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 10000
60 #define HIVEX_MAX_VALUES 1000
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);
73 /* Registry file, memory mapped if read-only, or malloc'd if writing. */
76 struct ntreg_header *hdr;
79 /* Use a bitmap to store which file offsets are valid (point to a
80 * used block). We only need to store 1 bit per 32 bits of the file
81 * (because blocks are 4-byte aligned). We found that the average
82 * block size in a registry file is ~50 bytes. So roughly 1 in 12
83 * bits in the bitmap will be set, making it likely a more efficient
84 * structure than a hash table.
87 #define BITMAP_SET(bitmap,off) (bitmap[(off)>>5] |= 1 << (((off)>>2)&7))
88 #define BITMAP_CLR(bitmap,off) (bitmap[(off)>>5] &= ~ (1 << (((off)>>2)&7)))
89 #define BITMAP_TST(bitmap,off) (bitmap[(off)>>5] & (1 << (((off)>>2)&7)))
90 #define IS_VALID_BLOCK(h,off) \
91 (((off) & 3) == 0 && \
93 (off) < (h)->size && \
94 BITMAP_TST((h)->bitmap,(off)))
96 /* Fields from the header, extracted from little-endianness hell. */
97 size_t rootoffs; /* Root key offset (always an nk-block). */
98 size_t endpages; /* Offset of end of pages. */
101 size_t endblocks; /* Offset to next block allocation (0
102 if not allocated anything yet). */
105 /* NB. All fields are little endian. */
106 struct ntreg_header {
107 char magic[4]; /* "regf" */
110 char last_modified[8];
111 uint32_t major_ver; /* 1 */
112 uint32_t minor_ver; /* 3 */
113 uint32_t unknown5; /* 0 */
114 uint32_t unknown6; /* 1 */
115 uint32_t offset; /* offset of root key record - 4KB */
116 uint32_t blocks; /* pointer AFTER last hbin in file - 4KB */
117 uint32_t unknown7; /* 1 */
119 char name[64]; /* original file name of hive */
120 char unknown_guid1[16];
121 char unknown_guid2[16];
124 char unknown_guid3[16];
129 uint32_t csum; /* checksum: xor of dwords 0-0x1fb. */
131 char unknown11[3528];
133 char unknown_guid4[16];
134 char unknown_guid5[16];
135 char unknown_guid6[16];
139 } __attribute__((__packed__));
141 struct ntreg_hbin_page {
142 char magic[4]; /* "hbin" */
143 uint32_t offset_first; /* offset from 1st block */
144 uint32_t page_size; /* size of this page (multiple of 4KB) */
146 /* Linked list of blocks follows here. */
147 } __attribute__((__packed__));
149 struct ntreg_hbin_block {
150 int32_t seg_len; /* length of this block (-ve for used block) */
151 char id[2]; /* the block type (eg. "nk" for nk record) */
152 /* Block data follows here. */
153 } __attribute__((__packed__));
155 #define BLOCK_ID_EQ(h,offs,eqid) \
156 (STREQLEN (((struct ntreg_hbin_block *)((h)->addr + (offs)))->id, (eqid), 2))
159 block_len (hive_h *h, size_t blkoff, int *used)
161 struct ntreg_hbin_block *block;
162 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
164 int32_t len = le32toh (block->seg_len);
175 struct ntreg_nk_record {
176 int32_t seg_len; /* length (always -ve because used) */
177 char id[2]; /* "nk" */
181 uint32_t parent; /* offset of owner/parent */
182 uint32_t nr_subkeys; /* number of subkeys */
183 uint32_t nr_subkeys_volatile;
184 uint32_t subkey_lf; /* lf record containing list of subkeys */
185 uint32_t subkey_lf_volatile;
186 uint32_t nr_values; /* number of values */
187 uint32_t vallist; /* value-list record */
188 uint32_t sk; /* offset of sk-record */
189 uint32_t classname; /* offset of classname record */
190 uint16_t max_subkey_name_len; /* maximum length of a subkey name in bytes
191 if the subkey was reencoded as UTF-16LE */
194 uint32_t max_vk_name_len; /* maximum length of any vk name in bytes
195 if the name was reencoded as UTF-16LE */
196 uint32_t max_vk_data_len; /* maximum length of any vk data in bytes */
198 uint16_t name_len; /* length of name */
199 uint16_t classname_len; /* length of classname */
200 char name[1]; /* name follows here */
201 } __attribute__((__packed__));
203 struct ntreg_lf_record {
205 char id[2]; /* "lf"|"lh" */
206 uint16_t nr_keys; /* number of keys in this record */
208 uint32_t offset; /* offset of nk-record for this subkey */
209 char hash[4]; /* hash of subkey name */
211 } __attribute__((__packed__));
213 struct ntreg_ri_record {
215 char id[2]; /* "ri" */
216 uint16_t nr_offsets; /* number of pointers to lh records */
217 uint32_t offset[1]; /* list of pointers to lh records */
218 } __attribute__((__packed__));
220 /* This has no ID header. */
221 struct ntreg_value_list {
223 uint32_t offset[1]; /* list of pointers to vk records */
224 } __attribute__((__packed__));
226 struct ntreg_vk_record {
227 int32_t seg_len; /* length (always -ve because used) */
228 char id[2]; /* "vk" */
229 uint16_t name_len; /* length of name */
230 /* length of the data:
231 * If data_len is <= 4, then it's stored inline.
232 * If data_len is 0x80000000, then it's an inline dword.
233 * Top bit may be set or not set at random.
236 uint32_t data_offset; /* pointer to the data (or data if inline) */
237 uint32_t data_type; /* type of the data */
238 uint16_t flags; /* bit 0 set => key name ASCII,
239 bit 0 clr => key name UTF-16.
240 Only seen ASCII here in the wild.
241 NB: this is CLEAR for default key. */
243 char name[1]; /* key name follows here */
244 } __attribute__((__packed__));
246 struct ntreg_sk_record {
247 int32_t seg_len; /* length (always -ve because used) */
248 char id[2]; /* "sk" */
250 uint32_t sk_next; /* linked into a circular list */
252 uint32_t refcount; /* reference count */
253 uint32_t sec_len; /* length of security info */
254 char sec_desc[1]; /* security info follows */
255 } __attribute__((__packed__));
258 header_checksum (const hive_h *h)
260 uint32_t *daddr = (uint32_t *) h->addr;
264 for (i = 0; i < 0x1fc / 4; ++i) {
265 sum ^= le32toh (*daddr);
273 hivex_open (const char *filename, int flags)
277 assert (sizeof (struct ntreg_header) == 0x1000);
278 assert (offsetof (struct ntreg_header, csum) == 0x1fc);
280 h = calloc (1, sizeof *h);
284 h->msglvl = flags & HIVEX_OPEN_MSGLVL_MASK;
286 const char *debug = getenv ("HIVEX_DEBUG");
287 if (debug && STREQ (debug, "1"))
291 fprintf (stderr, "hivex_open: created handle %p\n", h);
293 h->writable = !!(flags & HIVEX_OPEN_WRITE);
294 h->filename = strdup (filename);
295 if (h->filename == NULL)
298 h->fd = open (filename, O_RDONLY | O_CLOEXEC);
303 if (fstat (h->fd, &statbuf) == -1)
306 h->size = statbuf.st_size;
309 h->addr = mmap (NULL, h->size, PROT_READ, MAP_SHARED, h->fd, 0);
310 if (h->addr == MAP_FAILED)
314 fprintf (stderr, "hivex_open: mapped file at %p\n", h->addr);
316 h->addr = malloc (h->size);
320 if (full_read (h->fd, h->addr, h->size) < h->size)
325 if (h->hdr->magic[0] != 'r' ||
326 h->hdr->magic[1] != 'e' ||
327 h->hdr->magic[2] != 'g' ||
328 h->hdr->magic[3] != 'f') {
329 fprintf (stderr, "hivex: %s: not a Windows NT Registry hive file\n",
335 /* Check major version. */
336 uint32_t major_ver = le32toh (h->hdr->major_ver);
337 if (major_ver != 1) {
339 "hivex: %s: hive file major version %" PRIu32 " (expected 1)\n",
340 filename, major_ver);
345 h->bitmap = calloc (1 + h->size / 32, 1);
346 if (h->bitmap == NULL)
349 /* Header checksum. */
350 uint32_t sum = header_checksum (h);
351 if (sum != le32toh (h->hdr->csum)) {
352 fprintf (stderr, "hivex: %s: bad checksum in hive header\n", filename);
357 if (h->msglvl >= 2) {
358 char *name = windows_utf16_to_utf8 (h->hdr->name, 64);
361 "hivex_open: header fields:\n"
362 " file version %" PRIu32 ".%" PRIu32 "\n"
363 " sequence nos %" PRIu32 " %" PRIu32 "\n"
364 " (sequences nos should match if hive was synched at shutdown)\n"
365 " original file name %s\n"
366 " (only 32 chars are stored, name is probably truncated)\n"
367 " root offset 0x%x + 0x1000\n"
368 " end of last page 0x%x + 0x1000 (total file size 0x%zx)\n"
369 " checksum 0x%x (calculated 0x%x)\n",
370 major_ver, le32toh (h->hdr->minor_ver),
371 le32toh (h->hdr->sequence1), le32toh (h->hdr->sequence2),
372 name ? name : "(conversion failed)",
373 le32toh (h->hdr->offset),
374 le32toh (h->hdr->blocks), h->size,
375 le32toh (h->hdr->csum), sum);
379 h->rootoffs = le32toh (h->hdr->offset) + 0x1000;
380 h->endpages = le32toh (h->hdr->blocks) + 0x1000;
383 fprintf (stderr, "hivex_open: root offset = 0x%zx\n", h->rootoffs);
385 /* We'll set this flag when we see a block with the root offset (ie.
388 int seen_root_block = 0, bad_root_block = 0;
390 /* Collect some stats. */
391 size_t pages = 0; /* Number of hbin pages read. */
392 size_t smallest_page = SIZE_MAX, largest_page = 0;
393 size_t blocks = 0; /* Total number of blocks found. */
394 size_t smallest_block = SIZE_MAX, largest_block = 0, blocks_bytes = 0;
395 size_t used_blocks = 0; /* Total number of used blocks found. */
396 size_t used_size = 0; /* Total size (bytes) of used blocks. */
398 /* Read the pages and blocks. The aim here is to be robust against
399 * corrupt or malicious registries. So we make sure the loops
400 * always make forward progress. We add the address of each block
401 * we read to a hash table so pointers will only reference the start
405 struct ntreg_hbin_page *page;
406 for (off = 0x1000; off < h->size; off += le32toh (page->page_size)) {
407 if (off >= h->endpages)
410 page = (struct ntreg_hbin_page *) (h->addr + off);
411 if (page->magic[0] != 'h' ||
412 page->magic[1] != 'b' ||
413 page->magic[2] != 'i' ||
414 page->magic[3] != 'n') {
415 fprintf (stderr, "hivex: %s: trailing garbage at end of file (at 0x%zx, after %zu pages)\n",
416 filename, off, pages);
421 size_t page_size = le32toh (page->page_size);
423 fprintf (stderr, "hivex_open: page at 0x%zx, size %zu\n", off, page_size);
425 if (page_size < smallest_page) smallest_page = page_size;
426 if (page_size > largest_page) largest_page = page_size;
428 if (page_size <= sizeof (struct ntreg_hbin_page) ||
429 (page_size & 0x0fff) != 0) {
430 fprintf (stderr, "hivex: %s: page size %zu at 0x%zx, bad registry\n",
431 filename, page_size, off);
436 /* Read the blocks in this page. */
438 struct ntreg_hbin_block *block;
440 for (blkoff = off + 0x20;
441 blkoff < off + page_size;
445 int is_root = blkoff == h->rootoffs;
449 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
451 seg_len = block_len (h, blkoff, &used);
452 if (seg_len <= 4 || (seg_len & 3) != 0) {
453 fprintf (stderr, "hivex: %s: block size %" PRIu32 " at 0x%zx, bad registry\n",
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)
536 munmap (h->addr, h->size);
546 /*----------------------------------------------------------------------
551 hivex_root (hive_h *h)
553 hive_node_h ret = h->rootoffs;
554 if (!IS_VALID_BLOCK (h, ret)) {
562 hivex_node_name (hive_h *h, hive_node_h node)
564 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
569 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
571 /* AFAIK the node name is always plain ASCII, so no conversion
572 * to UTF-8 is necessary. However we do need to nul-terminate
576 /* nk->name_len is unsigned, 16 bit, so this is safe ... However
577 * we have to make sure the length doesn't exceed the block length.
579 size_t len = le16toh (nk->name_len);
580 size_t seg_len = block_len (h, node, NULL);
581 if (sizeof (struct ntreg_nk_record) + len - 1 > seg_len) {
583 fprintf (stderr, "hivex_node_name: returning EFAULT because node name is too long (%zu, %zu)\n",
589 char *ret = malloc (len + 1);
592 memcpy (ret, nk->name, len);
598 /* I think the documentation for the sk and classname fields in the nk
599 * record is wrong, or else the offset field is in the wrong place.
600 * Otherwise this makes no sense. Disabled this for now -- it's not
601 * useful for reading the registry anyway.
605 hivex_node_security (hive_h *h, hive_node_h node)
607 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
612 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
614 hive_node_h ret = le32toh (nk->sk);
616 if (!IS_VALID_BLOCK (h, ret)) {
624 hivex_node_classname (hive_h *h, hive_node_h node)
626 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
631 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
633 hive_node_h ret = le32toh (nk->classname);
635 if (!IS_VALID_BLOCK (h, ret)) {
643 /* Structure for returning 0-terminated lists of offsets (nodes,
653 init_offset_list (struct offset_list *list)
657 list->offsets = NULL;
660 #define INIT_OFFSET_LIST(name) \
661 struct offset_list name; \
662 init_offset_list (&name)
664 /* Preallocates the offset_list, but doesn't make the contents longer. */
666 grow_offset_list (struct offset_list *list, size_t alloc)
668 assert (alloc >= list->len);
669 size_t *p = realloc (list->offsets, alloc * sizeof (size_t));
678 add_to_offset_list (struct offset_list *list, size_t offset)
680 if (list->len >= list->alloc) {
681 if (grow_offset_list (list, list->alloc ? list->alloc * 2 : 4) == -1)
684 list->offsets[list->len] = offset;
690 free_offset_list (struct offset_list *list)
692 free (list->offsets);
696 return_offset_list (struct offset_list *list)
698 if (add_to_offset_list (list, 0) == -1)
700 return list->offsets; /* caller frees */
703 /* Iterate over children, returning child nodes and intermediate blocks. */
704 #define GET_CHILDREN_NO_CHECK_NK 1
707 get_children (hive_h *h, hive_node_h node,
708 hive_node_h **children_ret, size_t **blocks_ret,
711 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
716 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
718 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
720 INIT_OFFSET_LIST (children);
721 INIT_OFFSET_LIST (blocks);
723 /* Deal with the common "no subkeys" case quickly. */
724 if (nr_subkeys_in_nk == 0)
727 /* Arbitrarily limit the number of subkeys we will ever deal with. */
728 if (nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS) {
733 /* Preallocate space for the children. */
734 if (grow_offset_list (&children, nr_subkeys_in_nk) == -1)
737 /* The subkey_lf field can point either to an lf-record, which is
738 * the common case, or if there are lots of subkeys, to an
741 size_t subkey_lf = le32toh (nk->subkey_lf);
743 if (!IS_VALID_BLOCK (h, subkey_lf)) {
745 fprintf (stderr, "hivex_node_children: returning EFAULT because subkey_lf is not a valid block (0x%zx)\n",
751 if (add_to_offset_list (&blocks, subkey_lf) == -1)
754 struct ntreg_hbin_block *block =
755 (struct ntreg_hbin_block *) (h->addr + subkey_lf);
757 /* Points to lf-record? (Note, also "lh" but that is basically the
758 * same as "lf" as far as we are concerned here).
760 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
761 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
763 /* Check number of subkeys in the nk-record matches number of subkeys
766 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
769 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu, nr_subkeys_in_lf = %zu\n",
770 nr_subkeys_in_nk, nr_subkeys_in_lf);
772 if (nr_subkeys_in_nk != nr_subkeys_in_lf) {
777 size_t len = block_len (h, subkey_lf, NULL);
778 if (8 + nr_subkeys_in_lf * 8 > len) {
780 fprintf (stderr, "hivex_node_children: returning EFAULT because too many subkeys (%zu, %zu)\n",
781 nr_subkeys_in_lf, len);
787 for (i = 0; i < nr_subkeys_in_lf; ++i) {
788 hive_node_h subkey = le32toh (lf->keys[i].offset);
790 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
791 if (!IS_VALID_BLOCK (h, subkey)) {
793 fprintf (stderr, "hivex_node_children: returning EFAULT because subkey is not a valid block (0x%zx)\n",
799 if (add_to_offset_list (&children, subkey) == -1)
804 /* Points to ri-record? */
805 else if (block->id[0] == 'r' && block->id[1] == 'i') {
806 struct ntreg_ri_record *ri = (struct ntreg_ri_record *) block;
808 size_t nr_offsets = le16toh (ri->nr_offsets);
810 /* Count total number of children. */
812 for (i = 0; i < nr_offsets; ++i) {
813 hive_node_h offset = le32toh (ri->offset[i]);
815 if (!IS_VALID_BLOCK (h, offset)) {
817 fprintf (stderr, "hivex_node_children: returning EFAULT because ri-offset is not a valid block (0x%zx)\n",
822 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
824 fprintf (stderr, "get_children: returning ENOTSUP because ri-record offset does not point to lf/lh (0x%zx)\n",
830 if (add_to_offset_list (&blocks, offset) == -1)
833 struct ntreg_lf_record *lf =
834 (struct ntreg_lf_record *) (h->addr + offset);
836 count += le16toh (lf->nr_keys);
840 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu, counted = %zu\n",
841 nr_subkeys_in_nk, count);
843 if (nr_subkeys_in_nk != count) {
848 /* Copy list of children. Note nr_subkeys_in_nk is limited to
849 * something reasonable above.
851 for (i = 0; i < nr_offsets; ++i) {
852 hive_node_h offset = le32toh (ri->offset[i]);
854 if (!IS_VALID_BLOCK (h, offset)) {
856 fprintf (stderr, "hivex_node_children: returning EFAULT because ri-offset is not a valid block (0x%zx)\n",
861 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
863 fprintf (stderr, "get_children: returning ENOTSUP because ri-record offset does not point to lf/lh (0x%zx)\n",
869 struct ntreg_lf_record *lf =
870 (struct ntreg_lf_record *) (h->addr + offset);
873 for (j = 0; j < le16toh (lf->nr_keys); ++j) {
874 hive_node_h subkey = le32toh (lf->keys[j].offset);
876 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
877 if (!IS_VALID_BLOCK (h, subkey)) {
879 fprintf (stderr, "hivex_node_children: returning EFAULT because indirect subkey is not a valid block (0x%zx)\n",
885 if (add_to_offset_list (&children, subkey) == -1)
891 /* else not supported, set errno and fall through */
893 fprintf (stderr, "get_children: returning ENOTSUP because subkey block is not lf/lh/ri (0x%zx, %d, %d)\n",
894 subkey_lf, block->id[0], block->id[1]);
897 free_offset_list (&children);
898 free_offset_list (&blocks);
902 *children_ret = return_offset_list (&children);
903 *blocks_ret = return_offset_list (&blocks);
904 if (!*children_ret || !*blocks_ret)
910 hivex_node_children (hive_h *h, hive_node_h node)
912 hive_node_h *children;
915 if (get_children (h, node, &children, &blocks, 0) == -1)
922 /* Very inefficient, but at least having a separate API call
923 * allows us to make it more efficient in future.
926 hivex_node_get_child (hive_h *h, hive_node_h node, const char *nname)
928 hive_node_h *children = NULL;
932 children = hivex_node_children (h, node);
933 if (!children) goto error;
936 for (i = 0; children[i] != 0; ++i) {
937 name = hivex_node_name (h, children[i]);
938 if (!name) goto error;
939 if (STRCASEEQ (name, nname)) {
943 free (name); name = NULL;
953 hivex_node_parent (hive_h *h, hive_node_h node)
955 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
960 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
962 hive_node_h ret = le32toh (nk->parent);
964 if (!IS_VALID_BLOCK (h, ret)) {
966 fprintf (stderr, "hivex_node_parent: returning EFAULT because parent is not a valid block (0x%zx)\n",
975 get_values (hive_h *h, hive_node_h node,
976 hive_value_h **values_ret, size_t **blocks_ret)
978 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
983 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
985 size_t nr_values = le32toh (nk->nr_values);
988 fprintf (stderr, "hivex_node_values: nr_values = %zu\n", nr_values);
990 INIT_OFFSET_LIST (values);
991 INIT_OFFSET_LIST (blocks);
993 /* Deal with the common "no values" case quickly. */
997 /* Arbitrarily limit the number of values we will ever deal with. */
998 if (nr_values > HIVEX_MAX_VALUES) {
1003 /* Preallocate space for the values. */
1004 if (grow_offset_list (&values, nr_values) == -1)
1007 /* Get the value list and check it looks reasonable. */
1008 size_t vlist_offset = le32toh (nk->vallist);
1009 vlist_offset += 0x1000;
1010 if (!IS_VALID_BLOCK (h, vlist_offset)) {
1012 fprintf (stderr, "hivex_node_values: returning EFAULT because value list is not a valid block (0x%zx)\n",
1018 if (add_to_offset_list (&blocks, vlist_offset) == -1)
1021 struct ntreg_value_list *vlist =
1022 (struct ntreg_value_list *) (h->addr + vlist_offset);
1024 size_t len = block_len (h, vlist_offset, NULL);
1025 if (4 + nr_values * 4 > len) {
1027 fprintf (stderr, "hivex_node_values: returning EFAULT because value list is too long (%zu, %zu)\n",
1034 for (i = 0; i < nr_values; ++i) {
1035 hive_node_h value = vlist->offset[i];
1037 if (!IS_VALID_BLOCK (h, value)) {
1039 fprintf (stderr, "hivex_node_values: returning EFAULT because value is not a valid block (0x%zx)\n",
1044 if (add_to_offset_list (&values, value) == -1)
1049 *values_ret = return_offset_list (&values);
1050 *blocks_ret = return_offset_list (&blocks);
1051 if (!*values_ret || !*blocks_ret)
1056 free_offset_list (&values);
1057 free_offset_list (&blocks);
1062 hivex_node_values (hive_h *h, hive_node_h node)
1064 hive_value_h *values;
1067 if (get_values (h, node, &values, &blocks) == -1)
1074 /* Very inefficient, but at least having a separate API call
1075 * allows us to make it more efficient in future.
1078 hivex_node_get_value (hive_h *h, hive_node_h node, const char *key)
1080 hive_value_h *values = NULL;
1082 hive_value_h ret = 0;
1084 values = hivex_node_values (h, node);
1085 if (!values) goto error;
1088 for (i = 0; values[i] != 0; ++i) {
1089 name = hivex_value_key (h, values[i]);
1090 if (!name) goto error;
1091 if (STRCASEEQ (name, key)) {
1095 free (name); name = NULL;
1105 hivex_value_key (hive_h *h, hive_value_h value)
1107 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1112 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1114 /* AFAIK the key is always plain ASCII, so no conversion to UTF-8 is
1115 * necessary. However we do need to nul-terminate the string.
1118 /* vk->name_len is unsigned, 16 bit, so this is safe ... However
1119 * we have to make sure the length doesn't exceed the block length.
1121 size_t len = le16toh (vk->name_len);
1122 size_t seg_len = block_len (h, value, NULL);
1123 if (sizeof (struct ntreg_vk_record) + len - 1 > seg_len) {
1125 fprintf (stderr, "hivex_value_key: returning EFAULT because key length is too long (%zu, %zu)\n",
1131 char *ret = malloc (len + 1);
1134 memcpy (ret, vk->name, len);
1140 hivex_value_type (hive_h *h, hive_value_h value, hive_type *t, size_t *len)
1142 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1147 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1150 *t = le32toh (vk->data_type);
1153 *len = le32toh (vk->data_len);
1154 if (*len == 0x80000000) { /* special case */
1156 if (t) *t = hive_t_dword;
1165 hivex_value_value (hive_h *h, hive_value_h value,
1166 hive_type *t_rtn, size_t *len_rtn)
1168 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1173 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1178 t = le32toh (vk->data_type);
1180 len = le32toh (vk->data_len);
1181 if (len == 0x80000000) { /* special case */
1188 fprintf (stderr, "hivex_value_value: value=0x%zx, t=%d, len=%zu\n",
1196 /* Arbitrarily limit the length that we will read. */
1197 if (len > HIVEX_MAX_VALUE_LEN) {
1202 char *ret = malloc (len);
1206 /* If length is <= 4 it's always stored inline. */
1208 memcpy (ret, (char *) &vk->data_offset, len);
1212 size_t data_offset = le32toh (vk->data_offset);
1213 data_offset += 0x1000;
1214 if (!IS_VALID_BLOCK (h, data_offset)) {
1216 fprintf (stderr, "hivex_value_value: returning EFAULT because data offset is not a valid block (0x%zx)\n",
1223 /* Check that the declared size isn't larger than the block its in.
1225 * XXX Some apparently valid registries are seen to have this,
1226 * so turn this into a warning and substitute the smaller length
1229 size_t blen = block_len (h, data_offset, NULL);
1230 if (len > blen - 4 /* subtract 4 for block header */) {
1232 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",
1233 data_offset, len, blen);
1237 char *data = h->addr + data_offset + 4;
1238 memcpy (ret, data, len);
1243 windows_utf16_to_utf8 (/* const */ char *input, size_t len)
1245 iconv_t ic = iconv_open ("UTF-8", "UTF-16");
1246 if (ic == (iconv_t) -1)
1249 /* iconv(3) has an insane interface ... */
1251 /* Mostly UTF-8 will be smaller, so this is a good initial guess. */
1252 size_t outalloc = len;
1256 size_t outlen = outalloc;
1257 char *out = malloc (outlen + 1);
1267 size_t r = iconv (ic, &inp, &inlen, &outp, &outlen);
1268 if (r == (size_t) -1) {
1269 if (errno == E2BIG) {
1270 size_t prev = outalloc;
1271 /* Try again with a larger output buffer. */
1274 if (outalloc < prev)
1279 /* Else some conversion failure, eg. EILSEQ, EINVAL. */
1295 hivex_value_string (hive_h *h, hive_value_h value)
1299 char *data = hivex_value_value (h, value, &t, &len);
1304 if (t != hive_t_string && t != hive_t_expand_string && t != hive_t_link) {
1310 char *ret = windows_utf16_to_utf8 (data, len);
1319 free_strings (char **argv)
1324 for (i = 0; argv[i] != NULL; ++i)
1330 /* Get the length of a UTF-16 format string. Handle the string as
1331 * pairs of bytes, looking for the first \0\0 pair.
1334 utf16_string_len_in_bytes (const char *str)
1338 while (str[0] || str[1]) {
1346 /* http://blogs.msdn.com/oldnewthing/archive/2009/10/08/9904646.aspx */
1348 hivex_value_multiple_strings (hive_h *h, hive_value_h value)
1352 char *data = hivex_value_value (h, value, &t, &len);
1357 if (t != hive_t_multiple_strings) {
1363 size_t nr_strings = 0;
1364 char **ret = malloc ((1 + nr_strings) * sizeof (char *));
1374 while (p < data + len && (plen = utf16_string_len_in_bytes (p)) > 0) {
1376 char **ret2 = realloc (ret, (1 + nr_strings) * sizeof (char *));
1384 ret[nr_strings-1] = windows_utf16_to_utf8 (p, plen);
1385 ret[nr_strings] = NULL;
1386 if (ret[nr_strings-1] == NULL) {
1392 p += plen + 2 /* skip over UTF-16 \0\0 at the end of this string */;
1400 hivex_value_dword (hive_h *h, hive_value_h value)
1404 char *data = hivex_value_value (h, value, &t, &len);
1409 if ((t != hive_t_dword && t != hive_t_dword_be) || len != 4) {
1415 int32_t ret = *(int32_t*)data;
1417 if (t == hive_t_dword) /* little endian */
1418 ret = le32toh (ret);
1420 ret = be32toh (ret);
1426 hivex_value_qword (hive_h *h, hive_value_h value)
1430 char *data = hivex_value_value (h, value, &t, &len);
1435 if (t != hive_t_qword || len != 8) {
1441 int64_t ret = *(int64_t*)data;
1443 ret = le64toh (ret); /* always little endian */
1448 /*----------------------------------------------------------------------
1453 hivex_visit (hive_h *h, const struct hivex_visitor *visitor, size_t len,
1454 void *opaque, int flags)
1456 return hivex_visit_node (h, hivex_root (h), visitor, len, opaque, flags);
1459 static int hivex__visit_node (hive_h *h, hive_node_h node, const struct hivex_visitor *vtor, char *unvisited, void *opaque, int flags);
1462 hivex_visit_node (hive_h *h, hive_node_h node,
1463 const struct hivex_visitor *visitor, size_t len, void *opaque,
1466 struct hivex_visitor vtor;
1467 memset (&vtor, 0, sizeof vtor);
1469 /* Note that len might be larger *or smaller* than the expected size. */
1470 size_t copysize = len <= sizeof vtor ? len : sizeof vtor;
1471 memcpy (&vtor, visitor, copysize);
1473 /* This bitmap records unvisited nodes, so we don't loop if the
1474 * registry contains cycles.
1476 char *unvisited = malloc (1 + h->size / 32);
1477 if (unvisited == NULL)
1479 memcpy (unvisited, h->bitmap, 1 + h->size / 32);
1481 int r = hivex__visit_node (h, node, &vtor, unvisited, opaque, flags);
1487 hivex__visit_node (hive_h *h, hive_node_h node,
1488 const struct hivex_visitor *vtor, char *unvisited,
1489 void *opaque, int flags)
1491 int skip_bad = flags & HIVEX_VISIT_SKIP_BAD;
1493 hive_value_h *values = NULL;
1494 hive_node_h *children = NULL;
1500 /* Return -1 on all callback errors. However on internal errors,
1501 * check if skip_bad is set and suppress those errors if so.
1505 if (!BITMAP_TST (unvisited, node)) {
1507 fprintf (stderr, "hivex__visit_node: contains cycle: visited node 0x%zx already\n",
1511 return skip_bad ? 0 : -1;
1513 BITMAP_CLR (unvisited, node);
1515 name = hivex_node_name (h, node);
1516 if (!name) return skip_bad ? 0 : -1;
1517 if (vtor->node_start && vtor->node_start (h, opaque, node, name) == -1)
1520 values = hivex_node_values (h, node);
1522 ret = skip_bad ? 0 : -1;
1526 for (i = 0; values[i] != 0; ++i) {
1530 if (hivex_value_type (h, values[i], &t, &len) == -1) {
1531 ret = skip_bad ? 0 : -1;
1535 key = hivex_value_key (h, values[i]);
1537 ret = skip_bad ? 0 : -1;
1541 if (vtor->value_any) {
1542 str = hivex_value_value (h, values[i], &t, &len);
1544 ret = skip_bad ? 0 : -1;
1547 if (vtor->value_any (h, opaque, node, values[i], t, len, key, str) == -1)
1549 free (str); str = NULL;
1554 str = hivex_value_value (h, values[i], &t, &len);
1556 ret = skip_bad ? 0 : -1;
1559 if (t != hive_t_none) {
1560 ret = skip_bad ? 0 : -1;
1563 if (vtor->value_none &&
1564 vtor->value_none (h, opaque, node, values[i], t, len, key, str) == -1)
1566 free (str); str = NULL;
1570 case hive_t_expand_string:
1572 str = hivex_value_string (h, values[i]);
1574 if (errno != EILSEQ && errno != EINVAL) {
1575 ret = skip_bad ? 0 : -1;
1578 if (vtor->value_string_invalid_utf16) {
1579 str = hivex_value_value (h, values[i], &t, &len);
1580 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i], t, len, key, str) == -1)
1582 free (str); str = NULL;
1586 if (vtor->value_string &&
1587 vtor->value_string (h, opaque, node, values[i], t, len, key, str) == -1)
1589 free (str); str = NULL;
1593 case hive_t_dword_be: {
1594 int32_t i32 = hivex_value_dword (h, values[i]);
1595 if (vtor->value_dword &&
1596 vtor->value_dword (h, opaque, node, values[i], t, len, key, i32) == -1)
1601 case hive_t_qword: {
1602 int64_t i64 = hivex_value_qword (h, values[i]);
1603 if (vtor->value_qword &&
1604 vtor->value_qword (h, opaque, node, values[i], t, len, key, i64) == -1)
1610 str = hivex_value_value (h, values[i], &t, &len);
1612 ret = skip_bad ? 0 : -1;
1615 if (t != hive_t_binary) {
1616 ret = skip_bad ? 0 : -1;
1619 if (vtor->value_binary &&
1620 vtor->value_binary (h, opaque, node, values[i], t, len, key, str) == -1)
1622 free (str); str = NULL;
1625 case hive_t_multiple_strings:
1626 strs = hivex_value_multiple_strings (h, values[i]);
1628 if (errno != EILSEQ && errno != EINVAL) {
1629 ret = skip_bad ? 0 : -1;
1632 if (vtor->value_string_invalid_utf16) {
1633 str = hivex_value_value (h, values[i], &t, &len);
1634 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i], t, len, key, str) == -1)
1636 free (str); str = NULL;
1640 if (vtor->value_multiple_strings &&
1641 vtor->value_multiple_strings (h, opaque, node, values[i], t, len, key, strs) == -1)
1643 free_strings (strs); strs = NULL;
1646 case hive_t_resource_list:
1647 case hive_t_full_resource_description:
1648 case hive_t_resource_requirements_list:
1650 str = hivex_value_value (h, values[i], &t, &len);
1652 ret = skip_bad ? 0 : -1;
1655 if (vtor->value_other &&
1656 vtor->value_other (h, opaque, node, values[i], t, len, key, str) == -1)
1658 free (str); str = NULL;
1663 free (key); key = NULL;
1666 children = hivex_node_children (h, node);
1667 if (children == NULL) {
1668 ret = skip_bad ? 0 : -1;
1672 for (i = 0; children[i] != 0; ++i) {
1674 fprintf (stderr, "hivex__visit_node: %s: visiting subkey %d (0x%zx)\n",
1675 name, i, children[i]);
1677 if (hivex__visit_node (h, children[i], vtor, unvisited, opaque, flags) == -1)
1681 if (vtor->node_end && vtor->node_end (h, opaque, node, name) == -1)
1692 free_strings (strs);
1696 /*----------------------------------------------------------------------
1700 /* Allocate an hbin (page), extending the malloc'd space if necessary,
1701 * and updating the hive handle fields (but NOT the hive disk header
1702 * -- the hive disk header is updated when we commit). This function
1703 * also extends the bitmap if necessary.
1705 * 'allocation_hint' is the size of the block allocation we would like
1706 * to make. Normally registry blocks are very small (avg 50 bytes)
1707 * and are contained in standard-sized pages (4KB), but the registry
1708 * can support blocks which are larger than a standard page, in which
1709 * case it creates a page of 8KB, 12KB etc.
1712 * > 0 : offset of first usable byte of new page (after page header)
1713 * 0 : error (errno set)
1716 allocate_page (hive_h *h, size_t allocation_hint)
1718 /* In almost all cases this will be 1. */
1719 size_t nr_4k_pages =
1720 1 + (allocation_hint + sizeof (struct ntreg_hbin_page) - 1) / 4096;
1721 assert (nr_4k_pages >= 1);
1723 /* 'extend' is the number of bytes to extend the file by. Note that
1724 * hives found in the wild often contain slack between 'endpages'
1725 * and the actual end of the file, so we don't always need to make
1728 ssize_t extend = h->endpages + nr_4k_pages * 4096 - h->size;
1730 if (h->msglvl >= 2) {
1731 fprintf (stderr, "allocate_page: current endpages = 0x%zx, current size = 0x%zx\n",
1732 h->endpages, h->size);
1733 fprintf (stderr, "allocate_page: extending file by %zd bytes (<= 0 if no extension)\n",
1738 size_t oldsize = h->size;
1739 size_t newsize = h->size + extend;
1740 char *newaddr = realloc (h->addr, newsize);
1741 if (newaddr == NULL)
1744 size_t oldbitmapsize = 1 + oldsize / 32;
1745 size_t newbitmapsize = 1 + newsize / 32;
1746 char *newbitmap = realloc (h->bitmap, newbitmapsize);
1747 if (newbitmap == NULL) {
1754 h->bitmap = newbitmap;
1756 memset (h->addr + oldsize, 0, newsize - oldsize);
1757 memset (h->bitmap + oldbitmapsize, 0, newbitmapsize - oldbitmapsize);
1760 size_t offset = h->endpages;
1761 h->endpages += nr_4k_pages * 4096;
1764 fprintf (stderr, "allocate_page: new endpages = 0x%zx, new size = 0x%zx\n",
1765 h->endpages, h->size);
1767 /* Write the hbin header. */
1768 struct ntreg_hbin_page *page =
1769 (struct ntreg_hbin_page *) (h->addr + offset);
1770 page->magic[0] = 'h';
1771 page->magic[1] = 'b';
1772 page->magic[2] = 'i';
1773 page->magic[3] = 'n';
1774 page->offset_first = htole32 (offset - 0x1000);
1775 page->page_size = htole32 (nr_4k_pages * 4096);
1776 memset (page->unknown, 0, sizeof (page->unknown));
1779 fprintf (stderr, "allocate_page: new page at 0x%zx\n", offset);
1781 /* Offset of first usable byte after the header. */
1782 return offset + sizeof (struct ntreg_hbin_page);
1785 /* Allocate a single block, first allocating an hbin (page) at the end
1786 * of the current file if necessary. NB. To keep the implementation
1787 * simple and more likely to be correct, we do not reuse existing free
1790 * seg_len is the size of the block (this INCLUDES the block header).
1791 * The header of the block is initialized to -seg_len (negative to
1792 * indicate used). id[2] is the block ID (type), eg. "nk" for nk-
1793 * record. The block bitmap is updated to show this block as valid.
1794 * The rest of the contents of the block will be zero.
1797 * > 0 : offset of new block
1798 * 0 : error (errno set)
1801 allocate_block (hive_h *h, size_t seg_len, const char id[2])
1809 /* The caller probably forgot to include the header. Note that
1810 * value lists have no ID field, so seg_len == 4 would be possible
1811 * for them, albeit unusual.
1814 fprintf (stderr, "allocate_block: refusing too small allocation (%zu), returning ERANGE\n",
1820 /* Refuse really large allocations. */
1821 if (seg_len > HIVEX_MAX_ALLOCATION) {
1823 fprintf (stderr, "allocate_block: refusing large allocation (%zu), returning ERANGE\n",
1829 /* Round up allocation to multiple of 8 bytes. All blocks must be
1830 * on an 8 byte boundary.
1832 seg_len = (seg_len + 7) & ~7;
1834 /* Allocate a new page if necessary. */
1835 if (h->endblocks == 0 || h->endblocks + seg_len > h->endpages) {
1836 size_t newendblocks = allocate_page (h, seg_len);
1837 if (newendblocks == 0)
1839 h->endblocks = newendblocks;
1842 size_t offset = h->endblocks;
1845 fprintf (stderr, "allocate_block: new block at 0x%zx, size %zu\n",
1848 struct ntreg_hbin_block *blockhdr =
1849 (struct ntreg_hbin_block *) (h->addr + offset);
1851 blockhdr->seg_len = htole32 (- (int32_t) seg_len);
1852 if (id[0] && id[1] && seg_len >= sizeof (struct ntreg_hbin_block)) {
1853 blockhdr->id[0] = id[0];
1854 blockhdr->id[1] = id[1];
1857 BITMAP_SET (h->bitmap, offset);
1859 h->endblocks += seg_len;
1861 /* If there is space after the last block in the last page, then we
1862 * have to put a dummy free block header here to mark the rest of
1865 ssize_t rem = h->endpages - h->endblocks;
1868 fprintf (stderr, "allocate_block: marking remainder of page free starting at 0x%zx, size %zd\n",
1873 blockhdr = (struct ntreg_hbin_block *) (h->addr + h->endblocks);
1874 blockhdr->seg_len = htole32 ((int32_t) rem);
1880 /* 'offset' must point to a valid, used block. This function marks
1881 * the block unused (by updating the seg_len field) and invalidates
1882 * the bitmap. It does NOT do this recursively, so to avoid creating
1883 * unreachable used blocks, callers may have to recurse over the hive
1884 * structures. Also callers must ensure there are no references to
1885 * this block from other parts of the hive.
1888 mark_block_unused (hive_h *h, size_t offset)
1890 assert (h->writable);
1891 assert (IS_VALID_BLOCK (h, offset));
1894 fprintf (stderr, "mark_block_unused: marking 0x%zx unused\n", offset);
1896 struct ntreg_hbin_block *blockhdr =
1897 (struct ntreg_hbin_block *) (h->addr + offset);
1899 size_t seg_len = block_len (h, offset, NULL);
1900 blockhdr->seg_len = htole32 (seg_len);
1902 BITMAP_CLR (h->bitmap, offset);
1905 /* Delete all existing values at this node. */
1907 delete_values (hive_h *h, hive_node_h node)
1909 assert (h->writable);
1911 hive_value_h *values;
1913 if (get_values (h, node, &values, &blocks) == -1)
1917 for (i = 0; blocks[i] != 0; ++i)
1918 mark_block_unused (h, blocks[i]);
1922 for (i = 0; values[i] != 0; ++i) {
1923 struct ntreg_vk_record *vk =
1924 (struct ntreg_vk_record *) (h->addr + values[i]);
1927 len = le32toh (vk->data_len);
1928 if (len == 0x80000000) /* special case */
1932 if (len > 4) { /* non-inline, so remove data block */
1933 size_t data_offset = le32toh (vk->data_offset);
1934 data_offset += 0x1000;
1935 mark_block_unused (h, data_offset);
1938 /* remove vk record */
1939 mark_block_unused (h, values[i]);
1944 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
1945 nk->nr_values = htole32 (0);
1946 nk->vallist = htole32 (0xffffffff);
1952 hivex_commit (hive_h *h, const char *filename, int flags)
1964 filename = filename ? : h->filename;
1965 int fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
1969 /* Update the header fields. */
1970 uint32_t sequence = le32toh (h->hdr->sequence1);
1972 h->hdr->sequence1 = htole32 (sequence);
1973 h->hdr->sequence2 = htole32 (sequence);
1974 /* XXX Ought to update h->hdr->last_modified. */
1975 h->hdr->blocks = htole32 (h->endpages - 0x1000);
1977 /* Recompute header checksum. */
1978 uint32_t sum = header_checksum (h);
1979 h->hdr->csum = htole32 (sum);
1982 fprintf (stderr, "hivex_commit: new header checksum: 0x%x\n", sum);
1984 if (full_write (fd, h->addr, h->size) != h->size) {
1991 if (close (fd) == -1)
1997 /* Calculate the hash for a lf or lh record offset.
2000 calc_hash (const char *type, const char *name, char *ret)
2002 size_t len = strlen (name);
2004 if (STRPREFIX (type, "lf"))
2005 /* Old-style, not used in current registries. */
2006 memcpy (ret, name, len < 4 ? len : 4);
2008 /* New-style for lh-records. */
2011 for (i = 0; i < len; ++i) {
2012 c = c_toupper (name[i]);
2016 *((uint32_t *) ret) = htole32 (h);
2020 /* Create a completely new lh-record containing just the single node. */
2022 new_lh_record (hive_h *h, const char *name, hive_node_h node)
2024 static const char id[2] = { 'l', 'h' };
2025 size_t seg_len = sizeof (struct ntreg_lf_record);
2026 size_t offset = allocate_block (h, seg_len, id);
2030 struct ntreg_lf_record *lh = (struct ntreg_lf_record *) (h->addr + offset);
2031 lh->nr_keys = htole16 (1);
2032 lh->keys[0].offset = htole32 (node - 0x1000);
2033 calc_hash ("lh", name, lh->keys[0].hash);
2038 /* Insert node into existing lf/lh-record at position.
2039 * This allocates a new record and marks the old one as unused.
2042 insert_lf_record (hive_h *h, size_t old_offs, size_t posn,
2043 const char *name, hive_node_h node)
2045 assert (IS_VALID_BLOCK (h, old_offs));
2047 /* Work around C stupidity.
2048 * http://www.redhat.com/archives/libguestfs/2010-February/msg00056.html
2050 int test = BLOCK_ID_EQ (h, old_offs, "lf") || BLOCK_ID_EQ (h, old_offs, "lh");
2053 struct ntreg_lf_record *old_lf =
2054 (struct ntreg_lf_record *) (h->addr + old_offs);
2055 size_t nr_keys = le16toh (old_lf->nr_keys);
2057 nr_keys++; /* in new record ... */
2059 size_t seg_len = sizeof (struct ntreg_lf_record) + (nr_keys-1) * 8;
2060 size_t new_offs = allocate_block (h, seg_len, old_lf->id);
2064 struct ntreg_lf_record *new_lf =
2065 (struct ntreg_lf_record *) (h->addr + new_offs);
2066 new_lf->nr_keys = htole16 (nr_keys);
2068 /* Copy the keys until we reach posn, insert the new key there, then
2069 * copy the remaining keys.
2072 for (i = 0; i < posn; ++i)
2073 new_lf->keys[i] = old_lf->keys[i];
2075 new_lf->keys[i].offset = htole32 (node - 0x1000);
2076 calc_hash (new_lf->id, name, new_lf->keys[i].hash);
2078 for (i = posn+1; i < nr_keys; ++i)
2079 new_lf->keys[i] = old_lf->keys[i-1];
2081 /* Old block is unused, return new block. */
2082 mark_block_unused (h, old_offs);
2086 /* Compare name with name in nk-record. */
2088 compare_name_with_nk_name (hive_h *h, const char *name, hive_node_h nk_offs)
2090 assert (IS_VALID_BLOCK (h, nk_offs));
2091 assert (BLOCK_ID_EQ (h, nk_offs, "nk"));
2093 /* Name in nk is not necessarily nul-terminated. */
2094 char *nname = hivex_node_name (h, nk_offs);
2096 /* Unfortunately we don't have a way to return errors here. */
2098 perror ("compare_name_with_nk_name");
2102 int r = strcasecmp (name, nname);
2109 hivex_node_add_child (hive_h *h, hive_node_h parent, const char *name)
2116 if (!IS_VALID_BLOCK (h, parent) || !BLOCK_ID_EQ (h, parent, "nk")) {
2121 if (name == NULL || strlen (name) == 0) {
2126 if (hivex_node_get_child (h, parent, name) != 0) {
2131 /* Create the new nk-record. */
2132 static const char nk_id[2] = { 'n', 'k' };
2133 size_t seg_len = sizeof (struct ntreg_nk_record) + strlen (name);
2134 hive_node_h node = allocate_block (h, seg_len, nk_id);
2139 fprintf (stderr, "hivex_node_add_child: allocated new nk-record for child at 0x%zx\n", node);
2141 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2142 nk->flags = htole16 (0x0020); /* key is ASCII. */
2143 nk->parent = htole32 (parent - 0x1000);
2144 nk->subkey_lf = htole32 (0xffffffff);
2145 nk->subkey_lf_volatile = htole32 (0xffffffff);
2146 nk->vallist = htole32 (0xffffffff);
2147 nk->classname = htole32 (0xffffffff);
2148 nk->name_len = htole16 (strlen (name));
2149 strcpy (nk->name, name);
2151 /* Inherit parent sk. */
2152 struct ntreg_nk_record *parent_nk =
2153 (struct ntreg_nk_record *) (h->addr + parent);
2154 size_t parent_sk_offset = le32toh (parent_nk->sk);
2155 parent_sk_offset += 0x1000;
2156 if (!IS_VALID_BLOCK (h, parent_sk_offset) ||
2157 !BLOCK_ID_EQ (h, parent_sk_offset, "sk")) {
2159 fprintf (stderr, "hivex_node_add_child: returning EFAULT because parent sk is not a valid block (%zu)\n",
2164 struct ntreg_sk_record *sk =
2165 (struct ntreg_sk_record *) (h->addr + parent_sk_offset);
2166 sk->refcount = htole32 (le32toh (sk->refcount) + 1);
2167 nk->sk = htole32 (parent_sk_offset - 0x1000);
2169 /* Inherit parent timestamp. */
2170 memcpy (nk->timestamp, parent_nk->timestamp, sizeof (parent_nk->timestamp));
2172 /* What I found out the hard way (not documented anywhere): the
2173 * subkeys in lh-records must be kept sorted. If you just add a
2174 * subkey in a non-sorted position (eg. just add it at the end) then
2175 * Windows won't see the subkey _and_ Windows will corrupt the hive
2176 * itself when it modifies or saves it.
2178 * So use get_children() to get a list of intermediate
2179 * lf/lh-records. get_children() returns these in reading order
2180 * (which is sorted), so we look for the lf/lh-records in sequence
2181 * until we find the key name just after the one we are inserting,
2182 * and we insert the subkey just before it.
2184 * The only other case is the no-subkeys case, where we have to
2185 * create a brand new lh-record.
2187 hive_node_h *unused;
2190 if (get_children (h, parent, &unused, &blocks, 0) == -1)
2195 size_t nr_subkeys_in_parent_nk = le32toh (parent_nk->nr_subkeys);
2196 if (nr_subkeys_in_parent_nk == 0) { /* No subkeys case. */
2197 /* Free up any existing intermediate blocks. */
2198 for (i = 0; blocks[i] != 0; ++i)
2199 mark_block_unused (h, blocks[i]);
2200 size_t lh_offs = new_lh_record (h, name, node);
2207 fprintf (stderr, "hivex_node_add_child: no keys, allocated new lh-record at 0x%zx\n", lh_offs);
2209 parent_nk->subkey_lf = htole32 (lh_offs - 0x1000);
2211 else { /* Insert subkeys case. */
2212 size_t old_offs = 0, new_offs = 0;
2213 struct ntreg_lf_record *old_lf = NULL;
2215 /* Find lf/lh key name just after the one we are inserting. */
2216 for (i = 0; blocks[i] != 0; ++i) {
2217 if (BLOCK_ID_EQ (h, blocks[i], "lf") ||
2218 BLOCK_ID_EQ (h, blocks[i], "lh")) {
2219 old_offs = blocks[i];
2220 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2221 for (j = 0; j < le16toh (old_lf->nr_keys); ++j) {
2222 hive_node_h nk_offs = le32toh (old_lf->keys[j].offset);
2224 if (compare_name_with_nk_name (h, name, nk_offs) < 0)
2230 /* Insert it at the end.
2231 * old_offs points to the last lf record, set j.
2233 assert (old_offs != 0); /* should never happen if nr_subkeys > 0 */
2234 j = le16toh (old_lf->nr_keys);
2239 fprintf (stderr, "hivex_node_add_child: insert key in existing lh-record at 0x%zx, posn %zu\n", old_offs, j);
2241 new_offs = insert_lf_record (h, old_offs, j, name, node);
2242 if (new_offs == 0) {
2248 fprintf (stderr, "hivex_node_add_child: new lh-record at 0x%zx\n",
2251 /* If the lf/lh-record was directly referenced by the parent nk,
2252 * then update the parent nk.
2254 if (le32toh (parent_nk->subkey_lf) + 0x1000 == old_offs)
2255 parent_nk->subkey_lf = htole32 (new_offs - 0x1000);
2256 /* Else we have to look for the intermediate ri-record and update
2260 for (i = 0; blocks[i] != 0; ++i) {
2261 if (BLOCK_ID_EQ (h, blocks[i], "ri")) {
2262 struct ntreg_ri_record *ri =
2263 (struct ntreg_ri_record *) (h->addr + blocks[i]);
2264 for (j = 0; j < le16toh (ri->nr_offsets); ++j)
2265 if (le32toh (ri->offset[j] + 0x1000) == old_offs) {
2266 ri->offset[j] = htole32 (new_offs - 0x1000);
2272 /* Not found .. This is an internal error. */
2274 fprintf (stderr, "hivex_node_add_child: returning ENOTSUP because could not find ri->lf link\n");
2286 /* Update nr_subkeys in parent nk. */
2287 nr_subkeys_in_parent_nk++;
2288 parent_nk->nr_subkeys = htole32 (nr_subkeys_in_parent_nk);
2290 /* Update max_subkey_name_len in parent nk. */
2291 uint16_t max = le16toh (parent_nk->max_subkey_name_len);
2292 if (max < strlen (name) * 2) /* *2 because "recoded" in UTF16-LE. */
2293 parent_nk->max_subkey_name_len = htole16 (strlen (name) * 2);
2298 /* Decrement the refcount of an sk-record, and if it reaches zero,
2299 * unlink it from the chain and delete it.
2302 delete_sk (hive_h *h, size_t sk_offset)
2304 if (!IS_VALID_BLOCK (h, sk_offset) || !BLOCK_ID_EQ (h, sk_offset, "sk")) {
2306 fprintf (stderr, "delete_sk: not an sk record: 0x%zx\n", sk_offset);
2311 struct ntreg_sk_record *sk = (struct ntreg_sk_record *) (h->addr + sk_offset);
2313 if (sk->refcount == 0) {
2315 fprintf (stderr, "delete_sk: sk record already has refcount 0: 0x%zx\n",
2323 if (sk->refcount == 0) {
2324 size_t sk_prev_offset = sk->sk_prev;
2325 sk_prev_offset += 0x1000;
2327 size_t sk_next_offset = sk->sk_next;
2328 sk_next_offset += 0x1000;
2330 /* Update sk_prev/sk_next SKs, unless they both point back to this
2331 * cell in which case we are deleting the last SK.
2333 if (sk_prev_offset != sk_offset && sk_next_offset != sk_offset) {
2334 struct ntreg_sk_record *sk_prev =
2335 (struct ntreg_sk_record *) (h->addr + sk_prev_offset);
2336 struct ntreg_sk_record *sk_next =
2337 (struct ntreg_sk_record *) (h->addr + sk_next_offset);
2339 sk_prev->sk_next = htole32 (sk_next_offset - 0x1000);
2340 sk_next->sk_prev = htole32 (sk_prev_offset - 0x1000);
2343 /* Refcount is zero so really delete this block. */
2344 mark_block_unused (h, sk_offset);
2350 /* Callback from hivex_node_delete_child which is called to delete a
2351 * node AFTER its subnodes have been visited. The subnodes have been
2352 * deleted but we still have to delete any lf/lh/li/ri records and the
2353 * value list block and values, followed by deleting the node itself.
2356 delete_node (hive_h *h, void *opaque, hive_node_h node, const char *name)
2358 /* Get the intermediate blocks. The subkeys have already been
2359 * deleted by this point, so tell get_children() not to check for
2360 * validity of the nk-records.
2362 hive_node_h *unused;
2364 if (get_children (h, node, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK) == -1)
2368 /* We don't care what's in these intermediate blocks, so we can just
2369 * delete them unconditionally.
2372 for (i = 0; blocks[i] != 0; ++i)
2373 mark_block_unused (h, blocks[i]);
2377 /* Delete the values in the node. */
2378 if (delete_values (h, node) == -1)
2381 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2383 /* If the NK references an SK, delete it. */
2384 size_t sk_offs = le32toh (nk->sk);
2385 if (sk_offs != 0xffffffff) {
2387 if (delete_sk (h, sk_offs) == -1)
2389 nk->sk = htole32 (0xffffffff);
2392 /* If the NK references a classname, delete it. */
2393 size_t cl_offs = le32toh (nk->classname);
2394 if (cl_offs != 0xffffffff) {
2396 mark_block_unused (h, cl_offs);
2397 nk->classname = htole32 (0xffffffff);
2400 /* Delete the node itself. */
2401 mark_block_unused (h, node);
2407 hivex_node_delete_child (hive_h *h, hive_node_h node)
2414 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2419 if (node == hivex_root (h)) {
2421 fprintf (stderr, "hivex_node_delete_child: cannot delete root node\n");
2426 hive_node_h parent = hivex_node_parent (h, node);
2430 /* Delete node and all its children and values recursively. */
2431 static const struct hivex_visitor visitor = { .node_end = delete_node };
2432 if (hivex_visit_node (h, node, &visitor, sizeof visitor, NULL, 0) == -1)
2435 /* Delete the link from parent to child. We need to find the lf/lh
2436 * record which contains the offset and remove the offset from that
2437 * record, then decrement the element count in that record, and
2438 * decrement the overall number of subkeys stored in the parent
2441 hive_node_h *unused;
2443 if (get_children (h, parent, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK)== -1)
2448 for (i = 0; blocks[i] != 0; ++i) {
2449 struct ntreg_hbin_block *block =
2450 (struct ntreg_hbin_block *) (h->addr + blocks[i]);
2452 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
2453 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
2455 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
2457 for (j = 0; j < nr_subkeys_in_lf; ++j)
2458 if (le32toh (lf->keys[j].offset) + 0x1000 == node) {
2459 for (; j < nr_subkeys_in_lf - 1; ++j)
2460 memcpy (&lf->keys[j], &lf->keys[j+1], sizeof (lf->keys[j]));
2461 lf->nr_keys = htole16 (nr_subkeys_in_lf - 1);
2467 fprintf (stderr, "hivex_node_delete_child: could not find parent to child link\n");
2472 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + parent);
2473 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
2474 nk->nr_subkeys = htole32 (nr_subkeys_in_nk - 1);
2477 fprintf (stderr, "hivex_node_delete_child: updating nr_subkeys in parent 0x%zx to %zu\n",
2478 parent, nr_subkeys_in_nk);
2484 hivex_node_set_values (hive_h *h, hive_node_h node,
2485 size_t nr_values, const hive_set_value *values,
2493 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2498 /* Delete all existing values. */
2499 if (delete_values (h, node) == -1)
2505 /* Allocate value list node. Value lists have no id field. */
2506 static const char nul_id[2] = { 0, 0 };
2508 sizeof (struct ntreg_value_list) + (nr_values - 1) * sizeof (uint32_t);
2509 size_t vallist_offs = allocate_block (h, seg_len, nul_id);
2510 if (vallist_offs == 0)
2513 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2514 nk->nr_values = htole32 (nr_values);
2515 nk->vallist = htole32 (vallist_offs - 0x1000);
2517 struct ntreg_value_list *vallist =
2518 (struct ntreg_value_list *) (h->addr + vallist_offs);
2521 for (i = 0; i < nr_values; ++i) {
2522 /* Allocate vk record to store this (key, value) pair. */
2523 static const char vk_id[2] = { 'v', 'k' };
2524 seg_len = sizeof (struct ntreg_vk_record) + strlen (values[i].key);
2525 size_t vk_offs = allocate_block (h, seg_len, vk_id);
2529 vallist->offset[i] = htole32 (vk_offs - 0x1000);
2531 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2532 size_t name_len = strlen (values[i].key);
2533 vk->name_len = htole16 (name_len);
2534 strcpy (vk->name, values[i].key);
2535 vk->data_type = htole32 (values[i].t);
2536 vk->data_len = htole16 (values[i].len);
2537 vk->flags = name_len == 0 ? 0 : 1;
2539 if (values[i].len <= 4) /* Store data inline. */
2540 memcpy (&vk->data_offset, values[i].value, values[i].len);
2542 size_t offs = allocate_block (h, values[i].len + 4, nul_id);
2545 memcpy (h->addr + offs + 4, values[i].value, values[i].len);
2546 vk->data_offset = htole32 (offs - 0x1000);
2549 if (name_len * 2 > le32toh (nk->max_vk_name_len))
2550 /* * 2 for UTF16-LE "reencoding" */
2551 nk->max_vk_name_len = htole32 (name_len * 2);
2552 if (values[i].len > le32toh (nk->max_vk_data_len))
2553 nk->max_vk_data_len = htole32 (values[i].len);