1 // SPDX-License-Identifier: GPL-2.0-only
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 #include <linux/unistd.h>
41 #include <linux/vmalloc.h>
42 #include <linux/sched/signal.h>
44 #include <linux/drbd_limits.h>
46 #include "drbd_protocol.h"
47 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
49 #include "drbd_debugfs.h"
51 static DEFINE_MUTEX(drbd_main_mutex
);
52 static int drbd_open(struct gendisk
*disk
, blk_mode_t mode
);
53 static void drbd_release(struct gendisk
*gd
);
54 static void md_sync_timer_fn(struct timer_list
*t
);
55 static int w_bitmap_io(struct drbd_work
*w
, int unused
);
57 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
58 "Lars Ellenberg <lars@linbit.com>");
59 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION
);
60 MODULE_VERSION(REL_VERSION
);
61 MODULE_LICENSE("GPL");
62 MODULE_PARM_DESC(minor_count
, "Approximate number of drbd devices ("
63 __stringify(DRBD_MINOR_COUNT_MIN
) "-" __stringify(DRBD_MINOR_COUNT_MAX
) ")");
64 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR
);
66 #include <linux/moduleparam.h>
67 /* thanks to these macros, if compiled into the kernel (not-module),
68 * these become boot parameters (e.g., drbd.minor_count) */
70 #ifdef CONFIG_DRBD_FAULT_INJECTION
71 int drbd_enable_faults
;
73 static int drbd_fault_count
;
74 static int drbd_fault_devs
;
75 /* bitmap of enabled faults */
76 module_param_named(enable_faults
, drbd_enable_faults
, int, 0664);
77 /* fault rate % value - applies to all enabled faults */
78 module_param_named(fault_rate
, drbd_fault_rate
, int, 0664);
79 /* count of faults inserted */
80 module_param_named(fault_count
, drbd_fault_count
, int, 0664);
81 /* bitmap of devices to insert faults on */
82 module_param_named(fault_devs
, drbd_fault_devs
, int, 0644);
85 /* module parameters we can keep static */
86 static bool drbd_allow_oos
; /* allow_open_on_secondary */
87 static bool drbd_disable_sendpage
;
88 MODULE_PARM_DESC(allow_oos
, "DONT USE!");
89 module_param_named(allow_oos
, drbd_allow_oos
, bool, 0);
90 module_param_named(disable_sendpage
, drbd_disable_sendpage
, bool, 0644);
92 /* module parameters we share */
93 int drbd_proc_details
; /* Detail level in proc drbd*/
94 module_param_named(proc_details
, drbd_proc_details
, int, 0644);
95 /* module parameters shared with defaults */
96 unsigned int drbd_minor_count
= DRBD_MINOR_COUNT_DEF
;
97 /* Module parameter for setting the user mode helper program
98 * to run. Default is /sbin/drbdadm */
99 char drbd_usermode_helper
[80] = "/sbin/drbdadm";
100 module_param_named(minor_count
, drbd_minor_count
, uint
, 0444);
101 module_param_string(usermode_helper
, drbd_usermode_helper
, sizeof(drbd_usermode_helper
), 0644);
103 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
104 * as member "struct gendisk *vdisk;"
106 struct idr drbd_devices
;
107 struct list_head drbd_resources
;
108 struct mutex resources_mutex
;
110 struct kmem_cache
*drbd_request_cache
;
111 struct kmem_cache
*drbd_ee_cache
; /* peer requests */
112 struct kmem_cache
*drbd_bm_ext_cache
; /* bitmap extents */
113 struct kmem_cache
*drbd_al_ext_cache
; /* activity log extents */
114 mempool_t drbd_request_mempool
;
115 mempool_t drbd_ee_mempool
;
116 mempool_t drbd_md_io_page_pool
;
117 struct bio_set drbd_md_io_bio_set
;
118 struct bio_set drbd_io_bio_set
;
120 /* I do not use a standard mempool, because:
121 1) I want to hand out the pre-allocated objects first.
122 2) I want to be able to interrupt sleeping allocation with a signal.
123 Note: This is a single linked list, the next pointer is the private
124 member of struct page.
126 struct page
*drbd_pp_pool
;
127 DEFINE_SPINLOCK(drbd_pp_lock
);
129 wait_queue_head_t drbd_pp_wait
;
131 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state
, 5 * HZ
, 5);
133 static const struct block_device_operations drbd_ops
= {
134 .owner
= THIS_MODULE
,
135 .submit_bio
= drbd_submit_bio
,
137 .release
= drbd_release
,
141 /* When checking with sparse, and this is an inline function, sparse will
142 give tons of false positives. When this is a real functions sparse works.
144 int _get_ldev_if_state(struct drbd_device
*device
, enum drbd_disk_state mins
)
148 atomic_inc(&device
->local_cnt
);
149 io_allowed
= (device
->state
.disk
>= mins
);
151 if (atomic_dec_and_test(&device
->local_cnt
))
152 wake_up(&device
->misc_wait
);
160 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
161 * @connection: DRBD connection.
162 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
163 * @set_size: Expected number of requests before that barrier.
165 * In case the passed barrier_nr or set_size does not match the oldest
166 * epoch of not yet barrier-acked requests, this function will cause a
167 * termination of the connection.
169 void tl_release(struct drbd_connection
*connection
, unsigned int barrier_nr
,
170 unsigned int set_size
)
172 struct drbd_request
*r
;
173 struct drbd_request
*req
= NULL
, *tmp
= NULL
;
174 int expect_epoch
= 0;
177 spin_lock_irq(&connection
->resource
->req_lock
);
179 /* find oldest not yet barrier-acked write request,
180 * count writes in its epoch. */
181 list_for_each_entry(r
, &connection
->transfer_log
, tl_requests
) {
182 const unsigned s
= r
->rq_state
;
186 if (!(s
& RQ_NET_MASK
))
191 expect_epoch
= req
->epoch
;
194 if (r
->epoch
!= expect_epoch
)
198 /* if (s & RQ_DONE): not expected */
199 /* if (!(s & RQ_NET_MASK)): not expected */
204 /* first some paranoia code */
206 drbd_err(connection
, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
210 if (expect_epoch
!= barrier_nr
) {
211 drbd_err(connection
, "BAD! BarrierAck #%u received, expected #%u!\n",
212 barrier_nr
, expect_epoch
);
216 if (expect_size
!= set_size
) {
217 drbd_err(connection
, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
218 barrier_nr
, set_size
, expect_size
);
222 /* Clean up list of requests processed during current epoch. */
223 /* this extra list walk restart is paranoia,
224 * to catch requests being barrier-acked "unexpectedly".
225 * It usually should find the same req again, or some READ preceding it. */
226 list_for_each_entry(req
, &connection
->transfer_log
, tl_requests
)
227 if (req
->epoch
== expect_epoch
) {
231 req
= list_prepare_entry(tmp
, &connection
->transfer_log
, tl_requests
);
232 list_for_each_entry_safe_from(req
, r
, &connection
->transfer_log
, tl_requests
) {
233 struct drbd_peer_device
*peer_device
;
234 if (req
->epoch
!= expect_epoch
)
236 peer_device
= conn_peer_device(connection
, req
->device
->vnr
);
237 _req_mod(req
, BARRIER_ACKED
, peer_device
);
239 spin_unlock_irq(&connection
->resource
->req_lock
);
244 spin_unlock_irq(&connection
->resource
->req_lock
);
245 conn_request_state(connection
, NS(conn
, C_PROTOCOL_ERROR
), CS_HARD
);
250 * _tl_restart() - Walks the transfer log, and applies an action to all requests
251 * @connection: DRBD connection to operate on.
252 * @what: The action/event to perform with all request objects
254 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
255 * RESTART_FROZEN_DISK_IO.
257 /* must hold resource->req_lock */
258 void _tl_restart(struct drbd_connection
*connection
, enum drbd_req_event what
)
260 struct drbd_peer_device
*peer_device
;
261 struct drbd_request
*req
, *r
;
263 list_for_each_entry_safe(req
, r
, &connection
->transfer_log
, tl_requests
) {
264 peer_device
= conn_peer_device(connection
, req
->device
->vnr
);
265 _req_mod(req
, what
, peer_device
);
269 void tl_restart(struct drbd_connection
*connection
, enum drbd_req_event what
)
271 spin_lock_irq(&connection
->resource
->req_lock
);
272 _tl_restart(connection
, what
);
273 spin_unlock_irq(&connection
->resource
->req_lock
);
277 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
278 * @connection: DRBD connection.
280 * This is called after the connection to the peer was lost. The storage covered
281 * by the requests on the transfer gets marked as our of sync. Called from the
282 * receiver thread and the worker thread.
284 void tl_clear(struct drbd_connection
*connection
)
286 tl_restart(connection
, CONNECTION_LOST_WHILE_PENDING
);
290 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
291 * @device: DRBD device.
293 void tl_abort_disk_io(struct drbd_device
*device
)
295 struct drbd_connection
*connection
= first_peer_device(device
)->connection
;
296 struct drbd_request
*req
, *r
;
298 spin_lock_irq(&connection
->resource
->req_lock
);
299 list_for_each_entry_safe(req
, r
, &connection
->transfer_log
, tl_requests
) {
300 if (!(req
->rq_state
& RQ_LOCAL_PENDING
))
302 if (req
->device
!= device
)
304 _req_mod(req
, ABORT_DISK_IO
, NULL
);
306 spin_unlock_irq(&connection
->resource
->req_lock
);
309 static int drbd_thread_setup(void *arg
)
311 struct drbd_thread
*thi
= (struct drbd_thread
*) arg
;
312 struct drbd_resource
*resource
= thi
->resource
;
316 snprintf(current
->comm
, sizeof(current
->comm
), "drbd_%c_%s",
320 allow_kernel_signal(DRBD_SIGKILL
);
321 allow_kernel_signal(SIGXCPU
);
323 retval
= thi
->function(thi
);
325 spin_lock_irqsave(&thi
->t_lock
, flags
);
327 /* if the receiver has been "EXITING", the last thing it did
328 * was set the conn state to "StandAlone",
329 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
330 * and receiver thread will be "started".
331 * drbd_thread_start needs to set "RESTARTING" in that case.
332 * t_state check and assignment needs to be within the same spinlock,
333 * so either thread_start sees EXITING, and can remap to RESTARTING,
334 * or thread_start see NONE, and can proceed as normal.
337 if (thi
->t_state
== RESTARTING
) {
338 drbd_info(resource
, "Restarting %s thread\n", thi
->name
);
339 thi
->t_state
= RUNNING
;
340 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
347 complete_all(&thi
->stop
);
348 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
350 drbd_info(resource
, "Terminating %s\n", current
->comm
);
352 /* Release mod reference taken when thread was started */
355 kref_put(&thi
->connection
->kref
, drbd_destroy_connection
);
356 kref_put(&resource
->kref
, drbd_destroy_resource
);
357 module_put(THIS_MODULE
);
361 static void drbd_thread_init(struct drbd_resource
*resource
, struct drbd_thread
*thi
,
362 int (*func
) (struct drbd_thread
*), const char *name
)
364 spin_lock_init(&thi
->t_lock
);
367 thi
->function
= func
;
368 thi
->resource
= resource
;
369 thi
->connection
= NULL
;
373 int drbd_thread_start(struct drbd_thread
*thi
)
375 struct drbd_resource
*resource
= thi
->resource
;
376 struct task_struct
*nt
;
379 /* is used from state engine doing drbd_thread_stop_nowait,
380 * while holding the req lock irqsave */
381 spin_lock_irqsave(&thi
->t_lock
, flags
);
383 switch (thi
->t_state
) {
385 drbd_info(resource
, "Starting %s thread (from %s [%d])\n",
386 thi
->name
, current
->comm
, current
->pid
);
388 /* Get ref on module for thread - this is released when thread exits */
389 if (!try_module_get(THIS_MODULE
)) {
390 drbd_err(resource
, "Failed to get module reference in drbd_thread_start\n");
391 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
395 kref_get(&resource
->kref
);
397 kref_get(&thi
->connection
->kref
);
399 init_completion(&thi
->stop
);
400 thi
->reset_cpu_mask
= 1;
401 thi
->t_state
= RUNNING
;
402 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
403 flush_signals(current
); /* otherw. may get -ERESTARTNOINTR */
405 nt
= kthread_create(drbd_thread_setup
, (void *) thi
,
406 "drbd_%c_%s", thi
->name
[0], thi
->resource
->name
);
409 drbd_err(resource
, "Couldn't start thread\n");
412 kref_put(&thi
->connection
->kref
, drbd_destroy_connection
);
413 kref_put(&resource
->kref
, drbd_destroy_resource
);
414 module_put(THIS_MODULE
);
417 spin_lock_irqsave(&thi
->t_lock
, flags
);
419 thi
->t_state
= RUNNING
;
420 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
424 thi
->t_state
= RESTARTING
;
425 drbd_info(resource
, "Restarting %s thread (from %s [%d])\n",
426 thi
->name
, current
->comm
, current
->pid
);
431 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
439 void _drbd_thread_stop(struct drbd_thread
*thi
, int restart
, int wait
)
443 enum drbd_thread_state ns
= restart
? RESTARTING
: EXITING
;
445 /* may be called from state engine, holding the req lock irqsave */
446 spin_lock_irqsave(&thi
->t_lock
, flags
);
448 if (thi
->t_state
== NONE
) {
449 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
451 drbd_thread_start(thi
);
455 if (thi
->t_state
!= ns
) {
456 if (thi
->task
== NULL
) {
457 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
463 init_completion(&thi
->stop
);
464 if (thi
->task
!= current
)
465 send_sig(DRBD_SIGKILL
, thi
->task
, 1);
468 spin_unlock_irqrestore(&thi
->t_lock
, flags
);
471 wait_for_completion(&thi
->stop
);
474 int conn_lowest_minor(struct drbd_connection
*connection
)
476 struct drbd_peer_device
*peer_device
;
477 int vnr
= 0, minor
= -1;
480 peer_device
= idr_get_next(&connection
->peer_devices
, &vnr
);
482 minor
= device_to_minor(peer_device
->device
);
490 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
492 * Forces all threads of a resource onto the same CPU. This is beneficial for
493 * DRBD's performance. May be overwritten by user's configuration.
495 static void drbd_calc_cpu_mask(cpumask_var_t
*cpu_mask
)
497 unsigned int *resources_per_cpu
, min_index
= ~0;
499 resources_per_cpu
= kcalloc(nr_cpu_ids
, sizeof(*resources_per_cpu
),
501 if (resources_per_cpu
) {
502 struct drbd_resource
*resource
;
503 unsigned int cpu
, min
= ~0;
506 for_each_resource_rcu(resource
, &drbd_resources
) {
507 for_each_cpu(cpu
, resource
->cpu_mask
)
508 resources_per_cpu
[cpu
]++;
511 for_each_online_cpu(cpu
) {
512 if (resources_per_cpu
[cpu
] < min
) {
513 min
= resources_per_cpu
[cpu
];
517 kfree(resources_per_cpu
);
519 if (min_index
== ~0) {
520 cpumask_setall(*cpu_mask
);
523 cpumask_set_cpu(min_index
, *cpu_mask
);
527 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
528 * @thi: drbd_thread object
530 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
533 void drbd_thread_current_set_cpu(struct drbd_thread
*thi
)
535 struct drbd_resource
*resource
= thi
->resource
;
536 struct task_struct
*p
= current
;
538 if (!thi
->reset_cpu_mask
)
540 thi
->reset_cpu_mask
= 0;
541 set_cpus_allowed_ptr(p
, resource
->cpu_mask
);
544 #define drbd_calc_cpu_mask(A) ({})
548 * drbd_header_size - size of a packet header
550 * The header size is a multiple of 8, so any payload following the header is
551 * word aligned on 64-bit architectures. (The bitmap send and receive code
554 unsigned int drbd_header_size(struct drbd_connection
*connection
)
556 if (connection
->agreed_pro_version
>= 100) {
557 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100
), 8));
558 return sizeof(struct p_header100
);
560 BUILD_BUG_ON(sizeof(struct p_header80
) !=
561 sizeof(struct p_header95
));
562 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80
), 8));
563 return sizeof(struct p_header80
);
567 static unsigned int prepare_header80(struct p_header80
*h
, enum drbd_packet cmd
, int size
)
569 h
->magic
= cpu_to_be32(DRBD_MAGIC
);
570 h
->command
= cpu_to_be16(cmd
);
571 h
->length
= cpu_to_be16(size
);
572 return sizeof(struct p_header80
);
575 static unsigned int prepare_header95(struct p_header95
*h
, enum drbd_packet cmd
, int size
)
577 h
->magic
= cpu_to_be16(DRBD_MAGIC_BIG
);
578 h
->command
= cpu_to_be16(cmd
);
579 h
->length
= cpu_to_be32(size
);
580 return sizeof(struct p_header95
);
583 static unsigned int prepare_header100(struct p_header100
*h
, enum drbd_packet cmd
,
586 h
->magic
= cpu_to_be32(DRBD_MAGIC_100
);
587 h
->volume
= cpu_to_be16(vnr
);
588 h
->command
= cpu_to_be16(cmd
);
589 h
->length
= cpu_to_be32(size
);
591 return sizeof(struct p_header100
);
594 static unsigned int prepare_header(struct drbd_connection
*connection
, int vnr
,
595 void *buffer
, enum drbd_packet cmd
, int size
)
597 if (connection
->agreed_pro_version
>= 100)
598 return prepare_header100(buffer
, cmd
, size
, vnr
);
599 else if (connection
->agreed_pro_version
>= 95 &&
600 size
> DRBD_MAX_SIZE_H80_PACKET
)
601 return prepare_header95(buffer
, cmd
, size
);
603 return prepare_header80(buffer
, cmd
, size
);
606 static void *__conn_prepare_command(struct drbd_connection
*connection
,
607 struct drbd_socket
*sock
)
611 return sock
->sbuf
+ drbd_header_size(connection
);
614 void *conn_prepare_command(struct drbd_connection
*connection
, struct drbd_socket
*sock
)
618 mutex_lock(&sock
->mutex
);
619 p
= __conn_prepare_command(connection
, sock
);
621 mutex_unlock(&sock
->mutex
);
626 void *drbd_prepare_command(struct drbd_peer_device
*peer_device
, struct drbd_socket
*sock
)
628 return conn_prepare_command(peer_device
->connection
, sock
);
631 static int __send_command(struct drbd_connection
*connection
, int vnr
,
632 struct drbd_socket
*sock
, enum drbd_packet cmd
,
633 unsigned int header_size
, void *data
,
640 * Called with @data == NULL and the size of the data blocks in @size
641 * for commands that send data blocks. For those commands, omit the
642 * MSG_MORE flag: this will increase the likelihood that data blocks
643 * which are page aligned on the sender will end up page aligned on the
646 msg_flags
= data
? MSG_MORE
: 0;
648 header_size
+= prepare_header(connection
, vnr
, sock
->sbuf
, cmd
,
650 err
= drbd_send_all(connection
, sock
->socket
, sock
->sbuf
, header_size
,
653 err
= drbd_send_all(connection
, sock
->socket
, data
, size
, 0);
654 /* DRBD protocol "pings" are latency critical.
655 * This is supposed to trigger tcp_push_pending_frames() */
656 if (!err
&& (cmd
== P_PING
|| cmd
== P_PING_ACK
))
657 tcp_sock_set_nodelay(sock
->socket
->sk
);
662 static int __conn_send_command(struct drbd_connection
*connection
, struct drbd_socket
*sock
,
663 enum drbd_packet cmd
, unsigned int header_size
,
664 void *data
, unsigned int size
)
666 return __send_command(connection
, 0, sock
, cmd
, header_size
, data
, size
);
669 int conn_send_command(struct drbd_connection
*connection
, struct drbd_socket
*sock
,
670 enum drbd_packet cmd
, unsigned int header_size
,
671 void *data
, unsigned int size
)
675 err
= __conn_send_command(connection
, sock
, cmd
, header_size
, data
, size
);
676 mutex_unlock(&sock
->mutex
);
680 int drbd_send_command(struct drbd_peer_device
*peer_device
, struct drbd_socket
*sock
,
681 enum drbd_packet cmd
, unsigned int header_size
,
682 void *data
, unsigned int size
)
686 err
= __send_command(peer_device
->connection
, peer_device
->device
->vnr
,
687 sock
, cmd
, header_size
, data
, size
);
688 mutex_unlock(&sock
->mutex
);
692 int drbd_send_ping(struct drbd_connection
*connection
)
694 struct drbd_socket
*sock
;
696 sock
= &connection
->meta
;
697 if (!conn_prepare_command(connection
, sock
))
699 return conn_send_command(connection
, sock
, P_PING
, 0, NULL
, 0);
702 int drbd_send_ping_ack(struct drbd_connection
*connection
)
704 struct drbd_socket
*sock
;
706 sock
= &connection
->meta
;
707 if (!conn_prepare_command(connection
, sock
))
709 return conn_send_command(connection
, sock
, P_PING_ACK
, 0, NULL
, 0);
712 int drbd_send_sync_param(struct drbd_peer_device
*peer_device
)
714 struct drbd_socket
*sock
;
715 struct p_rs_param_95
*p
;
717 const int apv
= peer_device
->connection
->agreed_pro_version
;
718 enum drbd_packet cmd
;
720 struct disk_conf
*dc
;
722 sock
= &peer_device
->connection
->data
;
723 p
= drbd_prepare_command(peer_device
, sock
);
728 nc
= rcu_dereference(peer_device
->connection
->net_conf
);
730 size
= apv
<= 87 ? sizeof(struct p_rs_param
)
731 : apv
== 88 ? sizeof(struct p_rs_param
)
732 + strlen(nc
->verify_alg
) + 1
733 : apv
<= 94 ? sizeof(struct p_rs_param_89
)
734 : /* apv >= 95 */ sizeof(struct p_rs_param_95
);
736 cmd
= apv
>= 89 ? P_SYNC_PARAM89
: P_SYNC_PARAM
;
738 /* initialize verify_alg and csums_alg */
739 BUILD_BUG_ON(sizeof(p
->algs
) != 2 * SHARED_SECRET_MAX
);
740 memset(&p
->algs
, 0, sizeof(p
->algs
));
742 if (get_ldev(peer_device
->device
)) {
743 dc
= rcu_dereference(peer_device
->device
->ldev
->disk_conf
);
744 p
->resync_rate
= cpu_to_be32(dc
->resync_rate
);
745 p
->c_plan_ahead
= cpu_to_be32(dc
->c_plan_ahead
);
746 p
->c_delay_target
= cpu_to_be32(dc
->c_delay_target
);
747 p
->c_fill_target
= cpu_to_be32(dc
->c_fill_target
);
748 p
->c_max_rate
= cpu_to_be32(dc
->c_max_rate
);
749 put_ldev(peer_device
->device
);
751 p
->resync_rate
= cpu_to_be32(DRBD_RESYNC_RATE_DEF
);
752 p
->c_plan_ahead
= cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF
);
753 p
->c_delay_target
= cpu_to_be32(DRBD_C_DELAY_TARGET_DEF
);
754 p
->c_fill_target
= cpu_to_be32(DRBD_C_FILL_TARGET_DEF
);
755 p
->c_max_rate
= cpu_to_be32(DRBD_C_MAX_RATE_DEF
);
759 strcpy(p
->verify_alg
, nc
->verify_alg
);
761 strcpy(p
->csums_alg
, nc
->csums_alg
);
764 return drbd_send_command(peer_device
, sock
, cmd
, size
, NULL
, 0);
767 int __drbd_send_protocol(struct drbd_connection
*connection
, enum drbd_packet cmd
)
769 struct drbd_socket
*sock
;
770 struct p_protocol
*p
;
774 sock
= &connection
->data
;
775 p
= __conn_prepare_command(connection
, sock
);
780 nc
= rcu_dereference(connection
->net_conf
);
782 if (nc
->tentative
&& connection
->agreed_pro_version
< 92) {
784 drbd_err(connection
, "--dry-run is not supported by peer");
789 if (connection
->agreed_pro_version
>= 87)
790 size
+= strlen(nc
->integrity_alg
) + 1;
792 p
->protocol
= cpu_to_be32(nc
->wire_protocol
);
793 p
->after_sb_0p
= cpu_to_be32(nc
->after_sb_0p
);
794 p
->after_sb_1p
= cpu_to_be32(nc
->after_sb_1p
);
795 p
->after_sb_2p
= cpu_to_be32(nc
->after_sb_2p
);
796 p
->two_primaries
= cpu_to_be32(nc
->two_primaries
);
798 if (nc
->discard_my_data
)
799 cf
|= CF_DISCARD_MY_DATA
;
802 p
->conn_flags
= cpu_to_be32(cf
);
804 if (connection
->agreed_pro_version
>= 87)
805 strcpy(p
->integrity_alg
, nc
->integrity_alg
);
808 return __conn_send_command(connection
, sock
, cmd
, size
, NULL
, 0);
811 int drbd_send_protocol(struct drbd_connection
*connection
)
815 mutex_lock(&connection
->data
.mutex
);
816 err
= __drbd_send_protocol(connection
, P_PROTOCOL
);
817 mutex_unlock(&connection
->data
.mutex
);
822 static int _drbd_send_uuids(struct drbd_peer_device
*peer_device
, u64 uuid_flags
)
824 struct drbd_device
*device
= peer_device
->device
;
825 struct drbd_socket
*sock
;
829 if (!get_ldev_if_state(device
, D_NEGOTIATING
))
832 sock
= &peer_device
->connection
->data
;
833 p
= drbd_prepare_command(peer_device
, sock
);
838 spin_lock_irq(&device
->ldev
->md
.uuid_lock
);
839 for (i
= UI_CURRENT
; i
< UI_SIZE
; i
++)
840 p
->uuid
[i
] = cpu_to_be64(device
->ldev
->md
.uuid
[i
]);
841 spin_unlock_irq(&device
->ldev
->md
.uuid_lock
);
843 device
->comm_bm_set
= drbd_bm_total_weight(device
);
844 p
->uuid
[UI_SIZE
] = cpu_to_be64(device
->comm_bm_set
);
846 uuid_flags
|= rcu_dereference(peer_device
->connection
->net_conf
)->discard_my_data
? 1 : 0;
848 uuid_flags
|= test_bit(CRASHED_PRIMARY
, &device
->flags
) ? 2 : 0;
849 uuid_flags
|= device
->new_state_tmp
.disk
== D_INCONSISTENT
? 4 : 0;
850 p
->uuid
[UI_FLAGS
] = cpu_to_be64(uuid_flags
);
853 return drbd_send_command(peer_device
, sock
, P_UUIDS
, sizeof(*p
), NULL
, 0);
856 int drbd_send_uuids(struct drbd_peer_device
*peer_device
)
858 return _drbd_send_uuids(peer_device
, 0);
861 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device
*peer_device
)
863 return _drbd_send_uuids(peer_device
, 8);
866 void drbd_print_uuids(struct drbd_device
*device
, const char *text
)
868 if (get_ldev_if_state(device
, D_NEGOTIATING
)) {
869 u64
*uuid
= device
->ldev
->md
.uuid
;
870 drbd_info(device
, "%s %016llX:%016llX:%016llX:%016llX\n",
872 (unsigned long long)uuid
[UI_CURRENT
],
873 (unsigned long long)uuid
[UI_BITMAP
],
874 (unsigned long long)uuid
[UI_HISTORY_START
],
875 (unsigned long long)uuid
[UI_HISTORY_END
]);
878 drbd_info(device
, "%s effective data uuid: %016llX\n",
880 (unsigned long long)device
->ed_uuid
);
884 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device
*peer_device
)
886 struct drbd_device
*device
= peer_device
->device
;
887 struct drbd_socket
*sock
;
891 D_ASSERT(device
, device
->state
.disk
== D_UP_TO_DATE
);
893 uuid
= device
->ldev
->md
.uuid
[UI_BITMAP
];
894 if (uuid
&& uuid
!= UUID_JUST_CREATED
)
895 uuid
= uuid
+ UUID_NEW_BM_OFFSET
;
897 get_random_bytes(&uuid
, sizeof(u64
));
898 drbd_uuid_set(device
, UI_BITMAP
, uuid
);
899 drbd_print_uuids(device
, "updated sync UUID");
900 drbd_md_sync(device
);
902 sock
= &peer_device
->connection
->data
;
903 p
= drbd_prepare_command(peer_device
, sock
);
905 p
->uuid
= cpu_to_be64(uuid
);
906 drbd_send_command(peer_device
, sock
, P_SYNC_UUID
, sizeof(*p
), NULL
, 0);
910 int drbd_send_sizes(struct drbd_peer_device
*peer_device
, int trigger_reply
, enum dds_flags flags
)
912 struct drbd_device
*device
= peer_device
->device
;
913 struct drbd_socket
*sock
;
915 sector_t d_size
, u_size
;
917 unsigned int max_bio_size
;
918 unsigned int packet_size
;
920 sock
= &peer_device
->connection
->data
;
921 p
= drbd_prepare_command(peer_device
, sock
);
925 packet_size
= sizeof(*p
);
926 if (peer_device
->connection
->agreed_features
& DRBD_FF_WSAME
)
927 packet_size
+= sizeof(p
->qlim
[0]);
929 memset(p
, 0, packet_size
);
930 if (get_ldev_if_state(device
, D_NEGOTIATING
)) {
931 struct block_device
*bdev
= device
->ldev
->backing_bdev
;
932 struct request_queue
*q
= bdev_get_queue(bdev
);
934 d_size
= drbd_get_max_capacity(device
->ldev
);
936 u_size
= rcu_dereference(device
->ldev
->disk_conf
)->disk_size
;
938 q_order_type
= drbd_queue_order_type(device
);
939 max_bio_size
= queue_max_hw_sectors(q
) << 9;
940 max_bio_size
= min(max_bio_size
, DRBD_MAX_BIO_SIZE
);
941 p
->qlim
->physical_block_size
=
942 cpu_to_be32(bdev_physical_block_size(bdev
));
943 p
->qlim
->logical_block_size
=
944 cpu_to_be32(bdev_logical_block_size(bdev
));
945 p
->qlim
->alignment_offset
=
946 cpu_to_be32(bdev_alignment_offset(bdev
));
947 p
->qlim
->io_min
= cpu_to_be32(bdev_io_min(bdev
));
948 p
->qlim
->io_opt
= cpu_to_be32(bdev_io_opt(bdev
));
949 p
->qlim
->discard_enabled
= !!bdev_max_discard_sectors(bdev
);
952 struct request_queue
*q
= device
->rq_queue
;
954 p
->qlim
->physical_block_size
=
955 cpu_to_be32(queue_physical_block_size(q
));
956 p
->qlim
->logical_block_size
=
957 cpu_to_be32(queue_logical_block_size(q
));
958 p
->qlim
->alignment_offset
= 0;
959 p
->qlim
->io_min
= cpu_to_be32(queue_io_min(q
));
960 p
->qlim
->io_opt
= cpu_to_be32(queue_io_opt(q
));
961 p
->qlim
->discard_enabled
= 0;
965 q_order_type
= QUEUE_ORDERED_NONE
;
966 max_bio_size
= DRBD_MAX_BIO_SIZE
; /* ... multiple BIOs per peer_request */
969 if (peer_device
->connection
->agreed_pro_version
<= 94)
970 max_bio_size
= min(max_bio_size
, DRBD_MAX_SIZE_H80_PACKET
);
971 else if (peer_device
->connection
->agreed_pro_version
< 100)
972 max_bio_size
= min(max_bio_size
, DRBD_MAX_BIO_SIZE_P95
);
974 p
->d_size
= cpu_to_be64(d_size
);
975 p
->u_size
= cpu_to_be64(u_size
);
979 p
->c_size
= cpu_to_be64(get_capacity(device
->vdisk
));
980 p
->max_bio_size
= cpu_to_be32(max_bio_size
);
981 p
->queue_order_type
= cpu_to_be16(q_order_type
);
982 p
->dds_flags
= cpu_to_be16(flags
);
984 return drbd_send_command(peer_device
, sock
, P_SIZES
, packet_size
, NULL
, 0);
988 * drbd_send_current_state() - Sends the drbd state to the peer
989 * @peer_device: DRBD peer device.
991 int drbd_send_current_state(struct drbd_peer_device
*peer_device
)
993 struct drbd_socket
*sock
;
996 sock
= &peer_device
->connection
->data
;
997 p
= drbd_prepare_command(peer_device
, sock
);
1000 p
->state
= cpu_to_be32(peer_device
->device
->state
.i
); /* Within the send mutex */
1001 return drbd_send_command(peer_device
, sock
, P_STATE
, sizeof(*p
), NULL
, 0);
1005 * drbd_send_state() - After a state change, sends the new state to the peer
1006 * @peer_device: DRBD peer device.
1007 * @state: the state to send, not necessarily the current state.
1009 * Each state change queues an "after_state_ch" work, which will eventually
1010 * send the resulting new state to the peer. If more state changes happen
1011 * between queuing and processing of the after_state_ch work, we still
1012 * want to send each intermediary state in the order it occurred.
1014 int drbd_send_state(struct drbd_peer_device
*peer_device
, union drbd_state state
)
1016 struct drbd_socket
*sock
;
1019 sock
= &peer_device
->connection
->data
;
1020 p
= drbd_prepare_command(peer_device
, sock
);
1023 p
->state
= cpu_to_be32(state
.i
); /* Within the send mutex */
1024 return drbd_send_command(peer_device
, sock
, P_STATE
, sizeof(*p
), NULL
, 0);
1027 int drbd_send_state_req(struct drbd_peer_device
*peer_device
, union drbd_state mask
, union drbd_state val
)
1029 struct drbd_socket
*sock
;
1030 struct p_req_state
*p
;
1032 sock
= &peer_device
->connection
->data
;
1033 p
= drbd_prepare_command(peer_device
, sock
);
1036 p
->mask
= cpu_to_be32(mask
.i
);
1037 p
->val
= cpu_to_be32(val
.i
);
1038 return drbd_send_command(peer_device
, sock
, P_STATE_CHG_REQ
, sizeof(*p
), NULL
, 0);
1041 int conn_send_state_req(struct drbd_connection
*connection
, union drbd_state mask
, union drbd_state val
)
1043 enum drbd_packet cmd
;
1044 struct drbd_socket
*sock
;
1045 struct p_req_state
*p
;
1047 cmd
= connection
->agreed_pro_version
< 100 ? P_STATE_CHG_REQ
: P_CONN_ST_CHG_REQ
;
1048 sock
= &connection
->data
;
1049 p
= conn_prepare_command(connection
, sock
);
1052 p
->mask
= cpu_to_be32(mask
.i
);
1053 p
->val
= cpu_to_be32(val
.i
);
1054 return conn_send_command(connection
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1057 void drbd_send_sr_reply(struct drbd_peer_device
*peer_device
, enum drbd_state_rv retcode
)
1059 struct drbd_socket
*sock
;
1060 struct p_req_state_reply
*p
;
1062 sock
= &peer_device
->connection
->meta
;
1063 p
= drbd_prepare_command(peer_device
, sock
);
1065 p
->retcode
= cpu_to_be32(retcode
);
1066 drbd_send_command(peer_device
, sock
, P_STATE_CHG_REPLY
, sizeof(*p
), NULL
, 0);
1070 void conn_send_sr_reply(struct drbd_connection
*connection
, enum drbd_state_rv retcode
)
1072 struct drbd_socket
*sock
;
1073 struct p_req_state_reply
*p
;
1074 enum drbd_packet cmd
= connection
->agreed_pro_version
< 100 ? P_STATE_CHG_REPLY
: P_CONN_ST_CHG_REPLY
;
1076 sock
= &connection
->meta
;
1077 p
= conn_prepare_command(connection
, sock
);
1079 p
->retcode
= cpu_to_be32(retcode
);
1080 conn_send_command(connection
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1084 static void dcbp_set_code(struct p_compressed_bm
*p
, enum drbd_bitmap_code code
)
1086 BUG_ON(code
& ~0xf);
1087 p
->encoding
= (p
->encoding
& ~0xf) | code
;
1090 static void dcbp_set_start(struct p_compressed_bm
*p
, int set
)
1092 p
->encoding
= (p
->encoding
& ~0x80) | (set
? 0x80 : 0);
1095 static void dcbp_set_pad_bits(struct p_compressed_bm
*p
, int n
)
1098 p
->encoding
= (p
->encoding
& (~0x7 << 4)) | (n
<< 4);
1101 static int fill_bitmap_rle_bits(struct drbd_device
*device
,
1102 struct p_compressed_bm
*p
,
1104 struct bm_xfer_ctx
*c
)
1106 struct bitstream bs
;
1107 unsigned long plain_bits
;
1114 /* may we use this feature? */
1116 use_rle
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
)->use_rle
;
1118 if (!use_rle
|| first_peer_device(device
)->connection
->agreed_pro_version
< 90)
1121 if (c
->bit_offset
>= c
->bm_bits
)
1122 return 0; /* nothing to do. */
1124 /* use at most thus many bytes */
1125 bitstream_init(&bs
, p
->code
, size
, 0);
1126 memset(p
->code
, 0, size
);
1127 /* plain bits covered in this code string */
1130 /* p->encoding & 0x80 stores whether the first run length is set.
1131 * bit offset is implicit.
1132 * start with toggle == 2 to be able to tell the first iteration */
1135 /* see how much plain bits we can stuff into one packet
1136 * using RLE and VLI. */
1138 tmp
= (toggle
== 0) ? _drbd_bm_find_next_zero(device
, c
->bit_offset
)
1139 : _drbd_bm_find_next(device
, c
->bit_offset
);
1142 rl
= tmp
- c
->bit_offset
;
1144 if (toggle
== 2) { /* first iteration */
1146 /* the first checked bit was set,
1147 * store start value, */
1148 dcbp_set_start(p
, 1);
1149 /* but skip encoding of zero run length */
1153 dcbp_set_start(p
, 0);
1156 /* paranoia: catch zero runlength.
1157 * can only happen if bitmap is modified while we scan it. */
1159 drbd_err(device
, "unexpected zero runlength while encoding bitmap "
1160 "t:%u bo:%lu\n", toggle
, c
->bit_offset
);
1164 bits
= vli_encode_bits(&bs
, rl
);
1165 if (bits
== -ENOBUFS
) /* buffer full */
1168 drbd_err(device
, "error while encoding bitmap: %d\n", bits
);
1174 c
->bit_offset
= tmp
;
1175 } while (c
->bit_offset
< c
->bm_bits
);
1177 len
= bs
.cur
.b
- p
->code
+ !!bs
.cur
.bit
;
1179 if (plain_bits
< (len
<< 3)) {
1180 /* incompressible with this method.
1181 * we need to rewind both word and bit position. */
1182 c
->bit_offset
-= plain_bits
;
1183 bm_xfer_ctx_bit_to_word_offset(c
);
1184 c
->bit_offset
= c
->word_offset
* BITS_PER_LONG
;
1188 /* RLE + VLI was able to compress it just fine.
1189 * update c->word_offset. */
1190 bm_xfer_ctx_bit_to_word_offset(c
);
1192 /* store pad_bits */
1193 dcbp_set_pad_bits(p
, (8 - bs
.cur
.bit
) & 0x7);
1199 * send_bitmap_rle_or_plain
1201 * Return 0 when done, 1 when another iteration is needed, and a negative error
1202 * code upon failure.
1205 send_bitmap_rle_or_plain(struct drbd_peer_device
*peer_device
, struct bm_xfer_ctx
*c
)
1207 struct drbd_device
*device
= peer_device
->device
;
1208 struct drbd_socket
*sock
= &peer_device
->connection
->data
;
1209 unsigned int header_size
= drbd_header_size(peer_device
->connection
);
1210 struct p_compressed_bm
*p
= sock
->sbuf
+ header_size
;
1213 len
= fill_bitmap_rle_bits(device
, p
,
1214 DRBD_SOCKET_BUFFER_SIZE
- header_size
- sizeof(*p
), c
);
1219 dcbp_set_code(p
, RLE_VLI_Bits
);
1220 err
= __send_command(peer_device
->connection
, device
->vnr
, sock
,
1221 P_COMPRESSED_BITMAP
, sizeof(*p
) + len
,
1224 c
->bytes
[0] += header_size
+ sizeof(*p
) + len
;
1226 if (c
->bit_offset
>= c
->bm_bits
)
1229 /* was not compressible.
1230 * send a buffer full of plain text bits instead. */
1231 unsigned int data_size
;
1232 unsigned long num_words
;
1233 unsigned long *p
= sock
->sbuf
+ header_size
;
1235 data_size
= DRBD_SOCKET_BUFFER_SIZE
- header_size
;
1236 num_words
= min_t(size_t, data_size
/ sizeof(*p
),
1237 c
->bm_words
- c
->word_offset
);
1238 len
= num_words
* sizeof(*p
);
1240 drbd_bm_get_lel(device
, c
->word_offset
, num_words
, p
);
1241 err
= __send_command(peer_device
->connection
, device
->vnr
, sock
, P_BITMAP
,
1243 c
->word_offset
+= num_words
;
1244 c
->bit_offset
= c
->word_offset
* BITS_PER_LONG
;
1247 c
->bytes
[1] += header_size
+ len
;
1249 if (c
->bit_offset
> c
->bm_bits
)
1250 c
->bit_offset
= c
->bm_bits
;
1254 INFO_bm_xfer_stats(peer_device
, "send", c
);
1262 /* See the comment at receive_bitmap() */
1263 static int _drbd_send_bitmap(struct drbd_device
*device
,
1264 struct drbd_peer_device
*peer_device
)
1266 struct bm_xfer_ctx c
;
1269 if (!expect(device
, device
->bitmap
))
1272 if (get_ldev(device
)) {
1273 if (drbd_md_test_flag(device
->ldev
, MDF_FULL_SYNC
)) {
1274 drbd_info(device
, "Writing the whole bitmap, MDF_FullSync was set.\n");
1275 drbd_bm_set_all(device
);
1276 if (drbd_bm_write(device
, peer_device
)) {
1277 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1278 * but otherwise process as per normal - need to tell other
1279 * side that a full resync is required! */
1280 drbd_err(device
, "Failed to write bitmap to disk!\n");
1282 drbd_md_clear_flag(device
, MDF_FULL_SYNC
);
1283 drbd_md_sync(device
);
1289 c
= (struct bm_xfer_ctx
) {
1290 .bm_bits
= drbd_bm_bits(device
),
1291 .bm_words
= drbd_bm_words(device
),
1295 err
= send_bitmap_rle_or_plain(peer_device
, &c
);
1301 int drbd_send_bitmap(struct drbd_device
*device
, struct drbd_peer_device
*peer_device
)
1303 struct drbd_socket
*sock
= &peer_device
->connection
->data
;
1306 mutex_lock(&sock
->mutex
);
1308 err
= !_drbd_send_bitmap(device
, peer_device
);
1309 mutex_unlock(&sock
->mutex
);
1313 void drbd_send_b_ack(struct drbd_connection
*connection
, u32 barrier_nr
, u32 set_size
)
1315 struct drbd_socket
*sock
;
1316 struct p_barrier_ack
*p
;
1318 if (connection
->cstate
< C_WF_REPORT_PARAMS
)
1321 sock
= &connection
->meta
;
1322 p
= conn_prepare_command(connection
, sock
);
1325 p
->barrier
= barrier_nr
;
1326 p
->set_size
= cpu_to_be32(set_size
);
1327 conn_send_command(connection
, sock
, P_BARRIER_ACK
, sizeof(*p
), NULL
, 0);
1331 * _drbd_send_ack() - Sends an ack packet
1332 * @peer_device: DRBD peer device.
1333 * @cmd: Packet command code.
1334 * @sector: sector, needs to be in big endian byte order
1335 * @blksize: size in byte, needs to be in big endian byte order
1336 * @block_id: Id, big endian byte order
1338 static int _drbd_send_ack(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1339 u64 sector
, u32 blksize
, u64 block_id
)
1341 struct drbd_socket
*sock
;
1342 struct p_block_ack
*p
;
1344 if (peer_device
->device
->state
.conn
< C_CONNECTED
)
1347 sock
= &peer_device
->connection
->meta
;
1348 p
= drbd_prepare_command(peer_device
, sock
);
1352 p
->block_id
= block_id
;
1353 p
->blksize
= blksize
;
1354 p
->seq_num
= cpu_to_be32(atomic_inc_return(&peer_device
->device
->packet_seq
));
1355 return drbd_send_command(peer_device
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1358 /* dp->sector and dp->block_id already/still in network byte order,
1359 * data_size is payload size according to dp->head,
1360 * and may need to be corrected for digest size. */
1361 void drbd_send_ack_dp(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1362 struct p_data
*dp
, int data_size
)
1364 if (peer_device
->connection
->peer_integrity_tfm
)
1365 data_size
-= crypto_shash_digestsize(peer_device
->connection
->peer_integrity_tfm
);
1366 _drbd_send_ack(peer_device
, cmd
, dp
->sector
, cpu_to_be32(data_size
),
1370 void drbd_send_ack_rp(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1371 struct p_block_req
*rp
)
1373 _drbd_send_ack(peer_device
, cmd
, rp
->sector
, rp
->blksize
, rp
->block_id
);
1377 * drbd_send_ack() - Sends an ack packet
1378 * @peer_device: DRBD peer device
1379 * @cmd: packet command code
1380 * @peer_req: peer request
1382 int drbd_send_ack(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1383 struct drbd_peer_request
*peer_req
)
1385 return _drbd_send_ack(peer_device
, cmd
,
1386 cpu_to_be64(peer_req
->i
.sector
),
1387 cpu_to_be32(peer_req
->i
.size
),
1388 peer_req
->block_id
);
1391 /* This function misuses the block_id field to signal if the blocks
1392 * are is sync or not. */
1393 int drbd_send_ack_ex(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1394 sector_t sector
, int blksize
, u64 block_id
)
1396 return _drbd_send_ack(peer_device
, cmd
,
1397 cpu_to_be64(sector
),
1398 cpu_to_be32(blksize
),
1399 cpu_to_be64(block_id
));
1402 int drbd_send_rs_deallocated(struct drbd_peer_device
*peer_device
,
1403 struct drbd_peer_request
*peer_req
)
1405 struct drbd_socket
*sock
;
1406 struct p_block_desc
*p
;
1408 sock
= &peer_device
->connection
->data
;
1409 p
= drbd_prepare_command(peer_device
, sock
);
1412 p
->sector
= cpu_to_be64(peer_req
->i
.sector
);
1413 p
->blksize
= cpu_to_be32(peer_req
->i
.size
);
1415 return drbd_send_command(peer_device
, sock
, P_RS_DEALLOCATED
, sizeof(*p
), NULL
, 0);
1418 int drbd_send_drequest(struct drbd_peer_device
*peer_device
, int cmd
,
1419 sector_t sector
, int size
, u64 block_id
)
1421 struct drbd_socket
*sock
;
1422 struct p_block_req
*p
;
1424 sock
= &peer_device
->connection
->data
;
1425 p
= drbd_prepare_command(peer_device
, sock
);
1428 p
->sector
= cpu_to_be64(sector
);
1429 p
->block_id
= block_id
;
1430 p
->blksize
= cpu_to_be32(size
);
1431 return drbd_send_command(peer_device
, sock
, cmd
, sizeof(*p
), NULL
, 0);
1434 int drbd_send_drequest_csum(struct drbd_peer_device
*peer_device
, sector_t sector
, int size
,
1435 void *digest
, int digest_size
, enum drbd_packet cmd
)
1437 struct drbd_socket
*sock
;
1438 struct p_block_req
*p
;
1440 /* FIXME: Put the digest into the preallocated socket buffer. */
1442 sock
= &peer_device
->connection
->data
;
1443 p
= drbd_prepare_command(peer_device
, sock
);
1446 p
->sector
= cpu_to_be64(sector
);
1447 p
->block_id
= ID_SYNCER
/* unused */;
1448 p
->blksize
= cpu_to_be32(size
);
1449 return drbd_send_command(peer_device
, sock
, cmd
, sizeof(*p
), digest
, digest_size
);
1452 int drbd_send_ov_request(struct drbd_peer_device
*peer_device
, sector_t sector
, int size
)
1454 struct drbd_socket
*sock
;
1455 struct p_block_req
*p
;
1457 sock
= &peer_device
->connection
->data
;
1458 p
= drbd_prepare_command(peer_device
, sock
);
1461 p
->sector
= cpu_to_be64(sector
);
1462 p
->block_id
= ID_SYNCER
/* unused */;
1463 p
->blksize
= cpu_to_be32(size
);
1464 return drbd_send_command(peer_device
, sock
, P_OV_REQUEST
, sizeof(*p
), NULL
, 0);
1467 /* called on sndtimeo
1468 * returns false if we should retry,
1469 * true if we think connection is dead
1471 static int we_should_drop_the_connection(struct drbd_connection
*connection
, struct socket
*sock
)
1474 /* long elapsed = (long)(jiffies - device->last_received); */
1476 drop_it
= connection
->meta
.socket
== sock
1477 || !connection
->ack_receiver
.task
1478 || get_t_state(&connection
->ack_receiver
) != RUNNING
1479 || connection
->cstate
< C_WF_REPORT_PARAMS
;
1484 drop_it
= !--connection
->ko_count
;
1486 drbd_err(connection
, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1487 current
->comm
, current
->pid
, connection
->ko_count
);
1488 request_ping(connection
);
1491 return drop_it
; /* && (device->state == R_PRIMARY) */;
1494 static void drbd_update_congested(struct drbd_connection
*connection
)
1496 struct sock
*sk
= connection
->data
.socket
->sk
;
1497 if (sk
->sk_wmem_queued
> sk
->sk_sndbuf
* 4 / 5)
1498 set_bit(NET_CONGESTED
, &connection
->flags
);
1501 /* The idea of sendpage seems to be to put some kind of reference
1502 * to the page into the skb, and to hand it over to the NIC. In
1503 * this process get_page() gets called.
1505 * As soon as the page was really sent over the network put_page()
1506 * gets called by some part of the network layer. [ NIC driver? ]
1508 * [ get_page() / put_page() increment/decrement the count. If count
1509 * reaches 0 the page will be freed. ]
1511 * This works nicely with pages from FSs.
1512 * But this means that in protocol A we might signal IO completion too early!
1514 * In order not to corrupt data during a resync we must make sure
1515 * that we do not reuse our own buffer pages (EEs) to early, therefore
1516 * we have the net_ee list.
1518 * XFS seems to have problems, still, it submits pages with page_count == 0!
1519 * As a workaround, we disable sendpage on pages
1520 * with page_count == 0 or PageSlab.
1522 static int _drbd_no_send_page(struct drbd_peer_device
*peer_device
, struct page
*page
,
1523 int offset
, size_t size
, unsigned msg_flags
)
1525 struct socket
*socket
;
1529 socket
= peer_device
->connection
->data
.socket
;
1530 addr
= kmap(page
) + offset
;
1531 err
= drbd_send_all(peer_device
->connection
, socket
, addr
, size
, msg_flags
);
1534 peer_device
->device
->send_cnt
+= size
>> 9;
1538 static int _drbd_send_page(struct drbd_peer_device
*peer_device
, struct page
*page
,
1539 int offset
, size_t size
, unsigned msg_flags
)
1541 struct socket
*socket
= peer_device
->connection
->data
.socket
;
1542 struct msghdr msg
= { .msg_flags
= msg_flags
, };
1543 struct bio_vec bvec
;
1547 /* e.g. XFS meta- & log-data is in slab pages, which have a
1548 * page_count of 0 and/or have PageSlab() set.
1549 * we cannot use send_page for those, as that does get_page();
1550 * put_page(); and would cause either a VM_BUG directly, or
1551 * __page_cache_release a page that would actually still be referenced
1552 * by someone, leading to some obscure delayed Oops somewhere else. */
1553 if (!drbd_disable_sendpage
&& sendpage_ok(page
))
1554 msg
.msg_flags
|= MSG_NOSIGNAL
| MSG_SPLICE_PAGES
;
1556 drbd_update_congested(peer_device
->connection
);
1560 bvec_set_page(&bvec
, page
, len
, offset
);
1561 iov_iter_bvec(&msg
.msg_iter
, ITER_SOURCE
, &bvec
, 1, len
);
1563 sent
= sock_sendmsg(socket
, &msg
);
1565 if (sent
== -EAGAIN
) {
1566 if (we_should_drop_the_connection(peer_device
->connection
, socket
))
1570 drbd_warn(peer_device
->device
, "%s: size=%d len=%d sent=%d\n",
1571 __func__
, (int)size
, len
, sent
);
1578 } while (len
> 0 /* THINK && device->cstate >= C_CONNECTED*/);
1579 clear_bit(NET_CONGESTED
, &peer_device
->connection
->flags
);
1583 peer_device
->device
->send_cnt
+= size
>> 9;
1588 static int _drbd_send_bio(struct drbd_peer_device
*peer_device
, struct bio
*bio
)
1590 struct bio_vec bvec
;
1591 struct bvec_iter iter
;
1593 /* hint all but last page with MSG_MORE */
1594 bio_for_each_segment(bvec
, bio
, iter
) {
1597 err
= _drbd_no_send_page(peer_device
, bvec
.bv_page
,
1598 bvec
.bv_offset
, bvec
.bv_len
,
1599 bio_iter_last(bvec
, iter
)
1607 static int _drbd_send_zc_bio(struct drbd_peer_device
*peer_device
, struct bio
*bio
)
1609 struct bio_vec bvec
;
1610 struct bvec_iter iter
;
1612 /* hint all but last page with MSG_MORE */
1613 bio_for_each_segment(bvec
, bio
, iter
) {
1616 err
= _drbd_send_page(peer_device
, bvec
.bv_page
,
1617 bvec
.bv_offset
, bvec
.bv_len
,
1618 bio_iter_last(bvec
, iter
) ? 0 : MSG_MORE
);
1625 static int _drbd_send_zc_ee(struct drbd_peer_device
*peer_device
,
1626 struct drbd_peer_request
*peer_req
)
1628 struct page
*page
= peer_req
->pages
;
1629 unsigned len
= peer_req
->i
.size
;
1632 /* hint all but last page with MSG_MORE */
1633 page_chain_for_each(page
) {
1634 unsigned l
= min_t(unsigned, len
, PAGE_SIZE
);
1636 err
= _drbd_send_page(peer_device
, page
, 0, l
,
1637 page_chain_next(page
) ? MSG_MORE
: 0);
1645 static u32
bio_flags_to_wire(struct drbd_connection
*connection
,
1648 if (connection
->agreed_pro_version
>= 95)
1649 return (bio
->bi_opf
& REQ_SYNC
? DP_RW_SYNC
: 0) |
1650 (bio
->bi_opf
& REQ_FUA
? DP_FUA
: 0) |
1651 (bio
->bi_opf
& REQ_PREFLUSH
? DP_FLUSH
: 0) |
1652 (bio_op(bio
) == REQ_OP_DISCARD
? DP_DISCARD
: 0) |
1653 (bio_op(bio
) == REQ_OP_WRITE_ZEROES
?
1654 ((connection
->agreed_features
& DRBD_FF_WZEROES
) ?
1655 (DP_ZEROES
|(!(bio
->bi_opf
& REQ_NOUNMAP
) ? DP_DISCARD
: 0))
1659 return bio
->bi_opf
& REQ_SYNC
? DP_RW_SYNC
: 0;
1662 /* Used to send write or TRIM aka REQ_OP_DISCARD requests
1663 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1665 int drbd_send_dblock(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
1667 struct drbd_device
*device
= peer_device
->device
;
1668 struct drbd_socket
*sock
;
1671 unsigned int dp_flags
= 0;
1675 sock
= &peer_device
->connection
->data
;
1676 p
= drbd_prepare_command(peer_device
, sock
);
1677 digest_size
= peer_device
->connection
->integrity_tfm
?
1678 crypto_shash_digestsize(peer_device
->connection
->integrity_tfm
) : 0;
1682 p
->sector
= cpu_to_be64(req
->i
.sector
);
1683 p
->block_id
= (unsigned long)req
;
1684 p
->seq_num
= cpu_to_be32(atomic_inc_return(&device
->packet_seq
));
1685 dp_flags
= bio_flags_to_wire(peer_device
->connection
, req
->master_bio
);
1686 if (device
->state
.conn
>= C_SYNC_SOURCE
&&
1687 device
->state
.conn
<= C_PAUSED_SYNC_T
)
1688 dp_flags
|= DP_MAY_SET_IN_SYNC
;
1689 if (peer_device
->connection
->agreed_pro_version
>= 100) {
1690 if (req
->rq_state
& RQ_EXP_RECEIVE_ACK
)
1691 dp_flags
|= DP_SEND_RECEIVE_ACK
;
1692 /* During resync, request an explicit write ack,
1693 * even in protocol != C */
1694 if (req
->rq_state
& RQ_EXP_WRITE_ACK
1695 || (dp_flags
& DP_MAY_SET_IN_SYNC
))
1696 dp_flags
|= DP_SEND_WRITE_ACK
;
1698 p
->dp_flags
= cpu_to_be32(dp_flags
);
1700 if (dp_flags
& (DP_DISCARD
|DP_ZEROES
)) {
1701 enum drbd_packet cmd
= (dp_flags
& DP_ZEROES
) ? P_ZEROES
: P_TRIM
;
1702 struct p_trim
*t
= (struct p_trim
*)p
;
1703 t
->size
= cpu_to_be32(req
->i
.size
);
1704 err
= __send_command(peer_device
->connection
, device
->vnr
, sock
, cmd
, sizeof(*t
), NULL
, 0);
1709 /* our digest is still only over the payload.
1710 * TRIM does not carry any payload. */
1712 drbd_csum_bio(peer_device
->connection
->integrity_tfm
, req
->master_bio
, digest_out
);
1713 err
= __send_command(peer_device
->connection
, device
->vnr
, sock
, P_DATA
,
1714 sizeof(*p
) + digest_size
, NULL
, req
->i
.size
);
1716 /* For protocol A, we have to memcpy the payload into
1717 * socket buffers, as we may complete right away
1718 * as soon as we handed it over to tcp, at which point the data
1719 * pages may become invalid.
1721 * For data-integrity enabled, we copy it as well, so we can be
1722 * sure that even if the bio pages may still be modified, it
1723 * won't change the data on the wire, thus if the digest checks
1724 * out ok after sending on this side, but does not fit on the
1725 * receiving side, we sure have detected corruption elsewhere.
1727 if (!(req
->rq_state
& (RQ_EXP_RECEIVE_ACK
| RQ_EXP_WRITE_ACK
)) || digest_size
)
1728 err
= _drbd_send_bio(peer_device
, req
->master_bio
);
1730 err
= _drbd_send_zc_bio(peer_device
, req
->master_bio
);
1732 /* double check digest, sometimes buffers have been modified in flight. */
1733 if (digest_size
> 0 && digest_size
<= 64) {
1734 /* 64 byte, 512 bit, is the largest digest size
1735 * currently supported in kernel crypto. */
1736 unsigned char digest
[64];
1737 drbd_csum_bio(peer_device
->connection
->integrity_tfm
, req
->master_bio
, digest
);
1738 if (memcmp(p
+ 1, digest
, digest_size
)) {
1740 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1741 (unsigned long long)req
->i
.sector
, req
->i
.size
);
1743 } /* else if (digest_size > 64) {
1744 ... Be noisy about digest too large ...
1748 mutex_unlock(&sock
->mutex
); /* locked by drbd_prepare_command() */
1753 /* answer packet, used to send data back for read requests:
1754 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1755 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1757 int drbd_send_block(struct drbd_peer_device
*peer_device
, enum drbd_packet cmd
,
1758 struct drbd_peer_request
*peer_req
)
1760 struct drbd_device
*device
= peer_device
->device
;
1761 struct drbd_socket
*sock
;
1766 sock
= &peer_device
->connection
->data
;
1767 p
= drbd_prepare_command(peer_device
, sock
);
1769 digest_size
= peer_device
->connection
->integrity_tfm
?
1770 crypto_shash_digestsize(peer_device
->connection
->integrity_tfm
) : 0;
1774 p
->sector
= cpu_to_be64(peer_req
->i
.sector
);
1775 p
->block_id
= peer_req
->block_id
;
1776 p
->seq_num
= 0; /* unused */
1779 drbd_csum_ee(peer_device
->connection
->integrity_tfm
, peer_req
, p
+ 1);
1780 err
= __send_command(peer_device
->connection
, device
->vnr
, sock
, cmd
, sizeof(*p
) + digest_size
, NULL
, peer_req
->i
.size
);
1782 err
= _drbd_send_zc_ee(peer_device
, peer_req
);
1783 mutex_unlock(&sock
->mutex
); /* locked by drbd_prepare_command() */
1788 int drbd_send_out_of_sync(struct drbd_peer_device
*peer_device
, struct drbd_request
*req
)
1790 struct drbd_socket
*sock
;
1791 struct p_block_desc
*p
;
1793 sock
= &peer_device
->connection
->data
;
1794 p
= drbd_prepare_command(peer_device
, sock
);
1797 p
->sector
= cpu_to_be64(req
->i
.sector
);
1798 p
->blksize
= cpu_to_be32(req
->i
.size
);
1799 return drbd_send_command(peer_device
, sock
, P_OUT_OF_SYNC
, sizeof(*p
), NULL
, 0);
1803 drbd_send distinguishes two cases:
1805 Packets sent via the data socket "sock"
1806 and packets sent via the meta data socket "msock"
1809 -----------------+-------------------------+------------------------------
1810 timeout conf.timeout / 2 conf.timeout / 2
1811 timeout action send a ping via msock Abort communication
1812 and close all sockets
1816 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1818 int drbd_send(struct drbd_connection
*connection
, struct socket
*sock
,
1819 void *buf
, size_t size
, unsigned msg_flags
)
1821 struct kvec iov
= {.iov_base
= buf
, .iov_len
= size
};
1822 struct msghdr msg
= {.msg_flags
= msg_flags
| MSG_NOSIGNAL
};
1828 /* THINK if (signal_pending) return ... ? */
1830 iov_iter_kvec(&msg
.msg_iter
, ITER_SOURCE
, &iov
, 1, size
);
1832 if (sock
== connection
->data
.socket
) {
1834 connection
->ko_count
= rcu_dereference(connection
->net_conf
)->ko_count
;
1836 drbd_update_congested(connection
);
1839 rv
= sock_sendmsg(sock
, &msg
);
1840 if (rv
== -EAGAIN
) {
1841 if (we_should_drop_the_connection(connection
, sock
))
1847 flush_signals(current
);
1853 } while (sent
< size
);
1855 if (sock
== connection
->data
.socket
)
1856 clear_bit(NET_CONGESTED
, &connection
->flags
);
1859 if (rv
!= -EAGAIN
) {
1860 drbd_err(connection
, "%s_sendmsg returned %d\n",
1861 sock
== connection
->meta
.socket
? "msock" : "sock",
1863 conn_request_state(connection
, NS(conn
, C_BROKEN_PIPE
), CS_HARD
);
1865 conn_request_state(connection
, NS(conn
, C_TIMEOUT
), CS_HARD
);
1872 * drbd_send_all - Send an entire buffer
1874 * Returns 0 upon success and a negative error value otherwise.
1876 int drbd_send_all(struct drbd_connection
*connection
, struct socket
*sock
, void *buffer
,
1877 size_t size
, unsigned msg_flags
)
1881 err
= drbd_send(connection
, sock
, buffer
, size
, msg_flags
);
1889 static int drbd_open(struct gendisk
*disk
, blk_mode_t mode
)
1891 struct drbd_device
*device
= disk
->private_data
;
1892 unsigned long flags
;
1895 mutex_lock(&drbd_main_mutex
);
1896 spin_lock_irqsave(&device
->resource
->req_lock
, flags
);
1897 /* to have a stable device->state.role
1898 * and no race with updating open_cnt */
1900 if (device
->state
.role
!= R_PRIMARY
) {
1901 if (mode
& BLK_OPEN_WRITE
)
1903 else if (!drbd_allow_oos
)
1909 spin_unlock_irqrestore(&device
->resource
->req_lock
, flags
);
1910 mutex_unlock(&drbd_main_mutex
);
1915 static void drbd_release(struct gendisk
*gd
)
1917 struct drbd_device
*device
= gd
->private_data
;
1919 mutex_lock(&drbd_main_mutex
);
1921 mutex_unlock(&drbd_main_mutex
);
1924 /* need to hold resource->req_lock */
1925 void drbd_queue_unplug(struct drbd_device
*device
)
1927 if (device
->state
.pdsk
>= D_INCONSISTENT
&& device
->state
.conn
>= C_CONNECTED
) {
1928 D_ASSERT(device
, device
->state
.role
== R_PRIMARY
);
1929 if (test_and_clear_bit(UNPLUG_REMOTE
, &device
->flags
)) {
1930 drbd_queue_work_if_unqueued(
1931 &first_peer_device(device
)->connection
->sender_work
,
1932 &device
->unplug_work
);
1937 static void drbd_set_defaults(struct drbd_device
*device
)
1939 /* Beware! The actual layout differs
1940 * between big endian and little endian */
1941 device
->state
= (union drbd_dev_state
) {
1942 { .role
= R_SECONDARY
,
1944 .conn
= C_STANDALONE
,
1950 void drbd_init_set_defaults(struct drbd_device
*device
)
1952 /* the memset(,0,) did most of this.
1953 * note: only assignments, no allocation in here */
1955 drbd_set_defaults(device
);
1957 atomic_set(&device
->ap_bio_cnt
, 0);
1958 atomic_set(&device
->ap_actlog_cnt
, 0);
1959 atomic_set(&device
->ap_pending_cnt
, 0);
1960 atomic_set(&device
->rs_pending_cnt
, 0);
1961 atomic_set(&device
->unacked_cnt
, 0);
1962 atomic_set(&device
->local_cnt
, 0);
1963 atomic_set(&device
->pp_in_use_by_net
, 0);
1964 atomic_set(&device
->rs_sect_in
, 0);
1965 atomic_set(&device
->rs_sect_ev
, 0);
1966 atomic_set(&device
->ap_in_flight
, 0);
1967 atomic_set(&device
->md_io
.in_use
, 0);
1969 mutex_init(&device
->own_state_mutex
);
1970 device
->state_mutex
= &device
->own_state_mutex
;
1972 spin_lock_init(&device
->al_lock
);
1973 spin_lock_init(&device
->peer_seq_lock
);
1975 INIT_LIST_HEAD(&device
->active_ee
);
1976 INIT_LIST_HEAD(&device
->sync_ee
);
1977 INIT_LIST_HEAD(&device
->done_ee
);
1978 INIT_LIST_HEAD(&device
->read_ee
);
1979 INIT_LIST_HEAD(&device
->net_ee
);
1980 INIT_LIST_HEAD(&device
->resync_reads
);
1981 INIT_LIST_HEAD(&device
->resync_work
.list
);
1982 INIT_LIST_HEAD(&device
->unplug_work
.list
);
1983 INIT_LIST_HEAD(&device
->bm_io_work
.w
.list
);
1984 INIT_LIST_HEAD(&device
->pending_master_completion
[0]);
1985 INIT_LIST_HEAD(&device
->pending_master_completion
[1]);
1986 INIT_LIST_HEAD(&device
->pending_completion
[0]);
1987 INIT_LIST_HEAD(&device
->pending_completion
[1]);
1989 device
->resync_work
.cb
= w_resync_timer
;
1990 device
->unplug_work
.cb
= w_send_write_hint
;
1991 device
->bm_io_work
.w
.cb
= w_bitmap_io
;
1993 timer_setup(&device
->resync_timer
, resync_timer_fn
, 0);
1994 timer_setup(&device
->md_sync_timer
, md_sync_timer_fn
, 0);
1995 timer_setup(&device
->start_resync_timer
, start_resync_timer_fn
, 0);
1996 timer_setup(&device
->request_timer
, request_timer_fn
, 0);
1998 init_waitqueue_head(&device
->misc_wait
);
1999 init_waitqueue_head(&device
->state_wait
);
2000 init_waitqueue_head(&device
->ee_wait
);
2001 init_waitqueue_head(&device
->al_wait
);
2002 init_waitqueue_head(&device
->seq_wait
);
2004 device
->resync_wenr
= LC_FREE
;
2005 device
->peer_max_bio_size
= DRBD_MAX_BIO_SIZE_SAFE
;
2006 device
->local_max_bio_size
= DRBD_MAX_BIO_SIZE_SAFE
;
2009 void drbd_set_my_capacity(struct drbd_device
*device
, sector_t size
)
2013 set_capacity_and_notify(device
->vdisk
, size
);
2015 drbd_info(device
, "size = %s (%llu KB)\n",
2016 ppsize(ppb
, size
>>1), (unsigned long long)size
>>1);
2019 void drbd_device_cleanup(struct drbd_device
*device
)
2022 if (first_peer_device(device
)->connection
->receiver
.t_state
!= NONE
)
2023 drbd_err(device
, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2024 first_peer_device(device
)->connection
->receiver
.t_state
);
2026 device
->al_writ_cnt
=
2027 device
->bm_writ_cnt
=
2035 device
->rs_failed
= 0;
2036 device
->rs_last_events
= 0;
2037 device
->rs_last_sect_ev
= 0;
2038 for (i
= 0; i
< DRBD_SYNC_MARKS
; i
++) {
2039 device
->rs_mark_left
[i
] = 0;
2040 device
->rs_mark_time
[i
] = 0;
2042 D_ASSERT(device
, first_peer_device(device
)->connection
->net_conf
== NULL
);
2044 set_capacity_and_notify(device
->vdisk
, 0);
2045 if (device
->bitmap
) {
2046 /* maybe never allocated. */
2047 drbd_bm_resize(device
, 0, 1);
2048 drbd_bm_cleanup(device
);
2051 drbd_backing_dev_free(device
, device
->ldev
);
2052 device
->ldev
= NULL
;
2054 clear_bit(AL_SUSPENDED
, &device
->flags
);
2056 D_ASSERT(device
, list_empty(&device
->active_ee
));
2057 D_ASSERT(device
, list_empty(&device
->sync_ee
));
2058 D_ASSERT(device
, list_empty(&device
->done_ee
));
2059 D_ASSERT(device
, list_empty(&device
->read_ee
));
2060 D_ASSERT(device
, list_empty(&device
->net_ee
));
2061 D_ASSERT(device
, list_empty(&device
->resync_reads
));
2062 D_ASSERT(device
, list_empty(&first_peer_device(device
)->connection
->sender_work
.q
));
2063 D_ASSERT(device
, list_empty(&device
->resync_work
.list
));
2064 D_ASSERT(device
, list_empty(&device
->unplug_work
.list
));
2066 drbd_set_defaults(device
);
2070 static void drbd_destroy_mempools(void)
2074 while (drbd_pp_pool
) {
2075 page
= drbd_pp_pool
;
2076 drbd_pp_pool
= (struct page
*)page_private(page
);
2081 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2083 bioset_exit(&drbd_io_bio_set
);
2084 bioset_exit(&drbd_md_io_bio_set
);
2085 mempool_exit(&drbd_md_io_page_pool
);
2086 mempool_exit(&drbd_ee_mempool
);
2087 mempool_exit(&drbd_request_mempool
);
2088 kmem_cache_destroy(drbd_ee_cache
);
2089 kmem_cache_destroy(drbd_request_cache
);
2090 kmem_cache_destroy(drbd_bm_ext_cache
);
2091 kmem_cache_destroy(drbd_al_ext_cache
);
2093 drbd_ee_cache
= NULL
;
2094 drbd_request_cache
= NULL
;
2095 drbd_bm_ext_cache
= NULL
;
2096 drbd_al_ext_cache
= NULL
;
2101 static int drbd_create_mempools(void)
2104 const int number
= (DRBD_MAX_BIO_SIZE
/PAGE_SIZE
) * drbd_minor_count
;
2108 drbd_request_cache
= kmem_cache_create(
2109 "drbd_req", sizeof(struct drbd_request
), 0, 0, NULL
);
2110 if (drbd_request_cache
== NULL
)
2113 drbd_ee_cache
= kmem_cache_create(
2114 "drbd_ee", sizeof(struct drbd_peer_request
), 0, 0, NULL
);
2115 if (drbd_ee_cache
== NULL
)
2118 drbd_bm_ext_cache
= kmem_cache_create(
2119 "drbd_bm", sizeof(struct bm_extent
), 0, 0, NULL
);
2120 if (drbd_bm_ext_cache
== NULL
)
2123 drbd_al_ext_cache
= kmem_cache_create(
2124 "drbd_al", sizeof(struct lc_element
), 0, 0, NULL
);
2125 if (drbd_al_ext_cache
== NULL
)
2129 ret
= bioset_init(&drbd_io_bio_set
, BIO_POOL_SIZE
, 0, 0);
2133 ret
= bioset_init(&drbd_md_io_bio_set
, DRBD_MIN_POOL_PAGES
, 0,
2138 ret
= mempool_init_page_pool(&drbd_md_io_page_pool
, DRBD_MIN_POOL_PAGES
, 0);
2142 ret
= mempool_init_slab_pool(&drbd_request_mempool
, number
,
2143 drbd_request_cache
);
2147 ret
= mempool_init_slab_pool(&drbd_ee_mempool
, number
, drbd_ee_cache
);
2151 for (i
= 0; i
< number
; i
++) {
2152 page
= alloc_page(GFP_HIGHUSER
);
2155 set_page_private(page
, (unsigned long)drbd_pp_pool
);
2156 drbd_pp_pool
= page
;
2158 drbd_pp_vacant
= number
;
2163 drbd_destroy_mempools(); /* in case we allocated some */
2167 static void drbd_release_all_peer_reqs(struct drbd_device
*device
)
2171 rr
= drbd_free_peer_reqs(device
, &device
->active_ee
);
2173 drbd_err(device
, "%d EEs in active list found!\n", rr
);
2175 rr
= drbd_free_peer_reqs(device
, &device
->sync_ee
);
2177 drbd_err(device
, "%d EEs in sync list found!\n", rr
);
2179 rr
= drbd_free_peer_reqs(device
, &device
->read_ee
);
2181 drbd_err(device
, "%d EEs in read list found!\n", rr
);
2183 rr
= drbd_free_peer_reqs(device
, &device
->done_ee
);
2185 drbd_err(device
, "%d EEs in done list found!\n", rr
);
2187 rr
= drbd_free_peer_reqs(device
, &device
->net_ee
);
2189 drbd_err(device
, "%d EEs in net list found!\n", rr
);
2192 /* caution. no locking. */
2193 void drbd_destroy_device(struct kref
*kref
)
2195 struct drbd_device
*device
= container_of(kref
, struct drbd_device
, kref
);
2196 struct drbd_resource
*resource
= device
->resource
;
2197 struct drbd_peer_device
*peer_device
, *tmp_peer_device
;
2199 timer_shutdown_sync(&device
->request_timer
);
2201 /* paranoia asserts */
2202 D_ASSERT(device
, device
->open_cnt
== 0);
2203 /* end paranoia asserts */
2205 /* cleanup stuff that may have been allocated during
2206 * device (re-)configuration or state changes */
2208 drbd_backing_dev_free(device
, device
->ldev
);
2209 device
->ldev
= NULL
;
2211 drbd_release_all_peer_reqs(device
);
2213 lc_destroy(device
->act_log
);
2214 lc_destroy(device
->resync
);
2216 kfree(device
->p_uuid
);
2217 /* device->p_uuid = NULL; */
2219 if (device
->bitmap
) /* should no longer be there. */
2220 drbd_bm_cleanup(device
);
2221 __free_page(device
->md_io
.page
);
2222 put_disk(device
->vdisk
);
2223 kfree(device
->rs_plan_s
);
2225 /* not for_each_connection(connection, resource):
2226 * those may have been cleaned up and disassociated already.
2228 for_each_peer_device_safe(peer_device
, tmp_peer_device
, device
) {
2229 kref_put(&peer_device
->connection
->kref
, drbd_destroy_connection
);
2232 if (device
->submit
.wq
)
2233 destroy_workqueue(device
->submit
.wq
);
2235 kref_put(&resource
->kref
, drbd_destroy_resource
);
2238 /* One global retry thread, if we need to push back some bio and have it
2239 * reinserted through our make request function.
2241 static struct retry_worker
{
2242 struct workqueue_struct
*wq
;
2243 struct work_struct worker
;
2246 struct list_head writes
;
2249 static void do_retry(struct work_struct
*ws
)
2251 struct retry_worker
*retry
= container_of(ws
, struct retry_worker
, worker
);
2253 struct drbd_request
*req
, *tmp
;
2255 spin_lock_irq(&retry
->lock
);
2256 list_splice_init(&retry
->writes
, &writes
);
2257 spin_unlock_irq(&retry
->lock
);
2259 list_for_each_entry_safe(req
, tmp
, &writes
, tl_requests
) {
2260 struct drbd_device
*device
= req
->device
;
2261 struct bio
*bio
= req
->master_bio
;
2265 expect(device
, atomic_read(&req
->completion_ref
) == 0) &&
2266 expect(device
, req
->rq_state
& RQ_POSTPONED
) &&
2267 expect(device
, (req
->rq_state
& RQ_LOCAL_PENDING
) == 0 ||
2268 (req
->rq_state
& RQ_LOCAL_ABORTED
) != 0);
2271 drbd_err(device
, "req=%p completion_ref=%d rq_state=%x\n",
2272 req
, atomic_read(&req
->completion_ref
),
2275 /* We still need to put one kref associated with the
2276 * "completion_ref" going zero in the code path that queued it
2277 * here. The request object may still be referenced by a
2278 * frozen local req->private_bio, in case we force-detached.
2280 kref_put(&req
->kref
, drbd_req_destroy
);
2282 /* A single suspended or otherwise blocking device may stall
2283 * all others as well. Fortunately, this code path is to
2284 * recover from a situation that "should not happen":
2285 * concurrent writes in multi-primary setup.
2286 * In a "normal" lifecycle, this workqueue is supposed to be
2287 * destroyed without ever doing anything.
2288 * If it turns out to be an issue anyways, we can do per
2289 * resource (replication group) or per device (minor) retry
2290 * workqueues instead.
2293 /* We are not just doing submit_bio_noacct(),
2294 * as we want to keep the start_time information. */
2296 __drbd_make_request(device
, bio
);
2300 /* called via drbd_req_put_completion_ref(),
2301 * holds resource->req_lock */
2302 void drbd_restart_request(struct drbd_request
*req
)
2304 unsigned long flags
;
2305 spin_lock_irqsave(&retry
.lock
, flags
);
2306 list_move_tail(&req
->tl_requests
, &retry
.writes
);
2307 spin_unlock_irqrestore(&retry
.lock
, flags
);
2309 /* Drop the extra reference that would otherwise
2310 * have been dropped by complete_master_bio.
2311 * do_retry() needs to grab a new one. */
2312 dec_ap_bio(req
->device
);
2314 queue_work(retry
.wq
, &retry
.worker
);
2317 void drbd_destroy_resource(struct kref
*kref
)
2319 struct drbd_resource
*resource
=
2320 container_of(kref
, struct drbd_resource
, kref
);
2322 idr_destroy(&resource
->devices
);
2323 free_cpumask_var(resource
->cpu_mask
);
2324 kfree(resource
->name
);
2328 void drbd_free_resource(struct drbd_resource
*resource
)
2330 struct drbd_connection
*connection
, *tmp
;
2332 for_each_connection_safe(connection
, tmp
, resource
) {
2333 list_del(&connection
->connections
);
2334 drbd_debugfs_connection_cleanup(connection
);
2335 kref_put(&connection
->kref
, drbd_destroy_connection
);
2337 drbd_debugfs_resource_cleanup(resource
);
2338 kref_put(&resource
->kref
, drbd_destroy_resource
);
2341 static void drbd_cleanup(void)
2344 struct drbd_device
*device
;
2345 struct drbd_resource
*resource
, *tmp
;
2347 /* first remove proc,
2348 * drbdsetup uses it's presence to detect
2349 * whether DRBD is loaded.
2350 * If we would get stuck in proc removal,
2351 * but have netlink already deregistered,
2352 * some drbdsetup commands may wait forever
2356 remove_proc_entry("drbd", NULL
);
2359 destroy_workqueue(retry
.wq
);
2361 drbd_genl_unregister();
2363 idr_for_each_entry(&drbd_devices
, device
, i
)
2364 drbd_delete_device(device
);
2366 /* not _rcu since, no other updater anymore. Genl already unregistered */
2367 for_each_resource_safe(resource
, tmp
, &drbd_resources
) {
2368 list_del(&resource
->resources
);
2369 drbd_free_resource(resource
);
2372 drbd_debugfs_cleanup();
2374 drbd_destroy_mempools();
2375 unregister_blkdev(DRBD_MAJOR
, "drbd");
2377 idr_destroy(&drbd_devices
);
2379 pr_info("module cleanup done.\n");
2382 static void drbd_init_workqueue(struct drbd_work_queue
* wq
)
2384 spin_lock_init(&wq
->q_lock
);
2385 INIT_LIST_HEAD(&wq
->q
);
2386 init_waitqueue_head(&wq
->q_wait
);
2389 struct completion_work
{
2391 struct completion done
;
2394 static int w_complete(struct drbd_work
*w
, int cancel
)
2396 struct completion_work
*completion_work
=
2397 container_of(w
, struct completion_work
, w
);
2399 complete(&completion_work
->done
);
2403 void drbd_flush_workqueue(struct drbd_work_queue
*work_queue
)
2405 struct completion_work completion_work
;
2407 completion_work
.w
.cb
= w_complete
;
2408 init_completion(&completion_work
.done
);
2409 drbd_queue_work(work_queue
, &completion_work
.w
);
2410 wait_for_completion(&completion_work
.done
);
2413 struct drbd_resource
*drbd_find_resource(const char *name
)
2415 struct drbd_resource
*resource
;
2417 if (!name
|| !name
[0])
2421 for_each_resource_rcu(resource
, &drbd_resources
) {
2422 if (!strcmp(resource
->name
, name
)) {
2423 kref_get(&resource
->kref
);
2433 struct drbd_connection
*conn_get_by_addrs(void *my_addr
, int my_addr_len
,
2434 void *peer_addr
, int peer_addr_len
)
2436 struct drbd_resource
*resource
;
2437 struct drbd_connection
*connection
;
2440 for_each_resource_rcu(resource
, &drbd_resources
) {
2441 for_each_connection_rcu(connection
, resource
) {
2442 if (connection
->my_addr_len
== my_addr_len
&&
2443 connection
->peer_addr_len
== peer_addr_len
&&
2444 !memcmp(&connection
->my_addr
, my_addr
, my_addr_len
) &&
2445 !memcmp(&connection
->peer_addr
, peer_addr
, peer_addr_len
)) {
2446 kref_get(&connection
->kref
);
2457 static int drbd_alloc_socket(struct drbd_socket
*socket
)
2459 socket
->rbuf
= (void *) __get_free_page(GFP_KERNEL
);
2462 socket
->sbuf
= (void *) __get_free_page(GFP_KERNEL
);
2468 static void drbd_free_socket(struct drbd_socket
*socket
)
2470 free_page((unsigned long) socket
->sbuf
);
2471 free_page((unsigned long) socket
->rbuf
);
2474 void conn_free_crypto(struct drbd_connection
*connection
)
2476 drbd_free_sock(connection
);
2478 crypto_free_shash(connection
->csums_tfm
);
2479 crypto_free_shash(connection
->verify_tfm
);
2480 crypto_free_shash(connection
->cram_hmac_tfm
);
2481 crypto_free_shash(connection
->integrity_tfm
);
2482 crypto_free_shash(connection
->peer_integrity_tfm
);
2483 kfree(connection
->int_dig_in
);
2484 kfree(connection
->int_dig_vv
);
2486 connection
->csums_tfm
= NULL
;
2487 connection
->verify_tfm
= NULL
;
2488 connection
->cram_hmac_tfm
= NULL
;
2489 connection
->integrity_tfm
= NULL
;
2490 connection
->peer_integrity_tfm
= NULL
;
2491 connection
->int_dig_in
= NULL
;
2492 connection
->int_dig_vv
= NULL
;
2495 int set_resource_options(struct drbd_resource
*resource
, struct res_opts
*res_opts
)
2497 struct drbd_connection
*connection
;
2498 cpumask_var_t new_cpu_mask
;
2501 if (!zalloc_cpumask_var(&new_cpu_mask
, GFP_KERNEL
))
2504 /* silently ignore cpu mask on UP kernel */
2505 if (nr_cpu_ids
> 1 && res_opts
->cpu_mask
[0] != 0) {
2506 err
= bitmap_parse(res_opts
->cpu_mask
, DRBD_CPU_MASK_SIZE
,
2507 cpumask_bits(new_cpu_mask
), nr_cpu_ids
);
2508 if (err
== -EOVERFLOW
) {
2509 /* So what. mask it out. */
2510 cpumask_var_t tmp_cpu_mask
;
2511 if (zalloc_cpumask_var(&tmp_cpu_mask
, GFP_KERNEL
)) {
2512 cpumask_setall(tmp_cpu_mask
);
2513 cpumask_and(new_cpu_mask
, new_cpu_mask
, tmp_cpu_mask
);
2514 drbd_warn(resource
, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2516 strlen(res_opts
->cpu_mask
) > 12 ? "..." : "",
2518 free_cpumask_var(tmp_cpu_mask
);
2523 drbd_warn(resource
, "bitmap_parse() failed with %d\n", err
);
2524 /* retcode = ERR_CPU_MASK_PARSE; */
2528 resource
->res_opts
= *res_opts
;
2529 if (cpumask_empty(new_cpu_mask
))
2530 drbd_calc_cpu_mask(&new_cpu_mask
);
2531 if (!cpumask_equal(resource
->cpu_mask
, new_cpu_mask
)) {
2532 cpumask_copy(resource
->cpu_mask
, new_cpu_mask
);
2533 for_each_connection_rcu(connection
, resource
) {
2534 connection
->receiver
.reset_cpu_mask
= 1;
2535 connection
->ack_receiver
.reset_cpu_mask
= 1;
2536 connection
->worker
.reset_cpu_mask
= 1;
2542 free_cpumask_var(new_cpu_mask
);
2547 struct drbd_resource
*drbd_create_resource(const char *name
)
2549 struct drbd_resource
*resource
;
2551 resource
= kzalloc(sizeof(struct drbd_resource
), GFP_KERNEL
);
2554 resource
->name
= kstrdup(name
, GFP_KERNEL
);
2555 if (!resource
->name
)
2556 goto fail_free_resource
;
2557 if (!zalloc_cpumask_var(&resource
->cpu_mask
, GFP_KERNEL
))
2558 goto fail_free_name
;
2559 kref_init(&resource
->kref
);
2560 idr_init(&resource
->devices
);
2561 INIT_LIST_HEAD(&resource
->connections
);
2562 resource
->write_ordering
= WO_BDEV_FLUSH
;
2563 list_add_tail_rcu(&resource
->resources
, &drbd_resources
);
2564 mutex_init(&resource
->conf_update
);
2565 mutex_init(&resource
->adm_mutex
);
2566 spin_lock_init(&resource
->req_lock
);
2567 drbd_debugfs_resource_add(resource
);
2571 kfree(resource
->name
);
2578 /* caller must be under adm_mutex */
2579 struct drbd_connection
*conn_create(const char *name
, struct res_opts
*res_opts
)
2581 struct drbd_resource
*resource
;
2582 struct drbd_connection
*connection
;
2584 connection
= kzalloc(sizeof(struct drbd_connection
), GFP_KERNEL
);
2588 if (drbd_alloc_socket(&connection
->data
))
2590 if (drbd_alloc_socket(&connection
->meta
))
2593 connection
->current_epoch
= kzalloc(sizeof(struct drbd_epoch
), GFP_KERNEL
);
2594 if (!connection
->current_epoch
)
2597 INIT_LIST_HEAD(&connection
->transfer_log
);
2599 INIT_LIST_HEAD(&connection
->current_epoch
->list
);
2600 connection
->epochs
= 1;
2601 spin_lock_init(&connection
->epoch_lock
);
2603 connection
->send
.seen_any_write_yet
= false;
2604 connection
->send
.current_epoch_nr
= 0;
2605 connection
->send
.current_epoch_writes
= 0;
2607 resource
= drbd_create_resource(name
);
2611 connection
->cstate
= C_STANDALONE
;
2612 mutex_init(&connection
->cstate_mutex
);
2613 init_waitqueue_head(&connection
->ping_wait
);
2614 idr_init(&connection
->peer_devices
);
2616 drbd_init_workqueue(&connection
->sender_work
);
2617 mutex_init(&connection
->data
.mutex
);
2618 mutex_init(&connection
->meta
.mutex
);
2620 drbd_thread_init(resource
, &connection
->receiver
, drbd_receiver
, "receiver");
2621 connection
->receiver
.connection
= connection
;
2622 drbd_thread_init(resource
, &connection
->worker
, drbd_worker
, "worker");
2623 connection
->worker
.connection
= connection
;
2624 drbd_thread_init(resource
, &connection
->ack_receiver
, drbd_ack_receiver
, "ack_recv");
2625 connection
->ack_receiver
.connection
= connection
;
2627 kref_init(&connection
->kref
);
2629 connection
->resource
= resource
;
2631 if (set_resource_options(resource
, res_opts
))
2634 kref_get(&resource
->kref
);
2635 list_add_tail_rcu(&connection
->connections
, &resource
->connections
);
2636 drbd_debugfs_connection_add(connection
);
2640 list_del(&resource
->resources
);
2641 drbd_free_resource(resource
);
2643 kfree(connection
->current_epoch
);
2644 drbd_free_socket(&connection
->meta
);
2645 drbd_free_socket(&connection
->data
);
2650 void drbd_destroy_connection(struct kref
*kref
)
2652 struct drbd_connection
*connection
= container_of(kref
, struct drbd_connection
, kref
);
2653 struct drbd_resource
*resource
= connection
->resource
;
2655 if (atomic_read(&connection
->current_epoch
->epoch_size
) != 0)
2656 drbd_err(connection
, "epoch_size:%d\n", atomic_read(&connection
->current_epoch
->epoch_size
));
2657 kfree(connection
->current_epoch
);
2659 idr_destroy(&connection
->peer_devices
);
2661 drbd_free_socket(&connection
->meta
);
2662 drbd_free_socket(&connection
->data
);
2663 kfree(connection
->int_dig_in
);
2664 kfree(connection
->int_dig_vv
);
2666 kref_put(&resource
->kref
, drbd_destroy_resource
);
2669 static int init_submitter(struct drbd_device
*device
)
2671 /* opencoded create_singlethread_workqueue(),
2672 * to be able to say "drbd%d", ..., minor */
2674 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM
, device
->minor
);
2675 if (!device
->submit
.wq
)
2678 INIT_WORK(&device
->submit
.worker
, do_submit
);
2679 INIT_LIST_HEAD(&device
->submit
.writes
);
2683 enum drbd_ret_code
drbd_create_device(struct drbd_config_context
*adm_ctx
, unsigned int minor
)
2685 struct drbd_resource
*resource
= adm_ctx
->resource
;
2686 struct drbd_connection
*connection
, *n
;
2687 struct drbd_device
*device
;
2688 struct drbd_peer_device
*peer_device
, *tmp_peer_device
;
2689 struct gendisk
*disk
;
2691 int vnr
= adm_ctx
->volume
;
2692 enum drbd_ret_code err
= ERR_NOMEM
;
2693 struct queue_limits lim
= {
2695 * Setting the max_hw_sectors to an odd value of 8kibyte here.
2696 * This triggers a max_bio_size message upon first attach or
2699 .max_hw_sectors
= DRBD_MAX_BIO_SIZE_SAFE
>> 8,
2702 device
= minor_to_device(minor
);
2704 return ERR_MINOR_OR_VOLUME_EXISTS
;
2706 /* GFP_KERNEL, we are outside of all write-out paths */
2707 device
= kzalloc(sizeof(struct drbd_device
), GFP_KERNEL
);
2710 kref_init(&device
->kref
);
2712 kref_get(&resource
->kref
);
2713 device
->resource
= resource
;
2714 device
->minor
= minor
;
2717 drbd_init_set_defaults(device
);
2719 disk
= blk_alloc_disk(&lim
, NUMA_NO_NODE
);
2721 err
= PTR_ERR(disk
);
2725 device
->vdisk
= disk
;
2726 device
->rq_queue
= disk
->queue
;
2728 set_disk_ro(disk
, true);
2730 disk
->major
= DRBD_MAJOR
;
2731 disk
->first_minor
= minor
;
2733 disk
->fops
= &drbd_ops
;
2734 disk
->flags
|= GENHD_FL_NO_PART
;
2735 sprintf(disk
->disk_name
, "drbd%d", minor
);
2736 disk
->private_data
= device
;
2738 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES
, disk
->queue
);
2739 blk_queue_write_cache(disk
->queue
, true, true);
2741 device
->md_io
.page
= alloc_page(GFP_KERNEL
);
2742 if (!device
->md_io
.page
)
2743 goto out_no_io_page
;
2745 if (drbd_bm_init(device
))
2747 device
->read_requests
= RB_ROOT
;
2748 device
->write_requests
= RB_ROOT
;
2750 id
= idr_alloc(&drbd_devices
, device
, minor
, minor
+ 1, GFP_KERNEL
);
2753 err
= ERR_MINOR_OR_VOLUME_EXISTS
;
2754 goto out_no_minor_idr
;
2756 kref_get(&device
->kref
);
2758 id
= idr_alloc(&resource
->devices
, device
, vnr
, vnr
+ 1, GFP_KERNEL
);
2761 err
= ERR_MINOR_OR_VOLUME_EXISTS
;
2762 goto out_idr_remove_minor
;
2764 kref_get(&device
->kref
);
2766 INIT_LIST_HEAD(&device
->peer_devices
);
2767 INIT_LIST_HEAD(&device
->pending_bitmap_io
);
2768 for_each_connection(connection
, resource
) {
2769 peer_device
= kzalloc(sizeof(struct drbd_peer_device
), GFP_KERNEL
);
2771 goto out_idr_remove_from_resource
;
2772 peer_device
->connection
= connection
;
2773 peer_device
->device
= device
;
2775 list_add(&peer_device
->peer_devices
, &device
->peer_devices
);
2776 kref_get(&device
->kref
);
2778 id
= idr_alloc(&connection
->peer_devices
, peer_device
, vnr
, vnr
+ 1, GFP_KERNEL
);
2781 err
= ERR_INVALID_REQUEST
;
2782 goto out_idr_remove_from_resource
;
2784 kref_get(&connection
->kref
);
2785 INIT_WORK(&peer_device
->send_acks_work
, drbd_send_acks_wf
);
2788 if (init_submitter(device
)) {
2790 goto out_idr_remove_from_resource
;
2793 err
= add_disk(disk
);
2795 goto out_destroy_workqueue
;
2797 /* inherit the connection state */
2798 device
->state
.conn
= first_connection(resource
)->cstate
;
2799 if (device
->state
.conn
== C_WF_REPORT_PARAMS
) {
2800 for_each_peer_device(peer_device
, device
)
2801 drbd_connected(peer_device
);
2803 /* move to create_peer_device() */
2804 for_each_peer_device(peer_device
, device
)
2805 drbd_debugfs_peer_device_add(peer_device
);
2806 drbd_debugfs_device_add(device
);
2809 out_destroy_workqueue
:
2810 destroy_workqueue(device
->submit
.wq
);
2811 out_idr_remove_from_resource
:
2812 for_each_connection_safe(connection
, n
, resource
) {
2813 peer_device
= idr_remove(&connection
->peer_devices
, vnr
);
2815 kref_put(&connection
->kref
, drbd_destroy_connection
);
2817 for_each_peer_device_safe(peer_device
, tmp_peer_device
, device
) {
2818 list_del(&peer_device
->peer_devices
);
2821 idr_remove(&resource
->devices
, vnr
);
2822 out_idr_remove_minor
:
2823 idr_remove(&drbd_devices
, minor
);
2826 drbd_bm_cleanup(device
);
2828 __free_page(device
->md_io
.page
);
2832 kref_put(&resource
->kref
, drbd_destroy_resource
);
2837 void drbd_delete_device(struct drbd_device
*device
)
2839 struct drbd_resource
*resource
= device
->resource
;
2840 struct drbd_connection
*connection
;
2841 struct drbd_peer_device
*peer_device
;
2843 /* move to free_peer_device() */
2844 for_each_peer_device(peer_device
, device
)
2845 drbd_debugfs_peer_device_cleanup(peer_device
);
2846 drbd_debugfs_device_cleanup(device
);
2847 for_each_connection(connection
, resource
) {
2848 idr_remove(&connection
->peer_devices
, device
->vnr
);
2849 kref_put(&device
->kref
, drbd_destroy_device
);
2851 idr_remove(&resource
->devices
, device
->vnr
);
2852 kref_put(&device
->kref
, drbd_destroy_device
);
2853 idr_remove(&drbd_devices
, device_to_minor(device
));
2854 kref_put(&device
->kref
, drbd_destroy_device
);
2855 del_gendisk(device
->vdisk
);
2857 kref_put(&device
->kref
, drbd_destroy_device
);
2860 static int __init
drbd_init(void)
2864 if (drbd_minor_count
< DRBD_MINOR_COUNT_MIN
|| drbd_minor_count
> DRBD_MINOR_COUNT_MAX
) {
2865 pr_err("invalid minor_count (%d)\n", drbd_minor_count
);
2869 drbd_minor_count
= DRBD_MINOR_COUNT_DEF
;
2873 err
= register_blkdev(DRBD_MAJOR
, "drbd");
2875 pr_err("unable to register block device major %d\n",
2881 * allocate all necessary structs
2883 init_waitqueue_head(&drbd_pp_wait
);
2885 drbd_proc
= NULL
; /* play safe for drbd_cleanup */
2886 idr_init(&drbd_devices
);
2888 mutex_init(&resources_mutex
);
2889 INIT_LIST_HEAD(&drbd_resources
);
2891 err
= drbd_genl_register();
2893 pr_err("unable to register generic netlink family\n");
2897 err
= drbd_create_mempools();
2902 drbd_proc
= proc_create_single("drbd", S_IFREG
| 0444 , NULL
, drbd_seq_show
);
2904 pr_err("unable to register proc file\n");
2908 retry
.wq
= create_singlethread_workqueue("drbd-reissue");
2910 pr_err("unable to create retry workqueue\n");
2913 INIT_WORK(&retry
.worker
, do_retry
);
2914 spin_lock_init(&retry
.lock
);
2915 INIT_LIST_HEAD(&retry
.writes
);
2917 drbd_debugfs_init();
2919 pr_info("initialized. "
2920 "Version: " REL_VERSION
" (api:%d/proto:%d-%d)\n",
2921 GENL_MAGIC_VERSION
, PRO_VERSION_MIN
, PRO_VERSION_MAX
);
2922 pr_info("%s\n", drbd_buildtag());
2923 pr_info("registered as block device major %d\n", DRBD_MAJOR
);
2924 return 0; /* Success! */
2929 pr_err("ran out of memory\n");
2931 pr_err("initialization failure\n");
2935 static void drbd_free_one_sock(struct drbd_socket
*ds
)
2938 mutex_lock(&ds
->mutex
);
2941 mutex_unlock(&ds
->mutex
);
2943 /* so debugfs does not need to mutex_lock() */
2945 kernel_sock_shutdown(s
, SHUT_RDWR
);
2950 void drbd_free_sock(struct drbd_connection
*connection
)
2952 if (connection
->data
.socket
)
2953 drbd_free_one_sock(&connection
->data
);
2954 if (connection
->meta
.socket
)
2955 drbd_free_one_sock(&connection
->meta
);
2958 /* meta data management */
2960 void conn_md_sync(struct drbd_connection
*connection
)
2962 struct drbd_peer_device
*peer_device
;
2966 idr_for_each_entry(&connection
->peer_devices
, peer_device
, vnr
) {
2967 struct drbd_device
*device
= peer_device
->device
;
2969 kref_get(&device
->kref
);
2971 drbd_md_sync(device
);
2972 kref_put(&device
->kref
, drbd_destroy_device
);
2978 /* aligned 4kByte */
2979 struct meta_data_on_disk
{
2980 u64 la_size_sect
; /* last agreed size. */
2981 u64 uuid
[UI_SIZE
]; /* UUIDs. */
2984 u32 flags
; /* MDF */
2987 u32 al_offset
; /* offset to this block */
2988 u32 al_nr_extents
; /* important for restoring the AL (userspace) */
2989 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2990 u32 bm_offset
; /* offset to the bitmap, from here */
2991 u32 bm_bytes_per_bit
; /* BM_BLOCK_SIZE */
2992 u32 la_peer_max_bio_size
; /* last peer max_bio_size */
2994 /* see al_tr_number_to_on_disk_sector() */
2996 u32 al_stripe_size_4k
;
2998 u8 reserved_u8
[4096 - (7*8 + 10*4)];
3003 void drbd_md_write(struct drbd_device
*device
, void *b
)
3005 struct meta_data_on_disk
*buffer
= b
;
3009 memset(buffer
, 0, sizeof(*buffer
));
3011 buffer
->la_size_sect
= cpu_to_be64(get_capacity(device
->vdisk
));
3012 for (i
= UI_CURRENT
; i
< UI_SIZE
; i
++)
3013 buffer
->uuid
[i
] = cpu_to_be64(device
->ldev
->md
.uuid
[i
]);
3014 buffer
->flags
= cpu_to_be32(device
->ldev
->md
.flags
);
3015 buffer
->magic
= cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN
);
3017 buffer
->md_size_sect
= cpu_to_be32(device
->ldev
->md
.md_size_sect
);
3018 buffer
->al_offset
= cpu_to_be32(device
->ldev
->md
.al_offset
);
3019 buffer
->al_nr_extents
= cpu_to_be32(device
->act_log
->nr_elements
);
3020 buffer
->bm_bytes_per_bit
= cpu_to_be32(BM_BLOCK_SIZE
);
3021 buffer
->device_uuid
= cpu_to_be64(device
->ldev
->md
.device_uuid
);
3023 buffer
->bm_offset
= cpu_to_be32(device
->ldev
->md
.bm_offset
);
3024 buffer
->la_peer_max_bio_size
= cpu_to_be32(device
->peer_max_bio_size
);
3026 buffer
->al_stripes
= cpu_to_be32(device
->ldev
->md
.al_stripes
);
3027 buffer
->al_stripe_size_4k
= cpu_to_be32(device
->ldev
->md
.al_stripe_size_4k
);
3029 D_ASSERT(device
, drbd_md_ss(device
->ldev
) == device
->ldev
->md
.md_offset
);
3030 sector
= device
->ldev
->md
.md_offset
;
3032 if (drbd_md_sync_page_io(device
, device
->ldev
, sector
, REQ_OP_WRITE
)) {
3033 /* this was a try anyways ... */
3034 drbd_err(device
, "meta data update failed!\n");
3035 drbd_chk_io_error(device
, 1, DRBD_META_IO_ERROR
);
3040 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3041 * @device: DRBD device.
3043 void drbd_md_sync(struct drbd_device
*device
)
3045 struct meta_data_on_disk
*buffer
;
3047 /* Don't accidentally change the DRBD meta data layout. */
3048 BUILD_BUG_ON(UI_SIZE
!= 4);
3049 BUILD_BUG_ON(sizeof(struct meta_data_on_disk
) != 4096);
3051 del_timer(&device
->md_sync_timer
);
3052 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3053 if (!test_and_clear_bit(MD_DIRTY
, &device
->flags
))
3056 /* We use here D_FAILED and not D_ATTACHING because we try to write
3057 * metadata even if we detach due to a disk failure! */
3058 if (!get_ldev_if_state(device
, D_FAILED
))
3061 buffer
= drbd_md_get_buffer(device
, __func__
);
3065 drbd_md_write(device
, buffer
);
3067 /* Update device->ldev->md.la_size_sect,
3068 * since we updated it on metadata. */
3069 device
->ldev
->md
.la_size_sect
= get_capacity(device
->vdisk
);
3071 drbd_md_put_buffer(device
);
3076 static int check_activity_log_stripe_size(struct drbd_device
*device
,
3077 struct meta_data_on_disk
*on_disk
,
3078 struct drbd_md
*in_core
)
3080 u32 al_stripes
= be32_to_cpu(on_disk
->al_stripes
);
3081 u32 al_stripe_size_4k
= be32_to_cpu(on_disk
->al_stripe_size_4k
);
3084 /* both not set: default to old fixed size activity log */
3085 if (al_stripes
== 0 && al_stripe_size_4k
== 0) {
3087 al_stripe_size_4k
= MD_32kB_SECT
/8;
3090 /* some paranoia plausibility checks */
3092 /* we need both values to be set */
3093 if (al_stripes
== 0 || al_stripe_size_4k
== 0)
3096 al_size_4k
= (u64
)al_stripes
* al_stripe_size_4k
;
3098 /* Upper limit of activity log area, to avoid potential overflow
3099 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3100 * than 72 * 4k blocks total only increases the amount of history,
3101 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3102 if (al_size_4k
> (16 * 1024 * 1024/4))
3105 /* Lower limit: we need at least 8 transaction slots (32kB)
3106 * to not break existing setups */
3107 if (al_size_4k
< MD_32kB_SECT
/8)
3110 in_core
->al_stripe_size_4k
= al_stripe_size_4k
;
3111 in_core
->al_stripes
= al_stripes
;
3112 in_core
->al_size_4k
= al_size_4k
;
3116 drbd_err(device
, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3117 al_stripes
, al_stripe_size_4k
);
3121 static int check_offsets_and_sizes(struct drbd_device
*device
, struct drbd_backing_dev
*bdev
)
3123 sector_t capacity
= drbd_get_capacity(bdev
->md_bdev
);
3124 struct drbd_md
*in_core
= &bdev
->md
;
3125 s32 on_disk_al_sect
;
3126 s32 on_disk_bm_sect
;
3128 /* The on-disk size of the activity log, calculated from offsets, and
3129 * the size of the activity log calculated from the stripe settings,
3131 * Though we could relax this a bit: it is ok, if the striped activity log
3132 * fits in the available on-disk activity log size.
3133 * Right now, that would break how resize is implemented.
3134 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3135 * of possible unused padding space in the on disk layout. */
3136 if (in_core
->al_offset
< 0) {
3137 if (in_core
->bm_offset
> in_core
->al_offset
)
3139 on_disk_al_sect
= -in_core
->al_offset
;
3140 on_disk_bm_sect
= in_core
->al_offset
- in_core
->bm_offset
;
3142 if (in_core
->al_offset
!= MD_4kB_SECT
)
3144 if (in_core
->bm_offset
< in_core
->al_offset
+ in_core
->al_size_4k
* MD_4kB_SECT
)
3147 on_disk_al_sect
= in_core
->bm_offset
- MD_4kB_SECT
;
3148 on_disk_bm_sect
= in_core
->md_size_sect
- in_core
->bm_offset
;
3151 /* old fixed size meta data is exactly that: fixed. */
3152 if (in_core
->meta_dev_idx
>= 0) {
3153 if (in_core
->md_size_sect
!= MD_128MB_SECT
3154 || in_core
->al_offset
!= MD_4kB_SECT
3155 || in_core
->bm_offset
!= MD_4kB_SECT
+ MD_32kB_SECT
3156 || in_core
->al_stripes
!= 1
3157 || in_core
->al_stripe_size_4k
!= MD_32kB_SECT
/8)
3161 if (capacity
< in_core
->md_size_sect
)
3163 if (capacity
- in_core
->md_size_sect
< drbd_md_first_sector(bdev
))
3166 /* should be aligned, and at least 32k */
3167 if ((on_disk_al_sect
& 7) || (on_disk_al_sect
< MD_32kB_SECT
))
3170 /* should fit (for now: exactly) into the available on-disk space;
3171 * overflow prevention is in check_activity_log_stripe_size() above. */
3172 if (on_disk_al_sect
!= in_core
->al_size_4k
* MD_4kB_SECT
)
3175 /* again, should be aligned */
3176 if (in_core
->bm_offset
& 7)
3179 /* FIXME check for device grow with flex external meta data? */
3181 /* can the available bitmap space cover the last agreed device size? */
3182 if (on_disk_bm_sect
< (in_core
->la_size_sect
+7)/MD_4kB_SECT
/8/512)
3188 drbd_err(device
, "meta data offsets don't make sense: idx=%d "
3189 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3190 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3191 in_core
->meta_dev_idx
,
3192 in_core
->al_stripes
, in_core
->al_stripe_size_4k
,
3193 in_core
->al_offset
, in_core
->bm_offset
, in_core
->md_size_sect
,
3194 (unsigned long long)in_core
->la_size_sect
,
3195 (unsigned long long)capacity
);
3202 * drbd_md_read() - Reads in the meta data super block
3203 * @device: DRBD device.
3204 * @bdev: Device from which the meta data should be read in.
3206 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3207 * something goes wrong.
3209 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3210 * even before @bdev is assigned to @device->ldev.
3212 int drbd_md_read(struct drbd_device
*device
, struct drbd_backing_dev
*bdev
)
3214 struct meta_data_on_disk
*buffer
;
3216 int i
, rv
= NO_ERROR
;
3218 if (device
->state
.disk
!= D_DISKLESS
)
3219 return ERR_DISK_CONFIGURED
;
3221 buffer
= drbd_md_get_buffer(device
, __func__
);
3225 /* First, figure out where our meta data superblock is located,
3227 bdev
->md
.meta_dev_idx
= bdev
->disk_conf
->meta_dev_idx
;
3228 bdev
->md
.md_offset
= drbd_md_ss(bdev
);
3229 /* Even for (flexible or indexed) external meta data,
3230 * initially restrict us to the 4k superblock for now.
3231 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3232 bdev
->md
.md_size_sect
= 8;
3234 if (drbd_md_sync_page_io(device
, bdev
, bdev
->md
.md_offset
,
3236 /* NOTE: can't do normal error processing here as this is
3237 called BEFORE disk is attached */
3238 drbd_err(device
, "Error while reading metadata.\n");
3239 rv
= ERR_IO_MD_DISK
;
3243 magic
= be32_to_cpu(buffer
->magic
);
3244 flags
= be32_to_cpu(buffer
->flags
);
3245 if (magic
== DRBD_MD_MAGIC_84_UNCLEAN
||
3246 (magic
== DRBD_MD_MAGIC_08
&& !(flags
& MDF_AL_CLEAN
))) {
3247 /* btw: that's Activity Log clean, not "all" clean. */
3248 drbd_err(device
, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3249 rv
= ERR_MD_UNCLEAN
;
3253 rv
= ERR_MD_INVALID
;
3254 if (magic
!= DRBD_MD_MAGIC_08
) {
3255 if (magic
== DRBD_MD_MAGIC_07
)
3256 drbd_err(device
, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3258 drbd_err(device
, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3262 if (be32_to_cpu(buffer
->bm_bytes_per_bit
) != BM_BLOCK_SIZE
) {
3263 drbd_err(device
, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3264 be32_to_cpu(buffer
->bm_bytes_per_bit
), BM_BLOCK_SIZE
);
3269 /* convert to in_core endian */
3270 bdev
->md
.la_size_sect
= be64_to_cpu(buffer
->la_size_sect
);
3271 for (i
= UI_CURRENT
; i
< UI_SIZE
; i
++)
3272 bdev
->md
.uuid
[i
] = be64_to_cpu(buffer
->uuid
[i
]);
3273 bdev
->md
.flags
= be32_to_cpu(buffer
->flags
);
3274 bdev
->md
.device_uuid
= be64_to_cpu(buffer
->device_uuid
);
3276 bdev
->md
.md_size_sect
= be32_to_cpu(buffer
->md_size_sect
);
3277 bdev
->md
.al_offset
= be32_to_cpu(buffer
->al_offset
);
3278 bdev
->md
.bm_offset
= be32_to_cpu(buffer
->bm_offset
);
3280 if (check_activity_log_stripe_size(device
, buffer
, &bdev
->md
))
3282 if (check_offsets_and_sizes(device
, bdev
))
3285 if (be32_to_cpu(buffer
->bm_offset
) != bdev
->md
.bm_offset
) {
3286 drbd_err(device
, "unexpected bm_offset: %d (expected %d)\n",
3287 be32_to_cpu(buffer
->bm_offset
), bdev
->md
.bm_offset
);
3290 if (be32_to_cpu(buffer
->md_size_sect
) != bdev
->md
.md_size_sect
) {
3291 drbd_err(device
, "unexpected md_size: %u (expected %u)\n",
3292 be32_to_cpu(buffer
->md_size_sect
), bdev
->md
.md_size_sect
);
3298 spin_lock_irq(&device
->resource
->req_lock
);
3299 if (device
->state
.conn
< C_CONNECTED
) {
3301 peer
= be32_to_cpu(buffer
->la_peer_max_bio_size
);
3302 peer
= max(peer
, DRBD_MAX_BIO_SIZE_SAFE
);
3303 device
->peer_max_bio_size
= peer
;
3305 spin_unlock_irq(&device
->resource
->req_lock
);
3308 drbd_md_put_buffer(device
);
3314 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3315 * @device: DRBD device.
3317 * Call this function if you change anything that should be written to
3318 * the meta-data super block. This function sets MD_DIRTY, and starts a
3319 * timer that ensures that within five seconds you have to call drbd_md_sync().
3321 void drbd_md_mark_dirty(struct drbd_device
*device
)
3323 if (!test_and_set_bit(MD_DIRTY
, &device
->flags
))
3324 mod_timer(&device
->md_sync_timer
, jiffies
+ 5*HZ
);
3327 void drbd_uuid_move_history(struct drbd_device
*device
) __must_hold(local
)
3331 for (i
= UI_HISTORY_START
; i
< UI_HISTORY_END
; i
++)
3332 device
->ldev
->md
.uuid
[i
+1] = device
->ldev
->md
.uuid
[i
];
3335 void __drbd_uuid_set(struct drbd_device
*device
, int idx
, u64 val
) __must_hold(local
)
3337 if (idx
== UI_CURRENT
) {
3338 if (device
->state
.role
== R_PRIMARY
)
3343 drbd_set_ed_uuid(device
, val
);
3346 device
->ldev
->md
.uuid
[idx
] = val
;
3347 drbd_md_mark_dirty(device
);
3350 void _drbd_uuid_set(struct drbd_device
*device
, int idx
, u64 val
) __must_hold(local
)
3352 unsigned long flags
;
3353 spin_lock_irqsave(&device
->ldev
->md
.uuid_lock
, flags
);
3354 __drbd_uuid_set(device
, idx
, val
);
3355 spin_unlock_irqrestore(&device
->ldev
->md
.uuid_lock
, flags
);
3358 void drbd_uuid_set(struct drbd_device
*device
, int idx
, u64 val
) __must_hold(local
)
3360 unsigned long flags
;
3361 spin_lock_irqsave(&device
->ldev
->md
.uuid_lock
, flags
);
3362 if (device
->ldev
->md
.uuid
[idx
]) {
3363 drbd_uuid_move_history(device
);
3364 device
->ldev
->md
.uuid
[UI_HISTORY_START
] = device
->ldev
->md
.uuid
[idx
];
3366 __drbd_uuid_set(device
, idx
, val
);
3367 spin_unlock_irqrestore(&device
->ldev
->md
.uuid_lock
, flags
);
3371 * drbd_uuid_new_current() - Creates a new current UUID
3372 * @device: DRBD device.
3374 * Creates a new current UUID, and rotates the old current UUID into
3375 * the bitmap slot. Causes an incremental resync upon next connect.
3377 void drbd_uuid_new_current(struct drbd_device
*device
) __must_hold(local
)
3380 unsigned long long bm_uuid
;
3382 get_random_bytes(&val
, sizeof(u64
));
3384 spin_lock_irq(&device
->ldev
->md
.uuid_lock
);
3385 bm_uuid
= device
->ldev
->md
.uuid
[UI_BITMAP
];
3388 drbd_warn(device
, "bm UUID was already set: %llX\n", bm_uuid
);
3390 device
->ldev
->md
.uuid
[UI_BITMAP
] = device
->ldev
->md
.uuid
[UI_CURRENT
];
3391 __drbd_uuid_set(device
, UI_CURRENT
, val
);
3392 spin_unlock_irq(&device
->ldev
->md
.uuid_lock
);
3394 drbd_print_uuids(device
, "new current UUID");
3395 /* get it to stable storage _now_ */
3396 drbd_md_sync(device
);
3399 void drbd_uuid_set_bm(struct drbd_device
*device
, u64 val
) __must_hold(local
)
3401 unsigned long flags
;
3402 if (device
->ldev
->md
.uuid
[UI_BITMAP
] == 0 && val
== 0)
3405 spin_lock_irqsave(&device
->ldev
->md
.uuid_lock
, flags
);
3407 drbd_uuid_move_history(device
);
3408 device
->ldev
->md
.uuid
[UI_HISTORY_START
] = device
->ldev
->md
.uuid
[UI_BITMAP
];
3409 device
->ldev
->md
.uuid
[UI_BITMAP
] = 0;
3411 unsigned long long bm_uuid
= device
->ldev
->md
.uuid
[UI_BITMAP
];
3413 drbd_warn(device
, "bm UUID was already set: %llX\n", bm_uuid
);
3415 device
->ldev
->md
.uuid
[UI_BITMAP
] = val
& ~((u64
)1);
3417 spin_unlock_irqrestore(&device
->ldev
->md
.uuid_lock
, flags
);
3419 drbd_md_mark_dirty(device
);
3423 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3424 * @device: DRBD device.
3426 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3428 int drbd_bmio_set_n_write(struct drbd_device
*device
,
3429 struct drbd_peer_device
*peer_device
) __must_hold(local
)
3434 drbd_md_set_flag(device
, MDF_FULL_SYNC
);
3435 drbd_md_sync(device
);
3436 drbd_bm_set_all(device
);
3438 rv
= drbd_bm_write(device
, peer_device
);
3441 drbd_md_clear_flag(device
, MDF_FULL_SYNC
);
3442 drbd_md_sync(device
);
3449 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3450 * @device: DRBD device.
3452 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3454 int drbd_bmio_clear_n_write(struct drbd_device
*device
,
3455 struct drbd_peer_device
*peer_device
) __must_hold(local
)
3458 drbd_resume_al(device
);
3459 drbd_bm_clear_all(device
);
3460 return drbd_bm_write(device
, peer_device
);
3463 static int w_bitmap_io(struct drbd_work
*w
, int unused
)
3465 struct drbd_device
*device
=
3466 container_of(w
, struct drbd_device
, bm_io_work
.w
);
3467 struct bm_io_work
*work
= &device
->bm_io_work
;
3470 if (work
->flags
!= BM_LOCKED_CHANGE_ALLOWED
) {
3471 int cnt
= atomic_read(&device
->ap_bio_cnt
);
3473 drbd_err(device
, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3477 if (get_ldev(device
)) {
3478 drbd_bm_lock(device
, work
->why
, work
->flags
);
3479 rv
= work
->io_fn(device
, work
->peer_device
);
3480 drbd_bm_unlock(device
);
3484 clear_bit_unlock(BITMAP_IO
, &device
->flags
);
3485 wake_up(&device
->misc_wait
);
3488 work
->done(device
, rv
);
3490 clear_bit(BITMAP_IO_QUEUED
, &device
->flags
);
3498 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3499 * @device: DRBD device.
3500 * @io_fn: IO callback to be called when bitmap IO is possible
3501 * @done: callback to be called after the bitmap IO was performed
3502 * @why: Descriptive text of the reason for doing the IO
3503 * @flags: Bitmap flags
3505 * While IO on the bitmap happens we freeze application IO thus we ensure
3506 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3507 * called from worker context. It MUST NOT be used while a previous such
3508 * work is still pending!
3510 * Its worker function encloses the call of io_fn() by get_ldev() and
3513 void drbd_queue_bitmap_io(struct drbd_device
*device
,
3514 int (*io_fn
)(struct drbd_device
*, struct drbd_peer_device
*),
3515 void (*done
)(struct drbd_device
*, int),
3516 char *why
, enum bm_flag flags
,
3517 struct drbd_peer_device
*peer_device
)
3519 D_ASSERT(device
, current
== peer_device
->connection
->worker
.task
);
3521 D_ASSERT(device
, !test_bit(BITMAP_IO_QUEUED
, &device
->flags
));
3522 D_ASSERT(device
, !test_bit(BITMAP_IO
, &device
->flags
));
3523 D_ASSERT(device
, list_empty(&device
->bm_io_work
.w
.list
));
3524 if (device
->bm_io_work
.why
)
3525 drbd_err(device
, "FIXME going to queue '%s' but '%s' still pending?\n",
3526 why
, device
->bm_io_work
.why
);
3528 device
->bm_io_work
.peer_device
= peer_device
;
3529 device
->bm_io_work
.io_fn
= io_fn
;
3530 device
->bm_io_work
.done
= done
;
3531 device
->bm_io_work
.why
= why
;
3532 device
->bm_io_work
.flags
= flags
;
3534 spin_lock_irq(&device
->resource
->req_lock
);
3535 set_bit(BITMAP_IO
, &device
->flags
);
3536 /* don't wait for pending application IO if the caller indicates that
3537 * application IO does not conflict anyways. */
3538 if (flags
== BM_LOCKED_CHANGE_ALLOWED
|| atomic_read(&device
->ap_bio_cnt
) == 0) {
3539 if (!test_and_set_bit(BITMAP_IO_QUEUED
, &device
->flags
))
3540 drbd_queue_work(&peer_device
->connection
->sender_work
,
3541 &device
->bm_io_work
.w
);
3543 spin_unlock_irq(&device
->resource
->req_lock
);
3547 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3548 * @device: DRBD device.
3549 * @io_fn: IO callback to be called when bitmap IO is possible
3550 * @why: Descriptive text of the reason for doing the IO
3551 * @flags: Bitmap flags
3553 * freezes application IO while that the actual IO operations runs. This
3554 * functions MAY NOT be called from worker context.
3556 int drbd_bitmap_io(struct drbd_device
*device
,
3557 int (*io_fn
)(struct drbd_device
*, struct drbd_peer_device
*),
3558 char *why
, enum bm_flag flags
,
3559 struct drbd_peer_device
*peer_device
)
3561 /* Only suspend io, if some operation is supposed to be locked out */
3562 const bool do_suspend_io
= flags
& (BM_DONT_CLEAR
|BM_DONT_SET
|BM_DONT_TEST
);
3565 D_ASSERT(device
, current
!= first_peer_device(device
)->connection
->worker
.task
);
3568 drbd_suspend_io(device
);
3570 drbd_bm_lock(device
, why
, flags
);
3571 rv
= io_fn(device
, peer_device
);
3572 drbd_bm_unlock(device
);
3575 drbd_resume_io(device
);
3580 void drbd_md_set_flag(struct drbd_device
*device
, int flag
) __must_hold(local
)
3582 if ((device
->ldev
->md
.flags
& flag
) != flag
) {
3583 drbd_md_mark_dirty(device
);
3584 device
->ldev
->md
.flags
|= flag
;
3588 void drbd_md_clear_flag(struct drbd_device
*device
, int flag
) __must_hold(local
)
3590 if ((device
->ldev
->md
.flags
& flag
) != 0) {
3591 drbd_md_mark_dirty(device
);
3592 device
->ldev
->md
.flags
&= ~flag
;
3595 int drbd_md_test_flag(struct drbd_backing_dev
*bdev
, int flag
)
3597 return (bdev
->md
.flags
& flag
) != 0;
3600 static void md_sync_timer_fn(struct timer_list
*t
)
3602 struct drbd_device
*device
= from_timer(device
, t
, md_sync_timer
);
3603 drbd_device_post_work(device
, MD_SYNC
);
3606 const char *cmdname(enum drbd_packet cmd
)
3608 /* THINK may need to become several global tables
3609 * when we want to support more than
3610 * one PRO_VERSION */
3611 static const char *cmdnames
[] = {
3614 [P_DATA_REPLY
] = "DataReply",
3615 [P_RS_DATA_REPLY
] = "RSDataReply",
3616 [P_BARRIER
] = "Barrier",
3617 [P_BITMAP
] = "ReportBitMap",
3618 [P_BECOME_SYNC_TARGET
] = "BecomeSyncTarget",
3619 [P_BECOME_SYNC_SOURCE
] = "BecomeSyncSource",
3620 [P_UNPLUG_REMOTE
] = "UnplugRemote",
3621 [P_DATA_REQUEST
] = "DataRequest",
3622 [P_RS_DATA_REQUEST
] = "RSDataRequest",
3623 [P_SYNC_PARAM
] = "SyncParam",
3624 [P_PROTOCOL
] = "ReportProtocol",
3625 [P_UUIDS
] = "ReportUUIDs",
3626 [P_SIZES
] = "ReportSizes",
3627 [P_STATE
] = "ReportState",
3628 [P_SYNC_UUID
] = "ReportSyncUUID",
3629 [P_AUTH_CHALLENGE
] = "AuthChallenge",
3630 [P_AUTH_RESPONSE
] = "AuthResponse",
3631 [P_STATE_CHG_REQ
] = "StateChgRequest",
3633 [P_PING_ACK
] = "PingAck",
3634 [P_RECV_ACK
] = "RecvAck",
3635 [P_WRITE_ACK
] = "WriteAck",
3636 [P_RS_WRITE_ACK
] = "RSWriteAck",
3637 [P_SUPERSEDED
] = "Superseded",
3638 [P_NEG_ACK
] = "NegAck",
3639 [P_NEG_DREPLY
] = "NegDReply",
3640 [P_NEG_RS_DREPLY
] = "NegRSDReply",
3641 [P_BARRIER_ACK
] = "BarrierAck",
3642 [P_STATE_CHG_REPLY
] = "StateChgReply",
3643 [P_OV_REQUEST
] = "OVRequest",
3644 [P_OV_REPLY
] = "OVReply",
3645 [P_OV_RESULT
] = "OVResult",
3646 [P_CSUM_RS_REQUEST
] = "CsumRSRequest",
3647 [P_RS_IS_IN_SYNC
] = "CsumRSIsInSync",
3648 [P_SYNC_PARAM89
] = "SyncParam89",
3649 [P_COMPRESSED_BITMAP
] = "CBitmap",
3650 [P_DELAY_PROBE
] = "DelayProbe",
3651 [P_OUT_OF_SYNC
] = "OutOfSync",
3652 [P_RS_CANCEL
] = "RSCancel",
3653 [P_CONN_ST_CHG_REQ
] = "conn_st_chg_req",
3654 [P_CONN_ST_CHG_REPLY
] = "conn_st_chg_reply",
3655 [P_PROTOCOL_UPDATE
] = "protocol_update",
3657 [P_RS_THIN_REQ
] = "rs_thin_req",
3658 [P_RS_DEALLOCATED
] = "rs_deallocated",
3659 [P_WSAME
] = "WriteSame",
3660 [P_ZEROES
] = "Zeroes",
3662 /* enum drbd_packet, but not commands - obsoleted flags:
3668 /* too big for the array: 0xfffX */
3669 if (cmd
== P_INITIAL_META
)
3670 return "InitialMeta";
3671 if (cmd
== P_INITIAL_DATA
)
3672 return "InitialData";
3673 if (cmd
== P_CONNECTION_FEATURES
)
3674 return "ConnectionFeatures";
3675 if (cmd
>= ARRAY_SIZE(cmdnames
))
3677 return cmdnames
[cmd
];
3681 * drbd_wait_misc - wait for a request to make progress
3682 * @device: device associated with the request
3683 * @i: the struct drbd_interval embedded in struct drbd_request or
3684 * struct drbd_peer_request
3686 int drbd_wait_misc(struct drbd_device
*device
, struct drbd_interval
*i
)
3688 struct net_conf
*nc
;
3693 nc
= rcu_dereference(first_peer_device(device
)->connection
->net_conf
);
3698 timeout
= nc
->ko_count
? nc
->timeout
* HZ
/ 10 * nc
->ko_count
: MAX_SCHEDULE_TIMEOUT
;
3701 /* Indicate to wake up device->misc_wait on progress. */
3703 prepare_to_wait(&device
->misc_wait
, &wait
, TASK_INTERRUPTIBLE
);
3704 spin_unlock_irq(&device
->resource
->req_lock
);
3705 timeout
= schedule_timeout(timeout
);
3706 finish_wait(&device
->misc_wait
, &wait
);
3707 spin_lock_irq(&device
->resource
->req_lock
);
3708 if (!timeout
|| device
->state
.conn
< C_CONNECTED
)
3710 if (signal_pending(current
))
3711 return -ERESTARTSYS
;
3715 void lock_all_resources(void)
3717 struct drbd_resource
*resource
;
3718 int __maybe_unused i
= 0;
3720 mutex_lock(&resources_mutex
);
3721 local_irq_disable();
3722 for_each_resource(resource
, &drbd_resources
)
3723 spin_lock_nested(&resource
->req_lock
, i
++);
3726 void unlock_all_resources(void)
3728 struct drbd_resource
*resource
;
3730 for_each_resource(resource
, &drbd_resources
)
3731 spin_unlock(&resource
->req_lock
);
3733 mutex_unlock(&resources_mutex
);
3736 #ifdef CONFIG_DRBD_FAULT_INJECTION
3737 /* Fault insertion support including random number generator shamelessly
3738 * stolen from kernel/rcutorture.c */
3739 struct fault_random_state
{
3740 unsigned long state
;
3741 unsigned long count
;
3744 #define FAULT_RANDOM_MULT 39916801 /* prime */
3745 #define FAULT_RANDOM_ADD 479001701 /* prime */
3746 #define FAULT_RANDOM_REFRESH 10000
3749 * Crude but fast random-number generator. Uses a linear congruential
3750 * generator, with occasional help from get_random_bytes().
3752 static unsigned long
3753 _drbd_fault_random(struct fault_random_state
*rsp
)
3757 if (!rsp
->count
--) {
3758 get_random_bytes(&refresh
, sizeof(refresh
));
3759 rsp
->state
+= refresh
;
3760 rsp
->count
= FAULT_RANDOM_REFRESH
;
3762 rsp
->state
= rsp
->state
* FAULT_RANDOM_MULT
+ FAULT_RANDOM_ADD
;
3763 return swahw32(rsp
->state
);
3767 _drbd_fault_str(unsigned int type
) {
3768 static char *_faults
[] = {
3769 [DRBD_FAULT_MD_WR
] = "Meta-data write",
3770 [DRBD_FAULT_MD_RD
] = "Meta-data read",
3771 [DRBD_FAULT_RS_WR
] = "Resync write",
3772 [DRBD_FAULT_RS_RD
] = "Resync read",
3773 [DRBD_FAULT_DT_WR
] = "Data write",
3774 [DRBD_FAULT_DT_RD
] = "Data read",
3775 [DRBD_FAULT_DT_RA
] = "Data read ahead",
3776 [DRBD_FAULT_BM_ALLOC
] = "BM allocation",
3777 [DRBD_FAULT_AL_EE
] = "EE allocation",
3778 [DRBD_FAULT_RECEIVE
] = "receive data corruption",
3781 return (type
< DRBD_FAULT_MAX
) ? _faults
[type
] : "**Unknown**";
3785 _drbd_insert_fault(struct drbd_device
*device
, unsigned int type
)
3787 static struct fault_random_state rrs
= {0, 0};
3789 unsigned int ret
= (
3790 (drbd_fault_devs
== 0 ||
3791 ((1 << device_to_minor(device
)) & drbd_fault_devs
) != 0) &&
3792 (((_drbd_fault_random(&rrs
) % 100) + 1) <= drbd_fault_rate
));
3797 if (drbd_ratelimit())
3798 drbd_warn(device
, "***Simulating %s failure\n",
3799 _drbd_fault_str(type
));
3806 module_init(drbd_init
)
3807 module_exit(drbd_cleanup
)
3809 EXPORT_SYMBOL(drbd_conn_str
);
3810 EXPORT_SYMBOL(drbd_role_str
);
3811 EXPORT_SYMBOL(drbd_disk_str
);
3812 EXPORT_SYMBOL(drbd_set_st_err_str
);