xref: /dragonfly/sbin/hammer/cmd_recover.c (revision 25a2db75) 
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 struct recover_dict {
38 	struct recover_dict *next;
39 	struct recover_dict *parent;
40 	int64_t	obj_id;
41 	uint8_t obj_type;
42 	uint8_t flags;
43 	uint16_t llid;
44 	int64_t	size;
45 	char	*name;
46 };
47 
48 #define DICTF_MADEDIR	0x01
49 #define DICTF_MADEFILE	0x02
50 #define DICTF_PARENT	0x04	/* parent attached for real */
51 #define DICTF_TRAVERSED	0x80
52 
53 static void recover_top(char *ptr);
54 static void recover_elm(hammer_btree_leaf_elm_t leaf);
55 static struct recover_dict *get_dict(int64_t obj_id, uint16_t llid);
56 static char *recover_path(struct recover_dict *dict);
57 static void sanitize_string(char *str);
58 
59 static const char *TargetDir;
60 static int CachedFd = -1;
61 static char *CachedPath;
62 
63 void
64 hammer_cmd_recover(const char *target_dir)
65 {
66 	struct buffer_info *data_buffer;
67 	struct volume_info *scan;
68 	struct volume_info *volume;
69 	hammer_off_t off;
70 	hammer_off_t off_end;
71 	char *ptr;
72 
73 	AssertOnFailure = 0;
74 	TargetDir = target_dir;
75 
76 	printf("Running raw scan of HAMMER image, recovering to %s\n",
77 		TargetDir);
78 	mkdir(TargetDir, 0777);
79 
80 	data_buffer = NULL;
81 	TAILQ_FOREACH(scan, &VolList, entry) {
82 		volume = get_volume(scan->vol_no);
83 
84 		off = HAMMER_ZONE_RAW_BUFFER + 0;
85 		off |= HAMMER_VOL_ENCODE(volume->vol_no);
86 		off_end = off + (volume->ondisk->vol_buf_end - volume->ondisk->vol_buf_beg);
87 		while (off < off_end) {
88 			ptr = get_buffer_data(off, &data_buffer, 0);
89 			if (ptr) {
90 				recover_top(ptr);
91 				off += HAMMER_BUFSIZE;
92 			}
93 		}
94 	}
95 	if (data_buffer)
96 		rel_buffer(data_buffer);
97 
98 	if (CachedPath) {
99 		free(CachedPath);
100 		close(CachedFd);
101 		CachedPath = NULL;
102 		CachedFd = -1;
103 	}
104 
105 	AssertOnFailure = 1;
106 }
107 
108 /*
109  * Top level recovery processor.  Assume the data is a B-Tree node.
110  * If the CRC is good we attempt to process the node, building the
111  * object space and creating the dictionary as we go.
112  */
113 static void
114 recover_top(char *ptr)
115 {
116 	struct hammer_node_ondisk *node;
117 	hammer_btree_elm_t elm;
118 	int maxcount;
119 	int i;
120 
121 	for (node = (void *)ptr; (char *)node < ptr + HAMMER_BUFSIZE; ++node) {
122 		if (crc32(&node->crc + 1, HAMMER_BTREE_CRCSIZE) ==
123 		    node->crc &&
124 		    node->type == HAMMER_BTREE_TYPE_LEAF) {
125 			/*
126 			 * Scan elements
127 			 */
128 			maxcount = HAMMER_BTREE_LEAF_ELMS;
129 			for (i = 0; i < node->count && i < maxcount; ++i) {
130 				elm = &node->elms[i];
131 				if (elm->base.btype != 'R')
132 					continue;
133 				recover_elm(&elm->leaf);
134 			}
135 		}
136 	}
137 }
138 
139 static void
140 recover_elm(hammer_btree_leaf_elm_t leaf)
141 {
142 	struct buffer_info *data_buffer = NULL;
143 	struct recover_dict *dict;
144 	struct recover_dict *dict2;
145 	hammer_data_ondisk_t ondisk;
146 	hammer_off_t data_offset;
147 	struct stat st;
148 	int chunk;
149 	int len;
150 	int zfill;
151 	int64_t file_offset;
152 	uint16_t llid;
153 	size_t nlen;
154 	int fd;
155 	char *name;
156 	char *path1;
157 	char *path2;
158 
159 	/*
160 	 * Ignore deleted records
161 	 */
162 	if (leaf->delete_ts)
163 		return;
164 	if ((data_offset = leaf->data_offset) != 0)
165 		ondisk = get_buffer_data(data_offset, &data_buffer, 0);
166 	else
167 		ondisk = NULL;
168 	if (ondisk == NULL)
169 		goto done;
170 
171 	len = leaf->data_len;
172 	chunk = HAMMER_BUFSIZE - ((int)data_offset & HAMMER_BUFMASK);
173 	if (chunk > len)
174 		chunk = len;
175 
176 	if (len < 0 || len > HAMMER_XBUFSIZE || len > chunk)
177 		goto done;
178 
179 	llid = leaf->base.localization >> 16;
180 
181 	dict = get_dict(leaf->base.obj_id, llid);
182 
183 	switch(leaf->base.rec_type) {
184 	case HAMMER_RECTYPE_INODE:
185 		/*
186 		 * We found an inode which also tells us where the file
187 		 * or directory is in the directory hierarchy.
188 		 */
189 		if (VerboseOpt) {
190 			printf("file %016jx:%05d inode found\n",
191 				(uintmax_t)leaf->base.obj_id, llid);
192 		}
193 		path1 = recover_path(dict);
194 
195 		/*
196 		 * Attach the inode to its parent.  This isn't strictly
197 		 * necessary because the information is also in the
198 		 * directory entries, but if we do not find the directory
199 		 * entry this ensures that the files will still be
200 		 * reasonably well organized in their proper directories.
201 		 */
202 		if ((dict->flags & DICTF_PARENT) == 0 &&
203 		    dict->obj_id != 1 && ondisk->inode.parent_obj_id != 0) {
204 			dict->flags |= DICTF_PARENT;
205 			dict->parent = get_dict(ondisk->inode.parent_obj_id,
206 						llid);
207 			if (dict->parent &&
208 			    (dict->parent->flags & DICTF_MADEDIR) == 0) {
209 				dict->parent->flags |= DICTF_MADEDIR;
210 				path2 = recover_path(dict->parent);
211 				printf("mkdir %s\n", path2);
212 				mkdir(path2, 0777);
213 				free(path2);
214 				path2 = NULL;
215 			}
216 		}
217 		if (dict->obj_type == 0)
218 			dict->obj_type = ondisk->inode.obj_type;
219 		dict->size = ondisk->inode.size;
220 		path2 = recover_path(dict);
221 
222 		if (lstat(path1, &st) == 0) {
223 			if (ondisk->inode.obj_type == HAMMER_OBJTYPE_REGFILE) {
224 				truncate(path1, dict->size);
225 				/* chmod(path1, 0666); */
226 			}
227 			if (strcmp(path1, path2)) {
228 				printf("Rename %s -> %s\n", path1, path2);
229 				rename(path1, path2);
230 			}
231 		} else if (ondisk->inode.obj_type == HAMMER_OBJTYPE_REGFILE) {
232 			printf("mkinode (file) %s\n", path2);
233 			fd = open(path2, O_RDWR|O_CREAT, 0666);
234 			if (fd > 0)
235 				close(fd);
236 		} else if (ondisk->inode.obj_type == HAMMER_OBJTYPE_DIRECTORY) {
237 			printf("mkinode (dir) %s\n", path2);
238 			mkdir(path2, 0777);
239 			dict->flags |= DICTF_MADEDIR;
240 		}
241 		free(path1);
242 		free(path2);
243 		break;
244 	case HAMMER_RECTYPE_DATA:
245 		/*
246 		 * File record data
247 		 */
248 		if (leaf->base.obj_id == 0)
249 			break;
250 		if (VerboseOpt) {
251 			printf("file %016jx:%05d data %016jx,%d\n",
252 				(uintmax_t)leaf->base.obj_id,
253 				llid,
254 				(uintmax_t)leaf->base.key - len,
255 				len);
256 		}
257 
258 		/*
259 		 * Update the dictionary entry
260 		 */
261 		if (dict->obj_type == 0)
262 			dict->obj_type = HAMMER_OBJTYPE_REGFILE;
263 
264 		/*
265 		 * If the parent directory has not been created we
266 		 * have to create it (typically a PFS%05d)
267 		 */
268 		if (dict->parent &&
269 		    (dict->parent->flags & DICTF_MADEDIR) == 0) {
270 			dict->parent->flags |= DICTF_MADEDIR;
271 			path2 = recover_path(dict->parent);
272 			printf("mkdir %s\n", path2);
273 			mkdir(path2, 0777);
274 			free(path2);
275 			path2 = NULL;
276 		}
277 
278 		/*
279 		 * Create the file if necessary, report file creations
280 		 */
281 		path1 = recover_path(dict);
282 		if (CachedPath && strcmp(CachedPath, path1) == 0) {
283 			fd = CachedFd;
284 		} else {
285 			fd = open(path1, O_CREAT|O_RDWR, 0666);
286 		}
287 		if (fd < 0) {
288 			printf("Unable to create %s: %s\n",
289 				path1, strerror(errno));
290 			free(path1);
291 			break;
292 		}
293 		if ((dict->flags & DICTF_MADEFILE) == 0) {
294 			dict->flags |= DICTF_MADEFILE;
295 			printf("mkfile %s\n", path1);
296 		}
297 
298 		/*
299 		 * And write the record.  A HAMMER data block is aligned
300 		 * and may contain trailing zeros after the file EOF.  The
301 		 * inode record is required to get the actual file size.
302 		 *
303 		 * However, when the inode record is not available
304 		 * we can do a sparse write and that will get it right
305 		 * most of the time even if the inode record is never
306 		 * found.
307 		 */
308 		file_offset = (int64_t)leaf->base.key - len;
309 		lseek(fd, (off_t)file_offset, SEEK_SET);
310 		while (len) {
311 			if (dict->size == -1) {
312 				for (zfill = chunk - 1; zfill >= 0; --zfill) {
313 					if (((char *)ondisk)[zfill])
314 						break;
315 				}
316 				++zfill;
317 			} else {
318 				zfill = chunk;
319 			}
320 
321 			if (zfill)
322 				write(fd, ondisk, zfill);
323 			if (zfill < chunk)
324 				lseek(fd, chunk - zfill, SEEK_CUR);
325 
326 			len -= chunk;
327 			data_offset += chunk;
328 			file_offset += chunk;
329 			ondisk = get_buffer_data(data_offset, &data_buffer, 0);
330 			if (ondisk == NULL)
331 				break;
332 			chunk = HAMMER_BUFSIZE -
333 				((int)data_offset & HAMMER_BUFMASK);
334 			if (chunk > len)
335 				chunk = len;
336 		}
337 		if (dict->size >= 0 && file_offset > dict->size) {
338 			ftruncate(fd, dict->size);
339 			/* fchmod(fd, 0666); */
340 		}
341 
342 		if (fd == CachedFd) {
343 			free(path1);
344 		} else if (CachedPath) {
345 			free(CachedPath);
346 			close(CachedFd);
347 			CachedPath = path1;
348 			CachedFd = fd;
349 		} else {
350 			CachedPath = path1;
351 			CachedFd = fd;
352 		}
353 		break;
354 	case HAMMER_RECTYPE_DIRENTRY:
355 		nlen = len - offsetof(struct hammer_entry_data, name[0]);
356 		if ((int)nlen < 0)	/* illegal length */
357 			break;
358 		if (ondisk->entry.obj_id == 0 || ondisk->entry.obj_id == 1)
359 			break;
360 		name = malloc(nlen + 1);
361 		bcopy(ondisk->entry.name, name, nlen);
362 		name[nlen] = 0;
363 		sanitize_string(name);
364 
365 		/*
366 		 * We can't deal with hardlinks so if the object already
367 		 * has a name assigned to it we just keep using that name.
368 		 */
369 		dict2 = get_dict(ondisk->entry.obj_id, llid);
370 		path1 = recover_path(dict2);
371 
372 		if (dict2->name == NULL)
373 			dict2->name = name;
374 		else
375 			free(name);
376 
377 		/*
378 		 * Attach dict2 to its directory (dict), create the
379 		 * directory (dict) if necessary.  We must ensure
380 		 * that the directory entry exists in order to be
381 		 * able to properly rename() the file without creating
382 		 * a namespace conflict.
383 		 */
384 		if ((dict2->flags & DICTF_PARENT) == 0) {
385 			dict2->flags |= DICTF_PARENT;
386 			dict2->parent = dict;
387 			if ((dict->flags & DICTF_MADEDIR) == 0) {
388 				dict->flags |= DICTF_MADEDIR;
389 				path2 = recover_path(dict);
390 				printf("mkdir %s\n", path2);
391 				mkdir(path2, 0777);
392 				free(path2);
393 				path2 = NULL;
394 			}
395 		}
396 		path2 = recover_path(dict2);
397 		if (strcmp(path1, path2) != 0 && lstat(path1, &st) == 0) {
398 			printf("Rename %s -> %s\n", path1, path2);
399 			rename(path1, path2);
400 		}
401 		free(path1);
402 		free(path2);
403 
404 		printf("dir  %016jx:%05d entry %016jx \"%s\"\n",
405 			(uintmax_t)leaf->base.obj_id,
406 			llid,
407 			(uintmax_t)ondisk->entry.obj_id,
408 			name);
409 		break;
410 	default:
411 		/*
412 		 * Ignore any other record types
413 		 */
414 		break;
415 	}
416 done:
417 	if (data_buffer)
418 		rel_buffer(data_buffer);
419 }
420 
421 #define RD_HSIZE	32768
422 #define RD_HMASK	(RD_HSIZE - 1)
423 
424 struct recover_dict *RDHash[RD_HSIZE];
425 
426 static
427 struct recover_dict *
428 get_dict(int64_t obj_id, uint16_t llid)
429 {
430 	struct recover_dict *dict;
431 	int i;
432 
433 	if (obj_id == 0)
434 		return(NULL);
435 
436 	i = crc32(&obj_id, sizeof(obj_id)) & RD_HMASK;
437 	for (dict = RDHash[i]; dict; dict = dict->next) {
438 		if (dict->obj_id == obj_id &&
439 		    dict->llid == llid) {
440 			break;
441 		}
442 	}
443 	if (dict == NULL) {
444 		dict = malloc(sizeof(*dict));
445 		bzero(dict, sizeof(*dict));
446 		dict->obj_id = obj_id;
447 		dict->llid = llid;
448 		dict->next = RDHash[i];
449 		dict->size = -1;
450 		RDHash[i] = dict;
451 
452 		/*
453 		 * Always connect dangling dictionary entries to object 1
454 		 * (the root of the PFS).
455 		 *
456 		 * DICTF_PARENT will not be set until we know what the
457 		 * real parent directory object is.
458 		 */
459 		if (dict->obj_id != 1)
460 			dict->parent = get_dict(1, llid);
461 	}
462 	return(dict);
463 }
464 
465 struct path_info {
466 	enum { PI_FIGURE, PI_LOAD } state;
467 	uint16_t llid;
468 	char *base;
469 	char *next;
470 	int len;
471 };
472 
473 static void recover_path_helper(struct recover_dict *, struct path_info *);
474 
475 static
476 char *
477 recover_path(struct recover_dict *dict)
478 {
479 	struct path_info info;
480 
481 	bzero(&info, sizeof(info));
482 	info.llid = dict->llid;
483 	info.state = PI_FIGURE;
484 	recover_path_helper(dict, &info);
485 	info.base = malloc(info.len);
486 	info.next = info.base;
487 	info.state = PI_LOAD;
488 	recover_path_helper(dict, &info);
489 
490 	return(info.base);
491 }
492 
493 static
494 void
495 recover_path_helper(struct recover_dict *dict, struct path_info *info)
496 {
497 	/*
498 	 * Calculate path element length
499 	 */
500 	dict->flags |= DICTF_TRAVERSED;
501 
502 	switch(info->state) {
503 	case PI_FIGURE:
504 		if (dict->obj_id == 1)
505 			info->len += 8;
506 		else if (dict->name)
507 			info->len += strlen(dict->name);
508 		else
509 			info->len += 6 + 16;
510 		++info->len;
511 
512 		if (dict->parent &&
513 		    (dict->parent->flags & DICTF_TRAVERSED) == 0) {
514 			recover_path_helper(dict->parent, info);
515 		} else {
516 			info->len += strlen(TargetDir) + 1;
517 		}
518 		break;
519 	case PI_LOAD:
520 		if (dict->parent &&
521 		    (dict->parent->flags & DICTF_TRAVERSED) == 0) {
522 			recover_path_helper(dict->parent, info);
523 		} else {
524 			strcpy(info->next, TargetDir);
525 			info->next += strlen(info->next);
526 		}
527 
528 		*info->next++ = '/';
529 		if (dict->obj_id == 1) {
530 			snprintf(info->next, 8+1, "PFS%05d", info->llid);
531 		} else if (dict->name) {
532 			strcpy(info->next, dict->name);
533 		} else {
534 			snprintf(info->next, 6+16+1, "obj_0x%016jx",
535 				(uintmax_t)dict->obj_id);
536 		}
537 		info->next += strlen(info->next);
538 		break;
539 	}
540 	dict->flags &= ~DICTF_TRAVERSED;
541 }
542 
543 static
544 void
545 sanitize_string(char *str)
546 {
547 	while (*str) {
548 		if (!isprint(*str))
549 			*str = 'x';
550 		++str;
551 	}
552 }
553