xref: /openbmc/linux/drivers/block/drbd/drbd_main.c (revision b240b419db5d624ce7a5a397d6f62a1a686009ec)
1 /*
2    drbd.c
3 
4    This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
5 
6    Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7    Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8    Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
9 
10    Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
11    from Logicworks, Inc. for making SDP replication support possible.
12 
13    drbd is free software; you can redistribute it and/or modify
14    it under the terms of the GNU General Public License as published by
15    the Free Software Foundation; either version 2, or (at your option)
16    any later version.
17 
18    drbd is distributed in the hope that it will be useful,
19    but WITHOUT ANY WARRANTY; without even the implied warranty of
20    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21    GNU General Public License for more details.
22 
23    You should have received a copy of the GNU General Public License
24    along with drbd; see the file COPYING.  If not, write to
25    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
26 
27  */
28 
29 #define pr_fmt(fmt)	KBUILD_MODNAME ": " fmt
30 
31 #include <linux/module.h>
32 #include <linux/jiffies.h>
33 #include <linux/drbd.h>
34 #include <linux/uaccess.h>
35 #include <asm/types.h>
36 #include <net/sock.h>
37 #include <linux/ctype.h>
38 #include <linux/mutex.h>
39 #include <linux/fs.h>
40 #include <linux/file.h>
41 #include <linux/proc_fs.h>
42 #include <linux/init.h>
43 #include <linux/mm.h>
44 #include <linux/memcontrol.h>
45 #include <linux/mm_inline.h>
46 #include <linux/slab.h>
47 #include <linux/random.h>
48 #include <linux/reboot.h>
49 #include <linux/notifier.h>
50 #include <linux/kthread.h>
51 #include <linux/workqueue.h>
52 #define __KERNEL_SYSCALLS__
53 #include <linux/unistd.h>
54 #include <linux/vmalloc.h>
55 #include <linux/sched/signal.h>
56 
57 #include <linux/drbd_limits.h>
58 #include "drbd_int.h"
59 #include "drbd_protocol.h"
60 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
61 #include "drbd_vli.h"
62 #include "drbd_debugfs.h"
63 
64 static DEFINE_MUTEX(drbd_main_mutex);
65 static int drbd_open(struct block_device *bdev, fmode_t mode);
66 static void drbd_release(struct gendisk *gd, fmode_t mode);
67 static void md_sync_timer_fn(struct timer_list *t);
68 static int w_bitmap_io(struct drbd_work *w, int unused);
69 
70 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
71 	      "Lars Ellenberg <lars@linbit.com>");
72 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
73 MODULE_VERSION(REL_VERSION);
74 MODULE_LICENSE("GPL");
75 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
76 		 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
77 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
78 
79 #include <linux/moduleparam.h>
80 /* thanks to these macros, if compiled into the kernel (not-module),
81  * these become boot parameters (e.g., drbd.minor_count) */
82 
83 #ifdef CONFIG_DRBD_FAULT_INJECTION
84 int drbd_enable_faults;
85 int drbd_fault_rate;
86 static int drbd_fault_count;
87 static int drbd_fault_devs;
88 /* bitmap of enabled faults */
89 module_param_named(enable_faults, drbd_enable_faults, int, 0664);
90 /* fault rate % value - applies to all enabled faults */
91 module_param_named(fault_rate, drbd_fault_rate, int, 0664);
92 /* count of faults inserted */
93 module_param_named(fault_count, drbd_fault_count, int, 0664);
94 /* bitmap of devices to insert faults on */
95 module_param_named(fault_devs, drbd_fault_devs, int, 0644);
96 #endif
97 
98 /* module parameters we can keep static */
99 static bool drbd_allow_oos; /* allow_open_on_secondary */
100 static bool drbd_disable_sendpage;
101 MODULE_PARM_DESC(allow_oos, "DONT USE!");
102 module_param_named(allow_oos, drbd_allow_oos, bool, 0);
103 module_param_named(disable_sendpage, drbd_disable_sendpage, bool, 0644);
104 
105 /* module parameters we share */
106 int drbd_proc_details; /* Detail level in proc drbd*/
107 module_param_named(proc_details, drbd_proc_details, int, 0644);
108 /* module parameters shared with defaults */
109 unsigned int drbd_minor_count = DRBD_MINOR_COUNT_DEF;
110 /* Module parameter for setting the user mode helper program
111  * to run. Default is /sbin/drbdadm */
112 char drbd_usermode_helper[80] = "/sbin/drbdadm";
113 module_param_named(minor_count, drbd_minor_count, uint, 0444);
114 module_param_string(usermode_helper, drbd_usermode_helper, sizeof(drbd_usermode_helper), 0644);
115 
116 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
117  * as member "struct gendisk *vdisk;"
118  */
119 struct idr drbd_devices;
120 struct list_head drbd_resources;
121 struct mutex resources_mutex;
122 
123 struct kmem_cache *drbd_request_cache;
124 struct kmem_cache *drbd_ee_cache;	/* peer requests */
125 struct kmem_cache *drbd_bm_ext_cache;	/* bitmap extents */
126 struct kmem_cache *drbd_al_ext_cache;	/* activity log extents */
127 mempool_t *drbd_request_mempool;
128 mempool_t *drbd_ee_mempool;
129 mempool_t *drbd_md_io_page_pool;
130 struct bio_set *drbd_md_io_bio_set;
131 struct bio_set *drbd_io_bio_set;
132 
133 /* I do not use a standard mempool, because:
134    1) I want to hand out the pre-allocated objects first.
135    2) I want to be able to interrupt sleeping allocation with a signal.
136    Note: This is a single linked list, the next pointer is the private
137 	 member of struct page.
138  */
139 struct page *drbd_pp_pool;
140 spinlock_t   drbd_pp_lock;
141 int          drbd_pp_vacant;
142 wait_queue_head_t drbd_pp_wait;
143 
144 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
145 
146 static const struct block_device_operations drbd_ops = {
147 	.owner =   THIS_MODULE,
148 	.open =    drbd_open,
149 	.release = drbd_release,
150 };
151 
152 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
153 {
154 	struct bio *bio;
155 
156 	if (!drbd_md_io_bio_set)
157 		return bio_alloc(gfp_mask, 1);
158 
159 	bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
160 	if (!bio)
161 		return NULL;
162 	return bio;
163 }
164 
165 #ifdef __CHECKER__
166 /* When checking with sparse, and this is an inline function, sparse will
167    give tons of false positives. When this is a real functions sparse works.
168  */
169 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
170 {
171 	int io_allowed;
172 
173 	atomic_inc(&device->local_cnt);
174 	io_allowed = (device->state.disk >= mins);
175 	if (!io_allowed) {
176 		if (atomic_dec_and_test(&device->local_cnt))
177 			wake_up(&device->misc_wait);
178 	}
179 	return io_allowed;
180 }
181 
182 #endif
183 
184 /**
185  * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
186  * @connection:	DRBD connection.
187  * @barrier_nr:	Expected identifier of the DRBD write barrier packet.
188  * @set_size:	Expected number of requests before that barrier.
189  *
190  * In case the passed barrier_nr or set_size does not match the oldest
191  * epoch of not yet barrier-acked requests, this function will cause a
192  * termination of the connection.
193  */
194 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
195 		unsigned int set_size)
196 {
197 	struct drbd_request *r;
198 	struct drbd_request *req = NULL;
199 	int expect_epoch = 0;
200 	int expect_size = 0;
201 
202 	spin_lock_irq(&connection->resource->req_lock);
203 
204 	/* find oldest not yet barrier-acked write request,
205 	 * count writes in its epoch. */
206 	list_for_each_entry(r, &connection->transfer_log, tl_requests) {
207 		const unsigned s = r->rq_state;
208 		if (!req) {
209 			if (!(s & RQ_WRITE))
210 				continue;
211 			if (!(s & RQ_NET_MASK))
212 				continue;
213 			if (s & RQ_NET_DONE)
214 				continue;
215 			req = r;
216 			expect_epoch = req->epoch;
217 			expect_size ++;
218 		} else {
219 			if (r->epoch != expect_epoch)
220 				break;
221 			if (!(s & RQ_WRITE))
222 				continue;
223 			/* if (s & RQ_DONE): not expected */
224 			/* if (!(s & RQ_NET_MASK)): not expected */
225 			expect_size++;
226 		}
227 	}
228 
229 	/* first some paranoia code */
230 	if (req == NULL) {
231 		drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
232 			 barrier_nr);
233 		goto bail;
234 	}
235 	if (expect_epoch != barrier_nr) {
236 		drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
237 			 barrier_nr, expect_epoch);
238 		goto bail;
239 	}
240 
241 	if (expect_size != set_size) {
242 		drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
243 			 barrier_nr, set_size, expect_size);
244 		goto bail;
245 	}
246 
247 	/* Clean up list of requests processed during current epoch. */
248 	/* this extra list walk restart is paranoia,
249 	 * to catch requests being barrier-acked "unexpectedly".
250 	 * It usually should find the same req again, or some READ preceding it. */
251 	list_for_each_entry(req, &connection->transfer_log, tl_requests)
252 		if (req->epoch == expect_epoch)
253 			break;
254 	list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
255 		if (req->epoch != expect_epoch)
256 			break;
257 		_req_mod(req, BARRIER_ACKED);
258 	}
259 	spin_unlock_irq(&connection->resource->req_lock);
260 
261 	return;
262 
263 bail:
264 	spin_unlock_irq(&connection->resource->req_lock);
265 	conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
266 }
267 
268 
269 /**
270  * _tl_restart() - Walks the transfer log, and applies an action to all requests
271  * @connection:	DRBD connection to operate on.
272  * @what:       The action/event to perform with all request objects
273  *
274  * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
275  * RESTART_FROZEN_DISK_IO.
276  */
277 /* must hold resource->req_lock */
278 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
279 {
280 	struct drbd_request *req, *r;
281 
282 	list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
283 		_req_mod(req, what);
284 }
285 
286 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
287 {
288 	spin_lock_irq(&connection->resource->req_lock);
289 	_tl_restart(connection, what);
290 	spin_unlock_irq(&connection->resource->req_lock);
291 }
292 
293 /**
294  * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
295  * @device:	DRBD device.
296  *
297  * This is called after the connection to the peer was lost. The storage covered
298  * by the requests on the transfer gets marked as our of sync. Called from the
299  * receiver thread and the worker thread.
300  */
301 void tl_clear(struct drbd_connection *connection)
302 {
303 	tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
304 }
305 
306 /**
307  * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
308  * @device:	DRBD device.
309  */
310 void tl_abort_disk_io(struct drbd_device *device)
311 {
312 	struct drbd_connection *connection = first_peer_device(device)->connection;
313 	struct drbd_request *req, *r;
314 
315 	spin_lock_irq(&connection->resource->req_lock);
316 	list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
317 		if (!(req->rq_state & RQ_LOCAL_PENDING))
318 			continue;
319 		if (req->device != device)
320 			continue;
321 		_req_mod(req, ABORT_DISK_IO);
322 	}
323 	spin_unlock_irq(&connection->resource->req_lock);
324 }
325 
326 static int drbd_thread_setup(void *arg)
327 {
328 	struct drbd_thread *thi = (struct drbd_thread *) arg;
329 	struct drbd_resource *resource = thi->resource;
330 	unsigned long flags;
331 	int retval;
332 
333 	snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
334 		 thi->name[0],
335 		 resource->name);
336 
337 restart:
338 	retval = thi->function(thi);
339 
340 	spin_lock_irqsave(&thi->t_lock, flags);
341 
342 	/* if the receiver has been "EXITING", the last thing it did
343 	 * was set the conn state to "StandAlone",
344 	 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
345 	 * and receiver thread will be "started".
346 	 * drbd_thread_start needs to set "RESTARTING" in that case.
347 	 * t_state check and assignment needs to be within the same spinlock,
348 	 * so either thread_start sees EXITING, and can remap to RESTARTING,
349 	 * or thread_start see NONE, and can proceed as normal.
350 	 */
351 
352 	if (thi->t_state == RESTARTING) {
353 		drbd_info(resource, "Restarting %s thread\n", thi->name);
354 		thi->t_state = RUNNING;
355 		spin_unlock_irqrestore(&thi->t_lock, flags);
356 		goto restart;
357 	}
358 
359 	thi->task = NULL;
360 	thi->t_state = NONE;
361 	smp_mb();
362 	complete_all(&thi->stop);
363 	spin_unlock_irqrestore(&thi->t_lock, flags);
364 
365 	drbd_info(resource, "Terminating %s\n", current->comm);
366 
367 	/* Release mod reference taken when thread was started */
368 
369 	if (thi->connection)
370 		kref_put(&thi->connection->kref, drbd_destroy_connection);
371 	kref_put(&resource->kref, drbd_destroy_resource);
372 	module_put(THIS_MODULE);
373 	return retval;
374 }
375 
376 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
377 			     int (*func) (struct drbd_thread *), const char *name)
378 {
379 	spin_lock_init(&thi->t_lock);
380 	thi->task    = NULL;
381 	thi->t_state = NONE;
382 	thi->function = func;
383 	thi->resource = resource;
384 	thi->connection = NULL;
385 	thi->name = name;
386 }
387 
388 int drbd_thread_start(struct drbd_thread *thi)
389 {
390 	struct drbd_resource *resource = thi->resource;
391 	struct task_struct *nt;
392 	unsigned long flags;
393 
394 	/* is used from state engine doing drbd_thread_stop_nowait,
395 	 * while holding the req lock irqsave */
396 	spin_lock_irqsave(&thi->t_lock, flags);
397 
398 	switch (thi->t_state) {
399 	case NONE:
400 		drbd_info(resource, "Starting %s thread (from %s [%d])\n",
401 			 thi->name, current->comm, current->pid);
402 
403 		/* Get ref on module for thread - this is released when thread exits */
404 		if (!try_module_get(THIS_MODULE)) {
405 			drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
406 			spin_unlock_irqrestore(&thi->t_lock, flags);
407 			return false;
408 		}
409 
410 		kref_get(&resource->kref);
411 		if (thi->connection)
412 			kref_get(&thi->connection->kref);
413 
414 		init_completion(&thi->stop);
415 		thi->reset_cpu_mask = 1;
416 		thi->t_state = RUNNING;
417 		spin_unlock_irqrestore(&thi->t_lock, flags);
418 		flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
419 
420 		nt = kthread_create(drbd_thread_setup, (void *) thi,
421 				    "drbd_%c_%s", thi->name[0], thi->resource->name);
422 
423 		if (IS_ERR(nt)) {
424 			drbd_err(resource, "Couldn't start thread\n");
425 
426 			if (thi->connection)
427 				kref_put(&thi->connection->kref, drbd_destroy_connection);
428 			kref_put(&resource->kref, drbd_destroy_resource);
429 			module_put(THIS_MODULE);
430 			return false;
431 		}
432 		spin_lock_irqsave(&thi->t_lock, flags);
433 		thi->task = nt;
434 		thi->t_state = RUNNING;
435 		spin_unlock_irqrestore(&thi->t_lock, flags);
436 		wake_up_process(nt);
437 		break;
438 	case EXITING:
439 		thi->t_state = RESTARTING;
440 		drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
441 				thi->name, current->comm, current->pid);
442 		/* fall through */
443 	case RUNNING:
444 	case RESTARTING:
445 	default:
446 		spin_unlock_irqrestore(&thi->t_lock, flags);
447 		break;
448 	}
449 
450 	return true;
451 }
452 
453 
454 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
455 {
456 	unsigned long flags;
457 
458 	enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
459 
460 	/* may be called from state engine, holding the req lock irqsave */
461 	spin_lock_irqsave(&thi->t_lock, flags);
462 
463 	if (thi->t_state == NONE) {
464 		spin_unlock_irqrestore(&thi->t_lock, flags);
465 		if (restart)
466 			drbd_thread_start(thi);
467 		return;
468 	}
469 
470 	if (thi->t_state != ns) {
471 		if (thi->task == NULL) {
472 			spin_unlock_irqrestore(&thi->t_lock, flags);
473 			return;
474 		}
475 
476 		thi->t_state = ns;
477 		smp_mb();
478 		init_completion(&thi->stop);
479 		if (thi->task != current)
480 			force_sig(DRBD_SIGKILL, thi->task);
481 	}
482 
483 	spin_unlock_irqrestore(&thi->t_lock, flags);
484 
485 	if (wait)
486 		wait_for_completion(&thi->stop);
487 }
488 
489 int conn_lowest_minor(struct drbd_connection *connection)
490 {
491 	struct drbd_peer_device *peer_device;
492 	int vnr = 0, minor = -1;
493 
494 	rcu_read_lock();
495 	peer_device = idr_get_next(&connection->peer_devices, &vnr);
496 	if (peer_device)
497 		minor = device_to_minor(peer_device->device);
498 	rcu_read_unlock();
499 
500 	return minor;
501 }
502 
503 #ifdef CONFIG_SMP
504 /**
505  * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
506  *
507  * Forces all threads of a resource onto the same CPU. This is beneficial for
508  * DRBD's performance. May be overwritten by user's configuration.
509  */
510 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
511 {
512 	unsigned int *resources_per_cpu, min_index = ~0;
513 
514 	resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL);
515 	if (resources_per_cpu) {
516 		struct drbd_resource *resource;
517 		unsigned int cpu, min = ~0;
518 
519 		rcu_read_lock();
520 		for_each_resource_rcu(resource, &drbd_resources) {
521 			for_each_cpu(cpu, resource->cpu_mask)
522 				resources_per_cpu[cpu]++;
523 		}
524 		rcu_read_unlock();
525 		for_each_online_cpu(cpu) {
526 			if (resources_per_cpu[cpu] < min) {
527 				min = resources_per_cpu[cpu];
528 				min_index = cpu;
529 			}
530 		}
531 		kfree(resources_per_cpu);
532 	}
533 	if (min_index == ~0) {
534 		cpumask_setall(*cpu_mask);
535 		return;
536 	}
537 	cpumask_set_cpu(min_index, *cpu_mask);
538 }
539 
540 /**
541  * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
542  * @device:	DRBD device.
543  * @thi:	drbd_thread object
544  *
545  * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
546  * prematurely.
547  */
548 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
549 {
550 	struct drbd_resource *resource = thi->resource;
551 	struct task_struct *p = current;
552 
553 	if (!thi->reset_cpu_mask)
554 		return;
555 	thi->reset_cpu_mask = 0;
556 	set_cpus_allowed_ptr(p, resource->cpu_mask);
557 }
558 #else
559 #define drbd_calc_cpu_mask(A) ({})
560 #endif
561 
562 /**
563  * drbd_header_size  -  size of a packet header
564  *
565  * The header size is a multiple of 8, so any payload following the header is
566  * word aligned on 64-bit architectures.  (The bitmap send and receive code
567  * relies on this.)
568  */
569 unsigned int drbd_header_size(struct drbd_connection *connection)
570 {
571 	if (connection->agreed_pro_version >= 100) {
572 		BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
573 		return sizeof(struct p_header100);
574 	} else {
575 		BUILD_BUG_ON(sizeof(struct p_header80) !=
576 			     sizeof(struct p_header95));
577 		BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
578 		return sizeof(struct p_header80);
579 	}
580 }
581 
582 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
583 {
584 	h->magic   = cpu_to_be32(DRBD_MAGIC);
585 	h->command = cpu_to_be16(cmd);
586 	h->length  = cpu_to_be16(size);
587 	return sizeof(struct p_header80);
588 }
589 
590 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
591 {
592 	h->magic   = cpu_to_be16(DRBD_MAGIC_BIG);
593 	h->command = cpu_to_be16(cmd);
594 	h->length = cpu_to_be32(size);
595 	return sizeof(struct p_header95);
596 }
597 
598 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
599 				      int size, int vnr)
600 {
601 	h->magic = cpu_to_be32(DRBD_MAGIC_100);
602 	h->volume = cpu_to_be16(vnr);
603 	h->command = cpu_to_be16(cmd);
604 	h->length = cpu_to_be32(size);
605 	h->pad = 0;
606 	return sizeof(struct p_header100);
607 }
608 
609 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
610 				   void *buffer, enum drbd_packet cmd, int size)
611 {
612 	if (connection->agreed_pro_version >= 100)
613 		return prepare_header100(buffer, cmd, size, vnr);
614 	else if (connection->agreed_pro_version >= 95 &&
615 		 size > DRBD_MAX_SIZE_H80_PACKET)
616 		return prepare_header95(buffer, cmd, size);
617 	else
618 		return prepare_header80(buffer, cmd, size);
619 }
620 
621 static void *__conn_prepare_command(struct drbd_connection *connection,
622 				    struct drbd_socket *sock)
623 {
624 	if (!sock->socket)
625 		return NULL;
626 	return sock->sbuf + drbd_header_size(connection);
627 }
628 
629 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
630 {
631 	void *p;
632 
633 	mutex_lock(&sock->mutex);
634 	p = __conn_prepare_command(connection, sock);
635 	if (!p)
636 		mutex_unlock(&sock->mutex);
637 
638 	return p;
639 }
640 
641 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
642 {
643 	return conn_prepare_command(peer_device->connection, sock);
644 }
645 
646 static int __send_command(struct drbd_connection *connection, int vnr,
647 			  struct drbd_socket *sock, enum drbd_packet cmd,
648 			  unsigned int header_size, void *data,
649 			  unsigned int size)
650 {
651 	int msg_flags;
652 	int err;
653 
654 	/*
655 	 * Called with @data == NULL and the size of the data blocks in @size
656 	 * for commands that send data blocks.  For those commands, omit the
657 	 * MSG_MORE flag: this will increase the likelihood that data blocks
658 	 * which are page aligned on the sender will end up page aligned on the
659 	 * receiver.
660 	 */
661 	msg_flags = data ? MSG_MORE : 0;
662 
663 	header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
664 				      header_size + size);
665 	err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
666 			    msg_flags);
667 	if (data && !err)
668 		err = drbd_send_all(connection, sock->socket, data, size, 0);
669 	/* DRBD protocol "pings" are latency critical.
670 	 * This is supposed to trigger tcp_push_pending_frames() */
671 	if (!err && (cmd == P_PING || cmd == P_PING_ACK))
672 		drbd_tcp_nodelay(sock->socket);
673 
674 	return err;
675 }
676 
677 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
678 			       enum drbd_packet cmd, unsigned int header_size,
679 			       void *data, unsigned int size)
680 {
681 	return __send_command(connection, 0, sock, cmd, header_size, data, size);
682 }
683 
684 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
685 		      enum drbd_packet cmd, unsigned int header_size,
686 		      void *data, unsigned int size)
687 {
688 	int err;
689 
690 	err = __conn_send_command(connection, sock, cmd, header_size, data, size);
691 	mutex_unlock(&sock->mutex);
692 	return err;
693 }
694 
695 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
696 		      enum drbd_packet cmd, unsigned int header_size,
697 		      void *data, unsigned int size)
698 {
699 	int err;
700 
701 	err = __send_command(peer_device->connection, peer_device->device->vnr,
702 			     sock, cmd, header_size, data, size);
703 	mutex_unlock(&sock->mutex);
704 	return err;
705 }
706 
707 int drbd_send_ping(struct drbd_connection *connection)
708 {
709 	struct drbd_socket *sock;
710 
711 	sock = &connection->meta;
712 	if (!conn_prepare_command(connection, sock))
713 		return -EIO;
714 	return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
715 }
716 
717 int drbd_send_ping_ack(struct drbd_connection *connection)
718 {
719 	struct drbd_socket *sock;
720 
721 	sock = &connection->meta;
722 	if (!conn_prepare_command(connection, sock))
723 		return -EIO;
724 	return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
725 }
726 
727 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
728 {
729 	struct drbd_socket *sock;
730 	struct p_rs_param_95 *p;
731 	int size;
732 	const int apv = peer_device->connection->agreed_pro_version;
733 	enum drbd_packet cmd;
734 	struct net_conf *nc;
735 	struct disk_conf *dc;
736 
737 	sock = &peer_device->connection->data;
738 	p = drbd_prepare_command(peer_device, sock);
739 	if (!p)
740 		return -EIO;
741 
742 	rcu_read_lock();
743 	nc = rcu_dereference(peer_device->connection->net_conf);
744 
745 	size = apv <= 87 ? sizeof(struct p_rs_param)
746 		: apv == 88 ? sizeof(struct p_rs_param)
747 			+ strlen(nc->verify_alg) + 1
748 		: apv <= 94 ? sizeof(struct p_rs_param_89)
749 		: /* apv >= 95 */ sizeof(struct p_rs_param_95);
750 
751 	cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
752 
753 	/* initialize verify_alg and csums_alg */
754 	memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
755 
756 	if (get_ldev(peer_device->device)) {
757 		dc = rcu_dereference(peer_device->device->ldev->disk_conf);
758 		p->resync_rate = cpu_to_be32(dc->resync_rate);
759 		p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
760 		p->c_delay_target = cpu_to_be32(dc->c_delay_target);
761 		p->c_fill_target = cpu_to_be32(dc->c_fill_target);
762 		p->c_max_rate = cpu_to_be32(dc->c_max_rate);
763 		put_ldev(peer_device->device);
764 	} else {
765 		p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
766 		p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
767 		p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
768 		p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
769 		p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
770 	}
771 
772 	if (apv >= 88)
773 		strcpy(p->verify_alg, nc->verify_alg);
774 	if (apv >= 89)
775 		strcpy(p->csums_alg, nc->csums_alg);
776 	rcu_read_unlock();
777 
778 	return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
779 }
780 
781 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
782 {
783 	struct drbd_socket *sock;
784 	struct p_protocol *p;
785 	struct net_conf *nc;
786 	int size, cf;
787 
788 	sock = &connection->data;
789 	p = __conn_prepare_command(connection, sock);
790 	if (!p)
791 		return -EIO;
792 
793 	rcu_read_lock();
794 	nc = rcu_dereference(connection->net_conf);
795 
796 	if (nc->tentative && connection->agreed_pro_version < 92) {
797 		rcu_read_unlock();
798 		mutex_unlock(&sock->mutex);
799 		drbd_err(connection, "--dry-run is not supported by peer");
800 		return -EOPNOTSUPP;
801 	}
802 
803 	size = sizeof(*p);
804 	if (connection->agreed_pro_version >= 87)
805 		size += strlen(nc->integrity_alg) + 1;
806 
807 	p->protocol      = cpu_to_be32(nc->wire_protocol);
808 	p->after_sb_0p   = cpu_to_be32(nc->after_sb_0p);
809 	p->after_sb_1p   = cpu_to_be32(nc->after_sb_1p);
810 	p->after_sb_2p   = cpu_to_be32(nc->after_sb_2p);
811 	p->two_primaries = cpu_to_be32(nc->two_primaries);
812 	cf = 0;
813 	if (nc->discard_my_data)
814 		cf |= CF_DISCARD_MY_DATA;
815 	if (nc->tentative)
816 		cf |= CF_DRY_RUN;
817 	p->conn_flags    = cpu_to_be32(cf);
818 
819 	if (connection->agreed_pro_version >= 87)
820 		strcpy(p->integrity_alg, nc->integrity_alg);
821 	rcu_read_unlock();
822 
823 	return __conn_send_command(connection, sock, cmd, size, NULL, 0);
824 }
825 
826 int drbd_send_protocol(struct drbd_connection *connection)
827 {
828 	int err;
829 
830 	mutex_lock(&connection->data.mutex);
831 	err = __drbd_send_protocol(connection, P_PROTOCOL);
832 	mutex_unlock(&connection->data.mutex);
833 
834 	return err;
835 }
836 
837 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
838 {
839 	struct drbd_device *device = peer_device->device;
840 	struct drbd_socket *sock;
841 	struct p_uuids *p;
842 	int i;
843 
844 	if (!get_ldev_if_state(device, D_NEGOTIATING))
845 		return 0;
846 
847 	sock = &peer_device->connection->data;
848 	p = drbd_prepare_command(peer_device, sock);
849 	if (!p) {
850 		put_ldev(device);
851 		return -EIO;
852 	}
853 	spin_lock_irq(&device->ldev->md.uuid_lock);
854 	for (i = UI_CURRENT; i < UI_SIZE; i++)
855 		p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
856 	spin_unlock_irq(&device->ldev->md.uuid_lock);
857 
858 	device->comm_bm_set = drbd_bm_total_weight(device);
859 	p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
860 	rcu_read_lock();
861 	uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
862 	rcu_read_unlock();
863 	uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
864 	uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
865 	p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
866 
867 	put_ldev(device);
868 	return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
869 }
870 
871 int drbd_send_uuids(struct drbd_peer_device *peer_device)
872 {
873 	return _drbd_send_uuids(peer_device, 0);
874 }
875 
876 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
877 {
878 	return _drbd_send_uuids(peer_device, 8);
879 }
880 
881 void drbd_print_uuids(struct drbd_device *device, const char *text)
882 {
883 	if (get_ldev_if_state(device, D_NEGOTIATING)) {
884 		u64 *uuid = device->ldev->md.uuid;
885 		drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
886 		     text,
887 		     (unsigned long long)uuid[UI_CURRENT],
888 		     (unsigned long long)uuid[UI_BITMAP],
889 		     (unsigned long long)uuid[UI_HISTORY_START],
890 		     (unsigned long long)uuid[UI_HISTORY_END]);
891 		put_ldev(device);
892 	} else {
893 		drbd_info(device, "%s effective data uuid: %016llX\n",
894 				text,
895 				(unsigned long long)device->ed_uuid);
896 	}
897 }
898 
899 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
900 {
901 	struct drbd_device *device = peer_device->device;
902 	struct drbd_socket *sock;
903 	struct p_rs_uuid *p;
904 	u64 uuid;
905 
906 	D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
907 
908 	uuid = device->ldev->md.uuid[UI_BITMAP];
909 	if (uuid && uuid != UUID_JUST_CREATED)
910 		uuid = uuid + UUID_NEW_BM_OFFSET;
911 	else
912 		get_random_bytes(&uuid, sizeof(u64));
913 	drbd_uuid_set(device, UI_BITMAP, uuid);
914 	drbd_print_uuids(device, "updated sync UUID");
915 	drbd_md_sync(device);
916 
917 	sock = &peer_device->connection->data;
918 	p = drbd_prepare_command(peer_device, sock);
919 	if (p) {
920 		p->uuid = cpu_to_be64(uuid);
921 		drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
922 	}
923 }
924 
925 /* communicated if (agreed_features & DRBD_FF_WSAME) */
926 static void
927 assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p,
928 					struct request_queue *q)
929 {
930 	if (q) {
931 		p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
932 		p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
933 		p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
934 		p->qlim->io_min = cpu_to_be32(queue_io_min(q));
935 		p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
936 		p->qlim->discard_enabled = blk_queue_discard(q);
937 		p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
938 	} else {
939 		q = device->rq_queue;
940 		p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
941 		p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
942 		p->qlim->alignment_offset = 0;
943 		p->qlim->io_min = cpu_to_be32(queue_io_min(q));
944 		p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
945 		p->qlim->discard_enabled = 0;
946 		p->qlim->write_same_capable = 0;
947 	}
948 }
949 
950 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
951 {
952 	struct drbd_device *device = peer_device->device;
953 	struct drbd_socket *sock;
954 	struct p_sizes *p;
955 	sector_t d_size, u_size;
956 	int q_order_type;
957 	unsigned int max_bio_size;
958 	unsigned int packet_size;
959 
960 	sock = &peer_device->connection->data;
961 	p = drbd_prepare_command(peer_device, sock);
962 	if (!p)
963 		return -EIO;
964 
965 	packet_size = sizeof(*p);
966 	if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
967 		packet_size += sizeof(p->qlim[0]);
968 
969 	memset(p, 0, packet_size);
970 	if (get_ldev_if_state(device, D_NEGOTIATING)) {
971 		struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
972 		d_size = drbd_get_max_capacity(device->ldev);
973 		rcu_read_lock();
974 		u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
975 		rcu_read_unlock();
976 		q_order_type = drbd_queue_order_type(device);
977 		max_bio_size = queue_max_hw_sectors(q) << 9;
978 		max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
979 		assign_p_sizes_qlim(device, p, q);
980 		put_ldev(device);
981 	} else {
982 		d_size = 0;
983 		u_size = 0;
984 		q_order_type = QUEUE_ORDERED_NONE;
985 		max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
986 		assign_p_sizes_qlim(device, p, NULL);
987 	}
988 
989 	if (peer_device->connection->agreed_pro_version <= 94)
990 		max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
991 	else if (peer_device->connection->agreed_pro_version < 100)
992 		max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
993 
994 	p->d_size = cpu_to_be64(d_size);
995 	p->u_size = cpu_to_be64(u_size);
996 	p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
997 	p->max_bio_size = cpu_to_be32(max_bio_size);
998 	p->queue_order_type = cpu_to_be16(q_order_type);
999 	p->dds_flags = cpu_to_be16(flags);
1000 
1001 	return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
1002 }
1003 
1004 /**
1005  * drbd_send_current_state() - Sends the drbd state to the peer
1006  * @peer_device:	DRBD peer device.
1007  */
1008 int drbd_send_current_state(struct drbd_peer_device *peer_device)
1009 {
1010 	struct drbd_socket *sock;
1011 	struct p_state *p;
1012 
1013 	sock = &peer_device->connection->data;
1014 	p = drbd_prepare_command(peer_device, sock);
1015 	if (!p)
1016 		return -EIO;
1017 	p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1018 	return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1019 }
1020 
1021 /**
1022  * drbd_send_state() - After a state change, sends the new state to the peer
1023  * @peer_device:      DRBD peer device.
1024  * @state:     the state to send, not necessarily the current state.
1025  *
1026  * Each state change queues an "after_state_ch" work, which will eventually
1027  * send the resulting new state to the peer. If more state changes happen
1028  * between queuing and processing of the after_state_ch work, we still
1029  * want to send each intermediary state in the order it occurred.
1030  */
1031 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1032 {
1033 	struct drbd_socket *sock;
1034 	struct p_state *p;
1035 
1036 	sock = &peer_device->connection->data;
1037 	p = drbd_prepare_command(peer_device, sock);
1038 	if (!p)
1039 		return -EIO;
1040 	p->state = cpu_to_be32(state.i); /* Within the send mutex */
1041 	return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1042 }
1043 
1044 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1045 {
1046 	struct drbd_socket *sock;
1047 	struct p_req_state *p;
1048 
1049 	sock = &peer_device->connection->data;
1050 	p = drbd_prepare_command(peer_device, 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 drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1056 }
1057 
1058 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1059 {
1060 	enum drbd_packet cmd;
1061 	struct drbd_socket *sock;
1062 	struct p_req_state *p;
1063 
1064 	cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1065 	sock = &connection->data;
1066 	p = conn_prepare_command(connection, sock);
1067 	if (!p)
1068 		return -EIO;
1069 	p->mask = cpu_to_be32(mask.i);
1070 	p->val = cpu_to_be32(val.i);
1071 	return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1072 }
1073 
1074 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1075 {
1076 	struct drbd_socket *sock;
1077 	struct p_req_state_reply *p;
1078 
1079 	sock = &peer_device->connection->meta;
1080 	p = drbd_prepare_command(peer_device, sock);
1081 	if (p) {
1082 		p->retcode = cpu_to_be32(retcode);
1083 		drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1084 	}
1085 }
1086 
1087 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1088 {
1089 	struct drbd_socket *sock;
1090 	struct p_req_state_reply *p;
1091 	enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1092 
1093 	sock = &connection->meta;
1094 	p = conn_prepare_command(connection, sock);
1095 	if (p) {
1096 		p->retcode = cpu_to_be32(retcode);
1097 		conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1098 	}
1099 }
1100 
1101 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1102 {
1103 	BUG_ON(code & ~0xf);
1104 	p->encoding = (p->encoding & ~0xf) | code;
1105 }
1106 
1107 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1108 {
1109 	p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1110 }
1111 
1112 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1113 {
1114 	BUG_ON(n & ~0x7);
1115 	p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1116 }
1117 
1118 static int fill_bitmap_rle_bits(struct drbd_device *device,
1119 			 struct p_compressed_bm *p,
1120 			 unsigned int size,
1121 			 struct bm_xfer_ctx *c)
1122 {
1123 	struct bitstream bs;
1124 	unsigned long plain_bits;
1125 	unsigned long tmp;
1126 	unsigned long rl;
1127 	unsigned len;
1128 	unsigned toggle;
1129 	int bits, use_rle;
1130 
1131 	/* may we use this feature? */
1132 	rcu_read_lock();
1133 	use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1134 	rcu_read_unlock();
1135 	if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1136 		return 0;
1137 
1138 	if (c->bit_offset >= c->bm_bits)
1139 		return 0; /* nothing to do. */
1140 
1141 	/* use at most thus many bytes */
1142 	bitstream_init(&bs, p->code, size, 0);
1143 	memset(p->code, 0, size);
1144 	/* plain bits covered in this code string */
1145 	plain_bits = 0;
1146 
1147 	/* p->encoding & 0x80 stores whether the first run length is set.
1148 	 * bit offset is implicit.
1149 	 * start with toggle == 2 to be able to tell the first iteration */
1150 	toggle = 2;
1151 
1152 	/* see how much plain bits we can stuff into one packet
1153 	 * using RLE and VLI. */
1154 	do {
1155 		tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1156 				    : _drbd_bm_find_next(device, c->bit_offset);
1157 		if (tmp == -1UL)
1158 			tmp = c->bm_bits;
1159 		rl = tmp - c->bit_offset;
1160 
1161 		if (toggle == 2) { /* first iteration */
1162 			if (rl == 0) {
1163 				/* the first checked bit was set,
1164 				 * store start value, */
1165 				dcbp_set_start(p, 1);
1166 				/* but skip encoding of zero run length */
1167 				toggle = !toggle;
1168 				continue;
1169 			}
1170 			dcbp_set_start(p, 0);
1171 		}
1172 
1173 		/* paranoia: catch zero runlength.
1174 		 * can only happen if bitmap is modified while we scan it. */
1175 		if (rl == 0) {
1176 			drbd_err(device, "unexpected zero runlength while encoding bitmap "
1177 			    "t:%u bo:%lu\n", toggle, c->bit_offset);
1178 			return -1;
1179 		}
1180 
1181 		bits = vli_encode_bits(&bs, rl);
1182 		if (bits == -ENOBUFS) /* buffer full */
1183 			break;
1184 		if (bits <= 0) {
1185 			drbd_err(device, "error while encoding bitmap: %d\n", bits);
1186 			return 0;
1187 		}
1188 
1189 		toggle = !toggle;
1190 		plain_bits += rl;
1191 		c->bit_offset = tmp;
1192 	} while (c->bit_offset < c->bm_bits);
1193 
1194 	len = bs.cur.b - p->code + !!bs.cur.bit;
1195 
1196 	if (plain_bits < (len << 3)) {
1197 		/* incompressible with this method.
1198 		 * we need to rewind both word and bit position. */
1199 		c->bit_offset -= plain_bits;
1200 		bm_xfer_ctx_bit_to_word_offset(c);
1201 		c->bit_offset = c->word_offset * BITS_PER_LONG;
1202 		return 0;
1203 	}
1204 
1205 	/* RLE + VLI was able to compress it just fine.
1206 	 * update c->word_offset. */
1207 	bm_xfer_ctx_bit_to_word_offset(c);
1208 
1209 	/* store pad_bits */
1210 	dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1211 
1212 	return len;
1213 }
1214 
1215 /**
1216  * send_bitmap_rle_or_plain
1217  *
1218  * Return 0 when done, 1 when another iteration is needed, and a negative error
1219  * code upon failure.
1220  */
1221 static int
1222 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1223 {
1224 	struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1225 	unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1226 	struct p_compressed_bm *p = sock->sbuf + header_size;
1227 	int len, err;
1228 
1229 	len = fill_bitmap_rle_bits(device, p,
1230 			DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1231 	if (len < 0)
1232 		return -EIO;
1233 
1234 	if (len) {
1235 		dcbp_set_code(p, RLE_VLI_Bits);
1236 		err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1237 				     P_COMPRESSED_BITMAP, sizeof(*p) + len,
1238 				     NULL, 0);
1239 		c->packets[0]++;
1240 		c->bytes[0] += header_size + sizeof(*p) + len;
1241 
1242 		if (c->bit_offset >= c->bm_bits)
1243 			len = 0; /* DONE */
1244 	} else {
1245 		/* was not compressible.
1246 		 * send a buffer full of plain text bits instead. */
1247 		unsigned int data_size;
1248 		unsigned long num_words;
1249 		unsigned long *p = sock->sbuf + header_size;
1250 
1251 		data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1252 		num_words = min_t(size_t, data_size / sizeof(*p),
1253 				  c->bm_words - c->word_offset);
1254 		len = num_words * sizeof(*p);
1255 		if (len)
1256 			drbd_bm_get_lel(device, c->word_offset, num_words, p);
1257 		err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1258 		c->word_offset += num_words;
1259 		c->bit_offset = c->word_offset * BITS_PER_LONG;
1260 
1261 		c->packets[1]++;
1262 		c->bytes[1] += header_size + len;
1263 
1264 		if (c->bit_offset > c->bm_bits)
1265 			c->bit_offset = c->bm_bits;
1266 	}
1267 	if (!err) {
1268 		if (len == 0) {
1269 			INFO_bm_xfer_stats(device, "send", c);
1270 			return 0;
1271 		} else
1272 			return 1;
1273 	}
1274 	return -EIO;
1275 }
1276 
1277 /* See the comment at receive_bitmap() */
1278 static int _drbd_send_bitmap(struct drbd_device *device)
1279 {
1280 	struct bm_xfer_ctx c;
1281 	int err;
1282 
1283 	if (!expect(device->bitmap))
1284 		return false;
1285 
1286 	if (get_ldev(device)) {
1287 		if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1288 			drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1289 			drbd_bm_set_all(device);
1290 			if (drbd_bm_write(device)) {
1291 				/* write_bm did fail! Leave full sync flag set in Meta P_DATA
1292 				 * but otherwise process as per normal - need to tell other
1293 				 * side that a full resync is required! */
1294 				drbd_err(device, "Failed to write bitmap to disk!\n");
1295 			} else {
1296 				drbd_md_clear_flag(device, MDF_FULL_SYNC);
1297 				drbd_md_sync(device);
1298 			}
1299 		}
1300 		put_ldev(device);
1301 	}
1302 
1303 	c = (struct bm_xfer_ctx) {
1304 		.bm_bits = drbd_bm_bits(device),
1305 		.bm_words = drbd_bm_words(device),
1306 	};
1307 
1308 	do {
1309 		err = send_bitmap_rle_or_plain(device, &c);
1310 	} while (err > 0);
1311 
1312 	return err == 0;
1313 }
1314 
1315 int drbd_send_bitmap(struct drbd_device *device)
1316 {
1317 	struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1318 	int err = -1;
1319 
1320 	mutex_lock(&sock->mutex);
1321 	if (sock->socket)
1322 		err = !_drbd_send_bitmap(device);
1323 	mutex_unlock(&sock->mutex);
1324 	return err;
1325 }
1326 
1327 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1328 {
1329 	struct drbd_socket *sock;
1330 	struct p_barrier_ack *p;
1331 
1332 	if (connection->cstate < C_WF_REPORT_PARAMS)
1333 		return;
1334 
1335 	sock = &connection->meta;
1336 	p = conn_prepare_command(connection, sock);
1337 	if (!p)
1338 		return;
1339 	p->barrier = barrier_nr;
1340 	p->set_size = cpu_to_be32(set_size);
1341 	conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1342 }
1343 
1344 /**
1345  * _drbd_send_ack() - Sends an ack packet
1346  * @device:	DRBD device.
1347  * @cmd:	Packet command code.
1348  * @sector:	sector, needs to be in big endian byte order
1349  * @blksize:	size in byte, needs to be in big endian byte order
1350  * @block_id:	Id, big endian byte order
1351  */
1352 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1353 			  u64 sector, u32 blksize, u64 block_id)
1354 {
1355 	struct drbd_socket *sock;
1356 	struct p_block_ack *p;
1357 
1358 	if (peer_device->device->state.conn < C_CONNECTED)
1359 		return -EIO;
1360 
1361 	sock = &peer_device->connection->meta;
1362 	p = drbd_prepare_command(peer_device, sock);
1363 	if (!p)
1364 		return -EIO;
1365 	p->sector = sector;
1366 	p->block_id = block_id;
1367 	p->blksize = blksize;
1368 	p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1369 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1370 }
1371 
1372 /* dp->sector and dp->block_id already/still in network byte order,
1373  * data_size is payload size according to dp->head,
1374  * and may need to be corrected for digest size. */
1375 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1376 		      struct p_data *dp, int data_size)
1377 {
1378 	if (peer_device->connection->peer_integrity_tfm)
1379 		data_size -= crypto_ahash_digestsize(peer_device->connection->peer_integrity_tfm);
1380 	_drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1381 		       dp->block_id);
1382 }
1383 
1384 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1385 		      struct p_block_req *rp)
1386 {
1387 	_drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1388 }
1389 
1390 /**
1391  * drbd_send_ack() - Sends an ack packet
1392  * @device:	DRBD device
1393  * @cmd:	packet command code
1394  * @peer_req:	peer request
1395  */
1396 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1397 		  struct drbd_peer_request *peer_req)
1398 {
1399 	return _drbd_send_ack(peer_device, cmd,
1400 			      cpu_to_be64(peer_req->i.sector),
1401 			      cpu_to_be32(peer_req->i.size),
1402 			      peer_req->block_id);
1403 }
1404 
1405 /* This function misuses the block_id field to signal if the blocks
1406  * are is sync or not. */
1407 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1408 		     sector_t sector, int blksize, u64 block_id)
1409 {
1410 	return _drbd_send_ack(peer_device, cmd,
1411 			      cpu_to_be64(sector),
1412 			      cpu_to_be32(blksize),
1413 			      cpu_to_be64(block_id));
1414 }
1415 
1416 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1417 			     struct drbd_peer_request *peer_req)
1418 {
1419 	struct drbd_socket *sock;
1420 	struct p_block_desc *p;
1421 
1422 	sock = &peer_device->connection->data;
1423 	p = drbd_prepare_command(peer_device, sock);
1424 	if (!p)
1425 		return -EIO;
1426 	p->sector = cpu_to_be64(peer_req->i.sector);
1427 	p->blksize = cpu_to_be32(peer_req->i.size);
1428 	p->pad = 0;
1429 	return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1430 }
1431 
1432 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1433 		       sector_t sector, int size, u64 block_id)
1434 {
1435 	struct drbd_socket *sock;
1436 	struct p_block_req *p;
1437 
1438 	sock = &peer_device->connection->data;
1439 	p = drbd_prepare_command(peer_device, sock);
1440 	if (!p)
1441 		return -EIO;
1442 	p->sector = cpu_to_be64(sector);
1443 	p->block_id = block_id;
1444 	p->blksize = cpu_to_be32(size);
1445 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1446 }
1447 
1448 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1449 			    void *digest, int digest_size, enum drbd_packet cmd)
1450 {
1451 	struct drbd_socket *sock;
1452 	struct p_block_req *p;
1453 
1454 	/* FIXME: Put the digest into the preallocated socket buffer.  */
1455 
1456 	sock = &peer_device->connection->data;
1457 	p = drbd_prepare_command(peer_device, sock);
1458 	if (!p)
1459 		return -EIO;
1460 	p->sector = cpu_to_be64(sector);
1461 	p->block_id = ID_SYNCER /* unused */;
1462 	p->blksize = cpu_to_be32(size);
1463 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1464 }
1465 
1466 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1467 {
1468 	struct drbd_socket *sock;
1469 	struct p_block_req *p;
1470 
1471 	sock = &peer_device->connection->data;
1472 	p = drbd_prepare_command(peer_device, sock);
1473 	if (!p)
1474 		return -EIO;
1475 	p->sector = cpu_to_be64(sector);
1476 	p->block_id = ID_SYNCER /* unused */;
1477 	p->blksize = cpu_to_be32(size);
1478 	return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1479 }
1480 
1481 /* called on sndtimeo
1482  * returns false if we should retry,
1483  * true if we think connection is dead
1484  */
1485 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1486 {
1487 	int drop_it;
1488 	/* long elapsed = (long)(jiffies - device->last_received); */
1489 
1490 	drop_it =   connection->meta.socket == sock
1491 		|| !connection->ack_receiver.task
1492 		|| get_t_state(&connection->ack_receiver) != RUNNING
1493 		|| connection->cstate < C_WF_REPORT_PARAMS;
1494 
1495 	if (drop_it)
1496 		return true;
1497 
1498 	drop_it = !--connection->ko_count;
1499 	if (!drop_it) {
1500 		drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1501 			 current->comm, current->pid, connection->ko_count);
1502 		request_ping(connection);
1503 	}
1504 
1505 	return drop_it; /* && (device->state == R_PRIMARY) */;
1506 }
1507 
1508 static void drbd_update_congested(struct drbd_connection *connection)
1509 {
1510 	struct sock *sk = connection->data.socket->sk;
1511 	if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1512 		set_bit(NET_CONGESTED, &connection->flags);
1513 }
1514 
1515 /* The idea of sendpage seems to be to put some kind of reference
1516  * to the page into the skb, and to hand it over to the NIC. In
1517  * this process get_page() gets called.
1518  *
1519  * As soon as the page was really sent over the network put_page()
1520  * gets called by some part of the network layer. [ NIC driver? ]
1521  *
1522  * [ get_page() / put_page() increment/decrement the count. If count
1523  *   reaches 0 the page will be freed. ]
1524  *
1525  * This works nicely with pages from FSs.
1526  * But this means that in protocol A we might signal IO completion too early!
1527  *
1528  * In order not to corrupt data during a resync we must make sure
1529  * that we do not reuse our own buffer pages (EEs) to early, therefore
1530  * we have the net_ee list.
1531  *
1532  * XFS seems to have problems, still, it submits pages with page_count == 0!
1533  * As a workaround, we disable sendpage on pages
1534  * with page_count == 0 or PageSlab.
1535  */
1536 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1537 			      int offset, size_t size, unsigned msg_flags)
1538 {
1539 	struct socket *socket;
1540 	void *addr;
1541 	int err;
1542 
1543 	socket = peer_device->connection->data.socket;
1544 	addr = kmap(page) + offset;
1545 	err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1546 	kunmap(page);
1547 	if (!err)
1548 		peer_device->device->send_cnt += size >> 9;
1549 	return err;
1550 }
1551 
1552 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1553 		    int offset, size_t size, unsigned msg_flags)
1554 {
1555 	struct socket *socket = peer_device->connection->data.socket;
1556 	int len = size;
1557 	int err = -EIO;
1558 
1559 	/* e.g. XFS meta- & log-data is in slab pages, which have a
1560 	 * page_count of 0 and/or have PageSlab() set.
1561 	 * we cannot use send_page for those, as that does get_page();
1562 	 * put_page(); and would cause either a VM_BUG directly, or
1563 	 * __page_cache_release a page that would actually still be referenced
1564 	 * by someone, leading to some obscure delayed Oops somewhere else. */
1565 	if (drbd_disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1566 		return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1567 
1568 	msg_flags |= MSG_NOSIGNAL;
1569 	drbd_update_congested(peer_device->connection);
1570 	do {
1571 		int sent;
1572 
1573 		sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1574 		if (sent <= 0) {
1575 			if (sent == -EAGAIN) {
1576 				if (we_should_drop_the_connection(peer_device->connection, socket))
1577 					break;
1578 				continue;
1579 			}
1580 			drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1581 			     __func__, (int)size, len, sent);
1582 			if (sent < 0)
1583 				err = sent;
1584 			break;
1585 		}
1586 		len    -= sent;
1587 		offset += sent;
1588 	} while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1589 	clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1590 
1591 	if (len == 0) {
1592 		err = 0;
1593 		peer_device->device->send_cnt += size >> 9;
1594 	}
1595 	return err;
1596 }
1597 
1598 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1599 {
1600 	struct bio_vec bvec;
1601 	struct bvec_iter iter;
1602 
1603 	/* hint all but last page with MSG_MORE */
1604 	bio_for_each_segment(bvec, bio, iter) {
1605 		int err;
1606 
1607 		err = _drbd_no_send_page(peer_device, bvec.bv_page,
1608 					 bvec.bv_offset, bvec.bv_len,
1609 					 bio_iter_last(bvec, iter)
1610 					 ? 0 : MSG_MORE);
1611 		if (err)
1612 			return err;
1613 		/* REQ_OP_WRITE_SAME has only one segment */
1614 		if (bio_op(bio) == REQ_OP_WRITE_SAME)
1615 			break;
1616 	}
1617 	return 0;
1618 }
1619 
1620 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1621 {
1622 	struct bio_vec bvec;
1623 	struct bvec_iter iter;
1624 
1625 	/* hint all but last page with MSG_MORE */
1626 	bio_for_each_segment(bvec, bio, iter) {
1627 		int err;
1628 
1629 		err = _drbd_send_page(peer_device, bvec.bv_page,
1630 				      bvec.bv_offset, bvec.bv_len,
1631 				      bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1632 		if (err)
1633 			return err;
1634 		/* REQ_OP_WRITE_SAME has only one segment */
1635 		if (bio_op(bio) == REQ_OP_WRITE_SAME)
1636 			break;
1637 	}
1638 	return 0;
1639 }
1640 
1641 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1642 			    struct drbd_peer_request *peer_req)
1643 {
1644 	struct page *page = peer_req->pages;
1645 	unsigned len = peer_req->i.size;
1646 	int err;
1647 
1648 	/* hint all but last page with MSG_MORE */
1649 	page_chain_for_each(page) {
1650 		unsigned l = min_t(unsigned, len, PAGE_SIZE);
1651 
1652 		err = _drbd_send_page(peer_device, page, 0, l,
1653 				      page_chain_next(page) ? MSG_MORE : 0);
1654 		if (err)
1655 			return err;
1656 		len -= l;
1657 	}
1658 	return 0;
1659 }
1660 
1661 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1662 			     struct bio *bio)
1663 {
1664 	if (connection->agreed_pro_version >= 95)
1665 		return  (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1666 			(bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1667 			(bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1668 			(bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1669 			(bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1670 			(bio_op(bio) == REQ_OP_WRITE_ZEROES ? DP_DISCARD : 0);
1671 	else
1672 		return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1673 }
1674 
1675 /* Used to send write or TRIM aka REQ_DISCARD requests
1676  * R_PRIMARY -> Peer	(P_DATA, P_TRIM)
1677  */
1678 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1679 {
1680 	struct drbd_device *device = peer_device->device;
1681 	struct drbd_socket *sock;
1682 	struct p_data *p;
1683 	struct p_wsame *wsame = NULL;
1684 	void *digest_out;
1685 	unsigned int dp_flags = 0;
1686 	int digest_size;
1687 	int err;
1688 
1689 	sock = &peer_device->connection->data;
1690 	p = drbd_prepare_command(peer_device, sock);
1691 	digest_size = peer_device->connection->integrity_tfm ?
1692 		      crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1693 
1694 	if (!p)
1695 		return -EIO;
1696 	p->sector = cpu_to_be64(req->i.sector);
1697 	p->block_id = (unsigned long)req;
1698 	p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1699 	dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1700 	if (device->state.conn >= C_SYNC_SOURCE &&
1701 	    device->state.conn <= C_PAUSED_SYNC_T)
1702 		dp_flags |= DP_MAY_SET_IN_SYNC;
1703 	if (peer_device->connection->agreed_pro_version >= 100) {
1704 		if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1705 			dp_flags |= DP_SEND_RECEIVE_ACK;
1706 		/* During resync, request an explicit write ack,
1707 		 * even in protocol != C */
1708 		if (req->rq_state & RQ_EXP_WRITE_ACK
1709 		|| (dp_flags & DP_MAY_SET_IN_SYNC))
1710 			dp_flags |= DP_SEND_WRITE_ACK;
1711 	}
1712 	p->dp_flags = cpu_to_be32(dp_flags);
1713 
1714 	if (dp_flags & DP_DISCARD) {
1715 		struct p_trim *t = (struct p_trim*)p;
1716 		t->size = cpu_to_be32(req->i.size);
1717 		err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1718 		goto out;
1719 	}
1720 	if (dp_flags & DP_WSAME) {
1721 		/* this will only work if DRBD_FF_WSAME is set AND the
1722 		 * handshake agreed that all nodes and backend devices are
1723 		 * WRITE_SAME capable and agree on logical_block_size */
1724 		wsame = (struct p_wsame*)p;
1725 		digest_out = wsame + 1;
1726 		wsame->size = cpu_to_be32(req->i.size);
1727 	} else
1728 		digest_out = p + 1;
1729 
1730 	/* our digest is still only over the payload.
1731 	 * TRIM does not carry any payload. */
1732 	if (digest_size)
1733 		drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1734 	if (wsame) {
1735 		err =
1736 		    __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1737 				   sizeof(*wsame) + digest_size, NULL,
1738 				   bio_iovec(req->master_bio).bv_len);
1739 	} else
1740 		err =
1741 		    __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1742 				   sizeof(*p) + digest_size, NULL, req->i.size);
1743 	if (!err) {
1744 		/* For protocol A, we have to memcpy the payload into
1745 		 * socket buffers, as we may complete right away
1746 		 * as soon as we handed it over to tcp, at which point the data
1747 		 * pages may become invalid.
1748 		 *
1749 		 * For data-integrity enabled, we copy it as well, so we can be
1750 		 * sure that even if the bio pages may still be modified, it
1751 		 * won't change the data on the wire, thus if the digest checks
1752 		 * out ok after sending on this side, but does not fit on the
1753 		 * receiving side, we sure have detected corruption elsewhere.
1754 		 */
1755 		if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1756 			err = _drbd_send_bio(peer_device, req->master_bio);
1757 		else
1758 			err = _drbd_send_zc_bio(peer_device, req->master_bio);
1759 
1760 		/* double check digest, sometimes buffers have been modified in flight. */
1761 		if (digest_size > 0 && digest_size <= 64) {
1762 			/* 64 byte, 512 bit, is the largest digest size
1763 			 * currently supported in kernel crypto. */
1764 			unsigned char digest[64];
1765 			drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1766 			if (memcmp(p + 1, digest, digest_size)) {
1767 				drbd_warn(device,
1768 					"Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1769 					(unsigned long long)req->i.sector, req->i.size);
1770 			}
1771 		} /* else if (digest_size > 64) {
1772 		     ... Be noisy about digest too large ...
1773 		} */
1774 	}
1775 out:
1776 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1777 
1778 	return err;
1779 }
1780 
1781 /* answer packet, used to send data back for read requests:
1782  *  Peer       -> (diskless) R_PRIMARY   (P_DATA_REPLY)
1783  *  C_SYNC_SOURCE -> C_SYNC_TARGET         (P_RS_DATA_REPLY)
1784  */
1785 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1786 		    struct drbd_peer_request *peer_req)
1787 {
1788 	struct drbd_device *device = peer_device->device;
1789 	struct drbd_socket *sock;
1790 	struct p_data *p;
1791 	int err;
1792 	int digest_size;
1793 
1794 	sock = &peer_device->connection->data;
1795 	p = drbd_prepare_command(peer_device, sock);
1796 
1797 	digest_size = peer_device->connection->integrity_tfm ?
1798 		      crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1799 
1800 	if (!p)
1801 		return -EIO;
1802 	p->sector = cpu_to_be64(peer_req->i.sector);
1803 	p->block_id = peer_req->block_id;
1804 	p->seq_num = 0;  /* unused */
1805 	p->dp_flags = 0;
1806 	if (digest_size)
1807 		drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1808 	err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1809 	if (!err)
1810 		err = _drbd_send_zc_ee(peer_device, peer_req);
1811 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1812 
1813 	return err;
1814 }
1815 
1816 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1817 {
1818 	struct drbd_socket *sock;
1819 	struct p_block_desc *p;
1820 
1821 	sock = &peer_device->connection->data;
1822 	p = drbd_prepare_command(peer_device, sock);
1823 	if (!p)
1824 		return -EIO;
1825 	p->sector = cpu_to_be64(req->i.sector);
1826 	p->blksize = cpu_to_be32(req->i.size);
1827 	return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1828 }
1829 
1830 /*
1831   drbd_send distinguishes two cases:
1832 
1833   Packets sent via the data socket "sock"
1834   and packets sent via the meta data socket "msock"
1835 
1836 		    sock                      msock
1837   -----------------+-------------------------+------------------------------
1838   timeout           conf.timeout / 2          conf.timeout / 2
1839   timeout action    send a ping via msock     Abort communication
1840 					      and close all sockets
1841 */
1842 
1843 /*
1844  * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1845  */
1846 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1847 	      void *buf, size_t size, unsigned msg_flags)
1848 {
1849 	struct kvec iov = {.iov_base = buf, .iov_len = size};
1850 	struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL};
1851 	int rv, sent = 0;
1852 
1853 	if (!sock)
1854 		return -EBADR;
1855 
1856 	/* THINK  if (signal_pending) return ... ? */
1857 
1858 	iov_iter_kvec(&msg.msg_iter, WRITE | ITER_KVEC, &iov, 1, size);
1859 
1860 	if (sock == connection->data.socket) {
1861 		rcu_read_lock();
1862 		connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1863 		rcu_read_unlock();
1864 		drbd_update_congested(connection);
1865 	}
1866 	do {
1867 		rv = sock_sendmsg(sock, &msg);
1868 		if (rv == -EAGAIN) {
1869 			if (we_should_drop_the_connection(connection, sock))
1870 				break;
1871 			else
1872 				continue;
1873 		}
1874 		if (rv == -EINTR) {
1875 			flush_signals(current);
1876 			rv = 0;
1877 		}
1878 		if (rv < 0)
1879 			break;
1880 		sent += rv;
1881 	} while (sent < size);
1882 
1883 	if (sock == connection->data.socket)
1884 		clear_bit(NET_CONGESTED, &connection->flags);
1885 
1886 	if (rv <= 0) {
1887 		if (rv != -EAGAIN) {
1888 			drbd_err(connection, "%s_sendmsg returned %d\n",
1889 				 sock == connection->meta.socket ? "msock" : "sock",
1890 				 rv);
1891 			conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1892 		} else
1893 			conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1894 	}
1895 
1896 	return sent;
1897 }
1898 
1899 /**
1900  * drbd_send_all  -  Send an entire buffer
1901  *
1902  * Returns 0 upon success and a negative error value otherwise.
1903  */
1904 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1905 		  size_t size, unsigned msg_flags)
1906 {
1907 	int err;
1908 
1909 	err = drbd_send(connection, sock, buffer, size, msg_flags);
1910 	if (err < 0)
1911 		return err;
1912 	if (err != size)
1913 		return -EIO;
1914 	return 0;
1915 }
1916 
1917 static int drbd_open(struct block_device *bdev, fmode_t mode)
1918 {
1919 	struct drbd_device *device = bdev->bd_disk->private_data;
1920 	unsigned long flags;
1921 	int rv = 0;
1922 
1923 	mutex_lock(&drbd_main_mutex);
1924 	spin_lock_irqsave(&device->resource->req_lock, flags);
1925 	/* to have a stable device->state.role
1926 	 * and no race with updating open_cnt */
1927 
1928 	if (device->state.role != R_PRIMARY) {
1929 		if (mode & FMODE_WRITE)
1930 			rv = -EROFS;
1931 		else if (!drbd_allow_oos)
1932 			rv = -EMEDIUMTYPE;
1933 	}
1934 
1935 	if (!rv)
1936 		device->open_cnt++;
1937 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
1938 	mutex_unlock(&drbd_main_mutex);
1939 
1940 	return rv;
1941 }
1942 
1943 static void drbd_release(struct gendisk *gd, fmode_t mode)
1944 {
1945 	struct drbd_device *device = gd->private_data;
1946 	mutex_lock(&drbd_main_mutex);
1947 	device->open_cnt--;
1948 	mutex_unlock(&drbd_main_mutex);
1949 }
1950 
1951 /* need to hold resource->req_lock */
1952 void drbd_queue_unplug(struct drbd_device *device)
1953 {
1954 	if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1955 		D_ASSERT(device, device->state.role == R_PRIMARY);
1956 		if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1957 			drbd_queue_work_if_unqueued(
1958 				&first_peer_device(device)->connection->sender_work,
1959 				&device->unplug_work);
1960 		}
1961 	}
1962 }
1963 
1964 static void drbd_set_defaults(struct drbd_device *device)
1965 {
1966 	/* Beware! The actual layout differs
1967 	 * between big endian and little endian */
1968 	device->state = (union drbd_dev_state) {
1969 		{ .role = R_SECONDARY,
1970 		  .peer = R_UNKNOWN,
1971 		  .conn = C_STANDALONE,
1972 		  .disk = D_DISKLESS,
1973 		  .pdsk = D_UNKNOWN,
1974 		} };
1975 }
1976 
1977 void drbd_init_set_defaults(struct drbd_device *device)
1978 {
1979 	/* the memset(,0,) did most of this.
1980 	 * note: only assignments, no allocation in here */
1981 
1982 	drbd_set_defaults(device);
1983 
1984 	atomic_set(&device->ap_bio_cnt, 0);
1985 	atomic_set(&device->ap_actlog_cnt, 0);
1986 	atomic_set(&device->ap_pending_cnt, 0);
1987 	atomic_set(&device->rs_pending_cnt, 0);
1988 	atomic_set(&device->unacked_cnt, 0);
1989 	atomic_set(&device->local_cnt, 0);
1990 	atomic_set(&device->pp_in_use_by_net, 0);
1991 	atomic_set(&device->rs_sect_in, 0);
1992 	atomic_set(&device->rs_sect_ev, 0);
1993 	atomic_set(&device->ap_in_flight, 0);
1994 	atomic_set(&device->md_io.in_use, 0);
1995 
1996 	mutex_init(&device->own_state_mutex);
1997 	device->state_mutex = &device->own_state_mutex;
1998 
1999 	spin_lock_init(&device->al_lock);
2000 	spin_lock_init(&device->peer_seq_lock);
2001 
2002 	INIT_LIST_HEAD(&device->active_ee);
2003 	INIT_LIST_HEAD(&device->sync_ee);
2004 	INIT_LIST_HEAD(&device->done_ee);
2005 	INIT_LIST_HEAD(&device->read_ee);
2006 	INIT_LIST_HEAD(&device->net_ee);
2007 	INIT_LIST_HEAD(&device->resync_reads);
2008 	INIT_LIST_HEAD(&device->resync_work.list);
2009 	INIT_LIST_HEAD(&device->unplug_work.list);
2010 	INIT_LIST_HEAD(&device->bm_io_work.w.list);
2011 	INIT_LIST_HEAD(&device->pending_master_completion[0]);
2012 	INIT_LIST_HEAD(&device->pending_master_completion[1]);
2013 	INIT_LIST_HEAD(&device->pending_completion[0]);
2014 	INIT_LIST_HEAD(&device->pending_completion[1]);
2015 
2016 	device->resync_work.cb  = w_resync_timer;
2017 	device->unplug_work.cb  = w_send_write_hint;
2018 	device->bm_io_work.w.cb = w_bitmap_io;
2019 
2020 	timer_setup(&device->resync_timer, resync_timer_fn, 0);
2021 	timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0);
2022 	timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0);
2023 	timer_setup(&device->request_timer, request_timer_fn, 0);
2024 
2025 	init_waitqueue_head(&device->misc_wait);
2026 	init_waitqueue_head(&device->state_wait);
2027 	init_waitqueue_head(&device->ee_wait);
2028 	init_waitqueue_head(&device->al_wait);
2029 	init_waitqueue_head(&device->seq_wait);
2030 
2031 	device->resync_wenr = LC_FREE;
2032 	device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2033 	device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2034 }
2035 
2036 void drbd_device_cleanup(struct drbd_device *device)
2037 {
2038 	int i;
2039 	if (first_peer_device(device)->connection->receiver.t_state != NONE)
2040 		drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2041 				first_peer_device(device)->connection->receiver.t_state);
2042 
2043 	device->al_writ_cnt  =
2044 	device->bm_writ_cnt  =
2045 	device->read_cnt     =
2046 	device->recv_cnt     =
2047 	device->send_cnt     =
2048 	device->writ_cnt     =
2049 	device->p_size       =
2050 	device->rs_start     =
2051 	device->rs_total     =
2052 	device->rs_failed    = 0;
2053 	device->rs_last_events = 0;
2054 	device->rs_last_sect_ev = 0;
2055 	for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2056 		device->rs_mark_left[i] = 0;
2057 		device->rs_mark_time[i] = 0;
2058 	}
2059 	D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2060 
2061 	drbd_set_my_capacity(device, 0);
2062 	if (device->bitmap) {
2063 		/* maybe never allocated. */
2064 		drbd_bm_resize(device, 0, 1);
2065 		drbd_bm_cleanup(device);
2066 	}
2067 
2068 	drbd_backing_dev_free(device, device->ldev);
2069 	device->ldev = NULL;
2070 
2071 	clear_bit(AL_SUSPENDED, &device->flags);
2072 
2073 	D_ASSERT(device, list_empty(&device->active_ee));
2074 	D_ASSERT(device, list_empty(&device->sync_ee));
2075 	D_ASSERT(device, list_empty(&device->done_ee));
2076 	D_ASSERT(device, list_empty(&device->read_ee));
2077 	D_ASSERT(device, list_empty(&device->net_ee));
2078 	D_ASSERT(device, list_empty(&device->resync_reads));
2079 	D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2080 	D_ASSERT(device, list_empty(&device->resync_work.list));
2081 	D_ASSERT(device, list_empty(&device->unplug_work.list));
2082 
2083 	drbd_set_defaults(device);
2084 }
2085 
2086 
2087 static void drbd_destroy_mempools(void)
2088 {
2089 	struct page *page;
2090 
2091 	while (drbd_pp_pool) {
2092 		page = drbd_pp_pool;
2093 		drbd_pp_pool = (struct page *)page_private(page);
2094 		__free_page(page);
2095 		drbd_pp_vacant--;
2096 	}
2097 
2098 	/* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2099 
2100 	if (drbd_io_bio_set)
2101 		bioset_free(drbd_io_bio_set);
2102 	if (drbd_md_io_bio_set)
2103 		bioset_free(drbd_md_io_bio_set);
2104 	if (drbd_md_io_page_pool)
2105 		mempool_destroy(drbd_md_io_page_pool);
2106 	if (drbd_ee_mempool)
2107 		mempool_destroy(drbd_ee_mempool);
2108 	if (drbd_request_mempool)
2109 		mempool_destroy(drbd_request_mempool);
2110 	if (drbd_ee_cache)
2111 		kmem_cache_destroy(drbd_ee_cache);
2112 	if (drbd_request_cache)
2113 		kmem_cache_destroy(drbd_request_cache);
2114 	if (drbd_bm_ext_cache)
2115 		kmem_cache_destroy(drbd_bm_ext_cache);
2116 	if (drbd_al_ext_cache)
2117 		kmem_cache_destroy(drbd_al_ext_cache);
2118 
2119 	drbd_io_bio_set      = NULL;
2120 	drbd_md_io_bio_set   = NULL;
2121 	drbd_md_io_page_pool = NULL;
2122 	drbd_ee_mempool      = NULL;
2123 	drbd_request_mempool = NULL;
2124 	drbd_ee_cache        = NULL;
2125 	drbd_request_cache   = NULL;
2126 	drbd_bm_ext_cache    = NULL;
2127 	drbd_al_ext_cache    = NULL;
2128 
2129 	return;
2130 }
2131 
2132 static int drbd_create_mempools(void)
2133 {
2134 	struct page *page;
2135 	const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2136 	int i;
2137 
2138 	/* prepare our caches and mempools */
2139 	drbd_request_mempool = NULL;
2140 	drbd_ee_cache        = NULL;
2141 	drbd_request_cache   = NULL;
2142 	drbd_bm_ext_cache    = NULL;
2143 	drbd_al_ext_cache    = NULL;
2144 	drbd_pp_pool         = NULL;
2145 	drbd_md_io_page_pool = NULL;
2146 	drbd_md_io_bio_set   = NULL;
2147 	drbd_io_bio_set      = NULL;
2148 
2149 	/* caches */
2150 	drbd_request_cache = kmem_cache_create(
2151 		"drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2152 	if (drbd_request_cache == NULL)
2153 		goto Enomem;
2154 
2155 	drbd_ee_cache = kmem_cache_create(
2156 		"drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2157 	if (drbd_ee_cache == NULL)
2158 		goto Enomem;
2159 
2160 	drbd_bm_ext_cache = kmem_cache_create(
2161 		"drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2162 	if (drbd_bm_ext_cache == NULL)
2163 		goto Enomem;
2164 
2165 	drbd_al_ext_cache = kmem_cache_create(
2166 		"drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2167 	if (drbd_al_ext_cache == NULL)
2168 		goto Enomem;
2169 
2170 	/* mempools */
2171 	drbd_io_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
2172 	if (drbd_io_bio_set == NULL)
2173 		goto Enomem;
2174 
2175 	drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0,
2176 					   BIOSET_NEED_BVECS);
2177 	if (drbd_md_io_bio_set == NULL)
2178 		goto Enomem;
2179 
2180 	drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2181 	if (drbd_md_io_page_pool == NULL)
2182 		goto Enomem;
2183 
2184 	drbd_request_mempool = mempool_create_slab_pool(number,
2185 		drbd_request_cache);
2186 	if (drbd_request_mempool == NULL)
2187 		goto Enomem;
2188 
2189 	drbd_ee_mempool = mempool_create_slab_pool(number, drbd_ee_cache);
2190 	if (drbd_ee_mempool == NULL)
2191 		goto Enomem;
2192 
2193 	/* drbd's page pool */
2194 	spin_lock_init(&drbd_pp_lock);
2195 
2196 	for (i = 0; i < number; i++) {
2197 		page = alloc_page(GFP_HIGHUSER);
2198 		if (!page)
2199 			goto Enomem;
2200 		set_page_private(page, (unsigned long)drbd_pp_pool);
2201 		drbd_pp_pool = page;
2202 	}
2203 	drbd_pp_vacant = number;
2204 
2205 	return 0;
2206 
2207 Enomem:
2208 	drbd_destroy_mempools(); /* in case we allocated some */
2209 	return -ENOMEM;
2210 }
2211 
2212 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2213 {
2214 	int rr;
2215 
2216 	rr = drbd_free_peer_reqs(device, &device->active_ee);
2217 	if (rr)
2218 		drbd_err(device, "%d EEs in active list found!\n", rr);
2219 
2220 	rr = drbd_free_peer_reqs(device, &device->sync_ee);
2221 	if (rr)
2222 		drbd_err(device, "%d EEs in sync list found!\n", rr);
2223 
2224 	rr = drbd_free_peer_reqs(device, &device->read_ee);
2225 	if (rr)
2226 		drbd_err(device, "%d EEs in read list found!\n", rr);
2227 
2228 	rr = drbd_free_peer_reqs(device, &device->done_ee);
2229 	if (rr)
2230 		drbd_err(device, "%d EEs in done list found!\n", rr);
2231 
2232 	rr = drbd_free_peer_reqs(device, &device->net_ee);
2233 	if (rr)
2234 		drbd_err(device, "%d EEs in net list found!\n", rr);
2235 }
2236 
2237 /* caution. no locking. */
2238 void drbd_destroy_device(struct kref *kref)
2239 {
2240 	struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2241 	struct drbd_resource *resource = device->resource;
2242 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2243 
2244 	del_timer_sync(&device->request_timer);
2245 
2246 	/* paranoia asserts */
2247 	D_ASSERT(device, device->open_cnt == 0);
2248 	/* end paranoia asserts */
2249 
2250 	/* cleanup stuff that may have been allocated during
2251 	 * device (re-)configuration or state changes */
2252 
2253 	if (device->this_bdev)
2254 		bdput(device->this_bdev);
2255 
2256 	drbd_backing_dev_free(device, device->ldev);
2257 	device->ldev = NULL;
2258 
2259 	drbd_release_all_peer_reqs(device);
2260 
2261 	lc_destroy(device->act_log);
2262 	lc_destroy(device->resync);
2263 
2264 	kfree(device->p_uuid);
2265 	/* device->p_uuid = NULL; */
2266 
2267 	if (device->bitmap) /* should no longer be there. */
2268 		drbd_bm_cleanup(device);
2269 	__free_page(device->md_io.page);
2270 	put_disk(device->vdisk);
2271 	blk_cleanup_queue(device->rq_queue);
2272 	kfree(device->rs_plan_s);
2273 
2274 	/* not for_each_connection(connection, resource):
2275 	 * those may have been cleaned up and disassociated already.
2276 	 */
2277 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2278 		kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2279 		kfree(peer_device);
2280 	}
2281 	memset(device, 0xfd, sizeof(*device));
2282 	kfree(device);
2283 	kref_put(&resource->kref, drbd_destroy_resource);
2284 }
2285 
2286 /* One global retry thread, if we need to push back some bio and have it
2287  * reinserted through our make request function.
2288  */
2289 static struct retry_worker {
2290 	struct workqueue_struct *wq;
2291 	struct work_struct worker;
2292 
2293 	spinlock_t lock;
2294 	struct list_head writes;
2295 } retry;
2296 
2297 static void do_retry(struct work_struct *ws)
2298 {
2299 	struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2300 	LIST_HEAD(writes);
2301 	struct drbd_request *req, *tmp;
2302 
2303 	spin_lock_irq(&retry->lock);
2304 	list_splice_init(&retry->writes, &writes);
2305 	spin_unlock_irq(&retry->lock);
2306 
2307 	list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2308 		struct drbd_device *device = req->device;
2309 		struct bio *bio = req->master_bio;
2310 		unsigned long start_jif = req->start_jif;
2311 		bool expected;
2312 
2313 		expected =
2314 			expect(atomic_read(&req->completion_ref) == 0) &&
2315 			expect(req->rq_state & RQ_POSTPONED) &&
2316 			expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2317 				(req->rq_state & RQ_LOCAL_ABORTED) != 0);
2318 
2319 		if (!expected)
2320 			drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2321 				req, atomic_read(&req->completion_ref),
2322 				req->rq_state);
2323 
2324 		/* We still need to put one kref associated with the
2325 		 * "completion_ref" going zero in the code path that queued it
2326 		 * here.  The request object may still be referenced by a
2327 		 * frozen local req->private_bio, in case we force-detached.
2328 		 */
2329 		kref_put(&req->kref, drbd_req_destroy);
2330 
2331 		/* A single suspended or otherwise blocking device may stall
2332 		 * all others as well.  Fortunately, this code path is to
2333 		 * recover from a situation that "should not happen":
2334 		 * concurrent writes in multi-primary setup.
2335 		 * In a "normal" lifecycle, this workqueue is supposed to be
2336 		 * destroyed without ever doing anything.
2337 		 * If it turns out to be an issue anyways, we can do per
2338 		 * resource (replication group) or per device (minor) retry
2339 		 * workqueues instead.
2340 		 */
2341 
2342 		/* We are not just doing generic_make_request(),
2343 		 * as we want to keep the start_time information. */
2344 		inc_ap_bio(device);
2345 		__drbd_make_request(device, bio, start_jif);
2346 	}
2347 }
2348 
2349 /* called via drbd_req_put_completion_ref(),
2350  * holds resource->req_lock */
2351 void drbd_restart_request(struct drbd_request *req)
2352 {
2353 	unsigned long flags;
2354 	spin_lock_irqsave(&retry.lock, flags);
2355 	list_move_tail(&req->tl_requests, &retry.writes);
2356 	spin_unlock_irqrestore(&retry.lock, flags);
2357 
2358 	/* Drop the extra reference that would otherwise
2359 	 * have been dropped by complete_master_bio.
2360 	 * do_retry() needs to grab a new one. */
2361 	dec_ap_bio(req->device);
2362 
2363 	queue_work(retry.wq, &retry.worker);
2364 }
2365 
2366 void drbd_destroy_resource(struct kref *kref)
2367 {
2368 	struct drbd_resource *resource =
2369 		container_of(kref, struct drbd_resource, kref);
2370 
2371 	idr_destroy(&resource->devices);
2372 	free_cpumask_var(resource->cpu_mask);
2373 	kfree(resource->name);
2374 	memset(resource, 0xf2, sizeof(*resource));
2375 	kfree(resource);
2376 }
2377 
2378 void drbd_free_resource(struct drbd_resource *resource)
2379 {
2380 	struct drbd_connection *connection, *tmp;
2381 
2382 	for_each_connection_safe(connection, tmp, resource) {
2383 		list_del(&connection->connections);
2384 		drbd_debugfs_connection_cleanup(connection);
2385 		kref_put(&connection->kref, drbd_destroy_connection);
2386 	}
2387 	drbd_debugfs_resource_cleanup(resource);
2388 	kref_put(&resource->kref, drbd_destroy_resource);
2389 }
2390 
2391 static void drbd_cleanup(void)
2392 {
2393 	unsigned int i;
2394 	struct drbd_device *device;
2395 	struct drbd_resource *resource, *tmp;
2396 
2397 	/* first remove proc,
2398 	 * drbdsetup uses it's presence to detect
2399 	 * whether DRBD is loaded.
2400 	 * If we would get stuck in proc removal,
2401 	 * but have netlink already deregistered,
2402 	 * some drbdsetup commands may wait forever
2403 	 * for an answer.
2404 	 */
2405 	if (drbd_proc)
2406 		remove_proc_entry("drbd", NULL);
2407 
2408 	if (retry.wq)
2409 		destroy_workqueue(retry.wq);
2410 
2411 	drbd_genl_unregister();
2412 
2413 	idr_for_each_entry(&drbd_devices, device, i)
2414 		drbd_delete_device(device);
2415 
2416 	/* not _rcu since, no other updater anymore. Genl already unregistered */
2417 	for_each_resource_safe(resource, tmp, &drbd_resources) {
2418 		list_del(&resource->resources);
2419 		drbd_free_resource(resource);
2420 	}
2421 
2422 	drbd_debugfs_cleanup();
2423 
2424 	drbd_destroy_mempools();
2425 	unregister_blkdev(DRBD_MAJOR, "drbd");
2426 
2427 	idr_destroy(&drbd_devices);
2428 
2429 	pr_info("module cleanup done.\n");
2430 }
2431 
2432 /**
2433  * drbd_congested() - Callback for the flusher thread
2434  * @congested_data:	User data
2435  * @bdi_bits:		Bits the BDI flusher thread is currently interested in
2436  *
2437  * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested.
2438  */
2439 static int drbd_congested(void *congested_data, int bdi_bits)
2440 {
2441 	struct drbd_device *device = congested_data;
2442 	struct request_queue *q;
2443 	char reason = '-';
2444 	int r = 0;
2445 
2446 	if (!may_inc_ap_bio(device)) {
2447 		/* DRBD has frozen IO */
2448 		r = bdi_bits;
2449 		reason = 'd';
2450 		goto out;
2451 	}
2452 
2453 	if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
2454 		r |= (1 << WB_async_congested);
2455 		/* Without good local data, we would need to read from remote,
2456 		 * and that would need the worker thread as well, which is
2457 		 * currently blocked waiting for that usermode helper to
2458 		 * finish.
2459 		 */
2460 		if (!get_ldev_if_state(device, D_UP_TO_DATE))
2461 			r |= (1 << WB_sync_congested);
2462 		else
2463 			put_ldev(device);
2464 		r &= bdi_bits;
2465 		reason = 'c';
2466 		goto out;
2467 	}
2468 
2469 	if (get_ldev(device)) {
2470 		q = bdev_get_queue(device->ldev->backing_bdev);
2471 		r = bdi_congested(q->backing_dev_info, bdi_bits);
2472 		put_ldev(device);
2473 		if (r)
2474 			reason = 'b';
2475 	}
2476 
2477 	if (bdi_bits & (1 << WB_async_congested) &&
2478 	    test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
2479 		r |= (1 << WB_async_congested);
2480 		reason = reason == 'b' ? 'a' : 'n';
2481 	}
2482 
2483 out:
2484 	device->congestion_reason = reason;
2485 	return r;
2486 }
2487 
2488 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2489 {
2490 	spin_lock_init(&wq->q_lock);
2491 	INIT_LIST_HEAD(&wq->q);
2492 	init_waitqueue_head(&wq->q_wait);
2493 }
2494 
2495 struct completion_work {
2496 	struct drbd_work w;
2497 	struct completion done;
2498 };
2499 
2500 static int w_complete(struct drbd_work *w, int cancel)
2501 {
2502 	struct completion_work *completion_work =
2503 		container_of(w, struct completion_work, w);
2504 
2505 	complete(&completion_work->done);
2506 	return 0;
2507 }
2508 
2509 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2510 {
2511 	struct completion_work completion_work;
2512 
2513 	completion_work.w.cb = w_complete;
2514 	init_completion(&completion_work.done);
2515 	drbd_queue_work(work_queue, &completion_work.w);
2516 	wait_for_completion(&completion_work.done);
2517 }
2518 
2519 struct drbd_resource *drbd_find_resource(const char *name)
2520 {
2521 	struct drbd_resource *resource;
2522 
2523 	if (!name || !name[0])
2524 		return NULL;
2525 
2526 	rcu_read_lock();
2527 	for_each_resource_rcu(resource, &drbd_resources) {
2528 		if (!strcmp(resource->name, name)) {
2529 			kref_get(&resource->kref);
2530 			goto found;
2531 		}
2532 	}
2533 	resource = NULL;
2534 found:
2535 	rcu_read_unlock();
2536 	return resource;
2537 }
2538 
2539 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2540 				     void *peer_addr, int peer_addr_len)
2541 {
2542 	struct drbd_resource *resource;
2543 	struct drbd_connection *connection;
2544 
2545 	rcu_read_lock();
2546 	for_each_resource_rcu(resource, &drbd_resources) {
2547 		for_each_connection_rcu(connection, resource) {
2548 			if (connection->my_addr_len == my_addr_len &&
2549 			    connection->peer_addr_len == peer_addr_len &&
2550 			    !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2551 			    !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2552 				kref_get(&connection->kref);
2553 				goto found;
2554 			}
2555 		}
2556 	}
2557 	connection = NULL;
2558 found:
2559 	rcu_read_unlock();
2560 	return connection;
2561 }
2562 
2563 static int drbd_alloc_socket(struct drbd_socket *socket)
2564 {
2565 	socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2566 	if (!socket->rbuf)
2567 		return -ENOMEM;
2568 	socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2569 	if (!socket->sbuf)
2570 		return -ENOMEM;
2571 	return 0;
2572 }
2573 
2574 static void drbd_free_socket(struct drbd_socket *socket)
2575 {
2576 	free_page((unsigned long) socket->sbuf);
2577 	free_page((unsigned long) socket->rbuf);
2578 }
2579 
2580 void conn_free_crypto(struct drbd_connection *connection)
2581 {
2582 	drbd_free_sock(connection);
2583 
2584 	crypto_free_ahash(connection->csums_tfm);
2585 	crypto_free_ahash(connection->verify_tfm);
2586 	crypto_free_shash(connection->cram_hmac_tfm);
2587 	crypto_free_ahash(connection->integrity_tfm);
2588 	crypto_free_ahash(connection->peer_integrity_tfm);
2589 	kfree(connection->int_dig_in);
2590 	kfree(connection->int_dig_vv);
2591 
2592 	connection->csums_tfm = NULL;
2593 	connection->verify_tfm = NULL;
2594 	connection->cram_hmac_tfm = NULL;
2595 	connection->integrity_tfm = NULL;
2596 	connection->peer_integrity_tfm = NULL;
2597 	connection->int_dig_in = NULL;
2598 	connection->int_dig_vv = NULL;
2599 }
2600 
2601 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2602 {
2603 	struct drbd_connection *connection;
2604 	cpumask_var_t new_cpu_mask;
2605 	int err;
2606 
2607 	if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2608 		return -ENOMEM;
2609 
2610 	/* silently ignore cpu mask on UP kernel */
2611 	if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2612 		err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2613 				   cpumask_bits(new_cpu_mask), nr_cpu_ids);
2614 		if (err == -EOVERFLOW) {
2615 			/* So what. mask it out. */
2616 			cpumask_var_t tmp_cpu_mask;
2617 			if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2618 				cpumask_setall(tmp_cpu_mask);
2619 				cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2620 				drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2621 					res_opts->cpu_mask,
2622 					strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2623 					nr_cpu_ids);
2624 				free_cpumask_var(tmp_cpu_mask);
2625 				err = 0;
2626 			}
2627 		}
2628 		if (err) {
2629 			drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2630 			/* retcode = ERR_CPU_MASK_PARSE; */
2631 			goto fail;
2632 		}
2633 	}
2634 	resource->res_opts = *res_opts;
2635 	if (cpumask_empty(new_cpu_mask))
2636 		drbd_calc_cpu_mask(&new_cpu_mask);
2637 	if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2638 		cpumask_copy(resource->cpu_mask, new_cpu_mask);
2639 		for_each_connection_rcu(connection, resource) {
2640 			connection->receiver.reset_cpu_mask = 1;
2641 			connection->ack_receiver.reset_cpu_mask = 1;
2642 			connection->worker.reset_cpu_mask = 1;
2643 		}
2644 	}
2645 	err = 0;
2646 
2647 fail:
2648 	free_cpumask_var(new_cpu_mask);
2649 	return err;
2650 
2651 }
2652 
2653 struct drbd_resource *drbd_create_resource(const char *name)
2654 {
2655 	struct drbd_resource *resource;
2656 
2657 	resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2658 	if (!resource)
2659 		goto fail;
2660 	resource->name = kstrdup(name, GFP_KERNEL);
2661 	if (!resource->name)
2662 		goto fail_free_resource;
2663 	if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2664 		goto fail_free_name;
2665 	kref_init(&resource->kref);
2666 	idr_init(&resource->devices);
2667 	INIT_LIST_HEAD(&resource->connections);
2668 	resource->write_ordering = WO_BDEV_FLUSH;
2669 	list_add_tail_rcu(&resource->resources, &drbd_resources);
2670 	mutex_init(&resource->conf_update);
2671 	mutex_init(&resource->adm_mutex);
2672 	spin_lock_init(&resource->req_lock);
2673 	drbd_debugfs_resource_add(resource);
2674 	return resource;
2675 
2676 fail_free_name:
2677 	kfree(resource->name);
2678 fail_free_resource:
2679 	kfree(resource);
2680 fail:
2681 	return NULL;
2682 }
2683 
2684 /* caller must be under adm_mutex */
2685 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2686 {
2687 	struct drbd_resource *resource;
2688 	struct drbd_connection *connection;
2689 
2690 	connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2691 	if (!connection)
2692 		return NULL;
2693 
2694 	if (drbd_alloc_socket(&connection->data))
2695 		goto fail;
2696 	if (drbd_alloc_socket(&connection->meta))
2697 		goto fail;
2698 
2699 	connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2700 	if (!connection->current_epoch)
2701 		goto fail;
2702 
2703 	INIT_LIST_HEAD(&connection->transfer_log);
2704 
2705 	INIT_LIST_HEAD(&connection->current_epoch->list);
2706 	connection->epochs = 1;
2707 	spin_lock_init(&connection->epoch_lock);
2708 
2709 	connection->send.seen_any_write_yet = false;
2710 	connection->send.current_epoch_nr = 0;
2711 	connection->send.current_epoch_writes = 0;
2712 
2713 	resource = drbd_create_resource(name);
2714 	if (!resource)
2715 		goto fail;
2716 
2717 	connection->cstate = C_STANDALONE;
2718 	mutex_init(&connection->cstate_mutex);
2719 	init_waitqueue_head(&connection->ping_wait);
2720 	idr_init(&connection->peer_devices);
2721 
2722 	drbd_init_workqueue(&connection->sender_work);
2723 	mutex_init(&connection->data.mutex);
2724 	mutex_init(&connection->meta.mutex);
2725 
2726 	drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2727 	connection->receiver.connection = connection;
2728 	drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2729 	connection->worker.connection = connection;
2730 	drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2731 	connection->ack_receiver.connection = connection;
2732 
2733 	kref_init(&connection->kref);
2734 
2735 	connection->resource = resource;
2736 
2737 	if (set_resource_options(resource, res_opts))
2738 		goto fail_resource;
2739 
2740 	kref_get(&resource->kref);
2741 	list_add_tail_rcu(&connection->connections, &resource->connections);
2742 	drbd_debugfs_connection_add(connection);
2743 	return connection;
2744 
2745 fail_resource:
2746 	list_del(&resource->resources);
2747 	drbd_free_resource(resource);
2748 fail:
2749 	kfree(connection->current_epoch);
2750 	drbd_free_socket(&connection->meta);
2751 	drbd_free_socket(&connection->data);
2752 	kfree(connection);
2753 	return NULL;
2754 }
2755 
2756 void drbd_destroy_connection(struct kref *kref)
2757 {
2758 	struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2759 	struct drbd_resource *resource = connection->resource;
2760 
2761 	if (atomic_read(&connection->current_epoch->epoch_size) !=  0)
2762 		drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2763 	kfree(connection->current_epoch);
2764 
2765 	idr_destroy(&connection->peer_devices);
2766 
2767 	drbd_free_socket(&connection->meta);
2768 	drbd_free_socket(&connection->data);
2769 	kfree(connection->int_dig_in);
2770 	kfree(connection->int_dig_vv);
2771 	memset(connection, 0xfc, sizeof(*connection));
2772 	kfree(connection);
2773 	kref_put(&resource->kref, drbd_destroy_resource);
2774 }
2775 
2776 static int init_submitter(struct drbd_device *device)
2777 {
2778 	/* opencoded create_singlethread_workqueue(),
2779 	 * to be able to say "drbd%d", ..., minor */
2780 	device->submit.wq =
2781 		alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2782 	if (!device->submit.wq)
2783 		return -ENOMEM;
2784 
2785 	INIT_WORK(&device->submit.worker, do_submit);
2786 	INIT_LIST_HEAD(&device->submit.writes);
2787 	return 0;
2788 }
2789 
2790 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2791 {
2792 	struct drbd_resource *resource = adm_ctx->resource;
2793 	struct drbd_connection *connection;
2794 	struct drbd_device *device;
2795 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2796 	struct gendisk *disk;
2797 	struct request_queue *q;
2798 	int id;
2799 	int vnr = adm_ctx->volume;
2800 	enum drbd_ret_code err = ERR_NOMEM;
2801 
2802 	device = minor_to_device(minor);
2803 	if (device)
2804 		return ERR_MINOR_OR_VOLUME_EXISTS;
2805 
2806 	/* GFP_KERNEL, we are outside of all write-out paths */
2807 	device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2808 	if (!device)
2809 		return ERR_NOMEM;
2810 	kref_init(&device->kref);
2811 
2812 	kref_get(&resource->kref);
2813 	device->resource = resource;
2814 	device->minor = minor;
2815 	device->vnr = vnr;
2816 
2817 	drbd_init_set_defaults(device);
2818 
2819 	q = blk_alloc_queue_node(GFP_KERNEL, NUMA_NO_NODE, &resource->req_lock);
2820 	if (!q)
2821 		goto out_no_q;
2822 	device->rq_queue = q;
2823 	q->queuedata   = device;
2824 
2825 	disk = alloc_disk(1);
2826 	if (!disk)
2827 		goto out_no_disk;
2828 	device->vdisk = disk;
2829 
2830 	set_disk_ro(disk, true);
2831 
2832 	disk->queue = q;
2833 	disk->major = DRBD_MAJOR;
2834 	disk->first_minor = minor;
2835 	disk->fops = &drbd_ops;
2836 	sprintf(disk->disk_name, "drbd%d", minor);
2837 	disk->private_data = device;
2838 
2839 	device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2840 	/* we have no partitions. we contain only ourselves. */
2841 	device->this_bdev->bd_contains = device->this_bdev;
2842 
2843 	q->backing_dev_info->congested_fn = drbd_congested;
2844 	q->backing_dev_info->congested_data = device;
2845 
2846 	blk_queue_make_request(q, drbd_make_request);
2847 	blk_queue_write_cache(q, true, true);
2848 	/* Setting the max_hw_sectors to an odd value of 8kibyte here
2849 	   This triggers a max_bio_size message upon first attach or connect */
2850 	blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2851 
2852 	device->md_io.page = alloc_page(GFP_KERNEL);
2853 	if (!device->md_io.page)
2854 		goto out_no_io_page;
2855 
2856 	if (drbd_bm_init(device))
2857 		goto out_no_bitmap;
2858 	device->read_requests = RB_ROOT;
2859 	device->write_requests = RB_ROOT;
2860 
2861 	id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2862 	if (id < 0) {
2863 		if (id == -ENOSPC)
2864 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2865 		goto out_no_minor_idr;
2866 	}
2867 	kref_get(&device->kref);
2868 
2869 	id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2870 	if (id < 0) {
2871 		if (id == -ENOSPC)
2872 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2873 		goto out_idr_remove_minor;
2874 	}
2875 	kref_get(&device->kref);
2876 
2877 	INIT_LIST_HEAD(&device->peer_devices);
2878 	INIT_LIST_HEAD(&device->pending_bitmap_io);
2879 	for_each_connection(connection, resource) {
2880 		peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2881 		if (!peer_device)
2882 			goto out_idr_remove_from_resource;
2883 		peer_device->connection = connection;
2884 		peer_device->device = device;
2885 
2886 		list_add(&peer_device->peer_devices, &device->peer_devices);
2887 		kref_get(&device->kref);
2888 
2889 		id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2890 		if (id < 0) {
2891 			if (id == -ENOSPC)
2892 				err = ERR_INVALID_REQUEST;
2893 			goto out_idr_remove_from_resource;
2894 		}
2895 		kref_get(&connection->kref);
2896 		INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2897 	}
2898 
2899 	if (init_submitter(device)) {
2900 		err = ERR_NOMEM;
2901 		goto out_idr_remove_vol;
2902 	}
2903 
2904 	add_disk(disk);
2905 
2906 	/* inherit the connection state */
2907 	device->state.conn = first_connection(resource)->cstate;
2908 	if (device->state.conn == C_WF_REPORT_PARAMS) {
2909 		for_each_peer_device(peer_device, device)
2910 			drbd_connected(peer_device);
2911 	}
2912 	/* move to create_peer_device() */
2913 	for_each_peer_device(peer_device, device)
2914 		drbd_debugfs_peer_device_add(peer_device);
2915 	drbd_debugfs_device_add(device);
2916 	return NO_ERROR;
2917 
2918 out_idr_remove_vol:
2919 	idr_remove(&connection->peer_devices, vnr);
2920 out_idr_remove_from_resource:
2921 	for_each_connection(connection, resource) {
2922 		peer_device = idr_remove(&connection->peer_devices, vnr);
2923 		if (peer_device)
2924 			kref_put(&connection->kref, drbd_destroy_connection);
2925 	}
2926 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2927 		list_del(&peer_device->peer_devices);
2928 		kfree(peer_device);
2929 	}
2930 	idr_remove(&resource->devices, vnr);
2931 out_idr_remove_minor:
2932 	idr_remove(&drbd_devices, minor);
2933 	synchronize_rcu();
2934 out_no_minor_idr:
2935 	drbd_bm_cleanup(device);
2936 out_no_bitmap:
2937 	__free_page(device->md_io.page);
2938 out_no_io_page:
2939 	put_disk(disk);
2940 out_no_disk:
2941 	blk_cleanup_queue(q);
2942 out_no_q:
2943 	kref_put(&resource->kref, drbd_destroy_resource);
2944 	kfree(device);
2945 	return err;
2946 }
2947 
2948 void drbd_delete_device(struct drbd_device *device)
2949 {
2950 	struct drbd_resource *resource = device->resource;
2951 	struct drbd_connection *connection;
2952 	struct drbd_peer_device *peer_device;
2953 
2954 	/* move to free_peer_device() */
2955 	for_each_peer_device(peer_device, device)
2956 		drbd_debugfs_peer_device_cleanup(peer_device);
2957 	drbd_debugfs_device_cleanup(device);
2958 	for_each_connection(connection, resource) {
2959 		idr_remove(&connection->peer_devices, device->vnr);
2960 		kref_put(&device->kref, drbd_destroy_device);
2961 	}
2962 	idr_remove(&resource->devices, device->vnr);
2963 	kref_put(&device->kref, drbd_destroy_device);
2964 	idr_remove(&drbd_devices, device_to_minor(device));
2965 	kref_put(&device->kref, drbd_destroy_device);
2966 	del_gendisk(device->vdisk);
2967 	synchronize_rcu();
2968 	kref_put(&device->kref, drbd_destroy_device);
2969 }
2970 
2971 static int __init drbd_init(void)
2972 {
2973 	int err;
2974 
2975 	if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2976 		pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2977 #ifdef MODULE
2978 		return -EINVAL;
2979 #else
2980 		drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2981 #endif
2982 	}
2983 
2984 	err = register_blkdev(DRBD_MAJOR, "drbd");
2985 	if (err) {
2986 		pr_err("unable to register block device major %d\n",
2987 		       DRBD_MAJOR);
2988 		return err;
2989 	}
2990 
2991 	/*
2992 	 * allocate all necessary structs
2993 	 */
2994 	init_waitqueue_head(&drbd_pp_wait);
2995 
2996 	drbd_proc = NULL; /* play safe for drbd_cleanup */
2997 	idr_init(&drbd_devices);
2998 
2999 	mutex_init(&resources_mutex);
3000 	INIT_LIST_HEAD(&drbd_resources);
3001 
3002 	err = drbd_genl_register();
3003 	if (err) {
3004 		pr_err("unable to register generic netlink family\n");
3005 		goto fail;
3006 	}
3007 
3008 	err = drbd_create_mempools();
3009 	if (err)
3010 		goto fail;
3011 
3012 	err = -ENOMEM;
3013 	drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
3014 	if (!drbd_proc)	{
3015 		pr_err("unable to register proc file\n");
3016 		goto fail;
3017 	}
3018 
3019 	retry.wq = create_singlethread_workqueue("drbd-reissue");
3020 	if (!retry.wq) {
3021 		pr_err("unable to create retry workqueue\n");
3022 		goto fail;
3023 	}
3024 	INIT_WORK(&retry.worker, do_retry);
3025 	spin_lock_init(&retry.lock);
3026 	INIT_LIST_HEAD(&retry.writes);
3027 
3028 	if (drbd_debugfs_init())
3029 		pr_notice("failed to initialize debugfs -- will not be available\n");
3030 
3031 	pr_info("initialized. "
3032 	       "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
3033 	       API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
3034 	pr_info("%s\n", drbd_buildtag());
3035 	pr_info("registered as block device major %d\n", DRBD_MAJOR);
3036 	return 0; /* Success! */
3037 
3038 fail:
3039 	drbd_cleanup();
3040 	if (err == -ENOMEM)
3041 		pr_err("ran out of memory\n");
3042 	else
3043 		pr_err("initialization failure\n");
3044 	return err;
3045 }
3046 
3047 static void drbd_free_one_sock(struct drbd_socket *ds)
3048 {
3049 	struct socket *s;
3050 	mutex_lock(&ds->mutex);
3051 	s = ds->socket;
3052 	ds->socket = NULL;
3053 	mutex_unlock(&ds->mutex);
3054 	if (s) {
3055 		/* so debugfs does not need to mutex_lock() */
3056 		synchronize_rcu();
3057 		kernel_sock_shutdown(s, SHUT_RDWR);
3058 		sock_release(s);
3059 	}
3060 }
3061 
3062 void drbd_free_sock(struct drbd_connection *connection)
3063 {
3064 	if (connection->data.socket)
3065 		drbd_free_one_sock(&connection->data);
3066 	if (connection->meta.socket)
3067 		drbd_free_one_sock(&connection->meta);
3068 }
3069 
3070 /* meta data management */
3071 
3072 void conn_md_sync(struct drbd_connection *connection)
3073 {
3074 	struct drbd_peer_device *peer_device;
3075 	int vnr;
3076 
3077 	rcu_read_lock();
3078 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3079 		struct drbd_device *device = peer_device->device;
3080 
3081 		kref_get(&device->kref);
3082 		rcu_read_unlock();
3083 		drbd_md_sync(device);
3084 		kref_put(&device->kref, drbd_destroy_device);
3085 		rcu_read_lock();
3086 	}
3087 	rcu_read_unlock();
3088 }
3089 
3090 /* aligned 4kByte */
3091 struct meta_data_on_disk {
3092 	u64 la_size_sect;      /* last agreed size. */
3093 	u64 uuid[UI_SIZE];   /* UUIDs. */
3094 	u64 device_uuid;
3095 	u64 reserved_u64_1;
3096 	u32 flags;             /* MDF */
3097 	u32 magic;
3098 	u32 md_size_sect;
3099 	u32 al_offset;         /* offset to this block */
3100 	u32 al_nr_extents;     /* important for restoring the AL (userspace) */
3101 	      /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3102 	u32 bm_offset;         /* offset to the bitmap, from here */
3103 	u32 bm_bytes_per_bit;  /* BM_BLOCK_SIZE */
3104 	u32 la_peer_max_bio_size;   /* last peer max_bio_size */
3105 
3106 	/* see al_tr_number_to_on_disk_sector() */
3107 	u32 al_stripes;
3108 	u32 al_stripe_size_4k;
3109 
3110 	u8 reserved_u8[4096 - (7*8 + 10*4)];
3111 } __packed;
3112 
3113 
3114 
3115 void drbd_md_write(struct drbd_device *device, void *b)
3116 {
3117 	struct meta_data_on_disk *buffer = b;
3118 	sector_t sector;
3119 	int i;
3120 
3121 	memset(buffer, 0, sizeof(*buffer));
3122 
3123 	buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3124 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3125 		buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3126 	buffer->flags = cpu_to_be32(device->ldev->md.flags);
3127 	buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3128 
3129 	buffer->md_size_sect  = cpu_to_be32(device->ldev->md.md_size_sect);
3130 	buffer->al_offset     = cpu_to_be32(device->ldev->md.al_offset);
3131 	buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3132 	buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3133 	buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3134 
3135 	buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3136 	buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3137 
3138 	buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3139 	buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3140 
3141 	D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3142 	sector = device->ldev->md.md_offset;
3143 
3144 	if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3145 		/* this was a try anyways ... */
3146 		drbd_err(device, "meta data update failed!\n");
3147 		drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3148 	}
3149 }
3150 
3151 /**
3152  * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3153  * @device:	DRBD device.
3154  */
3155 void drbd_md_sync(struct drbd_device *device)
3156 {
3157 	struct meta_data_on_disk *buffer;
3158 
3159 	/* Don't accidentally change the DRBD meta data layout. */
3160 	BUILD_BUG_ON(UI_SIZE != 4);
3161 	BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3162 
3163 	del_timer(&device->md_sync_timer);
3164 	/* timer may be rearmed by drbd_md_mark_dirty() now. */
3165 	if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3166 		return;
3167 
3168 	/* We use here D_FAILED and not D_ATTACHING because we try to write
3169 	 * metadata even if we detach due to a disk failure! */
3170 	if (!get_ldev_if_state(device, D_FAILED))
3171 		return;
3172 
3173 	buffer = drbd_md_get_buffer(device, __func__);
3174 	if (!buffer)
3175 		goto out;
3176 
3177 	drbd_md_write(device, buffer);
3178 
3179 	/* Update device->ldev->md.la_size_sect,
3180 	 * since we updated it on metadata. */
3181 	device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3182 
3183 	drbd_md_put_buffer(device);
3184 out:
3185 	put_ldev(device);
3186 }
3187 
3188 static int check_activity_log_stripe_size(struct drbd_device *device,
3189 		struct meta_data_on_disk *on_disk,
3190 		struct drbd_md *in_core)
3191 {
3192 	u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3193 	u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3194 	u64 al_size_4k;
3195 
3196 	/* both not set: default to old fixed size activity log */
3197 	if (al_stripes == 0 && al_stripe_size_4k == 0) {
3198 		al_stripes = 1;
3199 		al_stripe_size_4k = MD_32kB_SECT/8;
3200 	}
3201 
3202 	/* some paranoia plausibility checks */
3203 
3204 	/* we need both values to be set */
3205 	if (al_stripes == 0 || al_stripe_size_4k == 0)
3206 		goto err;
3207 
3208 	al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3209 
3210 	/* Upper limit of activity log area, to avoid potential overflow
3211 	 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3212 	 * than 72 * 4k blocks total only increases the amount of history,
3213 	 * limiting this arbitrarily to 16 GB is not a real limitation ;-)  */
3214 	if (al_size_4k > (16 * 1024 * 1024/4))
3215 		goto err;
3216 
3217 	/* Lower limit: we need at least 8 transaction slots (32kB)
3218 	 * to not break existing setups */
3219 	if (al_size_4k < MD_32kB_SECT/8)
3220 		goto err;
3221 
3222 	in_core->al_stripe_size_4k = al_stripe_size_4k;
3223 	in_core->al_stripes = al_stripes;
3224 	in_core->al_size_4k = al_size_4k;
3225 
3226 	return 0;
3227 err:
3228 	drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3229 			al_stripes, al_stripe_size_4k);
3230 	return -EINVAL;
3231 }
3232 
3233 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3234 {
3235 	sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3236 	struct drbd_md *in_core = &bdev->md;
3237 	s32 on_disk_al_sect;
3238 	s32 on_disk_bm_sect;
3239 
3240 	/* The on-disk size of the activity log, calculated from offsets, and
3241 	 * the size of the activity log calculated from the stripe settings,
3242 	 * should match.
3243 	 * Though we could relax this a bit: it is ok, if the striped activity log
3244 	 * fits in the available on-disk activity log size.
3245 	 * Right now, that would break how resize is implemented.
3246 	 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3247 	 * of possible unused padding space in the on disk layout. */
3248 	if (in_core->al_offset < 0) {
3249 		if (in_core->bm_offset > in_core->al_offset)
3250 			goto err;
3251 		on_disk_al_sect = -in_core->al_offset;
3252 		on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3253 	} else {
3254 		if (in_core->al_offset != MD_4kB_SECT)
3255 			goto err;
3256 		if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3257 			goto err;
3258 
3259 		on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3260 		on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3261 	}
3262 
3263 	/* old fixed size meta data is exactly that: fixed. */
3264 	if (in_core->meta_dev_idx >= 0) {
3265 		if (in_core->md_size_sect != MD_128MB_SECT
3266 		||  in_core->al_offset != MD_4kB_SECT
3267 		||  in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3268 		||  in_core->al_stripes != 1
3269 		||  in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3270 			goto err;
3271 	}
3272 
3273 	if (capacity < in_core->md_size_sect)
3274 		goto err;
3275 	if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3276 		goto err;
3277 
3278 	/* should be aligned, and at least 32k */
3279 	if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3280 		goto err;
3281 
3282 	/* should fit (for now: exactly) into the available on-disk space;
3283 	 * overflow prevention is in check_activity_log_stripe_size() above. */
3284 	if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3285 		goto err;
3286 
3287 	/* again, should be aligned */
3288 	if (in_core->bm_offset & 7)
3289 		goto err;
3290 
3291 	/* FIXME check for device grow with flex external meta data? */
3292 
3293 	/* can the available bitmap space cover the last agreed device size? */
3294 	if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3295 		goto err;
3296 
3297 	return 0;
3298 
3299 err:
3300 	drbd_err(device, "meta data offsets don't make sense: idx=%d "
3301 			"al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3302 			"md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3303 			in_core->meta_dev_idx,
3304 			in_core->al_stripes, in_core->al_stripe_size_4k,
3305 			in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3306 			(unsigned long long)in_core->la_size_sect,
3307 			(unsigned long long)capacity);
3308 
3309 	return -EINVAL;
3310 }
3311 
3312 
3313 /**
3314  * drbd_md_read() - Reads in the meta data super block
3315  * @device:	DRBD device.
3316  * @bdev:	Device from which the meta data should be read in.
3317  *
3318  * Return NO_ERROR on success, and an enum drbd_ret_code in case
3319  * something goes wrong.
3320  *
3321  * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3322  * even before @bdev is assigned to @device->ldev.
3323  */
3324 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3325 {
3326 	struct meta_data_on_disk *buffer;
3327 	u32 magic, flags;
3328 	int i, rv = NO_ERROR;
3329 
3330 	if (device->state.disk != D_DISKLESS)
3331 		return ERR_DISK_CONFIGURED;
3332 
3333 	buffer = drbd_md_get_buffer(device, __func__);
3334 	if (!buffer)
3335 		return ERR_NOMEM;
3336 
3337 	/* First, figure out where our meta data superblock is located,
3338 	 * and read it. */
3339 	bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3340 	bdev->md.md_offset = drbd_md_ss(bdev);
3341 	/* Even for (flexible or indexed) external meta data,
3342 	 * initially restrict us to the 4k superblock for now.
3343 	 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3344 	bdev->md.md_size_sect = 8;
3345 
3346 	if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3347 				 REQ_OP_READ)) {
3348 		/* NOTE: can't do normal error processing here as this is
3349 		   called BEFORE disk is attached */
3350 		drbd_err(device, "Error while reading metadata.\n");
3351 		rv = ERR_IO_MD_DISK;
3352 		goto err;
3353 	}
3354 
3355 	magic = be32_to_cpu(buffer->magic);
3356 	flags = be32_to_cpu(buffer->flags);
3357 	if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3358 	    (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3359 			/* btw: that's Activity Log clean, not "all" clean. */
3360 		drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3361 		rv = ERR_MD_UNCLEAN;
3362 		goto err;
3363 	}
3364 
3365 	rv = ERR_MD_INVALID;
3366 	if (magic != DRBD_MD_MAGIC_08) {
3367 		if (magic == DRBD_MD_MAGIC_07)
3368 			drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3369 		else
3370 			drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3371 		goto err;
3372 	}
3373 
3374 	if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3375 		drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3376 		    be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3377 		goto err;
3378 	}
3379 
3380 
3381 	/* convert to in_core endian */
3382 	bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3383 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3384 		bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3385 	bdev->md.flags = be32_to_cpu(buffer->flags);
3386 	bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3387 
3388 	bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3389 	bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3390 	bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3391 
3392 	if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3393 		goto err;
3394 	if (check_offsets_and_sizes(device, bdev))
3395 		goto err;
3396 
3397 	if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3398 		drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3399 		    be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3400 		goto err;
3401 	}
3402 	if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3403 		drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3404 		    be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3405 		goto err;
3406 	}
3407 
3408 	rv = NO_ERROR;
3409 
3410 	spin_lock_irq(&device->resource->req_lock);
3411 	if (device->state.conn < C_CONNECTED) {
3412 		unsigned int peer;
3413 		peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3414 		peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3415 		device->peer_max_bio_size = peer;
3416 	}
3417 	spin_unlock_irq(&device->resource->req_lock);
3418 
3419  err:
3420 	drbd_md_put_buffer(device);
3421 
3422 	return rv;
3423 }
3424 
3425 /**
3426  * drbd_md_mark_dirty() - Mark meta data super block as dirty
3427  * @device:	DRBD device.
3428  *
3429  * Call this function if you change anything that should be written to
3430  * the meta-data super block. This function sets MD_DIRTY, and starts a
3431  * timer that ensures that within five seconds you have to call drbd_md_sync().
3432  */
3433 #ifdef DEBUG
3434 void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
3435 {
3436 	if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3437 		mod_timer(&device->md_sync_timer, jiffies + HZ);
3438 		device->last_md_mark_dirty.line = line;
3439 		device->last_md_mark_dirty.func = func;
3440 	}
3441 }
3442 #else
3443 void drbd_md_mark_dirty(struct drbd_device *device)
3444 {
3445 	if (!test_and_set_bit(MD_DIRTY, &device->flags))
3446 		mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3447 }
3448 #endif
3449 
3450 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3451 {
3452 	int i;
3453 
3454 	for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3455 		device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3456 }
3457 
3458 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3459 {
3460 	if (idx == UI_CURRENT) {
3461 		if (device->state.role == R_PRIMARY)
3462 			val |= 1;
3463 		else
3464 			val &= ~((u64)1);
3465 
3466 		drbd_set_ed_uuid(device, val);
3467 	}
3468 
3469 	device->ldev->md.uuid[idx] = val;
3470 	drbd_md_mark_dirty(device);
3471 }
3472 
3473 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3474 {
3475 	unsigned long flags;
3476 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3477 	__drbd_uuid_set(device, idx, val);
3478 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3479 }
3480 
3481 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3482 {
3483 	unsigned long flags;
3484 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3485 	if (device->ldev->md.uuid[idx]) {
3486 		drbd_uuid_move_history(device);
3487 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3488 	}
3489 	__drbd_uuid_set(device, idx, val);
3490 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3491 }
3492 
3493 /**
3494  * drbd_uuid_new_current() - Creates a new current UUID
3495  * @device:	DRBD device.
3496  *
3497  * Creates a new current UUID, and rotates the old current UUID into
3498  * the bitmap slot. Causes an incremental resync upon next connect.
3499  */
3500 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3501 {
3502 	u64 val;
3503 	unsigned long long bm_uuid;
3504 
3505 	get_random_bytes(&val, sizeof(u64));
3506 
3507 	spin_lock_irq(&device->ldev->md.uuid_lock);
3508 	bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3509 
3510 	if (bm_uuid)
3511 		drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3512 
3513 	device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3514 	__drbd_uuid_set(device, UI_CURRENT, val);
3515 	spin_unlock_irq(&device->ldev->md.uuid_lock);
3516 
3517 	drbd_print_uuids(device, "new current UUID");
3518 	/* get it to stable storage _now_ */
3519 	drbd_md_sync(device);
3520 }
3521 
3522 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3523 {
3524 	unsigned long flags;
3525 	if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3526 		return;
3527 
3528 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3529 	if (val == 0) {
3530 		drbd_uuid_move_history(device);
3531 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3532 		device->ldev->md.uuid[UI_BITMAP] = 0;
3533 	} else {
3534 		unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3535 		if (bm_uuid)
3536 			drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3537 
3538 		device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3539 	}
3540 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3541 
3542 	drbd_md_mark_dirty(device);
3543 }
3544 
3545 /**
3546  * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3547  * @device:	DRBD device.
3548  *
3549  * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3550  */
3551 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3552 {
3553 	int rv = -EIO;
3554 
3555 	drbd_md_set_flag(device, MDF_FULL_SYNC);
3556 	drbd_md_sync(device);
3557 	drbd_bm_set_all(device);
3558 
3559 	rv = drbd_bm_write(device);
3560 
3561 	if (!rv) {
3562 		drbd_md_clear_flag(device, MDF_FULL_SYNC);
3563 		drbd_md_sync(device);
3564 	}
3565 
3566 	return rv;
3567 }
3568 
3569 /**
3570  * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3571  * @device:	DRBD device.
3572  *
3573  * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3574  */
3575 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3576 {
3577 	drbd_resume_al(device);
3578 	drbd_bm_clear_all(device);
3579 	return drbd_bm_write(device);
3580 }
3581 
3582 static int w_bitmap_io(struct drbd_work *w, int unused)
3583 {
3584 	struct drbd_device *device =
3585 		container_of(w, struct drbd_device, bm_io_work.w);
3586 	struct bm_io_work *work = &device->bm_io_work;
3587 	int rv = -EIO;
3588 
3589 	if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3590 		int cnt = atomic_read(&device->ap_bio_cnt);
3591 		if (cnt)
3592 			drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3593 					cnt, work->why);
3594 	}
3595 
3596 	if (get_ldev(device)) {
3597 		drbd_bm_lock(device, work->why, work->flags);
3598 		rv = work->io_fn(device);
3599 		drbd_bm_unlock(device);
3600 		put_ldev(device);
3601 	}
3602 
3603 	clear_bit_unlock(BITMAP_IO, &device->flags);
3604 	wake_up(&device->misc_wait);
3605 
3606 	if (work->done)
3607 		work->done(device, rv);
3608 
3609 	clear_bit(BITMAP_IO_QUEUED, &device->flags);
3610 	work->why = NULL;
3611 	work->flags = 0;
3612 
3613 	return 0;
3614 }
3615 
3616 /**
3617  * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3618  * @device:	DRBD device.
3619  * @io_fn:	IO callback to be called when bitmap IO is possible
3620  * @done:	callback to be called after the bitmap IO was performed
3621  * @why:	Descriptive text of the reason for doing the IO
3622  *
3623  * While IO on the bitmap happens we freeze application IO thus we ensure
3624  * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3625  * called from worker context. It MUST NOT be used while a previous such
3626  * work is still pending!
3627  *
3628  * Its worker function encloses the call of io_fn() by get_ldev() and
3629  * put_ldev().
3630  */
3631 void drbd_queue_bitmap_io(struct drbd_device *device,
3632 			  int (*io_fn)(struct drbd_device *),
3633 			  void (*done)(struct drbd_device *, int),
3634 			  char *why, enum bm_flag flags)
3635 {
3636 	D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3637 
3638 	D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3639 	D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3640 	D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3641 	if (device->bm_io_work.why)
3642 		drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3643 			why, device->bm_io_work.why);
3644 
3645 	device->bm_io_work.io_fn = io_fn;
3646 	device->bm_io_work.done = done;
3647 	device->bm_io_work.why = why;
3648 	device->bm_io_work.flags = flags;
3649 
3650 	spin_lock_irq(&device->resource->req_lock);
3651 	set_bit(BITMAP_IO, &device->flags);
3652 	/* don't wait for pending application IO if the caller indicates that
3653 	 * application IO does not conflict anyways. */
3654 	if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3655 		if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3656 			drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3657 					&device->bm_io_work.w);
3658 	}
3659 	spin_unlock_irq(&device->resource->req_lock);
3660 }
3661 
3662 /**
3663  * drbd_bitmap_io() -  Does an IO operation on the whole bitmap
3664  * @device:	DRBD device.
3665  * @io_fn:	IO callback to be called when bitmap IO is possible
3666  * @why:	Descriptive text of the reason for doing the IO
3667  *
3668  * freezes application IO while that the actual IO operations runs. This
3669  * functions MAY NOT be called from worker context.
3670  */
3671 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3672 		char *why, enum bm_flag flags)
3673 {
3674 	/* Only suspend io, if some operation is supposed to be locked out */
3675 	const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3676 	int rv;
3677 
3678 	D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3679 
3680 	if (do_suspend_io)
3681 		drbd_suspend_io(device);
3682 
3683 	drbd_bm_lock(device, why, flags);
3684 	rv = io_fn(device);
3685 	drbd_bm_unlock(device);
3686 
3687 	if (do_suspend_io)
3688 		drbd_resume_io(device);
3689 
3690 	return rv;
3691 }
3692 
3693 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3694 {
3695 	if ((device->ldev->md.flags & flag) != flag) {
3696 		drbd_md_mark_dirty(device);
3697 		device->ldev->md.flags |= flag;
3698 	}
3699 }
3700 
3701 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3702 {
3703 	if ((device->ldev->md.flags & flag) != 0) {
3704 		drbd_md_mark_dirty(device);
3705 		device->ldev->md.flags &= ~flag;
3706 	}
3707 }
3708 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3709 {
3710 	return (bdev->md.flags & flag) != 0;
3711 }
3712 
3713 static void md_sync_timer_fn(struct timer_list *t)
3714 {
3715 	struct drbd_device *device = from_timer(device, t, md_sync_timer);
3716 	drbd_device_post_work(device, MD_SYNC);
3717 }
3718 
3719 const char *cmdname(enum drbd_packet cmd)
3720 {
3721 	/* THINK may need to become several global tables
3722 	 * when we want to support more than
3723 	 * one PRO_VERSION */
3724 	static const char *cmdnames[] = {
3725 		[P_DATA]	        = "Data",
3726 		[P_WSAME]	        = "WriteSame",
3727 		[P_TRIM]	        = "Trim",
3728 		[P_DATA_REPLY]	        = "DataReply",
3729 		[P_RS_DATA_REPLY]	= "RSDataReply",
3730 		[P_BARRIER]	        = "Barrier",
3731 		[P_BITMAP]	        = "ReportBitMap",
3732 		[P_BECOME_SYNC_TARGET]  = "BecomeSyncTarget",
3733 		[P_BECOME_SYNC_SOURCE]  = "BecomeSyncSource",
3734 		[P_UNPLUG_REMOTE]	= "UnplugRemote",
3735 		[P_DATA_REQUEST]	= "DataRequest",
3736 		[P_RS_DATA_REQUEST]     = "RSDataRequest",
3737 		[P_SYNC_PARAM]	        = "SyncParam",
3738 		[P_SYNC_PARAM89]	= "SyncParam89",
3739 		[P_PROTOCOL]            = "ReportProtocol",
3740 		[P_UUIDS]	        = "ReportUUIDs",
3741 		[P_SIZES]	        = "ReportSizes",
3742 		[P_STATE]	        = "ReportState",
3743 		[P_SYNC_UUID]           = "ReportSyncUUID",
3744 		[P_AUTH_CHALLENGE]      = "AuthChallenge",
3745 		[P_AUTH_RESPONSE]	= "AuthResponse",
3746 		[P_PING]		= "Ping",
3747 		[P_PING_ACK]	        = "PingAck",
3748 		[P_RECV_ACK]	        = "RecvAck",
3749 		[P_WRITE_ACK]	        = "WriteAck",
3750 		[P_RS_WRITE_ACK]	= "RSWriteAck",
3751 		[P_SUPERSEDED]          = "Superseded",
3752 		[P_NEG_ACK]	        = "NegAck",
3753 		[P_NEG_DREPLY]	        = "NegDReply",
3754 		[P_NEG_RS_DREPLY]	= "NegRSDReply",
3755 		[P_BARRIER_ACK]	        = "BarrierAck",
3756 		[P_STATE_CHG_REQ]       = "StateChgRequest",
3757 		[P_STATE_CHG_REPLY]     = "StateChgReply",
3758 		[P_OV_REQUEST]          = "OVRequest",
3759 		[P_OV_REPLY]            = "OVReply",
3760 		[P_OV_RESULT]           = "OVResult",
3761 		[P_CSUM_RS_REQUEST]     = "CsumRSRequest",
3762 		[P_RS_IS_IN_SYNC]	= "CsumRSIsInSync",
3763 		[P_COMPRESSED_BITMAP]   = "CBitmap",
3764 		[P_DELAY_PROBE]         = "DelayProbe",
3765 		[P_OUT_OF_SYNC]		= "OutOfSync",
3766 		[P_RETRY_WRITE]		= "RetryWrite",
3767 		[P_RS_CANCEL]		= "RSCancel",
3768 		[P_CONN_ST_CHG_REQ]	= "conn_st_chg_req",
3769 		[P_CONN_ST_CHG_REPLY]	= "conn_st_chg_reply",
3770 		[P_RETRY_WRITE]		= "retry_write",
3771 		[P_PROTOCOL_UPDATE]	= "protocol_update",
3772 		[P_RS_THIN_REQ]         = "rs_thin_req",
3773 		[P_RS_DEALLOCATED]      = "rs_deallocated",
3774 
3775 		/* enum drbd_packet, but not commands - obsoleted flags:
3776 		 *	P_MAY_IGNORE
3777 		 *	P_MAX_OPT_CMD
3778 		 */
3779 	};
3780 
3781 	/* too big for the array: 0xfffX */
3782 	if (cmd == P_INITIAL_META)
3783 		return "InitialMeta";
3784 	if (cmd == P_INITIAL_DATA)
3785 		return "InitialData";
3786 	if (cmd == P_CONNECTION_FEATURES)
3787 		return "ConnectionFeatures";
3788 	if (cmd >= ARRAY_SIZE(cmdnames))
3789 		return "Unknown";
3790 	return cmdnames[cmd];
3791 }
3792 
3793 /**
3794  * drbd_wait_misc  -  wait for a request to make progress
3795  * @device:	device associated with the request
3796  * @i:		the struct drbd_interval embedded in struct drbd_request or
3797  *		struct drbd_peer_request
3798  */
3799 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3800 {
3801 	struct net_conf *nc;
3802 	DEFINE_WAIT(wait);
3803 	long timeout;
3804 
3805 	rcu_read_lock();
3806 	nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3807 	if (!nc) {
3808 		rcu_read_unlock();
3809 		return -ETIMEDOUT;
3810 	}
3811 	timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3812 	rcu_read_unlock();
3813 
3814 	/* Indicate to wake up device->misc_wait on progress.  */
3815 	i->waiting = true;
3816 	prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3817 	spin_unlock_irq(&device->resource->req_lock);
3818 	timeout = schedule_timeout(timeout);
3819 	finish_wait(&device->misc_wait, &wait);
3820 	spin_lock_irq(&device->resource->req_lock);
3821 	if (!timeout || device->state.conn < C_CONNECTED)
3822 		return -ETIMEDOUT;
3823 	if (signal_pending(current))
3824 		return -ERESTARTSYS;
3825 	return 0;
3826 }
3827 
3828 void lock_all_resources(void)
3829 {
3830 	struct drbd_resource *resource;
3831 	int __maybe_unused i = 0;
3832 
3833 	mutex_lock(&resources_mutex);
3834 	local_irq_disable();
3835 	for_each_resource(resource, &drbd_resources)
3836 		spin_lock_nested(&resource->req_lock, i++);
3837 }
3838 
3839 void unlock_all_resources(void)
3840 {
3841 	struct drbd_resource *resource;
3842 
3843 	for_each_resource(resource, &drbd_resources)
3844 		spin_unlock(&resource->req_lock);
3845 	local_irq_enable();
3846 	mutex_unlock(&resources_mutex);
3847 }
3848 
3849 #ifdef CONFIG_DRBD_FAULT_INJECTION
3850 /* Fault insertion support including random number generator shamelessly
3851  * stolen from kernel/rcutorture.c */
3852 struct fault_random_state {
3853 	unsigned long state;
3854 	unsigned long count;
3855 };
3856 
3857 #define FAULT_RANDOM_MULT 39916801  /* prime */
3858 #define FAULT_RANDOM_ADD	479001701 /* prime */
3859 #define FAULT_RANDOM_REFRESH 10000
3860 
3861 /*
3862  * Crude but fast random-number generator.  Uses a linear congruential
3863  * generator, with occasional help from get_random_bytes().
3864  */
3865 static unsigned long
3866 _drbd_fault_random(struct fault_random_state *rsp)
3867 {
3868 	long refresh;
3869 
3870 	if (!rsp->count--) {
3871 		get_random_bytes(&refresh, sizeof(refresh));
3872 		rsp->state += refresh;
3873 		rsp->count = FAULT_RANDOM_REFRESH;
3874 	}
3875 	rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3876 	return swahw32(rsp->state);
3877 }
3878 
3879 static char *
3880 _drbd_fault_str(unsigned int type) {
3881 	static char *_faults[] = {
3882 		[DRBD_FAULT_MD_WR] = "Meta-data write",
3883 		[DRBD_FAULT_MD_RD] = "Meta-data read",
3884 		[DRBD_FAULT_RS_WR] = "Resync write",
3885 		[DRBD_FAULT_RS_RD] = "Resync read",
3886 		[DRBD_FAULT_DT_WR] = "Data write",
3887 		[DRBD_FAULT_DT_RD] = "Data read",
3888 		[DRBD_FAULT_DT_RA] = "Data read ahead",
3889 		[DRBD_FAULT_BM_ALLOC] = "BM allocation",
3890 		[DRBD_FAULT_AL_EE] = "EE allocation",
3891 		[DRBD_FAULT_RECEIVE] = "receive data corruption",
3892 	};
3893 
3894 	return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3895 }
3896 
3897 unsigned int
3898 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3899 {
3900 	static struct fault_random_state rrs = {0, 0};
3901 
3902 	unsigned int ret = (
3903 		(drbd_fault_devs == 0 ||
3904 			((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3905 		(((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3906 
3907 	if (ret) {
3908 		drbd_fault_count++;
3909 
3910 		if (__ratelimit(&drbd_ratelimit_state))
3911 			drbd_warn(device, "***Simulating %s failure\n",
3912 				_drbd_fault_str(type));
3913 	}
3914 
3915 	return ret;
3916 }
3917 #endif
3918 
3919 const char *drbd_buildtag(void)
3920 {
3921 	/* DRBD built from external sources has here a reference to the
3922 	   git hash of the source code. */
3923 
3924 	static char buildtag[38] = "\0uilt-in";
3925 
3926 	if (buildtag[0] == 0) {
3927 #ifdef MODULE
3928 		sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3929 #else
3930 		buildtag[0] = 'b';
3931 #endif
3932 	}
3933 
3934 	return buildtag;
3935 }
3936 
3937 module_init(drbd_init)
3938 module_exit(drbd_cleanup)
3939 
3940 EXPORT_SYMBOL(drbd_conn_str);
3941 EXPORT_SYMBOL(drbd_role_str);
3942 EXPORT_SYMBOL(drbd_disk_str);
3943 EXPORT_SYMBOL(drbd_set_st_err_str);
3944