4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
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>.
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)
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.
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.
26 #include <linux/module.h>
28 #include <linux/slab.h>
29 #include <linux/drbd.h>
34 static bool drbd_may_do_local_read(struct drbd_device
*device
, sector_t sector
, int size
);
36 /* Update disk stats at start of I/O request */
37 static void _drbd_start_io_acct(struct drbd_device
*device
, struct drbd_request
*req
)
39 generic_start_io_acct(bio_data_dir(req
->master_bio
), req
->i
.size
>> 9,
40 &device
->vdisk
->part0
);
43 /* Update disk stats when completing request upwards */
44 static void _drbd_end_io_acct(struct drbd_device
*device
, struct drbd_request
*req
)
46 generic_end_io_acct(bio_data_dir(req
->master_bio
),
47 &device
->vdisk
->part0
, req
->start_jif
);
50 static struct drbd_request
*drbd_req_new(struct drbd_device
*device
, struct bio
*bio_src
)
52 struct drbd_request
*req
;
54 req
= mempool_alloc(drbd_request_mempool
, GFP_NOIO
);
57 memset(req
, 0, sizeof(*req
));
59 drbd_req_make_private_bio(req
, bio_src
);
60 req
->rq_state
= (bio_data_dir(bio_src
) == WRITE
? RQ_WRITE
: 0)
61 | (bio_op(bio_src
) == REQ_OP_WRITE_SAME
? RQ_WSAME
: 0)
62 | (bio_op(bio_src
) == REQ_OP_WRITE_ZEROES
? RQ_UNMAP
: 0)
63 | (bio_op(bio_src
) == REQ_OP_DISCARD
? RQ_UNMAP
: 0);
65 req
->master_bio
= bio_src
;
68 drbd_clear_interval(&req
->i
);
69 req
->i
.sector
= bio_src
->bi_iter
.bi_sector
;
70 req
->i
.size
= bio_src
->bi_iter
.bi_size
;
72 req
->i
.waiting
= false;
74 INIT_LIST_HEAD(&req
->tl_requests
);
75 INIT_LIST_HEAD(&req
->w
.list
);
76 INIT_LIST_HEAD(&req
->req_pending_master_completion
);
77 INIT_LIST_HEAD(&req
->req_pending_local
);
79 /* one reference to be put by __drbd_make_request */
80 atomic_set(&req
->completion_ref
, 1);
81 /* one kref as long as completion_ref > 0 */
82 kref_init(&req
->kref
);
86 static void drbd_remove_request_interval(struct rb_root
*root
,
87 struct drbd_request
*req
)
89 struct drbd_device
*device
= req
->device
;
90 struct drbd_interval
*i
= &req
->i
;
92 drbd_remove_interval(root
, i
);
94 /* Wake up any processes waiting for this request to complete. */
96 wake_up(&device
->misc_wait
);
99 void drbd_req_destroy(struct kref
*kref
)
101 struct drbd_request
*req
= container_of(kref
, struct drbd_request
, kref
);
102 struct drbd_device
*device
= req
->device
;
103 const unsigned s
= req
->rq_state
;
105 if ((req
->master_bio
&& !(s
& RQ_POSTPONED
)) ||
106 atomic_read(&req
->completion_ref
) ||
107 (s
& RQ_LOCAL_PENDING
) ||
108 ((s
& RQ_NET_MASK
) && !(s
& RQ_NET_DONE
))) {
109 drbd_err(device
, "drbd_req_destroy: Logic BUG rq_state = 0x%x, completion_ref = %d\n",
110 s
, atomic_read(&req
->completion_ref
));
114 /* If called from mod_rq_state (expected normal case) or
115 * drbd_send_and_submit (the less likely normal path), this holds the
116 * req_lock, and req->tl_requests will typicaly be on ->transfer_log,
117 * though it may be still empty (never added to the transfer log).
119 * If called from do_retry(), we do NOT hold the req_lock, but we are
120 * still allowed to unconditionally list_del(&req->tl_requests),
121 * because it will be on a local on-stack list only. */
122 list_del_init(&req
->tl_requests
);
124 /* finally remove the request from the conflict detection
125 * respective block_id verification interval tree. */
126 if (!drbd_interval_empty(&req
->i
)) {
127 struct rb_root
*root
;
130 root
= &device
->write_requests
;
132 root
= &device
->read_requests
;
133 drbd_remove_request_interval(root
, req
);
134 } else if (s
& (RQ_NET_MASK
& ~RQ_NET_DONE
) && req
->i
.size
!= 0)
135 drbd_err(device
, "drbd_req_destroy: Logic BUG: interval empty, but: rq_state=0x%x, sect=%llu, size=%u\n",
136 s
, (unsigned long long)req
->i
.sector
, req
->i
.size
);
138 /* if it was a write, we may have to set the corresponding
139 * bit(s) out-of-sync first. If it had a local part, we need to
140 * release the reference to the activity log. */
142 /* Set out-of-sync unless both OK flags are set
143 * (local only or remote failed).
144 * Other places where we set out-of-sync:
145 * READ with local io-error */
147 /* There is a special case:
148 * we may notice late that IO was suspended,
149 * and postpone, or schedule for retry, a write,
150 * before it even was submitted or sent.
151 * In that case we do not want to touch the bitmap at all.
153 if ((s
& (RQ_POSTPONED
|RQ_LOCAL_MASK
|RQ_NET_MASK
)) != RQ_POSTPONED
) {
154 if (!(s
& RQ_NET_OK
) || !(s
& RQ_LOCAL_OK
))
155 drbd_set_out_of_sync(device
, req
->i
.sector
, req
->i
.size
);
157 if ((s
& RQ_NET_OK
) && (s
& RQ_LOCAL_OK
) && (s
& RQ_NET_SIS
))
158 drbd_set_in_sync(device
, req
->i
.sector
, req
->i
.size
);
161 /* one might be tempted to move the drbd_al_complete_io
162 * to the local io completion callback drbd_request_endio.
163 * but, if this was a mirror write, we may only
164 * drbd_al_complete_io after this is RQ_NET_DONE,
165 * otherwise the extent could be dropped from the al
166 * before it has actually been written on the peer.
167 * if we crash before our peer knows about the request,
168 * but after the extent has been dropped from the al,
169 * we would forget to resync the corresponding extent.
171 if (s
& RQ_IN_ACT_LOG
) {
172 if (get_ldev_if_state(device
, D_FAILED
)) {
173 drbd_al_complete_io(device
, &req
->i
);
175 } else if (__ratelimit(&drbd_ratelimit_state
)) {
176 drbd_warn(device
, "Should have called drbd_al_complete_io(, %llu, %u), "
177 "but my Disk seems to have failed :(\n",
178 (unsigned long long) req
->i
.sector
, req
->i
.size
);
183 mempool_free(req
, drbd_request_mempool
);
186 static void wake_all_senders(struct drbd_connection
*connection
)
188 wake_up(&connection
->sender_work
.q_wait
);
191 /* must hold resource->req_lock */
192 void start_new_tl_epoch(struct drbd_connection
*connection
)
194 /* no point closing an epoch, if it is empty, anyways. */
195 if (connection
->current_tle_writes
== 0)
198 connection
->current_tle_writes
= 0;
199 atomic_inc(&connection
->current_tle_nr
);
200 wake_all_senders(connection
);
203 void complete_master_bio(struct drbd_device
*device
,
204 struct bio_and_error
*m
)
206 m
->bio
->bi_status
= errno_to_blk_status(m
->error
);
212 /* Helper for __req_mod().
213 * Set m->bio to the master bio, if it is fit to be completed,
214 * or leave it alone (it is initialized to NULL in __req_mod),
215 * if it has already been completed, or cannot be completed yet.
216 * If m->bio is set, the error status to be returned is placed in m->error.
219 void drbd_req_complete(struct drbd_request
*req
, struct bio_and_error
*m
)
221 const unsigned s
= req
->rq_state
;
222 struct drbd_device
*device
= req
->device
;
225 /* we must not complete the master bio, while it is
226 * still being processed by _drbd_send_zc_bio (drbd_send_dblock)
227 * not yet acknowledged by the peer
228 * not yet completed by the local io subsystem
229 * these flags may get cleared in any order by
232 * the bio_endio completion callbacks.
234 if ((s
& RQ_LOCAL_PENDING
&& !(s
& RQ_LOCAL_ABORTED
)) ||
235 (s
& RQ_NET_QUEUED
) || (s
& RQ_NET_PENDING
) ||
236 (s
& RQ_COMPLETION_SUSP
)) {
237 drbd_err(device
, "drbd_req_complete: Logic BUG rq_state = 0x%x\n", s
);
241 if (!req
->master_bio
) {
242 drbd_err(device
, "drbd_req_complete: Logic BUG, master_bio == NULL!\n");
247 * figure out whether to report success or failure.
249 * report success when at least one of the operations succeeded.
250 * or, to put the other way,
251 * only report failure, when both operations failed.
253 * what to do about the failures is handled elsewhere.
254 * what we need to do here is just: complete the master_bio.
256 * local completion error, if any, has been stored as ERR_PTR
257 * in private_bio within drbd_request_endio.
259 ok
= (s
& RQ_LOCAL_OK
) || (s
& RQ_NET_OK
);
260 error
= PTR_ERR(req
->private_bio
);
262 /* Before we can signal completion to the upper layers,
263 * we may need to close the current transfer log epoch.
264 * We are within the request lock, so we can simply compare
265 * the request epoch number with the current transfer log
266 * epoch number. If they match, increase the current_tle_nr,
267 * and reset the transfer log epoch write_cnt.
269 if (op_is_write(bio_op(req
->master_bio
)) &&
270 req
->epoch
== atomic_read(&first_peer_device(device
)->connection
->current_tle_nr
))
271 start_new_tl_epoch(first_peer_device(device
)->connection
);
273 /* Update disk stats */
274 _drbd_end_io_acct(device
, req
);
277 * have it be pushed back to the retry work queue,
278 * so it will re-enter __drbd_make_request(),
279 * and be re-assigned to a suitable local or remote path,
280 * or failed if we do not have access to good data anymore.
282 * Unless it was failed early by __drbd_make_request(),
283 * because no path was available, in which case
284 * it was not even added to the transfer_log.
286 * read-ahead may fail, and will not be retried.
288 * WRITE should have used all available paths already.
291 bio_op(req
->master_bio
) == REQ_OP_READ
&&
292 !(req
->master_bio
->bi_opf
& REQ_RAHEAD
) &&
293 !list_empty(&req
->tl_requests
))
294 req
->rq_state
|= RQ_POSTPONED
;
296 if (!(req
->rq_state
& RQ_POSTPONED
)) {
297 m
->error
= ok
? 0 : (error
?: -EIO
);
298 m
->bio
= req
->master_bio
;
299 req
->master_bio
= NULL
;
300 /* We leave it in the tree, to be able to verify later
301 * write-acks in protocol != C during resync.
302 * But we mark it as "complete", so it won't be counted as
303 * conflict in a multi-primary setup. */
304 req
->i
.completed
= true;
308 wake_up(&device
->misc_wait
);
310 /* Either we are about to complete to upper layers,
311 * or we will restart this request.
312 * In either case, the request object will be destroyed soon,
313 * so better remove it from all lists. */
314 list_del_init(&req
->req_pending_master_completion
);
317 /* still holds resource->req_lock */
318 static void drbd_req_put_completion_ref(struct drbd_request
*req
, struct bio_and_error
*m
, int put
)
320 struct drbd_device
*device
= req
->device
;
321 D_ASSERT(device
, m
|| (req
->rq_state
& RQ_POSTPONED
));
326 if (!atomic_sub_and_test(put
, &req
->completion_ref
))
329 drbd_req_complete(req
, m
);
331 /* local completion may still come in later,
332 * we need to keep the req object around. */
333 if (req
->rq_state
& RQ_LOCAL_ABORTED
)
336 if (req
->rq_state
& RQ_POSTPONED
) {
337 /* don't destroy the req object just yet,
338 * but queue it for retry */
339 drbd_restart_request(req
);
343 kref_put(&req
->kref
, drbd_req_destroy
);
346 static void set_if_null_req_next(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
348 struct drbd_connection
*connection
= peer_device
? peer_device
->connection
: NULL
;
351 if (connection
->req_next
== NULL
)
352 connection
->req_next
= req
;
355 static void advance_conn_req_next(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
357 struct drbd_connection
*connection
= peer_device
? peer_device
->connection
: NULL
;
360 if (connection
->req_next
!= req
)
362 list_for_each_entry_continue(req
, &connection
->transfer_log
, tl_requests
) {
363 const unsigned s
= req
->rq_state
;
364 if (s
& RQ_NET_QUEUED
)
367 if (&req
->tl_requests
== &connection
->transfer_log
)
369 connection
->req_next
= req
;
372 static void set_if_null_req_ack_pending(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
374 struct drbd_connection
*connection
= peer_device
? peer_device
->connection
: NULL
;
377 if (connection
->req_ack_pending
== NULL
)
378 connection
->req_ack_pending
= req
;
381 static void advance_conn_req_ack_pending(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
383 struct drbd_connection
*connection
= peer_device
? peer_device
->connection
: NULL
;
386 if (connection
->req_ack_pending
!= req
)
388 list_for_each_entry_continue(req
, &connection
->transfer_log
, tl_requests
) {
389 const unsigned s
= req
->rq_state
;
390 if ((s
& RQ_NET_SENT
) && (s
& RQ_NET_PENDING
))
393 if (&req
->tl_requests
== &connection
->transfer_log
)
395 connection
->req_ack_pending
= req
;
398 static void set_if_null_req_not_net_done(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
400 struct drbd_connection
*connection
= peer_device
? peer_device
->connection
: NULL
;
403 if (connection
->req_not_net_done
== NULL
)
404 connection
->req_not_net_done
= req
;
407 static void advance_conn_req_not_net_done(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
409 struct drbd_connection
*connection
= peer_device
? peer_device
->connection
: NULL
;
412 if (connection
->req_not_net_done
!= req
)
414 list_for_each_entry_continue(req
, &connection
->transfer_log
, tl_requests
) {
415 const unsigned s
= req
->rq_state
;
416 if ((s
& RQ_NET_SENT
) && !(s
& RQ_NET_DONE
))
419 if (&req
->tl_requests
== &connection
->transfer_log
)
421 connection
->req_not_net_done
= req
;
424 /* I'd like this to be the only place that manipulates
425 * req->completion_ref and req->kref. */
426 static void mod_rq_state(struct drbd_request
*req
, struct bio_and_error
*m
,
429 struct drbd_device
*device
= req
->device
;
430 struct drbd_peer_device
*peer_device
= first_peer_device(device
);
431 unsigned s
= req
->rq_state
;
434 if (drbd_suspended(device
) && !((s
| clear
) & RQ_COMPLETION_SUSP
))
435 set
|= RQ_COMPLETION_SUSP
;
439 req
->rq_state
&= ~clear
;
440 req
->rq_state
|= set
;
443 if (req
->rq_state
== s
)
446 /* intent: get references */
448 kref_get(&req
->kref
);
450 if (!(s
& RQ_LOCAL_PENDING
) && (set
& RQ_LOCAL_PENDING
))
451 atomic_inc(&req
->completion_ref
);
453 if (!(s
& RQ_NET_PENDING
) && (set
& RQ_NET_PENDING
)) {
454 inc_ap_pending(device
);
455 atomic_inc(&req
->completion_ref
);
458 if (!(s
& RQ_NET_QUEUED
) && (set
& RQ_NET_QUEUED
)) {
459 atomic_inc(&req
->completion_ref
);
460 set_if_null_req_next(peer_device
, req
);
463 if (!(s
& RQ_EXP_BARR_ACK
) && (set
& RQ_EXP_BARR_ACK
))
464 kref_get(&req
->kref
); /* wait for the DONE */
466 if (!(s
& RQ_NET_SENT
) && (set
& RQ_NET_SENT
)) {
467 /* potentially already completed in the ack_receiver thread */
468 if (!(s
& RQ_NET_DONE
)) {
469 atomic_add(req
->i
.size
>> 9, &device
->ap_in_flight
);
470 set_if_null_req_not_net_done(peer_device
, req
);
472 if (req
->rq_state
& RQ_NET_PENDING
)
473 set_if_null_req_ack_pending(peer_device
, req
);
476 if (!(s
& RQ_COMPLETION_SUSP
) && (set
& RQ_COMPLETION_SUSP
))
477 atomic_inc(&req
->completion_ref
);
479 /* progress: put references */
481 if ((s
& RQ_COMPLETION_SUSP
) && (clear
& RQ_COMPLETION_SUSP
))
484 if (!(s
& RQ_LOCAL_ABORTED
) && (set
& RQ_LOCAL_ABORTED
)) {
485 D_ASSERT(device
, req
->rq_state
& RQ_LOCAL_PENDING
);
489 if ((s
& RQ_LOCAL_PENDING
) && (clear
& RQ_LOCAL_PENDING
)) {
490 if (req
->rq_state
& RQ_LOCAL_ABORTED
)
491 kref_put(&req
->kref
, drbd_req_destroy
);
494 list_del_init(&req
->req_pending_local
);
497 if ((s
& RQ_NET_PENDING
) && (clear
& RQ_NET_PENDING
)) {
498 dec_ap_pending(device
);
500 req
->acked_jif
= jiffies
;
501 advance_conn_req_ack_pending(peer_device
, req
);
504 if ((s
& RQ_NET_QUEUED
) && (clear
& RQ_NET_QUEUED
)) {
506 advance_conn_req_next(peer_device
, req
);
509 if (!(s
& RQ_NET_DONE
) && (set
& RQ_NET_DONE
)) {
511 atomic_sub(req
->i
.size
>> 9, &device
->ap_in_flight
);
512 if (s
& RQ_EXP_BARR_ACK
)
513 kref_put(&req
->kref
, drbd_req_destroy
);
514 req
->net_done_jif
= jiffies
;
516 /* in ahead/behind mode, or just in case,
517 * before we finally destroy this request,
518 * the caching pointers must not reference it anymore */
519 advance_conn_req_next(peer_device
, req
);
520 advance_conn_req_ack_pending(peer_device
, req
);
521 advance_conn_req_not_net_done(peer_device
, req
);
524 /* potentially complete and destroy */
526 /* If we made progress, retry conflicting peer requests, if any. */
528 wake_up(&device
->misc_wait
);
530 drbd_req_put_completion_ref(req
, m
, c_put
);
531 kref_put(&req
->kref
, drbd_req_destroy
);
534 static void drbd_report_io_error(struct drbd_device
*device
, struct drbd_request
*req
)
536 char b
[BDEVNAME_SIZE
];
538 if (!__ratelimit(&drbd_ratelimit_state
))
541 drbd_warn(device
, "local %s IO error sector %llu+%u on %s\n",
542 (req
->rq_state
& RQ_WRITE
) ? "WRITE" : "READ",
543 (unsigned long long)req
->i
.sector
,
545 bdevname(device
->ldev
->backing_bdev
, b
));
548 /* Helper for HANDED_OVER_TO_NETWORK.
549 * Is this a protocol A write (neither WRITE_ACK nor RECEIVE_ACK expected)?
550 * Is it also still "PENDING"?
551 * --> If so, clear PENDING and set NET_OK below.
552 * If it is a protocol A write, but not RQ_PENDING anymore, neg-ack was faster
553 * (and we must not set RQ_NET_OK) */
554 static inline bool is_pending_write_protocol_A(struct drbd_request
*req
)
556 return (req
->rq_state
&
557 (RQ_WRITE
|RQ_NET_PENDING
|RQ_EXP_WRITE_ACK
|RQ_EXP_RECEIVE_ACK
))
558 == (RQ_WRITE
|RQ_NET_PENDING
);
561 /* obviously this could be coded as many single functions
562 * instead of one huge switch,
563 * or by putting the code directly in the respective locations
564 * (as it has been before).
566 * but having it this way
567 * enforces that it is all in this one place, where it is easier to audit,
568 * it makes it obvious that whatever "event" "happens" to a request should
569 * happen "atomically" within the req_lock,
570 * and it enforces that we have to think in a very structured manner
571 * about the "events" that may happen to a request during its life time ...
573 int __req_mod(struct drbd_request
*req
, enum drbd_req_event what
,
574 struct bio_and_error
*m
)
576 struct drbd_device
*const device
= req
->device
;
577 struct drbd_peer_device
*const peer_device
= first_peer_device(device
);
578 struct drbd_connection
*const connection
= peer_device
? peer_device
->connection
: NULL
;
587 drbd_err(device
, "LOGIC BUG in %s:%u\n", __FILE__
, __LINE__
);
590 /* does not happen...
591 * initialization done in drbd_req_new
596 case TO_BE_SENT
: /* via network */
597 /* reached via __drbd_make_request
598 * and from w_read_retry_remote */
599 D_ASSERT(device
, !(req
->rq_state
& RQ_NET_MASK
));
601 nc
= rcu_dereference(connection
->net_conf
);
602 p
= nc
->wire_protocol
;
605 p
== DRBD_PROT_C
? RQ_EXP_WRITE_ACK
:
606 p
== DRBD_PROT_B
? RQ_EXP_RECEIVE_ACK
: 0;
607 mod_rq_state(req
, m
, 0, RQ_NET_PENDING
);
610 case TO_BE_SUBMITTED
: /* locally */
611 /* reached via __drbd_make_request */
612 D_ASSERT(device
, !(req
->rq_state
& RQ_LOCAL_MASK
));
613 mod_rq_state(req
, m
, 0, RQ_LOCAL_PENDING
);
617 if (req
->rq_state
& RQ_WRITE
)
618 device
->writ_cnt
+= req
->i
.size
>> 9;
620 device
->read_cnt
+= req
->i
.size
>> 9;
622 mod_rq_state(req
, m
, RQ_LOCAL_PENDING
,
623 RQ_LOCAL_COMPLETED
|RQ_LOCAL_OK
);
627 mod_rq_state(req
, m
, 0, RQ_LOCAL_ABORTED
);
630 case WRITE_COMPLETED_WITH_ERROR
:
631 drbd_report_io_error(device
, req
);
632 __drbd_chk_io_error(device
, DRBD_WRITE_ERROR
);
633 mod_rq_state(req
, m
, RQ_LOCAL_PENDING
, RQ_LOCAL_COMPLETED
);
636 case READ_COMPLETED_WITH_ERROR
:
637 drbd_set_out_of_sync(device
, req
->i
.sector
, req
->i
.size
);
638 drbd_report_io_error(device
, req
);
639 __drbd_chk_io_error(device
, DRBD_READ_ERROR
);
641 case READ_AHEAD_COMPLETED_WITH_ERROR
:
642 /* it is legal to fail read-ahead, no __drbd_chk_io_error in that case. */
643 mod_rq_state(req
, m
, RQ_LOCAL_PENDING
, RQ_LOCAL_COMPLETED
);
646 case DISCARD_COMPLETED_NOTSUPP
:
647 case DISCARD_COMPLETED_WITH_ERROR
:
648 /* I'd rather not detach from local disk just because it
649 * failed a REQ_DISCARD. */
650 mod_rq_state(req
, m
, RQ_LOCAL_PENDING
, RQ_LOCAL_COMPLETED
);
653 case QUEUE_FOR_NET_READ
:
656 * or target area marked as invalid,
657 * or just got an io-error. */
658 /* from __drbd_make_request
659 * or from bio_endio during read io-error recovery */
661 /* So we can verify the handle in the answer packet.
662 * Corresponding drbd_remove_request_interval is in
663 * drbd_req_complete() */
664 D_ASSERT(device
, drbd_interval_empty(&req
->i
));
665 drbd_insert_interval(&device
->read_requests
, &req
->i
);
667 set_bit(UNPLUG_REMOTE
, &device
->flags
);
669 D_ASSERT(device
, req
->rq_state
& RQ_NET_PENDING
);
670 D_ASSERT(device
, (req
->rq_state
& RQ_LOCAL_MASK
) == 0);
671 mod_rq_state(req
, m
, 0, RQ_NET_QUEUED
);
672 req
->w
.cb
= w_send_read_req
;
673 drbd_queue_work(&connection
->sender_work
,
677 case QUEUE_FOR_NET_WRITE
:
678 /* assert something? */
679 /* from __drbd_make_request only */
681 /* Corresponding drbd_remove_request_interval is in
682 * drbd_req_complete() */
683 D_ASSERT(device
, drbd_interval_empty(&req
->i
));
684 drbd_insert_interval(&device
->write_requests
, &req
->i
);
687 * In case the req ended up on the transfer log before being
688 * queued on the worker, it could lead to this request being
689 * missed during cleanup after connection loss.
690 * So we have to do both operations here,
691 * within the same lock that protects the transfer log.
693 * _req_add_to_epoch(req); this has to be after the
694 * _maybe_start_new_epoch(req); which happened in
695 * __drbd_make_request, because we now may set the bit
696 * again ourselves to close the current epoch.
698 * Add req to the (now) current epoch (barrier). */
700 /* otherwise we may lose an unplug, which may cause some remote
701 * io-scheduler timeout to expire, increasing maximum latency,
702 * hurting performance. */
703 set_bit(UNPLUG_REMOTE
, &device
->flags
);
705 /* queue work item to send data */
706 D_ASSERT(device
, req
->rq_state
& RQ_NET_PENDING
);
707 mod_rq_state(req
, m
, 0, RQ_NET_QUEUED
|RQ_EXP_BARR_ACK
);
708 req
->w
.cb
= w_send_dblock
;
709 drbd_queue_work(&connection
->sender_work
,
712 /* close the epoch, in case it outgrew the limit */
714 nc
= rcu_dereference(connection
->net_conf
);
715 p
= nc
->max_epoch_size
;
717 if (connection
->current_tle_writes
>= p
)
718 start_new_tl_epoch(connection
);
722 case QUEUE_FOR_SEND_OOS
:
723 mod_rq_state(req
, m
, 0, RQ_NET_QUEUED
);
724 req
->w
.cb
= w_send_out_of_sync
;
725 drbd_queue_work(&connection
->sender_work
,
729 case READ_RETRY_REMOTE_CANCELED
:
732 /* real cleanup will be done from tl_clear. just update flags
733 * so it is no longer marked as on the worker queue */
734 mod_rq_state(req
, m
, RQ_NET_QUEUED
, 0);
737 case HANDED_OVER_TO_NETWORK
:
738 /* assert something? */
739 if (is_pending_write_protocol_A(req
))
740 /* this is what is dangerous about protocol A:
741 * pretend it was successfully written on the peer. */
742 mod_rq_state(req
, m
, RQ_NET_QUEUED
|RQ_NET_PENDING
,
743 RQ_NET_SENT
|RQ_NET_OK
);
745 mod_rq_state(req
, m
, RQ_NET_QUEUED
, RQ_NET_SENT
);
746 /* It is still not yet RQ_NET_DONE until the
747 * corresponding epoch barrier got acked as well,
748 * so we know what to dirty on connection loss. */
751 case OOS_HANDED_TO_NETWORK
:
752 /* Was not set PENDING, no longer QUEUED, so is now DONE
753 * as far as this connection is concerned. */
754 mod_rq_state(req
, m
, RQ_NET_QUEUED
, RQ_NET_DONE
);
757 case CONNECTION_LOST_WHILE_PENDING
:
758 /* transfer log cleanup after connection loss */
760 RQ_NET_OK
|RQ_NET_PENDING
|RQ_COMPLETION_SUSP
,
764 case CONFLICT_RESOLVED
:
765 /* for superseded conflicting writes of multiple primaries,
766 * there is no need to keep anything in the tl, potential
767 * node crashes are covered by the activity log.
769 * If this request had been marked as RQ_POSTPONED before,
770 * it will actually not be completed, but "restarted",
771 * resubmitted from the retry worker context. */
772 D_ASSERT(device
, req
->rq_state
& RQ_NET_PENDING
);
773 D_ASSERT(device
, req
->rq_state
& RQ_EXP_WRITE_ACK
);
774 mod_rq_state(req
, m
, RQ_NET_PENDING
, RQ_NET_DONE
|RQ_NET_OK
);
777 case WRITE_ACKED_BY_PEER_AND_SIS
:
778 req
->rq_state
|= RQ_NET_SIS
;
779 case WRITE_ACKED_BY_PEER
:
780 /* Normal operation protocol C: successfully written on peer.
781 * During resync, even in protocol != C,
782 * we requested an explicit write ack anyways.
783 * Which means we cannot even assert anything here.
784 * Nothing more to do here.
785 * We want to keep the tl in place for all protocols, to cater
786 * for volatile write-back caches on lower level devices. */
788 case RECV_ACKED_BY_PEER
:
789 D_ASSERT(device
, req
->rq_state
& RQ_EXP_RECEIVE_ACK
);
790 /* protocol B; pretends to be successfully written on peer.
791 * see also notes above in HANDED_OVER_TO_NETWORK about
794 mod_rq_state(req
, m
, RQ_NET_PENDING
, RQ_NET_OK
);
798 D_ASSERT(device
, req
->rq_state
& RQ_EXP_WRITE_ACK
);
799 /* If this node has already detected the write conflict, the
800 * worker will be waiting on misc_wait. Wake it up once this
801 * request has completed locally.
803 D_ASSERT(device
, req
->rq_state
& RQ_NET_PENDING
);
804 req
->rq_state
|= RQ_POSTPONED
;
806 wake_up(&device
->misc_wait
);
807 /* Do not clear RQ_NET_PENDING. This request will make further
808 * progress via restart_conflicting_writes() or
809 * fail_postponed_requests(). Hopefully. */
813 mod_rq_state(req
, m
, RQ_NET_OK
|RQ_NET_PENDING
, 0);
816 case FAIL_FROZEN_DISK_IO
:
817 if (!(req
->rq_state
& RQ_LOCAL_COMPLETED
))
819 mod_rq_state(req
, m
, RQ_COMPLETION_SUSP
, 0);
822 case RESTART_FROZEN_DISK_IO
:
823 if (!(req
->rq_state
& RQ_LOCAL_COMPLETED
))
827 RQ_COMPLETION_SUSP
|RQ_LOCAL_COMPLETED
,
831 if (bio_data_dir(req
->master_bio
) == WRITE
)
834 get_ldev(device
); /* always succeeds in this call path */
835 req
->w
.cb
= w_restart_disk_io
;
836 drbd_queue_work(&connection
->sender_work
,
841 /* Simply complete (local only) READs. */
842 if (!(req
->rq_state
& RQ_WRITE
) && !req
->w
.cb
) {
843 mod_rq_state(req
, m
, RQ_COMPLETION_SUSP
, 0);
847 /* If RQ_NET_OK is already set, we got a P_WRITE_ACK or P_RECV_ACK
848 before the connection loss (B&C only); only P_BARRIER_ACK
849 (or the local completion?) was missing when we suspended.
850 Throwing them out of the TL here by pretending we got a BARRIER_ACK.
851 During connection handshake, we ensure that the peer was not rebooted. */
852 if (!(req
->rq_state
& RQ_NET_OK
)) {
853 /* FIXME could this possibly be a req->dw.cb == w_send_out_of_sync?
854 * in that case we must not set RQ_NET_PENDING. */
856 mod_rq_state(req
, m
, RQ_COMPLETION_SUSP
, RQ_NET_QUEUED
|RQ_NET_PENDING
);
858 /* w.cb expected to be w_send_dblock, or w_send_read_req */
859 drbd_queue_work(&connection
->sender_work
,
861 rv
= req
->rq_state
& RQ_WRITE
? MR_WRITE
: MR_READ
;
862 } /* else: FIXME can this happen? */
865 /* else, fall through to BARRIER_ACKED */
868 /* barrier ack for READ requests does not make sense */
869 if (!(req
->rq_state
& RQ_WRITE
))
872 if (req
->rq_state
& RQ_NET_PENDING
) {
873 /* barrier came in before all requests were acked.
874 * this is bad, because if the connection is lost now,
875 * we won't be able to clean them up... */
876 drbd_err(device
, "FIXME (BARRIER_ACKED but pending)\n");
878 /* Allowed to complete requests, even while suspended.
879 * As this is called for all requests within a matching epoch,
880 * we need to filter, and only set RQ_NET_DONE for those that
881 * have actually been on the wire. */
882 mod_rq_state(req
, m
, RQ_COMPLETION_SUSP
,
883 (req
->rq_state
& RQ_NET_MASK
) ? RQ_NET_DONE
: 0);
887 D_ASSERT(device
, req
->rq_state
& RQ_NET_PENDING
);
888 mod_rq_state(req
, m
, RQ_NET_PENDING
, RQ_NET_OK
|RQ_NET_DONE
);
891 case QUEUE_AS_DRBD_BARRIER
:
892 start_new_tl_epoch(connection
);
893 mod_rq_state(req
, m
, 0, RQ_NET_OK
|RQ_NET_DONE
);
900 /* we may do a local read if:
901 * - we are consistent (of course),
902 * - or we are generally inconsistent,
903 * BUT we are still/already IN SYNC for this area.
904 * since size may be bigger than BM_BLOCK_SIZE,
905 * we may need to check several bits.
907 static bool drbd_may_do_local_read(struct drbd_device
*device
, sector_t sector
, int size
)
909 unsigned long sbnr
, ebnr
;
910 sector_t esector
, nr_sectors
;
912 if (device
->state
.disk
== D_UP_TO_DATE
)
914 if (device
->state
.disk
!= D_INCONSISTENT
)
916 esector
= sector
+ (size
>> 9) - 1;
917 nr_sectors
= drbd_get_capacity(device
->this_bdev
);
918 D_ASSERT(device
, sector
< nr_sectors
);
919 D_ASSERT(device
, esector
< nr_sectors
);
921 sbnr
= BM_SECT_TO_BIT(sector
);
922 ebnr
= BM_SECT_TO_BIT(esector
);
924 return drbd_bm_count_bits(device
, sbnr
, ebnr
) == 0;
927 static bool remote_due_to_read_balancing(struct drbd_device
*device
, sector_t sector
,
928 enum drbd_read_balancing rbm
)
930 struct backing_dev_info
*bdi
;
934 case RB_CONGESTED_REMOTE
:
935 bdi
= device
->ldev
->backing_bdev
->bd_disk
->queue
->backing_dev_info
;
936 return bdi_read_congested(bdi
);
937 case RB_LEAST_PENDING
:
938 return atomic_read(&device
->local_cnt
) >
939 atomic_read(&device
->ap_pending_cnt
) + atomic_read(&device
->rs_pending_cnt
);
940 case RB_32K_STRIPING
: /* stripe_shift = 15 */
941 case RB_64K_STRIPING
:
942 case RB_128K_STRIPING
:
943 case RB_256K_STRIPING
:
944 case RB_512K_STRIPING
:
945 case RB_1M_STRIPING
: /* stripe_shift = 20 */
946 stripe_shift
= (rbm
- RB_32K_STRIPING
+ 15);
947 return (sector
>> (stripe_shift
- 9)) & 1;
949 return test_and_change_bit(READ_BALANCE_RR
, &device
->flags
);
950 case RB_PREFER_REMOTE
:
952 case RB_PREFER_LOCAL
:
959 * complete_conflicting_writes - wait for any conflicting write requests
961 * The write_requests tree contains all active write requests which we
962 * currently know about. Wait for any requests to complete which conflict with
965 * Only way out: remove the conflicting intervals from the tree.
967 static void complete_conflicting_writes(struct drbd_request
*req
)
970 struct drbd_device
*device
= req
->device
;
971 struct drbd_interval
*i
;
972 sector_t sector
= req
->i
.sector
;
973 int size
= req
->i
.size
;
976 drbd_for_each_overlap(i
, &device
->write_requests
, sector
, size
) {
977 /* Ignore, if already completed to upper layers. */
980 /* Handle the first found overlap. After the schedule
981 * we have to restart the tree walk. */
987 /* Indicate to wake up device->misc_wait on progress. */
988 prepare_to_wait(&device
->misc_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
990 spin_unlock_irq(&device
->resource
->req_lock
);
992 spin_lock_irq(&device
->resource
->req_lock
);
994 finish_wait(&device
->misc_wait
, &wait
);
997 /* called within req_lock */
998 static void maybe_pull_ahead(struct drbd_device
*device
)
1000 struct drbd_connection
*connection
= first_peer_device(device
)->connection
;
1001 struct net_conf
*nc
;
1002 bool congested
= false;
1003 enum drbd_on_congestion on_congestion
;
1006 nc
= rcu_dereference(connection
->net_conf
);
1007 on_congestion
= nc
? nc
->on_congestion
: OC_BLOCK
;
1009 if (on_congestion
== OC_BLOCK
||
1010 connection
->agreed_pro_version
< 96)
1013 if (on_congestion
== OC_PULL_AHEAD
&& device
->state
.conn
== C_AHEAD
)
1014 return; /* nothing to do ... */
1016 /* If I don't even have good local storage, we can not reasonably try
1017 * to pull ahead of the peer. We also need the local reference to make
1018 * sure device->act_log is there.
1020 if (!get_ldev_if_state(device
, D_UP_TO_DATE
))
1023 if (nc
->cong_fill
&&
1024 atomic_read(&device
->ap_in_flight
) >= nc
->cong_fill
) {
1025 drbd_info(device
, "Congestion-fill threshold reached\n");
1029 if (device
->act_log
->used
>= nc
->cong_extents
) {
1030 drbd_info(device
, "Congestion-extents threshold reached\n");
1035 /* start a new epoch for non-mirrored writes */
1036 start_new_tl_epoch(first_peer_device(device
)->connection
);
1038 if (on_congestion
== OC_PULL_AHEAD
)
1039 _drbd_set_state(_NS(device
, conn
, C_AHEAD
), 0, NULL
);
1040 else /*nc->on_congestion == OC_DISCONNECT */
1041 _drbd_set_state(_NS(device
, conn
, C_DISCONNECTING
), 0, NULL
);
1046 /* If this returns false, and req->private_bio is still set,
1047 * this should be submitted locally.
1049 * If it returns false, but req->private_bio is not set,
1050 * we do not have access to good data :(
1052 * Otherwise, this destroys req->private_bio, if any,
1055 static bool do_remote_read(struct drbd_request
*req
)
1057 struct drbd_device
*device
= req
->device
;
1058 enum drbd_read_balancing rbm
;
1060 if (req
->private_bio
) {
1061 if (!drbd_may_do_local_read(device
,
1062 req
->i
.sector
, req
->i
.size
)) {
1063 bio_put(req
->private_bio
);
1064 req
->private_bio
= NULL
;
1069 if (device
->state
.pdsk
!= D_UP_TO_DATE
)
1072 if (req
->private_bio
== NULL
)
1075 /* TODO: improve read balancing decisions, take into account drbd
1076 * protocol, pending requests etc. */
1079 rbm
= rcu_dereference(device
->ldev
->disk_conf
)->read_balancing
;
1082 if (rbm
== RB_PREFER_LOCAL
&& req
->private_bio
)
1083 return false; /* submit locally */
1085 if (remote_due_to_read_balancing(device
, req
->i
.sector
, rbm
)) {
1086 if (req
->private_bio
) {
1087 bio_put(req
->private_bio
);
1088 req
->private_bio
= NULL
;
1097 bool drbd_should_do_remote(union drbd_dev_state s
)
1099 return s
.pdsk
== D_UP_TO_DATE
||
1100 (s
.pdsk
>= D_INCONSISTENT
&&
1101 s
.conn
>= C_WF_BITMAP_T
&&
1103 /* Before proto 96 that was >= CONNECTED instead of >= C_WF_BITMAP_T.
1104 That is equivalent since before 96 IO was frozen in the C_WF_BITMAP*
1108 static bool drbd_should_send_out_of_sync(union drbd_dev_state s
)
1110 return s
.conn
== C_AHEAD
|| s
.conn
== C_WF_BITMAP_S
;
1111 /* pdsk = D_INCONSISTENT as a consequence. Protocol 96 check not necessary
1112 since we enter state C_AHEAD only if proto >= 96 */
1115 /* returns number of connections (== 1, for drbd 8.4)
1116 * expected to actually write this data,
1117 * which does NOT include those that we are L_AHEAD for. */
1118 static int drbd_process_write_request(struct drbd_request
*req
)
1120 struct drbd_device
*device
= req
->device
;
1121 int remote
, send_oos
;
1123 remote
= drbd_should_do_remote(device
->state
);
1124 send_oos
= drbd_should_send_out_of_sync(device
->state
);
1126 /* Need to replicate writes. Unless it is an empty flush,
1127 * which is better mapped to a DRBD P_BARRIER packet,
1128 * also for drbd wire protocol compatibility reasons.
1129 * If this was a flush, just start a new epoch.
1130 * Unless the current epoch was empty anyways, or we are not currently
1131 * replicating, in which case there is no point. */
1132 if (unlikely(req
->i
.size
== 0)) {
1133 /* The only size==0 bios we expect are empty flushes. */
1134 D_ASSERT(device
, req
->master_bio
->bi_opf
& REQ_PREFLUSH
);
1136 _req_mod(req
, QUEUE_AS_DRBD_BARRIER
);
1140 if (!remote
&& !send_oos
)
1143 D_ASSERT(device
, !(remote
&& send_oos
));
1146 _req_mod(req
, TO_BE_SENT
);
1147 _req_mod(req
, QUEUE_FOR_NET_WRITE
);
1148 } else if (drbd_set_out_of_sync(device
, req
->i
.sector
, req
->i
.size
))
1149 _req_mod(req
, QUEUE_FOR_SEND_OOS
);
1154 static void drbd_process_discard_req(struct drbd_request
*req
)
1156 struct block_device
*bdev
= req
->device
->ldev
->backing_bdev
;
1158 if (blkdev_issue_zeroout(bdev
, req
->i
.sector
, req
->i
.size
>> 9,
1160 req
->private_bio
->bi_status
= BLK_STS_IOERR
;
1161 bio_endio(req
->private_bio
);
1165 drbd_submit_req_private_bio(struct drbd_request
*req
)
1167 struct drbd_device
*device
= req
->device
;
1168 struct bio
*bio
= req
->private_bio
;
1171 if (bio_op(bio
) != REQ_OP_READ
)
1172 type
= DRBD_FAULT_DT_WR
;
1173 else if (bio
->bi_opf
& REQ_RAHEAD
)
1174 type
= DRBD_FAULT_DT_RA
;
1176 type
= DRBD_FAULT_DT_RD
;
1178 bio
->bi_bdev
= device
->ldev
->backing_bdev
;
1180 /* State may have changed since we grabbed our reference on the
1181 * ->ldev member. Double check, and short-circuit to endio.
1182 * In case the last activity log transaction failed to get on
1183 * stable storage, and this is a WRITE, we may not even submit
1185 if (get_ldev(device
)) {
1186 if (drbd_insert_fault(device
, type
))
1188 else if (bio_op(bio
) == REQ_OP_WRITE_ZEROES
||
1189 bio_op(bio
) == REQ_OP_DISCARD
)
1190 drbd_process_discard_req(req
);
1192 generic_make_request(bio
);
1198 static void drbd_queue_write(struct drbd_device
*device
, struct drbd_request
*req
)
1200 spin_lock_irq(&device
->resource
->req_lock
);
1201 list_add_tail(&req
->tl_requests
, &device
->submit
.writes
);
1202 list_add_tail(&req
->req_pending_master_completion
,
1203 &device
->pending_master_completion
[1 /* WRITE */]);
1204 spin_unlock_irq(&device
->resource
->req_lock
);
1205 queue_work(device
->submit
.wq
, &device
->submit
.worker
);
1206 /* do_submit() may sleep internally on al_wait, too */
1207 wake_up(&device
->al_wait
);
1210 /* returns the new drbd_request pointer, if the caller is expected to
1211 * drbd_send_and_submit() it (to save latency), or NULL if we queued the
1212 * request on the submitter thread.
1213 * Returns ERR_PTR(-ENOMEM) if we cannot allocate a drbd_request.
1215 static struct drbd_request
*
1216 drbd_request_prepare(struct drbd_device
*device
, struct bio
*bio
, unsigned long start_jif
)
1218 const int rw
= bio_data_dir(bio
);
1219 struct drbd_request
*req
;
1221 /* allocate outside of all locks; */
1222 req
= drbd_req_new(device
, bio
);
1225 /* only pass the error to the upper layers.
1226 * if user cannot handle io errors, that's not our business. */
1227 drbd_err(device
, "could not kmalloc() req\n");
1228 bio
->bi_status
= BLK_STS_RESOURCE
;
1230 return ERR_PTR(-ENOMEM
);
1232 req
->start_jif
= start_jif
;
1234 if (!get_ldev(device
)) {
1235 bio_put(req
->private_bio
);
1236 req
->private_bio
= NULL
;
1239 /* Update disk stats */
1240 _drbd_start_io_acct(device
, req
);
1242 /* process discards always from our submitter thread */
1243 if ((bio_op(bio
) & REQ_OP_WRITE_ZEROES
) ||
1244 (bio_op(bio
) & REQ_OP_DISCARD
))
1245 goto queue_for_submitter_thread
;
1247 if (rw
== WRITE
&& req
->private_bio
&& req
->i
.size
1248 && !test_bit(AL_SUSPENDED
, &device
->flags
)) {
1249 if (!drbd_al_begin_io_fastpath(device
, &req
->i
))
1250 goto queue_for_submitter_thread
;
1251 req
->rq_state
|= RQ_IN_ACT_LOG
;
1252 req
->in_actlog_jif
= jiffies
;
1256 queue_for_submitter_thread
:
1257 atomic_inc(&device
->ap_actlog_cnt
);
1258 drbd_queue_write(device
, req
);
1262 /* Require at least one path to current data.
1263 * We don't want to allow writes on C_STANDALONE D_INCONSISTENT:
1264 * We would not allow to read what was written,
1265 * we would not have bumped the data generation uuids,
1266 * we would cause data divergence for all the wrong reasons.
1268 * If we don't see at least one D_UP_TO_DATE, we will fail this request,
1269 * which either returns EIO, or, if OND_SUSPEND_IO is set, suspends IO,
1270 * and queues for retry later.
1272 static bool may_do_writes(struct drbd_device
*device
)
1274 const union drbd_dev_state s
= device
->state
;
1275 return s
.disk
== D_UP_TO_DATE
|| s
.pdsk
== D_UP_TO_DATE
;
1278 static void drbd_send_and_submit(struct drbd_device
*device
, struct drbd_request
*req
)
1280 struct drbd_resource
*resource
= device
->resource
;
1281 const int rw
= bio_data_dir(req
->master_bio
);
1282 struct bio_and_error m
= { NULL
, };
1283 bool no_remote
= false;
1284 bool submit_private_bio
= false;
1286 spin_lock_irq(&resource
->req_lock
);
1288 /* This may temporarily give up the req_lock,
1289 * but will re-aquire it before it returns here.
1290 * Needs to be before the check on drbd_suspended() */
1291 complete_conflicting_writes(req
);
1292 /* no more giving up req_lock from now on! */
1294 /* check for congestion, and potentially stop sending
1295 * full data updates, but start sending "dirty bits" only. */
1296 maybe_pull_ahead(device
);
1300 if (drbd_suspended(device
)) {
1301 /* push back and retry: */
1302 req
->rq_state
|= RQ_POSTPONED
;
1303 if (req
->private_bio
) {
1304 bio_put(req
->private_bio
);
1305 req
->private_bio
= NULL
;
1311 /* We fail READ early, if we can not serve it.
1312 * We must do this before req is registered on any lists.
1313 * Otherwise, drbd_req_complete() will queue failed READ for retry. */
1315 if (!do_remote_read(req
) && !req
->private_bio
)
1319 /* which transfer log epoch does this belong to? */
1320 req
->epoch
= atomic_read(&first_peer_device(device
)->connection
->current_tle_nr
);
1322 /* no point in adding empty flushes to the transfer log,
1323 * they are mapped to drbd barriers already. */
1324 if (likely(req
->i
.size
!=0)) {
1326 first_peer_device(device
)->connection
->current_tle_writes
++;
1328 list_add_tail(&req
->tl_requests
, &first_peer_device(device
)->connection
->transfer_log
);
1332 if (req
->private_bio
&& !may_do_writes(device
)) {
1333 bio_put(req
->private_bio
);
1334 req
->private_bio
= NULL
;
1338 if (!drbd_process_write_request(req
))
1341 /* We either have a private_bio, or we can read from remote.
1342 * Otherwise we had done the goto nodata above. */
1343 if (req
->private_bio
== NULL
) {
1344 _req_mod(req
, TO_BE_SENT
);
1345 _req_mod(req
, QUEUE_FOR_NET_READ
);
1350 /* If it took the fast path in drbd_request_prepare, add it here.
1351 * The slow path has added it already. */
1352 if (list_empty(&req
->req_pending_master_completion
))
1353 list_add_tail(&req
->req_pending_master_completion
,
1354 &device
->pending_master_completion
[rw
== WRITE
]);
1355 if (req
->private_bio
) {
1356 /* needs to be marked within the same spinlock */
1357 req
->pre_submit_jif
= jiffies
;
1358 list_add_tail(&req
->req_pending_local
,
1359 &device
->pending_completion
[rw
== WRITE
]);
1360 _req_mod(req
, TO_BE_SUBMITTED
);
1361 /* but we need to give up the spinlock to submit */
1362 submit_private_bio
= true;
1363 } else if (no_remote
) {
1365 if (__ratelimit(&drbd_ratelimit_state
))
1366 drbd_err(device
, "IO ERROR: neither local nor remote data, sector %llu+%u\n",
1367 (unsigned long long)req
->i
.sector
, req
->i
.size
>> 9);
1368 /* A write may have been queued for send_oos, however.
1369 * So we can not simply free it, we must go through drbd_req_put_completion_ref() */
1373 drbd_req_put_completion_ref(req
, &m
, 1);
1374 spin_unlock_irq(&resource
->req_lock
);
1376 /* Even though above is a kref_put(), this is safe.
1377 * As long as we still need to submit our private bio,
1378 * we hold a completion ref, and the request cannot disappear.
1379 * If however this request did not even have a private bio to submit
1380 * (e.g. remote read), req may already be invalid now.
1381 * That's why we cannot check on req->private_bio. */
1382 if (submit_private_bio
)
1383 drbd_submit_req_private_bio(req
);
1385 complete_master_bio(device
, &m
);
1388 void __drbd_make_request(struct drbd_device
*device
, struct bio
*bio
, unsigned long start_jif
)
1390 struct drbd_request
*req
= drbd_request_prepare(device
, bio
, start_jif
);
1391 if (IS_ERR_OR_NULL(req
))
1393 drbd_send_and_submit(device
, req
);
1396 static void submit_fast_path(struct drbd_device
*device
, struct list_head
*incoming
)
1398 struct drbd_request
*req
, *tmp
;
1399 list_for_each_entry_safe(req
, tmp
, incoming
, tl_requests
) {
1400 const int rw
= bio_data_dir(req
->master_bio
);
1402 if (rw
== WRITE
/* rw != WRITE should not even end up here! */
1403 && req
->private_bio
&& req
->i
.size
1404 && !test_bit(AL_SUSPENDED
, &device
->flags
)) {
1405 if (!drbd_al_begin_io_fastpath(device
, &req
->i
))
1408 req
->rq_state
|= RQ_IN_ACT_LOG
;
1409 req
->in_actlog_jif
= jiffies
;
1410 atomic_dec(&device
->ap_actlog_cnt
);
1413 list_del_init(&req
->tl_requests
);
1414 drbd_send_and_submit(device
, req
);
1418 static bool prepare_al_transaction_nonblock(struct drbd_device
*device
,
1419 struct list_head
*incoming
,
1420 struct list_head
*pending
,
1421 struct list_head
*later
)
1423 struct drbd_request
*req
, *tmp
;
1427 spin_lock_irq(&device
->al_lock
);
1428 list_for_each_entry_safe(req
, tmp
, incoming
, tl_requests
) {
1429 err
= drbd_al_begin_io_nonblock(device
, &req
->i
);
1430 if (err
== -ENOBUFS
)
1435 list_move_tail(&req
->tl_requests
, later
);
1437 list_move_tail(&req
->tl_requests
, pending
);
1439 spin_unlock_irq(&device
->al_lock
);
1441 wake_up(&device
->al_wait
);
1442 return !list_empty(pending
);
1445 void send_and_submit_pending(struct drbd_device
*device
, struct list_head
*pending
)
1447 struct drbd_request
*req
, *tmp
;
1449 list_for_each_entry_safe(req
, tmp
, pending
, tl_requests
) {
1450 req
->rq_state
|= RQ_IN_ACT_LOG
;
1451 req
->in_actlog_jif
= jiffies
;
1452 atomic_dec(&device
->ap_actlog_cnt
);
1453 list_del_init(&req
->tl_requests
);
1454 drbd_send_and_submit(device
, req
);
1458 void do_submit(struct work_struct
*ws
)
1460 struct drbd_device
*device
= container_of(ws
, struct drbd_device
, submit
.worker
);
1461 LIST_HEAD(incoming
); /* from drbd_make_request() */
1462 LIST_HEAD(pending
); /* to be submitted after next AL-transaction commit */
1463 LIST_HEAD(busy
); /* blocked by resync requests */
1465 /* grab new incoming requests */
1466 spin_lock_irq(&device
->resource
->req_lock
);
1467 list_splice_tail_init(&device
->submit
.writes
, &incoming
);
1468 spin_unlock_irq(&device
->resource
->req_lock
);
1473 /* move used-to-be-busy back to front of incoming */
1474 list_splice_init(&busy
, &incoming
);
1475 submit_fast_path(device
, &incoming
);
1476 if (list_empty(&incoming
))
1480 prepare_to_wait(&device
->al_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
1482 list_splice_init(&busy
, &incoming
);
1483 prepare_al_transaction_nonblock(device
, &incoming
, &pending
, &busy
);
1484 if (!list_empty(&pending
))
1489 /* If all currently "hot" activity log extents are kept busy by
1490 * incoming requests, we still must not totally starve new
1491 * requests to "cold" extents.
1492 * Something left on &incoming means there had not been
1493 * enough update slots available, and the activity log
1494 * has been marked as "starving".
1496 * Try again now, without looking for new requests,
1497 * effectively blocking all new requests until we made
1498 * at least _some_ progress with what we currently have.
1500 if (!list_empty(&incoming
))
1503 /* Nothing moved to pending, but nothing left
1504 * on incoming: all moved to busy!
1505 * Grab new and iterate. */
1506 spin_lock_irq(&device
->resource
->req_lock
);
1507 list_splice_tail_init(&device
->submit
.writes
, &incoming
);
1508 spin_unlock_irq(&device
->resource
->req_lock
);
1510 finish_wait(&device
->al_wait
, &wait
);
1512 /* If the transaction was full, before all incoming requests
1513 * had been processed, skip ahead to commit, and iterate
1514 * without splicing in more incoming requests from upper layers.
1516 * Else, if all incoming have been processed,
1517 * they have become either "pending" (to be submitted after
1518 * next transaction commit) or "busy" (blocked by resync).
1520 * Maybe more was queued, while we prepared the transaction?
1521 * Try to stuff those into this transaction as well.
1522 * Be strictly non-blocking here,
1523 * we already have something to commit.
1525 * Commit if we don't make any more progres.
1528 while (list_empty(&incoming
)) {
1529 LIST_HEAD(more_pending
);
1530 LIST_HEAD(more_incoming
);
1533 /* It is ok to look outside the lock,
1534 * it's only an optimization anyways */
1535 if (list_empty(&device
->submit
.writes
))
1538 spin_lock_irq(&device
->resource
->req_lock
);
1539 list_splice_tail_init(&device
->submit
.writes
, &more_incoming
);
1540 spin_unlock_irq(&device
->resource
->req_lock
);
1542 if (list_empty(&more_incoming
))
1545 made_progress
= prepare_al_transaction_nonblock(device
, &more_incoming
, &more_pending
, &busy
);
1547 list_splice_tail_init(&more_pending
, &pending
);
1548 list_splice_tail_init(&more_incoming
, &incoming
);
1553 drbd_al_begin_io_commit(device
);
1554 send_and_submit_pending(device
, &pending
);
1558 blk_qc_t
drbd_make_request(struct request_queue
*q
, struct bio
*bio
)
1560 struct drbd_device
*device
= (struct drbd_device
*) q
->queuedata
;
1561 unsigned long start_jif
;
1563 blk_queue_split(q
, &bio
);
1565 start_jif
= jiffies
;
1568 * what we "blindly" assume:
1570 D_ASSERT(device
, IS_ALIGNED(bio
->bi_iter
.bi_size
, 512));
1573 __drbd_make_request(device
, bio
, start_jif
);
1574 return BLK_QC_T_NONE
;
1577 static bool net_timeout_reached(struct drbd_request
*net_req
,
1578 struct drbd_connection
*connection
,
1579 unsigned long now
, unsigned long ent
,
1580 unsigned int ko_count
, unsigned int timeout
)
1582 struct drbd_device
*device
= net_req
->device
;
1584 if (!time_after(now
, net_req
->pre_send_jif
+ ent
))
1587 if (time_in_range(now
, connection
->last_reconnect_jif
, connection
->last_reconnect_jif
+ ent
))
1590 if (net_req
->rq_state
& RQ_NET_PENDING
) {
1591 drbd_warn(device
, "Remote failed to finish a request within %ums > ko-count (%u) * timeout (%u * 0.1s)\n",
1592 jiffies_to_msecs(now
- net_req
->pre_send_jif
), ko_count
, timeout
);
1596 /* We received an ACK already (or are using protocol A),
1597 * but are waiting for the epoch closing barrier ack.
1598 * Check if we sent the barrier already. We should not blame the peer
1599 * for being unresponsive, if we did not even ask it yet. */
1600 if (net_req
->epoch
== connection
->send
.current_epoch_nr
) {
1602 "We did not send a P_BARRIER for %ums > ko-count (%u) * timeout (%u * 0.1s); drbd kernel thread blocked?\n",
1603 jiffies_to_msecs(now
- net_req
->pre_send_jif
), ko_count
, timeout
);
1607 /* Worst case: we may have been blocked for whatever reason, then
1608 * suddenly are able to send a lot of requests (and epoch separating
1609 * barriers) in quick succession.
1610 * The timestamp of the net_req may be much too old and not correspond
1611 * to the sending time of the relevant unack'ed barrier packet, so
1612 * would trigger a spurious timeout. The latest barrier packet may
1613 * have a too recent timestamp to trigger the timeout, potentially miss
1614 * a timeout. Right now we don't have a place to conveniently store
1616 * But in this particular situation, the application requests are still
1617 * completed to upper layers, DRBD should still "feel" responsive.
1618 * No need yet to kill this connection, it may still recover.
1619 * If not, eventually we will have queued enough into the network for
1620 * us to block. From that point of view, the timestamp of the last sent
1621 * barrier packet is relevant enough.
1623 if (time_after(now
, connection
->send
.last_sent_barrier_jif
+ ent
)) {
1624 drbd_warn(device
, "Remote failed to answer a P_BARRIER (sent at %lu jif; now=%lu jif) within %ums > ko-count (%u) * timeout (%u * 0.1s)\n",
1625 connection
->send
.last_sent_barrier_jif
, now
,
1626 jiffies_to_msecs(now
- connection
->send
.last_sent_barrier_jif
), ko_count
, timeout
);
1632 /* A request is considered timed out, if
1633 * - we have some effective timeout from the configuration,
1634 * with some state restrictions applied,
1635 * - the oldest request is waiting for a response from the network
1636 * resp. the local disk,
1637 * - the oldest request is in fact older than the effective timeout,
1638 * - the connection was established (resp. disk was attached)
1639 * for longer than the timeout already.
1640 * Note that for 32bit jiffies and very stable connections/disks,
1641 * we may have a wrap around, which is catched by
1642 * !time_in_range(now, last_..._jif, last_..._jif + timeout).
1644 * Side effect: once per 32bit wrap-around interval, which means every
1645 * ~198 days with 250 HZ, we have a window where the timeout would need
1646 * to expire twice (worst case) to become effective. Good enough.
1649 void request_timer_fn(unsigned long data
)
1651 struct drbd_device
*device
= (struct drbd_device
*) data
;
1652 struct drbd_connection
*connection
= first_peer_device(device
)->connection
;
1653 struct drbd_request
*req_read
, *req_write
, *req_peer
; /* oldest request */
1654 struct net_conf
*nc
;
1655 unsigned long oldest_submit_jif
;
1656 unsigned long ent
= 0, dt
= 0, et
, nt
; /* effective timeout = ko_count * timeout */
1658 unsigned int ko_count
= 0, timeout
= 0;
1661 nc
= rcu_dereference(connection
->net_conf
);
1662 if (nc
&& device
->state
.conn
>= C_WF_REPORT_PARAMS
) {
1663 ko_count
= nc
->ko_count
;
1664 timeout
= nc
->timeout
;
1667 if (get_ldev(device
)) { /* implicit state.disk >= D_INCONSISTENT */
1668 dt
= rcu_dereference(device
->ldev
->disk_conf
)->disk_timeout
* HZ
/ 10;
1674 ent
= timeout
* HZ
/10 * ko_count
;
1675 et
= min_not_zero(dt
, ent
);
1678 return; /* Recurring timer stopped */
1683 spin_lock_irq(&device
->resource
->req_lock
);
1684 req_read
= list_first_entry_or_null(&device
->pending_completion
[0], struct drbd_request
, req_pending_local
);
1685 req_write
= list_first_entry_or_null(&device
->pending_completion
[1], struct drbd_request
, req_pending_local
);
1687 /* maybe the oldest request waiting for the peer is in fact still
1688 * blocking in tcp sendmsg. That's ok, though, that's handled via the
1689 * socket send timeout, requesting a ping, and bumping ko-count in
1690 * we_should_drop_the_connection().
1693 /* check the oldest request we did successfully sent,
1694 * but which is still waiting for an ACK. */
1695 req_peer
= connection
->req_ack_pending
;
1697 /* if we don't have such request (e.g. protocoll A)
1698 * check the oldest requests which is still waiting on its epoch
1699 * closing barrier ack. */
1701 req_peer
= connection
->req_not_net_done
;
1703 /* evaluate the oldest peer request only in one timer! */
1704 if (req_peer
&& req_peer
->device
!= device
)
1707 /* do we have something to evaluate? */
1708 if (req_peer
== NULL
&& req_write
== NULL
&& req_read
== NULL
)
1712 (req_write
&& req_read
)
1713 ? ( time_before(req_write
->pre_submit_jif
, req_read
->pre_submit_jif
)
1714 ? req_write
->pre_submit_jif
: req_read
->pre_submit_jif
)
1715 : req_write
? req_write
->pre_submit_jif
1716 : req_read
? req_read
->pre_submit_jif
: now
;
1718 if (ent
&& req_peer
&& net_timeout_reached(req_peer
, connection
, now
, ent
, ko_count
, timeout
))
1719 _conn_request_state(connection
, NS(conn
, C_TIMEOUT
), CS_VERBOSE
| CS_HARD
);
1721 if (dt
&& oldest_submit_jif
!= now
&&
1722 time_after(now
, oldest_submit_jif
+ dt
) &&
1723 !time_in_range(now
, device
->last_reattach_jif
, device
->last_reattach_jif
+ dt
)) {
1724 drbd_warn(device
, "Local backing device failed to meet the disk-timeout\n");
1725 __drbd_chk_io_error(device
, DRBD_FORCE_DETACH
);
1728 /* Reschedule timer for the nearest not already expired timeout.
1729 * Fallback to now + min(effective network timeout, disk timeout). */
1730 ent
= (ent
&& req_peer
&& time_before(now
, req_peer
->pre_send_jif
+ ent
))
1731 ? req_peer
->pre_send_jif
+ ent
: now
+ et
;
1732 dt
= (dt
&& oldest_submit_jif
!= now
&& time_before(now
, oldest_submit_jif
+ dt
))
1733 ? oldest_submit_jif
+ dt
: now
+ et
;
1734 nt
= time_before(ent
, dt
) ? ent
: dt
;
1736 spin_unlock_irq(&device
->resource
->req_lock
);
1737 mod_timer(&device
->request_timer
, nt
);