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 DECLARE_RWSEM(drbd_cfg_rwsem);
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 */
457 idr_for_each_entry(&tconn->volumes, mdev, vnr)
458 clear_bit(CREATE_BARRIER, &mdev->flags);
461 spin_unlock_irq(&tconn->req_lock);
464 void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what)
466 spin_lock_irq(&tconn->req_lock);
467 _tl_restart(tconn, what);
468 spin_unlock_irq(&tconn->req_lock);
471 static int drbd_thread_setup(void *arg)
473 struct drbd_thread *thi = (struct drbd_thread *) arg;
474 struct drbd_tconn *tconn = thi->tconn;
478 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
479 thi->name[0], thi->tconn->name);
482 retval = thi->function(thi);
484 spin_lock_irqsave(&thi->t_lock, flags);
486 /* if the receiver has been "EXITING", the last thing it did
487 * was set the conn state to "StandAlone",
488 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
489 * and receiver thread will be "started".
490 * drbd_thread_start needs to set "RESTARTING" in that case.
491 * t_state check and assignment needs to be within the same spinlock,
492 * so either thread_start sees EXITING, and can remap to RESTARTING,
493 * or thread_start see NONE, and can proceed as normal.
496 if (thi->t_state == RESTARTING) {
497 conn_info(tconn, "Restarting %s thread\n", thi->name);
498 thi->t_state = RUNNING;
499 spin_unlock_irqrestore(&thi->t_lock, flags);
506 complete_all(&thi->stop);
507 spin_unlock_irqrestore(&thi->t_lock, flags);
509 conn_info(tconn, "Terminating %s\n", current->comm);
511 /* Release mod reference taken when thread was started */
513 kref_put(&tconn->kref, &conn_destroy);
514 module_put(THIS_MODULE);
518 static void drbd_thread_init(struct drbd_tconn *tconn, struct drbd_thread *thi,
519 int (*func) (struct drbd_thread *), char *name)
521 spin_lock_init(&thi->t_lock);
524 thi->function = func;
526 strncpy(thi->name, name, ARRAY_SIZE(thi->name));
529 int drbd_thread_start(struct drbd_thread *thi)
531 struct drbd_tconn *tconn = thi->tconn;
532 struct task_struct *nt;
535 /* is used from state engine doing drbd_thread_stop_nowait,
536 * while holding the req lock irqsave */
537 spin_lock_irqsave(&thi->t_lock, flags);
539 switch (thi->t_state) {
541 conn_info(tconn, "Starting %s thread (from %s [%d])\n",
542 thi->name, current->comm, current->pid);
544 /* Get ref on module for thread - this is released when thread exits */
545 if (!try_module_get(THIS_MODULE)) {
546 conn_err(tconn, "Failed to get module reference in drbd_thread_start\n");
547 spin_unlock_irqrestore(&thi->t_lock, flags);
551 kref_get(&thi->tconn->kref);
553 init_completion(&thi->stop);
554 thi->reset_cpu_mask = 1;
555 thi->t_state = RUNNING;
556 spin_unlock_irqrestore(&thi->t_lock, flags);
557 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
559 nt = kthread_create(drbd_thread_setup, (void *) thi,
560 "drbd_%c_%s", thi->name[0], thi->tconn->name);
563 conn_err(tconn, "Couldn't start thread\n");
565 kref_put(&tconn->kref, &conn_destroy);
566 module_put(THIS_MODULE);
569 spin_lock_irqsave(&thi->t_lock, flags);
571 thi->t_state = RUNNING;
572 spin_unlock_irqrestore(&thi->t_lock, flags);
576 thi->t_state = RESTARTING;
577 conn_info(tconn, "Restarting %s thread (from %s [%d])\n",
578 thi->name, current->comm, current->pid);
583 spin_unlock_irqrestore(&thi->t_lock, flags);
591 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
595 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
597 /* may be called from state engine, holding the req lock irqsave */
598 spin_lock_irqsave(&thi->t_lock, flags);
600 if (thi->t_state == NONE) {
601 spin_unlock_irqrestore(&thi->t_lock, flags);
603 drbd_thread_start(thi);
607 if (thi->t_state != ns) {
608 if (thi->task == NULL) {
609 spin_unlock_irqrestore(&thi->t_lock, flags);
615 init_completion(&thi->stop);
616 if (thi->task != current)
617 force_sig(DRBD_SIGKILL, thi->task);
620 spin_unlock_irqrestore(&thi->t_lock, flags);
623 wait_for_completion(&thi->stop);
626 static struct drbd_thread *drbd_task_to_thread(struct drbd_tconn *tconn, struct task_struct *task)
628 struct drbd_thread *thi =
629 task == tconn->receiver.task ? &tconn->receiver :
630 task == tconn->asender.task ? &tconn->asender :
631 task == tconn->worker.task ? &tconn->worker : NULL;
636 char *drbd_task_to_thread_name(struct drbd_tconn *tconn, struct task_struct *task)
638 struct drbd_thread *thi = drbd_task_to_thread(tconn, task);
639 return thi ? thi->name : task->comm;
642 int conn_lowest_minor(struct drbd_tconn *tconn)
644 struct drbd_conf *mdev;
648 mdev = idr_get_next(&tconn->volumes, &vnr);
649 m = mdev ? mdev_to_minor(mdev) : -1;
657 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
658 * @mdev: DRBD device.
660 * Forces all threads of a device onto the same CPU. This is beneficial for
661 * DRBD's performance. May be overwritten by user's configuration.
663 void drbd_calc_cpu_mask(struct drbd_tconn *tconn)
668 if (cpumask_weight(tconn->cpu_mask))
671 ord = conn_lowest_minor(tconn) % cpumask_weight(cpu_online_mask);
672 for_each_online_cpu(cpu) {
674 cpumask_set_cpu(cpu, tconn->cpu_mask);
678 /* should not be reached */
679 cpumask_setall(tconn->cpu_mask);
683 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
684 * @mdev: DRBD device.
685 * @thi: drbd_thread object
687 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
690 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
692 struct task_struct *p = current;
694 if (!thi->reset_cpu_mask)
696 thi->reset_cpu_mask = 0;
697 set_cpus_allowed_ptr(p, thi->tconn->cpu_mask);
702 * drbd_header_size - size of a packet header
704 * The header size is a multiple of 8, so any payload following the header is
705 * word aligned on 64-bit architectures. (The bitmap send and receive code
708 unsigned int drbd_header_size(struct drbd_tconn *tconn)
710 if (tconn->agreed_pro_version >= 100) {
711 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
712 return sizeof(struct p_header100);
714 BUILD_BUG_ON(sizeof(struct p_header80) !=
715 sizeof(struct p_header95));
716 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
717 return sizeof(struct p_header80);
721 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
723 h->magic = cpu_to_be32(DRBD_MAGIC);
724 h->command = cpu_to_be16(cmd);
725 h->length = cpu_to_be16(size);
726 return sizeof(struct p_header80);
729 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
731 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
732 h->command = cpu_to_be16(cmd);
733 h->length = cpu_to_be32(size);
734 return sizeof(struct p_header95);
737 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
740 h->magic = cpu_to_be32(DRBD_MAGIC_100);
741 h->volume = cpu_to_be16(vnr);
742 h->command = cpu_to_be16(cmd);
743 h->length = cpu_to_be32(size);
745 return sizeof(struct p_header100);
748 static unsigned int prepare_header(struct drbd_tconn *tconn, int vnr,
749 void *buffer, enum drbd_packet cmd, int size)
751 if (tconn->agreed_pro_version >= 100)
752 return prepare_header100(buffer, cmd, size, vnr);
753 else if (tconn->agreed_pro_version >= 95 &&
754 size > DRBD_MAX_SIZE_H80_PACKET)
755 return prepare_header95(buffer, cmd, size);
757 return prepare_header80(buffer, cmd, size);
760 static void *__conn_prepare_command(struct drbd_tconn *tconn,
761 struct drbd_socket *sock)
765 return sock->sbuf + drbd_header_size(tconn);
768 void *conn_prepare_command(struct drbd_tconn *tconn, struct drbd_socket *sock)
772 mutex_lock(&sock->mutex);
773 p = __conn_prepare_command(tconn, sock);
775 mutex_unlock(&sock->mutex);
780 void *drbd_prepare_command(struct drbd_conf *mdev, struct drbd_socket *sock)
782 return conn_prepare_command(mdev->tconn, sock);
785 static int __send_command(struct drbd_tconn *tconn, int vnr,
786 struct drbd_socket *sock, enum drbd_packet cmd,
787 unsigned int header_size, void *data,
794 * Called with @data == NULL and the size of the data blocks in @size
795 * for commands that send data blocks. For those commands, omit the
796 * MSG_MORE flag: this will increase the likelihood that data blocks
797 * which are page aligned on the sender will end up page aligned on the
800 msg_flags = data ? MSG_MORE : 0;
802 header_size += prepare_header(tconn, vnr, sock->sbuf, cmd,
804 err = drbd_send_all(tconn, sock->socket, sock->sbuf, header_size,
807 err = drbd_send_all(tconn, sock->socket, data, size, 0);
811 static int __conn_send_command(struct drbd_tconn *tconn, struct drbd_socket *sock,
812 enum drbd_packet cmd, unsigned int header_size,
813 void *data, unsigned int size)
815 return __send_command(tconn, 0, sock, cmd, header_size, data, size);
818 int conn_send_command(struct drbd_tconn *tconn, struct drbd_socket *sock,
819 enum drbd_packet cmd, unsigned int header_size,
820 void *data, unsigned int size)
824 err = __conn_send_command(tconn, sock, cmd, header_size, data, size);
825 mutex_unlock(&sock->mutex);
829 int drbd_send_command(struct drbd_conf *mdev, struct drbd_socket *sock,
830 enum drbd_packet cmd, unsigned int header_size,
831 void *data, unsigned int size)
835 err = __send_command(mdev->tconn, mdev->vnr, sock, cmd, header_size,
837 mutex_unlock(&sock->mutex);
841 int drbd_send_ping(struct drbd_tconn *tconn)
843 struct drbd_socket *sock;
846 if (!conn_prepare_command(tconn, sock))
848 return conn_send_command(tconn, sock, P_PING, 0, NULL, 0);
851 int drbd_send_ping_ack(struct drbd_tconn *tconn)
853 struct drbd_socket *sock;
856 if (!conn_prepare_command(tconn, sock))
858 return conn_send_command(tconn, sock, P_PING_ACK, 0, NULL, 0);
861 int drbd_send_sync_param(struct drbd_conf *mdev)
863 struct drbd_socket *sock;
864 struct p_rs_param_95 *p;
866 const int apv = mdev->tconn->agreed_pro_version;
867 enum drbd_packet cmd;
870 sock = &mdev->tconn->data;
871 p = drbd_prepare_command(mdev, sock);
876 nc = rcu_dereference(mdev->tconn->net_conf);
878 size = apv <= 87 ? sizeof(struct p_rs_param)
879 : apv == 88 ? sizeof(struct p_rs_param)
880 + strlen(nc->verify_alg) + 1
881 : apv <= 94 ? sizeof(struct p_rs_param_89)
882 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
884 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
886 /* initialize verify_alg and csums_alg */
887 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
889 if (get_ldev(mdev)) {
890 p->rate = cpu_to_be32(mdev->ldev->dc.resync_rate);
891 p->c_plan_ahead = cpu_to_be32(mdev->ldev->dc.c_plan_ahead);
892 p->c_delay_target = cpu_to_be32(mdev->ldev->dc.c_delay_target);
893 p->c_fill_target = cpu_to_be32(mdev->ldev->dc.c_fill_target);
894 p->c_max_rate = cpu_to_be32(mdev->ldev->dc.c_max_rate);
897 p->rate = cpu_to_be32(DRBD_RATE_DEF);
898 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
899 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
900 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
901 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
905 strcpy(p->verify_alg, nc->verify_alg);
907 strcpy(p->csums_alg, nc->csums_alg);
910 return drbd_send_command(mdev, sock, cmd, size, NULL, 0);
913 int __drbd_send_protocol(struct drbd_tconn *tconn)
915 struct drbd_socket *sock;
916 struct p_protocol *p;
921 p = __conn_prepare_command(tconn, sock);
926 nc = rcu_dereference(tconn->net_conf);
928 if (nc->dry_run && tconn->agreed_pro_version < 92) {
930 mutex_unlock(&sock->mutex);
931 conn_err(tconn, "--dry-run is not supported by peer");
936 if (tconn->agreed_pro_version >= 87)
937 size += strlen(nc->integrity_alg) + 1;
939 p->protocol = cpu_to_be32(nc->wire_protocol);
940 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
941 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
942 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
943 p->two_primaries = cpu_to_be32(nc->two_primaries);
949 p->conn_flags = cpu_to_be32(cf);
951 if (tconn->agreed_pro_version >= 87)
952 strcpy(p->integrity_alg, nc->integrity_alg);
955 return __conn_send_command(tconn, sock, P_PROTOCOL, size, NULL, 0);
958 int drbd_send_protocol(struct drbd_tconn *tconn)
962 mutex_lock(&tconn->data.mutex);
963 err = __drbd_send_protocol(tconn);
964 mutex_unlock(&tconn->data.mutex);
969 int _drbd_send_uuids(struct drbd_conf *mdev, u64 uuid_flags)
971 struct drbd_socket *sock;
975 if (!get_ldev_if_state(mdev, D_NEGOTIATING))
978 sock = &mdev->tconn->data;
979 p = drbd_prepare_command(mdev, sock);
984 for (i = UI_CURRENT; i < UI_SIZE; i++)
985 p->uuid[i] = mdev->ldev ? cpu_to_be64(mdev->ldev->md.uuid[i]) : 0;
987 mdev->comm_bm_set = drbd_bm_total_weight(mdev);
988 p->uuid[UI_SIZE] = cpu_to_be64(mdev->comm_bm_set);
990 uuid_flags |= rcu_dereference(mdev->tconn->net_conf)->want_lose ? 1 : 0;
992 uuid_flags |= test_bit(CRASHED_PRIMARY, &mdev->flags) ? 2 : 0;
993 uuid_flags |= mdev->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
994 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
997 return drbd_send_command(mdev, sock, P_UUIDS, sizeof(*p), NULL, 0);
1000 int drbd_send_uuids(struct drbd_conf *mdev)
1002 return _drbd_send_uuids(mdev, 0);
1005 int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev)
1007 return _drbd_send_uuids(mdev, 8);
1010 void drbd_print_uuids(struct drbd_conf *mdev, const char *text)
1012 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
1013 u64 *uuid = mdev->ldev->md.uuid;
1014 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX\n",
1016 (unsigned long long)uuid[UI_CURRENT],
1017 (unsigned long long)uuid[UI_BITMAP],
1018 (unsigned long long)uuid[UI_HISTORY_START],
1019 (unsigned long long)uuid[UI_HISTORY_END]);
1022 dev_info(DEV, "%s effective data uuid: %016llX\n",
1024 (unsigned long long)mdev->ed_uuid);
1028 void drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev)
1030 struct drbd_socket *sock;
1031 struct p_rs_uuid *p;
1034 D_ASSERT(mdev->state.disk == D_UP_TO_DATE);
1036 uuid = mdev->ldev->md.uuid[UI_BITMAP] + UUID_NEW_BM_OFFSET;
1037 drbd_uuid_set(mdev, UI_BITMAP, uuid);
1038 drbd_print_uuids(mdev, "updated sync UUID");
1041 sock = &mdev->tconn->data;
1042 p = drbd_prepare_command(mdev, sock);
1044 p->uuid = cpu_to_be64(uuid);
1045 drbd_send_command(mdev, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
1049 int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags)
1051 struct drbd_socket *sock;
1053 sector_t d_size, u_size;
1054 int q_order_type, max_bio_size;
1056 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
1057 D_ASSERT(mdev->ldev->backing_bdev);
1058 d_size = drbd_get_max_capacity(mdev->ldev);
1059 u_size = mdev->ldev->dc.disk_size;
1060 q_order_type = drbd_queue_order_type(mdev);
1061 max_bio_size = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1062 max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE);
1067 q_order_type = QUEUE_ORDERED_NONE;
1068 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
1071 sock = &mdev->tconn->data;
1072 p = drbd_prepare_command(mdev, sock);
1075 p->d_size = cpu_to_be64(d_size);
1076 p->u_size = cpu_to_be64(u_size);
1077 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(mdev->this_bdev));
1078 p->max_bio_size = cpu_to_be32(max_bio_size);
1079 p->queue_order_type = cpu_to_be16(q_order_type);
1080 p->dds_flags = cpu_to_be16(flags);
1081 return drbd_send_command(mdev, sock, P_SIZES, sizeof(*p), NULL, 0);
1085 * drbd_send_state() - Sends the drbd state to the peer
1086 * @mdev: DRBD device.
1088 int drbd_send_state(struct drbd_conf *mdev)
1090 struct drbd_socket *sock;
1093 sock = &mdev->tconn->data;
1094 p = drbd_prepare_command(mdev, sock);
1097 p->state = cpu_to_be32(mdev->state.i); /* Within the send mutex */
1098 return drbd_send_command(mdev, sock, P_STATE, sizeof(*p), NULL, 0);
1101 int drbd_send_state_req(struct drbd_conf *mdev, union drbd_state mask, union drbd_state val)
1103 struct drbd_socket *sock;
1104 struct p_req_state *p;
1106 sock = &mdev->tconn->data;
1107 p = drbd_prepare_command(mdev, sock);
1110 p->mask = cpu_to_be32(mask.i);
1111 p->val = cpu_to_be32(val.i);
1112 return drbd_send_command(mdev, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1116 int conn_send_state_req(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val)
1118 enum drbd_packet cmd;
1119 struct drbd_socket *sock;
1120 struct p_req_state *p;
1122 cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1123 sock = &tconn->data;
1124 p = conn_prepare_command(tconn, sock);
1127 p->mask = cpu_to_be32(mask.i);
1128 p->val = cpu_to_be32(val.i);
1129 return conn_send_command(tconn, sock, cmd, sizeof(*p), NULL, 0);
1132 void drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode)
1134 struct drbd_socket *sock;
1135 struct p_req_state_reply *p;
1137 sock = &mdev->tconn->meta;
1138 p = drbd_prepare_command(mdev, sock);
1140 p->retcode = cpu_to_be32(retcode);
1141 drbd_send_command(mdev, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1145 void conn_send_sr_reply(struct drbd_tconn *tconn, enum drbd_state_rv retcode)
1147 struct drbd_socket *sock;
1148 struct p_req_state_reply *p;
1149 enum drbd_packet cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1151 sock = &tconn->meta;
1152 p = conn_prepare_command(tconn, sock);
1154 p->retcode = cpu_to_be32(retcode);
1155 conn_send_command(tconn, sock, cmd, sizeof(*p), NULL, 0);
1159 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1161 BUG_ON(code & ~0xf);
1162 p->encoding = (p->encoding & ~0xf) | code;
1165 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1167 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1170 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1173 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1176 int fill_bitmap_rle_bits(struct drbd_conf *mdev,
1177 struct p_compressed_bm *p,
1179 struct bm_xfer_ctx *c)
1181 struct bitstream bs;
1182 unsigned long plain_bits;
1189 /* may we use this feature? */
1191 use_rle = rcu_dereference(mdev->tconn->net_conf)->use_rle;
1193 if (!use_rle || mdev->tconn->agreed_pro_version < 90)
1196 if (c->bit_offset >= c->bm_bits)
1197 return 0; /* nothing to do. */
1199 /* use at most thus many bytes */
1200 bitstream_init(&bs, p->code, size, 0);
1201 memset(p->code, 0, size);
1202 /* plain bits covered in this code string */
1205 /* p->encoding & 0x80 stores whether the first run length is set.
1206 * bit offset is implicit.
1207 * start with toggle == 2 to be able to tell the first iteration */
1210 /* see how much plain bits we can stuff into one packet
1211 * using RLE and VLI. */
1213 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(mdev, c->bit_offset)
1214 : _drbd_bm_find_next(mdev, c->bit_offset);
1217 rl = tmp - c->bit_offset;
1219 if (toggle == 2) { /* first iteration */
1221 /* the first checked bit was set,
1222 * store start value, */
1223 dcbp_set_start(p, 1);
1224 /* but skip encoding of zero run length */
1228 dcbp_set_start(p, 0);
1231 /* paranoia: catch zero runlength.
1232 * can only happen if bitmap is modified while we scan it. */
1234 dev_err(DEV, "unexpected zero runlength while encoding bitmap "
1235 "t:%u bo:%lu\n", toggle, c->bit_offset);
1239 bits = vli_encode_bits(&bs, rl);
1240 if (bits == -ENOBUFS) /* buffer full */
1243 dev_err(DEV, "error while encoding bitmap: %d\n", bits);
1249 c->bit_offset = tmp;
1250 } while (c->bit_offset < c->bm_bits);
1252 len = bs.cur.b - p->code + !!bs.cur.bit;
1254 if (plain_bits < (len << 3)) {
1255 /* incompressible with this method.
1256 * we need to rewind both word and bit position. */
1257 c->bit_offset -= plain_bits;
1258 bm_xfer_ctx_bit_to_word_offset(c);
1259 c->bit_offset = c->word_offset * BITS_PER_LONG;
1263 /* RLE + VLI was able to compress it just fine.
1264 * update c->word_offset. */
1265 bm_xfer_ctx_bit_to_word_offset(c);
1267 /* store pad_bits */
1268 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1274 * send_bitmap_rle_or_plain
1276 * Return 0 when done, 1 when another iteration is needed, and a negative error
1277 * code upon failure.
1280 send_bitmap_rle_or_plain(struct drbd_conf *mdev, struct bm_xfer_ctx *c)
1282 struct drbd_socket *sock = &mdev->tconn->data;
1283 unsigned int header_size = drbd_header_size(mdev->tconn);
1284 struct p_compressed_bm *p = sock->sbuf + header_size;
1287 len = fill_bitmap_rle_bits(mdev, p,
1288 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1293 dcbp_set_code(p, RLE_VLI_Bits);
1294 err = __send_command(mdev->tconn, mdev->vnr, sock,
1295 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1298 c->bytes[0] += header_size + sizeof(*p) + len;
1300 if (c->bit_offset >= c->bm_bits)
1303 /* was not compressible.
1304 * send a buffer full of plain text bits instead. */
1305 unsigned int data_size;
1306 unsigned long num_words;
1307 unsigned long *p = sock->sbuf + header_size;
1309 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1310 num_words = min_t(size_t, data_size / sizeof(*p),
1311 c->bm_words - c->word_offset);
1312 len = num_words * sizeof(*p);
1314 drbd_bm_get_lel(mdev, c->word_offset, num_words, p);
1315 err = __send_command(mdev->tconn, mdev->vnr, sock, P_BITMAP, len, NULL, 0);
1316 c->word_offset += num_words;
1317 c->bit_offset = c->word_offset * BITS_PER_LONG;
1320 c->bytes[1] += header_size + len;
1322 if (c->bit_offset > c->bm_bits)
1323 c->bit_offset = c->bm_bits;
1327 INFO_bm_xfer_stats(mdev, "send", c);
1335 /* See the comment at receive_bitmap() */
1336 static int _drbd_send_bitmap(struct drbd_conf *mdev)
1338 struct bm_xfer_ctx c;
1341 if (!expect(mdev->bitmap))
1344 if (get_ldev(mdev)) {
1345 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1346 dev_info(DEV, "Writing the whole bitmap, MDF_FullSync was set.\n");
1347 drbd_bm_set_all(mdev);
1348 if (drbd_bm_write(mdev)) {
1349 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1350 * but otherwise process as per normal - need to tell other
1351 * side that a full resync is required! */
1352 dev_err(DEV, "Failed to write bitmap to disk!\n");
1354 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
1361 c = (struct bm_xfer_ctx) {
1362 .bm_bits = drbd_bm_bits(mdev),
1363 .bm_words = drbd_bm_words(mdev),
1367 err = send_bitmap_rle_or_plain(mdev, &c);
1373 int drbd_send_bitmap(struct drbd_conf *mdev)
1375 struct drbd_socket *sock = &mdev->tconn->data;
1378 mutex_lock(&sock->mutex);
1380 err = !_drbd_send_bitmap(mdev);
1381 mutex_unlock(&sock->mutex);
1385 void drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, u32 set_size)
1387 struct drbd_socket *sock;
1388 struct p_barrier_ack *p;
1390 if (mdev->state.conn < C_CONNECTED)
1393 sock = &mdev->tconn->meta;
1394 p = drbd_prepare_command(mdev, sock);
1397 p->barrier = barrier_nr;
1398 p->set_size = cpu_to_be32(set_size);
1399 drbd_send_command(mdev, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1403 * _drbd_send_ack() - Sends an ack packet
1404 * @mdev: DRBD device.
1405 * @cmd: Packet command code.
1406 * @sector: sector, needs to be in big endian byte order
1407 * @blksize: size in byte, needs to be in big endian byte order
1408 * @block_id: Id, big endian byte order
1410 static int _drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1411 u64 sector, u32 blksize, u64 block_id)
1413 struct drbd_socket *sock;
1414 struct p_block_ack *p;
1416 if (mdev->state.conn < C_CONNECTED)
1419 sock = &mdev->tconn->meta;
1420 p = drbd_prepare_command(mdev, sock);
1424 p->block_id = block_id;
1425 p->blksize = blksize;
1426 p->seq_num = cpu_to_be32(atomic_inc_return(&mdev->packet_seq));
1427 return drbd_send_command(mdev, sock, cmd, sizeof(*p), NULL, 0);
1430 /* dp->sector and dp->block_id already/still in network byte order,
1431 * data_size is payload size according to dp->head,
1432 * and may need to be corrected for digest size. */
1433 void drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd,
1434 struct p_data *dp, int data_size)
1436 if (mdev->tconn->peer_integrity_tfm)
1437 data_size -= crypto_hash_digestsize(mdev->tconn->peer_integrity_tfm);
1438 _drbd_send_ack(mdev, cmd, dp->sector, cpu_to_be32(data_size),
1442 void drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd,
1443 struct p_block_req *rp)
1445 _drbd_send_ack(mdev, cmd, rp->sector, rp->blksize, rp->block_id);
1449 * drbd_send_ack() - Sends an ack packet
1450 * @mdev: DRBD device
1451 * @cmd: packet command code
1452 * @peer_req: peer request
1454 int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1455 struct drbd_peer_request *peer_req)
1457 return _drbd_send_ack(mdev, cmd,
1458 cpu_to_be64(peer_req->i.sector),
1459 cpu_to_be32(peer_req->i.size),
1460 peer_req->block_id);
1463 /* This function misuses the block_id field to signal if the blocks
1464 * are is sync or not. */
1465 int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd,
1466 sector_t sector, int blksize, u64 block_id)
1468 return _drbd_send_ack(mdev, cmd,
1469 cpu_to_be64(sector),
1470 cpu_to_be32(blksize),
1471 cpu_to_be64(block_id));
1474 int drbd_send_drequest(struct drbd_conf *mdev, int cmd,
1475 sector_t sector, int size, u64 block_id)
1477 struct drbd_socket *sock;
1478 struct p_block_req *p;
1480 sock = &mdev->tconn->data;
1481 p = drbd_prepare_command(mdev, sock);
1484 p->sector = cpu_to_be64(sector);
1485 p->block_id = block_id;
1486 p->blksize = cpu_to_be32(size);
1487 return drbd_send_command(mdev, sock, cmd, sizeof(*p), NULL, 0);
1490 int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, int size,
1491 void *digest, int digest_size, enum drbd_packet cmd)
1493 struct drbd_socket *sock;
1494 struct p_block_req *p;
1496 /* FIXME: Put the digest into the preallocated socket buffer. */
1498 sock = &mdev->tconn->data;
1499 p = drbd_prepare_command(mdev, sock);
1502 p->sector = cpu_to_be64(sector);
1503 p->block_id = ID_SYNCER /* unused */;
1504 p->blksize = cpu_to_be32(size);
1505 return drbd_send_command(mdev, sock, cmd, sizeof(*p),
1506 digest, digest_size);
1509 int drbd_send_ov_request(struct drbd_conf *mdev, sector_t sector, int size)
1511 struct drbd_socket *sock;
1512 struct p_block_req *p;
1514 sock = &mdev->tconn->data;
1515 p = drbd_prepare_command(mdev, sock);
1518 p->sector = cpu_to_be64(sector);
1519 p->block_id = ID_SYNCER /* unused */;
1520 p->blksize = cpu_to_be32(size);
1521 return drbd_send_command(mdev, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1524 /* called on sndtimeo
1525 * returns false if we should retry,
1526 * true if we think connection is dead
1528 static int we_should_drop_the_connection(struct drbd_tconn *tconn, struct socket *sock)
1531 /* long elapsed = (long)(jiffies - mdev->last_received); */
1533 drop_it = tconn->meta.socket == sock
1534 || !tconn->asender.task
1535 || get_t_state(&tconn->asender) != RUNNING
1536 || tconn->cstate < C_WF_REPORT_PARAMS;
1541 drop_it = !--tconn->ko_count;
1543 conn_err(tconn, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1544 current->comm, current->pid, tconn->ko_count);
1545 request_ping(tconn);
1548 return drop_it; /* && (mdev->state == R_PRIMARY) */;
1551 static void drbd_update_congested(struct drbd_tconn *tconn)
1553 struct sock *sk = tconn->data.socket->sk;
1554 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1555 set_bit(NET_CONGESTED, &tconn->flags);
1558 /* The idea of sendpage seems to be to put some kind of reference
1559 * to the page into the skb, and to hand it over to the NIC. In
1560 * this process get_page() gets called.
1562 * As soon as the page was really sent over the network put_page()
1563 * gets called by some part of the network layer. [ NIC driver? ]
1565 * [ get_page() / put_page() increment/decrement the count. If count
1566 * reaches 0 the page will be freed. ]
1568 * This works nicely with pages from FSs.
1569 * But this means that in protocol A we might signal IO completion too early!
1571 * In order not to corrupt data during a resync we must make sure
1572 * that we do not reuse our own buffer pages (EEs) to early, therefore
1573 * we have the net_ee list.
1575 * XFS seems to have problems, still, it submits pages with page_count == 0!
1576 * As a workaround, we disable sendpage on pages
1577 * with page_count == 0 or PageSlab.
1579 static int _drbd_no_send_page(struct drbd_conf *mdev, struct page *page,
1580 int offset, size_t size, unsigned msg_flags)
1582 struct socket *socket;
1586 socket = mdev->tconn->data.socket;
1587 addr = kmap(page) + offset;
1588 err = drbd_send_all(mdev->tconn, socket, addr, size, msg_flags);
1591 mdev->send_cnt += size >> 9;
1595 static int _drbd_send_page(struct drbd_conf *mdev, struct page *page,
1596 int offset, size_t size, unsigned msg_flags)
1598 struct socket *socket = mdev->tconn->data.socket;
1599 mm_segment_t oldfs = get_fs();
1603 /* e.g. XFS meta- & log-data is in slab pages, which have a
1604 * page_count of 0 and/or have PageSlab() set.
1605 * we cannot use send_page for those, as that does get_page();
1606 * put_page(); and would cause either a VM_BUG directly, or
1607 * __page_cache_release a page that would actually still be referenced
1608 * by someone, leading to some obscure delayed Oops somewhere else. */
1609 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1610 return _drbd_no_send_page(mdev, page, offset, size, msg_flags);
1612 msg_flags |= MSG_NOSIGNAL;
1613 drbd_update_congested(mdev->tconn);
1618 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1620 if (sent == -EAGAIN) {
1621 if (we_should_drop_the_connection(mdev->tconn, socket))
1625 dev_warn(DEV, "%s: size=%d len=%d sent=%d\n",
1626 __func__, (int)size, len, sent);
1633 } while (len > 0 /* THINK && mdev->cstate >= C_CONNECTED*/);
1635 clear_bit(NET_CONGESTED, &mdev->tconn->flags);
1639 mdev->send_cnt += size >> 9;
1644 static int _drbd_send_bio(struct drbd_conf *mdev, struct bio *bio)
1646 struct bio_vec *bvec;
1648 /* hint all but last page with MSG_MORE */
1649 __bio_for_each_segment(bvec, bio, i, 0) {
1652 err = _drbd_no_send_page(mdev, bvec->bv_page,
1653 bvec->bv_offset, bvec->bv_len,
1654 i == bio->bi_vcnt - 1 ? 0 : MSG_MORE);
1661 static int _drbd_send_zc_bio(struct drbd_conf *mdev, struct bio *bio)
1663 struct bio_vec *bvec;
1665 /* hint all but last page with MSG_MORE */
1666 __bio_for_each_segment(bvec, bio, i, 0) {
1669 err = _drbd_send_page(mdev, bvec->bv_page,
1670 bvec->bv_offset, bvec->bv_len,
1671 i == bio->bi_vcnt - 1 ? 0 : MSG_MORE);
1678 static int _drbd_send_zc_ee(struct drbd_conf *mdev,
1679 struct drbd_peer_request *peer_req)
1681 struct page *page = peer_req->pages;
1682 unsigned len = peer_req->i.size;
1685 /* hint all but last page with MSG_MORE */
1686 page_chain_for_each(page) {
1687 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1689 err = _drbd_send_page(mdev, page, 0, l,
1690 page_chain_next(page) ? MSG_MORE : 0);
1698 static u32 bio_flags_to_wire(struct drbd_conf *mdev, unsigned long bi_rw)
1700 if (mdev->tconn->agreed_pro_version >= 95)
1701 return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) |
1702 (bi_rw & REQ_FUA ? DP_FUA : 0) |
1703 (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) |
1704 (bi_rw & REQ_DISCARD ? DP_DISCARD : 0);
1706 return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0;
1709 /* Used to send write requests
1710 * R_PRIMARY -> Peer (P_DATA)
1712 int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req)
1714 struct drbd_socket *sock;
1716 unsigned int dp_flags = 0;
1720 dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_tfm) ?
1721 crypto_hash_digestsize(mdev->tconn->integrity_tfm) : 0;
1723 sock = &mdev->tconn->data;
1724 p = drbd_prepare_command(mdev, sock);
1727 p->sector = cpu_to_be64(req->i.sector);
1728 p->block_id = (unsigned long)req;
1729 p->seq_num = cpu_to_be32(req->seq_num = atomic_inc_return(&mdev->packet_seq));
1730 dp_flags = bio_flags_to_wire(mdev, req->master_bio->bi_rw);
1731 if (mdev->state.conn >= C_SYNC_SOURCE &&
1732 mdev->state.conn <= C_PAUSED_SYNC_T)
1733 dp_flags |= DP_MAY_SET_IN_SYNC;
1734 if (mdev->tconn->agreed_pro_version >= 100) {
1735 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1736 dp_flags |= DP_SEND_RECEIVE_ACK;
1737 if (req->rq_state & RQ_EXP_WRITE_ACK)
1738 dp_flags |= DP_SEND_WRITE_ACK;
1740 p->dp_flags = cpu_to_be32(dp_flags);
1742 drbd_csum_bio(mdev, mdev->tconn->integrity_tfm, req->master_bio, p + 1);
1743 err = __send_command(mdev->tconn, mdev->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size);
1745 /* For protocol A, we have to memcpy the payload into
1746 * socket buffers, as we may complete right away
1747 * as soon as we handed it over to tcp, at which point the data
1748 * pages may become invalid.
1750 * For data-integrity enabled, we copy it as well, so we can be
1751 * sure that even if the bio pages may still be modified, it
1752 * won't change the data on the wire, thus if the digest checks
1753 * out ok after sending on this side, but does not fit on the
1754 * receiving side, we sure have detected corruption elsewhere.
1756 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs)
1757 err = _drbd_send_bio(mdev, req->master_bio);
1759 err = _drbd_send_zc_bio(mdev, req->master_bio);
1761 /* double check digest, sometimes buffers have been modified in flight. */
1762 if (dgs > 0 && dgs <= 64) {
1763 /* 64 byte, 512 bit, is the largest digest size
1764 * currently supported in kernel crypto. */
1765 unsigned char digest[64];
1766 drbd_csum_bio(mdev, mdev->tconn->integrity_tfm, req->master_bio, digest);
1767 if (memcmp(p + 1, digest, dgs)) {
1769 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1770 (unsigned long long)req->i.sector, req->i.size);
1772 } /* else if (dgs > 64) {
1773 ... Be noisy about digest too large ...
1776 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1781 /* answer packet, used to send data back for read requests:
1782 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1783 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1785 int drbd_send_block(struct drbd_conf *mdev, enum drbd_packet cmd,
1786 struct drbd_peer_request *peer_req)
1788 struct drbd_socket *sock;
1793 dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_tfm) ?
1794 crypto_hash_digestsize(mdev->tconn->integrity_tfm) : 0;
1796 sock = &mdev->tconn->data;
1797 p = drbd_prepare_command(mdev, sock);
1800 p->sector = cpu_to_be64(peer_req->i.sector);
1801 p->block_id = peer_req->block_id;
1802 p->seq_num = 0; /* unused */
1804 drbd_csum_ee(mdev, mdev->tconn->integrity_tfm, peer_req, p + 1);
1805 err = __send_command(mdev->tconn, mdev->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size);
1807 err = _drbd_send_zc_ee(mdev, peer_req);
1808 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1813 int drbd_send_out_of_sync(struct drbd_conf *mdev, struct drbd_request *req)
1815 struct drbd_socket *sock;
1816 struct p_block_desc *p;
1818 sock = &mdev->tconn->data;
1819 p = drbd_prepare_command(mdev, sock);
1822 p->sector = cpu_to_be64(req->i.sector);
1823 p->blksize = cpu_to_be32(req->i.size);
1824 return drbd_send_command(mdev, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1828 drbd_send distinguishes two cases:
1830 Packets sent via the data socket "sock"
1831 and packets sent via the meta data socket "msock"
1834 -----------------+-------------------------+------------------------------
1835 timeout conf.timeout / 2 conf.timeout / 2
1836 timeout action send a ping via msock Abort communication
1837 and close all sockets
1841 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1843 int drbd_send(struct drbd_tconn *tconn, struct socket *sock,
1844 void *buf, size_t size, unsigned msg_flags)
1853 /* THINK if (signal_pending) return ... ? */
1858 msg.msg_name = NULL;
1859 msg.msg_namelen = 0;
1860 msg.msg_control = NULL;
1861 msg.msg_controllen = 0;
1862 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1864 if (sock == tconn->data.socket) {
1866 tconn->ko_count = rcu_dereference(tconn->net_conf)->ko_count;
1868 drbd_update_congested(tconn);
1872 * tcp_sendmsg does _not_ use its size parameter at all ?
1874 * -EAGAIN on timeout, -EINTR on signal.
1877 * do we need to block DRBD_SIG if sock == &meta.socket ??
1878 * otherwise wake_asender() might interrupt some send_*Ack !
1880 rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
1881 if (rv == -EAGAIN) {
1882 if (we_should_drop_the_connection(tconn, sock))
1888 flush_signals(current);
1896 } while (sent < size);
1898 if (sock == tconn->data.socket)
1899 clear_bit(NET_CONGESTED, &tconn->flags);
1902 if (rv != -EAGAIN) {
1903 conn_err(tconn, "%s_sendmsg returned %d\n",
1904 sock == tconn->meta.socket ? "msock" : "sock",
1906 conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD);
1908 conn_request_state(tconn, NS(conn, C_TIMEOUT), CS_HARD);
1915 * drbd_send_all - Send an entire buffer
1917 * Returns 0 upon success and a negative error value otherwise.
1919 int drbd_send_all(struct drbd_tconn *tconn, struct socket *sock, void *buffer,
1920 size_t size, unsigned msg_flags)
1924 err = drbd_send(tconn, sock, buffer, size, msg_flags);
1932 static int drbd_open(struct block_device *bdev, fmode_t mode)
1934 struct drbd_conf *mdev = bdev->bd_disk->private_data;
1935 unsigned long flags;
1938 mutex_lock(&drbd_main_mutex);
1939 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
1940 /* to have a stable mdev->state.role
1941 * and no race with updating open_cnt */
1943 if (mdev->state.role != R_PRIMARY) {
1944 if (mode & FMODE_WRITE)
1946 else if (!allow_oos)
1952 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
1953 mutex_unlock(&drbd_main_mutex);
1958 static int drbd_release(struct gendisk *gd, fmode_t mode)
1960 struct drbd_conf *mdev = gd->private_data;
1961 mutex_lock(&drbd_main_mutex);
1963 mutex_unlock(&drbd_main_mutex);
1967 static void drbd_set_defaults(struct drbd_conf *mdev)
1969 /* Beware! The actual layout differs
1970 * between big endian and little endian */
1971 mdev->state = (union drbd_dev_state) {
1972 { .role = R_SECONDARY,
1974 .conn = C_STANDALONE,
1980 void drbd_init_set_defaults(struct drbd_conf *mdev)
1982 /* the memset(,0,) did most of this.
1983 * note: only assignments, no allocation in here */
1985 drbd_set_defaults(mdev);
1987 atomic_set(&mdev->ap_bio_cnt, 0);
1988 atomic_set(&mdev->ap_pending_cnt, 0);
1989 atomic_set(&mdev->rs_pending_cnt, 0);
1990 atomic_set(&mdev->unacked_cnt, 0);
1991 atomic_set(&mdev->local_cnt, 0);
1992 atomic_set(&mdev->pp_in_use_by_net, 0);
1993 atomic_set(&mdev->rs_sect_in, 0);
1994 atomic_set(&mdev->rs_sect_ev, 0);
1995 atomic_set(&mdev->ap_in_flight, 0);
1997 mutex_init(&mdev->md_io_mutex);
1998 mutex_init(&mdev->own_state_mutex);
1999 mdev->state_mutex = &mdev->own_state_mutex;
2001 spin_lock_init(&mdev->al_lock);
2002 spin_lock_init(&mdev->peer_seq_lock);
2003 spin_lock_init(&mdev->epoch_lock);
2005 INIT_LIST_HEAD(&mdev->active_ee);
2006 INIT_LIST_HEAD(&mdev->sync_ee);
2007 INIT_LIST_HEAD(&mdev->done_ee);
2008 INIT_LIST_HEAD(&mdev->read_ee);
2009 INIT_LIST_HEAD(&mdev->net_ee);
2010 INIT_LIST_HEAD(&mdev->resync_reads);
2011 INIT_LIST_HEAD(&mdev->resync_work.list);
2012 INIT_LIST_HEAD(&mdev->unplug_work.list);
2013 INIT_LIST_HEAD(&mdev->go_diskless.list);
2014 INIT_LIST_HEAD(&mdev->md_sync_work.list);
2015 INIT_LIST_HEAD(&mdev->start_resync_work.list);
2016 INIT_LIST_HEAD(&mdev->bm_io_work.w.list);
2018 mdev->resync_work.cb = w_resync_timer;
2019 mdev->unplug_work.cb = w_send_write_hint;
2020 mdev->go_diskless.cb = w_go_diskless;
2021 mdev->md_sync_work.cb = w_md_sync;
2022 mdev->bm_io_work.w.cb = w_bitmap_io;
2023 mdev->start_resync_work.cb = w_start_resync;
2025 mdev->resync_work.mdev = mdev;
2026 mdev->unplug_work.mdev = mdev;
2027 mdev->go_diskless.mdev = mdev;
2028 mdev->md_sync_work.mdev = mdev;
2029 mdev->bm_io_work.w.mdev = mdev;
2030 mdev->start_resync_work.mdev = mdev;
2032 init_timer(&mdev->resync_timer);
2033 init_timer(&mdev->md_sync_timer);
2034 init_timer(&mdev->start_resync_timer);
2035 init_timer(&mdev->request_timer);
2036 mdev->resync_timer.function = resync_timer_fn;
2037 mdev->resync_timer.data = (unsigned long) mdev;
2038 mdev->md_sync_timer.function = md_sync_timer_fn;
2039 mdev->md_sync_timer.data = (unsigned long) mdev;
2040 mdev->start_resync_timer.function = start_resync_timer_fn;
2041 mdev->start_resync_timer.data = (unsigned long) mdev;
2042 mdev->request_timer.function = request_timer_fn;
2043 mdev->request_timer.data = (unsigned long) mdev;
2045 init_waitqueue_head(&mdev->misc_wait);
2046 init_waitqueue_head(&mdev->state_wait);
2047 init_waitqueue_head(&mdev->ee_wait);
2048 init_waitqueue_head(&mdev->al_wait);
2049 init_waitqueue_head(&mdev->seq_wait);
2051 /* mdev->tconn->agreed_pro_version gets initialized in drbd_connect() */
2052 mdev->write_ordering = WO_bdev_flush;
2053 mdev->resync_wenr = LC_FREE;
2054 mdev->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2055 mdev->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2058 void drbd_mdev_cleanup(struct drbd_conf *mdev)
2061 if (mdev->tconn->receiver.t_state != NONE)
2062 dev_err(DEV, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2063 mdev->tconn->receiver.t_state);
2065 /* no need to lock it, I'm the only thread alive */
2066 if (atomic_read(&mdev->current_epoch->epoch_size) != 0)
2067 dev_err(DEV, "epoch_size:%d\n", atomic_read(&mdev->current_epoch->epoch_size));
2077 mdev->rs_failed = 0;
2078 mdev->rs_last_events = 0;
2079 mdev->rs_last_sect_ev = 0;
2080 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2081 mdev->rs_mark_left[i] = 0;
2082 mdev->rs_mark_time[i] = 0;
2084 D_ASSERT(mdev->tconn->net_conf == NULL);
2086 drbd_set_my_capacity(mdev, 0);
2088 /* maybe never allocated. */
2089 drbd_bm_resize(mdev, 0, 1);
2090 drbd_bm_cleanup(mdev);
2093 drbd_free_bc(mdev->ldev);
2096 clear_bit(AL_SUSPENDED, &mdev->flags);
2098 D_ASSERT(list_empty(&mdev->active_ee));
2099 D_ASSERT(list_empty(&mdev->sync_ee));
2100 D_ASSERT(list_empty(&mdev->done_ee));
2101 D_ASSERT(list_empty(&mdev->read_ee));
2102 D_ASSERT(list_empty(&mdev->net_ee));
2103 D_ASSERT(list_empty(&mdev->resync_reads));
2104 D_ASSERT(list_empty(&mdev->tconn->data.work.q));
2105 D_ASSERT(list_empty(&mdev->tconn->meta.work.q));
2106 D_ASSERT(list_empty(&mdev->resync_work.list));
2107 D_ASSERT(list_empty(&mdev->unplug_work.list));
2108 D_ASSERT(list_empty(&mdev->go_diskless.list));
2110 drbd_set_defaults(mdev);
2114 static void drbd_destroy_mempools(void)
2118 while (drbd_pp_pool) {
2119 page = drbd_pp_pool;
2120 drbd_pp_pool = (struct page *)page_private(page);
2125 /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */
2127 if (drbd_md_io_bio_set)
2128 bioset_free(drbd_md_io_bio_set);
2129 if (drbd_md_io_page_pool)
2130 mempool_destroy(drbd_md_io_page_pool);
2131 if (drbd_ee_mempool)
2132 mempool_destroy(drbd_ee_mempool);
2133 if (drbd_request_mempool)
2134 mempool_destroy(drbd_request_mempool);
2136 kmem_cache_destroy(drbd_ee_cache);
2137 if (drbd_request_cache)
2138 kmem_cache_destroy(drbd_request_cache);
2139 if (drbd_bm_ext_cache)
2140 kmem_cache_destroy(drbd_bm_ext_cache);
2141 if (drbd_al_ext_cache)
2142 kmem_cache_destroy(drbd_al_ext_cache);
2144 drbd_md_io_bio_set = NULL;
2145 drbd_md_io_page_pool = NULL;
2146 drbd_ee_mempool = NULL;
2147 drbd_request_mempool = NULL;
2148 drbd_ee_cache = NULL;
2149 drbd_request_cache = NULL;
2150 drbd_bm_ext_cache = NULL;
2151 drbd_al_ext_cache = NULL;
2156 static int drbd_create_mempools(void)
2159 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
2162 /* prepare our caches and mempools */
2163 drbd_request_mempool = NULL;
2164 drbd_ee_cache = NULL;
2165 drbd_request_cache = NULL;
2166 drbd_bm_ext_cache = NULL;
2167 drbd_al_ext_cache = NULL;
2168 drbd_pp_pool = NULL;
2169 drbd_md_io_page_pool = NULL;
2170 drbd_md_io_bio_set = NULL;
2173 drbd_request_cache = kmem_cache_create(
2174 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2175 if (drbd_request_cache == NULL)
2178 drbd_ee_cache = kmem_cache_create(
2179 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2180 if (drbd_ee_cache == NULL)
2183 drbd_bm_ext_cache = kmem_cache_create(
2184 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2185 if (drbd_bm_ext_cache == NULL)
2188 drbd_al_ext_cache = kmem_cache_create(
2189 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2190 if (drbd_al_ext_cache == NULL)
2194 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2195 if (drbd_md_io_bio_set == NULL)
2198 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2199 if (drbd_md_io_page_pool == NULL)
2202 drbd_request_mempool = mempool_create(number,
2203 mempool_alloc_slab, mempool_free_slab, drbd_request_cache);
2204 if (drbd_request_mempool == NULL)
2207 drbd_ee_mempool = mempool_create(number,
2208 mempool_alloc_slab, mempool_free_slab, drbd_ee_cache);
2209 if (drbd_ee_mempool == NULL)
2212 /* drbd's page pool */
2213 spin_lock_init(&drbd_pp_lock);
2215 for (i = 0; i < number; i++) {
2216 page = alloc_page(GFP_HIGHUSER);
2219 set_page_private(page, (unsigned long)drbd_pp_pool);
2220 drbd_pp_pool = page;
2222 drbd_pp_vacant = number;
2227 drbd_destroy_mempools(); /* in case we allocated some */
2231 static int drbd_notify_sys(struct notifier_block *this, unsigned long code,
2234 /* just so we have it. you never know what interesting things we
2235 * might want to do here some day...
2241 static struct notifier_block drbd_notifier = {
2242 .notifier_call = drbd_notify_sys,
2245 static void drbd_release_all_peer_reqs(struct drbd_conf *mdev)
2249 rr = drbd_free_peer_reqs(mdev, &mdev->active_ee);
2251 dev_err(DEV, "%d EEs in active list found!\n", rr);
2253 rr = drbd_free_peer_reqs(mdev, &mdev->sync_ee);
2255 dev_err(DEV, "%d EEs in sync list found!\n", rr);
2257 rr = drbd_free_peer_reqs(mdev, &mdev->read_ee);
2259 dev_err(DEV, "%d EEs in read list found!\n", rr);
2261 rr = drbd_free_peer_reqs(mdev, &mdev->done_ee);
2263 dev_err(DEV, "%d EEs in done list found!\n", rr);
2265 rr = drbd_free_peer_reqs(mdev, &mdev->net_ee);
2267 dev_err(DEV, "%d EEs in net list found!\n", rr);
2270 /* caution. no locking. */
2271 void drbd_delete_device(struct drbd_conf *mdev)
2273 struct drbd_tconn *tconn = mdev->tconn;
2275 idr_remove(&mdev->tconn->volumes, mdev->vnr);
2276 idr_remove(&minors, mdev_to_minor(mdev));
2279 /* paranoia asserts */
2280 D_ASSERT(mdev->open_cnt == 0);
2281 D_ASSERT(list_empty(&mdev->tconn->data.work.q));
2282 /* end paranoia asserts */
2284 del_gendisk(mdev->vdisk);
2286 /* cleanup stuff that may have been allocated during
2287 * device (re-)configuration or state changes */
2289 if (mdev->this_bdev)
2290 bdput(mdev->this_bdev);
2292 drbd_free_bc(mdev->ldev);
2295 drbd_release_all_peer_reqs(mdev);
2297 lc_destroy(mdev->act_log);
2298 lc_destroy(mdev->resync);
2300 kfree(mdev->p_uuid);
2301 /* mdev->p_uuid = NULL; */
2303 kfree(mdev->current_epoch);
2304 if (mdev->bitmap) /* should no longer be there. */
2305 drbd_bm_cleanup(mdev);
2306 __free_page(mdev->md_io_page);
2307 put_disk(mdev->vdisk);
2308 blk_cleanup_queue(mdev->rq_queue);
2311 kref_put(&tconn->kref, &conn_destroy);
2314 static void drbd_cleanup(void)
2317 struct drbd_conf *mdev;
2319 unregister_reboot_notifier(&drbd_notifier);
2321 /* first remove proc,
2322 * drbdsetup uses it's presence to detect
2323 * whether DRBD is loaded.
2324 * If we would get stuck in proc removal,
2325 * but have netlink already deregistered,
2326 * some drbdsetup commands may wait forever
2330 remove_proc_entry("drbd", NULL);
2332 drbd_genl_unregister();
2334 down_write(&drbd_cfg_rwsem);
2335 idr_for_each_entry(&minors, mdev, i)
2336 drbd_delete_device(mdev);
2337 up_write(&drbd_cfg_rwsem);
2339 drbd_destroy_mempools();
2340 unregister_blkdev(DRBD_MAJOR, "drbd");
2342 idr_destroy(&minors);
2344 printk(KERN_INFO "drbd: module cleanup done.\n");
2348 * drbd_congested() - Callback for pdflush
2349 * @congested_data: User data
2350 * @bdi_bits: Bits pdflush is currently interested in
2352 * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2354 static int drbd_congested(void *congested_data, int bdi_bits)
2356 struct drbd_conf *mdev = congested_data;
2357 struct request_queue *q;
2361 if (!may_inc_ap_bio(mdev)) {
2362 /* DRBD has frozen IO */
2368 if (get_ldev(mdev)) {
2369 q = bdev_get_queue(mdev->ldev->backing_bdev);
2370 r = bdi_congested(&q->backing_dev_info, bdi_bits);
2376 if (bdi_bits & (1 << BDI_async_congested) && test_bit(NET_CONGESTED, &mdev->tconn->flags)) {
2377 r |= (1 << BDI_async_congested);
2378 reason = reason == 'b' ? 'a' : 'n';
2382 mdev->congestion_reason = reason;
2386 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2388 sema_init(&wq->s, 0);
2389 spin_lock_init(&wq->q_lock);
2390 INIT_LIST_HEAD(&wq->q);
2393 struct drbd_tconn *conn_get_by_name(const char *name)
2395 struct drbd_tconn *tconn;
2397 if (!name || !name[0])
2400 down_read(&drbd_cfg_rwsem);
2401 list_for_each_entry(tconn, &drbd_tconns, all_tconn) {
2402 if (!strcmp(tconn->name, name)) {
2403 kref_get(&tconn->kref);
2409 up_read(&drbd_cfg_rwsem);
2413 static int drbd_alloc_socket(struct drbd_socket *socket)
2415 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2418 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2424 static void drbd_free_socket(struct drbd_socket *socket)
2426 free_page((unsigned long) socket->sbuf);
2427 free_page((unsigned long) socket->rbuf);
2430 void conn_free_crypto(struct drbd_tconn *tconn)
2432 drbd_free_sock(tconn);
2434 crypto_free_hash(tconn->csums_tfm);
2435 crypto_free_hash(tconn->verify_tfm);
2436 crypto_free_hash(tconn->cram_hmac_tfm);
2437 crypto_free_hash(tconn->integrity_tfm);
2438 crypto_free_hash(tconn->peer_integrity_tfm);
2439 kfree(tconn->int_dig_in);
2440 kfree(tconn->int_dig_vv);
2442 tconn->csums_tfm = NULL;
2443 tconn->verify_tfm = NULL;
2444 tconn->cram_hmac_tfm = NULL;
2445 tconn->integrity_tfm = NULL;
2446 tconn->peer_integrity_tfm = NULL;
2447 tconn->int_dig_in = NULL;
2448 tconn->int_dig_vv = NULL;
2451 struct drbd_tconn *conn_create(const char *name)
2453 struct drbd_tconn *tconn;
2455 tconn = kzalloc(sizeof(struct drbd_tconn), GFP_KERNEL);
2459 tconn->name = kstrdup(name, GFP_KERNEL);
2463 if (drbd_alloc_socket(&tconn->data))
2465 if (drbd_alloc_socket(&tconn->meta))
2468 if (!zalloc_cpumask_var(&tconn->cpu_mask, GFP_KERNEL))
2471 if (!tl_init(tconn))
2474 tconn->cstate = C_STANDALONE;
2475 mutex_init(&tconn->cstate_mutex);
2476 spin_lock_init(&tconn->req_lock);
2477 mutex_init(&tconn->net_conf_update);
2478 init_waitqueue_head(&tconn->ping_wait);
2479 idr_init(&tconn->volumes);
2481 drbd_init_workqueue(&tconn->data.work);
2482 mutex_init(&tconn->data.mutex);
2484 drbd_init_workqueue(&tconn->meta.work);
2485 mutex_init(&tconn->meta.mutex);
2487 drbd_thread_init(tconn, &tconn->receiver, drbdd_init, "receiver");
2488 drbd_thread_init(tconn, &tconn->worker, drbd_worker, "worker");
2489 drbd_thread_init(tconn, &tconn->asender, drbd_asender, "asender");
2491 drbd_set_res_opts_default(&tconn->res_opts);
2493 down_write(&drbd_cfg_rwsem);
2494 kref_init(&tconn->kref);
2495 list_add_tail(&tconn->all_tconn, &drbd_tconns);
2496 up_write(&drbd_cfg_rwsem);
2502 free_cpumask_var(tconn->cpu_mask);
2503 drbd_free_socket(&tconn->meta);
2504 drbd_free_socket(&tconn->data);
2511 void conn_destroy(struct kref *kref)
2513 struct drbd_tconn *tconn = container_of(kref, struct drbd_tconn, kref);
2515 idr_destroy(&tconn->volumes);
2517 free_cpumask_var(tconn->cpu_mask);
2518 drbd_free_socket(&tconn->meta);
2519 drbd_free_socket(&tconn->data);
2521 kfree(tconn->int_dig_in);
2522 kfree(tconn->int_dig_vv);
2526 enum drbd_ret_code conn_new_minor(struct drbd_tconn *tconn, unsigned int minor, int vnr)
2528 struct drbd_conf *mdev;
2529 struct gendisk *disk;
2530 struct request_queue *q;
2532 int minor_got = minor;
2533 enum drbd_ret_code err = ERR_NOMEM;
2535 mdev = minor_to_mdev(minor);
2537 return ERR_MINOR_EXISTS;
2539 /* GFP_KERNEL, we are outside of all write-out paths */
2540 mdev = kzalloc(sizeof(struct drbd_conf), GFP_KERNEL);
2544 kref_get(&tconn->kref);
2545 mdev->tconn = tconn;
2547 mdev->minor = minor;
2550 drbd_init_set_defaults(mdev);
2552 q = blk_alloc_queue(GFP_KERNEL);
2556 q->queuedata = mdev;
2558 disk = alloc_disk(1);
2563 set_disk_ro(disk, true);
2566 disk->major = DRBD_MAJOR;
2567 disk->first_minor = minor;
2568 disk->fops = &drbd_ops;
2569 sprintf(disk->disk_name, "drbd%d", minor);
2570 disk->private_data = mdev;
2572 mdev->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2573 /* we have no partitions. we contain only ourselves. */
2574 mdev->this_bdev->bd_contains = mdev->this_bdev;
2576 q->backing_dev_info.congested_fn = drbd_congested;
2577 q->backing_dev_info.congested_data = mdev;
2579 blk_queue_make_request(q, drbd_make_request);
2580 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2581 This triggers a max_bio_size message upon first attach or connect */
2582 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2583 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2584 blk_queue_merge_bvec(q, drbd_merge_bvec);
2585 q->queue_lock = &mdev->tconn->req_lock; /* needed since we use */
2587 mdev->md_io_page = alloc_page(GFP_KERNEL);
2588 if (!mdev->md_io_page)
2589 goto out_no_io_page;
2591 if (drbd_bm_init(mdev))
2593 mdev->read_requests = RB_ROOT;
2594 mdev->write_requests = RB_ROOT;
2596 mdev->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2597 if (!mdev->current_epoch)
2600 INIT_LIST_HEAD(&mdev->current_epoch->list);
2603 if (!idr_pre_get(&minors, GFP_KERNEL))
2604 goto out_no_minor_idr;
2605 if (idr_get_new_above(&minors, mdev, minor, &minor_got))
2606 goto out_no_minor_idr;
2607 if (minor_got != minor) {
2608 err = ERR_MINOR_EXISTS;
2609 drbd_msg_put_info("requested minor exists already");
2610 goto out_idr_remove_minor;
2613 if (!idr_pre_get(&tconn->volumes, GFP_KERNEL))
2614 goto out_idr_remove_minor;
2615 if (idr_get_new_above(&tconn->volumes, mdev, vnr, &vnr_got))
2616 goto out_idr_remove_minor;
2617 if (vnr_got != vnr) {
2618 err = ERR_INVALID_REQUEST;
2619 drbd_msg_put_info("requested volume exists already");
2620 goto out_idr_remove_vol;
2624 /* inherit the connection state */
2625 mdev->state.conn = tconn->cstate;
2626 if (mdev->state.conn == C_WF_REPORT_PARAMS)
2627 drbd_connected(vnr, mdev, tconn);
2632 idr_remove(&tconn->volumes, vnr_got);
2633 out_idr_remove_minor:
2634 idr_remove(&minors, minor_got);
2637 kfree(mdev->current_epoch);
2639 drbd_bm_cleanup(mdev);
2641 __free_page(mdev->md_io_page);
2645 blk_cleanup_queue(q);
2648 kref_put(&tconn->kref, &conn_destroy);
2652 int __init drbd_init(void)
2656 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2658 "drbd: invalid minor_count (%d)\n", minor_count);
2666 err = register_blkdev(DRBD_MAJOR, "drbd");
2669 "drbd: unable to register block device major %d\n",
2674 err = drbd_genl_register();
2676 printk(KERN_ERR "drbd: unable to register generic netlink family\n");
2681 register_reboot_notifier(&drbd_notifier);
2684 * allocate all necessary structs
2688 init_waitqueue_head(&drbd_pp_wait);
2690 drbd_proc = NULL; /* play safe for drbd_cleanup */
2693 err = drbd_create_mempools();
2697 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
2699 printk(KERN_ERR "drbd: unable to register proc file\n");
2703 rwlock_init(&global_state_lock);
2704 INIT_LIST_HEAD(&drbd_tconns);
2706 printk(KERN_INFO "drbd: initialized. "
2707 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2708 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2709 printk(KERN_INFO "drbd: %s\n", drbd_buildtag());
2710 printk(KERN_INFO "drbd: registered as block device major %d\n",
2713 return 0; /* Success! */
2718 /* currently always the case */
2719 printk(KERN_ERR "drbd: ran out of memory\n");
2721 printk(KERN_ERR "drbd: initialization failure\n");
2725 void drbd_free_bc(struct drbd_backing_dev *ldev)
2730 blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2731 blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2736 void drbd_free_sock(struct drbd_tconn *tconn)
2738 if (tconn->data.socket) {
2739 mutex_lock(&tconn->data.mutex);
2740 kernel_sock_shutdown(tconn->data.socket, SHUT_RDWR);
2741 sock_release(tconn->data.socket);
2742 tconn->data.socket = NULL;
2743 mutex_unlock(&tconn->data.mutex);
2745 if (tconn->meta.socket) {
2746 mutex_lock(&tconn->meta.mutex);
2747 kernel_sock_shutdown(tconn->meta.socket, SHUT_RDWR);
2748 sock_release(tconn->meta.socket);
2749 tconn->meta.socket = NULL;
2750 mutex_unlock(&tconn->meta.mutex);
2754 /* meta data management */
2756 struct meta_data_on_disk {
2757 u64 la_size; /* last agreed size. */
2758 u64 uuid[UI_SIZE]; /* UUIDs. */
2761 u32 flags; /* MDF */
2764 u32 al_offset; /* offset to this block */
2765 u32 al_nr_extents; /* important for restoring the AL */
2766 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2767 u32 bm_offset; /* offset to the bitmap, from here */
2768 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
2769 u32 la_peer_max_bio_size; /* last peer max_bio_size */
2770 u32 reserved_u32[3];
2775 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
2776 * @mdev: DRBD device.
2778 void drbd_md_sync(struct drbd_conf *mdev)
2780 struct meta_data_on_disk *buffer;
2784 del_timer(&mdev->md_sync_timer);
2785 /* timer may be rearmed by drbd_md_mark_dirty() now. */
2786 if (!test_and_clear_bit(MD_DIRTY, &mdev->flags))
2789 /* We use here D_FAILED and not D_ATTACHING because we try to write
2790 * metadata even if we detach due to a disk failure! */
2791 if (!get_ldev_if_state(mdev, D_FAILED))
2794 mutex_lock(&mdev->md_io_mutex);
2795 buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page);
2796 memset(buffer, 0, 512);
2798 buffer->la_size = cpu_to_be64(drbd_get_capacity(mdev->this_bdev));
2799 for (i = UI_CURRENT; i < UI_SIZE; i++)
2800 buffer->uuid[i] = cpu_to_be64(mdev->ldev->md.uuid[i]);
2801 buffer->flags = cpu_to_be32(mdev->ldev->md.flags);
2802 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC);
2804 buffer->md_size_sect = cpu_to_be32(mdev->ldev->md.md_size_sect);
2805 buffer->al_offset = cpu_to_be32(mdev->ldev->md.al_offset);
2806 buffer->al_nr_extents = cpu_to_be32(mdev->act_log->nr_elements);
2807 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
2808 buffer->device_uuid = cpu_to_be64(mdev->ldev->md.device_uuid);
2810 buffer->bm_offset = cpu_to_be32(mdev->ldev->md.bm_offset);
2811 buffer->la_peer_max_bio_size = cpu_to_be32(mdev->peer_max_bio_size);
2813 D_ASSERT(drbd_md_ss__(mdev, mdev->ldev) == mdev->ldev->md.md_offset);
2814 sector = mdev->ldev->md.md_offset;
2816 if (drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) {
2817 /* this was a try anyways ... */
2818 dev_err(DEV, "meta data update failed!\n");
2819 drbd_chk_io_error(mdev, 1, true);
2822 /* Update mdev->ldev->md.la_size_sect,
2823 * since we updated it on metadata. */
2824 mdev->ldev->md.la_size_sect = drbd_get_capacity(mdev->this_bdev);
2826 mutex_unlock(&mdev->md_io_mutex);
2831 * drbd_md_read() - Reads in the meta data super block
2832 * @mdev: DRBD device.
2833 * @bdev: Device from which the meta data should be read in.
2835 * Return 0 (NO_ERROR) on success, and an enum drbd_ret_code in case
2836 * something goes wrong. Currently only: ERR_IO_MD_DISK, ERR_MD_INVALID.
2838 int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
2840 struct meta_data_on_disk *buffer;
2841 int i, rv = NO_ERROR;
2843 if (!get_ldev_if_state(mdev, D_ATTACHING))
2844 return ERR_IO_MD_DISK;
2846 mutex_lock(&mdev->md_io_mutex);
2847 buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page);
2849 if (drbd_md_sync_page_io(mdev, bdev, bdev->md.md_offset, READ)) {
2850 /* NOTE: can't do normal error processing here as this is
2851 called BEFORE disk is attached */
2852 dev_err(DEV, "Error while reading metadata.\n");
2853 rv = ERR_IO_MD_DISK;
2857 if (buffer->magic != cpu_to_be32(DRBD_MD_MAGIC)) {
2858 dev_err(DEV, "Error while reading metadata, magic not found.\n");
2859 rv = ERR_MD_INVALID;
2862 if (be32_to_cpu(buffer->al_offset) != bdev->md.al_offset) {
2863 dev_err(DEV, "unexpected al_offset: %d (expected %d)\n",
2864 be32_to_cpu(buffer->al_offset), bdev->md.al_offset);
2865 rv = ERR_MD_INVALID;
2868 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
2869 dev_err(DEV, "unexpected bm_offset: %d (expected %d)\n",
2870 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
2871 rv = ERR_MD_INVALID;
2874 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
2875 dev_err(DEV, "unexpected md_size: %u (expected %u)\n",
2876 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
2877 rv = ERR_MD_INVALID;
2881 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
2882 dev_err(DEV, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
2883 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
2884 rv = ERR_MD_INVALID;
2888 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size);
2889 for (i = UI_CURRENT; i < UI_SIZE; i++)
2890 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
2891 bdev->md.flags = be32_to_cpu(buffer->flags);
2892 bdev->dc.al_extents = be32_to_cpu(buffer->al_nr_extents);
2893 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
2895 spin_lock_irq(&mdev->tconn->req_lock);
2896 if (mdev->state.conn < C_CONNECTED) {
2898 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
2899 peer = max_t(int, peer, DRBD_MAX_BIO_SIZE_SAFE);
2900 mdev->peer_max_bio_size = peer;
2902 spin_unlock_irq(&mdev->tconn->req_lock);
2904 if (bdev->dc.al_extents < 7)
2905 bdev->dc.al_extents = 127;
2908 mutex_unlock(&mdev->md_io_mutex);
2915 * drbd_md_mark_dirty() - Mark meta data super block as dirty
2916 * @mdev: DRBD device.
2918 * Call this function if you change anything that should be written to
2919 * the meta-data super block. This function sets MD_DIRTY, and starts a
2920 * timer that ensures that within five seconds you have to call drbd_md_sync().
2923 void drbd_md_mark_dirty_(struct drbd_conf *mdev, unsigned int line, const char *func)
2925 if (!test_and_set_bit(MD_DIRTY, &mdev->flags)) {
2926 mod_timer(&mdev->md_sync_timer, jiffies + HZ);
2927 mdev->last_md_mark_dirty.line = line;
2928 mdev->last_md_mark_dirty.func = func;
2932 void drbd_md_mark_dirty(struct drbd_conf *mdev)
2934 if (!test_and_set_bit(MD_DIRTY, &mdev->flags))
2935 mod_timer(&mdev->md_sync_timer, jiffies + 5*HZ);
2939 static void drbd_uuid_move_history(struct drbd_conf *mdev) __must_hold(local)
2943 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
2944 mdev->ldev->md.uuid[i+1] = mdev->ldev->md.uuid[i];
2947 void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
2949 if (idx == UI_CURRENT) {
2950 if (mdev->state.role == R_PRIMARY)
2955 drbd_set_ed_uuid(mdev, val);
2958 mdev->ldev->md.uuid[idx] = val;
2959 drbd_md_mark_dirty(mdev);
2963 void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
2965 if (mdev->ldev->md.uuid[idx]) {
2966 drbd_uuid_move_history(mdev);
2967 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[idx];
2969 _drbd_uuid_set(mdev, idx, val);
2973 * drbd_uuid_new_current() - Creates a new current UUID
2974 * @mdev: DRBD device.
2976 * Creates a new current UUID, and rotates the old current UUID into
2977 * the bitmap slot. Causes an incremental resync upon next connect.
2979 void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local)
2982 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
2985 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
2987 mdev->ldev->md.uuid[UI_BITMAP] = mdev->ldev->md.uuid[UI_CURRENT];
2989 get_random_bytes(&val, sizeof(u64));
2990 _drbd_uuid_set(mdev, UI_CURRENT, val);
2991 drbd_print_uuids(mdev, "new current UUID");
2992 /* get it to stable storage _now_ */
2996 void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local)
2998 if (mdev->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3002 drbd_uuid_move_history(mdev);
3003 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP];
3004 mdev->ldev->md.uuid[UI_BITMAP] = 0;
3006 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
3008 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
3010 mdev->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3012 drbd_md_mark_dirty(mdev);
3016 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3017 * @mdev: DRBD device.
3019 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3021 int drbd_bmio_set_n_write(struct drbd_conf *mdev)
3025 if (get_ldev_if_state(mdev, D_ATTACHING)) {
3026 drbd_md_set_flag(mdev, MDF_FULL_SYNC);
3028 drbd_bm_set_all(mdev);
3030 rv = drbd_bm_write(mdev);
3033 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
3044 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3045 * @mdev: DRBD device.
3047 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3049 int drbd_bmio_clear_n_write(struct drbd_conf *mdev)
3053 drbd_resume_al(mdev);
3054 if (get_ldev_if_state(mdev, D_ATTACHING)) {
3055 drbd_bm_clear_all(mdev);
3056 rv = drbd_bm_write(mdev);
3063 static int w_bitmap_io(struct drbd_work *w, int unused)
3065 struct bm_io_work *work = container_of(w, struct bm_io_work, w);
3066 struct drbd_conf *mdev = w->mdev;
3069 D_ASSERT(atomic_read(&mdev->ap_bio_cnt) == 0);
3071 if (get_ldev(mdev)) {
3072 drbd_bm_lock(mdev, work->why, work->flags);
3073 rv = work->io_fn(mdev);
3074 drbd_bm_unlock(mdev);
3078 clear_bit_unlock(BITMAP_IO, &mdev->flags);
3079 wake_up(&mdev->misc_wait);
3082 work->done(mdev, rv);
3084 clear_bit(BITMAP_IO_QUEUED, &mdev->flags);
3091 void drbd_ldev_destroy(struct drbd_conf *mdev)
3093 lc_destroy(mdev->resync);
3094 mdev->resync = NULL;
3095 lc_destroy(mdev->act_log);
3096 mdev->act_log = NULL;
3098 drbd_free_bc(mdev->ldev);
3099 mdev->ldev = NULL;);
3101 clear_bit(GO_DISKLESS, &mdev->flags);
3104 static int w_go_diskless(struct drbd_work *w, int unused)
3106 struct drbd_conf *mdev = w->mdev;
3108 D_ASSERT(mdev->state.disk == D_FAILED);
3109 /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will
3110 * inc/dec it frequently. Once we are D_DISKLESS, no one will touch
3111 * the protected members anymore, though, so once put_ldev reaches zero
3112 * again, it will be safe to free them. */
3113 drbd_force_state(mdev, NS(disk, D_DISKLESS));
3117 void drbd_go_diskless(struct drbd_conf *mdev)
3119 D_ASSERT(mdev->state.disk == D_FAILED);
3120 if (!test_and_set_bit(GO_DISKLESS, &mdev->flags))
3121 drbd_queue_work(&mdev->tconn->data.work, &mdev->go_diskless);
3125 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3126 * @mdev: DRBD device.
3127 * @io_fn: IO callback to be called when bitmap IO is possible
3128 * @done: callback to be called after the bitmap IO was performed
3129 * @why: Descriptive text of the reason for doing the IO
3131 * While IO on the bitmap happens we freeze application IO thus we ensure
3132 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3133 * called from worker context. It MUST NOT be used while a previous such
3134 * work is still pending!
3136 void drbd_queue_bitmap_io(struct drbd_conf *mdev,
3137 int (*io_fn)(struct drbd_conf *),
3138 void (*done)(struct drbd_conf *, int),
3139 char *why, enum bm_flag flags)
3141 D_ASSERT(current == mdev->tconn->worker.task);
3143 D_ASSERT(!test_bit(BITMAP_IO_QUEUED, &mdev->flags));
3144 D_ASSERT(!test_bit(BITMAP_IO, &mdev->flags));
3145 D_ASSERT(list_empty(&mdev->bm_io_work.w.list));
3146 if (mdev->bm_io_work.why)
3147 dev_err(DEV, "FIXME going to queue '%s' but '%s' still pending?\n",
3148 why, mdev->bm_io_work.why);
3150 mdev->bm_io_work.io_fn = io_fn;
3151 mdev->bm_io_work.done = done;
3152 mdev->bm_io_work.why = why;
3153 mdev->bm_io_work.flags = flags;
3155 spin_lock_irq(&mdev->tconn->req_lock);
3156 set_bit(BITMAP_IO, &mdev->flags);
3157 if (atomic_read(&mdev->ap_bio_cnt) == 0) {
3158 if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags))
3159 drbd_queue_work(&mdev->tconn->data.work, &mdev->bm_io_work.w);
3161 spin_unlock_irq(&mdev->tconn->req_lock);
3165 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3166 * @mdev: DRBD device.
3167 * @io_fn: IO callback to be called when bitmap IO is possible
3168 * @why: Descriptive text of the reason for doing the IO
3170 * freezes application IO while that the actual IO operations runs. This
3171 * functions MAY NOT be called from worker context.
3173 int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *),
3174 char *why, enum bm_flag flags)
3178 D_ASSERT(current != mdev->tconn->worker.task);
3180 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3181 drbd_suspend_io(mdev);
3183 drbd_bm_lock(mdev, why, flags);
3185 drbd_bm_unlock(mdev);
3187 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3188 drbd_resume_io(mdev);
3193 void drbd_md_set_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3195 if ((mdev->ldev->md.flags & flag) != flag) {
3196 drbd_md_mark_dirty(mdev);
3197 mdev->ldev->md.flags |= flag;
3201 void drbd_md_clear_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3203 if ((mdev->ldev->md.flags & flag) != 0) {
3204 drbd_md_mark_dirty(mdev);
3205 mdev->ldev->md.flags &= ~flag;
3208 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3210 return (bdev->md.flags & flag) != 0;
3213 static void md_sync_timer_fn(unsigned long data)
3215 struct drbd_conf *mdev = (struct drbd_conf *) data;
3217 drbd_queue_work_front(&mdev->tconn->data.work, &mdev->md_sync_work);
3220 static int w_md_sync(struct drbd_work *w, int unused)
3222 struct drbd_conf *mdev = w->mdev;
3224 dev_warn(DEV, "md_sync_timer expired! Worker calls drbd_md_sync().\n");
3226 dev_warn(DEV, "last md_mark_dirty: %s:%u\n",
3227 mdev->last_md_mark_dirty.func, mdev->last_md_mark_dirty.line);
3233 const char *cmdname(enum drbd_packet cmd)
3235 /* THINK may need to become several global tables
3236 * when we want to support more than
3237 * one PRO_VERSION */
3238 static const char *cmdnames[] = {
3240 [P_DATA_REPLY] = "DataReply",
3241 [P_RS_DATA_REPLY] = "RSDataReply",
3242 [P_BARRIER] = "Barrier",
3243 [P_BITMAP] = "ReportBitMap",
3244 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3245 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3246 [P_UNPLUG_REMOTE] = "UnplugRemote",
3247 [P_DATA_REQUEST] = "DataRequest",
3248 [P_RS_DATA_REQUEST] = "RSDataRequest",
3249 [P_SYNC_PARAM] = "SyncParam",
3250 [P_SYNC_PARAM89] = "SyncParam89",
3251 [P_PROTOCOL] = "ReportProtocol",
3252 [P_UUIDS] = "ReportUUIDs",
3253 [P_SIZES] = "ReportSizes",
3254 [P_STATE] = "ReportState",
3255 [P_SYNC_UUID] = "ReportSyncUUID",
3256 [P_AUTH_CHALLENGE] = "AuthChallenge",
3257 [P_AUTH_RESPONSE] = "AuthResponse",
3259 [P_PING_ACK] = "PingAck",
3260 [P_RECV_ACK] = "RecvAck",
3261 [P_WRITE_ACK] = "WriteAck",
3262 [P_RS_WRITE_ACK] = "RSWriteAck",
3263 [P_DISCARD_WRITE] = "DiscardWrite",
3264 [P_NEG_ACK] = "NegAck",
3265 [P_NEG_DREPLY] = "NegDReply",
3266 [P_NEG_RS_DREPLY] = "NegRSDReply",
3267 [P_BARRIER_ACK] = "BarrierAck",
3268 [P_STATE_CHG_REQ] = "StateChgRequest",
3269 [P_STATE_CHG_REPLY] = "StateChgReply",
3270 [P_OV_REQUEST] = "OVRequest",
3271 [P_OV_REPLY] = "OVReply",
3272 [P_OV_RESULT] = "OVResult",
3273 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3274 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3275 [P_COMPRESSED_BITMAP] = "CBitmap",
3276 [P_DELAY_PROBE] = "DelayProbe",
3277 [P_OUT_OF_SYNC] = "OutOfSync",
3278 [P_RETRY_WRITE] = "RetryWrite",
3279 [P_RS_CANCEL] = "RSCancel",
3280 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3281 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3283 /* enum drbd_packet, but not commands - obsoleted flags:
3289 /* too big for the array: 0xfffX */
3290 if (cmd == P_INITIAL_META)
3291 return "InitialMeta";
3292 if (cmd == P_INITIAL_DATA)
3293 return "InitialData";
3294 if (cmd == P_CONNECTION_FEATURES)
3295 return "ConnectionFeatures";
3296 if (cmd >= ARRAY_SIZE(cmdnames))
3298 return cmdnames[cmd];
3302 * drbd_wait_misc - wait for a request to make progress
3303 * @mdev: device associated with the request
3304 * @i: the struct drbd_interval embedded in struct drbd_request or
3305 * struct drbd_peer_request
3307 int drbd_wait_misc(struct drbd_conf *mdev, struct drbd_interval *i)
3309 struct net_conf *nc;
3314 nc = rcu_dereference(mdev->tconn->net_conf);
3319 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3322 /* Indicate to wake up mdev->misc_wait on progress. */
3324 prepare_to_wait(&mdev->misc_wait, &wait, TASK_INTERRUPTIBLE);
3325 spin_unlock_irq(&mdev->tconn->req_lock);
3326 timeout = schedule_timeout(timeout);
3327 finish_wait(&mdev->misc_wait, &wait);
3328 spin_lock_irq(&mdev->tconn->req_lock);
3329 if (!timeout || mdev->state.conn < C_CONNECTED)
3331 if (signal_pending(current))
3332 return -ERESTARTSYS;
3336 #ifdef CONFIG_DRBD_FAULT_INJECTION
3337 /* Fault insertion support including random number generator shamelessly
3338 * stolen from kernel/rcutorture.c */
3339 struct fault_random_state {
3340 unsigned long state;
3341 unsigned long count;
3344 #define FAULT_RANDOM_MULT 39916801 /* prime */
3345 #define FAULT_RANDOM_ADD 479001701 /* prime */
3346 #define FAULT_RANDOM_REFRESH 10000
3349 * Crude but fast random-number generator. Uses a linear congruential
3350 * generator, with occasional help from get_random_bytes().
3352 static unsigned long
3353 _drbd_fault_random(struct fault_random_state *rsp)
3357 if (!rsp->count--) {
3358 get_random_bytes(&refresh, sizeof(refresh));
3359 rsp->state += refresh;
3360 rsp->count = FAULT_RANDOM_REFRESH;
3362 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3363 return swahw32(rsp->state);
3367 _drbd_fault_str(unsigned int type) {
3368 static char *_faults[] = {
3369 [DRBD_FAULT_MD_WR] = "Meta-data write",
3370 [DRBD_FAULT_MD_RD] = "Meta-data read",
3371 [DRBD_FAULT_RS_WR] = "Resync write",
3372 [DRBD_FAULT_RS_RD] = "Resync read",
3373 [DRBD_FAULT_DT_WR] = "Data write",
3374 [DRBD_FAULT_DT_RD] = "Data read",
3375 [DRBD_FAULT_DT_RA] = "Data read ahead",
3376 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3377 [DRBD_FAULT_AL_EE] = "EE allocation",
3378 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3381 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3385 _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type)
3387 static struct fault_random_state rrs = {0, 0};
3389 unsigned int ret = (
3391 ((1 << mdev_to_minor(mdev)) & fault_devs) != 0) &&
3392 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3397 if (__ratelimit(&drbd_ratelimit_state))
3398 dev_warn(DEV, "***Simulating %s failure\n",
3399 _drbd_fault_str(type));
3406 const char *drbd_buildtag(void)
3408 /* DRBD built from external sources has here a reference to the
3409 git hash of the source code. */
3411 static char buildtag[38] = "\0uilt-in";
3413 if (buildtag[0] == 0) {
3414 #ifdef CONFIG_MODULES
3415 if (THIS_MODULE != NULL)
3416 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3425 module_init(drbd_init)
3426 module_exit(drbd_cleanup)
3428 EXPORT_SYMBOL(drbd_conn_str);
3429 EXPORT_SYMBOL(drbd_role_str);
3430 EXPORT_SYMBOL(drbd_disk_str);
3431 EXPORT_SYMBOL(drbd_set_st_err_str);