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 */
124 struct kmem_cache *drbd_request_cache;
125 struct kmem_cache *drbd_ee_cache; /* peer requests */
126 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
127 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
128 mempool_t *drbd_request_mempool;
129 mempool_t *drbd_ee_mempool;
130 mempool_t *drbd_md_io_page_pool;
131 struct bio_set *drbd_md_io_bio_set;
133 /* I do not use a standard mempool, because:
134 1) I want to hand out the pre-allocated objects first.
135 2) I want to be able to interrupt sleeping allocation with a signal.
136 Note: This is a single linked list, the next pointer is the private
137 member of struct page.
139 struct page *drbd_pp_pool;
140 spinlock_t drbd_pp_lock;
142 wait_queue_head_t drbd_pp_wait;
144 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
146 static const struct block_device_operations drbd_ops = {
147 .owner = THIS_MODULE,
149 .release = drbd_release,
152 static void bio_destructor_drbd(struct bio *bio)
154 bio_free(bio, drbd_md_io_bio_set);
157 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
161 if (!drbd_md_io_bio_set)
162 return bio_alloc(gfp_mask, 1);
164 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
167 bio->bi_destructor = bio_destructor_drbd;
172 /* When checking with sparse, and this is an inline function, sparse will
173 give tons of false positives. When this is a real functions sparse works.
175 int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins)
179 atomic_inc(&mdev->local_cnt);
180 io_allowed = (mdev->state.disk >= mins);
182 if (atomic_dec_and_test(&mdev->local_cnt))
183 wake_up(&mdev->misc_wait);
191 * DOC: The transfer log
193 * The transfer log is a single linked list of &struct drbd_tl_epoch objects.
194 * mdev->tconn->newest_tle points to the head, mdev->tconn->oldest_tle points to the tail
195 * of the list. There is always at least one &struct drbd_tl_epoch object.
197 * Each &struct drbd_tl_epoch has a circular double linked list of requests
200 static int tl_init(struct drbd_tconn *tconn)
202 struct drbd_tl_epoch *b;
204 /* during device minor initialization, we may well use GFP_KERNEL */
205 b = kmalloc(sizeof(struct drbd_tl_epoch), GFP_KERNEL);
208 INIT_LIST_HEAD(&b->requests);
209 INIT_LIST_HEAD(&b->w.list);
213 b->w.cb = NULL; /* if this is != NULL, we need to dec_ap_pending in tl_clear */
215 tconn->oldest_tle = b;
216 tconn->newest_tle = b;
217 INIT_LIST_HEAD(&tconn->out_of_sequence_requests);
218 INIT_LIST_HEAD(&tconn->barrier_acked_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_splice_init(&b->requests, &tconn->barrier_acked_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);
379 if (what == RESEND) {
380 b->n_writes = n_writes;
381 if (b->w.cb == NULL) {
382 b->w.cb = w_send_barrier;
383 inc_ap_pending(b->w.mdev);
384 set_bit(CREATE_BARRIER, &b->w.mdev->flags);
387 drbd_queue_work(&tconn->data.work, &b->w);
392 list_add(&carry_reads, &b->requests);
393 /* there could still be requests on that ring list,
394 * in case local io is still pending */
395 list_del(&b->requests);
397 /* dec_ap_pending corresponding to queue_barrier.
398 * the newest barrier may not have been queued yet,
399 * in which case w.cb is still NULL. */
401 dec_ap_pending(b->w.mdev);
403 if (b == tconn->newest_tle) {
404 /* recycle, but reinit! */
406 conn_err(tconn, "ASSERT FAILED tmp == NULL");
407 INIT_LIST_HEAD(&b->requests);
408 list_splice(&carry_reads, &b->requests);
409 INIT_LIST_HEAD(&b->w.list);
411 b->br_number = net_random();
421 list_splice(&carry_reads, &b->requests);
424 /* Actions operating on the disk state, also want to work on
425 requests that got barrier acked. */
427 case FAIL_FROZEN_DISK_IO:
428 case RESTART_FROZEN_DISK_IO:
429 list_for_each_safe(le, tle, &tconn->barrier_acked_requests) {
430 req = list_entry(le, struct drbd_request, tl_requests);
433 case CONNECTION_LOST_WHILE_PENDING:
437 conn_err(tconn, "what = %d in _tl_restart()\n", what);
442 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
443 * @mdev: DRBD device.
445 * This is called after the connection to the peer was lost. The storage covered
446 * by the requests on the transfer gets marked as our of sync. Called from the
447 * receiver thread and the worker thread.
449 void tl_clear(struct drbd_tconn *tconn)
451 struct drbd_conf *mdev;
452 struct list_head *le, *tle;
453 struct drbd_request *r;
456 spin_lock_irq(&tconn->req_lock);
458 _tl_restart(tconn, CONNECTION_LOST_WHILE_PENDING);
460 /* we expect this list to be empty. */
461 if (!list_empty(&tconn->out_of_sequence_requests))
462 conn_err(tconn, "ASSERT FAILED list_empty(&out_of_sequence_requests)\n");
464 /* but just in case, clean it up anyways! */
465 list_for_each_safe(le, tle, &tconn->out_of_sequence_requests) {
466 r = list_entry(le, struct drbd_request, tl_requests);
467 /* It would be nice to complete outside of spinlock.
468 * But this is easier for now. */
469 _req_mod(r, CONNECTION_LOST_WHILE_PENDING);
472 /* ensure bit indicating barrier is required is clear */
474 idr_for_each_entry(&tconn->volumes, mdev, vnr)
475 clear_bit(CREATE_BARRIER, &mdev->flags);
478 spin_unlock_irq(&tconn->req_lock);
481 void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what)
483 spin_lock_irq(&tconn->req_lock);
484 _tl_restart(tconn, what);
485 spin_unlock_irq(&tconn->req_lock);
489 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain mdev in the TL
490 * @mdev: DRBD device.
492 void tl_abort_disk_io(struct drbd_conf *mdev)
494 struct drbd_tconn *tconn = mdev->tconn;
495 struct drbd_tl_epoch *b;
496 struct list_head *le, *tle;
497 struct drbd_request *req;
499 spin_lock_irq(&tconn->req_lock);
500 b = tconn->oldest_tle;
502 list_for_each_safe(le, tle, &b->requests) {
503 req = list_entry(le, struct drbd_request, tl_requests);
504 if (req->w.mdev == mdev)
505 _req_mod(req, ABORT_DISK_IO);
510 list_for_each_safe(le, tle, &tconn->barrier_acked_requests) {
511 req = list_entry(le, struct drbd_request, tl_requests);
512 if (req->w.mdev == mdev)
513 _req_mod(req, ABORT_DISK_IO);
516 spin_unlock_irq(&tconn->req_lock);
519 static int drbd_thread_setup(void *arg)
521 struct drbd_thread *thi = (struct drbd_thread *) arg;
522 struct drbd_tconn *tconn = thi->tconn;
526 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
527 thi->name[0], thi->tconn->name);
530 retval = thi->function(thi);
532 spin_lock_irqsave(&thi->t_lock, flags);
534 /* if the receiver has been "EXITING", the last thing it did
535 * was set the conn state to "StandAlone",
536 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
537 * and receiver thread will be "started".
538 * drbd_thread_start needs to set "RESTARTING" in that case.
539 * t_state check and assignment needs to be within the same spinlock,
540 * so either thread_start sees EXITING, and can remap to RESTARTING,
541 * or thread_start see NONE, and can proceed as normal.
544 if (thi->t_state == RESTARTING) {
545 conn_info(tconn, "Restarting %s thread\n", thi->name);
546 thi->t_state = RUNNING;
547 spin_unlock_irqrestore(&thi->t_lock, flags);
554 complete_all(&thi->stop);
555 spin_unlock_irqrestore(&thi->t_lock, flags);
557 conn_info(tconn, "Terminating %s\n", current->comm);
559 /* Release mod reference taken when thread was started */
561 kref_put(&tconn->kref, &conn_destroy);
562 module_put(THIS_MODULE);
566 static void drbd_thread_init(struct drbd_tconn *tconn, struct drbd_thread *thi,
567 int (*func) (struct drbd_thread *), char *name)
569 spin_lock_init(&thi->t_lock);
572 thi->function = func;
574 strncpy(thi->name, name, ARRAY_SIZE(thi->name));
577 int drbd_thread_start(struct drbd_thread *thi)
579 struct drbd_tconn *tconn = thi->tconn;
580 struct task_struct *nt;
583 /* is used from state engine doing drbd_thread_stop_nowait,
584 * while holding the req lock irqsave */
585 spin_lock_irqsave(&thi->t_lock, flags);
587 switch (thi->t_state) {
589 conn_info(tconn, "Starting %s thread (from %s [%d])\n",
590 thi->name, current->comm, current->pid);
592 /* Get ref on module for thread - this is released when thread exits */
593 if (!try_module_get(THIS_MODULE)) {
594 conn_err(tconn, "Failed to get module reference in drbd_thread_start\n");
595 spin_unlock_irqrestore(&thi->t_lock, flags);
599 kref_get(&thi->tconn->kref);
601 init_completion(&thi->stop);
602 thi->reset_cpu_mask = 1;
603 thi->t_state = RUNNING;
604 spin_unlock_irqrestore(&thi->t_lock, flags);
605 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
607 nt = kthread_create(drbd_thread_setup, (void *) thi,
608 "drbd_%c_%s", thi->name[0], thi->tconn->name);
611 conn_err(tconn, "Couldn't start thread\n");
613 kref_put(&tconn->kref, &conn_destroy);
614 module_put(THIS_MODULE);
617 spin_lock_irqsave(&thi->t_lock, flags);
619 thi->t_state = RUNNING;
620 spin_unlock_irqrestore(&thi->t_lock, flags);
624 thi->t_state = RESTARTING;
625 conn_info(tconn, "Restarting %s thread (from %s [%d])\n",
626 thi->name, current->comm, current->pid);
631 spin_unlock_irqrestore(&thi->t_lock, flags);
639 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
643 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
645 /* may be called from state engine, holding the req lock irqsave */
646 spin_lock_irqsave(&thi->t_lock, flags);
648 if (thi->t_state == NONE) {
649 spin_unlock_irqrestore(&thi->t_lock, flags);
651 drbd_thread_start(thi);
655 if (thi->t_state != ns) {
656 if (thi->task == NULL) {
657 spin_unlock_irqrestore(&thi->t_lock, flags);
663 init_completion(&thi->stop);
664 if (thi->task != current)
665 force_sig(DRBD_SIGKILL, thi->task);
668 spin_unlock_irqrestore(&thi->t_lock, flags);
671 wait_for_completion(&thi->stop);
674 static struct drbd_thread *drbd_task_to_thread(struct drbd_tconn *tconn, struct task_struct *task)
676 struct drbd_thread *thi =
677 task == tconn->receiver.task ? &tconn->receiver :
678 task == tconn->asender.task ? &tconn->asender :
679 task == tconn->worker.task ? &tconn->worker : NULL;
684 char *drbd_task_to_thread_name(struct drbd_tconn *tconn, struct task_struct *task)
686 struct drbd_thread *thi = drbd_task_to_thread(tconn, task);
687 return thi ? thi->name : task->comm;
690 int conn_lowest_minor(struct drbd_tconn *tconn)
692 struct drbd_conf *mdev;
696 mdev = idr_get_next(&tconn->volumes, &vnr);
697 m = mdev ? mdev_to_minor(mdev) : -1;
705 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
706 * @mdev: DRBD device.
708 * Forces all threads of a device onto the same CPU. This is beneficial for
709 * DRBD's performance. May be overwritten by user's configuration.
711 void drbd_calc_cpu_mask(struct drbd_tconn *tconn)
716 if (cpumask_weight(tconn->cpu_mask))
719 ord = conn_lowest_minor(tconn) % cpumask_weight(cpu_online_mask);
720 for_each_online_cpu(cpu) {
722 cpumask_set_cpu(cpu, tconn->cpu_mask);
726 /* should not be reached */
727 cpumask_setall(tconn->cpu_mask);
731 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
732 * @mdev: DRBD device.
733 * @thi: drbd_thread object
735 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
738 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
740 struct task_struct *p = current;
742 if (!thi->reset_cpu_mask)
744 thi->reset_cpu_mask = 0;
745 set_cpus_allowed_ptr(p, thi->tconn->cpu_mask);
750 * drbd_header_size - size of a packet header
752 * The header size is a multiple of 8, so any payload following the header is
753 * word aligned on 64-bit architectures. (The bitmap send and receive code
756 unsigned int drbd_header_size(struct drbd_tconn *tconn)
758 if (tconn->agreed_pro_version >= 100) {
759 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
760 return sizeof(struct p_header100);
762 BUILD_BUG_ON(sizeof(struct p_header80) !=
763 sizeof(struct p_header95));
764 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
765 return sizeof(struct p_header80);
769 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
771 h->magic = cpu_to_be32(DRBD_MAGIC);
772 h->command = cpu_to_be16(cmd);
773 h->length = cpu_to_be16(size);
774 return sizeof(struct p_header80);
777 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
779 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
780 h->command = cpu_to_be16(cmd);
781 h->length = cpu_to_be32(size);
782 return sizeof(struct p_header95);
785 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
788 h->magic = cpu_to_be32(DRBD_MAGIC_100);
789 h->volume = cpu_to_be16(vnr);
790 h->command = cpu_to_be16(cmd);
791 h->length = cpu_to_be32(size);
793 return sizeof(struct p_header100);
796 static unsigned int prepare_header(struct drbd_tconn *tconn, int vnr,
797 void *buffer, enum drbd_packet cmd, int size)
799 if (tconn->agreed_pro_version >= 100)
800 return prepare_header100(buffer, cmd, size, vnr);
801 else if (tconn->agreed_pro_version >= 95 &&
802 size > DRBD_MAX_SIZE_H80_PACKET)
803 return prepare_header95(buffer, cmd, size);
805 return prepare_header80(buffer, cmd, size);
808 static void *__conn_prepare_command(struct drbd_tconn *tconn,
809 struct drbd_socket *sock)
813 return sock->sbuf + drbd_header_size(tconn);
816 void *conn_prepare_command(struct drbd_tconn *tconn, struct drbd_socket *sock)
820 mutex_lock(&sock->mutex);
821 p = __conn_prepare_command(tconn, sock);
823 mutex_unlock(&sock->mutex);
828 void *drbd_prepare_command(struct drbd_conf *mdev, struct drbd_socket *sock)
830 return conn_prepare_command(mdev->tconn, sock);
833 static int __send_command(struct drbd_tconn *tconn, int vnr,
834 struct drbd_socket *sock, enum drbd_packet cmd,
835 unsigned int header_size, void *data,
842 * Called with @data == NULL and the size of the data blocks in @size
843 * for commands that send data blocks. For those commands, omit the
844 * MSG_MORE flag: this will increase the likelihood that data blocks
845 * which are page aligned on the sender will end up page aligned on the
848 msg_flags = data ? MSG_MORE : 0;
850 header_size += prepare_header(tconn, vnr, sock->sbuf, cmd,
852 err = drbd_send_all(tconn, sock->socket, sock->sbuf, header_size,
855 err = drbd_send_all(tconn, sock->socket, data, size, 0);
859 static int __conn_send_command(struct drbd_tconn *tconn, struct drbd_socket *sock,
860 enum drbd_packet cmd, unsigned int header_size,
861 void *data, unsigned int size)
863 return __send_command(tconn, 0, sock, cmd, header_size, data, size);
866 int conn_send_command(struct drbd_tconn *tconn, struct drbd_socket *sock,
867 enum drbd_packet cmd, unsigned int header_size,
868 void *data, unsigned int size)
872 err = __conn_send_command(tconn, sock, cmd, header_size, data, size);
873 mutex_unlock(&sock->mutex);
877 int drbd_send_command(struct drbd_conf *mdev, struct drbd_socket *sock,
878 enum drbd_packet cmd, unsigned int header_size,
879 void *data, unsigned int size)
883 err = __send_command(mdev->tconn, mdev->vnr, sock, cmd, header_size,
885 mutex_unlock(&sock->mutex);
889 int drbd_send_ping(struct drbd_tconn *tconn)
891 struct drbd_socket *sock;
894 if (!conn_prepare_command(tconn, sock))
896 return conn_send_command(tconn, sock, P_PING, 0, NULL, 0);
899 int drbd_send_ping_ack(struct drbd_tconn *tconn)
901 struct drbd_socket *sock;
904 if (!conn_prepare_command(tconn, sock))
906 return conn_send_command(tconn, sock, P_PING_ACK, 0, NULL, 0);
909 int drbd_send_sync_param(struct drbd_conf *mdev)
911 struct drbd_socket *sock;
912 struct p_rs_param_95 *p;
914 const int apv = mdev->tconn->agreed_pro_version;
915 enum drbd_packet cmd;
917 struct disk_conf *dc;
919 sock = &mdev->tconn->data;
920 p = drbd_prepare_command(mdev, sock);
925 nc = rcu_dereference(mdev->tconn->net_conf);
927 size = apv <= 87 ? sizeof(struct p_rs_param)
928 : apv == 88 ? sizeof(struct p_rs_param)
929 + strlen(nc->verify_alg) + 1
930 : apv <= 94 ? sizeof(struct p_rs_param_89)
931 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
933 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
935 /* initialize verify_alg and csums_alg */
936 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
938 if (get_ldev(mdev)) {
939 dc = rcu_dereference(mdev->ldev->disk_conf);
940 p->resync_rate = cpu_to_be32(dc->resync_rate);
941 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
942 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
943 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
944 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
947 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
948 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
949 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
950 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
951 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
955 strcpy(p->verify_alg, nc->verify_alg);
957 strcpy(p->csums_alg, nc->csums_alg);
960 return drbd_send_command(mdev, sock, cmd, size, NULL, 0);
963 int __drbd_send_protocol(struct drbd_tconn *tconn, enum drbd_packet cmd)
965 struct drbd_socket *sock;
966 struct p_protocol *p;
971 p = __conn_prepare_command(tconn, sock);
976 nc = rcu_dereference(tconn->net_conf);
978 if (nc->tentative && tconn->agreed_pro_version < 92) {
980 mutex_unlock(&sock->mutex);
981 conn_err(tconn, "--dry-run is not supported by peer");
986 if (tconn->agreed_pro_version >= 87)
987 size += strlen(nc->integrity_alg) + 1;
989 p->protocol = cpu_to_be32(nc->wire_protocol);
990 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
991 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
992 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
993 p->two_primaries = cpu_to_be32(nc->two_primaries);
995 if (nc->discard_my_data)
996 cf |= CF_DISCARD_MY_DATA;
999 p->conn_flags = cpu_to_be32(cf);
1001 if (tconn->agreed_pro_version >= 87)
1002 strcpy(p->integrity_alg, nc->integrity_alg);
1005 return __conn_send_command(tconn, sock, cmd, size, NULL, 0);
1008 int drbd_send_protocol(struct drbd_tconn *tconn)
1012 mutex_lock(&tconn->data.mutex);
1013 err = __drbd_send_protocol(tconn, P_PROTOCOL);
1014 mutex_unlock(&tconn->data.mutex);
1019 int _drbd_send_uuids(struct drbd_conf *mdev, u64 uuid_flags)
1021 struct drbd_socket *sock;
1025 if (!get_ldev_if_state(mdev, D_NEGOTIATING))
1028 sock = &mdev->tconn->data;
1029 p = drbd_prepare_command(mdev, sock);
1034 for (i = UI_CURRENT; i < UI_SIZE; i++)
1035 p->uuid[i] = mdev->ldev ? cpu_to_be64(mdev->ldev->md.uuid[i]) : 0;
1037 mdev->comm_bm_set = drbd_bm_total_weight(mdev);
1038 p->uuid[UI_SIZE] = cpu_to_be64(mdev->comm_bm_set);
1040 uuid_flags |= rcu_dereference(mdev->tconn->net_conf)->discard_my_data ? 1 : 0;
1042 uuid_flags |= test_bit(CRASHED_PRIMARY, &mdev->flags) ? 2 : 0;
1043 uuid_flags |= mdev->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
1044 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
1047 return drbd_send_command(mdev, sock, P_UUIDS, sizeof(*p), NULL, 0);
1050 int drbd_send_uuids(struct drbd_conf *mdev)
1052 return _drbd_send_uuids(mdev, 0);
1055 int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev)
1057 return _drbd_send_uuids(mdev, 8);
1060 void drbd_print_uuids(struct drbd_conf *mdev, const char *text)
1062 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
1063 u64 *uuid = mdev->ldev->md.uuid;
1064 dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX\n",
1066 (unsigned long long)uuid[UI_CURRENT],
1067 (unsigned long long)uuid[UI_BITMAP],
1068 (unsigned long long)uuid[UI_HISTORY_START],
1069 (unsigned long long)uuid[UI_HISTORY_END]);
1072 dev_info(DEV, "%s effective data uuid: %016llX\n",
1074 (unsigned long long)mdev->ed_uuid);
1078 void drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev)
1080 struct drbd_socket *sock;
1081 struct p_rs_uuid *p;
1084 D_ASSERT(mdev->state.disk == D_UP_TO_DATE);
1086 uuid = mdev->ldev->md.uuid[UI_BITMAP] + UUID_NEW_BM_OFFSET;
1087 drbd_uuid_set(mdev, UI_BITMAP, uuid);
1088 drbd_print_uuids(mdev, "updated sync UUID");
1091 sock = &mdev->tconn->data;
1092 p = drbd_prepare_command(mdev, sock);
1094 p->uuid = cpu_to_be64(uuid);
1095 drbd_send_command(mdev, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
1099 int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags)
1101 struct drbd_socket *sock;
1103 sector_t d_size, u_size;
1104 int q_order_type, max_bio_size;
1106 if (get_ldev_if_state(mdev, D_NEGOTIATING)) {
1107 D_ASSERT(mdev->ldev->backing_bdev);
1108 d_size = drbd_get_max_capacity(mdev->ldev);
1110 u_size = rcu_dereference(mdev->ldev->disk_conf)->disk_size;
1112 q_order_type = drbd_queue_order_type(mdev);
1113 max_bio_size = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9;
1114 max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE);
1119 q_order_type = QUEUE_ORDERED_NONE;
1120 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
1123 sock = &mdev->tconn->data;
1124 p = drbd_prepare_command(mdev, sock);
1128 if (mdev->tconn->agreed_pro_version <= 94)
1129 max_bio_size = min_t(int, max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
1130 else if (mdev->tconn->agreed_pro_version < 100)
1131 max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE_P95);
1133 p->d_size = cpu_to_be64(d_size);
1134 p->u_size = cpu_to_be64(u_size);
1135 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(mdev->this_bdev));
1136 p->max_bio_size = cpu_to_be32(max_bio_size);
1137 p->queue_order_type = cpu_to_be16(q_order_type);
1138 p->dds_flags = cpu_to_be16(flags);
1139 return drbd_send_command(mdev, sock, P_SIZES, sizeof(*p), NULL, 0);
1143 * drbd_send_state() - Sends the drbd state to the peer
1144 * @mdev: DRBD device.
1146 int drbd_send_state(struct drbd_conf *mdev)
1148 struct drbd_socket *sock;
1151 sock = &mdev->tconn->data;
1152 p = drbd_prepare_command(mdev, sock);
1155 p->state = cpu_to_be32(mdev->state.i); /* Within the send mutex */
1156 return drbd_send_command(mdev, sock, P_STATE, sizeof(*p), NULL, 0);
1159 int drbd_send_state_req(struct drbd_conf *mdev, union drbd_state mask, union drbd_state val)
1161 struct drbd_socket *sock;
1162 struct p_req_state *p;
1164 sock = &mdev->tconn->data;
1165 p = drbd_prepare_command(mdev, sock);
1168 p->mask = cpu_to_be32(mask.i);
1169 p->val = cpu_to_be32(val.i);
1170 return drbd_send_command(mdev, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1174 int conn_send_state_req(struct drbd_tconn *tconn, union drbd_state mask, union drbd_state val)
1176 enum drbd_packet cmd;
1177 struct drbd_socket *sock;
1178 struct p_req_state *p;
1180 cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1181 sock = &tconn->data;
1182 p = conn_prepare_command(tconn, sock);
1185 p->mask = cpu_to_be32(mask.i);
1186 p->val = cpu_to_be32(val.i);
1187 return conn_send_command(tconn, sock, cmd, sizeof(*p), NULL, 0);
1190 void drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode)
1192 struct drbd_socket *sock;
1193 struct p_req_state_reply *p;
1195 sock = &mdev->tconn->meta;
1196 p = drbd_prepare_command(mdev, sock);
1198 p->retcode = cpu_to_be32(retcode);
1199 drbd_send_command(mdev, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1203 void conn_send_sr_reply(struct drbd_tconn *tconn, enum drbd_state_rv retcode)
1205 struct drbd_socket *sock;
1206 struct p_req_state_reply *p;
1207 enum drbd_packet cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1209 sock = &tconn->meta;
1210 p = conn_prepare_command(tconn, sock);
1212 p->retcode = cpu_to_be32(retcode);
1213 conn_send_command(tconn, sock, cmd, sizeof(*p), NULL, 0);
1217 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1219 BUG_ON(code & ~0xf);
1220 p->encoding = (p->encoding & ~0xf) | code;
1223 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1225 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1228 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1231 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1234 int fill_bitmap_rle_bits(struct drbd_conf *mdev,
1235 struct p_compressed_bm *p,
1237 struct bm_xfer_ctx *c)
1239 struct bitstream bs;
1240 unsigned long plain_bits;
1247 /* may we use this feature? */
1249 use_rle = rcu_dereference(mdev->tconn->net_conf)->use_rle;
1251 if (!use_rle || mdev->tconn->agreed_pro_version < 90)
1254 if (c->bit_offset >= c->bm_bits)
1255 return 0; /* nothing to do. */
1257 /* use at most thus many bytes */
1258 bitstream_init(&bs, p->code, size, 0);
1259 memset(p->code, 0, size);
1260 /* plain bits covered in this code string */
1263 /* p->encoding & 0x80 stores whether the first run length is set.
1264 * bit offset is implicit.
1265 * start with toggle == 2 to be able to tell the first iteration */
1268 /* see how much plain bits we can stuff into one packet
1269 * using RLE and VLI. */
1271 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(mdev, c->bit_offset)
1272 : _drbd_bm_find_next(mdev, c->bit_offset);
1275 rl = tmp - c->bit_offset;
1277 if (toggle == 2) { /* first iteration */
1279 /* the first checked bit was set,
1280 * store start value, */
1281 dcbp_set_start(p, 1);
1282 /* but skip encoding of zero run length */
1286 dcbp_set_start(p, 0);
1289 /* paranoia: catch zero runlength.
1290 * can only happen if bitmap is modified while we scan it. */
1292 dev_err(DEV, "unexpected zero runlength while encoding bitmap "
1293 "t:%u bo:%lu\n", toggle, c->bit_offset);
1297 bits = vli_encode_bits(&bs, rl);
1298 if (bits == -ENOBUFS) /* buffer full */
1301 dev_err(DEV, "error while encoding bitmap: %d\n", bits);
1307 c->bit_offset = tmp;
1308 } while (c->bit_offset < c->bm_bits);
1310 len = bs.cur.b - p->code + !!bs.cur.bit;
1312 if (plain_bits < (len << 3)) {
1313 /* incompressible with this method.
1314 * we need to rewind both word and bit position. */
1315 c->bit_offset -= plain_bits;
1316 bm_xfer_ctx_bit_to_word_offset(c);
1317 c->bit_offset = c->word_offset * BITS_PER_LONG;
1321 /* RLE + VLI was able to compress it just fine.
1322 * update c->word_offset. */
1323 bm_xfer_ctx_bit_to_word_offset(c);
1325 /* store pad_bits */
1326 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1332 * send_bitmap_rle_or_plain
1334 * Return 0 when done, 1 when another iteration is needed, and a negative error
1335 * code upon failure.
1338 send_bitmap_rle_or_plain(struct drbd_conf *mdev, struct bm_xfer_ctx *c)
1340 struct drbd_socket *sock = &mdev->tconn->data;
1341 unsigned int header_size = drbd_header_size(mdev->tconn);
1342 struct p_compressed_bm *p = sock->sbuf + header_size;
1345 len = fill_bitmap_rle_bits(mdev, p,
1346 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1351 dcbp_set_code(p, RLE_VLI_Bits);
1352 err = __send_command(mdev->tconn, mdev->vnr, sock,
1353 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1356 c->bytes[0] += header_size + sizeof(*p) + len;
1358 if (c->bit_offset >= c->bm_bits)
1361 /* was not compressible.
1362 * send a buffer full of plain text bits instead. */
1363 unsigned int data_size;
1364 unsigned long num_words;
1365 unsigned long *p = sock->sbuf + header_size;
1367 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1368 num_words = min_t(size_t, data_size / sizeof(*p),
1369 c->bm_words - c->word_offset);
1370 len = num_words * sizeof(*p);
1372 drbd_bm_get_lel(mdev, c->word_offset, num_words, p);
1373 err = __send_command(mdev->tconn, mdev->vnr, sock, P_BITMAP, len, NULL, 0);
1374 c->word_offset += num_words;
1375 c->bit_offset = c->word_offset * BITS_PER_LONG;
1378 c->bytes[1] += header_size + len;
1380 if (c->bit_offset > c->bm_bits)
1381 c->bit_offset = c->bm_bits;
1385 INFO_bm_xfer_stats(mdev, "send", c);
1393 /* See the comment at receive_bitmap() */
1394 static int _drbd_send_bitmap(struct drbd_conf *mdev)
1396 struct bm_xfer_ctx c;
1399 if (!expect(mdev->bitmap))
1402 if (get_ldev(mdev)) {
1403 if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) {
1404 dev_info(DEV, "Writing the whole bitmap, MDF_FullSync was set.\n");
1405 drbd_bm_set_all(mdev);
1406 if (drbd_bm_write(mdev)) {
1407 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1408 * but otherwise process as per normal - need to tell other
1409 * side that a full resync is required! */
1410 dev_err(DEV, "Failed to write bitmap to disk!\n");
1412 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
1419 c = (struct bm_xfer_ctx) {
1420 .bm_bits = drbd_bm_bits(mdev),
1421 .bm_words = drbd_bm_words(mdev),
1425 err = send_bitmap_rle_or_plain(mdev, &c);
1431 int drbd_send_bitmap(struct drbd_conf *mdev)
1433 struct drbd_socket *sock = &mdev->tconn->data;
1436 mutex_lock(&sock->mutex);
1438 err = !_drbd_send_bitmap(mdev);
1439 mutex_unlock(&sock->mutex);
1443 void drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, u32 set_size)
1445 struct drbd_socket *sock;
1446 struct p_barrier_ack *p;
1448 if (mdev->state.conn < C_CONNECTED)
1451 sock = &mdev->tconn->meta;
1452 p = drbd_prepare_command(mdev, sock);
1455 p->barrier = barrier_nr;
1456 p->set_size = cpu_to_be32(set_size);
1457 drbd_send_command(mdev, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1461 * _drbd_send_ack() - Sends an ack packet
1462 * @mdev: DRBD device.
1463 * @cmd: Packet command code.
1464 * @sector: sector, needs to be in big endian byte order
1465 * @blksize: size in byte, needs to be in big endian byte order
1466 * @block_id: Id, big endian byte order
1468 static int _drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1469 u64 sector, u32 blksize, u64 block_id)
1471 struct drbd_socket *sock;
1472 struct p_block_ack *p;
1474 if (mdev->state.conn < C_CONNECTED)
1477 sock = &mdev->tconn->meta;
1478 p = drbd_prepare_command(mdev, sock);
1482 p->block_id = block_id;
1483 p->blksize = blksize;
1484 p->seq_num = cpu_to_be32(atomic_inc_return(&mdev->packet_seq));
1485 return drbd_send_command(mdev, sock, cmd, sizeof(*p), NULL, 0);
1488 /* dp->sector and dp->block_id already/still in network byte order,
1489 * data_size is payload size according to dp->head,
1490 * and may need to be corrected for digest size. */
1491 void drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd,
1492 struct p_data *dp, int data_size)
1494 if (mdev->tconn->peer_integrity_tfm)
1495 data_size -= crypto_hash_digestsize(mdev->tconn->peer_integrity_tfm);
1496 _drbd_send_ack(mdev, cmd, dp->sector, cpu_to_be32(data_size),
1500 void drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd,
1501 struct p_block_req *rp)
1503 _drbd_send_ack(mdev, cmd, rp->sector, rp->blksize, rp->block_id);
1507 * drbd_send_ack() - Sends an ack packet
1508 * @mdev: DRBD device
1509 * @cmd: packet command code
1510 * @peer_req: peer request
1512 int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd,
1513 struct drbd_peer_request *peer_req)
1515 return _drbd_send_ack(mdev, cmd,
1516 cpu_to_be64(peer_req->i.sector),
1517 cpu_to_be32(peer_req->i.size),
1518 peer_req->block_id);
1521 /* This function misuses the block_id field to signal if the blocks
1522 * are is sync or not. */
1523 int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd,
1524 sector_t sector, int blksize, u64 block_id)
1526 return _drbd_send_ack(mdev, cmd,
1527 cpu_to_be64(sector),
1528 cpu_to_be32(blksize),
1529 cpu_to_be64(block_id));
1532 int drbd_send_drequest(struct drbd_conf *mdev, int cmd,
1533 sector_t sector, int size, u64 block_id)
1535 struct drbd_socket *sock;
1536 struct p_block_req *p;
1538 sock = &mdev->tconn->data;
1539 p = drbd_prepare_command(mdev, sock);
1542 p->sector = cpu_to_be64(sector);
1543 p->block_id = block_id;
1544 p->blksize = cpu_to_be32(size);
1545 return drbd_send_command(mdev, sock, cmd, sizeof(*p), NULL, 0);
1548 int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, int size,
1549 void *digest, int digest_size, enum drbd_packet cmd)
1551 struct drbd_socket *sock;
1552 struct p_block_req *p;
1554 /* FIXME: Put the digest into the preallocated socket buffer. */
1556 sock = &mdev->tconn->data;
1557 p = drbd_prepare_command(mdev, sock);
1560 p->sector = cpu_to_be64(sector);
1561 p->block_id = ID_SYNCER /* unused */;
1562 p->blksize = cpu_to_be32(size);
1563 return drbd_send_command(mdev, sock, cmd, sizeof(*p),
1564 digest, digest_size);
1567 int drbd_send_ov_request(struct drbd_conf *mdev, sector_t sector, int size)
1569 struct drbd_socket *sock;
1570 struct p_block_req *p;
1572 sock = &mdev->tconn->data;
1573 p = drbd_prepare_command(mdev, sock);
1576 p->sector = cpu_to_be64(sector);
1577 p->block_id = ID_SYNCER /* unused */;
1578 p->blksize = cpu_to_be32(size);
1579 return drbd_send_command(mdev, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1582 /* called on sndtimeo
1583 * returns false if we should retry,
1584 * true if we think connection is dead
1586 static int we_should_drop_the_connection(struct drbd_tconn *tconn, struct socket *sock)
1589 /* long elapsed = (long)(jiffies - mdev->last_received); */
1591 drop_it = tconn->meta.socket == sock
1592 || !tconn->asender.task
1593 || get_t_state(&tconn->asender) != RUNNING
1594 || tconn->cstate < C_WF_REPORT_PARAMS;
1599 drop_it = !--tconn->ko_count;
1601 conn_err(tconn, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1602 current->comm, current->pid, tconn->ko_count);
1603 request_ping(tconn);
1606 return drop_it; /* && (mdev->state == R_PRIMARY) */;
1609 static void drbd_update_congested(struct drbd_tconn *tconn)
1611 struct sock *sk = tconn->data.socket->sk;
1612 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1613 set_bit(NET_CONGESTED, &tconn->flags);
1616 /* The idea of sendpage seems to be to put some kind of reference
1617 * to the page into the skb, and to hand it over to the NIC. In
1618 * this process get_page() gets called.
1620 * As soon as the page was really sent over the network put_page()
1621 * gets called by some part of the network layer. [ NIC driver? ]
1623 * [ get_page() / put_page() increment/decrement the count. If count
1624 * reaches 0 the page will be freed. ]
1626 * This works nicely with pages from FSs.
1627 * But this means that in protocol A we might signal IO completion too early!
1629 * In order not to corrupt data during a resync we must make sure
1630 * that we do not reuse our own buffer pages (EEs) to early, therefore
1631 * we have the net_ee list.
1633 * XFS seems to have problems, still, it submits pages with page_count == 0!
1634 * As a workaround, we disable sendpage on pages
1635 * with page_count == 0 or PageSlab.
1637 static int _drbd_no_send_page(struct drbd_conf *mdev, struct page *page,
1638 int offset, size_t size, unsigned msg_flags)
1640 struct socket *socket;
1644 socket = mdev->tconn->data.socket;
1645 addr = kmap(page) + offset;
1646 err = drbd_send_all(mdev->tconn, socket, addr, size, msg_flags);
1649 mdev->send_cnt += size >> 9;
1653 static int _drbd_send_page(struct drbd_conf *mdev, struct page *page,
1654 int offset, size_t size, unsigned msg_flags)
1656 struct socket *socket = mdev->tconn->data.socket;
1657 mm_segment_t oldfs = get_fs();
1661 /* e.g. XFS meta- & log-data is in slab pages, which have a
1662 * page_count of 0 and/or have PageSlab() set.
1663 * we cannot use send_page for those, as that does get_page();
1664 * put_page(); and would cause either a VM_BUG directly, or
1665 * __page_cache_release a page that would actually still be referenced
1666 * by someone, leading to some obscure delayed Oops somewhere else. */
1667 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1668 return _drbd_no_send_page(mdev, page, offset, size, msg_flags);
1670 msg_flags |= MSG_NOSIGNAL;
1671 drbd_update_congested(mdev->tconn);
1676 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1678 if (sent == -EAGAIN) {
1679 if (we_should_drop_the_connection(mdev->tconn, socket))
1683 dev_warn(DEV, "%s: size=%d len=%d sent=%d\n",
1684 __func__, (int)size, len, sent);
1691 } while (len > 0 /* THINK && mdev->cstate >= C_CONNECTED*/);
1693 clear_bit(NET_CONGESTED, &mdev->tconn->flags);
1697 mdev->send_cnt += size >> 9;
1702 static int _drbd_send_bio(struct drbd_conf *mdev, struct bio *bio)
1704 struct bio_vec *bvec;
1706 /* hint all but last page with MSG_MORE */
1707 __bio_for_each_segment(bvec, bio, i, 0) {
1710 err = _drbd_no_send_page(mdev, bvec->bv_page,
1711 bvec->bv_offset, bvec->bv_len,
1712 i == bio->bi_vcnt - 1 ? 0 : MSG_MORE);
1719 static int _drbd_send_zc_bio(struct drbd_conf *mdev, struct bio *bio)
1721 struct bio_vec *bvec;
1723 /* hint all but last page with MSG_MORE */
1724 __bio_for_each_segment(bvec, bio, i, 0) {
1727 err = _drbd_send_page(mdev, bvec->bv_page,
1728 bvec->bv_offset, bvec->bv_len,
1729 i == bio->bi_vcnt - 1 ? 0 : MSG_MORE);
1736 static int _drbd_send_zc_ee(struct drbd_conf *mdev,
1737 struct drbd_peer_request *peer_req)
1739 struct page *page = peer_req->pages;
1740 unsigned len = peer_req->i.size;
1743 /* hint all but last page with MSG_MORE */
1744 page_chain_for_each(page) {
1745 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1747 err = _drbd_send_page(mdev, page, 0, l,
1748 page_chain_next(page) ? MSG_MORE : 0);
1756 static u32 bio_flags_to_wire(struct drbd_conf *mdev, unsigned long bi_rw)
1758 if (mdev->tconn->agreed_pro_version >= 95)
1759 return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) |
1760 (bi_rw & REQ_FUA ? DP_FUA : 0) |
1761 (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) |
1762 (bi_rw & REQ_DISCARD ? DP_DISCARD : 0);
1764 return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0;
1767 /* Used to send write requests
1768 * R_PRIMARY -> Peer (P_DATA)
1770 int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req)
1772 struct drbd_socket *sock;
1774 unsigned int dp_flags = 0;
1778 sock = &mdev->tconn->data;
1779 p = drbd_prepare_command(mdev, sock);
1780 dgs = mdev->tconn->integrity_tfm ? crypto_hash_digestsize(mdev->tconn->integrity_tfm) : 0;
1784 p->sector = cpu_to_be64(req->i.sector);
1785 p->block_id = (unsigned long)req;
1786 p->seq_num = cpu_to_be32(req->seq_num = atomic_inc_return(&mdev->packet_seq));
1787 dp_flags = bio_flags_to_wire(mdev, req->master_bio->bi_rw);
1788 if (mdev->state.conn >= C_SYNC_SOURCE &&
1789 mdev->state.conn <= C_PAUSED_SYNC_T)
1790 dp_flags |= DP_MAY_SET_IN_SYNC;
1791 if (mdev->tconn->agreed_pro_version >= 100) {
1792 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1793 dp_flags |= DP_SEND_RECEIVE_ACK;
1794 if (req->rq_state & RQ_EXP_WRITE_ACK)
1795 dp_flags |= DP_SEND_WRITE_ACK;
1797 p->dp_flags = cpu_to_be32(dp_flags);
1799 drbd_csum_bio(mdev, mdev->tconn->integrity_tfm, req->master_bio, p + 1);
1800 err = __send_command(mdev->tconn, mdev->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size);
1802 /* For protocol A, we have to memcpy the payload into
1803 * socket buffers, as we may complete right away
1804 * as soon as we handed it over to tcp, at which point the data
1805 * pages may become invalid.
1807 * For data-integrity enabled, we copy it as well, so we can be
1808 * sure that even if the bio pages may still be modified, it
1809 * won't change the data on the wire, thus if the digest checks
1810 * out ok after sending on this side, but does not fit on the
1811 * receiving side, we sure have detected corruption elsewhere.
1813 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs)
1814 err = _drbd_send_bio(mdev, req->master_bio);
1816 err = _drbd_send_zc_bio(mdev, req->master_bio);
1818 /* double check digest, sometimes buffers have been modified in flight. */
1819 if (dgs > 0 && dgs <= 64) {
1820 /* 64 byte, 512 bit, is the largest digest size
1821 * currently supported in kernel crypto. */
1822 unsigned char digest[64];
1823 drbd_csum_bio(mdev, mdev->tconn->integrity_tfm, req->master_bio, digest);
1824 if (memcmp(p + 1, digest, dgs)) {
1826 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1827 (unsigned long long)req->i.sector, req->i.size);
1829 } /* else if (dgs > 64) {
1830 ... Be noisy about digest too large ...
1833 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1838 /* answer packet, used to send data back for read requests:
1839 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1840 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1842 int drbd_send_block(struct drbd_conf *mdev, enum drbd_packet cmd,
1843 struct drbd_peer_request *peer_req)
1845 struct drbd_socket *sock;
1850 sock = &mdev->tconn->data;
1851 p = drbd_prepare_command(mdev, sock);
1853 dgs = mdev->tconn->integrity_tfm ? crypto_hash_digestsize(mdev->tconn->integrity_tfm) : 0;
1857 p->sector = cpu_to_be64(peer_req->i.sector);
1858 p->block_id = peer_req->block_id;
1859 p->seq_num = 0; /* unused */
1861 drbd_csum_ee(mdev, mdev->tconn->integrity_tfm, peer_req, p + 1);
1862 err = __send_command(mdev->tconn, mdev->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size);
1864 err = _drbd_send_zc_ee(mdev, peer_req);
1865 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1870 int drbd_send_out_of_sync(struct drbd_conf *mdev, struct drbd_request *req)
1872 struct drbd_socket *sock;
1873 struct p_block_desc *p;
1875 sock = &mdev->tconn->data;
1876 p = drbd_prepare_command(mdev, sock);
1879 p->sector = cpu_to_be64(req->i.sector);
1880 p->blksize = cpu_to_be32(req->i.size);
1881 return drbd_send_command(mdev, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1885 drbd_send distinguishes two cases:
1887 Packets sent via the data socket "sock"
1888 and packets sent via the meta data socket "msock"
1891 -----------------+-------------------------+------------------------------
1892 timeout conf.timeout / 2 conf.timeout / 2
1893 timeout action send a ping via msock Abort communication
1894 and close all sockets
1898 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1900 int drbd_send(struct drbd_tconn *tconn, struct socket *sock,
1901 void *buf, size_t size, unsigned msg_flags)
1910 /* THINK if (signal_pending) return ... ? */
1915 msg.msg_name = NULL;
1916 msg.msg_namelen = 0;
1917 msg.msg_control = NULL;
1918 msg.msg_controllen = 0;
1919 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1921 if (sock == tconn->data.socket) {
1923 tconn->ko_count = rcu_dereference(tconn->net_conf)->ko_count;
1925 drbd_update_congested(tconn);
1929 * tcp_sendmsg does _not_ use its size parameter at all ?
1931 * -EAGAIN on timeout, -EINTR on signal.
1934 * do we need to block DRBD_SIG if sock == &meta.socket ??
1935 * otherwise wake_asender() might interrupt some send_*Ack !
1937 rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
1938 if (rv == -EAGAIN) {
1939 if (we_should_drop_the_connection(tconn, sock))
1945 flush_signals(current);
1953 } while (sent < size);
1955 if (sock == tconn->data.socket)
1956 clear_bit(NET_CONGESTED, &tconn->flags);
1959 if (rv != -EAGAIN) {
1960 conn_err(tconn, "%s_sendmsg returned %d\n",
1961 sock == tconn->meta.socket ? "msock" : "sock",
1963 conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD);
1965 conn_request_state(tconn, NS(conn, C_TIMEOUT), CS_HARD);
1972 * drbd_send_all - Send an entire buffer
1974 * Returns 0 upon success and a negative error value otherwise.
1976 int drbd_send_all(struct drbd_tconn *tconn, struct socket *sock, void *buffer,
1977 size_t size, unsigned msg_flags)
1981 err = drbd_send(tconn, sock, buffer, size, msg_flags);
1989 static int drbd_open(struct block_device *bdev, fmode_t mode)
1991 struct drbd_conf *mdev = bdev->bd_disk->private_data;
1992 unsigned long flags;
1995 mutex_lock(&drbd_main_mutex);
1996 spin_lock_irqsave(&mdev->tconn->req_lock, flags);
1997 /* to have a stable mdev->state.role
1998 * and no race with updating open_cnt */
2000 if (mdev->state.role != R_PRIMARY) {
2001 if (mode & FMODE_WRITE)
2003 else if (!allow_oos)
2009 spin_unlock_irqrestore(&mdev->tconn->req_lock, flags);
2010 mutex_unlock(&drbd_main_mutex);
2015 static int drbd_release(struct gendisk *gd, fmode_t mode)
2017 struct drbd_conf *mdev = gd->private_data;
2018 mutex_lock(&drbd_main_mutex);
2020 mutex_unlock(&drbd_main_mutex);
2024 static void drbd_set_defaults(struct drbd_conf *mdev)
2026 /* Beware! The actual layout differs
2027 * between big endian and little endian */
2028 mdev->state = (union drbd_dev_state) {
2029 { .role = R_SECONDARY,
2031 .conn = C_STANDALONE,
2037 void drbd_init_set_defaults(struct drbd_conf *mdev)
2039 /* the memset(,0,) did most of this.
2040 * note: only assignments, no allocation in here */
2042 drbd_set_defaults(mdev);
2044 atomic_set(&mdev->ap_bio_cnt, 0);
2045 atomic_set(&mdev->ap_pending_cnt, 0);
2046 atomic_set(&mdev->rs_pending_cnt, 0);
2047 atomic_set(&mdev->unacked_cnt, 0);
2048 atomic_set(&mdev->local_cnt, 0);
2049 atomic_set(&mdev->pp_in_use_by_net, 0);
2050 atomic_set(&mdev->rs_sect_in, 0);
2051 atomic_set(&mdev->rs_sect_ev, 0);
2052 atomic_set(&mdev->ap_in_flight, 0);
2053 atomic_set(&mdev->md_io_in_use, 0);
2055 mutex_init(&mdev->own_state_mutex);
2056 mdev->state_mutex = &mdev->own_state_mutex;
2058 spin_lock_init(&mdev->al_lock);
2059 spin_lock_init(&mdev->peer_seq_lock);
2060 spin_lock_init(&mdev->epoch_lock);
2062 INIT_LIST_HEAD(&mdev->active_ee);
2063 INIT_LIST_HEAD(&mdev->sync_ee);
2064 INIT_LIST_HEAD(&mdev->done_ee);
2065 INIT_LIST_HEAD(&mdev->read_ee);
2066 INIT_LIST_HEAD(&mdev->net_ee);
2067 INIT_LIST_HEAD(&mdev->resync_reads);
2068 INIT_LIST_HEAD(&mdev->resync_work.list);
2069 INIT_LIST_HEAD(&mdev->unplug_work.list);
2070 INIT_LIST_HEAD(&mdev->go_diskless.list);
2071 INIT_LIST_HEAD(&mdev->md_sync_work.list);
2072 INIT_LIST_HEAD(&mdev->start_resync_work.list);
2073 INIT_LIST_HEAD(&mdev->bm_io_work.w.list);
2075 mdev->resync_work.cb = w_resync_timer;
2076 mdev->unplug_work.cb = w_send_write_hint;
2077 mdev->go_diskless.cb = w_go_diskless;
2078 mdev->md_sync_work.cb = w_md_sync;
2079 mdev->bm_io_work.w.cb = w_bitmap_io;
2080 mdev->start_resync_work.cb = w_start_resync;
2082 mdev->resync_work.mdev = mdev;
2083 mdev->unplug_work.mdev = mdev;
2084 mdev->go_diskless.mdev = mdev;
2085 mdev->md_sync_work.mdev = mdev;
2086 mdev->bm_io_work.w.mdev = mdev;
2087 mdev->start_resync_work.mdev = mdev;
2089 init_timer(&mdev->resync_timer);
2090 init_timer(&mdev->md_sync_timer);
2091 init_timer(&mdev->start_resync_timer);
2092 init_timer(&mdev->request_timer);
2093 mdev->resync_timer.function = resync_timer_fn;
2094 mdev->resync_timer.data = (unsigned long) mdev;
2095 mdev->md_sync_timer.function = md_sync_timer_fn;
2096 mdev->md_sync_timer.data = (unsigned long) mdev;
2097 mdev->start_resync_timer.function = start_resync_timer_fn;
2098 mdev->start_resync_timer.data = (unsigned long) mdev;
2099 mdev->request_timer.function = request_timer_fn;
2100 mdev->request_timer.data = (unsigned long) mdev;
2102 init_waitqueue_head(&mdev->misc_wait);
2103 init_waitqueue_head(&mdev->state_wait);
2104 init_waitqueue_head(&mdev->ee_wait);
2105 init_waitqueue_head(&mdev->al_wait);
2106 init_waitqueue_head(&mdev->seq_wait);
2108 mdev->write_ordering = WO_bdev_flush;
2109 mdev->resync_wenr = LC_FREE;
2110 mdev->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2111 mdev->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2114 void drbd_mdev_cleanup(struct drbd_conf *mdev)
2117 if (mdev->tconn->receiver.t_state != NONE)
2118 dev_err(DEV, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2119 mdev->tconn->receiver.t_state);
2121 /* no need to lock it, I'm the only thread alive */
2122 if (atomic_read(&mdev->current_epoch->epoch_size) != 0)
2123 dev_err(DEV, "epoch_size:%d\n", atomic_read(&mdev->current_epoch->epoch_size));
2133 mdev->rs_failed = 0;
2134 mdev->rs_last_events = 0;
2135 mdev->rs_last_sect_ev = 0;
2136 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2137 mdev->rs_mark_left[i] = 0;
2138 mdev->rs_mark_time[i] = 0;
2140 D_ASSERT(mdev->tconn->net_conf == NULL);
2142 drbd_set_my_capacity(mdev, 0);
2144 /* maybe never allocated. */
2145 drbd_bm_resize(mdev, 0, 1);
2146 drbd_bm_cleanup(mdev);
2149 drbd_free_bc(mdev->ldev);
2152 clear_bit(AL_SUSPENDED, &mdev->flags);
2154 D_ASSERT(list_empty(&mdev->active_ee));
2155 D_ASSERT(list_empty(&mdev->sync_ee));
2156 D_ASSERT(list_empty(&mdev->done_ee));
2157 D_ASSERT(list_empty(&mdev->read_ee));
2158 D_ASSERT(list_empty(&mdev->net_ee));
2159 D_ASSERT(list_empty(&mdev->resync_reads));
2160 D_ASSERT(list_empty(&mdev->tconn->data.work.q));
2161 D_ASSERT(list_empty(&mdev->tconn->meta.work.q));
2162 D_ASSERT(list_empty(&mdev->resync_work.list));
2163 D_ASSERT(list_empty(&mdev->unplug_work.list));
2164 D_ASSERT(list_empty(&mdev->go_diskless.list));
2166 drbd_set_defaults(mdev);
2170 static void drbd_destroy_mempools(void)
2174 while (drbd_pp_pool) {
2175 page = drbd_pp_pool;
2176 drbd_pp_pool = (struct page *)page_private(page);
2181 /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */
2183 if (drbd_md_io_bio_set)
2184 bioset_free(drbd_md_io_bio_set);
2185 if (drbd_md_io_page_pool)
2186 mempool_destroy(drbd_md_io_page_pool);
2187 if (drbd_ee_mempool)
2188 mempool_destroy(drbd_ee_mempool);
2189 if (drbd_request_mempool)
2190 mempool_destroy(drbd_request_mempool);
2192 kmem_cache_destroy(drbd_ee_cache);
2193 if (drbd_request_cache)
2194 kmem_cache_destroy(drbd_request_cache);
2195 if (drbd_bm_ext_cache)
2196 kmem_cache_destroy(drbd_bm_ext_cache);
2197 if (drbd_al_ext_cache)
2198 kmem_cache_destroy(drbd_al_ext_cache);
2200 drbd_md_io_bio_set = NULL;
2201 drbd_md_io_page_pool = NULL;
2202 drbd_ee_mempool = NULL;
2203 drbd_request_mempool = NULL;
2204 drbd_ee_cache = NULL;
2205 drbd_request_cache = NULL;
2206 drbd_bm_ext_cache = NULL;
2207 drbd_al_ext_cache = NULL;
2212 static int drbd_create_mempools(void)
2215 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
2218 /* prepare our caches and mempools */
2219 drbd_request_mempool = NULL;
2220 drbd_ee_cache = NULL;
2221 drbd_request_cache = NULL;
2222 drbd_bm_ext_cache = NULL;
2223 drbd_al_ext_cache = NULL;
2224 drbd_pp_pool = NULL;
2225 drbd_md_io_page_pool = NULL;
2226 drbd_md_io_bio_set = NULL;
2229 drbd_request_cache = kmem_cache_create(
2230 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2231 if (drbd_request_cache == NULL)
2234 drbd_ee_cache = kmem_cache_create(
2235 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2236 if (drbd_ee_cache == NULL)
2239 drbd_bm_ext_cache = kmem_cache_create(
2240 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2241 if (drbd_bm_ext_cache == NULL)
2244 drbd_al_ext_cache = kmem_cache_create(
2245 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2246 if (drbd_al_ext_cache == NULL)
2250 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2251 if (drbd_md_io_bio_set == NULL)
2254 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2255 if (drbd_md_io_page_pool == NULL)
2258 drbd_request_mempool = mempool_create(number,
2259 mempool_alloc_slab, mempool_free_slab, drbd_request_cache);
2260 if (drbd_request_mempool == NULL)
2263 drbd_ee_mempool = mempool_create(number,
2264 mempool_alloc_slab, mempool_free_slab, drbd_ee_cache);
2265 if (drbd_ee_mempool == NULL)
2268 /* drbd's page pool */
2269 spin_lock_init(&drbd_pp_lock);
2271 for (i = 0; i < number; i++) {
2272 page = alloc_page(GFP_HIGHUSER);
2275 set_page_private(page, (unsigned long)drbd_pp_pool);
2276 drbd_pp_pool = page;
2278 drbd_pp_vacant = number;
2283 drbd_destroy_mempools(); /* in case we allocated some */
2287 static int drbd_notify_sys(struct notifier_block *this, unsigned long code,
2290 /* just so we have it. you never know what interesting things we
2291 * might want to do here some day...
2297 static struct notifier_block drbd_notifier = {
2298 .notifier_call = drbd_notify_sys,
2301 static void drbd_release_all_peer_reqs(struct drbd_conf *mdev)
2305 rr = drbd_free_peer_reqs(mdev, &mdev->active_ee);
2307 dev_err(DEV, "%d EEs in active list found!\n", rr);
2309 rr = drbd_free_peer_reqs(mdev, &mdev->sync_ee);
2311 dev_err(DEV, "%d EEs in sync list found!\n", rr);
2313 rr = drbd_free_peer_reqs(mdev, &mdev->read_ee);
2315 dev_err(DEV, "%d EEs in read list found!\n", rr);
2317 rr = drbd_free_peer_reqs(mdev, &mdev->done_ee);
2319 dev_err(DEV, "%d EEs in done list found!\n", rr);
2321 rr = drbd_free_peer_reqs(mdev, &mdev->net_ee);
2323 dev_err(DEV, "%d EEs in net list found!\n", rr);
2326 /* caution. no locking. */
2327 void drbd_minor_destroy(struct kref *kref)
2329 struct drbd_conf *mdev = container_of(kref, struct drbd_conf, kref);
2330 struct drbd_tconn *tconn = mdev->tconn;
2332 del_timer_sync(&mdev->request_timer);
2334 /* paranoia asserts */
2335 D_ASSERT(mdev->open_cnt == 0);
2336 D_ASSERT(list_empty(&mdev->tconn->data.work.q));
2337 /* end paranoia asserts */
2339 /* cleanup stuff that may have been allocated during
2340 * device (re-)configuration or state changes */
2342 if (mdev->this_bdev)
2343 bdput(mdev->this_bdev);
2345 drbd_free_bc(mdev->ldev);
2348 drbd_release_all_peer_reqs(mdev);
2350 lc_destroy(mdev->act_log);
2351 lc_destroy(mdev->resync);
2353 kfree(mdev->p_uuid);
2354 /* mdev->p_uuid = NULL; */
2356 kfree(mdev->current_epoch);
2357 if (mdev->bitmap) /* should no longer be there. */
2358 drbd_bm_cleanup(mdev);
2359 __free_page(mdev->md_io_page);
2360 put_disk(mdev->vdisk);
2361 blk_cleanup_queue(mdev->rq_queue);
2362 kfree(mdev->rs_plan_s);
2365 kref_put(&tconn->kref, &conn_destroy);
2368 static void drbd_cleanup(void)
2371 struct drbd_conf *mdev;
2372 struct drbd_tconn *tconn, *tmp;
2374 unregister_reboot_notifier(&drbd_notifier);
2376 /* first remove proc,
2377 * drbdsetup uses it's presence to detect
2378 * whether DRBD is loaded.
2379 * If we would get stuck in proc removal,
2380 * but have netlink already deregistered,
2381 * some drbdsetup commands may wait forever
2385 remove_proc_entry("drbd", NULL);
2387 drbd_genl_unregister();
2389 idr_for_each_entry(&minors, mdev, i) {
2390 idr_remove(&minors, mdev_to_minor(mdev));
2391 idr_remove(&mdev->tconn->volumes, mdev->vnr);
2392 del_gendisk(mdev->vdisk);
2393 /* synchronize_rcu(); No other threads running at this point */
2394 kref_put(&mdev->kref, &drbd_minor_destroy);
2397 /* not _rcu since, no other updater anymore. Genl already unregistered */
2398 list_for_each_entry_safe(tconn, tmp, &drbd_tconns, all_tconn) {
2399 list_del(&tconn->all_tconn); /* not _rcu no proc, not other threads */
2400 /* synchronize_rcu(); */
2401 kref_put(&tconn->kref, &conn_destroy);
2404 drbd_destroy_mempools();
2405 unregister_blkdev(DRBD_MAJOR, "drbd");
2407 idr_destroy(&minors);
2409 printk(KERN_INFO "drbd: module cleanup done.\n");
2413 * drbd_congested() - Callback for pdflush
2414 * @congested_data: User data
2415 * @bdi_bits: Bits pdflush is currently interested in
2417 * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2419 static int drbd_congested(void *congested_data, int bdi_bits)
2421 struct drbd_conf *mdev = congested_data;
2422 struct request_queue *q;
2426 if (!may_inc_ap_bio(mdev)) {
2427 /* DRBD has frozen IO */
2433 if (get_ldev(mdev)) {
2434 q = bdev_get_queue(mdev->ldev->backing_bdev);
2435 r = bdi_congested(&q->backing_dev_info, bdi_bits);
2441 if (bdi_bits & (1 << BDI_async_congested) && test_bit(NET_CONGESTED, &mdev->tconn->flags)) {
2442 r |= (1 << BDI_async_congested);
2443 reason = reason == 'b' ? 'a' : 'n';
2447 mdev->congestion_reason = reason;
2451 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2453 sema_init(&wq->s, 0);
2454 spin_lock_init(&wq->q_lock);
2455 INIT_LIST_HEAD(&wq->q);
2458 struct drbd_tconn *conn_get_by_name(const char *name)
2460 struct drbd_tconn *tconn;
2462 if (!name || !name[0])
2466 list_for_each_entry_rcu(tconn, &drbd_tconns, all_tconn) {
2467 if (!strcmp(tconn->name, name)) {
2468 kref_get(&tconn->kref);
2478 struct drbd_tconn *conn_get_by_addrs(void *my_addr, int my_addr_len,
2479 void *peer_addr, int peer_addr_len)
2481 struct drbd_tconn *tconn;
2484 list_for_each_entry_rcu(tconn, &drbd_tconns, all_tconn) {
2485 if (tconn->my_addr_len == my_addr_len &&
2486 tconn->peer_addr_len == peer_addr_len &&
2487 !memcmp(&tconn->my_addr, my_addr, my_addr_len) &&
2488 !memcmp(&tconn->peer_addr, peer_addr, peer_addr_len)) {
2489 kref_get(&tconn->kref);
2499 static int drbd_alloc_socket(struct drbd_socket *socket)
2501 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2504 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2510 static void drbd_free_socket(struct drbd_socket *socket)
2512 free_page((unsigned long) socket->sbuf);
2513 free_page((unsigned long) socket->rbuf);
2516 void conn_free_crypto(struct drbd_tconn *tconn)
2518 drbd_free_sock(tconn);
2520 crypto_free_hash(tconn->csums_tfm);
2521 crypto_free_hash(tconn->verify_tfm);
2522 crypto_free_hash(tconn->cram_hmac_tfm);
2523 crypto_free_hash(tconn->integrity_tfm);
2524 crypto_free_hash(tconn->peer_integrity_tfm);
2525 kfree(tconn->int_dig_in);
2526 kfree(tconn->int_dig_vv);
2528 tconn->csums_tfm = NULL;
2529 tconn->verify_tfm = NULL;
2530 tconn->cram_hmac_tfm = NULL;
2531 tconn->integrity_tfm = NULL;
2532 tconn->peer_integrity_tfm = NULL;
2533 tconn->int_dig_in = NULL;
2534 tconn->int_dig_vv = NULL;
2537 int set_resource_options(struct drbd_tconn *tconn, struct res_opts *res_opts)
2539 cpumask_var_t new_cpu_mask;
2542 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2545 retcode = ERR_NOMEM;
2546 drbd_msg_put_info("unable to allocate cpumask");
2549 /* silently ignore cpu mask on UP kernel */
2550 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2551 /* FIXME: Get rid of constant 32 here */
2552 err = __bitmap_parse(res_opts->cpu_mask, 32, 0,
2553 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2555 conn_warn(tconn, "__bitmap_parse() failed with %d\n", err);
2556 /* retcode = ERR_CPU_MASK_PARSE; */
2560 tconn->res_opts = *res_opts;
2561 if (!cpumask_equal(tconn->cpu_mask, new_cpu_mask)) {
2562 cpumask_copy(tconn->cpu_mask, new_cpu_mask);
2563 drbd_calc_cpu_mask(tconn);
2564 tconn->receiver.reset_cpu_mask = 1;
2565 tconn->asender.reset_cpu_mask = 1;
2566 tconn->worker.reset_cpu_mask = 1;
2571 free_cpumask_var(new_cpu_mask);
2576 /* caller must be under genl_lock() */
2577 struct drbd_tconn *conn_create(const char *name, struct res_opts *res_opts)
2579 struct drbd_tconn *tconn;
2581 tconn = kzalloc(sizeof(struct drbd_tconn), GFP_KERNEL);
2585 tconn->name = kstrdup(name, GFP_KERNEL);
2589 if (drbd_alloc_socket(&tconn->data))
2591 if (drbd_alloc_socket(&tconn->meta))
2594 if (!zalloc_cpumask_var(&tconn->cpu_mask, GFP_KERNEL))
2597 if (set_resource_options(tconn, res_opts))
2600 if (!tl_init(tconn))
2603 tconn->cstate = C_STANDALONE;
2604 mutex_init(&tconn->cstate_mutex);
2605 spin_lock_init(&tconn->req_lock);
2606 mutex_init(&tconn->conf_update);
2607 init_waitqueue_head(&tconn->ping_wait);
2608 idr_init(&tconn->volumes);
2610 drbd_init_workqueue(&tconn->data.work);
2611 mutex_init(&tconn->data.mutex);
2613 drbd_init_workqueue(&tconn->meta.work);
2614 mutex_init(&tconn->meta.mutex);
2616 drbd_thread_init(tconn, &tconn->receiver, drbdd_init, "receiver");
2617 drbd_thread_init(tconn, &tconn->worker, drbd_worker, "worker");
2618 drbd_thread_init(tconn, &tconn->asender, drbd_asender, "asender");
2620 kref_init(&tconn->kref);
2621 list_add_tail_rcu(&tconn->all_tconn, &drbd_tconns);
2627 free_cpumask_var(tconn->cpu_mask);
2628 drbd_free_socket(&tconn->meta);
2629 drbd_free_socket(&tconn->data);
2636 void conn_destroy(struct kref *kref)
2638 struct drbd_tconn *tconn = container_of(kref, struct drbd_tconn, kref);
2640 idr_destroy(&tconn->volumes);
2642 free_cpumask_var(tconn->cpu_mask);
2643 drbd_free_socket(&tconn->meta);
2644 drbd_free_socket(&tconn->data);
2646 kfree(tconn->int_dig_in);
2647 kfree(tconn->int_dig_vv);
2651 enum drbd_ret_code conn_new_minor(struct drbd_tconn *tconn, unsigned int minor, int vnr)
2653 struct drbd_conf *mdev;
2654 struct gendisk *disk;
2655 struct request_queue *q;
2657 int minor_got = minor;
2658 enum drbd_ret_code err = ERR_NOMEM;
2660 mdev = minor_to_mdev(minor);
2662 return ERR_MINOR_EXISTS;
2664 /* GFP_KERNEL, we are outside of all write-out paths */
2665 mdev = kzalloc(sizeof(struct drbd_conf), GFP_KERNEL);
2669 kref_get(&tconn->kref);
2670 mdev->tconn = tconn;
2672 mdev->minor = minor;
2675 drbd_init_set_defaults(mdev);
2677 q = blk_alloc_queue(GFP_KERNEL);
2681 q->queuedata = mdev;
2683 disk = alloc_disk(1);
2688 set_disk_ro(disk, true);
2691 disk->major = DRBD_MAJOR;
2692 disk->first_minor = minor;
2693 disk->fops = &drbd_ops;
2694 sprintf(disk->disk_name, "drbd%d", minor);
2695 disk->private_data = mdev;
2697 mdev->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2698 /* we have no partitions. we contain only ourselves. */
2699 mdev->this_bdev->bd_contains = mdev->this_bdev;
2701 q->backing_dev_info.congested_fn = drbd_congested;
2702 q->backing_dev_info.congested_data = mdev;
2704 blk_queue_make_request(q, drbd_make_request);
2705 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2706 This triggers a max_bio_size message upon first attach or connect */
2707 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2708 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2709 blk_queue_merge_bvec(q, drbd_merge_bvec);
2710 q->queue_lock = &mdev->tconn->req_lock; /* needed since we use */
2712 mdev->md_io_page = alloc_page(GFP_KERNEL);
2713 if (!mdev->md_io_page)
2714 goto out_no_io_page;
2716 if (drbd_bm_init(mdev))
2718 mdev->read_requests = RB_ROOT;
2719 mdev->write_requests = RB_ROOT;
2721 mdev->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2722 if (!mdev->current_epoch)
2725 INIT_LIST_HEAD(&mdev->current_epoch->list);
2728 if (!idr_pre_get(&minors, GFP_KERNEL))
2729 goto out_no_minor_idr;
2730 if (idr_get_new_above(&minors, mdev, minor, &minor_got))
2731 goto out_no_minor_idr;
2732 if (minor_got != minor) {
2733 err = ERR_MINOR_EXISTS;
2734 drbd_msg_put_info("requested minor exists already");
2735 goto out_idr_remove_minor;
2738 if (!idr_pre_get(&tconn->volumes, GFP_KERNEL))
2739 goto out_idr_remove_minor;
2740 if (idr_get_new_above(&tconn->volumes, mdev, vnr, &vnr_got))
2741 goto out_idr_remove_minor;
2742 if (vnr_got != vnr) {
2743 err = ERR_INVALID_REQUEST;
2744 drbd_msg_put_info("requested volume exists already");
2745 goto out_idr_remove_vol;
2748 kref_init(&mdev->kref); /* one ref for both idrs and the the add_disk */
2750 /* inherit the connection state */
2751 mdev->state.conn = tconn->cstate;
2752 if (mdev->state.conn == C_WF_REPORT_PARAMS)
2753 drbd_connected(mdev);
2758 idr_remove(&tconn->volumes, vnr_got);
2759 out_idr_remove_minor:
2760 idr_remove(&minors, minor_got);
2763 kfree(mdev->current_epoch);
2765 drbd_bm_cleanup(mdev);
2767 __free_page(mdev->md_io_page);
2771 blk_cleanup_queue(q);
2774 kref_put(&tconn->kref, &conn_destroy);
2778 int __init drbd_init(void)
2782 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2784 "drbd: invalid minor_count (%d)\n", minor_count);
2788 minor_count = DRBD_MINOR_COUNT_DEF;
2792 err = register_blkdev(DRBD_MAJOR, "drbd");
2795 "drbd: unable to register block device major %d\n",
2800 err = drbd_genl_register();
2802 printk(KERN_ERR "drbd: unable to register generic netlink family\n");
2807 register_reboot_notifier(&drbd_notifier);
2810 * allocate all necessary structs
2814 init_waitqueue_head(&drbd_pp_wait);
2816 drbd_proc = NULL; /* play safe for drbd_cleanup */
2819 err = drbd_create_mempools();
2823 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
2825 printk(KERN_ERR "drbd: unable to register proc file\n");
2829 rwlock_init(&global_state_lock);
2830 INIT_LIST_HEAD(&drbd_tconns);
2832 printk(KERN_INFO "drbd: initialized. "
2833 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2834 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2835 printk(KERN_INFO "drbd: %s\n", drbd_buildtag());
2836 printk(KERN_INFO "drbd: registered as block device major %d\n",
2839 return 0; /* Success! */
2844 /* currently always the case */
2845 printk(KERN_ERR "drbd: ran out of memory\n");
2847 printk(KERN_ERR "drbd: initialization failure\n");
2851 void drbd_free_bc(struct drbd_backing_dev *ldev)
2856 blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2857 blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2862 void drbd_free_sock(struct drbd_tconn *tconn)
2864 if (tconn->data.socket) {
2865 mutex_lock(&tconn->data.mutex);
2866 kernel_sock_shutdown(tconn->data.socket, SHUT_RDWR);
2867 sock_release(tconn->data.socket);
2868 tconn->data.socket = NULL;
2869 mutex_unlock(&tconn->data.mutex);
2871 if (tconn->meta.socket) {
2872 mutex_lock(&tconn->meta.mutex);
2873 kernel_sock_shutdown(tconn->meta.socket, SHUT_RDWR);
2874 sock_release(tconn->meta.socket);
2875 tconn->meta.socket = NULL;
2876 mutex_unlock(&tconn->meta.mutex);
2880 /* meta data management */
2882 struct meta_data_on_disk {
2883 u64 la_size; /* last agreed size. */
2884 u64 uuid[UI_SIZE]; /* UUIDs. */
2887 u32 flags; /* MDF */
2890 u32 al_offset; /* offset to this block */
2891 u32 al_nr_extents; /* important for restoring the AL */
2892 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
2893 u32 bm_offset; /* offset to the bitmap, from here */
2894 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
2895 u32 la_peer_max_bio_size; /* last peer max_bio_size */
2896 u32 reserved_u32[3];
2901 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
2902 * @mdev: DRBD device.
2904 void drbd_md_sync(struct drbd_conf *mdev)
2906 struct meta_data_on_disk *buffer;
2910 del_timer(&mdev->md_sync_timer);
2911 /* timer may be rearmed by drbd_md_mark_dirty() now. */
2912 if (!test_and_clear_bit(MD_DIRTY, &mdev->flags))
2915 /* We use here D_FAILED and not D_ATTACHING because we try to write
2916 * metadata even if we detach due to a disk failure! */
2917 if (!get_ldev_if_state(mdev, D_FAILED))
2920 buffer = drbd_md_get_buffer(mdev);
2924 memset(buffer, 0, 512);
2926 buffer->la_size = cpu_to_be64(drbd_get_capacity(mdev->this_bdev));
2927 for (i = UI_CURRENT; i < UI_SIZE; i++)
2928 buffer->uuid[i] = cpu_to_be64(mdev->ldev->md.uuid[i]);
2929 buffer->flags = cpu_to_be32(mdev->ldev->md.flags);
2930 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
2932 buffer->md_size_sect = cpu_to_be32(mdev->ldev->md.md_size_sect);
2933 buffer->al_offset = cpu_to_be32(mdev->ldev->md.al_offset);
2934 buffer->al_nr_extents = cpu_to_be32(mdev->act_log->nr_elements);
2935 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
2936 buffer->device_uuid = cpu_to_be64(mdev->ldev->md.device_uuid);
2938 buffer->bm_offset = cpu_to_be32(mdev->ldev->md.bm_offset);
2939 buffer->la_peer_max_bio_size = cpu_to_be32(mdev->peer_max_bio_size);
2941 D_ASSERT(drbd_md_ss__(mdev, mdev->ldev) == mdev->ldev->md.md_offset);
2942 sector = mdev->ldev->md.md_offset;
2944 if (drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) {
2945 /* this was a try anyways ... */
2946 dev_err(DEV, "meta data update failed!\n");
2947 drbd_chk_io_error(mdev, 1, true);
2950 /* Update mdev->ldev->md.la_size_sect,
2951 * since we updated it on metadata. */
2952 mdev->ldev->md.la_size_sect = drbd_get_capacity(mdev->this_bdev);
2954 drbd_md_put_buffer(mdev);
2960 * drbd_md_read() - Reads in the meta data super block
2961 * @mdev: DRBD device.
2962 * @bdev: Device from which the meta data should be read in.
2964 * Return 0 (NO_ERROR) on success, and an enum drbd_ret_code in case
2965 * something goes wrong.
2967 int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
2969 struct meta_data_on_disk *buffer;
2971 int i, rv = NO_ERROR;
2973 if (!get_ldev_if_state(mdev, D_ATTACHING))
2974 return ERR_IO_MD_DISK;
2976 buffer = drbd_md_get_buffer(mdev);
2980 if (drbd_md_sync_page_io(mdev, bdev, bdev->md.md_offset, READ)) {
2981 /* NOTE: can't do normal error processing here as this is
2982 called BEFORE disk is attached */
2983 dev_err(DEV, "Error while reading metadata.\n");
2984 rv = ERR_IO_MD_DISK;
2988 magic = be32_to_cpu(buffer->magic);
2989 flags = be32_to_cpu(buffer->flags);
2990 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
2991 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
2992 /* btw: that's Activity Log clean, not "all" clean. */
2993 dev_err(DEV, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
2994 rv = ERR_MD_UNCLEAN;
2997 if (magic != DRBD_MD_MAGIC_08) {
2998 if (magic == DRBD_MD_MAGIC_07)
2999 dev_err(DEV, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3001 dev_err(DEV, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3002 rv = ERR_MD_INVALID;
3005 if (be32_to_cpu(buffer->al_offset) != bdev->md.al_offset) {
3006 dev_err(DEV, "unexpected al_offset: %d (expected %d)\n",
3007 be32_to_cpu(buffer->al_offset), bdev->md.al_offset);
3008 rv = ERR_MD_INVALID;
3011 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3012 dev_err(DEV, "unexpected bm_offset: %d (expected %d)\n",
3013 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3014 rv = ERR_MD_INVALID;
3017 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3018 dev_err(DEV, "unexpected md_size: %u (expected %u)\n",
3019 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3020 rv = ERR_MD_INVALID;
3024 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3025 dev_err(DEV, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3026 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3027 rv = ERR_MD_INVALID;
3031 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size);
3032 for (i = UI_CURRENT; i < UI_SIZE; i++)
3033 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3034 bdev->md.flags = be32_to_cpu(buffer->flags);
3035 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3037 spin_lock_irq(&mdev->tconn->req_lock);
3038 if (mdev->state.conn < C_CONNECTED) {
3040 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3041 peer = max_t(int, peer, DRBD_MAX_BIO_SIZE_SAFE);
3042 mdev->peer_max_bio_size = peer;
3044 spin_unlock_irq(&mdev->tconn->req_lock);
3047 drbd_md_put_buffer(mdev);
3055 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3056 * @mdev: DRBD device.
3058 * Call this function if you change anything that should be written to
3059 * the meta-data super block. This function sets MD_DIRTY, and starts a
3060 * timer that ensures that within five seconds you have to call drbd_md_sync().
3063 void drbd_md_mark_dirty_(struct drbd_conf *mdev, unsigned int line, const char *func)
3065 if (!test_and_set_bit(MD_DIRTY, &mdev->flags)) {
3066 mod_timer(&mdev->md_sync_timer, jiffies + HZ);
3067 mdev->last_md_mark_dirty.line = line;
3068 mdev->last_md_mark_dirty.func = func;
3072 void drbd_md_mark_dirty(struct drbd_conf *mdev)
3074 if (!test_and_set_bit(MD_DIRTY, &mdev->flags))
3075 mod_timer(&mdev->md_sync_timer, jiffies + 5*HZ);
3079 static void drbd_uuid_move_history(struct drbd_conf *mdev) __must_hold(local)
3083 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3084 mdev->ldev->md.uuid[i+1] = mdev->ldev->md.uuid[i];
3087 void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
3089 if (idx == UI_CURRENT) {
3090 if (mdev->state.role == R_PRIMARY)
3095 drbd_set_ed_uuid(mdev, val);
3098 mdev->ldev->md.uuid[idx] = val;
3099 drbd_md_mark_dirty(mdev);
3103 void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local)
3105 if (mdev->ldev->md.uuid[idx]) {
3106 drbd_uuid_move_history(mdev);
3107 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[idx];
3109 _drbd_uuid_set(mdev, idx, val);
3113 * drbd_uuid_new_current() - Creates a new current UUID
3114 * @mdev: DRBD device.
3116 * Creates a new current UUID, and rotates the old current UUID into
3117 * the bitmap slot. Causes an incremental resync upon next connect.
3119 void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local)
3122 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
3125 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
3127 mdev->ldev->md.uuid[UI_BITMAP] = mdev->ldev->md.uuid[UI_CURRENT];
3129 get_random_bytes(&val, sizeof(u64));
3130 _drbd_uuid_set(mdev, UI_CURRENT, val);
3131 drbd_print_uuids(mdev, "new current UUID");
3132 /* get it to stable storage _now_ */
3136 void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local)
3138 if (mdev->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3142 drbd_uuid_move_history(mdev);
3143 mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP];
3144 mdev->ldev->md.uuid[UI_BITMAP] = 0;
3146 unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP];
3148 dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid);
3150 mdev->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3152 drbd_md_mark_dirty(mdev);
3156 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3157 * @mdev: DRBD device.
3159 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3161 int drbd_bmio_set_n_write(struct drbd_conf *mdev)
3165 if (get_ldev_if_state(mdev, D_ATTACHING)) {
3166 drbd_md_set_flag(mdev, MDF_FULL_SYNC);
3168 drbd_bm_set_all(mdev);
3170 rv = drbd_bm_write(mdev);
3173 drbd_md_clear_flag(mdev, MDF_FULL_SYNC);
3184 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3185 * @mdev: DRBD device.
3187 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3189 int drbd_bmio_clear_n_write(struct drbd_conf *mdev)
3193 drbd_resume_al(mdev);
3194 if (get_ldev_if_state(mdev, D_ATTACHING)) {
3195 drbd_bm_clear_all(mdev);
3196 rv = drbd_bm_write(mdev);
3203 static int w_bitmap_io(struct drbd_work *w, int unused)
3205 struct bm_io_work *work = container_of(w, struct bm_io_work, w);
3206 struct drbd_conf *mdev = w->mdev;
3209 D_ASSERT(atomic_read(&mdev->ap_bio_cnt) == 0);
3211 if (get_ldev(mdev)) {
3212 drbd_bm_lock(mdev, work->why, work->flags);
3213 rv = work->io_fn(mdev);
3214 drbd_bm_unlock(mdev);
3218 clear_bit_unlock(BITMAP_IO, &mdev->flags);
3219 wake_up(&mdev->misc_wait);
3222 work->done(mdev, rv);
3224 clear_bit(BITMAP_IO_QUEUED, &mdev->flags);
3231 void drbd_ldev_destroy(struct drbd_conf *mdev)
3233 lc_destroy(mdev->resync);
3234 mdev->resync = NULL;
3235 lc_destroy(mdev->act_log);
3236 mdev->act_log = NULL;
3238 drbd_free_bc(mdev->ldev);
3239 mdev->ldev = NULL;);
3241 clear_bit(GO_DISKLESS, &mdev->flags);
3244 static int w_go_diskless(struct drbd_work *w, int unused)
3246 struct drbd_conf *mdev = w->mdev;
3248 D_ASSERT(mdev->state.disk == D_FAILED);
3249 /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will
3250 * inc/dec it frequently. Once we are D_DISKLESS, no one will touch
3251 * the protected members anymore, though, so once put_ldev reaches zero
3252 * again, it will be safe to free them. */
3253 drbd_force_state(mdev, NS(disk, D_DISKLESS));
3257 void drbd_go_diskless(struct drbd_conf *mdev)
3259 D_ASSERT(mdev->state.disk == D_FAILED);
3260 if (!test_and_set_bit(GO_DISKLESS, &mdev->flags))
3261 drbd_queue_work(&mdev->tconn->data.work, &mdev->go_diskless);
3265 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3266 * @mdev: DRBD device.
3267 * @io_fn: IO callback to be called when bitmap IO is possible
3268 * @done: callback to be called after the bitmap IO was performed
3269 * @why: Descriptive text of the reason for doing the IO
3271 * While IO on the bitmap happens we freeze application IO thus we ensure
3272 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3273 * called from worker context. It MUST NOT be used while a previous such
3274 * work is still pending!
3276 void drbd_queue_bitmap_io(struct drbd_conf *mdev,
3277 int (*io_fn)(struct drbd_conf *),
3278 void (*done)(struct drbd_conf *, int),
3279 char *why, enum bm_flag flags)
3281 D_ASSERT(current == mdev->tconn->worker.task);
3283 D_ASSERT(!test_bit(BITMAP_IO_QUEUED, &mdev->flags));
3284 D_ASSERT(!test_bit(BITMAP_IO, &mdev->flags));
3285 D_ASSERT(list_empty(&mdev->bm_io_work.w.list));
3286 if (mdev->bm_io_work.why)
3287 dev_err(DEV, "FIXME going to queue '%s' but '%s' still pending?\n",
3288 why, mdev->bm_io_work.why);
3290 mdev->bm_io_work.io_fn = io_fn;
3291 mdev->bm_io_work.done = done;
3292 mdev->bm_io_work.why = why;
3293 mdev->bm_io_work.flags = flags;
3295 spin_lock_irq(&mdev->tconn->req_lock);
3296 set_bit(BITMAP_IO, &mdev->flags);
3297 if (atomic_read(&mdev->ap_bio_cnt) == 0) {
3298 if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags))
3299 drbd_queue_work(&mdev->tconn->data.work, &mdev->bm_io_work.w);
3301 spin_unlock_irq(&mdev->tconn->req_lock);
3305 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3306 * @mdev: DRBD device.
3307 * @io_fn: IO callback to be called when bitmap IO is possible
3308 * @why: Descriptive text of the reason for doing the IO
3310 * freezes application IO while that the actual IO operations runs. This
3311 * functions MAY NOT be called from worker context.
3313 int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *),
3314 char *why, enum bm_flag flags)
3318 D_ASSERT(current != mdev->tconn->worker.task);
3320 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3321 drbd_suspend_io(mdev);
3323 drbd_bm_lock(mdev, why, flags);
3325 drbd_bm_unlock(mdev);
3327 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3328 drbd_resume_io(mdev);
3333 void drbd_md_set_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3335 if ((mdev->ldev->md.flags & flag) != flag) {
3336 drbd_md_mark_dirty(mdev);
3337 mdev->ldev->md.flags |= flag;
3341 void drbd_md_clear_flag(struct drbd_conf *mdev, int flag) __must_hold(local)
3343 if ((mdev->ldev->md.flags & flag) != 0) {
3344 drbd_md_mark_dirty(mdev);
3345 mdev->ldev->md.flags &= ~flag;
3348 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3350 return (bdev->md.flags & flag) != 0;
3353 static void md_sync_timer_fn(unsigned long data)
3355 struct drbd_conf *mdev = (struct drbd_conf *) data;
3357 drbd_queue_work_front(&mdev->tconn->data.work, &mdev->md_sync_work);
3360 static int w_md_sync(struct drbd_work *w, int unused)
3362 struct drbd_conf *mdev = w->mdev;
3364 dev_warn(DEV, "md_sync_timer expired! Worker calls drbd_md_sync().\n");
3366 dev_warn(DEV, "last md_mark_dirty: %s:%u\n",
3367 mdev->last_md_mark_dirty.func, mdev->last_md_mark_dirty.line);
3373 const char *cmdname(enum drbd_packet cmd)
3375 /* THINK may need to become several global tables
3376 * when we want to support more than
3377 * one PRO_VERSION */
3378 static const char *cmdnames[] = {
3380 [P_DATA_REPLY] = "DataReply",
3381 [P_RS_DATA_REPLY] = "RSDataReply",
3382 [P_BARRIER] = "Barrier",
3383 [P_BITMAP] = "ReportBitMap",
3384 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3385 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3386 [P_UNPLUG_REMOTE] = "UnplugRemote",
3387 [P_DATA_REQUEST] = "DataRequest",
3388 [P_RS_DATA_REQUEST] = "RSDataRequest",
3389 [P_SYNC_PARAM] = "SyncParam",
3390 [P_SYNC_PARAM89] = "SyncParam89",
3391 [P_PROTOCOL] = "ReportProtocol",
3392 [P_UUIDS] = "ReportUUIDs",
3393 [P_SIZES] = "ReportSizes",
3394 [P_STATE] = "ReportState",
3395 [P_SYNC_UUID] = "ReportSyncUUID",
3396 [P_AUTH_CHALLENGE] = "AuthChallenge",
3397 [P_AUTH_RESPONSE] = "AuthResponse",
3399 [P_PING_ACK] = "PingAck",
3400 [P_RECV_ACK] = "RecvAck",
3401 [P_WRITE_ACK] = "WriteAck",
3402 [P_RS_WRITE_ACK] = "RSWriteAck",
3403 [P_DISCARD_WRITE] = "DiscardWrite",
3404 [P_NEG_ACK] = "NegAck",
3405 [P_NEG_DREPLY] = "NegDReply",
3406 [P_NEG_RS_DREPLY] = "NegRSDReply",
3407 [P_BARRIER_ACK] = "BarrierAck",
3408 [P_STATE_CHG_REQ] = "StateChgRequest",
3409 [P_STATE_CHG_REPLY] = "StateChgReply",
3410 [P_OV_REQUEST] = "OVRequest",
3411 [P_OV_REPLY] = "OVReply",
3412 [P_OV_RESULT] = "OVResult",
3413 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3414 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3415 [P_COMPRESSED_BITMAP] = "CBitmap",
3416 [P_DELAY_PROBE] = "DelayProbe",
3417 [P_OUT_OF_SYNC] = "OutOfSync",
3418 [P_RETRY_WRITE] = "RetryWrite",
3419 [P_RS_CANCEL] = "RSCancel",
3420 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3421 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3422 [P_RETRY_WRITE] = "retry_write",
3423 [P_PROTOCOL_UPDATE] = "protocol_update",
3425 /* enum drbd_packet, but not commands - obsoleted flags:
3431 /* too big for the array: 0xfffX */
3432 if (cmd == P_INITIAL_META)
3433 return "InitialMeta";
3434 if (cmd == P_INITIAL_DATA)
3435 return "InitialData";
3436 if (cmd == P_CONNECTION_FEATURES)
3437 return "ConnectionFeatures";
3438 if (cmd >= ARRAY_SIZE(cmdnames))
3440 return cmdnames[cmd];
3444 * drbd_wait_misc - wait for a request to make progress
3445 * @mdev: device associated with the request
3446 * @i: the struct drbd_interval embedded in struct drbd_request or
3447 * struct drbd_peer_request
3449 int drbd_wait_misc(struct drbd_conf *mdev, struct drbd_interval *i)
3451 struct net_conf *nc;
3456 nc = rcu_dereference(mdev->tconn->net_conf);
3461 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3464 /* Indicate to wake up mdev->misc_wait on progress. */
3466 prepare_to_wait(&mdev->misc_wait, &wait, TASK_INTERRUPTIBLE);
3467 spin_unlock_irq(&mdev->tconn->req_lock);
3468 timeout = schedule_timeout(timeout);
3469 finish_wait(&mdev->misc_wait, &wait);
3470 spin_lock_irq(&mdev->tconn->req_lock);
3471 if (!timeout || mdev->state.conn < C_CONNECTED)
3473 if (signal_pending(current))
3474 return -ERESTARTSYS;
3478 #ifdef CONFIG_DRBD_FAULT_INJECTION
3479 /* Fault insertion support including random number generator shamelessly
3480 * stolen from kernel/rcutorture.c */
3481 struct fault_random_state {
3482 unsigned long state;
3483 unsigned long count;
3486 #define FAULT_RANDOM_MULT 39916801 /* prime */
3487 #define FAULT_RANDOM_ADD 479001701 /* prime */
3488 #define FAULT_RANDOM_REFRESH 10000
3491 * Crude but fast random-number generator. Uses a linear congruential
3492 * generator, with occasional help from get_random_bytes().
3494 static unsigned long
3495 _drbd_fault_random(struct fault_random_state *rsp)
3499 if (!rsp->count--) {
3500 get_random_bytes(&refresh, sizeof(refresh));
3501 rsp->state += refresh;
3502 rsp->count = FAULT_RANDOM_REFRESH;
3504 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3505 return swahw32(rsp->state);
3509 _drbd_fault_str(unsigned int type) {
3510 static char *_faults[] = {
3511 [DRBD_FAULT_MD_WR] = "Meta-data write",
3512 [DRBD_FAULT_MD_RD] = "Meta-data read",
3513 [DRBD_FAULT_RS_WR] = "Resync write",
3514 [DRBD_FAULT_RS_RD] = "Resync read",
3515 [DRBD_FAULT_DT_WR] = "Data write",
3516 [DRBD_FAULT_DT_RD] = "Data read",
3517 [DRBD_FAULT_DT_RA] = "Data read ahead",
3518 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3519 [DRBD_FAULT_AL_EE] = "EE allocation",
3520 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3523 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3527 _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type)
3529 static struct fault_random_state rrs = {0, 0};
3531 unsigned int ret = (
3533 ((1 << mdev_to_minor(mdev)) & fault_devs) != 0) &&
3534 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3539 if (__ratelimit(&drbd_ratelimit_state))
3540 dev_warn(DEV, "***Simulating %s failure\n",
3541 _drbd_fault_str(type));
3548 const char *drbd_buildtag(void)
3550 /* DRBD built from external sources has here a reference to the
3551 git hash of the source code. */
3553 static char buildtag[38] = "\0uilt-in";
3555 if (buildtag[0] == 0) {
3556 #ifdef CONFIG_MODULES
3557 if (THIS_MODULE != NULL)
3558 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3567 module_init(drbd_init)
3568 module_exit(drbd_cleanup)
3570 EXPORT_SYMBOL(drbd_conn_str);
3571 EXPORT_SYMBOL(drbd_role_str);
3572 EXPORT_SYMBOL(drbd_disk_str);
3573 EXPORT_SYMBOL(drbd_set_st_err_str);