rage-iodine 1.7.58

data/ext/iodine/http.c

@@ -0,0 +1,2736 @@
+/*
+Copyright: Boaz Segev, 2016-2019
+License: MIT
+
+Feel free to copy, use and enjoy according to the license provided.
+*/
+
+#include <fio.h>
+
+#include <http1.h>
+#include <http_internal.h>
+
+#include <ctype.h>
+#include <fcntl.h>
+#include <signal.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include <pthread.h>
+
+#ifndef HAVE_TM_TM_ZONE
+#if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) ||     \
+    defined(__DragonFly__) || defined(__bsdi__) || defined(__ultrix) ||        \
+    (defined(__APPLE__) && defined(__MACH__)) ||                               \
+    (defined(__sun) && !defined(__SVR4))
+/* Known BSD systems */
+#define HAVE_TM_TM_ZONE 1
+#elif defined(__GLIBC__) && defined(_BSD_SOURCE)
+/* GNU systems with _BSD_SOURCE */
+#define HAVE_TM_TM_ZONE 1
+#else
+#define HAVE_TM_TM_ZONE 0
+#endif
+#endif
+
+#ifndef __MINGW32__
+/* *****************************************************************************
+SSL/TLS patch
+***************************************************************************** */
+
+/**
+ * Adds an ALPN protocol callback to the SSL/TLS context.
+ *
+ * The first protocol added will act as the default protocol to be selected.
+ */
+void __attribute__((weak))
+fio_tls_alpn_add(void *tls, const char *protocol_name,
+                 void (*callback)(intptr_t uuid, void *udata_connection,
+                                  void *udata_tls),
+                 void *udata_tls, void (*on_cleanup)(void *udata_tls)) {
+  FIO_LOG_FATAL("HTTP SSL/TLS required but unavailable!");
+  exit(-1);
+  (void)tls;
+  (void)protocol_name;
+  (void)callback;
+  (void)on_cleanup;
+  (void)udata_tls;
+}
+#pragma weak fio_tls_alpn_add
+#endif
+
+/* *****************************************************************************
+Small Helpers
+***************************************************************************** */
+static inline int hex2byte(uint8_t *dest, const uint8_t *source);
+
+static inline void add_content_length(http_s *r, uintptr_t length) {
+  static uint64_t cl_hash = 0;
+  if (!cl_hash)
+    cl_hash = fiobj_hash_string("content-length", 14);
+  if (!fiobj_hash_get2(r->private_data.out_headers, cl_hash)) {
+    fiobj_hash_set(r->private_data.out_headers, HTTP_HEADER_CONTENT_LENGTH,
+                   fiobj_num_new(length));
+  }
+}
+static inline void add_content_type(http_s *r) {
+  static uint64_t ct_hash = 0;
+  if (!ct_hash)
+    ct_hash = fiobj_hash_string("content-type", 12);
+  if (!fiobj_hash_get2(r->private_data.out_headers, ct_hash)) {
+    fiobj_hash_set(r->private_data.out_headers, HTTP_HEADER_CONTENT_TYPE,
+                   http_mimetype_find2(r->path));
+  }
+}
+
+static FIOBJ current_date;
+static time_t last_date_added;
+static fio_lock_i date_lock;
+static inline void add_date(http_s *r) {
+  static uint64_t date_hash = 0;
+  if (!date_hash)
+    date_hash = fiobj_hash_string("date", 4);
+
+  if (fio_last_tick().tv_sec > last_date_added) {
+    fio_lock(&date_lock);
+    if (fio_last_tick().tv_sec > last_date_added) { /* retest inside lock */
+      /* 32 chars are ok for a while, but http_time2str below has a buffer sized
+       * 48 chars and does a memcpy ... */
+      FIOBJ tmp = fiobj_str_buf(32);
+      FIOBJ old = current_date;
+      fiobj_str_resize(
+          tmp, http_time2str(fiobj_obj2cstr(tmp).data, fio_last_tick().tv_sec));
+      last_date_added = fio_last_tick().tv_sec;
+      current_date = tmp;
+      fiobj_free(old);
+    }
+    fio_unlock(&date_lock);
+  }
+
+  if (!fiobj_hash_get2(r->private_data.out_headers, date_hash)) {
+    fiobj_hash_set(r->private_data.out_headers, HTTP_HEADER_DATE,
+                   fiobj_dup(current_date));
+  }
+}
+
+struct header_writer_s {
+  FIOBJ dest;
+  FIOBJ name;
+  FIOBJ value;
+};
+
+static int write_header(FIOBJ o, void *w_) {
+  struct header_writer_s *w = w_;
+  if (!o)
+    return 0;
+  if (fiobj_hash_key_in_loop()) {
+    w->name = fiobj_hash_key_in_loop();
+  }
+  if (FIOBJ_TYPE_IS(o, FIOBJ_T_ARRAY)) {
+    fiobj_each1(o, 0, write_header, w);
+    return 0;
+  }
+  fio_str_info_s name = fiobj_obj2cstr(w->name);
+  fio_str_info_s str = fiobj_obj2cstr(o);
+  if (!str.data)
+    return 0;
+  fiobj_str_write(w->dest, name.data, name.len);
+  fiobj_str_write(w->dest, ":", 1);
+  fiobj_str_write(w->dest, str.data, str.len);
+  fiobj_str_write(w->dest, "\r\n", 2);
+  return 0;
+}
+
+static char invalid_cookie_name_char[256];
+
+static char invalid_cookie_value_char[256];
+/* *****************************************************************************
+The Request / Response type and functions
+***************************************************************************** */
+static const char hex_chars[] = {'0', '1', '2', '3', '4', '5', '6', '7',
+                                 '8', '9', 'A', 'B', 'C', 'D', 'E', 'F'};
+
+/**
+ * Sets a response header, taking ownership of the value object, but NOT the
+ * name object (so name objects could be reused in future responses).
+ *
+ * Returns -1 on error and 0 on success.
+ */
+int http_set_header(http_s *r, FIOBJ name, FIOBJ value) {
+  if (HTTP_INVALID_HANDLE(r) || !name) {
+    fiobj_free(value);
+    return -1;
+  }
+  set_header_add(r->private_data.out_headers, name, value);
+  return 0;
+}
+/**
+ * Sets a response header.
+ *
+ * Returns -1 on error and 0 on success.
+ */
+int http_set_header2(http_s *r, fio_str_info_s n, fio_str_info_s v) {
+  if (HTTP_INVALID_HANDLE(r) || !n.data || !n.len || (v.data && !v.len))
+    return -1;
+  FIOBJ tmp = fiobj_str_new(n.data, n.len);
+  int ret = http_set_header(r, tmp, fiobj_str_new(v.data, v.len));
+  fiobj_free(tmp);
+  return ret;
+}
+
+/**
+ * Sets a response cookie, taking ownership of the value object, but NOT the
+ * name object (so name objects could be reused in future responses).
+ *
+ * Returns -1 on error and 0 on success.
+ */
+#undef http_set_cookie
+int http_set_cookie(http_s *h, http_cookie_args_s cookie) {
+#if DEBUG
+  FIO_ASSERT(h, "Can't set cookie for NULL HTTP handler!");
+#endif
+  if (HTTP_INVALID_HANDLE(h) || cookie.name_len >= 32768 ||
+      cookie.value_len >= 131072) {
+    return -1;
+  }
+
+  static int warn_illegal = 0;
+
+  /* write name and value while auto-correcting encoding issues */
+  size_t capa = cookie.name_len + cookie.value_len + 128;
+  size_t len = 0;
+  FIOBJ c = fiobj_str_buf(capa);
+  fio_str_info_s t = fiobj_obj2cstr(c);
+
+#define copy_cookie_ch(ch_var)                                                 \
+  if (invalid_cookie_##ch_var##_char[(uint8_t)cookie.ch_var[tmp]]) {           \
+    if (!warn_illegal) {                                                       \
+      ++warn_illegal;                                                          \
+      FIO_LOG_WARNING("illegal char 0x%.2x in cookie " #ch_var " (in %s)\n"    \
+                      "         automatic %% encoding applied",                \
+                      cookie.ch_var[tmp], cookie.ch_var);                      \
+    }                                                                          \
+    t.data[len++] = '%';                                                       \
+    t.data[len++] = hex_chars[((uint8_t)cookie.ch_var[tmp] >> 4) & 0x0F];      \
+    t.data[len++] = hex_chars[(uint8_t)cookie.ch_var[tmp] & 0x0F];             \
+  } else {                                                                     \
+    t.data[len++] = cookie.ch_var[tmp];                                        \
+  }                                                                            \
+  tmp += 1;                                                                    \
+  if (capa <= len + 3) {                                                       \
+    capa += 32;                                                                \
+    fiobj_str_capa_assert(c, capa);                                            \
+    t = fiobj_obj2cstr(c);                                                     \
+  }
+
+  if (cookie.name) {
+    size_t tmp = 0;
+    if (cookie.name_len) {
+      while (tmp < cookie.name_len) {
+        copy_cookie_ch(name);
+      }
+    } else {
+      while (cookie.name[tmp]) {
+        copy_cookie_ch(name);
+      }
+    }
+  }
+  t.data[len++] = '=';
+  if (cookie.value) {
+    size_t tmp = 0;
+    if (cookie.value_len) {
+      while (tmp < cookie.value_len) {
+        copy_cookie_ch(value);
+      }
+    } else {
+      while (cookie.value[tmp]) {
+        copy_cookie_ch(value);
+      }
+    }
+  } else
+    cookie.max_age = -1;
+
+  if (http_settings(h) && http_settings(h)->is_client) {
+    if (!cookie.value) {
+      fiobj_free(c);
+      return -1;
+    }
+    set_header_add(h->private_data.out_headers, HTTP_HEADER_COOKIE, c);
+    return 0;
+  }
+
+  t.data[len++] = ';';
+  t.data[len++] = ' ';
+
+  if (h->status_str || !h->status) { /* on first request status == 0 */
+    static uint64_t cookie_hash;
+    if (!cookie_hash)
+      cookie_hash = fiobj_hash_string("cookie", 6);
+    FIOBJ tmp = fiobj_hash_get2(h->private_data.out_headers, cookie_hash);
+    if (!tmp) {
+      set_header_add(h->private_data.out_headers, HTTP_HEADER_COOKIE, c);
+    } else {
+      fiobj_str_join(tmp, c);
+      fiobj_free(c);
+    }
+    return 0;
+  }
+
+  if (capa <= len + 40) {
+    capa = len + 40;
+    fiobj_str_capa_assert(c, capa);
+    t = fiobj_obj2cstr(c);
+  }
+  if (cookie.max_age) {
+    memcpy(t.data + len, "Max-Age=", 8);
+    len += 8;
+    len += fio_ltoa(t.data + len, cookie.max_age, 10);
+    t.data[len++] = ';';
+    t.data[len++] = ' ';
+  }
+  fiobj_str_resize(c, len);
+
+  if (cookie.domain && cookie.domain_len) {
+    fiobj_str_write(c, "domain=", 7);
+    fiobj_str_write(c, cookie.domain, cookie.domain_len);
+    fiobj_str_write(c, ";", 1);
+    t.data[len++] = ' ';
+  }
+  if (cookie.path && cookie.path_len) {
+    fiobj_str_write(c, "path=", 5);
+    fiobj_str_write(c, cookie.path, cookie.path_len);
+    fiobj_str_write(c, ";", 1);
+    t.data[len++] = ' ';
+  }
+  if (cookie.http_only) {
+    fiobj_str_write(c, "HttpOnly;", 9);
+  }
+  if (cookie.secure) {
+    fiobj_str_write(c, "secure;", 7);
+  }
+  set_header_add(h->private_data.out_headers, HTTP_HEADER_SET_COOKIE, c);
+  return 0;
+}
+#define http_set_cookie(http__req__, ...)                                      \
+  http_set_cookie((http__req__), (http_cookie_args_s){__VA_ARGS__})
+
+/**
+ * Sends the response headers and body.
+ *
+ * Returns -1 on error and 0 on success.
+ *
+ * AFTER THIS FUNCTION IS CALLED, THE `http_s` OBJECT IS NO LONGER VALID.
+ */
+int http_send_body(http_s *r, void *data, uintptr_t length) {
+  if (HTTP_INVALID_HANDLE(r))
+    return -1;
+  if (!length || !data) {
+    http_finish(r);
+    return 0;
+  }
+  add_content_length(r, length);
+  // add_content_type(r);
+  add_date(r);
+  return ((http_vtable_s *)r->private_data.vtbl)
+      ->http_send_body(r, data, length);
+}
+/**
+ * Sends the response headers and the specified file (the response's body).
+ *
+ * Returns -1 on error and 0 on success.
+ *
+ * AFTER THIS FUNCTION IS CALLED, THE `http_s` OBJECT IS NO LONGER VALID.
+ */
+int http_sendfile(http_s *r, int fd, uintptr_t length, uintptr_t offset) {
+  if (HTTP_INVALID_HANDLE(r)) {
+    close(fd);
+    return -1;
+  };
+  add_content_length(r, length);
+  add_content_type(r);
+  add_date(r);
+  return ((http_vtable_s *)r->private_data.vtbl)
+      ->http_sendfile(r, fd, length, offset);
+}
+
+static inline int http_test_encoded_path(const char *mem, size_t len) {
+  const char *pos = NULL;
+  const char *end = mem + len;
+  while (mem < end && (pos = memchr(mem, '/', (size_t)len))) {
+    len = end - pos;
+    mem = pos + 1;
+    if (pos[1] == '/')
+      return -1;
+    if (len > 3 && pos[1] == '.' && pos[2] == '.' && pos[3] == '/')
+      return -1;
+  }
+  return 0;
+}
+
+/**
+ * Sends the response headers and the specified file (the response's body).
+ *
+ * Returns -1 eton error and 0 on success.
+ *
+ * AFTER THIS FUNCTION IS CALLED, THE `http_s` OBJECT IS NO LONGER VALID.
+ */
+int http_sendfile2(http_s *h, const char *prefix, size_t prefix_len,
+                   const char *encoded, size_t encoded_len) {
+  if (HTTP_INVALID_HANDLE(h))
+    return -1;
+  struct stat file_data = {.st_size = 0};
+  static uint64_t accept_enc_hash = 0;
+  if (!accept_enc_hash)
+    accept_enc_hash = fiobj_hash_string("accept-encoding", 15);
+  static uint64_t range_hash = 0;
+  if (!range_hash)
+    range_hash = fiobj_hash_string("range", 5);
+
+  /* create filename string */
+  FIOBJ filename = fiobj_str_tmp();
+  if (prefix && prefix_len) {
+    /* start with prefix path */
+    if (encoded && prefix[prefix_len - 1] == '/' && encoded[0] == '/')
+      --prefix_len;
+    fiobj_str_capa_assert(filename, prefix_len + encoded_len + 4);
+    fiobj_str_write(filename, prefix, prefix_len);
+  }
+  {
+    /* decode filename in cases where it's URL encoded */
+    fio_str_info_s tmp = fiobj_obj2cstr(filename);
+    if (encoded) {
+      char *pos = (char *)encoded;
+      const char *end = encoded + encoded_len;
+      while (pos < end) {
+        if (*pos == '%') {
+          // decode hex value (this is a percent encoded value).
+          if (hex2byte((uint8_t *)tmp.data + tmp.len, (uint8_t *)pos + 1))
+            return -1;
+          tmp.len++;
+          pos += 3;
+        } else
+          tmp.data[tmp.len++] = *(pos++);
+      }
+      tmp.data[tmp.len] = 0;
+      fiobj_str_resize(filename, tmp.len);
+      /* test for path manipulations after decoding */
+      if (http_test_encoded_path(tmp.data + prefix_len, tmp.len - prefix_len))
+        return -1;
+    }
+    if (tmp.data[tmp.len - 1] == '/')
+      fiobj_str_write(filename, "index.html", 10);
+  }
+  /* test for file existance  */
+
+  int file = -1;
+  uint8_t is_gz = 0;
+
+  fio_str_info_s s = fiobj_obj2cstr(filename);
+  {
+    FIOBJ tmp = fiobj_hash_get2(h->headers, accept_enc_hash);
+    if (!tmp)
+      goto no_gzip_support;
+    fio_str_info_s ac_str = fiobj_obj2cstr(tmp);
+    if (!ac_str.data || !strstr(ac_str.data, "gzip"))
+      goto no_gzip_support;
+    if (s.data[s.len - 3] != '.' || s.data[s.len - 2] != 'g' ||
+        s.data[s.len - 1] != 'z') {
+      fiobj_str_write(filename, ".gz", 3);
+      s = fiobj_obj2cstr(filename);
+      if (!stat(s.data, &file_data) &&
+          (S_ISREG(file_data.st_mode) || S_ISLNK(file_data.st_mode))) {
+        is_gz = 1;
+        goto found_file;
+      }
+      fiobj_str_resize(filename, s.len - 3);
+    }
+  }
+no_gzip_support:
+  if (stat(s.data, &file_data) ||
+      !(S_ISREG(file_data.st_mode) || S_ISLNK(file_data.st_mode)))
+    return -1;
+found_file:
+  /* set last-modified */
+  {
+    FIOBJ tmp = fiobj_str_buf(32);
+    fiobj_str_resize(
+        tmp, http_time2str(fiobj_obj2cstr(tmp).data, file_data.st_mtime));
+    http_set_header(h, HTTP_HEADER_LAST_MODIFIED, tmp);
+  }
+  /* set cache-control */
+  http_set_header(h, HTTP_HEADER_CACHE_CONTROL, fiobj_dup(HTTP_HVALUE_MAX_AGE));
+  /* set & test etag */
+  uint64_t etag = (uint64_t)file_data.st_size;
+  etag ^= (uint64_t)file_data.st_mtime;
+  etag = fiobj_hash_string(&etag, sizeof(uint64_t));
+  FIOBJ etag_str = fiobj_str_buf(32);
+  fiobj_str_resize(etag_str,
+                   fio_base64_encode(fiobj_obj2cstr(etag_str).data,
+                                     (void *)&etag, sizeof(uint64_t)));
+  /* set */
+  http_set_header(h, HTTP_HEADER_ETAG, etag_str);
+  /* test */
+  {
+    static uint64_t none_match_hash = 0;
+    if (!none_match_hash)
+      none_match_hash = fiobj_hash_string("if-none-match", 13);
+    FIOBJ tmp2 = fiobj_hash_get2(h->headers, none_match_hash);
+    if (tmp2 && fiobj_iseq(tmp2, etag_str)) {
+      h->status = 304;
+      http_finish(h);
+      return 0;
+    }
+  }
+  /* handle range requests */
+  int64_t offset = 0;
+  int64_t length = file_data.st_size;
+  {
+    static uint64_t ifrange_hash = 0;
+    if (!ifrange_hash)
+      ifrange_hash = fiobj_hash_string("if-range", 8);
+    FIOBJ tmp = fiobj_hash_get2(h->headers, ifrange_hash);
+    if (tmp && fiobj_iseq(tmp, etag_str)) {
+      fiobj_hash_delete2(h->headers, range_hash);
+    } else {
+      tmp = fiobj_hash_get2(h->headers, range_hash);
+      if (tmp) {
+        /* range ahead... */
+        if (FIOBJ_TYPE_IS(tmp, FIOBJ_T_ARRAY))
+          tmp = fiobj_ary_index(tmp, 0);
+        fio_str_info_s range = fiobj_obj2cstr(tmp);
+        if (!range.data || memcmp("bytes=", range.data, 6))
+          goto open_file;
+        char *pos = range.data + 6;
+        int64_t start_at = 0, end_at = 0;
+        start_at = fio_atol(&pos);
+        if (start_at >= file_data.st_size)
+          goto open_file;
+        if (start_at >= 0) {
+          pos++;
+          end_at = fio_atol(&pos);
+          if (end_at <= 0)
+            goto open_file;
+        }
+        /* we ignore multimple ranges, only responding with the first range. */
+        if (start_at < 0) {
+          if (0 - start_at < file_data.st_size) {
+            offset = file_data.st_size - start_at;
+            length = 0 - start_at;
+          }
+        } else if (end_at) {
+          offset = start_at;
+          length = end_at - start_at + 1;
+          if (length + start_at > file_data.st_size || length <= 0)
+            length = length - start_at;
+        } else {
+          offset = start_at;
+          length = length - start_at;
+        }
+        h->status = 206;
+
+        {
+          FIOBJ cranges = fiobj_str_buf(1);
+          fiobj_str_printf(cranges, "bytes %lu-%lu/%lu",
+                           (unsigned long)start_at,
+                           (unsigned long)(start_at + length - 1),
+                           (unsigned long)file_data.st_size);
+          http_set_header(h, HTTP_HEADER_CONTENT_RANGE, cranges);
+        }
+        http_set_header(h, HTTP_HEADER_ACCEPT_RANGES,
+                        fiobj_dup(HTTP_HVALUE_BYTES));
+      }
+    }
+  }
+  /* test for an OPTIONS request or invalid methods */
+  s = fiobj_obj2cstr(h->method);
+  switch (s.len) {
+  case 7:
+    if (!strncasecmp("options", s.data, 7)) {
+      http_set_header2(h, (fio_str_info_s){.data = (char *)"allow", .len = 5},
+                       (fio_str_info_s){.data = (char *)"GET, HEAD", .len = 9});
+      h->status = 200;
+      http_finish(h);
+      return 0;
+    }
+    break;
+  case 3:
+    if (!strncasecmp("get", s.data, 3))
+      goto open_file;
+    break;
+  case 4:
+    if (!strncasecmp("head", s.data, 4)) {
+      http_set_header(h, HTTP_HEADER_CONTENT_LENGTH, fiobj_num_new(length));
+      http_finish(h);
+      return 0;
+    }
+    goto open_file;
+    break;
+  }
+  http_send_error(h, 403);
+  return 0;
+open_file:
+  s = fiobj_obj2cstr(filename);
+  file = open(s.data, O_RDONLY);
+  if (file == -1) {
+    FIO_LOG_ERROR("(HTTP) couldn't open file %s!\n", s.data);
+    perror("     ");
+    http_send_error(h, 500);
+    return 0;
+  }
+  {
+    FIOBJ tmp = 0;
+    uintptr_t pos = 0;
+    if (is_gz) {
+      http_set_header(h, HTTP_HEADER_CONTENT_ENCODING,
+                      fiobj_dup(HTTP_HVALUE_GZIP));
+
+      pos = s.len - 4;
+      while (pos && s.data[pos] != '.')
+        pos--;
+      pos++; /* assuming, but that's fine. */
+      tmp = http_mimetype_find(s.data + pos, s.len - pos - 3);
+
+    } else {
+      pos = s.len - 1;
+      while (pos && s.data[pos] != '.')
+        pos--;
+      pos++; /* assuming, but that's fine. */
+      tmp = http_mimetype_find(s.data + pos, s.len - pos);
+    }
+    if (tmp)
+      http_set_header(h, HTTP_HEADER_CONTENT_TYPE, tmp);
+  }
+  http_sendfile(h, file, length, offset);
+  return 0;
+}
+
+/**
+ * Sends an HTTP error response.
+ *
+ * Returns -1 on error and 0 on success.
+ *
+ * AFTER THIS FUNCTION IS CALLED, THE `http_s` OBJECT IS NO LONGER VALID.
+ *
+ * The `uuid` argument is optional and will be used only if the `http_s`
+ * argument is set to NULL.
+ */
+int http_send_error(http_s *r, size_t error) {
+  if (!r || !r->private_data.out_headers) {
+    return -1;
+  }
+  if (error < 100 || error >= 1000)
+    error = 500;
+  r->status = error;
+  char buffer[16];
+  buffer[0] = '/';
+  size_t pos = 1 + fio_ltoa(buffer + 1, error, 10);
+  buffer[pos++] = '.';
+  buffer[pos++] = 'h';
+  buffer[pos++] = 't';
+  buffer[pos++] = 'm';
+  buffer[pos++] = 'l';
+  buffer[pos] = 0;
+  if (http_sendfile2(r, http2protocol(r)->settings->public_folder,
+                     http2protocol(r)->settings->public_folder_length, buffer,
+                     pos)) {
+    http_set_header(r, HTTP_HEADER_CONTENT_TYPE,
+                    http_mimetype_find((char *)"txt", 3));
+    fio_str_info_s t = http_status2str(error);
+    http_send_body(r, t.data, t.len);
+  }
+  return 0;
+}
+
+/**
+ * Sends the response headers for a header only response.
+ *
+ * AFTER THIS FUNCTION IS CALLED, THE `http_s` OBJECT IS NO LONGER VALID.
+ */
+void http_finish(http_s *r) {
+  if (!r || !r->private_data.vtbl) {
+    return;
+  }
+  add_content_length(r, 0);
+  add_date(r);
+  ((http_vtable_s *)r->private_data.vtbl)->http_finish(r);
+}
+/**
+ * Pushes a data response when supported (HTTP/2 only).
+ *
+ * Returns -1 on error and 0 on success.
+ */
+int http_push_data(http_s *r, void *data, uintptr_t length, FIOBJ mime_type) {
+  if (!r || !(http_fio_protocol_s *)r->private_data.flag)
+    return -1;
+  return ((http_vtable_s *)r->private_data.vtbl)
+      ->http_push_data(r, data, length, mime_type);
+}
+/**
+ * Pushes a file response when supported (HTTP/2 only).
+ *
+ * If `mime_type` is NULL, an attempt at automatic detection using
+ * `filename` will be made.
+ *
+ * Returns -1 on error and 0 on success.
+ */
+int http_push_file(http_s *h, FIOBJ filename, FIOBJ mime_type) {
+  if (HTTP_INVALID_HANDLE(h))
+    return -1;
+  return ((http_vtable_s *)h->private_data.vtbl)
+      ->http_push_file(h, filename, mime_type);
+}
+
+/**
+ * Upgrades an HTTP/1.1 connection to a Websocket connection.
+ */
+#undef http_upgrade2ws
+int http_upgrade2ws(http_s *h, websocket_settings_s args) {
+  if (!h) {
+    FIO_LOG_ERROR("`http_upgrade2ws` requires a valid `http_s` handle.");
+    goto error;
+  }
+  if (HTTP_INVALID_HANDLE(h))
+    goto error;
+  return ((http_vtable_s *)h->private_data.vtbl)->http2websocket(h, &args);
+error:
+  if (args.on_close)
+    args.on_close(-1, args.udata);
+  return -1;
+}
+
+/* *****************************************************************************
+Pause / Resume
+***************************************************************************** */
+
+/** Returns the `udata` associated with the paused opaque handle */
+void *http_paused_udata_get(http_pause_handle_s *http) { return http->udata; }
+
+/**
+ * Sets the `udata` associated with the paused opaque handle, returning the
+ * old value.
+ */
+void *http_paused_udata_set(http_pause_handle_s *http, void *udata) {
+  void *old = http->udata;
+  http->udata = udata;
+  return old;
+}
+
+/* perform the pause task outside of the connection's lock */
+static void http_pause_wrapper(void *h_, void *task_) {
+  void (*task)(void *h) = (void (*)(void *h))((uintptr_t)task_);
+  task(h_);
+}
+
+/* perform the resume task within of the connection's lock */
+static void http_resume_wrapper(intptr_t uuid, fio_protocol_s *p_, void *arg) {
+  http_fio_protocol_s *p = (http_fio_protocol_s *)p_;
+  http_pause_handle_s *http = arg;
+  http_s *h = http->h;
+  h->udata = http->udata;
+  http_vtable_s *vtbl = (http_vtable_s *)h->private_data.vtbl;
+  if (http->task)
+    http->task(h);
+  vtbl->http_on_resume(h, p);
+  fio_free(http);
+  (void)uuid;
+}
+
+/* perform the resume task fallback */
+static void http_resume_fallback_wrapper(intptr_t uuid, void *arg) {
+  http_pause_handle_s *http = arg;
+  if (http->fallback)
+    http->fallback(http->udata);
+  fio_free(http);
+  (void)uuid;
+}
+
+/**
+ * Defers the request / response handling for later.
+ */
+void http_pause(http_s *h, void (*task)(http_pause_handle_s *http)) {
+  if (HTTP_INVALID_HANDLE(h)) {
+    return;
+  }
+  http_fio_protocol_s *p = (http_fio_protocol_s *)h->private_data.flag;
+  http_vtable_s *vtbl = (http_vtable_s *)h->private_data.vtbl;
+  http_pause_handle_s *http = fio_malloc(sizeof(*http));
+  FIO_ASSERT_ALLOC(http);
+  *http = (http_pause_handle_s){
+      .uuid = p->uuid,
+      .h = h,
+      .udata = h->udata,
+  };
+  vtbl->http_on_pause(h, p);
+  fio_defer(http_pause_wrapper, http, (void *)((uintptr_t)task));
+}
+
+/**
+ * Defers the request / response handling for later.
+ */
+void http_resume(http_pause_handle_s *http, void (*task)(http_s *h),
+                 void (*fallback)(void *udata)) {
+  if (!http)
+    return;
+  http->task = task;
+  http->fallback = fallback;
+  fio_defer_io_task(http->uuid, .udata = http, .type = FIO_PR_LOCK_TASK,
+                    .task = http_resume_wrapper,
+                    .fallback = http_resume_fallback_wrapper);
+}
+
+/**
+ * Hijacks the socket away from the HTTP protocol and away from facil.io.
+ */
+intptr_t http_hijack(http_s *h, fio_str_info_s *leftover) {
+  if (!h)
+    return -1;
+  return ((http_vtable_s *)h->private_data.vtbl)->http_hijack(h, leftover);
+}
+
+/* *****************************************************************************
+Setting the default settings and allocating a persistent copy
+***************************************************************************** */
+
+static void http_on_request_fallback(http_s *h) { http_send_error(h, 404); }
+static void http_on_upgrade_fallback(http_s *h, char *p, size_t i) {
+  http_send_error(h, 400);
+  (void)p;
+  (void)i;
+}
+static void http_on_response_fallback(http_s *h) { http_send_error(h, 400); }
+
+static http_settings_s *http_settings_new(http_settings_s arg_settings) {
+  /* TODO: improve locality by unifying malloc to a single call */
+  if (!arg_settings.on_request)
+    arg_settings.on_request = http_on_request_fallback;
+  if (!arg_settings.on_response)
+    arg_settings.on_response = http_on_response_fallback;
+  if (!arg_settings.on_upgrade)
+    arg_settings.on_upgrade = http_on_upgrade_fallback;
+
+  if (!arg_settings.max_body_size)
+    arg_settings.max_body_size = HTTP_DEFAULT_BODY_LIMIT;
+  if (!arg_settings.timeout)
+    arg_settings.timeout = 40;
+  if (!arg_settings.ws_max_msg_size)
+    arg_settings.ws_max_msg_size = 262144; /** defaults to ~250KB */
+  if (!arg_settings.ws_timeout)
+    arg_settings.ws_timeout = 40; /* defaults to 40 seconds */
+  if (!arg_settings.max_header_size)
+    arg_settings.max_header_size = 32 * 1024; /* defaults to 32Kib seconds */
+  if (arg_settings.max_clients <= 0 ||
+      (size_t)(arg_settings.max_clients + HTTP_BUSY_UNLESS_HAS_FDS) >
+          fio_capa()) {
+    arg_settings.max_clients = fio_capa();
+    if ((ssize_t)arg_settings.max_clients - HTTP_BUSY_UNLESS_HAS_FDS > 0)
+      arg_settings.max_clients -= HTTP_BUSY_UNLESS_HAS_FDS;
+  }
+
+  http_settings_s *settings = malloc(sizeof(*settings) + sizeof(void *));
+  *settings = arg_settings;
+
+  if (settings->public_folder) {
+    settings->public_folder_length = strlen(settings->public_folder);
+    if (settings->public_folder[0] == '~' &&
+        settings->public_folder[1] == '/' && getenv("HOME")) {
+      char *home = getenv("HOME");
+      size_t home_len = strlen(home);
+      char *tmp = malloc(settings->public_folder_length + home_len + 1);
+      FIO_ASSERT_ALLOC(tmp);
+      memcpy(tmp, home, home_len);
+      if (home[home_len - 1] == '/')
+        --home_len;
+      memcpy(tmp + home_len, settings->public_folder + 1,
+             settings->public_folder_length); // copy also the NULL
+      settings->public_folder = tmp;
+      settings->public_folder_length = strlen(settings->public_folder);
+    } else {
+      settings->public_folder = malloc(settings->public_folder_length + 1);
+      FIO_ASSERT_ALLOC(settings->public_folder);
+      memcpy((void *)settings->public_folder, arg_settings.public_folder,
+             settings->public_folder_length);
+      ((uint8_t *)settings->public_folder)[settings->public_folder_length] = 0;
+    }
+  }
+  return settings;
+}
+
+static void http_settings_free(http_settings_s *s) {
+  free((void *)s->public_folder);
+  free(s);
+}
+/* *****************************************************************************
+Listening to HTTP connections
+***************************************************************************** */
+
+static uint8_t fio_http_at_capa = 0;
+
+static void http_on_server_protocol_http1(intptr_t uuid, void *set,
+                                          void *ignr_) {
+  if ((unsigned int)fio_uuid2fd(uuid) >=
+      ((http_settings_s *)set)->max_clients) {
+    if (fio_uuid2fd(uuid) != -1) {
+      if (!fio_http_at_capa)
+        FIO_LOG_WARNING("HTTP server at capacity");
+      fio_http_at_capa = 1;
+      http_send_error2(503, uuid, set);
+      fio_close(uuid);
+    }
+    return;
+  }
+  fio_http_at_capa = 0;
+  fio_protocol_s *pr = http1_new(uuid, set, NULL, 0);
+  if (!pr)
+    fio_close(uuid);
+  else
+    fio_timeout_set(uuid, ((http_settings_s *)set)->timeout);
+  (void)ignr_;
+}
+
+static void http_on_open(intptr_t uuid, void *set) {
+  http_on_server_protocol_http1(uuid, set, NULL);
+}
+
+static void http_on_finish(intptr_t uuid, void *set) {
+  http_settings_s *settings = set;
+
+  if (settings->on_finish)
+    settings->on_finish(settings);
+
+  http_settings_free(settings);
+  (void)uuid;
+}
+
+/**
+ * Listens to HTTP connections at the specified `port`.
+ *
+ * Leave as NULL to ignore IP binding.
+ *
+ * Returns -1 on error and 0 on success.
+ */
+#undef http_listen
+intptr_t http_listen(const char *port, const char *binding,
+                     struct http_settings_s arg_settings) {
+  if (arg_settings.on_request == NULL) {
+    FIO_LOG_ERROR("http_listen requires the .on_request parameter "
+                  "to be set\n");
+    kill(0, SIGINT);
+    exit(11);
+  }
+
+  http_settings_s *settings = http_settings_new(arg_settings);
+  settings->is_client = 0;
+#ifndef __MINGW32__
+  if (settings->tls) {
+    fio_tls_alpn_add(settings->tls, "http/1.1", http_on_server_protocol_http1,
+                     NULL, NULL);
+  }
+#endif
+
+  return fio_listen(.port = port, .address = binding, .tls = arg_settings.tls,
+                    .on_finish = http_on_finish, .on_open = http_on_open,
+                    .udata = settings);
+}
+/** Listens to HTTP connections at the specified `port` and `binding`. */
+#define http_listen(port, binding, ...)                                        \
+  http_listen((port), (binding), (struct http_settings_s)(__VA_ARGS__))
+
+/**
+ * Returns the settings used to setup the connection.
+ *
+ * Returns NULL on error (i.e., connection was lost).
+ */
+struct http_settings_s *http_settings(http_s *r) {
+  return ((http_fio_protocol_s *)r->private_data.flag)->settings;
+}
+
+/**
+ * Returns the direct address of the connected peer (likely an intermediary).
+ */
+fio_str_info_s http_peer_addr(http_s *h) {
+  return fio_peer_addr(((http_fio_protocol_s *)h->private_data.flag)->uuid);
+}
+
+/* *****************************************************************************
+HTTP client connections
+***************************************************************************** */
+
+static void http_on_close_client(intptr_t uuid, fio_protocol_s *protocol) {
+  http_fio_protocol_s *p = (http_fio_protocol_s *)protocol;
+  http_settings_s *set = p->settings;
+  void (**original)(intptr_t, fio_protocol_s *) =
+      (void (**)(intptr_t, fio_protocol_s *))(set + 1);
+  if (set->on_finish)
+    set->on_finish(set);
+
+  original[0](uuid, protocol);
+  http_settings_free(set);
+}
+
+static void http_on_open_client_perform(http_settings_s *set) {
+  http_s *h = set->udata;
+  set->on_response(h);
+}
+static void http_on_open_client_http1(intptr_t uuid, void *set_,
+                                      void *ignore_) {
+  http_settings_s *set = set_;
+  http_s *h = set->udata;
+  fio_timeout_set(uuid, set->timeout);
+  fio_protocol_s *pr = http1_new(uuid, set, NULL, 0);
+  if (!pr) {
+    fio_close(uuid);
+    return;
+  }
+  { /* store the original on_close at the end of the struct, we wrap it. */
+    void (**original)(intptr_t, fio_protocol_s *) =
+        (void (**)(intptr_t, fio_protocol_s *))(set + 1);
+    *original = pr->on_close;
+    pr->on_close = http_on_close_client;
+  }
+  h->private_data.flag = (uintptr_t)pr;
+  h->private_data.vtbl = http1_vtable();
+  http_on_open_client_perform(set);
+  (void)ignore_;
+}
+
+static void http_on_open_client(intptr_t uuid, void *set_) {
+  http_on_open_client_http1(uuid, set_, NULL);
+}
+
+static void http_on_client_failed(intptr_t uuid, void *set_) {
+  http_settings_s *set = set_;
+  http_s *h = set->udata;
+  set->udata = h->udata;
+  http_s_destroy(h, 0);
+  fio_free(h);
+  if (set->on_finish)
+    set->on_finish(set);
+  http_settings_free(set);
+  (void)uuid;
+}
+
+intptr_t http_connect__(void); /* sublime text marker */
+/**
+ * Connects to an HTTP server as a client.
+ *
+ * Upon a successful connection, the `on_response` callback is called with an
+ * empty `http_s*` handler (status == 0). Use the same API to set it's content
+ * and send the request to the server. The next`on_response` will contain the
+ * response.
+ *
+ * `address` should contain a full URL style address for the server. i.e.:
+ *           "http:/www.example.com:8080/"
+ *
+ * Returns -1 on error and 0 on success. the `on_finish` callback is always
+ * called.
+ */
+intptr_t http_connect FIO_IGNORE_MACRO(const char *url,
+                                       const char *unix_address,
+                                       struct http_settings_s arg_settings) {
+  if (!arg_settings.on_response && !arg_settings.on_upgrade) {
+    FIO_LOG_ERROR("http_connect requires either an on_response "
+                  " or an on_upgrade callback.\n");
+    errno = EINVAL;
+    goto on_error;
+  }
+  size_t len = 0, h_len = 0;
+  char *a = NULL, *p = NULL, *host = NULL;
+  uint8_t is_websocket = 0;
+  uint8_t is_secure = 0;
+  FIOBJ path = FIOBJ_INVALID;
+  if (!url && !unix_address) {
+    FIO_LOG_ERROR("http_connect requires a valid address.");
+    errno = EINVAL;
+    goto on_error;
+  }
+  if (url) {
+    fio_url_s u = fio_url_parse(url, strlen(url));
+    if (u.scheme.data &&
+        (u.scheme.len == 2 || (u.scheme.len == 3 && u.scheme.data[2] == 's')) &&
+        u.scheme.data[0] == 'w' && u.scheme.data[1] == 's') {
+      is_websocket = 1;
+      is_secure = (u.scheme.len == 3);
+    } else if (u.scheme.data &&
+               (u.scheme.len == 4 ||
+                (u.scheme.len == 5 && u.scheme.data[4] == 's')) &&
+               u.scheme.data[0] == 'h' && u.scheme.data[1] == 't' &&
+               u.scheme.data[2] == 't' && u.scheme.data[3] == 'p') {
+      is_secure = (u.scheme.len == 5);
+    }
+    if (is_secure && !arg_settings.tls) {
+      FIO_LOG_ERROR("Secure connections (%.*s) require a TLS object.",
+                    (int)u.scheme.len, u.scheme.data);
+      errno = EINVAL;
+      goto on_error;
+    }
+    if (u.path.data) {
+      path = fiobj_str_new(
+          u.path.data, strlen(u.path.data)); /* copy query and target as well */
+    }
+    if (unix_address) {
+      a = (char *)unix_address;
+      h_len = len = strlen(a);
+      host = a;
+    } else {
+      if (!u.host.data) {
+        FIO_LOG_ERROR("http_connect requires a valid address.");
+        errno = EINVAL;
+        goto on_error;
+      }
+      /***** no more error handling, since memory is allocated *****/
+      /* copy address */
+      a = fio_malloc(u.host.len + 1);
+      memcpy(a, u.host.data, u.host.len);
+      a[u.host.len] = 0;
+      len = u.host.len;
+      /* copy port */
+      if (u.port.data) {
+        p = fio_malloc(u.port.len + 1);
+        memcpy(p, u.port.data, u.port.len);
+        p[u.port.len] = 0;
+      } else if (is_secure) {
+        p = fio_malloc(3 + 1);
+        memcpy(p, "443", 3);
+        p[3] = 0;
+      } else {
+        p = fio_malloc(2 + 1);
+        memcpy(p, "80", 2);
+        p[2] = 0;
+      }
+    }
+    if (u.host.data) {
+      host = u.host.data;
+      h_len = u.host.len;
+    }
+  }
+
+  /* set settings */
+  if (!arg_settings.timeout)
+    arg_settings.timeout = 30;
+  http_settings_s *settings = http_settings_new(arg_settings);
+  settings->is_client = 1;
+  // if (settings->tls) {
+  //   fio_tls_alpn_add(settings->tls, "http/1.1", http_on_open_client_http1,
+  //                     NULL, NULL);
+  // }
+
+  if (!arg_settings.ws_timeout)
+    settings->ws_timeout = 0; /* allow server to dictate timeout */
+  if (!arg_settings.timeout)
+    settings->timeout = 0; /* allow server to dictate timeout */
+  http_s *h = fio_malloc(sizeof(*h));
+  FIO_ASSERT(h, "HTTP Client handler allocation failed");
+  http_s_new(h, 0, http1_vtable());
+  h->udata = arg_settings.udata;
+  h->status = 0;
+  h->path = path;
+  settings->udata = h;
+  settings->tls = arg_settings.tls;
+  if (host)
+    http_set_header2(h, (fio_str_info_s){.data = (char *)"host", .len = 4},
+                     (fio_str_info_s){.data = host, .len = h_len});
+  intptr_t ret;
+  if (is_websocket) {
+    /* force HTTP/1.1 */
+    ret = fio_connect(.address = a, .port = p, .on_fail = http_on_client_failed,
+                      .on_connect = http_on_open_client, .udata = settings,
+                      .tls = arg_settings.tls);
+    (void)0;
+  } else {
+    /* Allow for any HTTP version */
+    ret = fio_connect(.address = a, .port = p, .on_fail = http_on_client_failed,
+                      .on_connect = http_on_open_client, .udata = settings,
+                      .tls = arg_settings.tls);
+    (void)0;
+  }
+  if (a != unix_address)
+    fio_free(a);
+  fio_free(p);
+  return ret;
+on_error:
+  if (arg_settings.on_finish)
+    arg_settings.on_finish(&arg_settings);
+  return -1;
+}
+
+/* *****************************************************************************
+HTTP Websocket Connect
+***************************************************************************** */
+
+#undef http_upgrade2ws
+static void on_websocket_http_connected(http_s *h) {
+  websocket_settings_s *s = h->udata;
+  h->udata = http_settings(h)->udata = NULL;
+  if (!h->path) {
+    FIO_LOG_WARNING("(websocket client) path not specified in "
+                    "address, assuming root!");
+    h->path = fiobj_str_new("/", 1);
+  }
+  http_upgrade2ws(h, *s);
+  fio_free(s);
+}
+
+static void on_websocket_http_connection_finished(http_settings_s *settings) {
+  websocket_settings_s *s = settings->udata;
+  if (s) {
+    if (s->on_close)
+      s->on_close(0, s->udata);
+    fio_free(s);
+  }
+}
+
+#undef websocket_connect
+int websocket_connect(const char *address, websocket_settings_s settings) {
+  websocket_settings_s *s = fio_malloc(sizeof(*s));
+  FIO_ASSERT_ALLOC(s);
+  *s = settings;
+  return http_connect(address, NULL, .on_request = on_websocket_http_connected,
+                      .on_response = on_websocket_http_connected,
+                      .on_finish = on_websocket_http_connection_finished,
+                      .udata = s);
+}
+#define websocket_connect(address, ...)                                        \
+  websocket_connect((address), (websocket_settings_s){__VA_ARGS__})
+
+/* *****************************************************************************
+EventSource Support (SSE)
+
+Note:
+
+* `http_sse_subscribe` and `http_sse_unsubscribe` are implemented in the
+  http_internal logical unit.
+
+***************************************************************************** */
+
+static inline void http_sse_copy2str(FIOBJ dest, char *prefix, size_t pre_len,
+                                     fio_str_info_s data) {
+  if (!data.len)
+    return;
+  const char *stop = data.data + data.len;
+  while (data.len) {
+    fiobj_str_write(dest, prefix, pre_len);
+    char *pos = data.data;
+    while (pos < stop && *pos != '\n' && *pos != '\r')
+      ++pos;
+    fiobj_str_write(dest, data.data, (uintptr_t)(pos - data.data));
+    fiobj_str_write(dest, "\r\n", 2);
+    if (*pos == '\r')
+      ++pos;
+    if (*pos == '\n')
+      ++pos;
+    data.len -= (uintptr_t)(pos - data.data);
+    data.data = pos;
+  }
+}
+
+/** The on message callback. the `*msg` pointer is to a temporary object. */
+static void http_sse_on_message(fio_msg_s *msg) {
+  http_sse_internal_s *sse = msg->udata1;
+  struct http_sse_subscribe_args *args = msg->udata2;
+  /* perform a callback */
+  fio_protocol_s *pr = fio_protocol_try_lock(sse->uuid, FIO_PR_LOCK_TASK);
+  if (!pr)
+    goto postpone;
+  args->on_message(&sse->sse, msg->channel, msg->msg, args->udata);
+  fio_protocol_unlock(pr, FIO_PR_LOCK_TASK);
+  return;
+postpone:
+  if (errno == EBADF)
+    return;
+  fio_message_defer(msg);
+  return;
+}
+
+static void http_sse_on_message__direct(http_sse_s *sse, fio_str_info_s channel,
+                                        fio_str_info_s msg, void *udata) {
+  http_sse_write(sse, .data = msg);
+  (void)udata;
+  (void)channel;
+}
+/** An optional callback for when a subscription is fully canceled. */
+static void http_sse_on_unsubscribe(void *sse_, void *args_) {
+  http_sse_internal_s *sse = sse_;
+  struct http_sse_subscribe_args *args = args_;
+  if (args->on_unsubscribe)
+    args->on_unsubscribe(args->udata);
+  fio_free(args);
+  http_sse_try_free(sse);
+}
+
+/** This macro allows easy access to the `http_sse_subscribe` function. */
+#undef http_sse_subscribe
+/**
+ * Subscribes to a channel. See {struct http_sse_subscribe_args} for possible
+ * arguments.
+ *
+ * Returns a subscription ID on success and 0 on failure.
+ *
+ * All subscriptions are automatically revoked once the connection is closed.
+ *
+ * If the connections subscribes to the same channel more than once, messages
+ * will be merged. However, another subscription ID will be assigned, and two
+ * calls to {http_sse_unsubscribe} will be required in order to unregister from
+ * the channel.
+ */
+uintptr_t http_sse_subscribe(http_sse_s *sse_,
+                             struct http_sse_subscribe_args args) {
+  http_sse_internal_s *sse = FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse_);
+  if (sse->uuid == -1)
+    return 0;
+  if (!args.on_message)
+    args.on_message = http_sse_on_message__direct;
+  struct http_sse_subscribe_args *udata = fio_malloc(sizeof(*udata));
+  FIO_ASSERT_ALLOC(udata);
+  *udata = args;
+
+  fio_atomic_add(&sse->ref, 1);
+  subscription_s *sub =
+      fio_subscribe(.channel = args.channel, .on_message = http_sse_on_message,
+                    .on_unsubscribe = http_sse_on_unsubscribe, .udata1 = sse,
+                    .udata2 = udata, .match = args.match);
+  if (!sub)
+    return 0;
+
+  fio_lock(&sse->lock);
+  fio_ls_s *pos = fio_ls_push(&sse->subscriptions, sub);
+  fio_unlock(&sse->lock);
+  return (uintptr_t)pos;
+}
+
+/**
+ * Cancels a subscription and invalidates the subscription object.
+ */
+void http_sse_unsubscribe(http_sse_s *sse_, uintptr_t subscription) {
+  if (!sse_ || !subscription)
+    return;
+  http_sse_internal_s *sse = FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse_);
+  subscription_s *sub = (subscription_s *)((fio_ls_s *)subscription)->obj;
+  fio_lock(&sse->lock);
+  fio_ls_remove((fio_ls_s *)subscription);
+  fio_unlock(&sse->lock);
+  fio_unsubscribe(sub);
+}
+
+#undef http_upgrade2sse
+/**
+ * Upgrades an HTTP connection to an EventSource (SSE) connection.
+ *
+ * Thie `http_s` handle will be invalid after this call.
+ *
+ * On HTTP/1.1 connections, this will preclude future requests using the same
+ * connection.
+ */
+int http_upgrade2sse(http_s *h, http_sse_s sse) {
+  if (HTTP_INVALID_HANDLE(h)) {
+    if (sse.on_close)
+      sse.on_close(&sse);
+    return -1;
+  }
+  return ((http_vtable_s *)h->private_data.vtbl)->http_upgrade2sse(h, &sse);
+}
+
+/**
+ * Sets the ping interval for SSE connections.
+ */
+void http_sse_set_timout(http_sse_s *sse_, uint8_t timeout) {
+  if (!sse_)
+    return;
+  http_sse_internal_s *sse = FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse_);
+  fio_timeout_set(sse->uuid, timeout);
+}
+
+#undef http_sse_write
+/**
+ * Writes data to an EventSource (SSE) connection.
+ */
+int http_sse_write(http_sse_s *sse, struct http_sse_write_args args) {
+  if (!sse || !(args.id.len + args.data.len + args.event.len) ||
+      fio_is_closed(FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse)->uuid))
+    return -1;
+  FIOBJ buf;
+  {
+    /* best guess at data length, ignoring missing fields and multiline data */
+    const size_t total = 4 + args.id.len + 2 + 7 + args.event.len + 2 + 6 +
+                         args.data.len + 2 + 7 + 10 + 4;
+    buf = fiobj_str_buf(total);
+  }
+  http_sse_copy2str(buf, (char *)"id: ", 4, args.id);
+  http_sse_copy2str(buf, (char *)"event: ", 7, args.event);
+  if (args.retry) {
+    FIOBJ i = fiobj_num_new(args.retry);
+    fiobj_str_write(buf, (char *)"retry: ", 7);
+    fiobj_str_join(buf, i);
+    fiobj_free(i);
+  }
+  http_sse_copy2str(buf, (char *)"data: ", 6, args.data);
+  fiobj_str_write(buf, "\r\n", 2);
+  return FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse)
+      ->vtable->http_sse_write(sse, buf);
+}
+
+/**
+ * Get the connection's UUID (for fio_defer and similar use cases).
+ */
+intptr_t http_sse2uuid(http_sse_s *sse) {
+  if (!sse ||
+      fio_is_closed(FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse)->uuid))
+    return -1;
+  return FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse)->uuid;
+}
+
+/**
+ * Closes an EventSource (SSE) connection.
+ */
+int http_sse_close(http_sse_s *sse) {
+  if (!sse ||
+      fio_is_closed(FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse)->uuid))
+    return -1;
+  return FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse)
+      ->vtable->http_sse_close(sse);
+}
+
+/**
+ * Duplicates an SSE handle by reference, remember to http_sse_free.
+ *
+ * Returns the same object (increases a reference count, no allocation is made).
+ */
+http_sse_s *http_sse_dup(http_sse_s *sse) {
+  fio_atomic_add(&FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse)->ref, 1);
+  return sse;
+}
+
+/**
+ * Frees an SSE handle by reference (decreases the reference count).
+ */
+void http_sse_free(http_sse_s *sse) {
+  http_sse_try_free(FIO_LS_EMBD_OBJ(http_sse_internal_s, sse, sse));
+}
+
+/* *****************************************************************************
+HTTP GET and POST parsing helpers
+***************************************************************************** */
+
+/** URL decodes a string, returning a `FIOBJ`. */
+static inline FIOBJ http_urlstr2fiobj(char *s, size_t len) {
+  FIOBJ o = fiobj_str_buf(len);
+  ssize_t l = http_decode_url(fiobj_obj2cstr(o).data, s, len);
+  if (l < 0) {
+    fiobj_free(o);
+    return fiobj_str_new(NULL, 0); /* empty string */
+  }
+  fiobj_str_resize(o, (size_t)l);
+  return o;
+}
+
+/** converts a string into a `FIOBJ`. */
+static inline FIOBJ http_str2fiobj(char *s, size_t len, uint8_t encoded) {
+  switch (len) {
+  case 0:
+    return fiobj_str_new(NULL, 0); /* empty string */
+  case 4:
+    if (!strncasecmp(s, "true", 4))
+      return fiobj_true();
+    if (!strncasecmp(s, "null", 4))
+      return fiobj_null();
+    break;
+  case 5:
+    if (!strncasecmp(s, "false", 5))
+      return fiobj_false();
+  }
+  {
+    char *end = s;
+    const uint64_t tmp = fio_atol(&end);
+    if (end == s + len)
+      return fiobj_num_new(tmp);
+  }
+  {
+    char *end = s;
+    const double tmp = fio_atof(&end);
+    if (end == s + len)
+      return fiobj_float_new(tmp);
+  }
+  if (encoded)
+    return http_urlstr2fiobj(s, len);
+  return fiobj_str_new(s, len);
+}
+
+/** Parses the query part of an HTTP request/response. Uses `http_add2hash`. */
+void http_parse_query(http_s *h) {
+  if (!h->query)
+    return;
+  if (!h->params)
+    h->params = fiobj_hash_new();
+  fio_str_info_s q = fiobj_obj2cstr(h->query);
+  do {
+    char *cut = memchr(q.data, '&', q.len);
+    if (!cut)
+      cut = q.data + q.len;
+    char *cut2 = memchr(q.data, '=', (cut - q.data));
+    if (cut2) {
+      /* we only add named elements... */
+      http_add2hash(h->params, q.data, (size_t)(cut2 - q.data), (cut2 + 1),
+                    (size_t)(cut - (cut2 + 1)), 1);
+    }
+    if (cut[0] == '&') {
+      /* protecting against some ...less informed... clients */
+      if (cut[1] == 'a' && cut[2] == 'm' && cut[3] == 'p' && cut[4] == ';')
+        cut += 5;
+      else
+        cut += 1;
+    }
+    q.len -= (uintptr_t)(cut - q.data);
+    q.data = cut;
+  } while (q.len);
+}
+
+static inline void http_parse_cookies_cookie_str(FIOBJ dest, FIOBJ str,
+                                                 uint8_t is_url_encoded) {
+  if (!FIOBJ_TYPE_IS(str, FIOBJ_T_STRING))
+    return;
+  fio_str_info_s s = fiobj_obj2cstr(str);
+  while (s.len) {
+    if (s.data[0] == ' ') {
+      ++s.data;
+      --s.len;
+      continue;
+    }
+    char *cut = memchr(s.data, '=', s.len);
+    if (!cut)
+      cut = s.data;
+    char *cut2 = memchr(cut, ';', s.len - (cut - s.data));
+    if (!cut2)
+      cut2 = s.data + s.len;
+    http_add2hash(dest, s.data, cut - s.data, cut + 1, (cut2 - (cut + 1)),
+                  is_url_encoded);
+    if ((size_t)((cut2 + 1) - s.data) > s.len)
+      s.len = 0;
+    else
+      s.len -= ((cut2 + 1) - s.data);
+    s.data = cut2 + 1;
+  }
+}
+static inline void http_parse_cookies_setcookie_str(FIOBJ dest, FIOBJ str,
+                                                    uint8_t is_url_encoded) {
+  if (!FIOBJ_TYPE_IS(str, FIOBJ_T_STRING))
+    return;
+  fio_str_info_s s = fiobj_obj2cstr(str);
+  char *cut = memchr(s.data, '=', s.len);
+  if (!cut)
+    cut = s.data;
+  char *cut2 = memchr(cut, ';', s.len - (cut - s.data));
+  if (!cut2)
+    cut2 = s.data + s.len;
+  if (cut2 > cut)
+    http_add2hash(dest, s.data, cut - s.data, cut + 1, (cut2 - (cut + 1)),
+                  is_url_encoded);
+}
+
+/** Parses any Cookie / Set-Cookie headers, using the `http_add2hash` scheme. */
+void http_parse_cookies(http_s *h, uint8_t is_url_encoded) {
+  if (!h->headers)
+    return;
+  if (h->cookies && fiobj_hash_count(h->cookies)) {
+    FIO_LOG_WARNING("(http) attempting to parse cookies more than once.");
+    return;
+  }
+  static uint64_t setcookie_header_hash;
+  if (!setcookie_header_hash)
+    setcookie_header_hash = fiobj_obj2hash(HTTP_HEADER_SET_COOKIE);
+  FIOBJ c = fiobj_hash_get2(h->headers, fiobj_obj2hash(HTTP_HEADER_COOKIE));
+  if (c) {
+    if (!h->cookies)
+      h->cookies = fiobj_hash_new();
+    if (FIOBJ_TYPE_IS(c, FIOBJ_T_ARRAY)) {
+      /* Array of Strings */
+      size_t count = fiobj_ary_count(c);
+      for (size_t i = 0; i < count; ++i) {
+        http_parse_cookies_cookie_str(
+            h->cookies, fiobj_ary_index(c, (int64_t)i), is_url_encoded);
+      }
+    } else {
+      /* single string */
+      http_parse_cookies_cookie_str(h->cookies, c, is_url_encoded);
+    }
+  }
+  c = fiobj_hash_get2(h->headers, fiobj_obj2hash(HTTP_HEADER_SET_COOKIE));
+  if (c) {
+    if (!h->cookies)
+      h->cookies = fiobj_hash_new();
+    if (FIOBJ_TYPE_IS(c, FIOBJ_T_ARRAY)) {
+      /* Array of Strings */
+      size_t count = fiobj_ary_count(c);
+      for (size_t i = 0; i < count; ++i) {
+        http_parse_cookies_setcookie_str(
+            h->cookies, fiobj_ary_index(c, (int64_t)i), is_url_encoded);
+      }
+    } else {
+      /* single string */
+      http_parse_cookies_setcookie_str(h->cookies, c, is_url_encoded);
+    }
+  }
+}
+
+/**
+ * Adds a named parameter to the hash, resolving nesting references.
+ *
+ * i.e.:
+ *
+ * * "name[]" references a nested Array (nested in the Hash).
+ * * "name[key]" references a nested Hash.
+ * * "name[][key]" references a nested Hash within an array. Hash keys will be
+ *   unique (repeating a key advances the hash).
+ * * These rules can be nested (i.e. "name[][key1][][key2]...")
+ * * "name[][]" is an error (there's no way for the parser to analyze
+ *    dimensions)
+ *
+ * Note: names can't begin with "[" or end with "]" as these are reserved
+ *       characters.
+ */
+int http_add2hash2(FIOBJ dest, char *name, size_t name_len, FIOBJ val,
+                   uint8_t encoded) {
+  if (!name)
+    goto error;
+  FIOBJ nested_ary = FIOBJ_INVALID;
+  char *cut1;
+  /* we can't start with an empty object name */
+  while (name_len && name[0] == '[') {
+    --name_len;
+    ++name;
+  }
+  if (!name_len) {
+    /* an empty name is an error */
+    goto error;
+  }
+  uint32_t nesting = ((uint32_t)~0);
+rebase:
+  /* test for nesting level limit (limit at 32) */
+  if (!nesting)
+    goto error;
+  /* start clearing away bits. */
+  nesting >>= 1;
+  /* since we might be rebasing, notice that "name" might be "name]" */
+  cut1 = memchr(name, '[', name_len);
+  if (!cut1)
+    goto place_in_hash;
+  /* simple case "name=" (the "=" was already removed) */
+  if (cut1 == name) {
+    /* an empty name is an error */
+    goto error;
+  }
+  if (cut1 + 1 == name + name_len) {
+    /* we have name[= ... autocorrect */
+    name_len -= 1;
+    goto place_in_array;
+  }
+
+  if (cut1[1] == ']') {
+    /* Nested Array "name[]..." */
+
+    /* Test for name[]= */
+    if ((cut1 + 2) == name + name_len) {
+      name_len -= 2;
+      goto place_in_array;
+    }
+
+    /* Test for a nested Array format error */
+    if (cut1[2] != '[' || cut1[3] == ']') { /* error, we can't parse this */
+      goto error;
+    }
+
+    /* we have name[][key...= */
+
+    /* ensure array exists and it's an array + set nested_ary */
+    const size_t len = ((cut1[-1] == ']') ? (size_t)((cut1 - 1) - name)
+                                          : (size_t)(cut1 - name));
+    const uint64_t hash =
+        fiobj_hash_string(name, len); /* hash the current name */
+    nested_ary = fiobj_hash_get2(dest, hash);
+    if (!nested_ary) {
+      /* create a new nested array */
+      FIOBJ key =
+          encoded ? http_urlstr2fiobj(name, len) : fiobj_str_new(name, len);
+      nested_ary = fiobj_ary_new2(4);
+      fiobj_hash_set(dest, key, nested_ary);
+      fiobj_free(key);
+    } else if (!FIOBJ_TYPE_IS(nested_ary, FIOBJ_T_ARRAY)) {
+      /* convert existing object to an array - auto error correction */
+      FIOBJ key =
+          encoded ? http_urlstr2fiobj(name, len) : fiobj_str_new(name, len);
+      FIOBJ tmp = fiobj_ary_new2(4);
+      fiobj_ary_push(tmp, nested_ary);
+      nested_ary = tmp;
+      fiobj_hash_set(dest, key, nested_ary);
+      fiobj_free(key);
+    }
+
+    /* test if last object in the array is a hash - create hash if not */
+    dest = fiobj_ary_index(nested_ary, -1);
+    if (!dest || !FIOBJ_TYPE_IS(dest, FIOBJ_T_HASH)) {
+      dest = fiobj_hash_new();
+      fiobj_ary_push(nested_ary, dest);
+    }
+
+    /* rebase `name` to `key` and restart. */
+    cut1 += 3; /* consume "[][" */
+    name_len -= (size_t)(cut1 - name);
+    name = cut1;
+    goto rebase;
+
+  } else {
+    /* we have name[key]... */
+    const size_t len = ((cut1[-1] == ']') ? (size_t)((cut1 - 1) - name)
+                                          : (size_t)(cut1 - name));
+    const uint64_t hash =
+        fiobj_hash_string(name, len); /* hash the current name */
+    FIOBJ tmp = fiobj_hash_get2(dest, hash);
+    if (!tmp) {
+      /* hash doesn't exist, create it */
+      FIOBJ key =
+          encoded ? http_urlstr2fiobj(name, len) : fiobj_str_new(name, len);
+      tmp = fiobj_hash_new();
+      fiobj_hash_set(dest, key, tmp);
+      fiobj_free(key);
+    } else if (!FIOBJ_TYPE_IS(tmp, FIOBJ_T_HASH)) {
+      /* type error, referencing an existing object that isn't a Hash */
+      goto error;
+    }
+    dest = tmp;
+    /* no rollback is possible once we enter the new nesting level... */
+    nested_ary = FIOBJ_INVALID;
+    /* rebase `name` to `key` and restart. */
+    cut1 += 1; /* consume "[" */
+    name_len -= (size_t)(cut1 - name);
+    name = cut1;
+    goto rebase;
+  }
+
+place_in_hash:
+  if (name[name_len - 1] == ']')
+    --name_len;
+  {
+    FIOBJ key = encoded ? http_urlstr2fiobj(name, name_len)
+                        : fiobj_str_new(name, name_len);
+    FIOBJ old = fiobj_hash_replace(dest, key, val);
+    if (old) {
+      if (nested_ary) {
+        fiobj_hash_replace(dest, key, old);
+        old = fiobj_hash_new();
+        fiobj_hash_set(old, key, val);
+        fiobj_ary_push(nested_ary, old);
+      } else {
+        if (!FIOBJ_TYPE_IS(old, FIOBJ_T_ARRAY)) {
+          FIOBJ tmp = fiobj_ary_new2(4);
+          fiobj_ary_push(tmp, old);
+          old = tmp;
+        }
+        fiobj_ary_push(old, val);
+        fiobj_hash_replace(dest, key, old);
+      }
+    }
+    fiobj_free(key);
+  }
+  return 0;
+
+place_in_array:
+  if (name[name_len - 1] == ']')
+    --name_len;
+  {
+    uint64_t hash = fiobj_hash_string(name, name_len);
+    FIOBJ ary = fiobj_hash_get2(dest, hash);
+    if (!ary) {
+      FIOBJ key = encoded ? http_urlstr2fiobj(name, name_len)
+                          : fiobj_str_new(name, name_len);
+      ary = fiobj_ary_new2(4);
+      fiobj_hash_set(dest, key, ary);
+      fiobj_free(key);
+    } else if (!FIOBJ_TYPE_IS(ary, FIOBJ_T_ARRAY)) {
+      FIOBJ tmp = fiobj_ary_new2(4);
+      fiobj_ary_push(tmp, ary);
+      ary = tmp;
+      FIOBJ key = encoded ? http_urlstr2fiobj(name, name_len)
+                          : fiobj_str_new(name, name_len);
+      fiobj_hash_replace(dest, key, ary);
+      fiobj_free(key);
+    }
+    fiobj_ary_push(ary, val);
+  }
+  return 0;
+error:
+  fiobj_free(val);
+  errno = EOPNOTSUPP;
+  return -1;
+}
+
+/**
+ * Adds a named parameter to the hash, resolving nesting references.
+ *
+ * i.e.:
+ *
+ * * "name[]" references a nested Array (nested in the Hash).
+ * * "name[key]" references a nested Hash.
+ * * "name[][key]" references a nested Hash within an array. Hash keys will be
+ *   unique (repeating a key advances the hash).
+ * * These rules can be nested (i.e. "name[][key1][][key2]...")
+ * * "name[][]" is an error (there's no way for the parser to analyze
+ *    dimensions)
+ *
+ * Note: names can't begin with "[" or end with "]" as these are reserved
+ *       characters.
+ */
+int http_add2hash(FIOBJ dest, char *name, size_t name_len, char *value,
+                  size_t value_len, uint8_t encoded) {
+  return http_add2hash2(dest, name, name_len,
+                        http_str2fiobj(value, value_len, encoded), encoded);
+}
+
+/* *****************************************************************************
+HTTP Body Parsing
+***************************************************************************** */
+#include <http_mime_parser.h>
+
+typedef struct {
+  http_mime_parser_s p;
+  http_s *h;
+  fio_str_info_s buffer;
+  size_t pos;
+  size_t partial_offset;
+  size_t partial_length;
+  FIOBJ partial_name;
+} http_fio_mime_s;
+
+#define http_mime_parser2fio(parser) ((http_fio_mime_s *)(parser))
+
+/** Called when all the data is available at once. */
+static void http_mime_parser_on_data(http_mime_parser_s *parser, void *name,
+                                     size_t name_len, void *filename,
+                                     size_t filename_len, void *mimetype,
+                                     size_t mimetype_len, void *value,
+                                     size_t value_len) {
+  if (!filename_len) {
+    http_add2hash(http_mime_parser2fio(parser)->h->params, name, name_len,
+                  value, value_len, 0);
+    return;
+  }
+  FIOBJ n = fiobj_str_new(name, name_len);
+  fiobj_str_write(n, "[data]", 6);
+  fio_str_info_s tmp = fiobj_obj2cstr(n);
+  http_add2hash(http_mime_parser2fio(parser)->h->params, tmp.data, tmp.len,
+                value, value_len, 0);
+  fiobj_str_resize(n, name_len);
+  fiobj_str_write(n, "[name]", 6);
+  tmp = fiobj_obj2cstr(n);
+  http_add2hash(http_mime_parser2fio(parser)->h->params, tmp.data, tmp.len,
+                filename, filename_len, 0);
+  if (mimetype_len) {
+    fiobj_str_resize(n, name_len);
+    fiobj_str_write(n, "[type]", 6);
+    tmp = fiobj_obj2cstr(n);
+    http_add2hash(http_mime_parser2fio(parser)->h->params, tmp.data, tmp.len,
+                  mimetype, mimetype_len, 0);
+  }
+  fiobj_free(n);
+}
+
+/** Called when the data didn't fit in the buffer. Data will be streamed. */
+static void http_mime_parser_on_partial_start(
+    http_mime_parser_s *parser, void *name, size_t name_len, void *filename,
+    size_t filename_len, void *mimetype, size_t mimetype_len) {
+  http_mime_parser2fio(parser)->partial_length = 0;
+  http_mime_parser2fio(parser)->partial_offset = 0;
+  http_mime_parser2fio(parser)->partial_name = fiobj_str_new(name, name_len);
+
+  if (!filename)
+    return;
+
+  fiobj_str_write(http_mime_parser2fio(parser)->partial_name, "[type]", 6);
+  fio_str_info_s tmp =
+      fiobj_obj2cstr(http_mime_parser2fio(parser)->partial_name);
+  http_add2hash(http_mime_parser2fio(parser)->h->params, tmp.data, tmp.len,
+                mimetype, mimetype_len, 0);
+
+  fiobj_str_resize(http_mime_parser2fio(parser)->partial_name, name_len);
+  fiobj_str_write(http_mime_parser2fio(parser)->partial_name, "[name]", 6);
+  tmp = fiobj_obj2cstr(http_mime_parser2fio(parser)->partial_name);
+  http_add2hash(http_mime_parser2fio(parser)->h->params, tmp.data, tmp.len,
+                filename, filename_len, 0);
+
+  fiobj_str_resize(http_mime_parser2fio(parser)->partial_name, name_len);
+  fiobj_str_write(http_mime_parser2fio(parser)->partial_name, "[data]", 6);
+}
+
+/** Called when partial data is available. */
+static void http_mime_parser_on_partial_data(http_mime_parser_s *parser,
+                                             void *value, size_t value_len) {
+  if (!http_mime_parser2fio(parser)->partial_offset)
+    http_mime_parser2fio(parser)->partial_offset =
+        http_mime_parser2fio(parser)->pos +
+        ((uintptr_t)value -
+         (uintptr_t)http_mime_parser2fio(parser)->buffer.data);
+  http_mime_parser2fio(parser)->partial_length += value_len;
+  (void)value;
+}
+
+/** Called when the partial data is complete. */
+static void http_mime_parser_on_partial_end(http_mime_parser_s *parser) {
+
+  fio_str_info_s tmp =
+      fiobj_obj2cstr(http_mime_parser2fio(parser)->partial_name);
+  FIOBJ o = FIOBJ_INVALID;
+  if (!http_mime_parser2fio(parser)->partial_length)
+    return;
+  if (http_mime_parser2fio(parser)->partial_length < 42) {
+    /* short data gets a new object */
+    o = fiobj_str_new(http_mime_parser2fio(parser)->buffer.data +
+                          http_mime_parser2fio(parser)->partial_offset,
+                      http_mime_parser2fio(parser)->partial_length);
+  } else {
+    /* longer data gets a reference object (memory collision concerns) */
+    o = fiobj_data_slice(http_mime_parser2fio(parser)->h->body,
+                         http_mime_parser2fio(parser)->partial_offset,
+                         http_mime_parser2fio(parser)->partial_length);
+  }
+  http_add2hash2(http_mime_parser2fio(parser)->h->params, tmp.data, tmp.len, o,
+                 0);
+  fiobj_free(http_mime_parser2fio(parser)->partial_name);
+  http_mime_parser2fio(parser)->partial_name = FIOBJ_INVALID;
+  http_mime_parser2fio(parser)->partial_offset = 0;
+}
+
+/**
+ * Called when URL decoding is required.
+ *
+ * Should support inplace decoding (`dest == encoded`).
+ *
+ * Should return the length of the decoded string.
+ */
+static inline size_t http_mime_decode_url(char *dest, const char *encoded,
+                                          size_t length) {
+  return http_decode_url(dest, encoded, length);
+}
+
+/**
+ * Attempts to decode the request's body.
+ *
+ * Supported Types include:
+ * * application/x-www-form-urlencoded
+ * * application/json
+ * * multipart/form-data
+ */
+int http_parse_body(http_s *h) {
+  static uint64_t content_type_hash;
+  if (!h->body)
+    return -1;
+  if (!content_type_hash)
+    content_type_hash = fiobj_hash_string("content-type", 12);
+  FIOBJ ct = fiobj_hash_get2(h->headers, content_type_hash);
+  fio_str_info_s content_type = fiobj_obj2cstr(ct);
+  if (content_type.len < 16)
+    return -1;
+  if (content_type.len >= 33 &&
+      !strncasecmp("application/x-www-form-urlencoded", content_type.data,
+                   33)) {
+    if (!h->params)
+      h->params = fiobj_hash_new();
+    FIOBJ tmp = h->query;
+    h->query = h->body;
+    http_parse_query(h);
+    h->query = tmp;
+    return 0;
+  }
+  if (content_type.len >= 16 &&
+      !strncasecmp("application/json", content_type.data, 16)) {
+    content_type = fiobj_obj2cstr(h->body);
+    if (h->params)
+      return -1;
+    if (fiobj_json2obj(&h->params, content_type.data, content_type.len) == 0)
+      return -1;
+    if (FIOBJ_TYPE_IS(h->params, FIOBJ_T_HASH))
+      return 0;
+    FIOBJ tmp = h->params;
+    FIOBJ key = fiobj_str_new("JSON", 4);
+    h->params = fiobj_hash_new2(4);
+    fiobj_hash_set(h->params, key, tmp);
+    fiobj_free(key);
+    return 0;
+  }
+
+  http_fio_mime_s p = {.h = h};
+  if (http_mime_parser_init(&p.p, content_type.data, content_type.len))
+    return -1;
+  if (!h->params)
+    h->params = fiobj_hash_new();
+
+  do {
+    size_t cons = http_mime_parse(&p.p, p.buffer.data, p.buffer.len);
+    p.pos += cons;
+    p.buffer = fiobj_data_pread(h->body, p.pos, 4096);
+  } while (p.buffer.data && !p.p.done && !p.p.error);
+  fiobj_free(p.partial_name);
+  p.partial_name = FIOBJ_INVALID;
+  return 0;
+}
+
+/* *****************************************************************************
+HTTP Helper functions that could be used globally
+***************************************************************************** */
+
+/**
+ * Returns a String object representing the unparsed HTTP request (HTTP
+ * version is capped at HTTP/1.1). Mostly usable for proxy usage and
+ * debugging.
+ */
+FIOBJ http_req2str(http_s *h) {
+  if (HTTP_INVALID_HANDLE(h) || !fiobj_hash_count(h->headers))
+    return FIOBJ_INVALID;
+
+  struct header_writer_s w;
+  w.dest = fiobj_str_buf(0);
+  if (h->status_str) {
+    fiobj_str_join(w.dest, h->version);
+    fiobj_str_write(w.dest, " ", 1);
+    fiobj_str_join(w.dest, fiobj_num_tmp(h->status));
+    fiobj_str_write(w.dest, " ", 1);
+    fiobj_str_join(w.dest, h->status_str);
+    fiobj_str_write(w.dest, "\r\n", 2);
+  } else {
+    fiobj_str_join(w.dest, h->method);
+    fiobj_str_write(w.dest, " ", 1);
+    fiobj_str_join(w.dest, h->path);
+    if (h->query) {
+      fiobj_str_write(w.dest, "?", 1);
+      fiobj_str_join(w.dest, h->query);
+    }
+    {
+      fio_str_info_s t = fiobj_obj2cstr(h->version);
+      if (t.len < 6 || t.data[5] != '1')
+        fiobj_str_write(w.dest, " HTTP/1.1\r\n", 10);
+      else {
+        fiobj_str_write(w.dest, " ", 1);
+        fiobj_str_join(w.dest, h->version);
+        fiobj_str_write(w.dest, "\r\n", 2);
+      }
+    }
+  }
+
+  fiobj_each1(h->headers, 0, write_header, &w);
+  fiobj_str_write(w.dest, "\r\n", 2);
+  if (h->body) {
+    // fiobj_data_seek(h->body, 0);
+    // fio_str_info_s t = fiobj_data_read(h->body, 0);
+    // fiobj_str_write(w.dest, t.data, t.len);
+    fiobj_str_join(w.dest, h->body);
+  }
+  return w.dest;
+}
+
+void http_write_log(http_s *h) {
+  FIOBJ l = fiobj_str_buf(128);
+
+  intptr_t bytes_sent = fiobj_obj2num(fiobj_hash_get2(
+      h->private_data.out_headers, fiobj_obj2hash(HTTP_HEADER_CONTENT_LENGTH)));
+
+  struct timespec start, end;
+  clock_gettime(CLOCK_REALTIME, &end);
+  start = h->received_at;
+
+  {
+    // TODO Guess IP address from headers (forwarded) where possible
+    fio_str_info_s peer = fio_peer_addr(http2protocol(h)->uuid);
+    fiobj_str_write(l, peer.data, peer.len);
+  }
+  fio_str_info_s buff = fiobj_obj2cstr(l);
+
+  if (buff.len == 0) {
+    memcpy(buff.data, "[unknown]", 9);
+    buff.len = 9;
+  }
+  memcpy(buff.data + buff.len, " - - [", 6);
+  buff.len += 6;
+  fiobj_str_resize(l, buff.len);
+  {
+    FIOBJ date;
+    fio_lock(&date_lock);
+    date = fiobj_dup(current_date);
+    fio_unlock(&date_lock);
+    fiobj_str_join(l, current_date);
+    fiobj_free(date);
+  }
+  fiobj_str_write(l, "] \"", 3);
+  fiobj_str_join(l, h->method);
+  fiobj_str_write(l, " ", 1);
+  fiobj_str_join(l, h->path);
+  fiobj_str_write(l, " ", 1);
+  fiobj_str_join(l, h->version);
+  fiobj_str_write(l, "\" ", 2);
+  if (bytes_sent > 0) {
+    fiobj_str_write_i(l, h->status);
+    fiobj_str_write(l, " ", 1);
+    fiobj_str_write_i(l, bytes_sent);
+    fiobj_str_write(l, "b ", 2);
+  } else {
+    fiobj_str_join(l, fiobj_num_tmp(h->status));
+    fiobj_str_write(l, " -- ", 4);
+  }
+
+  bytes_sent = ((end.tv_sec - start.tv_sec) * 1000) +
+               ((end.tv_nsec - start.tv_nsec) / 1000000);
+  fiobj_str_write_i(l, bytes_sent);
+  fiobj_str_write(l, "ms\r\n", 4);
+
+  buff = fiobj_obj2cstr(l);
+  fwrite(buff.data, 1, buff.len, stderr);
+  fiobj_free(l);
+}
+
+/**
+A faster (yet less localized) alternative to `gmtime_r`.
+
+See the libc `gmtime_r` documentation for details.
+
+Falls back to `gmtime_r` for dates before epoch.
+*/
+struct tm *http_gmtime(time_t timer, struct tm *tm) {
+  ssize_t a, b;
+#if HAVE_TM_TM_ZONE || defined(BSD)
+  *tm = (struct tm){
+      .tm_isdst = 0,
+      .tm_zone = (char *)"UTC",
+  };
+#else
+  *tm = (struct tm){
+      .tm_isdst = 0,
+  };
+#endif
+
+  // convert seconds from epoch to days from epoch + extract data
+  if (timer >= 0) {
+    // for seconds up to weekdays, we reduce the reminder every step.
+    a = (ssize_t)timer;
+    b = a / 60; // b == time in minutes
+    tm->tm_sec = a - (b * 60);
+    a = b / 60; // b == time in hours
+    tm->tm_min = b - (a * 60);
+    b = a / 24; // b == time in days since epoch
+    tm->tm_hour = a - (b * 24);
+    // b == number of days since epoch
+    // day of epoch was a thursday. Add + 4 so sunday == 0...
+    tm->tm_wday = (b + 4) % 7;
+  } else {
+    // for seconds up to weekdays, we reduce the reminder every step.
+    a = (ssize_t)timer;
+    b = a / 60; // b == time in minutes
+    if (b * 60 != a) {
+      /* seconds passed */
+      tm->tm_sec = (a - (b * 60)) + 60;
+      --b;
+    } else {
+      /* no seconds */
+      tm->tm_sec = 0;
+    }
+    a = b / 60; // b == time in hours
+    if (a * 60 != b) {
+      /* minutes passed */
+      tm->tm_min = (b - (a * 60)) + 60;
+      --a;
+    } else {
+      /* no minutes */
+      tm->tm_min = 0;
+    }
+    b = a / 24; // b == time in days since epoch?
+    if (b * 24 != a) {
+      /* hours passed */
+      tm->tm_hour = (a - (b * 24)) + 24;
+      --b;
+    } else {
+      /* no hours */
+      tm->tm_hour = 0;
+    }
+    // day of epoch was a thursday. Add + 4 so sunday == 0...
+    tm->tm_wday = ((b - 3) % 7);
+    if (tm->tm_wday)
+      tm->tm_wday += 7;
+    /* b == days from epoch */
+  }
+
+  // at this point we can apply the algorithm described here:
+  // http://howardhinnant.github.io/date_algorithms.html#civil_from_days
+  // Credit to Howard Hinnant.
+  {
+    b += 719468L; // adjust to March 1st, 2000 (post leap of 400 year era)
+    // 146,097 = days in era (400 years)
+    const size_t era = (b >= 0 ? b : b - 146096) / 146097;
+    const uint32_t doe = (b - (era * 146097)); // day of era
+    const uint16_t yoe =
+        (doe - doe / 1460 + doe / 36524 - doe / 146096) / 365; // year of era
+    a = yoe;
+    a += era * 400; // a == year number, assuming year starts on March 1st...
+    const uint16_t doy = doe - (365 * yoe + yoe / 4 - yoe / 100);
+    const uint16_t mp = (5U * doy + 2) / 153;
+    const uint16_t d = doy - (153U * mp + 2) / 5 + 1;
+    const uint8_t m = mp + (mp < 10 ? 2 : -10);
+    a += (m <= 1);
+    tm->tm_year = a - 1900; // tm_year == years since 1900
+    tm->tm_mon = m;
+    tm->tm_mday = d;
+    const uint8_t is_leap = (a % 4 == 0 && (a % 100 != 0 || a % 400 == 0));
+    tm->tm_yday = (doy + (is_leap) + 28 + 31) % (365 + is_leap);
+  }
+
+  return tm;
+}
+
+static const char *DAY_NAMES[] = {"Sun", "Mon", "Tue", "Wed",
+                                  "Thu", "Fri", "Sat"};
+static const char *MONTH_NAMES[] = {"Jan ", "Feb ", "Mar ", "Apr ",
+                                    "May ", "Jun ", "Jul ", "Aug ",
+                                    "Sep ", "Oct ", "Nov ", "Dec "};
+static const char *GMT_STR = "GMT";
+
+size_t http_date2rfc7231(char *target, struct tm *tmbuf) {
+  /* note: day of month is always 2 digits */
+  char *pos = target;
+  uint16_t tmp;
+  pos[0] = DAY_NAMES[tmbuf->tm_wday][0];
+  pos[1] = DAY_NAMES[tmbuf->tm_wday][1];
+  pos[2] = DAY_NAMES[tmbuf->tm_wday][2];
+  pos[3] = ',';
+  pos[4] = ' ';
+  pos += 5;
+  tmp = tmbuf->tm_mday / 10;
+  pos[0] = '0' + tmp;
+  pos[1] = '0' + (tmbuf->tm_mday - (tmp * 10));
+  pos += 2;
+  *(pos++) = ' ';
+  pos[0] = MONTH_NAMES[tmbuf->tm_mon][0];
+  pos[1] = MONTH_NAMES[tmbuf->tm_mon][1];
+  pos[2] = MONTH_NAMES[tmbuf->tm_mon][2];
+  pos[3] = ' ';
+  pos += 4;
+  // write year.
+  pos += fio_ltoa(pos, tmbuf->tm_year + 1900, 10);
+  *(pos++) = ' ';
+  tmp = tmbuf->tm_hour / 10;
+  pos[0] = '0' + tmp;
+  pos[1] = '0' + (tmbuf->tm_hour - (tmp * 10));
+  pos[2] = ':';
+  tmp = tmbuf->tm_min / 10;
+  pos[3] = '0' + tmp;
+  pos[4] = '0' + (tmbuf->tm_min - (tmp * 10));
+  pos[5] = ':';
+  tmp = tmbuf->tm_sec / 10;
+  pos[6] = '0' + tmp;
+  pos[7] = '0' + (tmbuf->tm_sec - (tmp * 10));
+  pos += 8;
+  pos[0] = ' ';
+  pos[1] = GMT_STR[0];
+  pos[2] = GMT_STR[1];
+  pos[3] = GMT_STR[2];
+  pos[4] = 0;
+  pos += 4;
+  return pos - target;
+}
+
+size_t http_date2str(char *target, struct tm *tmbuf);
+
+size_t http_date2rfc2822(char *target, struct tm *tmbuf) {
+  /* note: day of month is either 1 or 2 digits */
+  char *pos = target;
+  uint16_t tmp;
+  pos[0] = DAY_NAMES[tmbuf->tm_wday][0];
+  pos[1] = DAY_NAMES[tmbuf->tm_wday][1];
+  pos[2] = DAY_NAMES[tmbuf->tm_wday][2];
+  pos[3] = ',';
+  pos[4] = ' ';
+  pos += 5;
+  if (tmbuf->tm_mday < 10) {
+    *pos = '0' + tmbuf->tm_mday;
+    ++pos;
+  } else {
+    tmp = tmbuf->tm_mday / 10;
+    pos[0] = '0' + tmp;
+    pos[1] = '0' + (tmbuf->tm_mday - (tmp * 10));
+    pos += 2;
+  }
+  *(pos++) = '-';
+  pos[0] = MONTH_NAMES[tmbuf->tm_mon][0];
+  pos[1] = MONTH_NAMES[tmbuf->tm_mon][1];
+  pos[2] = MONTH_NAMES[tmbuf->tm_mon][2];
+  pos += 3;
+  *(pos++) = '-';
+  // write year.
+  pos += fio_ltoa(pos, tmbuf->tm_year + 1900, 10);
+  *(pos++) = ' ';
+  tmp = tmbuf->tm_hour / 10;
+  pos[0] = '0' + tmp;
+  pos[1] = '0' + (tmbuf->tm_hour - (tmp * 10));
+  pos[2] = ':';
+  tmp = tmbuf->tm_min / 10;
+  pos[3] = '0' + tmp;
+  pos[4] = '0' + (tmbuf->tm_min - (tmp * 10));
+  pos[5] = ':';
+  tmp = tmbuf->tm_sec / 10;
+  pos[6] = '0' + tmp;
+  pos[7] = '0' + (tmbuf->tm_sec - (tmp * 10));
+  pos += 8;
+  pos[0] = ' ';
+  pos[1] = GMT_STR[0];
+  pos[2] = GMT_STR[1];
+  pos[3] = GMT_STR[2];
+  pos[4] = 0;
+  pos += 4;
+  return pos - target;
+}
+
+/* HTTP header format for Cookie ages */
+size_t http_date2rfc2109(char *target, struct tm *tmbuf) {
+  /* note: day of month is always 2 digits */
+  char *pos = target;
+  uint16_t tmp;
+  pos[0] = DAY_NAMES[tmbuf->tm_wday][0];
+  pos[1] = DAY_NAMES[tmbuf->tm_wday][1];
+  pos[2] = DAY_NAMES[tmbuf->tm_wday][2];
+  pos[3] = ',';
+  pos[4] = ' ';
+  pos += 5;
+  tmp = tmbuf->tm_mday / 10;
+  pos[0] = '0' + tmp;
+  pos[1] = '0' + (tmbuf->tm_mday - (tmp * 10));
+  pos += 2;
+  *(pos++) = ' ';
+  pos[0] = MONTH_NAMES[tmbuf->tm_mon][0];
+  pos[1] = MONTH_NAMES[tmbuf->tm_mon][1];
+  pos[2] = MONTH_NAMES[tmbuf->tm_mon][2];
+  pos[3] = ' ';
+  pos += 4;
+  // write year.
+  pos += fio_ltoa(pos, tmbuf->tm_year + 1900, 10);
+  *(pos++) = ' ';
+  tmp = tmbuf->tm_hour / 10;
+  pos[0] = '0' + tmp;
+  pos[1] = '0' + (tmbuf->tm_hour - (tmp * 10));
+  pos[2] = ':';
+  tmp = tmbuf->tm_min / 10;
+  pos[3] = '0' + tmp;
+  pos[4] = '0' + (tmbuf->tm_min - (tmp * 10));
+  pos[5] = ':';
+  tmp = tmbuf->tm_sec / 10;
+  pos[6] = '0' + tmp;
+  pos[7] = '0' + (tmbuf->tm_sec - (tmp * 10));
+  pos += 8;
+  *pos++ = ' ';
+  *pos++ = '-';
+  *pos++ = '0';
+  *pos++ = '0';
+  *pos++ = '0';
+  *pos++ = '0';
+  *pos = 0;
+  return pos - target;
+}
+
+static pthread_key_t cached_tick_key;
+static pthread_key_t cached_httpdate_key;
+static pthread_key_t cached_len_key;
+static pthread_once_t cached_once = PTHREAD_ONCE_INIT;
+static void init_cached_key(void) {
+  pthread_key_create(&cached_tick_key, free);
+  pthread_key_create(&cached_httpdate_key, free);
+  pthread_key_create(&cached_len_key, free);
+}
+static void init_cached_key_ptr(void) {
+  time_t *cached_tick = malloc(sizeof(time_t));
+  FIO_ASSERT_ALLOC(cached_tick);
+  memset(cached_tick, 0, sizeof(time_t));
+  char *cached_httpdate = malloc(sizeof(char) * 48);
+  FIO_ASSERT_ALLOC(cached_tick);
+  memset(cached_httpdate, 0, 48);
+  size_t *cached_len = malloc(sizeof(size_t));
+  *cached_len = 0;
+  FIO_ASSERT_ALLOC(cached_len);
+  pthread_setspecific(cached_tick_key, cached_tick);
+  pthread_setspecific(cached_httpdate_key, cached_httpdate);
+  pthread_setspecific(cached_len_key, cached_len);
+}
+
+/**
+ * Prints Unix time to a HTTP time formatted string.
+ *
+ * This variation implements cached results for faster processing, at the
+ * price of a less accurate string.
+ */
+size_t http_time2str(char *target, const time_t t) {
+  /* pre-print time every 1 or 2 seconds or so. */
+  pthread_once(&cached_once, init_cached_key);
+  char *cached_httpdate = pthread_getspecific(cached_httpdate_key);
+  if (!cached_httpdate) {
+    init_cached_key_ptr();
+    cached_httpdate = pthread_getspecific(cached_httpdate_key);
+  }
+  time_t *cached_tick = pthread_getspecific(cached_tick_key);
+  size_t *cached_len = pthread_getspecific(cached_len_key);
+  time_t last_tick = fio_last_tick().tv_sec;
+  if ((t | 7) < last_tick) {
+    /* this is a custom time, not "now", pass through */
+    struct tm tm;
+    http_gmtime(t, &tm);
+    return http_date2str(target, &tm);
+  }
+  if (last_tick > *cached_tick) {
+    struct tm tm;
+    *cached_tick = last_tick; /* refresh every second */
+    http_gmtime(last_tick, &tm);
+    *cached_len = http_date2str(cached_httpdate, &tm);
+  }
+  memcpy(target, cached_httpdate, *cached_len);
+  return *cached_len;
+}
+
+/* Credit to Jonathan Leffler for the idea of a unified conditional */
+#define hex_val(c)                                                             \
+  (((c) >= '0' && (c) <= '9')                 ? ((c)-48)                       \
+   : (((c) | 32) >= 'a' && ((c) | 32) <= 'f') ? (((c) | 32) - 87)              \
+                                              : ({                             \
+                                                  return -1;                   \
+                                                  0;                           \
+                                                }))
+static inline int hex2byte(uint8_t *dest, const uint8_t *source) {
+  if (source[0] >= '0' && source[0] <= '9')
+    *dest = (source[0] - '0');
+  else if ((source[0] | 32) >= 'a' && (source[0] | 32) <= 'f')
+    *dest = (source[0] | 32) - ('a' - 10);
+  else
+    return -1;
+  *dest <<= 4;
+  if (source[1] >= '0' && source[1] <= '9')
+    *dest |= (source[1] - '0');
+  else if ((source[1] | 32) >= 'a' && (source[1] | 32) <= 'f')
+    *dest |= (source[1] | 32) - ('a' - 10);
+  else
+    return -1;
+  return 0;
+}
+
+ssize_t http_decode_url(char *dest, const char *url_data, size_t length) {
+  char *pos = dest;
+  const char *end = url_data + length;
+  while (url_data < end) {
+    if (*url_data == '+') {
+      // decode space
+      *(pos++) = ' ';
+      ++url_data;
+    } else if (*url_data == '%') {
+      // decode hex value
+      // this is a percent encoded value.
+      if (hex2byte((uint8_t *)pos, (uint8_t *)&url_data[1]))
+        return -1;
+      pos++;
+      url_data += 3;
+    } else
+      *(pos++) = *(url_data++);
+  }
+  *pos = 0;
+  return pos - dest;
+}
+
+ssize_t http_decode_url_unsafe(char *dest, const char *url_data) {
+  char *pos = dest;
+  while (*url_data) {
+    if (*url_data == '+') {
+      // decode space
+      *(pos++) = ' ';
+      ++url_data;
+    } else if (*url_data == '%') {
+      // decode hex value
+      // this is a percent encoded value.
+      if (hex2byte((uint8_t *)pos, (uint8_t *)&url_data[1]))
+        return -1;
+      pos++;
+      url_data += 3;
+    } else
+      *(pos++) = *(url_data++);
+  }
+  *pos = 0;
+  return pos - dest;
+}
+
+ssize_t http_decode_path(char *dest, const char *url_data, size_t length) {
+  char *pos = dest;
+  const char *end = url_data + length;
+  while (url_data < end) {
+    if (*url_data == '%') {
+      // decode hex value
+      // this is a percent encoded value.
+      if (hex2byte((uint8_t *)pos, (uint8_t *)&url_data[1]))
+        return -1;
+      pos++;
+      url_data += 3;
+    } else
+      *(pos++) = *(url_data++);
+  }
+  *pos = 0;
+  return pos - dest;
+}
+
+ssize_t http_decode_path_unsafe(char *dest, const char *url_data) {
+  char *pos = dest;
+  while (*url_data) {
+    if (*url_data == '%') {
+      // decode hex value
+      // this is a percent encoded value.
+      if (hex2byte((uint8_t *)pos, (uint8_t *)&url_data[1]))
+        return -1;
+      pos++;
+      url_data += 3;
+    } else
+      *(pos++) = *(url_data++);
+  }
+  *pos = 0;
+  return pos - dest;
+}
+
+/* *****************************************************************************
+Lookup Tables / functions
+***************************************************************************** */
+
+#define FIO_FORCE_MALLOC_TMP 1 /* use malloc for the mime registry */
+#define FIO_SET_NAME fio_mime_set
+#define FIO_SET_OBJ_TYPE FIOBJ
+#define FIO_SET_OBJ_COMPARE(o1, o2) (1)
+#define FIO_SET_OBJ_COPY(dest, o) (dest) = fiobj_dup((o))
+#define FIO_SET_OBJ_DESTROY(o) fiobj_free((o))
+
+#include <fio.h>
+
+static fio_mime_set_s fio_http_mime_types = FIO_SET_INIT;
+
+#define LONGEST_FILE_EXTENSION_LENGTH 15
+
+/** Registers a Mime-Type to be associated with the file extension. */
+void http_mimetype_register(char *file_ext, size_t file_ext_len,
+                            FIOBJ mime_type_str) {
+  uintptr_t hash = FIO_HASH_FN(file_ext, file_ext_len, 0, 0);
+  if (mime_type_str == FIOBJ_INVALID) {
+    fio_mime_set_remove(&fio_http_mime_types, hash, FIOBJ_INVALID, NULL);
+  } else {
+    FIOBJ old = FIOBJ_INVALID;
+    fio_mime_set_overwrite(&fio_http_mime_types, hash, mime_type_str, &old);
+    if (old != FIOBJ_INVALID) {
+      FIO_LOG_WARNING("mime-type collision: %.*s was %s, now %s",
+                      (int)file_ext_len, file_ext, fiobj_obj2cstr(old).data,
+                      fiobj_obj2cstr(mime_type_str).data);
+      fiobj_free(old);
+    }
+    fiobj_free(mime_type_str); /* move ownership to the registry */
+  }
+}
+
+/** Registers a Mime-Type to be associated with the file extension. */
+void http_mimetype_stats(void) {
+  FIO_LOG_DEBUG("HTTP MIME hash storage count/capa: %zu / %zu",
+                fio_mime_set_count(&fio_http_mime_types),
+                fio_mime_set_capa(&fio_http_mime_types));
+}
+
+/**
+ * Finds the mime-type associated with the file extension.
+ *  Remember to call `fiobj_free`.
+ */
+FIOBJ http_mimetype_find(char *file_ext, size_t file_ext_len) {
+  uintptr_t hash = FIO_HASH_FN(file_ext, file_ext_len, 0, 0);
+  return fiobj_dup(
+      fio_mime_set_find(&fio_http_mime_types, hash, FIOBJ_INVALID));
+}
+
+static pthread_key_t buffer_key;
+static pthread_once_t buffer_once = PTHREAD_ONCE_INIT;
+static void init_buffer_key(void) { pthread_key_create(&buffer_key, free); }
+static void init_buffer_ptr(void) {
+  char *buffer = malloc(sizeof(char) * (LONGEST_FILE_EXTENSION_LENGTH + 1));
+  FIO_ASSERT_ALLOC(buffer);
+  memset(buffer, 0, sizeof(char) * (LONGEST_FILE_EXTENSION_LENGTH + 1));
+  pthread_setspecific(buffer_key, buffer);
+}
+/**
+ * Finds the mime-type associated with the URL.
+ *  Remember to call `fiobj_free`.
+ */
+FIOBJ http_mimetype_find2(FIOBJ url) {
+  pthread_once(&buffer_once, init_buffer_key);
+  char *buffer = pthread_getspecific(buffer_key);
+  if (!buffer) {
+    init_buffer_ptr();
+    buffer = pthread_getspecific(buffer_key);
+  }
+  fio_str_info_s ext = {.data = NULL};
+  FIOBJ mimetype;
+  if (!url)
+    goto finish;
+  fio_str_info_s tmp = fiobj_obj2cstr(url);
+  uint8_t steps = 1;
+  while (tmp.len > steps || steps >= LONGEST_FILE_EXTENSION_LENGTH) {
+    switch (tmp.data[tmp.len - steps]) {
+    case '.':
+      --steps;
+      if (steps) {
+        ext.len = steps;
+        ext.data = buffer;
+        buffer[steps] = 0;
+        for (size_t i = 1; i <= steps; ++i) {
+          buffer[steps - i] = tolower(tmp.data[tmp.len - i]);
+        }
+      }
+    /* fallthrough */
+    case '/':
+      goto finish;
+      break;
+    }
+    ++steps;
+  }
+finish:
+  mimetype = http_mimetype_find(ext.data, ext.len);
+  if (!mimetype)
+    mimetype = fiobj_dup(HTTP_HVALUE_CONTENT_TYPE_DEFAULT);
+  return mimetype;
+}
+
+/** Clears the Mime-Type registry (it will be empty afterthis call). */
+void http_mimetype_clear(void) {
+  fio_mime_set_free(&fio_http_mime_types);
+  fiobj_free(current_date);
+  current_date = FIOBJ_INVALID;
+  last_date_added = 0;
+}
+
+/**
+* Create with Ruby using:
+
+a = []
+256.times {|i| a[i] = 1;}
+('a'.ord..'z'.ord).each {|i| a[i] = 0;}
+('A'.ord..'Z'.ord).each {|i| a[i] = 0;}
+('0'.ord..'9'.ord).each {|i| a[i] = 0;}
+"!#$%&'*+-.^_`|~".bytes.each {|i| a[i] = 0;}
+p a; nil
+"!#$%&'()*+-./:<=>?@[]^_`{|}~".bytes.each {|i| a[i] = 0;} # for values
+p a; nil
+*/
+static char invalid_cookie_name_char[256] = {
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 1,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+
+static char invalid_cookie_value_char[256] = {
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+    0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1};
+
+// clang-format off
+#define HTTP_SET_STATUS_STR(status, str) [status-100] = { .data = (char *)(#status " " str), .len = (sizeof( #status " " str) - 1) }
+// clang-format on
+
+/** Returns the status as a C string struct */
+fio_str_info_s http_status2str(uintptr_t status) {
+  static const fio_str_info_s status2str[] = {
+      HTTP_SET_STATUS_STR(100, "Continue"),
+      HTTP_SET_STATUS_STR(101, "Switching Protocols"),
+      HTTP_SET_STATUS_STR(102, "Processing"),
+      HTTP_SET_STATUS_STR(103, "Early Hints"),
+      HTTP_SET_STATUS_STR(200, "OK"),
+      HTTP_SET_STATUS_STR(201, "Created"),
+      HTTP_SET_STATUS_STR(202, "Accepted"),
+      HTTP_SET_STATUS_STR(203, "Non-Authoritative Information"),
+      HTTP_SET_STATUS_STR(204, "No Content"),
+      HTTP_SET_STATUS_STR(205, "Reset Content"),
+      HTTP_SET_STATUS_STR(206, "Partial Content"),
+      HTTP_SET_STATUS_STR(207, "Multi-Status"),
+      HTTP_SET_STATUS_STR(208, "Already Reported"),
+      HTTP_SET_STATUS_STR(226, "IM Used"),
+      HTTP_SET_STATUS_STR(300, "Multiple Choices"),
+      HTTP_SET_STATUS_STR(301, "Moved Permanently"),
+      HTTP_SET_STATUS_STR(302, "Found"),
+      HTTP_SET_STATUS_STR(303, "See Other"),
+      HTTP_SET_STATUS_STR(304, "Not Modified"),
+      HTTP_SET_STATUS_STR(305, "Use Proxy"),
+      HTTP_SET_STATUS_STR(306, "(Unused), "),
+      HTTP_SET_STATUS_STR(307, "Temporary Redirect"),
+      HTTP_SET_STATUS_STR(308, "Permanent Redirect"),
+      HTTP_SET_STATUS_STR(400, "Bad Request"),
+      HTTP_SET_STATUS_STR(403, "Forbidden"),
+      HTTP_SET_STATUS_STR(404, "Not Found"),
+      HTTP_SET_STATUS_STR(401, "Unauthorized"),
+      HTTP_SET_STATUS_STR(402, "Payment Required"),
+      HTTP_SET_STATUS_STR(405, "Method Not Allowed"),
+      HTTP_SET_STATUS_STR(406, "Not Acceptable"),
+      HTTP_SET_STATUS_STR(407, "Proxy Authentication Required"),
+      HTTP_SET_STATUS_STR(408, "Request Timeout"),
+      HTTP_SET_STATUS_STR(409, "Conflict"),
+      HTTP_SET_STATUS_STR(410, "Gone"),
+      HTTP_SET_STATUS_STR(411, "Length Required"),
+      HTTP_SET_STATUS_STR(412, "Precondition Failed"),
+      HTTP_SET_STATUS_STR(413, "Payload Too Large"),
+      HTTP_SET_STATUS_STR(414, "URI Too Long"),
+      HTTP_SET_STATUS_STR(415, "Unsupported Media Type"),
+      HTTP_SET_STATUS_STR(416, "Range Not Satisfiable"),
+      HTTP_SET_STATUS_STR(417, "Expectation Failed"),
+      HTTP_SET_STATUS_STR(418, "I'm a teapot"),
+      HTTP_SET_STATUS_STR(421, "Misdirected Request"),
+      HTTP_SET_STATUS_STR(422, "Unprocessable Entity"),
+      HTTP_SET_STATUS_STR(423, "Locked"),
+      HTTP_SET_STATUS_STR(424, "Failed Dependency"),
+      HTTP_SET_STATUS_STR(425, "Unassigned"),
+      HTTP_SET_STATUS_STR(426, "Upgrade Required"),
+      HTTP_SET_STATUS_STR(427, "Unassigned"),
+      HTTP_SET_STATUS_STR(428, "Precondition Required"),
+      HTTP_SET_STATUS_STR(429, "Too Many Requests"),
+      HTTP_SET_STATUS_STR(430, "Unassigned"),
+      HTTP_SET_STATUS_STR(431, "Request Header Fields Too Large"),
+      HTTP_SET_STATUS_STR(500, "Internal Server Error"),
+      HTTP_SET_STATUS_STR(501, "Not Implemented"),
+      HTTP_SET_STATUS_STR(502, "Bad Gateway"),
+      HTTP_SET_STATUS_STR(503, "Service Unavailable"),
+      HTTP_SET_STATUS_STR(504, "Gateway Timeout"),
+      HTTP_SET_STATUS_STR(505, "HTTP Version Not Supported"),
+      HTTP_SET_STATUS_STR(506, "Variant Also Negotiates"),
+      HTTP_SET_STATUS_STR(507, "Insufficient Storage"),
+      HTTP_SET_STATUS_STR(508, "Loop Detected"),
+      HTTP_SET_STATUS_STR(509, "Unassigned"),
+      HTTP_SET_STATUS_STR(510, "Not Extended"),
+      HTTP_SET_STATUS_STR(511, "Network Authentication Required"),
+  };
+  fio_str_info_s ret = (fio_str_info_s){.len = 0, .data = NULL};
+  if (status >= 100 &&
+      (status - 100) < sizeof(status2str) / sizeof(status2str[0]))
+    ret = status2str[status - 100];
+  if (!ret.data) {
+    ret = status2str[400];
+  }
+  return ret;
+}
+#undef HTTP_SET_STATUS_STR
+
+#if DEBUG
+void http_tests(void) {
+  fprintf(stderr, "=== Testing HTTP helpers\n");
+  FIOBJ html_mime = http_mimetype_find("html", 4);
+  FIO_ASSERT(html_mime,
+             "HTML mime-type not found! Mime-Type registry invalid!\n");
+  fiobj_free(html_mime);
+}
+#endif