xref: /dragonfly/sbin/hammer/cmd_recover.c (revision d9f85b33)
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 *bigblock_t;
56 
57 static void recover_top(char *ptr, hammer_off_t offset);
58 static void recover_elm(hammer_btree_leaf_elm_t leaf);
59 static struct recover_dict *get_dict(int64_t obj_id, uint16_t pfs_id);
60 static char *recover_path(struct recover_dict *dict);
61 static void sanitize_string(char *str);
62 static hammer_off_t scan_raw_limit(void);
63 static void scan_bigblocks(int target_zone);
64 static void free_bigblocks(void);
65 static void add_bigblock_entry(hammer_off_t offset,
66 	hammer_blockmap_layer1_t layer1, hammer_blockmap_layer2_t layer2);
67 static bigblock_t get_bigblock_entry(hammer_off_t offset);
68 
69 static const char *TargetDir;
70 static int CachedFd = -1;
71 static char *CachedPath;
72 
73 typedef struct bigblock {
74 	RB_ENTRY(bigblock) entry;
75 	hammer_off_t phys_offset; /* zone-2 */
76 	struct hammer_blockmap_layer1 layer1;
77 	struct hammer_blockmap_layer2 layer2;
78 } *bigblock_t;
79 
80 static int
81 bigblock_cmp(bigblock_t b1, bigblock_t b2)
82 {
83 	if (b1->phys_offset < b2->phys_offset)
84 		return(-1);
85 	if (b1->phys_offset > b2->phys_offset)
86 		return(1);
87 	return(0);
88 }
89 
90 RB_HEAD(bigblock_rb_tree, bigblock) ZoneTree = RB_INITIALIZER(&ZoneTree);
91 RB_PROTOTYPE2(bigblock_rb_tree, bigblock, entry, bigblock_cmp, hammer_off_t);
92 RB_GENERATE2(bigblock_rb_tree, bigblock, entry, bigblock_cmp, hammer_off_t,
93 	phys_offset);
94 
95 /*
96  * There was a hidden bug here while iterating zone-2 offset as
97  * shown in an example below.
98  *
99  * If a volume was once used as HAMMER filesystem which consists of
100  * multiple volumes whose usage has reached beyond the first volume,
101  * and then later re-formatted only using 1 volume, hammer recover is
102  * likely to hit assertion in get_buffer() due to having access to
103  * invalid volume (vol1,2,...) from old filesystem data.
104  *
105  * To avoid this, now the command only scans upto the last big-block
106  * that's actually used for filesystem data or meta-data at the moment,
107  * if all layer1/2 entries have correct CRC values. This also avoids
108  * recovery of irrelevant files from old filesystem.
109  *
110  * It also doesn't scan beyond append offset of big-blocks in B-Tree
111  * zone to avoid recovery of irrelevant files from old filesystem,
112  * if layer1/2 entries for those big-blocks have correct CRC values.
113  *
114  * |-----vol0-----|-----vol1-----|-----vol2-----| old filesystem
115  * <-----------------------> used by old filesystem
116  *
117  * |-----vol0-----| new filesystem
118  * <-----> used by new filesystem
119  *        <-------> unused, invalid data from old filesystem
120  *              <-> B-Tree nodes likely to point to vol1
121  */
122 
123 void
124 hammer_cmd_recover(char **av, int ac)
125 {
126 	buffer_info_t data_buffer;
127 	volume_info_t volume;
128 	bigblock_t b = NULL;
129 	hammer_off_t off;
130 	hammer_off_t off_end;
131 	hammer_off_t off_blk;
132 	hammer_off_t raw_limit = 0;
133 	hammer_off_t zone_limit = 0;
134 	char *ptr;
135 	int i;
136 	int target_zone = HAMMER_ZONE_BTREE_INDEX;
137 	int full = 0;
138 	int quick = 0;
139 
140 	if (ac < 1) {
141 		errx(1, "hammer recover <target_dir> [full|quick]");
142 		/* not reached */
143 	}
144 
145 	TargetDir = av[0];
146 	if (ac > 1) {
147 		if (!strcmp(av[1], "full"))
148 			full = 1;
149 		if (!strcmp(av[1], "quick"))
150 			quick = 1;
151 	}
152 	assert(!full || !quick);
153 
154 	if (mkdir(TargetDir, 0777) == -1) {
155 		if (errno != EEXIST) {
156 			err(1, "mkdir");
157 			/* not reached */
158 		}
159 	}
160 
161 	printf("Running %sraw scan of HAMMER image, recovering to %s\n",
162 		full ? "full " : quick ? "quick " : "",
163 		TargetDir);
164 
165 	if (!full) {
166 		scan_bigblocks(target_zone);
167 		raw_limit = scan_raw_limit();
168 		if (raw_limit) {
169 			raw_limit += HAMMER_BIGBLOCK_SIZE;
170 			assert(hammer_is_zone_raw_buffer(raw_limit));
171 		}
172 	}
173 
174 	if (quick) {
175 		assert(!full);
176 		if (!RB_EMPTY(&ZoneTree)) {
177 			printf("Found zone-%d big-blocks at\n", target_zone);
178 			RB_FOREACH(b, bigblock_rb_tree, &ZoneTree)
179 				printf("%016jx\n", b->phys_offset);
180 
181 			b = RB_MAX(bigblock_rb_tree, &ZoneTree);
182 			zone_limit = b->phys_offset + HAMMER_BIGBLOCK_SIZE;
183 			assert(hammer_is_zone_raw_buffer(zone_limit));
184 		}
185 	}
186 
187 	if (raw_limit || zone_limit) {
188 #define _fmt "Scanning zone-%d big-blocks till %016jx"
189 		if (!raw_limit) /* unlikely */
190 			printf(_fmt" ???", target_zone, zone_limit);
191 		else if (!zone_limit)
192 			printf(_fmt, HAMMER_ZONE_RAW_BUFFER_INDEX, raw_limit);
193 		else if (raw_limit >= zone_limit)
194 			printf(_fmt, target_zone, zone_limit);
195 		else /* unlikely */
196 			printf(_fmt" ???", HAMMER_ZONE_RAW_BUFFER_INDEX, raw_limit);
197 		printf("\n");
198 	}
199 
200 	data_buffer = NULL;
201 	for (i = 0; i < HAMMER_MAX_VOLUMES; i++) {
202 		volume = get_volume(i);
203 		if (volume == NULL)
204 			continue;
205 
206 		printf("Scanning volume %d size %s\n",
207 			volume->vol_no, sizetostr(volume->size));
208 		off = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no, 0);
209 		off_end = off + HAMMER_VOL_BUF_SIZE(volume->ondisk);
210 
211 		while (off < off_end) {
212 			off_blk = off & HAMMER_BIGBLOCK_MASK64;
213 			if (off_blk == 0)
214 				b = get_bigblock_entry(off);
215 
216 			if (raw_limit) {
217 				if (off >= raw_limit) {
218 					printf("Done %016jx\n", (uintmax_t)off);
219 					goto end;
220 				}
221 			}
222 			if (zone_limit) {
223 				if (off >= zone_limit) {
224 					printf("Done %016jx\n", (uintmax_t)off);
225 					goto end;
226 				}
227 				if (b == NULL) {
228 					off = HAMMER_ZONE_LAYER2_NEXT_OFFSET(off);
229 					continue;
230 				}
231 			}
232 
233 			if (b) {
234 				if (hammer_crc_test_layer1(HammerVersion,
235 							   &b->layer1) &&
236 				    hammer_crc_test_layer2(HammerVersion,
237 							   &b->layer2) &&
238 				    off_blk >= b->layer2.append_off) {
239 					off = HAMMER_ZONE_LAYER2_NEXT_OFFSET(off);
240 					continue;
241 				}
242 			}
243 
244 			ptr = get_buffer_data(off, &data_buffer, 0);
245 			if (ptr)
246 				recover_top(ptr, off);
247 			off += HAMMER_BUFSIZE;
248 		}
249 	}
250 end:
251 	rel_buffer(data_buffer);
252 	free_bigblocks();
253 
254 	if (CachedPath) {
255 		free(CachedPath);
256 		close(CachedFd);
257 		CachedPath = NULL;
258 		CachedFd = -1;
259 	}
260 }
261 
262 static __inline
263 void
264 print_node(hammer_node_ondisk_t node, hammer_off_t offset)
265 {
266 	char buf[HAMMER_BTREE_LEAF_ELMS + 1];
267 	int maxcount = hammer_node_max_elements(node->type);
268 	int i;
269 
270 	for (i = 0; i < node->count && i < maxcount; ++i)
271 		buf[i] = hammer_elm_btype(&node->elms[i]);
272 	buf[i] = '\0';
273 
274 	printf("%016jx %c %d %s\n", offset, node->type, node->count, buf);
275 }
276 
277 /*
278  * Top level recovery processor.  Assume the data is a B-Tree node.
279  * If the CRC is good we attempt to process the node, building the
280  * object space and creating the dictionary as we go.
281  */
282 static
283 void
284 recover_top(char *ptr, hammer_off_t offset)
285 {
286 	hammer_node_ondisk_t node;
287 	hammer_btree_elm_t elm;
288 	int maxcount;
289 	int i;
290 	int isnode;
291 
292 	for (node = (void *)ptr; (char *)node < ptr + HAMMER_BUFSIZE; ++node) {
293 		isnode = hammer_crc_test_btree(HammerVersion, node);
294 		maxcount = hammer_node_max_elements(node->type);
295 
296 		if (DebugOpt) {
297 			if (isnode)
298 				print_node(node, offset);
299 			else if (DebugOpt > 1)
300 				printf("%016jx -\n", offset);
301 		}
302 		offset += sizeof(*node);
303 
304 		if (isnode && node->type == HAMMER_BTREE_TYPE_LEAF) {
305 			for (i = 0; i < node->count && i < maxcount; ++i) {
306 				elm = &node->elms[i];
307 				if (elm->base.btype == HAMMER_BTREE_TYPE_RECORD)
308 					recover_elm(&elm->leaf);
309 			}
310 		}
311 	}
312 }
313 
314 static
315 void
316 recover_elm(hammer_btree_leaf_elm_t leaf)
317 {
318 	buffer_info_t data_buffer = NULL;
319 	struct recover_dict *dict;
320 	struct recover_dict *dict2;
321 	hammer_data_ondisk_t ondisk;
322 	hammer_off_t data_offset;
323 	struct stat st;
324 	int chunk;
325 	int len;
326 	int zfill;
327 	int64_t file_offset;
328 	uint16_t pfs_id;
329 	size_t nlen;
330 	int fd;
331 	char *name;
332 	char *path1;
333 	char *path2;
334 
335 	/*
336 	 * Ignore deleted records
337 	 */
338 	if (leaf->delete_ts)
339 		return;
340 
341 	/*
342 	 * If we're running full scan, it's possible that data_offset
343 	 * refers to old filesystem data that we can't physically access.
344 	 */
345 	data_offset = leaf->data_offset;
346 	if (get_volume(HAMMER_VOL_DECODE(data_offset)) == NULL)
347 		return;
348 
349 	if (data_offset != 0)
350 		ondisk = get_buffer_data(data_offset, &data_buffer, 0);
351 	else
352 		ondisk = NULL;
353 	if (ondisk == NULL)
354 		goto done;
355 
356 	len = leaf->data_len;
357 	chunk = HAMMER_BUFSIZE - ((int)data_offset & HAMMER_BUFMASK);
358 	if (chunk > len)
359 		chunk = len;
360 
361 	if (len < 0 || len > HAMMER_XBUFSIZE || len > chunk)
362 		goto done;
363 
364 	pfs_id = lo_to_pfs(leaf->base.localization);
365 
366 	/*
367 	 * Note that meaning of leaf->base.obj_id differs depending
368 	 * on record type.  For a direntry, leaf->base.obj_id points
369 	 * to its parent inode that this entry is a part of, but not
370 	 * its corresponding inode.
371 	 */
372 	dict = get_dict(leaf->base.obj_id, pfs_id);
373 
374 	switch(leaf->base.rec_type) {
375 	case HAMMER_RECTYPE_INODE:
376 		/*
377 		 * We found an inode which also tells us where the file
378 		 * or directory is in the directory hierarchy.
379 		 */
380 		if (VerboseOpt) {
381 			printf("inode %016jx:%05d found\n",
382 				(uintmax_t)leaf->base.obj_id, pfs_id);
383 		}
384 		path1 = recover_path(dict);
385 
386 		/*
387 		 * Attach the inode to its parent.  This isn't strictly
388 		 * necessary because the information is also in the
389 		 * directory entries, but if we do not find the directory
390 		 * entry this ensures that the files will still be
391 		 * reasonably well organized in their proper directories.
392 		 */
393 		if ((dict->flags & DICTF_PARENT) == 0 &&
394 		    dict->obj_id != HAMMER_OBJID_ROOT &&
395 		    ondisk->inode.parent_obj_id != 0) {
396 			dict->flags |= DICTF_PARENT;
397 			dict->parent = get_dict(ondisk->inode.parent_obj_id,
398 						pfs_id);
399 			if (dict->parent &&
400 			    (dict->parent->flags & DICTF_MADEDIR) == 0) {
401 				dict->parent->flags |= DICTF_MADEDIR;
402 				path2 = recover_path(dict->parent);
403 				printf("mkdir %s\n", path2);
404 				mkdir(path2, 0777);
405 				free(path2);
406 				path2 = NULL;
407 			}
408 		}
409 		if (dict->obj_type == 0)
410 			dict->obj_type = ondisk->inode.obj_type;
411 		dict->size = ondisk->inode.size;
412 		path2 = recover_path(dict);
413 
414 		if (lstat(path1, &st) == 0) {
415 			if (ondisk->inode.obj_type == HAMMER_OBJTYPE_REGFILE) {
416 				truncate(path1, dict->size);
417 				/* chmod(path1, 0666); */
418 			}
419 			if (strcmp(path1, path2)) {
420 				printf("Rename (inode) %s -> %s\n", path1, path2);
421 				rename(path1, path2);
422 			}
423 		} else if (ondisk->inode.obj_type == HAMMER_OBJTYPE_REGFILE) {
424 			printf("mkinode (file) %s\n", path2);
425 			fd = open(path2, O_RDWR|O_CREAT, 0666);
426 			if (fd > 0)
427 				close(fd);
428 		} else if (ondisk->inode.obj_type == HAMMER_OBJTYPE_DIRECTORY) {
429 			printf("mkinode (dir) %s\n", path2);
430 			mkdir(path2, 0777);
431 			dict->flags |= DICTF_MADEDIR;
432 		}
433 		free(path1);
434 		free(path2);
435 		break;
436 	case HAMMER_RECTYPE_DATA:
437 		/*
438 		 * File record data
439 		 */
440 		if (leaf->base.obj_id == 0)
441 			break;
442 		if (VerboseOpt) {
443 			printf("inode %016jx:%05d data %016jx,%d\n",
444 				(uintmax_t)leaf->base.obj_id,
445 				pfs_id,
446 				(uintmax_t)leaf->base.key - len,
447 				len);
448 		}
449 
450 		/*
451 		 * Update the dictionary entry
452 		 */
453 		if (dict->obj_type == 0)
454 			dict->obj_type = HAMMER_OBJTYPE_REGFILE;
455 
456 		/*
457 		 * If the parent directory has not been created we
458 		 * have to create it (typically a PFS%05d)
459 		 */
460 		if (dict->parent &&
461 		    (dict->parent->flags & DICTF_MADEDIR) == 0) {
462 			dict->parent->flags |= DICTF_MADEDIR;
463 			path2 = recover_path(dict->parent);
464 			printf("mkdir %s\n", path2);
465 			mkdir(path2, 0777);
466 			free(path2);
467 			path2 = NULL;
468 		}
469 
470 		/*
471 		 * Create the file if necessary, report file creations
472 		 */
473 		path1 = recover_path(dict);
474 		if (CachedPath && strcmp(CachedPath, path1) == 0)
475 			fd = CachedFd;
476 		else
477 			fd = open(path1, O_CREAT|O_RDWR, 0666);
478 		if (fd < 0) {
479 			printf("Unable to create %s: %s\n",
480 				path1, strerror(errno));
481 			free(path1);
482 			break;
483 		}
484 		if ((dict->flags & DICTF_MADEFILE) == 0) {
485 			dict->flags |= DICTF_MADEFILE;
486 			printf("mkfile %s\n", path1);
487 		}
488 
489 		/*
490 		 * And write the record.  A HAMMER data block is aligned
491 		 * and may contain trailing zeros after the file EOF.  The
492 		 * inode record is required to get the actual file size.
493 		 *
494 		 * However, when the inode record is not available
495 		 * we can do a sparse write and that will get it right
496 		 * most of the time even if the inode record is never
497 		 * found.
498 		 */
499 		file_offset = (int64_t)leaf->base.key - len;
500 		lseek(fd, (off_t)file_offset, SEEK_SET);
501 		while (len) {
502 			if (dict->size == -1) {
503 				for (zfill = chunk - 1; zfill >= 0; --zfill) {
504 					if (((char *)ondisk)[zfill])
505 						break;
506 				}
507 				++zfill;
508 			} else {
509 				zfill = chunk;
510 			}
511 
512 			if (zfill)
513 				write(fd, ondisk, zfill);
514 			if (zfill < chunk)
515 				lseek(fd, chunk - zfill, SEEK_CUR);
516 
517 			len -= chunk;
518 			data_offset += chunk;
519 			file_offset += chunk;
520 			ondisk = get_buffer_data(data_offset, &data_buffer, 0);
521 			if (ondisk == NULL)
522 				break;
523 			chunk = HAMMER_BUFSIZE -
524 				((int)data_offset & HAMMER_BUFMASK);
525 			if (chunk > len)
526 				chunk = len;
527 		}
528 		if (dict->size >= 0 && file_offset > dict->size) {
529 			ftruncate(fd, dict->size);
530 			/* fchmod(fd, 0666); */
531 		}
532 
533 		if (fd == CachedFd) {
534 			free(path1);
535 		} else if (CachedPath) {
536 			free(CachedPath);
537 			close(CachedFd);
538 			CachedPath = path1;
539 			CachedFd = fd;
540 		} else {
541 			CachedPath = path1;
542 			CachedFd = fd;
543 		}
544 		break;
545 	case HAMMER_RECTYPE_DIRENTRY:
546 		nlen = len - HAMMER_ENTRY_NAME_OFF;
547 		if ((int)nlen < 0)	/* illegal length */
548 			break;
549 		if (ondisk->entry.obj_id == 0 ||
550 		    ondisk->entry.obj_id == HAMMER_OBJID_ROOT) {
551 			break;
552 		}
553 		name = malloc(nlen + 1);
554 		bcopy(ondisk->entry.name, name, nlen);
555 		name[nlen] = 0;
556 		sanitize_string(name);
557 
558 		if (VerboseOpt) {
559 			printf("dir %016jx:%05d entry %016jx \"%s\"\n",
560 				(uintmax_t)leaf->base.obj_id,
561 				pfs_id,
562 				(uintmax_t)ondisk->entry.obj_id,
563 				name);
564 		}
565 
566 		/*
567 		 * We can't deal with hardlinks so if the object already
568 		 * has a name assigned to it we just keep using that name.
569 		 */
570 		dict2 = get_dict(ondisk->entry.obj_id, pfs_id);
571 		path1 = recover_path(dict2);
572 
573 		if (dict2->name == NULL)
574 			dict2->name = name;
575 		else
576 			free(name);
577 
578 		/*
579 		 * Attach dict2 to its directory (dict), create the
580 		 * directory (dict) if necessary.  We must ensure
581 		 * that the directory entry exists in order to be
582 		 * able to properly rename() the file without creating
583 		 * a namespace conflict.
584 		 */
585 		if ((dict2->flags & DICTF_PARENT) == 0) {
586 			dict2->flags |= DICTF_PARENT;
587 			dict2->parent = dict;
588 			if ((dict->flags & DICTF_MADEDIR) == 0) {
589 				dict->flags |= DICTF_MADEDIR;
590 				path2 = recover_path(dict);
591 				printf("mkdir %s\n", path2);
592 				mkdir(path2, 0777);
593 				free(path2);
594 				path2 = NULL;
595 			}
596 		}
597 		path2 = recover_path(dict2);
598 		if (strcmp(path1, path2) != 0 && lstat(path1, &st) == 0) {
599 			printf("Rename (entry) %s -> %s\n", path1, path2);
600 			rename(path1, path2);
601 		}
602 		free(path1);
603 		free(path2);
604 		break;
605 	default:
606 		/*
607 		 * Ignore any other record types
608 		 */
609 		break;
610 	}
611 done:
612 	rel_buffer(data_buffer);
613 }
614 
615 #define RD_HSIZE	32768
616 #define RD_HMASK	(RD_HSIZE - 1)
617 
618 struct recover_dict *RDHash[RD_HSIZE];
619 
620 static
621 struct recover_dict *
622 get_dict(int64_t obj_id, uint16_t pfs_id)
623 {
624 	struct recover_dict *dict;
625 	int i;
626 
627 	if (obj_id == 0)
628 		return(NULL);
629 
630 	i = crc32(&obj_id, sizeof(obj_id)) & RD_HMASK;
631 	for (dict = RDHash[i]; dict; dict = dict->next) {
632 		if (dict->obj_id == obj_id && dict->pfs_id == pfs_id)
633 			break;
634 	}
635 
636 	if (dict == NULL) {
637 		dict = malloc(sizeof(*dict));
638 		bzero(dict, sizeof(*dict));
639 		dict->obj_id = obj_id;
640 		dict->pfs_id = pfs_id;
641 		dict->next = RDHash[i];
642 		dict->size = -1;
643 		RDHash[i] = dict;
644 
645 		/*
646 		 * Always connect dangling dictionary entries to object 1
647 		 * (the root of the PFS).
648 		 *
649 		 * DICTF_PARENT will not be set until we know what the
650 		 * real parent directory object is.
651 		 */
652 		if (dict->obj_id != HAMMER_OBJID_ROOT)
653 			dict->parent = get_dict(HAMMER_OBJID_ROOT, pfs_id);
654 	}
655 	return(dict);
656 }
657 
658 struct path_info {
659 	enum { PI_FIGURE, PI_LOAD } state;
660 	uint16_t pfs_id;
661 	char *base;
662 	char *next;
663 	int len;
664 };
665 
666 static void recover_path_helper(struct recover_dict *, struct path_info *);
667 
668 static
669 char *
670 recover_path(struct recover_dict *dict)
671 {
672 	struct path_info info;
673 
674 	/* Find info.len first */
675 	bzero(&info, sizeof(info));
676 	info.state = PI_FIGURE;
677 	recover_path_helper(dict, &info);
678 
679 	/* Fill in the path */
680 	info.pfs_id = dict->pfs_id;
681 	info.base = malloc(info.len);
682 	info.next = info.base;
683 	info.state = PI_LOAD;
684 	recover_path_helper(dict, &info);
685 
686 	/* Return the path */
687 	return(info.base);
688 }
689 
690 #define STRLEN_OBJID	22	/* "obj_0x%016jx" */
691 #define STRLEN_PFSID	8	/* "PFS%05d" */
692 
693 static
694 void
695 recover_path_helper(struct recover_dict *dict, struct path_info *info)
696 {
697 	/*
698 	 * Calculate path element length
699 	 */
700 	dict->flags |= DICTF_TRAVERSED;
701 
702 	switch(info->state) {
703 	case PI_FIGURE:
704 		if (dict->obj_id == HAMMER_OBJID_ROOT)
705 			info->len += STRLEN_PFSID;
706 		else if (dict->name)
707 			info->len += strlen(dict->name);
708 		else
709 			info->len += STRLEN_OBJID;
710 		++info->len;
711 
712 		if (dict->parent &&
713 		    (dict->parent->flags & DICTF_TRAVERSED) == 0) {
714 			recover_path_helper(dict->parent, info);
715 		} else {
716 			info->len += strlen(TargetDir) + 1;
717 		}
718 		break;
719 	case PI_LOAD:
720 		if (dict->parent &&
721 		    (dict->parent->flags & DICTF_TRAVERSED) == 0) {
722 			recover_path_helper(dict->parent, info);
723 		} else {
724 			strcpy(info->next, TargetDir);
725 			info->next += strlen(info->next);
726 		}
727 
728 		*info->next++ = '/';
729 		if (dict->obj_id == HAMMER_OBJID_ROOT) {
730 			snprintf(info->next, STRLEN_PFSID + 1,
731 				"PFS%05d", info->pfs_id);
732 		} else if (dict->name) {
733 			strcpy(info->next, dict->name);
734 		} else {
735 			snprintf(info->next, STRLEN_OBJID + 1,
736 				"obj_0x%016jx", (uintmax_t)dict->obj_id);
737 		}
738 		info->next += strlen(info->next);
739 		break;
740 	}
741 	dict->flags &= ~DICTF_TRAVERSED;
742 }
743 
744 static
745 void
746 sanitize_string(char *str)
747 {
748 	while (*str) {
749 		if (!isprint(*str))
750 			*str = 'x';
751 		++str;
752 	}
753 }
754 
755 static
756 hammer_off_t
757 scan_raw_limit(void)
758 {
759 	volume_info_t volume;
760 	hammer_blockmap_t rootmap;
761 	hammer_blockmap_layer1_t layer1;
762 	hammer_blockmap_layer2_t layer2;
763 	buffer_info_t buffer1 = NULL;
764 	buffer_info_t buffer2 = NULL;
765 	hammer_off_t layer1_offset;
766 	hammer_off_t layer2_offset;
767 	hammer_off_t phys_offset;
768 	hammer_off_t block_offset;
769 	hammer_off_t offset = 0;
770 	int zone = HAMMER_ZONE_FREEMAP_INDEX;
771 
772 	volume = get_root_volume();
773 	rootmap = &volume->ondisk->vol0_blockmap[zone];
774 	assert(rootmap->phys_offset != 0);
775 
776 	for (phys_offset = HAMMER_ZONE_ENCODE(zone, 0);
777 	     phys_offset < HAMMER_ZONE_ENCODE(zone, HAMMER_OFF_LONG_MASK);
778 	     phys_offset += HAMMER_BLOCKMAP_LAYER2) {
779 		/*
780 		 * Dive layer 1.
781 		 */
782 		layer1_offset = rootmap->phys_offset +
783 				HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
784 		layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
785 
786 		if (!hammer_crc_test_layer1(HammerVersion, layer1)) {
787 			offset = 0; /* failed */
788 			goto end;
789 		}
790 		if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL)
791 			continue;
792 
793 		for (block_offset = 0;
794 		     block_offset < HAMMER_BLOCKMAP_LAYER2;
795 		     block_offset += HAMMER_BIGBLOCK_SIZE) {
796 			/*
797 			 * Dive layer 2, each entry represents a big-block.
798 			 */
799 			layer2_offset = layer1->phys_offset +
800 					HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
801 			layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
802 
803 			if (!hammer_crc_test_layer2(HammerVersion, layer2)) {
804 				offset = 0; /* failed */
805 				goto end;
806 			}
807 			if (layer2->zone == HAMMER_ZONE_UNAVAIL_INDEX) {
808 				break;
809 			} else if (layer2->zone && layer2->zone != zone) {
810 				offset = phys_offset + block_offset;
811 			}
812 		}
813 	}
814 end:
815 	rel_buffer(buffer1);
816 	rel_buffer(buffer2);
817 
818 	return(hammer_xlate_to_zone2(offset));
819 }
820 
821 static
822 void
823 scan_bigblocks(int target_zone)
824 {
825 	volume_info_t volume;
826 	hammer_blockmap_t rootmap;
827 	hammer_blockmap_layer1_t layer1;
828 	hammer_blockmap_layer2_t layer2;
829 	buffer_info_t buffer1 = NULL;
830 	buffer_info_t buffer2 = NULL;
831 	hammer_off_t layer1_offset;
832 	hammer_off_t layer2_offset;
833 	hammer_off_t phys_offset;
834 	hammer_off_t block_offset;
835 	hammer_off_t offset = 0;
836 	int zone = HAMMER_ZONE_FREEMAP_INDEX;
837 
838 	volume = get_root_volume();
839 	rootmap = &volume->ondisk->vol0_blockmap[zone];
840 	assert(rootmap->phys_offset != 0);
841 
842 	for (phys_offset = HAMMER_ZONE_ENCODE(zone, 0);
843 	     phys_offset < HAMMER_ZONE_ENCODE(zone, HAMMER_OFF_LONG_MASK);
844 	     phys_offset += HAMMER_BLOCKMAP_LAYER2) {
845 		/*
846 		 * Dive layer 1.
847 		 */
848 		layer1_offset = rootmap->phys_offset +
849 				HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
850 		layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
851 
852 		/*
853 		if (!hammer_crc_test_layer1(HammerVersion, layer1)) {
854 		}
855 		*/
856 		if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL)
857 			continue;
858 
859 		for (block_offset = 0;
860 		     block_offset < HAMMER_BLOCKMAP_LAYER2;
861 		     block_offset += HAMMER_BIGBLOCK_SIZE) {
862 			offset = phys_offset + block_offset;
863 			/*
864 			 * Dive layer 2, each entry represents a big-block.
865 			 */
866 			layer2_offset = layer1->phys_offset +
867 					HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
868 			layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
869 
870 			/*
871 			if (!hammer_crc_test_layer2(HammerVersion, layer2)) {
872 			}
873 			*/
874 			if (layer2->zone == target_zone) {
875 				add_bigblock_entry(offset, layer1, layer2);
876 			} else if (layer2->zone == HAMMER_ZONE_UNAVAIL_INDEX) {
877 				break;
878 			}
879 		}
880 	}
881 	rel_buffer(buffer1);
882 	rel_buffer(buffer2);
883 }
884 
885 static
886 void
887 free_bigblocks(void)
888 {
889 	bigblock_t b;
890 
891 	while ((b = RB_ROOT(&ZoneTree)) != NULL) {
892 		RB_REMOVE(bigblock_rb_tree, &ZoneTree, b);
893 		free(b);
894 	}
895 	assert(RB_EMPTY(&ZoneTree));
896 }
897 
898 static
899 void
900 add_bigblock_entry(hammer_off_t offset,
901 	hammer_blockmap_layer1_t layer1, hammer_blockmap_layer2_t layer2)
902 {
903 	bigblock_t b;
904 
905 	b = calloc(1, sizeof(*b));
906 	b->phys_offset = hammer_xlate_to_zone2(offset);
907 	assert((b->phys_offset & HAMMER_BIGBLOCK_MASK64) == 0);
908 	bcopy(layer1, &b->layer1, sizeof(*layer1));
909 	bcopy(layer2, &b->layer2, sizeof(*layer2));
910 
911 	RB_INSERT(bigblock_rb_tree, &ZoneTree, b);
912 }
913 
914 static
915 bigblock_t
916 get_bigblock_entry(hammer_off_t offset)
917 {
918 	bigblock_t b;
919 
920 	offset = hammer_xlate_to_zone2(offset);
921 	offset &= ~HAMMER_BIGBLOCK_MASK64;
922 
923 	b = RB_LOOKUP(bigblock_rb_tree, &ZoneTree, offset);
924 	if (b)
925 		return(b);
926 	return(NULL);
927 }
928