2021-02-28 01:18:25 +03:00
|
|
|
module vweb
|
|
|
|
|
|
|
|
import io
|
2021-03-04 19:00:03 +03:00
|
|
|
import strings
|
2021-02-28 01:18:25 +03:00
|
|
|
import net.http
|
|
|
|
import net.urllib
|
|
|
|
|
2021-03-04 19:00:03 +03:00
|
|
|
fn parse_request(mut reader io.BufferedReader) ?http.Request {
|
2021-02-28 01:18:25 +03:00
|
|
|
// request line
|
|
|
|
mut line := reader.read_line() ?
|
|
|
|
method, target, version := parse_request_line(line) ?
|
|
|
|
|
|
|
|
// headers
|
2021-03-01 13:50:52 +03:00
|
|
|
mut h := http.new_header()
|
2021-02-28 01:18:25 +03:00
|
|
|
line = reader.read_line() ?
|
|
|
|
for line != '' {
|
2021-03-01 13:50:52 +03:00
|
|
|
key, value := parse_header(line) ?
|
|
|
|
h.add_str(key, value) ?
|
2021-02-28 01:18:25 +03:00
|
|
|
line = reader.read_line() ?
|
|
|
|
}
|
|
|
|
|
2021-03-01 13:50:52 +03:00
|
|
|
// create map[string]string from headers
|
|
|
|
// TODO: replace headers and lheaders with http.Header type
|
|
|
|
mut headers := map[string]string{}
|
|
|
|
mut lheaders := map[string]string{}
|
|
|
|
for key in h.keys() {
|
|
|
|
values := h.values_str(key).join('; ')
|
|
|
|
headers[key] = values
|
|
|
|
lheaders[key.to_lower()] = values
|
2021-02-28 01:18:25 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// body
|
2021-03-02 22:02:17 +03:00
|
|
|
mut body := []byte{}
|
2021-03-01 13:50:52 +03:00
|
|
|
if length := h.get(.content_length) {
|
|
|
|
n := length.int()
|
2021-03-02 22:02:17 +03:00
|
|
|
if n > 0 {
|
|
|
|
body = []byte{len: n}
|
2021-03-06 22:04:51 +03:00
|
|
|
reader.read(mut body) or {}
|
2021-03-02 22:02:17 +03:00
|
|
|
}
|
2021-02-28 01:18:25 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
return http.Request{
|
|
|
|
method: method
|
|
|
|
url: target.str()
|
2021-03-01 13:50:52 +03:00
|
|
|
headers: headers
|
|
|
|
lheaders: lheaders
|
2021-03-02 22:02:17 +03:00
|
|
|
data: body.bytestr()
|
2021-02-28 01:18:25 +03:00
|
|
|
version: version
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fn parse_request_line(s string) ?(http.Method, urllib.URL, http.Version) {
|
|
|
|
words := s.split(' ')
|
|
|
|
if words.len != 3 {
|
|
|
|
return error('malformed request line')
|
|
|
|
}
|
|
|
|
method := http.method_from_str(words[0])
|
|
|
|
target := urllib.parse(words[1]) ?
|
|
|
|
version := http.version_from_str(words[2])
|
|
|
|
if version == .unknown {
|
|
|
|
return error('unsupported version')
|
|
|
|
}
|
|
|
|
|
|
|
|
return method, target, version
|
|
|
|
}
|
|
|
|
|
2021-03-01 13:50:52 +03:00
|
|
|
fn parse_header(s string) ?(string, string) {
|
2021-02-28 01:18:25 +03:00
|
|
|
if ':' !in s {
|
|
|
|
return error('missing colon in header')
|
|
|
|
}
|
|
|
|
words := s.split_nth(':', 2)
|
|
|
|
// TODO: parse quoted text according to the RFC
|
2021-03-01 13:50:52 +03:00
|
|
|
return words[0], words[1].trim_left(' \t')
|
2021-02-28 01:18:25 +03:00
|
|
|
}
|
|
|
|
|
2021-03-04 19:00:03 +03:00
|
|
|
// Parse URL encoded key=value&key=value forms
|
|
|
|
fn parse_form(body string) map[string]string {
|
|
|
|
words := body.split('&')
|
|
|
|
mut form := map[string]string{}
|
|
|
|
for word in words {
|
|
|
|
kv := word.split_nth('=', 2)
|
|
|
|
if kv.len != 2 {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
key := urllib.query_unescape(kv[0]) or { continue }
|
|
|
|
val := urllib.query_unescape(kv[1]) or { continue }
|
|
|
|
form[key] = val
|
|
|
|
}
|
|
|
|
return form
|
|
|
|
// }
|
|
|
|
// todo: parse form-data and application/json
|
|
|
|
// ...
|
|
|
|
}
|
2021-02-28 01:18:25 +03:00
|
|
|
|
2021-03-04 19:00:03 +03:00
|
|
|
fn parse_multipart_form(body string, boundary string) (map[string]string, map[string][]FileData) {
|
|
|
|
sections := body.split(boundary)
|
|
|
|
fields := sections[1..sections.len - 1]
|
|
|
|
mut form := map[string]string{}
|
|
|
|
mut files := map[string][]FileData{}
|
|
|
|
|
|
|
|
for field in fields {
|
|
|
|
// TODO: do not split into lines; do same parsing for HTTP body
|
|
|
|
lines := field.split_into_lines()[1..]
|
|
|
|
disposition := parse_disposition(lines[0])
|
|
|
|
// Grab everything between the double quotes
|
|
|
|
name := disposition['name'] or { continue }
|
|
|
|
// Parse files
|
|
|
|
// TODO: filename*
|
|
|
|
if 'filename' in disposition {
|
|
|
|
filename := disposition['filename']
|
|
|
|
// Parse Content-Type header
|
|
|
|
if lines.len == 1 || !lines[1].to_lower().starts_with('content-type:') {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
mut ct := lines[1].split_nth(':', 2)[1]
|
|
|
|
ct = ct.trim_left(' \t')
|
|
|
|
data := lines_to_string(field.len, lines, 3, lines.len - 1)
|
|
|
|
files[name] << FileData{
|
|
|
|
filename: filename
|
|
|
|
content_type: ct
|
|
|
|
data: data
|
|
|
|
}
|
|
|
|
continue
|
2021-02-28 01:18:25 +03:00
|
|
|
}
|
2021-03-04 19:00:03 +03:00
|
|
|
data := lines_to_string(field.len, lines, 2, lines.len - 1)
|
|
|
|
form[name] = data
|
|
|
|
}
|
|
|
|
return form, files
|
|
|
|
}
|
|
|
|
|
|
|
|
// Parse the Content-Disposition header of a multipart form
|
|
|
|
// Returns a map of the key="value" pairs
|
|
|
|
// Example: parse_disposition('Content-Disposition: form-data; name="a"; filename="b"') == {'name': 'a', 'filename': 'b'}
|
|
|
|
fn parse_disposition(line string) map[string]string {
|
|
|
|
mut data := map[string]string{}
|
|
|
|
for word in line.split(';') {
|
|
|
|
kv := word.split_nth('=', 2)
|
|
|
|
if kv.len != 2 {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
key, value := kv[0].to_lower().trim_left(' \t'), kv[1]
|
|
|
|
if value.starts_with('"') && value.ends_with('"') {
|
|
|
|
data[key] = value[1..value.len - 1]
|
|
|
|
} else {
|
|
|
|
data[key] = value
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return data
|
|
|
|
}
|
|
|
|
|
|
|
|
[manualfree]
|
|
|
|
fn lines_to_string(len int, lines []string, start int, end int) string {
|
|
|
|
mut sb := strings.new_builder(len)
|
|
|
|
for i in start .. end {
|
|
|
|
sb.writeln(lines[i])
|
2021-02-28 01:18:25 +03:00
|
|
|
}
|
2021-03-04 19:00:03 +03:00
|
|
|
res := sb.str()
|
|
|
|
unsafe { sb.free() }
|
|
|
|
return res
|
2021-02-28 01:18:25 +03:00
|
|
|
}
|