4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10 drbd is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
15 drbd is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with drbd; see the file COPYING. If not, write to
22 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
26 #include <linux/module.h>
28 #include <asm/uaccess.h>
31 #include <linux/drbd.h>
33 #include <linux/file.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/string.h>
46 #include <linux/scatterlist.h>
54 struct drbd_epoch *epoch;
63 static int drbd_do_handshake(struct drbd_conf *mdev);
64 static int drbd_do_auth(struct drbd_conf *mdev);
66 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event);
67 static int e_end_block(struct drbd_conf *, struct drbd_work *, int);
69 static struct drbd_epoch *previous_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch)
71 struct drbd_epoch *prev;
72 spin_lock(&mdev->epoch_lock);
73 prev = list_entry(epoch->list.prev, struct drbd_epoch, list);
74 if (prev == epoch || prev == mdev->current_epoch)
76 spin_unlock(&mdev->epoch_lock);
80 #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN)
83 * some helper functions to deal with single linked page lists,
84 * page->private being our "next" pointer.
87 /* If at least n pages are linked at head, get n pages off.
88 * Otherwise, don't modify head, and return NULL.
89 * Locking is the responsibility of the caller.
91 static struct page *page_chain_del(struct page **head, int n)
105 tmp = page_chain_next(page);
107 break; /* found sufficient pages */
109 /* insufficient pages, don't use any of them. */
114 /* add end of list marker for the returned list */
115 set_page_private(page, 0);
116 /* actual return value, and adjustment of head */
122 /* may be used outside of locks to find the tail of a (usually short)
123 * "private" page chain, before adding it back to a global chain head
124 * with page_chain_add() under a spinlock. */
125 static struct page *page_chain_tail(struct page *page, int *len)
129 while ((tmp = page_chain_next(page)))
136 static int page_chain_free(struct page *page)
140 page_chain_for_each_safe(page, tmp) {
147 static void page_chain_add(struct page **head,
148 struct page *chain_first, struct page *chain_last)
152 tmp = page_chain_tail(chain_first, NULL);
153 BUG_ON(tmp != chain_last);
156 /* add chain to head */
157 set_page_private(chain_last, (unsigned long)*head);
161 static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number)
163 struct page *page = NULL;
164 struct page *tmp = NULL;
167 /* Yes, testing drbd_pp_vacant outside the lock is racy.
168 * So what. It saves a spin_lock. */
169 if (drbd_pp_vacant >= number) {
170 spin_lock(&drbd_pp_lock);
171 page = page_chain_del(&drbd_pp_pool, number);
173 drbd_pp_vacant -= number;
174 spin_unlock(&drbd_pp_lock);
179 /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD
180 * "criss-cross" setup, that might cause write-out on some other DRBD,
181 * which in turn might block on the other node at this very place. */
182 for (i = 0; i < number; i++) {
183 tmp = alloc_page(GFP_TRY);
186 set_page_private(tmp, (unsigned long)page);
193 /* Not enough pages immediately available this time.
194 * No need to jump around here, drbd_pp_alloc will retry this
195 * function "soon". */
197 tmp = page_chain_tail(page, NULL);
198 spin_lock(&drbd_pp_lock);
199 page_chain_add(&drbd_pp_pool, page, tmp);
201 spin_unlock(&drbd_pp_lock);
206 /* kick lower level device, if we have more than (arbitrary number)
207 * reference counts on it, which typically are locally submitted io
208 * requests. don't use unacked_cnt, so we speed up proto A and B, too. */
209 static void maybe_kick_lo(struct drbd_conf *mdev)
211 if (atomic_read(&mdev->local_cnt) >= mdev->net_conf->unplug_watermark)
215 static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed)
217 struct drbd_epoch_entry *e;
218 struct list_head *le, *tle;
220 /* The EEs are always appended to the end of the list. Since
221 they are sent in order over the wire, they have to finish
222 in order. As soon as we see the first not finished we can
223 stop to examine the list... */
225 list_for_each_safe(le, tle, &mdev->net_ee) {
226 e = list_entry(le, struct drbd_epoch_entry, w.list);
227 if (drbd_ee_has_active_page(e))
229 list_move(le, to_be_freed);
233 static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev)
235 LIST_HEAD(reclaimed);
236 struct drbd_epoch_entry *e, *t;
239 spin_lock_irq(&mdev->req_lock);
240 reclaim_net_ee(mdev, &reclaimed);
241 spin_unlock_irq(&mdev->req_lock);
243 list_for_each_entry_safe(e, t, &reclaimed, w.list)
244 drbd_free_net_ee(mdev, e);
248 * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled)
249 * @mdev: DRBD device.
250 * @number: number of pages requested
251 * @retry: whether to retry, if not enough pages are available right now
253 * Tries to allocate number pages, first from our own page pool, then from
254 * the kernel, unless this allocation would exceed the max_buffers setting.
255 * Possibly retry until DRBD frees sufficient pages somewhere else.
257 * Returns a page chain linked via page->private.
259 static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry)
261 struct page *page = NULL;
264 /* Yes, we may run up to @number over max_buffers. If we
265 * follow it strictly, the admin will get it wrong anyways. */
266 if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers)
267 page = drbd_pp_first_pages_or_try_alloc(mdev, number);
269 while (page == NULL) {
270 prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE);
272 drbd_kick_lo_and_reclaim_net(mdev);
274 if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) {
275 page = drbd_pp_first_pages_or_try_alloc(mdev, number);
283 if (signal_pending(current)) {
284 dev_warn(DEV, "drbd_pp_alloc interrupted!\n");
290 finish_wait(&drbd_pp_wait, &wait);
293 atomic_add(number, &mdev->pp_in_use);
297 /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc.
298 * Is also used from inside an other spin_lock_irq(&mdev->req_lock);
299 * Either links the page chain back to the global pool,
300 * or returns all pages to the system. */
301 static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net)
303 atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use;
306 if (drbd_pp_vacant > (DRBD_MAX_SEGMENT_SIZE/PAGE_SIZE)*minor_count)
307 i = page_chain_free(page);
310 tmp = page_chain_tail(page, &i);
311 spin_lock(&drbd_pp_lock);
312 page_chain_add(&drbd_pp_pool, page, tmp);
314 spin_unlock(&drbd_pp_lock);
316 i = atomic_sub_return(i, a);
318 dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n",
319 is_net ? "pp_in_use_by_net" : "pp_in_use", i);
320 wake_up(&drbd_pp_wait);
324 You need to hold the req_lock:
325 _drbd_wait_ee_list_empty()
327 You must not have the req_lock:
333 drbd_process_done_ee()
335 drbd_wait_ee_list_empty()
338 struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev,
341 unsigned int data_size,
342 gfp_t gfp_mask) __must_hold(local)
344 struct drbd_epoch_entry *e;
346 unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT;
348 if (FAULT_ACTIVE(mdev, DRBD_FAULT_AL_EE))
351 e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM);
353 if (!(gfp_mask & __GFP_NOWARN))
354 dev_err(DEV, "alloc_ee: Allocation of an EE failed\n");
358 page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT));
362 INIT_HLIST_NODE(&e->colision);
366 atomic_set(&e->pending_bios, 0);
375 mempool_free(e, drbd_ee_mempool);
379 void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, int is_net)
381 if (e->flags & EE_HAS_DIGEST)
383 drbd_pp_free(mdev, e->pages, is_net);
384 D_ASSERT(atomic_read(&e->pending_bios) == 0);
385 D_ASSERT(hlist_unhashed(&e->colision));
386 mempool_free(e, drbd_ee_mempool);
389 int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list)
391 LIST_HEAD(work_list);
392 struct drbd_epoch_entry *e, *t;
394 int is_net = list == &mdev->net_ee;
396 spin_lock_irq(&mdev->req_lock);
397 list_splice_init(list, &work_list);
398 spin_unlock_irq(&mdev->req_lock);
400 list_for_each_entry_safe(e, t, &work_list, w.list) {
401 drbd_free_some_ee(mdev, e, is_net);
409 * This function is called from _asender only_
410 * but see also comments in _req_mod(,barrier_acked)
411 * and receive_Barrier.
413 * Move entries from net_ee to done_ee, if ready.
414 * Grab done_ee, call all callbacks, free the entries.
415 * The callbacks typically send out ACKs.
417 static int drbd_process_done_ee(struct drbd_conf *mdev)
419 LIST_HEAD(work_list);
420 LIST_HEAD(reclaimed);
421 struct drbd_epoch_entry *e, *t;
422 int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS);
424 spin_lock_irq(&mdev->req_lock);
425 reclaim_net_ee(mdev, &reclaimed);
426 list_splice_init(&mdev->done_ee, &work_list);
427 spin_unlock_irq(&mdev->req_lock);
429 list_for_each_entry_safe(e, t, &reclaimed, w.list)
430 drbd_free_net_ee(mdev, e);
432 /* possible callbacks here:
433 * e_end_block, and e_end_resync_block, e_send_discard_ack.
434 * all ignore the last argument.
436 list_for_each_entry_safe(e, t, &work_list, w.list) {
437 /* list_del not necessary, next/prev members not touched */
438 ok = e->w.cb(mdev, &e->w, !ok) && ok;
439 drbd_free_ee(mdev, e);
441 wake_up(&mdev->ee_wait);
446 void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head)
450 /* avoids spin_lock/unlock
451 * and calling prepare_to_wait in the fast path */
452 while (!list_empty(head)) {
453 prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE);
454 spin_unlock_irq(&mdev->req_lock);
457 finish_wait(&mdev->ee_wait, &wait);
458 spin_lock_irq(&mdev->req_lock);
462 void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head)
464 spin_lock_irq(&mdev->req_lock);
465 _drbd_wait_ee_list_empty(mdev, head);
466 spin_unlock_irq(&mdev->req_lock);
469 /* see also kernel_accept; which is only present since 2.6.18.
470 * also we want to log which part of it failed, exactly */
471 static int drbd_accept(struct drbd_conf *mdev, const char **what,
472 struct socket *sock, struct socket **newsock)
474 struct sock *sk = sock->sk;
478 err = sock->ops->listen(sock, 5);
482 *what = "sock_create_lite";
483 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
489 err = sock->ops->accept(sock, *newsock, 0);
491 sock_release(*newsock);
495 (*newsock)->ops = sock->ops;
501 static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock,
502 void *buf, size_t size, int flags)
509 struct msghdr msg = {
511 .msg_iov = (struct iovec *)&iov,
512 .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL)
518 rv = sock_recvmsg(sock, &msg, size, msg.msg_flags);
524 static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size)
531 struct msghdr msg = {
533 .msg_iov = (struct iovec *)&iov,
534 .msg_flags = MSG_WAITALL | MSG_NOSIGNAL
542 rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags);
547 * ECONNRESET other side closed the connection
548 * ERESTARTSYS (on sock) we got a signal
552 if (rv == -ECONNRESET)
553 dev_info(DEV, "sock was reset by peer\n");
554 else if (rv != -ERESTARTSYS)
555 dev_err(DEV, "sock_recvmsg returned %d\n", rv);
557 } else if (rv == 0) {
558 dev_info(DEV, "sock was shut down by peer\n");
561 /* signal came in, or peer/link went down,
562 * after we read a partial message
564 /* D_ASSERT(signal_pending(current)); */
572 drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE));
578 * On individual connections, the socket buffer size must be set prior to the
579 * listen(2) or connect(2) calls in order to have it take effect.
580 * This is our wrapper to do so.
582 static void drbd_setbufsize(struct socket *sock, unsigned int snd,
585 /* open coded SO_SNDBUF, SO_RCVBUF */
587 sock->sk->sk_sndbuf = snd;
588 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
591 sock->sk->sk_rcvbuf = rcv;
592 sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
596 static struct socket *drbd_try_connect(struct drbd_conf *mdev)
600 struct sockaddr_in6 src_in6;
602 int disconnect_on_error = 1;
604 if (!get_net_conf(mdev))
607 what = "sock_create_kern";
608 err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family,
609 SOCK_STREAM, IPPROTO_TCP, &sock);
615 sock->sk->sk_rcvtimeo =
616 sock->sk->sk_sndtimeo = mdev->net_conf->try_connect_int*HZ;
617 drbd_setbufsize(sock, mdev->net_conf->sndbuf_size,
618 mdev->net_conf->rcvbuf_size);
620 /* explicitly bind to the configured IP as source IP
621 * for the outgoing connections.
622 * This is needed for multihomed hosts and to be
623 * able to use lo: interfaces for drbd.
624 * Make sure to use 0 as port number, so linux selects
625 * a free one dynamically.
627 memcpy(&src_in6, mdev->net_conf->my_addr,
628 min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6)));
629 if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6)
630 src_in6.sin6_port = 0;
632 ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */
634 what = "bind before connect";
635 err = sock->ops->bind(sock,
636 (struct sockaddr *) &src_in6,
637 mdev->net_conf->my_addr_len);
641 /* connect may fail, peer not yet available.
642 * stay C_WF_CONNECTION, don't go Disconnecting! */
643 disconnect_on_error = 0;
645 err = sock->ops->connect(sock,
646 (struct sockaddr *)mdev->net_conf->peer_addr,
647 mdev->net_conf->peer_addr_len, 0);
656 /* timeout, busy, signal pending */
657 case ETIMEDOUT: case EAGAIN: case EINPROGRESS:
658 case EINTR: case ERESTARTSYS:
659 /* peer not (yet) available, network problem */
660 case ECONNREFUSED: case ENETUNREACH:
661 case EHOSTDOWN: case EHOSTUNREACH:
662 disconnect_on_error = 0;
665 dev_err(DEV, "%s failed, err = %d\n", what, err);
667 if (disconnect_on_error)
668 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
674 static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev)
677 struct socket *s_estab = NULL, *s_listen;
680 if (!get_net_conf(mdev))
683 what = "sock_create_kern";
684 err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family,
685 SOCK_STREAM, IPPROTO_TCP, &s_listen);
691 timeo = mdev->net_conf->try_connect_int * HZ;
692 timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */
694 s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */
695 s_listen->sk->sk_rcvtimeo = timeo;
696 s_listen->sk->sk_sndtimeo = timeo;
697 drbd_setbufsize(s_listen, mdev->net_conf->sndbuf_size,
698 mdev->net_conf->rcvbuf_size);
700 what = "bind before listen";
701 err = s_listen->ops->bind(s_listen,
702 (struct sockaddr *) mdev->net_conf->my_addr,
703 mdev->net_conf->my_addr_len);
707 err = drbd_accept(mdev, &what, s_listen, &s_estab);
711 sock_release(s_listen);
713 if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) {
714 dev_err(DEV, "%s failed, err = %d\n", what, err);
715 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
723 static int drbd_send_fp(struct drbd_conf *mdev,
724 struct socket *sock, enum drbd_packets cmd)
726 struct p_header80 *h = &mdev->data.sbuf.header.h80;
728 return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0);
731 static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock)
733 struct p_header80 *h = &mdev->data.rbuf.header.h80;
736 rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0);
738 if (rr == sizeof(*h) && h->magic == BE_DRBD_MAGIC)
739 return be16_to_cpu(h->command);
745 * drbd_socket_okay() - Free the socket if its connection is not okay
746 * @mdev: DRBD device.
747 * @sock: pointer to the pointer to the socket.
749 static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock)
757 rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK);
759 if (rr > 0 || rr == -EAGAIN) {
770 * 1 yes, we have a valid connection
771 * 0 oops, did not work out, please try again
772 * -1 peer talks different language,
773 * no point in trying again, please go standalone.
774 * -2 We do not have a network config...
776 static int drbd_connect(struct drbd_conf *mdev)
778 struct socket *s, *sock, *msock;
781 D_ASSERT(!mdev->data.socket);
783 if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS)
786 clear_bit(DISCARD_CONCURRENT, &mdev->flags);
793 /* 3 tries, this should take less than a second! */
794 s = drbd_try_connect(mdev);
797 /* give the other side time to call bind() & listen() */
798 __set_current_state(TASK_INTERRUPTIBLE);
799 schedule_timeout(HZ / 10);
804 drbd_send_fp(mdev, s, P_HAND_SHAKE_S);
808 drbd_send_fp(mdev, s, P_HAND_SHAKE_M);
812 dev_err(DEV, "Logic error in drbd_connect()\n");
813 goto out_release_sockets;
818 __set_current_state(TASK_INTERRUPTIBLE);
819 schedule_timeout(HZ / 10);
820 ok = drbd_socket_okay(mdev, &sock);
821 ok = drbd_socket_okay(mdev, &msock) && ok;
827 s = drbd_wait_for_connect(mdev);
829 try = drbd_recv_fp(mdev, s);
830 drbd_socket_okay(mdev, &sock);
831 drbd_socket_okay(mdev, &msock);
835 dev_warn(DEV, "initial packet S crossed\n");
842 dev_warn(DEV, "initial packet M crossed\n");
846 set_bit(DISCARD_CONCURRENT, &mdev->flags);
849 dev_warn(DEV, "Error receiving initial packet\n");
856 if (mdev->state.conn <= C_DISCONNECTING)
857 goto out_release_sockets;
858 if (signal_pending(current)) {
859 flush_signals(current);
861 if (get_t_state(&mdev->receiver) == Exiting)
862 goto out_release_sockets;
866 ok = drbd_socket_okay(mdev, &sock);
867 ok = drbd_socket_okay(mdev, &msock) && ok;
873 msock->sk->sk_reuse = 1; /* SO_REUSEADDR */
874 sock->sk->sk_reuse = 1; /* SO_REUSEADDR */
876 sock->sk->sk_allocation = GFP_NOIO;
877 msock->sk->sk_allocation = GFP_NOIO;
879 sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK;
880 msock->sk->sk_priority = TC_PRIO_INTERACTIVE;
883 * sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
884 * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
885 * first set it to the P_HAND_SHAKE timeout,
886 * which we set to 4x the configured ping_timeout. */
887 sock->sk->sk_sndtimeo =
888 sock->sk->sk_rcvtimeo = mdev->net_conf->ping_timeo*4*HZ/10;
890 msock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
891 msock->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
893 /* we don't want delays.
894 * we use TCP_CORK where apropriate, though */
895 drbd_tcp_nodelay(sock);
896 drbd_tcp_nodelay(msock);
898 mdev->data.socket = sock;
899 mdev->meta.socket = msock;
900 mdev->last_received = jiffies;
902 D_ASSERT(mdev->asender.task == NULL);
904 h = drbd_do_handshake(mdev);
908 if (mdev->cram_hmac_tfm) {
909 /* drbd_request_state(mdev, NS(conn, WFAuth)); */
910 switch (drbd_do_auth(mdev)) {
912 dev_err(DEV, "Authentication of peer failed\n");
915 dev_err(DEV, "Authentication of peer failed, trying again.\n");
920 if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS)
923 sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
924 sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
926 atomic_set(&mdev->packet_seq, 0);
929 drbd_thread_start(&mdev->asender);
931 if (mdev->agreed_pro_version < 95 && get_ldev(mdev)) {
932 drbd_setup_queue_param(mdev, DRBD_MAX_SIZE_H80_PACKET);
936 if (!drbd_send_protocol(mdev))
938 drbd_send_sync_param(mdev, &mdev->sync_conf);
939 drbd_send_sizes(mdev, 0, 0);
940 drbd_send_uuids(mdev);
941 drbd_send_state(mdev);
942 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
943 clear_bit(RESIZE_PENDING, &mdev->flags);
955 static int drbd_recv_header(struct drbd_conf *mdev, enum drbd_packets *cmd, unsigned int *packet_size)
957 union p_header *h = &mdev->data.rbuf.header;
960 r = drbd_recv(mdev, h, sizeof(*h));
961 if (unlikely(r != sizeof(*h))) {
962 dev_err(DEV, "short read expecting header on sock: r=%d\n", r);
966 if (likely(h->h80.magic == BE_DRBD_MAGIC)) {
967 *cmd = be16_to_cpu(h->h80.command);
968 *packet_size = be16_to_cpu(h->h80.length);
969 } else if (h->h95.magic == BE_DRBD_MAGIC_BIG) {
970 *cmd = be16_to_cpu(h->h95.command);
971 *packet_size = be32_to_cpu(h->h95.length);
973 dev_err(DEV, "magic?? on data m: 0x%lx c: %d l: %d\n",
974 (long)be32_to_cpu(h->h80.magic),
975 h->h80.command, h->h80.length);
978 mdev->last_received = jiffies;
983 static enum finish_epoch drbd_flush_after_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch)
987 if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) {
988 rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL,
989 NULL, BLKDEV_IFL_WAIT);
991 dev_err(DEV, "local disk flush failed with status %d\n", rv);
992 /* would rather check on EOPNOTSUPP, but that is not reliable.
993 * don't try again for ANY return value != 0
994 * if (rv == -EOPNOTSUPP) */
995 drbd_bump_write_ordering(mdev, WO_drain_io);
1000 return drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE);
1003 static int w_flush(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1005 struct flush_work *fw = (struct flush_work *)w;
1006 struct drbd_epoch *epoch = fw->epoch;
1010 if (!test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags))
1011 drbd_flush_after_epoch(mdev, epoch);
1013 drbd_may_finish_epoch(mdev, epoch, EV_PUT |
1014 (mdev->state.conn < C_CONNECTED ? EV_CLEANUP : 0));
1020 * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it.
1021 * @mdev: DRBD device.
1022 * @epoch: Epoch object.
1025 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev,
1026 struct drbd_epoch *epoch,
1027 enum epoch_event ev)
1029 int finish, epoch_size;
1030 struct drbd_epoch *next_epoch;
1031 int schedule_flush = 0;
1032 enum finish_epoch rv = FE_STILL_LIVE;
1034 spin_lock(&mdev->epoch_lock);
1039 epoch_size = atomic_read(&epoch->epoch_size);
1041 switch (ev & ~EV_CLEANUP) {
1043 atomic_dec(&epoch->active);
1045 case EV_GOT_BARRIER_NR:
1046 set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags);
1048 /* Special case: If we just switched from WO_bio_barrier to
1049 WO_bdev_flush we should not finish the current epoch */
1050 if (test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags) && epoch_size == 1 &&
1051 mdev->write_ordering != WO_bio_barrier &&
1052 epoch == mdev->current_epoch)
1053 clear_bit(DE_CONTAINS_A_BARRIER, &epoch->flags);
1055 case EV_BARRIER_DONE:
1056 set_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags);
1058 case EV_BECAME_LAST:
1063 if (epoch_size != 0 &&
1064 atomic_read(&epoch->active) == 0 &&
1065 test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) &&
1066 epoch->list.prev == &mdev->current_epoch->list &&
1067 !test_bit(DE_IS_FINISHING, &epoch->flags)) {
1068 /* Nearly all conditions are met to finish that epoch... */
1069 if (test_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags) ||
1070 mdev->write_ordering == WO_none ||
1071 (epoch_size == 1 && test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) ||
1074 set_bit(DE_IS_FINISHING, &epoch->flags);
1075 } else if (!test_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags) &&
1076 mdev->write_ordering == WO_bio_barrier) {
1077 atomic_inc(&epoch->active);
1082 if (!(ev & EV_CLEANUP)) {
1083 spin_unlock(&mdev->epoch_lock);
1084 drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size);
1085 spin_lock(&mdev->epoch_lock);
1089 if (mdev->current_epoch != epoch) {
1090 next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list);
1091 list_del(&epoch->list);
1092 ev = EV_BECAME_LAST | (ev & EV_CLEANUP);
1096 if (rv == FE_STILL_LIVE)
1100 atomic_set(&epoch->epoch_size, 0);
1101 /* atomic_set(&epoch->active, 0); is already zero */
1102 if (rv == FE_STILL_LIVE)
1113 spin_unlock(&mdev->epoch_lock);
1115 if (schedule_flush) {
1116 struct flush_work *fw;
1117 fw = kmalloc(sizeof(*fw), GFP_ATOMIC);
1121 drbd_queue_work(&mdev->data.work, &fw->w);
1123 dev_warn(DEV, "Could not kmalloc a flush_work obj\n");
1124 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags);
1125 /* That is not a recursion, only one level */
1126 drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE);
1127 drbd_may_finish_epoch(mdev, epoch, EV_PUT);
1135 * drbd_bump_write_ordering() - Fall back to an other write ordering method
1136 * @mdev: DRBD device.
1137 * @wo: Write ordering method to try.
1139 void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local)
1141 enum write_ordering_e pwo;
1142 static char *write_ordering_str[] = {
1144 [WO_drain_io] = "drain",
1145 [WO_bdev_flush] = "flush",
1146 [WO_bio_barrier] = "barrier",
1149 pwo = mdev->write_ordering;
1151 if (wo == WO_bio_barrier && mdev->ldev->dc.no_disk_barrier)
1153 if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush)
1155 if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain)
1157 mdev->write_ordering = wo;
1158 if (pwo != mdev->write_ordering || wo == WO_bio_barrier)
1159 dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]);
1164 * @mdev: DRBD device.
1166 * @rw: flag field, see bio->bi_rw
1168 /* TODO allocate from our own bio_set. */
1169 int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e,
1170 const unsigned rw, const int fault_type)
1172 struct bio *bios = NULL;
1174 struct page *page = e->pages;
1175 sector_t sector = e->sector;
1176 unsigned ds = e->size;
1177 unsigned n_bios = 0;
1178 unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT;
1180 /* In most cases, we will only need one bio. But in case the lower
1181 * level restrictions happen to be different at this offset on this
1182 * side than those of the sending peer, we may need to submit the
1183 * request in more than one bio. */
1185 bio = bio_alloc(GFP_NOIO, nr_pages);
1187 dev_err(DEV, "submit_ee: Allocation of a bio failed\n");
1190 /* > e->sector, unless this is the first bio */
1191 bio->bi_sector = sector;
1192 bio->bi_bdev = mdev->ldev->backing_bdev;
1193 /* we special case some flags in the multi-bio case, see below
1194 * (REQ_UNPLUG, REQ_HARDBARRIER) */
1196 bio->bi_private = e;
1197 bio->bi_end_io = drbd_endio_sec;
1199 bio->bi_next = bios;
1203 page_chain_for_each(page) {
1204 unsigned len = min_t(unsigned, ds, PAGE_SIZE);
1205 if (!bio_add_page(bio, page, len, 0)) {
1206 /* a single page must always be possible! */
1207 BUG_ON(bio->bi_vcnt == 0);
1214 D_ASSERT(page == NULL);
1217 atomic_set(&e->pending_bios, n_bios);
1220 bios = bios->bi_next;
1221 bio->bi_next = NULL;
1223 /* strip off REQ_UNPLUG unless it is the last bio */
1225 bio->bi_rw &= ~REQ_UNPLUG;
1227 drbd_generic_make_request(mdev, fault_type, bio);
1229 /* strip off REQ_HARDBARRIER,
1230 * unless it is the first or last bio */
1231 if (bios && bios->bi_next)
1232 bios->bi_rw &= ~REQ_HARDBARRIER;
1234 maybe_kick_lo(mdev);
1240 bios = bios->bi_next;
1247 * w_e_reissue() - Worker callback; Resubmit a bio, without REQ_HARDBARRIER set
1248 * @mdev: DRBD device.
1250 * @cancel: The connection will be closed anyways (unused in this callback)
1252 int w_e_reissue(struct drbd_conf *mdev, struct drbd_work *w, int cancel) __releases(local)
1254 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1255 /* We leave DE_CONTAINS_A_BARRIER and EE_IS_BARRIER in place,
1256 (and DE_BARRIER_IN_NEXT_EPOCH_ISSUED in the previous Epoch)
1257 so that we can finish that epoch in drbd_may_finish_epoch().
1258 That is necessary if we already have a long chain of Epochs, before
1259 we realize that REQ_HARDBARRIER is actually not supported */
1261 /* As long as the -ENOTSUPP on the barrier is reported immediately
1262 that will never trigger. If it is reported late, we will just
1263 print that warning and continue correctly for all future requests
1264 with WO_bdev_flush */
1265 if (previous_epoch(mdev, e->epoch))
1266 dev_warn(DEV, "Write ordering was not enforced (one time event)\n");
1268 /* we still have a local reference,
1269 * get_ldev was done in receive_Data. */
1271 e->w.cb = e_end_block;
1272 if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_DT_WR) != 0) {
1273 /* drbd_submit_ee fails for one reason only:
1274 * if was not able to allocate sufficient bios.
1275 * requeue, try again later. */
1276 e->w.cb = w_e_reissue;
1277 drbd_queue_work(&mdev->data.work, &e->w);
1282 static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1284 int rv, issue_flush;
1285 struct p_barrier *p = &mdev->data.rbuf.barrier;
1286 struct drbd_epoch *epoch;
1290 if (mdev->net_conf->wire_protocol != DRBD_PROT_C)
1293 mdev->current_epoch->barrier_nr = p->barrier;
1294 rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR);
1296 /* P_BARRIER_ACK may imply that the corresponding extent is dropped from
1297 * the activity log, which means it would not be resynced in case the
1298 * R_PRIMARY crashes now.
1299 * Therefore we must send the barrier_ack after the barrier request was
1301 switch (mdev->write_ordering) {
1302 case WO_bio_barrier:
1304 if (rv == FE_RECYCLED)
1310 if (rv == FE_STILL_LIVE) {
1311 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags);
1312 drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
1313 rv = drbd_flush_after_epoch(mdev, mdev->current_epoch);
1315 if (rv == FE_RECYCLED)
1318 /* The asender will send all the ACKs and barrier ACKs out, since
1319 all EEs moved from the active_ee to the done_ee. We need to
1320 provide a new epoch object for the EEs that come in soon */
1324 /* receiver context, in the writeout path of the other node.
1325 * avoid potential distributed deadlock */
1326 epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO);
1328 dev_warn(DEV, "Allocation of an epoch failed, slowing down\n");
1329 issue_flush = !test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags);
1330 drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
1332 rv = drbd_flush_after_epoch(mdev, mdev->current_epoch);
1333 if (rv == FE_RECYCLED)
1337 drbd_wait_ee_list_empty(mdev, &mdev->done_ee);
1343 atomic_set(&epoch->epoch_size, 0);
1344 atomic_set(&epoch->active, 0);
1346 spin_lock(&mdev->epoch_lock);
1347 if (atomic_read(&mdev->current_epoch->epoch_size)) {
1348 list_add(&epoch->list, &mdev->current_epoch->list);
1349 mdev->current_epoch = epoch;
1352 /* The current_epoch got recycled while we allocated this one... */
1355 spin_unlock(&mdev->epoch_lock);
1360 /* used from receive_RSDataReply (recv_resync_read)
1361 * and from receive_Data */
1362 static struct drbd_epoch_entry *
1363 read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local)
1365 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
1366 struct drbd_epoch_entry *e;
1369 void *dig_in = mdev->int_dig_in;
1370 void *dig_vv = mdev->int_dig_vv;
1371 unsigned long *data;
1373 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1374 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1377 rr = drbd_recv(mdev, dig_in, dgs);
1379 dev_warn(DEV, "short read receiving data digest: read %d expected %d\n",
1387 ERR_IF(data_size & 0x1ff) return NULL;
1388 ERR_IF(data_size > DRBD_MAX_SEGMENT_SIZE) return NULL;
1390 /* even though we trust out peer,
1391 * we sometimes have to double check. */
1392 if (sector + (data_size>>9) > capacity) {
1393 dev_err(DEV, "capacity: %llus < sector: %llus + size: %u\n",
1394 (unsigned long long)capacity,
1395 (unsigned long long)sector, data_size);
1399 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
1400 * "criss-cross" setup, that might cause write-out on some other DRBD,
1401 * which in turn might block on the other node at this very place. */
1402 e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO);
1408 page_chain_for_each(page) {
1409 unsigned len = min_t(int, ds, PAGE_SIZE);
1411 rr = drbd_recv(mdev, data, len);
1412 if (FAULT_ACTIVE(mdev, DRBD_FAULT_RECEIVE)) {
1413 dev_err(DEV, "Fault injection: Corrupting data on receive\n");
1414 data[0] = data[0] ^ (unsigned long)-1;
1418 drbd_free_ee(mdev, e);
1419 dev_warn(DEV, "short read receiving data: read %d expected %d\n",
1427 drbd_csum_ee(mdev, mdev->integrity_r_tfm, e, dig_vv);
1428 if (memcmp(dig_in, dig_vv, dgs)) {
1429 dev_err(DEV, "Digest integrity check FAILED.\n");
1430 drbd_bcast_ee(mdev, "digest failed",
1431 dgs, dig_in, dig_vv, e);
1432 drbd_free_ee(mdev, e);
1436 mdev->recv_cnt += data_size>>9;
1440 /* drbd_drain_block() just takes a data block
1441 * out of the socket input buffer, and discards it.
1443 static int drbd_drain_block(struct drbd_conf *mdev, int data_size)
1452 page = drbd_pp_alloc(mdev, 1, 1);
1456 rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE));
1457 if (rr != min_t(int, data_size, PAGE_SIZE)) {
1459 dev_warn(DEV, "short read receiving data: read %d expected %d\n",
1460 rr, min_t(int, data_size, PAGE_SIZE));
1466 drbd_pp_free(mdev, page, 0);
1470 static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req,
1471 sector_t sector, int data_size)
1473 struct bio_vec *bvec;
1475 int dgs, rr, i, expect;
1476 void *dig_in = mdev->int_dig_in;
1477 void *dig_vv = mdev->int_dig_vv;
1479 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1480 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1483 rr = drbd_recv(mdev, dig_in, dgs);
1485 dev_warn(DEV, "short read receiving data reply digest: read %d expected %d\n",
1493 /* optimistically update recv_cnt. if receiving fails below,
1494 * we disconnect anyways, and counters will be reset. */
1495 mdev->recv_cnt += data_size>>9;
1497 bio = req->master_bio;
1498 D_ASSERT(sector == bio->bi_sector);
1500 bio_for_each_segment(bvec, bio, i) {
1501 expect = min_t(int, data_size, bvec->bv_len);
1502 rr = drbd_recv(mdev,
1503 kmap(bvec->bv_page)+bvec->bv_offset,
1505 kunmap(bvec->bv_page);
1507 dev_warn(DEV, "short read receiving data reply: "
1508 "read %d expected %d\n",
1516 drbd_csum_bio(mdev, mdev->integrity_r_tfm, bio, dig_vv);
1517 if (memcmp(dig_in, dig_vv, dgs)) {
1518 dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n");
1523 D_ASSERT(data_size == 0);
1527 /* e_end_resync_block() is called via
1528 * drbd_process_done_ee() by asender only */
1529 static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1531 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1532 sector_t sector = e->sector;
1535 D_ASSERT(hlist_unhashed(&e->colision));
1537 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1538 drbd_set_in_sync(mdev, sector, e->size);
1539 ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e);
1541 /* Record failure to sync */
1542 drbd_rs_failed_io(mdev, sector, e->size);
1544 ok = drbd_send_ack(mdev, P_NEG_ACK, e);
1551 static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local)
1553 struct drbd_epoch_entry *e;
1555 e = read_in_block(mdev, ID_SYNCER, sector, data_size);
1559 dec_rs_pending(mdev);
1562 /* corresponding dec_unacked() in e_end_resync_block()
1563 * respective _drbd_clear_done_ee */
1565 e->w.cb = e_end_resync_block;
1567 spin_lock_irq(&mdev->req_lock);
1568 list_add(&e->w.list, &mdev->sync_ee);
1569 spin_unlock_irq(&mdev->req_lock);
1571 atomic_add(data_size >> 9, &mdev->rs_sect_ev);
1572 if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0)
1575 drbd_free_ee(mdev, e);
1581 static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1583 struct drbd_request *req;
1586 struct p_data *p = &mdev->data.rbuf.data;
1588 sector = be64_to_cpu(p->sector);
1590 spin_lock_irq(&mdev->req_lock);
1591 req = _ar_id_to_req(mdev, p->block_id, sector);
1592 spin_unlock_irq(&mdev->req_lock);
1593 if (unlikely(!req)) {
1594 dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n");
1598 /* hlist_del(&req->colision) is done in _req_may_be_done, to avoid
1599 * special casing it there for the various failure cases.
1600 * still no race with drbd_fail_pending_reads */
1601 ok = recv_dless_read(mdev, req, sector, data_size);
1604 req_mod(req, data_received);
1605 /* else: nothing. handled from drbd_disconnect...
1606 * I don't think we may complete this just yet
1607 * in case we are "on-disconnect: freeze" */
1612 static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1616 struct p_data *p = &mdev->data.rbuf.data;
1618 sector = be64_to_cpu(p->sector);
1619 D_ASSERT(p->block_id == ID_SYNCER);
1621 if (get_ldev(mdev)) {
1622 /* data is submitted to disk within recv_resync_read.
1623 * corresponding put_ldev done below on error,
1624 * or in drbd_endio_write_sec. */
1625 ok = recv_resync_read(mdev, sector, data_size);
1627 if (__ratelimit(&drbd_ratelimit_state))
1628 dev_err(DEV, "Can not write resync data to local disk.\n");
1630 ok = drbd_drain_block(mdev, data_size);
1632 drbd_send_ack_dp(mdev, P_NEG_ACK, p);
1635 atomic_add(data_size >> 9, &mdev->rs_sect_in);
1640 /* e_end_block() is called via drbd_process_done_ee().
1641 * this means this function only runs in the asender thread
1643 static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1645 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1646 sector_t sector = e->sector;
1647 struct drbd_epoch *epoch;
1650 if (e->flags & EE_IS_BARRIER) {
1651 epoch = previous_epoch(mdev, e->epoch);
1653 drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE + (cancel ? EV_CLEANUP : 0));
1656 if (mdev->net_conf->wire_protocol == DRBD_PROT_C) {
1657 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1658 pcmd = (mdev->state.conn >= C_SYNC_SOURCE &&
1659 mdev->state.conn <= C_PAUSED_SYNC_T &&
1660 e->flags & EE_MAY_SET_IN_SYNC) ?
1661 P_RS_WRITE_ACK : P_WRITE_ACK;
1662 ok &= drbd_send_ack(mdev, pcmd, e);
1663 if (pcmd == P_RS_WRITE_ACK)
1664 drbd_set_in_sync(mdev, sector, e->size);
1666 ok = drbd_send_ack(mdev, P_NEG_ACK, e);
1667 /* we expect it to be marked out of sync anyways...
1668 * maybe assert this? */
1672 /* we delete from the conflict detection hash _after_ we sent out the
1673 * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */
1674 if (mdev->net_conf->two_primaries) {
1675 spin_lock_irq(&mdev->req_lock);
1676 D_ASSERT(!hlist_unhashed(&e->colision));
1677 hlist_del_init(&e->colision);
1678 spin_unlock_irq(&mdev->req_lock);
1680 D_ASSERT(hlist_unhashed(&e->colision));
1683 drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0));
1688 static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1690 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1693 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1694 ok = drbd_send_ack(mdev, P_DISCARD_ACK, e);
1696 spin_lock_irq(&mdev->req_lock);
1697 D_ASSERT(!hlist_unhashed(&e->colision));
1698 hlist_del_init(&e->colision);
1699 spin_unlock_irq(&mdev->req_lock);
1706 /* Called from receive_Data.
1707 * Synchronize packets on sock with packets on msock.
1709 * This is here so even when a P_DATA packet traveling via sock overtook an Ack
1710 * packet traveling on msock, they are still processed in the order they have
1713 * Note: we don't care for Ack packets overtaking P_DATA packets.
1715 * In case packet_seq is larger than mdev->peer_seq number, there are
1716 * outstanding packets on the msock. We wait for them to arrive.
1717 * In case we are the logically next packet, we update mdev->peer_seq
1718 * ourselves. Correctly handles 32bit wrap around.
1720 * Assume we have a 10 GBit connection, that is about 1<<30 byte per second,
1721 * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds
1722 * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have
1723 * 1<<9 == 512 seconds aka ages for the 32bit wrap around...
1725 * returns 0 if we may process the packet,
1726 * -ERESTARTSYS if we were interrupted (by disconnect signal). */
1727 static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq)
1733 spin_lock(&mdev->peer_seq_lock);
1735 prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE);
1736 if (seq_le(packet_seq, mdev->peer_seq+1))
1738 if (signal_pending(current)) {
1742 p_seq = mdev->peer_seq;
1743 spin_unlock(&mdev->peer_seq_lock);
1744 timeout = schedule_timeout(30*HZ);
1745 spin_lock(&mdev->peer_seq_lock);
1746 if (timeout == 0 && p_seq == mdev->peer_seq) {
1748 dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n");
1752 finish_wait(&mdev->seq_wait, &wait);
1753 if (mdev->peer_seq+1 == packet_seq)
1755 spin_unlock(&mdev->peer_seq_lock);
1759 static unsigned long write_flags_to_bio(struct drbd_conf *mdev, u32 dpf)
1761 if (mdev->agreed_pro_version >= 95)
1762 return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) |
1763 (dpf & DP_UNPLUG ? REQ_UNPLUG : 0) |
1764 (dpf & DP_FUA ? REQ_FUA : 0) |
1765 (dpf & DP_FLUSH ? REQ_FUA : 0) |
1766 (dpf & DP_DISCARD ? REQ_DISCARD : 0);
1768 return dpf & DP_RW_SYNC ? (REQ_SYNC | REQ_UNPLUG) : 0;
1771 /* mirrored write */
1772 static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1775 struct drbd_epoch_entry *e;
1776 struct p_data *p = &mdev->data.rbuf.data;
1780 if (!get_ldev(mdev)) {
1781 if (__ratelimit(&drbd_ratelimit_state))
1782 dev_err(DEV, "Can not write mirrored data block "
1783 "to local disk.\n");
1784 spin_lock(&mdev->peer_seq_lock);
1785 if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num))
1787 spin_unlock(&mdev->peer_seq_lock);
1789 drbd_send_ack_dp(mdev, P_NEG_ACK, p);
1790 atomic_inc(&mdev->current_epoch->epoch_size);
1791 return drbd_drain_block(mdev, data_size);
1794 /* get_ldev(mdev) successful.
1795 * Corresponding put_ldev done either below (on various errors),
1796 * or in drbd_endio_write_sec, if we successfully submit the data at
1797 * the end of this function. */
1799 sector = be64_to_cpu(p->sector);
1800 e = read_in_block(mdev, p->block_id, sector, data_size);
1806 e->w.cb = e_end_block;
1808 spin_lock(&mdev->epoch_lock);
1809 e->epoch = mdev->current_epoch;
1810 atomic_inc(&e->epoch->epoch_size);
1811 atomic_inc(&e->epoch->active);
1813 if (mdev->write_ordering == WO_bio_barrier && atomic_read(&e->epoch->epoch_size) == 1) {
1814 struct drbd_epoch *epoch;
1815 /* Issue a barrier if we start a new epoch, and the previous epoch
1816 was not a epoch containing a single request which already was
1818 epoch = list_entry(e->epoch->list.prev, struct drbd_epoch, list);
1819 if (epoch == e->epoch) {
1820 set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags);
1821 rw |= REQ_HARDBARRIER;
1822 e->flags |= EE_IS_BARRIER;
1824 if (atomic_read(&epoch->epoch_size) > 1 ||
1825 !test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) {
1826 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags);
1827 set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags);
1828 rw |= REQ_HARDBARRIER;
1829 e->flags |= EE_IS_BARRIER;
1833 spin_unlock(&mdev->epoch_lock);
1835 dp_flags = be32_to_cpu(p->dp_flags);
1836 rw |= write_flags_to_bio(mdev, dp_flags);
1838 if (dp_flags & DP_MAY_SET_IN_SYNC)
1839 e->flags |= EE_MAY_SET_IN_SYNC;
1841 /* I'm the receiver, I do hold a net_cnt reference. */
1842 if (!mdev->net_conf->two_primaries) {
1843 spin_lock_irq(&mdev->req_lock);
1845 /* don't get the req_lock yet,
1846 * we may sleep in drbd_wait_peer_seq */
1847 const int size = e->size;
1848 const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1850 struct drbd_request *i;
1851 struct hlist_node *n;
1852 struct hlist_head *slot;
1855 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1856 BUG_ON(mdev->ee_hash == NULL);
1857 BUG_ON(mdev->tl_hash == NULL);
1859 /* conflict detection and handling:
1860 * 1. wait on the sequence number,
1861 * in case this data packet overtook ACK packets.
1862 * 2. check our hash tables for conflicting requests.
1863 * we only need to walk the tl_hash, since an ee can not
1864 * have a conflict with an other ee: on the submitting
1865 * node, the corresponding req had already been conflicting,
1866 * and a conflicting req is never sent.
1868 * Note: for two_primaries, we are protocol C,
1869 * so there cannot be any request that is DONE
1870 * but still on the transfer log.
1872 * unconditionally add to the ee_hash.
1874 * if no conflicting request is found:
1877 * if any conflicting request is found
1878 * that has not yet been acked,
1879 * AND I have the "discard concurrent writes" flag:
1880 * queue (via done_ee) the P_DISCARD_ACK; OUT.
1882 * if any conflicting request is found:
1883 * block the receiver, waiting on misc_wait
1884 * until no more conflicting requests are there,
1885 * or we get interrupted (disconnect).
1887 * we do not just write after local io completion of those
1888 * requests, but only after req is done completely, i.e.
1889 * we wait for the P_DISCARD_ACK to arrive!
1891 * then proceed normally, i.e. submit.
1893 if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num)))
1894 goto out_interrupted;
1896 spin_lock_irq(&mdev->req_lock);
1898 hlist_add_head(&e->colision, ee_hash_slot(mdev, sector));
1900 #define OVERLAPS overlaps(i->sector, i->size, sector, size)
1901 slot = tl_hash_slot(mdev, sector);
1904 int have_unacked = 0;
1905 int have_conflict = 0;
1906 prepare_to_wait(&mdev->misc_wait, &wait,
1907 TASK_INTERRUPTIBLE);
1908 hlist_for_each_entry(i, n, slot, colision) {
1910 /* only ALERT on first iteration,
1911 * we may be woken up early... */
1913 dev_alert(DEV, "%s[%u] Concurrent local write detected!"
1914 " new: %llus +%u; pending: %llus +%u\n",
1915 current->comm, current->pid,
1916 (unsigned long long)sector, size,
1917 (unsigned long long)i->sector, i->size);
1918 if (i->rq_state & RQ_NET_PENDING)
1927 /* Discard Ack only for the _first_ iteration */
1928 if (first && discard && have_unacked) {
1929 dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n",
1930 (unsigned long long)sector);
1932 e->w.cb = e_send_discard_ack;
1933 list_add_tail(&e->w.list, &mdev->done_ee);
1935 spin_unlock_irq(&mdev->req_lock);
1937 /* we could probably send that P_DISCARD_ACK ourselves,
1938 * but I don't like the receiver using the msock */
1942 finish_wait(&mdev->misc_wait, &wait);
1946 if (signal_pending(current)) {
1947 hlist_del_init(&e->colision);
1949 spin_unlock_irq(&mdev->req_lock);
1951 finish_wait(&mdev->misc_wait, &wait);
1952 goto out_interrupted;
1955 spin_unlock_irq(&mdev->req_lock);
1958 dev_alert(DEV, "Concurrent write! [W AFTERWARDS] "
1959 "sec=%llus\n", (unsigned long long)sector);
1960 } else if (discard) {
1961 /* we had none on the first iteration.
1962 * there must be none now. */
1963 D_ASSERT(have_unacked == 0);
1966 spin_lock_irq(&mdev->req_lock);
1968 finish_wait(&mdev->misc_wait, &wait);
1971 list_add(&e->w.list, &mdev->active_ee);
1972 spin_unlock_irq(&mdev->req_lock);
1974 switch (mdev->net_conf->wire_protocol) {
1977 /* corresponding dec_unacked() in e_end_block()
1978 * respective _drbd_clear_done_ee */
1981 /* I really don't like it that the receiver thread
1982 * sends on the msock, but anyways */
1983 drbd_send_ack(mdev, P_RECV_ACK, e);
1990 if (mdev->state.pdsk == D_DISKLESS) {
1991 /* In case we have the only disk of the cluster, */
1992 drbd_set_out_of_sync(mdev, e->sector, e->size);
1993 e->flags |= EE_CALL_AL_COMPLETE_IO;
1994 drbd_al_begin_io(mdev, e->sector);
1997 if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0)
2001 /* yes, the epoch_size now is imbalanced.
2002 * but we drop the connection anyways, so we don't have a chance to
2003 * receive a barrier... atomic_inc(&mdev->epoch_size); */
2005 drbd_free_ee(mdev, e);
2009 /* We may throttle resync, if the lower device seems to be busy,
2010 * and current sync rate is above c_min_rate.
2012 * To decide whether or not the lower device is busy, we use a scheme similar
2013 * to MD RAID is_mddev_idle(): if the partition stats reveal "significant"
2014 * (more than 64 sectors) of activity we cannot account for with our own resync
2015 * activity, it obviously is "busy".
2017 * The current sync rate used here uses only the most recent two step marks,
2018 * to have a short time average so we can react faster.
2020 int drbd_rs_should_slow_down(struct drbd_conf *mdev)
2022 struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk;
2023 unsigned long db, dt, dbdt;
2027 /* feature disabled? */
2028 if (mdev->sync_conf.c_min_rate == 0)
2031 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
2032 (int)part_stat_read(&disk->part0, sectors[1]) -
2033 atomic_read(&mdev->rs_sect_ev);
2034 if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) {
2035 unsigned long rs_left;
2038 mdev->rs_last_events = curr_events;
2040 /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP,
2042 i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-2) % DRBD_SYNC_MARKS;
2043 rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed;
2045 dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ;
2048 db = mdev->rs_mark_left[i] - rs_left;
2049 dbdt = Bit2KB(db/dt);
2051 if (dbdt > mdev->sync_conf.c_min_rate)
2058 static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size)
2061 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
2062 struct drbd_epoch_entry *e;
2063 struct digest_info *di = NULL;
2065 unsigned int fault_type;
2066 struct p_block_req *p = &mdev->data.rbuf.block_req;
2068 sector = be64_to_cpu(p->sector);
2069 size = be32_to_cpu(p->blksize);
2071 if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) {
2072 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
2073 (unsigned long long)sector, size);
2076 if (sector + (size>>9) > capacity) {
2077 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
2078 (unsigned long long)sector, size);
2082 if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) {
2083 if (__ratelimit(&drbd_ratelimit_state))
2084 dev_err(DEV, "Can not satisfy peer's read request, "
2085 "no local data.\n");
2086 drbd_send_ack_rp(mdev, cmd == P_DATA_REQUEST ? P_NEG_DREPLY :
2087 P_NEG_RS_DREPLY , p);
2091 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
2092 * "criss-cross" setup, that might cause write-out on some other DRBD,
2093 * which in turn might block on the other node at this very place. */
2094 e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO);
2101 case P_DATA_REQUEST:
2102 e->w.cb = w_e_end_data_req;
2103 fault_type = DRBD_FAULT_DT_RD;
2104 /* application IO, don't drbd_rs_begin_io */
2107 case P_RS_DATA_REQUEST:
2108 e->w.cb = w_e_end_rsdata_req;
2109 fault_type = DRBD_FAULT_RS_RD;
2113 case P_CSUM_RS_REQUEST:
2114 fault_type = DRBD_FAULT_RS_RD;
2115 di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO);
2119 di->digest_size = digest_size;
2120 di->digest = (((char *)di)+sizeof(struct digest_info));
2123 e->flags |= EE_HAS_DIGEST;
2125 if (drbd_recv(mdev, di->digest, digest_size) != digest_size)
2128 if (cmd == P_CSUM_RS_REQUEST) {
2129 D_ASSERT(mdev->agreed_pro_version >= 89);
2130 e->w.cb = w_e_end_csum_rs_req;
2131 } else if (cmd == P_OV_REPLY) {
2132 e->w.cb = w_e_end_ov_reply;
2133 dec_rs_pending(mdev);
2134 /* drbd_rs_begin_io done when we sent this request,
2135 * but accounting still needs to be done. */
2136 goto submit_for_resync;
2141 if (mdev->state.conn >= C_CONNECTED &&
2142 mdev->state.conn != C_VERIFY_T)
2143 dev_warn(DEV, "ASSERT FAILED: got P_OV_REQUEST while being %s\n",
2144 drbd_conn_str(mdev->state.conn));
2145 if (mdev->ov_start_sector == ~(sector_t)0 &&
2146 mdev->agreed_pro_version >= 90) {
2147 mdev->ov_start_sector = sector;
2148 mdev->ov_position = sector;
2149 mdev->ov_left = mdev->rs_total - BM_SECT_TO_BIT(sector);
2150 dev_info(DEV, "Online Verify start sector: %llu\n",
2151 (unsigned long long)sector);
2153 e->w.cb = w_e_end_ov_req;
2154 fault_type = DRBD_FAULT_RS_RD;
2158 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n",
2160 fault_type = DRBD_FAULT_MAX;
2164 /* Throttle, drbd_rs_begin_io and submit should become asynchronous
2165 * wrt the receiver, but it is not as straightforward as it may seem.
2166 * Various places in the resync start and stop logic assume resync
2167 * requests are processed in order, requeuing this on the worker thread
2168 * introduces a bunch of new code for synchronization between threads.
2170 * Unlimited throttling before drbd_rs_begin_io may stall the resync
2171 * "forever", throttling after drbd_rs_begin_io will lock that extent
2172 * for application writes for the same time. For now, just throttle
2173 * here, where the rest of the code expects the receiver to sleep for
2177 /* Throttle before drbd_rs_begin_io, as that locks out application IO;
2178 * this defers syncer requests for some time, before letting at least
2179 * on request through. The resync controller on the receiving side
2180 * will adapt to the incoming rate accordingly.
2182 * We cannot throttle here if remote is Primary/SyncTarget:
2183 * we would also throttle its application reads.
2184 * In that case, throttling is done on the SyncTarget only.
2186 if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev))
2188 if (drbd_rs_begin_io(mdev, e->sector))
2192 atomic_add(size >> 9, &mdev->rs_sect_ev);
2196 spin_lock_irq(&mdev->req_lock);
2197 list_add_tail(&e->w.list, &mdev->read_ee);
2198 spin_unlock_irq(&mdev->req_lock);
2200 if (drbd_submit_ee(mdev, e, READ, fault_type) == 0)
2205 drbd_free_ee(mdev, e);
2209 static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local)
2211 int self, peer, rv = -100;
2212 unsigned long ch_self, ch_peer;
2214 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2215 peer = mdev->p_uuid[UI_BITMAP] & 1;
2217 ch_peer = mdev->p_uuid[UI_SIZE];
2218 ch_self = mdev->comm_bm_set;
2220 switch (mdev->net_conf->after_sb_0p) {
2222 case ASB_DISCARD_SECONDARY:
2223 case ASB_CALL_HELPER:
2224 dev_err(DEV, "Configuration error.\n");
2226 case ASB_DISCONNECT:
2228 case ASB_DISCARD_YOUNGER_PRI:
2229 if (self == 0 && peer == 1) {
2233 if (self == 1 && peer == 0) {
2237 /* Else fall through to one of the other strategies... */
2238 case ASB_DISCARD_OLDER_PRI:
2239 if (self == 0 && peer == 1) {
2243 if (self == 1 && peer == 0) {
2247 /* Else fall through to one of the other strategies... */
2248 dev_warn(DEV, "Discard younger/older primary did not find a decision\n"
2249 "Using discard-least-changes instead\n");
2250 case ASB_DISCARD_ZERO_CHG:
2251 if (ch_peer == 0 && ch_self == 0) {
2252 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2256 if (ch_peer == 0) { rv = 1; break; }
2257 if (ch_self == 0) { rv = -1; break; }
2259 if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG)
2261 case ASB_DISCARD_LEAST_CHG:
2262 if (ch_self < ch_peer)
2264 else if (ch_self > ch_peer)
2266 else /* ( ch_self == ch_peer ) */
2267 /* Well, then use something else. */
2268 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2271 case ASB_DISCARD_LOCAL:
2274 case ASB_DISCARD_REMOTE:
2281 static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local)
2283 int self, peer, hg, rv = -100;
2285 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2286 peer = mdev->p_uuid[UI_BITMAP] & 1;
2288 switch (mdev->net_conf->after_sb_1p) {
2289 case ASB_DISCARD_YOUNGER_PRI:
2290 case ASB_DISCARD_OLDER_PRI:
2291 case ASB_DISCARD_LEAST_CHG:
2292 case ASB_DISCARD_LOCAL:
2293 case ASB_DISCARD_REMOTE:
2294 dev_err(DEV, "Configuration error.\n");
2296 case ASB_DISCONNECT:
2299 hg = drbd_asb_recover_0p(mdev);
2300 if (hg == -1 && mdev->state.role == R_SECONDARY)
2302 if (hg == 1 && mdev->state.role == R_PRIMARY)
2306 rv = drbd_asb_recover_0p(mdev);
2308 case ASB_DISCARD_SECONDARY:
2309 return mdev->state.role == R_PRIMARY ? 1 : -1;
2310 case ASB_CALL_HELPER:
2311 hg = drbd_asb_recover_0p(mdev);
2312 if (hg == -1 && mdev->state.role == R_PRIMARY) {
2313 self = drbd_set_role(mdev, R_SECONDARY, 0);
2314 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2315 * we might be here in C_WF_REPORT_PARAMS which is transient.
2316 * we do not need to wait for the after state change work either. */
2317 self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2318 if (self != SS_SUCCESS) {
2319 drbd_khelper(mdev, "pri-lost-after-sb");
2321 dev_warn(DEV, "Successfully gave up primary role.\n");
2331 static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local)
2333 int self, peer, hg, rv = -100;
2335 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2336 peer = mdev->p_uuid[UI_BITMAP] & 1;
2338 switch (mdev->net_conf->after_sb_2p) {
2339 case ASB_DISCARD_YOUNGER_PRI:
2340 case ASB_DISCARD_OLDER_PRI:
2341 case ASB_DISCARD_LEAST_CHG:
2342 case ASB_DISCARD_LOCAL:
2343 case ASB_DISCARD_REMOTE:
2345 case ASB_DISCARD_SECONDARY:
2346 dev_err(DEV, "Configuration error.\n");
2349 rv = drbd_asb_recover_0p(mdev);
2351 case ASB_DISCONNECT:
2353 case ASB_CALL_HELPER:
2354 hg = drbd_asb_recover_0p(mdev);
2356 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2357 * we might be here in C_WF_REPORT_PARAMS which is transient.
2358 * we do not need to wait for the after state change work either. */
2359 self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2360 if (self != SS_SUCCESS) {
2361 drbd_khelper(mdev, "pri-lost-after-sb");
2363 dev_warn(DEV, "Successfully gave up primary role.\n");
2373 static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid,
2374 u64 bits, u64 flags)
2377 dev_info(DEV, "%s uuid info vanished while I was looking!\n", text);
2380 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n",
2382 (unsigned long long)uuid[UI_CURRENT],
2383 (unsigned long long)uuid[UI_BITMAP],
2384 (unsigned long long)uuid[UI_HISTORY_START],
2385 (unsigned long long)uuid[UI_HISTORY_END],
2386 (unsigned long long)bits,
2387 (unsigned long long)flags);
2391 100 after split brain try auto recover
2392 2 C_SYNC_SOURCE set BitMap
2393 1 C_SYNC_SOURCE use BitMap
2395 -1 C_SYNC_TARGET use BitMap
2396 -2 C_SYNC_TARGET set BitMap
2397 -100 after split brain, disconnect
2398 -1000 unrelated data
2400 static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local)
2405 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2406 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2409 if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED)
2413 if ((self == UUID_JUST_CREATED || self == (u64)0) &&
2414 peer != UUID_JUST_CREATED)
2418 if (self != UUID_JUST_CREATED &&
2419 (peer == UUID_JUST_CREATED || peer == (u64)0))
2423 int rct, dc; /* roles at crash time */
2425 if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) {
2427 if (mdev->agreed_pro_version < 91)
2430 if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) &&
2431 (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) {
2432 dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n");
2433 drbd_uuid_set_bm(mdev, 0UL);
2435 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2436 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2439 dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n");
2446 if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) {
2448 if (mdev->agreed_pro_version < 91)
2451 if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) &&
2452 (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) {
2453 dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n");
2455 mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START];
2456 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP];
2457 mdev->p_uuid[UI_BITMAP] = 0UL;
2459 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2462 dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n");
2469 /* Common power [off|failure] */
2470 rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) +
2471 (mdev->p_uuid[UI_FLAGS] & 2);
2472 /* lowest bit is set when we were primary,
2473 * next bit (weight 2) is set when peer was primary */
2477 case 0: /* !self_pri && !peer_pri */ return 0;
2478 case 1: /* self_pri && !peer_pri */ return 1;
2479 case 2: /* !self_pri && peer_pri */ return -1;
2480 case 3: /* self_pri && peer_pri */
2481 dc = test_bit(DISCARD_CONCURRENT, &mdev->flags);
2487 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2492 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2494 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2495 peer = mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1);
2497 /* The last P_SYNC_UUID did not get though. Undo the last start of
2498 resync as sync source modifications of the peer's UUIDs. */
2500 if (mdev->agreed_pro_version < 91)
2503 mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START];
2504 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1];
2510 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2511 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2512 peer = mdev->p_uuid[i] & ~((u64)1);
2518 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2519 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2524 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2526 self = mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1);
2527 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2529 /* The last P_SYNC_UUID did not get though. Undo the last start of
2530 resync as sync source modifications of our UUIDs. */
2532 if (mdev->agreed_pro_version < 91)
2535 _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]);
2536 _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]);
2538 dev_info(DEV, "Undid last start of resync:\n");
2540 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2541 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2549 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2550 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2551 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2557 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2558 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2559 if (self == peer && self != ((u64)0))
2563 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2564 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2565 for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) {
2566 peer = mdev->p_uuid[j] & ~((u64)1);
2575 /* drbd_sync_handshake() returns the new conn state on success, or
2576 CONN_MASK (-1) on failure.
2578 static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role,
2579 enum drbd_disk_state peer_disk) __must_hold(local)
2582 enum drbd_conns rv = C_MASK;
2583 enum drbd_disk_state mydisk;
2585 mydisk = mdev->state.disk;
2586 if (mydisk == D_NEGOTIATING)
2587 mydisk = mdev->new_state_tmp.disk;
2589 dev_info(DEV, "drbd_sync_handshake:\n");
2590 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0);
2591 drbd_uuid_dump(mdev, "peer", mdev->p_uuid,
2592 mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2594 hg = drbd_uuid_compare(mdev, &rule_nr);
2596 dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr);
2599 dev_alert(DEV, "Unrelated data, aborting!\n");
2603 dev_alert(DEV, "To resolve this both sides have to support at least protocol\n");
2607 if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) ||
2608 (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) {
2609 int f = (hg == -100) || abs(hg) == 2;
2610 hg = mydisk > D_INCONSISTENT ? 1 : -1;
2613 dev_info(DEV, "Becoming sync %s due to disk states.\n",
2614 hg > 0 ? "source" : "target");
2618 drbd_khelper(mdev, "initial-split-brain");
2620 if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) {
2621 int pcount = (mdev->state.role == R_PRIMARY)
2622 + (peer_role == R_PRIMARY);
2623 int forced = (hg == -100);
2627 hg = drbd_asb_recover_0p(mdev);
2630 hg = drbd_asb_recover_1p(mdev);
2633 hg = drbd_asb_recover_2p(mdev);
2636 if (abs(hg) < 100) {
2637 dev_warn(DEV, "Split-Brain detected, %d primaries, "
2638 "automatically solved. Sync from %s node\n",
2639 pcount, (hg < 0) ? "peer" : "this");
2641 dev_warn(DEV, "Doing a full sync, since"
2642 " UUIDs where ambiguous.\n");
2649 if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1))
2651 if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1))
2655 dev_warn(DEV, "Split-Brain detected, manually solved. "
2656 "Sync from %s node\n",
2657 (hg < 0) ? "peer" : "this");
2661 /* FIXME this log message is not correct if we end up here
2662 * after an attempted attach on a diskless node.
2663 * We just refuse to attach -- well, we drop the "connection"
2664 * to that disk, in a way... */
2665 dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n");
2666 drbd_khelper(mdev, "split-brain");
2670 if (hg > 0 && mydisk <= D_INCONSISTENT) {
2671 dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n");
2675 if (hg < 0 && /* by intention we do not use mydisk here. */
2676 mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) {
2677 switch (mdev->net_conf->rr_conflict) {
2678 case ASB_CALL_HELPER:
2679 drbd_khelper(mdev, "pri-lost");
2681 case ASB_DISCONNECT:
2682 dev_err(DEV, "I shall become SyncTarget, but I am primary!\n");
2685 dev_warn(DEV, "Becoming SyncTarget, violating the stable-data"
2690 if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) {
2692 dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n");
2694 dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.",
2695 drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET),
2696 abs(hg) >= 2 ? "full" : "bit-map based");
2701 dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n");
2702 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake"))
2706 if (hg > 0) { /* become sync source. */
2708 } else if (hg < 0) { /* become sync target */
2712 if (drbd_bm_total_weight(mdev)) {
2713 dev_info(DEV, "No resync, but %lu bits in bitmap!\n",
2714 drbd_bm_total_weight(mdev));
2721 /* returns 1 if invalid */
2722 static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self)
2724 /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */
2725 if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) ||
2726 (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL))
2729 /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */
2730 if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL ||
2731 self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL)
2734 /* everything else is valid if they are equal on both sides. */
2738 /* everything es is invalid. */
2742 static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
2744 struct p_protocol *p = &mdev->data.rbuf.protocol;
2745 int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p;
2746 int p_want_lose, p_two_primaries, cf;
2747 char p_integrity_alg[SHARED_SECRET_MAX] = "";
2749 p_proto = be32_to_cpu(p->protocol);
2750 p_after_sb_0p = be32_to_cpu(p->after_sb_0p);
2751 p_after_sb_1p = be32_to_cpu(p->after_sb_1p);
2752 p_after_sb_2p = be32_to_cpu(p->after_sb_2p);
2753 p_two_primaries = be32_to_cpu(p->two_primaries);
2754 cf = be32_to_cpu(p->conn_flags);
2755 p_want_lose = cf & CF_WANT_LOSE;
2757 clear_bit(CONN_DRY_RUN, &mdev->flags);
2759 if (cf & CF_DRY_RUN)
2760 set_bit(CONN_DRY_RUN, &mdev->flags);
2762 if (p_proto != mdev->net_conf->wire_protocol) {
2763 dev_err(DEV, "incompatible communication protocols\n");
2767 if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) {
2768 dev_err(DEV, "incompatible after-sb-0pri settings\n");
2772 if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) {
2773 dev_err(DEV, "incompatible after-sb-1pri settings\n");
2777 if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) {
2778 dev_err(DEV, "incompatible after-sb-2pri settings\n");
2782 if (p_want_lose && mdev->net_conf->want_lose) {
2783 dev_err(DEV, "both sides have the 'want_lose' flag set\n");
2787 if (p_two_primaries != mdev->net_conf->two_primaries) {
2788 dev_err(DEV, "incompatible setting of the two-primaries options\n");
2792 if (mdev->agreed_pro_version >= 87) {
2793 unsigned char *my_alg = mdev->net_conf->integrity_alg;
2795 if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size)
2798 p_integrity_alg[SHARED_SECRET_MAX-1] = 0;
2799 if (strcmp(p_integrity_alg, my_alg)) {
2800 dev_err(DEV, "incompatible setting of the data-integrity-alg\n");
2803 dev_info(DEV, "data-integrity-alg: %s\n",
2804 my_alg[0] ? my_alg : (unsigned char *)"<not-used>");
2810 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2815 * input: alg name, feature name
2816 * return: NULL (alg name was "")
2817 * ERR_PTR(error) if something goes wrong
2818 * or the crypto hash ptr, if it worked out ok. */
2819 struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev,
2820 const char *alg, const char *name)
2822 struct crypto_hash *tfm;
2827 tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC);
2829 dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n",
2830 alg, name, PTR_ERR(tfm));
2833 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
2834 crypto_free_hash(tfm);
2835 dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name);
2836 return ERR_PTR(-EINVAL);
2841 static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size)
2844 struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95;
2845 unsigned int header_size, data_size, exp_max_sz;
2846 struct crypto_hash *verify_tfm = NULL;
2847 struct crypto_hash *csums_tfm = NULL;
2848 const int apv = mdev->agreed_pro_version;
2849 int *rs_plan_s = NULL;
2852 exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param)
2853 : apv == 88 ? sizeof(struct p_rs_param)
2855 : apv <= 94 ? sizeof(struct p_rs_param_89)
2856 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
2858 if (packet_size > exp_max_sz) {
2859 dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n",
2860 packet_size, exp_max_sz);
2865 header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80);
2866 data_size = packet_size - header_size;
2867 } else if (apv <= 94) {
2868 header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80);
2869 data_size = packet_size - header_size;
2870 D_ASSERT(data_size == 0);
2872 header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80);
2873 data_size = packet_size - header_size;
2874 D_ASSERT(data_size == 0);
2877 /* initialize verify_alg and csums_alg */
2878 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
2880 if (drbd_recv(mdev, &p->head.payload, header_size) != header_size)
2883 mdev->sync_conf.rate = be32_to_cpu(p->rate);
2887 if (data_size > SHARED_SECRET_MAX) {
2888 dev_err(DEV, "verify-alg too long, "
2889 "peer wants %u, accepting only %u byte\n",
2890 data_size, SHARED_SECRET_MAX);
2894 if (drbd_recv(mdev, p->verify_alg, data_size) != data_size)
2897 /* we expect NUL terminated string */
2898 /* but just in case someone tries to be evil */
2899 D_ASSERT(p->verify_alg[data_size-1] == 0);
2900 p->verify_alg[data_size-1] = 0;
2902 } else /* apv >= 89 */ {
2903 /* we still expect NUL terminated strings */
2904 /* but just in case someone tries to be evil */
2905 D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0);
2906 D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0);
2907 p->verify_alg[SHARED_SECRET_MAX-1] = 0;
2908 p->csums_alg[SHARED_SECRET_MAX-1] = 0;
2911 if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) {
2912 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2913 dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n",
2914 mdev->sync_conf.verify_alg, p->verify_alg);
2917 verify_tfm = drbd_crypto_alloc_digest_safe(mdev,
2918 p->verify_alg, "verify-alg");
2919 if (IS_ERR(verify_tfm)) {
2925 if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) {
2926 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2927 dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n",
2928 mdev->sync_conf.csums_alg, p->csums_alg);
2931 csums_tfm = drbd_crypto_alloc_digest_safe(mdev,
2932 p->csums_alg, "csums-alg");
2933 if (IS_ERR(csums_tfm)) {
2940 mdev->sync_conf.rate = be32_to_cpu(p->rate);
2941 mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead);
2942 mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target);
2943 mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target);
2944 mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate);
2946 fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
2947 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
2948 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
2950 dev_err(DEV, "kmalloc of fifo_buffer failed");
2956 spin_lock(&mdev->peer_seq_lock);
2957 /* lock against drbd_nl_syncer_conf() */
2959 strcpy(mdev->sync_conf.verify_alg, p->verify_alg);
2960 mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1;
2961 crypto_free_hash(mdev->verify_tfm);
2962 mdev->verify_tfm = verify_tfm;
2963 dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg);
2966 strcpy(mdev->sync_conf.csums_alg, p->csums_alg);
2967 mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1;
2968 crypto_free_hash(mdev->csums_tfm);
2969 mdev->csums_tfm = csums_tfm;
2970 dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg);
2972 if (fifo_size != mdev->rs_plan_s.size) {
2973 kfree(mdev->rs_plan_s.values);
2974 mdev->rs_plan_s.values = rs_plan_s;
2975 mdev->rs_plan_s.size = fifo_size;
2976 mdev->rs_planed = 0;
2978 spin_unlock(&mdev->peer_seq_lock);
2983 /* just for completeness: actually not needed,
2984 * as this is not reached if csums_tfm was ok. */
2985 crypto_free_hash(csums_tfm);
2986 /* but free the verify_tfm again, if csums_tfm did not work out */
2987 crypto_free_hash(verify_tfm);
2988 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2992 static void drbd_setup_order_type(struct drbd_conf *mdev, int peer)
2994 /* sorry, we currently have no working implementation
2995 * of distributed TCQ */
2998 /* warn if the arguments differ by more than 12.5% */
2999 static void warn_if_differ_considerably(struct drbd_conf *mdev,
3000 const char *s, sector_t a, sector_t b)
3003 if (a == 0 || b == 0)
3005 d = (a > b) ? (a - b) : (b - a);
3006 if (d > (a>>3) || d > (b>>3))
3007 dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s,
3008 (unsigned long long)a, (unsigned long long)b);
3011 static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3013 struct p_sizes *p = &mdev->data.rbuf.sizes;
3014 enum determine_dev_size dd = unchanged;
3015 unsigned int max_seg_s;
3016 sector_t p_size, p_usize, my_usize;
3017 int ldsc = 0; /* local disk size changed */
3018 enum dds_flags ddsf;
3020 p_size = be64_to_cpu(p->d_size);
3021 p_usize = be64_to_cpu(p->u_size);
3023 if (p_size == 0 && mdev->state.disk == D_DISKLESS) {
3024 dev_err(DEV, "some backing storage is needed\n");
3025 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3029 /* just store the peer's disk size for now.
3030 * we still need to figure out whether we accept that. */
3031 mdev->p_size = p_size;
3033 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
3034 if (get_ldev(mdev)) {
3035 warn_if_differ_considerably(mdev, "lower level device sizes",
3036 p_size, drbd_get_max_capacity(mdev->ldev));
3037 warn_if_differ_considerably(mdev, "user requested size",
3038 p_usize, mdev->ldev->dc.disk_size);
3040 /* if this is the first connect, or an otherwise expected
3041 * param exchange, choose the minimum */
3042 if (mdev->state.conn == C_WF_REPORT_PARAMS)
3043 p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size,
3046 my_usize = mdev->ldev->dc.disk_size;
3048 if (mdev->ldev->dc.disk_size != p_usize) {
3049 mdev->ldev->dc.disk_size = p_usize;
3050 dev_info(DEV, "Peer sets u_size to %lu sectors\n",
3051 (unsigned long)mdev->ldev->dc.disk_size);
3054 /* Never shrink a device with usable data during connect.
3055 But allow online shrinking if we are connected. */
3056 if (drbd_new_dev_size(mdev, mdev->ldev, 0) <
3057 drbd_get_capacity(mdev->this_bdev) &&
3058 mdev->state.disk >= D_OUTDATED &&
3059 mdev->state.conn < C_CONNECTED) {
3060 dev_err(DEV, "The peer's disk size is too small!\n");
3061 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3062 mdev->ldev->dc.disk_size = my_usize;
3070 ddsf = be16_to_cpu(p->dds_flags);
3071 if (get_ldev(mdev)) {
3072 dd = drbd_determin_dev_size(mdev, ddsf);
3074 if (dd == dev_size_error)
3078 /* I am diskless, need to accept the peer's size. */
3079 drbd_set_my_capacity(mdev, p_size);
3082 if (get_ldev(mdev)) {
3083 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) {
3084 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
3088 if (mdev->agreed_pro_version < 94)
3089 max_seg_s = be32_to_cpu(p->max_segment_size);
3090 else /* drbd 8.3.8 onwards */
3091 max_seg_s = DRBD_MAX_SEGMENT_SIZE;
3093 if (max_seg_s != queue_max_segment_size(mdev->rq_queue))
3094 drbd_setup_queue_param(mdev, max_seg_s);
3096 drbd_setup_order_type(mdev, be16_to_cpu(p->queue_order_type));
3100 if (mdev->state.conn > C_WF_REPORT_PARAMS) {
3101 if (be64_to_cpu(p->c_size) !=
3102 drbd_get_capacity(mdev->this_bdev) || ldsc) {
3103 /* we have different sizes, probably peer
3104 * needs to know my new size... */
3105 drbd_send_sizes(mdev, 0, ddsf);
3107 if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) ||
3108 (dd == grew && mdev->state.conn == C_CONNECTED)) {
3109 if (mdev->state.pdsk >= D_INCONSISTENT &&
3110 mdev->state.disk >= D_INCONSISTENT) {
3111 if (ddsf & DDSF_NO_RESYNC)
3112 dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n");
3114 resync_after_online_grow(mdev);
3116 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
3123 static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3125 struct p_uuids *p = &mdev->data.rbuf.uuids;
3129 p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO);
3131 for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++)
3132 p_uuid[i] = be64_to_cpu(p->uuid[i]);
3134 kfree(mdev->p_uuid);
3135 mdev->p_uuid = p_uuid;
3137 if (mdev->state.conn < C_CONNECTED &&
3138 mdev->state.disk < D_INCONSISTENT &&
3139 mdev->state.role == R_PRIMARY &&
3140 (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) {
3141 dev_err(DEV, "Can only connect to data with current UUID=%016llX\n",
3142 (unsigned long long)mdev->ed_uuid);
3143 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3147 if (get_ldev(mdev)) {
3148 int skip_initial_sync =
3149 mdev->state.conn == C_CONNECTED &&
3150 mdev->agreed_pro_version >= 90 &&
3151 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
3152 (p_uuid[UI_FLAGS] & 8);
3153 if (skip_initial_sync) {
3154 dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n");
3155 drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
3156 "clear_n_write from receive_uuids");
3157 _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]);
3158 _drbd_uuid_set(mdev, UI_BITMAP, 0);
3159 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3164 } else if (mdev->state.disk < D_INCONSISTENT &&
3165 mdev->state.role == R_PRIMARY) {
3166 /* I am a diskless primary, the peer just created a new current UUID
3168 drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3171 /* Before we test for the disk state, we should wait until an eventually
3172 ongoing cluster wide state change is finished. That is important if
3173 we are primary and are detaching from our disk. We need to see the
3174 new disk state... */
3175 wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags));
3176 if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT)
3177 drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3183 * convert_state() - Converts the peer's view of the cluster state to our point of view
3184 * @ps: The state as seen by the peer.
3186 static union drbd_state convert_state(union drbd_state ps)
3188 union drbd_state ms;
3190 static enum drbd_conns c_tab[] = {
3191 [C_CONNECTED] = C_CONNECTED,
3193 [C_STARTING_SYNC_S] = C_STARTING_SYNC_T,
3194 [C_STARTING_SYNC_T] = C_STARTING_SYNC_S,
3195 [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */
3196 [C_VERIFY_S] = C_VERIFY_T,
3202 ms.conn = c_tab[ps.conn];
3207 ms.peer_isp = (ps.aftr_isp | ps.user_isp);
3212 static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3214 struct p_req_state *p = &mdev->data.rbuf.req_state;
3215 union drbd_state mask, val;
3218 mask.i = be32_to_cpu(p->mask);
3219 val.i = be32_to_cpu(p->val);
3221 if (test_bit(DISCARD_CONCURRENT, &mdev->flags) &&
3222 test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) {
3223 drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG);
3227 mask = convert_state(mask);
3228 val = convert_state(val);
3230 rv = drbd_change_state(mdev, CS_VERBOSE, mask, val);
3232 drbd_send_sr_reply(mdev, rv);
3238 static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3240 struct p_state *p = &mdev->data.rbuf.state;
3241 enum drbd_conns nconn, oconn;
3242 union drbd_state ns, peer_state;
3243 enum drbd_disk_state real_peer_disk;
3244 enum chg_state_flags cs_flags;
3247 peer_state.i = be32_to_cpu(p->state);
3249 real_peer_disk = peer_state.disk;
3250 if (peer_state.disk == D_NEGOTIATING) {
3251 real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT;
3252 dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk));
3255 spin_lock_irq(&mdev->req_lock);
3257 oconn = nconn = mdev->state.conn;
3258 spin_unlock_irq(&mdev->req_lock);
3260 if (nconn == C_WF_REPORT_PARAMS)
3261 nconn = C_CONNECTED;
3263 if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING &&
3264 get_ldev_if_state(mdev, D_NEGOTIATING)) {
3265 int cr; /* consider resync */
3267 /* if we established a new connection */
3268 cr = (oconn < C_CONNECTED);
3269 /* if we had an established connection
3270 * and one of the nodes newly attaches a disk */
3271 cr |= (oconn == C_CONNECTED &&
3272 (peer_state.disk == D_NEGOTIATING ||
3273 mdev->state.disk == D_NEGOTIATING));
3274 /* if we have both been inconsistent, and the peer has been
3275 * forced to be UpToDate with --overwrite-data */
3276 cr |= test_bit(CONSIDER_RESYNC, &mdev->flags);
3277 /* if we had been plain connected, and the admin requested to
3278 * start a sync by "invalidate" or "invalidate-remote" */
3279 cr |= (oconn == C_CONNECTED &&
3280 (peer_state.conn >= C_STARTING_SYNC_S &&
3281 peer_state.conn <= C_WF_BITMAP_T));
3284 nconn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk);
3287 if (nconn == C_MASK) {
3288 nconn = C_CONNECTED;
3289 if (mdev->state.disk == D_NEGOTIATING) {
3290 drbd_force_state(mdev, NS(disk, D_DISKLESS));
3291 } else if (peer_state.disk == D_NEGOTIATING) {
3292 dev_err(DEV, "Disk attach process on the peer node was aborted.\n");
3293 peer_state.disk = D_DISKLESS;
3294 real_peer_disk = D_DISKLESS;
3296 if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags))
3298 D_ASSERT(oconn == C_WF_REPORT_PARAMS);
3299 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3305 spin_lock_irq(&mdev->req_lock);
3306 if (mdev->state.conn != oconn)
3308 clear_bit(CONSIDER_RESYNC, &mdev->flags);
3309 ns.i = mdev->state.i;
3311 ns.peer = peer_state.role;
3312 ns.pdsk = real_peer_disk;
3313 ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp);
3314 if ((nconn == C_CONNECTED || nconn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING)
3315 ns.disk = mdev->new_state_tmp.disk;
3316 cs_flags = CS_VERBOSE + (oconn < C_CONNECTED && nconn >= C_CONNECTED ? 0 : CS_HARD);
3317 if (ns.pdsk == D_CONSISTENT && ns.susp && nconn == C_CONNECTED && oconn < C_CONNECTED &&
3318 test_bit(NEW_CUR_UUID, &mdev->flags)) {
3319 /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this
3320 for temporal network outages! */
3321 spin_unlock_irq(&mdev->req_lock);
3322 dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n");
3324 drbd_uuid_new_current(mdev);
3325 clear_bit(NEW_CUR_UUID, &mdev->flags);
3326 drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0));
3329 rv = _drbd_set_state(mdev, ns, cs_flags, NULL);
3331 spin_unlock_irq(&mdev->req_lock);
3333 if (rv < SS_SUCCESS) {
3334 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3338 if (oconn > C_WF_REPORT_PARAMS) {
3339 if (nconn > C_CONNECTED && peer_state.conn <= C_CONNECTED &&
3340 peer_state.disk != D_NEGOTIATING ) {
3341 /* we want resync, peer has not yet decided to sync... */
3342 /* Nowadays only used when forcing a node into primary role and
3343 setting its disk to UpToDate with that */
3344 drbd_send_uuids(mdev);
3345 drbd_send_state(mdev);
3349 mdev->net_conf->want_lose = 0;
3351 drbd_md_sync(mdev); /* update connected indicator, la_size, ... */
3356 static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3358 struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid;
3360 wait_event(mdev->misc_wait,
3361 mdev->state.conn == C_WF_SYNC_UUID ||
3362 mdev->state.conn < C_CONNECTED ||
3363 mdev->state.disk < D_NEGOTIATING);
3365 /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */
3367 /* Here the _drbd_uuid_ functions are right, current should
3368 _not_ be rotated into the history */
3369 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
3370 _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid));
3371 _drbd_uuid_set(mdev, UI_BITMAP, 0UL);
3373 drbd_start_resync(mdev, C_SYNC_TARGET);
3377 dev_err(DEV, "Ignoring SyncUUID packet!\n");
3382 enum receive_bitmap_ret { OK, DONE, FAILED };
3384 static enum receive_bitmap_ret
3385 receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size,
3386 unsigned long *buffer, struct bm_xfer_ctx *c)
3388 unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset);
3389 unsigned want = num_words * sizeof(long);
3391 if (want != data_size) {
3392 dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size);
3397 if (drbd_recv(mdev, buffer, want) != want)
3400 drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer);
3402 c->word_offset += num_words;
3403 c->bit_offset = c->word_offset * BITS_PER_LONG;
3404 if (c->bit_offset > c->bm_bits)
3405 c->bit_offset = c->bm_bits;
3410 static enum receive_bitmap_ret
3411 recv_bm_rle_bits(struct drbd_conf *mdev,
3412 struct p_compressed_bm *p,
3413 struct bm_xfer_ctx *c)
3415 struct bitstream bs;
3419 unsigned long s = c->bit_offset;
3421 int len = p->head.length - (sizeof(*p) - sizeof(p->head));
3422 int toggle = DCBP_get_start(p);
3426 bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p));
3428 bits = bitstream_get_bits(&bs, &look_ahead, 64);
3432 for (have = bits; have > 0; s += rl, toggle = !toggle) {
3433 bits = vli_decode_bits(&rl, look_ahead);
3439 if (e >= c->bm_bits) {
3440 dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e);
3443 _drbd_bm_set_bits(mdev, s, e);
3447 dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n",
3448 have, bits, look_ahead,
3449 (unsigned int)(bs.cur.b - p->code),
3450 (unsigned int)bs.buf_len);
3453 look_ahead >>= bits;
3456 bits = bitstream_get_bits(&bs, &tmp, 64 - have);
3459 look_ahead |= tmp << have;
3464 bm_xfer_ctx_bit_to_word_offset(c);
3466 return (s == c->bm_bits) ? DONE : OK;
3469 static enum receive_bitmap_ret
3470 decode_bitmap_c(struct drbd_conf *mdev,
3471 struct p_compressed_bm *p,
3472 struct bm_xfer_ctx *c)
3474 if (DCBP_get_code(p) == RLE_VLI_Bits)
3475 return recv_bm_rle_bits(mdev, p, c);
3477 /* other variants had been implemented for evaluation,
3478 * but have been dropped as this one turned out to be "best"
3479 * during all our tests. */
3481 dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding);
3482 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3486 void INFO_bm_xfer_stats(struct drbd_conf *mdev,
3487 const char *direction, struct bm_xfer_ctx *c)
3489 /* what would it take to transfer it "plaintext" */
3490 unsigned plain = sizeof(struct p_header80) *
3491 ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1)
3492 + c->bm_words * sizeof(long);
3493 unsigned total = c->bytes[0] + c->bytes[1];
3496 /* total can not be zero. but just in case: */
3500 /* don't report if not compressed */
3504 /* total < plain. check for overflow, still */
3505 r = (total > UINT_MAX/1000) ? (total / (plain/1000))
3506 : (1000 * total / plain);
3512 dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), "
3513 "total %u; compression: %u.%u%%\n",
3515 c->bytes[1], c->packets[1],
3516 c->bytes[0], c->packets[0],
3517 total, r/10, r % 10);
3520 /* Since we are processing the bitfield from lower addresses to higher,
3521 it does not matter if the process it in 32 bit chunks or 64 bit
3522 chunks as long as it is little endian. (Understand it as byte stream,
3523 beginning with the lowest byte...) If we would use big endian
3524 we would need to process it from the highest address to the lowest,
3525 in order to be agnostic to the 32 vs 64 bits issue.
3527 returns 0 on failure, 1 if we successfully received it. */
3528 static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3530 struct bm_xfer_ctx c;
3532 enum receive_bitmap_ret ret;
3534 struct p_header80 *h = &mdev->data.rbuf.header.h80;
3536 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
3538 drbd_bm_lock(mdev, "receive bitmap");
3540 /* maybe we should use some per thread scratch page,
3541 * and allocate that during initial device creation? */
3542 buffer = (unsigned long *) __get_free_page(GFP_NOIO);
3544 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__);
3548 c = (struct bm_xfer_ctx) {
3549 .bm_bits = drbd_bm_bits(mdev),
3550 .bm_words = drbd_bm_words(mdev),
3554 if (cmd == P_BITMAP) {
3555 ret = receive_bitmap_plain(mdev, data_size, buffer, &c);
3556 } else if (cmd == P_COMPRESSED_BITMAP) {
3557 /* MAYBE: sanity check that we speak proto >= 90,
3558 * and the feature is enabled! */
3559 struct p_compressed_bm *p;
3561 if (data_size > BM_PACKET_PAYLOAD_BYTES) {
3562 dev_err(DEV, "ReportCBitmap packet too large\n");
3565 /* use the page buff */
3567 memcpy(p, h, sizeof(*h));
3568 if (drbd_recv(mdev, p->head.payload, data_size) != data_size)
3570 if (p->head.length <= (sizeof(*p) - sizeof(p->head))) {
3571 dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", p->head.length);
3574 ret = decode_bitmap_c(mdev, p, &c);
3576 dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd);
3580 c.packets[cmd == P_BITMAP]++;
3581 c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size;
3586 if (!drbd_recv_header(mdev, &cmd, &data_size))
3588 } while (ret == OK);
3592 INFO_bm_xfer_stats(mdev, "receive", &c);
3594 if (mdev->state.conn == C_WF_BITMAP_T) {
3595 ok = !drbd_send_bitmap(mdev);
3598 /* Omit CS_ORDERED with this state transition to avoid deadlocks. */
3599 ok = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
3600 D_ASSERT(ok == SS_SUCCESS);
3601 } else if (mdev->state.conn != C_WF_BITMAP_S) {
3602 /* admin may have requested C_DISCONNECTING,
3603 * other threads may have noticed network errors */
3604 dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n",
3605 drbd_conn_str(mdev->state.conn));
3610 drbd_bm_unlock(mdev);
3611 if (ok && mdev->state.conn == C_WF_BITMAP_S)
3612 drbd_start_resync(mdev, C_SYNC_SOURCE);
3613 free_page((unsigned long) buffer);
3617 static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3619 /* TODO zero copy sink :) */
3620 static char sink[128];
3623 dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n",
3628 want = min_t(int, size, sizeof(sink));
3629 r = drbd_recv(mdev, sink, want);
3630 ERR_IF(r <= 0) break;
3636 static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3638 if (mdev->state.disk >= D_INCONSISTENT)
3641 /* Make sure we've acked all the TCP data associated
3642 * with the data requests being unplugged */
3643 drbd_tcp_quickack(mdev->data.socket);
3648 typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive);
3653 drbd_cmd_handler_f function;
3656 static struct data_cmd drbd_cmd_handler[] = {
3657 [P_DATA] = { 1, sizeof(struct p_data), receive_Data },
3658 [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply },
3659 [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } ,
3660 [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } ,
3661 [P_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } ,
3662 [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } ,
3663 [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header80), receive_UnplugRemote },
3664 [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3665 [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3666 [P_SYNC_PARAM] = { 1, sizeof(struct p_header80), receive_SyncParam },
3667 [P_SYNC_PARAM89] = { 1, sizeof(struct p_header80), receive_SyncParam },
3668 [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol },
3669 [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids },
3670 [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes },
3671 [P_STATE] = { 0, sizeof(struct p_state), receive_state },
3672 [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state },
3673 [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid },
3674 [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3675 [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest },
3676 [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest },
3677 [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip },
3678 /* anything missing from this table is in
3679 * the asender_tbl, see get_asender_cmd */
3680 [P_MAX_CMD] = { 0, 0, NULL },
3683 /* All handler functions that expect a sub-header get that sub-heder in
3684 mdev->data.rbuf.header.head.payload.
3686 Usually in mdev->data.rbuf.header.head the callback can find the usual
3687 p_header, but they may not rely on that. Since there is also p_header95 !
3690 static void drbdd(struct drbd_conf *mdev)
3692 union p_header *header = &mdev->data.rbuf.header;
3693 unsigned int packet_size;
3694 enum drbd_packets cmd;
3695 size_t shs; /* sub header size */
3698 while (get_t_state(&mdev->receiver) == Running) {
3699 drbd_thread_current_set_cpu(mdev);
3700 if (!drbd_recv_header(mdev, &cmd, &packet_size))
3703 if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) {
3704 dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size);
3708 shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header);
3709 rv = drbd_recv(mdev, &header->h80.payload, shs);
3710 if (unlikely(rv != shs)) {
3711 dev_err(DEV, "short read while reading sub header: rv=%d\n", rv);
3715 if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) {
3716 dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size);
3720 rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs);
3722 if (unlikely(!rv)) {
3723 dev_err(DEV, "error receiving %s, l: %d!\n",
3724 cmdname(cmd), packet_size);
3731 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3735 void drbd_flush_workqueue(struct drbd_conf *mdev)
3737 struct drbd_wq_barrier barr;
3739 barr.w.cb = w_prev_work_done;
3740 init_completion(&barr.done);
3741 drbd_queue_work(&mdev->data.work, &barr.w);
3742 wait_for_completion(&barr.done);
3745 void drbd_free_tl_hash(struct drbd_conf *mdev)
3747 struct hlist_head *h;
3749 spin_lock_irq(&mdev->req_lock);
3751 if (!mdev->tl_hash || mdev->state.conn != C_STANDALONE) {
3752 spin_unlock_irq(&mdev->req_lock);
3756 for (h = mdev->ee_hash; h < mdev->ee_hash + mdev->ee_hash_s; h++)
3758 dev_err(DEV, "ASSERT FAILED ee_hash[%u].first == %p, expected NULL\n",
3759 (int)(h - mdev->ee_hash), h->first);
3760 kfree(mdev->ee_hash);
3761 mdev->ee_hash = NULL;
3762 mdev->ee_hash_s = 0;
3765 for (h = mdev->tl_hash; h < mdev->tl_hash + mdev->tl_hash_s; h++)
3767 dev_err(DEV, "ASSERT FAILED tl_hash[%u] == %p, expected NULL\n",
3768 (int)(h - mdev->tl_hash), h->first);
3769 kfree(mdev->tl_hash);
3770 mdev->tl_hash = NULL;
3771 mdev->tl_hash_s = 0;
3772 spin_unlock_irq(&mdev->req_lock);
3775 static void drbd_disconnect(struct drbd_conf *mdev)
3777 enum drbd_fencing_p fp;
3778 union drbd_state os, ns;
3779 int rv = SS_UNKNOWN_ERROR;
3782 if (mdev->state.conn == C_STANDALONE)
3784 if (mdev->state.conn >= C_WF_CONNECTION)
3785 dev_err(DEV, "ASSERT FAILED cstate = %s, expected < WFConnection\n",
3786 drbd_conn_str(mdev->state.conn));
3788 /* asender does not clean up anything. it must not interfere, either */
3789 drbd_thread_stop(&mdev->asender);
3790 drbd_free_sock(mdev);
3792 /* wait for current activity to cease. */
3793 spin_lock_irq(&mdev->req_lock);
3794 _drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
3795 _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee);
3796 _drbd_wait_ee_list_empty(mdev, &mdev->read_ee);
3797 spin_unlock_irq(&mdev->req_lock);
3799 /* We do not have data structures that would allow us to
3800 * get the rs_pending_cnt down to 0 again.
3801 * * On C_SYNC_TARGET we do not have any data structures describing
3802 * the pending RSDataRequest's we have sent.
3803 * * On C_SYNC_SOURCE there is no data structure that tracks
3804 * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget.
3805 * And no, it is not the sum of the reference counts in the
3806 * resync_LRU. The resync_LRU tracks the whole operation including
3807 * the disk-IO, while the rs_pending_cnt only tracks the blocks
3809 drbd_rs_cancel_all(mdev);
3811 mdev->rs_failed = 0;
3812 atomic_set(&mdev->rs_pending_cnt, 0);
3813 wake_up(&mdev->misc_wait);
3815 /* make sure syncer is stopped and w_resume_next_sg queued */
3816 del_timer_sync(&mdev->resync_timer);
3817 resync_timer_fn((unsigned long)mdev);
3819 /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier,
3820 * w_make_resync_request etc. which may still be on the worker queue
3821 * to be "canceled" */
3822 drbd_flush_workqueue(mdev);
3824 /* This also does reclaim_net_ee(). If we do this too early, we might
3825 * miss some resync ee and pages.*/
3826 drbd_process_done_ee(mdev);
3828 kfree(mdev->p_uuid);
3829 mdev->p_uuid = NULL;
3831 if (!mdev->state.susp)
3834 dev_info(DEV, "Connection closed\n");
3839 if (get_ldev(mdev)) {
3840 fp = mdev->ldev->dc.fencing;
3844 if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN)
3845 drbd_try_outdate_peer_async(mdev);
3847 spin_lock_irq(&mdev->req_lock);
3849 if (os.conn >= C_UNCONNECTED) {
3850 /* Do not restart in case we are C_DISCONNECTING */
3852 ns.conn = C_UNCONNECTED;
3853 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
3855 spin_unlock_irq(&mdev->req_lock);
3857 if (os.conn == C_DISCONNECTING) {
3858 wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0);
3860 if (!mdev->state.susp) {
3861 /* we must not free the tl_hash
3862 * while application io is still on the fly */
3863 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
3864 drbd_free_tl_hash(mdev);
3867 crypto_free_hash(mdev->cram_hmac_tfm);
3868 mdev->cram_hmac_tfm = NULL;
3870 kfree(mdev->net_conf);
3871 mdev->net_conf = NULL;
3872 drbd_request_state(mdev, NS(conn, C_STANDALONE));
3875 /* tcp_close and release of sendpage pages can be deferred. I don't
3876 * want to use SO_LINGER, because apparently it can be deferred for
3877 * more than 20 seconds (longest time I checked).
3879 * Actually we don't care for exactly when the network stack does its
3880 * put_page(), but release our reference on these pages right here.
3882 i = drbd_release_ee(mdev, &mdev->net_ee);
3884 dev_info(DEV, "net_ee not empty, killed %u entries\n", i);
3885 i = atomic_read(&mdev->pp_in_use_by_net);
3887 dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i);
3888 i = atomic_read(&mdev->pp_in_use);
3890 dev_info(DEV, "pp_in_use = %d, expected 0\n", i);
3892 D_ASSERT(list_empty(&mdev->read_ee));
3893 D_ASSERT(list_empty(&mdev->active_ee));
3894 D_ASSERT(list_empty(&mdev->sync_ee));
3895 D_ASSERT(list_empty(&mdev->done_ee));
3897 /* ok, no more ee's on the fly, it is safe to reset the epoch_size */
3898 atomic_set(&mdev->current_epoch->epoch_size, 0);
3899 D_ASSERT(list_empty(&mdev->current_epoch->list));
3903 * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version
3904 * we can agree on is stored in agreed_pro_version.
3906 * feature flags and the reserved array should be enough room for future
3907 * enhancements of the handshake protocol, and possible plugins...
3909 * for now, they are expected to be zero, but ignored.
3911 static int drbd_send_handshake(struct drbd_conf *mdev)
3913 /* ASSERT current == mdev->receiver ... */
3914 struct p_handshake *p = &mdev->data.sbuf.handshake;
3917 if (mutex_lock_interruptible(&mdev->data.mutex)) {
3918 dev_err(DEV, "interrupted during initial handshake\n");
3919 return 0; /* interrupted. not ok. */
3922 if (mdev->data.socket == NULL) {
3923 mutex_unlock(&mdev->data.mutex);
3927 memset(p, 0, sizeof(*p));
3928 p->protocol_min = cpu_to_be32(PRO_VERSION_MIN);
3929 p->protocol_max = cpu_to_be32(PRO_VERSION_MAX);
3930 ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE,
3931 (struct p_header80 *)p, sizeof(*p), 0 );
3932 mutex_unlock(&mdev->data.mutex);
3938 * 1 yes, we have a valid connection
3939 * 0 oops, did not work out, please try again
3940 * -1 peer talks different language,
3941 * no point in trying again, please go standalone.
3943 static int drbd_do_handshake(struct drbd_conf *mdev)
3945 /* ASSERT current == mdev->receiver ... */
3946 struct p_handshake *p = &mdev->data.rbuf.handshake;
3947 const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80);
3948 unsigned int length;
3949 enum drbd_packets cmd;
3952 rv = drbd_send_handshake(mdev);
3956 rv = drbd_recv_header(mdev, &cmd, &length);
3960 if (cmd != P_HAND_SHAKE) {
3961 dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n",
3966 if (length != expect) {
3967 dev_err(DEV, "expected HandShake length: %u, received: %u\n",
3972 rv = drbd_recv(mdev, &p->head.payload, expect);
3975 dev_err(DEV, "short read receiving handshake packet: l=%u\n", rv);
3979 p->protocol_min = be32_to_cpu(p->protocol_min);
3980 p->protocol_max = be32_to_cpu(p->protocol_max);
3981 if (p->protocol_max == 0)
3982 p->protocol_max = p->protocol_min;
3984 if (PRO_VERSION_MAX < p->protocol_min ||
3985 PRO_VERSION_MIN > p->protocol_max)
3988 mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max);
3990 dev_info(DEV, "Handshake successful: "
3991 "Agreed network protocol version %d\n", mdev->agreed_pro_version);
3996 dev_err(DEV, "incompatible DRBD dialects: "
3997 "I support %d-%d, peer supports %d-%d\n",
3998 PRO_VERSION_MIN, PRO_VERSION_MAX,
3999 p->protocol_min, p->protocol_max);
4003 #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE)
4004 static int drbd_do_auth(struct drbd_conf *mdev)
4006 dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n");
4007 dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n");
4011 #define CHALLENGE_LEN 64
4015 0 - failed, try again (network error),
4016 -1 - auth failed, don't try again.
4019 static int drbd_do_auth(struct drbd_conf *mdev)
4021 char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */
4022 struct scatterlist sg;
4023 char *response = NULL;
4024 char *right_response = NULL;
4025 char *peers_ch = NULL;
4026 unsigned int key_len = strlen(mdev->net_conf->shared_secret);
4027 unsigned int resp_size;
4028 struct hash_desc desc;
4029 enum drbd_packets cmd;
4030 unsigned int length;
4033 desc.tfm = mdev->cram_hmac_tfm;
4036 rv = crypto_hash_setkey(mdev->cram_hmac_tfm,
4037 (u8 *)mdev->net_conf->shared_secret, key_len);
4039 dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv);
4044 get_random_bytes(my_challenge, CHALLENGE_LEN);
4046 rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN);
4050 rv = drbd_recv_header(mdev, &cmd, &length);
4054 if (cmd != P_AUTH_CHALLENGE) {
4055 dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n",
4061 if (length > CHALLENGE_LEN * 2) {
4062 dev_err(DEV, "expected AuthChallenge payload too big.\n");
4067 peers_ch = kmalloc(length, GFP_NOIO);
4068 if (peers_ch == NULL) {
4069 dev_err(DEV, "kmalloc of peers_ch failed\n");
4074 rv = drbd_recv(mdev, peers_ch, length);
4077 dev_err(DEV, "short read AuthChallenge: l=%u\n", rv);
4082 resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm);
4083 response = kmalloc(resp_size, GFP_NOIO);
4084 if (response == NULL) {
4085 dev_err(DEV, "kmalloc of response failed\n");
4090 sg_init_table(&sg, 1);
4091 sg_set_buf(&sg, peers_ch, length);
4093 rv = crypto_hash_digest(&desc, &sg, sg.length, response);
4095 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4100 rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size);
4104 rv = drbd_recv_header(mdev, &cmd, &length);
4108 if (cmd != P_AUTH_RESPONSE) {
4109 dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n",
4115 if (length != resp_size) {
4116 dev_err(DEV, "expected AuthResponse payload of wrong size\n");
4121 rv = drbd_recv(mdev, response , resp_size);
4123 if (rv != resp_size) {
4124 dev_err(DEV, "short read receiving AuthResponse: l=%u\n", rv);
4129 right_response = kmalloc(resp_size, GFP_NOIO);
4130 if (right_response == NULL) {
4131 dev_err(DEV, "kmalloc of right_response failed\n");
4136 sg_set_buf(&sg, my_challenge, CHALLENGE_LEN);
4138 rv = crypto_hash_digest(&desc, &sg, sg.length, right_response);
4140 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4145 rv = !memcmp(response, right_response, resp_size);
4148 dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n",
4149 resp_size, mdev->net_conf->cram_hmac_alg);
4156 kfree(right_response);
4162 int drbdd_init(struct drbd_thread *thi)
4164 struct drbd_conf *mdev = thi->mdev;
4165 unsigned int minor = mdev_to_minor(mdev);
4168 sprintf(current->comm, "drbd%d_receiver", minor);
4170 dev_info(DEV, "receiver (re)started\n");
4173 h = drbd_connect(mdev);
4175 drbd_disconnect(mdev);
4176 __set_current_state(TASK_INTERRUPTIBLE);
4177 schedule_timeout(HZ);
4180 dev_warn(DEV, "Discarding network configuration.\n");
4181 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4186 if (get_net_conf(mdev)) {
4192 drbd_disconnect(mdev);
4194 dev_info(DEV, "receiver terminated\n");
4198 /* ********* acknowledge sender ******** */
4200 static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h)
4202 struct p_req_state_reply *p = (struct p_req_state_reply *)h;
4204 int retcode = be32_to_cpu(p->retcode);
4206 if (retcode >= SS_SUCCESS) {
4207 set_bit(CL_ST_CHG_SUCCESS, &mdev->flags);
4209 set_bit(CL_ST_CHG_FAIL, &mdev->flags);
4210 dev_err(DEV, "Requested state change failed by peer: %s (%d)\n",
4211 drbd_set_st_err_str(retcode), retcode);
4213 wake_up(&mdev->state_wait);
4218 static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h)
4220 return drbd_send_ping_ack(mdev);
4224 static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h)
4226 /* restore idle timeout */
4227 mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
4228 if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags))
4229 wake_up(&mdev->misc_wait);
4234 static int got_IsInSync(struct drbd_conf *mdev, struct p_header80 *h)
4236 struct p_block_ack *p = (struct p_block_ack *)h;
4237 sector_t sector = be64_to_cpu(p->sector);
4238 int blksize = be32_to_cpu(p->blksize);
4240 D_ASSERT(mdev->agreed_pro_version >= 89);
4242 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4244 drbd_rs_complete_io(mdev, sector);
4245 drbd_set_in_sync(mdev, sector, blksize);
4246 /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */
4247 mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT);
4248 dec_rs_pending(mdev);
4249 atomic_add(blksize >> 9, &mdev->rs_sect_in);
4254 /* when we receive the ACK for a write request,
4255 * verify that we actually know about it */
4256 static struct drbd_request *_ack_id_to_req(struct drbd_conf *mdev,
4257 u64 id, sector_t sector)
4259 struct hlist_head *slot = tl_hash_slot(mdev, sector);
4260 struct hlist_node *n;
4261 struct drbd_request *req;
4263 hlist_for_each_entry(req, n, slot, colision) {
4264 if ((unsigned long)req == (unsigned long)id) {
4265 if (req->sector != sector) {
4266 dev_err(DEV, "_ack_id_to_req: found req %p but it has "
4267 "wrong sector (%llus versus %llus)\n", req,
4268 (unsigned long long)req->sector,
4269 (unsigned long long)sector);
4275 dev_err(DEV, "_ack_id_to_req: failed to find req %p, sector %llus in list\n",
4276 (void *)(unsigned long)id, (unsigned long long)sector);
4280 typedef struct drbd_request *(req_validator_fn)
4281 (struct drbd_conf *mdev, u64 id, sector_t sector);
4283 static int validate_req_change_req_state(struct drbd_conf *mdev,
4284 u64 id, sector_t sector, req_validator_fn validator,
4285 const char *func, enum drbd_req_event what)
4287 struct drbd_request *req;
4288 struct bio_and_error m;
4290 spin_lock_irq(&mdev->req_lock);
4291 req = validator(mdev, id, sector);
4292 if (unlikely(!req)) {
4293 spin_unlock_irq(&mdev->req_lock);
4294 dev_err(DEV, "%s: got a corrupt block_id/sector pair\n", func);
4297 __req_mod(req, what, &m);
4298 spin_unlock_irq(&mdev->req_lock);
4301 complete_master_bio(mdev, &m);
4305 static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h)
4307 struct p_block_ack *p = (struct p_block_ack *)h;
4308 sector_t sector = be64_to_cpu(p->sector);
4309 int blksize = be32_to_cpu(p->blksize);
4310 enum drbd_req_event what;
4312 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4314 if (is_syncer_block_id(p->block_id)) {
4315 drbd_set_in_sync(mdev, sector, blksize);
4316 dec_rs_pending(mdev);
4319 switch (be16_to_cpu(h->command)) {
4320 case P_RS_WRITE_ACK:
4321 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4322 what = write_acked_by_peer_and_sis;
4325 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4326 what = write_acked_by_peer;
4329 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B);
4330 what = recv_acked_by_peer;
4333 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4334 what = conflict_discarded_by_peer;
4341 return validate_req_change_req_state(mdev, p->block_id, sector,
4342 _ack_id_to_req, __func__ , what);
4345 static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h)
4347 struct p_block_ack *p = (struct p_block_ack *)h;
4348 sector_t sector = be64_to_cpu(p->sector);
4350 if (__ratelimit(&drbd_ratelimit_state))
4351 dev_warn(DEV, "Got NegAck packet. Peer is in troubles?\n");
4353 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4355 if (is_syncer_block_id(p->block_id)) {
4356 int size = be32_to_cpu(p->blksize);
4357 dec_rs_pending(mdev);
4358 drbd_rs_failed_io(mdev, sector, size);
4361 return validate_req_change_req_state(mdev, p->block_id, sector,
4362 _ack_id_to_req, __func__ , neg_acked);
4365 static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h)
4367 struct p_block_ack *p = (struct p_block_ack *)h;
4368 sector_t sector = be64_to_cpu(p->sector);
4370 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4371 dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n",
4372 (unsigned long long)sector, be32_to_cpu(p->blksize));
4374 return validate_req_change_req_state(mdev, p->block_id, sector,
4375 _ar_id_to_req, __func__ , neg_acked);
4378 static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h)
4382 struct p_block_ack *p = (struct p_block_ack *)h;
4384 sector = be64_to_cpu(p->sector);
4385 size = be32_to_cpu(p->blksize);
4387 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4389 dec_rs_pending(mdev);
4391 if (get_ldev_if_state(mdev, D_FAILED)) {
4392 drbd_rs_complete_io(mdev, sector);
4393 drbd_rs_failed_io(mdev, sector, size);
4400 static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h)
4402 struct p_barrier_ack *p = (struct p_barrier_ack *)h;
4404 tl_release(mdev, p->barrier, be32_to_cpu(p->set_size));
4409 static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h)
4411 struct p_block_ack *p = (struct p_block_ack *)h;
4412 struct drbd_work *w;
4416 sector = be64_to_cpu(p->sector);
4417 size = be32_to_cpu(p->blksize);
4419 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4421 if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC)
4422 drbd_ov_oos_found(mdev, sector, size);
4426 drbd_rs_complete_io(mdev, sector);
4427 dec_rs_pending(mdev);
4429 if (--mdev->ov_left == 0) {
4430 w = kmalloc(sizeof(*w), GFP_NOIO);
4432 w->cb = w_ov_finished;
4433 drbd_queue_work_front(&mdev->data.work, w);
4435 dev_err(DEV, "kmalloc(w) failed.");
4437 drbd_resync_finished(mdev);
4443 static int got_skip(struct drbd_conf *mdev, struct p_header80 *h)
4448 struct asender_cmd {
4450 int (*process)(struct drbd_conf *mdev, struct p_header80 *h);
4453 static struct asender_cmd *get_asender_cmd(int cmd)
4455 static struct asender_cmd asender_tbl[] = {
4456 /* anything missing from this table is in
4457 * the drbd_cmd_handler (drbd_default_handler) table,
4458 * see the beginning of drbdd() */
4459 [P_PING] = { sizeof(struct p_header80), got_Ping },
4460 [P_PING_ACK] = { sizeof(struct p_header80), got_PingAck },
4461 [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4462 [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4463 [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4464 [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4465 [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck },
4466 [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply },
4467 [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply},
4468 [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult },
4469 [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck },
4470 [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply },
4471 [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync },
4472 [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip },
4473 [P_MAX_CMD] = { 0, NULL },
4475 if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL)
4477 return &asender_tbl[cmd];
4480 int drbd_asender(struct drbd_thread *thi)
4482 struct drbd_conf *mdev = thi->mdev;
4483 struct p_header80 *h = &mdev->meta.rbuf.header.h80;
4484 struct asender_cmd *cmd = NULL;
4489 int expect = sizeof(struct p_header80);
4492 sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev));
4494 current->policy = SCHED_RR; /* Make this a realtime task! */
4495 current->rt_priority = 2; /* more important than all other tasks */
4497 while (get_t_state(thi) == Running) {
4498 drbd_thread_current_set_cpu(mdev);
4499 if (test_and_clear_bit(SEND_PING, &mdev->flags)) {
4500 ERR_IF(!drbd_send_ping(mdev)) goto reconnect;
4501 mdev->meta.socket->sk->sk_rcvtimeo =
4502 mdev->net_conf->ping_timeo*HZ/10;
4505 /* conditionally cork;
4506 * it may hurt latency if we cork without much to send */
4507 if (!mdev->net_conf->no_cork &&
4508 3 < atomic_read(&mdev->unacked_cnt))
4509 drbd_tcp_cork(mdev->meta.socket);
4511 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4512 flush_signals(current);
4513 if (!drbd_process_done_ee(mdev)) {
4514 dev_err(DEV, "process_done_ee() = NOT_OK\n");
4517 /* to avoid race with newly queued ACKs */
4518 set_bit(SIGNAL_ASENDER, &mdev->flags);
4519 spin_lock_irq(&mdev->req_lock);
4520 empty = list_empty(&mdev->done_ee);
4521 spin_unlock_irq(&mdev->req_lock);
4522 /* new ack may have been queued right here,
4523 * but then there is also a signal pending,
4524 * and we start over... */
4528 /* but unconditionally uncork unless disabled */
4529 if (!mdev->net_conf->no_cork)
4530 drbd_tcp_uncork(mdev->meta.socket);
4532 /* short circuit, recv_msg would return EINTR anyways. */
4533 if (signal_pending(current))
4536 rv = drbd_recv_short(mdev, mdev->meta.socket,
4537 buf, expect-received, 0);
4538 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4540 flush_signals(current);
4543 * -EINTR (on meta) we got a signal
4544 * -EAGAIN (on meta) rcvtimeo expired
4545 * -ECONNRESET other side closed the connection
4546 * -ERESTARTSYS (on data) we got a signal
4547 * rv < 0 other than above: unexpected error!
4548 * rv == expected: full header or command
4549 * rv < expected: "woken" by signal during receive
4550 * rv == 0 : "connection shut down by peer"
4552 if (likely(rv > 0)) {
4555 } else if (rv == 0) {
4556 dev_err(DEV, "meta connection shut down by peer.\n");
4558 } else if (rv == -EAGAIN) {
4559 if (mdev->meta.socket->sk->sk_rcvtimeo ==
4560 mdev->net_conf->ping_timeo*HZ/10) {
4561 dev_err(DEV, "PingAck did not arrive in time.\n");
4564 set_bit(SEND_PING, &mdev->flags);
4566 } else if (rv == -EINTR) {
4569 dev_err(DEV, "sock_recvmsg returned %d\n", rv);
4573 if (received == expect && cmd == NULL) {
4574 if (unlikely(h->magic != BE_DRBD_MAGIC)) {
4575 dev_err(DEV, "magic?? on meta m: 0x%lx c: %d l: %d\n",
4576 (long)be32_to_cpu(h->magic),
4577 h->command, h->length);
4580 cmd = get_asender_cmd(be16_to_cpu(h->command));
4581 len = be16_to_cpu(h->length);
4582 if (unlikely(cmd == NULL)) {
4583 dev_err(DEV, "unknown command?? on meta m: 0x%lx c: %d l: %d\n",
4584 (long)be32_to_cpu(h->magic),
4585 h->command, h->length);
4588 expect = cmd->pkt_size;
4589 ERR_IF(len != expect-sizeof(struct p_header80))
4592 if (received == expect) {
4593 D_ASSERT(cmd != NULL);
4594 if (!cmd->process(mdev, h))
4599 expect = sizeof(struct p_header80);
4606 drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE));
4610 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4612 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4614 D_ASSERT(mdev->state.conn < C_CONNECTED);
4615 dev_info(DEV, "asender terminated\n");