src/or/cpuworker.c

   1 /* Copyright (c) 2003-2004, Roger Dingledine.
   2  * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
   3  * Copyright (c) 2007-2011, The Tor Project, Inc. */
   4 /* See LICENSE for licensing information */
   5
   6 /**
   7  * \file cpuworker.c
   8  * \brief Implements a farm of 'CPU worker' processes to perform
   9  * CPU-intensive tasks in another thread or process, to not
  10  * interrupt the main thread.
  11  *
  12  * Right now, we only use this for processing onionskins.
  13  **/
  14
  15 #include "or.h"
  16
  17 /** The maximum number of cpuworker processes we will keep around. */
  18 #define MAX_CPUWORKERS 16
  19 /** The minimum number of cpuworker processes we will keep around. */
  20 #define MIN_CPUWORKERS 1
  21
  22 /** The tag specifies which circuit this onionskin was from. */
  23 #define TAG_LEN 10
  24 /** How many bytes are sent from the cpuworker back to tor? */
  25 #define LEN_ONION_RESPONSE \
  26   (1+TAG_LEN+ONIONSKIN_REPLY_LEN+CPATH_KEY_MATERIAL_LEN)
  27
  28 /** How many cpuworkers we have running right now. */
  29 static int num_cpuworkers=0;
  30 /** How many of the running cpuworkers have an assigned task right now. */
  31 static int num_cpuworkers_busy=0;
  32 /** We need to spawn new cpuworkers whenever we rotate the onion keys
  33  * on platforms where execution contexts==processes.  This variable stores
  34  * the last time we got a key rotation event. */
  35 static time_t last_rotation_time=0;
  36
  37 static void cpuworker_main(void *data) ATTR_NORETURN;
  38 static int spawn_cpuworker(void);
  39 static void spawn_enough_cpuworkers(void);
  40 static void process_pending_task(connection_t *cpuworker);
  41
  42 /** Initialize the cpuworker subsystem.
  43  */
  44 void
  45 cpu_init(void)
  46 {
  47   cpuworkers_rotate();
  48 }
  49
  50 /** Called when we're done sending a request to a cpuworker. */
  51 int
  52 connection_cpu_finished_flushing(connection_t *conn)
  53 {
  54   tor_assert(conn);
  55   tor_assert(conn->type == CONN_TYPE_CPUWORKER);
  56   connection_stop_writing(conn);
  57   return 0;
  58 }
  59
  60 /** Pack global_id and circ_id; set *tag to the result. (See note on
  61  * cpuworker_main for wire format.) */
  62 static void
  63 tag_pack(char *tag, uint64_t conn_id, circid_t circ_id)
  64 {
  65   /*XXXX RETHINK THIS WHOLE MESS !!!! !NM NM NM NM*/
  66   set_uint64(tag, conn_id);
  67   set_uint16(tag+8, circ_id);
  68 }
  69
  70 /** Unpack <b>tag</b> into addr, port, and circ_id.
  71  */
  72 static void
  73 tag_unpack(const char *tag, uint64_t *conn_id, circid_t *circ_id)
  74 {
  75   *conn_id = get_uint64(tag);
  76   *circ_id = get_uint16(tag+8);
  77 }
  78
  79 /** Called when the onion key has changed and we need to spawn new
  80  * cpuworkers.  Close all currently idle cpuworkers, and mark the last
  81  * rotation time as now.
  82  */
  83 void
  84 cpuworkers_rotate(void)
  85 {
  86   connection_t *cpuworker;
  87   while ((cpuworker = connection_get_by_type_state(CONN_TYPE_CPUWORKER,
  88                                                    CPUWORKER_STATE_IDLE))) {
  89     connection_mark_for_close(cpuworker);
  90     --num_cpuworkers;
  91   }
  92   last_rotation_time = time(NULL);
  93   if (server_mode(get_options()))
  94     spawn_enough_cpuworkers();
  95 }
  96
  97 /** If the cpuworker closes the connection,
  98  * mark it as closed and spawn a new one as needed. */
  99 int
 100 connection_cpu_reached_eof(connection_t *conn)
 101 {
 102   log_warn(LD_GENERAL,"Read eof. CPU worker died unexpectedly.");
 103   if (conn->state != CPUWORKER_STATE_IDLE) {
 104     /* the circ associated with this cpuworker will have to wait until
 105      * it gets culled in run_connection_housekeeping(), since we have
 106      * no way to find out which circ it was. */
 107     log_warn(LD_GENERAL,"...and it left a circuit queued; abandoning circ.");
 108     num_cpuworkers_busy--;
 109   }
 110   num_cpuworkers--;
 111   spawn_enough_cpuworkers(); /* try to regrow. hope we don't end up
 112                                 spinning. */
 113   connection_mark_for_close(conn);
 114   return 0;
 115 }
 116
 117 /** Called when we get data from a cpuworker.  If the answer is not complete,
 118  * wait for a complete answer. If the answer is complete,
 119  * process it as appropriate.
 120  */
 121 int
 122 connection_cpu_process_inbuf(connection_t *conn)
 123 {
 124   char success;
 125   char buf[LEN_ONION_RESPONSE];
 126   uint64_t conn_id;
 127   circid_t circ_id;
 128   connection_t *tmp_conn;
 129   or_connection_t *p_conn = NULL;
 130   circuit_t *circ;
 131
 132   tor_assert(conn);
 133   tor_assert(conn->type == CONN_TYPE_CPUWORKER);
 134
 135   if (!buf_datalen(conn->inbuf))
 136     return 0;
 137
 138   if (conn->state == CPUWORKER_STATE_BUSY_ONION) {
 139     if (buf_datalen(conn->inbuf) < LEN_ONION_RESPONSE) /* answer available? */
 140       return 0; /* not yet */
 141     tor_assert(buf_datalen(conn->inbuf) == LEN_ONION_RESPONSE);
 142
 143     connection_fetch_from_buf(&success,1,conn);
 144     connection_fetch_from_buf(buf,LEN_ONION_RESPONSE-1,conn);
 145
 146     /* parse out the circ it was talking about */
 147     tag_unpack(buf, &conn_id, &circ_id);
 148     circ = NULL;
 149     tmp_conn = connection_get_by_global_id(conn_id);
 150     if (tmp_conn && !tmp_conn->marked_for_close &&
 151         tmp_conn->type == CONN_TYPE_OR)
 152       p_conn = TO_OR_CONN(tmp_conn);
 153
 154     if (p_conn)
 155       circ = circuit_get_by_circid_orconn(circ_id, p_conn);
 156
 157     if (success == 0) {
 158       log_debug(LD_OR,
 159                 "decoding onionskin failed. "
 160                 "(Old key or bad software.) Closing.");
 161       if (circ)
 162         circuit_mark_for_close(circ, END_CIRC_REASON_TORPROTOCOL);
 163       goto done_processing;
 164     }
 165     if (!circ) {
 166       /* This happens because somebody sends us a destroy cell and the
 167        * circuit goes away, while the cpuworker is working. This is also
 168        * why our tag doesn't include a pointer to the circ, because we'd
 169        * never know if it's still valid.
 170        */
 171       log_debug(LD_OR,"processed onion for a circ that's gone. Dropping.");
 172       goto done_processing;
 173     }
 174     tor_assert(! CIRCUIT_IS_ORIGIN(circ));
 175     if (onionskin_answer(TO_OR_CIRCUIT(circ), CELL_CREATED, buf+TAG_LEN,
 176                          buf+TAG_LEN+ONIONSKIN_REPLY_LEN) < 0) {
 177       log_warn(LD_OR,"onionskin_answer failed. Closing.");
 178       circuit_mark_for_close(circ, END_CIRC_REASON_INTERNAL);
 179       goto done_processing;
 180     }
 181     log_debug(LD_OR,"onionskin_answer succeeded. Yay.");
 182   } else {
 183     tor_assert(0); /* don't ask me to do handshakes yet */
 184   }
 185
 186 done_processing:
 187   conn->state = CPUWORKER_STATE_IDLE;
 188   num_cpuworkers_busy--;
 189   if (conn->timestamp_created < last_rotation_time) {
 190     connection_mark_for_close(conn);
 191     num_cpuworkers--;
 192     spawn_enough_cpuworkers();
 193   } else {
 194     process_pending_task(conn);
 195   }
 196   return 0;
 197 }
 198
 199 /** Implement a cpuworker.  'data' is an fdarray as returned by socketpair.
 200  * Read and writes from fdarray[1].  Reads requests, writes answers.
 201  *
 202  *   Request format:
 203  *          Task type           [1 byte, always CPUWORKER_TASK_ONION]
 204  *          Opaque tag          TAG_LEN
 205  *          Onionskin challenge ONIONSKIN_CHALLENGE_LEN
 206  *   Response format:
 207  *          Success/failure     [1 byte, boolean.]
 208  *          Opaque tag          TAG_LEN
 209  *          Onionskin challenge ONIONSKIN_REPLY_LEN
 210  *          Negotiated keys     KEY_LEN*2+DIGEST_LEN*2
 211  *
 212  *  (Note: this _should_ be by addr/port, since we're concerned with specific
 213  * connections, not with routers (where we'd use identity).)
 214  */
 215 static void
 216 cpuworker_main(void *data)
 217 {
 218   char question[ONIONSKIN_CHALLENGE_LEN];
 219   uint8_t question_type;
 220   int *fdarray = data;
 221   int fd;
 222
 223   /* variables for onion processing */
 224   char keys[CPATH_KEY_MATERIAL_LEN];
 225   char reply_to_proxy[ONIONSKIN_REPLY_LEN];
 226   char buf[LEN_ONION_RESPONSE];
 227   char tag[TAG_LEN];
 228   crypto_pk_env_t *onion_key = NULL, *last_onion_key = NULL;
 229
 230   fd = fdarray[1]; /* this side is ours */
 231 #ifndef TOR_IS_MULTITHREADED
 232   tor_close_socket(fdarray[0]); /* this is the side of the socketpair the
 233                                  * parent uses */
 234   tor_free_all(1); /* so the child doesn't hold the parent's fd's open */
 235   handle_signals(0); /* ignore interrupts from the keyboard, etc */
 236 #endif
 237   tor_free(data);
 238
 239   dup_onion_keys(&onion_key, &last_onion_key);
 240
 241   for (;;) {
 242     ssize_t r;
 243
 244     if ((r = recv(fd, &question_type, 1, 0)) != 1) {
 245 //      log_fn(LOG_ERR,"read type failed. Exiting.");
 246       if (r == 0) {
 247         log_info(LD_OR,
 248                  "CPU worker exiting because Tor process closed connection "
 249                  "(either rotated keys or died).");
 250       } else {
 251         log_info(LD_OR,
 252                  "CPU worker exiting because of error on connection to Tor "
 253                  "process.");
 254         log_info(LD_OR,"(Error on %d was %s)",
 255                  fd, tor_socket_strerror(tor_socket_errno(fd)));
 256       }
 257       goto end;
 258     }
 259     tor_assert(question_type == CPUWORKER_TASK_ONION);
 260
 261     if (read_all(fd, tag, TAG_LEN, 1) != TAG_LEN) {
 262       log_err(LD_BUG,"read tag failed. Exiting.");
 263       goto end;
 264     }
 265
 266     if (read_all(fd, question, ONIONSKIN_CHALLENGE_LEN, 1) !=
 267         ONIONSKIN_CHALLENGE_LEN) {
 268       log_err(LD_BUG,"read question failed. Exiting.");
 269       goto end;
 270     }
 271
 272     if (question_type == CPUWORKER_TASK_ONION) {
 273       if (onion_skin_server_handshake(question, onion_key, last_onion_key,
 274           reply_to_proxy, keys, CPATH_KEY_MATERIAL_LEN) < 0) {
 275         /* failure */
 276         log_debug(LD_OR,"onion_skin_server_handshake failed.");
 277         *buf = 0; /* indicate failure in first byte */
 278         memcpy(buf+1,tag,TAG_LEN);
 279         /* send all zeros as answer */
 280         memset(buf+1+TAG_LEN, 0, LEN_ONION_RESPONSE-(1+TAG_LEN));
 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         tor_free(onionskin);
 449         return -1;
 450       }
 451       return 0;
 452     }
 453
 454     if (!cpuworker)
 455       cpuworker = connection_get_by_type_state(CONN_TYPE_CPUWORKER,
 456                                                CPUWORKER_STATE_IDLE);
 457
 458     tor_assert(cpuworker);
 459
 460     if (!circ->p_conn) {
 461       log_info(LD_OR,"circ->p_conn gone. Failing circ.");
 462       tor_free(onionskin);
 463       return -1;
 464     }
 465     tag_pack(tag, circ->p_conn->_base.global_identifier,
 466              circ->p_circ_id);
 467
 468     cpuworker->state = CPUWORKER_STATE_BUSY_ONION;
 469     /* touch the lastwritten timestamp, since that's how we check to
 470      * see how long it's been since we asked the question, and sometimes
 471      * we check before the first call to connection_handle_write(). */
 472     cpuworker->timestamp_lastwritten = time(NULL);
 473     num_cpuworkers_busy++;
 474
 475     qbuf[0] = CPUWORKER_TASK_ONION;
 476     connection_write_to_buf(qbuf, 1, cpuworker);
 477     connection_write_to_buf(tag, sizeof(tag), cpuworker);
 478     connection_write_to_buf(onionskin, ONIONSKIN_CHALLENGE_LEN, cpuworker);
 479     tor_free(onionskin);
 480   }
 481   return 0;
 482 }
 483