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(unsigned long data);
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 /* allow_open_on_secondary */
81 MODULE_PARM_DESC(allow_oos, "DONT USE!");
82 /* thanks to these macros, if compiled into the kernel (not-module),
83 * this becomes the boot parameter drbd.minor_count */
84 module_param(minor_count, uint, 0444);
85 module_param(disable_sendpage, bool, 0644);
86 module_param(allow_oos, bool, 0);
87 module_param(proc_details, int, 0644);
89 #ifdef CONFIG_DRBD_FAULT_INJECTION
92 static int fault_count;
94 /* bitmap of enabled faults */
95 module_param(enable_faults, int, 0664);
96 /* fault rate % value - applies to all enabled faults */
97 module_param(fault_rate, int, 0664);
98 /* count of faults inserted */
99 module_param(fault_count, int, 0664);
100 /* bitmap of devices to insert faults on */
101 module_param(fault_devs, int, 0644);
104 /* module parameter, defined */
105 unsigned int minor_count = DRBD_MINOR_COUNT_DEF;
106 bool disable_sendpage;
108 int proc_details; /* Detail level in proc drbd*/
110 /* Module parameter for setting the user mode helper program
111 * to run. Default is /sbin/drbdadm */
112 char usermode_helper[80] = "/sbin/drbdadm";
114 module_param_string(usermode_helper, usermode_helper, sizeof(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;
132 /* I do not use a standard mempool, because:
133 1) I want to hand out the pre-allocated objects first.
134 2) I want to be able to interrupt sleeping allocation with a signal.
135 Note: This is a single linked list, the next pointer is the private
136 member of struct page.
138 struct page *drbd_pp_pool;
139 spinlock_t drbd_pp_lock;
141 wait_queue_head_t drbd_pp_wait;
143 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
145 static const struct block_device_operations drbd_ops = {
146 .owner = THIS_MODULE,
148 .release = drbd_release,
151 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
155 if (!drbd_md_io_bio_set)
156 return bio_alloc(gfp_mask, 1);
158 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
165 /* When checking with sparse, and this is an inline function, sparse will
166 give tons of false positives. When this is a real functions sparse works.
168 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
172 atomic_inc(&device->local_cnt);
173 io_allowed = (device->state.disk >= mins);
175 if (atomic_dec_and_test(&device->local_cnt))
176 wake_up(&device->misc_wait);
184 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
185 * @connection: DRBD connection.
186 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
187 * @set_size: Expected number of requests before that barrier.
189 * In case the passed barrier_nr or set_size does not match the oldest
190 * epoch of not yet barrier-acked requests, this function will cause a
191 * termination of the connection.
193 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
194 unsigned int set_size)
196 struct drbd_request *r;
197 struct drbd_request *req = NULL;
198 int expect_epoch = 0;
201 spin_lock_irq(&connection->resource->req_lock);
203 /* find oldest not yet barrier-acked write request,
204 * count writes in its epoch. */
205 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
206 const unsigned s = r->rq_state;
210 if (!(s & RQ_NET_MASK))
215 expect_epoch = req->epoch;
218 if (r->epoch != expect_epoch)
222 /* if (s & RQ_DONE): not expected */
223 /* if (!(s & RQ_NET_MASK)): not expected */
228 /* first some paranoia code */
230 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
234 if (expect_epoch != barrier_nr) {
235 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
236 barrier_nr, expect_epoch);
240 if (expect_size != set_size) {
241 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
242 barrier_nr, set_size, expect_size);
246 /* Clean up list of requests processed during current epoch. */
247 /* this extra list walk restart is paranoia,
248 * to catch requests being barrier-acked "unexpectedly".
249 * It usually should find the same req again, or some READ preceding it. */
250 list_for_each_entry(req, &connection->transfer_log, tl_requests)
251 if (req->epoch == expect_epoch)
253 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
254 if (req->epoch != expect_epoch)
256 _req_mod(req, BARRIER_ACKED);
258 spin_unlock_irq(&connection->resource->req_lock);
263 spin_unlock_irq(&connection->resource->req_lock);
264 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
269 * _tl_restart() - Walks the transfer log, and applies an action to all requests
270 * @connection: DRBD connection to operate on.
271 * @what: The action/event to perform with all request objects
273 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
274 * RESTART_FROZEN_DISK_IO.
276 /* must hold resource->req_lock */
277 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
279 struct drbd_request *req, *r;
281 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
285 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
287 spin_lock_irq(&connection->resource->req_lock);
288 _tl_restart(connection, what);
289 spin_unlock_irq(&connection->resource->req_lock);
293 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
294 * @device: DRBD device.
296 * This is called after the connection to the peer was lost. The storage covered
297 * by the requests on the transfer gets marked as our of sync. Called from the
298 * receiver thread and the worker thread.
300 void tl_clear(struct drbd_connection *connection)
302 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
306 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
307 * @device: DRBD device.
309 void tl_abort_disk_io(struct drbd_device *device)
311 struct drbd_connection *connection = first_peer_device(device)->connection;
312 struct drbd_request *req, *r;
314 spin_lock_irq(&connection->resource->req_lock);
315 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
316 if (!(req->rq_state & RQ_LOCAL_PENDING))
318 if (req->device != device)
320 _req_mod(req, ABORT_DISK_IO);
322 spin_unlock_irq(&connection->resource->req_lock);
325 static int drbd_thread_setup(void *arg)
327 struct drbd_thread *thi = (struct drbd_thread *) arg;
328 struct drbd_resource *resource = thi->resource;
332 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
337 retval = thi->function(thi);
339 spin_lock_irqsave(&thi->t_lock, flags);
341 /* if the receiver has been "EXITING", the last thing it did
342 * was set the conn state to "StandAlone",
343 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
344 * and receiver thread will be "started".
345 * drbd_thread_start needs to set "RESTARTING" in that case.
346 * t_state check and assignment needs to be within the same spinlock,
347 * so either thread_start sees EXITING, and can remap to RESTARTING,
348 * or thread_start see NONE, and can proceed as normal.
351 if (thi->t_state == RESTARTING) {
352 drbd_info(resource, "Restarting %s thread\n", thi->name);
353 thi->t_state = RUNNING;
354 spin_unlock_irqrestore(&thi->t_lock, flags);
361 complete_all(&thi->stop);
362 spin_unlock_irqrestore(&thi->t_lock, flags);
364 drbd_info(resource, "Terminating %s\n", current->comm);
366 /* Release mod reference taken when thread was started */
369 kref_put(&thi->connection->kref, drbd_destroy_connection);
370 kref_put(&resource->kref, drbd_destroy_resource);
371 module_put(THIS_MODULE);
375 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
376 int (*func) (struct drbd_thread *), const char *name)
378 spin_lock_init(&thi->t_lock);
381 thi->function = func;
382 thi->resource = resource;
383 thi->connection = NULL;
387 int drbd_thread_start(struct drbd_thread *thi)
389 struct drbd_resource *resource = thi->resource;
390 struct task_struct *nt;
393 /* is used from state engine doing drbd_thread_stop_nowait,
394 * while holding the req lock irqsave */
395 spin_lock_irqsave(&thi->t_lock, flags);
397 switch (thi->t_state) {
399 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
400 thi->name, current->comm, current->pid);
402 /* Get ref on module for thread - this is released when thread exits */
403 if (!try_module_get(THIS_MODULE)) {
404 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
405 spin_unlock_irqrestore(&thi->t_lock, flags);
409 kref_get(&resource->kref);
411 kref_get(&thi->connection->kref);
413 init_completion(&thi->stop);
414 thi->reset_cpu_mask = 1;
415 thi->t_state = RUNNING;
416 spin_unlock_irqrestore(&thi->t_lock, flags);
417 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
419 nt = kthread_create(drbd_thread_setup, (void *) thi,
420 "drbd_%c_%s", thi->name[0], thi->resource->name);
423 drbd_err(resource, "Couldn't start thread\n");
426 kref_put(&thi->connection->kref, drbd_destroy_connection);
427 kref_put(&resource->kref, drbd_destroy_resource);
428 module_put(THIS_MODULE);
431 spin_lock_irqsave(&thi->t_lock, flags);
433 thi->t_state = RUNNING;
434 spin_unlock_irqrestore(&thi->t_lock, flags);
438 thi->t_state = RESTARTING;
439 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
440 thi->name, current->comm, current->pid);
445 spin_unlock_irqrestore(&thi->t_lock, flags);
453 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
457 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
459 /* may be called from state engine, holding the req lock irqsave */
460 spin_lock_irqsave(&thi->t_lock, flags);
462 if (thi->t_state == NONE) {
463 spin_unlock_irqrestore(&thi->t_lock, flags);
465 drbd_thread_start(thi);
469 if (thi->t_state != ns) {
470 if (thi->task == NULL) {
471 spin_unlock_irqrestore(&thi->t_lock, flags);
477 init_completion(&thi->stop);
478 if (thi->task != current)
479 force_sig(DRBD_SIGKILL, thi->task);
482 spin_unlock_irqrestore(&thi->t_lock, flags);
485 wait_for_completion(&thi->stop);
488 int conn_lowest_minor(struct drbd_connection *connection)
490 struct drbd_peer_device *peer_device;
491 int vnr = 0, minor = -1;
494 peer_device = idr_get_next(&connection->peer_devices, &vnr);
496 minor = device_to_minor(peer_device->device);
504 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
506 * Forces all threads of a resource onto the same CPU. This is beneficial for
507 * DRBD's performance. May be overwritten by user's configuration.
509 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
511 unsigned int *resources_per_cpu, min_index = ~0;
513 resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL);
514 if (resources_per_cpu) {
515 struct drbd_resource *resource;
516 unsigned int cpu, min = ~0;
519 for_each_resource_rcu(resource, &drbd_resources) {
520 for_each_cpu(cpu, resource->cpu_mask)
521 resources_per_cpu[cpu]++;
524 for_each_online_cpu(cpu) {
525 if (resources_per_cpu[cpu] < min) {
526 min = resources_per_cpu[cpu];
530 kfree(resources_per_cpu);
532 if (min_index == ~0) {
533 cpumask_setall(*cpu_mask);
536 cpumask_set_cpu(min_index, *cpu_mask);
540 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
541 * @device: DRBD device.
542 * @thi: drbd_thread object
544 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
547 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
549 struct drbd_resource *resource = thi->resource;
550 struct task_struct *p = current;
552 if (!thi->reset_cpu_mask)
554 thi->reset_cpu_mask = 0;
555 set_cpus_allowed_ptr(p, resource->cpu_mask);
558 #define drbd_calc_cpu_mask(A) ({})
562 * drbd_header_size - size of a packet header
564 * The header size is a multiple of 8, so any payload following the header is
565 * word aligned on 64-bit architectures. (The bitmap send and receive code
568 unsigned int drbd_header_size(struct drbd_connection *connection)
570 if (connection->agreed_pro_version >= 100) {
571 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
572 return sizeof(struct p_header100);
574 BUILD_BUG_ON(sizeof(struct p_header80) !=
575 sizeof(struct p_header95));
576 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
577 return sizeof(struct p_header80);
581 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
583 h->magic = cpu_to_be32(DRBD_MAGIC);
584 h->command = cpu_to_be16(cmd);
585 h->length = cpu_to_be16(size);
586 return sizeof(struct p_header80);
589 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
591 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
592 h->command = cpu_to_be16(cmd);
593 h->length = cpu_to_be32(size);
594 return sizeof(struct p_header95);
597 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
600 h->magic = cpu_to_be32(DRBD_MAGIC_100);
601 h->volume = cpu_to_be16(vnr);
602 h->command = cpu_to_be16(cmd);
603 h->length = cpu_to_be32(size);
605 return sizeof(struct p_header100);
608 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
609 void *buffer, enum drbd_packet cmd, int size)
611 if (connection->agreed_pro_version >= 100)
612 return prepare_header100(buffer, cmd, size, vnr);
613 else if (connection->agreed_pro_version >= 95 &&
614 size > DRBD_MAX_SIZE_H80_PACKET)
615 return prepare_header95(buffer, cmd, size);
617 return prepare_header80(buffer, cmd, size);
620 static void *__conn_prepare_command(struct drbd_connection *connection,
621 struct drbd_socket *sock)
625 return sock->sbuf + drbd_header_size(connection);
628 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
632 mutex_lock(&sock->mutex);
633 p = __conn_prepare_command(connection, sock);
635 mutex_unlock(&sock->mutex);
640 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
642 return conn_prepare_command(peer_device->connection, sock);
645 static int __send_command(struct drbd_connection *connection, int vnr,
646 struct drbd_socket *sock, enum drbd_packet cmd,
647 unsigned int header_size, void *data,
654 * Called with @data == NULL and the size of the data blocks in @size
655 * for commands that send data blocks. For those commands, omit the
656 * MSG_MORE flag: this will increase the likelihood that data blocks
657 * which are page aligned on the sender will end up page aligned on the
660 msg_flags = data ? MSG_MORE : 0;
662 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
664 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
667 err = drbd_send_all(connection, sock->socket, data, size, 0);
668 /* DRBD protocol "pings" are latency critical.
669 * This is supposed to trigger tcp_push_pending_frames() */
670 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
671 drbd_tcp_nodelay(sock->socket);
676 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
677 enum drbd_packet cmd, unsigned int header_size,
678 void *data, unsigned int size)
680 return __send_command(connection, 0, sock, cmd, header_size, data, size);
683 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
684 enum drbd_packet cmd, unsigned int header_size,
685 void *data, unsigned int size)
689 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
690 mutex_unlock(&sock->mutex);
694 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
695 enum drbd_packet cmd, unsigned int header_size,
696 void *data, unsigned int size)
700 err = __send_command(peer_device->connection, peer_device->device->vnr,
701 sock, cmd, header_size, data, size);
702 mutex_unlock(&sock->mutex);
706 int drbd_send_ping(struct drbd_connection *connection)
708 struct drbd_socket *sock;
710 sock = &connection->meta;
711 if (!conn_prepare_command(connection, sock))
713 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
716 int drbd_send_ping_ack(struct drbd_connection *connection)
718 struct drbd_socket *sock;
720 sock = &connection->meta;
721 if (!conn_prepare_command(connection, sock))
723 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
726 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
728 struct drbd_socket *sock;
729 struct p_rs_param_95 *p;
731 const int apv = peer_device->connection->agreed_pro_version;
732 enum drbd_packet cmd;
734 struct disk_conf *dc;
736 sock = &peer_device->connection->data;
737 p = drbd_prepare_command(peer_device, sock);
742 nc = rcu_dereference(peer_device->connection->net_conf);
744 size = apv <= 87 ? sizeof(struct p_rs_param)
745 : apv == 88 ? sizeof(struct p_rs_param)
746 + strlen(nc->verify_alg) + 1
747 : apv <= 94 ? sizeof(struct p_rs_param_89)
748 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
750 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
752 /* initialize verify_alg and csums_alg */
753 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
755 if (get_ldev(peer_device->device)) {
756 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
757 p->resync_rate = cpu_to_be32(dc->resync_rate);
758 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
759 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
760 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
761 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
762 put_ldev(peer_device->device);
764 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
765 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
766 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
767 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
768 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
772 strcpy(p->verify_alg, nc->verify_alg);
774 strcpy(p->csums_alg, nc->csums_alg);
777 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
780 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
782 struct drbd_socket *sock;
783 struct p_protocol *p;
787 sock = &connection->data;
788 p = __conn_prepare_command(connection, sock);
793 nc = rcu_dereference(connection->net_conf);
795 if (nc->tentative && connection->agreed_pro_version < 92) {
797 mutex_unlock(&sock->mutex);
798 drbd_err(connection, "--dry-run is not supported by peer");
803 if (connection->agreed_pro_version >= 87)
804 size += strlen(nc->integrity_alg) + 1;
806 p->protocol = cpu_to_be32(nc->wire_protocol);
807 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
808 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
809 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
810 p->two_primaries = cpu_to_be32(nc->two_primaries);
812 if (nc->discard_my_data)
813 cf |= CF_DISCARD_MY_DATA;
816 p->conn_flags = cpu_to_be32(cf);
818 if (connection->agreed_pro_version >= 87)
819 strcpy(p->integrity_alg, nc->integrity_alg);
822 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
825 int drbd_send_protocol(struct drbd_connection *connection)
829 mutex_lock(&connection->data.mutex);
830 err = __drbd_send_protocol(connection, P_PROTOCOL);
831 mutex_unlock(&connection->data.mutex);
836 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
838 struct drbd_device *device = peer_device->device;
839 struct drbd_socket *sock;
843 if (!get_ldev_if_state(device, D_NEGOTIATING))
846 sock = &peer_device->connection->data;
847 p = drbd_prepare_command(peer_device, sock);
852 spin_lock_irq(&device->ldev->md.uuid_lock);
853 for (i = UI_CURRENT; i < UI_SIZE; i++)
854 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
855 spin_unlock_irq(&device->ldev->md.uuid_lock);
857 device->comm_bm_set = drbd_bm_total_weight(device);
858 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
860 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
862 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
863 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
864 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
867 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
870 int drbd_send_uuids(struct drbd_peer_device *peer_device)
872 return _drbd_send_uuids(peer_device, 0);
875 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
877 return _drbd_send_uuids(peer_device, 8);
880 void drbd_print_uuids(struct drbd_device *device, const char *text)
882 if (get_ldev_if_state(device, D_NEGOTIATING)) {
883 u64 *uuid = device->ldev->md.uuid;
884 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
886 (unsigned long long)uuid[UI_CURRENT],
887 (unsigned long long)uuid[UI_BITMAP],
888 (unsigned long long)uuid[UI_HISTORY_START],
889 (unsigned long long)uuid[UI_HISTORY_END]);
892 drbd_info(device, "%s effective data uuid: %016llX\n",
894 (unsigned long long)device->ed_uuid);
898 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
900 struct drbd_device *device = peer_device->device;
901 struct drbd_socket *sock;
905 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
907 uuid = device->ldev->md.uuid[UI_BITMAP];
908 if (uuid && uuid != UUID_JUST_CREATED)
909 uuid = uuid + UUID_NEW_BM_OFFSET;
911 get_random_bytes(&uuid, sizeof(u64));
912 drbd_uuid_set(device, UI_BITMAP, uuid);
913 drbd_print_uuids(device, "updated sync UUID");
914 drbd_md_sync(device);
916 sock = &peer_device->connection->data;
917 p = drbd_prepare_command(peer_device, sock);
919 p->uuid = cpu_to_be64(uuid);
920 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
924 /* communicated if (agreed_features & DRBD_FF_WSAME) */
925 void assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p, struct request_queue *q)
928 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
929 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
930 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
931 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
932 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
933 p->qlim->discard_enabled = blk_queue_discard(q);
934 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
936 q = device->rq_queue;
937 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
938 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
939 p->qlim->alignment_offset = 0;
940 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
941 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
942 p->qlim->discard_enabled = 0;
943 p->qlim->write_same_capable = 0;
947 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
949 struct drbd_device *device = peer_device->device;
950 struct drbd_socket *sock;
952 sector_t d_size, u_size;
954 unsigned int max_bio_size;
955 unsigned int packet_size;
957 sock = &peer_device->connection->data;
958 p = drbd_prepare_command(peer_device, sock);
962 packet_size = sizeof(*p);
963 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
964 packet_size += sizeof(p->qlim[0]);
966 memset(p, 0, packet_size);
967 if (get_ldev_if_state(device, D_NEGOTIATING)) {
968 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
969 d_size = drbd_get_max_capacity(device->ldev);
971 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
973 q_order_type = drbd_queue_order_type(device);
974 max_bio_size = queue_max_hw_sectors(q) << 9;
975 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
976 assign_p_sizes_qlim(device, p, q);
981 q_order_type = QUEUE_ORDERED_NONE;
982 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
983 assign_p_sizes_qlim(device, p, NULL);
986 if (peer_device->connection->agreed_pro_version <= 94)
987 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
988 else if (peer_device->connection->agreed_pro_version < 100)
989 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
991 p->d_size = cpu_to_be64(d_size);
992 p->u_size = cpu_to_be64(u_size);
993 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
994 p->max_bio_size = cpu_to_be32(max_bio_size);
995 p->queue_order_type = cpu_to_be16(q_order_type);
996 p->dds_flags = cpu_to_be16(flags);
998 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
1002 * drbd_send_current_state() - Sends the drbd state to the peer
1003 * @peer_device: DRBD peer device.
1005 int drbd_send_current_state(struct drbd_peer_device *peer_device)
1007 struct drbd_socket *sock;
1010 sock = &peer_device->connection->data;
1011 p = drbd_prepare_command(peer_device, sock);
1014 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1015 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1019 * drbd_send_state() - After a state change, sends the new state to the peer
1020 * @peer_device: DRBD peer device.
1021 * @state: the state to send, not necessarily the current state.
1023 * Each state change queues an "after_state_ch" work, which will eventually
1024 * send the resulting new state to the peer. If more state changes happen
1025 * between queuing and processing of the after_state_ch work, we still
1026 * want to send each intermediary state in the order it occurred.
1028 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1030 struct drbd_socket *sock;
1033 sock = &peer_device->connection->data;
1034 p = drbd_prepare_command(peer_device, sock);
1037 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1038 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1041 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1043 struct drbd_socket *sock;
1044 struct p_req_state *p;
1046 sock = &peer_device->connection->data;
1047 p = drbd_prepare_command(peer_device, sock);
1050 p->mask = cpu_to_be32(mask.i);
1051 p->val = cpu_to_be32(val.i);
1052 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1055 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1057 enum drbd_packet cmd;
1058 struct drbd_socket *sock;
1059 struct p_req_state *p;
1061 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1062 sock = &connection->data;
1063 p = conn_prepare_command(connection, sock);
1066 p->mask = cpu_to_be32(mask.i);
1067 p->val = cpu_to_be32(val.i);
1068 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1071 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1073 struct drbd_socket *sock;
1074 struct p_req_state_reply *p;
1076 sock = &peer_device->connection->meta;
1077 p = drbd_prepare_command(peer_device, sock);
1079 p->retcode = cpu_to_be32(retcode);
1080 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1084 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1086 struct drbd_socket *sock;
1087 struct p_req_state_reply *p;
1088 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1090 sock = &connection->meta;
1091 p = conn_prepare_command(connection, sock);
1093 p->retcode = cpu_to_be32(retcode);
1094 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1098 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1100 BUG_ON(code & ~0xf);
1101 p->encoding = (p->encoding & ~0xf) | code;
1104 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1106 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1109 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1112 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1115 static int fill_bitmap_rle_bits(struct drbd_device *device,
1116 struct p_compressed_bm *p,
1118 struct bm_xfer_ctx *c)
1120 struct bitstream bs;
1121 unsigned long plain_bits;
1128 /* may we use this feature? */
1130 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1132 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1135 if (c->bit_offset >= c->bm_bits)
1136 return 0; /* nothing to do. */
1138 /* use at most thus many bytes */
1139 bitstream_init(&bs, p->code, size, 0);
1140 memset(p->code, 0, size);
1141 /* plain bits covered in this code string */
1144 /* p->encoding & 0x80 stores whether the first run length is set.
1145 * bit offset is implicit.
1146 * start with toggle == 2 to be able to tell the first iteration */
1149 /* see how much plain bits we can stuff into one packet
1150 * using RLE and VLI. */
1152 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1153 : _drbd_bm_find_next(device, c->bit_offset);
1156 rl = tmp - c->bit_offset;
1158 if (toggle == 2) { /* first iteration */
1160 /* the first checked bit was set,
1161 * store start value, */
1162 dcbp_set_start(p, 1);
1163 /* but skip encoding of zero run length */
1167 dcbp_set_start(p, 0);
1170 /* paranoia: catch zero runlength.
1171 * can only happen if bitmap is modified while we scan it. */
1173 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1174 "t:%u bo:%lu\n", toggle, c->bit_offset);
1178 bits = vli_encode_bits(&bs, rl);
1179 if (bits == -ENOBUFS) /* buffer full */
1182 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1188 c->bit_offset = tmp;
1189 } while (c->bit_offset < c->bm_bits);
1191 len = bs.cur.b - p->code + !!bs.cur.bit;
1193 if (plain_bits < (len << 3)) {
1194 /* incompressible with this method.
1195 * we need to rewind both word and bit position. */
1196 c->bit_offset -= plain_bits;
1197 bm_xfer_ctx_bit_to_word_offset(c);
1198 c->bit_offset = c->word_offset * BITS_PER_LONG;
1202 /* RLE + VLI was able to compress it just fine.
1203 * update c->word_offset. */
1204 bm_xfer_ctx_bit_to_word_offset(c);
1206 /* store pad_bits */
1207 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1213 * send_bitmap_rle_or_plain
1215 * Return 0 when done, 1 when another iteration is needed, and a negative error
1216 * code upon failure.
1219 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1221 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1222 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1223 struct p_compressed_bm *p = sock->sbuf + header_size;
1226 len = fill_bitmap_rle_bits(device, p,
1227 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1232 dcbp_set_code(p, RLE_VLI_Bits);
1233 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1234 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1237 c->bytes[0] += header_size + sizeof(*p) + len;
1239 if (c->bit_offset >= c->bm_bits)
1242 /* was not compressible.
1243 * send a buffer full of plain text bits instead. */
1244 unsigned int data_size;
1245 unsigned long num_words;
1246 unsigned long *p = sock->sbuf + header_size;
1248 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1249 num_words = min_t(size_t, data_size / sizeof(*p),
1250 c->bm_words - c->word_offset);
1251 len = num_words * sizeof(*p);
1253 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1254 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1255 c->word_offset += num_words;
1256 c->bit_offset = c->word_offset * BITS_PER_LONG;
1259 c->bytes[1] += header_size + len;
1261 if (c->bit_offset > c->bm_bits)
1262 c->bit_offset = c->bm_bits;
1266 INFO_bm_xfer_stats(device, "send", c);
1274 /* See the comment at receive_bitmap() */
1275 static int _drbd_send_bitmap(struct drbd_device *device)
1277 struct bm_xfer_ctx c;
1280 if (!expect(device->bitmap))
1283 if (get_ldev(device)) {
1284 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1285 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1286 drbd_bm_set_all(device);
1287 if (drbd_bm_write(device)) {
1288 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1289 * but otherwise process as per normal - need to tell other
1290 * side that a full resync is required! */
1291 drbd_err(device, "Failed to write bitmap to disk!\n");
1293 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1294 drbd_md_sync(device);
1300 c = (struct bm_xfer_ctx) {
1301 .bm_bits = drbd_bm_bits(device),
1302 .bm_words = drbd_bm_words(device),
1306 err = send_bitmap_rle_or_plain(device, &c);
1312 int drbd_send_bitmap(struct drbd_device *device)
1314 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1317 mutex_lock(&sock->mutex);
1319 err = !_drbd_send_bitmap(device);
1320 mutex_unlock(&sock->mutex);
1324 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1326 struct drbd_socket *sock;
1327 struct p_barrier_ack *p;
1329 if (connection->cstate < C_WF_REPORT_PARAMS)
1332 sock = &connection->meta;
1333 p = conn_prepare_command(connection, sock);
1336 p->barrier = barrier_nr;
1337 p->set_size = cpu_to_be32(set_size);
1338 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1342 * _drbd_send_ack() - Sends an ack packet
1343 * @device: DRBD device.
1344 * @cmd: Packet command code.
1345 * @sector: sector, needs to be in big endian byte order
1346 * @blksize: size in byte, needs to be in big endian byte order
1347 * @block_id: Id, big endian byte order
1349 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1350 u64 sector, u32 blksize, u64 block_id)
1352 struct drbd_socket *sock;
1353 struct p_block_ack *p;
1355 if (peer_device->device->state.conn < C_CONNECTED)
1358 sock = &peer_device->connection->meta;
1359 p = drbd_prepare_command(peer_device, sock);
1363 p->block_id = block_id;
1364 p->blksize = blksize;
1365 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1366 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1369 /* dp->sector and dp->block_id already/still in network byte order,
1370 * data_size is payload size according to dp->head,
1371 * and may need to be corrected for digest size. */
1372 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1373 struct p_data *dp, int data_size)
1375 if (peer_device->connection->peer_integrity_tfm)
1376 data_size -= crypto_ahash_digestsize(peer_device->connection->peer_integrity_tfm);
1377 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1381 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1382 struct p_block_req *rp)
1384 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1388 * drbd_send_ack() - Sends an ack packet
1389 * @device: DRBD device
1390 * @cmd: packet command code
1391 * @peer_req: peer request
1393 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1394 struct drbd_peer_request *peer_req)
1396 return _drbd_send_ack(peer_device, cmd,
1397 cpu_to_be64(peer_req->i.sector),
1398 cpu_to_be32(peer_req->i.size),
1399 peer_req->block_id);
1402 /* This function misuses the block_id field to signal if the blocks
1403 * are is sync or not. */
1404 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1405 sector_t sector, int blksize, u64 block_id)
1407 return _drbd_send_ack(peer_device, cmd,
1408 cpu_to_be64(sector),
1409 cpu_to_be32(blksize),
1410 cpu_to_be64(block_id));
1413 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1414 struct drbd_peer_request *peer_req)
1416 struct drbd_socket *sock;
1417 struct p_block_desc *p;
1419 sock = &peer_device->connection->data;
1420 p = drbd_prepare_command(peer_device, sock);
1423 p->sector = cpu_to_be64(peer_req->i.sector);
1424 p->blksize = cpu_to_be32(peer_req->i.size);
1426 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1429 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1430 sector_t sector, int size, u64 block_id)
1432 struct drbd_socket *sock;
1433 struct p_block_req *p;
1435 sock = &peer_device->connection->data;
1436 p = drbd_prepare_command(peer_device, sock);
1439 p->sector = cpu_to_be64(sector);
1440 p->block_id = block_id;
1441 p->blksize = cpu_to_be32(size);
1442 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1445 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1446 void *digest, int digest_size, enum drbd_packet cmd)
1448 struct drbd_socket *sock;
1449 struct p_block_req *p;
1451 /* FIXME: Put the digest into the preallocated socket buffer. */
1453 sock = &peer_device->connection->data;
1454 p = drbd_prepare_command(peer_device, sock);
1457 p->sector = cpu_to_be64(sector);
1458 p->block_id = ID_SYNCER /* unused */;
1459 p->blksize = cpu_to_be32(size);
1460 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1463 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1465 struct drbd_socket *sock;
1466 struct p_block_req *p;
1468 sock = &peer_device->connection->data;
1469 p = drbd_prepare_command(peer_device, sock);
1472 p->sector = cpu_to_be64(sector);
1473 p->block_id = ID_SYNCER /* unused */;
1474 p->blksize = cpu_to_be32(size);
1475 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1478 /* called on sndtimeo
1479 * returns false if we should retry,
1480 * true if we think connection is dead
1482 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1485 /* long elapsed = (long)(jiffies - device->last_received); */
1487 drop_it = connection->meta.socket == sock
1488 || !connection->ack_receiver.task
1489 || get_t_state(&connection->ack_receiver) != RUNNING
1490 || connection->cstate < C_WF_REPORT_PARAMS;
1495 drop_it = !--connection->ko_count;
1497 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1498 current->comm, current->pid, connection->ko_count);
1499 request_ping(connection);
1502 return drop_it; /* && (device->state == R_PRIMARY) */;
1505 static void drbd_update_congested(struct drbd_connection *connection)
1507 struct sock *sk = connection->data.socket->sk;
1508 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1509 set_bit(NET_CONGESTED, &connection->flags);
1512 /* The idea of sendpage seems to be to put some kind of reference
1513 * to the page into the skb, and to hand it over to the NIC. In
1514 * this process get_page() gets called.
1516 * As soon as the page was really sent over the network put_page()
1517 * gets called by some part of the network layer. [ NIC driver? ]
1519 * [ get_page() / put_page() increment/decrement the count. If count
1520 * reaches 0 the page will be freed. ]
1522 * This works nicely with pages from FSs.
1523 * But this means that in protocol A we might signal IO completion too early!
1525 * In order not to corrupt data during a resync we must make sure
1526 * that we do not reuse our own buffer pages (EEs) to early, therefore
1527 * we have the net_ee list.
1529 * XFS seems to have problems, still, it submits pages with page_count == 0!
1530 * As a workaround, we disable sendpage on pages
1531 * with page_count == 0 or PageSlab.
1533 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1534 int offset, size_t size, unsigned msg_flags)
1536 struct socket *socket;
1540 socket = peer_device->connection->data.socket;
1541 addr = kmap(page) + offset;
1542 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1545 peer_device->device->send_cnt += size >> 9;
1549 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1550 int offset, size_t size, unsigned msg_flags)
1552 struct socket *socket = peer_device->connection->data.socket;
1553 mm_segment_t oldfs = get_fs();
1557 /* e.g. XFS meta- & log-data is in slab pages, which have a
1558 * page_count of 0 and/or have PageSlab() set.
1559 * we cannot use send_page for those, as that does get_page();
1560 * put_page(); and would cause either a VM_BUG directly, or
1561 * __page_cache_release a page that would actually still be referenced
1562 * by someone, leading to some obscure delayed Oops somewhere else. */
1563 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1564 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1566 msg_flags |= MSG_NOSIGNAL;
1567 drbd_update_congested(peer_device->connection);
1572 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1574 if (sent == -EAGAIN) {
1575 if (we_should_drop_the_connection(peer_device->connection, socket))
1579 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1580 __func__, (int)size, len, sent);
1587 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1589 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1593 peer_device->device->send_cnt += size >> 9;
1598 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1600 struct bio_vec bvec;
1601 struct bvec_iter iter;
1603 /* hint all but last page with MSG_MORE */
1604 bio_for_each_segment(bvec, bio, iter) {
1607 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1608 bvec.bv_offset, bvec.bv_len,
1609 bio_iter_last(bvec, iter)
1613 /* REQ_OP_WRITE_SAME has only one segment */
1614 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1620 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1622 struct bio_vec bvec;
1623 struct bvec_iter iter;
1625 /* hint all but last page with MSG_MORE */
1626 bio_for_each_segment(bvec, bio, iter) {
1629 err = _drbd_send_page(peer_device, bvec.bv_page,
1630 bvec.bv_offset, bvec.bv_len,
1631 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1634 /* REQ_OP_WRITE_SAME has only one segment */
1635 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1641 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1642 struct drbd_peer_request *peer_req)
1644 struct page *page = peer_req->pages;
1645 unsigned len = peer_req->i.size;
1648 /* hint all but last page with MSG_MORE */
1649 page_chain_for_each(page) {
1650 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1652 err = _drbd_send_page(peer_device, page, 0, l,
1653 page_chain_next(page) ? MSG_MORE : 0);
1661 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1664 if (connection->agreed_pro_version >= 95)
1665 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1666 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1667 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1668 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1669 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1670 (bio_op(bio) == REQ_OP_WRITE_ZEROES ? DP_DISCARD : 0);
1672 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1675 /* Used to send write or TRIM aka REQ_DISCARD requests
1676 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1678 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1680 struct drbd_device *device = peer_device->device;
1681 struct drbd_socket *sock;
1683 struct p_wsame *wsame = NULL;
1685 unsigned int dp_flags = 0;
1689 sock = &peer_device->connection->data;
1690 p = drbd_prepare_command(peer_device, sock);
1691 digest_size = peer_device->connection->integrity_tfm ?
1692 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1696 p->sector = cpu_to_be64(req->i.sector);
1697 p->block_id = (unsigned long)req;
1698 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1699 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1700 if (device->state.conn >= C_SYNC_SOURCE &&
1701 device->state.conn <= C_PAUSED_SYNC_T)
1702 dp_flags |= DP_MAY_SET_IN_SYNC;
1703 if (peer_device->connection->agreed_pro_version >= 100) {
1704 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1705 dp_flags |= DP_SEND_RECEIVE_ACK;
1706 /* During resync, request an explicit write ack,
1707 * even in protocol != C */
1708 if (req->rq_state & RQ_EXP_WRITE_ACK
1709 || (dp_flags & DP_MAY_SET_IN_SYNC))
1710 dp_flags |= DP_SEND_WRITE_ACK;
1712 p->dp_flags = cpu_to_be32(dp_flags);
1714 if (dp_flags & DP_DISCARD) {
1715 struct p_trim *t = (struct p_trim*)p;
1716 t->size = cpu_to_be32(req->i.size);
1717 err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1720 if (dp_flags & DP_WSAME) {
1721 /* this will only work if DRBD_FF_WSAME is set AND the
1722 * handshake agreed that all nodes and backend devices are
1723 * WRITE_SAME capable and agree on logical_block_size */
1724 wsame = (struct p_wsame*)p;
1725 digest_out = wsame + 1;
1726 wsame->size = cpu_to_be32(req->i.size);
1730 /* our digest is still only over the payload.
1731 * TRIM does not carry any payload. */
1733 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1736 __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1737 sizeof(*wsame) + digest_size, NULL,
1738 bio_iovec(req->master_bio).bv_len);
1741 __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1742 sizeof(*p) + digest_size, NULL, req->i.size);
1744 /* For protocol A, we have to memcpy the payload into
1745 * socket buffers, as we may complete right away
1746 * as soon as we handed it over to tcp, at which point the data
1747 * pages may become invalid.
1749 * For data-integrity enabled, we copy it as well, so we can be
1750 * sure that even if the bio pages may still be modified, it
1751 * won't change the data on the wire, thus if the digest checks
1752 * out ok after sending on this side, but does not fit on the
1753 * receiving side, we sure have detected corruption elsewhere.
1755 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1756 err = _drbd_send_bio(peer_device, req->master_bio);
1758 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1760 /* double check digest, sometimes buffers have been modified in flight. */
1761 if (digest_size > 0 && digest_size <= 64) {
1762 /* 64 byte, 512 bit, is the largest digest size
1763 * currently supported in kernel crypto. */
1764 unsigned char digest[64];
1765 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1766 if (memcmp(p + 1, digest, digest_size)) {
1768 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1769 (unsigned long long)req->i.sector, req->i.size);
1771 } /* else if (digest_size > 64) {
1772 ... Be noisy about digest too large ...
1776 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1781 /* answer packet, used to send data back for read requests:
1782 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1783 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1785 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1786 struct drbd_peer_request *peer_req)
1788 struct drbd_device *device = peer_device->device;
1789 struct drbd_socket *sock;
1794 sock = &peer_device->connection->data;
1795 p = drbd_prepare_command(peer_device, sock);
1797 digest_size = peer_device->connection->integrity_tfm ?
1798 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1802 p->sector = cpu_to_be64(peer_req->i.sector);
1803 p->block_id = peer_req->block_id;
1804 p->seq_num = 0; /* unused */
1807 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1808 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1810 err = _drbd_send_zc_ee(peer_device, peer_req);
1811 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1816 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1818 struct drbd_socket *sock;
1819 struct p_block_desc *p;
1821 sock = &peer_device->connection->data;
1822 p = drbd_prepare_command(peer_device, sock);
1825 p->sector = cpu_to_be64(req->i.sector);
1826 p->blksize = cpu_to_be32(req->i.size);
1827 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1831 drbd_send distinguishes two cases:
1833 Packets sent via the data socket "sock"
1834 and packets sent via the meta data socket "msock"
1837 -----------------+-------------------------+------------------------------
1838 timeout conf.timeout / 2 conf.timeout / 2
1839 timeout action send a ping via msock Abort communication
1840 and close all sockets
1844 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1846 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1847 void *buf, size_t size, unsigned msg_flags)
1849 struct kvec iov = {.iov_base = buf, .iov_len = size};
1856 /* THINK if (signal_pending) return ... ? */
1858 msg.msg_name = NULL;
1859 msg.msg_namelen = 0;
1860 msg.msg_control = NULL;
1861 msg.msg_controllen = 0;
1862 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1864 iov_iter_kvec(&msg.msg_iter, WRITE | ITER_KVEC, &iov, 1, size);
1866 if (sock == connection->data.socket) {
1868 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1870 drbd_update_congested(connection);
1873 rv = sock_sendmsg(sock, &msg);
1874 if (rv == -EAGAIN) {
1875 if (we_should_drop_the_connection(connection, sock))
1881 flush_signals(current);
1887 } while (sent < size);
1889 if (sock == connection->data.socket)
1890 clear_bit(NET_CONGESTED, &connection->flags);
1893 if (rv != -EAGAIN) {
1894 drbd_err(connection, "%s_sendmsg returned %d\n",
1895 sock == connection->meta.socket ? "msock" : "sock",
1897 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1899 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1906 * drbd_send_all - Send an entire buffer
1908 * Returns 0 upon success and a negative error value otherwise.
1910 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1911 size_t size, unsigned msg_flags)
1915 err = drbd_send(connection, sock, buffer, size, msg_flags);
1923 static int drbd_open(struct block_device *bdev, fmode_t mode)
1925 struct drbd_device *device = bdev->bd_disk->private_data;
1926 unsigned long flags;
1929 mutex_lock(&drbd_main_mutex);
1930 spin_lock_irqsave(&device->resource->req_lock, flags);
1931 /* to have a stable device->state.role
1932 * and no race with updating open_cnt */
1934 if (device->state.role != R_PRIMARY) {
1935 if (mode & FMODE_WRITE)
1937 else if (!allow_oos)
1943 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1944 mutex_unlock(&drbd_main_mutex);
1949 static void drbd_release(struct gendisk *gd, fmode_t mode)
1951 struct drbd_device *device = gd->private_data;
1952 mutex_lock(&drbd_main_mutex);
1954 mutex_unlock(&drbd_main_mutex);
1957 static void drbd_set_defaults(struct drbd_device *device)
1959 /* Beware! The actual layout differs
1960 * between big endian and little endian */
1961 device->state = (union drbd_dev_state) {
1962 { .role = R_SECONDARY,
1964 .conn = C_STANDALONE,
1970 void drbd_init_set_defaults(struct drbd_device *device)
1972 /* the memset(,0,) did most of this.
1973 * note: only assignments, no allocation in here */
1975 drbd_set_defaults(device);
1977 atomic_set(&device->ap_bio_cnt, 0);
1978 atomic_set(&device->ap_actlog_cnt, 0);
1979 atomic_set(&device->ap_pending_cnt, 0);
1980 atomic_set(&device->rs_pending_cnt, 0);
1981 atomic_set(&device->unacked_cnt, 0);
1982 atomic_set(&device->local_cnt, 0);
1983 atomic_set(&device->pp_in_use_by_net, 0);
1984 atomic_set(&device->rs_sect_in, 0);
1985 atomic_set(&device->rs_sect_ev, 0);
1986 atomic_set(&device->ap_in_flight, 0);
1987 atomic_set(&device->md_io.in_use, 0);
1989 mutex_init(&device->own_state_mutex);
1990 device->state_mutex = &device->own_state_mutex;
1992 spin_lock_init(&device->al_lock);
1993 spin_lock_init(&device->peer_seq_lock);
1995 INIT_LIST_HEAD(&device->active_ee);
1996 INIT_LIST_HEAD(&device->sync_ee);
1997 INIT_LIST_HEAD(&device->done_ee);
1998 INIT_LIST_HEAD(&device->read_ee);
1999 INIT_LIST_HEAD(&device->net_ee);
2000 INIT_LIST_HEAD(&device->resync_reads);
2001 INIT_LIST_HEAD(&device->resync_work.list);
2002 INIT_LIST_HEAD(&device->unplug_work.list);
2003 INIT_LIST_HEAD(&device->bm_io_work.w.list);
2004 INIT_LIST_HEAD(&device->pending_master_completion[0]);
2005 INIT_LIST_HEAD(&device->pending_master_completion[1]);
2006 INIT_LIST_HEAD(&device->pending_completion[0]);
2007 INIT_LIST_HEAD(&device->pending_completion[1]);
2009 device->resync_work.cb = w_resync_timer;
2010 device->unplug_work.cb = w_send_write_hint;
2011 device->bm_io_work.w.cb = w_bitmap_io;
2013 init_timer(&device->resync_timer);
2014 init_timer(&device->md_sync_timer);
2015 init_timer(&device->start_resync_timer);
2016 init_timer(&device->request_timer);
2017 device->resync_timer.function = resync_timer_fn;
2018 device->resync_timer.data = (unsigned long) device;
2019 device->md_sync_timer.function = md_sync_timer_fn;
2020 device->md_sync_timer.data = (unsigned long) device;
2021 device->start_resync_timer.function = start_resync_timer_fn;
2022 device->start_resync_timer.data = (unsigned long) device;
2023 device->request_timer.function = request_timer_fn;
2024 device->request_timer.data = (unsigned long) device;
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 if (drbd_md_io_bio_set)
2102 bioset_free(drbd_md_io_bio_set);
2103 if (drbd_md_io_page_pool)
2104 mempool_destroy(drbd_md_io_page_pool);
2105 if (drbd_ee_mempool)
2106 mempool_destroy(drbd_ee_mempool);
2107 if (drbd_request_mempool)
2108 mempool_destroy(drbd_request_mempool);
2110 kmem_cache_destroy(drbd_ee_cache);
2111 if (drbd_request_cache)
2112 kmem_cache_destroy(drbd_request_cache);
2113 if (drbd_bm_ext_cache)
2114 kmem_cache_destroy(drbd_bm_ext_cache);
2115 if (drbd_al_ext_cache)
2116 kmem_cache_destroy(drbd_al_ext_cache);
2118 drbd_md_io_bio_set = NULL;
2119 drbd_md_io_page_pool = NULL;
2120 drbd_ee_mempool = NULL;
2121 drbd_request_mempool = NULL;
2122 drbd_ee_cache = NULL;
2123 drbd_request_cache = NULL;
2124 drbd_bm_ext_cache = NULL;
2125 drbd_al_ext_cache = NULL;
2130 static int drbd_create_mempools(void)
2133 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
2136 /* prepare our caches and mempools */
2137 drbd_request_mempool = NULL;
2138 drbd_ee_cache = NULL;
2139 drbd_request_cache = NULL;
2140 drbd_bm_ext_cache = NULL;
2141 drbd_al_ext_cache = NULL;
2142 drbd_pp_pool = NULL;
2143 drbd_md_io_page_pool = NULL;
2144 drbd_md_io_bio_set = NULL;
2147 drbd_request_cache = kmem_cache_create(
2148 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2149 if (drbd_request_cache == NULL)
2152 drbd_ee_cache = kmem_cache_create(
2153 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2154 if (drbd_ee_cache == NULL)
2157 drbd_bm_ext_cache = kmem_cache_create(
2158 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2159 if (drbd_bm_ext_cache == NULL)
2162 drbd_al_ext_cache = kmem_cache_create(
2163 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2164 if (drbd_al_ext_cache == NULL)
2168 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2169 if (drbd_md_io_bio_set == NULL)
2172 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2173 if (drbd_md_io_page_pool == NULL)
2176 drbd_request_mempool = mempool_create_slab_pool(number,
2177 drbd_request_cache);
2178 if (drbd_request_mempool == NULL)
2181 drbd_ee_mempool = mempool_create_slab_pool(number, drbd_ee_cache);
2182 if (drbd_ee_mempool == NULL)
2185 /* drbd's page pool */
2186 spin_lock_init(&drbd_pp_lock);
2188 for (i = 0; i < number; i++) {
2189 page = alloc_page(GFP_HIGHUSER);
2192 set_page_private(page, (unsigned long)drbd_pp_pool);
2193 drbd_pp_pool = page;
2195 drbd_pp_vacant = number;
2200 drbd_destroy_mempools(); /* in case we allocated some */
2204 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2208 rr = drbd_free_peer_reqs(device, &device->active_ee);
2210 drbd_err(device, "%d EEs in active list found!\n", rr);
2212 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2214 drbd_err(device, "%d EEs in sync list found!\n", rr);
2216 rr = drbd_free_peer_reqs(device, &device->read_ee);
2218 drbd_err(device, "%d EEs in read list found!\n", rr);
2220 rr = drbd_free_peer_reqs(device, &device->done_ee);
2222 drbd_err(device, "%d EEs in done list found!\n", rr);
2224 rr = drbd_free_peer_reqs(device, &device->net_ee);
2226 drbd_err(device, "%d EEs in net list found!\n", rr);
2229 /* caution. no locking. */
2230 void drbd_destroy_device(struct kref *kref)
2232 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2233 struct drbd_resource *resource = device->resource;
2234 struct drbd_peer_device *peer_device, *tmp_peer_device;
2236 del_timer_sync(&device->request_timer);
2238 /* paranoia asserts */
2239 D_ASSERT(device, device->open_cnt == 0);
2240 /* end paranoia asserts */
2242 /* cleanup stuff that may have been allocated during
2243 * device (re-)configuration or state changes */
2245 if (device->this_bdev)
2246 bdput(device->this_bdev);
2248 drbd_backing_dev_free(device, device->ldev);
2249 device->ldev = NULL;
2251 drbd_release_all_peer_reqs(device);
2253 lc_destroy(device->act_log);
2254 lc_destroy(device->resync);
2256 kfree(device->p_uuid);
2257 /* device->p_uuid = NULL; */
2259 if (device->bitmap) /* should no longer be there. */
2260 drbd_bm_cleanup(device);
2261 __free_page(device->md_io.page);
2262 put_disk(device->vdisk);
2263 blk_cleanup_queue(device->rq_queue);
2264 kfree(device->rs_plan_s);
2266 /* not for_each_connection(connection, resource):
2267 * those may have been cleaned up and disassociated already.
2269 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2270 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2273 memset(device, 0xfd, sizeof(*device));
2275 kref_put(&resource->kref, drbd_destroy_resource);
2278 /* One global retry thread, if we need to push back some bio and have it
2279 * reinserted through our make request function.
2281 static struct retry_worker {
2282 struct workqueue_struct *wq;
2283 struct work_struct worker;
2286 struct list_head writes;
2289 static void do_retry(struct work_struct *ws)
2291 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2293 struct drbd_request *req, *tmp;
2295 spin_lock_irq(&retry->lock);
2296 list_splice_init(&retry->writes, &writes);
2297 spin_unlock_irq(&retry->lock);
2299 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2300 struct drbd_device *device = req->device;
2301 struct bio *bio = req->master_bio;
2302 unsigned long start_jif = req->start_jif;
2306 expect(atomic_read(&req->completion_ref) == 0) &&
2307 expect(req->rq_state & RQ_POSTPONED) &&
2308 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2309 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2312 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2313 req, atomic_read(&req->completion_ref),
2316 /* We still need to put one kref associated with the
2317 * "completion_ref" going zero in the code path that queued it
2318 * here. The request object may still be referenced by a
2319 * frozen local req->private_bio, in case we force-detached.
2321 kref_put(&req->kref, drbd_req_destroy);
2323 /* A single suspended or otherwise blocking device may stall
2324 * all others as well. Fortunately, this code path is to
2325 * recover from a situation that "should not happen":
2326 * concurrent writes in multi-primary setup.
2327 * In a "normal" lifecycle, this workqueue is supposed to be
2328 * destroyed without ever doing anything.
2329 * If it turns out to be an issue anyways, we can do per
2330 * resource (replication group) or per device (minor) retry
2331 * workqueues instead.
2334 /* We are not just doing generic_make_request(),
2335 * as we want to keep the start_time information. */
2337 __drbd_make_request(device, bio, start_jif);
2341 /* called via drbd_req_put_completion_ref(),
2342 * holds resource->req_lock */
2343 void drbd_restart_request(struct drbd_request *req)
2345 unsigned long flags;
2346 spin_lock_irqsave(&retry.lock, flags);
2347 list_move_tail(&req->tl_requests, &retry.writes);
2348 spin_unlock_irqrestore(&retry.lock, flags);
2350 /* Drop the extra reference that would otherwise
2351 * have been dropped by complete_master_bio.
2352 * do_retry() needs to grab a new one. */
2353 dec_ap_bio(req->device);
2355 queue_work(retry.wq, &retry.worker);
2358 void drbd_destroy_resource(struct kref *kref)
2360 struct drbd_resource *resource =
2361 container_of(kref, struct drbd_resource, kref);
2363 idr_destroy(&resource->devices);
2364 free_cpumask_var(resource->cpu_mask);
2365 kfree(resource->name);
2366 memset(resource, 0xf2, sizeof(*resource));
2370 void drbd_free_resource(struct drbd_resource *resource)
2372 struct drbd_connection *connection, *tmp;
2374 for_each_connection_safe(connection, tmp, resource) {
2375 list_del(&connection->connections);
2376 drbd_debugfs_connection_cleanup(connection);
2377 kref_put(&connection->kref, drbd_destroy_connection);
2379 drbd_debugfs_resource_cleanup(resource);
2380 kref_put(&resource->kref, drbd_destroy_resource);
2383 static void drbd_cleanup(void)
2386 struct drbd_device *device;
2387 struct drbd_resource *resource, *tmp;
2389 /* first remove proc,
2390 * drbdsetup uses it's presence to detect
2391 * whether DRBD is loaded.
2392 * If we would get stuck in proc removal,
2393 * but have netlink already deregistered,
2394 * some drbdsetup commands may wait forever
2398 remove_proc_entry("drbd", NULL);
2401 destroy_workqueue(retry.wq);
2403 drbd_genl_unregister();
2404 drbd_debugfs_cleanup();
2406 idr_for_each_entry(&drbd_devices, device, i)
2407 drbd_delete_device(device);
2409 /* not _rcu since, no other updater anymore. Genl already unregistered */
2410 for_each_resource_safe(resource, tmp, &drbd_resources) {
2411 list_del(&resource->resources);
2412 drbd_free_resource(resource);
2415 drbd_destroy_mempools();
2416 unregister_blkdev(DRBD_MAJOR, "drbd");
2418 idr_destroy(&drbd_devices);
2420 pr_info("module cleanup done.\n");
2424 * drbd_congested() - Callback for the flusher thread
2425 * @congested_data: User data
2426 * @bdi_bits: Bits the BDI flusher thread is currently interested in
2428 * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested.
2430 static int drbd_congested(void *congested_data, int bdi_bits)
2432 struct drbd_device *device = congested_data;
2433 struct request_queue *q;
2437 if (!may_inc_ap_bio(device)) {
2438 /* DRBD has frozen IO */
2444 if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
2445 r |= (1 << WB_async_congested);
2446 /* Without good local data, we would need to read from remote,
2447 * and that would need the worker thread as well, which is
2448 * currently blocked waiting for that usermode helper to
2451 if (!get_ldev_if_state(device, D_UP_TO_DATE))
2452 r |= (1 << WB_sync_congested);
2460 if (get_ldev(device)) {
2461 q = bdev_get_queue(device->ldev->backing_bdev);
2462 r = bdi_congested(q->backing_dev_info, bdi_bits);
2468 if (bdi_bits & (1 << WB_async_congested) &&
2469 test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
2470 r |= (1 << WB_async_congested);
2471 reason = reason == 'b' ? 'a' : 'n';
2475 device->congestion_reason = reason;
2479 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2481 spin_lock_init(&wq->q_lock);
2482 INIT_LIST_HEAD(&wq->q);
2483 init_waitqueue_head(&wq->q_wait);
2486 struct completion_work {
2488 struct completion done;
2491 static int w_complete(struct drbd_work *w, int cancel)
2493 struct completion_work *completion_work =
2494 container_of(w, struct completion_work, w);
2496 complete(&completion_work->done);
2500 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2502 struct completion_work completion_work;
2504 completion_work.w.cb = w_complete;
2505 init_completion(&completion_work.done);
2506 drbd_queue_work(work_queue, &completion_work.w);
2507 wait_for_completion(&completion_work.done);
2510 struct drbd_resource *drbd_find_resource(const char *name)
2512 struct drbd_resource *resource;
2514 if (!name || !name[0])
2518 for_each_resource_rcu(resource, &drbd_resources) {
2519 if (!strcmp(resource->name, name)) {
2520 kref_get(&resource->kref);
2530 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2531 void *peer_addr, int peer_addr_len)
2533 struct drbd_resource *resource;
2534 struct drbd_connection *connection;
2537 for_each_resource_rcu(resource, &drbd_resources) {
2538 for_each_connection_rcu(connection, resource) {
2539 if (connection->my_addr_len == my_addr_len &&
2540 connection->peer_addr_len == peer_addr_len &&
2541 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2542 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2543 kref_get(&connection->kref);
2554 static int drbd_alloc_socket(struct drbd_socket *socket)
2556 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2559 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2565 static void drbd_free_socket(struct drbd_socket *socket)
2567 free_page((unsigned long) socket->sbuf);
2568 free_page((unsigned long) socket->rbuf);
2571 void conn_free_crypto(struct drbd_connection *connection)
2573 drbd_free_sock(connection);
2575 crypto_free_ahash(connection->csums_tfm);
2576 crypto_free_ahash(connection->verify_tfm);
2577 crypto_free_shash(connection->cram_hmac_tfm);
2578 crypto_free_ahash(connection->integrity_tfm);
2579 crypto_free_ahash(connection->peer_integrity_tfm);
2580 kfree(connection->int_dig_in);
2581 kfree(connection->int_dig_vv);
2583 connection->csums_tfm = NULL;
2584 connection->verify_tfm = NULL;
2585 connection->cram_hmac_tfm = NULL;
2586 connection->integrity_tfm = NULL;
2587 connection->peer_integrity_tfm = NULL;
2588 connection->int_dig_in = NULL;
2589 connection->int_dig_vv = NULL;
2592 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2594 struct drbd_connection *connection;
2595 cpumask_var_t new_cpu_mask;
2598 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2601 /* silently ignore cpu mask on UP kernel */
2602 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2603 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2604 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2605 if (err == -EOVERFLOW) {
2606 /* So what. mask it out. */
2607 cpumask_var_t tmp_cpu_mask;
2608 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2609 cpumask_setall(tmp_cpu_mask);
2610 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2611 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2613 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2615 free_cpumask_var(tmp_cpu_mask);
2620 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2621 /* retcode = ERR_CPU_MASK_PARSE; */
2625 resource->res_opts = *res_opts;
2626 if (cpumask_empty(new_cpu_mask))
2627 drbd_calc_cpu_mask(&new_cpu_mask);
2628 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2629 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2630 for_each_connection_rcu(connection, resource) {
2631 connection->receiver.reset_cpu_mask = 1;
2632 connection->ack_receiver.reset_cpu_mask = 1;
2633 connection->worker.reset_cpu_mask = 1;
2639 free_cpumask_var(new_cpu_mask);
2644 struct drbd_resource *drbd_create_resource(const char *name)
2646 struct drbd_resource *resource;
2648 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2651 resource->name = kstrdup(name, GFP_KERNEL);
2652 if (!resource->name)
2653 goto fail_free_resource;
2654 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2655 goto fail_free_name;
2656 kref_init(&resource->kref);
2657 idr_init(&resource->devices);
2658 INIT_LIST_HEAD(&resource->connections);
2659 resource->write_ordering = WO_BDEV_FLUSH;
2660 list_add_tail_rcu(&resource->resources, &drbd_resources);
2661 mutex_init(&resource->conf_update);
2662 mutex_init(&resource->adm_mutex);
2663 spin_lock_init(&resource->req_lock);
2664 drbd_debugfs_resource_add(resource);
2668 kfree(resource->name);
2675 /* caller must be under adm_mutex */
2676 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2678 struct drbd_resource *resource;
2679 struct drbd_connection *connection;
2681 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2685 if (drbd_alloc_socket(&connection->data))
2687 if (drbd_alloc_socket(&connection->meta))
2690 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2691 if (!connection->current_epoch)
2694 INIT_LIST_HEAD(&connection->transfer_log);
2696 INIT_LIST_HEAD(&connection->current_epoch->list);
2697 connection->epochs = 1;
2698 spin_lock_init(&connection->epoch_lock);
2700 connection->send.seen_any_write_yet = false;
2701 connection->send.current_epoch_nr = 0;
2702 connection->send.current_epoch_writes = 0;
2704 resource = drbd_create_resource(name);
2708 connection->cstate = C_STANDALONE;
2709 mutex_init(&connection->cstate_mutex);
2710 init_waitqueue_head(&connection->ping_wait);
2711 idr_init(&connection->peer_devices);
2713 drbd_init_workqueue(&connection->sender_work);
2714 mutex_init(&connection->data.mutex);
2715 mutex_init(&connection->meta.mutex);
2717 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2718 connection->receiver.connection = connection;
2719 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2720 connection->worker.connection = connection;
2721 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2722 connection->ack_receiver.connection = connection;
2724 kref_init(&connection->kref);
2726 connection->resource = resource;
2728 if (set_resource_options(resource, res_opts))
2731 kref_get(&resource->kref);
2732 list_add_tail_rcu(&connection->connections, &resource->connections);
2733 drbd_debugfs_connection_add(connection);
2737 list_del(&resource->resources);
2738 drbd_free_resource(resource);
2740 kfree(connection->current_epoch);
2741 drbd_free_socket(&connection->meta);
2742 drbd_free_socket(&connection->data);
2747 void drbd_destroy_connection(struct kref *kref)
2749 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2750 struct drbd_resource *resource = connection->resource;
2752 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
2753 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2754 kfree(connection->current_epoch);
2756 idr_destroy(&connection->peer_devices);
2758 drbd_free_socket(&connection->meta);
2759 drbd_free_socket(&connection->data);
2760 kfree(connection->int_dig_in);
2761 kfree(connection->int_dig_vv);
2762 memset(connection, 0xfc, sizeof(*connection));
2764 kref_put(&resource->kref, drbd_destroy_resource);
2767 static int init_submitter(struct drbd_device *device)
2769 /* opencoded create_singlethread_workqueue(),
2770 * to be able to say "drbd%d", ..., minor */
2772 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2773 if (!device->submit.wq)
2776 INIT_WORK(&device->submit.worker, do_submit);
2777 INIT_LIST_HEAD(&device->submit.writes);
2781 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2783 struct drbd_resource *resource = adm_ctx->resource;
2784 struct drbd_connection *connection;
2785 struct drbd_device *device;
2786 struct drbd_peer_device *peer_device, *tmp_peer_device;
2787 struct gendisk *disk;
2788 struct request_queue *q;
2790 int vnr = adm_ctx->volume;
2791 enum drbd_ret_code err = ERR_NOMEM;
2793 device = minor_to_device(minor);
2795 return ERR_MINOR_OR_VOLUME_EXISTS;
2797 /* GFP_KERNEL, we are outside of all write-out paths */
2798 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2801 kref_init(&device->kref);
2803 kref_get(&resource->kref);
2804 device->resource = resource;
2805 device->minor = minor;
2808 drbd_init_set_defaults(device);
2810 q = blk_alloc_queue(GFP_KERNEL);
2813 device->rq_queue = q;
2814 q->queuedata = device;
2816 disk = alloc_disk(1);
2819 device->vdisk = disk;
2821 set_disk_ro(disk, true);
2824 disk->major = DRBD_MAJOR;
2825 disk->first_minor = minor;
2826 disk->fops = &drbd_ops;
2827 sprintf(disk->disk_name, "drbd%d", minor);
2828 disk->private_data = device;
2830 device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2831 /* we have no partitions. we contain only ourselves. */
2832 device->this_bdev->bd_contains = device->this_bdev;
2834 q->backing_dev_info->congested_fn = drbd_congested;
2835 q->backing_dev_info->congested_data = device;
2837 blk_queue_make_request(q, drbd_make_request);
2838 blk_queue_write_cache(q, true, true);
2839 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2840 This triggers a max_bio_size message upon first attach or connect */
2841 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2842 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2843 q->queue_lock = &resource->req_lock;
2845 device->md_io.page = alloc_page(GFP_KERNEL);
2846 if (!device->md_io.page)
2847 goto out_no_io_page;
2849 if (drbd_bm_init(device))
2851 device->read_requests = RB_ROOT;
2852 device->write_requests = RB_ROOT;
2854 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2857 err = ERR_MINOR_OR_VOLUME_EXISTS;
2858 goto out_no_minor_idr;
2860 kref_get(&device->kref);
2862 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2865 err = ERR_MINOR_OR_VOLUME_EXISTS;
2866 goto out_idr_remove_minor;
2868 kref_get(&device->kref);
2870 INIT_LIST_HEAD(&device->peer_devices);
2871 INIT_LIST_HEAD(&device->pending_bitmap_io);
2872 for_each_connection(connection, resource) {
2873 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2875 goto out_idr_remove_from_resource;
2876 peer_device->connection = connection;
2877 peer_device->device = device;
2879 list_add(&peer_device->peer_devices, &device->peer_devices);
2880 kref_get(&device->kref);
2882 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2885 err = ERR_INVALID_REQUEST;
2886 goto out_idr_remove_from_resource;
2888 kref_get(&connection->kref);
2889 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2892 if (init_submitter(device)) {
2894 goto out_idr_remove_vol;
2899 /* inherit the connection state */
2900 device->state.conn = first_connection(resource)->cstate;
2901 if (device->state.conn == C_WF_REPORT_PARAMS) {
2902 for_each_peer_device(peer_device, device)
2903 drbd_connected(peer_device);
2905 /* move to create_peer_device() */
2906 for_each_peer_device(peer_device, device)
2907 drbd_debugfs_peer_device_add(peer_device);
2908 drbd_debugfs_device_add(device);
2912 idr_remove(&connection->peer_devices, vnr);
2913 out_idr_remove_from_resource:
2914 for_each_connection(connection, resource) {
2915 peer_device = idr_remove(&connection->peer_devices, vnr);
2917 kref_put(&connection->kref, drbd_destroy_connection);
2919 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2920 list_del(&peer_device->peer_devices);
2923 idr_remove(&resource->devices, vnr);
2924 out_idr_remove_minor:
2925 idr_remove(&drbd_devices, minor);
2928 drbd_bm_cleanup(device);
2930 __free_page(device->md_io.page);
2934 blk_cleanup_queue(q);
2936 kref_put(&resource->kref, drbd_destroy_resource);
2941 void drbd_delete_device(struct drbd_device *device)
2943 struct drbd_resource *resource = device->resource;
2944 struct drbd_connection *connection;
2945 struct drbd_peer_device *peer_device;
2947 /* move to free_peer_device() */
2948 for_each_peer_device(peer_device, device)
2949 drbd_debugfs_peer_device_cleanup(peer_device);
2950 drbd_debugfs_device_cleanup(device);
2951 for_each_connection(connection, resource) {
2952 idr_remove(&connection->peer_devices, device->vnr);
2953 kref_put(&device->kref, drbd_destroy_device);
2955 idr_remove(&resource->devices, device->vnr);
2956 kref_put(&device->kref, drbd_destroy_device);
2957 idr_remove(&drbd_devices, device_to_minor(device));
2958 kref_put(&device->kref, drbd_destroy_device);
2959 del_gendisk(device->vdisk);
2961 kref_put(&device->kref, drbd_destroy_device);
2964 static int __init drbd_init(void)
2968 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2969 pr_err("invalid minor_count (%d)\n", minor_count);
2973 minor_count = DRBD_MINOR_COUNT_DEF;
2977 err = register_blkdev(DRBD_MAJOR, "drbd");
2979 pr_err("unable to register block device major %d\n",
2985 * allocate all necessary structs
2987 init_waitqueue_head(&drbd_pp_wait);
2989 drbd_proc = NULL; /* play safe for drbd_cleanup */
2990 idr_init(&drbd_devices);
2992 mutex_init(&resources_mutex);
2993 INIT_LIST_HEAD(&drbd_resources);
2995 err = drbd_genl_register();
2997 pr_err("unable to register generic netlink family\n");
3001 err = drbd_create_mempools();
3006 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
3008 pr_err("unable to register proc file\n");
3012 retry.wq = create_singlethread_workqueue("drbd-reissue");
3014 pr_err("unable to create retry workqueue\n");
3017 INIT_WORK(&retry.worker, do_retry);
3018 spin_lock_init(&retry.lock);
3019 INIT_LIST_HEAD(&retry.writes);
3021 if (drbd_debugfs_init())
3022 pr_notice("failed to initialize debugfs -- will not be available\n");
3024 pr_info("initialized. "
3025 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
3026 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
3027 pr_info("%s\n", drbd_buildtag());
3028 pr_info("registered as block device major %d\n", DRBD_MAJOR);
3029 return 0; /* Success! */
3034 pr_err("ran out of memory\n");
3036 pr_err("initialization failure\n");
3040 static void drbd_free_one_sock(struct drbd_socket *ds)
3043 mutex_lock(&ds->mutex);
3046 mutex_unlock(&ds->mutex);
3048 /* so debugfs does not need to mutex_lock() */
3050 kernel_sock_shutdown(s, SHUT_RDWR);
3055 void drbd_free_sock(struct drbd_connection *connection)
3057 if (connection->data.socket)
3058 drbd_free_one_sock(&connection->data);
3059 if (connection->meta.socket)
3060 drbd_free_one_sock(&connection->meta);
3063 /* meta data management */
3065 void conn_md_sync(struct drbd_connection *connection)
3067 struct drbd_peer_device *peer_device;
3071 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3072 struct drbd_device *device = peer_device->device;
3074 kref_get(&device->kref);
3076 drbd_md_sync(device);
3077 kref_put(&device->kref, drbd_destroy_device);
3083 /* aligned 4kByte */
3084 struct meta_data_on_disk {
3085 u64 la_size_sect; /* last agreed size. */
3086 u64 uuid[UI_SIZE]; /* UUIDs. */
3089 u32 flags; /* MDF */
3092 u32 al_offset; /* offset to this block */
3093 u32 al_nr_extents; /* important for restoring the AL (userspace) */
3094 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3095 u32 bm_offset; /* offset to the bitmap, from here */
3096 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
3097 u32 la_peer_max_bio_size; /* last peer max_bio_size */
3099 /* see al_tr_number_to_on_disk_sector() */
3101 u32 al_stripe_size_4k;
3103 u8 reserved_u8[4096 - (7*8 + 10*4)];
3108 void drbd_md_write(struct drbd_device *device, void *b)
3110 struct meta_data_on_disk *buffer = b;
3114 memset(buffer, 0, sizeof(*buffer));
3116 buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3117 for (i = UI_CURRENT; i < UI_SIZE; i++)
3118 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3119 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3120 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3122 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3123 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3124 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3125 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3126 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3128 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3129 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3131 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3132 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3134 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3135 sector = device->ldev->md.md_offset;
3137 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3138 /* this was a try anyways ... */
3139 drbd_err(device, "meta data update failed!\n");
3140 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3145 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3146 * @device: DRBD device.
3148 void drbd_md_sync(struct drbd_device *device)
3150 struct meta_data_on_disk *buffer;
3152 /* Don't accidentally change the DRBD meta data layout. */
3153 BUILD_BUG_ON(UI_SIZE != 4);
3154 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3156 del_timer(&device->md_sync_timer);
3157 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3158 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3161 /* We use here D_FAILED and not D_ATTACHING because we try to write
3162 * metadata even if we detach due to a disk failure! */
3163 if (!get_ldev_if_state(device, D_FAILED))
3166 buffer = drbd_md_get_buffer(device, __func__);
3170 drbd_md_write(device, buffer);
3172 /* Update device->ldev->md.la_size_sect,
3173 * since we updated it on metadata. */
3174 device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3176 drbd_md_put_buffer(device);
3181 static int check_activity_log_stripe_size(struct drbd_device *device,
3182 struct meta_data_on_disk *on_disk,
3183 struct drbd_md *in_core)
3185 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3186 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3189 /* both not set: default to old fixed size activity log */
3190 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3192 al_stripe_size_4k = MD_32kB_SECT/8;
3195 /* some paranoia plausibility checks */
3197 /* we need both values to be set */
3198 if (al_stripes == 0 || al_stripe_size_4k == 0)
3201 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3203 /* Upper limit of activity log area, to avoid potential overflow
3204 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3205 * than 72 * 4k blocks total only increases the amount of history,
3206 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3207 if (al_size_4k > (16 * 1024 * 1024/4))
3210 /* Lower limit: we need at least 8 transaction slots (32kB)
3211 * to not break existing setups */
3212 if (al_size_4k < MD_32kB_SECT/8)
3215 in_core->al_stripe_size_4k = al_stripe_size_4k;
3216 in_core->al_stripes = al_stripes;
3217 in_core->al_size_4k = al_size_4k;
3221 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3222 al_stripes, al_stripe_size_4k);
3226 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3228 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3229 struct drbd_md *in_core = &bdev->md;
3230 s32 on_disk_al_sect;
3231 s32 on_disk_bm_sect;
3233 /* The on-disk size of the activity log, calculated from offsets, and
3234 * the size of the activity log calculated from the stripe settings,
3236 * Though we could relax this a bit: it is ok, if the striped activity log
3237 * fits in the available on-disk activity log size.
3238 * Right now, that would break how resize is implemented.
3239 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3240 * of possible unused padding space in the on disk layout. */
3241 if (in_core->al_offset < 0) {
3242 if (in_core->bm_offset > in_core->al_offset)
3244 on_disk_al_sect = -in_core->al_offset;
3245 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3247 if (in_core->al_offset != MD_4kB_SECT)
3249 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3252 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3253 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3256 /* old fixed size meta data is exactly that: fixed. */
3257 if (in_core->meta_dev_idx >= 0) {
3258 if (in_core->md_size_sect != MD_128MB_SECT
3259 || in_core->al_offset != MD_4kB_SECT
3260 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3261 || in_core->al_stripes != 1
3262 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3266 if (capacity < in_core->md_size_sect)
3268 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3271 /* should be aligned, and at least 32k */
3272 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3275 /* should fit (for now: exactly) into the available on-disk space;
3276 * overflow prevention is in check_activity_log_stripe_size() above. */
3277 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3280 /* again, should be aligned */
3281 if (in_core->bm_offset & 7)
3284 /* FIXME check for device grow with flex external meta data? */
3286 /* can the available bitmap space cover the last agreed device size? */
3287 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3293 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3294 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3295 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3296 in_core->meta_dev_idx,
3297 in_core->al_stripes, in_core->al_stripe_size_4k,
3298 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3299 (unsigned long long)in_core->la_size_sect,
3300 (unsigned long long)capacity);
3307 * drbd_md_read() - Reads in the meta data super block
3308 * @device: DRBD device.
3309 * @bdev: Device from which the meta data should be read in.
3311 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3312 * something goes wrong.
3314 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3315 * even before @bdev is assigned to @device->ldev.
3317 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3319 struct meta_data_on_disk *buffer;
3321 int i, rv = NO_ERROR;
3323 if (device->state.disk != D_DISKLESS)
3324 return ERR_DISK_CONFIGURED;
3326 buffer = drbd_md_get_buffer(device, __func__);
3330 /* First, figure out where our meta data superblock is located,
3332 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3333 bdev->md.md_offset = drbd_md_ss(bdev);
3334 /* Even for (flexible or indexed) external meta data,
3335 * initially restrict us to the 4k superblock for now.
3336 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3337 bdev->md.md_size_sect = 8;
3339 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3341 /* NOTE: can't do normal error processing here as this is
3342 called BEFORE disk is attached */
3343 drbd_err(device, "Error while reading metadata.\n");
3344 rv = ERR_IO_MD_DISK;
3348 magic = be32_to_cpu(buffer->magic);
3349 flags = be32_to_cpu(buffer->flags);
3350 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3351 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3352 /* btw: that's Activity Log clean, not "all" clean. */
3353 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3354 rv = ERR_MD_UNCLEAN;
3358 rv = ERR_MD_INVALID;
3359 if (magic != DRBD_MD_MAGIC_08) {
3360 if (magic == DRBD_MD_MAGIC_07)
3361 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3363 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3367 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3368 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3369 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3374 /* convert to in_core endian */
3375 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3376 for (i = UI_CURRENT; i < UI_SIZE; i++)
3377 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3378 bdev->md.flags = be32_to_cpu(buffer->flags);
3379 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3381 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3382 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3383 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3385 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3387 if (check_offsets_and_sizes(device, bdev))
3390 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3391 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3392 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3395 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3396 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3397 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3403 spin_lock_irq(&device->resource->req_lock);
3404 if (device->state.conn < C_CONNECTED) {
3406 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3407 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3408 device->peer_max_bio_size = peer;
3410 spin_unlock_irq(&device->resource->req_lock);
3413 drbd_md_put_buffer(device);
3419 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3420 * @device: DRBD device.
3422 * Call this function if you change anything that should be written to
3423 * the meta-data super block. This function sets MD_DIRTY, and starts a
3424 * timer that ensures that within five seconds you have to call drbd_md_sync().
3427 void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
3429 if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3430 mod_timer(&device->md_sync_timer, jiffies + HZ);
3431 device->last_md_mark_dirty.line = line;
3432 device->last_md_mark_dirty.func = func;
3436 void drbd_md_mark_dirty(struct drbd_device *device)
3438 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3439 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3443 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3447 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3448 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3451 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3453 if (idx == UI_CURRENT) {
3454 if (device->state.role == R_PRIMARY)
3459 drbd_set_ed_uuid(device, val);
3462 device->ldev->md.uuid[idx] = val;
3463 drbd_md_mark_dirty(device);
3466 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3468 unsigned long flags;
3469 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3470 __drbd_uuid_set(device, idx, val);
3471 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3474 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3476 unsigned long flags;
3477 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3478 if (device->ldev->md.uuid[idx]) {
3479 drbd_uuid_move_history(device);
3480 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3482 __drbd_uuid_set(device, idx, val);
3483 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3487 * drbd_uuid_new_current() - Creates a new current UUID
3488 * @device: DRBD device.
3490 * Creates a new current UUID, and rotates the old current UUID into
3491 * the bitmap slot. Causes an incremental resync upon next connect.
3493 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3496 unsigned long long bm_uuid;
3498 get_random_bytes(&val, sizeof(u64));
3500 spin_lock_irq(&device->ldev->md.uuid_lock);
3501 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3504 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3506 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3507 __drbd_uuid_set(device, UI_CURRENT, val);
3508 spin_unlock_irq(&device->ldev->md.uuid_lock);
3510 drbd_print_uuids(device, "new current UUID");
3511 /* get it to stable storage _now_ */
3512 drbd_md_sync(device);
3515 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3517 unsigned long flags;
3518 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3521 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3523 drbd_uuid_move_history(device);
3524 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3525 device->ldev->md.uuid[UI_BITMAP] = 0;
3527 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3529 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3531 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3533 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3535 drbd_md_mark_dirty(device);
3539 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3540 * @device: DRBD device.
3542 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3544 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3548 drbd_md_set_flag(device, MDF_FULL_SYNC);
3549 drbd_md_sync(device);
3550 drbd_bm_set_all(device);
3552 rv = drbd_bm_write(device);
3555 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3556 drbd_md_sync(device);
3563 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3564 * @device: DRBD device.
3566 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3568 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3570 drbd_resume_al(device);
3571 drbd_bm_clear_all(device);
3572 return drbd_bm_write(device);
3575 static int w_bitmap_io(struct drbd_work *w, int unused)
3577 struct drbd_device *device =
3578 container_of(w, struct drbd_device, bm_io_work.w);
3579 struct bm_io_work *work = &device->bm_io_work;
3582 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3583 int cnt = atomic_read(&device->ap_bio_cnt);
3585 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3589 if (get_ldev(device)) {
3590 drbd_bm_lock(device, work->why, work->flags);
3591 rv = work->io_fn(device);
3592 drbd_bm_unlock(device);
3596 clear_bit_unlock(BITMAP_IO, &device->flags);
3597 wake_up(&device->misc_wait);
3600 work->done(device, rv);
3602 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3610 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3611 * @device: DRBD device.
3612 * @io_fn: IO callback to be called when bitmap IO is possible
3613 * @done: callback to be called after the bitmap IO was performed
3614 * @why: Descriptive text of the reason for doing the IO
3616 * While IO on the bitmap happens we freeze application IO thus we ensure
3617 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3618 * called from worker context. It MUST NOT be used while a previous such
3619 * work is still pending!
3621 * Its worker function encloses the call of io_fn() by get_ldev() and
3624 void drbd_queue_bitmap_io(struct drbd_device *device,
3625 int (*io_fn)(struct drbd_device *),
3626 void (*done)(struct drbd_device *, int),
3627 char *why, enum bm_flag flags)
3629 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3631 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3632 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3633 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3634 if (device->bm_io_work.why)
3635 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3636 why, device->bm_io_work.why);
3638 device->bm_io_work.io_fn = io_fn;
3639 device->bm_io_work.done = done;
3640 device->bm_io_work.why = why;
3641 device->bm_io_work.flags = flags;
3643 spin_lock_irq(&device->resource->req_lock);
3644 set_bit(BITMAP_IO, &device->flags);
3645 /* don't wait for pending application IO if the caller indicates that
3646 * application IO does not conflict anyways. */
3647 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3648 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3649 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3650 &device->bm_io_work.w);
3652 spin_unlock_irq(&device->resource->req_lock);
3656 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3657 * @device: DRBD device.
3658 * @io_fn: IO callback to be called when bitmap IO is possible
3659 * @why: Descriptive text of the reason for doing the IO
3661 * freezes application IO while that the actual IO operations runs. This
3662 * functions MAY NOT be called from worker context.
3664 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3665 char *why, enum bm_flag flags)
3667 /* Only suspend io, if some operation is supposed to be locked out */
3668 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3671 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3674 drbd_suspend_io(device);
3676 drbd_bm_lock(device, why, flags);
3678 drbd_bm_unlock(device);
3681 drbd_resume_io(device);
3686 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3688 if ((device->ldev->md.flags & flag) != flag) {
3689 drbd_md_mark_dirty(device);
3690 device->ldev->md.flags |= flag;
3694 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3696 if ((device->ldev->md.flags & flag) != 0) {
3697 drbd_md_mark_dirty(device);
3698 device->ldev->md.flags &= ~flag;
3701 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3703 return (bdev->md.flags & flag) != 0;
3706 static void md_sync_timer_fn(unsigned long data)
3708 struct drbd_device *device = (struct drbd_device *) data;
3709 drbd_device_post_work(device, MD_SYNC);
3712 const char *cmdname(enum drbd_packet cmd)
3714 /* THINK may need to become several global tables
3715 * when we want to support more than
3716 * one PRO_VERSION */
3717 static const char *cmdnames[] = {
3719 [P_WSAME] = "WriteSame",
3721 [P_DATA_REPLY] = "DataReply",
3722 [P_RS_DATA_REPLY] = "RSDataReply",
3723 [P_BARRIER] = "Barrier",
3724 [P_BITMAP] = "ReportBitMap",
3725 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3726 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3727 [P_UNPLUG_REMOTE] = "UnplugRemote",
3728 [P_DATA_REQUEST] = "DataRequest",
3729 [P_RS_DATA_REQUEST] = "RSDataRequest",
3730 [P_SYNC_PARAM] = "SyncParam",
3731 [P_SYNC_PARAM89] = "SyncParam89",
3732 [P_PROTOCOL] = "ReportProtocol",
3733 [P_UUIDS] = "ReportUUIDs",
3734 [P_SIZES] = "ReportSizes",
3735 [P_STATE] = "ReportState",
3736 [P_SYNC_UUID] = "ReportSyncUUID",
3737 [P_AUTH_CHALLENGE] = "AuthChallenge",
3738 [P_AUTH_RESPONSE] = "AuthResponse",
3740 [P_PING_ACK] = "PingAck",
3741 [P_RECV_ACK] = "RecvAck",
3742 [P_WRITE_ACK] = "WriteAck",
3743 [P_RS_WRITE_ACK] = "RSWriteAck",
3744 [P_SUPERSEDED] = "Superseded",
3745 [P_NEG_ACK] = "NegAck",
3746 [P_NEG_DREPLY] = "NegDReply",
3747 [P_NEG_RS_DREPLY] = "NegRSDReply",
3748 [P_BARRIER_ACK] = "BarrierAck",
3749 [P_STATE_CHG_REQ] = "StateChgRequest",
3750 [P_STATE_CHG_REPLY] = "StateChgReply",
3751 [P_OV_REQUEST] = "OVRequest",
3752 [P_OV_REPLY] = "OVReply",
3753 [P_OV_RESULT] = "OVResult",
3754 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3755 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3756 [P_COMPRESSED_BITMAP] = "CBitmap",
3757 [P_DELAY_PROBE] = "DelayProbe",
3758 [P_OUT_OF_SYNC] = "OutOfSync",
3759 [P_RETRY_WRITE] = "RetryWrite",
3760 [P_RS_CANCEL] = "RSCancel",
3761 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3762 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3763 [P_RETRY_WRITE] = "retry_write",
3764 [P_PROTOCOL_UPDATE] = "protocol_update",
3765 [P_RS_THIN_REQ] = "rs_thin_req",
3766 [P_RS_DEALLOCATED] = "rs_deallocated",
3768 /* enum drbd_packet, but not commands - obsoleted flags:
3774 /* too big for the array: 0xfffX */
3775 if (cmd == P_INITIAL_META)
3776 return "InitialMeta";
3777 if (cmd == P_INITIAL_DATA)
3778 return "InitialData";
3779 if (cmd == P_CONNECTION_FEATURES)
3780 return "ConnectionFeatures";
3781 if (cmd >= ARRAY_SIZE(cmdnames))
3783 return cmdnames[cmd];
3787 * drbd_wait_misc - wait for a request to make progress
3788 * @device: device associated with the request
3789 * @i: the struct drbd_interval embedded in struct drbd_request or
3790 * struct drbd_peer_request
3792 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3794 struct net_conf *nc;
3799 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3804 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3807 /* Indicate to wake up device->misc_wait on progress. */
3809 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3810 spin_unlock_irq(&device->resource->req_lock);
3811 timeout = schedule_timeout(timeout);
3812 finish_wait(&device->misc_wait, &wait);
3813 spin_lock_irq(&device->resource->req_lock);
3814 if (!timeout || device->state.conn < C_CONNECTED)
3816 if (signal_pending(current))
3817 return -ERESTARTSYS;
3821 void lock_all_resources(void)
3823 struct drbd_resource *resource;
3824 int __maybe_unused i = 0;
3826 mutex_lock(&resources_mutex);
3827 local_irq_disable();
3828 for_each_resource(resource, &drbd_resources)
3829 spin_lock_nested(&resource->req_lock, i++);
3832 void unlock_all_resources(void)
3834 struct drbd_resource *resource;
3836 for_each_resource(resource, &drbd_resources)
3837 spin_unlock(&resource->req_lock);
3839 mutex_unlock(&resources_mutex);
3842 #ifdef CONFIG_DRBD_FAULT_INJECTION
3843 /* Fault insertion support including random number generator shamelessly
3844 * stolen from kernel/rcutorture.c */
3845 struct fault_random_state {
3846 unsigned long state;
3847 unsigned long count;
3850 #define FAULT_RANDOM_MULT 39916801 /* prime */
3851 #define FAULT_RANDOM_ADD 479001701 /* prime */
3852 #define FAULT_RANDOM_REFRESH 10000
3855 * Crude but fast random-number generator. Uses a linear congruential
3856 * generator, with occasional help from get_random_bytes().
3858 static unsigned long
3859 _drbd_fault_random(struct fault_random_state *rsp)
3863 if (!rsp->count--) {
3864 get_random_bytes(&refresh, sizeof(refresh));
3865 rsp->state += refresh;
3866 rsp->count = FAULT_RANDOM_REFRESH;
3868 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3869 return swahw32(rsp->state);
3873 _drbd_fault_str(unsigned int type) {
3874 static char *_faults[] = {
3875 [DRBD_FAULT_MD_WR] = "Meta-data write",
3876 [DRBD_FAULT_MD_RD] = "Meta-data read",
3877 [DRBD_FAULT_RS_WR] = "Resync write",
3878 [DRBD_FAULT_RS_RD] = "Resync read",
3879 [DRBD_FAULT_DT_WR] = "Data write",
3880 [DRBD_FAULT_DT_RD] = "Data read",
3881 [DRBD_FAULT_DT_RA] = "Data read ahead",
3882 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3883 [DRBD_FAULT_AL_EE] = "EE allocation",
3884 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3887 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3891 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3893 static struct fault_random_state rrs = {0, 0};
3895 unsigned int ret = (
3897 ((1 << device_to_minor(device)) & fault_devs) != 0) &&
3898 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3903 if (__ratelimit(&drbd_ratelimit_state))
3904 drbd_warn(device, "***Simulating %s failure\n",
3905 _drbd_fault_str(type));
3912 const char *drbd_buildtag(void)
3914 /* DRBD built from external sources has here a reference to the
3915 git hash of the source code. */
3917 static char buildtag[38] = "\0uilt-in";
3919 if (buildtag[0] == 0) {
3921 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3930 module_init(drbd_init)
3931 module_exit(drbd_cleanup)
3933 EXPORT_SYMBOL(drbd_conn_str);
3934 EXPORT_SYMBOL(drbd_role_str);
3935 EXPORT_SYMBOL(drbd_disk_str);
3936 EXPORT_SYMBOL(drbd_set_st_err_str);