?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
drbd: check for corrupt or malicous sector addresses when receiving data
[linux-2.6/libata-dev.git] / drivers / block / drbd / drbd_receiver.c
blob3a36bc814e773f996c30414eea94763061fcd096
1 /*
2 drbd_receiver.c
3
4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
5
6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
9
10 drbd is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
13 any later version.
14
15 drbd is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
19
20 You should have received a copy of the GNU General Public License
21 along with drbd; see the file COPYING. If not, write to
22 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
23 */
24
25
26 #include <linux/module.h>
27
28 #include <asm/uaccess.h>
29 #include <net/sock.h>
30
31 #include <linux/drbd.h>
32 #include <linux/fs.h>
33 #include <linux/file.h>
34 #include <linux/in.h>
35 #include <linux/mm.h>
36 #include <linux/memcontrol.h>
37 #include <linux/mm_inline.h>
38 #include <linux/slab.h>
39 #include <linux/smp_lock.h>
40 #include <linux/pkt_sched.h>
41 #define __KERNEL_SYSCALLS__
42 #include <linux/unistd.h>
43 #include <linux/vmalloc.h>
44 #include <linux/random.h>
45 #include <linux/mm.h>
46 #include <linux/string.h>
47 #include <linux/scatterlist.h>
48 #include "drbd_int.h"
49 #include "drbd_req.h"
50
51 #include "drbd_vli.h"
52
53 struct flush_work {
54 struct drbd_work w;
55 struct drbd_epoch *epoch;
56 };
57
58 enum finish_epoch {
59 FE_STILL_LIVE,
60 FE_DESTROYED,
61 FE_RECYCLED,
62 };
63
64 static int drbd_do_handshake(struct drbd_conf *mdev);
65 static int drbd_do_auth(struct drbd_conf *mdev);
66
67 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event);
68 static int e_end_block(struct drbd_conf *, struct drbd_work *, int);
69
70 static struct drbd_epoch *previous_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch)
71 {
72 struct drbd_epoch *prev;
73 spin_lock(&mdev->epoch_lock);
74 prev = list_entry(epoch->list.prev, struct drbd_epoch, list);
75 if (prev == epoch || prev == mdev->current_epoch)
76 prev = NULL;
77 spin_unlock(&mdev->epoch_lock);
78 return prev;
79 }
80
81 #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN)
82
83 static struct page *drbd_pp_first_page_or_try_alloc(struct drbd_conf *mdev)
84 {
85 struct page *page = NULL;
86
87 /* Yes, testing drbd_pp_vacant outside the lock is racy.
88 * So what. It saves a spin_lock. */
89 if (drbd_pp_vacant > 0) {
90 spin_lock(&drbd_pp_lock);
91 page = drbd_pp_pool;
92 if (page) {
93 drbd_pp_pool = (struct page *)page_private(page);
94 set_page_private(page, 0); /* just to be polite */
95 drbd_pp_vacant--;
96 }
97 spin_unlock(&drbd_pp_lock);
98 }
99 /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD
100 * "criss-cross" setup, that might cause write-out on some other DRBD,
101 * which in turn might block on the other node at this very place. */
102 if (!page)
103 page = alloc_page(GFP_TRY);
104 if (page)
105 atomic_inc(&mdev->pp_in_use);
106 return page;
107 }
108
109 /* kick lower level device, if we have more than (arbitrary number)
110 * reference counts on it, which typically are locally submitted io
111 * requests. don't use unacked_cnt, so we speed up proto A and B, too. */
112 static void maybe_kick_lo(struct drbd_conf *mdev)
113 {
114 if (atomic_read(&mdev->local_cnt) >= mdev->net_conf->unplug_watermark)
115 drbd_kick_lo(mdev);
116 }
117
118 static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed)
119 {
120 struct drbd_epoch_entry *e;
121 struct list_head *le, *tle;
122
123 /* The EEs are always appended to the end of the list. Since
124 they are sent in order over the wire, they have to finish
125 in order. As soon as we see the first not finished we can
126 stop to examine the list... */
127
128 list_for_each_safe(le, tle, &mdev->net_ee) {
129 e = list_entry(le, struct drbd_epoch_entry, w.list);
130 if (drbd_bio_has_active_page(e->private_bio))
131 break;
132 list_move(le, to_be_freed);
133 }
134 }
135
136 static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev)
137 {
138 LIST_HEAD(reclaimed);
139 struct drbd_epoch_entry *e, *t;
140
141 maybe_kick_lo(mdev);
142 spin_lock_irq(&mdev->req_lock);
143 reclaim_net_ee(mdev, &reclaimed);
144 spin_unlock_irq(&mdev->req_lock);
145
146 list_for_each_entry_safe(e, t, &reclaimed, w.list)
147 drbd_free_ee(mdev, e);
148 }
149
150 /**
151 * drbd_pp_alloc() - Returns a page, fails only if a signal comes in
152 * @mdev: DRBD device.
153 * @retry: whether or not to retry allocation forever (or until signalled)
154 *
155 * Tries to allocate a page, first from our own page pool, then from the
156 * kernel, unless this allocation would exceed the max_buffers setting.
157 * If @retry is non-zero, retry until DRBD frees a page somewhere else.
158 */
159 static struct page *drbd_pp_alloc(struct drbd_conf *mdev, int retry)
160 {
161 struct page *page = NULL;
162 DEFINE_WAIT(wait);
163
164 if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) {
165 page = drbd_pp_first_page_or_try_alloc(mdev);
166 if (page)
167 return page;
168 }
169
170 for (;;) {
171 prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE);
172
173 drbd_kick_lo_and_reclaim_net(mdev);
174
175 if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) {
176 page = drbd_pp_first_page_or_try_alloc(mdev);
177 if (page)
178 break;
179 }
180
181 if (!retry)
182 break;
183
184 if (signal_pending(current)) {
185 dev_warn(DEV, "drbd_pp_alloc interrupted!\n");
186 break;
187 }
188
189 schedule();
190 }
191 finish_wait(&drbd_pp_wait, &wait);
192
193 return page;
194 }
195
196 /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc.
197 * Is also used from inside an other spin_lock_irq(&mdev->req_lock) */
198 static void drbd_pp_free(struct drbd_conf *mdev, struct page *page)
199 {
200 int free_it;
201
202 spin_lock(&drbd_pp_lock);
203 if (drbd_pp_vacant > (DRBD_MAX_SEGMENT_SIZE/PAGE_SIZE)*minor_count) {
204 free_it = 1;
205 } else {
206 set_page_private(page, (unsigned long)drbd_pp_pool);
207 drbd_pp_pool = page;
208 drbd_pp_vacant++;
209 free_it = 0;
210 }
211 spin_unlock(&drbd_pp_lock);
212
213 atomic_dec(&mdev->pp_in_use);
214
215 if (free_it)
216 __free_page(page);
217
218 wake_up(&drbd_pp_wait);
219 }
220
221 static void drbd_pp_free_bio_pages(struct drbd_conf *mdev, struct bio *bio)
222 {
223 struct page *p_to_be_freed = NULL;
224 struct page *page;
225 struct bio_vec *bvec;
226 int i;
227
228 spin_lock(&drbd_pp_lock);
229 __bio_for_each_segment(bvec, bio, i, 0) {
230 if (drbd_pp_vacant > (DRBD_MAX_SEGMENT_SIZE/PAGE_SIZE)*minor_count) {
231 set_page_private(bvec->bv_page, (unsigned long)p_to_be_freed);
232 p_to_be_freed = bvec->bv_page;
233 } else {
234 set_page_private(bvec->bv_page, (unsigned long)drbd_pp_pool);
235 drbd_pp_pool = bvec->bv_page;
236 drbd_pp_vacant++;
237 }
238 }
239 spin_unlock(&drbd_pp_lock);
240 atomic_sub(bio->bi_vcnt, &mdev->pp_in_use);
241
242 while (p_to_be_freed) {
243 page = p_to_be_freed;
244 p_to_be_freed = (struct page *)page_private(page);
245 set_page_private(page, 0); /* just to be polite */
246 put_page(page);
247 }
248
249 wake_up(&drbd_pp_wait);
250 }
251
252 /*
253 You need to hold the req_lock:
254 _drbd_wait_ee_list_empty()
255
256 You must not have the req_lock:
257 drbd_free_ee()
258 drbd_alloc_ee()
259 drbd_init_ee()
260 drbd_release_ee()
261 drbd_ee_fix_bhs()
262 drbd_process_done_ee()
263 drbd_clear_done_ee()
264 drbd_wait_ee_list_empty()
265 */
266
267 struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev,
268 u64 id,
269 sector_t sector,
270 unsigned int data_size,
271 gfp_t gfp_mask) __must_hold(local)
272 {
273 struct request_queue *q;
274 struct drbd_epoch_entry *e;
275 struct page *page;
276 struct bio *bio;
277 unsigned int ds;
278
279 if (FAULT_ACTIVE(mdev, DRBD_FAULT_AL_EE))
280 return NULL;
281
282 e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM);
283 if (!e) {
284 if (!(gfp_mask & __GFP_NOWARN))
285 dev_err(DEV, "alloc_ee: Allocation of an EE failed\n");
286 return NULL;
287 }
288
289 bio = bio_alloc(gfp_mask & ~__GFP_HIGHMEM, div_ceil(data_size, PAGE_SIZE));
290 if (!bio) {
291 if (!(gfp_mask & __GFP_NOWARN))
292 dev_err(DEV, "alloc_ee: Allocation of a bio failed\n");
293 goto fail1;
294 }
295
296 bio->bi_bdev = mdev->ldev->backing_bdev;
297 bio->bi_sector = sector;
298
299 ds = data_size;
300 while (ds) {
301 page = drbd_pp_alloc(mdev, (gfp_mask & __GFP_WAIT));
302 if (!page) {
303 if (!(gfp_mask & __GFP_NOWARN))
304 dev_err(DEV, "alloc_ee: Allocation of a page failed\n");
305 goto fail2;
306 }
307 if (!bio_add_page(bio, page, min_t(int, ds, PAGE_SIZE), 0)) {
308 drbd_pp_free(mdev, page);
309 dev_err(DEV, "alloc_ee: bio_add_page(s=%llu,"
310 "data_size=%u,ds=%u) failed\n",
311 (unsigned long long)sector, data_size, ds);
312
313 q = bdev_get_queue(bio->bi_bdev);
314 if (q->merge_bvec_fn) {
315 struct bvec_merge_data bvm = {
316 .bi_bdev = bio->bi_bdev,
317 .bi_sector = bio->bi_sector,
318 .bi_size = bio->bi_size,
319 .bi_rw = bio->bi_rw,
320 };
321 int l = q->merge_bvec_fn(q, &bvm,
322 &bio->bi_io_vec[bio->bi_vcnt]);
323 dev_err(DEV, "merge_bvec_fn() = %d\n", l);
324 }
325
326 /* dump more of the bio. */
327 dev_err(DEV, "bio->bi_max_vecs = %d\n", bio->bi_max_vecs);
328 dev_err(DEV, "bio->bi_vcnt = %d\n", bio->bi_vcnt);
329 dev_err(DEV, "bio->bi_size = %d\n", bio->bi_size);
330 dev_err(DEV, "bio->bi_phys_segments = %d\n", bio->bi_phys_segments);
331
332 goto fail2;
333 break;
334 }
335 ds -= min_t(int, ds, PAGE_SIZE);
336 }
337
338 D_ASSERT(data_size == bio->bi_size);
339
340 bio->bi_private = e;
341 e->mdev = mdev;
342 e->sector = sector;
343 e->size = bio->bi_size;
344
345 e->private_bio = bio;
346 e->block_id = id;
347 INIT_HLIST_NODE(&e->colision);
348 e->epoch = NULL;
349 e->flags = 0;
350
351 return e;
352
353 fail2:
354 drbd_pp_free_bio_pages(mdev, bio);
355 bio_put(bio);
356 fail1:
357 mempool_free(e, drbd_ee_mempool);
358
359 return NULL;
360 }
361
362 void drbd_free_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e)
363 {
364 struct bio *bio = e->private_bio;
365 drbd_pp_free_bio_pages(mdev, bio);
366 bio_put(bio);
367 D_ASSERT(hlist_unhashed(&e->colision));
368 mempool_free(e, drbd_ee_mempool);
369 }
370
371 int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list)
372 {
373 LIST_HEAD(work_list);
374 struct drbd_epoch_entry *e, *t;
375 int count = 0;
376
377 spin_lock_irq(&mdev->req_lock);
378 list_splice_init(list, &work_list);
379 spin_unlock_irq(&mdev->req_lock);
380
381 list_for_each_entry_safe(e, t, &work_list, w.list) {
382 drbd_free_ee(mdev, e);
383 count++;
384 }
385 return count;
386 }
387
388
389 /*
390 * This function is called from _asender only_
391 * but see also comments in _req_mod(,barrier_acked)
392 * and receive_Barrier.
393 *
394 * Move entries from net_ee to done_ee, if ready.
395 * Grab done_ee, call all callbacks, free the entries.
396 * The callbacks typically send out ACKs.
397 */
398 static int drbd_process_done_ee(struct drbd_conf *mdev)
399 {
400 LIST_HEAD(work_list);
401 LIST_HEAD(reclaimed);
402 struct drbd_epoch_entry *e, *t;
403 int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS);
404
405 spin_lock_irq(&mdev->req_lock);
406 reclaim_net_ee(mdev, &reclaimed);
407 list_splice_init(&mdev->done_ee, &work_list);
408 spin_unlock_irq(&mdev->req_lock);
409
410 list_for_each_entry_safe(e, t, &reclaimed, w.list)
411 drbd_free_ee(mdev, e);
412
413 /* possible callbacks here:
414 * e_end_block, and e_end_resync_block, e_send_discard_ack.
415 * all ignore the last argument.
416 */
417 list_for_each_entry_safe(e, t, &work_list, w.list) {
418 /* list_del not necessary, next/prev members not touched */
419 ok = e->w.cb(mdev, &e->w, !ok) && ok;
420 drbd_free_ee(mdev, e);
421 }
422 wake_up(&mdev->ee_wait);
423
424 return ok;
425 }
426
427 void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head)
428 {
429 DEFINE_WAIT(wait);
430
431 /* avoids spin_lock/unlock
432 * and calling prepare_to_wait in the fast path */
433 while (!list_empty(head)) {
434 prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE);
435 spin_unlock_irq(&mdev->req_lock);
436 drbd_kick_lo(mdev);
437 schedule();
438 finish_wait(&mdev->ee_wait, &wait);
439 spin_lock_irq(&mdev->req_lock);
440 }
441 }
442
443 void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head)
444 {
445 spin_lock_irq(&mdev->req_lock);
446 _drbd_wait_ee_list_empty(mdev, head);
447 spin_unlock_irq(&mdev->req_lock);
448 }
449
450 /* see also kernel_accept; which is only present since 2.6.18.
451 * also we want to log which part of it failed, exactly */
452 static int drbd_accept(struct drbd_conf *mdev, const char **what,
453 struct socket *sock, struct socket **newsock)
454 {
455 struct sock *sk = sock->sk;
456 int err = 0;
457
458 *what = "listen";
459 err = sock->ops->listen(sock, 5);
460 if (err < 0)
461 goto out;
462
463 *what = "sock_create_lite";
464 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
465 newsock);
466 if (err < 0)
467 goto out;
468
469 *what = "accept";
470 err = sock->ops->accept(sock, *newsock, 0);
471 if (err < 0) {
472 sock_release(*newsock);
473 *newsock = NULL;
474 goto out;
475 }
476 (*newsock)->ops = sock->ops;
477
478 out:
479 return err;
480 }
481
482 static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock,
483 void *buf, size_t size, int flags)
484 {
485 mm_segment_t oldfs;
486 struct kvec iov = {
487 .iov_base = buf,
488 .iov_len = size,
489 };
490 struct msghdr msg = {
491 .msg_iovlen = 1,
492 .msg_iov = (struct iovec *)&iov,
493 .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL)
494 };
495 int rv;
496
497 oldfs = get_fs();
498 set_fs(KERNEL_DS);
499 rv = sock_recvmsg(sock, &msg, size, msg.msg_flags);
500 set_fs(oldfs);
501
502 return rv;
503 }
504
505 static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size)
506 {
507 mm_segment_t oldfs;
508 struct kvec iov = {
509 .iov_base = buf,
510 .iov_len = size,
511 };
512 struct msghdr msg = {
513 .msg_iovlen = 1,
514 .msg_iov = (struct iovec *)&iov,
515 .msg_flags = MSG_WAITALL | MSG_NOSIGNAL
516 };
517 int rv;
518
519 oldfs = get_fs();
520 set_fs(KERNEL_DS);
521
522 for (;;) {
523 rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags);
524 if (rv == size)
525 break;
526
527 /* Note:
528 * ECONNRESET other side closed the connection
529 * ERESTARTSYS (on sock) we got a signal
530 */
531
532 if (rv < 0) {
533 if (rv == -ECONNRESET)
534 dev_info(DEV, "sock was reset by peer\n");
535 else if (rv != -ERESTARTSYS)
536 dev_err(DEV, "sock_recvmsg returned %d\n", rv);
537 break;
538 } else if (rv == 0) {
539 dev_info(DEV, "sock was shut down by peer\n");
540 break;
541 } else {
542 /* signal came in, or peer/link went down,
543 * after we read a partial message
544 */
545 /* D_ASSERT(signal_pending(current)); */
546 break;
547 }
548 };
549
550 set_fs(oldfs);
551
552 if (rv != size)
553 drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE));
554
555 return rv;
556 }
557
558 static struct socket *drbd_try_connect(struct drbd_conf *mdev)
559 {
560 const char *what;
561 struct socket *sock;
562 struct sockaddr_in6 src_in6;
563 int err;
564 int disconnect_on_error = 1;
565
566 if (!get_net_conf(mdev))
567 return NULL;
568
569 what = "sock_create_kern";
570 err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family,
571 SOCK_STREAM, IPPROTO_TCP, &sock);
572 if (err < 0) {
573 sock = NULL;
574 goto out;
575 }
576
577 sock->sk->sk_rcvtimeo =
578 sock->sk->sk_sndtimeo = mdev->net_conf->try_connect_int*HZ;
579
580 /* explicitly bind to the configured IP as source IP
581 * for the outgoing connections.
582 * This is needed for multihomed hosts and to be
583 * able to use lo: interfaces for drbd.
584 * Make sure to use 0 as port number, so linux selects
585 * a free one dynamically.
586 */
587 memcpy(&src_in6, mdev->net_conf->my_addr,
588 min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6)));
589 if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6)
590 src_in6.sin6_port = 0;
591 else
592 ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */
593
594 what = "bind before connect";
595 err = sock->ops->bind(sock,
596 (struct sockaddr *) &src_in6,
597 mdev->net_conf->my_addr_len);
598 if (err < 0)
599 goto out;
600
601 /* connect may fail, peer not yet available.
602 * stay C_WF_CONNECTION, don't go Disconnecting! */
603 disconnect_on_error = 0;
604 what = "connect";
605 err = sock->ops->connect(sock,
606 (struct sockaddr *)mdev->net_conf->peer_addr,
607 mdev->net_conf->peer_addr_len, 0);
608
609 out:
610 if (err < 0) {
611 if (sock) {
612 sock_release(sock);
613 sock = NULL;
614 }
615 switch (-err) {
616 /* timeout, busy, signal pending */
617 case ETIMEDOUT: case EAGAIN: case EINPROGRESS:
618 case EINTR: case ERESTARTSYS:
619 /* peer not (yet) available, network problem */
620 case ECONNREFUSED: case ENETUNREACH:
621 case EHOSTDOWN: case EHOSTUNREACH:
622 disconnect_on_error = 0;
623 break;
624 default:
625 dev_err(DEV, "%s failed, err = %d\n", what, err);
626 }
627 if (disconnect_on_error)
628 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
629 }
630 put_net_conf(mdev);
631 return sock;
632 }
633
634 static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev)
635 {
636 int timeo, err;
637 struct socket *s_estab = NULL, *s_listen;
638 const char *what;
639
640 if (!get_net_conf(mdev))
641 return NULL;
642
643 what = "sock_create_kern";
644 err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family,
645 SOCK_STREAM, IPPROTO_TCP, &s_listen);
646 if (err) {
647 s_listen = NULL;
648 goto out;
649 }
650
651 timeo = mdev->net_conf->try_connect_int * HZ;
652 timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */
653
654 s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */
655 s_listen->sk->sk_rcvtimeo = timeo;
656 s_listen->sk->sk_sndtimeo = timeo;
657
658 what = "bind before listen";
659 err = s_listen->ops->bind(s_listen,
660 (struct sockaddr *) mdev->net_conf->my_addr,
661 mdev->net_conf->my_addr_len);
662 if (err < 0)
663 goto out;
664
665 err = drbd_accept(mdev, &what, s_listen, &s_estab);
666
667 out:
668 if (s_listen)
669 sock_release(s_listen);
670 if (err < 0) {
671 if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) {
672 dev_err(DEV, "%s failed, err = %d\n", what, err);
673 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
674 }
675 }
676 put_net_conf(mdev);
677
678 return s_estab;
679 }
680
681 static int drbd_send_fp(struct drbd_conf *mdev,
682 struct socket *sock, enum drbd_packets cmd)
683 {
684 struct p_header *h = (struct p_header *) &mdev->data.sbuf.header;
685
686 return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0);
687 }
688
689 static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock)
690 {
691 struct p_header *h = (struct p_header *) &mdev->data.sbuf.header;
692 int rr;
693
694 rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0);
695
696 if (rr == sizeof(*h) && h->magic == BE_DRBD_MAGIC)
697 return be16_to_cpu(h->command);
698
699 return 0xffff;
700 }
701
702 /**
703 * drbd_socket_okay() - Free the socket if its connection is not okay
704 * @mdev: DRBD device.
705 * @sock: pointer to the pointer to the socket.
706 */
707 static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock)
708 {
709 int rr;
710 char tb[4];
711
712 if (!*sock)
713 return FALSE;
714
715 rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK);
716
717 if (rr > 0 || rr == -EAGAIN) {
718 return TRUE;
719 } else {
720 sock_release(*sock);
721 *sock = NULL;
722 return FALSE;
723 }
724 }
725
726 /*
727 * return values:
728 * 1 yes, we have a valid connection
729 * 0 oops, did not work out, please try again
730 * -1 peer talks different language,
731 * no point in trying again, please go standalone.
732 * -2 We do not have a network config...
733 */
734 static int drbd_connect(struct drbd_conf *mdev)
735 {
736 struct socket *s, *sock, *msock;
737 int try, h, ok;
738
739 D_ASSERT(!mdev->data.socket);
740
741 if (test_and_clear_bit(CREATE_BARRIER, &mdev->flags))
742 dev_err(DEV, "CREATE_BARRIER flag was set in drbd_connect - now cleared!\n");
743
744 if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS)
745 return -2;
746
747 clear_bit(DISCARD_CONCURRENT, &mdev->flags);
748
749 sock = NULL;
750 msock = NULL;
751
752 do {
753 for (try = 0;;) {
754 /* 3 tries, this should take less than a second! */
755 s = drbd_try_connect(mdev);
756 if (s || ++try >= 3)
757 break;
758 /* give the other side time to call bind() & listen() */
759 __set_current_state(TASK_INTERRUPTIBLE);
760 schedule_timeout(HZ / 10);
761 }
762
763 if (s) {
764 if (!sock) {
765 drbd_send_fp(mdev, s, P_HAND_SHAKE_S);
766 sock = s;
767 s = NULL;
768 } else if (!msock) {
769 drbd_send_fp(mdev, s, P_HAND_SHAKE_M);
770 msock = s;
771 s = NULL;
772 } else {
773 dev_err(DEV, "Logic error in drbd_connect()\n");
774 goto out_release_sockets;
775 }
776 }
777
778 if (sock && msock) {
779 __set_current_state(TASK_INTERRUPTIBLE);
780 schedule_timeout(HZ / 10);
781 ok = drbd_socket_okay(mdev, &sock);
782 ok = drbd_socket_okay(mdev, &msock) && ok;
783 if (ok)
784 break;
785 }
786
787 retry:
788 s = drbd_wait_for_connect(mdev);
789 if (s) {
790 try = drbd_recv_fp(mdev, s);
791 drbd_socket_okay(mdev, &sock);
792 drbd_socket_okay(mdev, &msock);
793 switch (try) {
794 case P_HAND_SHAKE_S:
795 if (sock) {
796 dev_warn(DEV, "initial packet S crossed\n");
797 sock_release(sock);
798 }
799 sock = s;
800 break;
801 case P_HAND_SHAKE_M:
802 if (msock) {
803 dev_warn(DEV, "initial packet M crossed\n");
804 sock_release(msock);
805 }
806 msock = s;
807 set_bit(DISCARD_CONCURRENT, &mdev->flags);
808 break;
809 default:
810 dev_warn(DEV, "Error receiving initial packet\n");
811 sock_release(s);
812 if (random32() & 1)
813 goto retry;
814 }
815 }
816
817 if (mdev->state.conn <= C_DISCONNECTING)
818 goto out_release_sockets;
819 if (signal_pending(current)) {
820 flush_signals(current);
821 smp_rmb();
822 if (get_t_state(&mdev->receiver) == Exiting)
823 goto out_release_sockets;
824 }
825
826 if (sock && msock) {
827 ok = drbd_socket_okay(mdev, &sock);
828 ok = drbd_socket_okay(mdev, &msock) && ok;
829 if (ok)
830 break;
831 }
832 } while (1);
833
834 msock->sk->sk_reuse = 1; /* SO_REUSEADDR */
835 sock->sk->sk_reuse = 1; /* SO_REUSEADDR */
836
837 sock->sk->sk_allocation = GFP_NOIO;
838 msock->sk->sk_allocation = GFP_NOIO;
839
840 sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK;
841 msock->sk->sk_priority = TC_PRIO_INTERACTIVE;
842
843 if (mdev->net_conf->sndbuf_size) {
844 sock->sk->sk_sndbuf = mdev->net_conf->sndbuf_size;
845 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
846 }
847
848 if (mdev->net_conf->rcvbuf_size) {
849 sock->sk->sk_rcvbuf = mdev->net_conf->rcvbuf_size;
850 sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
851 }
852
853 /* NOT YET ...
854 * sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
855 * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
856 * first set it to the P_HAND_SHAKE timeout,
857 * which we set to 4x the configured ping_timeout. */
858 sock->sk->sk_sndtimeo =
859 sock->sk->sk_rcvtimeo = mdev->net_conf->ping_timeo*4*HZ/10;
860
861 msock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
862 msock->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
863
864 /* we don't want delays.
865 * we use TCP_CORK where apropriate, though */
866 drbd_tcp_nodelay(sock);
867 drbd_tcp_nodelay(msock);
868
869 mdev->data.socket = sock;
870 mdev->meta.socket = msock;
871 mdev->last_received = jiffies;
872
873 D_ASSERT(mdev->asender.task == NULL);
874
875 h = drbd_do_handshake(mdev);
876 if (h <= 0)
877 return h;
878
879 if (mdev->cram_hmac_tfm) {
880 /* drbd_request_state(mdev, NS(conn, WFAuth)); */
881 switch (drbd_do_auth(mdev)) {
882 case -1:
883 dev_err(DEV, "Authentication of peer failed\n");
884 return -1;
885 case 0:
886 dev_err(DEV, "Authentication of peer failed, trying again.\n");
887 return 0;
888 }
889 }
890
891 if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS)
892 return 0;
893
894 sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
895 sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
896
897 atomic_set(&mdev->packet_seq, 0);
898 mdev->peer_seq = 0;
899
900 drbd_thread_start(&mdev->asender);
901
902 if (!drbd_send_protocol(mdev))
903 return -1;
904 drbd_send_sync_param(mdev, &mdev->sync_conf);
905 drbd_send_sizes(mdev, 0);
906 drbd_send_uuids(mdev);
907 drbd_send_state(mdev);
908 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
909 clear_bit(RESIZE_PENDING, &mdev->flags);
910
911 return 1;
912
913 out_release_sockets:
914 if (sock)
915 sock_release(sock);
916 if (msock)
917 sock_release(msock);
918 return -1;
919 }
920
921 static int drbd_recv_header(struct drbd_conf *mdev, struct p_header *h)
922 {
923 int r;
924
925 r = drbd_recv(mdev, h, sizeof(*h));
926
927 if (unlikely(r != sizeof(*h))) {
928 dev_err(DEV, "short read expecting header on sock: r=%d\n", r);
929 return FALSE;
930 };
931 h->command = be16_to_cpu(h->command);
932 h->length = be16_to_cpu(h->length);
933 if (unlikely(h->magic != BE_DRBD_MAGIC)) {
934 dev_err(DEV, "magic?? on data m: 0x%lx c: %d l: %d\n",
935 (long)be32_to_cpu(h->magic),
936 h->command, h->length);
937 return FALSE;
938 }
939 mdev->last_received = jiffies;
940
941 return TRUE;
942 }
943
944 static enum finish_epoch drbd_flush_after_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch)
945 {
946 int rv;
947
948 if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) {
949 rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL,
950 NULL, BLKDEV_IFL_WAIT);
951 if (rv) {
952 dev_err(DEV, "local disk flush failed with status %d\n", rv);
953 /* would rather check on EOPNOTSUPP, but that is not reliable.
954 * don't try again for ANY return value != 0
955 * if (rv == -EOPNOTSUPP) */
956 drbd_bump_write_ordering(mdev, WO_drain_io);
957 }
958 put_ldev(mdev);
959 }
960
961 return drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE);
962 }
963
964 static int w_flush(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
965 {
966 struct flush_work *fw = (struct flush_work *)w;
967 struct drbd_epoch *epoch = fw->epoch;
968
969 kfree(w);
970
971 if (!test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags))
972 drbd_flush_after_epoch(mdev, epoch);
973
974 drbd_may_finish_epoch(mdev, epoch, EV_PUT |
975 (mdev->state.conn < C_CONNECTED ? EV_CLEANUP : 0));
976
977 return 1;
978 }
979
980 /**
981 * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it.
982 * @mdev: DRBD device.
983 * @epoch: Epoch object.
984 * @ev: Epoch event.
985 */
986 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev,
987 struct drbd_epoch *epoch,
988 enum epoch_event ev)
989 {
990 int finish, epoch_size;
991 struct drbd_epoch *next_epoch;
992 int schedule_flush = 0;
993 enum finish_epoch rv = FE_STILL_LIVE;
994
995 spin_lock(&mdev->epoch_lock);
996 do {
997 next_epoch = NULL;
998 finish = 0;
999
1000 epoch_size = atomic_read(&epoch->epoch_size);
1001
1002 switch (ev & ~EV_CLEANUP) {
1003 case EV_PUT:
1004 atomic_dec(&epoch->active);
1005 break;
1006 case EV_GOT_BARRIER_NR:
1007 set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags);
1008
1009 /* Special case: If we just switched from WO_bio_barrier to
1010 WO_bdev_flush we should not finish the current epoch */
1011 if (test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags) && epoch_size == 1 &&
1012 mdev->write_ordering != WO_bio_barrier &&
1013 epoch == mdev->current_epoch)
1014 clear_bit(DE_CONTAINS_A_BARRIER, &epoch->flags);
1015 break;
1016 case EV_BARRIER_DONE:
1017 set_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags);
1018 break;
1019 case EV_BECAME_LAST:
1020 /* nothing to do*/
1021 break;
1022 }
1023
1024 if (epoch_size != 0 &&
1025 atomic_read(&epoch->active) == 0 &&
1026 test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) &&
1027 epoch->list.prev == &mdev->current_epoch->list &&
1028 !test_bit(DE_IS_FINISHING, &epoch->flags)) {
1029 /* Nearly all conditions are met to finish that epoch... */
1030 if (test_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags) ||
1031 mdev->write_ordering == WO_none ||
1032 (epoch_size == 1 && test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) ||
1033 ev & EV_CLEANUP) {
1034 finish = 1;
1035 set_bit(DE_IS_FINISHING, &epoch->flags);
1036 } else if (!test_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags) &&
1037 mdev->write_ordering == WO_bio_barrier) {
1038 atomic_inc(&epoch->active);
1039 schedule_flush = 1;
1040 }
1041 }
1042 if (finish) {
1043 if (!(ev & EV_CLEANUP)) {
1044 spin_unlock(&mdev->epoch_lock);
1045 drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size);
1046 spin_lock(&mdev->epoch_lock);
1047 }
1048 dec_unacked(mdev);
1049
1050 if (mdev->current_epoch != epoch) {
1051 next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list);
1052 list_del(&epoch->list);
1053 ev = EV_BECAME_LAST | (ev & EV_CLEANUP);
1054 mdev->epochs--;
1055 kfree(epoch);
1056
1057 if (rv == FE_STILL_LIVE)
1058 rv = FE_DESTROYED;
1059 } else {
1060 epoch->flags = 0;
1061 atomic_set(&epoch->epoch_size, 0);
1062 /* atomic_set(&epoch->active, 0); is alrady zero */
1063 if (rv == FE_STILL_LIVE)
1064 rv = FE_RECYCLED;
1065 }
1066 }
1067
1068 if (!next_epoch)
1069 break;
1070
1071 epoch = next_epoch;
1072 } while (1);
1073
1074 spin_unlock(&mdev->epoch_lock);
1075
1076 if (schedule_flush) {
1077 struct flush_work *fw;
1078 fw = kmalloc(sizeof(*fw), GFP_ATOMIC);
1079 if (fw) {
1080 fw->w.cb = w_flush;
1081 fw->epoch = epoch;
1082 drbd_queue_work(&mdev->data.work, &fw->w);
1083 } else {
1084 dev_warn(DEV, "Could not kmalloc a flush_work obj\n");
1085 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags);
1086 /* That is not a recursion, only one level */
1087 drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE);
1088 drbd_may_finish_epoch(mdev, epoch, EV_PUT);
1089 }
1090 }
1091
1092 return rv;
1093 }
1094
1095 /**
1096 * drbd_bump_write_ordering() - Fall back to an other write ordering method
1097 * @mdev: DRBD device.
1098 * @wo: Write ordering method to try.
1099 */
1100 void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local)
1101 {
1102 enum write_ordering_e pwo;
1103 static char *write_ordering_str[] = {
1104 [WO_none] = "none",
1105 [WO_drain_io] = "drain",
1106 [WO_bdev_flush] = "flush",
1107 [WO_bio_barrier] = "barrier",
1108 };
1109
1110 pwo = mdev->write_ordering;
1111 wo = min(pwo, wo);
1112 if (wo == WO_bio_barrier && mdev->ldev->dc.no_disk_barrier)
1113 wo = WO_bdev_flush;
1114 if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush)
1115 wo = WO_drain_io;
1116 if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain)
1117 wo = WO_none;
1118 mdev->write_ordering = wo;
1119 if (pwo != mdev->write_ordering || wo == WO_bio_barrier)
1120 dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]);
1121 }
1122
1123 /**
1124 * w_e_reissue() - Worker callback; Resubmit a bio, without BIO_RW_BARRIER set
1125 * @mdev: DRBD device.
1126 * @w: work object.
1127 * @cancel: The connection will be closed anyways (unused in this callback)
1128 */
1129 int w_e_reissue(struct drbd_conf *mdev, struct drbd_work *w, int cancel) __releases(local)
1130 {
1131 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1132 struct bio *bio = e->private_bio;
1133
1134 /* We leave DE_CONTAINS_A_BARRIER and EE_IS_BARRIER in place,
1135 (and DE_BARRIER_IN_NEXT_EPOCH_ISSUED in the previous Epoch)
1136 so that we can finish that epoch in drbd_may_finish_epoch().
1137 That is necessary if we already have a long chain of Epochs, before
1138 we realize that BIO_RW_BARRIER is actually not supported */
1139
1140 /* As long as the -ENOTSUPP on the barrier is reported immediately
1141 that will never trigger. If it is reported late, we will just
1142 print that warning and continue correctly for all future requests
1143 with WO_bdev_flush */
1144 if (previous_epoch(mdev, e->epoch))
1145 dev_warn(DEV, "Write ordering was not enforced (one time event)\n");
1146
1147 /* prepare bio for re-submit,
1148 * re-init volatile members */
1149 /* we still have a local reference,
1150 * get_ldev was done in receive_Data. */
1151 bio->bi_bdev = mdev->ldev->backing_bdev;
1152 bio->bi_sector = e->sector;
1153 bio->bi_size = e->size;
1154 bio->bi_idx = 0;
1155
1156 bio->bi_flags &= ~(BIO_POOL_MASK - 1);
1157 bio->bi_flags |= 1 << BIO_UPTODATE;
1158
1159 /* don't know whether this is necessary: */
1160 bio->bi_phys_segments = 0;
1161 bio->bi_next = NULL;
1162
1163 /* these should be unchanged: */
1164 /* bio->bi_end_io = drbd_endio_write_sec; */
1165 /* bio->bi_vcnt = whatever; */
1166
1167 e->w.cb = e_end_block;
1168
1169 /* This is no longer a barrier request. */
1170 bio->bi_rw &= ~(1UL << BIO_RW_BARRIER);
1171
1172 drbd_generic_make_request(mdev, DRBD_FAULT_DT_WR, bio);
1173
1174 return 1;
1175 }
1176
1177 static int receive_Barrier(struct drbd_conf *mdev, struct p_header *h)
1178 {
1179 int rv, issue_flush;
1180 struct p_barrier *p = (struct p_barrier *)h;
1181 struct drbd_epoch *epoch;
1182
1183 ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE;
1184
1185 rv = drbd_recv(mdev, h->payload, h->length);
1186 ERR_IF(rv != h->length) return FALSE;
1187
1188 inc_unacked(mdev);
1189
1190 if (mdev->net_conf->wire_protocol != DRBD_PROT_C)
1191 drbd_kick_lo(mdev);
1192
1193 mdev->current_epoch->barrier_nr = p->barrier;
1194 rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR);
1195
1196 /* P_BARRIER_ACK may imply that the corresponding extent is dropped from
1197 * the activity log, which means it would not be resynced in case the
1198 * R_PRIMARY crashes now.
1199 * Therefore we must send the barrier_ack after the barrier request was
1200 * completed. */
1201 switch (mdev->write_ordering) {
1202 case WO_bio_barrier:
1203 case WO_none:
1204 if (rv == FE_RECYCLED)
1205 return TRUE;
1206 break;
1207
1208 case WO_bdev_flush:
1209 case WO_drain_io:
1210 if (rv == FE_STILL_LIVE) {
1211 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags);
1212 drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
1213 rv = drbd_flush_after_epoch(mdev, mdev->current_epoch);
1214 }
1215 if (rv == FE_RECYCLED)
1216 return TRUE;
1217
1218 /* The asender will send all the ACKs and barrier ACKs out, since
1219 all EEs moved from the active_ee to the done_ee. We need to
1220 provide a new epoch object for the EEs that come in soon */
1221 break;
1222 }
1223
1224 /* receiver context, in the writeout path of the other node.
1225 * avoid potential distributed deadlock */
1226 epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO);
1227 if (!epoch) {
1228 dev_warn(DEV, "Allocation of an epoch failed, slowing down\n");
1229 issue_flush = !test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags);
1230 drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
1231 if (issue_flush) {
1232 rv = drbd_flush_after_epoch(mdev, mdev->current_epoch);
1233 if (rv == FE_RECYCLED)
1234 return TRUE;
1235 }
1236
1237 drbd_wait_ee_list_empty(mdev, &mdev->done_ee);
1238
1239 return TRUE;
1240 }
1241
1242 epoch->flags = 0;
1243 atomic_set(&epoch->epoch_size, 0);
1244 atomic_set(&epoch->active, 0);
1245
1246 spin_lock(&mdev->epoch_lock);
1247 if (atomic_read(&mdev->current_epoch->epoch_size)) {
1248 list_add(&epoch->list, &mdev->current_epoch->list);
1249 mdev->current_epoch = epoch;
1250 mdev->epochs++;
1251 } else {
1252 /* The current_epoch got recycled while we allocated this one... */
1253 kfree(epoch);
1254 }
1255 spin_unlock(&mdev->epoch_lock);
1256
1257 return TRUE;
1258 }
1259
1260 /* used from receive_RSDataReply (recv_resync_read)
1261 * and from receive_Data */
1262 static struct drbd_epoch_entry *
1263 read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local)
1264 {
1265 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
1266 struct drbd_epoch_entry *e;
1267 struct bio_vec *bvec;
1268 struct page *page;
1269 struct bio *bio;
1270 int dgs, ds, i, rr;
1271 void *dig_in = mdev->int_dig_in;
1272 void *dig_vv = mdev->int_dig_vv;
1273
1274 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1275 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1276
1277 if (dgs) {
1278 rr = drbd_recv(mdev, dig_in, dgs);
1279 if (rr != dgs) {
1280 dev_warn(DEV, "short read receiving data digest: read %d expected %d\n",
1281 rr, dgs);
1282 return NULL;
1283 }
1284 }
1285
1286 data_size -= dgs;
1287
1288 ERR_IF(data_size & 0x1ff) return NULL;
1289 ERR_IF(data_size > DRBD_MAX_SEGMENT_SIZE) return NULL;
1290
1291 /* even though we trust out peer,
1292 * we sometimes have to double check. */
1293 if (sector + (data_size>>9) > capacity) {
1294 dev_err(DEV, "capacity: %llus < sector: %llus + size: %u\n",
1295 (unsigned long long)capacity,
1296 (unsigned long long)sector, data_size);
1297 return NULL;
1298 }
1299
1300 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
1301 * "criss-cross" setup, that might cause write-out on some other DRBD,
1302 * which in turn might block on the other node at this very place. */
1303 e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO);
1304 if (!e)
1305 return NULL;
1306 bio = e->private_bio;
1307 ds = data_size;
1308 bio_for_each_segment(bvec, bio, i) {
1309 page = bvec->bv_page;
1310 rr = drbd_recv(mdev, kmap(page), min_t(int, ds, PAGE_SIZE));
1311 kunmap(page);
1312 if (rr != min_t(int, ds, PAGE_SIZE)) {
1313 drbd_free_ee(mdev, e);
1314 dev_warn(DEV, "short read receiving data: read %d expected %d\n",
1315 rr, min_t(int, ds, PAGE_SIZE));
1316 return NULL;
1317 }
1318 ds -= rr;
1319 }
1320
1321 if (dgs) {
1322 drbd_csum(mdev, mdev->integrity_r_tfm, bio, dig_vv);
1323 if (memcmp(dig_in, dig_vv, dgs)) {
1324 dev_err(DEV, "Digest integrity check FAILED.\n");
1325 drbd_bcast_ee(mdev, "digest failed",
1326 dgs, dig_in, dig_vv, e);
1327 drbd_free_ee(mdev, e);
1328 return NULL;
1329 }
1330 }
1331 mdev->recv_cnt += data_size>>9;
1332 return e;
1333 }
1334
1335 /* drbd_drain_block() just takes a data block
1336 * out of the socket input buffer, and discards it.
1337 */
1338 static int drbd_drain_block(struct drbd_conf *mdev, int data_size)
1339 {
1340 struct page *page;
1341 int rr, rv = 1;
1342 void *data;
1343
1344 if (!data_size)
1345 return TRUE;
1346
1347 page = drbd_pp_alloc(mdev, 1);
1348
1349 data = kmap(page);
1350 while (data_size) {
1351 rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE));
1352 if (rr != min_t(int, data_size, PAGE_SIZE)) {
1353 rv = 0;
1354 dev_warn(DEV, "short read receiving data: read %d expected %d\n",
1355 rr, min_t(int, data_size, PAGE_SIZE));
1356 break;
1357 }
1358 data_size -= rr;
1359 }
1360 kunmap(page);
1361 drbd_pp_free(mdev, page);
1362 return rv;
1363 }
1364
1365 static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req,
1366 sector_t sector, int data_size)
1367 {
1368 struct bio_vec *bvec;
1369 struct bio *bio;
1370 int dgs, rr, i, expect;
1371 void *dig_in = mdev->int_dig_in;
1372 void *dig_vv = mdev->int_dig_vv;
1373
1374 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1375 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1376
1377 if (dgs) {
1378 rr = drbd_recv(mdev, dig_in, dgs);
1379 if (rr != dgs) {
1380 dev_warn(DEV, "short read receiving data reply digest: read %d expected %d\n",
1381 rr, dgs);
1382 return 0;
1383 }
1384 }
1385
1386 data_size -= dgs;
1387
1388 /* optimistically update recv_cnt. if receiving fails below,
1389 * we disconnect anyways, and counters will be reset. */
1390 mdev->recv_cnt += data_size>>9;
1391
1392 bio = req->master_bio;
1393 D_ASSERT(sector == bio->bi_sector);
1394
1395 bio_for_each_segment(bvec, bio, i) {
1396 expect = min_t(int, data_size, bvec->bv_len);
1397 rr = drbd_recv(mdev,
1398 kmap(bvec->bv_page)+bvec->bv_offset,
1399 expect);
1400 kunmap(bvec->bv_page);
1401 if (rr != expect) {
1402 dev_warn(DEV, "short read receiving data reply: "
1403 "read %d expected %d\n",
1404 rr, expect);
1405 return 0;
1406 }
1407 data_size -= rr;
1408 }
1409
1410 if (dgs) {
1411 drbd_csum(mdev, mdev->integrity_r_tfm, bio, dig_vv);
1412 if (memcmp(dig_in, dig_vv, dgs)) {
1413 dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n");
1414 return 0;
1415 }
1416 }
1417
1418 D_ASSERT(data_size == 0);
1419 return 1;
1420 }
1421
1422 /* e_end_resync_block() is called via
1423 * drbd_process_done_ee() by asender only */
1424 static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1425 {
1426 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1427 sector_t sector = e->sector;
1428 int ok;
1429
1430 D_ASSERT(hlist_unhashed(&e->colision));
1431
1432 if (likely(drbd_bio_uptodate(e->private_bio))) {
1433 drbd_set_in_sync(mdev, sector, e->size);
1434 ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e);
1435 } else {
1436 /* Record failure to sync */
1437 drbd_rs_failed_io(mdev, sector, e->size);
1438
1439 ok = drbd_send_ack(mdev, P_NEG_ACK, e);
1440 }
1441 dec_unacked(mdev);
1442
1443 return ok;
1444 }
1445
1446 static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local)
1447 {
1448 struct drbd_epoch_entry *e;
1449
1450 e = read_in_block(mdev, ID_SYNCER, sector, data_size);
1451 if (!e) {
1452 put_ldev(mdev);
1453 return FALSE;
1454 }
1455
1456 dec_rs_pending(mdev);
1457
1458 e->private_bio->bi_end_io = drbd_endio_write_sec;
1459 e->private_bio->bi_rw = WRITE;
1460 e->w.cb = e_end_resync_block;
1461
1462 inc_unacked(mdev);
1463 /* corresponding dec_unacked() in e_end_resync_block()
1464 * respective _drbd_clear_done_ee */
1465
1466 spin_lock_irq(&mdev->req_lock);
1467 list_add(&e->w.list, &mdev->sync_ee);
1468 spin_unlock_irq(&mdev->req_lock);
1469
1470 drbd_generic_make_request(mdev, DRBD_FAULT_RS_WR, e->private_bio);
1471 /* accounting done in endio */
1472
1473 maybe_kick_lo(mdev);
1474 return TRUE;
1475 }
1476
1477 static int receive_DataReply(struct drbd_conf *mdev, struct p_header *h)
1478 {
1479 struct drbd_request *req;
1480 sector_t sector;
1481 unsigned int header_size, data_size;
1482 int ok;
1483 struct p_data *p = (struct p_data *)h;
1484
1485 header_size = sizeof(*p) - sizeof(*h);
1486 data_size = h->length - header_size;
1487
1488 ERR_IF(data_size == 0) return FALSE;
1489
1490 if (drbd_recv(mdev, h->payload, header_size) != header_size)
1491 return FALSE;
1492
1493 sector = be64_to_cpu(p->sector);
1494
1495 spin_lock_irq(&mdev->req_lock);
1496 req = _ar_id_to_req(mdev, p->block_id, sector);
1497 spin_unlock_irq(&mdev->req_lock);
1498 if (unlikely(!req)) {
1499 dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n");
1500 return FALSE;
1501 }
1502
1503 /* hlist_del(&req->colision) is done in _req_may_be_done, to avoid
1504 * special casing it there for the various failure cases.
1505 * still no race with drbd_fail_pending_reads */
1506 ok = recv_dless_read(mdev, req, sector, data_size);
1507
1508 if (ok)
1509 req_mod(req, data_received);
1510 /* else: nothing. handled from drbd_disconnect...
1511 * I don't think we may complete this just yet
1512 * in case we are "on-disconnect: freeze" */
1513
1514 return ok;
1515 }
1516
1517 static int receive_RSDataReply(struct drbd_conf *mdev, struct p_header *h)
1518 {
1519 sector_t sector;
1520 unsigned int header_size, data_size;
1521 int ok;
1522 struct p_data *p = (struct p_data *)h;
1523
1524 header_size = sizeof(*p) - sizeof(*h);
1525 data_size = h->length - header_size;
1526
1527 ERR_IF(data_size == 0) return FALSE;
1528
1529 if (drbd_recv(mdev, h->payload, header_size) != header_size)
1530 return FALSE;
1531
1532 sector = be64_to_cpu(p->sector);
1533 D_ASSERT(p->block_id == ID_SYNCER);
1534
1535 if (get_ldev(mdev)) {
1536 /* data is submitted to disk within recv_resync_read.
1537 * corresponding put_ldev done below on error,
1538 * or in drbd_endio_write_sec. */
1539 ok = recv_resync_read(mdev, sector, data_size);
1540 } else {
1541 if (__ratelimit(&drbd_ratelimit_state))
1542 dev_err(DEV, "Can not write resync data to local disk.\n");
1543
1544 ok = drbd_drain_block(mdev, data_size);
1545
1546 drbd_send_ack_dp(mdev, P_NEG_ACK, p);
1547 }
1548
1549 return ok;
1550 }
1551
1552 /* e_end_block() is called via drbd_process_done_ee().
1553 * this means this function only runs in the asender thread
1554 */
1555 static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1556 {
1557 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1558 sector_t sector = e->sector;
1559 struct drbd_epoch *epoch;
1560 int ok = 1, pcmd;
1561
1562 if (e->flags & EE_IS_BARRIER) {
1563 epoch = previous_epoch(mdev, e->epoch);
1564 if (epoch)
1565 drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE + (cancel ? EV_CLEANUP : 0));
1566 }
1567
1568 if (mdev->net_conf->wire_protocol == DRBD_PROT_C) {
1569 if (likely(drbd_bio_uptodate(e->private_bio))) {
1570 pcmd = (mdev->state.conn >= C_SYNC_SOURCE &&
1571 mdev->state.conn <= C_PAUSED_SYNC_T &&
1572 e->flags & EE_MAY_SET_IN_SYNC) ?
1573 P_RS_WRITE_ACK : P_WRITE_ACK;
1574 ok &= drbd_send_ack(mdev, pcmd, e);
1575 if (pcmd == P_RS_WRITE_ACK)
1576 drbd_set_in_sync(mdev, sector, e->size);
1577 } else {
1578 ok = drbd_send_ack(mdev, P_NEG_ACK, e);
1579 /* we expect it to be marked out of sync anyways...
1580 * maybe assert this? */
1581 }
1582 dec_unacked(mdev);
1583 }
1584 /* we delete from the conflict detection hash _after_ we sent out the
1585 * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */
1586 if (mdev->net_conf->two_primaries) {
1587 spin_lock_irq(&mdev->req_lock);
1588 D_ASSERT(!hlist_unhashed(&e->colision));
1589 hlist_del_init(&e->colision);
1590 spin_unlock_irq(&mdev->req_lock);
1591 } else {
1592 D_ASSERT(hlist_unhashed(&e->colision));
1593 }
1594
1595 drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0));
1596
1597 return ok;
1598 }
1599
1600 static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1601 {
1602 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1603 int ok = 1;
1604
1605 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1606 ok = drbd_send_ack(mdev, P_DISCARD_ACK, e);
1607
1608 spin_lock_irq(&mdev->req_lock);
1609 D_ASSERT(!hlist_unhashed(&e->colision));
1610 hlist_del_init(&e->colision);
1611 spin_unlock_irq(&mdev->req_lock);
1612
1613 dec_unacked(mdev);
1614
1615 return ok;
1616 }
1617
1618 /* Called from receive_Data.
1619 * Synchronize packets on sock with packets on msock.
1620 *
1621 * This is here so even when a P_DATA packet traveling via sock overtook an Ack
1622 * packet traveling on msock, they are still processed in the order they have
1623 * been sent.
1624 *
1625 * Note: we don't care for Ack packets overtaking P_DATA packets.
1626 *
1627 * In case packet_seq is larger than mdev->peer_seq number, there are
1628 * outstanding packets on the msock. We wait for them to arrive.
1629 * In case we are the logically next packet, we update mdev->peer_seq
1630 * ourselves. Correctly handles 32bit wrap around.
1631 *
1632 * Assume we have a 10 GBit connection, that is about 1<<30 byte per second,
1633 * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds
1634 * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have
1635 * 1<<9 == 512 seconds aka ages for the 32bit wrap around...
1636 *
1637 * returns 0 if we may process the packet,
1638 * -ERESTARTSYS if we were interrupted (by disconnect signal). */
1639 static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq)
1640 {
1641 DEFINE_WAIT(wait);
1642 unsigned int p_seq;
1643 long timeout;
1644 int ret = 0;
1645 spin_lock(&mdev->peer_seq_lock);
1646 for (;;) {
1647 prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE);
1648 if (seq_le(packet_seq, mdev->peer_seq+1))
1649 break;
1650 if (signal_pending(current)) {
1651 ret = -ERESTARTSYS;
1652 break;
1653 }
1654 p_seq = mdev->peer_seq;
1655 spin_unlock(&mdev->peer_seq_lock);
1656 timeout = schedule_timeout(30*HZ);
1657 spin_lock(&mdev->peer_seq_lock);
1658 if (timeout == 0 && p_seq == mdev->peer_seq) {
1659 ret = -ETIMEDOUT;
1660 dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n");
1661 break;
1662 }
1663 }
1664 finish_wait(&mdev->seq_wait, &wait);
1665 if (mdev->peer_seq+1 == packet_seq)
1666 mdev->peer_seq++;
1667 spin_unlock(&mdev->peer_seq_lock);
1668 return ret;
1669 }
1670
1671 /* mirrored write */
1672 static int receive_Data(struct drbd_conf *mdev, struct p_header *h)
1673 {
1674 sector_t sector;
1675 struct drbd_epoch_entry *e;
1676 struct p_data *p = (struct p_data *)h;
1677 int header_size, data_size;
1678 int rw = WRITE;
1679 u32 dp_flags;
1680
1681 header_size = sizeof(*p) - sizeof(*h);
1682 data_size = h->length - header_size;
1683
1684 ERR_IF(data_size == 0) return FALSE;
1685
1686 if (drbd_recv(mdev, h->payload, header_size) != header_size)
1687 return FALSE;
1688
1689 if (!get_ldev(mdev)) {
1690 if (__ratelimit(&drbd_ratelimit_state))
1691 dev_err(DEV, "Can not write mirrored data block "
1692 "to local disk.\n");
1693 spin_lock(&mdev->peer_seq_lock);
1694 if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num))
1695 mdev->peer_seq++;
1696 spin_unlock(&mdev->peer_seq_lock);
1697
1698 drbd_send_ack_dp(mdev, P_NEG_ACK, p);
1699 atomic_inc(&mdev->current_epoch->epoch_size);
1700 return drbd_drain_block(mdev, data_size);
1701 }
1702
1703 /* get_ldev(mdev) successful.
1704 * Corresponding put_ldev done either below (on various errors),
1705 * or in drbd_endio_write_sec, if we successfully submit the data at
1706 * the end of this function. */
1707
1708 sector = be64_to_cpu(p->sector);
1709 e = read_in_block(mdev, p->block_id, sector, data_size);
1710 if (!e) {
1711 put_ldev(mdev);
1712 return FALSE;
1713 }
1714
1715 e->private_bio->bi_end_io = drbd_endio_write_sec;
1716 e->w.cb = e_end_block;
1717
1718 spin_lock(&mdev->epoch_lock);
1719 e->epoch = mdev->current_epoch;
1720 atomic_inc(&e->epoch->epoch_size);
1721 atomic_inc(&e->epoch->active);
1722
1723 if (mdev->write_ordering == WO_bio_barrier && atomic_read(&e->epoch->epoch_size) == 1) {
1724 struct drbd_epoch *epoch;
1725 /* Issue a barrier if we start a new epoch, and the previous epoch
1726 was not a epoch containing a single request which already was
1727 a Barrier. */
1728 epoch = list_entry(e->epoch->list.prev, struct drbd_epoch, list);
1729 if (epoch == e->epoch) {
1730 set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags);
1731 rw |= (1<<BIO_RW_BARRIER);
1732 e->flags |= EE_IS_BARRIER;
1733 } else {
1734 if (atomic_read(&epoch->epoch_size) > 1 ||
1735 !test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) {
1736 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags);
1737 set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags);
1738 rw |= (1<<BIO_RW_BARRIER);
1739 e->flags |= EE_IS_BARRIER;
1740 }
1741 }
1742 }
1743 spin_unlock(&mdev->epoch_lock);
1744
1745 dp_flags = be32_to_cpu(p->dp_flags);
1746 if (dp_flags & DP_HARDBARRIER) {
1747 dev_err(DEV, "ASSERT FAILED would have submitted barrier request\n");
1748 /* rw |= (1<<BIO_RW_BARRIER); */
1749 }
1750 if (dp_flags & DP_RW_SYNC)
1751 rw |= (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
1752 if (dp_flags & DP_MAY_SET_IN_SYNC)
1753 e->flags |= EE_MAY_SET_IN_SYNC;
1754
1755 /* I'm the receiver, I do hold a net_cnt reference. */
1756 if (!mdev->net_conf->two_primaries) {
1757 spin_lock_irq(&mdev->req_lock);
1758 } else {
1759 /* don't get the req_lock yet,
1760 * we may sleep in drbd_wait_peer_seq */
1761 const int size = e->size;
1762 const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1763 DEFINE_WAIT(wait);
1764 struct drbd_request *i;
1765 struct hlist_node *n;
1766 struct hlist_head *slot;
1767 int first;
1768
1769 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1770 BUG_ON(mdev->ee_hash == NULL);
1771 BUG_ON(mdev->tl_hash == NULL);
1772
1773 /* conflict detection and handling:
1774 * 1. wait on the sequence number,
1775 * in case this data packet overtook ACK packets.
1776 * 2. check our hash tables for conflicting requests.
1777 * we only need to walk the tl_hash, since an ee can not
1778 * have a conflict with an other ee: on the submitting
1779 * node, the corresponding req had already been conflicting,
1780 * and a conflicting req is never sent.
1781 *
1782 * Note: for two_primaries, we are protocol C,
1783 * so there cannot be any request that is DONE
1784 * but still on the transfer log.
1785 *
1786 * unconditionally add to the ee_hash.
1787 *
1788 * if no conflicting request is found:
1789 * submit.
1790 *
1791 * if any conflicting request is found
1792 * that has not yet been acked,
1793 * AND I have the "discard concurrent writes" flag:
1794 * queue (via done_ee) the P_DISCARD_ACK; OUT.
1795 *
1796 * if any conflicting request is found:
1797 * block the receiver, waiting on misc_wait
1798 * until no more conflicting requests are there,
1799 * or we get interrupted (disconnect).
1800 *
1801 * we do not just write after local io completion of those
1802 * requests, but only after req is done completely, i.e.
1803 * we wait for the P_DISCARD_ACK to arrive!
1804 *
1805 * then proceed normally, i.e. submit.
1806 */
1807 if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num)))
1808 goto out_interrupted;
1809
1810 spin_lock_irq(&mdev->req_lock);
1811
1812 hlist_add_head(&e->colision, ee_hash_slot(mdev, sector));
1813
1814 #define OVERLAPS overlaps(i->sector, i->size, sector, size)
1815 slot = tl_hash_slot(mdev, sector);
1816 first = 1;
1817 for (;;) {
1818 int have_unacked = 0;
1819 int have_conflict = 0;
1820 prepare_to_wait(&mdev->misc_wait, &wait,
1821 TASK_INTERRUPTIBLE);
1822 hlist_for_each_entry(i, n, slot, colision) {
1823 if (OVERLAPS) {
1824 /* only ALERT on first iteration,
1825 * we may be woken up early... */
1826 if (first)
1827 dev_alert(DEV, "%s[%u] Concurrent local write detected!"
1828 " new: %llus +%u; pending: %llus +%u\n",
1829 current->comm, current->pid,
1830 (unsigned long long)sector, size,
1831 (unsigned long long)i->sector, i->size);
1832 if (i->rq_state & RQ_NET_PENDING)
1833 ++have_unacked;
1834 ++have_conflict;
1835 }
1836 }
1837 #undef OVERLAPS
1838 if (!have_conflict)
1839 break;
1840
1841 /* Discard Ack only for the _first_ iteration */
1842 if (first && discard && have_unacked) {
1843 dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n",
1844 (unsigned long long)sector);
1845 inc_unacked(mdev);
1846 e->w.cb = e_send_discard_ack;
1847 list_add_tail(&e->w.list, &mdev->done_ee);
1848
1849 spin_unlock_irq(&mdev->req_lock);
1850
1851 /* we could probably send that P_DISCARD_ACK ourselves,
1852 * but I don't like the receiver using the msock */
1853
1854 put_ldev(mdev);
1855 wake_asender(mdev);
1856 finish_wait(&mdev->misc_wait, &wait);
1857 return TRUE;
1858 }
1859
1860 if (signal_pending(current)) {
1861 hlist_del_init(&e->colision);
1862
1863 spin_unlock_irq(&mdev->req_lock);
1864
1865 finish_wait(&mdev->misc_wait, &wait);
1866 goto out_interrupted;
1867 }
1868
1869 spin_unlock_irq(&mdev->req_lock);
1870 if (first) {
1871 first = 0;
1872 dev_alert(DEV, "Concurrent write! [W AFTERWARDS] "
1873 "sec=%llus\n", (unsigned long long)sector);
1874 } else if (discard) {
1875 /* we had none on the first iteration.
1876 * there must be none now. */
1877 D_ASSERT(have_unacked == 0);
1878 }
1879 schedule();
1880 spin_lock_irq(&mdev->req_lock);
1881 }
1882 finish_wait(&mdev->misc_wait, &wait);
1883 }
1884
1885 list_add(&e->w.list, &mdev->active_ee);
1886 spin_unlock_irq(&mdev->req_lock);
1887
1888 switch (mdev->net_conf->wire_protocol) {
1889 case DRBD_PROT_C:
1890 inc_unacked(mdev);
1891 /* corresponding dec_unacked() in e_end_block()
1892 * respective _drbd_clear_done_ee */
1893 break;
1894 case DRBD_PROT_B:
1895 /* I really don't like it that the receiver thread
1896 * sends on the msock, but anyways */
1897 drbd_send_ack(mdev, P_RECV_ACK, e);
1898 break;
1899 case DRBD_PROT_A:
1900 /* nothing to do */
1901 break;
1902 }
1903
1904 if (mdev->state.pdsk == D_DISKLESS) {
1905 /* In case we have the only disk of the cluster, */
1906 drbd_set_out_of_sync(mdev, e->sector, e->size);
1907 e->flags |= EE_CALL_AL_COMPLETE_IO;
1908 drbd_al_begin_io(mdev, e->sector);
1909 }
1910
1911 e->private_bio->bi_rw = rw;
1912 drbd_generic_make_request(mdev, DRBD_FAULT_DT_WR, e->private_bio);
1913 /* accounting done in endio */
1914
1915 maybe_kick_lo(mdev);
1916 return TRUE;
1917
1918 out_interrupted:
1919 /* yes, the epoch_size now is imbalanced.
1920 * but we drop the connection anyways, so we don't have a chance to
1921 * receive a barrier... atomic_inc(&mdev->epoch_size); */
1922 put_ldev(mdev);
1923 drbd_free_ee(mdev, e);
1924 return FALSE;
1925 }
1926
1927 static int receive_DataRequest(struct drbd_conf *mdev, struct p_header *h)
1928 {
1929 sector_t sector;
1930 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
1931 struct drbd_epoch_entry *e;
1932 struct digest_info *di = NULL;
1933 int size, digest_size;
1934 unsigned int fault_type;
1935 struct p_block_req *p =
1936 (struct p_block_req *)h;
1937 const int brps = sizeof(*p)-sizeof(*h);
1938
1939 if (drbd_recv(mdev, h->payload, brps) != brps)
1940 return FALSE;
1941
1942 sector = be64_to_cpu(p->sector);
1943 size = be32_to_cpu(p->blksize);
1944
1945 if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) {
1946 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
1947 (unsigned long long)sector, size);
1948 return FALSE;
1949 }
1950 if (sector + (size>>9) > capacity) {
1951 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
1952 (unsigned long long)sector, size);
1953 return FALSE;
1954 }
1955
1956 if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) {
1957 if (__ratelimit(&drbd_ratelimit_state))
1958 dev_err(DEV, "Can not satisfy peer's read request, "
1959 "no local data.\n");
1960 drbd_send_ack_rp(mdev, h->command == P_DATA_REQUEST ? P_NEG_DREPLY :
1961 P_NEG_RS_DREPLY , p);
1962 return drbd_drain_block(mdev, h->length - brps);
1963 }
1964
1965 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
1966 * "criss-cross" setup, that might cause write-out on some other DRBD,
1967 * which in turn might block on the other node at this very place. */
1968 e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO);
1969 if (!e) {
1970 put_ldev(mdev);
1971 return FALSE;
1972 }
1973
1974 e->private_bio->bi_rw = READ;
1975 e->private_bio->bi_end_io = drbd_endio_read_sec;
1976
1977 switch (h->command) {
1978 case P_DATA_REQUEST:
1979 e->w.cb = w_e_end_data_req;
1980 fault_type = DRBD_FAULT_DT_RD;
1981 break;
1982 case P_RS_DATA_REQUEST:
1983 e->w.cb = w_e_end_rsdata_req;
1984 fault_type = DRBD_FAULT_RS_RD;
1985 /* Eventually this should become asynchronously. Currently it
1986 * blocks the whole receiver just to delay the reading of a
1987 * resync data block.
1988 * the drbd_work_queue mechanism is made for this...
1989 */
1990 if (!drbd_rs_begin_io(mdev, sector)) {
1991 /* we have been interrupted,
1992 * probably connection lost! */
1993 D_ASSERT(signal_pending(current));
1994 goto out_free_e;
1995 }
1996 break;
1997
1998 case P_OV_REPLY:
1999 case P_CSUM_RS_REQUEST:
2000 fault_type = DRBD_FAULT_RS_RD;
2001 digest_size = h->length - brps ;
2002 di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO);
2003 if (!di)
2004 goto out_free_e;
2005
2006 di->digest_size = digest_size;
2007 di->digest = (((char *)di)+sizeof(struct digest_info));
2008
2009 if (drbd_recv(mdev, di->digest, digest_size) != digest_size)
2010 goto out_free_e;
2011
2012 e->block_id = (u64)(unsigned long)di;
2013 if (h->command == P_CSUM_RS_REQUEST) {
2014 D_ASSERT(mdev->agreed_pro_version >= 89);
2015 e->w.cb = w_e_end_csum_rs_req;
2016 } else if (h->command == P_OV_REPLY) {
2017 e->w.cb = w_e_end_ov_reply;
2018 dec_rs_pending(mdev);
2019 break;
2020 }
2021
2022 if (!drbd_rs_begin_io(mdev, sector)) {
2023 /* we have been interrupted, probably connection lost! */
2024 D_ASSERT(signal_pending(current));
2025 goto out_free_e;
2026 }
2027 break;
2028
2029 case P_OV_REQUEST:
2030 if (mdev->state.conn >= C_CONNECTED &&
2031 mdev->state.conn != C_VERIFY_T)
2032 dev_warn(DEV, "ASSERT FAILED: got P_OV_REQUEST while being %s\n",
2033 drbd_conn_str(mdev->state.conn));
2034 if (mdev->ov_start_sector == ~(sector_t)0 &&
2035 mdev->agreed_pro_version >= 90) {
2036 mdev->ov_start_sector = sector;
2037 mdev->ov_position = sector;
2038 mdev->ov_left = mdev->rs_total - BM_SECT_TO_BIT(sector);
2039 dev_info(DEV, "Online Verify start sector: %llu\n",
2040 (unsigned long long)sector);
2041 }
2042 e->w.cb = w_e_end_ov_req;
2043 fault_type = DRBD_FAULT_RS_RD;
2044 /* Eventually this should become asynchronous. Currently it
2045 * blocks the whole receiver just to delay the reading of a
2046 * resync data block.
2047 * the drbd_work_queue mechanism is made for this...
2048 */
2049 if (!drbd_rs_begin_io(mdev, sector)) {
2050 /* we have been interrupted,
2051 * probably connection lost! */
2052 D_ASSERT(signal_pending(current));
2053 goto out_free_e;
2054 }
2055 break;
2056
2057
2058 default:
2059 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n",
2060 cmdname(h->command));
2061 fault_type = DRBD_FAULT_MAX;
2062 }
2063
2064 spin_lock_irq(&mdev->req_lock);
2065 list_add(&e->w.list, &mdev->read_ee);
2066 spin_unlock_irq(&mdev->req_lock);
2067
2068 inc_unacked(mdev);
2069
2070 drbd_generic_make_request(mdev, fault_type, e->private_bio);
2071 maybe_kick_lo(mdev);
2072
2073 return TRUE;
2074
2075 out_free_e:
2076 kfree(di);
2077 put_ldev(mdev);
2078 drbd_free_ee(mdev, e);
2079 return FALSE;
2080 }
2081
2082 static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local)
2083 {
2084 int self, peer, rv = -100;
2085 unsigned long ch_self, ch_peer;
2086
2087 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2088 peer = mdev->p_uuid[UI_BITMAP] & 1;
2089
2090 ch_peer = mdev->p_uuid[UI_SIZE];
2091 ch_self = mdev->comm_bm_set;
2092
2093 switch (mdev->net_conf->after_sb_0p) {
2094 case ASB_CONSENSUS:
2095 case ASB_DISCARD_SECONDARY:
2096 case ASB_CALL_HELPER:
2097 dev_err(DEV, "Configuration error.\n");
2098 break;
2099 case ASB_DISCONNECT:
2100 break;
2101 case ASB_DISCARD_YOUNGER_PRI:
2102 if (self == 0 && peer == 1) {
2103 rv = -1;
2104 break;
2105 }
2106 if (self == 1 && peer == 0) {
2107 rv = 1;
2108 break;
2109 }
2110 /* Else fall through to one of the other strategies... */
2111 case ASB_DISCARD_OLDER_PRI:
2112 if (self == 0 && peer == 1) {
2113 rv = 1;
2114 break;
2115 }
2116 if (self == 1 && peer == 0) {
2117 rv = -1;
2118 break;
2119 }
2120 /* Else fall through to one of the other strategies... */
2121 dev_warn(DEV, "Discard younger/older primary did not find a decision\n"
2122 "Using discard-least-changes instead\n");
2123 case ASB_DISCARD_ZERO_CHG:
2124 if (ch_peer == 0 && ch_self == 0) {
2125 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2126 ? -1 : 1;
2127 break;
2128 } else {
2129 if (ch_peer == 0) { rv = 1; break; }
2130 if (ch_self == 0) { rv = -1; break; }
2131 }
2132 if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG)
2133 break;
2134 case ASB_DISCARD_LEAST_CHG:
2135 if (ch_self < ch_peer)
2136 rv = -1;
2137 else if (ch_self > ch_peer)
2138 rv = 1;
2139 else /* ( ch_self == ch_peer ) */
2140 /* Well, then use something else. */
2141 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2142 ? -1 : 1;
2143 break;
2144 case ASB_DISCARD_LOCAL:
2145 rv = -1;
2146 break;
2147 case ASB_DISCARD_REMOTE:
2148 rv = 1;
2149 }
2150
2151 return rv;
2152 }
2153
2154 static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local)
2155 {
2156 int self, peer, hg, rv = -100;
2157
2158 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2159 peer = mdev->p_uuid[UI_BITMAP] & 1;
2160
2161 switch (mdev->net_conf->after_sb_1p) {
2162 case ASB_DISCARD_YOUNGER_PRI:
2163 case ASB_DISCARD_OLDER_PRI:
2164 case ASB_DISCARD_LEAST_CHG:
2165 case ASB_DISCARD_LOCAL:
2166 case ASB_DISCARD_REMOTE:
2167 dev_err(DEV, "Configuration error.\n");
2168 break;
2169 case ASB_DISCONNECT:
2170 break;
2171 case ASB_CONSENSUS:
2172 hg = drbd_asb_recover_0p(mdev);
2173 if (hg == -1 && mdev->state.role == R_SECONDARY)
2174 rv = hg;
2175 if (hg == 1 && mdev->state.role == R_PRIMARY)
2176 rv = hg;
2177 break;
2178 case ASB_VIOLENTLY:
2179 rv = drbd_asb_recover_0p(mdev);
2180 break;
2181 case ASB_DISCARD_SECONDARY:
2182 return mdev->state.role == R_PRIMARY ? 1 : -1;
2183 case ASB_CALL_HELPER:
2184 hg = drbd_asb_recover_0p(mdev);
2185 if (hg == -1 && mdev->state.role == R_PRIMARY) {
2186 self = drbd_set_role(mdev, R_SECONDARY, 0);
2187 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2188 * we might be here in C_WF_REPORT_PARAMS which is transient.
2189 * we do not need to wait for the after state change work either. */
2190 self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2191 if (self != SS_SUCCESS) {
2192 drbd_khelper(mdev, "pri-lost-after-sb");
2193 } else {
2194 dev_warn(DEV, "Successfully gave up primary role.\n");
2195 rv = hg;
2196 }
2197 } else
2198 rv = hg;
2199 }
2200
2201 return rv;
2202 }
2203
2204 static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local)
2205 {
2206 int self, peer, hg, rv = -100;
2207
2208 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2209 peer = mdev->p_uuid[UI_BITMAP] & 1;
2210
2211 switch (mdev->net_conf->after_sb_2p) {
2212 case ASB_DISCARD_YOUNGER_PRI:
2213 case ASB_DISCARD_OLDER_PRI:
2214 case ASB_DISCARD_LEAST_CHG:
2215 case ASB_DISCARD_LOCAL:
2216 case ASB_DISCARD_REMOTE:
2217 case ASB_CONSENSUS:
2218 case ASB_DISCARD_SECONDARY:
2219 dev_err(DEV, "Configuration error.\n");
2220 break;
2221 case ASB_VIOLENTLY:
2222 rv = drbd_asb_recover_0p(mdev);
2223 break;
2224 case ASB_DISCONNECT:
2225 break;
2226 case ASB_CALL_HELPER:
2227 hg = drbd_asb_recover_0p(mdev);
2228 if (hg == -1) {
2229 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2230 * we might be here in C_WF_REPORT_PARAMS which is transient.
2231 * we do not need to wait for the after state change work either. */
2232 self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2233 if (self != SS_SUCCESS) {
2234 drbd_khelper(mdev, "pri-lost-after-sb");
2235 } else {
2236 dev_warn(DEV, "Successfully gave up primary role.\n");
2237 rv = hg;
2238 }
2239 } else
2240 rv = hg;
2241 }
2242
2243 return rv;
2244 }
2245
2246 static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid,
2247 u64 bits, u64 flags)
2248 {
2249 if (!uuid) {
2250 dev_info(DEV, "%s uuid info vanished while I was looking!\n", text);
2251 return;
2252 }
2253 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n",
2254 text,
2255 (unsigned long long)uuid[UI_CURRENT],
2256 (unsigned long long)uuid[UI_BITMAP],
2257 (unsigned long long)uuid[UI_HISTORY_START],
2258 (unsigned long long)uuid[UI_HISTORY_END],
2259 (unsigned long long)bits,
2260 (unsigned long long)flags);
2261 }
2262
2263 /*
2264 100 after split brain try auto recover
2265 2 C_SYNC_SOURCE set BitMap
2266 1 C_SYNC_SOURCE use BitMap
2267 0 no Sync
2268 -1 C_SYNC_TARGET use BitMap
2269 -2 C_SYNC_TARGET set BitMap
2270 -100 after split brain, disconnect
2271 -1000 unrelated data
2272 */
2273 static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local)
2274 {
2275 u64 self, peer;
2276 int i, j;
2277
2278 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2279 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2280
2281 *rule_nr = 10;
2282 if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED)
2283 return 0;
2284
2285 *rule_nr = 20;
2286 if ((self == UUID_JUST_CREATED || self == (u64)0) &&
2287 peer != UUID_JUST_CREATED)
2288 return -2;
2289
2290 *rule_nr = 30;
2291 if (self != UUID_JUST_CREATED &&
2292 (peer == UUID_JUST_CREATED || peer == (u64)0))
2293 return 2;
2294
2295 if (self == peer) {
2296 int rct, dc; /* roles at crash time */
2297
2298 if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) {
2299
2300 if (mdev->agreed_pro_version < 91)
2301 return -1001;
2302
2303 if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) &&
2304 (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) {
2305 dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n");
2306 drbd_uuid_set_bm(mdev, 0UL);
2307
2308 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2309 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2310 *rule_nr = 34;
2311 } else {
2312 dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n");
2313 *rule_nr = 36;
2314 }
2315
2316 return 1;
2317 }
2318
2319 if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) {
2320
2321 if (mdev->agreed_pro_version < 91)
2322 return -1001;
2323
2324 if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) &&
2325 (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) {
2326 dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n");
2327
2328 mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START];
2329 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP];
2330 mdev->p_uuid[UI_BITMAP] = 0UL;
2331
2332 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2333 *rule_nr = 35;
2334 } else {
2335 dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n");
2336 *rule_nr = 37;
2337 }
2338
2339 return -1;
2340 }
2341
2342 /* Common power [off|failure] */
2343 rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) +
2344 (mdev->p_uuid[UI_FLAGS] & 2);
2345 /* lowest bit is set when we were primary,
2346 * next bit (weight 2) is set when peer was primary */
2347 *rule_nr = 40;
2348
2349 switch (rct) {
2350 case 0: /* !self_pri && !peer_pri */ return 0;
2351 case 1: /* self_pri && !peer_pri */ return 1;
2352 case 2: /* !self_pri && peer_pri */ return -1;
2353 case 3: /* self_pri && peer_pri */
2354 dc = test_bit(DISCARD_CONCURRENT, &mdev->flags);
2355 return dc ? -1 : 1;
2356 }
2357 }
2358
2359 *rule_nr = 50;
2360 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2361 if (self == peer)
2362 return -1;
2363
2364 *rule_nr = 51;
2365 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2366 if (self == peer) {
2367 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2368 peer = mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1);
2369 if (self == peer) {
2370 /* The last P_SYNC_UUID did not get though. Undo the last start of
2371 resync as sync source modifications of the peer's UUIDs. */
2372
2373 if (mdev->agreed_pro_version < 91)
2374 return -1001;
2375
2376 mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START];
2377 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1];
2378 return -1;
2379 }
2380 }
2381
2382 *rule_nr = 60;
2383 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2384 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2385 peer = mdev->p_uuid[i] & ~((u64)1);
2386 if (self == peer)
2387 return -2;
2388 }
2389
2390 *rule_nr = 70;
2391 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2392 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2393 if (self == peer)
2394 return 1;
2395
2396 *rule_nr = 71;
2397 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2398 if (self == peer) {
2399 self = mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1);
2400 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2401 if (self == peer) {
2402 /* The last P_SYNC_UUID did not get though. Undo the last start of
2403 resync as sync source modifications of our UUIDs. */
2404
2405 if (mdev->agreed_pro_version < 91)
2406 return -1001;
2407
2408 _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]);
2409 _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]);
2410
2411 dev_info(DEV, "Undid last start of resync:\n");
2412
2413 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2414 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2415
2416 return 1;
2417 }
2418 }
2419
2420
2421 *rule_nr = 80;
2422 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2423 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2424 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2425 if (self == peer)
2426 return 2;
2427 }
2428
2429 *rule_nr = 90;
2430 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2431 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2432 if (self == peer && self != ((u64)0))
2433 return 100;
2434
2435 *rule_nr = 100;
2436 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2437 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2438 for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) {
2439 peer = mdev->p_uuid[j] & ~((u64)1);
2440 if (self == peer)
2441 return -100;
2442 }
2443 }
2444
2445 return -1000;
2446 }
2447
2448 /* drbd_sync_handshake() returns the new conn state on success, or
2449 CONN_MASK (-1) on failure.
2450 */
2451 static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role,
2452 enum drbd_disk_state peer_disk) __must_hold(local)
2453 {
2454 int hg, rule_nr;
2455 enum drbd_conns rv = C_MASK;
2456 enum drbd_disk_state mydisk;
2457
2458 mydisk = mdev->state.disk;
2459 if (mydisk == D_NEGOTIATING)
2460 mydisk = mdev->new_state_tmp.disk;
2461
2462 dev_info(DEV, "drbd_sync_handshake:\n");
2463 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0);
2464 drbd_uuid_dump(mdev, "peer", mdev->p_uuid,
2465 mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2466
2467 hg = drbd_uuid_compare(mdev, &rule_nr);
2468
2469 dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr);
2470
2471 if (hg == -1000) {
2472 dev_alert(DEV, "Unrelated data, aborting!\n");
2473 return C_MASK;
2474 }
2475 if (hg == -1001) {
2476 dev_alert(DEV, "To resolve this both sides have to support at least protocol\n");
2477 return C_MASK;
2478 }
2479
2480 if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) ||
2481 (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) {
2482 int f = (hg == -100) || abs(hg) == 2;
2483 hg = mydisk > D_INCONSISTENT ? 1 : -1;
2484 if (f)
2485 hg = hg*2;
2486 dev_info(DEV, "Becoming sync %s due to disk states.\n",
2487 hg > 0 ? "source" : "target");
2488 }
2489
2490 if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) {
2491 int pcount = (mdev->state.role == R_PRIMARY)
2492 + (peer_role == R_PRIMARY);
2493 int forced = (hg == -100);
2494
2495 switch (pcount) {
2496 case 0:
2497 hg = drbd_asb_recover_0p(mdev);
2498 break;
2499 case 1:
2500 hg = drbd_asb_recover_1p(mdev);
2501 break;
2502 case 2:
2503 hg = drbd_asb_recover_2p(mdev);
2504 break;
2505 }
2506 if (abs(hg) < 100) {
2507 dev_warn(DEV, "Split-Brain detected, %d primaries, "
2508 "automatically solved. Sync from %s node\n",
2509 pcount, (hg < 0) ? "peer" : "this");
2510 if (forced) {
2511 dev_warn(DEV, "Doing a full sync, since"
2512 " UUIDs where ambiguous.\n");
2513 hg = hg*2;
2514 }
2515 }
2516 }
2517
2518 if (hg == -100) {
2519 if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1))
2520 hg = -1;
2521 if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1))
2522 hg = 1;
2523
2524 if (abs(hg) < 100)
2525 dev_warn(DEV, "Split-Brain detected, manually solved. "
2526 "Sync from %s node\n",
2527 (hg < 0) ? "peer" : "this");
2528 }
2529
2530 if (hg == -100) {
2531 /* FIXME this log message is not correct if we end up here
2532 * after an attempted attach on a diskless node.
2533 * We just refuse to attach -- well, we drop the "connection"
2534 * to that disk, in a way... */
2535 dev_alert(DEV, "Split-Brain detected, dropping connection!\n");
2536 drbd_khelper(mdev, "split-brain");
2537 return C_MASK;
2538 }
2539
2540 if (hg > 0 && mydisk <= D_INCONSISTENT) {
2541 dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n");
2542 return C_MASK;
2543 }
2544
2545 if (hg < 0 && /* by intention we do not use mydisk here. */
2546 mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) {
2547 switch (mdev->net_conf->rr_conflict) {
2548 case ASB_CALL_HELPER:
2549 drbd_khelper(mdev, "pri-lost");
2550 /* fall through */
2551 case ASB_DISCONNECT:
2552 dev_err(DEV, "I shall become SyncTarget, but I am primary!\n");
2553 return C_MASK;
2554 case ASB_VIOLENTLY:
2555 dev_warn(DEV, "Becoming SyncTarget, violating the stable-data"
2556 "assumption\n");
2557 }
2558 }
2559
2560 if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) {
2561 if (hg == 0)
2562 dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n");
2563 else
2564 dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.",
2565 drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET),
2566 abs(hg) >= 2 ? "full" : "bit-map based");
2567 return C_MASK;
2568 }
2569
2570 if (abs(hg) >= 2) {
2571 dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n");
2572 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake"))
2573 return C_MASK;
2574 }
2575
2576 if (hg > 0) { /* become sync source. */
2577 rv = C_WF_BITMAP_S;
2578 } else if (hg < 0) { /* become sync target */
2579 rv = C_WF_BITMAP_T;
2580 } else {
2581 rv = C_CONNECTED;
2582 if (drbd_bm_total_weight(mdev)) {
2583 dev_info(DEV, "No resync, but %lu bits in bitmap!\n",
2584 drbd_bm_total_weight(mdev));
2585 }
2586 }
2587
2588 return rv;
2589 }
2590
2591 /* returns 1 if invalid */
2592 static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self)
2593 {
2594 /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */
2595 if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) ||
2596 (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL))
2597 return 0;
2598
2599 /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */
2600 if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL ||
2601 self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL)
2602 return 1;
2603
2604 /* everything else is valid if they are equal on both sides. */
2605 if (peer == self)
2606 return 0;
2607
2608 /* everything es is invalid. */
2609 return 1;
2610 }
2611
2612 static int receive_protocol(struct drbd_conf *mdev, struct p_header *h)
2613 {
2614 struct p_protocol *p = (struct p_protocol *)h;
2615 int header_size, data_size;
2616 int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p;
2617 int p_want_lose, p_two_primaries, cf;
2618 char p_integrity_alg[SHARED_SECRET_MAX] = "";
2619
2620 header_size = sizeof(*p) - sizeof(*h);
2621 data_size = h->length - header_size;
2622
2623 if (drbd_recv(mdev, h->payload, header_size) != header_size)
2624 return FALSE;
2625
2626 p_proto = be32_to_cpu(p->protocol);
2627 p_after_sb_0p = be32_to_cpu(p->after_sb_0p);
2628 p_after_sb_1p = be32_to_cpu(p->after_sb_1p);
2629 p_after_sb_2p = be32_to_cpu(p->after_sb_2p);
2630 p_two_primaries = be32_to_cpu(p->two_primaries);
2631 cf = be32_to_cpu(p->conn_flags);
2632 p_want_lose = cf & CF_WANT_LOSE;
2633
2634 clear_bit(CONN_DRY_RUN, &mdev->flags);
2635
2636 if (cf & CF_DRY_RUN)
2637 set_bit(CONN_DRY_RUN, &mdev->flags);
2638
2639 if (p_proto != mdev->net_conf->wire_protocol) {
2640 dev_err(DEV, "incompatible communication protocols\n");
2641 goto disconnect;
2642 }
2643
2644 if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) {
2645 dev_err(DEV, "incompatible after-sb-0pri settings\n");
2646 goto disconnect;
2647 }
2648
2649 if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) {
2650 dev_err(DEV, "incompatible after-sb-1pri settings\n");
2651 goto disconnect;
2652 }
2653
2654 if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) {
2655 dev_err(DEV, "incompatible after-sb-2pri settings\n");
2656 goto disconnect;
2657 }
2658
2659 if (p_want_lose && mdev->net_conf->want_lose) {
2660 dev_err(DEV, "both sides have the 'want_lose' flag set\n");
2661 goto disconnect;
2662 }
2663
2664 if (p_two_primaries != mdev->net_conf->two_primaries) {
2665 dev_err(DEV, "incompatible setting of the two-primaries options\n");
2666 goto disconnect;
2667 }
2668
2669 if (mdev->agreed_pro_version >= 87) {
2670 unsigned char *my_alg = mdev->net_conf->integrity_alg;
2671
2672 if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size)
2673 return FALSE;
2674
2675 p_integrity_alg[SHARED_SECRET_MAX-1] = 0;
2676 if (strcmp(p_integrity_alg, my_alg)) {
2677 dev_err(DEV, "incompatible setting of the data-integrity-alg\n");
2678 goto disconnect;
2679 }
2680 dev_info(DEV, "data-integrity-alg: %s\n",
2681 my_alg[0] ? my_alg : (unsigned char *)"<not-used>");
2682 }
2683
2684 return TRUE;
2685
2686 disconnect:
2687 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2688 return FALSE;
2689 }
2690
2691 /* helper function
2692 * input: alg name, feature name
2693 * return: NULL (alg name was "")
2694 * ERR_PTR(error) if something goes wrong
2695 * or the crypto hash ptr, if it worked out ok. */
2696 struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev,
2697 const char *alg, const char *name)
2698 {
2699 struct crypto_hash *tfm;
2700
2701 if (!alg[0])
2702 return NULL;
2703
2704 tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC);
2705 if (IS_ERR(tfm)) {
2706 dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n",
2707 alg, name, PTR_ERR(tfm));
2708 return tfm;
2709 }
2710 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
2711 crypto_free_hash(tfm);
2712 dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name);
2713 return ERR_PTR(-EINVAL);
2714 }
2715 return tfm;
2716 }
2717
2718 static int receive_SyncParam(struct drbd_conf *mdev, struct p_header *h)
2719 {
2720 int ok = TRUE;
2721 struct p_rs_param_89 *p = (struct p_rs_param_89 *)h;
2722 unsigned int header_size, data_size, exp_max_sz;
2723 struct crypto_hash *verify_tfm = NULL;
2724 struct crypto_hash *csums_tfm = NULL;
2725 const int apv = mdev->agreed_pro_version;
2726
2727 exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param)
2728 : apv == 88 ? sizeof(struct p_rs_param)
2729 + SHARED_SECRET_MAX
2730 : /* 89 */ sizeof(struct p_rs_param_89);
2731
2732 if (h->length > exp_max_sz) {
2733 dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n",
2734 h->length, exp_max_sz);
2735 return FALSE;
2736 }
2737
2738 if (apv <= 88) {
2739 header_size = sizeof(struct p_rs_param) - sizeof(*h);
2740 data_size = h->length - header_size;
2741 } else /* apv >= 89 */ {
2742 header_size = sizeof(struct p_rs_param_89) - sizeof(*h);
2743 data_size = h->length - header_size;
2744 D_ASSERT(data_size == 0);
2745 }
2746
2747 /* initialize verify_alg and csums_alg */
2748 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
2749
2750 if (drbd_recv(mdev, h->payload, header_size) != header_size)
2751 return FALSE;
2752
2753 mdev->sync_conf.rate = be32_to_cpu(p->rate);
2754
2755 if (apv >= 88) {
2756 if (apv == 88) {
2757 if (data_size > SHARED_SECRET_MAX) {
2758 dev_err(DEV, "verify-alg too long, "
2759 "peer wants %u, accepting only %u byte\n",
2760 data_size, SHARED_SECRET_MAX);
2761 return FALSE;
2762 }
2763
2764 if (drbd_recv(mdev, p->verify_alg, data_size) != data_size)
2765 return FALSE;
2766
2767 /* we expect NUL terminated string */
2768 /* but just in case someone tries to be evil */
2769 D_ASSERT(p->verify_alg[data_size-1] == 0);
2770 p->verify_alg[data_size-1] = 0;
2771
2772 } else /* apv >= 89 */ {
2773 /* we still expect NUL terminated strings */
2774 /* but just in case someone tries to be evil */
2775 D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0);
2776 D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0);
2777 p->verify_alg[SHARED_SECRET_MAX-1] = 0;
2778 p->csums_alg[SHARED_SECRET_MAX-1] = 0;
2779 }
2780
2781 if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) {
2782 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2783 dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n",
2784 mdev->sync_conf.verify_alg, p->verify_alg);
2785 goto disconnect;
2786 }
2787 verify_tfm = drbd_crypto_alloc_digest_safe(mdev,
2788 p->verify_alg, "verify-alg");
2789 if (IS_ERR(verify_tfm)) {
2790 verify_tfm = NULL;
2791 goto disconnect;
2792 }
2793 }
2794
2795 if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) {
2796 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2797 dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n",
2798 mdev->sync_conf.csums_alg, p->csums_alg);
2799 goto disconnect;
2800 }
2801 csums_tfm = drbd_crypto_alloc_digest_safe(mdev,
2802 p->csums_alg, "csums-alg");
2803 if (IS_ERR(csums_tfm)) {
2804 csums_tfm = NULL;
2805 goto disconnect;
2806 }
2807 }
2808
2809
2810 spin_lock(&mdev->peer_seq_lock);
2811 /* lock against drbd_nl_syncer_conf() */
2812 if (verify_tfm) {
2813 strcpy(mdev->sync_conf.verify_alg, p->verify_alg);
2814 mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1;
2815 crypto_free_hash(mdev->verify_tfm);
2816 mdev->verify_tfm = verify_tfm;
2817 dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg);
2818 }
2819 if (csums_tfm) {
2820 strcpy(mdev->sync_conf.csums_alg, p->csums_alg);
2821 mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1;
2822 crypto_free_hash(mdev->csums_tfm);
2823 mdev->csums_tfm = csums_tfm;
2824 dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg);
2825 }
2826 spin_unlock(&mdev->peer_seq_lock);
2827 }
2828
2829 return ok;
2830 disconnect:
2831 /* just for completeness: actually not needed,
2832 * as this is not reached if csums_tfm was ok. */
2833 crypto_free_hash(csums_tfm);
2834 /* but free the verify_tfm again, if csums_tfm did not work out */
2835 crypto_free_hash(verify_tfm);
2836 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2837 return FALSE;
2838 }
2839
2840 static void drbd_setup_order_type(struct drbd_conf *mdev, int peer)
2841 {
2842 /* sorry, we currently have no working implementation
2843 * of distributed TCQ */
2844 }
2845
2846 /* warn if the arguments differ by more than 12.5% */
2847 static void warn_if_differ_considerably(struct drbd_conf *mdev,
2848 const char *s, sector_t a, sector_t b)
2849 {
2850 sector_t d;
2851 if (a == 0 || b == 0)
2852 return;
2853 d = (a > b) ? (a - b) : (b - a);
2854 if (d > (a>>3) || d > (b>>3))
2855 dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s,
2856 (unsigned long long)a, (unsigned long long)b);
2857 }
2858
2859 static int receive_sizes(struct drbd_conf *mdev, struct p_header *h)
2860 {
2861 struct p_sizes *p = (struct p_sizes *)h;
2862 enum determine_dev_size dd = unchanged;
2863 unsigned int max_seg_s;
2864 sector_t p_size, p_usize, my_usize;
2865 int ldsc = 0; /* local disk size changed */
2866
2867 ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE;
2868 if (drbd_recv(mdev, h->payload, h->length) != h->length)
2869 return FALSE;
2870
2871 p_size = be64_to_cpu(p->d_size);
2872 p_usize = be64_to_cpu(p->u_size);
2873
2874 if (p_size == 0 && mdev->state.disk == D_DISKLESS) {
2875 dev_err(DEV, "some backing storage is needed\n");
2876 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2877 return FALSE;
2878 }
2879
2880 /* just store the peer's disk size for now.
2881 * we still need to figure out whether we accept that. */
2882 mdev->p_size = p_size;
2883
2884 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
2885 if (get_ldev(mdev)) {
2886 warn_if_differ_considerably(mdev, "lower level device sizes",
2887 p_size, drbd_get_max_capacity(mdev->ldev));
2888 warn_if_differ_considerably(mdev, "user requested size",
2889 p_usize, mdev->ldev->dc.disk_size);
2890
2891 /* if this is the first connect, or an otherwise expected
2892 * param exchange, choose the minimum */
2893 if (mdev->state.conn == C_WF_REPORT_PARAMS)
2894 p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size,
2895 p_usize);
2896
2897 my_usize = mdev->ldev->dc.disk_size;
2898
2899 if (mdev->ldev->dc.disk_size != p_usize) {
2900 mdev->ldev->dc.disk_size = p_usize;
2901 dev_info(DEV, "Peer sets u_size to %lu sectors\n",
2902 (unsigned long)mdev->ldev->dc.disk_size);
2903 }
2904
2905 /* Never shrink a device with usable data during connect.
2906 But allow online shrinking if we are connected. */
2907 if (drbd_new_dev_size(mdev, mdev->ldev, 0) <
2908 drbd_get_capacity(mdev->this_bdev) &&
2909 mdev->state.disk >= D_OUTDATED &&
2910 mdev->state.conn < C_CONNECTED) {
2911 dev_err(DEV, "The peer's disk size is too small!\n");
2912 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2913 mdev->ldev->dc.disk_size = my_usize;
2914 put_ldev(mdev);
2915 return FALSE;
2916 }
2917 put_ldev(mdev);
2918 }
2919 #undef min_not_zero
2920
2921 if (get_ldev(mdev)) {
2922 dd = drbd_determin_dev_size(mdev, 0);
2923 put_ldev(mdev);
2924 if (dd == dev_size_error)
2925 return FALSE;
2926 drbd_md_sync(mdev);
2927 } else {
2928 /* I am diskless, need to accept the peer's size. */
2929 drbd_set_my_capacity(mdev, p_size);
2930 }
2931
2932 if (get_ldev(mdev)) {
2933 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) {
2934 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
2935 ldsc = 1;
2936 }
2937
2938 max_seg_s = be32_to_cpu(p->max_segment_size);
2939 if (max_seg_s != queue_max_segment_size(mdev->rq_queue))
2940 drbd_setup_queue_param(mdev, max_seg_s);
2941
2942 drbd_setup_order_type(mdev, be32_to_cpu(p->queue_order_type));
2943 put_ldev(mdev);
2944 }
2945
2946 if (mdev->state.conn > C_WF_REPORT_PARAMS) {
2947 if (be64_to_cpu(p->c_size) !=
2948 drbd_get_capacity(mdev->this_bdev) || ldsc) {
2949 /* we have different sizes, probably peer
2950 * needs to know my new size... */
2951 drbd_send_sizes(mdev, 0);
2952 }
2953 if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) ||
2954 (dd == grew && mdev->state.conn == C_CONNECTED)) {
2955 if (mdev->state.pdsk >= D_INCONSISTENT &&
2956 mdev->state.disk >= D_INCONSISTENT)
2957 resync_after_online_grow(mdev);
2958 else
2959 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
2960 }
2961 }
2962
2963 return TRUE;
2964 }
2965
2966 static int receive_uuids(struct drbd_conf *mdev, struct p_header *h)
2967 {
2968 struct p_uuids *p = (struct p_uuids *)h;
2969 u64 *p_uuid;
2970 int i;
2971
2972 ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE;
2973 if (drbd_recv(mdev, h->payload, h->length) != h->length)
2974 return FALSE;
2975
2976 p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO);
2977
2978 for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++)
2979 p_uuid[i] = be64_to_cpu(p->uuid[i]);
2980
2981 kfree(mdev->p_uuid);
2982 mdev->p_uuid = p_uuid;
2983
2984 if (mdev->state.conn < C_CONNECTED &&
2985 mdev->state.disk < D_INCONSISTENT &&
2986 mdev->state.role == R_PRIMARY &&
2987 (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) {
2988 dev_err(DEV, "Can only connect to data with current UUID=%016llX\n",
2989 (unsigned long long)mdev->ed_uuid);
2990 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2991 return FALSE;
2992 }
2993
2994 if (get_ldev(mdev)) {
2995 int skip_initial_sync =
2996 mdev->state.conn == C_CONNECTED &&
2997 mdev->agreed_pro_version >= 90 &&
2998 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
2999 (p_uuid[UI_FLAGS] & 8);
3000 if (skip_initial_sync) {
3001 dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n");
3002 drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
3003 "clear_n_write from receive_uuids");
3004 _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]);
3005 _drbd_uuid_set(mdev, UI_BITMAP, 0);
3006 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3007 CS_VERBOSE, NULL);
3008 drbd_md_sync(mdev);
3009 }
3010 put_ldev(mdev);
3011 }
3012
3013 /* Before we test for the disk state, we should wait until an eventually
3014 ongoing cluster wide state change is finished. That is important if
3015 we are primary and are detaching from our disk. We need to see the
3016 new disk state... */
3017 wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags));
3018 if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT)
3019 drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3020
3021 return TRUE;
3022 }
3023
3024 /**
3025 * convert_state() - Converts the peer's view of the cluster state to our point of view
3026 * @ps: The state as seen by the peer.
3027 */
3028 static union drbd_state convert_state(union drbd_state ps)
3029 {
3030 union drbd_state ms;
3031
3032 static enum drbd_conns c_tab[] = {
3033 [C_CONNECTED] = C_CONNECTED,
3034
3035 [C_STARTING_SYNC_S] = C_STARTING_SYNC_T,
3036 [C_STARTING_SYNC_T] = C_STARTING_SYNC_S,
3037 [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */
3038 [C_VERIFY_S] = C_VERIFY_T,
3039 [C_MASK] = C_MASK,
3040 };
3041
3042 ms.i = ps.i;
3043
3044 ms.conn = c_tab[ps.conn];
3045 ms.peer = ps.role;
3046 ms.role = ps.peer;
3047 ms.pdsk = ps.disk;
3048 ms.disk = ps.pdsk;
3049 ms.peer_isp = (ps.aftr_isp | ps.user_isp);
3050
3051 return ms;
3052 }
3053
3054 static int receive_req_state(struct drbd_conf *mdev, struct p_header *h)
3055 {
3056 struct p_req_state *p = (struct p_req_state *)h;
3057 union drbd_state mask, val;
3058 int rv;
3059
3060 ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE;
3061 if (drbd_recv(mdev, h->payload, h->length) != h->length)
3062 return FALSE;
3063
3064 mask.i = be32_to_cpu(p->mask);
3065 val.i = be32_to_cpu(p->val);
3066
3067 if (test_bit(DISCARD_CONCURRENT, &mdev->flags) &&
3068 test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) {
3069 drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG);
3070 return TRUE;
3071 }
3072
3073 mask = convert_state(mask);
3074 val = convert_state(val);
3075
3076 rv = drbd_change_state(mdev, CS_VERBOSE, mask, val);
3077
3078 drbd_send_sr_reply(mdev, rv);
3079 drbd_md_sync(mdev);
3080
3081 return TRUE;
3082 }
3083
3084 static int receive_state(struct drbd_conf *mdev, struct p_header *h)
3085 {
3086 struct p_state *p = (struct p_state *)h;
3087 enum drbd_conns nconn, oconn;
3088 union drbd_state ns, peer_state;
3089 enum drbd_disk_state real_peer_disk;
3090 int rv;
3091
3092 ERR_IF(h->length != (sizeof(*p)-sizeof(*h)))
3093 return FALSE;
3094
3095 if (drbd_recv(mdev, h->payload, h->length) != h->length)
3096 return FALSE;
3097
3098 peer_state.i = be32_to_cpu(p->state);
3099
3100 real_peer_disk = peer_state.disk;
3101 if (peer_state.disk == D_NEGOTIATING) {
3102 real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT;
3103 dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk));
3104 }
3105
3106 spin_lock_irq(&mdev->req_lock);
3107 retry:
3108 oconn = nconn = mdev->state.conn;
3109 spin_unlock_irq(&mdev->req_lock);
3110
3111 if (nconn == C_WF_REPORT_PARAMS)
3112 nconn = C_CONNECTED;
3113
3114 if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING &&
3115 get_ldev_if_state(mdev, D_NEGOTIATING)) {
3116 int cr; /* consider resync */
3117
3118 /* if we established a new connection */
3119 cr = (oconn < C_CONNECTED);
3120 /* if we had an established connection
3121 * and one of the nodes newly attaches a disk */
3122 cr |= (oconn == C_CONNECTED &&
3123 (peer_state.disk == D_NEGOTIATING ||
3124 mdev->state.disk == D_NEGOTIATING));
3125 /* if we have both been inconsistent, and the peer has been
3126 * forced to be UpToDate with --overwrite-data */
3127 cr |= test_bit(CONSIDER_RESYNC, &mdev->flags);
3128 /* if we had been plain connected, and the admin requested to
3129 * start a sync by "invalidate" or "invalidate-remote" */
3130 cr |= (oconn == C_CONNECTED &&
3131 (peer_state.conn >= C_STARTING_SYNC_S &&
3132 peer_state.conn <= C_WF_BITMAP_T));
3133
3134 if (cr)
3135 nconn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk);
3136
3137 put_ldev(mdev);
3138 if (nconn == C_MASK) {
3139 nconn = C_CONNECTED;
3140 if (mdev->state.disk == D_NEGOTIATING) {
3141 drbd_force_state(mdev, NS(disk, D_DISKLESS));
3142 } else if (peer_state.disk == D_NEGOTIATING) {
3143 dev_err(DEV, "Disk attach process on the peer node was aborted.\n");
3144 peer_state.disk = D_DISKLESS;
3145 real_peer_disk = D_DISKLESS;
3146 } else {
3147 if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags))
3148 return FALSE;
3149 D_ASSERT(oconn == C_WF_REPORT_PARAMS);
3150 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3151 return FALSE;
3152 }
3153 }
3154 }
3155
3156 spin_lock_irq(&mdev->req_lock);
3157 if (mdev->state.conn != oconn)
3158 goto retry;
3159 clear_bit(CONSIDER_RESYNC, &mdev->flags);
3160 ns.i = mdev->state.i;
3161 ns.conn = nconn;
3162 ns.peer = peer_state.role;
3163 ns.pdsk = real_peer_disk;
3164 ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp);
3165 if ((nconn == C_CONNECTED || nconn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING)
3166 ns.disk = mdev->new_state_tmp.disk;
3167
3168 rv = _drbd_set_state(mdev, ns, CS_VERBOSE | CS_HARD, NULL);
3169 ns = mdev->state;
3170 spin_unlock_irq(&mdev->req_lock);
3171
3172 if (rv < SS_SUCCESS) {
3173 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3174 return FALSE;
3175 }
3176
3177 if (oconn > C_WF_REPORT_PARAMS) {
3178 if (nconn > C_CONNECTED && peer_state.conn <= C_CONNECTED &&
3179 peer_state.disk != D_NEGOTIATING ) {
3180 /* we want resync, peer has not yet decided to sync... */
3181 /* Nowadays only used when forcing a node into primary role and
3182 setting its disk to UpToDate with that */
3183 drbd_send_uuids(mdev);
3184 drbd_send_state(mdev);
3185 }
3186 }
3187
3188 mdev->net_conf->want_lose = 0;
3189
3190 drbd_md_sync(mdev); /* update connected indicator, la_size, ... */
3191
3192 return TRUE;
3193 }
3194
3195 static int receive_sync_uuid(struct drbd_conf *mdev, struct p_header *h)
3196 {
3197 struct p_rs_uuid *p = (struct p_rs_uuid *)h;
3198
3199 wait_event(mdev->misc_wait,
3200 mdev->state.conn == C_WF_SYNC_UUID ||
3201 mdev->state.conn < C_CONNECTED ||
3202 mdev->state.disk < D_NEGOTIATING);
3203
3204 /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */
3205
3206 ERR_IF(h->length != (sizeof(*p)-sizeof(*h))) return FALSE;
3207 if (drbd_recv(mdev, h->payload, h->length) != h->length)
3208 return FALSE;
3209
3210 /* Here the _drbd_uuid_ functions are right, current should
3211 _not_ be rotated into the history */
3212 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
3213 _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid));
3214 _drbd_uuid_set(mdev, UI_BITMAP, 0UL);
3215
3216 drbd_start_resync(mdev, C_SYNC_TARGET);
3217
3218 put_ldev(mdev);
3219 } else
3220 dev_err(DEV, "Ignoring SyncUUID packet!\n");
3221
3222 return TRUE;
3223 }
3224
3225 enum receive_bitmap_ret { OK, DONE, FAILED };
3226
3227 static enum receive_bitmap_ret
3228 receive_bitmap_plain(struct drbd_conf *mdev, struct p_header *h,
3229 unsigned long *buffer, struct bm_xfer_ctx *c)
3230 {
3231 unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset);
3232 unsigned want = num_words * sizeof(long);
3233
3234 if (want != h->length) {
3235 dev_err(DEV, "%s:want (%u) != h->length (%u)\n", __func__, want, h->length);
3236 return FAILED;
3237 }
3238 if (want == 0)
3239 return DONE;
3240 if (drbd_recv(mdev, buffer, want) != want)
3241 return FAILED;
3242
3243 drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer);
3244
3245 c->word_offset += num_words;
3246 c->bit_offset = c->word_offset * BITS_PER_LONG;
3247 if (c->bit_offset > c->bm_bits)
3248 c->bit_offset = c->bm_bits;
3249
3250 return OK;
3251 }
3252
3253 static enum receive_bitmap_ret
3254 recv_bm_rle_bits(struct drbd_conf *mdev,
3255 struct p_compressed_bm *p,
3256 struct bm_xfer_ctx *c)
3257 {
3258 struct bitstream bs;
3259 u64 look_ahead;
3260 u64 rl;
3261 u64 tmp;
3262 unsigned long s = c->bit_offset;
3263 unsigned long e;
3264 int len = p->head.length - (sizeof(*p) - sizeof(p->head));
3265 int toggle = DCBP_get_start(p);
3266 int have;
3267 int bits;
3268
3269 bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p));
3270
3271 bits = bitstream_get_bits(&bs, &look_ahead, 64);
3272 if (bits < 0)
3273 return FAILED;
3274
3275 for (have = bits; have > 0; s += rl, toggle = !toggle) {
3276 bits = vli_decode_bits(&rl, look_ahead);
3277 if (bits <= 0)
3278 return FAILED;
3279
3280 if (toggle) {
3281 e = s + rl -1;
3282 if (e >= c->bm_bits) {
3283 dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e);
3284 return FAILED;
3285 }
3286 _drbd_bm_set_bits(mdev, s, e);
3287 }
3288
3289 if (have < bits) {
3290 dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n",
3291 have, bits, look_ahead,
3292 (unsigned int)(bs.cur.b - p->code),
3293 (unsigned int)bs.buf_len);
3294 return FAILED;
3295 }
3296 look_ahead >>= bits;
3297 have -= bits;
3298
3299 bits = bitstream_get_bits(&bs, &tmp, 64 - have);
3300 if (bits < 0)
3301 return FAILED;
3302 look_ahead |= tmp << have;
3303 have += bits;
3304 }
3305
3306 c->bit_offset = s;
3307 bm_xfer_ctx_bit_to_word_offset(c);
3308
3309 return (s == c->bm_bits) ? DONE : OK;
3310 }
3311
3312 static enum receive_bitmap_ret
3313 decode_bitmap_c(struct drbd_conf *mdev,
3314 struct p_compressed_bm *p,
3315 struct bm_xfer_ctx *c)
3316 {
3317 if (DCBP_get_code(p) == RLE_VLI_Bits)
3318 return recv_bm_rle_bits(mdev, p, c);
3319
3320 /* other variants had been implemented for evaluation,
3321 * but have been dropped as this one turned out to be "best"
3322 * during all our tests. */
3323
3324 dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding);
3325 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3326 return FAILED;
3327 }
3328
3329 void INFO_bm_xfer_stats(struct drbd_conf *mdev,
3330 const char *direction, struct bm_xfer_ctx *c)
3331 {
3332 /* what would it take to transfer it "plaintext" */
3333 unsigned plain = sizeof(struct p_header) *
3334 ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1)
3335 + c->bm_words * sizeof(long);
3336 unsigned total = c->bytes[0] + c->bytes[1];
3337 unsigned r;
3338
3339 /* total can not be zero. but just in case: */
3340 if (total == 0)
3341 return;
3342
3343 /* don't report if not compressed */
3344 if (total >= plain)
3345 return;
3346
3347 /* total < plain. check for overflow, still */
3348 r = (total > UINT_MAX/1000) ? (total / (plain/1000))
3349 : (1000 * total / plain);
3350
3351 if (r > 1000)
3352 r = 1000;
3353
3354 r = 1000 - r;
3355 dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), "
3356 "total %u; compression: %u.%u%%\n",
3357 direction,
3358 c->bytes[1], c->packets[1],
3359 c->bytes[0], c->packets[0],
3360 total, r/10, r % 10);
3361 }
3362
3363 /* Since we are processing the bitfield from lower addresses to higher,
3364 it does not matter if the process it in 32 bit chunks or 64 bit
3365 chunks as long as it is little endian. (Understand it as byte stream,
3366 beginning with the lowest byte...) If we would use big endian
3367 we would need to process it from the highest address to the lowest,
3368 in order to be agnostic to the 32 vs 64 bits issue.
3369
3370 returns 0 on failure, 1 if we successfully received it. */
3371 static int receive_bitmap(struct drbd_conf *mdev, struct p_header *h)
3372 {
3373 struct bm_xfer_ctx c;
3374 void *buffer;
3375 enum receive_bitmap_ret ret;
3376 int ok = FALSE;
3377
3378 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
3379
3380 drbd_bm_lock(mdev, "receive bitmap");
3381
3382 /* maybe we should use some per thread scratch page,
3383 * and allocate that during initial device creation? */
3384 buffer = (unsigned long *) __get_free_page(GFP_NOIO);
3385 if (!buffer) {
3386 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__);
3387 goto out;
3388 }
3389
3390 c = (struct bm_xfer_ctx) {
3391 .bm_bits = drbd_bm_bits(mdev),
3392 .bm_words = drbd_bm_words(mdev),
3393 };
3394
3395 do {
3396 if (h->command == P_BITMAP) {
3397 ret = receive_bitmap_plain(mdev, h, buffer, &c);
3398 } else if (h->command == P_COMPRESSED_BITMAP) {
3399 /* MAYBE: sanity check that we speak proto >= 90,
3400 * and the feature is enabled! */
3401 struct p_compressed_bm *p;
3402
3403 if (h->length > BM_PACKET_PAYLOAD_BYTES) {
3404 dev_err(DEV, "ReportCBitmap packet too large\n");
3405 goto out;
3406 }
3407 /* use the page buff */
3408 p = buffer;
3409 memcpy(p, h, sizeof(*h));
3410 if (drbd_recv(mdev, p->head.payload, h->length) != h->length)
3411 goto out;
3412 if (p->head.length <= (sizeof(*p) - sizeof(p->head))) {
3413 dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", p->head.length);
3414 return FAILED;
3415 }
3416 ret = decode_bitmap_c(mdev, p, &c);
3417 } else {
3418 dev_warn(DEV, "receive_bitmap: h->command neither ReportBitMap nor ReportCBitMap (is 0x%x)", h->command);
3419 goto out;
3420 }
3421
3422 c.packets[h->command == P_BITMAP]++;
3423 c.bytes[h->command == P_BITMAP] += sizeof(struct p_header) + h->length;
3424
3425 if (ret != OK)
3426 break;
3427
3428 if (!drbd_recv_header(mdev, h))
3429 goto out;
3430 } while (ret == OK);
3431 if (ret == FAILED)
3432 goto out;
3433
3434 INFO_bm_xfer_stats(mdev, "receive", &c);
3435
3436 if (mdev->state.conn == C_WF_BITMAP_T) {
3437 ok = !drbd_send_bitmap(mdev);
3438 if (!ok)
3439 goto out;
3440 /* Omit CS_ORDERED with this state transition to avoid deadlocks. */
3441 ok = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
3442 D_ASSERT(ok == SS_SUCCESS);
3443 } else if (mdev->state.conn != C_WF_BITMAP_S) {
3444 /* admin may have requested C_DISCONNECTING,
3445 * other threads may have noticed network errors */
3446 dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n",
3447 drbd_conn_str(mdev->state.conn));
3448 }
3449
3450 ok = TRUE;
3451 out:
3452 drbd_bm_unlock(mdev);
3453 if (ok && mdev->state.conn == C_WF_BITMAP_S)
3454 drbd_start_resync(mdev, C_SYNC_SOURCE);
3455 free_page((unsigned long) buffer);
3456 return ok;
3457 }
3458
3459 static int receive_skip(struct drbd_conf *mdev, struct p_header *h)
3460 {
3461 /* TODO zero copy sink :) */
3462 static char sink[128];
3463 int size, want, r;
3464
3465 dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n",
3466 h->command, h->length);
3467
3468 size = h->length;
3469 while (size > 0) {
3470 want = min_t(int, size, sizeof(sink));
3471 r = drbd_recv(mdev, sink, want);
3472 ERR_IF(r <= 0) break;
3473 size -= r;
3474 }
3475 return size == 0;
3476 }
3477
3478 static int receive_UnplugRemote(struct drbd_conf *mdev, struct p_header *h)
3479 {
3480 if (mdev->state.disk >= D_INCONSISTENT)
3481 drbd_kick_lo(mdev);
3482
3483 /* Make sure we've acked all the TCP data associated
3484 * with the data requests being unplugged */
3485 drbd_tcp_quickack(mdev->data.socket);
3486
3487 return TRUE;
3488 }
3489
3490 typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, struct p_header *);
3491
3492 static drbd_cmd_handler_f drbd_default_handler[] = {
3493 [P_DATA] = receive_Data,
3494 [P_DATA_REPLY] = receive_DataReply,
3495 [P_RS_DATA_REPLY] = receive_RSDataReply,
3496 [P_BARRIER] = receive_Barrier,
3497 [P_BITMAP] = receive_bitmap,
3498 [P_COMPRESSED_BITMAP] = receive_bitmap,
3499 [P_UNPLUG_REMOTE] = receive_UnplugRemote,
3500 [P_DATA_REQUEST] = receive_DataRequest,
3501 [P_RS_DATA_REQUEST] = receive_DataRequest,
3502 [P_SYNC_PARAM] = receive_SyncParam,
3503 [P_SYNC_PARAM89] = receive_SyncParam,
3504 [P_PROTOCOL] = receive_protocol,
3505 [P_UUIDS] = receive_uuids,
3506 [P_SIZES] = receive_sizes,
3507 [P_STATE] = receive_state,
3508 [P_STATE_CHG_REQ] = receive_req_state,
3509 [P_SYNC_UUID] = receive_sync_uuid,
3510 [P_OV_REQUEST] = receive_DataRequest,
3511 [P_OV_REPLY] = receive_DataRequest,
3512 [P_CSUM_RS_REQUEST] = receive_DataRequest,
3513 /* anything missing from this table is in
3514 * the asender_tbl, see get_asender_cmd */
3515 [P_MAX_CMD] = NULL,
3516 };
3517
3518 static drbd_cmd_handler_f *drbd_cmd_handler = drbd_default_handler;
3519 static drbd_cmd_handler_f *drbd_opt_cmd_handler;
3520
3521 static void drbdd(struct drbd_conf *mdev)
3522 {
3523 drbd_cmd_handler_f handler;
3524 struct p_header *header = &mdev->data.rbuf.header;
3525
3526 while (get_t_state(&mdev->receiver) == Running) {
3527 drbd_thread_current_set_cpu(mdev);
3528 if (!drbd_recv_header(mdev, header)) {
3529 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3530 break;
3531 }
3532
3533 if (header->command < P_MAX_CMD)
3534 handler = drbd_cmd_handler[header->command];
3535 else if (P_MAY_IGNORE < header->command
3536 && header->command < P_MAX_OPT_CMD)
3537 handler = drbd_opt_cmd_handler[header->command-P_MAY_IGNORE];
3538 else if (header->command > P_MAX_OPT_CMD)
3539 handler = receive_skip;
3540 else
3541 handler = NULL;
3542
3543 if (unlikely(!handler)) {
3544 dev_err(DEV, "unknown packet type %d, l: %d!\n",
3545 header->command, header->length);
3546 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3547 break;
3548 }
3549 if (unlikely(!handler(mdev, header))) {
3550 dev_err(DEV, "error receiving %s, l: %d!\n",
3551 cmdname(header->command), header->length);
3552 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3553 break;
3554 }
3555 }
3556 }
3557
3558 static void drbd_fail_pending_reads(struct drbd_conf *mdev)
3559 {
3560 struct hlist_head *slot;
3561 struct hlist_node *pos;
3562 struct hlist_node *tmp;
3563 struct drbd_request *req;
3564 int i;
3565
3566 /*
3567 * Application READ requests
3568 */
3569 spin_lock_irq(&mdev->req_lock);
3570 for (i = 0; i < APP_R_HSIZE; i++) {
3571 slot = mdev->app_reads_hash+i;
3572 hlist_for_each_entry_safe(req, pos, tmp, slot, colision) {
3573 /* it may (but should not any longer!)
3574 * be on the work queue; if that assert triggers,
3575 * we need to also grab the
3576 * spin_lock_irq(&mdev->data.work.q_lock);
3577 * and list_del_init here. */
3578 D_ASSERT(list_empty(&req->w.list));
3579 /* It would be nice to complete outside of spinlock.
3580 * But this is easier for now. */
3581 _req_mod(req, connection_lost_while_pending);
3582 }
3583 }
3584 for (i = 0; i < APP_R_HSIZE; i++)
3585 if (!hlist_empty(mdev->app_reads_hash+i))
3586 dev_warn(DEV, "ASSERT FAILED: app_reads_hash[%d].first: "
3587 "%p, should be NULL\n", i, mdev->app_reads_hash[i].first);
3588
3589 memset(mdev->app_reads_hash, 0, APP_R_HSIZE*sizeof(void *));
3590 spin_unlock_irq(&mdev->req_lock);
3591 }
3592
3593 void drbd_flush_workqueue(struct drbd_conf *mdev)
3594 {
3595 struct drbd_wq_barrier barr;
3596
3597 barr.w.cb = w_prev_work_done;
3598 init_completion(&barr.done);
3599 drbd_queue_work(&mdev->data.work, &barr.w);
3600 wait_for_completion(&barr.done);
3601 }
3602
3603 static void drbd_disconnect(struct drbd_conf *mdev)
3604 {
3605 enum drbd_fencing_p fp;
3606 union drbd_state os, ns;
3607 int rv = SS_UNKNOWN_ERROR;
3608 unsigned int i;
3609
3610 if (mdev->state.conn == C_STANDALONE)
3611 return;
3612 if (mdev->state.conn >= C_WF_CONNECTION)
3613 dev_err(DEV, "ASSERT FAILED cstate = %s, expected < WFConnection\n",
3614 drbd_conn_str(mdev->state.conn));
3615
3616 /* asender does not clean up anything. it must not interfere, either */
3617 drbd_thread_stop(&mdev->asender);
3618 drbd_free_sock(mdev);
3619
3620 spin_lock_irq(&mdev->req_lock);
3621 _drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
3622 _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee);
3623 _drbd_wait_ee_list_empty(mdev, &mdev->read_ee);
3624 spin_unlock_irq(&mdev->req_lock);
3625
3626 /* We do not have data structures that would allow us to
3627 * get the rs_pending_cnt down to 0 again.
3628 * * On C_SYNC_TARGET we do not have any data structures describing
3629 * the pending RSDataRequest's we have sent.
3630 * * On C_SYNC_SOURCE there is no data structure that tracks
3631 * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget.
3632 * And no, it is not the sum of the reference counts in the
3633 * resync_LRU. The resync_LRU tracks the whole operation including
3634 * the disk-IO, while the rs_pending_cnt only tracks the blocks
3635 * on the fly. */
3636 drbd_rs_cancel_all(mdev);
3637 mdev->rs_total = 0;
3638 mdev->rs_failed = 0;
3639 atomic_set(&mdev->rs_pending_cnt, 0);
3640 wake_up(&mdev->misc_wait);
3641
3642 /* make sure syncer is stopped and w_resume_next_sg queued */
3643 del_timer_sync(&mdev->resync_timer);
3644 set_bit(STOP_SYNC_TIMER, &mdev->flags);
3645 resync_timer_fn((unsigned long)mdev);
3646
3647 /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier,
3648 * w_make_resync_request etc. which may still be on the worker queue
3649 * to be "canceled" */
3650 drbd_flush_workqueue(mdev);
3651
3652 /* This also does reclaim_net_ee(). If we do this too early, we might
3653 * miss some resync ee and pages.*/
3654 drbd_process_done_ee(mdev);
3655
3656 kfree(mdev->p_uuid);
3657 mdev->p_uuid = NULL;
3658
3659 if (!mdev->state.susp)
3660 tl_clear(mdev);
3661
3662 drbd_fail_pending_reads(mdev);
3663
3664 dev_info(DEV, "Connection closed\n");
3665
3666 drbd_md_sync(mdev);
3667
3668 fp = FP_DONT_CARE;
3669 if (get_ldev(mdev)) {
3670 fp = mdev->ldev->dc.fencing;
3671 put_ldev(mdev);
3672 }
3673
3674 if (mdev->state.role == R_PRIMARY) {
3675 if (fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) {
3676 enum drbd_disk_state nps = drbd_try_outdate_peer(mdev);
3677 drbd_request_state(mdev, NS(pdsk, nps));
3678 }
3679 }
3680
3681 spin_lock_irq(&mdev->req_lock);
3682 os = mdev->state;
3683 if (os.conn >= C_UNCONNECTED) {
3684 /* Do not restart in case we are C_DISCONNECTING */
3685 ns = os;
3686 ns.conn = C_UNCONNECTED;
3687 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
3688 }
3689 spin_unlock_irq(&mdev->req_lock);
3690
3691 if (os.conn == C_DISCONNECTING) {
3692 struct hlist_head *h;
3693 wait_event(mdev->misc_wait, atomic_read(&mdev->net_cnt) == 0);
3694
3695 /* we must not free the tl_hash
3696 * while application io is still on the fly */
3697 wait_event(mdev->misc_wait, atomic_read(&mdev->ap_bio_cnt) == 0);
3698
3699 spin_lock_irq(&mdev->req_lock);
3700 /* paranoia code */
3701 for (h = mdev->ee_hash; h < mdev->ee_hash + mdev->ee_hash_s; h++)
3702 if (h->first)
3703 dev_err(DEV, "ASSERT FAILED ee_hash[%u].first == %p, expected NULL\n",
3704 (int)(h - mdev->ee_hash), h->first);
3705 kfree(mdev->ee_hash);
3706 mdev->ee_hash = NULL;
3707 mdev->ee_hash_s = 0;
3708
3709 /* paranoia code */
3710 for (h = mdev->tl_hash; h < mdev->tl_hash + mdev->tl_hash_s; h++)
3711 if (h->first)
3712 dev_err(DEV, "ASSERT FAILED tl_hash[%u] == %p, expected NULL\n",
3713 (int)(h - mdev->tl_hash), h->first);
3714 kfree(mdev->tl_hash);
3715 mdev->tl_hash = NULL;
3716 mdev->tl_hash_s = 0;
3717 spin_unlock_irq(&mdev->req_lock);
3718
3719 crypto_free_hash(mdev->cram_hmac_tfm);
3720 mdev->cram_hmac_tfm = NULL;
3721
3722 kfree(mdev->net_conf);
3723 mdev->net_conf = NULL;
3724 drbd_request_state(mdev, NS(conn, C_STANDALONE));
3725 }
3726
3727 /* tcp_close and release of sendpage pages can be deferred. I don't
3728 * want to use SO_LINGER, because apparently it can be deferred for
3729 * more than 20 seconds (longest time I checked).
3730 *
3731 * Actually we don't care for exactly when the network stack does its
3732 * put_page(), but release our reference on these pages right here.
3733 */
3734 i = drbd_release_ee(mdev, &mdev->net_ee);
3735 if (i)
3736 dev_info(DEV, "net_ee not empty, killed %u entries\n", i);
3737 i = atomic_read(&mdev->pp_in_use);
3738 if (i)
3739 dev_info(DEV, "pp_in_use = %u, expected 0\n", i);
3740
3741 D_ASSERT(list_empty(&mdev->read_ee));
3742 D_ASSERT(list_empty(&mdev->active_ee));
3743 D_ASSERT(list_empty(&mdev->sync_ee));
3744 D_ASSERT(list_empty(&mdev->done_ee));
3745
3746 /* ok, no more ee's on the fly, it is safe to reset the epoch_size */
3747 atomic_set(&mdev->current_epoch->epoch_size, 0);
3748 D_ASSERT(list_empty(&mdev->current_epoch->list));
3749 }
3750
3751 /*
3752 * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version
3753 * we can agree on is stored in agreed_pro_version.
3754 *
3755 * feature flags and the reserved array should be enough room for future
3756 * enhancements of the handshake protocol, and possible plugins...
3757 *
3758 * for now, they are expected to be zero, but ignored.
3759 */
3760 static int drbd_send_handshake(struct drbd_conf *mdev)
3761 {
3762 /* ASSERT current == mdev->receiver ... */
3763 struct p_handshake *p = &mdev->data.sbuf.handshake;
3764 int ok;
3765
3766 if (mutex_lock_interruptible(&mdev->data.mutex)) {
3767 dev_err(DEV, "interrupted during initial handshake\n");
3768 return 0; /* interrupted. not ok. */
3769 }
3770
3771 if (mdev->data.socket == NULL) {
3772 mutex_unlock(&mdev->data.mutex);
3773 return 0;
3774 }
3775
3776 memset(p, 0, sizeof(*p));
3777 p->protocol_min = cpu_to_be32(PRO_VERSION_MIN);
3778 p->protocol_max = cpu_to_be32(PRO_VERSION_MAX);
3779 ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE,
3780 (struct p_header *)p, sizeof(*p), 0 );
3781 mutex_unlock(&mdev->data.mutex);
3782 return ok;
3783 }
3784
3785 /*
3786 * return values:
3787 * 1 yes, we have a valid connection
3788 * 0 oops, did not work out, please try again
3789 * -1 peer talks different language,
3790 * no point in trying again, please go standalone.
3791 */
3792 static int drbd_do_handshake(struct drbd_conf *mdev)
3793 {
3794 /* ASSERT current == mdev->receiver ... */
3795 struct p_handshake *p = &mdev->data.rbuf.handshake;
3796 const int expect = sizeof(struct p_handshake)
3797 -sizeof(struct p_header);
3798 int rv;
3799
3800 rv = drbd_send_handshake(mdev);
3801 if (!rv)
3802 return 0;
3803
3804 rv = drbd_recv_header(mdev, &p->head);
3805 if (!rv)
3806 return 0;
3807
3808 if (p->head.command != P_HAND_SHAKE) {
3809 dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n",
3810 cmdname(p->head.command), p->head.command);
3811 return -1;
3812 }
3813
3814 if (p->head.length != expect) {
3815 dev_err(DEV, "expected HandShake length: %u, received: %u\n",
3816 expect, p->head.length);
3817 return -1;
3818 }
3819
3820 rv = drbd_recv(mdev, &p->head.payload, expect);
3821
3822 if (rv != expect) {
3823 dev_err(DEV, "short read receiving handshake packet: l=%u\n", rv);
3824 return 0;
3825 }
3826
3827 p->protocol_min = be32_to_cpu(p->protocol_min);
3828 p->protocol_max = be32_to_cpu(p->protocol_max);
3829 if (p->protocol_max == 0)
3830 p->protocol_max = p->protocol_min;
3831
3832 if (PRO_VERSION_MAX < p->protocol_min ||
3833 PRO_VERSION_MIN > p->protocol_max)
3834 goto incompat;
3835
3836 mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max);
3837
3838 dev_info(DEV, "Handshake successful: "
3839 "Agreed network protocol version %d\n", mdev->agreed_pro_version);
3840
3841 return 1;
3842
3843 incompat:
3844 dev_err(DEV, "incompatible DRBD dialects: "
3845 "I support %d-%d, peer supports %d-%d\n",
3846 PRO_VERSION_MIN, PRO_VERSION_MAX,
3847 p->protocol_min, p->protocol_max);
3848 return -1;
3849 }
3850
3851 #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE)
3852 static int drbd_do_auth(struct drbd_conf *mdev)
3853 {
3854 dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n");
3855 dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n");
3856 return -1;
3857 }
3858 #else
3859 #define CHALLENGE_LEN 64
3860
3861 /* Return value:
3862 1 - auth succeeded,
3863 0 - failed, try again (network error),
3864 -1 - auth failed, don't try again.
3865 */
3866
3867 static int drbd_do_auth(struct drbd_conf *mdev)
3868 {
3869 char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */
3870 struct scatterlist sg;
3871 char *response = NULL;
3872 char *right_response = NULL;
3873 char *peers_ch = NULL;
3874 struct p_header p;
3875 unsigned int key_len = strlen(mdev->net_conf->shared_secret);
3876 unsigned int resp_size;
3877 struct hash_desc desc;
3878 int rv;
3879
3880 desc.tfm = mdev->cram_hmac_tfm;
3881 desc.flags = 0;
3882
3883 rv = crypto_hash_setkey(mdev->cram_hmac_tfm,
3884 (u8 *)mdev->net_conf->shared_secret, key_len);
3885 if (rv) {
3886 dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv);
3887 rv = -1;
3888 goto fail;
3889 }
3890
3891 get_random_bytes(my_challenge, CHALLENGE_LEN);
3892
3893 rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN);
3894 if (!rv)
3895 goto fail;
3896
3897 rv = drbd_recv_header(mdev, &p);
3898 if (!rv)
3899 goto fail;
3900
3901 if (p.command != P_AUTH_CHALLENGE) {
3902 dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n",
3903 cmdname(p.command), p.command);
3904 rv = 0;
3905 goto fail;
3906 }
3907
3908 if (p.length > CHALLENGE_LEN*2) {
3909 dev_err(DEV, "expected AuthChallenge payload too big.\n");
3910 rv = -1;
3911 goto fail;
3912 }
3913
3914 peers_ch = kmalloc(p.length, GFP_NOIO);
3915 if (peers_ch == NULL) {
3916 dev_err(DEV, "kmalloc of peers_ch failed\n");
3917 rv = -1;
3918 goto fail;
3919 }
3920
3921 rv = drbd_recv(mdev, peers_ch, p.length);
3922
3923 if (rv != p.length) {
3924 dev_err(DEV, "short read AuthChallenge: l=%u\n", rv);
3925 rv = 0;
3926 goto fail;
3927 }
3928
3929 resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm);
3930 response = kmalloc(resp_size, GFP_NOIO);
3931 if (response == NULL) {
3932 dev_err(DEV, "kmalloc of response failed\n");
3933 rv = -1;
3934 goto fail;
3935 }
3936
3937 sg_init_table(&sg, 1);
3938 sg_set_buf(&sg, peers_ch, p.length);
3939
3940 rv = crypto_hash_digest(&desc, &sg, sg.length, response);
3941 if (rv) {
3942 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
3943 rv = -1;
3944 goto fail;
3945 }
3946
3947 rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size);
3948 if (!rv)
3949 goto fail;
3950
3951 rv = drbd_recv_header(mdev, &p);
3952 if (!rv)
3953 goto fail;
3954
3955 if (p.command != P_AUTH_RESPONSE) {
3956 dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n",
3957 cmdname(p.command), p.command);
3958 rv = 0;
3959 goto fail;
3960 }
3961
3962 if (p.length != resp_size) {
3963 dev_err(DEV, "expected AuthResponse payload of wrong size\n");
3964 rv = 0;
3965 goto fail;
3966 }
3967
3968 rv = drbd_recv(mdev, response , resp_size);
3969
3970 if (rv != resp_size) {
3971 dev_err(DEV, "short read receiving AuthResponse: l=%u\n", rv);
3972 rv = 0;
3973 goto fail;
3974 }
3975
3976 right_response = kmalloc(resp_size, GFP_NOIO);
3977 if (right_response == NULL) {
3978 dev_err(DEV, "kmalloc of right_response failed\n");
3979 rv = -1;
3980 goto fail;
3981 }
3982
3983 sg_set_buf(&sg, my_challenge, CHALLENGE_LEN);
3984
3985 rv = crypto_hash_digest(&desc, &sg, sg.length, right_response);
3986 if (rv) {
3987 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
3988 rv = -1;
3989 goto fail;
3990 }
3991
3992 rv = !memcmp(response, right_response, resp_size);
3993
3994 if (rv)
3995 dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n",
3996 resp_size, mdev->net_conf->cram_hmac_alg);
3997 else
3998 rv = -1;
3999
4000 fail:
4001 kfree(peers_ch);
4002 kfree(response);
4003 kfree(right_response);
4004
4005 return rv;
4006 }
4007 #endif
4008
4009 int drbdd_init(struct drbd_thread *thi)
4010 {
4011 struct drbd_conf *mdev = thi->mdev;
4012 unsigned int minor = mdev_to_minor(mdev);
4013 int h;
4014
4015 sprintf(current->comm, "drbd%d_receiver", minor);
4016
4017 dev_info(DEV, "receiver (re)started\n");
4018
4019 do {
4020 h = drbd_connect(mdev);
4021 if (h == 0) {
4022 drbd_disconnect(mdev);
4023 __set_current_state(TASK_INTERRUPTIBLE);
4024 schedule_timeout(HZ);
4025 }
4026 if (h == -1) {
4027 dev_warn(DEV, "Discarding network configuration.\n");
4028 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4029 }
4030 } while (h == 0);
4031
4032 if (h > 0) {
4033 if (get_net_conf(mdev)) {
4034 drbdd(mdev);
4035 put_net_conf(mdev);
4036 }
4037 }
4038
4039 drbd_disconnect(mdev);
4040
4041 dev_info(DEV, "receiver terminated\n");
4042 return 0;
4043 }
4044
4045 /* ********* acknowledge sender ******** */
4046
4047 static int got_RqSReply(struct drbd_conf *mdev, struct p_header *h)
4048 {
4049 struct p_req_state_reply *p = (struct p_req_state_reply *)h;
4050
4051 int retcode = be32_to_cpu(p->retcode);
4052
4053 if (retcode >= SS_SUCCESS) {
4054 set_bit(CL_ST_CHG_SUCCESS, &mdev->flags);
4055 } else {
4056 set_bit(CL_ST_CHG_FAIL, &mdev->flags);
4057 dev_err(DEV, "Requested state change failed by peer: %s (%d)\n",
4058 drbd_set_st_err_str(retcode), retcode);
4059 }
4060 wake_up(&mdev->state_wait);
4061
4062 return TRUE;
4063 }
4064
4065 static int got_Ping(struct drbd_conf *mdev, struct p_header *h)
4066 {
4067 return drbd_send_ping_ack(mdev);
4068
4069 }
4070
4071 static int got_PingAck(struct drbd_conf *mdev, struct p_header *h)
4072 {
4073 /* restore idle timeout */
4074 mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
4075 if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags))
4076 wake_up(&mdev->misc_wait);
4077
4078 return TRUE;
4079 }
4080
4081 static int got_IsInSync(struct drbd_conf *mdev, struct p_header *h)
4082 {
4083 struct p_block_ack *p = (struct p_block_ack *)h;
4084 sector_t sector = be64_to_cpu(p->sector);
4085 int blksize = be32_to_cpu(p->blksize);
4086
4087 D_ASSERT(mdev->agreed_pro_version >= 89);
4088
4089 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4090
4091 drbd_rs_complete_io(mdev, sector);
4092 drbd_set_in_sync(mdev, sector, blksize);
4093 /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */
4094 mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT);
4095 dec_rs_pending(mdev);
4096
4097 return TRUE;
4098 }
4099
4100 /* when we receive the ACK for a write request,
4101 * verify that we actually know about it */
4102 static struct drbd_request *_ack_id_to_req(struct drbd_conf *mdev,
4103 u64 id, sector_t sector)
4104 {
4105 struct hlist_head *slot = tl_hash_slot(mdev, sector);
4106 struct hlist_node *n;
4107 struct drbd_request *req;
4108
4109 hlist_for_each_entry(req, n, slot, colision) {
4110 if ((unsigned long)req == (unsigned long)id) {
4111 if (req->sector != sector) {
4112 dev_err(DEV, "_ack_id_to_req: found req %p but it has "
4113 "wrong sector (%llus versus %llus)\n", req,
4114 (unsigned long long)req->sector,
4115 (unsigned long long)sector);
4116 break;
4117 }
4118 return req;
4119 }
4120 }
4121 dev_err(DEV, "_ack_id_to_req: failed to find req %p, sector %llus in list\n",
4122 (void *)(unsigned long)id, (unsigned long long)sector);
4123 return NULL;
4124 }
4125
4126 typedef struct drbd_request *(req_validator_fn)
4127 (struct drbd_conf *mdev, u64 id, sector_t sector);
4128
4129 static int validate_req_change_req_state(struct drbd_conf *mdev,
4130 u64 id, sector_t sector, req_validator_fn validator,
4131 const char *func, enum drbd_req_event what)
4132 {
4133 struct drbd_request *req;
4134 struct bio_and_error m;
4135
4136 spin_lock_irq(&mdev->req_lock);
4137 req = validator(mdev, id, sector);
4138 if (unlikely(!req)) {
4139 spin_unlock_irq(&mdev->req_lock);
4140 dev_err(DEV, "%s: got a corrupt block_id/sector pair\n", func);
4141 return FALSE;
4142 }
4143 __req_mod(req, what, &m);
4144 spin_unlock_irq(&mdev->req_lock);
4145
4146 if (m.bio)
4147 complete_master_bio(mdev, &m);
4148 return TRUE;
4149 }
4150
4151 static int got_BlockAck(struct drbd_conf *mdev, struct p_header *h)
4152 {
4153 struct p_block_ack *p = (struct p_block_ack *)h;
4154 sector_t sector = be64_to_cpu(p->sector);
4155 int blksize = be32_to_cpu(p->blksize);
4156 enum drbd_req_event what;
4157
4158 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4159
4160 if (is_syncer_block_id(p->block_id)) {
4161 drbd_set_in_sync(mdev, sector, blksize);
4162 dec_rs_pending(mdev);
4163 return TRUE;
4164 }
4165 switch (be16_to_cpu(h->command)) {
4166 case P_RS_WRITE_ACK:
4167 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4168 what = write_acked_by_peer_and_sis;
4169 break;
4170 case P_WRITE_ACK:
4171 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4172 what = write_acked_by_peer;
4173 break;
4174 case P_RECV_ACK:
4175 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B);
4176 what = recv_acked_by_peer;
4177 break;
4178 case P_DISCARD_ACK:
4179 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4180 what = conflict_discarded_by_peer;
4181 break;
4182 default:
4183 D_ASSERT(0);
4184 return FALSE;
4185 }
4186
4187 return validate_req_change_req_state(mdev, p->block_id, sector,
4188 _ack_id_to_req, __func__ , what);
4189 }
4190
4191 static int got_NegAck(struct drbd_conf *mdev, struct p_header *h)
4192 {
4193 struct p_block_ack *p = (struct p_block_ack *)h;
4194 sector_t sector = be64_to_cpu(p->sector);
4195
4196 if (__ratelimit(&drbd_ratelimit_state))
4197 dev_warn(DEV, "Got NegAck packet. Peer is in troubles?\n");
4198
4199 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4200
4201 if (is_syncer_block_id(p->block_id)) {
4202 int size = be32_to_cpu(p->blksize);
4203 dec_rs_pending(mdev);
4204 drbd_rs_failed_io(mdev, sector, size);
4205 return TRUE;
4206 }
4207 return validate_req_change_req_state(mdev, p->block_id, sector,
4208 _ack_id_to_req, __func__ , neg_acked);
4209 }
4210
4211 static int got_NegDReply(struct drbd_conf *mdev, struct p_header *h)
4212 {
4213 struct p_block_ack *p = (struct p_block_ack *)h;
4214 sector_t sector = be64_to_cpu(p->sector);
4215
4216 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4217 dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n",
4218 (unsigned long long)sector, be32_to_cpu(p->blksize));
4219
4220 return validate_req_change_req_state(mdev, p->block_id, sector,
4221 _ar_id_to_req, __func__ , neg_acked);
4222 }
4223
4224 static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header *h)
4225 {
4226 sector_t sector;
4227 int size;
4228 struct p_block_ack *p = (struct p_block_ack *)h;
4229
4230 sector = be64_to_cpu(p->sector);
4231 size = be32_to_cpu(p->blksize);
4232
4233 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4234
4235 dec_rs_pending(mdev);
4236
4237 if (get_ldev_if_state(mdev, D_FAILED)) {
4238 drbd_rs_complete_io(mdev, sector);
4239 drbd_rs_failed_io(mdev, sector, size);
4240 put_ldev(mdev);
4241 }
4242
4243 return TRUE;
4244 }
4245
4246 static int got_BarrierAck(struct drbd_conf *mdev, struct p_header *h)
4247 {
4248 struct p_barrier_ack *p = (struct p_barrier_ack *)h;
4249
4250 tl_release(mdev, p->barrier, be32_to_cpu(p->set_size));
4251
4252 return TRUE;
4253 }
4254
4255 static int got_OVResult(struct drbd_conf *mdev, struct p_header *h)
4256 {
4257 struct p_block_ack *p = (struct p_block_ack *)h;
4258 struct drbd_work *w;
4259 sector_t sector;
4260 int size;
4261
4262 sector = be64_to_cpu(p->sector);
4263 size = be32_to_cpu(p->blksize);
4264
4265 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4266
4267 if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC)
4268 drbd_ov_oos_found(mdev, sector, size);
4269 else
4270 ov_oos_print(mdev);
4271
4272 drbd_rs_complete_io(mdev, sector);
4273 dec_rs_pending(mdev);
4274
4275 if (--mdev->ov_left == 0) {
4276 w = kmalloc(sizeof(*w), GFP_NOIO);
4277 if (w) {
4278 w->cb = w_ov_finished;
4279 drbd_queue_work_front(&mdev->data.work, w);
4280 } else {
4281 dev_err(DEV, "kmalloc(w) failed.");
4282 ov_oos_print(mdev);
4283 drbd_resync_finished(mdev);
4284 }
4285 }
4286 return TRUE;
4287 }
4288
4289 struct asender_cmd {
4290 size_t pkt_size;
4291 int (*process)(struct drbd_conf *mdev, struct p_header *h);
4292 };
4293
4294 static struct asender_cmd *get_asender_cmd(int cmd)
4295 {
4296 static struct asender_cmd asender_tbl[] = {
4297 /* anything missing from this table is in
4298 * the drbd_cmd_handler (drbd_default_handler) table,
4299 * see the beginning of drbdd() */
4300 [P_PING] = { sizeof(struct p_header), got_Ping },
4301 [P_PING_ACK] = { sizeof(struct p_header), got_PingAck },
4302 [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4303 [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4304 [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4305 [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4306 [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck },
4307 [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply },
4308 [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply},
4309 [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult },
4310 [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck },
4311 [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply },
4312 [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync },
4313 [P_MAX_CMD] = { 0, NULL },
4314 };
4315 if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL)
4316 return NULL;
4317 return &asender_tbl[cmd];
4318 }
4319
4320 int drbd_asender(struct drbd_thread *thi)
4321 {
4322 struct drbd_conf *mdev = thi->mdev;
4323 struct p_header *h = &mdev->meta.rbuf.header;
4324 struct asender_cmd *cmd = NULL;
4325
4326 int rv, len;
4327 void *buf = h;
4328 int received = 0;
4329 int expect = sizeof(struct p_header);
4330 int empty;
4331
4332 sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev));
4333
4334 current->policy = SCHED_RR; /* Make this a realtime task! */
4335 current->rt_priority = 2; /* more important than all other tasks */
4336
4337 while (get_t_state(thi) == Running) {
4338 drbd_thread_current_set_cpu(mdev);
4339 if (test_and_clear_bit(SEND_PING, &mdev->flags)) {
4340 ERR_IF(!drbd_send_ping(mdev)) goto reconnect;
4341 mdev->meta.socket->sk->sk_rcvtimeo =
4342 mdev->net_conf->ping_timeo*HZ/10;
4343 }
4344
4345 /* conditionally cork;
4346 * it may hurt latency if we cork without much to send */
4347 if (!mdev->net_conf->no_cork &&
4348 3 < atomic_read(&mdev->unacked_cnt))
4349 drbd_tcp_cork(mdev->meta.socket);
4350 while (1) {
4351 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4352 flush_signals(current);
4353 if (!drbd_process_done_ee(mdev)) {
4354 dev_err(DEV, "process_done_ee() = NOT_OK\n");
4355 goto reconnect;
4356 }
4357 /* to avoid race with newly queued ACKs */
4358 set_bit(SIGNAL_ASENDER, &mdev->flags);
4359 spin_lock_irq(&mdev->req_lock);
4360 empty = list_empty(&mdev->done_ee);
4361 spin_unlock_irq(&mdev->req_lock);
4362 /* new ack may have been queued right here,
4363 * but then there is also a signal pending,
4364 * and we start over... */
4365 if (empty)
4366 break;
4367 }
4368 /* but unconditionally uncork unless disabled */
4369 if (!mdev->net_conf->no_cork)
4370 drbd_tcp_uncork(mdev->meta.socket);
4371
4372 /* short circuit, recv_msg would return EINTR anyways. */
4373 if (signal_pending(current))
4374 continue;
4375
4376 rv = drbd_recv_short(mdev, mdev->meta.socket,
4377 buf, expect-received, 0);
4378 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4379
4380 flush_signals(current);
4381
4382 /* Note:
4383 * -EINTR (on meta) we got a signal
4384 * -EAGAIN (on meta) rcvtimeo expired
4385 * -ECONNRESET other side closed the connection
4386 * -ERESTARTSYS (on data) we got a signal
4387 * rv < 0 other than above: unexpected error!
4388 * rv == expected: full header or command
4389 * rv < expected: "woken" by signal during receive
4390 * rv == 0 : "connection shut down by peer"
4391 */
4392 if (likely(rv > 0)) {
4393 received += rv;
4394 buf += rv;
4395 } else if (rv == 0) {
4396 dev_err(DEV, "meta connection shut down by peer.\n");
4397 goto reconnect;
4398 } else if (rv == -EAGAIN) {
4399 if (mdev->meta.socket->sk->sk_rcvtimeo ==
4400 mdev->net_conf->ping_timeo*HZ/10) {
4401 dev_err(DEV, "PingAck did not arrive in time.\n");
4402 goto reconnect;
4403 }
4404 set_bit(SEND_PING, &mdev->flags);
4405 continue;
4406 } else if (rv == -EINTR) {
4407 continue;
4408 } else {
4409 dev_err(DEV, "sock_recvmsg returned %d\n", rv);
4410 goto reconnect;
4411 }
4412
4413 if (received == expect && cmd == NULL) {
4414 if (unlikely(h->magic != BE_DRBD_MAGIC)) {
4415 dev_err(DEV, "magic?? on meta m: 0x%lx c: %d l: %d\n",
4416 (long)be32_to_cpu(h->magic),
4417 h->command, h->length);
4418 goto reconnect;
4419 }
4420 cmd = get_asender_cmd(be16_to_cpu(h->command));
4421 len = be16_to_cpu(h->length);
4422 if (unlikely(cmd == NULL)) {
4423 dev_err(DEV, "unknown command?? on meta m: 0x%lx c: %d l: %d\n",
4424 (long)be32_to_cpu(h->magic),
4425 h->command, h->length);
4426 goto disconnect;
4427 }
4428 expect = cmd->pkt_size;
4429 ERR_IF(len != expect-sizeof(struct p_header))
4430 goto reconnect;
4431 }
4432 if (received == expect) {
4433 D_ASSERT(cmd != NULL);
4434 if (!cmd->process(mdev, h))
4435 goto reconnect;
4436
4437 buf = h;
4438 received = 0;
4439 expect = sizeof(struct p_header);
4440 cmd = NULL;
4441 }
4442 }
4443
4444 if (0) {
4445 reconnect:
4446 drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE));
4447 }
4448 if (0) {
4449 disconnect:
4450 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4451 }
4452 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4453
4454 D_ASSERT(mdev->state.conn < C_CONNECTED);
4455 dev_info(DEV, "asender terminated\n");
4456
4457 return 0;
4458 }
Linux 2.6 libata-dev (mirror)