drbd_req.c 35.0 KB
Newer Older
P
Philipp Reisner 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
/*
   drbd_req.c

   This file is part of DRBD by Philipp Reisner and Lars Ellenberg.

   Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.

   drbd is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 2, or (at your option)
   any later version.

   drbd is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with drbd; see the file COPYING.  If not, write to
   the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.

 */

#include <linux/module.h>

#include <linux/slab.h>
#include <linux/drbd.h>
#include "drbd_int.h"
#include "drbd_req.h"


34 35
static bool drbd_may_do_local_read(struct drbd_conf *mdev, sector_t sector, int size);

P
Philipp Reisner 已提交
36 37 38 39 40 41
/* Update disk stats at start of I/O request */
static void _drbd_start_io_acct(struct drbd_conf *mdev, struct drbd_request *req, struct bio *bio)
{
	const int rw = bio_data_dir(bio);
	int cpu;
	cpu = part_stat_lock();
42
	part_round_stats(cpu, &mdev->vdisk->part0);
P
Philipp Reisner 已提交
43 44
	part_stat_inc(cpu, &mdev->vdisk->part0, ios[rw]);
	part_stat_add(cpu, &mdev->vdisk->part0, sectors[rw], bio_sectors(bio));
P
Philipp Reisner 已提交
45 46
	(void) cpu; /* The macro invocations above want the cpu argument, I do not like
		       the compiler warning about cpu only assigned but never used... */
47
	part_inc_in_flight(&mdev->vdisk->part0, rw);
P
Philipp Reisner 已提交
48 49 50 51 52 53 54 55 56 57 58 59
	part_stat_unlock();
}

/* Update disk stats when completing request upwards */
static void _drbd_end_io_acct(struct drbd_conf *mdev, struct drbd_request *req)
{
	int rw = bio_data_dir(req->master_bio);
	unsigned long duration = jiffies - req->start_time;
	int cpu;
	cpu = part_stat_lock();
	part_stat_add(cpu, &mdev->vdisk->part0, ticks[rw], duration);
	part_round_stats(cpu, &mdev->vdisk->part0);
60
	part_dec_in_flight(&mdev->vdisk->part0, rw);
P
Philipp Reisner 已提交
61 62 63
	part_stat_unlock();
}

64 65 66 67 68 69 70 71 72 73 74
static struct drbd_request *drbd_req_new(struct drbd_conf *mdev,
					       struct bio *bio_src)
{
	struct drbd_request *req;

	req = mempool_alloc(drbd_request_mempool, GFP_NOIO);
	if (!req)
		return NULL;

	drbd_req_make_private_bio(req, bio_src);
	req->rq_state    = bio_data_dir(bio_src) == WRITE ? RQ_WRITE : 0;
75
	req->w.mdev      = mdev;
76 77
	req->master_bio  = bio_src;
	req->epoch       = 0;
78

79 80 81
	drbd_clear_interval(&req->i);
	req->i.sector     = bio_src->bi_sector;
	req->i.size      = bio_src->bi_size;
82
	req->i.local = true;
83 84
	req->i.waiting = false;

85 86 87 88 89 90 91 92 93 94 95 96
	INIT_LIST_HEAD(&req->tl_requests);
	INIT_LIST_HEAD(&req->w.list);

	return req;
}

static void drbd_req_free(struct drbd_request *req)
{
	mempool_free(req, drbd_request_mempool);
}

/* rw is bio_data_dir(), only READ or WRITE */
P
Philipp Reisner 已提交
97 98 99
static void _req_is_done(struct drbd_conf *mdev, struct drbd_request *req, const int rw)
{
	const unsigned long s = req->rq_state;
100 101 102 103 104 105 106

	/* remove it from the transfer log.
	 * well, only if it had been there in the first
	 * place... if it had not (local only or conflicting
	 * and never sent), it should still be "empty" as
	 * initialized in drbd_req_new(), so we can list_del() it
	 * here unconditionally */
107
	list_del_init(&req->tl_requests);
108

P
Philipp Reisner 已提交
109 110 111 112 113 114 115 116 117
	/* if it was a write, we may have to set the corresponding
	 * bit(s) out-of-sync first. If it had a local part, we need to
	 * release the reference to the activity log. */
	if (rw == WRITE) {
		/* Set out-of-sync unless both OK flags are set
		 * (local only or remote failed).
		 * Other places where we set out-of-sync:
		 * READ with local io-error */
		if (!(s & RQ_NET_OK) || !(s & RQ_LOCAL_OK))
118
			drbd_set_out_of_sync(mdev, req->i.sector, req->i.size);
P
Philipp Reisner 已提交
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);
P
Philipp Reisner 已提交
122 123

		/* one might be tempted to move the drbd_al_complete_io
124
		 * to the local io completion callback drbd_request_endio.
P
Philipp Reisner 已提交
125 126 127 128 129 130 131 132 133 134
		 * but, if this was a mirror write, we may only
		 * drbd_al_complete_io after this is RQ_NET_DONE,
		 * otherwise the extent could be dropped from the al
		 * before it has actually been written on the peer.
		 * if we crash before our peer knows about the request,
		 * but after the extent has been dropped from the al,
		 * we would forget to resync the corresponding extent.
		 */
		if (s & RQ_LOCAL_MASK) {
			if (get_ldev_if_state(mdev, D_FAILED)) {
135
				if (s & RQ_IN_ACT_LOG)
136
					drbd_al_complete_io(mdev, &req->i);
P
Philipp Reisner 已提交
137 138
				put_ldev(mdev);
			} else if (__ratelimit(&drbd_ratelimit_state)) {
139 140 141
				dev_warn(DEV, "Should have called drbd_al_complete_io(, %llu, %u), "
					 "but my Disk seems to have failed :(\n",
					 (unsigned long long) req->i.sector, req->i.size);
P
Philipp Reisner 已提交
142 143 144 145
			}
		}
	}

146
	if (s & RQ_POSTPONED)
147
		drbd_restart_request(req);
148 149
	else
		drbd_req_free(req);
P
Philipp Reisner 已提交
150 151
}

152 153
static void wake_all_senders(struct drbd_tconn *tconn) {
	wake_up(&tconn->sender_work.q_wait);
P
Philipp Reisner 已提交
154 155
}

156 157
/* must hold resource->req_lock */
static void start_new_tl_epoch(struct drbd_tconn *tconn)
P
Philipp Reisner 已提交
158
{
159 160 161
	tconn->current_tle_writes = 0;
	atomic_inc(&tconn->current_tle_nr);
	wake_all_senders(tconn);
P
Philipp Reisner 已提交
162 163 164 165 166 167 168 169 170
}

void complete_master_bio(struct drbd_conf *mdev,
		struct bio_and_error *m)
{
	bio_endio(m->bio, m->error);
	dec_ap_bio(mdev);
}

171 172 173 174

static void drbd_remove_request_interval(struct rb_root *root,
					 struct drbd_request *req)
{
175
	struct drbd_conf *mdev = req->w.mdev;
176 177 178 179 180 181 182 183 184
	struct drbd_interval *i = &req->i;

	drbd_remove_interval(root, i);

	/* Wake up any processes waiting for this request to complete.  */
	if (i->waiting)
		wake_up(&mdev->misc_wait);
}

185 186 187 188 189 190 191 192 193 194
static void maybe_wakeup_conflicting_requests(struct drbd_request *req)
{
	const unsigned long s = req->rq_state;
	if (s & RQ_LOCAL_PENDING && !(s & RQ_LOCAL_ABORTED))
		return;
	if (req->i.waiting)
		/* Retry all conflicting peer requests.  */
		wake_up(&req->w.mdev->misc_wait);
}

195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224
static
void req_may_be_done(struct drbd_request *req)
{
	const unsigned long s = req->rq_state;
	struct drbd_conf *mdev = req->w.mdev;
	int rw = req->rq_state & RQ_WRITE ? WRITE : READ;

	/* req->master_bio still present means: Not yet completed.
	 *
	 * Unless this is RQ_POSTPONED, which will cause _req_is_done() to
	 * queue it on the retry workqueue instead of destroying it.
	 */
	if (req->master_bio && !(s & RQ_POSTPONED))
		return;

	/* Local still pending, even though master_bio is already completed?
	 * may happen for RQ_LOCAL_ABORTED requests. */
	if (s & RQ_LOCAL_PENDING)
		return;

	if ((s & RQ_NET_MASK) == 0 || (s & RQ_NET_DONE)) {
		/* this is disconnected (local only) operation,
		 * or protocol A, B, or C P_BARRIER_ACK,
		 * or killed from the transfer log due to connection loss. */
		_req_is_done(mdev, req, rw);
	}
	/* else: network part and not DONE yet. that is
	 * protocol A, B, or C, barrier ack still pending... */
}

P
Philipp Reisner 已提交
225 226 227 228 229 230
/* Helper for __req_mod().
 * Set m->bio to the master bio, if it is fit to be completed,
 * or leave it alone (it is initialized to NULL in __req_mod),
 * if it has already been completed, or cannot be completed yet.
 * If m->bio is set, the error status to be returned is placed in m->error.
 */
231 232
static
void req_may_be_completed(struct drbd_request *req, struct bio_and_error *m)
P
Philipp Reisner 已提交
233 234
{
	const unsigned long s = req->rq_state;
235
	struct drbd_conf *mdev = req->w.mdev;
P
Philipp Reisner 已提交
236 237 238 239 240 241 242 243 244 245

	/* we must not complete the master bio, while it is
	 *	still being processed by _drbd_send_zc_bio (drbd_send_dblock)
	 *	not yet acknowledged by the peer
	 *	not yet completed by the local io subsystem
	 * these flags may get cleared in any order by
	 *	the worker,
	 *	the receiver,
	 *	the bio_endio completion callbacks.
	 */
246
	if (s & RQ_LOCAL_PENDING && !(s & RQ_LOCAL_ABORTED))
247
		return;
P
Philipp Reisner 已提交
248 249 250 251 252 253
	if (s & RQ_NET_QUEUED)
		return;
	if (s & RQ_NET_PENDING)
		return;

	if (req->master_bio) {
254 255
		int rw = bio_rw(req->master_bio);

256
		/* this is DATA_RECEIVED (remote read)
P
Philipp Reisner 已提交
257 258
		 * or protocol C P_WRITE_ACK
		 * or protocol B P_RECV_ACK
259
		 * or protocol A "HANDED_OVER_TO_NETWORK" (SendAck)
P
Philipp Reisner 已提交
260 261 262 263 264 265 266 267 268 269 270 271 272 273 274
		 * or canceled or failed,
		 * or killed from the transfer log due to connection loss.
		 */

		/*
		 * figure out whether to report success or failure.
		 *
		 * report success when at least one of the operations succeeded.
		 * or, to put the other way,
		 * only report failure, when both operations failed.
		 *
		 * what to do about the failures is handled elsewhere.
		 * what we need to do here is just: complete the master_bio.
		 *
		 * local completion error, if any, has been stored as ERR_PTR
275
		 * in private_bio within drbd_request_endio.
P
Philipp Reisner 已提交
276 277 278 279 280 281
		 */
		int ok = (s & RQ_LOCAL_OK) || (s & RQ_NET_OK);
		int error = PTR_ERR(req->private_bio);

		/* remove the request from the conflict detection
		 * respective block_id verification hash */
282 283 284 285 286 287 288
		if (!drbd_interval_empty(&req->i)) {
			struct rb_root *root;

			if (rw == WRITE)
				root = &mdev->write_requests;
			else
				root = &mdev->read_requests;
289
			drbd_remove_request_interval(root, req);
290
		} else if (!(s & RQ_POSTPONED))
P
Philipp Reisner 已提交
291
			D_ASSERT((s & (RQ_NET_MASK & ~RQ_NET_DONE)) == 0);
P
Philipp Reisner 已提交
292

293 294 295 296 297 298 299 300 301 302
		/* Before we can signal completion to the upper layers,
		 * we may need to close the current transfer log epoch.
		 * We are within the request lock, so we can simply compare
		 * the request epoch number with the current transfer log
		 * epoch number.  If they match, increase the current_tle_nr,
		 * and reset the transfer log epoch write_cnt.
		 */
		if (rw == WRITE &&
		    req->epoch == atomic_read(&mdev->tconn->current_tle_nr))
			start_new_tl_epoch(mdev->tconn);
P
Philipp Reisner 已提交
303 304 305 306

		/* Update disk stats */
		_drbd_end_io_acct(mdev, req);

307 308 309 310 311 312 313 314 315 316 317 318
		/* if READ failed,
		 * have it be pushed back to the retry work queue,
		 * so it will re-enter __drbd_make_request,
		 * and be re-assigned to a suitable local or remote path,
		 * or failed if we do not have access to good data anymore.
		 * READA may fail.
		 * WRITE should have used all available paths already.
		 */
		if (!ok && rw == READ)
			req->rq_state |= RQ_POSTPONED;

		if (!(req->rq_state & RQ_POSTPONED)) {
319 320
			m->error = ok ? 0 : (error ?: -EIO);
			m->bio = req->master_bio;
321 322 323 324 325 326 327 328
			req->master_bio = NULL;
		} else {
			/* Assert that this will be _req_is_done()
			 * with this very invokation. */
			/* FIXME:
			 * what about (RQ_LOCAL_PENDING | RQ_LOCAL_ABORTED)?
			 */
			D_ASSERT(!(s & RQ_LOCAL_PENDING));
329
			D_ASSERT((s & RQ_NET_MASK) == 0 || (s & RQ_NET_DONE));
330
		}
P
Philipp Reisner 已提交
331
	}
332
	req_may_be_done(req);
P
Philipp Reisner 已提交
333 334
}

335
static void req_may_be_completed_not_susp(struct drbd_request *req, struct bio_and_error *m)
336
{
337
	struct drbd_conf *mdev = req->w.mdev;
338

339
	if (!drbd_suspended(mdev))
340
		req_may_be_completed(req, m);
341 342
}

P
Philipp Reisner 已提交
343 344 345 346 347 348 349 350 351 352 353 354
/* obviously this could be coded as many single functions
 * instead of one huge switch,
 * or by putting the code directly in the respective locations
 * (as it has been before).
 *
 * but having it this way
 *  enforces that it is all in this one place, where it is easier to audit,
 *  it makes it obvious that whatever "event" "happens" to a request should
 *  happen "atomically" within the req_lock,
 *  and it enforces that we have to think in a very structured manner
 *  about the "events" that may happen to a request during its life time ...
 */
355
int __req_mod(struct drbd_request *req, enum drbd_req_event what,
P
Philipp Reisner 已提交
356 357
		struct bio_and_error *m)
{
358
	struct drbd_conf *mdev = req->w.mdev;
359
	struct net_conf *nc;
360
	int p, rv = 0;
361 362 363

	if (m)
		m->bio = NULL;
P
Philipp Reisner 已提交
364 365 366 367 368 369 370 371

	switch (what) {
	default:
		dev_err(DEV, "LOGIC BUG in %s:%u\n", __FILE__ , __LINE__);
		break;

	/* does not happen...
	 * initialization done in drbd_req_new
372
	case CREATED:
P
Philipp Reisner 已提交
373 374 375
		break;
		*/

376
	case TO_BE_SENT: /* via network */
377
		/* reached via __drbd_make_request
P
Philipp Reisner 已提交
378 379 380
		 * and from w_read_retry_remote */
		D_ASSERT(!(req->rq_state & RQ_NET_MASK));
		req->rq_state |= RQ_NET_PENDING;
381 382 383 384
		rcu_read_lock();
		nc = rcu_dereference(mdev->tconn->net_conf);
		p = nc->wire_protocol;
		rcu_read_unlock();
385 386 387
		req->rq_state |=
			p == DRBD_PROT_C ? RQ_EXP_WRITE_ACK :
			p == DRBD_PROT_B ? RQ_EXP_RECEIVE_ACK : 0;
P
Philipp Reisner 已提交
388 389 390
		inc_ap_pending(mdev);
		break;

391
	case TO_BE_SUBMITTED: /* locally */
392
		/* reached via __drbd_make_request */
P
Philipp Reisner 已提交
393 394 395 396
		D_ASSERT(!(req->rq_state & RQ_LOCAL_MASK));
		req->rq_state |= RQ_LOCAL_PENDING;
		break;

397
	case COMPLETED_OK:
398
		if (req->rq_state & RQ_WRITE)
399
			mdev->writ_cnt += req->i.size >> 9;
P
Philipp Reisner 已提交
400
		else
401
			mdev->read_cnt += req->i.size >> 9;
P
Philipp Reisner 已提交
402 403 404 405

		req->rq_state |= (RQ_LOCAL_COMPLETED|RQ_LOCAL_OK);
		req->rq_state &= ~RQ_LOCAL_PENDING;

406
		maybe_wakeup_conflicting_requests(req);
407
		req_may_be_completed_not_susp(req, m);
P
Philipp Reisner 已提交
408 409
		break;

410 411
	case ABORT_DISK_IO:
		req->rq_state |= RQ_LOCAL_ABORTED;
412
		req_may_be_completed_not_susp(req, m);
413 414
		break;

415
	case WRITE_COMPLETED_WITH_ERROR:
P
Philipp Reisner 已提交
416 417 418
		req->rq_state |= RQ_LOCAL_COMPLETED;
		req->rq_state &= ~RQ_LOCAL_PENDING;

419
		__drbd_chk_io_error(mdev, false);
420
		maybe_wakeup_conflicting_requests(req);
421
		req_may_be_completed_not_susp(req, m);
P
Philipp Reisner 已提交
422 423
		break;

424
	case READ_AHEAD_COMPLETED_WITH_ERROR:
P
Philipp Reisner 已提交
425 426 427
		/* it is legal to fail READA */
		req->rq_state |= RQ_LOCAL_COMPLETED;
		req->rq_state &= ~RQ_LOCAL_PENDING;
428
		req_may_be_completed_not_susp(req, m);
P
Philipp Reisner 已提交
429 430
		break;

431
	case READ_COMPLETED_WITH_ERROR:
432
		drbd_set_out_of_sync(mdev, req->i.sector, req->i.size);
P
Philipp Reisner 已提交
433 434 435 436 437 438

		req->rq_state |= RQ_LOCAL_COMPLETED;
		req->rq_state &= ~RQ_LOCAL_PENDING;

		D_ASSERT(!(req->rq_state & RQ_NET_MASK));

439
		__drbd_chk_io_error(mdev, false);
440
		req_may_be_completed_not_susp(req, m);
441
		break;
P
Philipp Reisner 已提交
442

443
	case QUEUE_FOR_NET_READ:
P
Philipp Reisner 已提交
444 445 446 447
		/* READ or READA, and
		 * no local disk,
		 * or target area marked as invalid,
		 * or just got an io-error. */
448
		/* from __drbd_make_request
P
Philipp Reisner 已提交
449 450
		 * or from bio_endio during read io-error recovery */

451 452 453
		/* So we can verify the handle in the answer packet.
		 * Corresponding drbd_remove_request_interval is in
		 * req_may_be_completed() */
454
		D_ASSERT(drbd_interval_empty(&req->i));
455
		drbd_insert_interval(&mdev->read_requests, &req->i);
P
Philipp Reisner 已提交
456

457
		set_bit(UNPLUG_REMOTE, &mdev->flags);
P
Philipp Reisner 已提交
458 459

		D_ASSERT(req->rq_state & RQ_NET_PENDING);
460
		D_ASSERT((req->rq_state & RQ_LOCAL_MASK) == 0);
P
Philipp Reisner 已提交
461
		req->rq_state |= RQ_NET_QUEUED;
462
		req->w.cb = w_send_read_req;
463
		drbd_queue_work(&mdev->tconn->sender_work, &req->w);
P
Philipp Reisner 已提交
464 465
		break;

466
	case QUEUE_FOR_NET_WRITE:
P
Philipp Reisner 已提交
467
		/* assert something? */
468
		/* from __drbd_make_request only */
P
Philipp Reisner 已提交
469

470 471
		/* Corresponding drbd_remove_request_interval is in
		 * req_may_be_completed() */
472
		D_ASSERT(drbd_interval_empty(&req->i));
473
		drbd_insert_interval(&mdev->write_requests, &req->i);
P
Philipp Reisner 已提交
474 475 476 477 478 479 480 481 482 483

		/* NOTE
		 * In case the req ended up on the transfer log before being
		 * queued on the worker, it could lead to this request being
		 * missed during cleanup after connection loss.
		 * So we have to do both operations here,
		 * within the same lock that protects the transfer log.
		 *
		 * _req_add_to_epoch(req); this has to be after the
		 * _maybe_start_new_epoch(req); which happened in
484
		 * __drbd_make_request, because we now may set the bit
P
Philipp Reisner 已提交
485 486 487 488
		 * again ourselves to close the current epoch.
		 *
		 * Add req to the (now) current epoch (barrier). */

489 490 491 492 493
		/* otherwise we may lose an unplug, which may cause some remote
		 * io-scheduler timeout to expire, increasing maximum latency,
		 * hurting performance. */
		set_bit(UNPLUG_REMOTE, &mdev->flags);

P
Philipp Reisner 已提交
494 495 496 497
		/* queue work item to send data */
		D_ASSERT(req->rq_state & RQ_NET_PENDING);
		req->rq_state |= RQ_NET_QUEUED;
		req->w.cb =  w_send_dblock;
498
		drbd_queue_work(&mdev->tconn->sender_work, &req->w);
P
Philipp Reisner 已提交
499 500

		/* close the epoch, in case it outgrew the limit */
501 502 503 504
		rcu_read_lock();
		nc = rcu_dereference(mdev->tconn->net_conf);
		p = nc->max_epoch_size;
		rcu_read_unlock();
505 506
		if (mdev->tconn->current_tle_writes >= p)
			start_new_tl_epoch(mdev->tconn);
P
Philipp Reisner 已提交
507 508 509

		break;

510
	case QUEUE_FOR_SEND_OOS:
511
		req->rq_state |= RQ_NET_QUEUED;
512
		req->w.cb =  w_send_out_of_sync;
513
		drbd_queue_work(&mdev->tconn->sender_work, &req->w);
514 515
		break;

516
	case READ_RETRY_REMOTE_CANCELED:
517 518
	case SEND_CANCELED:
	case SEND_FAILED:
P
Philipp Reisner 已提交
519 520 521 522 523
		/* real cleanup will be done from tl_clear.  just update flags
		 * so it is no longer marked as on the worker queue */
		req->rq_state &= ~RQ_NET_QUEUED;
		/* if we did it right, tl_clear should be scheduled only after
		 * this, so this should not be necessary! */
524
		req_may_be_completed_not_susp(req, m);
P
Philipp Reisner 已提交
525 526
		break;

527
	case HANDED_OVER_TO_NETWORK:
P
Philipp Reisner 已提交
528
		/* assert something? */
529
		if (bio_data_dir(req->master_bio) == WRITE)
530
			atomic_add(req->i.size >> 9, &mdev->ap_in_flight);
531

P
Philipp Reisner 已提交
532
		if (bio_data_dir(req->master_bio) == WRITE &&
533
		    !(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK))) {
P
Philipp Reisner 已提交
534 535 536 537 538 539 540 541 542 543 544 545 546
			/* this is what is dangerous about protocol A:
			 * pretend it was successfully written on the peer. */
			if (req->rq_state & RQ_NET_PENDING) {
				dec_ap_pending(mdev);
				req->rq_state &= ~RQ_NET_PENDING;
				req->rq_state |= RQ_NET_OK;
			} /* else: neg-ack was faster... */
			/* it is still not yet RQ_NET_DONE until the
			 * corresponding epoch barrier got acked as well,
			 * so we know what to dirty on connection loss */
		}
		req->rq_state &= ~RQ_NET_QUEUED;
		req->rq_state |= RQ_NET_SENT;
547
		req_may_be_completed_not_susp(req, m);
548 549 550 551 552 553 554
		break;

	case OOS_HANDED_TO_NETWORK:
		/* Was not set PENDING, no longer QUEUED, so is now DONE
		 * as far as this connection is concerned. */
		req->rq_state &= ~RQ_NET_QUEUED;
		req->rq_state |= RQ_NET_DONE;
555
		req_may_be_completed_not_susp(req, m);
P
Philipp Reisner 已提交
556 557
		break;

558
	case CONNECTION_LOST_WHILE_PENDING:
P
Philipp Reisner 已提交
559 560 561 562
		/* transfer log cleanup after connection loss */
		/* assert something? */
		if (req->rq_state & RQ_NET_PENDING)
			dec_ap_pending(mdev);
563 564 565

		p = !(req->rq_state & RQ_WRITE) && req->rq_state & RQ_NET_PENDING;

P
Philipp Reisner 已提交
566 567
		req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING);
		req->rq_state |= RQ_NET_DONE;
568
		if (req->rq_state & RQ_NET_SENT && req->rq_state & RQ_WRITE)
569
			atomic_sub(req->i.size >> 9, &mdev->ap_in_flight);
570

571
		req_may_be_completed(req, m); /* Allowed while state.susp */
P
Philipp Reisner 已提交
572 573
		break;

574
	case DISCARD_WRITE:
P
Philipp Reisner 已提交
575 576 577 578 579
		/* for discarded conflicting writes of multiple primaries,
		 * there is no need to keep anything in the tl, potential
		 * node crashes are covered by the activity log. */
		req->rq_state |= RQ_NET_DONE;
		/* fall through */
580
	case WRITE_ACKED_BY_PEER_AND_SIS:
581
	case WRITE_ACKED_BY_PEER:
582 583
		if (what == WRITE_ACKED_BY_PEER_AND_SIS)
			req->rq_state |= RQ_NET_SIS;
584
		D_ASSERT(req->rq_state & RQ_EXP_WRITE_ACK);
P
Philipp Reisner 已提交
585
		/* protocol C; successfully written on peer.
586
		 * Nothing more to do here.
P
Philipp Reisner 已提交
587
		 * We want to keep the tl in place for all protocols, to cater
588
		 * for volatile write-back caches on lower level devices. */
P
Philipp Reisner 已提交
589

590
		goto ack_common;
591
	case RECV_ACKED_BY_PEER:
592
		D_ASSERT(req->rq_state & RQ_EXP_RECEIVE_ACK);
P
Philipp Reisner 已提交
593
		/* protocol B; pretends to be successfully written on peer.
594
		 * see also notes above in HANDED_OVER_TO_NETWORK about
P
Philipp Reisner 已提交
595
		 * protocol != C */
596
	ack_common:
P
Philipp Reisner 已提交
597 598 599
		req->rq_state |= RQ_NET_OK;
		D_ASSERT(req->rq_state & RQ_NET_PENDING);
		dec_ap_pending(mdev);
600
		atomic_sub(req->i.size >> 9, &mdev->ap_in_flight);
P
Philipp Reisner 已提交
601
		req->rq_state &= ~RQ_NET_PENDING;
602
		maybe_wakeup_conflicting_requests(req);
603
		req_may_be_completed_not_susp(req, m);
P
Philipp Reisner 已提交
604 605
		break;

606
	case POSTPONE_WRITE:
607 608
		D_ASSERT(req->rq_state & RQ_EXP_WRITE_ACK);
		/* If this node has already detected the write conflict, the
609 610 611 612 613
		 * worker will be waiting on misc_wait.  Wake it up once this
		 * request has completed locally.
		 */
		D_ASSERT(req->rq_state & RQ_NET_PENDING);
		req->rq_state |= RQ_POSTPONED;
614
		maybe_wakeup_conflicting_requests(req);
615
		req_may_be_completed_not_susp(req, m);
616 617
		break;

618
	case NEG_ACKED:
P
Philipp Reisner 已提交
619
		/* assert something? */
620
		if (req->rq_state & RQ_NET_PENDING) {
P
Philipp Reisner 已提交
621
			dec_ap_pending(mdev);
622 623
			if (req->rq_state & RQ_WRITE)
				atomic_sub(req->i.size >> 9, &mdev->ap_in_flight);
624
		}
P
Philipp Reisner 已提交
625 626 627
		req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING);

		req->rq_state |= RQ_NET_DONE;
628

629
		maybe_wakeup_conflicting_requests(req);
630
		req_may_be_completed_not_susp(req, m);
631
		/* else: done by HANDED_OVER_TO_NETWORK */
P
Philipp Reisner 已提交
632 633
		break;

634
	case FAIL_FROZEN_DISK_IO:
635 636 637
		if (!(req->rq_state & RQ_LOCAL_COMPLETED))
			break;

638
		req_may_be_completed(req, m); /* Allowed while state.susp */
639 640
		break;

641
	case RESTART_FROZEN_DISK_IO:
642 643 644 645 646 647 648 649 650 651 652
		if (!(req->rq_state & RQ_LOCAL_COMPLETED))
			break;

		req->rq_state &= ~RQ_LOCAL_COMPLETED;

		rv = MR_READ;
		if (bio_data_dir(req->master_bio) == WRITE)
			rv = MR_WRITE;

		get_ldev(mdev);
		req->w.cb = w_restart_disk_io;
653
		drbd_queue_work(&mdev->tconn->sender_work, &req->w);
654 655
		break;

656
	case RESEND:
657
		/* If RQ_NET_OK is already set, we got a P_WRITE_ACK or P_RECV_ACK
658
		   before the connection loss (B&C only); only P_BARRIER_ACK was missing.
659 660
		   Throwing them out of the TL here by pretending we got a BARRIER_ACK.
		   During connection handshake, we ensure that the peer was not rebooted. */
661 662
		if (!(req->rq_state & RQ_NET_OK)) {
			if (req->w.cb) {
663
				/* w.cb expected to be w_send_dblock, or w_send_read_req */
664
				drbd_queue_work(&mdev->tconn->sender_work, &req->w);
665 666 667 668
				rv = req->rq_state & RQ_WRITE ? MR_WRITE : MR_READ;
			}
			break;
		}
669
		/* else, fall through to BARRIER_ACKED */
670

671
	case BARRIER_ACKED:
672 673 674
		if (!(req->rq_state & RQ_WRITE))
			break;

P
Philipp Reisner 已提交
675
		if (req->rq_state & RQ_NET_PENDING) {
676
			/* barrier came in before all requests were acked.
P
Philipp Reisner 已提交
677 678
			 * this is bad, because if the connection is lost now,
			 * we won't be able to clean them up... */
679
			dev_err(DEV, "FIXME (BARRIER_ACKED but pending)\n");
P
Philipp Reisner 已提交
680
		}
681 682
		if ((req->rq_state & RQ_NET_MASK) != 0) {
			req->rq_state |= RQ_NET_DONE;
683
			if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)))
684
				atomic_sub(req->i.size>>9, &mdev->ap_in_flight);
685
		}
686
		req_may_be_done(req); /* Allowed while state.susp */
P
Philipp Reisner 已提交
687 688
		break;

689
	case DATA_RECEIVED:
P
Philipp Reisner 已提交
690 691 692 693
		D_ASSERT(req->rq_state & RQ_NET_PENDING);
		dec_ap_pending(mdev);
		req->rq_state &= ~RQ_NET_PENDING;
		req->rq_state |= (RQ_NET_OK|RQ_NET_DONE);
694
		req_may_be_completed_not_susp(req, m);
P
Philipp Reisner 已提交
695 696
		break;
	};
697 698

	return rv;
P
Philipp Reisner 已提交
699 700 701 702 703 704 705 706 707
}

/* we may do a local read if:
 * - we are consistent (of course),
 * - or we are generally inconsistent,
 *   BUT we are still/already IN SYNC for this area.
 *   since size may be bigger than BM_BLOCK_SIZE,
 *   we may need to check several bits.
 */
708
static bool drbd_may_do_local_read(struct drbd_conf *mdev, sector_t sector, int size)
P
Philipp Reisner 已提交
709 710 711 712 713
{
	unsigned long sbnr, ebnr;
	sector_t esector, nr_sectors;

	if (mdev->state.disk == D_UP_TO_DATE)
714
		return true;
715
	if (mdev->state.disk != D_INCONSISTENT)
716
		return false;
P
Philipp Reisner 已提交
717
	esector = sector + (size >> 9) - 1;
718
	nr_sectors = drbd_get_capacity(mdev->this_bdev);
P
Philipp Reisner 已提交
719 720 721 722 723 724
	D_ASSERT(sector  < nr_sectors);
	D_ASSERT(esector < nr_sectors);

	sbnr = BM_SECT_TO_BIT(sector);
	ebnr = BM_SECT_TO_BIT(esector);

725
	return drbd_bm_count_bits(mdev, sbnr, ebnr) == 0;
P
Philipp Reisner 已提交
726 727
}

728
static bool remote_due_to_read_balancing(struct drbd_conf *mdev, sector_t sector)
729 730 731
{
	enum drbd_read_balancing rbm;
	struct backing_dev_info *bdi;
732
	int stripe_shift;
733 734 735 736 737 738 739 740 741 742 743 744 745 746 747

	if (mdev->state.pdsk < D_UP_TO_DATE)
		return false;

	rcu_read_lock();
	rbm = rcu_dereference(mdev->ldev->disk_conf)->read_balancing;
	rcu_read_unlock();

	switch (rbm) {
	case RB_CONGESTED_REMOTE:
		bdi = &mdev->ldev->backing_bdev->bd_disk->queue->backing_dev_info;
		return bdi_read_congested(bdi);
	case RB_LEAST_PENDING:
		return atomic_read(&mdev->local_cnt) >
			atomic_read(&mdev->ap_pending_cnt) + atomic_read(&mdev->rs_pending_cnt);
748 749 750 751 752 753 754 755
	case RB_32K_STRIPING:  /* stripe_shift = 15 */
	case RB_64K_STRIPING:
	case RB_128K_STRIPING:
	case RB_256K_STRIPING:
	case RB_512K_STRIPING:
	case RB_1M_STRIPING:   /* stripe_shift = 20 */
		stripe_shift = (rbm - RB_32K_STRIPING + 15);
		return (sector >> (stripe_shift - 9)) & 1;
756 757 758 759 760 761 762 763 764 765
	case RB_ROUND_ROBIN:
		return test_and_change_bit(READ_BALANCE_RR, &mdev->flags);
	case RB_PREFER_REMOTE:
		return true;
	case RB_PREFER_LOCAL:
	default:
		return false;
	}
}

766 767 768 769 770 771
/*
 * complete_conflicting_writes  -  wait for any conflicting write requests
 *
 * The write_requests tree contains all active write requests which we
 * currently know about.  Wait for any requests to complete which conflict with
 * the new one.
772 773
 *
 * Only way out: remove the conflicting intervals from the tree.
774
 */
775
static void complete_conflicting_writes(struct drbd_request *req)
776
{
777 778 779 780 781 782 783 784 785
	DEFINE_WAIT(wait);
	struct drbd_conf *mdev = req->w.mdev;
	struct drbd_interval *i;
	sector_t sector = req->i.sector;
	int size = req->i.size;

	i = drbd_find_overlap(&mdev->write_requests, sector, size);
	if (!i)
		return;
786

787 788
	for (;;) {
		prepare_to_wait(&mdev->misc_wait, &wait, TASK_UNINTERRUPTIBLE);
789 790
		i = drbd_find_overlap(&mdev->write_requests, sector, size);
		if (!i)
791 792 793 794 795 796
			break;
		/* Indicate to wake up device->misc_wait on progress.  */
		i->waiting = true;
		spin_unlock_irq(&mdev->tconn->req_lock);
		schedule();
		spin_lock_irq(&mdev->tconn->req_lock);
797
	}
798
	finish_wait(&mdev->misc_wait, &wait);
799 800
}

801
int __drbd_make_request(struct drbd_conf *mdev, struct bio *bio, unsigned long start_time)
P
Philipp Reisner 已提交
802 803 804 805 806
{
	const int rw = bio_rw(bio);
	const int size = bio->bi_size;
	const sector_t sector = bio->bi_sector;
	struct drbd_request *req;
807
	struct net_conf *nc;
808
	int local, remote, send_oos = 0;
809
	int err = 0;
810
	int ret = 0;
811
	union drbd_dev_state s;
P
Philipp Reisner 已提交
812 813 814 815 816 817 818 819 820 821 822

	/* allocate outside of all locks; */
	req = drbd_req_new(mdev, bio);
	if (!req) {
		dec_ap_bio(mdev);
		/* only pass the error to the upper layers.
		 * if user cannot handle io errors, that's not our business. */
		dev_err(DEV, "could not kmalloc() req\n");
		bio_endio(bio, -ENOMEM);
		return 0;
	}
823
	req->start_time = start_time;
P
Philipp Reisner 已提交
824 825 826 827 828 829 830 831 832 833 834

	local = get_ldev(mdev);
	if (!local) {
		bio_put(req->private_bio); /* or we get a bio leak */
		req->private_bio = NULL;
	}
	if (rw == WRITE) {
		remote = 1;
	} else {
		/* READ || READA */
		if (local) {
835 836
			if (!drbd_may_do_local_read(mdev, sector, size) ||
			    remote_due_to_read_balancing(mdev, sector)) {
P
Philipp Reisner 已提交
837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868
				/* we could kick the syncer to
				 * sync this extent asap, wait for
				 * it, then continue locally.
				 * Or just issue the request remotely.
				 */
				local = 0;
				bio_put(req->private_bio);
				req->private_bio = NULL;
				put_ldev(mdev);
			}
		}
		remote = !local && mdev->state.pdsk >= D_UP_TO_DATE;
	}

	/* If we have a disk, but a READA request is mapped to remote,
	 * we are R_PRIMARY, D_INCONSISTENT, SyncTarget.
	 * Just fail that READA request right here.
	 *
	 * THINK: maybe fail all READA when not local?
	 *        or make this configurable...
	 *        if network is slow, READA won't do any good.
	 */
	if (rw == READA && mdev->state.disk >= D_INCONSISTENT && !local) {
		err = -EWOULDBLOCK;
		goto fail_and_free_req;
	}

	/* For WRITES going to the local disk, grab a reference on the target
	 * extent.  This waits for any resync activity in the corresponding
	 * resync extent to finish, and, if necessary, pulls in the target
	 * extent into the activity log, which involves further disk io because
	 * of transactional on-disk meta data updates. */
869 870
	if (rw == WRITE && local && !test_bit(AL_SUSPENDED, &mdev->flags)) {
		req->rq_state |= RQ_IN_ACT_LOG;
871
		drbd_al_begin_io(mdev, &req->i);
872
	}
P
Philipp Reisner 已提交
873

874 875 876
	s = mdev->state;
	remote = remote && drbd_should_do_remote(s);
	send_oos = rw == WRITE && drbd_should_send_out_of_sync(s);
877
	D_ASSERT(!(remote && send_oos));
P
Philipp Reisner 已提交
878

879
	if (!(local || remote) && !drbd_suspended(mdev)) {
L
Lars Ellenberg 已提交
880 881
		if (__ratelimit(&drbd_ratelimit_state))
			dev_err(DEV, "IO ERROR: neither local nor remote disk\n");
882
		err = -EIO;
P
Philipp Reisner 已提交
883 884 885 886
		goto fail_free_complete;
	}

	/* GOOD, everything prepared, grab the spin_lock */
887
	spin_lock_irq(&mdev->tconn->req_lock);
P
Philipp Reisner 已提交
888

889
	if (rw == WRITE) {
890 891 892 893
		/* This may temporarily give up the req_lock,
		 * but will re-aquire it before it returns here.
		 * Needs to be before the check on drbd_suspended() */
		complete_conflicting_writes(req);
894 895
	}

896
	if (drbd_suspended(mdev)) {
897 898
		/* If we got suspended, use the retry mechanism in
		   drbd_make_request() to restart processing of this
899
		   bio. In the next call to drbd_make_request
900 901
		   we sleep in inc_ap_bio() */
		ret = 1;
902
		spin_unlock_irq(&mdev->tconn->req_lock);
903 904 905
		goto fail_free_complete;
	}

906
	if (remote || send_oos) {
907
		remote = drbd_should_do_remote(mdev->state);
908
		send_oos = rw == WRITE && drbd_should_send_out_of_sync(mdev->state);
909
		D_ASSERT(!(remote && send_oos));
910 911

		if (!(remote || send_oos))
P
Philipp Reisner 已提交
912 913 914
			dev_warn(DEV, "lost connection while grabbing the req_lock!\n");
		if (!(local || remote)) {
			dev_err(DEV, "IO ERROR: neither local nor remote disk\n");
915
			spin_unlock_irq(&mdev->tconn->req_lock);
916
			err = -EIO;
P
Philipp Reisner 已提交
917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939
			goto fail_free_complete;
		}
	}

	/* Update disk stats */
	_drbd_start_io_acct(mdev, req, bio);

	/* NOTE
	 * Actually, 'local' may be wrong here already, since we may have failed
	 * to write to the meta data, and may become wrong anytime because of
	 * local io-error for some other request, which would lead to us
	 * "detaching" the local disk.
	 *
	 * 'remote' may become wrong any time because the network could fail.
	 *
	 * This is a harmless race condition, though, since it is handled
	 * correctly at the appropriate places; so it just defers the failure
	 * of the respective operation.
	 */

	/* mark them early for readability.
	 * this just sets some state flags. */
	if (remote)
940
		_req_mod(req, TO_BE_SENT);
P
Philipp Reisner 已提交
941
	if (local)
942
		_req_mod(req, TO_BE_SUBMITTED);
P
Philipp Reisner 已提交
943

944 945 946 947 948 949
	/* which transfer log epoch does this belong to? */
	req->epoch = atomic_read(&mdev->tconn->current_tle_nr);
	if (rw == WRITE)
		mdev->tconn->current_tle_writes++;

	list_add_tail(&req->tl_requests, &mdev->tconn->transfer_log);
950

P
Philipp Reisner 已提交
951 952 953 954 955 956 957 958
	/* NOTE remote first: to get the concurrent write detection right,
	 * we must register the request before start of local IO.  */
	if (remote) {
		/* either WRITE and C_CONNECTED,
		 * or READ, and no local disk,
		 * or READ, but not in sync.
		 */
		_req_mod(req, (rw == WRITE)
959 960
				? QUEUE_FOR_NET_WRITE
				: QUEUE_FOR_NET_READ);
P
Philipp Reisner 已提交
961
	}
962
	if (send_oos && drbd_set_out_of_sync(mdev, sector, size))
963
		_req_mod(req, QUEUE_FOR_SEND_OOS);
964

965 966
	rcu_read_lock();
	nc = rcu_dereference(mdev->tconn->net_conf);
967
	if (remote &&
968
	    nc->on_congestion != OC_BLOCK && mdev->tconn->agreed_pro_version >= 96) {
969 970
		int congested = 0;

971 972
		if (nc->cong_fill &&
		    atomic_read(&mdev->ap_in_flight) >= nc->cong_fill) {
973 974 975 976
			dev_info(DEV, "Congestion-fill threshold reached\n");
			congested = 1;
		}

977
		if (mdev->act_log->used >= nc->cong_extents) {
978 979 980 981
			dev_info(DEV, "Congestion-extents threshold reached\n");
			congested = 1;
		}

982
		if (congested) {
983 984 985
			if (mdev->tconn->current_tle_writes)
				/* start a new epoch for non-mirrored writes */
				start_new_tl_epoch(mdev->tconn);
986

987
			if (nc->on_congestion == OC_PULL_AHEAD)
988
				_drbd_set_state(_NS(mdev, conn, C_AHEAD), 0, NULL);
989
			else  /*nc->on_congestion == OC_DISCONNECT */
990 991 992
				_drbd_set_state(_NS(mdev, conn, C_DISCONNECTING), 0, NULL);
		}
	}
993
	rcu_read_unlock();
994

995
	spin_unlock_irq(&mdev->tconn->req_lock);
P
Philipp Reisner 已提交
996 997 998 999

	if (local) {
		req->private_bio->bi_bdev = mdev->ldev->backing_bdev;

1000 1001 1002 1003 1004 1005
		/* State may have changed since we grabbed our reference on the
		 * mdev->ldev member. Double check, and short-circuit to endio.
		 * In case the last activity log transaction failed to get on
		 * stable storage, and this is a WRITE, we may not even submit
		 * this bio. */
		if (get_ldev(mdev)) {
1006 1007 1008
			if (drbd_insert_fault(mdev,   rw == WRITE ? DRBD_FAULT_DT_WR
						    : rw == READ  ? DRBD_FAULT_DT_RD
						    :               DRBD_FAULT_DT_RA))
1009 1010 1011 1012 1013
				bio_endio(req->private_bio, -EIO);
			else
				generic_make_request(req->private_bio);
			put_ldev(mdev);
		} else
P
Philipp Reisner 已提交
1014 1015 1016 1017 1018 1019
			bio_endio(req->private_bio, -EIO);
	}

	return 0;

fail_free_complete:
1020
	if (req->rq_state & RQ_IN_ACT_LOG)
1021
		drbd_al_complete_io(mdev, &req->i);
P
Philipp Reisner 已提交
1022
fail_and_free_req:
1023
	if (local) {
P
Philipp Reisner 已提交
1024 1025 1026 1027
		bio_put(req->private_bio);
		req->private_bio = NULL;
		put_ldev(mdev);
	}
1028 1029 1030
	if (!ret)
		bio_endio(bio, err);

P
Philipp Reisner 已提交
1031 1032 1033
	drbd_req_free(req);
	dec_ap_bio(mdev);

1034
	return ret;
P
Philipp Reisner 已提交
1035 1036
}

1037
int drbd_make_request(struct request_queue *q, struct bio *bio)
P
Philipp Reisner 已提交
1038 1039
{
	struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata;
1040
	unsigned long start_time;
P
Philipp Reisner 已提交
1041

1042 1043
	start_time = jiffies;

P
Philipp Reisner 已提交
1044 1045 1046 1047
	/*
	 * what we "blindly" assume:
	 */
	D_ASSERT(bio->bi_size > 0);
1048
	D_ASSERT(IS_ALIGNED(bio->bi_size, 512));
P
Philipp Reisner 已提交
1049

1050 1051 1052 1053 1054
	do {
		inc_ap_bio(mdev);
	} while (__drbd_make_request(mdev, bio, start_time));

	return 0;
P
Philipp Reisner 已提交
1055 1056
}

1057 1058 1059
/* This is called by bio_add_page().
 *
 * q->max_hw_sectors and other global limits are already enforced there.
P
Philipp Reisner 已提交
1060
 *
1061 1062 1063 1064
 * We need to call down to our lower level device,
 * in case it has special restrictions.
 *
 * We also may need to enforce configured max-bio-bvecs limits.
P
Philipp Reisner 已提交
1065 1066
 *
 * As long as the BIO is empty we have to allow at least one bvec,
1067
 * regardless of size and offset, so no need to ask lower levels.
P
Philipp Reisner 已提交
1068 1069 1070 1071 1072
 */
int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec)
{
	struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata;
	unsigned int bio_size = bvm->bi_size;
1073 1074 1075 1076
	int limit = DRBD_MAX_BIO_SIZE;
	int backing_limit;

	if (bio_size && get_ldev(mdev)) {
P
Philipp Reisner 已提交
1077 1078
		struct request_queue * const b =
			mdev->ldev->backing_bdev->bd_disk->queue;
1079
		if (b->merge_bvec_fn) {
P
Philipp Reisner 已提交
1080 1081 1082 1083 1084 1085 1086
			backing_limit = b->merge_bvec_fn(b, bvm, bvec);
			limit = min(limit, backing_limit);
		}
		put_ldev(mdev);
	}
	return limit;
}
1087

1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099
struct drbd_request *find_oldest_request(struct drbd_tconn *tconn)
{
	/* Walk the transfer log,
	 * and find the oldest not yet completed request */
	struct drbd_request *r;
	list_for_each_entry(r, &tconn->transfer_log, tl_requests) {
		if (r->rq_state & (RQ_NET_PENDING|RQ_LOCAL_PENDING))
			return r;
	}
	return NULL;
}

1100 1101 1102
void request_timer_fn(unsigned long data)
{
	struct drbd_conf *mdev = (struct drbd_conf *) data;
1103
	struct drbd_tconn *tconn = mdev->tconn;
1104
	struct drbd_request *req; /* oldest request */
1105
	struct net_conf *nc;
1106
	unsigned long ent = 0, dt = 0, et, nt; /* effective timeout = ko_count * timeout */
1107
	unsigned long now;
1108 1109 1110

	rcu_read_lock();
	nc = rcu_dereference(tconn->net_conf);
1111 1112
	if (nc && mdev->state.conn >= C_WF_REPORT_PARAMS)
		ent = nc->timeout * HZ/10 * nc->ko_count;
1113

1114
	if (get_ldev(mdev)) { /* implicit state.disk >= D_INCONSISTENT */
1115 1116 1117
		dt = rcu_dereference(mdev->ldev->disk_conf)->disk_timeout * HZ / 10;
		put_ldev(mdev);
	}
1118
	rcu_read_unlock();
1119

1120 1121
	et = min_not_zero(dt, ent);

1122
	if (!et)
1123 1124
		return; /* Recurring timer stopped */

1125 1126
	now = jiffies;

1127
	spin_lock_irq(&tconn->req_lock);
1128 1129
	req = find_oldest_request(tconn);
	if (!req) {
1130
		spin_unlock_irq(&tconn->req_lock);
1131
		mod_timer(&mdev->request_timer, now + et);
1132 1133 1134
		return;
	}

1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155
	/* The request is considered timed out, if
	 * - we have some effective timeout from the configuration,
	 *   with above state restrictions applied,
	 * - the oldest request is waiting for a response from the network
	 *   resp. the local disk,
	 * - the oldest request is in fact older than the effective timeout,
	 * - the connection was established (resp. disk was attached)
	 *   for longer than the timeout already.
	 * Note that for 32bit jiffies and very stable connections/disks,
	 * we may have a wrap around, which is catched by
	 *   !time_in_range(now, last_..._jif, last_..._jif + timeout).
	 *
	 * Side effect: once per 32bit wrap-around interval, which means every
	 * ~198 days with 250 HZ, we have a window where the timeout would need
	 * to expire twice (worst case) to become effective. Good enough.
	 */
	if (ent && req->rq_state & RQ_NET_PENDING &&
		 time_after(now, req->start_time + ent) &&
		!time_in_range(now, tconn->last_reconnect_jif, tconn->last_reconnect_jif + ent)) {
		dev_warn(DEV, "Remote failed to finish a request within ko-count * timeout\n");
		_drbd_set_state(_NS(mdev, conn, C_TIMEOUT), CS_VERBOSE | CS_HARD, NULL);
1156
	}
1157 1158 1159 1160 1161
	if (dt && req->rq_state & RQ_LOCAL_PENDING && req->w.mdev == mdev &&
		 time_after(now, req->start_time + dt) &&
		!time_in_range(now, mdev->last_reattach_jif, mdev->last_reattach_jif + dt)) {
		dev_warn(DEV, "Local backing device failed to meet the disk-timeout\n");
		__drbd_chk_io_error(mdev, 1);
1162
	}
1163
	nt = (time_after(now, req->start_time + et) ? now : req->start_time) + et;
1164
	spin_unlock_irq(&tconn->req_lock);
1165
	mod_timer(&mdev->request_timer, nt);
1166
}