xref: /linux/drivers/block/drbd/drbd_main.c (revision 4ae68b26c3ab5a82aa271e6e9fc9b1a06e1d6b40)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3    drbd.c
4 
5    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 
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>.
10 
11    Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
12    from Logicworks, Inc. for making SDP replication support possible.
13 
14 
15  */
16 
17 #define pr_fmt(fmt)	KBUILD_MODNAME ": " fmt
18 
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>
24 #include <net/sock.h>
25 #include <linux/ctype.h>
26 #include <linux/mutex.h>
27 #include <linux/fs.h>
28 #include <linux/file.h>
29 #include <linux/proc_fs.h>
30 #include <linux/init.h>
31 #include <linux/mm.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>
43 
44 #include <linux/drbd_limits.h>
45 #include "drbd_int.h"
46 #include "drbd_protocol.h"
47 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
48 #include "drbd_vli.h"
49 #include "drbd_debugfs.h"
50 
51 static DEFINE_MUTEX(drbd_main_mutex);
52 static int drbd_open(struct gendisk *disk, blk_mode_t mode);
53 static void drbd_release(struct gendisk *gd);
54 static void md_sync_timer_fn(struct timer_list *t);
55 static int w_bitmap_io(struct drbd_work *w, int unused);
56 
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);
65 
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) */
69 
70 #ifdef CONFIG_DRBD_FAULT_INJECTION
71 int drbd_enable_faults;
72 int drbd_fault_rate;
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);
83 #endif
84 
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);
91 
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);
102 
103 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
104  * as member "struct gendisk *vdisk;"
105  */
106 struct idr drbd_devices;
107 struct list_head drbd_resources;
108 struct mutex resources_mutex;
109 
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;
119 
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.
125  */
126 struct page *drbd_pp_pool;
127 DEFINE_SPINLOCK(drbd_pp_lock);
128 int          drbd_pp_vacant;
129 wait_queue_head_t drbd_pp_wait;
130 
131 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
132 
133 static const struct block_device_operations drbd_ops = {
134 	.owner		= THIS_MODULE,
135 	.submit_bio	= drbd_submit_bio,
136 	.open		= drbd_open,
137 	.release	= drbd_release,
138 };
139 
140 #ifdef __CHECKER__
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.
143  */
144 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
145 {
146 	int io_allowed;
147 
148 	atomic_inc(&device->local_cnt);
149 	io_allowed = (device->state.disk >= mins);
150 	if (!io_allowed) {
151 		if (atomic_dec_and_test(&device->local_cnt))
152 			wake_up(&device->misc_wait);
153 	}
154 	return io_allowed;
155 }
156 
157 #endif
158 
159 /**
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.
164  *
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.
168  */
169 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
170 		unsigned int set_size)
171 {
172 	struct drbd_request *r;
173 	struct drbd_request *req = NULL, *tmp = NULL;
174 	int expect_epoch = 0;
175 	int expect_size = 0;
176 
177 	spin_lock_irq(&connection->resource->req_lock);
178 
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;
183 		if (!req) {
184 			if (!(s & RQ_WRITE))
185 				continue;
186 			if (!(s & RQ_NET_MASK))
187 				continue;
188 			if (s & RQ_NET_DONE)
189 				continue;
190 			req = r;
191 			expect_epoch = req->epoch;
192 			expect_size ++;
193 		} else {
194 			if (r->epoch != expect_epoch)
195 				break;
196 			if (!(s & RQ_WRITE))
197 				continue;
198 			/* if (s & RQ_DONE): not expected */
199 			/* if (!(s & RQ_NET_MASK)): not expected */
200 			expect_size++;
201 		}
202 	}
203 
204 	/* first some paranoia code */
205 	if (req == NULL) {
206 		drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
207 			 barrier_nr);
208 		goto bail;
209 	}
210 	if (expect_epoch != barrier_nr) {
211 		drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
212 			 barrier_nr, expect_epoch);
213 		goto bail;
214 	}
215 
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);
219 		goto bail;
220 	}
221 
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) {
228 			tmp = req;
229 			break;
230 		}
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)
235 			break;
236 		peer_device = conn_peer_device(connection, req->device->vnr);
237 		_req_mod(req, BARRIER_ACKED, peer_device);
238 	}
239 	spin_unlock_irq(&connection->resource->req_lock);
240 
241 	return;
242 
243 bail:
244 	spin_unlock_irq(&connection->resource->req_lock);
245 	conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
246 }
247 
248 
249 /**
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
253  *
254  * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
255  * RESTART_FROZEN_DISK_IO.
256  */
257 /* must hold resource->req_lock */
258 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
259 {
260 	struct drbd_peer_device *peer_device;
261 	struct drbd_request *req, *r;
262 
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);
266 	}
267 }
268 
269 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
270 {
271 	spin_lock_irq(&connection->resource->req_lock);
272 	_tl_restart(connection, what);
273 	spin_unlock_irq(&connection->resource->req_lock);
274 }
275 
276 /**
277  * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
278  * @connection:	DRBD connection.
279  *
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.
283  */
284 void tl_clear(struct drbd_connection *connection)
285 {
286 	tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
287 }
288 
289 /**
290  * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
291  * @device:	DRBD device.
292  */
293 void tl_abort_disk_io(struct drbd_device *device)
294 {
295 	struct drbd_connection *connection = first_peer_device(device)->connection;
296 	struct drbd_request *req, *r;
297 
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))
301 			continue;
302 		if (req->device != device)
303 			continue;
304 		_req_mod(req, ABORT_DISK_IO, NULL);
305 	}
306 	spin_unlock_irq(&connection->resource->req_lock);
307 }
308 
309 static int drbd_thread_setup(void *arg)
310 {
311 	struct drbd_thread *thi = (struct drbd_thread *) arg;
312 	struct drbd_resource *resource = thi->resource;
313 	unsigned long flags;
314 	int retval;
315 
316 	snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
317 		 thi->name[0],
318 		 resource->name);
319 
320 	allow_kernel_signal(DRBD_SIGKILL);
321 	allow_kernel_signal(SIGXCPU);
322 restart:
323 	retval = thi->function(thi);
324 
325 	spin_lock_irqsave(&thi->t_lock, flags);
326 
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.
335 	 */
336 
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);
341 		goto restart;
342 	}
343 
344 	thi->task = NULL;
345 	thi->t_state = NONE;
346 	smp_mb();
347 	complete_all(&thi->stop);
348 	spin_unlock_irqrestore(&thi->t_lock, flags);
349 
350 	drbd_info(resource, "Terminating %s\n", current->comm);
351 
352 	/* Release mod reference taken when thread was started */
353 
354 	if (thi->connection)
355 		kref_put(&thi->connection->kref, drbd_destroy_connection);
356 	kref_put(&resource->kref, drbd_destroy_resource);
357 	module_put(THIS_MODULE);
358 	return retval;
359 }
360 
361 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
362 			     int (*func) (struct drbd_thread *), const char *name)
363 {
364 	spin_lock_init(&thi->t_lock);
365 	thi->task    = NULL;
366 	thi->t_state = NONE;
367 	thi->function = func;
368 	thi->resource = resource;
369 	thi->connection = NULL;
370 	thi->name = name;
371 }
372 
373 int drbd_thread_start(struct drbd_thread *thi)
374 {
375 	struct drbd_resource *resource = thi->resource;
376 	struct task_struct *nt;
377 	unsigned long flags;
378 
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);
382 
383 	switch (thi->t_state) {
384 	case NONE:
385 		drbd_info(resource, "Starting %s thread (from %s [%d])\n",
386 			 thi->name, current->comm, current->pid);
387 
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);
392 			return false;
393 		}
394 
395 		kref_get(&resource->kref);
396 		if (thi->connection)
397 			kref_get(&thi->connection->kref);
398 
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 */
404 
405 		nt = kthread_create(drbd_thread_setup, (void *) thi,
406 				    "drbd_%c_%s", thi->name[0], thi->resource->name);
407 
408 		if (IS_ERR(nt)) {
409 			drbd_err(resource, "Couldn't start thread\n");
410 
411 			if (thi->connection)
412 				kref_put(&thi->connection->kref, drbd_destroy_connection);
413 			kref_put(&resource->kref, drbd_destroy_resource);
414 			module_put(THIS_MODULE);
415 			return false;
416 		}
417 		spin_lock_irqsave(&thi->t_lock, flags);
418 		thi->task = nt;
419 		thi->t_state = RUNNING;
420 		spin_unlock_irqrestore(&thi->t_lock, flags);
421 		wake_up_process(nt);
422 		break;
423 	case EXITING:
424 		thi->t_state = RESTARTING;
425 		drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
426 				thi->name, current->comm, current->pid);
427 		fallthrough;
428 	case RUNNING:
429 	case RESTARTING:
430 	default:
431 		spin_unlock_irqrestore(&thi->t_lock, flags);
432 		break;
433 	}
434 
435 	return true;
436 }
437 
438 
439 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
440 {
441 	unsigned long flags;
442 
443 	enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
444 
445 	/* may be called from state engine, holding the req lock irqsave */
446 	spin_lock_irqsave(&thi->t_lock, flags);
447 
448 	if (thi->t_state == NONE) {
449 		spin_unlock_irqrestore(&thi->t_lock, flags);
450 		if (restart)
451 			drbd_thread_start(thi);
452 		return;
453 	}
454 
455 	if (thi->t_state != ns) {
456 		if (thi->task == NULL) {
457 			spin_unlock_irqrestore(&thi->t_lock, flags);
458 			return;
459 		}
460 
461 		thi->t_state = ns;
462 		smp_mb();
463 		init_completion(&thi->stop);
464 		if (thi->task != current)
465 			send_sig(DRBD_SIGKILL, thi->task, 1);
466 	}
467 
468 	spin_unlock_irqrestore(&thi->t_lock, flags);
469 
470 	if (wait)
471 		wait_for_completion(&thi->stop);
472 }
473 
474 int conn_lowest_minor(struct drbd_connection *connection)
475 {
476 	struct drbd_peer_device *peer_device;
477 	int vnr = 0, minor = -1;
478 
479 	rcu_read_lock();
480 	peer_device = idr_get_next(&connection->peer_devices, &vnr);
481 	if (peer_device)
482 		minor = device_to_minor(peer_device->device);
483 	rcu_read_unlock();
484 
485 	return minor;
486 }
487 
488 #ifdef CONFIG_SMP
489 /*
490  * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
491  *
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.
494  */
495 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
496 {
497 	unsigned int *resources_per_cpu, min_index = ~0;
498 
499 	resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu),
500 				    GFP_KERNEL);
501 	if (resources_per_cpu) {
502 		struct drbd_resource *resource;
503 		unsigned int cpu, min = ~0;
504 
505 		rcu_read_lock();
506 		for_each_resource_rcu(resource, &drbd_resources) {
507 			for_each_cpu(cpu, resource->cpu_mask)
508 				resources_per_cpu[cpu]++;
509 		}
510 		rcu_read_unlock();
511 		for_each_online_cpu(cpu) {
512 			if (resources_per_cpu[cpu] < min) {
513 				min = resources_per_cpu[cpu];
514 				min_index = cpu;
515 			}
516 		}
517 		kfree(resources_per_cpu);
518 	}
519 	if (min_index == ~0) {
520 		cpumask_setall(*cpu_mask);
521 		return;
522 	}
523 	cpumask_set_cpu(min_index, *cpu_mask);
524 }
525 
526 /**
527  * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
528  * @thi:	drbd_thread object
529  *
530  * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
531  * prematurely.
532  */
533 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
534 {
535 	struct drbd_resource *resource = thi->resource;
536 	struct task_struct *p = current;
537 
538 	if (!thi->reset_cpu_mask)
539 		return;
540 	thi->reset_cpu_mask = 0;
541 	set_cpus_allowed_ptr(p, resource->cpu_mask);
542 }
543 #else
544 #define drbd_calc_cpu_mask(A) ({})
545 #endif
546 
547 /*
548  * drbd_header_size  -  size of a packet header
549  *
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
552  * relies on this.)
553  */
554 unsigned int drbd_header_size(struct drbd_connection *connection)
555 {
556 	if (connection->agreed_pro_version >= 100) {
557 		BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
558 		return sizeof(struct p_header100);
559 	} else {
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);
564 	}
565 }
566 
567 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
568 {
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);
573 }
574 
575 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
576 {
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);
581 }
582 
583 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
584 				      int size, int vnr)
585 {
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);
590 	h->pad = 0;
591 	return sizeof(struct p_header100);
592 }
593 
594 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
595 				   void *buffer, enum drbd_packet cmd, int size)
596 {
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);
602 	else
603 		return prepare_header80(buffer, cmd, size);
604 }
605 
606 static void *__conn_prepare_command(struct drbd_connection *connection,
607 				    struct drbd_socket *sock)
608 {
609 	if (!sock->socket)
610 		return NULL;
611 	return sock->sbuf + drbd_header_size(connection);
612 }
613 
614 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
615 {
616 	void *p;
617 
618 	mutex_lock(&sock->mutex);
619 	p = __conn_prepare_command(connection, sock);
620 	if (!p)
621 		mutex_unlock(&sock->mutex);
622 
623 	return p;
624 }
625 
626 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
627 {
628 	return conn_prepare_command(peer_device->connection, sock);
629 }
630 
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,
634 			  unsigned int size)
635 {
636 	int msg_flags;
637 	int err;
638 
639 	/*
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
644 	 * receiver.
645 	 */
646 	msg_flags = data ? MSG_MORE : 0;
647 
648 	header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
649 				      header_size + size);
650 	err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
651 			    msg_flags);
652 	if (data && !err)
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);
658 
659 	return err;
660 }
661 
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)
665 {
666 	return __send_command(connection, 0, sock, cmd, header_size, data, size);
667 }
668 
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)
672 {
673 	int err;
674 
675 	err = __conn_send_command(connection, sock, cmd, header_size, data, size);
676 	mutex_unlock(&sock->mutex);
677 	return err;
678 }
679 
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)
683 {
684 	int err;
685 
686 	err = __send_command(peer_device->connection, peer_device->device->vnr,
687 			     sock, cmd, header_size, data, size);
688 	mutex_unlock(&sock->mutex);
689 	return err;
690 }
691 
692 int drbd_send_ping(struct drbd_connection *connection)
693 {
694 	struct drbd_socket *sock;
695 
696 	sock = &connection->meta;
697 	if (!conn_prepare_command(connection, sock))
698 		return -EIO;
699 	return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
700 }
701 
702 int drbd_send_ping_ack(struct drbd_connection *connection)
703 {
704 	struct drbd_socket *sock;
705 
706 	sock = &connection->meta;
707 	if (!conn_prepare_command(connection, sock))
708 		return -EIO;
709 	return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
710 }
711 
712 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
713 {
714 	struct drbd_socket *sock;
715 	struct p_rs_param_95 *p;
716 	int size;
717 	const int apv = peer_device->connection->agreed_pro_version;
718 	enum drbd_packet cmd;
719 	struct net_conf *nc;
720 	struct disk_conf *dc;
721 
722 	sock = &peer_device->connection->data;
723 	p = drbd_prepare_command(peer_device, sock);
724 	if (!p)
725 		return -EIO;
726 
727 	rcu_read_lock();
728 	nc = rcu_dereference(peer_device->connection->net_conf);
729 
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);
735 
736 	cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
737 
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));
741 
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);
750 	} else {
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);
756 	}
757 
758 	if (apv >= 88)
759 		strcpy(p->verify_alg, nc->verify_alg);
760 	if (apv >= 89)
761 		strcpy(p->csums_alg, nc->csums_alg);
762 	rcu_read_unlock();
763 
764 	return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
765 }
766 
767 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
768 {
769 	struct drbd_socket *sock;
770 	struct p_protocol *p;
771 	struct net_conf *nc;
772 	int size, cf;
773 
774 	sock = &connection->data;
775 	p = __conn_prepare_command(connection, sock);
776 	if (!p)
777 		return -EIO;
778 
779 	rcu_read_lock();
780 	nc = rcu_dereference(connection->net_conf);
781 
782 	if (nc->tentative && connection->agreed_pro_version < 92) {
783 		rcu_read_unlock();
784 		drbd_err(connection, "--dry-run is not supported by peer");
785 		return -EOPNOTSUPP;
786 	}
787 
788 	size = sizeof(*p);
789 	if (connection->agreed_pro_version >= 87)
790 		size += strlen(nc->integrity_alg) + 1;
791 
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);
797 	cf = 0;
798 	if (nc->discard_my_data)
799 		cf |= CF_DISCARD_MY_DATA;
800 	if (nc->tentative)
801 		cf |= CF_DRY_RUN;
802 	p->conn_flags    = cpu_to_be32(cf);
803 
804 	if (connection->agreed_pro_version >= 87)
805 		strcpy(p->integrity_alg, nc->integrity_alg);
806 	rcu_read_unlock();
807 
808 	return __conn_send_command(connection, sock, cmd, size, NULL, 0);
809 }
810 
811 int drbd_send_protocol(struct drbd_connection *connection)
812 {
813 	int err;
814 
815 	mutex_lock(&connection->data.mutex);
816 	err = __drbd_send_protocol(connection, P_PROTOCOL);
817 	mutex_unlock(&connection->data.mutex);
818 
819 	return err;
820 }
821 
822 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
823 {
824 	struct drbd_device *device = peer_device->device;
825 	struct drbd_socket *sock;
826 	struct p_uuids *p;
827 	int i;
828 
829 	if (!get_ldev_if_state(device, D_NEGOTIATING))
830 		return 0;
831 
832 	sock = &peer_device->connection->data;
833 	p = drbd_prepare_command(peer_device, sock);
834 	if (!p) {
835 		put_ldev(device);
836 		return -EIO;
837 	}
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);
842 
843 	device->comm_bm_set = drbd_bm_total_weight(device);
844 	p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
845 	rcu_read_lock();
846 	uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
847 	rcu_read_unlock();
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);
851 
852 	put_ldev(device);
853 	return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
854 }
855 
856 int drbd_send_uuids(struct drbd_peer_device *peer_device)
857 {
858 	return _drbd_send_uuids(peer_device, 0);
859 }
860 
861 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
862 {
863 	return _drbd_send_uuids(peer_device, 8);
864 }
865 
866 void drbd_print_uuids(struct drbd_device *device, const char *text)
867 {
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",
871 		     text,
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]);
876 		put_ldev(device);
877 	} else {
878 		drbd_info(device, "%s effective data uuid: %016llX\n",
879 				text,
880 				(unsigned long long)device->ed_uuid);
881 	}
882 }
883 
884 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
885 {
886 	struct drbd_device *device = peer_device->device;
887 	struct drbd_socket *sock;
888 	struct p_rs_uuid *p;
889 	u64 uuid;
890 
891 	D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
892 
893 	uuid = device->ldev->md.uuid[UI_BITMAP];
894 	if (uuid && uuid != UUID_JUST_CREATED)
895 		uuid = uuid + UUID_NEW_BM_OFFSET;
896 	else
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);
901 
902 	sock = &peer_device->connection->data;
903 	p = drbd_prepare_command(peer_device, sock);
904 	if (p) {
905 		p->uuid = cpu_to_be64(uuid);
906 		drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
907 	}
908 }
909 
910 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
911 {
912 	struct drbd_device *device = peer_device->device;
913 	struct drbd_socket *sock;
914 	struct p_sizes *p;
915 	sector_t d_size, u_size;
916 	int q_order_type;
917 	unsigned int max_bio_size;
918 	unsigned int packet_size;
919 
920 	sock = &peer_device->connection->data;
921 	p = drbd_prepare_command(peer_device, sock);
922 	if (!p)
923 		return -EIO;
924 
925 	packet_size = sizeof(*p);
926 	if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
927 		packet_size += sizeof(p->qlim[0]);
928 
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);
933 
934 		d_size = drbd_get_max_capacity(device->ldev);
935 		rcu_read_lock();
936 		u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
937 		rcu_read_unlock();
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);
950 		put_ldev(device);
951 	} else {
952 		struct request_queue *q = device->rq_queue;
953 
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;
962 
963 		d_size = 0;
964 		u_size = 0;
965 		q_order_type = QUEUE_ORDERED_NONE;
966 		max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
967 	}
968 
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);
973 
974 	p->d_size = cpu_to_be64(d_size);
975 	p->u_size = cpu_to_be64(u_size);
976 	if (trigger_reply)
977 		p->c_size = 0;
978 	else
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);
983 
984 	return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
985 }
986 
987 /**
988  * drbd_send_current_state() - Sends the drbd state to the peer
989  * @peer_device:	DRBD peer device.
990  */
991 int drbd_send_current_state(struct drbd_peer_device *peer_device)
992 {
993 	struct drbd_socket *sock;
994 	struct p_state *p;
995 
996 	sock = &peer_device->connection->data;
997 	p = drbd_prepare_command(peer_device, sock);
998 	if (!p)
999 		return -EIO;
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);
1002 }
1003 
1004 /**
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.
1008  *
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.
1013  */
1014 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1015 {
1016 	struct drbd_socket *sock;
1017 	struct p_state *p;
1018 
1019 	sock = &peer_device->connection->data;
1020 	p = drbd_prepare_command(peer_device, sock);
1021 	if (!p)
1022 		return -EIO;
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);
1025 }
1026 
1027 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1028 {
1029 	struct drbd_socket *sock;
1030 	struct p_req_state *p;
1031 
1032 	sock = &peer_device->connection->data;
1033 	p = drbd_prepare_command(peer_device, sock);
1034 	if (!p)
1035 		return -EIO;
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);
1039 }
1040 
1041 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1042 {
1043 	enum drbd_packet cmd;
1044 	struct drbd_socket *sock;
1045 	struct p_req_state *p;
1046 
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);
1050 	if (!p)
1051 		return -EIO;
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);
1055 }
1056 
1057 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1058 {
1059 	struct drbd_socket *sock;
1060 	struct p_req_state_reply *p;
1061 
1062 	sock = &peer_device->connection->meta;
1063 	p = drbd_prepare_command(peer_device, sock);
1064 	if (p) {
1065 		p->retcode = cpu_to_be32(retcode);
1066 		drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1067 	}
1068 }
1069 
1070 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1071 {
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;
1075 
1076 	sock = &connection->meta;
1077 	p = conn_prepare_command(connection, sock);
1078 	if (p) {
1079 		p->retcode = cpu_to_be32(retcode);
1080 		conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1081 	}
1082 }
1083 
1084 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1085 {
1086 	BUG_ON(code & ~0xf);
1087 	p->encoding = (p->encoding & ~0xf) | code;
1088 }
1089 
1090 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1091 {
1092 	p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1093 }
1094 
1095 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1096 {
1097 	BUG_ON(n & ~0x7);
1098 	p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1099 }
1100 
1101 static int fill_bitmap_rle_bits(struct drbd_device *device,
1102 			 struct p_compressed_bm *p,
1103 			 unsigned int size,
1104 			 struct bm_xfer_ctx *c)
1105 {
1106 	struct bitstream bs;
1107 	unsigned long plain_bits;
1108 	unsigned long tmp;
1109 	unsigned long rl;
1110 	unsigned len;
1111 	unsigned toggle;
1112 	int bits, use_rle;
1113 
1114 	/* may we use this feature? */
1115 	rcu_read_lock();
1116 	use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1117 	rcu_read_unlock();
1118 	if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1119 		return 0;
1120 
1121 	if (c->bit_offset >= c->bm_bits)
1122 		return 0; /* nothing to do. */
1123 
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 */
1128 	plain_bits = 0;
1129 
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 */
1133 	toggle = 2;
1134 
1135 	/* see how much plain bits we can stuff into one packet
1136 	 * using RLE and VLI. */
1137 	do {
1138 		tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1139 				    : _drbd_bm_find_next(device, c->bit_offset);
1140 		if (tmp == -1UL)
1141 			tmp = c->bm_bits;
1142 		rl = tmp - c->bit_offset;
1143 
1144 		if (toggle == 2) { /* first iteration */
1145 			if (rl == 0) {
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 */
1150 				toggle = !toggle;
1151 				continue;
1152 			}
1153 			dcbp_set_start(p, 0);
1154 		}
1155 
1156 		/* paranoia: catch zero runlength.
1157 		 * can only happen if bitmap is modified while we scan it. */
1158 		if (rl == 0) {
1159 			drbd_err(device, "unexpected zero runlength while encoding bitmap "
1160 			    "t:%u bo:%lu\n", toggle, c->bit_offset);
1161 			return -1;
1162 		}
1163 
1164 		bits = vli_encode_bits(&bs, rl);
1165 		if (bits == -ENOBUFS) /* buffer full */
1166 			break;
1167 		if (bits <= 0) {
1168 			drbd_err(device, "error while encoding bitmap: %d\n", bits);
1169 			return 0;
1170 		}
1171 
1172 		toggle = !toggle;
1173 		plain_bits += rl;
1174 		c->bit_offset = tmp;
1175 	} while (c->bit_offset < c->bm_bits);
1176 
1177 	len = bs.cur.b - p->code + !!bs.cur.bit;
1178 
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;
1185 		return 0;
1186 	}
1187 
1188 	/* RLE + VLI was able to compress it just fine.
1189 	 * update c->word_offset. */
1190 	bm_xfer_ctx_bit_to_word_offset(c);
1191 
1192 	/* store pad_bits */
1193 	dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1194 
1195 	return len;
1196 }
1197 
1198 /*
1199  * send_bitmap_rle_or_plain
1200  *
1201  * Return 0 when done, 1 when another iteration is needed, and a negative error
1202  * code upon failure.
1203  */
1204 static int
1205 send_bitmap_rle_or_plain(struct drbd_peer_device *peer_device, struct bm_xfer_ctx *c)
1206 {
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;
1211 	int len, err;
1212 
1213 	len = fill_bitmap_rle_bits(device, p,
1214 			DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1215 	if (len < 0)
1216 		return -EIO;
1217 
1218 	if (len) {
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,
1222 				     NULL, 0);
1223 		c->packets[0]++;
1224 		c->bytes[0] += header_size + sizeof(*p) + len;
1225 
1226 		if (c->bit_offset >= c->bm_bits)
1227 			len = 0; /* DONE */
1228 	} else {
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;
1234 
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);
1239 		if (len)
1240 			drbd_bm_get_lel(device, c->word_offset, num_words, p);
1241 		err = __send_command(peer_device->connection, device->vnr, sock, P_BITMAP,
1242 				     len, NULL, 0);
1243 		c->word_offset += num_words;
1244 		c->bit_offset = c->word_offset * BITS_PER_LONG;
1245 
1246 		c->packets[1]++;
1247 		c->bytes[1] += header_size + len;
1248 
1249 		if (c->bit_offset > c->bm_bits)
1250 			c->bit_offset = c->bm_bits;
1251 	}
1252 	if (!err) {
1253 		if (len == 0) {
1254 			INFO_bm_xfer_stats(peer_device, "send", c);
1255 			return 0;
1256 		} else
1257 			return 1;
1258 	}
1259 	return -EIO;
1260 }
1261 
1262 /* See the comment at receive_bitmap() */
1263 static int _drbd_send_bitmap(struct drbd_device *device,
1264 			    struct drbd_peer_device *peer_device)
1265 {
1266 	struct bm_xfer_ctx c;
1267 	int err;
1268 
1269 	if (!expect(device, device->bitmap))
1270 		return false;
1271 
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");
1281 			} else {
1282 				drbd_md_clear_flag(device, MDF_FULL_SYNC);
1283 				drbd_md_sync(device);
1284 			}
1285 		}
1286 		put_ldev(device);
1287 	}
1288 
1289 	c = (struct bm_xfer_ctx) {
1290 		.bm_bits = drbd_bm_bits(device),
1291 		.bm_words = drbd_bm_words(device),
1292 	};
1293 
1294 	do {
1295 		err = send_bitmap_rle_or_plain(peer_device, &c);
1296 	} while (err > 0);
1297 
1298 	return err == 0;
1299 }
1300 
1301 int drbd_send_bitmap(struct drbd_device *device, struct drbd_peer_device *peer_device)
1302 {
1303 	struct drbd_socket *sock = &peer_device->connection->data;
1304 	int err = -1;
1305 
1306 	mutex_lock(&sock->mutex);
1307 	if (sock->socket)
1308 		err = !_drbd_send_bitmap(device, peer_device);
1309 	mutex_unlock(&sock->mutex);
1310 	return err;
1311 }
1312 
1313 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1314 {
1315 	struct drbd_socket *sock;
1316 	struct p_barrier_ack *p;
1317 
1318 	if (connection->cstate < C_WF_REPORT_PARAMS)
1319 		return;
1320 
1321 	sock = &connection->meta;
1322 	p = conn_prepare_command(connection, sock);
1323 	if (!p)
1324 		return;
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);
1328 }
1329 
1330 /**
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
1337  */
1338 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1339 			  u64 sector, u32 blksize, u64 block_id)
1340 {
1341 	struct drbd_socket *sock;
1342 	struct p_block_ack *p;
1343 
1344 	if (peer_device->device->state.conn < C_CONNECTED)
1345 		return -EIO;
1346 
1347 	sock = &peer_device->connection->meta;
1348 	p = drbd_prepare_command(peer_device, sock);
1349 	if (!p)
1350 		return -EIO;
1351 	p->sector = sector;
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);
1356 }
1357 
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)
1363 {
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),
1367 		       dp->block_id);
1368 }
1369 
1370 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1371 		      struct p_block_req *rp)
1372 {
1373 	_drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1374 }
1375 
1376 /**
1377  * drbd_send_ack() - Sends an ack packet
1378  * @peer_device:	DRBD peer device
1379  * @cmd:		packet command code
1380  * @peer_req:		peer request
1381  */
1382 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1383 		  struct drbd_peer_request *peer_req)
1384 {
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);
1389 }
1390 
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)
1395 {
1396 	return _drbd_send_ack(peer_device, cmd,
1397 			      cpu_to_be64(sector),
1398 			      cpu_to_be32(blksize),
1399 			      cpu_to_be64(block_id));
1400 }
1401 
1402 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1403 			     struct drbd_peer_request *peer_req)
1404 {
1405 	struct drbd_socket *sock;
1406 	struct p_block_desc *p;
1407 
1408 	sock = &peer_device->connection->data;
1409 	p = drbd_prepare_command(peer_device, sock);
1410 	if (!p)
1411 		return -EIO;
1412 	p->sector = cpu_to_be64(peer_req->i.sector);
1413 	p->blksize = cpu_to_be32(peer_req->i.size);
1414 	p->pad = 0;
1415 	return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1416 }
1417 
1418 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1419 		       sector_t sector, int size, u64 block_id)
1420 {
1421 	struct drbd_socket *sock;
1422 	struct p_block_req *p;
1423 
1424 	sock = &peer_device->connection->data;
1425 	p = drbd_prepare_command(peer_device, sock);
1426 	if (!p)
1427 		return -EIO;
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);
1432 }
1433 
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)
1436 {
1437 	struct drbd_socket *sock;
1438 	struct p_block_req *p;
1439 
1440 	/* FIXME: Put the digest into the preallocated socket buffer.  */
1441 
1442 	sock = &peer_device->connection->data;
1443 	p = drbd_prepare_command(peer_device, sock);
1444 	if (!p)
1445 		return -EIO;
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);
1450 }
1451 
1452 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1453 {
1454 	struct drbd_socket *sock;
1455 	struct p_block_req *p;
1456 
1457 	sock = &peer_device->connection->data;
1458 	p = drbd_prepare_command(peer_device, sock);
1459 	if (!p)
1460 		return -EIO;
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);
1465 }
1466 
1467 /* called on sndtimeo
1468  * returns false if we should retry,
1469  * true if we think connection is dead
1470  */
1471 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1472 {
1473 	int drop_it;
1474 	/* long elapsed = (long)(jiffies - device->last_received); */
1475 
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;
1480 
1481 	if (drop_it)
1482 		return true;
1483 
1484 	drop_it = !--connection->ko_count;
1485 	if (!drop_it) {
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);
1489 	}
1490 
1491 	return drop_it; /* && (device->state == R_PRIMARY) */;
1492 }
1493 
1494 static void drbd_update_congested(struct drbd_connection *connection)
1495 {
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);
1499 }
1500 
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.
1504  *
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? ]
1507  *
1508  * [ get_page() / put_page() increment/decrement the count. If count
1509  *   reaches 0 the page will be freed. ]
1510  *
1511  * This works nicely with pages from FSs.
1512  * But this means that in protocol A we might signal IO completion too early!
1513  *
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.
1517  *
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.
1521  */
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)
1524 {
1525 	struct socket *socket;
1526 	void *addr;
1527 	int err;
1528 
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);
1532 	kunmap(page);
1533 	if (!err)
1534 		peer_device->device->send_cnt += size >> 9;
1535 	return err;
1536 }
1537 
1538 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1539 		    int offset, size_t size, unsigned msg_flags)
1540 {
1541 	struct socket *socket = peer_device->connection->data.socket;
1542 	struct msghdr msg = { .msg_flags = msg_flags, };
1543 	struct bio_vec bvec;
1544 	int len = size;
1545 	int err = -EIO;
1546 
1547 	/* e.g. XFS meta- & log-data is in slab pages, which have a
1548 	 * page_count of 0 and/or have PageSlab() set.
1549 	 * we cannot use send_page for those, as that does get_page();
1550 	 * put_page(); and would cause either a VM_BUG directly, or
1551 	 * __page_cache_release a page that would actually still be referenced
1552 	 * by someone, leading to some obscure delayed Oops somewhere else. */
1553 	if (!drbd_disable_sendpage && sendpage_ok(page))
1554 		msg.msg_flags |= MSG_NOSIGNAL | MSG_SPLICE_PAGES;
1555 
1556 	drbd_update_congested(peer_device->connection);
1557 	do {
1558 		int sent;
1559 
1560 		bvec_set_page(&bvec, page, offset, len);
1561 		iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, len);
1562 
1563 		sent = sock_sendmsg(socket, &msg);
1564 		if (sent <= 0) {
1565 			if (sent == -EAGAIN) {
1566 				if (we_should_drop_the_connection(peer_device->connection, socket))
1567 					break;
1568 				continue;
1569 			}
1570 			drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1571 			     __func__, (int)size, len, sent);
1572 			if (sent < 0)
1573 				err = sent;
1574 			break;
1575 		}
1576 		len    -= sent;
1577 		offset += sent;
1578 	} while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1579 	clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1580 
1581 	if (len == 0) {
1582 		err = 0;
1583 		peer_device->device->send_cnt += size >> 9;
1584 	}
1585 	return err;
1586 }
1587 
1588 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1589 {
1590 	struct bio_vec bvec;
1591 	struct bvec_iter iter;
1592 
1593 	/* hint all but last page with MSG_MORE */
1594 	bio_for_each_segment(bvec, bio, iter) {
1595 		int err;
1596 
1597 		err = _drbd_no_send_page(peer_device, bvec.bv_page,
1598 					 bvec.bv_offset, bvec.bv_len,
1599 					 bio_iter_last(bvec, iter)
1600 					 ? 0 : MSG_MORE);
1601 		if (err)
1602 			return err;
1603 	}
1604 	return 0;
1605 }
1606 
1607 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1608 {
1609 	struct bio_vec bvec;
1610 	struct bvec_iter iter;
1611 
1612 	/* hint all but last page with MSG_MORE */
1613 	bio_for_each_segment(bvec, bio, iter) {
1614 		int err;
1615 
1616 		err = _drbd_send_page(peer_device, bvec.bv_page,
1617 				      bvec.bv_offset, bvec.bv_len,
1618 				      bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1619 		if (err)
1620 			return err;
1621 	}
1622 	return 0;
1623 }
1624 
1625 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1626 			    struct drbd_peer_request *peer_req)
1627 {
1628 	struct page *page = peer_req->pages;
1629 	unsigned len = peer_req->i.size;
1630 	int err;
1631 
1632 	/* hint all but last page with MSG_MORE */
1633 	page_chain_for_each(page) {
1634 		unsigned l = min_t(unsigned, len, PAGE_SIZE);
1635 
1636 		err = _drbd_send_page(peer_device, page, 0, l,
1637 				      page_chain_next(page) ? MSG_MORE : 0);
1638 		if (err)
1639 			return err;
1640 		len -= l;
1641 	}
1642 	return 0;
1643 }
1644 
1645 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1646 			     struct bio *bio)
1647 {
1648 	if (connection->agreed_pro_version >= 95)
1649 		return  (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1650 			(bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1651 			(bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1652 			(bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1653 			(bio_op(bio) == REQ_OP_WRITE_ZEROES ?
1654 			  ((connection->agreed_features & DRBD_FF_WZEROES) ?
1655 			   (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0))
1656 			   : DP_DISCARD)
1657 			: 0);
1658 	else
1659 		return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1660 }
1661 
1662 /* Used to send write or TRIM aka REQ_OP_DISCARD requests
1663  * R_PRIMARY -> Peer	(P_DATA, P_TRIM)
1664  */
1665 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1666 {
1667 	struct drbd_device *device = peer_device->device;
1668 	struct drbd_socket *sock;
1669 	struct p_data *p;
1670 	void *digest_out;
1671 	unsigned int dp_flags = 0;
1672 	int digest_size;
1673 	int err;
1674 
1675 	sock = &peer_device->connection->data;
1676 	p = drbd_prepare_command(peer_device, sock);
1677 	digest_size = peer_device->connection->integrity_tfm ?
1678 		      crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1679 
1680 	if (!p)
1681 		return -EIO;
1682 	p->sector = cpu_to_be64(req->i.sector);
1683 	p->block_id = (unsigned long)req;
1684 	p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1685 	dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1686 	if (device->state.conn >= C_SYNC_SOURCE &&
1687 	    device->state.conn <= C_PAUSED_SYNC_T)
1688 		dp_flags |= DP_MAY_SET_IN_SYNC;
1689 	if (peer_device->connection->agreed_pro_version >= 100) {
1690 		if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1691 			dp_flags |= DP_SEND_RECEIVE_ACK;
1692 		/* During resync, request an explicit write ack,
1693 		 * even in protocol != C */
1694 		if (req->rq_state & RQ_EXP_WRITE_ACK
1695 		|| (dp_flags & DP_MAY_SET_IN_SYNC))
1696 			dp_flags |= DP_SEND_WRITE_ACK;
1697 	}
1698 	p->dp_flags = cpu_to_be32(dp_flags);
1699 
1700 	if (dp_flags & (DP_DISCARD|DP_ZEROES)) {
1701 		enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM;
1702 		struct p_trim *t = (struct p_trim*)p;
1703 		t->size = cpu_to_be32(req->i.size);
1704 		err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0);
1705 		goto out;
1706 	}
1707 	digest_out = p + 1;
1708 
1709 	/* our digest is still only over the payload.
1710 	 * TRIM does not carry any payload. */
1711 	if (digest_size)
1712 		drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1713 	err = __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1714 			     sizeof(*p) + digest_size, NULL, req->i.size);
1715 	if (!err) {
1716 		/* For protocol A, we have to memcpy the payload into
1717 		 * socket buffers, as we may complete right away
1718 		 * as soon as we handed it over to tcp, at which point the data
1719 		 * pages may become invalid.
1720 		 *
1721 		 * For data-integrity enabled, we copy it as well, so we can be
1722 		 * sure that even if the bio pages may still be modified, it
1723 		 * won't change the data on the wire, thus if the digest checks
1724 		 * out ok after sending on this side, but does not fit on the
1725 		 * receiving side, we sure have detected corruption elsewhere.
1726 		 */
1727 		if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1728 			err = _drbd_send_bio(peer_device, req->master_bio);
1729 		else
1730 			err = _drbd_send_zc_bio(peer_device, req->master_bio);
1731 
1732 		/* double check digest, sometimes buffers have been modified in flight. */
1733 		if (digest_size > 0 && digest_size <= 64) {
1734 			/* 64 byte, 512 bit, is the largest digest size
1735 			 * currently supported in kernel crypto. */
1736 			unsigned char digest[64];
1737 			drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1738 			if (memcmp(p + 1, digest, digest_size)) {
1739 				drbd_warn(device,
1740 					"Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1741 					(unsigned long long)req->i.sector, req->i.size);
1742 			}
1743 		} /* else if (digest_size > 64) {
1744 		     ... Be noisy about digest too large ...
1745 		} */
1746 	}
1747 out:
1748 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1749 
1750 	return err;
1751 }
1752 
1753 /* answer packet, used to send data back for read requests:
1754  *  Peer       -> (diskless) R_PRIMARY   (P_DATA_REPLY)
1755  *  C_SYNC_SOURCE -> C_SYNC_TARGET         (P_RS_DATA_REPLY)
1756  */
1757 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1758 		    struct drbd_peer_request *peer_req)
1759 {
1760 	struct drbd_device *device = peer_device->device;
1761 	struct drbd_socket *sock;
1762 	struct p_data *p;
1763 	int err;
1764 	int digest_size;
1765 
1766 	sock = &peer_device->connection->data;
1767 	p = drbd_prepare_command(peer_device, sock);
1768 
1769 	digest_size = peer_device->connection->integrity_tfm ?
1770 		      crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1771 
1772 	if (!p)
1773 		return -EIO;
1774 	p->sector = cpu_to_be64(peer_req->i.sector);
1775 	p->block_id = peer_req->block_id;
1776 	p->seq_num = 0;  /* unused */
1777 	p->dp_flags = 0;
1778 	if (digest_size)
1779 		drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1780 	err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1781 	if (!err)
1782 		err = _drbd_send_zc_ee(peer_device, peer_req);
1783 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1784 
1785 	return err;
1786 }
1787 
1788 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1789 {
1790 	struct drbd_socket *sock;
1791 	struct p_block_desc *p;
1792 
1793 	sock = &peer_device->connection->data;
1794 	p = drbd_prepare_command(peer_device, sock);
1795 	if (!p)
1796 		return -EIO;
1797 	p->sector = cpu_to_be64(req->i.sector);
1798 	p->blksize = cpu_to_be32(req->i.size);
1799 	return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1800 }
1801 
1802 /*
1803   drbd_send distinguishes two cases:
1804 
1805   Packets sent via the data socket "sock"
1806   and packets sent via the meta data socket "msock"
1807 
1808 		    sock                      msock
1809   -----------------+-------------------------+------------------------------
1810   timeout           conf.timeout / 2          conf.timeout / 2
1811   timeout action    send a ping via msock     Abort communication
1812 					      and close all sockets
1813 */
1814 
1815 /*
1816  * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1817  */
1818 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1819 	      void *buf, size_t size, unsigned msg_flags)
1820 {
1821 	struct kvec iov = {.iov_base = buf, .iov_len = size};
1822 	struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL};
1823 	int rv, sent = 0;
1824 
1825 	if (!sock)
1826 		return -EBADR;
1827 
1828 	/* THINK  if (signal_pending) return ... ? */
1829 
1830 	iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, &iov, 1, size);
1831 
1832 	if (sock == connection->data.socket) {
1833 		rcu_read_lock();
1834 		connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1835 		rcu_read_unlock();
1836 		drbd_update_congested(connection);
1837 	}
1838 	do {
1839 		rv = sock_sendmsg(sock, &msg);
1840 		if (rv == -EAGAIN) {
1841 			if (we_should_drop_the_connection(connection, sock))
1842 				break;
1843 			else
1844 				continue;
1845 		}
1846 		if (rv == -EINTR) {
1847 			flush_signals(current);
1848 			rv = 0;
1849 		}
1850 		if (rv < 0)
1851 			break;
1852 		sent += rv;
1853 	} while (sent < size);
1854 
1855 	if (sock == connection->data.socket)
1856 		clear_bit(NET_CONGESTED, &connection->flags);
1857 
1858 	if (rv <= 0) {
1859 		if (rv != -EAGAIN) {
1860 			drbd_err(connection, "%s_sendmsg returned %d\n",
1861 				 sock == connection->meta.socket ? "msock" : "sock",
1862 				 rv);
1863 			conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1864 		} else
1865 			conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1866 	}
1867 
1868 	return sent;
1869 }
1870 
1871 /*
1872  * drbd_send_all  -  Send an entire buffer
1873  *
1874  * Returns 0 upon success and a negative error value otherwise.
1875  */
1876 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1877 		  size_t size, unsigned msg_flags)
1878 {
1879 	int err;
1880 
1881 	err = drbd_send(connection, sock, buffer, size, msg_flags);
1882 	if (err < 0)
1883 		return err;
1884 	if (err != size)
1885 		return -EIO;
1886 	return 0;
1887 }
1888 
1889 static int drbd_open(struct gendisk *disk, blk_mode_t mode)
1890 {
1891 	struct drbd_device *device = disk->private_data;
1892 	unsigned long flags;
1893 	int rv = 0;
1894 
1895 	mutex_lock(&drbd_main_mutex);
1896 	spin_lock_irqsave(&device->resource->req_lock, flags);
1897 	/* to have a stable device->state.role
1898 	 * and no race with updating open_cnt */
1899 
1900 	if (device->state.role != R_PRIMARY) {
1901 		if (mode & BLK_OPEN_WRITE)
1902 			rv = -EROFS;
1903 		else if (!drbd_allow_oos)
1904 			rv = -EMEDIUMTYPE;
1905 	}
1906 
1907 	if (!rv)
1908 		device->open_cnt++;
1909 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
1910 	mutex_unlock(&drbd_main_mutex);
1911 
1912 	return rv;
1913 }
1914 
1915 static void drbd_release(struct gendisk *gd)
1916 {
1917 	struct drbd_device *device = gd->private_data;
1918 
1919 	mutex_lock(&drbd_main_mutex);
1920 	device->open_cnt--;
1921 	mutex_unlock(&drbd_main_mutex);
1922 }
1923 
1924 /* need to hold resource->req_lock */
1925 void drbd_queue_unplug(struct drbd_device *device)
1926 {
1927 	if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1928 		D_ASSERT(device, device->state.role == R_PRIMARY);
1929 		if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1930 			drbd_queue_work_if_unqueued(
1931 				&first_peer_device(device)->connection->sender_work,
1932 				&device->unplug_work);
1933 		}
1934 	}
1935 }
1936 
1937 static void drbd_set_defaults(struct drbd_device *device)
1938 {
1939 	/* Beware! The actual layout differs
1940 	 * between big endian and little endian */
1941 	device->state = (union drbd_dev_state) {
1942 		{ .role = R_SECONDARY,
1943 		  .peer = R_UNKNOWN,
1944 		  .conn = C_STANDALONE,
1945 		  .disk = D_DISKLESS,
1946 		  .pdsk = D_UNKNOWN,
1947 		} };
1948 }
1949 
1950 void drbd_init_set_defaults(struct drbd_device *device)
1951 {
1952 	/* the memset(,0,) did most of this.
1953 	 * note: only assignments, no allocation in here */
1954 
1955 	drbd_set_defaults(device);
1956 
1957 	atomic_set(&device->ap_bio_cnt, 0);
1958 	atomic_set(&device->ap_actlog_cnt, 0);
1959 	atomic_set(&device->ap_pending_cnt, 0);
1960 	atomic_set(&device->rs_pending_cnt, 0);
1961 	atomic_set(&device->unacked_cnt, 0);
1962 	atomic_set(&device->local_cnt, 0);
1963 	atomic_set(&device->pp_in_use_by_net, 0);
1964 	atomic_set(&device->rs_sect_in, 0);
1965 	atomic_set(&device->rs_sect_ev, 0);
1966 	atomic_set(&device->ap_in_flight, 0);
1967 	atomic_set(&device->md_io.in_use, 0);
1968 
1969 	mutex_init(&device->own_state_mutex);
1970 	device->state_mutex = &device->own_state_mutex;
1971 
1972 	spin_lock_init(&device->al_lock);
1973 	spin_lock_init(&device->peer_seq_lock);
1974 
1975 	INIT_LIST_HEAD(&device->active_ee);
1976 	INIT_LIST_HEAD(&device->sync_ee);
1977 	INIT_LIST_HEAD(&device->done_ee);
1978 	INIT_LIST_HEAD(&device->read_ee);
1979 	INIT_LIST_HEAD(&device->net_ee);
1980 	INIT_LIST_HEAD(&device->resync_reads);
1981 	INIT_LIST_HEAD(&device->resync_work.list);
1982 	INIT_LIST_HEAD(&device->unplug_work.list);
1983 	INIT_LIST_HEAD(&device->bm_io_work.w.list);
1984 	INIT_LIST_HEAD(&device->pending_master_completion[0]);
1985 	INIT_LIST_HEAD(&device->pending_master_completion[1]);
1986 	INIT_LIST_HEAD(&device->pending_completion[0]);
1987 	INIT_LIST_HEAD(&device->pending_completion[1]);
1988 
1989 	device->resync_work.cb  = w_resync_timer;
1990 	device->unplug_work.cb  = w_send_write_hint;
1991 	device->bm_io_work.w.cb = w_bitmap_io;
1992 
1993 	timer_setup(&device->resync_timer, resync_timer_fn, 0);
1994 	timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0);
1995 	timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0);
1996 	timer_setup(&device->request_timer, request_timer_fn, 0);
1997 
1998 	init_waitqueue_head(&device->misc_wait);
1999 	init_waitqueue_head(&device->state_wait);
2000 	init_waitqueue_head(&device->ee_wait);
2001 	init_waitqueue_head(&device->al_wait);
2002 	init_waitqueue_head(&device->seq_wait);
2003 
2004 	device->resync_wenr = LC_FREE;
2005 	device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2006 	device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2007 }
2008 
2009 void drbd_set_my_capacity(struct drbd_device *device, sector_t size)
2010 {
2011 	char ppb[10];
2012 
2013 	set_capacity_and_notify(device->vdisk, size);
2014 
2015 	drbd_info(device, "size = %s (%llu KB)\n",
2016 		ppsize(ppb, size>>1), (unsigned long long)size>>1);
2017 }
2018 
2019 void drbd_device_cleanup(struct drbd_device *device)
2020 {
2021 	int i;
2022 	if (first_peer_device(device)->connection->receiver.t_state != NONE)
2023 		drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2024 				first_peer_device(device)->connection->receiver.t_state);
2025 
2026 	device->al_writ_cnt  =
2027 	device->bm_writ_cnt  =
2028 	device->read_cnt     =
2029 	device->recv_cnt     =
2030 	device->send_cnt     =
2031 	device->writ_cnt     =
2032 	device->p_size       =
2033 	device->rs_start     =
2034 	device->rs_total     =
2035 	device->rs_failed    = 0;
2036 	device->rs_last_events = 0;
2037 	device->rs_last_sect_ev = 0;
2038 	for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2039 		device->rs_mark_left[i] = 0;
2040 		device->rs_mark_time[i] = 0;
2041 	}
2042 	D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2043 
2044 	set_capacity_and_notify(device->vdisk, 0);
2045 	if (device->bitmap) {
2046 		/* maybe never allocated. */
2047 		drbd_bm_resize(device, 0, 1);
2048 		drbd_bm_cleanup(device);
2049 	}
2050 
2051 	drbd_backing_dev_free(device, device->ldev);
2052 	device->ldev = NULL;
2053 
2054 	clear_bit(AL_SUSPENDED, &device->flags);
2055 
2056 	D_ASSERT(device, list_empty(&device->active_ee));
2057 	D_ASSERT(device, list_empty(&device->sync_ee));
2058 	D_ASSERT(device, list_empty(&device->done_ee));
2059 	D_ASSERT(device, list_empty(&device->read_ee));
2060 	D_ASSERT(device, list_empty(&device->net_ee));
2061 	D_ASSERT(device, list_empty(&device->resync_reads));
2062 	D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2063 	D_ASSERT(device, list_empty(&device->resync_work.list));
2064 	D_ASSERT(device, list_empty(&device->unplug_work.list));
2065 
2066 	drbd_set_defaults(device);
2067 }
2068 
2069 
2070 static void drbd_destroy_mempools(void)
2071 {
2072 	struct page *page;
2073 
2074 	while (drbd_pp_pool) {
2075 		page = drbd_pp_pool;
2076 		drbd_pp_pool = (struct page *)page_private(page);
2077 		__free_page(page);
2078 		drbd_pp_vacant--;
2079 	}
2080 
2081 	/* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2082 
2083 	bioset_exit(&drbd_io_bio_set);
2084 	bioset_exit(&drbd_md_io_bio_set);
2085 	mempool_exit(&drbd_md_io_page_pool);
2086 	mempool_exit(&drbd_ee_mempool);
2087 	mempool_exit(&drbd_request_mempool);
2088 	kmem_cache_destroy(drbd_ee_cache);
2089 	kmem_cache_destroy(drbd_request_cache);
2090 	kmem_cache_destroy(drbd_bm_ext_cache);
2091 	kmem_cache_destroy(drbd_al_ext_cache);
2092 
2093 	drbd_ee_cache        = NULL;
2094 	drbd_request_cache   = NULL;
2095 	drbd_bm_ext_cache    = NULL;
2096 	drbd_al_ext_cache    = NULL;
2097 
2098 	return;
2099 }
2100 
2101 static int drbd_create_mempools(void)
2102 {
2103 	struct page *page;
2104 	const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2105 	int i, ret;
2106 
2107 	/* caches */
2108 	drbd_request_cache = kmem_cache_create(
2109 		"drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2110 	if (drbd_request_cache == NULL)
2111 		goto Enomem;
2112 
2113 	drbd_ee_cache = kmem_cache_create(
2114 		"drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2115 	if (drbd_ee_cache == NULL)
2116 		goto Enomem;
2117 
2118 	drbd_bm_ext_cache = kmem_cache_create(
2119 		"drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2120 	if (drbd_bm_ext_cache == NULL)
2121 		goto Enomem;
2122 
2123 	drbd_al_ext_cache = kmem_cache_create(
2124 		"drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2125 	if (drbd_al_ext_cache == NULL)
2126 		goto Enomem;
2127 
2128 	/* mempools */
2129 	ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0);
2130 	if (ret)
2131 		goto Enomem;
2132 
2133 	ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0,
2134 			  BIOSET_NEED_BVECS);
2135 	if (ret)
2136 		goto Enomem;
2137 
2138 	ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0);
2139 	if (ret)
2140 		goto Enomem;
2141 
2142 	ret = mempool_init_slab_pool(&drbd_request_mempool, number,
2143 				     drbd_request_cache);
2144 	if (ret)
2145 		goto Enomem;
2146 
2147 	ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache);
2148 	if (ret)
2149 		goto Enomem;
2150 
2151 	for (i = 0; i < number; i++) {
2152 		page = alloc_page(GFP_HIGHUSER);
2153 		if (!page)
2154 			goto Enomem;
2155 		set_page_private(page, (unsigned long)drbd_pp_pool);
2156 		drbd_pp_pool = page;
2157 	}
2158 	drbd_pp_vacant = number;
2159 
2160 	return 0;
2161 
2162 Enomem:
2163 	drbd_destroy_mempools(); /* in case we allocated some */
2164 	return -ENOMEM;
2165 }
2166 
2167 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2168 {
2169 	int rr;
2170 
2171 	rr = drbd_free_peer_reqs(device, &device->active_ee);
2172 	if (rr)
2173 		drbd_err(device, "%d EEs in active list found!\n", rr);
2174 
2175 	rr = drbd_free_peer_reqs(device, &device->sync_ee);
2176 	if (rr)
2177 		drbd_err(device, "%d EEs in sync list found!\n", rr);
2178 
2179 	rr = drbd_free_peer_reqs(device, &device->read_ee);
2180 	if (rr)
2181 		drbd_err(device, "%d EEs in read list found!\n", rr);
2182 
2183 	rr = drbd_free_peer_reqs(device, &device->done_ee);
2184 	if (rr)
2185 		drbd_err(device, "%d EEs in done list found!\n", rr);
2186 
2187 	rr = drbd_free_peer_reqs(device, &device->net_ee);
2188 	if (rr)
2189 		drbd_err(device, "%d EEs in net list found!\n", rr);
2190 }
2191 
2192 /* caution. no locking. */
2193 void drbd_destroy_device(struct kref *kref)
2194 {
2195 	struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2196 	struct drbd_resource *resource = device->resource;
2197 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2198 
2199 	timer_shutdown_sync(&device->request_timer);
2200 
2201 	/* paranoia asserts */
2202 	D_ASSERT(device, device->open_cnt == 0);
2203 	/* end paranoia asserts */
2204 
2205 	/* cleanup stuff that may have been allocated during
2206 	 * device (re-)configuration or state changes */
2207 
2208 	drbd_backing_dev_free(device, device->ldev);
2209 	device->ldev = NULL;
2210 
2211 	drbd_release_all_peer_reqs(device);
2212 
2213 	lc_destroy(device->act_log);
2214 	lc_destroy(device->resync);
2215 
2216 	kfree(device->p_uuid);
2217 	/* device->p_uuid = NULL; */
2218 
2219 	if (device->bitmap) /* should no longer be there. */
2220 		drbd_bm_cleanup(device);
2221 	__free_page(device->md_io.page);
2222 	put_disk(device->vdisk);
2223 	kfree(device->rs_plan_s);
2224 
2225 	/* not for_each_connection(connection, resource):
2226 	 * those may have been cleaned up and disassociated already.
2227 	 */
2228 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2229 		kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2230 		kfree(peer_device);
2231 	}
2232 	if (device->submit.wq)
2233 		destroy_workqueue(device->submit.wq);
2234 	kfree(device);
2235 	kref_put(&resource->kref, drbd_destroy_resource);
2236 }
2237 
2238 /* One global retry thread, if we need to push back some bio and have it
2239  * reinserted through our make request function.
2240  */
2241 static struct retry_worker {
2242 	struct workqueue_struct *wq;
2243 	struct work_struct worker;
2244 
2245 	spinlock_t lock;
2246 	struct list_head writes;
2247 } retry;
2248 
2249 static void do_retry(struct work_struct *ws)
2250 {
2251 	struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2252 	LIST_HEAD(writes);
2253 	struct drbd_request *req, *tmp;
2254 
2255 	spin_lock_irq(&retry->lock);
2256 	list_splice_init(&retry->writes, &writes);
2257 	spin_unlock_irq(&retry->lock);
2258 
2259 	list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2260 		struct drbd_device *device = req->device;
2261 		struct bio *bio = req->master_bio;
2262 		bool expected;
2263 
2264 		expected =
2265 			expect(device, atomic_read(&req->completion_ref) == 0) &&
2266 			expect(device, req->rq_state & RQ_POSTPONED) &&
2267 			expect(device, (req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2268 				(req->rq_state & RQ_LOCAL_ABORTED) != 0);
2269 
2270 		if (!expected)
2271 			drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2272 				req, atomic_read(&req->completion_ref),
2273 				req->rq_state);
2274 
2275 		/* We still need to put one kref associated with the
2276 		 * "completion_ref" going zero in the code path that queued it
2277 		 * here.  The request object may still be referenced by a
2278 		 * frozen local req->private_bio, in case we force-detached.
2279 		 */
2280 		kref_put(&req->kref, drbd_req_destroy);
2281 
2282 		/* A single suspended or otherwise blocking device may stall
2283 		 * all others as well.  Fortunately, this code path is to
2284 		 * recover from a situation that "should not happen":
2285 		 * concurrent writes in multi-primary setup.
2286 		 * In a "normal" lifecycle, this workqueue is supposed to be
2287 		 * destroyed without ever doing anything.
2288 		 * If it turns out to be an issue anyways, we can do per
2289 		 * resource (replication group) or per device (minor) retry
2290 		 * workqueues instead.
2291 		 */
2292 
2293 		/* We are not just doing submit_bio_noacct(),
2294 		 * as we want to keep the start_time information. */
2295 		inc_ap_bio(device);
2296 		__drbd_make_request(device, bio);
2297 	}
2298 }
2299 
2300 /* called via drbd_req_put_completion_ref(),
2301  * holds resource->req_lock */
2302 void drbd_restart_request(struct drbd_request *req)
2303 {
2304 	unsigned long flags;
2305 	spin_lock_irqsave(&retry.lock, flags);
2306 	list_move_tail(&req->tl_requests, &retry.writes);
2307 	spin_unlock_irqrestore(&retry.lock, flags);
2308 
2309 	/* Drop the extra reference that would otherwise
2310 	 * have been dropped by complete_master_bio.
2311 	 * do_retry() needs to grab a new one. */
2312 	dec_ap_bio(req->device);
2313 
2314 	queue_work(retry.wq, &retry.worker);
2315 }
2316 
2317 void drbd_destroy_resource(struct kref *kref)
2318 {
2319 	struct drbd_resource *resource =
2320 		container_of(kref, struct drbd_resource, kref);
2321 
2322 	idr_destroy(&resource->devices);
2323 	free_cpumask_var(resource->cpu_mask);
2324 	kfree(resource->name);
2325 	kfree(resource);
2326 }
2327 
2328 void drbd_free_resource(struct drbd_resource *resource)
2329 {
2330 	struct drbd_connection *connection, *tmp;
2331 
2332 	for_each_connection_safe(connection, tmp, resource) {
2333 		list_del(&connection->connections);
2334 		drbd_debugfs_connection_cleanup(connection);
2335 		kref_put(&connection->kref, drbd_destroy_connection);
2336 	}
2337 	drbd_debugfs_resource_cleanup(resource);
2338 	kref_put(&resource->kref, drbd_destroy_resource);
2339 }
2340 
2341 static void drbd_cleanup(void)
2342 {
2343 	unsigned int i;
2344 	struct drbd_device *device;
2345 	struct drbd_resource *resource, *tmp;
2346 
2347 	/* first remove proc,
2348 	 * drbdsetup uses it's presence to detect
2349 	 * whether DRBD is loaded.
2350 	 * If we would get stuck in proc removal,
2351 	 * but have netlink already deregistered,
2352 	 * some drbdsetup commands may wait forever
2353 	 * for an answer.
2354 	 */
2355 	if (drbd_proc)
2356 		remove_proc_entry("drbd", NULL);
2357 
2358 	if (retry.wq)
2359 		destroy_workqueue(retry.wq);
2360 
2361 	drbd_genl_unregister();
2362 
2363 	idr_for_each_entry(&drbd_devices, device, i)
2364 		drbd_delete_device(device);
2365 
2366 	/* not _rcu since, no other updater anymore. Genl already unregistered */
2367 	for_each_resource_safe(resource, tmp, &drbd_resources) {
2368 		list_del(&resource->resources);
2369 		drbd_free_resource(resource);
2370 	}
2371 
2372 	drbd_debugfs_cleanup();
2373 
2374 	drbd_destroy_mempools();
2375 	unregister_blkdev(DRBD_MAJOR, "drbd");
2376 
2377 	idr_destroy(&drbd_devices);
2378 
2379 	pr_info("module cleanup done.\n");
2380 }
2381 
2382 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2383 {
2384 	spin_lock_init(&wq->q_lock);
2385 	INIT_LIST_HEAD(&wq->q);
2386 	init_waitqueue_head(&wq->q_wait);
2387 }
2388 
2389 struct completion_work {
2390 	struct drbd_work w;
2391 	struct completion done;
2392 };
2393 
2394 static int w_complete(struct drbd_work *w, int cancel)
2395 {
2396 	struct completion_work *completion_work =
2397 		container_of(w, struct completion_work, w);
2398 
2399 	complete(&completion_work->done);
2400 	return 0;
2401 }
2402 
2403 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2404 {
2405 	struct completion_work completion_work;
2406 
2407 	completion_work.w.cb = w_complete;
2408 	init_completion(&completion_work.done);
2409 	drbd_queue_work(work_queue, &completion_work.w);
2410 	wait_for_completion(&completion_work.done);
2411 }
2412 
2413 struct drbd_resource *drbd_find_resource(const char *name)
2414 {
2415 	struct drbd_resource *resource;
2416 
2417 	if (!name || !name[0])
2418 		return NULL;
2419 
2420 	rcu_read_lock();
2421 	for_each_resource_rcu(resource, &drbd_resources) {
2422 		if (!strcmp(resource->name, name)) {
2423 			kref_get(&resource->kref);
2424 			goto found;
2425 		}
2426 	}
2427 	resource = NULL;
2428 found:
2429 	rcu_read_unlock();
2430 	return resource;
2431 }
2432 
2433 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2434 				     void *peer_addr, int peer_addr_len)
2435 {
2436 	struct drbd_resource *resource;
2437 	struct drbd_connection *connection;
2438 
2439 	rcu_read_lock();
2440 	for_each_resource_rcu(resource, &drbd_resources) {
2441 		for_each_connection_rcu(connection, resource) {
2442 			if (connection->my_addr_len == my_addr_len &&
2443 			    connection->peer_addr_len == peer_addr_len &&
2444 			    !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2445 			    !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2446 				kref_get(&connection->kref);
2447 				goto found;
2448 			}
2449 		}
2450 	}
2451 	connection = NULL;
2452 found:
2453 	rcu_read_unlock();
2454 	return connection;
2455 }
2456 
2457 static int drbd_alloc_socket(struct drbd_socket *socket)
2458 {
2459 	socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2460 	if (!socket->rbuf)
2461 		return -ENOMEM;
2462 	socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2463 	if (!socket->sbuf)
2464 		return -ENOMEM;
2465 	return 0;
2466 }
2467 
2468 static void drbd_free_socket(struct drbd_socket *socket)
2469 {
2470 	free_page((unsigned long) socket->sbuf);
2471 	free_page((unsigned long) socket->rbuf);
2472 }
2473 
2474 void conn_free_crypto(struct drbd_connection *connection)
2475 {
2476 	drbd_free_sock(connection);
2477 
2478 	crypto_free_shash(connection->csums_tfm);
2479 	crypto_free_shash(connection->verify_tfm);
2480 	crypto_free_shash(connection->cram_hmac_tfm);
2481 	crypto_free_shash(connection->integrity_tfm);
2482 	crypto_free_shash(connection->peer_integrity_tfm);
2483 	kfree(connection->int_dig_in);
2484 	kfree(connection->int_dig_vv);
2485 
2486 	connection->csums_tfm = NULL;
2487 	connection->verify_tfm = NULL;
2488 	connection->cram_hmac_tfm = NULL;
2489 	connection->integrity_tfm = NULL;
2490 	connection->peer_integrity_tfm = NULL;
2491 	connection->int_dig_in = NULL;
2492 	connection->int_dig_vv = NULL;
2493 }
2494 
2495 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2496 {
2497 	struct drbd_connection *connection;
2498 	cpumask_var_t new_cpu_mask;
2499 	int err;
2500 
2501 	if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2502 		return -ENOMEM;
2503 
2504 	/* silently ignore cpu mask on UP kernel */
2505 	if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2506 		err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2507 				   cpumask_bits(new_cpu_mask), nr_cpu_ids);
2508 		if (err == -EOVERFLOW) {
2509 			/* So what. mask it out. */
2510 			cpumask_var_t tmp_cpu_mask;
2511 			if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2512 				cpumask_setall(tmp_cpu_mask);
2513 				cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2514 				drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2515 					res_opts->cpu_mask,
2516 					strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2517 					nr_cpu_ids);
2518 				free_cpumask_var(tmp_cpu_mask);
2519 				err = 0;
2520 			}
2521 		}
2522 		if (err) {
2523 			drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2524 			/* retcode = ERR_CPU_MASK_PARSE; */
2525 			goto fail;
2526 		}
2527 	}
2528 	resource->res_opts = *res_opts;
2529 	if (cpumask_empty(new_cpu_mask))
2530 		drbd_calc_cpu_mask(&new_cpu_mask);
2531 	if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2532 		cpumask_copy(resource->cpu_mask, new_cpu_mask);
2533 		for_each_connection_rcu(connection, resource) {
2534 			connection->receiver.reset_cpu_mask = 1;
2535 			connection->ack_receiver.reset_cpu_mask = 1;
2536 			connection->worker.reset_cpu_mask = 1;
2537 		}
2538 	}
2539 	err = 0;
2540 
2541 fail:
2542 	free_cpumask_var(new_cpu_mask);
2543 	return err;
2544 
2545 }
2546 
2547 struct drbd_resource *drbd_create_resource(const char *name)
2548 {
2549 	struct drbd_resource *resource;
2550 
2551 	resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2552 	if (!resource)
2553 		goto fail;
2554 	resource->name = kstrdup(name, GFP_KERNEL);
2555 	if (!resource->name)
2556 		goto fail_free_resource;
2557 	if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2558 		goto fail_free_name;
2559 	kref_init(&resource->kref);
2560 	idr_init(&resource->devices);
2561 	INIT_LIST_HEAD(&resource->connections);
2562 	resource->write_ordering = WO_BDEV_FLUSH;
2563 	list_add_tail_rcu(&resource->resources, &drbd_resources);
2564 	mutex_init(&resource->conf_update);
2565 	mutex_init(&resource->adm_mutex);
2566 	spin_lock_init(&resource->req_lock);
2567 	drbd_debugfs_resource_add(resource);
2568 	return resource;
2569 
2570 fail_free_name:
2571 	kfree(resource->name);
2572 fail_free_resource:
2573 	kfree(resource);
2574 fail:
2575 	return NULL;
2576 }
2577 
2578 /* caller must be under adm_mutex */
2579 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2580 {
2581 	struct drbd_resource *resource;
2582 	struct drbd_connection *connection;
2583 
2584 	connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2585 	if (!connection)
2586 		return NULL;
2587 
2588 	if (drbd_alloc_socket(&connection->data))
2589 		goto fail;
2590 	if (drbd_alloc_socket(&connection->meta))
2591 		goto fail;
2592 
2593 	connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2594 	if (!connection->current_epoch)
2595 		goto fail;
2596 
2597 	INIT_LIST_HEAD(&connection->transfer_log);
2598 
2599 	INIT_LIST_HEAD(&connection->current_epoch->list);
2600 	connection->epochs = 1;
2601 	spin_lock_init(&connection->epoch_lock);
2602 
2603 	connection->send.seen_any_write_yet = false;
2604 	connection->send.current_epoch_nr = 0;
2605 	connection->send.current_epoch_writes = 0;
2606 
2607 	resource = drbd_create_resource(name);
2608 	if (!resource)
2609 		goto fail;
2610 
2611 	connection->cstate = C_STANDALONE;
2612 	mutex_init(&connection->cstate_mutex);
2613 	init_waitqueue_head(&connection->ping_wait);
2614 	idr_init(&connection->peer_devices);
2615 
2616 	drbd_init_workqueue(&connection->sender_work);
2617 	mutex_init(&connection->data.mutex);
2618 	mutex_init(&connection->meta.mutex);
2619 
2620 	drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2621 	connection->receiver.connection = connection;
2622 	drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2623 	connection->worker.connection = connection;
2624 	drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2625 	connection->ack_receiver.connection = connection;
2626 
2627 	kref_init(&connection->kref);
2628 
2629 	connection->resource = resource;
2630 
2631 	if (set_resource_options(resource, res_opts))
2632 		goto fail_resource;
2633 
2634 	kref_get(&resource->kref);
2635 	list_add_tail_rcu(&connection->connections, &resource->connections);
2636 	drbd_debugfs_connection_add(connection);
2637 	return connection;
2638 
2639 fail_resource:
2640 	list_del(&resource->resources);
2641 	drbd_free_resource(resource);
2642 fail:
2643 	kfree(connection->current_epoch);
2644 	drbd_free_socket(&connection->meta);
2645 	drbd_free_socket(&connection->data);
2646 	kfree(connection);
2647 	return NULL;
2648 }
2649 
2650 void drbd_destroy_connection(struct kref *kref)
2651 {
2652 	struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2653 	struct drbd_resource *resource = connection->resource;
2654 
2655 	if (atomic_read(&connection->current_epoch->epoch_size) !=  0)
2656 		drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2657 	kfree(connection->current_epoch);
2658 
2659 	idr_destroy(&connection->peer_devices);
2660 
2661 	drbd_free_socket(&connection->meta);
2662 	drbd_free_socket(&connection->data);
2663 	kfree(connection->int_dig_in);
2664 	kfree(connection->int_dig_vv);
2665 	kfree(connection);
2666 	kref_put(&resource->kref, drbd_destroy_resource);
2667 }
2668 
2669 static int init_submitter(struct drbd_device *device)
2670 {
2671 	/* opencoded create_singlethread_workqueue(),
2672 	 * to be able to say "drbd%d", ..., minor */
2673 	device->submit.wq =
2674 		alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2675 	if (!device->submit.wq)
2676 		return -ENOMEM;
2677 
2678 	INIT_WORK(&device->submit.worker, do_submit);
2679 	INIT_LIST_HEAD(&device->submit.writes);
2680 	return 0;
2681 }
2682 
2683 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2684 {
2685 	struct drbd_resource *resource = adm_ctx->resource;
2686 	struct drbd_connection *connection, *n;
2687 	struct drbd_device *device;
2688 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2689 	struct gendisk *disk;
2690 	int id;
2691 	int vnr = adm_ctx->volume;
2692 	enum drbd_ret_code err = ERR_NOMEM;
2693 
2694 	device = minor_to_device(minor);
2695 	if (device)
2696 		return ERR_MINOR_OR_VOLUME_EXISTS;
2697 
2698 	/* GFP_KERNEL, we are outside of all write-out paths */
2699 	device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2700 	if (!device)
2701 		return ERR_NOMEM;
2702 	kref_init(&device->kref);
2703 
2704 	kref_get(&resource->kref);
2705 	device->resource = resource;
2706 	device->minor = minor;
2707 	device->vnr = vnr;
2708 
2709 	drbd_init_set_defaults(device);
2710 
2711 	disk = blk_alloc_disk(NUMA_NO_NODE);
2712 	if (!disk)
2713 		goto out_no_disk;
2714 
2715 	device->vdisk = disk;
2716 	device->rq_queue = disk->queue;
2717 
2718 	set_disk_ro(disk, true);
2719 
2720 	disk->major = DRBD_MAJOR;
2721 	disk->first_minor = minor;
2722 	disk->minors = 1;
2723 	disk->fops = &drbd_ops;
2724 	disk->flags |= GENHD_FL_NO_PART;
2725 	sprintf(disk->disk_name, "drbd%d", minor);
2726 	disk->private_data = device;
2727 
2728 	blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, disk->queue);
2729 	blk_queue_write_cache(disk->queue, true, true);
2730 	/* Setting the max_hw_sectors to an odd value of 8kibyte here
2731 	   This triggers a max_bio_size message upon first attach or connect */
2732 	blk_queue_max_hw_sectors(disk->queue, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2733 
2734 	device->md_io.page = alloc_page(GFP_KERNEL);
2735 	if (!device->md_io.page)
2736 		goto out_no_io_page;
2737 
2738 	if (drbd_bm_init(device))
2739 		goto out_no_bitmap;
2740 	device->read_requests = RB_ROOT;
2741 	device->write_requests = RB_ROOT;
2742 
2743 	id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2744 	if (id < 0) {
2745 		if (id == -ENOSPC)
2746 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2747 		goto out_no_minor_idr;
2748 	}
2749 	kref_get(&device->kref);
2750 
2751 	id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2752 	if (id < 0) {
2753 		if (id == -ENOSPC)
2754 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2755 		goto out_idr_remove_minor;
2756 	}
2757 	kref_get(&device->kref);
2758 
2759 	INIT_LIST_HEAD(&device->peer_devices);
2760 	INIT_LIST_HEAD(&device->pending_bitmap_io);
2761 	for_each_connection(connection, resource) {
2762 		peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2763 		if (!peer_device)
2764 			goto out_idr_remove_from_resource;
2765 		peer_device->connection = connection;
2766 		peer_device->device = device;
2767 
2768 		list_add(&peer_device->peer_devices, &device->peer_devices);
2769 		kref_get(&device->kref);
2770 
2771 		id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2772 		if (id < 0) {
2773 			if (id == -ENOSPC)
2774 				err = ERR_INVALID_REQUEST;
2775 			goto out_idr_remove_from_resource;
2776 		}
2777 		kref_get(&connection->kref);
2778 		INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2779 	}
2780 
2781 	if (init_submitter(device)) {
2782 		err = ERR_NOMEM;
2783 		goto out_idr_remove_from_resource;
2784 	}
2785 
2786 	err = add_disk(disk);
2787 	if (err)
2788 		goto out_destroy_workqueue;
2789 
2790 	/* inherit the connection state */
2791 	device->state.conn = first_connection(resource)->cstate;
2792 	if (device->state.conn == C_WF_REPORT_PARAMS) {
2793 		for_each_peer_device(peer_device, device)
2794 			drbd_connected(peer_device);
2795 	}
2796 	/* move to create_peer_device() */
2797 	for_each_peer_device(peer_device, device)
2798 		drbd_debugfs_peer_device_add(peer_device);
2799 	drbd_debugfs_device_add(device);
2800 	return NO_ERROR;
2801 
2802 out_destroy_workqueue:
2803 	destroy_workqueue(device->submit.wq);
2804 out_idr_remove_from_resource:
2805 	for_each_connection_safe(connection, n, resource) {
2806 		peer_device = idr_remove(&connection->peer_devices, vnr);
2807 		if (peer_device)
2808 			kref_put(&connection->kref, drbd_destroy_connection);
2809 	}
2810 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2811 		list_del(&peer_device->peer_devices);
2812 		kfree(peer_device);
2813 	}
2814 	idr_remove(&resource->devices, vnr);
2815 out_idr_remove_minor:
2816 	idr_remove(&drbd_devices, minor);
2817 	synchronize_rcu();
2818 out_no_minor_idr:
2819 	drbd_bm_cleanup(device);
2820 out_no_bitmap:
2821 	__free_page(device->md_io.page);
2822 out_no_io_page:
2823 	put_disk(disk);
2824 out_no_disk:
2825 	kref_put(&resource->kref, drbd_destroy_resource);
2826 	kfree(device);
2827 	return err;
2828 }
2829 
2830 void drbd_delete_device(struct drbd_device *device)
2831 {
2832 	struct drbd_resource *resource = device->resource;
2833 	struct drbd_connection *connection;
2834 	struct drbd_peer_device *peer_device;
2835 
2836 	/* move to free_peer_device() */
2837 	for_each_peer_device(peer_device, device)
2838 		drbd_debugfs_peer_device_cleanup(peer_device);
2839 	drbd_debugfs_device_cleanup(device);
2840 	for_each_connection(connection, resource) {
2841 		idr_remove(&connection->peer_devices, device->vnr);
2842 		kref_put(&device->kref, drbd_destroy_device);
2843 	}
2844 	idr_remove(&resource->devices, device->vnr);
2845 	kref_put(&device->kref, drbd_destroy_device);
2846 	idr_remove(&drbd_devices, device_to_minor(device));
2847 	kref_put(&device->kref, drbd_destroy_device);
2848 	del_gendisk(device->vdisk);
2849 	synchronize_rcu();
2850 	kref_put(&device->kref, drbd_destroy_device);
2851 }
2852 
2853 static int __init drbd_init(void)
2854 {
2855 	int err;
2856 
2857 	if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2858 		pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2859 #ifdef MODULE
2860 		return -EINVAL;
2861 #else
2862 		drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2863 #endif
2864 	}
2865 
2866 	err = register_blkdev(DRBD_MAJOR, "drbd");
2867 	if (err) {
2868 		pr_err("unable to register block device major %d\n",
2869 		       DRBD_MAJOR);
2870 		return err;
2871 	}
2872 
2873 	/*
2874 	 * allocate all necessary structs
2875 	 */
2876 	init_waitqueue_head(&drbd_pp_wait);
2877 
2878 	drbd_proc = NULL; /* play safe for drbd_cleanup */
2879 	idr_init(&drbd_devices);
2880 
2881 	mutex_init(&resources_mutex);
2882 	INIT_LIST_HEAD(&drbd_resources);
2883 
2884 	err = drbd_genl_register();
2885 	if (err) {
2886 		pr_err("unable to register generic netlink family\n");
2887 		goto fail;
2888 	}
2889 
2890 	err = drbd_create_mempools();
2891 	if (err)
2892 		goto fail;
2893 
2894 	err = -ENOMEM;
2895 	drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show);
2896 	if (!drbd_proc)	{
2897 		pr_err("unable to register proc file\n");
2898 		goto fail;
2899 	}
2900 
2901 	retry.wq = create_singlethread_workqueue("drbd-reissue");
2902 	if (!retry.wq) {
2903 		pr_err("unable to create retry workqueue\n");
2904 		goto fail;
2905 	}
2906 	INIT_WORK(&retry.worker, do_retry);
2907 	spin_lock_init(&retry.lock);
2908 	INIT_LIST_HEAD(&retry.writes);
2909 
2910 	drbd_debugfs_init();
2911 
2912 	pr_info("initialized. "
2913 	       "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2914 	       GENL_MAGIC_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2915 	pr_info("%s\n", drbd_buildtag());
2916 	pr_info("registered as block device major %d\n", DRBD_MAJOR);
2917 	return 0; /* Success! */
2918 
2919 fail:
2920 	drbd_cleanup();
2921 	if (err == -ENOMEM)
2922 		pr_err("ran out of memory\n");
2923 	else
2924 		pr_err("initialization failure\n");
2925 	return err;
2926 }
2927 
2928 static void drbd_free_one_sock(struct drbd_socket *ds)
2929 {
2930 	struct socket *s;
2931 	mutex_lock(&ds->mutex);
2932 	s = ds->socket;
2933 	ds->socket = NULL;
2934 	mutex_unlock(&ds->mutex);
2935 	if (s) {
2936 		/* so debugfs does not need to mutex_lock() */
2937 		synchronize_rcu();
2938 		kernel_sock_shutdown(s, SHUT_RDWR);
2939 		sock_release(s);
2940 	}
2941 }
2942 
2943 void drbd_free_sock(struct drbd_connection *connection)
2944 {
2945 	if (connection->data.socket)
2946 		drbd_free_one_sock(&connection->data);
2947 	if (connection->meta.socket)
2948 		drbd_free_one_sock(&connection->meta);
2949 }
2950 
2951 /* meta data management */
2952 
2953 void conn_md_sync(struct drbd_connection *connection)
2954 {
2955 	struct drbd_peer_device *peer_device;
2956 	int vnr;
2957 
2958 	rcu_read_lock();
2959 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2960 		struct drbd_device *device = peer_device->device;
2961 
2962 		kref_get(&device->kref);
2963 		rcu_read_unlock();
2964 		drbd_md_sync(device);
2965 		kref_put(&device->kref, drbd_destroy_device);
2966 		rcu_read_lock();
2967 	}
2968 	rcu_read_unlock();
2969 }
2970 
2971 /* aligned 4kByte */
2972 struct meta_data_on_disk {
2973 	u64 la_size_sect;      /* last agreed size. */
2974 	u64 uuid[UI_SIZE];   /* UUIDs. */
2975 	u64 device_uuid;
2976 	u64 reserved_u64_1;
2977 	u32 flags;             /* MDF */
2978 	u32 magic;
2979 	u32 md_size_sect;
2980 	u32 al_offset;         /* offset to this block */
2981 	u32 al_nr_extents;     /* important for restoring the AL (userspace) */
2982 	      /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2983 	u32 bm_offset;         /* offset to the bitmap, from here */
2984 	u32 bm_bytes_per_bit;  /* BM_BLOCK_SIZE */
2985 	u32 la_peer_max_bio_size;   /* last peer max_bio_size */
2986 
2987 	/* see al_tr_number_to_on_disk_sector() */
2988 	u32 al_stripes;
2989 	u32 al_stripe_size_4k;
2990 
2991 	u8 reserved_u8[4096 - (7*8 + 10*4)];
2992 } __packed;
2993 
2994 
2995 
2996 void drbd_md_write(struct drbd_device *device, void *b)
2997 {
2998 	struct meta_data_on_disk *buffer = b;
2999 	sector_t sector;
3000 	int i;
3001 
3002 	memset(buffer, 0, sizeof(*buffer));
3003 
3004 	buffer->la_size_sect = cpu_to_be64(get_capacity(device->vdisk));
3005 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3006 		buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3007 	buffer->flags = cpu_to_be32(device->ldev->md.flags);
3008 	buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3009 
3010 	buffer->md_size_sect  = cpu_to_be32(device->ldev->md.md_size_sect);
3011 	buffer->al_offset     = cpu_to_be32(device->ldev->md.al_offset);
3012 	buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3013 	buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3014 	buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3015 
3016 	buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3017 	buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3018 
3019 	buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3020 	buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3021 
3022 	D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3023 	sector = device->ldev->md.md_offset;
3024 
3025 	if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3026 		/* this was a try anyways ... */
3027 		drbd_err(device, "meta data update failed!\n");
3028 		drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3029 	}
3030 }
3031 
3032 /**
3033  * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3034  * @device:	DRBD device.
3035  */
3036 void drbd_md_sync(struct drbd_device *device)
3037 {
3038 	struct meta_data_on_disk *buffer;
3039 
3040 	/* Don't accidentally change the DRBD meta data layout. */
3041 	BUILD_BUG_ON(UI_SIZE != 4);
3042 	BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3043 
3044 	del_timer(&device->md_sync_timer);
3045 	/* timer may be rearmed by drbd_md_mark_dirty() now. */
3046 	if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3047 		return;
3048 
3049 	/* We use here D_FAILED and not D_ATTACHING because we try to write
3050 	 * metadata even if we detach due to a disk failure! */
3051 	if (!get_ldev_if_state(device, D_FAILED))
3052 		return;
3053 
3054 	buffer = drbd_md_get_buffer(device, __func__);
3055 	if (!buffer)
3056 		goto out;
3057 
3058 	drbd_md_write(device, buffer);
3059 
3060 	/* Update device->ldev->md.la_size_sect,
3061 	 * since we updated it on metadata. */
3062 	device->ldev->md.la_size_sect = get_capacity(device->vdisk);
3063 
3064 	drbd_md_put_buffer(device);
3065 out:
3066 	put_ldev(device);
3067 }
3068 
3069 static int check_activity_log_stripe_size(struct drbd_device *device,
3070 		struct meta_data_on_disk *on_disk,
3071 		struct drbd_md *in_core)
3072 {
3073 	u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3074 	u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3075 	u64 al_size_4k;
3076 
3077 	/* both not set: default to old fixed size activity log */
3078 	if (al_stripes == 0 && al_stripe_size_4k == 0) {
3079 		al_stripes = 1;
3080 		al_stripe_size_4k = MD_32kB_SECT/8;
3081 	}
3082 
3083 	/* some paranoia plausibility checks */
3084 
3085 	/* we need both values to be set */
3086 	if (al_stripes == 0 || al_stripe_size_4k == 0)
3087 		goto err;
3088 
3089 	al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3090 
3091 	/* Upper limit of activity log area, to avoid potential overflow
3092 	 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3093 	 * than 72 * 4k blocks total only increases the amount of history,
3094 	 * limiting this arbitrarily to 16 GB is not a real limitation ;-)  */
3095 	if (al_size_4k > (16 * 1024 * 1024/4))
3096 		goto err;
3097 
3098 	/* Lower limit: we need at least 8 transaction slots (32kB)
3099 	 * to not break existing setups */
3100 	if (al_size_4k < MD_32kB_SECT/8)
3101 		goto err;
3102 
3103 	in_core->al_stripe_size_4k = al_stripe_size_4k;
3104 	in_core->al_stripes = al_stripes;
3105 	in_core->al_size_4k = al_size_4k;
3106 
3107 	return 0;
3108 err:
3109 	drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3110 			al_stripes, al_stripe_size_4k);
3111 	return -EINVAL;
3112 }
3113 
3114 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3115 {
3116 	sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3117 	struct drbd_md *in_core = &bdev->md;
3118 	s32 on_disk_al_sect;
3119 	s32 on_disk_bm_sect;
3120 
3121 	/* The on-disk size of the activity log, calculated from offsets, and
3122 	 * the size of the activity log calculated from the stripe settings,
3123 	 * should match.
3124 	 * Though we could relax this a bit: it is ok, if the striped activity log
3125 	 * fits in the available on-disk activity log size.
3126 	 * Right now, that would break how resize is implemented.
3127 	 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3128 	 * of possible unused padding space in the on disk layout. */
3129 	if (in_core->al_offset < 0) {
3130 		if (in_core->bm_offset > in_core->al_offset)
3131 			goto err;
3132 		on_disk_al_sect = -in_core->al_offset;
3133 		on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3134 	} else {
3135 		if (in_core->al_offset != MD_4kB_SECT)
3136 			goto err;
3137 		if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3138 			goto err;
3139 
3140 		on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3141 		on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3142 	}
3143 
3144 	/* old fixed size meta data is exactly that: fixed. */
3145 	if (in_core->meta_dev_idx >= 0) {
3146 		if (in_core->md_size_sect != MD_128MB_SECT
3147 		||  in_core->al_offset != MD_4kB_SECT
3148 		||  in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3149 		||  in_core->al_stripes != 1
3150 		||  in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3151 			goto err;
3152 	}
3153 
3154 	if (capacity < in_core->md_size_sect)
3155 		goto err;
3156 	if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3157 		goto err;
3158 
3159 	/* should be aligned, and at least 32k */
3160 	if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3161 		goto err;
3162 
3163 	/* should fit (for now: exactly) into the available on-disk space;
3164 	 * overflow prevention is in check_activity_log_stripe_size() above. */
3165 	if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3166 		goto err;
3167 
3168 	/* again, should be aligned */
3169 	if (in_core->bm_offset & 7)
3170 		goto err;
3171 
3172 	/* FIXME check for device grow with flex external meta data? */
3173 
3174 	/* can the available bitmap space cover the last agreed device size? */
3175 	if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3176 		goto err;
3177 
3178 	return 0;
3179 
3180 err:
3181 	drbd_err(device, "meta data offsets don't make sense: idx=%d "
3182 			"al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3183 			"md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3184 			in_core->meta_dev_idx,
3185 			in_core->al_stripes, in_core->al_stripe_size_4k,
3186 			in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3187 			(unsigned long long)in_core->la_size_sect,
3188 			(unsigned long long)capacity);
3189 
3190 	return -EINVAL;
3191 }
3192 
3193 
3194 /**
3195  * drbd_md_read() - Reads in the meta data super block
3196  * @device:	DRBD device.
3197  * @bdev:	Device from which the meta data should be read in.
3198  *
3199  * Return NO_ERROR on success, and an enum drbd_ret_code in case
3200  * something goes wrong.
3201  *
3202  * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3203  * even before @bdev is assigned to @device->ldev.
3204  */
3205 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3206 {
3207 	struct meta_data_on_disk *buffer;
3208 	u32 magic, flags;
3209 	int i, rv = NO_ERROR;
3210 
3211 	if (device->state.disk != D_DISKLESS)
3212 		return ERR_DISK_CONFIGURED;
3213 
3214 	buffer = drbd_md_get_buffer(device, __func__);
3215 	if (!buffer)
3216 		return ERR_NOMEM;
3217 
3218 	/* First, figure out where our meta data superblock is located,
3219 	 * and read it. */
3220 	bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3221 	bdev->md.md_offset = drbd_md_ss(bdev);
3222 	/* Even for (flexible or indexed) external meta data,
3223 	 * initially restrict us to the 4k superblock for now.
3224 	 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3225 	bdev->md.md_size_sect = 8;
3226 
3227 	if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3228 				 REQ_OP_READ)) {
3229 		/* NOTE: can't do normal error processing here as this is
3230 		   called BEFORE disk is attached */
3231 		drbd_err(device, "Error while reading metadata.\n");
3232 		rv = ERR_IO_MD_DISK;
3233 		goto err;
3234 	}
3235 
3236 	magic = be32_to_cpu(buffer->magic);
3237 	flags = be32_to_cpu(buffer->flags);
3238 	if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3239 	    (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3240 			/* btw: that's Activity Log clean, not "all" clean. */
3241 		drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3242 		rv = ERR_MD_UNCLEAN;
3243 		goto err;
3244 	}
3245 
3246 	rv = ERR_MD_INVALID;
3247 	if (magic != DRBD_MD_MAGIC_08) {
3248 		if (magic == DRBD_MD_MAGIC_07)
3249 			drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3250 		else
3251 			drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3252 		goto err;
3253 	}
3254 
3255 	if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3256 		drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3257 		    be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3258 		goto err;
3259 	}
3260 
3261 
3262 	/* convert to in_core endian */
3263 	bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3264 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3265 		bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3266 	bdev->md.flags = be32_to_cpu(buffer->flags);
3267 	bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3268 
3269 	bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3270 	bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3271 	bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3272 
3273 	if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3274 		goto err;
3275 	if (check_offsets_and_sizes(device, bdev))
3276 		goto err;
3277 
3278 	if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3279 		drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3280 		    be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3281 		goto err;
3282 	}
3283 	if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3284 		drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3285 		    be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3286 		goto err;
3287 	}
3288 
3289 	rv = NO_ERROR;
3290 
3291 	spin_lock_irq(&device->resource->req_lock);
3292 	if (device->state.conn < C_CONNECTED) {
3293 		unsigned int peer;
3294 		peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3295 		peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3296 		device->peer_max_bio_size = peer;
3297 	}
3298 	spin_unlock_irq(&device->resource->req_lock);
3299 
3300  err:
3301 	drbd_md_put_buffer(device);
3302 
3303 	return rv;
3304 }
3305 
3306 /**
3307  * drbd_md_mark_dirty() - Mark meta data super block as dirty
3308  * @device:	DRBD device.
3309  *
3310  * Call this function if you change anything that should be written to
3311  * the meta-data super block. This function sets MD_DIRTY, and starts a
3312  * timer that ensures that within five seconds you have to call drbd_md_sync().
3313  */
3314 void drbd_md_mark_dirty(struct drbd_device *device)
3315 {
3316 	if (!test_and_set_bit(MD_DIRTY, &device->flags))
3317 		mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3318 }
3319 
3320 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3321 {
3322 	int i;
3323 
3324 	for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3325 		device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3326 }
3327 
3328 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3329 {
3330 	if (idx == UI_CURRENT) {
3331 		if (device->state.role == R_PRIMARY)
3332 			val |= 1;
3333 		else
3334 			val &= ~((u64)1);
3335 
3336 		drbd_set_ed_uuid(device, val);
3337 	}
3338 
3339 	device->ldev->md.uuid[idx] = val;
3340 	drbd_md_mark_dirty(device);
3341 }
3342 
3343 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3344 {
3345 	unsigned long flags;
3346 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3347 	__drbd_uuid_set(device, idx, val);
3348 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3349 }
3350 
3351 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3352 {
3353 	unsigned long flags;
3354 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3355 	if (device->ldev->md.uuid[idx]) {
3356 		drbd_uuid_move_history(device);
3357 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3358 	}
3359 	__drbd_uuid_set(device, idx, val);
3360 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3361 }
3362 
3363 /**
3364  * drbd_uuid_new_current() - Creates a new current UUID
3365  * @device:	DRBD device.
3366  *
3367  * Creates a new current UUID, and rotates the old current UUID into
3368  * the bitmap slot. Causes an incremental resync upon next connect.
3369  */
3370 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3371 {
3372 	u64 val;
3373 	unsigned long long bm_uuid;
3374 
3375 	get_random_bytes(&val, sizeof(u64));
3376 
3377 	spin_lock_irq(&device->ldev->md.uuid_lock);
3378 	bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3379 
3380 	if (bm_uuid)
3381 		drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3382 
3383 	device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3384 	__drbd_uuid_set(device, UI_CURRENT, val);
3385 	spin_unlock_irq(&device->ldev->md.uuid_lock);
3386 
3387 	drbd_print_uuids(device, "new current UUID");
3388 	/* get it to stable storage _now_ */
3389 	drbd_md_sync(device);
3390 }
3391 
3392 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3393 {
3394 	unsigned long flags;
3395 	if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3396 		return;
3397 
3398 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3399 	if (val == 0) {
3400 		drbd_uuid_move_history(device);
3401 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3402 		device->ldev->md.uuid[UI_BITMAP] = 0;
3403 	} else {
3404 		unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3405 		if (bm_uuid)
3406 			drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3407 
3408 		device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3409 	}
3410 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3411 
3412 	drbd_md_mark_dirty(device);
3413 }
3414 
3415 /**
3416  * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3417  * @device:	DRBD device.
3418  *
3419  * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3420  */
3421 int drbd_bmio_set_n_write(struct drbd_device *device,
3422 			  struct drbd_peer_device *peer_device) __must_hold(local)
3423 
3424 {
3425 	int rv = -EIO;
3426 
3427 	drbd_md_set_flag(device, MDF_FULL_SYNC);
3428 	drbd_md_sync(device);
3429 	drbd_bm_set_all(device);
3430 
3431 	rv = drbd_bm_write(device, peer_device);
3432 
3433 	if (!rv) {
3434 		drbd_md_clear_flag(device, MDF_FULL_SYNC);
3435 		drbd_md_sync(device);
3436 	}
3437 
3438 	return rv;
3439 }
3440 
3441 /**
3442  * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3443  * @device:	DRBD device.
3444  *
3445  * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3446  */
3447 int drbd_bmio_clear_n_write(struct drbd_device *device,
3448 			  struct drbd_peer_device *peer_device) __must_hold(local)
3449 
3450 {
3451 	drbd_resume_al(device);
3452 	drbd_bm_clear_all(device);
3453 	return drbd_bm_write(device, peer_device);
3454 }
3455 
3456 static int w_bitmap_io(struct drbd_work *w, int unused)
3457 {
3458 	struct drbd_device *device =
3459 		container_of(w, struct drbd_device, bm_io_work.w);
3460 	struct bm_io_work *work = &device->bm_io_work;
3461 	int rv = -EIO;
3462 
3463 	if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3464 		int cnt = atomic_read(&device->ap_bio_cnt);
3465 		if (cnt)
3466 			drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3467 					cnt, work->why);
3468 	}
3469 
3470 	if (get_ldev(device)) {
3471 		drbd_bm_lock(device, work->why, work->flags);
3472 		rv = work->io_fn(device, work->peer_device);
3473 		drbd_bm_unlock(device);
3474 		put_ldev(device);
3475 	}
3476 
3477 	clear_bit_unlock(BITMAP_IO, &device->flags);
3478 	wake_up(&device->misc_wait);
3479 
3480 	if (work->done)
3481 		work->done(device, rv);
3482 
3483 	clear_bit(BITMAP_IO_QUEUED, &device->flags);
3484 	work->why = NULL;
3485 	work->flags = 0;
3486 
3487 	return 0;
3488 }
3489 
3490 /**
3491  * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3492  * @device:	DRBD device.
3493  * @io_fn:	IO callback to be called when bitmap IO is possible
3494  * @done:	callback to be called after the bitmap IO was performed
3495  * @why:	Descriptive text of the reason for doing the IO
3496  * @flags:	Bitmap flags
3497  *
3498  * While IO on the bitmap happens we freeze application IO thus we ensure
3499  * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3500  * called from worker context. It MUST NOT be used while a previous such
3501  * work is still pending!
3502  *
3503  * Its worker function encloses the call of io_fn() by get_ldev() and
3504  * put_ldev().
3505  */
3506 void drbd_queue_bitmap_io(struct drbd_device *device,
3507 			  int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
3508 			  void (*done)(struct drbd_device *, int),
3509 			  char *why, enum bm_flag flags,
3510 			  struct drbd_peer_device *peer_device)
3511 {
3512 	D_ASSERT(device, current == peer_device->connection->worker.task);
3513 
3514 	D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3515 	D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3516 	D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3517 	if (device->bm_io_work.why)
3518 		drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3519 			why, device->bm_io_work.why);
3520 
3521 	device->bm_io_work.peer_device = peer_device;
3522 	device->bm_io_work.io_fn = io_fn;
3523 	device->bm_io_work.done = done;
3524 	device->bm_io_work.why = why;
3525 	device->bm_io_work.flags = flags;
3526 
3527 	spin_lock_irq(&device->resource->req_lock);
3528 	set_bit(BITMAP_IO, &device->flags);
3529 	/* don't wait for pending application IO if the caller indicates that
3530 	 * application IO does not conflict anyways. */
3531 	if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3532 		if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3533 			drbd_queue_work(&peer_device->connection->sender_work,
3534 					&device->bm_io_work.w);
3535 	}
3536 	spin_unlock_irq(&device->resource->req_lock);
3537 }
3538 
3539 /**
3540  * drbd_bitmap_io() -  Does an IO operation on the whole bitmap
3541  * @device:	DRBD device.
3542  * @io_fn:	IO callback to be called when bitmap IO is possible
3543  * @why:	Descriptive text of the reason for doing the IO
3544  * @flags:	Bitmap flags
3545  *
3546  * freezes application IO while that the actual IO operations runs. This
3547  * functions MAY NOT be called from worker context.
3548  */
3549 int drbd_bitmap_io(struct drbd_device *device,
3550 		int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
3551 		char *why, enum bm_flag flags,
3552 		struct drbd_peer_device *peer_device)
3553 {
3554 	/* Only suspend io, if some operation is supposed to be locked out */
3555 	const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3556 	int rv;
3557 
3558 	D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3559 
3560 	if (do_suspend_io)
3561 		drbd_suspend_io(device);
3562 
3563 	drbd_bm_lock(device, why, flags);
3564 	rv = io_fn(device, peer_device);
3565 	drbd_bm_unlock(device);
3566 
3567 	if (do_suspend_io)
3568 		drbd_resume_io(device);
3569 
3570 	return rv;
3571 }
3572 
3573 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3574 {
3575 	if ((device->ldev->md.flags & flag) != flag) {
3576 		drbd_md_mark_dirty(device);
3577 		device->ldev->md.flags |= flag;
3578 	}
3579 }
3580 
3581 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3582 {
3583 	if ((device->ldev->md.flags & flag) != 0) {
3584 		drbd_md_mark_dirty(device);
3585 		device->ldev->md.flags &= ~flag;
3586 	}
3587 }
3588 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3589 {
3590 	return (bdev->md.flags & flag) != 0;
3591 }
3592 
3593 static void md_sync_timer_fn(struct timer_list *t)
3594 {
3595 	struct drbd_device *device = from_timer(device, t, md_sync_timer);
3596 	drbd_device_post_work(device, MD_SYNC);
3597 }
3598 
3599 const char *cmdname(enum drbd_packet cmd)
3600 {
3601 	/* THINK may need to become several global tables
3602 	 * when we want to support more than
3603 	 * one PRO_VERSION */
3604 	static const char *cmdnames[] = {
3605 
3606 		[P_DATA]	        = "Data",
3607 		[P_DATA_REPLY]	        = "DataReply",
3608 		[P_RS_DATA_REPLY]	= "RSDataReply",
3609 		[P_BARRIER]	        = "Barrier",
3610 		[P_BITMAP]	        = "ReportBitMap",
3611 		[P_BECOME_SYNC_TARGET]  = "BecomeSyncTarget",
3612 		[P_BECOME_SYNC_SOURCE]  = "BecomeSyncSource",
3613 		[P_UNPLUG_REMOTE]	= "UnplugRemote",
3614 		[P_DATA_REQUEST]	= "DataRequest",
3615 		[P_RS_DATA_REQUEST]     = "RSDataRequest",
3616 		[P_SYNC_PARAM]	        = "SyncParam",
3617 		[P_PROTOCOL]            = "ReportProtocol",
3618 		[P_UUIDS]	        = "ReportUUIDs",
3619 		[P_SIZES]	        = "ReportSizes",
3620 		[P_STATE]	        = "ReportState",
3621 		[P_SYNC_UUID]           = "ReportSyncUUID",
3622 		[P_AUTH_CHALLENGE]      = "AuthChallenge",
3623 		[P_AUTH_RESPONSE]	= "AuthResponse",
3624 		[P_STATE_CHG_REQ]       = "StateChgRequest",
3625 		[P_PING]		= "Ping",
3626 		[P_PING_ACK]	        = "PingAck",
3627 		[P_RECV_ACK]	        = "RecvAck",
3628 		[P_WRITE_ACK]	        = "WriteAck",
3629 		[P_RS_WRITE_ACK]	= "RSWriteAck",
3630 		[P_SUPERSEDED]          = "Superseded",
3631 		[P_NEG_ACK]	        = "NegAck",
3632 		[P_NEG_DREPLY]	        = "NegDReply",
3633 		[P_NEG_RS_DREPLY]	= "NegRSDReply",
3634 		[P_BARRIER_ACK]	        = "BarrierAck",
3635 		[P_STATE_CHG_REPLY]     = "StateChgReply",
3636 		[P_OV_REQUEST]          = "OVRequest",
3637 		[P_OV_REPLY]            = "OVReply",
3638 		[P_OV_RESULT]           = "OVResult",
3639 		[P_CSUM_RS_REQUEST]     = "CsumRSRequest",
3640 		[P_RS_IS_IN_SYNC]	= "CsumRSIsInSync",
3641 		[P_SYNC_PARAM89]	= "SyncParam89",
3642 		[P_COMPRESSED_BITMAP]   = "CBitmap",
3643 		[P_DELAY_PROBE]         = "DelayProbe",
3644 		[P_OUT_OF_SYNC]		= "OutOfSync",
3645 		[P_RS_CANCEL]		= "RSCancel",
3646 		[P_CONN_ST_CHG_REQ]	= "conn_st_chg_req",
3647 		[P_CONN_ST_CHG_REPLY]	= "conn_st_chg_reply",
3648 		[P_PROTOCOL_UPDATE]	= "protocol_update",
3649 		[P_TRIM]	        = "Trim",
3650 		[P_RS_THIN_REQ]         = "rs_thin_req",
3651 		[P_RS_DEALLOCATED]      = "rs_deallocated",
3652 		[P_WSAME]	        = "WriteSame",
3653 		[P_ZEROES]		= "Zeroes",
3654 
3655 		/* enum drbd_packet, but not commands - obsoleted flags:
3656 		 *	P_MAY_IGNORE
3657 		 *	P_MAX_OPT_CMD
3658 		 */
3659 	};
3660 
3661 	/* too big for the array: 0xfffX */
3662 	if (cmd == P_INITIAL_META)
3663 		return "InitialMeta";
3664 	if (cmd == P_INITIAL_DATA)
3665 		return "InitialData";
3666 	if (cmd == P_CONNECTION_FEATURES)
3667 		return "ConnectionFeatures";
3668 	if (cmd >= ARRAY_SIZE(cmdnames))
3669 		return "Unknown";
3670 	return cmdnames[cmd];
3671 }
3672 
3673 /**
3674  * drbd_wait_misc  -  wait for a request to make progress
3675  * @device:	device associated with the request
3676  * @i:		the struct drbd_interval embedded in struct drbd_request or
3677  *		struct drbd_peer_request
3678  */
3679 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3680 {
3681 	struct net_conf *nc;
3682 	DEFINE_WAIT(wait);
3683 	long timeout;
3684 
3685 	rcu_read_lock();
3686 	nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3687 	if (!nc) {
3688 		rcu_read_unlock();
3689 		return -ETIMEDOUT;
3690 	}
3691 	timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3692 	rcu_read_unlock();
3693 
3694 	/* Indicate to wake up device->misc_wait on progress.  */
3695 	i->waiting = true;
3696 	prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3697 	spin_unlock_irq(&device->resource->req_lock);
3698 	timeout = schedule_timeout(timeout);
3699 	finish_wait(&device->misc_wait, &wait);
3700 	spin_lock_irq(&device->resource->req_lock);
3701 	if (!timeout || device->state.conn < C_CONNECTED)
3702 		return -ETIMEDOUT;
3703 	if (signal_pending(current))
3704 		return -ERESTARTSYS;
3705 	return 0;
3706 }
3707 
3708 void lock_all_resources(void)
3709 {
3710 	struct drbd_resource *resource;
3711 	int __maybe_unused i = 0;
3712 
3713 	mutex_lock(&resources_mutex);
3714 	local_irq_disable();
3715 	for_each_resource(resource, &drbd_resources)
3716 		spin_lock_nested(&resource->req_lock, i++);
3717 }
3718 
3719 void unlock_all_resources(void)
3720 {
3721 	struct drbd_resource *resource;
3722 
3723 	for_each_resource(resource, &drbd_resources)
3724 		spin_unlock(&resource->req_lock);
3725 	local_irq_enable();
3726 	mutex_unlock(&resources_mutex);
3727 }
3728 
3729 #ifdef CONFIG_DRBD_FAULT_INJECTION
3730 /* Fault insertion support including random number generator shamelessly
3731  * stolen from kernel/rcutorture.c */
3732 struct fault_random_state {
3733 	unsigned long state;
3734 	unsigned long count;
3735 };
3736 
3737 #define FAULT_RANDOM_MULT 39916801  /* prime */
3738 #define FAULT_RANDOM_ADD	479001701 /* prime */
3739 #define FAULT_RANDOM_REFRESH 10000
3740 
3741 /*
3742  * Crude but fast random-number generator.  Uses a linear congruential
3743  * generator, with occasional help from get_random_bytes().
3744  */
3745 static unsigned long
3746 _drbd_fault_random(struct fault_random_state *rsp)
3747 {
3748 	long refresh;
3749 
3750 	if (!rsp->count--) {
3751 		get_random_bytes(&refresh, sizeof(refresh));
3752 		rsp->state += refresh;
3753 		rsp->count = FAULT_RANDOM_REFRESH;
3754 	}
3755 	rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3756 	return swahw32(rsp->state);
3757 }
3758 
3759 static char *
3760 _drbd_fault_str(unsigned int type) {
3761 	static char *_faults[] = {
3762 		[DRBD_FAULT_MD_WR] = "Meta-data write",
3763 		[DRBD_FAULT_MD_RD] = "Meta-data read",
3764 		[DRBD_FAULT_RS_WR] = "Resync write",
3765 		[DRBD_FAULT_RS_RD] = "Resync read",
3766 		[DRBD_FAULT_DT_WR] = "Data write",
3767 		[DRBD_FAULT_DT_RD] = "Data read",
3768 		[DRBD_FAULT_DT_RA] = "Data read ahead",
3769 		[DRBD_FAULT_BM_ALLOC] = "BM allocation",
3770 		[DRBD_FAULT_AL_EE] = "EE allocation",
3771 		[DRBD_FAULT_RECEIVE] = "receive data corruption",
3772 	};
3773 
3774 	return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3775 }
3776 
3777 unsigned int
3778 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3779 {
3780 	static struct fault_random_state rrs = {0, 0};
3781 
3782 	unsigned int ret = (
3783 		(drbd_fault_devs == 0 ||
3784 			((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3785 		(((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3786 
3787 	if (ret) {
3788 		drbd_fault_count++;
3789 
3790 		if (drbd_ratelimit())
3791 			drbd_warn(device, "***Simulating %s failure\n",
3792 				_drbd_fault_str(type));
3793 	}
3794 
3795 	return ret;
3796 }
3797 #endif
3798 
3799 module_init(drbd_init)
3800 module_exit(drbd_cleanup)
3801 
3802 EXPORT_SYMBOL(drbd_conn_str);
3803 EXPORT_SYMBOL(drbd_role_str);
3804 EXPORT_SYMBOL(drbd_disk_str);
3805 EXPORT_SYMBOL(drbd_set_st_err_str);
3806