lighttpd 1.4.x https://www.lighttpd.net/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
lighttpd1.4/src/buffer.c

936 lines
16 KiB

#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <assert.h>
#include <ctype.h>
#include "buffer.h"
static const char hex_chars[] = "0123456789abcdef";
/**
* init the buffer
*
*/
buffer* buffer_init(void) {
buffer *b;
b = malloc(sizeof(*b));
assert(b);
b->ptr = NULL;
b->size = 0;
b->used = 0;
return b;
}
/**
* free the buffer
*
*/
void buffer_free(buffer *b) {
if (!b) return;
free(b->ptr);
free(b);
}
void buffer_reset(buffer *b) {
if (!b) return;
/* limit don't reuse buffer larger than ... bytes */
if (b->size > BUFFER_MAX_REUSE_SIZE) {
free(b->ptr);
b->ptr = NULL;
b->size = 0;
}
b->used = 0;
}
/**
*
* allocate (if neccessary) enough space for 'size' bytes and
* set the 'used' counter to 0
*
*/
#define BUFFER_PIECE_SIZE 64
int buffer_prepare_copy(buffer *b, size_t size) {
if (!b) return -1;
if ((0 == b->size) ||
(size > b->size)) {
if (b->size) free(b->ptr);
b->size = size;
/* always allocate a multiply of BUFFER_PIECE_SIZE */
b->size += BUFFER_PIECE_SIZE - (b->size % BUFFER_PIECE_SIZE);
b->ptr = malloc(b->size);
assert(b->ptr);
}
b->used = 0;
return 0;
}
/**
*
* increase the internal buffer (if neccessary) to append another 'size' byte
* ->used isn't changed
*
*/
int buffer_prepare_append(buffer *b, size_t size) {
if (!b) return -1;
if (0 == b->size) {
b->size = size;
/* always allocate a multiply of BUFFER_PIECE_SIZE */
b->size += BUFFER_PIECE_SIZE - (b->size % BUFFER_PIECE_SIZE);
b->ptr = malloc(b->size);
b->used = 0;
assert(b->ptr);
} else if (b->used + size > b->size) {
b->size += size;
/* always allocate a multiply of BUFFER_PIECE_SIZE */
b->size += BUFFER_PIECE_SIZE - (b->size % BUFFER_PIECE_SIZE);
b->ptr = realloc(b->ptr, b->size);
assert(b->ptr);
}
return 0;
}
int buffer_copy_string(buffer *b, const char *s) {
size_t s_len;
if (!s || !b) return -1;
s_len = strlen(s) + 1;
buffer_prepare_copy(b, s_len);
memcpy(b->ptr, s, s_len);
b->used = s_len;
return 0;
}
int buffer_copy_string_len(buffer *b, const char *s, size_t s_len) {
if (!s || !b) return -1;
#if 0
/* removed optimization as we have to keep the empty string
* in some cases for the config handling
*
* url.access-deny = ( "" )
*/
if (s_len == 0) return 0;
#endif
buffer_prepare_copy(b, s_len + 1);
memcpy(b->ptr, s, s_len);
b->ptr[s_len] = '\0';
b->used = s_len + 1;
return 0;
}
int buffer_copy_string_buffer(buffer *b, const buffer *src) {
if (!src) return -1;
if (src->used == 0) {
b->used = 0;
return 0;
}
return buffer_copy_string_len(b, src->ptr, src->used - 1);
}
int buffer_append_string(buffer *b, const char *s) {
size_t s_len;
if (!s || !b) return -1;
s_len = strlen(s);
buffer_prepare_append(b, s_len + 1);
if (b->used == 0)
b->used++;
memcpy(b->ptr + b->used - 1, s, s_len + 1);
b->used += s_len;
return 0;
}
int buffer_append_string_rfill(buffer *b, const char *s, size_t maxlen) {
size_t s_len;
if (!s || !b) return -1;
s_len = strlen(s);
buffer_prepare_append(b, maxlen + 1);
if (b->used == 0)
b->used++;
memcpy(b->ptr + b->used - 1, s, s_len);
if (maxlen > s_len) {
memset(b->ptr + b->used - 1 + s_len, ' ', maxlen - s_len);
}
b->used += maxlen;
b->ptr[b->used - 1] = '\0';
return 0;
}
/**
* append a string to the end of the buffer
*
* the resulting buffer is terminated with a '\0'
* s is treated as a un-terminated string (a \0 is handled a normal character)
*
* @param b a buffer
* @param s the string
* @param s_len size of the string (without the terminating \0)
*/
int buffer_append_string_len(buffer *b, const char *s, size_t s_len) {
if (!s || !b) return -1;
if (s_len == 0) return 0;
buffer_prepare_append(b, s_len + 1);
if (b->used == 0)
b->used++;
memcpy(b->ptr + b->used - 1, s, s_len);
b->used += s_len;
b->ptr[b->used - 1] = '\0';
return 0;
}
int buffer_append_string_buffer(buffer *b, const buffer *src) {
if (!src) return -1;
if (src->used == 0) return 0;
return buffer_append_string_len(b, src->ptr, src->used - 1);
}
int buffer_append_memory(buffer *b, const char *s, size_t s_len) {
if (!s || !b) return -1;
if (s_len == 0) return 0;
buffer_prepare_append(b, s_len);
memcpy(b->ptr + b->used, s, s_len);
b->used += s_len;
return 0;
}
int buffer_copy_memory(buffer *b, const char *s, size_t s_len) {
if (!s || !b) return -1;
b->used = 0;
return buffer_append_memory(b, s, s_len);
}
int buffer_append_hex(buffer *b, unsigned long value) {
char *buf;
int shift = 0;
unsigned long copy = value;
while (copy) {
copy >>= 4;
shift++;
}
if (shift == 0)
shift++;
if (shift & 0x01)
shift++;
buffer_prepare_append(b, shift + 1);
if (b->used == 0)
b->used++;
buf = b->ptr + (b->used - 1);
b->used += shift;
shift <<= 2;
while (shift > 0) {
shift -= 4;
*(buf++) = hex_chars[(value >> shift) & 0x0F];
}
*buf = '\0';
return 0;
}
int ltostr(char *buf, long val) {
char swap;
char *end;
int len = 1;
if (val < 0) {
len++;
*(buf++) = '-';
val = -val;
}
end = buf;
while (val > 9) {
*(end++) = '0' + (val % 10);
val = val / 10;
}
*(end) = '0' + val;
*(end + 1) = '\0';
len += end - buf;
while (buf < end) {
swap = *end;
*end = *buf;
*buf = swap;
buf++;
end--;
}
return len;
}
int buffer_append_long(buffer *b, long val) {
if (!b) return -1;
buffer_prepare_append(b, 32);
if (b->used == 0)
b->used++;
b->used += ltostr(b->ptr + (b->used - 1), val);
return 0;
}
int buffer_copy_long(buffer *b, long val) {
if (!b) return -1;
b->used = 0;
return buffer_append_long(b, val);
}
#if !defined(SIZEOF_LONG) || (SIZEOF_LONG != SIZEOF_OFF_T)
int buffer_append_off_t(buffer *b, off_t val) {
char swap;
char *end;
char *start;
int len = 1;
if (!b) return -1;
buffer_prepare_append(b, 32);
if (b->used == 0)
b->used++;
start = b->ptr + (b->used - 1);
if (val < 0) {
len++;
*(start++) = '-';
val = -val;
}
end = start;
while (val > 9) {
*(end++) = '0' + (val % 10);
val = val / 10;
}
*(end) = '0' + val;
*(end + 1) = '\0';
len += end - start;
while (start < end) {
swap = *end;
*end = *start;
*start = swap;
start++;
end--;
}
b->used += len;
return 0;
}
int buffer_copy_off_t(buffer *b, off_t val) {
if (!b) return -1;
b->used = 0;
return buffer_append_off_t(b, val);
}
#endif /* !defined(SIZEOF_LONG) || (SIZEOF_LONG != SIZEOF_OFF_T) */
char int2hex(char c) {
return hex_chars[(c & 0x0F)];
}
/* converts hex char (0-9, A-Z, a-z) to decimal.
* returns 0xFF on invalid input.
*/
char hex2int(unsigned char hex) {
hex = hex - '0';
if (hex > 9) {
hex = (hex + '0' - 1) | 0x20;
hex = hex - 'a' + 11;
}
if (hex > 15)
hex = 0xFF;
return hex;
}
/**
* init the buffer
*
*/
buffer_array* buffer_array_init(void) {
buffer_array *b;
b = malloc(sizeof(*b));
assert(b);
b->ptr = NULL;
b->size = 0;
b->used = 0;
return b;
}
/**
* free the buffer_array
*
*/
void buffer_array_free(buffer_array *b) {
size_t i;
if (!b) return;
for (i = 0; i < b->size; i++) {
if (b->ptr[i]) buffer_free(b->ptr[i]);
}
free(b->ptr);
free(b);
}
buffer *buffer_array_append_get_buffer(buffer_array *b) {
size_t i;
if (b->size == 0) {
b->size = 16;
b->ptr = malloc(sizeof(*b->ptr) * b->size);
assert(b->ptr);
for (i = 0; i < b->size; i++) {
b->ptr[i] = NULL;
}
} else if (b->size == b->used) {
b->size += 16;
b->ptr = realloc(b->ptr, sizeof(*b->ptr) * b->size);
assert(b->ptr);
for (i = b->used; i < b->size; i++) {
b->ptr[i] = NULL;
}
}
if (b->ptr[b->used] == NULL) {
b->ptr[b->used] = buffer_init();
}
b->ptr[b->used]->used = 0;
return b->ptr[b->used++];
}
char * buffer_search_string_len(buffer *b, const char *needle, size_t len) {
size_t i;
if (len == 0) return NULL;
if (needle == NULL) return NULL;
if (b->used < len) return NULL;
for(i = 0; i < b->used - len; i++) {
if (0 == memcmp(b->ptr + i, needle, len)) {
return b->ptr + i;
}
}
return NULL;
}
buffer *buffer_init_string(const char *str) {
buffer *b = buffer_init();
buffer_copy_string(b, str);
return b;
}
int buffer_is_empty(buffer *b) {
return (b->used == 0);
}
/**
* check if two buffer contain the same data
*
* this is a optimized 32/64bit compare function.
*
* it is assumed that the leftmost byte have the most equality.
* That why the comparision is done right to left
*
*/
int buffer_is_equal(buffer *a, buffer *b) {
size_t i;
if (a->used != b->used) return 0;
if (a->used == 0) return 1;
for (i = a->used - 1; i < a->used && i % (sizeof(size_t)); i --) {
if (a->ptr[i] != b->ptr[i]) return 0;
}
for (i -= (sizeof(size_t)); i < a->used; i -= (sizeof(size_t))) {
if (*((size_t *)(a->ptr + i)) !=
*((size_t *)(b->ptr + i))) return 0;
}
return 1;
}
int buffer_is_equal_string(buffer *a, const char *s, size_t b_len) {
buffer b;
b.ptr = (char *)s;
b.used = b_len + 1;
return buffer_is_equal(a, &b);
}
/* simple-assumption:
*
* most parts are equal and doing a case conversion needs time
*
*/
int buffer_caseless_compare(const char *a, size_t a_len, const char *b, size_t b_len) {
size_t ndx = 0, max_ndx;
size_t *al, *bl;
size_t mask = sizeof(*al) - 1;
al = (size_t *)a;
bl = (size_t *)b;
/* is the alignment correct ? */
if ( ((size_t)al & mask) == 0 &&
((size_t)bl & mask) == 0 ) {
max_ndx = ((a_len < b_len) ? a_len : b_len) & ~mask;
for (; ndx < max_ndx; ndx += sizeof(*al)) {
if (*al != *bl) break;
al++; bl++;
}
}
a = (char *)al;
b = (char *)bl;
max_ndx = ((a_len < b_len) ? a_len : b_len);
for (; ndx < max_ndx; ndx++) {
char a1 = *a++, b1 = *b++;
if (a1 != b1) {
if ((a1 >= 'A' && a1 <= 'Z') && (b1 >= 'a' && b1 <= 'z'))
a1 |= 32;
else if ((a1 >= 'a' && a1 <= 'z') && (b1 >= 'A' && b1 <= 'Z'))
b1 |= 32;
if ((a1 - b1) != 0) return (a1 - b1);
}
}
return 0;
}
/**
* check if the rightmost bytes of the string are equal.
*
*
*/
int buffer_is_equal_right_len(buffer *b1, buffer *b2, size_t len) {
/* no, len -> equal */
if (len == 0) return 1;
/* len > 0, but empty buffers -> not equal */
if (b1->used == 0 || b2->used == 0) return 0;
/* buffers too small -> not equal */
if (b1->used - 1 < len || b1->used - 1 < len) return 0;
if (0 == strncmp(b1->ptr + b1->used - 1 - len,
b2->ptr + b2->used - 1 - len, len)) {
return 1;
}
return 0;
}
int buffer_copy_string_hex(buffer *b, const char *in, size_t in_len) {
size_t i;
/* BO protection */
if (in_len * 2 < in_len) return -1;
buffer_prepare_copy(b, in_len * 2 + 1);
for (i = 0; i < in_len; i++) {
b->ptr[b->used++] = hex_chars[(in[i] >> 4) & 0x0F];
b->ptr[b->used++] = hex_chars[in[i] & 0x0F];
}
b->ptr[b->used++] = '\0';
return 0;
}
int buffer_append_string_hex(buffer *b, const char *in, size_t in_len) {
size_t i;
/* BO protection */
if (in_len * 2 < in_len) return -1;
if (b->used > 0 && b->ptr[b->used - 1] == '\0') {
b->used--;
}
buffer_prepare_append(b, in_len * 2 + 1);
for (i = 0; i < in_len; i++) {
b->ptr[b->used++] = hex_chars[(in[i] >> 4) & 0x0F];
b->ptr[b->used++] = hex_chars[in[i] & 0x0F];
}
b->ptr[b->used++] = '\0';
return 0;
}
int buffer_append_string_url_encoded(buffer *b, const char *s) {
unsigned char *ds, *d;
size_t d_len;
if (!s || !b) return -1;
if (b->ptr[b->used - 1] != '\0') {
SEGFAULT();
}
/* count to-be-encoded-characters */
for (ds = (unsigned char *)s, d_len = 0; *ds; ds++) {
if (*ds < 32 || *ds > 126) {
d_len += 3;
} else {
switch (*ds) {
case '$':
case '&':
case '+':
case ',':
case '/':
case ':':
case ';':
case '=':
case '?':
case '@':
case ' ':
case '#':
case '%':
case '<':
case '>':
case '"':
case '\'':
d_len += 3;
break;
default:
d_len ++;
break;
}
}
}
buffer_prepare_append(b, d_len);
for (ds = (unsigned char *)s, d = (unsigned char *)b->ptr + b->used - 1, d_len = 0; *ds; ds++) {
if (*ds < 32 || *ds > 126) {
d[d_len++] = '%';
d[d_len++] = hex_chars[((*ds) >> 4) & 0x0F];
d[d_len++] = hex_chars[(*ds) & 0x0F];
} else {
switch (*ds) {
case '$':
case '&':
case '+':
case ',':
case '/':
case ':':
case ';':
case '=':
case '?':
case '@':
case ' ':
case '#':
case '%':
case '<':
case '>':
case '"':
case '\'':
d[d_len++] = '%';
d[d_len++] = hex_chars[((*ds) >> 4) & 0x0F];
d[d_len++] = hex_chars[(*ds) & 0x0F];
break;
default:
d[d_len++] = *ds;
break;
}
}
}
b->ptr[b->used + d_len - 1] = '\0';
b->used += d_len;
return 0;
}
int buffer_append_string_html_encoded(buffer *b, const char *s) {
unsigned char *ds, *d;
size_t d_len;
if (!s || !b) return -1;
if (b->ptr[b->used - 1] != '\0') {
SEGFAULT();
}
/* count to-be-encoded-characters */
for (ds = (unsigned char *)s, d_len = 0; *ds; ds++) {
d_len++;
if (*ds == '<' || *ds == '>')
d_len += 4 - 1;
else if (*ds == '&')
d_len += 5 - 1;
}
buffer_prepare_append(b, d_len);
for (ds = (unsigned char *)s, d = (unsigned char *)b->ptr + b->used - 1, d_len = 0; *ds; ds++) {
switch (*ds) {
case '>':
d[d_len++] = '&';
d[d_len++] = 'g';
d[d_len++] = 't';
d[d_len++] = ';';
break;
case '<':
d[d_len++] = '&';
d[d_len++] = 'l';
d[d_len++] = 't';
d[d_len++] = ';';
break;
case '&':
d[d_len++] = '&';
d[d_len++] = 'a';
d[d_len++] = 'm';
d[d_len++] = 'p';
d[d_len++] = ';';
break;
default:
d[d_len++] = *ds;
break;
}
}
b->ptr[b->used + d_len - 1] = '\0';
b->used += d_len;
return 0;
}
/* decodes url-special-chars inplace.
* replaces non-printable characters with '_'
*/
int buffer_urldecode(buffer *url) {
unsigned char high, low;
const char *src;
char *dst;
if (!url || !url->ptr) return -1;
src = (const char*) url->ptr;
dst = (char*) url->ptr;
while ((*src) != '\0') {
#if 1
if (*src == '+') {
*dst = ' ';
} else
#endif
if (*src == '%') {
*dst = '%';
high = hex2int(*(src + 1));
if (high != 0xFF) {
low = hex2int(*(src + 2));
if (low != 0xFF) {
high = (high << 4) | low;
/* map control-characters out */
if (high < 32 || high == 127) high = '_';
*dst = high;
src += 2;
}
}
} else {
*dst = *src;
}
dst++;
src++;
}
*dst = '\0';
url->used = (dst - url->ptr) + 1;
return 0;
}
/* Remove "/../", "//", "/./" parts from path.
*
* /blah/.. gets /
* /blah/../foo gets /foo
* /abc/./xyz gets /abc/xyz
* /abc//xyz gets /abc/xyz
*
* NOTE: src and dest can point to the same buffer, in which case,
* the operation is performed in-place.
*/
int buffer_path_simplify(buffer *dest, buffer *src)
{
int toklen;
char c, pre1;
char *start, *slash, *walk, *out;
unsigned short pre;
if (src == NULL || src->ptr == NULL || dest == NULL)
return -1;
if (src == dest)
buffer_prepare_append(dest, 1);
else
buffer_prepare_copy(dest, src->used + 1);
walk = src->ptr;
start = dest->ptr;
out = dest->ptr;
slash = dest->ptr;
while (*walk == ' ') {
walk++;
}
pre1 = *(walk++);
c = *(walk++);
pre = pre1;
if (pre1 != '/') {
pre = ('/' << 8) | pre1;
*(out++) = '/';
}
*(out++) = pre1;
while (1) {
if (c == '/' || c == '\0') {
toklen = out - slash;
if (toklen == 3 && pre == (('.' << 8) | '.')) {
out = slash;
if (out > start) {
out--;
while (out > start && *out != '/') {
out--;
}
}
if (c == '\0')
out++;
} else if (toklen == 1 || pre == (('/' << 8) | '.')) {
out = slash;
if (c == '\0')
out++;
}
slash = out;
}
if (c == '\0')
break;
pre1 = c;
pre = (pre << 8) | pre1;
c = *walk;
*out = pre1;
out++;
walk++;
}
*out = '\0';
dest->used = (out - start) + 1;
return 0;
}
inline int light_isdigit(int c) {
return (c >= '0' && c <= '9');
}
inline int light_isxdigit(int c) {
if (light_isdigit(c)) return 1;
c |= 32;
return (c >= 'a' && c <= 'f');
}
inline int light_isalpha(int c) {
c |= 32;
return (c >= 'a' && c <= 'z');
}
inline int light_isalnum(int c) {
return light_isdigit(c) || light_isalpha(c);
}