drivers/staging/lustre/lustre/ptlrpc/ptlrpcd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/ptlrpc/ptlrpcd.c
  37  */
  38
  39 /** \defgroup ptlrpcd PortalRPC daemon
  40  *
  41  * ptlrpcd is a special thread with its own set where other user might add
  42  * requests when they don't want to wait for their completion.
  43  * PtlRPCD will take care of sending such requests and then processing their
  44  * replies and calling completion callbacks as necessary.
  45  * The callbacks are called directly from ptlrpcd context.
  46  * It is important to never significantly block (esp. on RPCs!) within such
  47  * completion handler or a deadlock might occur where ptlrpcd enters some
  48  * callback that attempts to send another RPC and wait for it to return,
  49  * during which time ptlrpcd is completely blocked, so e.g. if import
  50  * fails, recovery cannot progress because connection requests are also
  51  * sent by ptlrpcd.
  52  *
  53  * @{
  54  */
  55
  56 #define DEBUG_SUBSYSTEM S_RPC
  57
  58 # include <linux/libcfs/libcfs.h>
  59
  60 #include <lustre_net.h>
  61 # include <lustre_lib.h>
  62
  63 #include <lustre_ha.h>
  64 #include <obd_class.h>   /* for obd_zombie */
  65 #include <obd_support.h> /* for OBD_FAIL_CHECK */
  66 #include <cl_object.h> /* cl_env_{get,put}() */
  67 #include <lprocfs_status.h>
  68
  69 #include "ptlrpc_internal.h"
  70
  71 struct ptlrpcd {
  72         int             pd_size;
  73         int             pd_index;
  74         int             pd_nthreads;
  75         struct ptlrpcd_ctl pd_thread_rcv;
  76         struct ptlrpcd_ctl pd_threads[0];
  77 };
  78
  79 static int max_ptlrpcds;
  80 CFS_MODULE_PARM(max_ptlrpcds, "i", int, 0644,
  81                 "Max ptlrpcd thread count to be started.");
  82
  83 static int ptlrpcd_bind_policy = PDB_POLICY_PAIR;
  84 CFS_MODULE_PARM(ptlrpcd_bind_policy, "i", int, 0644,
  85                 "Ptlrpcd threads binding mode.");
  86 static struct ptlrpcd *ptlrpcds;
  87
  88 struct mutex ptlrpcd_mutex;
  89 static int ptlrpcd_users = 0;
  90
  91 void ptlrpcd_wake(struct ptlrpc_request *req)
  92 {
  93         struct ptlrpc_request_set *rq_set = req->rq_set;
  94
  95         LASSERT(rq_set != NULL);
  96
  97         wake_up(&rq_set->set_waitq);
  98 }
  99 EXPORT_SYMBOL(ptlrpcd_wake);
 100
 101 static struct ptlrpcd_ctl *
 102 ptlrpcd_select_pc(struct ptlrpc_request *req, pdl_policy_t policy, int index)
 103 {
 104         int idx = 0;
 105
 106         if (req != NULL && req->rq_send_state != LUSTRE_IMP_FULL)
 107                 return &ptlrpcds->pd_thread_rcv;
 108
 109         switch (policy) {
 110         case PDL_POLICY_SAME:
 111                 idx = smp_processor_id() % ptlrpcds->pd_nthreads;
 112                 break;
 113         case PDL_POLICY_LOCAL:
 114                 /* Before CPU partition patches available, process it the same
 115                  * as "PDL_POLICY_ROUND". */
 116 # ifdef CFS_CPU_MODE_NUMA
 117 # warning "fix this code to use new CPU partition APIs"
 118 # endif
 119                 /* Fall through to PDL_POLICY_ROUND until the CPU
 120                  * CPU partition patches are available. */
 121                 index = -1;
 122         case PDL_POLICY_PREFERRED:
 123                 if (index >= 0 && index < num_online_cpus()) {
 124                         idx = index % ptlrpcds->pd_nthreads;
 125                         break;
 126                 }
 127                 /* Fall through to PDL_POLICY_ROUND for bad index. */
 128         default:
 129                 /* Fall through to PDL_POLICY_ROUND for unknown policy. */
 130         case PDL_POLICY_ROUND:
 131                 /* We do not care whether it is strict load balance. */
 132                 idx = ptlrpcds->pd_index + 1;
 133                 if (idx == smp_processor_id())
 134                         idx++;
 135                 idx %= ptlrpcds->pd_nthreads;
 136                 ptlrpcds->pd_index = idx;
 137                 break;
 138         }
 139
 140         return &ptlrpcds->pd_threads[idx];
 141 }
 142
 143 /**
 144  * Move all request from an existing request set to the ptlrpcd queue.
 145  * All requests from the set must be in phase RQ_PHASE_NEW.
 146  */
 147 void ptlrpcd_add_rqset(struct ptlrpc_request_set *set)
 148 {
 149         struct list_head *tmp, *pos;
 150         struct ptlrpcd_ctl *pc;
 151         struct ptlrpc_request_set *new;
 152         int count, i;
 153
 154         pc = ptlrpcd_select_pc(NULL, PDL_POLICY_LOCAL, -1);
 155         new = pc->pc_set;
 156
 157         list_for_each_safe(pos, tmp, &set->set_requests) {
 158                 struct ptlrpc_request *req =
 159                         list_entry(pos, struct ptlrpc_request,
 160                                        rq_set_chain);
 161
 162                 LASSERT(req->rq_phase == RQ_PHASE_NEW);
 163                 req->rq_set = new;
 164                 req->rq_queued_time = cfs_time_current();
 165         }
 166
 167         spin_lock(&new->set_new_req_lock);
 168         list_splice_init(&set->set_requests, &new->set_new_requests);
 169         i = atomic_read(&set->set_remaining);
 170         count = atomic_add_return(i, &new->set_new_count);
 171         atomic_set(&set->set_remaining, 0);
 172         spin_unlock(&new->set_new_req_lock);
 173         if (count == i) {
 174                 wake_up(&new->set_waitq);
 175
 176                 /* XXX: It maybe unnecessary to wakeup all the partners. But to
 177                  *      guarantee the async RPC can be processed ASAP, we have
 178                  *      no other better choice. It maybe fixed in future. */
 179                 for (i = 0; i < pc->pc_npartners; i++)
 180                         wake_up(&pc->pc_partners[i]->pc_set->set_waitq);
 181         }
 182 }
 183 EXPORT_SYMBOL(ptlrpcd_add_rqset);
 184
 185 /**
 186  * Return transferred RPCs count.
 187  */
 188 static int ptlrpcd_steal_rqset(struct ptlrpc_request_set *des,
 189                                struct ptlrpc_request_set *src)
 190 {
 191         struct list_head *tmp, *pos;
 192         struct ptlrpc_request *req;
 193         int rc = 0;
 194
 195         spin_lock(&src->set_new_req_lock);
 196         if (likely(!list_empty(&src->set_new_requests))) {
 197                 list_for_each_safe(pos, tmp, &src->set_new_requests) {
 198                         req = list_entry(pos, struct ptlrpc_request,
 199                                              rq_set_chain);
 200                         req->rq_set = des;
 201                 }
 202                 list_splice_init(&src->set_new_requests,
 203                                      &des->set_requests);
 204                 rc = atomic_read(&src->set_new_count);
 205                 atomic_add(rc, &des->set_remaining);
 206                 atomic_set(&src->set_new_count, 0);
 207         }
 208         spin_unlock(&src->set_new_req_lock);
 209         return rc;
 210 }
 211
 212 /**
 213  * Requests that are added to the ptlrpcd queue are sent via
 214  * ptlrpcd_check->ptlrpc_check_set().
 215  */
 216 void ptlrpcd_add_req(struct ptlrpc_request *req, pdl_policy_t policy, int idx)
 217 {
 218         struct ptlrpcd_ctl *pc;
 219
 220         if (req->rq_reqmsg)
 221                 lustre_msg_set_jobid(req->rq_reqmsg, NULL);
 222
 223         spin_lock(&req->rq_lock);
 224         if (req->rq_invalid_rqset) {
 225                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(5),
 226                                                      back_to_sleep, NULL);
 227
 228                 req->rq_invalid_rqset = 0;
 229                 spin_unlock(&req->rq_lock);
 230                 l_wait_event(req->rq_set_waitq, (req->rq_set == NULL), &lwi);
 231         } else if (req->rq_set) {
 232                 /* If we have a vaid "rq_set", just reuse it to avoid double
 233                  * linked. */
 234                 LASSERT(req->rq_phase == RQ_PHASE_NEW);
 235                 LASSERT(req->rq_send_state == LUSTRE_IMP_REPLAY);
 236
 237                 /* ptlrpc_check_set will decrease the count */
 238                 atomic_inc(&req->rq_set->set_remaining);
 239                 spin_unlock(&req->rq_lock);
 240                 wake_up(&req->rq_set->set_waitq);
 241                 return;
 242         } else {
 243                 spin_unlock(&req->rq_lock);
 244         }
 245
 246         pc = ptlrpcd_select_pc(req, policy, idx);
 247
 248         DEBUG_REQ(D_INFO, req, "add req [%p] to pc [%s:%d]",
 249                   req, pc->pc_name, pc->pc_index);
 250
 251         ptlrpc_set_add_new_req(pc, req);
 252 }
 253 EXPORT_SYMBOL(ptlrpcd_add_req);
 254
 255 static inline void ptlrpc_reqset_get(struct ptlrpc_request_set *set)
 256 {
 257         atomic_inc(&set->set_refcount);
 258 }
 259
 260 /**
 261  * Check if there is more work to do on ptlrpcd set.
 262  * Returns 1 if yes.
 263  */
 264 static int ptlrpcd_check(struct lu_env *env, struct ptlrpcd_ctl *pc)
 265 {
 266         struct list_head *tmp, *pos;
 267         struct ptlrpc_request *req;
 268         struct ptlrpc_request_set *set = pc->pc_set;
 269         int rc = 0;
 270         int rc2;
 271
 272         if (atomic_read(&set->set_new_count)) {
 273                 spin_lock(&set->set_new_req_lock);
 274                 if (likely(!list_empty(&set->set_new_requests))) {
 275                         list_splice_init(&set->set_new_requests,
 276                                              &set->set_requests);
 277                         atomic_add(atomic_read(&set->set_new_count),
 278                                        &set->set_remaining);
 279                         atomic_set(&set->set_new_count, 0);
 280                         /*
 281                          * Need to calculate its timeout.
 282                          */
 283                         rc = 1;
 284                 }
 285                 spin_unlock(&set->set_new_req_lock);
 286         }
 287
 288         /* We should call lu_env_refill() before handling new requests to make
 289          * sure that env key the requests depending on really exists.
 290          */
 291         rc2 = lu_env_refill(env);
 292         if (rc2 != 0) {
 293                 /*
 294                  * XXX This is very awkward situation, because
 295                  * execution can neither continue (request
 296                  * interpreters assume that env is set up), nor repeat
 297                  * the loop (as this potentially results in a tight
 298                  * loop of -ENOMEM's).
 299                  *
 300                  * Fortunately, refill only ever does something when
 301                  * new modules are loaded, i.e., early during boot up.
 302                  */
 303                 CERROR("Failure to refill session: %d\n", rc2);
 304                 return rc;
 305         }
 306
 307         if (atomic_read(&set->set_remaining))
 308                 rc |= ptlrpc_check_set(env, set);
 309
 310         if (!list_empty(&set->set_requests)) {
 311                 /*
 312                  * XXX: our set never completes, so we prune the completed
 313                  * reqs after each iteration. boy could this be smarter.
 314                  */
 315                 list_for_each_safe(pos, tmp, &set->set_requests) {
 316                         req = list_entry(pos, struct ptlrpc_request,
 317                                              rq_set_chain);
 318                         if (req->rq_phase != RQ_PHASE_COMPLETE)
 319                                 continue;
 320
 321                         list_del_init(&req->rq_set_chain);
 322                         req->rq_set = NULL;
 323                         ptlrpc_req_finished(req);
 324                 }
 325         }
 326
 327         if (rc == 0) {
 328                 /*
 329                  * If new requests have been added, make sure to wake up.
 330                  */
 331                 rc = atomic_read(&set->set_new_count);
 332
 333                 /* If we have nothing to do, check whether we can take some
 334                  * work from our partner threads. */
 335                 if (rc == 0 && pc->pc_npartners > 0) {
 336                         struct ptlrpcd_ctl *partner;
 337                         struct ptlrpc_request_set *ps;
 338                         int first = pc->pc_cursor;
 339
 340                         do {
 341                                 partner = pc->pc_partners[pc->pc_cursor++];
 342                                 if (pc->pc_cursor >= pc->pc_npartners)
 343                                         pc->pc_cursor = 0;
 344                                 if (partner == NULL)
 345                                         continue;
 346
 347                                 spin_lock(&partner->pc_lock);
 348                                 ps = partner->pc_set;
 349                                 if (ps == NULL) {
 350                                         spin_unlock(&partner->pc_lock);
 351                                         continue;
 352                                 }
 353
 354                                 ptlrpc_reqset_get(ps);
 355                                 spin_unlock(&partner->pc_lock);
 356
 357                                 if (atomic_read(&ps->set_new_count)) {
 358                                         rc = ptlrpcd_steal_rqset(set, ps);
 359                                         if (rc > 0)
 360                                                 CDEBUG(D_RPCTRACE, "transfer %d"
 361                                                        " async RPCs [%d->%d]\n",
 362                                                         rc, partner->pc_index,
 363                                                         pc->pc_index);
 364                                 }
 365                                 ptlrpc_reqset_put(ps);
 366                         } while (rc == 0 && pc->pc_cursor != first);
 367                 }
 368         }
 369
 370         return rc;
 371 }
 372
 373 /**
 374  * Main ptlrpcd thread.
 375  * ptlrpc's code paths like to execute in process context, so we have this
 376  * thread which spins on a set which contains the rpcs and sends them.
 377  *
 378  */
 379 static int ptlrpcd(void *arg)
 380 {
 381         struct ptlrpcd_ctl *pc = arg;
 382         struct ptlrpc_request_set *set = pc->pc_set;
 383         struct lu_env env = { .le_ses = NULL };
 384         int rc, exit = 0;
 385
 386         unshare_fs_struct();
 387 #if defined(CONFIG_SMP)
 388         if (test_bit(LIOD_BIND, &pc->pc_flags)) {
 389                 int index = pc->pc_index;
 390
 391                 if (index >= 0 && index < num_possible_cpus()) {
 392                         while (!cpu_online(index)) {
 393                                 if (++index >= num_possible_cpus())
 394                                         index = 0;
 395                         }
 396                         set_cpus_allowed_ptr(current,
 397                                         cpumask_of_node(cpu_to_node(index)));
 398                 }
 399         }
 400 #endif
 401         /*
 402          * XXX So far only "client" ptlrpcd uses an environment. In
 403          * the future, ptlrpcd thread (or a thread-set) has to given
 404          * an argument, describing its "scope".
 405          */
 406         rc = lu_context_init(&env.le_ctx,
 407                              LCT_CL_THREAD|LCT_REMEMBER|LCT_NOREF);
 408         complete(&pc->pc_starting);
 409
 410         if (rc != 0)
 411                 return rc;
 412
 413         /*
 414          * This mainloop strongly resembles ptlrpc_set_wait() except that our
 415          * set never completes.  ptlrpcd_check() calls ptlrpc_check_set() when
 416          * there are requests in the set. New requests come in on the set's
 417          * new_req_list and ptlrpcd_check() moves them into the set.
 418          */
 419         do {
 420                 struct l_wait_info lwi;
 421                 int timeout;
 422
 423                 timeout = ptlrpc_set_next_timeout(set);
 424                 lwi = LWI_TIMEOUT(cfs_time_seconds(timeout ? timeout : 1),
 425                                   ptlrpc_expired_set, set);
 426
 427                 lu_context_enter(&env.le_ctx);
 428                 l_wait_event(set->set_waitq,
 429                              ptlrpcd_check(&env, pc), &lwi);
 430                 lu_context_exit(&env.le_ctx);
 431
 432                 /*
 433                  * Abort inflight rpcs for forced stop case.
 434                  */
 435                 if (test_bit(LIOD_STOP, &pc->pc_flags)) {
 436                         if (test_bit(LIOD_FORCE, &pc->pc_flags))
 437                                 ptlrpc_abort_set(set);
 438                         exit++;
 439                 }
 440
 441                 /*
 442                  * Let's make one more loop to make sure that ptlrpcd_check()
 443                  * copied all raced new rpcs into the set so we can kill them.
 444                  */
 445         } while (exit < 2);
 446
 447         /*
 448          * Wait for inflight requests to drain.
 449          */
 450         if (!list_empty(&set->set_requests))
 451                 ptlrpc_set_wait(set);
 452         lu_context_fini(&env.le_ctx);
 453
 454         complete(&pc->pc_finishing);
 455
 456         return 0;
 457 }
 458
 459 /* XXX: We want multiple CPU cores to share the async RPC load. So we start many
 460  *      ptlrpcd threads. We also want to reduce the ptlrpcd overhead caused by
 461  *      data transfer cross-CPU cores. So we bind ptlrpcd thread to specified
 462  *      CPU core. But binding all ptlrpcd threads maybe cause response delay
 463  *      because of some CPU core(s) busy with other loads.
 464  *
 465  *      For example: "ls -l", some async RPCs for statahead are assigned to
 466  *      ptlrpcd_0, and ptlrpcd_0 is bound to CPU_0, but CPU_0 may be quite busy
 467  *      with other non-ptlrpcd, like "ls -l" itself (we want to the "ls -l"
 468  *      thread, statahead thread, and ptlrpcd thread can run in parallel), under
 469  *      such case, the statahead async RPCs can not be processed in time, it is
 470  *      unexpected. If ptlrpcd_0 can be re-scheduled on other CPU core, it may
 471  *      be better. But it breaks former data transfer policy.
 472  *
 473  *      So we shouldn't be blind for avoiding the data transfer. We make some
 474  *      compromise: divide the ptlrpcd threds pool into two parts. One part is
 475  *      for bound mode, each ptlrpcd thread in this part is bound to some CPU
 476  *      core. The other part is for free mode, all the ptlrpcd threads in the
 477  *      part can be scheduled on any CPU core. We specify some partnership
 478  *      between bound mode ptlrpcd thread(s) and free mode ptlrpcd thread(s),
 479  *      and the async RPC load within the partners are shared.
 480  *
 481  *      It can partly avoid data transfer cross-CPU (if the bound mode ptlrpcd
 482  *      thread can be scheduled in time), and try to guarantee the async RPC
 483  *      processed ASAP (as long as the free mode ptlrpcd thread can be scheduled
 484  *      on any CPU core).
 485  *
 486  *      As for how to specify the partnership between bound mode ptlrpcd
 487  *      thread(s) and free mode ptlrpcd thread(s), the simplest way is to use
 488  *      <free bound> pair. In future, we can specify some more complex
 489  *      partnership based on the patches for CPU partition. But before such
 490  *      patches are available, we prefer to use the simplest one.
 491  */
 492 # ifdef CFS_CPU_MODE_NUMA
 493 # warning "fix ptlrpcd_bind() to use new CPU partition APIs"
 494 # endif
 495 static int ptlrpcd_bind(int index, int max)
 496 {
 497         struct ptlrpcd_ctl *pc;
 498         int rc = 0;
 499 #if defined(CONFIG_NUMA)
 500         cpumask_t mask;
 501 #endif
 502
 503         LASSERT(index <= max - 1);
 504         pc = &ptlrpcds->pd_threads[index];
 505         switch (ptlrpcd_bind_policy) {
 506         case PDB_POLICY_NONE:
 507                 pc->pc_npartners = -1;
 508                 break;
 509         case PDB_POLICY_FULL:
 510                 pc->pc_npartners = 0;
 511                 set_bit(LIOD_BIND, &pc->pc_flags);
 512                 break;
 513         case PDB_POLICY_PAIR:
 514                 LASSERT(max % 2 == 0);
 515                 pc->pc_npartners = 1;
 516                 break;
 517         case PDB_POLICY_NEIGHBOR:
 518 #if defined(CONFIG_NUMA)
 519         {
 520                 int i;
 521                 mask = *cpumask_of_node(cpu_to_node(index));
 522                 for (i = max; i < num_online_cpus(); i++)
 523                         cpu_clear(i, mask);
 524                 pc->pc_npartners = cpus_weight(mask) - 1;
 525                 set_bit(LIOD_BIND, &pc->pc_flags);
 526         }
 527 #else
 528                 LASSERT(max >= 3);
 529                 pc->pc_npartners = 2;
 530 #endif
 531                 break;
 532         default:
 533                 CERROR("unknown ptlrpcd bind policy %d\n", ptlrpcd_bind_policy);
 534                 rc = -EINVAL;
 535         }
 536
 537         if (rc == 0 && pc->pc_npartners > 0) {
 538                 OBD_ALLOC(pc->pc_partners,
 539                           sizeof(struct ptlrpcd_ctl *) * pc->pc_npartners);
 540                 if (pc->pc_partners == NULL) {
 541                         pc->pc_npartners = 0;
 542                         rc = -ENOMEM;
 543                 } else {
 544                         switch (ptlrpcd_bind_policy) {
 545                         case PDB_POLICY_PAIR:
 546                                 if (index & 0x1) {
 547                                         set_bit(LIOD_BIND, &pc->pc_flags);
 548                                         pc->pc_partners[0] = &ptlrpcds->
 549                                                 pd_threads[index - 1];
 550                                         ptlrpcds->pd_threads[index - 1].
 551                                                 pc_partners[0] = pc;
 552                                 }
 553                                 break;
 554                         case PDB_POLICY_NEIGHBOR:
 555 #if defined(CONFIG_NUMA)
 556                         {
 557                                 struct ptlrpcd_ctl *ppc;
 558                                 int i, pidx;
 559                                 /* partners are cores in the same NUMA node.
 560                                  * setup partnership only with ptlrpcd threads
 561                                  * that are already initialized
 562                                  */
 563                                 for (pidx = 0, i = 0; i < index; i++) {
 564                                         if (cpu_isset(i, mask)) {
 565                                                 ppc = &ptlrpcds->pd_threads[i];
 566                                                 pc->pc_partners[pidx++] = ppc;
 567                                                 ppc->pc_partners[ppc->
 568                                                           pc_npartners++] = pc;
 569                                         }
 570                                 }
 571                                 /* adjust number of partners to the number
 572                                  * of partnership really setup */
 573                                 pc->pc_npartners = pidx;
 574                         }
 575 #else
 576                                 if (index & 0x1)
 577                                         set_bit(LIOD_BIND, &pc->pc_flags);
 578                                 if (index > 0) {
 579                                         pc->pc_partners[0] = &ptlrpcds->
 580                                                 pd_threads[index - 1];
 581                                         ptlrpcds->pd_threads[index - 1].
 582                                                 pc_partners[1] = pc;
 583                                         if (index == max - 1) {
 584                                                 pc->pc_partners[1] =
 585                                                 &ptlrpcds->pd_threads[0];
 586                                                 ptlrpcds->pd_threads[0].
 587                                                 pc_partners[0] = pc;
 588                                         }
 589                                 }
 590 #endif
 591                                 break;
 592                         }
 593                 }
 594         }
 595
 596         return rc;
 597 }
 598
 599
 600 int ptlrpcd_start(int index, int max, const char *name, struct ptlrpcd_ctl *pc)
 601 {
 602         int rc;
 603         int env = 0;
 604
 605         /*
 606          * Do not allow start second thread for one pc.
 607          */
 608         if (test_and_set_bit(LIOD_START, &pc->pc_flags)) {
 609                 CWARN("Starting second thread (%s) for same pc %p\n",
 610                       name, pc);
 611                 return 0;
 612         }
 613
 614         pc->pc_index = index;
 615         init_completion(&pc->pc_starting);
 616         init_completion(&pc->pc_finishing);
 617         spin_lock_init(&pc->pc_lock);
 618         strlcpy(pc->pc_name, name, sizeof(pc->pc_name));
 619         pc->pc_set = ptlrpc_prep_set();
 620         if (pc->pc_set == NULL)
 621                 GOTO(out, rc = -ENOMEM);
 622         /*
 623          * So far only "client" ptlrpcd uses an environment. In the future,
 624          * ptlrpcd thread (or a thread-set) has to be given an argument,
 625          * describing its "scope".
 626          */
 627         rc = lu_context_init(&pc->pc_env.le_ctx, LCT_CL_THREAD|LCT_REMEMBER);
 628         if (rc != 0)
 629                 GOTO(out, rc);
 630
 631         env = 1;
 632         {
 633                 struct task_struct *task;
 634
 635                 if (index >= 0) {
 636                         rc = ptlrpcd_bind(index, max);
 637                         if (rc < 0)
 638                                 GOTO(out, rc);
 639                 }
 640
 641                 task = kthread_run(ptlrpcd, pc, "%s", pc->pc_name);
 642                 if (IS_ERR(task))
 643                         GOTO(out, rc = PTR_ERR(task));
 644
 645                 rc = 0;
 646                 wait_for_completion(&pc->pc_starting);
 647         }
 648 out:
 649         if (rc) {
 650                 if (pc->pc_set != NULL) {
 651                         struct ptlrpc_request_set *set = pc->pc_set;
 652
 653                         spin_lock(&pc->pc_lock);
 654                         pc->pc_set = NULL;
 655                         spin_unlock(&pc->pc_lock);
 656                         ptlrpc_set_destroy(set);
 657                 }
 658                 if (env != 0)
 659                         lu_context_fini(&pc->pc_env.le_ctx);
 660                 clear_bit(LIOD_BIND, &pc->pc_flags);
 661                 clear_bit(LIOD_START, &pc->pc_flags);
 662         }
 663         return rc;
 664 }
 665
 666 void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force)
 667 {
 668         if (!test_bit(LIOD_START, &pc->pc_flags)) {
 669                 CWARN("Thread for pc %p was not started\n", pc);
 670                 return;
 671         }
 672
 673         set_bit(LIOD_STOP, &pc->pc_flags);
 674         if (force)
 675                 set_bit(LIOD_FORCE, &pc->pc_flags);
 676         wake_up(&pc->pc_set->set_waitq);
 677 }
 678
 679 void ptlrpcd_free(struct ptlrpcd_ctl *pc)
 680 {
 681         struct ptlrpc_request_set *set = pc->pc_set;
 682
 683         if (!test_bit(LIOD_START, &pc->pc_flags)) {
 684                 CWARN("Thread for pc %p was not started\n", pc);
 685                 goto out;
 686         }
 687
 688         wait_for_completion(&pc->pc_finishing);
 689         lu_context_fini(&pc->pc_env.le_ctx);
 690
 691         spin_lock(&pc->pc_lock);
 692         pc->pc_set = NULL;
 693         spin_unlock(&pc->pc_lock);
 694         ptlrpc_set_destroy(set);
 695
 696         clear_bit(LIOD_START, &pc->pc_flags);
 697         clear_bit(LIOD_STOP, &pc->pc_flags);
 698         clear_bit(LIOD_FORCE, &pc->pc_flags);
 699         clear_bit(LIOD_BIND, &pc->pc_flags);
 700
 701 out:
 702         if (pc->pc_npartners > 0) {
 703                 LASSERT(pc->pc_partners != NULL);
 704
 705                 OBD_FREE(pc->pc_partners,
 706                          sizeof(struct ptlrpcd_ctl *) * pc->pc_npartners);
 707                 pc->pc_partners = NULL;
 708         }
 709         pc->pc_npartners = 0;
 710 }
 711
 712 static void ptlrpcd_fini(void)
 713 {
 714         int i;
 715
 716         if (ptlrpcds != NULL) {
 717                 for (i = 0; i < ptlrpcds->pd_nthreads; i++)
 718                         ptlrpcd_stop(&ptlrpcds->pd_threads[i], 0);
 719                 for (i = 0; i < ptlrpcds->pd_nthreads; i++)
 720                         ptlrpcd_free(&ptlrpcds->pd_threads[i]);
 721                 ptlrpcd_stop(&ptlrpcds->pd_thread_rcv, 0);
 722                 ptlrpcd_free(&ptlrpcds->pd_thread_rcv);
 723                 OBD_FREE(ptlrpcds, ptlrpcds->pd_size);
 724                 ptlrpcds = NULL;
 725         }
 726 }
 727
 728 static int ptlrpcd_init(void)
 729 {
 730         int nthreads = num_online_cpus();
 731         char name[16];
 732         int size, i = -1, j, rc = 0;
 733
 734         if (max_ptlrpcds > 0 && max_ptlrpcds < nthreads)
 735                 nthreads = max_ptlrpcds;
 736         if (nthreads < 2)
 737                 nthreads = 2;
 738         if (nthreads < 3 && ptlrpcd_bind_policy == PDB_POLICY_NEIGHBOR)
 739                 ptlrpcd_bind_policy = PDB_POLICY_PAIR;
 740         else if (nthreads % 2 != 0 && ptlrpcd_bind_policy == PDB_POLICY_PAIR)
 741                 nthreads &= ~1; /* make sure it is even */
 742
 743         size = offsetof(struct ptlrpcd, pd_threads[nthreads]);
 744         OBD_ALLOC(ptlrpcds, size);
 745         if (ptlrpcds == NULL)
 746                 GOTO(out, rc = -ENOMEM);
 747
 748         snprintf(name, sizeof(name), "ptlrpcd_rcv");
 749         set_bit(LIOD_RECOVERY, &ptlrpcds->pd_thread_rcv.pc_flags);
 750         rc = ptlrpcd_start(-1, nthreads, name, &ptlrpcds->pd_thread_rcv);
 751         if (rc < 0)
 752                 GOTO(out, rc);
 753
 754         /* XXX: We start nthreads ptlrpc daemons. Each of them can process any
 755          *      non-recovery async RPC to improve overall async RPC efficiency.
 756          *
 757          *      But there are some issues with async I/O RPCs and async non-I/O
 758          *      RPCs processed in the same set under some cases. The ptlrpcd may
 759          *      be blocked by some async I/O RPC(s), then will cause other async
 760          *      non-I/O RPC(s) can not be processed in time.
 761          *
 762          *      Maybe we should distinguish blocked async RPCs from non-blocked
 763          *      async RPCs, and process them in different ptlrpcd sets to avoid
 764          *      unnecessary dependency. But how to distribute async RPCs load
 765          *      among all the ptlrpc daemons becomes another trouble. */
 766         for (i = 0; i < nthreads; i++) {
 767                 snprintf(name, sizeof(name), "ptlrpcd_%d", i);
 768                 rc = ptlrpcd_start(i, nthreads, name, &ptlrpcds->pd_threads[i]);
 769                 if (rc < 0)
 770                         GOTO(out, rc);
 771         }
 772
 773         ptlrpcds->pd_size = size;
 774         ptlrpcds->pd_index = 0;
 775         ptlrpcds->pd_nthreads = nthreads;
 776
 777 out:
 778         if (rc != 0 && ptlrpcds != NULL) {
 779                 for (j = 0; j <= i; j++)
 780                         ptlrpcd_stop(&ptlrpcds->pd_threads[j], 0);
 781                 for (j = 0; j <= i; j++)
 782                         ptlrpcd_free(&ptlrpcds->pd_threads[j]);
 783                 ptlrpcd_stop(&ptlrpcds->pd_thread_rcv, 0);
 784                 ptlrpcd_free(&ptlrpcds->pd_thread_rcv);
 785                 OBD_FREE(ptlrpcds, size);
 786                 ptlrpcds = NULL;
 787         }
 788
 789         return 0;
 790 }
 791
 792 int ptlrpcd_addref(void)
 793 {
 794         int rc = 0;
 795
 796         mutex_lock(&ptlrpcd_mutex);
 797         if (++ptlrpcd_users == 1)
 798                 rc = ptlrpcd_init();
 799         mutex_unlock(&ptlrpcd_mutex);
 800         return rc;
 801 }
 802 EXPORT_SYMBOL(ptlrpcd_addref);
 803
 804 void ptlrpcd_decref(void)
 805 {
 806         mutex_lock(&ptlrpcd_mutex);
 807         if (--ptlrpcd_users == 0)
 808                 ptlrpcd_fini();
 809         mutex_unlock(&ptlrpcd_mutex);
 810 }
 811 EXPORT_SYMBOL(ptlrpcd_decref);
 812 /** @} ptlrpcd */