1 /* hivex - Windows Registry "hive" extraction library.
2 * Copyright (C) 2009-2010 Red Hat Inc.
3 * Derived from code by Petter Nordahl-Hagen under a compatible license:
4 * Copyright (c) 1997-2007 Petter Nordahl-Hagen.
5 * Derived from code by Markus Stephany under a compatible license:
6 * Copyright (c) 2000-2004, Markus Stephany.
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation;
11 * version 2.1 of the License.
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * See file LICENSE for the full license.
38 #include "full-read.h"
39 #include "full-write.h"
45 #define STREQ(a,b) (strcmp((a),(b)) == 0)
46 #define STRCASEEQ(a,b) (strcasecmp((a),(b)) == 0)
47 //#define STRNEQ(a,b) (strcmp((a),(b)) != 0)
48 //#define STRCASENEQ(a,b) (strcasecmp((a),(b)) != 0)
49 #define STREQLEN(a,b,n) (strncmp((a),(b),(n)) == 0)
50 //#define STRCASEEQLEN(a,b,n) (strncasecmp((a),(b),(n)) == 0)
51 //#define STRNEQLEN(a,b,n) (strncmp((a),(b),(n)) != 0)
52 //#define STRCASENEQLEN(a,b,n) (strncasecmp((a),(b),(n)) != 0)
53 #define STRPREFIX(a,b) (strncmp((a),(b),strlen((b))) == 0)
56 #include "byte_conversions.h"
58 /* These limits are in place to stop really stupid stuff and/or exploits. */
59 #define HIVEX_MAX_SUBKEYS 15000
60 #define HIVEX_MAX_VALUES 10000
61 #define HIVEX_MAX_VALUE_LEN 1000000
62 #define HIVEX_MAX_ALLOCATION 1000000
64 static char *windows_utf16_to_utf8 (/* const */ char *input, size_t len);
65 static size_t utf16_string_len_in_bytes_max (const char *str, size_t len);
74 /* Registry file, memory mapped if read-only, or malloc'd if writing. */
77 struct ntreg_header *hdr;
80 /* Use a bitmap to store which file offsets are valid (point to a
81 * used block). We only need to store 1 bit per 32 bits of the file
82 * (because blocks are 4-byte aligned). We found that the average
83 * block size in a registry file is ~50 bytes. So roughly 1 in 12
84 * bits in the bitmap will be set, making it likely a more efficient
85 * structure than a hash table.
88 #define BITMAP_SET(bitmap,off) (bitmap[(off)>>5] |= 1 << (((off)>>2)&7))
89 #define BITMAP_CLR(bitmap,off) (bitmap[(off)>>5] &= ~ (1 << (((off)>>2)&7)))
90 #define BITMAP_TST(bitmap,off) (bitmap[(off)>>5] & (1 << (((off)>>2)&7)))
91 #define IS_VALID_BLOCK(h,off) \
92 (((off) & 3) == 0 && \
94 (off) < (h)->size && \
95 BITMAP_TST((h)->bitmap,(off)))
97 /* Fields from the header, extracted from little-endianness hell. */
98 size_t rootoffs; /* Root key offset (always an nk-block). */
99 size_t endpages; /* Offset of end of pages. */
102 size_t endblocks; /* Offset to next block allocation (0
103 if not allocated anything yet). */
106 /* NB. All fields are little endian. */
107 struct ntreg_header {
108 char magic[4]; /* "regf" */
111 char last_modified[8];
112 uint32_t major_ver; /* 1 */
113 uint32_t minor_ver; /* 3 */
114 uint32_t unknown5; /* 0 */
115 uint32_t unknown6; /* 1 */
116 uint32_t offset; /* offset of root key record - 4KB */
117 uint32_t blocks; /* pointer AFTER last hbin in file - 4KB */
118 uint32_t unknown7; /* 1 */
120 char name[64]; /* original file name of hive */
121 char unknown_guid1[16];
122 char unknown_guid2[16];
125 char unknown_guid3[16];
130 uint32_t csum; /* checksum: xor of dwords 0-0x1fb. */
132 char unknown11[3528];
134 char unknown_guid4[16];
135 char unknown_guid5[16];
136 char unknown_guid6[16];
140 } __attribute__((__packed__));
142 struct ntreg_hbin_page {
143 char magic[4]; /* "hbin" */
144 uint32_t offset_first; /* offset from 1st block */
145 uint32_t page_size; /* size of this page (multiple of 4KB) */
147 /* Linked list of blocks follows here. */
148 } __attribute__((__packed__));
150 struct ntreg_hbin_block {
151 int32_t seg_len; /* length of this block (-ve for used block) */
152 char id[2]; /* the block type (eg. "nk" for nk record) */
153 /* Block data follows here. */
154 } __attribute__((__packed__));
156 #define BLOCK_ID_EQ(h,offs,eqid) \
157 (STREQLEN (((struct ntreg_hbin_block *)((h)->addr + (offs)))->id, (eqid), 2))
160 block_len (hive_h *h, size_t blkoff, int *used)
162 struct ntreg_hbin_block *block;
163 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
165 int32_t len = le32toh (block->seg_len);
176 struct ntreg_nk_record {
177 int32_t seg_len; /* length (always -ve because used) */
178 char id[2]; /* "nk" */
182 uint32_t parent; /* offset of owner/parent */
183 uint32_t nr_subkeys; /* number of subkeys */
184 uint32_t nr_subkeys_volatile;
185 uint32_t subkey_lf; /* lf record containing list of subkeys */
186 uint32_t subkey_lf_volatile;
187 uint32_t nr_values; /* number of values */
188 uint32_t vallist; /* value-list record */
189 uint32_t sk; /* offset of sk-record */
190 uint32_t classname; /* offset of classname record */
191 uint16_t max_subkey_name_len; /* maximum length of a subkey name in bytes
192 if the subkey was reencoded as UTF-16LE */
195 uint32_t max_vk_name_len; /* maximum length of any vk name in bytes
196 if the name was reencoded as UTF-16LE */
197 uint32_t max_vk_data_len; /* maximum length of any vk data in bytes */
199 uint16_t name_len; /* length of name */
200 uint16_t classname_len; /* length of classname */
201 char name[1]; /* name follows here */
202 } __attribute__((__packed__));
204 struct ntreg_lf_record {
206 char id[2]; /* "lf"|"lh" */
207 uint16_t nr_keys; /* number of keys in this record */
209 uint32_t offset; /* offset of nk-record for this subkey */
210 char hash[4]; /* hash of subkey name */
212 } __attribute__((__packed__));
214 struct ntreg_ri_record {
216 char id[2]; /* "ri" */
217 uint16_t nr_offsets; /* number of pointers to lh records */
218 uint32_t offset[1]; /* list of pointers to lh records */
219 } __attribute__((__packed__));
221 /* This has no ID header. */
222 struct ntreg_value_list {
224 uint32_t offset[1]; /* list of pointers to vk records */
225 } __attribute__((__packed__));
227 struct ntreg_vk_record {
228 int32_t seg_len; /* length (always -ve because used) */
229 char id[2]; /* "vk" */
230 uint16_t name_len; /* length of name */
231 /* length of the data:
232 * If data_len is <= 4, then it's stored inline.
233 * Top bit is set to indicate inline.
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);
272 #define HIVEX_OPEN_MSGLVL_MASK (HIVEX_OPEN_VERBOSE|HIVEX_OPEN_DEBUG)
275 hivex_open (const char *filename, int flags)
279 assert (sizeof (struct ntreg_header) == 0x1000);
280 assert (offsetof (struct ntreg_header, csum) == 0x1fc);
282 h = calloc (1, sizeof *h);
286 h->msglvl = flags & HIVEX_OPEN_MSGLVL_MASK;
288 const char *debug = getenv ("HIVEX_DEBUG");
289 if (debug && STREQ (debug, "1"))
293 fprintf (stderr, "hivex_open: created handle %p\n", h);
295 h->writable = !!(flags & HIVEX_OPEN_WRITE);
296 h->filename = strdup (filename);
297 if (h->filename == NULL)
300 h->fd = open (filename, O_RDONLY | O_CLOEXEC);
305 if (fstat (h->fd, &statbuf) == -1)
308 h->size = statbuf.st_size;
311 h->addr = mmap (NULL, h->size, PROT_READ, MAP_SHARED, h->fd, 0);
312 if (h->addr == MAP_FAILED)
316 fprintf (stderr, "hivex_open: mapped file at %p\n", h->addr);
318 h->addr = malloc (h->size);
322 if (full_read (h->fd, h->addr, h->size) < h->size)
327 if (h->hdr->magic[0] != 'r' ||
328 h->hdr->magic[1] != 'e' ||
329 h->hdr->magic[2] != 'g' ||
330 h->hdr->magic[3] != 'f') {
331 fprintf (stderr, "hivex: %s: not a Windows NT Registry hive file\n",
337 /* Check major version. */
338 uint32_t major_ver = le32toh (h->hdr->major_ver);
339 if (major_ver != 1) {
341 "hivex: %s: hive file major version %" PRIu32 " (expected 1)\n",
342 filename, major_ver);
347 h->bitmap = calloc (1 + h->size / 32, 1);
348 if (h->bitmap == NULL)
351 /* Header checksum. */
352 uint32_t sum = header_checksum (h);
353 if (sum != le32toh (h->hdr->csum)) {
354 fprintf (stderr, "hivex: %s: bad checksum in hive header\n", filename);
359 if (h->msglvl >= 2) {
360 char *name = windows_utf16_to_utf8 (h->hdr->name, 64);
363 "hivex_open: header fields:\n"
364 " file version %" PRIu32 ".%" PRIu32 "\n"
365 " sequence nos %" PRIu32 " %" PRIu32 "\n"
366 " (sequences nos should match if hive was synched at shutdown)\n"
367 " original file name %s\n"
368 " (only 32 chars are stored, name is probably truncated)\n"
369 " root offset 0x%x + 0x1000\n"
370 " end of last page 0x%x + 0x1000 (total file size 0x%zx)\n"
371 " checksum 0x%x (calculated 0x%x)\n",
372 major_ver, le32toh (h->hdr->minor_ver),
373 le32toh (h->hdr->sequence1), le32toh (h->hdr->sequence2),
374 name ? name : "(conversion failed)",
375 le32toh (h->hdr->offset),
376 le32toh (h->hdr->blocks), h->size,
377 le32toh (h->hdr->csum), sum);
381 h->rootoffs = le32toh (h->hdr->offset) + 0x1000;
382 h->endpages = le32toh (h->hdr->blocks) + 0x1000;
385 fprintf (stderr, "hivex_open: root offset = 0x%zx\n", h->rootoffs);
387 /* We'll set this flag when we see a block with the root offset (ie.
390 int seen_root_block = 0, bad_root_block = 0;
392 /* Collect some stats. */
393 size_t pages = 0; /* Number of hbin pages read. */
394 size_t smallest_page = SIZE_MAX, largest_page = 0;
395 size_t blocks = 0; /* Total number of blocks found. */
396 size_t smallest_block = SIZE_MAX, largest_block = 0, blocks_bytes = 0;
397 size_t used_blocks = 0; /* Total number of used blocks found. */
398 size_t used_size = 0; /* Total size (bytes) of used blocks. */
400 /* Read the pages and blocks. The aim here is to be robust against
401 * corrupt or malicious registries. So we make sure the loops
402 * always make forward progress. We add the address of each block
403 * we read to a hash table so pointers will only reference the start
407 struct ntreg_hbin_page *page;
408 for (off = 0x1000; off < h->size; off += le32toh (page->page_size)) {
409 if (off >= h->endpages)
412 page = (struct ntreg_hbin_page *) (h->addr + off);
413 if (page->magic[0] != 'h' ||
414 page->magic[1] != 'b' ||
415 page->magic[2] != 'i' ||
416 page->magic[3] != 'n') {
417 fprintf (stderr, "hivex: %s: trailing garbage at end of file (at 0x%zx, after %zu pages)\n",
418 filename, off, pages);
423 size_t page_size = le32toh (page->page_size);
425 fprintf (stderr, "hivex_open: page at 0x%zx, size %zu\n", off, page_size);
427 if (page_size < smallest_page) smallest_page = page_size;
428 if (page_size > largest_page) largest_page = page_size;
430 if (page_size <= sizeof (struct ntreg_hbin_page) ||
431 (page_size & 0x0fff) != 0) {
432 fprintf (stderr, "hivex: %s: page size %zu at 0x%zx, bad registry\n",
433 filename, page_size, off);
438 /* Read the blocks in this page. */
440 struct ntreg_hbin_block *block;
442 for (blkoff = off + 0x20;
443 blkoff < off + page_size;
447 int is_root = blkoff == h->rootoffs;
451 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
453 seg_len = block_len (h, blkoff, &used);
454 if (seg_len <= 4 || (seg_len & 3) != 0) {
455 fprintf (stderr, "hivex: %s: block size %" PRIu32 " at 0x%zx, bad registry\n",
456 filename, le32toh (block->seg_len), blkoff);
462 fprintf (stderr, "hivex_open: %s block id %d,%d at 0x%zx size %zu%s\n",
463 used ? "used" : "free", block->id[0], block->id[1], blkoff,
464 seg_len, is_root ? " (root)" : "");
466 blocks_bytes += seg_len;
467 if (seg_len < smallest_block) smallest_block = seg_len;
468 if (seg_len > largest_block) largest_block = seg_len;
470 if (is_root && !used)
475 used_size += seg_len;
477 /* Root block must be an nk-block. */
478 if (is_root && (block->id[0] != 'n' || block->id[1] != 'k'))
481 /* Note this blkoff is a valid address. */
482 BITMAP_SET (h->bitmap, blkoff);
487 if (!seen_root_block) {
488 fprintf (stderr, "hivex: %s: no root block found\n", filename);
493 if (bad_root_block) {
494 fprintf (stderr, "hivex: %s: bad root block (free or not nk)\n", filename);
501 "hivex_open: successfully read Windows Registry hive file:\n"
502 " pages: %zu [sml: %zu, lge: %zu]\n"
503 " blocks: %zu [sml: %zu, avg: %zu, lge: %zu]\n"
504 " blocks used: %zu\n"
505 " bytes used: %zu\n",
506 pages, smallest_page, largest_page,
507 blocks, smallest_block, blocks_bytes / blocks, largest_block,
508 used_blocks, used_size);
516 if (h->addr && h->size && h->addr != MAP_FAILED) {
518 munmap (h->addr, h->size);
532 hivex_close (hive_h *h)
537 fprintf (stderr, "hivex_close\n");
541 munmap (h->addr, h->size);
551 /*----------------------------------------------------------------------
556 hivex_root (hive_h *h)
558 hive_node_h ret = h->rootoffs;
559 if (!IS_VALID_BLOCK (h, ret)) {
567 hivex_node_name (hive_h *h, hive_node_h node)
569 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
574 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
576 /* AFAIK the node name is always plain ASCII, so no conversion
577 * to UTF-8 is necessary. However we do need to nul-terminate
581 /* nk->name_len is unsigned, 16 bit, so this is safe ... However
582 * we have to make sure the length doesn't exceed the block length.
584 size_t len = le16toh (nk->name_len);
585 size_t seg_len = block_len (h, node, NULL);
586 if (sizeof (struct ntreg_nk_record) + len - 1 > seg_len) {
588 fprintf (stderr, "hivex_node_name: returning EFAULT because node name is too long (%zu, %zu)\n",
594 char *ret = malloc (len + 1);
597 memcpy (ret, nk->name, len);
603 /* I think the documentation for the sk and classname fields in the nk
604 * record is wrong, or else the offset field is in the wrong place.
605 * Otherwise this makes no sense. Disabled this for now -- it's not
606 * useful for reading the registry anyway.
610 hivex_node_security (hive_h *h, hive_node_h node)
612 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
617 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
619 hive_node_h ret = le32toh (nk->sk);
621 if (!IS_VALID_BLOCK (h, ret)) {
629 hivex_node_classname (hive_h *h, hive_node_h node)
631 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
636 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
638 hive_node_h ret = le32toh (nk->classname);
640 if (!IS_VALID_BLOCK (h, ret)) {
648 /* Structure for returning 0-terminated lists of offsets (nodes,
658 init_offset_list (struct offset_list *list)
662 list->offsets = NULL;
665 #define INIT_OFFSET_LIST(name) \
666 struct offset_list name; \
667 init_offset_list (&name)
669 /* Preallocates the offset_list, but doesn't make the contents longer. */
671 grow_offset_list (struct offset_list *list, size_t alloc)
673 assert (alloc >= list->len);
674 size_t *p = realloc (list->offsets, alloc * sizeof (size_t));
683 add_to_offset_list (struct offset_list *list, size_t offset)
685 if (list->len >= list->alloc) {
686 if (grow_offset_list (list, list->alloc ? list->alloc * 2 : 4) == -1)
689 list->offsets[list->len] = offset;
695 free_offset_list (struct offset_list *list)
697 free (list->offsets);
701 return_offset_list (struct offset_list *list)
703 if (add_to_offset_list (list, 0) == -1)
705 return list->offsets; /* caller frees */
708 /* Iterate over children, returning child nodes and intermediate blocks. */
709 #define GET_CHILDREN_NO_CHECK_NK 1
712 get_children (hive_h *h, hive_node_h node,
713 hive_node_h **children_ret, size_t **blocks_ret,
716 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
721 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
723 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
725 INIT_OFFSET_LIST (children);
726 INIT_OFFSET_LIST (blocks);
728 /* Deal with the common "no subkeys" case quickly. */
729 if (nr_subkeys_in_nk == 0)
732 /* Arbitrarily limit the number of subkeys we will ever deal with. */
733 if (nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS) {
735 fprintf (stderr, "hivex: get_children: returning ERANGE because nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS (%zu > %d)\n",
736 nr_subkeys_in_nk, HIVEX_MAX_SUBKEYS);
741 /* Preallocate space for the children. */
742 if (grow_offset_list (&children, nr_subkeys_in_nk) == -1)
745 /* The subkey_lf field can point either to an lf-record, which is
746 * the common case, or if there are lots of subkeys, to an
749 size_t subkey_lf = le32toh (nk->subkey_lf);
751 if (!IS_VALID_BLOCK (h, subkey_lf)) {
753 fprintf (stderr, "hivex_node_children: returning EFAULT because subkey_lf is not a valid block (0x%zx)\n",
759 if (add_to_offset_list (&blocks, subkey_lf) == -1)
762 struct ntreg_hbin_block *block =
763 (struct ntreg_hbin_block *) (h->addr + subkey_lf);
765 /* Points to lf-record? (Note, also "lh" but that is basically the
766 * same as "lf" as far as we are concerned here).
768 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
769 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
771 /* Check number of subkeys in the nk-record matches number of subkeys
774 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
777 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu, nr_subkeys_in_lf = %zu\n",
778 nr_subkeys_in_nk, nr_subkeys_in_lf);
780 if (nr_subkeys_in_nk != nr_subkeys_in_lf) {
785 size_t len = block_len (h, subkey_lf, NULL);
786 if (8 + nr_subkeys_in_lf * 8 > len) {
788 fprintf (stderr, "hivex_node_children: returning EFAULT because too many subkeys (%zu, %zu)\n",
789 nr_subkeys_in_lf, len);
795 for (i = 0; i < nr_subkeys_in_lf; ++i) {
796 hive_node_h subkey = le32toh (lf->keys[i].offset);
798 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
799 if (!IS_VALID_BLOCK (h, subkey)) {
801 fprintf (stderr, "hivex_node_children: returning EFAULT because subkey is not a valid block (0x%zx)\n",
807 if (add_to_offset_list (&children, subkey) == -1)
812 /* Points to ri-record? */
813 else if (block->id[0] == 'r' && block->id[1] == 'i') {
814 struct ntreg_ri_record *ri = (struct ntreg_ri_record *) block;
816 size_t nr_offsets = le16toh (ri->nr_offsets);
818 /* Count total number of children. */
820 for (i = 0; i < nr_offsets; ++i) {
821 hive_node_h offset = le32toh (ri->offset[i]);
823 if (!IS_VALID_BLOCK (h, offset)) {
825 fprintf (stderr, "hivex_node_children: returning EFAULT because ri-offset is not a valid block (0x%zx)\n",
830 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
832 fprintf (stderr, "get_children: returning ENOTSUP because ri-record offset does not point to lf/lh (0x%zx)\n",
838 if (add_to_offset_list (&blocks, offset) == -1)
841 struct ntreg_lf_record *lf =
842 (struct ntreg_lf_record *) (h->addr + offset);
844 count += le16toh (lf->nr_keys);
848 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu, counted = %zu\n",
849 nr_subkeys_in_nk, count);
851 if (nr_subkeys_in_nk != count) {
856 /* Copy list of children. Note nr_subkeys_in_nk is limited to
857 * something reasonable above.
859 for (i = 0; i < nr_offsets; ++i) {
860 hive_node_h offset = le32toh (ri->offset[i]);
862 if (!IS_VALID_BLOCK (h, offset)) {
864 fprintf (stderr, "hivex_node_children: returning EFAULT because ri-offset is not a valid block (0x%zx)\n",
869 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
871 fprintf (stderr, "get_children: returning ENOTSUP because ri-record offset does not point to lf/lh (0x%zx)\n",
877 struct ntreg_lf_record *lf =
878 (struct ntreg_lf_record *) (h->addr + offset);
881 for (j = 0; j < le16toh (lf->nr_keys); ++j) {
882 hive_node_h subkey = le32toh (lf->keys[j].offset);
884 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
885 if (!IS_VALID_BLOCK (h, subkey)) {
887 fprintf (stderr, "hivex_node_children: returning EFAULT because indirect subkey is not a valid block (0x%zx)\n",
893 if (add_to_offset_list (&children, subkey) == -1)
899 /* else not supported, set errno and fall through */
901 fprintf (stderr, "get_children: returning ENOTSUP because subkey block is not lf/lh/ri (0x%zx, %d, %d)\n",
902 subkey_lf, block->id[0], block->id[1]);
905 free_offset_list (&children);
906 free_offset_list (&blocks);
910 *children_ret = return_offset_list (&children);
911 *blocks_ret = return_offset_list (&blocks);
912 if (!*children_ret || !*blocks_ret)
918 hivex_node_children (hive_h *h, hive_node_h node)
920 hive_node_h *children;
923 if (get_children (h, node, &children, &blocks, 0) == -1)
930 /* Very inefficient, but at least having a separate API call
931 * allows us to make it more efficient in future.
934 hivex_node_get_child (hive_h *h, hive_node_h node, const char *nname)
936 hive_node_h *children = NULL;
940 children = hivex_node_children (h, node);
941 if (!children) goto error;
944 for (i = 0; children[i] != 0; ++i) {
945 name = hivex_node_name (h, children[i]);
946 if (!name) goto error;
947 if (STRCASEEQ (name, nname)) {
951 free (name); name = NULL;
961 hivex_node_parent (hive_h *h, hive_node_h node)
963 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
968 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
970 hive_node_h ret = le32toh (nk->parent);
972 if (!IS_VALID_BLOCK (h, ret)) {
974 fprintf (stderr, "hivex_node_parent: returning EFAULT because parent is not a valid block (0x%zx)\n",
983 get_values (hive_h *h, hive_node_h node,
984 hive_value_h **values_ret, size_t **blocks_ret)
986 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
991 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
993 size_t nr_values = le32toh (nk->nr_values);
996 fprintf (stderr, "hivex_node_values: nr_values = %zu\n", nr_values);
998 INIT_OFFSET_LIST (values);
999 INIT_OFFSET_LIST (blocks);
1001 /* Deal with the common "no values" case quickly. */
1005 /* Arbitrarily limit the number of values we will ever deal with. */
1006 if (nr_values > HIVEX_MAX_VALUES) {
1008 fprintf (stderr, "hivex: get_values: returning ERANGE because nr_values > HIVEX_MAX_VALUES (%zu > %d)\n",
1009 nr_values, HIVEX_MAX_VALUES);
1014 /* Preallocate space for the values. */
1015 if (grow_offset_list (&values, nr_values) == -1)
1018 /* Get the value list and check it looks reasonable. */
1019 size_t vlist_offset = le32toh (nk->vallist);
1020 vlist_offset += 0x1000;
1021 if (!IS_VALID_BLOCK (h, vlist_offset)) {
1023 fprintf (stderr, "hivex_node_values: returning EFAULT because value list is not a valid block (0x%zx)\n",
1029 if (add_to_offset_list (&blocks, vlist_offset) == -1)
1032 struct ntreg_value_list *vlist =
1033 (struct ntreg_value_list *) (h->addr + vlist_offset);
1035 size_t len = block_len (h, vlist_offset, NULL);
1036 if (4 + nr_values * 4 > len) {
1038 fprintf (stderr, "hivex_node_values: returning EFAULT because value list is too long (%zu, %zu)\n",
1045 for (i = 0; i < nr_values; ++i) {
1046 hive_node_h value = vlist->offset[i];
1048 if (!IS_VALID_BLOCK (h, value)) {
1050 fprintf (stderr, "hivex_node_values: returning EFAULT because value is not a valid block (0x%zx)\n",
1055 if (add_to_offset_list (&values, value) == -1)
1060 *values_ret = return_offset_list (&values);
1061 *blocks_ret = return_offset_list (&blocks);
1062 if (!*values_ret || !*blocks_ret)
1067 free_offset_list (&values);
1068 free_offset_list (&blocks);
1073 hivex_node_values (hive_h *h, hive_node_h node)
1075 hive_value_h *values;
1078 if (get_values (h, node, &values, &blocks) == -1)
1085 /* Very inefficient, but at least having a separate API call
1086 * allows us to make it more efficient in future.
1089 hivex_node_get_value (hive_h *h, hive_node_h node, const char *key)
1091 hive_value_h *values = NULL;
1093 hive_value_h ret = 0;
1095 values = hivex_node_values (h, node);
1096 if (!values) goto error;
1099 for (i = 0; values[i] != 0; ++i) {
1100 name = hivex_value_key (h, values[i]);
1101 if (!name) goto error;
1102 if (STRCASEEQ (name, key)) {
1106 free (name); name = NULL;
1116 hivex_value_key (hive_h *h, hive_value_h value)
1118 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1123 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1125 /* AFAIK the key is always plain ASCII, so no conversion to UTF-8 is
1126 * necessary. However we do need to nul-terminate the string.
1129 /* vk->name_len is unsigned, 16 bit, so this is safe ... However
1130 * we have to make sure the length doesn't exceed the block length.
1132 size_t len = le16toh (vk->name_len);
1133 size_t seg_len = block_len (h, value, NULL);
1134 if (sizeof (struct ntreg_vk_record) + len - 1 > seg_len) {
1136 fprintf (stderr, "hivex_value_key: returning EFAULT because key length is too long (%zu, %zu)\n",
1142 char *ret = malloc (len + 1);
1145 memcpy (ret, vk->name, len);
1151 hivex_value_type (hive_h *h, hive_value_h value, hive_type *t, size_t *len)
1153 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1158 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1161 *t = le32toh (vk->data_type);
1164 *len = le32toh (vk->data_len);
1165 *len &= 0x7fffffff; /* top bit indicates if data is stored inline */
1172 hivex_value_value (hive_h *h, hive_value_h value,
1173 hive_type *t_rtn, size_t *len_rtn)
1175 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1180 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1186 t = le32toh (vk->data_type);
1188 len = le32toh (vk->data_len);
1189 is_inline = !!(len & 0x80000000);
1193 fprintf (stderr, "hivex_value_value: value=0x%zx, t=%d, len=%zu, inline=%d\n",
1194 value, t, len, is_inline);
1201 if (is_inline && len > 4) {
1206 /* Arbitrarily limit the length that we will read. */
1207 if (len > HIVEX_MAX_VALUE_LEN) {
1209 fprintf (stderr, "hivex_value_value: returning ERANGE because data length > HIVEX_MAX_VALUE_LEN (%zu > %d)\n",
1210 len, HIVEX_MAX_SUBKEYS);
1215 char *ret = malloc (len);
1220 memcpy (ret, (char *) &vk->data_offset, len);
1224 size_t data_offset = le32toh (vk->data_offset);
1225 data_offset += 0x1000;
1226 if (!IS_VALID_BLOCK (h, data_offset)) {
1228 fprintf (stderr, "hivex_value_value: returning EFAULT because data offset is not a valid block (0x%zx)\n",
1235 /* Check that the declared size isn't larger than the block its in.
1237 * XXX Some apparently valid registries are seen to have this,
1238 * so turn this into a warning and substitute the smaller length
1241 size_t blen = block_len (h, data_offset, NULL);
1242 if (len > blen - 4 /* subtract 4 for block header */) {
1244 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",
1245 data_offset, len, blen);
1248 /* Return the smaller length to the caller too. */
1253 char *data = h->addr + data_offset + 4;
1254 memcpy (ret, data, len);
1259 windows_utf16_to_utf8 (/* const */ char *input, size_t len)
1261 iconv_t ic = iconv_open ("UTF-8", "UTF-16");
1262 if (ic == (iconv_t) -1)
1265 /* iconv(3) has an insane interface ... */
1267 /* Mostly UTF-8 will be smaller, so this is a good initial guess. */
1268 size_t outalloc = len;
1272 size_t outlen = outalloc;
1273 char *out = malloc (outlen + 1);
1283 size_t r = iconv (ic, &inp, &inlen, &outp, &outlen);
1284 if (r == (size_t) -1) {
1285 if (errno == E2BIG) {
1287 size_t prev = outalloc;
1288 /* Try again with a larger output buffer. */
1291 if (outalloc < prev) {
1299 /* Else some conversion failure, eg. EILSEQ, EINVAL. */
1315 hivex_value_string (hive_h *h, hive_value_h value)
1319 char *data = hivex_value_value (h, value, &t, &len);
1324 if (t != hive_t_string && t != hive_t_expand_string && t != hive_t_link) {
1330 /* Deal with the case where Windows has allocated a large buffer
1331 * full of random junk, and only the first few bytes of the buffer
1332 * contain a genuine UTF-16 string.
1334 * In this case, iconv would try to process the junk bytes as UTF-16
1335 * and inevitably find an illegal sequence (EILSEQ). Instead, stop
1336 * after we find the first \0\0.
1338 * (Found by Hilko Bengen in a fresh Windows XP SOFTWARE hive).
1340 size_t slen = utf16_string_len_in_bytes_max (data, len);
1344 char *ret = windows_utf16_to_utf8 (data, len);
1353 free_strings (char **argv)
1358 for (i = 0; argv[i] != NULL; ++i)
1364 /* Get the length of a UTF-16 format string. Handle the string as
1365 * pairs of bytes, looking for the first \0\0 pair. Only read up to
1366 * 'len' maximum bytes.
1369 utf16_string_len_in_bytes_max (const char *str, size_t len)
1373 while (len >= 2 && (str[0] || str[1])) {
1382 /* http://blogs.msdn.com/oldnewthing/archive/2009/10/08/9904646.aspx */
1384 hivex_value_multiple_strings (hive_h *h, hive_value_h value)
1388 char *data = hivex_value_value (h, value, &t, &len);
1393 if (t != hive_t_multiple_strings) {
1399 size_t nr_strings = 0;
1400 char **ret = malloc ((1 + nr_strings) * sizeof (char *));
1410 while (p < data + len &&
1411 (plen = utf16_string_len_in_bytes_max (p, data + len - p)) > 0) {
1413 char **ret2 = realloc (ret, (1 + nr_strings) * sizeof (char *));
1421 ret[nr_strings-1] = windows_utf16_to_utf8 (p, plen);
1422 ret[nr_strings] = NULL;
1423 if (ret[nr_strings-1] == NULL) {
1429 p += plen + 2 /* skip over UTF-16 \0\0 at the end of this string */;
1437 hivex_value_dword (hive_h *h, hive_value_h value)
1441 char *data = hivex_value_value (h, value, &t, &len);
1446 if ((t != hive_t_dword && t != hive_t_dword_be) || len != 4) {
1452 int32_t ret = *(int32_t*)data;
1454 if (t == hive_t_dword) /* little endian */
1455 ret = le32toh (ret);
1457 ret = be32toh (ret);
1463 hivex_value_qword (hive_h *h, hive_value_h value)
1467 char *data = hivex_value_value (h, value, &t, &len);
1472 if (t != hive_t_qword || len != 8) {
1478 int64_t ret = *(int64_t*)data;
1480 ret = le64toh (ret); /* always little endian */
1485 /*----------------------------------------------------------------------
1490 hivex_visit (hive_h *h, const struct hivex_visitor *visitor, size_t len,
1491 void *opaque, int flags)
1493 return hivex_visit_node (h, hivex_root (h), visitor, len, opaque, flags);
1496 static int hivex__visit_node (hive_h *h, hive_node_h node, const struct hivex_visitor *vtor, char *unvisited, void *opaque, int flags);
1499 hivex_visit_node (hive_h *h, hive_node_h node,
1500 const struct hivex_visitor *visitor, size_t len, void *opaque,
1503 struct hivex_visitor vtor;
1504 memset (&vtor, 0, sizeof vtor);
1506 /* Note that len might be larger *or smaller* than the expected size. */
1507 size_t copysize = len <= sizeof vtor ? len : sizeof vtor;
1508 memcpy (&vtor, visitor, copysize);
1510 /* This bitmap records unvisited nodes, so we don't loop if the
1511 * registry contains cycles.
1513 char *unvisited = malloc (1 + h->size / 32);
1514 if (unvisited == NULL)
1516 memcpy (unvisited, h->bitmap, 1 + h->size / 32);
1518 int r = hivex__visit_node (h, node, &vtor, unvisited, opaque, flags);
1524 hivex__visit_node (hive_h *h, hive_node_h node,
1525 const struct hivex_visitor *vtor, char *unvisited,
1526 void *opaque, int flags)
1528 int skip_bad = flags & HIVEX_VISIT_SKIP_BAD;
1530 hive_value_h *values = NULL;
1531 hive_node_h *children = NULL;
1537 /* Return -1 on all callback errors. However on internal errors,
1538 * check if skip_bad is set and suppress those errors if so.
1542 if (!BITMAP_TST (unvisited, node)) {
1544 fprintf (stderr, "hivex__visit_node: contains cycle: visited node 0x%zx already\n",
1548 return skip_bad ? 0 : -1;
1550 BITMAP_CLR (unvisited, node);
1552 name = hivex_node_name (h, node);
1553 if (!name) return skip_bad ? 0 : -1;
1554 if (vtor->node_start && vtor->node_start (h, opaque, node, name) == -1)
1557 values = hivex_node_values (h, node);
1559 ret = skip_bad ? 0 : -1;
1563 for (i = 0; values[i] != 0; ++i) {
1567 if (hivex_value_type (h, values[i], &t, &len) == -1) {
1568 ret = skip_bad ? 0 : -1;
1572 key = hivex_value_key (h, values[i]);
1574 ret = skip_bad ? 0 : -1;
1578 if (vtor->value_any) {
1579 str = hivex_value_value (h, values[i], &t, &len);
1581 ret = skip_bad ? 0 : -1;
1584 if (vtor->value_any (h, opaque, node, values[i], t, len, key, str) == -1)
1586 free (str); str = NULL;
1591 str = hivex_value_value (h, values[i], &t, &len);
1593 ret = skip_bad ? 0 : -1;
1596 if (t != hive_t_none) {
1597 ret = skip_bad ? 0 : -1;
1600 if (vtor->value_none &&
1601 vtor->value_none (h, opaque, node, values[i], t, len, key, str) == -1)
1603 free (str); str = NULL;
1607 case hive_t_expand_string:
1609 str = hivex_value_string (h, values[i]);
1611 if (errno != EILSEQ && errno != EINVAL) {
1612 ret = skip_bad ? 0 : -1;
1615 if (vtor->value_string_invalid_utf16) {
1616 str = hivex_value_value (h, values[i], &t, &len);
1617 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i], t, len, key, str) == -1)
1619 free (str); str = NULL;
1623 if (vtor->value_string &&
1624 vtor->value_string (h, opaque, node, values[i], t, len, key, str) == -1)
1626 free (str); str = NULL;
1630 case hive_t_dword_be: {
1631 int32_t i32 = hivex_value_dword (h, values[i]);
1632 if (vtor->value_dword &&
1633 vtor->value_dword (h, opaque, node, values[i], t, len, key, i32) == -1)
1638 case hive_t_qword: {
1639 int64_t i64 = hivex_value_qword (h, values[i]);
1640 if (vtor->value_qword &&
1641 vtor->value_qword (h, opaque, node, values[i], t, len, key, i64) == -1)
1647 str = hivex_value_value (h, values[i], &t, &len);
1649 ret = skip_bad ? 0 : -1;
1652 if (t != hive_t_binary) {
1653 ret = skip_bad ? 0 : -1;
1656 if (vtor->value_binary &&
1657 vtor->value_binary (h, opaque, node, values[i], t, len, key, str) == -1)
1659 free (str); str = NULL;
1662 case hive_t_multiple_strings:
1663 strs = hivex_value_multiple_strings (h, values[i]);
1665 if (errno != EILSEQ && errno != EINVAL) {
1666 ret = skip_bad ? 0 : -1;
1669 if (vtor->value_string_invalid_utf16) {
1670 str = hivex_value_value (h, values[i], &t, &len);
1671 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i], t, len, key, str) == -1)
1673 free (str); str = NULL;
1677 if (vtor->value_multiple_strings &&
1678 vtor->value_multiple_strings (h, opaque, node, values[i], t, len, key, strs) == -1)
1680 free_strings (strs); strs = NULL;
1683 case hive_t_resource_list:
1684 case hive_t_full_resource_description:
1685 case hive_t_resource_requirements_list:
1687 str = hivex_value_value (h, values[i], &t, &len);
1689 ret = skip_bad ? 0 : -1;
1692 if (vtor->value_other &&
1693 vtor->value_other (h, opaque, node, values[i], t, len, key, str) == -1)
1695 free (str); str = NULL;
1700 free (key); key = NULL;
1703 children = hivex_node_children (h, node);
1704 if (children == NULL) {
1705 ret = skip_bad ? 0 : -1;
1709 for (i = 0; children[i] != 0; ++i) {
1711 fprintf (stderr, "hivex__visit_node: %s: visiting subkey %d (0x%zx)\n",
1712 name, i, children[i]);
1714 if (hivex__visit_node (h, children[i], vtor, unvisited, opaque, flags) == -1)
1718 if (vtor->node_end && vtor->node_end (h, opaque, node, name) == -1)
1729 free_strings (strs);
1733 /*----------------------------------------------------------------------
1737 /* Allocate an hbin (page), extending the malloc'd space if necessary,
1738 * and updating the hive handle fields (but NOT the hive disk header
1739 * -- the hive disk header is updated when we commit). This function
1740 * also extends the bitmap if necessary.
1742 * 'allocation_hint' is the size of the block allocation we would like
1743 * to make. Normally registry blocks are very small (avg 50 bytes)
1744 * and are contained in standard-sized pages (4KB), but the registry
1745 * can support blocks which are larger than a standard page, in which
1746 * case it creates a page of 8KB, 12KB etc.
1749 * > 0 : offset of first usable byte of new page (after page header)
1750 * 0 : error (errno set)
1753 allocate_page (hive_h *h, size_t allocation_hint)
1755 /* In almost all cases this will be 1. */
1756 size_t nr_4k_pages =
1757 1 + (allocation_hint + sizeof (struct ntreg_hbin_page) - 1) / 4096;
1758 assert (nr_4k_pages >= 1);
1760 /* 'extend' is the number of bytes to extend the file by. Note that
1761 * hives found in the wild often contain slack between 'endpages'
1762 * and the actual end of the file, so we don't always need to make
1765 ssize_t extend = h->endpages + nr_4k_pages * 4096 - h->size;
1767 if (h->msglvl >= 2) {
1768 fprintf (stderr, "allocate_page: current endpages = 0x%zx, current size = 0x%zx\n",
1769 h->endpages, h->size);
1770 fprintf (stderr, "allocate_page: extending file by %zd bytes (<= 0 if no extension)\n",
1775 size_t oldsize = h->size;
1776 size_t newsize = h->size + extend;
1777 char *newaddr = realloc (h->addr, newsize);
1778 if (newaddr == NULL)
1781 size_t oldbitmapsize = 1 + oldsize / 32;
1782 size_t newbitmapsize = 1 + newsize / 32;
1783 char *newbitmap = realloc (h->bitmap, newbitmapsize);
1784 if (newbitmap == NULL) {
1791 h->bitmap = newbitmap;
1793 memset (h->addr + oldsize, 0, newsize - oldsize);
1794 memset (h->bitmap + oldbitmapsize, 0, newbitmapsize - oldbitmapsize);
1797 size_t offset = h->endpages;
1798 h->endpages += nr_4k_pages * 4096;
1801 fprintf (stderr, "allocate_page: new endpages = 0x%zx, new size = 0x%zx\n",
1802 h->endpages, h->size);
1804 /* Write the hbin header. */
1805 struct ntreg_hbin_page *page =
1806 (struct ntreg_hbin_page *) (h->addr + offset);
1807 page->magic[0] = 'h';
1808 page->magic[1] = 'b';
1809 page->magic[2] = 'i';
1810 page->magic[3] = 'n';
1811 page->offset_first = htole32 (offset - 0x1000);
1812 page->page_size = htole32 (nr_4k_pages * 4096);
1813 memset (page->unknown, 0, sizeof (page->unknown));
1816 fprintf (stderr, "allocate_page: new page at 0x%zx\n", offset);
1818 /* Offset of first usable byte after the header. */
1819 return offset + sizeof (struct ntreg_hbin_page);
1822 /* Allocate a single block, first allocating an hbin (page) at the end
1823 * of the current file if necessary. NB. To keep the implementation
1824 * simple and more likely to be correct, we do not reuse existing free
1827 * seg_len is the size of the block (this INCLUDES the block header).
1828 * The header of the block is initialized to -seg_len (negative to
1829 * indicate used). id[2] is the block ID (type), eg. "nk" for nk-
1830 * record. The block bitmap is updated to show this block as valid.
1831 * The rest of the contents of the block will be zero.
1833 * **NB** Because allocate_block may reallocate the memory, all
1834 * pointers into the memory become potentially invalid. I really
1835 * love writing in C, can't you tell?
1838 * > 0 : offset of new block
1839 * 0 : error (errno set)
1842 allocate_block (hive_h *h, size_t seg_len, const char id[2])
1850 /* The caller probably forgot to include the header. Note that
1851 * value lists have no ID field, so seg_len == 4 would be possible
1852 * for them, albeit unusual.
1855 fprintf (stderr, "allocate_block: refusing too small allocation (%zu), returning ERANGE\n",
1861 /* Refuse really large allocations. */
1862 if (seg_len > HIVEX_MAX_ALLOCATION) {
1864 fprintf (stderr, "allocate_block: refusing large allocation (%zu), returning ERANGE\n",
1870 /* Round up allocation to multiple of 8 bytes. All blocks must be
1871 * on an 8 byte boundary.
1873 seg_len = (seg_len + 7) & ~7;
1875 /* Allocate a new page if necessary. */
1876 if (h->endblocks == 0 || h->endblocks + seg_len > h->endpages) {
1877 size_t newendblocks = allocate_page (h, seg_len);
1878 if (newendblocks == 0)
1880 h->endblocks = newendblocks;
1883 size_t offset = h->endblocks;
1886 fprintf (stderr, "allocate_block: new block at 0x%zx, size %zu\n",
1889 struct ntreg_hbin_block *blockhdr =
1890 (struct ntreg_hbin_block *) (h->addr + offset);
1892 memset (blockhdr, 0, seg_len);
1894 blockhdr->seg_len = htole32 (- (int32_t) seg_len);
1895 if (id[0] && id[1] && seg_len >= sizeof (struct ntreg_hbin_block)) {
1896 blockhdr->id[0] = id[0];
1897 blockhdr->id[1] = id[1];
1900 BITMAP_SET (h->bitmap, offset);
1902 h->endblocks += seg_len;
1904 /* If there is space after the last block in the last page, then we
1905 * have to put a dummy free block header here to mark the rest of
1908 ssize_t rem = h->endpages - h->endblocks;
1911 fprintf (stderr, "allocate_block: marking remainder of page free starting at 0x%zx, size %zd\n",
1916 blockhdr = (struct ntreg_hbin_block *) (h->addr + h->endblocks);
1917 blockhdr->seg_len = htole32 ((int32_t) rem);
1923 /* 'offset' must point to a valid, used block. This function marks
1924 * the block unused (by updating the seg_len field) and invalidates
1925 * the bitmap. It does NOT do this recursively, so to avoid creating
1926 * unreachable used blocks, callers may have to recurse over the hive
1927 * structures. Also callers must ensure there are no references to
1928 * this block from other parts of the hive.
1931 mark_block_unused (hive_h *h, size_t offset)
1933 assert (h->writable);
1934 assert (IS_VALID_BLOCK (h, offset));
1937 fprintf (stderr, "mark_block_unused: marking 0x%zx unused\n", offset);
1939 struct ntreg_hbin_block *blockhdr =
1940 (struct ntreg_hbin_block *) (h->addr + offset);
1942 size_t seg_len = block_len (h, offset, NULL);
1943 blockhdr->seg_len = htole32 (seg_len);
1945 BITMAP_CLR (h->bitmap, offset);
1948 /* Delete all existing values at this node. */
1950 delete_values (hive_h *h, hive_node_h node)
1952 assert (h->writable);
1954 hive_value_h *values;
1956 if (get_values (h, node, &values, &blocks) == -1)
1960 for (i = 0; blocks[i] != 0; ++i)
1961 mark_block_unused (h, blocks[i]);
1965 for (i = 0; values[i] != 0; ++i) {
1966 struct ntreg_vk_record *vk =
1967 (struct ntreg_vk_record *) (h->addr + values[i]);
1971 len = le32toh (vk->data_len);
1972 is_inline = !!(len & 0x80000000); /* top bit indicates is inline */
1975 if (!is_inline) { /* non-inline, so remove data block */
1976 size_t data_offset = le32toh (vk->data_offset);
1977 data_offset += 0x1000;
1978 mark_block_unused (h, data_offset);
1981 /* remove vk record */
1982 mark_block_unused (h, values[i]);
1987 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
1988 nk->nr_values = htole32 (0);
1989 nk->vallist = htole32 (0xffffffff);
1995 hivex_commit (hive_h *h, const char *filename, int flags)
2007 filename = filename ? : h->filename;
2008 int fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
2012 /* Update the header fields. */
2013 uint32_t sequence = le32toh (h->hdr->sequence1);
2015 h->hdr->sequence1 = htole32 (sequence);
2016 h->hdr->sequence2 = htole32 (sequence);
2017 /* XXX Ought to update h->hdr->last_modified. */
2018 h->hdr->blocks = htole32 (h->endpages - 0x1000);
2020 /* Recompute header checksum. */
2021 uint32_t sum = header_checksum (h);
2022 h->hdr->csum = htole32 (sum);
2025 fprintf (stderr, "hivex_commit: new header checksum: 0x%x\n", sum);
2027 if (full_write (fd, h->addr, h->size) != h->size) {
2034 if (close (fd) == -1)
2040 /* Calculate the hash for a lf or lh record offset.
2043 calc_hash (const char *type, const char *name, char *ret)
2045 size_t len = strlen (name);
2047 if (STRPREFIX (type, "lf"))
2048 /* Old-style, not used in current registries. */
2049 memcpy (ret, name, len < 4 ? len : 4);
2051 /* New-style for lh-records. */
2054 for (i = 0; i < len; ++i) {
2055 c = c_toupper (name[i]);
2059 *((uint32_t *) ret) = htole32 (h);
2063 /* Create a completely new lh-record containing just the single node. */
2065 new_lh_record (hive_h *h, const char *name, hive_node_h node)
2067 static const char id[2] = { 'l', 'h' };
2068 size_t seg_len = sizeof (struct ntreg_lf_record);
2069 size_t offset = allocate_block (h, seg_len, id);
2073 struct ntreg_lf_record *lh = (struct ntreg_lf_record *) (h->addr + offset);
2074 lh->nr_keys = htole16 (1);
2075 lh->keys[0].offset = htole32 (node - 0x1000);
2076 calc_hash ("lh", name, lh->keys[0].hash);
2081 /* Insert node into existing lf/lh-record at position.
2082 * This allocates a new record and marks the old one as unused.
2085 insert_lf_record (hive_h *h, size_t old_offs, size_t posn,
2086 const char *name, hive_node_h node)
2088 assert (IS_VALID_BLOCK (h, old_offs));
2090 /* Work around C stupidity.
2091 * http://www.redhat.com/archives/libguestfs/2010-February/msg00056.html
2093 int test = BLOCK_ID_EQ (h, old_offs, "lf") || BLOCK_ID_EQ (h, old_offs, "lh");
2096 struct ntreg_lf_record *old_lf =
2097 (struct ntreg_lf_record *) (h->addr + old_offs);
2098 size_t nr_keys = le16toh (old_lf->nr_keys);
2100 nr_keys++; /* in new record ... */
2102 size_t seg_len = sizeof (struct ntreg_lf_record) + (nr_keys-1) * 8;
2104 /* Copy the old_lf->id in case it moves during allocate_block. */
2106 memcpy (id, old_lf->id, sizeof id);
2108 size_t new_offs = allocate_block (h, seg_len, id);
2112 /* old_lf could have been invalidated by allocate_block. */
2113 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2115 struct ntreg_lf_record *new_lf =
2116 (struct ntreg_lf_record *) (h->addr + new_offs);
2117 new_lf->nr_keys = htole16 (nr_keys);
2119 /* Copy the keys until we reach posn, insert the new key there, then
2120 * copy the remaining keys.
2123 for (i = 0; i < posn; ++i)
2124 new_lf->keys[i] = old_lf->keys[i];
2126 new_lf->keys[i].offset = htole32 (node - 0x1000);
2127 calc_hash (new_lf->id, name, new_lf->keys[i].hash);
2129 for (i = posn+1; i < nr_keys; ++i)
2130 new_lf->keys[i] = old_lf->keys[i-1];
2132 /* Old block is unused, return new block. */
2133 mark_block_unused (h, old_offs);
2137 /* Compare name with name in nk-record. */
2139 compare_name_with_nk_name (hive_h *h, const char *name, hive_node_h nk_offs)
2141 assert (IS_VALID_BLOCK (h, nk_offs));
2142 assert (BLOCK_ID_EQ (h, nk_offs, "nk"));
2144 /* Name in nk is not necessarily nul-terminated. */
2145 char *nname = hivex_node_name (h, nk_offs);
2147 /* Unfortunately we don't have a way to return errors here. */
2149 perror ("compare_name_with_nk_name");
2153 int r = strcasecmp (name, nname);
2160 hivex_node_add_child (hive_h *h, hive_node_h parent, const char *name)
2167 if (!IS_VALID_BLOCK (h, parent) || !BLOCK_ID_EQ (h, parent, "nk")) {
2172 if (name == NULL || strlen (name) == 0) {
2177 if (hivex_node_get_child (h, parent, name) != 0) {
2182 /* Create the new nk-record. */
2183 static const char nk_id[2] = { 'n', 'k' };
2184 size_t seg_len = sizeof (struct ntreg_nk_record) + strlen (name);
2185 hive_node_h node = allocate_block (h, seg_len, nk_id);
2190 fprintf (stderr, "hivex_node_add_child: allocated new nk-record for child at 0x%zx\n", node);
2192 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2193 nk->flags = htole16 (0x0020); /* key is ASCII. */
2194 nk->parent = htole32 (parent - 0x1000);
2195 nk->subkey_lf = htole32 (0xffffffff);
2196 nk->subkey_lf_volatile = htole32 (0xffffffff);
2197 nk->vallist = htole32 (0xffffffff);
2198 nk->classname = htole32 (0xffffffff);
2199 nk->name_len = htole16 (strlen (name));
2200 strcpy (nk->name, name);
2202 /* Inherit parent sk. */
2203 struct ntreg_nk_record *parent_nk =
2204 (struct ntreg_nk_record *) (h->addr + parent);
2205 size_t parent_sk_offset = le32toh (parent_nk->sk);
2206 parent_sk_offset += 0x1000;
2207 if (!IS_VALID_BLOCK (h, parent_sk_offset) ||
2208 !BLOCK_ID_EQ (h, parent_sk_offset, "sk")) {
2210 fprintf (stderr, "hivex_node_add_child: returning EFAULT because parent sk is not a valid block (%zu)\n",
2215 struct ntreg_sk_record *sk =
2216 (struct ntreg_sk_record *) (h->addr + parent_sk_offset);
2217 sk->refcount = htole32 (le32toh (sk->refcount) + 1);
2218 nk->sk = htole32 (parent_sk_offset - 0x1000);
2220 /* Inherit parent timestamp. */
2221 memcpy (nk->timestamp, parent_nk->timestamp, sizeof (parent_nk->timestamp));
2223 /* What I found out the hard way (not documented anywhere): the
2224 * subkeys in lh-records must be kept sorted. If you just add a
2225 * subkey in a non-sorted position (eg. just add it at the end) then
2226 * Windows won't see the subkey _and_ Windows will corrupt the hive
2227 * itself when it modifies or saves it.
2229 * So use get_children() to get a list of intermediate
2230 * lf/lh-records. get_children() returns these in reading order
2231 * (which is sorted), so we look for the lf/lh-records in sequence
2232 * until we find the key name just after the one we are inserting,
2233 * and we insert the subkey just before it.
2235 * The only other case is the no-subkeys case, where we have to
2236 * create a brand new lh-record.
2238 hive_node_h *unused;
2241 if (get_children (h, parent, &unused, &blocks, 0) == -1)
2246 size_t nr_subkeys_in_parent_nk = le32toh (parent_nk->nr_subkeys);
2247 if (nr_subkeys_in_parent_nk == 0) { /* No subkeys case. */
2248 /* Free up any existing intermediate blocks. */
2249 for (i = 0; blocks[i] != 0; ++i)
2250 mark_block_unused (h, blocks[i]);
2251 size_t lh_offs = new_lh_record (h, name, node);
2257 /* Recalculate pointers that could have been invalidated by
2258 * previous call to allocate_block (via new_lh_record).
2260 nk = (struct ntreg_nk_record *) (h->addr + node);
2261 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2264 fprintf (stderr, "hivex_node_add_child: no keys, allocated new lh-record at 0x%zx\n", lh_offs);
2266 parent_nk->subkey_lf = htole32 (lh_offs - 0x1000);
2268 else { /* Insert subkeys case. */
2269 size_t old_offs = 0, new_offs = 0;
2270 struct ntreg_lf_record *old_lf = NULL;
2272 /* Find lf/lh key name just after the one we are inserting. */
2273 for (i = 0; blocks[i] != 0; ++i) {
2274 if (BLOCK_ID_EQ (h, blocks[i], "lf") ||
2275 BLOCK_ID_EQ (h, blocks[i], "lh")) {
2276 old_offs = blocks[i];
2277 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2278 for (j = 0; j < le16toh (old_lf->nr_keys); ++j) {
2279 hive_node_h nk_offs = le32toh (old_lf->keys[j].offset);
2281 if (compare_name_with_nk_name (h, name, nk_offs) < 0)
2287 /* Insert it at the end.
2288 * old_offs points to the last lf record, set j.
2290 assert (old_offs != 0); /* should never happen if nr_subkeys > 0 */
2291 j = le16toh (old_lf->nr_keys);
2296 fprintf (stderr, "hivex_node_add_child: insert key in existing lh-record at 0x%zx, posn %zu\n", old_offs, j);
2298 new_offs = insert_lf_record (h, old_offs, j, name, node);
2299 if (new_offs == 0) {
2304 /* Recalculate pointers that could have been invalidated by
2305 * previous call to allocate_block (via insert_lf_record).
2307 nk = (struct ntreg_nk_record *) (h->addr + node);
2308 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2311 fprintf (stderr, "hivex_node_add_child: new lh-record at 0x%zx\n",
2314 /* If the lf/lh-record was directly referenced by the parent nk,
2315 * then update the parent nk.
2317 if (le32toh (parent_nk->subkey_lf) + 0x1000 == old_offs)
2318 parent_nk->subkey_lf = htole32 (new_offs - 0x1000);
2319 /* Else we have to look for the intermediate ri-record and update
2323 for (i = 0; blocks[i] != 0; ++i) {
2324 if (BLOCK_ID_EQ (h, blocks[i], "ri")) {
2325 struct ntreg_ri_record *ri =
2326 (struct ntreg_ri_record *) (h->addr + blocks[i]);
2327 for (j = 0; j < le16toh (ri->nr_offsets); ++j)
2328 if (le32toh (ri->offset[j] + 0x1000) == old_offs) {
2329 ri->offset[j] = htole32 (new_offs - 0x1000);
2335 /* Not found .. This is an internal error. */
2337 fprintf (stderr, "hivex_node_add_child: returning ENOTSUP because could not find ri->lf link\n");
2349 /* Update nr_subkeys in parent nk. */
2350 nr_subkeys_in_parent_nk++;
2351 parent_nk->nr_subkeys = htole32 (nr_subkeys_in_parent_nk);
2353 /* Update max_subkey_name_len in parent nk. */
2354 uint16_t max = le16toh (parent_nk->max_subkey_name_len);
2355 if (max < strlen (name) * 2) /* *2 because "recoded" in UTF16-LE. */
2356 parent_nk->max_subkey_name_len = htole16 (strlen (name) * 2);
2361 /* Decrement the refcount of an sk-record, and if it reaches zero,
2362 * unlink it from the chain and delete it.
2365 delete_sk (hive_h *h, size_t sk_offset)
2367 if (!IS_VALID_BLOCK (h, sk_offset) || !BLOCK_ID_EQ (h, sk_offset, "sk")) {
2369 fprintf (stderr, "delete_sk: not an sk record: 0x%zx\n", sk_offset);
2374 struct ntreg_sk_record *sk = (struct ntreg_sk_record *) (h->addr + sk_offset);
2376 if (sk->refcount == 0) {
2378 fprintf (stderr, "delete_sk: sk record already has refcount 0: 0x%zx\n",
2386 if (sk->refcount == 0) {
2387 size_t sk_prev_offset = sk->sk_prev;
2388 sk_prev_offset += 0x1000;
2390 size_t sk_next_offset = sk->sk_next;
2391 sk_next_offset += 0x1000;
2393 /* Update sk_prev/sk_next SKs, unless they both point back to this
2394 * cell in which case we are deleting the last SK.
2396 if (sk_prev_offset != sk_offset && sk_next_offset != sk_offset) {
2397 struct ntreg_sk_record *sk_prev =
2398 (struct ntreg_sk_record *) (h->addr + sk_prev_offset);
2399 struct ntreg_sk_record *sk_next =
2400 (struct ntreg_sk_record *) (h->addr + sk_next_offset);
2402 sk_prev->sk_next = htole32 (sk_next_offset - 0x1000);
2403 sk_next->sk_prev = htole32 (sk_prev_offset - 0x1000);
2406 /* Refcount is zero so really delete this block. */
2407 mark_block_unused (h, sk_offset);
2413 /* Callback from hivex_node_delete_child which is called to delete a
2414 * node AFTER its subnodes have been visited. The subnodes have been
2415 * deleted but we still have to delete any lf/lh/li/ri records and the
2416 * value list block and values, followed by deleting the node itself.
2419 delete_node (hive_h *h, void *opaque, hive_node_h node, const char *name)
2421 /* Get the intermediate blocks. The subkeys have already been
2422 * deleted by this point, so tell get_children() not to check for
2423 * validity of the nk-records.
2425 hive_node_h *unused;
2427 if (get_children (h, node, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK) == -1)
2431 /* We don't care what's in these intermediate blocks, so we can just
2432 * delete them unconditionally.
2435 for (i = 0; blocks[i] != 0; ++i)
2436 mark_block_unused (h, blocks[i]);
2440 /* Delete the values in the node. */
2441 if (delete_values (h, node) == -1)
2444 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2446 /* If the NK references an SK, delete it. */
2447 size_t sk_offs = le32toh (nk->sk);
2448 if (sk_offs != 0xffffffff) {
2450 if (delete_sk (h, sk_offs) == -1)
2452 nk->sk = htole32 (0xffffffff);
2455 /* If the NK references a classname, delete it. */
2456 size_t cl_offs = le32toh (nk->classname);
2457 if (cl_offs != 0xffffffff) {
2459 mark_block_unused (h, cl_offs);
2460 nk->classname = htole32 (0xffffffff);
2463 /* Delete the node itself. */
2464 mark_block_unused (h, node);
2470 hivex_node_delete_child (hive_h *h, hive_node_h node)
2477 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2482 if (node == hivex_root (h)) {
2484 fprintf (stderr, "hivex_node_delete_child: cannot delete root node\n");
2489 hive_node_h parent = hivex_node_parent (h, node);
2493 /* Delete node and all its children and values recursively. */
2494 static const struct hivex_visitor visitor = { .node_end = delete_node };
2495 if (hivex_visit_node (h, node, &visitor, sizeof visitor, NULL, 0) == -1)
2498 /* Delete the link from parent to child. We need to find the lf/lh
2499 * record which contains the offset and remove the offset from that
2500 * record, then decrement the element count in that record, and
2501 * decrement the overall number of subkeys stored in the parent
2504 hive_node_h *unused;
2506 if (get_children (h, parent, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK)== -1)
2511 for (i = 0; blocks[i] != 0; ++i) {
2512 struct ntreg_hbin_block *block =
2513 (struct ntreg_hbin_block *) (h->addr + blocks[i]);
2515 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
2516 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
2518 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
2520 for (j = 0; j < nr_subkeys_in_lf; ++j)
2521 if (le32toh (lf->keys[j].offset) + 0x1000 == node) {
2522 for (; j < nr_subkeys_in_lf - 1; ++j)
2523 memcpy (&lf->keys[j], &lf->keys[j+1], sizeof (lf->keys[j]));
2524 lf->nr_keys = htole16 (nr_subkeys_in_lf - 1);
2530 fprintf (stderr, "hivex_node_delete_child: could not find parent to child link\n");
2535 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + parent);
2536 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
2537 nk->nr_subkeys = htole32 (nr_subkeys_in_nk - 1);
2540 fprintf (stderr, "hivex_node_delete_child: updating nr_subkeys in parent 0x%zx to %zu\n",
2541 parent, nr_subkeys_in_nk);
2547 hivex_node_set_values (hive_h *h, hive_node_h node,
2548 size_t nr_values, const hive_set_value *values,
2556 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2561 /* Delete all existing values. */
2562 if (delete_values (h, node) == -1)
2568 /* Allocate value list node. Value lists have no id field. */
2569 static const char nul_id[2] = { 0, 0 };
2571 sizeof (struct ntreg_value_list) + (nr_values - 1) * sizeof (uint32_t);
2572 size_t vallist_offs = allocate_block (h, seg_len, nul_id);
2573 if (vallist_offs == 0)
2576 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2577 nk->nr_values = htole32 (nr_values);
2578 nk->vallist = htole32 (vallist_offs - 0x1000);
2580 struct ntreg_value_list *vallist =
2581 (struct ntreg_value_list *) (h->addr + vallist_offs);
2584 for (i = 0; i < nr_values; ++i) {
2585 /* Allocate vk record to store this (key, value) pair. */
2586 static const char vk_id[2] = { 'v', 'k' };
2587 seg_len = sizeof (struct ntreg_vk_record) + strlen (values[i].key);
2588 size_t vk_offs = allocate_block (h, seg_len, vk_id);
2592 /* Recalculate pointers that could have been invalidated by
2593 * previous call to allocate_block.
2595 nk = (struct ntreg_nk_record *) (h->addr + node);
2596 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2598 vallist->offset[i] = htole32 (vk_offs - 0x1000);
2600 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2601 size_t name_len = strlen (values[i].key);
2602 vk->name_len = htole16 (name_len);
2603 strcpy (vk->name, values[i].key);
2604 vk->data_type = htole32 (values[i].t);
2605 uint32_t len = values[i].len;
2606 if (len <= 4) /* store it inline => set MSB flag */
2608 vk->data_len = htole32 (len);
2609 vk->flags = name_len == 0 ? 0 : 1;
2611 if (values[i].len <= 4) /* store it inline */
2612 memcpy (&vk->data_offset, values[i].value, values[i].len);
2614 size_t offs = allocate_block (h, values[i].len + 4, nul_id);
2618 /* Recalculate pointers that could have been invalidated by
2619 * previous call to allocate_block.
2621 nk = (struct ntreg_nk_record *) (h->addr + node);
2622 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2623 vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2625 memcpy (h->addr + offs + 4, values[i].value, values[i].len);
2626 vk->data_offset = htole32 (offs - 0x1000);
2629 if (name_len * 2 > le32toh (nk->max_vk_name_len))
2630 /* * 2 for UTF16-LE "reencoding" */
2631 nk->max_vk_name_len = htole32 (name_len * 2);
2632 if (values[i].len > le32toh (nk->max_vk_data_len))
2633 nk->max_vk_data_len = htole32 (values[i].len);
2640 hivex_node_set_value (hive_h *h, hive_node_h node,
2641 const hive_set_value *val, int flags)
2643 hive_value_h *prev_values = hivex_node_values (h, node);
2644 if (prev_values == NULL)
2649 size_t nr_values = 0;
2650 for (hive_value_h *itr = prev_values; *itr != 0; ++itr)
2653 hive_set_value *values = malloc ((nr_values + 1) * (sizeof (hive_set_value)));
2655 goto leave_prev_values;
2658 int idx_of_val = -1;
2659 hive_value_h *prev_val;
2660 for (prev_val = prev_values; *prev_val != 0; ++prev_val) {
2664 hive_set_value *value = &values[prev_val - prev_values];
2666 char *valval = hivex_value_value (h, *prev_val, &t, &len);
2667 if (valval == NULL) goto leave_partial;
2670 value->value = valval;
2674 char *valkey = hivex_value_key (h, *prev_val);
2675 if (valkey == NULL) goto leave_partial;
2678 value->key = valkey;
2680 if (STRCASEEQ (valkey, val->key))
2681 idx_of_val = prev_val - prev_values;
2684 if (idx_of_val > -1) {
2685 free (values[idx_of_val].key);
2686 free (values[idx_of_val].value);
2688 idx_of_val = nr_values;
2692 hive_set_value *value = &values[idx_of_val];
2693 *value = (hive_set_value){
2694 .key = strdup (val->key),
2695 .value = malloc (val->len),
2700 if (value->key == NULL || value->value == NULL) goto leave_partial;
2701 memcpy (value->value, val->value, val->len);
2703 retval = hivex_node_set_values (h, node, nr_values, values, 0);
2706 for (int i = 0; i < alloc_ct; i += 2) {
2707 if (values[i / 2].value != NULL)
2708 free (values[i / 2].value);
2709 if (i + 1 < alloc_ct && values[i / 2].key != NULL)
2710 free (values[i / 2].key);