1 // SPDX-License-Identifier: GPL-2.0-or-later
5 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
7 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
8 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
9 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
11 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
12 from Logicworks, Inc. for making SDP replication support possible.
17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 #include <linux/module.h>
20 #include <linux/jiffies.h>
21 #include <linux/drbd.h>
22 #include <linux/uaccess.h>
23 #include <asm/types.h>
25 #include <linux/ctype.h>
26 #include <linux/mutex.h>
28 #include <linux/file.h>
29 #include <linux/proc_fs.h>
30 #include <linux/init.h>
32 #include <linux/memcontrol.h>
33 #include <linux/mm_inline.h>
34 #include <linux/slab.h>
35 #include <linux/random.h>
36 #include <linux/reboot.h>
37 #include <linux/notifier.h>
38 #include <linux/kthread.h>
39 #include <linux/workqueue.h>
40 #define __KERNEL_SYSCALLS__
41 #include <linux/unistd.h>
42 #include <linux/vmalloc.h>
43 #include <linux/sched/signal.h>
45 #include <linux/drbd_limits.h>
47 #include "drbd_protocol.h"
48 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
50 #include "drbd_debugfs.h"
52 static DEFINE_MUTEX(drbd_main_mutex);
53 static int drbd_open(struct block_device *bdev, fmode_t mode);
54 static void drbd_release(struct gendisk *gd, fmode_t mode);
55 static void md_sync_timer_fn(struct timer_list *t);
56 static int w_bitmap_io(struct drbd_work *w, int unused);
58 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
59 "Lars Ellenberg <lars@linbit.com>");
60 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
61 MODULE_VERSION(REL_VERSION);
62 MODULE_LICENSE("GPL");
63 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
64 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
65 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
67 #include <linux/moduleparam.h>
68 /* thanks to these macros, if compiled into the kernel (not-module),
69 * these become boot parameters (e.g., drbd.minor_count) */
71 #ifdef CONFIG_DRBD_FAULT_INJECTION
72 int drbd_enable_faults;
74 static int drbd_fault_count;
75 static int drbd_fault_devs;
76 /* bitmap of enabled faults */
77 module_param_named(enable_faults, drbd_enable_faults, int, 0664);
78 /* fault rate % value - applies to all enabled faults */
79 module_param_named(fault_rate, drbd_fault_rate, int, 0664);
80 /* count of faults inserted */
81 module_param_named(fault_count, drbd_fault_count, int, 0664);
82 /* bitmap of devices to insert faults on */
83 module_param_named(fault_devs, drbd_fault_devs, int, 0644);
86 /* module parameters we can keep static */
87 static bool drbd_allow_oos; /* allow_open_on_secondary */
88 static bool drbd_disable_sendpage;
89 MODULE_PARM_DESC(allow_oos, "DONT USE!");
90 module_param_named(allow_oos, drbd_allow_oos, bool, 0);
91 module_param_named(disable_sendpage, drbd_disable_sendpage, bool, 0644);
93 /* module parameters we share */
94 int drbd_proc_details; /* Detail level in proc drbd*/
95 module_param_named(proc_details, drbd_proc_details, int, 0644);
96 /* module parameters shared with defaults */
97 unsigned int drbd_minor_count = DRBD_MINOR_COUNT_DEF;
98 /* Module parameter for setting the user mode helper program
99 * to run. Default is /sbin/drbdadm */
100 char drbd_usermode_helper[80] = "/sbin/drbdadm";
101 module_param_named(minor_count, drbd_minor_count, uint, 0444);
102 module_param_string(usermode_helper, drbd_usermode_helper, sizeof(drbd_usermode_helper), 0644);
104 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
105 * as member "struct gendisk *vdisk;"
107 struct idr drbd_devices;
108 struct list_head drbd_resources;
109 struct mutex resources_mutex;
111 struct kmem_cache *drbd_request_cache;
112 struct kmem_cache *drbd_ee_cache; /* peer requests */
113 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
114 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
115 mempool_t drbd_request_mempool;
116 mempool_t drbd_ee_mempool;
117 mempool_t drbd_md_io_page_pool;
118 struct bio_set drbd_md_io_bio_set;
119 struct bio_set drbd_io_bio_set;
121 /* I do not use a standard mempool, because:
122 1) I want to hand out the pre-allocated objects first.
123 2) I want to be able to interrupt sleeping allocation with a signal.
124 Note: This is a single linked list, the next pointer is the private
125 member of struct page.
127 struct page *drbd_pp_pool;
128 DEFINE_SPINLOCK(drbd_pp_lock);
130 wait_queue_head_t drbd_pp_wait;
132 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
134 static const struct block_device_operations drbd_ops = {
135 .owner = THIS_MODULE,
136 .submit_bio = drbd_submit_bio,
138 .release = drbd_release,
142 /* When checking with sparse, and this is an inline function, sparse will
143 give tons of false positives. When this is a real functions sparse works.
145 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
149 atomic_inc(&device->local_cnt);
150 io_allowed = (device->state.disk >= mins);
152 if (atomic_dec_and_test(&device->local_cnt))
153 wake_up(&device->misc_wait);
161 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
162 * @connection: DRBD connection.
163 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
164 * @set_size: Expected number of requests before that barrier.
166 * In case the passed barrier_nr or set_size does not match the oldest
167 * epoch of not yet barrier-acked requests, this function will cause a
168 * termination of the connection.
170 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
171 unsigned int set_size)
173 struct drbd_request *r;
174 struct drbd_request *req = NULL;
175 int expect_epoch = 0;
178 spin_lock_irq(&connection->resource->req_lock);
180 /* find oldest not yet barrier-acked write request,
181 * count writes in its epoch. */
182 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
183 const unsigned s = r->rq_state;
187 if (!(s & RQ_NET_MASK))
192 expect_epoch = req->epoch;
195 if (r->epoch != expect_epoch)
199 /* if (s & RQ_DONE): not expected */
200 /* if (!(s & RQ_NET_MASK)): not expected */
205 /* first some paranoia code */
207 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
211 if (expect_epoch != barrier_nr) {
212 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
213 barrier_nr, expect_epoch);
217 if (expect_size != set_size) {
218 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
219 barrier_nr, set_size, expect_size);
223 /* Clean up list of requests processed during current epoch. */
224 /* this extra list walk restart is paranoia,
225 * to catch requests being barrier-acked "unexpectedly".
226 * It usually should find the same req again, or some READ preceding it. */
227 list_for_each_entry(req, &connection->transfer_log, tl_requests)
228 if (req->epoch == expect_epoch)
230 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
231 if (req->epoch != expect_epoch)
233 _req_mod(req, BARRIER_ACKED);
235 spin_unlock_irq(&connection->resource->req_lock);
240 spin_unlock_irq(&connection->resource->req_lock);
241 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
246 * _tl_restart() - Walks the transfer log, and applies an action to all requests
247 * @connection: DRBD connection to operate on.
248 * @what: The action/event to perform with all request objects
250 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
251 * RESTART_FROZEN_DISK_IO.
253 /* must hold resource->req_lock */
254 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
256 struct drbd_request *req, *r;
258 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
262 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
264 spin_lock_irq(&connection->resource->req_lock);
265 _tl_restart(connection, what);
266 spin_unlock_irq(&connection->resource->req_lock);
270 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
271 * @connection: DRBD connection.
273 * This is called after the connection to the peer was lost. The storage covered
274 * by the requests on the transfer gets marked as our of sync. Called from the
275 * receiver thread and the worker thread.
277 void tl_clear(struct drbd_connection *connection)
279 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
283 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
284 * @device: DRBD device.
286 void tl_abort_disk_io(struct drbd_device *device)
288 struct drbd_connection *connection = first_peer_device(device)->connection;
289 struct drbd_request *req, *r;
291 spin_lock_irq(&connection->resource->req_lock);
292 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
293 if (!(req->rq_state & RQ_LOCAL_PENDING))
295 if (req->device != device)
297 _req_mod(req, ABORT_DISK_IO);
299 spin_unlock_irq(&connection->resource->req_lock);
302 static int drbd_thread_setup(void *arg)
304 struct drbd_thread *thi = (struct drbd_thread *) arg;
305 struct drbd_resource *resource = thi->resource;
309 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
313 allow_kernel_signal(DRBD_SIGKILL);
314 allow_kernel_signal(SIGXCPU);
316 retval = thi->function(thi);
318 spin_lock_irqsave(&thi->t_lock, flags);
320 /* if the receiver has been "EXITING", the last thing it did
321 * was set the conn state to "StandAlone",
322 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
323 * and receiver thread will be "started".
324 * drbd_thread_start needs to set "RESTARTING" in that case.
325 * t_state check and assignment needs to be within the same spinlock,
326 * so either thread_start sees EXITING, and can remap to RESTARTING,
327 * or thread_start see NONE, and can proceed as normal.
330 if (thi->t_state == RESTARTING) {
331 drbd_info(resource, "Restarting %s thread\n", thi->name);
332 thi->t_state = RUNNING;
333 spin_unlock_irqrestore(&thi->t_lock, flags);
340 complete_all(&thi->stop);
341 spin_unlock_irqrestore(&thi->t_lock, flags);
343 drbd_info(resource, "Terminating %s\n", current->comm);
345 /* Release mod reference taken when thread was started */
348 kref_put(&thi->connection->kref, drbd_destroy_connection);
349 kref_put(&resource->kref, drbd_destroy_resource);
350 module_put(THIS_MODULE);
354 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
355 int (*func) (struct drbd_thread *), const char *name)
357 spin_lock_init(&thi->t_lock);
360 thi->function = func;
361 thi->resource = resource;
362 thi->connection = NULL;
366 int drbd_thread_start(struct drbd_thread *thi)
368 struct drbd_resource *resource = thi->resource;
369 struct task_struct *nt;
372 /* is used from state engine doing drbd_thread_stop_nowait,
373 * while holding the req lock irqsave */
374 spin_lock_irqsave(&thi->t_lock, flags);
376 switch (thi->t_state) {
378 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
379 thi->name, current->comm, current->pid);
381 /* Get ref on module for thread - this is released when thread exits */
382 if (!try_module_get(THIS_MODULE)) {
383 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
384 spin_unlock_irqrestore(&thi->t_lock, flags);
388 kref_get(&resource->kref);
390 kref_get(&thi->connection->kref);
392 init_completion(&thi->stop);
393 thi->reset_cpu_mask = 1;
394 thi->t_state = RUNNING;
395 spin_unlock_irqrestore(&thi->t_lock, flags);
396 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
398 nt = kthread_create(drbd_thread_setup, (void *) thi,
399 "drbd_%c_%s", thi->name[0], thi->resource->name);
402 drbd_err(resource, "Couldn't start thread\n");
405 kref_put(&thi->connection->kref, drbd_destroy_connection);
406 kref_put(&resource->kref, drbd_destroy_resource);
407 module_put(THIS_MODULE);
410 spin_lock_irqsave(&thi->t_lock, flags);
412 thi->t_state = RUNNING;
413 spin_unlock_irqrestore(&thi->t_lock, flags);
417 thi->t_state = RESTARTING;
418 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
419 thi->name, current->comm, current->pid);
424 spin_unlock_irqrestore(&thi->t_lock, flags);
432 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
436 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
438 /* may be called from state engine, holding the req lock irqsave */
439 spin_lock_irqsave(&thi->t_lock, flags);
441 if (thi->t_state == NONE) {
442 spin_unlock_irqrestore(&thi->t_lock, flags);
444 drbd_thread_start(thi);
448 if (thi->t_state != ns) {
449 if (thi->task == NULL) {
450 spin_unlock_irqrestore(&thi->t_lock, flags);
456 init_completion(&thi->stop);
457 if (thi->task != current)
458 send_sig(DRBD_SIGKILL, thi->task, 1);
461 spin_unlock_irqrestore(&thi->t_lock, flags);
464 wait_for_completion(&thi->stop);
467 int conn_lowest_minor(struct drbd_connection *connection)
469 struct drbd_peer_device *peer_device;
470 int vnr = 0, minor = -1;
473 peer_device = idr_get_next(&connection->peer_devices, &vnr);
475 minor = device_to_minor(peer_device->device);
483 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
485 * Forces all threads of a resource onto the same CPU. This is beneficial for
486 * DRBD's performance. May be overwritten by user's configuration.
488 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
490 unsigned int *resources_per_cpu, min_index = ~0;
492 resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu),
494 if (resources_per_cpu) {
495 struct drbd_resource *resource;
496 unsigned int cpu, min = ~0;
499 for_each_resource_rcu(resource, &drbd_resources) {
500 for_each_cpu(cpu, resource->cpu_mask)
501 resources_per_cpu[cpu]++;
504 for_each_online_cpu(cpu) {
505 if (resources_per_cpu[cpu] < min) {
506 min = resources_per_cpu[cpu];
510 kfree(resources_per_cpu);
512 if (min_index == ~0) {
513 cpumask_setall(*cpu_mask);
516 cpumask_set_cpu(min_index, *cpu_mask);
520 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
521 * @thi: drbd_thread object
523 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
526 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
528 struct drbd_resource *resource = thi->resource;
529 struct task_struct *p = current;
531 if (!thi->reset_cpu_mask)
533 thi->reset_cpu_mask = 0;
534 set_cpus_allowed_ptr(p, resource->cpu_mask);
537 #define drbd_calc_cpu_mask(A) ({})
541 * drbd_header_size - size of a packet header
543 * The header size is a multiple of 8, so any payload following the header is
544 * word aligned on 64-bit architectures. (The bitmap send and receive code
547 unsigned int drbd_header_size(struct drbd_connection *connection)
549 if (connection->agreed_pro_version >= 100) {
550 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
551 return sizeof(struct p_header100);
553 BUILD_BUG_ON(sizeof(struct p_header80) !=
554 sizeof(struct p_header95));
555 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
556 return sizeof(struct p_header80);
560 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
562 h->magic = cpu_to_be32(DRBD_MAGIC);
563 h->command = cpu_to_be16(cmd);
564 h->length = cpu_to_be16(size);
565 return sizeof(struct p_header80);
568 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
570 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
571 h->command = cpu_to_be16(cmd);
572 h->length = cpu_to_be32(size);
573 return sizeof(struct p_header95);
576 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
579 h->magic = cpu_to_be32(DRBD_MAGIC_100);
580 h->volume = cpu_to_be16(vnr);
581 h->command = cpu_to_be16(cmd);
582 h->length = cpu_to_be32(size);
584 return sizeof(struct p_header100);
587 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
588 void *buffer, enum drbd_packet cmd, int size)
590 if (connection->agreed_pro_version >= 100)
591 return prepare_header100(buffer, cmd, size, vnr);
592 else if (connection->agreed_pro_version >= 95 &&
593 size > DRBD_MAX_SIZE_H80_PACKET)
594 return prepare_header95(buffer, cmd, size);
596 return prepare_header80(buffer, cmd, size);
599 static void *__conn_prepare_command(struct drbd_connection *connection,
600 struct drbd_socket *sock)
604 return sock->sbuf + drbd_header_size(connection);
607 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
611 mutex_lock(&sock->mutex);
612 p = __conn_prepare_command(connection, sock);
614 mutex_unlock(&sock->mutex);
619 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
621 return conn_prepare_command(peer_device->connection, sock);
624 static int __send_command(struct drbd_connection *connection, int vnr,
625 struct drbd_socket *sock, enum drbd_packet cmd,
626 unsigned int header_size, void *data,
633 * Called with @data == NULL and the size of the data blocks in @size
634 * for commands that send data blocks. For those commands, omit the
635 * MSG_MORE flag: this will increase the likelihood that data blocks
636 * which are page aligned on the sender will end up page aligned on the
639 msg_flags = data ? MSG_MORE : 0;
641 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
643 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
646 err = drbd_send_all(connection, sock->socket, data, size, 0);
647 /* DRBD protocol "pings" are latency critical.
648 * This is supposed to trigger tcp_push_pending_frames() */
649 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
650 tcp_sock_set_nodelay(sock->socket->sk);
655 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
656 enum drbd_packet cmd, unsigned int header_size,
657 void *data, unsigned int size)
659 return __send_command(connection, 0, sock, cmd, header_size, data, size);
662 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)
668 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
669 mutex_unlock(&sock->mutex);
673 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
674 enum drbd_packet cmd, unsigned int header_size,
675 void *data, unsigned int size)
679 err = __send_command(peer_device->connection, peer_device->device->vnr,
680 sock, cmd, header_size, data, size);
681 mutex_unlock(&sock->mutex);
685 int drbd_send_ping(struct drbd_connection *connection)
687 struct drbd_socket *sock;
689 sock = &connection->meta;
690 if (!conn_prepare_command(connection, sock))
692 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
695 int drbd_send_ping_ack(struct drbd_connection *connection)
697 struct drbd_socket *sock;
699 sock = &connection->meta;
700 if (!conn_prepare_command(connection, sock))
702 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
705 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
707 struct drbd_socket *sock;
708 struct p_rs_param_95 *p;
710 const int apv = peer_device->connection->agreed_pro_version;
711 enum drbd_packet cmd;
713 struct disk_conf *dc;
715 sock = &peer_device->connection->data;
716 p = drbd_prepare_command(peer_device, sock);
721 nc = rcu_dereference(peer_device->connection->net_conf);
723 size = apv <= 87 ? sizeof(struct p_rs_param)
724 : apv == 88 ? sizeof(struct p_rs_param)
725 + strlen(nc->verify_alg) + 1
726 : apv <= 94 ? sizeof(struct p_rs_param_89)
727 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
729 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
731 /* initialize verify_alg and csums_alg */
732 BUILD_BUG_ON(sizeof(p->algs) != 2 * SHARED_SECRET_MAX);
733 memset(&p->algs, 0, sizeof(p->algs));
735 if (get_ldev(peer_device->device)) {
736 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
737 p->resync_rate = cpu_to_be32(dc->resync_rate);
738 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
739 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
740 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
741 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
742 put_ldev(peer_device->device);
744 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
745 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
746 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
747 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
748 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
752 strcpy(p->verify_alg, nc->verify_alg);
754 strcpy(p->csums_alg, nc->csums_alg);
757 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
760 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
762 struct drbd_socket *sock;
763 struct p_protocol *p;
767 sock = &connection->data;
768 p = __conn_prepare_command(connection, sock);
773 nc = rcu_dereference(connection->net_conf);
775 if (nc->tentative && connection->agreed_pro_version < 92) {
777 drbd_err(connection, "--dry-run is not supported by peer");
782 if (connection->agreed_pro_version >= 87)
783 size += strlen(nc->integrity_alg) + 1;
785 p->protocol = cpu_to_be32(nc->wire_protocol);
786 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
787 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
788 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
789 p->two_primaries = cpu_to_be32(nc->two_primaries);
791 if (nc->discard_my_data)
792 cf |= CF_DISCARD_MY_DATA;
795 p->conn_flags = cpu_to_be32(cf);
797 if (connection->agreed_pro_version >= 87)
798 strcpy(p->integrity_alg, nc->integrity_alg);
801 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
804 int drbd_send_protocol(struct drbd_connection *connection)
808 mutex_lock(&connection->data.mutex);
809 err = __drbd_send_protocol(connection, P_PROTOCOL);
810 mutex_unlock(&connection->data.mutex);
815 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
817 struct drbd_device *device = peer_device->device;
818 struct drbd_socket *sock;
822 if (!get_ldev_if_state(device, D_NEGOTIATING))
825 sock = &peer_device->connection->data;
826 p = drbd_prepare_command(peer_device, sock);
831 spin_lock_irq(&device->ldev->md.uuid_lock);
832 for (i = UI_CURRENT; i < UI_SIZE; i++)
833 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
834 spin_unlock_irq(&device->ldev->md.uuid_lock);
836 device->comm_bm_set = drbd_bm_total_weight(device);
837 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
839 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
841 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
842 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
843 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
846 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
849 int drbd_send_uuids(struct drbd_peer_device *peer_device)
851 return _drbd_send_uuids(peer_device, 0);
854 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
856 return _drbd_send_uuids(peer_device, 8);
859 void drbd_print_uuids(struct drbd_device *device, const char *text)
861 if (get_ldev_if_state(device, D_NEGOTIATING)) {
862 u64 *uuid = device->ldev->md.uuid;
863 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
865 (unsigned long long)uuid[UI_CURRENT],
866 (unsigned long long)uuid[UI_BITMAP],
867 (unsigned long long)uuid[UI_HISTORY_START],
868 (unsigned long long)uuid[UI_HISTORY_END]);
871 drbd_info(device, "%s effective data uuid: %016llX\n",
873 (unsigned long long)device->ed_uuid);
877 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
879 struct drbd_device *device = peer_device->device;
880 struct drbd_socket *sock;
884 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
886 uuid = device->ldev->md.uuid[UI_BITMAP];
887 if (uuid && uuid != UUID_JUST_CREATED)
888 uuid = uuid + UUID_NEW_BM_OFFSET;
890 get_random_bytes(&uuid, sizeof(u64));
891 drbd_uuid_set(device, UI_BITMAP, uuid);
892 drbd_print_uuids(device, "updated sync UUID");
893 drbd_md_sync(device);
895 sock = &peer_device->connection->data;
896 p = drbd_prepare_command(peer_device, sock);
898 p->uuid = cpu_to_be64(uuid);
899 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
903 /* communicated if (agreed_features & DRBD_FF_WSAME) */
905 assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p,
906 struct request_queue *q)
909 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
910 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
911 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
912 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
913 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
914 p->qlim->discard_enabled = blk_queue_discard(q);
915 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
917 q = device->rq_queue;
918 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
919 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
920 p->qlim->alignment_offset = 0;
921 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
922 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
923 p->qlim->discard_enabled = 0;
924 p->qlim->write_same_capable = 0;
928 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
930 struct drbd_device *device = peer_device->device;
931 struct drbd_socket *sock;
933 sector_t d_size, u_size;
935 unsigned int max_bio_size;
936 unsigned int packet_size;
938 sock = &peer_device->connection->data;
939 p = drbd_prepare_command(peer_device, sock);
943 packet_size = sizeof(*p);
944 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
945 packet_size += sizeof(p->qlim[0]);
947 memset(p, 0, packet_size);
948 if (get_ldev_if_state(device, D_NEGOTIATING)) {
949 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
950 d_size = drbd_get_max_capacity(device->ldev);
952 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
954 q_order_type = drbd_queue_order_type(device);
955 max_bio_size = queue_max_hw_sectors(q) << 9;
956 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
957 assign_p_sizes_qlim(device, p, q);
962 q_order_type = QUEUE_ORDERED_NONE;
963 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
964 assign_p_sizes_qlim(device, p, NULL);
967 if (peer_device->connection->agreed_pro_version <= 94)
968 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
969 else if (peer_device->connection->agreed_pro_version < 100)
970 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
972 p->d_size = cpu_to_be64(d_size);
973 p->u_size = cpu_to_be64(u_size);
977 p->c_size = cpu_to_be64(get_capacity(device->vdisk));
978 p->max_bio_size = cpu_to_be32(max_bio_size);
979 p->queue_order_type = cpu_to_be16(q_order_type);
980 p->dds_flags = cpu_to_be16(flags);
982 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
986 * drbd_send_current_state() - Sends the drbd state to the peer
987 * @peer_device: DRBD peer device.
989 int drbd_send_current_state(struct drbd_peer_device *peer_device)
991 struct drbd_socket *sock;
994 sock = &peer_device->connection->data;
995 p = drbd_prepare_command(peer_device, sock);
998 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
999 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1003 * drbd_send_state() - After a state change, sends the new state to the peer
1004 * @peer_device: DRBD peer device.
1005 * @state: the state to send, not necessarily the current state.
1007 * Each state change queues an "after_state_ch" work, which will eventually
1008 * send the resulting new state to the peer. If more state changes happen
1009 * between queuing and processing of the after_state_ch work, we still
1010 * want to send each intermediary state in the order it occurred.
1012 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1014 struct drbd_socket *sock;
1017 sock = &peer_device->connection->data;
1018 p = drbd_prepare_command(peer_device, sock);
1021 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1022 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1025 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1027 struct drbd_socket *sock;
1028 struct p_req_state *p;
1030 sock = &peer_device->connection->data;
1031 p = drbd_prepare_command(peer_device, sock);
1034 p->mask = cpu_to_be32(mask.i);
1035 p->val = cpu_to_be32(val.i);
1036 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1039 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1041 enum drbd_packet cmd;
1042 struct drbd_socket *sock;
1043 struct p_req_state *p;
1045 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1046 sock = &connection->data;
1047 p = conn_prepare_command(connection, sock);
1050 p->mask = cpu_to_be32(mask.i);
1051 p->val = cpu_to_be32(val.i);
1052 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1055 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1057 struct drbd_socket *sock;
1058 struct p_req_state_reply *p;
1060 sock = &peer_device->connection->meta;
1061 p = drbd_prepare_command(peer_device, sock);
1063 p->retcode = cpu_to_be32(retcode);
1064 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1068 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1070 struct drbd_socket *sock;
1071 struct p_req_state_reply *p;
1072 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1074 sock = &connection->meta;
1075 p = conn_prepare_command(connection, sock);
1077 p->retcode = cpu_to_be32(retcode);
1078 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1082 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1084 BUG_ON(code & ~0xf);
1085 p->encoding = (p->encoding & ~0xf) | code;
1088 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1090 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1093 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1096 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1099 static int fill_bitmap_rle_bits(struct drbd_device *device,
1100 struct p_compressed_bm *p,
1102 struct bm_xfer_ctx *c)
1104 struct bitstream bs;
1105 unsigned long plain_bits;
1112 /* may we use this feature? */
1114 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1116 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1119 if (c->bit_offset >= c->bm_bits)
1120 return 0; /* nothing to do. */
1122 /* use at most thus many bytes */
1123 bitstream_init(&bs, p->code, size, 0);
1124 memset(p->code, 0, size);
1125 /* plain bits covered in this code string */
1128 /* p->encoding & 0x80 stores whether the first run length is set.
1129 * bit offset is implicit.
1130 * start with toggle == 2 to be able to tell the first iteration */
1133 /* see how much plain bits we can stuff into one packet
1134 * using RLE and VLI. */
1136 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1137 : _drbd_bm_find_next(device, c->bit_offset);
1140 rl = tmp - c->bit_offset;
1142 if (toggle == 2) { /* first iteration */
1144 /* the first checked bit was set,
1145 * store start value, */
1146 dcbp_set_start(p, 1);
1147 /* but skip encoding of zero run length */
1151 dcbp_set_start(p, 0);
1154 /* paranoia: catch zero runlength.
1155 * can only happen if bitmap is modified while we scan it. */
1157 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1158 "t:%u bo:%lu\n", toggle, c->bit_offset);
1162 bits = vli_encode_bits(&bs, rl);
1163 if (bits == -ENOBUFS) /* buffer full */
1166 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1172 c->bit_offset = tmp;
1173 } while (c->bit_offset < c->bm_bits);
1175 len = bs.cur.b - p->code + !!bs.cur.bit;
1177 if (plain_bits < (len << 3)) {
1178 /* incompressible with this method.
1179 * we need to rewind both word and bit position. */
1180 c->bit_offset -= plain_bits;
1181 bm_xfer_ctx_bit_to_word_offset(c);
1182 c->bit_offset = c->word_offset * BITS_PER_LONG;
1186 /* RLE + VLI was able to compress it just fine.
1187 * update c->word_offset. */
1188 bm_xfer_ctx_bit_to_word_offset(c);
1190 /* store pad_bits */
1191 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1197 * send_bitmap_rle_or_plain
1199 * Return 0 when done, 1 when another iteration is needed, and a negative error
1200 * code upon failure.
1203 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1205 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1206 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1207 struct p_compressed_bm *p = sock->sbuf + header_size;
1210 len = fill_bitmap_rle_bits(device, p,
1211 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1216 dcbp_set_code(p, RLE_VLI_Bits);
1217 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1218 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1221 c->bytes[0] += header_size + sizeof(*p) + len;
1223 if (c->bit_offset >= c->bm_bits)
1226 /* was not compressible.
1227 * send a buffer full of plain text bits instead. */
1228 unsigned int data_size;
1229 unsigned long num_words;
1230 unsigned long *p = sock->sbuf + header_size;
1232 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1233 num_words = min_t(size_t, data_size / sizeof(*p),
1234 c->bm_words - c->word_offset);
1235 len = num_words * sizeof(*p);
1237 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1238 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1239 c->word_offset += num_words;
1240 c->bit_offset = c->word_offset * BITS_PER_LONG;
1243 c->bytes[1] += header_size + len;
1245 if (c->bit_offset > c->bm_bits)
1246 c->bit_offset = c->bm_bits;
1250 INFO_bm_xfer_stats(device, "send", c);
1258 /* See the comment at receive_bitmap() */
1259 static int _drbd_send_bitmap(struct drbd_device *device)
1261 struct bm_xfer_ctx c;
1264 if (!expect(device->bitmap))
1267 if (get_ldev(device)) {
1268 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1269 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1270 drbd_bm_set_all(device);
1271 if (drbd_bm_write(device)) {
1272 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1273 * but otherwise process as per normal - need to tell other
1274 * side that a full resync is required! */
1275 drbd_err(device, "Failed to write bitmap to disk!\n");
1277 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1278 drbd_md_sync(device);
1284 c = (struct bm_xfer_ctx) {
1285 .bm_bits = drbd_bm_bits(device),
1286 .bm_words = drbd_bm_words(device),
1290 err = send_bitmap_rle_or_plain(device, &c);
1296 int drbd_send_bitmap(struct drbd_device *device)
1298 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1301 mutex_lock(&sock->mutex);
1303 err = !_drbd_send_bitmap(device);
1304 mutex_unlock(&sock->mutex);
1308 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1310 struct drbd_socket *sock;
1311 struct p_barrier_ack *p;
1313 if (connection->cstate < C_WF_REPORT_PARAMS)
1316 sock = &connection->meta;
1317 p = conn_prepare_command(connection, sock);
1320 p->barrier = barrier_nr;
1321 p->set_size = cpu_to_be32(set_size);
1322 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1326 * _drbd_send_ack() - Sends an ack packet
1327 * @peer_device: DRBD peer device.
1328 * @cmd: Packet command code.
1329 * @sector: sector, needs to be in big endian byte order
1330 * @blksize: size in byte, needs to be in big endian byte order
1331 * @block_id: Id, big endian byte order
1333 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1334 u64 sector, u32 blksize, u64 block_id)
1336 struct drbd_socket *sock;
1337 struct p_block_ack *p;
1339 if (peer_device->device->state.conn < C_CONNECTED)
1342 sock = &peer_device->connection->meta;
1343 p = drbd_prepare_command(peer_device, sock);
1347 p->block_id = block_id;
1348 p->blksize = blksize;
1349 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1350 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1353 /* dp->sector and dp->block_id already/still in network byte order,
1354 * data_size is payload size according to dp->head,
1355 * and may need to be corrected for digest size. */
1356 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1357 struct p_data *dp, int data_size)
1359 if (peer_device->connection->peer_integrity_tfm)
1360 data_size -= crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm);
1361 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1365 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1366 struct p_block_req *rp)
1368 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1372 * drbd_send_ack() - Sends an ack packet
1373 * @peer_device: DRBD peer device
1374 * @cmd: packet command code
1375 * @peer_req: peer request
1377 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1378 struct drbd_peer_request *peer_req)
1380 return _drbd_send_ack(peer_device, cmd,
1381 cpu_to_be64(peer_req->i.sector),
1382 cpu_to_be32(peer_req->i.size),
1383 peer_req->block_id);
1386 /* This function misuses the block_id field to signal if the blocks
1387 * are is sync or not. */
1388 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1389 sector_t sector, int blksize, u64 block_id)
1391 return _drbd_send_ack(peer_device, cmd,
1392 cpu_to_be64(sector),
1393 cpu_to_be32(blksize),
1394 cpu_to_be64(block_id));
1397 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1398 struct drbd_peer_request *peer_req)
1400 struct drbd_socket *sock;
1401 struct p_block_desc *p;
1403 sock = &peer_device->connection->data;
1404 p = drbd_prepare_command(peer_device, sock);
1407 p->sector = cpu_to_be64(peer_req->i.sector);
1408 p->blksize = cpu_to_be32(peer_req->i.size);
1410 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1413 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1414 sector_t sector, int size, u64 block_id)
1416 struct drbd_socket *sock;
1417 struct p_block_req *p;
1419 sock = &peer_device->connection->data;
1420 p = drbd_prepare_command(peer_device, sock);
1423 p->sector = cpu_to_be64(sector);
1424 p->block_id = block_id;
1425 p->blksize = cpu_to_be32(size);
1426 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1429 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1430 void *digest, int digest_size, enum drbd_packet cmd)
1432 struct drbd_socket *sock;
1433 struct p_block_req *p;
1435 /* FIXME: Put the digest into the preallocated socket buffer. */
1437 sock = &peer_device->connection->data;
1438 p = drbd_prepare_command(peer_device, sock);
1441 p->sector = cpu_to_be64(sector);
1442 p->block_id = ID_SYNCER /* unused */;
1443 p->blksize = cpu_to_be32(size);
1444 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1447 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1449 struct drbd_socket *sock;
1450 struct p_block_req *p;
1452 sock = &peer_device->connection->data;
1453 p = drbd_prepare_command(peer_device, sock);
1456 p->sector = cpu_to_be64(sector);
1457 p->block_id = ID_SYNCER /* unused */;
1458 p->blksize = cpu_to_be32(size);
1459 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1462 /* called on sndtimeo
1463 * returns false if we should retry,
1464 * true if we think connection is dead
1466 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1469 /* long elapsed = (long)(jiffies - device->last_received); */
1471 drop_it = connection->meta.socket == sock
1472 || !connection->ack_receiver.task
1473 || get_t_state(&connection->ack_receiver) != RUNNING
1474 || connection->cstate < C_WF_REPORT_PARAMS;
1479 drop_it = !--connection->ko_count;
1481 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1482 current->comm, current->pid, connection->ko_count);
1483 request_ping(connection);
1486 return drop_it; /* && (device->state == R_PRIMARY) */;
1489 static void drbd_update_congested(struct drbd_connection *connection)
1491 struct sock *sk = connection->data.socket->sk;
1492 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1493 set_bit(NET_CONGESTED, &connection->flags);
1496 /* The idea of sendpage seems to be to put some kind of reference
1497 * to the page into the skb, and to hand it over to the NIC. In
1498 * this process get_page() gets called.
1500 * As soon as the page was really sent over the network put_page()
1501 * gets called by some part of the network layer. [ NIC driver? ]
1503 * [ get_page() / put_page() increment/decrement the count. If count
1504 * reaches 0 the page will be freed. ]
1506 * This works nicely with pages from FSs.
1507 * But this means that in protocol A we might signal IO completion too early!
1509 * In order not to corrupt data during a resync we must make sure
1510 * that we do not reuse our own buffer pages (EEs) to early, therefore
1511 * we have the net_ee list.
1513 * XFS seems to have problems, still, it submits pages with page_count == 0!
1514 * As a workaround, we disable sendpage on pages
1515 * with page_count == 0 or PageSlab.
1517 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1518 int offset, size_t size, unsigned msg_flags)
1520 struct socket *socket;
1524 socket = peer_device->connection->data.socket;
1525 addr = kmap(page) + offset;
1526 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1529 peer_device->device->send_cnt += size >> 9;
1533 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1534 int offset, size_t size, unsigned msg_flags)
1536 struct socket *socket = peer_device->connection->data.socket;
1540 /* e.g. XFS meta- & log-data is in slab pages, which have a
1541 * page_count of 0 and/or have PageSlab() set.
1542 * we cannot use send_page for those, as that does get_page();
1543 * put_page(); and would cause either a VM_BUG directly, or
1544 * __page_cache_release a page that would actually still be referenced
1545 * by someone, leading to some obscure delayed Oops somewhere else. */
1546 if (drbd_disable_sendpage || !sendpage_ok(page))
1547 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1549 msg_flags |= MSG_NOSIGNAL;
1550 drbd_update_congested(peer_device->connection);
1554 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1556 if (sent == -EAGAIN) {
1557 if (we_should_drop_the_connection(peer_device->connection, socket))
1561 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1562 __func__, (int)size, len, sent);
1569 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1570 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1574 peer_device->device->send_cnt += size >> 9;
1579 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1581 struct bio_vec bvec;
1582 struct bvec_iter iter;
1584 /* hint all but last page with MSG_MORE */
1585 bio_for_each_segment(bvec, bio, iter) {
1588 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1589 bvec.bv_offset, bvec.bv_len,
1590 bio_iter_last(bvec, iter)
1594 /* REQ_OP_WRITE_SAME has only one segment */
1595 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1601 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1603 struct bio_vec bvec;
1604 struct bvec_iter iter;
1606 /* hint all but last page with MSG_MORE */
1607 bio_for_each_segment(bvec, bio, iter) {
1610 err = _drbd_send_page(peer_device, bvec.bv_page,
1611 bvec.bv_offset, bvec.bv_len,
1612 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1615 /* REQ_OP_WRITE_SAME has only one segment */
1616 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1622 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1623 struct drbd_peer_request *peer_req)
1625 struct page *page = peer_req->pages;
1626 unsigned len = peer_req->i.size;
1629 /* hint all but last page with MSG_MORE */
1630 page_chain_for_each(page) {
1631 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1633 err = _drbd_send_page(peer_device, page, 0, l,
1634 page_chain_next(page) ? MSG_MORE : 0);
1642 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1645 if (connection->agreed_pro_version >= 95)
1646 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1647 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1648 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1649 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1650 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1651 (bio_op(bio) == REQ_OP_WRITE_ZEROES ?
1652 ((connection->agreed_features & DRBD_FF_WZEROES) ?
1653 (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0))
1657 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1660 /* Used to send write or TRIM aka REQ_OP_DISCARD requests
1661 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1663 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1665 struct drbd_device *device = peer_device->device;
1666 struct drbd_socket *sock;
1668 struct p_wsame *wsame = NULL;
1670 unsigned int dp_flags = 0;
1674 sock = &peer_device->connection->data;
1675 p = drbd_prepare_command(peer_device, sock);
1676 digest_size = peer_device->connection->integrity_tfm ?
1677 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1681 p->sector = cpu_to_be64(req->i.sector);
1682 p->block_id = (unsigned long)req;
1683 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1684 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1685 if (device->state.conn >= C_SYNC_SOURCE &&
1686 device->state.conn <= C_PAUSED_SYNC_T)
1687 dp_flags |= DP_MAY_SET_IN_SYNC;
1688 if (peer_device->connection->agreed_pro_version >= 100) {
1689 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1690 dp_flags |= DP_SEND_RECEIVE_ACK;
1691 /* During resync, request an explicit write ack,
1692 * even in protocol != C */
1693 if (req->rq_state & RQ_EXP_WRITE_ACK
1694 || (dp_flags & DP_MAY_SET_IN_SYNC))
1695 dp_flags |= DP_SEND_WRITE_ACK;
1697 p->dp_flags = cpu_to_be32(dp_flags);
1699 if (dp_flags & (DP_DISCARD|DP_ZEROES)) {
1700 enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM;
1701 struct p_trim *t = (struct p_trim*)p;
1702 t->size = cpu_to_be32(req->i.size);
1703 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0);
1706 if (dp_flags & DP_WSAME) {
1707 /* this will only work if DRBD_FF_WSAME is set AND the
1708 * handshake agreed that all nodes and backend devices are
1709 * WRITE_SAME capable and agree on logical_block_size */
1710 wsame = (struct p_wsame*)p;
1711 digest_out = wsame + 1;
1712 wsame->size = cpu_to_be32(req->i.size);
1716 /* our digest is still only over the payload.
1717 * TRIM does not carry any payload. */
1719 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1722 __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1723 sizeof(*wsame) + digest_size, NULL,
1724 bio_iovec(req->master_bio).bv_len);
1727 __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1728 sizeof(*p) + digest_size, NULL, req->i.size);
1730 /* For protocol A, we have to memcpy the payload into
1731 * socket buffers, as we may complete right away
1732 * as soon as we handed it over to tcp, at which point the data
1733 * pages may become invalid.
1735 * For data-integrity enabled, we copy it as well, so we can be
1736 * sure that even if the bio pages may still be modified, it
1737 * won't change the data on the wire, thus if the digest checks
1738 * out ok after sending on this side, but does not fit on the
1739 * receiving side, we sure have detected corruption elsewhere.
1741 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1742 err = _drbd_send_bio(peer_device, req->master_bio);
1744 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1746 /* double check digest, sometimes buffers have been modified in flight. */
1747 if (digest_size > 0 && digest_size <= 64) {
1748 /* 64 byte, 512 bit, is the largest digest size
1749 * currently supported in kernel crypto. */
1750 unsigned char digest[64];
1751 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1752 if (memcmp(p + 1, digest, digest_size)) {
1754 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1755 (unsigned long long)req->i.sector, req->i.size);
1757 } /* else if (digest_size > 64) {
1758 ... Be noisy about digest too large ...
1762 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1767 /* answer packet, used to send data back for read requests:
1768 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1769 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1771 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1772 struct drbd_peer_request *peer_req)
1774 struct drbd_device *device = peer_device->device;
1775 struct drbd_socket *sock;
1780 sock = &peer_device->connection->data;
1781 p = drbd_prepare_command(peer_device, sock);
1783 digest_size = peer_device->connection->integrity_tfm ?
1784 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1788 p->sector = cpu_to_be64(peer_req->i.sector);
1789 p->block_id = peer_req->block_id;
1790 p->seq_num = 0; /* unused */
1793 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1794 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1796 err = _drbd_send_zc_ee(peer_device, peer_req);
1797 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1802 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1804 struct drbd_socket *sock;
1805 struct p_block_desc *p;
1807 sock = &peer_device->connection->data;
1808 p = drbd_prepare_command(peer_device, sock);
1811 p->sector = cpu_to_be64(req->i.sector);
1812 p->blksize = cpu_to_be32(req->i.size);
1813 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1817 drbd_send distinguishes two cases:
1819 Packets sent via the data socket "sock"
1820 and packets sent via the meta data socket "msock"
1823 -----------------+-------------------------+------------------------------
1824 timeout conf.timeout / 2 conf.timeout / 2
1825 timeout action send a ping via msock Abort communication
1826 and close all sockets
1830 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1832 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1833 void *buf, size_t size, unsigned msg_flags)
1835 struct kvec iov = {.iov_base = buf, .iov_len = size};
1836 struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL};
1842 /* THINK if (signal_pending) return ... ? */
1844 iov_iter_kvec(&msg.msg_iter, WRITE, &iov, 1, size);
1846 if (sock == connection->data.socket) {
1848 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1850 drbd_update_congested(connection);
1853 rv = sock_sendmsg(sock, &msg);
1854 if (rv == -EAGAIN) {
1855 if (we_should_drop_the_connection(connection, sock))
1861 flush_signals(current);
1867 } while (sent < size);
1869 if (sock == connection->data.socket)
1870 clear_bit(NET_CONGESTED, &connection->flags);
1873 if (rv != -EAGAIN) {
1874 drbd_err(connection, "%s_sendmsg returned %d\n",
1875 sock == connection->meta.socket ? "msock" : "sock",
1877 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1879 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1886 * drbd_send_all - Send an entire buffer
1888 * Returns 0 upon success and a negative error value otherwise.
1890 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1891 size_t size, unsigned msg_flags)
1895 err = drbd_send(connection, sock, buffer, size, msg_flags);
1903 static int drbd_open(struct block_device *bdev, fmode_t mode)
1905 struct drbd_device *device = bdev->bd_disk->private_data;
1906 unsigned long flags;
1909 mutex_lock(&drbd_main_mutex);
1910 spin_lock_irqsave(&device->resource->req_lock, flags);
1911 /* to have a stable device->state.role
1912 * and no race with updating open_cnt */
1914 if (device->state.role != R_PRIMARY) {
1915 if (mode & FMODE_WRITE)
1917 else if (!drbd_allow_oos)
1923 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1924 mutex_unlock(&drbd_main_mutex);
1929 static void drbd_release(struct gendisk *gd, fmode_t mode)
1931 struct drbd_device *device = gd->private_data;
1932 mutex_lock(&drbd_main_mutex);
1934 mutex_unlock(&drbd_main_mutex);
1937 /* need to hold resource->req_lock */
1938 void drbd_queue_unplug(struct drbd_device *device)
1940 if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1941 D_ASSERT(device, device->state.role == R_PRIMARY);
1942 if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1943 drbd_queue_work_if_unqueued(
1944 &first_peer_device(device)->connection->sender_work,
1945 &device->unplug_work);
1950 static void drbd_set_defaults(struct drbd_device *device)
1952 /* Beware! The actual layout differs
1953 * between big endian and little endian */
1954 device->state = (union drbd_dev_state) {
1955 { .role = R_SECONDARY,
1957 .conn = C_STANDALONE,
1963 void drbd_init_set_defaults(struct drbd_device *device)
1965 /* the memset(,0,) did most of this.
1966 * note: only assignments, no allocation in here */
1968 drbd_set_defaults(device);
1970 atomic_set(&device->ap_bio_cnt, 0);
1971 atomic_set(&device->ap_actlog_cnt, 0);
1972 atomic_set(&device->ap_pending_cnt, 0);
1973 atomic_set(&device->rs_pending_cnt, 0);
1974 atomic_set(&device->unacked_cnt, 0);
1975 atomic_set(&device->local_cnt, 0);
1976 atomic_set(&device->pp_in_use_by_net, 0);
1977 atomic_set(&device->rs_sect_in, 0);
1978 atomic_set(&device->rs_sect_ev, 0);
1979 atomic_set(&device->ap_in_flight, 0);
1980 atomic_set(&device->md_io.in_use, 0);
1982 mutex_init(&device->own_state_mutex);
1983 device->state_mutex = &device->own_state_mutex;
1985 spin_lock_init(&device->al_lock);
1986 spin_lock_init(&device->peer_seq_lock);
1988 INIT_LIST_HEAD(&device->active_ee);
1989 INIT_LIST_HEAD(&device->sync_ee);
1990 INIT_LIST_HEAD(&device->done_ee);
1991 INIT_LIST_HEAD(&device->read_ee);
1992 INIT_LIST_HEAD(&device->net_ee);
1993 INIT_LIST_HEAD(&device->resync_reads);
1994 INIT_LIST_HEAD(&device->resync_work.list);
1995 INIT_LIST_HEAD(&device->unplug_work.list);
1996 INIT_LIST_HEAD(&device->bm_io_work.w.list);
1997 INIT_LIST_HEAD(&device->pending_master_completion[0]);
1998 INIT_LIST_HEAD(&device->pending_master_completion[1]);
1999 INIT_LIST_HEAD(&device->pending_completion[0]);
2000 INIT_LIST_HEAD(&device->pending_completion[1]);
2002 device->resync_work.cb = w_resync_timer;
2003 device->unplug_work.cb = w_send_write_hint;
2004 device->bm_io_work.w.cb = w_bitmap_io;
2006 timer_setup(&device->resync_timer, resync_timer_fn, 0);
2007 timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0);
2008 timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0);
2009 timer_setup(&device->request_timer, request_timer_fn, 0);
2011 init_waitqueue_head(&device->misc_wait);
2012 init_waitqueue_head(&device->state_wait);
2013 init_waitqueue_head(&device->ee_wait);
2014 init_waitqueue_head(&device->al_wait);
2015 init_waitqueue_head(&device->seq_wait);
2017 device->resync_wenr = LC_FREE;
2018 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2019 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2022 void drbd_set_my_capacity(struct drbd_device *device, sector_t size)
2026 set_capacity_and_notify(device->vdisk, size);
2028 drbd_info(device, "size = %s (%llu KB)\n",
2029 ppsize(ppb, size>>1), (unsigned long long)size>>1);
2032 void drbd_device_cleanup(struct drbd_device *device)
2035 if (first_peer_device(device)->connection->receiver.t_state != NONE)
2036 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2037 first_peer_device(device)->connection->receiver.t_state);
2039 device->al_writ_cnt =
2040 device->bm_writ_cnt =
2048 device->rs_failed = 0;
2049 device->rs_last_events = 0;
2050 device->rs_last_sect_ev = 0;
2051 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2052 device->rs_mark_left[i] = 0;
2053 device->rs_mark_time[i] = 0;
2055 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2057 set_capacity_and_notify(device->vdisk, 0);
2058 if (device->bitmap) {
2059 /* maybe never allocated. */
2060 drbd_bm_resize(device, 0, 1);
2061 drbd_bm_cleanup(device);
2064 drbd_backing_dev_free(device, device->ldev);
2065 device->ldev = NULL;
2067 clear_bit(AL_SUSPENDED, &device->flags);
2069 D_ASSERT(device, list_empty(&device->active_ee));
2070 D_ASSERT(device, list_empty(&device->sync_ee));
2071 D_ASSERT(device, list_empty(&device->done_ee));
2072 D_ASSERT(device, list_empty(&device->read_ee));
2073 D_ASSERT(device, list_empty(&device->net_ee));
2074 D_ASSERT(device, list_empty(&device->resync_reads));
2075 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2076 D_ASSERT(device, list_empty(&device->resync_work.list));
2077 D_ASSERT(device, list_empty(&device->unplug_work.list));
2079 drbd_set_defaults(device);
2083 static void drbd_destroy_mempools(void)
2087 while (drbd_pp_pool) {
2088 page = drbd_pp_pool;
2089 drbd_pp_pool = (struct page *)page_private(page);
2094 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2096 bioset_exit(&drbd_io_bio_set);
2097 bioset_exit(&drbd_md_io_bio_set);
2098 mempool_exit(&drbd_md_io_page_pool);
2099 mempool_exit(&drbd_ee_mempool);
2100 mempool_exit(&drbd_request_mempool);
2101 kmem_cache_destroy(drbd_ee_cache);
2102 kmem_cache_destroy(drbd_request_cache);
2103 kmem_cache_destroy(drbd_bm_ext_cache);
2104 kmem_cache_destroy(drbd_al_ext_cache);
2106 drbd_ee_cache = NULL;
2107 drbd_request_cache = NULL;
2108 drbd_bm_ext_cache = NULL;
2109 drbd_al_ext_cache = NULL;
2114 static int drbd_create_mempools(void)
2117 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2121 drbd_request_cache = kmem_cache_create(
2122 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2123 if (drbd_request_cache == NULL)
2126 drbd_ee_cache = kmem_cache_create(
2127 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2128 if (drbd_ee_cache == NULL)
2131 drbd_bm_ext_cache = kmem_cache_create(
2132 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2133 if (drbd_bm_ext_cache == NULL)
2136 drbd_al_ext_cache = kmem_cache_create(
2137 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2138 if (drbd_al_ext_cache == NULL)
2142 ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0);
2146 ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0,
2151 ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0);
2155 ret = mempool_init_slab_pool(&drbd_request_mempool, number,
2156 drbd_request_cache);
2160 ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache);
2164 for (i = 0; i < number; i++) {
2165 page = alloc_page(GFP_HIGHUSER);
2168 set_page_private(page, (unsigned long)drbd_pp_pool);
2169 drbd_pp_pool = page;
2171 drbd_pp_vacant = number;
2176 drbd_destroy_mempools(); /* in case we allocated some */
2180 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2184 rr = drbd_free_peer_reqs(device, &device->active_ee);
2186 drbd_err(device, "%d EEs in active list found!\n", rr);
2188 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2190 drbd_err(device, "%d EEs in sync list found!\n", rr);
2192 rr = drbd_free_peer_reqs(device, &device->read_ee);
2194 drbd_err(device, "%d EEs in read list found!\n", rr);
2196 rr = drbd_free_peer_reqs(device, &device->done_ee);
2198 drbd_err(device, "%d EEs in done list found!\n", rr);
2200 rr = drbd_free_peer_reqs(device, &device->net_ee);
2202 drbd_err(device, "%d EEs in net list found!\n", rr);
2205 /* caution. no locking. */
2206 void drbd_destroy_device(struct kref *kref)
2208 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2209 struct drbd_resource *resource = device->resource;
2210 struct drbd_peer_device *peer_device, *tmp_peer_device;
2212 del_timer_sync(&device->request_timer);
2214 /* paranoia asserts */
2215 D_ASSERT(device, device->open_cnt == 0);
2216 /* end paranoia asserts */
2218 /* cleanup stuff that may have been allocated during
2219 * device (re-)configuration or state changes */
2221 drbd_backing_dev_free(device, device->ldev);
2222 device->ldev = NULL;
2224 drbd_release_all_peer_reqs(device);
2226 lc_destroy(device->act_log);
2227 lc_destroy(device->resync);
2229 kfree(device->p_uuid);
2230 /* device->p_uuid = NULL; */
2232 if (device->bitmap) /* should no longer be there. */
2233 drbd_bm_cleanup(device);
2234 __free_page(device->md_io.page);
2235 blk_cleanup_disk(device->vdisk);
2236 kfree(device->rs_plan_s);
2238 /* not for_each_connection(connection, resource):
2239 * those may have been cleaned up and disassociated already.
2241 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2242 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2245 memset(device, 0xfd, sizeof(*device));
2247 kref_put(&resource->kref, drbd_destroy_resource);
2250 /* One global retry thread, if we need to push back some bio and have it
2251 * reinserted through our make request function.
2253 static struct retry_worker {
2254 struct workqueue_struct *wq;
2255 struct work_struct worker;
2258 struct list_head writes;
2261 static void do_retry(struct work_struct *ws)
2263 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2265 struct drbd_request *req, *tmp;
2267 spin_lock_irq(&retry->lock);
2268 list_splice_init(&retry->writes, &writes);
2269 spin_unlock_irq(&retry->lock);
2271 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2272 struct drbd_device *device = req->device;
2273 struct bio *bio = req->master_bio;
2277 expect(atomic_read(&req->completion_ref) == 0) &&
2278 expect(req->rq_state & RQ_POSTPONED) &&
2279 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2280 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2283 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2284 req, atomic_read(&req->completion_ref),
2287 /* We still need to put one kref associated with the
2288 * "completion_ref" going zero in the code path that queued it
2289 * here. The request object may still be referenced by a
2290 * frozen local req->private_bio, in case we force-detached.
2292 kref_put(&req->kref, drbd_req_destroy);
2294 /* A single suspended or otherwise blocking device may stall
2295 * all others as well. Fortunately, this code path is to
2296 * recover from a situation that "should not happen":
2297 * concurrent writes in multi-primary setup.
2298 * In a "normal" lifecycle, this workqueue is supposed to be
2299 * destroyed without ever doing anything.
2300 * If it turns out to be an issue anyways, we can do per
2301 * resource (replication group) or per device (minor) retry
2302 * workqueues instead.
2305 /* We are not just doing submit_bio_noacct(),
2306 * as we want to keep the start_time information. */
2308 __drbd_make_request(device, bio);
2312 /* called via drbd_req_put_completion_ref(),
2313 * holds resource->req_lock */
2314 void drbd_restart_request(struct drbd_request *req)
2316 unsigned long flags;
2317 spin_lock_irqsave(&retry.lock, flags);
2318 list_move_tail(&req->tl_requests, &retry.writes);
2319 spin_unlock_irqrestore(&retry.lock, flags);
2321 /* Drop the extra reference that would otherwise
2322 * have been dropped by complete_master_bio.
2323 * do_retry() needs to grab a new one. */
2324 dec_ap_bio(req->device);
2326 queue_work(retry.wq, &retry.worker);
2329 void drbd_destroy_resource(struct kref *kref)
2331 struct drbd_resource *resource =
2332 container_of(kref, struct drbd_resource, kref);
2334 idr_destroy(&resource->devices);
2335 free_cpumask_var(resource->cpu_mask);
2336 kfree(resource->name);
2337 memset(resource, 0xf2, sizeof(*resource));
2341 void drbd_free_resource(struct drbd_resource *resource)
2343 struct drbd_connection *connection, *tmp;
2345 for_each_connection_safe(connection, tmp, resource) {
2346 list_del(&connection->connections);
2347 drbd_debugfs_connection_cleanup(connection);
2348 kref_put(&connection->kref, drbd_destroy_connection);
2350 drbd_debugfs_resource_cleanup(resource);
2351 kref_put(&resource->kref, drbd_destroy_resource);
2354 static void drbd_cleanup(void)
2357 struct drbd_device *device;
2358 struct drbd_resource *resource, *tmp;
2360 /* first remove proc,
2361 * drbdsetup uses it's presence to detect
2362 * whether DRBD is loaded.
2363 * If we would get stuck in proc removal,
2364 * but have netlink already deregistered,
2365 * some drbdsetup commands may wait forever
2369 remove_proc_entry("drbd", NULL);
2372 destroy_workqueue(retry.wq);
2374 drbd_genl_unregister();
2376 idr_for_each_entry(&drbd_devices, device, i)
2377 drbd_delete_device(device);
2379 /* not _rcu since, no other updater anymore. Genl already unregistered */
2380 for_each_resource_safe(resource, tmp, &drbd_resources) {
2381 list_del(&resource->resources);
2382 drbd_free_resource(resource);
2385 drbd_debugfs_cleanup();
2387 drbd_destroy_mempools();
2388 unregister_blkdev(DRBD_MAJOR, "drbd");
2390 idr_destroy(&drbd_devices);
2392 pr_info("module cleanup done.\n");
2395 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2397 spin_lock_init(&wq->q_lock);
2398 INIT_LIST_HEAD(&wq->q);
2399 init_waitqueue_head(&wq->q_wait);
2402 struct completion_work {
2404 struct completion done;
2407 static int w_complete(struct drbd_work *w, int cancel)
2409 struct completion_work *completion_work =
2410 container_of(w, struct completion_work, w);
2412 complete(&completion_work->done);
2416 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2418 struct completion_work completion_work;
2420 completion_work.w.cb = w_complete;
2421 init_completion(&completion_work.done);
2422 drbd_queue_work(work_queue, &completion_work.w);
2423 wait_for_completion(&completion_work.done);
2426 struct drbd_resource *drbd_find_resource(const char *name)
2428 struct drbd_resource *resource;
2430 if (!name || !name[0])
2434 for_each_resource_rcu(resource, &drbd_resources) {
2435 if (!strcmp(resource->name, name)) {
2436 kref_get(&resource->kref);
2446 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2447 void *peer_addr, int peer_addr_len)
2449 struct drbd_resource *resource;
2450 struct drbd_connection *connection;
2453 for_each_resource_rcu(resource, &drbd_resources) {
2454 for_each_connection_rcu(connection, resource) {
2455 if (connection->my_addr_len == my_addr_len &&
2456 connection->peer_addr_len == peer_addr_len &&
2457 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2458 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2459 kref_get(&connection->kref);
2470 static int drbd_alloc_socket(struct drbd_socket *socket)
2472 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2475 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2481 static void drbd_free_socket(struct drbd_socket *socket)
2483 free_page((unsigned long) socket->sbuf);
2484 free_page((unsigned long) socket->rbuf);
2487 void conn_free_crypto(struct drbd_connection *connection)
2489 drbd_free_sock(connection);
2491 crypto_free_shash(connection->csums_tfm);
2492 crypto_free_shash(connection->verify_tfm);
2493 crypto_free_shash(connection->cram_hmac_tfm);
2494 crypto_free_shash(connection->integrity_tfm);
2495 crypto_free_shash(connection->peer_integrity_tfm);
2496 kfree(connection->int_dig_in);
2497 kfree(connection->int_dig_vv);
2499 connection->csums_tfm = NULL;
2500 connection->verify_tfm = NULL;
2501 connection->cram_hmac_tfm = NULL;
2502 connection->integrity_tfm = NULL;
2503 connection->peer_integrity_tfm = NULL;
2504 connection->int_dig_in = NULL;
2505 connection->int_dig_vv = NULL;
2508 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2510 struct drbd_connection *connection;
2511 cpumask_var_t new_cpu_mask;
2514 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2517 /* silently ignore cpu mask on UP kernel */
2518 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2519 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2520 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2521 if (err == -EOVERFLOW) {
2522 /* So what. mask it out. */
2523 cpumask_var_t tmp_cpu_mask;
2524 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2525 cpumask_setall(tmp_cpu_mask);
2526 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2527 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2529 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2531 free_cpumask_var(tmp_cpu_mask);
2536 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2537 /* retcode = ERR_CPU_MASK_PARSE; */
2541 resource->res_opts = *res_opts;
2542 if (cpumask_empty(new_cpu_mask))
2543 drbd_calc_cpu_mask(&new_cpu_mask);
2544 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2545 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2546 for_each_connection_rcu(connection, resource) {
2547 connection->receiver.reset_cpu_mask = 1;
2548 connection->ack_receiver.reset_cpu_mask = 1;
2549 connection->worker.reset_cpu_mask = 1;
2555 free_cpumask_var(new_cpu_mask);
2560 struct drbd_resource *drbd_create_resource(const char *name)
2562 struct drbd_resource *resource;
2564 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2567 resource->name = kstrdup(name, GFP_KERNEL);
2568 if (!resource->name)
2569 goto fail_free_resource;
2570 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2571 goto fail_free_name;
2572 kref_init(&resource->kref);
2573 idr_init(&resource->devices);
2574 INIT_LIST_HEAD(&resource->connections);
2575 resource->write_ordering = WO_BDEV_FLUSH;
2576 list_add_tail_rcu(&resource->resources, &drbd_resources);
2577 mutex_init(&resource->conf_update);
2578 mutex_init(&resource->adm_mutex);
2579 spin_lock_init(&resource->req_lock);
2580 drbd_debugfs_resource_add(resource);
2584 kfree(resource->name);
2591 /* caller must be under adm_mutex */
2592 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2594 struct drbd_resource *resource;
2595 struct drbd_connection *connection;
2597 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2601 if (drbd_alloc_socket(&connection->data))
2603 if (drbd_alloc_socket(&connection->meta))
2606 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2607 if (!connection->current_epoch)
2610 INIT_LIST_HEAD(&connection->transfer_log);
2612 INIT_LIST_HEAD(&connection->current_epoch->list);
2613 connection->epochs = 1;
2614 spin_lock_init(&connection->epoch_lock);
2616 connection->send.seen_any_write_yet = false;
2617 connection->send.current_epoch_nr = 0;
2618 connection->send.current_epoch_writes = 0;
2620 resource = drbd_create_resource(name);
2624 connection->cstate = C_STANDALONE;
2625 mutex_init(&connection->cstate_mutex);
2626 init_waitqueue_head(&connection->ping_wait);
2627 idr_init(&connection->peer_devices);
2629 drbd_init_workqueue(&connection->sender_work);
2630 mutex_init(&connection->data.mutex);
2631 mutex_init(&connection->meta.mutex);
2633 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2634 connection->receiver.connection = connection;
2635 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2636 connection->worker.connection = connection;
2637 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2638 connection->ack_receiver.connection = connection;
2640 kref_init(&connection->kref);
2642 connection->resource = resource;
2644 if (set_resource_options(resource, res_opts))
2647 kref_get(&resource->kref);
2648 list_add_tail_rcu(&connection->connections, &resource->connections);
2649 drbd_debugfs_connection_add(connection);
2653 list_del(&resource->resources);
2654 drbd_free_resource(resource);
2656 kfree(connection->current_epoch);
2657 drbd_free_socket(&connection->meta);
2658 drbd_free_socket(&connection->data);
2663 void drbd_destroy_connection(struct kref *kref)
2665 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2666 struct drbd_resource *resource = connection->resource;
2668 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
2669 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2670 kfree(connection->current_epoch);
2672 idr_destroy(&connection->peer_devices);
2674 drbd_free_socket(&connection->meta);
2675 drbd_free_socket(&connection->data);
2676 kfree(connection->int_dig_in);
2677 kfree(connection->int_dig_vv);
2678 memset(connection, 0xfc, sizeof(*connection));
2680 kref_put(&resource->kref, drbd_destroy_resource);
2683 static int init_submitter(struct drbd_device *device)
2685 /* opencoded create_singlethread_workqueue(),
2686 * to be able to say "drbd%d", ..., minor */
2688 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2689 if (!device->submit.wq)
2692 INIT_WORK(&device->submit.worker, do_submit);
2693 INIT_LIST_HEAD(&device->submit.writes);
2697 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2699 struct drbd_resource *resource = adm_ctx->resource;
2700 struct drbd_connection *connection;
2701 struct drbd_device *device;
2702 struct drbd_peer_device *peer_device, *tmp_peer_device;
2703 struct gendisk *disk;
2705 int vnr = adm_ctx->volume;
2706 enum drbd_ret_code err = ERR_NOMEM;
2708 device = minor_to_device(minor);
2710 return ERR_MINOR_OR_VOLUME_EXISTS;
2712 /* GFP_KERNEL, we are outside of all write-out paths */
2713 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2716 kref_init(&device->kref);
2718 kref_get(&resource->kref);
2719 device->resource = resource;
2720 device->minor = minor;
2723 drbd_init_set_defaults(device);
2725 disk = blk_alloc_disk(NUMA_NO_NODE);
2729 device->vdisk = disk;
2730 device->rq_queue = disk->queue;
2732 set_disk_ro(disk, true);
2734 disk->major = DRBD_MAJOR;
2735 disk->first_minor = minor;
2737 disk->fops = &drbd_ops;
2738 disk->flags |= GENHD_FL_NO_PART;
2739 sprintf(disk->disk_name, "drbd%d", minor);
2740 disk->private_data = device;
2742 blk_queue_write_cache(disk->queue, true, true);
2743 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2744 This triggers a max_bio_size message upon first attach or connect */
2745 blk_queue_max_hw_sectors(disk->queue, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2747 device->md_io.page = alloc_page(GFP_KERNEL);
2748 if (!device->md_io.page)
2749 goto out_no_io_page;
2751 if (drbd_bm_init(device))
2753 device->read_requests = RB_ROOT;
2754 device->write_requests = RB_ROOT;
2756 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2759 err = ERR_MINOR_OR_VOLUME_EXISTS;
2760 goto out_no_minor_idr;
2762 kref_get(&device->kref);
2764 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2767 err = ERR_MINOR_OR_VOLUME_EXISTS;
2768 goto out_idr_remove_minor;
2770 kref_get(&device->kref);
2772 INIT_LIST_HEAD(&device->peer_devices);
2773 INIT_LIST_HEAD(&device->pending_bitmap_io);
2774 for_each_connection(connection, resource) {
2775 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2777 goto out_idr_remove_from_resource;
2778 peer_device->connection = connection;
2779 peer_device->device = device;
2781 list_add(&peer_device->peer_devices, &device->peer_devices);
2782 kref_get(&device->kref);
2784 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2787 err = ERR_INVALID_REQUEST;
2788 goto out_idr_remove_from_resource;
2790 kref_get(&connection->kref);
2791 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2794 if (init_submitter(device)) {
2796 goto out_idr_remove_vol;
2799 err = add_disk(disk);
2801 goto out_idr_remove_vol;
2803 /* inherit the connection state */
2804 device->state.conn = first_connection(resource)->cstate;
2805 if (device->state.conn == C_WF_REPORT_PARAMS) {
2806 for_each_peer_device(peer_device, device)
2807 drbd_connected(peer_device);
2809 /* move to create_peer_device() */
2810 for_each_peer_device(peer_device, device)
2811 drbd_debugfs_peer_device_add(peer_device);
2812 drbd_debugfs_device_add(device);
2816 idr_remove(&connection->peer_devices, vnr);
2817 out_idr_remove_from_resource:
2818 for_each_connection(connection, resource) {
2819 peer_device = idr_remove(&connection->peer_devices, vnr);
2821 kref_put(&connection->kref, drbd_destroy_connection);
2823 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2824 list_del(&peer_device->peer_devices);
2827 idr_remove(&resource->devices, vnr);
2828 out_idr_remove_minor:
2829 idr_remove(&drbd_devices, minor);
2832 drbd_bm_cleanup(device);
2834 __free_page(device->md_io.page);
2836 blk_cleanup_disk(disk);
2838 kref_put(&resource->kref, drbd_destroy_resource);
2843 void drbd_delete_device(struct drbd_device *device)
2845 struct drbd_resource *resource = device->resource;
2846 struct drbd_connection *connection;
2847 struct drbd_peer_device *peer_device;
2849 /* move to free_peer_device() */
2850 for_each_peer_device(peer_device, device)
2851 drbd_debugfs_peer_device_cleanup(peer_device);
2852 drbd_debugfs_device_cleanup(device);
2853 for_each_connection(connection, resource) {
2854 idr_remove(&connection->peer_devices, device->vnr);
2855 kref_put(&device->kref, drbd_destroy_device);
2857 idr_remove(&resource->devices, device->vnr);
2858 kref_put(&device->kref, drbd_destroy_device);
2859 idr_remove(&drbd_devices, device_to_minor(device));
2860 kref_put(&device->kref, drbd_destroy_device);
2861 del_gendisk(device->vdisk);
2863 kref_put(&device->kref, drbd_destroy_device);
2866 static int __init drbd_init(void)
2870 if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2871 pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2875 drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2879 err = register_blkdev(DRBD_MAJOR, "drbd");
2881 pr_err("unable to register block device major %d\n",
2887 * allocate all necessary structs
2889 init_waitqueue_head(&drbd_pp_wait);
2891 drbd_proc = NULL; /* play safe for drbd_cleanup */
2892 idr_init(&drbd_devices);
2894 mutex_init(&resources_mutex);
2895 INIT_LIST_HEAD(&drbd_resources);
2897 err = drbd_genl_register();
2899 pr_err("unable to register generic netlink family\n");
2903 err = drbd_create_mempools();
2908 drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show);
2910 pr_err("unable to register proc file\n");
2914 retry.wq = create_singlethread_workqueue("drbd-reissue");
2916 pr_err("unable to create retry workqueue\n");
2919 INIT_WORK(&retry.worker, do_retry);
2920 spin_lock_init(&retry.lock);
2921 INIT_LIST_HEAD(&retry.writes);
2923 drbd_debugfs_init();
2925 pr_info("initialized. "
2926 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2927 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2928 pr_info("%s\n", drbd_buildtag());
2929 pr_info("registered as block device major %d\n", DRBD_MAJOR);
2930 return 0; /* Success! */
2935 pr_err("ran out of memory\n");
2937 pr_err("initialization failure\n");
2941 static void drbd_free_one_sock(struct drbd_socket *ds)
2944 mutex_lock(&ds->mutex);
2947 mutex_unlock(&ds->mutex);
2949 /* so debugfs does not need to mutex_lock() */
2951 kernel_sock_shutdown(s, SHUT_RDWR);
2956 void drbd_free_sock(struct drbd_connection *connection)
2958 if (connection->data.socket)
2959 drbd_free_one_sock(&connection->data);
2960 if (connection->meta.socket)
2961 drbd_free_one_sock(&connection->meta);
2964 /* meta data management */
2966 void conn_md_sync(struct drbd_connection *connection)
2968 struct drbd_peer_device *peer_device;
2972 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2973 struct drbd_device *device = peer_device->device;
2975 kref_get(&device->kref);
2977 drbd_md_sync(device);
2978 kref_put(&device->kref, drbd_destroy_device);
2984 /* aligned 4kByte */
2985 struct meta_data_on_disk {
2986 u64 la_size_sect; /* last agreed size. */
2987 u64 uuid[UI_SIZE]; /* UUIDs. */
2990 u32 flags; /* MDF */
2993 u32 al_offset; /* offset to this block */
2994 u32 al_nr_extents; /* important for restoring the AL (userspace) */
2995 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2996 u32 bm_offset; /* offset to the bitmap, from here */
2997 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
2998 u32 la_peer_max_bio_size; /* last peer max_bio_size */
3000 /* see al_tr_number_to_on_disk_sector() */
3002 u32 al_stripe_size_4k;
3004 u8 reserved_u8[4096 - (7*8 + 10*4)];
3009 void drbd_md_write(struct drbd_device *device, void *b)
3011 struct meta_data_on_disk *buffer = b;
3015 memset(buffer, 0, sizeof(*buffer));
3017 buffer->la_size_sect = cpu_to_be64(get_capacity(device->vdisk));
3018 for (i = UI_CURRENT; i < UI_SIZE; i++)
3019 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3020 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3021 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3023 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3024 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3025 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3026 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3027 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3029 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3030 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3032 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3033 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3035 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3036 sector = device->ldev->md.md_offset;
3038 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3039 /* this was a try anyways ... */
3040 drbd_err(device, "meta data update failed!\n");
3041 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3046 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3047 * @device: DRBD device.
3049 void drbd_md_sync(struct drbd_device *device)
3051 struct meta_data_on_disk *buffer;
3053 /* Don't accidentally change the DRBD meta data layout. */
3054 BUILD_BUG_ON(UI_SIZE != 4);
3055 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3057 del_timer(&device->md_sync_timer);
3058 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3059 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3062 /* We use here D_FAILED and not D_ATTACHING because we try to write
3063 * metadata even if we detach due to a disk failure! */
3064 if (!get_ldev_if_state(device, D_FAILED))
3067 buffer = drbd_md_get_buffer(device, __func__);
3071 drbd_md_write(device, buffer);
3073 /* Update device->ldev->md.la_size_sect,
3074 * since we updated it on metadata. */
3075 device->ldev->md.la_size_sect = get_capacity(device->vdisk);
3077 drbd_md_put_buffer(device);
3082 static int check_activity_log_stripe_size(struct drbd_device *device,
3083 struct meta_data_on_disk *on_disk,
3084 struct drbd_md *in_core)
3086 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3087 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3090 /* both not set: default to old fixed size activity log */
3091 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3093 al_stripe_size_4k = MD_32kB_SECT/8;
3096 /* some paranoia plausibility checks */
3098 /* we need both values to be set */
3099 if (al_stripes == 0 || al_stripe_size_4k == 0)
3102 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3104 /* Upper limit of activity log area, to avoid potential overflow
3105 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3106 * than 72 * 4k blocks total only increases the amount of history,
3107 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3108 if (al_size_4k > (16 * 1024 * 1024/4))
3111 /* Lower limit: we need at least 8 transaction slots (32kB)
3112 * to not break existing setups */
3113 if (al_size_4k < MD_32kB_SECT/8)
3116 in_core->al_stripe_size_4k = al_stripe_size_4k;
3117 in_core->al_stripes = al_stripes;
3118 in_core->al_size_4k = al_size_4k;
3122 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3123 al_stripes, al_stripe_size_4k);
3127 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3129 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3130 struct drbd_md *in_core = &bdev->md;
3131 s32 on_disk_al_sect;
3132 s32 on_disk_bm_sect;
3134 /* The on-disk size of the activity log, calculated from offsets, and
3135 * the size of the activity log calculated from the stripe settings,
3137 * Though we could relax this a bit: it is ok, if the striped activity log
3138 * fits in the available on-disk activity log size.
3139 * Right now, that would break how resize is implemented.
3140 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3141 * of possible unused padding space in the on disk layout. */
3142 if (in_core->al_offset < 0) {
3143 if (in_core->bm_offset > in_core->al_offset)
3145 on_disk_al_sect = -in_core->al_offset;
3146 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3148 if (in_core->al_offset != MD_4kB_SECT)
3150 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3153 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3154 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3157 /* old fixed size meta data is exactly that: fixed. */
3158 if (in_core->meta_dev_idx >= 0) {
3159 if (in_core->md_size_sect != MD_128MB_SECT
3160 || in_core->al_offset != MD_4kB_SECT
3161 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3162 || in_core->al_stripes != 1
3163 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3167 if (capacity < in_core->md_size_sect)
3169 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3172 /* should be aligned, and at least 32k */
3173 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3176 /* should fit (for now: exactly) into the available on-disk space;
3177 * overflow prevention is in check_activity_log_stripe_size() above. */
3178 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3181 /* again, should be aligned */
3182 if (in_core->bm_offset & 7)
3185 /* FIXME check for device grow with flex external meta data? */
3187 /* can the available bitmap space cover the last agreed device size? */
3188 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3194 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3195 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3196 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3197 in_core->meta_dev_idx,
3198 in_core->al_stripes, in_core->al_stripe_size_4k,
3199 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3200 (unsigned long long)in_core->la_size_sect,
3201 (unsigned long long)capacity);
3208 * drbd_md_read() - Reads in the meta data super block
3209 * @device: DRBD device.
3210 * @bdev: Device from which the meta data should be read in.
3212 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3213 * something goes wrong.
3215 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3216 * even before @bdev is assigned to @device->ldev.
3218 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3220 struct meta_data_on_disk *buffer;
3222 int i, rv = NO_ERROR;
3224 if (device->state.disk != D_DISKLESS)
3225 return ERR_DISK_CONFIGURED;
3227 buffer = drbd_md_get_buffer(device, __func__);
3231 /* First, figure out where our meta data superblock is located,
3233 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3234 bdev->md.md_offset = drbd_md_ss(bdev);
3235 /* Even for (flexible or indexed) external meta data,
3236 * initially restrict us to the 4k superblock for now.
3237 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3238 bdev->md.md_size_sect = 8;
3240 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3242 /* NOTE: can't do normal error processing here as this is
3243 called BEFORE disk is attached */
3244 drbd_err(device, "Error while reading metadata.\n");
3245 rv = ERR_IO_MD_DISK;
3249 magic = be32_to_cpu(buffer->magic);
3250 flags = be32_to_cpu(buffer->flags);
3251 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3252 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3253 /* btw: that's Activity Log clean, not "all" clean. */
3254 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3255 rv = ERR_MD_UNCLEAN;
3259 rv = ERR_MD_INVALID;
3260 if (magic != DRBD_MD_MAGIC_08) {
3261 if (magic == DRBD_MD_MAGIC_07)
3262 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3264 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3268 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3269 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3270 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3275 /* convert to in_core endian */
3276 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3277 for (i = UI_CURRENT; i < UI_SIZE; i++)
3278 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3279 bdev->md.flags = be32_to_cpu(buffer->flags);
3280 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3282 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3283 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3284 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3286 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3288 if (check_offsets_and_sizes(device, bdev))
3291 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3292 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3293 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3296 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3297 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3298 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3304 spin_lock_irq(&device->resource->req_lock);
3305 if (device->state.conn < C_CONNECTED) {
3307 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3308 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3309 device->peer_max_bio_size = peer;
3311 spin_unlock_irq(&device->resource->req_lock);
3314 drbd_md_put_buffer(device);
3320 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3321 * @device: DRBD device.
3323 * Call this function if you change anything that should be written to
3324 * the meta-data super block. This function sets MD_DIRTY, and starts a
3325 * timer that ensures that within five seconds you have to call drbd_md_sync().
3327 void drbd_md_mark_dirty(struct drbd_device *device)
3329 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3330 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3333 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3337 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3338 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3341 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3343 if (idx == UI_CURRENT) {
3344 if (device->state.role == R_PRIMARY)
3349 drbd_set_ed_uuid(device, val);
3352 device->ldev->md.uuid[idx] = val;
3353 drbd_md_mark_dirty(device);
3356 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3358 unsigned long flags;
3359 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3360 __drbd_uuid_set(device, idx, val);
3361 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3364 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3366 unsigned long flags;
3367 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3368 if (device->ldev->md.uuid[idx]) {
3369 drbd_uuid_move_history(device);
3370 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3372 __drbd_uuid_set(device, idx, val);
3373 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3377 * drbd_uuid_new_current() - Creates a new current UUID
3378 * @device: DRBD device.
3380 * Creates a new current UUID, and rotates the old current UUID into
3381 * the bitmap slot. Causes an incremental resync upon next connect.
3383 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3386 unsigned long long bm_uuid;
3388 get_random_bytes(&val, sizeof(u64));
3390 spin_lock_irq(&device->ldev->md.uuid_lock);
3391 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3394 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3396 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3397 __drbd_uuid_set(device, UI_CURRENT, val);
3398 spin_unlock_irq(&device->ldev->md.uuid_lock);
3400 drbd_print_uuids(device, "new current UUID");
3401 /* get it to stable storage _now_ */
3402 drbd_md_sync(device);
3405 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3407 unsigned long flags;
3408 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3411 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3413 drbd_uuid_move_history(device);
3414 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3415 device->ldev->md.uuid[UI_BITMAP] = 0;
3417 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3419 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3421 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3423 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3425 drbd_md_mark_dirty(device);
3429 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3430 * @device: DRBD device.
3432 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3434 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3438 drbd_md_set_flag(device, MDF_FULL_SYNC);
3439 drbd_md_sync(device);
3440 drbd_bm_set_all(device);
3442 rv = drbd_bm_write(device);
3445 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3446 drbd_md_sync(device);
3453 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3454 * @device: DRBD device.
3456 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3458 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3460 drbd_resume_al(device);
3461 drbd_bm_clear_all(device);
3462 return drbd_bm_write(device);
3465 static int w_bitmap_io(struct drbd_work *w, int unused)
3467 struct drbd_device *device =
3468 container_of(w, struct drbd_device, bm_io_work.w);
3469 struct bm_io_work *work = &device->bm_io_work;
3472 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3473 int cnt = atomic_read(&device->ap_bio_cnt);
3475 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3479 if (get_ldev(device)) {
3480 drbd_bm_lock(device, work->why, work->flags);
3481 rv = work->io_fn(device);
3482 drbd_bm_unlock(device);
3486 clear_bit_unlock(BITMAP_IO, &device->flags);
3487 wake_up(&device->misc_wait);
3490 work->done(device, rv);
3492 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3500 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3501 * @device: DRBD device.
3502 * @io_fn: IO callback to be called when bitmap IO is possible
3503 * @done: callback to be called after the bitmap IO was performed
3504 * @why: Descriptive text of the reason for doing the IO
3505 * @flags: Bitmap flags
3507 * While IO on the bitmap happens we freeze application IO thus we ensure
3508 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3509 * called from worker context. It MUST NOT be used while a previous such
3510 * work is still pending!
3512 * Its worker function encloses the call of io_fn() by get_ldev() and
3515 void drbd_queue_bitmap_io(struct drbd_device *device,
3516 int (*io_fn)(struct drbd_device *),
3517 void (*done)(struct drbd_device *, int),
3518 char *why, enum bm_flag flags)
3520 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3522 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3523 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3524 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3525 if (device->bm_io_work.why)
3526 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3527 why, device->bm_io_work.why);
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(&first_peer_device(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, int (*io_fn)(struct drbd_device *),
3557 char *why, enum bm_flag flags)
3559 /* Only suspend io, if some operation is supposed to be locked out */
3560 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3563 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3566 drbd_suspend_io(device);
3568 drbd_bm_lock(device, why, flags);
3570 drbd_bm_unlock(device);
3573 drbd_resume_io(device);
3578 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3580 if ((device->ldev->md.flags & flag) != flag) {
3581 drbd_md_mark_dirty(device);
3582 device->ldev->md.flags |= flag;
3586 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3588 if ((device->ldev->md.flags & flag) != 0) {
3589 drbd_md_mark_dirty(device);
3590 device->ldev->md.flags &= ~flag;
3593 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3595 return (bdev->md.flags & flag) != 0;
3598 static void md_sync_timer_fn(struct timer_list *t)
3600 struct drbd_device *device = from_timer(device, t, md_sync_timer);
3601 drbd_device_post_work(device, MD_SYNC);
3604 const char *cmdname(enum drbd_packet cmd)
3606 /* THINK may need to become several global tables
3607 * when we want to support more than
3608 * one PRO_VERSION */
3609 static const char *cmdnames[] = {
3611 [P_WSAME] = "WriteSame",
3613 [P_DATA_REPLY] = "DataReply",
3614 [P_RS_DATA_REPLY] = "RSDataReply",
3615 [P_BARRIER] = "Barrier",
3616 [P_BITMAP] = "ReportBitMap",
3617 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3618 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3619 [P_UNPLUG_REMOTE] = "UnplugRemote",
3620 [P_DATA_REQUEST] = "DataRequest",
3621 [P_RS_DATA_REQUEST] = "RSDataRequest",
3622 [P_SYNC_PARAM] = "SyncParam",
3623 [P_SYNC_PARAM89] = "SyncParam89",
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",
3632 [P_PING_ACK] = "PingAck",
3633 [P_RECV_ACK] = "RecvAck",
3634 [P_WRITE_ACK] = "WriteAck",
3635 [P_RS_WRITE_ACK] = "RSWriteAck",
3636 [P_SUPERSEDED] = "Superseded",
3637 [P_NEG_ACK] = "NegAck",
3638 [P_NEG_DREPLY] = "NegDReply",
3639 [P_NEG_RS_DREPLY] = "NegRSDReply",
3640 [P_BARRIER_ACK] = "BarrierAck",
3641 [P_STATE_CHG_REQ] = "StateChgRequest",
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_COMPRESSED_BITMAP] = "CBitmap",
3649 [P_DELAY_PROBE] = "DelayProbe",
3650 [P_OUT_OF_SYNC] = "OutOfSync",
3651 [P_RETRY_WRITE] = "RetryWrite",
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",
3656 [P_RS_THIN_REQ] = "rs_thin_req",
3657 [P_RS_DEALLOCATED] = "rs_deallocated",
3659 /* enum drbd_packet, but not commands - obsoleted flags:
3665 /* too big for the array: 0xfffX */
3666 if (cmd == P_INITIAL_META)
3667 return "InitialMeta";
3668 if (cmd == P_INITIAL_DATA)
3669 return "InitialData";
3670 if (cmd == P_CONNECTION_FEATURES)
3671 return "ConnectionFeatures";
3672 if (cmd >= ARRAY_SIZE(cmdnames))
3674 return cmdnames[cmd];
3678 * drbd_wait_misc - wait for a request to make progress
3679 * @device: device associated with the request
3680 * @i: the struct drbd_interval embedded in struct drbd_request or
3681 * struct drbd_peer_request
3683 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3685 struct net_conf *nc;
3690 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3695 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3698 /* Indicate to wake up device->misc_wait on progress. */
3700 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3701 spin_unlock_irq(&device->resource->req_lock);
3702 timeout = schedule_timeout(timeout);
3703 finish_wait(&device->misc_wait, &wait);
3704 spin_lock_irq(&device->resource->req_lock);
3705 if (!timeout || device->state.conn < C_CONNECTED)
3707 if (signal_pending(current))
3708 return -ERESTARTSYS;
3712 void lock_all_resources(void)
3714 struct drbd_resource *resource;
3715 int __maybe_unused i = 0;
3717 mutex_lock(&resources_mutex);
3718 local_irq_disable();
3719 for_each_resource(resource, &drbd_resources)
3720 spin_lock_nested(&resource->req_lock, i++);
3723 void unlock_all_resources(void)
3725 struct drbd_resource *resource;
3727 for_each_resource(resource, &drbd_resources)
3728 spin_unlock(&resource->req_lock);
3730 mutex_unlock(&resources_mutex);
3733 #ifdef CONFIG_DRBD_FAULT_INJECTION
3734 /* Fault insertion support including random number generator shamelessly
3735 * stolen from kernel/rcutorture.c */
3736 struct fault_random_state {
3737 unsigned long state;
3738 unsigned long count;
3741 #define FAULT_RANDOM_MULT 39916801 /* prime */
3742 #define FAULT_RANDOM_ADD 479001701 /* prime */
3743 #define FAULT_RANDOM_REFRESH 10000
3746 * Crude but fast random-number generator. Uses a linear congruential
3747 * generator, with occasional help from get_random_bytes().
3749 static unsigned long
3750 _drbd_fault_random(struct fault_random_state *rsp)
3754 if (!rsp->count--) {
3755 get_random_bytes(&refresh, sizeof(refresh));
3756 rsp->state += refresh;
3757 rsp->count = FAULT_RANDOM_REFRESH;
3759 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3760 return swahw32(rsp->state);
3764 _drbd_fault_str(unsigned int type) {
3765 static char *_faults[] = {
3766 [DRBD_FAULT_MD_WR] = "Meta-data write",
3767 [DRBD_FAULT_MD_RD] = "Meta-data read",
3768 [DRBD_FAULT_RS_WR] = "Resync write",
3769 [DRBD_FAULT_RS_RD] = "Resync read",
3770 [DRBD_FAULT_DT_WR] = "Data write",
3771 [DRBD_FAULT_DT_RD] = "Data read",
3772 [DRBD_FAULT_DT_RA] = "Data read ahead",
3773 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3774 [DRBD_FAULT_AL_EE] = "EE allocation",
3775 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3778 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3782 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3784 static struct fault_random_state rrs = {0, 0};
3786 unsigned int ret = (
3787 (drbd_fault_devs == 0 ||
3788 ((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3789 (((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3794 if (__ratelimit(&drbd_ratelimit_state))
3795 drbd_warn(device, "***Simulating %s failure\n",
3796 _drbd_fault_str(type));
3803 const char *drbd_buildtag(void)
3805 /* DRBD built from external sources has here a reference to the
3806 git hash of the source code. */
3808 static char buildtag[38] = "\0uilt-in";
3810 if (buildtag[0] == 0) {
3812 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3821 module_init(drbd_init)
3822 module_exit(drbd_cleanup)
3824 EXPORT_SYMBOL(drbd_conn_str);
3825 EXPORT_SYMBOL(drbd_role_str);
3826 EXPORT_SYMBOL(drbd_disk_str);
3827 EXPORT_SYMBOL(drbd_set_st_err_str);