src/or/cpuworker.c

   1 /* Copyright (c) 2003-2004, Roger Dingledine.
   2  * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
   3  * Copyright (c) 2007-2008, The Tor Project, Inc. */
   4 /* See LICENSE for licensing information */
   5 /* $Id$ */
   6 const char cpuworker_c_id[] =
   7   "$Id$";
   8
   9 /**
  10  * \file cpuworker.c
  11  * \brief Implements a farm of 'CPU worker' processes to perform
  12  * CPU-intensive tasks in another thread or process, to not
  13  * interrupt the main thread.
  14  *
  15  * Right now, we only use this for processing onionskins.
  16  **/
  17
  18 #include "or.h"
  19
  20 /** The maximum number of cpuworker processes we will keep around. */
  21 #define MAX_CPUWORKERS 16
  22 /** The minimum number of cpuworker processes we will keep around. */
  23 #define MIN_CPUWORKERS 1
  24
  25 /** The tag specifies which circuit this onionskin was from. */
  26 #define TAG_LEN 10
  27 /** How many bytes are sent from the cpuworker back to tor? */
  28 #define LEN_ONION_RESPONSE \
  29   (1+TAG_LEN+ONIONSKIN_REPLY_LEN+CPATH_KEY_MATERIAL_LEN)
  30
  31 /** How many cpuworkers we have running right now. */
  32 static int num_cpuworkers=0;
  33 /** How many of the running cpuworkers have an assigned task right now. */
  34 static int num_cpuworkers_busy=0;
  35 /** We need to spawn new cpuworkers whenever we rotate the onion keys
  36  * on platforms where execution contexts==processes.  This variable stores
  37  * the last time we got a key rotation event. */
  38 static time_t last_rotation_time=0;
  39
  40 static void cpuworker_main(void *data) ATTR_NORETURN;
  41 static int spawn_cpuworker(void);
  42 static void spawn_enough_cpuworkers(void);
  43 static void process_pending_task(connection_t *cpuworker);
  44
  45 /** Initialize the cpuworker subsystem.
  46  */
  47 void
  48 cpu_init(void)
  49 {
  50   cpuworkers_rotate();
  51 }
  52
  53 /** Called when we're done sending a request to a cpuworker. */
  54 int
  55 connection_cpu_finished_flushing(connection_t *conn)
  56 {
  57   tor_assert(conn);
  58   tor_assert(conn->type == CONN_TYPE_CPUWORKER);
  59   connection_stop_writing(conn);
  60   return 0;
  61 }
  62
  63 /** Pack global_id and circ_id; set *tag to the result. (See note on
  64  * cpuworker_main for wire format.) */
  65 static void
  66 tag_pack(char *tag, uint64_t conn_id, circid_t circ_id)
  67 {
  68   /*XXXX RETHINK THIS WHOLE MESS !!!! !NM NM NM NM*/
  69   *(uint64_t *)tag     = conn_id;
  70   *(uint16_t *)(tag+8) = circ_id;
  71 }
  72
  73 /** Unpack <b>tag</b> into addr, port, and circ_id.
  74  */
  75 static void
  76 tag_unpack(const char *tag, uint64_t *conn_id, circid_t *circ_id)
  77 {
  78   *conn_id = *(const uint64_t *)tag;
  79   *circ_id = *(const uint16_t *)(tag+8);
  80 }
  81
  82 /** Called when the onion key has changed and we need to spawn new
  83  * cpuworkers.  Close all currently idle cpuworkers, and mark the last
  84  * rotation time as now.
  85  */
  86 void
  87 cpuworkers_rotate(void)
  88 {
  89   connection_t *cpuworker;
  90   while ((cpuworker = connection_get_by_type_state(CONN_TYPE_CPUWORKER,
  91                                                    CPUWORKER_STATE_IDLE))) {
  92     connection_mark_for_close(cpuworker);
  93     --num_cpuworkers;
  94   }
  95   last_rotation_time = time(NULL);
  96   if (server_mode(get_options()))
  97     spawn_enough_cpuworkers();
  98 }
  99
 100 /** If the cpuworker closes the connection,
 101  * mark it as closed and spawn a new one as needed. */
 102 int
 103 connection_cpu_reached_eof(connection_t *conn)
 104 {
 105   log_warn(LD_GENERAL,"Read eof. CPU worker died unexpectedly.");
 106   if (conn->state != CPUWORKER_STATE_IDLE) {
 107     /* the circ associated with this cpuworker will have to wait until
 108      * it gets culled in run_connection_housekeeping(), since we have
 109      * no way to find out which circ it was. */
 110     log_warn(LD_GENERAL,"...and it left a circuit queued; abandoning circ.");
 111     num_cpuworkers_busy--;
 112   }
 113   num_cpuworkers--;
 114   spawn_enough_cpuworkers(); /* try to regrow. hope we don't end up
 115                                 spinning. */
 116   connection_mark_for_close(conn);
 117   return 0;
 118 }
 119
 120 /** Called when we get data from a cpuworker.  If the answer is not complete,
 121  * wait for a complete answer. If the answer is complete,
 122  * process it as appropriate.
 123  */
 124 int
 125 connection_cpu_process_inbuf(connection_t *conn)
 126 {
 127   char success;
 128   char buf[LEN_ONION_RESPONSE];
 129   uint64_t conn_id;
 130   circid_t circ_id;
 131   connection_t *tmp_conn;
 132   or_connection_t *p_conn = NULL;
 133   circuit_t *circ;
 134
 135   tor_assert(conn);
 136   tor_assert(conn->type == CONN_TYPE_CPUWORKER);
 137
 138   if (!buf_datalen(conn->inbuf))
 139     return 0;
 140
 141   if (conn->state == CPUWORKER_STATE_BUSY_ONION) {
 142     if (buf_datalen(conn->inbuf) < LEN_ONION_RESPONSE) /* answer available? */
 143       return 0; /* not yet */
 144     tor_assert(buf_datalen(conn->inbuf) == LEN_ONION_RESPONSE);
 145
 146     connection_fetch_from_buf(&success,1,conn);
 147     connection_fetch_from_buf(buf,LEN_ONION_RESPONSE-1,conn);
 148
 149     /* parse out the circ it was talking about */
 150     tag_unpack(buf, &conn_id, &circ_id);
 151     circ = NULL;
 152     tmp_conn = connection_get_by_global_id(conn_id);
 153     if (tmp_conn && !tmp_conn->marked_for_close &&
 154         tmp_conn->type == CONN_TYPE_OR)
 155       p_conn = TO_OR_CONN(tmp_conn);
 156
 157     if (p_conn)
 158       circ = circuit_get_by_circid_orconn(circ_id, p_conn);
 159
 160     if (success == 0) {
 161       log_debug(LD_OR,
 162                 "decoding onionskin failed. "
 163                 "(Old key or bad software.) Closing.");
 164       if (circ)
 165         circuit_mark_for_close(circ, END_CIRC_REASON_TORPROTOCOL);
 166       goto done_processing;
 167     }
 168     if (!circ) {
 169       /* This happens because somebody sends us a destroy cell and the
 170        * circuit goes away, while the cpuworker is working. This is also
 171        * why our tag doesn't include a pointer to the circ, because we'd
 172        * never know if it's still valid.
 173        */
 174       log_debug(LD_OR,"processed onion for a circ that's gone. Dropping.");
 175       goto done_processing;
 176     }
 177     tor_assert(! CIRCUIT_IS_ORIGIN(circ));
 178     if (onionskin_answer(TO_OR_CIRCUIT(circ), CELL_CREATED, buf+TAG_LEN,
 179                          buf+TAG_LEN+ONIONSKIN_REPLY_LEN) < 0) {
 180       log_warn(LD_OR,"onionskin_answer failed. Closing.");
 181       circuit_mark_for_close(circ, END_CIRC_REASON_INTERNAL);
 182       goto done_processing;
 183     }
 184     log_debug(LD_OR,"onionskin_answer succeeded. Yay.");
 185   } else {
 186     tor_assert(0); /* don't ask me to do handshakes yet */
 187   }
 188
 189 done_processing:
 190   conn->state = CPUWORKER_STATE_IDLE;
 191   num_cpuworkers_busy--;
 192   if (conn->timestamp_created < last_rotation_time) {
 193     connection_mark_for_close(conn);
 194     num_cpuworkers--;
 195     spawn_enough_cpuworkers();
 196   } else {
 197     process_pending_task(conn);
 198   }
 199   return 0;
 200 }
 201
 202 /** Implement a cpuworker.  'data' is an fdarray as returned by socketpair.
 203  * Read and writes from fdarray[1].  Reads requests, writes answers.
 204  *
 205  *   Request format:
 206  *          Task type           [1 byte, always CPUWORKER_TASK_ONION]
 207  *          Opaque tag          TAG_LEN
 208  *          Onionskin challenge ONIONSKIN_CHALLENGE_LEN
 209  *   Response format:
 210  *          Success/failure     [1 byte, boolean.]
 211  *          Opaque tag          TAG_LEN
 212  *          Onionskin challenge ONIONSKIN_REPLY_LEN
 213  *          Negotiated keys     KEY_LEN*2+DIGEST_LEN*2
 214  *
 215  *  (Note: this _should_ be by addr/port, since we're concerned with specific
 216  * connections, not with routers (where we'd use identity).)
 217  */
 218 static void
 219 cpuworker_main(void *data)
 220 {
 221   char question[ONIONSKIN_CHALLENGE_LEN];
 222   uint8_t question_type;
 223   int *fdarray = data;
 224   int fd;
 225
 226   /* variables for onion processing */
 227   char keys[CPATH_KEY_MATERIAL_LEN];
 228   char reply_to_proxy[ONIONSKIN_REPLY_LEN];
 229   char buf[LEN_ONION_RESPONSE];
 230   char tag[TAG_LEN];
 231   crypto_pk_env_t *onion_key = NULL, *last_onion_key = NULL;
 232
 233   fd = fdarray[1]; /* this side is ours */
 234 #ifndef TOR_IS_MULTITHREADED
 235   tor_close_socket(fdarray[0]); /* this is the side of the socketpair the
 236                                  * parent uses */
 237   tor_free_all(1); /* so the child doesn't hold the parent's fd's open */
 238   handle_signals(0); /* ignore interrupts from the keyboard, etc */
 239 #endif
 240   tor_free(data);
 241
 242   dup_onion_keys(&onion_key, &last_onion_key);
 243
 244   for (;;) {
 245     ssize_t r;
 246
 247     if ((r = recv(fd, &question_type, 1, 0)) != 1) {
 248 //      log_fn(LOG_ERR,"read type failed. Exiting.");
 249       if (r == 0) {
 250         log_info(LD_OR,
 251                  "CPU worker exiting because Tor process closed connection "
 252                  "(either rotated keys or died).");
 253       } else {
 254         log_info(LD_OR,
 255                  "CPU worker exiting because of error on connection to Tor "
 256                  "process.");
 257         log_info(LD_OR,"(Error on %d was %s)",
 258                  fd, tor_socket_strerror(tor_socket_errno(fd)));
 259       }
 260       goto end;
 261     }
 262     tor_assert(question_type == CPUWORKER_TASK_ONION);
 263
 264     if (read_all(fd, tag, TAG_LEN, 1) != TAG_LEN) {
 265       log_err(LD_BUG,"read tag failed. Exiting.");
 266       goto end;
 267     }
 268
 269     if (read_all(fd, question, ONIONSKIN_CHALLENGE_LEN, 1) !=
 270         ONIONSKIN_CHALLENGE_LEN) {
 271       log_err(LD_BUG,"read question failed. Exiting.");
 272       goto end;
 273     }
 274
 275     if (question_type == CPUWORKER_TASK_ONION) {
 276       if (onion_skin_server_handshake(question, onion_key, last_onion_key,
 277           reply_to_proxy, keys, CPATH_KEY_MATERIAL_LEN) < 0) {
 278         /* failure */
 279         log_debug(LD_OR,"onion_skin_server_handshake failed.");
 280         memset(buf,0,LEN_ONION_RESPONSE); /* send all zeros for failure */
 281       } else {
 282         /* success */
 283         log_debug(LD_OR,"onion_skin_server_handshake succeeded.");
 284         buf[0] = 1; /* 1 means success */
 285         memcpy(buf+1,tag,TAG_LEN);
 286         memcpy(buf+1+TAG_LEN,reply_to_proxy,ONIONSKIN_REPLY_LEN);
 287         memcpy(buf+1+TAG_LEN+ONIONSKIN_REPLY_LEN,keys,CPATH_KEY_MATERIAL_LEN);
 288       }
 289       if (write_all(fd, buf, LEN_ONION_RESPONSE, 1) != LEN_ONION_RESPONSE) {
 290         log_err(LD_BUG,"writing response buf failed. Exiting.");
 291         goto end;
 292       }
 293       log_debug(LD_OR,"finished writing response.");
 294     }
 295   }
 296  end:
 297   if (onion_key)
 298     crypto_free_pk_env(onion_key);
 299   if (last_onion_key)
 300     crypto_free_pk_env(last_onion_key);
 301   tor_close_socket(fd);
 302   crypto_thread_cleanup();
 303   spawn_exit();
 304 }
 305
 306 /** Launch a new cpuworker. Return 0 if we're happy, -1 if we failed.
 307  */
 308 static int
 309 spawn_cpuworker(void)
 310 {
 311   int *fdarray;
 312   int fd;
 313   connection_t *conn;
 314   int err;
 315
 316   fdarray = tor_malloc(sizeof(int)*2);
 317   if ((err = tor_socketpair(AF_UNIX, SOCK_STREAM, 0, fdarray)) < 0) {
 318     log_warn(LD_NET, "Couldn't construct socketpair for cpuworker: %s",
 319              tor_socket_strerror(-err));
 320     tor_free(fdarray);
 321     return -1;
 322   }
 323
 324   tor_assert(fdarray[0] >= 0);
 325   tor_assert(fdarray[1] >= 0);
 326
 327   fd = fdarray[0];
 328   spawn_func(cpuworker_main, (void*)fdarray);
 329   log_debug(LD_OR,"just spawned a cpu worker.");
 330 #ifndef TOR_IS_MULTITHREADED
 331   tor_close_socket(fdarray[1]); /* don't need the worker's side of the pipe */
 332   tor_free(fdarray);
 333 #endif
 334
 335   conn = connection_new(CONN_TYPE_CPUWORKER, AF_UNIX);
 336
 337   set_socket_nonblocking(fd);
 338
 339   /* set up conn so it's got all the data we need to remember */
 340   conn->s = fd;
 341   conn->address = tor_strdup("localhost");
 342
 343   if (connection_add(conn) < 0) { /* no space, forget it */
 344     log_warn(LD_NET,"connection_add for cpuworker failed. Giving up.");
 345     connection_free(conn); /* this closes fd */
 346     return -1;
 347   }
 348
 349   conn->state = CPUWORKER_STATE_IDLE;
 350   connection_start_reading(conn);
 351
 352   return 0; /* success */
 353 }
 354
 355 /** If we have too few or too many active cpuworkers, try to spawn new ones
 356  * or kill idle ones.
 357  */
 358 static void
 359 spawn_enough_cpuworkers(void)
 360 {
 361   int num_cpuworkers_needed = get_options()->NumCpus;
 362
 363   if (num_cpuworkers_needed < MIN_CPUWORKERS)
 364     num_cpuworkers_needed = MIN_CPUWORKERS;
 365   if (num_cpuworkers_needed > MAX_CPUWORKERS)
 366     num_cpuworkers_needed = MAX_CPUWORKERS;
 367
 368   while (num_cpuworkers < num_cpuworkers_needed) {
 369     if (spawn_cpuworker() < 0) {
 370       log_warn(LD_GENERAL,"Cpuworker spawn failed. Will try again later.");
 371       return;
 372     }
 373     num_cpuworkers++;
 374   }
 375 }
 376
 377 /** Take a pending task from the queue and assign it to 'cpuworker'. */
 378 static void
 379 process_pending_task(connection_t *cpuworker)
 380 {
 381   or_circuit_t *circ;
 382   char *onionskin = NULL;
 383
 384   tor_assert(cpuworker);
 385
 386   /* for now only process onion tasks */
 387
 388   circ = onion_next_task(&onionskin);
 389   if (!circ)
 390     return;
 391   if (assign_onionskin_to_cpuworker(cpuworker, circ, onionskin))
 392     log_warn(LD_OR,"assign_to_cpuworker failed. Ignoring.");
 393 }
 394
 395 /** How long should we let a cpuworker stay busy before we give
 396  * up on it and decide that we have a bug or infinite loop?
 397  * This value is high because some servers with low memory/cpu
 398  * sometimes spend an hour or more swapping, and Tor starves. */
 399 #define CPUWORKER_BUSY_TIMEOUT (60*60*12)
 400
 401 /** We have a bug that I can't find. Sometimes, very rarely, cpuworkers get
 402  * stuck in the 'busy' state, even though the cpuworker process thinks of
 403  * itself as idle. I don't know why. But here's a workaround to kill any
 404  * cpuworker that's been busy for more than CPUWORKER_BUSY_TIMEOUT.
 405  */
 406 static void
 407 cull_wedged_cpuworkers(void)
 408 {
 409   time_t now = time(NULL);
 410   smartlist_t *conns = get_connection_array();
 411   SMARTLIST_FOREACH(conns, connection_t *, conn,
 412   {
 413     if (!conn->marked_for_close &&
 414         conn->type == CONN_TYPE_CPUWORKER &&
 415         conn->state == CPUWORKER_STATE_BUSY_ONION &&
 416         conn->timestamp_lastwritten + CPUWORKER_BUSY_TIMEOUT < now) {
 417       log_notice(LD_BUG,
 418                  "closing wedged cpuworker. Can somebody find the bug?");
 419       num_cpuworkers_busy--;
 420       num_cpuworkers--;
 421       connection_mark_for_close(conn);
 422     }
 423   });
 424 }
 425
 426 /** Try to tell a cpuworker to perform the public key operations necessary to
 427  * respond to <b>onionskin</b> for the circuit <b>circ</b>.
 428  *
 429  * If <b>cpuworker</b> is defined, assert that he's idle, and use him. Else,
 430  * look for an idle cpuworker and use him. If none idle, queue task onto the
 431  * pending onion list and return.  Return 0 if we successfully assign the
 432  * task, or -1 on failure.
 433  */
 434 int
 435 assign_onionskin_to_cpuworker(connection_t *cpuworker,
 436                               or_circuit_t *circ, char *onionskin)
 437 {
 438   char qbuf[1];
 439   char tag[TAG_LEN];
 440
 441   cull_wedged_cpuworkers();
 442   spawn_enough_cpuworkers();
 443
 444   if (1) {
 445     if (num_cpuworkers_busy == num_cpuworkers) {
 446       log_debug(LD_OR,"No idle cpuworkers. Queuing.");
 447       if (onion_pending_add(circ, onionskin) < 0)
 448         return -1;
 449       return 0;
 450     }
 451
 452     if (!cpuworker)
 453       cpuworker = connection_get_by_type_state(CONN_TYPE_CPUWORKER,
 454                                                CPUWORKER_STATE_IDLE);
 455
 456     tor_assert(cpuworker);
 457
 458     if (!circ->p_conn) {
 459       log_info(LD_OR,"circ->p_conn gone. Failing circ.");
 460       tor_free(onionskin);
 461       return -1;
 462     }
 463     tag_pack(tag, circ->p_conn->_base.global_identifier,
 464              circ->p_circ_id);
 465
 466     cpuworker->state = CPUWORKER_STATE_BUSY_ONION;
 467     /* touch the lastwritten timestamp, since that's how we check to
 468      * see how long it's been since we asked the question, and sometimes
 469      * we check before the first call to connection_handle_write(). */
 470     cpuworker->timestamp_lastwritten = time(NULL);
 471     num_cpuworkers_busy++;
 472
 473     qbuf[0] = CPUWORKER_TASK_ONION;
 474     connection_write_to_buf(qbuf, 1, cpuworker);
 475     connection_write_to_buf(tag, sizeof(tag), cpuworker);
 476     connection_write_to_buf(onionskin, ONIONSKIN_CHALLENGE_LEN, cpuworker);
 477     tor_free(onionskin);
 478   }
 479   return 0;
 480 }
 481