root/daemons/attrd/attrd_sync.c

/* */

DEFINITIONS

1. next_key
2. free_waitlist_node
3. sync_point_str
4. attrd_add_client_to_waitlist
5. attrd_free_waitlist
6. attrd_remove_client_from_waitlist
7. attrd_ack_waitlist_clients
8. attrd_cluster_sync_point_update
9. attrd_request_sync_point
10. attrd_request_has_sync_point
11. free_action
12. confirmation_timeout_cb
13. attrd_do_not_expect_from_peer
14. attrd_do_not_wait_for_client
15. attrd_expect_confirmations
16. attrd_free_confirmations
17. attrd_handle_confirmation

   1 /*
   2  * Copyright 2022-2024 the Pacemaker project contributors
   3  *
   4  * The version control history for this file may have further details.
   5  *
   6  * This source code is licensed under the GNU General Public License version 2
   7  * or later (GPLv2+) WITHOUT ANY WARRANTY.
   8  */
   9 
  10 #include <crm_internal.h>
  11 
  12 #include <crm/common/xml.h>
  13 #include <crm/common/attrs_internal.h>
  14 
  15 #include "pacemaker-attrd.h"
  16 
  17 /* A hash table storing clients that are waiting on a sync point to be reached.
  18  * The key is waitlist_client - just a plain int.  The obvious key would be
  19  * the IPC client's ID, but this is not guaranteed to be unique.  A single client
  20  * could be waiting on a sync point for multiple attributes at the same time.
  21  *
  22  * It is not expected that this hash table will ever be especially large.
  23  */
  24 static GHashTable *waitlist = NULL;
  25 static int waitlist_client = 0;
  26 
  27 struct waitlist_node {
  28     /* What kind of sync point does this node describe? */
  29     enum attrd_sync_point sync_point;
  30 
  31     /* Information required to construct and send a reply to the client. */
  32     char *client_id;
  33     uint32_t ipc_id;
  34     uint32_t flags;
  35 };
  36 
  37 /* A hash table storing information on in-progress IPC requests that are awaiting
  38  * confirmations.  These requests are currently being processed by peer attrds and
  39  * we are waiting to receive confirmation messages from each peer indicating that
  40  * processing is complete.
  41  *
  42  * Multiple requests could be waiting on confirmations at the same time.
  43  *
  44  * The key is the unique callid for the IPC request, and the value is a
  45  * confirmation_action struct.
  46  */
  47 static GHashTable *expected_confirmations = NULL;
  48 
  49 /*!
  50  * \internal
  51  * \brief A structure describing a single IPC request that is awaiting confirmations
  52  */
  53 struct confirmation_action {
  54     /*!
  55      * \brief A list of peer attrds that we are waiting to receive confirmation
  56      *        messages from
  57      *
  58      * This list is dynamic - as confirmations arrive from peer attrds, they will
  59      * be removed from this list.  When the list is empty, all peers have processed
  60      * the request and the associated confirmation action will be taken.
  61      */
  62     GList *respondents;
  63 
  64     /*!
  65      * \brief A timer that will be used to remove the client should it time out
  66      *        before receiving all confirmations
  67      */
  68     mainloop_timer_t *timer;
  69 
  70     /*!
  71      * \brief A function to run when all confirmations have been received
  72      */
  73     attrd_confirmation_action_fn fn;
  74 
  75     /*!
  76      * \brief Information required to construct and send a reply to the client
  77      */
  78     char *client_id;
  79     uint32_t ipc_id;
  80     uint32_t flags;
  81 
  82     /*!
  83      * \brief The XML request containing the callid associated with this action
  84      */
  85     void *xml;
  86 };
  87 
  88 static void
  89 next_key(void)
     /*  */
  90 {
  91     do {
  92         waitlist_client++;
  93         if (waitlist_client < 0) {
  94             waitlist_client = 1;
  95         }
  96     } while (g_hash_table_contains(waitlist, GINT_TO_POINTER(waitlist_client)));
  97 }
  98 
  99 static void
 100 free_waitlist_node(gpointer data)
     /*  */
 101 {
 102     struct waitlist_node *wl = (struct waitlist_node *) data;
 103 
 104     free(wl->client_id);
 105     free(wl);
 106 }
 107 
 108 static const char *
 109 sync_point_str(enum attrd_sync_point sync_point)
     /*  */
 110 {
 111     if (sync_point == attrd_sync_point_local) {
 112         return PCMK__VALUE_LOCAL;
 113     } else if  (sync_point == attrd_sync_point_cluster) {
 114         return PCMK__VALUE_CLUSTER;
 115     } else {
 116         return PCMK_VALUE_UNKNOWN;
 117     }
 118 }
 119 
 120 /*!
 121  * \internal
 122  * \brief Add a client to the attrd waitlist
 123  *
 124  * Typically, a client receives an ACK for its XML IPC request immediately.  However,
 125  * some clients want to wait until their request has been processed and taken effect.
 126  * This is called a sync point.  Any client placed on this waitlist will have its
 127  * ACK message delayed until either its requested sync point is hit, or until it
 128  * times out.
 129  *
 130  * The XML IPC request must specify the type of sync point it wants to wait for.
 131  *
 132  * \param[in,out] request   The request describing the client to place on the waitlist.
 133  */
 134 void
 135 attrd_add_client_to_waitlist(pcmk__request_t *request)
     /*  */
 136 {
 137     const char *sync_point = attrd_request_sync_point(request->xml);
 138     struct waitlist_node *wl = NULL;
 139 
 140     if (sync_point == NULL) {
 141         return;
 142     }
 143 
 144     if (waitlist == NULL) {
 145         waitlist = pcmk__intkey_table(free_waitlist_node);
 146     }
 147 
 148     wl = pcmk__assert_alloc(1, sizeof(struct waitlist_node));
 149 
 150     if (pcmk__str_eq(sync_point, PCMK__VALUE_LOCAL, pcmk__str_none)) {
 151         wl->sync_point = attrd_sync_point_local;
 152     } else if (pcmk__str_eq(sync_point, PCMK__VALUE_CLUSTER, pcmk__str_none)) {
 153         wl->sync_point = attrd_sync_point_cluster;
 154     } else {
 155         free_waitlist_node(wl);
 156         return;
 157     }
 158 
 159     wl->client_id = pcmk__str_copy(request->ipc_client->id);
 160     wl->ipc_id = request->ipc_id;
 161     wl->flags = request->flags;
 162 
 163     next_key();
 164     pcmk__intkey_table_insert(waitlist, waitlist_client, wl);
 165 
 166     crm_trace("Added client %s to waitlist for %s sync point",
 167               wl->client_id, sync_point_str(wl->sync_point));
 168     crm_trace("%d clients now on waitlist", g_hash_table_size(waitlist));
 169 
 170     /* And then add the key to the request XML so we can uniquely identify
 171      * it when it comes time to issue the ACK.
 172      */
 173     crm_xml_add_int(request->xml, PCMK__XA_CALL_ID, waitlist_client);
 174 }
 175 
 176 /*!
 177  * \internal
 178  * \brief Free all memory associated with the waitlist.  This is most typically
 179  *        used when attrd shuts down.
 180  */
 181 void
 182 attrd_free_waitlist(void)
     /*  */
 183 {
 184     if (waitlist == NULL) {
 185         return;
 186     }
 187 
 188     g_hash_table_destroy(waitlist);
 189     waitlist = NULL;
 190 }
 191 
 192 /*!
 193  * \internal
 194  * \brief Unconditionally remove a client from the waitlist, such as when the client
 195  *        node disconnects from the cluster
 196  *
 197  * \param[in] client    The client to remove
 198  */
 199 void
 200 attrd_remove_client_from_waitlist(pcmk__client_t *client)
     /*  */
 201 {
 202     GHashTableIter iter;
 203     gpointer value;
 204 
 205     if (waitlist == NULL) {
 206         return;
 207     }
 208 
 209     g_hash_table_iter_init(&iter, waitlist);
 210 
 211     while (g_hash_table_iter_next(&iter, NULL, &value)) {
 212         struct waitlist_node *wl = (struct waitlist_node *) value;
 213 
 214         if (pcmk__str_eq(wl->client_id, client->id, pcmk__str_none)) {
 215             g_hash_table_iter_remove(&iter);
 216             crm_trace("%d clients now on waitlist", g_hash_table_size(waitlist));
 217         }
 218     }
 219 }
 220 
 221 /*!
 222  * \internal
 223  * \brief Send an IPC ACK message to all awaiting clients
 224  *
 225  * This function will search the waitlist for all clients that are currently awaiting
 226  * an ACK indicating their attrd operation is complete.  Only those clients with a
 227  * matching sync point type and callid from their original XML IPC request will be
 228  * ACKed.  Once they have received an ACK, they will be removed from the waitlist.
 229  *
 230  * \param[in] sync_point What kind of sync point have we hit?
 231  * \param[in] xml        The original XML IPC request.
 232  */
 233 void
 234 attrd_ack_waitlist_clients(enum attrd_sync_point sync_point, const xmlNode *xml)
     /*  */
 235 {
 236     int callid;
 237     gpointer value;
 238 
 239     if (waitlist == NULL) {
 240         return;
 241     }
 242 
 243     if (crm_element_value_int(xml, PCMK__XA_CALL_ID, &callid) == -1) {
 244         crm_warn("Could not get callid from request XML");
 245         return;
 246     }
 247 
 248     value = pcmk__intkey_table_lookup(waitlist, callid);
 249     if (value != NULL) {
 250         struct waitlist_node *wl = (struct waitlist_node *) value;
 251         pcmk__client_t *client = NULL;
 252 
 253         if (wl->sync_point != sync_point) {
 254             return;
 255         }
 256 
 257         crm_notice("Alerting client %s for reached %s sync point",
 258                    wl->client_id, sync_point_str(wl->sync_point));
 259 
 260         client = pcmk__find_client_by_id(wl->client_id);
 261         if (client == NULL) {
 262             return;
 263         }
 264 
 265         attrd_send_ack(client, wl->ipc_id, wl->flags | crm_ipc_client_response);
 266 
 267         /* And then remove the client so it doesn't get alerted again. */
 268         pcmk__intkey_table_remove(waitlist, callid);
 269 
 270         crm_trace("%d clients now on waitlist", g_hash_table_size(waitlist));
 271     }
 272 }
 273 
 274 /*!
 275  * \internal
 276  * \brief Action to take when a cluster sync point is hit for a
 277  *        PCMK__ATTRD_CMD_UPDATE* message.
 278  *
 279  * \param[in] xml  The request that should be passed along to
 280  *                 attrd_ack_waitlist_clients.  This should be the original
 281  *                 IPC request containing the callid for this update message.
 282  */
 283 int
 284 attrd_cluster_sync_point_update(xmlNode *xml)
     /*  */
 285 {
 286     crm_trace("Hit cluster sync point for attribute update");
 287     attrd_ack_waitlist_clients(attrd_sync_point_cluster, xml);
 288     return pcmk_rc_ok;
 289 }
 290 
 291 /*!
 292  * \internal
 293  * \brief Return the sync point attribute for an IPC request
 294  *
 295  * This function will check both the top-level element of \p xml for a sync
 296  * point attribute, as well as all of its \p op children, if any.  The latter
 297  * is useful for newer versions of attrd that can put multiple IPC requests
 298  * into a single message.
 299  *
 300  * \param[in] xml   An XML IPC request
 301  *
 302  * \note It is assumed that if one child element has a sync point attribute,
 303  *       all will have a sync point attribute and they will all be the same
 304  *       sync point.  No other configuration is supported.
 305  *
 306  * \return The sync point attribute of \p xml, or NULL if none.
 307  */
 308 const char *
 309 attrd_request_sync_point(xmlNode *xml)
     /*  */
 310 {
 311     CRM_CHECK(xml != NULL, return NULL);
 312 
 313     if (xml->children != NULL) {
 314         xmlNode *child = pcmk__xe_first_child(xml, PCMK_XE_OP,
 315                                               PCMK__XA_ATTR_SYNC_POINT, NULL);
 316 
 317         if (child) {
 318             return crm_element_value(child, PCMK__XA_ATTR_SYNC_POINT);
 319         } else {
 320             return NULL;
 321         }
 322 
 323     } else {
 324         return crm_element_value(xml, PCMK__XA_ATTR_SYNC_POINT);
 325     }
 326 }
 327 
 328 /*!
 329  * \internal
 330  * \brief Does an IPC request contain any sync point attribute?
 331  *
 332  * \param[in] xml   An XML IPC request
 333  *
 334  * \return true if there's a sync point attribute, false otherwise
 335  */
 336 bool
 337 attrd_request_has_sync_point(xmlNode *xml)
     /*  */
 338 {
 339     return attrd_request_sync_point(xml) != NULL;
 340 }
 341 
 342 static void
 343 free_action(gpointer data)
     /*  */
 344 {
 345     struct confirmation_action *action = (struct confirmation_action *) data;
 346     g_list_free_full(action->respondents, free);
 347     mainloop_timer_del(action->timer);
 348     free_xml(action->xml);
 349     free(action->client_id);
 350     free(action);
 351 }
 352 
 353 /* Remove an IPC request from the expected_confirmations table if the peer attrds
 354  * don't respond before the timeout is hit.  We set the timeout to 15s.  The exact
 355  * number isn't critical - we just want to make sure that the table eventually gets
 356  * cleared of things that didn't complete.
 357  */
 358 static gboolean
 359 confirmation_timeout_cb(gpointer data)
     /*  */
 360 {
 361     struct confirmation_action *action = (struct confirmation_action *) data;
 362 
 363     GHashTableIter iter;
 364     gpointer value;
 365 
 366     if (expected_confirmations == NULL) {
 367         return G_SOURCE_REMOVE;
 368     }
 369 
 370     g_hash_table_iter_init(&iter, expected_confirmations);
 371 
 372     while (g_hash_table_iter_next(&iter, NULL, &value)) {
 373         if (value == action) {
 374             pcmk__client_t *client = pcmk__find_client_by_id(action->client_id);
 375             if (client == NULL) {
 376                 return G_SOURCE_REMOVE;
 377             }
 378 
 379             crm_trace("Timed out waiting for confirmations for client %s", client->id);
 380             pcmk__ipc_send_ack(client, action->ipc_id,
 381                                action->flags|crm_ipc_client_response,
 382                                PCMK__XE_ACK, ATTRD_PROTOCOL_VERSION,
 383                                CRM_EX_TIMEOUT);
 384 
 385             g_hash_table_iter_remove(&iter);
 386             crm_trace("%d requests now in expected confirmations table", g_hash_table_size(expected_confirmations));
 387             break;
 388         }
 389     }
 390 
 391     return G_SOURCE_REMOVE;
 392 }
 393 
 394 /*!
 395  * \internal
 396  * \brief When a peer disconnects from the cluster, no longer wait for its confirmation
 397  *        for any IPC action.  If this peer is the last one being waited on, this will
 398  *        trigger the confirmation action.
 399  *
 400  * \param[in] host   The disconnecting peer attrd's uname
 401  */
 402 void
 403 attrd_do_not_expect_from_peer(const char *host)
     /*  */
 404 {
 405     GList *keys = NULL;
 406 
 407     if (expected_confirmations == NULL) {
 408         return;
 409     }
 410 
 411     keys = g_hash_table_get_keys(expected_confirmations);
 412 
 413     crm_trace("Removing peer %s from expected confirmations", host);
 414 
 415     for (GList *node = keys; node != NULL; node = node->next) {
 416         int callid = *(int *) node->data;
 417         attrd_handle_confirmation(callid, host);
 418     }
 419 
 420     g_list_free(keys);
 421 }
 422 
 423 /*!
 424  * \internal
 425  * \brief When a client disconnects from the cluster, no longer wait on confirmations
 426  *        for it.  Because the peer attrds may still be processing the original IPC
 427  *        message, they may still send us confirmations.  However, we will take no
 428  *        action on them.
 429  *
 430  * \param[in] client    The disconnecting client
 431  */
 432 void
 433 attrd_do_not_wait_for_client(pcmk__client_t *client)
     /*  */
 434 {
 435     GHashTableIter iter;
 436     gpointer value;
 437 
 438     if (expected_confirmations == NULL) {
 439         return;
 440     }
 441 
 442     g_hash_table_iter_init(&iter, expected_confirmations);
 443 
 444     while (g_hash_table_iter_next(&iter, NULL, &value)) {
 445         struct confirmation_action *action = (struct confirmation_action *) value;
 446 
 447         if (pcmk__str_eq(action->client_id, client->id, pcmk__str_none)) {
 448             crm_trace("Removing client %s from expected confirmations", client->id);
 449             g_hash_table_iter_remove(&iter);
 450             crm_trace("%d requests now in expected confirmations table", g_hash_table_size(expected_confirmations));
 451             break;
 452         }
 453     }
 454 }
 455 
 456 /*!
 457  * \internal
 458  * \brief Register some action to be taken when IPC request confirmations are
 459  *        received
 460  *
 461  * When this function is called, a list of all peer attrds that support confirming
 462  * requests is generated.  As confirmations from these peer attrds are received,
 463  * they are removed from this list.  When the list is empty, the registered action
 464  * will be called.
 465  *
 466  * \note This function should always be called before attrd_send_message is called
 467  *       to broadcast to the peers to ensure that we know what replies we are
 468  *       waiting on.  Otherwise, it is possible the peer could finish and confirm
 469  *       before we know to expect it.
 470  *
 471  * \param[in] request The request that is awaiting confirmations
 472  * \param[in] fn      A function to be run after all confirmations are received
 473  */
 474 void
 475 attrd_expect_confirmations(pcmk__request_t *request, attrd_confirmation_action_fn fn)
     /*  */
 476 {
 477     struct confirmation_action *action = NULL;
 478     GHashTableIter iter;
 479     gpointer host, ver;
 480     GList *respondents = NULL;
 481     int callid;
 482 
 483     if (expected_confirmations == NULL) {
 484         expected_confirmations = pcmk__intkey_table((GDestroyNotify) free_action);
 485     }
 486 
 487     if (crm_element_value_int(request->xml, PCMK__XA_CALL_ID, &callid) == -1) {
 488         crm_err("Could not get callid from xml");
 489         return;
 490     }
 491 
 492     if (pcmk__intkey_table_lookup(expected_confirmations, callid)) {
 493         crm_err("Already waiting on confirmations for call id %d", callid);
 494         return;
 495     }
 496 
 497     g_hash_table_iter_init(&iter, peer_protocol_vers);
 498     while (g_hash_table_iter_next(&iter, &host, &ver)) {
 499         if (ATTRD_SUPPORTS_CONFIRMATION(GPOINTER_TO_INT(ver))) {
 500             respondents = g_list_prepend(respondents,
 501                                          pcmk__str_copy((char *) host));
 502         }
 503     }
 504 
 505     action = pcmk__assert_alloc(1, sizeof(struct confirmation_action));
 506 
 507     action->respondents = respondents;
 508     action->fn = fn;
 509     action->xml = pcmk__xml_copy(NULL, request->xml);
 510     action->client_id = pcmk__str_copy(request->ipc_client->id);
 511     action->ipc_id = request->ipc_id;
 512     action->flags = request->flags;
 513 
 514     action->timer = mainloop_timer_add(NULL, 15000, FALSE, confirmation_timeout_cb, action);
 515     mainloop_timer_start(action->timer);
 516 
 517     pcmk__intkey_table_insert(expected_confirmations, callid, action);
 518     crm_trace("Callid %d now waiting on %d confirmations", callid, g_list_length(respondents));
 519     crm_trace("%d requests now in expected confirmations table", g_hash_table_size(expected_confirmations));
 520 }
 521 
 522 void
 523 attrd_free_confirmations(void)
     /*  */
 524 {
 525     if (expected_confirmations != NULL) {
 526         g_hash_table_destroy(expected_confirmations);
 527         expected_confirmations = NULL;
 528     }
 529 }
 530 
 531 /*!
 532  * \internal
 533  * \brief Process a confirmation message from a peer attrd
 534  *
 535  * This function is called every time a PCMK__ATTRD_CMD_CONFIRM message is
 536  * received from a peer attrd.  If this is the last confirmation we are waiting
 537  * on for a given operation, the registered action will be called.
 538  *
 539  * \param[in] callid The unique callid for the XML IPC request
 540  * \param[in] host   The confirming peer attrd's uname
 541  */
 542 void
 543 attrd_handle_confirmation(int callid, const char *host)
     /*  */
 544 {
 545     struct confirmation_action *action = NULL;
 546     GList *node = NULL;
 547 
 548     if (expected_confirmations == NULL) {
 549         return;
 550     }
 551 
 552     action = pcmk__intkey_table_lookup(expected_confirmations, callid);
 553     if (action == NULL) {
 554         return;
 555     }
 556 
 557     node = g_list_find_custom(action->respondents, host, (GCompareFunc) strcasecmp);
 558 
 559     if (node == NULL) {
 560         return;
 561     }
 562 
 563     action->respondents = g_list_remove(action->respondents, node->data);
 564     crm_trace("Callid %d now waiting on %d confirmations", callid, g_list_length(action->respondents));
 565 
 566     if (action->respondents == NULL) {
 567         action->fn(action->xml);
 568         pcmk__intkey_table_remove(expected_confirmations, callid);
 569         crm_trace("%d requests now in expected confirmations table", g_hash_table_size(expected_confirmations));
 570     }
 571 }

/* */