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 /* when we receive the answer for a read request,
1473 * verify that we actually know about it */
1474 static struct drbd_request *ar_id_to_req(struct drbd_conf *mdev, u64 id,
1477 struct hlist_head *slot = ar_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) {
1483 D_ASSERT(req->sector == sector);
1490 static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1492 struct drbd_request *req;
1495 struct p_data *p = &mdev->data.rbuf.data;
1497 sector = be64_to_cpu(p->sector);
1499 spin_lock_irq(&mdev->req_lock);
1500 req = ar_id_to_req(mdev, p->block_id, sector);
1501 spin_unlock_irq(&mdev->req_lock);
1502 if (unlikely(!req)) {
1503 dev_err(DEV, "Got a corrupt block_id/sector pair(1).\n");
1507 /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid
1508 * special casing it there for the various failure cases.
1509 * still no race with drbd_fail_pending_reads */
1510 ok = recv_dless_read(mdev, req, sector, data_size);
1513 req_mod(req, data_received);
1514 /* else: nothing. handled from drbd_disconnect...
1515 * I don't think we may complete this just yet
1516 * in case we are "on-disconnect: freeze" */
1521 static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1525 struct p_data *p = &mdev->data.rbuf.data;
1527 sector = be64_to_cpu(p->sector);
1528 D_ASSERT(p->block_id == ID_SYNCER);
1530 if (get_ldev(mdev)) {
1531 /* data is submitted to disk within recv_resync_read.
1532 * corresponding put_ldev done below on error,
1533 * or in drbd_endio_sec. */
1534 ok = recv_resync_read(mdev, sector, data_size);
1536 if (__ratelimit(&drbd_ratelimit_state))
1537 dev_err(DEV, "Can not write resync data to local disk.\n");
1539 ok = drbd_drain_block(mdev, data_size);
1541 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size);
1544 atomic_add(data_size >> 9, &mdev->rs_sect_in);
1549 /* e_end_block() is called via drbd_process_done_ee().
1550 * this means this function only runs in the asender thread
1552 static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1554 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1555 sector_t sector = e->sector;
1558 if (mdev->net_conf->wire_protocol == DRBD_PROT_C) {
1559 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1560 pcmd = (mdev->state.conn >= C_SYNC_SOURCE &&
1561 mdev->state.conn <= C_PAUSED_SYNC_T &&
1562 e->flags & EE_MAY_SET_IN_SYNC) ?
1563 P_RS_WRITE_ACK : P_WRITE_ACK;
1564 ok &= drbd_send_ack(mdev, pcmd, e);
1565 if (pcmd == P_RS_WRITE_ACK)
1566 drbd_set_in_sync(mdev, sector, e->size);
1568 ok = drbd_send_ack(mdev, P_NEG_ACK, e);
1569 /* we expect it to be marked out of sync anyways...
1570 * maybe assert this? */
1574 /* we delete from the conflict detection hash _after_ we sent out the
1575 * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */
1576 if (mdev->net_conf->two_primaries) {
1577 spin_lock_irq(&mdev->req_lock);
1578 D_ASSERT(!hlist_unhashed(&e->collision));
1579 hlist_del_init(&e->collision);
1580 spin_unlock_irq(&mdev->req_lock);
1582 D_ASSERT(hlist_unhashed(&e->collision));
1585 drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0));
1590 static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused)
1592 struct drbd_epoch_entry *e = (struct drbd_epoch_entry *)w;
1595 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1596 ok = drbd_send_ack(mdev, P_DISCARD_ACK, e);
1598 spin_lock_irq(&mdev->req_lock);
1599 D_ASSERT(!hlist_unhashed(&e->collision));
1600 hlist_del_init(&e->collision);
1601 spin_unlock_irq(&mdev->req_lock);
1608 /* Called from receive_Data.
1609 * Synchronize packets on sock with packets on msock.
1611 * This is here so even when a P_DATA packet traveling via sock overtook an Ack
1612 * packet traveling on msock, they are still processed in the order they have
1615 * Note: we don't care for Ack packets overtaking P_DATA packets.
1617 * In case packet_seq is larger than mdev->peer_seq number, there are
1618 * outstanding packets on the msock. We wait for them to arrive.
1619 * In case we are the logically next packet, we update mdev->peer_seq
1620 * ourselves. Correctly handles 32bit wrap around.
1622 * Assume we have a 10 GBit connection, that is about 1<<30 byte per second,
1623 * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds
1624 * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have
1625 * 1<<9 == 512 seconds aka ages for the 32bit wrap around...
1627 * returns 0 if we may process the packet,
1628 * -ERESTARTSYS if we were interrupted (by disconnect signal). */
1629 static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq)
1635 spin_lock(&mdev->peer_seq_lock);
1637 prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE);
1638 if (seq_le(packet_seq, mdev->peer_seq+1))
1640 if (signal_pending(current)) {
1644 p_seq = mdev->peer_seq;
1645 spin_unlock(&mdev->peer_seq_lock);
1646 timeout = schedule_timeout(30*HZ);
1647 spin_lock(&mdev->peer_seq_lock);
1648 if (timeout == 0 && p_seq == mdev->peer_seq) {
1650 dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n");
1654 finish_wait(&mdev->seq_wait, &wait);
1655 if (mdev->peer_seq+1 == packet_seq)
1657 spin_unlock(&mdev->peer_seq_lock);
1661 /* see also bio_flags_to_wire()
1662 * DRBD_REQ_*, because we need to semantically map the flags to data packet
1663 * flags and back. We may replicate to other kernel versions. */
1664 static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf)
1666 return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) |
1667 (dpf & DP_FUA ? REQ_FUA : 0) |
1668 (dpf & DP_FLUSH ? REQ_FLUSH : 0) |
1669 (dpf & DP_DISCARD ? REQ_DISCARD : 0);
1672 /* mirrored write */
1673 static int receive_Data(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
1676 struct drbd_epoch_entry *e;
1677 struct p_data *p = &mdev->data.rbuf.data;
1681 if (!get_ldev(mdev)) {
1682 spin_lock(&mdev->peer_seq_lock);
1683 if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num))
1685 spin_unlock(&mdev->peer_seq_lock);
1687 drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size);
1688 atomic_inc(&mdev->current_epoch->epoch_size);
1689 return drbd_drain_block(mdev, data_size);
1692 /* get_ldev(mdev) successful.
1693 * Corresponding put_ldev done either below (on various errors),
1694 * or in drbd_endio_sec, if we successfully submit the data at
1695 * the end of this function. */
1697 sector = be64_to_cpu(p->sector);
1698 e = read_in_block(mdev, p->block_id, sector, data_size);
1704 e->w.cb = e_end_block;
1706 dp_flags = be32_to_cpu(p->dp_flags);
1707 rw |= wire_flags_to_bio(mdev, dp_flags);
1709 if (dp_flags & DP_MAY_SET_IN_SYNC)
1710 e->flags |= EE_MAY_SET_IN_SYNC;
1712 spin_lock(&mdev->epoch_lock);
1713 e->epoch = mdev->current_epoch;
1714 atomic_inc(&e->epoch->epoch_size);
1715 atomic_inc(&e->epoch->active);
1716 spin_unlock(&mdev->epoch_lock);
1718 /* I'm the receiver, I do hold a net_cnt reference. */
1719 if (!mdev->net_conf->two_primaries) {
1720 spin_lock_irq(&mdev->req_lock);
1722 /* don't get the req_lock yet,
1723 * we may sleep in drbd_wait_peer_seq */
1724 const int size = e->size;
1725 const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags);
1727 struct drbd_request *i;
1728 struct hlist_node *n;
1729 struct hlist_head *slot;
1732 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
1733 BUG_ON(mdev->ee_hash == NULL);
1734 BUG_ON(mdev->tl_hash == NULL);
1736 /* conflict detection and handling:
1737 * 1. wait on the sequence number,
1738 * in case this data packet overtook ACK packets.
1739 * 2. check our hash tables for conflicting requests.
1740 * we only need to walk the tl_hash, since an ee can not
1741 * have a conflict with an other ee: on the submitting
1742 * node, the corresponding req had already been conflicting,
1743 * and a conflicting req is never sent.
1745 * Note: for two_primaries, we are protocol C,
1746 * so there cannot be any request that is DONE
1747 * but still on the transfer log.
1749 * unconditionally add to the ee_hash.
1751 * if no conflicting request is found:
1754 * if any conflicting request is found
1755 * that has not yet been acked,
1756 * AND I have the "discard concurrent writes" flag:
1757 * queue (via done_ee) the P_DISCARD_ACK; OUT.
1759 * if any conflicting request is found:
1760 * block the receiver, waiting on misc_wait
1761 * until no more conflicting requests are there,
1762 * or we get interrupted (disconnect).
1764 * we do not just write after local io completion of those
1765 * requests, but only after req is done completely, i.e.
1766 * we wait for the P_DISCARD_ACK to arrive!
1768 * then proceed normally, i.e. submit.
1770 if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num)))
1771 goto out_interrupted;
1773 spin_lock_irq(&mdev->req_lock);
1775 hlist_add_head(&e->collision, ee_hash_slot(mdev, sector));
1777 #define OVERLAPS overlaps(i->sector, i->size, sector, size)
1778 slot = tl_hash_slot(mdev, sector);
1781 int have_unacked = 0;
1782 int have_conflict = 0;
1783 prepare_to_wait(&mdev->misc_wait, &wait,
1784 TASK_INTERRUPTIBLE);
1785 hlist_for_each_entry(i, n, slot, collision) {
1787 /* only ALERT on first iteration,
1788 * we may be woken up early... */
1790 dev_alert(DEV, "%s[%u] Concurrent local write detected!"
1791 " new: %llus +%u; pending: %llus +%u\n",
1792 current->comm, current->pid,
1793 (unsigned long long)sector, size,
1794 (unsigned long long)i->sector, i->size);
1795 if (i->rq_state & RQ_NET_PENDING)
1804 /* Discard Ack only for the _first_ iteration */
1805 if (first && discard && have_unacked) {
1806 dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n",
1807 (unsigned long long)sector);
1809 e->w.cb = e_send_discard_ack;
1810 list_add_tail(&e->w.list, &mdev->done_ee);
1812 spin_unlock_irq(&mdev->req_lock);
1814 /* we could probably send that P_DISCARD_ACK ourselves,
1815 * but I don't like the receiver using the msock */
1819 finish_wait(&mdev->misc_wait, &wait);
1823 if (signal_pending(current)) {
1824 hlist_del_init(&e->collision);
1826 spin_unlock_irq(&mdev->req_lock);
1828 finish_wait(&mdev->misc_wait, &wait);
1829 goto out_interrupted;
1832 spin_unlock_irq(&mdev->req_lock);
1835 dev_alert(DEV, "Concurrent write! [W AFTERWARDS] "
1836 "sec=%llus\n", (unsigned long long)sector);
1837 } else if (discard) {
1838 /* we had none on the first iteration.
1839 * there must be none now. */
1840 D_ASSERT(have_unacked == 0);
1843 spin_lock_irq(&mdev->req_lock);
1845 finish_wait(&mdev->misc_wait, &wait);
1848 list_add(&e->w.list, &mdev->active_ee);
1849 spin_unlock_irq(&mdev->req_lock);
1851 switch (mdev->net_conf->wire_protocol) {
1854 /* corresponding dec_unacked() in e_end_block()
1855 * respective _drbd_clear_done_ee */
1858 /* I really don't like it that the receiver thread
1859 * sends on the msock, but anyways */
1860 drbd_send_ack(mdev, P_RECV_ACK, e);
1867 if (mdev->state.pdsk < D_INCONSISTENT) {
1868 /* In case we have the only disk of the cluster, */
1869 drbd_set_out_of_sync(mdev, e->sector, e->size);
1870 e->flags |= EE_CALL_AL_COMPLETE_IO;
1871 e->flags &= ~EE_MAY_SET_IN_SYNC;
1872 drbd_al_begin_io(mdev, e->sector);
1875 if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0)
1878 /* don't care for the reason here */
1879 dev_err(DEV, "submit failed, triggering re-connect\n");
1880 spin_lock_irq(&mdev->req_lock);
1881 list_del(&e->w.list);
1882 hlist_del_init(&e->collision);
1883 spin_unlock_irq(&mdev->req_lock);
1884 if (e->flags & EE_CALL_AL_COMPLETE_IO)
1885 drbd_al_complete_io(mdev, e->sector);
1888 drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + EV_CLEANUP);
1890 drbd_free_ee(mdev, e);
1894 /* We may throttle resync, if the lower device seems to be busy,
1895 * and current sync rate is above c_min_rate.
1897 * To decide whether or not the lower device is busy, we use a scheme similar
1898 * to MD RAID is_mddev_idle(): if the partition stats reveal "significant"
1899 * (more than 64 sectors) of activity we cannot account for with our own resync
1900 * activity, it obviously is "busy".
1902 * The current sync rate used here uses only the most recent two step marks,
1903 * to have a short time average so we can react faster.
1905 int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector)
1907 struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk;
1908 unsigned long db, dt, dbdt;
1909 struct lc_element *tmp;
1913 /* feature disabled? */
1914 if (mdev->sync_conf.c_min_rate == 0)
1917 spin_lock_irq(&mdev->al_lock);
1918 tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector));
1920 struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce);
1921 if (test_bit(BME_PRIORITY, &bm_ext->flags)) {
1922 spin_unlock_irq(&mdev->al_lock);
1925 /* Do not slow down if app IO is already waiting for this extent */
1927 spin_unlock_irq(&mdev->al_lock);
1929 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
1930 (int)part_stat_read(&disk->part0, sectors[1]) -
1931 atomic_read(&mdev->rs_sect_ev);
1933 if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) {
1934 unsigned long rs_left;
1937 mdev->rs_last_events = curr_events;
1939 /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP,
1941 i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS;
1943 if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T)
1944 rs_left = mdev->ov_left;
1946 rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed;
1948 dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ;
1951 db = mdev->rs_mark_left[i] - rs_left;
1952 dbdt = Bit2KB(db/dt);
1954 if (dbdt > mdev->sync_conf.c_min_rate)
1961 static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int digest_size)
1964 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
1965 struct drbd_epoch_entry *e;
1966 struct digest_info *di = NULL;
1968 unsigned int fault_type;
1969 struct p_block_req *p = &mdev->data.rbuf.block_req;
1971 sector = be64_to_cpu(p->sector);
1972 size = be32_to_cpu(p->blksize);
1974 if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) {
1975 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
1976 (unsigned long long)sector, size);
1979 if (sector + (size>>9) > capacity) {
1980 dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
1981 (unsigned long long)sector, size);
1985 if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) {
1988 case P_DATA_REQUEST:
1989 drbd_send_ack_rp(mdev, P_NEG_DREPLY, p);
1991 case P_RS_DATA_REQUEST:
1992 case P_CSUM_RS_REQUEST:
1994 drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p);
1998 dec_rs_pending(mdev);
1999 drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC);
2002 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n",
2005 if (verb && __ratelimit(&drbd_ratelimit_state))
2006 dev_err(DEV, "Can not satisfy peer's read request, "
2007 "no local data.\n");
2009 /* drain possibly payload */
2010 return drbd_drain_block(mdev, digest_size);
2013 /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD
2014 * "criss-cross" setup, that might cause write-out on some other DRBD,
2015 * which in turn might block on the other node at this very place. */
2016 e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO);
2023 case P_DATA_REQUEST:
2024 e->w.cb = w_e_end_data_req;
2025 fault_type = DRBD_FAULT_DT_RD;
2026 /* application IO, don't drbd_rs_begin_io */
2029 case P_RS_DATA_REQUEST:
2030 e->w.cb = w_e_end_rsdata_req;
2031 fault_type = DRBD_FAULT_RS_RD;
2032 /* used in the sector offset progress display */
2033 mdev->bm_resync_fo = BM_SECT_TO_BIT(sector);
2037 case P_CSUM_RS_REQUEST:
2038 fault_type = DRBD_FAULT_RS_RD;
2039 di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO);
2043 di->digest_size = digest_size;
2044 di->digest = (((char *)di)+sizeof(struct digest_info));
2047 e->flags |= EE_HAS_DIGEST;
2049 if (drbd_recv(mdev, di->digest, digest_size) != digest_size)
2052 if (cmd == P_CSUM_RS_REQUEST) {
2053 D_ASSERT(mdev->agreed_pro_version >= 89);
2054 e->w.cb = w_e_end_csum_rs_req;
2055 /* used in the sector offset progress display */
2056 mdev->bm_resync_fo = BM_SECT_TO_BIT(sector);
2057 } else if (cmd == P_OV_REPLY) {
2058 /* track progress, we may need to throttle */
2059 atomic_add(size >> 9, &mdev->rs_sect_in);
2060 e->w.cb = w_e_end_ov_reply;
2061 dec_rs_pending(mdev);
2062 /* drbd_rs_begin_io done when we sent this request,
2063 * but accounting still needs to be done. */
2064 goto submit_for_resync;
2069 if (mdev->ov_start_sector == ~(sector_t)0 &&
2070 mdev->agreed_pro_version >= 90) {
2071 unsigned long now = jiffies;
2073 mdev->ov_start_sector = sector;
2074 mdev->ov_position = sector;
2075 mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector);
2076 mdev->rs_total = mdev->ov_left;
2077 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2078 mdev->rs_mark_left[i] = mdev->ov_left;
2079 mdev->rs_mark_time[i] = now;
2081 dev_info(DEV, "Online Verify start sector: %llu\n",
2082 (unsigned long long)sector);
2084 e->w.cb = w_e_end_ov_req;
2085 fault_type = DRBD_FAULT_RS_RD;
2089 dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n",
2091 fault_type = DRBD_FAULT_MAX;
2095 /* Throttle, drbd_rs_begin_io and submit should become asynchronous
2096 * wrt the receiver, but it is not as straightforward as it may seem.
2097 * Various places in the resync start and stop logic assume resync
2098 * requests are processed in order, requeuing this on the worker thread
2099 * introduces a bunch of new code for synchronization between threads.
2101 * Unlimited throttling before drbd_rs_begin_io may stall the resync
2102 * "forever", throttling after drbd_rs_begin_io will lock that extent
2103 * for application writes for the same time. For now, just throttle
2104 * here, where the rest of the code expects the receiver to sleep for
2108 /* Throttle before drbd_rs_begin_io, as that locks out application IO;
2109 * this defers syncer requests for some time, before letting at least
2110 * on request through. The resync controller on the receiving side
2111 * will adapt to the incoming rate accordingly.
2113 * We cannot throttle here if remote is Primary/SyncTarget:
2114 * we would also throttle its application reads.
2115 * In that case, throttling is done on the SyncTarget only.
2117 if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector))
2118 schedule_timeout_uninterruptible(HZ/10);
2119 if (drbd_rs_begin_io(mdev, sector))
2123 atomic_add(size >> 9, &mdev->rs_sect_ev);
2127 spin_lock_irq(&mdev->req_lock);
2128 list_add_tail(&e->w.list, &mdev->read_ee);
2129 spin_unlock_irq(&mdev->req_lock);
2131 if (drbd_submit_ee(mdev, e, READ, fault_type) == 0)
2134 /* don't care for the reason here */
2135 dev_err(DEV, "submit failed, triggering re-connect\n");
2136 spin_lock_irq(&mdev->req_lock);
2137 list_del(&e->w.list);
2138 spin_unlock_irq(&mdev->req_lock);
2139 /* no drbd_rs_complete_io(), we are dropping the connection anyways */
2143 drbd_free_ee(mdev, e);
2147 static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local)
2149 int self, peer, rv = -100;
2150 unsigned long ch_self, ch_peer;
2152 self = mdev->ldev->md.uuid[UI_BITMAP] & 1;
2153 peer = mdev->p_uuid[UI_BITMAP] & 1;
2155 ch_peer = mdev->p_uuid[UI_SIZE];
2156 ch_self = mdev->comm_bm_set;
2158 switch (mdev->net_conf->after_sb_0p) {
2160 case ASB_DISCARD_SECONDARY:
2161 case ASB_CALL_HELPER:
2162 dev_err(DEV, "Configuration error.\n");
2164 case ASB_DISCONNECT:
2166 case ASB_DISCARD_YOUNGER_PRI:
2167 if (self == 0 && peer == 1) {
2171 if (self == 1 && peer == 0) {
2175 /* Else fall through to one of the other strategies... */
2176 case ASB_DISCARD_OLDER_PRI:
2177 if (self == 0 && peer == 1) {
2181 if (self == 1 && peer == 0) {
2185 /* Else fall through to one of the other strategies... */
2186 dev_warn(DEV, "Discard younger/older primary did not find a decision\n"
2187 "Using discard-least-changes instead\n");
2188 case ASB_DISCARD_ZERO_CHG:
2189 if (ch_peer == 0 && ch_self == 0) {
2190 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2194 if (ch_peer == 0) { rv = 1; break; }
2195 if (ch_self == 0) { rv = -1; break; }
2197 if (mdev->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG)
2199 case ASB_DISCARD_LEAST_CHG:
2200 if (ch_self < ch_peer)
2202 else if (ch_self > ch_peer)
2204 else /* ( ch_self == ch_peer ) */
2205 /* Well, then use something else. */
2206 rv = test_bit(DISCARD_CONCURRENT, &mdev->flags)
2209 case ASB_DISCARD_LOCAL:
2212 case ASB_DISCARD_REMOTE:
2219 static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local)
2223 switch (mdev->net_conf->after_sb_1p) {
2224 case ASB_DISCARD_YOUNGER_PRI:
2225 case ASB_DISCARD_OLDER_PRI:
2226 case ASB_DISCARD_LEAST_CHG:
2227 case ASB_DISCARD_LOCAL:
2228 case ASB_DISCARD_REMOTE:
2229 dev_err(DEV, "Configuration error.\n");
2231 case ASB_DISCONNECT:
2234 hg = drbd_asb_recover_0p(mdev);
2235 if (hg == -1 && mdev->state.role == R_SECONDARY)
2237 if (hg == 1 && mdev->state.role == R_PRIMARY)
2241 rv = drbd_asb_recover_0p(mdev);
2243 case ASB_DISCARD_SECONDARY:
2244 return mdev->state.role == R_PRIMARY ? 1 : -1;
2245 case ASB_CALL_HELPER:
2246 hg = drbd_asb_recover_0p(mdev);
2247 if (hg == -1 && mdev->state.role == R_PRIMARY) {
2248 enum drbd_state_rv rv2;
2250 drbd_set_role(mdev, R_SECONDARY, 0);
2251 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2252 * we might be here in C_WF_REPORT_PARAMS which is transient.
2253 * we do not need to wait for the after state change work either. */
2254 rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2255 if (rv2 != SS_SUCCESS) {
2256 drbd_khelper(mdev, "pri-lost-after-sb");
2258 dev_warn(DEV, "Successfully gave up primary role.\n");
2268 static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local)
2272 switch (mdev->net_conf->after_sb_2p) {
2273 case ASB_DISCARD_YOUNGER_PRI:
2274 case ASB_DISCARD_OLDER_PRI:
2275 case ASB_DISCARD_LEAST_CHG:
2276 case ASB_DISCARD_LOCAL:
2277 case ASB_DISCARD_REMOTE:
2279 case ASB_DISCARD_SECONDARY:
2280 dev_err(DEV, "Configuration error.\n");
2283 rv = drbd_asb_recover_0p(mdev);
2285 case ASB_DISCONNECT:
2287 case ASB_CALL_HELPER:
2288 hg = drbd_asb_recover_0p(mdev);
2290 enum drbd_state_rv rv2;
2292 /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE,
2293 * we might be here in C_WF_REPORT_PARAMS which is transient.
2294 * we do not need to wait for the after state change work either. */
2295 rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY));
2296 if (rv2 != SS_SUCCESS) {
2297 drbd_khelper(mdev, "pri-lost-after-sb");
2299 dev_warn(DEV, "Successfully gave up primary role.\n");
2309 static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid,
2310 u64 bits, u64 flags)
2313 dev_info(DEV, "%s uuid info vanished while I was looking!\n", text);
2316 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n",
2318 (unsigned long long)uuid[UI_CURRENT],
2319 (unsigned long long)uuid[UI_BITMAP],
2320 (unsigned long long)uuid[UI_HISTORY_START],
2321 (unsigned long long)uuid[UI_HISTORY_END],
2322 (unsigned long long)bits,
2323 (unsigned long long)flags);
2327 100 after split brain try auto recover
2328 2 C_SYNC_SOURCE set BitMap
2329 1 C_SYNC_SOURCE use BitMap
2331 -1 C_SYNC_TARGET use BitMap
2332 -2 C_SYNC_TARGET set BitMap
2333 -100 after split brain, disconnect
2334 -1000 unrelated data
2335 -1091 requires proto 91
2336 -1096 requires proto 96
2338 static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local)
2343 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2344 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2347 if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED)
2351 if ((self == UUID_JUST_CREATED || self == (u64)0) &&
2352 peer != UUID_JUST_CREATED)
2356 if (self != UUID_JUST_CREATED &&
2357 (peer == UUID_JUST_CREATED || peer == (u64)0))
2361 int rct, dc; /* roles at crash time */
2363 if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) {
2365 if (mdev->agreed_pro_version < 91)
2368 if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) &&
2369 (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) {
2370 dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n");
2371 drbd_uuid_set_bm(mdev, 0UL);
2373 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2374 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2377 dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n");
2384 if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) {
2386 if (mdev->agreed_pro_version < 91)
2389 if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) &&
2390 (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) {
2391 dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n");
2393 mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START];
2394 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP];
2395 mdev->p_uuid[UI_BITMAP] = 0UL;
2397 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2400 dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n");
2407 /* Common power [off|failure] */
2408 rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) +
2409 (mdev->p_uuid[UI_FLAGS] & 2);
2410 /* lowest bit is set when we were primary,
2411 * next bit (weight 2) is set when peer was primary */
2415 case 0: /* !self_pri && !peer_pri */ return 0;
2416 case 1: /* self_pri && !peer_pri */ return 1;
2417 case 2: /* !self_pri && peer_pri */ return -1;
2418 case 3: /* self_pri && peer_pri */
2419 dc = test_bit(DISCARD_CONCURRENT, &mdev->flags);
2425 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2430 peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1);
2432 if (mdev->agreed_pro_version < 96 ?
2433 (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) ==
2434 (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) :
2435 peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) {
2436 /* The last P_SYNC_UUID did not get though. Undo the last start of
2437 resync as sync source modifications of the peer's UUIDs. */
2439 if (mdev->agreed_pro_version < 91)
2442 mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START];
2443 mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1];
2445 dev_info(DEV, "Did not got last syncUUID packet, corrected:\n");
2446 drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2453 self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1);
2454 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2455 peer = mdev->p_uuid[i] & ~((u64)1);
2461 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2462 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2467 self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1);
2469 if (mdev->agreed_pro_version < 96 ?
2470 (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) ==
2471 (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) :
2472 self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) {
2473 /* The last P_SYNC_UUID did not get though. Undo the last start of
2474 resync as sync source modifications of our UUIDs. */
2476 if (mdev->agreed_pro_version < 91)
2479 _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]);
2480 _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]);
2482 dev_info(DEV, "Last syncUUID did not get through, corrected:\n");
2483 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid,
2484 mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0);
2492 peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1);
2493 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2494 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2500 self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1);
2501 peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1);
2502 if (self == peer && self != ((u64)0))
2506 for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) {
2507 self = mdev->ldev->md.uuid[i] & ~((u64)1);
2508 for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) {
2509 peer = mdev->p_uuid[j] & ~((u64)1);
2518 /* drbd_sync_handshake() returns the new conn state on success, or
2519 CONN_MASK (-1) on failure.
2521 static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role,
2522 enum drbd_disk_state peer_disk) __must_hold(local)
2525 enum drbd_conns rv = C_MASK;
2526 enum drbd_disk_state mydisk;
2528 mydisk = mdev->state.disk;
2529 if (mydisk == D_NEGOTIATING)
2530 mydisk = mdev->new_state_tmp.disk;
2532 dev_info(DEV, "drbd_sync_handshake:\n");
2533 drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0);
2534 drbd_uuid_dump(mdev, "peer", mdev->p_uuid,
2535 mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]);
2537 hg = drbd_uuid_compare(mdev, &rule_nr);
2539 dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr);
2542 dev_alert(DEV, "Unrelated data, aborting!\n");
2546 dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000);
2550 if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) ||
2551 (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) {
2552 int f = (hg == -100) || abs(hg) == 2;
2553 hg = mydisk > D_INCONSISTENT ? 1 : -1;
2556 dev_info(DEV, "Becoming sync %s due to disk states.\n",
2557 hg > 0 ? "source" : "target");
2561 drbd_khelper(mdev, "initial-split-brain");
2563 if (hg == 100 || (hg == -100 && mdev->net_conf->always_asbp)) {
2564 int pcount = (mdev->state.role == R_PRIMARY)
2565 + (peer_role == R_PRIMARY);
2566 int forced = (hg == -100);
2570 hg = drbd_asb_recover_0p(mdev);
2573 hg = drbd_asb_recover_1p(mdev);
2576 hg = drbd_asb_recover_2p(mdev);
2579 if (abs(hg) < 100) {
2580 dev_warn(DEV, "Split-Brain detected, %d primaries, "
2581 "automatically solved. Sync from %s node\n",
2582 pcount, (hg < 0) ? "peer" : "this");
2584 dev_warn(DEV, "Doing a full sync, since"
2585 " UUIDs where ambiguous.\n");
2592 if (mdev->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1))
2594 if (!mdev->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1))
2598 dev_warn(DEV, "Split-Brain detected, manually solved. "
2599 "Sync from %s node\n",
2600 (hg < 0) ? "peer" : "this");
2604 /* FIXME this log message is not correct if we end up here
2605 * after an attempted attach on a diskless node.
2606 * We just refuse to attach -- well, we drop the "connection"
2607 * to that disk, in a way... */
2608 dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n");
2609 drbd_khelper(mdev, "split-brain");
2613 if (hg > 0 && mydisk <= D_INCONSISTENT) {
2614 dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n");
2618 if (hg < 0 && /* by intention we do not use mydisk here. */
2619 mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) {
2620 switch (mdev->net_conf->rr_conflict) {
2621 case ASB_CALL_HELPER:
2622 drbd_khelper(mdev, "pri-lost");
2624 case ASB_DISCONNECT:
2625 dev_err(DEV, "I shall become SyncTarget, but I am primary!\n");
2628 dev_warn(DEV, "Becoming SyncTarget, violating the stable-data"
2633 if (mdev->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) {
2635 dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n");
2637 dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.",
2638 drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET),
2639 abs(hg) >= 2 ? "full" : "bit-map based");
2644 dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n");
2645 if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake",
2646 BM_LOCKED_SET_ALLOWED))
2650 if (hg > 0) { /* become sync source. */
2652 } else if (hg < 0) { /* become sync target */
2656 if (drbd_bm_total_weight(mdev)) {
2657 dev_info(DEV, "No resync, but %lu bits in bitmap!\n",
2658 drbd_bm_total_weight(mdev));
2665 /* returns 1 if invalid */
2666 static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self)
2668 /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */
2669 if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) ||
2670 (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL))
2673 /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */
2674 if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL ||
2675 self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL)
2678 /* everything else is valid if they are equal on both sides. */
2682 /* everything es is invalid. */
2686 static int receive_protocol(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
2688 struct p_protocol *p = &mdev->data.rbuf.protocol;
2689 int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p;
2690 int p_want_lose, p_two_primaries, cf;
2691 char p_integrity_alg[SHARED_SECRET_MAX] = "";
2693 p_proto = be32_to_cpu(p->protocol);
2694 p_after_sb_0p = be32_to_cpu(p->after_sb_0p);
2695 p_after_sb_1p = be32_to_cpu(p->after_sb_1p);
2696 p_after_sb_2p = be32_to_cpu(p->after_sb_2p);
2697 p_two_primaries = be32_to_cpu(p->two_primaries);
2698 cf = be32_to_cpu(p->conn_flags);
2699 p_want_lose = cf & CF_WANT_LOSE;
2701 clear_bit(CONN_DRY_RUN, &mdev->flags);
2703 if (cf & CF_DRY_RUN)
2704 set_bit(CONN_DRY_RUN, &mdev->flags);
2706 if (p_proto != mdev->net_conf->wire_protocol) {
2707 dev_err(DEV, "incompatible communication protocols\n");
2711 if (cmp_after_sb(p_after_sb_0p, mdev->net_conf->after_sb_0p)) {
2712 dev_err(DEV, "incompatible after-sb-0pri settings\n");
2716 if (cmp_after_sb(p_after_sb_1p, mdev->net_conf->after_sb_1p)) {
2717 dev_err(DEV, "incompatible after-sb-1pri settings\n");
2721 if (cmp_after_sb(p_after_sb_2p, mdev->net_conf->after_sb_2p)) {
2722 dev_err(DEV, "incompatible after-sb-2pri settings\n");
2726 if (p_want_lose && mdev->net_conf->want_lose) {
2727 dev_err(DEV, "both sides have the 'want_lose' flag set\n");
2731 if (p_two_primaries != mdev->net_conf->two_primaries) {
2732 dev_err(DEV, "incompatible setting of the two-primaries options\n");
2736 if (mdev->agreed_pro_version >= 87) {
2737 unsigned char *my_alg = mdev->net_conf->integrity_alg;
2739 if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size)
2742 p_integrity_alg[SHARED_SECRET_MAX-1] = 0;
2743 if (strcmp(p_integrity_alg, my_alg)) {
2744 dev_err(DEV, "incompatible setting of the data-integrity-alg\n");
2747 dev_info(DEV, "data-integrity-alg: %s\n",
2748 my_alg[0] ? my_alg : (unsigned char *)"<not-used>");
2754 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2759 * input: alg name, feature name
2760 * return: NULL (alg name was "")
2761 * ERR_PTR(error) if something goes wrong
2762 * or the crypto hash ptr, if it worked out ok. */
2763 struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev,
2764 const char *alg, const char *name)
2766 struct crypto_hash *tfm;
2771 tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC);
2773 dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n",
2774 alg, name, PTR_ERR(tfm));
2777 if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) {
2778 crypto_free_hash(tfm);
2779 dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name);
2780 return ERR_PTR(-EINVAL);
2785 static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int packet_size)
2788 struct p_rs_param_95 *p = &mdev->data.rbuf.rs_param_95;
2789 unsigned int header_size, data_size, exp_max_sz;
2790 struct crypto_hash *verify_tfm = NULL;
2791 struct crypto_hash *csums_tfm = NULL;
2792 const int apv = mdev->agreed_pro_version;
2793 int *rs_plan_s = NULL;
2796 exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param)
2797 : apv == 88 ? sizeof(struct p_rs_param)
2799 : apv <= 94 ? sizeof(struct p_rs_param_89)
2800 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
2802 if (packet_size > exp_max_sz) {
2803 dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n",
2804 packet_size, exp_max_sz);
2809 header_size = sizeof(struct p_rs_param) - sizeof(struct p_header80);
2810 data_size = packet_size - header_size;
2811 } else if (apv <= 94) {
2812 header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header80);
2813 data_size = packet_size - header_size;
2814 D_ASSERT(data_size == 0);
2816 header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header80);
2817 data_size = packet_size - header_size;
2818 D_ASSERT(data_size == 0);
2821 /* initialize verify_alg and csums_alg */
2822 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
2824 if (drbd_recv(mdev, &p->head.payload, header_size) != header_size)
2827 mdev->sync_conf.rate = be32_to_cpu(p->rate);
2831 if (data_size > SHARED_SECRET_MAX) {
2832 dev_err(DEV, "verify-alg too long, "
2833 "peer wants %u, accepting only %u byte\n",
2834 data_size, SHARED_SECRET_MAX);
2838 if (drbd_recv(mdev, p->verify_alg, data_size) != data_size)
2841 /* we expect NUL terminated string */
2842 /* but just in case someone tries to be evil */
2843 D_ASSERT(p->verify_alg[data_size-1] == 0);
2844 p->verify_alg[data_size-1] = 0;
2846 } else /* apv >= 89 */ {
2847 /* we still expect NUL terminated strings */
2848 /* but just in case someone tries to be evil */
2849 D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0);
2850 D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0);
2851 p->verify_alg[SHARED_SECRET_MAX-1] = 0;
2852 p->csums_alg[SHARED_SECRET_MAX-1] = 0;
2855 if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) {
2856 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2857 dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n",
2858 mdev->sync_conf.verify_alg, p->verify_alg);
2861 verify_tfm = drbd_crypto_alloc_digest_safe(mdev,
2862 p->verify_alg, "verify-alg");
2863 if (IS_ERR(verify_tfm)) {
2869 if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) {
2870 if (mdev->state.conn == C_WF_REPORT_PARAMS) {
2871 dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n",
2872 mdev->sync_conf.csums_alg, p->csums_alg);
2875 csums_tfm = drbd_crypto_alloc_digest_safe(mdev,
2876 p->csums_alg, "csums-alg");
2877 if (IS_ERR(csums_tfm)) {
2884 mdev->sync_conf.rate = be32_to_cpu(p->rate);
2885 mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead);
2886 mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target);
2887 mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target);
2888 mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate);
2890 fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ;
2891 if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) {
2892 rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL);
2894 dev_err(DEV, "kmalloc of fifo_buffer failed");
2900 spin_lock(&mdev->peer_seq_lock);
2901 /* lock against drbd_nl_syncer_conf() */
2903 strcpy(mdev->sync_conf.verify_alg, p->verify_alg);
2904 mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1;
2905 crypto_free_hash(mdev->verify_tfm);
2906 mdev->verify_tfm = verify_tfm;
2907 dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg);
2910 strcpy(mdev->sync_conf.csums_alg, p->csums_alg);
2911 mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1;
2912 crypto_free_hash(mdev->csums_tfm);
2913 mdev->csums_tfm = csums_tfm;
2914 dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg);
2916 if (fifo_size != mdev->rs_plan_s.size) {
2917 kfree(mdev->rs_plan_s.values);
2918 mdev->rs_plan_s.values = rs_plan_s;
2919 mdev->rs_plan_s.size = fifo_size;
2920 mdev->rs_planed = 0;
2922 spin_unlock(&mdev->peer_seq_lock);
2927 /* just for completeness: actually not needed,
2928 * as this is not reached if csums_tfm was ok. */
2929 crypto_free_hash(csums_tfm);
2930 /* but free the verify_tfm again, if csums_tfm did not work out */
2931 crypto_free_hash(verify_tfm);
2932 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2936 /* warn if the arguments differ by more than 12.5% */
2937 static void warn_if_differ_considerably(struct drbd_conf *mdev,
2938 const char *s, sector_t a, sector_t b)
2941 if (a == 0 || b == 0)
2943 d = (a > b) ? (a - b) : (b - a);
2944 if (d > (a>>3) || d > (b>>3))
2945 dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s,
2946 (unsigned long long)a, (unsigned long long)b);
2949 static int receive_sizes(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
2951 struct p_sizes *p = &mdev->data.rbuf.sizes;
2952 enum determine_dev_size dd = unchanged;
2953 sector_t p_size, p_usize, my_usize;
2954 int ldsc = 0; /* local disk size changed */
2955 enum dds_flags ddsf;
2957 p_size = be64_to_cpu(p->d_size);
2958 p_usize = be64_to_cpu(p->u_size);
2960 if (p_size == 0 && mdev->state.disk == D_DISKLESS) {
2961 dev_err(DEV, "some backing storage is needed\n");
2962 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2966 /* just store the peer's disk size for now.
2967 * we still need to figure out whether we accept that. */
2968 mdev->p_size = p_size;
2970 if (get_ldev(mdev)) {
2971 warn_if_differ_considerably(mdev, "lower level device sizes",
2972 p_size, drbd_get_max_capacity(mdev->ldev));
2973 warn_if_differ_considerably(mdev, "user requested size",
2974 p_usize, mdev->ldev->dc.disk_size);
2976 /* if this is the first connect, or an otherwise expected
2977 * param exchange, choose the minimum */
2978 if (mdev->state.conn == C_WF_REPORT_PARAMS)
2979 p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size,
2982 my_usize = mdev->ldev->dc.disk_size;
2984 if (mdev->ldev->dc.disk_size != p_usize) {
2985 mdev->ldev->dc.disk_size = p_usize;
2986 dev_info(DEV, "Peer sets u_size to %lu sectors\n",
2987 (unsigned long)mdev->ldev->dc.disk_size);
2990 /* Never shrink a device with usable data during connect.
2991 But allow online shrinking if we are connected. */
2992 if (drbd_new_dev_size(mdev, mdev->ldev, 0) <
2993 drbd_get_capacity(mdev->this_bdev) &&
2994 mdev->state.disk >= D_OUTDATED &&
2995 mdev->state.conn < C_CONNECTED) {
2996 dev_err(DEV, "The peer's disk size is too small!\n");
2997 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
2998 mdev->ldev->dc.disk_size = my_usize;
3005 ddsf = be16_to_cpu(p->dds_flags);
3006 if (get_ldev(mdev)) {
3007 dd = drbd_determine_dev_size(mdev, ddsf);
3009 if (dd == dev_size_error)
3013 /* I am diskless, need to accept the peer's size. */
3014 drbd_set_my_capacity(mdev, p_size);
3017 mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size);
3018 drbd_reconsider_max_bio_size(mdev);
3020 if (get_ldev(mdev)) {
3021 if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) {
3022 mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev);
3029 if (mdev->state.conn > C_WF_REPORT_PARAMS) {
3030 if (be64_to_cpu(p->c_size) !=
3031 drbd_get_capacity(mdev->this_bdev) || ldsc) {
3032 /* we have different sizes, probably peer
3033 * needs to know my new size... */
3034 drbd_send_sizes(mdev, 0, ddsf);
3036 if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) ||
3037 (dd == grew && mdev->state.conn == C_CONNECTED)) {
3038 if (mdev->state.pdsk >= D_INCONSISTENT &&
3039 mdev->state.disk >= D_INCONSISTENT) {
3040 if (ddsf & DDSF_NO_RESYNC)
3041 dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n");
3043 resync_after_online_grow(mdev);
3045 set_bit(RESYNC_AFTER_NEG, &mdev->flags);
3052 static int receive_uuids(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3054 struct p_uuids *p = &mdev->data.rbuf.uuids;
3056 int i, updated_uuids = 0;
3058 p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO);
3060 for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++)
3061 p_uuid[i] = be64_to_cpu(p->uuid[i]);
3063 kfree(mdev->p_uuid);
3064 mdev->p_uuid = p_uuid;
3066 if (mdev->state.conn < C_CONNECTED &&
3067 mdev->state.disk < D_INCONSISTENT &&
3068 mdev->state.role == R_PRIMARY &&
3069 (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) {
3070 dev_err(DEV, "Can only connect to data with current UUID=%016llX\n",
3071 (unsigned long long)mdev->ed_uuid);
3072 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3076 if (get_ldev(mdev)) {
3077 int skip_initial_sync =
3078 mdev->state.conn == C_CONNECTED &&
3079 mdev->agreed_pro_version >= 90 &&
3080 mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
3081 (p_uuid[UI_FLAGS] & 8);
3082 if (skip_initial_sync) {
3083 dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n");
3084 drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write,
3085 "clear_n_write from receive_uuids",
3086 BM_LOCKED_TEST_ALLOWED);
3087 _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]);
3088 _drbd_uuid_set(mdev, UI_BITMAP, 0);
3089 _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
3095 } else if (mdev->state.disk < D_INCONSISTENT &&
3096 mdev->state.role == R_PRIMARY) {
3097 /* I am a diskless primary, the peer just created a new current UUID
3099 updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3102 /* Before we test for the disk state, we should wait until an eventually
3103 ongoing cluster wide state change is finished. That is important if
3104 we are primary and are detaching from our disk. We need to see the
3105 new disk state... */
3106 wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags));
3107 if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT)
3108 updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]);
3111 drbd_print_uuids(mdev, "receiver updated UUIDs to");
3117 * convert_state() - Converts the peer's view of the cluster state to our point of view
3118 * @ps: The state as seen by the peer.
3120 static union drbd_state convert_state(union drbd_state ps)
3122 union drbd_state ms;
3124 static enum drbd_conns c_tab[] = {
3125 [C_CONNECTED] = C_CONNECTED,
3127 [C_STARTING_SYNC_S] = C_STARTING_SYNC_T,
3128 [C_STARTING_SYNC_T] = C_STARTING_SYNC_S,
3129 [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */
3130 [C_VERIFY_S] = C_VERIFY_T,
3136 ms.conn = c_tab[ps.conn];
3141 ms.peer_isp = (ps.aftr_isp | ps.user_isp);
3146 static int receive_req_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3148 struct p_req_state *p = &mdev->data.rbuf.req_state;
3149 union drbd_state mask, val;
3150 enum drbd_state_rv rv;
3152 mask.i = be32_to_cpu(p->mask);
3153 val.i = be32_to_cpu(p->val);
3155 if (test_bit(DISCARD_CONCURRENT, &mdev->flags) &&
3156 test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) {
3157 drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG);
3161 mask = convert_state(mask);
3162 val = convert_state(val);
3164 rv = drbd_change_state(mdev, CS_VERBOSE, mask, val);
3166 drbd_send_sr_reply(mdev, rv);
3172 static int receive_state(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3174 struct p_state *p = &mdev->data.rbuf.state;
3175 union drbd_state os, ns, peer_state;
3176 enum drbd_disk_state real_peer_disk;
3177 enum chg_state_flags cs_flags;
3180 peer_state.i = be32_to_cpu(p->state);
3182 real_peer_disk = peer_state.disk;
3183 if (peer_state.disk == D_NEGOTIATING) {
3184 real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT;
3185 dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk));
3188 spin_lock_irq(&mdev->req_lock);
3190 os = ns = mdev->state;
3191 spin_unlock_irq(&mdev->req_lock);
3193 /* peer says his disk is uptodate, while we think it is inconsistent,
3194 * and this happens while we think we have a sync going on. */
3195 if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE &&
3196 os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) {
3197 /* If we are (becoming) SyncSource, but peer is still in sync
3198 * preparation, ignore its uptodate-ness to avoid flapping, it
3199 * will change to inconsistent once the peer reaches active
3201 * It may have changed syncer-paused flags, however, so we
3202 * cannot ignore this completely. */
3203 if (peer_state.conn > C_CONNECTED &&
3204 peer_state.conn < C_SYNC_SOURCE)
3205 real_peer_disk = D_INCONSISTENT;
3207 /* if peer_state changes to connected at the same time,
3208 * it explicitly notifies us that it finished resync.
3209 * Maybe we should finish it up, too? */
3210 else if (os.conn >= C_SYNC_SOURCE &&
3211 peer_state.conn == C_CONNECTED) {
3212 if (drbd_bm_total_weight(mdev) <= mdev->rs_failed)
3213 drbd_resync_finished(mdev);
3218 /* peer says his disk is inconsistent, while we think it is uptodate,
3219 * and this happens while the peer still thinks we have a sync going on,
3220 * but we think we are already done with the sync.
3221 * We ignore this to avoid flapping pdsk.
3222 * This should not happen, if the peer is a recent version of drbd. */
3223 if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT &&
3224 os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE)
3225 real_peer_disk = D_UP_TO_DATE;
3227 if (ns.conn == C_WF_REPORT_PARAMS)
3228 ns.conn = C_CONNECTED;
3230 if (peer_state.conn == C_AHEAD)
3233 if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING &&
3234 get_ldev_if_state(mdev, D_NEGOTIATING)) {
3235 int cr; /* consider resync */
3237 /* if we established a new connection */
3238 cr = (os.conn < C_CONNECTED);
3239 /* if we had an established connection
3240 * and one of the nodes newly attaches a disk */
3241 cr |= (os.conn == C_CONNECTED &&
3242 (peer_state.disk == D_NEGOTIATING ||
3243 os.disk == D_NEGOTIATING));
3244 /* if we have both been inconsistent, and the peer has been
3245 * forced to be UpToDate with --overwrite-data */
3246 cr |= test_bit(CONSIDER_RESYNC, &mdev->flags);
3247 /* if we had been plain connected, and the admin requested to
3248 * start a sync by "invalidate" or "invalidate-remote" */
3249 cr |= (os.conn == C_CONNECTED &&
3250 (peer_state.conn >= C_STARTING_SYNC_S &&
3251 peer_state.conn <= C_WF_BITMAP_T));
3254 ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk);
3257 if (ns.conn == C_MASK) {
3258 ns.conn = C_CONNECTED;
3259 if (mdev->state.disk == D_NEGOTIATING) {
3260 drbd_force_state(mdev, NS(disk, D_FAILED));
3261 } else if (peer_state.disk == D_NEGOTIATING) {
3262 dev_err(DEV, "Disk attach process on the peer node was aborted.\n");
3263 peer_state.disk = D_DISKLESS;
3264 real_peer_disk = D_DISKLESS;
3266 if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags))
3268 D_ASSERT(os.conn == C_WF_REPORT_PARAMS);
3269 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3275 spin_lock_irq(&mdev->req_lock);
3276 if (mdev->state.i != os.i)
3278 clear_bit(CONSIDER_RESYNC, &mdev->flags);
3279 ns.peer = peer_state.role;
3280 ns.pdsk = real_peer_disk;
3281 ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp);
3282 if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING)
3283 ns.disk = mdev->new_state_tmp.disk;
3284 cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD);
3285 if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED &&
3286 test_bit(NEW_CUR_UUID, &mdev->flags)) {
3287 /* Do not allow tl_restart(resend) for a rebooted peer. We can only allow this
3288 for temporal network outages! */
3289 spin_unlock_irq(&mdev->req_lock);
3290 dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n");
3292 drbd_uuid_new_current(mdev);
3293 clear_bit(NEW_CUR_UUID, &mdev->flags);
3294 drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0));
3297 rv = _drbd_set_state(mdev, ns, cs_flags, NULL);
3299 spin_unlock_irq(&mdev->req_lock);
3301 if (rv < SS_SUCCESS) {
3302 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
3306 if (os.conn > C_WF_REPORT_PARAMS) {
3307 if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED &&
3308 peer_state.disk != D_NEGOTIATING ) {
3309 /* we want resync, peer has not yet decided to sync... */
3310 /* Nowadays only used when forcing a node into primary role and
3311 setting its disk to UpToDate with that */
3312 drbd_send_uuids(mdev);
3313 drbd_send_state(mdev);
3317 mdev->net_conf->want_lose = 0;
3319 drbd_md_sync(mdev); /* update connected indicator, la_size, ... */
3324 static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3326 struct p_rs_uuid *p = &mdev->data.rbuf.rs_uuid;
3328 wait_event(mdev->misc_wait,
3329 mdev->state.conn == C_WF_SYNC_UUID ||
3330 mdev->state.conn == C_BEHIND ||
3331 mdev->state.conn < C_CONNECTED ||
3332 mdev->state.disk < D_NEGOTIATING);
3334 /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */
3336 /* Here the _drbd_uuid_ functions are right, current should
3337 _not_ be rotated into the history */
3338 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
3339 _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid));
3340 _drbd_uuid_set(mdev, UI_BITMAP, 0UL);
3342 drbd_print_uuids(mdev, "updated sync uuid");
3343 drbd_start_resync(mdev, C_SYNC_TARGET);
3347 dev_err(DEV, "Ignoring SyncUUID packet!\n");
3353 * receive_bitmap_plain
3355 * Return 0 when done, 1 when another iteration is needed, and a negative error
3356 * code upon failure.
3359 receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size,
3360 unsigned long *buffer, struct bm_xfer_ctx *c)
3362 unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset);
3363 unsigned want = num_words * sizeof(long);
3366 if (want != data_size) {
3367 dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size);
3372 err = drbd_recv(mdev, buffer, want);
3379 drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer);
3381 c->word_offset += num_words;
3382 c->bit_offset = c->word_offset * BITS_PER_LONG;
3383 if (c->bit_offset > c->bm_bits)
3384 c->bit_offset = c->bm_bits;
3392 * Return 0 when done, 1 when another iteration is needed, and a negative error
3393 * code upon failure.
3396 recv_bm_rle_bits(struct drbd_conf *mdev,
3397 struct p_compressed_bm *p,
3398 struct bm_xfer_ctx *c)
3400 struct bitstream bs;
3404 unsigned long s = c->bit_offset;
3406 int len = be16_to_cpu(p->head.length) - (sizeof(*p) - sizeof(p->head));
3407 int toggle = DCBP_get_start(p);
3411 bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p));
3413 bits = bitstream_get_bits(&bs, &look_ahead, 64);
3417 for (have = bits; have > 0; s += rl, toggle = !toggle) {
3418 bits = vli_decode_bits(&rl, look_ahead);
3424 if (e >= c->bm_bits) {
3425 dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e);
3428 _drbd_bm_set_bits(mdev, s, e);
3432 dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n",
3433 have, bits, look_ahead,
3434 (unsigned int)(bs.cur.b - p->code),
3435 (unsigned int)bs.buf_len);
3438 look_ahead >>= bits;
3441 bits = bitstream_get_bits(&bs, &tmp, 64 - have);
3444 look_ahead |= tmp << have;
3449 bm_xfer_ctx_bit_to_word_offset(c);
3451 return (s != c->bm_bits);
3457 * Return 0 when done, 1 when another iteration is needed, and a negative error
3458 * code upon failure.
3461 decode_bitmap_c(struct drbd_conf *mdev,
3462 struct p_compressed_bm *p,
3463 struct bm_xfer_ctx *c)
3465 if (DCBP_get_code(p) == RLE_VLI_Bits)
3466 return recv_bm_rle_bits(mdev, p, c);
3468 /* other variants had been implemented for evaluation,
3469 * but have been dropped as this one turned out to be "best"
3470 * during all our tests. */
3472 dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding);
3473 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3477 void INFO_bm_xfer_stats(struct drbd_conf *mdev,
3478 const char *direction, struct bm_xfer_ctx *c)
3480 /* what would it take to transfer it "plaintext" */
3481 unsigned plain = sizeof(struct p_header80) *
3482 ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1)
3483 + c->bm_words * sizeof(long);
3484 unsigned total = c->bytes[0] + c->bytes[1];
3487 /* total can not be zero. but just in case: */
3491 /* don't report if not compressed */
3495 /* total < plain. check for overflow, still */
3496 r = (total > UINT_MAX/1000) ? (total / (plain/1000))
3497 : (1000 * total / plain);
3503 dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), "
3504 "total %u; compression: %u.%u%%\n",
3506 c->bytes[1], c->packets[1],
3507 c->bytes[0], c->packets[0],
3508 total, r/10, r % 10);
3511 /* Since we are processing the bitfield from lower addresses to higher,
3512 it does not matter if the process it in 32 bit chunks or 64 bit
3513 chunks as long as it is little endian. (Understand it as byte stream,
3514 beginning with the lowest byte...) If we would use big endian
3515 we would need to process it from the highest address to the lowest,
3516 in order to be agnostic to the 32 vs 64 bits issue.
3518 returns 0 on failure, 1 if we successfully received it. */
3519 static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3521 struct bm_xfer_ctx c;
3525 struct p_header80 *h = &mdev->data.rbuf.header.h80;
3527 drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED);
3528 /* you are supposed to send additional out-of-sync information
3529 * if you actually set bits during this phase */
3531 /* maybe we should use some per thread scratch page,
3532 * and allocate that during initial device creation? */
3533 buffer = (unsigned long *) __get_free_page(GFP_NOIO);
3535 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__);
3539 c = (struct bm_xfer_ctx) {
3540 .bm_bits = drbd_bm_bits(mdev),
3541 .bm_words = drbd_bm_words(mdev),
3545 if (cmd == P_BITMAP) {
3546 err = receive_bitmap_plain(mdev, data_size, buffer, &c);
3547 } else if (cmd == P_COMPRESSED_BITMAP) {
3548 /* MAYBE: sanity check that we speak proto >= 90,
3549 * and the feature is enabled! */
3550 struct p_compressed_bm *p;
3552 if (data_size > BM_PACKET_PAYLOAD_BYTES) {
3553 dev_err(DEV, "ReportCBitmap packet too large\n");
3556 /* use the page buff */
3558 memcpy(p, h, sizeof(*h));
3559 if (drbd_recv(mdev, p->head.payload, data_size) != data_size)
3561 if (data_size <= (sizeof(*p) - sizeof(p->head))) {
3562 dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size);
3565 err = decode_bitmap_c(mdev, p, &c);
3567 dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd);
3571 c.packets[cmd == P_BITMAP]++;
3572 c.bytes[cmd == P_BITMAP] += sizeof(struct p_header80) + data_size;
3579 if (!drbd_recv_header(mdev, &cmd, &data_size))
3583 INFO_bm_xfer_stats(mdev, "receive", &c);
3585 if (mdev->state.conn == C_WF_BITMAP_T) {
3586 enum drbd_state_rv rv;
3588 ok = !drbd_send_bitmap(mdev);
3591 /* Omit CS_ORDERED with this state transition to avoid deadlocks. */
3592 rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE);
3593 D_ASSERT(rv == SS_SUCCESS);
3594 } else if (mdev->state.conn != C_WF_BITMAP_S) {
3595 /* admin may have requested C_DISCONNECTING,
3596 * other threads may have noticed network errors */
3597 dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n",
3598 drbd_conn_str(mdev->state.conn));
3603 drbd_bm_unlock(mdev);
3604 if (ok && mdev->state.conn == C_WF_BITMAP_S)
3605 drbd_start_resync(mdev, C_SYNC_SOURCE);
3606 free_page((unsigned long) buffer);
3610 static int receive_skip(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3612 /* TODO zero copy sink :) */
3613 static char sink[128];
3616 dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n",
3621 want = min_t(int, size, sizeof(sink));
3622 r = drbd_recv(mdev, sink, want);
3623 ERR_IF(r <= 0) break;
3629 static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3631 /* Make sure we've acked all the TCP data associated
3632 * with the data requests being unplugged */
3633 drbd_tcp_quickack(mdev->data.socket);
3638 static int receive_out_of_sync(struct drbd_conf *mdev, enum drbd_packets cmd, unsigned int data_size)
3640 struct p_block_desc *p = &mdev->data.rbuf.block_desc;
3642 switch (mdev->state.conn) {
3643 case C_WF_SYNC_UUID:
3648 dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n",
3649 drbd_conn_str(mdev->state.conn));
3652 drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize));
3657 typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packets cmd, unsigned int to_receive);
3662 drbd_cmd_handler_f function;
3665 static struct data_cmd drbd_cmd_handler[] = {
3666 [P_DATA] = { 1, sizeof(struct p_data), receive_Data },
3667 [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply },
3668 [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } ,
3669 [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } ,
3670 [P_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } ,
3671 [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header80), receive_bitmap } ,
3672 [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header80), receive_UnplugRemote },
3673 [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3674 [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3675 [P_SYNC_PARAM] = { 1, sizeof(struct p_header80), receive_SyncParam },
3676 [P_SYNC_PARAM89] = { 1, sizeof(struct p_header80), receive_SyncParam },
3677 [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol },
3678 [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids },
3679 [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes },
3680 [P_STATE] = { 0, sizeof(struct p_state), receive_state },
3681 [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state },
3682 [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid },
3683 [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest },
3684 [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest },
3685 [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest },
3686 [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip },
3687 [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync },
3688 /* anything missing from this table is in
3689 * the asender_tbl, see get_asender_cmd */
3690 [P_MAX_CMD] = { 0, 0, NULL },
3693 /* All handler functions that expect a sub-header get that sub-heder in
3694 mdev->data.rbuf.header.head.payload.
3696 Usually in mdev->data.rbuf.header.head the callback can find the usual
3697 p_header, but they may not rely on that. Since there is also p_header95 !
3700 static void drbdd(struct drbd_conf *mdev)
3702 union p_header *header = &mdev->data.rbuf.header;
3703 unsigned int packet_size;
3704 enum drbd_packets cmd;
3705 size_t shs; /* sub header size */
3708 while (get_t_state(&mdev->receiver) == Running) {
3709 drbd_thread_current_set_cpu(mdev);
3710 if (!drbd_recv_header(mdev, &cmd, &packet_size))
3713 if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) {
3714 dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size);
3718 shs = drbd_cmd_handler[cmd].pkt_size - sizeof(union p_header);
3719 if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) {
3720 dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size);
3725 rv = drbd_recv(mdev, &header->h80.payload, shs);
3726 if (unlikely(rv != shs)) {
3727 if (!signal_pending(current))
3728 dev_warn(DEV, "short read while reading sub header: rv=%d\n", rv);
3733 rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs);
3735 if (unlikely(!rv)) {
3736 dev_err(DEV, "error receiving %s, l: %d!\n",
3737 cmdname(cmd), packet_size);
3744 drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR));
3746 /* If we leave here, we probably want to update at least the
3747 * "Connected" indicator on stable storage. Do so explicitly here. */
3751 void drbd_flush_workqueue(struct drbd_conf *mdev)
3753 struct drbd_wq_barrier barr;
3755 barr.w.cb = w_prev_work_done;
3756 init_completion(&barr.done);
3757 drbd_queue_work(&mdev->data.work, &barr.w);
3758 wait_for_completion(&barr.done);
3761 static void drbd_disconnect(struct drbd_conf *mdev)
3763 enum drbd_fencing_p fp;
3764 union drbd_state os, ns;
3765 int rv = SS_UNKNOWN_ERROR;
3768 if (mdev->state.conn == C_STANDALONE)
3771 /* asender does not clean up anything. it must not interfere, either */
3772 drbd_thread_stop(&mdev->asender);
3773 drbd_free_sock(mdev);
3775 /* wait for current activity to cease. */
3776 spin_lock_irq(&mdev->req_lock);
3777 _drbd_wait_ee_list_empty(mdev, &mdev->active_ee);
3778 _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee);
3779 _drbd_wait_ee_list_empty(mdev, &mdev->read_ee);
3780 spin_unlock_irq(&mdev->req_lock);
3782 /* We do not have data structures that would allow us to
3783 * get the rs_pending_cnt down to 0 again.
3784 * * On C_SYNC_TARGET we do not have any data structures describing
3785 * the pending RSDataRequest's we have sent.
3786 * * On C_SYNC_SOURCE there is no data structure that tracks
3787 * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget.
3788 * And no, it is not the sum of the reference counts in the
3789 * resync_LRU. The resync_LRU tracks the whole operation including
3790 * the disk-IO, while the rs_pending_cnt only tracks the blocks
3792 drbd_rs_cancel_all(mdev);
3794 mdev->rs_failed = 0;
3795 atomic_set(&mdev->rs_pending_cnt, 0);
3796 wake_up(&mdev->misc_wait);
3798 del_timer(&mdev->request_timer);
3800 /* make sure syncer is stopped and w_resume_next_sg queued */
3801 del_timer_sync(&mdev->resync_timer);
3802 resync_timer_fn((unsigned long)mdev);
3804 /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier,
3805 * w_make_resync_request etc. which may still be on the worker queue
3806 * to be "canceled" */
3807 drbd_flush_workqueue(mdev);
3809 /* This also does reclaim_net_ee(). If we do this too early, we might
3810 * miss some resync ee and pages.*/
3811 drbd_process_done_ee(mdev);
3813 kfree(mdev->p_uuid);
3814 mdev->p_uuid = NULL;
3816 if (!is_susp(mdev->state))
3819 dev_info(DEV, "Connection closed\n");
3824 if (get_ldev(mdev)) {
3825 fp = mdev->ldev->dc.fencing;
3829 if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN)
3830 drbd_try_outdate_peer_async(mdev);
3832 spin_lock_irq(&mdev->req_lock);
3834 if (os.conn >= C_UNCONNECTED) {
3835 /* Do not restart in case we are C_DISCONNECTING */
3837 ns.conn = C_UNCONNECTED;
3838 rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
3840 spin_unlock_irq(&mdev->req_lock);
3842 if (os.conn == C_DISCONNECTING) {
3843 wait_event(mdev->net_cnt_wait, atomic_read(&mdev->net_cnt) == 0);
3845 crypto_free_hash(mdev->cram_hmac_tfm);
3846 mdev->cram_hmac_tfm = NULL;
3848 kfree(mdev->net_conf);
3849 mdev->net_conf = NULL;
3850 drbd_request_state(mdev, NS(conn, C_STANDALONE));
3853 /* serialize with bitmap writeout triggered by the state change,
3855 wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags));
3857 /* tcp_close and release of sendpage pages can be deferred. I don't
3858 * want to use SO_LINGER, because apparently it can be deferred for
3859 * more than 20 seconds (longest time I checked).
3861 * Actually we don't care for exactly when the network stack does its
3862 * put_page(), but release our reference on these pages right here.
3864 i = drbd_release_ee(mdev, &mdev->net_ee);
3866 dev_info(DEV, "net_ee not empty, killed %u entries\n", i);
3867 i = atomic_read(&mdev->pp_in_use_by_net);
3869 dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i);
3870 i = atomic_read(&mdev->pp_in_use);
3872 dev_info(DEV, "pp_in_use = %d, expected 0\n", i);
3874 D_ASSERT(list_empty(&mdev->read_ee));
3875 D_ASSERT(list_empty(&mdev->active_ee));
3876 D_ASSERT(list_empty(&mdev->sync_ee));
3877 D_ASSERT(list_empty(&mdev->done_ee));
3879 /* ok, no more ee's on the fly, it is safe to reset the epoch_size */
3880 atomic_set(&mdev->current_epoch->epoch_size, 0);
3881 D_ASSERT(list_empty(&mdev->current_epoch->list));
3885 * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version
3886 * we can agree on is stored in agreed_pro_version.
3888 * feature flags and the reserved array should be enough room for future
3889 * enhancements of the handshake protocol, and possible plugins...
3891 * for now, they are expected to be zero, but ignored.
3893 static int drbd_send_handshake(struct drbd_conf *mdev)
3895 /* ASSERT current == mdev->receiver ... */
3896 struct p_handshake *p = &mdev->data.sbuf.handshake;
3899 if (mutex_lock_interruptible(&mdev->data.mutex)) {
3900 dev_err(DEV, "interrupted during initial handshake\n");
3901 return 0; /* interrupted. not ok. */
3904 if (mdev->data.socket == NULL) {
3905 mutex_unlock(&mdev->data.mutex);
3909 memset(p, 0, sizeof(*p));
3910 p->protocol_min = cpu_to_be32(PRO_VERSION_MIN);
3911 p->protocol_max = cpu_to_be32(PRO_VERSION_MAX);
3912 ok = _drbd_send_cmd( mdev, mdev->data.socket, P_HAND_SHAKE,
3913 (struct p_header80 *)p, sizeof(*p), 0 );
3914 mutex_unlock(&mdev->data.mutex);
3920 * 1 yes, we have a valid connection
3921 * 0 oops, did not work out, please try again
3922 * -1 peer talks different language,
3923 * no point in trying again, please go standalone.
3925 static int drbd_do_handshake(struct drbd_conf *mdev)
3927 /* ASSERT current == mdev->receiver ... */
3928 struct p_handshake *p = &mdev->data.rbuf.handshake;
3929 const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80);
3930 unsigned int length;
3931 enum drbd_packets cmd;
3934 rv = drbd_send_handshake(mdev);
3938 rv = drbd_recv_header(mdev, &cmd, &length);
3942 if (cmd != P_HAND_SHAKE) {
3943 dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n",
3948 if (length != expect) {
3949 dev_err(DEV, "expected HandShake length: %u, received: %u\n",
3954 rv = drbd_recv(mdev, &p->head.payload, expect);
3957 if (!signal_pending(current))
3958 dev_warn(DEV, "short read receiving handshake packet: l=%u\n", rv);
3962 p->protocol_min = be32_to_cpu(p->protocol_min);
3963 p->protocol_max = be32_to_cpu(p->protocol_max);
3964 if (p->protocol_max == 0)
3965 p->protocol_max = p->protocol_min;
3967 if (PRO_VERSION_MAX < p->protocol_min ||
3968 PRO_VERSION_MIN > p->protocol_max)
3971 mdev->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max);
3973 dev_info(DEV, "Handshake successful: "
3974 "Agreed network protocol version %d\n", mdev->agreed_pro_version);
3979 dev_err(DEV, "incompatible DRBD dialects: "
3980 "I support %d-%d, peer supports %d-%d\n",
3981 PRO_VERSION_MIN, PRO_VERSION_MAX,
3982 p->protocol_min, p->protocol_max);
3986 #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE)
3987 static int drbd_do_auth(struct drbd_conf *mdev)
3989 dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n");
3990 dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n");
3994 #define CHALLENGE_LEN 64
3998 0 - failed, try again (network error),
3999 -1 - auth failed, don't try again.
4002 static int drbd_do_auth(struct drbd_conf *mdev)
4004 char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */
4005 struct scatterlist sg;
4006 char *response = NULL;
4007 char *right_response = NULL;
4008 char *peers_ch = NULL;
4009 unsigned int key_len = strlen(mdev->net_conf->shared_secret);
4010 unsigned int resp_size;
4011 struct hash_desc desc;
4012 enum drbd_packets cmd;
4013 unsigned int length;
4016 desc.tfm = mdev->cram_hmac_tfm;
4019 rv = crypto_hash_setkey(mdev->cram_hmac_tfm,
4020 (u8 *)mdev->net_conf->shared_secret, key_len);
4022 dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv);
4027 get_random_bytes(my_challenge, CHALLENGE_LEN);
4029 rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN);
4033 rv = drbd_recv_header(mdev, &cmd, &length);
4037 if (cmd != P_AUTH_CHALLENGE) {
4038 dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n",
4044 if (length > CHALLENGE_LEN * 2) {
4045 dev_err(DEV, "expected AuthChallenge payload too big.\n");
4050 peers_ch = kmalloc(length, GFP_NOIO);
4051 if (peers_ch == NULL) {
4052 dev_err(DEV, "kmalloc of peers_ch failed\n");
4057 rv = drbd_recv(mdev, peers_ch, length);
4060 if (!signal_pending(current))
4061 dev_warn(DEV, "short read AuthChallenge: l=%u\n", rv);
4066 resp_size = crypto_hash_digestsize(mdev->cram_hmac_tfm);
4067 response = kmalloc(resp_size, GFP_NOIO);
4068 if (response == NULL) {
4069 dev_err(DEV, "kmalloc of response failed\n");
4074 sg_init_table(&sg, 1);
4075 sg_set_buf(&sg, peers_ch, length);
4077 rv = crypto_hash_digest(&desc, &sg, sg.length, response);
4079 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4084 rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size);
4088 rv = drbd_recv_header(mdev, &cmd, &length);
4092 if (cmd != P_AUTH_RESPONSE) {
4093 dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n",
4099 if (length != resp_size) {
4100 dev_err(DEV, "expected AuthResponse payload of wrong size\n");
4105 rv = drbd_recv(mdev, response , resp_size);
4107 if (rv != resp_size) {
4108 if (!signal_pending(current))
4109 dev_warn(DEV, "short read receiving AuthResponse: l=%u\n", rv);
4114 right_response = kmalloc(resp_size, GFP_NOIO);
4115 if (right_response == NULL) {
4116 dev_err(DEV, "kmalloc of right_response failed\n");
4121 sg_set_buf(&sg, my_challenge, CHALLENGE_LEN);
4123 rv = crypto_hash_digest(&desc, &sg, sg.length, right_response);
4125 dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv);
4130 rv = !memcmp(response, right_response, resp_size);
4133 dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n",
4134 resp_size, mdev->net_conf->cram_hmac_alg);
4141 kfree(right_response);
4147 int drbdd_init(struct drbd_thread *thi)
4149 struct drbd_conf *mdev = thi->mdev;
4150 unsigned int minor = mdev_to_minor(mdev);
4153 sprintf(current->comm, "drbd%d_receiver", minor);
4155 dev_info(DEV, "receiver (re)started\n");
4158 h = drbd_connect(mdev);
4160 drbd_disconnect(mdev);
4161 schedule_timeout_interruptible(HZ);
4164 dev_warn(DEV, "Discarding network configuration.\n");
4165 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4170 if (get_net_conf(mdev)) {
4176 drbd_disconnect(mdev);
4178 dev_info(DEV, "receiver terminated\n");
4182 /* ********* acknowledge sender ******** */
4184 static int got_RqSReply(struct drbd_conf *mdev, struct p_header80 *h)
4186 struct p_req_state_reply *p = (struct p_req_state_reply *)h;
4188 int retcode = be32_to_cpu(p->retcode);
4190 if (retcode >= SS_SUCCESS) {
4191 set_bit(CL_ST_CHG_SUCCESS, &mdev->flags);
4193 set_bit(CL_ST_CHG_FAIL, &mdev->flags);
4194 dev_err(DEV, "Requested state change failed by peer: %s (%d)\n",
4195 drbd_set_st_err_str(retcode), retcode);
4197 wake_up(&mdev->state_wait);
4202 static int got_Ping(struct drbd_conf *mdev, struct p_header80 *h)
4204 return drbd_send_ping_ack(mdev);
4208 static int got_PingAck(struct drbd_conf *mdev, struct p_header80 *h)
4210 /* restore idle timeout */
4211 mdev->meta.socket->sk->sk_rcvtimeo = mdev->net_conf->ping_int*HZ;
4212 if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags))
4213 wake_up(&mdev->misc_wait);
4218 static int got_IsInSync(struct drbd_conf *mdev, struct p_header80 *h)
4220 struct p_block_ack *p = (struct p_block_ack *)h;
4221 sector_t sector = be64_to_cpu(p->sector);
4222 int blksize = be32_to_cpu(p->blksize);
4224 D_ASSERT(mdev->agreed_pro_version >= 89);
4226 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4228 if (get_ldev(mdev)) {
4229 drbd_rs_complete_io(mdev, sector);
4230 drbd_set_in_sync(mdev, sector, blksize);
4231 /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */
4232 mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT);
4235 dec_rs_pending(mdev);
4236 atomic_add(blksize >> 9, &mdev->rs_sect_in);
4241 /* when we receive the ACK for a write request,
4242 * verify that we actually know about it */
4243 static struct drbd_request *ack_id_to_req(struct drbd_conf *mdev, u64 id,
4246 struct hlist_head *slot = tl_hash_slot(mdev, sector);
4247 struct hlist_node *n;
4248 struct drbd_request *req;
4250 hlist_for_each_entry(req, n, slot, collision) {
4251 if ((unsigned long)req == (unsigned long)id) {
4252 if (req->sector != sector) {
4253 dev_err(DEV, "ack_id_to_req: found req %p but it has "
4254 "wrong sector (%llus versus %llus)\n", req,
4255 (unsigned long long)req->sector,
4256 (unsigned long long)sector);
4265 static int validate_req_change_req_state(struct drbd_conf *mdev,
4266 u64 id, sector_t sector,
4267 struct drbd_request *(*validator)(struct drbd_conf *, u64, sector_t),
4268 const char *func, enum drbd_req_event what)
4270 struct drbd_request *req;
4271 struct bio_and_error m;
4273 spin_lock_irq(&mdev->req_lock);
4274 req = validator(mdev, id, sector);
4275 if (unlikely(!req)) {
4276 spin_unlock_irq(&mdev->req_lock);
4278 dev_err(DEV, "%s: failed to find req %p, sector %llus\n", func,
4279 (void *)(unsigned long)id, (unsigned long long)sector);
4282 __req_mod(req, what, &m);
4283 spin_unlock_irq(&mdev->req_lock);
4286 complete_master_bio(mdev, &m);
4290 static int got_BlockAck(struct drbd_conf *mdev, struct p_header80 *h)
4292 struct p_block_ack *p = (struct p_block_ack *)h;
4293 sector_t sector = be64_to_cpu(p->sector);
4294 int blksize = be32_to_cpu(p->blksize);
4295 enum drbd_req_event what;
4297 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4299 if (p->block_id == ID_SYNCER) {
4300 drbd_set_in_sync(mdev, sector, blksize);
4301 dec_rs_pending(mdev);
4304 switch (be16_to_cpu(h->command)) {
4305 case P_RS_WRITE_ACK:
4306 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4307 what = write_acked_by_peer_and_sis;
4310 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4311 what = write_acked_by_peer;
4314 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_B);
4315 what = recv_acked_by_peer;
4318 D_ASSERT(mdev->net_conf->wire_protocol == DRBD_PROT_C);
4319 what = conflict_discarded_by_peer;
4326 return validate_req_change_req_state(mdev, p->block_id, sector,
4327 ack_id_to_req, __func__, what);
4330 static int got_NegAck(struct drbd_conf *mdev, struct p_header80 *h)
4332 struct p_block_ack *p = (struct p_block_ack *)h;
4333 sector_t sector = be64_to_cpu(p->sector);
4334 int size = be32_to_cpu(p->blksize);
4335 struct drbd_request *req;
4336 struct bio_and_error m;
4338 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4340 if (p->block_id == ID_SYNCER) {
4341 dec_rs_pending(mdev);
4342 drbd_rs_failed_io(mdev, sector, size);
4346 spin_lock_irq(&mdev->req_lock);
4347 req = ack_id_to_req(mdev, p->block_id, sector);
4349 spin_unlock_irq(&mdev->req_lock);
4350 if (mdev->net_conf->wire_protocol == DRBD_PROT_A ||
4351 mdev->net_conf->wire_protocol == DRBD_PROT_B) {
4352 /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs.
4353 The master bio might already be completed, therefore the
4354 request is no longer in the collision hash.
4355 => Do not try to validate block_id as request. */
4356 /* In Protocol B we might already have got a P_RECV_ACK
4357 but then get a P_NEG_ACK after wards. */
4358 drbd_set_out_of_sync(mdev, sector, size);
4361 dev_err(DEV, "%s: failed to find req %p, sector %llus\n", __func__,
4362 (void *)(unsigned long)p->block_id, (unsigned long long)sector);
4366 __req_mod(req, neg_acked, &m);
4367 spin_unlock_irq(&mdev->req_lock);
4370 complete_master_bio(mdev, &m);
4374 static int got_NegDReply(struct drbd_conf *mdev, struct p_header80 *h)
4376 struct p_block_ack *p = (struct p_block_ack *)h;
4377 sector_t sector = be64_to_cpu(p->sector);
4379 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4380 dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n",
4381 (unsigned long long)sector, be32_to_cpu(p->blksize));
4383 return validate_req_change_req_state(mdev, p->block_id, sector,
4384 ar_id_to_req, __func__ , neg_acked);
4387 static int got_NegRSDReply(struct drbd_conf *mdev, struct p_header80 *h)
4391 struct p_block_ack *p = (struct p_block_ack *)h;
4393 sector = be64_to_cpu(p->sector);
4394 size = be32_to_cpu(p->blksize);
4396 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4398 dec_rs_pending(mdev);
4400 if (get_ldev_if_state(mdev, D_FAILED)) {
4401 drbd_rs_complete_io(mdev, sector);
4402 switch (be16_to_cpu(h->command)) {
4403 case P_NEG_RS_DREPLY:
4404 drbd_rs_failed_io(mdev, sector, size);
4418 static int got_BarrierAck(struct drbd_conf *mdev, struct p_header80 *h)
4420 struct p_barrier_ack *p = (struct p_barrier_ack *)h;
4422 tl_release(mdev, p->barrier, be32_to_cpu(p->set_size));
4424 if (mdev->state.conn == C_AHEAD &&
4425 atomic_read(&mdev->ap_in_flight) == 0 &&
4426 !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags)) {
4427 mdev->start_resync_timer.expires = jiffies + HZ;
4428 add_timer(&mdev->start_resync_timer);
4434 static int got_OVResult(struct drbd_conf *mdev, struct p_header80 *h)
4436 struct p_block_ack *p = (struct p_block_ack *)h;
4437 struct drbd_work *w;
4441 sector = be64_to_cpu(p->sector);
4442 size = be32_to_cpu(p->blksize);
4444 update_peer_seq(mdev, be32_to_cpu(p->seq_num));
4446 if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC)
4447 drbd_ov_oos_found(mdev, sector, size);
4451 if (!get_ldev(mdev))
4454 drbd_rs_complete_io(mdev, sector);
4455 dec_rs_pending(mdev);
4459 /* let's advance progress step marks only for every other megabyte */
4460 if ((mdev->ov_left & 0x200) == 0x200)
4461 drbd_advance_rs_marks(mdev, mdev->ov_left);
4463 if (mdev->ov_left == 0) {
4464 w = kmalloc(sizeof(*w), GFP_NOIO);
4466 w->cb = w_ov_finished;
4467 drbd_queue_work_front(&mdev->data.work, w);
4469 dev_err(DEV, "kmalloc(w) failed.");
4471 drbd_resync_finished(mdev);
4478 static int got_skip(struct drbd_conf *mdev, struct p_header80 *h)
4483 struct asender_cmd {
4485 int (*process)(struct drbd_conf *mdev, struct p_header80 *h);
4488 static struct asender_cmd *get_asender_cmd(int cmd)
4490 static struct asender_cmd asender_tbl[] = {
4491 /* anything missing from this table is in
4492 * the drbd_cmd_handler (drbd_default_handler) table,
4493 * see the beginning of drbdd() */
4494 [P_PING] = { sizeof(struct p_header80), got_Ping },
4495 [P_PING_ACK] = { sizeof(struct p_header80), got_PingAck },
4496 [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4497 [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4498 [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4499 [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck },
4500 [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck },
4501 [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply },
4502 [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply},
4503 [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult },
4504 [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck },
4505 [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply },
4506 [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync },
4507 [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip },
4508 [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply},
4509 [P_MAX_CMD] = { 0, NULL },
4511 if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL)
4513 return &asender_tbl[cmd];
4516 int drbd_asender(struct drbd_thread *thi)
4518 struct drbd_conf *mdev = thi->mdev;
4519 struct p_header80 *h = &mdev->meta.rbuf.header.h80;
4520 struct asender_cmd *cmd = NULL;
4525 int expect = sizeof(struct p_header80);
4527 int ping_timeout_active = 0;
4529 sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev));
4531 current->policy = SCHED_RR; /* Make this a realtime task! */
4532 current->rt_priority = 2; /* more important than all other tasks */
4534 while (get_t_state(thi) == Running) {
4535 drbd_thread_current_set_cpu(mdev);
4536 if (test_and_clear_bit(SEND_PING, &mdev->flags)) {
4537 ERR_IF(!drbd_send_ping(mdev)) goto reconnect;
4538 mdev->meta.socket->sk->sk_rcvtimeo =
4539 mdev->net_conf->ping_timeo*HZ/10;
4540 ping_timeout_active = 1;
4543 /* conditionally cork;
4544 * it may hurt latency if we cork without much to send */
4545 if (!mdev->net_conf->no_cork &&
4546 3 < atomic_read(&mdev->unacked_cnt))
4547 drbd_tcp_cork(mdev->meta.socket);
4549 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4550 flush_signals(current);
4551 if (!drbd_process_done_ee(mdev))
4553 /* to avoid race with newly queued ACKs */
4554 set_bit(SIGNAL_ASENDER, &mdev->flags);
4555 spin_lock_irq(&mdev->req_lock);
4556 empty = list_empty(&mdev->done_ee);
4557 spin_unlock_irq(&mdev->req_lock);
4558 /* new ack may have been queued right here,
4559 * but then there is also a signal pending,
4560 * and we start over... */
4564 /* but unconditionally uncork unless disabled */
4565 if (!mdev->net_conf->no_cork)
4566 drbd_tcp_uncork(mdev->meta.socket);
4568 /* short circuit, recv_msg would return EINTR anyways. */
4569 if (signal_pending(current))
4572 rv = drbd_recv_short(mdev, mdev->meta.socket,
4573 buf, expect-received, 0);
4574 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4576 flush_signals(current);
4579 * -EINTR (on meta) we got a signal
4580 * -EAGAIN (on meta) rcvtimeo expired
4581 * -ECONNRESET other side closed the connection
4582 * -ERESTARTSYS (on data) we got a signal
4583 * rv < 0 other than above: unexpected error!
4584 * rv == expected: full header or command
4585 * rv < expected: "woken" by signal during receive
4586 * rv == 0 : "connection shut down by peer"
4588 if (likely(rv > 0)) {
4591 } else if (rv == 0) {
4592 dev_err(DEV, "meta connection shut down by peer.\n");
4594 } else if (rv == -EAGAIN) {
4595 /* If the data socket received something meanwhile,
4596 * that is good enough: peer is still alive. */
4597 if (time_after(mdev->last_received,
4598 jiffies - mdev->meta.socket->sk->sk_rcvtimeo))
4600 if (ping_timeout_active) {
4601 dev_err(DEV, "PingAck did not arrive in time.\n");
4604 set_bit(SEND_PING, &mdev->flags);
4606 } else if (rv == -EINTR) {
4609 dev_err(DEV, "sock_recvmsg returned %d\n", rv);
4613 if (received == expect && cmd == NULL) {
4614 if (unlikely(h->magic != cpu_to_be32(DRBD_MAGIC))) {
4615 dev_err(DEV, "magic?? on meta m: 0x%08x c: %d l: %d\n",
4616 be32_to_cpu(h->magic),
4617 be16_to_cpu(h->command),
4618 be16_to_cpu(h->length));
4621 cmd = get_asender_cmd(be16_to_cpu(h->command));
4622 len = be16_to_cpu(h->length);
4623 if (unlikely(cmd == NULL)) {
4624 dev_err(DEV, "unknown command?? on meta m: 0x%08x c: %d l: %d\n",
4625 be32_to_cpu(h->magic),
4626 be16_to_cpu(h->command),
4627 be16_to_cpu(h->length));
4630 expect = cmd->pkt_size;
4631 ERR_IF(len != expect-sizeof(struct p_header80))
4634 if (received == expect) {
4635 mdev->last_received = jiffies;
4636 D_ASSERT(cmd != NULL);
4637 if (!cmd->process(mdev, h))
4640 /* the idle_timeout (ping-int)
4641 * has been restored in got_PingAck() */
4642 if (cmd == get_asender_cmd(P_PING_ACK))
4643 ping_timeout_active = 0;
4647 expect = sizeof(struct p_header80);
4654 drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE));
4659 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
4662 clear_bit(SIGNAL_ASENDER, &mdev->flags);
4664 D_ASSERT(mdev->state.conn < C_CONNECTED);
4665 dev_info(DEV, "asender terminated\n");