xref: /linux/drivers/block/drbd/drbd_main.c (revision bd4af432cc71b5fbfe4833510359a6ad3ada250d)
1 // SPDX-License-Identifier: GPL-2.0-or-later
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 #define __KERNEL_SYSCALLS__
41 #include <linux/unistd.h>
42 #include <linux/vmalloc.h>
43 #include <linux/sched/signal.h>
44 
45 #include <linux/drbd_limits.h>
46 #include "drbd_int.h"
47 #include "drbd_protocol.h"
48 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
49 #include "drbd_vli.h"
50 #include "drbd_debugfs.h"
51 
52 static DEFINE_MUTEX(drbd_main_mutex);
53 static int drbd_open(struct block_device *bdev, fmode_t mode);
54 static void drbd_release(struct gendisk *gd, fmode_t mode);
55 static void md_sync_timer_fn(struct timer_list *t);
56 static int w_bitmap_io(struct drbd_work *w, int unused);
57 
58 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
59 	      "Lars Ellenberg <lars@linbit.com>");
60 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
61 MODULE_VERSION(REL_VERSION);
62 MODULE_LICENSE("GPL");
63 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
64 		 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
65 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
66 
67 #include <linux/moduleparam.h>
68 /* thanks to these macros, if compiled into the kernel (not-module),
69  * these become boot parameters (e.g., drbd.minor_count) */
70 
71 #ifdef CONFIG_DRBD_FAULT_INJECTION
72 int drbd_enable_faults;
73 int drbd_fault_rate;
74 static int drbd_fault_count;
75 static int drbd_fault_devs;
76 /* bitmap of enabled faults */
77 module_param_named(enable_faults, drbd_enable_faults, int, 0664);
78 /* fault rate % value - applies to all enabled faults */
79 module_param_named(fault_rate, drbd_fault_rate, int, 0664);
80 /* count of faults inserted */
81 module_param_named(fault_count, drbd_fault_count, int, 0664);
82 /* bitmap of devices to insert faults on */
83 module_param_named(fault_devs, drbd_fault_devs, int, 0644);
84 #endif
85 
86 /* module parameters we can keep static */
87 static bool drbd_allow_oos; /* allow_open_on_secondary */
88 static bool drbd_disable_sendpage;
89 MODULE_PARM_DESC(allow_oos, "DONT USE!");
90 module_param_named(allow_oos, drbd_allow_oos, bool, 0);
91 module_param_named(disable_sendpage, drbd_disable_sendpage, bool, 0644);
92 
93 /* module parameters we share */
94 int drbd_proc_details; /* Detail level in proc drbd*/
95 module_param_named(proc_details, drbd_proc_details, int, 0644);
96 /* module parameters shared with defaults */
97 unsigned int drbd_minor_count = DRBD_MINOR_COUNT_DEF;
98 /* Module parameter for setting the user mode helper program
99  * to run. Default is /sbin/drbdadm */
100 char drbd_usermode_helper[80] = "/sbin/drbdadm";
101 module_param_named(minor_count, drbd_minor_count, uint, 0444);
102 module_param_string(usermode_helper, drbd_usermode_helper, sizeof(drbd_usermode_helper), 0644);
103 
104 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
105  * as member "struct gendisk *vdisk;"
106  */
107 struct idr drbd_devices;
108 struct list_head drbd_resources;
109 struct mutex resources_mutex;
110 
111 struct kmem_cache *drbd_request_cache;
112 struct kmem_cache *drbd_ee_cache;	/* peer requests */
113 struct kmem_cache *drbd_bm_ext_cache;	/* bitmap extents */
114 struct kmem_cache *drbd_al_ext_cache;	/* activity log extents */
115 mempool_t drbd_request_mempool;
116 mempool_t drbd_ee_mempool;
117 mempool_t drbd_md_io_page_pool;
118 struct bio_set drbd_md_io_bio_set;
119 struct bio_set drbd_io_bio_set;
120 
121 /* I do not use a standard mempool, because:
122    1) I want to hand out the pre-allocated objects first.
123    2) I want to be able to interrupt sleeping allocation with a signal.
124    Note: This is a single linked list, the next pointer is the private
125 	 member of struct page.
126  */
127 struct page *drbd_pp_pool;
128 spinlock_t   drbd_pp_lock;
129 int          drbd_pp_vacant;
130 wait_queue_head_t drbd_pp_wait;
131 
132 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
133 
134 static const struct block_device_operations drbd_ops = {
135 	.owner =   THIS_MODULE,
136 	.open =    drbd_open,
137 	.release = drbd_release,
138 };
139 
140 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
141 {
142 	struct bio *bio;
143 
144 	if (!bioset_initialized(&drbd_md_io_bio_set))
145 		return bio_alloc(gfp_mask, 1);
146 
147 	bio = bio_alloc_bioset(gfp_mask, 1, &drbd_md_io_bio_set);
148 	if (!bio)
149 		return NULL;
150 	return bio;
151 }
152 
153 #ifdef __CHECKER__
154 /* When checking with sparse, and this is an inline function, sparse will
155    give tons of false positives. When this is a real functions sparse works.
156  */
157 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
158 {
159 	int io_allowed;
160 
161 	atomic_inc(&device->local_cnt);
162 	io_allowed = (device->state.disk >= mins);
163 	if (!io_allowed) {
164 		if (atomic_dec_and_test(&device->local_cnt))
165 			wake_up(&device->misc_wait);
166 	}
167 	return io_allowed;
168 }
169 
170 #endif
171 
172 /**
173  * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
174  * @connection:	DRBD connection.
175  * @barrier_nr:	Expected identifier of the DRBD write barrier packet.
176  * @set_size:	Expected number of requests before that barrier.
177  *
178  * In case the passed barrier_nr or set_size does not match the oldest
179  * epoch of not yet barrier-acked requests, this function will cause a
180  * termination of the connection.
181  */
182 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
183 		unsigned int set_size)
184 {
185 	struct drbd_request *r;
186 	struct drbd_request *req = NULL;
187 	int expect_epoch = 0;
188 	int expect_size = 0;
189 
190 	spin_lock_irq(&connection->resource->req_lock);
191 
192 	/* find oldest not yet barrier-acked write request,
193 	 * count writes in its epoch. */
194 	list_for_each_entry(r, &connection->transfer_log, tl_requests) {
195 		const unsigned s = r->rq_state;
196 		if (!req) {
197 			if (!(s & RQ_WRITE))
198 				continue;
199 			if (!(s & RQ_NET_MASK))
200 				continue;
201 			if (s & RQ_NET_DONE)
202 				continue;
203 			req = r;
204 			expect_epoch = req->epoch;
205 			expect_size ++;
206 		} else {
207 			if (r->epoch != expect_epoch)
208 				break;
209 			if (!(s & RQ_WRITE))
210 				continue;
211 			/* if (s & RQ_DONE): not expected */
212 			/* if (!(s & RQ_NET_MASK)): not expected */
213 			expect_size++;
214 		}
215 	}
216 
217 	/* first some paranoia code */
218 	if (req == NULL) {
219 		drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
220 			 barrier_nr);
221 		goto bail;
222 	}
223 	if (expect_epoch != barrier_nr) {
224 		drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
225 			 barrier_nr, expect_epoch);
226 		goto bail;
227 	}
228 
229 	if (expect_size != set_size) {
230 		drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
231 			 barrier_nr, set_size, expect_size);
232 		goto bail;
233 	}
234 
235 	/* Clean up list of requests processed during current epoch. */
236 	/* this extra list walk restart is paranoia,
237 	 * to catch requests being barrier-acked "unexpectedly".
238 	 * It usually should find the same req again, or some READ preceding it. */
239 	list_for_each_entry(req, &connection->transfer_log, tl_requests)
240 		if (req->epoch == expect_epoch)
241 			break;
242 	list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
243 		if (req->epoch != expect_epoch)
244 			break;
245 		_req_mod(req, BARRIER_ACKED);
246 	}
247 	spin_unlock_irq(&connection->resource->req_lock);
248 
249 	return;
250 
251 bail:
252 	spin_unlock_irq(&connection->resource->req_lock);
253 	conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
254 }
255 
256 
257 /**
258  * _tl_restart() - Walks the transfer log, and applies an action to all requests
259  * @connection:	DRBD connection to operate on.
260  * @what:       The action/event to perform with all request objects
261  *
262  * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
263  * RESTART_FROZEN_DISK_IO.
264  */
265 /* must hold resource->req_lock */
266 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
267 {
268 	struct drbd_request *req, *r;
269 
270 	list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
271 		_req_mod(req, what);
272 }
273 
274 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
275 {
276 	spin_lock_irq(&connection->resource->req_lock);
277 	_tl_restart(connection, what);
278 	spin_unlock_irq(&connection->resource->req_lock);
279 }
280 
281 /**
282  * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
283  * @device:	DRBD device.
284  *
285  * This is called after the connection to the peer was lost. The storage covered
286  * by the requests on the transfer gets marked as our of sync. Called from the
287  * receiver thread and the worker thread.
288  */
289 void tl_clear(struct drbd_connection *connection)
290 {
291 	tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
292 }
293 
294 /**
295  * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
296  * @device:	DRBD device.
297  */
298 void tl_abort_disk_io(struct drbd_device *device)
299 {
300 	struct drbd_connection *connection = first_peer_device(device)->connection;
301 	struct drbd_request *req, *r;
302 
303 	spin_lock_irq(&connection->resource->req_lock);
304 	list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
305 		if (!(req->rq_state & RQ_LOCAL_PENDING))
306 			continue;
307 		if (req->device != device)
308 			continue;
309 		_req_mod(req, ABORT_DISK_IO);
310 	}
311 	spin_unlock_irq(&connection->resource->req_lock);
312 }
313 
314 static int drbd_thread_setup(void *arg)
315 {
316 	struct drbd_thread *thi = (struct drbd_thread *) arg;
317 	struct drbd_resource *resource = thi->resource;
318 	unsigned long flags;
319 	int retval;
320 
321 	snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
322 		 thi->name[0],
323 		 resource->name);
324 
325 	allow_kernel_signal(DRBD_SIGKILL);
326 	allow_kernel_signal(SIGXCPU);
327 restart:
328 	retval = thi->function(thi);
329 
330 	spin_lock_irqsave(&thi->t_lock, flags);
331 
332 	/* if the receiver has been "EXITING", the last thing it did
333 	 * was set the conn state to "StandAlone",
334 	 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
335 	 * and receiver thread will be "started".
336 	 * drbd_thread_start needs to set "RESTARTING" in that case.
337 	 * t_state check and assignment needs to be within the same spinlock,
338 	 * so either thread_start sees EXITING, and can remap to RESTARTING,
339 	 * or thread_start see NONE, and can proceed as normal.
340 	 */
341 
342 	if (thi->t_state == RESTARTING) {
343 		drbd_info(resource, "Restarting %s thread\n", thi->name);
344 		thi->t_state = RUNNING;
345 		spin_unlock_irqrestore(&thi->t_lock, flags);
346 		goto restart;
347 	}
348 
349 	thi->task = NULL;
350 	thi->t_state = NONE;
351 	smp_mb();
352 	complete_all(&thi->stop);
353 	spin_unlock_irqrestore(&thi->t_lock, flags);
354 
355 	drbd_info(resource, "Terminating %s\n", current->comm);
356 
357 	/* Release mod reference taken when thread was started */
358 
359 	if (thi->connection)
360 		kref_put(&thi->connection->kref, drbd_destroy_connection);
361 	kref_put(&resource->kref, drbd_destroy_resource);
362 	module_put(THIS_MODULE);
363 	return retval;
364 }
365 
366 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
367 			     int (*func) (struct drbd_thread *), const char *name)
368 {
369 	spin_lock_init(&thi->t_lock);
370 	thi->task    = NULL;
371 	thi->t_state = NONE;
372 	thi->function = func;
373 	thi->resource = resource;
374 	thi->connection = NULL;
375 	thi->name = name;
376 }
377 
378 int drbd_thread_start(struct drbd_thread *thi)
379 {
380 	struct drbd_resource *resource = thi->resource;
381 	struct task_struct *nt;
382 	unsigned long flags;
383 
384 	/* is used from state engine doing drbd_thread_stop_nowait,
385 	 * while holding the req lock irqsave */
386 	spin_lock_irqsave(&thi->t_lock, flags);
387 
388 	switch (thi->t_state) {
389 	case NONE:
390 		drbd_info(resource, "Starting %s thread (from %s [%d])\n",
391 			 thi->name, current->comm, current->pid);
392 
393 		/* Get ref on module for thread - this is released when thread exits */
394 		if (!try_module_get(THIS_MODULE)) {
395 			drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
396 			spin_unlock_irqrestore(&thi->t_lock, flags);
397 			return false;
398 		}
399 
400 		kref_get(&resource->kref);
401 		if (thi->connection)
402 			kref_get(&thi->connection->kref);
403 
404 		init_completion(&thi->stop);
405 		thi->reset_cpu_mask = 1;
406 		thi->t_state = RUNNING;
407 		spin_unlock_irqrestore(&thi->t_lock, flags);
408 		flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
409 
410 		nt = kthread_create(drbd_thread_setup, (void *) thi,
411 				    "drbd_%c_%s", thi->name[0], thi->resource->name);
412 
413 		if (IS_ERR(nt)) {
414 			drbd_err(resource, "Couldn't start thread\n");
415 
416 			if (thi->connection)
417 				kref_put(&thi->connection->kref, drbd_destroy_connection);
418 			kref_put(&resource->kref, drbd_destroy_resource);
419 			module_put(THIS_MODULE);
420 			return false;
421 		}
422 		spin_lock_irqsave(&thi->t_lock, flags);
423 		thi->task = nt;
424 		thi->t_state = RUNNING;
425 		spin_unlock_irqrestore(&thi->t_lock, flags);
426 		wake_up_process(nt);
427 		break;
428 	case EXITING:
429 		thi->t_state = RESTARTING;
430 		drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
431 				thi->name, current->comm, current->pid);
432 		/* fall through */
433 	case RUNNING:
434 	case RESTARTING:
435 	default:
436 		spin_unlock_irqrestore(&thi->t_lock, flags);
437 		break;
438 	}
439 
440 	return true;
441 }
442 
443 
444 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
445 {
446 	unsigned long flags;
447 
448 	enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
449 
450 	/* may be called from state engine, holding the req lock irqsave */
451 	spin_lock_irqsave(&thi->t_lock, flags);
452 
453 	if (thi->t_state == NONE) {
454 		spin_unlock_irqrestore(&thi->t_lock, flags);
455 		if (restart)
456 			drbd_thread_start(thi);
457 		return;
458 	}
459 
460 	if (thi->t_state != ns) {
461 		if (thi->task == NULL) {
462 			spin_unlock_irqrestore(&thi->t_lock, flags);
463 			return;
464 		}
465 
466 		thi->t_state = ns;
467 		smp_mb();
468 		init_completion(&thi->stop);
469 		if (thi->task != current)
470 			send_sig(DRBD_SIGKILL, thi->task, 1);
471 	}
472 
473 	spin_unlock_irqrestore(&thi->t_lock, flags);
474 
475 	if (wait)
476 		wait_for_completion(&thi->stop);
477 }
478 
479 int conn_lowest_minor(struct drbd_connection *connection)
480 {
481 	struct drbd_peer_device *peer_device;
482 	int vnr = 0, minor = -1;
483 
484 	rcu_read_lock();
485 	peer_device = idr_get_next(&connection->peer_devices, &vnr);
486 	if (peer_device)
487 		minor = device_to_minor(peer_device->device);
488 	rcu_read_unlock();
489 
490 	return minor;
491 }
492 
493 #ifdef CONFIG_SMP
494 /**
495  * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
496  *
497  * Forces all threads of a resource onto the same CPU. This is beneficial for
498  * DRBD's performance. May be overwritten by user's configuration.
499  */
500 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
501 {
502 	unsigned int *resources_per_cpu, min_index = ~0;
503 
504 	resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu),
505 				    GFP_KERNEL);
506 	if (resources_per_cpu) {
507 		struct drbd_resource *resource;
508 		unsigned int cpu, min = ~0;
509 
510 		rcu_read_lock();
511 		for_each_resource_rcu(resource, &drbd_resources) {
512 			for_each_cpu(cpu, resource->cpu_mask)
513 				resources_per_cpu[cpu]++;
514 		}
515 		rcu_read_unlock();
516 		for_each_online_cpu(cpu) {
517 			if (resources_per_cpu[cpu] < min) {
518 				min = resources_per_cpu[cpu];
519 				min_index = cpu;
520 			}
521 		}
522 		kfree(resources_per_cpu);
523 	}
524 	if (min_index == ~0) {
525 		cpumask_setall(*cpu_mask);
526 		return;
527 	}
528 	cpumask_set_cpu(min_index, *cpu_mask);
529 }
530 
531 /**
532  * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
533  * @device:	DRBD device.
534  * @thi:	drbd_thread object
535  *
536  * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
537  * prematurely.
538  */
539 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
540 {
541 	struct drbd_resource *resource = thi->resource;
542 	struct task_struct *p = current;
543 
544 	if (!thi->reset_cpu_mask)
545 		return;
546 	thi->reset_cpu_mask = 0;
547 	set_cpus_allowed_ptr(p, resource->cpu_mask);
548 }
549 #else
550 #define drbd_calc_cpu_mask(A) ({})
551 #endif
552 
553 /**
554  * drbd_header_size  -  size of a packet header
555  *
556  * The header size is a multiple of 8, so any payload following the header is
557  * word aligned on 64-bit architectures.  (The bitmap send and receive code
558  * relies on this.)
559  */
560 unsigned int drbd_header_size(struct drbd_connection *connection)
561 {
562 	if (connection->agreed_pro_version >= 100) {
563 		BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
564 		return sizeof(struct p_header100);
565 	} else {
566 		BUILD_BUG_ON(sizeof(struct p_header80) !=
567 			     sizeof(struct p_header95));
568 		BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
569 		return sizeof(struct p_header80);
570 	}
571 }
572 
573 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
574 {
575 	h->magic   = cpu_to_be32(DRBD_MAGIC);
576 	h->command = cpu_to_be16(cmd);
577 	h->length  = cpu_to_be16(size);
578 	return sizeof(struct p_header80);
579 }
580 
581 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
582 {
583 	h->magic   = cpu_to_be16(DRBD_MAGIC_BIG);
584 	h->command = cpu_to_be16(cmd);
585 	h->length = cpu_to_be32(size);
586 	return sizeof(struct p_header95);
587 }
588 
589 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
590 				      int size, int vnr)
591 {
592 	h->magic = cpu_to_be32(DRBD_MAGIC_100);
593 	h->volume = cpu_to_be16(vnr);
594 	h->command = cpu_to_be16(cmd);
595 	h->length = cpu_to_be32(size);
596 	h->pad = 0;
597 	return sizeof(struct p_header100);
598 }
599 
600 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
601 				   void *buffer, enum drbd_packet cmd, int size)
602 {
603 	if (connection->agreed_pro_version >= 100)
604 		return prepare_header100(buffer, cmd, size, vnr);
605 	else if (connection->agreed_pro_version >= 95 &&
606 		 size > DRBD_MAX_SIZE_H80_PACKET)
607 		return prepare_header95(buffer, cmd, size);
608 	else
609 		return prepare_header80(buffer, cmd, size);
610 }
611 
612 static void *__conn_prepare_command(struct drbd_connection *connection,
613 				    struct drbd_socket *sock)
614 {
615 	if (!sock->socket)
616 		return NULL;
617 	return sock->sbuf + drbd_header_size(connection);
618 }
619 
620 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
621 {
622 	void *p;
623 
624 	mutex_lock(&sock->mutex);
625 	p = __conn_prepare_command(connection, sock);
626 	if (!p)
627 		mutex_unlock(&sock->mutex);
628 
629 	return p;
630 }
631 
632 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
633 {
634 	return conn_prepare_command(peer_device->connection, sock);
635 }
636 
637 static int __send_command(struct drbd_connection *connection, int vnr,
638 			  struct drbd_socket *sock, enum drbd_packet cmd,
639 			  unsigned int header_size, void *data,
640 			  unsigned int size)
641 {
642 	int msg_flags;
643 	int err;
644 
645 	/*
646 	 * Called with @data == NULL and the size of the data blocks in @size
647 	 * for commands that send data blocks.  For those commands, omit the
648 	 * MSG_MORE flag: this will increase the likelihood that data blocks
649 	 * which are page aligned on the sender will end up page aligned on the
650 	 * receiver.
651 	 */
652 	msg_flags = data ? MSG_MORE : 0;
653 
654 	header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
655 				      header_size + size);
656 	err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
657 			    msg_flags);
658 	if (data && !err)
659 		err = drbd_send_all(connection, sock->socket, data, size, 0);
660 	/* DRBD protocol "pings" are latency critical.
661 	 * This is supposed to trigger tcp_push_pending_frames() */
662 	if (!err && (cmd == P_PING || cmd == P_PING_ACK))
663 		drbd_tcp_nodelay(sock->socket);
664 
665 	return err;
666 }
667 
668 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
669 			       enum drbd_packet cmd, unsigned int header_size,
670 			       void *data, unsigned int size)
671 {
672 	return __send_command(connection, 0, sock, cmd, header_size, data, size);
673 }
674 
675 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
676 		      enum drbd_packet cmd, unsigned int header_size,
677 		      void *data, unsigned int size)
678 {
679 	int err;
680 
681 	err = __conn_send_command(connection, sock, cmd, header_size, data, size);
682 	mutex_unlock(&sock->mutex);
683 	return err;
684 }
685 
686 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
687 		      enum drbd_packet cmd, unsigned int header_size,
688 		      void *data, unsigned int size)
689 {
690 	int err;
691 
692 	err = __send_command(peer_device->connection, peer_device->device->vnr,
693 			     sock, cmd, header_size, data, size);
694 	mutex_unlock(&sock->mutex);
695 	return err;
696 }
697 
698 int drbd_send_ping(struct drbd_connection *connection)
699 {
700 	struct drbd_socket *sock;
701 
702 	sock = &connection->meta;
703 	if (!conn_prepare_command(connection, sock))
704 		return -EIO;
705 	return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
706 }
707 
708 int drbd_send_ping_ack(struct drbd_connection *connection)
709 {
710 	struct drbd_socket *sock;
711 
712 	sock = &connection->meta;
713 	if (!conn_prepare_command(connection, sock))
714 		return -EIO;
715 	return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
716 }
717 
718 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
719 {
720 	struct drbd_socket *sock;
721 	struct p_rs_param_95 *p;
722 	int size;
723 	const int apv = peer_device->connection->agreed_pro_version;
724 	enum drbd_packet cmd;
725 	struct net_conf *nc;
726 	struct disk_conf *dc;
727 
728 	sock = &peer_device->connection->data;
729 	p = drbd_prepare_command(peer_device, sock);
730 	if (!p)
731 		return -EIO;
732 
733 	rcu_read_lock();
734 	nc = rcu_dereference(peer_device->connection->net_conf);
735 
736 	size = apv <= 87 ? sizeof(struct p_rs_param)
737 		: apv == 88 ? sizeof(struct p_rs_param)
738 			+ strlen(nc->verify_alg) + 1
739 		: apv <= 94 ? sizeof(struct p_rs_param_89)
740 		: /* apv >= 95 */ sizeof(struct p_rs_param_95);
741 
742 	cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
743 
744 	/* initialize verify_alg and csums_alg */
745 	memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
746 
747 	if (get_ldev(peer_device->device)) {
748 		dc = rcu_dereference(peer_device->device->ldev->disk_conf);
749 		p->resync_rate = cpu_to_be32(dc->resync_rate);
750 		p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
751 		p->c_delay_target = cpu_to_be32(dc->c_delay_target);
752 		p->c_fill_target = cpu_to_be32(dc->c_fill_target);
753 		p->c_max_rate = cpu_to_be32(dc->c_max_rate);
754 		put_ldev(peer_device->device);
755 	} else {
756 		p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
757 		p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
758 		p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
759 		p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
760 		p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
761 	}
762 
763 	if (apv >= 88)
764 		strcpy(p->verify_alg, nc->verify_alg);
765 	if (apv >= 89)
766 		strcpy(p->csums_alg, nc->csums_alg);
767 	rcu_read_unlock();
768 
769 	return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
770 }
771 
772 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
773 {
774 	struct drbd_socket *sock;
775 	struct p_protocol *p;
776 	struct net_conf *nc;
777 	int size, cf;
778 
779 	sock = &connection->data;
780 	p = __conn_prepare_command(connection, sock);
781 	if (!p)
782 		return -EIO;
783 
784 	rcu_read_lock();
785 	nc = rcu_dereference(connection->net_conf);
786 
787 	if (nc->tentative && connection->agreed_pro_version < 92) {
788 		rcu_read_unlock();
789 		drbd_err(connection, "--dry-run is not supported by peer");
790 		return -EOPNOTSUPP;
791 	}
792 
793 	size = sizeof(*p);
794 	if (connection->agreed_pro_version >= 87)
795 		size += strlen(nc->integrity_alg) + 1;
796 
797 	p->protocol      = cpu_to_be32(nc->wire_protocol);
798 	p->after_sb_0p   = cpu_to_be32(nc->after_sb_0p);
799 	p->after_sb_1p   = cpu_to_be32(nc->after_sb_1p);
800 	p->after_sb_2p   = cpu_to_be32(nc->after_sb_2p);
801 	p->two_primaries = cpu_to_be32(nc->two_primaries);
802 	cf = 0;
803 	if (nc->discard_my_data)
804 		cf |= CF_DISCARD_MY_DATA;
805 	if (nc->tentative)
806 		cf |= CF_DRY_RUN;
807 	p->conn_flags    = cpu_to_be32(cf);
808 
809 	if (connection->agreed_pro_version >= 87)
810 		strcpy(p->integrity_alg, nc->integrity_alg);
811 	rcu_read_unlock();
812 
813 	return __conn_send_command(connection, sock, cmd, size, NULL, 0);
814 }
815 
816 int drbd_send_protocol(struct drbd_connection *connection)
817 {
818 	int err;
819 
820 	mutex_lock(&connection->data.mutex);
821 	err = __drbd_send_protocol(connection, P_PROTOCOL);
822 	mutex_unlock(&connection->data.mutex);
823 
824 	return err;
825 }
826 
827 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
828 {
829 	struct drbd_device *device = peer_device->device;
830 	struct drbd_socket *sock;
831 	struct p_uuids *p;
832 	int i;
833 
834 	if (!get_ldev_if_state(device, D_NEGOTIATING))
835 		return 0;
836 
837 	sock = &peer_device->connection->data;
838 	p = drbd_prepare_command(peer_device, sock);
839 	if (!p) {
840 		put_ldev(device);
841 		return -EIO;
842 	}
843 	spin_lock_irq(&device->ldev->md.uuid_lock);
844 	for (i = UI_CURRENT; i < UI_SIZE; i++)
845 		p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
846 	spin_unlock_irq(&device->ldev->md.uuid_lock);
847 
848 	device->comm_bm_set = drbd_bm_total_weight(device);
849 	p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
850 	rcu_read_lock();
851 	uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
852 	rcu_read_unlock();
853 	uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
854 	uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
855 	p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
856 
857 	put_ldev(device);
858 	return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
859 }
860 
861 int drbd_send_uuids(struct drbd_peer_device *peer_device)
862 {
863 	return _drbd_send_uuids(peer_device, 0);
864 }
865 
866 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
867 {
868 	return _drbd_send_uuids(peer_device, 8);
869 }
870 
871 void drbd_print_uuids(struct drbd_device *device, const char *text)
872 {
873 	if (get_ldev_if_state(device, D_NEGOTIATING)) {
874 		u64 *uuid = device->ldev->md.uuid;
875 		drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
876 		     text,
877 		     (unsigned long long)uuid[UI_CURRENT],
878 		     (unsigned long long)uuid[UI_BITMAP],
879 		     (unsigned long long)uuid[UI_HISTORY_START],
880 		     (unsigned long long)uuid[UI_HISTORY_END]);
881 		put_ldev(device);
882 	} else {
883 		drbd_info(device, "%s effective data uuid: %016llX\n",
884 				text,
885 				(unsigned long long)device->ed_uuid);
886 	}
887 }
888 
889 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
890 {
891 	struct drbd_device *device = peer_device->device;
892 	struct drbd_socket *sock;
893 	struct p_rs_uuid *p;
894 	u64 uuid;
895 
896 	D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
897 
898 	uuid = device->ldev->md.uuid[UI_BITMAP];
899 	if (uuid && uuid != UUID_JUST_CREATED)
900 		uuid = uuid + UUID_NEW_BM_OFFSET;
901 	else
902 		get_random_bytes(&uuid, sizeof(u64));
903 	drbd_uuid_set(device, UI_BITMAP, uuid);
904 	drbd_print_uuids(device, "updated sync UUID");
905 	drbd_md_sync(device);
906 
907 	sock = &peer_device->connection->data;
908 	p = drbd_prepare_command(peer_device, sock);
909 	if (p) {
910 		p->uuid = cpu_to_be64(uuid);
911 		drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
912 	}
913 }
914 
915 /* communicated if (agreed_features & DRBD_FF_WSAME) */
916 static void
917 assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p,
918 					struct request_queue *q)
919 {
920 	if (q) {
921 		p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
922 		p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
923 		p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
924 		p->qlim->io_min = cpu_to_be32(queue_io_min(q));
925 		p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
926 		p->qlim->discard_enabled = blk_queue_discard(q);
927 		p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
928 	} else {
929 		q = device->rq_queue;
930 		p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
931 		p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
932 		p->qlim->alignment_offset = 0;
933 		p->qlim->io_min = cpu_to_be32(queue_io_min(q));
934 		p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
935 		p->qlim->discard_enabled = 0;
936 		p->qlim->write_same_capable = 0;
937 	}
938 }
939 
940 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
941 {
942 	struct drbd_device *device = peer_device->device;
943 	struct drbd_socket *sock;
944 	struct p_sizes *p;
945 	sector_t d_size, u_size;
946 	int q_order_type;
947 	unsigned int max_bio_size;
948 	unsigned int packet_size;
949 
950 	sock = &peer_device->connection->data;
951 	p = drbd_prepare_command(peer_device, sock);
952 	if (!p)
953 		return -EIO;
954 
955 	packet_size = sizeof(*p);
956 	if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
957 		packet_size += sizeof(p->qlim[0]);
958 
959 	memset(p, 0, packet_size);
960 	if (get_ldev_if_state(device, D_NEGOTIATING)) {
961 		struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
962 		d_size = drbd_get_max_capacity(device->ldev);
963 		rcu_read_lock();
964 		u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
965 		rcu_read_unlock();
966 		q_order_type = drbd_queue_order_type(device);
967 		max_bio_size = queue_max_hw_sectors(q) << 9;
968 		max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
969 		assign_p_sizes_qlim(device, p, q);
970 		put_ldev(device);
971 	} else {
972 		d_size = 0;
973 		u_size = 0;
974 		q_order_type = QUEUE_ORDERED_NONE;
975 		max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
976 		assign_p_sizes_qlim(device, p, NULL);
977 	}
978 
979 	if (peer_device->connection->agreed_pro_version <= 94)
980 		max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
981 	else if (peer_device->connection->agreed_pro_version < 100)
982 		max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
983 
984 	p->d_size = cpu_to_be64(d_size);
985 	p->u_size = cpu_to_be64(u_size);
986 	p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
987 	p->max_bio_size = cpu_to_be32(max_bio_size);
988 	p->queue_order_type = cpu_to_be16(q_order_type);
989 	p->dds_flags = cpu_to_be16(flags);
990 
991 	return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
992 }
993 
994 /**
995  * drbd_send_current_state() - Sends the drbd state to the peer
996  * @peer_device:	DRBD peer device.
997  */
998 int drbd_send_current_state(struct drbd_peer_device *peer_device)
999 {
1000 	struct drbd_socket *sock;
1001 	struct p_state *p;
1002 
1003 	sock = &peer_device->connection->data;
1004 	p = drbd_prepare_command(peer_device, sock);
1005 	if (!p)
1006 		return -EIO;
1007 	p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1008 	return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1009 }
1010 
1011 /**
1012  * drbd_send_state() - After a state change, sends the new state to the peer
1013  * @peer_device:      DRBD peer device.
1014  * @state:     the state to send, not necessarily the current state.
1015  *
1016  * Each state change queues an "after_state_ch" work, which will eventually
1017  * send the resulting new state to the peer. If more state changes happen
1018  * between queuing and processing of the after_state_ch work, we still
1019  * want to send each intermediary state in the order it occurred.
1020  */
1021 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1022 {
1023 	struct drbd_socket *sock;
1024 	struct p_state *p;
1025 
1026 	sock = &peer_device->connection->data;
1027 	p = drbd_prepare_command(peer_device, sock);
1028 	if (!p)
1029 		return -EIO;
1030 	p->state = cpu_to_be32(state.i); /* Within the send mutex */
1031 	return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1032 }
1033 
1034 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1035 {
1036 	struct drbd_socket *sock;
1037 	struct p_req_state *p;
1038 
1039 	sock = &peer_device->connection->data;
1040 	p = drbd_prepare_command(peer_device, sock);
1041 	if (!p)
1042 		return -EIO;
1043 	p->mask = cpu_to_be32(mask.i);
1044 	p->val = cpu_to_be32(val.i);
1045 	return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1046 }
1047 
1048 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1049 {
1050 	enum drbd_packet cmd;
1051 	struct drbd_socket *sock;
1052 	struct p_req_state *p;
1053 
1054 	cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1055 	sock = &connection->data;
1056 	p = conn_prepare_command(connection, sock);
1057 	if (!p)
1058 		return -EIO;
1059 	p->mask = cpu_to_be32(mask.i);
1060 	p->val = cpu_to_be32(val.i);
1061 	return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1062 }
1063 
1064 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1065 {
1066 	struct drbd_socket *sock;
1067 	struct p_req_state_reply *p;
1068 
1069 	sock = &peer_device->connection->meta;
1070 	p = drbd_prepare_command(peer_device, sock);
1071 	if (p) {
1072 		p->retcode = cpu_to_be32(retcode);
1073 		drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1074 	}
1075 }
1076 
1077 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1078 {
1079 	struct drbd_socket *sock;
1080 	struct p_req_state_reply *p;
1081 	enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1082 
1083 	sock = &connection->meta;
1084 	p = conn_prepare_command(connection, sock);
1085 	if (p) {
1086 		p->retcode = cpu_to_be32(retcode);
1087 		conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1088 	}
1089 }
1090 
1091 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1092 {
1093 	BUG_ON(code & ~0xf);
1094 	p->encoding = (p->encoding & ~0xf) | code;
1095 }
1096 
1097 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1098 {
1099 	p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1100 }
1101 
1102 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1103 {
1104 	BUG_ON(n & ~0x7);
1105 	p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1106 }
1107 
1108 static int fill_bitmap_rle_bits(struct drbd_device *device,
1109 			 struct p_compressed_bm *p,
1110 			 unsigned int size,
1111 			 struct bm_xfer_ctx *c)
1112 {
1113 	struct bitstream bs;
1114 	unsigned long plain_bits;
1115 	unsigned long tmp;
1116 	unsigned long rl;
1117 	unsigned len;
1118 	unsigned toggle;
1119 	int bits, use_rle;
1120 
1121 	/* may we use this feature? */
1122 	rcu_read_lock();
1123 	use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1124 	rcu_read_unlock();
1125 	if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1126 		return 0;
1127 
1128 	if (c->bit_offset >= c->bm_bits)
1129 		return 0; /* nothing to do. */
1130 
1131 	/* use at most thus many bytes */
1132 	bitstream_init(&bs, p->code, size, 0);
1133 	memset(p->code, 0, size);
1134 	/* plain bits covered in this code string */
1135 	plain_bits = 0;
1136 
1137 	/* p->encoding & 0x80 stores whether the first run length is set.
1138 	 * bit offset is implicit.
1139 	 * start with toggle == 2 to be able to tell the first iteration */
1140 	toggle = 2;
1141 
1142 	/* see how much plain bits we can stuff into one packet
1143 	 * using RLE and VLI. */
1144 	do {
1145 		tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1146 				    : _drbd_bm_find_next(device, c->bit_offset);
1147 		if (tmp == -1UL)
1148 			tmp = c->bm_bits;
1149 		rl = tmp - c->bit_offset;
1150 
1151 		if (toggle == 2) { /* first iteration */
1152 			if (rl == 0) {
1153 				/* the first checked bit was set,
1154 				 * store start value, */
1155 				dcbp_set_start(p, 1);
1156 				/* but skip encoding of zero run length */
1157 				toggle = !toggle;
1158 				continue;
1159 			}
1160 			dcbp_set_start(p, 0);
1161 		}
1162 
1163 		/* paranoia: catch zero runlength.
1164 		 * can only happen if bitmap is modified while we scan it. */
1165 		if (rl == 0) {
1166 			drbd_err(device, "unexpected zero runlength while encoding bitmap "
1167 			    "t:%u bo:%lu\n", toggle, c->bit_offset);
1168 			return -1;
1169 		}
1170 
1171 		bits = vli_encode_bits(&bs, rl);
1172 		if (bits == -ENOBUFS) /* buffer full */
1173 			break;
1174 		if (bits <= 0) {
1175 			drbd_err(device, "error while encoding bitmap: %d\n", bits);
1176 			return 0;
1177 		}
1178 
1179 		toggle = !toggle;
1180 		plain_bits += rl;
1181 		c->bit_offset = tmp;
1182 	} while (c->bit_offset < c->bm_bits);
1183 
1184 	len = bs.cur.b - p->code + !!bs.cur.bit;
1185 
1186 	if (plain_bits < (len << 3)) {
1187 		/* incompressible with this method.
1188 		 * we need to rewind both word and bit position. */
1189 		c->bit_offset -= plain_bits;
1190 		bm_xfer_ctx_bit_to_word_offset(c);
1191 		c->bit_offset = c->word_offset * BITS_PER_LONG;
1192 		return 0;
1193 	}
1194 
1195 	/* RLE + VLI was able to compress it just fine.
1196 	 * update c->word_offset. */
1197 	bm_xfer_ctx_bit_to_word_offset(c);
1198 
1199 	/* store pad_bits */
1200 	dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1201 
1202 	return len;
1203 }
1204 
1205 /**
1206  * send_bitmap_rle_or_plain
1207  *
1208  * Return 0 when done, 1 when another iteration is needed, and a negative error
1209  * code upon failure.
1210  */
1211 static int
1212 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1213 {
1214 	struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1215 	unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1216 	struct p_compressed_bm *p = sock->sbuf + header_size;
1217 	int len, err;
1218 
1219 	len = fill_bitmap_rle_bits(device, p,
1220 			DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1221 	if (len < 0)
1222 		return -EIO;
1223 
1224 	if (len) {
1225 		dcbp_set_code(p, RLE_VLI_Bits);
1226 		err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1227 				     P_COMPRESSED_BITMAP, sizeof(*p) + len,
1228 				     NULL, 0);
1229 		c->packets[0]++;
1230 		c->bytes[0] += header_size + sizeof(*p) + len;
1231 
1232 		if (c->bit_offset >= c->bm_bits)
1233 			len = 0; /* DONE */
1234 	} else {
1235 		/* was not compressible.
1236 		 * send a buffer full of plain text bits instead. */
1237 		unsigned int data_size;
1238 		unsigned long num_words;
1239 		unsigned long *p = sock->sbuf + header_size;
1240 
1241 		data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1242 		num_words = min_t(size_t, data_size / sizeof(*p),
1243 				  c->bm_words - c->word_offset);
1244 		len = num_words * sizeof(*p);
1245 		if (len)
1246 			drbd_bm_get_lel(device, c->word_offset, num_words, p);
1247 		err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1248 		c->word_offset += num_words;
1249 		c->bit_offset = c->word_offset * BITS_PER_LONG;
1250 
1251 		c->packets[1]++;
1252 		c->bytes[1] += header_size + len;
1253 
1254 		if (c->bit_offset > c->bm_bits)
1255 			c->bit_offset = c->bm_bits;
1256 	}
1257 	if (!err) {
1258 		if (len == 0) {
1259 			INFO_bm_xfer_stats(device, "send", c);
1260 			return 0;
1261 		} else
1262 			return 1;
1263 	}
1264 	return -EIO;
1265 }
1266 
1267 /* See the comment at receive_bitmap() */
1268 static int _drbd_send_bitmap(struct drbd_device *device)
1269 {
1270 	struct bm_xfer_ctx c;
1271 	int err;
1272 
1273 	if (!expect(device->bitmap))
1274 		return false;
1275 
1276 	if (get_ldev(device)) {
1277 		if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1278 			drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1279 			drbd_bm_set_all(device);
1280 			if (drbd_bm_write(device)) {
1281 				/* write_bm did fail! Leave full sync flag set in Meta P_DATA
1282 				 * but otherwise process as per normal - need to tell other
1283 				 * side that a full resync is required! */
1284 				drbd_err(device, "Failed to write bitmap to disk!\n");
1285 			} else {
1286 				drbd_md_clear_flag(device, MDF_FULL_SYNC);
1287 				drbd_md_sync(device);
1288 			}
1289 		}
1290 		put_ldev(device);
1291 	}
1292 
1293 	c = (struct bm_xfer_ctx) {
1294 		.bm_bits = drbd_bm_bits(device),
1295 		.bm_words = drbd_bm_words(device),
1296 	};
1297 
1298 	do {
1299 		err = send_bitmap_rle_or_plain(device, &c);
1300 	} while (err > 0);
1301 
1302 	return err == 0;
1303 }
1304 
1305 int drbd_send_bitmap(struct drbd_device *device)
1306 {
1307 	struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1308 	int err = -1;
1309 
1310 	mutex_lock(&sock->mutex);
1311 	if (sock->socket)
1312 		err = !_drbd_send_bitmap(device);
1313 	mutex_unlock(&sock->mutex);
1314 	return err;
1315 }
1316 
1317 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1318 {
1319 	struct drbd_socket *sock;
1320 	struct p_barrier_ack *p;
1321 
1322 	if (connection->cstate < C_WF_REPORT_PARAMS)
1323 		return;
1324 
1325 	sock = &connection->meta;
1326 	p = conn_prepare_command(connection, sock);
1327 	if (!p)
1328 		return;
1329 	p->barrier = barrier_nr;
1330 	p->set_size = cpu_to_be32(set_size);
1331 	conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1332 }
1333 
1334 /**
1335  * _drbd_send_ack() - Sends an ack packet
1336  * @device:	DRBD device.
1337  * @cmd:	Packet command code.
1338  * @sector:	sector, needs to be in big endian byte order
1339  * @blksize:	size in byte, needs to be in big endian byte order
1340  * @block_id:	Id, big endian byte order
1341  */
1342 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1343 			  u64 sector, u32 blksize, u64 block_id)
1344 {
1345 	struct drbd_socket *sock;
1346 	struct p_block_ack *p;
1347 
1348 	if (peer_device->device->state.conn < C_CONNECTED)
1349 		return -EIO;
1350 
1351 	sock = &peer_device->connection->meta;
1352 	p = drbd_prepare_command(peer_device, sock);
1353 	if (!p)
1354 		return -EIO;
1355 	p->sector = sector;
1356 	p->block_id = block_id;
1357 	p->blksize = blksize;
1358 	p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1359 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1360 }
1361 
1362 /* dp->sector and dp->block_id already/still in network byte order,
1363  * data_size is payload size according to dp->head,
1364  * and may need to be corrected for digest size. */
1365 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1366 		      struct p_data *dp, int data_size)
1367 {
1368 	if (peer_device->connection->peer_integrity_tfm)
1369 		data_size -= crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm);
1370 	_drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1371 		       dp->block_id);
1372 }
1373 
1374 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1375 		      struct p_block_req *rp)
1376 {
1377 	_drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1378 }
1379 
1380 /**
1381  * drbd_send_ack() - Sends an ack packet
1382  * @device:	DRBD device
1383  * @cmd:	packet command code
1384  * @peer_req:	peer request
1385  */
1386 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1387 		  struct drbd_peer_request *peer_req)
1388 {
1389 	return _drbd_send_ack(peer_device, cmd,
1390 			      cpu_to_be64(peer_req->i.sector),
1391 			      cpu_to_be32(peer_req->i.size),
1392 			      peer_req->block_id);
1393 }
1394 
1395 /* This function misuses the block_id field to signal if the blocks
1396  * are is sync or not. */
1397 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1398 		     sector_t sector, int blksize, u64 block_id)
1399 {
1400 	return _drbd_send_ack(peer_device, cmd,
1401 			      cpu_to_be64(sector),
1402 			      cpu_to_be32(blksize),
1403 			      cpu_to_be64(block_id));
1404 }
1405 
1406 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1407 			     struct drbd_peer_request *peer_req)
1408 {
1409 	struct drbd_socket *sock;
1410 	struct p_block_desc *p;
1411 
1412 	sock = &peer_device->connection->data;
1413 	p = drbd_prepare_command(peer_device, sock);
1414 	if (!p)
1415 		return -EIO;
1416 	p->sector = cpu_to_be64(peer_req->i.sector);
1417 	p->blksize = cpu_to_be32(peer_req->i.size);
1418 	p->pad = 0;
1419 	return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1420 }
1421 
1422 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1423 		       sector_t sector, int size, u64 block_id)
1424 {
1425 	struct drbd_socket *sock;
1426 	struct p_block_req *p;
1427 
1428 	sock = &peer_device->connection->data;
1429 	p = drbd_prepare_command(peer_device, sock);
1430 	if (!p)
1431 		return -EIO;
1432 	p->sector = cpu_to_be64(sector);
1433 	p->block_id = block_id;
1434 	p->blksize = cpu_to_be32(size);
1435 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1436 }
1437 
1438 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1439 			    void *digest, int digest_size, enum drbd_packet cmd)
1440 {
1441 	struct drbd_socket *sock;
1442 	struct p_block_req *p;
1443 
1444 	/* FIXME: Put the digest into the preallocated socket buffer.  */
1445 
1446 	sock = &peer_device->connection->data;
1447 	p = drbd_prepare_command(peer_device, sock);
1448 	if (!p)
1449 		return -EIO;
1450 	p->sector = cpu_to_be64(sector);
1451 	p->block_id = ID_SYNCER /* unused */;
1452 	p->blksize = cpu_to_be32(size);
1453 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1454 }
1455 
1456 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1457 {
1458 	struct drbd_socket *sock;
1459 	struct p_block_req *p;
1460 
1461 	sock = &peer_device->connection->data;
1462 	p = drbd_prepare_command(peer_device, sock);
1463 	if (!p)
1464 		return -EIO;
1465 	p->sector = cpu_to_be64(sector);
1466 	p->block_id = ID_SYNCER /* unused */;
1467 	p->blksize = cpu_to_be32(size);
1468 	return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1469 }
1470 
1471 /* called on sndtimeo
1472  * returns false if we should retry,
1473  * true if we think connection is dead
1474  */
1475 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1476 {
1477 	int drop_it;
1478 	/* long elapsed = (long)(jiffies - device->last_received); */
1479 
1480 	drop_it =   connection->meta.socket == sock
1481 		|| !connection->ack_receiver.task
1482 		|| get_t_state(&connection->ack_receiver) != RUNNING
1483 		|| connection->cstate < C_WF_REPORT_PARAMS;
1484 
1485 	if (drop_it)
1486 		return true;
1487 
1488 	drop_it = !--connection->ko_count;
1489 	if (!drop_it) {
1490 		drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1491 			 current->comm, current->pid, connection->ko_count);
1492 		request_ping(connection);
1493 	}
1494 
1495 	return drop_it; /* && (device->state == R_PRIMARY) */;
1496 }
1497 
1498 static void drbd_update_congested(struct drbd_connection *connection)
1499 {
1500 	struct sock *sk = connection->data.socket->sk;
1501 	if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1502 		set_bit(NET_CONGESTED, &connection->flags);
1503 }
1504 
1505 /* The idea of sendpage seems to be to put some kind of reference
1506  * to the page into the skb, and to hand it over to the NIC. In
1507  * this process get_page() gets called.
1508  *
1509  * As soon as the page was really sent over the network put_page()
1510  * gets called by some part of the network layer. [ NIC driver? ]
1511  *
1512  * [ get_page() / put_page() increment/decrement the count. If count
1513  *   reaches 0 the page will be freed. ]
1514  *
1515  * This works nicely with pages from FSs.
1516  * But this means that in protocol A we might signal IO completion too early!
1517  *
1518  * In order not to corrupt data during a resync we must make sure
1519  * that we do not reuse our own buffer pages (EEs) to early, therefore
1520  * we have the net_ee list.
1521  *
1522  * XFS seems to have problems, still, it submits pages with page_count == 0!
1523  * As a workaround, we disable sendpage on pages
1524  * with page_count == 0 or PageSlab.
1525  */
1526 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1527 			      int offset, size_t size, unsigned msg_flags)
1528 {
1529 	struct socket *socket;
1530 	void *addr;
1531 	int err;
1532 
1533 	socket = peer_device->connection->data.socket;
1534 	addr = kmap(page) + offset;
1535 	err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1536 	kunmap(page);
1537 	if (!err)
1538 		peer_device->device->send_cnt += size >> 9;
1539 	return err;
1540 }
1541 
1542 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1543 		    int offset, size_t size, unsigned msg_flags)
1544 {
1545 	struct socket *socket = peer_device->connection->data.socket;
1546 	int len = size;
1547 	int err = -EIO;
1548 
1549 	/* e.g. XFS meta- & log-data is in slab pages, which have a
1550 	 * page_count of 0 and/or have PageSlab() set.
1551 	 * we cannot use send_page for those, as that does get_page();
1552 	 * put_page(); and would cause either a VM_BUG directly, or
1553 	 * __page_cache_release a page that would actually still be referenced
1554 	 * by someone, leading to some obscure delayed Oops somewhere else. */
1555 	if (drbd_disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1556 		return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1557 
1558 	msg_flags |= MSG_NOSIGNAL;
1559 	drbd_update_congested(peer_device->connection);
1560 	do {
1561 		int sent;
1562 
1563 		sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
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 		/* REQ_OP_WRITE_SAME has only one segment */
1604 		if (bio_op(bio) == REQ_OP_WRITE_SAME)
1605 			break;
1606 	}
1607 	return 0;
1608 }
1609 
1610 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1611 {
1612 	struct bio_vec bvec;
1613 	struct bvec_iter iter;
1614 
1615 	/* hint all but last page with MSG_MORE */
1616 	bio_for_each_segment(bvec, bio, iter) {
1617 		int err;
1618 
1619 		err = _drbd_send_page(peer_device, bvec.bv_page,
1620 				      bvec.bv_offset, bvec.bv_len,
1621 				      bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1622 		if (err)
1623 			return err;
1624 		/* REQ_OP_WRITE_SAME has only one segment */
1625 		if (bio_op(bio) == REQ_OP_WRITE_SAME)
1626 			break;
1627 	}
1628 	return 0;
1629 }
1630 
1631 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1632 			    struct drbd_peer_request *peer_req)
1633 {
1634 	struct page *page = peer_req->pages;
1635 	unsigned len = peer_req->i.size;
1636 	int err;
1637 
1638 	/* hint all but last page with MSG_MORE */
1639 	page_chain_for_each(page) {
1640 		unsigned l = min_t(unsigned, len, PAGE_SIZE);
1641 
1642 		err = _drbd_send_page(peer_device, page, 0, l,
1643 				      page_chain_next(page) ? MSG_MORE : 0);
1644 		if (err)
1645 			return err;
1646 		len -= l;
1647 	}
1648 	return 0;
1649 }
1650 
1651 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1652 			     struct bio *bio)
1653 {
1654 	if (connection->agreed_pro_version >= 95)
1655 		return  (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1656 			(bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1657 			(bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1658 			(bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1659 			(bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1660 			(bio_op(bio) == REQ_OP_WRITE_ZEROES ?
1661 			  ((connection->agreed_features & DRBD_FF_WZEROES) ?
1662 			   (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0))
1663 			   : DP_DISCARD)
1664 			: 0);
1665 	else
1666 		return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1667 }
1668 
1669 /* Used to send write or TRIM aka REQ_OP_DISCARD requests
1670  * R_PRIMARY -> Peer	(P_DATA, P_TRIM)
1671  */
1672 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1673 {
1674 	struct drbd_device *device = peer_device->device;
1675 	struct drbd_socket *sock;
1676 	struct p_data *p;
1677 	struct p_wsame *wsame = NULL;
1678 	void *digest_out;
1679 	unsigned int dp_flags = 0;
1680 	int digest_size;
1681 	int err;
1682 
1683 	sock = &peer_device->connection->data;
1684 	p = drbd_prepare_command(peer_device, sock);
1685 	digest_size = peer_device->connection->integrity_tfm ?
1686 		      crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1687 
1688 	if (!p)
1689 		return -EIO;
1690 	p->sector = cpu_to_be64(req->i.sector);
1691 	p->block_id = (unsigned long)req;
1692 	p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1693 	dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1694 	if (device->state.conn >= C_SYNC_SOURCE &&
1695 	    device->state.conn <= C_PAUSED_SYNC_T)
1696 		dp_flags |= DP_MAY_SET_IN_SYNC;
1697 	if (peer_device->connection->agreed_pro_version >= 100) {
1698 		if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1699 			dp_flags |= DP_SEND_RECEIVE_ACK;
1700 		/* During resync, request an explicit write ack,
1701 		 * even in protocol != C */
1702 		if (req->rq_state & RQ_EXP_WRITE_ACK
1703 		|| (dp_flags & DP_MAY_SET_IN_SYNC))
1704 			dp_flags |= DP_SEND_WRITE_ACK;
1705 	}
1706 	p->dp_flags = cpu_to_be32(dp_flags);
1707 
1708 	if (dp_flags & (DP_DISCARD|DP_ZEROES)) {
1709 		enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM;
1710 		struct p_trim *t = (struct p_trim*)p;
1711 		t->size = cpu_to_be32(req->i.size);
1712 		err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0);
1713 		goto out;
1714 	}
1715 	if (dp_flags & DP_WSAME) {
1716 		/* this will only work if DRBD_FF_WSAME is set AND the
1717 		 * handshake agreed that all nodes and backend devices are
1718 		 * WRITE_SAME capable and agree on logical_block_size */
1719 		wsame = (struct p_wsame*)p;
1720 		digest_out = wsame + 1;
1721 		wsame->size = cpu_to_be32(req->i.size);
1722 	} else
1723 		digest_out = p + 1;
1724 
1725 	/* our digest is still only over the payload.
1726 	 * TRIM does not carry any payload. */
1727 	if (digest_size)
1728 		drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1729 	if (wsame) {
1730 		err =
1731 		    __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1732 				   sizeof(*wsame) + digest_size, NULL,
1733 				   bio_iovec(req->master_bio).bv_len);
1734 	} else
1735 		err =
1736 		    __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1737 				   sizeof(*p) + digest_size, NULL, req->i.size);
1738 	if (!err) {
1739 		/* For protocol A, we have to memcpy the payload into
1740 		 * socket buffers, as we may complete right away
1741 		 * as soon as we handed it over to tcp, at which point the data
1742 		 * pages may become invalid.
1743 		 *
1744 		 * For data-integrity enabled, we copy it as well, so we can be
1745 		 * sure that even if the bio pages may still be modified, it
1746 		 * won't change the data on the wire, thus if the digest checks
1747 		 * out ok after sending on this side, but does not fit on the
1748 		 * receiving side, we sure have detected corruption elsewhere.
1749 		 */
1750 		if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1751 			err = _drbd_send_bio(peer_device, req->master_bio);
1752 		else
1753 			err = _drbd_send_zc_bio(peer_device, req->master_bio);
1754 
1755 		/* double check digest, sometimes buffers have been modified in flight. */
1756 		if (digest_size > 0 && digest_size <= 64) {
1757 			/* 64 byte, 512 bit, is the largest digest size
1758 			 * currently supported in kernel crypto. */
1759 			unsigned char digest[64];
1760 			drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1761 			if (memcmp(p + 1, digest, digest_size)) {
1762 				drbd_warn(device,
1763 					"Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1764 					(unsigned long long)req->i.sector, req->i.size);
1765 			}
1766 		} /* else if (digest_size > 64) {
1767 		     ... Be noisy about digest too large ...
1768 		} */
1769 	}
1770 out:
1771 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1772 
1773 	return err;
1774 }
1775 
1776 /* answer packet, used to send data back for read requests:
1777  *  Peer       -> (diskless) R_PRIMARY   (P_DATA_REPLY)
1778  *  C_SYNC_SOURCE -> C_SYNC_TARGET         (P_RS_DATA_REPLY)
1779  */
1780 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1781 		    struct drbd_peer_request *peer_req)
1782 {
1783 	struct drbd_device *device = peer_device->device;
1784 	struct drbd_socket *sock;
1785 	struct p_data *p;
1786 	int err;
1787 	int digest_size;
1788 
1789 	sock = &peer_device->connection->data;
1790 	p = drbd_prepare_command(peer_device, sock);
1791 
1792 	digest_size = peer_device->connection->integrity_tfm ?
1793 		      crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1794 
1795 	if (!p)
1796 		return -EIO;
1797 	p->sector = cpu_to_be64(peer_req->i.sector);
1798 	p->block_id = peer_req->block_id;
1799 	p->seq_num = 0;  /* unused */
1800 	p->dp_flags = 0;
1801 	if (digest_size)
1802 		drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1803 	err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1804 	if (!err)
1805 		err = _drbd_send_zc_ee(peer_device, peer_req);
1806 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1807 
1808 	return err;
1809 }
1810 
1811 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1812 {
1813 	struct drbd_socket *sock;
1814 	struct p_block_desc *p;
1815 
1816 	sock = &peer_device->connection->data;
1817 	p = drbd_prepare_command(peer_device, sock);
1818 	if (!p)
1819 		return -EIO;
1820 	p->sector = cpu_to_be64(req->i.sector);
1821 	p->blksize = cpu_to_be32(req->i.size);
1822 	return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1823 }
1824 
1825 /*
1826   drbd_send distinguishes two cases:
1827 
1828   Packets sent via the data socket "sock"
1829   and packets sent via the meta data socket "msock"
1830 
1831 		    sock                      msock
1832   -----------------+-------------------------+------------------------------
1833   timeout           conf.timeout / 2          conf.timeout / 2
1834   timeout action    send a ping via msock     Abort communication
1835 					      and close all sockets
1836 */
1837 
1838 /*
1839  * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1840  */
1841 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1842 	      void *buf, size_t size, unsigned msg_flags)
1843 {
1844 	struct kvec iov = {.iov_base = buf, .iov_len = size};
1845 	struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL};
1846 	int rv, sent = 0;
1847 
1848 	if (!sock)
1849 		return -EBADR;
1850 
1851 	/* THINK  if (signal_pending) return ... ? */
1852 
1853 	iov_iter_kvec(&msg.msg_iter, WRITE, &iov, 1, size);
1854 
1855 	if (sock == connection->data.socket) {
1856 		rcu_read_lock();
1857 		connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1858 		rcu_read_unlock();
1859 		drbd_update_congested(connection);
1860 	}
1861 	do {
1862 		rv = sock_sendmsg(sock, &msg);
1863 		if (rv == -EAGAIN) {
1864 			if (we_should_drop_the_connection(connection, sock))
1865 				break;
1866 			else
1867 				continue;
1868 		}
1869 		if (rv == -EINTR) {
1870 			flush_signals(current);
1871 			rv = 0;
1872 		}
1873 		if (rv < 0)
1874 			break;
1875 		sent += rv;
1876 	} while (sent < size);
1877 
1878 	if (sock == connection->data.socket)
1879 		clear_bit(NET_CONGESTED, &connection->flags);
1880 
1881 	if (rv <= 0) {
1882 		if (rv != -EAGAIN) {
1883 			drbd_err(connection, "%s_sendmsg returned %d\n",
1884 				 sock == connection->meta.socket ? "msock" : "sock",
1885 				 rv);
1886 			conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1887 		} else
1888 			conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1889 	}
1890 
1891 	return sent;
1892 }
1893 
1894 /**
1895  * drbd_send_all  -  Send an entire buffer
1896  *
1897  * Returns 0 upon success and a negative error value otherwise.
1898  */
1899 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1900 		  size_t size, unsigned msg_flags)
1901 {
1902 	int err;
1903 
1904 	err = drbd_send(connection, sock, buffer, size, msg_flags);
1905 	if (err < 0)
1906 		return err;
1907 	if (err != size)
1908 		return -EIO;
1909 	return 0;
1910 }
1911 
1912 static int drbd_open(struct block_device *bdev, fmode_t mode)
1913 {
1914 	struct drbd_device *device = bdev->bd_disk->private_data;
1915 	unsigned long flags;
1916 	int rv = 0;
1917 
1918 	mutex_lock(&drbd_main_mutex);
1919 	spin_lock_irqsave(&device->resource->req_lock, flags);
1920 	/* to have a stable device->state.role
1921 	 * and no race with updating open_cnt */
1922 
1923 	if (device->state.role != R_PRIMARY) {
1924 		if (mode & FMODE_WRITE)
1925 			rv = -EROFS;
1926 		else if (!drbd_allow_oos)
1927 			rv = -EMEDIUMTYPE;
1928 	}
1929 
1930 	if (!rv)
1931 		device->open_cnt++;
1932 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
1933 	mutex_unlock(&drbd_main_mutex);
1934 
1935 	return rv;
1936 }
1937 
1938 static void drbd_release(struct gendisk *gd, fmode_t mode)
1939 {
1940 	struct drbd_device *device = gd->private_data;
1941 	mutex_lock(&drbd_main_mutex);
1942 	device->open_cnt--;
1943 	mutex_unlock(&drbd_main_mutex);
1944 }
1945 
1946 /* need to hold resource->req_lock */
1947 void drbd_queue_unplug(struct drbd_device *device)
1948 {
1949 	if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1950 		D_ASSERT(device, device->state.role == R_PRIMARY);
1951 		if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1952 			drbd_queue_work_if_unqueued(
1953 				&first_peer_device(device)->connection->sender_work,
1954 				&device->unplug_work);
1955 		}
1956 	}
1957 }
1958 
1959 static void drbd_set_defaults(struct drbd_device *device)
1960 {
1961 	/* Beware! The actual layout differs
1962 	 * between big endian and little endian */
1963 	device->state = (union drbd_dev_state) {
1964 		{ .role = R_SECONDARY,
1965 		  .peer = R_UNKNOWN,
1966 		  .conn = C_STANDALONE,
1967 		  .disk = D_DISKLESS,
1968 		  .pdsk = D_UNKNOWN,
1969 		} };
1970 }
1971 
1972 void drbd_init_set_defaults(struct drbd_device *device)
1973 {
1974 	/* the memset(,0,) did most of this.
1975 	 * note: only assignments, no allocation in here */
1976 
1977 	drbd_set_defaults(device);
1978 
1979 	atomic_set(&device->ap_bio_cnt, 0);
1980 	atomic_set(&device->ap_actlog_cnt, 0);
1981 	atomic_set(&device->ap_pending_cnt, 0);
1982 	atomic_set(&device->rs_pending_cnt, 0);
1983 	atomic_set(&device->unacked_cnt, 0);
1984 	atomic_set(&device->local_cnt, 0);
1985 	atomic_set(&device->pp_in_use_by_net, 0);
1986 	atomic_set(&device->rs_sect_in, 0);
1987 	atomic_set(&device->rs_sect_ev, 0);
1988 	atomic_set(&device->ap_in_flight, 0);
1989 	atomic_set(&device->md_io.in_use, 0);
1990 
1991 	mutex_init(&device->own_state_mutex);
1992 	device->state_mutex = &device->own_state_mutex;
1993 
1994 	spin_lock_init(&device->al_lock);
1995 	spin_lock_init(&device->peer_seq_lock);
1996 
1997 	INIT_LIST_HEAD(&device->active_ee);
1998 	INIT_LIST_HEAD(&device->sync_ee);
1999 	INIT_LIST_HEAD(&device->done_ee);
2000 	INIT_LIST_HEAD(&device->read_ee);
2001 	INIT_LIST_HEAD(&device->net_ee);
2002 	INIT_LIST_HEAD(&device->resync_reads);
2003 	INIT_LIST_HEAD(&device->resync_work.list);
2004 	INIT_LIST_HEAD(&device->unplug_work.list);
2005 	INIT_LIST_HEAD(&device->bm_io_work.w.list);
2006 	INIT_LIST_HEAD(&device->pending_master_completion[0]);
2007 	INIT_LIST_HEAD(&device->pending_master_completion[1]);
2008 	INIT_LIST_HEAD(&device->pending_completion[0]);
2009 	INIT_LIST_HEAD(&device->pending_completion[1]);
2010 
2011 	device->resync_work.cb  = w_resync_timer;
2012 	device->unplug_work.cb  = w_send_write_hint;
2013 	device->bm_io_work.w.cb = w_bitmap_io;
2014 
2015 	timer_setup(&device->resync_timer, resync_timer_fn, 0);
2016 	timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0);
2017 	timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0);
2018 	timer_setup(&device->request_timer, request_timer_fn, 0);
2019 
2020 	init_waitqueue_head(&device->misc_wait);
2021 	init_waitqueue_head(&device->state_wait);
2022 	init_waitqueue_head(&device->ee_wait);
2023 	init_waitqueue_head(&device->al_wait);
2024 	init_waitqueue_head(&device->seq_wait);
2025 
2026 	device->resync_wenr = LC_FREE;
2027 	device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2028 	device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2029 }
2030 
2031 static void _drbd_set_my_capacity(struct drbd_device *device, sector_t size)
2032 {
2033 	/* set_capacity(device->this_bdev->bd_disk, size); */
2034 	set_capacity(device->vdisk, size);
2035 	device->this_bdev->bd_inode->i_size = (loff_t)size << 9;
2036 }
2037 
2038 void drbd_set_my_capacity(struct drbd_device *device, sector_t size)
2039 {
2040 	char ppb[10];
2041 	_drbd_set_my_capacity(device, size);
2042 	drbd_info(device, "size = %s (%llu KB)\n",
2043 		ppsize(ppb, size>>1), (unsigned long long)size>>1);
2044 }
2045 
2046 void drbd_device_cleanup(struct drbd_device *device)
2047 {
2048 	int i;
2049 	if (first_peer_device(device)->connection->receiver.t_state != NONE)
2050 		drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2051 				first_peer_device(device)->connection->receiver.t_state);
2052 
2053 	device->al_writ_cnt  =
2054 	device->bm_writ_cnt  =
2055 	device->read_cnt     =
2056 	device->recv_cnt     =
2057 	device->send_cnt     =
2058 	device->writ_cnt     =
2059 	device->p_size       =
2060 	device->rs_start     =
2061 	device->rs_total     =
2062 	device->rs_failed    = 0;
2063 	device->rs_last_events = 0;
2064 	device->rs_last_sect_ev = 0;
2065 	for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2066 		device->rs_mark_left[i] = 0;
2067 		device->rs_mark_time[i] = 0;
2068 	}
2069 	D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2070 
2071 	_drbd_set_my_capacity(device, 0);
2072 	if (device->bitmap) {
2073 		/* maybe never allocated. */
2074 		drbd_bm_resize(device, 0, 1);
2075 		drbd_bm_cleanup(device);
2076 	}
2077 
2078 	drbd_backing_dev_free(device, device->ldev);
2079 	device->ldev = NULL;
2080 
2081 	clear_bit(AL_SUSPENDED, &device->flags);
2082 
2083 	D_ASSERT(device, list_empty(&device->active_ee));
2084 	D_ASSERT(device, list_empty(&device->sync_ee));
2085 	D_ASSERT(device, list_empty(&device->done_ee));
2086 	D_ASSERT(device, list_empty(&device->read_ee));
2087 	D_ASSERT(device, list_empty(&device->net_ee));
2088 	D_ASSERT(device, list_empty(&device->resync_reads));
2089 	D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2090 	D_ASSERT(device, list_empty(&device->resync_work.list));
2091 	D_ASSERT(device, list_empty(&device->unplug_work.list));
2092 
2093 	drbd_set_defaults(device);
2094 }
2095 
2096 
2097 static void drbd_destroy_mempools(void)
2098 {
2099 	struct page *page;
2100 
2101 	while (drbd_pp_pool) {
2102 		page = drbd_pp_pool;
2103 		drbd_pp_pool = (struct page *)page_private(page);
2104 		__free_page(page);
2105 		drbd_pp_vacant--;
2106 	}
2107 
2108 	/* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2109 
2110 	bioset_exit(&drbd_io_bio_set);
2111 	bioset_exit(&drbd_md_io_bio_set);
2112 	mempool_exit(&drbd_md_io_page_pool);
2113 	mempool_exit(&drbd_ee_mempool);
2114 	mempool_exit(&drbd_request_mempool);
2115 	kmem_cache_destroy(drbd_ee_cache);
2116 	kmem_cache_destroy(drbd_request_cache);
2117 	kmem_cache_destroy(drbd_bm_ext_cache);
2118 	kmem_cache_destroy(drbd_al_ext_cache);
2119 
2120 	drbd_ee_cache        = NULL;
2121 	drbd_request_cache   = NULL;
2122 	drbd_bm_ext_cache    = NULL;
2123 	drbd_al_ext_cache    = NULL;
2124 
2125 	return;
2126 }
2127 
2128 static int drbd_create_mempools(void)
2129 {
2130 	struct page *page;
2131 	const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2132 	int i, ret;
2133 
2134 	/* caches */
2135 	drbd_request_cache = kmem_cache_create(
2136 		"drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2137 	if (drbd_request_cache == NULL)
2138 		goto Enomem;
2139 
2140 	drbd_ee_cache = kmem_cache_create(
2141 		"drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2142 	if (drbd_ee_cache == NULL)
2143 		goto Enomem;
2144 
2145 	drbd_bm_ext_cache = kmem_cache_create(
2146 		"drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2147 	if (drbd_bm_ext_cache == NULL)
2148 		goto Enomem;
2149 
2150 	drbd_al_ext_cache = kmem_cache_create(
2151 		"drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2152 	if (drbd_al_ext_cache == NULL)
2153 		goto Enomem;
2154 
2155 	/* mempools */
2156 	ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0);
2157 	if (ret)
2158 		goto Enomem;
2159 
2160 	ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0,
2161 			  BIOSET_NEED_BVECS);
2162 	if (ret)
2163 		goto Enomem;
2164 
2165 	ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0);
2166 	if (ret)
2167 		goto Enomem;
2168 
2169 	ret = mempool_init_slab_pool(&drbd_request_mempool, number,
2170 				     drbd_request_cache);
2171 	if (ret)
2172 		goto Enomem;
2173 
2174 	ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache);
2175 	if (ret)
2176 		goto Enomem;
2177 
2178 	/* drbd's page pool */
2179 	spin_lock_init(&drbd_pp_lock);
2180 
2181 	for (i = 0; i < number; i++) {
2182 		page = alloc_page(GFP_HIGHUSER);
2183 		if (!page)
2184 			goto Enomem;
2185 		set_page_private(page, (unsigned long)drbd_pp_pool);
2186 		drbd_pp_pool = page;
2187 	}
2188 	drbd_pp_vacant = number;
2189 
2190 	return 0;
2191 
2192 Enomem:
2193 	drbd_destroy_mempools(); /* in case we allocated some */
2194 	return -ENOMEM;
2195 }
2196 
2197 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2198 {
2199 	int rr;
2200 
2201 	rr = drbd_free_peer_reqs(device, &device->active_ee);
2202 	if (rr)
2203 		drbd_err(device, "%d EEs in active list found!\n", rr);
2204 
2205 	rr = drbd_free_peer_reqs(device, &device->sync_ee);
2206 	if (rr)
2207 		drbd_err(device, "%d EEs in sync list found!\n", rr);
2208 
2209 	rr = drbd_free_peer_reqs(device, &device->read_ee);
2210 	if (rr)
2211 		drbd_err(device, "%d EEs in read list found!\n", rr);
2212 
2213 	rr = drbd_free_peer_reqs(device, &device->done_ee);
2214 	if (rr)
2215 		drbd_err(device, "%d EEs in done list found!\n", rr);
2216 
2217 	rr = drbd_free_peer_reqs(device, &device->net_ee);
2218 	if (rr)
2219 		drbd_err(device, "%d EEs in net list found!\n", rr);
2220 }
2221 
2222 /* caution. no locking. */
2223 void drbd_destroy_device(struct kref *kref)
2224 {
2225 	struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2226 	struct drbd_resource *resource = device->resource;
2227 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2228 
2229 	del_timer_sync(&device->request_timer);
2230 
2231 	/* paranoia asserts */
2232 	D_ASSERT(device, device->open_cnt == 0);
2233 	/* end paranoia asserts */
2234 
2235 	/* cleanup stuff that may have been allocated during
2236 	 * device (re-)configuration or state changes */
2237 
2238 	if (device->this_bdev)
2239 		bdput(device->this_bdev);
2240 
2241 	drbd_backing_dev_free(device, device->ldev);
2242 	device->ldev = NULL;
2243 
2244 	drbd_release_all_peer_reqs(device);
2245 
2246 	lc_destroy(device->act_log);
2247 	lc_destroy(device->resync);
2248 
2249 	kfree(device->p_uuid);
2250 	/* device->p_uuid = NULL; */
2251 
2252 	if (device->bitmap) /* should no longer be there. */
2253 		drbd_bm_cleanup(device);
2254 	__free_page(device->md_io.page);
2255 	put_disk(device->vdisk);
2256 	blk_cleanup_queue(device->rq_queue);
2257 	kfree(device->rs_plan_s);
2258 
2259 	/* not for_each_connection(connection, resource):
2260 	 * those may have been cleaned up and disassociated already.
2261 	 */
2262 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2263 		kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2264 		kfree(peer_device);
2265 	}
2266 	memset(device, 0xfd, sizeof(*device));
2267 	kfree(device);
2268 	kref_put(&resource->kref, drbd_destroy_resource);
2269 }
2270 
2271 /* One global retry thread, if we need to push back some bio and have it
2272  * reinserted through our make request function.
2273  */
2274 static struct retry_worker {
2275 	struct workqueue_struct *wq;
2276 	struct work_struct worker;
2277 
2278 	spinlock_t lock;
2279 	struct list_head writes;
2280 } retry;
2281 
2282 static void do_retry(struct work_struct *ws)
2283 {
2284 	struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2285 	LIST_HEAD(writes);
2286 	struct drbd_request *req, *tmp;
2287 
2288 	spin_lock_irq(&retry->lock);
2289 	list_splice_init(&retry->writes, &writes);
2290 	spin_unlock_irq(&retry->lock);
2291 
2292 	list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2293 		struct drbd_device *device = req->device;
2294 		struct bio *bio = req->master_bio;
2295 		unsigned long start_jif = req->start_jif;
2296 		bool expected;
2297 
2298 		expected =
2299 			expect(atomic_read(&req->completion_ref) == 0) &&
2300 			expect(req->rq_state & RQ_POSTPONED) &&
2301 			expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2302 				(req->rq_state & RQ_LOCAL_ABORTED) != 0);
2303 
2304 		if (!expected)
2305 			drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2306 				req, atomic_read(&req->completion_ref),
2307 				req->rq_state);
2308 
2309 		/* We still need to put one kref associated with the
2310 		 * "completion_ref" going zero in the code path that queued it
2311 		 * here.  The request object may still be referenced by a
2312 		 * frozen local req->private_bio, in case we force-detached.
2313 		 */
2314 		kref_put(&req->kref, drbd_req_destroy);
2315 
2316 		/* A single suspended or otherwise blocking device may stall
2317 		 * all others as well.  Fortunately, this code path is to
2318 		 * recover from a situation that "should not happen":
2319 		 * concurrent writes in multi-primary setup.
2320 		 * In a "normal" lifecycle, this workqueue is supposed to be
2321 		 * destroyed without ever doing anything.
2322 		 * If it turns out to be an issue anyways, we can do per
2323 		 * resource (replication group) or per device (minor) retry
2324 		 * workqueues instead.
2325 		 */
2326 
2327 		/* We are not just doing generic_make_request(),
2328 		 * as we want to keep the start_time information. */
2329 		inc_ap_bio(device);
2330 		__drbd_make_request(device, bio, start_jif);
2331 	}
2332 }
2333 
2334 /* called via drbd_req_put_completion_ref(),
2335  * holds resource->req_lock */
2336 void drbd_restart_request(struct drbd_request *req)
2337 {
2338 	unsigned long flags;
2339 	spin_lock_irqsave(&retry.lock, flags);
2340 	list_move_tail(&req->tl_requests, &retry.writes);
2341 	spin_unlock_irqrestore(&retry.lock, flags);
2342 
2343 	/* Drop the extra reference that would otherwise
2344 	 * have been dropped by complete_master_bio.
2345 	 * do_retry() needs to grab a new one. */
2346 	dec_ap_bio(req->device);
2347 
2348 	queue_work(retry.wq, &retry.worker);
2349 }
2350 
2351 void drbd_destroy_resource(struct kref *kref)
2352 {
2353 	struct drbd_resource *resource =
2354 		container_of(kref, struct drbd_resource, kref);
2355 
2356 	idr_destroy(&resource->devices);
2357 	free_cpumask_var(resource->cpu_mask);
2358 	kfree(resource->name);
2359 	memset(resource, 0xf2, sizeof(*resource));
2360 	kfree(resource);
2361 }
2362 
2363 void drbd_free_resource(struct drbd_resource *resource)
2364 {
2365 	struct drbd_connection *connection, *tmp;
2366 
2367 	for_each_connection_safe(connection, tmp, resource) {
2368 		list_del(&connection->connections);
2369 		drbd_debugfs_connection_cleanup(connection);
2370 		kref_put(&connection->kref, drbd_destroy_connection);
2371 	}
2372 	drbd_debugfs_resource_cleanup(resource);
2373 	kref_put(&resource->kref, drbd_destroy_resource);
2374 }
2375 
2376 static void drbd_cleanup(void)
2377 {
2378 	unsigned int i;
2379 	struct drbd_device *device;
2380 	struct drbd_resource *resource, *tmp;
2381 
2382 	/* first remove proc,
2383 	 * drbdsetup uses it's presence to detect
2384 	 * whether DRBD is loaded.
2385 	 * If we would get stuck in proc removal,
2386 	 * but have netlink already deregistered,
2387 	 * some drbdsetup commands may wait forever
2388 	 * for an answer.
2389 	 */
2390 	if (drbd_proc)
2391 		remove_proc_entry("drbd", NULL);
2392 
2393 	if (retry.wq)
2394 		destroy_workqueue(retry.wq);
2395 
2396 	drbd_genl_unregister();
2397 
2398 	idr_for_each_entry(&drbd_devices, device, i)
2399 		drbd_delete_device(device);
2400 
2401 	/* not _rcu since, no other updater anymore. Genl already unregistered */
2402 	for_each_resource_safe(resource, tmp, &drbd_resources) {
2403 		list_del(&resource->resources);
2404 		drbd_free_resource(resource);
2405 	}
2406 
2407 	drbd_debugfs_cleanup();
2408 
2409 	drbd_destroy_mempools();
2410 	unregister_blkdev(DRBD_MAJOR, "drbd");
2411 
2412 	idr_destroy(&drbd_devices);
2413 
2414 	pr_info("module cleanup done.\n");
2415 }
2416 
2417 /**
2418  * drbd_congested() - Callback for the flusher thread
2419  * @congested_data:	User data
2420  * @bdi_bits:		Bits the BDI flusher thread is currently interested in
2421  *
2422  * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested.
2423  */
2424 static int drbd_congested(void *congested_data, int bdi_bits)
2425 {
2426 	struct drbd_device *device = congested_data;
2427 	struct request_queue *q;
2428 	char reason = '-';
2429 	int r = 0;
2430 
2431 	if (!may_inc_ap_bio(device)) {
2432 		/* DRBD has frozen IO */
2433 		r = bdi_bits;
2434 		reason = 'd';
2435 		goto out;
2436 	}
2437 
2438 	if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
2439 		r |= (1 << WB_async_congested);
2440 		/* Without good local data, we would need to read from remote,
2441 		 * and that would need the worker thread as well, which is
2442 		 * currently blocked waiting for that usermode helper to
2443 		 * finish.
2444 		 */
2445 		if (!get_ldev_if_state(device, D_UP_TO_DATE))
2446 			r |= (1 << WB_sync_congested);
2447 		else
2448 			put_ldev(device);
2449 		r &= bdi_bits;
2450 		reason = 'c';
2451 		goto out;
2452 	}
2453 
2454 	if (get_ldev(device)) {
2455 		q = bdev_get_queue(device->ldev->backing_bdev);
2456 		r = bdi_congested(q->backing_dev_info, bdi_bits);
2457 		put_ldev(device);
2458 		if (r)
2459 			reason = 'b';
2460 	}
2461 
2462 	if (bdi_bits & (1 << WB_async_congested) &&
2463 	    test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
2464 		r |= (1 << WB_async_congested);
2465 		reason = reason == 'b' ? 'a' : 'n';
2466 	}
2467 
2468 out:
2469 	device->congestion_reason = reason;
2470 	return r;
2471 }
2472 
2473 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2474 {
2475 	spin_lock_init(&wq->q_lock);
2476 	INIT_LIST_HEAD(&wq->q);
2477 	init_waitqueue_head(&wq->q_wait);
2478 }
2479 
2480 struct completion_work {
2481 	struct drbd_work w;
2482 	struct completion done;
2483 };
2484 
2485 static int w_complete(struct drbd_work *w, int cancel)
2486 {
2487 	struct completion_work *completion_work =
2488 		container_of(w, struct completion_work, w);
2489 
2490 	complete(&completion_work->done);
2491 	return 0;
2492 }
2493 
2494 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2495 {
2496 	struct completion_work completion_work;
2497 
2498 	completion_work.w.cb = w_complete;
2499 	init_completion(&completion_work.done);
2500 	drbd_queue_work(work_queue, &completion_work.w);
2501 	wait_for_completion(&completion_work.done);
2502 }
2503 
2504 struct drbd_resource *drbd_find_resource(const char *name)
2505 {
2506 	struct drbd_resource *resource;
2507 
2508 	if (!name || !name[0])
2509 		return NULL;
2510 
2511 	rcu_read_lock();
2512 	for_each_resource_rcu(resource, &drbd_resources) {
2513 		if (!strcmp(resource->name, name)) {
2514 			kref_get(&resource->kref);
2515 			goto found;
2516 		}
2517 	}
2518 	resource = NULL;
2519 found:
2520 	rcu_read_unlock();
2521 	return resource;
2522 }
2523 
2524 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2525 				     void *peer_addr, int peer_addr_len)
2526 {
2527 	struct drbd_resource *resource;
2528 	struct drbd_connection *connection;
2529 
2530 	rcu_read_lock();
2531 	for_each_resource_rcu(resource, &drbd_resources) {
2532 		for_each_connection_rcu(connection, resource) {
2533 			if (connection->my_addr_len == my_addr_len &&
2534 			    connection->peer_addr_len == peer_addr_len &&
2535 			    !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2536 			    !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2537 				kref_get(&connection->kref);
2538 				goto found;
2539 			}
2540 		}
2541 	}
2542 	connection = NULL;
2543 found:
2544 	rcu_read_unlock();
2545 	return connection;
2546 }
2547 
2548 static int drbd_alloc_socket(struct drbd_socket *socket)
2549 {
2550 	socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2551 	if (!socket->rbuf)
2552 		return -ENOMEM;
2553 	socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2554 	if (!socket->sbuf)
2555 		return -ENOMEM;
2556 	return 0;
2557 }
2558 
2559 static void drbd_free_socket(struct drbd_socket *socket)
2560 {
2561 	free_page((unsigned long) socket->sbuf);
2562 	free_page((unsigned long) socket->rbuf);
2563 }
2564 
2565 void conn_free_crypto(struct drbd_connection *connection)
2566 {
2567 	drbd_free_sock(connection);
2568 
2569 	crypto_free_shash(connection->csums_tfm);
2570 	crypto_free_shash(connection->verify_tfm);
2571 	crypto_free_shash(connection->cram_hmac_tfm);
2572 	crypto_free_shash(connection->integrity_tfm);
2573 	crypto_free_shash(connection->peer_integrity_tfm);
2574 	kfree(connection->int_dig_in);
2575 	kfree(connection->int_dig_vv);
2576 
2577 	connection->csums_tfm = NULL;
2578 	connection->verify_tfm = NULL;
2579 	connection->cram_hmac_tfm = NULL;
2580 	connection->integrity_tfm = NULL;
2581 	connection->peer_integrity_tfm = NULL;
2582 	connection->int_dig_in = NULL;
2583 	connection->int_dig_vv = NULL;
2584 }
2585 
2586 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2587 {
2588 	struct drbd_connection *connection;
2589 	cpumask_var_t new_cpu_mask;
2590 	int err;
2591 
2592 	if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2593 		return -ENOMEM;
2594 
2595 	/* silently ignore cpu mask on UP kernel */
2596 	if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2597 		err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2598 				   cpumask_bits(new_cpu_mask), nr_cpu_ids);
2599 		if (err == -EOVERFLOW) {
2600 			/* So what. mask it out. */
2601 			cpumask_var_t tmp_cpu_mask;
2602 			if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2603 				cpumask_setall(tmp_cpu_mask);
2604 				cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2605 				drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2606 					res_opts->cpu_mask,
2607 					strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2608 					nr_cpu_ids);
2609 				free_cpumask_var(tmp_cpu_mask);
2610 				err = 0;
2611 			}
2612 		}
2613 		if (err) {
2614 			drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2615 			/* retcode = ERR_CPU_MASK_PARSE; */
2616 			goto fail;
2617 		}
2618 	}
2619 	resource->res_opts = *res_opts;
2620 	if (cpumask_empty(new_cpu_mask))
2621 		drbd_calc_cpu_mask(&new_cpu_mask);
2622 	if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2623 		cpumask_copy(resource->cpu_mask, new_cpu_mask);
2624 		for_each_connection_rcu(connection, resource) {
2625 			connection->receiver.reset_cpu_mask = 1;
2626 			connection->ack_receiver.reset_cpu_mask = 1;
2627 			connection->worker.reset_cpu_mask = 1;
2628 		}
2629 	}
2630 	err = 0;
2631 
2632 fail:
2633 	free_cpumask_var(new_cpu_mask);
2634 	return err;
2635 
2636 }
2637 
2638 struct drbd_resource *drbd_create_resource(const char *name)
2639 {
2640 	struct drbd_resource *resource;
2641 
2642 	resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2643 	if (!resource)
2644 		goto fail;
2645 	resource->name = kstrdup(name, GFP_KERNEL);
2646 	if (!resource->name)
2647 		goto fail_free_resource;
2648 	if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2649 		goto fail_free_name;
2650 	kref_init(&resource->kref);
2651 	idr_init(&resource->devices);
2652 	INIT_LIST_HEAD(&resource->connections);
2653 	resource->write_ordering = WO_BDEV_FLUSH;
2654 	list_add_tail_rcu(&resource->resources, &drbd_resources);
2655 	mutex_init(&resource->conf_update);
2656 	mutex_init(&resource->adm_mutex);
2657 	spin_lock_init(&resource->req_lock);
2658 	drbd_debugfs_resource_add(resource);
2659 	return resource;
2660 
2661 fail_free_name:
2662 	kfree(resource->name);
2663 fail_free_resource:
2664 	kfree(resource);
2665 fail:
2666 	return NULL;
2667 }
2668 
2669 /* caller must be under adm_mutex */
2670 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2671 {
2672 	struct drbd_resource *resource;
2673 	struct drbd_connection *connection;
2674 
2675 	connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2676 	if (!connection)
2677 		return NULL;
2678 
2679 	if (drbd_alloc_socket(&connection->data))
2680 		goto fail;
2681 	if (drbd_alloc_socket(&connection->meta))
2682 		goto fail;
2683 
2684 	connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2685 	if (!connection->current_epoch)
2686 		goto fail;
2687 
2688 	INIT_LIST_HEAD(&connection->transfer_log);
2689 
2690 	INIT_LIST_HEAD(&connection->current_epoch->list);
2691 	connection->epochs = 1;
2692 	spin_lock_init(&connection->epoch_lock);
2693 
2694 	connection->send.seen_any_write_yet = false;
2695 	connection->send.current_epoch_nr = 0;
2696 	connection->send.current_epoch_writes = 0;
2697 
2698 	resource = drbd_create_resource(name);
2699 	if (!resource)
2700 		goto fail;
2701 
2702 	connection->cstate = C_STANDALONE;
2703 	mutex_init(&connection->cstate_mutex);
2704 	init_waitqueue_head(&connection->ping_wait);
2705 	idr_init(&connection->peer_devices);
2706 
2707 	drbd_init_workqueue(&connection->sender_work);
2708 	mutex_init(&connection->data.mutex);
2709 	mutex_init(&connection->meta.mutex);
2710 
2711 	drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2712 	connection->receiver.connection = connection;
2713 	drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2714 	connection->worker.connection = connection;
2715 	drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2716 	connection->ack_receiver.connection = connection;
2717 
2718 	kref_init(&connection->kref);
2719 
2720 	connection->resource = resource;
2721 
2722 	if (set_resource_options(resource, res_opts))
2723 		goto fail_resource;
2724 
2725 	kref_get(&resource->kref);
2726 	list_add_tail_rcu(&connection->connections, &resource->connections);
2727 	drbd_debugfs_connection_add(connection);
2728 	return connection;
2729 
2730 fail_resource:
2731 	list_del(&resource->resources);
2732 	drbd_free_resource(resource);
2733 fail:
2734 	kfree(connection->current_epoch);
2735 	drbd_free_socket(&connection->meta);
2736 	drbd_free_socket(&connection->data);
2737 	kfree(connection);
2738 	return NULL;
2739 }
2740 
2741 void drbd_destroy_connection(struct kref *kref)
2742 {
2743 	struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2744 	struct drbd_resource *resource = connection->resource;
2745 
2746 	if (atomic_read(&connection->current_epoch->epoch_size) !=  0)
2747 		drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2748 	kfree(connection->current_epoch);
2749 
2750 	idr_destroy(&connection->peer_devices);
2751 
2752 	drbd_free_socket(&connection->meta);
2753 	drbd_free_socket(&connection->data);
2754 	kfree(connection->int_dig_in);
2755 	kfree(connection->int_dig_vv);
2756 	memset(connection, 0xfc, sizeof(*connection));
2757 	kfree(connection);
2758 	kref_put(&resource->kref, drbd_destroy_resource);
2759 }
2760 
2761 static int init_submitter(struct drbd_device *device)
2762 {
2763 	/* opencoded create_singlethread_workqueue(),
2764 	 * to be able to say "drbd%d", ..., minor */
2765 	device->submit.wq =
2766 		alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2767 	if (!device->submit.wq)
2768 		return -ENOMEM;
2769 
2770 	INIT_WORK(&device->submit.worker, do_submit);
2771 	INIT_LIST_HEAD(&device->submit.writes);
2772 	return 0;
2773 }
2774 
2775 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2776 {
2777 	struct drbd_resource *resource = adm_ctx->resource;
2778 	struct drbd_connection *connection;
2779 	struct drbd_device *device;
2780 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2781 	struct gendisk *disk;
2782 	struct request_queue *q;
2783 	int id;
2784 	int vnr = adm_ctx->volume;
2785 	enum drbd_ret_code err = ERR_NOMEM;
2786 
2787 	device = minor_to_device(minor);
2788 	if (device)
2789 		return ERR_MINOR_OR_VOLUME_EXISTS;
2790 
2791 	/* GFP_KERNEL, we are outside of all write-out paths */
2792 	device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2793 	if (!device)
2794 		return ERR_NOMEM;
2795 	kref_init(&device->kref);
2796 
2797 	kref_get(&resource->kref);
2798 	device->resource = resource;
2799 	device->minor = minor;
2800 	device->vnr = vnr;
2801 
2802 	drbd_init_set_defaults(device);
2803 
2804 	q = blk_alloc_queue(drbd_make_request, NUMA_NO_NODE);
2805 	if (!q)
2806 		goto out_no_q;
2807 	device->rq_queue = q;
2808 	q->queuedata   = device;
2809 
2810 	disk = alloc_disk(1);
2811 	if (!disk)
2812 		goto out_no_disk;
2813 	device->vdisk = disk;
2814 
2815 	set_disk_ro(disk, true);
2816 
2817 	disk->queue = q;
2818 	disk->major = DRBD_MAJOR;
2819 	disk->first_minor = minor;
2820 	disk->fops = &drbd_ops;
2821 	sprintf(disk->disk_name, "drbd%d", minor);
2822 	disk->private_data = device;
2823 
2824 	device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2825 	/* we have no partitions. we contain only ourselves. */
2826 	device->this_bdev->bd_contains = device->this_bdev;
2827 
2828 	q->backing_dev_info->congested_fn = drbd_congested;
2829 	q->backing_dev_info->congested_data = device;
2830 
2831 	blk_queue_write_cache(q, true, true);
2832 	/* Setting the max_hw_sectors to an odd value of 8kibyte here
2833 	   This triggers a max_bio_size message upon first attach or connect */
2834 	blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2835 
2836 	device->md_io.page = alloc_page(GFP_KERNEL);
2837 	if (!device->md_io.page)
2838 		goto out_no_io_page;
2839 
2840 	if (drbd_bm_init(device))
2841 		goto out_no_bitmap;
2842 	device->read_requests = RB_ROOT;
2843 	device->write_requests = RB_ROOT;
2844 
2845 	id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2846 	if (id < 0) {
2847 		if (id == -ENOSPC)
2848 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2849 		goto out_no_minor_idr;
2850 	}
2851 	kref_get(&device->kref);
2852 
2853 	id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2854 	if (id < 0) {
2855 		if (id == -ENOSPC)
2856 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2857 		goto out_idr_remove_minor;
2858 	}
2859 	kref_get(&device->kref);
2860 
2861 	INIT_LIST_HEAD(&device->peer_devices);
2862 	INIT_LIST_HEAD(&device->pending_bitmap_io);
2863 	for_each_connection(connection, resource) {
2864 		peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2865 		if (!peer_device)
2866 			goto out_idr_remove_from_resource;
2867 		peer_device->connection = connection;
2868 		peer_device->device = device;
2869 
2870 		list_add(&peer_device->peer_devices, &device->peer_devices);
2871 		kref_get(&device->kref);
2872 
2873 		id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2874 		if (id < 0) {
2875 			if (id == -ENOSPC)
2876 				err = ERR_INVALID_REQUEST;
2877 			goto out_idr_remove_from_resource;
2878 		}
2879 		kref_get(&connection->kref);
2880 		INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2881 	}
2882 
2883 	if (init_submitter(device)) {
2884 		err = ERR_NOMEM;
2885 		goto out_idr_remove_vol;
2886 	}
2887 
2888 	add_disk(disk);
2889 
2890 	/* inherit the connection state */
2891 	device->state.conn = first_connection(resource)->cstate;
2892 	if (device->state.conn == C_WF_REPORT_PARAMS) {
2893 		for_each_peer_device(peer_device, device)
2894 			drbd_connected(peer_device);
2895 	}
2896 	/* move to create_peer_device() */
2897 	for_each_peer_device(peer_device, device)
2898 		drbd_debugfs_peer_device_add(peer_device);
2899 	drbd_debugfs_device_add(device);
2900 	return NO_ERROR;
2901 
2902 out_idr_remove_vol:
2903 	idr_remove(&connection->peer_devices, vnr);
2904 out_idr_remove_from_resource:
2905 	for_each_connection(connection, resource) {
2906 		peer_device = idr_remove(&connection->peer_devices, vnr);
2907 		if (peer_device)
2908 			kref_put(&connection->kref, drbd_destroy_connection);
2909 	}
2910 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2911 		list_del(&peer_device->peer_devices);
2912 		kfree(peer_device);
2913 	}
2914 	idr_remove(&resource->devices, vnr);
2915 out_idr_remove_minor:
2916 	idr_remove(&drbd_devices, minor);
2917 	synchronize_rcu();
2918 out_no_minor_idr:
2919 	drbd_bm_cleanup(device);
2920 out_no_bitmap:
2921 	__free_page(device->md_io.page);
2922 out_no_io_page:
2923 	put_disk(disk);
2924 out_no_disk:
2925 	blk_cleanup_queue(q);
2926 out_no_q:
2927 	kref_put(&resource->kref, drbd_destroy_resource);
2928 	kfree(device);
2929 	return err;
2930 }
2931 
2932 void drbd_delete_device(struct drbd_device *device)
2933 {
2934 	struct drbd_resource *resource = device->resource;
2935 	struct drbd_connection *connection;
2936 	struct drbd_peer_device *peer_device;
2937 
2938 	/* move to free_peer_device() */
2939 	for_each_peer_device(peer_device, device)
2940 		drbd_debugfs_peer_device_cleanup(peer_device);
2941 	drbd_debugfs_device_cleanup(device);
2942 	for_each_connection(connection, resource) {
2943 		idr_remove(&connection->peer_devices, device->vnr);
2944 		kref_put(&device->kref, drbd_destroy_device);
2945 	}
2946 	idr_remove(&resource->devices, device->vnr);
2947 	kref_put(&device->kref, drbd_destroy_device);
2948 	idr_remove(&drbd_devices, device_to_minor(device));
2949 	kref_put(&device->kref, drbd_destroy_device);
2950 	del_gendisk(device->vdisk);
2951 	synchronize_rcu();
2952 	kref_put(&device->kref, drbd_destroy_device);
2953 }
2954 
2955 static int __init drbd_init(void)
2956 {
2957 	int err;
2958 
2959 	if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2960 		pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2961 #ifdef MODULE
2962 		return -EINVAL;
2963 #else
2964 		drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2965 #endif
2966 	}
2967 
2968 	err = register_blkdev(DRBD_MAJOR, "drbd");
2969 	if (err) {
2970 		pr_err("unable to register block device major %d\n",
2971 		       DRBD_MAJOR);
2972 		return err;
2973 	}
2974 
2975 	/*
2976 	 * allocate all necessary structs
2977 	 */
2978 	init_waitqueue_head(&drbd_pp_wait);
2979 
2980 	drbd_proc = NULL; /* play safe for drbd_cleanup */
2981 	idr_init(&drbd_devices);
2982 
2983 	mutex_init(&resources_mutex);
2984 	INIT_LIST_HEAD(&drbd_resources);
2985 
2986 	err = drbd_genl_register();
2987 	if (err) {
2988 		pr_err("unable to register generic netlink family\n");
2989 		goto fail;
2990 	}
2991 
2992 	err = drbd_create_mempools();
2993 	if (err)
2994 		goto fail;
2995 
2996 	err = -ENOMEM;
2997 	drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show);
2998 	if (!drbd_proc)	{
2999 		pr_err("unable to register proc file\n");
3000 		goto fail;
3001 	}
3002 
3003 	retry.wq = create_singlethread_workqueue("drbd-reissue");
3004 	if (!retry.wq) {
3005 		pr_err("unable to create retry workqueue\n");
3006 		goto fail;
3007 	}
3008 	INIT_WORK(&retry.worker, do_retry);
3009 	spin_lock_init(&retry.lock);
3010 	INIT_LIST_HEAD(&retry.writes);
3011 
3012 	drbd_debugfs_init();
3013 
3014 	pr_info("initialized. "
3015 	       "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
3016 	       API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
3017 	pr_info("%s\n", drbd_buildtag());
3018 	pr_info("registered as block device major %d\n", DRBD_MAJOR);
3019 	return 0; /* Success! */
3020 
3021 fail:
3022 	drbd_cleanup();
3023 	if (err == -ENOMEM)
3024 		pr_err("ran out of memory\n");
3025 	else
3026 		pr_err("initialization failure\n");
3027 	return err;
3028 }
3029 
3030 static void drbd_free_one_sock(struct drbd_socket *ds)
3031 {
3032 	struct socket *s;
3033 	mutex_lock(&ds->mutex);
3034 	s = ds->socket;
3035 	ds->socket = NULL;
3036 	mutex_unlock(&ds->mutex);
3037 	if (s) {
3038 		/* so debugfs does not need to mutex_lock() */
3039 		synchronize_rcu();
3040 		kernel_sock_shutdown(s, SHUT_RDWR);
3041 		sock_release(s);
3042 	}
3043 }
3044 
3045 void drbd_free_sock(struct drbd_connection *connection)
3046 {
3047 	if (connection->data.socket)
3048 		drbd_free_one_sock(&connection->data);
3049 	if (connection->meta.socket)
3050 		drbd_free_one_sock(&connection->meta);
3051 }
3052 
3053 /* meta data management */
3054 
3055 void conn_md_sync(struct drbd_connection *connection)
3056 {
3057 	struct drbd_peer_device *peer_device;
3058 	int vnr;
3059 
3060 	rcu_read_lock();
3061 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3062 		struct drbd_device *device = peer_device->device;
3063 
3064 		kref_get(&device->kref);
3065 		rcu_read_unlock();
3066 		drbd_md_sync(device);
3067 		kref_put(&device->kref, drbd_destroy_device);
3068 		rcu_read_lock();
3069 	}
3070 	rcu_read_unlock();
3071 }
3072 
3073 /* aligned 4kByte */
3074 struct meta_data_on_disk {
3075 	u64 la_size_sect;      /* last agreed size. */
3076 	u64 uuid[UI_SIZE];   /* UUIDs. */
3077 	u64 device_uuid;
3078 	u64 reserved_u64_1;
3079 	u32 flags;             /* MDF */
3080 	u32 magic;
3081 	u32 md_size_sect;
3082 	u32 al_offset;         /* offset to this block */
3083 	u32 al_nr_extents;     /* important for restoring the AL (userspace) */
3084 	      /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3085 	u32 bm_offset;         /* offset to the bitmap, from here */
3086 	u32 bm_bytes_per_bit;  /* BM_BLOCK_SIZE */
3087 	u32 la_peer_max_bio_size;   /* last peer max_bio_size */
3088 
3089 	/* see al_tr_number_to_on_disk_sector() */
3090 	u32 al_stripes;
3091 	u32 al_stripe_size_4k;
3092 
3093 	u8 reserved_u8[4096 - (7*8 + 10*4)];
3094 } __packed;
3095 
3096 
3097 
3098 void drbd_md_write(struct drbd_device *device, void *b)
3099 {
3100 	struct meta_data_on_disk *buffer = b;
3101 	sector_t sector;
3102 	int i;
3103 
3104 	memset(buffer, 0, sizeof(*buffer));
3105 
3106 	buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3107 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3108 		buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3109 	buffer->flags = cpu_to_be32(device->ldev->md.flags);
3110 	buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3111 
3112 	buffer->md_size_sect  = cpu_to_be32(device->ldev->md.md_size_sect);
3113 	buffer->al_offset     = cpu_to_be32(device->ldev->md.al_offset);
3114 	buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3115 	buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3116 	buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3117 
3118 	buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3119 	buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3120 
3121 	buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3122 	buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3123 
3124 	D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3125 	sector = device->ldev->md.md_offset;
3126 
3127 	if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3128 		/* this was a try anyways ... */
3129 		drbd_err(device, "meta data update failed!\n");
3130 		drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3131 	}
3132 }
3133 
3134 /**
3135  * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3136  * @device:	DRBD device.
3137  */
3138 void drbd_md_sync(struct drbd_device *device)
3139 {
3140 	struct meta_data_on_disk *buffer;
3141 
3142 	/* Don't accidentally change the DRBD meta data layout. */
3143 	BUILD_BUG_ON(UI_SIZE != 4);
3144 	BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3145 
3146 	del_timer(&device->md_sync_timer);
3147 	/* timer may be rearmed by drbd_md_mark_dirty() now. */
3148 	if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3149 		return;
3150 
3151 	/* We use here D_FAILED and not D_ATTACHING because we try to write
3152 	 * metadata even if we detach due to a disk failure! */
3153 	if (!get_ldev_if_state(device, D_FAILED))
3154 		return;
3155 
3156 	buffer = drbd_md_get_buffer(device, __func__);
3157 	if (!buffer)
3158 		goto out;
3159 
3160 	drbd_md_write(device, buffer);
3161 
3162 	/* Update device->ldev->md.la_size_sect,
3163 	 * since we updated it on metadata. */
3164 	device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3165 
3166 	drbd_md_put_buffer(device);
3167 out:
3168 	put_ldev(device);
3169 }
3170 
3171 static int check_activity_log_stripe_size(struct drbd_device *device,
3172 		struct meta_data_on_disk *on_disk,
3173 		struct drbd_md *in_core)
3174 {
3175 	u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3176 	u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3177 	u64 al_size_4k;
3178 
3179 	/* both not set: default to old fixed size activity log */
3180 	if (al_stripes == 0 && al_stripe_size_4k == 0) {
3181 		al_stripes = 1;
3182 		al_stripe_size_4k = MD_32kB_SECT/8;
3183 	}
3184 
3185 	/* some paranoia plausibility checks */
3186 
3187 	/* we need both values to be set */
3188 	if (al_stripes == 0 || al_stripe_size_4k == 0)
3189 		goto err;
3190 
3191 	al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3192 
3193 	/* Upper limit of activity log area, to avoid potential overflow
3194 	 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3195 	 * than 72 * 4k blocks total only increases the amount of history,
3196 	 * limiting this arbitrarily to 16 GB is not a real limitation ;-)  */
3197 	if (al_size_4k > (16 * 1024 * 1024/4))
3198 		goto err;
3199 
3200 	/* Lower limit: we need at least 8 transaction slots (32kB)
3201 	 * to not break existing setups */
3202 	if (al_size_4k < MD_32kB_SECT/8)
3203 		goto err;
3204 
3205 	in_core->al_stripe_size_4k = al_stripe_size_4k;
3206 	in_core->al_stripes = al_stripes;
3207 	in_core->al_size_4k = al_size_4k;
3208 
3209 	return 0;
3210 err:
3211 	drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3212 			al_stripes, al_stripe_size_4k);
3213 	return -EINVAL;
3214 }
3215 
3216 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3217 {
3218 	sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3219 	struct drbd_md *in_core = &bdev->md;
3220 	s32 on_disk_al_sect;
3221 	s32 on_disk_bm_sect;
3222 
3223 	/* The on-disk size of the activity log, calculated from offsets, and
3224 	 * the size of the activity log calculated from the stripe settings,
3225 	 * should match.
3226 	 * Though we could relax this a bit: it is ok, if the striped activity log
3227 	 * fits in the available on-disk activity log size.
3228 	 * Right now, that would break how resize is implemented.
3229 	 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3230 	 * of possible unused padding space in the on disk layout. */
3231 	if (in_core->al_offset < 0) {
3232 		if (in_core->bm_offset > in_core->al_offset)
3233 			goto err;
3234 		on_disk_al_sect = -in_core->al_offset;
3235 		on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3236 	} else {
3237 		if (in_core->al_offset != MD_4kB_SECT)
3238 			goto err;
3239 		if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3240 			goto err;
3241 
3242 		on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3243 		on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3244 	}
3245 
3246 	/* old fixed size meta data is exactly that: fixed. */
3247 	if (in_core->meta_dev_idx >= 0) {
3248 		if (in_core->md_size_sect != MD_128MB_SECT
3249 		||  in_core->al_offset != MD_4kB_SECT
3250 		||  in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3251 		||  in_core->al_stripes != 1
3252 		||  in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3253 			goto err;
3254 	}
3255 
3256 	if (capacity < in_core->md_size_sect)
3257 		goto err;
3258 	if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3259 		goto err;
3260 
3261 	/* should be aligned, and at least 32k */
3262 	if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3263 		goto err;
3264 
3265 	/* should fit (for now: exactly) into the available on-disk space;
3266 	 * overflow prevention is in check_activity_log_stripe_size() above. */
3267 	if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3268 		goto err;
3269 
3270 	/* again, should be aligned */
3271 	if (in_core->bm_offset & 7)
3272 		goto err;
3273 
3274 	/* FIXME check for device grow with flex external meta data? */
3275 
3276 	/* can the available bitmap space cover the last agreed device size? */
3277 	if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3278 		goto err;
3279 
3280 	return 0;
3281 
3282 err:
3283 	drbd_err(device, "meta data offsets don't make sense: idx=%d "
3284 			"al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3285 			"md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3286 			in_core->meta_dev_idx,
3287 			in_core->al_stripes, in_core->al_stripe_size_4k,
3288 			in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3289 			(unsigned long long)in_core->la_size_sect,
3290 			(unsigned long long)capacity);
3291 
3292 	return -EINVAL;
3293 }
3294 
3295 
3296 /**
3297  * drbd_md_read() - Reads in the meta data super block
3298  * @device:	DRBD device.
3299  * @bdev:	Device from which the meta data should be read in.
3300  *
3301  * Return NO_ERROR on success, and an enum drbd_ret_code in case
3302  * something goes wrong.
3303  *
3304  * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3305  * even before @bdev is assigned to @device->ldev.
3306  */
3307 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3308 {
3309 	struct meta_data_on_disk *buffer;
3310 	u32 magic, flags;
3311 	int i, rv = NO_ERROR;
3312 
3313 	if (device->state.disk != D_DISKLESS)
3314 		return ERR_DISK_CONFIGURED;
3315 
3316 	buffer = drbd_md_get_buffer(device, __func__);
3317 	if (!buffer)
3318 		return ERR_NOMEM;
3319 
3320 	/* First, figure out where our meta data superblock is located,
3321 	 * and read it. */
3322 	bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3323 	bdev->md.md_offset = drbd_md_ss(bdev);
3324 	/* Even for (flexible or indexed) external meta data,
3325 	 * initially restrict us to the 4k superblock for now.
3326 	 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3327 	bdev->md.md_size_sect = 8;
3328 
3329 	if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3330 				 REQ_OP_READ)) {
3331 		/* NOTE: can't do normal error processing here as this is
3332 		   called BEFORE disk is attached */
3333 		drbd_err(device, "Error while reading metadata.\n");
3334 		rv = ERR_IO_MD_DISK;
3335 		goto err;
3336 	}
3337 
3338 	magic = be32_to_cpu(buffer->magic);
3339 	flags = be32_to_cpu(buffer->flags);
3340 	if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3341 	    (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3342 			/* btw: that's Activity Log clean, not "all" clean. */
3343 		drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3344 		rv = ERR_MD_UNCLEAN;
3345 		goto err;
3346 	}
3347 
3348 	rv = ERR_MD_INVALID;
3349 	if (magic != DRBD_MD_MAGIC_08) {
3350 		if (magic == DRBD_MD_MAGIC_07)
3351 			drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3352 		else
3353 			drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3354 		goto err;
3355 	}
3356 
3357 	if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3358 		drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3359 		    be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3360 		goto err;
3361 	}
3362 
3363 
3364 	/* convert to in_core endian */
3365 	bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3366 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3367 		bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3368 	bdev->md.flags = be32_to_cpu(buffer->flags);
3369 	bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3370 
3371 	bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3372 	bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3373 	bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3374 
3375 	if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3376 		goto err;
3377 	if (check_offsets_and_sizes(device, bdev))
3378 		goto err;
3379 
3380 	if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3381 		drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3382 		    be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3383 		goto err;
3384 	}
3385 	if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3386 		drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3387 		    be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3388 		goto err;
3389 	}
3390 
3391 	rv = NO_ERROR;
3392 
3393 	spin_lock_irq(&device->resource->req_lock);
3394 	if (device->state.conn < C_CONNECTED) {
3395 		unsigned int peer;
3396 		peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3397 		peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3398 		device->peer_max_bio_size = peer;
3399 	}
3400 	spin_unlock_irq(&device->resource->req_lock);
3401 
3402  err:
3403 	drbd_md_put_buffer(device);
3404 
3405 	return rv;
3406 }
3407 
3408 /**
3409  * drbd_md_mark_dirty() - Mark meta data super block as dirty
3410  * @device:	DRBD device.
3411  *
3412  * Call this function if you change anything that should be written to
3413  * the meta-data super block. This function sets MD_DIRTY, and starts a
3414  * timer that ensures that within five seconds you have to call drbd_md_sync().
3415  */
3416 void drbd_md_mark_dirty(struct drbd_device *device)
3417 {
3418 	if (!test_and_set_bit(MD_DIRTY, &device->flags))
3419 		mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3420 }
3421 
3422 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3423 {
3424 	int i;
3425 
3426 	for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3427 		device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3428 }
3429 
3430 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3431 {
3432 	if (idx == UI_CURRENT) {
3433 		if (device->state.role == R_PRIMARY)
3434 			val |= 1;
3435 		else
3436 			val &= ~((u64)1);
3437 
3438 		drbd_set_ed_uuid(device, val);
3439 	}
3440 
3441 	device->ldev->md.uuid[idx] = val;
3442 	drbd_md_mark_dirty(device);
3443 }
3444 
3445 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3446 {
3447 	unsigned long flags;
3448 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3449 	__drbd_uuid_set(device, idx, val);
3450 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3451 }
3452 
3453 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3454 {
3455 	unsigned long flags;
3456 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3457 	if (device->ldev->md.uuid[idx]) {
3458 		drbd_uuid_move_history(device);
3459 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3460 	}
3461 	__drbd_uuid_set(device, idx, val);
3462 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3463 }
3464 
3465 /**
3466  * drbd_uuid_new_current() - Creates a new current UUID
3467  * @device:	DRBD device.
3468  *
3469  * Creates a new current UUID, and rotates the old current UUID into
3470  * the bitmap slot. Causes an incremental resync upon next connect.
3471  */
3472 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3473 {
3474 	u64 val;
3475 	unsigned long long bm_uuid;
3476 
3477 	get_random_bytes(&val, sizeof(u64));
3478 
3479 	spin_lock_irq(&device->ldev->md.uuid_lock);
3480 	bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3481 
3482 	if (bm_uuid)
3483 		drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3484 
3485 	device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3486 	__drbd_uuid_set(device, UI_CURRENT, val);
3487 	spin_unlock_irq(&device->ldev->md.uuid_lock);
3488 
3489 	drbd_print_uuids(device, "new current UUID");
3490 	/* get it to stable storage _now_ */
3491 	drbd_md_sync(device);
3492 }
3493 
3494 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3495 {
3496 	unsigned long flags;
3497 	if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3498 		return;
3499 
3500 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3501 	if (val == 0) {
3502 		drbd_uuid_move_history(device);
3503 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3504 		device->ldev->md.uuid[UI_BITMAP] = 0;
3505 	} else {
3506 		unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3507 		if (bm_uuid)
3508 			drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3509 
3510 		device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3511 	}
3512 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3513 
3514 	drbd_md_mark_dirty(device);
3515 }
3516 
3517 /**
3518  * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3519  * @device:	DRBD device.
3520  *
3521  * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3522  */
3523 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3524 {
3525 	int rv = -EIO;
3526 
3527 	drbd_md_set_flag(device, MDF_FULL_SYNC);
3528 	drbd_md_sync(device);
3529 	drbd_bm_set_all(device);
3530 
3531 	rv = drbd_bm_write(device);
3532 
3533 	if (!rv) {
3534 		drbd_md_clear_flag(device, MDF_FULL_SYNC);
3535 		drbd_md_sync(device);
3536 	}
3537 
3538 	return rv;
3539 }
3540 
3541 /**
3542  * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3543  * @device:	DRBD device.
3544  *
3545  * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3546  */
3547 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3548 {
3549 	drbd_resume_al(device);
3550 	drbd_bm_clear_all(device);
3551 	return drbd_bm_write(device);
3552 }
3553 
3554 static int w_bitmap_io(struct drbd_work *w, int unused)
3555 {
3556 	struct drbd_device *device =
3557 		container_of(w, struct drbd_device, bm_io_work.w);
3558 	struct bm_io_work *work = &device->bm_io_work;
3559 	int rv = -EIO;
3560 
3561 	if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3562 		int cnt = atomic_read(&device->ap_bio_cnt);
3563 		if (cnt)
3564 			drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3565 					cnt, work->why);
3566 	}
3567 
3568 	if (get_ldev(device)) {
3569 		drbd_bm_lock(device, work->why, work->flags);
3570 		rv = work->io_fn(device);
3571 		drbd_bm_unlock(device);
3572 		put_ldev(device);
3573 	}
3574 
3575 	clear_bit_unlock(BITMAP_IO, &device->flags);
3576 	wake_up(&device->misc_wait);
3577 
3578 	if (work->done)
3579 		work->done(device, rv);
3580 
3581 	clear_bit(BITMAP_IO_QUEUED, &device->flags);
3582 	work->why = NULL;
3583 	work->flags = 0;
3584 
3585 	return 0;
3586 }
3587 
3588 /**
3589  * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3590  * @device:	DRBD device.
3591  * @io_fn:	IO callback to be called when bitmap IO is possible
3592  * @done:	callback to be called after the bitmap IO was performed
3593  * @why:	Descriptive text of the reason for doing the IO
3594  *
3595  * While IO on the bitmap happens we freeze application IO thus we ensure
3596  * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3597  * called from worker context. It MUST NOT be used while a previous such
3598  * work is still pending!
3599  *
3600  * Its worker function encloses the call of io_fn() by get_ldev() and
3601  * put_ldev().
3602  */
3603 void drbd_queue_bitmap_io(struct drbd_device *device,
3604 			  int (*io_fn)(struct drbd_device *),
3605 			  void (*done)(struct drbd_device *, int),
3606 			  char *why, enum bm_flag flags)
3607 {
3608 	D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3609 
3610 	D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3611 	D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3612 	D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3613 	if (device->bm_io_work.why)
3614 		drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3615 			why, device->bm_io_work.why);
3616 
3617 	device->bm_io_work.io_fn = io_fn;
3618 	device->bm_io_work.done = done;
3619 	device->bm_io_work.why = why;
3620 	device->bm_io_work.flags = flags;
3621 
3622 	spin_lock_irq(&device->resource->req_lock);
3623 	set_bit(BITMAP_IO, &device->flags);
3624 	/* don't wait for pending application IO if the caller indicates that
3625 	 * application IO does not conflict anyways. */
3626 	if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3627 		if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3628 			drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3629 					&device->bm_io_work.w);
3630 	}
3631 	spin_unlock_irq(&device->resource->req_lock);
3632 }
3633 
3634 /**
3635  * drbd_bitmap_io() -  Does an IO operation on the whole bitmap
3636  * @device:	DRBD device.
3637  * @io_fn:	IO callback to be called when bitmap IO is possible
3638  * @why:	Descriptive text of the reason for doing the IO
3639  *
3640  * freezes application IO while that the actual IO operations runs. This
3641  * functions MAY NOT be called from worker context.
3642  */
3643 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3644 		char *why, enum bm_flag flags)
3645 {
3646 	/* Only suspend io, if some operation is supposed to be locked out */
3647 	const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3648 	int rv;
3649 
3650 	D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3651 
3652 	if (do_suspend_io)
3653 		drbd_suspend_io(device);
3654 
3655 	drbd_bm_lock(device, why, flags);
3656 	rv = io_fn(device);
3657 	drbd_bm_unlock(device);
3658 
3659 	if (do_suspend_io)
3660 		drbd_resume_io(device);
3661 
3662 	return rv;
3663 }
3664 
3665 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3666 {
3667 	if ((device->ldev->md.flags & flag) != flag) {
3668 		drbd_md_mark_dirty(device);
3669 		device->ldev->md.flags |= flag;
3670 	}
3671 }
3672 
3673 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3674 {
3675 	if ((device->ldev->md.flags & flag) != 0) {
3676 		drbd_md_mark_dirty(device);
3677 		device->ldev->md.flags &= ~flag;
3678 	}
3679 }
3680 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3681 {
3682 	return (bdev->md.flags & flag) != 0;
3683 }
3684 
3685 static void md_sync_timer_fn(struct timer_list *t)
3686 {
3687 	struct drbd_device *device = from_timer(device, t, md_sync_timer);
3688 	drbd_device_post_work(device, MD_SYNC);
3689 }
3690 
3691 const char *cmdname(enum drbd_packet cmd)
3692 {
3693 	/* THINK may need to become several global tables
3694 	 * when we want to support more than
3695 	 * one PRO_VERSION */
3696 	static const char *cmdnames[] = {
3697 		[P_DATA]	        = "Data",
3698 		[P_WSAME]	        = "WriteSame",
3699 		[P_TRIM]	        = "Trim",
3700 		[P_DATA_REPLY]	        = "DataReply",
3701 		[P_RS_DATA_REPLY]	= "RSDataReply",
3702 		[P_BARRIER]	        = "Barrier",
3703 		[P_BITMAP]	        = "ReportBitMap",
3704 		[P_BECOME_SYNC_TARGET]  = "BecomeSyncTarget",
3705 		[P_BECOME_SYNC_SOURCE]  = "BecomeSyncSource",
3706 		[P_UNPLUG_REMOTE]	= "UnplugRemote",
3707 		[P_DATA_REQUEST]	= "DataRequest",
3708 		[P_RS_DATA_REQUEST]     = "RSDataRequest",
3709 		[P_SYNC_PARAM]	        = "SyncParam",
3710 		[P_SYNC_PARAM89]	= "SyncParam89",
3711 		[P_PROTOCOL]            = "ReportProtocol",
3712 		[P_UUIDS]	        = "ReportUUIDs",
3713 		[P_SIZES]	        = "ReportSizes",
3714 		[P_STATE]	        = "ReportState",
3715 		[P_SYNC_UUID]           = "ReportSyncUUID",
3716 		[P_AUTH_CHALLENGE]      = "AuthChallenge",
3717 		[P_AUTH_RESPONSE]	= "AuthResponse",
3718 		[P_PING]		= "Ping",
3719 		[P_PING_ACK]	        = "PingAck",
3720 		[P_RECV_ACK]	        = "RecvAck",
3721 		[P_WRITE_ACK]	        = "WriteAck",
3722 		[P_RS_WRITE_ACK]	= "RSWriteAck",
3723 		[P_SUPERSEDED]          = "Superseded",
3724 		[P_NEG_ACK]	        = "NegAck",
3725 		[P_NEG_DREPLY]	        = "NegDReply",
3726 		[P_NEG_RS_DREPLY]	= "NegRSDReply",
3727 		[P_BARRIER_ACK]	        = "BarrierAck",
3728 		[P_STATE_CHG_REQ]       = "StateChgRequest",
3729 		[P_STATE_CHG_REPLY]     = "StateChgReply",
3730 		[P_OV_REQUEST]          = "OVRequest",
3731 		[P_OV_REPLY]            = "OVReply",
3732 		[P_OV_RESULT]           = "OVResult",
3733 		[P_CSUM_RS_REQUEST]     = "CsumRSRequest",
3734 		[P_RS_IS_IN_SYNC]	= "CsumRSIsInSync",
3735 		[P_COMPRESSED_BITMAP]   = "CBitmap",
3736 		[P_DELAY_PROBE]         = "DelayProbe",
3737 		[P_OUT_OF_SYNC]		= "OutOfSync",
3738 		[P_RETRY_WRITE]		= "RetryWrite",
3739 		[P_RS_CANCEL]		= "RSCancel",
3740 		[P_CONN_ST_CHG_REQ]	= "conn_st_chg_req",
3741 		[P_CONN_ST_CHG_REPLY]	= "conn_st_chg_reply",
3742 		[P_RETRY_WRITE]		= "retry_write",
3743 		[P_PROTOCOL_UPDATE]	= "protocol_update",
3744 		[P_RS_THIN_REQ]         = "rs_thin_req",
3745 		[P_RS_DEALLOCATED]      = "rs_deallocated",
3746 
3747 		/* enum drbd_packet, but not commands - obsoleted flags:
3748 		 *	P_MAY_IGNORE
3749 		 *	P_MAX_OPT_CMD
3750 		 */
3751 	};
3752 
3753 	/* too big for the array: 0xfffX */
3754 	if (cmd == P_INITIAL_META)
3755 		return "InitialMeta";
3756 	if (cmd == P_INITIAL_DATA)
3757 		return "InitialData";
3758 	if (cmd == P_CONNECTION_FEATURES)
3759 		return "ConnectionFeatures";
3760 	if (cmd >= ARRAY_SIZE(cmdnames))
3761 		return "Unknown";
3762 	return cmdnames[cmd];
3763 }
3764 
3765 /**
3766  * drbd_wait_misc  -  wait for a request to make progress
3767  * @device:	device associated with the request
3768  * @i:		the struct drbd_interval embedded in struct drbd_request or
3769  *		struct drbd_peer_request
3770  */
3771 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3772 {
3773 	struct net_conf *nc;
3774 	DEFINE_WAIT(wait);
3775 	long timeout;
3776 
3777 	rcu_read_lock();
3778 	nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3779 	if (!nc) {
3780 		rcu_read_unlock();
3781 		return -ETIMEDOUT;
3782 	}
3783 	timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3784 	rcu_read_unlock();
3785 
3786 	/* Indicate to wake up device->misc_wait on progress.  */
3787 	i->waiting = true;
3788 	prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3789 	spin_unlock_irq(&device->resource->req_lock);
3790 	timeout = schedule_timeout(timeout);
3791 	finish_wait(&device->misc_wait, &wait);
3792 	spin_lock_irq(&device->resource->req_lock);
3793 	if (!timeout || device->state.conn < C_CONNECTED)
3794 		return -ETIMEDOUT;
3795 	if (signal_pending(current))
3796 		return -ERESTARTSYS;
3797 	return 0;
3798 }
3799 
3800 void lock_all_resources(void)
3801 {
3802 	struct drbd_resource *resource;
3803 	int __maybe_unused i = 0;
3804 
3805 	mutex_lock(&resources_mutex);
3806 	local_irq_disable();
3807 	for_each_resource(resource, &drbd_resources)
3808 		spin_lock_nested(&resource->req_lock, i++);
3809 }
3810 
3811 void unlock_all_resources(void)
3812 {
3813 	struct drbd_resource *resource;
3814 
3815 	for_each_resource(resource, &drbd_resources)
3816 		spin_unlock(&resource->req_lock);
3817 	local_irq_enable();
3818 	mutex_unlock(&resources_mutex);
3819 }
3820 
3821 #ifdef CONFIG_DRBD_FAULT_INJECTION
3822 /* Fault insertion support including random number generator shamelessly
3823  * stolen from kernel/rcutorture.c */
3824 struct fault_random_state {
3825 	unsigned long state;
3826 	unsigned long count;
3827 };
3828 
3829 #define FAULT_RANDOM_MULT 39916801  /* prime */
3830 #define FAULT_RANDOM_ADD	479001701 /* prime */
3831 #define FAULT_RANDOM_REFRESH 10000
3832 
3833 /*
3834  * Crude but fast random-number generator.  Uses a linear congruential
3835  * generator, with occasional help from get_random_bytes().
3836  */
3837 static unsigned long
3838 _drbd_fault_random(struct fault_random_state *rsp)
3839 {
3840 	long refresh;
3841 
3842 	if (!rsp->count--) {
3843 		get_random_bytes(&refresh, sizeof(refresh));
3844 		rsp->state += refresh;
3845 		rsp->count = FAULT_RANDOM_REFRESH;
3846 	}
3847 	rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3848 	return swahw32(rsp->state);
3849 }
3850 
3851 static char *
3852 _drbd_fault_str(unsigned int type) {
3853 	static char *_faults[] = {
3854 		[DRBD_FAULT_MD_WR] = "Meta-data write",
3855 		[DRBD_FAULT_MD_RD] = "Meta-data read",
3856 		[DRBD_FAULT_RS_WR] = "Resync write",
3857 		[DRBD_FAULT_RS_RD] = "Resync read",
3858 		[DRBD_FAULT_DT_WR] = "Data write",
3859 		[DRBD_FAULT_DT_RD] = "Data read",
3860 		[DRBD_FAULT_DT_RA] = "Data read ahead",
3861 		[DRBD_FAULT_BM_ALLOC] = "BM allocation",
3862 		[DRBD_FAULT_AL_EE] = "EE allocation",
3863 		[DRBD_FAULT_RECEIVE] = "receive data corruption",
3864 	};
3865 
3866 	return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3867 }
3868 
3869 unsigned int
3870 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3871 {
3872 	static struct fault_random_state rrs = {0, 0};
3873 
3874 	unsigned int ret = (
3875 		(drbd_fault_devs == 0 ||
3876 			((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3877 		(((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3878 
3879 	if (ret) {
3880 		drbd_fault_count++;
3881 
3882 		if (__ratelimit(&drbd_ratelimit_state))
3883 			drbd_warn(device, "***Simulating %s failure\n",
3884 				_drbd_fault_str(type));
3885 	}
3886 
3887 	return ret;
3888 }
3889 #endif
3890 
3891 const char *drbd_buildtag(void)
3892 {
3893 	/* DRBD built from external sources has here a reference to the
3894 	   git hash of the source code. */
3895 
3896 	static char buildtag[38] = "\0uilt-in";
3897 
3898 	if (buildtag[0] == 0) {
3899 #ifdef MODULE
3900 		sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3901 #else
3902 		buildtag[0] = 'b';
3903 #endif
3904 	}
3905 
3906 	return buildtag;
3907 }
3908 
3909 module_init(drbd_init)
3910 module_exit(drbd_cleanup)
3911 
3912 EXPORT_SYMBOL(drbd_conn_str);
3913 EXPORT_SYMBOL(drbd_role_str);
3914 EXPORT_SYMBOL(drbd_disk_str);
3915 EXPORT_SYMBOL(drbd_set_st_err_str);
3916