drivers/block/drbd/drbd_req.c

   1 /*
   2    drbd_req.c
   3
   4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
   5
   6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
   9
  10    drbd is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 2, or (at your option)
  13    any later version.
  14
  15    drbd is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with drbd; see the file COPYING.  If not, write to
  22    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  23
  24  */
  25
  26 #include <linux/module.h>
  27
  28 #include <linux/slab.h>
  29 #include <linux/drbd.h>
  30 #include "drbd_int.h"
  31 #include "drbd_req.h"
  32
  33
  34 static bool drbd_may_do_local_read(struct drbd_conf *mdev, sector_t sector, int size);
  35
  36 /* Update disk stats at start of I/O request */
  37 static void _drbd_start_io_acct(struct drbd_conf *mdev, struct drbd_request *req, struct bio *bio)
  38 {
  39         const int rw = bio_data_dir(bio);
  40         int cpu;
  41         cpu = part_stat_lock();
  42         part_round_stats(cpu, &mdev->vdisk->part0);
  43         part_stat_inc(cpu, &mdev->vdisk->part0, ios[rw]);
  44         part_stat_add(cpu, &mdev->vdisk->part0, sectors[rw], bio_sectors(bio));
  45         (void) cpu; /* The macro invocations above want the cpu argument, I do not like
  46                        the compiler warning about cpu only assigned but never used... */
  47         part_inc_in_flight(&mdev->vdisk->part0, rw);
  48         part_stat_unlock();
  49 }
  50
  51 /* Update disk stats when completing request upwards */
  52 static void _drbd_end_io_acct(struct drbd_conf *mdev, struct drbd_request *req)
  53 {
  54         int rw = bio_data_dir(req->master_bio);
  55         unsigned long duration = jiffies - req->start_time;
  56         int cpu;
  57         cpu = part_stat_lock();
  58         part_stat_add(cpu, &mdev->vdisk->part0, ticks[rw], duration);
  59         part_round_stats(cpu, &mdev->vdisk->part0);
  60         part_dec_in_flight(&mdev->vdisk->part0, rw);
  61         part_stat_unlock();
  62 }
  63
  64 static struct drbd_request *drbd_req_new(struct drbd_conf *mdev,
  65                                                struct bio *bio_src)
  66 {
  67         struct drbd_request *req;
  68
  69         req = mempool_alloc(drbd_request_mempool, GFP_NOIO);
  70         if (!req)
  71                 return NULL;
  72
  73         drbd_req_make_private_bio(req, bio_src);
  74         req->rq_state    = bio_data_dir(bio_src) == WRITE ? RQ_WRITE : 0;
  75         req->w.mdev      = mdev;
  76         req->master_bio  = bio_src;
  77         req->epoch       = 0;
  78
  79         drbd_clear_interval(&req->i);
  80         req->i.sector     = bio_src->bi_sector;
  81         req->i.size      = bio_src->bi_size;
  82         req->i.local = true;
  83         req->i.waiting = false;
  84
  85         INIT_LIST_HEAD(&req->tl_requests);
  86         INIT_LIST_HEAD(&req->w.list);
  87
  88         return req;
  89 }
  90
  91 static void drbd_req_free(struct drbd_request *req)
  92 {
  93         mempool_free(req, drbd_request_mempool);
  94 }
  95
  96 /* rw is bio_data_dir(), only READ or WRITE */
  97 static void _req_is_done(struct drbd_conf *mdev, struct drbd_request *req, const int rw)
  98 {
  99         const unsigned long s = req->rq_state;
 100
 101         /* remove it from the transfer log.
 102          * well, only if it had been there in the first
 103          * place... if it had not (local only or conflicting
 104          * and never sent), it should still be "empty" as
 105          * initialized in drbd_req_new(), so we can list_del() it
 106          * here unconditionally */
 107         list_del_init(&req->tl_requests);
 108
 109         /* if it was a write, we may have to set the corresponding
 110          * bit(s) out-of-sync first. If it had a local part, we need to
 111          * release the reference to the activity log. */
 112         if (rw == WRITE) {
 113                 /* Set out-of-sync unless both OK flags are set
 114                  * (local only or remote failed).
 115                  * Other places where we set out-of-sync:
 116                  * READ with local io-error */
 117                 if (!(s & RQ_NET_OK) || !(s & RQ_LOCAL_OK))
 118                         drbd_set_out_of_sync(mdev, req->i.sector, req->i.size);
 119
 120                 if ((s & RQ_NET_OK) && (s & RQ_LOCAL_OK) && (s & RQ_NET_SIS))
 121                         drbd_set_in_sync(mdev, req->i.sector, req->i.size);
 122
 123                 /* one might be tempted to move the drbd_al_complete_io
 124                  * to the local io completion callback drbd_request_endio.
 125                  * but, if this was a mirror write, we may only
 126                  * drbd_al_complete_io after this is RQ_NET_DONE,
 127                  * otherwise the extent could be dropped from the al
 128                  * before it has actually been written on the peer.
 129                  * if we crash before our peer knows about the request,
 130                  * but after the extent has been dropped from the al,
 131                  * we would forget to resync the corresponding extent.
 132                  */
 133                 if (s & RQ_LOCAL_MASK) {
 134                         if (get_ldev_if_state(mdev, D_FAILED)) {
 135                                 if (s & RQ_IN_ACT_LOG)
 136                                         drbd_al_complete_io(mdev, &req->i);
 137                                 put_ldev(mdev);
 138                         } else if (__ratelimit(&drbd_ratelimit_state)) {
 139                                 dev_warn(DEV, "Should have called drbd_al_complete_io(, %llu, %u), "
 140                                          "but my Disk seems to have failed :(\n",
 141                                          (unsigned long long) req->i.sector, req->i.size);
 142                         }
 143                 }
 144         }
 145
 146         if (s & RQ_POSTPONED)
 147                 drbd_restart_write(req);
 148         else
 149                 drbd_req_free(req);
 150 }
 151
 152 static void queue_barrier(struct drbd_conf *mdev)
 153 {
 154         struct drbd_tl_epoch *b;
 155         struct drbd_tconn *tconn = mdev->tconn;
 156
 157         /* We are within the req_lock. Once we queued the barrier for sending,
 158          * we set the CREATE_BARRIER bit. It is cleared as soon as a new
 159          * barrier/epoch object is added. This is the only place this bit is
 160          * set. It indicates that the barrier for this epoch is already queued,
 161          * and no new epoch has been created yet. */
 162         if (test_bit(CREATE_BARRIER, &tconn->flags))
 163                 return;
 164
 165         b = tconn->newest_tle;
 166         b->w.cb = w_send_barrier;
 167         b->w.mdev = mdev;
 168         /* inc_ap_pending done here, so we won't
 169          * get imbalanced on connection loss.
 170          * dec_ap_pending will be done in got_BarrierAck
 171          * or (on connection loss) in tl_clear.  */
 172         inc_ap_pending(mdev);
 173         drbd_queue_work(&tconn->data.work, &b->w);
 174         set_bit(CREATE_BARRIER, &tconn->flags);
 175 }
 176
 177 static void _about_to_complete_local_write(struct drbd_conf *mdev,
 178         struct drbd_request *req)
 179 {
 180         const unsigned long s = req->rq_state;
 181
 182         /* Before we can signal completion to the upper layers,
 183          * we may need to close the current epoch.
 184          * We can skip this, if this request has not even been sent, because we
 185          * did not have a fully established connection yet/anymore, during
 186          * bitmap exchange, or while we are C_AHEAD due to congestion policy.
 187          */
 188         if (mdev->state.conn >= C_CONNECTED &&
 189             (s & RQ_NET_SENT) != 0 &&
 190             req->epoch == mdev->tconn->newest_tle->br_number)
 191                 queue_barrier(mdev);
 192 }
 193
 194 void complete_master_bio(struct drbd_conf *mdev,
 195                 struct bio_and_error *m)
 196 {
 197         bio_endio(m->bio, m->error);
 198         dec_ap_bio(mdev);
 199 }
 200
 201
 202 static void drbd_remove_request_interval(struct rb_root *root,
 203                                          struct drbd_request *req)
 204 {
 205         struct drbd_conf *mdev = req->w.mdev;
 206         struct drbd_interval *i = &req->i;
 207
 208         drbd_remove_interval(root, i);
 209
 210         /* Wake up any processes waiting for this request to complete.  */
 211         if (i->waiting)
 212                 wake_up(&mdev->misc_wait);
 213 }
 214
 215 /* Helper for __req_mod().
 216  * Set m->bio to the master bio, if it is fit to be completed,
 217  * or leave it alone (it is initialized to NULL in __req_mod),
 218  * if it has already been completed, or cannot be completed yet.
 219  * If m->bio is set, the error status to be returned is placed in m->error.
 220  */
 221 void _req_may_be_done(struct drbd_request *req, struct bio_and_error *m)
 222 {
 223         const unsigned long s = req->rq_state;
 224         struct drbd_conf *mdev = req->w.mdev;
 225         int rw = req->rq_state & RQ_WRITE ? WRITE : READ;
 226
 227         /* we must not complete the master bio, while it is
 228          *      still being processed by _drbd_send_zc_bio (drbd_send_dblock)
 229          *      not yet acknowledged by the peer
 230          *      not yet completed by the local io subsystem
 231          * these flags may get cleared in any order by
 232          *      the worker,
 233          *      the receiver,
 234          *      the bio_endio completion callbacks.
 235          */
 236         if (s & RQ_LOCAL_PENDING && !(s & RQ_LOCAL_ABORTED))
 237                 return;
 238         if (req->i.waiting) {
 239                 /* Retry all conflicting peer requests.  */
 240                 wake_up(&mdev->misc_wait);
 241         }
 242         if (s & RQ_NET_QUEUED)
 243                 return;
 244         if (s & RQ_NET_PENDING)
 245                 return;
 246
 247         if (req->master_bio) {
 248                 /* this is DATA_RECEIVED (remote read)
 249                  * or protocol C P_WRITE_ACK
 250                  * or protocol B P_RECV_ACK
 251                  * or protocol A "HANDED_OVER_TO_NETWORK" (SendAck)
 252                  * or canceled or failed,
 253                  * or killed from the transfer log due to connection loss.
 254                  */
 255
 256                 /*
 257                  * figure out whether to report success or failure.
 258                  *
 259                  * report success when at least one of the operations succeeded.
 260                  * or, to put the other way,
 261                  * only report failure, when both operations failed.
 262                  *
 263                  * what to do about the failures is handled elsewhere.
 264                  * what we need to do here is just: complete the master_bio.
 265                  *
 266                  * local completion error, if any, has been stored as ERR_PTR
 267                  * in private_bio within drbd_request_endio.
 268                  */
 269                 int ok = (s & RQ_LOCAL_OK) || (s & RQ_NET_OK);
 270                 int error = PTR_ERR(req->private_bio);
 271
 272                 /* remove the request from the conflict detection
 273                  * respective block_id verification hash */
 274                 if (!drbd_interval_empty(&req->i)) {
 275                         struct rb_root *root;
 276
 277                         if (rw == WRITE)
 278                                 root = &mdev->write_requests;
 279                         else
 280                                 root = &mdev->read_requests;
 281                         drbd_remove_request_interval(root, req);
 282                 } else if (!(s & RQ_POSTPONED))
 283                         D_ASSERT((s & (RQ_NET_MASK & ~RQ_NET_DONE)) == 0);
 284
 285                 /* for writes we need to do some extra housekeeping */
 286                 if (rw == WRITE)
 287                         _about_to_complete_local_write(mdev, req);
 288
 289                 /* Update disk stats */
 290                 _drbd_end_io_acct(mdev, req);
 291
 292                 if (!(s & RQ_POSTPONED)) {
 293                         m->error = ok ? 0 : (error ?: -EIO);
 294                         m->bio = req->master_bio;
 295                         req->master_bio = NULL;
 296                 } else {
 297                         /* Assert that this will be _req_is_done()
 298                          * with this very invokation. */
 299                         /* FIXME:
 300                          * what about (RQ_LOCAL_PENDING | RQ_LOCAL_ABORTED)?
 301                          */
 302                         D_ASSERT(!(s & RQ_LOCAL_PENDING));
 303                         D_ASSERT(s & RQ_NET_DONE);
 304                 }
 305         }
 306
 307         if (s & RQ_LOCAL_PENDING)
 308                 return;
 309
 310         if ((s & RQ_NET_MASK) == 0 || (s & RQ_NET_DONE)) {
 311                 /* this is disconnected (local only) operation,
 312                  * or protocol A, B, or C P_BARRIER_ACK,
 313                  * or killed from the transfer log due to connection loss. */
 314                 _req_is_done(mdev, req, rw);
 315         }
 316         /* else: network part and not DONE yet. that is
 317          * protocol A, B, or C, barrier ack still pending... */
 318 }
 319
 320 static void _req_may_be_done_not_susp(struct drbd_request *req, struct bio_and_error *m)
 321 {
 322         struct drbd_conf *mdev = req->w.mdev;
 323
 324         if (!drbd_suspended(mdev))
 325                 _req_may_be_done(req, m);
 326 }
 327
 328 /* obviously this could be coded as many single functions
 329  * instead of one huge switch,
 330  * or by putting the code directly in the respective locations
 331  * (as it has been before).
 332  *
 333  * but having it this way
 334  *  enforces that it is all in this one place, where it is easier to audit,
 335  *  it makes it obvious that whatever "event" "happens" to a request should
 336  *  happen "atomically" within the req_lock,
 337  *  and it enforces that we have to think in a very structured manner
 338  *  about the "events" that may happen to a request during its life time ...
 339  */
 340 int __req_mod(struct drbd_request *req, enum drbd_req_event what,
 341                 struct bio_and_error *m)
 342 {
 343         struct drbd_conf *mdev = req->w.mdev;
 344         struct net_conf *nc;
 345         int p, rv = 0;
 346
 347         if (m)
 348                 m->bio = NULL;
 349
 350         switch (what) {
 351         default:
 352                 dev_err(DEV, "LOGIC BUG in %s:%u\n", __FILE__ , __LINE__);
 353                 break;
 354
 355         /* does not happen...
 356          * initialization done in drbd_req_new
 357         case CREATED:
 358                 break;
 359                 */
 360
 361         case TO_BE_SENT: /* via network */
 362                 /* reached via __drbd_make_request
 363                  * and from w_read_retry_remote */
 364                 D_ASSERT(!(req->rq_state & RQ_NET_MASK));
 365                 req->rq_state |= RQ_NET_PENDING;
 366                 rcu_read_lock();
 367                 nc = rcu_dereference(mdev->tconn->net_conf);
 368                 p = nc->wire_protocol;
 369                 rcu_read_unlock();
 370                 req->rq_state |=
 371                         p == DRBD_PROT_C ? RQ_EXP_WRITE_ACK :
 372                         p == DRBD_PROT_B ? RQ_EXP_RECEIVE_ACK : 0;
 373                 inc_ap_pending(mdev);
 374                 break;
 375
 376         case TO_BE_SUBMITTED: /* locally */
 377                 /* reached via __drbd_make_request */
 378                 D_ASSERT(!(req->rq_state & RQ_LOCAL_MASK));
 379                 req->rq_state |= RQ_LOCAL_PENDING;
 380                 break;
 381
 382         case COMPLETED_OK:
 383                 if (req->rq_state & RQ_WRITE)
 384                         mdev->writ_cnt += req->i.size >> 9;
 385                 else
 386                         mdev->read_cnt += req->i.size >> 9;
 387
 388                 req->rq_state |= (RQ_LOCAL_COMPLETED|RQ_LOCAL_OK);
 389                 req->rq_state &= ~RQ_LOCAL_PENDING;
 390
 391                 _req_may_be_done_not_susp(req, m);
 392                 put_ldev(mdev);
 393                 break;
 394
 395         case ABORT_DISK_IO:
 396                 req->rq_state |= RQ_LOCAL_ABORTED;
 397                 if (req->rq_state & RQ_WRITE)
 398                         _req_may_be_done_not_susp(req, m);
 399                 else
 400                         goto goto_queue_for_net_read;
 401                 break;
 402
 403         case WRITE_COMPLETED_WITH_ERROR:
 404                 req->rq_state |= RQ_LOCAL_COMPLETED;
 405                 req->rq_state &= ~RQ_LOCAL_PENDING;
 406
 407                 __drbd_chk_io_error(mdev, false);
 408                 _req_may_be_done_not_susp(req, m);
 409                 put_ldev(mdev);
 410                 break;
 411
 412         case READ_AHEAD_COMPLETED_WITH_ERROR:
 413                 /* it is legal to fail READA */
 414                 req->rq_state |= RQ_LOCAL_COMPLETED;
 415                 req->rq_state &= ~RQ_LOCAL_PENDING;
 416                 _req_may_be_done_not_susp(req, m);
 417                 put_ldev(mdev);
 418                 break;
 419
 420         case READ_COMPLETED_WITH_ERROR:
 421                 drbd_set_out_of_sync(mdev, req->i.sector, req->i.size);
 422
 423                 req->rq_state |= RQ_LOCAL_COMPLETED;
 424                 req->rq_state &= ~RQ_LOCAL_PENDING;
 425
 426                 D_ASSERT(!(req->rq_state & RQ_NET_MASK));
 427
 428                 __drbd_chk_io_error(mdev, false);
 429                 put_ldev(mdev);
 430
 431         goto_queue_for_net_read:
 432
 433                 /* no point in retrying if there is no good remote data,
 434                  * or we have no connection. */
 435                 if (mdev->state.pdsk != D_UP_TO_DATE) {
 436                         _req_may_be_done_not_susp(req, m);
 437                         break;
 438                 }
 439
 440                 /* _req_mod(req,TO_BE_SENT); oops, recursion... */
 441                 req->rq_state |= RQ_NET_PENDING;
 442                 inc_ap_pending(mdev);
 443                 /* fall through: _req_mod(req,QUEUE_FOR_NET_READ); */
 444
 445         case QUEUE_FOR_NET_READ:
 446                 /* READ or READA, and
 447                  * no local disk,
 448                  * or target area marked as invalid,
 449                  * or just got an io-error. */
 450                 /* from __drbd_make_request
 451                  * or from bio_endio during read io-error recovery */
 452
 453                 /* so we can verify the handle in the answer packet
 454                  * corresponding hlist_del is in _req_may_be_done() */
 455                 D_ASSERT(drbd_interval_empty(&req->i));
 456                 drbd_insert_interval(&mdev->read_requests, &req->i);
 457
 458                 set_bit(UNPLUG_REMOTE, &mdev->flags);
 459
 460                 D_ASSERT(req->rq_state & RQ_NET_PENDING);
 461                 req->rq_state |= RQ_NET_QUEUED;
 462                 req->w.cb = (req->rq_state & RQ_LOCAL_MASK)
 463                         ? w_read_retry_remote
 464                         : w_send_read_req;
 465                 drbd_queue_work(&mdev->tconn->data.work, &req->w);
 466                 break;
 467
 468         case QUEUE_FOR_NET_WRITE:
 469                 /* assert something? */
 470                 /* from __drbd_make_request only */
 471
 472                 /* corresponding hlist_del is in _req_may_be_done() */
 473                 D_ASSERT(drbd_interval_empty(&req->i));
 474                 drbd_insert_interval(&mdev->write_requests, &req->i);
 475
 476                 /* NOTE
 477                  * In case the req ended up on the transfer log before being
 478                  * queued on the worker, it could lead to this request being
 479                  * missed during cleanup after connection loss.
 480                  * So we have to do both operations here,
 481                  * within the same lock that protects the transfer log.
 482                  *
 483                  * _req_add_to_epoch(req); this has to be after the
 484                  * _maybe_start_new_epoch(req); which happened in
 485                  * __drbd_make_request, because we now may set the bit
 486                  * again ourselves to close the current epoch.
 487                  *
 488                  * Add req to the (now) current epoch (barrier). */
 489
 490                 /* otherwise we may lose an unplug, which may cause some remote
 491                  * io-scheduler timeout to expire, increasing maximum latency,
 492                  * hurting performance. */
 493                 set_bit(UNPLUG_REMOTE, &mdev->flags);
 494
 495                 /* see __drbd_make_request,
 496                  * just after it grabs the req_lock */
 497                 D_ASSERT(test_bit(CREATE_BARRIER, &mdev->tconn->flags) == 0);
 498
 499                 req->epoch = mdev->tconn->newest_tle->br_number;
 500
 501                 /* increment size of current epoch */
 502                 mdev->tconn->newest_tle->n_writes++;
 503
 504                 /* queue work item to send data */
 505                 D_ASSERT(req->rq_state & RQ_NET_PENDING);
 506                 req->rq_state |= RQ_NET_QUEUED;
 507                 req->w.cb =  w_send_dblock;
 508                 drbd_queue_work(&mdev->tconn->data.work, &req->w);
 509
 510                 /* close the epoch, in case it outgrew the limit */
 511                 rcu_read_lock();
 512                 nc = rcu_dereference(mdev->tconn->net_conf);
 513                 p = nc->max_epoch_size;
 514                 rcu_read_unlock();
 515                 if (mdev->tconn->newest_tle->n_writes >= p)
 516                         queue_barrier(mdev);
 517
 518                 break;
 519
 520         case QUEUE_FOR_SEND_OOS:
 521                 req->rq_state |= RQ_NET_QUEUED;
 522                 req->w.cb =  w_send_out_of_sync;
 523                 drbd_queue_work(&mdev->tconn->data.work, &req->w);
 524                 break;
 525
 526         case READ_RETRY_REMOTE_CANCELED:
 527         case SEND_CANCELED:
 528         case SEND_FAILED:
 529                 /* real cleanup will be done from tl_clear.  just update flags
 530                  * so it is no longer marked as on the worker queue */
 531                 req->rq_state &= ~RQ_NET_QUEUED;
 532                 /* if we did it right, tl_clear should be scheduled only after
 533                  * this, so this should not be necessary! */
 534                 _req_may_be_done_not_susp(req, m);
 535                 break;
 536
 537         case HANDED_OVER_TO_NETWORK:
 538                 /* assert something? */
 539                 if (bio_data_dir(req->master_bio) == WRITE)
 540                         atomic_add(req->i.size >> 9, &mdev->ap_in_flight);
 541
 542                 if (bio_data_dir(req->master_bio) == WRITE &&
 543                     !(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK))) {
 544                         /* this is what is dangerous about protocol A:
 545                          * pretend it was successfully written on the peer. */
 546                         if (req->rq_state & RQ_NET_PENDING) {
 547                                 dec_ap_pending(mdev);
 548                                 req->rq_state &= ~RQ_NET_PENDING;
 549                                 req->rq_state |= RQ_NET_OK;
 550                         } /* else: neg-ack was faster... */
 551                         /* it is still not yet RQ_NET_DONE until the
 552                          * corresponding epoch barrier got acked as well,
 553                          * so we know what to dirty on connection loss */
 554                 }
 555                 req->rq_state &= ~RQ_NET_QUEUED;
 556                 req->rq_state |= RQ_NET_SENT;
 557                 _req_may_be_done_not_susp(req, m);
 558                 break;
 559
 560         case OOS_HANDED_TO_NETWORK:
 561                 /* Was not set PENDING, no longer QUEUED, so is now DONE
 562                  * as far as this connection is concerned. */
 563                 req->rq_state &= ~RQ_NET_QUEUED;
 564                 req->rq_state |= RQ_NET_DONE;
 565                 _req_may_be_done_not_susp(req, m);
 566                 break;
 567
 568         case CONNECTION_LOST_WHILE_PENDING:
 569                 /* transfer log cleanup after connection loss */
 570                 /* assert something? */
 571                 if (req->rq_state & RQ_NET_PENDING)
 572                         dec_ap_pending(mdev);
 573
 574                 p = !(req->rq_state & RQ_WRITE) && req->rq_state & RQ_NET_PENDING;
 575
 576                 req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING);
 577                 req->rq_state |= RQ_NET_DONE;
 578                 if (req->rq_state & RQ_NET_SENT && req->rq_state & RQ_WRITE)
 579                         atomic_sub(req->i.size >> 9, &mdev->ap_in_flight);
 580
 581                 /* if it is still queued, we may not complete it here.
 582                  * it will be canceled soon. */
 583                 if (!(req->rq_state & RQ_NET_QUEUED)) {
 584                         if (p)
 585                                 goto goto_read_retry_local;
 586                         _req_may_be_done(req, m); /* Allowed while state.susp */
 587                 }
 588                 break;
 589
 590         case WRITE_ACKED_BY_PEER_AND_SIS:
 591                 req->rq_state |= RQ_NET_SIS;
 592         case DISCARD_WRITE:
 593                 /* for discarded conflicting writes of multiple primaries,
 594                  * there is no need to keep anything in the tl, potential
 595                  * node crashes are covered by the activity log. */
 596                 req->rq_state |= RQ_NET_DONE;
 597                 /* fall through */
 598         case WRITE_ACKED_BY_PEER:
 599                 D_ASSERT(req->rq_state & RQ_EXP_WRITE_ACK);
 600                 /* protocol C; successfully written on peer.
 601                  * Nothing to do here.
 602                  * We want to keep the tl in place for all protocols, to cater
 603                  * for volatile write-back caches on lower level devices.
 604                  *
 605                  * A barrier request is expected to have forced all prior
 606                  * requests onto stable storage, so completion of a barrier
 607                  * request could set NET_DONE right here, and not wait for the
 608                  * P_BARRIER_ACK, but that is an unnecessary optimization. */
 609
 610                 goto ack_common;
 611                 /* this makes it effectively the same as for: */
 612         case RECV_ACKED_BY_PEER:
 613                 D_ASSERT(req->rq_state & RQ_EXP_RECEIVE_ACK);
 614                 /* protocol B; pretends to be successfully written on peer.
 615                  * see also notes above in HANDED_OVER_TO_NETWORK about
 616                  * protocol != C */
 617         ack_common:
 618                 req->rq_state |= RQ_NET_OK;
 619                 D_ASSERT(req->rq_state & RQ_NET_PENDING);
 620                 dec_ap_pending(mdev);
 621                 atomic_sub(req->i.size >> 9, &mdev->ap_in_flight);
 622                 req->rq_state &= ~RQ_NET_PENDING;
 623                 _req_may_be_done_not_susp(req, m);
 624                 break;
 625
 626         case POSTPONE_WRITE:
 627                 D_ASSERT(req->rq_state & RQ_EXP_WRITE_ACK);
 628                 /* If this node has already detected the write conflict, the
 629                  * worker will be waiting on misc_wait.  Wake it up once this
 630                  * request has completed locally.
 631                  */
 632                 D_ASSERT(req->rq_state & RQ_NET_PENDING);
 633                 req->rq_state |= RQ_POSTPONED;
 634                 _req_may_be_done_not_susp(req, m);
 635                 break;
 636
 637         case NEG_ACKED:
 638                 /* assert something? */
 639                 if (req->rq_state & RQ_NET_PENDING) {
 640                         dec_ap_pending(mdev);
 641                         if (req->rq_state & RQ_WRITE)
 642                                 atomic_sub(req->i.size >> 9, &mdev->ap_in_flight);
 643                 }
 644                 req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING);
 645
 646                 req->rq_state |= RQ_NET_DONE;
 647
 648                 if (!(req->rq_state & RQ_WRITE))
 649                         goto goto_read_retry_local;
 650
 651                 _req_may_be_done_not_susp(req, m);
 652                 /* else: done by HANDED_OVER_TO_NETWORK */
 653                 break;
 654
 655         goto_read_retry_local:
 656                 if (!drbd_may_do_local_read(mdev, req->i.sector, req->i.size)) {
 657                         _req_may_be_done_not_susp(req, m);
 658                         break;
 659                 }
 660                 D_ASSERT(!(req->rq_state & RQ_LOCAL_PENDING));
 661                 req->rq_state |= RQ_LOCAL_PENDING;
 662
 663                 get_ldev(mdev);
 664                 req->w.cb = w_restart_disk_io;
 665                 drbd_queue_work(&mdev->tconn->data.work, &req->w);
 666                 break;
 667
 668         case FAIL_FROZEN_DISK_IO:
 669                 if (!(req->rq_state & RQ_LOCAL_COMPLETED))
 670                         break;
 671
 672                 _req_may_be_done(req, m); /* Allowed while state.susp */
 673                 break;
 674
 675         case RESTART_FROZEN_DISK_IO:
 676                 if (!(req->rq_state & RQ_LOCAL_COMPLETED))
 677                         break;
 678
 679                 req->rq_state &= ~RQ_LOCAL_COMPLETED;
 680
 681                 rv = MR_READ;
 682                 if (bio_data_dir(req->master_bio) == WRITE)
 683                         rv = MR_WRITE;
 684
 685                 get_ldev(mdev);
 686                 req->w.cb = w_restart_disk_io;
 687                 drbd_queue_work(&mdev->tconn->data.work, &req->w);
 688                 break;
 689
 690         case RESEND:
 691                 /* If RQ_NET_OK is already set, we got a P_WRITE_ACK or P_RECV_ACK
 692                    before the connection loss (B&C only); only P_BARRIER_ACK was missing.
 693                    Trowing them out of the TL here by pretending we got a BARRIER_ACK
 694                    We ensure that the peer was not rebooted */
 695                 if (!(req->rq_state & RQ_NET_OK)) {
 696                         if (req->w.cb) {
 697                                 drbd_queue_work(&mdev->tconn->data.work, &req->w);
 698                                 rv = req->rq_state & RQ_WRITE ? MR_WRITE : MR_READ;
 699                         }
 700                         break;
 701                 }
 702                 /* else, fall through to BARRIER_ACKED */
 703
 704         case BARRIER_ACKED:
 705                 if (!(req->rq_state & RQ_WRITE))
 706                         break;
 707
 708                 if (req->rq_state & RQ_NET_PENDING) {
 709                         /* barrier came in before all requests were acked.
 710                          * this is bad, because if the connection is lost now,
 711                          * we won't be able to clean them up... */
 712                         dev_err(DEV, "FIXME (BARRIER_ACKED but pending)\n");
 713                         list_move(&req->tl_requests, &mdev->tconn->out_of_sequence_requests);
 714                 }
 715                 if ((req->rq_state & RQ_NET_MASK) != 0) {
 716                         req->rq_state |= RQ_NET_DONE;
 717                         if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)))
 718                                 atomic_sub(req->i.size>>9, &mdev->ap_in_flight);
 719                 }
 720                 _req_may_be_done(req, m); /* Allowed while state.susp */
 721                 break;
 722
 723         case DATA_RECEIVED:
 724                 D_ASSERT(req->rq_state & RQ_NET_PENDING);
 725                 dec_ap_pending(mdev);
 726                 req->rq_state &= ~RQ_NET_PENDING;
 727                 req->rq_state |= (RQ_NET_OK|RQ_NET_DONE);
 728                 _req_may_be_done_not_susp(req, m);
 729                 break;
 730         };
 731
 732         return rv;
 733 }
 734
 735 /* we may do a local read if:
 736  * - we are consistent (of course),
 737  * - or we are generally inconsistent,
 738  *   BUT we are still/already IN SYNC for this area.
 739  *   since size may be bigger than BM_BLOCK_SIZE,
 740  *   we may need to check several bits.
 741  */
 742 static bool drbd_may_do_local_read(struct drbd_conf *mdev, sector_t sector, int size)
 743 {
 744         unsigned long sbnr, ebnr;
 745         sector_t esector, nr_sectors;
 746
 747         if (mdev->state.disk == D_UP_TO_DATE)
 748                 return true;
 749         if (mdev->state.disk != D_INCONSISTENT)
 750                 return false;
 751         esector = sector + (size >> 9) - 1;
 752         nr_sectors = drbd_get_capacity(mdev->this_bdev);
 753         D_ASSERT(sector  < nr_sectors);
 754         D_ASSERT(esector < nr_sectors);
 755
 756         sbnr = BM_SECT_TO_BIT(sector);
 757         ebnr = BM_SECT_TO_BIT(esector);
 758
 759         return drbd_bm_count_bits(mdev, sbnr, ebnr) == 0;
 760 }
 761
 762 static bool remote_due_to_read_balancing(struct drbd_conf *mdev, sector_t sector)
 763 {
 764         enum drbd_read_balancing rbm;
 765         struct backing_dev_info *bdi;
 766         int stripe_shift;
 767
 768         if (mdev->state.pdsk < D_UP_TO_DATE)
 769                 return false;
 770
 771         rcu_read_lock();
 772         rbm = rcu_dereference(mdev->ldev->disk_conf)->read_balancing;
 773         rcu_read_unlock();
 774
 775         switch (rbm) {
 776         case RB_CONGESTED_REMOTE:
 777                 bdi = &mdev->ldev->backing_bdev->bd_disk->queue->backing_dev_info;
 778                 return bdi_read_congested(bdi);
 779         case RB_LEAST_PENDING:
 780                 return atomic_read(&mdev->local_cnt) >
 781                         atomic_read(&mdev->ap_pending_cnt) + atomic_read(&mdev->rs_pending_cnt);
 782         case RB_32K_STRIPING:  /* stripe_shift = 15 */
 783         case RB_64K_STRIPING:
 784         case RB_128K_STRIPING:
 785         case RB_256K_STRIPING:
 786         case RB_512K_STRIPING:
 787         case RB_1M_STRIPING:   /* stripe_shift = 20 */
 788                 stripe_shift = (rbm - RB_32K_STRIPING + 15);
 789                 return (sector >> (stripe_shift - 9)) & 1;
 790         case RB_ROUND_ROBIN:
 791                 return test_and_change_bit(READ_BALANCE_RR, &mdev->flags);
 792         case RB_PREFER_REMOTE:
 793                 return true;
 794         case RB_PREFER_LOCAL:
 795         default:
 796                 return false;
 797         }
 798 }
 799
 800 /*
 801  * complete_conflicting_writes  -  wait for any conflicting write requests
 802  *
 803  * The write_requests tree contains all active write requests which we
 804  * currently know about.  Wait for any requests to complete which conflict with
 805  * the new one.
 806  */
 807 static int complete_conflicting_writes(struct drbd_conf *mdev,
 808                                        sector_t sector, int size)
 809 {
 810         for(;;) {
 811                 struct drbd_interval *i;
 812                 int err;
 813
 814                 i = drbd_find_overlap(&mdev->write_requests, sector, size);
 815                 if (!i)
 816                         return 0;
 817                 err = drbd_wait_misc(mdev, i);
 818                 if (err)
 819                         return err;
 820         }
 821 }
 822
 823 int __drbd_make_request(struct drbd_conf *mdev, struct bio *bio, unsigned long start_time)
 824 {
 825         const int rw = bio_rw(bio);
 826         const int size = bio->bi_size;
 827         const sector_t sector = bio->bi_sector;
 828         struct drbd_tl_epoch *b = NULL;
 829         struct drbd_request *req;
 830         struct net_conf *nc;
 831         int local, remote, send_oos = 0;
 832         int err;
 833         int ret = 0;
 834         union drbd_dev_state s;
 835
 836         /* allocate outside of all locks; */
 837         req = drbd_req_new(mdev, bio);
 838         if (!req) {
 839                 dec_ap_bio(mdev);
 840                 /* only pass the error to the upper layers.
 841                  * if user cannot handle io errors, that's not our business. */
 842                 dev_err(DEV, "could not kmalloc() req\n");
 843                 bio_endio(bio, -ENOMEM);
 844                 return 0;
 845         }
 846         req->start_time = start_time;
 847
 848         local = get_ldev(mdev);
 849         if (!local) {
 850                 bio_put(req->private_bio); /* or we get a bio leak */
 851                 req->private_bio = NULL;
 852         }
 853         if (rw == WRITE) {
 854                 remote = 1;
 855         } else {
 856                 /* READ || READA */
 857                 if (local) {
 858                         if (!drbd_may_do_local_read(mdev, sector, size) ||
 859                             remote_due_to_read_balancing(mdev, sector)) {
 860                                 /* we could kick the syncer to
 861                                  * sync this extent asap, wait for
 862                                  * it, then continue locally.
 863                                  * Or just issue the request remotely.
 864                                  */
 865                                 local = 0;
 866                                 bio_put(req->private_bio);
 867                                 req->private_bio = NULL;
 868                                 put_ldev(mdev);
 869                         }
 870                 }
 871                 remote = !local && mdev->state.pdsk >= D_UP_TO_DATE;
 872         }
 873
 874         /* If we have a disk, but a READA request is mapped to remote,
 875          * we are R_PRIMARY, D_INCONSISTENT, SyncTarget.
 876          * Just fail that READA request right here.
 877          *
 878          * THINK: maybe fail all READA when not local?
 879          *        or make this configurable...
 880          *        if network is slow, READA won't do any good.
 881          */
 882         if (rw == READA && mdev->state.disk >= D_INCONSISTENT && !local) {
 883                 err = -EWOULDBLOCK;
 884                 goto fail_and_free_req;
 885         }
 886
 887         /* For WRITES going to the local disk, grab a reference on the target
 888          * extent.  This waits for any resync activity in the corresponding
 889          * resync extent to finish, and, if necessary, pulls in the target
 890          * extent into the activity log, which involves further disk io because
 891          * of transactional on-disk meta data updates. */
 892         if (rw == WRITE && local && !test_bit(AL_SUSPENDED, &mdev->flags)) {
 893                 req->rq_state |= RQ_IN_ACT_LOG;
 894                 drbd_al_begin_io(mdev, &req->i);
 895         }
 896
 897         s = mdev->state;
 898         remote = remote && drbd_should_do_remote(s);
 899         send_oos = rw == WRITE && drbd_should_send_out_of_sync(s);
 900         D_ASSERT(!(remote && send_oos));
 901
 902         if (!(local || remote) && !drbd_suspended(mdev)) {
 903                 if (__ratelimit(&drbd_ratelimit_state))
 904                         dev_err(DEV, "IO ERROR: neither local nor remote disk\n");
 905                 err = -EIO;
 906                 goto fail_free_complete;
 907         }
 908
 909         /* For WRITE request, we have to make sure that we have an
 910          * unused_spare_tle, in case we need to start a new epoch.
 911          * I try to be smart and avoid to pre-allocate always "just in case",
 912          * but there is a race between testing the bit and pointer outside the
 913          * spinlock, and grabbing the spinlock.
 914          * if we lost that race, we retry.  */
 915         if (rw == WRITE && (remote || send_oos) &&
 916             mdev->tconn->unused_spare_tle == NULL &&
 917             test_bit(CREATE_BARRIER, &mdev->tconn->flags)) {
 918 allocate_barrier:
 919                 b = kmalloc(sizeof(struct drbd_tl_epoch), GFP_NOIO);
 920                 if (!b) {
 921                         dev_err(DEV, "Failed to alloc barrier.\n");
 922                         err = -ENOMEM;
 923                         goto fail_free_complete;
 924                 }
 925         }
 926
 927         /* GOOD, everything prepared, grab the spin_lock */
 928         spin_lock_irq(&mdev->tconn->req_lock);
 929
 930         if (rw == WRITE) {
 931                 err = complete_conflicting_writes(mdev, sector, size);
 932                 if (err) {
 933                         if (err != -ERESTARTSYS)
 934                                 _conn_request_state(mdev->tconn,
 935                                                     NS(conn, C_TIMEOUT),
 936                                                     CS_HARD);
 937                         spin_unlock_irq(&mdev->tconn->req_lock);
 938                         err = -EIO;
 939                         goto fail_free_complete;
 940                 }
 941         }
 942
 943         if (drbd_suspended(mdev)) {
 944                 /* If we got suspended, use the retry mechanism in
 945                    drbd_make_request() to restart processing of this
 946                    bio. In the next call to drbd_make_request
 947                    we sleep in inc_ap_bio() */
 948                 ret = 1;
 949                 spin_unlock_irq(&mdev->tconn->req_lock);
 950                 goto fail_free_complete;
 951         }
 952
 953         if (remote || send_oos) {
 954                 remote = drbd_should_do_remote(mdev->state);
 955                 send_oos = rw == WRITE && drbd_should_send_out_of_sync(mdev->state);
 956                 D_ASSERT(!(remote && send_oos));
 957
 958                 if (!(remote || send_oos))
 959                         dev_warn(DEV, "lost connection while grabbing the req_lock!\n");
 960                 if (!(local || remote)) {
 961                         dev_err(DEV, "IO ERROR: neither local nor remote disk\n");
 962                         spin_unlock_irq(&mdev->tconn->req_lock);
 963                         err = -EIO;
 964                         goto fail_free_complete;
 965                 }
 966         }
 967
 968         if (b && mdev->tconn->unused_spare_tle == NULL) {
 969                 mdev->tconn->unused_spare_tle = b;
 970                 b = NULL;
 971         }
 972         if (rw == WRITE && (remote || send_oos) &&
 973             mdev->tconn->unused_spare_tle == NULL &&
 974             test_bit(CREATE_BARRIER, &mdev->tconn->flags)) {
 975                 /* someone closed the current epoch
 976                  * while we were grabbing the spinlock */
 977                 spin_unlock_irq(&mdev->tconn->req_lock);
 978                 goto allocate_barrier;
 979         }
 980
 981
 982         /* Update disk stats */
 983         _drbd_start_io_acct(mdev, req, bio);
 984
 985         /* _maybe_start_new_epoch(mdev);
 986          * If we need to generate a write barrier packet, we have to add the
 987          * new epoch (barrier) object, and queue the barrier packet for sending,
 988          * and queue the req's data after it _within the same lock_, otherwise
 989          * we have race conditions were the reorder domains could be mixed up.
 990          *
 991          * Even read requests may start a new epoch and queue the corresponding
 992          * barrier packet.  To get the write ordering right, we only have to
 993          * make sure that, if this is a write request and it triggered a
 994          * barrier packet, this request is queued within the same spinlock. */
 995         if ((remote || send_oos) && mdev->tconn->unused_spare_tle &&
 996             test_and_clear_bit(CREATE_BARRIER, &mdev->tconn->flags)) {
 997                 _tl_add_barrier(mdev->tconn, mdev->tconn->unused_spare_tle);
 998                 mdev->tconn->unused_spare_tle = NULL;
 999         } else {
1000                 D_ASSERT(!(remote && rw == WRITE &&
1001                            test_bit(CREATE_BARRIER, &mdev->tconn->flags)));
1002         }
1003
1004         /* NOTE
1005          * Actually, 'local' may be wrong here already, since we may have failed
1006          * to write to the meta data, and may become wrong anytime because of
1007          * local io-error for some other request, which would lead to us
1008          * "detaching" the local disk.
1009          *
1010          * 'remote' may become wrong any time because the network could fail.
1011          *
1012          * This is a harmless race condition, though, since it is handled
1013          * correctly at the appropriate places; so it just defers the failure
1014          * of the respective operation.
1015          */
1016
1017         /* mark them early for readability.
1018          * this just sets some state flags. */
1019         if (remote)
1020                 _req_mod(req, TO_BE_SENT);
1021         if (local)
1022                 _req_mod(req, TO_BE_SUBMITTED);
1023
1024         list_add_tail(&req->tl_requests, &mdev->tconn->newest_tle->requests);
1025
1026         /* NOTE remote first: to get the concurrent write detection right,
1027          * we must register the request before start of local IO.  */
1028         if (remote) {
1029                 /* either WRITE and C_CONNECTED,
1030                  * or READ, and no local disk,
1031                  * or READ, but not in sync.
1032                  */
1033                 _req_mod(req, (rw == WRITE)
1034                                 ? QUEUE_FOR_NET_WRITE
1035                                 : QUEUE_FOR_NET_READ);
1036         }
1037         if (send_oos && drbd_set_out_of_sync(mdev, sector, size))
1038                 _req_mod(req, QUEUE_FOR_SEND_OOS);
1039
1040         rcu_read_lock();
1041         nc = rcu_dereference(mdev->tconn->net_conf);
1042         if (remote &&
1043             nc->on_congestion != OC_BLOCK && mdev->tconn->agreed_pro_version >= 96) {
1044                 int congested = 0;
1045
1046                 if (nc->cong_fill &&
1047                     atomic_read(&mdev->ap_in_flight) >= nc->cong_fill) {
1048                         dev_info(DEV, "Congestion-fill threshold reached\n");
1049                         congested = 1;
1050                 }
1051
1052                 if (mdev->act_log->used >= nc->cong_extents) {
1053                         dev_info(DEV, "Congestion-extents threshold reached\n");
1054                         congested = 1;
1055                 }
1056
1057                 if (congested) {
1058                         queue_barrier(mdev); /* last barrier, after mirrored writes */
1059
1060                         if (nc->on_congestion == OC_PULL_AHEAD)
1061                                 _drbd_set_state(_NS(mdev, conn, C_AHEAD), 0, NULL);
1062                         else  /*nc->on_congestion == OC_DISCONNECT */
1063                                 _drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), 0, NULL);
1064                 }
1065         }
1066         rcu_read_unlock();
1067
1068         spin_unlock_irq(&mdev->tconn->req_lock);
1069         kfree(b); /* if someone else has beaten us to it... */
1070
1071         if (local) {
1072                 req->private_bio->bi_bdev = mdev->ldev->backing_bdev;
1073
1074                 /* State may have changed since we grabbed our reference on the
1075                  * mdev->ldev member. Double check, and short-circuit to endio.
1076                  * In case the last activity log transaction failed to get on
1077                  * stable storage, and this is a WRITE, we may not even submit
1078                  * this bio. */
1079                 if (get_ldev(mdev)) {
1080                         if (drbd_insert_fault(mdev,   rw == WRITE ? DRBD_FAULT_DT_WR
1081                                                     : rw == READ  ? DRBD_FAULT_DT_RD
1082                                                     :               DRBD_FAULT_DT_RA))
1083                                 bio_endio(req->private_bio, -EIO);
1084                         else
1085                                 generic_make_request(req->private_bio);
1086                         put_ldev(mdev);
1087                 } else
1088                         bio_endio(req->private_bio, -EIO);
1089         }
1090
1091         return 0;
1092
1093 fail_free_complete:
1094         if (req->rq_state & RQ_IN_ACT_LOG)
1095                 drbd_al_complete_io(mdev, &req->i);
1096 fail_and_free_req:
1097         if (local) {
1098                 bio_put(req->private_bio);
1099                 req->private_bio = NULL;
1100                 put_ldev(mdev);
1101         }
1102         if (!ret)
1103                 bio_endio(bio, err);
1104
1105         drbd_req_free(req);
1106         dec_ap_bio(mdev);
1107         kfree(b);
1108
1109         return ret;
1110 }
1111
1112 int drbd_make_request(struct request_queue *q, struct bio *bio)
1113 {
1114         struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata;
1115         unsigned long start_time;
1116
1117         start_time = jiffies;
1118
1119         /*
1120          * what we "blindly" assume:
1121          */
1122         D_ASSERT(bio->bi_size > 0);
1123         D_ASSERT(IS_ALIGNED(bio->bi_size, 512));
1124
1125         do {
1126                 inc_ap_bio(mdev);
1127         } while (__drbd_make_request(mdev, bio, start_time));
1128
1129         return 0;
1130 }
1131
1132 /* This is called by bio_add_page().
1133  *
1134  * q->max_hw_sectors and other global limits are already enforced there.
1135  *
1136  * We need to call down to our lower level device,
1137  * in case it has special restrictions.
1138  *
1139  * We also may need to enforce configured max-bio-bvecs limits.
1140  *
1141  * As long as the BIO is empty we have to allow at least one bvec,
1142  * regardless of size and offset, so no need to ask lower levels.
1143  */
1144 int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec)
1145 {
1146         struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata;
1147         unsigned int bio_size = bvm->bi_size;
1148         int limit = DRBD_MAX_BIO_SIZE;
1149         int backing_limit;
1150
1151         if (bio_size && get_ldev(mdev)) {
1152                 struct request_queue * const b =
1153                         mdev->ldev->backing_bdev->bd_disk->queue;
1154                 if (b->merge_bvec_fn) {
1155                         backing_limit = b->merge_bvec_fn(b, bvm, bvec);
1156                         limit = min(limit, backing_limit);
1157                 }
1158                 put_ldev(mdev);
1159         }
1160         return limit;
1161 }
1162
1163 void request_timer_fn(unsigned long data)
1164 {
1165         struct drbd_conf *mdev = (struct drbd_conf *) data;
1166         struct drbd_tconn *tconn = mdev->tconn;
1167         struct drbd_request *req; /* oldest request */
1168         struct list_head *le;
1169         struct net_conf *nc;
1170         unsigned long ent = 0, dt = 0, et, nt; /* effective timeout = ko_count * timeout */
1171
1172         rcu_read_lock();
1173         nc = rcu_dereference(tconn->net_conf);
1174         ent = nc ? nc->timeout * HZ/10 * nc->ko_count : 0;
1175
1176         if (get_ldev(mdev)) {
1177                 dt = rcu_dereference(mdev->ldev->disk_conf)->disk_timeout * HZ / 10;
1178                 put_ldev(mdev);
1179         }
1180         rcu_read_unlock();
1181
1182         et = min_not_zero(dt, ent);
1183
1184         if (!et || (mdev->state.conn < C_WF_REPORT_PARAMS && mdev->state.disk <= D_FAILED))
1185                 return; /* Recurring timer stopped */
1186
1187         spin_lock_irq(&tconn->req_lock);
1188         le = &tconn->oldest_tle->requests;
1189         if (list_empty(le)) {
1190                 spin_unlock_irq(&tconn->req_lock);
1191                 mod_timer(&mdev->request_timer, jiffies + et);
1192                 return;
1193         }
1194
1195         le = le->prev;
1196         req = list_entry(le, struct drbd_request, tl_requests);
1197         if (ent && req->rq_state & RQ_NET_PENDING) {
1198                 if (time_is_before_eq_jiffies(req->start_time + ent)) {
1199                         dev_warn(DEV, "Remote failed to finish a request within ko-count * timeout\n");
1200                         _drbd_set_state(_NS(mdev, conn, C_TIMEOUT), CS_VERBOSE | CS_HARD, NULL);
1201                 }
1202         }
1203         if (dt && req->rq_state & RQ_LOCAL_PENDING && req->w.mdev == mdev) {
1204                 if (time_is_before_eq_jiffies(req->start_time + dt)) {
1205                         dev_warn(DEV, "Local backing device failed to meet the disk-timeout\n");
1206                         __drbd_chk_io_error(mdev, 1);
1207                 }
1208         }
1209         nt = (time_is_before_eq_jiffies(req->start_time + et) ? jiffies : req->start_time) + et;
1210         spin_unlock_irq(&tconn->req_lock);
1211         mod_timer(&mdev->request_timer, nt);
1212 }