1 // SPDX-License-Identifier: GPL-2.0-only
5 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
7 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
8 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
9 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
11 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
12 from Logicworks, Inc. for making SDP replication support possible.
17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 #include <linux/module.h>
20 #include <linux/jiffies.h>
21 #include <linux/drbd.h>
22 #include <linux/uaccess.h>
23 #include <asm/types.h>
25 #include <linux/ctype.h>
26 #include <linux/mutex.h>
28 #include <linux/file.h>
29 #include <linux/proc_fs.h>
30 #include <linux/init.h>
32 #include <linux/memcontrol.h>
33 #include <linux/mm_inline.h>
34 #include <linux/slab.h>
35 #include <linux/random.h>
36 #include <linux/reboot.h>
37 #include <linux/notifier.h>
38 #include <linux/kthread.h>
39 #include <linux/workqueue.h>
40 #include <linux/unistd.h>
41 #include <linux/vmalloc.h>
42 #include <linux/sched/signal.h>
44 #include <linux/drbd_limits.h>
46 #include "drbd_protocol.h"
47 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
49 #include "drbd_debugfs.h"
51 static DEFINE_MUTEX(drbd_main_mutex);
52 static int drbd_open(struct block_device *bdev, fmode_t mode);
53 static void drbd_release(struct gendisk *gd, fmode_t mode);
54 static void md_sync_timer_fn(struct timer_list *t);
55 static int w_bitmap_io(struct drbd_work *w, int unused);
57 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
58 "Lars Ellenberg <lars@linbit.com>");
59 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
60 MODULE_VERSION(REL_VERSION);
61 MODULE_LICENSE("GPL");
62 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
63 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
64 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
66 #include <linux/moduleparam.h>
67 /* thanks to these macros, if compiled into the kernel (not-module),
68 * these become boot parameters (e.g., drbd.minor_count) */
70 #ifdef CONFIG_DRBD_FAULT_INJECTION
71 int drbd_enable_faults;
73 static int drbd_fault_count;
74 static int drbd_fault_devs;
75 /* bitmap of enabled faults */
76 module_param_named(enable_faults, drbd_enable_faults, int, 0664);
77 /* fault rate % value - applies to all enabled faults */
78 module_param_named(fault_rate, drbd_fault_rate, int, 0664);
79 /* count of faults inserted */
80 module_param_named(fault_count, drbd_fault_count, int, 0664);
81 /* bitmap of devices to insert faults on */
82 module_param_named(fault_devs, drbd_fault_devs, int, 0644);
85 /* module parameters we can keep static */
86 static bool drbd_allow_oos; /* allow_open_on_secondary */
87 static bool drbd_disable_sendpage;
88 MODULE_PARM_DESC(allow_oos, "DONT USE!");
89 module_param_named(allow_oos, drbd_allow_oos, bool, 0);
90 module_param_named(disable_sendpage, drbd_disable_sendpage, bool, 0644);
92 /* module parameters we share */
93 int drbd_proc_details; /* Detail level in proc drbd*/
94 module_param_named(proc_details, drbd_proc_details, int, 0644);
95 /* module parameters shared with defaults */
96 unsigned int drbd_minor_count = DRBD_MINOR_COUNT_DEF;
97 /* Module parameter for setting the user mode helper program
98 * to run. Default is /sbin/drbdadm */
99 char drbd_usermode_helper[80] = "/sbin/drbdadm";
100 module_param_named(minor_count, drbd_minor_count, uint, 0444);
101 module_param_string(usermode_helper, drbd_usermode_helper, sizeof(drbd_usermode_helper), 0644);
103 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
104 * as member "struct gendisk *vdisk;"
106 struct idr drbd_devices;
107 struct list_head drbd_resources;
108 struct mutex resources_mutex;
110 struct kmem_cache *drbd_request_cache;
111 struct kmem_cache *drbd_ee_cache; /* peer requests */
112 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
113 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
114 mempool_t drbd_request_mempool;
115 mempool_t drbd_ee_mempool;
116 mempool_t drbd_md_io_page_pool;
117 struct bio_set drbd_md_io_bio_set;
118 struct bio_set drbd_io_bio_set;
120 /* I do not use a standard mempool, because:
121 1) I want to hand out the pre-allocated objects first.
122 2) I want to be able to interrupt sleeping allocation with a signal.
123 Note: This is a single linked list, the next pointer is the private
124 member of struct page.
126 struct page *drbd_pp_pool;
127 DEFINE_SPINLOCK(drbd_pp_lock);
129 wait_queue_head_t drbd_pp_wait;
131 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
133 static const struct block_device_operations drbd_ops = {
134 .owner = THIS_MODULE,
135 .submit_bio = drbd_submit_bio,
137 .release = drbd_release,
141 /* When checking with sparse, and this is an inline function, sparse will
142 give tons of false positives. When this is a real functions sparse works.
144 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
148 atomic_inc(&device->local_cnt);
149 io_allowed = (device->state.disk >= mins);
151 if (atomic_dec_and_test(&device->local_cnt))
152 wake_up(&device->misc_wait);
160 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
161 * @connection: DRBD connection.
162 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
163 * @set_size: Expected number of requests before that barrier.
165 * In case the passed barrier_nr or set_size does not match the oldest
166 * epoch of not yet barrier-acked requests, this function will cause a
167 * termination of the connection.
169 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
170 unsigned int set_size)
172 struct drbd_request *r;
173 struct drbd_request *req = NULL, *tmp = NULL;
174 int expect_epoch = 0;
177 spin_lock_irq(&connection->resource->req_lock);
179 /* find oldest not yet barrier-acked write request,
180 * count writes in its epoch. */
181 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
182 const unsigned s = r->rq_state;
186 if (!(s & RQ_NET_MASK))
191 expect_epoch = req->epoch;
194 if (r->epoch != expect_epoch)
198 /* if (s & RQ_DONE): not expected */
199 /* if (!(s & RQ_NET_MASK)): not expected */
204 /* first some paranoia code */
206 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
210 if (expect_epoch != barrier_nr) {
211 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
212 barrier_nr, expect_epoch);
216 if (expect_size != set_size) {
217 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
218 barrier_nr, set_size, expect_size);
222 /* Clean up list of requests processed during current epoch. */
223 /* this extra list walk restart is paranoia,
224 * to catch requests being barrier-acked "unexpectedly".
225 * It usually should find the same req again, or some READ preceding it. */
226 list_for_each_entry(req, &connection->transfer_log, tl_requests)
227 if (req->epoch == expect_epoch) {
231 req = list_prepare_entry(tmp, &connection->transfer_log, tl_requests);
232 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
233 struct drbd_peer_device *peer_device;
234 if (req->epoch != expect_epoch)
236 peer_device = conn_peer_device(connection, req->device->vnr);
237 _req_mod(req, BARRIER_ACKED, peer_device);
239 spin_unlock_irq(&connection->resource->req_lock);
244 spin_unlock_irq(&connection->resource->req_lock);
245 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
250 * _tl_restart() - Walks the transfer log, and applies an action to all requests
251 * @connection: DRBD connection to operate on.
252 * @what: The action/event to perform with all request objects
254 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
255 * RESTART_FROZEN_DISK_IO.
257 /* must hold resource->req_lock */
258 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
260 struct drbd_peer_device *peer_device;
261 struct drbd_request *req, *r;
263 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
264 peer_device = conn_peer_device(connection, req->device->vnr);
265 _req_mod(req, what, peer_device);
269 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
271 spin_lock_irq(&connection->resource->req_lock);
272 _tl_restart(connection, what);
273 spin_unlock_irq(&connection->resource->req_lock);
277 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
278 * @connection: DRBD connection.
280 * This is called after the connection to the peer was lost. The storage covered
281 * by the requests on the transfer gets marked as our of sync. Called from the
282 * receiver thread and the worker thread.
284 void tl_clear(struct drbd_connection *connection)
286 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
290 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
291 * @device: DRBD device.
293 void tl_abort_disk_io(struct drbd_device *device)
295 struct drbd_connection *connection = first_peer_device(device)->connection;
296 struct drbd_request *req, *r;
298 spin_lock_irq(&connection->resource->req_lock);
299 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
300 if (!(req->rq_state & RQ_LOCAL_PENDING))
302 if (req->device != device)
304 _req_mod(req, ABORT_DISK_IO, NULL);
306 spin_unlock_irq(&connection->resource->req_lock);
309 static int drbd_thread_setup(void *arg)
311 struct drbd_thread *thi = (struct drbd_thread *) arg;
312 struct drbd_resource *resource = thi->resource;
316 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
320 allow_kernel_signal(DRBD_SIGKILL);
321 allow_kernel_signal(SIGXCPU);
323 retval = thi->function(thi);
325 spin_lock_irqsave(&thi->t_lock, flags);
327 /* if the receiver has been "EXITING", the last thing it did
328 * was set the conn state to "StandAlone",
329 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
330 * and receiver thread will be "started".
331 * drbd_thread_start needs to set "RESTARTING" in that case.
332 * t_state check and assignment needs to be within the same spinlock,
333 * so either thread_start sees EXITING, and can remap to RESTARTING,
334 * or thread_start see NONE, and can proceed as normal.
337 if (thi->t_state == RESTARTING) {
338 drbd_info(resource, "Restarting %s thread\n", thi->name);
339 thi->t_state = RUNNING;
340 spin_unlock_irqrestore(&thi->t_lock, flags);
347 complete_all(&thi->stop);
348 spin_unlock_irqrestore(&thi->t_lock, flags);
350 drbd_info(resource, "Terminating %s\n", current->comm);
352 /* Release mod reference taken when thread was started */
355 kref_put(&thi->connection->kref, drbd_destroy_connection);
356 kref_put(&resource->kref, drbd_destroy_resource);
357 module_put(THIS_MODULE);
361 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
362 int (*func) (struct drbd_thread *), const char *name)
364 spin_lock_init(&thi->t_lock);
367 thi->function = func;
368 thi->resource = resource;
369 thi->connection = NULL;
373 int drbd_thread_start(struct drbd_thread *thi)
375 struct drbd_resource *resource = thi->resource;
376 struct task_struct *nt;
379 /* is used from state engine doing drbd_thread_stop_nowait,
380 * while holding the req lock irqsave */
381 spin_lock_irqsave(&thi->t_lock, flags);
383 switch (thi->t_state) {
385 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
386 thi->name, current->comm, current->pid);
388 /* Get ref on module for thread - this is released when thread exits */
389 if (!try_module_get(THIS_MODULE)) {
390 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
391 spin_unlock_irqrestore(&thi->t_lock, flags);
395 kref_get(&resource->kref);
397 kref_get(&thi->connection->kref);
399 init_completion(&thi->stop);
400 thi->reset_cpu_mask = 1;
401 thi->t_state = RUNNING;
402 spin_unlock_irqrestore(&thi->t_lock, flags);
403 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
405 nt = kthread_create(drbd_thread_setup, (void *) thi,
406 "drbd_%c_%s", thi->name[0], thi->resource->name);
409 drbd_err(resource, "Couldn't start thread\n");
412 kref_put(&thi->connection->kref, drbd_destroy_connection);
413 kref_put(&resource->kref, drbd_destroy_resource);
414 module_put(THIS_MODULE);
417 spin_lock_irqsave(&thi->t_lock, flags);
419 thi->t_state = RUNNING;
420 spin_unlock_irqrestore(&thi->t_lock, flags);
424 thi->t_state = RESTARTING;
425 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
426 thi->name, current->comm, current->pid);
431 spin_unlock_irqrestore(&thi->t_lock, flags);
439 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
443 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
445 /* may be called from state engine, holding the req lock irqsave */
446 spin_lock_irqsave(&thi->t_lock, flags);
448 if (thi->t_state == NONE) {
449 spin_unlock_irqrestore(&thi->t_lock, flags);
451 drbd_thread_start(thi);
455 if (thi->t_state != ns) {
456 if (thi->task == NULL) {
457 spin_unlock_irqrestore(&thi->t_lock, flags);
463 init_completion(&thi->stop);
464 if (thi->task != current)
465 send_sig(DRBD_SIGKILL, thi->task, 1);
468 spin_unlock_irqrestore(&thi->t_lock, flags);
471 wait_for_completion(&thi->stop);
474 int conn_lowest_minor(struct drbd_connection *connection)
476 struct drbd_peer_device *peer_device;
477 int vnr = 0, minor = -1;
480 peer_device = idr_get_next(&connection->peer_devices, &vnr);
482 minor = device_to_minor(peer_device->device);
490 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
492 * Forces all threads of a resource onto the same CPU. This is beneficial for
493 * DRBD's performance. May be overwritten by user's configuration.
495 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
497 unsigned int *resources_per_cpu, min_index = ~0;
499 resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu),
501 if (resources_per_cpu) {
502 struct drbd_resource *resource;
503 unsigned int cpu, min = ~0;
506 for_each_resource_rcu(resource, &drbd_resources) {
507 for_each_cpu(cpu, resource->cpu_mask)
508 resources_per_cpu[cpu]++;
511 for_each_online_cpu(cpu) {
512 if (resources_per_cpu[cpu] < min) {
513 min = resources_per_cpu[cpu];
517 kfree(resources_per_cpu);
519 if (min_index == ~0) {
520 cpumask_setall(*cpu_mask);
523 cpumask_set_cpu(min_index, *cpu_mask);
527 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
528 * @thi: drbd_thread object
530 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
533 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
535 struct drbd_resource *resource = thi->resource;
536 struct task_struct *p = current;
538 if (!thi->reset_cpu_mask)
540 thi->reset_cpu_mask = 0;
541 set_cpus_allowed_ptr(p, resource->cpu_mask);
544 #define drbd_calc_cpu_mask(A) ({})
548 * drbd_header_size - size of a packet header
550 * The header size is a multiple of 8, so any payload following the header is
551 * word aligned on 64-bit architectures. (The bitmap send and receive code
554 unsigned int drbd_header_size(struct drbd_connection *connection)
556 if (connection->agreed_pro_version >= 100) {
557 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
558 return sizeof(struct p_header100);
560 BUILD_BUG_ON(sizeof(struct p_header80) !=
561 sizeof(struct p_header95));
562 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
563 return sizeof(struct p_header80);
567 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
569 h->magic = cpu_to_be32(DRBD_MAGIC);
570 h->command = cpu_to_be16(cmd);
571 h->length = cpu_to_be16(size);
572 return sizeof(struct p_header80);
575 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
577 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
578 h->command = cpu_to_be16(cmd);
579 h->length = cpu_to_be32(size);
580 return sizeof(struct p_header95);
583 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
586 h->magic = cpu_to_be32(DRBD_MAGIC_100);
587 h->volume = cpu_to_be16(vnr);
588 h->command = cpu_to_be16(cmd);
589 h->length = cpu_to_be32(size);
591 return sizeof(struct p_header100);
594 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
595 void *buffer, enum drbd_packet cmd, int size)
597 if (connection->agreed_pro_version >= 100)
598 return prepare_header100(buffer, cmd, size, vnr);
599 else if (connection->agreed_pro_version >= 95 &&
600 size > DRBD_MAX_SIZE_H80_PACKET)
601 return prepare_header95(buffer, cmd, size);
603 return prepare_header80(buffer, cmd, size);
606 static void *__conn_prepare_command(struct drbd_connection *connection,
607 struct drbd_socket *sock)
611 return sock->sbuf + drbd_header_size(connection);
614 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
618 mutex_lock(&sock->mutex);
619 p = __conn_prepare_command(connection, sock);
621 mutex_unlock(&sock->mutex);
626 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
628 return conn_prepare_command(peer_device->connection, sock);
631 static int __send_command(struct drbd_connection *connection, int vnr,
632 struct drbd_socket *sock, enum drbd_packet cmd,
633 unsigned int header_size, void *data,
640 * Called with @data == NULL and the size of the data blocks in @size
641 * for commands that send data blocks. For those commands, omit the
642 * MSG_MORE flag: this will increase the likelihood that data blocks
643 * which are page aligned on the sender will end up page aligned on the
646 msg_flags = data ? MSG_MORE : 0;
648 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
650 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
653 err = drbd_send_all(connection, sock->socket, data, size, 0);
654 /* DRBD protocol "pings" are latency critical.
655 * This is supposed to trigger tcp_push_pending_frames() */
656 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
657 tcp_sock_set_nodelay(sock->socket->sk);
662 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
663 enum drbd_packet cmd, unsigned int header_size,
664 void *data, unsigned int size)
666 return __send_command(connection, 0, sock, cmd, header_size, data, size);
669 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
670 enum drbd_packet cmd, unsigned int header_size,
671 void *data, unsigned int size)
675 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
676 mutex_unlock(&sock->mutex);
680 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
681 enum drbd_packet cmd, unsigned int header_size,
682 void *data, unsigned int size)
686 err = __send_command(peer_device->connection, peer_device->device->vnr,
687 sock, cmd, header_size, data, size);
688 mutex_unlock(&sock->mutex);
692 int drbd_send_ping(struct drbd_connection *connection)
694 struct drbd_socket *sock;
696 sock = &connection->meta;
697 if (!conn_prepare_command(connection, sock))
699 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
702 int drbd_send_ping_ack(struct drbd_connection *connection)
704 struct drbd_socket *sock;
706 sock = &connection->meta;
707 if (!conn_prepare_command(connection, sock))
709 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
712 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
714 struct drbd_socket *sock;
715 struct p_rs_param_95 *p;
717 const int apv = peer_device->connection->agreed_pro_version;
718 enum drbd_packet cmd;
720 struct disk_conf *dc;
722 sock = &peer_device->connection->data;
723 p = drbd_prepare_command(peer_device, sock);
728 nc = rcu_dereference(peer_device->connection->net_conf);
730 size = apv <= 87 ? sizeof(struct p_rs_param)
731 : apv == 88 ? sizeof(struct p_rs_param)
732 + strlen(nc->verify_alg) + 1
733 : apv <= 94 ? sizeof(struct p_rs_param_89)
734 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
736 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
738 /* initialize verify_alg and csums_alg */
739 BUILD_BUG_ON(sizeof(p->algs) != 2 * SHARED_SECRET_MAX);
740 memset(&p->algs, 0, sizeof(p->algs));
742 if (get_ldev(peer_device->device)) {
743 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
744 p->resync_rate = cpu_to_be32(dc->resync_rate);
745 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
746 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
747 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
748 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
749 put_ldev(peer_device->device);
751 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
752 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
753 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
754 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
755 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
759 strcpy(p->verify_alg, nc->verify_alg);
761 strcpy(p->csums_alg, nc->csums_alg);
764 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
767 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
769 struct drbd_socket *sock;
770 struct p_protocol *p;
774 sock = &connection->data;
775 p = __conn_prepare_command(connection, sock);
780 nc = rcu_dereference(connection->net_conf);
782 if (nc->tentative && connection->agreed_pro_version < 92) {
784 drbd_err(connection, "--dry-run is not supported by peer");
789 if (connection->agreed_pro_version >= 87)
790 size += strlen(nc->integrity_alg) + 1;
792 p->protocol = cpu_to_be32(nc->wire_protocol);
793 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
794 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
795 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
796 p->two_primaries = cpu_to_be32(nc->two_primaries);
798 if (nc->discard_my_data)
799 cf |= CF_DISCARD_MY_DATA;
802 p->conn_flags = cpu_to_be32(cf);
804 if (connection->agreed_pro_version >= 87)
805 strcpy(p->integrity_alg, nc->integrity_alg);
808 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
811 int drbd_send_protocol(struct drbd_connection *connection)
815 mutex_lock(&connection->data.mutex);
816 err = __drbd_send_protocol(connection, P_PROTOCOL);
817 mutex_unlock(&connection->data.mutex);
822 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
824 struct drbd_device *device = peer_device->device;
825 struct drbd_socket *sock;
829 if (!get_ldev_if_state(device, D_NEGOTIATING))
832 sock = &peer_device->connection->data;
833 p = drbd_prepare_command(peer_device, sock);
838 spin_lock_irq(&device->ldev->md.uuid_lock);
839 for (i = UI_CURRENT; i < UI_SIZE; i++)
840 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
841 spin_unlock_irq(&device->ldev->md.uuid_lock);
843 device->comm_bm_set = drbd_bm_total_weight(device);
844 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
846 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
848 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
849 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
850 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
853 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
856 int drbd_send_uuids(struct drbd_peer_device *peer_device)
858 return _drbd_send_uuids(peer_device, 0);
861 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
863 return _drbd_send_uuids(peer_device, 8);
866 void drbd_print_uuids(struct drbd_device *device, const char *text)
868 if (get_ldev_if_state(device, D_NEGOTIATING)) {
869 u64 *uuid = device->ldev->md.uuid;
870 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
872 (unsigned long long)uuid[UI_CURRENT],
873 (unsigned long long)uuid[UI_BITMAP],
874 (unsigned long long)uuid[UI_HISTORY_START],
875 (unsigned long long)uuid[UI_HISTORY_END]);
878 drbd_info(device, "%s effective data uuid: %016llX\n",
880 (unsigned long long)device->ed_uuid);
884 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
886 struct drbd_device *device = peer_device->device;
887 struct drbd_socket *sock;
891 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
893 uuid = device->ldev->md.uuid[UI_BITMAP];
894 if (uuid && uuid != UUID_JUST_CREATED)
895 uuid = uuid + UUID_NEW_BM_OFFSET;
897 get_random_bytes(&uuid, sizeof(u64));
898 drbd_uuid_set(device, UI_BITMAP, uuid);
899 drbd_print_uuids(device, "updated sync UUID");
900 drbd_md_sync(device);
902 sock = &peer_device->connection->data;
903 p = drbd_prepare_command(peer_device, sock);
905 p->uuid = cpu_to_be64(uuid);
906 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
910 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
912 struct drbd_device *device = peer_device->device;
913 struct drbd_socket *sock;
915 sector_t d_size, u_size;
917 unsigned int max_bio_size;
918 unsigned int packet_size;
920 sock = &peer_device->connection->data;
921 p = drbd_prepare_command(peer_device, sock);
925 packet_size = sizeof(*p);
926 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
927 packet_size += sizeof(p->qlim[0]);
929 memset(p, 0, packet_size);
930 if (get_ldev_if_state(device, D_NEGOTIATING)) {
931 struct block_device *bdev = device->ldev->backing_bdev;
932 struct request_queue *q = bdev_get_queue(bdev);
934 d_size = drbd_get_max_capacity(device->ldev);
936 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
938 q_order_type = drbd_queue_order_type(device);
939 max_bio_size = queue_max_hw_sectors(q) << 9;
940 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
941 p->qlim->physical_block_size =
942 cpu_to_be32(bdev_physical_block_size(bdev));
943 p->qlim->logical_block_size =
944 cpu_to_be32(bdev_logical_block_size(bdev));
945 p->qlim->alignment_offset =
946 cpu_to_be32(bdev_alignment_offset(bdev));
947 p->qlim->io_min = cpu_to_be32(bdev_io_min(bdev));
948 p->qlim->io_opt = cpu_to_be32(bdev_io_opt(bdev));
949 p->qlim->discard_enabled = !!bdev_max_discard_sectors(bdev);
952 struct request_queue *q = device->rq_queue;
954 p->qlim->physical_block_size =
955 cpu_to_be32(queue_physical_block_size(q));
956 p->qlim->logical_block_size =
957 cpu_to_be32(queue_logical_block_size(q));
958 p->qlim->alignment_offset = 0;
959 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
960 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
961 p->qlim->discard_enabled = 0;
965 q_order_type = QUEUE_ORDERED_NONE;
966 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
969 if (peer_device->connection->agreed_pro_version <= 94)
970 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
971 else if (peer_device->connection->agreed_pro_version < 100)
972 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
974 p->d_size = cpu_to_be64(d_size);
975 p->u_size = cpu_to_be64(u_size);
979 p->c_size = cpu_to_be64(get_capacity(device->vdisk));
980 p->max_bio_size = cpu_to_be32(max_bio_size);
981 p->queue_order_type = cpu_to_be16(q_order_type);
982 p->dds_flags = cpu_to_be16(flags);
984 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
988 * drbd_send_current_state() - Sends the drbd state to the peer
989 * @peer_device: DRBD peer device.
991 int drbd_send_current_state(struct drbd_peer_device *peer_device)
993 struct drbd_socket *sock;
996 sock = &peer_device->connection->data;
997 p = drbd_prepare_command(peer_device, sock);
1000 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1001 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1005 * drbd_send_state() - After a state change, sends the new state to the peer
1006 * @peer_device: DRBD peer device.
1007 * @state: the state to send, not necessarily the current state.
1009 * Each state change queues an "after_state_ch" work, which will eventually
1010 * send the resulting new state to the peer. If more state changes happen
1011 * between queuing and processing of the after_state_ch work, we still
1012 * want to send each intermediary state in the order it occurred.
1014 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1016 struct drbd_socket *sock;
1019 sock = &peer_device->connection->data;
1020 p = drbd_prepare_command(peer_device, sock);
1023 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1024 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1027 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1029 struct drbd_socket *sock;
1030 struct p_req_state *p;
1032 sock = &peer_device->connection->data;
1033 p = drbd_prepare_command(peer_device, sock);
1036 p->mask = cpu_to_be32(mask.i);
1037 p->val = cpu_to_be32(val.i);
1038 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1041 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1043 enum drbd_packet cmd;
1044 struct drbd_socket *sock;
1045 struct p_req_state *p;
1047 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1048 sock = &connection->data;
1049 p = conn_prepare_command(connection, sock);
1052 p->mask = cpu_to_be32(mask.i);
1053 p->val = cpu_to_be32(val.i);
1054 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1057 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1059 struct drbd_socket *sock;
1060 struct p_req_state_reply *p;
1062 sock = &peer_device->connection->meta;
1063 p = drbd_prepare_command(peer_device, sock);
1065 p->retcode = cpu_to_be32(retcode);
1066 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1070 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1072 struct drbd_socket *sock;
1073 struct p_req_state_reply *p;
1074 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1076 sock = &connection->meta;
1077 p = conn_prepare_command(connection, sock);
1079 p->retcode = cpu_to_be32(retcode);
1080 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1084 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1086 BUG_ON(code & ~0xf);
1087 p->encoding = (p->encoding & ~0xf) | code;
1090 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1092 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1095 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1098 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1101 static int fill_bitmap_rle_bits(struct drbd_device *device,
1102 struct p_compressed_bm *p,
1104 struct bm_xfer_ctx *c)
1106 struct bitstream bs;
1107 unsigned long plain_bits;
1114 /* may we use this feature? */
1116 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1118 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1121 if (c->bit_offset >= c->bm_bits)
1122 return 0; /* nothing to do. */
1124 /* use at most thus many bytes */
1125 bitstream_init(&bs, p->code, size, 0);
1126 memset(p->code, 0, size);
1127 /* plain bits covered in this code string */
1130 /* p->encoding & 0x80 stores whether the first run length is set.
1131 * bit offset is implicit.
1132 * start with toggle == 2 to be able to tell the first iteration */
1135 /* see how much plain bits we can stuff into one packet
1136 * using RLE and VLI. */
1138 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1139 : _drbd_bm_find_next(device, c->bit_offset);
1142 rl = tmp - c->bit_offset;
1144 if (toggle == 2) { /* first iteration */
1146 /* the first checked bit was set,
1147 * store start value, */
1148 dcbp_set_start(p, 1);
1149 /* but skip encoding of zero run length */
1153 dcbp_set_start(p, 0);
1156 /* paranoia: catch zero runlength.
1157 * can only happen if bitmap is modified while we scan it. */
1159 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1160 "t:%u bo:%lu\n", toggle, c->bit_offset);
1164 bits = vli_encode_bits(&bs, rl);
1165 if (bits == -ENOBUFS) /* buffer full */
1168 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1174 c->bit_offset = tmp;
1175 } while (c->bit_offset < c->bm_bits);
1177 len = bs.cur.b - p->code + !!bs.cur.bit;
1179 if (plain_bits < (len << 3)) {
1180 /* incompressible with this method.
1181 * we need to rewind both word and bit position. */
1182 c->bit_offset -= plain_bits;
1183 bm_xfer_ctx_bit_to_word_offset(c);
1184 c->bit_offset = c->word_offset * BITS_PER_LONG;
1188 /* RLE + VLI was able to compress it just fine.
1189 * update c->word_offset. */
1190 bm_xfer_ctx_bit_to_word_offset(c);
1192 /* store pad_bits */
1193 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1199 * send_bitmap_rle_or_plain
1201 * Return 0 when done, 1 when another iteration is needed, and a negative error
1202 * code upon failure.
1205 send_bitmap_rle_or_plain(struct drbd_peer_device *peer_device, struct bm_xfer_ctx *c)
1207 struct drbd_device *device = peer_device->device;
1208 struct drbd_socket *sock = &peer_device->connection->data;
1209 unsigned int header_size = drbd_header_size(peer_device->connection);
1210 struct p_compressed_bm *p = sock->sbuf + header_size;
1213 len = fill_bitmap_rle_bits(device, p,
1214 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1219 dcbp_set_code(p, RLE_VLI_Bits);
1220 err = __send_command(peer_device->connection, device->vnr, sock,
1221 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1224 c->bytes[0] += header_size + sizeof(*p) + len;
1226 if (c->bit_offset >= c->bm_bits)
1229 /* was not compressible.
1230 * send a buffer full of plain text bits instead. */
1231 unsigned int data_size;
1232 unsigned long num_words;
1233 unsigned long *p = sock->sbuf + header_size;
1235 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1236 num_words = min_t(size_t, data_size / sizeof(*p),
1237 c->bm_words - c->word_offset);
1238 len = num_words * sizeof(*p);
1240 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1241 err = __send_command(peer_device->connection, device->vnr, sock, P_BITMAP,
1243 c->word_offset += num_words;
1244 c->bit_offset = c->word_offset * BITS_PER_LONG;
1247 c->bytes[1] += header_size + len;
1249 if (c->bit_offset > c->bm_bits)
1250 c->bit_offset = c->bm_bits;
1254 INFO_bm_xfer_stats(peer_device, "send", c);
1262 /* See the comment at receive_bitmap() */
1263 static int _drbd_send_bitmap(struct drbd_device *device,
1264 struct drbd_peer_device *peer_device)
1266 struct bm_xfer_ctx c;
1269 if (!expect(device, device->bitmap))
1272 if (get_ldev(device)) {
1273 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1274 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1275 drbd_bm_set_all(device);
1276 if (drbd_bm_write(device, peer_device)) {
1277 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1278 * but otherwise process as per normal - need to tell other
1279 * side that a full resync is required! */
1280 drbd_err(device, "Failed to write bitmap to disk!\n");
1282 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1283 drbd_md_sync(device);
1289 c = (struct bm_xfer_ctx) {
1290 .bm_bits = drbd_bm_bits(device),
1291 .bm_words = drbd_bm_words(device),
1295 err = send_bitmap_rle_or_plain(peer_device, &c);
1301 int drbd_send_bitmap(struct drbd_device *device, struct drbd_peer_device *peer_device)
1303 struct drbd_socket *sock = &peer_device->connection->data;
1306 mutex_lock(&sock->mutex);
1308 err = !_drbd_send_bitmap(device, peer_device);
1309 mutex_unlock(&sock->mutex);
1313 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1315 struct drbd_socket *sock;
1316 struct p_barrier_ack *p;
1318 if (connection->cstate < C_WF_REPORT_PARAMS)
1321 sock = &connection->meta;
1322 p = conn_prepare_command(connection, sock);
1325 p->barrier = barrier_nr;
1326 p->set_size = cpu_to_be32(set_size);
1327 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1331 * _drbd_send_ack() - Sends an ack packet
1332 * @peer_device: DRBD peer device.
1333 * @cmd: Packet command code.
1334 * @sector: sector, needs to be in big endian byte order
1335 * @blksize: size in byte, needs to be in big endian byte order
1336 * @block_id: Id, big endian byte order
1338 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1339 u64 sector, u32 blksize, u64 block_id)
1341 struct drbd_socket *sock;
1342 struct p_block_ack *p;
1344 if (peer_device->device->state.conn < C_CONNECTED)
1347 sock = &peer_device->connection->meta;
1348 p = drbd_prepare_command(peer_device, sock);
1352 p->block_id = block_id;
1353 p->blksize = blksize;
1354 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1355 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1358 /* dp->sector and dp->block_id already/still in network byte order,
1359 * data_size is payload size according to dp->head,
1360 * and may need to be corrected for digest size. */
1361 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1362 struct p_data *dp, int data_size)
1364 if (peer_device->connection->peer_integrity_tfm)
1365 data_size -= crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm);
1366 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1370 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1371 struct p_block_req *rp)
1373 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1377 * drbd_send_ack() - Sends an ack packet
1378 * @peer_device: DRBD peer device
1379 * @cmd: packet command code
1380 * @peer_req: peer request
1382 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1383 struct drbd_peer_request *peer_req)
1385 return _drbd_send_ack(peer_device, cmd,
1386 cpu_to_be64(peer_req->i.sector),
1387 cpu_to_be32(peer_req->i.size),
1388 peer_req->block_id);
1391 /* This function misuses the block_id field to signal if the blocks
1392 * are is sync or not. */
1393 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1394 sector_t sector, int blksize, u64 block_id)
1396 return _drbd_send_ack(peer_device, cmd,
1397 cpu_to_be64(sector),
1398 cpu_to_be32(blksize),
1399 cpu_to_be64(block_id));
1402 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1403 struct drbd_peer_request *peer_req)
1405 struct drbd_socket *sock;
1406 struct p_block_desc *p;
1408 sock = &peer_device->connection->data;
1409 p = drbd_prepare_command(peer_device, sock);
1412 p->sector = cpu_to_be64(peer_req->i.sector);
1413 p->blksize = cpu_to_be32(peer_req->i.size);
1415 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1418 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1419 sector_t sector, int size, u64 block_id)
1421 struct drbd_socket *sock;
1422 struct p_block_req *p;
1424 sock = &peer_device->connection->data;
1425 p = drbd_prepare_command(peer_device, sock);
1428 p->sector = cpu_to_be64(sector);
1429 p->block_id = block_id;
1430 p->blksize = cpu_to_be32(size);
1431 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1434 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1435 void *digest, int digest_size, enum drbd_packet cmd)
1437 struct drbd_socket *sock;
1438 struct p_block_req *p;
1440 /* FIXME: Put the digest into the preallocated socket buffer. */
1442 sock = &peer_device->connection->data;
1443 p = drbd_prepare_command(peer_device, sock);
1446 p->sector = cpu_to_be64(sector);
1447 p->block_id = ID_SYNCER /* unused */;
1448 p->blksize = cpu_to_be32(size);
1449 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1452 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1454 struct drbd_socket *sock;
1455 struct p_block_req *p;
1457 sock = &peer_device->connection->data;
1458 p = drbd_prepare_command(peer_device, sock);
1461 p->sector = cpu_to_be64(sector);
1462 p->block_id = ID_SYNCER /* unused */;
1463 p->blksize = cpu_to_be32(size);
1464 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1467 /* called on sndtimeo
1468 * returns false if we should retry,
1469 * true if we think connection is dead
1471 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1474 /* long elapsed = (long)(jiffies - device->last_received); */
1476 drop_it = connection->meta.socket == sock
1477 || !connection->ack_receiver.task
1478 || get_t_state(&connection->ack_receiver) != RUNNING
1479 || connection->cstate < C_WF_REPORT_PARAMS;
1484 drop_it = !--connection->ko_count;
1486 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1487 current->comm, current->pid, connection->ko_count);
1488 request_ping(connection);
1491 return drop_it; /* && (device->state == R_PRIMARY) */;
1494 static void drbd_update_congested(struct drbd_connection *connection)
1496 struct sock *sk = connection->data.socket->sk;
1497 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1498 set_bit(NET_CONGESTED, &connection->flags);
1501 /* The idea of sendpage seems to be to put some kind of reference
1502 * to the page into the skb, and to hand it over to the NIC. In
1503 * this process get_page() gets called.
1505 * As soon as the page was really sent over the network put_page()
1506 * gets called by some part of the network layer. [ NIC driver? ]
1508 * [ get_page() / put_page() increment/decrement the count. If count
1509 * reaches 0 the page will be freed. ]
1511 * This works nicely with pages from FSs.
1512 * But this means that in protocol A we might signal IO completion too early!
1514 * In order not to corrupt data during a resync we must make sure
1515 * that we do not reuse our own buffer pages (EEs) to early, therefore
1516 * we have the net_ee list.
1518 * XFS seems to have problems, still, it submits pages with page_count == 0!
1519 * As a workaround, we disable sendpage on pages
1520 * with page_count == 0 or PageSlab.
1522 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1523 int offset, size_t size, unsigned msg_flags)
1525 struct socket *socket;
1529 socket = peer_device->connection->data.socket;
1530 addr = kmap(page) + offset;
1531 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1534 peer_device->device->send_cnt += size >> 9;
1538 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1539 int offset, size_t size, unsigned msg_flags)
1541 struct socket *socket = peer_device->connection->data.socket;
1545 /* e.g. XFS meta- & log-data is in slab pages, which have a
1546 * page_count of 0 and/or have PageSlab() set.
1547 * we cannot use send_page for those, as that does get_page();
1548 * put_page(); and would cause either a VM_BUG directly, or
1549 * __page_cache_release a page that would actually still be referenced
1550 * by someone, leading to some obscure delayed Oops somewhere else. */
1551 if (drbd_disable_sendpage || !sendpage_ok(page))
1552 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1554 msg_flags |= MSG_NOSIGNAL;
1555 drbd_update_congested(peer_device->connection);
1559 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1561 if (sent == -EAGAIN) {
1562 if (we_should_drop_the_connection(peer_device->connection, socket))
1566 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1567 __func__, (int)size, len, sent);
1574 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1575 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1579 peer_device->device->send_cnt += size >> 9;
1584 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1586 struct bio_vec bvec;
1587 struct bvec_iter iter;
1589 /* hint all but last page with MSG_MORE */
1590 bio_for_each_segment(bvec, bio, iter) {
1593 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1594 bvec.bv_offset, bvec.bv_len,
1595 bio_iter_last(bvec, iter)
1603 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1605 struct bio_vec bvec;
1606 struct bvec_iter iter;
1608 /* hint all but last page with MSG_MORE */
1609 bio_for_each_segment(bvec, bio, iter) {
1612 err = _drbd_send_page(peer_device, bvec.bv_page,
1613 bvec.bv_offset, bvec.bv_len,
1614 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1621 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1622 struct drbd_peer_request *peer_req)
1624 struct page *page = peer_req->pages;
1625 unsigned len = peer_req->i.size;
1628 /* hint all but last page with MSG_MORE */
1629 page_chain_for_each(page) {
1630 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1632 err = _drbd_send_page(peer_device, page, 0, l,
1633 page_chain_next(page) ? MSG_MORE : 0);
1641 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1644 if (connection->agreed_pro_version >= 95)
1645 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1646 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1647 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1648 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1649 (bio_op(bio) == REQ_OP_WRITE_ZEROES ?
1650 ((connection->agreed_features & DRBD_FF_WZEROES) ?
1651 (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0))
1655 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1658 /* Used to send write or TRIM aka REQ_OP_DISCARD requests
1659 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1661 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1663 struct drbd_device *device = peer_device->device;
1664 struct drbd_socket *sock;
1667 unsigned int dp_flags = 0;
1671 sock = &peer_device->connection->data;
1672 p = drbd_prepare_command(peer_device, sock);
1673 digest_size = peer_device->connection->integrity_tfm ?
1674 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1678 p->sector = cpu_to_be64(req->i.sector);
1679 p->block_id = (unsigned long)req;
1680 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1681 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1682 if (device->state.conn >= C_SYNC_SOURCE &&
1683 device->state.conn <= C_PAUSED_SYNC_T)
1684 dp_flags |= DP_MAY_SET_IN_SYNC;
1685 if (peer_device->connection->agreed_pro_version >= 100) {
1686 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1687 dp_flags |= DP_SEND_RECEIVE_ACK;
1688 /* During resync, request an explicit write ack,
1689 * even in protocol != C */
1690 if (req->rq_state & RQ_EXP_WRITE_ACK
1691 || (dp_flags & DP_MAY_SET_IN_SYNC))
1692 dp_flags |= DP_SEND_WRITE_ACK;
1694 p->dp_flags = cpu_to_be32(dp_flags);
1696 if (dp_flags & (DP_DISCARD|DP_ZEROES)) {
1697 enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM;
1698 struct p_trim *t = (struct p_trim*)p;
1699 t->size = cpu_to_be32(req->i.size);
1700 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0);
1705 /* our digest is still only over the payload.
1706 * TRIM does not carry any payload. */
1708 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1709 err = __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1710 sizeof(*p) + digest_size, NULL, req->i.size);
1712 /* For protocol A, we have to memcpy the payload into
1713 * socket buffers, as we may complete right away
1714 * as soon as we handed it over to tcp, at which point the data
1715 * pages may become invalid.
1717 * For data-integrity enabled, we copy it as well, so we can be
1718 * sure that even if the bio pages may still be modified, it
1719 * won't change the data on the wire, thus if the digest checks
1720 * out ok after sending on this side, but does not fit on the
1721 * receiving side, we sure have detected corruption elsewhere.
1723 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1724 err = _drbd_send_bio(peer_device, req->master_bio);
1726 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1728 /* double check digest, sometimes buffers have been modified in flight. */
1729 if (digest_size > 0 && digest_size <= 64) {
1730 /* 64 byte, 512 bit, is the largest digest size
1731 * currently supported in kernel crypto. */
1732 unsigned char digest[64];
1733 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1734 if (memcmp(p + 1, digest, digest_size)) {
1736 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1737 (unsigned long long)req->i.sector, req->i.size);
1739 } /* else if (digest_size > 64) {
1740 ... Be noisy about digest too large ...
1744 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1749 /* answer packet, used to send data back for read requests:
1750 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1751 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1753 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1754 struct drbd_peer_request *peer_req)
1756 struct drbd_device *device = peer_device->device;
1757 struct drbd_socket *sock;
1762 sock = &peer_device->connection->data;
1763 p = drbd_prepare_command(peer_device, sock);
1765 digest_size = peer_device->connection->integrity_tfm ?
1766 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1770 p->sector = cpu_to_be64(peer_req->i.sector);
1771 p->block_id = peer_req->block_id;
1772 p->seq_num = 0; /* unused */
1775 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1776 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1778 err = _drbd_send_zc_ee(peer_device, peer_req);
1779 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1784 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1786 struct drbd_socket *sock;
1787 struct p_block_desc *p;
1789 sock = &peer_device->connection->data;
1790 p = drbd_prepare_command(peer_device, sock);
1793 p->sector = cpu_to_be64(req->i.sector);
1794 p->blksize = cpu_to_be32(req->i.size);
1795 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1799 drbd_send distinguishes two cases:
1801 Packets sent via the data socket "sock"
1802 and packets sent via the meta data socket "msock"
1805 -----------------+-------------------------+------------------------------
1806 timeout conf.timeout / 2 conf.timeout / 2
1807 timeout action send a ping via msock Abort communication
1808 and close all sockets
1812 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1814 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1815 void *buf, size_t size, unsigned msg_flags)
1817 struct kvec iov = {.iov_base = buf, .iov_len = size};
1818 struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL};
1824 /* THINK if (signal_pending) return ... ? */
1826 iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, &iov, 1, size);
1828 if (sock == connection->data.socket) {
1830 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1832 drbd_update_congested(connection);
1835 rv = sock_sendmsg(sock, &msg);
1836 if (rv == -EAGAIN) {
1837 if (we_should_drop_the_connection(connection, sock))
1843 flush_signals(current);
1849 } while (sent < size);
1851 if (sock == connection->data.socket)
1852 clear_bit(NET_CONGESTED, &connection->flags);
1855 if (rv != -EAGAIN) {
1856 drbd_err(connection, "%s_sendmsg returned %d\n",
1857 sock == connection->meta.socket ? "msock" : "sock",
1859 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1861 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1868 * drbd_send_all - Send an entire buffer
1870 * Returns 0 upon success and a negative error value otherwise.
1872 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1873 size_t size, unsigned msg_flags)
1877 err = drbd_send(connection, sock, buffer, size, msg_flags);
1885 static int drbd_open(struct block_device *bdev, fmode_t mode)
1887 struct drbd_device *device = bdev->bd_disk->private_data;
1888 unsigned long flags;
1891 mutex_lock(&drbd_main_mutex);
1892 spin_lock_irqsave(&device->resource->req_lock, flags);
1893 /* to have a stable device->state.role
1894 * and no race with updating open_cnt */
1896 if (device->state.role != R_PRIMARY) {
1897 if (mode & FMODE_WRITE)
1899 else if (!drbd_allow_oos)
1905 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1906 mutex_unlock(&drbd_main_mutex);
1911 static void drbd_release(struct gendisk *gd, fmode_t mode)
1913 struct drbd_device *device = gd->private_data;
1914 mutex_lock(&drbd_main_mutex);
1916 mutex_unlock(&drbd_main_mutex);
1919 /* need to hold resource->req_lock */
1920 void drbd_queue_unplug(struct drbd_device *device)
1922 if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1923 D_ASSERT(device, device->state.role == R_PRIMARY);
1924 if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1925 drbd_queue_work_if_unqueued(
1926 &first_peer_device(device)->connection->sender_work,
1927 &device->unplug_work);
1932 static void drbd_set_defaults(struct drbd_device *device)
1934 /* Beware! The actual layout differs
1935 * between big endian and little endian */
1936 device->state = (union drbd_dev_state) {
1937 { .role = R_SECONDARY,
1939 .conn = C_STANDALONE,
1945 void drbd_init_set_defaults(struct drbd_device *device)
1947 /* the memset(,0,) did most of this.
1948 * note: only assignments, no allocation in here */
1950 drbd_set_defaults(device);
1952 atomic_set(&device->ap_bio_cnt, 0);
1953 atomic_set(&device->ap_actlog_cnt, 0);
1954 atomic_set(&device->ap_pending_cnt, 0);
1955 atomic_set(&device->rs_pending_cnt, 0);
1956 atomic_set(&device->unacked_cnt, 0);
1957 atomic_set(&device->local_cnt, 0);
1958 atomic_set(&device->pp_in_use_by_net, 0);
1959 atomic_set(&device->rs_sect_in, 0);
1960 atomic_set(&device->rs_sect_ev, 0);
1961 atomic_set(&device->ap_in_flight, 0);
1962 atomic_set(&device->md_io.in_use, 0);
1964 mutex_init(&device->own_state_mutex);
1965 device->state_mutex = &device->own_state_mutex;
1967 spin_lock_init(&device->al_lock);
1968 spin_lock_init(&device->peer_seq_lock);
1970 INIT_LIST_HEAD(&device->active_ee);
1971 INIT_LIST_HEAD(&device->sync_ee);
1972 INIT_LIST_HEAD(&device->done_ee);
1973 INIT_LIST_HEAD(&device->read_ee);
1974 INIT_LIST_HEAD(&device->net_ee);
1975 INIT_LIST_HEAD(&device->resync_reads);
1976 INIT_LIST_HEAD(&device->resync_work.list);
1977 INIT_LIST_HEAD(&device->unplug_work.list);
1978 INIT_LIST_HEAD(&device->bm_io_work.w.list);
1979 INIT_LIST_HEAD(&device->pending_master_completion[0]);
1980 INIT_LIST_HEAD(&device->pending_master_completion[1]);
1981 INIT_LIST_HEAD(&device->pending_completion[0]);
1982 INIT_LIST_HEAD(&device->pending_completion[1]);
1984 device->resync_work.cb = w_resync_timer;
1985 device->unplug_work.cb = w_send_write_hint;
1986 device->bm_io_work.w.cb = w_bitmap_io;
1988 timer_setup(&device->resync_timer, resync_timer_fn, 0);
1989 timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0);
1990 timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0);
1991 timer_setup(&device->request_timer, request_timer_fn, 0);
1993 init_waitqueue_head(&device->misc_wait);
1994 init_waitqueue_head(&device->state_wait);
1995 init_waitqueue_head(&device->ee_wait);
1996 init_waitqueue_head(&device->al_wait);
1997 init_waitqueue_head(&device->seq_wait);
1999 device->resync_wenr = LC_FREE;
2000 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2001 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2004 void drbd_set_my_capacity(struct drbd_device *device, sector_t size)
2008 set_capacity_and_notify(device->vdisk, size);
2010 drbd_info(device, "size = %s (%llu KB)\n",
2011 ppsize(ppb, size>>1), (unsigned long long)size>>1);
2014 void drbd_device_cleanup(struct drbd_device *device)
2017 if (first_peer_device(device)->connection->receiver.t_state != NONE)
2018 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2019 first_peer_device(device)->connection->receiver.t_state);
2021 device->al_writ_cnt =
2022 device->bm_writ_cnt =
2030 device->rs_failed = 0;
2031 device->rs_last_events = 0;
2032 device->rs_last_sect_ev = 0;
2033 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2034 device->rs_mark_left[i] = 0;
2035 device->rs_mark_time[i] = 0;
2037 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2039 set_capacity_and_notify(device->vdisk, 0);
2040 if (device->bitmap) {
2041 /* maybe never allocated. */
2042 drbd_bm_resize(device, 0, 1);
2043 drbd_bm_cleanup(device);
2046 drbd_backing_dev_free(device, device->ldev);
2047 device->ldev = NULL;
2049 clear_bit(AL_SUSPENDED, &device->flags);
2051 D_ASSERT(device, list_empty(&device->active_ee));
2052 D_ASSERT(device, list_empty(&device->sync_ee));
2053 D_ASSERT(device, list_empty(&device->done_ee));
2054 D_ASSERT(device, list_empty(&device->read_ee));
2055 D_ASSERT(device, list_empty(&device->net_ee));
2056 D_ASSERT(device, list_empty(&device->resync_reads));
2057 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2058 D_ASSERT(device, list_empty(&device->resync_work.list));
2059 D_ASSERT(device, list_empty(&device->unplug_work.list));
2061 drbd_set_defaults(device);
2065 static void drbd_destroy_mempools(void)
2069 while (drbd_pp_pool) {
2070 page = drbd_pp_pool;
2071 drbd_pp_pool = (struct page *)page_private(page);
2076 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2078 bioset_exit(&drbd_io_bio_set);
2079 bioset_exit(&drbd_md_io_bio_set);
2080 mempool_exit(&drbd_md_io_page_pool);
2081 mempool_exit(&drbd_ee_mempool);
2082 mempool_exit(&drbd_request_mempool);
2083 kmem_cache_destroy(drbd_ee_cache);
2084 kmem_cache_destroy(drbd_request_cache);
2085 kmem_cache_destroy(drbd_bm_ext_cache);
2086 kmem_cache_destroy(drbd_al_ext_cache);
2088 drbd_ee_cache = NULL;
2089 drbd_request_cache = NULL;
2090 drbd_bm_ext_cache = NULL;
2091 drbd_al_ext_cache = NULL;
2096 static int drbd_create_mempools(void)
2099 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2103 drbd_request_cache = kmem_cache_create(
2104 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2105 if (drbd_request_cache == NULL)
2108 drbd_ee_cache = kmem_cache_create(
2109 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2110 if (drbd_ee_cache == NULL)
2113 drbd_bm_ext_cache = kmem_cache_create(
2114 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2115 if (drbd_bm_ext_cache == NULL)
2118 drbd_al_ext_cache = kmem_cache_create(
2119 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2120 if (drbd_al_ext_cache == NULL)
2124 ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0);
2128 ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0,
2133 ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0);
2137 ret = mempool_init_slab_pool(&drbd_request_mempool, number,
2138 drbd_request_cache);
2142 ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache);
2146 for (i = 0; i < number; i++) {
2147 page = alloc_page(GFP_HIGHUSER);
2150 set_page_private(page, (unsigned long)drbd_pp_pool);
2151 drbd_pp_pool = page;
2153 drbd_pp_vacant = number;
2158 drbd_destroy_mempools(); /* in case we allocated some */
2162 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2166 rr = drbd_free_peer_reqs(device, &device->active_ee);
2168 drbd_err(device, "%d EEs in active list found!\n", rr);
2170 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2172 drbd_err(device, "%d EEs in sync list found!\n", rr);
2174 rr = drbd_free_peer_reqs(device, &device->read_ee);
2176 drbd_err(device, "%d EEs in read list found!\n", rr);
2178 rr = drbd_free_peer_reqs(device, &device->done_ee);
2180 drbd_err(device, "%d EEs in done list found!\n", rr);
2182 rr = drbd_free_peer_reqs(device, &device->net_ee);
2184 drbd_err(device, "%d EEs in net list found!\n", rr);
2187 /* caution. no locking. */
2188 void drbd_destroy_device(struct kref *kref)
2190 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2191 struct drbd_resource *resource = device->resource;
2192 struct drbd_peer_device *peer_device, *tmp_peer_device;
2194 timer_shutdown_sync(&device->request_timer);
2196 /* paranoia asserts */
2197 D_ASSERT(device, device->open_cnt == 0);
2198 /* end paranoia asserts */
2200 /* cleanup stuff that may have been allocated during
2201 * device (re-)configuration or state changes */
2203 drbd_backing_dev_free(device, device->ldev);
2204 device->ldev = NULL;
2206 drbd_release_all_peer_reqs(device);
2208 lc_destroy(device->act_log);
2209 lc_destroy(device->resync);
2211 kfree(device->p_uuid);
2212 /* device->p_uuid = NULL; */
2214 if (device->bitmap) /* should no longer be there. */
2215 drbd_bm_cleanup(device);
2216 __free_page(device->md_io.page);
2217 put_disk(device->vdisk);
2218 kfree(device->rs_plan_s);
2220 /* not for_each_connection(connection, resource):
2221 * those may have been cleaned up and disassociated already.
2223 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2224 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2227 if (device->submit.wq)
2228 destroy_workqueue(device->submit.wq);
2230 kref_put(&resource->kref, drbd_destroy_resource);
2233 /* One global retry thread, if we need to push back some bio and have it
2234 * reinserted through our make request function.
2236 static struct retry_worker {
2237 struct workqueue_struct *wq;
2238 struct work_struct worker;
2241 struct list_head writes;
2244 static void do_retry(struct work_struct *ws)
2246 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2248 struct drbd_request *req, *tmp;
2250 spin_lock_irq(&retry->lock);
2251 list_splice_init(&retry->writes, &writes);
2252 spin_unlock_irq(&retry->lock);
2254 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2255 struct drbd_device *device = req->device;
2256 struct bio *bio = req->master_bio;
2260 expect(device, atomic_read(&req->completion_ref) == 0) &&
2261 expect(device, req->rq_state & RQ_POSTPONED) &&
2262 expect(device, (req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2263 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2266 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2267 req, atomic_read(&req->completion_ref),
2270 /* We still need to put one kref associated with the
2271 * "completion_ref" going zero in the code path that queued it
2272 * here. The request object may still be referenced by a
2273 * frozen local req->private_bio, in case we force-detached.
2275 kref_put(&req->kref, drbd_req_destroy);
2277 /* A single suspended or otherwise blocking device may stall
2278 * all others as well. Fortunately, this code path is to
2279 * recover from a situation that "should not happen":
2280 * concurrent writes in multi-primary setup.
2281 * In a "normal" lifecycle, this workqueue is supposed to be
2282 * destroyed without ever doing anything.
2283 * If it turns out to be an issue anyways, we can do per
2284 * resource (replication group) or per device (minor) retry
2285 * workqueues instead.
2288 /* We are not just doing submit_bio_noacct(),
2289 * as we want to keep the start_time information. */
2291 __drbd_make_request(device, bio);
2295 /* called via drbd_req_put_completion_ref(),
2296 * holds resource->req_lock */
2297 void drbd_restart_request(struct drbd_request *req)
2299 unsigned long flags;
2300 spin_lock_irqsave(&retry.lock, flags);
2301 list_move_tail(&req->tl_requests, &retry.writes);
2302 spin_unlock_irqrestore(&retry.lock, flags);
2304 /* Drop the extra reference that would otherwise
2305 * have been dropped by complete_master_bio.
2306 * do_retry() needs to grab a new one. */
2307 dec_ap_bio(req->device);
2309 queue_work(retry.wq, &retry.worker);
2312 void drbd_destroy_resource(struct kref *kref)
2314 struct drbd_resource *resource =
2315 container_of(kref, struct drbd_resource, kref);
2317 idr_destroy(&resource->devices);
2318 free_cpumask_var(resource->cpu_mask);
2319 kfree(resource->name);
2323 void drbd_free_resource(struct drbd_resource *resource)
2325 struct drbd_connection *connection, *tmp;
2327 for_each_connection_safe(connection, tmp, resource) {
2328 list_del(&connection->connections);
2329 drbd_debugfs_connection_cleanup(connection);
2330 kref_put(&connection->kref, drbd_destroy_connection);
2332 drbd_debugfs_resource_cleanup(resource);
2333 kref_put(&resource->kref, drbd_destroy_resource);
2336 static void drbd_cleanup(void)
2339 struct drbd_device *device;
2340 struct drbd_resource *resource, *tmp;
2342 /* first remove proc,
2343 * drbdsetup uses it's presence to detect
2344 * whether DRBD is loaded.
2345 * If we would get stuck in proc removal,
2346 * but have netlink already deregistered,
2347 * some drbdsetup commands may wait forever
2351 remove_proc_entry("drbd", NULL);
2354 destroy_workqueue(retry.wq);
2356 drbd_genl_unregister();
2358 idr_for_each_entry(&drbd_devices, device, i)
2359 drbd_delete_device(device);
2361 /* not _rcu since, no other updater anymore. Genl already unregistered */
2362 for_each_resource_safe(resource, tmp, &drbd_resources) {
2363 list_del(&resource->resources);
2364 drbd_free_resource(resource);
2367 drbd_debugfs_cleanup();
2369 drbd_destroy_mempools();
2370 unregister_blkdev(DRBD_MAJOR, "drbd");
2372 idr_destroy(&drbd_devices);
2374 pr_info("module cleanup done.\n");
2377 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2379 spin_lock_init(&wq->q_lock);
2380 INIT_LIST_HEAD(&wq->q);
2381 init_waitqueue_head(&wq->q_wait);
2384 struct completion_work {
2386 struct completion done;
2389 static int w_complete(struct drbd_work *w, int cancel)
2391 struct completion_work *completion_work =
2392 container_of(w, struct completion_work, w);
2394 complete(&completion_work->done);
2398 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2400 struct completion_work completion_work;
2402 completion_work.w.cb = w_complete;
2403 init_completion(&completion_work.done);
2404 drbd_queue_work(work_queue, &completion_work.w);
2405 wait_for_completion(&completion_work.done);
2408 struct drbd_resource *drbd_find_resource(const char *name)
2410 struct drbd_resource *resource;
2412 if (!name || !name[0])
2416 for_each_resource_rcu(resource, &drbd_resources) {
2417 if (!strcmp(resource->name, name)) {
2418 kref_get(&resource->kref);
2428 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2429 void *peer_addr, int peer_addr_len)
2431 struct drbd_resource *resource;
2432 struct drbd_connection *connection;
2435 for_each_resource_rcu(resource, &drbd_resources) {
2436 for_each_connection_rcu(connection, resource) {
2437 if (connection->my_addr_len == my_addr_len &&
2438 connection->peer_addr_len == peer_addr_len &&
2439 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2440 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2441 kref_get(&connection->kref);
2452 static int drbd_alloc_socket(struct drbd_socket *socket)
2454 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2457 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2463 static void drbd_free_socket(struct drbd_socket *socket)
2465 free_page((unsigned long) socket->sbuf);
2466 free_page((unsigned long) socket->rbuf);
2469 void conn_free_crypto(struct drbd_connection *connection)
2471 drbd_free_sock(connection);
2473 crypto_free_shash(connection->csums_tfm);
2474 crypto_free_shash(connection->verify_tfm);
2475 crypto_free_shash(connection->cram_hmac_tfm);
2476 crypto_free_shash(connection->integrity_tfm);
2477 crypto_free_shash(connection->peer_integrity_tfm);
2478 kfree(connection->int_dig_in);
2479 kfree(connection->int_dig_vv);
2481 connection->csums_tfm = NULL;
2482 connection->verify_tfm = NULL;
2483 connection->cram_hmac_tfm = NULL;
2484 connection->integrity_tfm = NULL;
2485 connection->peer_integrity_tfm = NULL;
2486 connection->int_dig_in = NULL;
2487 connection->int_dig_vv = NULL;
2490 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2492 struct drbd_connection *connection;
2493 cpumask_var_t new_cpu_mask;
2496 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2499 /* silently ignore cpu mask on UP kernel */
2500 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2501 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2502 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2503 if (err == -EOVERFLOW) {
2504 /* So what. mask it out. */
2505 cpumask_var_t tmp_cpu_mask;
2506 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2507 cpumask_setall(tmp_cpu_mask);
2508 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2509 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2511 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2513 free_cpumask_var(tmp_cpu_mask);
2518 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2519 /* retcode = ERR_CPU_MASK_PARSE; */
2523 resource->res_opts = *res_opts;
2524 if (cpumask_empty(new_cpu_mask))
2525 drbd_calc_cpu_mask(&new_cpu_mask);
2526 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2527 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2528 for_each_connection_rcu(connection, resource) {
2529 connection->receiver.reset_cpu_mask = 1;
2530 connection->ack_receiver.reset_cpu_mask = 1;
2531 connection->worker.reset_cpu_mask = 1;
2537 free_cpumask_var(new_cpu_mask);
2542 struct drbd_resource *drbd_create_resource(const char *name)
2544 struct drbd_resource *resource;
2546 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2549 resource->name = kstrdup(name, GFP_KERNEL);
2550 if (!resource->name)
2551 goto fail_free_resource;
2552 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2553 goto fail_free_name;
2554 kref_init(&resource->kref);
2555 idr_init(&resource->devices);
2556 INIT_LIST_HEAD(&resource->connections);
2557 resource->write_ordering = WO_BDEV_FLUSH;
2558 list_add_tail_rcu(&resource->resources, &drbd_resources);
2559 mutex_init(&resource->conf_update);
2560 mutex_init(&resource->adm_mutex);
2561 spin_lock_init(&resource->req_lock);
2562 drbd_debugfs_resource_add(resource);
2566 kfree(resource->name);
2573 /* caller must be under adm_mutex */
2574 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2576 struct drbd_resource *resource;
2577 struct drbd_connection *connection;
2579 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2583 if (drbd_alloc_socket(&connection->data))
2585 if (drbd_alloc_socket(&connection->meta))
2588 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2589 if (!connection->current_epoch)
2592 INIT_LIST_HEAD(&connection->transfer_log);
2594 INIT_LIST_HEAD(&connection->current_epoch->list);
2595 connection->epochs = 1;
2596 spin_lock_init(&connection->epoch_lock);
2598 connection->send.seen_any_write_yet = false;
2599 connection->send.current_epoch_nr = 0;
2600 connection->send.current_epoch_writes = 0;
2602 resource = drbd_create_resource(name);
2606 connection->cstate = C_STANDALONE;
2607 mutex_init(&connection->cstate_mutex);
2608 init_waitqueue_head(&connection->ping_wait);
2609 idr_init(&connection->peer_devices);
2611 drbd_init_workqueue(&connection->sender_work);
2612 mutex_init(&connection->data.mutex);
2613 mutex_init(&connection->meta.mutex);
2615 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2616 connection->receiver.connection = connection;
2617 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2618 connection->worker.connection = connection;
2619 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2620 connection->ack_receiver.connection = connection;
2622 kref_init(&connection->kref);
2624 connection->resource = resource;
2626 if (set_resource_options(resource, res_opts))
2629 kref_get(&resource->kref);
2630 list_add_tail_rcu(&connection->connections, &resource->connections);
2631 drbd_debugfs_connection_add(connection);
2635 list_del(&resource->resources);
2636 drbd_free_resource(resource);
2638 kfree(connection->current_epoch);
2639 drbd_free_socket(&connection->meta);
2640 drbd_free_socket(&connection->data);
2645 void drbd_destroy_connection(struct kref *kref)
2647 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2648 struct drbd_resource *resource = connection->resource;
2650 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
2651 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2652 kfree(connection->current_epoch);
2654 idr_destroy(&connection->peer_devices);
2656 drbd_free_socket(&connection->meta);
2657 drbd_free_socket(&connection->data);
2658 kfree(connection->int_dig_in);
2659 kfree(connection->int_dig_vv);
2661 kref_put(&resource->kref, drbd_destroy_resource);
2664 static int init_submitter(struct drbd_device *device)
2666 /* opencoded create_singlethread_workqueue(),
2667 * to be able to say "drbd%d", ..., minor */
2669 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2670 if (!device->submit.wq)
2673 INIT_WORK(&device->submit.worker, do_submit);
2674 INIT_LIST_HEAD(&device->submit.writes);
2678 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2680 struct drbd_resource *resource = adm_ctx->resource;
2681 struct drbd_connection *connection, *n;
2682 struct drbd_device *device;
2683 struct drbd_peer_device *peer_device, *tmp_peer_device;
2684 struct gendisk *disk;
2686 int vnr = adm_ctx->volume;
2687 enum drbd_ret_code err = ERR_NOMEM;
2689 device = minor_to_device(minor);
2691 return ERR_MINOR_OR_VOLUME_EXISTS;
2693 /* GFP_KERNEL, we are outside of all write-out paths */
2694 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2697 kref_init(&device->kref);
2699 kref_get(&resource->kref);
2700 device->resource = resource;
2701 device->minor = minor;
2704 drbd_init_set_defaults(device);
2706 disk = blk_alloc_disk(NUMA_NO_NODE);
2710 device->vdisk = disk;
2711 device->rq_queue = disk->queue;
2713 set_disk_ro(disk, true);
2715 disk->major = DRBD_MAJOR;
2716 disk->first_minor = minor;
2718 disk->fops = &drbd_ops;
2719 disk->flags |= GENHD_FL_NO_PART;
2720 sprintf(disk->disk_name, "drbd%d", minor);
2721 disk->private_data = device;
2723 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, disk->queue);
2724 blk_queue_write_cache(disk->queue, true, true);
2725 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2726 This triggers a max_bio_size message upon first attach or connect */
2727 blk_queue_max_hw_sectors(disk->queue, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2729 device->md_io.page = alloc_page(GFP_KERNEL);
2730 if (!device->md_io.page)
2731 goto out_no_io_page;
2733 if (drbd_bm_init(device))
2735 device->read_requests = RB_ROOT;
2736 device->write_requests = RB_ROOT;
2738 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2741 err = ERR_MINOR_OR_VOLUME_EXISTS;
2742 goto out_no_minor_idr;
2744 kref_get(&device->kref);
2746 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2749 err = ERR_MINOR_OR_VOLUME_EXISTS;
2750 goto out_idr_remove_minor;
2752 kref_get(&device->kref);
2754 INIT_LIST_HEAD(&device->peer_devices);
2755 INIT_LIST_HEAD(&device->pending_bitmap_io);
2756 for_each_connection(connection, resource) {
2757 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2759 goto out_idr_remove_from_resource;
2760 peer_device->connection = connection;
2761 peer_device->device = device;
2763 list_add(&peer_device->peer_devices, &device->peer_devices);
2764 kref_get(&device->kref);
2766 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2769 err = ERR_INVALID_REQUEST;
2770 goto out_idr_remove_from_resource;
2772 kref_get(&connection->kref);
2773 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2776 if (init_submitter(device)) {
2778 goto out_idr_remove_from_resource;
2781 err = add_disk(disk);
2783 goto out_destroy_workqueue;
2785 /* inherit the connection state */
2786 device->state.conn = first_connection(resource)->cstate;
2787 if (device->state.conn == C_WF_REPORT_PARAMS) {
2788 for_each_peer_device(peer_device, device)
2789 drbd_connected(peer_device);
2791 /* move to create_peer_device() */
2792 for_each_peer_device(peer_device, device)
2793 drbd_debugfs_peer_device_add(peer_device);
2794 drbd_debugfs_device_add(device);
2797 out_destroy_workqueue:
2798 destroy_workqueue(device->submit.wq);
2799 out_idr_remove_from_resource:
2800 for_each_connection_safe(connection, n, resource) {
2801 peer_device = idr_remove(&connection->peer_devices, vnr);
2803 kref_put(&connection->kref, drbd_destroy_connection);
2805 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2806 list_del(&peer_device->peer_devices);
2809 idr_remove(&resource->devices, vnr);
2810 out_idr_remove_minor:
2811 idr_remove(&drbd_devices, minor);
2814 drbd_bm_cleanup(device);
2816 __free_page(device->md_io.page);
2820 kref_put(&resource->kref, drbd_destroy_resource);
2825 void drbd_delete_device(struct drbd_device *device)
2827 struct drbd_resource *resource = device->resource;
2828 struct drbd_connection *connection;
2829 struct drbd_peer_device *peer_device;
2831 /* move to free_peer_device() */
2832 for_each_peer_device(peer_device, device)
2833 drbd_debugfs_peer_device_cleanup(peer_device);
2834 drbd_debugfs_device_cleanup(device);
2835 for_each_connection(connection, resource) {
2836 idr_remove(&connection->peer_devices, device->vnr);
2837 kref_put(&device->kref, drbd_destroy_device);
2839 idr_remove(&resource->devices, device->vnr);
2840 kref_put(&device->kref, drbd_destroy_device);
2841 idr_remove(&drbd_devices, device_to_minor(device));
2842 kref_put(&device->kref, drbd_destroy_device);
2843 del_gendisk(device->vdisk);
2845 kref_put(&device->kref, drbd_destroy_device);
2848 static int __init drbd_init(void)
2852 if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2853 pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2857 drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2861 err = register_blkdev(DRBD_MAJOR, "drbd");
2863 pr_err("unable to register block device major %d\n",
2869 * allocate all necessary structs
2871 init_waitqueue_head(&drbd_pp_wait);
2873 drbd_proc = NULL; /* play safe for drbd_cleanup */
2874 idr_init(&drbd_devices);
2876 mutex_init(&resources_mutex);
2877 INIT_LIST_HEAD(&drbd_resources);
2879 err = drbd_genl_register();
2881 pr_err("unable to register generic netlink family\n");
2885 err = drbd_create_mempools();
2890 drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show);
2892 pr_err("unable to register proc file\n");
2896 retry.wq = create_singlethread_workqueue("drbd-reissue");
2898 pr_err("unable to create retry workqueue\n");
2901 INIT_WORK(&retry.worker, do_retry);
2902 spin_lock_init(&retry.lock);
2903 INIT_LIST_HEAD(&retry.writes);
2905 drbd_debugfs_init();
2907 pr_info("initialized. "
2908 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2909 GENL_MAGIC_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2910 pr_info("%s\n", drbd_buildtag());
2911 pr_info("registered as block device major %d\n", DRBD_MAJOR);
2912 return 0; /* Success! */
2917 pr_err("ran out of memory\n");
2919 pr_err("initialization failure\n");
2923 static void drbd_free_one_sock(struct drbd_socket *ds)
2926 mutex_lock(&ds->mutex);
2929 mutex_unlock(&ds->mutex);
2931 /* so debugfs does not need to mutex_lock() */
2933 kernel_sock_shutdown(s, SHUT_RDWR);
2938 void drbd_free_sock(struct drbd_connection *connection)
2940 if (connection->data.socket)
2941 drbd_free_one_sock(&connection->data);
2942 if (connection->meta.socket)
2943 drbd_free_one_sock(&connection->meta);
2946 /* meta data management */
2948 void conn_md_sync(struct drbd_connection *connection)
2950 struct drbd_peer_device *peer_device;
2954 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2955 struct drbd_device *device = peer_device->device;
2957 kref_get(&device->kref);
2959 drbd_md_sync(device);
2960 kref_put(&device->kref, drbd_destroy_device);
2966 /* aligned 4kByte */
2967 struct meta_data_on_disk {
2968 u64 la_size_sect; /* last agreed size. */
2969 u64 uuid[UI_SIZE]; /* UUIDs. */
2972 u32 flags; /* MDF */
2975 u32 al_offset; /* offset to this block */
2976 u32 al_nr_extents; /* important for restoring the AL (userspace) */
2977 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2978 u32 bm_offset; /* offset to the bitmap, from here */
2979 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
2980 u32 la_peer_max_bio_size; /* last peer max_bio_size */
2982 /* see al_tr_number_to_on_disk_sector() */
2984 u32 al_stripe_size_4k;
2986 u8 reserved_u8[4096 - (7*8 + 10*4)];
2991 void drbd_md_write(struct drbd_device *device, void *b)
2993 struct meta_data_on_disk *buffer = b;
2997 memset(buffer, 0, sizeof(*buffer));
2999 buffer->la_size_sect = cpu_to_be64(get_capacity(device->vdisk));
3000 for (i = UI_CURRENT; i < UI_SIZE; i++)
3001 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3002 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3003 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3005 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3006 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3007 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3008 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3009 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3011 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3012 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3014 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3015 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3017 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3018 sector = device->ldev->md.md_offset;
3020 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3021 /* this was a try anyways ... */
3022 drbd_err(device, "meta data update failed!\n");
3023 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3028 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3029 * @device: DRBD device.
3031 void drbd_md_sync(struct drbd_device *device)
3033 struct meta_data_on_disk *buffer;
3035 /* Don't accidentally change the DRBD meta data layout. */
3036 BUILD_BUG_ON(UI_SIZE != 4);
3037 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3039 del_timer(&device->md_sync_timer);
3040 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3041 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3044 /* We use here D_FAILED and not D_ATTACHING because we try to write
3045 * metadata even if we detach due to a disk failure! */
3046 if (!get_ldev_if_state(device, D_FAILED))
3049 buffer = drbd_md_get_buffer(device, __func__);
3053 drbd_md_write(device, buffer);
3055 /* Update device->ldev->md.la_size_sect,
3056 * since we updated it on metadata. */
3057 device->ldev->md.la_size_sect = get_capacity(device->vdisk);
3059 drbd_md_put_buffer(device);
3064 static int check_activity_log_stripe_size(struct drbd_device *device,
3065 struct meta_data_on_disk *on_disk,
3066 struct drbd_md *in_core)
3068 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3069 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3072 /* both not set: default to old fixed size activity log */
3073 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3075 al_stripe_size_4k = MD_32kB_SECT/8;
3078 /* some paranoia plausibility checks */
3080 /* we need both values to be set */
3081 if (al_stripes == 0 || al_stripe_size_4k == 0)
3084 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3086 /* Upper limit of activity log area, to avoid potential overflow
3087 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3088 * than 72 * 4k blocks total only increases the amount of history,
3089 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3090 if (al_size_4k > (16 * 1024 * 1024/4))
3093 /* Lower limit: we need at least 8 transaction slots (32kB)
3094 * to not break existing setups */
3095 if (al_size_4k < MD_32kB_SECT/8)
3098 in_core->al_stripe_size_4k = al_stripe_size_4k;
3099 in_core->al_stripes = al_stripes;
3100 in_core->al_size_4k = al_size_4k;
3104 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3105 al_stripes, al_stripe_size_4k);
3109 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3111 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3112 struct drbd_md *in_core = &bdev->md;
3113 s32 on_disk_al_sect;
3114 s32 on_disk_bm_sect;
3116 /* The on-disk size of the activity log, calculated from offsets, and
3117 * the size of the activity log calculated from the stripe settings,
3119 * Though we could relax this a bit: it is ok, if the striped activity log
3120 * fits in the available on-disk activity log size.
3121 * Right now, that would break how resize is implemented.
3122 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3123 * of possible unused padding space in the on disk layout. */
3124 if (in_core->al_offset < 0) {
3125 if (in_core->bm_offset > in_core->al_offset)
3127 on_disk_al_sect = -in_core->al_offset;
3128 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3130 if (in_core->al_offset != MD_4kB_SECT)
3132 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3135 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3136 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3139 /* old fixed size meta data is exactly that: fixed. */
3140 if (in_core->meta_dev_idx >= 0) {
3141 if (in_core->md_size_sect != MD_128MB_SECT
3142 || in_core->al_offset != MD_4kB_SECT
3143 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3144 || in_core->al_stripes != 1
3145 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3149 if (capacity < in_core->md_size_sect)
3151 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3154 /* should be aligned, and at least 32k */
3155 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3158 /* should fit (for now: exactly) into the available on-disk space;
3159 * overflow prevention is in check_activity_log_stripe_size() above. */
3160 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3163 /* again, should be aligned */
3164 if (in_core->bm_offset & 7)
3167 /* FIXME check for device grow with flex external meta data? */
3169 /* can the available bitmap space cover the last agreed device size? */
3170 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3176 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3177 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3178 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3179 in_core->meta_dev_idx,
3180 in_core->al_stripes, in_core->al_stripe_size_4k,
3181 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3182 (unsigned long long)in_core->la_size_sect,
3183 (unsigned long long)capacity);
3190 * drbd_md_read() - Reads in the meta data super block
3191 * @device: DRBD device.
3192 * @bdev: Device from which the meta data should be read in.
3194 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3195 * something goes wrong.
3197 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3198 * even before @bdev is assigned to @device->ldev.
3200 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3202 struct meta_data_on_disk *buffer;
3204 int i, rv = NO_ERROR;
3206 if (device->state.disk != D_DISKLESS)
3207 return ERR_DISK_CONFIGURED;
3209 buffer = drbd_md_get_buffer(device, __func__);
3213 /* First, figure out where our meta data superblock is located,
3215 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3216 bdev->md.md_offset = drbd_md_ss(bdev);
3217 /* Even for (flexible or indexed) external meta data,
3218 * initially restrict us to the 4k superblock for now.
3219 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3220 bdev->md.md_size_sect = 8;
3222 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3224 /* NOTE: can't do normal error processing here as this is
3225 called BEFORE disk is attached */
3226 drbd_err(device, "Error while reading metadata.\n");
3227 rv = ERR_IO_MD_DISK;
3231 magic = be32_to_cpu(buffer->magic);
3232 flags = be32_to_cpu(buffer->flags);
3233 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3234 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3235 /* btw: that's Activity Log clean, not "all" clean. */
3236 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3237 rv = ERR_MD_UNCLEAN;
3241 rv = ERR_MD_INVALID;
3242 if (magic != DRBD_MD_MAGIC_08) {
3243 if (magic == DRBD_MD_MAGIC_07)
3244 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3246 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3250 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3251 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3252 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3257 /* convert to in_core endian */
3258 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3259 for (i = UI_CURRENT; i < UI_SIZE; i++)
3260 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3261 bdev->md.flags = be32_to_cpu(buffer->flags);
3262 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3264 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3265 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3266 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3268 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3270 if (check_offsets_and_sizes(device, bdev))
3273 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3274 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3275 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3278 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3279 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3280 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3286 spin_lock_irq(&device->resource->req_lock);
3287 if (device->state.conn < C_CONNECTED) {
3289 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3290 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3291 device->peer_max_bio_size = peer;
3293 spin_unlock_irq(&device->resource->req_lock);
3296 drbd_md_put_buffer(device);
3302 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3303 * @device: DRBD device.
3305 * Call this function if you change anything that should be written to
3306 * the meta-data super block. This function sets MD_DIRTY, and starts a
3307 * timer that ensures that within five seconds you have to call drbd_md_sync().
3309 void drbd_md_mark_dirty(struct drbd_device *device)
3311 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3312 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3315 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3319 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3320 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3323 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3325 if (idx == UI_CURRENT) {
3326 if (device->state.role == R_PRIMARY)
3331 drbd_set_ed_uuid(device, val);
3334 device->ldev->md.uuid[idx] = val;
3335 drbd_md_mark_dirty(device);
3338 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3340 unsigned long flags;
3341 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3342 __drbd_uuid_set(device, idx, val);
3343 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3346 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3348 unsigned long flags;
3349 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3350 if (device->ldev->md.uuid[idx]) {
3351 drbd_uuid_move_history(device);
3352 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3354 __drbd_uuid_set(device, idx, val);
3355 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3359 * drbd_uuid_new_current() - Creates a new current UUID
3360 * @device: DRBD device.
3362 * Creates a new current UUID, and rotates the old current UUID into
3363 * the bitmap slot. Causes an incremental resync upon next connect.
3365 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3368 unsigned long long bm_uuid;
3370 get_random_bytes(&val, sizeof(u64));
3372 spin_lock_irq(&device->ldev->md.uuid_lock);
3373 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3376 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3378 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3379 __drbd_uuid_set(device, UI_CURRENT, val);
3380 spin_unlock_irq(&device->ldev->md.uuid_lock);
3382 drbd_print_uuids(device, "new current UUID");
3383 /* get it to stable storage _now_ */
3384 drbd_md_sync(device);
3387 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3389 unsigned long flags;
3390 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3393 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3395 drbd_uuid_move_history(device);
3396 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3397 device->ldev->md.uuid[UI_BITMAP] = 0;
3399 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3401 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3403 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3405 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3407 drbd_md_mark_dirty(device);
3411 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3412 * @device: DRBD device.
3414 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3416 int drbd_bmio_set_n_write(struct drbd_device *device,
3417 struct drbd_peer_device *peer_device) __must_hold(local)
3422 drbd_md_set_flag(device, MDF_FULL_SYNC);
3423 drbd_md_sync(device);
3424 drbd_bm_set_all(device);
3426 rv = drbd_bm_write(device, peer_device);
3429 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3430 drbd_md_sync(device);
3437 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3438 * @device: DRBD device.
3440 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3442 int drbd_bmio_clear_n_write(struct drbd_device *device,
3443 struct drbd_peer_device *peer_device) __must_hold(local)
3446 drbd_resume_al(device);
3447 drbd_bm_clear_all(device);
3448 return drbd_bm_write(device, peer_device);
3451 static int w_bitmap_io(struct drbd_work *w, int unused)
3453 struct drbd_device *device =
3454 container_of(w, struct drbd_device, bm_io_work.w);
3455 struct bm_io_work *work = &device->bm_io_work;
3458 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3459 int cnt = atomic_read(&device->ap_bio_cnt);
3461 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3465 if (get_ldev(device)) {
3466 drbd_bm_lock(device, work->why, work->flags);
3467 rv = work->io_fn(device, work->peer_device);
3468 drbd_bm_unlock(device);
3472 clear_bit_unlock(BITMAP_IO, &device->flags);
3473 wake_up(&device->misc_wait);
3476 work->done(device, rv);
3478 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3486 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3487 * @device: DRBD device.
3488 * @io_fn: IO callback to be called when bitmap IO is possible
3489 * @done: callback to be called after the bitmap IO was performed
3490 * @why: Descriptive text of the reason for doing the IO
3491 * @flags: Bitmap flags
3493 * While IO on the bitmap happens we freeze application IO thus we ensure
3494 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3495 * called from worker context. It MUST NOT be used while a previous such
3496 * work is still pending!
3498 * Its worker function encloses the call of io_fn() by get_ldev() and
3501 void drbd_queue_bitmap_io(struct drbd_device *device,
3502 int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
3503 void (*done)(struct drbd_device *, int),
3504 char *why, enum bm_flag flags,
3505 struct drbd_peer_device *peer_device)
3507 D_ASSERT(device, current == peer_device->connection->worker.task);
3509 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3510 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3511 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3512 if (device->bm_io_work.why)
3513 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3514 why, device->bm_io_work.why);
3516 device->bm_io_work.peer_device = peer_device;
3517 device->bm_io_work.io_fn = io_fn;
3518 device->bm_io_work.done = done;
3519 device->bm_io_work.why = why;
3520 device->bm_io_work.flags = flags;
3522 spin_lock_irq(&device->resource->req_lock);
3523 set_bit(BITMAP_IO, &device->flags);
3524 /* don't wait for pending application IO if the caller indicates that
3525 * application IO does not conflict anyways. */
3526 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3527 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3528 drbd_queue_work(&peer_device->connection->sender_work,
3529 &device->bm_io_work.w);
3531 spin_unlock_irq(&device->resource->req_lock);
3535 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3536 * @device: DRBD device.
3537 * @io_fn: IO callback to be called when bitmap IO is possible
3538 * @why: Descriptive text of the reason for doing the IO
3539 * @flags: Bitmap flags
3541 * freezes application IO while that the actual IO operations runs. This
3542 * functions MAY NOT be called from worker context.
3544 int drbd_bitmap_io(struct drbd_device *device,
3545 int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
3546 char *why, enum bm_flag flags,
3547 struct drbd_peer_device *peer_device)
3549 /* Only suspend io, if some operation is supposed to be locked out */
3550 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3553 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3556 drbd_suspend_io(device);
3558 drbd_bm_lock(device, why, flags);
3559 rv = io_fn(device, peer_device);
3560 drbd_bm_unlock(device);
3563 drbd_resume_io(device);
3568 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3570 if ((device->ldev->md.flags & flag) != flag) {
3571 drbd_md_mark_dirty(device);
3572 device->ldev->md.flags |= flag;
3576 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3578 if ((device->ldev->md.flags & flag) != 0) {
3579 drbd_md_mark_dirty(device);
3580 device->ldev->md.flags &= ~flag;
3583 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3585 return (bdev->md.flags & flag) != 0;
3588 static void md_sync_timer_fn(struct timer_list *t)
3590 struct drbd_device *device = from_timer(device, t, md_sync_timer);
3591 drbd_device_post_work(device, MD_SYNC);
3594 const char *cmdname(enum drbd_packet cmd)
3596 /* THINK may need to become several global tables
3597 * when we want to support more than
3598 * one PRO_VERSION */
3599 static const char *cmdnames[] = {
3602 [P_DATA_REPLY] = "DataReply",
3603 [P_RS_DATA_REPLY] = "RSDataReply",
3604 [P_BARRIER] = "Barrier",
3605 [P_BITMAP] = "ReportBitMap",
3606 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3607 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3608 [P_UNPLUG_REMOTE] = "UnplugRemote",
3609 [P_DATA_REQUEST] = "DataRequest",
3610 [P_RS_DATA_REQUEST] = "RSDataRequest",
3611 [P_SYNC_PARAM] = "SyncParam",
3612 [P_PROTOCOL] = "ReportProtocol",
3613 [P_UUIDS] = "ReportUUIDs",
3614 [P_SIZES] = "ReportSizes",
3615 [P_STATE] = "ReportState",
3616 [P_SYNC_UUID] = "ReportSyncUUID",
3617 [P_AUTH_CHALLENGE] = "AuthChallenge",
3618 [P_AUTH_RESPONSE] = "AuthResponse",
3619 [P_STATE_CHG_REQ] = "StateChgRequest",
3621 [P_PING_ACK] = "PingAck",
3622 [P_RECV_ACK] = "RecvAck",
3623 [P_WRITE_ACK] = "WriteAck",
3624 [P_RS_WRITE_ACK] = "RSWriteAck",
3625 [P_SUPERSEDED] = "Superseded",
3626 [P_NEG_ACK] = "NegAck",
3627 [P_NEG_DREPLY] = "NegDReply",
3628 [P_NEG_RS_DREPLY] = "NegRSDReply",
3629 [P_BARRIER_ACK] = "BarrierAck",
3630 [P_STATE_CHG_REPLY] = "StateChgReply",
3631 [P_OV_REQUEST] = "OVRequest",
3632 [P_OV_REPLY] = "OVReply",
3633 [P_OV_RESULT] = "OVResult",
3634 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3635 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3636 [P_SYNC_PARAM89] = "SyncParam89",
3637 [P_COMPRESSED_BITMAP] = "CBitmap",
3638 [P_DELAY_PROBE] = "DelayProbe",
3639 [P_OUT_OF_SYNC] = "OutOfSync",
3640 [P_RS_CANCEL] = "RSCancel",
3641 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3642 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3643 [P_PROTOCOL_UPDATE] = "protocol_update",
3645 [P_RS_THIN_REQ] = "rs_thin_req",
3646 [P_RS_DEALLOCATED] = "rs_deallocated",
3647 [P_WSAME] = "WriteSame",
3648 [P_ZEROES] = "Zeroes",
3650 /* enum drbd_packet, but not commands - obsoleted flags:
3656 /* too big for the array: 0xfffX */
3657 if (cmd == P_INITIAL_META)
3658 return "InitialMeta";
3659 if (cmd == P_INITIAL_DATA)
3660 return "InitialData";
3661 if (cmd == P_CONNECTION_FEATURES)
3662 return "ConnectionFeatures";
3663 if (cmd >= ARRAY_SIZE(cmdnames))
3665 return cmdnames[cmd];
3669 * drbd_wait_misc - wait for a request to make progress
3670 * @device: device associated with the request
3671 * @i: the struct drbd_interval embedded in struct drbd_request or
3672 * struct drbd_peer_request
3674 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3676 struct net_conf *nc;
3681 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3686 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3689 /* Indicate to wake up device->misc_wait on progress. */
3691 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3692 spin_unlock_irq(&device->resource->req_lock);
3693 timeout = schedule_timeout(timeout);
3694 finish_wait(&device->misc_wait, &wait);
3695 spin_lock_irq(&device->resource->req_lock);
3696 if (!timeout || device->state.conn < C_CONNECTED)
3698 if (signal_pending(current))
3699 return -ERESTARTSYS;
3703 void lock_all_resources(void)
3705 struct drbd_resource *resource;
3706 int __maybe_unused i = 0;
3708 mutex_lock(&resources_mutex);
3709 local_irq_disable();
3710 for_each_resource(resource, &drbd_resources)
3711 spin_lock_nested(&resource->req_lock, i++);
3714 void unlock_all_resources(void)
3716 struct drbd_resource *resource;
3718 for_each_resource(resource, &drbd_resources)
3719 spin_unlock(&resource->req_lock);
3721 mutex_unlock(&resources_mutex);
3724 #ifdef CONFIG_DRBD_FAULT_INJECTION
3725 /* Fault insertion support including random number generator shamelessly
3726 * stolen from kernel/rcutorture.c */
3727 struct fault_random_state {
3728 unsigned long state;
3729 unsigned long count;
3732 #define FAULT_RANDOM_MULT 39916801 /* prime */
3733 #define FAULT_RANDOM_ADD 479001701 /* prime */
3734 #define FAULT_RANDOM_REFRESH 10000
3737 * Crude but fast random-number generator. Uses a linear congruential
3738 * generator, with occasional help from get_random_bytes().
3740 static unsigned long
3741 _drbd_fault_random(struct fault_random_state *rsp)
3745 if (!rsp->count--) {
3746 get_random_bytes(&refresh, sizeof(refresh));
3747 rsp->state += refresh;
3748 rsp->count = FAULT_RANDOM_REFRESH;
3750 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3751 return swahw32(rsp->state);
3755 _drbd_fault_str(unsigned int type) {
3756 static char *_faults[] = {
3757 [DRBD_FAULT_MD_WR] = "Meta-data write",
3758 [DRBD_FAULT_MD_RD] = "Meta-data read",
3759 [DRBD_FAULT_RS_WR] = "Resync write",
3760 [DRBD_FAULT_RS_RD] = "Resync read",
3761 [DRBD_FAULT_DT_WR] = "Data write",
3762 [DRBD_FAULT_DT_RD] = "Data read",
3763 [DRBD_FAULT_DT_RA] = "Data read ahead",
3764 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3765 [DRBD_FAULT_AL_EE] = "EE allocation",
3766 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3769 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3773 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3775 static struct fault_random_state rrs = {0, 0};
3777 unsigned int ret = (
3778 (drbd_fault_devs == 0 ||
3779 ((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3780 (((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3785 if (drbd_ratelimit())
3786 drbd_warn(device, "***Simulating %s failure\n",
3787 _drbd_fault_str(type));
3794 module_init(drbd_init)
3795 module_exit(drbd_cleanup)
3797 EXPORT_SYMBOL(drbd_conn_str);
3798 EXPORT_SYMBOL(drbd_role_str);
3799 EXPORT_SYMBOL(drbd_disk_str);
3800 EXPORT_SYMBOL(drbd_set_st_err_str);