summaryrefslogtreecommitdiffhomepage
path: root/libs/web/luasrc
diff options
context:
space:
mode:
Diffstat (limited to 'libs/web/luasrc')
-rw-r--r--libs/web/luasrc/http/protocol.lua572
1 files changed, 0 insertions, 572 deletions
diff --git a/libs/web/luasrc/http/protocol.lua b/libs/web/luasrc/http/protocol.lua
deleted file mode 100644
index 6901291b9a..0000000000
--- a/libs/web/luasrc/http/protocol.lua
+++ /dev/null
@@ -1,572 +0,0 @@
---[[
-
-HTTP protocol implementation for LuCI
-(c) 2008 Freifunk Leipzig / Jo-Philipp Wich <xm@leipzig.freifunk.net>
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
- http://www.apache.org/licenses/LICENSE-2.0
-
-$Id$
-
-]]--
-
-module("luci.http.protocol", package.seeall)
-
-require("luci.util")
-
-
-HTTP_MAX_CONTENT = 1024^2 -- 1 MB maximum content size
-HTTP_MAX_READBUF = 1024 -- 1 kB read buffer size
-
-HTTP_DEFAULT_CTYPE = "text/html" -- default content type
-HTTP_DEFAULT_VERSION = "1.0" -- HTTP default version
-
-
--- Decode an urlencoded string.
--- Returns the decoded value.
-function urldecode( str )
-
- local function __chrdec( hex )
- return string.char( tonumber( hex, 16 ) )
- end
-
- if type(str) == "string" then
- str = str:gsub( "+", " " ):gsub( "%%([a-fA-F0-9][a-fA-F0-9])", __chrdec )
- end
-
- return str
-end
-
-
--- Extract and split urlencoded data pairs, separated bei either "&" or ";" from given url.
--- Returns a table value with urldecoded values.
-function urldecode_params( url )
-
- local params = { }
-
- if url:find("?") then
- url = url:gsub( "^.+%?([^?]+)", "%1" )
- end
-
- for i, pair in ipairs(luci.util.split( url, "[&;]+", nil, true )) do
-
- -- find key and value
- local key = urldecode( pair:match("^([^=]+)") )
- local val = urldecode( pair:match("^[^=]+=(.+)$") )
-
- -- store
- if type(key) == "string" and key:len() > 0 then
- if type(val) ~= "string" then val = "" end
-
- if not params[key] then
- params[key] = val
- elseif type(params[key]) ~= "table" then
- params[key] = { params[key], val }
- else
- table.insert( params[key], val )
- end
- end
- end
-
- return params
-end
-
-
--- Encode given string in urlencoded format.
--- Returns the encoded string.
-function urlencode( str )
-
- local function __chrenc( chr )
- return string.format(
- "%%%02x", string.byte( chr )
- )
- end
-
- if type(str) == "string" then
- str = str:gsub(
- "([^a-zA-Z0-9$_%-%.+!*'(),])",
- __chrenc
- )
- end
-
- return str
-end
-
-
--- Encode given table to urlencoded string.
--- Returns the encoded string.
-function urlencode_params( tbl )
- local enc = ""
-
- for k, v in pairs(tbl) do
- enc = enc .. ( enc and "&" or "" ) ..
- urlencode(k) .. "=" ..
- urlencode(v)
- end
-
- return enc
-end
-
-
--- Decode MIME encoded data.
--- Returns a table with decoded values.
-function mimedecode( data, boundary, filecb )
-
- local params = { }
-
- -- create a line reader
- local reader = _linereader( data, HTTP_MAX_READBUF )
-
- -- state variables
- local in_part = false
- local in_file = false
- local in_fbeg = false
- local in_size = true
-
- local filename
- local buffer
- local field
- local clen = 0
-
- -- try to read all mime parts
- for line, eol in reader do
-
- -- update content length
- clen = clen + line:len()
-
- if clen >= HTTP_MAX_CONTENT then
- in_size = false
- end
-
- -- when no boundary is given, try to find it
- if not boundary then
- boundary = line:match("^%-%-([^\r\n]+)\r?\n$")
- end
-
- -- Got a valid boundary line or reached max allowed size.
- if ( boundary and line:sub(1,2) == "--" and line:len() > #boundary + 2 and
- line:sub( 3, 2 + #boundary ) == boundary ) or not in_size
- then
- -- Flush the data of the previous mime part.
- -- When field and/or buffer are set to nil we should discard
- -- the previous section entirely due to format violations.
- if type(field) == "string" and field:len() > 0 and
- type(buffer) == "string"
- then
- -- According to the rfc the \r\n preceeding a boundary
- -- is assumed to be part of the boundary itself.
- -- Since we are reading line by line here, this crlf
- -- is part of the last line of our section content,
- -- so strip it before storing the buffer.
- buffer = buffer:gsub("\r?\n$","")
-
- -- If we're in a file part and a file callback has been provided
- -- then do a final call and send eof.
- if in_file and type(filecb) == "function" then
- filecb( field, filename, buffer, true )
- params[field] = filename
-
- -- Store buffer.
- else
- params[field] = buffer
- end
- end
-
- -- Reset vars
- buffer = ""
- filename = nil
- field = nil
- in_file = false
-
- -- Abort here if we reached maximum allowed size
- if not in_size then break end
-
- -- Do we got the last boundary?
- if line:len() > #boundary + 4 and
- line:sub( #boundary + 2, #boundary + 4 ) == "--"
- then
- -- No more processing
- in_part = false
-
- -- It's a middle boundary
- else
-
- -- Read headers
- local hlen, headers = extract_headers( reader )
-
- -- Check for valid headers
- if headers['Content-Disposition'] then
-
- -- Got no content type header, assume content-type "text/plain"
- if not headers['Content-Type'] then
- headers['Content-Type'] = 'text/plain'
- end
-
- -- Find field name
- local hdrvals = luci.util.split(
- headers['Content-Disposition'], '; '
- )
-
- -- Valid form data part?
- if hdrvals[1] == "form-data" and hdrvals[2]:match("^name=") then
-
- -- Store field identifier
- field = hdrvals[2]:match('^name="(.+)"$')
-
- -- Do we got a file upload field?
- if #hdrvals == 3 and hdrvals[3]:match("^filename=") then
- in_file = true
- if_fbeg = true
- filename = hdrvals[3]:match('^filename="(.+)"$')
- end
-
- -- Entering next part processing
- in_part = true
- end
- end
- end
-
- -- Processing content
- elseif in_part then
-
- -- XXX: Would be really good to switch from line based to
- -- buffered reading here.
-
-
- -- If we're in a file part and a file callback has been provided
- -- then call the callback and reset the buffer.
- if in_file and type(filecb) == "function" then
-
- -- If we're not processing the first chunk, then call
- if not in_fbeg then
- filecb( field, filename, buffer, false )
- buffer = ""
-
- -- Clear in_fbeg flag after first run
- else
- in_fbeg = false
- end
- end
-
- -- Append date to buffer
- buffer = buffer .. line
- end
- end
-
- return params
-end
-
-
--- Extract "magic", the first line of a http message.
--- Returns the message type ("get", "post" or "response"), the requested uri
--- if it is a valid http request or the status code if the line descripes a
--- http response. For requests the third parameter is nil, for responses it
--- contains the human readable status description.
-function extract_magic( reader )
-
- for line in reader do
- -- Is it a request?
- local method, uri = line:match("^([A-Z]+) ([^ ]+) HTTP/[01]%.[019]\r?\n$")
-
- -- Yup, it is
- if method then
- return method:lower(), uri, nil
-
- -- Is it a response?
- else
- local code, message = line:match("^HTTP/[01]%.[019] ([0-9]+) ([^\r\n]+)\r?\n$")
-
- -- Is a response
- if code then
- return "response", code + 0, message
-
- -- Can't handle it
- else
- return nil
- end
- end
- end
-end
-
-
--- Extract headers from given string.
--- Returns a table of extracted headers and the remainder of the parsed data.
-function extract_headers( reader, tbl )
-
- local headers = tbl or { }
- local count = 0
-
- -- Iterate line by line
- for line in reader do
-
- -- Look for a valid header format
- local hdr, val = line:match( "^([A-Z][A-Za-z0-9%-_]+): +([^\r\n]+)\r?\n$" )
-
- if type(hdr) == "string" and hdr:len() > 0 and
- type(val) == "string" and val:len() > 0
- then
- count = count + line:len()
- headers[hdr] = val
-
- elseif line:match("^\r?\n$") then
-
- return count + line:len(), headers
-
- else
- -- junk data, don't add length
- return count, headers
- end
- end
-
- return count, headers
-end
-
-
--- Parse a http message
-function parse_message( data, filecb )
-
- local reader = _linereader( data, HTTP_MAX_READBUF )
- local message = parse_message_header( reader )
-
- if message then
- parse_message_body( reader, message, filecb )
- end
-
- return message
-end
-
-
--- Parse a http message header
-function parse_message_header( data )
-
- -- Create a line reader
- local reader = _linereader( data, HTTP_MAX_READBUF )
- local message = { }
-
- -- Try to extract magic
- local method, arg1, arg2 = extract_magic( reader )
-
- -- Does it looks like a valid message?
- if method then
-
- message.request_method = method
- message.status_code = arg2 and arg1 or 200
- message.status_message = arg2 or nil
- message.request_uri = arg2 and nil or arg1
-
- if method == "response" then
- message.type = "response"
- else
- message.type = "request"
- end
-
- -- Parse headers?
- local hlen, hdrs = extract_headers( reader )
-
- -- Valid headers?
- if hlen > 2 and type(hdrs) == "table" then
-
- message.headers = hdrs
-
- -- Process get parameters
- if ( method == "get" or method == "post" ) and
- message.request_uri:match("?")
- then
- message.params = urldecode_params( message.request_uri )
- else
- message.params = { }
- end
-
- -- Populate common environment variables
- message.env = {
- CONTENT_LENGTH = hdrs['Content-Length'];
- CONTENT_TYPE = hdrs['Content-Type'];
- REQUEST_METHOD = message.request_method;
- REQUEST_URI = message.request_uri;
- SCRIPT_NAME = message.request_uri:gsub("?.+$","");
- SCRIPT_FILENAME = "" -- XXX implement me
- }
-
- -- Populate HTTP_* environment variables
- for i, hdr in ipairs( {
- 'Accept',
- 'Accept-Charset',
- 'Accept-Encoding',
- 'Accept-Language',
- 'Connection',
- 'Cookie',
- 'Host',
- 'Referer',
- 'User-Agent',
- } ) do
- local var = 'HTTP_' .. hdr:upper():gsub("%-","_")
- local val = hdrs[hdr]
-
- message.env[var] = val
- end
-
-
- return message
- end
- end
-end
-
-
--- Parse a http message body
-function parse_message_body( reader, message, filecb )
-
- if type(message) == "table" then
- local env = message.env
-
- local clen = ( env.CONTENT_LENGTH or HTTP_MAX_CONTENT ) + 0
-
- -- Process post method
- if env.REQUEST_METHOD:lower() == "post" and env.CONTENT_TYPE then
-
- -- Is it multipart/form-data ?
- if env.CONTENT_TYPE:match("^multipart/form%-data") then
-
- -- Read multipart/mime data
- for k, v in pairs( mimedecode(
- reader,
- env.CONTENT_TYPE:match("boundary=(.+)"),
- filecb
- ) ) do
- message.params[k] = v
- end
-
- -- Is it x-www-form-urlencoded?
- elseif env.CONTENT_TYPE:match('^application/x%-www%-form%-urlencoded') then
-
- -- Read post data
- local post_data = ""
-
- for chunk, eol in reader do
-
- post_data = post_data .. chunk
-
- -- Abort on eol or if maximum allowed size or content length is reached
- if eol or #post_data >= HTTP_MAX_CONTENT or #post_data > clen then
- break
- end
- end
-
- -- Parse params
- for k, v in pairs( urldecode_params( post_data ) ) do
- message.params[k] = v
- end
-
- -- Unhandled encoding
- -- If a file callback is given then feed it line by line, else
- -- store whole buffer in message.content
- else
-
- local len = 0
-
- for chunk in reader do
-
- len = len + #chunk
-
- -- We have a callback, feed it.
- if type(filecb) == "function" then
-
- filecb( "_post", nil, chunk, false )
-
- -- Append to .content buffer.
- else
- message.content =
- type(message.content) == "string"
- and message.content .. chunk
- or chunk
- end
-
- -- Abort if maximum allowed size or content length is reached
- if len >= HTTP_MAX_CONTENT or len >= clen then
- break
- end
- end
-
- -- Send eof to callback
- if type(filecb) == "function" then
- filecb( "_post", nil, "", true )
- end
- end
- end
- end
-end
-
-
--- Wrap given object into a line read iterator
-function _linereader( obj, bufsz )
-
- bufsz = ( bufsz and bufsz >= 256 ) and bufsz or 256
-
- local __read = function() return nil end
- local __eof = function(x) return type(x) ~= "string" or #x == 0 end
-
- local _pos = 1
- local _buf = ""
- local _eof = nil
-
- -- object is string
- if type(obj) == "string" then
-
- __read = function() return obj:sub( _pos, _pos + bufsz - #_buf - 1 ) end
-
- -- object implements a receive() or read() function
- elseif (type(obj) == "userdata" or type(obj) == "table") and ( type(obj.receive) == "function" or type(obj.read) == "function" ) then
-
- if type(obj.read) == "function" then
- __read = function() return obj:read( bufsz - #_buf ) end
- else
- __read = function() return obj:receive( bufsz - #_buf ) end
- end
-
- -- object is a function
- elseif type(obj) == "function" then
-
- return obj
-
- -- no usable data type
- else
-
- -- dummy iterator
- return __read
- end
-
-
- -- generic block to line algorithm
- return function()
- if not _eof then
- local buffer = __read()
-
- if __eof( buffer ) then
- buffer = ""
- end
-
- _pos = _pos + #buffer
- buffer = _buf .. buffer
-
- local crlf, endpos = buffer:find("\r?\n")
-
-
- if crlf then
- _buf = buffer:sub( endpos + 1, #buffer )
- return buffer:sub( 1, endpos ), true
- else
- -- check for eof
- _eof = __eof( buffer )
-
- -- clear overflow buffer
- _buf = ""
-
- return buffer, false
- end
- else
- return nil
- end
- end
-end