1 /* hivex - Windows Registry "hive" extraction library.
2 * Copyright (C) 2009-2010 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"
45 #define STREQ(a,b) (strcmp((a),(b)) == 0)
46 #define STRCASEEQ(a,b) (strcasecmp((a),(b)) == 0)
47 //#define STRNEQ(a,b) (strcmp((a),(b)) != 0)
48 //#define STRCASENEQ(a,b) (strcasecmp((a),(b)) != 0)
49 #define STREQLEN(a,b,n) (strncmp((a),(b),(n)) == 0)
50 //#define STRCASEEQLEN(a,b,n) (strncasecmp((a),(b),(n)) == 0)
51 //#define STRNEQLEN(a,b,n) (strncmp((a),(b),(n)) != 0)
52 //#define STRCASENEQLEN(a,b,n) (strncasecmp((a),(b),(n)) != 0)
53 #define STRPREFIX(a,b) (strncmp((a),(b),strlen((b))) == 0)
56 #include "byte_conversions.h"
58 /* These limits are in place to stop really stupid stuff and/or exploits. */
59 #define HIVEX_MAX_SUBKEYS 15000
60 #define HIVEX_MAX_VALUES 10000
61 #define HIVEX_MAX_VALUE_LEN 1000000
62 #define HIVEX_MAX_ALLOCATION 1000000
64 static char *windows_utf16_to_utf8 (/* const */ char *input, size_t len);
65 static size_t utf16_string_len_in_bytes (const char *str);
66 static size_t utf16_string_len_in_bytes_max (const char *str, size_t len);
75 /* Registry file, memory mapped if read-only, or malloc'd if writing. */
78 struct ntreg_header *hdr;
81 /* Use a bitmap to store which file offsets are valid (point to a
82 * used block). We only need to store 1 bit per 32 bits of the file
83 * (because blocks are 4-byte aligned). We found that the average
84 * block size in a registry file is ~50 bytes. So roughly 1 in 12
85 * bits in the bitmap will be set, making it likely a more efficient
86 * structure than a hash table.
89 #define BITMAP_SET(bitmap,off) (bitmap[(off)>>5] |= 1 << (((off)>>2)&7))
90 #define BITMAP_CLR(bitmap,off) (bitmap[(off)>>5] &= ~ (1 << (((off)>>2)&7)))
91 #define BITMAP_TST(bitmap,off) (bitmap[(off)>>5] & (1 << (((off)>>2)&7)))
92 #define IS_VALID_BLOCK(h,off) \
93 (((off) & 3) == 0 && \
95 (off) < (h)->size && \
96 BITMAP_TST((h)->bitmap,(off)))
98 /* Fields from the header, extracted from little-endianness hell. */
99 size_t rootoffs; /* Root key offset (always an nk-block). */
100 size_t endpages; /* Offset of end of pages. */
103 size_t endblocks; /* Offset to next block allocation (0
104 if not allocated anything yet). */
107 /* NB. All fields are little endian. */
108 struct ntreg_header {
109 char magic[4]; /* "regf" */
112 char last_modified[8];
113 uint32_t major_ver; /* 1 */
114 uint32_t minor_ver; /* 3 */
115 uint32_t unknown5; /* 0 */
116 uint32_t unknown6; /* 1 */
117 uint32_t offset; /* offset of root key record - 4KB */
118 uint32_t blocks; /* pointer AFTER last hbin in file - 4KB */
119 uint32_t unknown7; /* 1 */
121 char name[64]; /* original file name of hive */
122 char unknown_guid1[16];
123 char unknown_guid2[16];
126 char unknown_guid3[16];
131 uint32_t csum; /* checksum: xor of dwords 0-0x1fb. */
133 char unknown11[3528];
135 char unknown_guid4[16];
136 char unknown_guid5[16];
137 char unknown_guid6[16];
141 } __attribute__((__packed__));
143 struct ntreg_hbin_page {
144 char magic[4]; /* "hbin" */
145 uint32_t offset_first; /* offset from 1st block */
146 uint32_t page_size; /* size of this page (multiple of 4KB) */
148 /* Linked list of blocks follows here. */
149 } __attribute__((__packed__));
151 struct ntreg_hbin_block {
152 int32_t seg_len; /* length of this block (-ve for used block) */
153 char id[2]; /* the block type (eg. "nk" for nk record) */
154 /* Block data follows here. */
155 } __attribute__((__packed__));
157 #define BLOCK_ID_EQ(h,offs,eqid) \
158 (STREQLEN (((struct ntreg_hbin_block *)((h)->addr + (offs)))->id, (eqid), 2))
161 block_len (hive_h *h, size_t blkoff, int *used)
163 struct ntreg_hbin_block *block;
164 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
166 int32_t len = le32toh (block->seg_len);
177 struct ntreg_nk_record {
178 int32_t seg_len; /* length (always -ve because used) */
179 char id[2]; /* "nk" */
183 uint32_t parent; /* offset of owner/parent */
184 uint32_t nr_subkeys; /* number of subkeys */
185 uint32_t nr_subkeys_volatile;
186 uint32_t subkey_lf; /* lf record containing list of subkeys */
187 uint32_t subkey_lf_volatile;
188 uint32_t nr_values; /* number of values */
189 uint32_t vallist; /* value-list record */
190 uint32_t sk; /* offset of sk-record */
191 uint32_t classname; /* offset of classname record */
192 uint16_t max_subkey_name_len; /* maximum length of a subkey name in bytes
193 if the subkey was reencoded as UTF-16LE */
196 uint32_t max_vk_name_len; /* maximum length of any vk name in bytes
197 if the name was reencoded as UTF-16LE */
198 uint32_t max_vk_data_len; /* maximum length of any vk data in bytes */
200 uint16_t name_len; /* length of name */
201 uint16_t classname_len; /* length of classname */
202 char name[1]; /* name follows here */
203 } __attribute__((__packed__));
205 struct ntreg_lf_record {
207 char id[2]; /* "lf"|"lh" */
208 uint16_t nr_keys; /* number of keys in this record */
210 uint32_t offset; /* offset of nk-record for this subkey */
211 char hash[4]; /* hash of subkey name */
213 } __attribute__((__packed__));
215 struct ntreg_ri_record {
217 char id[2]; /* "ri" */
218 uint16_t nr_offsets; /* number of pointers to lh records */
219 uint32_t offset[1]; /* list of pointers to lh records */
220 } __attribute__((__packed__));
222 /* This has no ID header. */
223 struct ntreg_value_list {
225 uint32_t offset[1]; /* list of pointers to vk records */
226 } __attribute__((__packed__));
228 struct ntreg_vk_record {
229 int32_t seg_len; /* length (always -ve because used) */
230 char id[2]; /* "vk" */
231 uint16_t name_len; /* length of name */
232 /* length of the data:
233 * If data_len is <= 4, then it's stored inline.
234 * Top bit is set to indicate inline.
237 uint32_t data_offset; /* pointer to the data (or data if inline) */
238 uint32_t data_type; /* type of the data */
239 uint16_t flags; /* bit 0 set => key name ASCII,
240 bit 0 clr => key name UTF-16.
241 Only seen ASCII here in the wild.
242 NB: this is CLEAR for default key. */
244 char name[1]; /* key name follows here */
245 } __attribute__((__packed__));
247 struct ntreg_sk_record {
248 int32_t seg_len; /* length (always -ve because used) */
249 char id[2]; /* "sk" */
251 uint32_t sk_next; /* linked into a circular list */
253 uint32_t refcount; /* reference count */
254 uint32_t sec_len; /* length of security info */
255 char sec_desc[1]; /* security info follows */
256 } __attribute__((__packed__));
259 header_checksum (const hive_h *h)
261 uint32_t *daddr = (uint32_t *) h->addr;
265 for (i = 0; i < 0x1fc / 4; ++i) {
266 sum ^= le32toh (*daddr);
273 #define HIVEX_OPEN_MSGLVL_MASK (HIVEX_OPEN_VERBOSE|HIVEX_OPEN_DEBUG)
276 hivex_open (const char *filename, int flags)
280 assert (sizeof (struct ntreg_header) == 0x1000);
281 assert (offsetof (struct ntreg_header, csum) == 0x1fc);
283 h = calloc (1, sizeof *h);
287 h->msglvl = flags & HIVEX_OPEN_MSGLVL_MASK;
289 const char *debug = getenv ("HIVEX_DEBUG");
290 if (debug && STREQ (debug, "1"))
294 fprintf (stderr, "hivex_open: created handle %p\n", h);
296 h->writable = !!(flags & HIVEX_OPEN_WRITE);
297 h->filename = strdup (filename);
298 if (h->filename == NULL)
301 h->fd = open (filename, O_RDONLY | O_CLOEXEC);
306 if (fstat (h->fd, &statbuf) == -1)
309 h->size = statbuf.st_size;
312 h->addr = mmap (NULL, h->size, PROT_READ, MAP_SHARED, h->fd, 0);
313 if (h->addr == MAP_FAILED)
317 fprintf (stderr, "hivex_open: mapped file at %p\n", h->addr);
319 h->addr = malloc (h->size);
323 if (full_read (h->fd, h->addr, h->size) < h->size)
328 if (h->hdr->magic[0] != 'r' ||
329 h->hdr->magic[1] != 'e' ||
330 h->hdr->magic[2] != 'g' ||
331 h->hdr->magic[3] != 'f') {
332 fprintf (stderr, "hivex: %s: not a Windows NT Registry hive file\n",
338 /* Check major version. */
339 uint32_t major_ver = le32toh (h->hdr->major_ver);
340 if (major_ver != 1) {
342 "hivex: %s: hive file major version %" PRIu32 " (expected 1)\n",
343 filename, major_ver);
348 h->bitmap = calloc (1 + h->size / 32, 1);
349 if (h->bitmap == NULL)
352 /* Header checksum. */
353 uint32_t sum = header_checksum (h);
354 if (sum != le32toh (h->hdr->csum)) {
355 fprintf (stderr, "hivex: %s: bad checksum in hive header\n", filename);
360 if (h->msglvl >= 2) {
361 char *name = windows_utf16_to_utf8 (h->hdr->name, 64);
364 "hivex_open: header fields:\n"
365 " file version %" PRIu32 ".%" PRIu32 "\n"
366 " sequence nos %" PRIu32 " %" PRIu32 "\n"
367 " (sequences nos should match if hive was synched at shutdown)\n"
368 " original file name %s\n"
369 " (only 32 chars are stored, name is probably truncated)\n"
370 " root offset 0x%x + 0x1000\n"
371 " end of last page 0x%x + 0x1000 (total file size 0x%zx)\n"
372 " checksum 0x%x (calculated 0x%x)\n",
373 major_ver, le32toh (h->hdr->minor_ver),
374 le32toh (h->hdr->sequence1), le32toh (h->hdr->sequence2),
375 name ? name : "(conversion failed)",
376 le32toh (h->hdr->offset),
377 le32toh (h->hdr->blocks), h->size,
378 le32toh (h->hdr->csum), sum);
382 h->rootoffs = le32toh (h->hdr->offset) + 0x1000;
383 h->endpages = le32toh (h->hdr->blocks) + 0x1000;
386 fprintf (stderr, "hivex_open: root offset = 0x%zx\n", h->rootoffs);
388 /* We'll set this flag when we see a block with the root offset (ie.
391 int seen_root_block = 0, bad_root_block = 0;
393 /* Collect some stats. */
394 size_t pages = 0; /* Number of hbin pages read. */
395 size_t smallest_page = SIZE_MAX, largest_page = 0;
396 size_t blocks = 0; /* Total number of blocks found. */
397 size_t smallest_block = SIZE_MAX, largest_block = 0, blocks_bytes = 0;
398 size_t used_blocks = 0; /* Total number of used blocks found. */
399 size_t used_size = 0; /* Total size (bytes) of used blocks. */
401 /* Read the pages and blocks. The aim here is to be robust against
402 * corrupt or malicious registries. So we make sure the loops
403 * always make forward progress. We add the address of each block
404 * we read to a hash table so pointers will only reference the start
408 struct ntreg_hbin_page *page;
409 for (off = 0x1000; off < h->size; off += le32toh (page->page_size)) {
410 if (off >= h->endpages)
413 page = (struct ntreg_hbin_page *) (h->addr + off);
414 if (page->magic[0] != 'h' ||
415 page->magic[1] != 'b' ||
416 page->magic[2] != 'i' ||
417 page->magic[3] != 'n') {
418 fprintf (stderr, "hivex: %s: trailing garbage at end of file (at 0x%zx, after %zu pages)\n",
419 filename, off, pages);
424 size_t page_size = le32toh (page->page_size);
426 fprintf (stderr, "hivex_open: page at 0x%zx, size %zu\n", off, page_size);
428 if (page_size < smallest_page) smallest_page = page_size;
429 if (page_size > largest_page) largest_page = page_size;
431 if (page_size <= sizeof (struct ntreg_hbin_page) ||
432 (page_size & 0x0fff) != 0) {
433 fprintf (stderr, "hivex: %s: page size %zu at 0x%zx, bad registry\n",
434 filename, page_size, off);
439 /* Read the blocks in this page. */
441 struct ntreg_hbin_block *block;
443 for (blkoff = off + 0x20;
444 blkoff < off + page_size;
448 int is_root = blkoff == h->rootoffs;
452 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
454 seg_len = block_len (h, blkoff, &used);
455 if (seg_len <= 4 || (seg_len & 3) != 0) {
456 fprintf (stderr, "hivex: %s: block size %" PRIu32 " at 0x%zx, bad registry\n",
457 filename, le32toh (block->seg_len), blkoff);
463 fprintf (stderr, "hivex_open: %s block id %d,%d at 0x%zx size %zu%s\n",
464 used ? "used" : "free", block->id[0], block->id[1], blkoff,
465 seg_len, is_root ? " (root)" : "");
467 blocks_bytes += seg_len;
468 if (seg_len < smallest_block) smallest_block = seg_len;
469 if (seg_len > largest_block) largest_block = seg_len;
471 if (is_root && !used)
476 used_size += seg_len;
478 /* Root block must be an nk-block. */
479 if (is_root && (block->id[0] != 'n' || block->id[1] != 'k'))
482 /* Note this blkoff is a valid address. */
483 BITMAP_SET (h->bitmap, blkoff);
488 if (!seen_root_block) {
489 fprintf (stderr, "hivex: %s: no root block found\n", filename);
494 if (bad_root_block) {
495 fprintf (stderr, "hivex: %s: bad root block (free or not nk)\n", filename);
502 "hivex_open: successfully read Windows Registry hive file:\n"
503 " pages: %zu [sml: %zu, lge: %zu]\n"
504 " blocks: %zu [sml: %zu, avg: %zu, lge: %zu]\n"
505 " blocks used: %zu\n"
506 " bytes used: %zu\n",
507 pages, smallest_page, largest_page,
508 blocks, smallest_block, blocks_bytes / blocks, largest_block,
509 used_blocks, used_size);
517 if (h->addr && h->size && h->addr != MAP_FAILED) {
519 munmap (h->addr, h->size);
533 hivex_close (hive_h *h)
538 fprintf (stderr, "hivex_close\n");
542 munmap (h->addr, h->size);
552 /*----------------------------------------------------------------------
557 hivex_root (hive_h *h)
559 hive_node_h ret = h->rootoffs;
560 if (!IS_VALID_BLOCK (h, ret)) {
568 hivex_node_name (hive_h *h, hive_node_h node)
570 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
575 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
577 /* AFAIK the node name is always plain ASCII, so no conversion
578 * to UTF-8 is necessary. However we do need to nul-terminate
582 /* nk->name_len is unsigned, 16 bit, so this is safe ... However
583 * we have to make sure the length doesn't exceed the block length.
585 size_t len = le16toh (nk->name_len);
586 size_t seg_len = block_len (h, node, NULL);
587 if (sizeof (struct ntreg_nk_record) + len - 1 > seg_len) {
589 fprintf (stderr, "hivex_node_name: returning EFAULT because node name is too long (%zu, %zu)\n",
595 char *ret = malloc (len + 1);
598 memcpy (ret, nk->name, len);
604 /* I think the documentation for the sk and classname fields in the nk
605 * record is wrong, or else the offset field is in the wrong place.
606 * Otherwise this makes no sense. Disabled this for now -- it's not
607 * useful for reading the registry anyway.
611 hivex_node_security (hive_h *h, hive_node_h node)
613 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
618 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
620 hive_node_h ret = le32toh (nk->sk);
622 if (!IS_VALID_BLOCK (h, ret)) {
630 hivex_node_classname (hive_h *h, hive_node_h node)
632 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
637 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
639 hive_node_h ret = le32toh (nk->classname);
641 if (!IS_VALID_BLOCK (h, ret)) {
649 /* Structure for returning 0-terminated lists of offsets (nodes,
659 init_offset_list (struct offset_list *list)
663 list->offsets = NULL;
666 #define INIT_OFFSET_LIST(name) \
667 struct offset_list name; \
668 init_offset_list (&name)
670 /* Preallocates the offset_list, but doesn't make the contents longer. */
672 grow_offset_list (struct offset_list *list, size_t alloc)
674 assert (alloc >= list->len);
675 size_t *p = realloc (list->offsets, alloc * sizeof (size_t));
684 add_to_offset_list (struct offset_list *list, size_t offset)
686 if (list->len >= list->alloc) {
687 if (grow_offset_list (list, list->alloc ? list->alloc * 2 : 4) == -1)
690 list->offsets[list->len] = offset;
696 free_offset_list (struct offset_list *list)
698 free (list->offsets);
702 return_offset_list (struct offset_list *list)
704 if (add_to_offset_list (list, 0) == -1)
706 return list->offsets; /* caller frees */
709 /* Iterate over children, returning child nodes and intermediate blocks. */
710 #define GET_CHILDREN_NO_CHECK_NK 1
713 get_children (hive_h *h, hive_node_h node,
714 hive_node_h **children_ret, size_t **blocks_ret,
717 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
722 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
724 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
726 INIT_OFFSET_LIST (children);
727 INIT_OFFSET_LIST (blocks);
729 /* Deal with the common "no subkeys" case quickly. */
730 if (nr_subkeys_in_nk == 0)
733 /* Arbitrarily limit the number of subkeys we will ever deal with. */
734 if (nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS) {
736 fprintf (stderr, "hivex: get_children: returning ERANGE because nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS (%zu > %d)\n",
737 nr_subkeys_in_nk, HIVEX_MAX_SUBKEYS);
742 /* Preallocate space for the children. */
743 if (grow_offset_list (&children, nr_subkeys_in_nk) == -1)
746 /* The subkey_lf field can point either to an lf-record, which is
747 * the common case, or if there are lots of subkeys, to an
750 size_t subkey_lf = le32toh (nk->subkey_lf);
752 if (!IS_VALID_BLOCK (h, subkey_lf)) {
754 fprintf (stderr, "hivex_node_children: returning EFAULT 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, nr_subkeys_in_lf = %zu\n",
779 nr_subkeys_in_nk, nr_subkeys_in_lf);
781 if (nr_subkeys_in_nk != nr_subkeys_in_lf) {
786 size_t len = block_len (h, subkey_lf, NULL);
787 if (8 + nr_subkeys_in_lf * 8 > len) {
789 fprintf (stderr, "hivex_node_children: returning EFAULT because too many subkeys (%zu, %zu)\n",
790 nr_subkeys_in_lf, len);
796 for (i = 0; i < nr_subkeys_in_lf; ++i) {
797 hive_node_h subkey = le32toh (lf->keys[i].offset);
799 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
800 if (!IS_VALID_BLOCK (h, subkey)) {
802 fprintf (stderr, "hivex_node_children: returning EFAULT because subkey is not a valid block (0x%zx)\n",
808 if (add_to_offset_list (&children, subkey) == -1)
813 /* Points to ri-record? */
814 else if (block->id[0] == 'r' && block->id[1] == 'i') {
815 struct ntreg_ri_record *ri = (struct ntreg_ri_record *) block;
817 size_t nr_offsets = le16toh (ri->nr_offsets);
819 /* Count total number of children. */
821 for (i = 0; i < nr_offsets; ++i) {
822 hive_node_h offset = le32toh (ri->offset[i]);
824 if (!IS_VALID_BLOCK (h, offset)) {
826 fprintf (stderr, "hivex_node_children: returning EFAULT because ri-offset is not a valid block (0x%zx)\n",
831 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
833 fprintf (stderr, "get_children: returning ENOTSUP because ri-record offset does not point to lf/lh (0x%zx)\n",
839 if (add_to_offset_list (&blocks, offset) == -1)
842 struct ntreg_lf_record *lf =
843 (struct ntreg_lf_record *) (h->addr + offset);
845 count += le16toh (lf->nr_keys);
849 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu, counted = %zu\n",
850 nr_subkeys_in_nk, count);
852 if (nr_subkeys_in_nk != count) {
857 /* Copy list of children. Note nr_subkeys_in_nk is limited to
858 * something reasonable above.
860 for (i = 0; i < nr_offsets; ++i) {
861 hive_node_h offset = le32toh (ri->offset[i]);
863 if (!IS_VALID_BLOCK (h, offset)) {
865 fprintf (stderr, "hivex_node_children: returning EFAULT because ri-offset is not a valid block (0x%zx)\n",
870 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
872 fprintf (stderr, "get_children: returning ENOTSUP because ri-record offset does not point to lf/lh (0x%zx)\n",
878 struct ntreg_lf_record *lf =
879 (struct ntreg_lf_record *) (h->addr + offset);
882 for (j = 0; j < le16toh (lf->nr_keys); ++j) {
883 hive_node_h subkey = le32toh (lf->keys[j].offset);
885 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
886 if (!IS_VALID_BLOCK (h, subkey)) {
888 fprintf (stderr, "hivex_node_children: returning EFAULT because indirect subkey is not a valid block (0x%zx)\n",
894 if (add_to_offset_list (&children, subkey) == -1)
900 /* else not supported, set errno and fall through */
902 fprintf (stderr, "get_children: returning ENOTSUP because subkey block is not lf/lh/ri (0x%zx, %d, %d)\n",
903 subkey_lf, block->id[0], block->id[1]);
906 free_offset_list (&children);
907 free_offset_list (&blocks);
911 *children_ret = return_offset_list (&children);
912 *blocks_ret = return_offset_list (&blocks);
913 if (!*children_ret || !*blocks_ret)
919 hivex_node_children (hive_h *h, hive_node_h node)
921 hive_node_h *children;
924 if (get_children (h, node, &children, &blocks, 0) == -1)
931 /* Very inefficient, but at least having a separate API call
932 * allows us to make it more efficient in future.
935 hivex_node_get_child (hive_h *h, hive_node_h node, const char *nname)
937 hive_node_h *children = NULL;
941 children = hivex_node_children (h, node);
942 if (!children) goto error;
945 for (i = 0; children[i] != 0; ++i) {
946 name = hivex_node_name (h, children[i]);
947 if (!name) goto error;
948 if (STRCASEEQ (name, nname)) {
952 free (name); name = NULL;
962 hivex_node_parent (hive_h *h, hive_node_h node)
964 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
969 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
971 hive_node_h ret = le32toh (nk->parent);
973 if (!IS_VALID_BLOCK (h, ret)) {
975 fprintf (stderr, "hivex_node_parent: returning EFAULT because parent is not a valid block (0x%zx)\n",
984 get_values (hive_h *h, hive_node_h node,
985 hive_value_h **values_ret, size_t **blocks_ret)
987 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
992 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
994 size_t nr_values = le32toh (nk->nr_values);
997 fprintf (stderr, "hivex_node_values: nr_values = %zu\n", nr_values);
999 INIT_OFFSET_LIST (values);
1000 INIT_OFFSET_LIST (blocks);
1002 /* Deal with the common "no values" case quickly. */
1006 /* Arbitrarily limit the number of values we will ever deal with. */
1007 if (nr_values > HIVEX_MAX_VALUES) {
1009 fprintf (stderr, "hivex: get_values: returning ERANGE because nr_values > HIVEX_MAX_VALUES (%zu > %d)\n",
1010 nr_values, HIVEX_MAX_VALUES);
1015 /* Preallocate space for the values. */
1016 if (grow_offset_list (&values, nr_values) == -1)
1019 /* Get the value list and check it looks reasonable. */
1020 size_t vlist_offset = le32toh (nk->vallist);
1021 vlist_offset += 0x1000;
1022 if (!IS_VALID_BLOCK (h, vlist_offset)) {
1024 fprintf (stderr, "hivex_node_values: returning EFAULT because value list is not a valid block (0x%zx)\n",
1030 if (add_to_offset_list (&blocks, vlist_offset) == -1)
1033 struct ntreg_value_list *vlist =
1034 (struct ntreg_value_list *) (h->addr + vlist_offset);
1036 size_t len = block_len (h, vlist_offset, NULL);
1037 if (4 + nr_values * 4 > len) {
1039 fprintf (stderr, "hivex_node_values: returning EFAULT because value list is too long (%zu, %zu)\n",
1046 for (i = 0; i < nr_values; ++i) {
1047 hive_node_h value = vlist->offset[i];
1049 if (!IS_VALID_BLOCK (h, value)) {
1051 fprintf (stderr, "hivex_node_values: returning EFAULT because value is not a valid block (0x%zx)\n",
1056 if (add_to_offset_list (&values, value) == -1)
1061 *values_ret = return_offset_list (&values);
1062 *blocks_ret = return_offset_list (&blocks);
1063 if (!*values_ret || !*blocks_ret)
1068 free_offset_list (&values);
1069 free_offset_list (&blocks);
1074 hivex_node_values (hive_h *h, hive_node_h node)
1076 hive_value_h *values;
1079 if (get_values (h, node, &values, &blocks) == -1)
1086 /* Very inefficient, but at least having a separate API call
1087 * allows us to make it more efficient in future.
1090 hivex_node_get_value (hive_h *h, hive_node_h node, const char *key)
1092 hive_value_h *values = NULL;
1094 hive_value_h ret = 0;
1096 values = hivex_node_values (h, node);
1097 if (!values) goto error;
1100 for (i = 0; values[i] != 0; ++i) {
1101 name = hivex_value_key (h, values[i]);
1102 if (!name) goto error;
1103 if (STRCASEEQ (name, key)) {
1107 free (name); name = NULL;
1117 hivex_value_key (hive_h *h, hive_value_h value)
1119 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1124 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1126 /* AFAIK the key is always plain ASCII, so no conversion to UTF-8 is
1127 * necessary. However we do need to nul-terminate the string.
1130 /* vk->name_len is unsigned, 16 bit, so this is safe ... However
1131 * we have to make sure the length doesn't exceed the block length.
1133 size_t len = le16toh (vk->name_len);
1134 size_t seg_len = block_len (h, value, NULL);
1135 if (sizeof (struct ntreg_vk_record) + len - 1 > seg_len) {
1137 fprintf (stderr, "hivex_value_key: returning EFAULT because key length is too long (%zu, %zu)\n",
1143 char *ret = malloc (len + 1);
1146 memcpy (ret, vk->name, len);
1152 hivex_value_type (hive_h *h, hive_value_h value, hive_type *t, size_t *len)
1154 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1159 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1162 *t = le32toh (vk->data_type);
1165 *len = le32toh (vk->data_len);
1166 *len &= 0x7fffffff; /* top bit indicates if data is stored inline */
1173 hivex_value_value (hive_h *h, hive_value_h value,
1174 hive_type *t_rtn, size_t *len_rtn)
1176 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1181 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1187 t = le32toh (vk->data_type);
1189 len = le32toh (vk->data_len);
1190 is_inline = !!(len & 0x80000000);
1194 fprintf (stderr, "hivex_value_value: value=0x%zx, t=%d, len=%zu, inline=%d\n",
1195 value, t, len, is_inline);
1202 if (is_inline && len > 4) {
1207 /* Arbitrarily limit the length that we will read. */
1208 if (len > HIVEX_MAX_VALUE_LEN) {
1210 fprintf (stderr, "hivex_value_value: returning ERANGE because data length > HIVEX_MAX_VALUE_LEN (%zu > %d)\n",
1211 len, HIVEX_MAX_SUBKEYS);
1216 char *ret = malloc (len);
1221 memcpy (ret, (char *) &vk->data_offset, len);
1225 size_t data_offset = le32toh (vk->data_offset);
1226 data_offset += 0x1000;
1227 if (!IS_VALID_BLOCK (h, data_offset)) {
1229 fprintf (stderr, "hivex_value_value: returning EFAULT because data offset is not a valid block (0x%zx)\n",
1236 /* Check that the declared size isn't larger than the block its in.
1238 * XXX Some apparently valid registries are seen to have this,
1239 * so turn this into a warning and substitute the smaller length
1242 size_t blen = block_len (h, data_offset, NULL);
1243 if (len > blen - 4 /* subtract 4 for block header */) {
1245 fprintf (stderr, "hivex_value_value: warning: declared data length is longer than the block it is in (data 0x%zx, data len %zu, block len %zu)\n",
1246 data_offset, len, blen);
1249 /* Return the smaller length to the caller too. */
1254 char *data = h->addr + data_offset + 4;
1255 memcpy (ret, data, len);
1260 windows_utf16_to_utf8 (/* const */ char *input, size_t len)
1262 iconv_t ic = iconv_open ("UTF-8", "UTF-16");
1263 if (ic == (iconv_t) -1)
1266 /* iconv(3) has an insane interface ... */
1268 /* Mostly UTF-8 will be smaller, so this is a good initial guess. */
1269 size_t outalloc = len;
1273 size_t outlen = outalloc;
1274 char *out = malloc (outlen + 1);
1284 size_t r = iconv (ic, &inp, &inlen, &outp, &outlen);
1285 if (r == (size_t) -1) {
1286 if (errno == E2BIG) {
1288 size_t prev = outalloc;
1289 /* Try again with a larger output buffer. */
1292 if (outalloc < prev) {
1300 /* Else some conversion failure, eg. EILSEQ, EINVAL. */
1316 hivex_value_string (hive_h *h, hive_value_h value)
1320 char *data = hivex_value_value (h, value, &t, &len);
1325 if (t != hive_t_string && t != hive_t_expand_string && t != hive_t_link) {
1331 /* Deal with the case where Windows has allocated a large buffer
1332 * full of random junk, and only the first few bytes of the buffer
1333 * contain a genuine UTF-16 string.
1335 * In this case, iconv would try to process the junk bytes as UTF-16
1336 * and inevitably find an illegal sequence (EILSEQ). Instead, stop
1337 * after we find the first \0\0.
1339 * (Found by Hilko Bengen in a fresh Windows XP SOFTWARE hive).
1341 size_t slen = utf16_string_len_in_bytes_max (data, len);
1345 char *ret = windows_utf16_to_utf8 (data, len);
1354 free_strings (char **argv)
1359 for (i = 0; argv[i] != NULL; ++i)
1365 /* Get the length of a UTF-16 format string. Handle the string as
1366 * pairs of bytes, looking for the first \0\0 pair.
1369 utf16_string_len_in_bytes (const char *str)
1373 while (str[0] || str[1]) {
1381 /* As for utf16_string_len_in_bytes but only read up to a maximum length. */
1383 utf16_string_len_in_bytes_max (const char *str, size_t len)
1387 while (len > 0 && (str[0] || str[1])) {
1396 /* http://blogs.msdn.com/oldnewthing/archive/2009/10/08/9904646.aspx */
1398 hivex_value_multiple_strings (hive_h *h, hive_value_h value)
1402 char *data = hivex_value_value (h, value, &t, &len);
1407 if (t != hive_t_multiple_strings) {
1413 size_t nr_strings = 0;
1414 char **ret = malloc ((1 + nr_strings) * sizeof (char *));
1424 while (p < data + len && (plen = utf16_string_len_in_bytes (p)) > 0) {
1426 char **ret2 = realloc (ret, (1 + nr_strings) * sizeof (char *));
1434 ret[nr_strings-1] = windows_utf16_to_utf8 (p, plen);
1435 ret[nr_strings] = NULL;
1436 if (ret[nr_strings-1] == NULL) {
1442 p += plen + 2 /* skip over UTF-16 \0\0 at the end of this string */;
1450 hivex_value_dword (hive_h *h, hive_value_h value)
1454 char *data = hivex_value_value (h, value, &t, &len);
1459 if ((t != hive_t_dword && t != hive_t_dword_be) || len != 4) {
1465 int32_t ret = *(int32_t*)data;
1467 if (t == hive_t_dword) /* little endian */
1468 ret = le32toh (ret);
1470 ret = be32toh (ret);
1476 hivex_value_qword (hive_h *h, hive_value_h value)
1480 char *data = hivex_value_value (h, value, &t, &len);
1485 if (t != hive_t_qword || len != 8) {
1491 int64_t ret = *(int64_t*)data;
1493 ret = le64toh (ret); /* always little endian */
1498 /*----------------------------------------------------------------------
1503 hivex_visit (hive_h *h, const struct hivex_visitor *visitor, size_t len,
1504 void *opaque, int flags)
1506 return hivex_visit_node (h, hivex_root (h), visitor, len, opaque, flags);
1509 static int hivex__visit_node (hive_h *h, hive_node_h node, const struct hivex_visitor *vtor, char *unvisited, void *opaque, int flags);
1512 hivex_visit_node (hive_h *h, hive_node_h node,
1513 const struct hivex_visitor *visitor, size_t len, void *opaque,
1516 struct hivex_visitor vtor;
1517 memset (&vtor, 0, sizeof vtor);
1519 /* Note that len might be larger *or smaller* than the expected size. */
1520 size_t copysize = len <= sizeof vtor ? len : sizeof vtor;
1521 memcpy (&vtor, visitor, copysize);
1523 /* This bitmap records unvisited nodes, so we don't loop if the
1524 * registry contains cycles.
1526 char *unvisited = malloc (1 + h->size / 32);
1527 if (unvisited == NULL)
1529 memcpy (unvisited, h->bitmap, 1 + h->size / 32);
1531 int r = hivex__visit_node (h, node, &vtor, unvisited, opaque, flags);
1537 hivex__visit_node (hive_h *h, hive_node_h node,
1538 const struct hivex_visitor *vtor, char *unvisited,
1539 void *opaque, int flags)
1541 int skip_bad = flags & HIVEX_VISIT_SKIP_BAD;
1543 hive_value_h *values = NULL;
1544 hive_node_h *children = NULL;
1550 /* Return -1 on all callback errors. However on internal errors,
1551 * check if skip_bad is set and suppress those errors if so.
1555 if (!BITMAP_TST (unvisited, node)) {
1557 fprintf (stderr, "hivex__visit_node: contains cycle: visited node 0x%zx already\n",
1561 return skip_bad ? 0 : -1;
1563 BITMAP_CLR (unvisited, node);
1565 name = hivex_node_name (h, node);
1566 if (!name) return skip_bad ? 0 : -1;
1567 if (vtor->node_start && vtor->node_start (h, opaque, node, name) == -1)
1570 values = hivex_node_values (h, node);
1572 ret = skip_bad ? 0 : -1;
1576 for (i = 0; values[i] != 0; ++i) {
1580 if (hivex_value_type (h, values[i], &t, &len) == -1) {
1581 ret = skip_bad ? 0 : -1;
1585 key = hivex_value_key (h, values[i]);
1587 ret = skip_bad ? 0 : -1;
1591 if (vtor->value_any) {
1592 str = hivex_value_value (h, values[i], &t, &len);
1594 ret = skip_bad ? 0 : -1;
1597 if (vtor->value_any (h, opaque, node, values[i], t, len, key, str) == -1)
1599 free (str); str = NULL;
1604 str = hivex_value_value (h, values[i], &t, &len);
1606 ret = skip_bad ? 0 : -1;
1609 if (t != hive_t_none) {
1610 ret = skip_bad ? 0 : -1;
1613 if (vtor->value_none &&
1614 vtor->value_none (h, opaque, node, values[i], t, len, key, str) == -1)
1616 free (str); str = NULL;
1620 case hive_t_expand_string:
1622 str = hivex_value_string (h, values[i]);
1624 if (errno != EILSEQ && errno != EINVAL) {
1625 ret = skip_bad ? 0 : -1;
1628 if (vtor->value_string_invalid_utf16) {
1629 str = hivex_value_value (h, values[i], &t, &len);
1630 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i], t, len, key, str) == -1)
1632 free (str); str = NULL;
1636 if (vtor->value_string &&
1637 vtor->value_string (h, opaque, node, values[i], t, len, key, str) == -1)
1639 free (str); str = NULL;
1643 case hive_t_dword_be: {
1644 int32_t i32 = hivex_value_dword (h, values[i]);
1645 if (vtor->value_dword &&
1646 vtor->value_dword (h, opaque, node, values[i], t, len, key, i32) == -1)
1651 case hive_t_qword: {
1652 int64_t i64 = hivex_value_qword (h, values[i]);
1653 if (vtor->value_qword &&
1654 vtor->value_qword (h, opaque, node, values[i], t, len, key, i64) == -1)
1660 str = hivex_value_value (h, values[i], &t, &len);
1662 ret = skip_bad ? 0 : -1;
1665 if (t != hive_t_binary) {
1666 ret = skip_bad ? 0 : -1;
1669 if (vtor->value_binary &&
1670 vtor->value_binary (h, opaque, node, values[i], t, len, key, str) == -1)
1672 free (str); str = NULL;
1675 case hive_t_multiple_strings:
1676 strs = hivex_value_multiple_strings (h, values[i]);
1678 if (errno != EILSEQ && errno != EINVAL) {
1679 ret = skip_bad ? 0 : -1;
1682 if (vtor->value_string_invalid_utf16) {
1683 str = hivex_value_value (h, values[i], &t, &len);
1684 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i], t, len, key, str) == -1)
1686 free (str); str = NULL;
1690 if (vtor->value_multiple_strings &&
1691 vtor->value_multiple_strings (h, opaque, node, values[i], t, len, key, strs) == -1)
1693 free_strings (strs); strs = NULL;
1696 case hive_t_resource_list:
1697 case hive_t_full_resource_description:
1698 case hive_t_resource_requirements_list:
1700 str = hivex_value_value (h, values[i], &t, &len);
1702 ret = skip_bad ? 0 : -1;
1705 if (vtor->value_other &&
1706 vtor->value_other (h, opaque, node, values[i], t, len, key, str) == -1)
1708 free (str); str = NULL;
1713 free (key); key = NULL;
1716 children = hivex_node_children (h, node);
1717 if (children == NULL) {
1718 ret = skip_bad ? 0 : -1;
1722 for (i = 0; children[i] != 0; ++i) {
1724 fprintf (stderr, "hivex__visit_node: %s: visiting subkey %d (0x%zx)\n",
1725 name, i, children[i]);
1727 if (hivex__visit_node (h, children[i], vtor, unvisited, opaque, flags) == -1)
1731 if (vtor->node_end && vtor->node_end (h, opaque, node, name) == -1)
1742 free_strings (strs);
1746 /*----------------------------------------------------------------------
1750 /* Allocate an hbin (page), extending the malloc'd space if necessary,
1751 * and updating the hive handle fields (but NOT the hive disk header
1752 * -- the hive disk header is updated when we commit). This function
1753 * also extends the bitmap if necessary.
1755 * 'allocation_hint' is the size of the block allocation we would like
1756 * to make. Normally registry blocks are very small (avg 50 bytes)
1757 * and are contained in standard-sized pages (4KB), but the registry
1758 * can support blocks which are larger than a standard page, in which
1759 * case it creates a page of 8KB, 12KB etc.
1762 * > 0 : offset of first usable byte of new page (after page header)
1763 * 0 : error (errno set)
1766 allocate_page (hive_h *h, size_t allocation_hint)
1768 /* In almost all cases this will be 1. */
1769 size_t nr_4k_pages =
1770 1 + (allocation_hint + sizeof (struct ntreg_hbin_page) - 1) / 4096;
1771 assert (nr_4k_pages >= 1);
1773 /* 'extend' is the number of bytes to extend the file by. Note that
1774 * hives found in the wild often contain slack between 'endpages'
1775 * and the actual end of the file, so we don't always need to make
1778 ssize_t extend = h->endpages + nr_4k_pages * 4096 - h->size;
1780 if (h->msglvl >= 2) {
1781 fprintf (stderr, "allocate_page: current endpages = 0x%zx, current size = 0x%zx\n",
1782 h->endpages, h->size);
1783 fprintf (stderr, "allocate_page: extending file by %zd bytes (<= 0 if no extension)\n",
1788 size_t oldsize = h->size;
1789 size_t newsize = h->size + extend;
1790 char *newaddr = realloc (h->addr, newsize);
1791 if (newaddr == NULL)
1794 size_t oldbitmapsize = 1 + oldsize / 32;
1795 size_t newbitmapsize = 1 + newsize / 32;
1796 char *newbitmap = realloc (h->bitmap, newbitmapsize);
1797 if (newbitmap == NULL) {
1804 h->bitmap = newbitmap;
1806 memset (h->addr + oldsize, 0, newsize - oldsize);
1807 memset (h->bitmap + oldbitmapsize, 0, newbitmapsize - oldbitmapsize);
1810 size_t offset = h->endpages;
1811 h->endpages += nr_4k_pages * 4096;
1814 fprintf (stderr, "allocate_page: new endpages = 0x%zx, new size = 0x%zx\n",
1815 h->endpages, h->size);
1817 /* Write the hbin header. */
1818 struct ntreg_hbin_page *page =
1819 (struct ntreg_hbin_page *) (h->addr + offset);
1820 page->magic[0] = 'h';
1821 page->magic[1] = 'b';
1822 page->magic[2] = 'i';
1823 page->magic[3] = 'n';
1824 page->offset_first = htole32 (offset - 0x1000);
1825 page->page_size = htole32 (nr_4k_pages * 4096);
1826 memset (page->unknown, 0, sizeof (page->unknown));
1829 fprintf (stderr, "allocate_page: new page at 0x%zx\n", offset);
1831 /* Offset of first usable byte after the header. */
1832 return offset + sizeof (struct ntreg_hbin_page);
1835 /* Allocate a single block, first allocating an hbin (page) at the end
1836 * of the current file if necessary. NB. To keep the implementation
1837 * simple and more likely to be correct, we do not reuse existing free
1840 * seg_len is the size of the block (this INCLUDES the block header).
1841 * The header of the block is initialized to -seg_len (negative to
1842 * indicate used). id[2] is the block ID (type), eg. "nk" for nk-
1843 * record. The block bitmap is updated to show this block as valid.
1844 * The rest of the contents of the block will be zero.
1846 * **NB** Because allocate_block may reallocate the memory, all
1847 * pointers into the memory become potentially invalid. I really
1848 * love writing in C, can't you tell?
1851 * > 0 : offset of new block
1852 * 0 : error (errno set)
1855 allocate_block (hive_h *h, size_t seg_len, const char id[2])
1863 /* The caller probably forgot to include the header. Note that
1864 * value lists have no ID field, so seg_len == 4 would be possible
1865 * for them, albeit unusual.
1868 fprintf (stderr, "allocate_block: refusing too small allocation (%zu), returning ERANGE\n",
1874 /* Refuse really large allocations. */
1875 if (seg_len > HIVEX_MAX_ALLOCATION) {
1877 fprintf (stderr, "allocate_block: refusing large allocation (%zu), returning ERANGE\n",
1883 /* Round up allocation to multiple of 8 bytes. All blocks must be
1884 * on an 8 byte boundary.
1886 seg_len = (seg_len + 7) & ~7;
1888 /* Allocate a new page if necessary. */
1889 if (h->endblocks == 0 || h->endblocks + seg_len > h->endpages) {
1890 size_t newendblocks = allocate_page (h, seg_len);
1891 if (newendblocks == 0)
1893 h->endblocks = newendblocks;
1896 size_t offset = h->endblocks;
1899 fprintf (stderr, "allocate_block: new block at 0x%zx, size %zu\n",
1902 struct ntreg_hbin_block *blockhdr =
1903 (struct ntreg_hbin_block *) (h->addr + offset);
1905 memset (blockhdr, 0, seg_len);
1907 blockhdr->seg_len = htole32 (- (int32_t) seg_len);
1908 if (id[0] && id[1] && seg_len >= sizeof (struct ntreg_hbin_block)) {
1909 blockhdr->id[0] = id[0];
1910 blockhdr->id[1] = id[1];
1913 BITMAP_SET (h->bitmap, offset);
1915 h->endblocks += seg_len;
1917 /* If there is space after the last block in the last page, then we
1918 * have to put a dummy free block header here to mark the rest of
1921 ssize_t rem = h->endpages - h->endblocks;
1924 fprintf (stderr, "allocate_block: marking remainder of page free starting at 0x%zx, size %zd\n",
1929 blockhdr = (struct ntreg_hbin_block *) (h->addr + h->endblocks);
1930 blockhdr->seg_len = htole32 ((int32_t) rem);
1936 /* 'offset' must point to a valid, used block. This function marks
1937 * the block unused (by updating the seg_len field) and invalidates
1938 * the bitmap. It does NOT do this recursively, so to avoid creating
1939 * unreachable used blocks, callers may have to recurse over the hive
1940 * structures. Also callers must ensure there are no references to
1941 * this block from other parts of the hive.
1944 mark_block_unused (hive_h *h, size_t offset)
1946 assert (h->writable);
1947 assert (IS_VALID_BLOCK (h, offset));
1950 fprintf (stderr, "mark_block_unused: marking 0x%zx unused\n", offset);
1952 struct ntreg_hbin_block *blockhdr =
1953 (struct ntreg_hbin_block *) (h->addr + offset);
1955 size_t seg_len = block_len (h, offset, NULL);
1956 blockhdr->seg_len = htole32 (seg_len);
1958 BITMAP_CLR (h->bitmap, offset);
1961 /* Delete all existing values at this node. */
1963 delete_values (hive_h *h, hive_node_h node)
1965 assert (h->writable);
1967 hive_value_h *values;
1969 if (get_values (h, node, &values, &blocks) == -1)
1973 for (i = 0; blocks[i] != 0; ++i)
1974 mark_block_unused (h, blocks[i]);
1978 for (i = 0; values[i] != 0; ++i) {
1979 struct ntreg_vk_record *vk =
1980 (struct ntreg_vk_record *) (h->addr + values[i]);
1984 len = le32toh (vk->data_len);
1985 is_inline = !!(len & 0x80000000); /* top bit indicates is inline */
1988 if (!is_inline) { /* non-inline, so remove data block */
1989 size_t data_offset = le32toh (vk->data_offset);
1990 data_offset += 0x1000;
1991 mark_block_unused (h, data_offset);
1994 /* remove vk record */
1995 mark_block_unused (h, values[i]);
2000 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2001 nk->nr_values = htole32 (0);
2002 nk->vallist = htole32 (0xffffffff);
2008 hivex_commit (hive_h *h, const char *filename, int flags)
2020 filename = filename ? : h->filename;
2021 int fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
2025 /* Update the header fields. */
2026 uint32_t sequence = le32toh (h->hdr->sequence1);
2028 h->hdr->sequence1 = htole32 (sequence);
2029 h->hdr->sequence2 = htole32 (sequence);
2030 /* XXX Ought to update h->hdr->last_modified. */
2031 h->hdr->blocks = htole32 (h->endpages - 0x1000);
2033 /* Recompute header checksum. */
2034 uint32_t sum = header_checksum (h);
2035 h->hdr->csum = htole32 (sum);
2038 fprintf (stderr, "hivex_commit: new header checksum: 0x%x\n", sum);
2040 if (full_write (fd, h->addr, h->size) != h->size) {
2047 if (close (fd) == -1)
2053 /* Calculate the hash for a lf or lh record offset.
2056 calc_hash (const char *type, const char *name, char *ret)
2058 size_t len = strlen (name);
2060 if (STRPREFIX (type, "lf"))
2061 /* Old-style, not used in current registries. */
2062 memcpy (ret, name, len < 4 ? len : 4);
2064 /* New-style for lh-records. */
2067 for (i = 0; i < len; ++i) {
2068 c = c_toupper (name[i]);
2072 *((uint32_t *) ret) = htole32 (h);
2076 /* Create a completely new lh-record containing just the single node. */
2078 new_lh_record (hive_h *h, const char *name, hive_node_h node)
2080 static const char id[2] = { 'l', 'h' };
2081 size_t seg_len = sizeof (struct ntreg_lf_record);
2082 size_t offset = allocate_block (h, seg_len, id);
2086 struct ntreg_lf_record *lh = (struct ntreg_lf_record *) (h->addr + offset);
2087 lh->nr_keys = htole16 (1);
2088 lh->keys[0].offset = htole32 (node - 0x1000);
2089 calc_hash ("lh", name, lh->keys[0].hash);
2094 /* Insert node into existing lf/lh-record at position.
2095 * This allocates a new record and marks the old one as unused.
2098 insert_lf_record (hive_h *h, size_t old_offs, size_t posn,
2099 const char *name, hive_node_h node)
2101 assert (IS_VALID_BLOCK (h, old_offs));
2103 /* Work around C stupidity.
2104 * http://www.redhat.com/archives/libguestfs/2010-February/msg00056.html
2106 int test = BLOCK_ID_EQ (h, old_offs, "lf") || BLOCK_ID_EQ (h, old_offs, "lh");
2109 struct ntreg_lf_record *old_lf =
2110 (struct ntreg_lf_record *) (h->addr + old_offs);
2111 size_t nr_keys = le16toh (old_lf->nr_keys);
2113 nr_keys++; /* in new record ... */
2115 size_t seg_len = sizeof (struct ntreg_lf_record) + (nr_keys-1) * 8;
2117 /* Copy the old_lf->id in case it moves during allocate_block. */
2119 memcpy (id, old_lf->id, sizeof id);
2121 size_t new_offs = allocate_block (h, seg_len, id);
2125 /* old_lf could have been invalidated by allocate_block. */
2126 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2128 struct ntreg_lf_record *new_lf =
2129 (struct ntreg_lf_record *) (h->addr + new_offs);
2130 new_lf->nr_keys = htole16 (nr_keys);
2132 /* Copy the keys until we reach posn, insert the new key there, then
2133 * copy the remaining keys.
2136 for (i = 0; i < posn; ++i)
2137 new_lf->keys[i] = old_lf->keys[i];
2139 new_lf->keys[i].offset = htole32 (node - 0x1000);
2140 calc_hash (new_lf->id, name, new_lf->keys[i].hash);
2142 for (i = posn+1; i < nr_keys; ++i)
2143 new_lf->keys[i] = old_lf->keys[i-1];
2145 /* Old block is unused, return new block. */
2146 mark_block_unused (h, old_offs);
2150 /* Compare name with name in nk-record. */
2152 compare_name_with_nk_name (hive_h *h, const char *name, hive_node_h nk_offs)
2154 assert (IS_VALID_BLOCK (h, nk_offs));
2155 assert (BLOCK_ID_EQ (h, nk_offs, "nk"));
2157 /* Name in nk is not necessarily nul-terminated. */
2158 char *nname = hivex_node_name (h, nk_offs);
2160 /* Unfortunately we don't have a way to return errors here. */
2162 perror ("compare_name_with_nk_name");
2166 int r = strcasecmp (name, nname);
2173 hivex_node_add_child (hive_h *h, hive_node_h parent, const char *name)
2180 if (!IS_VALID_BLOCK (h, parent) || !BLOCK_ID_EQ (h, parent, "nk")) {
2185 if (name == NULL || strlen (name) == 0) {
2190 if (hivex_node_get_child (h, parent, name) != 0) {
2195 /* Create the new nk-record. */
2196 static const char nk_id[2] = { 'n', 'k' };
2197 size_t seg_len = sizeof (struct ntreg_nk_record) + strlen (name);
2198 hive_node_h node = allocate_block (h, seg_len, nk_id);
2203 fprintf (stderr, "hivex_node_add_child: allocated new nk-record for child at 0x%zx\n", node);
2205 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2206 nk->flags = htole16 (0x0020); /* key is ASCII. */
2207 nk->parent = htole32 (parent - 0x1000);
2208 nk->subkey_lf = htole32 (0xffffffff);
2209 nk->subkey_lf_volatile = htole32 (0xffffffff);
2210 nk->vallist = htole32 (0xffffffff);
2211 nk->classname = htole32 (0xffffffff);
2212 nk->name_len = htole16 (strlen (name));
2213 strcpy (nk->name, name);
2215 /* Inherit parent sk. */
2216 struct ntreg_nk_record *parent_nk =
2217 (struct ntreg_nk_record *) (h->addr + parent);
2218 size_t parent_sk_offset = le32toh (parent_nk->sk);
2219 parent_sk_offset += 0x1000;
2220 if (!IS_VALID_BLOCK (h, parent_sk_offset) ||
2221 !BLOCK_ID_EQ (h, parent_sk_offset, "sk")) {
2223 fprintf (stderr, "hivex_node_add_child: returning EFAULT because parent sk is not a valid block (%zu)\n",
2228 struct ntreg_sk_record *sk =
2229 (struct ntreg_sk_record *) (h->addr + parent_sk_offset);
2230 sk->refcount = htole32 (le32toh (sk->refcount) + 1);
2231 nk->sk = htole32 (parent_sk_offset - 0x1000);
2233 /* Inherit parent timestamp. */
2234 memcpy (nk->timestamp, parent_nk->timestamp, sizeof (parent_nk->timestamp));
2236 /* What I found out the hard way (not documented anywhere): the
2237 * subkeys in lh-records must be kept sorted. If you just add a
2238 * subkey in a non-sorted position (eg. just add it at the end) then
2239 * Windows won't see the subkey _and_ Windows will corrupt the hive
2240 * itself when it modifies or saves it.
2242 * So use get_children() to get a list of intermediate
2243 * lf/lh-records. get_children() returns these in reading order
2244 * (which is sorted), so we look for the lf/lh-records in sequence
2245 * until we find the key name just after the one we are inserting,
2246 * and we insert the subkey just before it.
2248 * The only other case is the no-subkeys case, where we have to
2249 * create a brand new lh-record.
2251 hive_node_h *unused;
2254 if (get_children (h, parent, &unused, &blocks, 0) == -1)
2259 size_t nr_subkeys_in_parent_nk = le32toh (parent_nk->nr_subkeys);
2260 if (nr_subkeys_in_parent_nk == 0) { /* No subkeys case. */
2261 /* Free up any existing intermediate blocks. */
2262 for (i = 0; blocks[i] != 0; ++i)
2263 mark_block_unused (h, blocks[i]);
2264 size_t lh_offs = new_lh_record (h, name, node);
2270 /* Recalculate pointers that could have been invalidated by
2271 * previous call to allocate_block (via new_lh_record).
2273 nk = (struct ntreg_nk_record *) (h->addr + node);
2274 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2277 fprintf (stderr, "hivex_node_add_child: no keys, allocated new lh-record at 0x%zx\n", lh_offs);
2279 parent_nk->subkey_lf = htole32 (lh_offs - 0x1000);
2281 else { /* Insert subkeys case. */
2282 size_t old_offs = 0, new_offs = 0;
2283 struct ntreg_lf_record *old_lf = NULL;
2285 /* Find lf/lh key name just after the one we are inserting. */
2286 for (i = 0; blocks[i] != 0; ++i) {
2287 if (BLOCK_ID_EQ (h, blocks[i], "lf") ||
2288 BLOCK_ID_EQ (h, blocks[i], "lh")) {
2289 old_offs = blocks[i];
2290 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2291 for (j = 0; j < le16toh (old_lf->nr_keys); ++j) {
2292 hive_node_h nk_offs = le32toh (old_lf->keys[j].offset);
2294 if (compare_name_with_nk_name (h, name, nk_offs) < 0)
2300 /* Insert it at the end.
2301 * old_offs points to the last lf record, set j.
2303 assert (old_offs != 0); /* should never happen if nr_subkeys > 0 */
2304 j = le16toh (old_lf->nr_keys);
2309 fprintf (stderr, "hivex_node_add_child: insert key in existing lh-record at 0x%zx, posn %zu\n", old_offs, j);
2311 new_offs = insert_lf_record (h, old_offs, j, name, node);
2312 if (new_offs == 0) {
2317 /* Recalculate pointers that could have been invalidated by
2318 * previous call to allocate_block (via insert_lf_record).
2320 nk = (struct ntreg_nk_record *) (h->addr + node);
2321 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2324 fprintf (stderr, "hivex_node_add_child: new lh-record at 0x%zx\n",
2327 /* If the lf/lh-record was directly referenced by the parent nk,
2328 * then update the parent nk.
2330 if (le32toh (parent_nk->subkey_lf) + 0x1000 == old_offs)
2331 parent_nk->subkey_lf = htole32 (new_offs - 0x1000);
2332 /* Else we have to look for the intermediate ri-record and update
2336 for (i = 0; blocks[i] != 0; ++i) {
2337 if (BLOCK_ID_EQ (h, blocks[i], "ri")) {
2338 struct ntreg_ri_record *ri =
2339 (struct ntreg_ri_record *) (h->addr + blocks[i]);
2340 for (j = 0; j < le16toh (ri->nr_offsets); ++j)
2341 if (le32toh (ri->offset[j] + 0x1000) == old_offs) {
2342 ri->offset[j] = htole32 (new_offs - 0x1000);
2348 /* Not found .. This is an internal error. */
2350 fprintf (stderr, "hivex_node_add_child: returning ENOTSUP because could not find ri->lf link\n");
2362 /* Update nr_subkeys in parent nk. */
2363 nr_subkeys_in_parent_nk++;
2364 parent_nk->nr_subkeys = htole32 (nr_subkeys_in_parent_nk);
2366 /* Update max_subkey_name_len in parent nk. */
2367 uint16_t max = le16toh (parent_nk->max_subkey_name_len);
2368 if (max < strlen (name) * 2) /* *2 because "recoded" in UTF16-LE. */
2369 parent_nk->max_subkey_name_len = htole16 (strlen (name) * 2);
2374 /* Decrement the refcount of an sk-record, and if it reaches zero,
2375 * unlink it from the chain and delete it.
2378 delete_sk (hive_h *h, size_t sk_offset)
2380 if (!IS_VALID_BLOCK (h, sk_offset) || !BLOCK_ID_EQ (h, sk_offset, "sk")) {
2382 fprintf (stderr, "delete_sk: not an sk record: 0x%zx\n", sk_offset);
2387 struct ntreg_sk_record *sk = (struct ntreg_sk_record *) (h->addr + sk_offset);
2389 if (sk->refcount == 0) {
2391 fprintf (stderr, "delete_sk: sk record already has refcount 0: 0x%zx\n",
2399 if (sk->refcount == 0) {
2400 size_t sk_prev_offset = sk->sk_prev;
2401 sk_prev_offset += 0x1000;
2403 size_t sk_next_offset = sk->sk_next;
2404 sk_next_offset += 0x1000;
2406 /* Update sk_prev/sk_next SKs, unless they both point back to this
2407 * cell in which case we are deleting the last SK.
2409 if (sk_prev_offset != sk_offset && sk_next_offset != sk_offset) {
2410 struct ntreg_sk_record *sk_prev =
2411 (struct ntreg_sk_record *) (h->addr + sk_prev_offset);
2412 struct ntreg_sk_record *sk_next =
2413 (struct ntreg_sk_record *) (h->addr + sk_next_offset);
2415 sk_prev->sk_next = htole32 (sk_next_offset - 0x1000);
2416 sk_next->sk_prev = htole32 (sk_prev_offset - 0x1000);
2419 /* Refcount is zero so really delete this block. */
2420 mark_block_unused (h, sk_offset);
2426 /* Callback from hivex_node_delete_child which is called to delete a
2427 * node AFTER its subnodes have been visited. The subnodes have been
2428 * deleted but we still have to delete any lf/lh/li/ri records and the
2429 * value list block and values, followed by deleting the node itself.
2432 delete_node (hive_h *h, void *opaque, hive_node_h node, const char *name)
2434 /* Get the intermediate blocks. The subkeys have already been
2435 * deleted by this point, so tell get_children() not to check for
2436 * validity of the nk-records.
2438 hive_node_h *unused;
2440 if (get_children (h, node, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK) == -1)
2444 /* We don't care what's in these intermediate blocks, so we can just
2445 * delete them unconditionally.
2448 for (i = 0; blocks[i] != 0; ++i)
2449 mark_block_unused (h, blocks[i]);
2453 /* Delete the values in the node. */
2454 if (delete_values (h, node) == -1)
2457 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2459 /* If the NK references an SK, delete it. */
2460 size_t sk_offs = le32toh (nk->sk);
2461 if (sk_offs != 0xffffffff) {
2463 if (delete_sk (h, sk_offs) == -1)
2465 nk->sk = htole32 (0xffffffff);
2468 /* If the NK references a classname, delete it. */
2469 size_t cl_offs = le32toh (nk->classname);
2470 if (cl_offs != 0xffffffff) {
2472 mark_block_unused (h, cl_offs);
2473 nk->classname = htole32 (0xffffffff);
2476 /* Delete the node itself. */
2477 mark_block_unused (h, node);
2483 hivex_node_delete_child (hive_h *h, hive_node_h node)
2490 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2495 if (node == hivex_root (h)) {
2497 fprintf (stderr, "hivex_node_delete_child: cannot delete root node\n");
2502 hive_node_h parent = hivex_node_parent (h, node);
2506 /* Delete node and all its children and values recursively. */
2507 static const struct hivex_visitor visitor = { .node_end = delete_node };
2508 if (hivex_visit_node (h, node, &visitor, sizeof visitor, NULL, 0) == -1)
2511 /* Delete the link from parent to child. We need to find the lf/lh
2512 * record which contains the offset and remove the offset from that
2513 * record, then decrement the element count in that record, and
2514 * decrement the overall number of subkeys stored in the parent
2517 hive_node_h *unused;
2519 if (get_children (h, parent, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK)== -1)
2524 for (i = 0; blocks[i] != 0; ++i) {
2525 struct ntreg_hbin_block *block =
2526 (struct ntreg_hbin_block *) (h->addr + blocks[i]);
2528 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
2529 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
2531 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
2533 for (j = 0; j < nr_subkeys_in_lf; ++j)
2534 if (le32toh (lf->keys[j].offset) + 0x1000 == node) {
2535 for (; j < nr_subkeys_in_lf - 1; ++j)
2536 memcpy (&lf->keys[j], &lf->keys[j+1], sizeof (lf->keys[j]));
2537 lf->nr_keys = htole16 (nr_subkeys_in_lf - 1);
2543 fprintf (stderr, "hivex_node_delete_child: could not find parent to child link\n");
2548 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + parent);
2549 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
2550 nk->nr_subkeys = htole32 (nr_subkeys_in_nk - 1);
2553 fprintf (stderr, "hivex_node_delete_child: updating nr_subkeys in parent 0x%zx to %zu\n",
2554 parent, nr_subkeys_in_nk);
2560 hivex_node_set_values (hive_h *h, hive_node_h node,
2561 size_t nr_values, const hive_set_value *values,
2569 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2574 /* Delete all existing values. */
2575 if (delete_values (h, node) == -1)
2581 /* Allocate value list node. Value lists have no id field. */
2582 static const char nul_id[2] = { 0, 0 };
2584 sizeof (struct ntreg_value_list) + (nr_values - 1) * sizeof (uint32_t);
2585 size_t vallist_offs = allocate_block (h, seg_len, nul_id);
2586 if (vallist_offs == 0)
2589 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2590 nk->nr_values = htole32 (nr_values);
2591 nk->vallist = htole32 (vallist_offs - 0x1000);
2593 struct ntreg_value_list *vallist =
2594 (struct ntreg_value_list *) (h->addr + vallist_offs);
2597 for (i = 0; i < nr_values; ++i) {
2598 /* Allocate vk record to store this (key, value) pair. */
2599 static const char vk_id[2] = { 'v', 'k' };
2600 seg_len = sizeof (struct ntreg_vk_record) + strlen (values[i].key);
2601 size_t vk_offs = allocate_block (h, seg_len, vk_id);
2605 /* Recalculate pointers that could have been invalidated by
2606 * previous call to allocate_block.
2608 nk = (struct ntreg_nk_record *) (h->addr + node);
2609 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2611 vallist->offset[i] = htole32 (vk_offs - 0x1000);
2613 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2614 size_t name_len = strlen (values[i].key);
2615 vk->name_len = htole16 (name_len);
2616 strcpy (vk->name, values[i].key);
2617 vk->data_type = htole32 (values[i].t);
2618 uint32_t len = values[i].len;
2619 if (len <= 4) /* store it inline => set MSB flag */
2621 vk->data_len = htole32 (len);
2622 vk->flags = name_len == 0 ? 0 : 1;
2624 if (values[i].len <= 4) /* store it inline */
2625 memcpy (&vk->data_offset, values[i].value, values[i].len);
2627 size_t offs = allocate_block (h, values[i].len + 4, nul_id);
2631 /* Recalculate pointers that could have been invalidated by
2632 * previous call to allocate_block.
2634 nk = (struct ntreg_nk_record *) (h->addr + node);
2635 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2636 vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2638 memcpy (h->addr + offs + 4, values[i].value, values[i].len);
2639 vk->data_offset = htole32 (offs - 0x1000);
2642 if (name_len * 2 > le32toh (nk->max_vk_name_len))
2643 /* * 2 for UTF16-LE "reencoding" */
2644 nk->max_vk_name_len = htole32 (name_len * 2);
2645 if (values[i].len > le32toh (nk->max_vk_data_len))
2646 nk->max_vk_data_len = htole32 (values[i].len);
2653 hivex_node_set_value (hive_h *h, hive_node_h node,
2654 const hive_set_value *val, int flags)
2656 hive_value_h *prev_values = hivex_node_values (h, node);
2657 if (prev_values == NULL)
2662 size_t nr_values = 0;
2663 for (hive_value_h *itr = prev_values; *itr != 0; ++itr)
2666 hive_set_value *values = malloc ((nr_values + 1) * (sizeof (hive_set_value)));
2668 goto leave_prev_values;
2671 int idx_of_val = -1;
2672 hive_value_h *prev_val;
2673 for (prev_val = prev_values; *prev_val != 0; ++prev_val) {
2677 hive_set_value *value = &values[prev_val - prev_values];
2679 char *valval = hivex_value_value (h, *prev_val, &t, &len);
2680 if (valval == NULL) goto leave_partial;
2683 value->value = valval;
2687 char *valkey = hivex_value_key (h, *prev_val);
2688 if (valkey == NULL) goto leave_partial;
2691 value->key = valkey;
2693 if (STRCASEEQ (valkey, val->key))
2694 idx_of_val = prev_val - prev_values;
2697 if (idx_of_val > -1) {
2698 free (values[idx_of_val].key);
2699 free (values[idx_of_val].value);
2701 idx_of_val = nr_values;
2705 hive_set_value *value = &values[idx_of_val];
2706 *value = (hive_set_value){
2707 .key = strdup (val->key),
2708 .value = malloc (val->len),
2713 if (value->key == NULL || value->value == NULL) goto leave_partial;
2714 memcpy (value->value, val->value, val->len);
2716 retval = hivex_node_set_values (h, node, nr_values, values, 0);
2719 for (int i = 0; i < alloc_ct; i += 2) {
2720 if (values[i / 2].value != NULL)
2721 free (values[i / 2].value);
2722 if (i + 1 < alloc_ct && values[i / 2].key != NULL)
2723 free (values[i / 2].key);