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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/module.h>
32 #include <linux/jiffies.h>
33 #include <linux/drbd.h>
34 #include <linux/uaccess.h>
35 #include <asm/types.h>
37 #include <linux/ctype.h>
38 #include <linux/mutex.h>
40 #include <linux/file.h>
41 #include <linux/proc_fs.h>
42 #include <linux/init.h>
44 #include <linux/memcontrol.h>
45 #include <linux/mm_inline.h>
46 #include <linux/slab.h>
47 #include <linux/random.h>
48 #include <linux/reboot.h>
49 #include <linux/notifier.h>
50 #include <linux/kthread.h>
51 #include <linux/workqueue.h>
52 #define __KERNEL_SYSCALLS__
53 #include <linux/unistd.h>
54 #include <linux/vmalloc.h>
55 #include <linux/sched/signal.h>
57 #include <linux/drbd_limits.h>
59 #include "drbd_protocol.h"
60 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
62 #include "drbd_debugfs.h"
64 static DEFINE_MUTEX(drbd_main_mutex);
65 static int drbd_open(struct block_device *bdev, fmode_t mode);
66 static void drbd_release(struct gendisk *gd, fmode_t mode);
67 static void md_sync_timer_fn(struct timer_list *t);
68 static int w_bitmap_io(struct drbd_work *w, int unused);
70 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
71 "Lars Ellenberg <lars@linbit.com>");
72 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
73 MODULE_VERSION(REL_VERSION);
74 MODULE_LICENSE("GPL");
75 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
76 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
77 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
79 #include <linux/moduleparam.h>
80 /* thanks to these macros, if compiled into the kernel (not-module),
81 * these become boot parameters (e.g., drbd.minor_count) */
83 #ifdef CONFIG_DRBD_FAULT_INJECTION
84 int drbd_enable_faults;
86 static int drbd_fault_count;
87 static int drbd_fault_devs;
88 /* bitmap of enabled faults */
89 module_param_named(enable_faults, drbd_enable_faults, int, 0664);
90 /* fault rate % value - applies to all enabled faults */
91 module_param_named(fault_rate, drbd_fault_rate, int, 0664);
92 /* count of faults inserted */
93 module_param_named(fault_count, drbd_fault_count, int, 0664);
94 /* bitmap of devices to insert faults on */
95 module_param_named(fault_devs, drbd_fault_devs, int, 0644);
98 /* module parameters we can keep static */
99 static bool drbd_allow_oos; /* allow_open_on_secondary */
100 static bool drbd_disable_sendpage;
101 MODULE_PARM_DESC(allow_oos, "DONT USE!");
102 module_param_named(allow_oos, drbd_allow_oos, bool, 0);
103 module_param_named(disable_sendpage, drbd_disable_sendpage, bool, 0644);
105 /* module parameters we share */
106 int drbd_proc_details; /* Detail level in proc drbd*/
107 module_param_named(proc_details, drbd_proc_details, int, 0644);
108 /* module parameters shared with defaults */
109 unsigned int drbd_minor_count = DRBD_MINOR_COUNT_DEF;
110 /* Module parameter for setting the user mode helper program
111 * to run. Default is /sbin/drbdadm */
112 char drbd_usermode_helper[80] = "/sbin/drbdadm";
113 module_param_named(minor_count, drbd_minor_count, uint, 0444);
114 module_param_string(usermode_helper, drbd_usermode_helper, sizeof(drbd_usermode_helper), 0644);
116 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
117 * as member "struct gendisk *vdisk;"
119 struct idr drbd_devices;
120 struct list_head drbd_resources;
121 struct mutex resources_mutex;
123 struct kmem_cache *drbd_request_cache;
124 struct kmem_cache *drbd_ee_cache; /* peer requests */
125 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
126 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
127 mempool_t drbd_request_mempool;
128 mempool_t drbd_ee_mempool;
129 mempool_t drbd_md_io_page_pool;
130 struct bio_set drbd_md_io_bio_set;
131 struct bio_set drbd_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 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
156 if (!bioset_initialized(&drbd_md_io_bio_set))
157 return bio_alloc(gfp_mask, 1);
159 bio = bio_alloc_bioset(gfp_mask, 1, &drbd_md_io_bio_set);
166 /* When checking with sparse, and this is an inline function, sparse will
167 give tons of false positives. When this is a real functions sparse works.
169 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
173 atomic_inc(&device->local_cnt);
174 io_allowed = (device->state.disk >= mins);
176 if (atomic_dec_and_test(&device->local_cnt))
177 wake_up(&device->misc_wait);
185 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
186 * @connection: DRBD connection.
187 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
188 * @set_size: Expected number of requests before that barrier.
190 * In case the passed barrier_nr or set_size does not match the oldest
191 * epoch of not yet barrier-acked requests, this function will cause a
192 * termination of the connection.
194 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
195 unsigned int set_size)
197 struct drbd_request *r;
198 struct drbd_request *req = NULL;
199 int expect_epoch = 0;
202 spin_lock_irq(&connection->resource->req_lock);
204 /* find oldest not yet barrier-acked write request,
205 * count writes in its epoch. */
206 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
207 const unsigned s = r->rq_state;
211 if (!(s & RQ_NET_MASK))
216 expect_epoch = req->epoch;
219 if (r->epoch != expect_epoch)
223 /* if (s & RQ_DONE): not expected */
224 /* if (!(s & RQ_NET_MASK)): not expected */
229 /* first some paranoia code */
231 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
235 if (expect_epoch != barrier_nr) {
236 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
237 barrier_nr, expect_epoch);
241 if (expect_size != set_size) {
242 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
243 barrier_nr, set_size, expect_size);
247 /* Clean up list of requests processed during current epoch. */
248 /* this extra list walk restart is paranoia,
249 * to catch requests being barrier-acked "unexpectedly".
250 * It usually should find the same req again, or some READ preceding it. */
251 list_for_each_entry(req, &connection->transfer_log, tl_requests)
252 if (req->epoch == expect_epoch)
254 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
255 if (req->epoch != expect_epoch)
257 _req_mod(req, BARRIER_ACKED);
259 spin_unlock_irq(&connection->resource->req_lock);
264 spin_unlock_irq(&connection->resource->req_lock);
265 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
270 * _tl_restart() - Walks the transfer log, and applies an action to all requests
271 * @connection: DRBD connection to operate on.
272 * @what: The action/event to perform with all request objects
274 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
275 * RESTART_FROZEN_DISK_IO.
277 /* must hold resource->req_lock */
278 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
280 struct drbd_request *req, *r;
282 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
286 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
288 spin_lock_irq(&connection->resource->req_lock);
289 _tl_restart(connection, what);
290 spin_unlock_irq(&connection->resource->req_lock);
294 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
295 * @device: DRBD device.
297 * This is called after the connection to the peer was lost. The storage covered
298 * by the requests on the transfer gets marked as our of sync. Called from the
299 * receiver thread and the worker thread.
301 void tl_clear(struct drbd_connection *connection)
303 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
307 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
308 * @device: DRBD device.
310 void tl_abort_disk_io(struct drbd_device *device)
312 struct drbd_connection *connection = first_peer_device(device)->connection;
313 struct drbd_request *req, *r;
315 spin_lock_irq(&connection->resource->req_lock);
316 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
317 if (!(req->rq_state & RQ_LOCAL_PENDING))
319 if (req->device != device)
321 _req_mod(req, ABORT_DISK_IO);
323 spin_unlock_irq(&connection->resource->req_lock);
326 static int drbd_thread_setup(void *arg)
328 struct drbd_thread *thi = (struct drbd_thread *) arg;
329 struct drbd_resource *resource = thi->resource;
333 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
338 retval = thi->function(thi);
340 spin_lock_irqsave(&thi->t_lock, flags);
342 /* if the receiver has been "EXITING", the last thing it did
343 * was set the conn state to "StandAlone",
344 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
345 * and receiver thread will be "started".
346 * drbd_thread_start needs to set "RESTARTING" in that case.
347 * t_state check and assignment needs to be within the same spinlock,
348 * so either thread_start sees EXITING, and can remap to RESTARTING,
349 * or thread_start see NONE, and can proceed as normal.
352 if (thi->t_state == RESTARTING) {
353 drbd_info(resource, "Restarting %s thread\n", thi->name);
354 thi->t_state = RUNNING;
355 spin_unlock_irqrestore(&thi->t_lock, flags);
362 complete_all(&thi->stop);
363 spin_unlock_irqrestore(&thi->t_lock, flags);
365 drbd_info(resource, "Terminating %s\n", current->comm);
367 /* Release mod reference taken when thread was started */
370 kref_put(&thi->connection->kref, drbd_destroy_connection);
371 kref_put(&resource->kref, drbd_destroy_resource);
372 module_put(THIS_MODULE);
376 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
377 int (*func) (struct drbd_thread *), const char *name)
379 spin_lock_init(&thi->t_lock);
382 thi->function = func;
383 thi->resource = resource;
384 thi->connection = NULL;
388 int drbd_thread_start(struct drbd_thread *thi)
390 struct drbd_resource *resource = thi->resource;
391 struct task_struct *nt;
394 /* is used from state engine doing drbd_thread_stop_nowait,
395 * while holding the req lock irqsave */
396 spin_lock_irqsave(&thi->t_lock, flags);
398 switch (thi->t_state) {
400 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
401 thi->name, current->comm, current->pid);
403 /* Get ref on module for thread - this is released when thread exits */
404 if (!try_module_get(THIS_MODULE)) {
405 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
406 spin_unlock_irqrestore(&thi->t_lock, flags);
410 kref_get(&resource->kref);
412 kref_get(&thi->connection->kref);
414 init_completion(&thi->stop);
415 thi->reset_cpu_mask = 1;
416 thi->t_state = RUNNING;
417 spin_unlock_irqrestore(&thi->t_lock, flags);
418 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
420 nt = kthread_create(drbd_thread_setup, (void *) thi,
421 "drbd_%c_%s", thi->name[0], thi->resource->name);
424 drbd_err(resource, "Couldn't start thread\n");
427 kref_put(&thi->connection->kref, drbd_destroy_connection);
428 kref_put(&resource->kref, drbd_destroy_resource);
429 module_put(THIS_MODULE);
432 spin_lock_irqsave(&thi->t_lock, flags);
434 thi->t_state = RUNNING;
435 spin_unlock_irqrestore(&thi->t_lock, flags);
439 thi->t_state = RESTARTING;
440 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
441 thi->name, current->comm, current->pid);
446 spin_unlock_irqrestore(&thi->t_lock, flags);
454 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
458 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
460 /* may be called from state engine, holding the req lock irqsave */
461 spin_lock_irqsave(&thi->t_lock, flags);
463 if (thi->t_state == NONE) {
464 spin_unlock_irqrestore(&thi->t_lock, flags);
466 drbd_thread_start(thi);
470 if (thi->t_state != ns) {
471 if (thi->task == NULL) {
472 spin_unlock_irqrestore(&thi->t_lock, flags);
478 init_completion(&thi->stop);
479 if (thi->task != current)
480 force_sig(DRBD_SIGKILL, thi->task);
483 spin_unlock_irqrestore(&thi->t_lock, flags);
486 wait_for_completion(&thi->stop);
489 int conn_lowest_minor(struct drbd_connection *connection)
491 struct drbd_peer_device *peer_device;
492 int vnr = 0, minor = -1;
495 peer_device = idr_get_next(&connection->peer_devices, &vnr);
497 minor = device_to_minor(peer_device->device);
505 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
507 * Forces all threads of a resource onto the same CPU. This is beneficial for
508 * DRBD's performance. May be overwritten by user's configuration.
510 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
512 unsigned int *resources_per_cpu, min_index = ~0;
514 resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu),
516 if (resources_per_cpu) {
517 struct drbd_resource *resource;
518 unsigned int cpu, min = ~0;
521 for_each_resource_rcu(resource, &drbd_resources) {
522 for_each_cpu(cpu, resource->cpu_mask)
523 resources_per_cpu[cpu]++;
526 for_each_online_cpu(cpu) {
527 if (resources_per_cpu[cpu] < min) {
528 min = resources_per_cpu[cpu];
532 kfree(resources_per_cpu);
534 if (min_index == ~0) {
535 cpumask_setall(*cpu_mask);
538 cpumask_set_cpu(min_index, *cpu_mask);
542 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
543 * @device: DRBD device.
544 * @thi: drbd_thread object
546 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
549 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
551 struct drbd_resource *resource = thi->resource;
552 struct task_struct *p = current;
554 if (!thi->reset_cpu_mask)
556 thi->reset_cpu_mask = 0;
557 set_cpus_allowed_ptr(p, resource->cpu_mask);
560 #define drbd_calc_cpu_mask(A) ({})
564 * drbd_header_size - size of a packet header
566 * The header size is a multiple of 8, so any payload following the header is
567 * word aligned on 64-bit architectures. (The bitmap send and receive code
570 unsigned int drbd_header_size(struct drbd_connection *connection)
572 if (connection->agreed_pro_version >= 100) {
573 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
574 return sizeof(struct p_header100);
576 BUILD_BUG_ON(sizeof(struct p_header80) !=
577 sizeof(struct p_header95));
578 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
579 return sizeof(struct p_header80);
583 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
585 h->magic = cpu_to_be32(DRBD_MAGIC);
586 h->command = cpu_to_be16(cmd);
587 h->length = cpu_to_be16(size);
588 return sizeof(struct p_header80);
591 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
593 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
594 h->command = cpu_to_be16(cmd);
595 h->length = cpu_to_be32(size);
596 return sizeof(struct p_header95);
599 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
602 h->magic = cpu_to_be32(DRBD_MAGIC_100);
603 h->volume = cpu_to_be16(vnr);
604 h->command = cpu_to_be16(cmd);
605 h->length = cpu_to_be32(size);
607 return sizeof(struct p_header100);
610 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
611 void *buffer, enum drbd_packet cmd, int size)
613 if (connection->agreed_pro_version >= 100)
614 return prepare_header100(buffer, cmd, size, vnr);
615 else if (connection->agreed_pro_version >= 95 &&
616 size > DRBD_MAX_SIZE_H80_PACKET)
617 return prepare_header95(buffer, cmd, size);
619 return prepare_header80(buffer, cmd, size);
622 static void *__conn_prepare_command(struct drbd_connection *connection,
623 struct drbd_socket *sock)
627 return sock->sbuf + drbd_header_size(connection);
630 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
634 mutex_lock(&sock->mutex);
635 p = __conn_prepare_command(connection, sock);
637 mutex_unlock(&sock->mutex);
642 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
644 return conn_prepare_command(peer_device->connection, sock);
647 static int __send_command(struct drbd_connection *connection, int vnr,
648 struct drbd_socket *sock, enum drbd_packet cmd,
649 unsigned int header_size, void *data,
656 * Called with @data == NULL and the size of the data blocks in @size
657 * for commands that send data blocks. For those commands, omit the
658 * MSG_MORE flag: this will increase the likelihood that data blocks
659 * which are page aligned on the sender will end up page aligned on the
662 msg_flags = data ? MSG_MORE : 0;
664 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
666 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
669 err = drbd_send_all(connection, sock->socket, data, size, 0);
670 /* DRBD protocol "pings" are latency critical.
671 * This is supposed to trigger tcp_push_pending_frames() */
672 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
673 drbd_tcp_nodelay(sock->socket);
678 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
679 enum drbd_packet cmd, unsigned int header_size,
680 void *data, unsigned int size)
682 return __send_command(connection, 0, sock, cmd, header_size, data, size);
685 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
686 enum drbd_packet cmd, unsigned int header_size,
687 void *data, unsigned int size)
691 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
692 mutex_unlock(&sock->mutex);
696 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
697 enum drbd_packet cmd, unsigned int header_size,
698 void *data, unsigned int size)
702 err = __send_command(peer_device->connection, peer_device->device->vnr,
703 sock, cmd, header_size, data, size);
704 mutex_unlock(&sock->mutex);
708 int drbd_send_ping(struct drbd_connection *connection)
710 struct drbd_socket *sock;
712 sock = &connection->meta;
713 if (!conn_prepare_command(connection, sock))
715 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
718 int drbd_send_ping_ack(struct drbd_connection *connection)
720 struct drbd_socket *sock;
722 sock = &connection->meta;
723 if (!conn_prepare_command(connection, sock))
725 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
728 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
730 struct drbd_socket *sock;
731 struct p_rs_param_95 *p;
733 const int apv = peer_device->connection->agreed_pro_version;
734 enum drbd_packet cmd;
736 struct disk_conf *dc;
738 sock = &peer_device->connection->data;
739 p = drbd_prepare_command(peer_device, sock);
744 nc = rcu_dereference(peer_device->connection->net_conf);
746 size = apv <= 87 ? sizeof(struct p_rs_param)
747 : apv == 88 ? sizeof(struct p_rs_param)
748 + strlen(nc->verify_alg) + 1
749 : apv <= 94 ? sizeof(struct p_rs_param_89)
750 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
752 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
754 /* initialize verify_alg and csums_alg */
755 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
757 if (get_ldev(peer_device->device)) {
758 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
759 p->resync_rate = cpu_to_be32(dc->resync_rate);
760 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
761 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
762 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
763 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
764 put_ldev(peer_device->device);
766 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
767 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
768 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
769 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
770 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
774 strcpy(p->verify_alg, nc->verify_alg);
776 strcpy(p->csums_alg, nc->csums_alg);
779 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
782 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
784 struct drbd_socket *sock;
785 struct p_protocol *p;
789 sock = &connection->data;
790 p = __conn_prepare_command(connection, sock);
795 nc = rcu_dereference(connection->net_conf);
797 if (nc->tentative && connection->agreed_pro_version < 92) {
799 mutex_unlock(&sock->mutex);
800 drbd_err(connection, "--dry-run is not supported by peer");
805 if (connection->agreed_pro_version >= 87)
806 size += strlen(nc->integrity_alg) + 1;
808 p->protocol = cpu_to_be32(nc->wire_protocol);
809 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
810 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
811 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
812 p->two_primaries = cpu_to_be32(nc->two_primaries);
814 if (nc->discard_my_data)
815 cf |= CF_DISCARD_MY_DATA;
818 p->conn_flags = cpu_to_be32(cf);
820 if (connection->agreed_pro_version >= 87)
821 strcpy(p->integrity_alg, nc->integrity_alg);
824 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
827 int drbd_send_protocol(struct drbd_connection *connection)
831 mutex_lock(&connection->data.mutex);
832 err = __drbd_send_protocol(connection, P_PROTOCOL);
833 mutex_unlock(&connection->data.mutex);
838 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
840 struct drbd_device *device = peer_device->device;
841 struct drbd_socket *sock;
845 if (!get_ldev_if_state(device, D_NEGOTIATING))
848 sock = &peer_device->connection->data;
849 p = drbd_prepare_command(peer_device, sock);
854 spin_lock_irq(&device->ldev->md.uuid_lock);
855 for (i = UI_CURRENT; i < UI_SIZE; i++)
856 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
857 spin_unlock_irq(&device->ldev->md.uuid_lock);
859 device->comm_bm_set = drbd_bm_total_weight(device);
860 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
862 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
864 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
865 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
866 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
869 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
872 int drbd_send_uuids(struct drbd_peer_device *peer_device)
874 return _drbd_send_uuids(peer_device, 0);
877 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
879 return _drbd_send_uuids(peer_device, 8);
882 void drbd_print_uuids(struct drbd_device *device, const char *text)
884 if (get_ldev_if_state(device, D_NEGOTIATING)) {
885 u64 *uuid = device->ldev->md.uuid;
886 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
888 (unsigned long long)uuid[UI_CURRENT],
889 (unsigned long long)uuid[UI_BITMAP],
890 (unsigned long long)uuid[UI_HISTORY_START],
891 (unsigned long long)uuid[UI_HISTORY_END]);
894 drbd_info(device, "%s effective data uuid: %016llX\n",
896 (unsigned long long)device->ed_uuid);
900 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
902 struct drbd_device *device = peer_device->device;
903 struct drbd_socket *sock;
907 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
909 uuid = device->ldev->md.uuid[UI_BITMAP];
910 if (uuid && uuid != UUID_JUST_CREATED)
911 uuid = uuid + UUID_NEW_BM_OFFSET;
913 get_random_bytes(&uuid, sizeof(u64));
914 drbd_uuid_set(device, UI_BITMAP, uuid);
915 drbd_print_uuids(device, "updated sync UUID");
916 drbd_md_sync(device);
918 sock = &peer_device->connection->data;
919 p = drbd_prepare_command(peer_device, sock);
921 p->uuid = cpu_to_be64(uuid);
922 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
926 /* communicated if (agreed_features & DRBD_FF_WSAME) */
928 assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p,
929 struct request_queue *q)
932 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
933 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
934 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
935 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
936 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
937 p->qlim->discard_enabled = blk_queue_discard(q);
938 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
940 q = device->rq_queue;
941 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
942 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
943 p->qlim->alignment_offset = 0;
944 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
945 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
946 p->qlim->discard_enabled = 0;
947 p->qlim->write_same_capable = 0;
951 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
953 struct drbd_device *device = peer_device->device;
954 struct drbd_socket *sock;
956 sector_t d_size, u_size;
958 unsigned int max_bio_size;
959 unsigned int packet_size;
961 sock = &peer_device->connection->data;
962 p = drbd_prepare_command(peer_device, sock);
966 packet_size = sizeof(*p);
967 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
968 packet_size += sizeof(p->qlim[0]);
970 memset(p, 0, packet_size);
971 if (get_ldev_if_state(device, D_NEGOTIATING)) {
972 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
973 d_size = drbd_get_max_capacity(device->ldev);
975 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
977 q_order_type = drbd_queue_order_type(device);
978 max_bio_size = queue_max_hw_sectors(q) << 9;
979 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
980 assign_p_sizes_qlim(device, p, q);
985 q_order_type = QUEUE_ORDERED_NONE;
986 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
987 assign_p_sizes_qlim(device, p, NULL);
990 if (peer_device->connection->agreed_pro_version <= 94)
991 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
992 else if (peer_device->connection->agreed_pro_version < 100)
993 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
995 p->d_size = cpu_to_be64(d_size);
996 p->u_size = cpu_to_be64(u_size);
997 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
998 p->max_bio_size = cpu_to_be32(max_bio_size);
999 p->queue_order_type = cpu_to_be16(q_order_type);
1000 p->dds_flags = cpu_to_be16(flags);
1002 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
1006 * drbd_send_current_state() - Sends the drbd state to the peer
1007 * @peer_device: DRBD peer device.
1009 int drbd_send_current_state(struct drbd_peer_device *peer_device)
1011 struct drbd_socket *sock;
1014 sock = &peer_device->connection->data;
1015 p = drbd_prepare_command(peer_device, sock);
1018 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1019 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1023 * drbd_send_state() - After a state change, sends the new state to the peer
1024 * @peer_device: DRBD peer device.
1025 * @state: the state to send, not necessarily the current state.
1027 * Each state change queues an "after_state_ch" work, which will eventually
1028 * send the resulting new state to the peer. If more state changes happen
1029 * between queuing and processing of the after_state_ch work, we still
1030 * want to send each intermediary state in the order it occurred.
1032 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1034 struct drbd_socket *sock;
1037 sock = &peer_device->connection->data;
1038 p = drbd_prepare_command(peer_device, sock);
1041 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1042 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1045 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1047 struct drbd_socket *sock;
1048 struct p_req_state *p;
1050 sock = &peer_device->connection->data;
1051 p = drbd_prepare_command(peer_device, sock);
1054 p->mask = cpu_to_be32(mask.i);
1055 p->val = cpu_to_be32(val.i);
1056 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1059 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1061 enum drbd_packet cmd;
1062 struct drbd_socket *sock;
1063 struct p_req_state *p;
1065 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1066 sock = &connection->data;
1067 p = conn_prepare_command(connection, sock);
1070 p->mask = cpu_to_be32(mask.i);
1071 p->val = cpu_to_be32(val.i);
1072 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1075 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1077 struct drbd_socket *sock;
1078 struct p_req_state_reply *p;
1080 sock = &peer_device->connection->meta;
1081 p = drbd_prepare_command(peer_device, sock);
1083 p->retcode = cpu_to_be32(retcode);
1084 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1088 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1090 struct drbd_socket *sock;
1091 struct p_req_state_reply *p;
1092 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1094 sock = &connection->meta;
1095 p = conn_prepare_command(connection, sock);
1097 p->retcode = cpu_to_be32(retcode);
1098 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1102 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1104 BUG_ON(code & ~0xf);
1105 p->encoding = (p->encoding & ~0xf) | code;
1108 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1110 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1113 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1116 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1119 static int fill_bitmap_rle_bits(struct drbd_device *device,
1120 struct p_compressed_bm *p,
1122 struct bm_xfer_ctx *c)
1124 struct bitstream bs;
1125 unsigned long plain_bits;
1132 /* may we use this feature? */
1134 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1136 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1139 if (c->bit_offset >= c->bm_bits)
1140 return 0; /* nothing to do. */
1142 /* use at most thus many bytes */
1143 bitstream_init(&bs, p->code, size, 0);
1144 memset(p->code, 0, size);
1145 /* plain bits covered in this code string */
1148 /* p->encoding & 0x80 stores whether the first run length is set.
1149 * bit offset is implicit.
1150 * start with toggle == 2 to be able to tell the first iteration */
1153 /* see how much plain bits we can stuff into one packet
1154 * using RLE and VLI. */
1156 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1157 : _drbd_bm_find_next(device, c->bit_offset);
1160 rl = tmp - c->bit_offset;
1162 if (toggle == 2) { /* first iteration */
1164 /* the first checked bit was set,
1165 * store start value, */
1166 dcbp_set_start(p, 1);
1167 /* but skip encoding of zero run length */
1171 dcbp_set_start(p, 0);
1174 /* paranoia: catch zero runlength.
1175 * can only happen if bitmap is modified while we scan it. */
1177 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1178 "t:%u bo:%lu\n", toggle, c->bit_offset);
1182 bits = vli_encode_bits(&bs, rl);
1183 if (bits == -ENOBUFS) /* buffer full */
1186 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1192 c->bit_offset = tmp;
1193 } while (c->bit_offset < c->bm_bits);
1195 len = bs.cur.b - p->code + !!bs.cur.bit;
1197 if (plain_bits < (len << 3)) {
1198 /* incompressible with this method.
1199 * we need to rewind both word and bit position. */
1200 c->bit_offset -= plain_bits;
1201 bm_xfer_ctx_bit_to_word_offset(c);
1202 c->bit_offset = c->word_offset * BITS_PER_LONG;
1206 /* RLE + VLI was able to compress it just fine.
1207 * update c->word_offset. */
1208 bm_xfer_ctx_bit_to_word_offset(c);
1210 /* store pad_bits */
1211 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1217 * send_bitmap_rle_or_plain
1219 * Return 0 when done, 1 when another iteration is needed, and a negative error
1220 * code upon failure.
1223 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1225 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1226 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1227 struct p_compressed_bm *p = sock->sbuf + header_size;
1230 len = fill_bitmap_rle_bits(device, p,
1231 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1236 dcbp_set_code(p, RLE_VLI_Bits);
1237 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1238 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1241 c->bytes[0] += header_size + sizeof(*p) + len;
1243 if (c->bit_offset >= c->bm_bits)
1246 /* was not compressible.
1247 * send a buffer full of plain text bits instead. */
1248 unsigned int data_size;
1249 unsigned long num_words;
1250 unsigned long *p = sock->sbuf + header_size;
1252 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1253 num_words = min_t(size_t, data_size / sizeof(*p),
1254 c->bm_words - c->word_offset);
1255 len = num_words * sizeof(*p);
1257 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1258 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1259 c->word_offset += num_words;
1260 c->bit_offset = c->word_offset * BITS_PER_LONG;
1263 c->bytes[1] += header_size + len;
1265 if (c->bit_offset > c->bm_bits)
1266 c->bit_offset = c->bm_bits;
1270 INFO_bm_xfer_stats(device, "send", c);
1278 /* See the comment at receive_bitmap() */
1279 static int _drbd_send_bitmap(struct drbd_device *device)
1281 struct bm_xfer_ctx c;
1284 if (!expect(device->bitmap))
1287 if (get_ldev(device)) {
1288 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1289 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1290 drbd_bm_set_all(device);
1291 if (drbd_bm_write(device)) {
1292 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1293 * but otherwise process as per normal - need to tell other
1294 * side that a full resync is required! */
1295 drbd_err(device, "Failed to write bitmap to disk!\n");
1297 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1298 drbd_md_sync(device);
1304 c = (struct bm_xfer_ctx) {
1305 .bm_bits = drbd_bm_bits(device),
1306 .bm_words = drbd_bm_words(device),
1310 err = send_bitmap_rle_or_plain(device, &c);
1316 int drbd_send_bitmap(struct drbd_device *device)
1318 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1321 mutex_lock(&sock->mutex);
1323 err = !_drbd_send_bitmap(device);
1324 mutex_unlock(&sock->mutex);
1328 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1330 struct drbd_socket *sock;
1331 struct p_barrier_ack *p;
1333 if (connection->cstate < C_WF_REPORT_PARAMS)
1336 sock = &connection->meta;
1337 p = conn_prepare_command(connection, sock);
1340 p->barrier = barrier_nr;
1341 p->set_size = cpu_to_be32(set_size);
1342 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1346 * _drbd_send_ack() - Sends an ack packet
1347 * @device: DRBD device.
1348 * @cmd: Packet command code.
1349 * @sector: sector, needs to be in big endian byte order
1350 * @blksize: size in byte, needs to be in big endian byte order
1351 * @block_id: Id, big endian byte order
1353 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1354 u64 sector, u32 blksize, u64 block_id)
1356 struct drbd_socket *sock;
1357 struct p_block_ack *p;
1359 if (peer_device->device->state.conn < C_CONNECTED)
1362 sock = &peer_device->connection->meta;
1363 p = drbd_prepare_command(peer_device, sock);
1367 p->block_id = block_id;
1368 p->blksize = blksize;
1369 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1370 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1373 /* dp->sector and dp->block_id already/still in network byte order,
1374 * data_size is payload size according to dp->head,
1375 * and may need to be corrected for digest size. */
1376 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1377 struct p_data *dp, int data_size)
1379 if (peer_device->connection->peer_integrity_tfm)
1380 data_size -= crypto_ahash_digestsize(peer_device->connection->peer_integrity_tfm);
1381 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1385 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1386 struct p_block_req *rp)
1388 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1392 * drbd_send_ack() - Sends an ack packet
1393 * @device: DRBD device
1394 * @cmd: packet command code
1395 * @peer_req: peer request
1397 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1398 struct drbd_peer_request *peer_req)
1400 return _drbd_send_ack(peer_device, cmd,
1401 cpu_to_be64(peer_req->i.sector),
1402 cpu_to_be32(peer_req->i.size),
1403 peer_req->block_id);
1406 /* This function misuses the block_id field to signal if the blocks
1407 * are is sync or not. */
1408 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1409 sector_t sector, int blksize, u64 block_id)
1411 return _drbd_send_ack(peer_device, cmd,
1412 cpu_to_be64(sector),
1413 cpu_to_be32(blksize),
1414 cpu_to_be64(block_id));
1417 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1418 struct drbd_peer_request *peer_req)
1420 struct drbd_socket *sock;
1421 struct p_block_desc *p;
1423 sock = &peer_device->connection->data;
1424 p = drbd_prepare_command(peer_device, sock);
1427 p->sector = cpu_to_be64(peer_req->i.sector);
1428 p->blksize = cpu_to_be32(peer_req->i.size);
1430 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1433 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1434 sector_t sector, int size, u64 block_id)
1436 struct drbd_socket *sock;
1437 struct p_block_req *p;
1439 sock = &peer_device->connection->data;
1440 p = drbd_prepare_command(peer_device, sock);
1443 p->sector = cpu_to_be64(sector);
1444 p->block_id = block_id;
1445 p->blksize = cpu_to_be32(size);
1446 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1449 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1450 void *digest, int digest_size, enum drbd_packet cmd)
1452 struct drbd_socket *sock;
1453 struct p_block_req *p;
1455 /* FIXME: Put the digest into the preallocated socket buffer. */
1457 sock = &peer_device->connection->data;
1458 p = drbd_prepare_command(peer_device, sock);
1461 p->sector = cpu_to_be64(sector);
1462 p->block_id = ID_SYNCER /* unused */;
1463 p->blksize = cpu_to_be32(size);
1464 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1467 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1469 struct drbd_socket *sock;
1470 struct p_block_req *p;
1472 sock = &peer_device->connection->data;
1473 p = drbd_prepare_command(peer_device, sock);
1476 p->sector = cpu_to_be64(sector);
1477 p->block_id = ID_SYNCER /* unused */;
1478 p->blksize = cpu_to_be32(size);
1479 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1482 /* called on sndtimeo
1483 * returns false if we should retry,
1484 * true if we think connection is dead
1486 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1489 /* long elapsed = (long)(jiffies - device->last_received); */
1491 drop_it = connection->meta.socket == sock
1492 || !connection->ack_receiver.task
1493 || get_t_state(&connection->ack_receiver) != RUNNING
1494 || connection->cstate < C_WF_REPORT_PARAMS;
1499 drop_it = !--connection->ko_count;
1501 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1502 current->comm, current->pid, connection->ko_count);
1503 request_ping(connection);
1506 return drop_it; /* && (device->state == R_PRIMARY) */;
1509 static void drbd_update_congested(struct drbd_connection *connection)
1511 struct sock *sk = connection->data.socket->sk;
1512 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1513 set_bit(NET_CONGESTED, &connection->flags);
1516 /* The idea of sendpage seems to be to put some kind of reference
1517 * to the page into the skb, and to hand it over to the NIC. In
1518 * this process get_page() gets called.
1520 * As soon as the page was really sent over the network put_page()
1521 * gets called by some part of the network layer. [ NIC driver? ]
1523 * [ get_page() / put_page() increment/decrement the count. If count
1524 * reaches 0 the page will be freed. ]
1526 * This works nicely with pages from FSs.
1527 * But this means that in protocol A we might signal IO completion too early!
1529 * In order not to corrupt data during a resync we must make sure
1530 * that we do not reuse our own buffer pages (EEs) to early, therefore
1531 * we have the net_ee list.
1533 * XFS seems to have problems, still, it submits pages with page_count == 0!
1534 * As a workaround, we disable sendpage on pages
1535 * with page_count == 0 or PageSlab.
1537 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1538 int offset, size_t size, unsigned msg_flags)
1540 struct socket *socket;
1544 socket = peer_device->connection->data.socket;
1545 addr = kmap(page) + offset;
1546 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1549 peer_device->device->send_cnt += size >> 9;
1553 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1554 int offset, size_t size, unsigned msg_flags)
1556 struct socket *socket = peer_device->connection->data.socket;
1560 /* e.g. XFS meta- & log-data is in slab pages, which have a
1561 * page_count of 0 and/or have PageSlab() set.
1562 * we cannot use send_page for those, as that does get_page();
1563 * put_page(); and would cause either a VM_BUG directly, or
1564 * __page_cache_release a page that would actually still be referenced
1565 * by someone, leading to some obscure delayed Oops somewhere else. */
1566 if (drbd_disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1567 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1569 msg_flags |= MSG_NOSIGNAL;
1570 drbd_update_congested(peer_device->connection);
1574 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1576 if (sent == -EAGAIN) {
1577 if (we_should_drop_the_connection(peer_device->connection, socket))
1581 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1582 __func__, (int)size, len, sent);
1589 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1590 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1594 peer_device->device->send_cnt += size >> 9;
1599 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1601 struct bio_vec bvec;
1602 struct bvec_iter iter;
1604 /* hint all but last page with MSG_MORE */
1605 bio_for_each_segment(bvec, bio, iter) {
1608 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1609 bvec.bv_offset, bvec.bv_len,
1610 bio_iter_last(bvec, iter)
1614 /* REQ_OP_WRITE_SAME has only one segment */
1615 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1621 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1623 struct bio_vec bvec;
1624 struct bvec_iter iter;
1626 /* hint all but last page with MSG_MORE */
1627 bio_for_each_segment(bvec, bio, iter) {
1630 err = _drbd_send_page(peer_device, bvec.bv_page,
1631 bvec.bv_offset, bvec.bv_len,
1632 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1635 /* REQ_OP_WRITE_SAME has only one segment */
1636 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1642 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1643 struct drbd_peer_request *peer_req)
1645 struct page *page = peer_req->pages;
1646 unsigned len = peer_req->i.size;
1649 /* hint all but last page with MSG_MORE */
1650 page_chain_for_each(page) {
1651 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1653 err = _drbd_send_page(peer_device, page, 0, l,
1654 page_chain_next(page) ? MSG_MORE : 0);
1662 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1665 if (connection->agreed_pro_version >= 95)
1666 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1667 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1668 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1669 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1670 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1671 (bio_op(bio) == REQ_OP_WRITE_ZEROES ? DP_DISCARD : 0);
1673 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1676 /* Used to send write or TRIM aka REQ_DISCARD requests
1677 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1679 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1681 struct drbd_device *device = peer_device->device;
1682 struct drbd_socket *sock;
1684 struct p_wsame *wsame = NULL;
1686 unsigned int dp_flags = 0;
1690 sock = &peer_device->connection->data;
1691 p = drbd_prepare_command(peer_device, sock);
1692 digest_size = peer_device->connection->integrity_tfm ?
1693 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1697 p->sector = cpu_to_be64(req->i.sector);
1698 p->block_id = (unsigned long)req;
1699 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1700 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1701 if (device->state.conn >= C_SYNC_SOURCE &&
1702 device->state.conn <= C_PAUSED_SYNC_T)
1703 dp_flags |= DP_MAY_SET_IN_SYNC;
1704 if (peer_device->connection->agreed_pro_version >= 100) {
1705 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1706 dp_flags |= DP_SEND_RECEIVE_ACK;
1707 /* During resync, request an explicit write ack,
1708 * even in protocol != C */
1709 if (req->rq_state & RQ_EXP_WRITE_ACK
1710 || (dp_flags & DP_MAY_SET_IN_SYNC))
1711 dp_flags |= DP_SEND_WRITE_ACK;
1713 p->dp_flags = cpu_to_be32(dp_flags);
1715 if (dp_flags & DP_DISCARD) {
1716 struct p_trim *t = (struct p_trim*)p;
1717 t->size = cpu_to_be32(req->i.size);
1718 err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1721 if (dp_flags & DP_WSAME) {
1722 /* this will only work if DRBD_FF_WSAME is set AND the
1723 * handshake agreed that all nodes and backend devices are
1724 * WRITE_SAME capable and agree on logical_block_size */
1725 wsame = (struct p_wsame*)p;
1726 digest_out = wsame + 1;
1727 wsame->size = cpu_to_be32(req->i.size);
1731 /* our digest is still only over the payload.
1732 * TRIM does not carry any payload. */
1734 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1737 __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1738 sizeof(*wsame) + digest_size, NULL,
1739 bio_iovec(req->master_bio).bv_len);
1742 __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1743 sizeof(*p) + digest_size, 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)) || digest_size)
1757 err = _drbd_send_bio(peer_device, req->master_bio);
1759 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1761 /* double check digest, sometimes buffers have been modified in flight. */
1762 if (digest_size > 0 && digest_size <= 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(peer_device->connection->integrity_tfm, req->master_bio, digest);
1767 if (memcmp(p + 1, digest, digest_size)) {
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 (digest_size > 64) {
1773 ... Be noisy about digest too large ...
1777 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1782 /* answer packet, used to send data back for read requests:
1783 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1784 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1786 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1787 struct drbd_peer_request *peer_req)
1789 struct drbd_device *device = peer_device->device;
1790 struct drbd_socket *sock;
1795 sock = &peer_device->connection->data;
1796 p = drbd_prepare_command(peer_device, sock);
1798 digest_size = peer_device->connection->integrity_tfm ?
1799 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1803 p->sector = cpu_to_be64(peer_req->i.sector);
1804 p->block_id = peer_req->block_id;
1805 p->seq_num = 0; /* unused */
1808 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1809 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1811 err = _drbd_send_zc_ee(peer_device, peer_req);
1812 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1817 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1819 struct drbd_socket *sock;
1820 struct p_block_desc *p;
1822 sock = &peer_device->connection->data;
1823 p = drbd_prepare_command(peer_device, sock);
1826 p->sector = cpu_to_be64(req->i.sector);
1827 p->blksize = cpu_to_be32(req->i.size);
1828 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1832 drbd_send distinguishes two cases:
1834 Packets sent via the data socket "sock"
1835 and packets sent via the meta data socket "msock"
1838 -----------------+-------------------------+------------------------------
1839 timeout conf.timeout / 2 conf.timeout / 2
1840 timeout action send a ping via msock Abort communication
1841 and close all sockets
1845 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1847 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1848 void *buf, size_t size, unsigned msg_flags)
1850 struct kvec iov = {.iov_base = buf, .iov_len = size};
1851 struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL};
1857 /* THINK if (signal_pending) return ... ? */
1859 iov_iter_kvec(&msg.msg_iter, WRITE | ITER_KVEC, &iov, 1, size);
1861 if (sock == connection->data.socket) {
1863 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1865 drbd_update_congested(connection);
1868 rv = sock_sendmsg(sock, &msg);
1869 if (rv == -EAGAIN) {
1870 if (we_should_drop_the_connection(connection, sock))
1876 flush_signals(current);
1882 } while (sent < size);
1884 if (sock == connection->data.socket)
1885 clear_bit(NET_CONGESTED, &connection->flags);
1888 if (rv != -EAGAIN) {
1889 drbd_err(connection, "%s_sendmsg returned %d\n",
1890 sock == connection->meta.socket ? "msock" : "sock",
1892 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1894 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1901 * drbd_send_all - Send an entire buffer
1903 * Returns 0 upon success and a negative error value otherwise.
1905 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1906 size_t size, unsigned msg_flags)
1910 err = drbd_send(connection, sock, buffer, size, msg_flags);
1918 static int drbd_open(struct block_device *bdev, fmode_t mode)
1920 struct drbd_device *device = bdev->bd_disk->private_data;
1921 unsigned long flags;
1924 mutex_lock(&drbd_main_mutex);
1925 spin_lock_irqsave(&device->resource->req_lock, flags);
1926 /* to have a stable device->state.role
1927 * and no race with updating open_cnt */
1929 if (device->state.role != R_PRIMARY) {
1930 if (mode & FMODE_WRITE)
1932 else if (!drbd_allow_oos)
1938 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1939 mutex_unlock(&drbd_main_mutex);
1944 static void drbd_release(struct gendisk *gd, fmode_t mode)
1946 struct drbd_device *device = gd->private_data;
1947 mutex_lock(&drbd_main_mutex);
1949 mutex_unlock(&drbd_main_mutex);
1952 /* need to hold resource->req_lock */
1953 void drbd_queue_unplug(struct drbd_device *device)
1955 if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1956 D_ASSERT(device, device->state.role == R_PRIMARY);
1957 if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1958 drbd_queue_work_if_unqueued(
1959 &first_peer_device(device)->connection->sender_work,
1960 &device->unplug_work);
1965 static void drbd_set_defaults(struct drbd_device *device)
1967 /* Beware! The actual layout differs
1968 * between big endian and little endian */
1969 device->state = (union drbd_dev_state) {
1970 { .role = R_SECONDARY,
1972 .conn = C_STANDALONE,
1978 void drbd_init_set_defaults(struct drbd_device *device)
1980 /* the memset(,0,) did most of this.
1981 * note: only assignments, no allocation in here */
1983 drbd_set_defaults(device);
1985 atomic_set(&device->ap_bio_cnt, 0);
1986 atomic_set(&device->ap_actlog_cnt, 0);
1987 atomic_set(&device->ap_pending_cnt, 0);
1988 atomic_set(&device->rs_pending_cnt, 0);
1989 atomic_set(&device->unacked_cnt, 0);
1990 atomic_set(&device->local_cnt, 0);
1991 atomic_set(&device->pp_in_use_by_net, 0);
1992 atomic_set(&device->rs_sect_in, 0);
1993 atomic_set(&device->rs_sect_ev, 0);
1994 atomic_set(&device->ap_in_flight, 0);
1995 atomic_set(&device->md_io.in_use, 0);
1997 mutex_init(&device->own_state_mutex);
1998 device->state_mutex = &device->own_state_mutex;
2000 spin_lock_init(&device->al_lock);
2001 spin_lock_init(&device->peer_seq_lock);
2003 INIT_LIST_HEAD(&device->active_ee);
2004 INIT_LIST_HEAD(&device->sync_ee);
2005 INIT_LIST_HEAD(&device->done_ee);
2006 INIT_LIST_HEAD(&device->read_ee);
2007 INIT_LIST_HEAD(&device->net_ee);
2008 INIT_LIST_HEAD(&device->resync_reads);
2009 INIT_LIST_HEAD(&device->resync_work.list);
2010 INIT_LIST_HEAD(&device->unplug_work.list);
2011 INIT_LIST_HEAD(&device->bm_io_work.w.list);
2012 INIT_LIST_HEAD(&device->pending_master_completion[0]);
2013 INIT_LIST_HEAD(&device->pending_master_completion[1]);
2014 INIT_LIST_HEAD(&device->pending_completion[0]);
2015 INIT_LIST_HEAD(&device->pending_completion[1]);
2017 device->resync_work.cb = w_resync_timer;
2018 device->unplug_work.cb = w_send_write_hint;
2019 device->bm_io_work.w.cb = w_bitmap_io;
2021 timer_setup(&device->resync_timer, resync_timer_fn, 0);
2022 timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0);
2023 timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0);
2024 timer_setup(&device->request_timer, request_timer_fn, 0);
2026 init_waitqueue_head(&device->misc_wait);
2027 init_waitqueue_head(&device->state_wait);
2028 init_waitqueue_head(&device->ee_wait);
2029 init_waitqueue_head(&device->al_wait);
2030 init_waitqueue_head(&device->seq_wait);
2032 device->resync_wenr = LC_FREE;
2033 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2034 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2037 void drbd_device_cleanup(struct drbd_device *device)
2040 if (first_peer_device(device)->connection->receiver.t_state != NONE)
2041 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2042 first_peer_device(device)->connection->receiver.t_state);
2044 device->al_writ_cnt =
2045 device->bm_writ_cnt =
2053 device->rs_failed = 0;
2054 device->rs_last_events = 0;
2055 device->rs_last_sect_ev = 0;
2056 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2057 device->rs_mark_left[i] = 0;
2058 device->rs_mark_time[i] = 0;
2060 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2062 drbd_set_my_capacity(device, 0);
2063 if (device->bitmap) {
2064 /* maybe never allocated. */
2065 drbd_bm_resize(device, 0, 1);
2066 drbd_bm_cleanup(device);
2069 drbd_backing_dev_free(device, device->ldev);
2070 device->ldev = NULL;
2072 clear_bit(AL_SUSPENDED, &device->flags);
2074 D_ASSERT(device, list_empty(&device->active_ee));
2075 D_ASSERT(device, list_empty(&device->sync_ee));
2076 D_ASSERT(device, list_empty(&device->done_ee));
2077 D_ASSERT(device, list_empty(&device->read_ee));
2078 D_ASSERT(device, list_empty(&device->net_ee));
2079 D_ASSERT(device, list_empty(&device->resync_reads));
2080 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2081 D_ASSERT(device, list_empty(&device->resync_work.list));
2082 D_ASSERT(device, list_empty(&device->unplug_work.list));
2084 drbd_set_defaults(device);
2088 static void drbd_destroy_mempools(void)
2092 while (drbd_pp_pool) {
2093 page = drbd_pp_pool;
2094 drbd_pp_pool = (struct page *)page_private(page);
2099 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2101 bioset_exit(&drbd_io_bio_set);
2102 bioset_exit(&drbd_md_io_bio_set);
2103 mempool_exit(&drbd_md_io_page_pool);
2104 mempool_exit(&drbd_ee_mempool);
2105 mempool_exit(&drbd_request_mempool);
2107 kmem_cache_destroy(drbd_ee_cache);
2108 if (drbd_request_cache)
2109 kmem_cache_destroy(drbd_request_cache);
2110 if (drbd_bm_ext_cache)
2111 kmem_cache_destroy(drbd_bm_ext_cache);
2112 if (drbd_al_ext_cache)
2113 kmem_cache_destroy(drbd_al_ext_cache);
2115 drbd_ee_cache = NULL;
2116 drbd_request_cache = NULL;
2117 drbd_bm_ext_cache = NULL;
2118 drbd_al_ext_cache = NULL;
2123 static int drbd_create_mempools(void)
2126 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2130 drbd_request_cache = kmem_cache_create(
2131 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2132 if (drbd_request_cache == NULL)
2135 drbd_ee_cache = kmem_cache_create(
2136 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2137 if (drbd_ee_cache == NULL)
2140 drbd_bm_ext_cache = kmem_cache_create(
2141 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2142 if (drbd_bm_ext_cache == NULL)
2145 drbd_al_ext_cache = kmem_cache_create(
2146 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2147 if (drbd_al_ext_cache == NULL)
2151 ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0);
2155 ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0,
2160 ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0);
2164 ret = mempool_init_slab_pool(&drbd_request_mempool, number,
2165 drbd_request_cache);
2169 ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache);
2173 /* drbd's page pool */
2174 spin_lock_init(&drbd_pp_lock);
2176 for (i = 0; i < number; i++) {
2177 page = alloc_page(GFP_HIGHUSER);
2180 set_page_private(page, (unsigned long)drbd_pp_pool);
2181 drbd_pp_pool = page;
2183 drbd_pp_vacant = number;
2188 drbd_destroy_mempools(); /* in case we allocated some */
2192 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2196 rr = drbd_free_peer_reqs(device, &device->active_ee);
2198 drbd_err(device, "%d EEs in active list found!\n", rr);
2200 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2202 drbd_err(device, "%d EEs in sync list found!\n", rr);
2204 rr = drbd_free_peer_reqs(device, &device->read_ee);
2206 drbd_err(device, "%d EEs in read list found!\n", rr);
2208 rr = drbd_free_peer_reqs(device, &device->done_ee);
2210 drbd_err(device, "%d EEs in done list found!\n", rr);
2212 rr = drbd_free_peer_reqs(device, &device->net_ee);
2214 drbd_err(device, "%d EEs in net list found!\n", rr);
2217 /* caution. no locking. */
2218 void drbd_destroy_device(struct kref *kref)
2220 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2221 struct drbd_resource *resource = device->resource;
2222 struct drbd_peer_device *peer_device, *tmp_peer_device;
2224 del_timer_sync(&device->request_timer);
2226 /* paranoia asserts */
2227 D_ASSERT(device, device->open_cnt == 0);
2228 /* end paranoia asserts */
2230 /* cleanup stuff that may have been allocated during
2231 * device (re-)configuration or state changes */
2233 if (device->this_bdev)
2234 bdput(device->this_bdev);
2236 drbd_backing_dev_free(device, device->ldev);
2237 device->ldev = NULL;
2239 drbd_release_all_peer_reqs(device);
2241 lc_destroy(device->act_log);
2242 lc_destroy(device->resync);
2244 kfree(device->p_uuid);
2245 /* device->p_uuid = NULL; */
2247 if (device->bitmap) /* should no longer be there. */
2248 drbd_bm_cleanup(device);
2249 __free_page(device->md_io.page);
2250 put_disk(device->vdisk);
2251 blk_cleanup_queue(device->rq_queue);
2252 kfree(device->rs_plan_s);
2254 /* not for_each_connection(connection, resource):
2255 * those may have been cleaned up and disassociated already.
2257 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2258 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2261 memset(device, 0xfd, sizeof(*device));
2263 kref_put(&resource->kref, drbd_destroy_resource);
2266 /* One global retry thread, if we need to push back some bio and have it
2267 * reinserted through our make request function.
2269 static struct retry_worker {
2270 struct workqueue_struct *wq;
2271 struct work_struct worker;
2274 struct list_head writes;
2277 static void do_retry(struct work_struct *ws)
2279 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2281 struct drbd_request *req, *tmp;
2283 spin_lock_irq(&retry->lock);
2284 list_splice_init(&retry->writes, &writes);
2285 spin_unlock_irq(&retry->lock);
2287 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2288 struct drbd_device *device = req->device;
2289 struct bio *bio = req->master_bio;
2290 unsigned long start_jif = req->start_jif;
2294 expect(atomic_read(&req->completion_ref) == 0) &&
2295 expect(req->rq_state & RQ_POSTPONED) &&
2296 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2297 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2300 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2301 req, atomic_read(&req->completion_ref),
2304 /* We still need to put one kref associated with the
2305 * "completion_ref" going zero in the code path that queued it
2306 * here. The request object may still be referenced by a
2307 * frozen local req->private_bio, in case we force-detached.
2309 kref_put(&req->kref, drbd_req_destroy);
2311 /* A single suspended or otherwise blocking device may stall
2312 * all others as well. Fortunately, this code path is to
2313 * recover from a situation that "should not happen":
2314 * concurrent writes in multi-primary setup.
2315 * In a "normal" lifecycle, this workqueue is supposed to be
2316 * destroyed without ever doing anything.
2317 * If it turns out to be an issue anyways, we can do per
2318 * resource (replication group) or per device (minor) retry
2319 * workqueues instead.
2322 /* We are not just doing generic_make_request(),
2323 * as we want to keep the start_time information. */
2325 __drbd_make_request(device, bio, start_jif);
2329 /* called via drbd_req_put_completion_ref(),
2330 * holds resource->req_lock */
2331 void drbd_restart_request(struct drbd_request *req)
2333 unsigned long flags;
2334 spin_lock_irqsave(&retry.lock, flags);
2335 list_move_tail(&req->tl_requests, &retry.writes);
2336 spin_unlock_irqrestore(&retry.lock, flags);
2338 /* Drop the extra reference that would otherwise
2339 * have been dropped by complete_master_bio.
2340 * do_retry() needs to grab a new one. */
2341 dec_ap_bio(req->device);
2343 queue_work(retry.wq, &retry.worker);
2346 void drbd_destroy_resource(struct kref *kref)
2348 struct drbd_resource *resource =
2349 container_of(kref, struct drbd_resource, kref);
2351 idr_destroy(&resource->devices);
2352 free_cpumask_var(resource->cpu_mask);
2353 kfree(resource->name);
2354 memset(resource, 0xf2, sizeof(*resource));
2358 void drbd_free_resource(struct drbd_resource *resource)
2360 struct drbd_connection *connection, *tmp;
2362 for_each_connection_safe(connection, tmp, resource) {
2363 list_del(&connection->connections);
2364 drbd_debugfs_connection_cleanup(connection);
2365 kref_put(&connection->kref, drbd_destroy_connection);
2367 drbd_debugfs_resource_cleanup(resource);
2368 kref_put(&resource->kref, drbd_destroy_resource);
2371 static void drbd_cleanup(void)
2374 struct drbd_device *device;
2375 struct drbd_resource *resource, *tmp;
2377 /* first remove proc,
2378 * drbdsetup uses it's presence to detect
2379 * whether DRBD is loaded.
2380 * If we would get stuck in proc removal,
2381 * but have netlink already deregistered,
2382 * some drbdsetup commands may wait forever
2386 remove_proc_entry("drbd", NULL);
2389 destroy_workqueue(retry.wq);
2391 drbd_genl_unregister();
2393 idr_for_each_entry(&drbd_devices, device, i)
2394 drbd_delete_device(device);
2396 /* not _rcu since, no other updater anymore. Genl already unregistered */
2397 for_each_resource_safe(resource, tmp, &drbd_resources) {
2398 list_del(&resource->resources);
2399 drbd_free_resource(resource);
2402 drbd_debugfs_cleanup();
2404 drbd_destroy_mempools();
2405 unregister_blkdev(DRBD_MAJOR, "drbd");
2407 idr_destroy(&drbd_devices);
2409 pr_info("module cleanup done.\n");
2413 * drbd_congested() - Callback for the flusher thread
2414 * @congested_data: User data
2415 * @bdi_bits: Bits the BDI flusher thread is currently interested in
2417 * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested.
2419 static int drbd_congested(void *congested_data, int bdi_bits)
2421 struct drbd_device *device = congested_data;
2422 struct request_queue *q;
2426 if (!may_inc_ap_bio(device)) {
2427 /* DRBD has frozen IO */
2433 if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
2434 r |= (1 << WB_async_congested);
2435 /* Without good local data, we would need to read from remote,
2436 * and that would need the worker thread as well, which is
2437 * currently blocked waiting for that usermode helper to
2440 if (!get_ldev_if_state(device, D_UP_TO_DATE))
2441 r |= (1 << WB_sync_congested);
2449 if (get_ldev(device)) {
2450 q = bdev_get_queue(device->ldev->backing_bdev);
2451 r = bdi_congested(q->backing_dev_info, bdi_bits);
2457 if (bdi_bits & (1 << WB_async_congested) &&
2458 test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
2459 r |= (1 << WB_async_congested);
2460 reason = reason == 'b' ? 'a' : 'n';
2464 device->congestion_reason = reason;
2468 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2470 spin_lock_init(&wq->q_lock);
2471 INIT_LIST_HEAD(&wq->q);
2472 init_waitqueue_head(&wq->q_wait);
2475 struct completion_work {
2477 struct completion done;
2480 static int w_complete(struct drbd_work *w, int cancel)
2482 struct completion_work *completion_work =
2483 container_of(w, struct completion_work, w);
2485 complete(&completion_work->done);
2489 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2491 struct completion_work completion_work;
2493 completion_work.w.cb = w_complete;
2494 init_completion(&completion_work.done);
2495 drbd_queue_work(work_queue, &completion_work.w);
2496 wait_for_completion(&completion_work.done);
2499 struct drbd_resource *drbd_find_resource(const char *name)
2501 struct drbd_resource *resource;
2503 if (!name || !name[0])
2507 for_each_resource_rcu(resource, &drbd_resources) {
2508 if (!strcmp(resource->name, name)) {
2509 kref_get(&resource->kref);
2519 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2520 void *peer_addr, int peer_addr_len)
2522 struct drbd_resource *resource;
2523 struct drbd_connection *connection;
2526 for_each_resource_rcu(resource, &drbd_resources) {
2527 for_each_connection_rcu(connection, resource) {
2528 if (connection->my_addr_len == my_addr_len &&
2529 connection->peer_addr_len == peer_addr_len &&
2530 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2531 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2532 kref_get(&connection->kref);
2543 static int drbd_alloc_socket(struct drbd_socket *socket)
2545 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2548 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2554 static void drbd_free_socket(struct drbd_socket *socket)
2556 free_page((unsigned long) socket->sbuf);
2557 free_page((unsigned long) socket->rbuf);
2560 void conn_free_crypto(struct drbd_connection *connection)
2562 drbd_free_sock(connection);
2564 crypto_free_ahash(connection->csums_tfm);
2565 crypto_free_ahash(connection->verify_tfm);
2566 crypto_free_shash(connection->cram_hmac_tfm);
2567 crypto_free_ahash(connection->integrity_tfm);
2568 crypto_free_ahash(connection->peer_integrity_tfm);
2569 kfree(connection->int_dig_in);
2570 kfree(connection->int_dig_vv);
2572 connection->csums_tfm = NULL;
2573 connection->verify_tfm = NULL;
2574 connection->cram_hmac_tfm = NULL;
2575 connection->integrity_tfm = NULL;
2576 connection->peer_integrity_tfm = NULL;
2577 connection->int_dig_in = NULL;
2578 connection->int_dig_vv = NULL;
2581 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2583 struct drbd_connection *connection;
2584 cpumask_var_t new_cpu_mask;
2587 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2590 /* silently ignore cpu mask on UP kernel */
2591 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2592 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2593 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2594 if (err == -EOVERFLOW) {
2595 /* So what. mask it out. */
2596 cpumask_var_t tmp_cpu_mask;
2597 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2598 cpumask_setall(tmp_cpu_mask);
2599 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2600 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2602 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2604 free_cpumask_var(tmp_cpu_mask);
2609 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2610 /* retcode = ERR_CPU_MASK_PARSE; */
2614 resource->res_opts = *res_opts;
2615 if (cpumask_empty(new_cpu_mask))
2616 drbd_calc_cpu_mask(&new_cpu_mask);
2617 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2618 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2619 for_each_connection_rcu(connection, resource) {
2620 connection->receiver.reset_cpu_mask = 1;
2621 connection->ack_receiver.reset_cpu_mask = 1;
2622 connection->worker.reset_cpu_mask = 1;
2628 free_cpumask_var(new_cpu_mask);
2633 struct drbd_resource *drbd_create_resource(const char *name)
2635 struct drbd_resource *resource;
2637 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2640 resource->name = kstrdup(name, GFP_KERNEL);
2641 if (!resource->name)
2642 goto fail_free_resource;
2643 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2644 goto fail_free_name;
2645 kref_init(&resource->kref);
2646 idr_init(&resource->devices);
2647 INIT_LIST_HEAD(&resource->connections);
2648 resource->write_ordering = WO_BDEV_FLUSH;
2649 list_add_tail_rcu(&resource->resources, &drbd_resources);
2650 mutex_init(&resource->conf_update);
2651 mutex_init(&resource->adm_mutex);
2652 spin_lock_init(&resource->req_lock);
2653 drbd_debugfs_resource_add(resource);
2657 kfree(resource->name);
2664 /* caller must be under adm_mutex */
2665 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2667 struct drbd_resource *resource;
2668 struct drbd_connection *connection;
2670 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2674 if (drbd_alloc_socket(&connection->data))
2676 if (drbd_alloc_socket(&connection->meta))
2679 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2680 if (!connection->current_epoch)
2683 INIT_LIST_HEAD(&connection->transfer_log);
2685 INIT_LIST_HEAD(&connection->current_epoch->list);
2686 connection->epochs = 1;
2687 spin_lock_init(&connection->epoch_lock);
2689 connection->send.seen_any_write_yet = false;
2690 connection->send.current_epoch_nr = 0;
2691 connection->send.current_epoch_writes = 0;
2693 resource = drbd_create_resource(name);
2697 connection->cstate = C_STANDALONE;
2698 mutex_init(&connection->cstate_mutex);
2699 init_waitqueue_head(&connection->ping_wait);
2700 idr_init(&connection->peer_devices);
2702 drbd_init_workqueue(&connection->sender_work);
2703 mutex_init(&connection->data.mutex);
2704 mutex_init(&connection->meta.mutex);
2706 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2707 connection->receiver.connection = connection;
2708 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2709 connection->worker.connection = connection;
2710 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2711 connection->ack_receiver.connection = connection;
2713 kref_init(&connection->kref);
2715 connection->resource = resource;
2717 if (set_resource_options(resource, res_opts))
2720 kref_get(&resource->kref);
2721 list_add_tail_rcu(&connection->connections, &resource->connections);
2722 drbd_debugfs_connection_add(connection);
2726 list_del(&resource->resources);
2727 drbd_free_resource(resource);
2729 kfree(connection->current_epoch);
2730 drbd_free_socket(&connection->meta);
2731 drbd_free_socket(&connection->data);
2736 void drbd_destroy_connection(struct kref *kref)
2738 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2739 struct drbd_resource *resource = connection->resource;
2741 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
2742 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2743 kfree(connection->current_epoch);
2745 idr_destroy(&connection->peer_devices);
2747 drbd_free_socket(&connection->meta);
2748 drbd_free_socket(&connection->data);
2749 kfree(connection->int_dig_in);
2750 kfree(connection->int_dig_vv);
2751 memset(connection, 0xfc, sizeof(*connection));
2753 kref_put(&resource->kref, drbd_destroy_resource);
2756 static int init_submitter(struct drbd_device *device)
2758 /* opencoded create_singlethread_workqueue(),
2759 * to be able to say "drbd%d", ..., minor */
2761 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2762 if (!device->submit.wq)
2765 INIT_WORK(&device->submit.worker, do_submit);
2766 INIT_LIST_HEAD(&device->submit.writes);
2770 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2772 struct drbd_resource *resource = adm_ctx->resource;
2773 struct drbd_connection *connection;
2774 struct drbd_device *device;
2775 struct drbd_peer_device *peer_device, *tmp_peer_device;
2776 struct gendisk *disk;
2777 struct request_queue *q;
2779 int vnr = adm_ctx->volume;
2780 enum drbd_ret_code err = ERR_NOMEM;
2782 device = minor_to_device(minor);
2784 return ERR_MINOR_OR_VOLUME_EXISTS;
2786 /* GFP_KERNEL, we are outside of all write-out paths */
2787 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2790 kref_init(&device->kref);
2792 kref_get(&resource->kref);
2793 device->resource = resource;
2794 device->minor = minor;
2797 drbd_init_set_defaults(device);
2799 q = blk_alloc_queue_node(GFP_KERNEL, NUMA_NO_NODE, &resource->req_lock);
2802 device->rq_queue = q;
2803 q->queuedata = device;
2805 disk = alloc_disk(1);
2808 device->vdisk = disk;
2810 set_disk_ro(disk, true);
2813 disk->major = DRBD_MAJOR;
2814 disk->first_minor = minor;
2815 disk->fops = &drbd_ops;
2816 sprintf(disk->disk_name, "drbd%d", minor);
2817 disk->private_data = device;
2819 device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2820 /* we have no partitions. we contain only ourselves. */
2821 device->this_bdev->bd_contains = device->this_bdev;
2823 q->backing_dev_info->congested_fn = drbd_congested;
2824 q->backing_dev_info->congested_data = device;
2826 blk_queue_make_request(q, drbd_make_request);
2827 blk_queue_write_cache(q, true, true);
2828 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2829 This triggers a max_bio_size message upon first attach or connect */
2830 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2832 device->md_io.page = alloc_page(GFP_KERNEL);
2833 if (!device->md_io.page)
2834 goto out_no_io_page;
2836 if (drbd_bm_init(device))
2838 device->read_requests = RB_ROOT;
2839 device->write_requests = RB_ROOT;
2841 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2844 err = ERR_MINOR_OR_VOLUME_EXISTS;
2845 goto out_no_minor_idr;
2847 kref_get(&device->kref);
2849 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2852 err = ERR_MINOR_OR_VOLUME_EXISTS;
2853 goto out_idr_remove_minor;
2855 kref_get(&device->kref);
2857 INIT_LIST_HEAD(&device->peer_devices);
2858 INIT_LIST_HEAD(&device->pending_bitmap_io);
2859 for_each_connection(connection, resource) {
2860 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2862 goto out_idr_remove_from_resource;
2863 peer_device->connection = connection;
2864 peer_device->device = device;
2866 list_add(&peer_device->peer_devices, &device->peer_devices);
2867 kref_get(&device->kref);
2869 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2872 err = ERR_INVALID_REQUEST;
2873 goto out_idr_remove_from_resource;
2875 kref_get(&connection->kref);
2876 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2879 if (init_submitter(device)) {
2881 goto out_idr_remove_vol;
2886 /* inherit the connection state */
2887 device->state.conn = first_connection(resource)->cstate;
2888 if (device->state.conn == C_WF_REPORT_PARAMS) {
2889 for_each_peer_device(peer_device, device)
2890 drbd_connected(peer_device);
2892 /* move to create_peer_device() */
2893 for_each_peer_device(peer_device, device)
2894 drbd_debugfs_peer_device_add(peer_device);
2895 drbd_debugfs_device_add(device);
2899 idr_remove(&connection->peer_devices, vnr);
2900 out_idr_remove_from_resource:
2901 for_each_connection(connection, resource) {
2902 peer_device = idr_remove(&connection->peer_devices, vnr);
2904 kref_put(&connection->kref, drbd_destroy_connection);
2906 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2907 list_del(&peer_device->peer_devices);
2910 idr_remove(&resource->devices, vnr);
2911 out_idr_remove_minor:
2912 idr_remove(&drbd_devices, minor);
2915 drbd_bm_cleanup(device);
2917 __free_page(device->md_io.page);
2921 blk_cleanup_queue(q);
2923 kref_put(&resource->kref, drbd_destroy_resource);
2928 void drbd_delete_device(struct drbd_device *device)
2930 struct drbd_resource *resource = device->resource;
2931 struct drbd_connection *connection;
2932 struct drbd_peer_device *peer_device;
2934 /* move to free_peer_device() */
2935 for_each_peer_device(peer_device, device)
2936 drbd_debugfs_peer_device_cleanup(peer_device);
2937 drbd_debugfs_device_cleanup(device);
2938 for_each_connection(connection, resource) {
2939 idr_remove(&connection->peer_devices, device->vnr);
2940 kref_put(&device->kref, drbd_destroy_device);
2942 idr_remove(&resource->devices, device->vnr);
2943 kref_put(&device->kref, drbd_destroy_device);
2944 idr_remove(&drbd_devices, device_to_minor(device));
2945 kref_put(&device->kref, drbd_destroy_device);
2946 del_gendisk(device->vdisk);
2948 kref_put(&device->kref, drbd_destroy_device);
2951 static int __init drbd_init(void)
2955 if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2956 pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2960 drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2964 err = register_blkdev(DRBD_MAJOR, "drbd");
2966 pr_err("unable to register block device major %d\n",
2972 * allocate all necessary structs
2974 init_waitqueue_head(&drbd_pp_wait);
2976 drbd_proc = NULL; /* play safe for drbd_cleanup */
2977 idr_init(&drbd_devices);
2979 mutex_init(&resources_mutex);
2980 INIT_LIST_HEAD(&drbd_resources);
2982 err = drbd_genl_register();
2984 pr_err("unable to register generic netlink family\n");
2988 err = drbd_create_mempools();
2993 drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show);
2995 pr_err("unable to register proc file\n");
2999 retry.wq = create_singlethread_workqueue("drbd-reissue");
3001 pr_err("unable to create retry workqueue\n");
3004 INIT_WORK(&retry.worker, do_retry);
3005 spin_lock_init(&retry.lock);
3006 INIT_LIST_HEAD(&retry.writes);
3008 if (drbd_debugfs_init())
3009 pr_notice("failed to initialize debugfs -- will not be available\n");
3011 pr_info("initialized. "
3012 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
3013 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
3014 pr_info("%s\n", drbd_buildtag());
3015 pr_info("registered as block device major %d\n", DRBD_MAJOR);
3016 return 0; /* Success! */
3021 pr_err("ran out of memory\n");
3023 pr_err("initialization failure\n");
3027 static void drbd_free_one_sock(struct drbd_socket *ds)
3030 mutex_lock(&ds->mutex);
3033 mutex_unlock(&ds->mutex);
3035 /* so debugfs does not need to mutex_lock() */
3037 kernel_sock_shutdown(s, SHUT_RDWR);
3042 void drbd_free_sock(struct drbd_connection *connection)
3044 if (connection->data.socket)
3045 drbd_free_one_sock(&connection->data);
3046 if (connection->meta.socket)
3047 drbd_free_one_sock(&connection->meta);
3050 /* meta data management */
3052 void conn_md_sync(struct drbd_connection *connection)
3054 struct drbd_peer_device *peer_device;
3058 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3059 struct drbd_device *device = peer_device->device;
3061 kref_get(&device->kref);
3063 drbd_md_sync(device);
3064 kref_put(&device->kref, drbd_destroy_device);
3070 /* aligned 4kByte */
3071 struct meta_data_on_disk {
3072 u64 la_size_sect; /* last agreed size. */
3073 u64 uuid[UI_SIZE]; /* UUIDs. */
3076 u32 flags; /* MDF */
3079 u32 al_offset; /* offset to this block */
3080 u32 al_nr_extents; /* important for restoring the AL (userspace) */
3081 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3082 u32 bm_offset; /* offset to the bitmap, from here */
3083 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
3084 u32 la_peer_max_bio_size; /* last peer max_bio_size */
3086 /* see al_tr_number_to_on_disk_sector() */
3088 u32 al_stripe_size_4k;
3090 u8 reserved_u8[4096 - (7*8 + 10*4)];
3095 void drbd_md_write(struct drbd_device *device, void *b)
3097 struct meta_data_on_disk *buffer = b;
3101 memset(buffer, 0, sizeof(*buffer));
3103 buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3104 for (i = UI_CURRENT; i < UI_SIZE; i++)
3105 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3106 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3107 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3109 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3110 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3111 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3112 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3113 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3115 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3116 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3118 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3119 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3121 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3122 sector = device->ldev->md.md_offset;
3124 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3125 /* this was a try anyways ... */
3126 drbd_err(device, "meta data update failed!\n");
3127 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3132 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3133 * @device: DRBD device.
3135 void drbd_md_sync(struct drbd_device *device)
3137 struct meta_data_on_disk *buffer;
3139 /* Don't accidentally change the DRBD meta data layout. */
3140 BUILD_BUG_ON(UI_SIZE != 4);
3141 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3143 del_timer(&device->md_sync_timer);
3144 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3145 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3148 /* We use here D_FAILED and not D_ATTACHING because we try to write
3149 * metadata even if we detach due to a disk failure! */
3150 if (!get_ldev_if_state(device, D_FAILED))
3153 buffer = drbd_md_get_buffer(device, __func__);
3157 drbd_md_write(device, buffer);
3159 /* Update device->ldev->md.la_size_sect,
3160 * since we updated it on metadata. */
3161 device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3163 drbd_md_put_buffer(device);
3168 static int check_activity_log_stripe_size(struct drbd_device *device,
3169 struct meta_data_on_disk *on_disk,
3170 struct drbd_md *in_core)
3172 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3173 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3176 /* both not set: default to old fixed size activity log */
3177 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3179 al_stripe_size_4k = MD_32kB_SECT/8;
3182 /* some paranoia plausibility checks */
3184 /* we need both values to be set */
3185 if (al_stripes == 0 || al_stripe_size_4k == 0)
3188 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3190 /* Upper limit of activity log area, to avoid potential overflow
3191 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3192 * than 72 * 4k blocks total only increases the amount of history,
3193 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3194 if (al_size_4k > (16 * 1024 * 1024/4))
3197 /* Lower limit: we need at least 8 transaction slots (32kB)
3198 * to not break existing setups */
3199 if (al_size_4k < MD_32kB_SECT/8)
3202 in_core->al_stripe_size_4k = al_stripe_size_4k;
3203 in_core->al_stripes = al_stripes;
3204 in_core->al_size_4k = al_size_4k;
3208 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3209 al_stripes, al_stripe_size_4k);
3213 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3215 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3216 struct drbd_md *in_core = &bdev->md;
3217 s32 on_disk_al_sect;
3218 s32 on_disk_bm_sect;
3220 /* The on-disk size of the activity log, calculated from offsets, and
3221 * the size of the activity log calculated from the stripe settings,
3223 * Though we could relax this a bit: it is ok, if the striped activity log
3224 * fits in the available on-disk activity log size.
3225 * Right now, that would break how resize is implemented.
3226 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3227 * of possible unused padding space in the on disk layout. */
3228 if (in_core->al_offset < 0) {
3229 if (in_core->bm_offset > in_core->al_offset)
3231 on_disk_al_sect = -in_core->al_offset;
3232 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3234 if (in_core->al_offset != MD_4kB_SECT)
3236 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3239 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3240 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3243 /* old fixed size meta data is exactly that: fixed. */
3244 if (in_core->meta_dev_idx >= 0) {
3245 if (in_core->md_size_sect != MD_128MB_SECT
3246 || in_core->al_offset != MD_4kB_SECT
3247 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3248 || in_core->al_stripes != 1
3249 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3253 if (capacity < in_core->md_size_sect)
3255 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3258 /* should be aligned, and at least 32k */
3259 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3262 /* should fit (for now: exactly) into the available on-disk space;
3263 * overflow prevention is in check_activity_log_stripe_size() above. */
3264 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3267 /* again, should be aligned */
3268 if (in_core->bm_offset & 7)
3271 /* FIXME check for device grow with flex external meta data? */
3273 /* can the available bitmap space cover the last agreed device size? */
3274 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3280 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3281 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3282 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3283 in_core->meta_dev_idx,
3284 in_core->al_stripes, in_core->al_stripe_size_4k,
3285 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3286 (unsigned long long)in_core->la_size_sect,
3287 (unsigned long long)capacity);
3294 * drbd_md_read() - Reads in the meta data super block
3295 * @device: DRBD device.
3296 * @bdev: Device from which the meta data should be read in.
3298 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3299 * something goes wrong.
3301 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3302 * even before @bdev is assigned to @device->ldev.
3304 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3306 struct meta_data_on_disk *buffer;
3308 int i, rv = NO_ERROR;
3310 if (device->state.disk != D_DISKLESS)
3311 return ERR_DISK_CONFIGURED;
3313 buffer = drbd_md_get_buffer(device, __func__);
3317 /* First, figure out where our meta data superblock is located,
3319 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3320 bdev->md.md_offset = drbd_md_ss(bdev);
3321 /* Even for (flexible or indexed) external meta data,
3322 * initially restrict us to the 4k superblock for now.
3323 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3324 bdev->md.md_size_sect = 8;
3326 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3328 /* NOTE: can't do normal error processing here as this is
3329 called BEFORE disk is attached */
3330 drbd_err(device, "Error while reading metadata.\n");
3331 rv = ERR_IO_MD_DISK;
3335 magic = be32_to_cpu(buffer->magic);
3336 flags = be32_to_cpu(buffer->flags);
3337 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3338 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3339 /* btw: that's Activity Log clean, not "all" clean. */
3340 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3341 rv = ERR_MD_UNCLEAN;
3345 rv = ERR_MD_INVALID;
3346 if (magic != DRBD_MD_MAGIC_08) {
3347 if (magic == DRBD_MD_MAGIC_07)
3348 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3350 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3354 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3355 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3356 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3361 /* convert to in_core endian */
3362 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3363 for (i = UI_CURRENT; i < UI_SIZE; i++)
3364 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3365 bdev->md.flags = be32_to_cpu(buffer->flags);
3366 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3368 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3369 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3370 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3372 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3374 if (check_offsets_and_sizes(device, bdev))
3377 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3378 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3379 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3382 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3383 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3384 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3390 spin_lock_irq(&device->resource->req_lock);
3391 if (device->state.conn < C_CONNECTED) {
3393 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3394 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3395 device->peer_max_bio_size = peer;
3397 spin_unlock_irq(&device->resource->req_lock);
3400 drbd_md_put_buffer(device);
3406 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3407 * @device: DRBD device.
3409 * Call this function if you change anything that should be written to
3410 * the meta-data super block. This function sets MD_DIRTY, and starts a
3411 * timer that ensures that within five seconds you have to call drbd_md_sync().
3414 void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
3416 if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3417 mod_timer(&device->md_sync_timer, jiffies + HZ);
3418 device->last_md_mark_dirty.line = line;
3419 device->last_md_mark_dirty.func = func;
3423 void drbd_md_mark_dirty(struct drbd_device *device)
3425 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3426 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3430 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3434 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3435 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3438 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3440 if (idx == UI_CURRENT) {
3441 if (device->state.role == R_PRIMARY)
3446 drbd_set_ed_uuid(device, val);
3449 device->ldev->md.uuid[idx] = val;
3450 drbd_md_mark_dirty(device);
3453 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3455 unsigned long flags;
3456 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3457 __drbd_uuid_set(device, idx, val);
3458 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3461 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3463 unsigned long flags;
3464 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3465 if (device->ldev->md.uuid[idx]) {
3466 drbd_uuid_move_history(device);
3467 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3469 __drbd_uuid_set(device, idx, val);
3470 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3474 * drbd_uuid_new_current() - Creates a new current UUID
3475 * @device: DRBD device.
3477 * Creates a new current UUID, and rotates the old current UUID into
3478 * the bitmap slot. Causes an incremental resync upon next connect.
3480 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3483 unsigned long long bm_uuid;
3485 get_random_bytes(&val, sizeof(u64));
3487 spin_lock_irq(&device->ldev->md.uuid_lock);
3488 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3491 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3493 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3494 __drbd_uuid_set(device, UI_CURRENT, val);
3495 spin_unlock_irq(&device->ldev->md.uuid_lock);
3497 drbd_print_uuids(device, "new current UUID");
3498 /* get it to stable storage _now_ */
3499 drbd_md_sync(device);
3502 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3504 unsigned long flags;
3505 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3508 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3510 drbd_uuid_move_history(device);
3511 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3512 device->ldev->md.uuid[UI_BITMAP] = 0;
3514 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3516 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3518 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3520 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3522 drbd_md_mark_dirty(device);
3526 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3527 * @device: DRBD device.
3529 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3531 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3535 drbd_md_set_flag(device, MDF_FULL_SYNC);
3536 drbd_md_sync(device);
3537 drbd_bm_set_all(device);
3539 rv = drbd_bm_write(device);
3542 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3543 drbd_md_sync(device);
3550 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3551 * @device: DRBD device.
3553 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3555 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3557 drbd_resume_al(device);
3558 drbd_bm_clear_all(device);
3559 return drbd_bm_write(device);
3562 static int w_bitmap_io(struct drbd_work *w, int unused)
3564 struct drbd_device *device =
3565 container_of(w, struct drbd_device, bm_io_work.w);
3566 struct bm_io_work *work = &device->bm_io_work;
3569 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3570 int cnt = atomic_read(&device->ap_bio_cnt);
3572 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3576 if (get_ldev(device)) {
3577 drbd_bm_lock(device, work->why, work->flags);
3578 rv = work->io_fn(device);
3579 drbd_bm_unlock(device);
3583 clear_bit_unlock(BITMAP_IO, &device->flags);
3584 wake_up(&device->misc_wait);
3587 work->done(device, rv);
3589 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3597 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3598 * @device: DRBD device.
3599 * @io_fn: IO callback to be called when bitmap IO is possible
3600 * @done: callback to be called after the bitmap IO was performed
3601 * @why: Descriptive text of the reason for doing the IO
3603 * While IO on the bitmap happens we freeze application IO thus we ensure
3604 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3605 * called from worker context. It MUST NOT be used while a previous such
3606 * work is still pending!
3608 * Its worker function encloses the call of io_fn() by get_ldev() and
3611 void drbd_queue_bitmap_io(struct drbd_device *device,
3612 int (*io_fn)(struct drbd_device *),
3613 void (*done)(struct drbd_device *, int),
3614 char *why, enum bm_flag flags)
3616 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3618 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3619 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3620 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3621 if (device->bm_io_work.why)
3622 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3623 why, device->bm_io_work.why);
3625 device->bm_io_work.io_fn = io_fn;
3626 device->bm_io_work.done = done;
3627 device->bm_io_work.why = why;
3628 device->bm_io_work.flags = flags;
3630 spin_lock_irq(&device->resource->req_lock);
3631 set_bit(BITMAP_IO, &device->flags);
3632 /* don't wait for pending application IO if the caller indicates that
3633 * application IO does not conflict anyways. */
3634 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3635 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3636 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3637 &device->bm_io_work.w);
3639 spin_unlock_irq(&device->resource->req_lock);
3643 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3644 * @device: DRBD device.
3645 * @io_fn: IO callback to be called when bitmap IO is possible
3646 * @why: Descriptive text of the reason for doing the IO
3648 * freezes application IO while that the actual IO operations runs. This
3649 * functions MAY NOT be called from worker context.
3651 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3652 char *why, enum bm_flag flags)
3654 /* Only suspend io, if some operation is supposed to be locked out */
3655 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3658 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3661 drbd_suspend_io(device);
3663 drbd_bm_lock(device, why, flags);
3665 drbd_bm_unlock(device);
3668 drbd_resume_io(device);
3673 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3675 if ((device->ldev->md.flags & flag) != flag) {
3676 drbd_md_mark_dirty(device);
3677 device->ldev->md.flags |= flag;
3681 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3683 if ((device->ldev->md.flags & flag) != 0) {
3684 drbd_md_mark_dirty(device);
3685 device->ldev->md.flags &= ~flag;
3688 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3690 return (bdev->md.flags & flag) != 0;
3693 static void md_sync_timer_fn(struct timer_list *t)
3695 struct drbd_device *device = from_timer(device, t, md_sync_timer);
3696 drbd_device_post_work(device, MD_SYNC);
3699 const char *cmdname(enum drbd_packet cmd)
3701 /* THINK may need to become several global tables
3702 * when we want to support more than
3703 * one PRO_VERSION */
3704 static const char *cmdnames[] = {
3706 [P_WSAME] = "WriteSame",
3708 [P_DATA_REPLY] = "DataReply",
3709 [P_RS_DATA_REPLY] = "RSDataReply",
3710 [P_BARRIER] = "Barrier",
3711 [P_BITMAP] = "ReportBitMap",
3712 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3713 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3714 [P_UNPLUG_REMOTE] = "UnplugRemote",
3715 [P_DATA_REQUEST] = "DataRequest",
3716 [P_RS_DATA_REQUEST] = "RSDataRequest",
3717 [P_SYNC_PARAM] = "SyncParam",
3718 [P_SYNC_PARAM89] = "SyncParam89",
3719 [P_PROTOCOL] = "ReportProtocol",
3720 [P_UUIDS] = "ReportUUIDs",
3721 [P_SIZES] = "ReportSizes",
3722 [P_STATE] = "ReportState",
3723 [P_SYNC_UUID] = "ReportSyncUUID",
3724 [P_AUTH_CHALLENGE] = "AuthChallenge",
3725 [P_AUTH_RESPONSE] = "AuthResponse",
3727 [P_PING_ACK] = "PingAck",
3728 [P_RECV_ACK] = "RecvAck",
3729 [P_WRITE_ACK] = "WriteAck",
3730 [P_RS_WRITE_ACK] = "RSWriteAck",
3731 [P_SUPERSEDED] = "Superseded",
3732 [P_NEG_ACK] = "NegAck",
3733 [P_NEG_DREPLY] = "NegDReply",
3734 [P_NEG_RS_DREPLY] = "NegRSDReply",
3735 [P_BARRIER_ACK] = "BarrierAck",
3736 [P_STATE_CHG_REQ] = "StateChgRequest",
3737 [P_STATE_CHG_REPLY] = "StateChgReply",
3738 [P_OV_REQUEST] = "OVRequest",
3739 [P_OV_REPLY] = "OVReply",
3740 [P_OV_RESULT] = "OVResult",
3741 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3742 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3743 [P_COMPRESSED_BITMAP] = "CBitmap",
3744 [P_DELAY_PROBE] = "DelayProbe",
3745 [P_OUT_OF_SYNC] = "OutOfSync",
3746 [P_RETRY_WRITE] = "RetryWrite",
3747 [P_RS_CANCEL] = "RSCancel",
3748 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3749 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3750 [P_RETRY_WRITE] = "retry_write",
3751 [P_PROTOCOL_UPDATE] = "protocol_update",
3752 [P_RS_THIN_REQ] = "rs_thin_req",
3753 [P_RS_DEALLOCATED] = "rs_deallocated",
3755 /* enum drbd_packet, but not commands - obsoleted flags:
3761 /* too big for the array: 0xfffX */
3762 if (cmd == P_INITIAL_META)
3763 return "InitialMeta";
3764 if (cmd == P_INITIAL_DATA)
3765 return "InitialData";
3766 if (cmd == P_CONNECTION_FEATURES)
3767 return "ConnectionFeatures";
3768 if (cmd >= ARRAY_SIZE(cmdnames))
3770 return cmdnames[cmd];
3774 * drbd_wait_misc - wait for a request to make progress
3775 * @device: device associated with the request
3776 * @i: the struct drbd_interval embedded in struct drbd_request or
3777 * struct drbd_peer_request
3779 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3781 struct net_conf *nc;
3786 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3791 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3794 /* Indicate to wake up device->misc_wait on progress. */
3796 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3797 spin_unlock_irq(&device->resource->req_lock);
3798 timeout = schedule_timeout(timeout);
3799 finish_wait(&device->misc_wait, &wait);
3800 spin_lock_irq(&device->resource->req_lock);
3801 if (!timeout || device->state.conn < C_CONNECTED)
3803 if (signal_pending(current))
3804 return -ERESTARTSYS;
3808 void lock_all_resources(void)
3810 struct drbd_resource *resource;
3811 int __maybe_unused i = 0;
3813 mutex_lock(&resources_mutex);
3814 local_irq_disable();
3815 for_each_resource(resource, &drbd_resources)
3816 spin_lock_nested(&resource->req_lock, i++);
3819 void unlock_all_resources(void)
3821 struct drbd_resource *resource;
3823 for_each_resource(resource, &drbd_resources)
3824 spin_unlock(&resource->req_lock);
3826 mutex_unlock(&resources_mutex);
3829 #ifdef CONFIG_DRBD_FAULT_INJECTION
3830 /* Fault insertion support including random number generator shamelessly
3831 * stolen from kernel/rcutorture.c */
3832 struct fault_random_state {
3833 unsigned long state;
3834 unsigned long count;
3837 #define FAULT_RANDOM_MULT 39916801 /* prime */
3838 #define FAULT_RANDOM_ADD 479001701 /* prime */
3839 #define FAULT_RANDOM_REFRESH 10000
3842 * Crude but fast random-number generator. Uses a linear congruential
3843 * generator, with occasional help from get_random_bytes().
3845 static unsigned long
3846 _drbd_fault_random(struct fault_random_state *rsp)
3850 if (!rsp->count--) {
3851 get_random_bytes(&refresh, sizeof(refresh));
3852 rsp->state += refresh;
3853 rsp->count = FAULT_RANDOM_REFRESH;
3855 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3856 return swahw32(rsp->state);
3860 _drbd_fault_str(unsigned int type) {
3861 static char *_faults[] = {
3862 [DRBD_FAULT_MD_WR] = "Meta-data write",
3863 [DRBD_FAULT_MD_RD] = "Meta-data read",
3864 [DRBD_FAULT_RS_WR] = "Resync write",
3865 [DRBD_FAULT_RS_RD] = "Resync read",
3866 [DRBD_FAULT_DT_WR] = "Data write",
3867 [DRBD_FAULT_DT_RD] = "Data read",
3868 [DRBD_FAULT_DT_RA] = "Data read ahead",
3869 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3870 [DRBD_FAULT_AL_EE] = "EE allocation",
3871 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3874 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3878 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3880 static struct fault_random_state rrs = {0, 0};
3882 unsigned int ret = (
3883 (drbd_fault_devs == 0 ||
3884 ((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3885 (((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3890 if (__ratelimit(&drbd_ratelimit_state))
3891 drbd_warn(device, "***Simulating %s failure\n",
3892 _drbd_fault_str(type));
3899 const char *drbd_buildtag(void)
3901 /* DRBD built from external sources has here a reference to the
3902 git hash of the source code. */
3904 static char buildtag[38] = "\0uilt-in";
3906 if (buildtag[0] == 0) {
3908 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3917 module_init(drbd_init)
3918 module_exit(drbd_cleanup)
3920 EXPORT_SYMBOL(drbd_conn_str);
3921 EXPORT_SYMBOL(drbd_role_str);
3922 EXPORT_SYMBOL(drbd_disk_str);
3923 EXPORT_SYMBOL(drbd_set_st_err_str);