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>
48 #include <linux/workqueue.h>
49 #define __KERNEL_SYSCALLS__
50 #include <linux/unistd.h>
51 #include <linux/vmalloc.h>
53 #include <linux/drbd_limits.h>
55 #include "drbd_protocol.h"
56 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
60 static DEFINE_MUTEX(drbd_main_mutex
);
61 int drbd_worker(struct drbd_thread
*);
64 static int drbd_open(struct block_device
*bdev
, fmode_t mode
);
65 static void drbd_release(struct gendisk
*gd
, fmode_t mode
);
66 static int w_md_sync(struct drbd_work
*w
, int unused
);
67 static void md_sync_timer_fn(unsigned long data
);
68 static int w_bitmap_io(struct drbd_work
*w
, int unused
);
69 static int w_go_diskless(struct drbd_work
*w
, int unused
);
71 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
72 "Lars Ellenberg <lars@linbit.com>");
73 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION
);
74 MODULE_VERSION(REL_VERSION
);
75 MODULE_LICENSE("GPL");
76 MODULE_PARM_DESC(minor_count
, "Approximate number of drbd devices ("
77 __stringify(DRBD_MINOR_COUNT_MIN
) "-" __stringify(DRBD_MINOR_COUNT_MAX
) ")");
78 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR
);
80 #include <linux/moduleparam.h>
81 /* allow_open_on_secondary */
82 MODULE_PARM_DESC(allow_oos
, "DONT USE!");
83 /* thanks to these macros, if compiled into the kernel (not-module),
84 * this becomes the boot parameter drbd.minor_count */
85 module_param(minor_count
, uint
, 0444);
86 module_param(disable_sendpage
, bool, 0644);
87 module_param(allow_oos
, bool, 0);
88 module_param(proc_details
, int, 0644);
90 #ifdef CONFIG_DRBD_FAULT_INJECTION
93 static int fault_count
;
95 /* bitmap of enabled faults */
96 module_param(enable_faults
, int, 0664);
97 /* fault rate % value - applies to all enabled faults */
98 module_param(fault_rate
, int, 0664);
99 /* count of faults inserted */
100 module_param(fault_count
, int, 0664);
101 /* bitmap of devices to insert faults on */
102 module_param(fault_devs
, int, 0644);
105 /* module parameter, defined */
106 unsigned int minor_count
= DRBD_MINOR_COUNT_DEF
;
107 bool disable_sendpage
;
109 int proc_details
; /* Detail level in proc drbd*/
111 /* Module parameter for setting the user mode helper program
112 * to run. Default is /sbin/drbdadm */
113 char usermode_helper
[80] = "/sbin/drbdadm";
115 module_param_string(usermode_helper
, usermode_helper
, sizeof(usermode_helper
), 0644);
117 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
118 * as member "struct gendisk *vdisk;"
120 struct idr drbd_devices
;
121 struct list_head drbd_resources
;
123 struct kmem_cache
*drbd_request_cache
;
124 struct kmem_cache
*drbd_ee_cache
; /* peer requests */
125 struct kmem_cache
*drbd_bm_ext_cache
; /* bitmap extents */
126 struct kmem_cache
*drbd_al_ext_cache
; /* activity log extents */
127 mempool_t
*drbd_request_mempool
;
128 mempool_t
*drbd_ee_mempool
;
129 mempool_t
*drbd_md_io_page_pool
;
130 struct bio_set
*drbd_md_io_bio_set
;
132 /* I do not use a standard mempool, because:
133 1) I want to hand out the pre-allocated objects first.
134 2) I want to be able to interrupt sleeping allocation with a signal.
135 Note: This is a single linked list, the next pointer is the private
136 member of struct page.
138 struct page
*drbd_pp_pool
;
139 spinlock_t drbd_pp_lock
;
141 wait_queue_head_t drbd_pp_wait
;
143 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state
, 5 * HZ
, 5);
145 static const struct block_device_operations drbd_ops
= {
146 .owner
= THIS_MODULE
,
148 .release
= drbd_release
,
151 struct bio
*bio_alloc_drbd(gfp_t gfp_mask
)
155 if (!drbd_md_io_bio_set
)
156 return bio_alloc(gfp_mask
, 1);
158 bio
= bio_alloc_bioset(gfp_mask
, 1, drbd_md_io_bio_set
);
165 /* When checking with sparse, and this is an inline function, sparse will
166 give tons of false positives. When this is a real functions sparse works.
168 int _get_ldev_if_state(struct drbd_device
*device
, enum drbd_disk_state mins
)
172 atomic_inc(&device
->local_cnt
);
173 io_allowed
= (device
->state
.disk
>= mins
);
175 if (atomic_dec_and_test(&device
->local_cnt
))
176 wake_up(&device
->misc_wait
);
184 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
185 * @connection: DRBD connection.
186 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
187 * @set_size: Expected number of requests before that barrier.
189 * In case the passed barrier_nr or set_size does not match the oldest
190 * epoch of not yet barrier-acked requests, this function will cause a
191 * termination of the connection.
193 void tl_release(struct drbd_connection
*connection
, unsigned int barrier_nr
,
194 unsigned int set_size
)
196 struct drbd_request
*r
;
197 struct drbd_request
*req
= NULL
;
198 int expect_epoch
= 0;
201 spin_lock_irq(&connection
->req_lock
);
203 /* find oldest not yet barrier-acked write request,
204 * count writes in its epoch. */
205 list_for_each_entry(r
, &connection
->transfer_log
, tl_requests
) {
206 const unsigned s
= r
->rq_state
;
210 if (!(s
& RQ_NET_MASK
))
215 expect_epoch
= req
->epoch
;
218 if (r
->epoch
!= expect_epoch
)
222 /* if (s & RQ_DONE): not expected */
223 /* if (!(s & RQ_NET_MASK)): not expected */
228 /* first some paranoia code */
230 conn_err(connection
, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
234 if (expect_epoch
!= barrier_nr
) {
235 conn_err(connection
, "BAD! BarrierAck #%u received, expected #%u!\n",
236 barrier_nr
, expect_epoch
);
240 if (expect_size
!= set_size
) {
241 conn_err(connection
, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
242 barrier_nr
, set_size
, expect_size
);
246 /* Clean up list of requests processed during current epoch. */
247 /* this extra list walk restart is paranoia,
248 * to catch requests being barrier-acked "unexpectedly".
249 * It usually should find the same req again, or some READ preceding it. */
250 list_for_each_entry(req
, &connection
->transfer_log
, tl_requests
)
251 if (req
->epoch
== expect_epoch
)
253 list_for_each_entry_safe_from(req
, r
, &connection
->transfer_log
, tl_requests
) {
254 if (req
->epoch
!= expect_epoch
)
256 _req_mod(req
, BARRIER_ACKED
);
258 spin_unlock_irq(&connection
->req_lock
);
263 spin_unlock_irq(&connection
->req_lock
);
264 conn_request_state(connection
, NS(conn
, C_PROTOCOL_ERROR
), CS_HARD
);
269 * _tl_restart() - Walks the transfer log, and applies an action to all requests
270 * @device: DRBD device.
271 * @what: The action/event to perform with all request objects
273 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
274 * RESTART_FROZEN_DISK_IO.
276 /* must hold resource->req_lock */
277 void _tl_restart(struct drbd_connection
*connection
, enum drbd_req_event what
)
279 struct drbd_request
*req
, *r
;
281 list_for_each_entry_safe(req
, r
, &connection
->transfer_log
, tl_requests
)
285 void tl_restart(struct drbd_connection
*connection
, enum drbd_req_event what
)
287 spin_lock_irq(&connection
->req_lock
);
288 _tl_restart(connection
, what
);
289 spin_unlock_irq(&connection
->req_lock
);
293 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
294 * @device: DRBD device.
296 * This is called after the connection to the peer was lost. The storage covered
297 * by the requests on the transfer gets marked as our of sync. Called from the
298 * receiver thread and the worker thread.
300 void tl_clear(struct drbd_connection
*connection
)
302 tl_restart(connection
, CONNECTION_LOST_WHILE_PENDING
);
306 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
307 * @device: DRBD device.
309 void tl_abort_disk_io(struct drbd_device
*device
)
311 struct drbd_connection
*connection
= first_peer_device(device
)->connection
;
312 struct drbd_request
*req
, *r
;
314 spin_lock_irq(&connection
->req_lock
);
315 list_for_each_entry_safe(req
, r
, &connection
->transfer_log
, tl_requests
) {
316 if (!(req
->rq_state
& RQ_LOCAL_PENDING
))
318 if (req
->w
.device
!= device
)
320 _req_mod(req
, ABORT_DISK_IO
);
322 spin_unlock_irq(&connection
->req_lock
);
325 static int drbd_thread_setup(void *arg
)
327 struct drbd_thread
*thi
= (struct drbd_thread
*) arg
;
328 struct drbd_connection
*connection
= thi
->connection
;
332 snprintf(current
->comm
, sizeof(current
->comm
), "drbd_%c_%s",
334 thi
->connection
->resource
->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(connection
, "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(connection
, "Terminating %s\n", current
->comm
);
366 /* Release mod reference taken when thread was started */
368 kref_put(&connection
->kref
, drbd_destroy_connection
);
369 module_put(THIS_MODULE
);
373 static void drbd_thread_init(struct drbd_connection
*connection
, struct drbd_thread
*thi
,
374 int (*func
) (struct drbd_thread
*), char *name
)
376 spin_lock_init(&thi
->t_lock
);
379 thi
->function
= func
;
380 thi
->connection
= connection
;
381 strncpy(thi
->name
, name
, ARRAY_SIZE(thi
->name
));
384 int drbd_thread_start(struct drbd_thread
*thi
)
386 struct drbd_connection
*connection
= thi
->connection
;
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(connection
, "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(connection
, "Failed to get module reference in drbd_thread_start\n");
402 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
406 kref_get(&thi
->connection
->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
->connection
->resource
->name
);
418 conn_err(connection
, "Couldn't start thread\n");
420 kref_put(&connection
->kref
, drbd_destroy_connection
);
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(connection
, "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_connection
*connection
, struct task_struct
*task
)
483 struct drbd_thread
*thi
=
484 task
== connection
->receiver
.task
? &connection
->receiver
:
485 task
== connection
->asender
.task
? &connection
->asender
:
486 task
== connection
->worker
.task
? &connection
->worker
: NULL
;
491 char *drbd_task_to_thread_name(struct drbd_connection
*connection
, struct task_struct
*task
)
493 struct drbd_thread
*thi
= drbd_task_to_thread(connection
, task
);
494 return thi
? thi
->name
: task
->comm
;
497 int conn_lowest_minor(struct drbd_connection
*connection
)
499 struct drbd_device
*device
;
503 device
= idr_get_next(&connection
->volumes
, &vnr
);
504 m
= device
? device_to_minor(device
) : -1;
512 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
513 * @device: 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_connection
*connection
)
523 if (cpumask_weight(connection
->cpu_mask
))
526 ord
= conn_lowest_minor(connection
) % cpumask_weight(cpu_online_mask
);
527 for_each_online_cpu(cpu
) {
529 cpumask_set_cpu(cpu
, connection
->cpu_mask
);
533 /* should not be reached */
534 cpumask_setall(connection
->cpu_mask
);
538 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
539 * @device: 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
->connection
->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_connection
*connection
)
565 if (connection
->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_connection
*connection
, int vnr
,
604 void *buffer
, enum drbd_packet cmd
, int size
)
606 if (connection
->agreed_pro_version
>= 100)
607 return prepare_header100(buffer
, cmd
, size
, vnr
);
608 else if (connection
->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_connection
*connection
,
616 struct drbd_socket
*sock
)
620 return sock
->sbuf
+ drbd_header_size(connection
);
623 void *conn_prepare_command(struct drbd_connection
*connection
, struct drbd_socket
*sock
)
627 mutex_lock(&sock
->mutex
);
628 p
= __conn_prepare_command(connection
, sock
);
630 mutex_unlock(&sock
->mutex
);
635 void *drbd_prepare_command(struct drbd_device
*device
, struct drbd_socket
*sock
)
637 return conn_prepare_command(first_peer_device(device
)->connection
, sock
);
640 static int __send_command(struct drbd_connection
*connection
, 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(connection
, vnr
, sock
->sbuf
, cmd
,
659 err
= drbd_send_all(connection
, sock
->socket
, sock
->sbuf
, header_size
,
662 err
= drbd_send_all(connection
, sock
->socket
, data
, size
, 0);
666 static int __conn_send_command(struct drbd_connection
*connection
, struct drbd_socket
*sock
,
667 enum drbd_packet cmd
, unsigned int header_size
,
668 void *data
, unsigned int size
)
670 return __send_command(connection
, 0, sock
, cmd
, header_size
, data
, size
);
673 int conn_send_command(struct drbd_connection
*connection
, struct drbd_socket
*sock
,
674 enum drbd_packet cmd
, unsigned int header_size
,
675 void *data
, unsigned int size
)
679 err
= __conn_send_command(connection
, sock
, cmd
, header_size
, data
, size
);
680 mutex_unlock(&sock
->mutex
);
684 int drbd_send_command(struct drbd_device
*device
, struct drbd_socket
*sock
,
685 enum drbd_packet cmd
, unsigned int header_size
,
686 void *data
, unsigned int size
)
690 err
= __send_command(first_peer_device(device
)->connection
, device
->vnr
, sock
, cmd
, header_size
,
692 mutex_unlock(&sock
->mutex
);
696 int drbd_send_ping(struct drbd_connection
*connection
)
698 struct drbd_socket
*sock
;
700 sock
= &connection
->meta
;
701 if (!conn_prepare_command(connection
, sock
))
703 return conn_send_command(connection
, sock
, P_PING
, 0, NULL
, 0);
706 int drbd_send_ping_ack(struct drbd_connection
*connection
)
708 struct drbd_socket
*sock
;
710 sock
= &connection
->meta
;
711 if (!conn_prepare_command(connection
, sock
))
713 return conn_send_command(connection
, sock
, P_PING_ACK
, 0, NULL
, 0);
716 int drbd_send_sync_param(struct drbd_device
*device
)
718 struct drbd_socket
*sock
;
719 struct p_rs_param_95
*p
;
721 const int apv
= first_peer_device(device
)->connection
->agreed_pro_version
;
722 enum drbd_packet cmd
;
724 struct disk_conf
*dc
;
726 sock
= &first_peer_device(device
)->connection
->data
;
727 p
= drbd_prepare_command(device
, sock
);
732 nc
= rcu_dereference(first_peer_device(device
)->connection
->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(device
)) {
746 dc
= rcu_dereference(device
->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(device
, sock
, cmd
, size
, NULL
, 0);
770 int __drbd_send_protocol(struct drbd_connection
*connection
, enum drbd_packet cmd
)
772 struct drbd_socket
*sock
;
773 struct p_protocol
*p
;
777 sock
= &connection
->data
;
778 p
= __conn_prepare_command(connection
, sock
);
783 nc
= rcu_dereference(connection
->net_conf
);
785 if (nc
->tentative
&& connection
->agreed_pro_version
< 92) {
787 mutex_unlock(&sock
->mutex
);
788 conn_err(connection
, "--dry-run is not supported by peer");
793 if (connection
->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 (connection
->agreed_pro_version
>= 87)
809 strcpy(p
->integrity_alg
, nc
->integrity_alg
);
812 return __conn_send_command(connection
, sock
, cmd
, size
, NULL
, 0);
815 int drbd_send_protocol(struct drbd_connection
*connection
)
819 mutex_lock(&connection
->data
.mutex
);
820 err
= __drbd_send_protocol(connection
, P_PROTOCOL
);
821 mutex_unlock(&connection
->data
.mutex
);
826 static int _drbd_send_uuids(struct drbd_device
*device
, u64 uuid_flags
)
828 struct drbd_socket
*sock
;
832 if (!get_ldev_if_state(device
, D_NEGOTIATING
))
835 sock
= &first_peer_device(device
)->connection
->data
;
836 p
= drbd_prepare_command(device
, sock
);
841 spin_lock_irq(&device
->ldev
->md
.uuid_lock
);
842 for (i
= UI_CURRENT
; i
< UI_SIZE
; i
++)
843 p
->uuid
[i
] = cpu_to_be64(device
->ldev
->md
.uuid
[i
]);
844 spin_unlock_irq(&device
->ldev
->md
.uuid_lock
);
846 device
->comm_bm_set
= drbd_bm_total_weight(device
);
847 p
->uuid
[UI_SIZE
] = cpu_to_be64(device
->comm_bm_set
);
849 uuid_flags
|= rcu_dereference(first_peer_device(device
)->connection
->net_conf
)->discard_my_data
? 1 : 0;
851 uuid_flags
|= test_bit(CRASHED_PRIMARY
, &device
->flags
) ? 2 : 0;
852 uuid_flags
|= device
->new_state_tmp
.disk
== D_INCONSISTENT
? 4 : 0;
853 p
->uuid
[UI_FLAGS
] = cpu_to_be64(uuid_flags
);
856 return drbd_send_command(device
, sock
, P_UUIDS
, sizeof(*p
), NULL
, 0);
859 int drbd_send_uuids(struct drbd_device
*device
)
861 return _drbd_send_uuids(device
, 0);
864 int drbd_send_uuids_skip_initial_sync(struct drbd_device
*device
)
866 return _drbd_send_uuids(device
, 8);
869 void drbd_print_uuids(struct drbd_device
*device
, const char *text
)
871 if (get_ldev_if_state(device
, D_NEGOTIATING
)) {
872 u64
*uuid
= device
->ldev
->md
.uuid
;
873 dev_info(DEV
, "%s %016llX:%016llX:%016llX:%016llX\n",
875 (unsigned long long)uuid
[UI_CURRENT
],
876 (unsigned long long)uuid
[UI_BITMAP
],
877 (unsigned long long)uuid
[UI_HISTORY_START
],
878 (unsigned long long)uuid
[UI_HISTORY_END
]);
881 dev_info(DEV
, "%s effective data uuid: %016llX\n",
883 (unsigned long long)device
->ed_uuid
);
887 void drbd_gen_and_send_sync_uuid(struct drbd_device
*device
)
889 struct drbd_socket
*sock
;
893 D_ASSERT(device
->state
.disk
== D_UP_TO_DATE
);
895 uuid
= device
->ldev
->md
.uuid
[UI_BITMAP
];
896 if (uuid
&& uuid
!= UUID_JUST_CREATED
)
897 uuid
= uuid
+ UUID_NEW_BM_OFFSET
;
899 get_random_bytes(&uuid
, sizeof(u64
));
900 drbd_uuid_set(device
, UI_BITMAP
, uuid
);
901 drbd_print_uuids(device
, "updated sync UUID");
902 drbd_md_sync(device
);
904 sock
= &first_peer_device(device
)->connection
->data
;
905 p
= drbd_prepare_command(device
, sock
);
907 p
->uuid
= cpu_to_be64(uuid
);
908 drbd_send_command(device
, sock
, P_SYNC_UUID
, sizeof(*p
), NULL
, 0);
912 int drbd_send_sizes(struct drbd_device
*device
, int trigger_reply
, enum dds_flags flags
)
914 struct drbd_socket
*sock
;
916 sector_t d_size
, u_size
;
918 unsigned int max_bio_size
;
920 if (get_ldev_if_state(device
, D_NEGOTIATING
)) {
921 D_ASSERT(device
->ldev
->backing_bdev
);
922 d_size
= drbd_get_max_capacity(device
->ldev
);
924 u_size
= rcu_dereference(device
->ldev
->disk_conf
)->disk_size
;
926 q_order_type
= drbd_queue_order_type(device
);
927 max_bio_size
= queue_max_hw_sectors(device
->ldev
->backing_bdev
->bd_disk
->queue
) << 9;
928 max_bio_size
= min(max_bio_size
, DRBD_MAX_BIO_SIZE
);
933 q_order_type
= QUEUE_ORDERED_NONE
;
934 max_bio_size
= DRBD_MAX_BIO_SIZE
; /* ... multiple BIOs per peer_request */
937 sock
= &first_peer_device(device
)->connection
->data
;
938 p
= drbd_prepare_command(device
, sock
);
942 if (first_peer_device(device
)->connection
->agreed_pro_version
<= 94)
943 max_bio_size
= min(max_bio_size
, DRBD_MAX_SIZE_H80_PACKET
);
944 else if (first_peer_device(device
)->connection
->agreed_pro_version
< 100)
945 max_bio_size
= min(max_bio_size
, DRBD_MAX_BIO_SIZE_P95
);
947 p
->d_size
= cpu_to_be64(d_size
);
948 p
->u_size
= cpu_to_be64(u_size
);
949 p
->c_size
= cpu_to_be64(trigger_reply
? 0 : drbd_get_capacity(device
->this_bdev
));
950 p
->max_bio_size
= cpu_to_be32(max_bio_size
);
951 p
->queue_order_type
= cpu_to_be16(q_order_type
);
952 p
->dds_flags
= cpu_to_be16(flags
);
953 return drbd_send_command(device
, sock
, P_SIZES
, sizeof(*p
), NULL
, 0);
957 * drbd_send_current_state() - Sends the drbd state to the peer
958 * @device: DRBD device.
960 int drbd_send_current_state(struct drbd_device
*device
)
962 struct drbd_socket
*sock
;
965 sock
= &first_peer_device(device
)->connection
->data
;
966 p
= drbd_prepare_command(device
, sock
);
969 p
->state
= cpu_to_be32(device
->state
.i
); /* Within the send mutex */
970 return drbd_send_command(device
, sock
, P_STATE
, sizeof(*p
), NULL
, 0);
974 * drbd_send_state() - After a state change, sends the new state to the peer
975 * @device: DRBD device.
976 * @state: the state to send, not necessarily the current state.
978 * Each state change queues an "after_state_ch" work, which will eventually
979 * send the resulting new state to the peer. If more state changes happen
980 * between queuing and processing of the after_state_ch work, we still
981 * want to send each intermediary state in the order it occurred.
983 int drbd_send_state(struct drbd_device
*device
, union drbd_state state
)
985 struct drbd_socket
*sock
;
988 sock
= &first_peer_device(device
)->connection
->data
;
989 p
= drbd_prepare_command(device
, sock
);
992 p
->state
= cpu_to_be32(state
.i
); /* Within the send mutex */
993 return drbd_send_command(device
, sock
, P_STATE
, sizeof(*p
), NULL
, 0);
996 int drbd_send_state_req(struct drbd_device
*device
, union drbd_state mask
, union drbd_state val
)
998 struct drbd_socket
*sock
;
999 struct p_req_state
*p
;
1001 sock
= &first_peer_device(device
)->connection
->data
;
1002 p
= drbd_prepare_command(device
, sock
);
1005 p
->mask
= cpu_to_be32(mask
.i
);
1006 p
->val
= cpu_to_be32(val
.i
);
1007 return drbd_send_command(device
, sock
, P_STATE_CHG_REQ
, sizeof(*p
), NULL
, 0);
1010 int conn_send_state_req(struct drbd_connection
*connection
, union drbd_state mask
, union drbd_state val
)
1012 enum drbd_packet cmd
;
1013 struct drbd_socket
*sock
;
1014 struct p_req_state
*p
;
1016 cmd
= connection
->agreed_pro_version
< 100 ? P_STATE_CHG_REQ
: P_CONN_ST_CHG_REQ
;
1017 sock
= &connection
->data
;
1018 p
= conn_prepare_command(connection
, sock
);
1021 p
->mask
= cpu_to_be32(mask
.i
);
1022 p
->val
= cpu_to_be32(val
.i
);
1023 return conn_send_command(connection
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1026 void drbd_send_sr_reply(struct drbd_device
*device
, enum drbd_state_rv retcode
)
1028 struct drbd_socket
*sock
;
1029 struct p_req_state_reply
*p
;
1031 sock
= &first_peer_device(device
)->connection
->meta
;
1032 p
= drbd_prepare_command(device
, sock
);
1034 p
->retcode
= cpu_to_be32(retcode
);
1035 drbd_send_command(device
, sock
, P_STATE_CHG_REPLY
, sizeof(*p
), NULL
, 0);
1039 void conn_send_sr_reply(struct drbd_connection
*connection
, enum drbd_state_rv retcode
)
1041 struct drbd_socket
*sock
;
1042 struct p_req_state_reply
*p
;
1043 enum drbd_packet cmd
= connection
->agreed_pro_version
< 100 ? P_STATE_CHG_REPLY
: P_CONN_ST_CHG_REPLY
;
1045 sock
= &connection
->meta
;
1046 p
= conn_prepare_command(connection
, sock
);
1048 p
->retcode
= cpu_to_be32(retcode
);
1049 conn_send_command(connection
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1053 static void dcbp_set_code(struct p_compressed_bm
*p
, enum drbd_bitmap_code code
)
1055 BUG_ON(code
& ~0xf);
1056 p
->encoding
= (p
->encoding
& ~0xf) | code
;
1059 static void dcbp_set_start(struct p_compressed_bm
*p
, int set
)
1061 p
->encoding
= (p
->encoding
& ~0x80) | (set
? 0x80 : 0);
1064 static void dcbp_set_pad_bits(struct p_compressed_bm
*p
, int n
)
1067 p
->encoding
= (p
->encoding
& (~0x7 << 4)) | (n
<< 4);
1070 static int fill_bitmap_rle_bits(struct drbd_device
*device
,
1071 struct p_compressed_bm
*p
,
1073 struct bm_xfer_ctx
*c
)
1075 struct bitstream bs
;
1076 unsigned long plain_bits
;
1083 /* may we use this feature? */
1085 use_rle
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
)->use_rle
;
1087 if (!use_rle
|| first_peer_device(device
)->connection
->agreed_pro_version
< 90)
1090 if (c
->bit_offset
>= c
->bm_bits
)
1091 return 0; /* nothing to do. */
1093 /* use at most thus many bytes */
1094 bitstream_init(&bs
, p
->code
, size
, 0);
1095 memset(p
->code
, 0, size
);
1096 /* plain bits covered in this code string */
1099 /* p->encoding & 0x80 stores whether the first run length is set.
1100 * bit offset is implicit.
1101 * start with toggle == 2 to be able to tell the first iteration */
1104 /* see how much plain bits we can stuff into one packet
1105 * using RLE and VLI. */
1107 tmp
= (toggle
== 0) ? _drbd_bm_find_next_zero(device
, c
->bit_offset
)
1108 : _drbd_bm_find_next(device
, c
->bit_offset
);
1111 rl
= tmp
- c
->bit_offset
;
1113 if (toggle
== 2) { /* first iteration */
1115 /* the first checked bit was set,
1116 * store start value, */
1117 dcbp_set_start(p
, 1);
1118 /* but skip encoding of zero run length */
1122 dcbp_set_start(p
, 0);
1125 /* paranoia: catch zero runlength.
1126 * can only happen if bitmap is modified while we scan it. */
1128 dev_err(DEV
, "unexpected zero runlength while encoding bitmap "
1129 "t:%u bo:%lu\n", toggle
, c
->bit_offset
);
1133 bits
= vli_encode_bits(&bs
, rl
);
1134 if (bits
== -ENOBUFS
) /* buffer full */
1137 dev_err(DEV
, "error while encoding bitmap: %d\n", bits
);
1143 c
->bit_offset
= tmp
;
1144 } while (c
->bit_offset
< c
->bm_bits
);
1146 len
= bs
.cur
.b
- p
->code
+ !!bs
.cur
.bit
;
1148 if (plain_bits
< (len
<< 3)) {
1149 /* incompressible with this method.
1150 * we need to rewind both word and bit position. */
1151 c
->bit_offset
-= plain_bits
;
1152 bm_xfer_ctx_bit_to_word_offset(c
);
1153 c
->bit_offset
= c
->word_offset
* BITS_PER_LONG
;
1157 /* RLE + VLI was able to compress it just fine.
1158 * update c->word_offset. */
1159 bm_xfer_ctx_bit_to_word_offset(c
);
1161 /* store pad_bits */
1162 dcbp_set_pad_bits(p
, (8 - bs
.cur
.bit
) & 0x7);
1168 * send_bitmap_rle_or_plain
1170 * Return 0 when done, 1 when another iteration is needed, and a negative error
1171 * code upon failure.
1174 send_bitmap_rle_or_plain(struct drbd_device
*device
, struct bm_xfer_ctx
*c
)
1176 struct drbd_socket
*sock
= &first_peer_device(device
)->connection
->data
;
1177 unsigned int header_size
= drbd_header_size(first_peer_device(device
)->connection
);
1178 struct p_compressed_bm
*p
= sock
->sbuf
+ header_size
;
1181 len
= fill_bitmap_rle_bits(device
, p
,
1182 DRBD_SOCKET_BUFFER_SIZE
- header_size
- sizeof(*p
), c
);
1187 dcbp_set_code(p
, RLE_VLI_Bits
);
1188 err
= __send_command(first_peer_device(device
)->connection
, device
->vnr
, sock
,
1189 P_COMPRESSED_BITMAP
, sizeof(*p
) + len
,
1192 c
->bytes
[0] += header_size
+ sizeof(*p
) + len
;
1194 if (c
->bit_offset
>= c
->bm_bits
)
1197 /* was not compressible.
1198 * send a buffer full of plain text bits instead. */
1199 unsigned int data_size
;
1200 unsigned long num_words
;
1201 unsigned long *p
= sock
->sbuf
+ header_size
;
1203 data_size
= DRBD_SOCKET_BUFFER_SIZE
- header_size
;
1204 num_words
= min_t(size_t, data_size
/ sizeof(*p
),
1205 c
->bm_words
- c
->word_offset
);
1206 len
= num_words
* sizeof(*p
);
1208 drbd_bm_get_lel(device
, c
->word_offset
, num_words
, p
);
1209 err
= __send_command(first_peer_device(device
)->connection
, device
->vnr
, sock
, P_BITMAP
, len
, NULL
, 0);
1210 c
->word_offset
+= num_words
;
1211 c
->bit_offset
= c
->word_offset
* BITS_PER_LONG
;
1214 c
->bytes
[1] += header_size
+ len
;
1216 if (c
->bit_offset
> c
->bm_bits
)
1217 c
->bit_offset
= c
->bm_bits
;
1221 INFO_bm_xfer_stats(device
, "send", c
);
1229 /* See the comment at receive_bitmap() */
1230 static int _drbd_send_bitmap(struct drbd_device
*device
)
1232 struct bm_xfer_ctx c
;
1235 if (!expect(device
->bitmap
))
1238 if (get_ldev(device
)) {
1239 if (drbd_md_test_flag(device
->ldev
, MDF_FULL_SYNC
)) {
1240 dev_info(DEV
, "Writing the whole bitmap, MDF_FullSync was set.\n");
1241 drbd_bm_set_all(device
);
1242 if (drbd_bm_write(device
)) {
1243 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1244 * but otherwise process as per normal - need to tell other
1245 * side that a full resync is required! */
1246 dev_err(DEV
, "Failed to write bitmap to disk!\n");
1248 drbd_md_clear_flag(device
, MDF_FULL_SYNC
);
1249 drbd_md_sync(device
);
1255 c
= (struct bm_xfer_ctx
) {
1256 .bm_bits
= drbd_bm_bits(device
),
1257 .bm_words
= drbd_bm_words(device
),
1261 err
= send_bitmap_rle_or_plain(device
, &c
);
1267 int drbd_send_bitmap(struct drbd_device
*device
)
1269 struct drbd_socket
*sock
= &first_peer_device(device
)->connection
->data
;
1272 mutex_lock(&sock
->mutex
);
1274 err
= !_drbd_send_bitmap(device
);
1275 mutex_unlock(&sock
->mutex
);
1279 void drbd_send_b_ack(struct drbd_connection
*connection
, u32 barrier_nr
, u32 set_size
)
1281 struct drbd_socket
*sock
;
1282 struct p_barrier_ack
*p
;
1284 if (connection
->cstate
< C_WF_REPORT_PARAMS
)
1287 sock
= &connection
->meta
;
1288 p
= conn_prepare_command(connection
, sock
);
1291 p
->barrier
= barrier_nr
;
1292 p
->set_size
= cpu_to_be32(set_size
);
1293 conn_send_command(connection
, sock
, P_BARRIER_ACK
, sizeof(*p
), NULL
, 0);
1297 * _drbd_send_ack() - Sends an ack packet
1298 * @device: DRBD device.
1299 * @cmd: Packet command code.
1300 * @sector: sector, needs to be in big endian byte order
1301 * @blksize: size in byte, needs to be in big endian byte order
1302 * @block_id: Id, big endian byte order
1304 static int _drbd_send_ack(struct drbd_device
*device
, enum drbd_packet cmd
,
1305 u64 sector
, u32 blksize
, u64 block_id
)
1307 struct drbd_socket
*sock
;
1308 struct p_block_ack
*p
;
1310 if (device
->state
.conn
< C_CONNECTED
)
1313 sock
= &first_peer_device(device
)->connection
->meta
;
1314 p
= drbd_prepare_command(device
, sock
);
1318 p
->block_id
= block_id
;
1319 p
->blksize
= blksize
;
1320 p
->seq_num
= cpu_to_be32(atomic_inc_return(&device
->packet_seq
));
1321 return drbd_send_command(device
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1324 /* dp->sector and dp->block_id already/still in network byte order,
1325 * data_size is payload size according to dp->head,
1326 * and may need to be corrected for digest size. */
1327 void drbd_send_ack_dp(struct drbd_device
*device
, enum drbd_packet cmd
,
1328 struct p_data
*dp
, int data_size
)
1330 if (first_peer_device(device
)->connection
->peer_integrity_tfm
)
1331 data_size
-= crypto_hash_digestsize(first_peer_device(device
)->connection
->peer_integrity_tfm
);
1332 _drbd_send_ack(device
, cmd
, dp
->sector
, cpu_to_be32(data_size
),
1336 void drbd_send_ack_rp(struct drbd_device
*device
, enum drbd_packet cmd
,
1337 struct p_block_req
*rp
)
1339 _drbd_send_ack(device
, cmd
, rp
->sector
, rp
->blksize
, rp
->block_id
);
1343 * drbd_send_ack() - Sends an ack packet
1344 * @device: DRBD device
1345 * @cmd: packet command code
1346 * @peer_req: peer request
1348 int drbd_send_ack(struct drbd_device
*device
, enum drbd_packet cmd
,
1349 struct drbd_peer_request
*peer_req
)
1351 return _drbd_send_ack(device
, cmd
,
1352 cpu_to_be64(peer_req
->i
.sector
),
1353 cpu_to_be32(peer_req
->i
.size
),
1354 peer_req
->block_id
);
1357 /* This function misuses the block_id field to signal if the blocks
1358 * are is sync or not. */
1359 int drbd_send_ack_ex(struct drbd_device
*device
, enum drbd_packet cmd
,
1360 sector_t sector
, int blksize
, u64 block_id
)
1362 return _drbd_send_ack(device
, cmd
,
1363 cpu_to_be64(sector
),
1364 cpu_to_be32(blksize
),
1365 cpu_to_be64(block_id
));
1368 int drbd_send_drequest(struct drbd_device
*device
, int cmd
,
1369 sector_t sector
, int size
, u64 block_id
)
1371 struct drbd_socket
*sock
;
1372 struct p_block_req
*p
;
1374 sock
= &first_peer_device(device
)->connection
->data
;
1375 p
= drbd_prepare_command(device
, sock
);
1378 p
->sector
= cpu_to_be64(sector
);
1379 p
->block_id
= block_id
;
1380 p
->blksize
= cpu_to_be32(size
);
1381 return drbd_send_command(device
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1384 int drbd_send_drequest_csum(struct drbd_device
*device
, sector_t sector
, int size
,
1385 void *digest
, int digest_size
, enum drbd_packet cmd
)
1387 struct drbd_socket
*sock
;
1388 struct p_block_req
*p
;
1390 /* FIXME: Put the digest into the preallocated socket buffer. */
1392 sock
= &first_peer_device(device
)->connection
->data
;
1393 p
= drbd_prepare_command(device
, sock
);
1396 p
->sector
= cpu_to_be64(sector
);
1397 p
->block_id
= ID_SYNCER
/* unused */;
1398 p
->blksize
= cpu_to_be32(size
);
1399 return drbd_send_command(device
, sock
, cmd
, sizeof(*p
),
1400 digest
, digest_size
);
1403 int drbd_send_ov_request(struct drbd_device
*device
, sector_t sector
, int size
)
1405 struct drbd_socket
*sock
;
1406 struct p_block_req
*p
;
1408 sock
= &first_peer_device(device
)->connection
->data
;
1409 p
= drbd_prepare_command(device
, sock
);
1412 p
->sector
= cpu_to_be64(sector
);
1413 p
->block_id
= ID_SYNCER
/* unused */;
1414 p
->blksize
= cpu_to_be32(size
);
1415 return drbd_send_command(device
, sock
, P_OV_REQUEST
, sizeof(*p
), NULL
, 0);
1418 /* called on sndtimeo
1419 * returns false if we should retry,
1420 * true if we think connection is dead
1422 static int we_should_drop_the_connection(struct drbd_connection
*connection
, struct socket
*sock
)
1425 /* long elapsed = (long)(jiffies - device->last_received); */
1427 drop_it
= connection
->meta
.socket
== sock
1428 || !connection
->asender
.task
1429 || get_t_state(&connection
->asender
) != RUNNING
1430 || connection
->cstate
< C_WF_REPORT_PARAMS
;
1435 drop_it
= !--connection
->ko_count
;
1437 conn_err(connection
, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1438 current
->comm
, current
->pid
, connection
->ko_count
);
1439 request_ping(connection
);
1442 return drop_it
; /* && (device->state == R_PRIMARY) */;
1445 static void drbd_update_congested(struct drbd_connection
*connection
)
1447 struct sock
*sk
= connection
->data
.socket
->sk
;
1448 if (sk
->sk_wmem_queued
> sk
->sk_sndbuf
* 4 / 5)
1449 set_bit(NET_CONGESTED
, &connection
->flags
);
1452 /* The idea of sendpage seems to be to put some kind of reference
1453 * to the page into the skb, and to hand it over to the NIC. In
1454 * this process get_page() gets called.
1456 * As soon as the page was really sent over the network put_page()
1457 * gets called by some part of the network layer. [ NIC driver? ]
1459 * [ get_page() / put_page() increment/decrement the count. If count
1460 * reaches 0 the page will be freed. ]
1462 * This works nicely with pages from FSs.
1463 * But this means that in protocol A we might signal IO completion too early!
1465 * In order not to corrupt data during a resync we must make sure
1466 * that we do not reuse our own buffer pages (EEs) to early, therefore
1467 * we have the net_ee list.
1469 * XFS seems to have problems, still, it submits pages with page_count == 0!
1470 * As a workaround, we disable sendpage on pages
1471 * with page_count == 0 or PageSlab.
1473 static int _drbd_no_send_page(struct drbd_device
*device
, struct page
*page
,
1474 int offset
, size_t size
, unsigned msg_flags
)
1476 struct socket
*socket
;
1480 socket
= first_peer_device(device
)->connection
->data
.socket
;
1481 addr
= kmap(page
) + offset
;
1482 err
= drbd_send_all(first_peer_device(device
)->connection
, socket
, addr
, size
, msg_flags
);
1485 device
->send_cnt
+= size
>> 9;
1489 static int _drbd_send_page(struct drbd_device
*device
, struct page
*page
,
1490 int offset
, size_t size
, unsigned msg_flags
)
1492 struct socket
*socket
= first_peer_device(device
)->connection
->data
.socket
;
1493 mm_segment_t oldfs
= get_fs();
1497 /* e.g. XFS meta- & log-data is in slab pages, which have a
1498 * page_count of 0 and/or have PageSlab() set.
1499 * we cannot use send_page for those, as that does get_page();
1500 * put_page(); and would cause either a VM_BUG directly, or
1501 * __page_cache_release a page that would actually still be referenced
1502 * by someone, leading to some obscure delayed Oops somewhere else. */
1503 if (disable_sendpage
|| (page_count(page
) < 1) || PageSlab(page
))
1504 return _drbd_no_send_page(device
, page
, offset
, size
, msg_flags
);
1506 msg_flags
|= MSG_NOSIGNAL
;
1507 drbd_update_congested(first_peer_device(device
)->connection
);
1512 sent
= socket
->ops
->sendpage(socket
, page
, offset
, len
, msg_flags
);
1514 if (sent
== -EAGAIN
) {
1515 if (we_should_drop_the_connection(first_peer_device(device
)->connection
, socket
))
1519 dev_warn(DEV
, "%s: size=%d len=%d sent=%d\n",
1520 __func__
, (int)size
, len
, sent
);
1527 } while (len
> 0 /* THINK && device->cstate >= C_CONNECTED*/);
1529 clear_bit(NET_CONGESTED
, &first_peer_device(device
)->connection
->flags
);
1533 device
->send_cnt
+= size
>> 9;
1538 static int _drbd_send_bio(struct drbd_device
*device
, struct bio
*bio
)
1540 struct bio_vec bvec
;
1541 struct bvec_iter iter
;
1543 /* hint all but last page with MSG_MORE */
1544 bio_for_each_segment(bvec
, bio
, iter
) {
1547 err
= _drbd_no_send_page(device
, bvec
.bv_page
,
1548 bvec
.bv_offset
, bvec
.bv_len
,
1549 bio_iter_last(bvec
, iter
)
1557 static int _drbd_send_zc_bio(struct drbd_device
*device
, struct bio
*bio
)
1559 struct bio_vec bvec
;
1560 struct bvec_iter iter
;
1562 /* hint all but last page with MSG_MORE */
1563 bio_for_each_segment(bvec
, bio
, iter
) {
1566 err
= _drbd_send_page(device
, bvec
.bv_page
,
1567 bvec
.bv_offset
, bvec
.bv_len
,
1568 bio_iter_last(bvec
, iter
) ? 0 : MSG_MORE
);
1575 static int _drbd_send_zc_ee(struct drbd_device
*device
,
1576 struct drbd_peer_request
*peer_req
)
1578 struct page
*page
= peer_req
->pages
;
1579 unsigned len
= peer_req
->i
.size
;
1582 /* hint all but last page with MSG_MORE */
1583 page_chain_for_each(page
) {
1584 unsigned l
= min_t(unsigned, len
, PAGE_SIZE
);
1586 err
= _drbd_send_page(device
, page
, 0, l
,
1587 page_chain_next(page
) ? MSG_MORE
: 0);
1595 static u32
bio_flags_to_wire(struct drbd_device
*device
, unsigned long bi_rw
)
1597 if (first_peer_device(device
)->connection
->agreed_pro_version
>= 95)
1598 return (bi_rw
& REQ_SYNC
? DP_RW_SYNC
: 0) |
1599 (bi_rw
& REQ_FUA
? DP_FUA
: 0) |
1600 (bi_rw
& REQ_FLUSH
? DP_FLUSH
: 0) |
1601 (bi_rw
& REQ_DISCARD
? DP_DISCARD
: 0);
1603 return bi_rw
& REQ_SYNC
? DP_RW_SYNC
: 0;
1606 /* Used to send write requests
1607 * R_PRIMARY -> Peer (P_DATA)
1609 int drbd_send_dblock(struct drbd_device
*device
, struct drbd_request
*req
)
1611 struct drbd_socket
*sock
;
1613 unsigned int dp_flags
= 0;
1617 sock
= &first_peer_device(device
)->connection
->data
;
1618 p
= drbd_prepare_command(device
, sock
);
1619 dgs
= first_peer_device(device
)->connection
->integrity_tfm
?
1620 crypto_hash_digestsize(first_peer_device(device
)->connection
->integrity_tfm
) : 0;
1624 p
->sector
= cpu_to_be64(req
->i
.sector
);
1625 p
->block_id
= (unsigned long)req
;
1626 p
->seq_num
= cpu_to_be32(atomic_inc_return(&device
->packet_seq
));
1627 dp_flags
= bio_flags_to_wire(device
, req
->master_bio
->bi_rw
);
1628 if (device
->state
.conn
>= C_SYNC_SOURCE
&&
1629 device
->state
.conn
<= C_PAUSED_SYNC_T
)
1630 dp_flags
|= DP_MAY_SET_IN_SYNC
;
1631 if (first_peer_device(device
)->connection
->agreed_pro_version
>= 100) {
1632 if (req
->rq_state
& RQ_EXP_RECEIVE_ACK
)
1633 dp_flags
|= DP_SEND_RECEIVE_ACK
;
1634 if (req
->rq_state
& RQ_EXP_WRITE_ACK
)
1635 dp_flags
|= DP_SEND_WRITE_ACK
;
1637 p
->dp_flags
= cpu_to_be32(dp_flags
);
1639 drbd_csum_bio(device
, first_peer_device(device
)->connection
->integrity_tfm
, req
->master_bio
, p
+ 1);
1640 err
= __send_command(first_peer_device(device
)->connection
, device
->vnr
, sock
, P_DATA
, sizeof(*p
) + dgs
, NULL
, req
->i
.size
);
1642 /* For protocol A, we have to memcpy the payload into
1643 * socket buffers, as we may complete right away
1644 * as soon as we handed it over to tcp, at which point the data
1645 * pages may become invalid.
1647 * For data-integrity enabled, we copy it as well, so we can be
1648 * sure that even if the bio pages may still be modified, it
1649 * won't change the data on the wire, thus if the digest checks
1650 * out ok after sending on this side, but does not fit on the
1651 * receiving side, we sure have detected corruption elsewhere.
1653 if (!(req
->rq_state
& (RQ_EXP_RECEIVE_ACK
| RQ_EXP_WRITE_ACK
)) || dgs
)
1654 err
= _drbd_send_bio(device
, req
->master_bio
);
1656 err
= _drbd_send_zc_bio(device
, req
->master_bio
);
1658 /* double check digest, sometimes buffers have been modified in flight. */
1659 if (dgs
> 0 && dgs
<= 64) {
1660 /* 64 byte, 512 bit, is the largest digest size
1661 * currently supported in kernel crypto. */
1662 unsigned char digest
[64];
1663 drbd_csum_bio(device
, first_peer_device(device
)->connection
->integrity_tfm
, req
->master_bio
, digest
);
1664 if (memcmp(p
+ 1, digest
, dgs
)) {
1666 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1667 (unsigned long long)req
->i
.sector
, req
->i
.size
);
1669 } /* else if (dgs > 64) {
1670 ... Be noisy about digest too large ...
1673 mutex_unlock(&sock
->mutex
); /* locked by drbd_prepare_command() */
1678 /* answer packet, used to send data back for read requests:
1679 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1680 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1682 int drbd_send_block(struct drbd_device
*device
, enum drbd_packet cmd
,
1683 struct drbd_peer_request
*peer_req
)
1685 struct drbd_socket
*sock
;
1690 sock
= &first_peer_device(device
)->connection
->data
;
1691 p
= drbd_prepare_command(device
, sock
);
1693 dgs
= first_peer_device(device
)->connection
->integrity_tfm
?
1694 crypto_hash_digestsize(first_peer_device(device
)->connection
->integrity_tfm
) : 0;
1698 p
->sector
= cpu_to_be64(peer_req
->i
.sector
);
1699 p
->block_id
= peer_req
->block_id
;
1700 p
->seq_num
= 0; /* unused */
1703 drbd_csum_ee(device
, first_peer_device(device
)->connection
->integrity_tfm
, peer_req
, p
+ 1);
1704 err
= __send_command(first_peer_device(device
)->connection
, device
->vnr
, sock
, cmd
, sizeof(*p
) + dgs
, NULL
, peer_req
->i
.size
);
1706 err
= _drbd_send_zc_ee(device
, peer_req
);
1707 mutex_unlock(&sock
->mutex
); /* locked by drbd_prepare_command() */
1712 int drbd_send_out_of_sync(struct drbd_device
*device
, struct drbd_request
*req
)
1714 struct drbd_socket
*sock
;
1715 struct p_block_desc
*p
;
1717 sock
= &first_peer_device(device
)->connection
->data
;
1718 p
= drbd_prepare_command(device
, sock
);
1721 p
->sector
= cpu_to_be64(req
->i
.sector
);
1722 p
->blksize
= cpu_to_be32(req
->i
.size
);
1723 return drbd_send_command(device
, sock
, P_OUT_OF_SYNC
, sizeof(*p
), NULL
, 0);
1727 drbd_send distinguishes two cases:
1729 Packets sent via the data socket "sock"
1730 and packets sent via the meta data socket "msock"
1733 -----------------+-------------------------+------------------------------
1734 timeout conf.timeout / 2 conf.timeout / 2
1735 timeout action send a ping via msock Abort communication
1736 and close all sockets
1740 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1742 int drbd_send(struct drbd_connection
*connection
, struct socket
*sock
,
1743 void *buf
, size_t size
, unsigned msg_flags
)
1752 /* THINK if (signal_pending) return ... ? */
1757 msg
.msg_name
= NULL
;
1758 msg
.msg_namelen
= 0;
1759 msg
.msg_control
= NULL
;
1760 msg
.msg_controllen
= 0;
1761 msg
.msg_flags
= msg_flags
| MSG_NOSIGNAL
;
1763 if (sock
== connection
->data
.socket
) {
1765 connection
->ko_count
= rcu_dereference(connection
->net_conf
)->ko_count
;
1767 drbd_update_congested(connection
);
1771 * tcp_sendmsg does _not_ use its size parameter at all ?
1773 * -EAGAIN on timeout, -EINTR on signal.
1776 * do we need to block DRBD_SIG if sock == &meta.socket ??
1777 * otherwise wake_asender() might interrupt some send_*Ack !
1779 rv
= kernel_sendmsg(sock
, &msg
, &iov
, 1, size
);
1780 if (rv
== -EAGAIN
) {
1781 if (we_should_drop_the_connection(connection
, sock
))
1787 flush_signals(current
);
1795 } while (sent
< size
);
1797 if (sock
== connection
->data
.socket
)
1798 clear_bit(NET_CONGESTED
, &connection
->flags
);
1801 if (rv
!= -EAGAIN
) {
1802 conn_err(connection
, "%s_sendmsg returned %d\n",
1803 sock
== connection
->meta
.socket
? "msock" : "sock",
1805 conn_request_state(connection
, NS(conn
, C_BROKEN_PIPE
), CS_HARD
);
1807 conn_request_state(connection
, NS(conn
, C_TIMEOUT
), CS_HARD
);
1814 * drbd_send_all - Send an entire buffer
1816 * Returns 0 upon success and a negative error value otherwise.
1818 int drbd_send_all(struct drbd_connection
*connection
, struct socket
*sock
, void *buffer
,
1819 size_t size
, unsigned msg_flags
)
1823 err
= drbd_send(connection
, sock
, buffer
, size
, msg_flags
);
1831 static int drbd_open(struct block_device
*bdev
, fmode_t mode
)
1833 struct drbd_device
*device
= bdev
->bd_disk
->private_data
;
1834 unsigned long flags
;
1837 mutex_lock(&drbd_main_mutex
);
1838 spin_lock_irqsave(&first_peer_device(device
)->connection
->req_lock
, flags
);
1839 /* to have a stable device->state.role
1840 * and no race with updating open_cnt */
1842 if (device
->state
.role
!= R_PRIMARY
) {
1843 if (mode
& FMODE_WRITE
)
1845 else if (!allow_oos
)
1851 spin_unlock_irqrestore(&first_peer_device(device
)->connection
->req_lock
, flags
);
1852 mutex_unlock(&drbd_main_mutex
);
1857 static void drbd_release(struct gendisk
*gd
, fmode_t mode
)
1859 struct drbd_device
*device
= gd
->private_data
;
1860 mutex_lock(&drbd_main_mutex
);
1862 mutex_unlock(&drbd_main_mutex
);
1865 static void drbd_set_defaults(struct drbd_device
*device
)
1867 /* Beware! The actual layout differs
1868 * between big endian and little endian */
1869 device
->state
= (union drbd_dev_state
) {
1870 { .role
= R_SECONDARY
,
1872 .conn
= C_STANDALONE
,
1878 void drbd_init_set_defaults(struct drbd_device
*device
)
1880 /* the memset(,0,) did most of this.
1881 * note: only assignments, no allocation in here */
1883 drbd_set_defaults(device
);
1885 atomic_set(&device
->ap_bio_cnt
, 0);
1886 atomic_set(&device
->ap_pending_cnt
, 0);
1887 atomic_set(&device
->rs_pending_cnt
, 0);
1888 atomic_set(&device
->unacked_cnt
, 0);
1889 atomic_set(&device
->local_cnt
, 0);
1890 atomic_set(&device
->pp_in_use_by_net
, 0);
1891 atomic_set(&device
->rs_sect_in
, 0);
1892 atomic_set(&device
->rs_sect_ev
, 0);
1893 atomic_set(&device
->ap_in_flight
, 0);
1894 atomic_set(&device
->md_io_in_use
, 0);
1896 mutex_init(&device
->own_state_mutex
);
1897 device
->state_mutex
= &device
->own_state_mutex
;
1899 spin_lock_init(&device
->al_lock
);
1900 spin_lock_init(&device
->peer_seq_lock
);
1902 INIT_LIST_HEAD(&device
->active_ee
);
1903 INIT_LIST_HEAD(&device
->sync_ee
);
1904 INIT_LIST_HEAD(&device
->done_ee
);
1905 INIT_LIST_HEAD(&device
->read_ee
);
1906 INIT_LIST_HEAD(&device
->net_ee
);
1907 INIT_LIST_HEAD(&device
->resync_reads
);
1908 INIT_LIST_HEAD(&device
->resync_work
.list
);
1909 INIT_LIST_HEAD(&device
->unplug_work
.list
);
1910 INIT_LIST_HEAD(&device
->go_diskless
.list
);
1911 INIT_LIST_HEAD(&device
->md_sync_work
.list
);
1912 INIT_LIST_HEAD(&device
->start_resync_work
.list
);
1913 INIT_LIST_HEAD(&device
->bm_io_work
.w
.list
);
1915 device
->resync_work
.cb
= w_resync_timer
;
1916 device
->unplug_work
.cb
= w_send_write_hint
;
1917 device
->go_diskless
.cb
= w_go_diskless
;
1918 device
->md_sync_work
.cb
= w_md_sync
;
1919 device
->bm_io_work
.w
.cb
= w_bitmap_io
;
1920 device
->start_resync_work
.cb
= w_start_resync
;
1922 device
->resync_work
.device
= device
;
1923 device
->unplug_work
.device
= device
;
1924 device
->go_diskless
.device
= device
;
1925 device
->md_sync_work
.device
= device
;
1926 device
->bm_io_work
.w
.device
= device
;
1927 device
->start_resync_work
.device
= device
;
1929 init_timer(&device
->resync_timer
);
1930 init_timer(&device
->md_sync_timer
);
1931 init_timer(&device
->start_resync_timer
);
1932 init_timer(&device
->request_timer
);
1933 device
->resync_timer
.function
= resync_timer_fn
;
1934 device
->resync_timer
.data
= (unsigned long) device
;
1935 device
->md_sync_timer
.function
= md_sync_timer_fn
;
1936 device
->md_sync_timer
.data
= (unsigned long) device
;
1937 device
->start_resync_timer
.function
= start_resync_timer_fn
;
1938 device
->start_resync_timer
.data
= (unsigned long) device
;
1939 device
->request_timer
.function
= request_timer_fn
;
1940 device
->request_timer
.data
= (unsigned long) device
;
1942 init_waitqueue_head(&device
->misc_wait
);
1943 init_waitqueue_head(&device
->state_wait
);
1944 init_waitqueue_head(&device
->ee_wait
);
1945 init_waitqueue_head(&device
->al_wait
);
1946 init_waitqueue_head(&device
->seq_wait
);
1948 device
->resync_wenr
= LC_FREE
;
1949 device
->peer_max_bio_size
= DRBD_MAX_BIO_SIZE_SAFE
;
1950 device
->local_max_bio_size
= DRBD_MAX_BIO_SIZE_SAFE
;
1953 void drbd_device_cleanup(struct drbd_device
*device
)
1956 if (first_peer_device(device
)->connection
->receiver
.t_state
!= NONE
)
1957 dev_err(DEV
, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
1958 first_peer_device(device
)->connection
->receiver
.t_state
);
1960 device
->al_writ_cnt
=
1961 device
->bm_writ_cnt
=
1969 device
->rs_failed
= 0;
1970 device
->rs_last_events
= 0;
1971 device
->rs_last_sect_ev
= 0;
1972 for (i
= 0; i
< DRBD_SYNC_MARKS
; i
++) {
1973 device
->rs_mark_left
[i
] = 0;
1974 device
->rs_mark_time
[i
] = 0;
1976 D_ASSERT(first_peer_device(device
)->connection
->net_conf
== NULL
);
1978 drbd_set_my_capacity(device
, 0);
1979 if (device
->bitmap
) {
1980 /* maybe never allocated. */
1981 drbd_bm_resize(device
, 0, 1);
1982 drbd_bm_cleanup(device
);
1985 drbd_free_bc(device
->ldev
);
1986 device
->ldev
= NULL
;
1988 clear_bit(AL_SUSPENDED
, &device
->flags
);
1990 D_ASSERT(list_empty(&device
->active_ee
));
1991 D_ASSERT(list_empty(&device
->sync_ee
));
1992 D_ASSERT(list_empty(&device
->done_ee
));
1993 D_ASSERT(list_empty(&device
->read_ee
));
1994 D_ASSERT(list_empty(&device
->net_ee
));
1995 D_ASSERT(list_empty(&device
->resync_reads
));
1996 D_ASSERT(list_empty(&first_peer_device(device
)->connection
->sender_work
.q
));
1997 D_ASSERT(list_empty(&device
->resync_work
.list
));
1998 D_ASSERT(list_empty(&device
->unplug_work
.list
));
1999 D_ASSERT(list_empty(&device
->go_diskless
.list
));
2001 drbd_set_defaults(device
);
2005 static void drbd_destroy_mempools(void)
2009 while (drbd_pp_pool
) {
2010 page
= drbd_pp_pool
;
2011 drbd_pp_pool
= (struct page
*)page_private(page
);
2016 /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */
2018 if (drbd_md_io_bio_set
)
2019 bioset_free(drbd_md_io_bio_set
);
2020 if (drbd_md_io_page_pool
)
2021 mempool_destroy(drbd_md_io_page_pool
);
2022 if (drbd_ee_mempool
)
2023 mempool_destroy(drbd_ee_mempool
);
2024 if (drbd_request_mempool
)
2025 mempool_destroy(drbd_request_mempool
);
2027 kmem_cache_destroy(drbd_ee_cache
);
2028 if (drbd_request_cache
)
2029 kmem_cache_destroy(drbd_request_cache
);
2030 if (drbd_bm_ext_cache
)
2031 kmem_cache_destroy(drbd_bm_ext_cache
);
2032 if (drbd_al_ext_cache
)
2033 kmem_cache_destroy(drbd_al_ext_cache
);
2035 drbd_md_io_bio_set
= NULL
;
2036 drbd_md_io_page_pool
= NULL
;
2037 drbd_ee_mempool
= NULL
;
2038 drbd_request_mempool
= NULL
;
2039 drbd_ee_cache
= NULL
;
2040 drbd_request_cache
= NULL
;
2041 drbd_bm_ext_cache
= NULL
;
2042 drbd_al_ext_cache
= NULL
;
2047 static int drbd_create_mempools(void)
2050 const int number
= (DRBD_MAX_BIO_SIZE
/PAGE_SIZE
) * minor_count
;
2053 /* prepare our caches and mempools */
2054 drbd_request_mempool
= NULL
;
2055 drbd_ee_cache
= NULL
;
2056 drbd_request_cache
= NULL
;
2057 drbd_bm_ext_cache
= NULL
;
2058 drbd_al_ext_cache
= NULL
;
2059 drbd_pp_pool
= NULL
;
2060 drbd_md_io_page_pool
= NULL
;
2061 drbd_md_io_bio_set
= NULL
;
2064 drbd_request_cache
= kmem_cache_create(
2065 "drbd_req", sizeof(struct drbd_request
), 0, 0, NULL
);
2066 if (drbd_request_cache
== NULL
)
2069 drbd_ee_cache
= kmem_cache_create(
2070 "drbd_ee", sizeof(struct drbd_peer_request
), 0, 0, NULL
);
2071 if (drbd_ee_cache
== NULL
)
2074 drbd_bm_ext_cache
= kmem_cache_create(
2075 "drbd_bm", sizeof(struct bm_extent
), 0, 0, NULL
);
2076 if (drbd_bm_ext_cache
== NULL
)
2079 drbd_al_ext_cache
= kmem_cache_create(
2080 "drbd_al", sizeof(struct lc_element
), 0, 0, NULL
);
2081 if (drbd_al_ext_cache
== NULL
)
2085 drbd_md_io_bio_set
= bioset_create(DRBD_MIN_POOL_PAGES
, 0);
2086 if (drbd_md_io_bio_set
== NULL
)
2089 drbd_md_io_page_pool
= mempool_create_page_pool(DRBD_MIN_POOL_PAGES
, 0);
2090 if (drbd_md_io_page_pool
== NULL
)
2093 drbd_request_mempool
= mempool_create(number
,
2094 mempool_alloc_slab
, mempool_free_slab
, drbd_request_cache
);
2095 if (drbd_request_mempool
== NULL
)
2098 drbd_ee_mempool
= mempool_create(number
,
2099 mempool_alloc_slab
, mempool_free_slab
, drbd_ee_cache
);
2100 if (drbd_ee_mempool
== NULL
)
2103 /* drbd's page pool */
2104 spin_lock_init(&drbd_pp_lock
);
2106 for (i
= 0; i
< number
; i
++) {
2107 page
= alloc_page(GFP_HIGHUSER
);
2110 set_page_private(page
, (unsigned long)drbd_pp_pool
);
2111 drbd_pp_pool
= page
;
2113 drbd_pp_vacant
= number
;
2118 drbd_destroy_mempools(); /* in case we allocated some */
2122 static int drbd_notify_sys(struct notifier_block
*this, unsigned long code
,
2125 /* just so we have it. you never know what interesting things we
2126 * might want to do here some day...
2132 static struct notifier_block drbd_notifier
= {
2133 .notifier_call
= drbd_notify_sys
,
2136 static void drbd_release_all_peer_reqs(struct drbd_device
*device
)
2140 rr
= drbd_free_peer_reqs(device
, &device
->active_ee
);
2142 dev_err(DEV
, "%d EEs in active list found!\n", rr
);
2144 rr
= drbd_free_peer_reqs(device
, &device
->sync_ee
);
2146 dev_err(DEV
, "%d EEs in sync list found!\n", rr
);
2148 rr
= drbd_free_peer_reqs(device
, &device
->read_ee
);
2150 dev_err(DEV
, "%d EEs in read list found!\n", rr
);
2152 rr
= drbd_free_peer_reqs(device
, &device
->done_ee
);
2154 dev_err(DEV
, "%d EEs in done list found!\n", rr
);
2156 rr
= drbd_free_peer_reqs(device
, &device
->net_ee
);
2158 dev_err(DEV
, "%d EEs in net list found!\n", rr
);
2161 /* caution. no locking. */
2162 void drbd_destroy_device(struct kref
*kref
)
2164 struct drbd_device
*device
= container_of(kref
, struct drbd_device
, kref
);
2165 struct drbd_resource
*resource
= device
->resource
;
2166 struct drbd_connection
*connection
;
2168 del_timer_sync(&device
->request_timer
);
2170 /* paranoia asserts */
2171 D_ASSERT(device
->open_cnt
== 0);
2172 /* end paranoia asserts */
2174 /* cleanup stuff that may have been allocated during
2175 * device (re-)configuration or state changes */
2177 if (device
->this_bdev
)
2178 bdput(device
->this_bdev
);
2180 drbd_free_bc(device
->ldev
);
2181 device
->ldev
= NULL
;
2183 drbd_release_all_peer_reqs(device
);
2185 lc_destroy(device
->act_log
);
2186 lc_destroy(device
->resync
);
2188 kfree(device
->p_uuid
);
2189 /* device->p_uuid = NULL; */
2191 if (device
->bitmap
) /* should no longer be there. */
2192 drbd_bm_cleanup(device
);
2193 __free_page(device
->md_io_page
);
2194 put_disk(device
->vdisk
);
2195 blk_cleanup_queue(device
->rq_queue
);
2196 kfree(device
->rs_plan_s
);
2197 kfree(first_peer_device(device
));
2200 for_each_connection(connection
, resource
)
2201 kref_put(&connection
->kref
, drbd_destroy_connection
);
2202 kref_put(&resource
->kref
, drbd_destroy_resource
);
2205 /* One global retry thread, if we need to push back some bio and have it
2206 * reinserted through our make request function.
2208 static struct retry_worker
{
2209 struct workqueue_struct
*wq
;
2210 struct work_struct worker
;
2213 struct list_head writes
;
2216 static void do_retry(struct work_struct
*ws
)
2218 struct retry_worker
*retry
= container_of(ws
, struct retry_worker
, worker
);
2220 struct drbd_request
*req
, *tmp
;
2222 spin_lock_irq(&retry
->lock
);
2223 list_splice_init(&retry
->writes
, &writes
);
2224 spin_unlock_irq(&retry
->lock
);
2226 list_for_each_entry_safe(req
, tmp
, &writes
, tl_requests
) {
2227 struct drbd_device
*device
= req
->w
.device
;
2228 struct bio
*bio
= req
->master_bio
;
2229 unsigned long start_time
= req
->start_time
;
2233 expect(atomic_read(&req
->completion_ref
) == 0) &&
2234 expect(req
->rq_state
& RQ_POSTPONED
) &&
2235 expect((req
->rq_state
& RQ_LOCAL_PENDING
) == 0 ||
2236 (req
->rq_state
& RQ_LOCAL_ABORTED
) != 0);
2239 dev_err(DEV
, "req=%p completion_ref=%d rq_state=%x\n",
2240 req
, atomic_read(&req
->completion_ref
),
2243 /* We still need to put one kref associated with the
2244 * "completion_ref" going zero in the code path that queued it
2245 * here. The request object may still be referenced by a
2246 * frozen local req->private_bio, in case we force-detached.
2248 kref_put(&req
->kref
, drbd_req_destroy
);
2250 /* A single suspended or otherwise blocking device may stall
2251 * all others as well. Fortunately, this code path is to
2252 * recover from a situation that "should not happen":
2253 * concurrent writes in multi-primary setup.
2254 * In a "normal" lifecycle, this workqueue is supposed to be
2255 * destroyed without ever doing anything.
2256 * If it turns out to be an issue anyways, we can do per
2257 * resource (replication group) or per device (minor) retry
2258 * workqueues instead.
2261 /* We are not just doing generic_make_request(),
2262 * as we want to keep the start_time information. */
2264 __drbd_make_request(device
, bio
, start_time
);
2268 void drbd_restart_request(struct drbd_request
*req
)
2270 unsigned long flags
;
2271 spin_lock_irqsave(&retry
.lock
, flags
);
2272 list_move_tail(&req
->tl_requests
, &retry
.writes
);
2273 spin_unlock_irqrestore(&retry
.lock
, flags
);
2275 /* Drop the extra reference that would otherwise
2276 * have been dropped by complete_master_bio.
2277 * do_retry() needs to grab a new one. */
2278 dec_ap_bio(req
->w
.device
);
2280 queue_work(retry
.wq
, &retry
.worker
);
2283 void drbd_destroy_resource(struct kref
*kref
)
2285 struct drbd_resource
*resource
=
2286 container_of(kref
, struct drbd_resource
, kref
);
2288 idr_destroy(&resource
->devices
);
2289 kfree(resource
->name
);
2293 void drbd_free_resource(struct drbd_resource
*resource
)
2295 struct drbd_connection
*connection
, *tmp
;
2297 for_each_connection_safe(connection
, tmp
, resource
) {
2298 list_del(&connection
->connections
);
2299 kref_put(&connection
->kref
, drbd_destroy_connection
);
2301 kref_put(&resource
->kref
, drbd_destroy_resource
);
2304 static void drbd_cleanup(void)
2307 struct drbd_device
*device
;
2308 struct drbd_resource
*resource
, *tmp
;
2310 unregister_reboot_notifier(&drbd_notifier
);
2312 /* first remove proc,
2313 * drbdsetup uses it's presence to detect
2314 * whether DRBD is loaded.
2315 * If we would get stuck in proc removal,
2316 * but have netlink already deregistered,
2317 * some drbdsetup commands may wait forever
2321 remove_proc_entry("drbd", NULL
);
2324 destroy_workqueue(retry
.wq
);
2326 drbd_genl_unregister();
2328 idr_for_each_entry(&drbd_devices
, device
, i
)
2329 drbd_delete_minor(device
);
2331 /* not _rcu since, no other updater anymore. Genl already unregistered */
2332 for_each_resource_safe(resource
, tmp
, &drbd_resources
) {
2333 list_del(&resource
->resources
);
2334 drbd_free_resource(resource
);
2337 drbd_destroy_mempools();
2338 unregister_blkdev(DRBD_MAJOR
, "drbd");
2340 idr_destroy(&drbd_devices
);
2342 printk(KERN_INFO
"drbd: module cleanup done.\n");
2346 * drbd_congested() - Callback for the flusher thread
2347 * @congested_data: User data
2348 * @bdi_bits: Bits the BDI flusher thread is currently interested in
2350 * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2352 static int drbd_congested(void *congested_data
, int bdi_bits
)
2354 struct drbd_device
*device
= congested_data
;
2355 struct request_queue
*q
;
2359 if (!may_inc_ap_bio(device
)) {
2360 /* DRBD has frozen IO */
2366 if (test_bit(CALLBACK_PENDING
, &first_peer_device(device
)->connection
->flags
)) {
2367 r
|= (1 << BDI_async_congested
);
2368 /* Without good local data, we would need to read from remote,
2369 * and that would need the worker thread as well, which is
2370 * currently blocked waiting for that usermode helper to
2373 if (!get_ldev_if_state(device
, D_UP_TO_DATE
))
2374 r
|= (1 << BDI_sync_congested
);
2382 if (get_ldev(device
)) {
2383 q
= bdev_get_queue(device
->ldev
->backing_bdev
);
2384 r
= bdi_congested(&q
->backing_dev_info
, bdi_bits
);
2390 if (bdi_bits
& (1 << BDI_async_congested
) &&
2391 test_bit(NET_CONGESTED
, &first_peer_device(device
)->connection
->flags
)) {
2392 r
|= (1 << BDI_async_congested
);
2393 reason
= reason
== 'b' ? 'a' : 'n';
2397 device
->congestion_reason
= reason
;
2401 static void drbd_init_workqueue(struct drbd_work_queue
* wq
)
2403 spin_lock_init(&wq
->q_lock
);
2404 INIT_LIST_HEAD(&wq
->q
);
2405 init_waitqueue_head(&wq
->q_wait
);
2408 struct drbd_resource
*drbd_find_resource(const char *name
)
2410 struct drbd_resource
*resource
;
2412 if (!name
|| !name
[0])
2416 for_each_resource_rcu(resource
, &drbd_resources
) {
2417 if (!strcmp(resource
->name
, name
)) {
2418 kref_get(&resource
->kref
);
2428 struct drbd_connection
*conn_get_by_addrs(void *my_addr
, int my_addr_len
,
2429 void *peer_addr
, int peer_addr_len
)
2431 struct drbd_resource
*resource
;
2432 struct drbd_connection
*connection
;
2435 for_each_resource_rcu(resource
, &drbd_resources
) {
2436 for_each_connection_rcu(connection
, resource
) {
2437 if (connection
->my_addr_len
== my_addr_len
&&
2438 connection
->peer_addr_len
== peer_addr_len
&&
2439 !memcmp(&connection
->my_addr
, my_addr
, my_addr_len
) &&
2440 !memcmp(&connection
->peer_addr
, peer_addr
, peer_addr_len
)) {
2441 kref_get(&connection
->kref
);
2452 static int drbd_alloc_socket(struct drbd_socket
*socket
)
2454 socket
->rbuf
= (void *) __get_free_page(GFP_KERNEL
);
2457 socket
->sbuf
= (void *) __get_free_page(GFP_KERNEL
);
2463 static void drbd_free_socket(struct drbd_socket
*socket
)
2465 free_page((unsigned long) socket
->sbuf
);
2466 free_page((unsigned long) socket
->rbuf
);
2469 void conn_free_crypto(struct drbd_connection
*connection
)
2471 drbd_free_sock(connection
);
2473 crypto_free_hash(connection
->csums_tfm
);
2474 crypto_free_hash(connection
->verify_tfm
);
2475 crypto_free_hash(connection
->cram_hmac_tfm
);
2476 crypto_free_hash(connection
->integrity_tfm
);
2477 crypto_free_hash(connection
->peer_integrity_tfm
);
2478 kfree(connection
->int_dig_in
);
2479 kfree(connection
->int_dig_vv
);
2481 connection
->csums_tfm
= NULL
;
2482 connection
->verify_tfm
= NULL
;
2483 connection
->cram_hmac_tfm
= NULL
;
2484 connection
->integrity_tfm
= NULL
;
2485 connection
->peer_integrity_tfm
= NULL
;
2486 connection
->int_dig_in
= NULL
;
2487 connection
->int_dig_vv
= NULL
;
2490 int set_resource_options(struct drbd_resource
*resource
, struct res_opts
*res_opts
)
2492 struct drbd_connection
*connection
;
2493 cpumask_var_t new_cpu_mask
;
2496 if (!zalloc_cpumask_var(&new_cpu_mask
, GFP_KERNEL
))
2499 retcode = ERR_NOMEM;
2500 drbd_msg_put_info("unable to allocate cpumask");
2503 /* silently ignore cpu mask on UP kernel */
2504 if (nr_cpu_ids
> 1 && res_opts
->cpu_mask
[0] != 0) {
2505 /* FIXME: Get rid of constant 32 here */
2506 err
= bitmap_parse(res_opts
->cpu_mask
, 32,
2507 cpumask_bits(new_cpu_mask
), nr_cpu_ids
);
2509 conn_warn(connection
, "bitmap_parse() failed with %d\n", err
);
2510 /* retcode = ERR_CPU_MASK_PARSE; */
2514 resource
->res_opts
= *res_opts
;
2515 for_each_connection_rcu(connection
, resource
) {
2516 if (!cpumask_equal(connection
->cpu_mask
, new_cpu_mask
)) {
2517 cpumask_copy(connection
->cpu_mask
, new_cpu_mask
);
2518 drbd_calc_cpu_mask(connection
);
2519 connection
->receiver
.reset_cpu_mask
= 1;
2520 connection
->asender
.reset_cpu_mask
= 1;
2521 connection
->worker
.reset_cpu_mask
= 1;
2527 free_cpumask_var(new_cpu_mask
);
2532 struct drbd_resource
*drbd_create_resource(const char *name
)
2534 struct drbd_resource
*resource
;
2536 resource
= kmalloc(sizeof(struct drbd_resource
), GFP_KERNEL
);
2539 resource
->name
= kstrdup(name
, GFP_KERNEL
);
2540 if (!resource
->name
) {
2544 kref_init(&resource
->kref
);
2545 idr_init(&resource
->devices
);
2546 INIT_LIST_HEAD(&resource
->connections
);
2547 list_add_tail_rcu(&resource
->resources
, &drbd_resources
);
2551 /* caller must be under genl_lock() */
2552 struct drbd_connection
*conn_create(const char *name
, struct res_opts
*res_opts
)
2554 struct drbd_resource
*resource
;
2555 struct drbd_connection
*connection
;
2557 connection
= kzalloc(sizeof(struct drbd_connection
), GFP_KERNEL
);
2561 if (drbd_alloc_socket(&connection
->data
))
2563 if (drbd_alloc_socket(&connection
->meta
))
2566 if (!zalloc_cpumask_var(&connection
->cpu_mask
, GFP_KERNEL
))
2569 connection
->current_epoch
= kzalloc(sizeof(struct drbd_epoch
), GFP_KERNEL
);
2570 if (!connection
->current_epoch
)
2573 INIT_LIST_HEAD(&connection
->transfer_log
);
2575 INIT_LIST_HEAD(&connection
->current_epoch
->list
);
2576 connection
->epochs
= 1;
2577 spin_lock_init(&connection
->epoch_lock
);
2578 connection
->write_ordering
= WO_bdev_flush
;
2580 connection
->send
.seen_any_write_yet
= false;
2581 connection
->send
.current_epoch_nr
= 0;
2582 connection
->send
.current_epoch_writes
= 0;
2584 resource
= drbd_create_resource(name
);
2588 connection
->cstate
= C_STANDALONE
;
2589 mutex_init(&connection
->cstate_mutex
);
2590 spin_lock_init(&connection
->req_lock
);
2591 mutex_init(&connection
->conf_update
);
2592 init_waitqueue_head(&connection
->ping_wait
);
2593 idr_init(&connection
->volumes
);
2595 drbd_init_workqueue(&connection
->sender_work
);
2596 mutex_init(&connection
->data
.mutex
);
2597 mutex_init(&connection
->meta
.mutex
);
2599 drbd_thread_init(connection
, &connection
->receiver
, drbdd_init
, "receiver");
2600 drbd_thread_init(connection
, &connection
->worker
, drbd_worker
, "worker");
2601 drbd_thread_init(connection
, &connection
->asender
, drbd_asender
, "asender");
2603 kref_init(&connection
->kref
);
2605 connection
->resource
= resource
;
2607 if (set_resource_options(resource
, res_opts
))
2610 kref_get(&resource
->kref
);
2611 list_add_tail_rcu(&connection
->connections
, &resource
->connections
);
2615 list_del(&resource
->resources
);
2616 drbd_free_resource(resource
);
2618 kfree(connection
->current_epoch
);
2619 free_cpumask_var(connection
->cpu_mask
);
2620 drbd_free_socket(&connection
->meta
);
2621 drbd_free_socket(&connection
->data
);
2626 void drbd_destroy_connection(struct kref
*kref
)
2628 struct drbd_connection
*connection
= container_of(kref
, struct drbd_connection
, kref
);
2629 struct drbd_resource
*resource
= connection
->resource
;
2631 if (atomic_read(&connection
->current_epoch
->epoch_size
) != 0)
2632 conn_err(connection
, "epoch_size:%d\n", atomic_read(&connection
->current_epoch
->epoch_size
));
2633 kfree(connection
->current_epoch
);
2635 idr_destroy(&connection
->volumes
);
2637 free_cpumask_var(connection
->cpu_mask
);
2638 drbd_free_socket(&connection
->meta
);
2639 drbd_free_socket(&connection
->data
);
2640 kfree(connection
->int_dig_in
);
2641 kfree(connection
->int_dig_vv
);
2643 kref_put(&resource
->kref
, drbd_destroy_resource
);
2646 static int init_submitter(struct drbd_device
*device
)
2648 /* opencoded create_singlethread_workqueue(),
2649 * to be able to say "drbd%d", ..., minor */
2650 device
->submit
.wq
= alloc_workqueue("drbd%u_submit",
2651 WQ_UNBOUND
| WQ_MEM_RECLAIM
, 1, device
->minor
);
2652 if (!device
->submit
.wq
)
2655 INIT_WORK(&device
->submit
.worker
, do_submit
);
2656 spin_lock_init(&device
->submit
.lock
);
2657 INIT_LIST_HEAD(&device
->submit
.writes
);
2661 enum drbd_ret_code
drbd_create_minor(struct drbd_connection
*connection
, unsigned int minor
, int vnr
)
2663 struct drbd_resource
*resource
= connection
->resource
;
2664 struct drbd_device
*device
;
2665 struct drbd_peer_device
*peer_device
;
2666 struct gendisk
*disk
;
2667 struct request_queue
*q
;
2669 enum drbd_ret_code err
= ERR_NOMEM
;
2671 device
= minor_to_device(minor
);
2673 return ERR_MINOR_EXISTS
;
2675 /* GFP_KERNEL, we are outside of all write-out paths */
2676 device
= kzalloc(sizeof(struct drbd_device
), GFP_KERNEL
);
2679 kref_init(&device
->kref
);
2681 peer_device
= kzalloc(sizeof(struct drbd_peer_device
), GFP_KERNEL
);
2683 goto out_no_peer_device
;
2685 INIT_LIST_HEAD(&device
->peer_devices
);
2686 list_add(&peer_device
->peer_devices
, &device
->peer_devices
);
2687 kref_get(&resource
->kref
);
2688 device
->resource
= resource
;
2689 kref_get(&connection
->kref
);
2690 peer_device
->connection
= connection
;
2691 peer_device
->device
= device
;
2693 device
->minor
= minor
;
2696 drbd_init_set_defaults(device
);
2698 q
= blk_alloc_queue(GFP_KERNEL
);
2701 device
->rq_queue
= q
;
2702 q
->queuedata
= device
;
2704 disk
= alloc_disk(1);
2707 device
->vdisk
= disk
;
2709 set_disk_ro(disk
, true);
2712 disk
->major
= DRBD_MAJOR
;
2713 disk
->first_minor
= minor
;
2714 disk
->fops
= &drbd_ops
;
2715 sprintf(disk
->disk_name
, "drbd%d", minor
);
2716 disk
->private_data
= device
;
2718 device
->this_bdev
= bdget(MKDEV(DRBD_MAJOR
, minor
));
2719 /* we have no partitions. we contain only ourselves. */
2720 device
->this_bdev
->bd_contains
= device
->this_bdev
;
2722 q
->backing_dev_info
.congested_fn
= drbd_congested
;
2723 q
->backing_dev_info
.congested_data
= device
;
2725 blk_queue_make_request(q
, drbd_make_request
);
2726 blk_queue_flush(q
, REQ_FLUSH
| REQ_FUA
);
2727 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2728 This triggers a max_bio_size message upon first attach or connect */
2729 blk_queue_max_hw_sectors(q
, DRBD_MAX_BIO_SIZE_SAFE
>> 8);
2730 blk_queue_bounce_limit(q
, BLK_BOUNCE_ANY
);
2731 blk_queue_merge_bvec(q
, drbd_merge_bvec
);
2732 q
->queue_lock
= &connection
->req_lock
;
2734 device
->md_io_page
= alloc_page(GFP_KERNEL
);
2735 if (!device
->md_io_page
)
2736 goto out_no_io_page
;
2738 if (drbd_bm_init(device
))
2740 device
->read_requests
= RB_ROOT
;
2741 device
->write_requests
= RB_ROOT
;
2743 id
= idr_alloc(&drbd_devices
, device
, minor
, minor
+ 1, GFP_KERNEL
);
2745 if (id
== -ENOSPC
) {
2746 err
= ERR_MINOR_EXISTS
;
2747 drbd_msg_put_info("requested minor exists already");
2749 goto out_no_minor_idr
;
2751 kref_get(&device
->kref
);
2753 id
= idr_alloc(&resource
->devices
, device
, vnr
, vnr
+ 1, GFP_KERNEL
);
2755 if (id
== -ENOSPC
) {
2756 err
= ERR_MINOR_EXISTS
;
2757 drbd_msg_put_info("requested minor exists already");
2759 goto out_idr_remove_minor
;
2761 kref_get(&device
->kref
);
2763 id
= idr_alloc(&connection
->volumes
, device
, vnr
, vnr
+ 1, GFP_KERNEL
);
2765 if (id
== -ENOSPC
) {
2766 err
= ERR_INVALID_REQUEST
;
2767 drbd_msg_put_info("requested volume exists already");
2769 goto out_idr_remove_from_resource
;
2771 kref_get(&device
->kref
);
2773 if (init_submitter(device
)) {
2775 drbd_msg_put_info("unable to create submit workqueue");
2776 goto out_idr_remove_vol
;
2781 /* inherit the connection state */
2782 device
->state
.conn
= connection
->cstate
;
2783 if (device
->state
.conn
== C_WF_REPORT_PARAMS
)
2784 drbd_connected(device
);
2789 idr_remove(&connection
->volumes
, vnr
);
2790 out_idr_remove_from_resource
:
2791 idr_remove(&resource
->devices
, vnr
);
2792 out_idr_remove_minor
:
2793 idr_remove(&drbd_devices
, minor
);
2796 drbd_bm_cleanup(device
);
2798 __free_page(device
->md_io_page
);
2802 blk_cleanup_queue(q
);
2804 kref_put(&connection
->kref
, drbd_destroy_connection
);
2805 kref_put(&resource
->kref
, drbd_destroy_resource
);
2811 void drbd_delete_minor(struct drbd_device
*device
)
2813 struct drbd_resource
*resource
= device
->resource
;
2814 struct drbd_connection
*connection
;
2817 for_each_connection(connection
, resource
) {
2818 idr_remove(&connection
->volumes
, device
->vnr
);
2821 idr_remove(&resource
->devices
, device
->vnr
);
2822 idr_remove(&drbd_devices
, device_to_minor(device
));
2823 del_gendisk(device
->vdisk
);
2825 kref_sub(&device
->kref
, refs
, drbd_destroy_device
);
2828 int __init
drbd_init(void)
2832 if (minor_count
< DRBD_MINOR_COUNT_MIN
|| minor_count
> DRBD_MINOR_COUNT_MAX
) {
2834 "drbd: invalid minor_count (%d)\n", minor_count
);
2838 minor_count
= DRBD_MINOR_COUNT_DEF
;
2842 err
= register_blkdev(DRBD_MAJOR
, "drbd");
2845 "drbd: unable to register block device major %d\n",
2850 register_reboot_notifier(&drbd_notifier
);
2853 * allocate all necessary structs
2855 init_waitqueue_head(&drbd_pp_wait
);
2857 drbd_proc
= NULL
; /* play safe for drbd_cleanup */
2858 idr_init(&drbd_devices
);
2860 rwlock_init(&global_state_lock
);
2861 INIT_LIST_HEAD(&drbd_resources
);
2863 err
= drbd_genl_register();
2865 printk(KERN_ERR
"drbd: unable to register generic netlink family\n");
2869 err
= drbd_create_mempools();
2874 drbd_proc
= proc_create_data("drbd", S_IFREG
| S_IRUGO
, NULL
, &drbd_proc_fops
, NULL
);
2876 printk(KERN_ERR
"drbd: unable to register proc file\n");
2880 retry
.wq
= create_singlethread_workqueue("drbd-reissue");
2882 printk(KERN_ERR
"drbd: unable to create retry workqueue\n");
2885 INIT_WORK(&retry
.worker
, do_retry
);
2886 spin_lock_init(&retry
.lock
);
2887 INIT_LIST_HEAD(&retry
.writes
);
2889 printk(KERN_INFO
"drbd: initialized. "
2890 "Version: " REL_VERSION
" (api:%d/proto:%d-%d)\n",
2891 API_VERSION
, PRO_VERSION_MIN
, PRO_VERSION_MAX
);
2892 printk(KERN_INFO
"drbd: %s\n", drbd_buildtag());
2893 printk(KERN_INFO
"drbd: registered as block device major %d\n",
2896 return 0; /* Success! */
2901 printk(KERN_ERR
"drbd: ran out of memory\n");
2903 printk(KERN_ERR
"drbd: initialization failure\n");
2907 void drbd_free_bc(struct drbd_backing_dev
*ldev
)
2912 blkdev_put(ldev
->backing_bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2913 blkdev_put(ldev
->md_bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2915 kfree(ldev
->disk_conf
);
2919 void drbd_free_sock(struct drbd_connection
*connection
)
2921 if (connection
->data
.socket
) {
2922 mutex_lock(&connection
->data
.mutex
);
2923 kernel_sock_shutdown(connection
->data
.socket
, SHUT_RDWR
);
2924 sock_release(connection
->data
.socket
);
2925 connection
->data
.socket
= NULL
;
2926 mutex_unlock(&connection
->data
.mutex
);
2928 if (connection
->meta
.socket
) {
2929 mutex_lock(&connection
->meta
.mutex
);
2930 kernel_sock_shutdown(connection
->meta
.socket
, SHUT_RDWR
);
2931 sock_release(connection
->meta
.socket
);
2932 connection
->meta
.socket
= NULL
;
2933 mutex_unlock(&connection
->meta
.mutex
);
2937 /* meta data management */
2939 void conn_md_sync(struct drbd_connection
*connection
)
2941 struct drbd_device
*device
;
2945 idr_for_each_entry(&connection
->volumes
, device
, vnr
) {
2946 kref_get(&device
->kref
);
2948 drbd_md_sync(device
);
2949 kref_put(&device
->kref
, drbd_destroy_device
);
2955 /* aligned 4kByte */
2956 struct meta_data_on_disk
{
2957 u64 la_size_sect
; /* last agreed size. */
2958 u64 uuid
[UI_SIZE
]; /* UUIDs. */
2961 u32 flags
; /* MDF */
2964 u32 al_offset
; /* offset to this block */
2965 u32 al_nr_extents
; /* important for restoring the AL (userspace) */
2966 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2967 u32 bm_offset
; /* offset to the bitmap, from here */
2968 u32 bm_bytes_per_bit
; /* BM_BLOCK_SIZE */
2969 u32 la_peer_max_bio_size
; /* last peer max_bio_size */
2971 /* see al_tr_number_to_on_disk_sector() */
2973 u32 al_stripe_size_4k
;
2975 u8 reserved_u8
[4096 - (7*8 + 10*4)];
2980 void drbd_md_write(struct drbd_device
*device
, void *b
)
2982 struct meta_data_on_disk
*buffer
= b
;
2986 memset(buffer
, 0, sizeof(*buffer
));
2988 buffer
->la_size_sect
= cpu_to_be64(drbd_get_capacity(device
->this_bdev
));
2989 for (i
= UI_CURRENT
; i
< UI_SIZE
; i
++)
2990 buffer
->uuid
[i
] = cpu_to_be64(device
->ldev
->md
.uuid
[i
]);
2991 buffer
->flags
= cpu_to_be32(device
->ldev
->md
.flags
);
2992 buffer
->magic
= cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN
);
2994 buffer
->md_size_sect
= cpu_to_be32(device
->ldev
->md
.md_size_sect
);
2995 buffer
->al_offset
= cpu_to_be32(device
->ldev
->md
.al_offset
);
2996 buffer
->al_nr_extents
= cpu_to_be32(device
->act_log
->nr_elements
);
2997 buffer
->bm_bytes_per_bit
= cpu_to_be32(BM_BLOCK_SIZE
);
2998 buffer
->device_uuid
= cpu_to_be64(device
->ldev
->md
.device_uuid
);
3000 buffer
->bm_offset
= cpu_to_be32(device
->ldev
->md
.bm_offset
);
3001 buffer
->la_peer_max_bio_size
= cpu_to_be32(device
->peer_max_bio_size
);
3003 buffer
->al_stripes
= cpu_to_be32(device
->ldev
->md
.al_stripes
);
3004 buffer
->al_stripe_size_4k
= cpu_to_be32(device
->ldev
->md
.al_stripe_size_4k
);
3006 D_ASSERT(drbd_md_ss(device
->ldev
) == device
->ldev
->md
.md_offset
);
3007 sector
= device
->ldev
->md
.md_offset
;
3009 if (drbd_md_sync_page_io(device
, device
->ldev
, sector
, WRITE
)) {
3010 /* this was a try anyways ... */
3011 dev_err(DEV
, "meta data update failed!\n");
3012 drbd_chk_io_error(device
, 1, DRBD_META_IO_ERROR
);
3017 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3018 * @device: DRBD device.
3020 void drbd_md_sync(struct drbd_device
*device
)
3022 struct meta_data_on_disk
*buffer
;
3024 /* Don't accidentally change the DRBD meta data layout. */
3025 BUILD_BUG_ON(UI_SIZE
!= 4);
3026 BUILD_BUG_ON(sizeof(struct meta_data_on_disk
) != 4096);
3028 del_timer(&device
->md_sync_timer
);
3029 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3030 if (!test_and_clear_bit(MD_DIRTY
, &device
->flags
))
3033 /* We use here D_FAILED and not D_ATTACHING because we try to write
3034 * metadata even if we detach due to a disk failure! */
3035 if (!get_ldev_if_state(device
, D_FAILED
))
3038 buffer
= drbd_md_get_buffer(device
);
3042 drbd_md_write(device
, buffer
);
3044 /* Update device->ldev->md.la_size_sect,
3045 * since we updated it on metadata. */
3046 device
->ldev
->md
.la_size_sect
= drbd_get_capacity(device
->this_bdev
);
3048 drbd_md_put_buffer(device
);
3053 static int check_activity_log_stripe_size(struct drbd_device
*device
,
3054 struct meta_data_on_disk
*on_disk
,
3055 struct drbd_md
*in_core
)
3057 u32 al_stripes
= be32_to_cpu(on_disk
->al_stripes
);
3058 u32 al_stripe_size_4k
= be32_to_cpu(on_disk
->al_stripe_size_4k
);
3061 /* both not set: default to old fixed size activity log */
3062 if (al_stripes
== 0 && al_stripe_size_4k
== 0) {
3064 al_stripe_size_4k
= MD_32kB_SECT
/8;
3067 /* some paranoia plausibility checks */
3069 /* we need both values to be set */
3070 if (al_stripes
== 0 || al_stripe_size_4k
== 0)
3073 al_size_4k
= (u64
)al_stripes
* al_stripe_size_4k
;
3075 /* Upper limit of activity log area, to avoid potential overflow
3076 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3077 * than 72 * 4k blocks total only increases the amount of history,
3078 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3079 if (al_size_4k
> (16 * 1024 * 1024/4))
3082 /* Lower limit: we need at least 8 transaction slots (32kB)
3083 * to not break existing setups */
3084 if (al_size_4k
< MD_32kB_SECT
/8)
3087 in_core
->al_stripe_size_4k
= al_stripe_size_4k
;
3088 in_core
->al_stripes
= al_stripes
;
3089 in_core
->al_size_4k
= al_size_4k
;
3093 dev_err(DEV
, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3094 al_stripes
, al_stripe_size_4k
);
3098 static int check_offsets_and_sizes(struct drbd_device
*device
, struct drbd_backing_dev
*bdev
)
3100 sector_t capacity
= drbd_get_capacity(bdev
->md_bdev
);
3101 struct drbd_md
*in_core
= &bdev
->md
;
3102 s32 on_disk_al_sect
;
3103 s32 on_disk_bm_sect
;
3105 /* The on-disk size of the activity log, calculated from offsets, and
3106 * the size of the activity log calculated from the stripe settings,
3108 * Though we could relax this a bit: it is ok, if the striped activity log
3109 * fits in the available on-disk activity log size.
3110 * Right now, that would break how resize is implemented.
3111 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3112 * of possible unused padding space in the on disk layout. */
3113 if (in_core
->al_offset
< 0) {
3114 if (in_core
->bm_offset
> in_core
->al_offset
)
3116 on_disk_al_sect
= -in_core
->al_offset
;
3117 on_disk_bm_sect
= in_core
->al_offset
- in_core
->bm_offset
;
3119 if (in_core
->al_offset
!= MD_4kB_SECT
)
3121 if (in_core
->bm_offset
< in_core
->al_offset
+ in_core
->al_size_4k
* MD_4kB_SECT
)
3124 on_disk_al_sect
= in_core
->bm_offset
- MD_4kB_SECT
;
3125 on_disk_bm_sect
= in_core
->md_size_sect
- in_core
->bm_offset
;
3128 /* old fixed size meta data is exactly that: fixed. */
3129 if (in_core
->meta_dev_idx
>= 0) {
3130 if (in_core
->md_size_sect
!= MD_128MB_SECT
3131 || in_core
->al_offset
!= MD_4kB_SECT
3132 || in_core
->bm_offset
!= MD_4kB_SECT
+ MD_32kB_SECT
3133 || in_core
->al_stripes
!= 1
3134 || in_core
->al_stripe_size_4k
!= MD_32kB_SECT
/8)
3138 if (capacity
< in_core
->md_size_sect
)
3140 if (capacity
- in_core
->md_size_sect
< drbd_md_first_sector(bdev
))
3143 /* should be aligned, and at least 32k */
3144 if ((on_disk_al_sect
& 7) || (on_disk_al_sect
< MD_32kB_SECT
))
3147 /* should fit (for now: exactly) into the available on-disk space;
3148 * overflow prevention is in check_activity_log_stripe_size() above. */
3149 if (on_disk_al_sect
!= in_core
->al_size_4k
* MD_4kB_SECT
)
3152 /* again, should be aligned */
3153 if (in_core
->bm_offset
& 7)
3156 /* FIXME check for device grow with flex external meta data? */
3158 /* can the available bitmap space cover the last agreed device size? */
3159 if (on_disk_bm_sect
< (in_core
->la_size_sect
+7)/MD_4kB_SECT
/8/512)
3165 dev_err(DEV
, "meta data offsets don't make sense: idx=%d "
3166 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3167 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3168 in_core
->meta_dev_idx
,
3169 in_core
->al_stripes
, in_core
->al_stripe_size_4k
,
3170 in_core
->al_offset
, in_core
->bm_offset
, in_core
->md_size_sect
,
3171 (unsigned long long)in_core
->la_size_sect
,
3172 (unsigned long long)capacity
);
3179 * drbd_md_read() - Reads in the meta data super block
3180 * @device: DRBD device.
3181 * @bdev: Device from which the meta data should be read in.
3183 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3184 * something goes wrong.
3186 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3187 * even before @bdev is assigned to @device->ldev.
3189 int drbd_md_read(struct drbd_device
*device
, struct drbd_backing_dev
*bdev
)
3191 struct meta_data_on_disk
*buffer
;
3193 int i
, rv
= NO_ERROR
;
3195 if (device
->state
.disk
!= D_DISKLESS
)
3196 return ERR_DISK_CONFIGURED
;
3198 buffer
= drbd_md_get_buffer(device
);
3202 /* First, figure out where our meta data superblock is located,
3204 bdev
->md
.meta_dev_idx
= bdev
->disk_conf
->meta_dev_idx
;
3205 bdev
->md
.md_offset
= drbd_md_ss(bdev
);
3207 if (drbd_md_sync_page_io(device
, bdev
, bdev
->md
.md_offset
, READ
)) {
3208 /* NOTE: can't do normal error processing here as this is
3209 called BEFORE disk is attached */
3210 dev_err(DEV
, "Error while reading metadata.\n");
3211 rv
= ERR_IO_MD_DISK
;
3215 magic
= be32_to_cpu(buffer
->magic
);
3216 flags
= be32_to_cpu(buffer
->flags
);
3217 if (magic
== DRBD_MD_MAGIC_84_UNCLEAN
||
3218 (magic
== DRBD_MD_MAGIC_08
&& !(flags
& MDF_AL_CLEAN
))) {
3219 /* btw: that's Activity Log clean, not "all" clean. */
3220 dev_err(DEV
, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3221 rv
= ERR_MD_UNCLEAN
;
3225 rv
= ERR_MD_INVALID
;
3226 if (magic
!= DRBD_MD_MAGIC_08
) {
3227 if (magic
== DRBD_MD_MAGIC_07
)
3228 dev_err(DEV
, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3230 dev_err(DEV
, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3234 if (be32_to_cpu(buffer
->bm_bytes_per_bit
) != BM_BLOCK_SIZE
) {
3235 dev_err(DEV
, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3236 be32_to_cpu(buffer
->bm_bytes_per_bit
), BM_BLOCK_SIZE
);
3241 /* convert to in_core endian */
3242 bdev
->md
.la_size_sect
= be64_to_cpu(buffer
->la_size_sect
);
3243 for (i
= UI_CURRENT
; i
< UI_SIZE
; i
++)
3244 bdev
->md
.uuid
[i
] = be64_to_cpu(buffer
->uuid
[i
]);
3245 bdev
->md
.flags
= be32_to_cpu(buffer
->flags
);
3246 bdev
->md
.device_uuid
= be64_to_cpu(buffer
->device_uuid
);
3248 bdev
->md
.md_size_sect
= be32_to_cpu(buffer
->md_size_sect
);
3249 bdev
->md
.al_offset
= be32_to_cpu(buffer
->al_offset
);
3250 bdev
->md
.bm_offset
= be32_to_cpu(buffer
->bm_offset
);
3252 if (check_activity_log_stripe_size(device
, buffer
, &bdev
->md
))
3254 if (check_offsets_and_sizes(device
, bdev
))
3257 if (be32_to_cpu(buffer
->bm_offset
) != bdev
->md
.bm_offset
) {
3258 dev_err(DEV
, "unexpected bm_offset: %d (expected %d)\n",
3259 be32_to_cpu(buffer
->bm_offset
), bdev
->md
.bm_offset
);
3262 if (be32_to_cpu(buffer
->md_size_sect
) != bdev
->md
.md_size_sect
) {
3263 dev_err(DEV
, "unexpected md_size: %u (expected %u)\n",
3264 be32_to_cpu(buffer
->md_size_sect
), bdev
->md
.md_size_sect
);
3270 spin_lock_irq(&first_peer_device(device
)->connection
->req_lock
);
3271 if (device
->state
.conn
< C_CONNECTED
) {
3273 peer
= be32_to_cpu(buffer
->la_peer_max_bio_size
);
3274 peer
= max(peer
, DRBD_MAX_BIO_SIZE_SAFE
);
3275 device
->peer_max_bio_size
= peer
;
3277 spin_unlock_irq(&first_peer_device(device
)->connection
->req_lock
);
3280 drbd_md_put_buffer(device
);
3286 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3287 * @device: DRBD device.
3289 * Call this function if you change anything that should be written to
3290 * the meta-data super block. This function sets MD_DIRTY, and starts a
3291 * timer that ensures that within five seconds you have to call drbd_md_sync().
3294 void drbd_md_mark_dirty_(struct drbd_device
*device
, unsigned int line
, const char *func
)
3296 if (!test_and_set_bit(MD_DIRTY
, &device
->flags
)) {
3297 mod_timer(&device
->md_sync_timer
, jiffies
+ HZ
);
3298 device
->last_md_mark_dirty
.line
= line
;
3299 device
->last_md_mark_dirty
.func
= func
;
3303 void drbd_md_mark_dirty(struct drbd_device
*device
)
3305 if (!test_and_set_bit(MD_DIRTY
, &device
->flags
))
3306 mod_timer(&device
->md_sync_timer
, jiffies
+ 5*HZ
);
3310 void drbd_uuid_move_history(struct drbd_device
*device
) __must_hold(local
)
3314 for (i
= UI_HISTORY_START
; i
< UI_HISTORY_END
; i
++)
3315 device
->ldev
->md
.uuid
[i
+1] = device
->ldev
->md
.uuid
[i
];
3318 void __drbd_uuid_set(struct drbd_device
*device
, int idx
, u64 val
) __must_hold(local
)
3320 if (idx
== UI_CURRENT
) {
3321 if (device
->state
.role
== R_PRIMARY
)
3326 drbd_set_ed_uuid(device
, val
);
3329 device
->ldev
->md
.uuid
[idx
] = val
;
3330 drbd_md_mark_dirty(device
);
3333 void _drbd_uuid_set(struct drbd_device
*device
, int idx
, u64 val
) __must_hold(local
)
3335 unsigned long flags
;
3336 spin_lock_irqsave(&device
->ldev
->md
.uuid_lock
, flags
);
3337 __drbd_uuid_set(device
, idx
, val
);
3338 spin_unlock_irqrestore(&device
->ldev
->md
.uuid_lock
, flags
);
3341 void drbd_uuid_set(struct drbd_device
*device
, int idx
, u64 val
) __must_hold(local
)
3343 unsigned long flags
;
3344 spin_lock_irqsave(&device
->ldev
->md
.uuid_lock
, flags
);
3345 if (device
->ldev
->md
.uuid
[idx
]) {
3346 drbd_uuid_move_history(device
);
3347 device
->ldev
->md
.uuid
[UI_HISTORY_START
] = device
->ldev
->md
.uuid
[idx
];
3349 __drbd_uuid_set(device
, idx
, val
);
3350 spin_unlock_irqrestore(&device
->ldev
->md
.uuid_lock
, flags
);
3354 * drbd_uuid_new_current() - Creates a new current UUID
3355 * @device: DRBD device.
3357 * Creates a new current UUID, and rotates the old current UUID into
3358 * the bitmap slot. Causes an incremental resync upon next connect.
3360 void drbd_uuid_new_current(struct drbd_device
*device
) __must_hold(local
)
3363 unsigned long long bm_uuid
;
3365 get_random_bytes(&val
, sizeof(u64
));
3367 spin_lock_irq(&device
->ldev
->md
.uuid_lock
);
3368 bm_uuid
= device
->ldev
->md
.uuid
[UI_BITMAP
];
3371 dev_warn(DEV
, "bm UUID was already set: %llX\n", bm_uuid
);
3373 device
->ldev
->md
.uuid
[UI_BITMAP
] = device
->ldev
->md
.uuid
[UI_CURRENT
];
3374 __drbd_uuid_set(device
, UI_CURRENT
, val
);
3375 spin_unlock_irq(&device
->ldev
->md
.uuid_lock
);
3377 drbd_print_uuids(device
, "new current UUID");
3378 /* get it to stable storage _now_ */
3379 drbd_md_sync(device
);
3382 void drbd_uuid_set_bm(struct drbd_device
*device
, u64 val
) __must_hold(local
)
3384 unsigned long flags
;
3385 if (device
->ldev
->md
.uuid
[UI_BITMAP
] == 0 && val
== 0)
3388 spin_lock_irqsave(&device
->ldev
->md
.uuid_lock
, flags
);
3390 drbd_uuid_move_history(device
);
3391 device
->ldev
->md
.uuid
[UI_HISTORY_START
] = device
->ldev
->md
.uuid
[UI_BITMAP
];
3392 device
->ldev
->md
.uuid
[UI_BITMAP
] = 0;
3394 unsigned long long bm_uuid
= device
->ldev
->md
.uuid
[UI_BITMAP
];
3396 dev_warn(DEV
, "bm UUID was already set: %llX\n", bm_uuid
);
3398 device
->ldev
->md
.uuid
[UI_BITMAP
] = val
& ~((u64
)1);
3400 spin_unlock_irqrestore(&device
->ldev
->md
.uuid_lock
, flags
);
3402 drbd_md_mark_dirty(device
);
3406 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3407 * @device: DRBD device.
3409 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3411 int drbd_bmio_set_n_write(struct drbd_device
*device
)
3415 if (get_ldev_if_state(device
, D_ATTACHING
)) {
3416 drbd_md_set_flag(device
, MDF_FULL_SYNC
);
3417 drbd_md_sync(device
);
3418 drbd_bm_set_all(device
);
3420 rv
= drbd_bm_write(device
);
3423 drbd_md_clear_flag(device
, MDF_FULL_SYNC
);
3424 drbd_md_sync(device
);
3434 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3435 * @device: DRBD device.
3437 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3439 int drbd_bmio_clear_n_write(struct drbd_device
*device
)
3443 drbd_resume_al(device
);
3444 if (get_ldev_if_state(device
, D_ATTACHING
)) {
3445 drbd_bm_clear_all(device
);
3446 rv
= drbd_bm_write(device
);
3453 static int w_bitmap_io(struct drbd_work
*w
, int unused
)
3455 struct bm_io_work
*work
= container_of(w
, struct bm_io_work
, w
);
3456 struct drbd_device
*device
= w
->device
;
3459 D_ASSERT(atomic_read(&device
->ap_bio_cnt
) == 0);
3461 if (get_ldev(device
)) {
3462 drbd_bm_lock(device
, work
->why
, work
->flags
);
3463 rv
= work
->io_fn(device
);
3464 drbd_bm_unlock(device
);
3468 clear_bit_unlock(BITMAP_IO
, &device
->flags
);
3469 wake_up(&device
->misc_wait
);
3472 work
->done(device
, rv
);
3474 clear_bit(BITMAP_IO_QUEUED
, &device
->flags
);
3481 void drbd_ldev_destroy(struct drbd_device
*device
)
3483 lc_destroy(device
->resync
);
3484 device
->resync
= NULL
;
3485 lc_destroy(device
->act_log
);
3486 device
->act_log
= NULL
;
3488 drbd_free_bc(device
->ldev
);
3489 device
->ldev
= NULL
;);
3491 clear_bit(GO_DISKLESS
, &device
->flags
);
3494 static int w_go_diskless(struct drbd_work
*w
, int unused
)
3496 struct drbd_device
*device
= w
->device
;
3498 D_ASSERT(device
->state
.disk
== D_FAILED
);
3499 /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will
3500 * inc/dec it frequently. Once we are D_DISKLESS, no one will touch
3501 * the protected members anymore, though, so once put_ldev reaches zero
3502 * again, it will be safe to free them. */
3504 /* Try to write changed bitmap pages, read errors may have just
3505 * set some bits outside the area covered by the activity log.
3507 * If we have an IO error during the bitmap writeout,
3508 * we will want a full sync next time, just in case.
3509 * (Do we want a specific meta data flag for this?)
3511 * If that does not make it to stable storage either,
3512 * we cannot do anything about that anymore.
3514 * We still need to check if both bitmap and ldev are present, we may
3515 * end up here after a failed attach, before ldev was even assigned.
3517 if (device
->bitmap
&& device
->ldev
) {
3518 /* An interrupted resync or similar is allowed to recounts bits
3520 * Any modifications would not be expected anymore, though.
3522 if (drbd_bitmap_io_from_worker(device
, drbd_bm_write
,
3523 "detach", BM_LOCKED_TEST_ALLOWED
)) {
3524 if (test_bit(WAS_READ_ERROR
, &device
->flags
)) {
3525 drbd_md_set_flag(device
, MDF_FULL_SYNC
);
3526 drbd_md_sync(device
);
3531 drbd_force_state(device
, NS(disk
, D_DISKLESS
));
3536 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3537 * @device: DRBD device.
3538 * @io_fn: IO callback to be called when bitmap IO is possible
3539 * @done: callback to be called after the bitmap IO was performed
3540 * @why: Descriptive text of the reason for doing the IO
3542 * While IO on the bitmap happens we freeze application IO thus we ensure
3543 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3544 * called from worker context. It MUST NOT be used while a previous such
3545 * work is still pending!
3547 void drbd_queue_bitmap_io(struct drbd_device
*device
,
3548 int (*io_fn
)(struct drbd_device
*),
3549 void (*done
)(struct drbd_device
*, int),
3550 char *why
, enum bm_flag flags
)
3552 D_ASSERT(current
== first_peer_device(device
)->connection
->worker
.task
);
3554 D_ASSERT(!test_bit(BITMAP_IO_QUEUED
, &device
->flags
));
3555 D_ASSERT(!test_bit(BITMAP_IO
, &device
->flags
));
3556 D_ASSERT(list_empty(&device
->bm_io_work
.w
.list
));
3557 if (device
->bm_io_work
.why
)
3558 dev_err(DEV
, "FIXME going to queue '%s' but '%s' still pending?\n",
3559 why
, device
->bm_io_work
.why
);
3561 device
->bm_io_work
.io_fn
= io_fn
;
3562 device
->bm_io_work
.done
= done
;
3563 device
->bm_io_work
.why
= why
;
3564 device
->bm_io_work
.flags
= flags
;
3566 spin_lock_irq(&first_peer_device(device
)->connection
->req_lock
);
3567 set_bit(BITMAP_IO
, &device
->flags
);
3568 if (atomic_read(&device
->ap_bio_cnt
) == 0) {
3569 if (!test_and_set_bit(BITMAP_IO_QUEUED
, &device
->flags
))
3570 drbd_queue_work(&first_peer_device(device
)->connection
->sender_work
, &device
->bm_io_work
.w
);
3572 spin_unlock_irq(&first_peer_device(device
)->connection
->req_lock
);
3576 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3577 * @device: DRBD device.
3578 * @io_fn: IO callback to be called when bitmap IO is possible
3579 * @why: Descriptive text of the reason for doing the IO
3581 * freezes application IO while that the actual IO operations runs. This
3582 * functions MAY NOT be called from worker context.
3584 int drbd_bitmap_io(struct drbd_device
*device
, int (*io_fn
)(struct drbd_device
*),
3585 char *why
, enum bm_flag flags
)
3589 D_ASSERT(current
!= first_peer_device(device
)->connection
->worker
.task
);
3591 if ((flags
& BM_LOCKED_SET_ALLOWED
) == 0)
3592 drbd_suspend_io(device
);
3594 drbd_bm_lock(device
, why
, flags
);
3596 drbd_bm_unlock(device
);
3598 if ((flags
& BM_LOCKED_SET_ALLOWED
) == 0)
3599 drbd_resume_io(device
);
3604 void drbd_md_set_flag(struct drbd_device
*device
, int flag
) __must_hold(local
)
3606 if ((device
->ldev
->md
.flags
& flag
) != flag
) {
3607 drbd_md_mark_dirty(device
);
3608 device
->ldev
->md
.flags
|= flag
;
3612 void drbd_md_clear_flag(struct drbd_device
*device
, int flag
) __must_hold(local
)
3614 if ((device
->ldev
->md
.flags
& flag
) != 0) {
3615 drbd_md_mark_dirty(device
);
3616 device
->ldev
->md
.flags
&= ~flag
;
3619 int drbd_md_test_flag(struct drbd_backing_dev
*bdev
, int flag
)
3621 return (bdev
->md
.flags
& flag
) != 0;
3624 static void md_sync_timer_fn(unsigned long data
)
3626 struct drbd_device
*device
= (struct drbd_device
*) data
;
3628 /* must not double-queue! */
3629 if (list_empty(&device
->md_sync_work
.list
))
3630 drbd_queue_work_front(&first_peer_device(device
)->connection
->sender_work
, &device
->md_sync_work
);
3633 static int w_md_sync(struct drbd_work
*w
, int unused
)
3635 struct drbd_device
*device
= w
->device
;
3637 dev_warn(DEV
, "md_sync_timer expired! Worker calls drbd_md_sync().\n");
3639 dev_warn(DEV
, "last md_mark_dirty: %s:%u\n",
3640 device
->last_md_mark_dirty
.func
, device
->last_md_mark_dirty
.line
);
3642 drbd_md_sync(device
);
3646 const char *cmdname(enum drbd_packet cmd
)
3648 /* THINK may need to become several global tables
3649 * when we want to support more than
3650 * one PRO_VERSION */
3651 static const char *cmdnames
[] = {
3653 [P_DATA_REPLY
] = "DataReply",
3654 [P_RS_DATA_REPLY
] = "RSDataReply",
3655 [P_BARRIER
] = "Barrier",
3656 [P_BITMAP
] = "ReportBitMap",
3657 [P_BECOME_SYNC_TARGET
] = "BecomeSyncTarget",
3658 [P_BECOME_SYNC_SOURCE
] = "BecomeSyncSource",
3659 [P_UNPLUG_REMOTE
] = "UnplugRemote",
3660 [P_DATA_REQUEST
] = "DataRequest",
3661 [P_RS_DATA_REQUEST
] = "RSDataRequest",
3662 [P_SYNC_PARAM
] = "SyncParam",
3663 [P_SYNC_PARAM89
] = "SyncParam89",
3664 [P_PROTOCOL
] = "ReportProtocol",
3665 [P_UUIDS
] = "ReportUUIDs",
3666 [P_SIZES
] = "ReportSizes",
3667 [P_STATE
] = "ReportState",
3668 [P_SYNC_UUID
] = "ReportSyncUUID",
3669 [P_AUTH_CHALLENGE
] = "AuthChallenge",
3670 [P_AUTH_RESPONSE
] = "AuthResponse",
3672 [P_PING_ACK
] = "PingAck",
3673 [P_RECV_ACK
] = "RecvAck",
3674 [P_WRITE_ACK
] = "WriteAck",
3675 [P_RS_WRITE_ACK
] = "RSWriteAck",
3676 [P_SUPERSEDED
] = "Superseded",
3677 [P_NEG_ACK
] = "NegAck",
3678 [P_NEG_DREPLY
] = "NegDReply",
3679 [P_NEG_RS_DREPLY
] = "NegRSDReply",
3680 [P_BARRIER_ACK
] = "BarrierAck",
3681 [P_STATE_CHG_REQ
] = "StateChgRequest",
3682 [P_STATE_CHG_REPLY
] = "StateChgReply",
3683 [P_OV_REQUEST
] = "OVRequest",
3684 [P_OV_REPLY
] = "OVReply",
3685 [P_OV_RESULT
] = "OVResult",
3686 [P_CSUM_RS_REQUEST
] = "CsumRSRequest",
3687 [P_RS_IS_IN_SYNC
] = "CsumRSIsInSync",
3688 [P_COMPRESSED_BITMAP
] = "CBitmap",
3689 [P_DELAY_PROBE
] = "DelayProbe",
3690 [P_OUT_OF_SYNC
] = "OutOfSync",
3691 [P_RETRY_WRITE
] = "RetryWrite",
3692 [P_RS_CANCEL
] = "RSCancel",
3693 [P_CONN_ST_CHG_REQ
] = "conn_st_chg_req",
3694 [P_CONN_ST_CHG_REPLY
] = "conn_st_chg_reply",
3695 [P_RETRY_WRITE
] = "retry_write",
3696 [P_PROTOCOL_UPDATE
] = "protocol_update",
3698 /* enum drbd_packet, but not commands - obsoleted flags:
3704 /* too big for the array: 0xfffX */
3705 if (cmd
== P_INITIAL_META
)
3706 return "InitialMeta";
3707 if (cmd
== P_INITIAL_DATA
)
3708 return "InitialData";
3709 if (cmd
== P_CONNECTION_FEATURES
)
3710 return "ConnectionFeatures";
3711 if (cmd
>= ARRAY_SIZE(cmdnames
))
3713 return cmdnames
[cmd
];
3717 * drbd_wait_misc - wait for a request to make progress
3718 * @device: device associated with the request
3719 * @i: the struct drbd_interval embedded in struct drbd_request or
3720 * struct drbd_peer_request
3722 int drbd_wait_misc(struct drbd_device
*device
, struct drbd_interval
*i
)
3724 struct net_conf
*nc
;
3729 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
3734 timeout
= nc
->ko_count
? nc
->timeout
* HZ
/ 10 * nc
->ko_count
: MAX_SCHEDULE_TIMEOUT
;
3737 /* Indicate to wake up device->misc_wait on progress. */
3739 prepare_to_wait(&device
->misc_wait
, &wait
, TASK_INTERRUPTIBLE
);
3740 spin_unlock_irq(&first_peer_device(device
)->connection
->req_lock
);
3741 timeout
= schedule_timeout(timeout
);
3742 finish_wait(&device
->misc_wait
, &wait
);
3743 spin_lock_irq(&first_peer_device(device
)->connection
->req_lock
);
3744 if (!timeout
|| device
->state
.conn
< C_CONNECTED
)
3746 if (signal_pending(current
))
3747 return -ERESTARTSYS
;
3751 #ifdef CONFIG_DRBD_FAULT_INJECTION
3752 /* Fault insertion support including random number generator shamelessly
3753 * stolen from kernel/rcutorture.c */
3754 struct fault_random_state
{
3755 unsigned long state
;
3756 unsigned long count
;
3759 #define FAULT_RANDOM_MULT 39916801 /* prime */
3760 #define FAULT_RANDOM_ADD 479001701 /* prime */
3761 #define FAULT_RANDOM_REFRESH 10000
3764 * Crude but fast random-number generator. Uses a linear congruential
3765 * generator, with occasional help from get_random_bytes().
3767 static unsigned long
3768 _drbd_fault_random(struct fault_random_state
*rsp
)
3772 if (!rsp
->count
--) {
3773 get_random_bytes(&refresh
, sizeof(refresh
));
3774 rsp
->state
+= refresh
;
3775 rsp
->count
= FAULT_RANDOM_REFRESH
;
3777 rsp
->state
= rsp
->state
* FAULT_RANDOM_MULT
+ FAULT_RANDOM_ADD
;
3778 return swahw32(rsp
->state
);
3782 _drbd_fault_str(unsigned int type
) {
3783 static char *_faults
[] = {
3784 [DRBD_FAULT_MD_WR
] = "Meta-data write",
3785 [DRBD_FAULT_MD_RD
] = "Meta-data read",
3786 [DRBD_FAULT_RS_WR
] = "Resync write",
3787 [DRBD_FAULT_RS_RD
] = "Resync read",
3788 [DRBD_FAULT_DT_WR
] = "Data write",
3789 [DRBD_FAULT_DT_RD
] = "Data read",
3790 [DRBD_FAULT_DT_RA
] = "Data read ahead",
3791 [DRBD_FAULT_BM_ALLOC
] = "BM allocation",
3792 [DRBD_FAULT_AL_EE
] = "EE allocation",
3793 [DRBD_FAULT_RECEIVE
] = "receive data corruption",
3796 return (type
< DRBD_FAULT_MAX
) ? _faults
[type
] : "**Unknown**";
3800 _drbd_insert_fault(struct drbd_device
*device
, unsigned int type
)
3802 static struct fault_random_state rrs
= {0, 0};
3804 unsigned int ret
= (
3806 ((1 << device_to_minor(device
)) & fault_devs
) != 0) &&
3807 (((_drbd_fault_random(&rrs
) % 100) + 1) <= fault_rate
));
3812 if (__ratelimit(&drbd_ratelimit_state
))
3813 dev_warn(DEV
, "***Simulating %s failure\n",
3814 _drbd_fault_str(type
));
3821 const char *drbd_buildtag(void)
3823 /* DRBD built from external sources has here a reference to the
3824 git hash of the source code. */
3826 static char buildtag
[38] = "\0uilt-in";
3828 if (buildtag
[0] == 0) {
3830 sprintf(buildtag
, "srcversion: %-24s", THIS_MODULE
->srcversion
);
3839 module_init(drbd_init
)
3840 module_exit(drbd_cleanup
)
3842 EXPORT_SYMBOL(drbd_conn_str
);
3843 EXPORT_SYMBOL(drbd_role_str
);
3844 EXPORT_SYMBOL(drbd_disk_str
);
3845 EXPORT_SYMBOL(drbd_set_st_err_str
);