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)
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 #include <linux/module.h>
30 #include <linux/drbd.h>
31 #include <asm/uaccess.h>
32 #include <asm/types.h>
34 #include <linux/ctype.h>
35 #include <linux/mutex.h>
37 #include <linux/file.h>
38 #include <linux/proc_fs.h>
39 #include <linux/init.h>
41 #include <linux/memcontrol.h>
42 #include <linux/mm_inline.h>
43 #include <linux/slab.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/notifier.h>
47 #include <linux/kthread.h>
49 #define __KERNEL_SYSCALLS__
50 #include <linux/unistd.h>
51 #include <linux/vmalloc.h>
53 #include <linux/drbd_limits.h>
55 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
59 static DEFINE_MUTEX(drbd_main_mutex);
60 int drbdd_init(struct drbd_thread *);
61 int drbd_worker(struct drbd_thread *);
62 int drbd_asender(struct drbd_thread *);
65 static int drbd_open(struct block_device *bdev, fmode_t mode);
66 static int drbd_release(struct gendisk *gd, fmode_t mode);
67 static int w_md_sync(struct drbd_work *w, int unused);
68 static void md_sync_timer_fn(unsigned long data);
69 static int w_bitmap_io(struct drbd_work *w, int unused);
70 static int w_go_diskless(struct drbd_work *w, int unused);
72 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
73 "Lars Ellenberg <lars@linbit.com>");
74 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
75 MODULE_VERSION(REL_VERSION);
76 MODULE_LICENSE("GPL");
77 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
78 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
79 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
81 #include <linux/moduleparam.h>
82 /* allow_open_on_secondary */
83 MODULE_PARM_DESC(allow_oos, "DONT USE!");
84 /* thanks to these macros, if compiled into the kernel (not-module),
85 * this becomes the boot parameter drbd.minor_count */
86 module_param(minor_count, uint, 0444);
87 module_param(disable_sendpage, bool, 0644);
88 module_param(allow_oos, bool, 0);
89 module_param(proc_details, int, 0644);
91 #ifdef CONFIG_DRBD_FAULT_INJECTION
94 static int fault_count;
96 /* bitmap of enabled faults */
97 module_param(enable_faults, int, 0664);
98 /* fault rate % value - applies to all enabled faults */
99 module_param(fault_rate, int, 0664);
100 /* count of faults inserted */
101 module_param(fault_count, int, 0664);
102 /* bitmap of devices to insert faults on */
103 module_param(fault_devs, int, 0644);
106 /* module parameter, defined */
107 unsigned int minor_count = DRBD_MINOR_COUNT_DEF;
108 int disable_sendpage;
110 int proc_details; /* Detail level in proc drbd*/
112 /* Module parameter for setting the user mode helper program
113 * to run. Default is /sbin/drbdadm */
114 char usermode_helper[80] = "/sbin/drbdadm";
116 module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644);
118 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
119 * as member "struct gendisk *vdisk;"
122 struct list_head drbd_tconns; /* list of struct drbd_tconn */
123 DEFINE_MUTEX(drbd_cfg_mutex);
125 struct kmem_cache *drbd_request_cache;
126 struct kmem_cache *drbd_ee_cache; /* peer requests */
127 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
128 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
129 mempool_t *drbd_request_mempool;
130 mempool_t *drbd_ee_mempool;
131 mempool_t *drbd_md_io_page_pool;
132 struct bio_set *drbd_md_io_bio_set;
134 /* I do not use a standard mempool, because:
135 1) I want to hand out the pre-allocated objects first.
136 2) I want to be able to interrupt sleeping allocation with a signal.
137 Note: This is a single linked list, the next pointer is the private
138 member of struct page.
140 struct page *drbd_pp_pool;
141 spinlock_t drbd_pp_lock;
143 wait_queue_head_t drbd_pp_wait;
145 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
147 static const struct block_device_operations drbd_ops = {
148 .owner = THIS_MODULE,
150 .release = drbd_release,
153 static void bio_destructor_drbd(struct bio *bio)
155 bio_free(bio, drbd_md_io_bio_set);
158 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
162 if (!drbd_md_io_bio_set)
163 return bio_alloc(gfp_mask, 1);
165 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
168 bio->bi_destructor = bio_destructor_drbd;
173 /* When checking with sparse, and this is an inline function, sparse will
174 give tons of false positives. When this is a real functions sparse works.
176 int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins)
180 atomic_inc(&mdev->local_cnt);
181 io_allowed = (mdev->state.disk >= mins);
183 if (atomic_dec_and_test(&mdev->local_cnt))
184 wake_up(&mdev->misc_wait);
192 * DOC: The transfer log
194 * The transfer log is a single linked list of &struct drbd_tl_epoch objects.
195 * mdev->tconn->newest_tle points to the head, mdev->tconn->oldest_tle points to the tail
196 * of the list. There is always at least one &struct drbd_tl_epoch object.
198 * Each &struct drbd_tl_epoch has a circular double linked list of requests
201 static int tl_init(struct drbd_tconn *tconn)
203 struct drbd_tl_epoch *b;
205 /* during device minor initialization, we may well use GFP_KERNEL */
206 b = kmalloc(sizeof(struct drbd_tl_epoch), GFP_KERNEL);
209 INIT_LIST_HEAD(&b->requests);
210 INIT_LIST_HEAD(&b->w.list);
214 b->w.cb = NULL; /* if this is != NULL, we need to dec_ap_pending in tl_clear */
216 tconn->oldest_tle = b;
217 tconn->newest_tle = b;
218 INIT_LIST_HEAD(&tconn->out_of_sequence_requests);
223 static void tl_cleanup(struct drbd_tconn *tconn)
225 if (tconn->oldest_tle != tconn->newest_tle)
226 conn_err(tconn, "ASSERT FAILED: oldest_tle == newest_tle\n");
227 if (!list_empty(&tconn->out_of_sequence_requests))
228 conn_err(tconn, "ASSERT FAILED: list_empty(out_of_sequence_requests)\n");
229 kfree(tconn->oldest_tle);
230 tconn->oldest_tle = NULL;
231 kfree(tconn->unused_spare_tle);
232 tconn->unused_spare_tle = NULL;
236 * _tl_add_barrier() - Adds a barrier to the transfer log
237 * @mdev: DRBD device.
238 * @new: Barrier to be added before the current head of the TL.
240 * The caller must hold the req_lock.
242 void _tl_add_barrier(struct drbd_tconn *tconn, struct drbd_tl_epoch *new)
244 struct drbd_tl_epoch *newest_before;
246 INIT_LIST_HEAD(&new->requests);
247 INIT_LIST_HEAD(&new->w.list);
248 new->w.cb = NULL; /* if this is != NULL, we need to dec_ap_pending in tl_clear */
252 newest_before = tconn->newest_tle;
253 /* never send a barrier number == 0, because that is special-cased
254 * when using TCQ for our write ordering code */
255 new->br_number = (newest_before->br_number+1) ?: 1;
256 if (tconn->newest_tle != new) {
257 tconn->newest_tle->next = new;
258 tconn->newest_tle = new;
263 * tl_release() - Free or recycle the oldest &struct drbd_tl_epoch object of the TL
264 * @mdev: DRBD device.
265 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
266 * @set_size: Expected number of requests before that barrier.
268 * In case the passed barrier_nr or set_size does not match the oldest
269 * &struct drbd_tl_epoch objects this function will cause a termination
272 void tl_release(struct drbd_tconn *tconn, unsigned int barrier_nr,
273 unsigned int set_size)
275 struct drbd_conf *mdev;
276 struct drbd_tl_epoch *b, *nob; /* next old barrier */
277 struct list_head *le, *tle;
278 struct drbd_request *r;
280 spin_lock_irq(&tconn->req_lock);
282 b = tconn->oldest_tle;
284 /* first some paranoia code */
286 conn_err(tconn, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
290 if (b->br_number != barrier_nr) {
291 conn_err(tconn, "BAD! BarrierAck #%u received, expected #%u!\n",
292 barrier_nr, b->br_number);
295 if (b->n_writes != set_size) {
296 conn_err(tconn, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
297 barrier_nr, set_size, b->n_writes);
301 /* Clean up list of requests processed during current epoch */
302 list_for_each_safe(le, tle, &b->requests) {
303 r = list_entry(le, struct drbd_request, tl_requests);
304 _req_mod(r, BARRIER_ACKED);
306 /* There could be requests on the list waiting for completion
307 of the write to the local disk. To avoid corruptions of
308 slab's data structures we have to remove the lists head.
310 Also there could have been a barrier ack out of sequence, overtaking
311 the write acks - which would be a bug and violating write ordering.
312 To not deadlock in case we lose connection while such requests are
313 still pending, we need some way to find them for the
314 _req_mode(CONNECTION_LOST_WHILE_PENDING).
316 These have been list_move'd to the out_of_sequence_requests list in
317 _req_mod(, BARRIER_ACKED) above.
319 list_del_init(&b->requests);
323 if (test_and_clear_bit(CREATE_BARRIER, &mdev->flags)) {
324 _tl_add_barrier(tconn, b);
326 tconn->oldest_tle = nob;
327 /* if nob == NULL b was the only barrier, and becomes the new
328 barrier. Therefore tconn->oldest_tle points already to b */
330 D_ASSERT(nob != NULL);
331 tconn->oldest_tle = nob;
335 spin_unlock_irq(&tconn->req_lock);
336 dec_ap_pending(mdev);
341 spin_unlock_irq(&tconn->req_lock);
342 conn_request_state(tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
347 * _tl_restart() - Walks the transfer log, and applies an action to all requests
348 * @mdev: DRBD device.
349 * @what: The action/event to perform with all request objects
351 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
352 * RESTART_FROZEN_DISK_IO.
354 void _tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what)
356 struct drbd_tl_epoch *b, *tmp, **pn;
357 struct list_head *le, *tle, carry_reads;
358 struct drbd_request *req;
359 int rv, n_writes, n_reads;
361 b = tconn->oldest_tle;
362 pn = &tconn->oldest_tle;
366 INIT_LIST_HEAD(&carry_reads);
367 list_for_each_safe(le, tle, &b->requests) {
368 req = list_entry(le, struct drbd_request, tl_requests);
369 rv = _req_mod(req, what);
371 n_writes += (rv & MR_WRITE) >> MR_WRITE_SHIFT;
372 n_reads += (rv & MR_READ) >> MR_READ_SHIFT;
377 if (what == RESEND) {
378 b->n_writes = n_writes;
379 if (b->w.cb == NULL) {
380 b->w.cb = w_send_barrier;
381 inc_ap_pending(b->w.mdev);
382 set_bit(CREATE_BARRIER, &b->w.mdev->flags);
385 drbd_queue_work(&tconn->data.work, &b->w);
390 list_add(&carry_reads, &b->requests);
391 /* there could still be requests on that ring list,
392 * in case local io is still pending */
393 list_del(&b->requests);
395 /* dec_ap_pending corresponding to queue_barrier.
396 * the newest barrier may not have been queued yet,
397 * in which case w.cb is still NULL. */
399 dec_ap_pending(b->w.mdev);
401 if (b == tconn->newest_tle) {
402 /* recycle, but reinit! */
404 conn_err(tconn, "ASSERT FAILED tmp == NULL");
405 INIT_LIST_HEAD(&b->requests);
406 list_splice(&carry_reads, &b->requests);
407 INIT_LIST_HEAD(&b->w.list);
409 b->br_number = net_random();
419 list_splice(&carry_reads, &b->requests);
425 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
426 * @mdev: DRBD device.
428 * This is called after the connection to the peer was lost. The storage covered
429 * by the requests on the transfer gets marked as our of sync. Called from the
430 * receiver thread and the worker thread.
432 void tl_clear(struct drbd_tconn *tconn)
434 struct drbd_conf *mdev;
435 struct list_head *le, *tle;
436 struct drbd_request *r;
439 spin_lock_irq(&tconn->req_lock);
441 _tl_restart(tconn, CONNECTION_LOST_WHILE_PENDING);
443 /* we expect this list to be empty. */
444 if (!list_empty(&tconn->out_of_sequence_requests))
445 conn_err(tconn, "ASSERT FAILED list_empty(&out_of_sequence_requests)\n");
447 /* but just in case, clean it up anyways! */
448 list_for_each_safe(le, tle, &tconn->out_of_sequence_requests) {
449 r = list_entry(le, struct drbd_request, tl_requests);
450 /* It would be nice to complete outside of spinlock.
451 * But this is easier for now. */
452 _req_mod(r, CONNECTION_LOST_WHILE_PENDING);
455 /* ensure bit indicating barrier is required is clear */
456 idr_for_each_entry(&tconn->volumes, mdev, vnr)
457 clear_bit(CREATE_BARRIER, &mdev->flags);
459 spin_unlock_irq(&tconn->req_lock);
462 void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what)
464 spin_lock_irq(&tconn->req_lock);
465 _tl_restart(tconn, what);
466 spin_unlock_irq(&tconn->req_lock);
469 static int drbd_thread_setup(void *arg)
471 struct drbd_thread *thi = (struct drbd_thread *) arg;
472 struct drbd_tconn *tconn = thi->tconn;
476 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
477 thi->name[0], thi->tconn->name);
480 retval = thi->function(thi);
482 spin_lock_irqsave(&thi->t_lock, flags);
484 /* if the receiver has been "EXITING", the last thing it did
485 * was set the conn state to "StandAlone",
486 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
487 * and receiver thread will be "started".
488 * drbd_thread_start needs to set "RESTARTING" in that case.
489 * t_state check and assignment needs to be within the same spinlock,
490 * so either thread_start sees EXITING, and can remap to RESTARTING,
491 * or thread_start see NONE, and can proceed as normal.
494 if (thi->t_state == RESTARTING) {
495 conn_info(tconn, "Restarting %s thread\n", thi->name);
496 thi->t_state = RUNNING;
497 spin_unlock_irqrestore(&thi->t_lock, flags);
504 complete(&thi->stop);
505 spin_unlock_irqrestore(&thi->t_lock, flags);
507 conn_info(tconn, "Terminating %s\n", current->comm);
509 /* Release mod reference taken when thread was started */
510 module_put(THIS_MODULE);
514 static void drbd_thread_init(struct drbd_tconn *tconn, struct drbd_thread *thi,
515 int (*func) (struct drbd_thread *), char *name)
517 spin_lock_init(&thi->t_lock);
520 thi->function = func;
522 strncpy(thi->name, name, ARRAY_SIZE(thi->name));
525 int drbd_thread_start(struct drbd_thread *thi)
527 struct drbd_tconn *tconn = thi->tconn;
528 struct task_struct *nt;
531 /* is used from state engine doing drbd_thread_stop_nowait,
532 * while holding the req lock irqsave */
533 spin_lock_irqsave(&thi->t_lock, flags);
535 switch (thi->t_state) {
537 conn_info(tconn, "Starting %s thread (from %s [%d])\n",
538 thi->name, current->comm, current->pid);
540 /* Get ref on module for thread - this is released when thread exits */
541 if (!try_module_get(THIS_MODULE)) {
542 conn_err(tconn, "Failed to get module reference in drbd_thread_start\n");
543 spin_unlock_irqrestore(&thi->t_lock, flags);
547 init_completion(&thi->stop);
548 thi->reset_cpu_mask = 1;
549 thi->t_state = RUNNING;
550 spin_unlock_irqrestore(&thi->t_lock, flags);
551 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
553 nt = kthread_create(drbd_thread_setup, (void *) thi,
554 "drbd_%c_%s", thi->name[0], thi->tconn->name);
557 conn_err(tconn, "Couldn't start thread\n");
559 module_put(THIS_MODULE);
562 spin_lock_irqsave(&thi->t_lock, flags);
564 thi->t_state = RUNNING;
565 spin_unlock_irqrestore(&thi->t_lock, flags);
569 thi->t_state = RESTARTING;
570 conn_info(tconn, "Restarting %s thread (from %s [%d])\n",
571 thi->name, current->comm, current->pid);
576 spin_unlock_irqrestore(&thi->t_lock, flags);
584 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
588 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
590 /* may be called from state engine, holding the req lock irqsave */
591 spin_lock_irqsave(&thi->t_lock, flags);
593 if (thi->t_state == NONE) {
594 spin_unlock_irqrestore(&thi->t_lock, flags);
596 drbd_thread_start(thi);
600 if (thi->t_state != ns) {
601 if (thi->task == NULL) {
602 spin_unlock_irqrestore(&thi->t_lock, flags);
608 init_completion(&thi->stop);
609 if (thi->task != current)
610 force_sig(DRBD_SIGKILL, thi->task);
613 spin_unlock_irqrestore(&thi->t_lock, flags);
616 wait_for_completion(&thi->stop);
619 static struct drbd_thread *drbd_task_to_thread(struct drbd_tconn *tconn, struct task_struct *task)
621 struct drbd_thread *thi =
622 task == tconn->receiver.task ? &tconn->receiver :
623 task == tconn->asender.task ? &tconn->asender :
624 task == tconn->worker.task ? &tconn->worker : NULL;
629 char *drbd_task_to_thread_name(struct drbd_tconn *tconn, struct task_struct *task)
631 struct drbd_thread *thi = drbd_task_to_thread(tconn, task);
632 return thi ? thi->name : task->comm;
635 int conn_lowest_minor(struct drbd_tconn *tconn)
638 struct drbd_conf *mdev;
640 mdev = idr_get_next(&tconn->volumes, &vnr);
643 return mdev_to_minor(mdev);
648 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
649 * @mdev: DRBD device.
651 * Forces all threads of a device onto the same CPU. This is beneficial for
652 * DRBD's performance. May be overwritten by user's configuration.
654 void drbd_calc_cpu_mask(struct drbd_tconn *tconn)
659 if (cpumask_weight(tconn->cpu_mask))
662 ord = conn_lowest_minor(tconn) % cpumask_weight(cpu_online_mask);
663 for_each_online_cpu(cpu) {
665 cpumask_set_cpu(cpu, tconn->cpu_mask);
669 /* should not be reached */
670 cpumask_setall(tconn->cpu_mask);
674 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
675 * @mdev: DRBD device.
676 * @thi: drbd_thread object
678 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
681 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
683 struct task_struct *p = current;
685 if (!thi->reset_cpu_mask)
687 thi->reset_cpu_mask = 0;
688 set_cpus_allowed_ptr(p, thi->tconn->cpu_mask);
692 static void prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
694 h->magic = cpu_to_be32(DRBD_MAGIC);
695 h->command = cpu_to_be16(cmd);
696 h->length = cpu_to_be16(size);
699 static void prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
701 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
702 h->command = cpu_to_be16(cmd);
703 h->length = cpu_to_be32(size);
706 static void _prepare_header(struct drbd_tconn *tconn, int vnr, struct p_header *h,
707 enum drbd_packet cmd, int size)
709 if (tconn->agreed_pro_version >= 95)
710 prepare_header95(&h->h95, cmd, size);
712 prepare_header80(&h->h80, cmd, size);
715 static void prepare_header(struct drbd_conf *mdev, struct p_header *h,
716 enum drbd_packet cmd, int size)
718 _prepare_header(mdev->tconn, mdev->vnr, h, cmd, size);
721 /* the appropriate socket mutex must be held already */
722 int _conn_send_cmd(struct drbd_tconn *tconn, int vnr, struct socket *sock,
723 enum drbd_packet cmd, struct p_header *h, size_t size,
728 _prepare_header(tconn, vnr, h, cmd, size - sizeof(struct p_header));
729 err = drbd_send_all(tconn, sock, h, size, msg_flags);
730 if (err && !signal_pending(current))
731 conn_warn(tconn, "short send %s size=%d\n",
732 cmdname(cmd), (int)size);
736 /* don't pass the socket. we may only look at it
737 * when we hold the appropriate socket mutex.
739 int conn_send_cmd(struct drbd_tconn *tconn, int vnr, struct drbd_socket *sock,
740 enum drbd_packet cmd, struct p_header *h, size_t size)
744 mutex_lock(&sock->mutex);
746 err = _conn_send_cmd(tconn, vnr, sock->socket, cmd, h, size, 0);
747 mutex_unlock(&sock->mutex);
751 int conn_send_cmd2(struct drbd_tconn *tconn, enum drbd_packet cmd, char *data,
757 prepare_header80(&h, cmd, size);
758 err = drbd_get_data_sock(tconn);
760 err = drbd_send_all(tconn, tconn->data.socket, &h, sizeof(h), 0);
762 err = drbd_send_all(tconn, tconn->data.socket, data, size, 0);
763 drbd_put_data_sock(tconn);
768 int drbd_send_ping(struct drbd_tconn *tconn)
771 return !conn_send_cmd(tconn, 0, &tconn->meta, P_PING, &h, sizeof(h));
774 int drbd_send_ping_ack(struct drbd_tconn *tconn)
777 return !conn_send_cmd(tconn, 0, &tconn->meta, P_PING_ACK, &h, sizeof(h));
780 int drbd_send_sync_param(struct drbd_conf *mdev)
782 struct p_rs_param_95 *p;
785 const int apv = mdev->tconn->agreed_pro_version;
787 size = apv <= 87 ? sizeof(struct p_rs_param)
788 : apv == 88 ? sizeof(struct p_rs_param)
789 + strlen(mdev->tconn->net_conf->verify_alg) + 1
790 : apv <= 94 ? sizeof(struct p_rs_param_89)
791 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
793 /* used from admin command context and receiver/worker context.
794 * to avoid kmalloc, grab the socket right here,
795 * then use the pre-allocated sbuf there */
796 mutex_lock(&mdev->tconn->data.mutex);
797 sock = mdev->tconn->data.socket;
799 if (likely(sock != NULL)) {
800 enum drbd_packet cmd =
801 apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
803 p = &mdev->tconn->data.sbuf.rs_param_95;
805 /* initialize verify_alg and csums_alg */
806 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
808 if (get_ldev(mdev)) {
809 p->rate = cpu_to_be32(mdev->ldev->dc.resync_rate);
810 p->c_plan_ahead = cpu_to_be32(mdev->ldev->dc.c_plan_ahead);
811 p->c_delay_target = cpu_to_be32(mdev->ldev->dc.c_delay_target);
812 p->c_fill_target = cpu_to_be32(mdev->ldev->dc.c_fill_target);
813 p->c_max_rate = cpu_to_be32(mdev->ldev->dc.c_max_rate);
816 p->rate = cpu_to_be32(DRBD_RATE_DEF);
817 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
818 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
819 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
820 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
824 strcpy(p->verify_alg, mdev->tconn->net_conf->verify_alg);
826 strcpy(p->csums_alg, mdev->tconn->net_conf->csums_alg);
828 err = _drbd_send_cmd(mdev, sock, cmd, &p->head, size, 0);
832 mutex_unlock(&mdev->tconn->data.mutex);
837 int drbd_send_protocol(struct drbd_tconn *tconn)
839 struct p_protocol *p;
842 size = sizeof(struct p_protocol);
844 if (tconn->agreed_pro_version >= 87)
845 size += strlen(tconn->net_conf->integrity_alg) + 1;
847 /* we must not recurse into our own queue,
848 * as that is blocked during handshake */
849 p = kmalloc(size, GFP_NOIO);
853 p->protocol = cpu_to_be32(tconn->net_conf->wire_protocol);
854 p->after_sb_0p = cpu_to_be32(tconn->net_conf->after_sb_0p);
855 p->after_sb_1p = cpu_to_be32(tconn->net_conf->after_sb_1p);
856 p->after_sb_2p = cpu_to_be32(tconn->net_conf->after_sb_2p);
857 p->two_primaries = cpu_to_be32(tconn->net_conf->two_primaries);
860 if (tconn->net_conf->want_lose)
862 if (tconn->net_conf->dry_run) {
863 if (tconn->agreed_pro_version >= 92)
866 conn_err(tconn, "--dry-run is not supported by peer");
871 p->conn_flags = cpu_to_be32(cf);
873 if (tconn->agreed_pro_version >= 87)
874 strcpy(p->integrity_alg, tconn->net_conf->integrity_alg);
876 err = conn_send_cmd2(tconn, P_PROTOCOL, p->head.payload, size - sizeof(struct p_header));
881 int _drbd_send_uuids(struct drbd_conf *mdev, u64 uuid_flags)
886 if (!get_ldev_if_state(mdev, D_NEGOTIATING))
889 for (i = UI_CURRENT; i < UI_SIZE; i++)
890 p.uuid[i] = mdev->ldev ? cpu_to_be64(mdev->ldev->md.uuid[i]) : 0;
892 mdev->comm_bm_set = drbd_bm_total_weight(mdev);
893 p.uuid[UI_SIZE] = cpu_to_be64(mdev->comm_bm_set);
894 uuid_flags |= mdev->tconn->net_conf->want_lose ? 1 : 0;
895 uuid_flags |= test_bit(CRASHED_PRIMARY, &mdev->flags) ? 2 : 0;
896 uuid_flags |= mdev->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
897 p.uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
901 return drbd_send_cmd(mdev, &mdev->tconn->data, P_UUIDS, &p.head, sizeof(p));
904 int drbd_send_uuids(struct drbd_conf *mdev)
906 return _drbd_send_uuids(mdev, 0);
909 int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev)
911 return _drbd_send_uuids(mdev, 8);
914 void drbd_print_uuids(struct drbd_conf *mdev, const char *text)
916 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
917 u64 *uuid = mdev->ldev->md.uuid;
918 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX\n",
920 (unsigned long long)uuid[UI_CURRENT],
921 (unsigned long long)uuid[UI_BITMAP],
922 (unsigned long long)uuid[UI_HISTORY_START],
923 (unsigned long long)uuid[UI_HISTORY_END]);
926 dev_info(DEV, "%s effective data uuid: %016llX\n",
928 (unsigned long long)mdev->ed_uuid);
932 void drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev)
937 D_ASSERT(mdev->state.disk == D_UP_TO_DATE);
939 uuid = mdev->ldev->md.uuid[UI_BITMAP] + UUID_NEW_BM_OFFSET;
940 drbd_uuid_set(mdev, UI_BITMAP, uuid);
941 drbd_print_uuids(mdev, "updated sync UUID");
943 p.uuid = cpu_to_be64(uuid);
945 drbd_send_cmd(mdev, &mdev->tconn->data, P_SYNC_UUID, &p.head, sizeof(p));
948 int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags)
951 sector_t d_size, u_size;
952 int q_order_type, max_bio_size;
954 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
955 D_ASSERT(mdev->ldev->backing_bdev);
956 d_size = drbd_get_max_capacity(mdev->ldev);
957 u_size = mdev->ldev->dc.disk_size;
958 q_order_type = drbd_queue_order_type(mdev);
959 max_bio_size = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
960 max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE);
965 q_order_type = QUEUE_ORDERED_NONE;
966 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
969 p.d_size = cpu_to_be64(d_size);
970 p.u_size = cpu_to_be64(u_size);
971 p.c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(mdev->this_bdev));
972 p.max_bio_size = cpu_to_be32(max_bio_size);
973 p.queue_order_type = cpu_to_be16(q_order_type);
974 p.dds_flags = cpu_to_be16(flags);
976 return drbd_send_cmd(mdev, &mdev->tconn->data, P_SIZES, &p.head, sizeof(p));
980 * drbd_send_state() - Sends the drbd state to the peer
981 * @mdev: DRBD device.
983 int drbd_send_state(struct drbd_conf *mdev)
989 mutex_lock(&mdev->tconn->data.mutex);
991 p.state = cpu_to_be32(mdev->state.i); /* Within the send mutex */
992 sock = mdev->tconn->data.socket;
994 if (likely(sock != NULL))
995 err = _drbd_send_cmd(mdev, sock, P_STATE, &p.head, sizeof(p), 0);
997 mutex_unlock(&mdev->tconn->data.mutex);
1002 int _conn_send_state_req(struct drbd_tconn *tconn, int vnr, enum drbd_packet cmd,
1003 union drbd_state mask, union drbd_state val)
1005 struct p_req_state p;
1007 p.mask = cpu_to_be32(mask.i);
1008 p.val = cpu_to_be32(val.i);
1010 return conn_send_cmd(tconn, vnr, &tconn->data, cmd, &p.head, sizeof(p));
1013 void drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode)
1015 struct p_req_state_reply p;
1017 p.retcode = cpu_to_be32(retcode);
1019 drbd_send_cmd(mdev, &mdev->tconn->meta, P_STATE_CHG_REPLY, &p.head, sizeof(p));
1022 int conn_send_sr_reply(struct drbd_tconn *tconn, enum drbd_state_rv retcode)
1024 struct p_req_state_reply p;
1025 enum drbd_packet cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1027 p.retcode = cpu_to_be32(retcode);
1029 return !conn_send_cmd(tconn, 0, &tconn->meta, cmd, &p.head, sizeof(p));
1032 int fill_bitmap_rle_bits(struct drbd_conf *mdev,
1033 struct p_compressed_bm *p,
1034 struct bm_xfer_ctx *c)
1036 struct bitstream bs;
1037 unsigned long plain_bits;
1044 /* may we use this feature? */
1045 if ((mdev->tconn->net_conf->use_rle == 0) ||
1046 (mdev->tconn->agreed_pro_version < 90))
1049 if (c->bit_offset >= c->bm_bits)
1050 return 0; /* nothing to do. */
1052 /* use at most thus many bytes */
1053 bitstream_init(&bs, p->code, BM_PACKET_VLI_BYTES_MAX, 0);
1054 memset(p->code, 0, BM_PACKET_VLI_BYTES_MAX);
1055 /* plain bits covered in this code string */
1058 /* p->encoding & 0x80 stores whether the first run length is set.
1059 * bit offset is implicit.
1060 * start with toggle == 2 to be able to tell the first iteration */
1063 /* see how much plain bits we can stuff into one packet
1064 * using RLE and VLI. */
1066 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(mdev, c->bit_offset)
1067 : _drbd_bm_find_next(mdev, c->bit_offset);
1070 rl = tmp - c->bit_offset;
1072 if (toggle == 2) { /* first iteration */
1074 /* the first checked bit was set,
1075 * store start value, */
1076 DCBP_set_start(p, 1);
1077 /* but skip encoding of zero run length */
1081 DCBP_set_start(p, 0);
1084 /* paranoia: catch zero runlength.
1085 * can only happen if bitmap is modified while we scan it. */
1087 dev_err(DEV, "unexpected zero runlength while encoding bitmap "
1088 "t:%u bo:%lu\n", toggle, c->bit_offset);
1092 bits = vli_encode_bits(&bs, rl);
1093 if (bits == -ENOBUFS) /* buffer full */
1096 dev_err(DEV, "error while encoding bitmap: %d\n", bits);
1102 c->bit_offset = tmp;
1103 } while (c->bit_offset < c->bm_bits);
1105 len = bs.cur.b - p->code + !!bs.cur.bit;
1107 if (plain_bits < (len << 3)) {
1108 /* incompressible with this method.
1109 * we need to rewind both word and bit position. */
1110 c->bit_offset -= plain_bits;
1111 bm_xfer_ctx_bit_to_word_offset(c);
1112 c->bit_offset = c->word_offset * BITS_PER_LONG;
1116 /* RLE + VLI was able to compress it just fine.
1117 * update c->word_offset. */
1118 bm_xfer_ctx_bit_to_word_offset(c);
1120 /* store pad_bits */
1121 DCBP_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1127 * send_bitmap_rle_or_plain
1129 * Return 0 when done, 1 when another iteration is needed, and a negative error
1130 * code upon failure.
1133 send_bitmap_rle_or_plain(struct drbd_conf *mdev,
1134 struct p_header *h, struct bm_xfer_ctx *c)
1136 struct p_compressed_bm *p = (void*)h;
1137 unsigned long num_words;
1140 len = fill_bitmap_rle_bits(mdev, p, c);
1146 DCBP_set_code(p, RLE_VLI_Bits);
1147 err = _drbd_send_cmd(mdev, mdev->tconn->data.socket,
1148 P_COMPRESSED_BITMAP, h,
1149 sizeof(*p) + len, 0);
1152 c->bytes[0] += sizeof(*p) + len;
1154 if (c->bit_offset >= c->bm_bits)
1157 /* was not compressible.
1158 * send a buffer full of plain text bits instead. */
1159 num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset);
1160 len = num_words * sizeof(long);
1162 drbd_bm_get_lel(mdev, c->word_offset, num_words, (unsigned long*)h->payload);
1163 err = _drbd_send_cmd(mdev, mdev->tconn->data.socket, P_BITMAP,
1164 h, sizeof(struct p_header80) + len, 0);
1165 c->word_offset += num_words;
1166 c->bit_offset = c->word_offset * BITS_PER_LONG;
1169 c->bytes[1] += sizeof(struct p_header80) + len;
1171 if (c->bit_offset > c->bm_bits)
1172 c->bit_offset = c->bm_bits;
1176 INFO_bm_xfer_stats(mdev, "send", c);
1184 /* See the comment at receive_bitmap() */
1185 static int _drbd_send_bitmap(struct drbd_conf *mdev)
1187 struct bm_xfer_ctx c;
1191 if (!expect(mdev->bitmap))
1194 /* maybe we should use some per thread scratch page,
1195 * and allocate that during initial device creation? */
1196 p = (struct p_header *) __get_free_page(GFP_NOIO);
1198 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__);
1202 if (get_ldev(mdev)) {
1203 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1204 dev_info(DEV, "Writing the whole bitmap, MDF_FullSync was set.\n");
1205 drbd_bm_set_all(mdev);
1206 if (drbd_bm_write(mdev)) {
1207 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1208 * but otherwise process as per normal - need to tell other
1209 * side that a full resync is required! */
1210 dev_err(DEV, "Failed to write bitmap to disk!\n");
1212 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
1219 c = (struct bm_xfer_ctx) {
1220 .bm_bits = drbd_bm_bits(mdev),
1221 .bm_words = drbd_bm_words(mdev),
1225 err = send_bitmap_rle_or_plain(mdev, p, &c);
1228 free_page((unsigned long) p);
1232 int drbd_send_bitmap(struct drbd_conf *mdev)
1236 if (drbd_get_data_sock(mdev->tconn))
1238 err = !_drbd_send_bitmap(mdev);
1239 drbd_put_data_sock(mdev->tconn);
1242 void drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, u32 set_size)
1244 struct p_barrier_ack p;
1246 p.barrier = barrier_nr;
1247 p.set_size = cpu_to_be32(set_size);
1249 if (mdev->state.conn >= C_CONNECTED)
1250 drbd_send_cmd(mdev, &mdev->tconn->meta, P_BARRIER_ACK, &p.head, sizeof(p));
1254 * _drbd_send_ack() - Sends an ack packet
1255 * @mdev: DRBD device.
1256 * @cmd: Packet command code.
1257 * @sector: sector, needs to be in big endian byte order
1258 * @blksize: size in byte, needs to be in big endian byte order
1259 * @block_id: Id, big endian byte order
1261 static int _drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1262 u64 sector, u32 blksize, u64 block_id)
1264 struct p_block_ack p;
1267 p.block_id = block_id;
1268 p.blksize = blksize;
1269 p.seq_num = cpu_to_be32(atomic_inc_return(&mdev->packet_seq));
1271 if (!mdev->tconn->meta.socket || mdev->state.conn < C_CONNECTED)
1273 return drbd_send_cmd(mdev, &mdev->tconn->meta, cmd, &p.head, sizeof(p));
1276 /* dp->sector and dp->block_id already/still in network byte order,
1277 * data_size is payload size according to dp->head,
1278 * and may need to be corrected for digest size. */
1279 void drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd,
1280 struct p_data *dp, int data_size)
1282 data_size -= (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ?
1283 crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0;
1284 _drbd_send_ack(mdev, cmd, dp->sector, cpu_to_be32(data_size),
1288 void drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd,
1289 struct p_block_req *rp)
1291 _drbd_send_ack(mdev, cmd, rp->sector, rp->blksize, rp->block_id);
1295 * drbd_send_ack() - Sends an ack packet
1296 * @mdev: DRBD device
1297 * @cmd: packet command code
1298 * @peer_req: peer request
1300 int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1301 struct drbd_peer_request *peer_req)
1303 return _drbd_send_ack(mdev, cmd,
1304 cpu_to_be64(peer_req->i.sector),
1305 cpu_to_be32(peer_req->i.size),
1306 peer_req->block_id);
1309 /* This function misuses the block_id field to signal if the blocks
1310 * are is sync or not. */
1311 int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd,
1312 sector_t sector, int blksize, u64 block_id)
1314 return _drbd_send_ack(mdev, cmd,
1315 cpu_to_be64(sector),
1316 cpu_to_be32(blksize),
1317 cpu_to_be64(block_id));
1320 int drbd_send_drequest(struct drbd_conf *mdev, int cmd,
1321 sector_t sector, int size, u64 block_id)
1323 struct p_block_req p;
1325 p.sector = cpu_to_be64(sector);
1326 p.block_id = block_id;
1327 p.blksize = cpu_to_be32(size);
1329 return drbd_send_cmd(mdev, &mdev->tconn->data, cmd, &p.head, sizeof(p));
1332 int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, int size,
1333 void *digest, int digest_size, enum drbd_packet cmd)
1336 struct p_block_req p;
1338 prepare_header(mdev, &p.head, cmd, sizeof(p) - sizeof(struct p_header) + digest_size);
1339 p.sector = cpu_to_be64(sector);
1340 p.block_id = ID_SYNCER /* unused */;
1341 p.blksize = cpu_to_be32(size);
1343 mutex_lock(&mdev->tconn->data.mutex);
1344 err = drbd_send_all(mdev->tconn, mdev->tconn->data.socket, &p, sizeof(p), 0);
1346 err = drbd_send_all(mdev->tconn, mdev->tconn->data.socket, digest, digest_size, 0);
1347 mutex_unlock(&mdev->tconn->data.mutex);
1351 int drbd_send_ov_request(struct drbd_conf *mdev, sector_t sector, int size)
1353 struct p_block_req p;
1355 p.sector = cpu_to_be64(sector);
1356 p.block_id = ID_SYNCER /* unused */;
1357 p.blksize = cpu_to_be32(size);
1359 return drbd_send_cmd(mdev, &mdev->tconn->data, P_OV_REQUEST, &p.head, sizeof(p));
1362 /* called on sndtimeo
1363 * returns false if we should retry,
1364 * true if we think connection is dead
1366 static int we_should_drop_the_connection(struct drbd_tconn *tconn, struct socket *sock)
1369 /* long elapsed = (long)(jiffies - mdev->last_received); */
1371 drop_it = tconn->meta.socket == sock
1372 || !tconn->asender.task
1373 || get_t_state(&tconn->asender) != RUNNING
1374 || tconn->cstate < C_WF_REPORT_PARAMS;
1379 drop_it = !--tconn->ko_count;
1381 conn_err(tconn, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1382 current->comm, current->pid, tconn->ko_count);
1383 request_ping(tconn);
1386 return drop_it; /* && (mdev->state == R_PRIMARY) */;
1389 static void drbd_update_congested(struct drbd_tconn *tconn)
1391 struct sock *sk = tconn->data.socket->sk;
1392 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1393 set_bit(NET_CONGESTED, &tconn->flags);
1396 /* The idea of sendpage seems to be to put some kind of reference
1397 * to the page into the skb, and to hand it over to the NIC. In
1398 * this process get_page() gets called.
1400 * As soon as the page was really sent over the network put_page()
1401 * gets called by some part of the network layer. [ NIC driver? ]
1403 * [ get_page() / put_page() increment/decrement the count. If count
1404 * reaches 0 the page will be freed. ]
1406 * This works nicely with pages from FSs.
1407 * But this means that in protocol A we might signal IO completion too early!
1409 * In order not to corrupt data during a resync we must make sure
1410 * that we do not reuse our own buffer pages (EEs) to early, therefore
1411 * we have the net_ee list.
1413 * XFS seems to have problems, still, it submits pages with page_count == 0!
1414 * As a workaround, we disable sendpage on pages
1415 * with page_count == 0 or PageSlab.
1417 static int _drbd_no_send_page(struct drbd_conf *mdev, struct page *page,
1418 int offset, size_t size, unsigned msg_flags)
1420 struct socket *socket;
1424 socket = mdev->tconn->data.socket;
1425 addr = kmap(page) + offset;
1426 err = drbd_send_all(mdev->tconn, socket, addr, size, msg_flags);
1429 mdev->send_cnt += size >> 9;
1433 static int _drbd_send_page(struct drbd_conf *mdev, struct page *page,
1434 int offset, size_t size, unsigned msg_flags)
1436 struct socket *socket = mdev->tconn->data.socket;
1437 mm_segment_t oldfs = get_fs();
1441 /* e.g. XFS meta- & log-data is in slab pages, which have a
1442 * page_count of 0 and/or have PageSlab() set.
1443 * we cannot use send_page for those, as that does get_page();
1444 * put_page(); and would cause either a VM_BUG directly, or
1445 * __page_cache_release a page that would actually still be referenced
1446 * by someone, leading to some obscure delayed Oops somewhere else. */
1447 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1448 return _drbd_no_send_page(mdev, page, offset, size, msg_flags);
1450 msg_flags |= MSG_NOSIGNAL;
1451 drbd_update_congested(mdev->tconn);
1456 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1458 if (sent == -EAGAIN) {
1459 if (we_should_drop_the_connection(mdev->tconn, socket))
1463 dev_warn(DEV, "%s: size=%d len=%d sent=%d\n",
1464 __func__, (int)size, len, sent);
1471 } while (len > 0 /* THINK && mdev->cstate >= C_CONNECTED*/);
1473 clear_bit(NET_CONGESTED, &mdev->tconn->flags);
1477 mdev->send_cnt += size >> 9;
1482 static int _drbd_send_bio(struct drbd_conf *mdev, struct bio *bio)
1484 struct bio_vec *bvec;
1486 /* hint all but last page with MSG_MORE */
1487 __bio_for_each_segment(bvec, bio, i, 0) {
1490 err = _drbd_no_send_page(mdev, bvec->bv_page,
1491 bvec->bv_offset, bvec->bv_len,
1492 i == bio->bi_vcnt - 1 ? 0 : MSG_MORE);
1499 static int _drbd_send_zc_bio(struct drbd_conf *mdev, struct bio *bio)
1501 struct bio_vec *bvec;
1503 /* hint all but last page with MSG_MORE */
1504 __bio_for_each_segment(bvec, bio, i, 0) {
1507 err = _drbd_send_page(mdev, bvec->bv_page,
1508 bvec->bv_offset, bvec->bv_len,
1509 i == bio->bi_vcnt - 1 ? 0 : MSG_MORE);
1516 static int _drbd_send_zc_ee(struct drbd_conf *mdev,
1517 struct drbd_peer_request *peer_req)
1519 struct page *page = peer_req->pages;
1520 unsigned len = peer_req->i.size;
1523 /* hint all but last page with MSG_MORE */
1524 page_chain_for_each(page) {
1525 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1527 err = _drbd_send_page(mdev, page, 0, l,
1528 page_chain_next(page) ? MSG_MORE : 0);
1536 static u32 bio_flags_to_wire(struct drbd_conf *mdev, unsigned long bi_rw)
1538 if (mdev->tconn->agreed_pro_version >= 95)
1539 return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) |
1540 (bi_rw & REQ_FUA ? DP_FUA : 0) |
1541 (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) |
1542 (bi_rw & REQ_DISCARD ? DP_DISCARD : 0);
1544 return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0;
1547 /* Used to send write requests
1548 * R_PRIMARY -> Peer (P_DATA)
1550 int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req)
1554 unsigned int dp_flags = 0;
1558 err = drbd_get_data_sock(mdev->tconn);
1562 dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_w_tfm) ?
1563 crypto_hash_digestsize(mdev->tconn->integrity_w_tfm) : 0;
1565 prepare_header(mdev, &p.head, P_DATA, sizeof(p) - sizeof(struct p_header) + dgs + req->i.size);
1566 p.sector = cpu_to_be64(req->i.sector);
1567 p.block_id = (unsigned long)req;
1568 p.seq_num = cpu_to_be32(req->seq_num = atomic_inc_return(&mdev->packet_seq));
1570 dp_flags = bio_flags_to_wire(mdev, req->master_bio->bi_rw);
1572 if (mdev->state.conn >= C_SYNC_SOURCE &&
1573 mdev->state.conn <= C_PAUSED_SYNC_T)
1574 dp_flags |= DP_MAY_SET_IN_SYNC;
1576 p.dp_flags = cpu_to_be32(dp_flags);
1577 set_bit(UNPLUG_REMOTE, &mdev->flags);
1578 err = drbd_send_all(mdev->tconn, mdev->tconn->data.socket, &p,
1579 sizeof(p), dgs ? MSG_MORE : 0);
1581 dgb = mdev->tconn->int_dig_out;
1582 drbd_csum_bio(mdev, mdev->tconn->integrity_w_tfm, req->master_bio, dgb);
1583 err = drbd_send_all(mdev->tconn, mdev->tconn->data.socket, dgb, dgs, 0);
1586 /* For protocol A, we have to memcpy the payload into
1587 * socket buffers, as we may complete right away
1588 * as soon as we handed it over to tcp, at which point the data
1589 * pages may become invalid.
1591 * For data-integrity enabled, we copy it as well, so we can be
1592 * sure that even if the bio pages may still be modified, it
1593 * won't change the data on the wire, thus if the digest checks
1594 * out ok after sending on this side, but does not fit on the
1595 * receiving side, we sure have detected corruption elsewhere.
1597 if (mdev->tconn->net_conf->wire_protocol == DRBD_PROT_A || dgs)
1598 err = _drbd_send_bio(mdev, req->master_bio);
1600 err = _drbd_send_zc_bio(mdev, req->master_bio);
1602 /* double check digest, sometimes buffers have been modified in flight. */
1603 if (dgs > 0 && dgs <= 64) {
1604 /* 64 byte, 512 bit, is the largest digest size
1605 * currently supported in kernel crypto. */
1606 unsigned char digest[64];
1607 drbd_csum_bio(mdev, mdev->tconn->integrity_w_tfm, req->master_bio, digest);
1608 if (memcmp(mdev->tconn->int_dig_out, digest, dgs)) {
1610 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1611 (unsigned long long)req->i.sector, req->i.size);
1613 } /* else if (dgs > 64) {
1614 ... Be noisy about digest too large ...
1618 drbd_put_data_sock(mdev->tconn);
1623 /* answer packet, used to send data back for read requests:
1624 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1625 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1627 int drbd_send_block(struct drbd_conf *mdev, enum drbd_packet cmd,
1628 struct drbd_peer_request *peer_req)
1635 dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_w_tfm) ?
1636 crypto_hash_digestsize(mdev->tconn->integrity_w_tfm) : 0;
1638 prepare_header(mdev, &p.head, cmd, sizeof(p) -
1639 sizeof(struct p_header80) +
1640 dgs + peer_req->i.size);
1641 p.sector = cpu_to_be64(peer_req->i.sector);
1642 p.block_id = peer_req->block_id;
1643 p.seq_num = 0; /* unused */
1645 /* Only called by our kernel thread.
1646 * This one may be interrupted by DRBD_SIG and/or DRBD_SIGKILL
1647 * in response to admin command or module unload.
1649 err = drbd_get_data_sock(mdev->tconn);
1652 err = drbd_send_all(mdev->tconn, mdev->tconn->data.socket, &p,
1653 sizeof(p), dgs ? MSG_MORE : 0);
1655 dgb = mdev->tconn->int_dig_out;
1656 drbd_csum_ee(mdev, mdev->tconn->integrity_w_tfm, peer_req, dgb);
1657 err = drbd_send_all(mdev->tconn, mdev->tconn->data.socket, dgb,
1661 err = _drbd_send_zc_ee(mdev, peer_req);
1662 drbd_put_data_sock(mdev->tconn);
1667 int drbd_send_out_of_sync(struct drbd_conf *mdev, struct drbd_request *req)
1669 struct p_block_desc p;
1671 p.sector = cpu_to_be64(req->i.sector);
1672 p.blksize = cpu_to_be32(req->i.size);
1674 return drbd_send_cmd(mdev, &mdev->tconn->data, P_OUT_OF_SYNC, &p.head, sizeof(p));
1678 drbd_send distinguishes two cases:
1680 Packets sent via the data socket "sock"
1681 and packets sent via the meta data socket "msock"
1684 -----------------+-------------------------+------------------------------
1685 timeout conf.timeout / 2 conf.timeout / 2
1686 timeout action send a ping via msock Abort communication
1687 and close all sockets
1691 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1693 int drbd_send(struct drbd_tconn *tconn, struct socket *sock,
1694 void *buf, size_t size, unsigned msg_flags)
1703 /* THINK if (signal_pending) return ... ? */
1708 msg.msg_name = NULL;
1709 msg.msg_namelen = 0;
1710 msg.msg_control = NULL;
1711 msg.msg_controllen = 0;
1712 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1714 if (sock == tconn->data.socket) {
1715 tconn->ko_count = tconn->net_conf->ko_count;
1716 drbd_update_congested(tconn);
1720 * tcp_sendmsg does _not_ use its size parameter at all ?
1722 * -EAGAIN on timeout, -EINTR on signal.
1725 * do we need to block DRBD_SIG if sock == &meta.socket ??
1726 * otherwise wake_asender() might interrupt some send_*Ack !
1728 rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
1729 if (rv == -EAGAIN) {
1730 if (we_should_drop_the_connection(tconn, sock))
1736 flush_signals(current);
1744 } while (sent < size);
1746 if (sock == tconn->data.socket)
1747 clear_bit(NET_CONGESTED, &tconn->flags);
1750 if (rv != -EAGAIN) {
1751 conn_err(tconn, "%s_sendmsg returned %d\n",
1752 sock == tconn->meta.socket ? "msock" : "sock",
1754 conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD);
1756 conn_request_state(tconn, NS(conn, C_TIMEOUT), CS_HARD);
1763 * drbd_send_all - Send an entire buffer
1765 * Returns 0 upon success and a negative error value otherwise.
1767 int drbd_send_all(struct drbd_tconn *tconn, struct socket *sock, void *buffer,
1768 size_t size, unsigned msg_flags)
1772 err = drbd_send(tconn, sock, buffer, size, msg_flags);
1780 static int drbd_open(struct block_device *bdev, fmode_t mode)
1782 struct drbd_conf *mdev = bdev->bd_disk->private_data;
1783 unsigned long flags;
1786 mutex_lock(&drbd_main_mutex);
1787 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
1788 /* to have a stable mdev->state.role
1789 * and no race with updating open_cnt */
1791 if (mdev->state.role != R_PRIMARY) {
1792 if (mode & FMODE_WRITE)
1794 else if (!allow_oos)
1800 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
1801 mutex_unlock(&drbd_main_mutex);
1806 static int drbd_release(struct gendisk *gd, fmode_t mode)
1808 struct drbd_conf *mdev = gd->private_data;
1809 mutex_lock(&drbd_main_mutex);
1811 mutex_unlock(&drbd_main_mutex);
1815 static void drbd_set_defaults(struct drbd_conf *mdev)
1817 /* Beware! The actual layout differs
1818 * between big endian and little endian */
1819 mdev->state = (union drbd_state) {
1820 { .role = R_SECONDARY,
1822 .conn = C_STANDALONE,
1831 void drbd_init_set_defaults(struct drbd_conf *mdev)
1833 /* the memset(,0,) did most of this.
1834 * note: only assignments, no allocation in here */
1836 drbd_set_defaults(mdev);
1838 atomic_set(&mdev->ap_bio_cnt, 0);
1839 atomic_set(&mdev->ap_pending_cnt, 0);
1840 atomic_set(&mdev->rs_pending_cnt, 0);
1841 atomic_set(&mdev->unacked_cnt, 0);
1842 atomic_set(&mdev->local_cnt, 0);
1843 atomic_set(&mdev->pp_in_use_by_net, 0);
1844 atomic_set(&mdev->rs_sect_in, 0);
1845 atomic_set(&mdev->rs_sect_ev, 0);
1846 atomic_set(&mdev->ap_in_flight, 0);
1848 mutex_init(&mdev->md_io_mutex);
1849 mutex_init(&mdev->own_state_mutex);
1850 mdev->state_mutex = &mdev->own_state_mutex;
1852 spin_lock_init(&mdev->al_lock);
1853 spin_lock_init(&mdev->peer_seq_lock);
1854 spin_lock_init(&mdev->epoch_lock);
1856 INIT_LIST_HEAD(&mdev->active_ee);
1857 INIT_LIST_HEAD(&mdev->sync_ee);
1858 INIT_LIST_HEAD(&mdev->done_ee);
1859 INIT_LIST_HEAD(&mdev->read_ee);
1860 INIT_LIST_HEAD(&mdev->net_ee);
1861 INIT_LIST_HEAD(&mdev->resync_reads);
1862 INIT_LIST_HEAD(&mdev->resync_work.list);
1863 INIT_LIST_HEAD(&mdev->unplug_work.list);
1864 INIT_LIST_HEAD(&mdev->go_diskless.list);
1865 INIT_LIST_HEAD(&mdev->md_sync_work.list);
1866 INIT_LIST_HEAD(&mdev->start_resync_work.list);
1867 INIT_LIST_HEAD(&mdev->bm_io_work.w.list);
1869 mdev->resync_work.cb = w_resync_timer;
1870 mdev->unplug_work.cb = w_send_write_hint;
1871 mdev->go_diskless.cb = w_go_diskless;
1872 mdev->md_sync_work.cb = w_md_sync;
1873 mdev->bm_io_work.w.cb = w_bitmap_io;
1874 mdev->start_resync_work.cb = w_start_resync;
1876 mdev->resync_work.mdev = mdev;
1877 mdev->unplug_work.mdev = mdev;
1878 mdev->go_diskless.mdev = mdev;
1879 mdev->md_sync_work.mdev = mdev;
1880 mdev->bm_io_work.w.mdev = mdev;
1881 mdev->start_resync_work.mdev = mdev;
1883 init_timer(&mdev->resync_timer);
1884 init_timer(&mdev->md_sync_timer);
1885 init_timer(&mdev->start_resync_timer);
1886 init_timer(&mdev->request_timer);
1887 mdev->resync_timer.function = resync_timer_fn;
1888 mdev->resync_timer.data = (unsigned long) mdev;
1889 mdev->md_sync_timer.function = md_sync_timer_fn;
1890 mdev->md_sync_timer.data = (unsigned long) mdev;
1891 mdev->start_resync_timer.function = start_resync_timer_fn;
1892 mdev->start_resync_timer.data = (unsigned long) mdev;
1893 mdev->request_timer.function = request_timer_fn;
1894 mdev->request_timer.data = (unsigned long) mdev;
1896 init_waitqueue_head(&mdev->misc_wait);
1897 init_waitqueue_head(&mdev->state_wait);
1898 init_waitqueue_head(&mdev->ee_wait);
1899 init_waitqueue_head(&mdev->al_wait);
1900 init_waitqueue_head(&mdev->seq_wait);
1902 /* mdev->tconn->agreed_pro_version gets initialized in drbd_connect() */
1903 mdev->write_ordering = WO_bdev_flush;
1904 mdev->resync_wenr = LC_FREE;
1905 mdev->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1906 mdev->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1909 void drbd_mdev_cleanup(struct drbd_conf *mdev)
1912 if (mdev->tconn->receiver.t_state != NONE)
1913 dev_err(DEV, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
1914 mdev->tconn->receiver.t_state);
1916 /* no need to lock it, I'm the only thread alive */
1917 if (atomic_read(&mdev->current_epoch->epoch_size) != 0)
1918 dev_err(DEV, "epoch_size:%d\n", atomic_read(&mdev->current_epoch->epoch_size));
1928 mdev->rs_failed = 0;
1929 mdev->rs_last_events = 0;
1930 mdev->rs_last_sect_ev = 0;
1931 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1932 mdev->rs_mark_left[i] = 0;
1933 mdev->rs_mark_time[i] = 0;
1935 D_ASSERT(mdev->tconn->net_conf == NULL);
1937 drbd_set_my_capacity(mdev, 0);
1939 /* maybe never allocated. */
1940 drbd_bm_resize(mdev, 0, 1);
1941 drbd_bm_cleanup(mdev);
1944 drbd_free_resources(mdev);
1945 clear_bit(AL_SUSPENDED, &mdev->flags);
1948 * currently we drbd_init_ee only on module load, so
1949 * we may do drbd_release_ee only on module unload!
1951 D_ASSERT(list_empty(&mdev->active_ee));
1952 D_ASSERT(list_empty(&mdev->sync_ee));
1953 D_ASSERT(list_empty(&mdev->done_ee));
1954 D_ASSERT(list_empty(&mdev->read_ee));
1955 D_ASSERT(list_empty(&mdev->net_ee));
1956 D_ASSERT(list_empty(&mdev->resync_reads));
1957 D_ASSERT(list_empty(&mdev->tconn->data.work.q));
1958 D_ASSERT(list_empty(&mdev->tconn->meta.work.q));
1959 D_ASSERT(list_empty(&mdev->resync_work.list));
1960 D_ASSERT(list_empty(&mdev->unplug_work.list));
1961 D_ASSERT(list_empty(&mdev->go_diskless.list));
1963 drbd_set_defaults(mdev);
1967 static void drbd_destroy_mempools(void)
1971 while (drbd_pp_pool) {
1972 page = drbd_pp_pool;
1973 drbd_pp_pool = (struct page *)page_private(page);
1978 /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */
1980 if (drbd_md_io_bio_set)
1981 bioset_free(drbd_md_io_bio_set);
1982 if (drbd_md_io_page_pool)
1983 mempool_destroy(drbd_md_io_page_pool);
1984 if (drbd_ee_mempool)
1985 mempool_destroy(drbd_ee_mempool);
1986 if (drbd_request_mempool)
1987 mempool_destroy(drbd_request_mempool);
1989 kmem_cache_destroy(drbd_ee_cache);
1990 if (drbd_request_cache)
1991 kmem_cache_destroy(drbd_request_cache);
1992 if (drbd_bm_ext_cache)
1993 kmem_cache_destroy(drbd_bm_ext_cache);
1994 if (drbd_al_ext_cache)
1995 kmem_cache_destroy(drbd_al_ext_cache);
1997 drbd_md_io_bio_set = NULL;
1998 drbd_md_io_page_pool = NULL;
1999 drbd_ee_mempool = NULL;
2000 drbd_request_mempool = NULL;
2001 drbd_ee_cache = NULL;
2002 drbd_request_cache = NULL;
2003 drbd_bm_ext_cache = NULL;
2004 drbd_al_ext_cache = NULL;
2009 static int drbd_create_mempools(void)
2012 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
2015 /* prepare our caches and mempools */
2016 drbd_request_mempool = NULL;
2017 drbd_ee_cache = NULL;
2018 drbd_request_cache = NULL;
2019 drbd_bm_ext_cache = NULL;
2020 drbd_al_ext_cache = NULL;
2021 drbd_pp_pool = NULL;
2022 drbd_md_io_page_pool = NULL;
2023 drbd_md_io_bio_set = NULL;
2026 drbd_request_cache = kmem_cache_create(
2027 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2028 if (drbd_request_cache == NULL)
2031 drbd_ee_cache = kmem_cache_create(
2032 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2033 if (drbd_ee_cache == NULL)
2036 drbd_bm_ext_cache = kmem_cache_create(
2037 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2038 if (drbd_bm_ext_cache == NULL)
2041 drbd_al_ext_cache = kmem_cache_create(
2042 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2043 if (drbd_al_ext_cache == NULL)
2047 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2048 if (drbd_md_io_bio_set == NULL)
2051 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2052 if (drbd_md_io_page_pool == NULL)
2055 drbd_request_mempool = mempool_create(number,
2056 mempool_alloc_slab, mempool_free_slab, drbd_request_cache);
2057 if (drbd_request_mempool == NULL)
2060 drbd_ee_mempool = mempool_create(number,
2061 mempool_alloc_slab, mempool_free_slab, drbd_ee_cache);
2062 if (drbd_ee_mempool == NULL)
2065 /* drbd's page pool */
2066 spin_lock_init(&drbd_pp_lock);
2068 for (i = 0; i < number; i++) {
2069 page = alloc_page(GFP_HIGHUSER);
2072 set_page_private(page, (unsigned long)drbd_pp_pool);
2073 drbd_pp_pool = page;
2075 drbd_pp_vacant = number;
2080 drbd_destroy_mempools(); /* in case we allocated some */
2084 static int drbd_notify_sys(struct notifier_block *this, unsigned long code,
2087 /* just so we have it. you never know what interesting things we
2088 * might want to do here some day...
2094 static struct notifier_block drbd_notifier = {
2095 .notifier_call = drbd_notify_sys,
2098 static void drbd_release_ee_lists(struct drbd_conf *mdev)
2102 rr = drbd_release_ee(mdev, &mdev->active_ee);
2104 dev_err(DEV, "%d EEs in active list found!\n", rr);
2106 rr = drbd_release_ee(mdev, &mdev->sync_ee);
2108 dev_err(DEV, "%d EEs in sync list found!\n", rr);
2110 rr = drbd_release_ee(mdev, &mdev->read_ee);
2112 dev_err(DEV, "%d EEs in read list found!\n", rr);
2114 rr = drbd_release_ee(mdev, &mdev->done_ee);
2116 dev_err(DEV, "%d EEs in done list found!\n", rr);
2118 rr = drbd_release_ee(mdev, &mdev->net_ee);
2120 dev_err(DEV, "%d EEs in net list found!\n", rr);
2123 /* caution. no locking. */
2124 void drbd_delete_device(unsigned int minor)
2126 struct drbd_conf *mdev = minor_to_mdev(minor);
2131 idr_remove(&mdev->tconn->volumes, mdev->vnr);
2132 idr_remove(&minors, minor);
2135 /* paranoia asserts */
2136 D_ASSERT(mdev->open_cnt == 0);
2137 D_ASSERT(list_empty(&mdev->tconn->data.work.q));
2138 /* end paranoia asserts */
2140 del_gendisk(mdev->vdisk);
2142 /* cleanup stuff that may have been allocated during
2143 * device (re-)configuration or state changes */
2145 if (mdev->this_bdev)
2146 bdput(mdev->this_bdev);
2148 drbd_free_resources(mdev);
2150 drbd_release_ee_lists(mdev);
2152 lc_destroy(mdev->act_log);
2153 lc_destroy(mdev->resync);
2155 kfree(mdev->p_uuid);
2156 /* mdev->p_uuid = NULL; */
2158 /* cleanup the rest that has been
2159 * allocated from drbd_new_device
2160 * and actually free the mdev itself */
2161 drbd_free_mdev(mdev);
2164 static void drbd_cleanup(void)
2167 struct drbd_conf *mdev;
2169 unregister_reboot_notifier(&drbd_notifier);
2171 /* first remove proc,
2172 * drbdsetup uses it's presence to detect
2173 * whether DRBD is loaded.
2174 * If we would get stuck in proc removal,
2175 * but have netlink already deregistered,
2176 * some drbdsetup commands may wait forever
2180 remove_proc_entry("drbd", NULL);
2182 drbd_genl_unregister();
2184 idr_for_each_entry(&minors, mdev, i)
2185 drbd_delete_device(i);
2186 drbd_destroy_mempools();
2187 unregister_blkdev(DRBD_MAJOR, "drbd");
2189 idr_destroy(&minors);
2191 printk(KERN_INFO "drbd: module cleanup done.\n");
2195 * drbd_congested() - Callback for pdflush
2196 * @congested_data: User data
2197 * @bdi_bits: Bits pdflush is currently interested in
2199 * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2201 static int drbd_congested(void *congested_data, int bdi_bits)
2203 struct drbd_conf *mdev = congested_data;
2204 struct request_queue *q;
2208 if (!may_inc_ap_bio(mdev)) {
2209 /* DRBD has frozen IO */
2215 if (get_ldev(mdev)) {
2216 q = bdev_get_queue(mdev->ldev->backing_bdev);
2217 r = bdi_congested(&q->backing_dev_info, bdi_bits);
2223 if (bdi_bits & (1 << BDI_async_congested) && test_bit(NET_CONGESTED, &mdev->tconn->flags)) {
2224 r |= (1 << BDI_async_congested);
2225 reason = reason == 'b' ? 'a' : 'n';
2229 mdev->congestion_reason = reason;
2233 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2235 sema_init(&wq->s, 0);
2236 spin_lock_init(&wq->q_lock);
2237 INIT_LIST_HEAD(&wq->q);
2240 struct drbd_tconn *conn_by_name(const char *name)
2242 struct drbd_tconn *tconn;
2244 if (!name || !name[0])
2247 mutex_lock(&drbd_cfg_mutex);
2248 list_for_each_entry(tconn, &drbd_tconns, all_tconn) {
2249 if (!strcmp(tconn->name, name))
2254 mutex_unlock(&drbd_cfg_mutex);
2258 struct drbd_tconn *drbd_new_tconn(const char *name)
2260 struct drbd_tconn *tconn;
2262 tconn = kzalloc(sizeof(struct drbd_tconn), GFP_KERNEL);
2266 tconn->name = kstrdup(name, GFP_KERNEL);
2270 if (!zalloc_cpumask_var(&tconn->cpu_mask, GFP_KERNEL))
2273 if (!tl_init(tconn))
2276 tconn->cstate = C_STANDALONE;
2277 mutex_init(&tconn->cstate_mutex);
2278 spin_lock_init(&tconn->req_lock);
2279 atomic_set(&tconn->net_cnt, 0);
2280 init_waitqueue_head(&tconn->net_cnt_wait);
2281 init_waitqueue_head(&tconn->ping_wait);
2282 idr_init(&tconn->volumes);
2284 drbd_init_workqueue(&tconn->data.work);
2285 mutex_init(&tconn->data.mutex);
2287 drbd_init_workqueue(&tconn->meta.work);
2288 mutex_init(&tconn->meta.mutex);
2290 drbd_thread_init(tconn, &tconn->receiver, drbdd_init, "receiver");
2291 drbd_thread_init(tconn, &tconn->worker, drbd_worker, "worker");
2292 drbd_thread_init(tconn, &tconn->asender, drbd_asender, "asender");
2294 tconn->res_opts = (struct res_opts) {
2295 {}, 0, /* cpu_mask */
2296 DRBD_ON_NO_DATA_DEF, /* on_no_data */
2299 mutex_lock(&drbd_cfg_mutex);
2300 list_add_tail(&tconn->all_tconn, &drbd_tconns);
2301 mutex_unlock(&drbd_cfg_mutex);
2307 free_cpumask_var(tconn->cpu_mask);
2314 void drbd_free_tconn(struct drbd_tconn *tconn)
2316 list_del(&tconn->all_tconn);
2317 idr_destroy(&tconn->volumes);
2319 free_cpumask_var(tconn->cpu_mask);
2321 kfree(tconn->int_dig_out);
2322 kfree(tconn->int_dig_in);
2323 kfree(tconn->int_dig_vv);
2327 enum drbd_ret_code conn_new_minor(struct drbd_tconn *tconn, unsigned int minor, int vnr)
2329 struct drbd_conf *mdev;
2330 struct gendisk *disk;
2331 struct request_queue *q;
2333 int minor_got = minor;
2334 enum drbd_ret_code err = ERR_NOMEM;
2336 mdev = minor_to_mdev(minor);
2338 return ERR_MINOR_EXISTS;
2340 /* GFP_KERNEL, we are outside of all write-out paths */
2341 mdev = kzalloc(sizeof(struct drbd_conf), GFP_KERNEL);
2345 mdev->tconn = tconn;
2346 mdev->minor = minor;
2349 drbd_init_set_defaults(mdev);
2351 q = blk_alloc_queue(GFP_KERNEL);
2355 q->queuedata = mdev;
2357 disk = alloc_disk(1);
2362 set_disk_ro(disk, true);
2365 disk->major = DRBD_MAJOR;
2366 disk->first_minor = minor;
2367 disk->fops = &drbd_ops;
2368 sprintf(disk->disk_name, "drbd%d", minor);
2369 disk->private_data = mdev;
2371 mdev->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2372 /* we have no partitions. we contain only ourselves. */
2373 mdev->this_bdev->bd_contains = mdev->this_bdev;
2375 q->backing_dev_info.congested_fn = drbd_congested;
2376 q->backing_dev_info.congested_data = mdev;
2378 blk_queue_make_request(q, drbd_make_request);
2379 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2380 This triggers a max_bio_size message upon first attach or connect */
2381 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2382 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2383 blk_queue_merge_bvec(q, drbd_merge_bvec);
2384 q->queue_lock = &mdev->tconn->req_lock; /* needed since we use */
2386 mdev->md_io_page = alloc_page(GFP_KERNEL);
2387 if (!mdev->md_io_page)
2388 goto out_no_io_page;
2390 if (drbd_bm_init(mdev))
2392 mdev->read_requests = RB_ROOT;
2393 mdev->write_requests = RB_ROOT;
2395 mdev->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2396 if (!mdev->current_epoch)
2399 INIT_LIST_HEAD(&mdev->current_epoch->list);
2402 if (!idr_pre_get(&minors, GFP_KERNEL))
2403 goto out_no_minor_idr;
2404 if (idr_get_new_above(&minors, mdev, minor, &minor_got))
2405 goto out_no_minor_idr;
2406 if (minor_got != minor) {
2407 err = ERR_MINOR_EXISTS;
2408 drbd_msg_put_info("requested minor exists already");
2409 goto out_idr_remove_minor;
2412 if (!idr_pre_get(&tconn->volumes, GFP_KERNEL))
2413 goto out_idr_remove_minor;
2414 if (idr_get_new_above(&tconn->volumes, mdev, vnr, &vnr_got))
2415 goto out_idr_remove_minor;
2416 if (vnr_got != vnr) {
2417 err = ERR_INVALID_REQUEST;
2418 drbd_msg_put_info("requested volume exists already");
2419 goto out_idr_remove_vol;
2423 /* inherit the connection state */
2424 mdev->state.conn = tconn->cstate;
2425 if (mdev->state.conn == C_WF_REPORT_PARAMS)
2426 drbd_connected(vnr, mdev, tconn);
2431 idr_remove(&tconn->volumes, vnr_got);
2432 out_idr_remove_minor:
2433 idr_remove(&minors, minor_got);
2436 kfree(mdev->current_epoch);
2438 drbd_bm_cleanup(mdev);
2440 __free_page(mdev->md_io_page);
2444 blk_cleanup_queue(q);
2450 /* counterpart of drbd_new_device.
2451 * last part of drbd_delete_device. */
2452 void drbd_free_mdev(struct drbd_conf *mdev)
2454 kfree(mdev->current_epoch);
2455 if (mdev->bitmap) /* should no longer be there. */
2456 drbd_bm_cleanup(mdev);
2457 __free_page(mdev->md_io_page);
2458 put_disk(mdev->vdisk);
2459 blk_cleanup_queue(mdev->rq_queue);
2464 int __init drbd_init(void)
2468 BUILD_BUG_ON(sizeof(struct p_header80) != sizeof(struct p_header95));
2469 BUILD_BUG_ON(sizeof(struct p_handshake) != 80);
2471 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2473 "drbd: invalid minor_count (%d)\n", minor_count);
2481 err = register_blkdev(DRBD_MAJOR, "drbd");
2484 "drbd: unable to register block device major %d\n",
2489 err = drbd_genl_register();
2491 printk(KERN_ERR "drbd: unable to register generic netlink family\n");
2496 register_reboot_notifier(&drbd_notifier);
2499 * allocate all necessary structs
2503 init_waitqueue_head(&drbd_pp_wait);
2505 drbd_proc = NULL; /* play safe for drbd_cleanup */
2508 err = drbd_create_mempools();
2512 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
2514 printk(KERN_ERR "drbd: unable to register proc file\n");
2518 rwlock_init(&global_state_lock);
2519 INIT_LIST_HEAD(&drbd_tconns);
2521 printk(KERN_INFO "drbd: initialized. "
2522 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2523 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2524 printk(KERN_INFO "drbd: %s\n", drbd_buildtag());
2525 printk(KERN_INFO "drbd: registered as block device major %d\n",
2528 return 0; /* Success! */
2533 /* currently always the case */
2534 printk(KERN_ERR "drbd: ran out of memory\n");
2536 printk(KERN_ERR "drbd: initialization failure\n");
2540 void drbd_free_bc(struct drbd_backing_dev *ldev)
2545 blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2546 blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2551 void drbd_free_sock(struct drbd_tconn *tconn)
2553 if (tconn->data.socket) {
2554 mutex_lock(&tconn->data.mutex);
2555 kernel_sock_shutdown(tconn->data.socket, SHUT_RDWR);
2556 sock_release(tconn->data.socket);
2557 tconn->data.socket = NULL;
2558 mutex_unlock(&tconn->data.mutex);
2560 if (tconn->meta.socket) {
2561 mutex_lock(&tconn->meta.mutex);
2562 kernel_sock_shutdown(tconn->meta.socket, SHUT_RDWR);
2563 sock_release(tconn->meta.socket);
2564 tconn->meta.socket = NULL;
2565 mutex_unlock(&tconn->meta.mutex);
2570 void drbd_free_resources(struct drbd_conf *mdev)
2572 crypto_free_hash(mdev->tconn->csums_tfm);
2573 mdev->tconn->csums_tfm = NULL;
2574 crypto_free_hash(mdev->tconn->verify_tfm);
2575 mdev->tconn->verify_tfm = NULL;
2576 crypto_free_hash(mdev->tconn->cram_hmac_tfm);
2577 mdev->tconn->cram_hmac_tfm = NULL;
2578 crypto_free_hash(mdev->tconn->integrity_w_tfm);
2579 mdev->tconn->integrity_w_tfm = NULL;
2580 crypto_free_hash(mdev->tconn->integrity_r_tfm);
2581 mdev->tconn->integrity_r_tfm = NULL;
2583 drbd_free_sock(mdev->tconn);
2586 drbd_free_bc(mdev->ldev);
2587 mdev->ldev = NULL;);
2590 /* meta data management */
2592 struct meta_data_on_disk {
2593 u64 la_size; /* last agreed size. */
2594 u64 uuid[UI_SIZE]; /* UUIDs. */
2597 u32 flags; /* MDF */
2600 u32 al_offset; /* offset to this block */
2601 u32 al_nr_extents; /* important for restoring the AL */
2602 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2603 u32 bm_offset; /* offset to the bitmap, from here */
2604 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
2605 u32 la_peer_max_bio_size; /* last peer max_bio_size */
2606 u32 reserved_u32[3];
2611 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
2612 * @mdev: DRBD device.
2614 void drbd_md_sync(struct drbd_conf *mdev)
2616 struct meta_data_on_disk *buffer;
2620 del_timer(&mdev->md_sync_timer);
2621 /* timer may be rearmed by drbd_md_mark_dirty() now. */
2622 if (!test_and_clear_bit(MD_DIRTY, &mdev->flags))
2625 /* We use here D_FAILED and not D_ATTACHING because we try to write
2626 * metadata even if we detach due to a disk failure! */
2627 if (!get_ldev_if_state(mdev, D_FAILED))
2630 mutex_lock(&mdev->md_io_mutex);
2631 buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page);
2632 memset(buffer, 0, 512);
2634 buffer->la_size = cpu_to_be64(drbd_get_capacity(mdev->this_bdev));
2635 for (i = UI_CURRENT; i < UI_SIZE; i++)
2636 buffer->uuid[i] = cpu_to_be64(mdev->ldev->md.uuid[i]);
2637 buffer->flags = cpu_to_be32(mdev->ldev->md.flags);
2638 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC);
2640 buffer->md_size_sect = cpu_to_be32(mdev->ldev->md.md_size_sect);
2641 buffer->al_offset = cpu_to_be32(mdev->ldev->md.al_offset);
2642 buffer->al_nr_extents = cpu_to_be32(mdev->act_log->nr_elements);
2643 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
2644 buffer->device_uuid = cpu_to_be64(mdev->ldev->md.device_uuid);
2646 buffer->bm_offset = cpu_to_be32(mdev->ldev->md.bm_offset);
2647 buffer->la_peer_max_bio_size = cpu_to_be32(mdev->peer_max_bio_size);
2649 D_ASSERT(drbd_md_ss__(mdev, mdev->ldev) == mdev->ldev->md.md_offset);
2650 sector = mdev->ldev->md.md_offset;
2652 if (drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) {
2653 /* this was a try anyways ... */
2654 dev_err(DEV, "meta data update failed!\n");
2655 drbd_chk_io_error(mdev, 1, true);
2658 /* Update mdev->ldev->md.la_size_sect,
2659 * since we updated it on metadata. */
2660 mdev->ldev->md.la_size_sect = drbd_get_capacity(mdev->this_bdev);
2662 mutex_unlock(&mdev->md_io_mutex);
2667 * drbd_md_read() - Reads in the meta data super block
2668 * @mdev: DRBD device.
2669 * @bdev: Device from which the meta data should be read in.
2671 * Return 0 (NO_ERROR) on success, and an enum drbd_ret_code in case
2672 * something goes wrong. Currently only: ERR_IO_MD_DISK, ERR_MD_INVALID.
2674 int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
2676 struct meta_data_on_disk *buffer;
2677 int i, rv = NO_ERROR;
2679 if (!get_ldev_if_state(mdev, D_ATTACHING))
2680 return ERR_IO_MD_DISK;
2682 mutex_lock(&mdev->md_io_mutex);
2683 buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page);
2685 if (drbd_md_sync_page_io(mdev, bdev, bdev->md.md_offset, READ)) {
2686 /* NOTE: can't do normal error processing here as this is
2687 called BEFORE disk is attached */
2688 dev_err(DEV, "Error while reading metadata.\n");
2689 rv = ERR_IO_MD_DISK;
2693 if (buffer->magic != cpu_to_be32(DRBD_MD_MAGIC)) {
2694 dev_err(DEV, "Error while reading metadata, magic not found.\n");
2695 rv = ERR_MD_INVALID;
2698 if (be32_to_cpu(buffer->al_offset) != bdev->md.al_offset) {
2699 dev_err(DEV, "unexpected al_offset: %d (expected %d)\n",
2700 be32_to_cpu(buffer->al_offset), bdev->md.al_offset);
2701 rv = ERR_MD_INVALID;
2704 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
2705 dev_err(DEV, "unexpected bm_offset: %d (expected %d)\n",
2706 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
2707 rv = ERR_MD_INVALID;
2710 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
2711 dev_err(DEV, "unexpected md_size: %u (expected %u)\n",
2712 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
2713 rv = ERR_MD_INVALID;
2717 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
2718 dev_err(DEV, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
2719 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
2720 rv = ERR_MD_INVALID;
2724 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size);
2725 for (i = UI_CURRENT; i < UI_SIZE; i++)
2726 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
2727 bdev->md.flags = be32_to_cpu(buffer->flags);
2728 bdev->dc.al_extents = be32_to_cpu(buffer->al_nr_extents);
2729 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
2731 spin_lock_irq(&mdev->tconn->req_lock);
2732 if (mdev->state.conn < C_CONNECTED) {
2734 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
2735 peer = max_t(int, peer, DRBD_MAX_BIO_SIZE_SAFE);
2736 mdev->peer_max_bio_size = peer;
2738 spin_unlock_irq(&mdev->tconn->req_lock);
2740 if (bdev->dc.al_extents < 7)
2741 bdev->dc.al_extents = 127;
2744 mutex_unlock(&mdev->md_io_mutex);
2751 * drbd_md_mark_dirty() - Mark meta data super block as dirty
2752 * @mdev: DRBD device.
2754 * Call this function if you change anything that should be written to
2755 * the meta-data super block. This function sets MD_DIRTY, and starts a
2756 * timer that ensures that within five seconds you have to call drbd_md_sync().
2759 void drbd_md_mark_dirty_(struct drbd_conf *mdev, unsigned int line, const char *func)
2761 if (!test_and_set_bit(MD_DIRTY, &mdev->flags)) {
2762 mod_timer(&mdev->md_sync_timer, jiffies + HZ);
2763 mdev->last_md_mark_dirty.line = line;
2764 mdev->last_md_mark_dirty.func = func;
2768 void drbd_md_mark_dirty(struct drbd_conf *mdev)
2770 if (!test_and_set_bit(MD_DIRTY, &mdev->flags))
2771 mod_timer(&mdev->md_sync_timer, jiffies + 5*HZ);
2775 static void drbd_uuid_move_history(struct drbd_conf *mdev) __must_hold(local)
2779 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
2780 mdev->ldev->md.uuid[i+1] = mdev->ldev->md.uuid[i];
2783 void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
2785 if (idx == UI_CURRENT) {
2786 if (mdev->state.role == R_PRIMARY)
2791 drbd_set_ed_uuid(mdev, val);
2794 mdev->ldev->md.uuid[idx] = val;
2795 drbd_md_mark_dirty(mdev);
2799 void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
2801 if (mdev->ldev->md.uuid[idx]) {
2802 drbd_uuid_move_history(mdev);
2803 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[idx];
2805 _drbd_uuid_set(mdev, idx, val);
2809 * drbd_uuid_new_current() - Creates a new current UUID
2810 * @mdev: DRBD device.
2812 * Creates a new current UUID, and rotates the old current UUID into
2813 * the bitmap slot. Causes an incremental resync upon next connect.
2815 void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local)
2818 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
2821 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
2823 mdev->ldev->md.uuid[UI_BITMAP] = mdev->ldev->md.uuid[UI_CURRENT];
2825 get_random_bytes(&val, sizeof(u64));
2826 _drbd_uuid_set(mdev, UI_CURRENT, val);
2827 drbd_print_uuids(mdev, "new current UUID");
2828 /* get it to stable storage _now_ */
2832 void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local)
2834 if (mdev->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
2838 drbd_uuid_move_history(mdev);
2839 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP];
2840 mdev->ldev->md.uuid[UI_BITMAP] = 0;
2842 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
2844 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
2846 mdev->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
2848 drbd_md_mark_dirty(mdev);
2852 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
2853 * @mdev: DRBD device.
2855 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
2857 int drbd_bmio_set_n_write(struct drbd_conf *mdev)
2861 if (get_ldev_if_state(mdev, D_ATTACHING)) {
2862 drbd_md_set_flag(mdev, MDF_FULL_SYNC);
2864 drbd_bm_set_all(mdev);
2866 rv = drbd_bm_write(mdev);
2869 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
2880 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
2881 * @mdev: DRBD device.
2883 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
2885 int drbd_bmio_clear_n_write(struct drbd_conf *mdev)
2889 drbd_resume_al(mdev);
2890 if (get_ldev_if_state(mdev, D_ATTACHING)) {
2891 drbd_bm_clear_all(mdev);
2892 rv = drbd_bm_write(mdev);
2899 static int w_bitmap_io(struct drbd_work *w, int unused)
2901 struct bm_io_work *work = container_of(w, struct bm_io_work, w);
2902 struct drbd_conf *mdev = w->mdev;
2905 D_ASSERT(atomic_read(&mdev->ap_bio_cnt) == 0);
2907 if (get_ldev(mdev)) {
2908 drbd_bm_lock(mdev, work->why, work->flags);
2909 rv = work->io_fn(mdev);
2910 drbd_bm_unlock(mdev);
2914 clear_bit_unlock(BITMAP_IO, &mdev->flags);
2915 wake_up(&mdev->misc_wait);
2918 work->done(mdev, rv);
2920 clear_bit(BITMAP_IO_QUEUED, &mdev->flags);
2927 void drbd_ldev_destroy(struct drbd_conf *mdev)
2929 lc_destroy(mdev->resync);
2930 mdev->resync = NULL;
2931 lc_destroy(mdev->act_log);
2932 mdev->act_log = NULL;
2934 drbd_free_bc(mdev->ldev);
2935 mdev->ldev = NULL;);
2937 clear_bit(GO_DISKLESS, &mdev->flags);
2940 static int w_go_diskless(struct drbd_work *w, int unused)
2942 struct drbd_conf *mdev = w->mdev;
2944 D_ASSERT(mdev->state.disk == D_FAILED);
2945 /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will
2946 * inc/dec it frequently. Once we are D_DISKLESS, no one will touch
2947 * the protected members anymore, though, so once put_ldev reaches zero
2948 * again, it will be safe to free them. */
2949 drbd_force_state(mdev, NS(disk, D_DISKLESS));
2953 void drbd_go_diskless(struct drbd_conf *mdev)
2955 D_ASSERT(mdev->state.disk == D_FAILED);
2956 if (!test_and_set_bit(GO_DISKLESS, &mdev->flags))
2957 drbd_queue_work(&mdev->tconn->data.work, &mdev->go_diskless);
2961 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
2962 * @mdev: DRBD device.
2963 * @io_fn: IO callback to be called when bitmap IO is possible
2964 * @done: callback to be called after the bitmap IO was performed
2965 * @why: Descriptive text of the reason for doing the IO
2967 * While IO on the bitmap happens we freeze application IO thus we ensure
2968 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
2969 * called from worker context. It MUST NOT be used while a previous such
2970 * work is still pending!
2972 void drbd_queue_bitmap_io(struct drbd_conf *mdev,
2973 int (*io_fn)(struct drbd_conf *),
2974 void (*done)(struct drbd_conf *, int),
2975 char *why, enum bm_flag flags)
2977 D_ASSERT(current == mdev->tconn->worker.task);
2979 D_ASSERT(!test_bit(BITMAP_IO_QUEUED, &mdev->flags));
2980 D_ASSERT(!test_bit(BITMAP_IO, &mdev->flags));
2981 D_ASSERT(list_empty(&mdev->bm_io_work.w.list));
2982 if (mdev->bm_io_work.why)
2983 dev_err(DEV, "FIXME going to queue '%s' but '%s' still pending?\n",
2984 why, mdev->bm_io_work.why);
2986 mdev->bm_io_work.io_fn = io_fn;
2987 mdev->bm_io_work.done = done;
2988 mdev->bm_io_work.why = why;
2989 mdev->bm_io_work.flags = flags;
2991 spin_lock_irq(&mdev->tconn->req_lock);
2992 set_bit(BITMAP_IO, &mdev->flags);
2993 if (atomic_read(&mdev->ap_bio_cnt) == 0) {
2994 if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags))
2995 drbd_queue_work(&mdev->tconn->data.work, &mdev->bm_io_work.w);
2997 spin_unlock_irq(&mdev->tconn->req_lock);
3001 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3002 * @mdev: DRBD device.
3003 * @io_fn: IO callback to be called when bitmap IO is possible
3004 * @why: Descriptive text of the reason for doing the IO
3006 * freezes application IO while that the actual IO operations runs. This
3007 * functions MAY NOT be called from worker context.
3009 int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *),
3010 char *why, enum bm_flag flags)
3014 D_ASSERT(current != mdev->tconn->worker.task);
3016 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3017 drbd_suspend_io(mdev);
3019 drbd_bm_lock(mdev, why, flags);
3021 drbd_bm_unlock(mdev);
3023 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3024 drbd_resume_io(mdev);
3029 void drbd_md_set_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3031 if ((mdev->ldev->md.flags & flag) != flag) {
3032 drbd_md_mark_dirty(mdev);
3033 mdev->ldev->md.flags |= flag;
3037 void drbd_md_clear_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3039 if ((mdev->ldev->md.flags & flag) != 0) {
3040 drbd_md_mark_dirty(mdev);
3041 mdev->ldev->md.flags &= ~flag;
3044 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3046 return (bdev->md.flags & flag) != 0;
3049 static void md_sync_timer_fn(unsigned long data)
3051 struct drbd_conf *mdev = (struct drbd_conf *) data;
3053 drbd_queue_work_front(&mdev->tconn->data.work, &mdev->md_sync_work);
3056 static int w_md_sync(struct drbd_work *w, int unused)
3058 struct drbd_conf *mdev = w->mdev;
3060 dev_warn(DEV, "md_sync_timer expired! Worker calls drbd_md_sync().\n");
3062 dev_warn(DEV, "last md_mark_dirty: %s:%u\n",
3063 mdev->last_md_mark_dirty.func, mdev->last_md_mark_dirty.line);
3069 const char *cmdname(enum drbd_packet cmd)
3071 /* THINK may need to become several global tables
3072 * when we want to support more than
3073 * one PRO_VERSION */
3074 static const char *cmdnames[] = {
3076 [P_DATA_REPLY] = "DataReply",
3077 [P_RS_DATA_REPLY] = "RSDataReply",
3078 [P_BARRIER] = "Barrier",
3079 [P_BITMAP] = "ReportBitMap",
3080 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3081 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3082 [P_UNPLUG_REMOTE] = "UnplugRemote",
3083 [P_DATA_REQUEST] = "DataRequest",
3084 [P_RS_DATA_REQUEST] = "RSDataRequest",
3085 [P_SYNC_PARAM] = "SyncParam",
3086 [P_SYNC_PARAM89] = "SyncParam89",
3087 [P_PROTOCOL] = "ReportProtocol",
3088 [P_UUIDS] = "ReportUUIDs",
3089 [P_SIZES] = "ReportSizes",
3090 [P_STATE] = "ReportState",
3091 [P_SYNC_UUID] = "ReportSyncUUID",
3092 [P_AUTH_CHALLENGE] = "AuthChallenge",
3093 [P_AUTH_RESPONSE] = "AuthResponse",
3095 [P_PING_ACK] = "PingAck",
3096 [P_RECV_ACK] = "RecvAck",
3097 [P_WRITE_ACK] = "WriteAck",
3098 [P_RS_WRITE_ACK] = "RSWriteAck",
3099 [P_DISCARD_WRITE] = "DiscardWrite",
3100 [P_NEG_ACK] = "NegAck",
3101 [P_NEG_DREPLY] = "NegDReply",
3102 [P_NEG_RS_DREPLY] = "NegRSDReply",
3103 [P_BARRIER_ACK] = "BarrierAck",
3104 [P_STATE_CHG_REQ] = "StateChgRequest",
3105 [P_STATE_CHG_REPLY] = "StateChgReply",
3106 [P_OV_REQUEST] = "OVRequest",
3107 [P_OV_REPLY] = "OVReply",
3108 [P_OV_RESULT] = "OVResult",
3109 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3110 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3111 [P_COMPRESSED_BITMAP] = "CBitmap",
3112 [P_DELAY_PROBE] = "DelayProbe",
3113 [P_OUT_OF_SYNC] = "OutOfSync",
3114 [P_RETRY_WRITE] = "RetryWrite",
3117 if (cmd == P_HAND_SHAKE_M)
3118 return "HandShakeM";
3119 if (cmd == P_HAND_SHAKE_S)
3120 return "HandShakeS";
3121 if (cmd == P_HAND_SHAKE)
3123 if (cmd >= ARRAY_SIZE(cmdnames))
3125 return cmdnames[cmd];
3129 * drbd_wait_misc - wait for a request to make progress
3130 * @mdev: device associated with the request
3131 * @i: the struct drbd_interval embedded in struct drbd_request or
3132 * struct drbd_peer_request
3134 int drbd_wait_misc(struct drbd_conf *mdev, struct drbd_interval *i)
3136 struct net_conf *net_conf = mdev->tconn->net_conf;
3142 timeout = MAX_SCHEDULE_TIMEOUT;
3143 if (net_conf->ko_count)
3144 timeout = net_conf->timeout * HZ / 10 * net_conf->ko_count;
3146 /* Indicate to wake up mdev->misc_wait on progress. */
3148 prepare_to_wait(&mdev->misc_wait, &wait, TASK_INTERRUPTIBLE);
3149 spin_unlock_irq(&mdev->tconn->req_lock);
3150 timeout = schedule_timeout(timeout);
3151 finish_wait(&mdev->misc_wait, &wait);
3152 spin_lock_irq(&mdev->tconn->req_lock);
3153 if (!timeout || mdev->state.conn < C_CONNECTED)
3155 if (signal_pending(current))
3156 return -ERESTARTSYS;
3160 #ifdef CONFIG_DRBD_FAULT_INJECTION
3161 /* Fault insertion support including random number generator shamelessly
3162 * stolen from kernel/rcutorture.c */
3163 struct fault_random_state {
3164 unsigned long state;
3165 unsigned long count;
3168 #define FAULT_RANDOM_MULT 39916801 /* prime */
3169 #define FAULT_RANDOM_ADD 479001701 /* prime */
3170 #define FAULT_RANDOM_REFRESH 10000
3173 * Crude but fast random-number generator. Uses a linear congruential
3174 * generator, with occasional help from get_random_bytes().
3176 static unsigned long
3177 _drbd_fault_random(struct fault_random_state *rsp)
3181 if (!rsp->count--) {
3182 get_random_bytes(&refresh, sizeof(refresh));
3183 rsp->state += refresh;
3184 rsp->count = FAULT_RANDOM_REFRESH;
3186 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3187 return swahw32(rsp->state);
3191 _drbd_fault_str(unsigned int type) {
3192 static char *_faults[] = {
3193 [DRBD_FAULT_MD_WR] = "Meta-data write",
3194 [DRBD_FAULT_MD_RD] = "Meta-data read",
3195 [DRBD_FAULT_RS_WR] = "Resync write",
3196 [DRBD_FAULT_RS_RD] = "Resync read",
3197 [DRBD_FAULT_DT_WR] = "Data write",
3198 [DRBD_FAULT_DT_RD] = "Data read",
3199 [DRBD_FAULT_DT_RA] = "Data read ahead",
3200 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3201 [DRBD_FAULT_AL_EE] = "EE allocation",
3202 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3205 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3209 _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type)
3211 static struct fault_random_state rrs = {0, 0};
3213 unsigned int ret = (
3215 ((1 << mdev_to_minor(mdev)) & fault_devs) != 0) &&
3216 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3221 if (__ratelimit(&drbd_ratelimit_state))
3222 dev_warn(DEV, "***Simulating %s failure\n",
3223 _drbd_fault_str(type));
3230 const char *drbd_buildtag(void)
3232 /* DRBD built from external sources has here a reference to the
3233 git hash of the source code. */
3235 static char buildtag[38] = "\0uilt-in";
3237 if (buildtag[0] == 0) {
3238 #ifdef CONFIG_MODULES
3239 if (THIS_MODULE != NULL)
3240 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3249 module_init(drbd_init)
3250 module_exit(drbd_cleanup)
3252 EXPORT_SYMBOL(drbd_conn_str);
3253 EXPORT_SYMBOL(drbd_role_str);
3254 EXPORT_SYMBOL(drbd_disk_str);
3255 EXPORT_SYMBOL(drbd_set_st_err_str);