http.c 94.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11
/********************************************************************
 *                                                                  *
 * THIS FILE IS PART OF THE libopusfile SOFTWARE CODEC SOURCE CODE. *
 * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS     *
 * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
 * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING.       *
 *                                                                  *
 * THE libopusfile SOURCE CODE IS (C) COPYRIGHT 2012                *
 * by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
 *                                                                  *
 ********************************************************************/
12 13 14 15 16 17
#include "internal.h"
#include <ctype.h>
#include <errno.h>
#include <limits.h>
#include <string.h>

18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
/*RFCs referenced in this file:
  RFC 1738: Uniform Resource Locators (URL)
  RFC 1945: Hypertext Transfer Protocol -- HTTP/1.0
  RFC 2068: Hypertext Transfer Protocol -- HTTP/1.1
  RFC 2145: Use and Interpretation of HTTP Version Numbers
  RFC 2246: The TLS Protocol Version 1.0
  RFC 2616: Hypertext Transfer Protocol -- HTTP/1.1
  RFC 2617: HTTP Authentication: Basic and Digest Access Authentication
  RFC 2817: Upgrading to TLS Within HTTP/1.1
  RFC 2818: HTTP Over TLS
  RFC 3492: Punycode: A Bootstring encoding of Unicode for Internationalized
   Domain Names in Applications (IDNA)
  RFC 3986: Uniform Resource Identifier (URI): Generic Syntax
  RFC 3987: Internationalized Resource Identifiers (IRIs)
  RFC 5894: Internationalized Domain Names for Applications (IDNA):
   Background, Explanation, and Rationale
  RFC 6066: Transport Layer Security (TLS) Extensions: Extension Definitions*/

36 37 38 39 40 41 42 43 44 45 46
typedef struct OpusParsedURL   OpusParsedURL;
typedef struct OpusStringBuf   OpusStringBuf;
typedef struct OpusHTTPConn    OpusHTTPConn;
typedef struct OpusHTTPStream  OpusHTTPStream;

static char *op_string_range_dup(const char *_start,const char *_end){
  size_t  len;
  char   *ret;
  OP_ASSERT(_start<=_end);
  len=_end-_start;
  ret=(char *)_ogg_malloc(sizeof(*ret)*(len+1));
47 48 49 50
  if(OP_LIKELY(ret!=NULL)){
    memcpy(ret,_start,sizeof(*ret)*(len));
    ret[len]='\0';
  }
51 52 53 54 55 56 57
  return ret;
}

static char *op_string_dup(const char *_s){
  return op_string_range_dup(_s,_s+strlen(_s));
}

58 59 60 61 62 63 64 65 66 67 68
static char *op_string_tolower(char *_s){
  int i;
  for(i=0;_s[i]!='\0';i++){
    int c;
    c=_s[i];
    if(c>='A'&&c<='Z')c+='a'-'A';
    _s[i]=(char)c;
  }
  return _s;
}

69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163
/*URI character classes (from RFC 3986).*/
#define OP_URL_ALPHA \
 "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
#define OP_URL_DIGIT       "01234567890"
#define OP_URL_HEXDIGIT    "01234567890ABCDEFabcdef"
/*Not a character class, but the characters allowed in <scheme>.*/
#define OP_URL_SCHEME      OP_URL_ALPHA OP_URL_DIGIT "+-."
#define OP_URL_GEN_DELIMS  "#/:?@[]"
#define OP_URL_SUB_DELIMS  "!$&'()*+,;="
#define OP_URL_RESERVED    OP_URL_GEN_DELIMS OP_URL_SUB_DELIMS
#define OP_URL_UNRESERVED  OP_URL_ALPHA OP_URL_DIGIT "-._~"
/*Not a character class, but the characters allowed in <pct-encoded>.*/
#define OP_URL_PCT_ENCODED "%"
/*Not a character class or production rule, but for convenience.*/
#define OP_URL_PCHAR_BASE \
 OP_URL_UNRESERVED OP_URL_PCT_ENCODED OP_URL_SUB_DELIMS
#define OP_URL_PCHAR       OP_URL_PCHAR_BASE ":@"
/*Not a character class, but the characters allowed in <userinfo> and
   <IP-literal>.*/
#define OP_URL_PCHAR_NA    OP_URL_PCHAR_BASE ":"
/*Not a character class, but the characters allowed in <segment-nz-nc>.*/
#define OP_URL_PCHAR_NC    OP_URL_PCHAR_BASE "@"
/*Not a character clsss, but the characters allowed in <path>.*/
#define OP_URL_PATH        OP_URL_PCHAR "/"
/*Not a character class, but the characters allowed in <query> / <fragment>.*/
#define OP_URL_QUERY_FRAG  OP_URL_PCHAR "/?"

/*Check the <% HEXDIG HEXDIG> escapes of a URL for validity.
  Return: 0 if valid, or a negative value on failure.*/
static int op_validate_url_escapes(const char *_s){
  int i;
  for(i=0;_s[i];i++){
    if(_s[i]=='%'){
      if(OP_UNLIKELY(!isxdigit(_s[i+1]))
       ||OP_UNLIKELY(!isxdigit(_s[i+2]))
       /*RFC 3986 says %00 "should be rejected if the application is not
          expecting to receive raw data within a component."*/
       ||OP_UNLIKELY(_s[i+1]=='0'&&_s[i+2]=='0')){
        return OP_FALSE;
      }
      i+=2;
    }
  }
  return 0;
}

/*Convert a hex digit to its actual value.
  _c: The hex digit to convert.
      Presumed to be valid ('0'...'9', 'A'...'F', or 'a'...'f').
  Return: The value of the digit, in the range [0,15].*/
static int op_hex_value(int _c){
  return _c>='a'?_c-'a'+10:_c>='A'?_c-'A'+10:_c-'0';
}

/*Unescape all the <% HEXDIG HEXDIG> sequences in a string in-place.
  This does no validity checking.*/
static char *op_unescape_url_component(char *_s){
  int i;
  int j;
  for(i=j=0;_s[i];i++,j++){
    if(_s[i]=='%'){
      _s[i]=(char)(op_hex_value(_s[i+1])<<4|op_hex_value(_s[i+2]));
      i+=2;
    }
  }
  return _s;
}

/*Parse a file: URL.
  This code is not meant to be fast: strspn() with large sets is likely to be
   slow, but it is very convenient.
  It is meant to be RFC 1738-compliant (as updated by RFC 3986).*/
static const char *op_parse_file_url(const char *_src){
  const char *scheme_end;
  const char *path;
  const char *path_end;
  scheme_end=_src+strspn(_src,OP_URL_SCHEME);
  if(OP_UNLIKELY(*scheme_end!=':')
   ||scheme_end-_src!=4||op_strncasecmp(_src,"file",4)!=0){
    /*Unsupported protocol.*/
    return NULL;
  }
  /*Make sure all escape sequences are valid to simplify unescaping later.*/
  if(OP_UNLIKELY(op_validate_url_escapes(scheme_end+1)<0))return NULL;
  if(scheme_end[1]=='/'&&scheme_end[2]=='/'){
    const char *host;
    /*file: URLs can have a host!
      Yeah, I was surprised, too, but that's what RFC 1738 says.
      It also says, "The file URL scheme is unusual in that it does not specify
       an Internet protocol or access method for such files; as such, its
       utility in network protocols between hosts is limited," which is a mild
       understatement.*/
    host=scheme_end+3;
    /*The empty host is what we expect.*/
    if(OP_LIKELY(*host!='/'))path=host;
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182
    else{
      const char *host_end;
      char        host_buf[28];
      /*RFC 1738 says localhost "is interpreted as `the machine from which the
         URL is being interpreted,'" so let's check for it.*/
      host_end=host+strspn(host,OP_URL_PCHAR_BASE);
      /*No <port> allowed.
        This also rejects IP-Literals.*/
      if(*host_end!='/')return NULL;
      /*An escaped "localhost" can take at most 27 characters.*/
      if(OP_UNLIKELY(host_end-host>27))return NULL;
      memcpy(host_buf,host,sizeof(*host_buf)*(host_end-host));
      host_buf[host_end-host]='\0';
      op_unescape_url_component(host_buf);
      op_string_tolower(host_buf);
      /*Some other host: give up.*/
      if(OP_UNLIKELY(strcmp(host_buf,"localhost")!=0))return NULL;
      path=host_end;
    }
183 184 185 186 187 188 189 190 191 192 193 194 195 196
  }
  else path=scheme_end+1;
  path_end=path+strspn(path,OP_URL_PATH);
  /*This will reject a <query> or <fragment> component, too.
    I don't know what to do with queries, but a temporal fragment would at
     least make sense.
    RFC 1738 pretty clearly defines a <searchpart> that's equivalent to the
     RFC 3986 <query> component for other schemes, but not the file: scheme,
     so I'm going to just reject it.*/
  if(*path_end!='\0')return NULL;
  return path;
}

#if defined(OP_ENABLE_HTTP)
197
# include <sys/ioctl.h>
198 199
# include <sys/types.h>
# include <sys/socket.h>
200
# include <sys/timeb.h>
201
# include <arpa/inet.h>
Timothy B. Terriberry's avatar
Timothy B. Terriberry committed
202
# include <netinet/in.h>
203
# include <netinet/tcp.h>
204 205 206 207 208 209
# include <fcntl.h>
# include <netdb.h>
# include <poll.h>
# include <unistd.h>
# include <openssl/ssl.h>

210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280
/*The maximum number of simultaneous connections.
  RFC 2616 says this SHOULD NOT be more than 2, but everyone on the modern web
   ignores that (e.g., IE 8 bumped theirs up from 2 to 6, Firefox uses 15).
  If it makes you feel better, we'll only ever actively read from one of these
   at a time.
  The others are kept around mainly to avoid slow-starting a new connection
   when seeking, and time out rapidly.*/
# define OP_NCONNS_MAX (4)

/*The number of redirections at which we give up.
  The value here is the current default in Firefox.
  RFC 2068 mandated a maximum of 5, but RFC 2616 relaxed that to "a client
   SHOULD detect infinite redirection loops."
  Fortunately, 20 is less than infinity.*/
# define OP_REDIRECT_LIMIT (20)

/*The maximum size of a response message (before the body).
  Responses larger than this will be discarded.
  The buffer for this is currently stack-allocated, which will have to change
   if you want to make it much larger.*/
# define OP_RESPONSE_SIZE_MAX (1024)

/*The number of milliseconds we will allow a connection to sit idle before we
   refuse to resurrect it.
  Apache as of 2.2 has reduced its default timeout to 5 seconds (from 15), so
   that's what we'll use here.*/
# define OP_CONNECTION_IDLE_TIMEOUT_MS (5*1000)

/*The number of milliseconds we will wait to send or receive data before giving
   up.*/
# define OP_POLL_TIMEOUT_MS (30*1000)

/*We will always attempt to read ahead at least this much in preference to
   opening a new connection.*/
# define OP_READAHEAD_THRESH_MIN (32*(opus_int32)1024)
/*The amount to read ahead per iteration of the read-ahead loop.
  16 kB is the largest size OpenSSL will return at once.*/
# define OP_READAHEAD_CHUNK_SIZE (16*1024)

/*The amount of data to request after a seek.
  This is a trade-off between read throughput after a seek vs. the the ability
   to quickly perform another seek with the same connection.*/
# define OP_PIPELINE_CHUNK_SIZE     (32*(opus_int32)1024)
/*Subsequent chunks are requested with larger and larger sizes until they pass
   this threshold, after which we just ask for the rest of the resource.*/
# define OP_PIPELINE_CHUNK_SIZE_MAX (1024*(opus_int32)1024)
/*This is the maximum number of requests we'll make with a single connection.
  Many servers will simply disconnect after we attempt some number of requests,
   possibly without sending a Connection: close header, meaning we won't
   discover it until we try to read beyond the end of the current chunk.
  We can reconnect when that happens, but this is slow.
  Instead, we impose a limit ourselves (set to the default for Apache
   installations and thus likely the most common value in use).*/
# define OP_PIPELINE_MAX_REQUESTS   (100)
/*This should be the number of requests, starting from a chunk size of
   OP_PIPELINE_CHUNK_SIZE and doubling each time, until we exceed
   OP_PIPELINE_CHUNK_SIZE_MAX and just request the rest of the file.
  We won't reuse a connection when seeking unless it has at least this many
   requests left, to reduce the chances we'll have to open a new connection
   while reading forward afterwards.*/
# define OP_PIPELINE_MIN_REQUESTS   (7)

/*Is this an https URL?
  For now we can simply check the last letter of the scheme.*/
# define OP_URL_IS_SSL(_url) ((_url)->scheme[4]=='s')

/*Does this URL use the default port for its scheme?*/
# define OP_URL_IS_DEFAULT_PORT(_url) \
 (!OP_URL_IS_SSL(_url)&&(_url)->port==80 \
 ||OP_URL_IS_SSL(_url)&&(_url)->port==443)

281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301
struct OpusParsedURL{
  /*Either "http" or "https".*/
  char     *scheme;
  /*The user name from the <userinfo> component, or NULL.*/
  char     *user;
  /*The password from the <userinfo> component, or NULL.*/
  char     *pass;
  /*The <host> component.
    This may not be NULL.*/
  char     *host;
  /*The <path> and <query> components.
    This may not be NULL.*/
  char     *path;
  /*The <port> component.
    This is set to the default port if the URL did not contain one.*/
  unsigned  port;
};

/*Parse a URL.
  This code is not meant to be fast: strspn() with large sets is likely to be
   slow, but it is very convenient.
302 303 304 305
  It is meant to be RFC 3986-compliant.
  We currently do not support IRIs (Internationalized Resource Identifiers,
   RFC 3987).
  Callers should translate them to URIs first.*/
306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348
static int op_parse_url_impl(OpusParsedURL *_dst,const char *_src){
  const char  *scheme_end;
  const char  *authority;
  const char  *userinfo_end;
  const char  *user;
  const char  *user_end;
  const char  *pass;
  const char  *hostport;
  const char  *hostport_end;
  const char  *host_end;
  const char  *port;
  opus_int32   port_num;
  const char  *port_end;
  const char  *path;
  const char  *path_end;
  const char  *uri_end;
  scheme_end=_src+strspn(_src,OP_URL_SCHEME);
  if(OP_UNLIKELY(*scheme_end!=':')
   ||OP_UNLIKELY(scheme_end-_src<4)||OP_UNLIKELY(scheme_end-_src>5)
   ||OP_UNLIKELY(op_strncasecmp(_src,"https",scheme_end-_src)!=0)){
    /*Unsupported protocol.*/
    return OP_EIMPL;
  }
  if(OP_UNLIKELY(scheme_end[1]!='/')||OP_UNLIKELY(scheme_end[2]!='/')){
    /*We require an <authority> component.*/
    return OP_EINVAL;
  }
  authority=scheme_end+3;
  /*Make sure all escape sequences are valid to simplify unescaping later.*/
  if(OP_UNLIKELY(op_validate_url_escapes(authority)<0))return OP_EINVAL;
  /*Look for a <userinfo> component.*/
  userinfo_end=authority+strspn(authority,OP_URL_PCHAR_NA);
  if(*userinfo_end=='@'){
    /*Found one.*/
    user=authority;
    /*Look for a password (yes, clear-text passwords are deprecated, I know,
       but what else are people supposed to use? use SSL if you care).*/
    user_end=authority+strspn(authority,OP_URL_PCHAR_BASE);
    if(*user_end==':')pass=user_end+1;
    else pass=NULL;
    hostport=userinfo_end+1;
  }
  else{
349 350 351
    /*We shouldn't have to initialize user_end, but gcc is too dumb to figure
       out that user!=NULL below means we didn't take this else branch.*/
    user=user_end=NULL;
352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394
    pass=NULL;
    hostport=authority;
  }
  /*Try to figure out where the <host> component ends.*/
  if(hostport[0]=='['){
    hostport++;
    /*We have an <IP-literal>, which can contain colons.*/
    hostport_end=host_end=hostport+strspn(hostport,OP_URL_PCHAR_NA);
    if(OP_UNLIKELY(*hostport_end++!=']'))return OP_EINVAL;
  }
  /*Currently we don't support IDNA (RFC 5894), because I don't want to deal
     with the policy about which domains should not be internationalized to
     avoid confusing similarities.
    Give this API Punycode (RFC 3492) domain names instead.*/
  else hostport_end=host_end=hostport+strspn(hostport,OP_URL_PCHAR_BASE);
  /*TODO: Validate host.*/
  /*Is there a port number?*/
  port_num=-1;
  if(*hostport_end==':'){
    int i;
    port=hostport_end+1;
    port_end=port+strspn(port,OP_URL_DIGIT);
    path=port_end;
    /*Not part of RFC 3986, but require port numbers in the range 0...65535.*/
    if(OP_LIKELY(port_end-port>0)){
      while(*port=='0')port++;
      if(OP_UNLIKELY(port_end-port>5))return OP_EINVAL;
      port_num=0;
      for(i=0;i<port_end-port;i++)port_num=port_num*10+port[i]-'0';
      if(OP_UNLIKELY(port_num>65535))return OP_EINVAL;
    }
  }
  else path=hostport_end;
  path_end=path+strspn(path,OP_URL_PATH);
  /*If the path is not empty, it must begin with a '/'.*/
  if(OP_LIKELY(path_end>path)&&OP_UNLIKELY(path[0]!='/'))return OP_EINVAL;
  /*Consume the <query> component, if any (right now we don't split this out
     from the <path> component).*/
  if(*path_end=='?')path_end=path_end+strspn(path_end,OP_URL_QUERY_FRAG);
  /*Discard the <fragment> component, if any.
    This doesn't get sent to the server.
    Some day we should add support for Media Fragment URIs
     <http://www.w3.org/TR/media-frags/>.*/
Timothy B. Terriberry's avatar
Timothy B. Terriberry committed
395
  if(*path_end=='#')uri_end=path_end+1+strspn(path_end+1,OP_URL_QUERY_FRAG);
396 397
  else uri_end=path_end;
  /*If there's anything left, this was not a valid URL.*/
Timothy B. Terriberry's avatar
Timothy B. Terriberry committed
398
  if(OP_UNLIKELY(*uri_end!='\0'))return OP_EINVAL;
399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512
  _dst->scheme=op_string_range_dup(_src,scheme_end);
  if(OP_UNLIKELY(_dst->scheme==NULL))return OP_EFAULT;
  op_string_tolower(_dst->scheme);
  if(user!=NULL){
    _dst->user=op_string_range_dup(user,user_end);
    if(OP_UNLIKELY(_dst->user==NULL))return OP_EFAULT;
    op_unescape_url_component(_dst->user);
    /*Unescaping might have created a ':' in the username.
      That's not allowed by RFC 2617's Basic Authentication Scheme.*/
    if(OP_UNLIKELY(strchr(_dst->user,':')!=NULL))return OP_EINVAL;
  }
  else _dst->user=NULL;
  if(pass!=NULL){
    _dst->pass=op_string_range_dup(pass,userinfo_end);
    if(OP_UNLIKELY(_dst->pass==NULL))return OP_EFAULT;
    op_unescape_url_component(_dst->pass);
  }
  else _dst->pass=NULL;
  _dst->host=op_string_range_dup(hostport,host_end);
  if(OP_UNLIKELY(_dst->host==NULL))return OP_EFAULT;
  if(port_num<0){
    if(_src[4]=='s')port_num=443;
    else port_num=80;
  }
  _dst->port=(unsigned)port_num;
  /*RFC 2616 says an empty <abs-path> component is equivalent to "/", and we
     MUST use the latter in the Request-URI.
    Reserve space for the slash here.*/
  if(path==path_end||path[0]=='?')path--;
  _dst->path=op_string_range_dup(path,path_end);
  if(OP_UNLIKELY(_dst->path==NULL))return OP_EFAULT;
  /*And force-set it here.*/
  _dst->path[0]='/';
  return 0;
}

static void op_parsed_url_init(OpusParsedURL *_url){
  memset(_url,0,sizeof(*_url));
}

static void op_parsed_url_clear(OpusParsedURL *_url){
  _ogg_free(_url->scheme);
  _ogg_free(_url->user);
  _ogg_free(_url->pass);
  _ogg_free(_url->host);
  _ogg_free(_url->path);
}

static int op_parse_url(OpusParsedURL *_dst,const char *_src){
  OpusParsedURL url;
  int           ret;
  op_parsed_url_init(&url);
  ret=op_parse_url_impl(&url,_src);
  if(OP_UNLIKELY(ret<0))op_parsed_url_clear(&url);
  else *_dst=*&url;
  return ret;
}

/*A buffer to hold growing strings.
  The main purpose of this is to consolidate allocation checks and simplify
   cleanup on a failed allocation.*/
struct OpusStringBuf{
  char *buf;
  int   nbuf;
  int   cbuf;
};

static void op_sb_init(OpusStringBuf *_sb){
  _sb->buf=NULL;
  _sb->nbuf=0;
  _sb->cbuf=0;
}

static void op_sb_clear(OpusStringBuf *_sb){
  _ogg_free(_sb->buf);
}

static int op_sb_ensure_capacity(OpusStringBuf *_sb,int _capacity){
  char *buf;
  int   cbuf;
  buf=_sb->buf;
  cbuf=_sb->cbuf;
  if(_capacity>=cbuf-1){
    if(OP_UNLIKELY(cbuf>=INT_MAX-1>>1))return OP_EFAULT;
    if(OP_UNLIKELY(_capacity>=INT_MAX-1))return OP_EFAULT;
    cbuf=OP_MAX(2*cbuf+1,_capacity+1);
    buf=_ogg_realloc(buf,sizeof(*buf)*cbuf);
    if(OP_UNLIKELY(buf==NULL))return OP_EFAULT;
    _sb->buf=buf;
    _sb->cbuf=cbuf;
  }
  return 0;
}

static int op_sb_append(OpusStringBuf *_sb,const char *_s,int _len){
  char *buf;
  int   nbuf;
  int   ret;
  nbuf=_sb->nbuf;
  if(OP_UNLIKELY(nbuf>INT_MAX-_len))return OP_EFAULT;
  ret=op_sb_ensure_capacity(_sb,nbuf+_len);
  if(OP_UNLIKELY(ret<0))return ret;
  buf=_sb->buf;
  memcpy(buf+nbuf,_s,sizeof(*buf)*_len);
  nbuf+=_len;
  buf[nbuf]='\0';
  _sb->nbuf=nbuf;
  return 0;
}

static int op_sb_append_string(OpusStringBuf *_sb,const char *_s){
  return op_sb_append(_sb,_s,strlen(_s));
}

513 514 515 516 517 518 519
static int op_sb_append_port(OpusStringBuf *_sb,unsigned _port){
  char port_buf[7];
  OP_ASSERT(_port<=65535U);
  sprintf(port_buf,":%u",_port);
  return op_sb_append_string(_sb,port_buf);
}

520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550
static int op_sb_append_nonnegative_int64(OpusStringBuf *_sb,opus_int64 _i){
  char digit;
  int  nbuf_start;
  int  ret;
  OP_ASSERT(_i>=0);
  nbuf_start=_sb->nbuf;
  ret=0;
  do{
    digit='0'+_i%10;
    ret|=op_sb_append(_sb,&digit,1);
    _i/=10;
  }
  while(_i>0);
  if(OP_LIKELY(ret>=0)){
    char *buf;
    int   nbuf_end;
    buf=_sb->buf;
    nbuf_end=_sb->nbuf-1;
    /*We've added the digits backwards.
      Reverse them.*/
    while(nbuf_start<nbuf_end){
      digit=buf[nbuf_start];
      buf[nbuf_start]=buf[nbuf_end];
      buf[nbuf_end]=digit;
      nbuf_start++;
      nbuf_end--;
    }
  }
  return ret;
}

551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592
static struct addrinfo *op_resolve(const char *_host,unsigned _port){
  struct addrinfo *addrs;
  struct addrinfo  hints;
  char             service[6];
  memset(&hints,0,sizeof(hints));
  hints.ai_socktype=SOCK_STREAM;
  OP_ASSERT(_port<=65535U);
  sprintf(service,"%u",_port);
  if(OP_LIKELY(!getaddrinfo(_host,service,&hints,&addrs)))return addrs;
  return NULL;
}

static int op_sock_set_nonblocking(int _fd,int _nonblocking){
  int flags;
  flags=fcntl(_fd,F_GETFL);
  if(OP_UNLIKELY(flags<0))return flags;
  if(_nonblocking)flags|=O_NONBLOCK;
  else flags&=~O_NONBLOCK;
  return fcntl(_fd,F_SETFL,flags);
}

/*Disable/enable write coalescing if we can.
  We always send whole requests at once and always parse the response headers
   before sending another one, so normally write coalescing just causes added
   delay.*/
static void op_sock_set_tcp_nodelay(int _fd,int _nodelay){
# if defined(TCP_NODELAY)&&(defined(IPPROTO_TCP)||defined(SOL_TCP))
#  if defined(IPPROTO_TCP)
#   define OP_SO_LEVEL IPPROTO_TCP
#  else
#   define OP_SO_LEVEL SOL_TCP
#  endif
  int ret;
  ret=setsockopt(_fd,OP_SO_LEVEL,TCP_NODELAY,&_nodelay,sizeof(_nodelay));
  /*It doesn't really matter if this call fails, but it would be interesting
     to hit a case where it does.*/
  OP_ASSERT(!ret);
# endif
}

/*A single physical connection to an HTTP server.
  We may have several of these open at once.*/
593 594 595
struct OpusHTTPConn{
  /*The current position indicator for this connection.*/
  opus_int64    pos;
596 597 598 599 600 601 602 603 604
  /*The position where the current request will end, or -1 if we're reading
     until EOF (an unseekable stream or the initial HTTP/1.0 request).*/
  opus_int64    end_pos;
  /*The position where next request we've sent will start, or -1 if we haven't
     sent the next request yet.*/
  opus_int64    next_pos;
  /*The end of the next request or -1 if we requested the rest of the resource.
    This is only set to a meaningful value if next_pos is not -1.*/
  opus_int64    next_end;
605
  /*The SSL connection, if this is https.*/
606
  SSL          *ssl_conn;
607
  /*The next connection in either the LRU or free list.*/
608
  OpusHTTPConn *next;
609 610 611 612 613 614 615
  /*The last time we blocked for reading from this connection.*/
  struct timeb  read_time;
  /*The number of bytes we've read since the last time we blocked.*/
  opus_int64    read_bytes;
  /*The estimated throughput of this connection, in bytes/s.*/
  opus_int64    read_rate;
  /*The socket we're reading from.*/
616
  int           fd;
617 618 619 620
  /*The number of remaining requests we are allowed on this connection.*/
  int           nrequests_left;
  /*The chunk size to use for pipelining requests.*/
  opus_int32    chunk_size;
621 622 623
};

static void op_http_conn_init(OpusHTTPConn *_conn){
624
  _conn->next_pos=-1;
625 626 627 628 629 630 631 632
  _conn->ssl_conn=NULL;
  _conn->next=NULL;
  _conn->fd=-1;
}

static void op_http_conn_clear(OpusHTTPConn *_conn){
  if(_conn->ssl_conn!=NULL)SSL_free(_conn->ssl_conn);
  /*SSL frees the BIO for us.*/
633
  if(_conn->fd>=0)close(_conn->fd);
634 635
}

636
/*The global stream state.*/
637 638 639 640 641
struct OpusHTTPStream{
  /*The list of connections.*/
  OpusHTTPConn     conns[OP_NCONNS_MAX];
  /*The context object used as a framework for TLS/SSL functions.*/
  SSL_CTX         *ssl_ctx;
642 643
  /*The cached session to reuse for future connections.*/
  SSL_SESSION     *ssl_session;
644 645
  /*The LRU list (ordered from MRU to LRU) of currently connected
     connections.*/
646 647 648 649 650 651 652 653 654 655 656 657 658 659 660
  OpusHTTPConn    *lru_head;
  /*The free list.*/
  OpusHTTPConn    *free_head;
  /*The URL to connect to.*/
  OpusParsedURL    url;
  /*Information about the address we connected to.*/
  struct addrinfo  addr_info;
  /*The address we connected to.*/
  union{
    struct sockaddr     s;
    struct sockaddr_in  v4;
    struct sockaddr_in6 v6;
  }                addr;
  /*A buffer used to build HTTP requests.*/
  OpusStringBuf    request;
661 662
  /*A buffer used to build proxy CONNECT requests.*/
  OpusStringBuf    proxy_connect;
663
  /*The Content-Length, if specified, or -1 otherwise.
664
    This will always be specified for seekable streams.*/
665 666 667
  opus_int64       content_length;
  /*The position indicator used when no connection is active.*/
  opus_int64       pos;
668 669 670 671 672 673 674 675 676 677 678 679 680
  /*The connection we're currently reading from.
    This can be -1 if no connection is active.*/
  int              cur_conni;
  /*Whether or not the server supports range requests.*/
  int              seekable;
  /*Whether or not the server supports HTTP/1.1 with persistent connections.*/
  int              pipeline;
  /*The offset of the tail of the request.
    Only the offset in the Range: header appears after this, allowing us to
     quickly edit the request to ask for a new range.*/
  int              request_tail;
  /*The estimated time required to open a new connection, in milliseconds.*/
  opus_int32       connect_rate;
681 682 683 684 685 686 687 688 689 690 691 692
};

static void op_http_stream_init(OpusHTTPStream *_stream){
  OpusHTTPConn **pnext;
  int ci;
  pnext=&_stream->free_head;
  for(ci=0;ci<OP_NCONNS_MAX;ci++){
    op_http_conn_init(_stream->conns+ci);
    *pnext=_stream->conns+ci;
    pnext=&_stream->conns[ci].next;
  }
  _stream->ssl_ctx=NULL;
693
  _stream->ssl_session=NULL;
694 695
  _stream->lru_head=NULL;
  op_parsed_url_init(&_stream->url);
696
  op_sb_init(&_stream->request);
697
  op_sb_init(&_stream->proxy_connect);
698 699 700
  _stream->seekable=0;
}

701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723
/*Close the connection and move it to the free list.
  _stream:     The stream containing the free list.
  _conn:       The connection to close.
  _penxt:      The linked-list pointer currently pointing to this connection.
  _gracefully: Whether or not to shut down cleanly.*/
static void op_http_conn_close(OpusHTTPStream *_stream,OpusHTTPConn *_conn,
 OpusHTTPConn **_pnext,int _gracefully){
  /*If we don't shut down gracefully, the server MUST NOT re-use our session
     according to RFC 2246, because it can't tell the difference between an
     abrupt close and a truncation attack.
    So we shut down gracefully if we can.
    However, we will not wait if this would block (it's not worth the savings
     from session resumption to do so).
    Clients (that's us) MAY resume a TLS session that ended with an incomplete
     close, according to RFC 2818, so that's no reason to make sure the server
     shut things down gracefully.
    It also says "client implementations MUST treat any premature closes as
     errors and the data received as potentially truncated," but libopusfile
     treats errors and potentially truncated data in unseekable streams just
     like a normal EOF.
    We warn about this in the docs, and give some suggestions if you truly want
     to avoid truncation attacks.*/
  if(_gracefully&&_conn->ssl_conn!=NULL)SSL_shutdown(_conn->ssl_conn);
724
  op_http_conn_clear(_conn);
725
  _conn->next_pos=-1;
726 727
  _conn->ssl_conn=NULL;
  _conn->fd=-1;
728 729
  OP_ASSERT(*_pnext==_conn);
  *_pnext=_conn->next;
730 731 732 733 734
  _conn->next=_stream->free_head;
  _stream->free_head=_conn;
}

static void op_http_stream_clear(OpusHTTPStream *_stream){
735 736 737
  while(_stream->lru_head!=NULL){
    op_http_conn_close(_stream,_stream->lru_head,&_stream->lru_head,0);
  }
738
  if(_stream->ssl_session!=NULL)SSL_SESSION_free(_stream->ssl_session);
739
  if(_stream->ssl_ctx!=NULL)SSL_CTX_free(_stream->ssl_ctx);
740
  op_sb_clear(&_stream->proxy_connect);
741 742 743 744
  op_sb_clear(&_stream->request);
  op_parsed_url_clear(&_stream->url);
}

745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785
static int op_http_conn_write_fully(OpusHTTPConn *_conn,
 const char *_buf,int _size){
  struct pollfd  fd;
  SSL           *ssl_conn;
  fd.fd=_conn->fd;
  ssl_conn=_conn->ssl_conn;
  while(_size>0){
    int err;
    if(ssl_conn!=NULL){
      int ret;
      ret=SSL_write(ssl_conn,_buf,_size);
      if(ret>0){
        /*Wrote some data.*/
        _buf+=ret;
        _size-=ret;
        continue;
      }
      /*Connection closed.*/
      else if(ret==0)return OP_FALSE;
      err=SSL_get_error(ssl_conn,ret);
      /*Yes, renegotiations can cause SSL_write() to block for reading.*/
      if(err==SSL_ERROR_WANT_READ)fd.events=POLLIN;
      else if(err==SSL_ERROR_WANT_WRITE)fd.events=POLLOUT;
      else return OP_FALSE;
    }
    else{
      ssize_t ret;
      errno=0;
      ret=write(fd.fd,_buf,_size);
      if(ret>0){
        _buf+=ret;
        _size-=ret;
        continue;
      }
      err=errno;
      if(err!=EAGAIN&&err!=EWOULDBLOCK)return OP_FALSE;
      fd.events=POLLOUT;
    }
    if(poll(&fd,1,OP_POLL_TIMEOUT_MS)<=0)return OP_FALSE;
  }
  return 0;
786 787
}

788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803
static int op_http_conn_estimate_available(OpusHTTPConn *_conn){
  int available;
  int ret;
  ret=ioctl(_conn->fd,FIONREAD,&available);
  if(ret<0)available=0;
  /*This requires the SSL read_ahead flag to be unset to work.
    We ignore partial records as well as the protocol overhead for any pending
     bytes.
    This means we might return somewhat less than can truly be read without
     blocking (if there's a partial record).
    This is okay, because we're using this value to estimate network transfer
     time, and we _have_ already received those bytes.
    We also might return slightly more (due to protocol overhead), but that's
     small enough that it probably doesn't matter.*/
  if(_conn->ssl_conn!=NULL)available+=SSL_pending(_conn->ssl_conn);
  return available;
804 805
}

806 807 808 809 810 811 812 813 814
static opus_int32 op_time_diff_ms(const struct timeb *_end,
 const struct timeb *_start){
  opus_int64 dtime;
  dtime=_end->time-_start->time;
  OP_ASSERT(_end->millitm<1000);
  OP_ASSERT(_start->millitm<1000);
  if(OP_UNLIKELY(dtime>(0x7FFFFFFF-1000)/1000))return 0x7FFFFFFF;
  if(OP_UNLIKELY(dtime<(-0x7FFFFFFF+999)/1000))return -0x7FFFFFFF-1;
  return (opus_int32)dtime*1000+_end->millitm-_start->millitm;
815 816
}

817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835
/*Update the read rate estimate for this connection.*/
static void op_http_conn_read_rate_update(OpusHTTPConn *_conn){
  struct timeb read_time;
  opus_int32   read_delta_ms;
  opus_int64   read_delta_bytes;
  opus_int64   read_rate;
  int          ret;
  read_delta_bytes=_conn->read_bytes;
  if(read_delta_bytes<=0)return;
  ret=ftime(&read_time);
  OP_ASSERT(!ret);
  read_delta_ms=op_time_diff_ms(&read_time,&_conn->read_time);
  read_rate=_conn->read_rate;
  read_delta_ms=OP_MAX(read_delta_ms,1);
  read_rate+=read_delta_bytes*1000/read_delta_ms-read_rate+4>>3;
  *&_conn->read_time=*&read_time;
  _conn->read_bytes=0;
  _conn->read_rate=read_rate;
}
836

837 838 839 840 841 842 843 844 845 846 847 848 849 850
/*Tries to read from the given connection.
  [out] _buf: Returns the data read.
  _size:      The size of the buffer.
  _blocking:  Whether or not to block until some data is retrieved.*/
static ptrdiff_t op_http_conn_read(OpusHTTPConn *_conn,
 char *_buf,ptrdiff_t _size,int _blocking){
  struct pollfd   fd;
  SSL            *ssl_conn;
  ptrdiff_t       nread;
  ptrdiff_t       nread_unblocked;
  fd.fd=_conn->fd;
  ssl_conn=_conn->ssl_conn;
  nread=nread_unblocked=0;
  do{
851
    int err;
852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894
    if(ssl_conn!=NULL){
      int ret;
      ret=SSL_read(ssl_conn,_buf+nread,_size-nread);
      if(ret>0){
        /*Read some data.
          Keep going to see if there's more.*/
        nread+=ret;
        nread_unblocked+=ret;
        continue;
      }
      /*Connection closed.*/
      else if(ret==0)break;
      /*If we already read some data, return it right now.*/
      if(nread>0)break;
      err=SSL_get_error(ssl_conn,ret);
      if(err==SSL_ERROR_WANT_READ)fd.events=POLLIN;
      /*Yes, renegotiations can cause SSL_read() to block for writing.*/
      else if(err==SSL_ERROR_WANT_WRITE)fd.events=POLLOUT;
      else return 0;
    }
    else{
      ssize_t ret;
      errno=0;
      ret=read(fd.fd,_buf+nread,_size-nread);
      if(ret>0){
        /*Read some data.
          Keep going to see if there's more.*/
        nread+=ret;
        nread_unblocked+=ret;
        continue;
      }
      /*If we already read some data, return it right now.*/
      if(nread>0)break;
      err=errno;
      if(err!=EAGAIN&&err!=EWOULDBLOCK)return 0;
      fd.events=POLLIN;
    }
    _conn->read_bytes+=nread_unblocked;
    op_http_conn_read_rate_update(_conn);
    nread_unblocked=0;
    if(!_blocking)break;
    /*Need to wait to get any data at all.*/
    if(poll(&fd,1,OP_POLL_TIMEOUT_MS)<=0)return 0;
895 896
  }
  while(nread<_size);
897
  _conn->read_bytes+=nread_unblocked;
898 899 900 901
  return nread;
}

/*Reads the entirety of a response to an HTTP request into a buffer.
902 903 904 905 906 907 908
  Actual parsing and validation is done later.
  _buf:  The buffer in which to read the response.
         No terminating NUL is appended.
  _size: The number of bytes available in the buffer.
  Return: The number of bytes in the response on success, OP_EREAD if the
           connection was closed before reading any data, or another negative
           value on any other error.*/
909 910 911 912 913 914 915 916 917 918 919 920
static int op_http_conn_read_response(OpusHTTPConn *_conn,
 char *_buf,int _size){
  /*The remaining size of the buffer.*/
  int size;
  /*How many characters we've yet to see from the "\r\n\r\n" terminator.*/
  int state;
  size=_size;
  state=4;
  while(size>=state){
    ptrdiff_t ret;
    int       len;
    ret=op_http_conn_read(_conn,_buf,state,1);
921
    if(ret<=0)return _size==size?OP_EREAD:OP_FALSE;
922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953
    /*We read some data.*/
    _buf+=ret;
    size-=ret;
    len=_size-size;
    /*Make sure the starting characters are "HTTP".
      Otherwise we could wind up waiting forever for a response from
       something that is not an HTTP server.*/
    if(len-ret<4&&op_strncasecmp(_buf-len,"HTTP",OP_MIN(len,4))!=0){
      return OP_FALSE;
    }
    /*How far along on the "\r\n\r\n" terminator are we?*/
    if(*(_buf-1)=='\n'){
      if(len>=2&&*(_buf-2)=='\r'){
        if(len>=4&&*(_buf-3)=='\n'&&*(_buf-4)=='\r')return len;
        state=2;
      }
      else state=4;
    }
    else if(*(_buf-1)=='\r'){
      state=3;
      if(len>=3&&*(_buf-2)=='\n'&&*(_buf-3)=='\r')state=1;
    }
    else state=4;
  }
  /*Not enough space left in the buffer to add the characters we'd need to get
     a valid terminator.*/
  return OP_EIMPL;
}

# define OP_HTTP_DIGIT "01234567890"

/*The Reason-Phrase is not allowed to contain control characters, except
954
   horizontal tab (HT: \011).*/
955 956 957 958 959 960 961 962
# define OP_HTTP_CREASON_PHRASE \
 "\001\002\003\004\005\006\007\010\012\013\014\015\016\017\020\021" \
 "\022\023\024\025\026\027\030\031\032\033\034\035\036\037\177"

# define OP_HTTP_CTLS \
 "\001\002\003\004\005\006\007\010\011\012\013\014\015\016\017\020" \
 "\021\022\023\024\025\026\027\030\031\032\033\034\035\036\037\177"

963 964
/*This also includes '\t', but we get that from OP_HTTP_CTLS.*/
# define OP_HTTP_SEPARATORS " \"(),/:;<=>?@[\\]{}"
965 966 967 968 969 970 971 972 973 974 975 976 977 978 979

/*TEXT can also include LWS, but that has structure, so we parse it
   separately.*/
# define OP_HTTP_CTOKEN OP_HTTP_CTLS OP_HTTP_SEPARATORS

/*Return: The amount of linear white space (LWS) at the start of _s.*/
static int op_http_lwsspn(const char *_s){
  int i;
  for(i=0;;){
    if(_s[0]=='\r'&&_s[1]=='\n'&&(_s[2]=='\t'||_s[2]==' '))i+=3;
    else if(_s[i]=='\t'||_s[i]==' ')i++;
    else return i;
  }
}

980 981
static char *op_http_parse_status_line(int *_v1_1_compat,
 char **_status_code,char *_response){
982 983
  char   *next;
  char   *status_code;
984
  int     v1_1_compat;
985 986 987 988 989 990 991 992 993 994
  size_t  d;
  /*RFC 2616 Section 6.1 does not say that the tokens in the Status-Line cannot
     be separated by optional LWS, but since it specifically calls out where
     spaces are to be placed and that CR and LF are not allowed except at the
     end, I am assuming this to be true.*/
  /*We already validated that this starts with "HTTP"*/
  OP_ASSERT(op_strncasecmp(_response,"HTTP",4)==0);
  next=_response+4;
  if(OP_UNLIKELY(*next++!='/'))return NULL;
  d=strspn(next,OP_HTTP_DIGIT);
995 996 997 998 999 1000 1001 1002
  /*"Leading zeros MUST be ignored by recipients."*/
  while(*next=='0'){
    next++;
    OP_ASSERT(d>0);
    d--;
  }
  /*We only support version 1.x*/
  if(OP_UNLIKELY(d!=1)||OP_UNLIKELY(*next++!='1'))return NULL;
1003 1004 1005
  if(OP_UNLIKELY(*next++!='.'))return NULL;
  d=strspn(next,OP_HTTP_DIGIT);
  if(OP_UNLIKELY(d<=0))return NULL;
1006 1007 1008 1009 1010 1011 1012 1013 1014
  /*"Leading zeros MUST be ignored by recipients."*/
  while(*next=='0'){
    next++;
    OP_ASSERT(d>0);
    d--;
  }
  /*We don't need to parse the version number.
    Any non-zero digit means it's greater than 1.*/
  v1_1_compat=d>0;
1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025
  next+=d;
  if(OP_UNLIKELY(*next++!=' '))return NULL;
  status_code=next;
  d=strspn(next,OP_HTTP_DIGIT);
  if(OP_UNLIKELY(d!=3))return NULL;
  next+=d;
  /*The Reason-Phrase can be empty, but the space must be here.*/
  if(OP_UNLIKELY(*next++!=' '))return NULL;
  next+=strcspn(next,OP_HTTP_CREASON_PHRASE);
  if(OP_UNLIKELY(*next++!='\r'))return NULL;
  if(OP_UNLIKELY(*next++!='\n'))return NULL;
1026
  if(_v1_1_compat!=NULL)*_v1_1_compat=v1_1_compat;
1027 1028 1029 1030
  *_status_code=status_code;
  return next;
}

1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044
/*Get the next response header.
  [out] _header: The header token, NUL-terminated, with leading and trailing
                  whitespace stripped, and converted to lower case (to simplify
                  case-insensitive comparisons), or NULL if there are no more
                  response headers.
  [out] _cdr:    The remaining contents of the header, excluding the initial
                  colon (':') and the terminating CRLF ("\r\n"),
                  NUL-terminated, and with leading and trailing whitespace
                  stripped, or NULL if there are no more response headers.
  [inout] _s:    On input, this points to the start of the current line of the
                  response headers.
                 On output, it points to the start of the first line following
                  this header, or NULL if there are no more response headers.
  Return: 0 on success, or a negative value on failure.*/
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
static int op_http_get_next_header(char **_header,char **_cdr,char **_s){
  char   *header;
  char   *header_end;
  char   *cdr;
  char   *cdr_end;
  char   *next;
  size_t  d;
  next=*_s;
  if(next[0]=='\r'&&next[1]=='\n'){
    /*No more headers.*/
    *_header=NULL;
    *_cdr=NULL;
    *_s=NULL;
    return 0;
  }
  header=next+op_http_lwsspn(next);
  d=strcspn(header,OP_HTTP_CTOKEN);
  if(OP_UNLIKELY(d<=0))return OP_FALSE;
  header_end=header+d;
  next=header_end+op_http_lwsspn(header_end);
  if(OP_UNLIKELY(*next++!=':'))return OP_FALSE;
  next+=op_http_lwsspn(next);
  cdr=next;
  do{
    cdr_end=next+strcspn(next,OP_HTTP_CTLS);
    next=cdr_end+op_http_lwsspn(cdr_end);
  }
  while(next>cdr_end);
  if(OP_UNLIKELY(*next++!='\r'))return OP_FALSE;
  if(OP_UNLIKELY(*next++!='\n'))return OP_FALSE;
  *header_end='\0';
  *cdr_end='\0';
  /*Field names are case-insensitive.*/
  op_string_tolower(header);
  *_header=header;
  *_cdr=cdr;
  *_s=next;
  return 0;
}

static opus_int64 op_http_parse_nonnegative_int64(const char **_next,
 const char *_cdr){
  const char *next;
  opus_int64  content_length;
  int         i;
  next=_cdr+strspn(_cdr,OP_HTTP_DIGIT);
  *_next=next;
  if(OP_UNLIKELY(next<=_cdr))return OP_FALSE;
  while(*_cdr=='0')_cdr++;
  if(OP_UNLIKELY(next-_cdr>19))return OP_EIMPL;
  content_length=0;
  for(i=0;i<next-_cdr;i++){
    int digit;
    digit=_cdr[i]-'0';
    /*Check for overflow.*/
    if(OP_UNLIKELY(content_length>(OP_INT64_MAX-9)/10+(digit<=7))){
      return OP_EIMPL;
    }
1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484
    content_length=content_length*10+digit;
  }
  return content_length;
}

static opus_int64 op_http_parse_content_length(const char *_cdr){
  const char *next;
  opus_int64  content_length;
  content_length=op_http_parse_nonnegative_int64(&next,_cdr);
  if(OP_UNLIKELY(*next!='\0'))return OP_FALSE;
  return content_length;
}

static int op_http_parse_content_range(opus_int64 *_first,opus_int64 *_last,
 opus_int64 *_length,const char *_cdr){
  opus_int64 first;
  opus_int64 last;
  opus_int64 length;
  size_t     d;
  if(OP_UNLIKELY(op_strncasecmp(_cdr,"bytes",5)!=0))return OP_FALSE;
  _cdr+=5;
  d=op_http_lwsspn(_cdr);
  if(OP_UNLIKELY(d<=0))return OP_FALSE;
  _cdr+=d;
  if(*_cdr!='*'){
    first=op_http_parse_nonnegative_int64(&_cdr,_cdr);
    if(OP_UNLIKELY(first<0))return (int)first;
    _cdr+=op_http_lwsspn(_cdr);
    if(*_cdr++!='-')return OP_FALSE;
    _cdr+=op_http_lwsspn(_cdr);
    last=op_http_parse_nonnegative_int64(&_cdr,_cdr);
    if(OP_UNLIKELY(last<0))return (int)last;
    _cdr+=op_http_lwsspn(_cdr);
  }
  else{
    /*This is for a 416 response (Requested range not satisfiable).*/
    first=last=-1;
    _cdr++;
  }
  if(OP_UNLIKELY(*_cdr++!='/'))return OP_FALSE;
  if(*_cdr!='*'){
    length=op_http_parse_nonnegative_int64(&_cdr,_cdr);
    if(OP_UNLIKELY(length<0))return (int)length;
  }
  else{
    /*The total length is unspecified.*/
    _cdr++;
    length=-1;
  }
  if(OP_UNLIKELY(*_cdr!='\0'))return OP_FALSE;
  if(OP_UNLIKELY(last<first))return OP_FALSE;
  if(length>=0&&OP_UNLIKELY(last>=length))return OP_FALSE;
  *_first=first;
  *_last=last;
  *_length=length;
  return 0;
}

/*Parse the Connection response header and look for a "close" token.
  Return: 1 if a "close" token is found, 0 if it's not found, and a negative
           value on error.*/
static int op_http_parse_connection(char *_cdr){
  size_t d;
  int    ret;
  ret=0;
  for(;;){
    d=strcspn(_cdr,OP_HTTP_CTOKEN);
    if(OP_UNLIKELY(d<=0))return OP_FALSE;
    if(op_strncasecmp(_cdr,"close",(int)d)==0)ret=1;
    /*We're supposed to strip and ignore any headers mentioned in the
       Connection header if this response is from an HTTP/1.0 server (to
       work around forwarding of hop-by-hop headers by old proxies), but the
       only hop-by-hop header we look at is Connection itself.
      Everything else is a well-defined end-to-end header, and going back and
       undoing the things we did based on already-examined headers would be
       hard (since we only scan them once, in a destructive manner).
      Therefore we just ignore all the other tokens.*/
    _cdr+=d;
    d=op_http_lwsspn(_cdr);
    if(d<=0)break;
    _cdr+=d;
  }
  return OP_UNLIKELY(*_cdr!='\0')?OP_FALSE:0;
}

typedef int (*op_ssl_step_func)(SSL *_ssl_conn);

/*Try to run an SSL function to completion (blocking if necessary).*/
static int op_do_ssl_step(SSL *_ssl_conn,int _fd,op_ssl_step_func _step){
  struct pollfd fd;
  fd.fd=_fd;
  for(;;){
    int ret;
    int err;
    ret=(*_step)(_ssl_conn);
    if(ret>=0)return ret;
    err=SSL_get_error(_ssl_conn,ret);
    if(err==SSL_ERROR_WANT_READ)fd.events=POLLIN;
    else if(err==SSL_ERROR_WANT_WRITE)fd.events=POLLOUT;
    else return OP_FALSE;
    if(poll(&fd,1,OP_POLL_TIMEOUT_MS)<=0)return OP_FALSE;
  }
}

/*Implement a BIO type that just indicates every operation should be retried.
  We use this when initializing an SSL connection via a proxy to allow the
   initial handshake to proceed all the way up to the first read attempt, and
   then return.
  This allows the TLS client hello message to be pipelined with the HTTP
   CONNECT request.*/

static int op_bio_retry_write(BIO *_b,const char *_buf,int _num){
  (void)_buf;
  (void)_num;
  BIO_clear_retry_flags(_b);
  BIO_set_retry_write(_b);
  return -1;
}

static int op_bio_retry_read(BIO *_b,char *_buf,int _num){
  (void)_buf;
  (void)_num;
  BIO_clear_retry_flags(_b);
  BIO_set_retry_read(_b);
  return -1;
}

static int op_bio_retry_puts(BIO *_b,const char *_str){
  return op_bio_retry_write(_b,_str,0);
}

static long op_bio_retry_ctrl(BIO *_b,int _cmd,long _num,void *_ptr){
  long ret;
  (void)_b;
  (void)_num;
  (void)_ptr;
  ret=0;
  switch(_cmd){
    case BIO_CTRL_RESET:
    case BIO_C_RESET_READ_REQUEST:{
      BIO_clear_retry_flags(_b);
      /*Fall through.*/
    }
    case BIO_CTRL_EOF:
    case BIO_CTRL_SET:
    case BIO_CTRL_SET_CLOSE:
    case BIO_CTRL_FLUSH:
    case BIO_CTRL_DUP:{
      ret=1;
    }break;
  }
  return ret;
}

static int op_bio_retry_new(BIO *_b){
  _b->init=1;
  _b->num=0;
  _b->ptr=NULL;
  return 1;
}

static int op_bio_retry_free(BIO *_b){
  return _b!=NULL;
}

/*This is not const because OpenSSL doesn't allow it, even though it won't
   write to it.*/
static BIO_METHOD op_bio_retry_method={
  BIO_TYPE_NULL,
  "retry",
  op_bio_retry_write,
  op_bio_retry_read,
  op_bio_retry_puts,
  NULL,
  op_bio_retry_ctrl,
  op_bio_retry_new,
  op_bio_retry_free,
  NULL
};

/*Establish a CONNECT tunnel and pipeline the start of the TLS handshake for
   proxying https URL requests.*/
int op_http_conn_establish_tunnel(OpusHTTPStream *_stream,
 OpusHTTPConn *_conn,int _fd,SSL *_ssl_conn,BIO *_ssl_bio){
  char  response[OP_RESPONSE_SIZE_MAX];
  BIO  *retry_bio;
  char *status_code;
  char *next;
  int   ret;
  _conn->ssl_conn=NULL;
  _conn->fd=_fd;
  OP_ASSERT(_stream->proxy_connect.nbuf>0);
  ret=op_http_conn_write_fully(_conn,
   _stream->proxy_connect.buf,_stream->proxy_connect.nbuf);
  if(OP_UNLIKELY(ret<0))return ret;
  retry_bio=BIO_new(&op_bio_retry_method);
  if(OP_UNLIKELY(retry_bio==NULL))return OP_EFAULT;
  SSL_set_bio(_ssl_conn,retry_bio,_ssl_bio);
  SSL_set_connect_state(_ssl_conn);
  ret=SSL_connect(_ssl_conn);
  /*This shouldn't succeed, since we can't read yet.*/
  OP_ASSERT(ret<0);
  SSL_set_bio(_ssl_conn,_ssl_bio,_ssl_bio);
  /*Only now do we disable write coalescing, to allow the CONNECT
     request and the start of the TLS handshake to be combined.*/
  op_sock_set_tcp_nodelay(_fd,1);
  ret=op_http_conn_read_response(_conn,response,sizeof(response));
  if(OP_UNLIKELY(ret<0))return ret;
  next=op_http_parse_status_line(NULL,&status_code,response);
  /*According to RFC 2817, "Any successful (2xx) response to a
     CONNECT request indicates that the proxy has established a
     connection to the requested host and port.*/
  if(OP_UNLIKELY(next==NULL)||OP_UNLIKELY(status_code[0]!='2'))return OP_FALSE;
  return 0;
}

/*Perform the TLS handshake on a new connection.*/
int op_http_conn_start_tls(OpusHTTPStream *_stream,OpusHTTPConn *_conn,
 int _fd,SSL *_ssl_conn){
  BIO *ssl_bio;
  int  ret;
  ssl_bio=BIO_new_socket(_fd,BIO_NOCLOSE);
  if(OP_LIKELY(ssl_bio==NULL))return OP_FALSE;
# if !defined(OPENSSL_NO_TLSEXT)
  /*Support for RFC 6066 Server Name Indication.*/
  SSL_set_tlsext_host_name(_ssl_conn,_stream->url.host);
# endif
  /*Resume a previous session if available.*/
  if(_stream->ssl_session!=NULL){
    SSL_set_session(_ssl_conn,_stream->ssl_session);
  }
  /*If we're proxying, establish the CONNECT tunnel.*/
  if(_stream->proxy_connect.nbuf>0){
    ret=op_http_conn_establish_tunnel(_stream,_conn,
     _fd,_ssl_conn,ssl_bio);
    if(OP_UNLIKELY(ret<0))return ret;
  }
  else{
    /*Otherwise, just use this socket directly.*/
    op_sock_set_tcp_nodelay(_fd,1);
    SSL_set_bio(_ssl_conn,ssl_bio,ssl_bio);
    SSL_set_connect_state(_ssl_conn);
  }
  ret=op_do_ssl_step(_ssl_conn,_fd,SSL_connect);
  if(OP_UNLIKELY(ret<=0))return OP_FALSE;
  if(_stream->ssl_session==NULL){
    /*Save a session for later resumption.*/
    ret=op_do_ssl_step(_ssl_conn,_fd,SSL_do_handshake);
    if(OP_UNLIKELY(ret<=0))return OP_FALSE;
    _stream->ssl_session=SSL_get1_session(_ssl_conn);
  }
  _conn->ssl_conn=_ssl_conn;
  _conn->fd=_fd;
  _conn->nrequests_left=OP_PIPELINE_MAX_REQUESTS;
  return 0;
}

/*Try to start a connection to the next address in the given list of a given
   type.
  _fd:           The socket to connect with.
  [inout] _addr: A pointer to the list of addresses.
                 This will be advanced to the first one that matches the given
                  address family (possibly the current one).
  _ai_family:    The address family to connect to.
  Return: 1        If the connection was successful.
          0        If the connection is in progress.
          OP_FALSE If the connection failed and there were no more addresses
                    left to try.
                    *_addr will be set to NULL in this case.*/
static int op_sock_connect_next(int _fd,
 struct addrinfo **_addr,int _ai_family){
  struct addrinfo *addr;
  addr=*_addr;
  for(;;){
    /*Move to the next address of the requested type.*/
    for(;addr!=NULL&&addr->ai_family!=_ai_family;addr=addr->ai_next);
    *_addr=addr;
    /*No more: failure.*/
    if(addr==NULL)return OP_FALSE;
    if(connect(_fd,addr->ai_addr,addr->ai_addrlen)>=0)return 1;
    if(OP_LIKELY(errno==EINPROGRESS))return 0;
  }
}

/*The number of address families to try connecting to simultaneously.*/
# define OP_NPROTOS (2)

static int op_http_connect(OpusHTTPStream *_stream,OpusHTTPConn *_conn,
 struct addrinfo *_addrs,struct timeb *_start_time){
  struct addrinfo *addr;
  struct addrinfo *addrs[OP_NPROTOS];
  struct pollfd    fds[OP_NPROTOS];
  int              ai_family;
  int              nprotos;
  int              ret;
  int              pi;
  int              pj;
  for(pi=0;pi<OP_NPROTOS;pi++)addrs[pi]=NULL;
  addr=_addrs;
  /*Try connecting via both IPv4 and IPv6 simultaneously, and keep the first
     one that succeeds.*/
  for(;addr!=NULL;addr=addr->ai_next){
    /*Give IPv6 a slight edge by putting it first in the list.*/
    if(addr->ai_family==AF_INET6){
      OP_ASSERT(addr->ai_addrlen<=sizeof(struct sockaddr_in6));
      if(addrs[0]==NULL)addrs[0]=addr;
    }
    else if(addr->ai_family==AF_INET){
      OP_ASSERT(addr->ai_addrlen<=sizeof(struct sockaddr_in));
      if(addrs[1]==NULL)addrs[1]=addr;
    }
  }
  /*Consolidate the list of addresses.*/
  for(pi=nprotos=0;pi<OP_NPROTOS;pi++){
    if(addrs[pi]!=NULL){
      addrs[nprotos]=addrs[pi];
      nprotos++;
    }
  }
  /*Pop the connection off the free list and put it on the LRU list.*/
  OP_ASSERT(_stream->free_head==_conn);
  _stream->free_head=_conn->next;
  _conn->next=_stream->lru_head;
  _stream->lru_head=_conn;
  ret=ftime(_start_time);
  OP_ASSERT(!ret);
  *&_conn->read_time=*_start_time;
  _conn->read_bytes=0;
  _conn->read_rate=0;
  /*Try to start a connection to each protocol.*/
  for(pi=0;pi<nprotos;pi++){
    ai_family=addrs[pi]->ai_family;
    fds[pi].fd=socket(ai_family,SOCK_STREAM,addrs[pi]->ai_protocol);
    fds[pi].events=POLLOUT;
    if(OP_LIKELY(fds[pi].fd>=0)){
      if(OP_LIKELY(op_sock_set_nonblocking(fds[pi].fd,1)>=0)){
        ret=op_sock_connect_next(fds[pi].fd,addrs+pi,ai_family);
        if(ret>1){
          /*It succeeded right away, so stop.*/
          nprotos=pi+1;
          break;
        }
        /*Otherwise go on to the next protocol, and skip the clean-up below.*/
        else if(ret==0)continue;
        /*Tried all the addresses for this protocol.*/
      }
      /*Clean up the socket.*/
      close(fds[pi].fd);
    }
    /*Remove this protocol from the list.*/
    memmove(addrs+pi,addrs+pi+1,sizeof(*addrs)*(nprotos-pi-1));
    nprotos--;
    pi--;
  }
  /*Wait for one of the connections to finish.*/
  while(pi>=nprotos&&nprotos>0&&poll(fds,nprotos,OP_POLL_TIMEOUT_MS)>0){
    for(pi=0;pi<nprotos;pi++){
      socklen_t errlen;
      int       err;
      /*Still waiting...*/
      if(!fds[pi].revents)continue;
      errlen=sizeof(err);
      if(getsockopt(fds[pi].fd,SOL_SOCKET,SO_ERROR,&err,&errlen)>=0&&err==0){
        /*Success!*/
        break;
      }
      /*Move on to the next address for this protocol.*/
      ai_family=addrs[pi]->ai_family;
      addrs[pi]=addrs[pi]->ai_next;
      ret=op_sock_connect_next(fds[pi].fd,addrs+pi,ai_family);
      /*It succeeded right away, so stop.*/
      if(ret>0)break;
      /*Otherwise go on to the next protocol, and skip the clean-up below.*/
      else if(ret==0)continue;
      /*Tried all the addresses for this protocol.
        Remove it from the list.*/
      close(fds[pi].fd);
      memmove(fds+pi,fds+pi+1,sizeof(*fds)*(nprotos-pi-1));
      memmove(addrs+pi,addrs+pi+1,sizeof(*addrs)*(nprotos-pi-1));
      nprotos--;
      pi--;
    }
1485
  }
1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519
  /*Close all the other sockets.*/
  for(pj=0;pj<nprotos;pj++)if(pi!=pj)close(fds[pj].fd);
  /*If none of them succeeded, we're done.*/
  if(pi>=nprotos)return OP_FALSE;
  /*Save this address for future connection attempts.*/
  if(addrs[pi]!=&_stream->addr_info){
    memcpy(&_stream->addr_info,addrs[pi],sizeof(_stream->addr_info));
    _stream->addr_info.ai_addr=&_stream->addr.s;
    _stream->addr_info.ai_next=NULL;
    memcpy(&_stream->addr,addrs[pi]->ai_addr,addrs[pi]->ai_addrlen);
  }
  if(OP_URL_IS_SSL(&_stream->url)){
    SSL *ssl_conn;
    /*Start the SSL connection.*/
    OP_ASSERT(_stream->ssl_ctx!=NULL);
    ssl_conn=SSL_new(_stream->ssl_ctx);
    if(OP_LIKELY(ssl_conn!=NULL)){
      ret=op_http_conn_start_tls(_stream,_conn,fds[pi].fd,ssl_conn);
      if(OP_LIKELY(ret>=0))return ret;
      SSL_free(ssl_conn);
    }
    close(fds[pi].fd);
    _conn->fd=-1;
    return OP_FALSE;
  }
  /*Just a normal non-SSL connection.*/
  _conn->ssl_conn=NULL;
  _conn->fd=fds[pi].fd;
  _conn->nrequests_left=OP_PIPELINE_MAX_REQUESTS;
  /*Disable write coalescing.
    We always send whole requests at once and always parse the response headers
     before sending another one.*/
  op_sock_set_tcp_nodelay(fds[pi].fd,1);
  return 0;
1520 1521
}

1522
# define OP_BASE64_LENGTH(_len) (((_len)+2)/3*4)
1523

1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545
static const char BASE64_TABLE[64]={
  'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
  'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
  'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
  'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
};

static char *op_base64_encode(char *_dst,const char *_src,int _len){
  unsigned s0;
  unsigned s1;
  unsigned s2;
  int      ngroups;
  int      i;
  ngroups=_len/3;
  for(i=0;i<ngroups;i++){
    s0=_src[3*i+0];
    s1=_src[3*i+1];
    s2=_src[3*i+2];
    _dst[4*i+0]=BASE64_TABLE[s0>>2];
    _dst[4*i+1]=BASE64_TABLE[s0&3<<4|s1>>4];
    _dst[4*i+2]=BASE64_TABLE[s1&15<<2|s2>>6];
    _dst[4*i+3]=BASE64_TABLE[s2&63];
1546
  }
1547 1548 1549 1550 1551 1552 1553 1554
  _len-=3*i;
  if(_len==1){
    s0=_src[3*i+0];
    _dst[4*i+0]=BASE64_TABLE[s0>>2];
    _dst[4*i+1]=BASE64_TABLE[s0&3<<4];
    _dst[4*i+2]='=';
    _dst[4*i+3]='=';
    i++;
1555
  }
1556 1557 1558 1559 1560 1561 1562 1563
  else if(_len==2){
    s0=_src[3*i+0];
    s1=_src[3*i+1];
    _dst[4*i+0]=BASE64_TABLE[s0>>2];
    _dst[4*i+1]=BASE64_TABLE[s0&3<<4|s1>>4];
    _dst[4*i+2]=BASE64_TABLE[s1&15<<2];
    _dst[4*i+3]='=';
    i++;
1564
  }
1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631
  _dst[4*i]='\0';
  return _dst+4*i;
}

/*Construct an HTTP authorization header using RFC 2617's Basic Authentication
   Scheme and append it to the given string buffer.*/
static int op_sb_append_basic_auth_header(OpusStringBuf *_sb,
 const char *_header,const char *_user,const char *_pass){
  int user_len;
  int pass_len;
  int user_pass_len;
  int base64_len;
  int nbuf_total;
  int ret;
  ret=op_sb_append_string(_sb,_header);
  ret|=op_sb_append(_sb,": Basic ",8);
  user_len=strlen(_user);
  pass_len=strlen(_pass);
  if(OP_UNLIKELY(pass_len>INT_MAX-user_len))return OP_EFAULT;
  if(OP_UNLIKELY(user_len+pass_len>(INT_MAX>>2)*3-3))return OP_EFAULT;
  user_pass_len=user_len+1+pass_len;
  base64_len=OP_BASE64_LENGTH(user_pass_len);
  /*Stick "user:pass" at the end of the buffer so we can Base64 encode it
     in-place.*/
  nbuf_total=_sb->nbuf;
  if(OP_UNLIKELY(base64_len>INT_MAX-nbuf_total))return OP_EFAULT;
  nbuf_total+=base64_len;
  ret|=op_sb_ensure_capacity(_sb,nbuf_total);
  if(OP_UNLIKELY(ret<0))return ret;
  _sb->nbuf=nbuf_total-user_pass_len;
  ret=op_sb_append(_sb,_user,user_len);
  OP_ASSERT(!ret);
  ret=op_sb_append(_sb,":",1);
  OP_ASSERT(!ret);
  ret=op_sb_append(_sb,_pass,pass_len);
  OP_ASSERT(!ret);
  op_base64_encode(_sb->buf+nbuf_total-base64_len,
   _sb->buf+nbuf_total-user_pass_len,user_pass_len);
  return op_sb_append(_sb,"\r\n",2);
}

static int op_http_allow_pipelining(const char *_server){
  /*Servers known to do bad things with pipelined requests.
    This list is taken from Gecko's nsHttpConnection::SupportsPipelining() (in
     netwerk/protocol/http/nsHttpConnection.cpp).*/
  static const char *BAD_SERVERS[]={
    "EFAServer/",
    "Microsoft-IIS/4.",
    "Microsoft-IIS/5.",
    "Netscape-Enterprise/3.",
    "Netscape-Enterprise/4.",
    "Netscape-Enterprise/5.",
    "Netscape-Enterprise/6.",
    "WebLogic 3.",
    "WebLogic 4.",
    "WebLogic 5.",
    "WebLogic 6.",
    "Winstone Servlet Engine v0."
  };
# define NBAD_SERVERS ((int)(sizeof(BAD_SERVERS)/sizeof(*BAD_SERVERS)))
  if(*_server>='E'&&*_server<='W'){
    int si;
    for(si=0;si<NBAD_SERVERS;si++){
      if(strncmp(_server,BAD_SERVERS[si],strlen(BAD_SERVERS[si]))==0){
        return 0;
      }
    }
1632
  }
1633 1634
  return 1;
# undef NBAD_SERVERS
1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646
}

static int op_http_stream_open(OpusHTTPStream *_stream,const char *_url,
 int _flags,const char *_proxy_host,unsigned _proxy_port,
 const char *_proxy_user,const char *_proxy_pass){
  struct addrinfo *addrs;
  const char      *last_host;
  unsigned         last_port;
  int              nredirs;
  int              ret;
  if(_proxy_host!=NULL&&OP_UNLIKELY(_proxy_port>65535U))return OP_EINVAL;
  last_host=NULL;
1647 1648 1649 1650
  /*We shouldn't have to initialize last_port, but gcc is too dumb to figure
     out that last_host!=NULL implies we've already taken one trip through the
     loop.*/
  last_port=0;
1651 1652 1653
  ret=op_parse_url(&_stream->url,_url);
  if(OP_UNLIKELY(ret<0))return ret;
  for(nredirs=0;nredirs<OP_REDIRECT_LIMIT;nredirs++){
1654 1655 1656 1657 1658 1659 1660
    struct timeb  start_time;
    struct timeb  end_time;
    char          response[OP_RESPONSE_SIZE_MAX];
    char         *next;
    char         *status_code;
    const char   *host;
    unsigned      port;
Timothy B. Terriberry's avatar
M