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 pfs_id
);
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
*volume
;
73 TargetDir
= target_dir
;
75 if (mkdir(TargetDir
, 0777) == -1) {
76 if (errno
!= EEXIST
) {
82 printf("Running raw scan of HAMMER image, recovering to %s\n",
86 TAILQ_FOREACH(volume
, &VolList
, entry
) {
88 printf("Scanning volume %d size %s\n",
89 volume
->vol_no
, sizetostr(volume
->size
));
90 off
= HAMMER_ENCODE_RAW_BUFFER(volume
->vol_no
, 0);
91 off_end
= off
+ HAMMER_VOL_BUF_SIZE(volume
->ondisk
);
92 while (off
< off_end
) {
93 ptr
= get_buffer_data(off
, &data_buffer
, 0);
96 off
+= HAMMER_BUFSIZE
;
100 rel_buffer(data_buffer
);
113 * Top level recovery processor. Assume the data is a B-Tree node.
114 * If the CRC is good we attempt to process the node, building the
115 * object space and creating the dictionary as we go.
118 recover_top(char *ptr
)
120 struct hammer_node_ondisk
*node
;
121 hammer_btree_elm_t elm
;
125 for (node
= (void *)ptr
; (char *)node
< ptr
+ HAMMER_BUFSIZE
; ++node
) {
126 if (crc32(&node
->crc
+ 1, HAMMER_BTREE_CRCSIZE
) ==
128 node
->type
== HAMMER_BTREE_TYPE_LEAF
) {
132 maxcount
= HAMMER_BTREE_LEAF_ELMS
;
133 for (i
= 0; i
< node
->count
&& i
< maxcount
; ++i
) {
134 elm
= &node
->elms
[i
];
135 if (elm
->base
.btype
!= 'R')
137 recover_elm(&elm
->leaf
);
144 recover_elm(hammer_btree_leaf_elm_t leaf
)
146 struct buffer_info
*data_buffer
= NULL
;
147 struct recover_dict
*dict
;
148 struct recover_dict
*dict2
;
149 hammer_data_ondisk_t ondisk
;
150 hammer_off_t data_offset
;
164 * Ignore deleted records
168 if ((data_offset
= leaf
->data_offset
) != 0)
169 ondisk
= get_buffer_data(data_offset
, &data_buffer
, 0);
175 len
= leaf
->data_len
;
176 chunk
= HAMMER_BUFSIZE
- ((int)data_offset
& HAMMER_BUFMASK
);
180 if (len
< 0 || len
> HAMMER_XBUFSIZE
|| len
> chunk
)
183 pfs_id
= lo_to_pfs(leaf
->base
.localization
);
185 dict
= get_dict(leaf
->base
.obj_id
, pfs_id
);
187 switch(leaf
->base
.rec_type
) {
188 case HAMMER_RECTYPE_INODE
:
190 * We found an inode which also tells us where the file
191 * or directory is in the directory hierarchy.
194 printf("file %016jx:%05d inode found\n",
195 (uintmax_t)leaf
->base
.obj_id
, pfs_id
);
197 path1
= recover_path(dict
);
200 * Attach the inode to its parent. This isn't strictly
201 * necessary because the information is also in the
202 * directory entries, but if we do not find the directory
203 * entry this ensures that the files will still be
204 * reasonably well organized in their proper directories.
206 if ((dict
->flags
& DICTF_PARENT
) == 0 &&
207 dict
->obj_id
!= HAMMER_OBJID_ROOT
&&
208 ondisk
->inode
.parent_obj_id
!= 0) {
209 dict
->flags
|= DICTF_PARENT
;
210 dict
->parent
= get_dict(ondisk
->inode
.parent_obj_id
,
213 (dict
->parent
->flags
& DICTF_MADEDIR
) == 0) {
214 dict
->parent
->flags
|= DICTF_MADEDIR
;
215 path2
= recover_path(dict
->parent
);
216 printf("mkdir %s\n", path2
);
222 if (dict
->obj_type
== 0)
223 dict
->obj_type
= ondisk
->inode
.obj_type
;
224 dict
->size
= ondisk
->inode
.size
;
225 path2
= recover_path(dict
);
227 if (lstat(path1
, &st
) == 0) {
228 if (ondisk
->inode
.obj_type
== HAMMER_OBJTYPE_REGFILE
) {
229 truncate(path1
, dict
->size
);
230 /* chmod(path1, 0666); */
232 if (strcmp(path1
, path2
)) {
233 printf("Rename %s -> %s\n", path1
, path2
);
234 rename(path1
, path2
);
236 } else if (ondisk
->inode
.obj_type
== HAMMER_OBJTYPE_REGFILE
) {
237 printf("mkinode (file) %s\n", path2
);
238 fd
= open(path2
, O_RDWR
|O_CREAT
, 0666);
241 } else if (ondisk
->inode
.obj_type
== HAMMER_OBJTYPE_DIRECTORY
) {
242 printf("mkinode (dir) %s\n", path2
);
244 dict
->flags
|= DICTF_MADEDIR
;
249 case HAMMER_RECTYPE_DATA
:
253 if (leaf
->base
.obj_id
== 0)
256 printf("file %016jx:%05d data %016jx,%d\n",
257 (uintmax_t)leaf
->base
.obj_id
,
259 (uintmax_t)leaf
->base
.key
- len
,
264 * Update the dictionary entry
266 if (dict
->obj_type
== 0)
267 dict
->obj_type
= HAMMER_OBJTYPE_REGFILE
;
270 * If the parent directory has not been created we
271 * have to create it (typically a PFS%05d)
274 (dict
->parent
->flags
& DICTF_MADEDIR
) == 0) {
275 dict
->parent
->flags
|= DICTF_MADEDIR
;
276 path2
= recover_path(dict
->parent
);
277 printf("mkdir %s\n", path2
);
284 * Create the file if necessary, report file creations
286 path1
= recover_path(dict
);
287 if (CachedPath
&& strcmp(CachedPath
, path1
) == 0) {
290 fd
= open(path1
, O_CREAT
|O_RDWR
, 0666);
293 printf("Unable to create %s: %s\n",
294 path1
, strerror(errno
));
298 if ((dict
->flags
& DICTF_MADEFILE
) == 0) {
299 dict
->flags
|= DICTF_MADEFILE
;
300 printf("mkfile %s\n", path1
);
304 * And write the record. A HAMMER data block is aligned
305 * and may contain trailing zeros after the file EOF. The
306 * inode record is required to get the actual file size.
308 * However, when the inode record is not available
309 * we can do a sparse write and that will get it right
310 * most of the time even if the inode record is never
313 file_offset
= (int64_t)leaf
->base
.key
- len
;
314 lseek(fd
, (off_t
)file_offset
, SEEK_SET
);
316 if (dict
->size
== -1) {
317 for (zfill
= chunk
- 1; zfill
>= 0; --zfill
) {
318 if (((char *)ondisk
)[zfill
])
327 write(fd
, ondisk
, zfill
);
329 lseek(fd
, chunk
- zfill
, SEEK_CUR
);
332 data_offset
+= chunk
;
333 file_offset
+= chunk
;
334 ondisk
= get_buffer_data(data_offset
, &data_buffer
, 0);
337 chunk
= HAMMER_BUFSIZE
-
338 ((int)data_offset
& HAMMER_BUFMASK
);
342 if (dict
->size
>= 0 && file_offset
> dict
->size
) {
343 ftruncate(fd
, dict
->size
);
344 /* fchmod(fd, 0666); */
347 if (fd
== CachedFd
) {
349 } else if (CachedPath
) {
359 case HAMMER_RECTYPE_DIRENTRY
:
360 nlen
= len
- offsetof(struct hammer_direntry_data
, name
[0]);
361 if ((int)nlen
< 0) /* illegal length */
363 if (ondisk
->entry
.obj_id
== 0 ||
364 ondisk
->entry
.obj_id
== HAMMER_OBJID_ROOT
)
366 name
= malloc(nlen
+ 1);
367 bcopy(ondisk
->entry
.name
, name
, nlen
);
369 sanitize_string(name
);
372 * We can't deal with hardlinks so if the object already
373 * has a name assigned to it we just keep using that name.
375 dict2
= get_dict(ondisk
->entry
.obj_id
, pfs_id
);
376 path1
= recover_path(dict2
);
378 if (dict2
->name
== NULL
)
384 * Attach dict2 to its directory (dict), create the
385 * directory (dict) if necessary. We must ensure
386 * that the directory entry exists in order to be
387 * able to properly rename() the file without creating
388 * a namespace conflict.
390 if ((dict2
->flags
& DICTF_PARENT
) == 0) {
391 dict2
->flags
|= DICTF_PARENT
;
392 dict2
->parent
= dict
;
393 if ((dict
->flags
& DICTF_MADEDIR
) == 0) {
394 dict
->flags
|= DICTF_MADEDIR
;
395 path2
= recover_path(dict
);
396 printf("mkdir %s\n", path2
);
402 path2
= recover_path(dict2
);
403 if (strcmp(path1
, path2
) != 0 && lstat(path1
, &st
) == 0) {
404 printf("Rename %s -> %s\n", path1
, path2
);
405 rename(path1
, path2
);
410 printf("dir %016jx:%05d entry %016jx \"%s\"\n",
411 (uintmax_t)leaf
->base
.obj_id
,
413 (uintmax_t)ondisk
->entry
.obj_id
,
418 * Ignore any other record types
423 rel_buffer(data_buffer
);
426 #define RD_HSIZE 32768
427 #define RD_HMASK (RD_HSIZE - 1)
429 struct recover_dict
*RDHash
[RD_HSIZE
];
432 struct recover_dict
*
433 get_dict(int64_t obj_id
, uint16_t pfs_id
)
435 struct recover_dict
*dict
;
441 i
= crc32(&obj_id
, sizeof(obj_id
)) & RD_HMASK
;
442 for (dict
= RDHash
[i
]; dict
; dict
= dict
->next
) {
443 if (dict
->obj_id
== obj_id
&&
444 dict
->pfs_id
== pfs_id
) {
449 dict
= malloc(sizeof(*dict
));
450 bzero(dict
, sizeof(*dict
));
451 dict
->obj_id
= obj_id
;
452 dict
->pfs_id
= pfs_id
;
453 dict
->next
= RDHash
[i
];
458 * Always connect dangling dictionary entries to object 1
459 * (the root of the PFS).
461 * DICTF_PARENT will not be set until we know what the
462 * real parent directory object is.
464 if (dict
->obj_id
!= HAMMER_OBJID_ROOT
)
465 dict
->parent
= get_dict(1, pfs_id
);
471 enum { PI_FIGURE
, PI_LOAD
} state
;
478 static void recover_path_helper(struct recover_dict
*, struct path_info
*);
482 recover_path(struct recover_dict
*dict
)
484 struct path_info info
;
486 bzero(&info
, sizeof(info
));
487 info
.pfs_id
= dict
->pfs_id
;
488 info
.state
= PI_FIGURE
;
489 recover_path_helper(dict
, &info
);
490 info
.base
= malloc(info
.len
);
491 info
.next
= info
.base
;
492 info
.state
= PI_LOAD
;
493 recover_path_helper(dict
, &info
);
500 recover_path_helper(struct recover_dict
*dict
, struct path_info
*info
)
503 * Calculate path element length
505 dict
->flags
|= DICTF_TRAVERSED
;
507 switch(info
->state
) {
509 if (dict
->obj_id
== HAMMER_OBJID_ROOT
)
512 info
->len
+= strlen(dict
->name
);
518 (dict
->parent
->flags
& DICTF_TRAVERSED
) == 0) {
519 recover_path_helper(dict
->parent
, info
);
521 info
->len
+= strlen(TargetDir
) + 1;
526 (dict
->parent
->flags
& DICTF_TRAVERSED
) == 0) {
527 recover_path_helper(dict
->parent
, info
);
529 strcpy(info
->next
, TargetDir
);
530 info
->next
+= strlen(info
->next
);
534 if (dict
->obj_id
== HAMMER_OBJID_ROOT
) {
535 snprintf(info
->next
, 8+1, "PFS%05d", info
->pfs_id
);
536 } else if (dict
->name
) {
537 strcpy(info
->next
, dict
->name
);
539 snprintf(info
->next
, 6+16+1, "obj_0x%016jx",
540 (uintmax_t)dict
->obj_id
);
542 info
->next
+= strlen(info
->next
);
545 dict
->flags
&= ~DICTF_TRAVERSED
;
550 sanitize_string(char *str
)