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"
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 "
418 "(at 0x%zx, after %zu pages)\n",
419 filename, off, pages);
424 size_t page_size = le32toh (page->page_size);
426 fprintf (stderr, "hivex_open: page at 0x%zx, size %zu\n", off, page_size);
428 if (page_size < smallest_page) smallest_page = page_size;
429 if (page_size > largest_page) largest_page = page_size;
431 if (page_size <= sizeof (struct ntreg_hbin_page) ||
432 (page_size & 0x0fff) != 0) {
433 fprintf (stderr, "hivex: %s: page size %zu at 0x%zx, bad registry\n",
434 filename, page_size, off);
439 /* Read the blocks in this page. */
441 struct ntreg_hbin_block *block;
443 for (blkoff = off + 0x20;
444 blkoff < off + page_size;
448 int is_root = blkoff == h->rootoffs;
452 block = (struct ntreg_hbin_block *) (h->addr + blkoff);
454 seg_len = block_len (h, blkoff, &used);
455 if (seg_len <= 4 || (seg_len & 3) != 0) {
456 fprintf (stderr, "hivex: %s: block size %" PRIu32 " at 0x%zx,"
458 filename, le32toh (block->seg_len), blkoff);
464 fprintf (stderr, "hivex_open: %s block id %d,%d at 0x%zx size %zu%s\n",
465 used ? "used" : "free", block->id[0], block->id[1], blkoff,
466 seg_len, is_root ? " (root)" : "");
468 blocks_bytes += seg_len;
469 if (seg_len < smallest_block) smallest_block = seg_len;
470 if (seg_len > largest_block) largest_block = seg_len;
472 if (is_root && !used)
477 used_size += seg_len;
479 /* Root block must be an nk-block. */
480 if (is_root && (block->id[0] != 'n' || block->id[1] != 'k'))
483 /* Note this blkoff is a valid address. */
484 BITMAP_SET (h->bitmap, blkoff);
489 if (!seen_root_block) {
490 fprintf (stderr, "hivex: %s: no root block found\n", filename);
495 if (bad_root_block) {
496 fprintf (stderr, "hivex: %s: bad root block (free or not nk)\n", filename);
503 "hivex_open: successfully read Windows Registry hive file:\n"
504 " pages: %zu [sml: %zu, lge: %zu]\n"
505 " blocks: %zu [sml: %zu, avg: %zu, lge: %zu]\n"
506 " blocks used: %zu\n"
507 " bytes used: %zu\n",
508 pages, smallest_page, largest_page,
509 blocks, smallest_block, blocks_bytes / blocks, largest_block,
510 used_blocks, used_size);
518 if (h->addr && h->size && h->addr != MAP_FAILED) {
520 munmap (h->addr, h->size);
534 hivex_close (hive_h *h)
539 fprintf (stderr, "hivex_close\n");
543 munmap (h->addr, h->size);
553 /*----------------------------------------------------------------------
558 hivex_root (hive_h *h)
560 hive_node_h ret = h->rootoffs;
561 if (!IS_VALID_BLOCK (h, ret)) {
569 hivex_node_name (hive_h *h, hive_node_h node)
571 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
576 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
578 /* AFAIK the node name is always plain ASCII, so no conversion
579 * to UTF-8 is necessary. However we do need to nul-terminate
583 /* nk->name_len is unsigned, 16 bit, so this is safe ... However
584 * we have to make sure the length doesn't exceed the block length.
586 size_t len = le16toh (nk->name_len);
587 size_t seg_len = block_len (h, node, NULL);
588 if (sizeof (struct ntreg_nk_record) + len - 1 > seg_len) {
590 fprintf (stderr, "hivex_node_name: returning EFAULT because node name"
591 " is too long (%zu, %zu)\n",
597 char *ret = malloc (len + 1);
600 memcpy (ret, nk->name, len);
606 /* I think the documentation for the sk and classname fields in the nk
607 * record is wrong, or else the offset field is in the wrong place.
608 * Otherwise this makes no sense. Disabled this for now -- it's not
609 * useful for reading the registry anyway.
613 hivex_node_security (hive_h *h, hive_node_h node)
615 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
620 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
622 hive_node_h ret = le32toh (nk->sk);
624 if (!IS_VALID_BLOCK (h, ret)) {
632 hivex_node_classname (hive_h *h, hive_node_h node)
634 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
639 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
641 hive_node_h ret = le32toh (nk->classname);
643 if (!IS_VALID_BLOCK (h, ret)) {
651 /* Structure for returning 0-terminated lists of offsets (nodes,
661 init_offset_list (struct offset_list *list)
665 list->offsets = NULL;
668 #define INIT_OFFSET_LIST(name) \
669 struct offset_list name; \
670 init_offset_list (&name)
672 /* Preallocates the offset_list, but doesn't make the contents longer. */
674 grow_offset_list (struct offset_list *list, size_t alloc)
676 assert (alloc >= list->len);
677 size_t *p = realloc (list->offsets, alloc * sizeof (size_t));
686 add_to_offset_list (struct offset_list *list, size_t offset)
688 if (list->len >= list->alloc) {
689 if (grow_offset_list (list, list->alloc ? list->alloc * 2 : 4) == -1)
692 list->offsets[list->len] = offset;
698 free_offset_list (struct offset_list *list)
700 free (list->offsets);
704 return_offset_list (struct offset_list *list)
706 if (add_to_offset_list (list, 0) == -1)
708 return list->offsets; /* caller frees */
711 /* Iterate over children, returning child nodes and intermediate blocks. */
712 #define GET_CHILDREN_NO_CHECK_NK 1
715 get_children (hive_h *h, hive_node_h node,
716 hive_node_h **children_ret, size_t **blocks_ret,
719 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
724 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
726 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
728 INIT_OFFSET_LIST (children);
729 INIT_OFFSET_LIST (blocks);
731 /* Deal with the common "no subkeys" case quickly. */
732 if (nr_subkeys_in_nk == 0)
735 /* Arbitrarily limit the number of subkeys we will ever deal with. */
736 if (nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS) {
738 fprintf (stderr, "hivex: get_children: returning ERANGE because "
739 "nr_subkeys_in_nk > HIVEX_MAX_SUBKEYS (%zu > %d)\n",
740 nr_subkeys_in_nk, HIVEX_MAX_SUBKEYS);
745 /* Preallocate space for the children. */
746 if (grow_offset_list (&children, nr_subkeys_in_nk) == -1)
749 /* The subkey_lf field can point either to an lf-record, which is
750 * the common case, or if there are lots of subkeys, to an
753 size_t subkey_lf = le32toh (nk->subkey_lf);
755 if (!IS_VALID_BLOCK (h, subkey_lf)) {
757 fprintf (stderr, "hivex_node_children: returning EFAULT"
758 " because subkey_lf is not a valid block (0x%zx)\n",
764 if (add_to_offset_list (&blocks, subkey_lf) == -1)
767 struct ntreg_hbin_block *block =
768 (struct ntreg_hbin_block *) (h->addr + subkey_lf);
770 /* Points to lf-record? (Note, also "lh" but that is basically the
771 * same as "lf" as far as we are concerned here).
773 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
774 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
776 /* Check number of subkeys in the nk-record matches number of subkeys
779 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
782 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu,"
783 " nr_subkeys_in_lf = %zu\n",
784 nr_subkeys_in_nk, nr_subkeys_in_lf);
786 if (nr_subkeys_in_nk != nr_subkeys_in_lf) {
791 size_t len = block_len (h, subkey_lf, NULL);
792 if (8 + nr_subkeys_in_lf * 8 > len) {
794 fprintf (stderr, "hivex_node_children: returning EFAULT"
795 " because too many subkeys (%zu, %zu)\n",
796 nr_subkeys_in_lf, len);
802 for (i = 0; i < nr_subkeys_in_lf; ++i) {
803 hive_node_h subkey = le32toh (lf->keys[i].offset);
805 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
806 if (!IS_VALID_BLOCK (h, subkey)) {
808 fprintf (stderr, "hivex_node_children: returning EFAULT"
809 " because subkey is not a valid block (0x%zx)\n",
815 if (add_to_offset_list (&children, subkey) == -1)
820 /* Points to ri-record? */
821 else if (block->id[0] == 'r' && block->id[1] == 'i') {
822 struct ntreg_ri_record *ri = (struct ntreg_ri_record *) block;
824 size_t nr_offsets = le16toh (ri->nr_offsets);
826 /* Count total number of children. */
828 for (i = 0; i < nr_offsets; ++i) {
829 hive_node_h offset = le32toh (ri->offset[i]);
831 if (!IS_VALID_BLOCK (h, offset)) {
833 fprintf (stderr, "hivex_node_children: returning EFAULT"
834 " because ri-offset is not a valid block (0x%zx)\n",
839 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
841 fprintf (stderr, "get_children: returning ENOTSUP"
842 " because ri-record offset does not point to lf/lh (0x%zx)\n",
848 if (add_to_offset_list (&blocks, offset) == -1)
851 struct ntreg_lf_record *lf =
852 (struct ntreg_lf_record *) (h->addr + offset);
854 count += le16toh (lf->nr_keys);
858 fprintf (stderr, "hivex_node_children: nr_subkeys_in_nk = %zu,"
860 nr_subkeys_in_nk, count);
862 if (nr_subkeys_in_nk != count) {
867 /* Copy list of children. Note nr_subkeys_in_nk is limited to
868 * something reasonable above.
870 for (i = 0; i < nr_offsets; ++i) {
871 hive_node_h offset = le32toh (ri->offset[i]);
873 if (!IS_VALID_BLOCK (h, offset)) {
875 fprintf (stderr, "hivex_node_children: returning EFAULT"
876 " because ri-offset is not a valid block (0x%zx)\n",
881 if (!BLOCK_ID_EQ (h, offset, "lf") && !BLOCK_ID_EQ (h, offset, "lh")) {
883 fprintf (stderr, "get_children: returning ENOTSUP"
884 " because ri-record offset does not point to lf/lh (0x%zx)\n",
890 struct ntreg_lf_record *lf =
891 (struct ntreg_lf_record *) (h->addr + offset);
894 for (j = 0; j < le16toh (lf->nr_keys); ++j) {
895 hive_node_h subkey = le32toh (lf->keys[j].offset);
897 if (!(flags & GET_CHILDREN_NO_CHECK_NK)) {
898 if (!IS_VALID_BLOCK (h, subkey)) {
900 fprintf (stderr, "hivex_node_children: returning EFAULT"
901 " because indirect subkey is not a valid block (0x%zx)\n",
907 if (add_to_offset_list (&children, subkey) == -1)
913 /* else not supported, set errno and fall through */
915 fprintf (stderr, "get_children: returning ENOTSUP"
916 " because subkey block is not lf/lh/ri (0x%zx, %d, %d)\n",
917 subkey_lf, block->id[0], block->id[1]);
920 free_offset_list (&children);
921 free_offset_list (&blocks);
925 *children_ret = return_offset_list (&children);
926 *blocks_ret = return_offset_list (&blocks);
927 if (!*children_ret || !*blocks_ret)
933 hivex_node_children (hive_h *h, hive_node_h node)
935 hive_node_h *children;
938 if (get_children (h, node, &children, &blocks, 0) == -1)
945 /* Very inefficient, but at least having a separate API call
946 * allows us to make it more efficient in future.
949 hivex_node_get_child (hive_h *h, hive_node_h node, const char *nname)
951 hive_node_h *children = NULL;
955 children = hivex_node_children (h, node);
956 if (!children) goto error;
959 for (i = 0; children[i] != 0; ++i) {
960 name = hivex_node_name (h, children[i]);
961 if (!name) goto error;
962 if (STRCASEEQ (name, nname)) {
966 free (name); name = NULL;
976 hivex_node_parent (hive_h *h, hive_node_h node)
978 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
983 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
985 hive_node_h ret = le32toh (nk->parent);
987 if (!IS_VALID_BLOCK (h, ret)) {
989 fprintf (stderr, "hivex_node_parent: returning EFAULT"
990 " because parent is not a valid block (0x%zx)\n",
999 get_values (hive_h *h, hive_node_h node,
1000 hive_value_h **values_ret, size_t **blocks_ret)
1002 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
1007 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
1009 size_t nr_values = le32toh (nk->nr_values);
1012 fprintf (stderr, "hivex_node_values: nr_values = %zu\n", nr_values);
1014 INIT_OFFSET_LIST (values);
1015 INIT_OFFSET_LIST (blocks);
1017 /* Deal with the common "no values" case quickly. */
1021 /* Arbitrarily limit the number of values we will ever deal with. */
1022 if (nr_values > HIVEX_MAX_VALUES) {
1024 fprintf (stderr, "hivex: get_values: returning ERANGE"
1025 " because nr_values > HIVEX_MAX_VALUES (%zu > %d)\n",
1026 nr_values, HIVEX_MAX_VALUES);
1031 /* Preallocate space for the values. */
1032 if (grow_offset_list (&values, nr_values) == -1)
1035 /* Get the value list and check it looks reasonable. */
1036 size_t vlist_offset = le32toh (nk->vallist);
1037 vlist_offset += 0x1000;
1038 if (!IS_VALID_BLOCK (h, vlist_offset)) {
1040 fprintf (stderr, "hivex_node_values: returning EFAULT"
1041 " because value list is not a valid block (0x%zx)\n",
1047 if (add_to_offset_list (&blocks, vlist_offset) == -1)
1050 struct ntreg_value_list *vlist =
1051 (struct ntreg_value_list *) (h->addr + vlist_offset);
1053 size_t len = block_len (h, vlist_offset, NULL);
1054 if (4 + nr_values * 4 > len) {
1056 fprintf (stderr, "hivex_node_values: returning EFAULT"
1057 " because value list is too long (%zu, %zu)\n",
1064 for (i = 0; i < nr_values; ++i) {
1065 hive_node_h value = vlist->offset[i];
1067 if (!IS_VALID_BLOCK (h, value)) {
1069 fprintf (stderr, "hivex_node_values: returning EFAULT"
1070 " because value is not a valid block (0x%zx)\n",
1075 if (add_to_offset_list (&values, value) == -1)
1080 *values_ret = return_offset_list (&values);
1081 *blocks_ret = return_offset_list (&blocks);
1082 if (!*values_ret || !*blocks_ret)
1087 free_offset_list (&values);
1088 free_offset_list (&blocks);
1093 hivex_node_values (hive_h *h, hive_node_h node)
1095 hive_value_h *values;
1098 if (get_values (h, node, &values, &blocks) == -1)
1105 /* Very inefficient, but at least having a separate API call
1106 * allows us to make it more efficient in future.
1109 hivex_node_get_value (hive_h *h, hive_node_h node, const char *key)
1111 hive_value_h *values = NULL;
1113 hive_value_h ret = 0;
1115 values = hivex_node_values (h, node);
1116 if (!values) goto error;
1119 for (i = 0; values[i] != 0; ++i) {
1120 name = hivex_value_key (h, values[i]);
1121 if (!name) goto error;
1122 if (STRCASEEQ (name, key)) {
1126 free (name); name = NULL;
1136 hivex_value_key (hive_h *h, hive_value_h value)
1138 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1143 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1145 /* AFAIK the key is always plain ASCII, so no conversion to UTF-8 is
1146 * necessary. However we do need to nul-terminate the string.
1149 /* vk->name_len is unsigned, 16 bit, so this is safe ... However
1150 * we have to make sure the length doesn't exceed the block length.
1152 size_t len = le16toh (vk->name_len);
1153 size_t seg_len = block_len (h, value, NULL);
1154 if (sizeof (struct ntreg_vk_record) + len - 1 > seg_len) {
1156 fprintf (stderr, "hivex_value_key: returning EFAULT"
1157 " because key length is too long (%zu, %zu)\n",
1163 char *ret = malloc (len + 1);
1166 memcpy (ret, vk->name, len);
1172 hivex_value_type (hive_h *h, hive_value_h value, hive_type *t, size_t *len)
1174 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1179 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1182 *t = le32toh (vk->data_type);
1185 *len = le32toh (vk->data_len);
1186 *len &= 0x7fffffff; /* top bit indicates if data is stored inline */
1193 hivex_value_value (hive_h *h, hive_value_h value,
1194 hive_type *t_rtn, size_t *len_rtn)
1196 if (!IS_VALID_BLOCK (h, value) || !BLOCK_ID_EQ (h, value, "vk")) {
1201 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + value);
1207 t = le32toh (vk->data_type);
1209 len = le32toh (vk->data_len);
1210 is_inline = !!(len & 0x80000000);
1214 fprintf (stderr, "hivex_value_value: value=0x%zx, t=%d, len=%zu, inline=%d\n",
1215 value, t, len, is_inline);
1222 if (is_inline && len > 4) {
1227 /* Arbitrarily limit the length that we will read. */
1228 if (len > HIVEX_MAX_VALUE_LEN) {
1230 fprintf (stderr, "hivex_value_value: returning ERANGE because data "
1231 "length > HIVEX_MAX_VALUE_LEN (%zu > %d)\n",
1232 len, HIVEX_MAX_SUBKEYS);
1237 char *ret = malloc (len);
1242 memcpy (ret, (char *) &vk->data_offset, len);
1246 size_t data_offset = le32toh (vk->data_offset);
1247 data_offset += 0x1000;
1248 if (!IS_VALID_BLOCK (h, data_offset)) {
1250 fprintf (stderr, "hivex_value_value: returning EFAULT because data "
1251 "offset is not a valid block (0x%zx)\n",
1258 /* Check that the declared size isn't larger than the block its in.
1260 * XXX Some apparently valid registries are seen to have this,
1261 * so turn this into a warning and substitute the smaller length
1264 size_t blen = block_len (h, data_offset, NULL);
1265 if (len > blen - 4 /* subtract 4 for block header */) {
1267 fprintf (stderr, "hivex_value_value: warning: declared data length "
1268 "is longer than the block it is in "
1269 "(data 0x%zx, data len %zu, block len %zu)\n",
1270 data_offset, len, blen);
1273 /* Return the smaller length to the caller too. */
1278 char *data = h->addr + data_offset + 4;
1279 memcpy (ret, data, len);
1284 windows_utf16_to_utf8 (/* const */ char *input, size_t len)
1286 iconv_t ic = iconv_open ("UTF-8", "UTF-16");
1287 if (ic == (iconv_t) -1)
1290 /* iconv(3) has an insane interface ... */
1292 /* Mostly UTF-8 will be smaller, so this is a good initial guess. */
1293 size_t outalloc = len;
1297 size_t outlen = outalloc;
1298 char *out = malloc (outlen + 1);
1308 size_t r = iconv (ic, &inp, &inlen, &outp, &outlen);
1309 if (r == (size_t) -1) {
1310 if (errno == E2BIG) {
1312 size_t prev = outalloc;
1313 /* Try again with a larger output buffer. */
1316 if (outalloc < prev) {
1324 /* Else some conversion failure, eg. EILSEQ, EINVAL. */
1340 hivex_value_string (hive_h *h, hive_value_h value)
1344 char *data = hivex_value_value (h, value, &t, &len);
1349 if (t != hive_t_string && t != hive_t_expand_string && t != hive_t_link) {
1355 /* Deal with the case where Windows has allocated a large buffer
1356 * full of random junk, and only the first few bytes of the buffer
1357 * contain a genuine UTF-16 string.
1359 * In this case, iconv would try to process the junk bytes as UTF-16
1360 * and inevitably find an illegal sequence (EILSEQ). Instead, stop
1361 * after we find the first \0\0.
1363 * (Found by Hilko Bengen in a fresh Windows XP SOFTWARE hive).
1365 size_t slen = utf16_string_len_in_bytes_max (data, len);
1369 char *ret = windows_utf16_to_utf8 (data, len);
1378 free_strings (char **argv)
1383 for (i = 0; argv[i] != NULL; ++i)
1389 /* Get the length of a UTF-16 format string. Handle the string as
1390 * pairs of bytes, looking for the first \0\0 pair. Only read up to
1391 * 'len' maximum bytes.
1394 utf16_string_len_in_bytes_max (const char *str, size_t len)
1398 while (len >= 2 && (str[0] || str[1])) {
1407 /* http://blogs.msdn.com/oldnewthing/archive/2009/10/08/9904646.aspx */
1409 hivex_value_multiple_strings (hive_h *h, hive_value_h value)
1413 char *data = hivex_value_value (h, value, &t, &len);
1418 if (t != hive_t_multiple_strings) {
1424 size_t nr_strings = 0;
1425 char **ret = malloc ((1 + nr_strings) * sizeof (char *));
1435 while (p < data + len &&
1436 (plen = utf16_string_len_in_bytes_max (p, data + len - p)) > 0) {
1438 char **ret2 = realloc (ret, (1 + nr_strings) * sizeof (char *));
1446 ret[nr_strings-1] = windows_utf16_to_utf8 (p, plen);
1447 ret[nr_strings] = NULL;
1448 if (ret[nr_strings-1] == NULL) {
1454 p += plen + 2 /* skip over UTF-16 \0\0 at the end of this string */;
1462 hivex_value_dword (hive_h *h, hive_value_h value)
1466 char *data = hivex_value_value (h, value, &t, &len);
1471 if ((t != hive_t_dword && t != hive_t_dword_be) || len != 4) {
1477 int32_t ret = *(int32_t*)data;
1479 if (t == hive_t_dword) /* little endian */
1480 ret = le32toh (ret);
1482 ret = be32toh (ret);
1488 hivex_value_qword (hive_h *h, hive_value_h value)
1492 char *data = hivex_value_value (h, value, &t, &len);
1497 if (t != hive_t_qword || len != 8) {
1503 int64_t ret = *(int64_t*)data;
1505 ret = le64toh (ret); /* always little endian */
1510 /*----------------------------------------------------------------------
1515 hivex_visit (hive_h *h, const struct hivex_visitor *visitor, size_t len,
1516 void *opaque, int flags)
1518 return hivex_visit_node (h, hivex_root (h), visitor, len, opaque, flags);
1521 static int hivex__visit_node (hive_h *h, hive_node_h node,
1522 const struct hivex_visitor *vtor,
1523 char *unvisited, void *opaque, int flags);
1526 hivex_visit_node (hive_h *h, hive_node_h node,
1527 const struct hivex_visitor *visitor, size_t len, void *opaque,
1530 struct hivex_visitor vtor;
1531 memset (&vtor, 0, sizeof vtor);
1533 /* Note that len might be larger *or smaller* than the expected size. */
1534 size_t copysize = len <= sizeof vtor ? len : sizeof vtor;
1535 memcpy (&vtor, visitor, copysize);
1537 /* This bitmap records unvisited nodes, so we don't loop if the
1538 * registry contains cycles.
1540 char *unvisited = malloc (1 + h->size / 32);
1541 if (unvisited == NULL)
1543 memcpy (unvisited, h->bitmap, 1 + h->size / 32);
1545 int r = hivex__visit_node (h, node, &vtor, unvisited, opaque, flags);
1551 hivex__visit_node (hive_h *h, hive_node_h node,
1552 const struct hivex_visitor *vtor, char *unvisited,
1553 void *opaque, int flags)
1555 int skip_bad = flags & HIVEX_VISIT_SKIP_BAD;
1557 hive_value_h *values = NULL;
1558 hive_node_h *children = NULL;
1564 /* Return -1 on all callback errors. However on internal errors,
1565 * check if skip_bad is set and suppress those errors if so.
1569 if (!BITMAP_TST (unvisited, node)) {
1571 fprintf (stderr, "hivex__visit_node: contains cycle:"
1572 " visited node 0x%zx already\n",
1576 return skip_bad ? 0 : -1;
1578 BITMAP_CLR (unvisited, node);
1580 name = hivex_node_name (h, node);
1581 if (!name) return skip_bad ? 0 : -1;
1582 if (vtor->node_start && vtor->node_start (h, opaque, node, name) == -1)
1585 values = hivex_node_values (h, node);
1587 ret = skip_bad ? 0 : -1;
1591 for (i = 0; values[i] != 0; ++i) {
1595 if (hivex_value_type (h, values[i], &t, &len) == -1) {
1596 ret = skip_bad ? 0 : -1;
1600 key = hivex_value_key (h, values[i]);
1602 ret = skip_bad ? 0 : -1;
1606 if (vtor->value_any) {
1607 str = hivex_value_value (h, values[i], &t, &len);
1609 ret = skip_bad ? 0 : -1;
1612 if (vtor->value_any (h, opaque, node, values[i], t, len, key, str) == -1)
1614 free (str); str = NULL;
1619 str = hivex_value_value (h, values[i], &t, &len);
1621 ret = skip_bad ? 0 : -1;
1624 if (t != hive_t_none) {
1625 ret = skip_bad ? 0 : -1;
1628 if (vtor->value_none &&
1629 vtor->value_none (h, opaque, node, values[i], t, len, key, str) == -1)
1631 free (str); str = NULL;
1635 case hive_t_expand_string:
1637 str = hivex_value_string (h, values[i]);
1639 if (errno != EILSEQ && errno != EINVAL) {
1640 ret = skip_bad ? 0 : -1;
1643 if (vtor->value_string_invalid_utf16) {
1644 str = hivex_value_value (h, values[i], &t, &len);
1645 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i],
1646 t, len, key, str) == -1)
1648 free (str); str = NULL;
1652 if (vtor->value_string &&
1653 vtor->value_string (h, opaque, node, values[i],
1654 t, len, key, str) == -1)
1656 free (str); str = NULL;
1660 case hive_t_dword_be: {
1661 int32_t i32 = hivex_value_dword (h, values[i]);
1662 if (vtor->value_dword &&
1663 vtor->value_dword (h, opaque, node, values[i],
1664 t, len, key, i32) == -1)
1669 case hive_t_qword: {
1670 int64_t i64 = hivex_value_qword (h, values[i]);
1671 if (vtor->value_qword &&
1672 vtor->value_qword (h, opaque, node, values[i],
1673 t, len, key, i64) == -1)
1679 str = hivex_value_value (h, values[i], &t, &len);
1681 ret = skip_bad ? 0 : -1;
1684 if (t != hive_t_binary) {
1685 ret = skip_bad ? 0 : -1;
1688 if (vtor->value_binary &&
1689 vtor->value_binary (h, opaque, node, values[i],
1690 t, len, key, str) == -1)
1692 free (str); str = NULL;
1695 case hive_t_multiple_strings:
1696 strs = hivex_value_multiple_strings (h, values[i]);
1698 if (errno != EILSEQ && errno != EINVAL) {
1699 ret = skip_bad ? 0 : -1;
1702 if (vtor->value_string_invalid_utf16) {
1703 str = hivex_value_value (h, values[i], &t, &len);
1704 if (vtor->value_string_invalid_utf16 (h, opaque, node, values[i],
1705 t, len, key, str) == -1)
1707 free (str); str = NULL;
1711 if (vtor->value_multiple_strings &&
1712 vtor->value_multiple_strings (h, opaque, node, values[i],
1713 t, len, key, strs) == -1)
1715 free_strings (strs); strs = NULL;
1718 case hive_t_resource_list:
1719 case hive_t_full_resource_description:
1720 case hive_t_resource_requirements_list:
1722 str = hivex_value_value (h, values[i], &t, &len);
1724 ret = skip_bad ? 0 : -1;
1727 if (vtor->value_other &&
1728 vtor->value_other (h, opaque, node, values[i],
1729 t, len, key, str) == -1)
1731 free (str); str = NULL;
1736 free (key); key = NULL;
1739 children = hivex_node_children (h, node);
1740 if (children == NULL) {
1741 ret = skip_bad ? 0 : -1;
1745 for (i = 0; children[i] != 0; ++i) {
1747 fprintf (stderr, "hivex__visit_node: %s: visiting subkey %d (0x%zx)\n",
1748 name, i, children[i]);
1750 if (hivex__visit_node (h, children[i], vtor, unvisited, opaque, flags) == -1)
1754 if (vtor->node_end && vtor->node_end (h, opaque, node, name) == -1)
1765 free_strings (strs);
1769 /*----------------------------------------------------------------------
1773 /* Allocate an hbin (page), extending the malloc'd space if necessary,
1774 * and updating the hive handle fields (but NOT the hive disk header
1775 * -- the hive disk header is updated when we commit). This function
1776 * also extends the bitmap if necessary.
1778 * 'allocation_hint' is the size of the block allocation we would like
1779 * to make. Normally registry blocks are very small (avg 50 bytes)
1780 * and are contained in standard-sized pages (4KB), but the registry
1781 * can support blocks which are larger than a standard page, in which
1782 * case it creates a page of 8KB, 12KB etc.
1785 * > 0 : offset of first usable byte of new page (after page header)
1786 * 0 : error (errno set)
1789 allocate_page (hive_h *h, size_t allocation_hint)
1791 /* In almost all cases this will be 1. */
1792 size_t nr_4k_pages =
1793 1 + (allocation_hint + sizeof (struct ntreg_hbin_page) - 1) / 4096;
1794 assert (nr_4k_pages >= 1);
1796 /* 'extend' is the number of bytes to extend the file by. Note that
1797 * hives found in the wild often contain slack between 'endpages'
1798 * and the actual end of the file, so we don't always need to make
1801 ssize_t extend = h->endpages + nr_4k_pages * 4096 - h->size;
1803 if (h->msglvl >= 2) {
1804 fprintf (stderr, "allocate_page: current endpages = 0x%zx,"
1805 " current size = 0x%zx\n",
1806 h->endpages, h->size);
1807 fprintf (stderr, "allocate_page: extending file by %zd bytes"
1808 " (<= 0 if no extension)\n",
1813 size_t oldsize = h->size;
1814 size_t newsize = h->size + extend;
1815 char *newaddr = realloc (h->addr, newsize);
1816 if (newaddr == NULL)
1819 size_t oldbitmapsize = 1 + oldsize / 32;
1820 size_t newbitmapsize = 1 + newsize / 32;
1821 char *newbitmap = realloc (h->bitmap, newbitmapsize);
1822 if (newbitmap == NULL) {
1829 h->bitmap = newbitmap;
1831 memset (h->addr + oldsize, 0, newsize - oldsize);
1832 memset (h->bitmap + oldbitmapsize, 0, newbitmapsize - oldbitmapsize);
1835 size_t offset = h->endpages;
1836 h->endpages += nr_4k_pages * 4096;
1839 fprintf (stderr, "allocate_page: new endpages = 0x%zx, new size = 0x%zx\n",
1840 h->endpages, h->size);
1842 /* Write the hbin header. */
1843 struct ntreg_hbin_page *page =
1844 (struct ntreg_hbin_page *) (h->addr + offset);
1845 page->magic[0] = 'h';
1846 page->magic[1] = 'b';
1847 page->magic[2] = 'i';
1848 page->magic[3] = 'n';
1849 page->offset_first = htole32 (offset - 0x1000);
1850 page->page_size = htole32 (nr_4k_pages * 4096);
1851 memset (page->unknown, 0, sizeof (page->unknown));
1854 fprintf (stderr, "allocate_page: new page at 0x%zx\n", offset);
1856 /* Offset of first usable byte after the header. */
1857 return offset + sizeof (struct ntreg_hbin_page);
1860 /* Allocate a single block, first allocating an hbin (page) at the end
1861 * of the current file if necessary. NB. To keep the implementation
1862 * simple and more likely to be correct, we do not reuse existing free
1865 * seg_len is the size of the block (this INCLUDES the block header).
1866 * The header of the block is initialized to -seg_len (negative to
1867 * indicate used). id[2] is the block ID (type), eg. "nk" for nk-
1868 * record. The block bitmap is updated to show this block as valid.
1869 * The rest of the contents of the block will be zero.
1871 * **NB** Because allocate_block may reallocate the memory, all
1872 * pointers into the memory become potentially invalid. I really
1873 * love writing in C, can't you tell?
1876 * > 0 : offset of new block
1877 * 0 : error (errno set)
1880 allocate_block (hive_h *h, size_t seg_len, const char id[2])
1888 /* The caller probably forgot to include the header. Note that
1889 * value lists have no ID field, so seg_len == 4 would be possible
1890 * for them, albeit unusual.
1893 fprintf (stderr, "allocate_block: refusing too small allocation (%zu),"
1894 " returning ERANGE\n", seg_len);
1899 /* Refuse really large allocations. */
1900 if (seg_len > HIVEX_MAX_ALLOCATION) {
1902 fprintf (stderr, "allocate_block: refusing large allocation (%zu),"
1903 " returning ERANGE\n", seg_len);
1908 /* Round up allocation to multiple of 8 bytes. All blocks must be
1909 * on an 8 byte boundary.
1911 seg_len = (seg_len + 7) & ~7;
1913 /* Allocate a new page if necessary. */
1914 if (h->endblocks == 0 || h->endblocks + seg_len > h->endpages) {
1915 size_t newendblocks = allocate_page (h, seg_len);
1916 if (newendblocks == 0)
1918 h->endblocks = newendblocks;
1921 size_t offset = h->endblocks;
1924 fprintf (stderr, "allocate_block: new block at 0x%zx, size %zu\n",
1927 struct ntreg_hbin_block *blockhdr =
1928 (struct ntreg_hbin_block *) (h->addr + offset);
1930 memset (blockhdr, 0, seg_len);
1932 blockhdr->seg_len = htole32 (- (int32_t) seg_len);
1933 if (id[0] && id[1] && seg_len >= sizeof (struct ntreg_hbin_block)) {
1934 blockhdr->id[0] = id[0];
1935 blockhdr->id[1] = id[1];
1938 BITMAP_SET (h->bitmap, offset);
1940 h->endblocks += seg_len;
1942 /* If there is space after the last block in the last page, then we
1943 * have to put a dummy free block header here to mark the rest of
1946 ssize_t rem = h->endpages - h->endblocks;
1949 fprintf (stderr, "allocate_block: marking remainder of page free"
1950 " starting at 0x%zx, size %zd\n", h->endblocks, rem);
1954 blockhdr = (struct ntreg_hbin_block *) (h->addr + h->endblocks);
1955 blockhdr->seg_len = htole32 ((int32_t) rem);
1961 /* 'offset' must point to a valid, used block. This function marks
1962 * the block unused (by updating the seg_len field) and invalidates
1963 * the bitmap. It does NOT do this recursively, so to avoid creating
1964 * unreachable used blocks, callers may have to recurse over the hive
1965 * structures. Also callers must ensure there are no references to
1966 * this block from other parts of the hive.
1969 mark_block_unused (hive_h *h, size_t offset)
1971 assert (h->writable);
1972 assert (IS_VALID_BLOCK (h, offset));
1975 fprintf (stderr, "mark_block_unused: marking 0x%zx unused\n", offset);
1977 struct ntreg_hbin_block *blockhdr =
1978 (struct ntreg_hbin_block *) (h->addr + offset);
1980 size_t seg_len = block_len (h, offset, NULL);
1981 blockhdr->seg_len = htole32 (seg_len);
1983 BITMAP_CLR (h->bitmap, offset);
1986 /* Delete all existing values at this node. */
1988 delete_values (hive_h *h, hive_node_h node)
1990 assert (h->writable);
1992 hive_value_h *values;
1994 if (get_values (h, node, &values, &blocks) == -1)
1998 for (i = 0; blocks[i] != 0; ++i)
1999 mark_block_unused (h, blocks[i]);
2003 for (i = 0; values[i] != 0; ++i) {
2004 struct ntreg_vk_record *vk =
2005 (struct ntreg_vk_record *) (h->addr + values[i]);
2009 len = le32toh (vk->data_len);
2010 is_inline = !!(len & 0x80000000); /* top bit indicates is inline */
2013 if (!is_inline) { /* non-inline, so remove data block */
2014 size_t data_offset = le32toh (vk->data_offset);
2015 data_offset += 0x1000;
2016 mark_block_unused (h, data_offset);
2019 /* remove vk record */
2020 mark_block_unused (h, values[i]);
2025 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2026 nk->nr_values = htole32 (0);
2027 nk->vallist = htole32 (0xffffffff);
2033 hivex_commit (hive_h *h, const char *filename, int flags)
2045 filename = filename ? : h->filename;
2046 int fd = open (filename, O_WRONLY|O_CREAT|O_TRUNC|O_NOCTTY, 0666);
2050 /* Update the header fields. */
2051 uint32_t sequence = le32toh (h->hdr->sequence1);
2053 h->hdr->sequence1 = htole32 (sequence);
2054 h->hdr->sequence2 = htole32 (sequence);
2055 /* XXX Ought to update h->hdr->last_modified. */
2056 h->hdr->blocks = htole32 (h->endpages - 0x1000);
2058 /* Recompute header checksum. */
2059 uint32_t sum = header_checksum (h);
2060 h->hdr->csum = htole32 (sum);
2063 fprintf (stderr, "hivex_commit: new header checksum: 0x%x\n", sum);
2065 if (full_write (fd, h->addr, h->size) != h->size) {
2072 if (close (fd) == -1)
2078 /* Calculate the hash for a lf or lh record offset.
2081 calc_hash (const char *type, const char *name, char *ret)
2083 size_t len = strlen (name);
2085 if (STRPREFIX (type, "lf"))
2086 /* Old-style, not used in current registries. */
2087 memcpy (ret, name, len < 4 ? len : 4);
2089 /* New-style for lh-records. */
2092 for (i = 0; i < len; ++i) {
2093 c = c_toupper (name[i]);
2097 *((uint32_t *) ret) = htole32 (h);
2101 /* Create a completely new lh-record containing just the single node. */
2103 new_lh_record (hive_h *h, const char *name, hive_node_h node)
2105 static const char id[2] = { 'l', 'h' };
2106 size_t seg_len = sizeof (struct ntreg_lf_record);
2107 size_t offset = allocate_block (h, seg_len, id);
2111 struct ntreg_lf_record *lh = (struct ntreg_lf_record *) (h->addr + offset);
2112 lh->nr_keys = htole16 (1);
2113 lh->keys[0].offset = htole32 (node - 0x1000);
2114 calc_hash ("lh", name, lh->keys[0].hash);
2119 /* Insert node into existing lf/lh-record at position.
2120 * This allocates a new record and marks the old one as unused.
2123 insert_lf_record (hive_h *h, size_t old_offs, size_t posn,
2124 const char *name, hive_node_h node)
2126 assert (IS_VALID_BLOCK (h, old_offs));
2128 /* Work around C stupidity.
2129 * http://www.redhat.com/archives/libguestfs/2010-February/msg00056.html
2131 int test = BLOCK_ID_EQ (h, old_offs, "lf") || BLOCK_ID_EQ (h, old_offs, "lh");
2134 struct ntreg_lf_record *old_lf =
2135 (struct ntreg_lf_record *) (h->addr + old_offs);
2136 size_t nr_keys = le16toh (old_lf->nr_keys);
2138 nr_keys++; /* in new record ... */
2140 size_t seg_len = sizeof (struct ntreg_lf_record) + (nr_keys-1) * 8;
2142 /* Copy the old_lf->id in case it moves during allocate_block. */
2144 memcpy (id, old_lf->id, sizeof id);
2146 size_t new_offs = allocate_block (h, seg_len, id);
2150 /* old_lf could have been invalidated by allocate_block. */
2151 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2153 struct ntreg_lf_record *new_lf =
2154 (struct ntreg_lf_record *) (h->addr + new_offs);
2155 new_lf->nr_keys = htole16 (nr_keys);
2157 /* Copy the keys until we reach posn, insert the new key there, then
2158 * copy the remaining keys.
2161 for (i = 0; i < posn; ++i)
2162 new_lf->keys[i] = old_lf->keys[i];
2164 new_lf->keys[i].offset = htole32 (node - 0x1000);
2165 calc_hash (new_lf->id, name, new_lf->keys[i].hash);
2167 for (i = posn+1; i < nr_keys; ++i)
2168 new_lf->keys[i] = old_lf->keys[i-1];
2170 /* Old block is unused, return new block. */
2171 mark_block_unused (h, old_offs);
2175 /* Compare name with name in nk-record. */
2177 compare_name_with_nk_name (hive_h *h, const char *name, hive_node_h nk_offs)
2179 assert (IS_VALID_BLOCK (h, nk_offs));
2180 assert (BLOCK_ID_EQ (h, nk_offs, "nk"));
2182 /* Name in nk is not necessarily nul-terminated. */
2183 char *nname = hivex_node_name (h, nk_offs);
2185 /* Unfortunately we don't have a way to return errors here. */
2187 perror ("compare_name_with_nk_name");
2191 int r = strcasecmp (name, nname);
2198 hivex_node_add_child (hive_h *h, hive_node_h parent, const char *name)
2205 if (!IS_VALID_BLOCK (h, parent) || !BLOCK_ID_EQ (h, parent, "nk")) {
2210 if (name == NULL || strlen (name) == 0) {
2215 if (hivex_node_get_child (h, parent, name) != 0) {
2220 /* Create the new nk-record. */
2221 static const char nk_id[2] = { 'n', 'k' };
2222 size_t seg_len = sizeof (struct ntreg_nk_record) + strlen (name);
2223 hive_node_h node = allocate_block (h, seg_len, nk_id);
2228 fprintf (stderr, "hivex_node_add_child: allocated new nk-record"
2229 " for child at 0x%zx\n", node);
2231 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2232 nk->flags = htole16 (0x0020); /* key is ASCII. */
2233 nk->parent = htole32 (parent - 0x1000);
2234 nk->subkey_lf = htole32 (0xffffffff);
2235 nk->subkey_lf_volatile = htole32 (0xffffffff);
2236 nk->vallist = htole32 (0xffffffff);
2237 nk->classname = htole32 (0xffffffff);
2238 nk->name_len = htole16 (strlen (name));
2239 strcpy (nk->name, name);
2241 /* Inherit parent sk. */
2242 struct ntreg_nk_record *parent_nk =
2243 (struct ntreg_nk_record *) (h->addr + parent);
2244 size_t parent_sk_offset = le32toh (parent_nk->sk);
2245 parent_sk_offset += 0x1000;
2246 if (!IS_VALID_BLOCK (h, parent_sk_offset) ||
2247 !BLOCK_ID_EQ (h, parent_sk_offset, "sk")) {
2249 fprintf (stderr, "hivex_node_add_child: returning EFAULT"
2250 " because parent sk is not a valid block (%zu)\n",
2255 struct ntreg_sk_record *sk =
2256 (struct ntreg_sk_record *) (h->addr + parent_sk_offset);
2257 sk->refcount = htole32 (le32toh (sk->refcount) + 1);
2258 nk->sk = htole32 (parent_sk_offset - 0x1000);
2260 /* Inherit parent timestamp. */
2261 memcpy (nk->timestamp, parent_nk->timestamp, sizeof (parent_nk->timestamp));
2263 /* What I found out the hard way (not documented anywhere): the
2264 * subkeys in lh-records must be kept sorted. If you just add a
2265 * subkey in a non-sorted position (eg. just add it at the end) then
2266 * Windows won't see the subkey _and_ Windows will corrupt the hive
2267 * itself when it modifies or saves it.
2269 * So use get_children() to get a list of intermediate
2270 * lf/lh-records. get_children() returns these in reading order
2271 * (which is sorted), so we look for the lf/lh-records in sequence
2272 * until we find the key name just after the one we are inserting,
2273 * and we insert the subkey just before it.
2275 * The only other case is the no-subkeys case, where we have to
2276 * create a brand new lh-record.
2278 hive_node_h *unused;
2281 if (get_children (h, parent, &unused, &blocks, 0) == -1)
2286 size_t nr_subkeys_in_parent_nk = le32toh (parent_nk->nr_subkeys);
2287 if (nr_subkeys_in_parent_nk == 0) { /* No subkeys case. */
2288 /* Free up any existing intermediate blocks. */
2289 for (i = 0; blocks[i] != 0; ++i)
2290 mark_block_unused (h, blocks[i]);
2291 size_t lh_offs = new_lh_record (h, name, node);
2297 /* Recalculate pointers that could have been invalidated by
2298 * previous call to allocate_block (via new_lh_record).
2300 nk = (struct ntreg_nk_record *) (h->addr + node);
2301 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2304 fprintf (stderr, "hivex_node_add_child: no keys, allocated new"
2305 " lh-record at 0x%zx\n", lh_offs);
2307 parent_nk->subkey_lf = htole32 (lh_offs - 0x1000);
2309 else { /* Insert subkeys case. */
2310 size_t old_offs = 0, new_offs = 0;
2311 struct ntreg_lf_record *old_lf = NULL;
2313 /* Find lf/lh key name just after the one we are inserting. */
2314 for (i = 0; blocks[i] != 0; ++i) {
2315 if (BLOCK_ID_EQ (h, blocks[i], "lf") ||
2316 BLOCK_ID_EQ (h, blocks[i], "lh")) {
2317 old_offs = blocks[i];
2318 old_lf = (struct ntreg_lf_record *) (h->addr + old_offs);
2319 for (j = 0; j < le16toh (old_lf->nr_keys); ++j) {
2320 hive_node_h nk_offs = le32toh (old_lf->keys[j].offset);
2322 if (compare_name_with_nk_name (h, name, nk_offs) < 0)
2328 /* Insert it at the end.
2329 * old_offs points to the last lf record, set j.
2331 assert (old_offs != 0); /* should never happen if nr_subkeys > 0 */
2332 j = le16toh (old_lf->nr_keys);
2337 fprintf (stderr, "hivex_node_add_child: insert key in existing"
2338 " lh-record at 0x%zx, posn %zu\n", old_offs, j);
2340 new_offs = insert_lf_record (h, old_offs, j, name, node);
2341 if (new_offs == 0) {
2346 /* Recalculate pointers that could have been invalidated by
2347 * previous call to allocate_block (via insert_lf_record).
2349 nk = (struct ntreg_nk_record *) (h->addr + node);
2350 parent_nk = (struct ntreg_nk_record *) (h->addr + parent);
2353 fprintf (stderr, "hivex_node_add_child: new lh-record at 0x%zx\n",
2356 /* If the lf/lh-record was directly referenced by the parent nk,
2357 * then update the parent nk.
2359 if (le32toh (parent_nk->subkey_lf) + 0x1000 == old_offs)
2360 parent_nk->subkey_lf = htole32 (new_offs - 0x1000);
2361 /* Else we have to look for the intermediate ri-record and update
2365 for (i = 0; blocks[i] != 0; ++i) {
2366 if (BLOCK_ID_EQ (h, blocks[i], "ri")) {
2367 struct ntreg_ri_record *ri =
2368 (struct ntreg_ri_record *) (h->addr + blocks[i]);
2369 for (j = 0; j < le16toh (ri->nr_offsets); ++j)
2370 if (le32toh (ri->offset[j] + 0x1000) == old_offs) {
2371 ri->offset[j] = htole32 (new_offs - 0x1000);
2377 /* Not found .. This is an internal error. */
2379 fprintf (stderr, "hivex_node_add_child: returning ENOTSUP"
2380 " because could not find ri->lf link\n");
2392 /* Update nr_subkeys in parent nk. */
2393 nr_subkeys_in_parent_nk++;
2394 parent_nk->nr_subkeys = htole32 (nr_subkeys_in_parent_nk);
2396 /* Update max_subkey_name_len in parent nk. */
2397 uint16_t max = le16toh (parent_nk->max_subkey_name_len);
2398 if (max < strlen (name) * 2) /* *2 because "recoded" in UTF16-LE. */
2399 parent_nk->max_subkey_name_len = htole16 (strlen (name) * 2);
2404 /* Decrement the refcount of an sk-record, and if it reaches zero,
2405 * unlink it from the chain and delete it.
2408 delete_sk (hive_h *h, size_t sk_offset)
2410 if (!IS_VALID_BLOCK (h, sk_offset) || !BLOCK_ID_EQ (h, sk_offset, "sk")) {
2412 fprintf (stderr, "delete_sk: not an sk record: 0x%zx\n", sk_offset);
2417 struct ntreg_sk_record *sk = (struct ntreg_sk_record *) (h->addr + sk_offset);
2419 if (sk->refcount == 0) {
2421 fprintf (stderr, "delete_sk: sk record already has refcount 0: 0x%zx\n",
2429 if (sk->refcount == 0) {
2430 size_t sk_prev_offset = sk->sk_prev;
2431 sk_prev_offset += 0x1000;
2433 size_t sk_next_offset = sk->sk_next;
2434 sk_next_offset += 0x1000;
2436 /* Update sk_prev/sk_next SKs, unless they both point back to this
2437 * cell in which case we are deleting the last SK.
2439 if (sk_prev_offset != sk_offset && sk_next_offset != sk_offset) {
2440 struct ntreg_sk_record *sk_prev =
2441 (struct ntreg_sk_record *) (h->addr + sk_prev_offset);
2442 struct ntreg_sk_record *sk_next =
2443 (struct ntreg_sk_record *) (h->addr + sk_next_offset);
2445 sk_prev->sk_next = htole32 (sk_next_offset - 0x1000);
2446 sk_next->sk_prev = htole32 (sk_prev_offset - 0x1000);
2449 /* Refcount is zero so really delete this block. */
2450 mark_block_unused (h, sk_offset);
2456 /* Callback from hivex_node_delete_child which is called to delete a
2457 * node AFTER its subnodes have been visited. The subnodes have been
2458 * deleted but we still have to delete any lf/lh/li/ri records and the
2459 * value list block and values, followed by deleting the node itself.
2462 delete_node (hive_h *h, void *opaque, hive_node_h node, const char *name)
2464 /* Get the intermediate blocks. The subkeys have already been
2465 * deleted by this point, so tell get_children() not to check for
2466 * validity of the nk-records.
2468 hive_node_h *unused;
2470 if (get_children (h, node, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK) == -1)
2474 /* We don't care what's in these intermediate blocks, so we can just
2475 * delete them unconditionally.
2478 for (i = 0; blocks[i] != 0; ++i)
2479 mark_block_unused (h, blocks[i]);
2483 /* Delete the values in the node. */
2484 if (delete_values (h, node) == -1)
2487 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2489 /* If the NK references an SK, delete it. */
2490 size_t sk_offs = le32toh (nk->sk);
2491 if (sk_offs != 0xffffffff) {
2493 if (delete_sk (h, sk_offs) == -1)
2495 nk->sk = htole32 (0xffffffff);
2498 /* If the NK references a classname, delete it. */
2499 size_t cl_offs = le32toh (nk->classname);
2500 if (cl_offs != 0xffffffff) {
2502 mark_block_unused (h, cl_offs);
2503 nk->classname = htole32 (0xffffffff);
2506 /* Delete the node itself. */
2507 mark_block_unused (h, node);
2513 hivex_node_delete_child (hive_h *h, hive_node_h node)
2520 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2525 if (node == hivex_root (h)) {
2527 fprintf (stderr, "hivex_node_delete_child: cannot delete root node\n");
2532 hive_node_h parent = hivex_node_parent (h, node);
2536 /* Delete node and all its children and values recursively. */
2537 static const struct hivex_visitor visitor = { .node_end = delete_node };
2538 if (hivex_visit_node (h, node, &visitor, sizeof visitor, NULL, 0) == -1)
2541 /* Delete the link from parent to child. We need to find the lf/lh
2542 * record which contains the offset and remove the offset from that
2543 * record, then decrement the element count in that record, and
2544 * decrement the overall number of subkeys stored in the parent
2547 hive_node_h *unused;
2549 if (get_children (h, parent, &unused, &blocks, GET_CHILDREN_NO_CHECK_NK)== -1)
2554 for (i = 0; blocks[i] != 0; ++i) {
2555 struct ntreg_hbin_block *block =
2556 (struct ntreg_hbin_block *) (h->addr + blocks[i]);
2558 if (block->id[0] == 'l' && (block->id[1] == 'f' || block->id[1] == 'h')) {
2559 struct ntreg_lf_record *lf = (struct ntreg_lf_record *) block;
2561 size_t nr_subkeys_in_lf = le16toh (lf->nr_keys);
2563 for (j = 0; j < nr_subkeys_in_lf; ++j)
2564 if (le32toh (lf->keys[j].offset) + 0x1000 == node) {
2565 for (; j < nr_subkeys_in_lf - 1; ++j)
2566 memcpy (&lf->keys[j], &lf->keys[j+1], sizeof (lf->keys[j]));
2567 lf->nr_keys = htole16 (nr_subkeys_in_lf - 1);
2573 fprintf (stderr, "hivex_node_delete_child: could not find parent"
2574 " to child link\n");
2579 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + parent);
2580 size_t nr_subkeys_in_nk = le32toh (nk->nr_subkeys);
2581 nk->nr_subkeys = htole32 (nr_subkeys_in_nk - 1);
2584 fprintf (stderr, "hivex_node_delete_child: updating nr_subkeys"
2585 " in parent 0x%zx to %zu\n", parent, nr_subkeys_in_nk);
2591 hivex_node_set_values (hive_h *h, hive_node_h node,
2592 size_t nr_values, const hive_set_value *values,
2600 if (!IS_VALID_BLOCK (h, node) || !BLOCK_ID_EQ (h, node, "nk")) {
2605 /* Delete all existing values. */
2606 if (delete_values (h, node) == -1)
2612 /* Allocate value list node. Value lists have no id field. */
2613 static const char nul_id[2] = { 0, 0 };
2615 sizeof (struct ntreg_value_list) + (nr_values - 1) * sizeof (uint32_t);
2616 size_t vallist_offs = allocate_block (h, seg_len, nul_id);
2617 if (vallist_offs == 0)
2620 struct ntreg_nk_record *nk = (struct ntreg_nk_record *) (h->addr + node);
2621 nk->nr_values = htole32 (nr_values);
2622 nk->vallist = htole32 (vallist_offs - 0x1000);
2624 struct ntreg_value_list *vallist =
2625 (struct ntreg_value_list *) (h->addr + vallist_offs);
2628 for (i = 0; i < nr_values; ++i) {
2629 /* Allocate vk record to store this (key, value) pair. */
2630 static const char vk_id[2] = { 'v', 'k' };
2631 seg_len = sizeof (struct ntreg_vk_record) + strlen (values[i].key);
2632 size_t vk_offs = allocate_block (h, seg_len, vk_id);
2636 /* Recalculate pointers that could have been invalidated by
2637 * previous call to allocate_block.
2639 nk = (struct ntreg_nk_record *) (h->addr + node);
2640 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2642 vallist->offset[i] = htole32 (vk_offs - 0x1000);
2644 struct ntreg_vk_record *vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2645 size_t name_len = strlen (values[i].key);
2646 vk->name_len = htole16 (name_len);
2647 strcpy (vk->name, values[i].key);
2648 vk->data_type = htole32 (values[i].t);
2649 uint32_t len = values[i].len;
2650 if (len <= 4) /* store it inline => set MSB flag */
2652 vk->data_len = htole32 (len);
2653 vk->flags = name_len == 0 ? 0 : 1;
2655 if (values[i].len <= 4) /* store it inline */
2656 memcpy (&vk->data_offset, values[i].value, values[i].len);
2658 size_t offs = allocate_block (h, values[i].len + 4, nul_id);
2662 /* Recalculate pointers that could have been invalidated by
2663 * previous call to allocate_block.
2665 nk = (struct ntreg_nk_record *) (h->addr + node);
2666 vallist = (struct ntreg_value_list *) (h->addr + vallist_offs);
2667 vk = (struct ntreg_vk_record *) (h->addr + vk_offs);
2669 memcpy (h->addr + offs + 4, values[i].value, values[i].len);
2670 vk->data_offset = htole32 (offs - 0x1000);
2673 if (name_len * 2 > le32toh (nk->max_vk_name_len))
2674 /* * 2 for UTF16-LE "reencoding" */
2675 nk->max_vk_name_len = htole32 (name_len * 2);
2676 if (values[i].len > le32toh (nk->max_vk_data_len))
2677 nk->max_vk_data_len = htole32 (values[i].len);
2684 hivex_node_set_value (hive_h *h, hive_node_h node,
2685 const hive_set_value *val, int flags)
2687 hive_value_h *prev_values = hivex_node_values (h, node);
2688 if (prev_values == NULL)
2693 size_t nr_values = 0;
2694 for (hive_value_h *itr = prev_values; *itr != 0; ++itr)
2697 hive_set_value *values = malloc ((nr_values + 1) * (sizeof (hive_set_value)));
2699 goto leave_prev_values;
2702 int idx_of_val = -1;
2703 hive_value_h *prev_val;
2704 for (prev_val = prev_values; *prev_val != 0; ++prev_val) {
2708 hive_set_value *value = &values[prev_val - prev_values];
2710 char *valval = hivex_value_value (h, *prev_val, &t, &len);
2711 if (valval == NULL) goto leave_partial;
2714 value->value = valval;
2718 char *valkey = hivex_value_key (h, *prev_val);
2719 if (valkey == NULL) goto leave_partial;
2722 value->key = valkey;
2724 if (STRCASEEQ (valkey, val->key))
2725 idx_of_val = prev_val - prev_values;
2728 if (idx_of_val > -1) {
2729 free (values[idx_of_val].key);
2730 free (values[idx_of_val].value);
2732 idx_of_val = nr_values;
2736 hive_set_value *value = &values[idx_of_val];
2737 *value = (hive_set_value){
2738 .key = strdup (val->key),
2739 .value = malloc (val->len),
2744 if (value->key == NULL || value->value == NULL) goto leave_partial;
2745 memcpy (value->value, val->value, val->len);
2747 retval = hivex_node_set_values (h, node, nr_values, values, 0);
2750 for (int i = 0; i < alloc_ct; i += 2) {
2751 if (values[i / 2].value != NULL)
2752 free (values[i / 2].value);
2753 if (i + 1 < alloc_ct && values[i / 2].key != NULL)
2754 free (values[i / 2].key);