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%08x c: %d l: %d\n",
974 be32_to_cpu(h->h80.magic),
975 be16_to_cpu(h->h80.command),
976 be16_to_cpu(h->h80.length));
979 mdev->last_received = jiffies;
984 static enum finish_epoch drbd_flush_after_epoch(struct drbd_conf *mdev, struct drbd_epoch *epoch)
988 if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) {
989 rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL,
992 dev_err(DEV, "local disk flush failed with status %d\n", rv);
993 /* would rather check on EOPNOTSUPP, but that is not reliable.
994 * don't try again for ANY return value != 0
995 * if (rv == -EOPNOTSUPP) */
996 drbd_bump_write_ordering(mdev, WO_drain_io);
1001 return drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE);
1004 static int w_flush(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1006 struct flush_work *fw = (struct flush_work *)w;
1007 struct drbd_epoch *epoch = fw->epoch;
1011 if (!test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags))
1012 drbd_flush_after_epoch(mdev, epoch);
1014 drbd_may_finish_epoch(mdev, epoch, EV_PUT |
1015 (mdev->state.conn < C_CONNECTED ? EV_CLEANUP : 0));
1021 * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it.
1022 * @mdev: DRBD device.
1023 * @epoch: Epoch object.
1026 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev,
1027 struct drbd_epoch *epoch,
1028 enum epoch_event ev)
1030 int finish, epoch_size;
1031 struct drbd_epoch *next_epoch;
1032 int schedule_flush = 0;
1033 enum finish_epoch rv = FE_STILL_LIVE;
1035 spin_lock(&mdev->epoch_lock);
1040 epoch_size = atomic_read(&epoch->epoch_size);
1042 switch (ev & ~EV_CLEANUP) {
1044 atomic_dec(&epoch->active);
1046 case EV_GOT_BARRIER_NR:
1047 set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags);
1049 /* Special case: If we just switched from WO_bio_barrier to
1050 WO_bdev_flush we should not finish the current epoch */
1051 if (test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags) && epoch_size == 1 &&
1052 mdev->write_ordering != WO_bio_barrier &&
1053 epoch == mdev->current_epoch)
1054 clear_bit(DE_CONTAINS_A_BARRIER, &epoch->flags);
1056 case EV_BARRIER_DONE:
1057 set_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags);
1059 case EV_BECAME_LAST:
1064 if (epoch_size != 0 &&
1065 atomic_read(&epoch->active) == 0 &&
1066 test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags) &&
1067 epoch->list.prev == &mdev->current_epoch->list &&
1068 !test_bit(DE_IS_FINISHING, &epoch->flags)) {
1069 /* Nearly all conditions are met to finish that epoch... */
1070 if (test_bit(DE_BARRIER_IN_NEXT_EPOCH_DONE, &epoch->flags) ||
1071 mdev->write_ordering == WO_none ||
1072 (epoch_size == 1 && test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) ||
1075 set_bit(DE_IS_FINISHING, &epoch->flags);
1076 } else if (!test_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags) &&
1077 mdev->write_ordering == WO_bio_barrier) {
1078 atomic_inc(&epoch->active);
1083 if (!(ev & EV_CLEANUP)) {
1084 spin_unlock(&mdev->epoch_lock);
1085 drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size);
1086 spin_lock(&mdev->epoch_lock);
1090 if (mdev->current_epoch != epoch) {
1091 next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list);
1092 list_del(&epoch->list);
1093 ev = EV_BECAME_LAST | (ev & EV_CLEANUP);
1097 if (rv == FE_STILL_LIVE)
1101 atomic_set(&epoch->epoch_size, 0);
1102 /* atomic_set(&epoch->active, 0); is already zero */
1103 if (rv == FE_STILL_LIVE)
1114 spin_unlock(&mdev->epoch_lock);
1116 if (schedule_flush) {
1117 struct flush_work *fw;
1118 fw = kmalloc(sizeof(*fw), GFP_ATOMIC);
1122 drbd_queue_work(&mdev->data.work, &fw->w);
1124 dev_warn(DEV, "Could not kmalloc a flush_work obj\n");
1125 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags);
1126 /* That is not a recursion, only one level */
1127 drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE);
1128 drbd_may_finish_epoch(mdev, epoch, EV_PUT);
1136 * drbd_bump_write_ordering() - Fall back to an other write ordering method
1137 * @mdev: DRBD device.
1138 * @wo: Write ordering method to try.
1140 void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local)
1142 enum write_ordering_e pwo;
1143 static char *write_ordering_str[] = {
1145 [WO_drain_io] = "drain",
1146 [WO_bdev_flush] = "flush",
1147 [WO_bio_barrier] = "barrier",
1150 pwo = mdev->write_ordering;
1152 if (wo == WO_bio_barrier && mdev->ldev->dc.no_disk_barrier)
1154 if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush)
1156 if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain)
1158 mdev->write_ordering = wo;
1159 if (pwo != mdev->write_ordering || wo == WO_bio_barrier)
1160 dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]);
1165 * @mdev: DRBD device.
1167 * @rw: flag field, see bio->bi_rw
1169 /* TODO allocate from our own bio_set. */
1170 int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e,
1171 const unsigned rw, const int fault_type)
1173 struct bio *bios = NULL;
1175 struct page *page = e->pages;
1176 sector_t sector = e->sector;
1177 unsigned ds = e->size;
1178 unsigned n_bios = 0;
1179 unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT;
1181 /* In most cases, we will only need one bio. But in case the lower
1182 * level restrictions happen to be different at this offset on this
1183 * side than those of the sending peer, we may need to submit the
1184 * request in more than one bio. */
1186 bio = bio_alloc(GFP_NOIO, nr_pages);
1188 dev_err(DEV, "submit_ee: Allocation of a bio failed\n");
1191 /* > e->sector, unless this is the first bio */
1192 bio->bi_sector = sector;
1193 bio->bi_bdev = mdev->ldev->backing_bdev;
1194 /* we special case some flags in the multi-bio case, see below
1195 * (REQ_UNPLUG, REQ_HARDBARRIER) */
1197 bio->bi_private = e;
1198 bio->bi_end_io = drbd_endio_sec;
1200 bio->bi_next = bios;
1204 page_chain_for_each(page) {
1205 unsigned len = min_t(unsigned, ds, PAGE_SIZE);
1206 if (!bio_add_page(bio, page, len, 0)) {
1207 /* a single page must always be possible! */
1208 BUG_ON(bio->bi_vcnt == 0);
1215 D_ASSERT(page == NULL);
1218 atomic_set(&e->pending_bios, n_bios);
1221 bios = bios->bi_next;
1222 bio->bi_next = NULL;
1224 /* strip off REQ_UNPLUG unless it is the last bio */
1226 bio->bi_rw &= ~REQ_UNPLUG;
1228 drbd_generic_make_request(mdev, fault_type, bio);
1230 /* strip off REQ_HARDBARRIER,
1231 * unless it is the first or last bio */
1232 if (bios && bios->bi_next)
1233 bios->bi_rw &= ~REQ_HARDBARRIER;
1235 maybe_kick_lo(mdev);
1241 bios = bios->bi_next;
1248 * w_e_reissue() - Worker callback; Resubmit a bio, without REQ_HARDBARRIER set
1249 * @mdev: DRBD device.
1251 * @cancel: The connection will be closed anyways (unused in this callback)
1253 int w_e_reissue(struct drbd_conf *mdev, struct drbd_work *w, int cancel) __releases(local)
1255 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1256 /* We leave DE_CONTAINS_A_BARRIER and EE_IS_BARRIER in place,
1257 (and DE_BARRIER_IN_NEXT_EPOCH_ISSUED in the previous Epoch)
1258 so that we can finish that epoch in drbd_may_finish_epoch().
1259 That is necessary if we already have a long chain of Epochs, before
1260 we realize that REQ_HARDBARRIER is actually not supported */
1262 /* As long as the -ENOTSUPP on the barrier is reported immediately
1263 that will never trigger. If it is reported late, we will just
1264 print that warning and continue correctly for all future requests
1265 with WO_bdev_flush */
1266 if (previous_epoch(mdev, e->epoch))
1267 dev_warn(DEV, "Write ordering was not enforced (one time event)\n");
1269 /* we still have a local reference,
1270 * get_ldev was done in receive_Data. */
1272 e->w.cb = e_end_block;
1273 if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_DT_WR) != 0) {
1274 /* drbd_submit_ee fails for one reason only:
1275 * if was not able to allocate sufficient bios.
1276 * requeue, try again later. */
1277 e->w.cb = w_e_reissue;
1278 drbd_queue_work(&mdev->data.work, &e->w);
1283 static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1285 int rv, issue_flush;
1286 struct p_barrier *p = &mdev->data.rbuf.barrier;
1287 struct drbd_epoch *epoch;
1291 if (mdev->net_conf->wire_protocol != DRBD_PROT_C)
1294 mdev->current_epoch->barrier_nr = p->barrier;
1295 rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR);
1297 /* P_BARRIER_ACK may imply that the corresponding extent is dropped from
1298 * the activity log, which means it would not be resynced in case the
1299 * R_PRIMARY crashes now.
1300 * Therefore we must send the barrier_ack after the barrier request was
1302 switch (mdev->write_ordering) {
1303 case WO_bio_barrier:
1305 if (rv == FE_RECYCLED)
1311 if (rv == FE_STILL_LIVE) {
1312 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags);
1313 drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
1314 rv = drbd_flush_after_epoch(mdev, mdev->current_epoch);
1316 if (rv == FE_RECYCLED)
1319 /* The asender will send all the ACKs and barrier ACKs out, since
1320 all EEs moved from the active_ee to the done_ee. We need to
1321 provide a new epoch object for the EEs that come in soon */
1325 /* receiver context, in the writeout path of the other node.
1326 * avoid potential distributed deadlock */
1327 epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO);
1329 dev_warn(DEV, "Allocation of an epoch failed, slowing down\n");
1330 issue_flush = !test_and_set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &mdev->current_epoch->flags);
1331 drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
1333 rv = drbd_flush_after_epoch(mdev, mdev->current_epoch);
1334 if (rv == FE_RECYCLED)
1338 drbd_wait_ee_list_empty(mdev, &mdev->done_ee);
1344 atomic_set(&epoch->epoch_size, 0);
1345 atomic_set(&epoch->active, 0);
1347 spin_lock(&mdev->epoch_lock);
1348 if (atomic_read(&mdev->current_epoch->epoch_size)) {
1349 list_add(&epoch->list, &mdev->current_epoch->list);
1350 mdev->current_epoch = epoch;
1353 /* The current_epoch got recycled while we allocated this one... */
1356 spin_unlock(&mdev->epoch_lock);
1361 /* used from receive_RSDataReply (recv_resync_read)
1362 * and from receive_Data */
1363 static struct drbd_epoch_entry *
1364 read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local)
1366 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
1367 struct drbd_epoch_entry *e;
1370 void *dig_in = mdev->int_dig_in;
1371 void *dig_vv = mdev->int_dig_vv;
1372 unsigned long *data;
1374 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1375 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1378 rr = drbd_recv(mdev, dig_in, dgs);
1380 dev_warn(DEV, "short read receiving data digest: read %d expected %d\n",
1388 ERR_IF(data_size & 0x1ff) return NULL;
1389 ERR_IF(data_size > DRBD_MAX_SEGMENT_SIZE) return NULL;
1391 /* even though we trust out peer,
1392 * we sometimes have to double check. */
1393 if (sector + (data_size>>9) > capacity) {
1394 dev_err(DEV, "capacity: %llus < sector: %llus + size: %u\n",
1395 (unsigned long long)capacity,
1396 (unsigned long long)sector, data_size);
1400 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
1401 * "criss-cross" setup, that might cause write-out on some other DRBD,
1402 * which in turn might block on the other node at this very place. */
1403 e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO);
1409 page_chain_for_each(page) {
1410 unsigned len = min_t(int, ds, PAGE_SIZE);
1412 rr = drbd_recv(mdev, data, len);
1413 if (FAULT_ACTIVE(mdev, DRBD_FAULT_RECEIVE)) {
1414 dev_err(DEV, "Fault injection: Corrupting data on receive\n");
1415 data[0] = data[0] ^ (unsigned long)-1;
1419 drbd_free_ee(mdev, e);
1420 dev_warn(DEV, "short read receiving data: read %d expected %d\n",
1428 drbd_csum_ee(mdev, mdev->integrity_r_tfm, e, dig_vv);
1429 if (memcmp(dig_in, dig_vv, dgs)) {
1430 dev_err(DEV, "Digest integrity check FAILED.\n");
1431 drbd_bcast_ee(mdev, "digest failed",
1432 dgs, dig_in, dig_vv, e);
1433 drbd_free_ee(mdev, e);
1437 mdev->recv_cnt += data_size>>9;
1441 /* drbd_drain_block() just takes a data block
1442 * out of the socket input buffer, and discards it.
1444 static int drbd_drain_block(struct drbd_conf *mdev, int data_size)
1453 page = drbd_pp_alloc(mdev, 1, 1);
1457 rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE));
1458 if (rr != min_t(int, data_size, PAGE_SIZE)) {
1460 dev_warn(DEV, "short read receiving data: read %d expected %d\n",
1461 rr, min_t(int, data_size, PAGE_SIZE));
1467 drbd_pp_free(mdev, page, 0);
1471 static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req,
1472 sector_t sector, int data_size)
1474 struct bio_vec *bvec;
1476 int dgs, rr, i, expect;
1477 void *dig_in = mdev->int_dig_in;
1478 void *dig_vv = mdev->int_dig_vv;
1480 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1481 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1484 rr = drbd_recv(mdev, dig_in, dgs);
1486 dev_warn(DEV, "short read receiving data reply digest: read %d expected %d\n",
1494 /* optimistically update recv_cnt. if receiving fails below,
1495 * we disconnect anyways, and counters will be reset. */
1496 mdev->recv_cnt += data_size>>9;
1498 bio = req->master_bio;
1499 D_ASSERT(sector == bio->bi_sector);
1501 bio_for_each_segment(bvec, bio, i) {
1502 expect = min_t(int, data_size, bvec->bv_len);
1503 rr = drbd_recv(mdev,
1504 kmap(bvec->bv_page)+bvec->bv_offset,
1506 kunmap(bvec->bv_page);
1508 dev_warn(DEV, "short read receiving data reply: "
1509 "read %d expected %d\n",
1517 drbd_csum_bio(mdev, mdev->integrity_r_tfm, bio, dig_vv);
1518 if (memcmp(dig_in, dig_vv, dgs)) {
1519 dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n");
1524 D_ASSERT(data_size == 0);
1528 /* e_end_resync_block() is called via
1529 * drbd_process_done_ee() by asender only */
1530 static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1532 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1533 sector_t sector = e->sector;
1536 D_ASSERT(hlist_unhashed(&e->colision));
1538 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1539 drbd_set_in_sync(mdev, sector, e->size);
1540 ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e);
1542 /* Record failure to sync */
1543 drbd_rs_failed_io(mdev, sector, e->size);
1545 ok = drbd_send_ack(mdev, P_NEG_ACK, e);
1552 static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local)
1554 struct drbd_epoch_entry *e;
1556 e = read_in_block(mdev, ID_SYNCER, sector, data_size);
1560 dec_rs_pending(mdev);
1563 /* corresponding dec_unacked() in e_end_resync_block()
1564 * respective _drbd_clear_done_ee */
1566 e->w.cb = e_end_resync_block;
1568 spin_lock_irq(&mdev->req_lock);
1569 list_add(&e->w.list, &mdev->sync_ee);
1570 spin_unlock_irq(&mdev->req_lock);
1572 atomic_add(data_size >> 9, &mdev->rs_sect_ev);
1573 if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0)
1576 /* drbd_submit_ee currently fails for one reason only:
1577 * not being able to allocate enough bios.
1578 * Is dropping the connection going to help? */
1579 spin_lock_irq(&mdev->req_lock);
1580 list_del(&e->w.list);
1581 spin_unlock_irq(&mdev->req_lock);
1583 drbd_free_ee(mdev, e);
1589 static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1591 struct drbd_request *req;
1594 struct p_data *p = &mdev->data.rbuf.data;
1596 sector = be64_to_cpu(p->sector);
1598 spin_lock_irq(&mdev->req_lock);
1599 req = _ar_id_to_req(mdev, p->block_id, sector);
1600 spin_unlock_irq(&mdev->req_lock);
1601 if (unlikely(!req)) {
1602 dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n");
1606 /* hlist_del(&req->colision) is done in _req_may_be_done, to avoid
1607 * special casing it there for the various failure cases.
1608 * still no race with drbd_fail_pending_reads */
1609 ok = recv_dless_read(mdev, req, sector, data_size);
1612 req_mod(req, data_received);
1613 /* else: nothing. handled from drbd_disconnect...
1614 * I don't think we may complete this just yet
1615 * in case we are "on-disconnect: freeze" */
1620 static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1624 struct p_data *p = &mdev->data.rbuf.data;
1626 sector = be64_to_cpu(p->sector);
1627 D_ASSERT(p->block_id == ID_SYNCER);
1629 if (get_ldev(mdev)) {
1630 /* data is submitted to disk within recv_resync_read.
1631 * corresponding put_ldev done below on error,
1632 * or in drbd_endio_write_sec. */
1633 ok = recv_resync_read(mdev, sector, data_size);
1635 if (__ratelimit(&drbd_ratelimit_state))
1636 dev_err(DEV, "Can not write resync data to local disk.\n");
1638 ok = drbd_drain_block(mdev, data_size);
1640 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size);
1643 atomic_add(data_size >> 9, &mdev->rs_sect_in);
1648 /* e_end_block() is called via drbd_process_done_ee().
1649 * this means this function only runs in the asender thread
1651 static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1653 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1654 sector_t sector = e->sector;
1655 struct drbd_epoch *epoch;
1658 if (e->flags & EE_IS_BARRIER) {
1659 epoch = previous_epoch(mdev, e->epoch);
1661 drbd_may_finish_epoch(mdev, epoch, EV_BARRIER_DONE + (cancel ? EV_CLEANUP : 0));
1664 if (mdev->net_conf->wire_protocol == DRBD_PROT_C) {
1665 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1666 pcmd = (mdev->state.conn >= C_SYNC_SOURCE &&
1667 mdev->state.conn <= C_PAUSED_SYNC_T &&
1668 e->flags & EE_MAY_SET_IN_SYNC) ?
1669 P_RS_WRITE_ACK : P_WRITE_ACK;
1670 ok &= drbd_send_ack(mdev, pcmd, e);
1671 if (pcmd == P_RS_WRITE_ACK)
1672 drbd_set_in_sync(mdev, sector, e->size);
1674 ok = drbd_send_ack(mdev, P_NEG_ACK, e);
1675 /* we expect it to be marked out of sync anyways...
1676 * maybe assert this? */
1680 /* we delete from the conflict detection hash _after_ we sent out the
1681 * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */
1682 if (mdev->net_conf->two_primaries) {
1683 spin_lock_irq(&mdev->req_lock);
1684 D_ASSERT(!hlist_unhashed(&e->colision));
1685 hlist_del_init(&e->colision);
1686 spin_unlock_irq(&mdev->req_lock);
1688 D_ASSERT(hlist_unhashed(&e->colision));
1691 drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0));
1696 static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1698 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1701 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1702 ok = drbd_send_ack(mdev, P_DISCARD_ACK, e);
1704 spin_lock_irq(&mdev->req_lock);
1705 D_ASSERT(!hlist_unhashed(&e->colision));
1706 hlist_del_init(&e->colision);
1707 spin_unlock_irq(&mdev->req_lock);
1714 /* Called from receive_Data.
1715 * Synchronize packets on sock with packets on msock.
1717 * This is here so even when a P_DATA packet traveling via sock overtook an Ack
1718 * packet traveling on msock, they are still processed in the order they have
1721 * Note: we don't care for Ack packets overtaking P_DATA packets.
1723 * In case packet_seq is larger than mdev->peer_seq number, there are
1724 * outstanding packets on the msock. We wait for them to arrive.
1725 * In case we are the logically next packet, we update mdev->peer_seq
1726 * ourselves. Correctly handles 32bit wrap around.
1728 * Assume we have a 10 GBit connection, that is about 1<<30 byte per second,
1729 * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds
1730 * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have
1731 * 1<<9 == 512 seconds aka ages for the 32bit wrap around...
1733 * returns 0 if we may process the packet,
1734 * -ERESTARTSYS if we were interrupted (by disconnect signal). */
1735 static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq)
1741 spin_lock(&mdev->peer_seq_lock);
1743 prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE);
1744 if (seq_le(packet_seq, mdev->peer_seq+1))
1746 if (signal_pending(current)) {
1750 p_seq = mdev->peer_seq;
1751 spin_unlock(&mdev->peer_seq_lock);
1752 timeout = schedule_timeout(30*HZ);
1753 spin_lock(&mdev->peer_seq_lock);
1754 if (timeout == 0 && p_seq == mdev->peer_seq) {
1756 dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n");
1760 finish_wait(&mdev->seq_wait, &wait);
1761 if (mdev->peer_seq+1 == packet_seq)
1763 spin_unlock(&mdev->peer_seq_lock);
1767 static unsigned long write_flags_to_bio(struct drbd_conf *mdev, u32 dpf)
1769 if (mdev->agreed_pro_version >= 95)
1770 return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) |
1771 (dpf & DP_UNPLUG ? REQ_UNPLUG : 0) |
1772 (dpf & DP_FUA ? REQ_FUA : 0) |
1773 (dpf & DP_FLUSH ? REQ_FUA : 0) |
1774 (dpf & DP_DISCARD ? REQ_DISCARD : 0);
1776 return dpf & DP_RW_SYNC ? (REQ_SYNC | REQ_UNPLUG) : 0;
1779 /* mirrored write */
1780 static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1783 struct drbd_epoch_entry *e;
1784 struct p_data *p = &mdev->data.rbuf.data;
1788 if (!get_ldev(mdev)) {
1789 if (__ratelimit(&drbd_ratelimit_state))
1790 dev_err(DEV, "Can not write mirrored data block "
1791 "to local disk.\n");
1792 spin_lock(&mdev->peer_seq_lock);
1793 if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num))
1795 spin_unlock(&mdev->peer_seq_lock);
1797 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size);
1798 atomic_inc(&mdev->current_epoch->epoch_size);
1799 return drbd_drain_block(mdev, data_size);
1802 /* get_ldev(mdev) successful.
1803 * Corresponding put_ldev done either below (on various errors),
1804 * or in drbd_endio_write_sec, if we successfully submit the data at
1805 * the end of this function. */
1807 sector = be64_to_cpu(p->sector);
1808 e = read_in_block(mdev, p->block_id, sector, data_size);
1814 e->w.cb = e_end_block;
1816 spin_lock(&mdev->epoch_lock);
1817 e->epoch = mdev->current_epoch;
1818 atomic_inc(&e->epoch->epoch_size);
1819 atomic_inc(&e->epoch->active);
1821 if (mdev->write_ordering == WO_bio_barrier && atomic_read(&e->epoch->epoch_size) == 1) {
1822 struct drbd_epoch *epoch;
1823 /* Issue a barrier if we start a new epoch, and the previous epoch
1824 was not a epoch containing a single request which already was
1826 epoch = list_entry(e->epoch->list.prev, struct drbd_epoch, list);
1827 if (epoch == e->epoch) {
1828 set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags);
1829 rw |= REQ_HARDBARRIER;
1830 e->flags |= EE_IS_BARRIER;
1832 if (atomic_read(&epoch->epoch_size) > 1 ||
1833 !test_bit(DE_CONTAINS_A_BARRIER, &epoch->flags)) {
1834 set_bit(DE_BARRIER_IN_NEXT_EPOCH_ISSUED, &epoch->flags);
1835 set_bit(DE_CONTAINS_A_BARRIER, &e->epoch->flags);
1836 rw |= REQ_HARDBARRIER;
1837 e->flags |= EE_IS_BARRIER;
1841 spin_unlock(&mdev->epoch_lock);
1843 dp_flags = be32_to_cpu(p->dp_flags);
1844 rw |= write_flags_to_bio(mdev, dp_flags);
1846 if (dp_flags & DP_MAY_SET_IN_SYNC)
1847 e->flags |= EE_MAY_SET_IN_SYNC;
1849 /* I'm the receiver, I do hold a net_cnt reference. */
1850 if (!mdev->net_conf->two_primaries) {
1851 spin_lock_irq(&mdev->req_lock);
1853 /* don't get the req_lock yet,
1854 * we may sleep in drbd_wait_peer_seq */
1855 const int size = e->size;
1856 const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1858 struct drbd_request *i;
1859 struct hlist_node *n;
1860 struct hlist_head *slot;
1863 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1864 BUG_ON(mdev->ee_hash == NULL);
1865 BUG_ON(mdev->tl_hash == NULL);
1867 /* conflict detection and handling:
1868 * 1. wait on the sequence number,
1869 * in case this data packet overtook ACK packets.
1870 * 2. check our hash tables for conflicting requests.
1871 * we only need to walk the tl_hash, since an ee can not
1872 * have a conflict with an other ee: on the submitting
1873 * node, the corresponding req had already been conflicting,
1874 * and a conflicting req is never sent.
1876 * Note: for two_primaries, we are protocol C,
1877 * so there cannot be any request that is DONE
1878 * but still on the transfer log.
1880 * unconditionally add to the ee_hash.
1882 * if no conflicting request is found:
1885 * if any conflicting request is found
1886 * that has not yet been acked,
1887 * AND I have the "discard concurrent writes" flag:
1888 * queue (via done_ee) the P_DISCARD_ACK; OUT.
1890 * if any conflicting request is found:
1891 * block the receiver, waiting on misc_wait
1892 * until no more conflicting requests are there,
1893 * or we get interrupted (disconnect).
1895 * we do not just write after local io completion of those
1896 * requests, but only after req is done completely, i.e.
1897 * we wait for the P_DISCARD_ACK to arrive!
1899 * then proceed normally, i.e. submit.
1901 if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num)))
1902 goto out_interrupted;
1904 spin_lock_irq(&mdev->req_lock);
1906 hlist_add_head(&e->colision, ee_hash_slot(mdev, sector));
1908 #define OVERLAPS overlaps(i->sector, i->size, sector, size)
1909 slot = tl_hash_slot(mdev, sector);
1912 int have_unacked = 0;
1913 int have_conflict = 0;
1914 prepare_to_wait(&mdev->misc_wait, &wait,
1915 TASK_INTERRUPTIBLE);
1916 hlist_for_each_entry(i, n, slot, colision) {
1918 /* only ALERT on first iteration,
1919 * we may be woken up early... */
1921 dev_alert(DEV, "%s[%u] Concurrent local write detected!"
1922 " new: %llus +%u; pending: %llus +%u\n",
1923 current->comm, current->pid,
1924 (unsigned long long)sector, size,
1925 (unsigned long long)i->sector, i->size);
1926 if (i->rq_state & RQ_NET_PENDING)
1935 /* Discard Ack only for the _first_ iteration */
1936 if (first && discard && have_unacked) {
1937 dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n",
1938 (unsigned long long)sector);
1940 e->w.cb = e_send_discard_ack;
1941 list_add_tail(&e->w.list, &mdev->done_ee);
1943 spin_unlock_irq(&mdev->req_lock);
1945 /* we could probably send that P_DISCARD_ACK ourselves,
1946 * but I don't like the receiver using the msock */
1950 finish_wait(&mdev->misc_wait, &wait);
1954 if (signal_pending(current)) {
1955 hlist_del_init(&e->colision);
1957 spin_unlock_irq(&mdev->req_lock);
1959 finish_wait(&mdev->misc_wait, &wait);
1960 goto out_interrupted;
1963 spin_unlock_irq(&mdev->req_lock);
1966 dev_alert(DEV, "Concurrent write! [W AFTERWARDS] "
1967 "sec=%llus\n", (unsigned long long)sector);
1968 } else if (discard) {
1969 /* we had none on the first iteration.
1970 * there must be none now. */
1971 D_ASSERT(have_unacked == 0);
1974 spin_lock_irq(&mdev->req_lock);
1976 finish_wait(&mdev->misc_wait, &wait);
1979 list_add(&e->w.list, &mdev->active_ee);
1980 spin_unlock_irq(&mdev->req_lock);
1982 switch (mdev->net_conf->wire_protocol) {
1985 /* corresponding dec_unacked() in e_end_block()
1986 * respective _drbd_clear_done_ee */
1989 /* I really don't like it that the receiver thread
1990 * sends on the msock, but anyways */
1991 drbd_send_ack(mdev, P_RECV_ACK, e);
1998 if (mdev->state.pdsk == D_DISKLESS) {
1999 /* In case we have the only disk of the cluster, */
2000 drbd_set_out_of_sync(mdev, e->sector, e->size);
2001 e->flags |= EE_CALL_AL_COMPLETE_IO;
2002 drbd_al_begin_io(mdev, e->sector);
2005 if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0)
2008 /* drbd_submit_ee currently fails for one reason only:
2009 * not being able to allocate enough bios.
2010 * Is dropping the connection going to help? */
2011 spin_lock_irq(&mdev->req_lock);
2012 list_del(&e->w.list);
2013 hlist_del_init(&e->colision);
2014 spin_unlock_irq(&mdev->req_lock);
2015 if (e->flags & EE_CALL_AL_COMPLETE_IO)
2016 drbd_al_complete_io(mdev, e->sector);
2019 /* yes, the epoch_size now is imbalanced.
2020 * but we drop the connection anyways, so we don't have a chance to
2021 * receive a barrier... atomic_inc(&mdev->epoch_size); */
2023 drbd_free_ee(mdev, e);
2027 /* We may throttle resync, if the lower device seems to be busy,
2028 * and current sync rate is above c_min_rate.
2030 * To decide whether or not the lower device is busy, we use a scheme similar
2031 * to MD RAID is_mddev_idle(): if the partition stats reveal "significant"
2032 * (more than 64 sectors) of activity we cannot account for with our own resync
2033 * activity, it obviously is "busy".
2035 * The current sync rate used here uses only the most recent two step marks,
2036 * to have a short time average so we can react faster.
2038 int drbd_rs_should_slow_down(struct drbd_conf *mdev)
2040 struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk;
2041 unsigned long db, dt, dbdt;
2045 /* feature disabled? */
2046 if (mdev->sync_conf.c_min_rate == 0)
2049 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
2050 (int)part_stat_read(&disk->part0, sectors[1]) -
2051 atomic_read(&mdev->rs_sect_ev);
2052 if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) {
2053 unsigned long rs_left;
2056 mdev->rs_last_events = curr_events;
2058 /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP,
2060 i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-2) % DRBD_SYNC_MARKS;
2061 rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed;
2063 dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ;
2066 db = mdev->rs_mark_left[i] - rs_left;
2067 dbdt = Bit2KB(db/dt);
2069 if (dbdt > mdev->sync_conf.c_min_rate)
2076 static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size)
2079 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
2080 struct drbd_epoch_entry *e;
2081 struct digest_info *di = NULL;
2083 unsigned int fault_type;
2084 struct p_block_req *p = &mdev->data.rbuf.block_req;
2086 sector = be64_to_cpu(p->sector);
2087 size = be32_to_cpu(p->blksize);
2089 if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) {
2090 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
2091 (unsigned long long)sector, size);
2094 if (sector + (size>>9) > capacity) {
2095 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
2096 (unsigned long long)sector, size);
2100 if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) {
2103 case P_DATA_REQUEST:
2104 drbd_send_ack_rp(mdev, P_NEG_DREPLY, p);
2106 case P_RS_DATA_REQUEST:
2107 case P_CSUM_RS_REQUEST:
2109 drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p);
2113 dec_rs_pending(mdev);
2114 drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC);
2117 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n",
2120 if (verb && __ratelimit(&drbd_ratelimit_state))
2121 dev_err(DEV, "Can not satisfy peer's read request, "
2122 "no local data.\n");
2124 /* drain possibly payload */
2125 return drbd_drain_block(mdev, digest_size);
2128 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
2129 * "criss-cross" setup, that might cause write-out on some other DRBD,
2130 * which in turn might block on the other node at this very place. */
2131 e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO);
2138 case P_DATA_REQUEST:
2139 e->w.cb = w_e_end_data_req;
2140 fault_type = DRBD_FAULT_DT_RD;
2141 /* application IO, don't drbd_rs_begin_io */
2144 case P_RS_DATA_REQUEST:
2145 e->w.cb = w_e_end_rsdata_req;
2146 fault_type = DRBD_FAULT_RS_RD;
2150 case P_CSUM_RS_REQUEST:
2151 fault_type = DRBD_FAULT_RS_RD;
2152 di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO);
2156 di->digest_size = digest_size;
2157 di->digest = (((char *)di)+sizeof(struct digest_info));
2160 e->flags |= EE_HAS_DIGEST;
2162 if (drbd_recv(mdev, di->digest, digest_size) != digest_size)
2165 if (cmd == P_CSUM_RS_REQUEST) {
2166 D_ASSERT(mdev->agreed_pro_version >= 89);
2167 e->w.cb = w_e_end_csum_rs_req;
2168 } else if (cmd == P_OV_REPLY) {
2169 e->w.cb = w_e_end_ov_reply;
2170 dec_rs_pending(mdev);
2171 /* drbd_rs_begin_io done when we sent this request,
2172 * but accounting still needs to be done. */
2173 goto submit_for_resync;
2178 if (mdev->ov_start_sector == ~(sector_t)0 &&
2179 mdev->agreed_pro_version >= 90) {
2180 mdev->ov_start_sector = sector;
2181 mdev->ov_position = sector;
2182 mdev->ov_left = mdev->rs_total - BM_SECT_TO_BIT(sector);
2183 dev_info(DEV, "Online Verify start sector: %llu\n",
2184 (unsigned long long)sector);
2186 e->w.cb = w_e_end_ov_req;
2187 fault_type = DRBD_FAULT_RS_RD;
2191 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n",
2193 fault_type = DRBD_FAULT_MAX;
2197 /* Throttle, drbd_rs_begin_io and submit should become asynchronous
2198 * wrt the receiver, but it is not as straightforward as it may seem.
2199 * Various places in the resync start and stop logic assume resync
2200 * requests are processed in order, requeuing this on the worker thread
2201 * introduces a bunch of new code for synchronization between threads.
2203 * Unlimited throttling before drbd_rs_begin_io may stall the resync
2204 * "forever", throttling after drbd_rs_begin_io will lock that extent
2205 * for application writes for the same time. For now, just throttle
2206 * here, where the rest of the code expects the receiver to sleep for
2210 /* Throttle before drbd_rs_begin_io, as that locks out application IO;
2211 * this defers syncer requests for some time, before letting at least
2212 * on request through. The resync controller on the receiving side
2213 * will adapt to the incoming rate accordingly.
2215 * We cannot throttle here if remote is Primary/SyncTarget:
2216 * we would also throttle its application reads.
2217 * In that case, throttling is done on the SyncTarget only.
2219 if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev))
2221 if (drbd_rs_begin_io(mdev, e->sector))
2225 atomic_add(size >> 9, &mdev->rs_sect_ev);
2229 spin_lock_irq(&mdev->req_lock);
2230 list_add_tail(&e->w.list, &mdev->read_ee);
2231 spin_unlock_irq(&mdev->req_lock);
2233 if (drbd_submit_ee(mdev, e, READ, fault_type) == 0)
2236 /* drbd_submit_ee currently fails for one reason only:
2237 * not being able to allocate enough bios.
2238 * Is dropping the connection going to help? */
2239 spin_lock_irq(&mdev->req_lock);
2240 list_del(&e->w.list);
2241 spin_unlock_irq(&mdev->req_lock);
2242 /* no drbd_rs_complete_io(), we are dropping the connection anyways */
2246 drbd_free_ee(mdev, e);
2250 static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local)
2252 int self, peer, rv = -100;
2253 unsigned long ch_self, ch_peer;
2255 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2256 peer = mdev->p_uuid[UI_BITMAP] & 1;
2258 ch_peer = mdev->p_uuid[UI_SIZE];
2259 ch_self = mdev->comm_bm_set;
2261 switch (mdev->net_conf->after_sb_0p) {
2263 case ASB_DISCARD_SECONDARY:
2264 case ASB_CALL_HELPER:
2265 dev_err(DEV, "Configuration error.\n");
2267 case ASB_DISCONNECT:
2269 case ASB_DISCARD_YOUNGER_PRI:
2270 if (self == 0 && peer == 1) {
2274 if (self == 1 && peer == 0) {
2278 /* Else fall through to one of the other strategies... */
2279 case ASB_DISCARD_OLDER_PRI:
2280 if (self == 0 && peer == 1) {
2284 if (self == 1 && peer == 0) {
2288 /* Else fall through to one of the other strategies... */
2289 dev_warn(DEV, "Discard younger/older primary did not find a decision\n"
2290 "Using discard-least-changes instead\n");
2291 case ASB_DISCARD_ZERO_CHG:
2292 if (ch_peer == 0 && ch_self == 0) {
2293 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2297 if (ch_peer == 0) { rv = 1; break; }
2298 if (ch_self == 0) { rv = -1; break; }
2300 if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG)
2302 case ASB_DISCARD_LEAST_CHG:
2303 if (ch_self < ch_peer)
2305 else if (ch_self > ch_peer)
2307 else /* ( ch_self == ch_peer ) */
2308 /* Well, then use something else. */
2309 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2312 case ASB_DISCARD_LOCAL:
2315 case ASB_DISCARD_REMOTE:
2322 static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local)
2324 int self, peer, hg, rv = -100;
2326 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2327 peer = mdev->p_uuid[UI_BITMAP] & 1;
2329 switch (mdev->net_conf->after_sb_1p) {
2330 case ASB_DISCARD_YOUNGER_PRI:
2331 case ASB_DISCARD_OLDER_PRI:
2332 case ASB_DISCARD_LEAST_CHG:
2333 case ASB_DISCARD_LOCAL:
2334 case ASB_DISCARD_REMOTE:
2335 dev_err(DEV, "Configuration error.\n");
2337 case ASB_DISCONNECT:
2340 hg = drbd_asb_recover_0p(mdev);
2341 if (hg == -1 && mdev->state.role == R_SECONDARY)
2343 if (hg == 1 && mdev->state.role == R_PRIMARY)
2347 rv = drbd_asb_recover_0p(mdev);
2349 case ASB_DISCARD_SECONDARY:
2350 return mdev->state.role == R_PRIMARY ? 1 : -1;
2351 case ASB_CALL_HELPER:
2352 hg = drbd_asb_recover_0p(mdev);
2353 if (hg == -1 && mdev->state.role == R_PRIMARY) {
2354 self = drbd_set_role(mdev, R_SECONDARY, 0);
2355 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2356 * we might be here in C_WF_REPORT_PARAMS which is transient.
2357 * we do not need to wait for the after state change work either. */
2358 self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2359 if (self != SS_SUCCESS) {
2360 drbd_khelper(mdev, "pri-lost-after-sb");
2362 dev_warn(DEV, "Successfully gave up primary role.\n");
2372 static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local)
2374 int self, peer, hg, rv = -100;
2376 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2377 peer = mdev->p_uuid[UI_BITMAP] & 1;
2379 switch (mdev->net_conf->after_sb_2p) {
2380 case ASB_DISCARD_YOUNGER_PRI:
2381 case ASB_DISCARD_OLDER_PRI:
2382 case ASB_DISCARD_LEAST_CHG:
2383 case ASB_DISCARD_LOCAL:
2384 case ASB_DISCARD_REMOTE:
2386 case ASB_DISCARD_SECONDARY:
2387 dev_err(DEV, "Configuration error.\n");
2390 rv = drbd_asb_recover_0p(mdev);
2392 case ASB_DISCONNECT:
2394 case ASB_CALL_HELPER:
2395 hg = drbd_asb_recover_0p(mdev);
2397 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2398 * we might be here in C_WF_REPORT_PARAMS which is transient.
2399 * we do not need to wait for the after state change work either. */
2400 self = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2401 if (self != SS_SUCCESS) {
2402 drbd_khelper(mdev, "pri-lost-after-sb");
2404 dev_warn(DEV, "Successfully gave up primary role.\n");
2414 static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid,
2415 u64 bits, u64 flags)
2418 dev_info(DEV, "%s uuid info vanished while I was looking!\n", text);
2421 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n",
2423 (unsigned long long)uuid[UI_CURRENT],
2424 (unsigned long long)uuid[UI_BITMAP],
2425 (unsigned long long)uuid[UI_HISTORY_START],
2426 (unsigned long long)uuid[UI_HISTORY_END],
2427 (unsigned long long)bits,
2428 (unsigned long long)flags);
2432 100 after split brain try auto recover
2433 2 C_SYNC_SOURCE set BitMap
2434 1 C_SYNC_SOURCE use BitMap
2436 -1 C_SYNC_TARGET use BitMap
2437 -2 C_SYNC_TARGET set BitMap
2438 -100 after split brain, disconnect
2439 -1000 unrelated data
2441 static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local)
2446 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2447 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2450 if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED)
2454 if ((self == UUID_JUST_CREATED || self == (u64)0) &&
2455 peer != UUID_JUST_CREATED)
2459 if (self != UUID_JUST_CREATED &&
2460 (peer == UUID_JUST_CREATED || peer == (u64)0))
2464 int rct, dc; /* roles at crash time */
2466 if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) {
2468 if (mdev->agreed_pro_version < 91)
2471 if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) &&
2472 (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) {
2473 dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n");
2474 drbd_uuid_set_bm(mdev, 0UL);
2476 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2477 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2480 dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n");
2487 if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) {
2489 if (mdev->agreed_pro_version < 91)
2492 if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) &&
2493 (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) {
2494 dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n");
2496 mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START];
2497 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP];
2498 mdev->p_uuid[UI_BITMAP] = 0UL;
2500 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2503 dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n");
2510 /* Common power [off|failure] */
2511 rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) +
2512 (mdev->p_uuid[UI_FLAGS] & 2);
2513 /* lowest bit is set when we were primary,
2514 * next bit (weight 2) is set when peer was primary */
2518 case 0: /* !self_pri && !peer_pri */ return 0;
2519 case 1: /* self_pri && !peer_pri */ return 1;
2520 case 2: /* !self_pri && peer_pri */ return -1;
2521 case 3: /* self_pri && peer_pri */
2522 dc = test_bit(DISCARD_CONCURRENT, &mdev->flags);
2528 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2533 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2535 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2536 peer = mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1);
2538 /* The last P_SYNC_UUID did not get though. Undo the last start of
2539 resync as sync source modifications of the peer's UUIDs. */
2541 if (mdev->agreed_pro_version < 91)
2544 mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START];
2545 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1];
2551 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2552 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2553 peer = mdev->p_uuid[i] & ~((u64)1);
2559 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2560 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2565 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2567 self = mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1);
2568 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2570 /* The last P_SYNC_UUID did not get though. Undo the last start of
2571 resync as sync source modifications of our UUIDs. */
2573 if (mdev->agreed_pro_version < 91)
2576 _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]);
2577 _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]);
2579 dev_info(DEV, "Undid last start of resync:\n");
2581 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2582 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2590 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2591 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2592 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2598 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2599 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2600 if (self == peer && self != ((u64)0))
2604 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2605 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2606 for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) {
2607 peer = mdev->p_uuid[j] & ~((u64)1);
2616 /* drbd_sync_handshake() returns the new conn state on success, or
2617 CONN_MASK (-1) on failure.
2619 static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role,
2620 enum drbd_disk_state peer_disk) __must_hold(local)
2623 enum drbd_conns rv = C_MASK;
2624 enum drbd_disk_state mydisk;
2626 mydisk = mdev->state.disk;
2627 if (mydisk == D_NEGOTIATING)
2628 mydisk = mdev->new_state_tmp.disk;
2630 dev_info(DEV, "drbd_sync_handshake:\n");
2631 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0);
2632 drbd_uuid_dump(mdev, "peer", mdev->p_uuid,
2633 mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2635 hg = drbd_uuid_compare(mdev, &rule_nr);
2637 dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr);
2640 dev_alert(DEV, "Unrelated data, aborting!\n");
2644 dev_alert(DEV, "To resolve this both sides have to support at least protocol\n");
2648 if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) ||
2649 (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) {
2650 int f = (hg == -100) || abs(hg) == 2;
2651 hg = mydisk > D_INCONSISTENT ? 1 : -1;
2654 dev_info(DEV, "Becoming sync %s due to disk states.\n",
2655 hg > 0 ? "source" : "target");
2659 drbd_khelper(mdev, "initial-split-brain");
2661 if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) {
2662 int pcount = (mdev->state.role == R_PRIMARY)
2663 + (peer_role == R_PRIMARY);
2664 int forced = (hg == -100);
2668 hg = drbd_asb_recover_0p(mdev);
2671 hg = drbd_asb_recover_1p(mdev);
2674 hg = drbd_asb_recover_2p(mdev);
2677 if (abs(hg) < 100) {
2678 dev_warn(DEV, "Split-Brain detected, %d primaries, "
2679 "automatically solved. Sync from %s node\n",
2680 pcount, (hg < 0) ? "peer" : "this");
2682 dev_warn(DEV, "Doing a full sync, since"
2683 " UUIDs where ambiguous.\n");
2690 if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1))
2692 if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1))
2696 dev_warn(DEV, "Split-Brain detected, manually solved. "
2697 "Sync from %s node\n",
2698 (hg < 0) ? "peer" : "this");
2702 /* FIXME this log message is not correct if we end up here
2703 * after an attempted attach on a diskless node.
2704 * We just refuse to attach -- well, we drop the "connection"
2705 * to that disk, in a way... */
2706 dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n");
2707 drbd_khelper(mdev, "split-brain");
2711 if (hg > 0 && mydisk <= D_INCONSISTENT) {
2712 dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n");
2716 if (hg < 0 && /* by intention we do not use mydisk here. */
2717 mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) {
2718 switch (mdev->net_conf->rr_conflict) {
2719 case ASB_CALL_HELPER:
2720 drbd_khelper(mdev, "pri-lost");
2722 case ASB_DISCONNECT:
2723 dev_err(DEV, "I shall become SyncTarget, but I am primary!\n");
2726 dev_warn(DEV, "Becoming SyncTarget, violating the stable-data"
2731 if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) {
2733 dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n");
2735 dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.",
2736 drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET),
2737 abs(hg) >= 2 ? "full" : "bit-map based");
2742 dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n");
2743 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake"))
2747 if (hg > 0) { /* become sync source. */
2749 } else if (hg < 0) { /* become sync target */
2753 if (drbd_bm_total_weight(mdev)) {
2754 dev_info(DEV, "No resync, but %lu bits in bitmap!\n",
2755 drbd_bm_total_weight(mdev));
2762 /* returns 1 if invalid */
2763 static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self)
2765 /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */
2766 if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) ||
2767 (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL))
2770 /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */
2771 if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL ||
2772 self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL)
2775 /* everything else is valid if they are equal on both sides. */
2779 /* everything es is invalid. */
2783 static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
2785 struct p_protocol *p = &mdev->data.rbuf.protocol;
2786 int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p;
2787 int p_want_lose, p_two_primaries, cf;
2788 char p_integrity_alg[SHARED_SECRET_MAX] = "";
2790 p_proto = be32_to_cpu(p->protocol);
2791 p_after_sb_0p = be32_to_cpu(p->after_sb_0p);
2792 p_after_sb_1p = be32_to_cpu(p->after_sb_1p);
2793 p_after_sb_2p = be32_to_cpu(p->after_sb_2p);
2794 p_two_primaries = be32_to_cpu(p->two_primaries);
2795 cf = be32_to_cpu(p->conn_flags);
2796 p_want_lose = cf & CF_WANT_LOSE;
2798 clear_bit(CONN_DRY_RUN, &mdev->flags);
2800 if (cf & CF_DRY_RUN)
2801 set_bit(CONN_DRY_RUN, &mdev->flags);
2803 if (p_proto != mdev->net_conf->wire_protocol) {
2804 dev_err(DEV, "incompatible communication protocols\n");
2808 if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) {
2809 dev_err(DEV, "incompatible after-sb-0pri settings\n");
2813 if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) {
2814 dev_err(DEV, "incompatible after-sb-1pri settings\n");
2818 if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) {
2819 dev_err(DEV, "incompatible after-sb-2pri settings\n");
2823 if (p_want_lose && mdev->net_conf->want_lose) {
2824 dev_err(DEV, "both sides have the 'want_lose' flag set\n");
2828 if (p_two_primaries != mdev->net_conf->two_primaries) {
2829 dev_err(DEV, "incompatible setting of the two-primaries options\n");
2833 if (mdev->agreed_pro_version >= 87) {
2834 unsigned char *my_alg = mdev->net_conf->integrity_alg;
2836 if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size)
2839 p_integrity_alg[SHARED_SECRET_MAX-1] = 0;
2840 if (strcmp(p_integrity_alg, my_alg)) {
2841 dev_err(DEV, "incompatible setting of the data-integrity-alg\n");
2844 dev_info(DEV, "data-integrity-alg: %s\n",
2845 my_alg[0] ? my_alg : (unsigned char *)"<not-used>");
2851 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2856 * input: alg name, feature name
2857 * return: NULL (alg name was "")
2858 * ERR_PTR(error) if something goes wrong
2859 * or the crypto hash ptr, if it worked out ok. */
2860 struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev,
2861 const char *alg, const char *name)
2863 struct crypto_hash *tfm;
2868 tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC);
2870 dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n",
2871 alg, name, PTR_ERR(tfm));
2874 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
2875 crypto_free_hash(tfm);
2876 dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name);
2877 return ERR_PTR(-EINVAL);
2882 static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size)
2885 struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95;
2886 unsigned int header_size, data_size, exp_max_sz;
2887 struct crypto_hash *verify_tfm = NULL;
2888 struct crypto_hash *csums_tfm = NULL;
2889 const int apv = mdev->agreed_pro_version;
2890 int *rs_plan_s = NULL;
2893 exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param)
2894 : apv == 88 ? sizeof(struct p_rs_param)
2896 : apv <= 94 ? sizeof(struct p_rs_param_89)
2897 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
2899 if (packet_size > exp_max_sz) {
2900 dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n",
2901 packet_size, exp_max_sz);
2906 header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80);
2907 data_size = packet_size - header_size;
2908 } else if (apv <= 94) {
2909 header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80);
2910 data_size = packet_size - header_size;
2911 D_ASSERT(data_size == 0);
2913 header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80);
2914 data_size = packet_size - header_size;
2915 D_ASSERT(data_size == 0);
2918 /* initialize verify_alg and csums_alg */
2919 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
2921 if (drbd_recv(mdev, &p->head.payload, header_size) != header_size)
2924 mdev->sync_conf.rate = be32_to_cpu(p->rate);
2928 if (data_size > SHARED_SECRET_MAX) {
2929 dev_err(DEV, "verify-alg too long, "
2930 "peer wants %u, accepting only %u byte\n",
2931 data_size, SHARED_SECRET_MAX);
2935 if (drbd_recv(mdev, p->verify_alg, data_size) != data_size)
2938 /* we expect NUL terminated string */
2939 /* but just in case someone tries to be evil */
2940 D_ASSERT(p->verify_alg[data_size-1] == 0);
2941 p->verify_alg[data_size-1] = 0;
2943 } else /* apv >= 89 */ {
2944 /* we still expect NUL terminated strings */
2945 /* but just in case someone tries to be evil */
2946 D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0);
2947 D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0);
2948 p->verify_alg[SHARED_SECRET_MAX-1] = 0;
2949 p->csums_alg[SHARED_SECRET_MAX-1] = 0;
2952 if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) {
2953 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2954 dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n",
2955 mdev->sync_conf.verify_alg, p->verify_alg);
2958 verify_tfm = drbd_crypto_alloc_digest_safe(mdev,
2959 p->verify_alg, "verify-alg");
2960 if (IS_ERR(verify_tfm)) {
2966 if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) {
2967 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2968 dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n",
2969 mdev->sync_conf.csums_alg, p->csums_alg);
2972 csums_tfm = drbd_crypto_alloc_digest_safe(mdev,
2973 p->csums_alg, "csums-alg");
2974 if (IS_ERR(csums_tfm)) {
2981 mdev->sync_conf.rate = be32_to_cpu(p->rate);
2982 mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead);
2983 mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target);
2984 mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target);
2985 mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate);
2987 fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
2988 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
2989 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
2991 dev_err(DEV, "kmalloc of fifo_buffer failed");
2997 spin_lock(&mdev->peer_seq_lock);
2998 /* lock against drbd_nl_syncer_conf() */
3000 strcpy(mdev->sync_conf.verify_alg, p->verify_alg);
3001 mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1;
3002 crypto_free_hash(mdev->verify_tfm);
3003 mdev->verify_tfm = verify_tfm;
3004 dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg);
3007 strcpy(mdev->sync_conf.csums_alg, p->csums_alg);
3008 mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1;
3009 crypto_free_hash(mdev->csums_tfm);
3010 mdev->csums_tfm = csums_tfm;
3011 dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg);
3013 if (fifo_size != mdev->rs_plan_s.size) {
3014 kfree(mdev->rs_plan_s.values);
3015 mdev->rs_plan_s.values = rs_plan_s;
3016 mdev->rs_plan_s.size = fifo_size;
3017 mdev->rs_planed = 0;
3019 spin_unlock(&mdev->peer_seq_lock);
3024 /* just for completeness: actually not needed,
3025 * as this is not reached if csums_tfm was ok. */
3026 crypto_free_hash(csums_tfm);
3027 /* but free the verify_tfm again, if csums_tfm did not work out */
3028 crypto_free_hash(verify_tfm);
3029 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3033 static void drbd_setup_order_type(struct drbd_conf *mdev, int peer)
3035 /* sorry, we currently have no working implementation
3036 * of distributed TCQ */
3039 /* warn if the arguments differ by more than 12.5% */
3040 static void warn_if_differ_considerably(struct drbd_conf *mdev,
3041 const char *s, sector_t a, sector_t b)
3044 if (a == 0 || b == 0)
3046 d = (a > b) ? (a - b) : (b - a);
3047 if (d > (a>>3) || d > (b>>3))
3048 dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s,
3049 (unsigned long long)a, (unsigned long long)b);
3052 static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3054 struct p_sizes *p = &mdev->data.rbuf.sizes;
3055 enum determine_dev_size dd = unchanged;
3056 unsigned int max_seg_s;
3057 sector_t p_size, p_usize, my_usize;
3058 int ldsc = 0; /* local disk size changed */
3059 enum dds_flags ddsf;
3061 p_size = be64_to_cpu(p->d_size);
3062 p_usize = be64_to_cpu(p->u_size);
3064 if (p_size == 0 && mdev->state.disk == D_DISKLESS) {
3065 dev_err(DEV, "some backing storage is needed\n");
3066 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3070 /* just store the peer's disk size for now.
3071 * we still need to figure out whether we accept that. */
3072 mdev->p_size = p_size;
3074 if (get_ldev(mdev)) {
3075 warn_if_differ_considerably(mdev, "lower level device sizes",
3076 p_size, drbd_get_max_capacity(mdev->ldev));
3077 warn_if_differ_considerably(mdev, "user requested size",
3078 p_usize, mdev->ldev->dc.disk_size);
3080 /* if this is the first connect, or an otherwise expected
3081 * param exchange, choose the minimum */
3082 if (mdev->state.conn == C_WF_REPORT_PARAMS)
3083 p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size,
3086 my_usize = mdev->ldev->dc.disk_size;
3088 if (mdev->ldev->dc.disk_size != p_usize) {
3089 mdev->ldev->dc.disk_size = p_usize;
3090 dev_info(DEV, "Peer sets u_size to %lu sectors\n",
3091 (unsigned long)mdev->ldev->dc.disk_size);
3094 /* Never shrink a device with usable data during connect.
3095 But allow online shrinking if we are connected. */
3096 if (drbd_new_dev_size(mdev, mdev->ldev, 0) <
3097 drbd_get_capacity(mdev->this_bdev) &&
3098 mdev->state.disk >= D_OUTDATED &&
3099 mdev->state.conn < C_CONNECTED) {
3100 dev_err(DEV, "The peer's disk size is too small!\n");
3101 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3102 mdev->ldev->dc.disk_size = my_usize;
3110 ddsf = be16_to_cpu(p->dds_flags);
3111 if (get_ldev(mdev)) {
3112 dd = drbd_determin_dev_size(mdev, ddsf);
3114 if (dd == dev_size_error)
3118 /* I am diskless, need to accept the peer's size. */
3119 drbd_set_my_capacity(mdev, p_size);
3122 if (get_ldev(mdev)) {
3123 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) {
3124 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
3128 if (mdev->agreed_pro_version < 94)
3129 max_seg_s = be32_to_cpu(p->max_segment_size);
3130 else if (mdev->agreed_pro_version == 94)
3131 max_seg_s = DRBD_MAX_SIZE_H80_PACKET;
3132 else /* drbd 8.3.8 onwards */
3133 max_seg_s = DRBD_MAX_SEGMENT_SIZE;
3135 if (max_seg_s != queue_max_segment_size(mdev->rq_queue))
3136 drbd_setup_queue_param(mdev, max_seg_s);
3138 drbd_setup_order_type(mdev, be16_to_cpu(p->queue_order_type));
3142 if (mdev->state.conn > C_WF_REPORT_PARAMS) {
3143 if (be64_to_cpu(p->c_size) !=
3144 drbd_get_capacity(mdev->this_bdev) || ldsc) {
3145 /* we have different sizes, probably peer
3146 * needs to know my new size... */
3147 drbd_send_sizes(mdev, 0, ddsf);
3149 if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) ||
3150 (dd == grew && mdev->state.conn == C_CONNECTED)) {
3151 if (mdev->state.pdsk >= D_INCONSISTENT &&
3152 mdev->state.disk >= D_INCONSISTENT) {
3153 if (ddsf & DDSF_NO_RESYNC)
3154 dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n");
3156 resync_after_online_grow(mdev);
3158 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
3165 static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3167 struct p_uuids *p = &mdev->data.rbuf.uuids;
3171 p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO);
3173 for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++)
3174 p_uuid[i] = be64_to_cpu(p->uuid[i]);
3176 kfree(mdev->p_uuid);
3177 mdev->p_uuid = p_uuid;
3179 if (mdev->state.conn < C_CONNECTED &&
3180 mdev->state.disk < D_INCONSISTENT &&
3181 mdev->state.role == R_PRIMARY &&
3182 (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) {
3183 dev_err(DEV, "Can only connect to data with current UUID=%016llX\n",
3184 (unsigned long long)mdev->ed_uuid);
3185 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3189 if (get_ldev(mdev)) {
3190 int skip_initial_sync =
3191 mdev->state.conn == C_CONNECTED &&
3192 mdev->agreed_pro_version >= 90 &&
3193 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
3194 (p_uuid[UI_FLAGS] & 8);
3195 if (skip_initial_sync) {
3196 dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n");
3197 drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
3198 "clear_n_write from receive_uuids");
3199 _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]);
3200 _drbd_uuid_set(mdev, UI_BITMAP, 0);
3201 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3206 } else if (mdev->state.disk < D_INCONSISTENT &&
3207 mdev->state.role == R_PRIMARY) {
3208 /* I am a diskless primary, the peer just created a new current UUID
3210 drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3213 /* Before we test for the disk state, we should wait until an eventually
3214 ongoing cluster wide state change is finished. That is important if
3215 we are primary and are detaching from our disk. We need to see the
3216 new disk state... */
3217 wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags));
3218 if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT)
3219 drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3225 * convert_state() - Converts the peer's view of the cluster state to our point of view
3226 * @ps: The state as seen by the peer.
3228 static union drbd_state convert_state(union drbd_state ps)
3230 union drbd_state ms;
3232 static enum drbd_conns c_tab[] = {
3233 [C_CONNECTED] = C_CONNECTED,
3235 [C_STARTING_SYNC_S] = C_STARTING_SYNC_T,
3236 [C_STARTING_SYNC_T] = C_STARTING_SYNC_S,
3237 [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */
3238 [C_VERIFY_S] = C_VERIFY_T,
3244 ms.conn = c_tab[ps.conn];
3249 ms.peer_isp = (ps.aftr_isp | ps.user_isp);
3254 static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3256 struct p_req_state *p = &mdev->data.rbuf.req_state;
3257 union drbd_state mask, val;
3260 mask.i = be32_to_cpu(p->mask);
3261 val.i = be32_to_cpu(p->val);
3263 if (test_bit(DISCARD_CONCURRENT, &mdev->flags) &&
3264 test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) {
3265 drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG);
3269 mask = convert_state(mask);
3270 val = convert_state(val);
3272 rv = drbd_change_state(mdev, CS_VERBOSE, mask, val);
3274 drbd_send_sr_reply(mdev, rv);
3280 static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3282 struct p_state *p = &mdev->data.rbuf.state;
3283 union drbd_state os, ns, peer_state;
3284 enum drbd_disk_state real_peer_disk;
3285 enum chg_state_flags cs_flags;
3288 peer_state.i = be32_to_cpu(p->state);
3290 real_peer_disk = peer_state.disk;
3291 if (peer_state.disk == D_NEGOTIATING) {
3292 real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT;
3293 dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk));
3296 spin_lock_irq(&mdev->req_lock);
3298 os = ns = mdev->state;
3299 spin_unlock_irq(&mdev->req_lock);
3301 /* peer says his disk is uptodate, while we think it is inconsistent,
3302 * and this happens while we think we have a sync going on. */
3303 if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE &&
3304 os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) {
3305 /* If we are (becoming) SyncSource, but peer is still in sync
3306 * preparation, ignore its uptodate-ness to avoid flapping, it
3307 * will change to inconsistent once the peer reaches active
3309 * It may have changed syncer-paused flags, however, so we
3310 * cannot ignore this completely. */
3311 if (peer_state.conn > C_CONNECTED &&
3312 peer_state.conn < C_SYNC_SOURCE)
3313 real_peer_disk = D_INCONSISTENT;
3315 /* if peer_state changes to connected at the same time,
3316 * it explicitly notifies us that it finished resync.
3317 * Maybe we should finish it up, too? */
3318 else if (os.conn >= C_SYNC_SOURCE &&
3319 peer_state.conn == C_CONNECTED) {
3320 if (drbd_bm_total_weight(mdev) <= mdev->rs_failed)
3321 drbd_resync_finished(mdev);
3326 /* peer says his disk is inconsistent, while we think it is uptodate,
3327 * and this happens while the peer still thinks we have a sync going on,
3328 * but we think we are already done with the sync.
3329 * We ignore this to avoid flapping pdsk.
3330 * This should not happen, if the peer is a recent version of drbd. */
3331 if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT &&
3332 os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE)
3333 real_peer_disk = D_UP_TO_DATE;
3335 if (ns.conn == C_WF_REPORT_PARAMS)
3336 ns.conn = C_CONNECTED;
3338 if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING &&
3339 get_ldev_if_state(mdev, D_NEGOTIATING)) {
3340 int cr; /* consider resync */
3342 /* if we established a new connection */
3343 cr = (os.conn < C_CONNECTED);
3344 /* if we had an established connection
3345 * and one of the nodes newly attaches a disk */
3346 cr |= (os.conn == C_CONNECTED &&
3347 (peer_state.disk == D_NEGOTIATING ||
3348 os.disk == D_NEGOTIATING));
3349 /* if we have both been inconsistent, and the peer has been
3350 * forced to be UpToDate with --overwrite-data */
3351 cr |= test_bit(CONSIDER_RESYNC, &mdev->flags);
3352 /* if we had been plain connected, and the admin requested to
3353 * start a sync by "invalidate" or "invalidate-remote" */
3354 cr |= (os.conn == C_CONNECTED &&
3355 (peer_state.conn >= C_STARTING_SYNC_S &&
3356 peer_state.conn <= C_WF_BITMAP_T));
3359 ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk);
3362 if (ns.conn == C_MASK) {
3363 ns.conn = C_CONNECTED;
3364 if (mdev->state.disk == D_NEGOTIATING) {
3365 drbd_force_state(mdev, NS(disk, D_DISKLESS));
3366 } else if (peer_state.disk == D_NEGOTIATING) {
3367 dev_err(DEV, "Disk attach process on the peer node was aborted.\n");
3368 peer_state.disk = D_DISKLESS;
3369 real_peer_disk = D_DISKLESS;
3371 if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags))
3373 D_ASSERT(os.conn == C_WF_REPORT_PARAMS);
3374 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3380 spin_lock_irq(&mdev->req_lock);
3381 if (mdev->state.i != os.i)
3383 clear_bit(CONSIDER_RESYNC, &mdev->flags);
3384 ns.peer = peer_state.role;
3385 ns.pdsk = real_peer_disk;
3386 ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp);
3387 if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING)
3388 ns.disk = mdev->new_state_tmp.disk;
3389 cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD);
3390 if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED &&
3391 test_bit(NEW_CUR_UUID, &mdev->flags)) {
3392 /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this
3393 for temporal network outages! */
3394 spin_unlock_irq(&mdev->req_lock);
3395 dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n");
3397 drbd_uuid_new_current(mdev);
3398 clear_bit(NEW_CUR_UUID, &mdev->flags);
3399 drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0));
3402 rv = _drbd_set_state(mdev, ns, cs_flags, NULL);
3404 spin_unlock_irq(&mdev->req_lock);
3406 if (rv < SS_SUCCESS) {
3407 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3411 if (os.conn > C_WF_REPORT_PARAMS) {
3412 if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED &&
3413 peer_state.disk != D_NEGOTIATING ) {
3414 /* we want resync, peer has not yet decided to sync... */
3415 /* Nowadays only used when forcing a node into primary role and
3416 setting its disk to UpToDate with that */
3417 drbd_send_uuids(mdev);
3418 drbd_send_state(mdev);
3422 mdev->net_conf->want_lose = 0;
3424 drbd_md_sync(mdev); /* update connected indicator, la_size, ... */
3429 static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3431 struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid;
3433 wait_event(mdev->misc_wait,
3434 mdev->state.conn == C_WF_SYNC_UUID ||
3435 mdev->state.conn < C_CONNECTED ||
3436 mdev->state.disk < D_NEGOTIATING);
3438 /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */
3440 /* Here the _drbd_uuid_ functions are right, current should
3441 _not_ be rotated into the history */
3442 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
3443 _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid));
3444 _drbd_uuid_set(mdev, UI_BITMAP, 0UL);
3446 drbd_start_resync(mdev, C_SYNC_TARGET);
3450 dev_err(DEV, "Ignoring SyncUUID packet!\n");
3455 enum receive_bitmap_ret { OK, DONE, FAILED };
3457 static enum receive_bitmap_ret
3458 receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size,
3459 unsigned long *buffer, struct bm_xfer_ctx *c)
3461 unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset);
3462 unsigned want = num_words * sizeof(long);
3464 if (want != data_size) {
3465 dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size);
3470 if (drbd_recv(mdev, buffer, want) != want)
3473 drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer);
3475 c->word_offset += num_words;
3476 c->bit_offset = c->word_offset * BITS_PER_LONG;
3477 if (c->bit_offset > c->bm_bits)
3478 c->bit_offset = c->bm_bits;
3483 static enum receive_bitmap_ret
3484 recv_bm_rle_bits(struct drbd_conf *mdev,
3485 struct p_compressed_bm *p,
3486 struct bm_xfer_ctx *c)
3488 struct bitstream bs;
3492 unsigned long s = c->bit_offset;
3494 int len = be16_to_cpu(p->head.length) - (sizeof(*p) - sizeof(p->head));
3495 int toggle = DCBP_get_start(p);
3499 bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p));
3501 bits = bitstream_get_bits(&bs, &look_ahead, 64);
3505 for (have = bits; have > 0; s += rl, toggle = !toggle) {
3506 bits = vli_decode_bits(&rl, look_ahead);
3512 if (e >= c->bm_bits) {
3513 dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e);
3516 _drbd_bm_set_bits(mdev, s, e);
3520 dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n",
3521 have, bits, look_ahead,
3522 (unsigned int)(bs.cur.b - p->code),
3523 (unsigned int)bs.buf_len);
3526 look_ahead >>= bits;
3529 bits = bitstream_get_bits(&bs, &tmp, 64 - have);
3532 look_ahead |= tmp << have;
3537 bm_xfer_ctx_bit_to_word_offset(c);
3539 return (s == c->bm_bits) ? DONE : OK;
3542 static enum receive_bitmap_ret
3543 decode_bitmap_c(struct drbd_conf *mdev,
3544 struct p_compressed_bm *p,
3545 struct bm_xfer_ctx *c)
3547 if (DCBP_get_code(p) == RLE_VLI_Bits)
3548 return recv_bm_rle_bits(mdev, p, c);
3550 /* other variants had been implemented for evaluation,
3551 * but have been dropped as this one turned out to be "best"
3552 * during all our tests. */
3554 dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding);
3555 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3559 void INFO_bm_xfer_stats(struct drbd_conf *mdev,
3560 const char *direction, struct bm_xfer_ctx *c)
3562 /* what would it take to transfer it "plaintext" */
3563 unsigned plain = sizeof(struct p_header80) *
3564 ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1)
3565 + c->bm_words * sizeof(long);
3566 unsigned total = c->bytes[0] + c->bytes[1];
3569 /* total can not be zero. but just in case: */
3573 /* don't report if not compressed */
3577 /* total < plain. check for overflow, still */
3578 r = (total > UINT_MAX/1000) ? (total / (plain/1000))
3579 : (1000 * total / plain);
3585 dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), "
3586 "total %u; compression: %u.%u%%\n",
3588 c->bytes[1], c->packets[1],
3589 c->bytes[0], c->packets[0],
3590 total, r/10, r % 10);
3593 /* Since we are processing the bitfield from lower addresses to higher,
3594 it does not matter if the process it in 32 bit chunks or 64 bit
3595 chunks as long as it is little endian. (Understand it as byte stream,
3596 beginning with the lowest byte...) If we would use big endian
3597 we would need to process it from the highest address to the lowest,
3598 in order to be agnostic to the 32 vs 64 bits issue.
3600 returns 0 on failure, 1 if we successfully received it. */
3601 static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3603 struct bm_xfer_ctx c;
3605 enum receive_bitmap_ret ret;
3607 struct p_header80 *h = &mdev->data.rbuf.header.h80;
3609 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
3611 drbd_bm_lock(mdev, "receive bitmap");
3613 /* maybe we should use some per thread scratch page,
3614 * and allocate that during initial device creation? */
3615 buffer = (unsigned long *) __get_free_page(GFP_NOIO);
3617 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__);
3621 c = (struct bm_xfer_ctx) {
3622 .bm_bits = drbd_bm_bits(mdev),
3623 .bm_words = drbd_bm_words(mdev),
3627 if (cmd == P_BITMAP) {
3628 ret = receive_bitmap_plain(mdev, data_size, buffer, &c);
3629 } else if (cmd == P_COMPRESSED_BITMAP) {
3630 /* MAYBE: sanity check that we speak proto >= 90,
3631 * and the feature is enabled! */
3632 struct p_compressed_bm *p;
3634 if (data_size > BM_PACKET_PAYLOAD_BYTES) {
3635 dev_err(DEV, "ReportCBitmap packet too large\n");
3638 /* use the page buff */
3640 memcpy(p, h, sizeof(*h));
3641 if (drbd_recv(mdev, p->head.payload, data_size) != data_size)
3643 if (data_size <= (sizeof(*p) - sizeof(p->head))) {
3644 dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size);
3647 ret = decode_bitmap_c(mdev, p, &c);
3649 dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd);
3653 c.packets[cmd == P_BITMAP]++;
3654 c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size;
3659 if (!drbd_recv_header(mdev, &cmd, &data_size))
3661 } while (ret == OK);
3665 INFO_bm_xfer_stats(mdev, "receive", &c);
3667 if (mdev->state.conn == C_WF_BITMAP_T) {
3668 ok = !drbd_send_bitmap(mdev);
3671 /* Omit CS_ORDERED with this state transition to avoid deadlocks. */
3672 ok = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
3673 D_ASSERT(ok == SS_SUCCESS);
3674 } else if (mdev->state.conn != C_WF_BITMAP_S) {
3675 /* admin may have requested C_DISCONNECTING,
3676 * other threads may have noticed network errors */
3677 dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n",
3678 drbd_conn_str(mdev->state.conn));
3683 drbd_bm_unlock(mdev);
3684 if (ok && mdev->state.conn == C_WF_BITMAP_S)
3685 drbd_start_resync(mdev, C_SYNC_SOURCE);
3686 free_page((unsigned long) buffer);
3690 static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3692 /* TODO zero copy sink :) */
3693 static char sink[128];
3696 dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n",
3701 want = min_t(int, size, sizeof(sink));
3702 r = drbd_recv(mdev, sink, want);
3703 ERR_IF(r <= 0) break;
3709 static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3711 if (mdev->state.disk >= D_INCONSISTENT)
3714 /* Make sure we've acked all the TCP data associated
3715 * with the data requests being unplugged */
3716 drbd_tcp_quickack(mdev->data.socket);
3721 typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive);
3726 drbd_cmd_handler_f function;
3729 static struct data_cmd drbd_cmd_handler[] = {
3730 [P_DATA] = { 1, sizeof(struct p_data), receive_Data },
3731 [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply },
3732 [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } ,
3733 [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } ,
3734 [P_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } ,
3735 [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } ,
3736 [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header80), receive_UnplugRemote },
3737 [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3738 [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3739 [P_SYNC_PARAM] = { 1, sizeof(struct p_header80), receive_SyncParam },
3740 [P_SYNC_PARAM89] = { 1, sizeof(struct p_header80), receive_SyncParam },
3741 [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol },
3742 [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids },
3743 [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes },
3744 [P_STATE] = { 0, sizeof(struct p_state), receive_state },
3745 [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state },
3746 [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid },
3747 [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3748 [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest },
3749 [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest },
3750 [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip },
3751 /* anything missing from this table is in
3752 * the asender_tbl, see get_asender_cmd */
3753 [P_MAX_CMD] = { 0, 0, NULL },
3756 /* All handler functions that expect a sub-header get that sub-heder in
3757 mdev->data.rbuf.header.head.payload.
3759 Usually in mdev->data.rbuf.header.head the callback can find the usual
3760 p_header, but they may not rely on that. Since there is also p_header95 !
3763 static void drbdd(struct drbd_conf *mdev)
3765 union p_header *header = &mdev->data.rbuf.header;
3766 unsigned int packet_size;
3767 enum drbd_packets cmd;
3768 size_t shs; /* sub header size */
3771 while (get_t_state(&mdev->receiver) == Running) {
3772 drbd_thread_current_set_cpu(mdev);
3773 if (!drbd_recv_header(mdev, &cmd, &packet_size))
3776 if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) {
3777 dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size);
3781 shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header);
3782 rv = drbd_recv(mdev, &header->h80.payload, shs);
3783 if (unlikely(rv != shs)) {
3784 dev_err(DEV, "short read while reading sub header: rv=%d\n", rv);
3788 if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) {
3789 dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size);
3793 rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs);
3795 if (unlikely(!rv)) {
3796 dev_err(DEV, "error receiving %s, l: %d!\n",
3797 cmdname(cmd), packet_size);
3804 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3806 /* If we leave here, we probably want to update at least the
3807 * "Connected" indicator on stable storage. Do so explicitly here. */
3811 void drbd_flush_workqueue(struct drbd_conf *mdev)
3813 struct drbd_wq_barrier barr;
3815 barr.w.cb = w_prev_work_done;
3816 init_completion(&barr.done);
3817 drbd_queue_work(&mdev->data.work, &barr.w);
3818 wait_for_completion(&barr.done);
3821 void drbd_free_tl_hash(struct drbd_conf *mdev)
3823 struct hlist_head *h;
3825 spin_lock_irq(&mdev->req_lock);
3827 if (!mdev->tl_hash || mdev->state.conn != C_STANDALONE) {
3828 spin_unlock_irq(&mdev->req_lock);
3832 for (h = mdev->ee_hash; h < mdev->ee_hash + mdev->ee_hash_s; h++)
3834 dev_err(DEV, "ASSERT FAILED ee_hash[%u].first == %p, expected NULL\n",
3835 (int)(h - mdev->ee_hash), h->first);
3836 kfree(mdev->ee_hash);
3837 mdev->ee_hash = NULL;
3838 mdev->ee_hash_s = 0;
3841 for (h = mdev->tl_hash; h < mdev->tl_hash + mdev->tl_hash_s; h++)
3843 dev_err(DEV, "ASSERT FAILED tl_hash[%u] == %p, expected NULL\n",
3844 (int)(h - mdev->tl_hash), h->first);
3845 kfree(mdev->tl_hash);
3846 mdev->tl_hash = NULL;
3847 mdev->tl_hash_s = 0;
3848 spin_unlock_irq(&mdev->req_lock);
3851 static void drbd_disconnect(struct drbd_conf *mdev)
3853 enum drbd_fencing_p fp;
3854 union drbd_state os, ns;
3855 int rv = SS_UNKNOWN_ERROR;
3858 if (mdev->state.conn == C_STANDALONE)
3860 if (mdev->state.conn >= C_WF_CONNECTION)
3861 dev_err(DEV, "ASSERT FAILED cstate = %s, expected < WFConnection\n",
3862 drbd_conn_str(mdev->state.conn));
3864 /* asender does not clean up anything. it must not interfere, either */
3865 drbd_thread_stop(&mdev->asender);
3866 drbd_free_sock(mdev);
3868 /* wait for current activity to cease. */
3869 spin_lock_irq(&mdev->req_lock);
3870 _drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
3871 _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee);
3872 _drbd_wait_ee_list_empty(mdev, &mdev->read_ee);
3873 spin_unlock_irq(&mdev->req_lock);
3875 /* We do not have data structures that would allow us to
3876 * get the rs_pending_cnt down to 0 again.
3877 * * On C_SYNC_TARGET we do not have any data structures describing
3878 * the pending RSDataRequest's we have sent.
3879 * * On C_SYNC_SOURCE there is no data structure that tracks
3880 * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget.
3881 * And no, it is not the sum of the reference counts in the
3882 * resync_LRU. The resync_LRU tracks the whole operation including
3883 * the disk-IO, while the rs_pending_cnt only tracks the blocks
3885 drbd_rs_cancel_all(mdev);
3887 mdev->rs_failed = 0;
3888 atomic_set(&mdev->rs_pending_cnt, 0);
3889 wake_up(&mdev->misc_wait);
3891 /* make sure syncer is stopped and w_resume_next_sg queued */
3892 del_timer_sync(&mdev->resync_timer);
3893 resync_timer_fn((unsigned long)mdev);
3895 /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier,
3896 * w_make_resync_request etc. which may still be on the worker queue
3897 * to be "canceled" */
3898 drbd_flush_workqueue(mdev);
3900 /* This also does reclaim_net_ee(). If we do this too early, we might
3901 * miss some resync ee and pages.*/
3902 drbd_process_done_ee(mdev);
3904 kfree(mdev->p_uuid);
3905 mdev->p_uuid = NULL;
3907 if (!is_susp(mdev->state))
3910 dev_info(DEV, "Connection closed\n");
3915 if (get_ldev(mdev)) {
3916 fp = mdev->ldev->dc.fencing;
3920 if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN)
3921 drbd_try_outdate_peer_async(mdev);
3923 spin_lock_irq(&mdev->req_lock);
3925 if (os.conn >= C_UNCONNECTED) {
3926 /* Do not restart in case we are C_DISCONNECTING */
3928 ns.conn = C_UNCONNECTED;
3929 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
3931 spin_unlock_irq(&mdev->req_lock);
3933 if (os.conn == C_DISCONNECTING) {
3934 wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0);
3936 if (!is_susp(mdev->state)) {
3937 /* we must not free the tl_hash
3938 * while application io is still on the fly */
3939 wait_event(mdev->misc_wait, !atomic_read(&mdev->ap_bio_cnt));
3940 drbd_free_tl_hash(mdev);
3943 crypto_free_hash(mdev->cram_hmac_tfm);
3944 mdev->cram_hmac_tfm = NULL;
3946 kfree(mdev->net_conf);
3947 mdev->net_conf = NULL;
3948 drbd_request_state(mdev, NS(conn, C_STANDALONE));
3951 /* tcp_close and release of sendpage pages can be deferred. I don't
3952 * want to use SO_LINGER, because apparently it can be deferred for
3953 * more than 20 seconds (longest time I checked).
3955 * Actually we don't care for exactly when the network stack does its
3956 * put_page(), but release our reference on these pages right here.
3958 i = drbd_release_ee(mdev, &mdev->net_ee);
3960 dev_info(DEV, "net_ee not empty, killed %u entries\n", i);
3961 i = atomic_read(&mdev->pp_in_use_by_net);
3963 dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i);
3964 i = atomic_read(&mdev->pp_in_use);
3966 dev_info(DEV, "pp_in_use = %d, expected 0\n", i);
3968 D_ASSERT(list_empty(&mdev->read_ee));
3969 D_ASSERT(list_empty(&mdev->active_ee));
3970 D_ASSERT(list_empty(&mdev->sync_ee));
3971 D_ASSERT(list_empty(&mdev->done_ee));
3973 /* ok, no more ee's on the fly, it is safe to reset the epoch_size */
3974 atomic_set(&mdev->current_epoch->epoch_size, 0);
3975 D_ASSERT(list_empty(&mdev->current_epoch->list));
3979 * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version
3980 * we can agree on is stored in agreed_pro_version.
3982 * feature flags and the reserved array should be enough room for future
3983 * enhancements of the handshake protocol, and possible plugins...
3985 * for now, they are expected to be zero, but ignored.
3987 static int drbd_send_handshake(struct drbd_conf *mdev)
3989 /* ASSERT current == mdev->receiver ... */
3990 struct p_handshake *p = &mdev->data.sbuf.handshake;
3993 if (mutex_lock_interruptible(&mdev->data.mutex)) {
3994 dev_err(DEV, "interrupted during initial handshake\n");
3995 return 0; /* interrupted. not ok. */
3998 if (mdev->data.socket == NULL) {
3999 mutex_unlock(&mdev->data.mutex);
4003 memset(p, 0, sizeof(*p));
4004 p->protocol_min = cpu_to_be32(PRO_VERSION_MIN);
4005 p->protocol_max = cpu_to_be32(PRO_VERSION_MAX);
4006 ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE,
4007 (struct p_header80 *)p, sizeof(*p), 0 );
4008 mutex_unlock(&mdev->data.mutex);
4014 * 1 yes, we have a valid connection
4015 * 0 oops, did not work out, please try again
4016 * -1 peer talks different language,
4017 * no point in trying again, please go standalone.
4019 static int drbd_do_handshake(struct drbd_conf *mdev)
4021 /* ASSERT current == mdev->receiver ... */
4022 struct p_handshake *p = &mdev->data.rbuf.handshake;
4023 const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80);
4024 unsigned int length;
4025 enum drbd_packets cmd;
4028 rv = drbd_send_handshake(mdev);
4032 rv = drbd_recv_header(mdev, &cmd, &length);
4036 if (cmd != P_HAND_SHAKE) {
4037 dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n",
4042 if (length != expect) {
4043 dev_err(DEV, "expected HandShake length: %u, received: %u\n",
4048 rv = drbd_recv(mdev, &p->head.payload, expect);
4051 dev_err(DEV, "short read receiving handshake packet: l=%u\n", rv);
4055 p->protocol_min = be32_to_cpu(p->protocol_min);
4056 p->protocol_max = be32_to_cpu(p->protocol_max);
4057 if (p->protocol_max == 0)
4058 p->protocol_max = p->protocol_min;
4060 if (PRO_VERSION_MAX < p->protocol_min ||
4061 PRO_VERSION_MIN > p->protocol_max)
4064 mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max);
4066 dev_info(DEV, "Handshake successful: "
4067 "Agreed network protocol version %d\n", mdev->agreed_pro_version);
4072 dev_err(DEV, "incompatible DRBD dialects: "
4073 "I support %d-%d, peer supports %d-%d\n",
4074 PRO_VERSION_MIN, PRO_VERSION_MAX,
4075 p->protocol_min, p->protocol_max);
4079 #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE)
4080 static int drbd_do_auth(struct drbd_conf *mdev)
4082 dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n");
4083 dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n");
4087 #define CHALLENGE_LEN 64
4091 0 - failed, try again (network error),
4092 -1 - auth failed, don't try again.
4095 static int drbd_do_auth(struct drbd_conf *mdev)
4097 char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */
4098 struct scatterlist sg;
4099 char *response = NULL;
4100 char *right_response = NULL;
4101 char *peers_ch = NULL;
4102 unsigned int key_len = strlen(mdev->net_conf->shared_secret);
4103 unsigned int resp_size;
4104 struct hash_desc desc;
4105 enum drbd_packets cmd;
4106 unsigned int length;
4109 desc.tfm = mdev->cram_hmac_tfm;
4112 rv = crypto_hash_setkey(mdev->cram_hmac_tfm,
4113 (u8 *)mdev->net_conf->shared_secret, key_len);
4115 dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv);
4120 get_random_bytes(my_challenge, CHALLENGE_LEN);
4122 rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN);
4126 rv = drbd_recv_header(mdev, &cmd, &length);
4130 if (cmd != P_AUTH_CHALLENGE) {
4131 dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n",
4137 if (length > CHALLENGE_LEN * 2) {
4138 dev_err(DEV, "expected AuthChallenge payload too big.\n");
4143 peers_ch = kmalloc(length, GFP_NOIO);
4144 if (peers_ch == NULL) {
4145 dev_err(DEV, "kmalloc of peers_ch failed\n");
4150 rv = drbd_recv(mdev, peers_ch, length);
4153 dev_err(DEV, "short read AuthChallenge: l=%u\n", rv);
4158 resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm);
4159 response = kmalloc(resp_size, GFP_NOIO);
4160 if (response == NULL) {
4161 dev_err(DEV, "kmalloc of response failed\n");
4166 sg_init_table(&sg, 1);
4167 sg_set_buf(&sg, peers_ch, length);
4169 rv = crypto_hash_digest(&desc, &sg, sg.length, response);
4171 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4176 rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size);
4180 rv = drbd_recv_header(mdev, &cmd, &length);
4184 if (cmd != P_AUTH_RESPONSE) {
4185 dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n",
4191 if (length != resp_size) {
4192 dev_err(DEV, "expected AuthResponse payload of wrong size\n");
4197 rv = drbd_recv(mdev, response , resp_size);
4199 if (rv != resp_size) {
4200 dev_err(DEV, "short read receiving AuthResponse: l=%u\n", rv);
4205 right_response = kmalloc(resp_size, GFP_NOIO);
4206 if (right_response == NULL) {
4207 dev_err(DEV, "kmalloc of right_response failed\n");
4212 sg_set_buf(&sg, my_challenge, CHALLENGE_LEN);
4214 rv = crypto_hash_digest(&desc, &sg, sg.length, right_response);
4216 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4221 rv = !memcmp(response, right_response, resp_size);
4224 dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n",
4225 resp_size, mdev->net_conf->cram_hmac_alg);
4232 kfree(right_response);
4238 int drbdd_init(struct drbd_thread *thi)
4240 struct drbd_conf *mdev = thi->mdev;
4241 unsigned int minor = mdev_to_minor(mdev);
4244 sprintf(current->comm, "drbd%d_receiver", minor);
4246 dev_info(DEV, "receiver (re)started\n");
4249 h = drbd_connect(mdev);
4251 drbd_disconnect(mdev);
4252 __set_current_state(TASK_INTERRUPTIBLE);
4253 schedule_timeout(HZ);
4256 dev_warn(DEV, "Discarding network configuration.\n");
4257 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4262 if (get_net_conf(mdev)) {
4268 drbd_disconnect(mdev);
4270 dev_info(DEV, "receiver terminated\n");
4274 /* ********* acknowledge sender ******** */
4276 static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h)
4278 struct p_req_state_reply *p = (struct p_req_state_reply *)h;
4280 int retcode = be32_to_cpu(p->retcode);
4282 if (retcode >= SS_SUCCESS) {
4283 set_bit(CL_ST_CHG_SUCCESS, &mdev->flags);
4285 set_bit(CL_ST_CHG_FAIL, &mdev->flags);
4286 dev_err(DEV, "Requested state change failed by peer: %s (%d)\n",
4287 drbd_set_st_err_str(retcode), retcode);
4289 wake_up(&mdev->state_wait);
4294 static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h)
4296 return drbd_send_ping_ack(mdev);
4300 static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h)
4302 /* restore idle timeout */
4303 mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
4304 if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags))
4305 wake_up(&mdev->misc_wait);
4310 static int got_IsInSync(struct drbd_conf *mdev, struct p_header80 *h)
4312 struct p_block_ack *p = (struct p_block_ack *)h;
4313 sector_t sector = be64_to_cpu(p->sector);
4314 int blksize = be32_to_cpu(p->blksize);
4316 D_ASSERT(mdev->agreed_pro_version >= 89);
4318 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4320 if (get_ldev(mdev)) {
4321 drbd_rs_complete_io(mdev, sector);
4322 drbd_set_in_sync(mdev, sector, blksize);
4323 /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */
4324 mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT);
4327 dec_rs_pending(mdev);
4328 atomic_add(blksize >> 9, &mdev->rs_sect_in);
4333 /* when we receive the ACK for a write request,
4334 * verify that we actually know about it */
4335 static struct drbd_request *_ack_id_to_req(struct drbd_conf *mdev,
4336 u64 id, sector_t sector)
4338 struct hlist_head *slot = tl_hash_slot(mdev, sector);
4339 struct hlist_node *n;
4340 struct drbd_request *req;
4342 hlist_for_each_entry(req, n, slot, colision) {
4343 if ((unsigned long)req == (unsigned long)id) {
4344 if (req->sector != sector) {
4345 dev_err(DEV, "_ack_id_to_req: found req %p but it has "
4346 "wrong sector (%llus versus %llus)\n", req,
4347 (unsigned long long)req->sector,
4348 (unsigned long long)sector);
4354 dev_err(DEV, "_ack_id_to_req: failed to find req %p, sector %llus in list\n",
4355 (void *)(unsigned long)id, (unsigned long long)sector);
4359 typedef struct drbd_request *(req_validator_fn)
4360 (struct drbd_conf *mdev, u64 id, sector_t sector);
4362 static int validate_req_change_req_state(struct drbd_conf *mdev,
4363 u64 id, sector_t sector, req_validator_fn validator,
4364 const char *func, enum drbd_req_event what)
4366 struct drbd_request *req;
4367 struct bio_and_error m;
4369 spin_lock_irq(&mdev->req_lock);
4370 req = validator(mdev, id, sector);
4371 if (unlikely(!req)) {
4372 spin_unlock_irq(&mdev->req_lock);
4373 dev_err(DEV, "%s: got a corrupt block_id/sector pair\n", func);
4376 __req_mod(req, what, &m);
4377 spin_unlock_irq(&mdev->req_lock);
4380 complete_master_bio(mdev, &m);
4384 static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h)
4386 struct p_block_ack *p = (struct p_block_ack *)h;
4387 sector_t sector = be64_to_cpu(p->sector);
4388 int blksize = be32_to_cpu(p->blksize);
4389 enum drbd_req_event what;
4391 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4393 if (is_syncer_block_id(p->block_id)) {
4394 drbd_set_in_sync(mdev, sector, blksize);
4395 dec_rs_pending(mdev);
4398 switch (be16_to_cpu(h->command)) {
4399 case P_RS_WRITE_ACK:
4400 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4401 what = write_acked_by_peer_and_sis;
4404 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4405 what = write_acked_by_peer;
4408 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B);
4409 what = recv_acked_by_peer;
4412 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4413 what = conflict_discarded_by_peer;
4420 return validate_req_change_req_state(mdev, p->block_id, sector,
4421 _ack_id_to_req, __func__ , what);
4424 static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h)
4426 struct p_block_ack *p = (struct p_block_ack *)h;
4427 sector_t sector = be64_to_cpu(p->sector);
4429 if (__ratelimit(&drbd_ratelimit_state))
4430 dev_warn(DEV, "Got NegAck packet. Peer is in troubles?\n");
4432 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4434 if (is_syncer_block_id(p->block_id)) {
4435 int size = be32_to_cpu(p->blksize);
4436 dec_rs_pending(mdev);
4437 drbd_rs_failed_io(mdev, sector, size);
4440 return validate_req_change_req_state(mdev, p->block_id, sector,
4441 _ack_id_to_req, __func__ , neg_acked);
4444 static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h)
4446 struct p_block_ack *p = (struct p_block_ack *)h;
4447 sector_t sector = be64_to_cpu(p->sector);
4449 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4450 dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n",
4451 (unsigned long long)sector, be32_to_cpu(p->blksize));
4453 return validate_req_change_req_state(mdev, p->block_id, sector,
4454 _ar_id_to_req, __func__ , neg_acked);
4457 static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h)
4461 struct p_block_ack *p = (struct p_block_ack *)h;
4463 sector = be64_to_cpu(p->sector);
4464 size = be32_to_cpu(p->blksize);
4466 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4468 dec_rs_pending(mdev);
4470 if (get_ldev_if_state(mdev, D_FAILED)) {
4471 drbd_rs_complete_io(mdev, sector);
4472 drbd_rs_failed_io(mdev, sector, size);
4479 static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h)
4481 struct p_barrier_ack *p = (struct p_barrier_ack *)h;
4483 tl_release(mdev, p->barrier, be32_to_cpu(p->set_size));
4488 static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h)
4490 struct p_block_ack *p = (struct p_block_ack *)h;
4491 struct drbd_work *w;
4495 sector = be64_to_cpu(p->sector);
4496 size = be32_to_cpu(p->blksize);
4498 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4500 if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC)
4501 drbd_ov_oos_found(mdev, sector, size);
4505 if (!get_ldev(mdev))
4508 drbd_rs_complete_io(mdev, sector);
4509 dec_rs_pending(mdev);
4511 if (--mdev->ov_left == 0) {
4512 w = kmalloc(sizeof(*w), GFP_NOIO);
4514 w->cb = w_ov_finished;
4515 drbd_queue_work_front(&mdev->data.work, w);
4517 dev_err(DEV, "kmalloc(w) failed.");
4519 drbd_resync_finished(mdev);
4526 static int got_skip(struct drbd_conf *mdev, struct p_header80 *h)
4531 struct asender_cmd {
4533 int (*process)(struct drbd_conf *mdev, struct p_header80 *h);
4536 static struct asender_cmd *get_asender_cmd(int cmd)
4538 static struct asender_cmd asender_tbl[] = {
4539 /* anything missing from this table is in
4540 * the drbd_cmd_handler (drbd_default_handler) table,
4541 * see the beginning of drbdd() */
4542 [P_PING] = { sizeof(struct p_header80), got_Ping },
4543 [P_PING_ACK] = { sizeof(struct p_header80), got_PingAck },
4544 [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4545 [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4546 [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4547 [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4548 [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck },
4549 [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply },
4550 [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply},
4551 [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult },
4552 [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck },
4553 [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply },
4554 [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync },
4555 [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip },
4556 [P_MAX_CMD] = { 0, NULL },
4558 if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL)
4560 return &asender_tbl[cmd];
4563 int drbd_asender(struct drbd_thread *thi)
4565 struct drbd_conf *mdev = thi->mdev;
4566 struct p_header80 *h = &mdev->meta.rbuf.header.h80;
4567 struct asender_cmd *cmd = NULL;
4572 int expect = sizeof(struct p_header80);
4575 sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev));
4577 current->policy = SCHED_RR; /* Make this a realtime task! */
4578 current->rt_priority = 2; /* more important than all other tasks */
4580 while (get_t_state(thi) == Running) {
4581 drbd_thread_current_set_cpu(mdev);
4582 if (test_and_clear_bit(SEND_PING, &mdev->flags)) {
4583 ERR_IF(!drbd_send_ping(mdev)) goto reconnect;
4584 mdev->meta.socket->sk->sk_rcvtimeo =
4585 mdev->net_conf->ping_timeo*HZ/10;
4588 /* conditionally cork;
4589 * it may hurt latency if we cork without much to send */
4590 if (!mdev->net_conf->no_cork &&
4591 3 < atomic_read(&mdev->unacked_cnt))
4592 drbd_tcp_cork(mdev->meta.socket);
4594 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4595 flush_signals(current);
4596 if (!drbd_process_done_ee(mdev))
4598 /* to avoid race with newly queued ACKs */
4599 set_bit(SIGNAL_ASENDER, &mdev->flags);
4600 spin_lock_irq(&mdev->req_lock);
4601 empty = list_empty(&mdev->done_ee);
4602 spin_unlock_irq(&mdev->req_lock);
4603 /* new ack may have been queued right here,
4604 * but then there is also a signal pending,
4605 * and we start over... */
4609 /* but unconditionally uncork unless disabled */
4610 if (!mdev->net_conf->no_cork)
4611 drbd_tcp_uncork(mdev->meta.socket);
4613 /* short circuit, recv_msg would return EINTR anyways. */
4614 if (signal_pending(current))
4617 rv = drbd_recv_short(mdev, mdev->meta.socket,
4618 buf, expect-received, 0);
4619 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4621 flush_signals(current);
4624 * -EINTR (on meta) we got a signal
4625 * -EAGAIN (on meta) rcvtimeo expired
4626 * -ECONNRESET other side closed the connection
4627 * -ERESTARTSYS (on data) we got a signal
4628 * rv < 0 other than above: unexpected error!
4629 * rv == expected: full header or command
4630 * rv < expected: "woken" by signal during receive
4631 * rv == 0 : "connection shut down by peer"
4633 if (likely(rv > 0)) {
4636 } else if (rv == 0) {
4637 dev_err(DEV, "meta connection shut down by peer.\n");
4639 } else if (rv == -EAGAIN) {
4640 if (mdev->meta.socket->sk->sk_rcvtimeo ==
4641 mdev->net_conf->ping_timeo*HZ/10) {
4642 dev_err(DEV, "PingAck did not arrive in time.\n");
4645 set_bit(SEND_PING, &mdev->flags);
4647 } else if (rv == -EINTR) {
4650 dev_err(DEV, "sock_recvmsg returned %d\n", rv);
4654 if (received == expect && cmd == NULL) {
4655 if (unlikely(h->magic != BE_DRBD_MAGIC)) {
4656 dev_err(DEV, "magic?? on meta m: 0x%08x c: %d l: %d\n",
4657 be32_to_cpu(h->magic),
4658 be16_to_cpu(h->command),
4659 be16_to_cpu(h->length));
4662 cmd = get_asender_cmd(be16_to_cpu(h->command));
4663 len = be16_to_cpu(h->length);
4664 if (unlikely(cmd == NULL)) {
4665 dev_err(DEV, "unknown command?? on meta m: 0x%08x c: %d l: %d\n",
4666 be32_to_cpu(h->magic),
4667 be16_to_cpu(h->command),
4668 be16_to_cpu(h->length));
4671 expect = cmd->pkt_size;
4672 ERR_IF(len != expect-sizeof(struct p_header80))
4675 if (received == expect) {
4676 D_ASSERT(cmd != NULL);
4677 if (!cmd->process(mdev, h))
4682 expect = sizeof(struct p_header80);
4689 drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE));
4694 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4697 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4699 D_ASSERT(mdev->state.conn < C_CONNECTED);
4700 dev_info(DEV, "asender terminated\n");