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 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
11 from Logicworks, Inc. for making SDP replication support possible.
13 drbd is free software; you can redistribute it and/or modify
14 it under the terms of the GNU General Public License as published by
15 the Free Software Foundation; either version 2, or (at your option)
18 drbd is distributed in the hope that it will be useful,
19 but WITHOUT ANY WARRANTY; without even the implied warranty of
20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 GNU General Public License for more details.
23 You should have received a copy of the GNU General Public License
24 along with drbd; see the file COPYING. If not, write to
25 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
29 #include <linux/module.h>
30 #include <linux/drbd.h>
31 #include <asm/uaccess.h>
32 #include <asm/types.h>
34 #include <linux/ctype.h>
35 #include <linux/mutex.h>
37 #include <linux/file.h>
38 #include <linux/proc_fs.h>
39 #include <linux/init.h>
41 #include <linux/memcontrol.h>
42 #include <linux/mm_inline.h>
43 #include <linux/slab.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/notifier.h>
47 #include <linux/kthread.h>
49 #define __KERNEL_SYSCALLS__
50 #include <linux/unistd.h>
51 #include <linux/vmalloc.h>
53 #include <linux/drbd_limits.h>
55 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
59 static DEFINE_MUTEX(drbd_main_mutex);
60 int drbdd_init(struct drbd_thread *);
61 int drbd_worker(struct drbd_thread *);
62 int drbd_asender(struct drbd_thread *);
65 static int drbd_open(struct block_device *bdev, fmode_t mode);
66 static int drbd_release(struct gendisk *gd, fmode_t mode);
67 static int w_md_sync(struct drbd_work *w, int unused);
68 static void md_sync_timer_fn(unsigned long data);
69 static int w_bitmap_io(struct drbd_work *w, int unused);
70 static int w_go_diskless(struct drbd_work *w, int unused);
72 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
73 "Lars Ellenberg <lars@linbit.com>");
74 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
75 MODULE_VERSION(REL_VERSION);
76 MODULE_LICENSE("GPL");
77 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
78 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
79 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
81 #include <linux/moduleparam.h>
82 /* allow_open_on_secondary */
83 MODULE_PARM_DESC(allow_oos, "DONT USE!");
84 /* thanks to these macros, if compiled into the kernel (not-module),
85 * this becomes the boot parameter drbd.minor_count */
86 module_param(minor_count, uint, 0444);
87 module_param(disable_sendpage, bool, 0644);
88 module_param(allow_oos, bool, 0);
89 module_param(proc_details, int, 0644);
91 #ifdef CONFIG_DRBD_FAULT_INJECTION
94 static int fault_count;
96 /* bitmap of enabled faults */
97 module_param(enable_faults, int, 0664);
98 /* fault rate % value - applies to all enabled faults */
99 module_param(fault_rate, int, 0664);
100 /* count of faults inserted */
101 module_param(fault_count, int, 0664);
102 /* bitmap of devices to insert faults on */
103 module_param(fault_devs, int, 0644);
106 /* module parameter, defined */
107 unsigned int minor_count = DRBD_MINOR_COUNT_DEF;
108 int disable_sendpage;
110 int proc_details; /* Detail level in proc drbd*/
112 /* Module parameter for setting the user mode helper program
113 * to run. Default is /sbin/drbdadm */
114 char usermode_helper[80] = "/sbin/drbdadm";
116 module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644);
118 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
119 * as member "struct gendisk *vdisk;"
122 struct list_head drbd_tconns; /* list of struct drbd_tconn */
124 struct kmem_cache *drbd_request_cache;
125 struct kmem_cache *drbd_ee_cache; /* peer requests */
126 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
127 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
128 mempool_t *drbd_request_mempool;
129 mempool_t *drbd_ee_mempool;
130 mempool_t *drbd_md_io_page_pool;
131 struct bio_set *drbd_md_io_bio_set;
133 /* I do not use a standard mempool, because:
134 1) I want to hand out the pre-allocated objects first.
135 2) I want to be able to interrupt sleeping allocation with a signal.
136 Note: This is a single linked list, the next pointer is the private
137 member of struct page.
139 struct page *drbd_pp_pool;
140 spinlock_t drbd_pp_lock;
142 wait_queue_head_t drbd_pp_wait;
144 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
146 static const struct block_device_operations drbd_ops = {
147 .owner = THIS_MODULE,
149 .release = drbd_release,
152 static void bio_destructor_drbd(struct bio *bio)
154 bio_free(bio, drbd_md_io_bio_set);
157 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
161 if (!drbd_md_io_bio_set)
162 return bio_alloc(gfp_mask, 1);
164 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
167 bio->bi_destructor = bio_destructor_drbd;
172 /* When checking with sparse, and this is an inline function, sparse will
173 give tons of false positives. When this is a real functions sparse works.
175 int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins)
179 atomic_inc(&mdev->local_cnt);
180 io_allowed = (mdev->state.disk >= mins);
182 if (atomic_dec_and_test(&mdev->local_cnt))
183 wake_up(&mdev->misc_wait);
191 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
192 * @tconn: DRBD connection.
193 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
194 * @set_size: Expected number of requests before that barrier.
196 * In case the passed barrier_nr or set_size does not match the oldest
197 * epoch of not yet barrier-acked requests, this function will cause a
198 * termination of the connection.
200 void tl_release(struct drbd_tconn *tconn, unsigned int barrier_nr,
201 unsigned int set_size)
203 struct drbd_request *r;
204 struct drbd_request *req = NULL;
205 int expect_epoch = 0;
208 spin_lock_irq(&tconn->req_lock);
210 /* find latest not yet barrier-acked write request,
211 * count writes in its epoch. */
212 list_for_each_entry(r, &tconn->transfer_log, tl_requests) {
213 const unsigned long s = r->rq_state;
217 if (!(s & RQ_NET_MASK))
222 expect_epoch = req->epoch;
225 if (r->epoch != expect_epoch)
229 /* if (s & RQ_DONE): not expected */
230 /* if (!(s & RQ_NET_MASK)): not expected */
235 /* first some paranoia code */
237 conn_err(tconn, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
241 if (expect_epoch != barrier_nr) {
242 conn_err(tconn, "BAD! BarrierAck #%u received, expected #%u!\n",
243 barrier_nr, expect_epoch);
247 if (expect_size != set_size) {
248 conn_err(tconn, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
249 barrier_nr, set_size, expect_size);
253 /* Clean up list of requests processed during current epoch */
254 list_for_each_entry_safe(req, r, &tconn->transfer_log, tl_requests) {
255 if (req->epoch != expect_epoch)
257 _req_mod(req, BARRIER_ACKED);
259 spin_unlock_irq(&tconn->req_lock);
264 spin_unlock_irq(&tconn->req_lock);
265 conn_request_state(tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
270 * _tl_restart() - Walks the transfer log, and applies an action to all requests
271 * @mdev: DRBD device.
272 * @what: The action/event to perform with all request objects
274 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
275 * RESTART_FROZEN_DISK_IO.
277 /* must hold resource->req_lock */
278 void _tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what)
280 struct drbd_request *req, *r;
282 list_for_each_entry_safe(req, r, &tconn->transfer_log, tl_requests)
286 void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what)
288 spin_lock_irq(&tconn->req_lock);
289 _tl_restart(tconn, what);
290 spin_unlock_irq(&tconn->req_lock);
294 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
295 * @mdev: DRBD device.
297 * This is called after the connection to the peer was lost. The storage covered
298 * by the requests on the transfer gets marked as our of sync. Called from the
299 * receiver thread and the worker thread.
301 void tl_clear(struct drbd_tconn *tconn)
303 tl_restart(tconn, CONNECTION_LOST_WHILE_PENDING);
307 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain mdev in the TL
308 * @mdev: DRBD device.
310 void tl_abort_disk_io(struct drbd_conf *mdev)
312 struct drbd_tconn *tconn = mdev->tconn;
313 struct drbd_request *req, *r;
315 spin_lock_irq(&tconn->req_lock);
316 list_for_each_entry_safe(req, r, &tconn->transfer_log, tl_requests) {
317 if (!(req->rq_state & RQ_LOCAL_PENDING))
319 if (req->w.mdev != mdev)
321 _req_mod(req, ABORT_DISK_IO);
323 spin_unlock_irq(&tconn->req_lock);
326 static int drbd_thread_setup(void *arg)
328 struct drbd_thread *thi = (struct drbd_thread *) arg;
329 struct drbd_tconn *tconn = thi->tconn;
333 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
334 thi->name[0], thi->tconn->name);
337 retval = thi->function(thi);
339 spin_lock_irqsave(&thi->t_lock, flags);
341 /* if the receiver has been "EXITING", the last thing it did
342 * was set the conn state to "StandAlone",
343 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
344 * and receiver thread will be "started".
345 * drbd_thread_start needs to set "RESTARTING" in that case.
346 * t_state check and assignment needs to be within the same spinlock,
347 * so either thread_start sees EXITING, and can remap to RESTARTING,
348 * or thread_start see NONE, and can proceed as normal.
351 if (thi->t_state == RESTARTING) {
352 conn_info(tconn, "Restarting %s thread\n", thi->name);
353 thi->t_state = RUNNING;
354 spin_unlock_irqrestore(&thi->t_lock, flags);
361 complete_all(&thi->stop);
362 spin_unlock_irqrestore(&thi->t_lock, flags);
364 conn_info(tconn, "Terminating %s\n", current->comm);
366 /* Release mod reference taken when thread was started */
368 kref_put(&tconn->kref, &conn_destroy);
369 module_put(THIS_MODULE);
373 static void drbd_thread_init(struct drbd_tconn *tconn, struct drbd_thread *thi,
374 int (*func) (struct drbd_thread *), char *name)
376 spin_lock_init(&thi->t_lock);
379 thi->function = func;
381 strncpy(thi->name, name, ARRAY_SIZE(thi->name));
384 int drbd_thread_start(struct drbd_thread *thi)
386 struct drbd_tconn *tconn = thi->tconn;
387 struct task_struct *nt;
390 /* is used from state engine doing drbd_thread_stop_nowait,
391 * while holding the req lock irqsave */
392 spin_lock_irqsave(&thi->t_lock, flags);
394 switch (thi->t_state) {
396 conn_info(tconn, "Starting %s thread (from %s [%d])\n",
397 thi->name, current->comm, current->pid);
399 /* Get ref on module for thread - this is released when thread exits */
400 if (!try_module_get(THIS_MODULE)) {
401 conn_err(tconn, "Failed to get module reference in drbd_thread_start\n");
402 spin_unlock_irqrestore(&thi->t_lock, flags);
406 kref_get(&thi->tconn->kref);
408 init_completion(&thi->stop);
409 thi->reset_cpu_mask = 1;
410 thi->t_state = RUNNING;
411 spin_unlock_irqrestore(&thi->t_lock, flags);
412 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
414 nt = kthread_create(drbd_thread_setup, (void *) thi,
415 "drbd_%c_%s", thi->name[0], thi->tconn->name);
418 conn_err(tconn, "Couldn't start thread\n");
420 kref_put(&tconn->kref, &conn_destroy);
421 module_put(THIS_MODULE);
424 spin_lock_irqsave(&thi->t_lock, flags);
426 thi->t_state = RUNNING;
427 spin_unlock_irqrestore(&thi->t_lock, flags);
431 thi->t_state = RESTARTING;
432 conn_info(tconn, "Restarting %s thread (from %s [%d])\n",
433 thi->name, current->comm, current->pid);
438 spin_unlock_irqrestore(&thi->t_lock, flags);
446 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
450 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
452 /* may be called from state engine, holding the req lock irqsave */
453 spin_lock_irqsave(&thi->t_lock, flags);
455 if (thi->t_state == NONE) {
456 spin_unlock_irqrestore(&thi->t_lock, flags);
458 drbd_thread_start(thi);
462 if (thi->t_state != ns) {
463 if (thi->task == NULL) {
464 spin_unlock_irqrestore(&thi->t_lock, flags);
470 init_completion(&thi->stop);
471 if (thi->task != current)
472 force_sig(DRBD_SIGKILL, thi->task);
475 spin_unlock_irqrestore(&thi->t_lock, flags);
478 wait_for_completion(&thi->stop);
481 static struct drbd_thread *drbd_task_to_thread(struct drbd_tconn *tconn, struct task_struct *task)
483 struct drbd_thread *thi =
484 task == tconn->receiver.task ? &tconn->receiver :
485 task == tconn->asender.task ? &tconn->asender :
486 task == tconn->worker.task ? &tconn->worker : NULL;
491 char *drbd_task_to_thread_name(struct drbd_tconn *tconn, struct task_struct *task)
493 struct drbd_thread *thi = drbd_task_to_thread(tconn, task);
494 return thi ? thi->name : task->comm;
497 int conn_lowest_minor(struct drbd_tconn *tconn)
499 struct drbd_conf *mdev;
503 mdev = idr_get_next(&tconn->volumes, &vnr);
504 m = mdev ? mdev_to_minor(mdev) : -1;
512 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
513 * @mdev: DRBD device.
515 * Forces all threads of a device onto the same CPU. This is beneficial for
516 * DRBD's performance. May be overwritten by user's configuration.
518 void drbd_calc_cpu_mask(struct drbd_tconn *tconn)
523 if (cpumask_weight(tconn->cpu_mask))
526 ord = conn_lowest_minor(tconn) % cpumask_weight(cpu_online_mask);
527 for_each_online_cpu(cpu) {
529 cpumask_set_cpu(cpu, tconn->cpu_mask);
533 /* should not be reached */
534 cpumask_setall(tconn->cpu_mask);
538 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
539 * @mdev: DRBD device.
540 * @thi: drbd_thread object
542 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
545 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
547 struct task_struct *p = current;
549 if (!thi->reset_cpu_mask)
551 thi->reset_cpu_mask = 0;
552 set_cpus_allowed_ptr(p, thi->tconn->cpu_mask);
557 * drbd_header_size - size of a packet header
559 * The header size is a multiple of 8, so any payload following the header is
560 * word aligned on 64-bit architectures. (The bitmap send and receive code
563 unsigned int drbd_header_size(struct drbd_tconn *tconn)
565 if (tconn->agreed_pro_version >= 100) {
566 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
567 return sizeof(struct p_header100);
569 BUILD_BUG_ON(sizeof(struct p_header80) !=
570 sizeof(struct p_header95));
571 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
572 return sizeof(struct p_header80);
576 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
578 h->magic = cpu_to_be32(DRBD_MAGIC);
579 h->command = cpu_to_be16(cmd);
580 h->length = cpu_to_be16(size);
581 return sizeof(struct p_header80);
584 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
586 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
587 h->command = cpu_to_be16(cmd);
588 h->length = cpu_to_be32(size);
589 return sizeof(struct p_header95);
592 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
595 h->magic = cpu_to_be32(DRBD_MAGIC_100);
596 h->volume = cpu_to_be16(vnr);
597 h->command = cpu_to_be16(cmd);
598 h->length = cpu_to_be32(size);
600 return sizeof(struct p_header100);
603 static unsigned int prepare_header(struct drbd_tconn *tconn, int vnr,
604 void *buffer, enum drbd_packet cmd, int size)
606 if (tconn->agreed_pro_version >= 100)
607 return prepare_header100(buffer, cmd, size, vnr);
608 else if (tconn->agreed_pro_version >= 95 &&
609 size > DRBD_MAX_SIZE_H80_PACKET)
610 return prepare_header95(buffer, cmd, size);
612 return prepare_header80(buffer, cmd, size);
615 static void *__conn_prepare_command(struct drbd_tconn *tconn,
616 struct drbd_socket *sock)
620 return sock->sbuf + drbd_header_size(tconn);
623 void *conn_prepare_command(struct drbd_tconn *tconn, struct drbd_socket *sock)
627 mutex_lock(&sock->mutex);
628 p = __conn_prepare_command(tconn, sock);
630 mutex_unlock(&sock->mutex);
635 void *drbd_prepare_command(struct drbd_conf *mdev, struct drbd_socket *sock)
637 return conn_prepare_command(mdev->tconn, sock);
640 static int __send_command(struct drbd_tconn *tconn, int vnr,
641 struct drbd_socket *sock, enum drbd_packet cmd,
642 unsigned int header_size, void *data,
649 * Called with @data == NULL and the size of the data blocks in @size
650 * for commands that send data blocks. For those commands, omit the
651 * MSG_MORE flag: this will increase the likelihood that data blocks
652 * which are page aligned on the sender will end up page aligned on the
655 msg_flags = data ? MSG_MORE : 0;
657 header_size += prepare_header(tconn, vnr, sock->sbuf, cmd,
659 err = drbd_send_all(tconn, sock->socket, sock->sbuf, header_size,
662 err = drbd_send_all(tconn, sock->socket, data, size, 0);
666 static int __conn_send_command(struct drbd_tconn *tconn, struct drbd_socket *sock,
667 enum drbd_packet cmd, unsigned int header_size,
668 void *data, unsigned int size)
670 return __send_command(tconn, 0, sock, cmd, header_size, data, size);
673 int conn_send_command(struct drbd_tconn *tconn, struct drbd_socket *sock,
674 enum drbd_packet cmd, unsigned int header_size,
675 void *data, unsigned int size)
679 err = __conn_send_command(tconn, sock, cmd, header_size, data, size);
680 mutex_unlock(&sock->mutex);
684 int drbd_send_command(struct drbd_conf *mdev, struct drbd_socket *sock,
685 enum drbd_packet cmd, unsigned int header_size,
686 void *data, unsigned int size)
690 err = __send_command(mdev->tconn, mdev->vnr, sock, cmd, header_size,
692 mutex_unlock(&sock->mutex);
696 int drbd_send_ping(struct drbd_tconn *tconn)
698 struct drbd_socket *sock;
701 if (!conn_prepare_command(tconn, sock))
703 return conn_send_command(tconn, sock, P_PING, 0, NULL, 0);
706 int drbd_send_ping_ack(struct drbd_tconn *tconn)
708 struct drbd_socket *sock;
711 if (!conn_prepare_command(tconn, sock))
713 return conn_send_command(tconn, sock, P_PING_ACK, 0, NULL, 0);
716 int drbd_send_sync_param(struct drbd_conf *mdev)
718 struct drbd_socket *sock;
719 struct p_rs_param_95 *p;
721 const int apv = mdev->tconn->agreed_pro_version;
722 enum drbd_packet cmd;
724 struct disk_conf *dc;
726 sock = &mdev->tconn->data;
727 p = drbd_prepare_command(mdev, sock);
732 nc = rcu_dereference(mdev->tconn->net_conf);
734 size = apv <= 87 ? sizeof(struct p_rs_param)
735 : apv == 88 ? sizeof(struct p_rs_param)
736 + strlen(nc->verify_alg) + 1
737 : apv <= 94 ? sizeof(struct p_rs_param_89)
738 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
740 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
742 /* initialize verify_alg and csums_alg */
743 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
745 if (get_ldev(mdev)) {
746 dc = rcu_dereference(mdev->ldev->disk_conf);
747 p->resync_rate = cpu_to_be32(dc->resync_rate);
748 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
749 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
750 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
751 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
754 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
755 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
756 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
757 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
758 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
762 strcpy(p->verify_alg, nc->verify_alg);
764 strcpy(p->csums_alg, nc->csums_alg);
767 return drbd_send_command(mdev, sock, cmd, size, NULL, 0);
770 int __drbd_send_protocol(struct drbd_tconn *tconn, enum drbd_packet cmd)
772 struct drbd_socket *sock;
773 struct p_protocol *p;
778 p = __conn_prepare_command(tconn, sock);
783 nc = rcu_dereference(tconn->net_conf);
785 if (nc->tentative && tconn->agreed_pro_version < 92) {
787 mutex_unlock(&sock->mutex);
788 conn_err(tconn, "--dry-run is not supported by peer");
793 if (tconn->agreed_pro_version >= 87)
794 size += strlen(nc->integrity_alg) + 1;
796 p->protocol = cpu_to_be32(nc->wire_protocol);
797 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
798 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
799 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
800 p->two_primaries = cpu_to_be32(nc->two_primaries);
802 if (nc->discard_my_data)
803 cf |= CF_DISCARD_MY_DATA;
806 p->conn_flags = cpu_to_be32(cf);
808 if (tconn->agreed_pro_version >= 87)
809 strcpy(p->integrity_alg, nc->integrity_alg);
812 return __conn_send_command(tconn, sock, cmd, size, NULL, 0);
815 int drbd_send_protocol(struct drbd_tconn *tconn)
819 mutex_lock(&tconn->data.mutex);
820 err = __drbd_send_protocol(tconn, P_PROTOCOL);
821 mutex_unlock(&tconn->data.mutex);
826 int _drbd_send_uuids(struct drbd_conf *mdev, u64 uuid_flags)
828 struct drbd_socket *sock;
832 if (!get_ldev_if_state(mdev, D_NEGOTIATING))
835 sock = &mdev->tconn->data;
836 p = drbd_prepare_command(mdev, sock);
841 for (i = UI_CURRENT; i < UI_SIZE; i++)
842 p->uuid[i] = mdev->ldev ? cpu_to_be64(mdev->ldev->md.uuid[i]) : 0;
844 mdev->comm_bm_set = drbd_bm_total_weight(mdev);
845 p->uuid[UI_SIZE] = cpu_to_be64(mdev->comm_bm_set);
847 uuid_flags |= rcu_dereference(mdev->tconn->net_conf)->discard_my_data ? 1 : 0;
849 uuid_flags |= test_bit(CRASHED_PRIMARY, &mdev->flags) ? 2 : 0;
850 uuid_flags |= mdev->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
851 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
854 return drbd_send_command(mdev, sock, P_UUIDS, sizeof(*p), NULL, 0);
857 int drbd_send_uuids(struct drbd_conf *mdev)
859 return _drbd_send_uuids(mdev, 0);
862 int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev)
864 return _drbd_send_uuids(mdev, 8);
867 void drbd_print_uuids(struct drbd_conf *mdev, const char *text)
869 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
870 u64 *uuid = mdev->ldev->md.uuid;
871 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX\n",
873 (unsigned long long)uuid[UI_CURRENT],
874 (unsigned long long)uuid[UI_BITMAP],
875 (unsigned long long)uuid[UI_HISTORY_START],
876 (unsigned long long)uuid[UI_HISTORY_END]);
879 dev_info(DEV, "%s effective data uuid: %016llX\n",
881 (unsigned long long)mdev->ed_uuid);
885 void drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev)
887 struct drbd_socket *sock;
891 D_ASSERT(mdev->state.disk == D_UP_TO_DATE);
893 uuid = mdev->ldev->md.uuid[UI_BITMAP];
894 if (uuid && uuid != UUID_JUST_CREATED)
895 uuid = uuid + UUID_NEW_BM_OFFSET;
897 get_random_bytes(&uuid, sizeof(u64));
898 drbd_uuid_set(mdev, UI_BITMAP, uuid);
899 drbd_print_uuids(mdev, "updated sync UUID");
902 sock = &mdev->tconn->data;
903 p = drbd_prepare_command(mdev, sock);
905 p->uuid = cpu_to_be64(uuid);
906 drbd_send_command(mdev, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
910 int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags)
912 struct drbd_socket *sock;
914 sector_t d_size, u_size;
915 int q_order_type, max_bio_size;
917 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
918 D_ASSERT(mdev->ldev->backing_bdev);
919 d_size = drbd_get_max_capacity(mdev->ldev);
921 u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
923 q_order_type = drbd_queue_order_type(mdev);
924 max_bio_size = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
925 max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE);
930 q_order_type = QUEUE_ORDERED_NONE;
931 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
934 sock = &mdev->tconn->data;
935 p = drbd_prepare_command(mdev, sock);
939 if (mdev->tconn->agreed_pro_version <= 94)
940 max_bio_size = min_t(int, max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
941 else if (mdev->tconn->agreed_pro_version < 100)
942 max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE_P95);
944 p->d_size = cpu_to_be64(d_size);
945 p->u_size = cpu_to_be64(u_size);
946 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(mdev->this_bdev));
947 p->max_bio_size = cpu_to_be32(max_bio_size);
948 p->queue_order_type = cpu_to_be16(q_order_type);
949 p->dds_flags = cpu_to_be16(flags);
950 return drbd_send_command(mdev, sock, P_SIZES, sizeof(*p), NULL, 0);
954 * drbd_send_current_state() - Sends the drbd state to the peer
955 * @mdev: DRBD device.
957 int drbd_send_current_state(struct drbd_conf *mdev)
959 struct drbd_socket *sock;
962 sock = &mdev->tconn->data;
963 p = drbd_prepare_command(mdev, sock);
966 p->state = cpu_to_be32(mdev->state.i); /* Within the send mutex */
967 return drbd_send_command(mdev, sock, P_STATE, sizeof(*p), NULL, 0);
971 * drbd_send_state() - After a state change, sends the new state to the peer
972 * @mdev: DRBD device.
973 * @state: the state to send, not necessarily the current state.
975 * Each state change queues an "after_state_ch" work, which will eventually
976 * send the resulting new state to the peer. If more state changes happen
977 * between queuing and processing of the after_state_ch work, we still
978 * want to send each intermediary state in the order it occurred.
980 int drbd_send_state(struct drbd_conf *mdev, union drbd_state state)
982 struct drbd_socket *sock;
985 sock = &mdev->tconn->data;
986 p = drbd_prepare_command(mdev, sock);
989 p->state = cpu_to_be32(state.i); /* Within the send mutex */
990 return drbd_send_command(mdev, sock, P_STATE, sizeof(*p), NULL, 0);
993 int drbd_send_state_req(struct drbd_conf *mdev, union drbd_state mask, union drbd_state val)
995 struct drbd_socket *sock;
996 struct p_req_state *p;
998 sock = &mdev->tconn->data;
999 p = drbd_prepare_command(mdev, sock);
1002 p->mask = cpu_to_be32(mask.i);
1003 p->val = cpu_to_be32(val.i);
1004 return drbd_send_command(mdev, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1007 int conn_send_state_req(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val)
1009 enum drbd_packet cmd;
1010 struct drbd_socket *sock;
1011 struct p_req_state *p;
1013 cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1014 sock = &tconn->data;
1015 p = conn_prepare_command(tconn, sock);
1018 p->mask = cpu_to_be32(mask.i);
1019 p->val = cpu_to_be32(val.i);
1020 return conn_send_command(tconn, sock, cmd, sizeof(*p), NULL, 0);
1023 void drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode)
1025 struct drbd_socket *sock;
1026 struct p_req_state_reply *p;
1028 sock = &mdev->tconn->meta;
1029 p = drbd_prepare_command(mdev, sock);
1031 p->retcode = cpu_to_be32(retcode);
1032 drbd_send_command(mdev, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1036 void conn_send_sr_reply(struct drbd_tconn *tconn, enum drbd_state_rv retcode)
1038 struct drbd_socket *sock;
1039 struct p_req_state_reply *p;
1040 enum drbd_packet cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1042 sock = &tconn->meta;
1043 p = conn_prepare_command(tconn, sock);
1045 p->retcode = cpu_to_be32(retcode);
1046 conn_send_command(tconn, sock, cmd, sizeof(*p), NULL, 0);
1050 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1052 BUG_ON(code & ~0xf);
1053 p->encoding = (p->encoding & ~0xf) | code;
1056 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1058 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1061 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1064 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1067 int fill_bitmap_rle_bits(struct drbd_conf *mdev,
1068 struct p_compressed_bm *p,
1070 struct bm_xfer_ctx *c)
1072 struct bitstream bs;
1073 unsigned long plain_bits;
1080 /* may we use this feature? */
1082 use_rle = rcu_dereference(mdev->tconn->net_conf)->use_rle;
1084 if (!use_rle || mdev->tconn->agreed_pro_version < 90)
1087 if (c->bit_offset >= c->bm_bits)
1088 return 0; /* nothing to do. */
1090 /* use at most thus many bytes */
1091 bitstream_init(&bs, p->code, size, 0);
1092 memset(p->code, 0, size);
1093 /* plain bits covered in this code string */
1096 /* p->encoding & 0x80 stores whether the first run length is set.
1097 * bit offset is implicit.
1098 * start with toggle == 2 to be able to tell the first iteration */
1101 /* see how much plain bits we can stuff into one packet
1102 * using RLE and VLI. */
1104 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(mdev, c->bit_offset)
1105 : _drbd_bm_find_next(mdev, c->bit_offset);
1108 rl = tmp - c->bit_offset;
1110 if (toggle == 2) { /* first iteration */
1112 /* the first checked bit was set,
1113 * store start value, */
1114 dcbp_set_start(p, 1);
1115 /* but skip encoding of zero run length */
1119 dcbp_set_start(p, 0);
1122 /* paranoia: catch zero runlength.
1123 * can only happen if bitmap is modified while we scan it. */
1125 dev_err(DEV, "unexpected zero runlength while encoding bitmap "
1126 "t:%u bo:%lu\n", toggle, c->bit_offset);
1130 bits = vli_encode_bits(&bs, rl);
1131 if (bits == -ENOBUFS) /* buffer full */
1134 dev_err(DEV, "error while encoding bitmap: %d\n", bits);
1140 c->bit_offset = tmp;
1141 } while (c->bit_offset < c->bm_bits);
1143 len = bs.cur.b - p->code + !!bs.cur.bit;
1145 if (plain_bits < (len << 3)) {
1146 /* incompressible with this method.
1147 * we need to rewind both word and bit position. */
1148 c->bit_offset -= plain_bits;
1149 bm_xfer_ctx_bit_to_word_offset(c);
1150 c->bit_offset = c->word_offset * BITS_PER_LONG;
1154 /* RLE + VLI was able to compress it just fine.
1155 * update c->word_offset. */
1156 bm_xfer_ctx_bit_to_word_offset(c);
1158 /* store pad_bits */
1159 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1165 * send_bitmap_rle_or_plain
1167 * Return 0 when done, 1 when another iteration is needed, and a negative error
1168 * code upon failure.
1171 send_bitmap_rle_or_plain(struct drbd_conf *mdev, struct bm_xfer_ctx *c)
1173 struct drbd_socket *sock = &mdev->tconn->data;
1174 unsigned int header_size = drbd_header_size(mdev->tconn);
1175 struct p_compressed_bm *p = sock->sbuf + header_size;
1178 len = fill_bitmap_rle_bits(mdev, p,
1179 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1184 dcbp_set_code(p, RLE_VLI_Bits);
1185 err = __send_command(mdev->tconn, mdev->vnr, sock,
1186 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1189 c->bytes[0] += header_size + sizeof(*p) + len;
1191 if (c->bit_offset >= c->bm_bits)
1194 /* was not compressible.
1195 * send a buffer full of plain text bits instead. */
1196 unsigned int data_size;
1197 unsigned long num_words;
1198 unsigned long *p = sock->sbuf + header_size;
1200 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1201 num_words = min_t(size_t, data_size / sizeof(*p),
1202 c->bm_words - c->word_offset);
1203 len = num_words * sizeof(*p);
1205 drbd_bm_get_lel(mdev, c->word_offset, num_words, p);
1206 err = __send_command(mdev->tconn, mdev->vnr, sock, P_BITMAP, len, NULL, 0);
1207 c->word_offset += num_words;
1208 c->bit_offset = c->word_offset * BITS_PER_LONG;
1211 c->bytes[1] += header_size + len;
1213 if (c->bit_offset > c->bm_bits)
1214 c->bit_offset = c->bm_bits;
1218 INFO_bm_xfer_stats(mdev, "send", c);
1226 /* See the comment at receive_bitmap() */
1227 static int _drbd_send_bitmap(struct drbd_conf *mdev)
1229 struct bm_xfer_ctx c;
1232 if (!expect(mdev->bitmap))
1235 if (get_ldev(mdev)) {
1236 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1237 dev_info(DEV, "Writing the whole bitmap, MDF_FullSync was set.\n");
1238 drbd_bm_set_all(mdev);
1239 if (drbd_bm_write(mdev)) {
1240 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1241 * but otherwise process as per normal - need to tell other
1242 * side that a full resync is required! */
1243 dev_err(DEV, "Failed to write bitmap to disk!\n");
1245 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
1252 c = (struct bm_xfer_ctx) {
1253 .bm_bits = drbd_bm_bits(mdev),
1254 .bm_words = drbd_bm_words(mdev),
1258 err = send_bitmap_rle_or_plain(mdev, &c);
1264 int drbd_send_bitmap(struct drbd_conf *mdev)
1266 struct drbd_socket *sock = &mdev->tconn->data;
1269 mutex_lock(&sock->mutex);
1271 err = !_drbd_send_bitmap(mdev);
1272 mutex_unlock(&sock->mutex);
1276 void drbd_send_b_ack(struct drbd_tconn *tconn, u32 barrier_nr, u32 set_size)
1278 struct drbd_socket *sock;
1279 struct p_barrier_ack *p;
1281 if (tconn->cstate < C_WF_REPORT_PARAMS)
1284 sock = &tconn->meta;
1285 p = conn_prepare_command(tconn, sock);
1288 p->barrier = barrier_nr;
1289 p->set_size = cpu_to_be32(set_size);
1290 conn_send_command(tconn, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1294 * _drbd_send_ack() - Sends an ack packet
1295 * @mdev: DRBD device.
1296 * @cmd: Packet command code.
1297 * @sector: sector, needs to be in big endian byte order
1298 * @blksize: size in byte, needs to be in big endian byte order
1299 * @block_id: Id, big endian byte order
1301 static int _drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1302 u64 sector, u32 blksize, u64 block_id)
1304 struct drbd_socket *sock;
1305 struct p_block_ack *p;
1307 if (mdev->state.conn < C_CONNECTED)
1310 sock = &mdev->tconn->meta;
1311 p = drbd_prepare_command(mdev, sock);
1315 p->block_id = block_id;
1316 p->blksize = blksize;
1317 p->seq_num = cpu_to_be32(atomic_inc_return(&mdev->packet_seq));
1318 return drbd_send_command(mdev, sock, cmd, sizeof(*p), NULL, 0);
1321 /* dp->sector and dp->block_id already/still in network byte order,
1322 * data_size is payload size according to dp->head,
1323 * and may need to be corrected for digest size. */
1324 void drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd,
1325 struct p_data *dp, int data_size)
1327 if (mdev->tconn->peer_integrity_tfm)
1328 data_size -= crypto_hash_digestsize(mdev->tconn->peer_integrity_tfm);
1329 _drbd_send_ack(mdev, cmd, dp->sector, cpu_to_be32(data_size),
1333 void drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd,
1334 struct p_block_req *rp)
1336 _drbd_send_ack(mdev, cmd, rp->sector, rp->blksize, rp->block_id);
1340 * drbd_send_ack() - Sends an ack packet
1341 * @mdev: DRBD device
1342 * @cmd: packet command code
1343 * @peer_req: peer request
1345 int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1346 struct drbd_peer_request *peer_req)
1348 return _drbd_send_ack(mdev, cmd,
1349 cpu_to_be64(peer_req->i.sector),
1350 cpu_to_be32(peer_req->i.size),
1351 peer_req->block_id);
1354 /* This function misuses the block_id field to signal if the blocks
1355 * are is sync or not. */
1356 int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd,
1357 sector_t sector, int blksize, u64 block_id)
1359 return _drbd_send_ack(mdev, cmd,
1360 cpu_to_be64(sector),
1361 cpu_to_be32(blksize),
1362 cpu_to_be64(block_id));
1365 int drbd_send_drequest(struct drbd_conf *mdev, int cmd,
1366 sector_t sector, int size, u64 block_id)
1368 struct drbd_socket *sock;
1369 struct p_block_req *p;
1371 sock = &mdev->tconn->data;
1372 p = drbd_prepare_command(mdev, sock);
1375 p->sector = cpu_to_be64(sector);
1376 p->block_id = block_id;
1377 p->blksize = cpu_to_be32(size);
1378 return drbd_send_command(mdev, sock, cmd, sizeof(*p), NULL, 0);
1381 int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, int size,
1382 void *digest, int digest_size, enum drbd_packet cmd)
1384 struct drbd_socket *sock;
1385 struct p_block_req *p;
1387 /* FIXME: Put the digest into the preallocated socket buffer. */
1389 sock = &mdev->tconn->data;
1390 p = drbd_prepare_command(mdev, sock);
1393 p->sector = cpu_to_be64(sector);
1394 p->block_id = ID_SYNCER /* unused */;
1395 p->blksize = cpu_to_be32(size);
1396 return drbd_send_command(mdev, sock, cmd, sizeof(*p),
1397 digest, digest_size);
1400 int drbd_send_ov_request(struct drbd_conf *mdev, sector_t sector, int size)
1402 struct drbd_socket *sock;
1403 struct p_block_req *p;
1405 sock = &mdev->tconn->data;
1406 p = drbd_prepare_command(mdev, sock);
1409 p->sector = cpu_to_be64(sector);
1410 p->block_id = ID_SYNCER /* unused */;
1411 p->blksize = cpu_to_be32(size);
1412 return drbd_send_command(mdev, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1415 /* called on sndtimeo
1416 * returns false if we should retry,
1417 * true if we think connection is dead
1419 static int we_should_drop_the_connection(struct drbd_tconn *tconn, struct socket *sock)
1422 /* long elapsed = (long)(jiffies - mdev->last_received); */
1424 drop_it = tconn->meta.socket == sock
1425 || !tconn->asender.task
1426 || get_t_state(&tconn->asender) != RUNNING
1427 || tconn->cstate < C_WF_REPORT_PARAMS;
1432 drop_it = !--tconn->ko_count;
1434 conn_err(tconn, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1435 current->comm, current->pid, tconn->ko_count);
1436 request_ping(tconn);
1439 return drop_it; /* && (mdev->state == R_PRIMARY) */;
1442 static void drbd_update_congested(struct drbd_tconn *tconn)
1444 struct sock *sk = tconn->data.socket->sk;
1445 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1446 set_bit(NET_CONGESTED, &tconn->flags);
1449 /* The idea of sendpage seems to be to put some kind of reference
1450 * to the page into the skb, and to hand it over to the NIC. In
1451 * this process get_page() gets called.
1453 * As soon as the page was really sent over the network put_page()
1454 * gets called by some part of the network layer. [ NIC driver? ]
1456 * [ get_page() / put_page() increment/decrement the count. If count
1457 * reaches 0 the page will be freed. ]
1459 * This works nicely with pages from FSs.
1460 * But this means that in protocol A we might signal IO completion too early!
1462 * In order not to corrupt data during a resync we must make sure
1463 * that we do not reuse our own buffer pages (EEs) to early, therefore
1464 * we have the net_ee list.
1466 * XFS seems to have problems, still, it submits pages with page_count == 0!
1467 * As a workaround, we disable sendpage on pages
1468 * with page_count == 0 or PageSlab.
1470 static int _drbd_no_send_page(struct drbd_conf *mdev, struct page *page,
1471 int offset, size_t size, unsigned msg_flags)
1473 struct socket *socket;
1477 socket = mdev->tconn->data.socket;
1478 addr = kmap(page) + offset;
1479 err = drbd_send_all(mdev->tconn, socket, addr, size, msg_flags);
1482 mdev->send_cnt += size >> 9;
1486 static int _drbd_send_page(struct drbd_conf *mdev, struct page *page,
1487 int offset, size_t size, unsigned msg_flags)
1489 struct socket *socket = mdev->tconn->data.socket;
1490 mm_segment_t oldfs = get_fs();
1494 /* e.g. XFS meta- & log-data is in slab pages, which have a
1495 * page_count of 0 and/or have PageSlab() set.
1496 * we cannot use send_page for those, as that does get_page();
1497 * put_page(); and would cause either a VM_BUG directly, or
1498 * __page_cache_release a page that would actually still be referenced
1499 * by someone, leading to some obscure delayed Oops somewhere else. */
1500 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1501 return _drbd_no_send_page(mdev, page, offset, size, msg_flags);
1503 msg_flags |= MSG_NOSIGNAL;
1504 drbd_update_congested(mdev->tconn);
1509 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1511 if (sent == -EAGAIN) {
1512 if (we_should_drop_the_connection(mdev->tconn, socket))
1516 dev_warn(DEV, "%s: size=%d len=%d sent=%d\n",
1517 __func__, (int)size, len, sent);
1524 } while (len > 0 /* THINK && mdev->cstate >= C_CONNECTED*/);
1526 clear_bit(NET_CONGESTED, &mdev->tconn->flags);
1530 mdev->send_cnt += size >> 9;
1535 static int _drbd_send_bio(struct drbd_conf *mdev, struct bio *bio)
1537 struct bio_vec *bvec;
1539 /* hint all but last page with MSG_MORE */
1540 bio_for_each_segment(bvec, bio, i) {
1543 err = _drbd_no_send_page(mdev, bvec->bv_page,
1544 bvec->bv_offset, bvec->bv_len,
1545 i == bio->bi_vcnt - 1 ? 0 : MSG_MORE);
1552 static int _drbd_send_zc_bio(struct drbd_conf *mdev, struct bio *bio)
1554 struct bio_vec *bvec;
1556 /* hint all but last page with MSG_MORE */
1557 bio_for_each_segment(bvec, bio, i) {
1560 err = _drbd_send_page(mdev, bvec->bv_page,
1561 bvec->bv_offset, bvec->bv_len,
1562 i == bio->bi_vcnt - 1 ? 0 : MSG_MORE);
1569 static int _drbd_send_zc_ee(struct drbd_conf *mdev,
1570 struct drbd_peer_request *peer_req)
1572 struct page *page = peer_req->pages;
1573 unsigned len = peer_req->i.size;
1576 /* hint all but last page with MSG_MORE */
1577 page_chain_for_each(page) {
1578 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1580 err = _drbd_send_page(mdev, page, 0, l,
1581 page_chain_next(page) ? MSG_MORE : 0);
1589 static u32 bio_flags_to_wire(struct drbd_conf *mdev, unsigned long bi_rw)
1591 if (mdev->tconn->agreed_pro_version >= 95)
1592 return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) |
1593 (bi_rw & REQ_FUA ? DP_FUA : 0) |
1594 (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) |
1595 (bi_rw & REQ_DISCARD ? DP_DISCARD : 0);
1597 return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0;
1600 /* Used to send write requests
1601 * R_PRIMARY -> Peer (P_DATA)
1603 int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req)
1605 struct drbd_socket *sock;
1607 unsigned int dp_flags = 0;
1611 sock = &mdev->tconn->data;
1612 p = drbd_prepare_command(mdev, sock);
1613 dgs = mdev->tconn->integrity_tfm ? crypto_hash_digestsize(mdev->tconn->integrity_tfm) : 0;
1617 p->sector = cpu_to_be64(req->i.sector);
1618 p->block_id = (unsigned long)req;
1619 p->seq_num = cpu_to_be32(atomic_inc_return(&mdev->packet_seq));
1620 dp_flags = bio_flags_to_wire(mdev, req->master_bio->bi_rw);
1621 if (mdev->state.conn >= C_SYNC_SOURCE &&
1622 mdev->state.conn <= C_PAUSED_SYNC_T)
1623 dp_flags |= DP_MAY_SET_IN_SYNC;
1624 if (mdev->tconn->agreed_pro_version >= 100) {
1625 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1626 dp_flags |= DP_SEND_RECEIVE_ACK;
1627 if (req->rq_state & RQ_EXP_WRITE_ACK)
1628 dp_flags |= DP_SEND_WRITE_ACK;
1630 p->dp_flags = cpu_to_be32(dp_flags);
1632 drbd_csum_bio(mdev, mdev->tconn->integrity_tfm, req->master_bio, p + 1);
1633 err = __send_command(mdev->tconn, mdev->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size);
1635 /* For protocol A, we have to memcpy the payload into
1636 * socket buffers, as we may complete right away
1637 * as soon as we handed it over to tcp, at which point the data
1638 * pages may become invalid.
1640 * For data-integrity enabled, we copy it as well, so we can be
1641 * sure that even if the bio pages may still be modified, it
1642 * won't change the data on the wire, thus if the digest checks
1643 * out ok after sending on this side, but does not fit on the
1644 * receiving side, we sure have detected corruption elsewhere.
1646 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs)
1647 err = _drbd_send_bio(mdev, req->master_bio);
1649 err = _drbd_send_zc_bio(mdev, req->master_bio);
1651 /* double check digest, sometimes buffers have been modified in flight. */
1652 if (dgs > 0 && dgs <= 64) {
1653 /* 64 byte, 512 bit, is the largest digest size
1654 * currently supported in kernel crypto. */
1655 unsigned char digest[64];
1656 drbd_csum_bio(mdev, mdev->tconn->integrity_tfm, req->master_bio, digest);
1657 if (memcmp(p + 1, digest, dgs)) {
1659 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1660 (unsigned long long)req->i.sector, req->i.size);
1662 } /* else if (dgs > 64) {
1663 ... Be noisy about digest too large ...
1666 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1671 /* answer packet, used to send data back for read requests:
1672 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1673 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1675 int drbd_send_block(struct drbd_conf *mdev, enum drbd_packet cmd,
1676 struct drbd_peer_request *peer_req)
1678 struct drbd_socket *sock;
1683 sock = &mdev->tconn->data;
1684 p = drbd_prepare_command(mdev, sock);
1686 dgs = mdev->tconn->integrity_tfm ? crypto_hash_digestsize(mdev->tconn->integrity_tfm) : 0;
1690 p->sector = cpu_to_be64(peer_req->i.sector);
1691 p->block_id = peer_req->block_id;
1692 p->seq_num = 0; /* unused */
1695 drbd_csum_ee(mdev, mdev->tconn->integrity_tfm, peer_req, p + 1);
1696 err = __send_command(mdev->tconn, mdev->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size);
1698 err = _drbd_send_zc_ee(mdev, peer_req);
1699 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1704 int drbd_send_out_of_sync(struct drbd_conf *mdev, struct drbd_request *req)
1706 struct drbd_socket *sock;
1707 struct p_block_desc *p;
1709 sock = &mdev->tconn->data;
1710 p = drbd_prepare_command(mdev, sock);
1713 p->sector = cpu_to_be64(req->i.sector);
1714 p->blksize = cpu_to_be32(req->i.size);
1715 return drbd_send_command(mdev, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1719 drbd_send distinguishes two cases:
1721 Packets sent via the data socket "sock"
1722 and packets sent via the meta data socket "msock"
1725 -----------------+-------------------------+------------------------------
1726 timeout conf.timeout / 2 conf.timeout / 2
1727 timeout action send a ping via msock Abort communication
1728 and close all sockets
1732 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1734 int drbd_send(struct drbd_tconn *tconn, struct socket *sock,
1735 void *buf, size_t size, unsigned msg_flags)
1744 /* THINK if (signal_pending) return ... ? */
1749 msg.msg_name = NULL;
1750 msg.msg_namelen = 0;
1751 msg.msg_control = NULL;
1752 msg.msg_controllen = 0;
1753 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1755 if (sock == tconn->data.socket) {
1757 tconn->ko_count = rcu_dereference(tconn->net_conf)->ko_count;
1759 drbd_update_congested(tconn);
1763 * tcp_sendmsg does _not_ use its size parameter at all ?
1765 * -EAGAIN on timeout, -EINTR on signal.
1768 * do we need to block DRBD_SIG if sock == &meta.socket ??
1769 * otherwise wake_asender() might interrupt some send_*Ack !
1771 rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
1772 if (rv == -EAGAIN) {
1773 if (we_should_drop_the_connection(tconn, sock))
1779 flush_signals(current);
1787 } while (sent < size);
1789 if (sock == tconn->data.socket)
1790 clear_bit(NET_CONGESTED, &tconn->flags);
1793 if (rv != -EAGAIN) {
1794 conn_err(tconn, "%s_sendmsg returned %d\n",
1795 sock == tconn->meta.socket ? "msock" : "sock",
1797 conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD);
1799 conn_request_state(tconn, NS(conn, C_TIMEOUT), CS_HARD);
1806 * drbd_send_all - Send an entire buffer
1808 * Returns 0 upon success and a negative error value otherwise.
1810 int drbd_send_all(struct drbd_tconn *tconn, struct socket *sock, void *buffer,
1811 size_t size, unsigned msg_flags)
1815 err = drbd_send(tconn, sock, buffer, size, msg_flags);
1823 static int drbd_open(struct block_device *bdev, fmode_t mode)
1825 struct drbd_conf *mdev = bdev->bd_disk->private_data;
1826 unsigned long flags;
1829 mutex_lock(&drbd_main_mutex);
1830 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
1831 /* to have a stable mdev->state.role
1832 * and no race with updating open_cnt */
1834 if (mdev->state.role != R_PRIMARY) {
1835 if (mode & FMODE_WRITE)
1837 else if (!allow_oos)
1843 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
1844 mutex_unlock(&drbd_main_mutex);
1849 static int drbd_release(struct gendisk *gd, fmode_t mode)
1851 struct drbd_conf *mdev = gd->private_data;
1852 mutex_lock(&drbd_main_mutex);
1854 mutex_unlock(&drbd_main_mutex);
1858 static void drbd_set_defaults(struct drbd_conf *mdev)
1860 /* Beware! The actual layout differs
1861 * between big endian and little endian */
1862 mdev->state = (union drbd_dev_state) {
1863 { .role = R_SECONDARY,
1865 .conn = C_STANDALONE,
1871 void drbd_init_set_defaults(struct drbd_conf *mdev)
1873 /* the memset(,0,) did most of this.
1874 * note: only assignments, no allocation in here */
1876 drbd_set_defaults(mdev);
1878 atomic_set(&mdev->ap_bio_cnt, 0);
1879 atomic_set(&mdev->ap_pending_cnt, 0);
1880 atomic_set(&mdev->rs_pending_cnt, 0);
1881 atomic_set(&mdev->unacked_cnt, 0);
1882 atomic_set(&mdev->local_cnt, 0);
1883 atomic_set(&mdev->pp_in_use_by_net, 0);
1884 atomic_set(&mdev->rs_sect_in, 0);
1885 atomic_set(&mdev->rs_sect_ev, 0);
1886 atomic_set(&mdev->ap_in_flight, 0);
1887 atomic_set(&mdev->md_io_in_use, 0);
1889 mutex_init(&mdev->own_state_mutex);
1890 mdev->state_mutex = &mdev->own_state_mutex;
1892 spin_lock_init(&mdev->al_lock);
1893 spin_lock_init(&mdev->peer_seq_lock);
1895 INIT_LIST_HEAD(&mdev->active_ee);
1896 INIT_LIST_HEAD(&mdev->sync_ee);
1897 INIT_LIST_HEAD(&mdev->done_ee);
1898 INIT_LIST_HEAD(&mdev->read_ee);
1899 INIT_LIST_HEAD(&mdev->net_ee);
1900 INIT_LIST_HEAD(&mdev->resync_reads);
1901 INIT_LIST_HEAD(&mdev->resync_work.list);
1902 INIT_LIST_HEAD(&mdev->unplug_work.list);
1903 INIT_LIST_HEAD(&mdev->go_diskless.list);
1904 INIT_LIST_HEAD(&mdev->md_sync_work.list);
1905 INIT_LIST_HEAD(&mdev->start_resync_work.list);
1906 INIT_LIST_HEAD(&mdev->bm_io_work.w.list);
1908 mdev->resync_work.cb = w_resync_timer;
1909 mdev->unplug_work.cb = w_send_write_hint;
1910 mdev->go_diskless.cb = w_go_diskless;
1911 mdev->md_sync_work.cb = w_md_sync;
1912 mdev->bm_io_work.w.cb = w_bitmap_io;
1913 mdev->start_resync_work.cb = w_start_resync;
1915 mdev->resync_work.mdev = mdev;
1916 mdev->unplug_work.mdev = mdev;
1917 mdev->go_diskless.mdev = mdev;
1918 mdev->md_sync_work.mdev = mdev;
1919 mdev->bm_io_work.w.mdev = mdev;
1920 mdev->start_resync_work.mdev = mdev;
1922 init_timer(&mdev->resync_timer);
1923 init_timer(&mdev->md_sync_timer);
1924 init_timer(&mdev->start_resync_timer);
1925 init_timer(&mdev->request_timer);
1926 mdev->resync_timer.function = resync_timer_fn;
1927 mdev->resync_timer.data = (unsigned long) mdev;
1928 mdev->md_sync_timer.function = md_sync_timer_fn;
1929 mdev->md_sync_timer.data = (unsigned long) mdev;
1930 mdev->start_resync_timer.function = start_resync_timer_fn;
1931 mdev->start_resync_timer.data = (unsigned long) mdev;
1932 mdev->request_timer.function = request_timer_fn;
1933 mdev->request_timer.data = (unsigned long) mdev;
1935 init_waitqueue_head(&mdev->misc_wait);
1936 init_waitqueue_head(&mdev->state_wait);
1937 init_waitqueue_head(&mdev->ee_wait);
1938 init_waitqueue_head(&mdev->al_wait);
1939 init_waitqueue_head(&mdev->seq_wait);
1941 mdev->resync_wenr = LC_FREE;
1942 mdev->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1943 mdev->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1946 void drbd_mdev_cleanup(struct drbd_conf *mdev)
1949 if (mdev->tconn->receiver.t_state != NONE)
1950 dev_err(DEV, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
1951 mdev->tconn->receiver.t_state);
1962 mdev->rs_failed = 0;
1963 mdev->rs_last_events = 0;
1964 mdev->rs_last_sect_ev = 0;
1965 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1966 mdev->rs_mark_left[i] = 0;
1967 mdev->rs_mark_time[i] = 0;
1969 D_ASSERT(mdev->tconn->net_conf == NULL);
1971 drbd_set_my_capacity(mdev, 0);
1973 /* maybe never allocated. */
1974 drbd_bm_resize(mdev, 0, 1);
1975 drbd_bm_cleanup(mdev);
1978 drbd_free_bc(mdev->ldev);
1981 clear_bit(AL_SUSPENDED, &mdev->flags);
1983 D_ASSERT(list_empty(&mdev->active_ee));
1984 D_ASSERT(list_empty(&mdev->sync_ee));
1985 D_ASSERT(list_empty(&mdev->done_ee));
1986 D_ASSERT(list_empty(&mdev->read_ee));
1987 D_ASSERT(list_empty(&mdev->net_ee));
1988 D_ASSERT(list_empty(&mdev->resync_reads));
1989 D_ASSERT(list_empty(&mdev->tconn->sender_work.q));
1990 D_ASSERT(list_empty(&mdev->resync_work.list));
1991 D_ASSERT(list_empty(&mdev->unplug_work.list));
1992 D_ASSERT(list_empty(&mdev->go_diskless.list));
1994 drbd_set_defaults(mdev);
1998 static void drbd_destroy_mempools(void)
2002 while (drbd_pp_pool) {
2003 page = drbd_pp_pool;
2004 drbd_pp_pool = (struct page *)page_private(page);
2009 /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */
2011 if (drbd_md_io_bio_set)
2012 bioset_free(drbd_md_io_bio_set);
2013 if (drbd_md_io_page_pool)
2014 mempool_destroy(drbd_md_io_page_pool);
2015 if (drbd_ee_mempool)
2016 mempool_destroy(drbd_ee_mempool);
2017 if (drbd_request_mempool)
2018 mempool_destroy(drbd_request_mempool);
2020 kmem_cache_destroy(drbd_ee_cache);
2021 if (drbd_request_cache)
2022 kmem_cache_destroy(drbd_request_cache);
2023 if (drbd_bm_ext_cache)
2024 kmem_cache_destroy(drbd_bm_ext_cache);
2025 if (drbd_al_ext_cache)
2026 kmem_cache_destroy(drbd_al_ext_cache);
2028 drbd_md_io_bio_set = NULL;
2029 drbd_md_io_page_pool = NULL;
2030 drbd_ee_mempool = NULL;
2031 drbd_request_mempool = NULL;
2032 drbd_ee_cache = NULL;
2033 drbd_request_cache = NULL;
2034 drbd_bm_ext_cache = NULL;
2035 drbd_al_ext_cache = NULL;
2040 static int drbd_create_mempools(void)
2043 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
2046 /* prepare our caches and mempools */
2047 drbd_request_mempool = NULL;
2048 drbd_ee_cache = NULL;
2049 drbd_request_cache = NULL;
2050 drbd_bm_ext_cache = NULL;
2051 drbd_al_ext_cache = NULL;
2052 drbd_pp_pool = NULL;
2053 drbd_md_io_page_pool = NULL;
2054 drbd_md_io_bio_set = NULL;
2057 drbd_request_cache = kmem_cache_create(
2058 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2059 if (drbd_request_cache == NULL)
2062 drbd_ee_cache = kmem_cache_create(
2063 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2064 if (drbd_ee_cache == NULL)
2067 drbd_bm_ext_cache = kmem_cache_create(
2068 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2069 if (drbd_bm_ext_cache == NULL)
2072 drbd_al_ext_cache = kmem_cache_create(
2073 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2074 if (drbd_al_ext_cache == NULL)
2078 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2079 if (drbd_md_io_bio_set == NULL)
2082 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2083 if (drbd_md_io_page_pool == NULL)
2086 drbd_request_mempool = mempool_create(number,
2087 mempool_alloc_slab, mempool_free_slab, drbd_request_cache);
2088 if (drbd_request_mempool == NULL)
2091 drbd_ee_mempool = mempool_create(number,
2092 mempool_alloc_slab, mempool_free_slab, drbd_ee_cache);
2093 if (drbd_ee_mempool == NULL)
2096 /* drbd's page pool */
2097 spin_lock_init(&drbd_pp_lock);
2099 for (i = 0; i < number; i++) {
2100 page = alloc_page(GFP_HIGHUSER);
2103 set_page_private(page, (unsigned long)drbd_pp_pool);
2104 drbd_pp_pool = page;
2106 drbd_pp_vacant = number;
2111 drbd_destroy_mempools(); /* in case we allocated some */
2115 static int drbd_notify_sys(struct notifier_block *this, unsigned long code,
2118 /* just so we have it. you never know what interesting things we
2119 * might want to do here some day...
2125 static struct notifier_block drbd_notifier = {
2126 .notifier_call = drbd_notify_sys,
2129 static void drbd_release_all_peer_reqs(struct drbd_conf *mdev)
2133 rr = drbd_free_peer_reqs(mdev, &mdev->active_ee);
2135 dev_err(DEV, "%d EEs in active list found!\n", rr);
2137 rr = drbd_free_peer_reqs(mdev, &mdev->sync_ee);
2139 dev_err(DEV, "%d EEs in sync list found!\n", rr);
2141 rr = drbd_free_peer_reqs(mdev, &mdev->read_ee);
2143 dev_err(DEV, "%d EEs in read list found!\n", rr);
2145 rr = drbd_free_peer_reqs(mdev, &mdev->done_ee);
2147 dev_err(DEV, "%d EEs in done list found!\n", rr);
2149 rr = drbd_free_peer_reqs(mdev, &mdev->net_ee);
2151 dev_err(DEV, "%d EEs in net list found!\n", rr);
2154 /* caution. no locking. */
2155 void drbd_minor_destroy(struct kref *kref)
2157 struct drbd_conf *mdev = container_of(kref, struct drbd_conf, kref);
2158 struct drbd_tconn *tconn = mdev->tconn;
2160 del_timer_sync(&mdev->request_timer);
2162 /* paranoia asserts */
2163 D_ASSERT(mdev->open_cnt == 0);
2164 /* end paranoia asserts */
2166 /* cleanup stuff that may have been allocated during
2167 * device (re-)configuration or state changes */
2169 if (mdev->this_bdev)
2170 bdput(mdev->this_bdev);
2172 drbd_free_bc(mdev->ldev);
2175 drbd_release_all_peer_reqs(mdev);
2177 lc_destroy(mdev->act_log);
2178 lc_destroy(mdev->resync);
2180 kfree(mdev->p_uuid);
2181 /* mdev->p_uuid = NULL; */
2183 if (mdev->bitmap) /* should no longer be there. */
2184 drbd_bm_cleanup(mdev);
2185 __free_page(mdev->md_io_page);
2186 put_disk(mdev->vdisk);
2187 blk_cleanup_queue(mdev->rq_queue);
2188 kfree(mdev->rs_plan_s);
2191 kref_put(&tconn->kref, &conn_destroy);
2194 /* One global retry thread, if we need to push back some bio and have it
2195 * reinserted through our make request function.
2197 static struct retry_worker {
2198 struct workqueue_struct *wq;
2199 struct work_struct worker;
2202 struct list_head writes;
2205 static void do_retry(struct work_struct *ws)
2207 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2209 struct drbd_request *req, *tmp;
2211 spin_lock_irq(&retry->lock);
2212 list_splice_init(&retry->writes, &writes);
2213 spin_unlock_irq(&retry->lock);
2215 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2216 struct drbd_conf *mdev = req->w.mdev;
2217 struct bio *bio = req->master_bio;
2218 unsigned long start_time = req->start_time;
2220 /* We have exclusive access to this request object.
2221 * If it had not been RQ_POSTPONED, the code path which queued
2222 * it here would have completed and freed it already.
2224 mempool_free(req, drbd_request_mempool);
2226 /* A single suspended or otherwise blocking device may stall
2227 * all others as well. Fortunately, this code path is to
2228 * recover from a situation that "should not happen":
2229 * concurrent writes in multi-primary setup.
2230 * In a "normal" lifecycle, this workqueue is supposed to be
2231 * destroyed without ever doing anything.
2232 * If it turns out to be an issue anyways, we can do per
2233 * resource (replication group) or per device (minor) retry
2234 * workqueues instead.
2237 /* We are not just doing generic_make_request(),
2238 * as we want to keep the start_time information. */
2241 } while(__drbd_make_request(mdev, bio, start_time));
2245 void drbd_restart_request(struct drbd_request *req)
2247 unsigned long flags;
2248 spin_lock_irqsave(&retry.lock, flags);
2249 list_move_tail(&req->tl_requests, &retry.writes);
2250 spin_unlock_irqrestore(&retry.lock, flags);
2252 /* Drop the extra reference that would otherwise
2253 * have been dropped by complete_master_bio.
2254 * do_retry() needs to grab a new one. */
2255 dec_ap_bio(req->w.mdev);
2257 queue_work(retry.wq, &retry.worker);
2261 static void drbd_cleanup(void)
2264 struct drbd_conf *mdev;
2265 struct drbd_tconn *tconn, *tmp;
2267 unregister_reboot_notifier(&drbd_notifier);
2269 /* first remove proc,
2270 * drbdsetup uses it's presence to detect
2271 * whether DRBD is loaded.
2272 * If we would get stuck in proc removal,
2273 * but have netlink already deregistered,
2274 * some drbdsetup commands may wait forever
2278 remove_proc_entry("drbd", NULL);
2281 destroy_workqueue(retry.wq);
2283 drbd_genl_unregister();
2285 idr_for_each_entry(&minors, mdev, i) {
2286 idr_remove(&minors, mdev_to_minor(mdev));
2287 idr_remove(&mdev->tconn->volumes, mdev->vnr);
2288 del_gendisk(mdev->vdisk);
2289 /* synchronize_rcu(); No other threads running at this point */
2290 kref_put(&mdev->kref, &drbd_minor_destroy);
2293 /* not _rcu since, no other updater anymore. Genl already unregistered */
2294 list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2295 list_del(&tconn->all_tconn); /* not _rcu no proc, not other threads */
2296 /* synchronize_rcu(); */
2297 kref_put(&tconn->kref, &conn_destroy);
2300 drbd_destroy_mempools();
2301 unregister_blkdev(DRBD_MAJOR, "drbd");
2303 idr_destroy(&minors);
2305 printk(KERN_INFO "drbd: module cleanup done.\n");
2309 * drbd_congested() - Callback for pdflush
2310 * @congested_data: User data
2311 * @bdi_bits: Bits pdflush is currently interested in
2313 * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2315 static int drbd_congested(void *congested_data, int bdi_bits)
2317 struct drbd_conf *mdev = congested_data;
2318 struct request_queue *q;
2322 if (!may_inc_ap_bio(mdev)) {
2323 /* DRBD has frozen IO */
2329 if (get_ldev(mdev)) {
2330 q = bdev_get_queue(mdev->ldev->backing_bdev);
2331 r = bdi_congested(&q->backing_dev_info, bdi_bits);
2337 if (bdi_bits & (1 << BDI_async_congested) && test_bit(NET_CONGESTED, &mdev->tconn->flags)) {
2338 r |= (1 << BDI_async_congested);
2339 reason = reason == 'b' ? 'a' : 'n';
2343 mdev->congestion_reason = reason;
2347 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2349 spin_lock_init(&wq->q_lock);
2350 INIT_LIST_HEAD(&wq->q);
2351 init_waitqueue_head(&wq->q_wait);
2354 struct drbd_tconn *conn_get_by_name(const char *name)
2356 struct drbd_tconn *tconn;
2358 if (!name || !name[0])
2362 list_for_each_entry_rcu(tconn, &drbd_tconns, all_tconn) {
2363 if (!strcmp(tconn->name, name)) {
2364 kref_get(&tconn->kref);
2374 struct drbd_tconn *conn_get_by_addrs(void *my_addr, int my_addr_len,
2375 void *peer_addr, int peer_addr_len)
2377 struct drbd_tconn *tconn;
2380 list_for_each_entry_rcu(tconn, &drbd_tconns, all_tconn) {
2381 if (tconn->my_addr_len == my_addr_len &&
2382 tconn->peer_addr_len == peer_addr_len &&
2383 !memcmp(&tconn->my_addr, my_addr, my_addr_len) &&
2384 !memcmp(&tconn->peer_addr, peer_addr, peer_addr_len)) {
2385 kref_get(&tconn->kref);
2395 static int drbd_alloc_socket(struct drbd_socket *socket)
2397 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2400 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2406 static void drbd_free_socket(struct drbd_socket *socket)
2408 free_page((unsigned long) socket->sbuf);
2409 free_page((unsigned long) socket->rbuf);
2412 void conn_free_crypto(struct drbd_tconn *tconn)
2414 drbd_free_sock(tconn);
2416 crypto_free_hash(tconn->csums_tfm);
2417 crypto_free_hash(tconn->verify_tfm);
2418 crypto_free_hash(tconn->cram_hmac_tfm);
2419 crypto_free_hash(tconn->integrity_tfm);
2420 crypto_free_hash(tconn->peer_integrity_tfm);
2421 kfree(tconn->int_dig_in);
2422 kfree(tconn->int_dig_vv);
2424 tconn->csums_tfm = NULL;
2425 tconn->verify_tfm = NULL;
2426 tconn->cram_hmac_tfm = NULL;
2427 tconn->integrity_tfm = NULL;
2428 tconn->peer_integrity_tfm = NULL;
2429 tconn->int_dig_in = NULL;
2430 tconn->int_dig_vv = NULL;
2433 int set_resource_options(struct drbd_tconn *tconn, struct res_opts *res_opts)
2435 cpumask_var_t new_cpu_mask;
2438 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2441 retcode = ERR_NOMEM;
2442 drbd_msg_put_info("unable to allocate cpumask");
2445 /* silently ignore cpu mask on UP kernel */
2446 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2447 /* FIXME: Get rid of constant 32 here */
2448 err = bitmap_parse(res_opts->cpu_mask, 32,
2449 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2451 conn_warn(tconn, "bitmap_parse() failed with %d\n", err);
2452 /* retcode = ERR_CPU_MASK_PARSE; */
2456 tconn->res_opts = *res_opts;
2457 if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2458 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2459 drbd_calc_cpu_mask(tconn);
2460 tconn->receiver.reset_cpu_mask = 1;
2461 tconn->asender.reset_cpu_mask = 1;
2462 tconn->worker.reset_cpu_mask = 1;
2467 free_cpumask_var(new_cpu_mask);
2472 /* caller must be under genl_lock() */
2473 struct drbd_tconn *conn_create(const char *name, struct res_opts *res_opts)
2475 struct drbd_tconn *tconn;
2477 tconn = kzalloc(sizeof(struct drbd_tconn), GFP_KERNEL);
2481 tconn->name = kstrdup(name, GFP_KERNEL);
2485 if (drbd_alloc_socket(&tconn->data))
2487 if (drbd_alloc_socket(&tconn->meta))
2490 if (!zalloc_cpumask_var(&tconn->cpu_mask, GFP_KERNEL))
2493 if (set_resource_options(tconn, res_opts))
2496 tconn->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2497 if (!tconn->current_epoch)
2500 INIT_LIST_HEAD(&tconn->transfer_log);
2502 INIT_LIST_HEAD(&tconn->current_epoch->list);
2504 spin_lock_init(&tconn->epoch_lock);
2505 tconn->write_ordering = WO_bdev_flush;
2507 tconn->send.seen_any_write_yet = false;
2508 tconn->send.current_epoch_nr = 0;
2509 tconn->send.current_epoch_writes = 0;
2511 tconn->cstate = C_STANDALONE;
2512 mutex_init(&tconn->cstate_mutex);
2513 spin_lock_init(&tconn->req_lock);
2514 mutex_init(&tconn->conf_update);
2515 init_waitqueue_head(&tconn->ping_wait);
2516 idr_init(&tconn->volumes);
2518 drbd_init_workqueue(&tconn->sender_work);
2519 mutex_init(&tconn->data.mutex);
2520 mutex_init(&tconn->meta.mutex);
2522 drbd_thread_init(tconn, &tconn->receiver, drbdd_init, "receiver");
2523 drbd_thread_init(tconn, &tconn->worker, drbd_worker, "worker");
2524 drbd_thread_init(tconn, &tconn->asender, drbd_asender, "asender");
2526 kref_init(&tconn->kref);
2527 list_add_tail_rcu(&tconn->all_tconn, &drbd_tconns);
2532 kfree(tconn->current_epoch);
2533 free_cpumask_var(tconn->cpu_mask);
2534 drbd_free_socket(&tconn->meta);
2535 drbd_free_socket(&tconn->data);
2542 void conn_destroy(struct kref *kref)
2544 struct drbd_tconn *tconn = container_of(kref, struct drbd_tconn, kref);
2546 if (atomic_read(&tconn->current_epoch->epoch_size) != 0)
2547 conn_err(tconn, "epoch_size:%d\n", atomic_read(&tconn->current_epoch->epoch_size));
2548 kfree(tconn->current_epoch);
2550 idr_destroy(&tconn->volumes);
2552 free_cpumask_var(tconn->cpu_mask);
2553 drbd_free_socket(&tconn->meta);
2554 drbd_free_socket(&tconn->data);
2556 kfree(tconn->int_dig_in);
2557 kfree(tconn->int_dig_vv);
2561 enum drbd_ret_code conn_new_minor(struct drbd_tconn *tconn, unsigned int minor, int vnr)
2563 struct drbd_conf *mdev;
2564 struct gendisk *disk;
2565 struct request_queue *q;
2567 int minor_got = minor;
2568 enum drbd_ret_code err = ERR_NOMEM;
2570 mdev = minor_to_mdev(minor);
2572 return ERR_MINOR_EXISTS;
2574 /* GFP_KERNEL, we are outside of all write-out paths */
2575 mdev = kzalloc(sizeof(struct drbd_conf), GFP_KERNEL);
2579 kref_get(&tconn->kref);
2580 mdev->tconn = tconn;
2582 mdev->minor = minor;
2585 drbd_init_set_defaults(mdev);
2587 q = blk_alloc_queue(GFP_KERNEL);
2591 q->queuedata = mdev;
2593 disk = alloc_disk(1);
2598 set_disk_ro(disk, true);
2601 disk->major = DRBD_MAJOR;
2602 disk->first_minor = minor;
2603 disk->fops = &drbd_ops;
2604 sprintf(disk->disk_name, "drbd%d", minor);
2605 disk->private_data = mdev;
2607 mdev->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2608 /* we have no partitions. we contain only ourselves. */
2609 mdev->this_bdev->bd_contains = mdev->this_bdev;
2611 q->backing_dev_info.congested_fn = drbd_congested;
2612 q->backing_dev_info.congested_data = mdev;
2614 blk_queue_make_request(q, drbd_make_request);
2615 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2616 This triggers a max_bio_size message upon first attach or connect */
2617 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2618 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2619 blk_queue_merge_bvec(q, drbd_merge_bvec);
2620 q->queue_lock = &mdev->tconn->req_lock; /* needed since we use */
2622 mdev->md_io_page = alloc_page(GFP_KERNEL);
2623 if (!mdev->md_io_page)
2624 goto out_no_io_page;
2626 if (drbd_bm_init(mdev))
2628 mdev->read_requests = RB_ROOT;
2629 mdev->write_requests = RB_ROOT;
2631 if (!idr_pre_get(&minors, GFP_KERNEL))
2632 goto out_no_minor_idr;
2633 if (idr_get_new_above(&minors, mdev, minor, &minor_got))
2634 goto out_no_minor_idr;
2635 if (minor_got != minor) {
2636 err = ERR_MINOR_EXISTS;
2637 drbd_msg_put_info("requested minor exists already");
2638 goto out_idr_remove_minor;
2641 if (!idr_pre_get(&tconn->volumes, GFP_KERNEL))
2642 goto out_idr_remove_minor;
2643 if (idr_get_new_above(&tconn->volumes, mdev, vnr, &vnr_got))
2644 goto out_idr_remove_minor;
2645 if (vnr_got != vnr) {
2646 err = ERR_INVALID_REQUEST;
2647 drbd_msg_put_info("requested volume exists already");
2648 goto out_idr_remove_vol;
2651 kref_init(&mdev->kref); /* one ref for both idrs and the the add_disk */
2653 /* inherit the connection state */
2654 mdev->state.conn = tconn->cstate;
2655 if (mdev->state.conn == C_WF_REPORT_PARAMS)
2656 drbd_connected(mdev);
2661 idr_remove(&tconn->volumes, vnr_got);
2662 out_idr_remove_minor:
2663 idr_remove(&minors, minor_got);
2666 drbd_bm_cleanup(mdev);
2668 __free_page(mdev->md_io_page);
2672 blk_cleanup_queue(q);
2675 kref_put(&tconn->kref, &conn_destroy);
2679 int __init drbd_init(void)
2683 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2685 "drbd: invalid minor_count (%d)\n", minor_count);
2689 minor_count = DRBD_MINOR_COUNT_DEF;
2693 err = register_blkdev(DRBD_MAJOR, "drbd");
2696 "drbd: unable to register block device major %d\n",
2701 err = drbd_genl_register();
2703 printk(KERN_ERR "drbd: unable to register generic netlink family\n");
2708 register_reboot_notifier(&drbd_notifier);
2711 * allocate all necessary structs
2715 init_waitqueue_head(&drbd_pp_wait);
2717 drbd_proc = NULL; /* play safe for drbd_cleanup */
2720 err = drbd_create_mempools();
2724 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
2726 printk(KERN_ERR "drbd: unable to register proc file\n");
2730 rwlock_init(&global_state_lock);
2731 INIT_LIST_HEAD(&drbd_tconns);
2733 retry.wq = create_singlethread_workqueue("drbd-reissue");
2735 printk(KERN_ERR "drbd: unable to create retry workqueue\n");
2738 INIT_WORK(&retry.worker, do_retry);
2739 spin_lock_init(&retry.lock);
2740 INIT_LIST_HEAD(&retry.writes);
2742 printk(KERN_INFO "drbd: initialized. "
2743 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2744 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2745 printk(KERN_INFO "drbd: %s\n", drbd_buildtag());
2746 printk(KERN_INFO "drbd: registered as block device major %d\n",
2749 return 0; /* Success! */
2754 /* currently always the case */
2755 printk(KERN_ERR "drbd: ran out of memory\n");
2757 printk(KERN_ERR "drbd: initialization failure\n");
2761 void drbd_free_bc(struct drbd_backing_dev *ldev)
2766 blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2767 blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2772 void drbd_free_sock(struct drbd_tconn *tconn)
2774 if (tconn->data.socket) {
2775 mutex_lock(&tconn->data.mutex);
2776 kernel_sock_shutdown(tconn->data.socket, SHUT_RDWR);
2777 sock_release(tconn->data.socket);
2778 tconn->data.socket = NULL;
2779 mutex_unlock(&tconn->data.mutex);
2781 if (tconn->meta.socket) {
2782 mutex_lock(&tconn->meta.mutex);
2783 kernel_sock_shutdown(tconn->meta.socket, SHUT_RDWR);
2784 sock_release(tconn->meta.socket);
2785 tconn->meta.socket = NULL;
2786 mutex_unlock(&tconn->meta.mutex);
2790 /* meta data management */
2792 struct meta_data_on_disk {
2793 u64 la_size; /* last agreed size. */
2794 u64 uuid[UI_SIZE]; /* UUIDs. */
2797 u32 flags; /* MDF */
2800 u32 al_offset; /* offset to this block */
2801 u32 al_nr_extents; /* important for restoring the AL */
2802 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2803 u32 bm_offset; /* offset to the bitmap, from here */
2804 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
2805 u32 la_peer_max_bio_size; /* last peer max_bio_size */
2806 u32 reserved_u32[3];
2811 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
2812 * @mdev: DRBD device.
2814 void drbd_md_sync(struct drbd_conf *mdev)
2816 struct meta_data_on_disk *buffer;
2820 del_timer(&mdev->md_sync_timer);
2821 /* timer may be rearmed by drbd_md_mark_dirty() now. */
2822 if (!test_and_clear_bit(MD_DIRTY, &mdev->flags))
2825 /* We use here D_FAILED and not D_ATTACHING because we try to write
2826 * metadata even if we detach due to a disk failure! */
2827 if (!get_ldev_if_state(mdev, D_FAILED))
2830 buffer = drbd_md_get_buffer(mdev);
2834 memset(buffer, 0, 512);
2836 buffer->la_size = cpu_to_be64(drbd_get_capacity(mdev->this_bdev));
2837 for (i = UI_CURRENT; i < UI_SIZE; i++)
2838 buffer->uuid[i] = cpu_to_be64(mdev->ldev->md.uuid[i]);
2839 buffer->flags = cpu_to_be32(mdev->ldev->md.flags);
2840 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
2842 buffer->md_size_sect = cpu_to_be32(mdev->ldev->md.md_size_sect);
2843 buffer->al_offset = cpu_to_be32(mdev->ldev->md.al_offset);
2844 buffer->al_nr_extents = cpu_to_be32(mdev->act_log->nr_elements);
2845 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
2846 buffer->device_uuid = cpu_to_be64(mdev->ldev->md.device_uuid);
2848 buffer->bm_offset = cpu_to_be32(mdev->ldev->md.bm_offset);
2849 buffer->la_peer_max_bio_size = cpu_to_be32(mdev->peer_max_bio_size);
2851 D_ASSERT(drbd_md_ss__(mdev, mdev->ldev) == mdev->ldev->md.md_offset);
2852 sector = mdev->ldev->md.md_offset;
2854 if (drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) {
2855 /* this was a try anyways ... */
2856 dev_err(DEV, "meta data update failed!\n");
2857 drbd_chk_io_error(mdev, 1, true);
2860 /* Update mdev->ldev->md.la_size_sect,
2861 * since we updated it on metadata. */
2862 mdev->ldev->md.la_size_sect = drbd_get_capacity(mdev->this_bdev);
2864 drbd_md_put_buffer(mdev);
2870 * drbd_md_read() - Reads in the meta data super block
2871 * @mdev: DRBD device.
2872 * @bdev: Device from which the meta data should be read in.
2874 * Return 0 (NO_ERROR) on success, and an enum drbd_ret_code in case
2875 * something goes wrong.
2877 int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
2879 struct meta_data_on_disk *buffer;
2881 int i, rv = NO_ERROR;
2883 if (!get_ldev_if_state(mdev, D_ATTACHING))
2884 return ERR_IO_MD_DISK;
2886 buffer = drbd_md_get_buffer(mdev);
2890 if (drbd_md_sync_page_io(mdev, bdev, bdev->md.md_offset, READ)) {
2891 /* NOTE: can't do normal error processing here as this is
2892 called BEFORE disk is attached */
2893 dev_err(DEV, "Error while reading metadata.\n");
2894 rv = ERR_IO_MD_DISK;
2898 magic = be32_to_cpu(buffer->magic);
2899 flags = be32_to_cpu(buffer->flags);
2900 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
2901 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
2902 /* btw: that's Activity Log clean, not "all" clean. */
2903 dev_err(DEV, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
2904 rv = ERR_MD_UNCLEAN;
2907 if (magic != DRBD_MD_MAGIC_08) {
2908 if (magic == DRBD_MD_MAGIC_07)
2909 dev_err(DEV, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
2911 dev_err(DEV, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
2912 rv = ERR_MD_INVALID;
2915 if (be32_to_cpu(buffer->al_offset) != bdev->md.al_offset) {
2916 dev_err(DEV, "unexpected al_offset: %d (expected %d)\n",
2917 be32_to_cpu(buffer->al_offset), bdev->md.al_offset);
2918 rv = ERR_MD_INVALID;
2921 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
2922 dev_err(DEV, "unexpected bm_offset: %d (expected %d)\n",
2923 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
2924 rv = ERR_MD_INVALID;
2927 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
2928 dev_err(DEV, "unexpected md_size: %u (expected %u)\n",
2929 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
2930 rv = ERR_MD_INVALID;
2934 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
2935 dev_err(DEV, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
2936 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
2937 rv = ERR_MD_INVALID;
2941 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size);
2942 for (i = UI_CURRENT; i < UI_SIZE; i++)
2943 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
2944 bdev->md.flags = be32_to_cpu(buffer->flags);
2945 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
2947 spin_lock_irq(&mdev->tconn->req_lock);
2948 if (mdev->state.conn < C_CONNECTED) {
2950 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
2951 peer = max_t(int, peer, DRBD_MAX_BIO_SIZE_SAFE);
2952 mdev->peer_max_bio_size = peer;
2954 spin_unlock_irq(&mdev->tconn->req_lock);
2957 drbd_md_put_buffer(mdev);
2965 * drbd_md_mark_dirty() - Mark meta data super block as dirty
2966 * @mdev: DRBD device.
2968 * Call this function if you change anything that should be written to
2969 * the meta-data super block. This function sets MD_DIRTY, and starts a
2970 * timer that ensures that within five seconds you have to call drbd_md_sync().
2973 void drbd_md_mark_dirty_(struct drbd_conf *mdev, unsigned int line, const char *func)
2975 if (!test_and_set_bit(MD_DIRTY, &mdev->flags)) {
2976 mod_timer(&mdev->md_sync_timer, jiffies + HZ);
2977 mdev->last_md_mark_dirty.line = line;
2978 mdev->last_md_mark_dirty.func = func;
2982 void drbd_md_mark_dirty(struct drbd_conf *mdev)
2984 if (!test_and_set_bit(MD_DIRTY, &mdev->flags))
2985 mod_timer(&mdev->md_sync_timer, jiffies + 5*HZ);
2989 static void drbd_uuid_move_history(struct drbd_conf *mdev) __must_hold(local)
2993 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
2994 mdev->ldev->md.uuid[i+1] = mdev->ldev->md.uuid[i];
2997 void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
2999 if (idx == UI_CURRENT) {
3000 if (mdev->state.role == R_PRIMARY)
3005 drbd_set_ed_uuid(mdev, val);
3008 mdev->ldev->md.uuid[idx] = val;
3009 drbd_md_mark_dirty(mdev);
3013 void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
3015 if (mdev->ldev->md.uuid[idx]) {
3016 drbd_uuid_move_history(mdev);
3017 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[idx];
3019 _drbd_uuid_set(mdev, idx, val);
3023 * drbd_uuid_new_current() - Creates a new current UUID
3024 * @mdev: DRBD device.
3026 * Creates a new current UUID, and rotates the old current UUID into
3027 * the bitmap slot. Causes an incremental resync upon next connect.
3029 void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local)
3032 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
3035 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
3037 mdev->ldev->md.uuid[UI_BITMAP] = mdev->ldev->md.uuid[UI_CURRENT];
3039 get_random_bytes(&val, sizeof(u64));
3040 _drbd_uuid_set(mdev, UI_CURRENT, val);
3041 drbd_print_uuids(mdev, "new current UUID");
3042 /* get it to stable storage _now_ */
3046 void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local)
3048 if (mdev->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3052 drbd_uuid_move_history(mdev);
3053 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP];
3054 mdev->ldev->md.uuid[UI_BITMAP] = 0;
3056 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
3058 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
3060 mdev->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3062 drbd_md_mark_dirty(mdev);
3066 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3067 * @mdev: DRBD device.
3069 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3071 int drbd_bmio_set_n_write(struct drbd_conf *mdev)
3075 if (get_ldev_if_state(mdev, D_ATTACHING)) {
3076 drbd_md_set_flag(mdev, MDF_FULL_SYNC);
3078 drbd_bm_set_all(mdev);
3080 rv = drbd_bm_write(mdev);
3083 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
3094 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3095 * @mdev: DRBD device.
3097 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3099 int drbd_bmio_clear_n_write(struct drbd_conf *mdev)
3103 drbd_resume_al(mdev);
3104 if (get_ldev_if_state(mdev, D_ATTACHING)) {
3105 drbd_bm_clear_all(mdev);
3106 rv = drbd_bm_write(mdev);
3113 static int w_bitmap_io(struct drbd_work *w, int unused)
3115 struct bm_io_work *work = container_of(w, struct bm_io_work, w);
3116 struct drbd_conf *mdev = w->mdev;
3119 D_ASSERT(atomic_read(&mdev->ap_bio_cnt) == 0);
3121 if (get_ldev(mdev)) {
3122 drbd_bm_lock(mdev, work->why, work->flags);
3123 rv = work->io_fn(mdev);
3124 drbd_bm_unlock(mdev);
3128 clear_bit_unlock(BITMAP_IO, &mdev->flags);
3129 wake_up(&mdev->misc_wait);
3132 work->done(mdev, rv);
3134 clear_bit(BITMAP_IO_QUEUED, &mdev->flags);
3141 void drbd_ldev_destroy(struct drbd_conf *mdev)
3143 lc_destroy(mdev->resync);
3144 mdev->resync = NULL;
3145 lc_destroy(mdev->act_log);
3146 mdev->act_log = NULL;
3148 drbd_free_bc(mdev->ldev);
3149 mdev->ldev = NULL;);
3151 clear_bit(GO_DISKLESS, &mdev->flags);
3154 static int w_go_diskless(struct drbd_work *w, int unused)
3156 struct drbd_conf *mdev = w->mdev;
3158 D_ASSERT(mdev->state.disk == D_FAILED);
3159 /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will
3160 * inc/dec it frequently. Once we are D_DISKLESS, no one will touch
3161 * the protected members anymore, though, so once put_ldev reaches zero
3162 * again, it will be safe to free them. */
3163 drbd_force_state(mdev, NS(disk, D_DISKLESS));
3167 void drbd_go_diskless(struct drbd_conf *mdev)
3169 D_ASSERT(mdev->state.disk == D_FAILED);
3170 if (!test_and_set_bit(GO_DISKLESS, &mdev->flags))
3171 drbd_queue_work(&mdev->tconn->sender_work, &mdev->go_diskless);
3175 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3176 * @mdev: DRBD device.
3177 * @io_fn: IO callback to be called when bitmap IO is possible
3178 * @done: callback to be called after the bitmap IO was performed
3179 * @why: Descriptive text of the reason for doing the IO
3181 * While IO on the bitmap happens we freeze application IO thus we ensure
3182 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3183 * called from worker context. It MUST NOT be used while a previous such
3184 * work is still pending!
3186 void drbd_queue_bitmap_io(struct drbd_conf *mdev,
3187 int (*io_fn)(struct drbd_conf *),
3188 void (*done)(struct drbd_conf *, int),
3189 char *why, enum bm_flag flags)
3191 D_ASSERT(current == mdev->tconn->worker.task);
3193 D_ASSERT(!test_bit(BITMAP_IO_QUEUED, &mdev->flags));
3194 D_ASSERT(!test_bit(BITMAP_IO, &mdev->flags));
3195 D_ASSERT(list_empty(&mdev->bm_io_work.w.list));
3196 if (mdev->bm_io_work.why)
3197 dev_err(DEV, "FIXME going to queue '%s' but '%s' still pending?\n",
3198 why, mdev->bm_io_work.why);
3200 mdev->bm_io_work.io_fn = io_fn;
3201 mdev->bm_io_work.done = done;
3202 mdev->bm_io_work.why = why;
3203 mdev->bm_io_work.flags = flags;
3205 spin_lock_irq(&mdev->tconn->req_lock);
3206 set_bit(BITMAP_IO, &mdev->flags);
3207 if (atomic_read(&mdev->ap_bio_cnt) == 0) {
3208 if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags))
3209 drbd_queue_work(&mdev->tconn->sender_work, &mdev->bm_io_work.w);
3211 spin_unlock_irq(&mdev->tconn->req_lock);
3215 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3216 * @mdev: DRBD device.
3217 * @io_fn: IO callback to be called when bitmap IO is possible
3218 * @why: Descriptive text of the reason for doing the IO
3220 * freezes application IO while that the actual IO operations runs. This
3221 * functions MAY NOT be called from worker context.
3223 int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *),
3224 char *why, enum bm_flag flags)
3228 D_ASSERT(current != mdev->tconn->worker.task);
3230 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3231 drbd_suspend_io(mdev);
3233 drbd_bm_lock(mdev, why, flags);
3235 drbd_bm_unlock(mdev);
3237 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3238 drbd_resume_io(mdev);
3243 void drbd_md_set_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3245 if ((mdev->ldev->md.flags & flag) != flag) {
3246 drbd_md_mark_dirty(mdev);
3247 mdev->ldev->md.flags |= flag;
3251 void drbd_md_clear_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3253 if ((mdev->ldev->md.flags & flag) != 0) {
3254 drbd_md_mark_dirty(mdev);
3255 mdev->ldev->md.flags &= ~flag;
3258 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3260 return (bdev->md.flags & flag) != 0;
3263 static void md_sync_timer_fn(unsigned long data)
3265 struct drbd_conf *mdev = (struct drbd_conf *) data;
3267 drbd_queue_work_front(&mdev->tconn->sender_work, &mdev->md_sync_work);
3270 static int w_md_sync(struct drbd_work *w, int unused)
3272 struct drbd_conf *mdev = w->mdev;
3274 dev_warn(DEV, "md_sync_timer expired! Worker calls drbd_md_sync().\n");
3276 dev_warn(DEV, "last md_mark_dirty: %s:%u\n",
3277 mdev->last_md_mark_dirty.func, mdev->last_md_mark_dirty.line);
3283 const char *cmdname(enum drbd_packet cmd)
3285 /* THINK may need to become several global tables
3286 * when we want to support more than
3287 * one PRO_VERSION */
3288 static const char *cmdnames[] = {
3290 [P_DATA_REPLY] = "DataReply",
3291 [P_RS_DATA_REPLY] = "RSDataReply",
3292 [P_BARRIER] = "Barrier",
3293 [P_BITMAP] = "ReportBitMap",
3294 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3295 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3296 [P_UNPLUG_REMOTE] = "UnplugRemote",
3297 [P_DATA_REQUEST] = "DataRequest",
3298 [P_RS_DATA_REQUEST] = "RSDataRequest",
3299 [P_SYNC_PARAM] = "SyncParam",
3300 [P_SYNC_PARAM89] = "SyncParam89",
3301 [P_PROTOCOL] = "ReportProtocol",
3302 [P_UUIDS] = "ReportUUIDs",
3303 [P_SIZES] = "ReportSizes",
3304 [P_STATE] = "ReportState",
3305 [P_SYNC_UUID] = "ReportSyncUUID",
3306 [P_AUTH_CHALLENGE] = "AuthChallenge",
3307 [P_AUTH_RESPONSE] = "AuthResponse",
3309 [P_PING_ACK] = "PingAck",
3310 [P_RECV_ACK] = "RecvAck",
3311 [P_WRITE_ACK] = "WriteAck",
3312 [P_RS_WRITE_ACK] = "RSWriteAck",
3313 [P_DISCARD_WRITE] = "DiscardWrite",
3314 [P_NEG_ACK] = "NegAck",
3315 [P_NEG_DREPLY] = "NegDReply",
3316 [P_NEG_RS_DREPLY] = "NegRSDReply",
3317 [P_BARRIER_ACK] = "BarrierAck",
3318 [P_STATE_CHG_REQ] = "StateChgRequest",
3319 [P_STATE_CHG_REPLY] = "StateChgReply",
3320 [P_OV_REQUEST] = "OVRequest",
3321 [P_OV_REPLY] = "OVReply",
3322 [P_OV_RESULT] = "OVResult",
3323 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3324 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3325 [P_COMPRESSED_BITMAP] = "CBitmap",
3326 [P_DELAY_PROBE] = "DelayProbe",
3327 [P_OUT_OF_SYNC] = "OutOfSync",
3328 [P_RETRY_WRITE] = "RetryWrite",
3329 [P_RS_CANCEL] = "RSCancel",
3330 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3331 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3332 [P_RETRY_WRITE] = "retry_write",
3333 [P_PROTOCOL_UPDATE] = "protocol_update",
3335 /* enum drbd_packet, but not commands - obsoleted flags:
3341 /* too big for the array: 0xfffX */
3342 if (cmd == P_INITIAL_META)
3343 return "InitialMeta";
3344 if (cmd == P_INITIAL_DATA)
3345 return "InitialData";
3346 if (cmd == P_CONNECTION_FEATURES)
3347 return "ConnectionFeatures";
3348 if (cmd >= ARRAY_SIZE(cmdnames))
3350 return cmdnames[cmd];
3354 * drbd_wait_misc - wait for a request to make progress
3355 * @mdev: device associated with the request
3356 * @i: the struct drbd_interval embedded in struct drbd_request or
3357 * struct drbd_peer_request
3359 int drbd_wait_misc(struct drbd_conf *mdev, struct drbd_interval *i)
3361 struct net_conf *nc;
3366 nc = rcu_dereference(mdev->tconn->net_conf);
3371 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3374 /* Indicate to wake up mdev->misc_wait on progress. */
3376 prepare_to_wait(&mdev->misc_wait, &wait, TASK_INTERRUPTIBLE);
3377 spin_unlock_irq(&mdev->tconn->req_lock);
3378 timeout = schedule_timeout(timeout);
3379 finish_wait(&mdev->misc_wait, &wait);
3380 spin_lock_irq(&mdev->tconn->req_lock);
3381 if (!timeout || mdev->state.conn < C_CONNECTED)
3383 if (signal_pending(current))
3384 return -ERESTARTSYS;
3388 #ifdef CONFIG_DRBD_FAULT_INJECTION
3389 /* Fault insertion support including random number generator shamelessly
3390 * stolen from kernel/rcutorture.c */
3391 struct fault_random_state {
3392 unsigned long state;
3393 unsigned long count;
3396 #define FAULT_RANDOM_MULT 39916801 /* prime */
3397 #define FAULT_RANDOM_ADD 479001701 /* prime */
3398 #define FAULT_RANDOM_REFRESH 10000
3401 * Crude but fast random-number generator. Uses a linear congruential
3402 * generator, with occasional help from get_random_bytes().
3404 static unsigned long
3405 _drbd_fault_random(struct fault_random_state *rsp)
3409 if (!rsp->count--) {
3410 get_random_bytes(&refresh, sizeof(refresh));
3411 rsp->state += refresh;
3412 rsp->count = FAULT_RANDOM_REFRESH;
3414 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3415 return swahw32(rsp->state);
3419 _drbd_fault_str(unsigned int type) {
3420 static char *_faults[] = {
3421 [DRBD_FAULT_MD_WR] = "Meta-data write",
3422 [DRBD_FAULT_MD_RD] = "Meta-data read",
3423 [DRBD_FAULT_RS_WR] = "Resync write",
3424 [DRBD_FAULT_RS_RD] = "Resync read",
3425 [DRBD_FAULT_DT_WR] = "Data write",
3426 [DRBD_FAULT_DT_RD] = "Data read",
3427 [DRBD_FAULT_DT_RA] = "Data read ahead",
3428 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3429 [DRBD_FAULT_AL_EE] = "EE allocation",
3430 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3433 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3437 _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type)
3439 static struct fault_random_state rrs = {0, 0};
3441 unsigned int ret = (
3443 ((1 << mdev_to_minor(mdev)) & fault_devs) != 0) &&
3444 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3449 if (__ratelimit(&drbd_ratelimit_state))
3450 dev_warn(DEV, "***Simulating %s failure\n",
3451 _drbd_fault_str(type));
3458 const char *drbd_buildtag(void)
3460 /* DRBD built from external sources has here a reference to the
3461 git hash of the source code. */
3463 static char buildtag[38] = "\0uilt-in";
3465 if (buildtag[0] == 0) {
3466 #ifdef CONFIG_MODULES
3467 if (THIS_MODULE != NULL)
3468 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3477 module_init(drbd_init)
3478 module_exit(drbd_cleanup)
3480 EXPORT_SYMBOL(drbd_conn_str);
3481 EXPORT_SYMBOL(drbd_role_str);
3482 EXPORT_SYMBOL(drbd_disk_str);
3483 EXPORT_SYMBOL(drbd_set_st_err_str);