* along with these programs. If not, see <http://www.gnu.org/licenses/>.
*/
-/* Integers are stored as 32 bit unsigned in "network" order, i.e. MSB first.
- All files start with a magic number.
+/* see documentation in libkmod/libkmod-index.c */
- Magic spells "BOOTFAST". Second one used on newer versioned binary files.
- */
-/* #define INDEX_MAGIC_OLD 0xB007FA57 */
#define INDEX_MAGIC 0xB007F457
-
-/* We use a version string to keep track of changes to the binary format
- * This is stored in the form: INDEX_MAJOR (hi) INDEX_MINOR (lo) just in
- * case we ever decide to have minor changes that are not incompatible.
- */
-
#define INDEX_VERSION_MAJOR 0x0002
#define INDEX_VERSION_MINOR 0x0001
#define INDEX_VERSION ((INDEX_VERSION_MAJOR<<16)|INDEX_VERSION_MINOR)
-
-/* The index file maps keys to values. Both keys and values are ASCII strings.
- Each key can have multiple values. Values are sorted by an integer priority.
-
- The reader also implements a wildcard search (including range expressions)
- where the keys in the index are treated as patterns.
- This feature is required for module aliases.
-*/
-
-/* Implementation is based on a radix tree, or "trie".
- Each arc from parent to child is labelled with a character.
- Each path from the root represents a string.
-
- == Example strings ==
-
- ask
- ate
- on
- once
- one
-
- == Key ==
- + Normal node
- * Marked node, representing a key and it's values.
-
- +
- |-a-+-s-+-k-*
- | |
- | `-t-+-e-*
- |
- `-o-+-n-*-c-+-e-*
- |
- `-e-*
-
- Naive implementations tend to be very space inefficient; child pointers
- are stored in arrays indexed by character, but most child pointers are null.
-
- Our implementation uses a scheme described by Wikipedia as a Patrica trie,
-
- "easiest to understand as a space-optimized trie where
- each node with only one child is merged with its child"
-
- +
- |-a-+-sk-*
- | |
- | `-te-*
- |
- `-on-*-ce-*
- |
- `-e-*
-
- We still use arrays of child pointers indexed by a single character;
- the remaining characters of the label are stored as a "prefix" in the child.
-
- The paper describing the original Patrica trie works on individiual bits -
- each node has a maximum of two children, which increases space efficiency.
- However for this application it is simpler to use the ASCII character set.
- Since the index file is read-only, it can be compressed by omitting null
- child pointers at the start and end of arrays.
-*/
-
-#define INDEX_PRIORITY_MIN UINT32_MAX
+#define INDEX_CHILDMAX 128
struct index_value {
struct index_value *next;
};
/* In-memory index (depmod only) */
-
-#define INDEX_CHILDMAX 128
struct index_node {
char *prefix; /* path compression */
struct index_value *values;
struct index_node *children[INDEX_CHILDMAX]; /* indexed by character */
};
-/* Disk format:
-
- uint32_t magic = INDEX_MAGIC;
- uint32_t version = INDEX_VERSION;
- uint32_t root_offset;
-
- (node_offset & INDEX_NODE_MASK) specifies the file offset of nodes:
-
- char[] prefix; // nul terminated
-
- char first;
- char last;
- uint32_t children[last - first + 1];
-
- uint32_t value_count;
- struct {
- uint32_t priority;
- char[] value; // nul terminated
- } values[value_count];
-
- (node_offset & INDEX_NODE_FLAGS) indicates which fields are present.
- Empty prefixes are omitted, leaf nodes omit the three child-related fields.
-
- This could be optimised further by adding a sparse child format
- (indicated using a new flag).
- */
/* Format of node offsets within index file */
enum node_offset {