1 // SPDX-License-Identifier: GPL-2.0-or-later
5 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
7 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
8 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
9 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
11 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
12 from Logicworks, Inc. for making SDP replication support possible.
17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 #include <linux/module.h>
20 #include <linux/jiffies.h>
21 #include <linux/drbd.h>
22 #include <linux/uaccess.h>
23 #include <asm/types.h>
25 #include <linux/ctype.h>
26 #include <linux/mutex.h>
28 #include <linux/file.h>
29 #include <linux/proc_fs.h>
30 #include <linux/init.h>
32 #include <linux/memcontrol.h>
33 #include <linux/mm_inline.h>
34 #include <linux/slab.h>
35 #include <linux/random.h>
36 #include <linux/reboot.h>
37 #include <linux/notifier.h>
38 #include <linux/kthread.h>
39 #include <linux/workqueue.h>
40 #define __KERNEL_SYSCALLS__
41 #include <linux/unistd.h>
42 #include <linux/vmalloc.h>
43 #include <linux/sched/signal.h>
45 #include <linux/drbd_limits.h>
47 #include "drbd_protocol.h"
48 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
50 #include "drbd_debugfs.h"
52 static DEFINE_MUTEX(drbd_main_mutex
);
53 static int drbd_open(struct block_device
*bdev
, fmode_t mode
);
54 static void drbd_release(struct gendisk
*gd
, fmode_t mode
);
55 static void md_sync_timer_fn(struct timer_list
*t
);
56 static int w_bitmap_io(struct drbd_work
*w
, int unused
);
58 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
59 "Lars Ellenberg <lars@linbit.com>");
60 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION
);
61 MODULE_VERSION(REL_VERSION
);
62 MODULE_LICENSE("GPL");
63 MODULE_PARM_DESC(minor_count
, "Approximate number of drbd devices ("
64 __stringify(DRBD_MINOR_COUNT_MIN
) "-" __stringify(DRBD_MINOR_COUNT_MAX
) ")");
65 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR
);
67 #include <linux/moduleparam.h>
68 /* thanks to these macros, if compiled into the kernel (not-module),
69 * these become boot parameters (e.g., drbd.minor_count) */
71 #ifdef CONFIG_DRBD_FAULT_INJECTION
72 int drbd_enable_faults
;
74 static int drbd_fault_count
;
75 static int drbd_fault_devs
;
76 /* bitmap of enabled faults */
77 module_param_named(enable_faults
, drbd_enable_faults
, int, 0664);
78 /* fault rate % value - applies to all enabled faults */
79 module_param_named(fault_rate
, drbd_fault_rate
, int, 0664);
80 /* count of faults inserted */
81 module_param_named(fault_count
, drbd_fault_count
, int, 0664);
82 /* bitmap of devices to insert faults on */
83 module_param_named(fault_devs
, drbd_fault_devs
, int, 0644);
86 /* module parameters we can keep static */
87 static bool drbd_allow_oos
; /* allow_open_on_secondary */
88 static bool drbd_disable_sendpage
;
89 MODULE_PARM_DESC(allow_oos
, "DONT USE!");
90 module_param_named(allow_oos
, drbd_allow_oos
, bool, 0);
91 module_param_named(disable_sendpage
, drbd_disable_sendpage
, bool, 0644);
93 /* module parameters we share */
94 int drbd_proc_details
; /* Detail level in proc drbd*/
95 module_param_named(proc_details
, drbd_proc_details
, int, 0644);
96 /* module parameters shared with defaults */
97 unsigned int drbd_minor_count
= DRBD_MINOR_COUNT_DEF
;
98 /* Module parameter for setting the user mode helper program
99 * to run. Default is /sbin/drbdadm */
100 char drbd_usermode_helper
[80] = "/sbin/drbdadm";
101 module_param_named(minor_count
, drbd_minor_count
, uint
, 0444);
102 module_param_string(usermode_helper
, drbd_usermode_helper
, sizeof(drbd_usermode_helper
), 0644);
104 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
105 * as member "struct gendisk *vdisk;"
107 struct idr drbd_devices
;
108 struct list_head drbd_resources
;
109 struct mutex resources_mutex
;
111 struct kmem_cache
*drbd_request_cache
;
112 struct kmem_cache
*drbd_ee_cache
; /* peer requests */
113 struct kmem_cache
*drbd_bm_ext_cache
; /* bitmap extents */
114 struct kmem_cache
*drbd_al_ext_cache
; /* activity log extents */
115 mempool_t drbd_request_mempool
;
116 mempool_t drbd_ee_mempool
;
117 mempool_t drbd_md_io_page_pool
;
118 struct bio_set drbd_md_io_bio_set
;
119 struct bio_set drbd_io_bio_set
;
121 /* I do not use a standard mempool, because:
122 1) I want to hand out the pre-allocated objects first.
123 2) I want to be able to interrupt sleeping allocation with a signal.
124 Note: This is a single linked list, the next pointer is the private
125 member of struct page.
127 struct page
*drbd_pp_pool
;
128 spinlock_t drbd_pp_lock
;
130 wait_queue_head_t drbd_pp_wait
;
132 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state
, 5 * HZ
, 5);
134 static const struct block_device_operations drbd_ops
= {
135 .owner
= THIS_MODULE
,
137 .release
= drbd_release
,
140 struct bio
*bio_alloc_drbd(gfp_t gfp_mask
)
144 if (!bioset_initialized(&drbd_md_io_bio_set
))
145 return bio_alloc(gfp_mask
, 1);
147 bio
= bio_alloc_bioset(gfp_mask
, 1, &drbd_md_io_bio_set
);
154 /* When checking with sparse, and this is an inline function, sparse will
155 give tons of false positives. When this is a real functions sparse works.
157 int _get_ldev_if_state(struct drbd_device
*device
, enum drbd_disk_state mins
)
161 atomic_inc(&device
->local_cnt
);
162 io_allowed
= (device
->state
.disk
>= mins
);
164 if (atomic_dec_and_test(&device
->local_cnt
))
165 wake_up(&device
->misc_wait
);
173 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
174 * @connection: DRBD connection.
175 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
176 * @set_size: Expected number of requests before that barrier.
178 * In case the passed barrier_nr or set_size does not match the oldest
179 * epoch of not yet barrier-acked requests, this function will cause a
180 * termination of the connection.
182 void tl_release(struct drbd_connection
*connection
, unsigned int barrier_nr
,
183 unsigned int set_size
)
185 struct drbd_request
*r
;
186 struct drbd_request
*req
= NULL
;
187 int expect_epoch
= 0;
190 spin_lock_irq(&connection
->resource
->req_lock
);
192 /* find oldest not yet barrier-acked write request,
193 * count writes in its epoch. */
194 list_for_each_entry(r
, &connection
->transfer_log
, tl_requests
) {
195 const unsigned s
= r
->rq_state
;
199 if (!(s
& RQ_NET_MASK
))
204 expect_epoch
= req
->epoch
;
207 if (r
->epoch
!= expect_epoch
)
211 /* if (s & RQ_DONE): not expected */
212 /* if (!(s & RQ_NET_MASK)): not expected */
217 /* first some paranoia code */
219 drbd_err(connection
, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
223 if (expect_epoch
!= barrier_nr
) {
224 drbd_err(connection
, "BAD! BarrierAck #%u received, expected #%u!\n",
225 barrier_nr
, expect_epoch
);
229 if (expect_size
!= set_size
) {
230 drbd_err(connection
, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
231 barrier_nr
, set_size
, expect_size
);
235 /* Clean up list of requests processed during current epoch. */
236 /* this extra list walk restart is paranoia,
237 * to catch requests being barrier-acked "unexpectedly".
238 * It usually should find the same req again, or some READ preceding it. */
239 list_for_each_entry(req
, &connection
->transfer_log
, tl_requests
)
240 if (req
->epoch
== expect_epoch
)
242 list_for_each_entry_safe_from(req
, r
, &connection
->transfer_log
, tl_requests
) {
243 if (req
->epoch
!= expect_epoch
)
245 _req_mod(req
, BARRIER_ACKED
);
247 spin_unlock_irq(&connection
->resource
->req_lock
);
252 spin_unlock_irq(&connection
->resource
->req_lock
);
253 conn_request_state(connection
, NS(conn
, C_PROTOCOL_ERROR
), CS_HARD
);
258 * _tl_restart() - Walks the transfer log, and applies an action to all requests
259 * @connection: DRBD connection to operate on.
260 * @what: The action/event to perform with all request objects
262 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
263 * RESTART_FROZEN_DISK_IO.
265 /* must hold resource->req_lock */
266 void _tl_restart(struct drbd_connection
*connection
, enum drbd_req_event what
)
268 struct drbd_request
*req
, *r
;
270 list_for_each_entry_safe(req
, r
, &connection
->transfer_log
, tl_requests
)
274 void tl_restart(struct drbd_connection
*connection
, enum drbd_req_event what
)
276 spin_lock_irq(&connection
->resource
->req_lock
);
277 _tl_restart(connection
, what
);
278 spin_unlock_irq(&connection
->resource
->req_lock
);
282 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
283 * @device: DRBD device.
285 * This is called after the connection to the peer was lost. The storage covered
286 * by the requests on the transfer gets marked as our of sync. Called from the
287 * receiver thread and the worker thread.
289 void tl_clear(struct drbd_connection
*connection
)
291 tl_restart(connection
, CONNECTION_LOST_WHILE_PENDING
);
295 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
296 * @device: DRBD device.
298 void tl_abort_disk_io(struct drbd_device
*device
)
300 struct drbd_connection
*connection
= first_peer_device(device
)->connection
;
301 struct drbd_request
*req
, *r
;
303 spin_lock_irq(&connection
->resource
->req_lock
);
304 list_for_each_entry_safe(req
, r
, &connection
->transfer_log
, tl_requests
) {
305 if (!(req
->rq_state
& RQ_LOCAL_PENDING
))
307 if (req
->device
!= device
)
309 _req_mod(req
, ABORT_DISK_IO
);
311 spin_unlock_irq(&connection
->resource
->req_lock
);
314 static int drbd_thread_setup(void *arg
)
316 struct drbd_thread
*thi
= (struct drbd_thread
*) arg
;
317 struct drbd_resource
*resource
= thi
->resource
;
321 snprintf(current
->comm
, sizeof(current
->comm
), "drbd_%c_%s",
326 retval
= thi
->function(thi
);
328 spin_lock_irqsave(&thi
->t_lock
, flags
);
330 /* if the receiver has been "EXITING", the last thing it did
331 * was set the conn state to "StandAlone",
332 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
333 * and receiver thread will be "started".
334 * drbd_thread_start needs to set "RESTARTING" in that case.
335 * t_state check and assignment needs to be within the same spinlock,
336 * so either thread_start sees EXITING, and can remap to RESTARTING,
337 * or thread_start see NONE, and can proceed as normal.
340 if (thi
->t_state
== RESTARTING
) {
341 drbd_info(resource
, "Restarting %s thread\n", thi
->name
);
342 thi
->t_state
= RUNNING
;
343 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
350 complete_all(&thi
->stop
);
351 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
353 drbd_info(resource
, "Terminating %s\n", current
->comm
);
355 /* Release mod reference taken when thread was started */
358 kref_put(&thi
->connection
->kref
, drbd_destroy_connection
);
359 kref_put(&resource
->kref
, drbd_destroy_resource
);
360 module_put(THIS_MODULE
);
364 static void drbd_thread_init(struct drbd_resource
*resource
, struct drbd_thread
*thi
,
365 int (*func
) (struct drbd_thread
*), const char *name
)
367 spin_lock_init(&thi
->t_lock
);
370 thi
->function
= func
;
371 thi
->resource
= resource
;
372 thi
->connection
= NULL
;
376 int drbd_thread_start(struct drbd_thread
*thi
)
378 struct drbd_resource
*resource
= thi
->resource
;
379 struct task_struct
*nt
;
382 /* is used from state engine doing drbd_thread_stop_nowait,
383 * while holding the req lock irqsave */
384 spin_lock_irqsave(&thi
->t_lock
, flags
);
386 switch (thi
->t_state
) {
388 drbd_info(resource
, "Starting %s thread (from %s [%d])\n",
389 thi
->name
, current
->comm
, current
->pid
);
391 /* Get ref on module for thread - this is released when thread exits */
392 if (!try_module_get(THIS_MODULE
)) {
393 drbd_err(resource
, "Failed to get module reference in drbd_thread_start\n");
394 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
398 kref_get(&resource
->kref
);
400 kref_get(&thi
->connection
->kref
);
402 init_completion(&thi
->stop
);
403 thi
->reset_cpu_mask
= 1;
404 thi
->t_state
= RUNNING
;
405 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
406 flush_signals(current
); /* otherw. may get -ERESTARTNOINTR */
408 nt
= kthread_create(drbd_thread_setup
, (void *) thi
,
409 "drbd_%c_%s", thi
->name
[0], thi
->resource
->name
);
412 drbd_err(resource
, "Couldn't start thread\n");
415 kref_put(&thi
->connection
->kref
, drbd_destroy_connection
);
416 kref_put(&resource
->kref
, drbd_destroy_resource
);
417 module_put(THIS_MODULE
);
420 spin_lock_irqsave(&thi
->t_lock
, flags
);
422 thi
->t_state
= RUNNING
;
423 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
427 thi
->t_state
= RESTARTING
;
428 drbd_info(resource
, "Restarting %s thread (from %s [%d])\n",
429 thi
->name
, current
->comm
, current
->pid
);
434 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
442 void _drbd_thread_stop(struct drbd_thread
*thi
, int restart
, int wait
)
446 enum drbd_thread_state ns
= restart
? RESTARTING
: EXITING
;
448 /* may be called from state engine, holding the req lock irqsave */
449 spin_lock_irqsave(&thi
->t_lock
, flags
);
451 if (thi
->t_state
== NONE
) {
452 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
454 drbd_thread_start(thi
);
458 if (thi
->t_state
!= ns
) {
459 if (thi
->task
== NULL
) {
460 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
466 init_completion(&thi
->stop
);
467 if (thi
->task
!= current
)
468 send_sig(DRBD_SIGKILL
, thi
->task
, 1);
471 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
474 wait_for_completion(&thi
->stop
);
477 int conn_lowest_minor(struct drbd_connection
*connection
)
479 struct drbd_peer_device
*peer_device
;
480 int vnr
= 0, minor
= -1;
483 peer_device
= idr_get_next(&connection
->peer_devices
, &vnr
);
485 minor
= device_to_minor(peer_device
->device
);
493 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
495 * Forces all threads of a resource onto the same CPU. This is beneficial for
496 * DRBD's performance. May be overwritten by user's configuration.
498 static void drbd_calc_cpu_mask(cpumask_var_t
*cpu_mask
)
500 unsigned int *resources_per_cpu
, min_index
= ~0;
502 resources_per_cpu
= kcalloc(nr_cpu_ids
, sizeof(*resources_per_cpu
),
504 if (resources_per_cpu
) {
505 struct drbd_resource
*resource
;
506 unsigned int cpu
, min
= ~0;
509 for_each_resource_rcu(resource
, &drbd_resources
) {
510 for_each_cpu(cpu
, resource
->cpu_mask
)
511 resources_per_cpu
[cpu
]++;
514 for_each_online_cpu(cpu
) {
515 if (resources_per_cpu
[cpu
] < min
) {
516 min
= resources_per_cpu
[cpu
];
520 kfree(resources_per_cpu
);
522 if (min_index
== ~0) {
523 cpumask_setall(*cpu_mask
);
526 cpumask_set_cpu(min_index
, *cpu_mask
);
530 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
531 * @device: DRBD device.
532 * @thi: drbd_thread object
534 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
537 void drbd_thread_current_set_cpu(struct drbd_thread
*thi
)
539 struct drbd_resource
*resource
= thi
->resource
;
540 struct task_struct
*p
= current
;
542 if (!thi
->reset_cpu_mask
)
544 thi
->reset_cpu_mask
= 0;
545 set_cpus_allowed_ptr(p
, resource
->cpu_mask
);
548 #define drbd_calc_cpu_mask(A) ({})
552 * drbd_header_size - size of a packet header
554 * The header size is a multiple of 8, so any payload following the header is
555 * word aligned on 64-bit architectures. (The bitmap send and receive code
558 unsigned int drbd_header_size(struct drbd_connection
*connection
)
560 if (connection
->agreed_pro_version
>= 100) {
561 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100
), 8));
562 return sizeof(struct p_header100
);
564 BUILD_BUG_ON(sizeof(struct p_header80
) !=
565 sizeof(struct p_header95
));
566 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80
), 8));
567 return sizeof(struct p_header80
);
571 static unsigned int prepare_header80(struct p_header80
*h
, enum drbd_packet cmd
, int size
)
573 h
->magic
= cpu_to_be32(DRBD_MAGIC
);
574 h
->command
= cpu_to_be16(cmd
);
575 h
->length
= cpu_to_be16(size
);
576 return sizeof(struct p_header80
);
579 static unsigned int prepare_header95(struct p_header95
*h
, enum drbd_packet cmd
, int size
)
581 h
->magic
= cpu_to_be16(DRBD_MAGIC_BIG
);
582 h
->command
= cpu_to_be16(cmd
);
583 h
->length
= cpu_to_be32(size
);
584 return sizeof(struct p_header95
);
587 static unsigned int prepare_header100(struct p_header100
*h
, enum drbd_packet cmd
,
590 h
->magic
= cpu_to_be32(DRBD_MAGIC_100
);
591 h
->volume
= cpu_to_be16(vnr
);
592 h
->command
= cpu_to_be16(cmd
);
593 h
->length
= cpu_to_be32(size
);
595 return sizeof(struct p_header100
);
598 static unsigned int prepare_header(struct drbd_connection
*connection
, int vnr
,
599 void *buffer
, enum drbd_packet cmd
, int size
)
601 if (connection
->agreed_pro_version
>= 100)
602 return prepare_header100(buffer
, cmd
, size
, vnr
);
603 else if (connection
->agreed_pro_version
>= 95 &&
604 size
> DRBD_MAX_SIZE_H80_PACKET
)
605 return prepare_header95(buffer
, cmd
, size
);
607 return prepare_header80(buffer
, cmd
, size
);
610 static void *__conn_prepare_command(struct drbd_connection
*connection
,
611 struct drbd_socket
*sock
)
615 return sock
->sbuf
+ drbd_header_size(connection
);
618 void *conn_prepare_command(struct drbd_connection
*connection
, struct drbd_socket
*sock
)
622 mutex_lock(&sock
->mutex
);
623 p
= __conn_prepare_command(connection
, sock
);
625 mutex_unlock(&sock
->mutex
);
630 void *drbd_prepare_command(struct drbd_peer_device
*peer_device
, struct drbd_socket
*sock
)
632 return conn_prepare_command(peer_device
->connection
, sock
);
635 static int __send_command(struct drbd_connection
*connection
, int vnr
,
636 struct drbd_socket
*sock
, enum drbd_packet cmd
,
637 unsigned int header_size
, void *data
,
644 * Called with @data == NULL and the size of the data blocks in @size
645 * for commands that send data blocks. For those commands, omit the
646 * MSG_MORE flag: this will increase the likelihood that data blocks
647 * which are page aligned on the sender will end up page aligned on the
650 msg_flags
= data
? MSG_MORE
: 0;
652 header_size
+= prepare_header(connection
, vnr
, sock
->sbuf
, cmd
,
654 err
= drbd_send_all(connection
, sock
->socket
, sock
->sbuf
, header_size
,
657 err
= drbd_send_all(connection
, sock
->socket
, data
, size
, 0);
658 /* DRBD protocol "pings" are latency critical.
659 * This is supposed to trigger tcp_push_pending_frames() */
660 if (!err
&& (cmd
== P_PING
|| cmd
== P_PING_ACK
))
661 drbd_tcp_nodelay(sock
->socket
);
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_peer_device
*peer_device
, struct drbd_socket
*sock
,
685 enum drbd_packet cmd
, unsigned int header_size
,
686 void *data
, unsigned int size
)
690 err
= __send_command(peer_device
->connection
, peer_device
->device
->vnr
,
691 sock
, cmd
, header_size
, data
, 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_peer_device
*peer_device
)
718 struct drbd_socket
*sock
;
719 struct p_rs_param_95
*p
;
721 const int apv
= peer_device
->connection
->agreed_pro_version
;
722 enum drbd_packet cmd
;
724 struct disk_conf
*dc
;
726 sock
= &peer_device
->connection
->data
;
727 p
= drbd_prepare_command(peer_device
, sock
);
732 nc
= rcu_dereference(peer_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(peer_device
->device
)) {
746 dc
= rcu_dereference(peer_device
->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
);
752 put_ldev(peer_device
->device
);
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(peer_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 drbd_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_peer_device
*peer_device
, u64 uuid_flags
)
828 struct drbd_device
*device
= peer_device
->device
;
829 struct drbd_socket
*sock
;
833 if (!get_ldev_if_state(device
, D_NEGOTIATING
))
836 sock
= &peer_device
->connection
->data
;
837 p
= drbd_prepare_command(peer_device
, sock
);
842 spin_lock_irq(&device
->ldev
->md
.uuid_lock
);
843 for (i
= UI_CURRENT
; i
< UI_SIZE
; i
++)
844 p
->uuid
[i
] = cpu_to_be64(device
->ldev
->md
.uuid
[i
]);
845 spin_unlock_irq(&device
->ldev
->md
.uuid_lock
);
847 device
->comm_bm_set
= drbd_bm_total_weight(device
);
848 p
->uuid
[UI_SIZE
] = cpu_to_be64(device
->comm_bm_set
);
850 uuid_flags
|= rcu_dereference(peer_device
->connection
->net_conf
)->discard_my_data
? 1 : 0;
852 uuid_flags
|= test_bit(CRASHED_PRIMARY
, &device
->flags
) ? 2 : 0;
853 uuid_flags
|= device
->new_state_tmp
.disk
== D_INCONSISTENT
? 4 : 0;
854 p
->uuid
[UI_FLAGS
] = cpu_to_be64(uuid_flags
);
857 return drbd_send_command(peer_device
, sock
, P_UUIDS
, sizeof(*p
), NULL
, 0);
860 int drbd_send_uuids(struct drbd_peer_device
*peer_device
)
862 return _drbd_send_uuids(peer_device
, 0);
865 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device
*peer_device
)
867 return _drbd_send_uuids(peer_device
, 8);
870 void drbd_print_uuids(struct drbd_device
*device
, const char *text
)
872 if (get_ldev_if_state(device
, D_NEGOTIATING
)) {
873 u64
*uuid
= device
->ldev
->md
.uuid
;
874 drbd_info(device
, "%s %016llX:%016llX:%016llX:%016llX\n",
876 (unsigned long long)uuid
[UI_CURRENT
],
877 (unsigned long long)uuid
[UI_BITMAP
],
878 (unsigned long long)uuid
[UI_HISTORY_START
],
879 (unsigned long long)uuid
[UI_HISTORY_END
]);
882 drbd_info(device
, "%s effective data uuid: %016llX\n",
884 (unsigned long long)device
->ed_uuid
);
888 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device
*peer_device
)
890 struct drbd_device
*device
= peer_device
->device
;
891 struct drbd_socket
*sock
;
895 D_ASSERT(device
, device
->state
.disk
== D_UP_TO_DATE
);
897 uuid
= device
->ldev
->md
.uuid
[UI_BITMAP
];
898 if (uuid
&& uuid
!= UUID_JUST_CREATED
)
899 uuid
= uuid
+ UUID_NEW_BM_OFFSET
;
901 get_random_bytes(&uuid
, sizeof(u64
));
902 drbd_uuid_set(device
, UI_BITMAP
, uuid
);
903 drbd_print_uuids(device
, "updated sync UUID");
904 drbd_md_sync(device
);
906 sock
= &peer_device
->connection
->data
;
907 p
= drbd_prepare_command(peer_device
, sock
);
909 p
->uuid
= cpu_to_be64(uuid
);
910 drbd_send_command(peer_device
, sock
, P_SYNC_UUID
, sizeof(*p
), NULL
, 0);
914 /* communicated if (agreed_features & DRBD_FF_WSAME) */
916 assign_p_sizes_qlim(struct drbd_device
*device
, struct p_sizes
*p
,
917 struct request_queue
*q
)
920 p
->qlim
->physical_block_size
= cpu_to_be32(queue_physical_block_size(q
));
921 p
->qlim
->logical_block_size
= cpu_to_be32(queue_logical_block_size(q
));
922 p
->qlim
->alignment_offset
= cpu_to_be32(queue_alignment_offset(q
));
923 p
->qlim
->io_min
= cpu_to_be32(queue_io_min(q
));
924 p
->qlim
->io_opt
= cpu_to_be32(queue_io_opt(q
));
925 p
->qlim
->discard_enabled
= blk_queue_discard(q
);
926 p
->qlim
->write_same_capable
= !!q
->limits
.max_write_same_sectors
;
928 q
= device
->rq_queue
;
929 p
->qlim
->physical_block_size
= cpu_to_be32(queue_physical_block_size(q
));
930 p
->qlim
->logical_block_size
= cpu_to_be32(queue_logical_block_size(q
));
931 p
->qlim
->alignment_offset
= 0;
932 p
->qlim
->io_min
= cpu_to_be32(queue_io_min(q
));
933 p
->qlim
->io_opt
= cpu_to_be32(queue_io_opt(q
));
934 p
->qlim
->discard_enabled
= 0;
935 p
->qlim
->write_same_capable
= 0;
939 int drbd_send_sizes(struct drbd_peer_device
*peer_device
, int trigger_reply
, enum dds_flags flags
)
941 struct drbd_device
*device
= peer_device
->device
;
942 struct drbd_socket
*sock
;
944 sector_t d_size
, u_size
;
946 unsigned int max_bio_size
;
947 unsigned int packet_size
;
949 sock
= &peer_device
->connection
->data
;
950 p
= drbd_prepare_command(peer_device
, sock
);
954 packet_size
= sizeof(*p
);
955 if (peer_device
->connection
->agreed_features
& DRBD_FF_WSAME
)
956 packet_size
+= sizeof(p
->qlim
[0]);
958 memset(p
, 0, packet_size
);
959 if (get_ldev_if_state(device
, D_NEGOTIATING
)) {
960 struct request_queue
*q
= bdev_get_queue(device
->ldev
->backing_bdev
);
961 d_size
= drbd_get_max_capacity(device
->ldev
);
963 u_size
= rcu_dereference(device
->ldev
->disk_conf
)->disk_size
;
965 q_order_type
= drbd_queue_order_type(device
);
966 max_bio_size
= queue_max_hw_sectors(q
) << 9;
967 max_bio_size
= min(max_bio_size
, DRBD_MAX_BIO_SIZE
);
968 assign_p_sizes_qlim(device
, p
, q
);
973 q_order_type
= QUEUE_ORDERED_NONE
;
974 max_bio_size
= DRBD_MAX_BIO_SIZE
; /* ... multiple BIOs per peer_request */
975 assign_p_sizes_qlim(device
, p
, NULL
);
978 if (peer_device
->connection
->agreed_pro_version
<= 94)
979 max_bio_size
= min(max_bio_size
, DRBD_MAX_SIZE_H80_PACKET
);
980 else if (peer_device
->connection
->agreed_pro_version
< 100)
981 max_bio_size
= min(max_bio_size
, DRBD_MAX_BIO_SIZE_P95
);
983 p
->d_size
= cpu_to_be64(d_size
);
984 p
->u_size
= cpu_to_be64(u_size
);
985 p
->c_size
= cpu_to_be64(trigger_reply
? 0 : drbd_get_capacity(device
->this_bdev
));
986 p
->max_bio_size
= cpu_to_be32(max_bio_size
);
987 p
->queue_order_type
= cpu_to_be16(q_order_type
);
988 p
->dds_flags
= cpu_to_be16(flags
);
990 return drbd_send_command(peer_device
, sock
, P_SIZES
, packet_size
, NULL
, 0);
994 * drbd_send_current_state() - Sends the drbd state to the peer
995 * @peer_device: DRBD peer device.
997 int drbd_send_current_state(struct drbd_peer_device
*peer_device
)
999 struct drbd_socket
*sock
;
1002 sock
= &peer_device
->connection
->data
;
1003 p
= drbd_prepare_command(peer_device
, sock
);
1006 p
->state
= cpu_to_be32(peer_device
->device
->state
.i
); /* Within the send mutex */
1007 return drbd_send_command(peer_device
, sock
, P_STATE
, sizeof(*p
), NULL
, 0);
1011 * drbd_send_state() - After a state change, sends the new state to the peer
1012 * @peer_device: DRBD peer device.
1013 * @state: the state to send, not necessarily the current state.
1015 * Each state change queues an "after_state_ch" work, which will eventually
1016 * send the resulting new state to the peer. If more state changes happen
1017 * between queuing and processing of the after_state_ch work, we still
1018 * want to send each intermediary state in the order it occurred.
1020 int drbd_send_state(struct drbd_peer_device
*peer_device
, union drbd_state state
)
1022 struct drbd_socket
*sock
;
1025 sock
= &peer_device
->connection
->data
;
1026 p
= drbd_prepare_command(peer_device
, sock
);
1029 p
->state
= cpu_to_be32(state
.i
); /* Within the send mutex */
1030 return drbd_send_command(peer_device
, sock
, P_STATE
, sizeof(*p
), NULL
, 0);
1033 int drbd_send_state_req(struct drbd_peer_device
*peer_device
, union drbd_state mask
, union drbd_state val
)
1035 struct drbd_socket
*sock
;
1036 struct p_req_state
*p
;
1038 sock
= &peer_device
->connection
->data
;
1039 p
= drbd_prepare_command(peer_device
, sock
);
1042 p
->mask
= cpu_to_be32(mask
.i
);
1043 p
->val
= cpu_to_be32(val
.i
);
1044 return drbd_send_command(peer_device
, sock
, P_STATE_CHG_REQ
, sizeof(*p
), NULL
, 0);
1047 int conn_send_state_req(struct drbd_connection
*connection
, union drbd_state mask
, union drbd_state val
)
1049 enum drbd_packet cmd
;
1050 struct drbd_socket
*sock
;
1051 struct p_req_state
*p
;
1053 cmd
= connection
->agreed_pro_version
< 100 ? P_STATE_CHG_REQ
: P_CONN_ST_CHG_REQ
;
1054 sock
= &connection
->data
;
1055 p
= conn_prepare_command(connection
, sock
);
1058 p
->mask
= cpu_to_be32(mask
.i
);
1059 p
->val
= cpu_to_be32(val
.i
);
1060 return conn_send_command(connection
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1063 void drbd_send_sr_reply(struct drbd_peer_device
*peer_device
, enum drbd_state_rv retcode
)
1065 struct drbd_socket
*sock
;
1066 struct p_req_state_reply
*p
;
1068 sock
= &peer_device
->connection
->meta
;
1069 p
= drbd_prepare_command(peer_device
, sock
);
1071 p
->retcode
= cpu_to_be32(retcode
);
1072 drbd_send_command(peer_device
, sock
, P_STATE_CHG_REPLY
, sizeof(*p
), NULL
, 0);
1076 void conn_send_sr_reply(struct drbd_connection
*connection
, enum drbd_state_rv retcode
)
1078 struct drbd_socket
*sock
;
1079 struct p_req_state_reply
*p
;
1080 enum drbd_packet cmd
= connection
->agreed_pro_version
< 100 ? P_STATE_CHG_REPLY
: P_CONN_ST_CHG_REPLY
;
1082 sock
= &connection
->meta
;
1083 p
= conn_prepare_command(connection
, sock
);
1085 p
->retcode
= cpu_to_be32(retcode
);
1086 conn_send_command(connection
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1090 static void dcbp_set_code(struct p_compressed_bm
*p
, enum drbd_bitmap_code code
)
1092 BUG_ON(code
& ~0xf);
1093 p
->encoding
= (p
->encoding
& ~0xf) | code
;
1096 static void dcbp_set_start(struct p_compressed_bm
*p
, int set
)
1098 p
->encoding
= (p
->encoding
& ~0x80) | (set
? 0x80 : 0);
1101 static void dcbp_set_pad_bits(struct p_compressed_bm
*p
, int n
)
1104 p
->encoding
= (p
->encoding
& (~0x7 << 4)) | (n
<< 4);
1107 static int fill_bitmap_rle_bits(struct drbd_device
*device
,
1108 struct p_compressed_bm
*p
,
1110 struct bm_xfer_ctx
*c
)
1112 struct bitstream bs
;
1113 unsigned long plain_bits
;
1120 /* may we use this feature? */
1122 use_rle
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
)->use_rle
;
1124 if (!use_rle
|| first_peer_device(device
)->connection
->agreed_pro_version
< 90)
1127 if (c
->bit_offset
>= c
->bm_bits
)
1128 return 0; /* nothing to do. */
1130 /* use at most thus many bytes */
1131 bitstream_init(&bs
, p
->code
, size
, 0);
1132 memset(p
->code
, 0, size
);
1133 /* plain bits covered in this code string */
1136 /* p->encoding & 0x80 stores whether the first run length is set.
1137 * bit offset is implicit.
1138 * start with toggle == 2 to be able to tell the first iteration */
1141 /* see how much plain bits we can stuff into one packet
1142 * using RLE and VLI. */
1144 tmp
= (toggle
== 0) ? _drbd_bm_find_next_zero(device
, c
->bit_offset
)
1145 : _drbd_bm_find_next(device
, c
->bit_offset
);
1148 rl
= tmp
- c
->bit_offset
;
1150 if (toggle
== 2) { /* first iteration */
1152 /* the first checked bit was set,
1153 * store start value, */
1154 dcbp_set_start(p
, 1);
1155 /* but skip encoding of zero run length */
1159 dcbp_set_start(p
, 0);
1162 /* paranoia: catch zero runlength.
1163 * can only happen if bitmap is modified while we scan it. */
1165 drbd_err(device
, "unexpected zero runlength while encoding bitmap "
1166 "t:%u bo:%lu\n", toggle
, c
->bit_offset
);
1170 bits
= vli_encode_bits(&bs
, rl
);
1171 if (bits
== -ENOBUFS
) /* buffer full */
1174 drbd_err(device
, "error while encoding bitmap: %d\n", bits
);
1180 c
->bit_offset
= tmp
;
1181 } while (c
->bit_offset
< c
->bm_bits
);
1183 len
= bs
.cur
.b
- p
->code
+ !!bs
.cur
.bit
;
1185 if (plain_bits
< (len
<< 3)) {
1186 /* incompressible with this method.
1187 * we need to rewind both word and bit position. */
1188 c
->bit_offset
-= plain_bits
;
1189 bm_xfer_ctx_bit_to_word_offset(c
);
1190 c
->bit_offset
= c
->word_offset
* BITS_PER_LONG
;
1194 /* RLE + VLI was able to compress it just fine.
1195 * update c->word_offset. */
1196 bm_xfer_ctx_bit_to_word_offset(c
);
1198 /* store pad_bits */
1199 dcbp_set_pad_bits(p
, (8 - bs
.cur
.bit
) & 0x7);
1205 * send_bitmap_rle_or_plain
1207 * Return 0 when done, 1 when another iteration is needed, and a negative error
1208 * code upon failure.
1211 send_bitmap_rle_or_plain(struct drbd_device
*device
, struct bm_xfer_ctx
*c
)
1213 struct drbd_socket
*sock
= &first_peer_device(device
)->connection
->data
;
1214 unsigned int header_size
= drbd_header_size(first_peer_device(device
)->connection
);
1215 struct p_compressed_bm
*p
= sock
->sbuf
+ header_size
;
1218 len
= fill_bitmap_rle_bits(device
, p
,
1219 DRBD_SOCKET_BUFFER_SIZE
- header_size
- sizeof(*p
), c
);
1224 dcbp_set_code(p
, RLE_VLI_Bits
);
1225 err
= __send_command(first_peer_device(device
)->connection
, device
->vnr
, sock
,
1226 P_COMPRESSED_BITMAP
, sizeof(*p
) + len
,
1229 c
->bytes
[0] += header_size
+ sizeof(*p
) + len
;
1231 if (c
->bit_offset
>= c
->bm_bits
)
1234 /* was not compressible.
1235 * send a buffer full of plain text bits instead. */
1236 unsigned int data_size
;
1237 unsigned long num_words
;
1238 unsigned long *p
= sock
->sbuf
+ header_size
;
1240 data_size
= DRBD_SOCKET_BUFFER_SIZE
- header_size
;
1241 num_words
= min_t(size_t, data_size
/ sizeof(*p
),
1242 c
->bm_words
- c
->word_offset
);
1243 len
= num_words
* sizeof(*p
);
1245 drbd_bm_get_lel(device
, c
->word_offset
, num_words
, p
);
1246 err
= __send_command(first_peer_device(device
)->connection
, device
->vnr
, sock
, P_BITMAP
, len
, NULL
, 0);
1247 c
->word_offset
+= num_words
;
1248 c
->bit_offset
= c
->word_offset
* BITS_PER_LONG
;
1251 c
->bytes
[1] += header_size
+ len
;
1253 if (c
->bit_offset
> c
->bm_bits
)
1254 c
->bit_offset
= c
->bm_bits
;
1258 INFO_bm_xfer_stats(device
, "send", c
);
1266 /* See the comment at receive_bitmap() */
1267 static int _drbd_send_bitmap(struct drbd_device
*device
)
1269 struct bm_xfer_ctx c
;
1272 if (!expect(device
->bitmap
))
1275 if (get_ldev(device
)) {
1276 if (drbd_md_test_flag(device
->ldev
, MDF_FULL_SYNC
)) {
1277 drbd_info(device
, "Writing the whole bitmap, MDF_FullSync was set.\n");
1278 drbd_bm_set_all(device
);
1279 if (drbd_bm_write(device
)) {
1280 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1281 * but otherwise process as per normal - need to tell other
1282 * side that a full resync is required! */
1283 drbd_err(device
, "Failed to write bitmap to disk!\n");
1285 drbd_md_clear_flag(device
, MDF_FULL_SYNC
);
1286 drbd_md_sync(device
);
1292 c
= (struct bm_xfer_ctx
) {
1293 .bm_bits
= drbd_bm_bits(device
),
1294 .bm_words
= drbd_bm_words(device
),
1298 err
= send_bitmap_rle_or_plain(device
, &c
);
1304 int drbd_send_bitmap(struct drbd_device
*device
)
1306 struct drbd_socket
*sock
= &first_peer_device(device
)->connection
->data
;
1309 mutex_lock(&sock
->mutex
);
1311 err
= !_drbd_send_bitmap(device
);
1312 mutex_unlock(&sock
->mutex
);
1316 void drbd_send_b_ack(struct drbd_connection
*connection
, u32 barrier_nr
, u32 set_size
)
1318 struct drbd_socket
*sock
;
1319 struct p_barrier_ack
*p
;
1321 if (connection
->cstate
< C_WF_REPORT_PARAMS
)
1324 sock
= &connection
->meta
;
1325 p
= conn_prepare_command(connection
, sock
);
1328 p
->barrier
= barrier_nr
;
1329 p
->set_size
= cpu_to_be32(set_size
);
1330 conn_send_command(connection
, sock
, P_BARRIER_ACK
, sizeof(*p
), NULL
, 0);
1334 * _drbd_send_ack() - Sends an ack packet
1335 * @device: DRBD device.
1336 * @cmd: Packet command code.
1337 * @sector: sector, needs to be in big endian byte order
1338 * @blksize: size in byte, needs to be in big endian byte order
1339 * @block_id: Id, big endian byte order
1341 static int _drbd_send_ack(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1342 u64 sector
, u32 blksize
, u64 block_id
)
1344 struct drbd_socket
*sock
;
1345 struct p_block_ack
*p
;
1347 if (peer_device
->device
->state
.conn
< C_CONNECTED
)
1350 sock
= &peer_device
->connection
->meta
;
1351 p
= drbd_prepare_command(peer_device
, sock
);
1355 p
->block_id
= block_id
;
1356 p
->blksize
= blksize
;
1357 p
->seq_num
= cpu_to_be32(atomic_inc_return(&peer_device
->device
->packet_seq
));
1358 return drbd_send_command(peer_device
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1361 /* dp->sector and dp->block_id already/still in network byte order,
1362 * data_size is payload size according to dp->head,
1363 * and may need to be corrected for digest size. */
1364 void drbd_send_ack_dp(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1365 struct p_data
*dp
, int data_size
)
1367 if (peer_device
->connection
->peer_integrity_tfm
)
1368 data_size
-= crypto_shash_digestsize(peer_device
->connection
->peer_integrity_tfm
);
1369 _drbd_send_ack(peer_device
, cmd
, dp
->sector
, cpu_to_be32(data_size
),
1373 void drbd_send_ack_rp(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1374 struct p_block_req
*rp
)
1376 _drbd_send_ack(peer_device
, cmd
, rp
->sector
, rp
->blksize
, rp
->block_id
);
1380 * drbd_send_ack() - Sends an ack packet
1381 * @device: DRBD device
1382 * @cmd: packet command code
1383 * @peer_req: peer request
1385 int drbd_send_ack(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1386 struct drbd_peer_request
*peer_req
)
1388 return _drbd_send_ack(peer_device
, cmd
,
1389 cpu_to_be64(peer_req
->i
.sector
),
1390 cpu_to_be32(peer_req
->i
.size
),
1391 peer_req
->block_id
);
1394 /* This function misuses the block_id field to signal if the blocks
1395 * are is sync or not. */
1396 int drbd_send_ack_ex(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1397 sector_t sector
, int blksize
, u64 block_id
)
1399 return _drbd_send_ack(peer_device
, cmd
,
1400 cpu_to_be64(sector
),
1401 cpu_to_be32(blksize
),
1402 cpu_to_be64(block_id
));
1405 int drbd_send_rs_deallocated(struct drbd_peer_device
*peer_device
,
1406 struct drbd_peer_request
*peer_req
)
1408 struct drbd_socket
*sock
;
1409 struct p_block_desc
*p
;
1411 sock
= &peer_device
->connection
->data
;
1412 p
= drbd_prepare_command(peer_device
, sock
);
1415 p
->sector
= cpu_to_be64(peer_req
->i
.sector
);
1416 p
->blksize
= cpu_to_be32(peer_req
->i
.size
);
1418 return drbd_send_command(peer_device
, sock
, P_RS_DEALLOCATED
, sizeof(*p
), NULL
, 0);
1421 int drbd_send_drequest(struct drbd_peer_device
*peer_device
, int cmd
,
1422 sector_t sector
, int size
, u64 block_id
)
1424 struct drbd_socket
*sock
;
1425 struct p_block_req
*p
;
1427 sock
= &peer_device
->connection
->data
;
1428 p
= drbd_prepare_command(peer_device
, sock
);
1431 p
->sector
= cpu_to_be64(sector
);
1432 p
->block_id
= block_id
;
1433 p
->blksize
= cpu_to_be32(size
);
1434 return drbd_send_command(peer_device
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1437 int drbd_send_drequest_csum(struct drbd_peer_device
*peer_device
, sector_t sector
, int size
,
1438 void *digest
, int digest_size
, enum drbd_packet cmd
)
1440 struct drbd_socket
*sock
;
1441 struct p_block_req
*p
;
1443 /* FIXME: Put the digest into the preallocated socket buffer. */
1445 sock
= &peer_device
->connection
->data
;
1446 p
= drbd_prepare_command(peer_device
, sock
);
1449 p
->sector
= cpu_to_be64(sector
);
1450 p
->block_id
= ID_SYNCER
/* unused */;
1451 p
->blksize
= cpu_to_be32(size
);
1452 return drbd_send_command(peer_device
, sock
, cmd
, sizeof(*p
), digest
, digest_size
);
1455 int drbd_send_ov_request(struct drbd_peer_device
*peer_device
, sector_t sector
, int size
)
1457 struct drbd_socket
*sock
;
1458 struct p_block_req
*p
;
1460 sock
= &peer_device
->connection
->data
;
1461 p
= drbd_prepare_command(peer_device
, sock
);
1464 p
->sector
= cpu_to_be64(sector
);
1465 p
->block_id
= ID_SYNCER
/* unused */;
1466 p
->blksize
= cpu_to_be32(size
);
1467 return drbd_send_command(peer_device
, sock
, P_OV_REQUEST
, sizeof(*p
), NULL
, 0);
1470 /* called on sndtimeo
1471 * returns false if we should retry,
1472 * true if we think connection is dead
1474 static int we_should_drop_the_connection(struct drbd_connection
*connection
, struct socket
*sock
)
1477 /* long elapsed = (long)(jiffies - device->last_received); */
1479 drop_it
= connection
->meta
.socket
== sock
1480 || !connection
->ack_receiver
.task
1481 || get_t_state(&connection
->ack_receiver
) != RUNNING
1482 || connection
->cstate
< C_WF_REPORT_PARAMS
;
1487 drop_it
= !--connection
->ko_count
;
1489 drbd_err(connection
, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1490 current
->comm
, current
->pid
, connection
->ko_count
);
1491 request_ping(connection
);
1494 return drop_it
; /* && (device->state == R_PRIMARY) */;
1497 static void drbd_update_congested(struct drbd_connection
*connection
)
1499 struct sock
*sk
= connection
->data
.socket
->sk
;
1500 if (sk
->sk_wmem_queued
> sk
->sk_sndbuf
* 4 / 5)
1501 set_bit(NET_CONGESTED
, &connection
->flags
);
1504 /* The idea of sendpage seems to be to put some kind of reference
1505 * to the page into the skb, and to hand it over to the NIC. In
1506 * this process get_page() gets called.
1508 * As soon as the page was really sent over the network put_page()
1509 * gets called by some part of the network layer. [ NIC driver? ]
1511 * [ get_page() / put_page() increment/decrement the count. If count
1512 * reaches 0 the page will be freed. ]
1514 * This works nicely with pages from FSs.
1515 * But this means that in protocol A we might signal IO completion too early!
1517 * In order not to corrupt data during a resync we must make sure
1518 * that we do not reuse our own buffer pages (EEs) to early, therefore
1519 * we have the net_ee list.
1521 * XFS seems to have problems, still, it submits pages with page_count == 0!
1522 * As a workaround, we disable sendpage on pages
1523 * with page_count == 0 or PageSlab.
1525 static int _drbd_no_send_page(struct drbd_peer_device
*peer_device
, struct page
*page
,
1526 int offset
, size_t size
, unsigned msg_flags
)
1528 struct socket
*socket
;
1532 socket
= peer_device
->connection
->data
.socket
;
1533 addr
= kmap(page
) + offset
;
1534 err
= drbd_send_all(peer_device
->connection
, socket
, addr
, size
, msg_flags
);
1537 peer_device
->device
->send_cnt
+= size
>> 9;
1541 static int _drbd_send_page(struct drbd_peer_device
*peer_device
, struct page
*page
,
1542 int offset
, size_t size
, unsigned msg_flags
)
1544 struct socket
*socket
= peer_device
->connection
->data
.socket
;
1548 /* e.g. XFS meta- & log-data is in slab pages, which have a
1549 * page_count of 0 and/or have PageSlab() set.
1550 * we cannot use send_page for those, as that does get_page();
1551 * put_page(); and would cause either a VM_BUG directly, or
1552 * __page_cache_release a page that would actually still be referenced
1553 * by someone, leading to some obscure delayed Oops somewhere else. */
1554 if (drbd_disable_sendpage
|| (page_count(page
) < 1) || PageSlab(page
))
1555 return _drbd_no_send_page(peer_device
, page
, offset
, size
, msg_flags
);
1557 msg_flags
|= MSG_NOSIGNAL
;
1558 drbd_update_congested(peer_device
->connection
);
1562 sent
= socket
->ops
->sendpage(socket
, page
, offset
, len
, msg_flags
);
1564 if (sent
== -EAGAIN
) {
1565 if (we_should_drop_the_connection(peer_device
->connection
, socket
))
1569 drbd_warn(peer_device
->device
, "%s: size=%d len=%d sent=%d\n",
1570 __func__
, (int)size
, len
, sent
);
1577 } while (len
> 0 /* THINK && device->cstate >= C_CONNECTED*/);
1578 clear_bit(NET_CONGESTED
, &peer_device
->connection
->flags
);
1582 peer_device
->device
->send_cnt
+= size
>> 9;
1587 static int _drbd_send_bio(struct drbd_peer_device
*peer_device
, struct bio
*bio
)
1589 struct bio_vec bvec
;
1590 struct bvec_iter iter
;
1592 /* hint all but last page with MSG_MORE */
1593 bio_for_each_segment(bvec
, bio
, iter
) {
1596 err
= _drbd_no_send_page(peer_device
, bvec
.bv_page
,
1597 bvec
.bv_offset
, bvec
.bv_len
,
1598 bio_iter_last(bvec
, iter
)
1602 /* REQ_OP_WRITE_SAME has only one segment */
1603 if (bio_op(bio
) == REQ_OP_WRITE_SAME
)
1609 static int _drbd_send_zc_bio(struct drbd_peer_device
*peer_device
, struct bio
*bio
)
1611 struct bio_vec bvec
;
1612 struct bvec_iter iter
;
1614 /* hint all but last page with MSG_MORE */
1615 bio_for_each_segment(bvec
, bio
, iter
) {
1618 err
= _drbd_send_page(peer_device
, bvec
.bv_page
,
1619 bvec
.bv_offset
, bvec
.bv_len
,
1620 bio_iter_last(bvec
, iter
) ? 0 : MSG_MORE
);
1623 /* REQ_OP_WRITE_SAME has only one segment */
1624 if (bio_op(bio
) == REQ_OP_WRITE_SAME
)
1630 static int _drbd_send_zc_ee(struct drbd_peer_device
*peer_device
,
1631 struct drbd_peer_request
*peer_req
)
1633 struct page
*page
= peer_req
->pages
;
1634 unsigned len
= peer_req
->i
.size
;
1637 /* hint all but last page with MSG_MORE */
1638 page_chain_for_each(page
) {
1639 unsigned l
= min_t(unsigned, len
, PAGE_SIZE
);
1641 err
= _drbd_send_page(peer_device
, page
, 0, l
,
1642 page_chain_next(page
) ? MSG_MORE
: 0);
1650 static u32
bio_flags_to_wire(struct drbd_connection
*connection
,
1653 if (connection
->agreed_pro_version
>= 95)
1654 return (bio
->bi_opf
& REQ_SYNC
? DP_RW_SYNC
: 0) |
1655 (bio
->bi_opf
& REQ_FUA
? DP_FUA
: 0) |
1656 (bio
->bi_opf
& REQ_PREFLUSH
? DP_FLUSH
: 0) |
1657 (bio_op(bio
) == REQ_OP_WRITE_SAME
? DP_WSAME
: 0) |
1658 (bio_op(bio
) == REQ_OP_DISCARD
? DP_DISCARD
: 0) |
1659 (bio_op(bio
) == REQ_OP_WRITE_ZEROES
?
1660 ((connection
->agreed_features
& DRBD_FF_WZEROES
) ?
1661 (DP_ZEROES
|(!(bio
->bi_opf
& REQ_NOUNMAP
) ? DP_DISCARD
: 0))
1665 return bio
->bi_opf
& REQ_SYNC
? DP_RW_SYNC
: 0;
1668 /* Used to send write or TRIM aka REQ_OP_DISCARD requests
1669 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1671 int drbd_send_dblock(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
1673 struct drbd_device
*device
= peer_device
->device
;
1674 struct drbd_socket
*sock
;
1676 struct p_wsame
*wsame
= NULL
;
1678 unsigned int dp_flags
= 0;
1682 sock
= &peer_device
->connection
->data
;
1683 p
= drbd_prepare_command(peer_device
, sock
);
1684 digest_size
= peer_device
->connection
->integrity_tfm
?
1685 crypto_shash_digestsize(peer_device
->connection
->integrity_tfm
) : 0;
1689 p
->sector
= cpu_to_be64(req
->i
.sector
);
1690 p
->block_id
= (unsigned long)req
;
1691 p
->seq_num
= cpu_to_be32(atomic_inc_return(&device
->packet_seq
));
1692 dp_flags
= bio_flags_to_wire(peer_device
->connection
, req
->master_bio
);
1693 if (device
->state
.conn
>= C_SYNC_SOURCE
&&
1694 device
->state
.conn
<= C_PAUSED_SYNC_T
)
1695 dp_flags
|= DP_MAY_SET_IN_SYNC
;
1696 if (peer_device
->connection
->agreed_pro_version
>= 100) {
1697 if (req
->rq_state
& RQ_EXP_RECEIVE_ACK
)
1698 dp_flags
|= DP_SEND_RECEIVE_ACK
;
1699 /* During resync, request an explicit write ack,
1700 * even in protocol != C */
1701 if (req
->rq_state
& RQ_EXP_WRITE_ACK
1702 || (dp_flags
& DP_MAY_SET_IN_SYNC
))
1703 dp_flags
|= DP_SEND_WRITE_ACK
;
1705 p
->dp_flags
= cpu_to_be32(dp_flags
);
1707 if (dp_flags
& (DP_DISCARD
|DP_ZEROES
)) {
1708 enum drbd_packet cmd
= (dp_flags
& DP_ZEROES
) ? P_ZEROES
: P_TRIM
;
1709 struct p_trim
*t
= (struct p_trim
*)p
;
1710 t
->size
= cpu_to_be32(req
->i
.size
);
1711 err
= __send_command(peer_device
->connection
, device
->vnr
, sock
, cmd
, sizeof(*t
), NULL
, 0);
1714 if (dp_flags
& DP_WSAME
) {
1715 /* this will only work if DRBD_FF_WSAME is set AND the
1716 * handshake agreed that all nodes and backend devices are
1717 * WRITE_SAME capable and agree on logical_block_size */
1718 wsame
= (struct p_wsame
*)p
;
1719 digest_out
= wsame
+ 1;
1720 wsame
->size
= cpu_to_be32(req
->i
.size
);
1724 /* our digest is still only over the payload.
1725 * TRIM does not carry any payload. */
1727 drbd_csum_bio(peer_device
->connection
->integrity_tfm
, req
->master_bio
, digest_out
);
1730 __send_command(peer_device
->connection
, device
->vnr
, sock
, P_WSAME
,
1731 sizeof(*wsame
) + digest_size
, NULL
,
1732 bio_iovec(req
->master_bio
).bv_len
);
1735 __send_command(peer_device
->connection
, device
->vnr
, sock
, P_DATA
,
1736 sizeof(*p
) + digest_size
, NULL
, req
->i
.size
);
1738 /* For protocol A, we have to memcpy the payload into
1739 * socket buffers, as we may complete right away
1740 * as soon as we handed it over to tcp, at which point the data
1741 * pages may become invalid.
1743 * For data-integrity enabled, we copy it as well, so we can be
1744 * sure that even if the bio pages may still be modified, it
1745 * won't change the data on the wire, thus if the digest checks
1746 * out ok after sending on this side, but does not fit on the
1747 * receiving side, we sure have detected corruption elsewhere.
1749 if (!(req
->rq_state
& (RQ_EXP_RECEIVE_ACK
| RQ_EXP_WRITE_ACK
)) || digest_size
)
1750 err
= _drbd_send_bio(peer_device
, req
->master_bio
);
1752 err
= _drbd_send_zc_bio(peer_device
, req
->master_bio
);
1754 /* double check digest, sometimes buffers have been modified in flight. */
1755 if (digest_size
> 0 && digest_size
<= 64) {
1756 /* 64 byte, 512 bit, is the largest digest size
1757 * currently supported in kernel crypto. */
1758 unsigned char digest
[64];
1759 drbd_csum_bio(peer_device
->connection
->integrity_tfm
, req
->master_bio
, digest
);
1760 if (memcmp(p
+ 1, digest
, digest_size
)) {
1762 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1763 (unsigned long long)req
->i
.sector
, req
->i
.size
);
1765 } /* else if (digest_size > 64) {
1766 ... Be noisy about digest too large ...
1770 mutex_unlock(&sock
->mutex
); /* locked by drbd_prepare_command() */
1775 /* answer packet, used to send data back for read requests:
1776 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1777 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1779 int drbd_send_block(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1780 struct drbd_peer_request
*peer_req
)
1782 struct drbd_device
*device
= peer_device
->device
;
1783 struct drbd_socket
*sock
;
1788 sock
= &peer_device
->connection
->data
;
1789 p
= drbd_prepare_command(peer_device
, sock
);
1791 digest_size
= peer_device
->connection
->integrity_tfm
?
1792 crypto_shash_digestsize(peer_device
->connection
->integrity_tfm
) : 0;
1796 p
->sector
= cpu_to_be64(peer_req
->i
.sector
);
1797 p
->block_id
= peer_req
->block_id
;
1798 p
->seq_num
= 0; /* unused */
1801 drbd_csum_ee(peer_device
->connection
->integrity_tfm
, peer_req
, p
+ 1);
1802 err
= __send_command(peer_device
->connection
, device
->vnr
, sock
, cmd
, sizeof(*p
) + digest_size
, NULL
, peer_req
->i
.size
);
1804 err
= _drbd_send_zc_ee(peer_device
, peer_req
);
1805 mutex_unlock(&sock
->mutex
); /* locked by drbd_prepare_command() */
1810 int drbd_send_out_of_sync(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
1812 struct drbd_socket
*sock
;
1813 struct p_block_desc
*p
;
1815 sock
= &peer_device
->connection
->data
;
1816 p
= drbd_prepare_command(peer_device
, sock
);
1819 p
->sector
= cpu_to_be64(req
->i
.sector
);
1820 p
->blksize
= cpu_to_be32(req
->i
.size
);
1821 return drbd_send_command(peer_device
, sock
, P_OUT_OF_SYNC
, sizeof(*p
), NULL
, 0);
1825 drbd_send distinguishes two cases:
1827 Packets sent via the data socket "sock"
1828 and packets sent via the meta data socket "msock"
1831 -----------------+-------------------------+------------------------------
1832 timeout conf.timeout / 2 conf.timeout / 2
1833 timeout action send a ping via msock Abort communication
1834 and close all sockets
1838 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1840 int drbd_send(struct drbd_connection
*connection
, struct socket
*sock
,
1841 void *buf
, size_t size
, unsigned msg_flags
)
1843 struct kvec iov
= {.iov_base
= buf
, .iov_len
= size
};
1844 struct msghdr msg
= {.msg_flags
= msg_flags
| MSG_NOSIGNAL
};
1850 /* THINK if (signal_pending) return ... ? */
1852 iov_iter_kvec(&msg
.msg_iter
, WRITE
, &iov
, 1, size
);
1854 if (sock
== connection
->data
.socket
) {
1856 connection
->ko_count
= rcu_dereference(connection
->net_conf
)->ko_count
;
1858 drbd_update_congested(connection
);
1861 rv
= sock_sendmsg(sock
, &msg
);
1862 if (rv
== -EAGAIN
) {
1863 if (we_should_drop_the_connection(connection
, sock
))
1869 flush_signals(current
);
1875 } while (sent
< size
);
1877 if (sock
== connection
->data
.socket
)
1878 clear_bit(NET_CONGESTED
, &connection
->flags
);
1881 if (rv
!= -EAGAIN
) {
1882 drbd_err(connection
, "%s_sendmsg returned %d\n",
1883 sock
== connection
->meta
.socket
? "msock" : "sock",
1885 conn_request_state(connection
, NS(conn
, C_BROKEN_PIPE
), CS_HARD
);
1887 conn_request_state(connection
, NS(conn
, C_TIMEOUT
), CS_HARD
);
1894 * drbd_send_all - Send an entire buffer
1896 * Returns 0 upon success and a negative error value otherwise.
1898 int drbd_send_all(struct drbd_connection
*connection
, struct socket
*sock
, void *buffer
,
1899 size_t size
, unsigned msg_flags
)
1903 err
= drbd_send(connection
, sock
, buffer
, size
, msg_flags
);
1911 static int drbd_open(struct block_device
*bdev
, fmode_t mode
)
1913 struct drbd_device
*device
= bdev
->bd_disk
->private_data
;
1914 unsigned long flags
;
1917 mutex_lock(&drbd_main_mutex
);
1918 spin_lock_irqsave(&device
->resource
->req_lock
, flags
);
1919 /* to have a stable device->state.role
1920 * and no race with updating open_cnt */
1922 if (device
->state
.role
!= R_PRIMARY
) {
1923 if (mode
& FMODE_WRITE
)
1925 else if (!drbd_allow_oos
)
1931 spin_unlock_irqrestore(&device
->resource
->req_lock
, flags
);
1932 mutex_unlock(&drbd_main_mutex
);
1937 static void drbd_release(struct gendisk
*gd
, fmode_t mode
)
1939 struct drbd_device
*device
= gd
->private_data
;
1940 mutex_lock(&drbd_main_mutex
);
1942 mutex_unlock(&drbd_main_mutex
);
1945 /* need to hold resource->req_lock */
1946 void drbd_queue_unplug(struct drbd_device
*device
)
1948 if (device
->state
.pdsk
>= D_INCONSISTENT
&& device
->state
.conn
>= C_CONNECTED
) {
1949 D_ASSERT(device
, device
->state
.role
== R_PRIMARY
);
1950 if (test_and_clear_bit(UNPLUG_REMOTE
, &device
->flags
)) {
1951 drbd_queue_work_if_unqueued(
1952 &first_peer_device(device
)->connection
->sender_work
,
1953 &device
->unplug_work
);
1958 static void drbd_set_defaults(struct drbd_device
*device
)
1960 /* Beware! The actual layout differs
1961 * between big endian and little endian */
1962 device
->state
= (union drbd_dev_state
) {
1963 { .role
= R_SECONDARY
,
1965 .conn
= C_STANDALONE
,
1971 void drbd_init_set_defaults(struct drbd_device
*device
)
1973 /* the memset(,0,) did most of this.
1974 * note: only assignments, no allocation in here */
1976 drbd_set_defaults(device
);
1978 atomic_set(&device
->ap_bio_cnt
, 0);
1979 atomic_set(&device
->ap_actlog_cnt
, 0);
1980 atomic_set(&device
->ap_pending_cnt
, 0);
1981 atomic_set(&device
->rs_pending_cnt
, 0);
1982 atomic_set(&device
->unacked_cnt
, 0);
1983 atomic_set(&device
->local_cnt
, 0);
1984 atomic_set(&device
->pp_in_use_by_net
, 0);
1985 atomic_set(&device
->rs_sect_in
, 0);
1986 atomic_set(&device
->rs_sect_ev
, 0);
1987 atomic_set(&device
->ap_in_flight
, 0);
1988 atomic_set(&device
->md_io
.in_use
, 0);
1990 mutex_init(&device
->own_state_mutex
);
1991 device
->state_mutex
= &device
->own_state_mutex
;
1993 spin_lock_init(&device
->al_lock
);
1994 spin_lock_init(&device
->peer_seq_lock
);
1996 INIT_LIST_HEAD(&device
->active_ee
);
1997 INIT_LIST_HEAD(&device
->sync_ee
);
1998 INIT_LIST_HEAD(&device
->done_ee
);
1999 INIT_LIST_HEAD(&device
->read_ee
);
2000 INIT_LIST_HEAD(&device
->net_ee
);
2001 INIT_LIST_HEAD(&device
->resync_reads
);
2002 INIT_LIST_HEAD(&device
->resync_work
.list
);
2003 INIT_LIST_HEAD(&device
->unplug_work
.list
);
2004 INIT_LIST_HEAD(&device
->bm_io_work
.w
.list
);
2005 INIT_LIST_HEAD(&device
->pending_master_completion
[0]);
2006 INIT_LIST_HEAD(&device
->pending_master_completion
[1]);
2007 INIT_LIST_HEAD(&device
->pending_completion
[0]);
2008 INIT_LIST_HEAD(&device
->pending_completion
[1]);
2010 device
->resync_work
.cb
= w_resync_timer
;
2011 device
->unplug_work
.cb
= w_send_write_hint
;
2012 device
->bm_io_work
.w
.cb
= w_bitmap_io
;
2014 timer_setup(&device
->resync_timer
, resync_timer_fn
, 0);
2015 timer_setup(&device
->md_sync_timer
, md_sync_timer_fn
, 0);
2016 timer_setup(&device
->start_resync_timer
, start_resync_timer_fn
, 0);
2017 timer_setup(&device
->request_timer
, request_timer_fn
, 0);
2019 init_waitqueue_head(&device
->misc_wait
);
2020 init_waitqueue_head(&device
->state_wait
);
2021 init_waitqueue_head(&device
->ee_wait
);
2022 init_waitqueue_head(&device
->al_wait
);
2023 init_waitqueue_head(&device
->seq_wait
);
2025 device
->resync_wenr
= LC_FREE
;
2026 device
->peer_max_bio_size
= DRBD_MAX_BIO_SIZE_SAFE
;
2027 device
->local_max_bio_size
= DRBD_MAX_BIO_SIZE_SAFE
;
2030 static void _drbd_set_my_capacity(struct drbd_device
*device
, sector_t size
)
2032 /* set_capacity(device->this_bdev->bd_disk, size); */
2033 set_capacity(device
->vdisk
, size
);
2034 device
->this_bdev
->bd_inode
->i_size
= (loff_t
)size
<< 9;
2037 void drbd_set_my_capacity(struct drbd_device
*device
, sector_t size
)
2040 _drbd_set_my_capacity(device
, size
);
2041 drbd_info(device
, "size = %s (%llu KB)\n",
2042 ppsize(ppb
, size
>>1), (unsigned long long)size
>>1);
2045 void drbd_device_cleanup(struct drbd_device
*device
)
2048 if (first_peer_device(device
)->connection
->receiver
.t_state
!= NONE
)
2049 drbd_err(device
, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2050 first_peer_device(device
)->connection
->receiver
.t_state
);
2052 device
->al_writ_cnt
=
2053 device
->bm_writ_cnt
=
2061 device
->rs_failed
= 0;
2062 device
->rs_last_events
= 0;
2063 device
->rs_last_sect_ev
= 0;
2064 for (i
= 0; i
< DRBD_SYNC_MARKS
; i
++) {
2065 device
->rs_mark_left
[i
] = 0;
2066 device
->rs_mark_time
[i
] = 0;
2068 D_ASSERT(device
, first_peer_device(device
)->connection
->net_conf
== NULL
);
2070 _drbd_set_my_capacity(device
, 0);
2071 if (device
->bitmap
) {
2072 /* maybe never allocated. */
2073 drbd_bm_resize(device
, 0, 1);
2074 drbd_bm_cleanup(device
);
2077 drbd_backing_dev_free(device
, device
->ldev
);
2078 device
->ldev
= NULL
;
2080 clear_bit(AL_SUSPENDED
, &device
->flags
);
2082 D_ASSERT(device
, list_empty(&device
->active_ee
));
2083 D_ASSERT(device
, list_empty(&device
->sync_ee
));
2084 D_ASSERT(device
, list_empty(&device
->done_ee
));
2085 D_ASSERT(device
, list_empty(&device
->read_ee
));
2086 D_ASSERT(device
, list_empty(&device
->net_ee
));
2087 D_ASSERT(device
, list_empty(&device
->resync_reads
));
2088 D_ASSERT(device
, list_empty(&first_peer_device(device
)->connection
->sender_work
.q
));
2089 D_ASSERT(device
, list_empty(&device
->resync_work
.list
));
2090 D_ASSERT(device
, list_empty(&device
->unplug_work
.list
));
2092 drbd_set_defaults(device
);
2096 static void drbd_destroy_mempools(void)
2100 while (drbd_pp_pool
) {
2101 page
= drbd_pp_pool
;
2102 drbd_pp_pool
= (struct page
*)page_private(page
);
2107 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2109 bioset_exit(&drbd_io_bio_set
);
2110 bioset_exit(&drbd_md_io_bio_set
);
2111 mempool_exit(&drbd_md_io_page_pool
);
2112 mempool_exit(&drbd_ee_mempool
);
2113 mempool_exit(&drbd_request_mempool
);
2114 kmem_cache_destroy(drbd_ee_cache
);
2115 kmem_cache_destroy(drbd_request_cache
);
2116 kmem_cache_destroy(drbd_bm_ext_cache
);
2117 kmem_cache_destroy(drbd_al_ext_cache
);
2119 drbd_ee_cache
= NULL
;
2120 drbd_request_cache
= NULL
;
2121 drbd_bm_ext_cache
= NULL
;
2122 drbd_al_ext_cache
= NULL
;
2127 static int drbd_create_mempools(void)
2130 const int number
= (DRBD_MAX_BIO_SIZE
/PAGE_SIZE
) * drbd_minor_count
;
2134 drbd_request_cache
= kmem_cache_create(
2135 "drbd_req", sizeof(struct drbd_request
), 0, 0, NULL
);
2136 if (drbd_request_cache
== NULL
)
2139 drbd_ee_cache
= kmem_cache_create(
2140 "drbd_ee", sizeof(struct drbd_peer_request
), 0, 0, NULL
);
2141 if (drbd_ee_cache
== NULL
)
2144 drbd_bm_ext_cache
= kmem_cache_create(
2145 "drbd_bm", sizeof(struct bm_extent
), 0, 0, NULL
);
2146 if (drbd_bm_ext_cache
== NULL
)
2149 drbd_al_ext_cache
= kmem_cache_create(
2150 "drbd_al", sizeof(struct lc_element
), 0, 0, NULL
);
2151 if (drbd_al_ext_cache
== NULL
)
2155 ret
= bioset_init(&drbd_io_bio_set
, BIO_POOL_SIZE
, 0, 0);
2159 ret
= bioset_init(&drbd_md_io_bio_set
, DRBD_MIN_POOL_PAGES
, 0,
2164 ret
= mempool_init_page_pool(&drbd_md_io_page_pool
, DRBD_MIN_POOL_PAGES
, 0);
2168 ret
= mempool_init_slab_pool(&drbd_request_mempool
, number
,
2169 drbd_request_cache
);
2173 ret
= mempool_init_slab_pool(&drbd_ee_mempool
, number
, drbd_ee_cache
);
2177 /* drbd's page pool */
2178 spin_lock_init(&drbd_pp_lock
);
2180 for (i
= 0; i
< number
; i
++) {
2181 page
= alloc_page(GFP_HIGHUSER
);
2184 set_page_private(page
, (unsigned long)drbd_pp_pool
);
2185 drbd_pp_pool
= page
;
2187 drbd_pp_vacant
= number
;
2192 drbd_destroy_mempools(); /* in case we allocated some */
2196 static void drbd_release_all_peer_reqs(struct drbd_device
*device
)
2200 rr
= drbd_free_peer_reqs(device
, &device
->active_ee
);
2202 drbd_err(device
, "%d EEs in active list found!\n", rr
);
2204 rr
= drbd_free_peer_reqs(device
, &device
->sync_ee
);
2206 drbd_err(device
, "%d EEs in sync list found!\n", rr
);
2208 rr
= drbd_free_peer_reqs(device
, &device
->read_ee
);
2210 drbd_err(device
, "%d EEs in read list found!\n", rr
);
2212 rr
= drbd_free_peer_reqs(device
, &device
->done_ee
);
2214 drbd_err(device
, "%d EEs in done list found!\n", rr
);
2216 rr
= drbd_free_peer_reqs(device
, &device
->net_ee
);
2218 drbd_err(device
, "%d EEs in net list found!\n", rr
);
2221 /* caution. no locking. */
2222 void drbd_destroy_device(struct kref
*kref
)
2224 struct drbd_device
*device
= container_of(kref
, struct drbd_device
, kref
);
2225 struct drbd_resource
*resource
= device
->resource
;
2226 struct drbd_peer_device
*peer_device
, *tmp_peer_device
;
2228 del_timer_sync(&device
->request_timer
);
2230 /* paranoia asserts */
2231 D_ASSERT(device
, device
->open_cnt
== 0);
2232 /* end paranoia asserts */
2234 /* cleanup stuff that may have been allocated during
2235 * device (re-)configuration or state changes */
2237 if (device
->this_bdev
)
2238 bdput(device
->this_bdev
);
2240 drbd_backing_dev_free(device
, device
->ldev
);
2241 device
->ldev
= NULL
;
2243 drbd_release_all_peer_reqs(device
);
2245 lc_destroy(device
->act_log
);
2246 lc_destroy(device
->resync
);
2248 kfree(device
->p_uuid
);
2249 /* device->p_uuid = NULL; */
2251 if (device
->bitmap
) /* should no longer be there. */
2252 drbd_bm_cleanup(device
);
2253 __free_page(device
->md_io
.page
);
2254 put_disk(device
->vdisk
);
2255 blk_cleanup_queue(device
->rq_queue
);
2256 kfree(device
->rs_plan_s
);
2258 /* not for_each_connection(connection, resource):
2259 * those may have been cleaned up and disassociated already.
2261 for_each_peer_device_safe(peer_device
, tmp_peer_device
, device
) {
2262 kref_put(&peer_device
->connection
->kref
, drbd_destroy_connection
);
2265 memset(device
, 0xfd, sizeof(*device
));
2267 kref_put(&resource
->kref
, drbd_destroy_resource
);
2270 /* One global retry thread, if we need to push back some bio and have it
2271 * reinserted through our make request function.
2273 static struct retry_worker
{
2274 struct workqueue_struct
*wq
;
2275 struct work_struct worker
;
2278 struct list_head writes
;
2281 static void do_retry(struct work_struct
*ws
)
2283 struct retry_worker
*retry
= container_of(ws
, struct retry_worker
, worker
);
2285 struct drbd_request
*req
, *tmp
;
2287 spin_lock_irq(&retry
->lock
);
2288 list_splice_init(&retry
->writes
, &writes
);
2289 spin_unlock_irq(&retry
->lock
);
2291 list_for_each_entry_safe(req
, tmp
, &writes
, tl_requests
) {
2292 struct drbd_device
*device
= req
->device
;
2293 struct bio
*bio
= req
->master_bio
;
2294 unsigned long start_jif
= req
->start_jif
;
2298 expect(atomic_read(&req
->completion_ref
) == 0) &&
2299 expect(req
->rq_state
& RQ_POSTPONED
) &&
2300 expect((req
->rq_state
& RQ_LOCAL_PENDING
) == 0 ||
2301 (req
->rq_state
& RQ_LOCAL_ABORTED
) != 0);
2304 drbd_err(device
, "req=%p completion_ref=%d rq_state=%x\n",
2305 req
, atomic_read(&req
->completion_ref
),
2308 /* We still need to put one kref associated with the
2309 * "completion_ref" going zero in the code path that queued it
2310 * here. The request object may still be referenced by a
2311 * frozen local req->private_bio, in case we force-detached.
2313 kref_put(&req
->kref
, drbd_req_destroy
);
2315 /* A single suspended or otherwise blocking device may stall
2316 * all others as well. Fortunately, this code path is to
2317 * recover from a situation that "should not happen":
2318 * concurrent writes in multi-primary setup.
2319 * In a "normal" lifecycle, this workqueue is supposed to be
2320 * destroyed without ever doing anything.
2321 * If it turns out to be an issue anyways, we can do per
2322 * resource (replication group) or per device (minor) retry
2323 * workqueues instead.
2326 /* We are not just doing generic_make_request(),
2327 * as we want to keep the start_time information. */
2329 __drbd_make_request(device
, bio
, start_jif
);
2333 /* called via drbd_req_put_completion_ref(),
2334 * holds resource->req_lock */
2335 void drbd_restart_request(struct drbd_request
*req
)
2337 unsigned long flags
;
2338 spin_lock_irqsave(&retry
.lock
, flags
);
2339 list_move_tail(&req
->tl_requests
, &retry
.writes
);
2340 spin_unlock_irqrestore(&retry
.lock
, flags
);
2342 /* Drop the extra reference that would otherwise
2343 * have been dropped by complete_master_bio.
2344 * do_retry() needs to grab a new one. */
2345 dec_ap_bio(req
->device
);
2347 queue_work(retry
.wq
, &retry
.worker
);
2350 void drbd_destroy_resource(struct kref
*kref
)
2352 struct drbd_resource
*resource
=
2353 container_of(kref
, struct drbd_resource
, kref
);
2355 idr_destroy(&resource
->devices
);
2356 free_cpumask_var(resource
->cpu_mask
);
2357 kfree(resource
->name
);
2358 memset(resource
, 0xf2, sizeof(*resource
));
2362 void drbd_free_resource(struct drbd_resource
*resource
)
2364 struct drbd_connection
*connection
, *tmp
;
2366 for_each_connection_safe(connection
, tmp
, resource
) {
2367 list_del(&connection
->connections
);
2368 drbd_debugfs_connection_cleanup(connection
);
2369 kref_put(&connection
->kref
, drbd_destroy_connection
);
2371 drbd_debugfs_resource_cleanup(resource
);
2372 kref_put(&resource
->kref
, drbd_destroy_resource
);
2375 static void drbd_cleanup(void)
2378 struct drbd_device
*device
;
2379 struct drbd_resource
*resource
, *tmp
;
2381 /* first remove proc,
2382 * drbdsetup uses it's presence to detect
2383 * whether DRBD is loaded.
2384 * If we would get stuck in proc removal,
2385 * but have netlink already deregistered,
2386 * some drbdsetup commands may wait forever
2390 remove_proc_entry("drbd", NULL
);
2393 destroy_workqueue(retry
.wq
);
2395 drbd_genl_unregister();
2397 idr_for_each_entry(&drbd_devices
, device
, i
)
2398 drbd_delete_device(device
);
2400 /* not _rcu since, no other updater anymore. Genl already unregistered */
2401 for_each_resource_safe(resource
, tmp
, &drbd_resources
) {
2402 list_del(&resource
->resources
);
2403 drbd_free_resource(resource
);
2406 drbd_debugfs_cleanup();
2408 drbd_destroy_mempools();
2409 unregister_blkdev(DRBD_MAJOR
, "drbd");
2411 idr_destroy(&drbd_devices
);
2413 pr_info("module cleanup done.\n");
2417 * drbd_congested() - Callback for the flusher thread
2418 * @congested_data: User data
2419 * @bdi_bits: Bits the BDI flusher thread is currently interested in
2421 * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested.
2423 static int drbd_congested(void *congested_data
, int bdi_bits
)
2425 struct drbd_device
*device
= congested_data
;
2426 struct request_queue
*q
;
2430 if (!may_inc_ap_bio(device
)) {
2431 /* DRBD has frozen IO */
2437 if (test_bit(CALLBACK_PENDING
, &first_peer_device(device
)->connection
->flags
)) {
2438 r
|= (1 << WB_async_congested
);
2439 /* Without good local data, we would need to read from remote,
2440 * and that would need the worker thread as well, which is
2441 * currently blocked waiting for that usermode helper to
2444 if (!get_ldev_if_state(device
, D_UP_TO_DATE
))
2445 r
|= (1 << WB_sync_congested
);
2453 if (get_ldev(device
)) {
2454 q
= bdev_get_queue(device
->ldev
->backing_bdev
);
2455 r
= bdi_congested(q
->backing_dev_info
, bdi_bits
);
2461 if (bdi_bits
& (1 << WB_async_congested
) &&
2462 test_bit(NET_CONGESTED
, &first_peer_device(device
)->connection
->flags
)) {
2463 r
|= (1 << WB_async_congested
);
2464 reason
= reason
== 'b' ? 'a' : 'n';
2468 device
->congestion_reason
= reason
;
2472 static void drbd_init_workqueue(struct drbd_work_queue
* wq
)
2474 spin_lock_init(&wq
->q_lock
);
2475 INIT_LIST_HEAD(&wq
->q
);
2476 init_waitqueue_head(&wq
->q_wait
);
2479 struct completion_work
{
2481 struct completion done
;
2484 static int w_complete(struct drbd_work
*w
, int cancel
)
2486 struct completion_work
*completion_work
=
2487 container_of(w
, struct completion_work
, w
);
2489 complete(&completion_work
->done
);
2493 void drbd_flush_workqueue(struct drbd_work_queue
*work_queue
)
2495 struct completion_work completion_work
;
2497 completion_work
.w
.cb
= w_complete
;
2498 init_completion(&completion_work
.done
);
2499 drbd_queue_work(work_queue
, &completion_work
.w
);
2500 wait_for_completion(&completion_work
.done
);
2503 struct drbd_resource
*drbd_find_resource(const char *name
)
2505 struct drbd_resource
*resource
;
2507 if (!name
|| !name
[0])
2511 for_each_resource_rcu(resource
, &drbd_resources
) {
2512 if (!strcmp(resource
->name
, name
)) {
2513 kref_get(&resource
->kref
);
2523 struct drbd_connection
*conn_get_by_addrs(void *my_addr
, int my_addr_len
,
2524 void *peer_addr
, int peer_addr_len
)
2526 struct drbd_resource
*resource
;
2527 struct drbd_connection
*connection
;
2530 for_each_resource_rcu(resource
, &drbd_resources
) {
2531 for_each_connection_rcu(connection
, resource
) {
2532 if (connection
->my_addr_len
== my_addr_len
&&
2533 connection
->peer_addr_len
== peer_addr_len
&&
2534 !memcmp(&connection
->my_addr
, my_addr
, my_addr_len
) &&
2535 !memcmp(&connection
->peer_addr
, peer_addr
, peer_addr_len
)) {
2536 kref_get(&connection
->kref
);
2547 static int drbd_alloc_socket(struct drbd_socket
*socket
)
2549 socket
->rbuf
= (void *) __get_free_page(GFP_KERNEL
);
2552 socket
->sbuf
= (void *) __get_free_page(GFP_KERNEL
);
2558 static void drbd_free_socket(struct drbd_socket
*socket
)
2560 free_page((unsigned long) socket
->sbuf
);
2561 free_page((unsigned long) socket
->rbuf
);
2564 void conn_free_crypto(struct drbd_connection
*connection
)
2566 drbd_free_sock(connection
);
2568 crypto_free_shash(connection
->csums_tfm
);
2569 crypto_free_shash(connection
->verify_tfm
);
2570 crypto_free_shash(connection
->cram_hmac_tfm
);
2571 crypto_free_shash(connection
->integrity_tfm
);
2572 crypto_free_shash(connection
->peer_integrity_tfm
);
2573 kfree(connection
->int_dig_in
);
2574 kfree(connection
->int_dig_vv
);
2576 connection
->csums_tfm
= NULL
;
2577 connection
->verify_tfm
= NULL
;
2578 connection
->cram_hmac_tfm
= NULL
;
2579 connection
->integrity_tfm
= NULL
;
2580 connection
->peer_integrity_tfm
= NULL
;
2581 connection
->int_dig_in
= NULL
;
2582 connection
->int_dig_vv
= NULL
;
2585 int set_resource_options(struct drbd_resource
*resource
, struct res_opts
*res_opts
)
2587 struct drbd_connection
*connection
;
2588 cpumask_var_t new_cpu_mask
;
2591 if (!zalloc_cpumask_var(&new_cpu_mask
, GFP_KERNEL
))
2594 /* silently ignore cpu mask on UP kernel */
2595 if (nr_cpu_ids
> 1 && res_opts
->cpu_mask
[0] != 0) {
2596 err
= bitmap_parse(res_opts
->cpu_mask
, DRBD_CPU_MASK_SIZE
,
2597 cpumask_bits(new_cpu_mask
), nr_cpu_ids
);
2598 if (err
== -EOVERFLOW
) {
2599 /* So what. mask it out. */
2600 cpumask_var_t tmp_cpu_mask
;
2601 if (zalloc_cpumask_var(&tmp_cpu_mask
, GFP_KERNEL
)) {
2602 cpumask_setall(tmp_cpu_mask
);
2603 cpumask_and(new_cpu_mask
, new_cpu_mask
, tmp_cpu_mask
);
2604 drbd_warn(resource
, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2606 strlen(res_opts
->cpu_mask
) > 12 ? "..." : "",
2608 free_cpumask_var(tmp_cpu_mask
);
2613 drbd_warn(resource
, "bitmap_parse() failed with %d\n", err
);
2614 /* retcode = ERR_CPU_MASK_PARSE; */
2618 resource
->res_opts
= *res_opts
;
2619 if (cpumask_empty(new_cpu_mask
))
2620 drbd_calc_cpu_mask(&new_cpu_mask
);
2621 if (!cpumask_equal(resource
->cpu_mask
, new_cpu_mask
)) {
2622 cpumask_copy(resource
->cpu_mask
, new_cpu_mask
);
2623 for_each_connection_rcu(connection
, resource
) {
2624 connection
->receiver
.reset_cpu_mask
= 1;
2625 connection
->ack_receiver
.reset_cpu_mask
= 1;
2626 connection
->worker
.reset_cpu_mask
= 1;
2632 free_cpumask_var(new_cpu_mask
);
2637 struct drbd_resource
*drbd_create_resource(const char *name
)
2639 struct drbd_resource
*resource
;
2641 resource
= kzalloc(sizeof(struct drbd_resource
), GFP_KERNEL
);
2644 resource
->name
= kstrdup(name
, GFP_KERNEL
);
2645 if (!resource
->name
)
2646 goto fail_free_resource
;
2647 if (!zalloc_cpumask_var(&resource
->cpu_mask
, GFP_KERNEL
))
2648 goto fail_free_name
;
2649 kref_init(&resource
->kref
);
2650 idr_init(&resource
->devices
);
2651 INIT_LIST_HEAD(&resource
->connections
);
2652 resource
->write_ordering
= WO_BDEV_FLUSH
;
2653 list_add_tail_rcu(&resource
->resources
, &drbd_resources
);
2654 mutex_init(&resource
->conf_update
);
2655 mutex_init(&resource
->adm_mutex
);
2656 spin_lock_init(&resource
->req_lock
);
2657 drbd_debugfs_resource_add(resource
);
2661 kfree(resource
->name
);
2668 /* caller must be under adm_mutex */
2669 struct drbd_connection
*conn_create(const char *name
, struct res_opts
*res_opts
)
2671 struct drbd_resource
*resource
;
2672 struct drbd_connection
*connection
;
2674 connection
= kzalloc(sizeof(struct drbd_connection
), GFP_KERNEL
);
2678 if (drbd_alloc_socket(&connection
->data
))
2680 if (drbd_alloc_socket(&connection
->meta
))
2683 connection
->current_epoch
= kzalloc(sizeof(struct drbd_epoch
), GFP_KERNEL
);
2684 if (!connection
->current_epoch
)
2687 INIT_LIST_HEAD(&connection
->transfer_log
);
2689 INIT_LIST_HEAD(&connection
->current_epoch
->list
);
2690 connection
->epochs
= 1;
2691 spin_lock_init(&connection
->epoch_lock
);
2693 connection
->send
.seen_any_write_yet
= false;
2694 connection
->send
.current_epoch_nr
= 0;
2695 connection
->send
.current_epoch_writes
= 0;
2697 resource
= drbd_create_resource(name
);
2701 connection
->cstate
= C_STANDALONE
;
2702 mutex_init(&connection
->cstate_mutex
);
2703 init_waitqueue_head(&connection
->ping_wait
);
2704 idr_init(&connection
->peer_devices
);
2706 drbd_init_workqueue(&connection
->sender_work
);
2707 mutex_init(&connection
->data
.mutex
);
2708 mutex_init(&connection
->meta
.mutex
);
2710 drbd_thread_init(resource
, &connection
->receiver
, drbd_receiver
, "receiver");
2711 connection
->receiver
.connection
= connection
;
2712 drbd_thread_init(resource
, &connection
->worker
, drbd_worker
, "worker");
2713 connection
->worker
.connection
= connection
;
2714 drbd_thread_init(resource
, &connection
->ack_receiver
, drbd_ack_receiver
, "ack_recv");
2715 connection
->ack_receiver
.connection
= connection
;
2717 kref_init(&connection
->kref
);
2719 connection
->resource
= resource
;
2721 if (set_resource_options(resource
, res_opts
))
2724 kref_get(&resource
->kref
);
2725 list_add_tail_rcu(&connection
->connections
, &resource
->connections
);
2726 drbd_debugfs_connection_add(connection
);
2730 list_del(&resource
->resources
);
2731 drbd_free_resource(resource
);
2733 kfree(connection
->current_epoch
);
2734 drbd_free_socket(&connection
->meta
);
2735 drbd_free_socket(&connection
->data
);
2740 void drbd_destroy_connection(struct kref
*kref
)
2742 struct drbd_connection
*connection
= container_of(kref
, struct drbd_connection
, kref
);
2743 struct drbd_resource
*resource
= connection
->resource
;
2745 if (atomic_read(&connection
->current_epoch
->epoch_size
) != 0)
2746 drbd_err(connection
, "epoch_size:%d\n", atomic_read(&connection
->current_epoch
->epoch_size
));
2747 kfree(connection
->current_epoch
);
2749 idr_destroy(&connection
->peer_devices
);
2751 drbd_free_socket(&connection
->meta
);
2752 drbd_free_socket(&connection
->data
);
2753 kfree(connection
->int_dig_in
);
2754 kfree(connection
->int_dig_vv
);
2755 memset(connection
, 0xfc, sizeof(*connection
));
2757 kref_put(&resource
->kref
, drbd_destroy_resource
);
2760 static int init_submitter(struct drbd_device
*device
)
2762 /* opencoded create_singlethread_workqueue(),
2763 * to be able to say "drbd%d", ..., minor */
2765 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM
, device
->minor
);
2766 if (!device
->submit
.wq
)
2769 INIT_WORK(&device
->submit
.worker
, do_submit
);
2770 INIT_LIST_HEAD(&device
->submit
.writes
);
2774 enum drbd_ret_code
drbd_create_device(struct drbd_config_context
*adm_ctx
, unsigned int minor
)
2776 struct drbd_resource
*resource
= adm_ctx
->resource
;
2777 struct drbd_connection
*connection
;
2778 struct drbd_device
*device
;
2779 struct drbd_peer_device
*peer_device
, *tmp_peer_device
;
2780 struct gendisk
*disk
;
2781 struct request_queue
*q
;
2783 int vnr
= adm_ctx
->volume
;
2784 enum drbd_ret_code err
= ERR_NOMEM
;
2786 device
= minor_to_device(minor
);
2788 return ERR_MINOR_OR_VOLUME_EXISTS
;
2790 /* GFP_KERNEL, we are outside of all write-out paths */
2791 device
= kzalloc(sizeof(struct drbd_device
), GFP_KERNEL
);
2794 kref_init(&device
->kref
);
2796 kref_get(&resource
->kref
);
2797 device
->resource
= resource
;
2798 device
->minor
= minor
;
2801 drbd_init_set_defaults(device
);
2803 q
= blk_alloc_queue_node(GFP_KERNEL
, NUMA_NO_NODE
);
2806 device
->rq_queue
= q
;
2807 q
->queuedata
= device
;
2809 disk
= alloc_disk(1);
2812 device
->vdisk
= disk
;
2814 set_disk_ro(disk
, true);
2817 disk
->major
= DRBD_MAJOR
;
2818 disk
->first_minor
= minor
;
2819 disk
->fops
= &drbd_ops
;
2820 sprintf(disk
->disk_name
, "drbd%d", minor
);
2821 disk
->private_data
= device
;
2823 device
->this_bdev
= bdget(MKDEV(DRBD_MAJOR
, minor
));
2824 /* we have no partitions. we contain only ourselves. */
2825 device
->this_bdev
->bd_contains
= device
->this_bdev
;
2827 q
->backing_dev_info
->congested_fn
= drbd_congested
;
2828 q
->backing_dev_info
->congested_data
= device
;
2830 blk_queue_make_request(q
, drbd_make_request
);
2831 blk_queue_write_cache(q
, true, true);
2832 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2833 This triggers a max_bio_size message upon first attach or connect */
2834 blk_queue_max_hw_sectors(q
, DRBD_MAX_BIO_SIZE_SAFE
>> 8);
2836 device
->md_io
.page
= alloc_page(GFP_KERNEL
);
2837 if (!device
->md_io
.page
)
2838 goto out_no_io_page
;
2840 if (drbd_bm_init(device
))
2842 device
->read_requests
= RB_ROOT
;
2843 device
->write_requests
= RB_ROOT
;
2845 id
= idr_alloc(&drbd_devices
, device
, minor
, minor
+ 1, GFP_KERNEL
);
2848 err
= ERR_MINOR_OR_VOLUME_EXISTS
;
2849 goto out_no_minor_idr
;
2851 kref_get(&device
->kref
);
2853 id
= idr_alloc(&resource
->devices
, device
, vnr
, vnr
+ 1, GFP_KERNEL
);
2856 err
= ERR_MINOR_OR_VOLUME_EXISTS
;
2857 goto out_idr_remove_minor
;
2859 kref_get(&device
->kref
);
2861 INIT_LIST_HEAD(&device
->peer_devices
);
2862 INIT_LIST_HEAD(&device
->pending_bitmap_io
);
2863 for_each_connection(connection
, resource
) {
2864 peer_device
= kzalloc(sizeof(struct drbd_peer_device
), GFP_KERNEL
);
2866 goto out_idr_remove_from_resource
;
2867 peer_device
->connection
= connection
;
2868 peer_device
->device
= device
;
2870 list_add(&peer_device
->peer_devices
, &device
->peer_devices
);
2871 kref_get(&device
->kref
);
2873 id
= idr_alloc(&connection
->peer_devices
, peer_device
, vnr
, vnr
+ 1, GFP_KERNEL
);
2876 err
= ERR_INVALID_REQUEST
;
2877 goto out_idr_remove_from_resource
;
2879 kref_get(&connection
->kref
);
2880 INIT_WORK(&peer_device
->send_acks_work
, drbd_send_acks_wf
);
2883 if (init_submitter(device
)) {
2885 goto out_idr_remove_vol
;
2890 /* inherit the connection state */
2891 device
->state
.conn
= first_connection(resource
)->cstate
;
2892 if (device
->state
.conn
== C_WF_REPORT_PARAMS
) {
2893 for_each_peer_device(peer_device
, device
)
2894 drbd_connected(peer_device
);
2896 /* move to create_peer_device() */
2897 for_each_peer_device(peer_device
, device
)
2898 drbd_debugfs_peer_device_add(peer_device
);
2899 drbd_debugfs_device_add(device
);
2903 idr_remove(&connection
->peer_devices
, vnr
);
2904 out_idr_remove_from_resource
:
2905 for_each_connection(connection
, resource
) {
2906 peer_device
= idr_remove(&connection
->peer_devices
, vnr
);
2908 kref_put(&connection
->kref
, drbd_destroy_connection
);
2910 for_each_peer_device_safe(peer_device
, tmp_peer_device
, device
) {
2911 list_del(&peer_device
->peer_devices
);
2914 idr_remove(&resource
->devices
, vnr
);
2915 out_idr_remove_minor
:
2916 idr_remove(&drbd_devices
, minor
);
2919 drbd_bm_cleanup(device
);
2921 __free_page(device
->md_io
.page
);
2925 blk_cleanup_queue(q
);
2927 kref_put(&resource
->kref
, drbd_destroy_resource
);
2932 void drbd_delete_device(struct drbd_device
*device
)
2934 struct drbd_resource
*resource
= device
->resource
;
2935 struct drbd_connection
*connection
;
2936 struct drbd_peer_device
*peer_device
;
2938 /* move to free_peer_device() */
2939 for_each_peer_device(peer_device
, device
)
2940 drbd_debugfs_peer_device_cleanup(peer_device
);
2941 drbd_debugfs_device_cleanup(device
);
2942 for_each_connection(connection
, resource
) {
2943 idr_remove(&connection
->peer_devices
, device
->vnr
);
2944 kref_put(&device
->kref
, drbd_destroy_device
);
2946 idr_remove(&resource
->devices
, device
->vnr
);
2947 kref_put(&device
->kref
, drbd_destroy_device
);
2948 idr_remove(&drbd_devices
, device_to_minor(device
));
2949 kref_put(&device
->kref
, drbd_destroy_device
);
2950 del_gendisk(device
->vdisk
);
2952 kref_put(&device
->kref
, drbd_destroy_device
);
2955 static int __init
drbd_init(void)
2959 if (drbd_minor_count
< DRBD_MINOR_COUNT_MIN
|| drbd_minor_count
> DRBD_MINOR_COUNT_MAX
) {
2960 pr_err("invalid minor_count (%d)\n", drbd_minor_count
);
2964 drbd_minor_count
= DRBD_MINOR_COUNT_DEF
;
2968 err
= register_blkdev(DRBD_MAJOR
, "drbd");
2970 pr_err("unable to register block device major %d\n",
2976 * allocate all necessary structs
2978 init_waitqueue_head(&drbd_pp_wait
);
2980 drbd_proc
= NULL
; /* play safe for drbd_cleanup */
2981 idr_init(&drbd_devices
);
2983 mutex_init(&resources_mutex
);
2984 INIT_LIST_HEAD(&drbd_resources
);
2986 err
= drbd_genl_register();
2988 pr_err("unable to register generic netlink family\n");
2992 err
= drbd_create_mempools();
2997 drbd_proc
= proc_create_single("drbd", S_IFREG
| 0444 , NULL
, drbd_seq_show
);
2999 pr_err("unable to register proc file\n");
3003 retry
.wq
= create_singlethread_workqueue("drbd-reissue");
3005 pr_err("unable to create retry workqueue\n");
3008 INIT_WORK(&retry
.worker
, do_retry
);
3009 spin_lock_init(&retry
.lock
);
3010 INIT_LIST_HEAD(&retry
.writes
);
3012 drbd_debugfs_init();
3014 pr_info("initialized. "
3015 "Version: " REL_VERSION
" (api:%d/proto:%d-%d)\n",
3016 API_VERSION
, PRO_VERSION_MIN
, PRO_VERSION_MAX
);
3017 pr_info("%s\n", drbd_buildtag());
3018 pr_info("registered as block device major %d\n", DRBD_MAJOR
);
3019 return 0; /* Success! */
3024 pr_err("ran out of memory\n");
3026 pr_err("initialization failure\n");
3030 static void drbd_free_one_sock(struct drbd_socket
*ds
)
3033 mutex_lock(&ds
->mutex
);
3036 mutex_unlock(&ds
->mutex
);
3038 /* so debugfs does not need to mutex_lock() */
3040 kernel_sock_shutdown(s
, SHUT_RDWR
);
3045 void drbd_free_sock(struct drbd_connection
*connection
)
3047 if (connection
->data
.socket
)
3048 drbd_free_one_sock(&connection
->data
);
3049 if (connection
->meta
.socket
)
3050 drbd_free_one_sock(&connection
->meta
);
3053 /* meta data management */
3055 void conn_md_sync(struct drbd_connection
*connection
)
3057 struct drbd_peer_device
*peer_device
;
3061 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
3062 struct drbd_device
*device
= peer_device
->device
;
3064 kref_get(&device
->kref
);
3066 drbd_md_sync(device
);
3067 kref_put(&device
->kref
, drbd_destroy_device
);
3073 /* aligned 4kByte */
3074 struct meta_data_on_disk
{
3075 u64 la_size_sect
; /* last agreed size. */
3076 u64 uuid
[UI_SIZE
]; /* UUIDs. */
3079 u32 flags
; /* MDF */
3082 u32 al_offset
; /* offset to this block */
3083 u32 al_nr_extents
; /* important for restoring the AL (userspace) */
3084 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3085 u32 bm_offset
; /* offset to the bitmap, from here */
3086 u32 bm_bytes_per_bit
; /* BM_BLOCK_SIZE */
3087 u32 la_peer_max_bio_size
; /* last peer max_bio_size */
3089 /* see al_tr_number_to_on_disk_sector() */
3091 u32 al_stripe_size_4k
;
3093 u8 reserved_u8
[4096 - (7*8 + 10*4)];
3098 void drbd_md_write(struct drbd_device
*device
, void *b
)
3100 struct meta_data_on_disk
*buffer
= b
;
3104 memset(buffer
, 0, sizeof(*buffer
));
3106 buffer
->la_size_sect
= cpu_to_be64(drbd_get_capacity(device
->this_bdev
));
3107 for (i
= UI_CURRENT
; i
< UI_SIZE
; i
++)
3108 buffer
->uuid
[i
] = cpu_to_be64(device
->ldev
->md
.uuid
[i
]);
3109 buffer
->flags
= cpu_to_be32(device
->ldev
->md
.flags
);
3110 buffer
->magic
= cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN
);
3112 buffer
->md_size_sect
= cpu_to_be32(device
->ldev
->md
.md_size_sect
);
3113 buffer
->al_offset
= cpu_to_be32(device
->ldev
->md
.al_offset
);
3114 buffer
->al_nr_extents
= cpu_to_be32(device
->act_log
->nr_elements
);
3115 buffer
->bm_bytes_per_bit
= cpu_to_be32(BM_BLOCK_SIZE
);
3116 buffer
->device_uuid
= cpu_to_be64(device
->ldev
->md
.device_uuid
);
3118 buffer
->bm_offset
= cpu_to_be32(device
->ldev
->md
.bm_offset
);
3119 buffer
->la_peer_max_bio_size
= cpu_to_be32(device
->peer_max_bio_size
);
3121 buffer
->al_stripes
= cpu_to_be32(device
->ldev
->md
.al_stripes
);
3122 buffer
->al_stripe_size_4k
= cpu_to_be32(device
->ldev
->md
.al_stripe_size_4k
);
3124 D_ASSERT(device
, drbd_md_ss(device
->ldev
) == device
->ldev
->md
.md_offset
);
3125 sector
= device
->ldev
->md
.md_offset
;
3127 if (drbd_md_sync_page_io(device
, device
->ldev
, sector
, REQ_OP_WRITE
)) {
3128 /* this was a try anyways ... */
3129 drbd_err(device
, "meta data update failed!\n");
3130 drbd_chk_io_error(device
, 1, DRBD_META_IO_ERROR
);
3135 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3136 * @device: DRBD device.
3138 void drbd_md_sync(struct drbd_device
*device
)
3140 struct meta_data_on_disk
*buffer
;
3142 /* Don't accidentally change the DRBD meta data layout. */
3143 BUILD_BUG_ON(UI_SIZE
!= 4);
3144 BUILD_BUG_ON(sizeof(struct meta_data_on_disk
) != 4096);
3146 del_timer(&device
->md_sync_timer
);
3147 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3148 if (!test_and_clear_bit(MD_DIRTY
, &device
->flags
))
3151 /* We use here D_FAILED and not D_ATTACHING because we try to write
3152 * metadata even if we detach due to a disk failure! */
3153 if (!get_ldev_if_state(device
, D_FAILED
))
3156 buffer
= drbd_md_get_buffer(device
, __func__
);
3160 drbd_md_write(device
, buffer
);
3162 /* Update device->ldev->md.la_size_sect,
3163 * since we updated it on metadata. */
3164 device
->ldev
->md
.la_size_sect
= drbd_get_capacity(device
->this_bdev
);
3166 drbd_md_put_buffer(device
);
3171 static int check_activity_log_stripe_size(struct drbd_device
*device
,
3172 struct meta_data_on_disk
*on_disk
,
3173 struct drbd_md
*in_core
)
3175 u32 al_stripes
= be32_to_cpu(on_disk
->al_stripes
);
3176 u32 al_stripe_size_4k
= be32_to_cpu(on_disk
->al_stripe_size_4k
);
3179 /* both not set: default to old fixed size activity log */
3180 if (al_stripes
== 0 && al_stripe_size_4k
== 0) {
3182 al_stripe_size_4k
= MD_32kB_SECT
/8;
3185 /* some paranoia plausibility checks */
3187 /* we need both values to be set */
3188 if (al_stripes
== 0 || al_stripe_size_4k
== 0)
3191 al_size_4k
= (u64
)al_stripes
* al_stripe_size_4k
;
3193 /* Upper limit of activity log area, to avoid potential overflow
3194 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3195 * than 72 * 4k blocks total only increases the amount of history,
3196 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3197 if (al_size_4k
> (16 * 1024 * 1024/4))
3200 /* Lower limit: we need at least 8 transaction slots (32kB)
3201 * to not break existing setups */
3202 if (al_size_4k
< MD_32kB_SECT
/8)
3205 in_core
->al_stripe_size_4k
= al_stripe_size_4k
;
3206 in_core
->al_stripes
= al_stripes
;
3207 in_core
->al_size_4k
= al_size_4k
;
3211 drbd_err(device
, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3212 al_stripes
, al_stripe_size_4k
);
3216 static int check_offsets_and_sizes(struct drbd_device
*device
, struct drbd_backing_dev
*bdev
)
3218 sector_t capacity
= drbd_get_capacity(bdev
->md_bdev
);
3219 struct drbd_md
*in_core
= &bdev
->md
;
3220 s32 on_disk_al_sect
;
3221 s32 on_disk_bm_sect
;
3223 /* The on-disk size of the activity log, calculated from offsets, and
3224 * the size of the activity log calculated from the stripe settings,
3226 * Though we could relax this a bit: it is ok, if the striped activity log
3227 * fits in the available on-disk activity log size.
3228 * Right now, that would break how resize is implemented.
3229 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3230 * of possible unused padding space in the on disk layout. */
3231 if (in_core
->al_offset
< 0) {
3232 if (in_core
->bm_offset
> in_core
->al_offset
)
3234 on_disk_al_sect
= -in_core
->al_offset
;
3235 on_disk_bm_sect
= in_core
->al_offset
- in_core
->bm_offset
;
3237 if (in_core
->al_offset
!= MD_4kB_SECT
)
3239 if (in_core
->bm_offset
< in_core
->al_offset
+ in_core
->al_size_4k
* MD_4kB_SECT
)
3242 on_disk_al_sect
= in_core
->bm_offset
- MD_4kB_SECT
;
3243 on_disk_bm_sect
= in_core
->md_size_sect
- in_core
->bm_offset
;
3246 /* old fixed size meta data is exactly that: fixed. */
3247 if (in_core
->meta_dev_idx
>= 0) {
3248 if (in_core
->md_size_sect
!= MD_128MB_SECT
3249 || in_core
->al_offset
!= MD_4kB_SECT
3250 || in_core
->bm_offset
!= MD_4kB_SECT
+ MD_32kB_SECT
3251 || in_core
->al_stripes
!= 1
3252 || in_core
->al_stripe_size_4k
!= MD_32kB_SECT
/8)
3256 if (capacity
< in_core
->md_size_sect
)
3258 if (capacity
- in_core
->md_size_sect
< drbd_md_first_sector(bdev
))
3261 /* should be aligned, and at least 32k */
3262 if ((on_disk_al_sect
& 7) || (on_disk_al_sect
< MD_32kB_SECT
))
3265 /* should fit (for now: exactly) into the available on-disk space;
3266 * overflow prevention is in check_activity_log_stripe_size() above. */
3267 if (on_disk_al_sect
!= in_core
->al_size_4k
* MD_4kB_SECT
)
3270 /* again, should be aligned */
3271 if (in_core
->bm_offset
& 7)
3274 /* FIXME check for device grow with flex external meta data? */
3276 /* can the available bitmap space cover the last agreed device size? */
3277 if (on_disk_bm_sect
< (in_core
->la_size_sect
+7)/MD_4kB_SECT
/8/512)
3283 drbd_err(device
, "meta data offsets don't make sense: idx=%d "
3284 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3285 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3286 in_core
->meta_dev_idx
,
3287 in_core
->al_stripes
, in_core
->al_stripe_size_4k
,
3288 in_core
->al_offset
, in_core
->bm_offset
, in_core
->md_size_sect
,
3289 (unsigned long long)in_core
->la_size_sect
,
3290 (unsigned long long)capacity
);
3297 * drbd_md_read() - Reads in the meta data super block
3298 * @device: DRBD device.
3299 * @bdev: Device from which the meta data should be read in.
3301 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3302 * something goes wrong.
3304 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3305 * even before @bdev is assigned to @device->ldev.
3307 int drbd_md_read(struct drbd_device
*device
, struct drbd_backing_dev
*bdev
)
3309 struct meta_data_on_disk
*buffer
;
3311 int i
, rv
= NO_ERROR
;
3313 if (device
->state
.disk
!= D_DISKLESS
)
3314 return ERR_DISK_CONFIGURED
;
3316 buffer
= drbd_md_get_buffer(device
, __func__
);
3320 /* First, figure out where our meta data superblock is located,
3322 bdev
->md
.meta_dev_idx
= bdev
->disk_conf
->meta_dev_idx
;
3323 bdev
->md
.md_offset
= drbd_md_ss(bdev
);
3324 /* Even for (flexible or indexed) external meta data,
3325 * initially restrict us to the 4k superblock for now.
3326 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3327 bdev
->md
.md_size_sect
= 8;
3329 if (drbd_md_sync_page_io(device
, bdev
, bdev
->md
.md_offset
,
3331 /* NOTE: can't do normal error processing here as this is
3332 called BEFORE disk is attached */
3333 drbd_err(device
, "Error while reading metadata.\n");
3334 rv
= ERR_IO_MD_DISK
;
3338 magic
= be32_to_cpu(buffer
->magic
);
3339 flags
= be32_to_cpu(buffer
->flags
);
3340 if (magic
== DRBD_MD_MAGIC_84_UNCLEAN
||
3341 (magic
== DRBD_MD_MAGIC_08
&& !(flags
& MDF_AL_CLEAN
))) {
3342 /* btw: that's Activity Log clean, not "all" clean. */
3343 drbd_err(device
, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3344 rv
= ERR_MD_UNCLEAN
;
3348 rv
= ERR_MD_INVALID
;
3349 if (magic
!= DRBD_MD_MAGIC_08
) {
3350 if (magic
== DRBD_MD_MAGIC_07
)
3351 drbd_err(device
, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3353 drbd_err(device
, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3357 if (be32_to_cpu(buffer
->bm_bytes_per_bit
) != BM_BLOCK_SIZE
) {
3358 drbd_err(device
, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3359 be32_to_cpu(buffer
->bm_bytes_per_bit
), BM_BLOCK_SIZE
);
3364 /* convert to in_core endian */
3365 bdev
->md
.la_size_sect
= be64_to_cpu(buffer
->la_size_sect
);
3366 for (i
= UI_CURRENT
; i
< UI_SIZE
; i
++)
3367 bdev
->md
.uuid
[i
] = be64_to_cpu(buffer
->uuid
[i
]);
3368 bdev
->md
.flags
= be32_to_cpu(buffer
->flags
);
3369 bdev
->md
.device_uuid
= be64_to_cpu(buffer
->device_uuid
);
3371 bdev
->md
.md_size_sect
= be32_to_cpu(buffer
->md_size_sect
);
3372 bdev
->md
.al_offset
= be32_to_cpu(buffer
->al_offset
);
3373 bdev
->md
.bm_offset
= be32_to_cpu(buffer
->bm_offset
);
3375 if (check_activity_log_stripe_size(device
, buffer
, &bdev
->md
))
3377 if (check_offsets_and_sizes(device
, bdev
))
3380 if (be32_to_cpu(buffer
->bm_offset
) != bdev
->md
.bm_offset
) {
3381 drbd_err(device
, "unexpected bm_offset: %d (expected %d)\n",
3382 be32_to_cpu(buffer
->bm_offset
), bdev
->md
.bm_offset
);
3385 if (be32_to_cpu(buffer
->md_size_sect
) != bdev
->md
.md_size_sect
) {
3386 drbd_err(device
, "unexpected md_size: %u (expected %u)\n",
3387 be32_to_cpu(buffer
->md_size_sect
), bdev
->md
.md_size_sect
);
3393 spin_lock_irq(&device
->resource
->req_lock
);
3394 if (device
->state
.conn
< C_CONNECTED
) {
3396 peer
= be32_to_cpu(buffer
->la_peer_max_bio_size
);
3397 peer
= max(peer
, DRBD_MAX_BIO_SIZE_SAFE
);
3398 device
->peer_max_bio_size
= peer
;
3400 spin_unlock_irq(&device
->resource
->req_lock
);
3403 drbd_md_put_buffer(device
);
3409 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3410 * @device: DRBD device.
3412 * Call this function if you change anything that should be written to
3413 * the meta-data super block. This function sets MD_DIRTY, and starts a
3414 * timer that ensures that within five seconds you have to call drbd_md_sync().
3417 void drbd_md_mark_dirty_(struct drbd_device
*device
, unsigned int line
, const char *func
)
3419 if (!test_and_set_bit(MD_DIRTY
, &device
->flags
)) {
3420 mod_timer(&device
->md_sync_timer
, jiffies
+ HZ
);
3421 device
->last_md_mark_dirty
.line
= line
;
3422 device
->last_md_mark_dirty
.func
= func
;
3426 void drbd_md_mark_dirty(struct drbd_device
*device
)
3428 if (!test_and_set_bit(MD_DIRTY
, &device
->flags
))
3429 mod_timer(&device
->md_sync_timer
, jiffies
+ 5*HZ
);
3433 void drbd_uuid_move_history(struct drbd_device
*device
) __must_hold(local
)
3437 for (i
= UI_HISTORY_START
; i
< UI_HISTORY_END
; i
++)
3438 device
->ldev
->md
.uuid
[i
+1] = device
->ldev
->md
.uuid
[i
];
3441 void __drbd_uuid_set(struct drbd_device
*device
, int idx
, u64 val
) __must_hold(local
)
3443 if (idx
== UI_CURRENT
) {
3444 if (device
->state
.role
== R_PRIMARY
)
3449 drbd_set_ed_uuid(device
, val
);
3452 device
->ldev
->md
.uuid
[idx
] = val
;
3453 drbd_md_mark_dirty(device
);
3456 void _drbd_uuid_set(struct drbd_device
*device
, int idx
, u64 val
) __must_hold(local
)
3458 unsigned long flags
;
3459 spin_lock_irqsave(&device
->ldev
->md
.uuid_lock
, flags
);
3460 __drbd_uuid_set(device
, idx
, val
);
3461 spin_unlock_irqrestore(&device
->ldev
->md
.uuid_lock
, flags
);
3464 void drbd_uuid_set(struct drbd_device
*device
, int idx
, u64 val
) __must_hold(local
)
3466 unsigned long flags
;
3467 spin_lock_irqsave(&device
->ldev
->md
.uuid_lock
, flags
);
3468 if (device
->ldev
->md
.uuid
[idx
]) {
3469 drbd_uuid_move_history(device
);
3470 device
->ldev
->md
.uuid
[UI_HISTORY_START
] = device
->ldev
->md
.uuid
[idx
];
3472 __drbd_uuid_set(device
, idx
, val
);
3473 spin_unlock_irqrestore(&device
->ldev
->md
.uuid_lock
, flags
);
3477 * drbd_uuid_new_current() - Creates a new current UUID
3478 * @device: DRBD device.
3480 * Creates a new current UUID, and rotates the old current UUID into
3481 * the bitmap slot. Causes an incremental resync upon next connect.
3483 void drbd_uuid_new_current(struct drbd_device
*device
) __must_hold(local
)
3486 unsigned long long bm_uuid
;
3488 get_random_bytes(&val
, sizeof(u64
));
3490 spin_lock_irq(&device
->ldev
->md
.uuid_lock
);
3491 bm_uuid
= device
->ldev
->md
.uuid
[UI_BITMAP
];
3494 drbd_warn(device
, "bm UUID was already set: %llX\n", bm_uuid
);
3496 device
->ldev
->md
.uuid
[UI_BITMAP
] = device
->ldev
->md
.uuid
[UI_CURRENT
];
3497 __drbd_uuid_set(device
, UI_CURRENT
, val
);
3498 spin_unlock_irq(&device
->ldev
->md
.uuid_lock
);
3500 drbd_print_uuids(device
, "new current UUID");
3501 /* get it to stable storage _now_ */
3502 drbd_md_sync(device
);
3505 void drbd_uuid_set_bm(struct drbd_device
*device
, u64 val
) __must_hold(local
)
3507 unsigned long flags
;
3508 if (device
->ldev
->md
.uuid
[UI_BITMAP
] == 0 && val
== 0)
3511 spin_lock_irqsave(&device
->ldev
->md
.uuid_lock
, flags
);
3513 drbd_uuid_move_history(device
);
3514 device
->ldev
->md
.uuid
[UI_HISTORY_START
] = device
->ldev
->md
.uuid
[UI_BITMAP
];
3515 device
->ldev
->md
.uuid
[UI_BITMAP
] = 0;
3517 unsigned long long bm_uuid
= device
->ldev
->md
.uuid
[UI_BITMAP
];
3519 drbd_warn(device
, "bm UUID was already set: %llX\n", bm_uuid
);
3521 device
->ldev
->md
.uuid
[UI_BITMAP
] = val
& ~((u64
)1);
3523 spin_unlock_irqrestore(&device
->ldev
->md
.uuid_lock
, flags
);
3525 drbd_md_mark_dirty(device
);
3529 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3530 * @device: DRBD device.
3532 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3534 int drbd_bmio_set_n_write(struct drbd_device
*device
) __must_hold(local
)
3538 drbd_md_set_flag(device
, MDF_FULL_SYNC
);
3539 drbd_md_sync(device
);
3540 drbd_bm_set_all(device
);
3542 rv
= drbd_bm_write(device
);
3545 drbd_md_clear_flag(device
, MDF_FULL_SYNC
);
3546 drbd_md_sync(device
);
3553 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3554 * @device: DRBD device.
3556 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3558 int drbd_bmio_clear_n_write(struct drbd_device
*device
) __must_hold(local
)
3560 drbd_resume_al(device
);
3561 drbd_bm_clear_all(device
);
3562 return drbd_bm_write(device
);
3565 static int w_bitmap_io(struct drbd_work
*w
, int unused
)
3567 struct drbd_device
*device
=
3568 container_of(w
, struct drbd_device
, bm_io_work
.w
);
3569 struct bm_io_work
*work
= &device
->bm_io_work
;
3572 if (work
->flags
!= BM_LOCKED_CHANGE_ALLOWED
) {
3573 int cnt
= atomic_read(&device
->ap_bio_cnt
);
3575 drbd_err(device
, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3579 if (get_ldev(device
)) {
3580 drbd_bm_lock(device
, work
->why
, work
->flags
);
3581 rv
= work
->io_fn(device
);
3582 drbd_bm_unlock(device
);
3586 clear_bit_unlock(BITMAP_IO
, &device
->flags
);
3587 wake_up(&device
->misc_wait
);
3590 work
->done(device
, rv
);
3592 clear_bit(BITMAP_IO_QUEUED
, &device
->flags
);
3600 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3601 * @device: DRBD device.
3602 * @io_fn: IO callback to be called when bitmap IO is possible
3603 * @done: callback to be called after the bitmap IO was performed
3604 * @why: Descriptive text of the reason for doing the IO
3606 * While IO on the bitmap happens we freeze application IO thus we ensure
3607 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3608 * called from worker context. It MUST NOT be used while a previous such
3609 * work is still pending!
3611 * Its worker function encloses the call of io_fn() by get_ldev() and
3614 void drbd_queue_bitmap_io(struct drbd_device
*device
,
3615 int (*io_fn
)(struct drbd_device
*),
3616 void (*done
)(struct drbd_device
*, int),
3617 char *why
, enum bm_flag flags
)
3619 D_ASSERT(device
, current
== first_peer_device(device
)->connection
->worker
.task
);
3621 D_ASSERT(device
, !test_bit(BITMAP_IO_QUEUED
, &device
->flags
));
3622 D_ASSERT(device
, !test_bit(BITMAP_IO
, &device
->flags
));
3623 D_ASSERT(device
, list_empty(&device
->bm_io_work
.w
.list
));
3624 if (device
->bm_io_work
.why
)
3625 drbd_err(device
, "FIXME going to queue '%s' but '%s' still pending?\n",
3626 why
, device
->bm_io_work
.why
);
3628 device
->bm_io_work
.io_fn
= io_fn
;
3629 device
->bm_io_work
.done
= done
;
3630 device
->bm_io_work
.why
= why
;
3631 device
->bm_io_work
.flags
= flags
;
3633 spin_lock_irq(&device
->resource
->req_lock
);
3634 set_bit(BITMAP_IO
, &device
->flags
);
3635 /* don't wait for pending application IO if the caller indicates that
3636 * application IO does not conflict anyways. */
3637 if (flags
== BM_LOCKED_CHANGE_ALLOWED
|| atomic_read(&device
->ap_bio_cnt
) == 0) {
3638 if (!test_and_set_bit(BITMAP_IO_QUEUED
, &device
->flags
))
3639 drbd_queue_work(&first_peer_device(device
)->connection
->sender_work
,
3640 &device
->bm_io_work
.w
);
3642 spin_unlock_irq(&device
->resource
->req_lock
);
3646 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3647 * @device: DRBD device.
3648 * @io_fn: IO callback to be called when bitmap IO is possible
3649 * @why: Descriptive text of the reason for doing the IO
3651 * freezes application IO while that the actual IO operations runs. This
3652 * functions MAY NOT be called from worker context.
3654 int drbd_bitmap_io(struct drbd_device
*device
, int (*io_fn
)(struct drbd_device
*),
3655 char *why
, enum bm_flag flags
)
3657 /* Only suspend io, if some operation is supposed to be locked out */
3658 const bool do_suspend_io
= flags
& (BM_DONT_CLEAR
|BM_DONT_SET
|BM_DONT_TEST
);
3661 D_ASSERT(device
, current
!= first_peer_device(device
)->connection
->worker
.task
);
3664 drbd_suspend_io(device
);
3666 drbd_bm_lock(device
, why
, flags
);
3668 drbd_bm_unlock(device
);
3671 drbd_resume_io(device
);
3676 void drbd_md_set_flag(struct drbd_device
*device
, int flag
) __must_hold(local
)
3678 if ((device
->ldev
->md
.flags
& flag
) != flag
) {
3679 drbd_md_mark_dirty(device
);
3680 device
->ldev
->md
.flags
|= flag
;
3684 void drbd_md_clear_flag(struct drbd_device
*device
, int flag
) __must_hold(local
)
3686 if ((device
->ldev
->md
.flags
& flag
) != 0) {
3687 drbd_md_mark_dirty(device
);
3688 device
->ldev
->md
.flags
&= ~flag
;
3691 int drbd_md_test_flag(struct drbd_backing_dev
*bdev
, int flag
)
3693 return (bdev
->md
.flags
& flag
) != 0;
3696 static void md_sync_timer_fn(struct timer_list
*t
)
3698 struct drbd_device
*device
= from_timer(device
, t
, md_sync_timer
);
3699 drbd_device_post_work(device
, MD_SYNC
);
3702 const char *cmdname(enum drbd_packet cmd
)
3704 /* THINK may need to become several global tables
3705 * when we want to support more than
3706 * one PRO_VERSION */
3707 static const char *cmdnames
[] = {
3709 [P_WSAME
] = "WriteSame",
3711 [P_DATA_REPLY
] = "DataReply",
3712 [P_RS_DATA_REPLY
] = "RSDataReply",
3713 [P_BARRIER
] = "Barrier",
3714 [P_BITMAP
] = "ReportBitMap",
3715 [P_BECOME_SYNC_TARGET
] = "BecomeSyncTarget",
3716 [P_BECOME_SYNC_SOURCE
] = "BecomeSyncSource",
3717 [P_UNPLUG_REMOTE
] = "UnplugRemote",
3718 [P_DATA_REQUEST
] = "DataRequest",
3719 [P_RS_DATA_REQUEST
] = "RSDataRequest",
3720 [P_SYNC_PARAM
] = "SyncParam",
3721 [P_SYNC_PARAM89
] = "SyncParam89",
3722 [P_PROTOCOL
] = "ReportProtocol",
3723 [P_UUIDS
] = "ReportUUIDs",
3724 [P_SIZES
] = "ReportSizes",
3725 [P_STATE
] = "ReportState",
3726 [P_SYNC_UUID
] = "ReportSyncUUID",
3727 [P_AUTH_CHALLENGE
] = "AuthChallenge",
3728 [P_AUTH_RESPONSE
] = "AuthResponse",
3730 [P_PING_ACK
] = "PingAck",
3731 [P_RECV_ACK
] = "RecvAck",
3732 [P_WRITE_ACK
] = "WriteAck",
3733 [P_RS_WRITE_ACK
] = "RSWriteAck",
3734 [P_SUPERSEDED
] = "Superseded",
3735 [P_NEG_ACK
] = "NegAck",
3736 [P_NEG_DREPLY
] = "NegDReply",
3737 [P_NEG_RS_DREPLY
] = "NegRSDReply",
3738 [P_BARRIER_ACK
] = "BarrierAck",
3739 [P_STATE_CHG_REQ
] = "StateChgRequest",
3740 [P_STATE_CHG_REPLY
] = "StateChgReply",
3741 [P_OV_REQUEST
] = "OVRequest",
3742 [P_OV_REPLY
] = "OVReply",
3743 [P_OV_RESULT
] = "OVResult",
3744 [P_CSUM_RS_REQUEST
] = "CsumRSRequest",
3745 [P_RS_IS_IN_SYNC
] = "CsumRSIsInSync",
3746 [P_COMPRESSED_BITMAP
] = "CBitmap",
3747 [P_DELAY_PROBE
] = "DelayProbe",
3748 [P_OUT_OF_SYNC
] = "OutOfSync",
3749 [P_RETRY_WRITE
] = "RetryWrite",
3750 [P_RS_CANCEL
] = "RSCancel",
3751 [P_CONN_ST_CHG_REQ
] = "conn_st_chg_req",
3752 [P_CONN_ST_CHG_REPLY
] = "conn_st_chg_reply",
3753 [P_RETRY_WRITE
] = "retry_write",
3754 [P_PROTOCOL_UPDATE
] = "protocol_update",
3755 [P_RS_THIN_REQ
] = "rs_thin_req",
3756 [P_RS_DEALLOCATED
] = "rs_deallocated",
3758 /* enum drbd_packet, but not commands - obsoleted flags:
3764 /* too big for the array: 0xfffX */
3765 if (cmd
== P_INITIAL_META
)
3766 return "InitialMeta";
3767 if (cmd
== P_INITIAL_DATA
)
3768 return "InitialData";
3769 if (cmd
== P_CONNECTION_FEATURES
)
3770 return "ConnectionFeatures";
3771 if (cmd
>= ARRAY_SIZE(cmdnames
))
3773 return cmdnames
[cmd
];
3777 * drbd_wait_misc - wait for a request to make progress
3778 * @device: device associated with the request
3779 * @i: the struct drbd_interval embedded in struct drbd_request or
3780 * struct drbd_peer_request
3782 int drbd_wait_misc(struct drbd_device
*device
, struct drbd_interval
*i
)
3784 struct net_conf
*nc
;
3789 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
3794 timeout
= nc
->ko_count
? nc
->timeout
* HZ
/ 10 * nc
->ko_count
: MAX_SCHEDULE_TIMEOUT
;
3797 /* Indicate to wake up device->misc_wait on progress. */
3799 prepare_to_wait(&device
->misc_wait
, &wait
, TASK_INTERRUPTIBLE
);
3800 spin_unlock_irq(&device
->resource
->req_lock
);
3801 timeout
= schedule_timeout(timeout
);
3802 finish_wait(&device
->misc_wait
, &wait
);
3803 spin_lock_irq(&device
->resource
->req_lock
);
3804 if (!timeout
|| device
->state
.conn
< C_CONNECTED
)
3806 if (signal_pending(current
))
3807 return -ERESTARTSYS
;
3811 void lock_all_resources(void)
3813 struct drbd_resource
*resource
;
3814 int __maybe_unused i
= 0;
3816 mutex_lock(&resources_mutex
);
3817 local_irq_disable();
3818 for_each_resource(resource
, &drbd_resources
)
3819 spin_lock_nested(&resource
->req_lock
, i
++);
3822 void unlock_all_resources(void)
3824 struct drbd_resource
*resource
;
3826 for_each_resource(resource
, &drbd_resources
)
3827 spin_unlock(&resource
->req_lock
);
3829 mutex_unlock(&resources_mutex
);
3832 #ifdef CONFIG_DRBD_FAULT_INJECTION
3833 /* Fault insertion support including random number generator shamelessly
3834 * stolen from kernel/rcutorture.c */
3835 struct fault_random_state
{
3836 unsigned long state
;
3837 unsigned long count
;
3840 #define FAULT_RANDOM_MULT 39916801 /* prime */
3841 #define FAULT_RANDOM_ADD 479001701 /* prime */
3842 #define FAULT_RANDOM_REFRESH 10000
3845 * Crude but fast random-number generator. Uses a linear congruential
3846 * generator, with occasional help from get_random_bytes().
3848 static unsigned long
3849 _drbd_fault_random(struct fault_random_state
*rsp
)
3853 if (!rsp
->count
--) {
3854 get_random_bytes(&refresh
, sizeof(refresh
));
3855 rsp
->state
+= refresh
;
3856 rsp
->count
= FAULT_RANDOM_REFRESH
;
3858 rsp
->state
= rsp
->state
* FAULT_RANDOM_MULT
+ FAULT_RANDOM_ADD
;
3859 return swahw32(rsp
->state
);
3863 _drbd_fault_str(unsigned int type
) {
3864 static char *_faults
[] = {
3865 [DRBD_FAULT_MD_WR
] = "Meta-data write",
3866 [DRBD_FAULT_MD_RD
] = "Meta-data read",
3867 [DRBD_FAULT_RS_WR
] = "Resync write",
3868 [DRBD_FAULT_RS_RD
] = "Resync read",
3869 [DRBD_FAULT_DT_WR
] = "Data write",
3870 [DRBD_FAULT_DT_RD
] = "Data read",
3871 [DRBD_FAULT_DT_RA
] = "Data read ahead",
3872 [DRBD_FAULT_BM_ALLOC
] = "BM allocation",
3873 [DRBD_FAULT_AL_EE
] = "EE allocation",
3874 [DRBD_FAULT_RECEIVE
] = "receive data corruption",
3877 return (type
< DRBD_FAULT_MAX
) ? _faults
[type
] : "**Unknown**";
3881 _drbd_insert_fault(struct drbd_device
*device
, unsigned int type
)
3883 static struct fault_random_state rrs
= {0, 0};
3885 unsigned int ret
= (
3886 (drbd_fault_devs
== 0 ||
3887 ((1 << device_to_minor(device
)) & drbd_fault_devs
) != 0) &&
3888 (((_drbd_fault_random(&rrs
) % 100) + 1) <= drbd_fault_rate
));
3893 if (__ratelimit(&drbd_ratelimit_state
))
3894 drbd_warn(device
, "***Simulating %s failure\n",
3895 _drbd_fault_str(type
));
3902 const char *drbd_buildtag(void)
3904 /* DRBD built from external sources has here a reference to the
3905 git hash of the source code. */
3907 static char buildtag
[38] = "\0uilt-in";
3909 if (buildtag
[0] == 0) {
3911 sprintf(buildtag
, "srcversion: %-24s", THIS_MODULE
->srcversion
);
3920 module_init(drbd_init
)
3921 module_exit(drbd_cleanup
)
3923 EXPORT_SYMBOL(drbd_conn_str
);
3924 EXPORT_SYMBOL(drbd_role_str
);
3925 EXPORT_SYMBOL(drbd_disk_str
);
3926 EXPORT_SYMBOL(drbd_set_st_err_str
);