sh_eth: fix EESIPR values for SH77{34|63}
[linux/fpc-iii.git] / drivers / block / drbd / drbd_main.c
blob83482721bc012739cf25ee627fd2b85b2fd094ab
1 /*
2 drbd.c
4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
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>.
10 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
11 from Logicworks, Inc. for making SDP replication support possible.
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.
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.
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.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/module.h>
32 #include <linux/jiffies.h>
33 #include <linux/drbd.h>
34 #include <linux/uaccess.h>
35 #include <asm/types.h>
36 #include <net/sock.h>
37 #include <linux/ctype.h>
38 #include <linux/mutex.h>
39 #include <linux/fs.h>
40 #include <linux/file.h>
41 #include <linux/proc_fs.h>
42 #include <linux/init.h>
43 #include <linux/mm.h>
44 #include <linux/memcontrol.h>
45 #include <linux/mm_inline.h>
46 #include <linux/slab.h>
47 #include <linux/random.h>
48 #include <linux/reboot.h>
49 #include <linux/notifier.h>
50 #include <linux/kthread.h>
51 #include <linux/workqueue.h>
52 #define __KERNEL_SYSCALLS__
53 #include <linux/unistd.h>
54 #include <linux/vmalloc.h>
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"
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);
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);
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);
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
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*/
109 /* Module parameter for setting the user mode helper program
110 * to run. Default is /sbin/drbdadm */
111 char usermode_helper[80] = "/sbin/drbdadm";
113 module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644);
115 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
116 * as member "struct gendisk *vdisk;"
118 struct idr drbd_devices;
119 struct list_head drbd_resources;
120 struct mutex resources_mutex;
122 struct kmem_cache *drbd_request_cache;
123 struct kmem_cache *drbd_ee_cache; /* peer requests */
124 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
125 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
126 mempool_t *drbd_request_mempool;
127 mempool_t *drbd_ee_mempool;
128 mempool_t *drbd_md_io_page_pool;
129 struct bio_set *drbd_md_io_bio_set;
131 /* I do not use a standard mempool, because:
132 1) I want to hand out the pre-allocated objects first.
133 2) I want to be able to interrupt sleeping allocation with a signal.
134 Note: This is a single linked list, the next pointer is the private
135 member of struct page.
137 struct page *drbd_pp_pool;
138 spinlock_t drbd_pp_lock;
139 int drbd_pp_vacant;
140 wait_queue_head_t drbd_pp_wait;
142 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
144 static const struct block_device_operations drbd_ops = {
145 .owner = THIS_MODULE,
146 .open = drbd_open,
147 .release = drbd_release,
150 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
152 struct bio *bio;
154 if (!drbd_md_io_bio_set)
155 return bio_alloc(gfp_mask, 1);
157 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
158 if (!bio)
159 return NULL;
160 return bio;
163 #ifdef __CHECKER__
164 /* When checking with sparse, and this is an inline function, sparse will
165 give tons of false positives. When this is a real functions sparse works.
167 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
169 int io_allowed;
171 atomic_inc(&device->local_cnt);
172 io_allowed = (device->state.disk >= mins);
173 if (!io_allowed) {
174 if (atomic_dec_and_test(&device->local_cnt))
175 wake_up(&device->misc_wait);
177 return io_allowed;
180 #endif
183 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
184 * @connection: DRBD connection.
185 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
186 * @set_size: Expected number of requests before that barrier.
188 * In case the passed barrier_nr or set_size does not match the oldest
189 * epoch of not yet barrier-acked requests, this function will cause a
190 * termination of the connection.
192 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
193 unsigned int set_size)
195 struct drbd_request *r;
196 struct drbd_request *req = NULL;
197 int expect_epoch = 0;
198 int expect_size = 0;
200 spin_lock_irq(&connection->resource->req_lock);
202 /* find oldest not yet barrier-acked write request,
203 * count writes in its epoch. */
204 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
205 const unsigned s = r->rq_state;
206 if (!req) {
207 if (!(s & RQ_WRITE))
208 continue;
209 if (!(s & RQ_NET_MASK))
210 continue;
211 if (s & RQ_NET_DONE)
212 continue;
213 req = r;
214 expect_epoch = req->epoch;
215 expect_size ++;
216 } else {
217 if (r->epoch != expect_epoch)
218 break;
219 if (!(s & RQ_WRITE))
220 continue;
221 /* if (s & RQ_DONE): not expected */
222 /* if (!(s & RQ_NET_MASK)): not expected */
223 expect_size++;
227 /* first some paranoia code */
228 if (req == NULL) {
229 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
230 barrier_nr);
231 goto bail;
233 if (expect_epoch != barrier_nr) {
234 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
235 barrier_nr, expect_epoch);
236 goto bail;
239 if (expect_size != set_size) {
240 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
241 barrier_nr, set_size, expect_size);
242 goto bail;
245 /* Clean up list of requests processed during current epoch. */
246 /* this extra list walk restart is paranoia,
247 * to catch requests being barrier-acked "unexpectedly".
248 * It usually should find the same req again, or some READ preceding it. */
249 list_for_each_entry(req, &connection->transfer_log, tl_requests)
250 if (req->epoch == expect_epoch)
251 break;
252 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
253 if (req->epoch != expect_epoch)
254 break;
255 _req_mod(req, BARRIER_ACKED);
257 spin_unlock_irq(&connection->resource->req_lock);
259 return;
261 bail:
262 spin_unlock_irq(&connection->resource->req_lock);
263 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
268 * _tl_restart() - Walks the transfer log, and applies an action to all requests
269 * @connection: DRBD connection to operate on.
270 * @what: The action/event to perform with all request objects
272 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
273 * RESTART_FROZEN_DISK_IO.
275 /* must hold resource->req_lock */
276 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
278 struct drbd_request *req, *r;
280 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
281 _req_mod(req, what);
284 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
286 spin_lock_irq(&connection->resource->req_lock);
287 _tl_restart(connection, what);
288 spin_unlock_irq(&connection->resource->req_lock);
292 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
293 * @device: DRBD device.
295 * This is called after the connection to the peer was lost. The storage covered
296 * by the requests on the transfer gets marked as our of sync. Called from the
297 * receiver thread and the worker thread.
299 void tl_clear(struct drbd_connection *connection)
301 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
305 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
306 * @device: DRBD device.
308 void tl_abort_disk_io(struct drbd_device *device)
310 struct drbd_connection *connection = first_peer_device(device)->connection;
311 struct drbd_request *req, *r;
313 spin_lock_irq(&connection->resource->req_lock);
314 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
315 if (!(req->rq_state & RQ_LOCAL_PENDING))
316 continue;
317 if (req->device != device)
318 continue;
319 _req_mod(req, ABORT_DISK_IO);
321 spin_unlock_irq(&connection->resource->req_lock);
324 static int drbd_thread_setup(void *arg)
326 struct drbd_thread *thi = (struct drbd_thread *) arg;
327 struct drbd_resource *resource = thi->resource;
328 unsigned long flags;
329 int retval;
331 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
332 thi->name[0],
333 resource->name);
335 restart:
336 retval = thi->function(thi);
338 spin_lock_irqsave(&thi->t_lock, flags);
340 /* if the receiver has been "EXITING", the last thing it did
341 * was set the conn state to "StandAlone",
342 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
343 * and receiver thread will be "started".
344 * drbd_thread_start needs to set "RESTARTING" in that case.
345 * t_state check and assignment needs to be within the same spinlock,
346 * so either thread_start sees EXITING, and can remap to RESTARTING,
347 * or thread_start see NONE, and can proceed as normal.
350 if (thi->t_state == RESTARTING) {
351 drbd_info(resource, "Restarting %s thread\n", thi->name);
352 thi->t_state = RUNNING;
353 spin_unlock_irqrestore(&thi->t_lock, flags);
354 goto restart;
357 thi->task = NULL;
358 thi->t_state = NONE;
359 smp_mb();
360 complete_all(&thi->stop);
361 spin_unlock_irqrestore(&thi->t_lock, flags);
363 drbd_info(resource, "Terminating %s\n", current->comm);
365 /* Release mod reference taken when thread was started */
367 if (thi->connection)
368 kref_put(&thi->connection->kref, drbd_destroy_connection);
369 kref_put(&resource->kref, drbd_destroy_resource);
370 module_put(THIS_MODULE);
371 return retval;
374 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
375 int (*func) (struct drbd_thread *), const char *name)
377 spin_lock_init(&thi->t_lock);
378 thi->task = NULL;
379 thi->t_state = NONE;
380 thi->function = func;
381 thi->resource = resource;
382 thi->connection = NULL;
383 thi->name = name;
386 int drbd_thread_start(struct drbd_thread *thi)
388 struct drbd_resource *resource = thi->resource;
389 struct task_struct *nt;
390 unsigned long flags;
392 /* is used from state engine doing drbd_thread_stop_nowait,
393 * while holding the req lock irqsave */
394 spin_lock_irqsave(&thi->t_lock, flags);
396 switch (thi->t_state) {
397 case NONE:
398 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
399 thi->name, current->comm, current->pid);
401 /* Get ref on module for thread - this is released when thread exits */
402 if (!try_module_get(THIS_MODULE)) {
403 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
404 spin_unlock_irqrestore(&thi->t_lock, flags);
405 return false;
408 kref_get(&resource->kref);
409 if (thi->connection)
410 kref_get(&thi->connection->kref);
412 init_completion(&thi->stop);
413 thi->reset_cpu_mask = 1;
414 thi->t_state = RUNNING;
415 spin_unlock_irqrestore(&thi->t_lock, flags);
416 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
418 nt = kthread_create(drbd_thread_setup, (void *) thi,
419 "drbd_%c_%s", thi->name[0], thi->resource->name);
421 if (IS_ERR(nt)) {
422 drbd_err(resource, "Couldn't start thread\n");
424 if (thi->connection)
425 kref_put(&thi->connection->kref, drbd_destroy_connection);
426 kref_put(&resource->kref, drbd_destroy_resource);
427 module_put(THIS_MODULE);
428 return false;
430 spin_lock_irqsave(&thi->t_lock, flags);
431 thi->task = nt;
432 thi->t_state = RUNNING;
433 spin_unlock_irqrestore(&thi->t_lock, flags);
434 wake_up_process(nt);
435 break;
436 case EXITING:
437 thi->t_state = RESTARTING;
438 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
439 thi->name, current->comm, current->pid);
440 /* fall through */
441 case RUNNING:
442 case RESTARTING:
443 default:
444 spin_unlock_irqrestore(&thi->t_lock, flags);
445 break;
448 return true;
452 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
454 unsigned long flags;
456 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
458 /* may be called from state engine, holding the req lock irqsave */
459 spin_lock_irqsave(&thi->t_lock, flags);
461 if (thi->t_state == NONE) {
462 spin_unlock_irqrestore(&thi->t_lock, flags);
463 if (restart)
464 drbd_thread_start(thi);
465 return;
468 if (thi->t_state != ns) {
469 if (thi->task == NULL) {
470 spin_unlock_irqrestore(&thi->t_lock, flags);
471 return;
474 thi->t_state = ns;
475 smp_mb();
476 init_completion(&thi->stop);
477 if (thi->task != current)
478 force_sig(DRBD_SIGKILL, thi->task);
481 spin_unlock_irqrestore(&thi->t_lock, flags);
483 if (wait)
484 wait_for_completion(&thi->stop);
487 int conn_lowest_minor(struct drbd_connection *connection)
489 struct drbd_peer_device *peer_device;
490 int vnr = 0, minor = -1;
492 rcu_read_lock();
493 peer_device = idr_get_next(&connection->peer_devices, &vnr);
494 if (peer_device)
495 minor = device_to_minor(peer_device->device);
496 rcu_read_unlock();
498 return minor;
501 #ifdef CONFIG_SMP
503 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
505 * Forces all threads of a resource onto the same CPU. This is beneficial for
506 * DRBD's performance. May be overwritten by user's configuration.
508 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
510 unsigned int *resources_per_cpu, min_index = ~0;
512 resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL);
513 if (resources_per_cpu) {
514 struct drbd_resource *resource;
515 unsigned int cpu, min = ~0;
517 rcu_read_lock();
518 for_each_resource_rcu(resource, &drbd_resources) {
519 for_each_cpu(cpu, resource->cpu_mask)
520 resources_per_cpu[cpu]++;
522 rcu_read_unlock();
523 for_each_online_cpu(cpu) {
524 if (resources_per_cpu[cpu] < min) {
525 min = resources_per_cpu[cpu];
526 min_index = cpu;
529 kfree(resources_per_cpu);
531 if (min_index == ~0) {
532 cpumask_setall(*cpu_mask);
533 return;
535 cpumask_set_cpu(min_index, *cpu_mask);
539 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
540 * @device: DRBD device.
541 * @thi: drbd_thread object
543 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
544 * prematurely.
546 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
548 struct drbd_resource *resource = thi->resource;
549 struct task_struct *p = current;
551 if (!thi->reset_cpu_mask)
552 return;
553 thi->reset_cpu_mask = 0;
554 set_cpus_allowed_ptr(p, resource->cpu_mask);
556 #else
557 #define drbd_calc_cpu_mask(A) ({})
558 #endif
561 * drbd_header_size - size of a packet header
563 * The header size is a multiple of 8, so any payload following the header is
564 * word aligned on 64-bit architectures. (The bitmap send and receive code
565 * relies on this.)
567 unsigned int drbd_header_size(struct drbd_connection *connection)
569 if (connection->agreed_pro_version >= 100) {
570 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
571 return sizeof(struct p_header100);
572 } else {
573 BUILD_BUG_ON(sizeof(struct p_header80) !=
574 sizeof(struct p_header95));
575 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
576 return sizeof(struct p_header80);
580 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
582 h->magic = cpu_to_be32(DRBD_MAGIC);
583 h->command = cpu_to_be16(cmd);
584 h->length = cpu_to_be16(size);
585 return sizeof(struct p_header80);
588 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
590 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
591 h->command = cpu_to_be16(cmd);
592 h->length = cpu_to_be32(size);
593 return sizeof(struct p_header95);
596 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
597 int size, int vnr)
599 h->magic = cpu_to_be32(DRBD_MAGIC_100);
600 h->volume = cpu_to_be16(vnr);
601 h->command = cpu_to_be16(cmd);
602 h->length = cpu_to_be32(size);
603 h->pad = 0;
604 return sizeof(struct p_header100);
607 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
608 void *buffer, enum drbd_packet cmd, int size)
610 if (connection->agreed_pro_version >= 100)
611 return prepare_header100(buffer, cmd, size, vnr);
612 else if (connection->agreed_pro_version >= 95 &&
613 size > DRBD_MAX_SIZE_H80_PACKET)
614 return prepare_header95(buffer, cmd, size);
615 else
616 return prepare_header80(buffer, cmd, size);
619 static void *__conn_prepare_command(struct drbd_connection *connection,
620 struct drbd_socket *sock)
622 if (!sock->socket)
623 return NULL;
624 return sock->sbuf + drbd_header_size(connection);
627 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
629 void *p;
631 mutex_lock(&sock->mutex);
632 p = __conn_prepare_command(connection, sock);
633 if (!p)
634 mutex_unlock(&sock->mutex);
636 return p;
639 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
641 return conn_prepare_command(peer_device->connection, sock);
644 static int __send_command(struct drbd_connection *connection, int vnr,
645 struct drbd_socket *sock, enum drbd_packet cmd,
646 unsigned int header_size, void *data,
647 unsigned int size)
649 int msg_flags;
650 int err;
653 * Called with @data == NULL and the size of the data blocks in @size
654 * for commands that send data blocks. For those commands, omit the
655 * MSG_MORE flag: this will increase the likelihood that data blocks
656 * which are page aligned on the sender will end up page aligned on the
657 * receiver.
659 msg_flags = data ? MSG_MORE : 0;
661 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
662 header_size + size);
663 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
664 msg_flags);
665 if (data && !err)
666 err = drbd_send_all(connection, sock->socket, data, size, 0);
667 /* DRBD protocol "pings" are latency critical.
668 * This is supposed to trigger tcp_push_pending_frames() */
669 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
670 drbd_tcp_nodelay(sock->socket);
672 return err;
675 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
676 enum drbd_packet cmd, unsigned int header_size,
677 void *data, unsigned int size)
679 return __send_command(connection, 0, sock, cmd, header_size, data, size);
682 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
683 enum drbd_packet cmd, unsigned int header_size,
684 void *data, unsigned int size)
686 int err;
688 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
689 mutex_unlock(&sock->mutex);
690 return err;
693 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
694 enum drbd_packet cmd, unsigned int header_size,
695 void *data, unsigned int size)
697 int err;
699 err = __send_command(peer_device->connection, peer_device->device->vnr,
700 sock, cmd, header_size, data, size);
701 mutex_unlock(&sock->mutex);
702 return err;
705 int drbd_send_ping(struct drbd_connection *connection)
707 struct drbd_socket *sock;
709 sock = &connection->meta;
710 if (!conn_prepare_command(connection, sock))
711 return -EIO;
712 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
715 int drbd_send_ping_ack(struct drbd_connection *connection)
717 struct drbd_socket *sock;
719 sock = &connection->meta;
720 if (!conn_prepare_command(connection, sock))
721 return -EIO;
722 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
725 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
727 struct drbd_socket *sock;
728 struct p_rs_param_95 *p;
729 int size;
730 const int apv = peer_device->connection->agreed_pro_version;
731 enum drbd_packet cmd;
732 struct net_conf *nc;
733 struct disk_conf *dc;
735 sock = &peer_device->connection->data;
736 p = drbd_prepare_command(peer_device, sock);
737 if (!p)
738 return -EIO;
740 rcu_read_lock();
741 nc = rcu_dereference(peer_device->connection->net_conf);
743 size = apv <= 87 ? sizeof(struct p_rs_param)
744 : apv == 88 ? sizeof(struct p_rs_param)
745 + strlen(nc->verify_alg) + 1
746 : apv <= 94 ? sizeof(struct p_rs_param_89)
747 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
749 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
751 /* initialize verify_alg and csums_alg */
752 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
754 if (get_ldev(peer_device->device)) {
755 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
756 p->resync_rate = cpu_to_be32(dc->resync_rate);
757 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
758 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
759 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
760 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
761 put_ldev(peer_device->device);
762 } else {
763 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
764 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
765 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
766 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
767 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
770 if (apv >= 88)
771 strcpy(p->verify_alg, nc->verify_alg);
772 if (apv >= 89)
773 strcpy(p->csums_alg, nc->csums_alg);
774 rcu_read_unlock();
776 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
779 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
781 struct drbd_socket *sock;
782 struct p_protocol *p;
783 struct net_conf *nc;
784 int size, cf;
786 sock = &connection->data;
787 p = __conn_prepare_command(connection, sock);
788 if (!p)
789 return -EIO;
791 rcu_read_lock();
792 nc = rcu_dereference(connection->net_conf);
794 if (nc->tentative && connection->agreed_pro_version < 92) {
795 rcu_read_unlock();
796 mutex_unlock(&sock->mutex);
797 drbd_err(connection, "--dry-run is not supported by peer");
798 return -EOPNOTSUPP;
801 size = sizeof(*p);
802 if (connection->agreed_pro_version >= 87)
803 size += strlen(nc->integrity_alg) + 1;
805 p->protocol = cpu_to_be32(nc->wire_protocol);
806 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
807 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
808 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
809 p->two_primaries = cpu_to_be32(nc->two_primaries);
810 cf = 0;
811 if (nc->discard_my_data)
812 cf |= CF_DISCARD_MY_DATA;
813 if (nc->tentative)
814 cf |= CF_DRY_RUN;
815 p->conn_flags = cpu_to_be32(cf);
817 if (connection->agreed_pro_version >= 87)
818 strcpy(p->integrity_alg, nc->integrity_alg);
819 rcu_read_unlock();
821 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
824 int drbd_send_protocol(struct drbd_connection *connection)
826 int err;
828 mutex_lock(&connection->data.mutex);
829 err = __drbd_send_protocol(connection, P_PROTOCOL);
830 mutex_unlock(&connection->data.mutex);
832 return err;
835 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
837 struct drbd_device *device = peer_device->device;
838 struct drbd_socket *sock;
839 struct p_uuids *p;
840 int i;
842 if (!get_ldev_if_state(device, D_NEGOTIATING))
843 return 0;
845 sock = &peer_device->connection->data;
846 p = drbd_prepare_command(peer_device, sock);
847 if (!p) {
848 put_ldev(device);
849 return -EIO;
851 spin_lock_irq(&device->ldev->md.uuid_lock);
852 for (i = UI_CURRENT; i < UI_SIZE; i++)
853 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
854 spin_unlock_irq(&device->ldev->md.uuid_lock);
856 device->comm_bm_set = drbd_bm_total_weight(device);
857 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
858 rcu_read_lock();
859 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
860 rcu_read_unlock();
861 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
862 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
863 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
865 put_ldev(device);
866 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
869 int drbd_send_uuids(struct drbd_peer_device *peer_device)
871 return _drbd_send_uuids(peer_device, 0);
874 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
876 return _drbd_send_uuids(peer_device, 8);
879 void drbd_print_uuids(struct drbd_device *device, const char *text)
881 if (get_ldev_if_state(device, D_NEGOTIATING)) {
882 u64 *uuid = device->ldev->md.uuid;
883 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
884 text,
885 (unsigned long long)uuid[UI_CURRENT],
886 (unsigned long long)uuid[UI_BITMAP],
887 (unsigned long long)uuid[UI_HISTORY_START],
888 (unsigned long long)uuid[UI_HISTORY_END]);
889 put_ldev(device);
890 } else {
891 drbd_info(device, "%s effective data uuid: %016llX\n",
892 text,
893 (unsigned long long)device->ed_uuid);
897 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
899 struct drbd_device *device = peer_device->device;
900 struct drbd_socket *sock;
901 struct p_rs_uuid *p;
902 u64 uuid;
904 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
906 uuid = device->ldev->md.uuid[UI_BITMAP];
907 if (uuid && uuid != UUID_JUST_CREATED)
908 uuid = uuid + UUID_NEW_BM_OFFSET;
909 else
910 get_random_bytes(&uuid, sizeof(u64));
911 drbd_uuid_set(device, UI_BITMAP, uuid);
912 drbd_print_uuids(device, "updated sync UUID");
913 drbd_md_sync(device);
915 sock = &peer_device->connection->data;
916 p = drbd_prepare_command(peer_device, sock);
917 if (p) {
918 p->uuid = cpu_to_be64(uuid);
919 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
923 /* communicated if (agreed_features & DRBD_FF_WSAME) */
924 void assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p, struct request_queue *q)
926 if (q) {
927 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
928 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
929 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
930 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
931 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
932 p->qlim->discard_enabled = blk_queue_discard(q);
933 p->qlim->discard_zeroes_data = queue_discard_zeroes_data(q);
934 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
935 } else {
936 q = device->rq_queue;
937 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
938 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
939 p->qlim->alignment_offset = 0;
940 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
941 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
942 p->qlim->discard_enabled = 0;
943 p->qlim->discard_zeroes_data = 0;
944 p->qlim->write_same_capable = 0;
948 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
950 struct drbd_device *device = peer_device->device;
951 struct drbd_socket *sock;
952 struct p_sizes *p;
953 sector_t d_size, u_size;
954 int q_order_type;
955 unsigned int max_bio_size;
956 unsigned int packet_size;
958 sock = &peer_device->connection->data;
959 p = drbd_prepare_command(peer_device, sock);
960 if (!p)
961 return -EIO;
963 packet_size = sizeof(*p);
964 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
965 packet_size += sizeof(p->qlim[0]);
967 memset(p, 0, packet_size);
968 if (get_ldev_if_state(device, D_NEGOTIATING)) {
969 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
970 d_size = drbd_get_max_capacity(device->ldev);
971 rcu_read_lock();
972 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
973 rcu_read_unlock();
974 q_order_type = drbd_queue_order_type(device);
975 max_bio_size = queue_max_hw_sectors(q) << 9;
976 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
977 assign_p_sizes_qlim(device, p, q);
978 put_ldev(device);
979 } else {
980 d_size = 0;
981 u_size = 0;
982 q_order_type = QUEUE_ORDERED_NONE;
983 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
984 assign_p_sizes_qlim(device, p, NULL);
987 if (peer_device->connection->agreed_pro_version <= 94)
988 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
989 else if (peer_device->connection->agreed_pro_version < 100)
990 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
992 p->d_size = cpu_to_be64(d_size);
993 p->u_size = cpu_to_be64(u_size);
994 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
995 p->max_bio_size = cpu_to_be32(max_bio_size);
996 p->queue_order_type = cpu_to_be16(q_order_type);
997 p->dds_flags = cpu_to_be16(flags);
999 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
1003 * drbd_send_current_state() - Sends the drbd state to the peer
1004 * @peer_device: DRBD peer device.
1006 int drbd_send_current_state(struct drbd_peer_device *peer_device)
1008 struct drbd_socket *sock;
1009 struct p_state *p;
1011 sock = &peer_device->connection->data;
1012 p = drbd_prepare_command(peer_device, sock);
1013 if (!p)
1014 return -EIO;
1015 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1016 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1020 * drbd_send_state() - After a state change, sends the new state to the peer
1021 * @peer_device: DRBD peer device.
1022 * @state: the state to send, not necessarily the current state.
1024 * Each state change queues an "after_state_ch" work, which will eventually
1025 * send the resulting new state to the peer. If more state changes happen
1026 * between queuing and processing of the after_state_ch work, we still
1027 * want to send each intermediary state in the order it occurred.
1029 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1031 struct drbd_socket *sock;
1032 struct p_state *p;
1034 sock = &peer_device->connection->data;
1035 p = drbd_prepare_command(peer_device, sock);
1036 if (!p)
1037 return -EIO;
1038 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1039 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1042 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1044 struct drbd_socket *sock;
1045 struct p_req_state *p;
1047 sock = &peer_device->connection->data;
1048 p = drbd_prepare_command(peer_device, sock);
1049 if (!p)
1050 return -EIO;
1051 p->mask = cpu_to_be32(mask.i);
1052 p->val = cpu_to_be32(val.i);
1053 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1056 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1058 enum drbd_packet cmd;
1059 struct drbd_socket *sock;
1060 struct p_req_state *p;
1062 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1063 sock = &connection->data;
1064 p = conn_prepare_command(connection, sock);
1065 if (!p)
1066 return -EIO;
1067 p->mask = cpu_to_be32(mask.i);
1068 p->val = cpu_to_be32(val.i);
1069 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1072 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1074 struct drbd_socket *sock;
1075 struct p_req_state_reply *p;
1077 sock = &peer_device->connection->meta;
1078 p = drbd_prepare_command(peer_device, sock);
1079 if (p) {
1080 p->retcode = cpu_to_be32(retcode);
1081 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1085 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1087 struct drbd_socket *sock;
1088 struct p_req_state_reply *p;
1089 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1091 sock = &connection->meta;
1092 p = conn_prepare_command(connection, sock);
1093 if (p) {
1094 p->retcode = cpu_to_be32(retcode);
1095 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1099 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1101 BUG_ON(code & ~0xf);
1102 p->encoding = (p->encoding & ~0xf) | code;
1105 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1107 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1110 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1112 BUG_ON(n & ~0x7);
1113 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1116 static int fill_bitmap_rle_bits(struct drbd_device *device,
1117 struct p_compressed_bm *p,
1118 unsigned int size,
1119 struct bm_xfer_ctx *c)
1121 struct bitstream bs;
1122 unsigned long plain_bits;
1123 unsigned long tmp;
1124 unsigned long rl;
1125 unsigned len;
1126 unsigned toggle;
1127 int bits, use_rle;
1129 /* may we use this feature? */
1130 rcu_read_lock();
1131 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1132 rcu_read_unlock();
1133 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1134 return 0;
1136 if (c->bit_offset >= c->bm_bits)
1137 return 0; /* nothing to do. */
1139 /* use at most thus many bytes */
1140 bitstream_init(&bs, p->code, size, 0);
1141 memset(p->code, 0, size);
1142 /* plain bits covered in this code string */
1143 plain_bits = 0;
1145 /* p->encoding & 0x80 stores whether the first run length is set.
1146 * bit offset is implicit.
1147 * start with toggle == 2 to be able to tell the first iteration */
1148 toggle = 2;
1150 /* see how much plain bits we can stuff into one packet
1151 * using RLE and VLI. */
1152 do {
1153 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1154 : _drbd_bm_find_next(device, c->bit_offset);
1155 if (tmp == -1UL)
1156 tmp = c->bm_bits;
1157 rl = tmp - c->bit_offset;
1159 if (toggle == 2) { /* first iteration */
1160 if (rl == 0) {
1161 /* the first checked bit was set,
1162 * store start value, */
1163 dcbp_set_start(p, 1);
1164 /* but skip encoding of zero run length */
1165 toggle = !toggle;
1166 continue;
1168 dcbp_set_start(p, 0);
1171 /* paranoia: catch zero runlength.
1172 * can only happen if bitmap is modified while we scan it. */
1173 if (rl == 0) {
1174 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1175 "t:%u bo:%lu\n", toggle, c->bit_offset);
1176 return -1;
1179 bits = vli_encode_bits(&bs, rl);
1180 if (bits == -ENOBUFS) /* buffer full */
1181 break;
1182 if (bits <= 0) {
1183 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1184 return 0;
1187 toggle = !toggle;
1188 plain_bits += rl;
1189 c->bit_offset = tmp;
1190 } while (c->bit_offset < c->bm_bits);
1192 len = bs.cur.b - p->code + !!bs.cur.bit;
1194 if (plain_bits < (len << 3)) {
1195 /* incompressible with this method.
1196 * we need to rewind both word and bit position. */
1197 c->bit_offset -= plain_bits;
1198 bm_xfer_ctx_bit_to_word_offset(c);
1199 c->bit_offset = c->word_offset * BITS_PER_LONG;
1200 return 0;
1203 /* RLE + VLI was able to compress it just fine.
1204 * update c->word_offset. */
1205 bm_xfer_ctx_bit_to_word_offset(c);
1207 /* store pad_bits */
1208 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1210 return len;
1214 * send_bitmap_rle_or_plain
1216 * Return 0 when done, 1 when another iteration is needed, and a negative error
1217 * code upon failure.
1219 static int
1220 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1222 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1223 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1224 struct p_compressed_bm *p = sock->sbuf + header_size;
1225 int len, err;
1227 len = fill_bitmap_rle_bits(device, p,
1228 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1229 if (len < 0)
1230 return -EIO;
1232 if (len) {
1233 dcbp_set_code(p, RLE_VLI_Bits);
1234 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1235 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1236 NULL, 0);
1237 c->packets[0]++;
1238 c->bytes[0] += header_size + sizeof(*p) + len;
1240 if (c->bit_offset >= c->bm_bits)
1241 len = 0; /* DONE */
1242 } else {
1243 /* was not compressible.
1244 * send a buffer full of plain text bits instead. */
1245 unsigned int data_size;
1246 unsigned long num_words;
1247 unsigned long *p = sock->sbuf + header_size;
1249 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1250 num_words = min_t(size_t, data_size / sizeof(*p),
1251 c->bm_words - c->word_offset);
1252 len = num_words * sizeof(*p);
1253 if (len)
1254 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1255 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1256 c->word_offset += num_words;
1257 c->bit_offset = c->word_offset * BITS_PER_LONG;
1259 c->packets[1]++;
1260 c->bytes[1] += header_size + len;
1262 if (c->bit_offset > c->bm_bits)
1263 c->bit_offset = c->bm_bits;
1265 if (!err) {
1266 if (len == 0) {
1267 INFO_bm_xfer_stats(device, "send", c);
1268 return 0;
1269 } else
1270 return 1;
1272 return -EIO;
1275 /* See the comment at receive_bitmap() */
1276 static int _drbd_send_bitmap(struct drbd_device *device)
1278 struct bm_xfer_ctx c;
1279 int err;
1281 if (!expect(device->bitmap))
1282 return false;
1284 if (get_ldev(device)) {
1285 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1286 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1287 drbd_bm_set_all(device);
1288 if (drbd_bm_write(device)) {
1289 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1290 * but otherwise process as per normal - need to tell other
1291 * side that a full resync is required! */
1292 drbd_err(device, "Failed to write bitmap to disk!\n");
1293 } else {
1294 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1295 drbd_md_sync(device);
1298 put_ldev(device);
1301 c = (struct bm_xfer_ctx) {
1302 .bm_bits = drbd_bm_bits(device),
1303 .bm_words = drbd_bm_words(device),
1306 do {
1307 err = send_bitmap_rle_or_plain(device, &c);
1308 } while (err > 0);
1310 return err == 0;
1313 int drbd_send_bitmap(struct drbd_device *device)
1315 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1316 int err = -1;
1318 mutex_lock(&sock->mutex);
1319 if (sock->socket)
1320 err = !_drbd_send_bitmap(device);
1321 mutex_unlock(&sock->mutex);
1322 return err;
1325 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1327 struct drbd_socket *sock;
1328 struct p_barrier_ack *p;
1330 if (connection->cstate < C_WF_REPORT_PARAMS)
1331 return;
1333 sock = &connection->meta;
1334 p = conn_prepare_command(connection, sock);
1335 if (!p)
1336 return;
1337 p->barrier = barrier_nr;
1338 p->set_size = cpu_to_be32(set_size);
1339 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1343 * _drbd_send_ack() - Sends an ack packet
1344 * @device: DRBD device.
1345 * @cmd: Packet command code.
1346 * @sector: sector, needs to be in big endian byte order
1347 * @blksize: size in byte, needs to be in big endian byte order
1348 * @block_id: Id, big endian byte order
1350 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1351 u64 sector, u32 blksize, u64 block_id)
1353 struct drbd_socket *sock;
1354 struct p_block_ack *p;
1356 if (peer_device->device->state.conn < C_CONNECTED)
1357 return -EIO;
1359 sock = &peer_device->connection->meta;
1360 p = drbd_prepare_command(peer_device, sock);
1361 if (!p)
1362 return -EIO;
1363 p->sector = sector;
1364 p->block_id = block_id;
1365 p->blksize = blksize;
1366 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1367 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1370 /* dp->sector and dp->block_id already/still in network byte order,
1371 * data_size is payload size according to dp->head,
1372 * and may need to be corrected for digest size. */
1373 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1374 struct p_data *dp, int data_size)
1376 if (peer_device->connection->peer_integrity_tfm)
1377 data_size -= crypto_ahash_digestsize(peer_device->connection->peer_integrity_tfm);
1378 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1379 dp->block_id);
1382 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1383 struct p_block_req *rp)
1385 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1389 * drbd_send_ack() - Sends an ack packet
1390 * @device: DRBD device
1391 * @cmd: packet command code
1392 * @peer_req: peer request
1394 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1395 struct drbd_peer_request *peer_req)
1397 return _drbd_send_ack(peer_device, cmd,
1398 cpu_to_be64(peer_req->i.sector),
1399 cpu_to_be32(peer_req->i.size),
1400 peer_req->block_id);
1403 /* This function misuses the block_id field to signal if the blocks
1404 * are is sync or not. */
1405 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1406 sector_t sector, int blksize, u64 block_id)
1408 return _drbd_send_ack(peer_device, cmd,
1409 cpu_to_be64(sector),
1410 cpu_to_be32(blksize),
1411 cpu_to_be64(block_id));
1414 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1415 struct drbd_peer_request *peer_req)
1417 struct drbd_socket *sock;
1418 struct p_block_desc *p;
1420 sock = &peer_device->connection->data;
1421 p = drbd_prepare_command(peer_device, sock);
1422 if (!p)
1423 return -EIO;
1424 p->sector = cpu_to_be64(peer_req->i.sector);
1425 p->blksize = cpu_to_be32(peer_req->i.size);
1426 p->pad = 0;
1427 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1430 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1431 sector_t sector, int size, u64 block_id)
1433 struct drbd_socket *sock;
1434 struct p_block_req *p;
1436 sock = &peer_device->connection->data;
1437 p = drbd_prepare_command(peer_device, sock);
1438 if (!p)
1439 return -EIO;
1440 p->sector = cpu_to_be64(sector);
1441 p->block_id = block_id;
1442 p->blksize = cpu_to_be32(size);
1443 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1446 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1447 void *digest, int digest_size, enum drbd_packet cmd)
1449 struct drbd_socket *sock;
1450 struct p_block_req *p;
1452 /* FIXME: Put the digest into the preallocated socket buffer. */
1454 sock = &peer_device->connection->data;
1455 p = drbd_prepare_command(peer_device, sock);
1456 if (!p)
1457 return -EIO;
1458 p->sector = cpu_to_be64(sector);
1459 p->block_id = ID_SYNCER /* unused */;
1460 p->blksize = cpu_to_be32(size);
1461 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1464 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1466 struct drbd_socket *sock;
1467 struct p_block_req *p;
1469 sock = &peer_device->connection->data;
1470 p = drbd_prepare_command(peer_device, sock);
1471 if (!p)
1472 return -EIO;
1473 p->sector = cpu_to_be64(sector);
1474 p->block_id = ID_SYNCER /* unused */;
1475 p->blksize = cpu_to_be32(size);
1476 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1479 /* called on sndtimeo
1480 * returns false if we should retry,
1481 * true if we think connection is dead
1483 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1485 int drop_it;
1486 /* long elapsed = (long)(jiffies - device->last_received); */
1488 drop_it = connection->meta.socket == sock
1489 || !connection->ack_receiver.task
1490 || get_t_state(&connection->ack_receiver) != RUNNING
1491 || connection->cstate < C_WF_REPORT_PARAMS;
1493 if (drop_it)
1494 return true;
1496 drop_it = !--connection->ko_count;
1497 if (!drop_it) {
1498 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1499 current->comm, current->pid, connection->ko_count);
1500 request_ping(connection);
1503 return drop_it; /* && (device->state == R_PRIMARY) */;
1506 static void drbd_update_congested(struct drbd_connection *connection)
1508 struct sock *sk = connection->data.socket->sk;
1509 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1510 set_bit(NET_CONGESTED, &connection->flags);
1513 /* The idea of sendpage seems to be to put some kind of reference
1514 * to the page into the skb, and to hand it over to the NIC. In
1515 * this process get_page() gets called.
1517 * As soon as the page was really sent over the network put_page()
1518 * gets called by some part of the network layer. [ NIC driver? ]
1520 * [ get_page() / put_page() increment/decrement the count. If count
1521 * reaches 0 the page will be freed. ]
1523 * This works nicely with pages from FSs.
1524 * But this means that in protocol A we might signal IO completion too early!
1526 * In order not to corrupt data during a resync we must make sure
1527 * that we do not reuse our own buffer pages (EEs) to early, therefore
1528 * we have the net_ee list.
1530 * XFS seems to have problems, still, it submits pages with page_count == 0!
1531 * As a workaround, we disable sendpage on pages
1532 * with page_count == 0 or PageSlab.
1534 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1535 int offset, size_t size, unsigned msg_flags)
1537 struct socket *socket;
1538 void *addr;
1539 int err;
1541 socket = peer_device->connection->data.socket;
1542 addr = kmap(page) + offset;
1543 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1544 kunmap(page);
1545 if (!err)
1546 peer_device->device->send_cnt += size >> 9;
1547 return err;
1550 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1551 int offset, size_t size, unsigned msg_flags)
1553 struct socket *socket = peer_device->connection->data.socket;
1554 mm_segment_t oldfs = get_fs();
1555 int len = size;
1556 int err = -EIO;
1558 /* e.g. XFS meta- & log-data is in slab pages, which have a
1559 * page_count of 0 and/or have PageSlab() set.
1560 * we cannot use send_page for those, as that does get_page();
1561 * put_page(); and would cause either a VM_BUG directly, or
1562 * __page_cache_release a page that would actually still be referenced
1563 * by someone, leading to some obscure delayed Oops somewhere else. */
1564 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1565 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1567 msg_flags |= MSG_NOSIGNAL;
1568 drbd_update_congested(peer_device->connection);
1569 set_fs(KERNEL_DS);
1570 do {
1571 int sent;
1573 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1574 if (sent <= 0) {
1575 if (sent == -EAGAIN) {
1576 if (we_should_drop_the_connection(peer_device->connection, socket))
1577 break;
1578 continue;
1580 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1581 __func__, (int)size, len, sent);
1582 if (sent < 0)
1583 err = sent;
1584 break;
1586 len -= sent;
1587 offset += sent;
1588 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1589 set_fs(oldfs);
1590 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1592 if (len == 0) {
1593 err = 0;
1594 peer_device->device->send_cnt += size >> 9;
1596 return err;
1599 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1601 struct bio_vec bvec;
1602 struct bvec_iter iter;
1604 /* hint all but last page with MSG_MORE */
1605 bio_for_each_segment(bvec, bio, iter) {
1606 int err;
1608 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1609 bvec.bv_offset, bvec.bv_len,
1610 bio_iter_last(bvec, iter)
1611 ? 0 : MSG_MORE);
1612 if (err)
1613 return err;
1614 /* REQ_OP_WRITE_SAME has only one segment */
1615 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1616 break;
1618 return 0;
1621 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1623 struct bio_vec bvec;
1624 struct bvec_iter iter;
1626 /* hint all but last page with MSG_MORE */
1627 bio_for_each_segment(bvec, bio, iter) {
1628 int err;
1630 err = _drbd_send_page(peer_device, bvec.bv_page,
1631 bvec.bv_offset, bvec.bv_len,
1632 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1633 if (err)
1634 return err;
1635 /* REQ_OP_WRITE_SAME has only one segment */
1636 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1637 break;
1639 return 0;
1642 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1643 struct drbd_peer_request *peer_req)
1645 struct page *page = peer_req->pages;
1646 unsigned len = peer_req->i.size;
1647 int err;
1649 /* hint all but last page with MSG_MORE */
1650 page_chain_for_each(page) {
1651 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1653 err = _drbd_send_page(peer_device, page, 0, l,
1654 page_chain_next(page) ? MSG_MORE : 0);
1655 if (err)
1656 return err;
1657 len -= l;
1659 return 0;
1662 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1663 struct bio *bio)
1665 if (connection->agreed_pro_version >= 95)
1666 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1667 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1668 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1669 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1670 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0);
1671 else
1672 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1675 /* Used to send write or TRIM aka REQ_DISCARD requests
1676 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1678 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1680 struct drbd_device *device = peer_device->device;
1681 struct drbd_socket *sock;
1682 struct p_data *p;
1683 struct p_wsame *wsame = NULL;
1684 void *digest_out;
1685 unsigned int dp_flags = 0;
1686 int digest_size;
1687 int err;
1689 sock = &peer_device->connection->data;
1690 p = drbd_prepare_command(peer_device, sock);
1691 digest_size = peer_device->connection->integrity_tfm ?
1692 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1694 if (!p)
1695 return -EIO;
1696 p->sector = cpu_to_be64(req->i.sector);
1697 p->block_id = (unsigned long)req;
1698 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1699 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1700 if (device->state.conn >= C_SYNC_SOURCE &&
1701 device->state.conn <= C_PAUSED_SYNC_T)
1702 dp_flags |= DP_MAY_SET_IN_SYNC;
1703 if (peer_device->connection->agreed_pro_version >= 100) {
1704 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1705 dp_flags |= DP_SEND_RECEIVE_ACK;
1706 /* During resync, request an explicit write ack,
1707 * even in protocol != C */
1708 if (req->rq_state & RQ_EXP_WRITE_ACK
1709 || (dp_flags & DP_MAY_SET_IN_SYNC))
1710 dp_flags |= DP_SEND_WRITE_ACK;
1712 p->dp_flags = cpu_to_be32(dp_flags);
1714 if (dp_flags & DP_DISCARD) {
1715 struct p_trim *t = (struct p_trim*)p;
1716 t->size = cpu_to_be32(req->i.size);
1717 err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1718 goto out;
1720 if (dp_flags & DP_WSAME) {
1721 /* this will only work if DRBD_FF_WSAME is set AND the
1722 * handshake agreed that all nodes and backend devices are
1723 * WRITE_SAME capable and agree on logical_block_size */
1724 wsame = (struct p_wsame*)p;
1725 digest_out = wsame + 1;
1726 wsame->size = cpu_to_be32(req->i.size);
1727 } else
1728 digest_out = p + 1;
1730 /* our digest is still only over the payload.
1731 * TRIM does not carry any payload. */
1732 if (digest_size)
1733 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1734 if (wsame) {
1735 err =
1736 __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1737 sizeof(*wsame) + digest_size, NULL,
1738 bio_iovec(req->master_bio).bv_len);
1739 } else
1740 err =
1741 __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1742 sizeof(*p) + digest_size, NULL, req->i.size);
1743 if (!err) {
1744 /* For protocol A, we have to memcpy the payload into
1745 * socket buffers, as we may complete right away
1746 * as soon as we handed it over to tcp, at which point the data
1747 * pages may become invalid.
1749 * For data-integrity enabled, we copy it as well, so we can be
1750 * sure that even if the bio pages may still be modified, it
1751 * won't change the data on the wire, thus if the digest checks
1752 * out ok after sending on this side, but does not fit on the
1753 * receiving side, we sure have detected corruption elsewhere.
1755 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1756 err = _drbd_send_bio(peer_device, req->master_bio);
1757 else
1758 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1760 /* double check digest, sometimes buffers have been modified in flight. */
1761 if (digest_size > 0 && digest_size <= 64) {
1762 /* 64 byte, 512 bit, is the largest digest size
1763 * currently supported in kernel crypto. */
1764 unsigned char digest[64];
1765 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1766 if (memcmp(p + 1, digest, digest_size)) {
1767 drbd_warn(device,
1768 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1769 (unsigned long long)req->i.sector, req->i.size);
1771 } /* else if (digest_size > 64) {
1772 ... Be noisy about digest too large ...
1773 } */
1775 out:
1776 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1778 return err;
1781 /* answer packet, used to send data back for read requests:
1782 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1783 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1785 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1786 struct drbd_peer_request *peer_req)
1788 struct drbd_device *device = peer_device->device;
1789 struct drbd_socket *sock;
1790 struct p_data *p;
1791 int err;
1792 int digest_size;
1794 sock = &peer_device->connection->data;
1795 p = drbd_prepare_command(peer_device, sock);
1797 digest_size = peer_device->connection->integrity_tfm ?
1798 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1800 if (!p)
1801 return -EIO;
1802 p->sector = cpu_to_be64(peer_req->i.sector);
1803 p->block_id = peer_req->block_id;
1804 p->seq_num = 0; /* unused */
1805 p->dp_flags = 0;
1806 if (digest_size)
1807 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1808 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1809 if (!err)
1810 err = _drbd_send_zc_ee(peer_device, peer_req);
1811 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1813 return err;
1816 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1818 struct drbd_socket *sock;
1819 struct p_block_desc *p;
1821 sock = &peer_device->connection->data;
1822 p = drbd_prepare_command(peer_device, sock);
1823 if (!p)
1824 return -EIO;
1825 p->sector = cpu_to_be64(req->i.sector);
1826 p->blksize = cpu_to_be32(req->i.size);
1827 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1831 drbd_send distinguishes two cases:
1833 Packets sent via the data socket "sock"
1834 and packets sent via the meta data socket "msock"
1836 sock msock
1837 -----------------+-------------------------+------------------------------
1838 timeout conf.timeout / 2 conf.timeout / 2
1839 timeout action send a ping via msock Abort communication
1840 and close all sockets
1844 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1846 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1847 void *buf, size_t size, unsigned msg_flags)
1849 struct kvec iov;
1850 struct msghdr msg;
1851 int rv, sent = 0;
1853 if (!sock)
1854 return -EBADR;
1856 /* THINK if (signal_pending) return ... ? */
1858 iov.iov_base = buf;
1859 iov.iov_len = size;
1861 msg.msg_name = NULL;
1862 msg.msg_namelen = 0;
1863 msg.msg_control = NULL;
1864 msg.msg_controllen = 0;
1865 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1867 if (sock == connection->data.socket) {
1868 rcu_read_lock();
1869 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1870 rcu_read_unlock();
1871 drbd_update_congested(connection);
1873 do {
1874 rv = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
1875 if (rv == -EAGAIN) {
1876 if (we_should_drop_the_connection(connection, sock))
1877 break;
1878 else
1879 continue;
1881 if (rv == -EINTR) {
1882 flush_signals(current);
1883 rv = 0;
1885 if (rv < 0)
1886 break;
1887 sent += rv;
1888 iov.iov_base += rv;
1889 iov.iov_len -= rv;
1890 } while (sent < size);
1892 if (sock == connection->data.socket)
1893 clear_bit(NET_CONGESTED, &connection->flags);
1895 if (rv <= 0) {
1896 if (rv != -EAGAIN) {
1897 drbd_err(connection, "%s_sendmsg returned %d\n",
1898 sock == connection->meta.socket ? "msock" : "sock",
1899 rv);
1900 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1901 } else
1902 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1905 return sent;
1909 * drbd_send_all - Send an entire buffer
1911 * Returns 0 upon success and a negative error value otherwise.
1913 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1914 size_t size, unsigned msg_flags)
1916 int err;
1918 err = drbd_send(connection, sock, buffer, size, msg_flags);
1919 if (err < 0)
1920 return err;
1921 if (err != size)
1922 return -EIO;
1923 return 0;
1926 static int drbd_open(struct block_device *bdev, fmode_t mode)
1928 struct drbd_device *device = bdev->bd_disk->private_data;
1929 unsigned long flags;
1930 int rv = 0;
1932 mutex_lock(&drbd_main_mutex);
1933 spin_lock_irqsave(&device->resource->req_lock, flags);
1934 /* to have a stable device->state.role
1935 * and no race with updating open_cnt */
1937 if (device->state.role != R_PRIMARY) {
1938 if (mode & FMODE_WRITE)
1939 rv = -EROFS;
1940 else if (!allow_oos)
1941 rv = -EMEDIUMTYPE;
1944 if (!rv)
1945 device->open_cnt++;
1946 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1947 mutex_unlock(&drbd_main_mutex);
1949 return rv;
1952 static void drbd_release(struct gendisk *gd, fmode_t mode)
1954 struct drbd_device *device = gd->private_data;
1955 mutex_lock(&drbd_main_mutex);
1956 device->open_cnt--;
1957 mutex_unlock(&drbd_main_mutex);
1960 static void drbd_set_defaults(struct drbd_device *device)
1962 /* Beware! The actual layout differs
1963 * between big endian and little endian */
1964 device->state = (union drbd_dev_state) {
1965 { .role = R_SECONDARY,
1966 .peer = R_UNKNOWN,
1967 .conn = C_STANDALONE,
1968 .disk = D_DISKLESS,
1969 .pdsk = D_UNKNOWN,
1970 } };
1973 void drbd_init_set_defaults(struct drbd_device *device)
1975 /* the memset(,0,) did most of this.
1976 * note: only assignments, no allocation in here */
1978 drbd_set_defaults(device);
1980 atomic_set(&device->ap_bio_cnt, 0);
1981 atomic_set(&device->ap_actlog_cnt, 0);
1982 atomic_set(&device->ap_pending_cnt, 0);
1983 atomic_set(&device->rs_pending_cnt, 0);
1984 atomic_set(&device->unacked_cnt, 0);
1985 atomic_set(&device->local_cnt, 0);
1986 atomic_set(&device->pp_in_use_by_net, 0);
1987 atomic_set(&device->rs_sect_in, 0);
1988 atomic_set(&device->rs_sect_ev, 0);
1989 atomic_set(&device->ap_in_flight, 0);
1990 atomic_set(&device->md_io.in_use, 0);
1992 mutex_init(&device->own_state_mutex);
1993 device->state_mutex = &device->own_state_mutex;
1995 spin_lock_init(&device->al_lock);
1996 spin_lock_init(&device->peer_seq_lock);
1998 INIT_LIST_HEAD(&device->active_ee);
1999 INIT_LIST_HEAD(&device->sync_ee);
2000 INIT_LIST_HEAD(&device->done_ee);
2001 INIT_LIST_HEAD(&device->read_ee);
2002 INIT_LIST_HEAD(&device->net_ee);
2003 INIT_LIST_HEAD(&device->resync_reads);
2004 INIT_LIST_HEAD(&device->resync_work.list);
2005 INIT_LIST_HEAD(&device->unplug_work.list);
2006 INIT_LIST_HEAD(&device->bm_io_work.w.list);
2007 INIT_LIST_HEAD(&device->pending_master_completion[0]);
2008 INIT_LIST_HEAD(&device->pending_master_completion[1]);
2009 INIT_LIST_HEAD(&device->pending_completion[0]);
2010 INIT_LIST_HEAD(&device->pending_completion[1]);
2012 device->resync_work.cb = w_resync_timer;
2013 device->unplug_work.cb = w_send_write_hint;
2014 device->bm_io_work.w.cb = w_bitmap_io;
2016 init_timer(&device->resync_timer);
2017 init_timer(&device->md_sync_timer);
2018 init_timer(&device->start_resync_timer);
2019 init_timer(&device->request_timer);
2020 device->resync_timer.function = resync_timer_fn;
2021 device->resync_timer.data = (unsigned long) device;
2022 device->md_sync_timer.function = md_sync_timer_fn;
2023 device->md_sync_timer.data = (unsigned long) device;
2024 device->start_resync_timer.function = start_resync_timer_fn;
2025 device->start_resync_timer.data = (unsigned long) device;
2026 device->request_timer.function = request_timer_fn;
2027 device->request_timer.data = (unsigned long) device;
2029 init_waitqueue_head(&device->misc_wait);
2030 init_waitqueue_head(&device->state_wait);
2031 init_waitqueue_head(&device->ee_wait);
2032 init_waitqueue_head(&device->al_wait);
2033 init_waitqueue_head(&device->seq_wait);
2035 device->resync_wenr = LC_FREE;
2036 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2037 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2040 void drbd_device_cleanup(struct drbd_device *device)
2042 int i;
2043 if (first_peer_device(device)->connection->receiver.t_state != NONE)
2044 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2045 first_peer_device(device)->connection->receiver.t_state);
2047 device->al_writ_cnt =
2048 device->bm_writ_cnt =
2049 device->read_cnt =
2050 device->recv_cnt =
2051 device->send_cnt =
2052 device->writ_cnt =
2053 device->p_size =
2054 device->rs_start =
2055 device->rs_total =
2056 device->rs_failed = 0;
2057 device->rs_last_events = 0;
2058 device->rs_last_sect_ev = 0;
2059 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2060 device->rs_mark_left[i] = 0;
2061 device->rs_mark_time[i] = 0;
2063 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2065 drbd_set_my_capacity(device, 0);
2066 if (device->bitmap) {
2067 /* maybe never allocated. */
2068 drbd_bm_resize(device, 0, 1);
2069 drbd_bm_cleanup(device);
2072 drbd_backing_dev_free(device, device->ldev);
2073 device->ldev = NULL;
2075 clear_bit(AL_SUSPENDED, &device->flags);
2077 D_ASSERT(device, list_empty(&device->active_ee));
2078 D_ASSERT(device, list_empty(&device->sync_ee));
2079 D_ASSERT(device, list_empty(&device->done_ee));
2080 D_ASSERT(device, list_empty(&device->read_ee));
2081 D_ASSERT(device, list_empty(&device->net_ee));
2082 D_ASSERT(device, list_empty(&device->resync_reads));
2083 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2084 D_ASSERT(device, list_empty(&device->resync_work.list));
2085 D_ASSERT(device, list_empty(&device->unplug_work.list));
2087 drbd_set_defaults(device);
2091 static void drbd_destroy_mempools(void)
2093 struct page *page;
2095 while (drbd_pp_pool) {
2096 page = drbd_pp_pool;
2097 drbd_pp_pool = (struct page *)page_private(page);
2098 __free_page(page);
2099 drbd_pp_vacant--;
2102 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2104 if (drbd_md_io_bio_set)
2105 bioset_free(drbd_md_io_bio_set);
2106 if (drbd_md_io_page_pool)
2107 mempool_destroy(drbd_md_io_page_pool);
2108 if (drbd_ee_mempool)
2109 mempool_destroy(drbd_ee_mempool);
2110 if (drbd_request_mempool)
2111 mempool_destroy(drbd_request_mempool);
2112 if (drbd_ee_cache)
2113 kmem_cache_destroy(drbd_ee_cache);
2114 if (drbd_request_cache)
2115 kmem_cache_destroy(drbd_request_cache);
2116 if (drbd_bm_ext_cache)
2117 kmem_cache_destroy(drbd_bm_ext_cache);
2118 if (drbd_al_ext_cache)
2119 kmem_cache_destroy(drbd_al_ext_cache);
2121 drbd_md_io_bio_set = NULL;
2122 drbd_md_io_page_pool = NULL;
2123 drbd_ee_mempool = NULL;
2124 drbd_request_mempool = NULL;
2125 drbd_ee_cache = NULL;
2126 drbd_request_cache = NULL;
2127 drbd_bm_ext_cache = NULL;
2128 drbd_al_ext_cache = NULL;
2130 return;
2133 static int drbd_create_mempools(void)
2135 struct page *page;
2136 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
2137 int i;
2139 /* prepare our caches and mempools */
2140 drbd_request_mempool = NULL;
2141 drbd_ee_cache = NULL;
2142 drbd_request_cache = NULL;
2143 drbd_bm_ext_cache = NULL;
2144 drbd_al_ext_cache = NULL;
2145 drbd_pp_pool = NULL;
2146 drbd_md_io_page_pool = NULL;
2147 drbd_md_io_bio_set = NULL;
2149 /* caches */
2150 drbd_request_cache = kmem_cache_create(
2151 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2152 if (drbd_request_cache == NULL)
2153 goto Enomem;
2155 drbd_ee_cache = kmem_cache_create(
2156 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2157 if (drbd_ee_cache == NULL)
2158 goto Enomem;
2160 drbd_bm_ext_cache = kmem_cache_create(
2161 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2162 if (drbd_bm_ext_cache == NULL)
2163 goto Enomem;
2165 drbd_al_ext_cache = kmem_cache_create(
2166 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2167 if (drbd_al_ext_cache == NULL)
2168 goto Enomem;
2170 /* mempools */
2171 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2172 if (drbd_md_io_bio_set == NULL)
2173 goto Enomem;
2175 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2176 if (drbd_md_io_page_pool == NULL)
2177 goto Enomem;
2179 drbd_request_mempool = mempool_create_slab_pool(number,
2180 drbd_request_cache);
2181 if (drbd_request_mempool == NULL)
2182 goto Enomem;
2184 drbd_ee_mempool = mempool_create_slab_pool(number, drbd_ee_cache);
2185 if (drbd_ee_mempool == NULL)
2186 goto Enomem;
2188 /* drbd's page pool */
2189 spin_lock_init(&drbd_pp_lock);
2191 for (i = 0; i < number; i++) {
2192 page = alloc_page(GFP_HIGHUSER);
2193 if (!page)
2194 goto Enomem;
2195 set_page_private(page, (unsigned long)drbd_pp_pool);
2196 drbd_pp_pool = page;
2198 drbd_pp_vacant = number;
2200 return 0;
2202 Enomem:
2203 drbd_destroy_mempools(); /* in case we allocated some */
2204 return -ENOMEM;
2207 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2209 int rr;
2211 rr = drbd_free_peer_reqs(device, &device->active_ee);
2212 if (rr)
2213 drbd_err(device, "%d EEs in active list found!\n", rr);
2215 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2216 if (rr)
2217 drbd_err(device, "%d EEs in sync list found!\n", rr);
2219 rr = drbd_free_peer_reqs(device, &device->read_ee);
2220 if (rr)
2221 drbd_err(device, "%d EEs in read list found!\n", rr);
2223 rr = drbd_free_peer_reqs(device, &device->done_ee);
2224 if (rr)
2225 drbd_err(device, "%d EEs in done list found!\n", rr);
2227 rr = drbd_free_peer_reqs(device, &device->net_ee);
2228 if (rr)
2229 drbd_err(device, "%d EEs in net list found!\n", rr);
2232 /* caution. no locking. */
2233 void drbd_destroy_device(struct kref *kref)
2235 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2236 struct drbd_resource *resource = device->resource;
2237 struct drbd_peer_device *peer_device, *tmp_peer_device;
2239 del_timer_sync(&device->request_timer);
2241 /* paranoia asserts */
2242 D_ASSERT(device, device->open_cnt == 0);
2243 /* end paranoia asserts */
2245 /* cleanup stuff that may have been allocated during
2246 * device (re-)configuration or state changes */
2248 if (device->this_bdev)
2249 bdput(device->this_bdev);
2251 drbd_backing_dev_free(device, device->ldev);
2252 device->ldev = NULL;
2254 drbd_release_all_peer_reqs(device);
2256 lc_destroy(device->act_log);
2257 lc_destroy(device->resync);
2259 kfree(device->p_uuid);
2260 /* device->p_uuid = NULL; */
2262 if (device->bitmap) /* should no longer be there. */
2263 drbd_bm_cleanup(device);
2264 __free_page(device->md_io.page);
2265 put_disk(device->vdisk);
2266 blk_cleanup_queue(device->rq_queue);
2267 kfree(device->rs_plan_s);
2269 /* not for_each_connection(connection, resource):
2270 * those may have been cleaned up and disassociated already.
2272 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2273 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2274 kfree(peer_device);
2276 memset(device, 0xfd, sizeof(*device));
2277 kfree(device);
2278 kref_put(&resource->kref, drbd_destroy_resource);
2281 /* One global retry thread, if we need to push back some bio and have it
2282 * reinserted through our make request function.
2284 static struct retry_worker {
2285 struct workqueue_struct *wq;
2286 struct work_struct worker;
2288 spinlock_t lock;
2289 struct list_head writes;
2290 } retry;
2292 static void do_retry(struct work_struct *ws)
2294 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2295 LIST_HEAD(writes);
2296 struct drbd_request *req, *tmp;
2298 spin_lock_irq(&retry->lock);
2299 list_splice_init(&retry->writes, &writes);
2300 spin_unlock_irq(&retry->lock);
2302 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2303 struct drbd_device *device = req->device;
2304 struct bio *bio = req->master_bio;
2305 unsigned long start_jif = req->start_jif;
2306 bool expected;
2308 expected =
2309 expect(atomic_read(&req->completion_ref) == 0) &&
2310 expect(req->rq_state & RQ_POSTPONED) &&
2311 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2312 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2314 if (!expected)
2315 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2316 req, atomic_read(&req->completion_ref),
2317 req->rq_state);
2319 /* We still need to put one kref associated with the
2320 * "completion_ref" going zero in the code path that queued it
2321 * here. The request object may still be referenced by a
2322 * frozen local req->private_bio, in case we force-detached.
2324 kref_put(&req->kref, drbd_req_destroy);
2326 /* A single suspended or otherwise blocking device may stall
2327 * all others as well. Fortunately, this code path is to
2328 * recover from a situation that "should not happen":
2329 * concurrent writes in multi-primary setup.
2330 * In a "normal" lifecycle, this workqueue is supposed to be
2331 * destroyed without ever doing anything.
2332 * If it turns out to be an issue anyways, we can do per
2333 * resource (replication group) or per device (minor) retry
2334 * workqueues instead.
2337 /* We are not just doing generic_make_request(),
2338 * as we want to keep the start_time information. */
2339 inc_ap_bio(device);
2340 __drbd_make_request(device, bio, start_jif);
2344 /* called via drbd_req_put_completion_ref(),
2345 * holds resource->req_lock */
2346 void drbd_restart_request(struct drbd_request *req)
2348 unsigned long flags;
2349 spin_lock_irqsave(&retry.lock, flags);
2350 list_move_tail(&req->tl_requests, &retry.writes);
2351 spin_unlock_irqrestore(&retry.lock, flags);
2353 /* Drop the extra reference that would otherwise
2354 * have been dropped by complete_master_bio.
2355 * do_retry() needs to grab a new one. */
2356 dec_ap_bio(req->device);
2358 queue_work(retry.wq, &retry.worker);
2361 void drbd_destroy_resource(struct kref *kref)
2363 struct drbd_resource *resource =
2364 container_of(kref, struct drbd_resource, kref);
2366 idr_destroy(&resource->devices);
2367 free_cpumask_var(resource->cpu_mask);
2368 kfree(resource->name);
2369 memset(resource, 0xf2, sizeof(*resource));
2370 kfree(resource);
2373 void drbd_free_resource(struct drbd_resource *resource)
2375 struct drbd_connection *connection, *tmp;
2377 for_each_connection_safe(connection, tmp, resource) {
2378 list_del(&connection->connections);
2379 drbd_debugfs_connection_cleanup(connection);
2380 kref_put(&connection->kref, drbd_destroy_connection);
2382 drbd_debugfs_resource_cleanup(resource);
2383 kref_put(&resource->kref, drbd_destroy_resource);
2386 static void drbd_cleanup(void)
2388 unsigned int i;
2389 struct drbd_device *device;
2390 struct drbd_resource *resource, *tmp;
2392 /* first remove proc,
2393 * drbdsetup uses it's presence to detect
2394 * whether DRBD is loaded.
2395 * If we would get stuck in proc removal,
2396 * but have netlink already deregistered,
2397 * some drbdsetup commands may wait forever
2398 * for an answer.
2400 if (drbd_proc)
2401 remove_proc_entry("drbd", NULL);
2403 if (retry.wq)
2404 destroy_workqueue(retry.wq);
2406 drbd_genl_unregister();
2407 drbd_debugfs_cleanup();
2409 idr_for_each_entry(&drbd_devices, device, i)
2410 drbd_delete_device(device);
2412 /* not _rcu since, no other updater anymore. Genl already unregistered */
2413 for_each_resource_safe(resource, tmp, &drbd_resources) {
2414 list_del(&resource->resources);
2415 drbd_free_resource(resource);
2418 drbd_destroy_mempools();
2419 unregister_blkdev(DRBD_MAJOR, "drbd");
2421 idr_destroy(&drbd_devices);
2423 pr_info("module cleanup done.\n");
2427 * drbd_congested() - Callback for the flusher thread
2428 * @congested_data: User data
2429 * @bdi_bits: Bits the BDI flusher thread is currently interested in
2431 * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested.
2433 static int drbd_congested(void *congested_data, int bdi_bits)
2435 struct drbd_device *device = congested_data;
2436 struct request_queue *q;
2437 char reason = '-';
2438 int r = 0;
2440 if (!may_inc_ap_bio(device)) {
2441 /* DRBD has frozen IO */
2442 r = bdi_bits;
2443 reason = 'd';
2444 goto out;
2447 if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
2448 r |= (1 << WB_async_congested);
2449 /* Without good local data, we would need to read from remote,
2450 * and that would need the worker thread as well, which is
2451 * currently blocked waiting for that usermode helper to
2452 * finish.
2454 if (!get_ldev_if_state(device, D_UP_TO_DATE))
2455 r |= (1 << WB_sync_congested);
2456 else
2457 put_ldev(device);
2458 r &= bdi_bits;
2459 reason = 'c';
2460 goto out;
2463 if (get_ldev(device)) {
2464 q = bdev_get_queue(device->ldev->backing_bdev);
2465 r = bdi_congested(&q->backing_dev_info, bdi_bits);
2466 put_ldev(device);
2467 if (r)
2468 reason = 'b';
2471 if (bdi_bits & (1 << WB_async_congested) &&
2472 test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
2473 r |= (1 << WB_async_congested);
2474 reason = reason == 'b' ? 'a' : 'n';
2477 out:
2478 device->congestion_reason = reason;
2479 return r;
2482 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2484 spin_lock_init(&wq->q_lock);
2485 INIT_LIST_HEAD(&wq->q);
2486 init_waitqueue_head(&wq->q_wait);
2489 struct completion_work {
2490 struct drbd_work w;
2491 struct completion done;
2494 static int w_complete(struct drbd_work *w, int cancel)
2496 struct completion_work *completion_work =
2497 container_of(w, struct completion_work, w);
2499 complete(&completion_work->done);
2500 return 0;
2503 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2505 struct completion_work completion_work;
2507 completion_work.w.cb = w_complete;
2508 init_completion(&completion_work.done);
2509 drbd_queue_work(work_queue, &completion_work.w);
2510 wait_for_completion(&completion_work.done);
2513 struct drbd_resource *drbd_find_resource(const char *name)
2515 struct drbd_resource *resource;
2517 if (!name || !name[0])
2518 return NULL;
2520 rcu_read_lock();
2521 for_each_resource_rcu(resource, &drbd_resources) {
2522 if (!strcmp(resource->name, name)) {
2523 kref_get(&resource->kref);
2524 goto found;
2527 resource = NULL;
2528 found:
2529 rcu_read_unlock();
2530 return resource;
2533 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2534 void *peer_addr, int peer_addr_len)
2536 struct drbd_resource *resource;
2537 struct drbd_connection *connection;
2539 rcu_read_lock();
2540 for_each_resource_rcu(resource, &drbd_resources) {
2541 for_each_connection_rcu(connection, resource) {
2542 if (connection->my_addr_len == my_addr_len &&
2543 connection->peer_addr_len == peer_addr_len &&
2544 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2545 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2546 kref_get(&connection->kref);
2547 goto found;
2551 connection = NULL;
2552 found:
2553 rcu_read_unlock();
2554 return connection;
2557 static int drbd_alloc_socket(struct drbd_socket *socket)
2559 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2560 if (!socket->rbuf)
2561 return -ENOMEM;
2562 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2563 if (!socket->sbuf)
2564 return -ENOMEM;
2565 return 0;
2568 static void drbd_free_socket(struct drbd_socket *socket)
2570 free_page((unsigned long) socket->sbuf);
2571 free_page((unsigned long) socket->rbuf);
2574 void conn_free_crypto(struct drbd_connection *connection)
2576 drbd_free_sock(connection);
2578 crypto_free_ahash(connection->csums_tfm);
2579 crypto_free_ahash(connection->verify_tfm);
2580 crypto_free_shash(connection->cram_hmac_tfm);
2581 crypto_free_ahash(connection->integrity_tfm);
2582 crypto_free_ahash(connection->peer_integrity_tfm);
2583 kfree(connection->int_dig_in);
2584 kfree(connection->int_dig_vv);
2586 connection->csums_tfm = NULL;
2587 connection->verify_tfm = NULL;
2588 connection->cram_hmac_tfm = NULL;
2589 connection->integrity_tfm = NULL;
2590 connection->peer_integrity_tfm = NULL;
2591 connection->int_dig_in = NULL;
2592 connection->int_dig_vv = NULL;
2595 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2597 struct drbd_connection *connection;
2598 cpumask_var_t new_cpu_mask;
2599 int err;
2601 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2602 return -ENOMEM;
2604 /* silently ignore cpu mask on UP kernel */
2605 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2606 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2607 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2608 if (err == -EOVERFLOW) {
2609 /* So what. mask it out. */
2610 cpumask_var_t tmp_cpu_mask;
2611 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2612 cpumask_setall(tmp_cpu_mask);
2613 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2614 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2615 res_opts->cpu_mask,
2616 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2617 nr_cpu_ids);
2618 free_cpumask_var(tmp_cpu_mask);
2619 err = 0;
2622 if (err) {
2623 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2624 /* retcode = ERR_CPU_MASK_PARSE; */
2625 goto fail;
2628 resource->res_opts = *res_opts;
2629 if (cpumask_empty(new_cpu_mask))
2630 drbd_calc_cpu_mask(&new_cpu_mask);
2631 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2632 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2633 for_each_connection_rcu(connection, resource) {
2634 connection->receiver.reset_cpu_mask = 1;
2635 connection->ack_receiver.reset_cpu_mask = 1;
2636 connection->worker.reset_cpu_mask = 1;
2639 err = 0;
2641 fail:
2642 free_cpumask_var(new_cpu_mask);
2643 return err;
2647 struct drbd_resource *drbd_create_resource(const char *name)
2649 struct drbd_resource *resource;
2651 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2652 if (!resource)
2653 goto fail;
2654 resource->name = kstrdup(name, GFP_KERNEL);
2655 if (!resource->name)
2656 goto fail_free_resource;
2657 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2658 goto fail_free_name;
2659 kref_init(&resource->kref);
2660 idr_init(&resource->devices);
2661 INIT_LIST_HEAD(&resource->connections);
2662 resource->write_ordering = WO_BDEV_FLUSH;
2663 list_add_tail_rcu(&resource->resources, &drbd_resources);
2664 mutex_init(&resource->conf_update);
2665 mutex_init(&resource->adm_mutex);
2666 spin_lock_init(&resource->req_lock);
2667 drbd_debugfs_resource_add(resource);
2668 return resource;
2670 fail_free_name:
2671 kfree(resource->name);
2672 fail_free_resource:
2673 kfree(resource);
2674 fail:
2675 return NULL;
2678 /* caller must be under adm_mutex */
2679 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2681 struct drbd_resource *resource;
2682 struct drbd_connection *connection;
2684 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2685 if (!connection)
2686 return NULL;
2688 if (drbd_alloc_socket(&connection->data))
2689 goto fail;
2690 if (drbd_alloc_socket(&connection->meta))
2691 goto fail;
2693 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2694 if (!connection->current_epoch)
2695 goto fail;
2697 INIT_LIST_HEAD(&connection->transfer_log);
2699 INIT_LIST_HEAD(&connection->current_epoch->list);
2700 connection->epochs = 1;
2701 spin_lock_init(&connection->epoch_lock);
2703 connection->send.seen_any_write_yet = false;
2704 connection->send.current_epoch_nr = 0;
2705 connection->send.current_epoch_writes = 0;
2707 resource = drbd_create_resource(name);
2708 if (!resource)
2709 goto fail;
2711 connection->cstate = C_STANDALONE;
2712 mutex_init(&connection->cstate_mutex);
2713 init_waitqueue_head(&connection->ping_wait);
2714 idr_init(&connection->peer_devices);
2716 drbd_init_workqueue(&connection->sender_work);
2717 mutex_init(&connection->data.mutex);
2718 mutex_init(&connection->meta.mutex);
2720 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2721 connection->receiver.connection = connection;
2722 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2723 connection->worker.connection = connection;
2724 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2725 connection->ack_receiver.connection = connection;
2727 kref_init(&connection->kref);
2729 connection->resource = resource;
2731 if (set_resource_options(resource, res_opts))
2732 goto fail_resource;
2734 kref_get(&resource->kref);
2735 list_add_tail_rcu(&connection->connections, &resource->connections);
2736 drbd_debugfs_connection_add(connection);
2737 return connection;
2739 fail_resource:
2740 list_del(&resource->resources);
2741 drbd_free_resource(resource);
2742 fail:
2743 kfree(connection->current_epoch);
2744 drbd_free_socket(&connection->meta);
2745 drbd_free_socket(&connection->data);
2746 kfree(connection);
2747 return NULL;
2750 void drbd_destroy_connection(struct kref *kref)
2752 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2753 struct drbd_resource *resource = connection->resource;
2755 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
2756 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2757 kfree(connection->current_epoch);
2759 idr_destroy(&connection->peer_devices);
2761 drbd_free_socket(&connection->meta);
2762 drbd_free_socket(&connection->data);
2763 kfree(connection->int_dig_in);
2764 kfree(connection->int_dig_vv);
2765 memset(connection, 0xfc, sizeof(*connection));
2766 kfree(connection);
2767 kref_put(&resource->kref, drbd_destroy_resource);
2770 static int init_submitter(struct drbd_device *device)
2772 /* opencoded create_singlethread_workqueue(),
2773 * to be able to say "drbd%d", ..., minor */
2774 device->submit.wq =
2775 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2776 if (!device->submit.wq)
2777 return -ENOMEM;
2779 INIT_WORK(&device->submit.worker, do_submit);
2780 INIT_LIST_HEAD(&device->submit.writes);
2781 return 0;
2784 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2786 struct drbd_resource *resource = adm_ctx->resource;
2787 struct drbd_connection *connection;
2788 struct drbd_device *device;
2789 struct drbd_peer_device *peer_device, *tmp_peer_device;
2790 struct gendisk *disk;
2791 struct request_queue *q;
2792 int id;
2793 int vnr = adm_ctx->volume;
2794 enum drbd_ret_code err = ERR_NOMEM;
2796 device = minor_to_device(minor);
2797 if (device)
2798 return ERR_MINOR_OR_VOLUME_EXISTS;
2800 /* GFP_KERNEL, we are outside of all write-out paths */
2801 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2802 if (!device)
2803 return ERR_NOMEM;
2804 kref_init(&device->kref);
2806 kref_get(&resource->kref);
2807 device->resource = resource;
2808 device->minor = minor;
2809 device->vnr = vnr;
2811 drbd_init_set_defaults(device);
2813 q = blk_alloc_queue(GFP_KERNEL);
2814 if (!q)
2815 goto out_no_q;
2816 device->rq_queue = q;
2817 q->queuedata = device;
2819 disk = alloc_disk(1);
2820 if (!disk)
2821 goto out_no_disk;
2822 device->vdisk = disk;
2824 set_disk_ro(disk, true);
2826 disk->queue = q;
2827 disk->major = DRBD_MAJOR;
2828 disk->first_minor = minor;
2829 disk->fops = &drbd_ops;
2830 sprintf(disk->disk_name, "drbd%d", minor);
2831 disk->private_data = device;
2833 device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2834 /* we have no partitions. we contain only ourselves. */
2835 device->this_bdev->bd_contains = device->this_bdev;
2837 q->backing_dev_info.congested_fn = drbd_congested;
2838 q->backing_dev_info.congested_data = device;
2840 blk_queue_make_request(q, drbd_make_request);
2841 blk_queue_write_cache(q, true, true);
2842 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2843 This triggers a max_bio_size message upon first attach or connect */
2844 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2845 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2846 q->queue_lock = &resource->req_lock;
2848 device->md_io.page = alloc_page(GFP_KERNEL);
2849 if (!device->md_io.page)
2850 goto out_no_io_page;
2852 if (drbd_bm_init(device))
2853 goto out_no_bitmap;
2854 device->read_requests = RB_ROOT;
2855 device->write_requests = RB_ROOT;
2857 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2858 if (id < 0) {
2859 if (id == -ENOSPC)
2860 err = ERR_MINOR_OR_VOLUME_EXISTS;
2861 goto out_no_minor_idr;
2863 kref_get(&device->kref);
2865 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2866 if (id < 0) {
2867 if (id == -ENOSPC)
2868 err = ERR_MINOR_OR_VOLUME_EXISTS;
2869 goto out_idr_remove_minor;
2871 kref_get(&device->kref);
2873 INIT_LIST_HEAD(&device->peer_devices);
2874 INIT_LIST_HEAD(&device->pending_bitmap_io);
2875 for_each_connection(connection, resource) {
2876 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2877 if (!peer_device)
2878 goto out_idr_remove_from_resource;
2879 peer_device->connection = connection;
2880 peer_device->device = device;
2882 list_add(&peer_device->peer_devices, &device->peer_devices);
2883 kref_get(&device->kref);
2885 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2886 if (id < 0) {
2887 if (id == -ENOSPC)
2888 err = ERR_INVALID_REQUEST;
2889 goto out_idr_remove_from_resource;
2891 kref_get(&connection->kref);
2892 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2895 if (init_submitter(device)) {
2896 err = ERR_NOMEM;
2897 goto out_idr_remove_vol;
2900 add_disk(disk);
2902 /* inherit the connection state */
2903 device->state.conn = first_connection(resource)->cstate;
2904 if (device->state.conn == C_WF_REPORT_PARAMS) {
2905 for_each_peer_device(peer_device, device)
2906 drbd_connected(peer_device);
2908 /* move to create_peer_device() */
2909 for_each_peer_device(peer_device, device)
2910 drbd_debugfs_peer_device_add(peer_device);
2911 drbd_debugfs_device_add(device);
2912 return NO_ERROR;
2914 out_idr_remove_vol:
2915 idr_remove(&connection->peer_devices, vnr);
2916 out_idr_remove_from_resource:
2917 for_each_connection(connection, resource) {
2918 peer_device = idr_find(&connection->peer_devices, vnr);
2919 if (peer_device) {
2920 idr_remove(&connection->peer_devices, vnr);
2921 kref_put(&connection->kref, drbd_destroy_connection);
2924 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2925 list_del(&peer_device->peer_devices);
2926 kfree(peer_device);
2928 idr_remove(&resource->devices, vnr);
2929 out_idr_remove_minor:
2930 idr_remove(&drbd_devices, minor);
2931 synchronize_rcu();
2932 out_no_minor_idr:
2933 drbd_bm_cleanup(device);
2934 out_no_bitmap:
2935 __free_page(device->md_io.page);
2936 out_no_io_page:
2937 put_disk(disk);
2938 out_no_disk:
2939 blk_cleanup_queue(q);
2940 out_no_q:
2941 kref_put(&resource->kref, drbd_destroy_resource);
2942 kfree(device);
2943 return err;
2946 void drbd_delete_device(struct drbd_device *device)
2948 struct drbd_resource *resource = device->resource;
2949 struct drbd_connection *connection;
2950 struct drbd_peer_device *peer_device;
2951 int refs = 3;
2953 /* move to free_peer_device() */
2954 for_each_peer_device(peer_device, device)
2955 drbd_debugfs_peer_device_cleanup(peer_device);
2956 drbd_debugfs_device_cleanup(device);
2957 for_each_connection(connection, resource) {
2958 idr_remove(&connection->peer_devices, device->vnr);
2959 refs++;
2961 idr_remove(&resource->devices, device->vnr);
2962 idr_remove(&drbd_devices, device_to_minor(device));
2963 del_gendisk(device->vdisk);
2964 synchronize_rcu();
2965 kref_sub(&device->kref, refs, drbd_destroy_device);
2968 static int __init drbd_init(void)
2970 int err;
2972 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2973 pr_err("invalid minor_count (%d)\n", minor_count);
2974 #ifdef MODULE
2975 return -EINVAL;
2976 #else
2977 minor_count = DRBD_MINOR_COUNT_DEF;
2978 #endif
2981 err = register_blkdev(DRBD_MAJOR, "drbd");
2982 if (err) {
2983 pr_err("unable to register block device major %d\n",
2984 DRBD_MAJOR);
2985 return err;
2989 * allocate all necessary structs
2991 init_waitqueue_head(&drbd_pp_wait);
2993 drbd_proc = NULL; /* play safe for drbd_cleanup */
2994 idr_init(&drbd_devices);
2996 mutex_init(&resources_mutex);
2997 INIT_LIST_HEAD(&drbd_resources);
2999 err = drbd_genl_register();
3000 if (err) {
3001 pr_err("unable to register generic netlink family\n");
3002 goto fail;
3005 err = drbd_create_mempools();
3006 if (err)
3007 goto fail;
3009 err = -ENOMEM;
3010 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
3011 if (!drbd_proc) {
3012 pr_err("unable to register proc file\n");
3013 goto fail;
3016 retry.wq = create_singlethread_workqueue("drbd-reissue");
3017 if (!retry.wq) {
3018 pr_err("unable to create retry workqueue\n");
3019 goto fail;
3021 INIT_WORK(&retry.worker, do_retry);
3022 spin_lock_init(&retry.lock);
3023 INIT_LIST_HEAD(&retry.writes);
3025 if (drbd_debugfs_init())
3026 pr_notice("failed to initialize debugfs -- will not be available\n");
3028 pr_info("initialized. "
3029 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
3030 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
3031 pr_info("%s\n", drbd_buildtag());
3032 pr_info("registered as block device major %d\n", DRBD_MAJOR);
3033 return 0; /* Success! */
3035 fail:
3036 drbd_cleanup();
3037 if (err == -ENOMEM)
3038 pr_err("ran out of memory\n");
3039 else
3040 pr_err("initialization failure\n");
3041 return err;
3044 static void drbd_free_one_sock(struct drbd_socket *ds)
3046 struct socket *s;
3047 mutex_lock(&ds->mutex);
3048 s = ds->socket;
3049 ds->socket = NULL;
3050 mutex_unlock(&ds->mutex);
3051 if (s) {
3052 /* so debugfs does not need to mutex_lock() */
3053 synchronize_rcu();
3054 kernel_sock_shutdown(s, SHUT_RDWR);
3055 sock_release(s);
3059 void drbd_free_sock(struct drbd_connection *connection)
3061 if (connection->data.socket)
3062 drbd_free_one_sock(&connection->data);
3063 if (connection->meta.socket)
3064 drbd_free_one_sock(&connection->meta);
3067 /* meta data management */
3069 void conn_md_sync(struct drbd_connection *connection)
3071 struct drbd_peer_device *peer_device;
3072 int vnr;
3074 rcu_read_lock();
3075 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3076 struct drbd_device *device = peer_device->device;
3078 kref_get(&device->kref);
3079 rcu_read_unlock();
3080 drbd_md_sync(device);
3081 kref_put(&device->kref, drbd_destroy_device);
3082 rcu_read_lock();
3084 rcu_read_unlock();
3087 /* aligned 4kByte */
3088 struct meta_data_on_disk {
3089 u64 la_size_sect; /* last agreed size. */
3090 u64 uuid[UI_SIZE]; /* UUIDs. */
3091 u64 device_uuid;
3092 u64 reserved_u64_1;
3093 u32 flags; /* MDF */
3094 u32 magic;
3095 u32 md_size_sect;
3096 u32 al_offset; /* offset to this block */
3097 u32 al_nr_extents; /* important for restoring the AL (userspace) */
3098 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3099 u32 bm_offset; /* offset to the bitmap, from here */
3100 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
3101 u32 la_peer_max_bio_size; /* last peer max_bio_size */
3103 /* see al_tr_number_to_on_disk_sector() */
3104 u32 al_stripes;
3105 u32 al_stripe_size_4k;
3107 u8 reserved_u8[4096 - (7*8 + 10*4)];
3108 } __packed;
3112 void drbd_md_write(struct drbd_device *device, void *b)
3114 struct meta_data_on_disk *buffer = b;
3115 sector_t sector;
3116 int i;
3118 memset(buffer, 0, sizeof(*buffer));
3120 buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3121 for (i = UI_CURRENT; i < UI_SIZE; i++)
3122 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3123 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3124 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3126 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3127 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3128 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3129 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3130 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3132 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3133 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3135 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3136 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3138 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3139 sector = device->ldev->md.md_offset;
3141 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3142 /* this was a try anyways ... */
3143 drbd_err(device, "meta data update failed!\n");
3144 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3149 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3150 * @device: DRBD device.
3152 void drbd_md_sync(struct drbd_device *device)
3154 struct meta_data_on_disk *buffer;
3156 /* Don't accidentally change the DRBD meta data layout. */
3157 BUILD_BUG_ON(UI_SIZE != 4);
3158 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3160 del_timer(&device->md_sync_timer);
3161 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3162 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3163 return;
3165 /* We use here D_FAILED and not D_ATTACHING because we try to write
3166 * metadata even if we detach due to a disk failure! */
3167 if (!get_ldev_if_state(device, D_FAILED))
3168 return;
3170 buffer = drbd_md_get_buffer(device, __func__);
3171 if (!buffer)
3172 goto out;
3174 drbd_md_write(device, buffer);
3176 /* Update device->ldev->md.la_size_sect,
3177 * since we updated it on metadata. */
3178 device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3180 drbd_md_put_buffer(device);
3181 out:
3182 put_ldev(device);
3185 static int check_activity_log_stripe_size(struct drbd_device *device,
3186 struct meta_data_on_disk *on_disk,
3187 struct drbd_md *in_core)
3189 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3190 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3191 u64 al_size_4k;
3193 /* both not set: default to old fixed size activity log */
3194 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3195 al_stripes = 1;
3196 al_stripe_size_4k = MD_32kB_SECT/8;
3199 /* some paranoia plausibility checks */
3201 /* we need both values to be set */
3202 if (al_stripes == 0 || al_stripe_size_4k == 0)
3203 goto err;
3205 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3207 /* Upper limit of activity log area, to avoid potential overflow
3208 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3209 * than 72 * 4k blocks total only increases the amount of history,
3210 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3211 if (al_size_4k > (16 * 1024 * 1024/4))
3212 goto err;
3214 /* Lower limit: we need at least 8 transaction slots (32kB)
3215 * to not break existing setups */
3216 if (al_size_4k < MD_32kB_SECT/8)
3217 goto err;
3219 in_core->al_stripe_size_4k = al_stripe_size_4k;
3220 in_core->al_stripes = al_stripes;
3221 in_core->al_size_4k = al_size_4k;
3223 return 0;
3224 err:
3225 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3226 al_stripes, al_stripe_size_4k);
3227 return -EINVAL;
3230 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3232 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3233 struct drbd_md *in_core = &bdev->md;
3234 s32 on_disk_al_sect;
3235 s32 on_disk_bm_sect;
3237 /* The on-disk size of the activity log, calculated from offsets, and
3238 * the size of the activity log calculated from the stripe settings,
3239 * should match.
3240 * Though we could relax this a bit: it is ok, if the striped activity log
3241 * fits in the available on-disk activity log size.
3242 * Right now, that would break how resize is implemented.
3243 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3244 * of possible unused padding space in the on disk layout. */
3245 if (in_core->al_offset < 0) {
3246 if (in_core->bm_offset > in_core->al_offset)
3247 goto err;
3248 on_disk_al_sect = -in_core->al_offset;
3249 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3250 } else {
3251 if (in_core->al_offset != MD_4kB_SECT)
3252 goto err;
3253 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3254 goto err;
3256 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3257 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3260 /* old fixed size meta data is exactly that: fixed. */
3261 if (in_core->meta_dev_idx >= 0) {
3262 if (in_core->md_size_sect != MD_128MB_SECT
3263 || in_core->al_offset != MD_4kB_SECT
3264 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3265 || in_core->al_stripes != 1
3266 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3267 goto err;
3270 if (capacity < in_core->md_size_sect)
3271 goto err;
3272 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3273 goto err;
3275 /* should be aligned, and at least 32k */
3276 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3277 goto err;
3279 /* should fit (for now: exactly) into the available on-disk space;
3280 * overflow prevention is in check_activity_log_stripe_size() above. */
3281 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3282 goto err;
3284 /* again, should be aligned */
3285 if (in_core->bm_offset & 7)
3286 goto err;
3288 /* FIXME check for device grow with flex external meta data? */
3290 /* can the available bitmap space cover the last agreed device size? */
3291 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3292 goto err;
3294 return 0;
3296 err:
3297 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3298 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3299 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3300 in_core->meta_dev_idx,
3301 in_core->al_stripes, in_core->al_stripe_size_4k,
3302 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3303 (unsigned long long)in_core->la_size_sect,
3304 (unsigned long long)capacity);
3306 return -EINVAL;
3311 * drbd_md_read() - Reads in the meta data super block
3312 * @device: DRBD device.
3313 * @bdev: Device from which the meta data should be read in.
3315 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3316 * something goes wrong.
3318 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3319 * even before @bdev is assigned to @device->ldev.
3321 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3323 struct meta_data_on_disk *buffer;
3324 u32 magic, flags;
3325 int i, rv = NO_ERROR;
3327 if (device->state.disk != D_DISKLESS)
3328 return ERR_DISK_CONFIGURED;
3330 buffer = drbd_md_get_buffer(device, __func__);
3331 if (!buffer)
3332 return ERR_NOMEM;
3334 /* First, figure out where our meta data superblock is located,
3335 * and read it. */
3336 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3337 bdev->md.md_offset = drbd_md_ss(bdev);
3338 /* Even for (flexible or indexed) external meta data,
3339 * initially restrict us to the 4k superblock for now.
3340 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3341 bdev->md.md_size_sect = 8;
3343 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3344 REQ_OP_READ)) {
3345 /* NOTE: can't do normal error processing here as this is
3346 called BEFORE disk is attached */
3347 drbd_err(device, "Error while reading metadata.\n");
3348 rv = ERR_IO_MD_DISK;
3349 goto err;
3352 magic = be32_to_cpu(buffer->magic);
3353 flags = be32_to_cpu(buffer->flags);
3354 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3355 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3356 /* btw: that's Activity Log clean, not "all" clean. */
3357 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3358 rv = ERR_MD_UNCLEAN;
3359 goto err;
3362 rv = ERR_MD_INVALID;
3363 if (magic != DRBD_MD_MAGIC_08) {
3364 if (magic == DRBD_MD_MAGIC_07)
3365 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3366 else
3367 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3368 goto err;
3371 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3372 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3373 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3374 goto err;
3378 /* convert to in_core endian */
3379 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3380 for (i = UI_CURRENT; i < UI_SIZE; i++)
3381 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3382 bdev->md.flags = be32_to_cpu(buffer->flags);
3383 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3385 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3386 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3387 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3389 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3390 goto err;
3391 if (check_offsets_and_sizes(device, bdev))
3392 goto err;
3394 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3395 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3396 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3397 goto err;
3399 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3400 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3401 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3402 goto err;
3405 rv = NO_ERROR;
3407 spin_lock_irq(&device->resource->req_lock);
3408 if (device->state.conn < C_CONNECTED) {
3409 unsigned int peer;
3410 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3411 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3412 device->peer_max_bio_size = peer;
3414 spin_unlock_irq(&device->resource->req_lock);
3416 err:
3417 drbd_md_put_buffer(device);
3419 return rv;
3423 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3424 * @device: DRBD device.
3426 * Call this function if you change anything that should be written to
3427 * the meta-data super block. This function sets MD_DIRTY, and starts a
3428 * timer that ensures that within five seconds you have to call drbd_md_sync().
3430 #ifdef DEBUG
3431 void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
3433 if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3434 mod_timer(&device->md_sync_timer, jiffies + HZ);
3435 device->last_md_mark_dirty.line = line;
3436 device->last_md_mark_dirty.func = func;
3439 #else
3440 void drbd_md_mark_dirty(struct drbd_device *device)
3442 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3443 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3445 #endif
3447 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3449 int i;
3451 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3452 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3455 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3457 if (idx == UI_CURRENT) {
3458 if (device->state.role == R_PRIMARY)
3459 val |= 1;
3460 else
3461 val &= ~((u64)1);
3463 drbd_set_ed_uuid(device, val);
3466 device->ldev->md.uuid[idx] = val;
3467 drbd_md_mark_dirty(device);
3470 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3472 unsigned long flags;
3473 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3474 __drbd_uuid_set(device, idx, val);
3475 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3478 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3480 unsigned long flags;
3481 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3482 if (device->ldev->md.uuid[idx]) {
3483 drbd_uuid_move_history(device);
3484 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3486 __drbd_uuid_set(device, idx, val);
3487 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3491 * drbd_uuid_new_current() - Creates a new current UUID
3492 * @device: DRBD device.
3494 * Creates a new current UUID, and rotates the old current UUID into
3495 * the bitmap slot. Causes an incremental resync upon next connect.
3497 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3499 u64 val;
3500 unsigned long long bm_uuid;
3502 get_random_bytes(&val, sizeof(u64));
3504 spin_lock_irq(&device->ldev->md.uuid_lock);
3505 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3507 if (bm_uuid)
3508 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3510 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3511 __drbd_uuid_set(device, UI_CURRENT, val);
3512 spin_unlock_irq(&device->ldev->md.uuid_lock);
3514 drbd_print_uuids(device, "new current UUID");
3515 /* get it to stable storage _now_ */
3516 drbd_md_sync(device);
3519 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3521 unsigned long flags;
3522 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3523 return;
3525 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3526 if (val == 0) {
3527 drbd_uuid_move_history(device);
3528 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3529 device->ldev->md.uuid[UI_BITMAP] = 0;
3530 } else {
3531 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3532 if (bm_uuid)
3533 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3535 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3537 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3539 drbd_md_mark_dirty(device);
3543 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3544 * @device: DRBD device.
3546 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3548 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3550 int rv = -EIO;
3552 drbd_md_set_flag(device, MDF_FULL_SYNC);
3553 drbd_md_sync(device);
3554 drbd_bm_set_all(device);
3556 rv = drbd_bm_write(device);
3558 if (!rv) {
3559 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3560 drbd_md_sync(device);
3563 return rv;
3567 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3568 * @device: DRBD device.
3570 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3572 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3574 drbd_resume_al(device);
3575 drbd_bm_clear_all(device);
3576 return drbd_bm_write(device);
3579 static int w_bitmap_io(struct drbd_work *w, int unused)
3581 struct drbd_device *device =
3582 container_of(w, struct drbd_device, bm_io_work.w);
3583 struct bm_io_work *work = &device->bm_io_work;
3584 int rv = -EIO;
3586 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3587 int cnt = atomic_read(&device->ap_bio_cnt);
3588 if (cnt)
3589 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3590 cnt, work->why);
3593 if (get_ldev(device)) {
3594 drbd_bm_lock(device, work->why, work->flags);
3595 rv = work->io_fn(device);
3596 drbd_bm_unlock(device);
3597 put_ldev(device);
3600 clear_bit_unlock(BITMAP_IO, &device->flags);
3601 wake_up(&device->misc_wait);
3603 if (work->done)
3604 work->done(device, rv);
3606 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3607 work->why = NULL;
3608 work->flags = 0;
3610 return 0;
3614 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3615 * @device: DRBD device.
3616 * @io_fn: IO callback to be called when bitmap IO is possible
3617 * @done: callback to be called after the bitmap IO was performed
3618 * @why: Descriptive text of the reason for doing the IO
3620 * While IO on the bitmap happens we freeze application IO thus we ensure
3621 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3622 * called from worker context. It MUST NOT be used while a previous such
3623 * work is still pending!
3625 * Its worker function encloses the call of io_fn() by get_ldev() and
3626 * put_ldev().
3628 void drbd_queue_bitmap_io(struct drbd_device *device,
3629 int (*io_fn)(struct drbd_device *),
3630 void (*done)(struct drbd_device *, int),
3631 char *why, enum bm_flag flags)
3633 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3635 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3636 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3637 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3638 if (device->bm_io_work.why)
3639 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3640 why, device->bm_io_work.why);
3642 device->bm_io_work.io_fn = io_fn;
3643 device->bm_io_work.done = done;
3644 device->bm_io_work.why = why;
3645 device->bm_io_work.flags = flags;
3647 spin_lock_irq(&device->resource->req_lock);
3648 set_bit(BITMAP_IO, &device->flags);
3649 /* don't wait for pending application IO if the caller indicates that
3650 * application IO does not conflict anyways. */
3651 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3652 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3653 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3654 &device->bm_io_work.w);
3656 spin_unlock_irq(&device->resource->req_lock);
3660 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3661 * @device: DRBD device.
3662 * @io_fn: IO callback to be called when bitmap IO is possible
3663 * @why: Descriptive text of the reason for doing the IO
3665 * freezes application IO while that the actual IO operations runs. This
3666 * functions MAY NOT be called from worker context.
3668 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3669 char *why, enum bm_flag flags)
3671 /* Only suspend io, if some operation is supposed to be locked out */
3672 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3673 int rv;
3675 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3677 if (do_suspend_io)
3678 drbd_suspend_io(device);
3680 drbd_bm_lock(device, why, flags);
3681 rv = io_fn(device);
3682 drbd_bm_unlock(device);
3684 if (do_suspend_io)
3685 drbd_resume_io(device);
3687 return rv;
3690 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3692 if ((device->ldev->md.flags & flag) != flag) {
3693 drbd_md_mark_dirty(device);
3694 device->ldev->md.flags |= flag;
3698 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3700 if ((device->ldev->md.flags & flag) != 0) {
3701 drbd_md_mark_dirty(device);
3702 device->ldev->md.flags &= ~flag;
3705 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3707 return (bdev->md.flags & flag) != 0;
3710 static void md_sync_timer_fn(unsigned long data)
3712 struct drbd_device *device = (struct drbd_device *) data;
3713 drbd_device_post_work(device, MD_SYNC);
3716 const char *cmdname(enum drbd_packet cmd)
3718 /* THINK may need to become several global tables
3719 * when we want to support more than
3720 * one PRO_VERSION */
3721 static const char *cmdnames[] = {
3722 [P_DATA] = "Data",
3723 [P_WSAME] = "WriteSame",
3724 [P_TRIM] = "Trim",
3725 [P_DATA_REPLY] = "DataReply",
3726 [P_RS_DATA_REPLY] = "RSDataReply",
3727 [P_BARRIER] = "Barrier",
3728 [P_BITMAP] = "ReportBitMap",
3729 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3730 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3731 [P_UNPLUG_REMOTE] = "UnplugRemote",
3732 [P_DATA_REQUEST] = "DataRequest",
3733 [P_RS_DATA_REQUEST] = "RSDataRequest",
3734 [P_SYNC_PARAM] = "SyncParam",
3735 [P_SYNC_PARAM89] = "SyncParam89",
3736 [P_PROTOCOL] = "ReportProtocol",
3737 [P_UUIDS] = "ReportUUIDs",
3738 [P_SIZES] = "ReportSizes",
3739 [P_STATE] = "ReportState",
3740 [P_SYNC_UUID] = "ReportSyncUUID",
3741 [P_AUTH_CHALLENGE] = "AuthChallenge",
3742 [P_AUTH_RESPONSE] = "AuthResponse",
3743 [P_PING] = "Ping",
3744 [P_PING_ACK] = "PingAck",
3745 [P_RECV_ACK] = "RecvAck",
3746 [P_WRITE_ACK] = "WriteAck",
3747 [P_RS_WRITE_ACK] = "RSWriteAck",
3748 [P_SUPERSEDED] = "Superseded",
3749 [P_NEG_ACK] = "NegAck",
3750 [P_NEG_DREPLY] = "NegDReply",
3751 [P_NEG_RS_DREPLY] = "NegRSDReply",
3752 [P_BARRIER_ACK] = "BarrierAck",
3753 [P_STATE_CHG_REQ] = "StateChgRequest",
3754 [P_STATE_CHG_REPLY] = "StateChgReply",
3755 [P_OV_REQUEST] = "OVRequest",
3756 [P_OV_REPLY] = "OVReply",
3757 [P_OV_RESULT] = "OVResult",
3758 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3759 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3760 [P_COMPRESSED_BITMAP] = "CBitmap",
3761 [P_DELAY_PROBE] = "DelayProbe",
3762 [P_OUT_OF_SYNC] = "OutOfSync",
3763 [P_RETRY_WRITE] = "RetryWrite",
3764 [P_RS_CANCEL] = "RSCancel",
3765 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3766 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3767 [P_RETRY_WRITE] = "retry_write",
3768 [P_PROTOCOL_UPDATE] = "protocol_update",
3769 [P_RS_THIN_REQ] = "rs_thin_req",
3770 [P_RS_DEALLOCATED] = "rs_deallocated",
3772 /* enum drbd_packet, but not commands - obsoleted flags:
3773 * P_MAY_IGNORE
3774 * P_MAX_OPT_CMD
3778 /* too big for the array: 0xfffX */
3779 if (cmd == P_INITIAL_META)
3780 return "InitialMeta";
3781 if (cmd == P_INITIAL_DATA)
3782 return "InitialData";
3783 if (cmd == P_CONNECTION_FEATURES)
3784 return "ConnectionFeatures";
3785 if (cmd >= ARRAY_SIZE(cmdnames))
3786 return "Unknown";
3787 return cmdnames[cmd];
3791 * drbd_wait_misc - wait for a request to make progress
3792 * @device: device associated with the request
3793 * @i: the struct drbd_interval embedded in struct drbd_request or
3794 * struct drbd_peer_request
3796 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3798 struct net_conf *nc;
3799 DEFINE_WAIT(wait);
3800 long timeout;
3802 rcu_read_lock();
3803 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3804 if (!nc) {
3805 rcu_read_unlock();
3806 return -ETIMEDOUT;
3808 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3809 rcu_read_unlock();
3811 /* Indicate to wake up device->misc_wait on progress. */
3812 i->waiting = true;
3813 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3814 spin_unlock_irq(&device->resource->req_lock);
3815 timeout = schedule_timeout(timeout);
3816 finish_wait(&device->misc_wait, &wait);
3817 spin_lock_irq(&device->resource->req_lock);
3818 if (!timeout || device->state.conn < C_CONNECTED)
3819 return -ETIMEDOUT;
3820 if (signal_pending(current))
3821 return -ERESTARTSYS;
3822 return 0;
3825 void lock_all_resources(void)
3827 struct drbd_resource *resource;
3828 int __maybe_unused i = 0;
3830 mutex_lock(&resources_mutex);
3831 local_irq_disable();
3832 for_each_resource(resource, &drbd_resources)
3833 spin_lock_nested(&resource->req_lock, i++);
3836 void unlock_all_resources(void)
3838 struct drbd_resource *resource;
3840 for_each_resource(resource, &drbd_resources)
3841 spin_unlock(&resource->req_lock);
3842 local_irq_enable();
3843 mutex_unlock(&resources_mutex);
3846 #ifdef CONFIG_DRBD_FAULT_INJECTION
3847 /* Fault insertion support including random number generator shamelessly
3848 * stolen from kernel/rcutorture.c */
3849 struct fault_random_state {
3850 unsigned long state;
3851 unsigned long count;
3854 #define FAULT_RANDOM_MULT 39916801 /* prime */
3855 #define FAULT_RANDOM_ADD 479001701 /* prime */
3856 #define FAULT_RANDOM_REFRESH 10000
3859 * Crude but fast random-number generator. Uses a linear congruential
3860 * generator, with occasional help from get_random_bytes().
3862 static unsigned long
3863 _drbd_fault_random(struct fault_random_state *rsp)
3865 long refresh;
3867 if (!rsp->count--) {
3868 get_random_bytes(&refresh, sizeof(refresh));
3869 rsp->state += refresh;
3870 rsp->count = FAULT_RANDOM_REFRESH;
3872 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3873 return swahw32(rsp->state);
3876 static char *
3877 _drbd_fault_str(unsigned int type) {
3878 static char *_faults[] = {
3879 [DRBD_FAULT_MD_WR] = "Meta-data write",
3880 [DRBD_FAULT_MD_RD] = "Meta-data read",
3881 [DRBD_FAULT_RS_WR] = "Resync write",
3882 [DRBD_FAULT_RS_RD] = "Resync read",
3883 [DRBD_FAULT_DT_WR] = "Data write",
3884 [DRBD_FAULT_DT_RD] = "Data read",
3885 [DRBD_FAULT_DT_RA] = "Data read ahead",
3886 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3887 [DRBD_FAULT_AL_EE] = "EE allocation",
3888 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3891 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3894 unsigned int
3895 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3897 static struct fault_random_state rrs = {0, 0};
3899 unsigned int ret = (
3900 (fault_devs == 0 ||
3901 ((1 << device_to_minor(device)) & fault_devs) != 0) &&
3902 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3904 if (ret) {
3905 fault_count++;
3907 if (__ratelimit(&drbd_ratelimit_state))
3908 drbd_warn(device, "***Simulating %s failure\n",
3909 _drbd_fault_str(type));
3912 return ret;
3914 #endif
3916 const char *drbd_buildtag(void)
3918 /* DRBD built from external sources has here a reference to the
3919 git hash of the source code. */
3921 static char buildtag[38] = "\0uilt-in";
3923 if (buildtag[0] == 0) {
3924 #ifdef MODULE
3925 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3926 #else
3927 buildtag[0] = 'b';
3928 #endif
3931 return buildtag;
3934 module_init(drbd_init)
3935 module_exit(drbd_cleanup)
3937 EXPORT_SYMBOL(drbd_conn_str);
3938 EXPORT_SYMBOL(drbd_role_str);
3939 EXPORT_SYMBOL(drbd_disk_str);
3940 EXPORT_SYMBOL(drbd_set_st_err_str);