2 * Copyright (c) 2010 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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
38 struct recover_dict
*next
;
39 struct recover_dict
*parent
;
48 #define DICTF_MADEDIR 0x01
49 #define DICTF_MADEFILE 0x02
50 #define DICTF_PARENT 0x04 /* parent attached for real */
51 #define DICTF_TRAVERSED 0x80
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
);
59 static const char *TargetDir
;
60 static int CachedFd
= -1;
61 static char *CachedPath
;
64 hammer_cmd_recover(const char *target_dir
)
66 struct buffer_info
*data_buffer
;
67 struct volume_info
*scan
;
68 struct volume_info
*volume
;
74 TargetDir
= target_dir
;
76 printf("Running raw scan of HAMMER image, recovering to %s\n",
78 mkdir(TargetDir
, 0777);
81 TAILQ_FOREACH(scan
, &VolList
, entry
) {
82 volume
= get_volume(scan
->vol_no
);
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);
91 off
+= HAMMER_BUFSIZE
;
95 rel_buffer(data_buffer
);
108 * Top level recovery processor. Assume the data is a B-Tree node.
109 * If the CRC is good we attempt to process the node, building the
110 * object space and creating the dictionary as we go.
113 recover_top(char *ptr
)
115 struct hammer_node_ondisk
*node
;
116 hammer_btree_elm_t elm
;
120 for (node
= (void *)ptr
; (char *)node
< ptr
+ HAMMER_BUFSIZE
; ++node
) {
121 if (crc32(&node
->crc
+ 1, HAMMER_BTREE_CRCSIZE
) ==
123 node
->type
== HAMMER_BTREE_TYPE_LEAF
) {
127 maxcount
= HAMMER_BTREE_LEAF_ELMS
;
128 for (i
= 0; i
< node
->count
&& i
< maxcount
; ++i
) {
129 elm
= &node
->elms
[i
];
130 if (elm
->base
.btype
!= 'R')
132 recover_elm(&elm
->leaf
);
139 recover_elm(hammer_btree_leaf_elm_t leaf
)
141 struct buffer_info
*data_buffer
= NULL
;
142 struct recover_dict
*dict
;
143 struct recover_dict
*dict2
;
144 hammer_data_ondisk_t ondisk
;
145 hammer_off_t data_offset
;
159 * Ignore deleted records
163 if ((data_offset
= leaf
->data_offset
) != 0)
164 ondisk
= get_buffer_data(data_offset
, &data_buffer
, 0);
170 len
= leaf
->data_len
;
171 chunk
= HAMMER_BUFSIZE
- ((int)data_offset
& HAMMER_BUFMASK
);
175 if (len
< 0 || len
> HAMMER_XBUFSIZE
|| len
> chunk
)
178 llid
= leaf
->base
.localization
>> 16;
180 dict
= get_dict(leaf
->base
.obj_id
, llid
);
182 switch(leaf
->base
.rec_type
) {
183 case HAMMER_RECTYPE_INODE
:
185 * We found an inode which also tells us where the file
186 * or directory is in the directory hierarchy.
189 printf("file %016jx:%05d inode found\n",
190 (uintmax_t)leaf
->base
.obj_id
, llid
);
192 path1
= recover_path(dict
);
195 * Attach the inode to its parent. This isn't strictly
196 * necessary because the information is also in the
197 * directory entries, but if we do not find the directory
198 * entry this ensures that the files will still be
199 * reasonably well organized in their proper directories.
201 if ((dict
->flags
& DICTF_PARENT
) == 0 &&
202 dict
->obj_id
!= 1 && ondisk
->inode
.parent_obj_id
!= 0) {
203 dict
->flags
|= DICTF_PARENT
;
204 dict
->parent
= get_dict(ondisk
->inode
.parent_obj_id
,
207 (dict
->parent
->flags
& DICTF_MADEDIR
) == 0) {
208 dict
->parent
->flags
|= DICTF_MADEDIR
;
209 path2
= recover_path(dict
->parent
);
210 printf("mkdir %s\n", path2
);
216 if (dict
->obj_type
== 0)
217 dict
->obj_type
= ondisk
->inode
.obj_type
;
218 dict
->size
= ondisk
->inode
.size
;
219 path2
= recover_path(dict
);
221 if (lstat(path1
, &st
) == 0) {
222 if (ondisk
->inode
.obj_type
== HAMMER_OBJTYPE_REGFILE
) {
223 truncate(path1
, dict
->size
);
224 /* chmod(path1, 0666); */
226 if (strcmp(path1
, path2
)) {
227 printf("Rename %s -> %s\n", path1
, path2
);
228 rename(path1
, path2
);
230 } else if (ondisk
->inode
.obj_type
== HAMMER_OBJTYPE_REGFILE
) {
231 printf("mkinode (file) %s\n", path2
);
232 fd
= open(path2
, O_RDWR
|O_CREAT
, 0666);
235 } else if (ondisk
->inode
.obj_type
== HAMMER_OBJTYPE_DIRECTORY
) {
236 printf("mkinode (dir) %s\n", path2
);
238 dict
->flags
|= DICTF_MADEDIR
;
243 case HAMMER_RECTYPE_DATA
:
247 if (leaf
->base
.obj_id
== 0)
250 printf("file %016jx:%05d data %016jx,%d\n",
251 (uintmax_t)leaf
->base
.obj_id
,
253 (uintmax_t)leaf
->base
.key
- len
,
258 * Update the dictionary entry
260 if (dict
->obj_type
== 0)
261 dict
->obj_type
= HAMMER_OBJTYPE_REGFILE
;
264 * If the parent directory has not been created we
265 * have to create it (typically a PFS%05d)
268 (dict
->parent
->flags
& DICTF_MADEDIR
) == 0) {
269 dict
->parent
->flags
|= DICTF_MADEDIR
;
270 path2
= recover_path(dict
->parent
);
271 printf("mkdir %s\n", path2
);
278 * Create the file if necessary, report file creations
280 path1
= recover_path(dict
);
281 if (CachedPath
&& strcmp(CachedPath
, path1
) == 0) {
284 fd
= open(path1
, O_CREAT
|O_RDWR
, 0666);
287 printf("Unable to create %s: %s\n",
288 path1
, strerror(errno
));
292 if ((dict
->flags
& DICTF_MADEFILE
) == 0) {
293 dict
->flags
|= DICTF_MADEFILE
;
294 printf("mkfile %s\n", path1
);
298 * And write the record. A HAMMER data block is aligned
299 * and may contain trailing zeros after the file EOF. The
300 * inode record is required to get the actual file size.
302 * However, when the inode record is not available
303 * we can do a sparse write and that will get it right
304 * most of the time even if the inode record is never
307 file_offset
= (int64_t)leaf
->base
.key
- len
;
308 lseek(fd
, (off_t
)file_offset
, SEEK_SET
);
310 if (dict
->size
== -1) {
311 for (zfill
= chunk
- 1; zfill
>= 0; --zfill
) {
312 if (((char *)ondisk
)[zfill
])
321 write(fd
, ondisk
, zfill
);
323 lseek(fd
, chunk
- zfill
, SEEK_CUR
);
326 data_offset
+= chunk
;
327 file_offset
+= chunk
;
328 ondisk
= get_buffer_data(data_offset
, &data_buffer
, 0);
331 chunk
= HAMMER_BUFSIZE
-
332 ((int)data_offset
& HAMMER_BUFMASK
);
336 if (dict
->size
>= 0 && file_offset
> dict
->size
) {
337 ftruncate(fd
, dict
->size
);
338 /* fchmod(fd, 0666); */
341 if (fd
== CachedFd
) {
343 } else if (CachedPath
) {
353 case HAMMER_RECTYPE_DIRENTRY
:
354 nlen
= len
- offsetof(struct hammer_entry_data
, name
[0]);
355 if ((int)nlen
< 0) /* illegal length */
357 if (ondisk
->entry
.obj_id
== 0 || ondisk
->entry
.obj_id
== 1)
359 name
= malloc(nlen
+ 1);
360 bcopy(ondisk
->entry
.name
, name
, nlen
);
362 sanitize_string(name
);
365 * We can't deal with hardlinks so if the object already
366 * has a name assigned to it we just keep using that name.
368 dict2
= get_dict(ondisk
->entry
.obj_id
, llid
);
369 path1
= recover_path(dict2
);
371 if (dict2
->name
== NULL
)
377 * Attach dict2 to its directory (dict), create the
378 * directory (dict) if necessary. We must ensure
379 * that the directory entry exists in order to be
380 * able to properly rename() the file without creating
381 * a namespace conflict.
383 if ((dict2
->flags
& DICTF_PARENT
) == 0) {
384 dict2
->flags
|= DICTF_PARENT
;
385 dict2
->parent
= dict
;
386 if ((dict
->flags
& DICTF_MADEDIR
) == 0) {
387 dict
->flags
|= DICTF_MADEDIR
;
388 path2
= recover_path(dict
);
389 printf("mkdir %s\n", path2
);
395 path2
= recover_path(dict2
);
396 if (strcmp(path1
, path2
) != 0 && lstat(path1
, &st
) == 0) {
397 printf("Rename %s -> %s\n", path1
, path2
);
398 rename(path1
, path2
);
403 printf("dir %016jx:%05d entry %016jx \"%s\"\n",
404 (uintmax_t)leaf
->base
.obj_id
,
406 (uintmax_t)ondisk
->entry
.obj_id
,
411 * Ignore any other record types
416 rel_buffer(data_buffer
);
419 #define RD_HSIZE 32768
420 #define RD_HMASK (RD_HSIZE - 1)
422 struct recover_dict
*RDHash
[RD_HSIZE
];
425 struct recover_dict
*
426 get_dict(int64_t obj_id
, uint16_t llid
)
428 struct recover_dict
*dict
;
434 i
= crc32(&obj_id
, sizeof(obj_id
)) & RD_HMASK
;
435 for (dict
= RDHash
[i
]; dict
; dict
= dict
->next
) {
436 if (dict
->obj_id
== obj_id
&&
437 dict
->llid
== llid
) {
442 dict
= malloc(sizeof(*dict
));
443 bzero(dict
, sizeof(*dict
));
444 dict
->obj_id
= obj_id
;
446 dict
->next
= RDHash
[i
];
451 * Always connect dangling dictionary entries to object 1
452 * (the root of the PFS).
454 * DICTF_PARENT will not be set until we know what the
455 * real parent directory object is.
457 if (dict
->obj_id
!= 1)
458 dict
->parent
= get_dict(1, llid
);
464 enum { PI_FIGURE
, PI_LOAD
} state
;
471 static void recover_path_helper(struct recover_dict
*, struct path_info
*);
475 recover_path(struct recover_dict
*dict
)
477 struct path_info info
;
479 bzero(&info
, sizeof(info
));
480 info
.llid
= dict
->llid
;
481 info
.state
= PI_FIGURE
;
482 recover_path_helper(dict
, &info
);
483 info
.base
= malloc(info
.len
);
484 info
.next
= info
.base
;
485 info
.state
= PI_LOAD
;
486 recover_path_helper(dict
, &info
);
493 recover_path_helper(struct recover_dict
*dict
, struct path_info
*info
)
496 * Calculate path element length
498 dict
->flags
|= DICTF_TRAVERSED
;
500 switch(info
->state
) {
502 if (dict
->obj_id
== 1)
505 info
->len
+= strlen(dict
->name
);
511 (dict
->parent
->flags
& DICTF_TRAVERSED
) == 0) {
512 recover_path_helper(dict
->parent
, info
);
514 info
->len
+= strlen(TargetDir
) + 1;
519 (dict
->parent
->flags
& DICTF_TRAVERSED
) == 0) {
520 recover_path_helper(dict
->parent
, info
);
522 strcpy(info
->next
, TargetDir
);
523 info
->next
+= strlen(info
->next
);
527 if (dict
->obj_id
== 1) {
528 snprintf(info
->next
, 8+1, "PFS%05d", info
->llid
);
529 } else if (dict
->name
) {
530 strcpy(info
->next
, dict
->name
);
532 snprintf(info
->next
, 6+16+1, "obj_0x%016jx",
533 (uintmax_t)dict
->obj_id
);
535 info
->next
+= strlen(info
->next
);
538 dict
->flags
&= ~DICTF_TRAVERSED
;
543 sanitize_string(char *str
)