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/pkt_sched.h>
40 #define __KERNEL_SYSCALLS__
41 #include <linux/unistd.h>
42 #include <linux/vmalloc.h>
43 #include <linux/random.h>
44 #include <linux/string.h>
45 #include <linux/scatterlist.h>
57 static int drbd_do_handshake(struct drbd_conf *mdev);
58 static int drbd_do_auth(struct drbd_conf *mdev);
60 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event);
61 static int e_end_block(struct drbd_conf *, struct drbd_work *, int);
64 #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN)
67 * some helper functions to deal with single linked page lists,
68 * page->private being our "next" pointer.
71 /* If at least n pages are linked at head, get n pages off.
72 * Otherwise, don't modify head, and return NULL.
73 * Locking is the responsibility of the caller.
75 static struct page *page_chain_del(struct page **head, int n)
89 tmp = page_chain_next(page);
91 break; /* found sufficient pages */
93 /* insufficient pages, don't use any of them. */
98 /* add end of list marker for the returned list */
99 set_page_private(page, 0);
100 /* actual return value, and adjustment of head */
106 /* may be used outside of locks to find the tail of a (usually short)
107 * "private" page chain, before adding it back to a global chain head
108 * with page_chain_add() under a spinlock. */
109 static struct page *page_chain_tail(struct page *page, int *len)
113 while ((tmp = page_chain_next(page)))
120 static int page_chain_free(struct page *page)
124 page_chain_for_each_safe(page, tmp) {
131 static void page_chain_add(struct page **head,
132 struct page *chain_first, struct page *chain_last)
136 tmp = page_chain_tail(chain_first, NULL);
137 BUG_ON(tmp != chain_last);
140 /* add chain to head */
141 set_page_private(chain_last, (unsigned long)*head);
145 static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number)
147 struct page *page = NULL;
148 struct page *tmp = NULL;
151 /* Yes, testing drbd_pp_vacant outside the lock is racy.
152 * So what. It saves a spin_lock. */
153 if (drbd_pp_vacant >= number) {
154 spin_lock(&drbd_pp_lock);
155 page = page_chain_del(&drbd_pp_pool, number);
157 drbd_pp_vacant -= number;
158 spin_unlock(&drbd_pp_lock);
163 /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD
164 * "criss-cross" setup, that might cause write-out on some other DRBD,
165 * which in turn might block on the other node at this very place. */
166 for (i = 0; i < number; i++) {
167 tmp = alloc_page(GFP_TRY);
170 set_page_private(tmp, (unsigned long)page);
177 /* Not enough pages immediately available this time.
178 * No need to jump around here, drbd_pp_alloc will retry this
179 * function "soon". */
181 tmp = page_chain_tail(page, NULL);
182 spin_lock(&drbd_pp_lock);
183 page_chain_add(&drbd_pp_pool, page, tmp);
185 spin_unlock(&drbd_pp_lock);
190 static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed)
192 struct drbd_epoch_entry *e;
193 struct list_head *le, *tle;
195 /* The EEs are always appended to the end of the list. Since
196 they are sent in order over the wire, they have to finish
197 in order. As soon as we see the first not finished we can
198 stop to examine the list... */
200 list_for_each_safe(le, tle, &mdev->net_ee) {
201 e = list_entry(le, struct drbd_epoch_entry, w.list);
202 if (drbd_ee_has_active_page(e))
204 list_move(le, to_be_freed);
208 static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev)
210 LIST_HEAD(reclaimed);
211 struct drbd_epoch_entry *e, *t;
213 spin_lock_irq(&mdev->req_lock);
214 reclaim_net_ee(mdev, &reclaimed);
215 spin_unlock_irq(&mdev->req_lock);
217 list_for_each_entry_safe(e, t, &reclaimed, w.list)
218 drbd_free_net_ee(mdev, e);
222 * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled)
223 * @mdev: DRBD device.
224 * @number: number of pages requested
225 * @retry: whether to retry, if not enough pages are available right now
227 * Tries to allocate number pages, first from our own page pool, then from
228 * the kernel, unless this allocation would exceed the max_buffers setting.
229 * Possibly retry until DRBD frees sufficient pages somewhere else.
231 * Returns a page chain linked via page->private.
233 static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry)
235 struct page *page = NULL;
238 /* Yes, we may run up to @number over max_buffers. If we
239 * follow it strictly, the admin will get it wrong anyways. */
240 if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers)
241 page = drbd_pp_first_pages_or_try_alloc(mdev, number);
243 while (page == NULL) {
244 prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE);
246 drbd_kick_lo_and_reclaim_net(mdev);
248 if (atomic_read(&mdev->pp_in_use) < mdev->net_conf->max_buffers) {
249 page = drbd_pp_first_pages_or_try_alloc(mdev, number);
257 if (signal_pending(current)) {
258 dev_warn(DEV, "drbd_pp_alloc interrupted!\n");
264 finish_wait(&drbd_pp_wait, &wait);
267 atomic_add(number, &mdev->pp_in_use);
271 /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc.
272 * Is also used from inside an other spin_lock_irq(&mdev->req_lock);
273 * Either links the page chain back to the global pool,
274 * or returns all pages to the system. */
275 static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net)
277 atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use;
280 if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE)*minor_count)
281 i = page_chain_free(page);
284 tmp = page_chain_tail(page, &i);
285 spin_lock(&drbd_pp_lock);
286 page_chain_add(&drbd_pp_pool, page, tmp);
288 spin_unlock(&drbd_pp_lock);
290 i = atomic_sub_return(i, a);
292 dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n",
293 is_net ? "pp_in_use_by_net" : "pp_in_use", i);
294 wake_up(&drbd_pp_wait);
298 You need to hold the req_lock:
299 _drbd_wait_ee_list_empty()
301 You must not have the req_lock:
307 drbd_process_done_ee()
309 drbd_wait_ee_list_empty()
312 struct drbd_epoch_entry *drbd_alloc_ee(struct drbd_conf *mdev,
315 unsigned int data_size,
316 gfp_t gfp_mask) __must_hold(local)
318 struct drbd_epoch_entry *e;
320 unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT;
322 if (drbd_insert_fault(mdev, DRBD_FAULT_AL_EE))
325 e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM);
327 if (!(gfp_mask & __GFP_NOWARN))
328 dev_err(DEV, "alloc_ee: Allocation of an EE failed\n");
332 page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT));
336 INIT_HLIST_NODE(&e->collision);
340 atomic_set(&e->pending_bios, 0);
345 * The block_id is opaque to the receiver. It is not endianness
346 * converted, and sent back to the sender unchanged.
353 mempool_free(e, drbd_ee_mempool);
357 void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e, int is_net)
359 if (e->flags & EE_HAS_DIGEST)
361 drbd_pp_free(mdev, e->pages, is_net);
362 D_ASSERT(atomic_read(&e->pending_bios) == 0);
363 D_ASSERT(hlist_unhashed(&e->collision));
364 mempool_free(e, drbd_ee_mempool);
367 int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list)
369 LIST_HEAD(work_list);
370 struct drbd_epoch_entry *e, *t;
372 int is_net = list == &mdev->net_ee;
374 spin_lock_irq(&mdev->req_lock);
375 list_splice_init(list, &work_list);
376 spin_unlock_irq(&mdev->req_lock);
378 list_for_each_entry_safe(e, t, &work_list, w.list) {
379 drbd_free_some_ee(mdev, e, is_net);
387 * This function is called from _asender only_
388 * but see also comments in _req_mod(,barrier_acked)
389 * and receive_Barrier.
391 * Move entries from net_ee to done_ee, if ready.
392 * Grab done_ee, call all callbacks, free the entries.
393 * The callbacks typically send out ACKs.
395 static int drbd_process_done_ee(struct drbd_conf *mdev)
397 LIST_HEAD(work_list);
398 LIST_HEAD(reclaimed);
399 struct drbd_epoch_entry *e, *t;
400 int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS);
402 spin_lock_irq(&mdev->req_lock);
403 reclaim_net_ee(mdev, &reclaimed);
404 list_splice_init(&mdev->done_ee, &work_list);
405 spin_unlock_irq(&mdev->req_lock);
407 list_for_each_entry_safe(e, t, &reclaimed, w.list)
408 drbd_free_net_ee(mdev, e);
410 /* possible callbacks here:
411 * e_end_block, and e_end_resync_block, e_send_discard_ack.
412 * all ignore the last argument.
414 list_for_each_entry_safe(e, t, &work_list, w.list) {
415 /* list_del not necessary, next/prev members not touched */
416 ok = e->w.cb(mdev, &e->w, !ok) && ok;
417 drbd_free_ee(mdev, e);
419 wake_up(&mdev->ee_wait);
424 void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head)
428 /* avoids spin_lock/unlock
429 * and calling prepare_to_wait in the fast path */
430 while (!list_empty(head)) {
431 prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE);
432 spin_unlock_irq(&mdev->req_lock);
434 finish_wait(&mdev->ee_wait, &wait);
435 spin_lock_irq(&mdev->req_lock);
439 void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head)
441 spin_lock_irq(&mdev->req_lock);
442 _drbd_wait_ee_list_empty(mdev, head);
443 spin_unlock_irq(&mdev->req_lock);
446 /* see also kernel_accept; which is only present since 2.6.18.
447 * also we want to log which part of it failed, exactly */
448 static int drbd_accept(struct drbd_conf *mdev, const char **what,
449 struct socket *sock, struct socket **newsock)
451 struct sock *sk = sock->sk;
455 err = sock->ops->listen(sock, 5);
459 *what = "sock_create_lite";
460 err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol,
466 err = sock->ops->accept(sock, *newsock, 0);
468 sock_release(*newsock);
472 (*newsock)->ops = sock->ops;
478 static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock,
479 void *buf, size_t size, int flags)
486 struct msghdr msg = {
488 .msg_iov = (struct iovec *)&iov,
489 .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL)
495 rv = sock_recvmsg(sock, &msg, size, msg.msg_flags);
501 static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size)
508 struct msghdr msg = {
510 .msg_iov = (struct iovec *)&iov,
511 .msg_flags = MSG_WAITALL | MSG_NOSIGNAL
519 rv = sock_recvmsg(mdev->data.socket, &msg, size, msg.msg_flags);
524 * ECONNRESET other side closed the connection
525 * ERESTARTSYS (on sock) we got a signal
529 if (rv == -ECONNRESET)
530 dev_info(DEV, "sock was reset by peer\n");
531 else if (rv != -ERESTARTSYS)
532 dev_err(DEV, "sock_recvmsg returned %d\n", rv);
534 } else if (rv == 0) {
535 dev_info(DEV, "sock was shut down by peer\n");
538 /* signal came in, or peer/link went down,
539 * after we read a partial message
541 /* D_ASSERT(signal_pending(current)); */
549 drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE));
555 * On individual connections, the socket buffer size must be set prior to the
556 * listen(2) or connect(2) calls in order to have it take effect.
557 * This is our wrapper to do so.
559 static void drbd_setbufsize(struct socket *sock, unsigned int snd,
562 /* open coded SO_SNDBUF, SO_RCVBUF */
564 sock->sk->sk_sndbuf = snd;
565 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
568 sock->sk->sk_rcvbuf = rcv;
569 sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
573 static struct socket *drbd_try_connect(struct drbd_conf *mdev)
577 struct sockaddr_in6 src_in6;
579 int disconnect_on_error = 1;
581 if (!get_net_conf(mdev))
584 what = "sock_create_kern";
585 err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family,
586 SOCK_STREAM, IPPROTO_TCP, &sock);
592 sock->sk->sk_rcvtimeo =
593 sock->sk->sk_sndtimeo = mdev->net_conf->try_connect_int*HZ;
594 drbd_setbufsize(sock, mdev->net_conf->sndbuf_size,
595 mdev->net_conf->rcvbuf_size);
597 /* explicitly bind to the configured IP as source IP
598 * for the outgoing connections.
599 * This is needed for multihomed hosts and to be
600 * able to use lo: interfaces for drbd.
601 * Make sure to use 0 as port number, so linux selects
602 * a free one dynamically.
604 memcpy(&src_in6, mdev->net_conf->my_addr,
605 min_t(int, mdev->net_conf->my_addr_len, sizeof(src_in6)));
606 if (((struct sockaddr *)mdev->net_conf->my_addr)->sa_family == AF_INET6)
607 src_in6.sin6_port = 0;
609 ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */
611 what = "bind before connect";
612 err = sock->ops->bind(sock,
613 (struct sockaddr *) &src_in6,
614 mdev->net_conf->my_addr_len);
618 /* connect may fail, peer not yet available.
619 * stay C_WF_CONNECTION, don't go Disconnecting! */
620 disconnect_on_error = 0;
622 err = sock->ops->connect(sock,
623 (struct sockaddr *)mdev->net_conf->peer_addr,
624 mdev->net_conf->peer_addr_len, 0);
633 /* timeout, busy, signal pending */
634 case ETIMEDOUT: case EAGAIN: case EINPROGRESS:
635 case EINTR: case ERESTARTSYS:
636 /* peer not (yet) available, network problem */
637 case ECONNREFUSED: case ENETUNREACH:
638 case EHOSTDOWN: case EHOSTUNREACH:
639 disconnect_on_error = 0;
642 dev_err(DEV, "%s failed, err = %d\n", what, err);
644 if (disconnect_on_error)
645 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
651 static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev)
654 struct socket *s_estab = NULL, *s_listen;
657 if (!get_net_conf(mdev))
660 what = "sock_create_kern";
661 err = sock_create_kern(((struct sockaddr *)mdev->net_conf->my_addr)->sa_family,
662 SOCK_STREAM, IPPROTO_TCP, &s_listen);
668 timeo = mdev->net_conf->try_connect_int * HZ;
669 timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */
671 s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */
672 s_listen->sk->sk_rcvtimeo = timeo;
673 s_listen->sk->sk_sndtimeo = timeo;
674 drbd_setbufsize(s_listen, mdev->net_conf->sndbuf_size,
675 mdev->net_conf->rcvbuf_size);
677 what = "bind before listen";
678 err = s_listen->ops->bind(s_listen,
679 (struct sockaddr *) mdev->net_conf->my_addr,
680 mdev->net_conf->my_addr_len);
684 err = drbd_accept(mdev, &what, s_listen, &s_estab);
688 sock_release(s_listen);
690 if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) {
691 dev_err(DEV, "%s failed, err = %d\n", what, err);
692 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
700 static int drbd_send_fp(struct drbd_conf *mdev,
701 struct socket *sock, enum drbd_packets cmd)
703 struct p_header80 *h = &mdev->data.sbuf.header.h80;
705 return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0);
708 static enum drbd_packets drbd_recv_fp(struct drbd_conf *mdev, struct socket *sock)
710 struct p_header80 *h = &mdev->data.rbuf.header.h80;
713 rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0);
715 if (rr == sizeof(*h) && h->magic == cpu_to_be32(DRBD_MAGIC))
716 return be16_to_cpu(h->command);
722 * drbd_socket_okay() - Free the socket if its connection is not okay
723 * @mdev: DRBD device.
724 * @sock: pointer to the pointer to the socket.
726 static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock)
734 rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK);
736 if (rr > 0 || rr == -EAGAIN) {
747 * 1 yes, we have a valid connection
748 * 0 oops, did not work out, please try again
749 * -1 peer talks different language,
750 * no point in trying again, please go standalone.
751 * -2 We do not have a network config...
753 static int drbd_connect(struct drbd_conf *mdev)
755 struct socket *s, *sock, *msock;
758 D_ASSERT(!mdev->data.socket);
760 if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS)
763 clear_bit(DISCARD_CONCURRENT, &mdev->flags);
770 /* 3 tries, this should take less than a second! */
771 s = drbd_try_connect(mdev);
774 /* give the other side time to call bind() & listen() */
775 schedule_timeout_interruptible(HZ / 10);
780 drbd_send_fp(mdev, s, P_HAND_SHAKE_S);
784 drbd_send_fp(mdev, s, P_HAND_SHAKE_M);
788 dev_err(DEV, "Logic error in drbd_connect()\n");
789 goto out_release_sockets;
794 schedule_timeout_interruptible(mdev->net_conf->ping_timeo*HZ/10);
795 ok = drbd_socket_okay(mdev, &sock);
796 ok = drbd_socket_okay(mdev, &msock) && ok;
802 s = drbd_wait_for_connect(mdev);
804 try = drbd_recv_fp(mdev, s);
805 drbd_socket_okay(mdev, &sock);
806 drbd_socket_okay(mdev, &msock);
810 dev_warn(DEV, "initial packet S crossed\n");
817 dev_warn(DEV, "initial packet M crossed\n");
821 set_bit(DISCARD_CONCURRENT, &mdev->flags);
824 dev_warn(DEV, "Error receiving initial packet\n");
831 if (mdev->state.conn <= C_DISCONNECTING)
832 goto out_release_sockets;
833 if (signal_pending(current)) {
834 flush_signals(current);
836 if (get_t_state(&mdev->receiver) == Exiting)
837 goto out_release_sockets;
841 ok = drbd_socket_okay(mdev, &sock);
842 ok = drbd_socket_okay(mdev, &msock) && ok;
848 msock->sk->sk_reuse = 1; /* SO_REUSEADDR */
849 sock->sk->sk_reuse = 1; /* SO_REUSEADDR */
851 sock->sk->sk_allocation = GFP_NOIO;
852 msock->sk->sk_allocation = GFP_NOIO;
854 sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK;
855 msock->sk->sk_priority = TC_PRIO_INTERACTIVE;
858 * sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
859 * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
860 * first set it to the P_HAND_SHAKE timeout,
861 * which we set to 4x the configured ping_timeout. */
862 sock->sk->sk_sndtimeo =
863 sock->sk->sk_rcvtimeo = mdev->net_conf->ping_timeo*4*HZ/10;
865 msock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
866 msock->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
868 /* we don't want delays.
869 * we use TCP_CORK where appropriate, though */
870 drbd_tcp_nodelay(sock);
871 drbd_tcp_nodelay(msock);
873 mdev->data.socket = sock;
874 mdev->meta.socket = msock;
875 mdev->last_received = jiffies;
877 D_ASSERT(mdev->asender.task == NULL);
879 h = drbd_do_handshake(mdev);
883 if (mdev->cram_hmac_tfm) {
884 /* drbd_request_state(mdev, NS(conn, WFAuth)); */
885 switch (drbd_do_auth(mdev)) {
887 dev_err(DEV, "Authentication of peer failed\n");
890 dev_err(DEV, "Authentication of peer failed, trying again.\n");
895 if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS)
898 sock->sk->sk_sndtimeo = mdev->net_conf->timeout*HZ/10;
899 sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
901 atomic_set(&mdev->packet_seq, 0);
904 drbd_thread_start(&mdev->asender);
906 if (drbd_send_protocol(mdev) == -1)
908 drbd_send_sync_param(mdev, &mdev->sync_conf);
909 drbd_send_sizes(mdev, 0, 0);
910 drbd_send_uuids(mdev);
911 drbd_send_state(mdev);
912 clear_bit(USE_DEGR_WFC_T, &mdev->flags);
913 clear_bit(RESIZE_PENDING, &mdev->flags);
914 mod_timer(&mdev->request_timer, jiffies + HZ); /* just start it here. */
926 static int drbd_recv_header(struct drbd_conf *mdev, enum drbd_packets *cmd, unsigned int *packet_size)
928 union p_header *h = &mdev->data.rbuf.header;
931 r = drbd_recv(mdev, h, sizeof(*h));
932 if (unlikely(r != sizeof(*h))) {
933 if (!signal_pending(current))
934 dev_warn(DEV, "short read expecting header on sock: r=%d\n", r);
938 if (likely(h->h80.magic == cpu_to_be32(DRBD_MAGIC))) {
939 *cmd = be16_to_cpu(h->h80.command);
940 *packet_size = be16_to_cpu(h->h80.length);
941 } else if (h->h95.magic == cpu_to_be16(DRBD_MAGIC_BIG)) {
942 *cmd = be16_to_cpu(h->h95.command);
943 *packet_size = be32_to_cpu(h->h95.length);
945 dev_err(DEV, "magic?? on data m: 0x%08x c: %d l: %d\n",
946 be32_to_cpu(h->h80.magic),
947 be16_to_cpu(h->h80.command),
948 be16_to_cpu(h->h80.length));
951 mdev->last_received = jiffies;
956 static void drbd_flush(struct drbd_conf *mdev)
960 if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) {
961 rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL,
964 dev_err(DEV, "local disk flush failed with status %d\n", rv);
965 /* would rather check on EOPNOTSUPP, but that is not reliable.
966 * don't try again for ANY return value != 0
967 * if (rv == -EOPNOTSUPP) */
968 drbd_bump_write_ordering(mdev, WO_drain_io);
975 * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it.
976 * @mdev: DRBD device.
977 * @epoch: Epoch object.
980 static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev,
981 struct drbd_epoch *epoch,
985 struct drbd_epoch *next_epoch;
986 enum finish_epoch rv = FE_STILL_LIVE;
988 spin_lock(&mdev->epoch_lock);
992 epoch_size = atomic_read(&epoch->epoch_size);
994 switch (ev & ~EV_CLEANUP) {
996 atomic_dec(&epoch->active);
998 case EV_GOT_BARRIER_NR:
999 set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags);
1001 case EV_BECAME_LAST:
1006 if (epoch_size != 0 &&
1007 atomic_read(&epoch->active) == 0 &&
1008 test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) {
1009 if (!(ev & EV_CLEANUP)) {
1010 spin_unlock(&mdev->epoch_lock);
1011 drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size);
1012 spin_lock(&mdev->epoch_lock);
1016 if (mdev->current_epoch != epoch) {
1017 next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list);
1018 list_del(&epoch->list);
1019 ev = EV_BECAME_LAST | (ev & EV_CLEANUP);
1023 if (rv == FE_STILL_LIVE)
1027 atomic_set(&epoch->epoch_size, 0);
1028 /* atomic_set(&epoch->active, 0); is already zero */
1029 if (rv == FE_STILL_LIVE)
1031 wake_up(&mdev->ee_wait);
1041 spin_unlock(&mdev->epoch_lock);
1047 * drbd_bump_write_ordering() - Fall back to an other write ordering method
1048 * @mdev: DRBD device.
1049 * @wo: Write ordering method to try.
1051 void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local)
1053 enum write_ordering_e pwo;
1054 static char *write_ordering_str[] = {
1056 [WO_drain_io] = "drain",
1057 [WO_bdev_flush] = "flush",
1060 pwo = mdev->write_ordering;
1062 if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush)
1064 if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain)
1066 mdev->write_ordering = wo;
1067 if (pwo != mdev->write_ordering || wo == WO_bdev_flush)
1068 dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]);
1073 * @mdev: DRBD device.
1075 * @rw: flag field, see bio->bi_rw
1077 * May spread the pages to multiple bios,
1078 * depending on bio_add_page restrictions.
1080 * Returns 0 if all bios have been submitted,
1081 * -ENOMEM if we could not allocate enough bios,
1082 * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a
1083 * single page to an empty bio (which should never happen and likely indicates
1084 * that the lower level IO stack is in some way broken). This has been observed
1085 * on certain Xen deployments.
1087 /* TODO allocate from our own bio_set. */
1088 int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_epoch_entry *e,
1089 const unsigned rw, const int fault_type)
1091 struct bio *bios = NULL;
1093 struct page *page = e->pages;
1094 sector_t sector = e->sector;
1095 unsigned ds = e->size;
1096 unsigned n_bios = 0;
1097 unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT;
1100 /* In most cases, we will only need one bio. But in case the lower
1101 * level restrictions happen to be different at this offset on this
1102 * side than those of the sending peer, we may need to submit the
1103 * request in more than one bio. */
1105 bio = bio_alloc(GFP_NOIO, nr_pages);
1107 dev_err(DEV, "submit_ee: Allocation of a bio failed\n");
1110 /* > e->sector, unless this is the first bio */
1111 bio->bi_sector = sector;
1112 bio->bi_bdev = mdev->ldev->backing_bdev;
1114 bio->bi_private = e;
1115 bio->bi_end_io = drbd_endio_sec;
1117 bio->bi_next = bios;
1121 page_chain_for_each(page) {
1122 unsigned len = min_t(unsigned, ds, PAGE_SIZE);
1123 if (!bio_add_page(bio, page, len, 0)) {
1124 /* A single page must always be possible!
1125 * But in case it fails anyways,
1126 * we deal with it, and complain (below). */
1127 if (bio->bi_vcnt == 0) {
1129 "bio_add_page failed for len=%u, "
1130 "bi_vcnt=0 (bi_sector=%llu)\n",
1131 len, (unsigned long long)bio->bi_sector);
1141 D_ASSERT(page == NULL);
1144 atomic_set(&e->pending_bios, n_bios);
1147 bios = bios->bi_next;
1148 bio->bi_next = NULL;
1150 drbd_generic_make_request(mdev, fault_type, bio);
1157 bios = bios->bi_next;
1163 static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1166 struct p_barrier *p = &mdev->data.rbuf.barrier;
1167 struct drbd_epoch *epoch;
1171 mdev->current_epoch->barrier_nr = p->barrier;
1172 rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR);
1174 /* P_BARRIER_ACK may imply that the corresponding extent is dropped from
1175 * the activity log, which means it would not be resynced in case the
1176 * R_PRIMARY crashes now.
1177 * Therefore we must send the barrier_ack after the barrier request was
1179 switch (mdev->write_ordering) {
1181 if (rv == FE_RECYCLED)
1184 /* receiver context, in the writeout path of the other node.
1185 * avoid potential distributed deadlock */
1186 epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO);
1190 dev_warn(DEV, "Allocation of an epoch failed, slowing down\n");
1195 drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
1198 if (atomic_read(&mdev->current_epoch->epoch_size)) {
1199 epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO);
1204 epoch = mdev->current_epoch;
1205 wait_event(mdev->ee_wait, atomic_read(&epoch->epoch_size) == 0);
1207 D_ASSERT(atomic_read(&epoch->active) == 0);
1208 D_ASSERT(epoch->flags == 0);
1212 dev_err(DEV, "Strangeness in mdev->write_ordering %d\n", mdev->write_ordering);
1217 atomic_set(&epoch->epoch_size, 0);
1218 atomic_set(&epoch->active, 0);
1220 spin_lock(&mdev->epoch_lock);
1221 if (atomic_read(&mdev->current_epoch->epoch_size)) {
1222 list_add(&epoch->list, &mdev->current_epoch->list);
1223 mdev->current_epoch = epoch;
1226 /* The current_epoch got recycled while we allocated this one... */
1229 spin_unlock(&mdev->epoch_lock);
1234 /* used from receive_RSDataReply (recv_resync_read)
1235 * and from receive_Data */
1236 static struct drbd_epoch_entry *
1237 read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, int data_size) __must_hold(local)
1239 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
1240 struct drbd_epoch_entry *e;
1243 void *dig_in = mdev->int_dig_in;
1244 void *dig_vv = mdev->int_dig_vv;
1245 unsigned long *data;
1247 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1248 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1251 rr = drbd_recv(mdev, dig_in, dgs);
1253 if (!signal_pending(current))
1255 "short read receiving data digest: read %d expected %d\n",
1263 ERR_IF(data_size == 0) return NULL;
1264 ERR_IF(data_size & 0x1ff) return NULL;
1265 ERR_IF(data_size > DRBD_MAX_BIO_SIZE) return NULL;
1267 /* even though we trust out peer,
1268 * we sometimes have to double check. */
1269 if (sector + (data_size>>9) > capacity) {
1270 dev_err(DEV, "request from peer beyond end of local disk: "
1271 "capacity: %llus < sector: %llus + size: %u\n",
1272 (unsigned long long)capacity,
1273 (unsigned long long)sector, data_size);
1277 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
1278 * "criss-cross" setup, that might cause write-out on some other DRBD,
1279 * which in turn might block on the other node at this very place. */
1280 e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO);
1286 page_chain_for_each(page) {
1287 unsigned len = min_t(int, ds, PAGE_SIZE);
1289 rr = drbd_recv(mdev, data, len);
1290 if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) {
1291 dev_err(DEV, "Fault injection: Corrupting data on receive\n");
1292 data[0] = data[0] ^ (unsigned long)-1;
1296 drbd_free_ee(mdev, e);
1297 if (!signal_pending(current))
1298 dev_warn(DEV, "short read receiving data: read %d expected %d\n",
1306 drbd_csum_ee(mdev, mdev->integrity_r_tfm, e, dig_vv);
1307 if (memcmp(dig_in, dig_vv, dgs)) {
1308 dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n",
1309 (unsigned long long)sector, data_size);
1310 drbd_bcast_ee(mdev, "digest failed",
1311 dgs, dig_in, dig_vv, e);
1312 drbd_free_ee(mdev, e);
1316 mdev->recv_cnt += data_size>>9;
1320 /* drbd_drain_block() just takes a data block
1321 * out of the socket input buffer, and discards it.
1323 static int drbd_drain_block(struct drbd_conf *mdev, int data_size)
1332 page = drbd_pp_alloc(mdev, 1, 1);
1336 rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE));
1337 if (rr != min_t(int, data_size, PAGE_SIZE)) {
1339 if (!signal_pending(current))
1341 "short read receiving data: read %d expected %d\n",
1342 rr, min_t(int, data_size, PAGE_SIZE));
1348 drbd_pp_free(mdev, page, 0);
1352 static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req,
1353 sector_t sector, int data_size)
1355 struct bio_vec *bvec;
1357 int dgs, rr, i, expect;
1358 void *dig_in = mdev->int_dig_in;
1359 void *dig_vv = mdev->int_dig_vv;
1361 dgs = (mdev->agreed_pro_version >= 87 && mdev->integrity_r_tfm) ?
1362 crypto_hash_digestsize(mdev->integrity_r_tfm) : 0;
1365 rr = drbd_recv(mdev, dig_in, dgs);
1367 if (!signal_pending(current))
1369 "short read receiving data reply digest: read %d expected %d\n",
1377 /* optimistically update recv_cnt. if receiving fails below,
1378 * we disconnect anyways, and counters will be reset. */
1379 mdev->recv_cnt += data_size>>9;
1381 bio = req->master_bio;
1382 D_ASSERT(sector == bio->bi_sector);
1384 bio_for_each_segment(bvec, bio, i) {
1385 expect = min_t(int, data_size, bvec->bv_len);
1386 rr = drbd_recv(mdev,
1387 kmap(bvec->bv_page)+bvec->bv_offset,
1389 kunmap(bvec->bv_page);
1391 if (!signal_pending(current))
1392 dev_warn(DEV, "short read receiving data reply: "
1393 "read %d expected %d\n",
1401 drbd_csum_bio(mdev, mdev->integrity_r_tfm, bio, dig_vv);
1402 if (memcmp(dig_in, dig_vv, dgs)) {
1403 dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n");
1408 D_ASSERT(data_size == 0);
1412 /* e_end_resync_block() is called via
1413 * drbd_process_done_ee() by asender only */
1414 static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1416 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1417 sector_t sector = e->sector;
1420 D_ASSERT(hlist_unhashed(&e->collision));
1422 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1423 drbd_set_in_sync(mdev, sector, e->size);
1424 ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e);
1426 /* Record failure to sync */
1427 drbd_rs_failed_io(mdev, sector, e->size);
1429 ok = drbd_send_ack(mdev, P_NEG_ACK, e);
1436 static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local)
1438 struct drbd_epoch_entry *e;
1440 e = read_in_block(mdev, ID_SYNCER, sector, data_size);
1444 dec_rs_pending(mdev);
1447 /* corresponding dec_unacked() in e_end_resync_block()
1448 * respective _drbd_clear_done_ee */
1450 e->w.cb = e_end_resync_block;
1452 spin_lock_irq(&mdev->req_lock);
1453 list_add(&e->w.list, &mdev->sync_ee);
1454 spin_unlock_irq(&mdev->req_lock);
1456 atomic_add(data_size >> 9, &mdev->rs_sect_ev);
1457 if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0)
1460 /* don't care for the reason here */
1461 dev_err(DEV, "submit failed, triggering re-connect\n");
1462 spin_lock_irq(&mdev->req_lock);
1463 list_del(&e->w.list);
1464 spin_unlock_irq(&mdev->req_lock);
1466 drbd_free_ee(mdev, e);
1472 static struct drbd_request *
1473 find_request(struct drbd_conf *mdev,
1474 struct hlist_head *(*hash_slot)(struct drbd_conf *, sector_t),
1475 u64 id, sector_t sector, const char *func)
1477 struct hlist_head *slot = hash_slot(mdev, sector);
1478 struct hlist_node *n;
1479 struct drbd_request *req;
1481 hlist_for_each_entry(req, n, slot, collision) {
1482 if ((unsigned long)req != (unsigned long)id)
1484 if (req->sector != sector) {
1485 dev_err(DEV, "%s: found request %lu but it has "
1486 "wrong sector (%llus versus %llus)\n",
1487 func, (unsigned long)req,
1488 (unsigned long long)req->sector,
1489 (unsigned long long)sector);
1497 static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1499 struct drbd_request *req;
1502 struct p_data *p = &mdev->data.rbuf.data;
1504 sector = be64_to_cpu(p->sector);
1506 spin_lock_irq(&mdev->req_lock);
1507 req = find_request(mdev, ar_hash_slot, p->block_id, sector, __func__);
1508 spin_unlock_irq(&mdev->req_lock);
1509 if (unlikely(!req)) {
1510 dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n");
1514 /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid
1515 * special casing it there for the various failure cases.
1516 * still no race with drbd_fail_pending_reads */
1517 ok = recv_dless_read(mdev, req, sector, data_size);
1520 req_mod(req, data_received);
1521 /* else: nothing. handled from drbd_disconnect...
1522 * I don't think we may complete this just yet
1523 * in case we are "on-disconnect: freeze" */
1528 static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1532 struct p_data *p = &mdev->data.rbuf.data;
1534 sector = be64_to_cpu(p->sector);
1535 D_ASSERT(p->block_id == ID_SYNCER);
1537 if (get_ldev(mdev)) {
1538 /* data is submitted to disk within recv_resync_read.
1539 * corresponding put_ldev done below on error,
1540 * or in drbd_endio_sec. */
1541 ok = recv_resync_read(mdev, sector, data_size);
1543 if (__ratelimit(&drbd_ratelimit_state))
1544 dev_err(DEV, "Can not write resync data to local disk.\n");
1546 ok = drbd_drain_block(mdev, data_size);
1548 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size);
1551 atomic_add(data_size >> 9, &mdev->rs_sect_in);
1556 /* e_end_block() is called via drbd_process_done_ee().
1557 * this means this function only runs in the asender thread
1559 static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1561 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1562 sector_t sector = e->sector;
1565 if (mdev->net_conf->wire_protocol == DRBD_PROT_C) {
1566 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1567 pcmd = (mdev->state.conn >= C_SYNC_SOURCE &&
1568 mdev->state.conn <= C_PAUSED_SYNC_T &&
1569 e->flags & EE_MAY_SET_IN_SYNC) ?
1570 P_RS_WRITE_ACK : P_WRITE_ACK;
1571 ok &= drbd_send_ack(mdev, pcmd, e);
1572 if (pcmd == P_RS_WRITE_ACK)
1573 drbd_set_in_sync(mdev, sector, e->size);
1575 ok = drbd_send_ack(mdev, P_NEG_ACK, e);
1576 /* we expect it to be marked out of sync anyways...
1577 * maybe assert this? */
1581 /* we delete from the conflict detection hash _after_ we sent out the
1582 * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */
1583 if (mdev->net_conf->two_primaries) {
1584 spin_lock_irq(&mdev->req_lock);
1585 D_ASSERT(!hlist_unhashed(&e->collision));
1586 hlist_del_init(&e->collision);
1587 spin_unlock_irq(&mdev->req_lock);
1589 D_ASSERT(hlist_unhashed(&e->collision));
1592 drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0));
1597 static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1599 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1602 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1603 ok = drbd_send_ack(mdev, P_DISCARD_ACK, e);
1605 spin_lock_irq(&mdev->req_lock);
1606 D_ASSERT(!hlist_unhashed(&e->collision));
1607 hlist_del_init(&e->collision);
1608 spin_unlock_irq(&mdev->req_lock);
1615 /* Called from receive_Data.
1616 * Synchronize packets on sock with packets on msock.
1618 * This is here so even when a P_DATA packet traveling via sock overtook an Ack
1619 * packet traveling on msock, they are still processed in the order they have
1622 * Note: we don't care for Ack packets overtaking P_DATA packets.
1624 * In case packet_seq is larger than mdev->peer_seq number, there are
1625 * outstanding packets on the msock. We wait for them to arrive.
1626 * In case we are the logically next packet, we update mdev->peer_seq
1627 * ourselves. Correctly handles 32bit wrap around.
1629 * Assume we have a 10 GBit connection, that is about 1<<30 byte per second,
1630 * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds
1631 * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have
1632 * 1<<9 == 512 seconds aka ages for the 32bit wrap around...
1634 * returns 0 if we may process the packet,
1635 * -ERESTARTSYS if we were interrupted (by disconnect signal). */
1636 static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq)
1642 spin_lock(&mdev->peer_seq_lock);
1644 prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE);
1645 if (seq_le(packet_seq, mdev->peer_seq+1))
1647 if (signal_pending(current)) {
1651 p_seq = mdev->peer_seq;
1652 spin_unlock(&mdev->peer_seq_lock);
1653 timeout = schedule_timeout(30*HZ);
1654 spin_lock(&mdev->peer_seq_lock);
1655 if (timeout == 0 && p_seq == mdev->peer_seq) {
1657 dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n");
1661 finish_wait(&mdev->seq_wait, &wait);
1662 if (mdev->peer_seq+1 == packet_seq)
1664 spin_unlock(&mdev->peer_seq_lock);
1668 /* see also bio_flags_to_wire()
1669 * DRBD_REQ_*, because we need to semantically map the flags to data packet
1670 * flags and back. We may replicate to other kernel versions. */
1671 static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf)
1673 return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) |
1674 (dpf & DP_FUA ? REQ_FUA : 0) |
1675 (dpf & DP_FLUSH ? REQ_FLUSH : 0) |
1676 (dpf & DP_DISCARD ? REQ_DISCARD : 0);
1679 /* mirrored write */
1680 static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1683 struct drbd_epoch_entry *e;
1684 struct p_data *p = &mdev->data.rbuf.data;
1688 if (!get_ldev(mdev)) {
1689 spin_lock(&mdev->peer_seq_lock);
1690 if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num))
1692 spin_unlock(&mdev->peer_seq_lock);
1694 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size);
1695 atomic_inc(&mdev->current_epoch->epoch_size);
1696 return drbd_drain_block(mdev, data_size);
1699 /* get_ldev(mdev) successful.
1700 * Corresponding put_ldev done either below (on various errors),
1701 * or in drbd_endio_sec, if we successfully submit the data at
1702 * the end of this function. */
1704 sector = be64_to_cpu(p->sector);
1705 e = read_in_block(mdev, p->block_id, sector, data_size);
1711 e->w.cb = e_end_block;
1713 dp_flags = be32_to_cpu(p->dp_flags);
1714 rw |= wire_flags_to_bio(mdev, dp_flags);
1716 if (dp_flags & DP_MAY_SET_IN_SYNC)
1717 e->flags |= EE_MAY_SET_IN_SYNC;
1719 spin_lock(&mdev->epoch_lock);
1720 e->epoch = mdev->current_epoch;
1721 atomic_inc(&e->epoch->epoch_size);
1722 atomic_inc(&e->epoch->active);
1723 spin_unlock(&mdev->epoch_lock);
1725 /* I'm the receiver, I do hold a net_cnt reference. */
1726 if (!mdev->net_conf->two_primaries) {
1727 spin_lock_irq(&mdev->req_lock);
1729 /* don't get the req_lock yet,
1730 * we may sleep in drbd_wait_peer_seq */
1731 const int size = e->size;
1732 const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1734 struct drbd_request *i;
1735 struct hlist_node *n;
1736 struct hlist_head *slot;
1739 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1740 BUG_ON(mdev->ee_hash == NULL);
1741 BUG_ON(mdev->tl_hash == NULL);
1743 /* conflict detection and handling:
1744 * 1. wait on the sequence number,
1745 * in case this data packet overtook ACK packets.
1746 * 2. check our hash tables for conflicting requests.
1747 * we only need to walk the tl_hash, since an ee can not
1748 * have a conflict with an other ee: on the submitting
1749 * node, the corresponding req had already been conflicting,
1750 * and a conflicting req is never sent.
1752 * Note: for two_primaries, we are protocol C,
1753 * so there cannot be any request that is DONE
1754 * but still on the transfer log.
1756 * unconditionally add to the ee_hash.
1758 * if no conflicting request is found:
1761 * if any conflicting request is found
1762 * that has not yet been acked,
1763 * AND I have the "discard concurrent writes" flag:
1764 * queue (via done_ee) the P_DISCARD_ACK; OUT.
1766 * if any conflicting request is found:
1767 * block the receiver, waiting on misc_wait
1768 * until no more conflicting requests are there,
1769 * or we get interrupted (disconnect).
1771 * we do not just write after local io completion of those
1772 * requests, but only after req is done completely, i.e.
1773 * we wait for the P_DISCARD_ACK to arrive!
1775 * then proceed normally, i.e. submit.
1777 if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num)))
1778 goto out_interrupted;
1780 spin_lock_irq(&mdev->req_lock);
1782 hlist_add_head(&e->collision, ee_hash_slot(mdev, sector));
1784 #define OVERLAPS overlaps(i->sector, i->size, sector, size)
1785 slot = tl_hash_slot(mdev, sector);
1788 int have_unacked = 0;
1789 int have_conflict = 0;
1790 prepare_to_wait(&mdev->misc_wait, &wait,
1791 TASK_INTERRUPTIBLE);
1792 hlist_for_each_entry(i, n, slot, collision) {
1794 /* only ALERT on first iteration,
1795 * we may be woken up early... */
1797 dev_alert(DEV, "%s[%u] Concurrent local write detected!"
1798 " new: %llus +%u; pending: %llus +%u\n",
1799 current->comm, current->pid,
1800 (unsigned long long)sector, size,
1801 (unsigned long long)i->sector, i->size);
1802 if (i->rq_state & RQ_NET_PENDING)
1811 /* Discard Ack only for the _first_ iteration */
1812 if (first && discard && have_unacked) {
1813 dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n",
1814 (unsigned long long)sector);
1816 e->w.cb = e_send_discard_ack;
1817 list_add_tail(&e->w.list, &mdev->done_ee);
1819 spin_unlock_irq(&mdev->req_lock);
1821 /* we could probably send that P_DISCARD_ACK ourselves,
1822 * but I don't like the receiver using the msock */
1826 finish_wait(&mdev->misc_wait, &wait);
1830 if (signal_pending(current)) {
1831 hlist_del_init(&e->collision);
1833 spin_unlock_irq(&mdev->req_lock);
1835 finish_wait(&mdev->misc_wait, &wait);
1836 goto out_interrupted;
1839 spin_unlock_irq(&mdev->req_lock);
1842 dev_alert(DEV, "Concurrent write! [W AFTERWARDS] "
1843 "sec=%llus\n", (unsigned long long)sector);
1844 } else if (discard) {
1845 /* we had none on the first iteration.
1846 * there must be none now. */
1847 D_ASSERT(have_unacked == 0);
1850 spin_lock_irq(&mdev->req_lock);
1852 finish_wait(&mdev->misc_wait, &wait);
1855 list_add(&e->w.list, &mdev->active_ee);
1856 spin_unlock_irq(&mdev->req_lock);
1858 switch (mdev->net_conf->wire_protocol) {
1861 /* corresponding dec_unacked() in e_end_block()
1862 * respective _drbd_clear_done_ee */
1865 /* I really don't like it that the receiver thread
1866 * sends on the msock, but anyways */
1867 drbd_send_ack(mdev, P_RECV_ACK, e);
1874 if (mdev->state.pdsk < D_INCONSISTENT) {
1875 /* In case we have the only disk of the cluster, */
1876 drbd_set_out_of_sync(mdev, e->sector, e->size);
1877 e->flags |= EE_CALL_AL_COMPLETE_IO;
1878 e->flags &= ~EE_MAY_SET_IN_SYNC;
1879 drbd_al_begin_io(mdev, e->sector);
1882 if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0)
1885 /* don't care for the reason here */
1886 dev_err(DEV, "submit failed, triggering re-connect\n");
1887 spin_lock_irq(&mdev->req_lock);
1888 list_del(&e->w.list);
1889 hlist_del_init(&e->collision);
1890 spin_unlock_irq(&mdev->req_lock);
1891 if (e->flags & EE_CALL_AL_COMPLETE_IO)
1892 drbd_al_complete_io(mdev, e->sector);
1895 drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + EV_CLEANUP);
1897 drbd_free_ee(mdev, e);
1901 /* We may throttle resync, if the lower device seems to be busy,
1902 * and current sync rate is above c_min_rate.
1904 * To decide whether or not the lower device is busy, we use a scheme similar
1905 * to MD RAID is_mddev_idle(): if the partition stats reveal "significant"
1906 * (more than 64 sectors) of activity we cannot account for with our own resync
1907 * activity, it obviously is "busy".
1909 * The current sync rate used here uses only the most recent two step marks,
1910 * to have a short time average so we can react faster.
1912 int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector)
1914 struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk;
1915 unsigned long db, dt, dbdt;
1916 struct lc_element *tmp;
1920 /* feature disabled? */
1921 if (mdev->sync_conf.c_min_rate == 0)
1924 spin_lock_irq(&mdev->al_lock);
1925 tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector));
1927 struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce);
1928 if (test_bit(BME_PRIORITY, &bm_ext->flags)) {
1929 spin_unlock_irq(&mdev->al_lock);
1932 /* Do not slow down if app IO is already waiting for this extent */
1934 spin_unlock_irq(&mdev->al_lock);
1936 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
1937 (int)part_stat_read(&disk->part0, sectors[1]) -
1938 atomic_read(&mdev->rs_sect_ev);
1940 if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) {
1941 unsigned long rs_left;
1944 mdev->rs_last_events = curr_events;
1946 /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP,
1948 i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS;
1950 if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T)
1951 rs_left = mdev->ov_left;
1953 rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed;
1955 dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ;
1958 db = mdev->rs_mark_left[i] - rs_left;
1959 dbdt = Bit2KB(db/dt);
1961 if (dbdt > mdev->sync_conf.c_min_rate)
1968 static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size)
1971 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
1972 struct drbd_epoch_entry *e;
1973 struct digest_info *di = NULL;
1975 unsigned int fault_type;
1976 struct p_block_req *p = &mdev->data.rbuf.block_req;
1978 sector = be64_to_cpu(p->sector);
1979 size = be32_to_cpu(p->blksize);
1981 if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) {
1982 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
1983 (unsigned long long)sector, size);
1986 if (sector + (size>>9) > capacity) {
1987 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
1988 (unsigned long long)sector, size);
1992 if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) {
1995 case P_DATA_REQUEST:
1996 drbd_send_ack_rp(mdev, P_NEG_DREPLY, p);
1998 case P_RS_DATA_REQUEST:
1999 case P_CSUM_RS_REQUEST:
2001 drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p);
2005 dec_rs_pending(mdev);
2006 drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC);
2009 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n",
2012 if (verb && __ratelimit(&drbd_ratelimit_state))
2013 dev_err(DEV, "Can not satisfy peer's read request, "
2014 "no local data.\n");
2016 /* drain possibly payload */
2017 return drbd_drain_block(mdev, digest_size);
2020 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
2021 * "criss-cross" setup, that might cause write-out on some other DRBD,
2022 * which in turn might block on the other node at this very place. */
2023 e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO);
2030 case P_DATA_REQUEST:
2031 e->w.cb = w_e_end_data_req;
2032 fault_type = DRBD_FAULT_DT_RD;
2033 /* application IO, don't drbd_rs_begin_io */
2036 case P_RS_DATA_REQUEST:
2037 e->w.cb = w_e_end_rsdata_req;
2038 fault_type = DRBD_FAULT_RS_RD;
2039 /* used in the sector offset progress display */
2040 mdev->bm_resync_fo = BM_SECT_TO_BIT(sector);
2044 case P_CSUM_RS_REQUEST:
2045 fault_type = DRBD_FAULT_RS_RD;
2046 di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO);
2050 di->digest_size = digest_size;
2051 di->digest = (((char *)di)+sizeof(struct digest_info));
2054 e->flags |= EE_HAS_DIGEST;
2056 if (drbd_recv(mdev, di->digest, digest_size) != digest_size)
2059 if (cmd == P_CSUM_RS_REQUEST) {
2060 D_ASSERT(mdev->agreed_pro_version >= 89);
2061 e->w.cb = w_e_end_csum_rs_req;
2062 /* used in the sector offset progress display */
2063 mdev->bm_resync_fo = BM_SECT_TO_BIT(sector);
2064 } else if (cmd == P_OV_REPLY) {
2065 /* track progress, we may need to throttle */
2066 atomic_add(size >> 9, &mdev->rs_sect_in);
2067 e->w.cb = w_e_end_ov_reply;
2068 dec_rs_pending(mdev);
2069 /* drbd_rs_begin_io done when we sent this request,
2070 * but accounting still needs to be done. */
2071 goto submit_for_resync;
2076 if (mdev->ov_start_sector == ~(sector_t)0 &&
2077 mdev->agreed_pro_version >= 90) {
2078 unsigned long now = jiffies;
2080 mdev->ov_start_sector = sector;
2081 mdev->ov_position = sector;
2082 mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector);
2083 mdev->rs_total = mdev->ov_left;
2084 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2085 mdev->rs_mark_left[i] = mdev->ov_left;
2086 mdev->rs_mark_time[i] = now;
2088 dev_info(DEV, "Online Verify start sector: %llu\n",
2089 (unsigned long long)sector);
2091 e->w.cb = w_e_end_ov_req;
2092 fault_type = DRBD_FAULT_RS_RD;
2096 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n",
2098 fault_type = DRBD_FAULT_MAX;
2102 /* Throttle, drbd_rs_begin_io and submit should become asynchronous
2103 * wrt the receiver, but it is not as straightforward as it may seem.
2104 * Various places in the resync start and stop logic assume resync
2105 * requests are processed in order, requeuing this on the worker thread
2106 * introduces a bunch of new code for synchronization between threads.
2108 * Unlimited throttling before drbd_rs_begin_io may stall the resync
2109 * "forever", throttling after drbd_rs_begin_io will lock that extent
2110 * for application writes for the same time. For now, just throttle
2111 * here, where the rest of the code expects the receiver to sleep for
2115 /* Throttle before drbd_rs_begin_io, as that locks out application IO;
2116 * this defers syncer requests for some time, before letting at least
2117 * on request through. The resync controller on the receiving side
2118 * will adapt to the incoming rate accordingly.
2120 * We cannot throttle here if remote is Primary/SyncTarget:
2121 * we would also throttle its application reads.
2122 * In that case, throttling is done on the SyncTarget only.
2124 if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector))
2125 schedule_timeout_uninterruptible(HZ/10);
2126 if (drbd_rs_begin_io(mdev, sector))
2130 atomic_add(size >> 9, &mdev->rs_sect_ev);
2134 spin_lock_irq(&mdev->req_lock);
2135 list_add_tail(&e->w.list, &mdev->read_ee);
2136 spin_unlock_irq(&mdev->req_lock);
2138 if (drbd_submit_ee(mdev, e, READ, fault_type) == 0)
2141 /* don't care for the reason here */
2142 dev_err(DEV, "submit failed, triggering re-connect\n");
2143 spin_lock_irq(&mdev->req_lock);
2144 list_del(&e->w.list);
2145 spin_unlock_irq(&mdev->req_lock);
2146 /* no drbd_rs_complete_io(), we are dropping the connection anyways */
2150 drbd_free_ee(mdev, e);
2154 static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local)
2156 int self, peer, rv = -100;
2157 unsigned long ch_self, ch_peer;
2159 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2160 peer = mdev->p_uuid[UI_BITMAP] & 1;
2162 ch_peer = mdev->p_uuid[UI_SIZE];
2163 ch_self = mdev->comm_bm_set;
2165 switch (mdev->net_conf->after_sb_0p) {
2167 case ASB_DISCARD_SECONDARY:
2168 case ASB_CALL_HELPER:
2169 dev_err(DEV, "Configuration error.\n");
2171 case ASB_DISCONNECT:
2173 case ASB_DISCARD_YOUNGER_PRI:
2174 if (self == 0 && peer == 1) {
2178 if (self == 1 && peer == 0) {
2182 /* Else fall through to one of the other strategies... */
2183 case ASB_DISCARD_OLDER_PRI:
2184 if (self == 0 && peer == 1) {
2188 if (self == 1 && peer == 0) {
2192 /* Else fall through to one of the other strategies... */
2193 dev_warn(DEV, "Discard younger/older primary did not find a decision\n"
2194 "Using discard-least-changes instead\n");
2195 case ASB_DISCARD_ZERO_CHG:
2196 if (ch_peer == 0 && ch_self == 0) {
2197 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2201 if (ch_peer == 0) { rv = 1; break; }
2202 if (ch_self == 0) { rv = -1; break; }
2204 if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG)
2206 case ASB_DISCARD_LEAST_CHG:
2207 if (ch_self < ch_peer)
2209 else if (ch_self > ch_peer)
2211 else /* ( ch_self == ch_peer ) */
2212 /* Well, then use something else. */
2213 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2216 case ASB_DISCARD_LOCAL:
2219 case ASB_DISCARD_REMOTE:
2226 static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local)
2230 switch (mdev->net_conf->after_sb_1p) {
2231 case ASB_DISCARD_YOUNGER_PRI:
2232 case ASB_DISCARD_OLDER_PRI:
2233 case ASB_DISCARD_LEAST_CHG:
2234 case ASB_DISCARD_LOCAL:
2235 case ASB_DISCARD_REMOTE:
2236 dev_err(DEV, "Configuration error.\n");
2238 case ASB_DISCONNECT:
2241 hg = drbd_asb_recover_0p(mdev);
2242 if (hg == -1 && mdev->state.role == R_SECONDARY)
2244 if (hg == 1 && mdev->state.role == R_PRIMARY)
2248 rv = drbd_asb_recover_0p(mdev);
2250 case ASB_DISCARD_SECONDARY:
2251 return mdev->state.role == R_PRIMARY ? 1 : -1;
2252 case ASB_CALL_HELPER:
2253 hg = drbd_asb_recover_0p(mdev);
2254 if (hg == -1 && mdev->state.role == R_PRIMARY) {
2255 enum drbd_state_rv rv2;
2257 drbd_set_role(mdev, R_SECONDARY, 0);
2258 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2259 * we might be here in C_WF_REPORT_PARAMS which is transient.
2260 * we do not need to wait for the after state change work either. */
2261 rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2262 if (rv2 != SS_SUCCESS) {
2263 drbd_khelper(mdev, "pri-lost-after-sb");
2265 dev_warn(DEV, "Successfully gave up primary role.\n");
2275 static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local)
2279 switch (mdev->net_conf->after_sb_2p) {
2280 case ASB_DISCARD_YOUNGER_PRI:
2281 case ASB_DISCARD_OLDER_PRI:
2282 case ASB_DISCARD_LEAST_CHG:
2283 case ASB_DISCARD_LOCAL:
2284 case ASB_DISCARD_REMOTE:
2286 case ASB_DISCARD_SECONDARY:
2287 dev_err(DEV, "Configuration error.\n");
2290 rv = drbd_asb_recover_0p(mdev);
2292 case ASB_DISCONNECT:
2294 case ASB_CALL_HELPER:
2295 hg = drbd_asb_recover_0p(mdev);
2297 enum drbd_state_rv rv2;
2299 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2300 * we might be here in C_WF_REPORT_PARAMS which is transient.
2301 * we do not need to wait for the after state change work either. */
2302 rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2303 if (rv2 != SS_SUCCESS) {
2304 drbd_khelper(mdev, "pri-lost-after-sb");
2306 dev_warn(DEV, "Successfully gave up primary role.\n");
2316 static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid,
2317 u64 bits, u64 flags)
2320 dev_info(DEV, "%s uuid info vanished while I was looking!\n", text);
2323 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n",
2325 (unsigned long long)uuid[UI_CURRENT],
2326 (unsigned long long)uuid[UI_BITMAP],
2327 (unsigned long long)uuid[UI_HISTORY_START],
2328 (unsigned long long)uuid[UI_HISTORY_END],
2329 (unsigned long long)bits,
2330 (unsigned long long)flags);
2334 100 after split brain try auto recover
2335 2 C_SYNC_SOURCE set BitMap
2336 1 C_SYNC_SOURCE use BitMap
2338 -1 C_SYNC_TARGET use BitMap
2339 -2 C_SYNC_TARGET set BitMap
2340 -100 after split brain, disconnect
2341 -1000 unrelated data
2342 -1091 requires proto 91
2343 -1096 requires proto 96
2345 static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local)
2350 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2351 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2354 if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED)
2358 if ((self == UUID_JUST_CREATED || self == (u64)0) &&
2359 peer != UUID_JUST_CREATED)
2363 if (self != UUID_JUST_CREATED &&
2364 (peer == UUID_JUST_CREATED || peer == (u64)0))
2368 int rct, dc; /* roles at crash time */
2370 if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) {
2372 if (mdev->agreed_pro_version < 91)
2375 if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) &&
2376 (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) {
2377 dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n");
2378 drbd_uuid_set_bm(mdev, 0UL);
2380 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2381 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2384 dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n");
2391 if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) {
2393 if (mdev->agreed_pro_version < 91)
2396 if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) &&
2397 (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) {
2398 dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n");
2400 mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START];
2401 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP];
2402 mdev->p_uuid[UI_BITMAP] = 0UL;
2404 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2407 dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n");
2414 /* Common power [off|failure] */
2415 rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) +
2416 (mdev->p_uuid[UI_FLAGS] & 2);
2417 /* lowest bit is set when we were primary,
2418 * next bit (weight 2) is set when peer was primary */
2422 case 0: /* !self_pri && !peer_pri */ return 0;
2423 case 1: /* self_pri && !peer_pri */ return 1;
2424 case 2: /* !self_pri && peer_pri */ return -1;
2425 case 3: /* self_pri && peer_pri */
2426 dc = test_bit(DISCARD_CONCURRENT, &mdev->flags);
2432 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2437 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2439 if (mdev->agreed_pro_version < 96 ?
2440 (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) ==
2441 (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) :
2442 peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) {
2443 /* The last P_SYNC_UUID did not get though. Undo the last start of
2444 resync as sync source modifications of the peer's UUIDs. */
2446 if (mdev->agreed_pro_version < 91)
2449 mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START];
2450 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1];
2452 dev_info(DEV, "Did not got last syncUUID packet, corrected:\n");
2453 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2460 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2461 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2462 peer = mdev->p_uuid[i] & ~((u64)1);
2468 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2469 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2474 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2476 if (mdev->agreed_pro_version < 96 ?
2477 (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) ==
2478 (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) :
2479 self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) {
2480 /* The last P_SYNC_UUID did not get though. Undo the last start of
2481 resync as sync source modifications of our UUIDs. */
2483 if (mdev->agreed_pro_version < 91)
2486 _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]);
2487 _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]);
2489 dev_info(DEV, "Last syncUUID did not get through, corrected:\n");
2490 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2491 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2499 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2500 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2501 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2507 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2508 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2509 if (self == peer && self != ((u64)0))
2513 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2514 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2515 for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) {
2516 peer = mdev->p_uuid[j] & ~((u64)1);
2525 /* drbd_sync_handshake() returns the new conn state on success, or
2526 CONN_MASK (-1) on failure.
2528 static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role,
2529 enum drbd_disk_state peer_disk) __must_hold(local)
2532 enum drbd_conns rv = C_MASK;
2533 enum drbd_disk_state mydisk;
2535 mydisk = mdev->state.disk;
2536 if (mydisk == D_NEGOTIATING)
2537 mydisk = mdev->new_state_tmp.disk;
2539 dev_info(DEV, "drbd_sync_handshake:\n");
2540 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0);
2541 drbd_uuid_dump(mdev, "peer", mdev->p_uuid,
2542 mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2544 hg = drbd_uuid_compare(mdev, &rule_nr);
2546 dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr);
2549 dev_alert(DEV, "Unrelated data, aborting!\n");
2553 dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000);
2557 if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) ||
2558 (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) {
2559 int f = (hg == -100) || abs(hg) == 2;
2560 hg = mydisk > D_INCONSISTENT ? 1 : -1;
2563 dev_info(DEV, "Becoming sync %s due to disk states.\n",
2564 hg > 0 ? "source" : "target");
2568 drbd_khelper(mdev, "initial-split-brain");
2570 if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) {
2571 int pcount = (mdev->state.role == R_PRIMARY)
2572 + (peer_role == R_PRIMARY);
2573 int forced = (hg == -100);
2577 hg = drbd_asb_recover_0p(mdev);
2580 hg = drbd_asb_recover_1p(mdev);
2583 hg = drbd_asb_recover_2p(mdev);
2586 if (abs(hg) < 100) {
2587 dev_warn(DEV, "Split-Brain detected, %d primaries, "
2588 "automatically solved. Sync from %s node\n",
2589 pcount, (hg < 0) ? "peer" : "this");
2591 dev_warn(DEV, "Doing a full sync, since"
2592 " UUIDs where ambiguous.\n");
2599 if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1))
2601 if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1))
2605 dev_warn(DEV, "Split-Brain detected, manually solved. "
2606 "Sync from %s node\n",
2607 (hg < 0) ? "peer" : "this");
2611 /* FIXME this log message is not correct if we end up here
2612 * after an attempted attach on a diskless node.
2613 * We just refuse to attach -- well, we drop the "connection"
2614 * to that disk, in a way... */
2615 dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n");
2616 drbd_khelper(mdev, "split-brain");
2620 if (hg > 0 && mydisk <= D_INCONSISTENT) {
2621 dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n");
2625 if (hg < 0 && /* by intention we do not use mydisk here. */
2626 mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) {
2627 switch (mdev->net_conf->rr_conflict) {
2628 case ASB_CALL_HELPER:
2629 drbd_khelper(mdev, "pri-lost");
2631 case ASB_DISCONNECT:
2632 dev_err(DEV, "I shall become SyncTarget, but I am primary!\n");
2635 dev_warn(DEV, "Becoming SyncTarget, violating the stable-data"
2640 if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) {
2642 dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n");
2644 dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.",
2645 drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET),
2646 abs(hg) >= 2 ? "full" : "bit-map based");
2651 dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n");
2652 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake",
2653 BM_LOCKED_SET_ALLOWED))
2657 if (hg > 0) { /* become sync source. */
2659 } else if (hg < 0) { /* become sync target */
2663 if (drbd_bm_total_weight(mdev)) {
2664 dev_info(DEV, "No resync, but %lu bits in bitmap!\n",
2665 drbd_bm_total_weight(mdev));
2672 /* returns 1 if invalid */
2673 static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self)
2675 /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */
2676 if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) ||
2677 (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL))
2680 /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */
2681 if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL ||
2682 self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL)
2685 /* everything else is valid if they are equal on both sides. */
2689 /* everything es is invalid. */
2693 static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
2695 struct p_protocol *p = &mdev->data.rbuf.protocol;
2696 int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p;
2697 int p_want_lose, p_two_primaries, cf;
2698 char p_integrity_alg[SHARED_SECRET_MAX] = "";
2700 p_proto = be32_to_cpu(p->protocol);
2701 p_after_sb_0p = be32_to_cpu(p->after_sb_0p);
2702 p_after_sb_1p = be32_to_cpu(p->after_sb_1p);
2703 p_after_sb_2p = be32_to_cpu(p->after_sb_2p);
2704 p_two_primaries = be32_to_cpu(p->two_primaries);
2705 cf = be32_to_cpu(p->conn_flags);
2706 p_want_lose = cf & CF_WANT_LOSE;
2708 clear_bit(CONN_DRY_RUN, &mdev->flags);
2710 if (cf & CF_DRY_RUN)
2711 set_bit(CONN_DRY_RUN, &mdev->flags);
2713 if (p_proto != mdev->net_conf->wire_protocol) {
2714 dev_err(DEV, "incompatible communication protocols\n");
2718 if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) {
2719 dev_err(DEV, "incompatible after-sb-0pri settings\n");
2723 if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) {
2724 dev_err(DEV, "incompatible after-sb-1pri settings\n");
2728 if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) {
2729 dev_err(DEV, "incompatible after-sb-2pri settings\n");
2733 if (p_want_lose && mdev->net_conf->want_lose) {
2734 dev_err(DEV, "both sides have the 'want_lose' flag set\n");
2738 if (p_two_primaries != mdev->net_conf->two_primaries) {
2739 dev_err(DEV, "incompatible setting of the two-primaries options\n");
2743 if (mdev->agreed_pro_version >= 87) {
2744 unsigned char *my_alg = mdev->net_conf->integrity_alg;
2746 if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size)
2749 p_integrity_alg[SHARED_SECRET_MAX-1] = 0;
2750 if (strcmp(p_integrity_alg, my_alg)) {
2751 dev_err(DEV, "incompatible setting of the data-integrity-alg\n");
2754 dev_info(DEV, "data-integrity-alg: %s\n",
2755 my_alg[0] ? my_alg : (unsigned char *)"<not-used>");
2761 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2766 * input: alg name, feature name
2767 * return: NULL (alg name was "")
2768 * ERR_PTR(error) if something goes wrong
2769 * or the crypto hash ptr, if it worked out ok. */
2770 struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev,
2771 const char *alg, const char *name)
2773 struct crypto_hash *tfm;
2778 tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC);
2780 dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n",
2781 alg, name, PTR_ERR(tfm));
2784 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
2785 crypto_free_hash(tfm);
2786 dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name);
2787 return ERR_PTR(-EINVAL);
2792 static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size)
2795 struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95;
2796 unsigned int header_size, data_size, exp_max_sz;
2797 struct crypto_hash *verify_tfm = NULL;
2798 struct crypto_hash *csums_tfm = NULL;
2799 const int apv = mdev->agreed_pro_version;
2800 int *rs_plan_s = NULL;
2803 exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param)
2804 : apv == 88 ? sizeof(struct p_rs_param)
2806 : apv <= 94 ? sizeof(struct p_rs_param_89)
2807 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
2809 if (packet_size > exp_max_sz) {
2810 dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n",
2811 packet_size, exp_max_sz);
2816 header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80);
2817 data_size = packet_size - header_size;
2818 } else if (apv <= 94) {
2819 header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80);
2820 data_size = packet_size - header_size;
2821 D_ASSERT(data_size == 0);
2823 header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80);
2824 data_size = packet_size - header_size;
2825 D_ASSERT(data_size == 0);
2828 /* initialize verify_alg and csums_alg */
2829 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
2831 if (drbd_recv(mdev, &p->head.payload, header_size) != header_size)
2834 mdev->sync_conf.rate = be32_to_cpu(p->rate);
2838 if (data_size > SHARED_SECRET_MAX) {
2839 dev_err(DEV, "verify-alg too long, "
2840 "peer wants %u, accepting only %u byte\n",
2841 data_size, SHARED_SECRET_MAX);
2845 if (drbd_recv(mdev, p->verify_alg, data_size) != data_size)
2848 /* we expect NUL terminated string */
2849 /* but just in case someone tries to be evil */
2850 D_ASSERT(p->verify_alg[data_size-1] == 0);
2851 p->verify_alg[data_size-1] = 0;
2853 } else /* apv >= 89 */ {
2854 /* we still expect NUL terminated strings */
2855 /* but just in case someone tries to be evil */
2856 D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0);
2857 D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0);
2858 p->verify_alg[SHARED_SECRET_MAX-1] = 0;
2859 p->csums_alg[SHARED_SECRET_MAX-1] = 0;
2862 if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) {
2863 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2864 dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n",
2865 mdev->sync_conf.verify_alg, p->verify_alg);
2868 verify_tfm = drbd_crypto_alloc_digest_safe(mdev,
2869 p->verify_alg, "verify-alg");
2870 if (IS_ERR(verify_tfm)) {
2876 if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) {
2877 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2878 dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n",
2879 mdev->sync_conf.csums_alg, p->csums_alg);
2882 csums_tfm = drbd_crypto_alloc_digest_safe(mdev,
2883 p->csums_alg, "csums-alg");
2884 if (IS_ERR(csums_tfm)) {
2891 mdev->sync_conf.rate = be32_to_cpu(p->rate);
2892 mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead);
2893 mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target);
2894 mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target);
2895 mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate);
2897 fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
2898 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
2899 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
2901 dev_err(DEV, "kmalloc of fifo_buffer failed");
2907 spin_lock(&mdev->peer_seq_lock);
2908 /* lock against drbd_nl_syncer_conf() */
2910 strcpy(mdev->sync_conf.verify_alg, p->verify_alg);
2911 mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1;
2912 crypto_free_hash(mdev->verify_tfm);
2913 mdev->verify_tfm = verify_tfm;
2914 dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg);
2917 strcpy(mdev->sync_conf.csums_alg, p->csums_alg);
2918 mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1;
2919 crypto_free_hash(mdev->csums_tfm);
2920 mdev->csums_tfm = csums_tfm;
2921 dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg);
2923 if (fifo_size != mdev->rs_plan_s.size) {
2924 kfree(mdev->rs_plan_s.values);
2925 mdev->rs_plan_s.values = rs_plan_s;
2926 mdev->rs_plan_s.size = fifo_size;
2927 mdev->rs_planed = 0;
2929 spin_unlock(&mdev->peer_seq_lock);
2934 /* just for completeness: actually not needed,
2935 * as this is not reached if csums_tfm was ok. */
2936 crypto_free_hash(csums_tfm);
2937 /* but free the verify_tfm again, if csums_tfm did not work out */
2938 crypto_free_hash(verify_tfm);
2939 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2943 /* warn if the arguments differ by more than 12.5% */
2944 static void warn_if_differ_considerably(struct drbd_conf *mdev,
2945 const char *s, sector_t a, sector_t b)
2948 if (a == 0 || b == 0)
2950 d = (a > b) ? (a - b) : (b - a);
2951 if (d > (a>>3) || d > (b>>3))
2952 dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s,
2953 (unsigned long long)a, (unsigned long long)b);
2956 static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
2958 struct p_sizes *p = &mdev->data.rbuf.sizes;
2959 enum determine_dev_size dd = unchanged;
2960 sector_t p_size, p_usize, my_usize;
2961 int ldsc = 0; /* local disk size changed */
2962 enum dds_flags ddsf;
2964 p_size = be64_to_cpu(p->d_size);
2965 p_usize = be64_to_cpu(p->u_size);
2967 if (p_size == 0 && mdev->state.disk == D_DISKLESS) {
2968 dev_err(DEV, "some backing storage is needed\n");
2969 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2973 /* just store the peer's disk size for now.
2974 * we still need to figure out whether we accept that. */
2975 mdev->p_size = p_size;
2977 if (get_ldev(mdev)) {
2978 warn_if_differ_considerably(mdev, "lower level device sizes",
2979 p_size, drbd_get_max_capacity(mdev->ldev));
2980 warn_if_differ_considerably(mdev, "user requested size",
2981 p_usize, mdev->ldev->dc.disk_size);
2983 /* if this is the first connect, or an otherwise expected
2984 * param exchange, choose the minimum */
2985 if (mdev->state.conn == C_WF_REPORT_PARAMS)
2986 p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size,
2989 my_usize = mdev->ldev->dc.disk_size;
2991 if (mdev->ldev->dc.disk_size != p_usize) {
2992 mdev->ldev->dc.disk_size = p_usize;
2993 dev_info(DEV, "Peer sets u_size to %lu sectors\n",
2994 (unsigned long)mdev->ldev->dc.disk_size);
2997 /* Never shrink a device with usable data during connect.
2998 But allow online shrinking if we are connected. */
2999 if (drbd_new_dev_size(mdev, mdev->ldev, 0) <
3000 drbd_get_capacity(mdev->this_bdev) &&
3001 mdev->state.disk >= D_OUTDATED &&
3002 mdev->state.conn < C_CONNECTED) {
3003 dev_err(DEV, "The peer's disk size is too small!\n");
3004 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3005 mdev->ldev->dc.disk_size = my_usize;
3012 ddsf = be16_to_cpu(p->dds_flags);
3013 if (get_ldev(mdev)) {
3014 dd = drbd_determine_dev_size(mdev, ddsf);
3016 if (dd == dev_size_error)
3020 /* I am diskless, need to accept the peer's size. */
3021 drbd_set_my_capacity(mdev, p_size);
3024 mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size);
3025 drbd_reconsider_max_bio_size(mdev);
3027 if (get_ldev(mdev)) {
3028 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) {
3029 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
3036 if (mdev->state.conn > C_WF_REPORT_PARAMS) {
3037 if (be64_to_cpu(p->c_size) !=
3038 drbd_get_capacity(mdev->this_bdev) || ldsc) {
3039 /* we have different sizes, probably peer
3040 * needs to know my new size... */
3041 drbd_send_sizes(mdev, 0, ddsf);
3043 if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) ||
3044 (dd == grew && mdev->state.conn == C_CONNECTED)) {
3045 if (mdev->state.pdsk >= D_INCONSISTENT &&
3046 mdev->state.disk >= D_INCONSISTENT) {
3047 if (ddsf & DDSF_NO_RESYNC)
3048 dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n");
3050 resync_after_online_grow(mdev);
3052 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
3059 static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3061 struct p_uuids *p = &mdev->data.rbuf.uuids;
3063 int i, updated_uuids = 0;
3065 p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO);
3067 for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++)
3068 p_uuid[i] = be64_to_cpu(p->uuid[i]);
3070 kfree(mdev->p_uuid);
3071 mdev->p_uuid = p_uuid;
3073 if (mdev->state.conn < C_CONNECTED &&
3074 mdev->state.disk < D_INCONSISTENT &&
3075 mdev->state.role == R_PRIMARY &&
3076 (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) {
3077 dev_err(DEV, "Can only connect to data with current UUID=%016llX\n",
3078 (unsigned long long)mdev->ed_uuid);
3079 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3083 if (get_ldev(mdev)) {
3084 int skip_initial_sync =
3085 mdev->state.conn == C_CONNECTED &&
3086 mdev->agreed_pro_version >= 90 &&
3087 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
3088 (p_uuid[UI_FLAGS] & 8);
3089 if (skip_initial_sync) {
3090 dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n");
3091 drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
3092 "clear_n_write from receive_uuids",
3093 BM_LOCKED_TEST_ALLOWED);
3094 _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]);
3095 _drbd_uuid_set(mdev, UI_BITMAP, 0);
3096 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3102 } else if (mdev->state.disk < D_INCONSISTENT &&
3103 mdev->state.role == R_PRIMARY) {
3104 /* I am a diskless primary, the peer just created a new current UUID
3106 updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3109 /* Before we test for the disk state, we should wait until an eventually
3110 ongoing cluster wide state change is finished. That is important if
3111 we are primary and are detaching from our disk. We need to see the
3112 new disk state... */
3113 wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags));
3114 if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT)
3115 updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3118 drbd_print_uuids(mdev, "receiver updated UUIDs to");
3124 * convert_state() - Converts the peer's view of the cluster state to our point of view
3125 * @ps: The state as seen by the peer.
3127 static union drbd_state convert_state(union drbd_state ps)
3129 union drbd_state ms;
3131 static enum drbd_conns c_tab[] = {
3132 [C_CONNECTED] = C_CONNECTED,
3134 [C_STARTING_SYNC_S] = C_STARTING_SYNC_T,
3135 [C_STARTING_SYNC_T] = C_STARTING_SYNC_S,
3136 [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */
3137 [C_VERIFY_S] = C_VERIFY_T,
3143 ms.conn = c_tab[ps.conn];
3148 ms.peer_isp = (ps.aftr_isp | ps.user_isp);
3153 static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3155 struct p_req_state *p = &mdev->data.rbuf.req_state;
3156 union drbd_state mask, val;
3157 enum drbd_state_rv rv;
3159 mask.i = be32_to_cpu(p->mask);
3160 val.i = be32_to_cpu(p->val);
3162 if (test_bit(DISCARD_CONCURRENT, &mdev->flags) &&
3163 test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) {
3164 drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG);
3168 mask = convert_state(mask);
3169 val = convert_state(val);
3171 rv = drbd_change_state(mdev, CS_VERBOSE, mask, val);
3173 drbd_send_sr_reply(mdev, rv);
3179 static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3181 struct p_state *p = &mdev->data.rbuf.state;
3182 union drbd_state os, ns, peer_state;
3183 enum drbd_disk_state real_peer_disk;
3184 enum chg_state_flags cs_flags;
3187 peer_state.i = be32_to_cpu(p->state);
3189 real_peer_disk = peer_state.disk;
3190 if (peer_state.disk == D_NEGOTIATING) {
3191 real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT;
3192 dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk));
3195 spin_lock_irq(&mdev->req_lock);
3197 os = ns = mdev->state;
3198 spin_unlock_irq(&mdev->req_lock);
3200 /* peer says his disk is uptodate, while we think it is inconsistent,
3201 * and this happens while we think we have a sync going on. */
3202 if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE &&
3203 os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) {
3204 /* If we are (becoming) SyncSource, but peer is still in sync
3205 * preparation, ignore its uptodate-ness to avoid flapping, it
3206 * will change to inconsistent once the peer reaches active
3208 * It may have changed syncer-paused flags, however, so we
3209 * cannot ignore this completely. */
3210 if (peer_state.conn > C_CONNECTED &&
3211 peer_state.conn < C_SYNC_SOURCE)
3212 real_peer_disk = D_INCONSISTENT;
3214 /* if peer_state changes to connected at the same time,
3215 * it explicitly notifies us that it finished resync.
3216 * Maybe we should finish it up, too? */
3217 else if (os.conn >= C_SYNC_SOURCE &&
3218 peer_state.conn == C_CONNECTED) {
3219 if (drbd_bm_total_weight(mdev) <= mdev->rs_failed)
3220 drbd_resync_finished(mdev);
3225 /* peer says his disk is inconsistent, while we think it is uptodate,
3226 * and this happens while the peer still thinks we have a sync going on,
3227 * but we think we are already done with the sync.
3228 * We ignore this to avoid flapping pdsk.
3229 * This should not happen, if the peer is a recent version of drbd. */
3230 if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT &&
3231 os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE)
3232 real_peer_disk = D_UP_TO_DATE;
3234 if (ns.conn == C_WF_REPORT_PARAMS)
3235 ns.conn = C_CONNECTED;
3237 if (peer_state.conn == C_AHEAD)
3240 if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING &&
3241 get_ldev_if_state(mdev, D_NEGOTIATING)) {
3242 int cr; /* consider resync */
3244 /* if we established a new connection */
3245 cr = (os.conn < C_CONNECTED);
3246 /* if we had an established connection
3247 * and one of the nodes newly attaches a disk */
3248 cr |= (os.conn == C_CONNECTED &&
3249 (peer_state.disk == D_NEGOTIATING ||
3250 os.disk == D_NEGOTIATING));
3251 /* if we have both been inconsistent, and the peer has been
3252 * forced to be UpToDate with --overwrite-data */
3253 cr |= test_bit(CONSIDER_RESYNC, &mdev->flags);
3254 /* if we had been plain connected, and the admin requested to
3255 * start a sync by "invalidate" or "invalidate-remote" */
3256 cr |= (os.conn == C_CONNECTED &&
3257 (peer_state.conn >= C_STARTING_SYNC_S &&
3258 peer_state.conn <= C_WF_BITMAP_T));
3261 ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk);
3264 if (ns.conn == C_MASK) {
3265 ns.conn = C_CONNECTED;
3266 if (mdev->state.disk == D_NEGOTIATING) {
3267 drbd_force_state(mdev, NS(disk, D_FAILED));
3268 } else if (peer_state.disk == D_NEGOTIATING) {
3269 dev_err(DEV, "Disk attach process on the peer node was aborted.\n");
3270 peer_state.disk = D_DISKLESS;
3271 real_peer_disk = D_DISKLESS;
3273 if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags))
3275 D_ASSERT(os.conn == C_WF_REPORT_PARAMS);
3276 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3282 spin_lock_irq(&mdev->req_lock);
3283 if (mdev->state.i != os.i)
3285 clear_bit(CONSIDER_RESYNC, &mdev->flags);
3286 ns.peer = peer_state.role;
3287 ns.pdsk = real_peer_disk;
3288 ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp);
3289 if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING)
3290 ns.disk = mdev->new_state_tmp.disk;
3291 cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD);
3292 if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED &&
3293 test_bit(NEW_CUR_UUID, &mdev->flags)) {
3294 /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this
3295 for temporal network outages! */
3296 spin_unlock_irq(&mdev->req_lock);
3297 dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n");
3299 drbd_uuid_new_current(mdev);
3300 clear_bit(NEW_CUR_UUID, &mdev->flags);
3301 drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0));
3304 rv = _drbd_set_state(mdev, ns, cs_flags, NULL);
3306 spin_unlock_irq(&mdev->req_lock);
3308 if (rv < SS_SUCCESS) {
3309 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3313 if (os.conn > C_WF_REPORT_PARAMS) {
3314 if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED &&
3315 peer_state.disk != D_NEGOTIATING ) {
3316 /* we want resync, peer has not yet decided to sync... */
3317 /* Nowadays only used when forcing a node into primary role and
3318 setting its disk to UpToDate with that */
3319 drbd_send_uuids(mdev);
3320 drbd_send_state(mdev);
3324 mdev->net_conf->want_lose = 0;
3326 drbd_md_sync(mdev); /* update connected indicator, la_size, ... */
3331 static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3333 struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid;
3335 wait_event(mdev->misc_wait,
3336 mdev->state.conn == C_WF_SYNC_UUID ||
3337 mdev->state.conn == C_BEHIND ||
3338 mdev->state.conn < C_CONNECTED ||
3339 mdev->state.disk < D_NEGOTIATING);
3341 /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */
3343 /* Here the _drbd_uuid_ functions are right, current should
3344 _not_ be rotated into the history */
3345 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
3346 _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid));
3347 _drbd_uuid_set(mdev, UI_BITMAP, 0UL);
3349 drbd_print_uuids(mdev, "updated sync uuid");
3350 drbd_start_resync(mdev, C_SYNC_TARGET);
3354 dev_err(DEV, "Ignoring SyncUUID packet!\n");
3360 * receive_bitmap_plain
3362 * Return 0 when done, 1 when another iteration is needed, and a negative error
3363 * code upon failure.
3366 receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size,
3367 unsigned long *buffer, struct bm_xfer_ctx *c)
3369 unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset);
3370 unsigned want = num_words * sizeof(long);
3373 if (want != data_size) {
3374 dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size);
3379 err = drbd_recv(mdev, buffer, want);
3386 drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer);
3388 c->word_offset += num_words;
3389 c->bit_offset = c->word_offset * BITS_PER_LONG;
3390 if (c->bit_offset > c->bm_bits)
3391 c->bit_offset = c->bm_bits;
3399 * Return 0 when done, 1 when another iteration is needed, and a negative error
3400 * code upon failure.
3403 recv_bm_rle_bits(struct drbd_conf *mdev,
3404 struct p_compressed_bm *p,
3405 struct bm_xfer_ctx *c)
3407 struct bitstream bs;
3411 unsigned long s = c->bit_offset;
3413 int len = be16_to_cpu(p->head.length) - (sizeof(*p) - sizeof(p->head));
3414 int toggle = DCBP_get_start(p);
3418 bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p));
3420 bits = bitstream_get_bits(&bs, &look_ahead, 64);
3424 for (have = bits; have > 0; s += rl, toggle = !toggle) {
3425 bits = vli_decode_bits(&rl, look_ahead);
3431 if (e >= c->bm_bits) {
3432 dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e);
3435 _drbd_bm_set_bits(mdev, s, e);
3439 dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n",
3440 have, bits, look_ahead,
3441 (unsigned int)(bs.cur.b - p->code),
3442 (unsigned int)bs.buf_len);
3445 look_ahead >>= bits;
3448 bits = bitstream_get_bits(&bs, &tmp, 64 - have);
3451 look_ahead |= tmp << have;
3456 bm_xfer_ctx_bit_to_word_offset(c);
3458 return (s != c->bm_bits);
3464 * Return 0 when done, 1 when another iteration is needed, and a negative error
3465 * code upon failure.
3468 decode_bitmap_c(struct drbd_conf *mdev,
3469 struct p_compressed_bm *p,
3470 struct bm_xfer_ctx *c)
3472 if (DCBP_get_code(p) == RLE_VLI_Bits)
3473 return recv_bm_rle_bits(mdev, p, c);
3475 /* other variants had been implemented for evaluation,
3476 * but have been dropped as this one turned out to be "best"
3477 * during all our tests. */
3479 dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding);
3480 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3484 void INFO_bm_xfer_stats(struct drbd_conf *mdev,
3485 const char *direction, struct bm_xfer_ctx *c)
3487 /* what would it take to transfer it "plaintext" */
3488 unsigned plain = sizeof(struct p_header80) *
3489 ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1)
3490 + c->bm_words * sizeof(long);
3491 unsigned total = c->bytes[0] + c->bytes[1];
3494 /* total can not be zero. but just in case: */
3498 /* don't report if not compressed */
3502 /* total < plain. check for overflow, still */
3503 r = (total > UINT_MAX/1000) ? (total / (plain/1000))
3504 : (1000 * total / plain);
3510 dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), "
3511 "total %u; compression: %u.%u%%\n",
3513 c->bytes[1], c->packets[1],
3514 c->bytes[0], c->packets[0],
3515 total, r/10, r % 10);
3518 /* Since we are processing the bitfield from lower addresses to higher,
3519 it does not matter if the process it in 32 bit chunks or 64 bit
3520 chunks as long as it is little endian. (Understand it as byte stream,
3521 beginning with the lowest byte...) If we would use big endian
3522 we would need to process it from the highest address to the lowest,
3523 in order to be agnostic to the 32 vs 64 bits issue.
3525 returns 0 on failure, 1 if we successfully received it. */
3526 static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3528 struct bm_xfer_ctx c;
3532 struct p_header80 *h = &mdev->data.rbuf.header.h80;
3534 drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED);
3535 /* you are supposed to send additional out-of-sync information
3536 * if you actually set bits during this phase */
3538 /* maybe we should use some per thread scratch page,
3539 * and allocate that during initial device creation? */
3540 buffer = (unsigned long *) __get_free_page(GFP_NOIO);
3542 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__);
3546 c = (struct bm_xfer_ctx) {
3547 .bm_bits = drbd_bm_bits(mdev),
3548 .bm_words = drbd_bm_words(mdev),
3552 if (cmd == P_BITMAP) {
3553 err = receive_bitmap_plain(mdev, data_size, buffer, &c);
3554 } else if (cmd == P_COMPRESSED_BITMAP) {
3555 /* MAYBE: sanity check that we speak proto >= 90,
3556 * and the feature is enabled! */
3557 struct p_compressed_bm *p;
3559 if (data_size > BM_PACKET_PAYLOAD_BYTES) {
3560 dev_err(DEV, "ReportCBitmap packet too large\n");
3563 /* use the page buff */
3565 memcpy(p, h, sizeof(*h));
3566 if (drbd_recv(mdev, p->head.payload, data_size) != data_size)
3568 if (data_size <= (sizeof(*p) - sizeof(p->head))) {
3569 dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size);
3572 err = decode_bitmap_c(mdev, p, &c);
3574 dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd);
3578 c.packets[cmd == P_BITMAP]++;
3579 c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size;
3586 if (!drbd_recv_header(mdev, &cmd, &data_size))
3590 INFO_bm_xfer_stats(mdev, "receive", &c);
3592 if (mdev->state.conn == C_WF_BITMAP_T) {
3593 enum drbd_state_rv rv;
3595 ok = !drbd_send_bitmap(mdev);
3598 /* Omit CS_ORDERED with this state transition to avoid deadlocks. */
3599 rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
3600 D_ASSERT(rv == SS_SUCCESS);
3601 } else if (mdev->state.conn != C_WF_BITMAP_S) {
3602 /* admin may have requested C_DISCONNECTING,
3603 * other threads may have noticed network errors */
3604 dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n",
3605 drbd_conn_str(mdev->state.conn));
3610 drbd_bm_unlock(mdev);
3611 if (ok && mdev->state.conn == C_WF_BITMAP_S)
3612 drbd_start_resync(mdev, C_SYNC_SOURCE);
3613 free_page((unsigned long) buffer);
3617 static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3619 /* TODO zero copy sink :) */
3620 static char sink[128];
3623 dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n",
3628 want = min_t(int, size, sizeof(sink));
3629 r = drbd_recv(mdev, sink, want);
3630 ERR_IF(r <= 0) break;
3636 static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3638 /* Make sure we've acked all the TCP data associated
3639 * with the data requests being unplugged */
3640 drbd_tcp_quickack(mdev->data.socket);
3645 static int receive_out_of_sync(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3647 struct p_block_desc *p = &mdev->data.rbuf.block_desc;
3649 switch (mdev->state.conn) {
3650 case C_WF_SYNC_UUID:
3655 dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n",
3656 drbd_conn_str(mdev->state.conn));
3659 drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize));
3664 typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive);
3669 drbd_cmd_handler_f function;
3672 static struct data_cmd drbd_cmd_handler[] = {
3673 [P_DATA] = { 1, sizeof(struct p_data), receive_Data },
3674 [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply },
3675 [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } ,
3676 [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } ,
3677 [P_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } ,
3678 [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } ,
3679 [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header80), receive_UnplugRemote },
3680 [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3681 [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3682 [P_SYNC_PARAM] = { 1, sizeof(struct p_header80), receive_SyncParam },
3683 [P_SYNC_PARAM89] = { 1, sizeof(struct p_header80), receive_SyncParam },
3684 [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol },
3685 [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids },
3686 [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes },
3687 [P_STATE] = { 0, sizeof(struct p_state), receive_state },
3688 [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state },
3689 [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid },
3690 [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3691 [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest },
3692 [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest },
3693 [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip },
3694 [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync },
3695 /* anything missing from this table is in
3696 * the asender_tbl, see get_asender_cmd */
3697 [P_MAX_CMD] = { 0, 0, NULL },
3700 /* All handler functions that expect a sub-header get that sub-heder in
3701 mdev->data.rbuf.header.head.payload.
3703 Usually in mdev->data.rbuf.header.head the callback can find the usual
3704 p_header, but they may not rely on that. Since there is also p_header95 !
3707 static void drbdd(struct drbd_conf *mdev)
3709 union p_header *header = &mdev->data.rbuf.header;
3710 unsigned int packet_size;
3711 enum drbd_packets cmd;
3712 size_t shs; /* sub header size */
3715 while (get_t_state(&mdev->receiver) == Running) {
3716 drbd_thread_current_set_cpu(mdev);
3717 if (!drbd_recv_header(mdev, &cmd, &packet_size))
3720 if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) {
3721 dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size);
3725 shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header);
3726 if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) {
3727 dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size);
3732 rv = drbd_recv(mdev, &header->h80.payload, shs);
3733 if (unlikely(rv != shs)) {
3734 if (!signal_pending(current))
3735 dev_warn(DEV, "short read while reading sub header: rv=%d\n", rv);
3740 rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs);
3742 if (unlikely(!rv)) {
3743 dev_err(DEV, "error receiving %s, l: %d!\n",
3744 cmdname(cmd), packet_size);
3751 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3753 /* If we leave here, we probably want to update at least the
3754 * "Connected" indicator on stable storage. Do so explicitly here. */
3758 void drbd_flush_workqueue(struct drbd_conf *mdev)
3760 struct drbd_wq_barrier barr;
3762 barr.w.cb = w_prev_work_done;
3763 init_completion(&barr.done);
3764 drbd_queue_work(&mdev->data.work, &barr.w);
3765 wait_for_completion(&barr.done);
3768 static void drbd_disconnect(struct drbd_conf *mdev)
3770 enum drbd_fencing_p fp;
3771 union drbd_state os, ns;
3772 int rv = SS_UNKNOWN_ERROR;
3775 if (mdev->state.conn == C_STANDALONE)
3778 /* asender does not clean up anything. it must not interfere, either */
3779 drbd_thread_stop(&mdev->asender);
3780 drbd_free_sock(mdev);
3782 /* wait for current activity to cease. */
3783 spin_lock_irq(&mdev->req_lock);
3784 _drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
3785 _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee);
3786 _drbd_wait_ee_list_empty(mdev, &mdev->read_ee);
3787 spin_unlock_irq(&mdev->req_lock);
3789 /* We do not have data structures that would allow us to
3790 * get the rs_pending_cnt down to 0 again.
3791 * * On C_SYNC_TARGET we do not have any data structures describing
3792 * the pending RSDataRequest's we have sent.
3793 * * On C_SYNC_SOURCE there is no data structure that tracks
3794 * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget.
3795 * And no, it is not the sum of the reference counts in the
3796 * resync_LRU. The resync_LRU tracks the whole operation including
3797 * the disk-IO, while the rs_pending_cnt only tracks the blocks
3799 drbd_rs_cancel_all(mdev);
3801 mdev->rs_failed = 0;
3802 atomic_set(&mdev->rs_pending_cnt, 0);
3803 wake_up(&mdev->misc_wait);
3805 del_timer(&mdev->request_timer);
3807 /* make sure syncer is stopped and w_resume_next_sg queued */
3808 del_timer_sync(&mdev->resync_timer);
3809 resync_timer_fn((unsigned long)mdev);
3811 /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier,
3812 * w_make_resync_request etc. which may still be on the worker queue
3813 * to be "canceled" */
3814 drbd_flush_workqueue(mdev);
3816 /* This also does reclaim_net_ee(). If we do this too early, we might
3817 * miss some resync ee and pages.*/
3818 drbd_process_done_ee(mdev);
3820 kfree(mdev->p_uuid);
3821 mdev->p_uuid = NULL;
3823 if (!is_susp(mdev->state))
3826 dev_info(DEV, "Connection closed\n");
3831 if (get_ldev(mdev)) {
3832 fp = mdev->ldev->dc.fencing;
3836 if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN)
3837 drbd_try_outdate_peer_async(mdev);
3839 spin_lock_irq(&mdev->req_lock);
3841 if (os.conn >= C_UNCONNECTED) {
3842 /* Do not restart in case we are C_DISCONNECTING */
3844 ns.conn = C_UNCONNECTED;
3845 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
3847 spin_unlock_irq(&mdev->req_lock);
3849 if (os.conn == C_DISCONNECTING) {
3850 wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0);
3852 crypto_free_hash(mdev->cram_hmac_tfm);
3853 mdev->cram_hmac_tfm = NULL;
3855 kfree(mdev->net_conf);
3856 mdev->net_conf = NULL;
3857 drbd_request_state(mdev, NS(conn, C_STANDALONE));
3860 /* serialize with bitmap writeout triggered by the state change,
3862 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
3864 /* tcp_close and release of sendpage pages can be deferred. I don't
3865 * want to use SO_LINGER, because apparently it can be deferred for
3866 * more than 20 seconds (longest time I checked).
3868 * Actually we don't care for exactly when the network stack does its
3869 * put_page(), but release our reference on these pages right here.
3871 i = drbd_release_ee(mdev, &mdev->net_ee);
3873 dev_info(DEV, "net_ee not empty, killed %u entries\n", i);
3874 i = atomic_read(&mdev->pp_in_use_by_net);
3876 dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i);
3877 i = atomic_read(&mdev->pp_in_use);
3879 dev_info(DEV, "pp_in_use = %d, expected 0\n", i);
3881 D_ASSERT(list_empty(&mdev->read_ee));
3882 D_ASSERT(list_empty(&mdev->active_ee));
3883 D_ASSERT(list_empty(&mdev->sync_ee));
3884 D_ASSERT(list_empty(&mdev->done_ee));
3886 /* ok, no more ee's on the fly, it is safe to reset the epoch_size */
3887 atomic_set(&mdev->current_epoch->epoch_size, 0);
3888 D_ASSERT(list_empty(&mdev->current_epoch->list));
3892 * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version
3893 * we can agree on is stored in agreed_pro_version.
3895 * feature flags and the reserved array should be enough room for future
3896 * enhancements of the handshake protocol, and possible plugins...
3898 * for now, they are expected to be zero, but ignored.
3900 static int drbd_send_handshake(struct drbd_conf *mdev)
3902 /* ASSERT current == mdev->receiver ... */
3903 struct p_handshake *p = &mdev->data.sbuf.handshake;
3906 if (mutex_lock_interruptible(&mdev->data.mutex)) {
3907 dev_err(DEV, "interrupted during initial handshake\n");
3908 return 0; /* interrupted. not ok. */
3911 if (mdev->data.socket == NULL) {
3912 mutex_unlock(&mdev->data.mutex);
3916 memset(p, 0, sizeof(*p));
3917 p->protocol_min = cpu_to_be32(PRO_VERSION_MIN);
3918 p->protocol_max = cpu_to_be32(PRO_VERSION_MAX);
3919 ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE,
3920 (struct p_header80 *)p, sizeof(*p), 0 );
3921 mutex_unlock(&mdev->data.mutex);
3927 * 1 yes, we have a valid connection
3928 * 0 oops, did not work out, please try again
3929 * -1 peer talks different language,
3930 * no point in trying again, please go standalone.
3932 static int drbd_do_handshake(struct drbd_conf *mdev)
3934 /* ASSERT current == mdev->receiver ... */
3935 struct p_handshake *p = &mdev->data.rbuf.handshake;
3936 const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80);
3937 unsigned int length;
3938 enum drbd_packets cmd;
3941 rv = drbd_send_handshake(mdev);
3945 rv = drbd_recv_header(mdev, &cmd, &length);
3949 if (cmd != P_HAND_SHAKE) {
3950 dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n",
3955 if (length != expect) {
3956 dev_err(DEV, "expected HandShake length: %u, received: %u\n",
3961 rv = drbd_recv(mdev, &p->head.payload, expect);
3964 if (!signal_pending(current))
3965 dev_warn(DEV, "short read receiving handshake packet: l=%u\n", rv);
3969 p->protocol_min = be32_to_cpu(p->protocol_min);
3970 p->protocol_max = be32_to_cpu(p->protocol_max);
3971 if (p->protocol_max == 0)
3972 p->protocol_max = p->protocol_min;
3974 if (PRO_VERSION_MAX < p->protocol_min ||
3975 PRO_VERSION_MIN > p->protocol_max)
3978 mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max);
3980 dev_info(DEV, "Handshake successful: "
3981 "Agreed network protocol version %d\n", mdev->agreed_pro_version);
3986 dev_err(DEV, "incompatible DRBD dialects: "
3987 "I support %d-%d, peer supports %d-%d\n",
3988 PRO_VERSION_MIN, PRO_VERSION_MAX,
3989 p->protocol_min, p->protocol_max);
3993 #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE)
3994 static int drbd_do_auth(struct drbd_conf *mdev)
3996 dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n");
3997 dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n");
4001 #define CHALLENGE_LEN 64
4005 0 - failed, try again (network error),
4006 -1 - auth failed, don't try again.
4009 static int drbd_do_auth(struct drbd_conf *mdev)
4011 char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */
4012 struct scatterlist sg;
4013 char *response = NULL;
4014 char *right_response = NULL;
4015 char *peers_ch = NULL;
4016 unsigned int key_len = strlen(mdev->net_conf->shared_secret);
4017 unsigned int resp_size;
4018 struct hash_desc desc;
4019 enum drbd_packets cmd;
4020 unsigned int length;
4023 desc.tfm = mdev->cram_hmac_tfm;
4026 rv = crypto_hash_setkey(mdev->cram_hmac_tfm,
4027 (u8 *)mdev->net_conf->shared_secret, key_len);
4029 dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv);
4034 get_random_bytes(my_challenge, CHALLENGE_LEN);
4036 rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN);
4040 rv = drbd_recv_header(mdev, &cmd, &length);
4044 if (cmd != P_AUTH_CHALLENGE) {
4045 dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n",
4051 if (length > CHALLENGE_LEN * 2) {
4052 dev_err(DEV, "expected AuthChallenge payload too big.\n");
4057 peers_ch = kmalloc(length, GFP_NOIO);
4058 if (peers_ch == NULL) {
4059 dev_err(DEV, "kmalloc of peers_ch failed\n");
4064 rv = drbd_recv(mdev, peers_ch, length);
4067 if (!signal_pending(current))
4068 dev_warn(DEV, "short read AuthChallenge: l=%u\n", rv);
4073 resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm);
4074 response = kmalloc(resp_size, GFP_NOIO);
4075 if (response == NULL) {
4076 dev_err(DEV, "kmalloc of response failed\n");
4081 sg_init_table(&sg, 1);
4082 sg_set_buf(&sg, peers_ch, length);
4084 rv = crypto_hash_digest(&desc, &sg, sg.length, response);
4086 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4091 rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size);
4095 rv = drbd_recv_header(mdev, &cmd, &length);
4099 if (cmd != P_AUTH_RESPONSE) {
4100 dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n",
4106 if (length != resp_size) {
4107 dev_err(DEV, "expected AuthResponse payload of wrong size\n");
4112 rv = drbd_recv(mdev, response , resp_size);
4114 if (rv != resp_size) {
4115 if (!signal_pending(current))
4116 dev_warn(DEV, "short read receiving AuthResponse: l=%u\n", rv);
4121 right_response = kmalloc(resp_size, GFP_NOIO);
4122 if (right_response == NULL) {
4123 dev_err(DEV, "kmalloc of right_response failed\n");
4128 sg_set_buf(&sg, my_challenge, CHALLENGE_LEN);
4130 rv = crypto_hash_digest(&desc, &sg, sg.length, right_response);
4132 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4137 rv = !memcmp(response, right_response, resp_size);
4140 dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n",
4141 resp_size, mdev->net_conf->cram_hmac_alg);
4148 kfree(right_response);
4154 int drbdd_init(struct drbd_thread *thi)
4156 struct drbd_conf *mdev = thi->mdev;
4157 unsigned int minor = mdev_to_minor(mdev);
4160 sprintf(current->comm, "drbd%d_receiver", minor);
4162 dev_info(DEV, "receiver (re)started\n");
4165 h = drbd_connect(mdev);
4167 drbd_disconnect(mdev);
4168 schedule_timeout_interruptible(HZ);
4171 dev_warn(DEV, "Discarding network configuration.\n");
4172 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4177 if (get_net_conf(mdev)) {
4183 drbd_disconnect(mdev);
4185 dev_info(DEV, "receiver terminated\n");
4189 /* ********* acknowledge sender ******** */
4191 static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h)
4193 struct p_req_state_reply *p = (struct p_req_state_reply *)h;
4195 int retcode = be32_to_cpu(p->retcode);
4197 if (retcode >= SS_SUCCESS) {
4198 set_bit(CL_ST_CHG_SUCCESS, &mdev->flags);
4200 set_bit(CL_ST_CHG_FAIL, &mdev->flags);
4201 dev_err(DEV, "Requested state change failed by peer: %s (%d)\n",
4202 drbd_set_st_err_str(retcode), retcode);
4204 wake_up(&mdev->state_wait);
4209 static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h)
4211 return drbd_send_ping_ack(mdev);
4215 static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h)
4217 /* restore idle timeout */
4218 mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
4219 if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags))
4220 wake_up(&mdev->misc_wait);
4225 static int got_IsInSync(struct drbd_conf *mdev, struct p_header80 *h)
4227 struct p_block_ack *p = (struct p_block_ack *)h;
4228 sector_t sector = be64_to_cpu(p->sector);
4229 int blksize = be32_to_cpu(p->blksize);
4231 D_ASSERT(mdev->agreed_pro_version >= 89);
4233 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4235 if (get_ldev(mdev)) {
4236 drbd_rs_complete_io(mdev, sector);
4237 drbd_set_in_sync(mdev, sector, blksize);
4238 /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */
4239 mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT);
4242 dec_rs_pending(mdev);
4243 atomic_add(blksize >> 9, &mdev->rs_sect_in);
4248 static int validate_req_change_req_state(struct drbd_conf *mdev,
4249 u64 id, sector_t sector,
4250 struct hlist_head *(*hash_slot)(struct drbd_conf *, sector_t),
4251 const char *func, enum drbd_req_event what)
4253 struct drbd_request *req;
4254 struct bio_and_error m;
4256 spin_lock_irq(&mdev->req_lock);
4257 req = find_request(mdev, hash_slot, id, sector, func);
4258 if (unlikely(!req)) {
4259 spin_unlock_irq(&mdev->req_lock);
4261 dev_err(DEV, "%s: failed to find req %p, sector %llus\n", func,
4262 (void *)(unsigned long)id, (unsigned long long)sector);
4265 __req_mod(req, what, &m);
4266 spin_unlock_irq(&mdev->req_lock);
4269 complete_master_bio(mdev, &m);
4273 static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h)
4275 struct p_block_ack *p = (struct p_block_ack *)h;
4276 sector_t sector = be64_to_cpu(p->sector);
4277 int blksize = be32_to_cpu(p->blksize);
4278 enum drbd_req_event what;
4280 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4282 if (p->block_id == ID_SYNCER) {
4283 drbd_set_in_sync(mdev, sector, blksize);
4284 dec_rs_pending(mdev);
4287 switch (be16_to_cpu(h->command)) {
4288 case P_RS_WRITE_ACK:
4289 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4290 what = write_acked_by_peer_and_sis;
4293 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4294 what = write_acked_by_peer;
4297 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B);
4298 what = recv_acked_by_peer;
4301 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4302 what = conflict_discarded_by_peer;
4309 return validate_req_change_req_state(mdev, p->block_id, sector,
4310 tl_hash_slot, __func__, what);
4313 static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h)
4315 struct p_block_ack *p = (struct p_block_ack *)h;
4316 sector_t sector = be64_to_cpu(p->sector);
4317 int size = be32_to_cpu(p->blksize);
4318 struct drbd_request *req;
4319 struct bio_and_error m;
4321 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4323 if (p->block_id == ID_SYNCER) {
4324 dec_rs_pending(mdev);
4325 drbd_rs_failed_io(mdev, sector, size);
4329 spin_lock_irq(&mdev->req_lock);
4330 req = find_request(mdev, tl_hash_slot, p->block_id, sector, __func__);
4332 spin_unlock_irq(&mdev->req_lock);
4333 if (mdev->net_conf->wire_protocol == DRBD_PROT_A ||
4334 mdev->net_conf->wire_protocol == DRBD_PROT_B) {
4335 /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs.
4336 The master bio might already be completed, therefore the
4337 request is no longer in the collision hash.
4338 => Do not try to validate block_id as request. */
4339 /* In Protocol B we might already have got a P_RECV_ACK
4340 but then get a P_NEG_ACK after wards. */
4341 drbd_set_out_of_sync(mdev, sector, size);
4344 dev_err(DEV, "%s: failed to find req %p, sector %llus\n", __func__,
4345 (void *)(unsigned long)p->block_id, (unsigned long long)sector);
4349 __req_mod(req, neg_acked, &m);
4350 spin_unlock_irq(&mdev->req_lock);
4353 complete_master_bio(mdev, &m);
4357 static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h)
4359 struct p_block_ack *p = (struct p_block_ack *)h;
4360 sector_t sector = be64_to_cpu(p->sector);
4362 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4363 dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n",
4364 (unsigned long long)sector, be32_to_cpu(p->blksize));
4366 return validate_req_change_req_state(mdev, p->block_id, sector,
4367 ar_hash_slot, __func__, neg_acked);
4370 static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h)
4374 struct p_block_ack *p = (struct p_block_ack *)h;
4376 sector = be64_to_cpu(p->sector);
4377 size = be32_to_cpu(p->blksize);
4379 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4381 dec_rs_pending(mdev);
4383 if (get_ldev_if_state(mdev, D_FAILED)) {
4384 drbd_rs_complete_io(mdev, sector);
4385 switch (be16_to_cpu(h->command)) {
4386 case P_NEG_RS_DREPLY:
4387 drbd_rs_failed_io(mdev, sector, size);
4401 static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h)
4403 struct p_barrier_ack *p = (struct p_barrier_ack *)h;
4405 tl_release(mdev, p->barrier, be32_to_cpu(p->set_size));
4407 if (mdev->state.conn == C_AHEAD &&
4408 atomic_read(&mdev->ap_in_flight) == 0 &&
4409 !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags)) {
4410 mdev->start_resync_timer.expires = jiffies + HZ;
4411 add_timer(&mdev->start_resync_timer);
4417 static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h)
4419 struct p_block_ack *p = (struct p_block_ack *)h;
4420 struct drbd_work *w;
4424 sector = be64_to_cpu(p->sector);
4425 size = be32_to_cpu(p->blksize);
4427 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4429 if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC)
4430 drbd_ov_oos_found(mdev, sector, size);
4434 if (!get_ldev(mdev))
4437 drbd_rs_complete_io(mdev, sector);
4438 dec_rs_pending(mdev);
4442 /* let's advance progress step marks only for every other megabyte */
4443 if ((mdev->ov_left & 0x200) == 0x200)
4444 drbd_advance_rs_marks(mdev, mdev->ov_left);
4446 if (mdev->ov_left == 0) {
4447 w = kmalloc(sizeof(*w), GFP_NOIO);
4449 w->cb = w_ov_finished;
4450 drbd_queue_work_front(&mdev->data.work, w);
4452 dev_err(DEV, "kmalloc(w) failed.");
4454 drbd_resync_finished(mdev);
4461 static int got_skip(struct drbd_conf *mdev, struct p_header80 *h)
4466 struct asender_cmd {
4468 int (*process)(struct drbd_conf *mdev, struct p_header80 *h);
4471 static struct asender_cmd *get_asender_cmd(int cmd)
4473 static struct asender_cmd asender_tbl[] = {
4474 /* anything missing from this table is in
4475 * the drbd_cmd_handler (drbd_default_handler) table,
4476 * see the beginning of drbdd() */
4477 [P_PING] = { sizeof(struct p_header80), got_Ping },
4478 [P_PING_ACK] = { sizeof(struct p_header80), got_PingAck },
4479 [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4480 [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4481 [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4482 [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4483 [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck },
4484 [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply },
4485 [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply},
4486 [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult },
4487 [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck },
4488 [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply },
4489 [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync },
4490 [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip },
4491 [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply},
4492 [P_MAX_CMD] = { 0, NULL },
4494 if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL)
4496 return &asender_tbl[cmd];
4499 int drbd_asender(struct drbd_thread *thi)
4501 struct drbd_conf *mdev = thi->mdev;
4502 struct p_header80 *h = &mdev->meta.rbuf.header.h80;
4503 struct asender_cmd *cmd = NULL;
4508 int expect = sizeof(struct p_header80);
4510 int ping_timeout_active = 0;
4512 sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev));
4514 current->policy = SCHED_RR; /* Make this a realtime task! */
4515 current->rt_priority = 2; /* more important than all other tasks */
4517 while (get_t_state(thi) == Running) {
4518 drbd_thread_current_set_cpu(mdev);
4519 if (test_and_clear_bit(SEND_PING, &mdev->flags)) {
4520 ERR_IF(!drbd_send_ping(mdev)) goto reconnect;
4521 mdev->meta.socket->sk->sk_rcvtimeo =
4522 mdev->net_conf->ping_timeo*HZ/10;
4523 ping_timeout_active = 1;
4526 /* conditionally cork;
4527 * it may hurt latency if we cork without much to send */
4528 if (!mdev->net_conf->no_cork &&
4529 3 < atomic_read(&mdev->unacked_cnt))
4530 drbd_tcp_cork(mdev->meta.socket);
4532 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4533 flush_signals(current);
4534 if (!drbd_process_done_ee(mdev))
4536 /* to avoid race with newly queued ACKs */
4537 set_bit(SIGNAL_ASENDER, &mdev->flags);
4538 spin_lock_irq(&mdev->req_lock);
4539 empty = list_empty(&mdev->done_ee);
4540 spin_unlock_irq(&mdev->req_lock);
4541 /* new ack may have been queued right here,
4542 * but then there is also a signal pending,
4543 * and we start over... */
4547 /* but unconditionally uncork unless disabled */
4548 if (!mdev->net_conf->no_cork)
4549 drbd_tcp_uncork(mdev->meta.socket);
4551 /* short circuit, recv_msg would return EINTR anyways. */
4552 if (signal_pending(current))
4555 rv = drbd_recv_short(mdev, mdev->meta.socket,
4556 buf, expect-received, 0);
4557 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4559 flush_signals(current);
4562 * -EINTR (on meta) we got a signal
4563 * -EAGAIN (on meta) rcvtimeo expired
4564 * -ECONNRESET other side closed the connection
4565 * -ERESTARTSYS (on data) we got a signal
4566 * rv < 0 other than above: unexpected error!
4567 * rv == expected: full header or command
4568 * rv < expected: "woken" by signal during receive
4569 * rv == 0 : "connection shut down by peer"
4571 if (likely(rv > 0)) {
4574 } else if (rv == 0) {
4575 dev_err(DEV, "meta connection shut down by peer.\n");
4577 } else if (rv == -EAGAIN) {
4578 /* If the data socket received something meanwhile,
4579 * that is good enough: peer is still alive. */
4580 if (time_after(mdev->last_received,
4581 jiffies - mdev->meta.socket->sk->sk_rcvtimeo))
4583 if (ping_timeout_active) {
4584 dev_err(DEV, "PingAck did not arrive in time.\n");
4587 set_bit(SEND_PING, &mdev->flags);
4589 } else if (rv == -EINTR) {
4592 dev_err(DEV, "sock_recvmsg returned %d\n", rv);
4596 if (received == expect && cmd == NULL) {
4597 if (unlikely(h->magic != cpu_to_be32(DRBD_MAGIC))) {
4598 dev_err(DEV, "magic?? on meta m: 0x%08x c: %d l: %d\n",
4599 be32_to_cpu(h->magic),
4600 be16_to_cpu(h->command),
4601 be16_to_cpu(h->length));
4604 cmd = get_asender_cmd(be16_to_cpu(h->command));
4605 len = be16_to_cpu(h->length);
4606 if (unlikely(cmd == NULL)) {
4607 dev_err(DEV, "unknown command?? on meta m: 0x%08x c: %d l: %d\n",
4608 be32_to_cpu(h->magic),
4609 be16_to_cpu(h->command),
4610 be16_to_cpu(h->length));
4613 expect = cmd->pkt_size;
4614 ERR_IF(len != expect-sizeof(struct p_header80))
4617 if (received == expect) {
4618 mdev->last_received = jiffies;
4619 D_ASSERT(cmd != NULL);
4620 if (!cmd->process(mdev, h))
4623 /* the idle_timeout (ping-int)
4624 * has been restored in got_PingAck() */
4625 if (cmd == get_asender_cmd(P_PING_ACK))
4626 ping_timeout_active = 0;
4630 expect = sizeof(struct p_header80);
4637 drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE));
4642 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4645 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4647 D_ASSERT(mdev->state.conn < C_CONNECTED);
4648 dev_info(DEV, "asender terminated\n");