xref: /linux/drivers/block/drbd/drbd_main.c (revision 93d90ad708b8da6efc0e487b66111aa9db7f70c7)
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 <asm/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 
56 #include <linux/drbd_limits.h>
57 #include "drbd_int.h"
58 #include "drbd_protocol.h"
59 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
60 #include "drbd_vli.h"
61 #include "drbd_debugfs.h"
62 
63 static DEFINE_MUTEX(drbd_main_mutex);
64 static int drbd_open(struct block_device *bdev, fmode_t mode);
65 static void drbd_release(struct gendisk *gd, fmode_t mode);
66 static void md_sync_timer_fn(unsigned long data);
67 static int w_bitmap_io(struct drbd_work *w, int unused);
68 
69 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
70 	      "Lars Ellenberg <lars@linbit.com>");
71 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
72 MODULE_VERSION(REL_VERSION);
73 MODULE_LICENSE("GPL");
74 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
75 		 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
76 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
77 
78 #include <linux/moduleparam.h>
79 /* allow_open_on_secondary */
80 MODULE_PARM_DESC(allow_oos, "DONT USE!");
81 /* thanks to these macros, if compiled into the kernel (not-module),
82  * this becomes the boot parameter drbd.minor_count */
83 module_param(minor_count, uint, 0444);
84 module_param(disable_sendpage, bool, 0644);
85 module_param(allow_oos, bool, 0);
86 module_param(proc_details, int, 0644);
87 
88 #ifdef CONFIG_DRBD_FAULT_INJECTION
89 int enable_faults;
90 int fault_rate;
91 static int fault_count;
92 int fault_devs;
93 /* bitmap of enabled faults */
94 module_param(enable_faults, int, 0664);
95 /* fault rate % value - applies to all enabled faults */
96 module_param(fault_rate, int, 0664);
97 /* count of faults inserted */
98 module_param(fault_count, int, 0664);
99 /* bitmap of devices to insert faults on */
100 module_param(fault_devs, int, 0644);
101 #endif
102 
103 /* module parameter, defined */
104 unsigned int minor_count = DRBD_MINOR_COUNT_DEF;
105 bool disable_sendpage;
106 bool allow_oos;
107 int proc_details;       /* Detail level in proc drbd*/
108 
109 /* Module parameter for setting the user mode helper program
110  * to run. Default is /sbin/drbdadm */
111 char usermode_helper[80] = "/sbin/drbdadm";
112 
113 module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644);
114 
115 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
116  * as member "struct gendisk *vdisk;"
117  */
118 struct idr drbd_devices;
119 struct list_head drbd_resources;
120 
121 struct kmem_cache *drbd_request_cache;
122 struct kmem_cache *drbd_ee_cache;	/* peer requests */
123 struct kmem_cache *drbd_bm_ext_cache;	/* bitmap extents */
124 struct kmem_cache *drbd_al_ext_cache;	/* activity log extents */
125 mempool_t *drbd_request_mempool;
126 mempool_t *drbd_ee_mempool;
127 mempool_t *drbd_md_io_page_pool;
128 struct bio_set *drbd_md_io_bio_set;
129 
130 /* I do not use a standard mempool, because:
131    1) I want to hand out the pre-allocated objects first.
132    2) I want to be able to interrupt sleeping allocation with a signal.
133    Note: This is a single linked list, the next pointer is the private
134 	 member of struct page.
135  */
136 struct page *drbd_pp_pool;
137 spinlock_t   drbd_pp_lock;
138 int          drbd_pp_vacant;
139 wait_queue_head_t drbd_pp_wait;
140 
141 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
142 
143 static const struct block_device_operations drbd_ops = {
144 	.owner =   THIS_MODULE,
145 	.open =    drbd_open,
146 	.release = drbd_release,
147 };
148 
149 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
150 {
151 	struct bio *bio;
152 
153 	if (!drbd_md_io_bio_set)
154 		return bio_alloc(gfp_mask, 1);
155 
156 	bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
157 	if (!bio)
158 		return NULL;
159 	return bio;
160 }
161 
162 #ifdef __CHECKER__
163 /* When checking with sparse, and this is an inline function, sparse will
164    give tons of false positives. When this is a real functions sparse works.
165  */
166 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
167 {
168 	int io_allowed;
169 
170 	atomic_inc(&device->local_cnt);
171 	io_allowed = (device->state.disk >= mins);
172 	if (!io_allowed) {
173 		if (atomic_dec_and_test(&device->local_cnt))
174 			wake_up(&device->misc_wait);
175 	}
176 	return io_allowed;
177 }
178 
179 #endif
180 
181 /**
182  * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
183  * @connection:	DRBD connection.
184  * @barrier_nr:	Expected identifier of the DRBD write barrier packet.
185  * @set_size:	Expected number of requests before that barrier.
186  *
187  * In case the passed barrier_nr or set_size does not match the oldest
188  * epoch of not yet barrier-acked requests, this function will cause a
189  * termination of the connection.
190  */
191 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
192 		unsigned int set_size)
193 {
194 	struct drbd_request *r;
195 	struct drbd_request *req = NULL;
196 	int expect_epoch = 0;
197 	int expect_size = 0;
198 
199 	spin_lock_irq(&connection->resource->req_lock);
200 
201 	/* find oldest not yet barrier-acked write request,
202 	 * count writes in its epoch. */
203 	list_for_each_entry(r, &connection->transfer_log, tl_requests) {
204 		const unsigned s = r->rq_state;
205 		if (!req) {
206 			if (!(s & RQ_WRITE))
207 				continue;
208 			if (!(s & RQ_NET_MASK))
209 				continue;
210 			if (s & RQ_NET_DONE)
211 				continue;
212 			req = r;
213 			expect_epoch = req->epoch;
214 			expect_size ++;
215 		} else {
216 			if (r->epoch != expect_epoch)
217 				break;
218 			if (!(s & RQ_WRITE))
219 				continue;
220 			/* if (s & RQ_DONE): not expected */
221 			/* if (!(s & RQ_NET_MASK)): not expected */
222 			expect_size++;
223 		}
224 	}
225 
226 	/* first some paranoia code */
227 	if (req == NULL) {
228 		drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
229 			 barrier_nr);
230 		goto bail;
231 	}
232 	if (expect_epoch != barrier_nr) {
233 		drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
234 			 barrier_nr, expect_epoch);
235 		goto bail;
236 	}
237 
238 	if (expect_size != set_size) {
239 		drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
240 			 barrier_nr, set_size, expect_size);
241 		goto bail;
242 	}
243 
244 	/* Clean up list of requests processed during current epoch. */
245 	/* this extra list walk restart is paranoia,
246 	 * to catch requests being barrier-acked "unexpectedly".
247 	 * It usually should find the same req again, or some READ preceding it. */
248 	list_for_each_entry(req, &connection->transfer_log, tl_requests)
249 		if (req->epoch == expect_epoch)
250 			break;
251 	list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
252 		if (req->epoch != expect_epoch)
253 			break;
254 		_req_mod(req, BARRIER_ACKED);
255 	}
256 	spin_unlock_irq(&connection->resource->req_lock);
257 
258 	return;
259 
260 bail:
261 	spin_unlock_irq(&connection->resource->req_lock);
262 	conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
263 }
264 
265 
266 /**
267  * _tl_restart() - Walks the transfer log, and applies an action to all requests
268  * @connection:	DRBD connection to operate on.
269  * @what:       The action/event to perform with all request objects
270  *
271  * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
272  * RESTART_FROZEN_DISK_IO.
273  */
274 /* must hold resource->req_lock */
275 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
276 {
277 	struct drbd_request *req, *r;
278 
279 	list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
280 		_req_mod(req, what);
281 }
282 
283 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
284 {
285 	spin_lock_irq(&connection->resource->req_lock);
286 	_tl_restart(connection, what);
287 	spin_unlock_irq(&connection->resource->req_lock);
288 }
289 
290 /**
291  * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
292  * @device:	DRBD device.
293  *
294  * This is called after the connection to the peer was lost. The storage covered
295  * by the requests on the transfer gets marked as our of sync. Called from the
296  * receiver thread and the worker thread.
297  */
298 void tl_clear(struct drbd_connection *connection)
299 {
300 	tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
301 }
302 
303 /**
304  * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
305  * @device:	DRBD device.
306  */
307 void tl_abort_disk_io(struct drbd_device *device)
308 {
309 	struct drbd_connection *connection = first_peer_device(device)->connection;
310 	struct drbd_request *req, *r;
311 
312 	spin_lock_irq(&connection->resource->req_lock);
313 	list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
314 		if (!(req->rq_state & RQ_LOCAL_PENDING))
315 			continue;
316 		if (req->device != device)
317 			continue;
318 		_req_mod(req, ABORT_DISK_IO);
319 	}
320 	spin_unlock_irq(&connection->resource->req_lock);
321 }
322 
323 static int drbd_thread_setup(void *arg)
324 {
325 	struct drbd_thread *thi = (struct drbd_thread *) arg;
326 	struct drbd_resource *resource = thi->resource;
327 	unsigned long flags;
328 	int retval;
329 
330 	snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
331 		 thi->name[0],
332 		 resource->name);
333 
334 restart:
335 	retval = thi->function(thi);
336 
337 	spin_lock_irqsave(&thi->t_lock, flags);
338 
339 	/* if the receiver has been "EXITING", the last thing it did
340 	 * was set the conn state to "StandAlone",
341 	 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
342 	 * and receiver thread will be "started".
343 	 * drbd_thread_start needs to set "RESTARTING" in that case.
344 	 * t_state check and assignment needs to be within the same spinlock,
345 	 * so either thread_start sees EXITING, and can remap to RESTARTING,
346 	 * or thread_start see NONE, and can proceed as normal.
347 	 */
348 
349 	if (thi->t_state == RESTARTING) {
350 		drbd_info(resource, "Restarting %s thread\n", thi->name);
351 		thi->t_state = RUNNING;
352 		spin_unlock_irqrestore(&thi->t_lock, flags);
353 		goto restart;
354 	}
355 
356 	thi->task = NULL;
357 	thi->t_state = NONE;
358 	smp_mb();
359 	complete_all(&thi->stop);
360 	spin_unlock_irqrestore(&thi->t_lock, flags);
361 
362 	drbd_info(resource, "Terminating %s\n", current->comm);
363 
364 	/* Release mod reference taken when thread was started */
365 
366 	if (thi->connection)
367 		kref_put(&thi->connection->kref, drbd_destroy_connection);
368 	kref_put(&resource->kref, drbd_destroy_resource);
369 	module_put(THIS_MODULE);
370 	return retval;
371 }
372 
373 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
374 			     int (*func) (struct drbd_thread *), const char *name)
375 {
376 	spin_lock_init(&thi->t_lock);
377 	thi->task    = NULL;
378 	thi->t_state = NONE;
379 	thi->function = func;
380 	thi->resource = resource;
381 	thi->connection = NULL;
382 	thi->name = name;
383 }
384 
385 int drbd_thread_start(struct drbd_thread *thi)
386 {
387 	struct drbd_resource *resource = thi->resource;
388 	struct task_struct *nt;
389 	unsigned long flags;
390 
391 	/* is used from state engine doing drbd_thread_stop_nowait,
392 	 * while holding the req lock irqsave */
393 	spin_lock_irqsave(&thi->t_lock, flags);
394 
395 	switch (thi->t_state) {
396 	case NONE:
397 		drbd_info(resource, "Starting %s thread (from %s [%d])\n",
398 			 thi->name, current->comm, current->pid);
399 
400 		/* Get ref on module for thread - this is released when thread exits */
401 		if (!try_module_get(THIS_MODULE)) {
402 			drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
403 			spin_unlock_irqrestore(&thi->t_lock, flags);
404 			return false;
405 		}
406 
407 		kref_get(&resource->kref);
408 		if (thi->connection)
409 			kref_get(&thi->connection->kref);
410 
411 		init_completion(&thi->stop);
412 		thi->reset_cpu_mask = 1;
413 		thi->t_state = RUNNING;
414 		spin_unlock_irqrestore(&thi->t_lock, flags);
415 		flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
416 
417 		nt = kthread_create(drbd_thread_setup, (void *) thi,
418 				    "drbd_%c_%s", thi->name[0], thi->resource->name);
419 
420 		if (IS_ERR(nt)) {
421 			drbd_err(resource, "Couldn't start thread\n");
422 
423 			if (thi->connection)
424 				kref_put(&thi->connection->kref, drbd_destroy_connection);
425 			kref_put(&resource->kref, drbd_destroy_resource);
426 			module_put(THIS_MODULE);
427 			return false;
428 		}
429 		spin_lock_irqsave(&thi->t_lock, flags);
430 		thi->task = nt;
431 		thi->t_state = RUNNING;
432 		spin_unlock_irqrestore(&thi->t_lock, flags);
433 		wake_up_process(nt);
434 		break;
435 	case EXITING:
436 		thi->t_state = RESTARTING;
437 		drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
438 				thi->name, current->comm, current->pid);
439 		/* fall through */
440 	case RUNNING:
441 	case RESTARTING:
442 	default:
443 		spin_unlock_irqrestore(&thi->t_lock, flags);
444 		break;
445 	}
446 
447 	return true;
448 }
449 
450 
451 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
452 {
453 	unsigned long flags;
454 
455 	enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
456 
457 	/* may be called from state engine, holding the req lock irqsave */
458 	spin_lock_irqsave(&thi->t_lock, flags);
459 
460 	if (thi->t_state == NONE) {
461 		spin_unlock_irqrestore(&thi->t_lock, flags);
462 		if (restart)
463 			drbd_thread_start(thi);
464 		return;
465 	}
466 
467 	if (thi->t_state != ns) {
468 		if (thi->task == NULL) {
469 			spin_unlock_irqrestore(&thi->t_lock, flags);
470 			return;
471 		}
472 
473 		thi->t_state = ns;
474 		smp_mb();
475 		init_completion(&thi->stop);
476 		if (thi->task != current)
477 			force_sig(DRBD_SIGKILL, thi->task);
478 	}
479 
480 	spin_unlock_irqrestore(&thi->t_lock, flags);
481 
482 	if (wait)
483 		wait_for_completion(&thi->stop);
484 }
485 
486 int conn_lowest_minor(struct drbd_connection *connection)
487 {
488 	struct drbd_peer_device *peer_device;
489 	int vnr = 0, minor = -1;
490 
491 	rcu_read_lock();
492 	peer_device = idr_get_next(&connection->peer_devices, &vnr);
493 	if (peer_device)
494 		minor = device_to_minor(peer_device->device);
495 	rcu_read_unlock();
496 
497 	return minor;
498 }
499 
500 #ifdef CONFIG_SMP
501 /**
502  * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
503  *
504  * Forces all threads of a resource onto the same CPU. This is beneficial for
505  * DRBD's performance. May be overwritten by user's configuration.
506  */
507 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
508 {
509 	unsigned int *resources_per_cpu, min_index = ~0;
510 
511 	resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL);
512 	if (resources_per_cpu) {
513 		struct drbd_resource *resource;
514 		unsigned int cpu, min = ~0;
515 
516 		rcu_read_lock();
517 		for_each_resource_rcu(resource, &drbd_resources) {
518 			for_each_cpu(cpu, resource->cpu_mask)
519 				resources_per_cpu[cpu]++;
520 		}
521 		rcu_read_unlock();
522 		for_each_online_cpu(cpu) {
523 			if (resources_per_cpu[cpu] < min) {
524 				min = resources_per_cpu[cpu];
525 				min_index = cpu;
526 			}
527 		}
528 		kfree(resources_per_cpu);
529 	}
530 	if (min_index == ~0) {
531 		cpumask_setall(*cpu_mask);
532 		return;
533 	}
534 	cpumask_set_cpu(min_index, *cpu_mask);
535 }
536 
537 /**
538  * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
539  * @device:	DRBD device.
540  * @thi:	drbd_thread object
541  *
542  * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
543  * prematurely.
544  */
545 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
546 {
547 	struct drbd_resource *resource = thi->resource;
548 	struct task_struct *p = current;
549 
550 	if (!thi->reset_cpu_mask)
551 		return;
552 	thi->reset_cpu_mask = 0;
553 	set_cpus_allowed_ptr(p, resource->cpu_mask);
554 }
555 #else
556 #define drbd_calc_cpu_mask(A) ({})
557 #endif
558 
559 /**
560  * drbd_header_size  -  size of a packet header
561  *
562  * The header size is a multiple of 8, so any payload following the header is
563  * word aligned on 64-bit architectures.  (The bitmap send and receive code
564  * relies on this.)
565  */
566 unsigned int drbd_header_size(struct drbd_connection *connection)
567 {
568 	if (connection->agreed_pro_version >= 100) {
569 		BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
570 		return sizeof(struct p_header100);
571 	} else {
572 		BUILD_BUG_ON(sizeof(struct p_header80) !=
573 			     sizeof(struct p_header95));
574 		BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
575 		return sizeof(struct p_header80);
576 	}
577 }
578 
579 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
580 {
581 	h->magic   = cpu_to_be32(DRBD_MAGIC);
582 	h->command = cpu_to_be16(cmd);
583 	h->length  = cpu_to_be16(size);
584 	return sizeof(struct p_header80);
585 }
586 
587 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
588 {
589 	h->magic   = cpu_to_be16(DRBD_MAGIC_BIG);
590 	h->command = cpu_to_be16(cmd);
591 	h->length = cpu_to_be32(size);
592 	return sizeof(struct p_header95);
593 }
594 
595 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
596 				      int size, int vnr)
597 {
598 	h->magic = cpu_to_be32(DRBD_MAGIC_100);
599 	h->volume = cpu_to_be16(vnr);
600 	h->command = cpu_to_be16(cmd);
601 	h->length = cpu_to_be32(size);
602 	h->pad = 0;
603 	return sizeof(struct p_header100);
604 }
605 
606 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
607 				   void *buffer, enum drbd_packet cmd, int size)
608 {
609 	if (connection->agreed_pro_version >= 100)
610 		return prepare_header100(buffer, cmd, size, vnr);
611 	else if (connection->agreed_pro_version >= 95 &&
612 		 size > DRBD_MAX_SIZE_H80_PACKET)
613 		return prepare_header95(buffer, cmd, size);
614 	else
615 		return prepare_header80(buffer, cmd, size);
616 }
617 
618 static void *__conn_prepare_command(struct drbd_connection *connection,
619 				    struct drbd_socket *sock)
620 {
621 	if (!sock->socket)
622 		return NULL;
623 	return sock->sbuf + drbd_header_size(connection);
624 }
625 
626 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
627 {
628 	void *p;
629 
630 	mutex_lock(&sock->mutex);
631 	p = __conn_prepare_command(connection, sock);
632 	if (!p)
633 		mutex_unlock(&sock->mutex);
634 
635 	return p;
636 }
637 
638 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
639 {
640 	return conn_prepare_command(peer_device->connection, sock);
641 }
642 
643 static int __send_command(struct drbd_connection *connection, int vnr,
644 			  struct drbd_socket *sock, enum drbd_packet cmd,
645 			  unsigned int header_size, void *data,
646 			  unsigned int size)
647 {
648 	int msg_flags;
649 	int err;
650 
651 	/*
652 	 * Called with @data == NULL and the size of the data blocks in @size
653 	 * for commands that send data blocks.  For those commands, omit the
654 	 * MSG_MORE flag: this will increase the likelihood that data blocks
655 	 * which are page aligned on the sender will end up page aligned on the
656 	 * receiver.
657 	 */
658 	msg_flags = data ? MSG_MORE : 0;
659 
660 	header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
661 				      header_size + size);
662 	err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
663 			    msg_flags);
664 	if (data && !err)
665 		err = drbd_send_all(connection, sock->socket, data, size, 0);
666 	/* DRBD protocol "pings" are latency critical.
667 	 * This is supposed to trigger tcp_push_pending_frames() */
668 	if (!err && (cmd == P_PING || cmd == P_PING_ACK))
669 		drbd_tcp_nodelay(sock->socket);
670 
671 	return err;
672 }
673 
674 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
675 			       enum drbd_packet cmd, unsigned int header_size,
676 			       void *data, unsigned int size)
677 {
678 	return __send_command(connection, 0, sock, cmd, header_size, data, size);
679 }
680 
681 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
682 		      enum drbd_packet cmd, unsigned int header_size,
683 		      void *data, unsigned int size)
684 {
685 	int err;
686 
687 	err = __conn_send_command(connection, sock, cmd, header_size, data, size);
688 	mutex_unlock(&sock->mutex);
689 	return err;
690 }
691 
692 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
693 		      enum drbd_packet cmd, unsigned int header_size,
694 		      void *data, unsigned int size)
695 {
696 	int err;
697 
698 	err = __send_command(peer_device->connection, peer_device->device->vnr,
699 			     sock, cmd, header_size, data, size);
700 	mutex_unlock(&sock->mutex);
701 	return err;
702 }
703 
704 int drbd_send_ping(struct drbd_connection *connection)
705 {
706 	struct drbd_socket *sock;
707 
708 	sock = &connection->meta;
709 	if (!conn_prepare_command(connection, sock))
710 		return -EIO;
711 	return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
712 }
713 
714 int drbd_send_ping_ack(struct drbd_connection *connection)
715 {
716 	struct drbd_socket *sock;
717 
718 	sock = &connection->meta;
719 	if (!conn_prepare_command(connection, sock))
720 		return -EIO;
721 	return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
722 }
723 
724 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
725 {
726 	struct drbd_socket *sock;
727 	struct p_rs_param_95 *p;
728 	int size;
729 	const int apv = peer_device->connection->agreed_pro_version;
730 	enum drbd_packet cmd;
731 	struct net_conf *nc;
732 	struct disk_conf *dc;
733 
734 	sock = &peer_device->connection->data;
735 	p = drbd_prepare_command(peer_device, sock);
736 	if (!p)
737 		return -EIO;
738 
739 	rcu_read_lock();
740 	nc = rcu_dereference(peer_device->connection->net_conf);
741 
742 	size = apv <= 87 ? sizeof(struct p_rs_param)
743 		: apv == 88 ? sizeof(struct p_rs_param)
744 			+ strlen(nc->verify_alg) + 1
745 		: apv <= 94 ? sizeof(struct p_rs_param_89)
746 		: /* apv >= 95 */ sizeof(struct p_rs_param_95);
747 
748 	cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
749 
750 	/* initialize verify_alg and csums_alg */
751 	memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
752 
753 	if (get_ldev(peer_device->device)) {
754 		dc = rcu_dereference(peer_device->device->ldev->disk_conf);
755 		p->resync_rate = cpu_to_be32(dc->resync_rate);
756 		p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
757 		p->c_delay_target = cpu_to_be32(dc->c_delay_target);
758 		p->c_fill_target = cpu_to_be32(dc->c_fill_target);
759 		p->c_max_rate = cpu_to_be32(dc->c_max_rate);
760 		put_ldev(peer_device->device);
761 	} else {
762 		p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
763 		p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
764 		p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
765 		p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
766 		p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
767 	}
768 
769 	if (apv >= 88)
770 		strcpy(p->verify_alg, nc->verify_alg);
771 	if (apv >= 89)
772 		strcpy(p->csums_alg, nc->csums_alg);
773 	rcu_read_unlock();
774 
775 	return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
776 }
777 
778 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
779 {
780 	struct drbd_socket *sock;
781 	struct p_protocol *p;
782 	struct net_conf *nc;
783 	int size, cf;
784 
785 	sock = &connection->data;
786 	p = __conn_prepare_command(connection, sock);
787 	if (!p)
788 		return -EIO;
789 
790 	rcu_read_lock();
791 	nc = rcu_dereference(connection->net_conf);
792 
793 	if (nc->tentative && connection->agreed_pro_version < 92) {
794 		rcu_read_unlock();
795 		mutex_unlock(&sock->mutex);
796 		drbd_err(connection, "--dry-run is not supported by peer");
797 		return -EOPNOTSUPP;
798 	}
799 
800 	size = sizeof(*p);
801 	if (connection->agreed_pro_version >= 87)
802 		size += strlen(nc->integrity_alg) + 1;
803 
804 	p->protocol      = cpu_to_be32(nc->wire_protocol);
805 	p->after_sb_0p   = cpu_to_be32(nc->after_sb_0p);
806 	p->after_sb_1p   = cpu_to_be32(nc->after_sb_1p);
807 	p->after_sb_2p   = cpu_to_be32(nc->after_sb_2p);
808 	p->two_primaries = cpu_to_be32(nc->two_primaries);
809 	cf = 0;
810 	if (nc->discard_my_data)
811 		cf |= CF_DISCARD_MY_DATA;
812 	if (nc->tentative)
813 		cf |= CF_DRY_RUN;
814 	p->conn_flags    = cpu_to_be32(cf);
815 
816 	if (connection->agreed_pro_version >= 87)
817 		strcpy(p->integrity_alg, nc->integrity_alg);
818 	rcu_read_unlock();
819 
820 	return __conn_send_command(connection, sock, cmd, size, NULL, 0);
821 }
822 
823 int drbd_send_protocol(struct drbd_connection *connection)
824 {
825 	int err;
826 
827 	mutex_lock(&connection->data.mutex);
828 	err = __drbd_send_protocol(connection, P_PROTOCOL);
829 	mutex_unlock(&connection->data.mutex);
830 
831 	return err;
832 }
833 
834 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
835 {
836 	struct drbd_device *device = peer_device->device;
837 	struct drbd_socket *sock;
838 	struct p_uuids *p;
839 	int i;
840 
841 	if (!get_ldev_if_state(device, D_NEGOTIATING))
842 		return 0;
843 
844 	sock = &peer_device->connection->data;
845 	p = drbd_prepare_command(peer_device, sock);
846 	if (!p) {
847 		put_ldev(device);
848 		return -EIO;
849 	}
850 	spin_lock_irq(&device->ldev->md.uuid_lock);
851 	for (i = UI_CURRENT; i < UI_SIZE; i++)
852 		p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
853 	spin_unlock_irq(&device->ldev->md.uuid_lock);
854 
855 	device->comm_bm_set = drbd_bm_total_weight(device);
856 	p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
857 	rcu_read_lock();
858 	uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
859 	rcu_read_unlock();
860 	uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
861 	uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
862 	p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
863 
864 	put_ldev(device);
865 	return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
866 }
867 
868 int drbd_send_uuids(struct drbd_peer_device *peer_device)
869 {
870 	return _drbd_send_uuids(peer_device, 0);
871 }
872 
873 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
874 {
875 	return _drbd_send_uuids(peer_device, 8);
876 }
877 
878 void drbd_print_uuids(struct drbd_device *device, const char *text)
879 {
880 	if (get_ldev_if_state(device, D_NEGOTIATING)) {
881 		u64 *uuid = device->ldev->md.uuid;
882 		drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
883 		     text,
884 		     (unsigned long long)uuid[UI_CURRENT],
885 		     (unsigned long long)uuid[UI_BITMAP],
886 		     (unsigned long long)uuid[UI_HISTORY_START],
887 		     (unsigned long long)uuid[UI_HISTORY_END]);
888 		put_ldev(device);
889 	} else {
890 		drbd_info(device, "%s effective data uuid: %016llX\n",
891 				text,
892 				(unsigned long long)device->ed_uuid);
893 	}
894 }
895 
896 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
897 {
898 	struct drbd_device *device = peer_device->device;
899 	struct drbd_socket *sock;
900 	struct p_rs_uuid *p;
901 	u64 uuid;
902 
903 	D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
904 
905 	uuid = device->ldev->md.uuid[UI_BITMAP];
906 	if (uuid && uuid != UUID_JUST_CREATED)
907 		uuid = uuid + UUID_NEW_BM_OFFSET;
908 	else
909 		get_random_bytes(&uuid, sizeof(u64));
910 	drbd_uuid_set(device, UI_BITMAP, uuid);
911 	drbd_print_uuids(device, "updated sync UUID");
912 	drbd_md_sync(device);
913 
914 	sock = &peer_device->connection->data;
915 	p = drbd_prepare_command(peer_device, sock);
916 	if (p) {
917 		p->uuid = cpu_to_be64(uuid);
918 		drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
919 	}
920 }
921 
922 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
923 {
924 	struct drbd_device *device = peer_device->device;
925 	struct drbd_socket *sock;
926 	struct p_sizes *p;
927 	sector_t d_size, u_size;
928 	int q_order_type;
929 	unsigned int max_bio_size;
930 
931 	if (get_ldev_if_state(device, D_NEGOTIATING)) {
932 		D_ASSERT(device, device->ldev->backing_bdev);
933 		d_size = drbd_get_max_capacity(device->ldev);
934 		rcu_read_lock();
935 		u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
936 		rcu_read_unlock();
937 		q_order_type = drbd_queue_order_type(device);
938 		max_bio_size = queue_max_hw_sectors(device->ldev->backing_bdev->bd_disk->queue) << 9;
939 		max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
940 		put_ldev(device);
941 	} else {
942 		d_size = 0;
943 		u_size = 0;
944 		q_order_type = QUEUE_ORDERED_NONE;
945 		max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
946 	}
947 
948 	sock = &peer_device->connection->data;
949 	p = drbd_prepare_command(peer_device, sock);
950 	if (!p)
951 		return -EIO;
952 
953 	if (peer_device->connection->agreed_pro_version <= 94)
954 		max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
955 	else if (peer_device->connection->agreed_pro_version < 100)
956 		max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
957 
958 	p->d_size = cpu_to_be64(d_size);
959 	p->u_size = cpu_to_be64(u_size);
960 	p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
961 	p->max_bio_size = cpu_to_be32(max_bio_size);
962 	p->queue_order_type = cpu_to_be16(q_order_type);
963 	p->dds_flags = cpu_to_be16(flags);
964 	return drbd_send_command(peer_device, sock, P_SIZES, sizeof(*p), NULL, 0);
965 }
966 
967 /**
968  * drbd_send_current_state() - Sends the drbd state to the peer
969  * @peer_device:	DRBD peer device.
970  */
971 int drbd_send_current_state(struct drbd_peer_device *peer_device)
972 {
973 	struct drbd_socket *sock;
974 	struct p_state *p;
975 
976 	sock = &peer_device->connection->data;
977 	p = drbd_prepare_command(peer_device, sock);
978 	if (!p)
979 		return -EIO;
980 	p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
981 	return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
982 }
983 
984 /**
985  * drbd_send_state() - After a state change, sends the new state to the peer
986  * @peer_device:      DRBD peer device.
987  * @state:     the state to send, not necessarily the current state.
988  *
989  * Each state change queues an "after_state_ch" work, which will eventually
990  * send the resulting new state to the peer. If more state changes happen
991  * between queuing and processing of the after_state_ch work, we still
992  * want to send each intermediary state in the order it occurred.
993  */
994 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
995 {
996 	struct drbd_socket *sock;
997 	struct p_state *p;
998 
999 	sock = &peer_device->connection->data;
1000 	p = drbd_prepare_command(peer_device, sock);
1001 	if (!p)
1002 		return -EIO;
1003 	p->state = cpu_to_be32(state.i); /* Within the send mutex */
1004 	return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1005 }
1006 
1007 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1008 {
1009 	struct drbd_socket *sock;
1010 	struct p_req_state *p;
1011 
1012 	sock = &peer_device->connection->data;
1013 	p = drbd_prepare_command(peer_device, sock);
1014 	if (!p)
1015 		return -EIO;
1016 	p->mask = cpu_to_be32(mask.i);
1017 	p->val = cpu_to_be32(val.i);
1018 	return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1019 }
1020 
1021 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1022 {
1023 	enum drbd_packet cmd;
1024 	struct drbd_socket *sock;
1025 	struct p_req_state *p;
1026 
1027 	cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1028 	sock = &connection->data;
1029 	p = conn_prepare_command(connection, sock);
1030 	if (!p)
1031 		return -EIO;
1032 	p->mask = cpu_to_be32(mask.i);
1033 	p->val = cpu_to_be32(val.i);
1034 	return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1035 }
1036 
1037 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1038 {
1039 	struct drbd_socket *sock;
1040 	struct p_req_state_reply *p;
1041 
1042 	sock = &peer_device->connection->meta;
1043 	p = drbd_prepare_command(peer_device, sock);
1044 	if (p) {
1045 		p->retcode = cpu_to_be32(retcode);
1046 		drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1047 	}
1048 }
1049 
1050 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1051 {
1052 	struct drbd_socket *sock;
1053 	struct p_req_state_reply *p;
1054 	enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1055 
1056 	sock = &connection->meta;
1057 	p = conn_prepare_command(connection, sock);
1058 	if (p) {
1059 		p->retcode = cpu_to_be32(retcode);
1060 		conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1061 	}
1062 }
1063 
1064 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1065 {
1066 	BUG_ON(code & ~0xf);
1067 	p->encoding = (p->encoding & ~0xf) | code;
1068 }
1069 
1070 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1071 {
1072 	p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1073 }
1074 
1075 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1076 {
1077 	BUG_ON(n & ~0x7);
1078 	p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1079 }
1080 
1081 static int fill_bitmap_rle_bits(struct drbd_device *device,
1082 			 struct p_compressed_bm *p,
1083 			 unsigned int size,
1084 			 struct bm_xfer_ctx *c)
1085 {
1086 	struct bitstream bs;
1087 	unsigned long plain_bits;
1088 	unsigned long tmp;
1089 	unsigned long rl;
1090 	unsigned len;
1091 	unsigned toggle;
1092 	int bits, use_rle;
1093 
1094 	/* may we use this feature? */
1095 	rcu_read_lock();
1096 	use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1097 	rcu_read_unlock();
1098 	if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1099 		return 0;
1100 
1101 	if (c->bit_offset >= c->bm_bits)
1102 		return 0; /* nothing to do. */
1103 
1104 	/* use at most thus many bytes */
1105 	bitstream_init(&bs, p->code, size, 0);
1106 	memset(p->code, 0, size);
1107 	/* plain bits covered in this code string */
1108 	plain_bits = 0;
1109 
1110 	/* p->encoding & 0x80 stores whether the first run length is set.
1111 	 * bit offset is implicit.
1112 	 * start with toggle == 2 to be able to tell the first iteration */
1113 	toggle = 2;
1114 
1115 	/* see how much plain bits we can stuff into one packet
1116 	 * using RLE and VLI. */
1117 	do {
1118 		tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1119 				    : _drbd_bm_find_next(device, c->bit_offset);
1120 		if (tmp == -1UL)
1121 			tmp = c->bm_bits;
1122 		rl = tmp - c->bit_offset;
1123 
1124 		if (toggle == 2) { /* first iteration */
1125 			if (rl == 0) {
1126 				/* the first checked bit was set,
1127 				 * store start value, */
1128 				dcbp_set_start(p, 1);
1129 				/* but skip encoding of zero run length */
1130 				toggle = !toggle;
1131 				continue;
1132 			}
1133 			dcbp_set_start(p, 0);
1134 		}
1135 
1136 		/* paranoia: catch zero runlength.
1137 		 * can only happen if bitmap is modified while we scan it. */
1138 		if (rl == 0) {
1139 			drbd_err(device, "unexpected zero runlength while encoding bitmap "
1140 			    "t:%u bo:%lu\n", toggle, c->bit_offset);
1141 			return -1;
1142 		}
1143 
1144 		bits = vli_encode_bits(&bs, rl);
1145 		if (bits == -ENOBUFS) /* buffer full */
1146 			break;
1147 		if (bits <= 0) {
1148 			drbd_err(device, "error while encoding bitmap: %d\n", bits);
1149 			return 0;
1150 		}
1151 
1152 		toggle = !toggle;
1153 		plain_bits += rl;
1154 		c->bit_offset = tmp;
1155 	} while (c->bit_offset < c->bm_bits);
1156 
1157 	len = bs.cur.b - p->code + !!bs.cur.bit;
1158 
1159 	if (plain_bits < (len << 3)) {
1160 		/* incompressible with this method.
1161 		 * we need to rewind both word and bit position. */
1162 		c->bit_offset -= plain_bits;
1163 		bm_xfer_ctx_bit_to_word_offset(c);
1164 		c->bit_offset = c->word_offset * BITS_PER_LONG;
1165 		return 0;
1166 	}
1167 
1168 	/* RLE + VLI was able to compress it just fine.
1169 	 * update c->word_offset. */
1170 	bm_xfer_ctx_bit_to_word_offset(c);
1171 
1172 	/* store pad_bits */
1173 	dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1174 
1175 	return len;
1176 }
1177 
1178 /**
1179  * send_bitmap_rle_or_plain
1180  *
1181  * Return 0 when done, 1 when another iteration is needed, and a negative error
1182  * code upon failure.
1183  */
1184 static int
1185 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1186 {
1187 	struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1188 	unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1189 	struct p_compressed_bm *p = sock->sbuf + header_size;
1190 	int len, err;
1191 
1192 	len = fill_bitmap_rle_bits(device, p,
1193 			DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1194 	if (len < 0)
1195 		return -EIO;
1196 
1197 	if (len) {
1198 		dcbp_set_code(p, RLE_VLI_Bits);
1199 		err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1200 				     P_COMPRESSED_BITMAP, sizeof(*p) + len,
1201 				     NULL, 0);
1202 		c->packets[0]++;
1203 		c->bytes[0] += header_size + sizeof(*p) + len;
1204 
1205 		if (c->bit_offset >= c->bm_bits)
1206 			len = 0; /* DONE */
1207 	} else {
1208 		/* was not compressible.
1209 		 * send a buffer full of plain text bits instead. */
1210 		unsigned int data_size;
1211 		unsigned long num_words;
1212 		unsigned long *p = sock->sbuf + header_size;
1213 
1214 		data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1215 		num_words = min_t(size_t, data_size / sizeof(*p),
1216 				  c->bm_words - c->word_offset);
1217 		len = num_words * sizeof(*p);
1218 		if (len)
1219 			drbd_bm_get_lel(device, c->word_offset, num_words, p);
1220 		err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1221 		c->word_offset += num_words;
1222 		c->bit_offset = c->word_offset * BITS_PER_LONG;
1223 
1224 		c->packets[1]++;
1225 		c->bytes[1] += header_size + len;
1226 
1227 		if (c->bit_offset > c->bm_bits)
1228 			c->bit_offset = c->bm_bits;
1229 	}
1230 	if (!err) {
1231 		if (len == 0) {
1232 			INFO_bm_xfer_stats(device, "send", c);
1233 			return 0;
1234 		} else
1235 			return 1;
1236 	}
1237 	return -EIO;
1238 }
1239 
1240 /* See the comment at receive_bitmap() */
1241 static int _drbd_send_bitmap(struct drbd_device *device)
1242 {
1243 	struct bm_xfer_ctx c;
1244 	int err;
1245 
1246 	if (!expect(device->bitmap))
1247 		return false;
1248 
1249 	if (get_ldev(device)) {
1250 		if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1251 			drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1252 			drbd_bm_set_all(device);
1253 			if (drbd_bm_write(device)) {
1254 				/* write_bm did fail! Leave full sync flag set in Meta P_DATA
1255 				 * but otherwise process as per normal - need to tell other
1256 				 * side that a full resync is required! */
1257 				drbd_err(device, "Failed to write bitmap to disk!\n");
1258 			} else {
1259 				drbd_md_clear_flag(device, MDF_FULL_SYNC);
1260 				drbd_md_sync(device);
1261 			}
1262 		}
1263 		put_ldev(device);
1264 	}
1265 
1266 	c = (struct bm_xfer_ctx) {
1267 		.bm_bits = drbd_bm_bits(device),
1268 		.bm_words = drbd_bm_words(device),
1269 	};
1270 
1271 	do {
1272 		err = send_bitmap_rle_or_plain(device, &c);
1273 	} while (err > 0);
1274 
1275 	return err == 0;
1276 }
1277 
1278 int drbd_send_bitmap(struct drbd_device *device)
1279 {
1280 	struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1281 	int err = -1;
1282 
1283 	mutex_lock(&sock->mutex);
1284 	if (sock->socket)
1285 		err = !_drbd_send_bitmap(device);
1286 	mutex_unlock(&sock->mutex);
1287 	return err;
1288 }
1289 
1290 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1291 {
1292 	struct drbd_socket *sock;
1293 	struct p_barrier_ack *p;
1294 
1295 	if (connection->cstate < C_WF_REPORT_PARAMS)
1296 		return;
1297 
1298 	sock = &connection->meta;
1299 	p = conn_prepare_command(connection, sock);
1300 	if (!p)
1301 		return;
1302 	p->barrier = barrier_nr;
1303 	p->set_size = cpu_to_be32(set_size);
1304 	conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1305 }
1306 
1307 /**
1308  * _drbd_send_ack() - Sends an ack packet
1309  * @device:	DRBD device.
1310  * @cmd:	Packet command code.
1311  * @sector:	sector, needs to be in big endian byte order
1312  * @blksize:	size in byte, needs to be in big endian byte order
1313  * @block_id:	Id, big endian byte order
1314  */
1315 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1316 			  u64 sector, u32 blksize, u64 block_id)
1317 {
1318 	struct drbd_socket *sock;
1319 	struct p_block_ack *p;
1320 
1321 	if (peer_device->device->state.conn < C_CONNECTED)
1322 		return -EIO;
1323 
1324 	sock = &peer_device->connection->meta;
1325 	p = drbd_prepare_command(peer_device, sock);
1326 	if (!p)
1327 		return -EIO;
1328 	p->sector = sector;
1329 	p->block_id = block_id;
1330 	p->blksize = blksize;
1331 	p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1332 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1333 }
1334 
1335 /* dp->sector and dp->block_id already/still in network byte order,
1336  * data_size is payload size according to dp->head,
1337  * and may need to be corrected for digest size. */
1338 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1339 		      struct p_data *dp, int data_size)
1340 {
1341 	if (peer_device->connection->peer_integrity_tfm)
1342 		data_size -= crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm);
1343 	_drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1344 		       dp->block_id);
1345 }
1346 
1347 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1348 		      struct p_block_req *rp)
1349 {
1350 	_drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1351 }
1352 
1353 /**
1354  * drbd_send_ack() - Sends an ack packet
1355  * @device:	DRBD device
1356  * @cmd:	packet command code
1357  * @peer_req:	peer request
1358  */
1359 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1360 		  struct drbd_peer_request *peer_req)
1361 {
1362 	return _drbd_send_ack(peer_device, cmd,
1363 			      cpu_to_be64(peer_req->i.sector),
1364 			      cpu_to_be32(peer_req->i.size),
1365 			      peer_req->block_id);
1366 }
1367 
1368 /* This function misuses the block_id field to signal if the blocks
1369  * are is sync or not. */
1370 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1371 		     sector_t sector, int blksize, u64 block_id)
1372 {
1373 	return _drbd_send_ack(peer_device, cmd,
1374 			      cpu_to_be64(sector),
1375 			      cpu_to_be32(blksize),
1376 			      cpu_to_be64(block_id));
1377 }
1378 
1379 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1380 		       sector_t sector, int size, u64 block_id)
1381 {
1382 	struct drbd_socket *sock;
1383 	struct p_block_req *p;
1384 
1385 	sock = &peer_device->connection->data;
1386 	p = drbd_prepare_command(peer_device, sock);
1387 	if (!p)
1388 		return -EIO;
1389 	p->sector = cpu_to_be64(sector);
1390 	p->block_id = block_id;
1391 	p->blksize = cpu_to_be32(size);
1392 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1393 }
1394 
1395 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1396 			    void *digest, int digest_size, enum drbd_packet cmd)
1397 {
1398 	struct drbd_socket *sock;
1399 	struct p_block_req *p;
1400 
1401 	/* FIXME: Put the digest into the preallocated socket buffer.  */
1402 
1403 	sock = &peer_device->connection->data;
1404 	p = drbd_prepare_command(peer_device, sock);
1405 	if (!p)
1406 		return -EIO;
1407 	p->sector = cpu_to_be64(sector);
1408 	p->block_id = ID_SYNCER /* unused */;
1409 	p->blksize = cpu_to_be32(size);
1410 	return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1411 }
1412 
1413 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1414 {
1415 	struct drbd_socket *sock;
1416 	struct p_block_req *p;
1417 
1418 	sock = &peer_device->connection->data;
1419 	p = drbd_prepare_command(peer_device, sock);
1420 	if (!p)
1421 		return -EIO;
1422 	p->sector = cpu_to_be64(sector);
1423 	p->block_id = ID_SYNCER /* unused */;
1424 	p->blksize = cpu_to_be32(size);
1425 	return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1426 }
1427 
1428 /* called on sndtimeo
1429  * returns false if we should retry,
1430  * true if we think connection is dead
1431  */
1432 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1433 {
1434 	int drop_it;
1435 	/* long elapsed = (long)(jiffies - device->last_received); */
1436 
1437 	drop_it =   connection->meta.socket == sock
1438 		|| !connection->asender.task
1439 		|| get_t_state(&connection->asender) != RUNNING
1440 		|| connection->cstate < C_WF_REPORT_PARAMS;
1441 
1442 	if (drop_it)
1443 		return true;
1444 
1445 	drop_it = !--connection->ko_count;
1446 	if (!drop_it) {
1447 		drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1448 			 current->comm, current->pid, connection->ko_count);
1449 		request_ping(connection);
1450 	}
1451 
1452 	return drop_it; /* && (device->state == R_PRIMARY) */;
1453 }
1454 
1455 static void drbd_update_congested(struct drbd_connection *connection)
1456 {
1457 	struct sock *sk = connection->data.socket->sk;
1458 	if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1459 		set_bit(NET_CONGESTED, &connection->flags);
1460 }
1461 
1462 /* The idea of sendpage seems to be to put some kind of reference
1463  * to the page into the skb, and to hand it over to the NIC. In
1464  * this process get_page() gets called.
1465  *
1466  * As soon as the page was really sent over the network put_page()
1467  * gets called by some part of the network layer. [ NIC driver? ]
1468  *
1469  * [ get_page() / put_page() increment/decrement the count. If count
1470  *   reaches 0 the page will be freed. ]
1471  *
1472  * This works nicely with pages from FSs.
1473  * But this means that in protocol A we might signal IO completion too early!
1474  *
1475  * In order not to corrupt data during a resync we must make sure
1476  * that we do not reuse our own buffer pages (EEs) to early, therefore
1477  * we have the net_ee list.
1478  *
1479  * XFS seems to have problems, still, it submits pages with page_count == 0!
1480  * As a workaround, we disable sendpage on pages
1481  * with page_count == 0 or PageSlab.
1482  */
1483 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1484 			      int offset, size_t size, unsigned msg_flags)
1485 {
1486 	struct socket *socket;
1487 	void *addr;
1488 	int err;
1489 
1490 	socket = peer_device->connection->data.socket;
1491 	addr = kmap(page) + offset;
1492 	err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1493 	kunmap(page);
1494 	if (!err)
1495 		peer_device->device->send_cnt += size >> 9;
1496 	return err;
1497 }
1498 
1499 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1500 		    int offset, size_t size, unsigned msg_flags)
1501 {
1502 	struct socket *socket = peer_device->connection->data.socket;
1503 	mm_segment_t oldfs = get_fs();
1504 	int len = size;
1505 	int err = -EIO;
1506 
1507 	/* e.g. XFS meta- & log-data is in slab pages, which have a
1508 	 * page_count of 0 and/or have PageSlab() set.
1509 	 * we cannot use send_page for those, as that does get_page();
1510 	 * put_page(); and would cause either a VM_BUG directly, or
1511 	 * __page_cache_release a page that would actually still be referenced
1512 	 * by someone, leading to some obscure delayed Oops somewhere else. */
1513 	if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1514 		return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1515 
1516 	msg_flags |= MSG_NOSIGNAL;
1517 	drbd_update_congested(peer_device->connection);
1518 	set_fs(KERNEL_DS);
1519 	do {
1520 		int sent;
1521 
1522 		sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1523 		if (sent <= 0) {
1524 			if (sent == -EAGAIN) {
1525 				if (we_should_drop_the_connection(peer_device->connection, socket))
1526 					break;
1527 				continue;
1528 			}
1529 			drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1530 			     __func__, (int)size, len, sent);
1531 			if (sent < 0)
1532 				err = sent;
1533 			break;
1534 		}
1535 		len    -= sent;
1536 		offset += sent;
1537 	} while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1538 	set_fs(oldfs);
1539 	clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1540 
1541 	if (len == 0) {
1542 		err = 0;
1543 		peer_device->device->send_cnt += size >> 9;
1544 	}
1545 	return err;
1546 }
1547 
1548 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1549 {
1550 	struct bio_vec bvec;
1551 	struct bvec_iter iter;
1552 
1553 	/* hint all but last page with MSG_MORE */
1554 	bio_for_each_segment(bvec, bio, iter) {
1555 		int err;
1556 
1557 		err = _drbd_no_send_page(peer_device, bvec.bv_page,
1558 					 bvec.bv_offset, bvec.bv_len,
1559 					 bio_iter_last(bvec, iter)
1560 					 ? 0 : MSG_MORE);
1561 		if (err)
1562 			return err;
1563 	}
1564 	return 0;
1565 }
1566 
1567 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1568 {
1569 	struct bio_vec bvec;
1570 	struct bvec_iter iter;
1571 
1572 	/* hint all but last page with MSG_MORE */
1573 	bio_for_each_segment(bvec, bio, iter) {
1574 		int err;
1575 
1576 		err = _drbd_send_page(peer_device, bvec.bv_page,
1577 				      bvec.bv_offset, bvec.bv_len,
1578 				      bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1579 		if (err)
1580 			return err;
1581 	}
1582 	return 0;
1583 }
1584 
1585 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1586 			    struct drbd_peer_request *peer_req)
1587 {
1588 	struct page *page = peer_req->pages;
1589 	unsigned len = peer_req->i.size;
1590 	int err;
1591 
1592 	/* hint all but last page with MSG_MORE */
1593 	page_chain_for_each(page) {
1594 		unsigned l = min_t(unsigned, len, PAGE_SIZE);
1595 
1596 		err = _drbd_send_page(peer_device, page, 0, l,
1597 				      page_chain_next(page) ? MSG_MORE : 0);
1598 		if (err)
1599 			return err;
1600 		len -= l;
1601 	}
1602 	return 0;
1603 }
1604 
1605 static u32 bio_flags_to_wire(struct drbd_connection *connection, unsigned long bi_rw)
1606 {
1607 	if (connection->agreed_pro_version >= 95)
1608 		return  (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) |
1609 			(bi_rw & REQ_FUA ? DP_FUA : 0) |
1610 			(bi_rw & REQ_FLUSH ? DP_FLUSH : 0) |
1611 			(bi_rw & REQ_DISCARD ? DP_DISCARD : 0);
1612 	else
1613 		return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0;
1614 }
1615 
1616 /* Used to send write or TRIM aka REQ_DISCARD requests
1617  * R_PRIMARY -> Peer	(P_DATA, P_TRIM)
1618  */
1619 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1620 {
1621 	struct drbd_device *device = peer_device->device;
1622 	struct drbd_socket *sock;
1623 	struct p_data *p;
1624 	unsigned int dp_flags = 0;
1625 	int digest_size;
1626 	int err;
1627 
1628 	sock = &peer_device->connection->data;
1629 	p = drbd_prepare_command(peer_device, sock);
1630 	digest_size = peer_device->connection->integrity_tfm ?
1631 		      crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
1632 
1633 	if (!p)
1634 		return -EIO;
1635 	p->sector = cpu_to_be64(req->i.sector);
1636 	p->block_id = (unsigned long)req;
1637 	p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1638 	dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio->bi_rw);
1639 	if (device->state.conn >= C_SYNC_SOURCE &&
1640 	    device->state.conn <= C_PAUSED_SYNC_T)
1641 		dp_flags |= DP_MAY_SET_IN_SYNC;
1642 	if (peer_device->connection->agreed_pro_version >= 100) {
1643 		if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1644 			dp_flags |= DP_SEND_RECEIVE_ACK;
1645 		/* During resync, request an explicit write ack,
1646 		 * even in protocol != C */
1647 		if (req->rq_state & RQ_EXP_WRITE_ACK
1648 		|| (dp_flags & DP_MAY_SET_IN_SYNC))
1649 			dp_flags |= DP_SEND_WRITE_ACK;
1650 	}
1651 	p->dp_flags = cpu_to_be32(dp_flags);
1652 
1653 	if (dp_flags & DP_DISCARD) {
1654 		struct p_trim *t = (struct p_trim*)p;
1655 		t->size = cpu_to_be32(req->i.size);
1656 		err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1657 		goto out;
1658 	}
1659 
1660 	/* our digest is still only over the payload.
1661 	 * TRIM does not carry any payload. */
1662 	if (digest_size)
1663 		drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, p + 1);
1664 	err = __send_command(peer_device->connection, device->vnr, sock, P_DATA, sizeof(*p) + digest_size, NULL, req->i.size);
1665 	if (!err) {
1666 		/* For protocol A, we have to memcpy the payload into
1667 		 * socket buffers, as we may complete right away
1668 		 * as soon as we handed it over to tcp, at which point the data
1669 		 * pages may become invalid.
1670 		 *
1671 		 * For data-integrity enabled, we copy it as well, so we can be
1672 		 * sure that even if the bio pages may still be modified, it
1673 		 * won't change the data on the wire, thus if the digest checks
1674 		 * out ok after sending on this side, but does not fit on the
1675 		 * receiving side, we sure have detected corruption elsewhere.
1676 		 */
1677 		if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1678 			err = _drbd_send_bio(peer_device, req->master_bio);
1679 		else
1680 			err = _drbd_send_zc_bio(peer_device, req->master_bio);
1681 
1682 		/* double check digest, sometimes buffers have been modified in flight. */
1683 		if (digest_size > 0 && digest_size <= 64) {
1684 			/* 64 byte, 512 bit, is the largest digest size
1685 			 * currently supported in kernel crypto. */
1686 			unsigned char digest[64];
1687 			drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1688 			if (memcmp(p + 1, digest, digest_size)) {
1689 				drbd_warn(device,
1690 					"Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1691 					(unsigned long long)req->i.sector, req->i.size);
1692 			}
1693 		} /* else if (digest_size > 64) {
1694 		     ... Be noisy about digest too large ...
1695 		} */
1696 	}
1697 out:
1698 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1699 
1700 	return err;
1701 }
1702 
1703 /* answer packet, used to send data back for read requests:
1704  *  Peer       -> (diskless) R_PRIMARY   (P_DATA_REPLY)
1705  *  C_SYNC_SOURCE -> C_SYNC_TARGET         (P_RS_DATA_REPLY)
1706  */
1707 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1708 		    struct drbd_peer_request *peer_req)
1709 {
1710 	struct drbd_device *device = peer_device->device;
1711 	struct drbd_socket *sock;
1712 	struct p_data *p;
1713 	int err;
1714 	int digest_size;
1715 
1716 	sock = &peer_device->connection->data;
1717 	p = drbd_prepare_command(peer_device, sock);
1718 
1719 	digest_size = peer_device->connection->integrity_tfm ?
1720 		      crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
1721 
1722 	if (!p)
1723 		return -EIO;
1724 	p->sector = cpu_to_be64(peer_req->i.sector);
1725 	p->block_id = peer_req->block_id;
1726 	p->seq_num = 0;  /* unused */
1727 	p->dp_flags = 0;
1728 	if (digest_size)
1729 		drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1730 	err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1731 	if (!err)
1732 		err = _drbd_send_zc_ee(peer_device, peer_req);
1733 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
1734 
1735 	return err;
1736 }
1737 
1738 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1739 {
1740 	struct drbd_socket *sock;
1741 	struct p_block_desc *p;
1742 
1743 	sock = &peer_device->connection->data;
1744 	p = drbd_prepare_command(peer_device, sock);
1745 	if (!p)
1746 		return -EIO;
1747 	p->sector = cpu_to_be64(req->i.sector);
1748 	p->blksize = cpu_to_be32(req->i.size);
1749 	return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1750 }
1751 
1752 /*
1753   drbd_send distinguishes two cases:
1754 
1755   Packets sent via the data socket "sock"
1756   and packets sent via the meta data socket "msock"
1757 
1758 		    sock                      msock
1759   -----------------+-------------------------+------------------------------
1760   timeout           conf.timeout / 2          conf.timeout / 2
1761   timeout action    send a ping via msock     Abort communication
1762 					      and close all sockets
1763 */
1764 
1765 /*
1766  * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1767  */
1768 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1769 	      void *buf, size_t size, unsigned msg_flags)
1770 {
1771 	struct kvec iov;
1772 	struct msghdr msg;
1773 	int rv, sent = 0;
1774 
1775 	if (!sock)
1776 		return -EBADR;
1777 
1778 	/* THINK  if (signal_pending) return ... ? */
1779 
1780 	iov.iov_base = buf;
1781 	iov.iov_len  = size;
1782 
1783 	msg.msg_name       = NULL;
1784 	msg.msg_namelen    = 0;
1785 	msg.msg_control    = NULL;
1786 	msg.msg_controllen = 0;
1787 	msg.msg_flags      = msg_flags | MSG_NOSIGNAL;
1788 
1789 	if (sock == connection->data.socket) {
1790 		rcu_read_lock();
1791 		connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1792 		rcu_read_unlock();
1793 		drbd_update_congested(connection);
1794 	}
1795 	do {
1796 		/* STRANGE
1797 		 * tcp_sendmsg does _not_ use its size parameter at all ?
1798 		 *
1799 		 * -EAGAIN on timeout, -EINTR on signal.
1800 		 */
1801 /* THINK
1802  * do we need to block DRBD_SIG if sock == &meta.socket ??
1803  * otherwise wake_asender() might interrupt some send_*Ack !
1804  */
1805 		rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
1806 		if (rv == -EAGAIN) {
1807 			if (we_should_drop_the_connection(connection, sock))
1808 				break;
1809 			else
1810 				continue;
1811 		}
1812 		if (rv == -EINTR) {
1813 			flush_signals(current);
1814 			rv = 0;
1815 		}
1816 		if (rv < 0)
1817 			break;
1818 		sent += rv;
1819 		iov.iov_base += rv;
1820 		iov.iov_len  -= rv;
1821 	} while (sent < size);
1822 
1823 	if (sock == connection->data.socket)
1824 		clear_bit(NET_CONGESTED, &connection->flags);
1825 
1826 	if (rv <= 0) {
1827 		if (rv != -EAGAIN) {
1828 			drbd_err(connection, "%s_sendmsg returned %d\n",
1829 				 sock == connection->meta.socket ? "msock" : "sock",
1830 				 rv);
1831 			conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1832 		} else
1833 			conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1834 	}
1835 
1836 	return sent;
1837 }
1838 
1839 /**
1840  * drbd_send_all  -  Send an entire buffer
1841  *
1842  * Returns 0 upon success and a negative error value otherwise.
1843  */
1844 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1845 		  size_t size, unsigned msg_flags)
1846 {
1847 	int err;
1848 
1849 	err = drbd_send(connection, sock, buffer, size, msg_flags);
1850 	if (err < 0)
1851 		return err;
1852 	if (err != size)
1853 		return -EIO;
1854 	return 0;
1855 }
1856 
1857 static int drbd_open(struct block_device *bdev, fmode_t mode)
1858 {
1859 	struct drbd_device *device = bdev->bd_disk->private_data;
1860 	unsigned long flags;
1861 	int rv = 0;
1862 
1863 	mutex_lock(&drbd_main_mutex);
1864 	spin_lock_irqsave(&device->resource->req_lock, flags);
1865 	/* to have a stable device->state.role
1866 	 * and no race with updating open_cnt */
1867 
1868 	if (device->state.role != R_PRIMARY) {
1869 		if (mode & FMODE_WRITE)
1870 			rv = -EROFS;
1871 		else if (!allow_oos)
1872 			rv = -EMEDIUMTYPE;
1873 	}
1874 
1875 	if (!rv)
1876 		device->open_cnt++;
1877 	spin_unlock_irqrestore(&device->resource->req_lock, flags);
1878 	mutex_unlock(&drbd_main_mutex);
1879 
1880 	return rv;
1881 }
1882 
1883 static void drbd_release(struct gendisk *gd, fmode_t mode)
1884 {
1885 	struct drbd_device *device = gd->private_data;
1886 	mutex_lock(&drbd_main_mutex);
1887 	device->open_cnt--;
1888 	mutex_unlock(&drbd_main_mutex);
1889 }
1890 
1891 static void drbd_set_defaults(struct drbd_device *device)
1892 {
1893 	/* Beware! The actual layout differs
1894 	 * between big endian and little endian */
1895 	device->state = (union drbd_dev_state) {
1896 		{ .role = R_SECONDARY,
1897 		  .peer = R_UNKNOWN,
1898 		  .conn = C_STANDALONE,
1899 		  .disk = D_DISKLESS,
1900 		  .pdsk = D_UNKNOWN,
1901 		} };
1902 }
1903 
1904 void drbd_init_set_defaults(struct drbd_device *device)
1905 {
1906 	/* the memset(,0,) did most of this.
1907 	 * note: only assignments, no allocation in here */
1908 
1909 	drbd_set_defaults(device);
1910 
1911 	atomic_set(&device->ap_bio_cnt, 0);
1912 	atomic_set(&device->ap_actlog_cnt, 0);
1913 	atomic_set(&device->ap_pending_cnt, 0);
1914 	atomic_set(&device->rs_pending_cnt, 0);
1915 	atomic_set(&device->unacked_cnt, 0);
1916 	atomic_set(&device->local_cnt, 0);
1917 	atomic_set(&device->pp_in_use_by_net, 0);
1918 	atomic_set(&device->rs_sect_in, 0);
1919 	atomic_set(&device->rs_sect_ev, 0);
1920 	atomic_set(&device->ap_in_flight, 0);
1921 	atomic_set(&device->md_io.in_use, 0);
1922 
1923 	mutex_init(&device->own_state_mutex);
1924 	device->state_mutex = &device->own_state_mutex;
1925 
1926 	spin_lock_init(&device->al_lock);
1927 	spin_lock_init(&device->peer_seq_lock);
1928 
1929 	INIT_LIST_HEAD(&device->active_ee);
1930 	INIT_LIST_HEAD(&device->sync_ee);
1931 	INIT_LIST_HEAD(&device->done_ee);
1932 	INIT_LIST_HEAD(&device->read_ee);
1933 	INIT_LIST_HEAD(&device->net_ee);
1934 	INIT_LIST_HEAD(&device->resync_reads);
1935 	INIT_LIST_HEAD(&device->resync_work.list);
1936 	INIT_LIST_HEAD(&device->unplug_work.list);
1937 	INIT_LIST_HEAD(&device->bm_io_work.w.list);
1938 	INIT_LIST_HEAD(&device->pending_master_completion[0]);
1939 	INIT_LIST_HEAD(&device->pending_master_completion[1]);
1940 	INIT_LIST_HEAD(&device->pending_completion[0]);
1941 	INIT_LIST_HEAD(&device->pending_completion[1]);
1942 
1943 	device->resync_work.cb  = w_resync_timer;
1944 	device->unplug_work.cb  = w_send_write_hint;
1945 	device->bm_io_work.w.cb = w_bitmap_io;
1946 
1947 	init_timer(&device->resync_timer);
1948 	init_timer(&device->md_sync_timer);
1949 	init_timer(&device->start_resync_timer);
1950 	init_timer(&device->request_timer);
1951 	device->resync_timer.function = resync_timer_fn;
1952 	device->resync_timer.data = (unsigned long) device;
1953 	device->md_sync_timer.function = md_sync_timer_fn;
1954 	device->md_sync_timer.data = (unsigned long) device;
1955 	device->start_resync_timer.function = start_resync_timer_fn;
1956 	device->start_resync_timer.data = (unsigned long) device;
1957 	device->request_timer.function = request_timer_fn;
1958 	device->request_timer.data = (unsigned long) device;
1959 
1960 	init_waitqueue_head(&device->misc_wait);
1961 	init_waitqueue_head(&device->state_wait);
1962 	init_waitqueue_head(&device->ee_wait);
1963 	init_waitqueue_head(&device->al_wait);
1964 	init_waitqueue_head(&device->seq_wait);
1965 
1966 	device->resync_wenr = LC_FREE;
1967 	device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1968 	device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1969 }
1970 
1971 void drbd_device_cleanup(struct drbd_device *device)
1972 {
1973 	int i;
1974 	if (first_peer_device(device)->connection->receiver.t_state != NONE)
1975 		drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
1976 				first_peer_device(device)->connection->receiver.t_state);
1977 
1978 	device->al_writ_cnt  =
1979 	device->bm_writ_cnt  =
1980 	device->read_cnt     =
1981 	device->recv_cnt     =
1982 	device->send_cnt     =
1983 	device->writ_cnt     =
1984 	device->p_size       =
1985 	device->rs_start     =
1986 	device->rs_total     =
1987 	device->rs_failed    = 0;
1988 	device->rs_last_events = 0;
1989 	device->rs_last_sect_ev = 0;
1990 	for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1991 		device->rs_mark_left[i] = 0;
1992 		device->rs_mark_time[i] = 0;
1993 	}
1994 	D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
1995 
1996 	drbd_set_my_capacity(device, 0);
1997 	if (device->bitmap) {
1998 		/* maybe never allocated. */
1999 		drbd_bm_resize(device, 0, 1);
2000 		drbd_bm_cleanup(device);
2001 	}
2002 
2003 	drbd_free_ldev(device->ldev);
2004 	device->ldev = NULL;
2005 
2006 	clear_bit(AL_SUSPENDED, &device->flags);
2007 
2008 	D_ASSERT(device, list_empty(&device->active_ee));
2009 	D_ASSERT(device, list_empty(&device->sync_ee));
2010 	D_ASSERT(device, list_empty(&device->done_ee));
2011 	D_ASSERT(device, list_empty(&device->read_ee));
2012 	D_ASSERT(device, list_empty(&device->net_ee));
2013 	D_ASSERT(device, list_empty(&device->resync_reads));
2014 	D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2015 	D_ASSERT(device, list_empty(&device->resync_work.list));
2016 	D_ASSERT(device, list_empty(&device->unplug_work.list));
2017 
2018 	drbd_set_defaults(device);
2019 }
2020 
2021 
2022 static void drbd_destroy_mempools(void)
2023 {
2024 	struct page *page;
2025 
2026 	while (drbd_pp_pool) {
2027 		page = drbd_pp_pool;
2028 		drbd_pp_pool = (struct page *)page_private(page);
2029 		__free_page(page);
2030 		drbd_pp_vacant--;
2031 	}
2032 
2033 	/* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2034 
2035 	if (drbd_md_io_bio_set)
2036 		bioset_free(drbd_md_io_bio_set);
2037 	if (drbd_md_io_page_pool)
2038 		mempool_destroy(drbd_md_io_page_pool);
2039 	if (drbd_ee_mempool)
2040 		mempool_destroy(drbd_ee_mempool);
2041 	if (drbd_request_mempool)
2042 		mempool_destroy(drbd_request_mempool);
2043 	if (drbd_ee_cache)
2044 		kmem_cache_destroy(drbd_ee_cache);
2045 	if (drbd_request_cache)
2046 		kmem_cache_destroy(drbd_request_cache);
2047 	if (drbd_bm_ext_cache)
2048 		kmem_cache_destroy(drbd_bm_ext_cache);
2049 	if (drbd_al_ext_cache)
2050 		kmem_cache_destroy(drbd_al_ext_cache);
2051 
2052 	drbd_md_io_bio_set   = NULL;
2053 	drbd_md_io_page_pool = NULL;
2054 	drbd_ee_mempool      = NULL;
2055 	drbd_request_mempool = NULL;
2056 	drbd_ee_cache        = NULL;
2057 	drbd_request_cache   = NULL;
2058 	drbd_bm_ext_cache    = NULL;
2059 	drbd_al_ext_cache    = NULL;
2060 
2061 	return;
2062 }
2063 
2064 static int drbd_create_mempools(void)
2065 {
2066 	struct page *page;
2067 	const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
2068 	int i;
2069 
2070 	/* prepare our caches and mempools */
2071 	drbd_request_mempool = NULL;
2072 	drbd_ee_cache        = NULL;
2073 	drbd_request_cache   = NULL;
2074 	drbd_bm_ext_cache    = NULL;
2075 	drbd_al_ext_cache    = NULL;
2076 	drbd_pp_pool         = NULL;
2077 	drbd_md_io_page_pool = NULL;
2078 	drbd_md_io_bio_set   = NULL;
2079 
2080 	/* caches */
2081 	drbd_request_cache = kmem_cache_create(
2082 		"drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2083 	if (drbd_request_cache == NULL)
2084 		goto Enomem;
2085 
2086 	drbd_ee_cache = kmem_cache_create(
2087 		"drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2088 	if (drbd_ee_cache == NULL)
2089 		goto Enomem;
2090 
2091 	drbd_bm_ext_cache = kmem_cache_create(
2092 		"drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2093 	if (drbd_bm_ext_cache == NULL)
2094 		goto Enomem;
2095 
2096 	drbd_al_ext_cache = kmem_cache_create(
2097 		"drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2098 	if (drbd_al_ext_cache == NULL)
2099 		goto Enomem;
2100 
2101 	/* mempools */
2102 	drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2103 	if (drbd_md_io_bio_set == NULL)
2104 		goto Enomem;
2105 
2106 	drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2107 	if (drbd_md_io_page_pool == NULL)
2108 		goto Enomem;
2109 
2110 	drbd_request_mempool = mempool_create(number,
2111 		mempool_alloc_slab, mempool_free_slab, drbd_request_cache);
2112 	if (drbd_request_mempool == NULL)
2113 		goto Enomem;
2114 
2115 	drbd_ee_mempool = mempool_create(number,
2116 		mempool_alloc_slab, mempool_free_slab, drbd_ee_cache);
2117 	if (drbd_ee_mempool == NULL)
2118 		goto Enomem;
2119 
2120 	/* drbd's page pool */
2121 	spin_lock_init(&drbd_pp_lock);
2122 
2123 	for (i = 0; i < number; i++) {
2124 		page = alloc_page(GFP_HIGHUSER);
2125 		if (!page)
2126 			goto Enomem;
2127 		set_page_private(page, (unsigned long)drbd_pp_pool);
2128 		drbd_pp_pool = page;
2129 	}
2130 	drbd_pp_vacant = number;
2131 
2132 	return 0;
2133 
2134 Enomem:
2135 	drbd_destroy_mempools(); /* in case we allocated some */
2136 	return -ENOMEM;
2137 }
2138 
2139 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2140 {
2141 	int rr;
2142 
2143 	rr = drbd_free_peer_reqs(device, &device->active_ee);
2144 	if (rr)
2145 		drbd_err(device, "%d EEs in active list found!\n", rr);
2146 
2147 	rr = drbd_free_peer_reqs(device, &device->sync_ee);
2148 	if (rr)
2149 		drbd_err(device, "%d EEs in sync list found!\n", rr);
2150 
2151 	rr = drbd_free_peer_reqs(device, &device->read_ee);
2152 	if (rr)
2153 		drbd_err(device, "%d EEs in read list found!\n", rr);
2154 
2155 	rr = drbd_free_peer_reqs(device, &device->done_ee);
2156 	if (rr)
2157 		drbd_err(device, "%d EEs in done list found!\n", rr);
2158 
2159 	rr = drbd_free_peer_reqs(device, &device->net_ee);
2160 	if (rr)
2161 		drbd_err(device, "%d EEs in net list found!\n", rr);
2162 }
2163 
2164 /* caution. no locking. */
2165 void drbd_destroy_device(struct kref *kref)
2166 {
2167 	struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2168 	struct drbd_resource *resource = device->resource;
2169 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2170 
2171 	del_timer_sync(&device->request_timer);
2172 
2173 	/* paranoia asserts */
2174 	D_ASSERT(device, device->open_cnt == 0);
2175 	/* end paranoia asserts */
2176 
2177 	/* cleanup stuff that may have been allocated during
2178 	 * device (re-)configuration or state changes */
2179 
2180 	if (device->this_bdev)
2181 		bdput(device->this_bdev);
2182 
2183 	drbd_free_ldev(device->ldev);
2184 	device->ldev = NULL;
2185 
2186 	drbd_release_all_peer_reqs(device);
2187 
2188 	lc_destroy(device->act_log);
2189 	lc_destroy(device->resync);
2190 
2191 	kfree(device->p_uuid);
2192 	/* device->p_uuid = NULL; */
2193 
2194 	if (device->bitmap) /* should no longer be there. */
2195 		drbd_bm_cleanup(device);
2196 	__free_page(device->md_io.page);
2197 	put_disk(device->vdisk);
2198 	blk_cleanup_queue(device->rq_queue);
2199 	kfree(device->rs_plan_s);
2200 
2201 	/* not for_each_connection(connection, resource):
2202 	 * those may have been cleaned up and disassociated already.
2203 	 */
2204 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2205 		kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2206 		kfree(peer_device);
2207 	}
2208 	memset(device, 0xfd, sizeof(*device));
2209 	kfree(device);
2210 	kref_put(&resource->kref, drbd_destroy_resource);
2211 }
2212 
2213 /* One global retry thread, if we need to push back some bio and have it
2214  * reinserted through our make request function.
2215  */
2216 static struct retry_worker {
2217 	struct workqueue_struct *wq;
2218 	struct work_struct worker;
2219 
2220 	spinlock_t lock;
2221 	struct list_head writes;
2222 } retry;
2223 
2224 static void do_retry(struct work_struct *ws)
2225 {
2226 	struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2227 	LIST_HEAD(writes);
2228 	struct drbd_request *req, *tmp;
2229 
2230 	spin_lock_irq(&retry->lock);
2231 	list_splice_init(&retry->writes, &writes);
2232 	spin_unlock_irq(&retry->lock);
2233 
2234 	list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2235 		struct drbd_device *device = req->device;
2236 		struct bio *bio = req->master_bio;
2237 		unsigned long start_jif = req->start_jif;
2238 		bool expected;
2239 
2240 		expected =
2241 			expect(atomic_read(&req->completion_ref) == 0) &&
2242 			expect(req->rq_state & RQ_POSTPONED) &&
2243 			expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2244 				(req->rq_state & RQ_LOCAL_ABORTED) != 0);
2245 
2246 		if (!expected)
2247 			drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2248 				req, atomic_read(&req->completion_ref),
2249 				req->rq_state);
2250 
2251 		/* We still need to put one kref associated with the
2252 		 * "completion_ref" going zero in the code path that queued it
2253 		 * here.  The request object may still be referenced by a
2254 		 * frozen local req->private_bio, in case we force-detached.
2255 		 */
2256 		kref_put(&req->kref, drbd_req_destroy);
2257 
2258 		/* A single suspended or otherwise blocking device may stall
2259 		 * all others as well.  Fortunately, this code path is to
2260 		 * recover from a situation that "should not happen":
2261 		 * concurrent writes in multi-primary setup.
2262 		 * In a "normal" lifecycle, this workqueue is supposed to be
2263 		 * destroyed without ever doing anything.
2264 		 * If it turns out to be an issue anyways, we can do per
2265 		 * resource (replication group) or per device (minor) retry
2266 		 * workqueues instead.
2267 		 */
2268 
2269 		/* We are not just doing generic_make_request(),
2270 		 * as we want to keep the start_time information. */
2271 		inc_ap_bio(device);
2272 		__drbd_make_request(device, bio, start_jif);
2273 	}
2274 }
2275 
2276 /* called via drbd_req_put_completion_ref(),
2277  * holds resource->req_lock */
2278 void drbd_restart_request(struct drbd_request *req)
2279 {
2280 	unsigned long flags;
2281 	spin_lock_irqsave(&retry.lock, flags);
2282 	list_move_tail(&req->tl_requests, &retry.writes);
2283 	spin_unlock_irqrestore(&retry.lock, flags);
2284 
2285 	/* Drop the extra reference that would otherwise
2286 	 * have been dropped by complete_master_bio.
2287 	 * do_retry() needs to grab a new one. */
2288 	dec_ap_bio(req->device);
2289 
2290 	queue_work(retry.wq, &retry.worker);
2291 }
2292 
2293 void drbd_destroy_resource(struct kref *kref)
2294 {
2295 	struct drbd_resource *resource =
2296 		container_of(kref, struct drbd_resource, kref);
2297 
2298 	idr_destroy(&resource->devices);
2299 	free_cpumask_var(resource->cpu_mask);
2300 	kfree(resource->name);
2301 	memset(resource, 0xf2, sizeof(*resource));
2302 	kfree(resource);
2303 }
2304 
2305 void drbd_free_resource(struct drbd_resource *resource)
2306 {
2307 	struct drbd_connection *connection, *tmp;
2308 
2309 	for_each_connection_safe(connection, tmp, resource) {
2310 		list_del(&connection->connections);
2311 		drbd_debugfs_connection_cleanup(connection);
2312 		kref_put(&connection->kref, drbd_destroy_connection);
2313 	}
2314 	drbd_debugfs_resource_cleanup(resource);
2315 	kref_put(&resource->kref, drbd_destroy_resource);
2316 }
2317 
2318 static void drbd_cleanup(void)
2319 {
2320 	unsigned int i;
2321 	struct drbd_device *device;
2322 	struct drbd_resource *resource, *tmp;
2323 
2324 	/* first remove proc,
2325 	 * drbdsetup uses it's presence to detect
2326 	 * whether DRBD is loaded.
2327 	 * If we would get stuck in proc removal,
2328 	 * but have netlink already deregistered,
2329 	 * some drbdsetup commands may wait forever
2330 	 * for an answer.
2331 	 */
2332 	if (drbd_proc)
2333 		remove_proc_entry("drbd", NULL);
2334 
2335 	if (retry.wq)
2336 		destroy_workqueue(retry.wq);
2337 
2338 	drbd_genl_unregister();
2339 	drbd_debugfs_cleanup();
2340 
2341 	idr_for_each_entry(&drbd_devices, device, i)
2342 		drbd_delete_device(device);
2343 
2344 	/* not _rcu since, no other updater anymore. Genl already unregistered */
2345 	for_each_resource_safe(resource, tmp, &drbd_resources) {
2346 		list_del(&resource->resources);
2347 		drbd_free_resource(resource);
2348 	}
2349 
2350 	drbd_destroy_mempools();
2351 	unregister_blkdev(DRBD_MAJOR, "drbd");
2352 
2353 	idr_destroy(&drbd_devices);
2354 
2355 	pr_info("module cleanup done.\n");
2356 }
2357 
2358 /**
2359  * drbd_congested() - Callback for the flusher thread
2360  * @congested_data:	User data
2361  * @bdi_bits:		Bits the BDI flusher thread is currently interested in
2362  *
2363  * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2364  */
2365 static int drbd_congested(void *congested_data, int bdi_bits)
2366 {
2367 	struct drbd_device *device = congested_data;
2368 	struct request_queue *q;
2369 	char reason = '-';
2370 	int r = 0;
2371 
2372 	if (!may_inc_ap_bio(device)) {
2373 		/* DRBD has frozen IO */
2374 		r = bdi_bits;
2375 		reason = 'd';
2376 		goto out;
2377 	}
2378 
2379 	if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
2380 		r |= (1 << BDI_async_congested);
2381 		/* Without good local data, we would need to read from remote,
2382 		 * and that would need the worker thread as well, which is
2383 		 * currently blocked waiting for that usermode helper to
2384 		 * finish.
2385 		 */
2386 		if (!get_ldev_if_state(device, D_UP_TO_DATE))
2387 			r |= (1 << BDI_sync_congested);
2388 		else
2389 			put_ldev(device);
2390 		r &= bdi_bits;
2391 		reason = 'c';
2392 		goto out;
2393 	}
2394 
2395 	if (get_ldev(device)) {
2396 		q = bdev_get_queue(device->ldev->backing_bdev);
2397 		r = bdi_congested(&q->backing_dev_info, bdi_bits);
2398 		put_ldev(device);
2399 		if (r)
2400 			reason = 'b';
2401 	}
2402 
2403 	if (bdi_bits & (1 << BDI_async_congested) &&
2404 	    test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
2405 		r |= (1 << BDI_async_congested);
2406 		reason = reason == 'b' ? 'a' : 'n';
2407 	}
2408 
2409 out:
2410 	device->congestion_reason = reason;
2411 	return r;
2412 }
2413 
2414 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2415 {
2416 	spin_lock_init(&wq->q_lock);
2417 	INIT_LIST_HEAD(&wq->q);
2418 	init_waitqueue_head(&wq->q_wait);
2419 }
2420 
2421 struct completion_work {
2422 	struct drbd_work w;
2423 	struct completion done;
2424 };
2425 
2426 static int w_complete(struct drbd_work *w, int cancel)
2427 {
2428 	struct completion_work *completion_work =
2429 		container_of(w, struct completion_work, w);
2430 
2431 	complete(&completion_work->done);
2432 	return 0;
2433 }
2434 
2435 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2436 {
2437 	struct completion_work completion_work;
2438 
2439 	completion_work.w.cb = w_complete;
2440 	init_completion(&completion_work.done);
2441 	drbd_queue_work(work_queue, &completion_work.w);
2442 	wait_for_completion(&completion_work.done);
2443 }
2444 
2445 struct drbd_resource *drbd_find_resource(const char *name)
2446 {
2447 	struct drbd_resource *resource;
2448 
2449 	if (!name || !name[0])
2450 		return NULL;
2451 
2452 	rcu_read_lock();
2453 	for_each_resource_rcu(resource, &drbd_resources) {
2454 		if (!strcmp(resource->name, name)) {
2455 			kref_get(&resource->kref);
2456 			goto found;
2457 		}
2458 	}
2459 	resource = NULL;
2460 found:
2461 	rcu_read_unlock();
2462 	return resource;
2463 }
2464 
2465 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2466 				     void *peer_addr, int peer_addr_len)
2467 {
2468 	struct drbd_resource *resource;
2469 	struct drbd_connection *connection;
2470 
2471 	rcu_read_lock();
2472 	for_each_resource_rcu(resource, &drbd_resources) {
2473 		for_each_connection_rcu(connection, resource) {
2474 			if (connection->my_addr_len == my_addr_len &&
2475 			    connection->peer_addr_len == peer_addr_len &&
2476 			    !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2477 			    !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2478 				kref_get(&connection->kref);
2479 				goto found;
2480 			}
2481 		}
2482 	}
2483 	connection = NULL;
2484 found:
2485 	rcu_read_unlock();
2486 	return connection;
2487 }
2488 
2489 static int drbd_alloc_socket(struct drbd_socket *socket)
2490 {
2491 	socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2492 	if (!socket->rbuf)
2493 		return -ENOMEM;
2494 	socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2495 	if (!socket->sbuf)
2496 		return -ENOMEM;
2497 	return 0;
2498 }
2499 
2500 static void drbd_free_socket(struct drbd_socket *socket)
2501 {
2502 	free_page((unsigned long) socket->sbuf);
2503 	free_page((unsigned long) socket->rbuf);
2504 }
2505 
2506 void conn_free_crypto(struct drbd_connection *connection)
2507 {
2508 	drbd_free_sock(connection);
2509 
2510 	crypto_free_hash(connection->csums_tfm);
2511 	crypto_free_hash(connection->verify_tfm);
2512 	crypto_free_hash(connection->cram_hmac_tfm);
2513 	crypto_free_hash(connection->integrity_tfm);
2514 	crypto_free_hash(connection->peer_integrity_tfm);
2515 	kfree(connection->int_dig_in);
2516 	kfree(connection->int_dig_vv);
2517 
2518 	connection->csums_tfm = NULL;
2519 	connection->verify_tfm = NULL;
2520 	connection->cram_hmac_tfm = NULL;
2521 	connection->integrity_tfm = NULL;
2522 	connection->peer_integrity_tfm = NULL;
2523 	connection->int_dig_in = NULL;
2524 	connection->int_dig_vv = NULL;
2525 }
2526 
2527 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2528 {
2529 	struct drbd_connection *connection;
2530 	cpumask_var_t new_cpu_mask;
2531 	int err;
2532 
2533 	if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2534 		return -ENOMEM;
2535 
2536 	/* silently ignore cpu mask on UP kernel */
2537 	if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2538 		err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2539 				   cpumask_bits(new_cpu_mask), nr_cpu_ids);
2540 		if (err == -EOVERFLOW) {
2541 			/* So what. mask it out. */
2542 			cpumask_var_t tmp_cpu_mask;
2543 			if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2544 				cpumask_setall(tmp_cpu_mask);
2545 				cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2546 				drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2547 					res_opts->cpu_mask,
2548 					strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2549 					nr_cpu_ids);
2550 				free_cpumask_var(tmp_cpu_mask);
2551 				err = 0;
2552 			}
2553 		}
2554 		if (err) {
2555 			drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2556 			/* retcode = ERR_CPU_MASK_PARSE; */
2557 			goto fail;
2558 		}
2559 	}
2560 	resource->res_opts = *res_opts;
2561 	if (cpumask_empty(new_cpu_mask))
2562 		drbd_calc_cpu_mask(&new_cpu_mask);
2563 	if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2564 		cpumask_copy(resource->cpu_mask, new_cpu_mask);
2565 		for_each_connection_rcu(connection, resource) {
2566 			connection->receiver.reset_cpu_mask = 1;
2567 			connection->asender.reset_cpu_mask = 1;
2568 			connection->worker.reset_cpu_mask = 1;
2569 		}
2570 	}
2571 	err = 0;
2572 
2573 fail:
2574 	free_cpumask_var(new_cpu_mask);
2575 	return err;
2576 
2577 }
2578 
2579 struct drbd_resource *drbd_create_resource(const char *name)
2580 {
2581 	struct drbd_resource *resource;
2582 
2583 	resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2584 	if (!resource)
2585 		goto fail;
2586 	resource->name = kstrdup(name, GFP_KERNEL);
2587 	if (!resource->name)
2588 		goto fail_free_resource;
2589 	if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2590 		goto fail_free_name;
2591 	kref_init(&resource->kref);
2592 	idr_init(&resource->devices);
2593 	INIT_LIST_HEAD(&resource->connections);
2594 	resource->write_ordering = WO_bdev_flush;
2595 	list_add_tail_rcu(&resource->resources, &drbd_resources);
2596 	mutex_init(&resource->conf_update);
2597 	mutex_init(&resource->adm_mutex);
2598 	spin_lock_init(&resource->req_lock);
2599 	drbd_debugfs_resource_add(resource);
2600 	return resource;
2601 
2602 fail_free_name:
2603 	kfree(resource->name);
2604 fail_free_resource:
2605 	kfree(resource);
2606 fail:
2607 	return NULL;
2608 }
2609 
2610 /* caller must be under adm_mutex */
2611 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2612 {
2613 	struct drbd_resource *resource;
2614 	struct drbd_connection *connection;
2615 
2616 	connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2617 	if (!connection)
2618 		return NULL;
2619 
2620 	if (drbd_alloc_socket(&connection->data))
2621 		goto fail;
2622 	if (drbd_alloc_socket(&connection->meta))
2623 		goto fail;
2624 
2625 	connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2626 	if (!connection->current_epoch)
2627 		goto fail;
2628 
2629 	INIT_LIST_HEAD(&connection->transfer_log);
2630 
2631 	INIT_LIST_HEAD(&connection->current_epoch->list);
2632 	connection->epochs = 1;
2633 	spin_lock_init(&connection->epoch_lock);
2634 
2635 	connection->send.seen_any_write_yet = false;
2636 	connection->send.current_epoch_nr = 0;
2637 	connection->send.current_epoch_writes = 0;
2638 
2639 	resource = drbd_create_resource(name);
2640 	if (!resource)
2641 		goto fail;
2642 
2643 	connection->cstate = C_STANDALONE;
2644 	mutex_init(&connection->cstate_mutex);
2645 	init_waitqueue_head(&connection->ping_wait);
2646 	idr_init(&connection->peer_devices);
2647 
2648 	drbd_init_workqueue(&connection->sender_work);
2649 	mutex_init(&connection->data.mutex);
2650 	mutex_init(&connection->meta.mutex);
2651 
2652 	drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2653 	connection->receiver.connection = connection;
2654 	drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2655 	connection->worker.connection = connection;
2656 	drbd_thread_init(resource, &connection->asender, drbd_asender, "asender");
2657 	connection->asender.connection = connection;
2658 
2659 	kref_init(&connection->kref);
2660 
2661 	connection->resource = resource;
2662 
2663 	if (set_resource_options(resource, res_opts))
2664 		goto fail_resource;
2665 
2666 	kref_get(&resource->kref);
2667 	list_add_tail_rcu(&connection->connections, &resource->connections);
2668 	drbd_debugfs_connection_add(connection);
2669 	return connection;
2670 
2671 fail_resource:
2672 	list_del(&resource->resources);
2673 	drbd_free_resource(resource);
2674 fail:
2675 	kfree(connection->current_epoch);
2676 	drbd_free_socket(&connection->meta);
2677 	drbd_free_socket(&connection->data);
2678 	kfree(connection);
2679 	return NULL;
2680 }
2681 
2682 void drbd_destroy_connection(struct kref *kref)
2683 {
2684 	struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2685 	struct drbd_resource *resource = connection->resource;
2686 
2687 	if (atomic_read(&connection->current_epoch->epoch_size) !=  0)
2688 		drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2689 	kfree(connection->current_epoch);
2690 
2691 	idr_destroy(&connection->peer_devices);
2692 
2693 	drbd_free_socket(&connection->meta);
2694 	drbd_free_socket(&connection->data);
2695 	kfree(connection->int_dig_in);
2696 	kfree(connection->int_dig_vv);
2697 	memset(connection, 0xfc, sizeof(*connection));
2698 	kfree(connection);
2699 	kref_put(&resource->kref, drbd_destroy_resource);
2700 }
2701 
2702 static int init_submitter(struct drbd_device *device)
2703 {
2704 	/* opencoded create_singlethread_workqueue(),
2705 	 * to be able to say "drbd%d", ..., minor */
2706 	device->submit.wq = alloc_workqueue("drbd%u_submit",
2707 			WQ_UNBOUND | WQ_MEM_RECLAIM, 1, device->minor);
2708 	if (!device->submit.wq)
2709 		return -ENOMEM;
2710 
2711 	INIT_WORK(&device->submit.worker, do_submit);
2712 	INIT_LIST_HEAD(&device->submit.writes);
2713 	return 0;
2714 }
2715 
2716 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2717 {
2718 	struct drbd_resource *resource = adm_ctx->resource;
2719 	struct drbd_connection *connection;
2720 	struct drbd_device *device;
2721 	struct drbd_peer_device *peer_device, *tmp_peer_device;
2722 	struct gendisk *disk;
2723 	struct request_queue *q;
2724 	int id;
2725 	int vnr = adm_ctx->volume;
2726 	enum drbd_ret_code err = ERR_NOMEM;
2727 
2728 	device = minor_to_device(minor);
2729 	if (device)
2730 		return ERR_MINOR_OR_VOLUME_EXISTS;
2731 
2732 	/* GFP_KERNEL, we are outside of all write-out paths */
2733 	device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2734 	if (!device)
2735 		return ERR_NOMEM;
2736 	kref_init(&device->kref);
2737 
2738 	kref_get(&resource->kref);
2739 	device->resource = resource;
2740 	device->minor = minor;
2741 	device->vnr = vnr;
2742 
2743 	drbd_init_set_defaults(device);
2744 
2745 	q = blk_alloc_queue(GFP_KERNEL);
2746 	if (!q)
2747 		goto out_no_q;
2748 	device->rq_queue = q;
2749 	q->queuedata   = device;
2750 
2751 	disk = alloc_disk(1);
2752 	if (!disk)
2753 		goto out_no_disk;
2754 	device->vdisk = disk;
2755 
2756 	set_disk_ro(disk, true);
2757 
2758 	disk->queue = q;
2759 	disk->major = DRBD_MAJOR;
2760 	disk->first_minor = minor;
2761 	disk->fops = &drbd_ops;
2762 	sprintf(disk->disk_name, "drbd%d", minor);
2763 	disk->private_data = device;
2764 
2765 	device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2766 	/* we have no partitions. we contain only ourselves. */
2767 	device->this_bdev->bd_contains = device->this_bdev;
2768 
2769 	q->backing_dev_info.congested_fn = drbd_congested;
2770 	q->backing_dev_info.congested_data = device;
2771 
2772 	blk_queue_make_request(q, drbd_make_request);
2773 	blk_queue_flush(q, REQ_FLUSH | REQ_FUA);
2774 	/* Setting the max_hw_sectors to an odd value of 8kibyte here
2775 	   This triggers a max_bio_size message upon first attach or connect */
2776 	blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2777 	blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2778 	blk_queue_merge_bvec(q, drbd_merge_bvec);
2779 	q->queue_lock = &resource->req_lock;
2780 
2781 	device->md_io.page = alloc_page(GFP_KERNEL);
2782 	if (!device->md_io.page)
2783 		goto out_no_io_page;
2784 
2785 	if (drbd_bm_init(device))
2786 		goto out_no_bitmap;
2787 	device->read_requests = RB_ROOT;
2788 	device->write_requests = RB_ROOT;
2789 
2790 	id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2791 	if (id < 0) {
2792 		if (id == -ENOSPC)
2793 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2794 		goto out_no_minor_idr;
2795 	}
2796 	kref_get(&device->kref);
2797 
2798 	id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2799 	if (id < 0) {
2800 		if (id == -ENOSPC)
2801 			err = ERR_MINOR_OR_VOLUME_EXISTS;
2802 		goto out_idr_remove_minor;
2803 	}
2804 	kref_get(&device->kref);
2805 
2806 	INIT_LIST_HEAD(&device->peer_devices);
2807 	INIT_LIST_HEAD(&device->pending_bitmap_io);
2808 	for_each_connection(connection, resource) {
2809 		peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2810 		if (!peer_device)
2811 			goto out_idr_remove_from_resource;
2812 		peer_device->connection = connection;
2813 		peer_device->device = device;
2814 
2815 		list_add(&peer_device->peer_devices, &device->peer_devices);
2816 		kref_get(&device->kref);
2817 
2818 		id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2819 		if (id < 0) {
2820 			if (id == -ENOSPC)
2821 				err = ERR_INVALID_REQUEST;
2822 			goto out_idr_remove_from_resource;
2823 		}
2824 		kref_get(&connection->kref);
2825 	}
2826 
2827 	if (init_submitter(device)) {
2828 		err = ERR_NOMEM;
2829 		goto out_idr_remove_vol;
2830 	}
2831 
2832 	add_disk(disk);
2833 
2834 	/* inherit the connection state */
2835 	device->state.conn = first_connection(resource)->cstate;
2836 	if (device->state.conn == C_WF_REPORT_PARAMS) {
2837 		for_each_peer_device(peer_device, device)
2838 			drbd_connected(peer_device);
2839 	}
2840 	/* move to create_peer_device() */
2841 	for_each_peer_device(peer_device, device)
2842 		drbd_debugfs_peer_device_add(peer_device);
2843 	drbd_debugfs_device_add(device);
2844 	return NO_ERROR;
2845 
2846 out_idr_remove_vol:
2847 	idr_remove(&connection->peer_devices, vnr);
2848 out_idr_remove_from_resource:
2849 	for_each_connection(connection, resource) {
2850 		peer_device = idr_find(&connection->peer_devices, vnr);
2851 		if (peer_device) {
2852 			idr_remove(&connection->peer_devices, vnr);
2853 			kref_put(&connection->kref, drbd_destroy_connection);
2854 		}
2855 	}
2856 	for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2857 		list_del(&peer_device->peer_devices);
2858 		kfree(peer_device);
2859 	}
2860 	idr_remove(&resource->devices, vnr);
2861 out_idr_remove_minor:
2862 	idr_remove(&drbd_devices, minor);
2863 	synchronize_rcu();
2864 out_no_minor_idr:
2865 	drbd_bm_cleanup(device);
2866 out_no_bitmap:
2867 	__free_page(device->md_io.page);
2868 out_no_io_page:
2869 	put_disk(disk);
2870 out_no_disk:
2871 	blk_cleanup_queue(q);
2872 out_no_q:
2873 	kref_put(&resource->kref, drbd_destroy_resource);
2874 	kfree(device);
2875 	return err;
2876 }
2877 
2878 void drbd_delete_device(struct drbd_device *device)
2879 {
2880 	struct drbd_resource *resource = device->resource;
2881 	struct drbd_connection *connection;
2882 	struct drbd_peer_device *peer_device;
2883 	int refs = 3;
2884 
2885 	/* move to free_peer_device() */
2886 	for_each_peer_device(peer_device, device)
2887 		drbd_debugfs_peer_device_cleanup(peer_device);
2888 	drbd_debugfs_device_cleanup(device);
2889 	for_each_connection(connection, resource) {
2890 		idr_remove(&connection->peer_devices, device->vnr);
2891 		refs++;
2892 	}
2893 	idr_remove(&resource->devices, device->vnr);
2894 	idr_remove(&drbd_devices, device_to_minor(device));
2895 	del_gendisk(device->vdisk);
2896 	synchronize_rcu();
2897 	kref_sub(&device->kref, refs, drbd_destroy_device);
2898 }
2899 
2900 static int __init drbd_init(void)
2901 {
2902 	int err;
2903 
2904 	if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2905 		pr_err("invalid minor_count (%d)\n", minor_count);
2906 #ifdef MODULE
2907 		return -EINVAL;
2908 #else
2909 		minor_count = DRBD_MINOR_COUNT_DEF;
2910 #endif
2911 	}
2912 
2913 	err = register_blkdev(DRBD_MAJOR, "drbd");
2914 	if (err) {
2915 		pr_err("unable to register block device major %d\n",
2916 		       DRBD_MAJOR);
2917 		return err;
2918 	}
2919 
2920 	/*
2921 	 * allocate all necessary structs
2922 	 */
2923 	init_waitqueue_head(&drbd_pp_wait);
2924 
2925 	drbd_proc = NULL; /* play safe for drbd_cleanup */
2926 	idr_init(&drbd_devices);
2927 
2928 	rwlock_init(&global_state_lock);
2929 	INIT_LIST_HEAD(&drbd_resources);
2930 
2931 	err = drbd_genl_register();
2932 	if (err) {
2933 		pr_err("unable to register generic netlink family\n");
2934 		goto fail;
2935 	}
2936 
2937 	err = drbd_create_mempools();
2938 	if (err)
2939 		goto fail;
2940 
2941 	err = -ENOMEM;
2942 	drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
2943 	if (!drbd_proc)	{
2944 		pr_err("unable to register proc file\n");
2945 		goto fail;
2946 	}
2947 
2948 	retry.wq = create_singlethread_workqueue("drbd-reissue");
2949 	if (!retry.wq) {
2950 		pr_err("unable to create retry workqueue\n");
2951 		goto fail;
2952 	}
2953 	INIT_WORK(&retry.worker, do_retry);
2954 	spin_lock_init(&retry.lock);
2955 	INIT_LIST_HEAD(&retry.writes);
2956 
2957 	if (drbd_debugfs_init())
2958 		pr_notice("failed to initialize debugfs -- will not be available\n");
2959 
2960 	pr_info("initialized. "
2961 	       "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2962 	       API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2963 	pr_info("%s\n", drbd_buildtag());
2964 	pr_info("registered as block device major %d\n", DRBD_MAJOR);
2965 	return 0; /* Success! */
2966 
2967 fail:
2968 	drbd_cleanup();
2969 	if (err == -ENOMEM)
2970 		pr_err("ran out of memory\n");
2971 	else
2972 		pr_err("initialization failure\n");
2973 	return err;
2974 }
2975 
2976 void drbd_free_ldev(struct drbd_backing_dev *ldev)
2977 {
2978 	if (ldev == NULL)
2979 		return;
2980 
2981 	blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2982 	blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2983 
2984 	kfree(ldev->disk_conf);
2985 	kfree(ldev);
2986 }
2987 
2988 static void drbd_free_one_sock(struct drbd_socket *ds)
2989 {
2990 	struct socket *s;
2991 	mutex_lock(&ds->mutex);
2992 	s = ds->socket;
2993 	ds->socket = NULL;
2994 	mutex_unlock(&ds->mutex);
2995 	if (s) {
2996 		/* so debugfs does not need to mutex_lock() */
2997 		synchronize_rcu();
2998 		kernel_sock_shutdown(s, SHUT_RDWR);
2999 		sock_release(s);
3000 	}
3001 }
3002 
3003 void drbd_free_sock(struct drbd_connection *connection)
3004 {
3005 	if (connection->data.socket)
3006 		drbd_free_one_sock(&connection->data);
3007 	if (connection->meta.socket)
3008 		drbd_free_one_sock(&connection->meta);
3009 }
3010 
3011 /* meta data management */
3012 
3013 void conn_md_sync(struct drbd_connection *connection)
3014 {
3015 	struct drbd_peer_device *peer_device;
3016 	int vnr;
3017 
3018 	rcu_read_lock();
3019 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3020 		struct drbd_device *device = peer_device->device;
3021 
3022 		kref_get(&device->kref);
3023 		rcu_read_unlock();
3024 		drbd_md_sync(device);
3025 		kref_put(&device->kref, drbd_destroy_device);
3026 		rcu_read_lock();
3027 	}
3028 	rcu_read_unlock();
3029 }
3030 
3031 /* aligned 4kByte */
3032 struct meta_data_on_disk {
3033 	u64 la_size_sect;      /* last agreed size. */
3034 	u64 uuid[UI_SIZE];   /* UUIDs. */
3035 	u64 device_uuid;
3036 	u64 reserved_u64_1;
3037 	u32 flags;             /* MDF */
3038 	u32 magic;
3039 	u32 md_size_sect;
3040 	u32 al_offset;         /* offset to this block */
3041 	u32 al_nr_extents;     /* important for restoring the AL (userspace) */
3042 	      /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3043 	u32 bm_offset;         /* offset to the bitmap, from here */
3044 	u32 bm_bytes_per_bit;  /* BM_BLOCK_SIZE */
3045 	u32 la_peer_max_bio_size;   /* last peer max_bio_size */
3046 
3047 	/* see al_tr_number_to_on_disk_sector() */
3048 	u32 al_stripes;
3049 	u32 al_stripe_size_4k;
3050 
3051 	u8 reserved_u8[4096 - (7*8 + 10*4)];
3052 } __packed;
3053 
3054 
3055 
3056 void drbd_md_write(struct drbd_device *device, void *b)
3057 {
3058 	struct meta_data_on_disk *buffer = b;
3059 	sector_t sector;
3060 	int i;
3061 
3062 	memset(buffer, 0, sizeof(*buffer));
3063 
3064 	buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3065 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3066 		buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3067 	buffer->flags = cpu_to_be32(device->ldev->md.flags);
3068 	buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3069 
3070 	buffer->md_size_sect  = cpu_to_be32(device->ldev->md.md_size_sect);
3071 	buffer->al_offset     = cpu_to_be32(device->ldev->md.al_offset);
3072 	buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3073 	buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3074 	buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3075 
3076 	buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3077 	buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3078 
3079 	buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3080 	buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3081 
3082 	D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3083 	sector = device->ldev->md.md_offset;
3084 
3085 	if (drbd_md_sync_page_io(device, device->ldev, sector, WRITE)) {
3086 		/* this was a try anyways ... */
3087 		drbd_err(device, "meta data update failed!\n");
3088 		drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3089 	}
3090 }
3091 
3092 /**
3093  * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3094  * @device:	DRBD device.
3095  */
3096 void drbd_md_sync(struct drbd_device *device)
3097 {
3098 	struct meta_data_on_disk *buffer;
3099 
3100 	/* Don't accidentally change the DRBD meta data layout. */
3101 	BUILD_BUG_ON(UI_SIZE != 4);
3102 	BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3103 
3104 	del_timer(&device->md_sync_timer);
3105 	/* timer may be rearmed by drbd_md_mark_dirty() now. */
3106 	if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3107 		return;
3108 
3109 	/* We use here D_FAILED and not D_ATTACHING because we try to write
3110 	 * metadata even if we detach due to a disk failure! */
3111 	if (!get_ldev_if_state(device, D_FAILED))
3112 		return;
3113 
3114 	buffer = drbd_md_get_buffer(device, __func__);
3115 	if (!buffer)
3116 		goto out;
3117 
3118 	drbd_md_write(device, buffer);
3119 
3120 	/* Update device->ldev->md.la_size_sect,
3121 	 * since we updated it on metadata. */
3122 	device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3123 
3124 	drbd_md_put_buffer(device);
3125 out:
3126 	put_ldev(device);
3127 }
3128 
3129 static int check_activity_log_stripe_size(struct drbd_device *device,
3130 		struct meta_data_on_disk *on_disk,
3131 		struct drbd_md *in_core)
3132 {
3133 	u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3134 	u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3135 	u64 al_size_4k;
3136 
3137 	/* both not set: default to old fixed size activity log */
3138 	if (al_stripes == 0 && al_stripe_size_4k == 0) {
3139 		al_stripes = 1;
3140 		al_stripe_size_4k = MD_32kB_SECT/8;
3141 	}
3142 
3143 	/* some paranoia plausibility checks */
3144 
3145 	/* we need both values to be set */
3146 	if (al_stripes == 0 || al_stripe_size_4k == 0)
3147 		goto err;
3148 
3149 	al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3150 
3151 	/* Upper limit of activity log area, to avoid potential overflow
3152 	 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3153 	 * than 72 * 4k blocks total only increases the amount of history,
3154 	 * limiting this arbitrarily to 16 GB is not a real limitation ;-)  */
3155 	if (al_size_4k > (16 * 1024 * 1024/4))
3156 		goto err;
3157 
3158 	/* Lower limit: we need at least 8 transaction slots (32kB)
3159 	 * to not break existing setups */
3160 	if (al_size_4k < MD_32kB_SECT/8)
3161 		goto err;
3162 
3163 	in_core->al_stripe_size_4k = al_stripe_size_4k;
3164 	in_core->al_stripes = al_stripes;
3165 	in_core->al_size_4k = al_size_4k;
3166 
3167 	return 0;
3168 err:
3169 	drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3170 			al_stripes, al_stripe_size_4k);
3171 	return -EINVAL;
3172 }
3173 
3174 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3175 {
3176 	sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3177 	struct drbd_md *in_core = &bdev->md;
3178 	s32 on_disk_al_sect;
3179 	s32 on_disk_bm_sect;
3180 
3181 	/* The on-disk size of the activity log, calculated from offsets, and
3182 	 * the size of the activity log calculated from the stripe settings,
3183 	 * should match.
3184 	 * Though we could relax this a bit: it is ok, if the striped activity log
3185 	 * fits in the available on-disk activity log size.
3186 	 * Right now, that would break how resize is implemented.
3187 	 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3188 	 * of possible unused padding space in the on disk layout. */
3189 	if (in_core->al_offset < 0) {
3190 		if (in_core->bm_offset > in_core->al_offset)
3191 			goto err;
3192 		on_disk_al_sect = -in_core->al_offset;
3193 		on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3194 	} else {
3195 		if (in_core->al_offset != MD_4kB_SECT)
3196 			goto err;
3197 		if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3198 			goto err;
3199 
3200 		on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3201 		on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3202 	}
3203 
3204 	/* old fixed size meta data is exactly that: fixed. */
3205 	if (in_core->meta_dev_idx >= 0) {
3206 		if (in_core->md_size_sect != MD_128MB_SECT
3207 		||  in_core->al_offset != MD_4kB_SECT
3208 		||  in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3209 		||  in_core->al_stripes != 1
3210 		||  in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3211 			goto err;
3212 	}
3213 
3214 	if (capacity < in_core->md_size_sect)
3215 		goto err;
3216 	if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3217 		goto err;
3218 
3219 	/* should be aligned, and at least 32k */
3220 	if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3221 		goto err;
3222 
3223 	/* should fit (for now: exactly) into the available on-disk space;
3224 	 * overflow prevention is in check_activity_log_stripe_size() above. */
3225 	if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3226 		goto err;
3227 
3228 	/* again, should be aligned */
3229 	if (in_core->bm_offset & 7)
3230 		goto err;
3231 
3232 	/* FIXME check for device grow with flex external meta data? */
3233 
3234 	/* can the available bitmap space cover the last agreed device size? */
3235 	if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3236 		goto err;
3237 
3238 	return 0;
3239 
3240 err:
3241 	drbd_err(device, "meta data offsets don't make sense: idx=%d "
3242 			"al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3243 			"md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3244 			in_core->meta_dev_idx,
3245 			in_core->al_stripes, in_core->al_stripe_size_4k,
3246 			in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3247 			(unsigned long long)in_core->la_size_sect,
3248 			(unsigned long long)capacity);
3249 
3250 	return -EINVAL;
3251 }
3252 
3253 
3254 /**
3255  * drbd_md_read() - Reads in the meta data super block
3256  * @device:	DRBD device.
3257  * @bdev:	Device from which the meta data should be read in.
3258  *
3259  * Return NO_ERROR on success, and an enum drbd_ret_code in case
3260  * something goes wrong.
3261  *
3262  * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3263  * even before @bdev is assigned to @device->ldev.
3264  */
3265 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3266 {
3267 	struct meta_data_on_disk *buffer;
3268 	u32 magic, flags;
3269 	int i, rv = NO_ERROR;
3270 
3271 	if (device->state.disk != D_DISKLESS)
3272 		return ERR_DISK_CONFIGURED;
3273 
3274 	buffer = drbd_md_get_buffer(device, __func__);
3275 	if (!buffer)
3276 		return ERR_NOMEM;
3277 
3278 	/* First, figure out where our meta data superblock is located,
3279 	 * and read it. */
3280 	bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3281 	bdev->md.md_offset = drbd_md_ss(bdev);
3282 
3283 	if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, READ)) {
3284 		/* NOTE: can't do normal error processing here as this is
3285 		   called BEFORE disk is attached */
3286 		drbd_err(device, "Error while reading metadata.\n");
3287 		rv = ERR_IO_MD_DISK;
3288 		goto err;
3289 	}
3290 
3291 	magic = be32_to_cpu(buffer->magic);
3292 	flags = be32_to_cpu(buffer->flags);
3293 	if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3294 	    (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3295 			/* btw: that's Activity Log clean, not "all" clean. */
3296 		drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3297 		rv = ERR_MD_UNCLEAN;
3298 		goto err;
3299 	}
3300 
3301 	rv = ERR_MD_INVALID;
3302 	if (magic != DRBD_MD_MAGIC_08) {
3303 		if (magic == DRBD_MD_MAGIC_07)
3304 			drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3305 		else
3306 			drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3307 		goto err;
3308 	}
3309 
3310 	if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3311 		drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3312 		    be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3313 		goto err;
3314 	}
3315 
3316 
3317 	/* convert to in_core endian */
3318 	bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3319 	for (i = UI_CURRENT; i < UI_SIZE; i++)
3320 		bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3321 	bdev->md.flags = be32_to_cpu(buffer->flags);
3322 	bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3323 
3324 	bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3325 	bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3326 	bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3327 
3328 	if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3329 		goto err;
3330 	if (check_offsets_and_sizes(device, bdev))
3331 		goto err;
3332 
3333 	if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3334 		drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3335 		    be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3336 		goto err;
3337 	}
3338 	if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3339 		drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3340 		    be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3341 		goto err;
3342 	}
3343 
3344 	rv = NO_ERROR;
3345 
3346 	spin_lock_irq(&device->resource->req_lock);
3347 	if (device->state.conn < C_CONNECTED) {
3348 		unsigned int peer;
3349 		peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3350 		peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3351 		device->peer_max_bio_size = peer;
3352 	}
3353 	spin_unlock_irq(&device->resource->req_lock);
3354 
3355  err:
3356 	drbd_md_put_buffer(device);
3357 
3358 	return rv;
3359 }
3360 
3361 /**
3362  * drbd_md_mark_dirty() - Mark meta data super block as dirty
3363  * @device:	DRBD device.
3364  *
3365  * Call this function if you change anything that should be written to
3366  * the meta-data super block. This function sets MD_DIRTY, and starts a
3367  * timer that ensures that within five seconds you have to call drbd_md_sync().
3368  */
3369 #ifdef DEBUG
3370 void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
3371 {
3372 	if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3373 		mod_timer(&device->md_sync_timer, jiffies + HZ);
3374 		device->last_md_mark_dirty.line = line;
3375 		device->last_md_mark_dirty.func = func;
3376 	}
3377 }
3378 #else
3379 void drbd_md_mark_dirty(struct drbd_device *device)
3380 {
3381 	if (!test_and_set_bit(MD_DIRTY, &device->flags))
3382 		mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3383 }
3384 #endif
3385 
3386 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3387 {
3388 	int i;
3389 
3390 	for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3391 		device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3392 }
3393 
3394 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3395 {
3396 	if (idx == UI_CURRENT) {
3397 		if (device->state.role == R_PRIMARY)
3398 			val |= 1;
3399 		else
3400 			val &= ~((u64)1);
3401 
3402 		drbd_set_ed_uuid(device, val);
3403 	}
3404 
3405 	device->ldev->md.uuid[idx] = val;
3406 	drbd_md_mark_dirty(device);
3407 }
3408 
3409 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3410 {
3411 	unsigned long flags;
3412 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3413 	__drbd_uuid_set(device, idx, val);
3414 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3415 }
3416 
3417 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3418 {
3419 	unsigned long flags;
3420 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3421 	if (device->ldev->md.uuid[idx]) {
3422 		drbd_uuid_move_history(device);
3423 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3424 	}
3425 	__drbd_uuid_set(device, idx, val);
3426 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3427 }
3428 
3429 /**
3430  * drbd_uuid_new_current() - Creates a new current UUID
3431  * @device:	DRBD device.
3432  *
3433  * Creates a new current UUID, and rotates the old current UUID into
3434  * the bitmap slot. Causes an incremental resync upon next connect.
3435  */
3436 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3437 {
3438 	u64 val;
3439 	unsigned long long bm_uuid;
3440 
3441 	get_random_bytes(&val, sizeof(u64));
3442 
3443 	spin_lock_irq(&device->ldev->md.uuid_lock);
3444 	bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3445 
3446 	if (bm_uuid)
3447 		drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3448 
3449 	device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3450 	__drbd_uuid_set(device, UI_CURRENT, val);
3451 	spin_unlock_irq(&device->ldev->md.uuid_lock);
3452 
3453 	drbd_print_uuids(device, "new current UUID");
3454 	/* get it to stable storage _now_ */
3455 	drbd_md_sync(device);
3456 }
3457 
3458 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3459 {
3460 	unsigned long flags;
3461 	if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3462 		return;
3463 
3464 	spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3465 	if (val == 0) {
3466 		drbd_uuid_move_history(device);
3467 		device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3468 		device->ldev->md.uuid[UI_BITMAP] = 0;
3469 	} else {
3470 		unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3471 		if (bm_uuid)
3472 			drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3473 
3474 		device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3475 	}
3476 	spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3477 
3478 	drbd_md_mark_dirty(device);
3479 }
3480 
3481 /**
3482  * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3483  * @device:	DRBD device.
3484  *
3485  * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3486  */
3487 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3488 {
3489 	int rv = -EIO;
3490 
3491 	drbd_md_set_flag(device, MDF_FULL_SYNC);
3492 	drbd_md_sync(device);
3493 	drbd_bm_set_all(device);
3494 
3495 	rv = drbd_bm_write(device);
3496 
3497 	if (!rv) {
3498 		drbd_md_clear_flag(device, MDF_FULL_SYNC);
3499 		drbd_md_sync(device);
3500 	}
3501 
3502 	return rv;
3503 }
3504 
3505 /**
3506  * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3507  * @device:	DRBD device.
3508  *
3509  * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3510  */
3511 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3512 {
3513 	drbd_resume_al(device);
3514 	drbd_bm_clear_all(device);
3515 	return drbd_bm_write(device);
3516 }
3517 
3518 static int w_bitmap_io(struct drbd_work *w, int unused)
3519 {
3520 	struct drbd_device *device =
3521 		container_of(w, struct drbd_device, bm_io_work.w);
3522 	struct bm_io_work *work = &device->bm_io_work;
3523 	int rv = -EIO;
3524 
3525 	D_ASSERT(device, atomic_read(&device->ap_bio_cnt) == 0);
3526 
3527 	if (get_ldev(device)) {
3528 		drbd_bm_lock(device, work->why, work->flags);
3529 		rv = work->io_fn(device);
3530 		drbd_bm_unlock(device);
3531 		put_ldev(device);
3532 	}
3533 
3534 	clear_bit_unlock(BITMAP_IO, &device->flags);
3535 	wake_up(&device->misc_wait);
3536 
3537 	if (work->done)
3538 		work->done(device, rv);
3539 
3540 	clear_bit(BITMAP_IO_QUEUED, &device->flags);
3541 	work->why = NULL;
3542 	work->flags = 0;
3543 
3544 	return 0;
3545 }
3546 
3547 /**
3548  * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3549  * @device:	DRBD device.
3550  * @io_fn:	IO callback to be called when bitmap IO is possible
3551  * @done:	callback to be called after the bitmap IO was performed
3552  * @why:	Descriptive text of the reason for doing the IO
3553  *
3554  * While IO on the bitmap happens we freeze application IO thus we ensure
3555  * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3556  * called from worker context. It MUST NOT be used while a previous such
3557  * work is still pending!
3558  *
3559  * Its worker function encloses the call of io_fn() by get_ldev() and
3560  * put_ldev().
3561  */
3562 void drbd_queue_bitmap_io(struct drbd_device *device,
3563 			  int (*io_fn)(struct drbd_device *),
3564 			  void (*done)(struct drbd_device *, int),
3565 			  char *why, enum bm_flag flags)
3566 {
3567 	D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3568 
3569 	D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3570 	D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3571 	D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3572 	if (device->bm_io_work.why)
3573 		drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3574 			why, device->bm_io_work.why);
3575 
3576 	device->bm_io_work.io_fn = io_fn;
3577 	device->bm_io_work.done = done;
3578 	device->bm_io_work.why = why;
3579 	device->bm_io_work.flags = flags;
3580 
3581 	spin_lock_irq(&device->resource->req_lock);
3582 	set_bit(BITMAP_IO, &device->flags);
3583 	if (atomic_read(&device->ap_bio_cnt) == 0) {
3584 		if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3585 			drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3586 					&device->bm_io_work.w);
3587 	}
3588 	spin_unlock_irq(&device->resource->req_lock);
3589 }
3590 
3591 /**
3592  * drbd_bitmap_io() -  Does an IO operation on the whole bitmap
3593  * @device:	DRBD device.
3594  * @io_fn:	IO callback to be called when bitmap IO is possible
3595  * @why:	Descriptive text of the reason for doing the IO
3596  *
3597  * freezes application IO while that the actual IO operations runs. This
3598  * functions MAY NOT be called from worker context.
3599  */
3600 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3601 		char *why, enum bm_flag flags)
3602 {
3603 	int rv;
3604 
3605 	D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3606 
3607 	if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3608 		drbd_suspend_io(device);
3609 
3610 	drbd_bm_lock(device, why, flags);
3611 	rv = io_fn(device);
3612 	drbd_bm_unlock(device);
3613 
3614 	if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3615 		drbd_resume_io(device);
3616 
3617 	return rv;
3618 }
3619 
3620 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3621 {
3622 	if ((device->ldev->md.flags & flag) != flag) {
3623 		drbd_md_mark_dirty(device);
3624 		device->ldev->md.flags |= flag;
3625 	}
3626 }
3627 
3628 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3629 {
3630 	if ((device->ldev->md.flags & flag) != 0) {
3631 		drbd_md_mark_dirty(device);
3632 		device->ldev->md.flags &= ~flag;
3633 	}
3634 }
3635 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3636 {
3637 	return (bdev->md.flags & flag) != 0;
3638 }
3639 
3640 static void md_sync_timer_fn(unsigned long data)
3641 {
3642 	struct drbd_device *device = (struct drbd_device *) data;
3643 	drbd_device_post_work(device, MD_SYNC);
3644 }
3645 
3646 const char *cmdname(enum drbd_packet cmd)
3647 {
3648 	/* THINK may need to become several global tables
3649 	 * when we want to support more than
3650 	 * one PRO_VERSION */
3651 	static const char *cmdnames[] = {
3652 		[P_DATA]	        = "Data",
3653 		[P_DATA_REPLY]	        = "DataReply",
3654 		[P_RS_DATA_REPLY]	= "RSDataReply",
3655 		[P_BARRIER]	        = "Barrier",
3656 		[P_BITMAP]	        = "ReportBitMap",
3657 		[P_BECOME_SYNC_TARGET]  = "BecomeSyncTarget",
3658 		[P_BECOME_SYNC_SOURCE]  = "BecomeSyncSource",
3659 		[P_UNPLUG_REMOTE]	= "UnplugRemote",
3660 		[P_DATA_REQUEST]	= "DataRequest",
3661 		[P_RS_DATA_REQUEST]     = "RSDataRequest",
3662 		[P_SYNC_PARAM]	        = "SyncParam",
3663 		[P_SYNC_PARAM89]	= "SyncParam89",
3664 		[P_PROTOCOL]            = "ReportProtocol",
3665 		[P_UUIDS]	        = "ReportUUIDs",
3666 		[P_SIZES]	        = "ReportSizes",
3667 		[P_STATE]	        = "ReportState",
3668 		[P_SYNC_UUID]           = "ReportSyncUUID",
3669 		[P_AUTH_CHALLENGE]      = "AuthChallenge",
3670 		[P_AUTH_RESPONSE]	= "AuthResponse",
3671 		[P_PING]		= "Ping",
3672 		[P_PING_ACK]	        = "PingAck",
3673 		[P_RECV_ACK]	        = "RecvAck",
3674 		[P_WRITE_ACK]	        = "WriteAck",
3675 		[P_RS_WRITE_ACK]	= "RSWriteAck",
3676 		[P_SUPERSEDED]          = "Superseded",
3677 		[P_NEG_ACK]	        = "NegAck",
3678 		[P_NEG_DREPLY]	        = "NegDReply",
3679 		[P_NEG_RS_DREPLY]	= "NegRSDReply",
3680 		[P_BARRIER_ACK]	        = "BarrierAck",
3681 		[P_STATE_CHG_REQ]       = "StateChgRequest",
3682 		[P_STATE_CHG_REPLY]     = "StateChgReply",
3683 		[P_OV_REQUEST]          = "OVRequest",
3684 		[P_OV_REPLY]            = "OVReply",
3685 		[P_OV_RESULT]           = "OVResult",
3686 		[P_CSUM_RS_REQUEST]     = "CsumRSRequest",
3687 		[P_RS_IS_IN_SYNC]	= "CsumRSIsInSync",
3688 		[P_COMPRESSED_BITMAP]   = "CBitmap",
3689 		[P_DELAY_PROBE]         = "DelayProbe",
3690 		[P_OUT_OF_SYNC]		= "OutOfSync",
3691 		[P_RETRY_WRITE]		= "RetryWrite",
3692 		[P_RS_CANCEL]		= "RSCancel",
3693 		[P_CONN_ST_CHG_REQ]	= "conn_st_chg_req",
3694 		[P_CONN_ST_CHG_REPLY]	= "conn_st_chg_reply",
3695 		[P_RETRY_WRITE]		= "retry_write",
3696 		[P_PROTOCOL_UPDATE]	= "protocol_update",
3697 
3698 		/* enum drbd_packet, but not commands - obsoleted flags:
3699 		 *	P_MAY_IGNORE
3700 		 *	P_MAX_OPT_CMD
3701 		 */
3702 	};
3703 
3704 	/* too big for the array: 0xfffX */
3705 	if (cmd == P_INITIAL_META)
3706 		return "InitialMeta";
3707 	if (cmd == P_INITIAL_DATA)
3708 		return "InitialData";
3709 	if (cmd == P_CONNECTION_FEATURES)
3710 		return "ConnectionFeatures";
3711 	if (cmd >= ARRAY_SIZE(cmdnames))
3712 		return "Unknown";
3713 	return cmdnames[cmd];
3714 }
3715 
3716 /**
3717  * drbd_wait_misc  -  wait for a request to make progress
3718  * @device:	device associated with the request
3719  * @i:		the struct drbd_interval embedded in struct drbd_request or
3720  *		struct drbd_peer_request
3721  */
3722 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3723 {
3724 	struct net_conf *nc;
3725 	DEFINE_WAIT(wait);
3726 	long timeout;
3727 
3728 	rcu_read_lock();
3729 	nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3730 	if (!nc) {
3731 		rcu_read_unlock();
3732 		return -ETIMEDOUT;
3733 	}
3734 	timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3735 	rcu_read_unlock();
3736 
3737 	/* Indicate to wake up device->misc_wait on progress.  */
3738 	i->waiting = true;
3739 	prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3740 	spin_unlock_irq(&device->resource->req_lock);
3741 	timeout = schedule_timeout(timeout);
3742 	finish_wait(&device->misc_wait, &wait);
3743 	spin_lock_irq(&device->resource->req_lock);
3744 	if (!timeout || device->state.conn < C_CONNECTED)
3745 		return -ETIMEDOUT;
3746 	if (signal_pending(current))
3747 		return -ERESTARTSYS;
3748 	return 0;
3749 }
3750 
3751 #ifdef CONFIG_DRBD_FAULT_INJECTION
3752 /* Fault insertion support including random number generator shamelessly
3753  * stolen from kernel/rcutorture.c */
3754 struct fault_random_state {
3755 	unsigned long state;
3756 	unsigned long count;
3757 };
3758 
3759 #define FAULT_RANDOM_MULT 39916801  /* prime */
3760 #define FAULT_RANDOM_ADD	479001701 /* prime */
3761 #define FAULT_RANDOM_REFRESH 10000
3762 
3763 /*
3764  * Crude but fast random-number generator.  Uses a linear congruential
3765  * generator, with occasional help from get_random_bytes().
3766  */
3767 static unsigned long
3768 _drbd_fault_random(struct fault_random_state *rsp)
3769 {
3770 	long refresh;
3771 
3772 	if (!rsp->count--) {
3773 		get_random_bytes(&refresh, sizeof(refresh));
3774 		rsp->state += refresh;
3775 		rsp->count = FAULT_RANDOM_REFRESH;
3776 	}
3777 	rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3778 	return swahw32(rsp->state);
3779 }
3780 
3781 static char *
3782 _drbd_fault_str(unsigned int type) {
3783 	static char *_faults[] = {
3784 		[DRBD_FAULT_MD_WR] = "Meta-data write",
3785 		[DRBD_FAULT_MD_RD] = "Meta-data read",
3786 		[DRBD_FAULT_RS_WR] = "Resync write",
3787 		[DRBD_FAULT_RS_RD] = "Resync read",
3788 		[DRBD_FAULT_DT_WR] = "Data write",
3789 		[DRBD_FAULT_DT_RD] = "Data read",
3790 		[DRBD_FAULT_DT_RA] = "Data read ahead",
3791 		[DRBD_FAULT_BM_ALLOC] = "BM allocation",
3792 		[DRBD_FAULT_AL_EE] = "EE allocation",
3793 		[DRBD_FAULT_RECEIVE] = "receive data corruption",
3794 	};
3795 
3796 	return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3797 }
3798 
3799 unsigned int
3800 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3801 {
3802 	static struct fault_random_state rrs = {0, 0};
3803 
3804 	unsigned int ret = (
3805 		(fault_devs == 0 ||
3806 			((1 << device_to_minor(device)) & fault_devs) != 0) &&
3807 		(((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3808 
3809 	if (ret) {
3810 		fault_count++;
3811 
3812 		if (__ratelimit(&drbd_ratelimit_state))
3813 			drbd_warn(device, "***Simulating %s failure\n",
3814 				_drbd_fault_str(type));
3815 	}
3816 
3817 	return ret;
3818 }
3819 #endif
3820 
3821 const char *drbd_buildtag(void)
3822 {
3823 	/* DRBD built from external sources has here a reference to the
3824 	   git hash of the source code. */
3825 
3826 	static char buildtag[38] = "\0uilt-in";
3827 
3828 	if (buildtag[0] == 0) {
3829 #ifdef MODULE
3830 		sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3831 #else
3832 		buildtag[0] = 'b';
3833 #endif
3834 	}
3835 
3836 	return buildtag;
3837 }
3838 
3839 module_init(drbd_init)
3840 module_exit(drbd_cleanup)
3841 
3842 EXPORT_SYMBOL(drbd_conn_str);
3843 EXPORT_SYMBOL(drbd_role_str);
3844 EXPORT_SYMBOL(drbd_disk_str);
3845 EXPORT_SYMBOL(drbd_set_st_err_str);
3846