unicorn-rupcio 6.1.0

data/ext/unicorn_http/unicorn_http.rl

@@ -0,0 +1,1054 @@
+/**
+ * Copyright (c) 2009 Eric Wong (all bugs are Eric's fault)
+ * Copyright (c) 2005 Zed A. Shaw
+ * You can redistribute it and/or modify it under the same terms as Ruby 1.8 or
+ * the GPLv2+ (GPLv3+ preferred)
+ */
+#include "ruby.h"
+#include "ext_help.h"
+#include <assert.h>
+#include <string.h>
+#include <sys/types.h>
+#include "common_field_optimization.h"
+#include "global_variables.h"
+#include "c_util.h"
+#include "epollexclusive.h"
+
+void init_unicorn_httpdate(void);
+
+#define UH_FL_CHUNKED  0x1
+#define UH_FL_HASBODY  0x2
+#define UH_FL_INBODY   0x4
+#define UH_FL_HASTRAILER 0x8
+#define UH_FL_INTRAILER 0x10
+#define UH_FL_INCHUNK  0x20
+#define UH_FL_REQEOF 0x40
+#define UH_FL_KAVERSION 0x80
+#define UH_FL_HASHEADER 0x100
+#define UH_FL_TO_CLEAR 0x200
+#define UH_FL_RESSTART 0x400 /* for check_client_connection */
+#define UH_FL_HIJACK 0x800
+#define UH_FL_RES_CHUNK_VER (1U << 12)
+#define UH_FL_RES_CHUNK_METHOD (1U << 13)
+
+/* all of these flags need to be set for keepalive to be supported */
+#define UH_FL_KEEPALIVE (UH_FL_KAVERSION | UH_FL_REQEOF | UH_FL_HASHEADER)
+
+/* we can only chunk responses for non-HEAD HTTP/1.1 requests */
+#define UH_FL_RES_CHUNKABLE (UH_FL_RES_CHUNK_VER | UH_FL_RES_CHUNK_METHOD)
+
+static unsigned int MAX_HEADER_LEN = 1024 * (80 + 32); /* same as Mongrel */
+
+/* this is only intended for use with Rainbows! */
+static VALUE set_maxhdrlen(VALUE self, VALUE len)
+{
+  return UINT2NUM(MAX_HEADER_LEN = NUM2UINT(len));
+}
+
+/* keep this small for other servers (e.g. yahns) since every client has one */
+struct http_parser {
+  int cs; /* Ragel internal state */
+  unsigned int flags;
+  unsigned int mark;
+  unsigned int offset;
+  union { /* these 2 fields don't nest */
+    unsigned int field;
+    unsigned int query;
+  } start;
+  union {
+    unsigned int field_len; /* only used during header processing */
+    unsigned int dest_offset; /* only used during body processing */
+  } s;
+  VALUE buf;
+  VALUE env;
+  VALUE cont; /* Qfalse: unset, Qnil: ignored header, T_STRING: append */
+  union {
+    off_t content;
+    off_t chunk;
+  } len;
+};
+
+static ID id_set_backtrace, id_is_chunked_p;
+static VALUE cHttpParser;
+
+static void finalize_header(struct http_parser *hp);
+
+static void parser_raise(VALUE klass, const char *msg)
+{
+  VALUE exc = rb_exc_new2(klass, msg);
+  VALUE bt = rb_ary_new();
+
+  rb_funcall(exc, id_set_backtrace, 1, bt);
+  rb_exc_raise(exc);
+}
+
+static inline unsigned int ulong2uint(unsigned long n)
+{
+  unsigned int i = (unsigned int)n;
+
+  if (sizeof(unsigned int) != sizeof(unsigned long)) {
+    if ((unsigned long)i != n) {
+      rb_raise(rb_eRangeError, "too large to be 32-bit uint: %lu", n);
+    }
+  }
+  return i;
+}
+
+#define REMAINING (unsigned long)(pe - p)
+#define LEN(AT, FPC) (ulong2uint(FPC - buffer) - hp->AT)
+#define MARK(M,FPC) (hp->M = ulong2uint((FPC) - buffer))
+#define PTR_TO(F) (buffer + hp->F)
+#define STR_NEW(M,FPC) rb_str_new(PTR_TO(M), LEN(M, FPC))
+#define STRIPPED_STR_NEW(M,FPC) stripped_str_new(PTR_TO(M), LEN(M, FPC))
+
+#define HP_FL_TEST(hp,fl) ((hp)->flags & (UH_FL_##fl))
+#define HP_FL_SET(hp,fl) ((hp)->flags |= (UH_FL_##fl))
+#define HP_FL_UNSET(hp,fl) ((hp)->flags &= ~(UH_FL_##fl))
+#define HP_FL_ALL(hp,fl) (HP_FL_TEST(hp, fl) == (UH_FL_##fl))
+
+static int is_lws(char c)
+{
+  return (c == ' ' || c == '\t');
+}
+
+static VALUE stripped_str_new(const char *str, long len)
+{
+  long end;
+
+  for (end = len - 1; end >= 0 && is_lws(str[end]); end--);
+
+  return rb_str_new(str, end + 1);
+}
+
+/*
+ * handles values of the "Connection:" header, keepalive is implied
+ * for HTTP/1.1 but needs to be explicitly enabled with HTTP/1.0
+ * Additionally, we require GET/HEAD requests to support keepalive.
+ */
+static void hp_keepalive_connection(struct http_parser *hp, VALUE val)
+{
+  if (STR_CSTR_CASE_EQ(val, "keep-alive")) {
+    /* basically have HTTP/1.0 masquerade as HTTP/1.1+ */
+    HP_FL_SET(hp, KAVERSION);
+  } else if (STR_CSTR_CASE_EQ(val, "close")) {
+    /*
+     * it doesn't matter what HTTP version or request method we have,
+     * if a client says "Connection: close", we disable keepalive
+     */
+    HP_FL_UNSET(hp, KAVERSION);
+  } else {
+    /*
+     * client could've sent anything, ignore it for now.  Maybe
+     * "HP_FL_UNSET(hp, KAVERSION);" just in case?
+     * Raising an exception might be too mean...
+     */
+  }
+}
+
+static void
+request_method(struct http_parser *hp, const char *ptr, size_t len)
+{
+  VALUE v = rb_str_new(ptr, len);
+
+  if (len != 4 || memcmp(ptr, "HEAD", 4))
+    HP_FL_SET(hp, RES_CHUNK_METHOD);
+
+  rb_hash_aset(hp->env, g_request_method, v);
+}
+
+static void
+http_version(struct http_parser *hp, const char *ptr, size_t len)
+{
+  VALUE v;
+
+  HP_FL_SET(hp, HASHEADER);
+
+  if (CONST_MEM_EQ("HTTP/1.1", ptr, len)) {
+    /* HTTP/1.1 implies keepalive unless "Connection: close" is set */
+    HP_FL_SET(hp, KAVERSION);
+    HP_FL_SET(hp, RES_CHUNK_VER);
+    v = g_http_11;
+  } else if (CONST_MEM_EQ("HTTP/1.0", ptr, len)) {
+    v = g_http_10;
+  } else {
+    v = rb_str_new(ptr, len);
+  }
+  rb_hash_aset(hp->env, g_server_protocol, v);
+  rb_hash_aset(hp->env, g_http_version, v);
+}
+
+static inline void hp_invalid_if_trailer(struct http_parser *hp)
+{
+  if (HP_FL_TEST(hp, INTRAILER))
+    parser_raise(eHttpParserError, "invalid Trailer");
+}
+
+static void write_cont_value(struct http_parser *hp,
+                             char *buffer, const char *p)
+{
+  char *vptr;
+  long end;
+  long len = LEN(mark, p);
+  long cont_len;
+
+  if (hp->cont == Qfalse)
+     parser_raise(eHttpParserError, "invalid continuation line");
+  if (NIL_P(hp->cont))
+     return; /* we're ignoring this header (probably Host:) */
+
+  assert(TYPE(hp->cont) == T_STRING && "continuation line is not a string");
+  assert(hp->mark > 0 && "impossible continuation line offset");
+
+  if (len == 0)
+    return;
+
+  cont_len = RSTRING_LEN(hp->cont);
+  if (cont_len > 0) {
+    --hp->mark;
+    len = LEN(mark, p);
+  }
+  vptr = PTR_TO(mark);
+
+  /* normalize tab to space */
+  if (cont_len > 0) {
+    assert((' ' == *vptr || '\t' == *vptr) && "invalid leading white space");
+    *vptr = ' ';
+  }
+
+  for (end = len - 1; end >= 0 && is_lws(vptr[end]); end--);
+  rb_str_buf_cat(hp->cont, vptr, end + 1);
+}
+
+static int is_chunked(VALUE v)
+{
+  /* common case first */
+  if (STR_CSTR_CASE_EQ(v, "chunked"))
+    return 1;
+
+  /*
+   * call Ruby function in unicorn/http_request.rb to deal with unlikely
+   * comma-delimited case
+   */
+  return rb_funcall(cHttpParser, id_is_chunked_p, 1, v) != Qfalse;
+}
+
+static void write_value(struct http_parser *hp, char *buffer, const char *p)
+{
+  VALUE f, v, e;
+  char *tip = PTR_TO(start.field);
+  size_t i;
+  static const size_t cl_len = sizeof("CONTENT_LENGTH") - 1;
+  static const size_t tl_len = sizeof("TRANSFER_ENCODING") - 1;
+
+  /* avoid confusion + smuggling attacks from `_' */
+  if (hp->s.field_len == cl_len && !memcmp(tip, "CONTENT_LENGTH", cl_len))
+        parser_raise(eHttpParserError, "invalid Content-Length");
+  else if (hp->s.field_len == tl_len &&
+           !memcmp(tip, "TRANSFER_ENCODING", tl_len))
+        parser_raise(eHttpParserError, "invalid Transfer-Encoding");
+
+  /* .tr('-', '_') since Rack expects it */
+  for (i = 0; i < hp->s.field_len; i++)
+	if (tip[i] == '-')
+		tip[i] = '_';
+
+  f = find_common_field(tip, hp->s.field_len);
+  VALIDATE_MAX_LENGTH(LEN(mark, p), FIELD_VALUE);
+  v = LEN(mark, p) == 0 ? rb_str_buf_new(128) : STRIPPED_STR_NEW(mark, p);
+  if (NIL_P(f)) {
+    const char *field = PTR_TO(start.field);
+    size_t flen = hp->s.field_len;
+
+    VALIDATE_MAX_LENGTH(flen, FIELD_NAME);
+
+    /*
+     * ignore "Version" headers since they conflict with the HTTP_VERSION
+     * rack env variable.
+     */
+    if (CONST_MEM_EQ("VERSION", field, flen)) {
+      hp->cont = Qnil;
+      return;
+    }
+    f = uncommon_field(field, flen);
+  } else if (f == g_http_connection) {
+    hp_keepalive_connection(hp, v);
+  } else if (f == g_content_length && !HP_FL_TEST(hp, CHUNKED)) {
+    if (hp->len.content)
+      parser_raise(eHttpParserError, "Content-Length already set");
+    hp->len.content = parse_length(RSTRING_PTR(v), RSTRING_LEN(v));
+    if (hp->len.content < 0)
+      parser_raise(eHttpParserError, "invalid Content-Length");
+    if (hp->len.content != 0)
+      HP_FL_SET(hp, HASBODY);
+    hp_invalid_if_trailer(hp);
+  } else if (f == g_http_transfer_encoding) {
+    if (is_chunked(v)) {
+      if (HP_FL_TEST(hp, CHUNKED))
+        /*
+         * RFC 7230 3.3.1:
+         * A sender MUST NOT apply chunked more than once to a message body
+         * (i.e., chunking an already chunked message is not allowed).
+         */
+        parser_raise(eHttpParserError, "Transfer-Encoding double chunked");
+
+      HP_FL_SET(hp, CHUNKED);
+      HP_FL_SET(hp, HASBODY);
+
+      /* RFC 7230 3.3.3, 3: favor chunked if Content-Length exists */
+      hp->len.content = 0;
+    } else if (HP_FL_TEST(hp, CHUNKED)) {
+      /*
+       * RFC 7230 3.3.3, point 3 states:
+       * If a Transfer-Encoding header field is present in a request and
+       * the chunked transfer coding is not the final encoding, the
+       * message body length cannot be determined reliably; the server
+       * MUST respond with the 400 (Bad Request) status code and then
+       * close the connection.
+       */
+      parser_raise(eHttpParserError, "invalid Transfer-Encoding");
+    }
+    hp_invalid_if_trailer(hp);
+  } else if (f == g_http_trailer) {
+    HP_FL_SET(hp, HASTRAILER);
+    hp_invalid_if_trailer(hp);
+  } else {
+    assert(TYPE(f) == T_STRING && "memoized object is not a string");
+    assert_frozen(f);
+  }
+
+  e = rb_hash_aref(hp->env, f);
+  if (NIL_P(e)) {
+    hp->cont = rb_hash_aset(hp->env, f, v);
+  } else if (f == g_http_host) {
+    /*
+     * ignored, absolute URLs in REQUEST_URI take precedence over
+     * the Host: header (ref: rfc 2616, section 5.2.1)
+     */
+     hp->cont = Qnil;
+  } else {
+    rb_str_buf_cat(e, ",", 1);
+    hp->cont = rb_str_buf_append(e, v);
+  }
+}
+
+/** Machine **/
+
+%%{
+  machine http_parser;
+
+  action mark {MARK(mark, fpc); }
+
+  action start_field { MARK(start.field, fpc); }
+  action upcase_field { upcase_char(deconst(fpc)); }
+  action downcase_char { downcase_char(deconst(fpc)); }
+  action write_field { hp->s.field_len = LEN(start.field, fpc); }
+  action start_value { MARK(mark, fpc); }
+  action write_value { write_value(hp, buffer, fpc); }
+  action write_cont_value { write_cont_value(hp, buffer, fpc); }
+  action request_method { request_method(hp, PTR_TO(mark), LEN(mark, fpc)); }
+  action scheme {
+    rb_hash_aset(hp->env, g_rack_url_scheme, STR_NEW(mark, fpc));
+  }
+  action host { rb_hash_aset(hp->env, g_http_host, STR_NEW(mark, fpc)); }
+  action request_uri {
+    VALUE str;
+
+    VALIDATE_MAX_URI_LENGTH(LEN(mark, fpc), REQUEST_URI);
+    str = rb_hash_aset(hp->env, g_request_uri, STR_NEW(mark, fpc));
+    /*
+     * "OPTIONS * HTTP/1.1\r\n" is a valid request, but we can't have '*'
+     * in REQUEST_PATH or PATH_INFO or else Rack::Lint will complain
+     */
+    if (STR_CSTR_EQ(str, "*")) {
+      str = rb_str_new(NULL, 0);
+      rb_hash_aset(hp->env, g_path_info, str);
+      rb_hash_aset(hp->env, g_request_path, str);
+    }
+  }
+  action fragment {
+    VALIDATE_MAX_URI_LENGTH(LEN(mark, fpc), FRAGMENT);
+    rb_hash_aset(hp->env, g_fragment, STR_NEW(mark, fpc));
+  }
+  action start_query {MARK(start.query, fpc); }
+  action query_string {
+    VALIDATE_MAX_URI_LENGTH(LEN(start.query, fpc), QUERY_STRING);
+    rb_hash_aset(hp->env, g_query_string, STR_NEW(start.query, fpc));
+  }
+  action http_version { http_version(hp, PTR_TO(mark), LEN(mark, fpc)); }
+  action request_path {
+    VALUE val;
+
+    VALIDATE_MAX_URI_LENGTH(LEN(mark, fpc), REQUEST_PATH);
+    val = rb_hash_aset(hp->env, g_request_path, STR_NEW(mark, fpc));
+
+    /* rack says PATH_INFO must start with "/" or be empty */
+    if (!STR_CSTR_EQ(val, "*"))
+      rb_hash_aset(hp->env, g_path_info, val);
+  }
+  action add_to_chunk_size {
+    hp->len.chunk = step_incr(hp->len.chunk, fc, 16);
+    if (hp->len.chunk < 0)
+      parser_raise(eHttpParserError, "invalid chunk size");
+  }
+  action header_done {
+    finalize_header(hp);
+
+    cs = http_parser_first_final;
+    if (HP_FL_TEST(hp, HASBODY)) {
+      HP_FL_SET(hp, INBODY);
+      if (HP_FL_TEST(hp, CHUNKED))
+        cs = http_parser_en_ChunkedBody;
+    } else {
+      HP_FL_SET(hp, REQEOF);
+      assert(!HP_FL_TEST(hp, CHUNKED) && "chunked encoding without body!");
+    }
+    /*
+     * go back to Ruby so we can call the Rack application, we'll reenter
+     * the parser iff the body needs to be processed.
+     */
+    goto post_exec;
+  }
+
+  action end_trailers {
+    cs = http_parser_first_final;
+    goto post_exec;
+  }
+
+  action end_chunked_body {
+    HP_FL_SET(hp, INTRAILER);
+    cs = http_parser_en_Trailers;
+    ++p;
+    assert(p <= pe && "buffer overflow after chunked body");
+    goto post_exec;
+  }
+
+  action skip_chunk_data {
+  skip_chunk_data_hack: {
+    size_t nr = MIN((size_t)hp->len.chunk, REMAINING);
+    memcpy(RSTRING_PTR(hp->cont) + hp->s.dest_offset, fpc, nr);
+    hp->s.dest_offset += nr;
+    hp->len.chunk -= nr;
+    p += nr;
+    assert(hp->len.chunk >= 0 && "negative chunk length");
+    if ((size_t)hp->len.chunk > REMAINING) {
+      HP_FL_SET(hp, INCHUNK);
+      goto post_exec;
+    } else {
+      fhold;
+      fgoto chunk_end;
+    }
+  }}
+
+  include unicorn_http_common "unicorn_http_common.rl";
+}%%
+
+/** Data **/
+%% write data;
+
+static void http_parser_init(struct http_parser *hp)
+{
+  int cs = 0;
+  hp->flags = 0;
+  hp->mark = 0;
+  hp->offset = 0;
+  hp->start.field = 0;
+  hp->s.field_len = 0;
+  hp->len.content = 0;
+  hp->cont = Qfalse; /* zero on MRI, should be optimized away by above */
+  %% write init;
+  hp->cs = cs;
+}
+
+/** exec **/
+static void
+http_parser_execute(struct http_parser *hp, char *buffer, size_t len)
+{
+  const char *p, *pe;
+  int cs = hp->cs;
+  size_t off = hp->offset;
+
+  if (cs == http_parser_first_final)
+    return;
+
+  assert(off <= len && "offset past end of buffer");
+
+  p = buffer+off;
+  pe = buffer+len;
+
+  assert((void *)(pe - p) == (void *)(len - off) &&
+         "pointers aren't same distance");
+
+  if (HP_FL_TEST(hp, INCHUNK)) {
+    HP_FL_UNSET(hp, INCHUNK);
+    goto skip_chunk_data_hack;
+  }
+  %% write exec;
+post_exec: /* "_out:" also goes here */
+  if (hp->cs != http_parser_error)
+    hp->cs = cs;
+  hp->offset = ulong2uint(p - buffer);
+
+  assert(p <= pe && "buffer overflow after parsing execute");
+  assert(hp->offset <= len && "offset longer than length");
+}
+
+static void hp_mark(void *ptr)
+{
+  struct http_parser *hp = ptr;
+
+  rb_gc_mark(hp->buf);
+  rb_gc_mark(hp->env);
+  rb_gc_mark(hp->cont);
+}
+
+static size_t hp_memsize(const void *ptr)
+{
+  return sizeof(struct http_parser);
+}
+
+static const rb_data_type_t hp_type = {
+  "unicorn_http",
+  { hp_mark, RUBY_TYPED_DEFAULT_FREE, hp_memsize, /* reserved */ },
+  /* parent, data, [ flags ] */
+};
+
+static struct http_parser *data_get(VALUE self)
+{
+  struct http_parser *hp;
+
+  TypedData_Get_Struct(self, struct http_parser, &hp_type, hp);
+  assert(hp && "failed to extract http_parser struct");
+  return hp;
+}
+
+/*
+ * set rack.url_scheme to "https" or "http", no others are allowed by Rack
+ * this resembles the Rack::Request#scheme method as of rack commit
+ * 35bb5ba6746b5d346de9202c004cc926039650c7
+ */
+static void set_url_scheme(VALUE env, VALUE *server_port)
+{
+  VALUE scheme = rb_hash_aref(env, g_rack_url_scheme);
+
+  if (NIL_P(scheme)) {
+    /*
+     * would anybody be horribly opposed to removing the X-Forwarded-SSL
+     * and X-Forwarded-Proto handling from this parser?  We've had it
+     * forever and nobody has said anything against it, either.
+     * Anyways, please send comments to our public mailing list:
+     * unicorn-public@yhbt.net (no HTML mail, no subscription necessary)
+     */
+    scheme = rb_hash_aref(env, g_http_x_forwarded_ssl);
+    if (!NIL_P(scheme) && STR_CSTR_EQ(scheme, "on")) {
+      *server_port = g_port_443;
+      scheme = g_https;
+    } else {
+      scheme = rb_hash_aref(env, g_http_x_forwarded_proto);
+      if (NIL_P(scheme)) {
+        scheme = g_http;
+      } else {
+        long len = RSTRING_LEN(scheme);
+        if (len >= 5 && !memcmp(RSTRING_PTR(scheme), "https", 5)) {
+          if (len != 5)
+            scheme = g_https;
+          *server_port = g_port_443;
+        } else {
+          scheme = g_http;
+        }
+      }
+    }
+    rb_hash_aset(env, g_rack_url_scheme, scheme);
+  } else if (STR_CSTR_EQ(scheme, "https")) {
+    *server_port = g_port_443;
+  } else {
+    assert(*server_port == g_port_80 && "server_port not set");
+  }
+}
+
+/*
+ * Parse and set the SERVER_NAME and SERVER_PORT variables
+ * Not supporting X-Forwarded-Host/X-Forwarded-Port in here since
+ * anybody who needs them is using an unsupported configuration and/or
+ * incompetent.  Rack::Request will handle X-Forwarded-{Port,Host} just
+ * fine.
+ */
+static void set_server_vars(VALUE env, VALUE *server_port)
+{
+  VALUE server_name = g_localhost;
+  VALUE host = rb_hash_aref(env, g_http_host);
+
+  if (!NIL_P(host)) {
+    char *host_ptr = RSTRING_PTR(host);
+    long host_len = RSTRING_LEN(host);
+    char *colon;
+
+    if (*host_ptr == '[') { /* ipv6 address format */
+      char *rbracket = memchr(host_ptr + 1, ']', host_len - 1);
+
+      if (rbracket)
+        colon = (rbracket[1] == ':') ? rbracket + 1 : NULL;
+      else
+        colon = memchr(host_ptr + 1, ':', host_len - 1);
+    } else {
+      colon = memchr(host_ptr, ':', host_len);
+    }
+
+    if (colon) {
+      long port_start = colon - host_ptr + 1;
+
+      server_name = rb_str_substr(host, 0, colon - host_ptr);
+      if ((host_len - port_start) > 0)
+        *server_port = rb_str_substr(host, port_start, host_len);
+    } else {
+      server_name = host;
+    }
+  }
+  rb_hash_aset(env, g_server_name, server_name);
+  rb_hash_aset(env, g_server_port, *server_port);
+}
+
+static void finalize_header(struct http_parser *hp)
+{
+  VALUE server_port = g_port_80;
+
+  set_url_scheme(hp->env, &server_port);
+  set_server_vars(hp->env, &server_port);
+
+  if (!HP_FL_TEST(hp, HASHEADER))
+    rb_hash_aset(hp->env, g_server_protocol, g_http_09);
+
+  /* rack requires QUERY_STRING */
+  if (NIL_P(rb_hash_aref(hp->env, g_query_string)))
+    rb_hash_aset(hp->env, g_query_string, rb_str_new(NULL, 0));
+}
+
+static VALUE HttpParser_alloc(VALUE klass)
+{
+  struct http_parser *hp;
+
+  return TypedData_Make_Struct(klass, struct http_parser, &hp_type, hp);
+}
+
+/**
+ * call-seq:
+ *    parser.new => parser
+ *
+ * Creates a new parser.
+ */
+static VALUE HttpParser_init(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+
+  http_parser_init(hp);
+  hp->buf = rb_str_new(NULL, 0);
+  hp->env = rb_hash_new();
+
+  return self;
+}
+
+/**
+ * call-seq:
+ *    parser.clear => parser
+ *
+ * Resets the parser to it's initial state so that you can reuse it
+ * rather than making new ones.
+ */
+static VALUE HttpParser_clear(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+
+  /* we can't safely reuse .buf and .env if hijacked */
+  if (HP_FL_TEST(hp, HIJACK))
+    return HttpParser_init(self);
+
+  http_parser_init(hp);
+  rb_hash_clear(hp->env);
+
+  return self;
+}
+
+static void advance_str(VALUE str, off_t nr)
+{
+  long len = RSTRING_LEN(str);
+
+  if (len == 0)
+    return;
+
+  rb_str_modify(str);
+
+  assert(nr <= len && "trying to advance past end of buffer");
+  len -= nr;
+  if (len > 0) /* unlikely, len is usually 0 */
+    memmove(RSTRING_PTR(str), RSTRING_PTR(str) + nr, len);
+  rb_str_set_len(str, len);
+}
+
+/**
+ * call-seq:
+ *   parser.content_length => nil or Integer
+ *
+ * Returns the number of bytes left to run through HttpParser#filter_body.
+ * This will initially be the value of the "Content-Length" HTTP header
+ * after header parsing is complete and will decrease in value as
+ * HttpParser#filter_body is called for each chunk.  This should return
+ * zero for requests with no body.
+ *
+ * This will return nil on "Transfer-Encoding: chunked" requests.
+ */
+static VALUE HttpParser_content_length(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+
+  return HP_FL_TEST(hp, CHUNKED) ? Qnil : OFFT2NUM(hp->len.content);
+}
+
+/**
+ * Document-method: parse
+ * call-seq:
+ *    parser.parse => env or nil
+ *
+ * Takes a Hash and a String of data, parses the String of data filling
+ * in the Hash returning the Hash if parsing is finished, nil otherwise
+ * When returning the env Hash, it may modify data to point to where
+ * body processing should begin.
+ *
+ * Raises HttpParserError if there are parsing errors.
+ */
+static VALUE HttpParser_parse(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+  VALUE data = hp->buf;
+
+  if (HP_FL_TEST(hp, TO_CLEAR))
+    HttpParser_clear(self);
+
+  http_parser_execute(hp, RSTRING_PTR(data), RSTRING_LEN(data));
+  if (hp->offset > MAX_HEADER_LEN)
+    parser_raise(e413, "HTTP header is too large");
+
+  if (hp->cs == http_parser_first_final ||
+      hp->cs == http_parser_en_ChunkedBody) {
+    advance_str(data, hp->offset + 1);
+    hp->offset = 0;
+    if (HP_FL_TEST(hp, INTRAILER))
+      HP_FL_SET(hp, REQEOF);
+
+    return hp->env;
+  }
+
+  if (hp->cs == http_parser_error)
+    parser_raise(eHttpParserError, "Invalid HTTP format, parsing fails.");
+
+  return Qnil;
+}
+
+/**
+ * Document-method: parse
+ * call-seq:
+ *    parser.add_parse(buffer) => env or nil
+ *
+ * adds the contents of +buffer+ to the internal buffer and attempts to
+ * continue parsing.  Returns the +env+ Hash on success or nil if more
+ * data is needed.
+ *
+ * Raises HttpParserError if there are parsing errors.
+ */
+static VALUE HttpParser_add_parse(VALUE self, VALUE buffer)
+{
+  struct http_parser *hp = data_get(self);
+
+  Check_Type(buffer, T_STRING);
+  rb_str_buf_append(hp->buf, buffer);
+
+  return HttpParser_parse(self);
+}
+
+/**
+ * Document-method: trailers
+ * call-seq:
+ *    parser.trailers(req, data) => req or nil
+ *
+ * This is an alias for HttpParser#headers
+ */
+
+/**
+ * Document-method: headers
+ */
+static VALUE HttpParser_headers(VALUE self, VALUE env, VALUE buf)
+{
+  struct http_parser *hp = data_get(self);
+
+  hp->env = env;
+  hp->buf = buf;
+
+  return HttpParser_parse(self);
+}
+
+static int chunked_eof(struct http_parser *hp)
+{
+  return ((hp->cs == http_parser_first_final) || HP_FL_TEST(hp, INTRAILER));
+}
+
+/**
+ * call-seq:
+ *    parser.body_eof? => true or false
+ *
+ * Detects if we're done filtering the body or not.  This can be used
+ * to detect when to stop calling HttpParser#filter_body.
+ */
+static VALUE HttpParser_body_eof(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+
+  if (HP_FL_TEST(hp, CHUNKED))
+    return chunked_eof(hp) ? Qtrue : Qfalse;
+
+  return hp->len.content == 0 ? Qtrue : Qfalse;
+}
+
+/**
+ * call-seq:
+ *    parser.keepalive? => true or false
+ *
+ * This should be used to detect if a request can really handle
+ * keepalives and pipelining.  Currently, the rules are:
+ *
+ * 1. MUST be a GET or HEAD request
+ * 2. MUST be HTTP/1.1 +or+ HTTP/1.0 with "Connection: keep-alive"
+ * 3. MUST NOT have "Connection: close" set
+ */
+static VALUE HttpParser_keepalive(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+
+  return HP_FL_ALL(hp, KEEPALIVE) ? Qtrue : Qfalse;
+}
+
+/* :nodoc: */
+static VALUE chunkable_response_p(VALUE self)
+{
+  const struct http_parser *hp = data_get(self);
+
+  return HP_FL_ALL(hp, RES_CHUNKABLE) ? Qtrue : Qfalse;
+}
+
+/**
+ * call-seq:
+ *    parser.next? => true or false
+ *
+ * Exactly like HttpParser#keepalive?, except it will reset the internal
+ * parser state on next parse if it returns true.
+ */
+static VALUE HttpParser_next(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+
+  if (HP_FL_ALL(hp, KEEPALIVE)) {
+    HP_FL_SET(hp, TO_CLEAR);
+    return Qtrue;
+  }
+  return Qfalse;
+}
+
+/**
+ * call-seq:
+ *    parser.headers? => true or false
+ *
+ * This should be used to detect if a request has headers (and if
+ * the response will have headers as well).  HTTP/0.9 requests
+ * should return false, all subsequent HTTP versions will return true
+ */
+static VALUE HttpParser_has_headers(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+
+  return HP_FL_TEST(hp, HASHEADER) ? Qtrue : Qfalse;
+}
+
+static VALUE HttpParser_buf(VALUE self)
+{
+  return data_get(self)->buf;
+}
+
+static VALUE HttpParser_env(VALUE self)
+{
+  return data_get(self)->env;
+}
+
+static VALUE HttpParser_hijacked_bang(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+
+  HP_FL_SET(hp, HIJACK);
+
+  return self;
+}
+
+/**
+ * call-seq:
+ *    parser.filter_body(dst, src) => nil/src
+ *
+ * Takes a String of +src+, will modify data if dechunking is done.
+ * Returns +nil+ if there is more data left to process.  Returns
+ * +src+ if body processing is complete. When returning +src+,
+ * it may modify +src+ so the start of the string points to where
+ * the body ended so that trailer processing can begin.
+ *
+ * Raises HttpParserError if there are dechunking errors.
+ * Basically this is a glorified memcpy(3) that copies +src+
+ * into +buf+ while filtering it through the dechunker.
+ */
+static VALUE HttpParser_filter_body(VALUE self, VALUE dst, VALUE src)
+{
+  struct http_parser *hp = data_get(self);
+  char *srcptr;
+  long srclen;
+
+  srcptr = RSTRING_PTR(src);
+  srclen = RSTRING_LEN(src);
+
+  StringValue(dst);
+
+  if (HP_FL_TEST(hp, CHUNKED)) {
+    if (!chunked_eof(hp)) {
+      rb_str_modify(dst);
+      rb_str_resize(dst, srclen); /* we can never copy more than srclen bytes */
+
+      hp->s.dest_offset = 0;
+      hp->cont = dst;
+      hp->buf = src;
+      http_parser_execute(hp, srcptr, srclen);
+      if (hp->cs == http_parser_error)
+        parser_raise(eHttpParserError, "Invalid HTTP format, parsing fails.");
+
+      assert(hp->s.dest_offset <= hp->offset &&
+             "destination buffer overflow");
+      advance_str(src, hp->offset);
+      rb_str_set_len(dst, hp->s.dest_offset);
+
+      if (RSTRING_LEN(dst) == 0 && chunked_eof(hp)) {
+        assert(hp->len.chunk == 0 && "chunk at EOF but more to parse");
+      } else {
+        src = Qnil;
+      }
+    }
+  } else {
+    /* no need to enter the Ragel machine for unchunked transfers */
+    assert(hp->len.content >= 0 && "negative Content-Length");
+    if (hp->len.content > 0) {
+      long nr = MIN(srclen, hp->len.content);
+
+      rb_str_modify(dst);
+      rb_str_resize(dst, nr);
+      /*
+       * using rb_str_replace() to avoid memcpy() doesn't help in
+       * most cases because a GC-aware programmer will pass an explicit
+       * buffer to env["rack.input"].read and reuse the buffer in a loop.
+       * This causes copy-on-write behavior to be triggered anyways
+       * when the +src+ buffer is modified (when reading off the socket).
+       */
+      hp->buf = src;
+      memcpy(RSTRING_PTR(dst), srcptr, nr);
+      hp->len.content -= nr;
+      if (hp->len.content == 0) {
+        HP_FL_SET(hp, REQEOF);
+        hp->cs = http_parser_first_final;
+      }
+      advance_str(src, nr);
+      src = Qnil;
+    }
+  }
+  hp->offset = 0; /* for trailer parsing */
+  return src;
+}
+
+static VALUE HttpParser_rssset(VALUE self, VALUE boolean)
+{
+  struct http_parser *hp = data_get(self);
+
+  if (RTEST(boolean))
+    HP_FL_SET(hp, RESSTART);
+  else
+    HP_FL_UNSET(hp, RESSTART);
+
+  return boolean; /* ignored by Ruby anyways */
+}
+
+static VALUE HttpParser_rssget(VALUE self)
+{
+  struct http_parser *hp = data_get(self);
+
+  return HP_FL_TEST(hp, RESSTART) ? Qtrue : Qfalse;
+}
+
+#define SET_GLOBAL(var,str) do { \
+  var = find_common_field(str, sizeof(str) - 1); \
+  assert(!NIL_P(var) && "missed global field"); \
+} while (0)
+
+void Init_unicorn_http(void)
+{
+  VALUE mUnicorn;
+
+  mUnicorn = rb_define_module("Unicorn");
+  cHttpParser = rb_define_class_under(mUnicorn, "HttpParser", rb_cObject);
+  eHttpParserError =
+         rb_define_class_under(mUnicorn, "HttpParserError", rb_eIOError);
+  e413 = rb_define_class_under(mUnicorn, "RequestEntityTooLargeError",
+                               eHttpParserError);
+  e414 = rb_define_class_under(mUnicorn, "RequestURITooLongError",
+                               eHttpParserError);
+
+  id_uminus = rb_intern("-@");
+  init_globals();
+  rb_define_alloc_func(cHttpParser, HttpParser_alloc);
+  rb_define_method(cHttpParser, "initialize", HttpParser_init, 0);
+  rb_define_method(cHttpParser, "clear", HttpParser_clear, 0);
+  rb_define_method(cHttpParser, "parse", HttpParser_parse, 0);
+  rb_define_method(cHttpParser, "add_parse", HttpParser_add_parse, 1);
+  rb_define_method(cHttpParser, "headers", HttpParser_headers, 2);
+  rb_define_method(cHttpParser, "trailers", HttpParser_headers, 2);
+  rb_define_method(cHttpParser, "filter_body", HttpParser_filter_body, 2);
+  rb_define_method(cHttpParser, "content_length", HttpParser_content_length, 0);
+  rb_define_method(cHttpParser, "body_eof?", HttpParser_body_eof, 0);
+  rb_define_method(cHttpParser, "keepalive?", HttpParser_keepalive, 0);
+  rb_define_method(cHttpParser, "chunkable_response?", chunkable_response_p, 0);
+  rb_define_method(cHttpParser, "headers?", HttpParser_has_headers, 0);
+  rb_define_method(cHttpParser, "next?", HttpParser_next, 0);
+  rb_define_method(cHttpParser, "buf", HttpParser_buf, 0);
+  rb_define_method(cHttpParser, "env", HttpParser_env, 0);
+  rb_define_method(cHttpParser, "hijacked!", HttpParser_hijacked_bang, 0);
+  rb_define_method(cHttpParser, "response_start_sent=", HttpParser_rssset, 1);
+  rb_define_method(cHttpParser, "response_start_sent", HttpParser_rssget, 0);
+
+  /*
+   * The maximum size a single chunk when using chunked transfer encoding.
+   * This is only a theoretical maximum used to detect errors in clients,
+   * it is highly unlikely to encounter clients that send more than
+   * several kilobytes at once.
+   */
+  rb_define_const(cHttpParser, "CHUNK_MAX", OFFT2NUM(UH_OFF_T_MAX));
+
+  /*
+   * The maximum size of the body as specified by Content-Length.
+   * This is only a theoretical maximum, the actual limit is subject
+   * to the limits of the file system used for +Dir.tmpdir+.
+   */
+  rb_define_const(cHttpParser, "LENGTH_MAX", OFFT2NUM(UH_OFF_T_MAX));
+
+  rb_define_singleton_method(cHttpParser, "max_header_len=", set_maxhdrlen, 1);
+
+  init_common_fields();
+  SET_GLOBAL(g_http_host, "HOST");
+  SET_GLOBAL(g_http_trailer, "TRAILER");
+  SET_GLOBAL(g_http_transfer_encoding, "TRANSFER_ENCODING");
+  SET_GLOBAL(g_content_length, "CONTENT_LENGTH");
+  SET_GLOBAL(g_http_connection, "CONNECTION");
+  id_set_backtrace = rb_intern("set_backtrace");
+  init_unicorn_httpdate();
+
+  id_is_chunked_p = rb_intern("is_chunked?");
+
+  init_epollexclusive(mUnicorn);
+}
+#undef SET_GLOBAL