1 /* hivex - Windows Registry "hive" extraction library.
2 * Copyright (C) 2009-2011 Red Hat Inc.
3 * Derived from code by Petter Nordahl-Hagen under a compatible license:
4 * Copyright (c) 1997-2007 Petter Nordahl-Hagen.
5 * Derived from code by Markus Stephany under a compatible license:
6 * Copyright (c) 2000-2004, Markus Stephany.
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation;
11 * version 2.1 of the License.
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * See file LICENSE for the full license.
38 #include "full-read.h"
39 #include "full-write.h"
41 #define STREQ(a,b) (strcmp((a),(b)) == 0)
42 #define STRCASEEQ(a,b) (strcasecmp((a),(b)) == 0)
43 //#define STRNEQ(a,b) (strcmp((a),(b)) != 0)
44 //#define STRCASENEQ(a,b) (strcasecmp((a),(b)) != 0)
45 #define STREQLEN(a,b,n) (strncmp((a),(b),(n)) == 0)
46 //#define STRCASEEQLEN(a,b,n) (strncasecmp((a),(b),(n)) == 0)
47 //#define STRNEQLEN(a,b,n) (strncmp((a),(b),(n)) != 0)
48 //#define STRCASENEQLEN(a,b,n) (strncasecmp((a),(b),(n)) != 0)
49 #define STRPREFIX(a,b) (strncmp((a),(b),strlen((b))) == 0)
52 #include "byte_conversions.h"
54 /* These limits are in place to stop really stupid stuff and/or exploits. */
55 #define HIVEX_MAX_SUBKEYS 15000
56 #define HIVEX_MAX_VALUES 10000
57 #define HIVEX_MAX_VALUE_LEN 1000000
58 #define HIVEX_MAX_ALLOCATION 1000000
60 static char *windows_utf16_to_utf8 (/* const */ char *input, size_t len);
61 static size_t utf16_string_len_in_bytes_max (const char *str, size_t len);
70 /* Registry file, memory mapped if read-only, or malloc'd if writing. */
73 struct ntreg_header *hdr;
76 /* Use a bitmap to store which file offsets are valid (point to a
77 * used block). We only need to store 1 bit per 32 bits of the file
78 * (because blocks are 4-byte aligned). We found that the average
79 * block size in a registry file is ~50 bytes. So roughly 1 in 12
80 * bits in the bitmap will be set, making it likely a more efficient
81 * structure than a hash table.
84 #define BITMAP_SET(bitmap,off) (bitmap[(off)>>5] |= 1 << (((off)>>2)&7))
85 #define BITMAP_CLR(bitmap,off) (bitmap[(off)>>5] &= ~ (1 << (((off)>>2)&7)))
86 #define BITMAP_TST(bitmap,off) (bitmap[(off)>>5] & (1 << (((off)>>2)&7)))
87 #define IS_VALID_BLOCK(h,off) \
88 (((off) & 3) == 0 && \
90 (off) < (h)->size && \
91 BITMAP_TST((h)->bitmap,(off)))
93 /* Fields from the header, extracted from little-endianness hell. */
94 size_t rootoffs; /* Root key offset (always an nk-block). */
95 size_t endpages; /* Offset of end of pages. */
98 size_t endblocks; /* Offset to next block allocation (0
99 if not allocated anything yet). */
102 /* NB. All fields are little endian. */
103 struct ntreg_header {
104 char magic[4]; /* "regf" */
107 char last_modified[8];
108 uint32_t major_ver; /* 1 */
109 uint32_t minor_ver; /* 3 */
110 uint32_t unknown5; /* 0 */
111 uint32_t unknown6; /* 1 */
112 uint32_t offset; /* offset of root key record - 4KB */
113 uint32_t blocks; /* pointer AFTER last hbin in file - 4KB */
114 uint32_t unknown7; /* 1 */
116 char name[64]; /* original file name of hive */
117 char unknown_guid1[16];
118 char unknown_guid2[16];
121 char unknown_guid3[16];
126 uint32_t csum; /* checksum: xor of dwords 0-0x1fb. */
128 char unknown11[3528];
130 char unknown_guid4[16];
131 char unknown_guid5[16];
132 char unknown_guid6[16];
136 } __attribute__((__packed__));
138 struct ntreg_hbin_page {
139 char magic[4]; /* "hbin" */
140 uint32_t offset_first; /* offset from 1st block */
141 uint32_t page_size; /* size of this page (multiple of 4KB) */
143 /* Linked list of blocks follows here. */
144 } __attribute__((__packed__));
146 struct ntreg_hbin_block {
147 int32_t seg_len; /* length of this block (-ve for used block) */
148 char id[2]; /* the block type (eg. "nk" for nk record) */
149 /* Block data follows here. */
150 } __attribute__((__packed__));
152 #define BLOCK_ID_EQ(h,offs,eqid) \
153 (STREQLEN (((struct ntreg_hbin_block *)((h)->addr + (offs)))->id, (eqid), 2))
156 block_len (hive_h *h, size_t blkoff, int *used)
158 struct ntreg_hbin_block *block;
159 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
161 int32_t len = le32toh (block->seg_len);
172 struct ntreg_nk_record {
173 int32_t seg_len; /* length (always -ve because used) */
174 char id[2]; /* "nk" */
178 uint32_t parent; /* offset of owner/parent */
179 uint32_t nr_subkeys; /* number of subkeys */
180 uint32_t nr_subkeys_volatile;
181 uint32_t subkey_lf; /* lf record containing list of subkeys */
182 uint32_t subkey_lf_volatile;
183 uint32_t nr_values; /* number of values */
184 uint32_t vallist; /* value-list record */
185 uint32_t sk; /* offset of sk-record */
186 uint32_t classname; /* offset of classname record */
187 uint16_t max_subkey_name_len; /* maximum length of a subkey name in bytes
188 if the subkey was reencoded as UTF-16LE */
191 uint32_t max_vk_name_len; /* maximum length of any vk name in bytes
192 if the name was reencoded as UTF-16LE */
193 uint32_t max_vk_data_len; /* maximum length of any vk data in bytes */
195 uint16_t name_len; /* length of name */
196 uint16_t classname_len; /* length of classname */
197 char name[1]; /* name follows here */
198 } __attribute__((__packed__));
200 struct ntreg_lf_record {
202 char id[2]; /* "lf"|"lh" */
203 uint16_t nr_keys; /* number of keys in this record */
205 uint32_t offset; /* offset of nk-record for this subkey */
206 char hash[4]; /* hash of subkey name */
208 } __attribute__((__packed__));
210 struct ntreg_ri_record {
212 char id[2]; /* "ri" */
213 uint16_t nr_offsets; /* number of pointers to lh records */
214 uint32_t offset[1]; /* list of pointers to lh records */
215 } __attribute__((__packed__));
217 /* This has no ID header. */
218 struct ntreg_value_list {
220 uint32_t offset[1]; /* list of pointers to vk records */
221 } __attribute__((__packed__));
223 struct ntreg_vk_record {
224 int32_t seg_len; /* length (always -ve because used) */
225 char id[2]; /* "vk" */
226 uint16_t name_len; /* length of name */
227 /* length of the data:
228 * If data_len is <= 4, then it's stored inline.
229 * Top bit is set to indicate inline.
232 uint32_t data_offset; /* pointer to the data (or data if inline) */
233 uint32_t data_type; /* type of the data */
234 uint16_t flags; /* bit 0 set => key name ASCII,
235 bit 0 clr => key name UTF-16.
236 Only seen ASCII here in the wild.
237 NB: this is CLEAR for default key. */
239 char name[1]; /* key name follows here */
240 } __attribute__((__packed__));
242 struct ntreg_sk_record {
243 int32_t seg_len; /* length (always -ve because used) */
244 char id[2]; /* "sk" */
246 uint32_t sk_next; /* linked into a circular list */
248 uint32_t refcount; /* reference count */
249 uint32_t sec_len; /* length of security info */
250 char sec_desc[1]; /* security info follows */
251 } __attribute__((__packed__));
254 header_checksum (const hive_h *h)
256 uint32_t *daddr = (uint32_t *) h->addr;
260 for (i = 0; i < 0x1fc / 4; ++i) {
261 sum ^= le32toh (*daddr);
268 #define HIVEX_OPEN_MSGLVL_MASK (HIVEX_OPEN_VERBOSE|HIVEX_OPEN_DEBUG)
271 hivex_open (const char *filename, int flags)
275 assert (sizeof (struct ntreg_header) == 0x1000);
276 assert (offsetof (struct ntreg_header, csum) == 0x1fc);
278 h = calloc (1, sizeof *h);
282 h->msglvl = flags & HIVEX_OPEN_MSGLVL_MASK;
284 const char *debug = getenv ("HIVEX_DEBUG");
285 if (debug && STREQ (debug, "1"))
289 fprintf (stderr, "hivex_open: created handle %p\n", h);
291 h->writable = !!(flags & HIVEX_OPEN_WRITE);
292 h->filename = strdup (filename);
293 if (h->filename == NULL)
296 h->fd = open (filename, O_RDONLY | O_CLOEXEC);
301 if (fstat (h->fd, &statbuf) == -1)
304 h->size = statbuf.st_size;
307 h->addr = mmap (NULL, h->size, PROT_READ, MAP_SHARED, h->fd, 0);
308 if (h->addr == MAP_FAILED)
312 fprintf (stderr, "hivex_open: mapped file at %p\n", h->addr);
314 h->addr = malloc (h->size);
318 if (full_read (h->fd, h->addr, h->size) < h->size)
323 if (h->hdr->magic[0] != 'r' ||
324 h->hdr->magic[1] != 'e' ||
325 h->hdr->magic[2] != 'g' ||
326 h->hdr->magic[3] != 'f') {
327 fprintf (stderr, "hivex: %s: not a Windows NT Registry hive file\n",
333 /* Check major version. */
334 uint32_t major_ver = le32toh (h->hdr->major_ver);
335 if (major_ver != 1) {
337 "hivex: %s: hive file major version %" PRIu32 " (expected 1)\n",
338 filename, major_ver);
343 h->bitmap = calloc (1 + h->size / 32, 1);
344 if (h->bitmap == NULL)
347 /* Header checksum. */
348 uint32_t sum = header_checksum (h);
349 if (sum != le32toh (h->hdr->csum)) {
350 fprintf (stderr, "hivex: %s: bad checksum in hive header\n", filename);
355 if (h->msglvl >= 2) {
356 char *name = windows_utf16_to_utf8 (h->hdr->name, 64);
359 "hivex_open: header fields:\n"
360 " file version %" PRIu32 ".%" PRIu32 "\n"
361 " sequence nos %" PRIu32 " %" PRIu32 "\n"
362 " (sequences nos should match if hive was synched at shutdown)\n"
363 " original file name %s\n"
364 " (only 32 chars are stored, name is probably truncated)\n"
365 " root offset 0x%x + 0x1000\n"
366 " end of last page 0x%x + 0x1000 (total file size 0x%zx)\n"
367 " checksum 0x%x (calculated 0x%x)\n",
368 major_ver, le32toh (h->hdr->minor_ver),
369 le32toh (h->hdr->sequence1), le32toh (h->hdr->sequence2),
370 name ? name : "(conversion failed)",
371 le32toh (h->hdr->offset),
372 le32toh (h->hdr->blocks), h->size,
373 le32toh (h->hdr->csum), sum);
377 h->rootoffs = le32toh (h->hdr->offset) + 0x1000;
378 h->endpages = le32toh (h->hdr->blocks) + 0x1000;
381 fprintf (stderr, "hivex_open: root offset = 0x%zx\n", h->rootoffs);
383 /* We'll set this flag when we see a block with the root offset (ie.
386 int seen_root_block = 0, bad_root_block = 0;
388 /* Collect some stats. */
389 size_t pages = 0; /* Number of hbin pages read. */
390 size_t smallest_page = SIZE_MAX, largest_page = 0;
391 size_t blocks = 0; /* Total number of blocks found. */
392 size_t smallest_block = SIZE_MAX, largest_block = 0, blocks_bytes = 0;
393 size_t used_blocks = 0; /* Total number of used blocks found. */
394 size_t used_size = 0; /* Total size (bytes) of used blocks. */
396 /* Read the pages and blocks. The aim here is to be robust against
397 * corrupt or malicious registries. So we make sure the loops
398 * always make forward progress. We add the address of each block
399 * we read to a hash table so pointers will only reference the start
403 struct ntreg_hbin_page *page;
404 for (off = 0x1000; off < h->size; off += le32toh (page->page_size)) {
405 if (off >= h->endpages)
408 page = (struct ntreg_hbin_page *) (h->addr + off);
409 if (page->magic[0] != 'h' ||
410 page->magic[1] != 'b' ||
411 page->magic[2] != 'i' ||
412 page->magic[3] != 'n') {
413 fprintf (stderr, "hivex: %s: trailing garbage at end of file "
414 "(at 0x%zx, after %zu pages)\n",
415 filename, off, pages);
420 size_t page_size = le32toh (page->page_size);
422 fprintf (stderr, "hivex_open: page at 0x%zx, size %zu\n", off, page_size);
424 if (page_size < smallest_page) smallest_page = page_size;
425 if (page_size > largest_page) largest_page = page_size;
427 if (page_size <= sizeof (struct ntreg_hbin_page) ||
428 (page_size & 0x0fff) != 0) {
429 fprintf (stderr, "hivex: %s: page size %zu at 0x%zx, bad registry\n",
430 filename, page_size, off);
435 /* Read the blocks in this page. */
437 struct ntreg_hbin_block *block;
439 for (blkoff = off + 0x20;
440 blkoff < off + page_size;
444 int is_root = blkoff == h->rootoffs;
448 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
450 seg_len = block_len (h, blkoff, &used);
451 if (seg_len <= 4 || (seg_len & 3) != 0) {
452 fprintf (stderr, "hivex: %s: block size %" PRIu32 " at 0x%zx,"
454 filename, le32toh (block->seg_len), blkoff);
460 fprintf (stderr, "hivex_open: %s block id %d,%d at 0x%zx size %zu%s\n",
461 used ? "used" : "free", block->id[0], block->id[1], blkoff,
462 seg_len, is_root ? " (root)" : "");
464 blocks_bytes += seg_len;
465 if (seg_len < smallest_block) smallest_block = seg_len;
466 if (seg_len > largest_block) largest_block = seg_len;
468 if (is_root && !used)
473 used_size += seg_len;
475 /* Root block must be an nk-block. */
476 if (is_root && (block->id[0] != 'n' || block->id[1] != 'k'))
479 /* Note this blkoff is a valid address. */
480 BITMAP_SET (h->bitmap, blkoff);
485 if (!seen_root_block) {
486 fprintf (stderr, "hivex: %s: no root block found\n", filename);
491 if (bad_root_block) {
492 fprintf (stderr, "hivex: %s: bad root block (free or not nk)\n", filename);
499 "hivex_open: successfully read Windows Registry hive file:\n"
500 " pages: %zu [sml: %zu, lge: %zu]\n"
501 " blocks: %zu [sml: %zu, avg: %zu, lge: %zu]\n"
502 " blocks used: %zu\n"
503 " bytes used: %zu\n",
504 pages, smallest_page, largest_page,
505 blocks, smallest_block, blocks_bytes / blocks, largest_block,
506 used_blocks, used_size);
514 if (h->addr && h->size && h->addr != MAP_FAILED) {
516 munmap (h->addr, h->size);
530 hivex_close (hive_h *h)
535 fprintf (stderr, "hivex_close\n");
539 munmap (h->addr, h->size);
549 /*----------------------------------------------------------------------
554 hivex_root (hive_h *h)
556 hive_node_h ret = h->rootoffs;
557 if (!IS_VALID_BLOCK (h, ret)) {
558 errno = HIVEX_NO_KEY;
565 hivex_node_name (hive_h *h, hive_node_h node)
567 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
572 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
574 /* AFAIK the node name is always plain ASCII, so no conversion
575 * to UTF-8 is necessary. However we do need to nul-terminate
579 /* nk->name_len is unsigned, 16 bit, so this is safe ... However
580 * we have to make sure the length doesn't exceed the block length.
582 size_t len = le16toh (nk->name_len);
583 size_t seg_len = block_len (h, node, NULL);
584 if (sizeof (struct ntreg_nk_record) + len - 1 > seg_len) {
586 fprintf (stderr, "hivex_node_name: returning EFAULT because node name"
587 " is too long (%zu, %zu)\n",
593 char *ret = malloc (len + 1);
596 memcpy (ret, nk->name, len);
602 /* I think the documentation for the sk and classname fields in the nk
603 * record is wrong, or else the offset field is in the wrong place.
604 * Otherwise this makes no sense. Disabled this for now -- it's not
605 * useful for reading the registry anyway.
609 hivex_node_security (hive_h *h, hive_node_h node)
611 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
616 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
618 hive_node_h ret = le32toh (nk->sk);
620 if (!IS_VALID_BLOCK (h, ret)) {
628 hivex_node_classname (hive_h *h, hive_node_h node)
630 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
635 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
637 hive_node_h ret = le32toh (nk->classname);
639 if (!IS_VALID_BLOCK (h, ret)) {
647 /* Structure for returning 0-terminated lists of offsets (nodes,
657 init_offset_list (struct offset_list *list)
661 list->offsets = NULL;
664 #define INIT_OFFSET_LIST(name) \
665 struct offset_list name; \
666 init_offset_list (&name)
668 /* Preallocates the offset_list, but doesn't make the contents longer. */
670 grow_offset_list (struct offset_list *list, size_t alloc)
672 assert (alloc >= list->len);
673 size_t *p = realloc (list->offsets, alloc * sizeof (size_t));
682 add_to_offset_list (struct offset_list *list, size_t offset)
684 if (list->len >= list->alloc) {
685 if (grow_offset_list (list, list->alloc ? list->alloc * 2 : 4) == -1)
688 list->offsets[list->len] = offset;
694 free_offset_list (struct offset_list *list)
696 free (list->offsets);
700 return_offset_list (struct offset_list *list)
702 if (add_to_offset_list (list, 0) == -1)
704 return list->offsets; /* caller frees */
707 /* Iterate over children, returning child nodes and intermediate blocks. */
708 #define GET_CHILDREN_NO_CHECK_NK 1
711 get_children (hive_h *h, hive_node_h node,
712 hive_node_h **children_ret, size_t **blocks_ret,
715 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
720 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
722 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
724 INIT_OFFSET_LIST (children);
725 INIT_OFFSET_LIST (blocks);
727 /* Deal with the common "no subkeys" case quickly. */
728 if (nr_subkeys_in_nk == 0)
731 /* Arbitrarily limit the number of subkeys we will ever deal with. */
732 if (nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS) {
734 fprintf (stderr, "hivex: get_children: returning ERANGE because "
735 "nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS (%zu > %d)\n",
736 nr_subkeys_in_nk, HIVEX_MAX_SUBKEYS);
741 /* Preallocate space for the children. */
742 if (grow_offset_list (&children, nr_subkeys_in_nk) == -1)
745 /* The subkey_lf field can point either to an lf-record, which is
746 * the common case, or if there are lots of subkeys, to an
749 size_t subkey_lf = le32toh (nk->subkey_lf);
751 if (!IS_VALID_BLOCK (h, subkey_lf)) {
753 fprintf (stderr, "hivex_node_children: returning EFAULT"
754 " because subkey_lf is not a valid block (0x%zx)\n",
760 if (add_to_offset_list (&blocks, subkey_lf) == -1)
763 struct ntreg_hbin_block *block =
764 (struct ntreg_hbin_block *) (h->addr + subkey_lf);
766 /* Points to lf-record? (Note, also "lh" but that is basically the
767 * same as "lf" as far as we are concerned here).
769 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
770 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
772 /* Check number of subkeys in the nk-record matches number of subkeys
775 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
778 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu,"
779 " nr_subkeys_in_lf = %zu\n",
780 nr_subkeys_in_nk, nr_subkeys_in_lf);
782 if (nr_subkeys_in_nk != nr_subkeys_in_lf) {
787 size_t len = block_len (h, subkey_lf, NULL);
788 if (8 + nr_subkeys_in_lf * 8 > len) {
790 fprintf (stderr, "hivex_node_children: returning EFAULT"
791 " because too many subkeys (%zu, %zu)\n",
792 nr_subkeys_in_lf, len);
798 for (i = 0; i < nr_subkeys_in_lf; ++i) {
799 hive_node_h subkey = le32toh (lf->keys[i].offset);
801 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
802 if (!IS_VALID_BLOCK (h, subkey)) {
804 fprintf (stderr, "hivex_node_children: returning EFAULT"
805 " because subkey is not a valid block (0x%zx)\n",
811 if (add_to_offset_list (&children, subkey) == -1)
816 /* Points to ri-record? */
817 else if (block->id[0] == 'r' && block->id[1] == 'i') {
818 struct ntreg_ri_record *ri = (struct ntreg_ri_record *) block;
820 size_t nr_offsets = le16toh (ri->nr_offsets);
822 /* Count total number of children. */
824 for (i = 0; i < nr_offsets; ++i) {
825 hive_node_h offset = le32toh (ri->offset[i]);
827 if (!IS_VALID_BLOCK (h, offset)) {
829 fprintf (stderr, "hivex_node_children: returning EFAULT"
830 " because ri-offset is not a valid block (0x%zx)\n",
835 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
837 fprintf (stderr, "get_children: returning ENOTSUP"
838 " because ri-record offset does not point to lf/lh (0x%zx)\n",
844 if (add_to_offset_list (&blocks, offset) == -1)
847 struct ntreg_lf_record *lf =
848 (struct ntreg_lf_record *) (h->addr + offset);
850 count += le16toh (lf->nr_keys);
854 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu,"
856 nr_subkeys_in_nk, count);
858 if (nr_subkeys_in_nk != count) {
863 /* Copy list of children. Note nr_subkeys_in_nk is limited to
864 * something reasonable above.
866 for (i = 0; i < nr_offsets; ++i) {
867 hive_node_h offset = le32toh (ri->offset[i]);
869 if (!IS_VALID_BLOCK (h, offset)) {
871 fprintf (stderr, "hivex_node_children: returning EFAULT"
872 " because ri-offset is not a valid block (0x%zx)\n",
877 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
879 fprintf (stderr, "get_children: returning ENOTSUP"
880 " because ri-record offset does not point to lf/lh (0x%zx)\n",
886 struct ntreg_lf_record *lf =
887 (struct ntreg_lf_record *) (h->addr + offset);
890 for (j = 0; j < le16toh (lf->nr_keys); ++j) {
891 hive_node_h subkey = le32toh (lf->keys[j].offset);
893 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
894 if (!IS_VALID_BLOCK (h, subkey)) {
896 fprintf (stderr, "hivex_node_children: returning EFAULT"
897 " because indirect subkey is not a valid block (0x%zx)\n",
903 if (add_to_offset_list (&children, subkey) == -1)
909 /* else not supported, set errno and fall through */
911 fprintf (stderr, "get_children: returning ENOTSUP"
912 " because subkey block is not lf/lh/ri (0x%zx, %d, %d)\n",
913 subkey_lf, block->id[0], block->id[1]);
916 free_offset_list (&children);
917 free_offset_list (&blocks);
921 *children_ret = return_offset_list (&children);
922 *blocks_ret = return_offset_list (&blocks);
923 if (!*children_ret || !*blocks_ret)
929 hivex_node_children (hive_h *h, hive_node_h node)
931 hive_node_h *children;
934 if (get_children (h, node, &children, &blocks, 0) == -1)
941 /* Very inefficient, but at least having a separate API call
942 * allows us to make it more efficient in future.
945 hivex_node_get_child (hive_h *h, hive_node_h node, const char *nname)
947 hive_node_h *children = NULL;
951 children = hivex_node_children (h, node);
952 if (!children) goto error;
955 for (i = 0; children[i] != 0; ++i) {
956 name = hivex_node_name (h, children[i]);
957 if (!name) goto error;
958 if (STRCASEEQ (name, nname)) {
962 free (name); name = NULL;
972 hivex_node_parent (hive_h *h, hive_node_h node)
974 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
979 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
981 hive_node_h ret = le32toh (nk->parent);
983 if (!IS_VALID_BLOCK (h, ret)) {
985 fprintf (stderr, "hivex_node_parent: returning EFAULT"
986 " because parent is not a valid block (0x%zx)\n",
995 get_values (hive_h *h, hive_node_h node,
996 hive_value_h **values_ret, size_t **blocks_ret)
998 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
1003 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
1005 size_t nr_values = le32toh (nk->nr_values);
1008 fprintf (stderr, "hivex_node_values: nr_values = %zu\n", nr_values);
1010 INIT_OFFSET_LIST (values);
1011 INIT_OFFSET_LIST (blocks);
1013 /* Deal with the common "no values" case quickly. */
1017 /* Arbitrarily limit the number of values we will ever deal with. */
1018 if (nr_values > HIVEX_MAX_VALUES) {
1020 fprintf (stderr, "hivex: get_values: returning ERANGE"
1021 " because nr_values > HIVEX_MAX_VALUES (%zu > %d)\n",
1022 nr_values, HIVEX_MAX_VALUES);
1027 /* Preallocate space for the values. */
1028 if (grow_offset_list (&values, nr_values) == -1)
1031 /* Get the value list and check it looks reasonable. */
1032 size_t vlist_offset = le32toh (nk->vallist);
1033 vlist_offset += 0x1000;
1034 if (!IS_VALID_BLOCK (h, vlist_offset)) {
1036 fprintf (stderr, "hivex_node_values: returning EFAULT"
1037 " because value list is not a valid block (0x%zx)\n",
1043 if (add_to_offset_list (&blocks, vlist_offset) == -1)
1046 struct ntreg_value_list *vlist =
1047 (struct ntreg_value_list *) (h->addr + vlist_offset);
1049 size_t len = block_len (h, vlist_offset, NULL);
1050 if (4 + nr_values * 4 > len) {
1052 fprintf (stderr, "hivex_node_values: returning EFAULT"
1053 " because value list is too long (%zu, %zu)\n",
1060 for (i = 0; i < nr_values; ++i) {
1061 hive_node_h value = le32toh (vlist->offset[i]);
1063 if (!IS_VALID_BLOCK (h, value)) {
1065 fprintf (stderr, "hivex_node_values: returning EFAULT"
1066 " because value is not a valid block (0x%zx)\n",
1071 if (add_to_offset_list (&values, value) == -1)
1076 *values_ret = return_offset_list (&values);
1077 *blocks_ret = return_offset_list (&blocks);
1078 if (!*values_ret || !*blocks_ret)
1083 free_offset_list (&values);
1084 free_offset_list (&blocks);
1089 hivex_node_values (hive_h *h, hive_node_h node)
1091 hive_value_h *values;
1094 if (get_values (h, node, &values, &blocks) == -1)
1101 /* Very inefficient, but at least having a separate API call
1102 * allows us to make it more efficient in future.
1105 hivex_node_get_value (hive_h *h, hive_node_h node, const char *key)
1107 hive_value_h *values = NULL;
1109 hive_value_h ret = 0;
1111 values = hivex_node_values (h, node);
1112 if (!values) goto error;
1115 for (i = 0; values[i] != 0; ++i) {
1116 name = hivex_value_key (h, values[i]);
1117 if (!name) goto error;
1118 if (STRCASEEQ (name, key)) {
1122 free (name); name = NULL;
1132 hivex_value_key (hive_h *h, hive_value_h value)
1134 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1139 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1141 /* AFAIK the key is always plain ASCII, so no conversion to UTF-8 is
1142 * necessary. However we do need to nul-terminate the string.
1145 /* vk->name_len is unsigned, 16 bit, so this is safe ... However
1146 * we have to make sure the length doesn't exceed the block length.
1148 size_t len = le16toh (vk->name_len);
1149 size_t seg_len = block_len (h, value, NULL);
1150 if (sizeof (struct ntreg_vk_record) + len - 1 > seg_len) {
1152 fprintf (stderr, "hivex_value_key: returning EFAULT"
1153 " because key length is too long (%zu, %zu)\n",
1159 char *ret = malloc (len + 1);
1162 memcpy (ret, vk->name, len);
1168 hivex_value_type (hive_h *h, hive_value_h value, hive_type *t, size_t *len)
1170 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1175 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1178 *t = le32toh (vk->data_type);
1181 *len = le32toh (vk->data_len);
1182 *len &= 0x7fffffff; /* top bit indicates if data is stored inline */
1189 hivex_value_value (hive_h *h, hive_value_h value,
1190 hive_type *t_rtn, size_t *len_rtn)
1192 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1197 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1203 t = le32toh (vk->data_type);
1205 len = le32toh (vk->data_len);
1206 is_inline = !!(len & 0x80000000);
1210 fprintf (stderr, "hivex_value_value: value=0x%zx, t=%d, len=%zu, inline=%d\n",
1211 value, t, len, is_inline);
1218 if (is_inline && len > 4) {
1223 /* Arbitrarily limit the length that we will read. */
1224 if (len > HIVEX_MAX_VALUE_LEN) {
1226 fprintf (stderr, "hivex_value_value: returning ERANGE because data "
1227 "length > HIVEX_MAX_VALUE_LEN (%zu > %d)\n",
1228 len, HIVEX_MAX_SUBKEYS);
1233 char *ret = malloc (len);
1238 memcpy (ret, (char *) &vk->data_offset, len);
1242 size_t data_offset = le32toh (vk->data_offset);
1243 data_offset += 0x1000;
1244 if (!IS_VALID_BLOCK (h, data_offset)) {
1246 fprintf (stderr, "hivex_value_value: returning EFAULT because data "
1247 "offset is not a valid block (0x%zx)\n",
1254 /* Check that the declared size isn't larger than the block its in.
1256 * XXX Some apparently valid registries are seen to have this,
1257 * so turn this into a warning and substitute the smaller length
1260 size_t blen = block_len (h, data_offset, NULL);
1261 if (len > blen - 4 /* subtract 4 for block header */) {
1263 fprintf (stderr, "hivex_value_value: warning: declared data length "
1264 "is longer than the block it is in "
1265 "(data 0x%zx, data len %zu, block len %zu)\n",
1266 data_offset, len, blen);
1269 /* Return the smaller length to the caller too. */
1274 char *data = h->addr + data_offset + 4;
1275 memcpy (ret, data, len);
1280 windows_utf16_to_utf8 (/* const */ char *input, size_t len)
1282 iconv_t ic = iconv_open ("UTF-8", "UTF-16");
1283 if (ic == (iconv_t) -1)
1286 /* iconv(3) has an insane interface ... */
1288 /* Mostly UTF-8 will be smaller, so this is a good initial guess. */
1289 size_t outalloc = len;
1293 size_t outlen = outalloc;
1294 char *out = malloc (outlen + 1);
1304 size_t r = iconv (ic, &inp, &inlen, &outp, &outlen);
1305 if (r == (size_t) -1) {
1306 if (errno == E2BIG) {
1308 size_t prev = outalloc;
1309 /* Try again with a larger output buffer. */
1312 if (outalloc < prev) {
1320 /* Else some conversion failure, eg. EILSEQ, EINVAL. */
1336 hivex_value_string (hive_h *h, hive_value_h value)
1340 char *data = hivex_value_value (h, value, &t, &len);
1345 if (t != hive_t_string && t != hive_t_expand_string && t != hive_t_link) {
1351 /* Deal with the case where Windows has allocated a large buffer
1352 * full of random junk, and only the first few bytes of the buffer
1353 * contain a genuine UTF-16 string.
1355 * In this case, iconv would try to process the junk bytes as UTF-16
1356 * and inevitably find an illegal sequence (EILSEQ). Instead, stop
1357 * after we find the first \0\0.
1359 * (Found by Hilko Bengen in a fresh Windows XP SOFTWARE hive).
1361 size_t slen = utf16_string_len_in_bytes_max (data, len);
1365 char *ret = windows_utf16_to_utf8 (data, len);
1374 free_strings (char **argv)
1379 for (i = 0; argv[i] != NULL; ++i)
1385 /* Get the length of a UTF-16 format string. Handle the string as
1386 * pairs of bytes, looking for the first \0\0 pair. Only read up to
1387 * 'len' maximum bytes.
1390 utf16_string_len_in_bytes_max (const char *str, size_t len)
1394 while (len >= 2 && (str[0] || str[1])) {
1403 /* http://blogs.msdn.com/oldnewthing/archive/2009/10/08/9904646.aspx */
1405 hivex_value_multiple_strings (hive_h *h, hive_value_h value)
1409 char *data = hivex_value_value (h, value, &t, &len);
1414 if (t != hive_t_multiple_strings) {
1420 size_t nr_strings = 0;
1421 char **ret = malloc ((1 + nr_strings) * sizeof (char *));
1431 while (p < data + len &&
1432 (plen = utf16_string_len_in_bytes_max (p, data + len - p)) > 0) {
1434 char **ret2 = realloc (ret, (1 + nr_strings) * sizeof (char *));
1442 ret[nr_strings-1] = windows_utf16_to_utf8 (p, plen);
1443 ret[nr_strings] = NULL;
1444 if (ret[nr_strings-1] == NULL) {
1450 p += plen + 2 /* skip over UTF-16 \0\0 at the end of this string */;
1458 hivex_value_dword (hive_h *h, hive_value_h value)
1462 char *data = hivex_value_value (h, value, &t, &len);
1467 if ((t != hive_t_dword && t != hive_t_dword_be) || len != 4) {
1473 int32_t ret = *(int32_t*)data;
1475 if (t == hive_t_dword) /* little endian */
1476 ret = le32toh (ret);
1478 ret = be32toh (ret);
1484 hivex_value_qword (hive_h *h, hive_value_h value)
1488 char *data = hivex_value_value (h, value, &t, &len);
1493 if (t != hive_t_qword || len != 8) {
1499 int64_t ret = *(int64_t*)data;
1501 ret = le64toh (ret); /* always little endian */
1506 /*----------------------------------------------------------------------
1511 hivex_visit (hive_h *h, const struct hivex_visitor *visitor, size_t len,
1512 void *opaque, int flags)
1514 return hivex_visit_node (h, hivex_root (h), visitor, len, opaque, flags);
1517 static int hivex__visit_node (hive_h *h, hive_node_h node,
1518 const struct hivex_visitor *vtor,
1519 char *unvisited, void *opaque, int flags);
1522 hivex_visit_node (hive_h *h, hive_node_h node,
1523 const struct hivex_visitor *visitor, size_t len, void *opaque,
1526 struct hivex_visitor vtor;
1527 memset (&vtor, 0, sizeof vtor);
1529 /* Note that len might be larger *or smaller* than the expected size. */
1530 size_t copysize = len <= sizeof vtor ? len : sizeof vtor;
1531 memcpy (&vtor, visitor, copysize);
1533 /* This bitmap records unvisited nodes, so we don't loop if the
1534 * registry contains cycles.
1536 char *unvisited = malloc (1 + h->size / 32);
1537 if (unvisited == NULL)
1539 memcpy (unvisited, h->bitmap, 1 + h->size / 32);
1541 int r = hivex__visit_node (h, node, &vtor, unvisited, opaque, flags);
1547 hivex__visit_node (hive_h *h, hive_node_h node,
1548 const struct hivex_visitor *vtor, char *unvisited,
1549 void *opaque, int flags)
1551 int skip_bad = flags & HIVEX_VISIT_SKIP_BAD;
1553 hive_value_h *values = NULL;
1554 hive_node_h *children = NULL;
1560 /* Return -1 on all callback errors. However on internal errors,
1561 * check if skip_bad is set and suppress those errors if so.
1565 if (!BITMAP_TST (unvisited, node)) {
1567 fprintf (stderr, "hivex__visit_node: contains cycle:"
1568 " visited node 0x%zx already\n",
1572 return skip_bad ? 0 : -1;
1574 BITMAP_CLR (unvisited, node);
1576 name = hivex_node_name (h, node);
1577 if (!name) return skip_bad ? 0 : -1;
1578 if (vtor->node_start && vtor->node_start (h, opaque, node, name) == -1)
1581 values = hivex_node_values (h, node);
1583 ret = skip_bad ? 0 : -1;
1587 for (i = 0; values[i] != 0; ++i) {
1591 if (hivex_value_type (h, values[i], &t, &len) == -1) {
1592 ret = skip_bad ? 0 : -1;
1596 key = hivex_value_key (h, values[i]);
1598 ret = skip_bad ? 0 : -1;
1602 if (vtor->value_any) {
1603 str = hivex_value_value (h, values[i], &t, &len);
1605 ret = skip_bad ? 0 : -1;
1608 if (vtor->value_any (h, opaque, node, values[i], t, len, key, str) == -1)
1610 free (str); str = NULL;
1615 str = hivex_value_value (h, values[i], &t, &len);
1617 ret = skip_bad ? 0 : -1;
1620 if (t != hive_t_none) {
1621 ret = skip_bad ? 0 : -1;
1624 if (vtor->value_none &&
1625 vtor->value_none (h, opaque, node, values[i], t, len, key, str) == -1)
1627 free (str); str = NULL;
1631 case hive_t_expand_string:
1633 str = hivex_value_string (h, values[i]);
1635 if (errno != EILSEQ && errno != EINVAL) {
1636 ret = skip_bad ? 0 : -1;
1639 if (vtor->value_string_invalid_utf16) {
1640 str = hivex_value_value (h, values[i], &t, &len);
1641 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i],
1642 t, len, key, str) == -1)
1644 free (str); str = NULL;
1648 if (vtor->value_string &&
1649 vtor->value_string (h, opaque, node, values[i],
1650 t, len, key, str) == -1)
1652 free (str); str = NULL;
1656 case hive_t_dword_be: {
1657 int32_t i32 = hivex_value_dword (h, values[i]);
1658 if (vtor->value_dword &&
1659 vtor->value_dword (h, opaque, node, values[i],
1660 t, len, key, i32) == -1)
1665 case hive_t_qword: {
1666 int64_t i64 = hivex_value_qword (h, values[i]);
1667 if (vtor->value_qword &&
1668 vtor->value_qword (h, opaque, node, values[i],
1669 t, len, key, i64) == -1)
1675 str = hivex_value_value (h, values[i], &t, &len);
1677 ret = skip_bad ? 0 : -1;
1680 if (t != hive_t_binary) {
1681 ret = skip_bad ? 0 : -1;
1684 if (vtor->value_binary &&
1685 vtor->value_binary (h, opaque, node, values[i],
1686 t, len, key, str) == -1)
1688 free (str); str = NULL;
1691 case hive_t_multiple_strings:
1692 strs = hivex_value_multiple_strings (h, values[i]);
1694 if (errno != EILSEQ && errno != EINVAL) {
1695 ret = skip_bad ? 0 : -1;
1698 if (vtor->value_string_invalid_utf16) {
1699 str = hivex_value_value (h, values[i], &t, &len);
1700 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i],
1701 t, len, key, str) == -1)
1703 free (str); str = NULL;
1707 if (vtor->value_multiple_strings &&
1708 vtor->value_multiple_strings (h, opaque, node, values[i],
1709 t, len, key, strs) == -1)
1711 free_strings (strs); strs = NULL;
1714 case hive_t_resource_list:
1715 case hive_t_full_resource_description:
1716 case hive_t_resource_requirements_list:
1718 str = hivex_value_value (h, values[i], &t, &len);
1720 ret = skip_bad ? 0 : -1;
1723 if (vtor->value_other &&
1724 vtor->value_other (h, opaque, node, values[i],
1725 t, len, key, str) == -1)
1727 free (str); str = NULL;
1732 free (key); key = NULL;
1735 children = hivex_node_children (h, node);
1736 if (children == NULL) {
1737 ret = skip_bad ? 0 : -1;
1741 for (i = 0; children[i] != 0; ++i) {
1743 fprintf (stderr, "hivex__visit_node: %s: visiting subkey %d (0x%zx)\n",
1744 name, i, children[i]);
1746 if (hivex__visit_node (h, children[i], vtor, unvisited, opaque, flags) == -1)
1750 if (vtor->node_end && vtor->node_end (h, opaque, node, name) == -1)
1761 free_strings (strs);
1765 /*----------------------------------------------------------------------
1769 /* Allocate an hbin (page), extending the malloc'd space if necessary,
1770 * and updating the hive handle fields (but NOT the hive disk header
1771 * -- the hive disk header is updated when we commit). This function
1772 * also extends the bitmap if necessary.
1774 * 'allocation_hint' is the size of the block allocation we would like
1775 * to make. Normally registry blocks are very small (avg 50 bytes)
1776 * and are contained in standard-sized pages (4KB), but the registry
1777 * can support blocks which are larger than a standard page, in which
1778 * case it creates a page of 8KB, 12KB etc.
1781 * > 0 : offset of first usable byte of new page (after page header)
1782 * 0 : error (errno set)
1785 allocate_page (hive_h *h, size_t allocation_hint)
1787 /* In almost all cases this will be 1. */
1788 size_t nr_4k_pages =
1789 1 + (allocation_hint + sizeof (struct ntreg_hbin_page) - 1) / 4096;
1790 assert (nr_4k_pages >= 1);
1792 /* 'extend' is the number of bytes to extend the file by. Note that
1793 * hives found in the wild often contain slack between 'endpages'
1794 * and the actual end of the file, so we don't always need to make
1797 ssize_t extend = h->endpages + nr_4k_pages * 4096 - h->size;
1799 if (h->msglvl >= 2) {
1800 fprintf (stderr, "allocate_page: current endpages = 0x%zx,"
1801 " current size = 0x%zx\n",
1802 h->endpages, h->size);
1803 fprintf (stderr, "allocate_page: extending file by %zd bytes"
1804 " (<= 0 if no extension)\n",
1809 size_t oldsize = h->size;
1810 size_t newsize = h->size + extend;
1811 char *newaddr = realloc (h->addr, newsize);
1812 if (newaddr == NULL)
1815 size_t oldbitmapsize = 1 + oldsize / 32;
1816 size_t newbitmapsize = 1 + newsize / 32;
1817 char *newbitmap = realloc (h->bitmap, newbitmapsize);
1818 if (newbitmap == NULL) {
1825 h->bitmap = newbitmap;
1827 memset (h->addr + oldsize, 0, newsize - oldsize);
1828 memset (h->bitmap + oldbitmapsize, 0, newbitmapsize - oldbitmapsize);
1831 size_t offset = h->endpages;
1832 h->endpages += nr_4k_pages * 4096;
1835 fprintf (stderr, "allocate_page: new endpages = 0x%zx, new size = 0x%zx\n",
1836 h->endpages, h->size);
1838 /* Write the hbin header. */
1839 struct ntreg_hbin_page *page =
1840 (struct ntreg_hbin_page *) (h->addr + offset);
1841 page->magic[0] = 'h';
1842 page->magic[1] = 'b';
1843 page->magic[2] = 'i';
1844 page->magic[3] = 'n';
1845 page->offset_first = htole32 (offset - 0x1000);
1846 page->page_size = htole32 (nr_4k_pages * 4096);
1847 memset (page->unknown, 0, sizeof (page->unknown));
1850 fprintf (stderr, "allocate_page: new page at 0x%zx\n", offset);
1852 /* Offset of first usable byte after the header. */
1853 return offset + sizeof (struct ntreg_hbin_page);
1856 /* Allocate a single block, first allocating an hbin (page) at the end
1857 * of the current file if necessary. NB. To keep the implementation
1858 * simple and more likely to be correct, we do not reuse existing free
1861 * seg_len is the size of the block (this INCLUDES the block header).
1862 * The header of the block is initialized to -seg_len (negative to
1863 * indicate used). id[2] is the block ID (type), eg. "nk" for nk-
1864 * record. The block bitmap is updated to show this block as valid.
1865 * The rest of the contents of the block will be zero.
1867 * **NB** Because allocate_block may reallocate the memory, all
1868 * pointers into the memory become potentially invalid. I really
1869 * love writing in C, can't you tell?
1872 * > 0 : offset of new block
1873 * 0 : error (errno set)
1876 allocate_block (hive_h *h, size_t seg_len, const char id[2])
1884 /* The caller probably forgot to include the header. Note that
1885 * value lists have no ID field, so seg_len == 4 would be possible
1886 * for them, albeit unusual.
1889 fprintf (stderr, "allocate_block: refusing too small allocation (%zu),"
1890 " returning ERANGE\n", seg_len);
1895 /* Refuse really large allocations. */
1896 if (seg_len > HIVEX_MAX_ALLOCATION) {
1898 fprintf (stderr, "allocate_block: refusing large allocation (%zu),"
1899 " returning ERANGE\n", seg_len);
1904 /* Round up allocation to multiple of 8 bytes. All blocks must be
1905 * on an 8 byte boundary.
1907 seg_len = (seg_len + 7) & ~7;
1909 /* Allocate a new page if necessary. */
1910 if (h->endblocks == 0 || h->endblocks + seg_len > h->endpages) {
1911 size_t newendblocks = allocate_page (h, seg_len);
1912 if (newendblocks == 0)
1914 h->endblocks = newendblocks;
1917 size_t offset = h->endblocks;
1920 fprintf (stderr, "allocate_block: new block at 0x%zx, size %zu\n",
1923 struct ntreg_hbin_block *blockhdr =
1924 (struct ntreg_hbin_block *) (h->addr + offset);
1926 memset (blockhdr, 0, seg_len);
1928 blockhdr->seg_len = htole32 (- (int32_t) seg_len);
1929 if (id[0] && id[1] && seg_len >= sizeof (struct ntreg_hbin_block)) {
1930 blockhdr->id[0] = id[0];
1931 blockhdr->id[1] = id[1];
1934 BITMAP_SET (h->bitmap, offset);
1936 h->endblocks += seg_len;
1938 /* If there is space after the last block in the last page, then we
1939 * have to put a dummy free block header here to mark the rest of
1942 ssize_t rem = h->endpages - h->endblocks;
1945 fprintf (stderr, "allocate_block: marking remainder of page free"
1946 " starting at 0x%zx, size %zd\n", h->endblocks, rem);
1950 blockhdr = (struct ntreg_hbin_block *) (h->addr + h->endblocks);
1951 blockhdr->seg_len = htole32 ((int32_t) rem);
1957 /* 'offset' must point to a valid, used block. This function marks
1958 * the block unused (by updating the seg_len field) and invalidates
1959 * the bitmap. It does NOT do this recursively, so to avoid creating
1960 * unreachable used blocks, callers may have to recurse over the hive
1961 * structures. Also callers must ensure there are no references to
1962 * this block from other parts of the hive.
1965 mark_block_unused (hive_h *h, size_t offset)
1967 assert (h->writable);
1968 assert (IS_VALID_BLOCK (h, offset));
1971 fprintf (stderr, "mark_block_unused: marking 0x%zx unused\n", offset);
1973 struct ntreg_hbin_block *blockhdr =
1974 (struct ntreg_hbin_block *) (h->addr + offset);
1976 size_t seg_len = block_len (h, offset, NULL);
1977 blockhdr->seg_len = htole32 (seg_len);
1979 BITMAP_CLR (h->bitmap, offset);
1982 /* Delete all existing values at this node. */
1984 delete_values (hive_h *h, hive_node_h node)
1986 assert (h->writable);
1988 hive_value_h *values;
1990 if (get_values (h, node, &values, &blocks) == -1)
1994 for (i = 0; blocks[i] != 0; ++i)
1995 mark_block_unused (h, blocks[i]);
1999 for (i = 0; values[i] != 0; ++i) {
2000 struct ntreg_vk_record *vk =
2001 (struct ntreg_vk_record *) (h->addr + values[i]);
2005 len = le32toh (vk->data_len);
2006 is_inline = !!(len & 0x80000000); /* top bit indicates is inline */
2009 if (!is_inline) { /* non-inline, so remove data block */
2010 size_t data_offset = le32toh (vk->data_offset);
2011 data_offset += 0x1000;
2012 mark_block_unused (h, data_offset);
2015 /* remove vk record */
2016 mark_block_unused (h, values[i]);
2021 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2022 nk->nr_values = htole32 (0);
2023 nk->vallist = htole32 (0xffffffff);
2029 hivex_commit (hive_h *h, const char *filename, int flags)
2041 filename = filename ? : h->filename;
2042 int fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
2046 /* Update the header fields. */
2047 uint32_t sequence = le32toh (h->hdr->sequence1);
2049 h->hdr->sequence1 = htole32 (sequence);
2050 h->hdr->sequence2 = htole32 (sequence);
2051 /* XXX Ought to update h->hdr->last_modified. */
2052 h->hdr->blocks = htole32 (h->endpages - 0x1000);
2054 /* Recompute header checksum. */
2055 uint32_t sum = header_checksum (h);
2056 h->hdr->csum = htole32 (sum);
2059 fprintf (stderr, "hivex_commit: new header checksum: 0x%x\n", sum);
2061 if (full_write (fd, h->addr, h->size) != h->size) {
2068 if (close (fd) == -1)
2074 /* Calculate the hash for a lf or lh record offset.
2077 calc_hash (const char *type, const char *name, char *ret)
2079 size_t len = strlen (name);
2081 if (STRPREFIX (type, "lf"))
2082 /* Old-style, not used in current registries. */
2083 memcpy (ret, name, len < 4 ? len : 4);
2085 /* New-style for lh-records. */
2088 for (i = 0; i < len; ++i) {
2089 c = c_toupper (name[i]);
2093 *((uint32_t *) ret) = htole32 (h);
2097 /* Create a completely new lh-record containing just the single node. */
2099 new_lh_record (hive_h *h, const char *name, hive_node_h node)
2101 static const char id[2] = { 'l', 'h' };
2102 size_t seg_len = sizeof (struct ntreg_lf_record);
2103 size_t offset = allocate_block (h, seg_len, id);
2107 struct ntreg_lf_record *lh = (struct ntreg_lf_record *) (h->addr + offset);
2108 lh->nr_keys = htole16 (1);
2109 lh->keys[0].offset = htole32 (node - 0x1000);
2110 calc_hash ("lh", name, lh->keys[0].hash);
2115 /* Insert node into existing lf/lh-record at position.
2116 * This allocates a new record and marks the old one as unused.
2119 insert_lf_record (hive_h *h, size_t old_offs, size_t posn,
2120 const char *name, hive_node_h node)
2122 assert (IS_VALID_BLOCK (h, old_offs));
2124 /* Work around C stupidity.
2125 * http://www.redhat.com/archives/libguestfs/2010-February/msg00056.html
2127 int test = BLOCK_ID_EQ (h, old_offs, "lf") || BLOCK_ID_EQ (h, old_offs, "lh");
2130 struct ntreg_lf_record *old_lf =
2131 (struct ntreg_lf_record *) (h->addr + old_offs);
2132 size_t nr_keys = le16toh (old_lf->nr_keys);
2134 nr_keys++; /* in new record ... */
2136 size_t seg_len = sizeof (struct ntreg_lf_record) + (nr_keys-1) * 8;
2138 /* Copy the old_lf->id in case it moves during allocate_block. */
2140 memcpy (id, old_lf->id, sizeof id);
2142 size_t new_offs = allocate_block (h, seg_len, id);
2146 /* old_lf could have been invalidated by allocate_block. */
2147 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2149 struct ntreg_lf_record *new_lf =
2150 (struct ntreg_lf_record *) (h->addr + new_offs);
2151 new_lf->nr_keys = htole16 (nr_keys);
2153 /* Copy the keys until we reach posn, insert the new key there, then
2154 * copy the remaining keys.
2157 for (i = 0; i < posn; ++i)
2158 new_lf->keys[i] = old_lf->keys[i];
2160 new_lf->keys[i].offset = htole32 (node - 0x1000);
2161 calc_hash (new_lf->id, name, new_lf->keys[i].hash);
2163 for (i = posn+1; i < nr_keys; ++i)
2164 new_lf->keys[i] = old_lf->keys[i-1];
2166 /* Old block is unused, return new block. */
2167 mark_block_unused (h, old_offs);
2171 /* Compare name with name in nk-record. */
2173 compare_name_with_nk_name (hive_h *h, const char *name, hive_node_h nk_offs)
2175 assert (IS_VALID_BLOCK (h, nk_offs));
2176 assert (BLOCK_ID_EQ (h, nk_offs, "nk"));
2178 /* Name in nk is not necessarily nul-terminated. */
2179 char *nname = hivex_node_name (h, nk_offs);
2181 /* Unfortunately we don't have a way to return errors here. */
2183 perror ("compare_name_with_nk_name");
2187 int r = strcasecmp (name, nname);
2194 hivex_node_add_child (hive_h *h, hive_node_h parent, const char *name)
2201 if (!IS_VALID_BLOCK (h, parent) || !BLOCK_ID_EQ (h, parent, "nk")) {
2206 if (name == NULL || strlen (name) == 0) {
2211 if (hivex_node_get_child (h, parent, name) != 0) {
2216 /* Create the new nk-record. */
2217 static const char nk_id[2] = { 'n', 'k' };
2218 size_t seg_len = sizeof (struct ntreg_nk_record) + strlen (name);
2219 hive_node_h node = allocate_block (h, seg_len, nk_id);
2224 fprintf (stderr, "hivex_node_add_child: allocated new nk-record"
2225 " for child at 0x%zx\n", node);
2227 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2228 nk->flags = htole16 (0x0020); /* key is ASCII. */
2229 nk->parent = htole32 (parent - 0x1000);
2230 nk->subkey_lf = htole32 (0xffffffff);
2231 nk->subkey_lf_volatile = htole32 (0xffffffff);
2232 nk->vallist = htole32 (0xffffffff);
2233 nk->classname = htole32 (0xffffffff);
2234 nk->name_len = htole16 (strlen (name));
2235 strcpy (nk->name, name);
2237 /* Inherit parent sk. */
2238 struct ntreg_nk_record *parent_nk =
2239 (struct ntreg_nk_record *) (h->addr + parent);
2240 size_t parent_sk_offset = le32toh (parent_nk->sk);
2241 parent_sk_offset += 0x1000;
2242 if (!IS_VALID_BLOCK (h, parent_sk_offset) ||
2243 !BLOCK_ID_EQ (h, parent_sk_offset, "sk")) {
2245 fprintf (stderr, "hivex_node_add_child: returning EFAULT"
2246 " because parent sk is not a valid block (%zu)\n",
2251 struct ntreg_sk_record *sk =
2252 (struct ntreg_sk_record *) (h->addr + parent_sk_offset);
2253 sk->refcount = htole32 (le32toh (sk->refcount) + 1);
2254 nk->sk = htole32 (parent_sk_offset - 0x1000);
2256 /* Inherit parent timestamp. */
2257 memcpy (nk->timestamp, parent_nk->timestamp, sizeof (parent_nk->timestamp));
2259 /* What I found out the hard way (not documented anywhere): the
2260 * subkeys in lh-records must be kept sorted. If you just add a
2261 * subkey in a non-sorted position (eg. just add it at the end) then
2262 * Windows won't see the subkey _and_ Windows will corrupt the hive
2263 * itself when it modifies or saves it.
2265 * So use get_children() to get a list of intermediate
2266 * lf/lh-records. get_children() returns these in reading order
2267 * (which is sorted), so we look for the lf/lh-records in sequence
2268 * until we find the key name just after the one we are inserting,
2269 * and we insert the subkey just before it.
2271 * The only other case is the no-subkeys case, where we have to
2272 * create a brand new lh-record.
2274 hive_node_h *unused;
2277 if (get_children (h, parent, &unused, &blocks, 0) == -1)
2282 size_t nr_subkeys_in_parent_nk = le32toh (parent_nk->nr_subkeys);
2283 if (nr_subkeys_in_parent_nk == 0) { /* No subkeys case. */
2284 /* Free up any existing intermediate blocks. */
2285 for (i = 0; blocks[i] != 0; ++i)
2286 mark_block_unused (h, blocks[i]);
2287 size_t lh_offs = new_lh_record (h, name, node);
2293 /* Recalculate pointers that could have been invalidated by
2294 * previous call to allocate_block (via new_lh_record).
2296 nk = (struct ntreg_nk_record *) (h->addr + node);
2297 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2300 fprintf (stderr, "hivex_node_add_child: no keys, allocated new"
2301 " lh-record at 0x%zx\n", lh_offs);
2303 parent_nk->subkey_lf = htole32 (lh_offs - 0x1000);
2305 else { /* Insert subkeys case. */
2306 size_t old_offs = 0, new_offs = 0;
2307 struct ntreg_lf_record *old_lf = NULL;
2309 /* Find lf/lh key name just after the one we are inserting. */
2310 for (i = 0; blocks[i] != 0; ++i) {
2311 if (BLOCK_ID_EQ (h, blocks[i], "lf") ||
2312 BLOCK_ID_EQ (h, blocks[i], "lh")) {
2313 old_offs = blocks[i];
2314 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2315 for (j = 0; j < le16toh (old_lf->nr_keys); ++j) {
2316 hive_node_h nk_offs = le32toh (old_lf->keys[j].offset);
2318 if (compare_name_with_nk_name (h, name, nk_offs) < 0)
2324 /* Insert it at the end.
2325 * old_offs points to the last lf record, set j.
2327 assert (old_offs != 0); /* should never happen if nr_subkeys > 0 */
2328 j = le16toh (old_lf->nr_keys);
2333 fprintf (stderr, "hivex_node_add_child: insert key in existing"
2334 " lh-record at 0x%zx, posn %zu\n", old_offs, j);
2336 new_offs = insert_lf_record (h, old_offs, j, name, node);
2337 if (new_offs == 0) {
2342 /* Recalculate pointers that could have been invalidated by
2343 * previous call to allocate_block (via insert_lf_record).
2345 nk = (struct ntreg_nk_record *) (h->addr + node);
2346 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2349 fprintf (stderr, "hivex_node_add_child: new lh-record at 0x%zx\n",
2352 /* If the lf/lh-record was directly referenced by the parent nk,
2353 * then update the parent nk.
2355 if (le32toh (parent_nk->subkey_lf) + 0x1000 == old_offs)
2356 parent_nk->subkey_lf = htole32 (new_offs - 0x1000);
2357 /* Else we have to look for the intermediate ri-record and update
2361 for (i = 0; blocks[i] != 0; ++i) {
2362 if (BLOCK_ID_EQ (h, blocks[i], "ri")) {
2363 struct ntreg_ri_record *ri =
2364 (struct ntreg_ri_record *) (h->addr + blocks[i]);
2365 for (j = 0; j < le16toh (ri->nr_offsets); ++j)
2366 if (le32toh (ri->offset[j] + 0x1000) == old_offs) {
2367 ri->offset[j] = htole32 (new_offs - 0x1000);
2373 /* Not found .. This is an internal error. */
2375 fprintf (stderr, "hivex_node_add_child: returning ENOTSUP"
2376 " because could not find ri->lf link\n");
2388 /* Update nr_subkeys in parent nk. */
2389 nr_subkeys_in_parent_nk++;
2390 parent_nk->nr_subkeys = htole32 (nr_subkeys_in_parent_nk);
2392 /* Update max_subkey_name_len in parent nk. */
2393 uint16_t max = le16toh (parent_nk->max_subkey_name_len);
2394 if (max < strlen (name) * 2) /* *2 because "recoded" in UTF16-LE. */
2395 parent_nk->max_subkey_name_len = htole16 (strlen (name) * 2);
2400 /* Decrement the refcount of an sk-record, and if it reaches zero,
2401 * unlink it from the chain and delete it.
2404 delete_sk (hive_h *h, size_t sk_offset)
2406 if (!IS_VALID_BLOCK (h, sk_offset) || !BLOCK_ID_EQ (h, sk_offset, "sk")) {
2408 fprintf (stderr, "delete_sk: not an sk record: 0x%zx\n", sk_offset);
2413 struct ntreg_sk_record *sk = (struct ntreg_sk_record *) (h->addr + sk_offset);
2415 if (sk->refcount == 0) {
2417 fprintf (stderr, "delete_sk: sk record already has refcount 0: 0x%zx\n",
2425 if (sk->refcount == 0) {
2426 size_t sk_prev_offset = sk->sk_prev;
2427 sk_prev_offset += 0x1000;
2429 size_t sk_next_offset = sk->sk_next;
2430 sk_next_offset += 0x1000;
2432 /* Update sk_prev/sk_next SKs, unless they both point back to this
2433 * cell in which case we are deleting the last SK.
2435 if (sk_prev_offset != sk_offset && sk_next_offset != sk_offset) {
2436 struct ntreg_sk_record *sk_prev =
2437 (struct ntreg_sk_record *) (h->addr + sk_prev_offset);
2438 struct ntreg_sk_record *sk_next =
2439 (struct ntreg_sk_record *) (h->addr + sk_next_offset);
2441 sk_prev->sk_next = htole32 (sk_next_offset - 0x1000);
2442 sk_next->sk_prev = htole32 (sk_prev_offset - 0x1000);
2445 /* Refcount is zero so really delete this block. */
2446 mark_block_unused (h, sk_offset);
2452 /* Callback from hivex_node_delete_child which is called to delete a
2453 * node AFTER its subnodes have been visited. The subnodes have been
2454 * deleted but we still have to delete any lf/lh/li/ri records and the
2455 * value list block and values, followed by deleting the node itself.
2458 delete_node (hive_h *h, void *opaque, hive_node_h node, const char *name)
2460 /* Get the intermediate blocks. The subkeys have already been
2461 * deleted by this point, so tell get_children() not to check for
2462 * validity of the nk-records.
2464 hive_node_h *unused;
2466 if (get_children (h, node, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK) == -1)
2470 /* We don't care what's in these intermediate blocks, so we can just
2471 * delete them unconditionally.
2474 for (i = 0; blocks[i] != 0; ++i)
2475 mark_block_unused (h, blocks[i]);
2479 /* Delete the values in the node. */
2480 if (delete_values (h, node) == -1)
2483 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2485 /* If the NK references an SK, delete it. */
2486 size_t sk_offs = le32toh (nk->sk);
2487 if (sk_offs != 0xffffffff) {
2489 if (delete_sk (h, sk_offs) == -1)
2491 nk->sk = htole32 (0xffffffff);
2494 /* If the NK references a classname, delete it. */
2495 size_t cl_offs = le32toh (nk->classname);
2496 if (cl_offs != 0xffffffff) {
2498 mark_block_unused (h, cl_offs);
2499 nk->classname = htole32 (0xffffffff);
2502 /* Delete the node itself. */
2503 mark_block_unused (h, node);
2509 hivex_node_delete_child (hive_h *h, hive_node_h node)
2516 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2521 if (node == hivex_root (h)) {
2523 fprintf (stderr, "hivex_node_delete_child: cannot delete root node\n");
2528 hive_node_h parent = hivex_node_parent (h, node);
2532 /* Delete node and all its children and values recursively. */
2533 static const struct hivex_visitor visitor = { .node_end = delete_node };
2534 if (hivex_visit_node (h, node, &visitor, sizeof visitor, NULL, 0) == -1)
2537 /* Delete the link from parent to child. We need to find the lf/lh
2538 * record which contains the offset and remove the offset from that
2539 * record, then decrement the element count in that record, and
2540 * decrement the overall number of subkeys stored in the parent
2543 hive_node_h *unused;
2545 if (get_children (h, parent, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK)== -1)
2550 for (i = 0; blocks[i] != 0; ++i) {
2551 struct ntreg_hbin_block *block =
2552 (struct ntreg_hbin_block *) (h->addr + blocks[i]);
2554 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
2555 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
2557 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
2559 for (j = 0; j < nr_subkeys_in_lf; ++j)
2560 if (le32toh (lf->keys[j].offset) + 0x1000 == node) {
2561 for (; j < nr_subkeys_in_lf - 1; ++j)
2562 memcpy (&lf->keys[j], &lf->keys[j+1], sizeof (lf->keys[j]));
2563 lf->nr_keys = htole16 (nr_subkeys_in_lf - 1);
2569 fprintf (stderr, "hivex_node_delete_child: could not find parent"
2570 " to child link\n");
2575 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + parent);
2576 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
2577 nk->nr_subkeys = htole32 (nr_subkeys_in_nk - 1);
2580 fprintf (stderr, "hivex_node_delete_child: updating nr_subkeys"
2581 " in parent 0x%zx to %zu\n", parent, nr_subkeys_in_nk);
2587 hivex_node_set_values (hive_h *h, hive_node_h node,
2588 size_t nr_values, const hive_set_value *values,
2596 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2601 /* Delete all existing values. */
2602 if (delete_values (h, node) == -1)
2608 /* Allocate value list node. Value lists have no id field. */
2609 static const char nul_id[2] = { 0, 0 };
2611 sizeof (struct ntreg_value_list) + (nr_values - 1) * sizeof (uint32_t);
2612 size_t vallist_offs = allocate_block (h, seg_len, nul_id);
2613 if (vallist_offs == 0)
2616 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2617 nk->nr_values = htole32 (nr_values);
2618 nk->vallist = htole32 (vallist_offs - 0x1000);
2620 struct ntreg_value_list *vallist =
2621 (struct ntreg_value_list *) (h->addr + vallist_offs);
2624 for (i = 0; i < nr_values; ++i) {
2625 /* Allocate vk record to store this (key, value) pair. */
2626 static const char vk_id[2] = { 'v', 'k' };
2627 seg_len = sizeof (struct ntreg_vk_record) + strlen (values[i].key);
2628 size_t vk_offs = allocate_block (h, seg_len, vk_id);
2632 /* Recalculate pointers that could have been invalidated by
2633 * previous call to allocate_block.
2635 nk = (struct ntreg_nk_record *) (h->addr + node);
2636 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2638 vallist->offset[i] = htole32 (vk_offs - 0x1000);
2640 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2641 size_t name_len = strlen (values[i].key);
2642 vk->name_len = htole16 (name_len);
2643 strcpy (vk->name, values[i].key);
2644 vk->data_type = htole32 (values[i].t);
2645 uint32_t len = values[i].len;
2646 if (len <= 4) /* store it inline => set MSB flag */
2648 vk->data_len = htole32 (len);
2649 vk->flags = name_len == 0 ? 0 : 1;
2651 if (values[i].len <= 4) /* store it inline */
2652 memcpy (&vk->data_offset, values[i].value, values[i].len);
2654 size_t offs = allocate_block (h, values[i].len + 4, nul_id);
2658 /* Recalculate pointers that could have been invalidated by
2659 * previous call to allocate_block.
2661 nk = (struct ntreg_nk_record *) (h->addr + node);
2662 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2663 vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2665 memcpy (h->addr + offs + 4, values[i].value, values[i].len);
2666 vk->data_offset = htole32 (offs - 0x1000);
2669 if (name_len * 2 > le32toh (nk->max_vk_name_len))
2670 /* * 2 for UTF16-LE "reencoding" */
2671 nk->max_vk_name_len = htole32 (name_len * 2);
2672 if (values[i].len > le32toh (nk->max_vk_data_len))
2673 nk->max_vk_data_len = htole32 (values[i].len);
2680 hivex_node_set_value (hive_h *h, hive_node_h node,
2681 const hive_set_value *val, int flags)
2683 hive_value_h *prev_values = hivex_node_values (h, node);
2684 if (prev_values == NULL)
2689 size_t nr_values = 0;
2690 for (hive_value_h *itr = prev_values; *itr != 0; ++itr)
2693 hive_set_value *values = malloc ((nr_values + 1) * (sizeof (hive_set_value)));
2695 goto leave_prev_values;
2698 int idx_of_val = -1;
2699 hive_value_h *prev_val;
2700 for (prev_val = prev_values; *prev_val != 0; ++prev_val) {
2704 hive_set_value *value = &values[prev_val - prev_values];
2706 char *valval = hivex_value_value (h, *prev_val, &t, &len);
2707 if (valval == NULL) goto leave_partial;
2710 value->value = valval;
2714 char *valkey = hivex_value_key (h, *prev_val);
2715 if (valkey == NULL) goto leave_partial;
2718 value->key = valkey;
2720 if (STRCASEEQ (valkey, val->key))
2721 idx_of_val = prev_val - prev_values;
2724 if (idx_of_val > -1) {
2725 free (values[idx_of_val].key);
2726 free (values[idx_of_val].value);
2728 idx_of_val = nr_values;
2732 hive_set_value *value = &values[idx_of_val];
2733 *value = (hive_set_value){
2734 .key = strdup (val->key),
2735 .value = malloc (val->len),
2740 if (value->key == NULL || value->value == NULL) goto leave_partial;
2741 memcpy (value->value, val->value, val->len);
2743 retval = hivex_node_set_values (h, node, nr_values, values, 0);
2746 for (int i = 0; i < alloc_ct; i += 2) {
2747 free (values[i / 2].value);
2748 if (i + 1 < alloc_ct && values[i / 2].key != NULL)
2749 free (values[i / 2].key);