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 >= 100 || size > DRBD_MAX_SIZE_H80_PACKET)
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_sync_param(struct drbd_conf *mdev)
770 struct p_rs_param_95 *p;
773 const int apv = mdev->tconn->agreed_pro_version;
775 size = apv <= 87 ? sizeof(struct p_rs_param)
776 : apv == 88 ? sizeof(struct p_rs_param)
777 + strlen(mdev->tconn->net_conf->verify_alg) + 1
778 : apv <= 94 ? sizeof(struct p_rs_param_89)
779 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
781 /* used from admin command context and receiver/worker context.
782 * to avoid kmalloc, grab the socket right here,
783 * then use the pre-allocated sbuf there */
784 mutex_lock(&mdev->tconn->data.mutex);
785 sock = mdev->tconn->data.socket;
787 if (likely(sock != NULL)) {
788 enum drbd_packet cmd =
789 apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
791 p = &mdev->tconn->data.sbuf.rs_param_95;
793 /* initialize verify_alg and csums_alg */
794 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
796 if (get_ldev(mdev)) {
797 p->rate = cpu_to_be32(mdev->ldev->dc.resync_rate);
798 p->c_plan_ahead = cpu_to_be32(mdev->ldev->dc.c_plan_ahead);
799 p->c_delay_target = cpu_to_be32(mdev->ldev->dc.c_delay_target);
800 p->c_fill_target = cpu_to_be32(mdev->ldev->dc.c_fill_target);
801 p->c_max_rate = cpu_to_be32(mdev->ldev->dc.c_max_rate);
804 p->rate = cpu_to_be32(DRBD_RATE_DEF);
805 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
806 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
807 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
808 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
812 strcpy(p->verify_alg, mdev->tconn->net_conf->verify_alg);
814 strcpy(p->csums_alg, mdev->tconn->net_conf->csums_alg);
816 err = _drbd_send_cmd(mdev, sock, cmd, &p->head, size, 0);
820 mutex_unlock(&mdev->tconn->data.mutex);
825 int drbd_send_protocol(struct drbd_tconn *tconn)
827 struct p_protocol *p;
830 size = sizeof(struct p_protocol);
832 if (tconn->agreed_pro_version >= 87)
833 size += strlen(tconn->net_conf->integrity_alg) + 1;
835 /* we must not recurse into our own queue,
836 * as that is blocked during handshake */
837 p = kmalloc(size, GFP_NOIO);
841 p->protocol = cpu_to_be32(tconn->net_conf->wire_protocol);
842 p->after_sb_0p = cpu_to_be32(tconn->net_conf->after_sb_0p);
843 p->after_sb_1p = cpu_to_be32(tconn->net_conf->after_sb_1p);
844 p->after_sb_2p = cpu_to_be32(tconn->net_conf->after_sb_2p);
845 p->two_primaries = cpu_to_be32(tconn->net_conf->two_primaries);
848 if (tconn->net_conf->want_lose)
850 if (tconn->net_conf->dry_run) {
851 if (tconn->agreed_pro_version >= 92)
854 conn_err(tconn, "--dry-run is not supported by peer");
859 p->conn_flags = cpu_to_be32(cf);
861 if (tconn->agreed_pro_version >= 87)
862 strcpy(p->integrity_alg, tconn->net_conf->integrity_alg);
864 err = conn_send_cmd2(tconn, P_PROTOCOL, p->head.payload, size - sizeof(struct p_header));
869 int _drbd_send_uuids(struct drbd_conf *mdev, u64 uuid_flags)
874 if (!get_ldev_if_state(mdev, D_NEGOTIATING))
877 for (i = UI_CURRENT; i < UI_SIZE; i++)
878 p.uuid[i] = mdev->ldev ? cpu_to_be64(mdev->ldev->md.uuid[i]) : 0;
880 mdev->comm_bm_set = drbd_bm_total_weight(mdev);
881 p.uuid[UI_SIZE] = cpu_to_be64(mdev->comm_bm_set);
882 uuid_flags |= mdev->tconn->net_conf->want_lose ? 1 : 0;
883 uuid_flags |= test_bit(CRASHED_PRIMARY, &mdev->flags) ? 2 : 0;
884 uuid_flags |= mdev->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
885 p.uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
889 return drbd_send_cmd(mdev, &mdev->tconn->data, P_UUIDS, &p.head, sizeof(p));
892 int drbd_send_uuids(struct drbd_conf *mdev)
894 return _drbd_send_uuids(mdev, 0);
897 int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev)
899 return _drbd_send_uuids(mdev, 8);
902 void drbd_print_uuids(struct drbd_conf *mdev, const char *text)
904 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
905 u64 *uuid = mdev->ldev->md.uuid;
906 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX\n",
908 (unsigned long long)uuid[UI_CURRENT],
909 (unsigned long long)uuid[UI_BITMAP],
910 (unsigned long long)uuid[UI_HISTORY_START],
911 (unsigned long long)uuid[UI_HISTORY_END]);
914 dev_info(DEV, "%s effective data uuid: %016llX\n",
916 (unsigned long long)mdev->ed_uuid);
920 void drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev)
925 D_ASSERT(mdev->state.disk == D_UP_TO_DATE);
927 uuid = mdev->ldev->md.uuid[UI_BITMAP] + UUID_NEW_BM_OFFSET;
928 drbd_uuid_set(mdev, UI_BITMAP, uuid);
929 drbd_print_uuids(mdev, "updated sync UUID");
931 p.uuid = cpu_to_be64(uuid);
933 drbd_send_cmd(mdev, &mdev->tconn->data, P_SYNC_UUID, &p.head, sizeof(p));
936 int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags)
939 sector_t d_size, u_size;
940 int q_order_type, max_bio_size;
942 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
943 D_ASSERT(mdev->ldev->backing_bdev);
944 d_size = drbd_get_max_capacity(mdev->ldev);
945 u_size = mdev->ldev->dc.disk_size;
946 q_order_type = drbd_queue_order_type(mdev);
947 max_bio_size = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
948 max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE);
953 q_order_type = QUEUE_ORDERED_NONE;
954 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
957 p.d_size = cpu_to_be64(d_size);
958 p.u_size = cpu_to_be64(u_size);
959 p.c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(mdev->this_bdev));
960 p.max_bio_size = cpu_to_be32(max_bio_size);
961 p.queue_order_type = cpu_to_be16(q_order_type);
962 p.dds_flags = cpu_to_be16(flags);
964 return drbd_send_cmd(mdev, &mdev->tconn->data, P_SIZES, &p.head, sizeof(p));
968 * drbd_send_state() - Sends the drbd state to the peer
969 * @mdev: DRBD device.
971 int drbd_send_state(struct drbd_conf *mdev)
977 mutex_lock(&mdev->tconn->data.mutex);
979 p.state = cpu_to_be32(mdev->state.i); /* Within the send mutex */
980 sock = mdev->tconn->data.socket;
982 if (likely(sock != NULL))
983 err = _drbd_send_cmd(mdev, sock, P_STATE, &p.head, sizeof(p), 0);
985 mutex_unlock(&mdev->tconn->data.mutex);
990 int _conn_send_state_req(struct drbd_tconn *tconn, int vnr, enum drbd_packet cmd,
991 union drbd_state mask, union drbd_state val)
993 struct p_req_state p;
995 p.mask = cpu_to_be32(mask.i);
996 p.val = cpu_to_be32(val.i);
998 return conn_send_cmd(tconn, vnr, &tconn->data, cmd, &p.head, sizeof(p));
1001 void drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode)
1003 struct p_req_state_reply p;
1005 p.retcode = cpu_to_be32(retcode);
1007 drbd_send_cmd(mdev, &mdev->tconn->meta, P_STATE_CHG_REPLY, &p.head, sizeof(p));
1010 int conn_send_sr_reply(struct drbd_tconn *tconn, enum drbd_state_rv retcode)
1012 struct p_req_state_reply p;
1013 enum drbd_packet cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1015 p.retcode = cpu_to_be32(retcode);
1017 return !conn_send_cmd(tconn, 0, &tconn->meta, cmd, &p.head, sizeof(p));
1020 int fill_bitmap_rle_bits(struct drbd_conf *mdev,
1021 struct p_compressed_bm *p,
1022 struct bm_xfer_ctx *c)
1024 struct bitstream bs;
1025 unsigned long plain_bits;
1032 /* may we use this feature? */
1033 if ((mdev->tconn->net_conf->use_rle == 0) ||
1034 (mdev->tconn->agreed_pro_version < 90))
1037 if (c->bit_offset >= c->bm_bits)
1038 return 0; /* nothing to do. */
1040 /* use at most thus many bytes */
1041 bitstream_init(&bs, p->code, BM_PACKET_VLI_BYTES_MAX, 0);
1042 memset(p->code, 0, BM_PACKET_VLI_BYTES_MAX);
1043 /* plain bits covered in this code string */
1046 /* p->encoding & 0x80 stores whether the first run length is set.
1047 * bit offset is implicit.
1048 * start with toggle == 2 to be able to tell the first iteration */
1051 /* see how much plain bits we can stuff into one packet
1052 * using RLE and VLI. */
1054 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(mdev, c->bit_offset)
1055 : _drbd_bm_find_next(mdev, c->bit_offset);
1058 rl = tmp - c->bit_offset;
1060 if (toggle == 2) { /* first iteration */
1062 /* the first checked bit was set,
1063 * store start value, */
1064 DCBP_set_start(p, 1);
1065 /* but skip encoding of zero run length */
1069 DCBP_set_start(p, 0);
1072 /* paranoia: catch zero runlength.
1073 * can only happen if bitmap is modified while we scan it. */
1075 dev_err(DEV, "unexpected zero runlength while encoding bitmap "
1076 "t:%u bo:%lu\n", toggle, c->bit_offset);
1080 bits = vli_encode_bits(&bs, rl);
1081 if (bits == -ENOBUFS) /* buffer full */
1084 dev_err(DEV, "error while encoding bitmap: %d\n", bits);
1090 c->bit_offset = tmp;
1091 } while (c->bit_offset < c->bm_bits);
1093 len = bs.cur.b - p->code + !!bs.cur.bit;
1095 if (plain_bits < (len << 3)) {
1096 /* incompressible with this method.
1097 * we need to rewind both word and bit position. */
1098 c->bit_offset -= plain_bits;
1099 bm_xfer_ctx_bit_to_word_offset(c);
1100 c->bit_offset = c->word_offset * BITS_PER_LONG;
1104 /* RLE + VLI was able to compress it just fine.
1105 * update c->word_offset. */
1106 bm_xfer_ctx_bit_to_word_offset(c);
1108 /* store pad_bits */
1109 DCBP_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1115 * send_bitmap_rle_or_plain
1117 * Return 0 when done, 1 when another iteration is needed, and a negative error
1118 * code upon failure.
1121 send_bitmap_rle_or_plain(struct drbd_conf *mdev,
1122 struct p_header *h, struct bm_xfer_ctx *c)
1124 struct p_compressed_bm *p = (void*)h;
1125 unsigned long num_words;
1129 len = fill_bitmap_rle_bits(mdev, p, c);
1135 DCBP_set_code(p, RLE_VLI_Bits);
1136 ok = !_drbd_send_cmd(mdev, mdev->tconn->data.socket, P_COMPRESSED_BITMAP, h,
1137 sizeof(*p) + len, 0);
1140 c->bytes[0] += sizeof(*p) + len;
1142 if (c->bit_offset >= c->bm_bits)
1145 /* was not compressible.
1146 * send a buffer full of plain text bits instead. */
1147 num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset);
1148 len = num_words * sizeof(long);
1150 drbd_bm_get_lel(mdev, c->word_offset, num_words, (unsigned long*)h->payload);
1151 ok = !_drbd_send_cmd(mdev, mdev->tconn->data.socket, P_BITMAP,
1152 h, sizeof(struct p_header80) + len, 0);
1153 c->word_offset += num_words;
1154 c->bit_offset = c->word_offset * BITS_PER_LONG;
1157 c->bytes[1] += sizeof(struct p_header80) + len;
1159 if (c->bit_offset > c->bm_bits)
1160 c->bit_offset = c->bm_bits;
1164 INFO_bm_xfer_stats(mdev, "send", c);
1172 /* See the comment at receive_bitmap() */
1173 int _drbd_send_bitmap(struct drbd_conf *mdev)
1175 struct bm_xfer_ctx c;
1179 if (!expect(mdev->bitmap))
1182 /* maybe we should use some per thread scratch page,
1183 * and allocate that during initial device creation? */
1184 p = (struct p_header *) __get_free_page(GFP_NOIO);
1186 dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__);
1190 if (get_ldev(mdev)) {
1191 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1192 dev_info(DEV, "Writing the whole bitmap, MDF_FullSync was set.\n");
1193 drbd_bm_set_all(mdev);
1194 if (drbd_bm_write(mdev)) {
1195 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1196 * but otherwise process as per normal - need to tell other
1197 * side that a full resync is required! */
1198 dev_err(DEV, "Failed to write bitmap to disk!\n");
1200 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
1207 c = (struct bm_xfer_ctx) {
1208 .bm_bits = drbd_bm_bits(mdev),
1209 .bm_words = drbd_bm_words(mdev),
1213 err = send_bitmap_rle_or_plain(mdev, p, &c);
1216 free_page((unsigned long) p);
1220 int drbd_send_bitmap(struct drbd_conf *mdev)
1224 if (drbd_get_data_sock(mdev->tconn))
1226 err = !_drbd_send_bitmap(mdev);
1227 drbd_put_data_sock(mdev->tconn);
1230 void drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, u32 set_size)
1232 struct p_barrier_ack p;
1234 p.barrier = barrier_nr;
1235 p.set_size = cpu_to_be32(set_size);
1237 if (mdev->state.conn >= C_CONNECTED)
1238 drbd_send_cmd(mdev, &mdev->tconn->meta, P_BARRIER_ACK, &p.head, sizeof(p));
1242 * _drbd_send_ack() - Sends an ack packet
1243 * @mdev: DRBD device.
1244 * @cmd: Packet command code.
1245 * @sector: sector, needs to be in big endian byte order
1246 * @blksize: size in byte, needs to be in big endian byte order
1247 * @block_id: Id, big endian byte order
1249 static int _drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1250 u64 sector, u32 blksize, u64 block_id)
1252 struct p_block_ack p;
1255 p.block_id = block_id;
1256 p.blksize = blksize;
1257 p.seq_num = cpu_to_be32(atomic_inc_return(&mdev->packet_seq));
1259 if (!mdev->tconn->meta.socket || mdev->state.conn < C_CONNECTED)
1261 return drbd_send_cmd(mdev, &mdev->tconn->meta, cmd, &p.head, sizeof(p));
1264 /* dp->sector and dp->block_id already/still in network byte order,
1265 * data_size is payload size according to dp->head,
1266 * and may need to be corrected for digest size. */
1267 void drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd,
1268 struct p_data *dp, int data_size)
1270 data_size -= (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ?
1271 crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0;
1272 _drbd_send_ack(mdev, cmd, dp->sector, cpu_to_be32(data_size),
1276 void drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd,
1277 struct p_block_req *rp)
1279 _drbd_send_ack(mdev, cmd, rp->sector, rp->blksize, rp->block_id);
1283 * drbd_send_ack() - Sends an ack packet
1284 * @mdev: DRBD device
1285 * @cmd: packet command code
1286 * @peer_req: peer request
1288 int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1289 struct drbd_peer_request *peer_req)
1291 return _drbd_send_ack(mdev, cmd,
1292 cpu_to_be64(peer_req->i.sector),
1293 cpu_to_be32(peer_req->i.size),
1294 peer_req->block_id);
1297 /* This function misuses the block_id field to signal if the blocks
1298 * are is sync or not. */
1299 int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd,
1300 sector_t sector, int blksize, u64 block_id)
1302 return _drbd_send_ack(mdev, cmd,
1303 cpu_to_be64(sector),
1304 cpu_to_be32(blksize),
1305 cpu_to_be64(block_id));
1308 int drbd_send_drequest(struct drbd_conf *mdev, int cmd,
1309 sector_t sector, int size, u64 block_id)
1312 struct p_block_req p;
1314 p.sector = cpu_to_be64(sector);
1315 p.block_id = block_id;
1316 p.blksize = cpu_to_be32(size);
1318 ok = !drbd_send_cmd(mdev, &mdev->tconn->data, cmd, &p.head, sizeof(p));
1322 int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, int size,
1323 void *digest, int digest_size, enum drbd_packet cmd)
1326 struct p_block_req p;
1328 prepare_header(mdev, &p.head, cmd, sizeof(p) - sizeof(struct p_header) + digest_size);
1329 p.sector = cpu_to_be64(sector);
1330 p.block_id = ID_SYNCER /* unused */;
1331 p.blksize = cpu_to_be32(size);
1333 mutex_lock(&mdev->tconn->data.mutex);
1335 ok = (sizeof(p) == drbd_send(mdev->tconn, mdev->tconn->data.socket, &p, sizeof(p), 0));
1336 ok = ok && (digest_size == drbd_send(mdev->tconn, mdev->tconn->data.socket, digest, digest_size, 0));
1338 mutex_unlock(&mdev->tconn->data.mutex);
1343 int drbd_send_ov_request(struct drbd_conf *mdev, sector_t sector, int size)
1345 struct p_block_req p;
1347 p.sector = cpu_to_be64(sector);
1348 p.block_id = ID_SYNCER /* unused */;
1349 p.blksize = cpu_to_be32(size);
1351 return drbd_send_cmd(mdev, &mdev->tconn->data, P_OV_REQUEST, &p.head, sizeof(p));
1354 /* called on sndtimeo
1355 * returns false if we should retry,
1356 * true if we think connection is dead
1358 static int we_should_drop_the_connection(struct drbd_tconn *tconn, struct socket *sock)
1361 /* long elapsed = (long)(jiffies - mdev->last_received); */
1363 drop_it = tconn->meta.socket == sock
1364 || !tconn->asender.task
1365 || get_t_state(&tconn->asender) != RUNNING
1366 || tconn->cstate < C_WF_REPORT_PARAMS;
1371 drop_it = !--tconn->ko_count;
1373 conn_err(tconn, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1374 current->comm, current->pid, tconn->ko_count);
1375 request_ping(tconn);
1378 return drop_it; /* && (mdev->state == R_PRIMARY) */;
1381 static void drbd_update_congested(struct drbd_tconn *tconn)
1383 struct sock *sk = tconn->data.socket->sk;
1384 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1385 set_bit(NET_CONGESTED, &tconn->flags);
1388 /* The idea of sendpage seems to be to put some kind of reference
1389 * to the page into the skb, and to hand it over to the NIC. In
1390 * this process get_page() gets called.
1392 * As soon as the page was really sent over the network put_page()
1393 * gets called by some part of the network layer. [ NIC driver? ]
1395 * [ get_page() / put_page() increment/decrement the count. If count
1396 * reaches 0 the page will be freed. ]
1398 * This works nicely with pages from FSs.
1399 * But this means that in protocol A we might signal IO completion too early!
1401 * In order not to corrupt data during a resync we must make sure
1402 * that we do not reuse our own buffer pages (EEs) to early, therefore
1403 * we have the net_ee list.
1405 * XFS seems to have problems, still, it submits pages with page_count == 0!
1406 * As a workaround, we disable sendpage on pages
1407 * with page_count == 0 or PageSlab.
1409 static int _drbd_no_send_page(struct drbd_conf *mdev, struct page *page,
1410 int offset, size_t size, unsigned msg_flags)
1412 int sent = drbd_send(mdev->tconn, mdev->tconn->data.socket, kmap(page) + offset, size, msg_flags);
1415 mdev->send_cnt += size>>9;
1416 return sent == size;
1419 static int _drbd_send_page(struct drbd_conf *mdev, struct page *page,
1420 int offset, size_t size, unsigned msg_flags)
1422 mm_segment_t oldfs = get_fs();
1426 /* e.g. XFS meta- & log-data is in slab pages, which have a
1427 * page_count of 0 and/or have PageSlab() set.
1428 * we cannot use send_page for those, as that does get_page();
1429 * put_page(); and would cause either a VM_BUG directly, or
1430 * __page_cache_release a page that would actually still be referenced
1431 * by someone, leading to some obscure delayed Oops somewhere else. */
1432 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1433 return _drbd_no_send_page(mdev, page, offset, size, msg_flags);
1435 msg_flags |= MSG_NOSIGNAL;
1436 drbd_update_congested(mdev->tconn);
1439 sent = mdev->tconn->data.socket->ops->sendpage(mdev->tconn->data.socket, page,
1442 if (sent == -EAGAIN) {
1443 if (we_should_drop_the_connection(mdev->tconn,
1444 mdev->tconn->data.socket))
1450 dev_warn(DEV, "%s: size=%d len=%d sent=%d\n",
1451 __func__, (int)size, len, sent);
1456 } while (len > 0 /* THINK && mdev->cstate >= C_CONNECTED*/);
1458 clear_bit(NET_CONGESTED, &mdev->tconn->flags);
1462 mdev->send_cnt += size>>9;
1466 static int _drbd_send_bio(struct drbd_conf *mdev, struct bio *bio)
1468 struct bio_vec *bvec;
1470 /* hint all but last page with MSG_MORE */
1471 __bio_for_each_segment(bvec, bio, i, 0) {
1472 if (!_drbd_no_send_page(mdev, bvec->bv_page,
1473 bvec->bv_offset, bvec->bv_len,
1474 i == bio->bi_vcnt -1 ? 0 : MSG_MORE))
1480 static int _drbd_send_zc_bio(struct drbd_conf *mdev, struct bio *bio)
1482 struct bio_vec *bvec;
1484 /* hint all but last page with MSG_MORE */
1485 __bio_for_each_segment(bvec, bio, i, 0) {
1486 if (!_drbd_send_page(mdev, bvec->bv_page,
1487 bvec->bv_offset, bvec->bv_len,
1488 i == bio->bi_vcnt -1 ? 0 : MSG_MORE))
1494 static int _drbd_send_zc_ee(struct drbd_conf *mdev,
1495 struct drbd_peer_request *peer_req)
1497 struct page *page = peer_req->pages;
1498 unsigned len = peer_req->i.size;
1500 /* hint all but last page with MSG_MORE */
1501 page_chain_for_each(page) {
1502 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1503 if (!_drbd_send_page(mdev, page, 0, l,
1504 page_chain_next(page) ? MSG_MORE : 0))
1511 static u32 bio_flags_to_wire(struct drbd_conf *mdev, unsigned long bi_rw)
1513 if (mdev->tconn->agreed_pro_version >= 95)
1514 return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) |
1515 (bi_rw & REQ_FUA ? DP_FUA : 0) |
1516 (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) |
1517 (bi_rw & REQ_DISCARD ? DP_DISCARD : 0);
1519 return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0;
1522 /* Used to send write requests
1523 * R_PRIMARY -> Peer (P_DATA)
1525 int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req)
1529 unsigned int dp_flags = 0;
1533 if (drbd_get_data_sock(mdev->tconn))
1536 dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_w_tfm) ?
1537 crypto_hash_digestsize(mdev->tconn->integrity_w_tfm) : 0;
1539 prepare_header(mdev, &p.head, P_DATA, sizeof(p) - sizeof(struct p_header) + dgs + req->i.size);
1540 p.sector = cpu_to_be64(req->i.sector);
1541 p.block_id = (unsigned long)req;
1542 p.seq_num = cpu_to_be32(req->seq_num = atomic_inc_return(&mdev->packet_seq));
1544 dp_flags = bio_flags_to_wire(mdev, req->master_bio->bi_rw);
1546 if (mdev->state.conn >= C_SYNC_SOURCE &&
1547 mdev->state.conn <= C_PAUSED_SYNC_T)
1548 dp_flags |= DP_MAY_SET_IN_SYNC;
1550 p.dp_flags = cpu_to_be32(dp_flags);
1551 set_bit(UNPLUG_REMOTE, &mdev->flags);
1553 drbd_send(mdev->tconn, mdev->tconn->data.socket, &p, sizeof(p), dgs ? MSG_MORE : 0));
1555 dgb = mdev->tconn->int_dig_out;
1556 drbd_csum_bio(mdev, mdev->tconn->integrity_w_tfm, req->master_bio, dgb);
1557 ok = dgs == drbd_send(mdev->tconn, mdev->tconn->data.socket, dgb, dgs, 0);
1560 /* For protocol A, we have to memcpy the payload into
1561 * socket buffers, as we may complete right away
1562 * as soon as we handed it over to tcp, at which point the data
1563 * pages may become invalid.
1565 * For data-integrity enabled, we copy it as well, so we can be
1566 * sure that even if the bio pages may still be modified, it
1567 * won't change the data on the wire, thus if the digest checks
1568 * out ok after sending on this side, but does not fit on the
1569 * receiving side, we sure have detected corruption elsewhere.
1571 if (mdev->tconn->net_conf->wire_protocol == DRBD_PROT_A || dgs)
1572 ok = _drbd_send_bio(mdev, req->master_bio);
1574 ok = _drbd_send_zc_bio(mdev, req->master_bio);
1576 /* double check digest, sometimes buffers have been modified in flight. */
1577 if (dgs > 0 && dgs <= 64) {
1578 /* 64 byte, 512 bit, is the largest digest size
1579 * currently supported in kernel crypto. */
1580 unsigned char digest[64];
1581 drbd_csum_bio(mdev, mdev->tconn->integrity_w_tfm, req->master_bio, digest);
1582 if (memcmp(mdev->tconn->int_dig_out, digest, dgs)) {
1584 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1585 (unsigned long long)req->i.sector, req->i.size);
1587 } /* else if (dgs > 64) {
1588 ... Be noisy about digest too large ...
1592 drbd_put_data_sock(mdev->tconn);
1597 /* answer packet, used to send data back for read requests:
1598 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1599 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1601 int drbd_send_block(struct drbd_conf *mdev, enum drbd_packet cmd,
1602 struct drbd_peer_request *peer_req)
1609 dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_w_tfm) ?
1610 crypto_hash_digestsize(mdev->tconn->integrity_w_tfm) : 0;
1612 prepare_header(mdev, &p.head, cmd, sizeof(p) -
1613 sizeof(struct p_header80) +
1614 dgs + peer_req->i.size);
1615 p.sector = cpu_to_be64(peer_req->i.sector);
1616 p.block_id = peer_req->block_id;
1617 p.seq_num = 0; /* unused */
1619 /* Only called by our kernel thread.
1620 * This one may be interrupted by DRBD_SIG and/or DRBD_SIGKILL
1621 * in response to admin command or module unload.
1623 if (drbd_get_data_sock(mdev->tconn))
1626 ok = sizeof(p) == drbd_send(mdev->tconn, mdev->tconn->data.socket, &p, sizeof(p), dgs ? MSG_MORE : 0);
1628 dgb = mdev->tconn->int_dig_out;
1629 drbd_csum_ee(mdev, mdev->tconn->integrity_w_tfm, peer_req, dgb);
1630 ok = dgs == drbd_send(mdev->tconn, mdev->tconn->data.socket, dgb, dgs, 0);
1633 ok = _drbd_send_zc_ee(mdev, peer_req);
1635 drbd_put_data_sock(mdev->tconn);
1640 int drbd_send_oos(struct drbd_conf *mdev, struct drbd_request *req)
1642 struct p_block_desc p;
1644 p.sector = cpu_to_be64(req->i.sector);
1645 p.blksize = cpu_to_be32(req->i.size);
1647 return !drbd_send_cmd(mdev, &mdev->tconn->data, P_OUT_OF_SYNC, &p.head, sizeof(p));
1651 drbd_send distinguishes two cases:
1653 Packets sent via the data socket "sock"
1654 and packets sent via the meta data socket "msock"
1657 -----------------+-------------------------+------------------------------
1658 timeout conf.timeout / 2 conf.timeout / 2
1659 timeout action send a ping via msock Abort communication
1660 and close all sockets
1664 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1666 int drbd_send(struct drbd_tconn *tconn, struct socket *sock,
1667 void *buf, size_t size, unsigned msg_flags)
1676 /* THINK if (signal_pending) return ... ? */
1681 msg.msg_name = NULL;
1682 msg.msg_namelen = 0;
1683 msg.msg_control = NULL;
1684 msg.msg_controllen = 0;
1685 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1687 if (sock == tconn->data.socket) {
1688 tconn->ko_count = tconn->net_conf->ko_count;
1689 drbd_update_congested(tconn);
1693 * tcp_sendmsg does _not_ use its size parameter at all ?
1695 * -EAGAIN on timeout, -EINTR on signal.
1698 * do we need to block DRBD_SIG if sock == &meta.socket ??
1699 * otherwise wake_asender() might interrupt some send_*Ack !
1701 rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
1702 if (rv == -EAGAIN) {
1703 if (we_should_drop_the_connection(tconn, sock))
1709 flush_signals(current);
1717 } while (sent < size);
1719 if (sock == tconn->data.socket)
1720 clear_bit(NET_CONGESTED, &tconn->flags);
1723 if (rv != -EAGAIN) {
1724 conn_err(tconn, "%s_sendmsg returned %d\n",
1725 sock == tconn->meta.socket ? "msock" : "sock",
1727 conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD);
1729 conn_request_state(tconn, NS(conn, C_TIMEOUT), CS_HARD);
1736 * drbd_send_all - Send an entire buffer
1738 * Returns 0 upon success and a negative error value otherwise.
1740 int drbd_send_all(struct drbd_tconn *tconn, struct socket *sock, void *buffer,
1741 size_t size, unsigned msg_flags)
1745 err = drbd_send(tconn, sock, buffer, size, msg_flags);
1753 static int drbd_open(struct block_device *bdev, fmode_t mode)
1755 struct drbd_conf *mdev = bdev->bd_disk->private_data;
1756 unsigned long flags;
1759 mutex_lock(&drbd_main_mutex);
1760 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
1761 /* to have a stable mdev->state.role
1762 * and no race with updating open_cnt */
1764 if (mdev->state.role != R_PRIMARY) {
1765 if (mode & FMODE_WRITE)
1767 else if (!allow_oos)
1773 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
1774 mutex_unlock(&drbd_main_mutex);
1779 static int drbd_release(struct gendisk *gd, fmode_t mode)
1781 struct drbd_conf *mdev = gd->private_data;
1782 mutex_lock(&drbd_main_mutex);
1784 mutex_unlock(&drbd_main_mutex);
1788 static void drbd_set_defaults(struct drbd_conf *mdev)
1790 /* Beware! The actual layout differs
1791 * between big endian and little endian */
1792 mdev->state = (union drbd_state) {
1793 { .role = R_SECONDARY,
1795 .conn = C_STANDALONE,
1804 void drbd_init_set_defaults(struct drbd_conf *mdev)
1806 /* the memset(,0,) did most of this.
1807 * note: only assignments, no allocation in here */
1809 drbd_set_defaults(mdev);
1811 atomic_set(&mdev->ap_bio_cnt, 0);
1812 atomic_set(&mdev->ap_pending_cnt, 0);
1813 atomic_set(&mdev->rs_pending_cnt, 0);
1814 atomic_set(&mdev->unacked_cnt, 0);
1815 atomic_set(&mdev->local_cnt, 0);
1816 atomic_set(&mdev->pp_in_use, 0);
1817 atomic_set(&mdev->pp_in_use_by_net, 0);
1818 atomic_set(&mdev->rs_sect_in, 0);
1819 atomic_set(&mdev->rs_sect_ev, 0);
1820 atomic_set(&mdev->ap_in_flight, 0);
1822 mutex_init(&mdev->md_io_mutex);
1823 mutex_init(&mdev->own_state_mutex);
1824 mdev->state_mutex = &mdev->own_state_mutex;
1826 spin_lock_init(&mdev->al_lock);
1827 spin_lock_init(&mdev->peer_seq_lock);
1828 spin_lock_init(&mdev->epoch_lock);
1830 INIT_LIST_HEAD(&mdev->active_ee);
1831 INIT_LIST_HEAD(&mdev->sync_ee);
1832 INIT_LIST_HEAD(&mdev->done_ee);
1833 INIT_LIST_HEAD(&mdev->read_ee);
1834 INIT_LIST_HEAD(&mdev->net_ee);
1835 INIT_LIST_HEAD(&mdev->resync_reads);
1836 INIT_LIST_HEAD(&mdev->resync_work.list);
1837 INIT_LIST_HEAD(&mdev->unplug_work.list);
1838 INIT_LIST_HEAD(&mdev->go_diskless.list);
1839 INIT_LIST_HEAD(&mdev->md_sync_work.list);
1840 INIT_LIST_HEAD(&mdev->start_resync_work.list);
1841 INIT_LIST_HEAD(&mdev->bm_io_work.w.list);
1843 mdev->resync_work.cb = w_resync_timer;
1844 mdev->unplug_work.cb = w_send_write_hint;
1845 mdev->go_diskless.cb = w_go_diskless;
1846 mdev->md_sync_work.cb = w_md_sync;
1847 mdev->bm_io_work.w.cb = w_bitmap_io;
1848 mdev->start_resync_work.cb = w_start_resync;
1850 mdev->resync_work.mdev = mdev;
1851 mdev->unplug_work.mdev = mdev;
1852 mdev->go_diskless.mdev = mdev;
1853 mdev->md_sync_work.mdev = mdev;
1854 mdev->bm_io_work.w.mdev = mdev;
1855 mdev->start_resync_work.mdev = mdev;
1857 init_timer(&mdev->resync_timer);
1858 init_timer(&mdev->md_sync_timer);
1859 init_timer(&mdev->start_resync_timer);
1860 init_timer(&mdev->request_timer);
1861 mdev->resync_timer.function = resync_timer_fn;
1862 mdev->resync_timer.data = (unsigned long) mdev;
1863 mdev->md_sync_timer.function = md_sync_timer_fn;
1864 mdev->md_sync_timer.data = (unsigned long) mdev;
1865 mdev->start_resync_timer.function = start_resync_timer_fn;
1866 mdev->start_resync_timer.data = (unsigned long) mdev;
1867 mdev->request_timer.function = request_timer_fn;
1868 mdev->request_timer.data = (unsigned long) mdev;
1870 init_waitqueue_head(&mdev->misc_wait);
1871 init_waitqueue_head(&mdev->state_wait);
1872 init_waitqueue_head(&mdev->ee_wait);
1873 init_waitqueue_head(&mdev->al_wait);
1874 init_waitqueue_head(&mdev->seq_wait);
1876 /* mdev->tconn->agreed_pro_version gets initialized in drbd_connect() */
1877 mdev->write_ordering = WO_bdev_flush;
1878 mdev->resync_wenr = LC_FREE;
1879 mdev->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1880 mdev->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1883 void drbd_mdev_cleanup(struct drbd_conf *mdev)
1886 if (mdev->tconn->receiver.t_state != NONE)
1887 dev_err(DEV, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
1888 mdev->tconn->receiver.t_state);
1890 /* no need to lock it, I'm the only thread alive */
1891 if (atomic_read(&mdev->current_epoch->epoch_size) != 0)
1892 dev_err(DEV, "epoch_size:%d\n", atomic_read(&mdev->current_epoch->epoch_size));
1902 mdev->rs_failed = 0;
1903 mdev->rs_last_events = 0;
1904 mdev->rs_last_sect_ev = 0;
1905 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1906 mdev->rs_mark_left[i] = 0;
1907 mdev->rs_mark_time[i] = 0;
1909 D_ASSERT(mdev->tconn->net_conf == NULL);
1911 drbd_set_my_capacity(mdev, 0);
1913 /* maybe never allocated. */
1914 drbd_bm_resize(mdev, 0, 1);
1915 drbd_bm_cleanup(mdev);
1918 drbd_free_resources(mdev);
1919 clear_bit(AL_SUSPENDED, &mdev->flags);
1922 * currently we drbd_init_ee only on module load, so
1923 * we may do drbd_release_ee only on module unload!
1925 D_ASSERT(list_empty(&mdev->active_ee));
1926 D_ASSERT(list_empty(&mdev->sync_ee));
1927 D_ASSERT(list_empty(&mdev->done_ee));
1928 D_ASSERT(list_empty(&mdev->read_ee));
1929 D_ASSERT(list_empty(&mdev->net_ee));
1930 D_ASSERT(list_empty(&mdev->resync_reads));
1931 D_ASSERT(list_empty(&mdev->tconn->data.work.q));
1932 D_ASSERT(list_empty(&mdev->tconn->meta.work.q));
1933 D_ASSERT(list_empty(&mdev->resync_work.list));
1934 D_ASSERT(list_empty(&mdev->unplug_work.list));
1935 D_ASSERT(list_empty(&mdev->go_diskless.list));
1937 drbd_set_defaults(mdev);
1941 static void drbd_destroy_mempools(void)
1945 while (drbd_pp_pool) {
1946 page = drbd_pp_pool;
1947 drbd_pp_pool = (struct page *)page_private(page);
1952 /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */
1954 if (drbd_md_io_bio_set)
1955 bioset_free(drbd_md_io_bio_set);
1956 if (drbd_md_io_page_pool)
1957 mempool_destroy(drbd_md_io_page_pool);
1958 if (drbd_ee_mempool)
1959 mempool_destroy(drbd_ee_mempool);
1960 if (drbd_request_mempool)
1961 mempool_destroy(drbd_request_mempool);
1963 kmem_cache_destroy(drbd_ee_cache);
1964 if (drbd_request_cache)
1965 kmem_cache_destroy(drbd_request_cache);
1966 if (drbd_bm_ext_cache)
1967 kmem_cache_destroy(drbd_bm_ext_cache);
1968 if (drbd_al_ext_cache)
1969 kmem_cache_destroy(drbd_al_ext_cache);
1971 drbd_md_io_bio_set = NULL;
1972 drbd_md_io_page_pool = NULL;
1973 drbd_ee_mempool = NULL;
1974 drbd_request_mempool = NULL;
1975 drbd_ee_cache = NULL;
1976 drbd_request_cache = NULL;
1977 drbd_bm_ext_cache = NULL;
1978 drbd_al_ext_cache = NULL;
1983 static int drbd_create_mempools(void)
1986 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
1989 /* prepare our caches and mempools */
1990 drbd_request_mempool = NULL;
1991 drbd_ee_cache = NULL;
1992 drbd_request_cache = NULL;
1993 drbd_bm_ext_cache = NULL;
1994 drbd_al_ext_cache = NULL;
1995 drbd_pp_pool = NULL;
1996 drbd_md_io_page_pool = NULL;
1997 drbd_md_io_bio_set = NULL;
2000 drbd_request_cache = kmem_cache_create(
2001 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2002 if (drbd_request_cache == NULL)
2005 drbd_ee_cache = kmem_cache_create(
2006 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2007 if (drbd_ee_cache == NULL)
2010 drbd_bm_ext_cache = kmem_cache_create(
2011 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2012 if (drbd_bm_ext_cache == NULL)
2015 drbd_al_ext_cache = kmem_cache_create(
2016 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2017 if (drbd_al_ext_cache == NULL)
2021 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2022 if (drbd_md_io_bio_set == NULL)
2025 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2026 if (drbd_md_io_page_pool == NULL)
2029 drbd_request_mempool = mempool_create(number,
2030 mempool_alloc_slab, mempool_free_slab, drbd_request_cache);
2031 if (drbd_request_mempool == NULL)
2034 drbd_ee_mempool = mempool_create(number,
2035 mempool_alloc_slab, mempool_free_slab, drbd_ee_cache);
2036 if (drbd_ee_mempool == NULL)
2039 /* drbd's page pool */
2040 spin_lock_init(&drbd_pp_lock);
2042 for (i = 0; i < number; i++) {
2043 page = alloc_page(GFP_HIGHUSER);
2046 set_page_private(page, (unsigned long)drbd_pp_pool);
2047 drbd_pp_pool = page;
2049 drbd_pp_vacant = number;
2054 drbd_destroy_mempools(); /* in case we allocated some */
2058 static int drbd_notify_sys(struct notifier_block *this, unsigned long code,
2061 /* just so we have it. you never know what interesting things we
2062 * might want to do here some day...
2068 static struct notifier_block drbd_notifier = {
2069 .notifier_call = drbd_notify_sys,
2072 static void drbd_release_ee_lists(struct drbd_conf *mdev)
2076 rr = drbd_release_ee(mdev, &mdev->active_ee);
2078 dev_err(DEV, "%d EEs in active list found!\n", rr);
2080 rr = drbd_release_ee(mdev, &mdev->sync_ee);
2082 dev_err(DEV, "%d EEs in sync list found!\n", rr);
2084 rr = drbd_release_ee(mdev, &mdev->read_ee);
2086 dev_err(DEV, "%d EEs in read list found!\n", rr);
2088 rr = drbd_release_ee(mdev, &mdev->done_ee);
2090 dev_err(DEV, "%d EEs in done list found!\n", rr);
2092 rr = drbd_release_ee(mdev, &mdev->net_ee);
2094 dev_err(DEV, "%d EEs in net list found!\n", rr);
2097 /* caution. no locking. */
2098 void drbd_delete_device(unsigned int minor)
2100 struct drbd_conf *mdev = minor_to_mdev(minor);
2105 idr_remove(&mdev->tconn->volumes, mdev->vnr);
2106 idr_remove(&minors, minor);
2109 /* paranoia asserts */
2110 D_ASSERT(mdev->open_cnt == 0);
2111 D_ASSERT(list_empty(&mdev->tconn->data.work.q));
2112 /* end paranoia asserts */
2114 del_gendisk(mdev->vdisk);
2116 /* cleanup stuff that may have been allocated during
2117 * device (re-)configuration or state changes */
2119 if (mdev->this_bdev)
2120 bdput(mdev->this_bdev);
2122 drbd_free_resources(mdev);
2124 drbd_release_ee_lists(mdev);
2126 lc_destroy(mdev->act_log);
2127 lc_destroy(mdev->resync);
2129 kfree(mdev->p_uuid);
2130 /* mdev->p_uuid = NULL; */
2132 /* cleanup the rest that has been
2133 * allocated from drbd_new_device
2134 * and actually free the mdev itself */
2135 drbd_free_mdev(mdev);
2138 static void drbd_cleanup(void)
2141 struct drbd_conf *mdev;
2143 unregister_reboot_notifier(&drbd_notifier);
2145 /* first remove proc,
2146 * drbdsetup uses it's presence to detect
2147 * whether DRBD is loaded.
2148 * If we would get stuck in proc removal,
2149 * but have netlink already deregistered,
2150 * some drbdsetup commands may wait forever
2154 remove_proc_entry("drbd", NULL);
2156 drbd_genl_unregister();
2158 idr_for_each_entry(&minors, mdev, i)
2159 drbd_delete_device(i);
2160 drbd_destroy_mempools();
2161 unregister_blkdev(DRBD_MAJOR, "drbd");
2163 idr_destroy(&minors);
2165 printk(KERN_INFO "drbd: module cleanup done.\n");
2169 * drbd_congested() - Callback for pdflush
2170 * @congested_data: User data
2171 * @bdi_bits: Bits pdflush is currently interested in
2173 * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2175 static int drbd_congested(void *congested_data, int bdi_bits)
2177 struct drbd_conf *mdev = congested_data;
2178 struct request_queue *q;
2182 if (!may_inc_ap_bio(mdev)) {
2183 /* DRBD has frozen IO */
2189 if (get_ldev(mdev)) {
2190 q = bdev_get_queue(mdev->ldev->backing_bdev);
2191 r = bdi_congested(&q->backing_dev_info, bdi_bits);
2197 if (bdi_bits & (1 << BDI_async_congested) && test_bit(NET_CONGESTED, &mdev->tconn->flags)) {
2198 r |= (1 << BDI_async_congested);
2199 reason = reason == 'b' ? 'a' : 'n';
2203 mdev->congestion_reason = reason;
2207 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2209 sema_init(&wq->s, 0);
2210 spin_lock_init(&wq->q_lock);
2211 INIT_LIST_HEAD(&wq->q);
2214 struct drbd_tconn *conn_by_name(const char *name)
2216 struct drbd_tconn *tconn;
2218 if (!name || !name[0])
2221 mutex_lock(&drbd_cfg_mutex);
2222 list_for_each_entry(tconn, &drbd_tconns, all_tconn) {
2223 if (!strcmp(tconn->name, name))
2228 mutex_unlock(&drbd_cfg_mutex);
2232 struct drbd_tconn *drbd_new_tconn(const char *name)
2234 struct drbd_tconn *tconn;
2236 tconn = kzalloc(sizeof(struct drbd_tconn), GFP_KERNEL);
2240 tconn->name = kstrdup(name, GFP_KERNEL);
2244 if (!zalloc_cpumask_var(&tconn->cpu_mask, GFP_KERNEL))
2247 if (!tl_init(tconn))
2250 tconn->cstate = C_STANDALONE;
2251 mutex_init(&tconn->cstate_mutex);
2252 spin_lock_init(&tconn->req_lock);
2253 atomic_set(&tconn->net_cnt, 0);
2254 init_waitqueue_head(&tconn->net_cnt_wait);
2255 init_waitqueue_head(&tconn->ping_wait);
2256 idr_init(&tconn->volumes);
2258 drbd_init_workqueue(&tconn->data.work);
2259 mutex_init(&tconn->data.mutex);
2261 drbd_init_workqueue(&tconn->meta.work);
2262 mutex_init(&tconn->meta.mutex);
2264 drbd_thread_init(tconn, &tconn->receiver, drbdd_init, "receiver");
2265 drbd_thread_init(tconn, &tconn->worker, drbd_worker, "worker");
2266 drbd_thread_init(tconn, &tconn->asender, drbd_asender, "asender");
2268 tconn->res_opts = (struct res_opts) {
2269 {}, 0, /* cpu_mask */
2270 DRBD_ON_NO_DATA_DEF, /* on_no_data */
2273 mutex_lock(&drbd_cfg_mutex);
2274 list_add_tail(&tconn->all_tconn, &drbd_tconns);
2275 mutex_unlock(&drbd_cfg_mutex);
2281 free_cpumask_var(tconn->cpu_mask);
2288 void drbd_free_tconn(struct drbd_tconn *tconn)
2290 list_del(&tconn->all_tconn);
2291 idr_destroy(&tconn->volumes);
2293 free_cpumask_var(tconn->cpu_mask);
2295 kfree(tconn->int_dig_out);
2296 kfree(tconn->int_dig_in);
2297 kfree(tconn->int_dig_vv);
2301 enum drbd_ret_code conn_new_minor(struct drbd_tconn *tconn, unsigned int minor, int vnr)
2303 struct drbd_conf *mdev;
2304 struct gendisk *disk;
2305 struct request_queue *q;
2307 int minor_got = minor;
2308 enum drbd_ret_code err = ERR_NOMEM;
2310 mdev = minor_to_mdev(minor);
2312 return ERR_MINOR_EXISTS;
2314 /* GFP_KERNEL, we are outside of all write-out paths */
2315 mdev = kzalloc(sizeof(struct drbd_conf), GFP_KERNEL);
2319 mdev->tconn = tconn;
2320 mdev->minor = minor;
2323 drbd_init_set_defaults(mdev);
2325 q = blk_alloc_queue(GFP_KERNEL);
2329 q->queuedata = mdev;
2331 disk = alloc_disk(1);
2336 set_disk_ro(disk, true);
2339 disk->major = DRBD_MAJOR;
2340 disk->first_minor = minor;
2341 disk->fops = &drbd_ops;
2342 sprintf(disk->disk_name, "drbd%d", minor);
2343 disk->private_data = mdev;
2345 mdev->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2346 /* we have no partitions. we contain only ourselves. */
2347 mdev->this_bdev->bd_contains = mdev->this_bdev;
2349 q->backing_dev_info.congested_fn = drbd_congested;
2350 q->backing_dev_info.congested_data = mdev;
2352 blk_queue_make_request(q, drbd_make_request);
2353 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2354 This triggers a max_bio_size message upon first attach or connect */
2355 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2356 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2357 blk_queue_merge_bvec(q, drbd_merge_bvec);
2358 q->queue_lock = &mdev->tconn->req_lock; /* needed since we use */
2360 mdev->md_io_page = alloc_page(GFP_KERNEL);
2361 if (!mdev->md_io_page)
2362 goto out_no_io_page;
2364 if (drbd_bm_init(mdev))
2366 mdev->read_requests = RB_ROOT;
2367 mdev->write_requests = RB_ROOT;
2369 mdev->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2370 if (!mdev->current_epoch)
2373 INIT_LIST_HEAD(&mdev->current_epoch->list);
2376 if (!idr_pre_get(&minors, GFP_KERNEL))
2377 goto out_no_minor_idr;
2378 if (idr_get_new_above(&minors, mdev, minor, &minor_got))
2379 goto out_no_minor_idr;
2380 if (minor_got != minor) {
2381 err = ERR_MINOR_EXISTS;
2382 drbd_msg_put_info("requested minor exists already");
2383 goto out_idr_remove_minor;
2386 if (!idr_pre_get(&tconn->volumes, GFP_KERNEL))
2387 goto out_idr_remove_minor;
2388 if (idr_get_new_above(&tconn->volumes, mdev, vnr, &vnr_got))
2389 goto out_idr_remove_minor;
2390 if (vnr_got != vnr) {
2391 err = ERR_INVALID_REQUEST;
2392 drbd_msg_put_info("requested volume exists already");
2393 goto out_idr_remove_vol;
2397 /* inherit the connection state */
2398 mdev->state.conn = tconn->cstate;
2399 if (mdev->state.conn == C_WF_REPORT_PARAMS)
2400 drbd_connected(vnr, mdev, tconn);
2405 idr_remove(&tconn->volumes, vnr_got);
2406 out_idr_remove_minor:
2407 idr_remove(&minors, minor_got);
2410 kfree(mdev->current_epoch);
2412 drbd_bm_cleanup(mdev);
2414 __free_page(mdev->md_io_page);
2418 blk_cleanup_queue(q);
2424 /* counterpart of drbd_new_device.
2425 * last part of drbd_delete_device. */
2426 void drbd_free_mdev(struct drbd_conf *mdev)
2428 kfree(mdev->current_epoch);
2429 if (mdev->bitmap) /* should no longer be there. */
2430 drbd_bm_cleanup(mdev);
2431 __free_page(mdev->md_io_page);
2432 put_disk(mdev->vdisk);
2433 blk_cleanup_queue(mdev->rq_queue);
2438 int __init drbd_init(void)
2442 BUILD_BUG_ON(sizeof(struct p_header80) != sizeof(struct p_header95));
2443 BUILD_BUG_ON(sizeof(struct p_handshake) != 80);
2445 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2447 "drbd: invalid minor_count (%d)\n", minor_count);
2455 err = register_blkdev(DRBD_MAJOR, "drbd");
2458 "drbd: unable to register block device major %d\n",
2463 err = drbd_genl_register();
2465 printk(KERN_ERR "drbd: unable to register generic netlink family\n");
2470 register_reboot_notifier(&drbd_notifier);
2473 * allocate all necessary structs
2477 init_waitqueue_head(&drbd_pp_wait);
2479 drbd_proc = NULL; /* play safe for drbd_cleanup */
2482 err = drbd_create_mempools();
2486 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
2488 printk(KERN_ERR "drbd: unable to register proc file\n");
2492 rwlock_init(&global_state_lock);
2493 INIT_LIST_HEAD(&drbd_tconns);
2495 printk(KERN_INFO "drbd: initialized. "
2496 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2497 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2498 printk(KERN_INFO "drbd: %s\n", drbd_buildtag());
2499 printk(KERN_INFO "drbd: registered as block device major %d\n",
2502 return 0; /* Success! */
2507 /* currently always the case */
2508 printk(KERN_ERR "drbd: ran out of memory\n");
2510 printk(KERN_ERR "drbd: initialization failure\n");
2514 void drbd_free_bc(struct drbd_backing_dev *ldev)
2519 blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2520 blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2525 void drbd_free_sock(struct drbd_tconn *tconn)
2527 if (tconn->data.socket) {
2528 mutex_lock(&tconn->data.mutex);
2529 kernel_sock_shutdown(tconn->data.socket, SHUT_RDWR);
2530 sock_release(tconn->data.socket);
2531 tconn->data.socket = NULL;
2532 mutex_unlock(&tconn->data.mutex);
2534 if (tconn->meta.socket) {
2535 mutex_lock(&tconn->meta.mutex);
2536 kernel_sock_shutdown(tconn->meta.socket, SHUT_RDWR);
2537 sock_release(tconn->meta.socket);
2538 tconn->meta.socket = NULL;
2539 mutex_unlock(&tconn->meta.mutex);
2544 void drbd_free_resources(struct drbd_conf *mdev)
2546 crypto_free_hash(mdev->tconn->csums_tfm);
2547 mdev->tconn->csums_tfm = NULL;
2548 crypto_free_hash(mdev->tconn->verify_tfm);
2549 mdev->tconn->verify_tfm = NULL;
2550 crypto_free_hash(mdev->tconn->cram_hmac_tfm);
2551 mdev->tconn->cram_hmac_tfm = NULL;
2552 crypto_free_hash(mdev->tconn->integrity_w_tfm);
2553 mdev->tconn->integrity_w_tfm = NULL;
2554 crypto_free_hash(mdev->tconn->integrity_r_tfm);
2555 mdev->tconn->integrity_r_tfm = NULL;
2557 drbd_free_sock(mdev->tconn);
2560 drbd_free_bc(mdev->ldev);
2561 mdev->ldev = NULL;);
2564 /* meta data management */
2566 struct meta_data_on_disk {
2567 u64 la_size; /* last agreed size. */
2568 u64 uuid[UI_SIZE]; /* UUIDs. */
2571 u32 flags; /* MDF */
2574 u32 al_offset; /* offset to this block */
2575 u32 al_nr_extents; /* important for restoring the AL */
2576 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2577 u32 bm_offset; /* offset to the bitmap, from here */
2578 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
2579 u32 la_peer_max_bio_size; /* last peer max_bio_size */
2580 u32 reserved_u32[3];
2585 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
2586 * @mdev: DRBD device.
2588 void drbd_md_sync(struct drbd_conf *mdev)
2590 struct meta_data_on_disk *buffer;
2594 del_timer(&mdev->md_sync_timer);
2595 /* timer may be rearmed by drbd_md_mark_dirty() now. */
2596 if (!test_and_clear_bit(MD_DIRTY, &mdev->flags))
2599 /* We use here D_FAILED and not D_ATTACHING because we try to write
2600 * metadata even if we detach due to a disk failure! */
2601 if (!get_ldev_if_state(mdev, D_FAILED))
2604 mutex_lock(&mdev->md_io_mutex);
2605 buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page);
2606 memset(buffer, 0, 512);
2608 buffer->la_size = cpu_to_be64(drbd_get_capacity(mdev->this_bdev));
2609 for (i = UI_CURRENT; i < UI_SIZE; i++)
2610 buffer->uuid[i] = cpu_to_be64(mdev->ldev->md.uuid[i]);
2611 buffer->flags = cpu_to_be32(mdev->ldev->md.flags);
2612 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC);
2614 buffer->md_size_sect = cpu_to_be32(mdev->ldev->md.md_size_sect);
2615 buffer->al_offset = cpu_to_be32(mdev->ldev->md.al_offset);
2616 buffer->al_nr_extents = cpu_to_be32(mdev->act_log->nr_elements);
2617 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
2618 buffer->device_uuid = cpu_to_be64(mdev->ldev->md.device_uuid);
2620 buffer->bm_offset = cpu_to_be32(mdev->ldev->md.bm_offset);
2621 buffer->la_peer_max_bio_size = cpu_to_be32(mdev->peer_max_bio_size);
2623 D_ASSERT(drbd_md_ss__(mdev, mdev->ldev) == mdev->ldev->md.md_offset);
2624 sector = mdev->ldev->md.md_offset;
2626 if (!drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) {
2627 /* this was a try anyways ... */
2628 dev_err(DEV, "meta data update failed!\n");
2629 drbd_chk_io_error(mdev, 1, true);
2632 /* Update mdev->ldev->md.la_size_sect,
2633 * since we updated it on metadata. */
2634 mdev->ldev->md.la_size_sect = drbd_get_capacity(mdev->this_bdev);
2636 mutex_unlock(&mdev->md_io_mutex);
2641 * drbd_md_read() - Reads in the meta data super block
2642 * @mdev: DRBD device.
2643 * @bdev: Device from which the meta data should be read in.
2645 * Return 0 (NO_ERROR) on success, and an enum drbd_ret_code in case
2646 * something goes wrong. Currently only: ERR_IO_MD_DISK, ERR_MD_INVALID.
2648 int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
2650 struct meta_data_on_disk *buffer;
2651 int i, rv = NO_ERROR;
2653 if (!get_ldev_if_state(mdev, D_ATTACHING))
2654 return ERR_IO_MD_DISK;
2656 mutex_lock(&mdev->md_io_mutex);
2657 buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page);
2659 if (!drbd_md_sync_page_io(mdev, bdev, bdev->md.md_offset, READ)) {
2660 /* NOTE: can't do normal error processing here as this is
2661 called BEFORE disk is attached */
2662 dev_err(DEV, "Error while reading metadata.\n");
2663 rv = ERR_IO_MD_DISK;
2667 if (buffer->magic != cpu_to_be32(DRBD_MD_MAGIC)) {
2668 dev_err(DEV, "Error while reading metadata, magic not found.\n");
2669 rv = ERR_MD_INVALID;
2672 if (be32_to_cpu(buffer->al_offset) != bdev->md.al_offset) {
2673 dev_err(DEV, "unexpected al_offset: %d (expected %d)\n",
2674 be32_to_cpu(buffer->al_offset), bdev->md.al_offset);
2675 rv = ERR_MD_INVALID;
2678 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
2679 dev_err(DEV, "unexpected bm_offset: %d (expected %d)\n",
2680 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
2681 rv = ERR_MD_INVALID;
2684 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
2685 dev_err(DEV, "unexpected md_size: %u (expected %u)\n",
2686 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
2687 rv = ERR_MD_INVALID;
2691 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
2692 dev_err(DEV, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
2693 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
2694 rv = ERR_MD_INVALID;
2698 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size);
2699 for (i = UI_CURRENT; i < UI_SIZE; i++)
2700 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
2701 bdev->md.flags = be32_to_cpu(buffer->flags);
2702 bdev->dc.al_extents = be32_to_cpu(buffer->al_nr_extents);
2703 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
2705 spin_lock_irq(&mdev->tconn->req_lock);
2706 if (mdev->state.conn < C_CONNECTED) {
2708 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
2709 peer = max_t(int, peer, DRBD_MAX_BIO_SIZE_SAFE);
2710 mdev->peer_max_bio_size = peer;
2712 spin_unlock_irq(&mdev->tconn->req_lock);
2714 if (bdev->dc.al_extents < 7)
2715 bdev->dc.al_extents = 127;
2718 mutex_unlock(&mdev->md_io_mutex);
2725 * drbd_md_mark_dirty() - Mark meta data super block as dirty
2726 * @mdev: DRBD device.
2728 * Call this function if you change anything that should be written to
2729 * the meta-data super block. This function sets MD_DIRTY, and starts a
2730 * timer that ensures that within five seconds you have to call drbd_md_sync().
2733 void drbd_md_mark_dirty_(struct drbd_conf *mdev, unsigned int line, const char *func)
2735 if (!test_and_set_bit(MD_DIRTY, &mdev->flags)) {
2736 mod_timer(&mdev->md_sync_timer, jiffies + HZ);
2737 mdev->last_md_mark_dirty.line = line;
2738 mdev->last_md_mark_dirty.func = func;
2742 void drbd_md_mark_dirty(struct drbd_conf *mdev)
2744 if (!test_and_set_bit(MD_DIRTY, &mdev->flags))
2745 mod_timer(&mdev->md_sync_timer, jiffies + 5*HZ);
2749 static void drbd_uuid_move_history(struct drbd_conf *mdev) __must_hold(local)
2753 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
2754 mdev->ldev->md.uuid[i+1] = mdev->ldev->md.uuid[i];
2757 void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
2759 if (idx == UI_CURRENT) {
2760 if (mdev->state.role == R_PRIMARY)
2765 drbd_set_ed_uuid(mdev, val);
2768 mdev->ldev->md.uuid[idx] = val;
2769 drbd_md_mark_dirty(mdev);
2773 void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
2775 if (mdev->ldev->md.uuid[idx]) {
2776 drbd_uuid_move_history(mdev);
2777 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[idx];
2779 _drbd_uuid_set(mdev, idx, val);
2783 * drbd_uuid_new_current() - Creates a new current UUID
2784 * @mdev: DRBD device.
2786 * Creates a new current UUID, and rotates the old current UUID into
2787 * the bitmap slot. Causes an incremental resync upon next connect.
2789 void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local)
2792 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
2795 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
2797 mdev->ldev->md.uuid[UI_BITMAP] = mdev->ldev->md.uuid[UI_CURRENT];
2799 get_random_bytes(&val, sizeof(u64));
2800 _drbd_uuid_set(mdev, UI_CURRENT, val);
2801 drbd_print_uuids(mdev, "new current UUID");
2802 /* get it to stable storage _now_ */
2806 void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local)
2808 if (mdev->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
2812 drbd_uuid_move_history(mdev);
2813 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP];
2814 mdev->ldev->md.uuid[UI_BITMAP] = 0;
2816 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
2818 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
2820 mdev->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
2822 drbd_md_mark_dirty(mdev);
2826 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
2827 * @mdev: DRBD device.
2829 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
2831 int drbd_bmio_set_n_write(struct drbd_conf *mdev)
2835 if (get_ldev_if_state(mdev, D_ATTACHING)) {
2836 drbd_md_set_flag(mdev, MDF_FULL_SYNC);
2838 drbd_bm_set_all(mdev);
2840 rv = drbd_bm_write(mdev);
2843 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
2854 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
2855 * @mdev: DRBD device.
2857 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
2859 int drbd_bmio_clear_n_write(struct drbd_conf *mdev)
2863 drbd_resume_al(mdev);
2864 if (get_ldev_if_state(mdev, D_ATTACHING)) {
2865 drbd_bm_clear_all(mdev);
2866 rv = drbd_bm_write(mdev);
2873 static int w_bitmap_io(struct drbd_work *w, int unused)
2875 struct bm_io_work *work = container_of(w, struct bm_io_work, w);
2876 struct drbd_conf *mdev = w->mdev;
2879 D_ASSERT(atomic_read(&mdev->ap_bio_cnt) == 0);
2881 if (get_ldev(mdev)) {
2882 drbd_bm_lock(mdev, work->why, work->flags);
2883 rv = work->io_fn(mdev);
2884 drbd_bm_unlock(mdev);
2888 clear_bit_unlock(BITMAP_IO, &mdev->flags);
2889 wake_up(&mdev->misc_wait);
2892 work->done(mdev, rv);
2894 clear_bit(BITMAP_IO_QUEUED, &mdev->flags);
2901 void drbd_ldev_destroy(struct drbd_conf *mdev)
2903 lc_destroy(mdev->resync);
2904 mdev->resync = NULL;
2905 lc_destroy(mdev->act_log);
2906 mdev->act_log = NULL;
2908 drbd_free_bc(mdev->ldev);
2909 mdev->ldev = NULL;);
2911 clear_bit(GO_DISKLESS, &mdev->flags);
2914 static int w_go_diskless(struct drbd_work *w, int unused)
2916 struct drbd_conf *mdev = w->mdev;
2918 D_ASSERT(mdev->state.disk == D_FAILED);
2919 /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will
2920 * inc/dec it frequently. Once we are D_DISKLESS, no one will touch
2921 * the protected members anymore, though, so once put_ldev reaches zero
2922 * again, it will be safe to free them. */
2923 drbd_force_state(mdev, NS(disk, D_DISKLESS));
2927 void drbd_go_diskless(struct drbd_conf *mdev)
2929 D_ASSERT(mdev->state.disk == D_FAILED);
2930 if (!test_and_set_bit(GO_DISKLESS, &mdev->flags))
2931 drbd_queue_work(&mdev->tconn->data.work, &mdev->go_diskless);
2935 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
2936 * @mdev: DRBD device.
2937 * @io_fn: IO callback to be called when bitmap IO is possible
2938 * @done: callback to be called after the bitmap IO was performed
2939 * @why: Descriptive text of the reason for doing the IO
2941 * While IO on the bitmap happens we freeze application IO thus we ensure
2942 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
2943 * called from worker context. It MUST NOT be used while a previous such
2944 * work is still pending!
2946 void drbd_queue_bitmap_io(struct drbd_conf *mdev,
2947 int (*io_fn)(struct drbd_conf *),
2948 void (*done)(struct drbd_conf *, int),
2949 char *why, enum bm_flag flags)
2951 D_ASSERT(current == mdev->tconn->worker.task);
2953 D_ASSERT(!test_bit(BITMAP_IO_QUEUED, &mdev->flags));
2954 D_ASSERT(!test_bit(BITMAP_IO, &mdev->flags));
2955 D_ASSERT(list_empty(&mdev->bm_io_work.w.list));
2956 if (mdev->bm_io_work.why)
2957 dev_err(DEV, "FIXME going to queue '%s' but '%s' still pending?\n",
2958 why, mdev->bm_io_work.why);
2960 mdev->bm_io_work.io_fn = io_fn;
2961 mdev->bm_io_work.done = done;
2962 mdev->bm_io_work.why = why;
2963 mdev->bm_io_work.flags = flags;
2965 spin_lock_irq(&mdev->tconn->req_lock);
2966 set_bit(BITMAP_IO, &mdev->flags);
2967 if (atomic_read(&mdev->ap_bio_cnt) == 0) {
2968 if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags))
2969 drbd_queue_work(&mdev->tconn->data.work, &mdev->bm_io_work.w);
2971 spin_unlock_irq(&mdev->tconn->req_lock);
2975 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
2976 * @mdev: DRBD device.
2977 * @io_fn: IO callback to be called when bitmap IO is possible
2978 * @why: Descriptive text of the reason for doing the IO
2980 * freezes application IO while that the actual IO operations runs. This
2981 * functions MAY NOT be called from worker context.
2983 int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *),
2984 char *why, enum bm_flag flags)
2988 D_ASSERT(current != mdev->tconn->worker.task);
2990 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
2991 drbd_suspend_io(mdev);
2993 drbd_bm_lock(mdev, why, flags);
2995 drbd_bm_unlock(mdev);
2997 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
2998 drbd_resume_io(mdev);
3003 void drbd_md_set_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3005 if ((mdev->ldev->md.flags & flag) != flag) {
3006 drbd_md_mark_dirty(mdev);
3007 mdev->ldev->md.flags |= flag;
3011 void drbd_md_clear_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3013 if ((mdev->ldev->md.flags & flag) != 0) {
3014 drbd_md_mark_dirty(mdev);
3015 mdev->ldev->md.flags &= ~flag;
3018 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3020 return (bdev->md.flags & flag) != 0;
3023 static void md_sync_timer_fn(unsigned long data)
3025 struct drbd_conf *mdev = (struct drbd_conf *) data;
3027 drbd_queue_work_front(&mdev->tconn->data.work, &mdev->md_sync_work);
3030 static int w_md_sync(struct drbd_work *w, int unused)
3032 struct drbd_conf *mdev = w->mdev;
3034 dev_warn(DEV, "md_sync_timer expired! Worker calls drbd_md_sync().\n");
3036 dev_warn(DEV, "last md_mark_dirty: %s:%u\n",
3037 mdev->last_md_mark_dirty.func, mdev->last_md_mark_dirty.line);
3043 const char *cmdname(enum drbd_packet cmd)
3045 /* THINK may need to become several global tables
3046 * when we want to support more than
3047 * one PRO_VERSION */
3048 static const char *cmdnames[] = {
3050 [P_DATA_REPLY] = "DataReply",
3051 [P_RS_DATA_REPLY] = "RSDataReply",
3052 [P_BARRIER] = "Barrier",
3053 [P_BITMAP] = "ReportBitMap",
3054 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3055 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3056 [P_UNPLUG_REMOTE] = "UnplugRemote",
3057 [P_DATA_REQUEST] = "DataRequest",
3058 [P_RS_DATA_REQUEST] = "RSDataRequest",
3059 [P_SYNC_PARAM] = "SyncParam",
3060 [P_SYNC_PARAM89] = "SyncParam89",
3061 [P_PROTOCOL] = "ReportProtocol",
3062 [P_UUIDS] = "ReportUUIDs",
3063 [P_SIZES] = "ReportSizes",
3064 [P_STATE] = "ReportState",
3065 [P_SYNC_UUID] = "ReportSyncUUID",
3066 [P_AUTH_CHALLENGE] = "AuthChallenge",
3067 [P_AUTH_RESPONSE] = "AuthResponse",
3069 [P_PING_ACK] = "PingAck",
3070 [P_RECV_ACK] = "RecvAck",
3071 [P_WRITE_ACK] = "WriteAck",
3072 [P_RS_WRITE_ACK] = "RSWriteAck",
3073 [P_DISCARD_WRITE] = "DiscardWrite",
3074 [P_NEG_ACK] = "NegAck",
3075 [P_NEG_DREPLY] = "NegDReply",
3076 [P_NEG_RS_DREPLY] = "NegRSDReply",
3077 [P_BARRIER_ACK] = "BarrierAck",
3078 [P_STATE_CHG_REQ] = "StateChgRequest",
3079 [P_STATE_CHG_REPLY] = "StateChgReply",
3080 [P_OV_REQUEST] = "OVRequest",
3081 [P_OV_REPLY] = "OVReply",
3082 [P_OV_RESULT] = "OVResult",
3083 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3084 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3085 [P_COMPRESSED_BITMAP] = "CBitmap",
3086 [P_DELAY_PROBE] = "DelayProbe",
3087 [P_OUT_OF_SYNC] = "OutOfSync",
3088 [P_RETRY_WRITE] = "RetryWrite",
3091 if (cmd == P_HAND_SHAKE_M)
3092 return "HandShakeM";
3093 if (cmd == P_HAND_SHAKE_S)
3094 return "HandShakeS";
3095 if (cmd == P_HAND_SHAKE)
3097 if (cmd >= ARRAY_SIZE(cmdnames))
3099 return cmdnames[cmd];
3103 * drbd_wait_misc - wait for a request to make progress
3104 * @mdev: device associated with the request
3105 * @i: the struct drbd_interval embedded in struct drbd_request or
3106 * struct drbd_peer_request
3108 int drbd_wait_misc(struct drbd_conf *mdev, struct drbd_interval *i)
3110 struct net_conf *net_conf = mdev->tconn->net_conf;
3116 timeout = MAX_SCHEDULE_TIMEOUT;
3117 if (net_conf->ko_count)
3118 timeout = net_conf->timeout * HZ / 10 * net_conf->ko_count;
3120 /* Indicate to wake up mdev->misc_wait on progress. */
3122 prepare_to_wait(&mdev->misc_wait, &wait, TASK_INTERRUPTIBLE);
3123 spin_unlock_irq(&mdev->tconn->req_lock);
3124 timeout = schedule_timeout(timeout);
3125 finish_wait(&mdev->misc_wait, &wait);
3126 spin_lock_irq(&mdev->tconn->req_lock);
3127 if (!timeout || mdev->state.conn < C_CONNECTED)
3129 if (signal_pending(current))
3130 return -ERESTARTSYS;
3134 #ifdef CONFIG_DRBD_FAULT_INJECTION
3135 /* Fault insertion support including random number generator shamelessly
3136 * stolen from kernel/rcutorture.c */
3137 struct fault_random_state {
3138 unsigned long state;
3139 unsigned long count;
3142 #define FAULT_RANDOM_MULT 39916801 /* prime */
3143 #define FAULT_RANDOM_ADD 479001701 /* prime */
3144 #define FAULT_RANDOM_REFRESH 10000
3147 * Crude but fast random-number generator. Uses a linear congruential
3148 * generator, with occasional help from get_random_bytes().
3150 static unsigned long
3151 _drbd_fault_random(struct fault_random_state *rsp)
3155 if (!rsp->count--) {
3156 get_random_bytes(&refresh, sizeof(refresh));
3157 rsp->state += refresh;
3158 rsp->count = FAULT_RANDOM_REFRESH;
3160 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3161 return swahw32(rsp->state);
3165 _drbd_fault_str(unsigned int type) {
3166 static char *_faults[] = {
3167 [DRBD_FAULT_MD_WR] = "Meta-data write",
3168 [DRBD_FAULT_MD_RD] = "Meta-data read",
3169 [DRBD_FAULT_RS_WR] = "Resync write",
3170 [DRBD_FAULT_RS_RD] = "Resync read",
3171 [DRBD_FAULT_DT_WR] = "Data write",
3172 [DRBD_FAULT_DT_RD] = "Data read",
3173 [DRBD_FAULT_DT_RA] = "Data read ahead",
3174 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3175 [DRBD_FAULT_AL_EE] = "EE allocation",
3176 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3179 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3183 _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type)
3185 static struct fault_random_state rrs = {0, 0};
3187 unsigned int ret = (
3189 ((1 << mdev_to_minor(mdev)) & fault_devs) != 0) &&
3190 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3195 if (__ratelimit(&drbd_ratelimit_state))
3196 dev_warn(DEV, "***Simulating %s failure\n",
3197 _drbd_fault_str(type));
3204 const char *drbd_buildtag(void)
3206 /* DRBD built from external sources has here a reference to the
3207 git hash of the source code. */
3209 static char buildtag[38] = "\0uilt-in";
3211 if (buildtag[0] == 0) {
3212 #ifdef CONFIG_MODULES
3213 if (THIS_MODULE != NULL)
3214 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3223 module_init(drbd_init)
3224 module_exit(drbd_cleanup)
3226 EXPORT_SYMBOL(drbd_conn_str);
3227 EXPORT_SYMBOL(drbd_role_str);
3228 EXPORT_SYMBOL(drbd_disk_str);
3229 EXPORT_SYMBOL(drbd_set_st_err_str);