485 lines
15 KiB
C
485 lines
15 KiB
C
#include <stdio.h>
|
|
|
|
/* file_context_node
|
|
* A node used in a linked list of file contexts.
|
|
* Each node contains the regular expression, the type and
|
|
* the context, as well as information about the regular
|
|
* expression. The regular expression data (meta, stem_len
|
|
* and str_len) can be filled in by using the fc_fill_data
|
|
* function after the regular expression has been loaded.
|
|
* next points to the next node in the linked list.
|
|
*/
|
|
struct file_context_node {
|
|
char* regex;
|
|
char* type;
|
|
char* context;
|
|
int meta;
|
|
int stem_len;
|
|
int str_len;
|
|
|
|
struct file_context_node* next;
|
|
};
|
|
|
|
|
|
|
|
/* file_context_bucket
|
|
* A node used in a linked list of buckets that contain
|
|
* file_context_node's.
|
|
* Each node contains a pointer to a file_context_node which
|
|
* is the header of its linked list. This linked list is the
|
|
* content of this bucket.
|
|
* next points to the next bucket in the linked list.
|
|
*/
|
|
struct file_context_bucket {
|
|
struct file_context_node* data;
|
|
|
|
struct file_context_bucket* next;
|
|
};
|
|
|
|
|
|
|
|
/* fc_merge
|
|
* Merges two sorted file context linked lists into one
|
|
* sorted one.
|
|
* Pass two lists a and b, and after the completion of fc_merge,
|
|
* the final list is contained in a, and b is empty.
|
|
*/
|
|
struct file_context_node* fc_merge( struct file_context_node* a, struct file_context_node* b )
|
|
{
|
|
struct file_context_node* a_current;
|
|
struct file_context_node* b_current;
|
|
struct file_context_node* temp;
|
|
struct file_context_node* jumpto;
|
|
|
|
/* If a is a empty list, and b is not,
|
|
* set a as b and proceed to the end. */
|
|
if( !a && b )
|
|
a = b;
|
|
/* If b is an empty list, leave a as it is. */
|
|
else if( !b ) { }
|
|
else {
|
|
/* Make it so the list a has the lesser
|
|
* first element always. */
|
|
if( fc_compare( a, b ) == 1 ) {
|
|
temp = a;
|
|
a = b;
|
|
b = temp;
|
|
}
|
|
a_current = a;
|
|
b_current = b;
|
|
|
|
/* Merge by inserting b's nodes inbetween a's nodes. */
|
|
while( a_current->next && b_current ) {
|
|
jumpto = a_current->next;
|
|
|
|
/* Insert b's nodes inbetween the current a node
|
|
* and the next a node.*/
|
|
while( b_current && a_current->next &&
|
|
fc_compare( a_current->next, b_current) != -1 ) { temp = a_current->next;
|
|
a_current->next = b_current;
|
|
b_current = b_current->next;
|
|
a_current->next->next = temp;
|
|
a_current = a_current->next;
|
|
}
|
|
|
|
/* Skip all the inserted node from b to the
|
|
* next node in the original a. */
|
|
a_current = jumpto;
|
|
}
|
|
|
|
|
|
/* if there is anything left in b to be inserted,
|
|
put it on the end */
|
|
if( b_current ) {
|
|
a_current->next = b_current;
|
|
}
|
|
}
|
|
|
|
b = NULL;
|
|
|
|
return a;
|
|
}
|
|
|
|
|
|
|
|
/* fc_merge_sort
|
|
* Sorts file contexts from least specific to more specific.
|
|
* The bucket linked list is passed and after the completion
|
|
* of the fc_merge_sort function, there is only one bucket
|
|
* (pointed to by master) that contains a linked list
|
|
* of all the file contexts, in sorted order.
|
|
* Explanation of the algorithm:
|
|
* The algorithm implemented in fc_merge_sort is an iterative
|
|
* implementation of merge sort.
|
|
* At first, each bucket has a linked list of file contexts
|
|
* that are 1 element each.
|
|
* Each pass, each odd numbered bucket is merged into the bucket
|
|
* before it. This halves the number of buckets each pass.
|
|
* It will continue passing over the buckets (as described above)
|
|
* until there is only one bucket left, containing the list of
|
|
* file contexts, sorted.
|
|
*/
|
|
void fc_merge_sort( struct file_context_bucket* master )
|
|
{
|
|
int i;
|
|
|
|
struct file_context_bucket* current;
|
|
struct file_context_bucket* temp;
|
|
|
|
struct file_context_node* ncurrent;
|
|
struct file_context_node* ntemp;
|
|
|
|
/* Loop until master is the only bucket left
|
|
* so that this will stop when master contains
|
|
* the sorted list. */
|
|
while( master->next ) {
|
|
current = master;
|
|
|
|
/* This loop merges buckets two-by-two. */
|
|
while( current ) {
|
|
if( current->next ) {
|
|
/* Merge the next one into the current one. */
|
|
current->data = fc_merge( current->data, current->next->data );
|
|
/* remove the next bucket that is now empty. */
|
|
temp = current->next;
|
|
current->next = current->next->next;
|
|
free( temp );
|
|
}
|
|
current = current->next;
|
|
}
|
|
}
|
|
}
|
|
|
|
/* fc_compare
|
|
* Compares two file contexts' regular expressions and returns:
|
|
* -1 if a is less specific than b
|
|
* 0 if a and be are equally specific
|
|
* 1 if a is more specific than b
|
|
* The comparison is based on the following statements,
|
|
* in order from most important to least important, given a and b:
|
|
* If a is a regular expression and b is not,
|
|
* -> a is less specific than b.
|
|
* If a's stem length is shorter than b's stem length,
|
|
* -> a is less specific than b.
|
|
* If a's string length is shorter than b's string length,
|
|
* -> a is less specific than b.
|
|
* If a does not have a specified type and b does not,
|
|
* -> a is less specific than b.
|
|
*/
|
|
int fc_compare( struct file_context_node* a, struct file_context_node* b )
|
|
{
|
|
/* Check to see if either a or b have meta characters
|
|
* and the other doesn't. */
|
|
if( a->meta && !b->meta )
|
|
return -1;
|
|
if( b->meta && !a->meta )
|
|
return 1;
|
|
|
|
/* Check to see if either a or b have a shorter stem
|
|
* length than the other. */
|
|
if( a->stem_len < b->stem_len )
|
|
return -1;
|
|
if( b->stem_len < a->stem_len )
|
|
return 1;
|
|
|
|
/* Check to see if either a or b have a shorter string
|
|
* length than the other. */
|
|
if( a->str_len < b->str_len )
|
|
return -1;
|
|
if( b->str_len < b->str_len )
|
|
return 1;
|
|
|
|
/* Check to see if either a or b has a specified type
|
|
* and the other doesn't. */
|
|
if( !a->type && b->type )
|
|
return -1;
|
|
if( !b->type && a->type )
|
|
return 1;
|
|
|
|
/* If none of the above conditions were satisfied,
|
|
* then a and b are equally specific. */
|
|
return 0;
|
|
}
|
|
|
|
|
|
|
|
/* fc_fill_data
|
|
* This processes a regular expression in a file context
|
|
* and sets the data held in file_context_node, namely
|
|
* meta, str_len and stem_len.
|
|
* The following changes are made to fc_node after the
|
|
* the completion of the function:
|
|
* fc_node->meta = 1 if regex has a meta character,
|
|
* 0 if not.
|
|
* fc_node->str_len = The string length of the regular
|
|
* expression.
|
|
* fc_node->stem_len = The number of characters up until
|
|
* the first meta character.
|
|
*/
|
|
void fc_fill_data( struct file_context_node* fc_node )
|
|
{
|
|
int c = 0;
|
|
|
|
fc_node->meta = 0;
|
|
fc_node->stem_len = 0;
|
|
fc_node->str_len = 0;
|
|
|
|
/* Process until the string termination character
|
|
* has been reached.
|
|
* Note: this while loop has been adapted from
|
|
* spec_hasMetaChars in matchpathcon.c from
|
|
* libselinux-1.22. */
|
|
while( fc_node->regex[c] != 0 ) {
|
|
switch( fc_node->regex[c] ) {
|
|
case '.':
|
|
case '^':
|
|
case '$':
|
|
case '?':
|
|
case '*':
|
|
case '+':
|
|
case '|':
|
|
case '[':
|
|
case '(':
|
|
case '{':
|
|
/* If a meta character is found,
|
|
* set meta to one */
|
|
fc_node->meta = 1;
|
|
break;
|
|
case '\\':
|
|
/* If a escape character is found,
|
|
* skip the next character. */
|
|
c++;
|
|
default:
|
|
/* If no meta character has been found yet,
|
|
* add one to the stem length. */
|
|
if( !fc_node->meta ) fc_node->stem_len++;
|
|
break;
|
|
}
|
|
|
|
fc_node->str_len++;
|
|
c++;
|
|
}
|
|
}
|
|
|
|
/* main
|
|
* This program takes in two arguments, the input filename and the
|
|
* output filename. The input file should be syntactically correct.
|
|
* Overall what is done in the main is read in the file and store each
|
|
* line of code, sort it, then output it to the output file.
|
|
*/
|
|
int main( int argc, char *argv[])
|
|
{
|
|
int i, j, lines;
|
|
int start, finish;
|
|
char* str;
|
|
struct file_context_node* temp;
|
|
struct file_context_node* head;
|
|
struct file_context_node* current;
|
|
struct file_context_node* array;
|
|
struct file_context_bucket* master;
|
|
struct file_context_bucket* bcurrent;
|
|
|
|
FILE *path;
|
|
char line_buf[ 127 ];
|
|
|
|
/* Check for the correct number of command line arguments. */
|
|
if( argc != 3 ) {
|
|
printf( "Error: invalid number of command line arguments.\n" );
|
|
return -1;
|
|
}
|
|
|
|
i = j = lines = 0;
|
|
|
|
/* Allocate the head of the file_context linked list. */
|
|
if( !( current = head = (struct file_context_node*)malloc( sizeof( struct file_context_node ) ) ) ) {
|
|
printf( "Error: failure allocating memory.\n" );
|
|
return -1;
|
|
}
|
|
|
|
/* Make sure to have a terminating character, always. */
|
|
line_buf[127] = 0;
|
|
|
|
/* Open the input file. */
|
|
if( !( path = fopen( argv[1], "r" ) ) ) {
|
|
printf( "Error: failure opening input file for read.\n" );
|
|
return -1;
|
|
}
|
|
|
|
/* Parse the file into a file_context linked list. */
|
|
while( fgets( line_buf, 126, path ) != NULL ) {
|
|
|
|
/* Get rid of whitespace from the front of the line. */
|
|
i = 0;
|
|
while( line_buf[i] && line_buf[i] <= ' ' ) i++;
|
|
|
|
/* Check if the line isn't empty and isn't a comment */
|
|
if( line_buf[i] && line_buf[i] != '#' ) {
|
|
/* Allocate a new node. */
|
|
temp = (struct file_context_node*)malloc( sizeof( struct file_context_node ) );
|
|
if( !temp ) {
|
|
printf( "Error: failure allocating memory.\n" );
|
|
return -1;
|
|
}
|
|
temp->next = NULL;
|
|
|
|
/* Parse out the regular expression from the line. */
|
|
start = i;
|
|
while( line_buf[i] > ' ' )i++;
|
|
finish = i;
|
|
|
|
/* Allocate a character array to hold the regular
|
|
* expression. */
|
|
temp->regex = (char*)malloc( sizeof( char ) * ( finish - start + 1) );
|
|
if( !( temp->regex ) ) {
|
|
printf( "Error: failure allocating memory.\n" );
|
|
return -1;
|
|
}
|
|
temp->regex[0] = 0;
|
|
|
|
/* Fill the regular expression array. */
|
|
temp->regex[ ( finish - start ) ] = 0;
|
|
for( j = 0; j < finish - start; j++ ) {
|
|
temp->regex[j] = line_buf[j + start];
|
|
}
|
|
|
|
/* Get rid of whitespace after the regular
|
|
* expression. */
|
|
while( line_buf[i] <= ' ' ) i++;
|
|
|
|
/* Parse out the type from the line (if it
|
|
* is there). */
|
|
if( line_buf[i] == '-' ) {
|
|
/* Allocate a character array to
|
|
* hold the type. */
|
|
temp->type = (char*)malloc( sizeof( char ) * 3 );
|
|
if( !( temp->type ) ) {
|
|
printf( "Error: failure allocating memory.\n" );
|
|
return -1;
|
|
}
|
|
|
|
/* Fill the type into the array. */
|
|
temp->type[0] = line_buf[i];
|
|
temp->type[1] = line_buf[i + 1];
|
|
i += 2;
|
|
temp->type[2] = 0;
|
|
|
|
/* Get rid of whitespace after the type. */
|
|
while( line_buf[i] <= ' ' ) i++;
|
|
}
|
|
|
|
/* Parse out the context from the line. */
|
|
start = i;
|
|
while( line_buf[i] > ' ' ) i++;
|
|
finish = i;
|
|
|
|
/* Allocate a character array to hold the context. */
|
|
temp->context = (char*)malloc( sizeof( char ) * ( finish - start + 1 ) );
|
|
if( !( temp->context ) ) {
|
|
printf( "Error: failure allocating memory.\n" );
|
|
return -1;
|
|
}
|
|
temp->context[0] = 0;
|
|
|
|
/* Fill the context array. */
|
|
temp->context[ ( finish - start ) ] = 0;
|
|
for( j = 0; j < finish - start; j++ ) {
|
|
temp->context[j] = line_buf[j + start];
|
|
}
|
|
|
|
/* Set all the data about the regular
|
|
* expression. */
|
|
fc_fill_data( temp );
|
|
|
|
/* Link this line of code at the end of
|
|
* the linked list. */
|
|
current->next = temp;
|
|
current = current->next;
|
|
lines++;
|
|
}
|
|
}
|
|
fclose( path );
|
|
|
|
/* Create the bucket linked list from the earlier linked list. */
|
|
current = head->next;
|
|
bcurrent = master = (struct file_context_bucket*)malloc( sizeof( struct file_context_bucket ) );
|
|
/* Go until all the nodes have been put in individual buckets. */
|
|
while( current ) {
|
|
/* Copy over the file context line into the bucket. */
|
|
bcurrent->data = current;
|
|
current = current->next;
|
|
|
|
/* Detatch the node in the bucket from the old list. */
|
|
bcurrent->data->next = NULL;
|
|
|
|
/* If there should be another bucket, put one at the end. */
|
|
if( current ) {
|
|
bcurrent->next = (struct file_context_bucket*) malloc( sizeof( struct file_context_bucket ) );
|
|
if( !( bcurrent->next ) ) {
|
|
printf( "Error: failure allocating memory.\n" );
|
|
return -1;
|
|
}
|
|
|
|
/* Make sure the new bucket thinks it's the end of the
|
|
* list. */
|
|
bcurrent->next->next = NULL;
|
|
|
|
bcurrent = bcurrent->next;
|
|
}
|
|
}
|
|
|
|
/* Sort the bucket list. */
|
|
fc_merge_sort( master );
|
|
|
|
/* Open the output file. */
|
|
if( !(path = fopen( argv[2], "w" ) ) ) {
|
|
printf( "Error: failure opening output file for write.\n" );
|
|
return -1;
|
|
}
|
|
|
|
/* Output the sorted file_context linked list to the output file. */
|
|
current = master->data;
|
|
while( current ) {
|
|
/* Output the regular expression. */
|
|
i = 0;
|
|
while( current->regex[i] != 0 ) {
|
|
fprintf( path, "%c", current->regex[i] );
|
|
i++;
|
|
}
|
|
fprintf( path, "\t" );
|
|
|
|
/* Output the type, if there is one. */
|
|
if( current->type ) {
|
|
i = 0;
|
|
while( current->type[i] != 0 ) {
|
|
fprintf( path, "%c", current->type[i] );
|
|
i++;
|
|
}
|
|
fprintf( path, "\t" );
|
|
}
|
|
|
|
/* Output the context. */
|
|
i = 0;
|
|
while( current->context[i] != 0 ) {
|
|
fprintf( path, "%c", current->context[i] );
|
|
i++;
|
|
}
|
|
fprintf( path, "\n" );
|
|
|
|
/* Remove the node. */
|
|
temp = current;
|
|
current = current->next;
|
|
|
|
free( temp->regex );
|
|
if( temp->type)
|
|
free( temp->type );
|
|
free( temp->context );
|
|
free( temp );
|
|
}
|
|
free( master );
|
|
|
|
fclose( path );
|
|
|
|
return 0;
|
|
}
|
|
|