ((int)nlen < 0)	/* illegal length */
547 			break;
548 		if (ondisk->entry.obj_id == 0 ||
549 		    ondisk->entry.obj_id == HAMMER_OBJID_ROOT) {
550 			break;
551 		}
552 		name = malloc(nlen + 1);
553 		bcopy(ondisk->entry.name, name, nlen);
554 		name[nlen] = 0;
555 		sanitize_string(name);
556 
557 		if (VerboseOpt) {
558 			printf("dir %016jx:%05d entry %016jx \"%s\"\n",
559 				(uintmax_t)leaf->base.obj_id,
560 				pfs_id,
561 				(uintmax_t)ondisk->entry.obj_id,
562 				name);
563 		}
564 
565 		/*
566 		 * We can't deal with hardlinks so if the object already
567 		 * has a name assigned to it we just keep using that name.
568 		 */
569 		dict2 = get_dict(ondisk->entry.obj_id, pfs_id);
570 		path1 = recover_path(dict2);
571 
572 		if (dict2->name == NULL)
573 			dict2->name = name;
574 		else
575 			free(name);
576 
577 		/*
578 		 * Attach dict2 to its directory (dict), create the
579 		 * directory (dict) if necessary.  We must ensure
580 		 * that the directory entry exists in order to be
581 		 * able to properly rename() the file without creating
582 		 * a namespace conflict.
583 		 */
584 		if ((dict2->flags & DICTF_PARENT) == 0) {
585 			dict2->flags |= DICTF_PARENT;
586 			dict2->parent = dict;
587 			if ((dict->flags & DICTF_MADEDIR) == 0) {
588 				dict->flags |= DICTF_MADEDIR;
589 				path2 = recover_path(dict);
590 				printf("mkdir %s\n", path2);
591 				mkdir(path2, 0777);
592 				free(path2);
593 				path2 = NULL;
594 			}
595 		}
596 		path2 = recover_path(dict2);
597 		if (strcmp(path1, path2) != 0 && lstat(path1, &st) == 0) {
598 			printf("Rename (entry) %s -> %s\n", path1, path2);
599 			rename(path1, path2);
600 		}
601 		free(path1);
602 		free(path2);
603 		break;
604 	default:
605 		/*
606 		 * Ignore any other record types
607 		 */
608 		break;
609 	}
610 done:
611 	rel_buffer(data_buffer);
612 }
613 
614 #define RD_HSIZE	32768
615 #define RD_HMASK	(RD_HSIZE - 1)
616 
617 static struct recover_dict *RDHash[RD_HSIZE];
618 
619 static
620 struct recover_dict *
621 get_dict(int64_t obj_id, uint16_t pfs_id)
622 {
623 	struct recover_dict *dict;
624 	int i;
625 
626 	if (obj_id == 0)
627 		return(NULL);
628 
629 	i = crc32(&obj_id, sizeof(obj_id)) & RD_HMASK;
630 	for (dict = RDHash[i]; dict; dict = dict->next) {
631 		if (dict->obj_id == obj_id && dict->pfs_id == pfs_id)
632 			break;
633 	}
634 
635 	if (dict == NULL) {
636 		dict = malloc(sizeof(*dict));
637 		bzero(dict, sizeof(*dict));
638 		dict->obj_id = obj_id;
639 		dict->pfs_id = pfs_id;
640 		dict->next = RDHash[i];
641 		dict->size = -1;
642 		RDHash[i] = dict;
643 
644 		/*
645 		 * Always connect dangling dictionary entries to object 1
646 		 * (the root of the PFS).
647 		 *
648 		 * DICTF_PARENT will not be set until we know what the
649 		 * real parent directory object is.
650 		 */
651 		if (dict->obj_id != HAMMER_OBJID_ROOT)
652 			dict->parent = get_dict(HAMMER_OBJID_ROOT, pfs_id);
653 	}
654 	return(dict);
655 }
656 
657 struct path_info {
658 	enum { PI_FIGURE, PI_LOAD } state;
659 	uint16_t pfs_id;
660 	char *base;
661 	char *next;
662 	int len;
663 };
664 
665 static void recover_path_helper(struct recover_dict *, struct path_info *);
666 
667 static
668 char *
669 recover_path(struct recover_dict *dict)
670 {
671 	struct path_info info;
672 
673 	/* Find info.len first */
674 	bzero(&info, sizeof(info));
675 	info.state = PI_FIGURE;
676 	recover_path_helper(dict, &info);
677 
678 	/* Fill in the path */
679 	info.pfs_id = dict->pfs_id;
680 	info.base = malloc(info.len);
681 	info.next = info.base;
682 	info.state = PI_LOAD;
683 	recover_path_helper(dict, &info);
684 
685 	/* Return the path */
686 	return(info.base);
687 }
688 
689 #define STRLEN_OBJID	22	/* "obj_0x%016jx" */
690 #define STRLEN_PFSID	8	/* "PFS%05d" */
691 
692 static
693 void
694 recover_path_helper(struct recover_dict *dict, struct path_info *info)
695 {
696 	/*
697 	 * Calculate path element length
698 	 */
699 	dict->flags |= DICTF_TRAVERSED;
700 
701 	switch(info->state) {
702 	case PI_FIGURE:
703 		if (dict->obj_id == HAMMER_OBJID_ROOT)
704 			info->len += STRLEN_PFSID;
705 		else if (dict->name)
706 			info->len += strlen(dict->name);
707 		else
708 			info->len += STRLEN_OBJID;
709 		++info->len;
710 
711 		if (dict->parent &&
712 		    (dict->parent->flags & DICTF_TRAVERSED) == 0) {
713 			recover_path_helper(dict->parent, info);
714 		} else {
715 			info->len += strlen(TargetDir) + 1;
716 		}
717 		break;
718 	case PI_LOAD:
719 		if (dict->parent &&
720 		    (dict->parent->flags & DICTF_TRAVERSED) == 0) {
721 			recover_path_helper(dict->parent, info);
722 		} else {
723 			strcpy(info->next, TargetDir);
724 			info->next += strlen(info->next);
725 		}
726 
727 		*info->next++ = '/';
728 		if (dict->obj_id == HAMMER_OBJID_ROOT) {
729 			snprintf(info->next, STRLEN_PFSID + 1,
730 				"PFS%05d", info->pfs_id);
731 		} else if (dict->name) {
732 			strcpy(info->next, dict->name);
733 		} else {
734 			snprintf(info->next, STRLEN_OBJID + 1,
735 				"obj_0x%016jx", (uintmax_t)dict->obj_id);
736 		}
737 		info->next += strlen(info->next);
738 		break;
739 	}
740 	dict->flags &= ~DICTF_TRAVERSED;
741 }
742 
743 static
744 void
745 sanitize_string(char *str)
746 {
747 	while (*str) {
748 		if (!isprint(*str))
749 			*str = 'x';
750 		++str;
751 	}
752 }
753 
754 static
755 hammer_off_t
756 scan_raw_limit(void)
757 {
758 	volume_info_t volume;
759 	hammer_blockmap_t rootmap;
760 	hammer_blockmap_layer1_t layer1;
761 	hammer_blockmap_layer2_t layer2;
762 	buffer_info_t buffer1 = NULL;
763 	buffer_info_t buffer2 = NULL;
764 	hammer_off_t layer1_offset;
765 	hammer_off_t layer2_offset;
766 	hammer_off_t phys_offset;
767 	hammer_off_t block_offset;
768 	hammer_off_t offset = 0;
769 	int zone = HAMMER_ZONE_FREEMAP_INDEX;
770 
771 	volume = get_root_volume();
772 	rootmap = &volume->ondisk->vol0_blockmap[zone];
773 	assert(rootmap->phys_offset != 0);
774 
775 	for (phys_offset = HAMMER_ZONE_ENCODE(zone, 0);
776 	     phys_offset < HAMMER_ZONE_ENCODE(zone, HAMMER_OFF_LONG_MASK);
777 	     phys_offset += HAMMER_BLOCKMAP_LAYER2) {
778 		/*
779 		 * Dive layer 1.
780 		 */
781 		layer1_offset = rootmap->phys_offset +
782 				HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
783 		layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
784 
785 		if (!hammer_crc_test_layer1(HammerVersion, layer1)) {
786 			offset = 0; /* failed */
787 			goto end;
788 		}
789 		if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL)
790 			continue;
791 
792 		for (block_offset = 0;
793 		     block_offset < HAMMER_BLOCKMAP_LAYER2;
794 		     block_offset += HAMMER_BIGBLOCK_SIZE) {
795 			/*
796 			 * Dive layer 2, each entry represents a big-block.
797 			 */
798 			layer2_offset = layer1->phys_offset +
799 					HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
800 			layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
801 
802 			if (!hammer_crc_test_layer2(HammerVersion, layer2)) {
803 				offset = 0; /* failed */
804 				goto end;
805 			}
806 			if (layer2->zone == HAMMER_ZONE_UNAVAIL_INDEX) {
807 				break;
808 			} else if (layer2->zone && layer2->zone != zone) {
809 				offset = phys_offset + block_offset;
810 			}
811 		}
812 	}
813 end:
814 	rel_buffer(buffer1);
815 	rel_buffer(buffer2);
816 
817 	return(hammer_xlate_to_zone2(offset));
818 }
819 
820 static
821 void
822 scan_bigblocks(int target_zone)
823 {
824 	volume_info_t volume;
825 	hammer_blockmap_t rootmap;
826 	hammer_blockmap_layer1_t layer1;
827 	hammer_blockmap_layer2_t layer2;
828 	buffer_info_t buffer1 = NULL;
829 	buffer_info_t buffer2 = NULL;
830 	hammer_off_t layer1_offset;
831 	hammer_off_t layer2_offset;
832 	hammer_off_t phys_offset;
833 	hammer_off_t block_offset;
834 	hammer_off_t offset = 0;
835 	int zone = HAMMER_ZONE_FREEMAP_INDEX;
836 
837 	volume = get_root_volume();
838 	rootmap = &volume->ondisk->vol0_blockmap[zone];
839 	assert(rootmap->phys_offset != 0);
840 
841 	for (phys_offset = HAMMER_ZONE_ENCODE(zone, 0);
842 	     phys_offset < HAMMER_ZONE_ENCODE(zone, HAMMER_OFF_LONG_MASK);
843 	     phys_offset += HAMMER_BLOCKMAP_LAYER2) {
844 		/*
845 		 * Dive layer 1.
846 		 */
847 		layer1_offset = rootmap->phys_offset +
848 				HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
849 		layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
850 
851 		/*
852 		if (!hammer_crc_test_layer1(HammerVersion, layer1)) {
853 		}
854 		*/
855 		if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL)
856 			continue;
857 
858 		for (block_offset = 0;
859 		     block_offset < HAMMER_BLOCKMAP_LAYER2;
860 		     block_offset += HAMMER_BIGBLOCK_SIZE) {
861 			offset = phys_offset + block_offset;
862 			/*
863 			 * Dive layer 2, each entry represents a big-block.
864 			 */
865 			layer2_offset = layer1->phys_offset +
866 					HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
867 			layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
868 
869 			/*
870 			if (!hammer_crc_test_layer2(HammerVersion, layer2)) {
871 			}
872 			*/
873 			if (layer2->zone == target_zone) {
874 				add_bigblock_entry(offset, layer1, layer2);
875 			} else if (layer2->zone == HAMMER_ZONE_UNAVAIL_INDEX) {
876 				break;
877 			}
878 		}
879 	}
880 	rel_buffer(buffer1);
881 	rel_buffer(buffer2);
882 }
883 
884 static
885 void
886 free_bigblocks(void)
887 {
888 	bigblock_t b;
889 
890 	while ((b = RB_ROOT(&ZoneTree)) != NULL) {
891 		RB_REMOVE(bigblock_rb_tree, &ZoneTree, b);
892 		free(b);
893 	}
894 	assert(RB_EMPTY(&ZoneTree));
895 }
896 
897 static
898 void
899 add_bigblock_entry(hammer_off_t offset,
900 	hammer_blockmap_layer1_t layer1, hammer_blockmap_layer2_t layer2)
901 {
902 	bigblock_t b;
903 
904 	b = calloc(1, sizeof(*b));
905 	b->phys_offset = hammer_xlate_to_zone2(offset);
906 	assert((b->phys_offset & HAMMER_BIGBLOCK_MASK64) == 0);
907 	bcopy(layer1, &b->layer1, sizeof(*layer1));
908 	bcopy(layer2, &b->layer2, sizeof(*layer2));
909 
910 	RB_INSERT(bigblock_rb_tree, &ZoneTree, b);
911 }
912 
913 static
914 bigblock_t
915 get_bigblock_entry(hammer_off_t offset)
916 {
917 	bigblock_t b;
918 
919 	offset = hammer_xlate_to_zone2(offset);
920 	offset &= ~HAMMER_BIGBLOCK_MASK64;
921 
922 	b = RB_LOOKUP(bigblock_rb_tree, &ZoneTree, offset);
923 	if (b)
924 		return(b);
925 	return(NULL);
926 }
927