1 /* hivex - Windows Registry "hive" extraction library.
2 * Copyright (C) 2009-2011 Red Hat Inc.
3 * Derived from code by Petter Nordahl-Hagen under a compatible license:
4 * Copyright (c) 1997-2007 Petter Nordahl-Hagen.
5 * Derived from code by Markus Stephany under a compatible license:
6 * Copyright (c) 2000-2004, Markus Stephany.
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation;
11 * version 2.1 of the License.
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * See file LICENSE for the full license.
38 #include "full-read.h"
39 #include "full-write.h"
41 #define STREQ(a,b) (strcmp((a),(b)) == 0)
42 #define STRCASEEQ(a,b) (strcasecmp((a),(b)) == 0)
43 //#define STRNEQ(a,b) (strcmp((a),(b)) != 0)
44 //#define STRCASENEQ(a,b) (strcasecmp((a),(b)) != 0)
45 #define STREQLEN(a,b,n) (strncmp((a),(b),(n)) == 0)
46 //#define STRCASEEQLEN(a,b,n) (strncasecmp((a),(b),(n)) == 0)
47 //#define STRNEQLEN(a,b,n) (strncmp((a),(b),(n)) != 0)
48 //#define STRCASENEQLEN(a,b,n) (strncasecmp((a),(b),(n)) != 0)
49 #define STRPREFIX(a,b) (strncmp((a),(b),strlen((b))) == 0)
52 #include "byte_conversions.h"
54 /* These limits are in place to stop really stupid stuff and/or exploits. */
55 #define HIVEX_MAX_SUBKEYS 15000
56 #define HIVEX_MAX_VALUES 10000
57 #define HIVEX_MAX_VALUE_LEN 1000000
58 #define HIVEX_MAX_ALLOCATION 1000000
60 static char *windows_utf16_to_utf8 (/* const */ char *input, size_t len);
61 static size_t utf16_string_len_in_bytes_max (const char *str, size_t len);
70 /* Registry file, memory mapped if read-only, or malloc'd if writing. */
73 struct ntreg_header *hdr;
76 /* Use a bitmap to store which file offsets are valid (point to a
77 * used block). We only need to store 1 bit per 32 bits of the file
78 * (because blocks are 4-byte aligned). We found that the average
79 * block size in a registry file is ~50 bytes. So roughly 1 in 12
80 * bits in the bitmap will be set, making it likely a more efficient
81 * structure than a hash table.
84 #define BITMAP_SET(bitmap,off) (bitmap[(off)>>5] |= 1 << (((off)>>2)&7))
85 #define BITMAP_CLR(bitmap,off) (bitmap[(off)>>5] &= ~ (1 << (((off)>>2)&7)))
86 #define BITMAP_TST(bitmap,off) (bitmap[(off)>>5] & (1 << (((off)>>2)&7)))
87 #define IS_VALID_BLOCK(h,off) \
88 (((off) & 3) == 0 && \
90 (off) < (h)->size && \
91 BITMAP_TST((h)->bitmap,(off)))
93 /* Fields from the header, extracted from little-endianness hell. */
94 size_t rootoffs; /* Root key offset (always an nk-block). */
95 size_t endpages; /* Offset of end of pages. */
98 size_t endblocks; /* Offset to next block allocation (0
99 if not allocated anything yet). */
102 /* NB. All fields are little endian. */
103 struct ntreg_header {
104 char magic[4]; /* "regf" */
107 char last_modified[8];
108 uint32_t major_ver; /* 1 */
109 uint32_t minor_ver; /* 3 */
110 uint32_t unknown5; /* 0 */
111 uint32_t unknown6; /* 1 */
112 uint32_t offset; /* offset of root key record - 4KB */
113 uint32_t blocks; /* pointer AFTER last hbin in file - 4KB */
114 uint32_t unknown7; /* 1 */
116 char name[64]; /* original file name of hive */
117 char unknown_guid1[16];
118 char unknown_guid2[16];
121 char unknown_guid3[16];
126 uint32_t csum; /* checksum: xor of dwords 0-0x1fb. */
128 char unknown11[3528];
130 char unknown_guid4[16];
131 char unknown_guid5[16];
132 char unknown_guid6[16];
136 } __attribute__((__packed__));
138 struct ntreg_hbin_page {
139 char magic[4]; /* "hbin" */
140 uint32_t offset_first; /* offset from 1st block */
141 uint32_t page_size; /* size of this page (multiple of 4KB) */
143 /* Linked list of blocks follows here. */
144 } __attribute__((__packed__));
146 struct ntreg_hbin_block {
147 int32_t seg_len; /* length of this block (-ve for used block) */
148 char id[2]; /* the block type (eg. "nk" for nk record) */
149 /* Block data follows here. */
150 } __attribute__((__packed__));
152 #define BLOCK_ID_EQ(h,offs,eqid) \
153 (STREQLEN (((struct ntreg_hbin_block *)((h)->addr + (offs)))->id, (eqid), 2))
156 block_len (hive_h *h, size_t blkoff, int *used)
158 struct ntreg_hbin_block *block;
159 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
161 int32_t len = le32toh (block->seg_len);
172 struct ntreg_nk_record {
173 int32_t seg_len; /* length (always -ve because used) */
174 char id[2]; /* "nk" */
178 uint32_t parent; /* offset of owner/parent */
179 uint32_t nr_subkeys; /* number of subkeys */
180 uint32_t nr_subkeys_volatile;
181 uint32_t subkey_lf; /* lf record containing list of subkeys */
182 uint32_t subkey_lf_volatile;
183 uint32_t nr_values; /* number of values */
184 uint32_t vallist; /* value-list record */
185 uint32_t sk; /* offset of sk-record */
186 uint32_t classname; /* offset of classname record */
187 uint16_t max_subkey_name_len; /* maximum length of a subkey name in bytes
188 if the subkey was reencoded as UTF-16LE */
191 uint32_t max_vk_name_len; /* maximum length of any vk name in bytes
192 if the name was reencoded as UTF-16LE */
193 uint32_t max_vk_data_len; /* maximum length of any vk data in bytes */
195 uint16_t name_len; /* length of name */
196 uint16_t classname_len; /* length of classname */
197 char name[1]; /* name follows here */
198 } __attribute__((__packed__));
200 struct ntreg_lf_record {
202 char id[2]; /* "lf"|"lh" */
203 uint16_t nr_keys; /* number of keys in this record */
205 uint32_t offset; /* offset of nk-record for this subkey */
206 char hash[4]; /* hash of subkey name */
208 } __attribute__((__packed__));
210 struct ntreg_ri_record {
212 char id[2]; /* "ri" */
213 uint16_t nr_offsets; /* number of pointers to lh records */
214 uint32_t offset[1]; /* list of pointers to lh records */
215 } __attribute__((__packed__));
217 /* This has no ID header. */
218 struct ntreg_value_list {
220 uint32_t offset[1]; /* list of pointers to vk records */
221 } __attribute__((__packed__));
223 struct ntreg_vk_record {
224 int32_t seg_len; /* length (always -ve because used) */
225 char id[2]; /* "vk" */
226 uint16_t name_len; /* length of name */
227 /* length of the data:
228 * If data_len is <= 4, then it's stored inline.
229 * Top bit is set to indicate inline.
232 uint32_t data_offset; /* pointer to the data (or data if inline) */
233 uint32_t data_type; /* type of the data */
234 uint16_t flags; /* bit 0 set => key name ASCII,
235 bit 0 clr => key name UTF-16.
236 Only seen ASCII here in the wild.
237 NB: this is CLEAR for default key. */
239 char name[1]; /* key name follows here */
240 } __attribute__((__packed__));
242 struct ntreg_sk_record {
243 int32_t seg_len; /* length (always -ve because used) */
244 char id[2]; /* "sk" */
246 uint32_t sk_next; /* linked into a circular list */
248 uint32_t refcount; /* reference count */
249 uint32_t sec_len; /* length of security info */
250 char sec_desc[1]; /* security info follows */
251 } __attribute__((__packed__));
254 header_checksum (const hive_h *h)
256 uint32_t *daddr = (uint32_t *) h->addr;
260 for (i = 0; i < 0x1fc / 4; ++i) {
261 sum ^= le32toh (*daddr);
268 #define HIVEX_OPEN_MSGLVL_MASK (HIVEX_OPEN_VERBOSE|HIVEX_OPEN_DEBUG)
271 hivex_open (const char *filename, int flags)
275 assert (sizeof (struct ntreg_header) == 0x1000);
276 assert (offsetof (struct ntreg_header, csum) == 0x1fc);
278 h = calloc (1, sizeof *h);
282 h->msglvl = flags & HIVEX_OPEN_MSGLVL_MASK;
284 const char *debug = getenv ("HIVEX_DEBUG");
285 if (debug && STREQ (debug, "1"))
289 fprintf (stderr, "hivex_open: created handle %p\n", h);
291 h->writable = !!(flags & HIVEX_OPEN_WRITE);
292 h->filename = strdup (filename);
293 if (h->filename == NULL)
296 h->fd = open (filename, O_RDONLY | O_CLOEXEC);
301 if (fstat (h->fd, &statbuf) == -1)
304 h->size = statbuf.st_size;
307 h->addr = mmap (NULL, h->size, PROT_READ, MAP_SHARED, h->fd, 0);
308 if (h->addr == MAP_FAILED)
312 fprintf (stderr, "hivex_open: mapped file at %p\n", h->addr);
314 h->addr = malloc (h->size);
318 if (full_read (h->fd, h->addr, h->size) < h->size)
321 /* We don't need the file descriptor along this path, since we
322 * have read all the data.
324 if (close (h->fd) == -1)
330 if (h->hdr->magic[0] != 'r' ||
331 h->hdr->magic[1] != 'e' ||
332 h->hdr->magic[2] != 'g' ||
333 h->hdr->magic[3] != 'f') {
334 fprintf (stderr, "hivex: %s: not a Windows NT Registry hive file\n",
340 /* Check major version. */
341 uint32_t major_ver = le32toh (h->hdr->major_ver);
342 if (major_ver != 1) {
344 "hivex: %s: hive file major version %" PRIu32 " (expected 1)\n",
345 filename, major_ver);
350 h->bitmap = calloc (1 + h->size / 32, 1);
351 if (h->bitmap == NULL)
354 /* Header checksum. */
355 uint32_t sum = header_checksum (h);
356 if (sum != le32toh (h->hdr->csum)) {
357 fprintf (stderr, "hivex: %s: bad checksum in hive header\n", filename);
362 if (h->msglvl >= 2) {
363 char *name = windows_utf16_to_utf8 (h->hdr->name, 64);
366 "hivex_open: header fields:\n"
367 " file version %" PRIu32 ".%" PRIu32 "\n"
368 " sequence nos %" PRIu32 " %" PRIu32 "\n"
369 " (sequences nos should match if hive was synched at shutdown)\n"
370 " original file name %s\n"
371 " (only 32 chars are stored, name is probably truncated)\n"
372 " root offset 0x%x + 0x1000\n"
373 " end of last page 0x%x + 0x1000 (total file size 0x%zx)\n"
374 " checksum 0x%x (calculated 0x%x)\n",
375 major_ver, le32toh (h->hdr->minor_ver),
376 le32toh (h->hdr->sequence1), le32toh (h->hdr->sequence2),
377 name ? name : "(conversion failed)",
378 le32toh (h->hdr->offset),
379 le32toh (h->hdr->blocks), h->size,
380 le32toh (h->hdr->csum), sum);
384 h->rootoffs = le32toh (h->hdr->offset) + 0x1000;
385 h->endpages = le32toh (h->hdr->blocks) + 0x1000;
388 fprintf (stderr, "hivex_open: root offset = 0x%zx\n", h->rootoffs);
390 /* We'll set this flag when we see a block with the root offset (ie.
393 int seen_root_block = 0, bad_root_block = 0;
395 /* Collect some stats. */
396 size_t pages = 0; /* Number of hbin pages read. */
397 size_t smallest_page = SIZE_MAX, largest_page = 0;
398 size_t blocks = 0; /* Total number of blocks found. */
399 size_t smallest_block = SIZE_MAX, largest_block = 0, blocks_bytes = 0;
400 size_t used_blocks = 0; /* Total number of used blocks found. */
401 size_t used_size = 0; /* Total size (bytes) of used blocks. */
403 /* Read the pages and blocks. The aim here is to be robust against
404 * corrupt or malicious registries. So we make sure the loops
405 * always make forward progress. We add the address of each block
406 * we read to a hash table so pointers will only reference the start
410 struct ntreg_hbin_page *page;
411 for (off = 0x1000; off < h->size; off += le32toh (page->page_size)) {
412 if (off >= h->endpages)
415 page = (struct ntreg_hbin_page *) (h->addr + off);
416 if (page->magic[0] != 'h' ||
417 page->magic[1] != 'b' ||
418 page->magic[2] != 'i' ||
419 page->magic[3] != 'n') {
420 fprintf (stderr, "hivex: %s: trailing garbage at end of file "
421 "(at 0x%zx, after %zu pages)\n",
422 filename, off, pages);
427 size_t page_size = le32toh (page->page_size);
429 fprintf (stderr, "hivex_open: page at 0x%zx, size %zu\n", off, page_size);
431 if (page_size < smallest_page) smallest_page = page_size;
432 if (page_size > largest_page) largest_page = page_size;
434 if (page_size <= sizeof (struct ntreg_hbin_page) ||
435 (page_size & 0x0fff) != 0) {
436 fprintf (stderr, "hivex: %s: page size %zu at 0x%zx, bad registry\n",
437 filename, page_size, off);
442 /* Read the blocks in this page. */
444 struct ntreg_hbin_block *block;
446 for (blkoff = off + 0x20;
447 blkoff < off + page_size;
451 int is_root = blkoff == h->rootoffs;
455 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
457 seg_len = block_len (h, blkoff, &used);
458 if (seg_len <= 4 || (seg_len & 3) != 0) {
459 fprintf (stderr, "hivex: %s: block size %" PRIu32 " at 0x%zx,"
461 filename, le32toh (block->seg_len), blkoff);
467 fprintf (stderr, "hivex_open: %s block id %d,%d at 0x%zx size %zu%s\n",
468 used ? "used" : "free", block->id[0], block->id[1], blkoff,
469 seg_len, is_root ? " (root)" : "");
471 blocks_bytes += seg_len;
472 if (seg_len < smallest_block) smallest_block = seg_len;
473 if (seg_len > largest_block) largest_block = seg_len;
475 if (is_root && !used)
480 used_size += seg_len;
482 /* Root block must be an nk-block. */
483 if (is_root && (block->id[0] != 'n' || block->id[1] != 'k'))
486 /* Note this blkoff is a valid address. */
487 BITMAP_SET (h->bitmap, blkoff);
492 if (!seen_root_block) {
493 fprintf (stderr, "hivex: %s: no root block found\n", filename);
498 if (bad_root_block) {
499 fprintf (stderr, "hivex: %s: bad root block (free or not nk)\n", filename);
506 "hivex_open: successfully read Windows Registry hive file:\n"
507 " pages: %zu [sml: %zu, lge: %zu]\n"
508 " blocks: %zu [sml: %zu, avg: %zu, lge: %zu]\n"
509 " blocks used: %zu\n"
510 " bytes used: %zu\n",
511 pages, smallest_page, largest_page,
512 blocks, smallest_block, blocks_bytes / blocks, largest_block,
513 used_blocks, used_size);
521 if (h->addr && h->size && h->addr != MAP_FAILED) {
523 munmap (h->addr, h->size);
537 hivex_close (hive_h *h)
542 fprintf (stderr, "hivex_close\n");
546 munmap (h->addr, h->size);
559 /*----------------------------------------------------------------------
564 hivex_root (hive_h *h)
566 hive_node_h ret = h->rootoffs;
567 if (!IS_VALID_BLOCK (h, ret)) {
568 errno = HIVEX_NO_KEY;
575 hivex_node_name (hive_h *h, hive_node_h node)
577 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
582 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
584 /* AFAIK the node name is always plain ASCII, so no conversion
585 * to UTF-8 is necessary. However we do need to nul-terminate
589 /* nk->name_len is unsigned, 16 bit, so this is safe ... However
590 * we have to make sure the length doesn't exceed the block length.
592 size_t len = le16toh (nk->name_len);
593 size_t seg_len = block_len (h, node, NULL);
594 if (sizeof (struct ntreg_nk_record) + len - 1 > seg_len) {
596 fprintf (stderr, "hivex_node_name: returning EFAULT because node name"
597 " is too long (%zu, %zu)\n",
603 char *ret = malloc (len + 1);
606 memcpy (ret, nk->name, len);
612 /* I think the documentation for the sk and classname fields in the nk
613 * record is wrong, or else the offset field is in the wrong place.
614 * Otherwise this makes no sense. Disabled this for now -- it's not
615 * useful for reading the registry anyway.
619 hivex_node_security (hive_h *h, hive_node_h node)
621 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
626 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
628 hive_node_h ret = le32toh (nk->sk);
630 if (!IS_VALID_BLOCK (h, ret)) {
638 hivex_node_classname (hive_h *h, hive_node_h node)
640 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
645 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
647 hive_node_h ret = le32toh (nk->classname);
649 if (!IS_VALID_BLOCK (h, ret)) {
657 /* Structure for returning 0-terminated lists of offsets (nodes,
667 init_offset_list (struct offset_list *list)
671 list->offsets = NULL;
674 #define INIT_OFFSET_LIST(name) \
675 struct offset_list name; \
676 init_offset_list (&name)
678 /* Preallocates the offset_list, but doesn't make the contents longer. */
680 grow_offset_list (struct offset_list *list, size_t alloc)
682 assert (alloc >= list->len);
683 size_t *p = realloc (list->offsets, alloc * sizeof (size_t));
692 add_to_offset_list (struct offset_list *list, size_t offset)
694 if (list->len >= list->alloc) {
695 if (grow_offset_list (list, list->alloc ? list->alloc * 2 : 4) == -1)
698 list->offsets[list->len] = offset;
704 free_offset_list (struct offset_list *list)
706 free (list->offsets);
710 return_offset_list (struct offset_list *list)
712 if (add_to_offset_list (list, 0) == -1)
714 return list->offsets; /* caller frees */
717 /* Iterate over children, returning child nodes and intermediate blocks. */
718 #define GET_CHILDREN_NO_CHECK_NK 1
721 get_children (hive_h *h, hive_node_h node,
722 hive_node_h **children_ret, size_t **blocks_ret,
725 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
730 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
732 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
734 INIT_OFFSET_LIST (children);
735 INIT_OFFSET_LIST (blocks);
737 /* Deal with the common "no subkeys" case quickly. */
738 if (nr_subkeys_in_nk == 0)
741 /* Arbitrarily limit the number of subkeys we will ever deal with. */
742 if (nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS) {
744 fprintf (stderr, "hivex: get_children: returning ERANGE because "
745 "nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS (%zu > %d)\n",
746 nr_subkeys_in_nk, HIVEX_MAX_SUBKEYS);
751 /* Preallocate space for the children. */
752 if (grow_offset_list (&children, nr_subkeys_in_nk) == -1)
755 /* The subkey_lf field can point either to an lf-record, which is
756 * the common case, or if there are lots of subkeys, to an
759 size_t subkey_lf = le32toh (nk->subkey_lf);
761 if (!IS_VALID_BLOCK (h, subkey_lf)) {
763 fprintf (stderr, "hivex_node_children: returning EFAULT"
764 " because subkey_lf is not a valid block (0x%zx)\n",
770 if (add_to_offset_list (&blocks, subkey_lf) == -1)
773 struct ntreg_hbin_block *block =
774 (struct ntreg_hbin_block *) (h->addr + subkey_lf);
776 /* Points to lf-record? (Note, also "lh" but that is basically the
777 * same as "lf" as far as we are concerned here).
779 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
780 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
782 /* Check number of subkeys in the nk-record matches number of subkeys
785 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
788 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu,"
789 " nr_subkeys_in_lf = %zu\n",
790 nr_subkeys_in_nk, nr_subkeys_in_lf);
792 if (nr_subkeys_in_nk != nr_subkeys_in_lf) {
797 size_t len = block_len (h, subkey_lf, NULL);
798 if (8 + nr_subkeys_in_lf * 8 > len) {
800 fprintf (stderr, "hivex_node_children: returning EFAULT"
801 " because too many subkeys (%zu, %zu)\n",
802 nr_subkeys_in_lf, len);
808 for (i = 0; i < nr_subkeys_in_lf; ++i) {
809 hive_node_h subkey = le32toh (lf->keys[i].offset);
811 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
812 if (!IS_VALID_BLOCK (h, subkey)) {
814 fprintf (stderr, "hivex_node_children: returning EFAULT"
815 " because subkey is not a valid block (0x%zx)\n",
821 if (add_to_offset_list (&children, subkey) == -1)
826 /* Points to ri-record? */
827 else if (block->id[0] == 'r' && block->id[1] == 'i') {
828 struct ntreg_ri_record *ri = (struct ntreg_ri_record *) block;
830 size_t nr_offsets = le16toh (ri->nr_offsets);
832 /* Count total number of children. */
834 for (i = 0; i < nr_offsets; ++i) {
835 hive_node_h offset = le32toh (ri->offset[i]);
837 if (!IS_VALID_BLOCK (h, offset)) {
839 fprintf (stderr, "hivex_node_children: returning EFAULT"
840 " because ri-offset is not a valid block (0x%zx)\n",
845 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
847 fprintf (stderr, "get_children: returning ENOTSUP"
848 " because ri-record offset does not point to lf/lh (0x%zx)\n",
854 if (add_to_offset_list (&blocks, offset) == -1)
857 struct ntreg_lf_record *lf =
858 (struct ntreg_lf_record *) (h->addr + offset);
860 count += le16toh (lf->nr_keys);
864 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu,"
866 nr_subkeys_in_nk, count);
868 if (nr_subkeys_in_nk != count) {
873 /* Copy list of children. Note nr_subkeys_in_nk is limited to
874 * something reasonable above.
876 for (i = 0; i < nr_offsets; ++i) {
877 hive_node_h offset = le32toh (ri->offset[i]);
879 if (!IS_VALID_BLOCK (h, offset)) {
881 fprintf (stderr, "hivex_node_children: returning EFAULT"
882 " because ri-offset is not a valid block (0x%zx)\n",
887 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
889 fprintf (stderr, "get_children: returning ENOTSUP"
890 " because ri-record offset does not point to lf/lh (0x%zx)\n",
896 struct ntreg_lf_record *lf =
897 (struct ntreg_lf_record *) (h->addr + offset);
900 for (j = 0; j < le16toh (lf->nr_keys); ++j) {
901 hive_node_h subkey = le32toh (lf->keys[j].offset);
903 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
904 if (!IS_VALID_BLOCK (h, subkey)) {
906 fprintf (stderr, "hivex_node_children: returning EFAULT"
907 " because indirect subkey is not a valid block (0x%zx)\n",
913 if (add_to_offset_list (&children, subkey) == -1)
919 /* else not supported, set errno and fall through */
921 fprintf (stderr, "get_children: returning ENOTSUP"
922 " because subkey block is not lf/lh/ri (0x%zx, %d, %d)\n",
923 subkey_lf, block->id[0], block->id[1]);
926 free_offset_list (&children);
927 free_offset_list (&blocks);
931 *children_ret = return_offset_list (&children);
932 *blocks_ret = return_offset_list (&blocks);
933 if (!*children_ret || !*blocks_ret)
939 hivex_node_children (hive_h *h, hive_node_h node)
941 hive_node_h *children;
944 if (get_children (h, node, &children, &blocks, 0) == -1)
951 /* Very inefficient, but at least having a separate API call
952 * allows us to make it more efficient in future.
955 hivex_node_get_child (hive_h *h, hive_node_h node, const char *nname)
957 hive_node_h *children = NULL;
961 children = hivex_node_children (h, node);
962 if (!children) goto error;
965 for (i = 0; children[i] != 0; ++i) {
966 name = hivex_node_name (h, children[i]);
967 if (!name) goto error;
968 if (STRCASEEQ (name, nname)) {
972 free (name); name = NULL;
982 hivex_node_parent (hive_h *h, hive_node_h node)
984 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
989 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
991 hive_node_h ret = le32toh (nk->parent);
993 if (!IS_VALID_BLOCK (h, ret)) {
995 fprintf (stderr, "hivex_node_parent: returning EFAULT"
996 " because parent is not a valid block (0x%zx)\n",
1005 get_values (hive_h *h, hive_node_h node,
1006 hive_value_h **values_ret, size_t **blocks_ret)
1008 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
1013 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
1015 size_t nr_values = le32toh (nk->nr_values);
1018 fprintf (stderr, "hivex_node_values: nr_values = %zu\n", nr_values);
1020 INIT_OFFSET_LIST (values);
1021 INIT_OFFSET_LIST (blocks);
1023 /* Deal with the common "no values" case quickly. */
1027 /* Arbitrarily limit the number of values we will ever deal with. */
1028 if (nr_values > HIVEX_MAX_VALUES) {
1030 fprintf (stderr, "hivex: get_values: returning ERANGE"
1031 " because nr_values > HIVEX_MAX_VALUES (%zu > %d)\n",
1032 nr_values, HIVEX_MAX_VALUES);
1037 /* Preallocate space for the values. */
1038 if (grow_offset_list (&values, nr_values) == -1)
1041 /* Get the value list and check it looks reasonable. */
1042 size_t vlist_offset = le32toh (nk->vallist);
1043 vlist_offset += 0x1000;
1044 if (!IS_VALID_BLOCK (h, vlist_offset)) {
1046 fprintf (stderr, "hivex_node_values: returning EFAULT"
1047 " because value list is not a valid block (0x%zx)\n",
1053 if (add_to_offset_list (&blocks, vlist_offset) == -1)
1056 struct ntreg_value_list *vlist =
1057 (struct ntreg_value_list *) (h->addr + vlist_offset);
1059 size_t len = block_len (h, vlist_offset, NULL);
1060 if (4 + nr_values * 4 > len) {
1062 fprintf (stderr, "hivex_node_values: returning EFAULT"
1063 " because value list is too long (%zu, %zu)\n",
1070 for (i = 0; i < nr_values; ++i) {
1071 hive_node_h value = le32toh (vlist->offset[i]);
1073 if (!IS_VALID_BLOCK (h, value)) {
1075 fprintf (stderr, "hivex_node_values: returning EFAULT"
1076 " because value is not a valid block (0x%zx)\n",
1081 if (add_to_offset_list (&values, value) == -1)
1086 *values_ret = return_offset_list (&values);
1087 *blocks_ret = return_offset_list (&blocks);
1088 if (!*values_ret || !*blocks_ret)
1093 free_offset_list (&values);
1094 free_offset_list (&blocks);
1099 hivex_node_values (hive_h *h, hive_node_h node)
1101 hive_value_h *values;
1104 if (get_values (h, node, &values, &blocks) == -1)
1111 /* Very inefficient, but at least having a separate API call
1112 * allows us to make it more efficient in future.
1115 hivex_node_get_value (hive_h *h, hive_node_h node, const char *key)
1117 hive_value_h *values = NULL;
1119 hive_value_h ret = 0;
1121 values = hivex_node_values (h, node);
1122 if (!values) goto error;
1125 for (i = 0; values[i] != 0; ++i) {
1126 name = hivex_value_key (h, values[i]);
1127 if (!name) goto error;
1128 if (STRCASEEQ (name, key)) {
1132 free (name); name = NULL;
1142 hivex_value_key (hive_h *h, hive_value_h value)
1144 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1149 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1151 /* AFAIK the key is always plain ASCII, so no conversion to UTF-8 is
1152 * necessary. However we do need to nul-terminate the string.
1155 /* vk->name_len is unsigned, 16 bit, so this is safe ... However
1156 * we have to make sure the length doesn't exceed the block length.
1158 size_t len = le16toh (vk->name_len);
1159 size_t seg_len = block_len (h, value, NULL);
1160 if (sizeof (struct ntreg_vk_record) + len - 1 > seg_len) {
1162 fprintf (stderr, "hivex_value_key: returning EFAULT"
1163 " because key length is too long (%zu, %zu)\n",
1169 char *ret = malloc (len + 1);
1172 memcpy (ret, vk->name, len);
1178 hivex_value_type (hive_h *h, hive_value_h value, hive_type *t, size_t *len)
1180 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1185 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1188 *t = le32toh (vk->data_type);
1191 *len = le32toh (vk->data_len);
1192 *len &= 0x7fffffff; /* top bit indicates if data is stored inline */
1199 hivex_value_value (hive_h *h, hive_value_h value,
1200 hive_type *t_rtn, size_t *len_rtn)
1202 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1207 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1213 t = le32toh (vk->data_type);
1215 len = le32toh (vk->data_len);
1216 is_inline = !!(len & 0x80000000);
1220 fprintf (stderr, "hivex_value_value: value=0x%zx, t=%d, len=%zu, inline=%d\n",
1221 value, t, len, is_inline);
1228 if (is_inline && len > 4) {
1233 /* Arbitrarily limit the length that we will read. */
1234 if (len > HIVEX_MAX_VALUE_LEN) {
1236 fprintf (stderr, "hivex_value_value: returning ERANGE because data "
1237 "length > HIVEX_MAX_VALUE_LEN (%zu > %d)\n",
1238 len, HIVEX_MAX_SUBKEYS);
1243 char *ret = malloc (len);
1248 memcpy (ret, (char *) &vk->data_offset, len);
1252 size_t data_offset = le32toh (vk->data_offset);
1253 data_offset += 0x1000;
1254 if (!IS_VALID_BLOCK (h, data_offset)) {
1256 fprintf (stderr, "hivex_value_value: returning EFAULT because data "
1257 "offset is not a valid block (0x%zx)\n",
1264 /* Check that the declared size isn't larger than the block its in.
1266 * XXX Some apparently valid registries are seen to have this,
1267 * so turn this into a warning and substitute the smaller length
1270 size_t blen = block_len (h, data_offset, NULL);
1271 if (len > blen - 4 /* subtract 4 for block header */) {
1273 fprintf (stderr, "hivex_value_value: warning: declared data length "
1274 "is longer than the block it is in "
1275 "(data 0x%zx, data len %zu, block len %zu)\n",
1276 data_offset, len, blen);
1279 /* Return the smaller length to the caller too. */
1284 char *data = h->addr + data_offset + 4;
1285 memcpy (ret, data, len);
1290 windows_utf16_to_utf8 (/* const */ char *input, size_t len)
1292 iconv_t ic = iconv_open ("UTF-8", "UTF-16");
1293 if (ic == (iconv_t) -1)
1296 /* iconv(3) has an insane interface ... */
1298 /* Mostly UTF-8 will be smaller, so this is a good initial guess. */
1299 size_t outalloc = len;
1303 size_t outlen = outalloc;
1304 char *out = malloc (outlen + 1);
1314 size_t r = iconv (ic, &inp, &inlen, &outp, &outlen);
1315 if (r == (size_t) -1) {
1316 if (errno == E2BIG) {
1318 size_t prev = outalloc;
1319 /* Try again with a larger output buffer. */
1322 if (outalloc < prev) {
1330 /* Else some conversion failure, eg. EILSEQ, EINVAL. */
1346 hivex_value_string (hive_h *h, hive_value_h value)
1350 char *data = hivex_value_value (h, value, &t, &len);
1355 if (t != hive_t_string && t != hive_t_expand_string && t != hive_t_link) {
1361 /* Deal with the case where Windows has allocated a large buffer
1362 * full of random junk, and only the first few bytes of the buffer
1363 * contain a genuine UTF-16 string.
1365 * In this case, iconv would try to process the junk bytes as UTF-16
1366 * and inevitably find an illegal sequence (EILSEQ). Instead, stop
1367 * after we find the first \0\0.
1369 * (Found by Hilko Bengen in a fresh Windows XP SOFTWARE hive).
1371 size_t slen = utf16_string_len_in_bytes_max (data, len);
1375 char *ret = windows_utf16_to_utf8 (data, len);
1384 free_strings (char **argv)
1389 for (i = 0; argv[i] != NULL; ++i)
1395 /* Get the length of a UTF-16 format string. Handle the string as
1396 * pairs of bytes, looking for the first \0\0 pair. Only read up to
1397 * 'len' maximum bytes.
1400 utf16_string_len_in_bytes_max (const char *str, size_t len)
1404 while (len >= 2 && (str[0] || str[1])) {
1413 /* http://blogs.msdn.com/oldnewthing/archive/2009/10/08/9904646.aspx */
1415 hivex_value_multiple_strings (hive_h *h, hive_value_h value)
1419 char *data = hivex_value_value (h, value, &t, &len);
1424 if (t != hive_t_multiple_strings) {
1430 size_t nr_strings = 0;
1431 char **ret = malloc ((1 + nr_strings) * sizeof (char *));
1441 while (p < data + len &&
1442 (plen = utf16_string_len_in_bytes_max (p, data + len - p)) > 0) {
1444 char **ret2 = realloc (ret, (1 + nr_strings) * sizeof (char *));
1452 ret[nr_strings-1] = windows_utf16_to_utf8 (p, plen);
1453 ret[nr_strings] = NULL;
1454 if (ret[nr_strings-1] == NULL) {
1460 p += plen + 2 /* skip over UTF-16 \0\0 at the end of this string */;
1468 hivex_value_dword (hive_h *h, hive_value_h value)
1472 char *data = hivex_value_value (h, value, &t, &len);
1477 if ((t != hive_t_dword && t != hive_t_dword_be) || len != 4) {
1483 int32_t ret = *(int32_t*)data;
1485 if (t == hive_t_dword) /* little endian */
1486 ret = le32toh (ret);
1488 ret = be32toh (ret);
1494 hivex_value_qword (hive_h *h, hive_value_h value)
1498 char *data = hivex_value_value (h, value, &t, &len);
1503 if (t != hive_t_qword || len != 8) {
1509 int64_t ret = *(int64_t*)data;
1511 ret = le64toh (ret); /* always little endian */
1516 /*----------------------------------------------------------------------
1521 hivex_visit (hive_h *h, const struct hivex_visitor *visitor, size_t len,
1522 void *opaque, int flags)
1524 return hivex_visit_node (h, hivex_root (h), visitor, len, opaque, flags);
1527 static int hivex__visit_node (hive_h *h, hive_node_h node,
1528 const struct hivex_visitor *vtor,
1529 char *unvisited, void *opaque, int flags);
1532 hivex_visit_node (hive_h *h, hive_node_h node,
1533 const struct hivex_visitor *visitor, size_t len, void *opaque,
1536 struct hivex_visitor vtor;
1537 memset (&vtor, 0, sizeof vtor);
1539 /* Note that len might be larger *or smaller* than the expected size. */
1540 size_t copysize = len <= sizeof vtor ? len : sizeof vtor;
1541 memcpy (&vtor, visitor, copysize);
1543 /* This bitmap records unvisited nodes, so we don't loop if the
1544 * registry contains cycles.
1546 char *unvisited = malloc (1 + h->size / 32);
1547 if (unvisited == NULL)
1549 memcpy (unvisited, h->bitmap, 1 + h->size / 32);
1551 int r = hivex__visit_node (h, node, &vtor, unvisited, opaque, flags);
1557 hivex__visit_node (hive_h *h, hive_node_h node,
1558 const struct hivex_visitor *vtor, char *unvisited,
1559 void *opaque, int flags)
1561 int skip_bad = flags & HIVEX_VISIT_SKIP_BAD;
1563 hive_value_h *values = NULL;
1564 hive_node_h *children = NULL;
1570 /* Return -1 on all callback errors. However on internal errors,
1571 * check if skip_bad is set and suppress those errors if so.
1575 if (!BITMAP_TST (unvisited, node)) {
1577 fprintf (stderr, "hivex__visit_node: contains cycle:"
1578 " visited node 0x%zx already\n",
1582 return skip_bad ? 0 : -1;
1584 BITMAP_CLR (unvisited, node);
1586 name = hivex_node_name (h, node);
1587 if (!name) return skip_bad ? 0 : -1;
1588 if (vtor->node_start && vtor->node_start (h, opaque, node, name) == -1)
1591 values = hivex_node_values (h, node);
1593 ret = skip_bad ? 0 : -1;
1597 for (i = 0; values[i] != 0; ++i) {
1601 if (hivex_value_type (h, values[i], &t, &len) == -1) {
1602 ret = skip_bad ? 0 : -1;
1606 key = hivex_value_key (h, values[i]);
1608 ret = skip_bad ? 0 : -1;
1612 if (vtor->value_any) {
1613 str = hivex_value_value (h, values[i], &t, &len);
1615 ret = skip_bad ? 0 : -1;
1618 if (vtor->value_any (h, opaque, node, values[i], t, len, key, str) == -1)
1620 free (str); str = NULL;
1625 str = hivex_value_value (h, values[i], &t, &len);
1627 ret = skip_bad ? 0 : -1;
1630 if (t != hive_t_none) {
1631 ret = skip_bad ? 0 : -1;
1634 if (vtor->value_none &&
1635 vtor->value_none (h, opaque, node, values[i], t, len, key, str) == -1)
1637 free (str); str = NULL;
1641 case hive_t_expand_string:
1643 str = hivex_value_string (h, values[i]);
1645 if (errno != EILSEQ && errno != EINVAL) {
1646 ret = skip_bad ? 0 : -1;
1649 if (vtor->value_string_invalid_utf16) {
1650 str = hivex_value_value (h, values[i], &t, &len);
1651 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i],
1652 t, len, key, str) == -1)
1654 free (str); str = NULL;
1658 if (vtor->value_string &&
1659 vtor->value_string (h, opaque, node, values[i],
1660 t, len, key, str) == -1)
1662 free (str); str = NULL;
1666 case hive_t_dword_be: {
1667 int32_t i32 = hivex_value_dword (h, values[i]);
1668 if (vtor->value_dword &&
1669 vtor->value_dword (h, opaque, node, values[i],
1670 t, len, key, i32) == -1)
1675 case hive_t_qword: {
1676 int64_t i64 = hivex_value_qword (h, values[i]);
1677 if (vtor->value_qword &&
1678 vtor->value_qword (h, opaque, node, values[i],
1679 t, len, key, i64) == -1)
1685 str = hivex_value_value (h, values[i], &t, &len);
1687 ret = skip_bad ? 0 : -1;
1690 if (t != hive_t_binary) {
1691 ret = skip_bad ? 0 : -1;
1694 if (vtor->value_binary &&
1695 vtor->value_binary (h, opaque, node, values[i],
1696 t, len, key, str) == -1)
1698 free (str); str = NULL;
1701 case hive_t_multiple_strings:
1702 strs = hivex_value_multiple_strings (h, values[i]);
1704 if (errno != EILSEQ && errno != EINVAL) {
1705 ret = skip_bad ? 0 : -1;
1708 if (vtor->value_string_invalid_utf16) {
1709 str = hivex_value_value (h, values[i], &t, &len);
1710 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i],
1711 t, len, key, str) == -1)
1713 free (str); str = NULL;
1717 if (vtor->value_multiple_strings &&
1718 vtor->value_multiple_strings (h, opaque, node, values[i],
1719 t, len, key, strs) == -1)
1721 free_strings (strs); strs = NULL;
1724 case hive_t_resource_list:
1725 case hive_t_full_resource_description:
1726 case hive_t_resource_requirements_list:
1728 str = hivex_value_value (h, values[i], &t, &len);
1730 ret = skip_bad ? 0 : -1;
1733 if (vtor->value_other &&
1734 vtor->value_other (h, opaque, node, values[i],
1735 t, len, key, str) == -1)
1737 free (str); str = NULL;
1742 free (key); key = NULL;
1745 children = hivex_node_children (h, node);
1746 if (children == NULL) {
1747 ret = skip_bad ? 0 : -1;
1751 for (i = 0; children[i] != 0; ++i) {
1753 fprintf (stderr, "hivex__visit_node: %s: visiting subkey %d (0x%zx)\n",
1754 name, i, children[i]);
1756 if (hivex__visit_node (h, children[i], vtor, unvisited, opaque, flags) == -1)
1760 if (vtor->node_end && vtor->node_end (h, opaque, node, name) == -1)
1771 free_strings (strs);
1775 /*----------------------------------------------------------------------
1779 /* Allocate an hbin (page), extending the malloc'd space if necessary,
1780 * and updating the hive handle fields (but NOT the hive disk header
1781 * -- the hive disk header is updated when we commit). This function
1782 * also extends the bitmap if necessary.
1784 * 'allocation_hint' is the size of the block allocation we would like
1785 * to make. Normally registry blocks are very small (avg 50 bytes)
1786 * and are contained in standard-sized pages (4KB), but the registry
1787 * can support blocks which are larger than a standard page, in which
1788 * case it creates a page of 8KB, 12KB etc.
1791 * > 0 : offset of first usable byte of new page (after page header)
1792 * 0 : error (errno set)
1795 allocate_page (hive_h *h, size_t allocation_hint)
1797 /* In almost all cases this will be 1. */
1798 size_t nr_4k_pages =
1799 1 + (allocation_hint + sizeof (struct ntreg_hbin_page) - 1) / 4096;
1800 assert (nr_4k_pages >= 1);
1802 /* 'extend' is the number of bytes to extend the file by. Note that
1803 * hives found in the wild often contain slack between 'endpages'
1804 * and the actual end of the file, so we don't always need to make
1807 ssize_t extend = h->endpages + nr_4k_pages * 4096 - h->size;
1809 if (h->msglvl >= 2) {
1810 fprintf (stderr, "allocate_page: current endpages = 0x%zx,"
1811 " current size = 0x%zx\n",
1812 h->endpages, h->size);
1813 fprintf (stderr, "allocate_page: extending file by %zd bytes"
1814 " (<= 0 if no extension)\n",
1819 size_t oldsize = h->size;
1820 size_t newsize = h->size + extend;
1821 char *newaddr = realloc (h->addr, newsize);
1822 if (newaddr == NULL)
1825 size_t oldbitmapsize = 1 + oldsize / 32;
1826 size_t newbitmapsize = 1 + newsize / 32;
1827 char *newbitmap = realloc (h->bitmap, newbitmapsize);
1828 if (newbitmap == NULL) {
1835 h->bitmap = newbitmap;
1837 memset (h->addr + oldsize, 0, newsize - oldsize);
1838 memset (h->bitmap + oldbitmapsize, 0, newbitmapsize - oldbitmapsize);
1841 size_t offset = h->endpages;
1842 h->endpages += nr_4k_pages * 4096;
1845 fprintf (stderr, "allocate_page: new endpages = 0x%zx, new size = 0x%zx\n",
1846 h->endpages, h->size);
1848 /* Write the hbin header. */
1849 struct ntreg_hbin_page *page =
1850 (struct ntreg_hbin_page *) (h->addr + offset);
1851 page->magic[0] = 'h';
1852 page->magic[1] = 'b';
1853 page->magic[2] = 'i';
1854 page->magic[3] = 'n';
1855 page->offset_first = htole32 (offset - 0x1000);
1856 page->page_size = htole32 (nr_4k_pages * 4096);
1857 memset (page->unknown, 0, sizeof (page->unknown));
1860 fprintf (stderr, "allocate_page: new page at 0x%zx\n", offset);
1862 /* Offset of first usable byte after the header. */
1863 return offset + sizeof (struct ntreg_hbin_page);
1866 /* Allocate a single block, first allocating an hbin (page) at the end
1867 * of the current file if necessary. NB. To keep the implementation
1868 * simple and more likely to be correct, we do not reuse existing free
1871 * seg_len is the size of the block (this INCLUDES the block header).
1872 * The header of the block is initialized to -seg_len (negative to
1873 * indicate used). id[2] is the block ID (type), eg. "nk" for nk-
1874 * record. The block bitmap is updated to show this block as valid.
1875 * The rest of the contents of the block will be zero.
1877 * **NB** Because allocate_block may reallocate the memory, all
1878 * pointers into the memory become potentially invalid. I really
1879 * love writing in C, can't you tell?
1882 * > 0 : offset of new block
1883 * 0 : error (errno set)
1886 allocate_block (hive_h *h, size_t seg_len, const char id[2])
1894 /* The caller probably forgot to include the header. Note that
1895 * value lists have no ID field, so seg_len == 4 would be possible
1896 * for them, albeit unusual.
1899 fprintf (stderr, "allocate_block: refusing too small allocation (%zu),"
1900 " returning ERANGE\n", seg_len);
1905 /* Refuse really large allocations. */
1906 if (seg_len > HIVEX_MAX_ALLOCATION) {
1908 fprintf (stderr, "allocate_block: refusing large allocation (%zu),"
1909 " returning ERANGE\n", seg_len);
1914 /* Round up allocation to multiple of 8 bytes. All blocks must be
1915 * on an 8 byte boundary.
1917 seg_len = (seg_len + 7) & ~7;
1919 /* Allocate a new page if necessary. */
1920 if (h->endblocks == 0 || h->endblocks + seg_len > h->endpages) {
1921 size_t newendblocks = allocate_page (h, seg_len);
1922 if (newendblocks == 0)
1924 h->endblocks = newendblocks;
1927 size_t offset = h->endblocks;
1930 fprintf (stderr, "allocate_block: new block at 0x%zx, size %zu\n",
1933 struct ntreg_hbin_block *blockhdr =
1934 (struct ntreg_hbin_block *) (h->addr + offset);
1936 memset (blockhdr, 0, seg_len);
1938 blockhdr->seg_len = htole32 (- (int32_t) seg_len);
1939 if (id[0] && id[1] && seg_len >= sizeof (struct ntreg_hbin_block)) {
1940 blockhdr->id[0] = id[0];
1941 blockhdr->id[1] = id[1];
1944 BITMAP_SET (h->bitmap, offset);
1946 h->endblocks += seg_len;
1948 /* If there is space after the last block in the last page, then we
1949 * have to put a dummy free block header here to mark the rest of
1952 ssize_t rem = h->endpages - h->endblocks;
1955 fprintf (stderr, "allocate_block: marking remainder of page free"
1956 " starting at 0x%zx, size %zd\n", h->endblocks, rem);
1960 blockhdr = (struct ntreg_hbin_block *) (h->addr + h->endblocks);
1961 blockhdr->seg_len = htole32 ((int32_t) rem);
1967 /* 'offset' must point to a valid, used block. This function marks
1968 * the block unused (by updating the seg_len field) and invalidates
1969 * the bitmap. It does NOT do this recursively, so to avoid creating
1970 * unreachable used blocks, callers may have to recurse over the hive
1971 * structures. Also callers must ensure there are no references to
1972 * this block from other parts of the hive.
1975 mark_block_unused (hive_h *h, size_t offset)
1977 assert (h->writable);
1978 assert (IS_VALID_BLOCK (h, offset));
1981 fprintf (stderr, "mark_block_unused: marking 0x%zx unused\n", offset);
1983 struct ntreg_hbin_block *blockhdr =
1984 (struct ntreg_hbin_block *) (h->addr + offset);
1986 size_t seg_len = block_len (h, offset, NULL);
1987 blockhdr->seg_len = htole32 (seg_len);
1989 BITMAP_CLR (h->bitmap, offset);
1992 /* Delete all existing values at this node. */
1994 delete_values (hive_h *h, hive_node_h node)
1996 assert (h->writable);
1998 hive_value_h *values;
2000 if (get_values (h, node, &values, &blocks) == -1)
2004 for (i = 0; blocks[i] != 0; ++i)
2005 mark_block_unused (h, blocks[i]);
2009 for (i = 0; values[i] != 0; ++i) {
2010 struct ntreg_vk_record *vk =
2011 (struct ntreg_vk_record *) (h->addr + values[i]);
2015 len = le32toh (vk->data_len);
2016 is_inline = !!(len & 0x80000000); /* top bit indicates is inline */
2019 if (!is_inline) { /* non-inline, so remove data block */
2020 size_t data_offset = le32toh (vk->data_offset);
2021 data_offset += 0x1000;
2022 mark_block_unused (h, data_offset);
2025 /* remove vk record */
2026 mark_block_unused (h, values[i]);
2031 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2032 nk->nr_values = htole32 (0);
2033 nk->vallist = htole32 (0xffffffff);
2039 hivex_commit (hive_h *h, const char *filename, int flags)
2051 filename = filename ? : h->filename;
2052 int fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
2056 /* Update the header fields. */
2057 uint32_t sequence = le32toh (h->hdr->sequence1);
2059 h->hdr->sequence1 = htole32 (sequence);
2060 h->hdr->sequence2 = htole32 (sequence);
2061 /* XXX Ought to update h->hdr->last_modified. */
2062 h->hdr->blocks = htole32 (h->endpages - 0x1000);
2064 /* Recompute header checksum. */
2065 uint32_t sum = header_checksum (h);
2066 h->hdr->csum = htole32 (sum);
2069 fprintf (stderr, "hivex_commit: new header checksum: 0x%x\n", sum);
2071 if (full_write (fd, h->addr, h->size) != h->size) {
2078 if (close (fd) == -1)
2084 /* Calculate the hash for a lf or lh record offset.
2087 calc_hash (const char *type, const char *name, char *ret)
2089 size_t len = strlen (name);
2091 if (STRPREFIX (type, "lf"))
2092 /* Old-style, not used in current registries. */
2093 memcpy (ret, name, len < 4 ? len : 4);
2095 /* New-style for lh-records. */
2098 for (i = 0; i < len; ++i) {
2099 c = c_toupper (name[i]);
2103 *((uint32_t *) ret) = htole32 (h);
2107 /* Create a completely new lh-record containing just the single node. */
2109 new_lh_record (hive_h *h, const char *name, hive_node_h node)
2111 static const char id[2] = { 'l', 'h' };
2112 size_t seg_len = sizeof (struct ntreg_lf_record);
2113 size_t offset = allocate_block (h, seg_len, id);
2117 struct ntreg_lf_record *lh = (struct ntreg_lf_record *) (h->addr + offset);
2118 lh->nr_keys = htole16 (1);
2119 lh->keys[0].offset = htole32 (node - 0x1000);
2120 calc_hash ("lh", name, lh->keys[0].hash);
2125 /* Insert node into existing lf/lh-record at position.
2126 * This allocates a new record and marks the old one as unused.
2129 insert_lf_record (hive_h *h, size_t old_offs, size_t posn,
2130 const char *name, hive_node_h node)
2132 assert (IS_VALID_BLOCK (h, old_offs));
2134 /* Work around C stupidity.
2135 * http://www.redhat.com/archives/libguestfs/2010-February/msg00056.html
2137 int test = BLOCK_ID_EQ (h, old_offs, "lf") || BLOCK_ID_EQ (h, old_offs, "lh");
2140 struct ntreg_lf_record *old_lf =
2141 (struct ntreg_lf_record *) (h->addr + old_offs);
2142 size_t nr_keys = le16toh (old_lf->nr_keys);
2144 nr_keys++; /* in new record ... */
2146 size_t seg_len = sizeof (struct ntreg_lf_record) + (nr_keys-1) * 8;
2148 /* Copy the old_lf->id in case it moves during allocate_block. */
2150 memcpy (id, old_lf->id, sizeof id);
2152 size_t new_offs = allocate_block (h, seg_len, id);
2156 /* old_lf could have been invalidated by allocate_block. */
2157 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2159 struct ntreg_lf_record *new_lf =
2160 (struct ntreg_lf_record *) (h->addr + new_offs);
2161 new_lf->nr_keys = htole16 (nr_keys);
2163 /* Copy the keys until we reach posn, insert the new key there, then
2164 * copy the remaining keys.
2167 for (i = 0; i < posn; ++i)
2168 new_lf->keys[i] = old_lf->keys[i];
2170 new_lf->keys[i].offset = htole32 (node - 0x1000);
2171 calc_hash (new_lf->id, name, new_lf->keys[i].hash);
2173 for (i = posn+1; i < nr_keys; ++i)
2174 new_lf->keys[i] = old_lf->keys[i-1];
2176 /* Old block is unused, return new block. */
2177 mark_block_unused (h, old_offs);
2181 /* Compare name with name in nk-record. */
2183 compare_name_with_nk_name (hive_h *h, const char *name, hive_node_h nk_offs)
2185 assert (IS_VALID_BLOCK (h, nk_offs));
2186 assert (BLOCK_ID_EQ (h, nk_offs, "nk"));
2188 /* Name in nk is not necessarily nul-terminated. */
2189 char *nname = hivex_node_name (h, nk_offs);
2191 /* Unfortunately we don't have a way to return errors here. */
2193 perror ("compare_name_with_nk_name");
2197 int r = strcasecmp (name, nname);
2204 hivex_node_add_child (hive_h *h, hive_node_h parent, const char *name)
2211 if (!IS_VALID_BLOCK (h, parent) || !BLOCK_ID_EQ (h, parent, "nk")) {
2216 if (name == NULL || strlen (name) == 0) {
2221 if (hivex_node_get_child (h, parent, name) != 0) {
2226 /* Create the new nk-record. */
2227 static const char nk_id[2] = { 'n', 'k' };
2228 size_t seg_len = sizeof (struct ntreg_nk_record) + strlen (name);
2229 hive_node_h node = allocate_block (h, seg_len, nk_id);
2234 fprintf (stderr, "hivex_node_add_child: allocated new nk-record"
2235 " for child at 0x%zx\n", node);
2237 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2238 nk->flags = htole16 (0x0020); /* key is ASCII. */
2239 nk->parent = htole32 (parent - 0x1000);
2240 nk->subkey_lf = htole32 (0xffffffff);
2241 nk->subkey_lf_volatile = htole32 (0xffffffff);
2242 nk->vallist = htole32 (0xffffffff);
2243 nk->classname = htole32 (0xffffffff);
2244 nk->name_len = htole16 (strlen (name));
2245 strcpy (nk->name, name);
2247 /* Inherit parent sk. */
2248 struct ntreg_nk_record *parent_nk =
2249 (struct ntreg_nk_record *) (h->addr + parent);
2250 size_t parent_sk_offset = le32toh (parent_nk->sk);
2251 parent_sk_offset += 0x1000;
2252 if (!IS_VALID_BLOCK (h, parent_sk_offset) ||
2253 !BLOCK_ID_EQ (h, parent_sk_offset, "sk")) {
2255 fprintf (stderr, "hivex_node_add_child: returning EFAULT"
2256 " because parent sk is not a valid block (%zu)\n",
2261 struct ntreg_sk_record *sk =
2262 (struct ntreg_sk_record *) (h->addr + parent_sk_offset);
2263 sk->refcount = htole32 (le32toh (sk->refcount) + 1);
2264 nk->sk = htole32 (parent_sk_offset - 0x1000);
2266 /* Inherit parent timestamp. */
2267 memcpy (nk->timestamp, parent_nk->timestamp, sizeof (parent_nk->timestamp));
2269 /* What I found out the hard way (not documented anywhere): the
2270 * subkeys in lh-records must be kept sorted. If you just add a
2271 * subkey in a non-sorted position (eg. just add it at the end) then
2272 * Windows won't see the subkey _and_ Windows will corrupt the hive
2273 * itself when it modifies or saves it.
2275 * So use get_children() to get a list of intermediate
2276 * lf/lh-records. get_children() returns these in reading order
2277 * (which is sorted), so we look for the lf/lh-records in sequence
2278 * until we find the key name just after the one we are inserting,
2279 * and we insert the subkey just before it.
2281 * The only other case is the no-subkeys case, where we have to
2282 * create a brand new lh-record.
2284 hive_node_h *unused;
2287 if (get_children (h, parent, &unused, &blocks, 0) == -1)
2292 size_t nr_subkeys_in_parent_nk = le32toh (parent_nk->nr_subkeys);
2293 if (nr_subkeys_in_parent_nk == 0) { /* No subkeys case. */
2294 /* Free up any existing intermediate blocks. */
2295 for (i = 0; blocks[i] != 0; ++i)
2296 mark_block_unused (h, blocks[i]);
2297 size_t lh_offs = new_lh_record (h, name, node);
2303 /* Recalculate pointers that could have been invalidated by
2304 * previous call to allocate_block (via new_lh_record).
2306 nk = (struct ntreg_nk_record *) (h->addr + node);
2307 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2310 fprintf (stderr, "hivex_node_add_child: no keys, allocated new"
2311 " lh-record at 0x%zx\n", lh_offs);
2313 parent_nk->subkey_lf = htole32 (lh_offs - 0x1000);
2315 else { /* Insert subkeys case. */
2316 size_t old_offs = 0, new_offs = 0;
2317 struct ntreg_lf_record *old_lf = NULL;
2319 /* Find lf/lh key name just after the one we are inserting. */
2320 for (i = 0; blocks[i] != 0; ++i) {
2321 if (BLOCK_ID_EQ (h, blocks[i], "lf") ||
2322 BLOCK_ID_EQ (h, blocks[i], "lh")) {
2323 old_offs = blocks[i];
2324 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2325 for (j = 0; j < le16toh (old_lf->nr_keys); ++j) {
2326 hive_node_h nk_offs = le32toh (old_lf->keys[j].offset);
2328 if (compare_name_with_nk_name (h, name, nk_offs) < 0)
2334 /* Insert it at the end.
2335 * old_offs points to the last lf record, set j.
2337 assert (old_offs != 0); /* should never happen if nr_subkeys > 0 */
2338 j = le16toh (old_lf->nr_keys);
2343 fprintf (stderr, "hivex_node_add_child: insert key in existing"
2344 " lh-record at 0x%zx, posn %zu\n", old_offs, j);
2346 new_offs = insert_lf_record (h, old_offs, j, name, node);
2347 if (new_offs == 0) {
2352 /* Recalculate pointers that could have been invalidated by
2353 * previous call to allocate_block (via insert_lf_record).
2355 nk = (struct ntreg_nk_record *) (h->addr + node);
2356 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2359 fprintf (stderr, "hivex_node_add_child: new lh-record at 0x%zx\n",
2362 /* If the lf/lh-record was directly referenced by the parent nk,
2363 * then update the parent nk.
2365 if (le32toh (parent_nk->subkey_lf) + 0x1000 == old_offs)
2366 parent_nk->subkey_lf = htole32 (new_offs - 0x1000);
2367 /* Else we have to look for the intermediate ri-record and update
2371 for (i = 0; blocks[i] != 0; ++i) {
2372 if (BLOCK_ID_EQ (h, blocks[i], "ri")) {
2373 struct ntreg_ri_record *ri =
2374 (struct ntreg_ri_record *) (h->addr + blocks[i]);
2375 for (j = 0; j < le16toh (ri->nr_offsets); ++j)
2376 if (le32toh (ri->offset[j] + 0x1000) == old_offs) {
2377 ri->offset[j] = htole32 (new_offs - 0x1000);
2383 /* Not found .. This is an internal error. */
2385 fprintf (stderr, "hivex_node_add_child: returning ENOTSUP"
2386 " because could not find ri->lf link\n");
2398 /* Update nr_subkeys in parent nk. */
2399 nr_subkeys_in_parent_nk++;
2400 parent_nk->nr_subkeys = htole32 (nr_subkeys_in_parent_nk);
2402 /* Update max_subkey_name_len in parent nk. */
2403 uint16_t max = le16toh (parent_nk->max_subkey_name_len);
2404 if (max < strlen (name) * 2) /* *2 because "recoded" in UTF16-LE. */
2405 parent_nk->max_subkey_name_len = htole16 (strlen (name) * 2);
2410 /* Decrement the refcount of an sk-record, and if it reaches zero,
2411 * unlink it from the chain and delete it.
2414 delete_sk (hive_h *h, size_t sk_offset)
2416 if (!IS_VALID_BLOCK (h, sk_offset) || !BLOCK_ID_EQ (h, sk_offset, "sk")) {
2418 fprintf (stderr, "delete_sk: not an sk record: 0x%zx\n", sk_offset);
2423 struct ntreg_sk_record *sk = (struct ntreg_sk_record *) (h->addr + sk_offset);
2425 if (sk->refcount == 0) {
2427 fprintf (stderr, "delete_sk: sk record already has refcount 0: 0x%zx\n",
2435 if (sk->refcount == 0) {
2436 size_t sk_prev_offset = sk->sk_prev;
2437 sk_prev_offset += 0x1000;
2439 size_t sk_next_offset = sk->sk_next;
2440 sk_next_offset += 0x1000;
2442 /* Update sk_prev/sk_next SKs, unless they both point back to this
2443 * cell in which case we are deleting the last SK.
2445 if (sk_prev_offset != sk_offset && sk_next_offset != sk_offset) {
2446 struct ntreg_sk_record *sk_prev =
2447 (struct ntreg_sk_record *) (h->addr + sk_prev_offset);
2448 struct ntreg_sk_record *sk_next =
2449 (struct ntreg_sk_record *) (h->addr + sk_next_offset);
2451 sk_prev->sk_next = htole32 (sk_next_offset - 0x1000);
2452 sk_next->sk_prev = htole32 (sk_prev_offset - 0x1000);
2455 /* Refcount is zero so really delete this block. */
2456 mark_block_unused (h, sk_offset);
2462 /* Callback from hivex_node_delete_child which is called to delete a
2463 * node AFTER its subnodes have been visited. The subnodes have been
2464 * deleted but we still have to delete any lf/lh/li/ri records and the
2465 * value list block and values, followed by deleting the node itself.
2468 delete_node (hive_h *h, void *opaque, hive_node_h node, const char *name)
2470 /* Get the intermediate blocks. The subkeys have already been
2471 * deleted by this point, so tell get_children() not to check for
2472 * validity of the nk-records.
2474 hive_node_h *unused;
2476 if (get_children (h, node, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK) == -1)
2480 /* We don't care what's in these intermediate blocks, so we can just
2481 * delete them unconditionally.
2484 for (i = 0; blocks[i] != 0; ++i)
2485 mark_block_unused (h, blocks[i]);
2489 /* Delete the values in the node. */
2490 if (delete_values (h, node) == -1)
2493 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2495 /* If the NK references an SK, delete it. */
2496 size_t sk_offs = le32toh (nk->sk);
2497 if (sk_offs != 0xffffffff) {
2499 if (delete_sk (h, sk_offs) == -1)
2501 nk->sk = htole32 (0xffffffff);
2504 /* If the NK references a classname, delete it. */
2505 size_t cl_offs = le32toh (nk->classname);
2506 if (cl_offs != 0xffffffff) {
2508 mark_block_unused (h, cl_offs);
2509 nk->classname = htole32 (0xffffffff);
2512 /* Delete the node itself. */
2513 mark_block_unused (h, node);
2519 hivex_node_delete_child (hive_h *h, hive_node_h node)
2526 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2531 if (node == hivex_root (h)) {
2533 fprintf (stderr, "hivex_node_delete_child: cannot delete root node\n");
2538 hive_node_h parent = hivex_node_parent (h, node);
2542 /* Delete node and all its children and values recursively. */
2543 static const struct hivex_visitor visitor = { .node_end = delete_node };
2544 if (hivex_visit_node (h, node, &visitor, sizeof visitor, NULL, 0) == -1)
2547 /* Delete the link from parent to child. We need to find the lf/lh
2548 * record which contains the offset and remove the offset from that
2549 * record, then decrement the element count in that record, and
2550 * decrement the overall number of subkeys stored in the parent
2553 hive_node_h *unused;
2555 if (get_children (h, parent, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK)== -1)
2560 for (i = 0; blocks[i] != 0; ++i) {
2561 struct ntreg_hbin_block *block =
2562 (struct ntreg_hbin_block *) (h->addr + blocks[i]);
2564 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
2565 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
2567 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
2569 for (j = 0; j < nr_subkeys_in_lf; ++j)
2570 if (le32toh (lf->keys[j].offset) + 0x1000 == node) {
2571 for (; j < nr_subkeys_in_lf - 1; ++j)
2572 memcpy (&lf->keys[j], &lf->keys[j+1], sizeof (lf->keys[j]));
2573 lf->nr_keys = htole16 (nr_subkeys_in_lf - 1);
2579 fprintf (stderr, "hivex_node_delete_child: could not find parent"
2580 " to child link\n");
2585 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + parent);
2586 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
2587 nk->nr_subkeys = htole32 (nr_subkeys_in_nk - 1);
2590 fprintf (stderr, "hivex_node_delete_child: updating nr_subkeys"
2591 " in parent 0x%zx to %zu\n", parent, nr_subkeys_in_nk);
2597 hivex_node_set_values (hive_h *h, hive_node_h node,
2598 size_t nr_values, const hive_set_value *values,
2606 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2611 /* Delete all existing values. */
2612 if (delete_values (h, node) == -1)
2618 /* Allocate value list node. Value lists have no id field. */
2619 static const char nul_id[2] = { 0, 0 };
2621 sizeof (struct ntreg_value_list) + (nr_values - 1) * sizeof (uint32_t);
2622 size_t vallist_offs = allocate_block (h, seg_len, nul_id);
2623 if (vallist_offs == 0)
2626 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2627 nk->nr_values = htole32 (nr_values);
2628 nk->vallist = htole32 (vallist_offs - 0x1000);
2630 struct ntreg_value_list *vallist =
2631 (struct ntreg_value_list *) (h->addr + vallist_offs);
2634 for (i = 0; i < nr_values; ++i) {
2635 /* Allocate vk record to store this (key, value) pair. */
2636 static const char vk_id[2] = { 'v', 'k' };
2637 seg_len = sizeof (struct ntreg_vk_record) + strlen (values[i].key);
2638 size_t vk_offs = allocate_block (h, seg_len, vk_id);
2642 /* Recalculate pointers that could have been invalidated by
2643 * previous call to allocate_block.
2645 nk = (struct ntreg_nk_record *) (h->addr + node);
2646 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2648 vallist->offset[i] = htole32 (vk_offs - 0x1000);
2650 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2651 size_t name_len = strlen (values[i].key);
2652 vk->name_len = htole16 (name_len);
2653 strcpy (vk->name, values[i].key);
2654 vk->data_type = htole32 (values[i].t);
2655 uint32_t len = values[i].len;
2656 if (len <= 4) /* store it inline => set MSB flag */
2658 vk->data_len = htole32 (len);
2659 vk->flags = name_len == 0 ? 0 : 1;
2661 if (values[i].len <= 4) /* store it inline */
2662 memcpy (&vk->data_offset, values[i].value, values[i].len);
2664 size_t offs = allocate_block (h, values[i].len + 4, nul_id);
2668 /* Recalculate pointers that could have been invalidated by
2669 * previous call to allocate_block.
2671 nk = (struct ntreg_nk_record *) (h->addr + node);
2672 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2673 vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2675 memcpy (h->addr + offs + 4, values[i].value, values[i].len);
2676 vk->data_offset = htole32 (offs - 0x1000);
2679 if (name_len * 2 > le32toh (nk->max_vk_name_len))
2680 /* * 2 for UTF16-LE "reencoding" */
2681 nk->max_vk_name_len = htole32 (name_len * 2);
2682 if (values[i].len > le32toh (nk->max_vk_data_len))
2683 nk->max_vk_data_len = htole32 (values[i].len);
2690 hivex_node_set_value (hive_h *h, hive_node_h node,
2691 const hive_set_value *val, int flags)
2693 hive_value_h *prev_values = hivex_node_values (h, node);
2694 if (prev_values == NULL)
2699 size_t nr_values = 0;
2700 for (hive_value_h *itr = prev_values; *itr != 0; ++itr)
2703 hive_set_value *values = malloc ((nr_values + 1) * (sizeof (hive_set_value)));
2705 goto leave_prev_values;
2708 int idx_of_val = -1;
2709 hive_value_h *prev_val;
2710 for (prev_val = prev_values; *prev_val != 0; ++prev_val) {
2714 hive_set_value *value = &values[prev_val - prev_values];
2716 char *valval = hivex_value_value (h, *prev_val, &t, &len);
2717 if (valval == NULL) goto leave_partial;
2720 value->value = valval;
2724 char *valkey = hivex_value_key (h, *prev_val);
2725 if (valkey == NULL) goto leave_partial;
2728 value->key = valkey;
2730 if (STRCASEEQ (valkey, val->key))
2731 idx_of_val = prev_val - prev_values;
2734 if (idx_of_val > -1) {
2735 free (values[idx_of_val].key);
2736 free (values[idx_of_val].value);
2738 idx_of_val = nr_values;
2742 hive_set_value *value = &values[idx_of_val];
2743 *value = (hive_set_value){
2744 .key = strdup (val->key),
2745 .value = malloc (val->len),
2750 if (value->key == NULL || value->value == NULL) goto leave_partial;
2751 memcpy (value->value, val->value, val->len);
2753 retval = hivex_node_set_values (h, node, nr_values, values, 0);
2756 for (int i = 0; i < alloc_ct; i += 2) {
2757 free (values[i / 2].value);
2758 if (i + 1 < alloc_ct && values[i / 2].key != NULL)
2759 free (values[i / 2].key);