xref: /linux/drivers/block/drbd/drbd_main.c (revision 87807f77a03d0271211b75f84b2a8b88f4e8e5d4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3    drbd.c
4 
5    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 
7    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
8    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
9    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10 
11    Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
12    from Logicworks, Inc. for making SDP replication support possible.
13 
14 
15  */
16 
17 #define pr_fmt(fmt)	KBUILD_MODNAME ": " fmt
18 
19 #include <linux/module.h>
20 #include <linux/jiffies.h>
21 #include <linux/drbd.h>
22 #include <linux/uaccess.h>
23 #include <asm/types.h>
24 #include <net/sock.h>
25 #include <linux/ctype.h>
26 #include <linux/mutex.h>
27 #include <linux/fs.h>
28 #include <linux/file.h>
29 #include <linux/proc_fs.h>
30 #include <linux/init.h>
31 #include <linux/mm.h>
32 #include <linux/memcontrol.h>
33 #include <linux/mm_inline.h>
34 #include <linux/slab.h>
35 #include <linux/random.h>
36 #include <linux/reboot.h>
37 #include <linux/notifier.h>
38 #include <linux/kthread.h>
39 #include <linux/workqueue.h>
40 #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 DEFINE_SPINLOCK(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 	.submit_bio	= drbd_submit_bio,
137 	.open		= drbd_open,
138 	.release	= drbd_release,
139 };
140 
141 #ifdef __CHECKER__
142 /* When checking with sparse, and this is an inline function, sparse will
143    give tons of false positives. When this is a real functions sparse works.
144  */
145 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
146 {
147 	int io_allowed;
148 
149 	atomic_inc(&device->local_cnt);
150 	io_allowed = (device->state.disk >= mins);
151 	if (!io_allowed) {
152 		if (atomic_dec_and_test(&device->local_cnt))
153 			wake_up(&device->misc_wait);
154 	}
155 	return io_allowed;
156 }
157 
158 #endif
159 
160 /**
161  * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
162  * @connection:	DRBD connection.
163  * @barrier_nr:	Expected identifier of the DRBD write barrier packet.
164  * @set_size:	Expected number of requests before that barrier.
165  *
166  * In case the passed barrier_nr or set_size does not match the oldest
167  * epoch of not yet barrier-acked requests, this function will cause a
168  * termination of the connection.
169  */
170 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
171 		unsigned int set_size)
172 {
173 	struct drbd_request *r;
174 	struct drbd_request *req = NULL, *tmp = NULL;
175 	int expect_epoch = 0;
176 	int expect_size = 0;
177 
178 	spin_lock_irq(&connection->resource->req_lock);
179 
180 	/* find oldest not yet barrier-acked write request,
181 	 * count writes in its epoch. */
182 	list_for_each_entry(r, &connection->transfer_log, tl_requests) {
183 		const unsigned s = r->rq_state;
184 		if (!req) {
185 			if (!(s & RQ_WRITE))
186 				continue;
187 			if (!(s & RQ_NET_MASK))
188 				continue;
189 			if (s & RQ_NET_DONE)
190 				continue;
191 			req = r;
192 			expect_epoch = req->epoch;
193 			expect_size ++;
194 		} else {
195 			if (r->epoch != expect_epoch)
196 				break;
197 			if (!(s & RQ_WRITE))
198 				continue;
199 			/* if (s & RQ_DONE): not expected */
200 			/* if (!(s & RQ_NET_MASK)): not expected */
201 			expect_size++;
202 		}
203 	}
204 
205 	/* first some paranoia code */
206 	if (req == NULL) {
207 		drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
208 			 barrier_nr);
209 		goto bail;
210 	}
211 	if (expect_epoch != barrier_nr) {
212 		drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
213 			 barrier_nr, expect_epoch);
214 		goto bail;
215 	}
216 
217 	if (expect_size != set_size) {
218 		drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
219 			 barrier_nr, set_size, expect_size);
220 		goto bail;
221 	}
222 
223 	/* Clean up list of requests processed during current epoch. */
224 	/* this extra list walk restart is paranoia,
225 	 * to catch requests being barrier-acked "unexpectedly".
226 	 * It usually should find the same req again, or some READ preceding it. */
227 	list_for_each_entry(req, &connection->transfer_log, tl_requests)
228 		if (req->epoch == expect_epoch) {
229 			tmp = req;
230 			break;
231 		}
232 	req = list_prepare_entry(tmp, &connection->transfer_log, tl_requests);
233 	list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
234 		struct drbd_peer_device *peer_device;
235 		if (req->epoch != expect_epoch)
236 			break;
237 		peer_device = conn_peer_device(connection, req->device->vnr);
238 		_req_mod(req, BARRIER_ACKED, peer_device);
239 	}
240 	spin_unlock_irq(&connection->resource->req_lock);
241 
242 	return;
243 
244 bail:
245 	spin_unlock_irq(&connection->resource->req_lock);
246 	conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
247 }
248 
249 
250 /**
251  * _tl_restart() - Walks the transfer log, and applies an action to all requests
252  * @connection:	DRBD connection to operate on.
253  * @what:       The action/event to perform with all request objects
254  *
255  * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
256  * RESTART_FROZEN_DISK_IO.
257  */
258 /* must hold resource->req_lock */
259 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
260 {
261 	struct drbd_peer_device *peer_device;
262 	struct drbd_request *req, *r;
263 
264 	list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
265 		peer_device = conn_peer_device(connection, req->device->vnr);
266 		_req_mod(req, what, peer_device);
267 	}
268 }
269 
270 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
271 {
272 	spin_lock_irq(&connection->resource->req_lock);
273 	_tl_restart(connection, what);
274 	spin_unlock_irq(&connection->resource->req_lock);
275 }
276 
277 /**
278  * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
279  * @connection:	DRBD connection.
280  *
281  * This is called after the connection to the peer was lost. The storage covered
282  * by the requests on the transfer gets marked as our of sync. Called from the
283  * receiver thread and the worker thread.
284  */
285 void tl_clear(struct drbd_connection *connection)
286 {
287 	tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
288 }
289 
290 /**
291  * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
292  * @device:	DRBD device.
293  */
294 void tl_abort_disk_io(struct drbd_device *device)
295 {
296 	struct drbd_connection *connection = first_peer_device(device)->connection;
297 	struct drbd_request *req, *r;
298 
299 	spin_lock_irq(&connection->resource->req_lock);
300 	list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
301 		if (!(req->rq_state & RQ_LOCAL_PENDING))
302 			continue;
303 		if (req->device != device)
304 			continue;
305 		_req_mod(req, ABORT_DISK_IO, NULL);
306 	}
307 	spin_unlock_irq(&connection->resource->req_lock);
308 }
309 
310 static int drbd_thread_setup(void *arg)
311 {
312 	struct drbd_thread *thi = (struct drbd_thread *) arg;
313 	struct drbd_resource *resource = thi->resource;
314 	unsigned long flags;
315 	int retval;
316 
317 	snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
318 		 thi->name[0],
319 		 resource->name);
320 
321 	allow_kernel_signal(DRBD_SIGKILL);
322 	allow_kernel_signal(SIGXCPU);
323 restart:
324 	retval = thi->function(thi);
325 
326 	spin_lock_irqsave(&thi->t_lock, flags);
327 
328 	/* if the receiver has been "EXITING", the last thing it did
329 	 * was set the conn state to "StandAlone",
330 	 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
331 	 * and receiver thread will be "started".
332 	 * drbd_thread_start needs to set "RESTARTING" in that case.
333 	 * t_state check and assignment needs to be within the same spinlock,
334 	 * so either thread_start sees EXITING, and can remap to RESTARTING,
335 	 * or thread_start see NONE, and can proceed as normal.
336 	 */
337 
338 	if (thi->t_state == RESTARTING) {
339 		drbd_info(resource, "Restarting %s thread\n", thi->name);
340 		thi->t_state = RUNNING;
341 		spin_unlock_irqrestore(&thi->t_lock, flags);
342 		goto restart;
343 	}
344 
345 	thi->task = NULL;
346 	thi->t_state = NONE;
347 	smp_mb();
348 	complete_all(&thi->stop);
349 	spin_unlock_irqrestore(&thi->t_lock, flags);
350 
351 	drbd_info(resource, "Terminating %s\n", current->comm);
352 
353 	/* Release mod reference taken when thread was started */
354 
355 	if (thi->connection)
356 		kref_put(&thi->connection->kref, drbd_destroy_connection);
357 	kref_put(&resource->kref, drbd_destroy_resource);
358 	module_put(THIS_MODULE);
359 	return retval;
360 }
361 
362 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
363 			     int (*func) (struct drbd_thread *), const char *name)
364 {
365 	spin_lock_init(&thi->t_lock);
366 	thi->task    = NULL;
367 	thi->t_state = NONE;
368 	thi->function = func;
369 	thi->resource = resource;
370 	thi->connection = NULL;
371 	thi->name = name;
372 }
373 
374 int drbd_thread_start(struct drbd_thread *thi)
375 {
376 	struct drbd_resource *resource = thi->resource;
377 	struct task_struct *nt;
378 	unsigned long flags;
379 
380 	/* is used from state engine doing drbd_thread_stop_nowait,
381 	 * while holding the req lock irqsave */
382 	spin_lock_irqsave(&thi->t_lock, flags);
383 
384 	switch (thi->t_state) {
385 	case NONE:
386 		drbd_info(resource, "Starting %s thread (from %s [%d])\n",
387 			 thi->name, current->comm, current->pid);
388 
389 		/* Get ref on module for thread - this is released when thread exits */
390 		if (!try_module_get(THIS_MODULE)) {
391 			drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
392 			spin_unlock_irqrestore(&thi->t_lock, flags);
393 			return false;
394 		}
395 
396 		kref_get(&resource->kref);
397 		if (thi->connection)
398 			kref_get(&thi->connection->kref);
399 
400 		init_completion(&thi->stop);
401 		thi->reset_cpu_mask = 1;
402 		thi->t_state = RUNNING;
403 		spin_unlock_irqrestore(&thi->t_lock, flags);
404 		flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
405 
406 		nt = kthread_create(drbd_thread_setup, (void *) thi,
407 				    "drbd_%c_%s", thi->name[0], thi->resource->name);
408 
409 		if (IS_ERR(nt)) {
410 			drbd_err(resource, "Couldn't start thread\n");
411 
412 			if (thi->connection)
413 				kref_put(&thi->connection->kref, drbd_destroy_connection);
414 			kref_put(&resource->kref, drbd_destroy_resource);
415 			module_put(THIS_MODULE);
416 			return false;
417 		}
418 		spin_lock_irqsave(&thi->t_lock, flags);
419 		thi->task = nt;
420 		thi->t_state = RUNNING;
421 		spin_unlock_irqrestore(&thi->t_lock, flags);
422 		wake_up_process(nt);
423 		break;
424 	case EXITING:
425 		thi->t_state = RESTARTING;
426 		drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
427 				thi->name, current->comm, current->pid);
428 		fallthrough;
429 	case RUNNING:
430 	case RESTARTING:
431 	default:
432 		spin_unlock_irqrestore(&thi->t_lock, flags);
433 		break;
434 	}
435 
436 	return true;
437 }
438 
439 
440 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
441 {
442 	unsigned long flags;
443 
444 	enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
445 
446 	/* may be called from state engine, holding the req lock irqsave */
447 	spin_lock_irqsave(&thi->t_lock, flags);
448 
449 	if (thi->t_state == NONE) {
450 		spin_unlock_irqrestore(&thi->t_lock, flags);
451 		if (restart)
452 			drbd_thread_start(thi);
453 		return;
454 	}
455 
456 	if (thi->t_state != ns) {
457 		if (thi->task == NULL) {
458 			spin_unlock_irqrestore(&thi->t_lock, flags);
459 			return;
460 		}
461 
462 		thi->t_state = ns;
463 		smp_mb();
464 		init_completion(&thi->stop);
465 		if (thi->task != current)
466 			send_sig(DRBD_SIGKILL, thi->task, 1);
467 	}
468 
469 	spin_unlock_irqrestore(&thi->t_lock, flags);
470 
471 	if (wait)
472 		wait_for_completion(&thi->stop);
473 }
474 
475 int conn_lowest_minor(struct drbd_connection *connection)
476 {
477 	struct drbd_peer_device *peer_device;
478 	int vnr = 0, minor = -1;
479 
480 	rcu_read_lock();
481 	peer_device = idr_get_next(&connection->peer_devices, &vnr);
482 	if (peer_device)
483 		minor = device_to_minor(peer_device->device);
484 	rcu_read_unlock();
485 
486 	return minor;
487 }
488 
489 #ifdef CONFIG_SMP
490 /*
491  * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
492  *
493  * Forces all threads of a resource onto the same CPU. This is beneficial for
494  * DRBD's performance. May be overwritten by user's configuration.
495  */
496 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
497 {
498 	unsigned int *resources_per_cpu, min_index = ~0;
499 
500 	resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu),
501 				    GFP_KERNEL);
502 	if (resources_per_cpu) {
503 		struct drbd_resource *resource;
504 		unsigned int cpu, min = ~0;
505 
506 		rcu_read_lock();
507 		for_each_resource_rcu(resource, &drbd_resources) {
508 			for_each_cpu(cpu, resource->cpu_mask)
509 				resources_per_cpu[cpu]++;
510 		}
511 		rcu_read_unlock();
512 		for_each_online_cpu(cpu) {
513 			if (resources_per_cpu[cpu] < min) {
514 				min = resources_per_cpu[cpu];
515 				min_index = cpu;
516 			}
517 		}
518 		kfree(resources_per_cpu);
519 	}
520 	if (min_index == ~0) {
521 		cpumask_setall(*cpu_mask);
522 		return;
523 	}
524 	cpumask_set_cpu(min_index, *cpu_mask);
525 }
526 
527 /**
528  * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
529  * @thi:	drbd_thread object
530  *
531  * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
532  * prematurely.
533  */
534 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
535 {
536 	struct drbd_resource *resource = thi->resource;
537 	struct task_struct *p = current;
538 
539 	if (!thi->reset_cpu_mask)
540 		return;
541 	thi->reset_cpu_mask = 0;
542 	set_cpus_allowed_ptr(p, resource->cpu_mask);
543 }
544 #else
545 #define drbd_calc_cpu_mask(A) ({})
546 #endif
547 
548 /*
549  * drbd_header_size  -  size of a packet header
550  *
551  * The header size is a multiple of 8, so any payload following the header is
552  * word aligned on 64-bit architectures.  (The bitmap send and receive code
553  * relies on this.)
554  */
555 unsigned int drbd_header_size(struct drbd_connection *connection)
556 {
557 	if (connection->agreed_pro_version >= 100) {
558 		BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
559 		return sizeof(struct p_header100);
560 	} else {
561 		BUILD_BUG_ON(sizeof(struct p_header80) !=
562 			     sizeof(struct p_header95));
563 		BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
564 		return sizeof(struct p_header80);
565 	}
566 }
567 
568 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
569 {
570 	h->magic   = cpu_to_be32(DRBD_MAGIC);
571 	h->command = cpu_to_be16(cmd);
572 	h->length  = cpu_to_be16(size);
573 	return sizeof(struct p_header80);
574 }
575 
576 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
577 {
578 	h->magic   = cpu_to_be16(DRBD_MAGIC_BIG);
579 	h->command = cpu_to_be16(cmd);
580 	h->length = cpu_to_be32(size);
581 	return sizeof(struct p_header95);
582 }
583 
584 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
585 				      int size, int vnr)
586 {
587 	h->magic = cpu_to_be32(DRBD_MAGIC_100);
588 	h->volume = cpu_to_be16(vnr);
589 	h->command = cpu_to_be16(cmd);
590 	h->length = cpu_to_be32(size);
591 	h->pad = 0;
592 	return sizeof(struct p_header100);
593 }
594 
595 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
596 				   void *buffer, enum drbd_packet cmd, int size)
597 {
598 	if (connection->agreed_pro_version >= 100)
599 		return prepare_header100(buffer, cmd, size, vnr);
600 	else if (connection->agreed_pro_version >= 95 &&
601 		 size > DRBD_MAX_SIZE_H80_PACKET)
602 		return prepare_header95(buffer, cmd, size);
603 	else
604 		return prepare_header80(buffer, cmd, size);
605 }
606 
607 static void *__conn_prepare_command(struct drbd_connection *connection,
608 				    struct drbd_socket *sock)
609 {
610 	if (!sock->socket)
611 		return NULL;
612 	return sock->sbuf + drbd_header_size(connection);
613 }
614 
615 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
616 {
617 	void *p;
618 
619 	mutex_lock(&sock->mutex);
620 	p = __conn_prepare_command(connection, sock);
621 	if (!p)
622 		mutex_unlock(&sock->mutex);
623 
624 	return p;
625 }
626 
627 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
628 {
629 	return conn_prepare_command(peer_device->connection, sock);
630 }
631 
632 static int __send_command(struct drbd_connection *connection, int vnr,
633 			  struct drbd_socket *sock, enum drbd_packet cmd,
634 			  unsigned int header_size, void *data,
635 			  unsigned int size)
636 {
637 	int msg_flags;
638 	int err;
639 
640 	/*
641 	 * Called with @data == NULL and the size of the data blocks in @size
642 	 * for commands that send data blocks.  For those commands, omit the
643 	 * MSG_MORE flag: this will increase the likelihood that data blocks
644 	 * which are page aligned on the sender will end up page aligned on the
645 	 * receiver.
646 	 */
647 	msg_flags = data ? MSG_MORE : 0;
648 
649 	header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
650 				      header_size + size);
651 	err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
652 			    msg_flags);
653 	if (data && !err)
654 		err = drbd_send_all(connection, sock->socket, data, size, 0);
655 	/* DRBD protocol "pings" are latency critical.
656 	 * This is supposed to trigger tcp_push_pending_frames() */
657 	if (!err && (cmd == P_PING || cmd == P_PING_ACK))
658 		tcp_sock_set_nodelay(sock->socket->sk);
659 
660 	return err;
661 }
662 
663 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
664 			       enum drbd_packet cmd, unsigned int header_size,
665 			       void *data, unsigned int size)
666 {
667 	return __send_command(connection, 0, sock, cmd, header_size, data, size);
668 }
669 
670 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
671 		      enum drbd_packet cmd, unsigned int header_size,
672 		      void *data, unsigned int size)
673 {
674 	int err;
675 
676 	err = __conn_send_command(connection, sock, cmd, header_size, data, size);
677 	mutex_unlock(&sock->mutex);
678 	return err;
679 }
680 
681 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
682 		      enum drbd_packet cmd, unsigned int header_size,
683 		      void *data, unsigned int size)
684 {
685 	int err;
686 
687 	err = __send_command(peer_device->connection, peer_device->device->vnr,
688 			     sock, cmd, header_size, data, size);
689 	mutex_unlock(&sock->mutex);
690 	return err;
691 }
692 
693 int drbd_send_ping(struct drbd_connection *connection)
694 {
695 	struct drbd_socket *sock;
696 
697 	sock = &connection->meta;
698 	if (!conn_prepare_command(connection, sock))
699 		return -EIO;
700 	return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
701 }
702 
703 int drbd_send_ping_ack(struct drbd_connection *connection)
704 {
705 	struct drbd_socket *sock;
706 
707 	sock = &connection->meta;
708 	if (!conn_prepare_command(connection, sock))
709 		return -EIO;
710 	return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
711 }
712 
713 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
714 {
715 	struct drbd_socket *sock;
716 	struct p_rs_param_95 *p;
717 	int size;
718 	const int apv = peer_device->connection->agreed_pro_version;
719 	enum drbd_packet cmd;
720 	struct net_conf *nc;
721 	struct disk_conf *dc;
722 
723 	sock = &peer_device->connection->data;
724 	p = drbd_prepare_command(peer_device, sock);
725 	if (!p)
726 		return -EIO;
727 
728 	rcu_read_lock();
729 	nc = rcu_dereference(peer_device->connection->net_conf);
730 
731 	size = apv <= 87 ? sizeof(struct p_rs_param)
732 		: apv == 88 ? sizeof(struct p_rs_param)
733 			+ strlen(nc->verify_alg) + 1
734 		: apv <= 94 ? sizeof(struct p_rs_param_89)
735 		: /* apv >= 95 */ sizeof(struct p_rs_param_95);
736 
737 	cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
738 
739 	/* initialize verify_alg and csums_alg */
740 	BUILD_BUG_ON(sizeof(p->algs) != 2 * SHARED_SECRET_MAX);
741 	memset(&p->algs, 0, sizeof(p->algs));
742 
743 	if (get_ldev(peer_device->device)) {
744 		dc = rcu_dereference(peer_device->device->ldev->disk_conf);
745 		p->resync_rate = cpu_to_be32(dc->resync_rate);
746 		p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
747 		p->c_delay_target = cpu_to_be32(dc->c_delay_target);
748 		p->c_fill_target = cpu_to_be32(dc->c_fill_target);
749 		p->c_max_rate = cpu_to_be32(dc->c_max_rate);
750 		put_ldev(peer_device->device);
751 	} else {
752 		p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
753 		p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
754 		p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
755 		p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
756 		p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
757 	}
758 
759 	if (apv >= 88)
760 		strcpy(p->verify_alg, nc->verify_alg);
761 	if (apv >= 89)
762 		strcpy(p->csums_alg, nc->csums_alg);
763 	rcu_read_unlock();
764 
765 	return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
766 }
767 
768 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
769 {
770 	struct drbd_socket *sock;
771 	struct p_protocol *p;
772 	struct net_conf *nc;
773 	int size, cf;
774 
775 	sock = &connection->data;
776 	p = __conn_prepare_command(connection, sock);
777 	if (!p)
778 		return -EIO;
779 
780 	rcu_read_lock();
781 	nc = rcu_dereference(connection->net_conf);
782 
783 	if (nc->tentative && connection->agreed_pro_version < 92) {
784 		rcu_read_unlock();
785 		drbd_err(connection, "--dry-run is not supported by peer");
786 		return -EOPNOTSUPP;
787 	}
788 
789 	size = sizeof(*p);
790 	if (connection->agreed_pro_version >= 87)
791 		size += strlen(nc->integrity_alg) + 1;
792 
793 	p->protocol      = cpu_to_be32(nc->wire_protocol);
794 	p->after_sb_0p   = cpu_to_be32(nc->after_sb_0p);
795 	p->after_sb_1p   = cpu_to_be32(nc->after_sb_1p);
796 	p->after_sb_2p   = cpu_to_be32(nc->after_sb_2p);
797 	p->two_primaries = cpu_to_be32(nc->two_primaries);
798 	cf = 0;
799 	if (nc->discard_my_data)
800 		cf |= CF_DISCARD_MY_DATA;
801 	if (nc->tentative)
802 		cf |= CF_DRY_RUN;
803 	p->conn_flags    = cpu_to_be32(cf);
804 
805 	if (connection->agreed_pro_version >= 87)
806 		strcpy(p->integrity_alg, nc->integrity_alg);
807 	rcu_read_unlock();
808 
809 	return __conn_send_command(connection, sock, cmd, size, NULL, 0);
810 }
811 
812 int drbd_send_protocol(struct drbd_connection *connection)
813 {
814 	int err;
815 
816 	mutex_lock(&connection->data.mutex);
817 	err = __drbd_send_protocol(connection, P_PROTOCOL);
818 	mutex_unlock(&connection->data.mutex);
819 
820 	return err;
821 }
822 
823 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
824 {
825 	struct drbd_device *device = peer_device->device;
826 	struct drbd_socket *sock;
827 	struct p_uuids *p;
828 	int i;
829 
830 	if (!get_ldev_if_state(device, D_NEGOTIATING))
831 		return 0;
832 
833 	sock = &peer_device->connection->data;
834 	p = drbd_prepare_command(peer_device, sock);
835 	if (!p) {
836 		put_ldev(device);
837 		return -EIO;
838 	}
839 	spin_lock_irq(&device->ldev->md.uuid_lock);
840 	for (i = UI_CURRENT; i < UI_SIZE; i++)
841 		p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
842 	spin_unlock_irq(&device->ldev->md.uuid_lock);
843 
844 	device->comm_bm_set = drbd_bm_total_weight(device);
845 	p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
846 	rcu_read_lock();
847 	uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
848 	rcu_read_unlock();
849 	uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
850 	uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
851 	p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
852 
853 	put_ldev(device);
854 	return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
855 }
856 
857 int drbd_send_uuids(struct drbd_peer_device *peer_device)
858 {
859 	return _drbd_send_uuids(peer_device, 0);
860 }
861 
862 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
863 {
864 	return _drbd_send_uuids(peer_device, 8);
865 }
866 
867 void drbd_print_uuids(struct drbd_device *device, const char *text)
868 {
869 	if (get_ldev_if_state(device, D_NEGOTIATING)) {
870 		u64 *uuid = device->ldev->md.uuid;
871 		drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
872 		     text,
873 		     (unsigned long long)uuid[UI_CURRENT],
874 		     (unsigned long long)uuid[UI_BITMAP],
875 		     (unsigned long long)uuid[UI_HISTORY_START],
876 		     (unsigned long long)uuid[UI_HISTORY_END]);
877 		put_ldev(device);
878 	} else {
879 		drbd_info(device, "%s effective data uuid: %016llX\n",
880 				text,
881 				(unsigned long long)device->ed_uuid);
882 	}
883 }
884 
885 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
886 {
887 	struct drbd_device *device = peer_device->device;
888 	struct drbd_socket *sock;
889 	struct p_rs_uuid *p;
890 	u64 uuid;
891 
892 	D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
893 
894 	uuid = device->ldev->md.uuid[UI_BITMAP];
895 	if (uuid && uuid != UUID_JUST_CREATED)
896 		uuid = uuid + UUID_NEW_BM_OFFSET;
897 	else
898 		get_random_bytes(&uuid, sizeof(u64));
899 	drbd_uuid_set(device, UI_BITMAP, uuid);
900 	drbd_print_uuids(device, "updated sync UUID");
901 	drbd_md_sync(device);
902 
903 	sock = &peer_device->connection->data;
904 	p = drbd_prepare_command(peer_device, sock);
905 	if (p) {
906 		p->uuid = cpu_to_be64(uuid);
907 		drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
908 	}
909 }
910 
911 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
912 {
913 	struct drbd_device *device = peer_device->device;
914 	struct drbd_socket *sock;
915 	struct p_sizes *p;
916 	sector_t d_size, u_size;
917 	int q_order_type;
918 	unsigned int max_bio_size;
919 	unsigned int packet_size;
920 
921 	sock = &peer_device->connection->data;
922 	p = drbd_prepare_command(peer_device, sock);
923 	if (!p)
924 		return -EIO;
925 
926 	packet_size = sizeof(*p);
927 	if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
928 		packet_size += sizeof(p->qlim[0]);
929 
930 	memset(p, 0, packet_size);
931 	if (get_ldev_if_state(device, D_NEGOTIATING)) {
932 		struct block_device *bdev = device->ldev->backing_bdev;
933 		struct request_queue *q = bdev_get_queue(bdev);
934 
935 		d_size = drbd_get_max_capacity(device->ldev);
936 		rcu_read_lock();
937 		u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
938 		rcu_read_unlock();
939 		q_order_type = drbd_queue_order_type(device);
940 		max_bio_size = queue_max_hw_sectors(q) << 9;
941 		max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
942 		p->qlim->physical_block_size =
943 			cpu_to_be32(bdev_physical_block_size(bdev));
944 		p->qlim->logical_block_size =
945 			cpu_to_be32(bdev_logical_block_size(bdev));
946 		p->qlim->alignment_offset =
947 			cpu_to_be32(bdev_alignment_offset(bdev));
948 		p->qlim->io_min = cpu_to_be32(bdev_io_min(bdev));
949 		p->qlim->io_opt = cpu_to_be32(bdev_io_opt(bdev));
950 		p->qlim->discard_enabled = !!bdev_max_discard_sectors(bdev);
951 		put_ldev(device);
952 	} else {
953 		struct request_queue *q = device->rq_queue;
954 
955 		p->qlim->physical_block_size =
956 			cpu_to_be32(queue_physical_block_size(q));
957 		p->qlim->logical_block_size =
958 			cpu_to_be32(queue_logical_block_size(q));
959 		p->qlim->alignment_offset = 0;
960 		p->qlim->io_min = cpu_to_be32(queue_io_min(q));
961 		p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
962 		p->qlim->discard_enabled = 0;
963 
964 		d_size = 0;
965 		u_size = 0;
966 		q_order_type = QUEUE_ORDERED_NONE;
967 		max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
968 	}
969 
970 	if (peer_device->connection->agreed_pro_version <= 94)
971 		max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
972 	else if (peer_device->connection->agreed_pro_version < 100)
973 		max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
974 
975 	p->d_size = cpu_to_be64(d_size);
976 	p->u_size = cpu_to_be64(u_size);
977 	if (trigger_reply)
978 		p->c_size = 0;
979 	else
980 		p->c_size = cpu_to_be64(get_capacity(device->vdisk));
981 	p->max_bio_size = cpu_to_be32(max_bio_size);
982 	p->queue_order_type = cpu_to_be16(q_order_type);
983 	p->dds_flags = cpu_to_be16(flags);
984 
985 	return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
986 }
987 
988 /**
989  * drbd_send_current_state() - Sends the drbd state to the peer
990  * @peer_device:	DRBD peer device.
991  */
992 int drbd_send_current_state(struct drbd_peer_device *peer_device)
993 {
994 	struct drbd_socket *sock;
995 	struct p_state *p;
996 
997 	sock = &peer_device->connection->data;
998 	p = drbd_prepare_command(peer_device, sock);
999 	if (!p)
1000 		return -EIO;
1001 	p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1002 	return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1003 }
1004 
1005 /**
1006  * drbd_send_state() - After a state change, sends the new state to the peer
1007  * @peer_device:      DRBD peer device.
1008  * @state:     the state to send, not necessarily the current state.
1009  *
1010  * Each state change queues an "after_state_ch" work, which will eventually
1011  * send the resulting new state to the peer. If more state changes happen
1012  * between queuing and processing of the after_state_ch work, we still
1013  * want to send each intermediary state in the order it occurred.
1014  */
1015 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1016 {
1017 	struct drbd_socket *sock;
1018 	struct p_state *p;
1019 
1020 	sock = &peer_device->connection->data;
1021 	p = drbd_prepare_command(peer_device, sock);
1022 	if (!p)
1023 		return -EIO;
1024 	p->state = cpu_to_be32(state.i); /* Within the send mutex */
1025 	return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1026 }
1027 
1028 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1029 {
1030 	struct drbd_socket *sock;
1031 	struct p_req_state *p;
1032 
1033 	sock = &peer_device->connection->data;
1034 	p = drbd_prepare_command(peer_device, sock);
1035 	if (!p)
1036 		return -EIO;
1037 	p->mask = cpu_to_be32(mask.i);
1038 	p->val = cpu_to_be32(val.i);
1039 	return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1040 }
1041 
1042 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1043 {
1044 	enum drbd_packet cmd;
1045 	struct drbd_socket *sock;
1046 	struct p_req_state *p;
1047 
1048 	cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1049 	sock = &connection->data;
1050 	p = conn_prepare_command(connection, sock);
1051 	if (!p)
1052 		return -EIO;
1053 	p->mask = cpu_to_be32(mask.i);
1054 	p->val = cpu_to_be32(val.i);
1055 	return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1056 }
1057 
1058 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1059 {
1060 	struct drbd_socket *sock;
1061 	struct p_req_state_reply *p;
1062 
1063 	sock = &peer_device->connection->meta;
1064 	p = drbd_prepare_command(peer_device, sock);
1065 	if (p) {
1066 		p->retcode = cpu_to_be32(retcode);
1067 		drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1068 	}
1069 }
1070 
1071 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1072 {
1073 	struct drbd_socket *sock;
1074 	struct p_req_state_reply *p;
1075 	enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1076 
1077 	sock = &connection->meta;
1078 	p = conn_prepare_command(connection, sock);
1079 	if (p) {
1080 		p->retcode = cpu_to_be32(retcode);
1081 		conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1082 	}
1083 }
1084 
1085 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1086 {
1087 	BUG_ON(code & ~0xf);
1088 	p->encoding = (p->encoding & ~0xf) | code;
1089 }
1090 
1091 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1092 {
1093 	p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1094 }
1095 
1096 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1097 {
1098 	BUG_ON(n & ~0x7);
1099 	p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1100 }
1101 
1102 static int fill_bitmap_rle_bits(struct drbd_device *device,
1103 			 struct p_compressed_bm *p,
1104 			 unsigned int size,
1105 			 struct bm_xfer_ctx *c)
1106 {
1107 	struct bitstream bs;
1108 	unsigned long plain_bits;
1109 	unsigned long tmp;
1110 	unsigned long rl;
1111 	unsigned len;
1112 	unsigned toggle;
1113 	int bits, use_rle;
1114 
1115 	/* may we use this feature? */
1116 	rcu_read_lock();
1117 	use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1118 	rcu_read_unlock();
1119 	if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1120 		return 0;
1121 
1122 	if (c->bit_offset >= c->bm_bits)
1123 		return 0; /* nothing to do. */
1124 
1125 	/* use at most thus many bytes */
1126 	bitstream_init(&bs, p->code, size, 0);
1127 	memset(p->code, 0, size);
1128 	/* plain bits covered in this code string */
1129 	plain_bits = 0;
1130 
1131 	/* p->encoding & 0x80 stores whether the first run length is set.
1132 	 * bit offset is implicit.
1133 	 * start with toggle == 2 to be able to tell the first iteration */
1134 	toggle = 2;
1135 
1136 	/* see how much plain bits we can stuff into one packet
1137 	 * using RLE and VLI. */
1138 	do {
1139 		tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1140 				    : _drbd_bm_find_next(device, c->bit_offset);
1141 		if (tmp == -1UL)
1142 			tmp = c->bm_bits;
1143 		rl = tmp - c->bit_offset;
1144 
1145 		if (toggle == 2) { /* first iteration */
1146 			if (rl == 0) {
1147 				/* the first checked bit was set,
1148 				 * store start value, */
1149 				dcbp_set_start(p, 1);
1150 				/* but skip encoding of zero run length */
1151 				toggle = !toggle;
1152 				continue;
1153 			}
1154 			dcbp_set_start(p, 0);
1155 		}
1156 
1157 		/* paranoia: catch zero runlength.
1158 		 * can only happen if bitmap is modified while we scan it. */
1159 		if (rl == 0) {
1160 			drbd_err(device, "unexpected zero runlength while encoding bitmap "
1161 			    "t:%u bo:%lu\n", toggle, c->bit_offset);
1162 			return -1;
1163 		}
1164 
1165 		bits = vli_encode_bits(&bs, rl);
1166 		if (bits == -ENOBUFS) /* buffer full */
1167 			break;
1168 		if (bits <= 0) {
1169 			drbd_err(device, "error while encoding bitmap: %d\n", bits);
1170 			return 0;
1171 		}
1172 
1173 		toggle = !toggle;
1174 		plain_bits += rl;
1175 		c->bit_offset = tmp;
1176 	} while (c->bit_offset < c->bm_bits);
1177 
1178 	len = bs.cur.b - p->code + !!bs.cur.bit;
1179 
1180 	if (plain_bits < (len << 3)) {
1181 		/* incompressible with this method.
1182 		 * we need to rewind both word and bit position. */
1183 		c->bit_offset -= plain_bits;
1184 		bm_xfer_ctx_bit_to_word_offset(c);
1185 		c->bit_offset = c->word_offset * BITS_PER_LONG;
1186 		return 0;
1187 	}
1188 
1189 	/* RLE + VLI was able to compress it just fine.
1190 	 * update c->word_offset. */
1191 	bm_xfer_ctx_bit_to_word_offset(c);
1192 
1193 	/* store pad_bits */
1194 	dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1195 
1196 	return len;
1197 }
1198 
1199 /*
1200  * send_bitmap_rle_or_plain
1201  *
1202  * Return 0 when done, 1 when another iteration is needed, and a negative error
1203  * code upon failure.
1204  */
1205 static int
1206 send_bitmap_rle_or_plain(struct drbd_peer_device *peer_device, struct bm_xfer_ctx *c)
1207 {
1208 	struct drbd_device *device = peer_device->device;
1209 	struct drbd_socket *sock = &peer_device->connection->data;
1210 	unsigned int header_size = drbd_header_size(peer_device->connection);
1211 	struct p_compressed_bm *p = sock->sbuf + header_size;
1212 	int len, err;
1213 
1214 	len = fill_bitmap_rle_bits(device, p,
1215 			DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1216 	if (len < 0)
1217 		return -EIO;
1218 
1219 	if (len) {
1220 		dcbp_set_code(p, RLE_VLI_Bits);
1221 		err = __send_command(peer_device->connection, device->vnr, sock,
1222 				     P_COMPRESSED_BITMAP, sizeof(*p) + len,
1223 				     NULL, 0);
1224 		c->packets[0]++;
1225 		c->bytes[0] += header_size + sizeof(*p) + len;
1226 
1227 		if (c->bit_offset >= c->bm_bits)
1228 			len = 0; /* DONE */
1229 	} else {
1230 		/* was not compressible.
1231 		 * send a buffer full of plain text bits instead. */
1232 		unsigned int data_size;
1233 		unsigned long num_words;
1234 		unsigned long *p = sock->sbuf + header_size;
1235 
1236 		data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1237 		num_words = min_t(size_t, data_size / sizeof(*p),
1238 				  c->bm_words - c->word_offset);
1239 		len = num_words * sizeof(*p);
1240 		if (len)
1241 			drbd_bm_get_lel(device, c->word_offset, num_words, p);
1242 		err = __send_command(peer_device->connection, device->vnr, sock, P_BITMAP,
1243 				     len, NULL, 0);
1244 		c->word_offset += num_words;
1245 		c->bit_offset = c->word_offset * BITS_PER_LONG;
1246 
1247 		c->packets[1]++;
1248 		c->bytes[1] += header_size + len;
1249 
1250 		if (c->bit_offset > c->bm_bits)
1251 			c->bit_offset = c->bm_bits;
1252 	}
1253 	if (!err) {
1254 		if (len == 0) {
1255 			INFO_bm_xfer_stats(peer_device, "send", c);
1256 			return 0;
1257 		} else
1258 			return 1;
1259 	}
1260 	return -EIO;
1261 }
1262 
1263 /* See the comment at receive_bitmap() */
1264 static int _drbd_send_bitmap(struct drbd_device *device,
1265 			    struct drbd_peer_device *peer_device)
1266 {
1267 	struct bm_xfer_ctx c;
1268 	int err;
1269 
1270 	if (!expect(device, device->bitmap))
1271 		return false;
1272 
1273 	if (get_ldev(device)) {
1274 		if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1275 			drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1276 			drbd_bm_set_all(device);
1277 			if (drbd_bm_write(device, peer_device)) {
1278 				/* write_bm did fail! Leave full sync flag set in Meta P_DATA
1279 				 * but otherwise process as per normal - need to tell other
1280 				 * side that a full resync is required! */
1281 				drbd_err(device, "Failed to write bitmap to disk!\n");
1282 			} else {
1283 				drbd_md_clear_flag(device, MDF_FULL_SYNC);
1284 				drbd_md_sync(device);
1285 			}
1286 		}
1287 		put_ldev(device);
1288 	}
1289 
1290 	c = (struct bm_xfer_ctx) {
1291 		.bm_bits = drbd_bm_bits(device),
1292 		.bm_words = drbd_bm_words(device),
1293 	};
1294 
1295 	do {
1296 		err = send_bitmap_rle_or_plain(peer_device, &c);
1297 	} while (err > 0);
1298 
1299 	return err == 0;
1300 }
1301 
1302 int drbd_send_bitmap(struct drbd_device *device, struct drbd_peer_device *peer_device)
1303 {
1304 	struct drbd_socket *sock = &peer_device->connection->data;
1305 	int err = -1;
1306 
1307 	mutex_lock(&sock->mutex);
1308 	if (sock->socket)
1309 		err = !_drbd_send_bitmap(device, peer_device);
1310 	mutex_unlock(&sock->mutex);
1311 	return err;
1312 }
1313 
1314 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1315 {
1316 	struct drbd_socket *sock;
1317 	struct p_barrier_ack *p;
1318 
1319 	if (connection->cstate < C_WF_REPORT_PARAMS)
1320 		return;
1321 
1322 	sock = &connection->meta;
1323 	p = conn_prepare_command(connection, sock);
1324 	if (!p)
1325 		return;
1326 	p->barrier = barrier_nr;
1327 	p->set_size = cpu_to_be32(set_size);
1328 	conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1329 }
1330 
1331 /**
1332  * _drbd_send_ack() - Sends an ack packet
1333  * @peer_device:	DRBD peer device.
1334  * @cmd:		Packet command code.
1335  * @sector:		sector, needs to be in big endian byte order
1336  * @blksize:		size in byte, needs to be in big endian byte order
1337  * @block_id:		Id, big endian byte order
1338  */
1339 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1340 			  u64 sector, u32 blksize, u64 block_id)
1341 {
1342 	struct drbd_socket *sock;
1343 	struct p_block_ack *p;
1344 
1345 	if (peer_device->device->state.conn < C_CONNECTED)
1346 		return -EIO;
1347 
1348 	sock = &peer_device->connection->meta;
1349 	p = drbd_prepare_command(peer_device, sock);
1350 	if (!p)
1351 		return -EIO;
1352 	p->sector = sector;
1353 	p->block_id = block_id;
1354 	p->blksize = blksize;
1355 	p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1356 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1357 }
1358 
1359 /* dp->sector and dp->block_id already/still in network byte order,
1360  * data_size is payload size according to dp->head,
1361  * and may need to be corrected for digest size. */
1362 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1363 		      struct p_data *dp, int data_size)
1364 {
1365 	if (peer_device->connection->peer_integrity_tfm)
1366 		data_size -= crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm);
1367 	_drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1368 		       dp->block_id);
1369 }
1370 
1371 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1372 		      struct p_block_req *rp)
1373 {
1374 	_drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1375 }
1376 
1377 /**
1378  * drbd_send_ack() - Sends an ack packet
1379  * @peer_device:	DRBD peer device
1380  * @cmd:		packet command code
1381  * @peer_req:		peer request
1382  */
1383 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1384 		  struct drbd_peer_request *peer_req)
1385 {
1386 	return _drbd_send_ack(peer_device, cmd,
1387 			      cpu_to_be64(peer_req->i.sector),
1388 			      cpu_to_be32(peer_req->i.size),
1389 			      peer_req->block_id);
1390 }
1391 
1392 /* This function misuses the block_id field to signal if the blocks
1393  * are is sync or not. */
1394 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1395 		     sector_t sector, int blksize, u64 block_id)
1396 {
1397 	return _drbd_send_ack(peer_device, cmd,
1398 			      cpu_to_be64(sector),
1399 			      cpu_to_be32(blksize),
1400 			      cpu_to_be64(block_id));
1401 }
1402 
1403 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1404 			     struct drbd_peer_request *peer_req)
1405 {
1406 	struct drbd_socket *sock;
1407 	struct p_block_desc *p;
1408 
1409 	sock = &peer_device->connection->data;
1410 	p = drbd_prepare_command(peer_device, sock);
1411 	if (!p)
1412 		return -EIO;
1413 	p->sector = cpu_to_be64(peer_req->i.sector);
1414 	p->blksize = cpu_to_be32(peer_req->i.size);
1415 	p->pad = 0;
1416 	return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1417 }
1418 
1419 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1420 		       sector_t sector, int size, u64 block_id)
1421 {
1422 	struct drbd_socket *sock;
1423 	struct p_block_req *p;
1424 
1425 	sock = &peer_device->connection->data;
1426 	p = drbd_prepare_command(peer_device, sock);
1427 	if (!p)
1428 		return -EIO;
1429 	p->sector = cpu_to_be64(sector);
1430 	p->block_id = block_id;
1431 	p->blksize = cpu_to_be32(size);
1432 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1433 }
1434 
1435 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1436 			    void *digest, int digest_size, enum drbd_packet cmd)
1437 {
1438 	struct drbd_socket *sock;
1439 	struct p_block_req *p;
1440 
1441 	/* FIXME: Put the digest into the preallocated socket buffer.  */
1442 
1443 	sock = &peer_device->connection->data;
1444 	p = drbd_prepare_command(peer_device, sock);
1445 	if (!p)
1446 		return -EIO;
1447 	p->sector = cpu_to_be64(sector);
1448 	p->block_id = ID_SYNCER /* unused */;
1449 	p->blksize = cpu_to_be32(size);
1450 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1451 }
1452 
1453 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1454 {
1455 	struct drbd_socket *sock;
1456 	struct p_block_req *p;
1457 
1458 	sock = &peer_device->connection->data;
1459 	p = drbd_prepare_command(peer_device, sock);
1460 	if (!p)
1461 		return -EIO;
1462 	p->sector = cpu_to_be64(sector);
1463 	p->block_id = ID_SYNCER /* unused */;
1464 	p->blksize = cpu_to_be32(size);
1465 	return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1466 }
1467 
1468 /* called on sndtimeo
1469  * returns false if we should retry,
1470  * true if we think connection is dead
1471  */
1472 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1473 {
1474 	int drop_it;
1475 	/* long elapsed = (long)(jiffies - device->last_received); */
1476 
1477 	drop_it =   connection->meta.socket == sock
1478 		|| !connection->ack_receiver.task
1479 		|| get_t_state(&connection->ack_receiver) != RUNNING
1480 		|| connection->cstate < C_WF_REPORT_PARAMS;
1481 
1482 	if (drop_it)
1483 		return true;
1484 
1485 	drop_it = !--connection->ko_count;
1486 	if (!drop_it) {
1487 		drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1488 			 current->comm, current->pid, connection->ko_count);
1489 		request_ping(connection);
1490 	}
1491 
1492 	return drop_it; /* && (device->state == R_PRIMARY) */;
1493 }
1494 
1495 static void drbd_update_congested(struct drbd_connection *connection)
1496 {
1497 	struct sock *sk = connection->data.socket->sk;
1498 	if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1499 		set_bit(NET_CONGESTED, &connection->flags);
1500 }
1501 
1502 /* The idea of sendpage seems to be to put some kind of reference
1503  * to the page into the skb, and to hand it over to the NIC. In
1504  * this process get_page() gets called.
1505  *
1506  * As soon as the page was really sent over the network put_page()
1507  * gets called by some part of the network layer. [ NIC driver? ]
1508  *
1509  * [ get_page() / put_page() increment/decrement the count. If count
1510  *   reaches 0 the page will be freed. ]
1511  *
1512  * This works nicely with pages from FSs.
1513  * But this means that in protocol A we might signal IO completion too early!
1514  *
1515  * In order not to corrupt data during a resync we must make sure
1516  * that we do not reuse our own buffer pages (EEs) to early, therefore
1517  * we have the net_ee list.
1518  *
1519  * XFS seems to have problems, still, it submits pages with page_count == 0!
1520  * As a workaround, we disable sendpage on pages
1521  * with page_count == 0 or PageSlab.
1522  */
1523 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1524 			      int offset, size_t size, unsigned msg_flags)
1525 {
1526 	struct socket *socket;
1527 	void *addr;
1528 	int err;
1529 
1530 	socket = peer_device->connection->data.socket;
1531 	addr = kmap(page) + offset;
1532 	err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1533 	kunmap(page);
1534 	if (!err)
1535 		peer_device->device->send_cnt += size >> 9;
1536 	return err;
1537 }
1538 
1539 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1540 		    int offset, size_t size, unsigned msg_flags)
1541 {
1542 	struct socket *socket = peer_device->connection->data.socket;
1543 	int len = size;
1544 	int err = -EIO;
1545 
1546 	/* e.g. XFS meta- & log-data is in slab pages, which have a
1547 	 * page_count of 0 and/or have PageSlab() set.
1548 	 * we cannot use send_page for those, as that does get_page();
1549 	 * put_page(); and would cause either a VM_BUG directly, or
1550 	 * __page_cache_release a page that would actually still be referenced
1551 	 * by someone, leading to some obscure delayed Oops somewhere else. */
1552 	if (drbd_disable_sendpage || !sendpage_ok(page))
1553 		return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1554 
1555 	msg_flags |= MSG_NOSIGNAL;
1556 	drbd_update_congested(peer_device->connection);
1557 	do {
1558 		int sent;
1559 
1560 		sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1561 		if (sent <= 0) {
1562 			if (sent == -EAGAIN) {
1563 				if (we_should_drop_the_connection(peer_device->connection, socket))
1564 					break;
1565 				continue;
1566 			}
1567 			drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1568 			     __func__, (int)size, len, sent);
1569 			if (sent < 0)
1570 				err = sent;
1571 			break;
1572 		}
1573 		len    -= sent;
1574 		offset += sent;
1575 	} while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1576 	clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1577 
1578 	if (len == 0) {
1579 		err = 0;
1580 		peer_device->device->send_cnt += size >> 9;
1581 	}
1582 	return err;
1583 }
1584 
1585 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1586 {
1587 	struct bio_vec bvec;
1588 	struct bvec_iter iter;
1589 
1590 	/* hint all but last page with MSG_MORE */
1591 	bio_for_each_segment(bvec, bio, iter) {
1592 		int err;
1593 
1594 		err = _drbd_no_send_page(peer_device, bvec.bv_page,
1595 					 bvec.bv_offset, bvec.bv_len,
1596 					 bio_iter_last(bvec, iter)
1597 					 ? 0 : MSG_MORE);
1598 		if (err)
1599 			return err;
1600 	}
1601 	return 0;
1602 }
1603 
1604 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1605 {
1606 	struct bio_vec bvec;
1607 	struct bvec_iter iter;
1608 
1609 	/* hint all but last page with MSG_MORE */
1610 	bio_for_each_segment(bvec, bio, iter) {
1611 		int err;
1612 
1613 		err = _drbd_send_page(peer_device, bvec.bv_page,
1614 				      bvec.bv_offset, bvec.bv_len,
1615 				      bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1616 		if (err)
1617 			return err;
1618 	}
1619 	return 0;
1620 }
1621 
1622 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1623 			    struct drbd_peer_request *peer_req)
1624 {
1625 	struct page *page = peer_req->pages;
1626 	unsigned len = peer_req->i.size;
1627 	int err;
1628 
1629 	/* hint all but last page with MSG_MORE */
1630 	page_chain_for_each(page) {
1631 		unsigned l = min_t(unsigned, len, PAGE_SIZE);
1632 
1633 		err = _drbd_send_page(peer_device, page, 0, l,
1634 				      page_chain_next(page) ? MSG_MORE : 0);
1635 		if (err)
1636 			return err;
1637 		len -= l;
1638 	}
1639 	return 0;
1640 }
1641 
1642 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1643 			     struct bio *bio)
1644 {
1645 	if (connection->agreed_pro_version >= 95)
1646 		return  (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1647 			(bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1648 			(bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1649 			(bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1650 			(bio_op(bio) == REQ_OP_WRITE_ZEROES ?
1651 			  ((connection->agreed_features & DRBD_FF_WZEROES) ?
1652 			   (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0))
1653 			   : DP_DISCARD)
1654 			: 0);
1655 	else
1656 		return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1657 }
1658 
1659 /* Used to send write or TRIM aka REQ_OP_DISCARD requests
1660  * R_PRIMARY -> Peer	(P_DATA, P_TRIM)
1661  */
1662 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1663 {
1664 	struct drbd_device *device = peer_device->device;
1665 	struct drbd_socket *sock;
1666 	struct p_data *p;
1667 	void *digest_out;
1668 	unsigned int dp_flags = 0;
1669 	int digest_size;
1670 	int err;
1671 
1672 	sock = &peer_device->connection->data;
1673 	p = drbd_prepare_command(peer_device, sock);
1674 	digest_size = peer_device->connection->integrity_tfm ?
1675 		      crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1676 
1677 	if (!p)
1678 		return -EIO;
1679 	p->sector = cpu_to_be64(req->i.sector);
1680 	p->block_id = (unsigned long)req;
1681 	p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1682 	dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1683 	if (device->state.conn >= C_SYNC_SOURCE &&
1684 	    device->state.conn <= C_PAUSED_SYNC_T)
1685 		dp_flags |= DP_MAY_SET_IN_SYNC;
1686 	if (peer_device->connection->agreed_pro_version >= 100) {
1687 		if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1688 			dp_flags |= DP_SEND_RECEIVE_ACK;
1689 		/* During resync, request an explicit write ack,
1690 		 * even in protocol != C */
1691 		if (req->rq_state & RQ_EXP_WRITE_ACK
1692 		|| (dp_flags & DP_MAY_SET_IN_SYNC))
1693 			dp_flags |= DP_SEND_WRITE_ACK;
1694 	}
1695 	p->dp_flags = cpu_to_be32(dp_flags);
1696 
1697 	if (dp_flags & (DP_DISCARD|DP_ZEROES)) {
1698 		enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM;
1699 		struct p_trim *t = (struct p_trim*)p;
1700 		t->size = cpu_to_be32(req->i.size);
1701 		err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0);
1702 		goto out;
1703 	}
1704 	digest_out = p + 1;
1705 
1706 	/* our digest is still only over the payload.
1707 	 * TRIM does not carry any payload. */
1708 	if (digest_size)
1709 		drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1710 	err = __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1711 			     sizeof(*p) + digest_size, NULL, req->i.size);
1712 	if (!err) {
1713 		/* For protocol A, we have to memcpy the payload into
1714 		 * socket buffers, as we may complete right away
1715 		 * as soon as we handed it over to tcp, at which point the data
1716 		 * pages may become invalid.
1717 		 *
1718 		 * For data-integrity enabled, we copy it as well, so we can be
1719 		 * sure that even if the bio pages may still be modified, it
1720 		 * won't change the data on the wire, thus if the digest checks
1721 		 * out ok after sending on this side, but does not fit on the
1722 		 * receiving side, we sure have detected corruption elsewhere.
1723 		 */
1724 		if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1725 			err = _drbd_send_bio(peer_device, req->master_bio);
1726 		else
1727 			err = _drbd_send_zc_bio(peer_device, req->master_bio);
1728 
1729 		/* double check digest, sometimes buffers have been modified in flight. */
1730 		if (digest_size > 0 && digest_size <= 64) {
1731 			/* 64 byte, 512 bit, is the largest digest size
1732 			 * currently supported in kernel crypto. */
1733 			unsigned char digest[64];
1734 			drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1735 			if (memcmp(p + 1, digest, digest_size)) {
1736 				drbd_warn(device,
1737 					"Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1738 					(unsigned long long)req->i.sector, req->i.size);
1739 			}
1740 		} /* else if (digest_size > 64) {
1741 		     ... Be noisy about digest too large ...
1742 		} */
1743 	}
1744 out:
1745 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1746 
1747 	return err;
1748 }
1749 
1750 /* answer packet, used to send data back for read requests:
1751  *  Peer       -> (diskless) R_PRIMARY   (P_DATA_REPLY)
1752  *  C_SYNC_SOURCE -> C_SYNC_TARGET         (P_RS_DATA_REPLY)
1753  */
1754 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1755 		    struct drbd_peer_request *peer_req)
1756 {
1757 	struct drbd_device *device = peer_device->device;
1758 	struct drbd_socket *sock;
1759 	struct p_data *p;
1760 	int err;
1761 	int digest_size;
1762 
1763 	sock = &peer_device->connection->data;
1764 	p = drbd_prepare_command(peer_device, sock);
1765 
1766 	digest_size = peer_device->connection->integrity_tfm ?
1767 		      crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1768 
1769 	if (!p)
1770 		return -EIO;
1771 	p->sector = cpu_to_be64(peer_req->i.sector);
1772 	p->block_id = peer_req->block_id;
1773 	p->seq_num = 0;  /* unused */
1774 	p->dp_flags = 0;
1775 	if (digest_size)
1776 		drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1777 	err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1778 	if (!err)
1779 		err = _drbd_send_zc_ee(peer_device, peer_req);
1780 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1781 
1782 	return err;
1783 }
1784 
1785 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1786 {
1787 	struct drbd_socket *sock;
1788 	struct p_block_desc *p;
1789 
1790 	sock = &peer_device->connection->data;
1791 	p = drbd_prepare_command(peer_device, sock);
1792 	if (!p)
1793 		return -EIO;
1794 	p->sector = cpu_to_be64(req->i.sector);
1795 	p->blksize = cpu_to_be32(req->i.size);
1796 	return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1797 }
1798 
1799 /*
1800   drbd_send distinguishes two cases:
1801 
1802   Packets sent via the data socket "sock"
1803   and packets sent via the meta data socket "msock"
1804 
1805 		    sock                      msock
1806   -----------------+-------------------------+------------------------------
1807   timeout           conf.timeout / 2          conf.timeout / 2
1808   timeout action    send a ping via msock     Abort communication
1809 					      and close all sockets
1810 */
1811 
1812 /*
1813  * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1814  */
1815 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1816 	      void *buf, size_t size, unsigned msg_flags)
1817 {
1818 	struct kvec iov = {.iov_base = buf, .iov_len = size};
1819 	struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL};
1820 	int rv, sent = 0;
1821 
1822 	if (!sock)
1823 		return -EBADR;
1824 
1825 	/* THINK  if (signal_pending) return ... ? */
1826 
1827 	iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, &iov, 1, size);
1828 
1829 	if (sock == connection->data.socket) {
1830 		rcu_read_lock();
1831 		connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1832 		rcu_read_unlock();
1833 		drbd_update_congested(connection);
1834 	}
1835 	do {
1836 		rv = sock_sendmsg(sock, &msg);
1837 		if (rv == -EAGAIN) {
1838 			if (we_should_drop_the_connection(connection, sock))
1839 				break;
1840 			else
1841 				continue;
1842 		}
1843 		if (rv == -EINTR) {
1844 			flush_signals(current);
1845 			rv = 0;
1846 		}
1847 		if (rv < 0)
1848 			break;
1849 		sent += rv;
1850 	} while (sent < size);
1851 
1852 	if (sock == connection->data.socket)
1853 		clear_bit(NET_CONGESTED, &connection->flags);
1854 
1855 	if (rv <= 0) {
1856 		if (rv != -EAGAIN) {
1857 			drbd_err(connection, "%s_sendmsg returned %d\n",
1858 				 sock == connection->meta.socket ? "msock" : "sock",
1859 				 rv);
1860 			conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1861 		} else
1862 			conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1863 	}
1864 
1865 	return sent;
1866 }
1867 
1868 /*
1869  * drbd_send_all  -  Send an entire buffer
1870  *
1871  * Returns 0 upon success and a negative error value otherwise.
1872  */
1873 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1874 		  size_t size, unsigned msg_flags)
1875 {
1876 	int err;
1877 
1878 	err = drbd_send(connection, sock, buffer, size, msg_flags);
1879 	if (err < 0)
1880 		return err;
1881 	if (err != size)
1882 		return -EIO;
1883 	return 0;
1884 }
1885 
1886 static int drbd_open(struct block_device *bdev, fmode_t mode)
1887 {
1888 	struct drbd_device *device = bdev->bd_disk->private_data;
1889 	unsigned long flags;
1890 	int rv = 0;
1891 
1892 	mutex_lock(&drbd_main_mutex);
1893 	spin_lock_irqsave(&device->resource->req_lock, flags);
1894 	/* to have a stable device->state.role
1895 	 * and no race with updating open_cnt */
1896 
1897 	if (device->state.role != R_PRIMARY) {
1898 		if (mode & FMODE_WRITE)
1899 			rv = -EROFS;
1900 		else if (!drbd_allow_oos)
1901 			rv = -EMEDIUMTYPE;
1902 	}
1903 
1904 	if (!rv)
1905 		device->open_cnt++;
1906 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
1907 	mutex_unlock(&drbd_main_mutex);
1908 
1909 	return rv;
1910 }
1911 
1912 static void drbd_release(struct gendisk *gd, fmode_t mode)
1913 {
1914 	struct drbd_device *device = gd->private_data;
1915 	mutex_lock(&drbd_main_mutex);
1916 	device->open_cnt--;
1917 	mutex_unlock(&drbd_main_mutex);
1918 }
1919 
1920 /* need to hold resource->req_lock */
1921 void drbd_queue_unplug(struct drbd_device *device)
1922 {
1923 	if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1924 		D_ASSERT(device, device->state.role == R_PRIMARY);
1925 		if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1926 			drbd_queue_work_if_unqueued(
1927 				&first_peer_device(device)->connection->sender_work,
1928 				&device->unplug_work);
1929 		}
1930 	}
1931 }
1932 
1933 static void drbd_set_defaults(struct drbd_device *device)
1934 {
1935 	/* Beware! The actual layout differs
1936 	 * between big endian and little endian */
1937 	device->state = (union drbd_dev_state) {
1938 		{ .role = R_SECONDARY,
1939 		  .peer = R_UNKNOWN,
1940 		  .conn = C_STANDALONE,
1941 		  .disk = D_DISKLESS,
1942 		  .pdsk = D_UNKNOWN,
1943 		} };
1944 }
1945 
1946 void drbd_init_set_defaults(struct drbd_device *device)
1947 {
1948 	/* the memset(,0,) did most of this.
1949 	 * note: only assignments, no allocation in here */
1950 
1951 	drbd_set_defaults(device);
1952 
1953 	atomic_set(&device->ap_bio_cnt, 0);
1954 	atomic_set(&device->ap_actlog_cnt, 0);
1955 	atomic_set(&device->ap_pending_cnt, 0);
1956 	atomic_set(&device->rs_pending_cnt, 0);
1957 	atomic_set(&device->unacked_cnt, 0);
1958 	atomic_set(&device->local_cnt, 0);
1959 	atomic_set(&device->pp_in_use_by_net, 0);
1960 	atomic_set(&device->rs_sect_in, 0);
1961 	atomic_set(&device->rs_sect_ev, 0);
1962 	atomic_set(&device->ap_in_flight, 0);
1963 	atomic_set(&device->md_io.in_use, 0);
1964 
1965 	mutex_init(&device->own_state_mutex);
1966 	device->state_mutex = &device->own_state_mutex;
1967 
1968 	spin_lock_init(&device->al_lock);
1969 	spin_lock_init(&device->peer_seq_lock);
1970 
1971 	INIT_LIST_HEAD(&device->active_ee);
1972 	INIT_LIST_HEAD(&device->sync_ee);
1973 	INIT_LIST_HEAD(&device->done_ee);
1974 	INIT_LIST_HEAD(&device->read_ee);
1975 	INIT_LIST_HEAD(&device->net_ee);
1976 	INIT_LIST_HEAD(&device->resync_reads);
1977 	INIT_LIST_HEAD(&device->resync_work.list);
1978 	INIT_LIST_HEAD(&device->unplug_work.list);
1979 	INIT_LIST_HEAD(&device->bm_io_work.w.list);
1980 	INIT_LIST_HEAD(&device->pending_master_completion[0]);
1981 	INIT_LIST_HEAD(&device->pending_master_completion[1]);
1982 	INIT_LIST_HEAD(&device->pending_completion[0]);
1983 	INIT_LIST_HEAD(&device->pending_completion[1]);
1984 
1985 	device->resync_work.cb  = w_resync_timer;
1986 	device->unplug_work.cb  = w_send_write_hint;
1987 	device->bm_io_work.w.cb = w_bitmap_io;
1988 
1989 	timer_setup(&device->resync_timer, resync_timer_fn, 0);
1990 	timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0);
1991 	timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0);
1992 	timer_setup(&device->request_timer, request_timer_fn, 0);
1993 
1994 	init_waitqueue_head(&device->misc_wait);
1995 	init_waitqueue_head(&device->state_wait);
1996 	init_waitqueue_head(&device->ee_wait);
1997 	init_waitqueue_head(&device->al_wait);
1998 	init_waitqueue_head(&device->seq_wait);
1999 
2000 	device->resync_wenr = LC_FREE;
2001 	device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2002 	device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2003 }
2004 
2005 void drbd_set_my_capacity(struct drbd_device *device, sector_t size)
2006 {
2007 	char ppb[10];
2008 
2009 	set_capacity_and_notify(device->vdisk, size);
2010 
2011 	drbd_info(device, "size = %s (%llu KB)\n",
2012 		ppsize(ppb, size>>1), (unsigned long long)size>>1);
2013 }
2014 
2015 void drbd_device_cleanup(struct drbd_device *device)
2016 {
2017 	int i;
2018 	if (first_peer_device(device)->connection->receiver.t_state != NONE)
2019 		drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2020 				first_peer_device(device)->connection->receiver.t_state);
2021 
2022 	device->al_writ_cnt  =
2023 	device->bm_writ_cnt  =
2024 	device->read_cnt     =
2025 	device->recv_cnt     =
2026 	device->send_cnt     =
2027 	device->writ_cnt     =
2028 	device->p_size       =
2029 	device->rs_start     =
2030 	device->rs_total     =
2031 	device->rs_failed    = 0;
2032 	device->rs_last_events = 0;
2033 	device->rs_last_sect_ev = 0;
2034 	for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2035 		device->rs_mark_left[i] = 0;
2036 		device->rs_mark_time[i] = 0;
2037 	}
2038 	D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2039 
2040 	set_capacity_and_notify(device->vdisk, 0);
2041 	if (device->bitmap) {
2042 		/* maybe never allocated. */
2043 		drbd_bm_resize(device, 0, 1);
2044 		drbd_bm_cleanup(device);
2045 	}
2046 
2047 	drbd_backing_dev_free(device, device->ldev);
2048 	device->ldev = NULL;
2049 
2050 	clear_bit(AL_SUSPENDED, &device->flags);
2051 
2052 	D_ASSERT(device, list_empty(&device->active_ee));
2053 	D_ASSERT(device, list_empty(&device->sync_ee));
2054 	D_ASSERT(device, list_empty(&device->done_ee));
2055 	D_ASSERT(device, list_empty(&device->read_ee));
2056 	D_ASSERT(device, list_empty(&device->net_ee));
2057 	D_ASSERT(device, list_empty(&device->resync_reads));
2058 	D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2059 	D_ASSERT(device, list_empty(&device->resync_work.list));
2060 	D_ASSERT(device, list_empty(&device->unplug_work.list));
2061 
2062 	drbd_set_defaults(device);
2063 }
2064 
2065 
2066 static void drbd_destroy_mempools(void)
2067 {
2068 	struct page *page;
2069 
2070 	while (drbd_pp_pool) {
2071 		page = drbd_pp_pool;
2072 		drbd_pp_pool = (struct page *)page_private(page);
2073 		__free_page(page);
2074 		drbd_pp_vacant--;
2075 	}
2076 
2077 	/* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2078 
2079 	bioset_exit(&drbd_io_bio_set);
2080 	bioset_exit(&drbd_md_io_bio_set);
2081 	mempool_exit(&drbd_md_io_page_pool);
2082 	mempool_exit(&drbd_ee_mempool);
2083 	mempool_exit(&drbd_request_mempool);
2084 	kmem_cache_destroy(drbd_ee_cache);
2085 	kmem_cache_destroy(drbd_request_cache);
2086 	kmem_cache_destroy(drbd_bm_ext_cache);
2087 	kmem_cache_destroy(drbd_al_ext_cache);
2088 
2089 	drbd_ee_cache        = NULL;
2090 	drbd_request_cache   = NULL;
2091 	drbd_bm_ext_cache    = NULL;
2092 	drbd_al_ext_cache    = NULL;
2093 
2094 	return;
2095 }
2096 
2097 static int drbd_create_mempools(void)
2098 {
2099 	struct page *page;
2100 	const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2101 	int i, ret;
2102 
2103 	/* caches */
2104 	drbd_request_cache = kmem_cache_create(
2105 		"drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2106 	if (drbd_request_cache == NULL)
2107 		goto Enomem;
2108 
2109 	drbd_ee_cache = kmem_cache_create(
2110 		"drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2111 	if (drbd_ee_cache == NULL)
2112 		goto Enomem;
2113 
2114 	drbd_bm_ext_cache = kmem_cache_create(
2115 		"drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2116 	if (drbd_bm_ext_cache == NULL)
2117 		goto Enomem;
2118 
2119 	drbd_al_ext_cache = kmem_cache_create(
2120 		"drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2121 	if (drbd_al_ext_cache == NULL)
2122 		goto Enomem;
2123 
2124 	/* mempools */
2125 	ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0);
2126 	if (ret)
2127 		goto Enomem;
2128 
2129 	ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0,
2130 			  BIOSET_NEED_BVECS);
2131 	if (ret)
2132 		goto Enomem;
2133 
2134 	ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0);
2135 	if (ret)
2136 		goto Enomem;
2137 
2138 	ret = mempool_init_slab_pool(&drbd_request_mempool, number,
2139 				     drbd_request_cache);
2140 	if (ret)
2141 		goto Enomem;
2142 
2143 	ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache);
2144 	if (ret)
2145 		goto Enomem;
2146 
2147 	for (i = 0; i < number; i++) {
2148 		page = alloc_page(GFP_HIGHUSER);
2149 		if (!page)
2150 			goto Enomem;
2151 		set_page_private(page, (unsigned long)drbd_pp_pool);
2152 		drbd_pp_pool = page;
2153 	}
2154 	drbd_pp_vacant = number;
2155 
2156 	return 0;
2157 
2158 Enomem:
2159 	drbd_destroy_mempools(); /* in case we allocated some */
2160 	return -ENOMEM;
2161 }
2162 
2163 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2164 {
2165 	int rr;
2166 
2167 	rr = drbd_free_peer_reqs(device, &device->active_ee);
2168 	if (rr)
2169 		drbd_err(device, "%d EEs in active list found!\n", rr);
2170 
2171 	rr = drbd_free_peer_reqs(device, &device->sync_ee);
2172 	if (rr)
2173 		drbd_err(device, "%d EEs in sync list found!\n", rr);
2174 
2175 	rr = drbd_free_peer_reqs(device, &device->read_ee);
2176 	if (rr)
2177 		drbd_err(device, "%d EEs in read list found!\n", rr);
2178 
2179 	rr = drbd_free_peer_reqs(device, &device->done_ee);
2180 	if (rr)
2181 		drbd_err(device, "%d EEs in done list found!\n", rr);
2182 
2183 	rr = drbd_free_peer_reqs(device, &device->net_ee);
2184 	if (rr)
2185 		drbd_err(device, "%d EEs in net list found!\n", rr);
2186 }
2187 
2188 /* caution. no locking. */
2189 void drbd_destroy_device(struct kref *kref)
2190 {
2191 	struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2192 	struct drbd_resource *resource = device->resource;
2193 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2194 
2195 	timer_shutdown_sync(&device->request_timer);
2196 
2197 	/* paranoia asserts */
2198 	D_ASSERT(device, device->open_cnt == 0);
2199 	/* end paranoia asserts */
2200 
2201 	/* cleanup stuff that may have been allocated during
2202 	 * device (re-)configuration or state changes */
2203 
2204 	drbd_backing_dev_free(device, device->ldev);
2205 	device->ldev = NULL;
2206 
2207 	drbd_release_all_peer_reqs(device);
2208 
2209 	lc_destroy(device->act_log);
2210 	lc_destroy(device->resync);
2211 
2212 	kfree(device->p_uuid);
2213 	/* device->p_uuid = NULL; */
2214 
2215 	if (device->bitmap) /* should no longer be there. */
2216 		drbd_bm_cleanup(device);
2217 	__free_page(device->md_io.page);
2218 	put_disk(device->vdisk);
2219 	kfree(device->rs_plan_s);
2220 
2221 	/* not for_each_connection(connection, resource):
2222 	 * those may have been cleaned up and disassociated already.
2223 	 */
2224 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2225 		kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2226 		kfree(peer_device);
2227 	}
2228 	if (device->submit.wq)
2229 		destroy_workqueue(device->submit.wq);
2230 	kfree(device);
2231 	kref_put(&resource->kref, drbd_destroy_resource);
2232 }
2233 
2234 /* One global retry thread, if we need to push back some bio and have it
2235  * reinserted through our make request function.
2236  */
2237 static struct retry_worker {
2238 	struct workqueue_struct *wq;
2239 	struct work_struct worker;
2240 
2241 	spinlock_t lock;
2242 	struct list_head writes;
2243 } retry;
2244 
2245 static void do_retry(struct work_struct *ws)
2246 {
2247 	struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2248 	LIST_HEAD(writes);
2249 	struct drbd_request *req, *tmp;
2250 
2251 	spin_lock_irq(&retry->lock);
2252 	list_splice_init(&retry->writes, &writes);
2253 	spin_unlock_irq(&retry->lock);
2254 
2255 	list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2256 		struct drbd_device *device = req->device;
2257 		struct bio *bio = req->master_bio;
2258 		bool expected;
2259 
2260 		expected =
2261 			expect(device, atomic_read(&req->completion_ref) == 0) &&
2262 			expect(device, req->rq_state & RQ_POSTPONED) &&
2263 			expect(device, (req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2264 				(req->rq_state & RQ_LOCAL_ABORTED) != 0);
2265 
2266 		if (!expected)
2267 			drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2268 				req, atomic_read(&req->completion_ref),
2269 				req->rq_state);
2270 
2271 		/* We still need to put one kref associated with the
2272 		 * "completion_ref" going zero in the code path that queued it
2273 		 * here.  The request object may still be referenced by a
2274 		 * frozen local req->private_bio, in case we force-detached.
2275 		 */
2276 		kref_put(&req->kref, drbd_req_destroy);
2277 
2278 		/* A single suspended or otherwise blocking device may stall
2279 		 * all others as well.  Fortunately, this code path is to
2280 		 * recover from a situation that "should not happen":
2281 		 * concurrent writes in multi-primary setup.
2282 		 * In a "normal" lifecycle, this workqueue is supposed to be
2283 		 * destroyed without ever doing anything.
2284 		 * If it turns out to be an issue anyways, we can do per
2285 		 * resource (replication group) or per device (minor) retry
2286 		 * workqueues instead.
2287 		 */
2288 
2289 		/* We are not just doing submit_bio_noacct(),
2290 		 * as we want to keep the start_time information. */
2291 		inc_ap_bio(device);
2292 		__drbd_make_request(device, bio);
2293 	}
2294 }
2295 
2296 /* called via drbd_req_put_completion_ref(),
2297  * holds resource->req_lock */
2298 void drbd_restart_request(struct drbd_request *req)
2299 {
2300 	unsigned long flags;
2301 	spin_lock_irqsave(&retry.lock, flags);
2302 	list_move_tail(&req->tl_requests, &retry.writes);
2303 	spin_unlock_irqrestore(&retry.lock, flags);
2304 
2305 	/* Drop the extra reference that would otherwise
2306 	 * have been dropped by complete_master_bio.
2307 	 * do_retry() needs to grab a new one. */
2308 	dec_ap_bio(req->device);
2309 
2310 	queue_work(retry.wq, &retry.worker);
2311 }
2312 
2313 void drbd_destroy_resource(struct kref *kref)
2314 {
2315 	struct drbd_resource *resource =
2316 		container_of(kref, struct drbd_resource, kref);
2317 
2318 	idr_destroy(&resource->devices);
2319 	free_cpumask_var(resource->cpu_mask);
2320 	kfree(resource->name);
2321 	kfree(resource);
2322 }
2323 
2324 void drbd_free_resource(struct drbd_resource *resource)
2325 {
2326 	struct drbd_connection *connection, *tmp;
2327 
2328 	for_each_connection_safe(connection, tmp, resource) {
2329 		list_del(&connection->connections);
2330 		drbd_debugfs_connection_cleanup(connection);
2331 		kref_put(&connection->kref, drbd_destroy_connection);
2332 	}
2333 	drbd_debugfs_resource_cleanup(resource);
2334 	kref_put(&resource->kref, drbd_destroy_resource);
2335 }
2336 
2337 static void drbd_cleanup(void)
2338 {
2339 	unsigned int i;
2340 	struct drbd_device *device;
2341 	struct drbd_resource *resource, *tmp;
2342 
2343 	/* first remove proc,
2344 	 * drbdsetup uses it's presence to detect
2345 	 * whether DRBD is loaded.
2346 	 * If we would get stuck in proc removal,
2347 	 * but have netlink already deregistered,
2348 	 * some drbdsetup commands may wait forever
2349 	 * for an answer.
2350 	 */
2351 	if (drbd_proc)
2352 		remove_proc_entry("drbd", NULL);
2353 
2354 	if (retry.wq)
2355 		destroy_workqueue(retry.wq);
2356 
2357 	drbd_genl_unregister();
2358 
2359 	idr_for_each_entry(&drbd_devices, device, i)
2360 		drbd_delete_device(device);
2361 
2362 	/* not _rcu since, no other updater anymore. Genl already unregistered */
2363 	for_each_resource_safe(resource, tmp, &drbd_resources) {
2364 		list_del(&resource->resources);
2365 		drbd_free_resource(resource);
2366 	}
2367 
2368 	drbd_debugfs_cleanup();
2369 
2370 	drbd_destroy_mempools();
2371 	unregister_blkdev(DRBD_MAJOR, "drbd");
2372 
2373 	idr_destroy(&drbd_devices);
2374 
2375 	pr_info("module cleanup done.\n");
2376 }
2377 
2378 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2379 {
2380 	spin_lock_init(&wq->q_lock);
2381 	INIT_LIST_HEAD(&wq->q);
2382 	init_waitqueue_head(&wq->q_wait);
2383 }
2384 
2385 struct completion_work {
2386 	struct drbd_work w;
2387 	struct completion done;
2388 };
2389 
2390 static int w_complete(struct drbd_work *w, int cancel)
2391 {
2392 	struct completion_work *completion_work =
2393 		container_of(w, struct completion_work, w);
2394 
2395 	complete(&completion_work->done);
2396 	return 0;
2397 }
2398 
2399 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2400 {
2401 	struct completion_work completion_work;
2402 
2403 	completion_work.w.cb = w_complete;
2404 	init_completion(&completion_work.done);
2405 	drbd_queue_work(work_queue, &completion_work.w);
2406 	wait_for_completion(&completion_work.done);
2407 }
2408 
2409 struct drbd_resource *drbd_find_resource(const char *name)
2410 {
2411 	struct drbd_resource *resource;
2412 
2413 	if (!name || !name[0])
2414 		return NULL;
2415 
2416 	rcu_read_lock();
2417 	for_each_resource_rcu(resource, &drbd_resources) {
2418 		if (!strcmp(resource->name, name)) {
2419 			kref_get(&resource->kref);
2420 			goto found;
2421 		}
2422 	}
2423 	resource = NULL;
2424 found:
2425 	rcu_read_unlock();
2426 	return resource;
2427 }
2428 
2429 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2430 				     void *peer_addr, int peer_addr_len)
2431 {
2432 	struct drbd_resource *resource;
2433 	struct drbd_connection *connection;
2434 
2435 	rcu_read_lock();
2436 	for_each_resource_rcu(resource, &drbd_resources) {
2437 		for_each_connection_rcu(connection, resource) {
2438 			if (connection->my_addr_len == my_addr_len &&
2439 			    connection->peer_addr_len == peer_addr_len &&
2440 			    !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2441 			    !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2442 				kref_get(&connection->kref);
2443 				goto found;
2444 			}
2445 		}
2446 	}
2447 	connection = NULL;
2448 found:
2449 	rcu_read_unlock();
2450 	return connection;
2451 }
2452 
2453 static int drbd_alloc_socket(struct drbd_socket *socket)
2454 {
2455 	socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2456 	if (!socket->rbuf)
2457 		return -ENOMEM;
2458 	socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2459 	if (!socket->sbuf)
2460 		return -ENOMEM;
2461 	return 0;
2462 }
2463 
2464 static void drbd_free_socket(struct drbd_socket *socket)
2465 {
2466 	free_page((unsigned long) socket->sbuf);
2467 	free_page((unsigned long) socket->rbuf);
2468 }
2469 
2470 void conn_free_crypto(struct drbd_connection *connection)
2471 {
2472 	drbd_free_sock(connection);
2473 
2474 	crypto_free_shash(connection->csums_tfm);
2475 	crypto_free_shash(connection->verify_tfm);
2476 	crypto_free_shash(connection->cram_hmac_tfm);
2477 	crypto_free_shash(connection->integrity_tfm);
2478 	crypto_free_shash(connection->peer_integrity_tfm);
2479 	kfree(connection->int_dig_in);
2480 	kfree(connection->int_dig_vv);
2481 
2482 	connection->csums_tfm = NULL;
2483 	connection->verify_tfm = NULL;
2484 	connection->cram_hmac_tfm = NULL;
2485 	connection->integrity_tfm = NULL;
2486 	connection->peer_integrity_tfm = NULL;
2487 	connection->int_dig_in = NULL;
2488 	connection->int_dig_vv = NULL;
2489 }
2490 
2491 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2492 {
2493 	struct drbd_connection *connection;
2494 	cpumask_var_t new_cpu_mask;
2495 	int err;
2496 
2497 	if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2498 		return -ENOMEM;
2499 
2500 	/* silently ignore cpu mask on UP kernel */
2501 	if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2502 		err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2503 				   cpumask_bits(new_cpu_mask), nr_cpu_ids);
2504 		if (err == -EOVERFLOW) {
2505 			/* So what. mask it out. */
2506 			cpumask_var_t tmp_cpu_mask;
2507 			if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2508 				cpumask_setall(tmp_cpu_mask);
2509 				cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2510 				drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2511 					res_opts->cpu_mask,
2512 					strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2513 					nr_cpu_ids);
2514 				free_cpumask_var(tmp_cpu_mask);
2515 				err = 0;
2516 			}
2517 		}
2518 		if (err) {
2519 			drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2520 			/* retcode = ERR_CPU_MASK_PARSE; */
2521 			goto fail;
2522 		}
2523 	}
2524 	resource->res_opts = *res_opts;
2525 	if (cpumask_empty(new_cpu_mask))
2526 		drbd_calc_cpu_mask(&new_cpu_mask);
2527 	if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2528 		cpumask_copy(resource->cpu_mask, new_cpu_mask);
2529 		for_each_connection_rcu(connection, resource) {
2530 			connection->receiver.reset_cpu_mask = 1;
2531 			connection->ack_receiver.reset_cpu_mask = 1;
2532 			connection->worker.reset_cpu_mask = 1;
2533 		}
2534 	}
2535 	err = 0;
2536 
2537 fail:
2538 	free_cpumask_var(new_cpu_mask);
2539 	return err;
2540 
2541 }
2542 
2543 struct drbd_resource *drbd_create_resource(const char *name)
2544 {
2545 	struct drbd_resource *resource;
2546 
2547 	resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2548 	if (!resource)
2549 		goto fail;
2550 	resource->name = kstrdup(name, GFP_KERNEL);
2551 	if (!resource->name)
2552 		goto fail_free_resource;
2553 	if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2554 		goto fail_free_name;
2555 	kref_init(&resource->kref);
2556 	idr_init(&resource->devices);
2557 	INIT_LIST_HEAD(&resource->connections);
2558 	resource->write_ordering = WO_BDEV_FLUSH;
2559 	list_add_tail_rcu(&resource->resources, &drbd_resources);
2560 	mutex_init(&resource->conf_update);
2561 	mutex_init(&resource->adm_mutex);
2562 	spin_lock_init(&resource->req_lock);
2563 	drbd_debugfs_resource_add(resource);
2564 	return resource;
2565 
2566 fail_free_name:
2567 	kfree(resource->name);
2568 fail_free_resource:
2569 	kfree(resource);
2570 fail:
2571 	return NULL;
2572 }
2573 
2574 /* caller must be under adm_mutex */
2575 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2576 {
2577 	struct drbd_resource *resource;
2578 	struct drbd_connection *connection;
2579 
2580 	connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2581 	if (!connection)
2582 		return NULL;
2583 
2584 	if (drbd_alloc_socket(&connection->data))
2585 		goto fail;
2586 	if (drbd_alloc_socket(&connection->meta))
2587 		goto fail;
2588 
2589 	connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2590 	if (!connection->current_epoch)
2591 		goto fail;
2592 
2593 	INIT_LIST_HEAD(&connection->transfer_log);
2594 
2595 	INIT_LIST_HEAD(&connection->current_epoch->list);
2596 	connection->epochs = 1;
2597 	spin_lock_init(&connection->epoch_lock);
2598 
2599 	connection->send.seen_any_write_yet = false;
2600 	connection->send.current_epoch_nr = 0;
2601 	connection->send.current_epoch_writes = 0;
2602 
2603 	resource = drbd_create_resource(name);
2604 	if (!resource)
2605 		goto fail;
2606 
2607 	connection->cstate = C_STANDALONE;
2608 	mutex_init(&connection->cstate_mutex);
2609 	init_waitqueue_head(&connection->ping_wait);
2610 	idr_init(&connection->peer_devices);
2611 
2612 	drbd_init_workqueue(&connection->sender_work);
2613 	mutex_init(&connection->data.mutex);
2614 	mutex_init(&connection->meta.mutex);
2615 
2616 	drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2617 	connection->receiver.connection = connection;
2618 	drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2619 	connection->worker.connection = connection;
2620 	drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2621 	connection->ack_receiver.connection = connection;
2622 
2623 	kref_init(&connection->kref);
2624 
2625 	connection->resource = resource;
2626 
2627 	if (set_resource_options(resource, res_opts))
2628 		goto fail_resource;
2629 
2630 	kref_get(&resource->kref);
2631 	list_add_tail_rcu(&connection->connections, &resource->connections);
2632 	drbd_debugfs_connection_add(connection);
2633 	return connection;
2634 
2635 fail_resource:
2636 	list_del(&resource->resources);
2637 	drbd_free_resource(resource);
2638 fail:
2639 	kfree(connection->current_epoch);
2640 	drbd_free_socket(&connection->meta);
2641 	drbd_free_socket(&connection->data);
2642 	kfree(connection);
2643 	return NULL;
2644 }
2645 
2646 void drbd_destroy_connection(struct kref *kref)
2647 {
2648 	struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2649 	struct drbd_resource *resource = connection->resource;
2650 
2651 	if (atomic_read(&connection->current_epoch->epoch_size) !=  0)
2652 		drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2653 	kfree(connection->current_epoch);
2654 
2655 	idr_destroy(&connection->peer_devices);
2656 
2657 	drbd_free_socket(&connection->meta);
2658 	drbd_free_socket(&connection->data);
2659 	kfree(connection->int_dig_in);
2660 	kfree(connection->int_dig_vv);
2661 	kfree(connection);
2662 	kref_put(&resource->kref, drbd_destroy_resource);
2663 }
2664 
2665 static int init_submitter(struct drbd_device *device)
2666 {
2667 	/* opencoded create_singlethread_workqueue(),
2668 	 * to be able to say "drbd%d", ..., minor */
2669 	device->submit.wq =
2670 		alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2671 	if (!device->submit.wq)
2672 		return -ENOMEM;
2673 
2674 	INIT_WORK(&device->submit.worker, do_submit);
2675 	INIT_LIST_HEAD(&device->submit.writes);
2676 	return 0;
2677 }
2678 
2679 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2680 {
2681 	struct drbd_resource *resource = adm_ctx->resource;
2682 	struct drbd_connection *connection, *n;
2683 	struct drbd_device *device;
2684 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2685 	struct gendisk *disk;
2686 	int id;
2687 	int vnr = adm_ctx->volume;
2688 	enum drbd_ret_code err = ERR_NOMEM;
2689 
2690 	device = minor_to_device(minor);
2691 	if (device)
2692 		return ERR_MINOR_OR_VOLUME_EXISTS;
2693 
2694 	/* GFP_KERNEL, we are outside of all write-out paths */
2695 	device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2696 	if (!device)
2697 		return ERR_NOMEM;
2698 	kref_init(&device->kref);
2699 
2700 	kref_get(&resource->kref);
2701 	device->resource = resource;
2702 	device->minor = minor;
2703 	device->vnr = vnr;
2704 
2705 	drbd_init_set_defaults(device);
2706 
2707 	disk = blk_alloc_disk(NUMA_NO_NODE);
2708 	if (!disk)
2709 		goto out_no_disk;
2710 
2711 	device->vdisk = disk;
2712 	device->rq_queue = disk->queue;
2713 
2714 	set_disk_ro(disk, true);
2715 
2716 	disk->major = DRBD_MAJOR;
2717 	disk->first_minor = minor;
2718 	disk->minors = 1;
2719 	disk->fops = &drbd_ops;
2720 	disk->flags |= GENHD_FL_NO_PART;
2721 	sprintf(disk->disk_name, "drbd%d", minor);
2722 	disk->private_data = device;
2723 
2724 	blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, disk->queue);
2725 	blk_queue_write_cache(disk->queue, true, true);
2726 	/* Setting the max_hw_sectors to an odd value of 8kibyte here
2727 	   This triggers a max_bio_size message upon first attach or connect */
2728 	blk_queue_max_hw_sectors(disk->queue, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2729 
2730 	device->md_io.page = alloc_page(GFP_KERNEL);
2731 	if (!device->md_io.page)
2732 		goto out_no_io_page;
2733 
2734 	if (drbd_bm_init(device))
2735 		goto out_no_bitmap;
2736 	device->read_requests = RB_ROOT;
2737 	device->write_requests = RB_ROOT;
2738 
2739 	id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2740 	if (id < 0) {
2741 		if (id == -ENOSPC)
2742 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2743 		goto out_no_minor_idr;
2744 	}
2745 	kref_get(&device->kref);
2746 
2747 	id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2748 	if (id < 0) {
2749 		if (id == -ENOSPC)
2750 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2751 		goto out_idr_remove_minor;
2752 	}
2753 	kref_get(&device->kref);
2754 
2755 	INIT_LIST_HEAD(&device->peer_devices);
2756 	INIT_LIST_HEAD(&device->pending_bitmap_io);
2757 	for_each_connection(connection, resource) {
2758 		peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2759 		if (!peer_device)
2760 			goto out_idr_remove_from_resource;
2761 		peer_device->connection = connection;
2762 		peer_device->device = device;
2763 
2764 		list_add(&peer_device->peer_devices, &device->peer_devices);
2765 		kref_get(&device->kref);
2766 
2767 		id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2768 		if (id < 0) {
2769 			if (id == -ENOSPC)
2770 				err = ERR_INVALID_REQUEST;
2771 			goto out_idr_remove_from_resource;
2772 		}
2773 		kref_get(&connection->kref);
2774 		INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2775 	}
2776 
2777 	if (init_submitter(device)) {
2778 		err = ERR_NOMEM;
2779 		goto out_idr_remove_from_resource;
2780 	}
2781 
2782 	err = add_disk(disk);
2783 	if (err)
2784 		goto out_destroy_workqueue;
2785 
2786 	/* inherit the connection state */
2787 	device->state.conn = first_connection(resource)->cstate;
2788 	if (device->state.conn == C_WF_REPORT_PARAMS) {
2789 		for_each_peer_device(peer_device, device)
2790 			drbd_connected(peer_device);
2791 	}
2792 	/* move to create_peer_device() */
2793 	for_each_peer_device(peer_device, device)
2794 		drbd_debugfs_peer_device_add(peer_device);
2795 	drbd_debugfs_device_add(device);
2796 	return NO_ERROR;
2797 
2798 out_destroy_workqueue:
2799 	destroy_workqueue(device->submit.wq);
2800 out_idr_remove_from_resource:
2801 	for_each_connection_safe(connection, n, resource) {
2802 		peer_device = idr_remove(&connection->peer_devices, vnr);
2803 		if (peer_device)
2804 			kref_put(&connection->kref, drbd_destroy_connection);
2805 	}
2806 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2807 		list_del(&peer_device->peer_devices);
2808 		kfree(peer_device);
2809 	}
2810 	idr_remove(&resource->devices, vnr);
2811 out_idr_remove_minor:
2812 	idr_remove(&drbd_devices, minor);
2813 	synchronize_rcu();
2814 out_no_minor_idr:
2815 	drbd_bm_cleanup(device);
2816 out_no_bitmap:
2817 	__free_page(device->md_io.page);
2818 out_no_io_page:
2819 	put_disk(disk);
2820 out_no_disk:
2821 	kref_put(&resource->kref, drbd_destroy_resource);
2822 	kfree(device);
2823 	return err;
2824 }
2825 
2826 void drbd_delete_device(struct drbd_device *device)
2827 {
2828 	struct drbd_resource *resource = device->resource;
2829 	struct drbd_connection *connection;
2830 	struct drbd_peer_device *peer_device;
2831 
2832 	/* move to free_peer_device() */
2833 	for_each_peer_device(peer_device, device)
2834 		drbd_debugfs_peer_device_cleanup(peer_device);
2835 	drbd_debugfs_device_cleanup(device);
2836 	for_each_connection(connection, resource) {
2837 		idr_remove(&connection->peer_devices, device->vnr);
2838 		kref_put(&device->kref, drbd_destroy_device);
2839 	}
2840 	idr_remove(&resource->devices, device->vnr);
2841 	kref_put(&device->kref, drbd_destroy_device);
2842 	idr_remove(&drbd_devices, device_to_minor(device));
2843 	kref_put(&device->kref, drbd_destroy_device);
2844 	del_gendisk(device->vdisk);
2845 	synchronize_rcu();
2846 	kref_put(&device->kref, drbd_destroy_device);
2847 }
2848 
2849 static int __init drbd_init(void)
2850 {
2851 	int err;
2852 
2853 	if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2854 		pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2855 #ifdef MODULE
2856 		return -EINVAL;
2857 #else
2858 		drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2859 #endif
2860 	}
2861 
2862 	err = register_blkdev(DRBD_MAJOR, "drbd");
2863 	if (err) {
2864 		pr_err("unable to register block device major %d\n",
2865 		       DRBD_MAJOR);
2866 		return err;
2867 	}
2868 
2869 	/*
2870 	 * allocate all necessary structs
2871 	 */
2872 	init_waitqueue_head(&drbd_pp_wait);
2873 
2874 	drbd_proc = NULL; /* play safe for drbd_cleanup */
2875 	idr_init(&drbd_devices);
2876 
2877 	mutex_init(&resources_mutex);
2878 	INIT_LIST_HEAD(&drbd_resources);
2879 
2880 	err = drbd_genl_register();
2881 	if (err) {
2882 		pr_err("unable to register generic netlink family\n");
2883 		goto fail;
2884 	}
2885 
2886 	err = drbd_create_mempools();
2887 	if (err)
2888 		goto fail;
2889 
2890 	err = -ENOMEM;
2891 	drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show);
2892 	if (!drbd_proc)	{
2893 		pr_err("unable to register proc file\n");
2894 		goto fail;
2895 	}
2896 
2897 	retry.wq = create_singlethread_workqueue("drbd-reissue");
2898 	if (!retry.wq) {
2899 		pr_err("unable to create retry workqueue\n");
2900 		goto fail;
2901 	}
2902 	INIT_WORK(&retry.worker, do_retry);
2903 	spin_lock_init(&retry.lock);
2904 	INIT_LIST_HEAD(&retry.writes);
2905 
2906 	drbd_debugfs_init();
2907 
2908 	pr_info("initialized. "
2909 	       "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2910 	       GENL_MAGIC_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2911 	pr_info("%s\n", drbd_buildtag());
2912 	pr_info("registered as block device major %d\n", DRBD_MAJOR);
2913 	return 0; /* Success! */
2914 
2915 fail:
2916 	drbd_cleanup();
2917 	if (err == -ENOMEM)
2918 		pr_err("ran out of memory\n");
2919 	else
2920 		pr_err("initialization failure\n");
2921 	return err;
2922 }
2923 
2924 static void drbd_free_one_sock(struct drbd_socket *ds)
2925 {
2926 	struct socket *s;
2927 	mutex_lock(&ds->mutex);
2928 	s = ds->socket;
2929 	ds->socket = NULL;
2930 	mutex_unlock(&ds->mutex);
2931 	if (s) {
2932 		/* so debugfs does not need to mutex_lock() */
2933 		synchronize_rcu();
2934 		kernel_sock_shutdown(s, SHUT_RDWR);
2935 		sock_release(s);
2936 	}
2937 }
2938 
2939 void drbd_free_sock(struct drbd_connection *connection)
2940 {
2941 	if (connection->data.socket)
2942 		drbd_free_one_sock(&connection->data);
2943 	if (connection->meta.socket)
2944 		drbd_free_one_sock(&connection->meta);
2945 }
2946 
2947 /* meta data management */
2948 
2949 void conn_md_sync(struct drbd_connection *connection)
2950 {
2951 	struct drbd_peer_device *peer_device;
2952 	int vnr;
2953 
2954 	rcu_read_lock();
2955 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2956 		struct drbd_device *device = peer_device->device;
2957 
2958 		kref_get(&device->kref);
2959 		rcu_read_unlock();
2960 		drbd_md_sync(device);
2961 		kref_put(&device->kref, drbd_destroy_device);
2962 		rcu_read_lock();
2963 	}
2964 	rcu_read_unlock();
2965 }
2966 
2967 /* aligned 4kByte */
2968 struct meta_data_on_disk {
2969 	u64 la_size_sect;      /* last agreed size. */
2970 	u64 uuid[UI_SIZE];   /* UUIDs. */
2971 	u64 device_uuid;
2972 	u64 reserved_u64_1;
2973 	u32 flags;             /* MDF */
2974 	u32 magic;
2975 	u32 md_size_sect;
2976 	u32 al_offset;         /* offset to this block */
2977 	u32 al_nr_extents;     /* important for restoring the AL (userspace) */
2978 	      /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2979 	u32 bm_offset;         /* offset to the bitmap, from here */
2980 	u32 bm_bytes_per_bit;  /* BM_BLOCK_SIZE */
2981 	u32 la_peer_max_bio_size;   /* last peer max_bio_size */
2982 
2983 	/* see al_tr_number_to_on_disk_sector() */
2984 	u32 al_stripes;
2985 	u32 al_stripe_size_4k;
2986 
2987 	u8 reserved_u8[4096 - (7*8 + 10*4)];
2988 } __packed;
2989 
2990 
2991 
2992 void drbd_md_write(struct drbd_device *device, void *b)
2993 {
2994 	struct meta_data_on_disk *buffer = b;
2995 	sector_t sector;
2996 	int i;
2997 
2998 	memset(buffer, 0, sizeof(*buffer));
2999 
3000 	buffer->la_size_sect = cpu_to_be64(get_capacity(device->vdisk));
3001 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3002 		buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3003 	buffer->flags = cpu_to_be32(device->ldev->md.flags);
3004 	buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3005 
3006 	buffer->md_size_sect  = cpu_to_be32(device->ldev->md.md_size_sect);
3007 	buffer->al_offset     = cpu_to_be32(device->ldev->md.al_offset);
3008 	buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3009 	buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3010 	buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3011 
3012 	buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3013 	buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3014 
3015 	buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3016 	buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3017 
3018 	D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3019 	sector = device->ldev->md.md_offset;
3020 
3021 	if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3022 		/* this was a try anyways ... */
3023 		drbd_err(device, "meta data update failed!\n");
3024 		drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3025 	}
3026 }
3027 
3028 /**
3029  * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3030  * @device:	DRBD device.
3031  */
3032 void drbd_md_sync(struct drbd_device *device)
3033 {
3034 	struct meta_data_on_disk *buffer;
3035 
3036 	/* Don't accidentally change the DRBD meta data layout. */
3037 	BUILD_BUG_ON(UI_SIZE != 4);
3038 	BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3039 
3040 	del_timer(&device->md_sync_timer);
3041 	/* timer may be rearmed by drbd_md_mark_dirty() now. */
3042 	if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3043 		return;
3044 
3045 	/* We use here D_FAILED and not D_ATTACHING because we try to write
3046 	 * metadata even if we detach due to a disk failure! */
3047 	if (!get_ldev_if_state(device, D_FAILED))
3048 		return;
3049 
3050 	buffer = drbd_md_get_buffer(device, __func__);
3051 	if (!buffer)
3052 		goto out;
3053 
3054 	drbd_md_write(device, buffer);
3055 
3056 	/* Update device->ldev->md.la_size_sect,
3057 	 * since we updated it on metadata. */
3058 	device->ldev->md.la_size_sect = get_capacity(device->vdisk);
3059 
3060 	drbd_md_put_buffer(device);
3061 out:
3062 	put_ldev(device);
3063 }
3064 
3065 static int check_activity_log_stripe_size(struct drbd_device *device,
3066 		struct meta_data_on_disk *on_disk,
3067 		struct drbd_md *in_core)
3068 {
3069 	u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3070 	u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3071 	u64 al_size_4k;
3072 
3073 	/* both not set: default to old fixed size activity log */
3074 	if (al_stripes == 0 && al_stripe_size_4k == 0) {
3075 		al_stripes = 1;
3076 		al_stripe_size_4k = MD_32kB_SECT/8;
3077 	}
3078 
3079 	/* some paranoia plausibility checks */
3080 
3081 	/* we need both values to be set */
3082 	if (al_stripes == 0 || al_stripe_size_4k == 0)
3083 		goto err;
3084 
3085 	al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3086 
3087 	/* Upper limit of activity log area, to avoid potential overflow
3088 	 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3089 	 * than 72 * 4k blocks total only increases the amount of history,
3090 	 * limiting this arbitrarily to 16 GB is not a real limitation ;-)  */
3091 	if (al_size_4k > (16 * 1024 * 1024/4))
3092 		goto err;
3093 
3094 	/* Lower limit: we need at least 8 transaction slots (32kB)
3095 	 * to not break existing setups */
3096 	if (al_size_4k < MD_32kB_SECT/8)
3097 		goto err;
3098 
3099 	in_core->al_stripe_size_4k = al_stripe_size_4k;
3100 	in_core->al_stripes = al_stripes;
3101 	in_core->al_size_4k = al_size_4k;
3102 
3103 	return 0;
3104 err:
3105 	drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3106 			al_stripes, al_stripe_size_4k);
3107 	return -EINVAL;
3108 }
3109 
3110 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3111 {
3112 	sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3113 	struct drbd_md *in_core = &bdev->md;
3114 	s32 on_disk_al_sect;
3115 	s32 on_disk_bm_sect;
3116 
3117 	/* The on-disk size of the activity log, calculated from offsets, and
3118 	 * the size of the activity log calculated from the stripe settings,
3119 	 * should match.
3120 	 * Though we could relax this a bit: it is ok, if the striped activity log
3121 	 * fits in the available on-disk activity log size.
3122 	 * Right now, that would break how resize is implemented.
3123 	 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3124 	 * of possible unused padding space in the on disk layout. */
3125 	if (in_core->al_offset < 0) {
3126 		if (in_core->bm_offset > in_core->al_offset)
3127 			goto err;
3128 		on_disk_al_sect = -in_core->al_offset;
3129 		on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3130 	} else {
3131 		if (in_core->al_offset != MD_4kB_SECT)
3132 			goto err;
3133 		if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3134 			goto err;
3135 
3136 		on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3137 		on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3138 	}
3139 
3140 	/* old fixed size meta data is exactly that: fixed. */
3141 	if (in_core->meta_dev_idx >= 0) {
3142 		if (in_core->md_size_sect != MD_128MB_SECT
3143 		||  in_core->al_offset != MD_4kB_SECT
3144 		||  in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3145 		||  in_core->al_stripes != 1
3146 		||  in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3147 			goto err;
3148 	}
3149 
3150 	if (capacity < in_core->md_size_sect)
3151 		goto err;
3152 	if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3153 		goto err;
3154 
3155 	/* should be aligned, and at least 32k */
3156 	if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3157 		goto err;
3158 
3159 	/* should fit (for now: exactly) into the available on-disk space;
3160 	 * overflow prevention is in check_activity_log_stripe_size() above. */
3161 	if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3162 		goto err;
3163 
3164 	/* again, should be aligned */
3165 	if (in_core->bm_offset & 7)
3166 		goto err;
3167 
3168 	/* FIXME check for device grow with flex external meta data? */
3169 
3170 	/* can the available bitmap space cover the last agreed device size? */
3171 	if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3172 		goto err;
3173 
3174 	return 0;
3175 
3176 err:
3177 	drbd_err(device, "meta data offsets don't make sense: idx=%d "
3178 			"al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3179 			"md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3180 			in_core->meta_dev_idx,
3181 			in_core->al_stripes, in_core->al_stripe_size_4k,
3182 			in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3183 			(unsigned long long)in_core->la_size_sect,
3184 			(unsigned long long)capacity);
3185 
3186 	return -EINVAL;
3187 }
3188 
3189 
3190 /**
3191  * drbd_md_read() - Reads in the meta data super block
3192  * @device:	DRBD device.
3193  * @bdev:	Device from which the meta data should be read in.
3194  *
3195  * Return NO_ERROR on success, and an enum drbd_ret_code in case
3196  * something goes wrong.
3197  *
3198  * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3199  * even before @bdev is assigned to @device->ldev.
3200  */
3201 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3202 {
3203 	struct meta_data_on_disk *buffer;
3204 	u32 magic, flags;
3205 	int i, rv = NO_ERROR;
3206 
3207 	if (device->state.disk != D_DISKLESS)
3208 		return ERR_DISK_CONFIGURED;
3209 
3210 	buffer = drbd_md_get_buffer(device, __func__);
3211 	if (!buffer)
3212 		return ERR_NOMEM;
3213 
3214 	/* First, figure out where our meta data superblock is located,
3215 	 * and read it. */
3216 	bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3217 	bdev->md.md_offset = drbd_md_ss(bdev);
3218 	/* Even for (flexible or indexed) external meta data,
3219 	 * initially restrict us to the 4k superblock for now.
3220 	 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3221 	bdev->md.md_size_sect = 8;
3222 
3223 	if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3224 				 REQ_OP_READ)) {
3225 		/* NOTE: can't do normal error processing here as this is
3226 		   called BEFORE disk is attached */
3227 		drbd_err(device, "Error while reading metadata.\n");
3228 		rv = ERR_IO_MD_DISK;
3229 		goto err;
3230 	}
3231 
3232 	magic = be32_to_cpu(buffer->magic);
3233 	flags = be32_to_cpu(buffer->flags);
3234 	if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3235 	    (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3236 			/* btw: that's Activity Log clean, not "all" clean. */
3237 		drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3238 		rv = ERR_MD_UNCLEAN;
3239 		goto err;
3240 	}
3241 
3242 	rv = ERR_MD_INVALID;
3243 	if (magic != DRBD_MD_MAGIC_08) {
3244 		if (magic == DRBD_MD_MAGIC_07)
3245 			drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3246 		else
3247 			drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3248 		goto err;
3249 	}
3250 
3251 	if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3252 		drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3253 		    be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3254 		goto err;
3255 	}
3256 
3257 
3258 	/* convert to in_core endian */
3259 	bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3260 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3261 		bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3262 	bdev->md.flags = be32_to_cpu(buffer->flags);
3263 	bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3264 
3265 	bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3266 	bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3267 	bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3268 
3269 	if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3270 		goto err;
3271 	if (check_offsets_and_sizes(device, bdev))
3272 		goto err;
3273 
3274 	if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3275 		drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3276 		    be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3277 		goto err;
3278 	}
3279 	if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3280 		drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3281 		    be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3282 		goto err;
3283 	}
3284 
3285 	rv = NO_ERROR;
3286 
3287 	spin_lock_irq(&device->resource->req_lock);
3288 	if (device->state.conn < C_CONNECTED) {
3289 		unsigned int peer;
3290 		peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3291 		peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3292 		device->peer_max_bio_size = peer;
3293 	}
3294 	spin_unlock_irq(&device->resource->req_lock);
3295 
3296  err:
3297 	drbd_md_put_buffer(device);
3298 
3299 	return rv;
3300 }
3301 
3302 /**
3303  * drbd_md_mark_dirty() - Mark meta data super block as dirty
3304  * @device:	DRBD device.
3305  *
3306  * Call this function if you change anything that should be written to
3307  * the meta-data super block. This function sets MD_DIRTY, and starts a
3308  * timer that ensures that within five seconds you have to call drbd_md_sync().
3309  */
3310 void drbd_md_mark_dirty(struct drbd_device *device)
3311 {
3312 	if (!test_and_set_bit(MD_DIRTY, &device->flags))
3313 		mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3314 }
3315 
3316 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3317 {
3318 	int i;
3319 
3320 	for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3321 		device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3322 }
3323 
3324 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3325 {
3326 	if (idx == UI_CURRENT) {
3327 		if (device->state.role == R_PRIMARY)
3328 			val |= 1;
3329 		else
3330 			val &= ~((u64)1);
3331 
3332 		drbd_set_ed_uuid(device, val);
3333 	}
3334 
3335 	device->ldev->md.uuid[idx] = val;
3336 	drbd_md_mark_dirty(device);
3337 }
3338 
3339 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3340 {
3341 	unsigned long flags;
3342 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3343 	__drbd_uuid_set(device, idx, val);
3344 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3345 }
3346 
3347 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3348 {
3349 	unsigned long flags;
3350 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3351 	if (device->ldev->md.uuid[idx]) {
3352 		drbd_uuid_move_history(device);
3353 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3354 	}
3355 	__drbd_uuid_set(device, idx, val);
3356 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3357 }
3358 
3359 /**
3360  * drbd_uuid_new_current() - Creates a new current UUID
3361  * @device:	DRBD device.
3362  *
3363  * Creates a new current UUID, and rotates the old current UUID into
3364  * the bitmap slot. Causes an incremental resync upon next connect.
3365  */
3366 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3367 {
3368 	u64 val;
3369 	unsigned long long bm_uuid;
3370 
3371 	get_random_bytes(&val, sizeof(u64));
3372 
3373 	spin_lock_irq(&device->ldev->md.uuid_lock);
3374 	bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3375 
3376 	if (bm_uuid)
3377 		drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3378 
3379 	device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3380 	__drbd_uuid_set(device, UI_CURRENT, val);
3381 	spin_unlock_irq(&device->ldev->md.uuid_lock);
3382 
3383 	drbd_print_uuids(device, "new current UUID");
3384 	/* get it to stable storage _now_ */
3385 	drbd_md_sync(device);
3386 }
3387 
3388 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3389 {
3390 	unsigned long flags;
3391 	if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3392 		return;
3393 
3394 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3395 	if (val == 0) {
3396 		drbd_uuid_move_history(device);
3397 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3398 		device->ldev->md.uuid[UI_BITMAP] = 0;
3399 	} else {
3400 		unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3401 		if (bm_uuid)
3402 			drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3403 
3404 		device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3405 	}
3406 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3407 
3408 	drbd_md_mark_dirty(device);
3409 }
3410 
3411 /**
3412  * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3413  * @device:	DRBD device.
3414  *
3415  * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3416  */
3417 int drbd_bmio_set_n_write(struct drbd_device *device,
3418 			  struct drbd_peer_device *peer_device) __must_hold(local)
3419 
3420 {
3421 	int rv = -EIO;
3422 
3423 	drbd_md_set_flag(device, MDF_FULL_SYNC);
3424 	drbd_md_sync(device);
3425 	drbd_bm_set_all(device);
3426 
3427 	rv = drbd_bm_write(device, peer_device);
3428 
3429 	if (!rv) {
3430 		drbd_md_clear_flag(device, MDF_FULL_SYNC);
3431 		drbd_md_sync(device);
3432 	}
3433 
3434 	return rv;
3435 }
3436 
3437 /**
3438  * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3439  * @device:	DRBD device.
3440  *
3441  * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3442  */
3443 int drbd_bmio_clear_n_write(struct drbd_device *device,
3444 			  struct drbd_peer_device *peer_device) __must_hold(local)
3445 
3446 {
3447 	drbd_resume_al(device);
3448 	drbd_bm_clear_all(device);
3449 	return drbd_bm_write(device, peer_device);
3450 }
3451 
3452 static int w_bitmap_io(struct drbd_work *w, int unused)
3453 {
3454 	struct drbd_device *device =
3455 		container_of(w, struct drbd_device, bm_io_work.w);
3456 	struct bm_io_work *work = &device->bm_io_work;
3457 	int rv = -EIO;
3458 
3459 	if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3460 		int cnt = atomic_read(&device->ap_bio_cnt);
3461 		if (cnt)
3462 			drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3463 					cnt, work->why);
3464 	}
3465 
3466 	if (get_ldev(device)) {
3467 		drbd_bm_lock(device, work->why, work->flags);
3468 		rv = work->io_fn(device, work->peer_device);
3469 		drbd_bm_unlock(device);
3470 		put_ldev(device);
3471 	}
3472 
3473 	clear_bit_unlock(BITMAP_IO, &device->flags);
3474 	wake_up(&device->misc_wait);
3475 
3476 	if (work->done)
3477 		work->done(device, rv);
3478 
3479 	clear_bit(BITMAP_IO_QUEUED, &device->flags);
3480 	work->why = NULL;
3481 	work->flags = 0;
3482 
3483 	return 0;
3484 }
3485 
3486 /**
3487  * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3488  * @device:	DRBD device.
3489  * @io_fn:	IO callback to be called when bitmap IO is possible
3490  * @done:	callback to be called after the bitmap IO was performed
3491  * @why:	Descriptive text of the reason for doing the IO
3492  * @flags:	Bitmap flags
3493  *
3494  * While IO on the bitmap happens we freeze application IO thus we ensure
3495  * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3496  * called from worker context. It MUST NOT be used while a previous such
3497  * work is still pending!
3498  *
3499  * Its worker function encloses the call of io_fn() by get_ldev() and
3500  * put_ldev().
3501  */
3502 void drbd_queue_bitmap_io(struct drbd_device *device,
3503 			  int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
3504 			  void (*done)(struct drbd_device *, int),
3505 			  char *why, enum bm_flag flags,
3506 			  struct drbd_peer_device *peer_device)
3507 {
3508 	D_ASSERT(device, current == peer_device->connection->worker.task);
3509 
3510 	D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3511 	D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3512 	D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3513 	if (device->bm_io_work.why)
3514 		drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3515 			why, device->bm_io_work.why);
3516 
3517 	device->bm_io_work.peer_device = peer_device;
3518 	device->bm_io_work.io_fn = io_fn;
3519 	device->bm_io_work.done = done;
3520 	device->bm_io_work.why = why;
3521 	device->bm_io_work.flags = flags;
3522 
3523 	spin_lock_irq(&device->resource->req_lock);
3524 	set_bit(BITMAP_IO, &device->flags);
3525 	/* don't wait for pending application IO if the caller indicates that
3526 	 * application IO does not conflict anyways. */
3527 	if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3528 		if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3529 			drbd_queue_work(&peer_device->connection->sender_work,
3530 					&device->bm_io_work.w);
3531 	}
3532 	spin_unlock_irq(&device->resource->req_lock);
3533 }
3534 
3535 /**
3536  * drbd_bitmap_io() -  Does an IO operation on the whole bitmap
3537  * @device:	DRBD device.
3538  * @io_fn:	IO callback to be called when bitmap IO is possible
3539  * @why:	Descriptive text of the reason for doing the IO
3540  * @flags:	Bitmap flags
3541  *
3542  * freezes application IO while that the actual IO operations runs. This
3543  * functions MAY NOT be called from worker context.
3544  */
3545 int drbd_bitmap_io(struct drbd_device *device,
3546 		int (*io_fn)(struct drbd_device *, struct drbd_peer_device *),
3547 		char *why, enum bm_flag flags,
3548 		struct drbd_peer_device *peer_device)
3549 {
3550 	/* Only suspend io, if some operation is supposed to be locked out */
3551 	const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3552 	int rv;
3553 
3554 	D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3555 
3556 	if (do_suspend_io)
3557 		drbd_suspend_io(device);
3558 
3559 	drbd_bm_lock(device, why, flags);
3560 	rv = io_fn(device, peer_device);
3561 	drbd_bm_unlock(device);
3562 
3563 	if (do_suspend_io)
3564 		drbd_resume_io(device);
3565 
3566 	return rv;
3567 }
3568 
3569 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3570 {
3571 	if ((device->ldev->md.flags & flag) != flag) {
3572 		drbd_md_mark_dirty(device);
3573 		device->ldev->md.flags |= flag;
3574 	}
3575 }
3576 
3577 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3578 {
3579 	if ((device->ldev->md.flags & flag) != 0) {
3580 		drbd_md_mark_dirty(device);
3581 		device->ldev->md.flags &= ~flag;
3582 	}
3583 }
3584 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3585 {
3586 	return (bdev->md.flags & flag) != 0;
3587 }
3588 
3589 static void md_sync_timer_fn(struct timer_list *t)
3590 {
3591 	struct drbd_device *device = from_timer(device, t, md_sync_timer);
3592 	drbd_device_post_work(device, MD_SYNC);
3593 }
3594 
3595 const char *cmdname(enum drbd_packet cmd)
3596 {
3597 	/* THINK may need to become several global tables
3598 	 * when we want to support more than
3599 	 * one PRO_VERSION */
3600 	static const char *cmdnames[] = {
3601 
3602 		[P_DATA]	        = "Data",
3603 		[P_DATA_REPLY]	        = "DataReply",
3604 		[P_RS_DATA_REPLY]	= "RSDataReply",
3605 		[P_BARRIER]	        = "Barrier",
3606 		[P_BITMAP]	        = "ReportBitMap",
3607 		[P_BECOME_SYNC_TARGET]  = "BecomeSyncTarget",
3608 		[P_BECOME_SYNC_SOURCE]  = "BecomeSyncSource",
3609 		[P_UNPLUG_REMOTE]	= "UnplugRemote",
3610 		[P_DATA_REQUEST]	= "DataRequest",
3611 		[P_RS_DATA_REQUEST]     = "RSDataRequest",
3612 		[P_SYNC_PARAM]	        = "SyncParam",
3613 		[P_PROTOCOL]            = "ReportProtocol",
3614 		[P_UUIDS]	        = "ReportUUIDs",
3615 		[P_SIZES]	        = "ReportSizes",
3616 		[P_STATE]	        = "ReportState",
3617 		[P_SYNC_UUID]           = "ReportSyncUUID",
3618 		[P_AUTH_CHALLENGE]      = "AuthChallenge",
3619 		[P_AUTH_RESPONSE]	= "AuthResponse",
3620 		[P_STATE_CHG_REQ]       = "StateChgRequest",
3621 		[P_PING]		= "Ping",
3622 		[P_PING_ACK]	        = "PingAck",
3623 		[P_RECV_ACK]	        = "RecvAck",
3624 		[P_WRITE_ACK]	        = "WriteAck",
3625 		[P_RS_WRITE_ACK]	= "RSWriteAck",
3626 		[P_SUPERSEDED]          = "Superseded",
3627 		[P_NEG_ACK]	        = "NegAck",
3628 		[P_NEG_DREPLY]	        = "NegDReply",
3629 		[P_NEG_RS_DREPLY]	= "NegRSDReply",
3630 		[P_BARRIER_ACK]	        = "BarrierAck",
3631 		[P_STATE_CHG_REPLY]     = "StateChgReply",
3632 		[P_OV_REQUEST]          = "OVRequest",
3633 		[P_OV_REPLY]            = "OVReply",
3634 		[P_OV_RESULT]           = "OVResult",
3635 		[P_CSUM_RS_REQUEST]     = "CsumRSRequest",
3636 		[P_RS_IS_IN_SYNC]	= "CsumRSIsInSync",
3637 		[P_SYNC_PARAM89]	= "SyncParam89",
3638 		[P_COMPRESSED_BITMAP]   = "CBitmap",
3639 		[P_DELAY_PROBE]         = "DelayProbe",
3640 		[P_OUT_OF_SYNC]		= "OutOfSync",
3641 		[P_RS_CANCEL]		= "RSCancel",
3642 		[P_CONN_ST_CHG_REQ]	= "conn_st_chg_req",
3643 		[P_CONN_ST_CHG_REPLY]	= "conn_st_chg_reply",
3644 		[P_PROTOCOL_UPDATE]	= "protocol_update",
3645 		[P_TRIM]	        = "Trim",
3646 		[P_RS_THIN_REQ]         = "rs_thin_req",
3647 		[P_RS_DEALLOCATED]      = "rs_deallocated",
3648 		[P_WSAME]	        = "WriteSame",
3649 		[P_ZEROES]		= "Zeroes",
3650 
3651 		/* enum drbd_packet, but not commands - obsoleted flags:
3652 		 *	P_MAY_IGNORE
3653 		 *	P_MAX_OPT_CMD
3654 		 */
3655 	};
3656 
3657 	/* too big for the array: 0xfffX */
3658 	if (cmd == P_INITIAL_META)
3659 		return "InitialMeta";
3660 	if (cmd == P_INITIAL_DATA)
3661 		return "InitialData";
3662 	if (cmd == P_CONNECTION_FEATURES)
3663 		return "ConnectionFeatures";
3664 	if (cmd >= ARRAY_SIZE(cmdnames))
3665 		return "Unknown";
3666 	return cmdnames[cmd];
3667 }
3668 
3669 /**
3670  * drbd_wait_misc  -  wait for a request to make progress
3671  * @device:	device associated with the request
3672  * @i:		the struct drbd_interval embedded in struct drbd_request or
3673  *		struct drbd_peer_request
3674  */
3675 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3676 {
3677 	struct net_conf *nc;
3678 	DEFINE_WAIT(wait);
3679 	long timeout;
3680 
3681 	rcu_read_lock();
3682 	nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3683 	if (!nc) {
3684 		rcu_read_unlock();
3685 		return -ETIMEDOUT;
3686 	}
3687 	timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3688 	rcu_read_unlock();
3689 
3690 	/* Indicate to wake up device->misc_wait on progress.  */
3691 	i->waiting = true;
3692 	prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3693 	spin_unlock_irq(&device->resource->req_lock);
3694 	timeout = schedule_timeout(timeout);
3695 	finish_wait(&device->misc_wait, &wait);
3696 	spin_lock_irq(&device->resource->req_lock);
3697 	if (!timeout || device->state.conn < C_CONNECTED)
3698 		return -ETIMEDOUT;
3699 	if (signal_pending(current))
3700 		return -ERESTARTSYS;
3701 	return 0;
3702 }
3703 
3704 void lock_all_resources(void)
3705 {
3706 	struct drbd_resource *resource;
3707 	int __maybe_unused i = 0;
3708 
3709 	mutex_lock(&resources_mutex);
3710 	local_irq_disable();
3711 	for_each_resource(resource, &drbd_resources)
3712 		spin_lock_nested(&resource->req_lock, i++);
3713 }
3714 
3715 void unlock_all_resources(void)
3716 {
3717 	struct drbd_resource *resource;
3718 
3719 	for_each_resource(resource, &drbd_resources)
3720 		spin_unlock(&resource->req_lock);
3721 	local_irq_enable();
3722 	mutex_unlock(&resources_mutex);
3723 }
3724 
3725 #ifdef CONFIG_DRBD_FAULT_INJECTION
3726 /* Fault insertion support including random number generator shamelessly
3727  * stolen from kernel/rcutorture.c */
3728 struct fault_random_state {
3729 	unsigned long state;
3730 	unsigned long count;
3731 };
3732 
3733 #define FAULT_RANDOM_MULT 39916801  /* prime */
3734 #define FAULT_RANDOM_ADD	479001701 /* prime */
3735 #define FAULT_RANDOM_REFRESH 10000
3736 
3737 /*
3738  * Crude but fast random-number generator.  Uses a linear congruential
3739  * generator, with occasional help from get_random_bytes().
3740  */
3741 static unsigned long
3742 _drbd_fault_random(struct fault_random_state *rsp)
3743 {
3744 	long refresh;
3745 
3746 	if (!rsp->count--) {
3747 		get_random_bytes(&refresh, sizeof(refresh));
3748 		rsp->state += refresh;
3749 		rsp->count = FAULT_RANDOM_REFRESH;
3750 	}
3751 	rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3752 	return swahw32(rsp->state);
3753 }
3754 
3755 static char *
3756 _drbd_fault_str(unsigned int type) {
3757 	static char *_faults[] = {
3758 		[DRBD_FAULT_MD_WR] = "Meta-data write",
3759 		[DRBD_FAULT_MD_RD] = "Meta-data read",
3760 		[DRBD_FAULT_RS_WR] = "Resync write",
3761 		[DRBD_FAULT_RS_RD] = "Resync read",
3762 		[DRBD_FAULT_DT_WR] = "Data write",
3763 		[DRBD_FAULT_DT_RD] = "Data read",
3764 		[DRBD_FAULT_DT_RA] = "Data read ahead",
3765 		[DRBD_FAULT_BM_ALLOC] = "BM allocation",
3766 		[DRBD_FAULT_AL_EE] = "EE allocation",
3767 		[DRBD_FAULT_RECEIVE] = "receive data corruption",
3768 	};
3769 
3770 	return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3771 }
3772 
3773 unsigned int
3774 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3775 {
3776 	static struct fault_random_state rrs = {0, 0};
3777 
3778 	unsigned int ret = (
3779 		(drbd_fault_devs == 0 ||
3780 			((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3781 		(((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3782 
3783 	if (ret) {
3784 		drbd_fault_count++;
3785 
3786 		if (drbd_ratelimit())
3787 			drbd_warn(device, "***Simulating %s failure\n",
3788 				_drbd_fault_str(type));
3789 	}
3790 
3791 	return ret;
3792 }
3793 #endif
3794 
3795 module_init(drbd_init)
3796 module_exit(drbd_cleanup)
3797 
3798 EXPORT_SYMBOL(drbd_conn_str);
3799 EXPORT_SYMBOL(drbd_role_str);
3800 EXPORT_SYMBOL(drbd_disk_str);
3801 EXPORT_SYMBOL(drbd_set_st_err_str);
3802