1
0
mirror of https://github.com/vlang/v.git synced 2023-08-10 21:13:21 +03:00
v/vlib/crypto/pem/decode.v

91 lines
3.1 KiB
V
Raw Normal View History

2023-01-19 18:01:11 +03:00
module pem
import encoding.base64
// decode_only reads `data` and returns the first parsed PEM Block. `none` is returned
// when a header is expected, but not present or when a start of '-----BEGIN' or end of '-----END'
// can't be found.
//
// use decode if you still need the unparsed rest of the string.
[inline]
pub fn decode_only(data string) ?Block {
block, _ := decode_internal(data)?
return block
}
// decode reads `data` and returns the first parsed PEM Block along with the rest of
2023-01-19 18:01:11 +03:00
// the string. `none` is returned when a header is expected, but not present
// or when a start of '-----BEGIN' or end of '-----END' can't be found.
//
// use decode_only if you do not need the unparsed rest of the string.
[direct_array_access; inline]
2023-01-19 18:01:11 +03:00
pub fn decode(data string) ?(Block, string) {
block, rest := decode_internal(data)?
return block, rest[rest.index(pem_end)? + pem_end.len..].all_after_first(pem_eol)
}
// decode_internal allows `decode` variations to deal with the rest of the data as
// they want to. for example Block.decode could have hindered performance with the final
// indexing into `rest` that `decode_partial` does.
[direct_array_access]
fn decode_internal(data string) ?(Block, string) {
// direct_array_access safety: since we use the string.index method here,
// we won't get an invalid index since it would otherwise return `none`
2023-01-19 18:01:11 +03:00
mut rest := data[data.index(pem_begin)?..]
mut block := Block.new(rest[pem_begin.len..].all_before(pem_eol))
2023-01-19 18:01:11 +03:00
block.headers, rest = parse_headers(rest[pem_begin.len..].all_after(pem_eol).trim_left(' \n\t\v\f\r'))?
block_end_index := rest.index(pem_end)?
b64_data := rest[..block_end_index].replace_each(['\r', '', '\n', '', '\t', '', ' ', ''])
block_data_len := block_end_index / 4 * 3
block.data = []u8{len: block_data_len, cap: block_data_len + 3, init: 0}
decoded_len := base64.decode_in_buffer(&b64_data, &block.data[0])
block.data = block.data[..decoded_len]
return block, rest
2023-01-19 18:01:11 +03:00
}
[direct_array_access]
2023-01-19 18:01:11 +03:00
fn parse_headers(block string) ?(map[string][]string, string) {
headers_str := block.all_before(pem_end).all_before('\n\n')
// check that something was split or if it's empty
if headers_str.len == block.all_before(pem_end).len || headers_str.len == 0 {
return map[string][]string{}, block
}
// seperate lines instead of iterating over them,
// so that we can manually index them
headers_seperated := headers_str.split_into_lines()
// index the key/value separator ':', otherwise
2023-01-19 18:01:11 +03:00
// return none because it should exist
// the initialisation of this function already tells us headers are present
mut colon_index := headers_seperated[0].index(colon) or { return none }
mut headers := map[string][]string{}
mut index := 0
for index < headers_seperated.len - 1 {
line := headers_seperated[index]
if line.len == 0 {
break
}
key := line[..colon_index].trim_space()
mut val := line[colon_index + 1..].trim_space()
for colon_index = 0; index < headers_seperated.len - 1 && colon_index == 0; {
index++
colon_index = headers_seperated[index].index(colon) or {
val += headers_seperated[index].trim_space()
0
}
}
headers[key] << val
}
return headers, block.all_after('\n\n')
}