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 <asm/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>
56 #include <linux/drbd_limits.h>
58 #include "drbd_protocol.h"
59 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
61 #include "drbd_debugfs.h"
63 static DEFINE_MUTEX(drbd_main_mutex);
64 static int drbd_open(struct block_device *bdev, fmode_t mode);
65 static void drbd_release(struct gendisk *gd, fmode_t mode);
66 static void md_sync_timer_fn(unsigned long data);
67 static int w_bitmap_io(struct drbd_work *w, int unused);
69 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
70 "Lars Ellenberg <lars@linbit.com>");
71 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
72 MODULE_VERSION(REL_VERSION);
73 MODULE_LICENSE("GPL");
74 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
75 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
76 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
78 #include <linux/moduleparam.h>
79 /* allow_open_on_secondary */
80 MODULE_PARM_DESC(allow_oos, "DONT USE!");
81 /* thanks to these macros, if compiled into the kernel (not-module),
82 * this becomes the boot parameter drbd.minor_count */
83 module_param(minor_count, uint, 0444);
84 module_param(disable_sendpage, bool, 0644);
85 module_param(allow_oos, bool, 0);
86 module_param(proc_details, int, 0644);
88 #ifdef CONFIG_DRBD_FAULT_INJECTION
91 static int fault_count;
93 /* bitmap of enabled faults */
94 module_param(enable_faults, int, 0664);
95 /* fault rate % value - applies to all enabled faults */
96 module_param(fault_rate, int, 0664);
97 /* count of faults inserted */
98 module_param(fault_count, int, 0664);
99 /* bitmap of devices to insert faults on */
100 module_param(fault_devs, int, 0644);
103 /* module parameter, defined */
104 unsigned int minor_count = DRBD_MINOR_COUNT_DEF;
105 bool disable_sendpage;
107 int proc_details; /* Detail level in proc drbd*/
109 /* Module parameter for setting the user mode helper program
110 * to run. Default is /sbin/drbdadm */
111 char usermode_helper[80] = "/sbin/drbdadm";
113 module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644);
115 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
116 * as member "struct gendisk *vdisk;"
118 struct idr drbd_devices;
119 struct list_head drbd_resources;
121 struct kmem_cache *drbd_request_cache;
122 struct kmem_cache *drbd_ee_cache; /* peer requests */
123 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
124 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
125 mempool_t *drbd_request_mempool;
126 mempool_t *drbd_ee_mempool;
127 mempool_t *drbd_md_io_page_pool;
128 struct bio_set *drbd_md_io_bio_set;
130 /* I do not use a standard mempool, because:
131 1) I want to hand out the pre-allocated objects first.
132 2) I want to be able to interrupt sleeping allocation with a signal.
133 Note: This is a single linked list, the next pointer is the private
134 member of struct page.
136 struct page *drbd_pp_pool;
137 spinlock_t drbd_pp_lock;
139 wait_queue_head_t drbd_pp_wait;
141 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
143 static const struct block_device_operations drbd_ops = {
144 .owner = THIS_MODULE,
146 .release = drbd_release,
149 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
153 if (!drbd_md_io_bio_set)
154 return bio_alloc(gfp_mask, 1);
156 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
163 /* When checking with sparse, and this is an inline function, sparse will
164 give tons of false positives. When this is a real functions sparse works.
166 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
170 atomic_inc(&device->local_cnt);
171 io_allowed = (device->state.disk >= mins);
173 if (atomic_dec_and_test(&device->local_cnt))
174 wake_up(&device->misc_wait);
182 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
183 * @connection: DRBD connection.
184 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
185 * @set_size: Expected number of requests before that barrier.
187 * In case the passed barrier_nr or set_size does not match the oldest
188 * epoch of not yet barrier-acked requests, this function will cause a
189 * termination of the connection.
191 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
192 unsigned int set_size)
194 struct drbd_request *r;
195 struct drbd_request *req = NULL;
196 int expect_epoch = 0;
199 spin_lock_irq(&connection->resource->req_lock);
201 /* find oldest not yet barrier-acked write request,
202 * count writes in its epoch. */
203 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
204 const unsigned s = r->rq_state;
208 if (!(s & RQ_NET_MASK))
213 expect_epoch = req->epoch;
216 if (r->epoch != expect_epoch)
220 /* if (s & RQ_DONE): not expected */
221 /* if (!(s & RQ_NET_MASK)): not expected */
226 /* first some paranoia code */
228 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
232 if (expect_epoch != barrier_nr) {
233 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
234 barrier_nr, expect_epoch);
238 if (expect_size != set_size) {
239 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
240 barrier_nr, set_size, expect_size);
244 /* Clean up list of requests processed during current epoch. */
245 /* this extra list walk restart is paranoia,
246 * to catch requests being barrier-acked "unexpectedly".
247 * It usually should find the same req again, or some READ preceding it. */
248 list_for_each_entry(req, &connection->transfer_log, tl_requests)
249 if (req->epoch == expect_epoch)
251 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
252 if (req->epoch != expect_epoch)
254 _req_mod(req, BARRIER_ACKED);
256 spin_unlock_irq(&connection->resource->req_lock);
261 spin_unlock_irq(&connection->resource->req_lock);
262 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
267 * _tl_restart() - Walks the transfer log, and applies an action to all requests
268 * @connection: DRBD connection to operate on.
269 * @what: The action/event to perform with all request objects
271 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
272 * RESTART_FROZEN_DISK_IO.
274 /* must hold resource->req_lock */
275 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
277 struct drbd_request *req, *r;
279 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
283 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
285 spin_lock_irq(&connection->resource->req_lock);
286 _tl_restart(connection, what);
287 spin_unlock_irq(&connection->resource->req_lock);
291 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
292 * @device: DRBD device.
294 * This is called after the connection to the peer was lost. The storage covered
295 * by the requests on the transfer gets marked as our of sync. Called from the
296 * receiver thread and the worker thread.
298 void tl_clear(struct drbd_connection *connection)
300 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
304 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
305 * @device: DRBD device.
307 void tl_abort_disk_io(struct drbd_device *device)
309 struct drbd_connection *connection = first_peer_device(device)->connection;
310 struct drbd_request *req, *r;
312 spin_lock_irq(&connection->resource->req_lock);
313 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
314 if (!(req->rq_state & RQ_LOCAL_PENDING))
316 if (req->device != device)
318 _req_mod(req, ABORT_DISK_IO);
320 spin_unlock_irq(&connection->resource->req_lock);
323 static int drbd_thread_setup(void *arg)
325 struct drbd_thread *thi = (struct drbd_thread *) arg;
326 struct drbd_resource *resource = thi->resource;
330 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
335 retval = thi->function(thi);
337 spin_lock_irqsave(&thi->t_lock, flags);
339 /* if the receiver has been "EXITING", the last thing it did
340 * was set the conn state to "StandAlone",
341 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
342 * and receiver thread will be "started".
343 * drbd_thread_start needs to set "RESTARTING" in that case.
344 * t_state check and assignment needs to be within the same spinlock,
345 * so either thread_start sees EXITING, and can remap to RESTARTING,
346 * or thread_start see NONE, and can proceed as normal.
349 if (thi->t_state == RESTARTING) {
350 drbd_info(resource, "Restarting %s thread\n", thi->name);
351 thi->t_state = RUNNING;
352 spin_unlock_irqrestore(&thi->t_lock, flags);
359 complete_all(&thi->stop);
360 spin_unlock_irqrestore(&thi->t_lock, flags);
362 drbd_info(resource, "Terminating %s\n", current->comm);
364 /* Release mod reference taken when thread was started */
367 kref_put(&thi->connection->kref, drbd_destroy_connection);
368 kref_put(&resource->kref, drbd_destroy_resource);
369 module_put(THIS_MODULE);
373 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
374 int (*func) (struct drbd_thread *), const char *name)
376 spin_lock_init(&thi->t_lock);
379 thi->function = func;
380 thi->resource = resource;
381 thi->connection = NULL;
385 int drbd_thread_start(struct drbd_thread *thi)
387 struct drbd_resource *resource = thi->resource;
388 struct task_struct *nt;
391 /* is used from state engine doing drbd_thread_stop_nowait,
392 * while holding the req lock irqsave */
393 spin_lock_irqsave(&thi->t_lock, flags);
395 switch (thi->t_state) {
397 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
398 thi->name, current->comm, current->pid);
400 /* Get ref on module for thread - this is released when thread exits */
401 if (!try_module_get(THIS_MODULE)) {
402 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
403 spin_unlock_irqrestore(&thi->t_lock, flags);
407 kref_get(&resource->kref);
409 kref_get(&thi->connection->kref);
411 init_completion(&thi->stop);
412 thi->reset_cpu_mask = 1;
413 thi->t_state = RUNNING;
414 spin_unlock_irqrestore(&thi->t_lock, flags);
415 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
417 nt = kthread_create(drbd_thread_setup, (void *) thi,
418 "drbd_%c_%s", thi->name[0], thi->resource->name);
421 drbd_err(resource, "Couldn't start thread\n");
424 kref_put(&thi->connection->kref, drbd_destroy_connection);
425 kref_put(&resource->kref, drbd_destroy_resource);
426 module_put(THIS_MODULE);
429 spin_lock_irqsave(&thi->t_lock, flags);
431 thi->t_state = RUNNING;
432 spin_unlock_irqrestore(&thi->t_lock, flags);
436 thi->t_state = RESTARTING;
437 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
438 thi->name, current->comm, current->pid);
443 spin_unlock_irqrestore(&thi->t_lock, flags);
451 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
455 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
457 /* may be called from state engine, holding the req lock irqsave */
458 spin_lock_irqsave(&thi->t_lock, flags);
460 if (thi->t_state == NONE) {
461 spin_unlock_irqrestore(&thi->t_lock, flags);
463 drbd_thread_start(thi);
467 if (thi->t_state != ns) {
468 if (thi->task == NULL) {
469 spin_unlock_irqrestore(&thi->t_lock, flags);
475 init_completion(&thi->stop);
476 if (thi->task != current)
477 force_sig(DRBD_SIGKILL, thi->task);
480 spin_unlock_irqrestore(&thi->t_lock, flags);
483 wait_for_completion(&thi->stop);
486 int conn_lowest_minor(struct drbd_connection *connection)
488 struct drbd_peer_device *peer_device;
489 int vnr = 0, minor = -1;
492 peer_device = idr_get_next(&connection->peer_devices, &vnr);
494 minor = device_to_minor(peer_device->device);
502 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
504 * Forces all threads of a resource onto the same CPU. This is beneficial for
505 * DRBD's performance. May be overwritten by user's configuration.
507 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
509 unsigned int *resources_per_cpu, min_index = ~0;
511 resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL);
512 if (resources_per_cpu) {
513 struct drbd_resource *resource;
514 unsigned int cpu, min = ~0;
517 for_each_resource_rcu(resource, &drbd_resources) {
518 for_each_cpu(cpu, resource->cpu_mask)
519 resources_per_cpu[cpu]++;
522 for_each_online_cpu(cpu) {
523 if (resources_per_cpu[cpu] < min) {
524 min = resources_per_cpu[cpu];
528 kfree(resources_per_cpu);
530 if (min_index == ~0) {
531 cpumask_setall(*cpu_mask);
534 cpumask_set_cpu(min_index, *cpu_mask);
538 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
539 * @device: DRBD device.
540 * @thi: drbd_thread object
542 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
545 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
547 struct drbd_resource *resource = thi->resource;
548 struct task_struct *p = current;
550 if (!thi->reset_cpu_mask)
552 thi->reset_cpu_mask = 0;
553 set_cpus_allowed_ptr(p, resource->cpu_mask);
556 #define drbd_calc_cpu_mask(A) ({})
560 * drbd_header_size - size of a packet header
562 * The header size is a multiple of 8, so any payload following the header is
563 * word aligned on 64-bit architectures. (The bitmap send and receive code
566 unsigned int drbd_header_size(struct drbd_connection *connection)
568 if (connection->agreed_pro_version >= 100) {
569 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
570 return sizeof(struct p_header100);
572 BUILD_BUG_ON(sizeof(struct p_header80) !=
573 sizeof(struct p_header95));
574 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
575 return sizeof(struct p_header80);
579 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
581 h->magic = cpu_to_be32(DRBD_MAGIC);
582 h->command = cpu_to_be16(cmd);
583 h->length = cpu_to_be16(size);
584 return sizeof(struct p_header80);
587 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
589 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
590 h->command = cpu_to_be16(cmd);
591 h->length = cpu_to_be32(size);
592 return sizeof(struct p_header95);
595 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
598 h->magic = cpu_to_be32(DRBD_MAGIC_100);
599 h->volume = cpu_to_be16(vnr);
600 h->command = cpu_to_be16(cmd);
601 h->length = cpu_to_be32(size);
603 return sizeof(struct p_header100);
606 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
607 void *buffer, enum drbd_packet cmd, int size)
609 if (connection->agreed_pro_version >= 100)
610 return prepare_header100(buffer, cmd, size, vnr);
611 else if (connection->agreed_pro_version >= 95 &&
612 size > DRBD_MAX_SIZE_H80_PACKET)
613 return prepare_header95(buffer, cmd, size);
615 return prepare_header80(buffer, cmd, size);
618 static void *__conn_prepare_command(struct drbd_connection *connection,
619 struct drbd_socket *sock)
623 return sock->sbuf + drbd_header_size(connection);
626 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
630 mutex_lock(&sock->mutex);
631 p = __conn_prepare_command(connection, sock);
633 mutex_unlock(&sock->mutex);
638 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
640 return conn_prepare_command(peer_device->connection, sock);
643 static int __send_command(struct drbd_connection *connection, int vnr,
644 struct drbd_socket *sock, enum drbd_packet cmd,
645 unsigned int header_size, void *data,
652 * Called with @data == NULL and the size of the data blocks in @size
653 * for commands that send data blocks. For those commands, omit the
654 * MSG_MORE flag: this will increase the likelihood that data blocks
655 * which are page aligned on the sender will end up page aligned on the
658 msg_flags = data ? MSG_MORE : 0;
660 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
662 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
665 err = drbd_send_all(connection, sock->socket, data, size, 0);
666 /* DRBD protocol "pings" are latency critical.
667 * This is supposed to trigger tcp_push_pending_frames() */
668 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
669 drbd_tcp_nodelay(sock->socket);
674 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
675 enum drbd_packet cmd, unsigned int header_size,
676 void *data, unsigned int size)
678 return __send_command(connection, 0, sock, cmd, header_size, data, size);
681 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
682 enum drbd_packet cmd, unsigned int header_size,
683 void *data, unsigned int size)
687 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
688 mutex_unlock(&sock->mutex);
692 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
693 enum drbd_packet cmd, unsigned int header_size,
694 void *data, unsigned int size)
698 err = __send_command(peer_device->connection, peer_device->device->vnr,
699 sock, cmd, header_size, data, size);
700 mutex_unlock(&sock->mutex);
704 int drbd_send_ping(struct drbd_connection *connection)
706 struct drbd_socket *sock;
708 sock = &connection->meta;
709 if (!conn_prepare_command(connection, sock))
711 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
714 int drbd_send_ping_ack(struct drbd_connection *connection)
716 struct drbd_socket *sock;
718 sock = &connection->meta;
719 if (!conn_prepare_command(connection, sock))
721 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
724 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
726 struct drbd_socket *sock;
727 struct p_rs_param_95 *p;
729 const int apv = peer_device->connection->agreed_pro_version;
730 enum drbd_packet cmd;
732 struct disk_conf *dc;
734 sock = &peer_device->connection->data;
735 p = drbd_prepare_command(peer_device, sock);
740 nc = rcu_dereference(peer_device->connection->net_conf);
742 size = apv <= 87 ? sizeof(struct p_rs_param)
743 : apv == 88 ? sizeof(struct p_rs_param)
744 + strlen(nc->verify_alg) + 1
745 : apv <= 94 ? sizeof(struct p_rs_param_89)
746 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
748 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
750 /* initialize verify_alg and csums_alg */
751 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
753 if (get_ldev(peer_device->device)) {
754 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
755 p->resync_rate = cpu_to_be32(dc->resync_rate);
756 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
757 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
758 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
759 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
760 put_ldev(peer_device->device);
762 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
763 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
764 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
765 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
766 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
770 strcpy(p->verify_alg, nc->verify_alg);
772 strcpy(p->csums_alg, nc->csums_alg);
775 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
778 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
780 struct drbd_socket *sock;
781 struct p_protocol *p;
785 sock = &connection->data;
786 p = __conn_prepare_command(connection, sock);
791 nc = rcu_dereference(connection->net_conf);
793 if (nc->tentative && connection->agreed_pro_version < 92) {
795 mutex_unlock(&sock->mutex);
796 drbd_err(connection, "--dry-run is not supported by peer");
801 if (connection->agreed_pro_version >= 87)
802 size += strlen(nc->integrity_alg) + 1;
804 p->protocol = cpu_to_be32(nc->wire_protocol);
805 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
806 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
807 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
808 p->two_primaries = cpu_to_be32(nc->two_primaries);
810 if (nc->discard_my_data)
811 cf |= CF_DISCARD_MY_DATA;
814 p->conn_flags = cpu_to_be32(cf);
816 if (connection->agreed_pro_version >= 87)
817 strcpy(p->integrity_alg, nc->integrity_alg);
820 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
823 int drbd_send_protocol(struct drbd_connection *connection)
827 mutex_lock(&connection->data.mutex);
828 err = __drbd_send_protocol(connection, P_PROTOCOL);
829 mutex_unlock(&connection->data.mutex);
834 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
836 struct drbd_device *device = peer_device->device;
837 struct drbd_socket *sock;
841 if (!get_ldev_if_state(device, D_NEGOTIATING))
844 sock = &peer_device->connection->data;
845 p = drbd_prepare_command(peer_device, sock);
850 spin_lock_irq(&device->ldev->md.uuid_lock);
851 for (i = UI_CURRENT; i < UI_SIZE; i++)
852 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
853 spin_unlock_irq(&device->ldev->md.uuid_lock);
855 device->comm_bm_set = drbd_bm_total_weight(device);
856 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
858 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
860 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
861 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
862 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
865 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
868 int drbd_send_uuids(struct drbd_peer_device *peer_device)
870 return _drbd_send_uuids(peer_device, 0);
873 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
875 return _drbd_send_uuids(peer_device, 8);
878 void drbd_print_uuids(struct drbd_device *device, const char *text)
880 if (get_ldev_if_state(device, D_NEGOTIATING)) {
881 u64 *uuid = device->ldev->md.uuid;
882 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
884 (unsigned long long)uuid[UI_CURRENT],
885 (unsigned long long)uuid[UI_BITMAP],
886 (unsigned long long)uuid[UI_HISTORY_START],
887 (unsigned long long)uuid[UI_HISTORY_END]);
890 drbd_info(device, "%s effective data uuid: %016llX\n",
892 (unsigned long long)device->ed_uuid);
896 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
898 struct drbd_device *device = peer_device->device;
899 struct drbd_socket *sock;
903 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
905 uuid = device->ldev->md.uuid[UI_BITMAP];
906 if (uuid && uuid != UUID_JUST_CREATED)
907 uuid = uuid + UUID_NEW_BM_OFFSET;
909 get_random_bytes(&uuid, sizeof(u64));
910 drbd_uuid_set(device, UI_BITMAP, uuid);
911 drbd_print_uuids(device, "updated sync UUID");
912 drbd_md_sync(device);
914 sock = &peer_device->connection->data;
915 p = drbd_prepare_command(peer_device, sock);
917 p->uuid = cpu_to_be64(uuid);
918 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
922 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
924 struct drbd_device *device = peer_device->device;
925 struct drbd_socket *sock;
927 sector_t d_size, u_size;
929 unsigned int max_bio_size;
931 if (get_ldev_if_state(device, D_NEGOTIATING)) {
932 D_ASSERT(device, device->ldev->backing_bdev);
933 d_size = drbd_get_max_capacity(device->ldev);
935 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
937 q_order_type = drbd_queue_order_type(device);
938 max_bio_size = queue_max_hw_sectors(device->ldev->backing_bdev->bd_disk->queue) << 9;
939 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
944 q_order_type = QUEUE_ORDERED_NONE;
945 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
948 sock = &peer_device->connection->data;
949 p = drbd_prepare_command(peer_device, sock);
953 if (peer_device->connection->agreed_pro_version <= 94)
954 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
955 else if (peer_device->connection->agreed_pro_version < 100)
956 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
958 p->d_size = cpu_to_be64(d_size);
959 p->u_size = cpu_to_be64(u_size);
960 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
961 p->max_bio_size = cpu_to_be32(max_bio_size);
962 p->queue_order_type = cpu_to_be16(q_order_type);
963 p->dds_flags = cpu_to_be16(flags);
964 return drbd_send_command(peer_device, sock, P_SIZES, sizeof(*p), NULL, 0);
968 * drbd_send_current_state() - Sends the drbd state to the peer
969 * @peer_device: DRBD peer device.
971 int drbd_send_current_state(struct drbd_peer_device *peer_device)
973 struct drbd_socket *sock;
976 sock = &peer_device->connection->data;
977 p = drbd_prepare_command(peer_device, sock);
980 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
981 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
985 * drbd_send_state() - After a state change, sends the new state to the peer
986 * @peer_device: DRBD peer device.
987 * @state: the state to send, not necessarily the current state.
989 * Each state change queues an "after_state_ch" work, which will eventually
990 * send the resulting new state to the peer. If more state changes happen
991 * between queuing and processing of the after_state_ch work, we still
992 * want to send each intermediary state in the order it occurred.
994 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
996 struct drbd_socket *sock;
999 sock = &peer_device->connection->data;
1000 p = drbd_prepare_command(peer_device, sock);
1003 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1004 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1007 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1009 struct drbd_socket *sock;
1010 struct p_req_state *p;
1012 sock = &peer_device->connection->data;
1013 p = drbd_prepare_command(peer_device, sock);
1016 p->mask = cpu_to_be32(mask.i);
1017 p->val = cpu_to_be32(val.i);
1018 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1021 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1023 enum drbd_packet cmd;
1024 struct drbd_socket *sock;
1025 struct p_req_state *p;
1027 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1028 sock = &connection->data;
1029 p = conn_prepare_command(connection, sock);
1032 p->mask = cpu_to_be32(mask.i);
1033 p->val = cpu_to_be32(val.i);
1034 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1037 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1039 struct drbd_socket *sock;
1040 struct p_req_state_reply *p;
1042 sock = &peer_device->connection->meta;
1043 p = drbd_prepare_command(peer_device, sock);
1045 p->retcode = cpu_to_be32(retcode);
1046 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1050 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1052 struct drbd_socket *sock;
1053 struct p_req_state_reply *p;
1054 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1056 sock = &connection->meta;
1057 p = conn_prepare_command(connection, sock);
1059 p->retcode = cpu_to_be32(retcode);
1060 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1064 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1066 BUG_ON(code & ~0xf);
1067 p->encoding = (p->encoding & ~0xf) | code;
1070 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1072 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1075 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1078 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1081 static int fill_bitmap_rle_bits(struct drbd_device *device,
1082 struct p_compressed_bm *p,
1084 struct bm_xfer_ctx *c)
1086 struct bitstream bs;
1087 unsigned long plain_bits;
1094 /* may we use this feature? */
1096 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1098 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1101 if (c->bit_offset >= c->bm_bits)
1102 return 0; /* nothing to do. */
1104 /* use at most thus many bytes */
1105 bitstream_init(&bs, p->code, size, 0);
1106 memset(p->code, 0, size);
1107 /* plain bits covered in this code string */
1110 /* p->encoding & 0x80 stores whether the first run length is set.
1111 * bit offset is implicit.
1112 * start with toggle == 2 to be able to tell the first iteration */
1115 /* see how much plain bits we can stuff into one packet
1116 * using RLE and VLI. */
1118 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1119 : _drbd_bm_find_next(device, c->bit_offset);
1122 rl = tmp - c->bit_offset;
1124 if (toggle == 2) { /* first iteration */
1126 /* the first checked bit was set,
1127 * store start value, */
1128 dcbp_set_start(p, 1);
1129 /* but skip encoding of zero run length */
1133 dcbp_set_start(p, 0);
1136 /* paranoia: catch zero runlength.
1137 * can only happen if bitmap is modified while we scan it. */
1139 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1140 "t:%u bo:%lu\n", toggle, c->bit_offset);
1144 bits = vli_encode_bits(&bs, rl);
1145 if (bits == -ENOBUFS) /* buffer full */
1148 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1154 c->bit_offset = tmp;
1155 } while (c->bit_offset < c->bm_bits);
1157 len = bs.cur.b - p->code + !!bs.cur.bit;
1159 if (plain_bits < (len << 3)) {
1160 /* incompressible with this method.
1161 * we need to rewind both word and bit position. */
1162 c->bit_offset -= plain_bits;
1163 bm_xfer_ctx_bit_to_word_offset(c);
1164 c->bit_offset = c->word_offset * BITS_PER_LONG;
1168 /* RLE + VLI was able to compress it just fine.
1169 * update c->word_offset. */
1170 bm_xfer_ctx_bit_to_word_offset(c);
1172 /* store pad_bits */
1173 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1179 * send_bitmap_rle_or_plain
1181 * Return 0 when done, 1 when another iteration is needed, and a negative error
1182 * code upon failure.
1185 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1187 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1188 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1189 struct p_compressed_bm *p = sock->sbuf + header_size;
1192 len = fill_bitmap_rle_bits(device, p,
1193 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1198 dcbp_set_code(p, RLE_VLI_Bits);
1199 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1200 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1203 c->bytes[0] += header_size + sizeof(*p) + len;
1205 if (c->bit_offset >= c->bm_bits)
1208 /* was not compressible.
1209 * send a buffer full of plain text bits instead. */
1210 unsigned int data_size;
1211 unsigned long num_words;
1212 unsigned long *p = sock->sbuf + header_size;
1214 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1215 num_words = min_t(size_t, data_size / sizeof(*p),
1216 c->bm_words - c->word_offset);
1217 len = num_words * sizeof(*p);
1219 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1220 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1221 c->word_offset += num_words;
1222 c->bit_offset = c->word_offset * BITS_PER_LONG;
1225 c->bytes[1] += header_size + len;
1227 if (c->bit_offset > c->bm_bits)
1228 c->bit_offset = c->bm_bits;
1232 INFO_bm_xfer_stats(device, "send", c);
1240 /* See the comment at receive_bitmap() */
1241 static int _drbd_send_bitmap(struct drbd_device *device)
1243 struct bm_xfer_ctx c;
1246 if (!expect(device->bitmap))
1249 if (get_ldev(device)) {
1250 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1251 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1252 drbd_bm_set_all(device);
1253 if (drbd_bm_write(device)) {
1254 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1255 * but otherwise process as per normal - need to tell other
1256 * side that a full resync is required! */
1257 drbd_err(device, "Failed to write bitmap to disk!\n");
1259 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1260 drbd_md_sync(device);
1266 c = (struct bm_xfer_ctx) {
1267 .bm_bits = drbd_bm_bits(device),
1268 .bm_words = drbd_bm_words(device),
1272 err = send_bitmap_rle_or_plain(device, &c);
1278 int drbd_send_bitmap(struct drbd_device *device)
1280 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1283 mutex_lock(&sock->mutex);
1285 err = !_drbd_send_bitmap(device);
1286 mutex_unlock(&sock->mutex);
1290 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1292 struct drbd_socket *sock;
1293 struct p_barrier_ack *p;
1295 if (connection->cstate < C_WF_REPORT_PARAMS)
1298 sock = &connection->meta;
1299 p = conn_prepare_command(connection, sock);
1302 p->barrier = barrier_nr;
1303 p->set_size = cpu_to_be32(set_size);
1304 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1308 * _drbd_send_ack() - Sends an ack packet
1309 * @device: DRBD device.
1310 * @cmd: Packet command code.
1311 * @sector: sector, needs to be in big endian byte order
1312 * @blksize: size in byte, needs to be in big endian byte order
1313 * @block_id: Id, big endian byte order
1315 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1316 u64 sector, u32 blksize, u64 block_id)
1318 struct drbd_socket *sock;
1319 struct p_block_ack *p;
1321 if (peer_device->device->state.conn < C_CONNECTED)
1324 sock = &peer_device->connection->meta;
1325 p = drbd_prepare_command(peer_device, sock);
1329 p->block_id = block_id;
1330 p->blksize = blksize;
1331 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1332 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1335 /* dp->sector and dp->block_id already/still in network byte order,
1336 * data_size is payload size according to dp->head,
1337 * and may need to be corrected for digest size. */
1338 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1339 struct p_data *dp, int data_size)
1341 if (peer_device->connection->peer_integrity_tfm)
1342 data_size -= crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm);
1343 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1347 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1348 struct p_block_req *rp)
1350 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1354 * drbd_send_ack() - Sends an ack packet
1355 * @device: DRBD device
1356 * @cmd: packet command code
1357 * @peer_req: peer request
1359 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1360 struct drbd_peer_request *peer_req)
1362 return _drbd_send_ack(peer_device, cmd,
1363 cpu_to_be64(peer_req->i.sector),
1364 cpu_to_be32(peer_req->i.size),
1365 peer_req->block_id);
1368 /* This function misuses the block_id field to signal if the blocks
1369 * are is sync or not. */
1370 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1371 sector_t sector, int blksize, u64 block_id)
1373 return _drbd_send_ack(peer_device, cmd,
1374 cpu_to_be64(sector),
1375 cpu_to_be32(blksize),
1376 cpu_to_be64(block_id));
1379 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1380 sector_t sector, int size, u64 block_id)
1382 struct drbd_socket *sock;
1383 struct p_block_req *p;
1385 sock = &peer_device->connection->data;
1386 p = drbd_prepare_command(peer_device, sock);
1389 p->sector = cpu_to_be64(sector);
1390 p->block_id = block_id;
1391 p->blksize = cpu_to_be32(size);
1392 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1395 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1396 void *digest, int digest_size, enum drbd_packet cmd)
1398 struct drbd_socket *sock;
1399 struct p_block_req *p;
1401 /* FIXME: Put the digest into the preallocated socket buffer. */
1403 sock = &peer_device->connection->data;
1404 p = drbd_prepare_command(peer_device, sock);
1407 p->sector = cpu_to_be64(sector);
1408 p->block_id = ID_SYNCER /* unused */;
1409 p->blksize = cpu_to_be32(size);
1410 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1413 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1415 struct drbd_socket *sock;
1416 struct p_block_req *p;
1418 sock = &peer_device->connection->data;
1419 p = drbd_prepare_command(peer_device, sock);
1422 p->sector = cpu_to_be64(sector);
1423 p->block_id = ID_SYNCER /* unused */;
1424 p->blksize = cpu_to_be32(size);
1425 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1428 /* called on sndtimeo
1429 * returns false if we should retry,
1430 * true if we think connection is dead
1432 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1435 /* long elapsed = (long)(jiffies - device->last_received); */
1437 drop_it = connection->meta.socket == sock
1438 || !connection->asender.task
1439 || get_t_state(&connection->asender) != RUNNING
1440 || connection->cstate < C_WF_REPORT_PARAMS;
1445 drop_it = !--connection->ko_count;
1447 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1448 current->comm, current->pid, connection->ko_count);
1449 request_ping(connection);
1452 return drop_it; /* && (device->state == R_PRIMARY) */;
1455 static void drbd_update_congested(struct drbd_connection *connection)
1457 struct sock *sk = connection->data.socket->sk;
1458 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1459 set_bit(NET_CONGESTED, &connection->flags);
1462 /* The idea of sendpage seems to be to put some kind of reference
1463 * to the page into the skb, and to hand it over to the NIC. In
1464 * this process get_page() gets called.
1466 * As soon as the page was really sent over the network put_page()
1467 * gets called by some part of the network layer. [ NIC driver? ]
1469 * [ get_page() / put_page() increment/decrement the count. If count
1470 * reaches 0 the page will be freed. ]
1472 * This works nicely with pages from FSs.
1473 * But this means that in protocol A we might signal IO completion too early!
1475 * In order not to corrupt data during a resync we must make sure
1476 * that we do not reuse our own buffer pages (EEs) to early, therefore
1477 * we have the net_ee list.
1479 * XFS seems to have problems, still, it submits pages with page_count == 0!
1480 * As a workaround, we disable sendpage on pages
1481 * with page_count == 0 or PageSlab.
1483 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1484 int offset, size_t size, unsigned msg_flags)
1486 struct socket *socket;
1490 socket = peer_device->connection->data.socket;
1491 addr = kmap(page) + offset;
1492 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1495 peer_device->device->send_cnt += size >> 9;
1499 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1500 int offset, size_t size, unsigned msg_flags)
1502 struct socket *socket = peer_device->connection->data.socket;
1503 mm_segment_t oldfs = get_fs();
1507 /* e.g. XFS meta- & log-data is in slab pages, which have a
1508 * page_count of 0 and/or have PageSlab() set.
1509 * we cannot use send_page for those, as that does get_page();
1510 * put_page(); and would cause either a VM_BUG directly, or
1511 * __page_cache_release a page that would actually still be referenced
1512 * by someone, leading to some obscure delayed Oops somewhere else. */
1513 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1514 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1516 msg_flags |= MSG_NOSIGNAL;
1517 drbd_update_congested(peer_device->connection);
1522 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1524 if (sent == -EAGAIN) {
1525 if (we_should_drop_the_connection(peer_device->connection, socket))
1529 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1530 __func__, (int)size, len, sent);
1537 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1539 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1543 peer_device->device->send_cnt += size >> 9;
1548 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1550 struct bio_vec bvec;
1551 struct bvec_iter iter;
1553 /* hint all but last page with MSG_MORE */
1554 bio_for_each_segment(bvec, bio, iter) {
1557 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1558 bvec.bv_offset, bvec.bv_len,
1559 bio_iter_last(bvec, iter)
1567 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1569 struct bio_vec bvec;
1570 struct bvec_iter iter;
1572 /* hint all but last page with MSG_MORE */
1573 bio_for_each_segment(bvec, bio, iter) {
1576 err = _drbd_send_page(peer_device, bvec.bv_page,
1577 bvec.bv_offset, bvec.bv_len,
1578 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1585 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1586 struct drbd_peer_request *peer_req)
1588 struct page *page = peer_req->pages;
1589 unsigned len = peer_req->i.size;
1592 /* hint all but last page with MSG_MORE */
1593 page_chain_for_each(page) {
1594 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1596 err = _drbd_send_page(peer_device, page, 0, l,
1597 page_chain_next(page) ? MSG_MORE : 0);
1605 static u32 bio_flags_to_wire(struct drbd_connection *connection, unsigned long bi_rw)
1607 if (connection->agreed_pro_version >= 95)
1608 return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) |
1609 (bi_rw & REQ_FUA ? DP_FUA : 0) |
1610 (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) |
1611 (bi_rw & REQ_DISCARD ? DP_DISCARD : 0);
1613 return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0;
1616 /* Used to send write or TRIM aka REQ_DISCARD requests
1617 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1619 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1621 struct drbd_device *device = peer_device->device;
1622 struct drbd_socket *sock;
1624 unsigned int dp_flags = 0;
1628 sock = &peer_device->connection->data;
1629 p = drbd_prepare_command(peer_device, sock);
1630 dgs = peer_device->connection->integrity_tfm ?
1631 crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
1635 p->sector = cpu_to_be64(req->i.sector);
1636 p->block_id = (unsigned long)req;
1637 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1638 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio->bi_rw);
1639 if (device->state.conn >= C_SYNC_SOURCE &&
1640 device->state.conn <= C_PAUSED_SYNC_T)
1641 dp_flags |= DP_MAY_SET_IN_SYNC;
1642 if (peer_device->connection->agreed_pro_version >= 100) {
1643 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1644 dp_flags |= DP_SEND_RECEIVE_ACK;
1645 /* During resync, request an explicit write ack,
1646 * even in protocol != C */
1647 if (req->rq_state & RQ_EXP_WRITE_ACK
1648 || (dp_flags & DP_MAY_SET_IN_SYNC))
1649 dp_flags |= DP_SEND_WRITE_ACK;
1651 p->dp_flags = cpu_to_be32(dp_flags);
1653 if (dp_flags & DP_DISCARD) {
1654 struct p_trim *t = (struct p_trim*)p;
1655 t->size = cpu_to_be32(req->i.size);
1656 err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1660 /* our digest is still only over the payload.
1661 * TRIM does not carry any payload. */
1663 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, p + 1);
1664 err = __send_command(peer_device->connection, device->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size);
1666 /* For protocol A, we have to memcpy the payload into
1667 * socket buffers, as we may complete right away
1668 * as soon as we handed it over to tcp, at which point the data
1669 * pages may become invalid.
1671 * For data-integrity enabled, we copy it as well, so we can be
1672 * sure that even if the bio pages may still be modified, it
1673 * won't change the data on the wire, thus if the digest checks
1674 * out ok after sending on this side, but does not fit on the
1675 * receiving side, we sure have detected corruption elsewhere.
1677 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs)
1678 err = _drbd_send_bio(peer_device, req->master_bio);
1680 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1682 /* double check digest, sometimes buffers have been modified in flight. */
1683 if (dgs > 0 && dgs <= 64) {
1684 /* 64 byte, 512 bit, is the largest digest size
1685 * currently supported in kernel crypto. */
1686 unsigned char digest[64];
1687 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1688 if (memcmp(p + 1, digest, dgs)) {
1690 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1691 (unsigned long long)req->i.sector, req->i.size);
1693 } /* else if (dgs > 64) {
1694 ... Be noisy about digest too large ...
1698 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1703 /* answer packet, used to send data back for read requests:
1704 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1705 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1707 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1708 struct drbd_peer_request *peer_req)
1710 struct drbd_device *device = peer_device->device;
1711 struct drbd_socket *sock;
1716 sock = &peer_device->connection->data;
1717 p = drbd_prepare_command(peer_device, sock);
1719 dgs = peer_device->connection->integrity_tfm ?
1720 crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
1724 p->sector = cpu_to_be64(peer_req->i.sector);
1725 p->block_id = peer_req->block_id;
1726 p->seq_num = 0; /* unused */
1729 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1730 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size);
1732 err = _drbd_send_zc_ee(peer_device, peer_req);
1733 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1738 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1740 struct drbd_socket *sock;
1741 struct p_block_desc *p;
1743 sock = &peer_device->connection->data;
1744 p = drbd_prepare_command(peer_device, sock);
1747 p->sector = cpu_to_be64(req->i.sector);
1748 p->blksize = cpu_to_be32(req->i.size);
1749 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1753 drbd_send distinguishes two cases:
1755 Packets sent via the data socket "sock"
1756 and packets sent via the meta data socket "msock"
1759 -----------------+-------------------------+------------------------------
1760 timeout conf.timeout / 2 conf.timeout / 2
1761 timeout action send a ping via msock Abort communication
1762 and close all sockets
1766 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1768 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1769 void *buf, size_t size, unsigned msg_flags)
1778 /* THINK if (signal_pending) return ... ? */
1783 msg.msg_name = NULL;
1784 msg.msg_namelen = 0;
1785 msg.msg_control = NULL;
1786 msg.msg_controllen = 0;
1787 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1789 if (sock == connection->data.socket) {
1791 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1793 drbd_update_congested(connection);
1797 * tcp_sendmsg does _not_ use its size parameter at all ?
1799 * -EAGAIN on timeout, -EINTR on signal.
1802 * do we need to block DRBD_SIG if sock == &meta.socket ??
1803 * otherwise wake_asender() might interrupt some send_*Ack !
1805 rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
1806 if (rv == -EAGAIN) {
1807 if (we_should_drop_the_connection(connection, sock))
1813 flush_signals(current);
1821 } while (sent < size);
1823 if (sock == connection->data.socket)
1824 clear_bit(NET_CONGESTED, &connection->flags);
1827 if (rv != -EAGAIN) {
1828 drbd_err(connection, "%s_sendmsg returned %d\n",
1829 sock == connection->meta.socket ? "msock" : "sock",
1831 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1833 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1840 * drbd_send_all - Send an entire buffer
1842 * Returns 0 upon success and a negative error value otherwise.
1844 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1845 size_t size, unsigned msg_flags)
1849 err = drbd_send(connection, sock, buffer, size, msg_flags);
1857 static int drbd_open(struct block_device *bdev, fmode_t mode)
1859 struct drbd_device *device = bdev->bd_disk->private_data;
1860 unsigned long flags;
1863 mutex_lock(&drbd_main_mutex);
1864 spin_lock_irqsave(&device->resource->req_lock, flags);
1865 /* to have a stable device->state.role
1866 * and no race with updating open_cnt */
1868 if (device->state.role != R_PRIMARY) {
1869 if (mode & FMODE_WRITE)
1871 else if (!allow_oos)
1877 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1878 mutex_unlock(&drbd_main_mutex);
1883 static void drbd_release(struct gendisk *gd, fmode_t mode)
1885 struct drbd_device *device = gd->private_data;
1886 mutex_lock(&drbd_main_mutex);
1888 mutex_unlock(&drbd_main_mutex);
1891 static void drbd_set_defaults(struct drbd_device *device)
1893 /* Beware! The actual layout differs
1894 * between big endian and little endian */
1895 device->state = (union drbd_dev_state) {
1896 { .role = R_SECONDARY,
1898 .conn = C_STANDALONE,
1904 void drbd_init_set_defaults(struct drbd_device *device)
1906 /* the memset(,0,) did most of this.
1907 * note: only assignments, no allocation in here */
1909 drbd_set_defaults(device);
1911 atomic_set(&device->ap_bio_cnt, 0);
1912 atomic_set(&device->ap_actlog_cnt, 0);
1913 atomic_set(&device->ap_pending_cnt, 0);
1914 atomic_set(&device->rs_pending_cnt, 0);
1915 atomic_set(&device->unacked_cnt, 0);
1916 atomic_set(&device->local_cnt, 0);
1917 atomic_set(&device->pp_in_use_by_net, 0);
1918 atomic_set(&device->rs_sect_in, 0);
1919 atomic_set(&device->rs_sect_ev, 0);
1920 atomic_set(&device->ap_in_flight, 0);
1921 atomic_set(&device->md_io.in_use, 0);
1923 mutex_init(&device->own_state_mutex);
1924 device->state_mutex = &device->own_state_mutex;
1926 spin_lock_init(&device->al_lock);
1927 spin_lock_init(&device->peer_seq_lock);
1929 INIT_LIST_HEAD(&device->active_ee);
1930 INIT_LIST_HEAD(&device->sync_ee);
1931 INIT_LIST_HEAD(&device->done_ee);
1932 INIT_LIST_HEAD(&device->read_ee);
1933 INIT_LIST_HEAD(&device->net_ee);
1934 INIT_LIST_HEAD(&device->resync_reads);
1935 INIT_LIST_HEAD(&device->resync_work.list);
1936 INIT_LIST_HEAD(&device->unplug_work.list);
1937 INIT_LIST_HEAD(&device->bm_io_work.w.list);
1938 INIT_LIST_HEAD(&device->pending_master_completion[0]);
1939 INIT_LIST_HEAD(&device->pending_master_completion[1]);
1940 INIT_LIST_HEAD(&device->pending_completion[0]);
1941 INIT_LIST_HEAD(&device->pending_completion[1]);
1943 device->resync_work.cb = w_resync_timer;
1944 device->unplug_work.cb = w_send_write_hint;
1945 device->bm_io_work.w.cb = w_bitmap_io;
1947 init_timer(&device->resync_timer);
1948 init_timer(&device->md_sync_timer);
1949 init_timer(&device->start_resync_timer);
1950 init_timer(&device->request_timer);
1951 device->resync_timer.function = resync_timer_fn;
1952 device->resync_timer.data = (unsigned long) device;
1953 device->md_sync_timer.function = md_sync_timer_fn;
1954 device->md_sync_timer.data = (unsigned long) device;
1955 device->start_resync_timer.function = start_resync_timer_fn;
1956 device->start_resync_timer.data = (unsigned long) device;
1957 device->request_timer.function = request_timer_fn;
1958 device->request_timer.data = (unsigned long) device;
1960 init_waitqueue_head(&device->misc_wait);
1961 init_waitqueue_head(&device->state_wait);
1962 init_waitqueue_head(&device->ee_wait);
1963 init_waitqueue_head(&device->al_wait);
1964 init_waitqueue_head(&device->seq_wait);
1966 device->resync_wenr = LC_FREE;
1967 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1968 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1971 void drbd_device_cleanup(struct drbd_device *device)
1974 if (first_peer_device(device)->connection->receiver.t_state != NONE)
1975 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
1976 first_peer_device(device)->connection->receiver.t_state);
1978 device->al_writ_cnt =
1979 device->bm_writ_cnt =
1987 device->rs_failed = 0;
1988 device->rs_last_events = 0;
1989 device->rs_last_sect_ev = 0;
1990 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1991 device->rs_mark_left[i] = 0;
1992 device->rs_mark_time[i] = 0;
1994 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
1996 drbd_set_my_capacity(device, 0);
1997 if (device->bitmap) {
1998 /* maybe never allocated. */
1999 drbd_bm_resize(device, 0, 1);
2000 drbd_bm_cleanup(device);
2003 drbd_free_ldev(device->ldev);
2004 device->ldev = NULL;
2006 clear_bit(AL_SUSPENDED, &device->flags);
2008 D_ASSERT(device, list_empty(&device->active_ee));
2009 D_ASSERT(device, list_empty(&device->sync_ee));
2010 D_ASSERT(device, list_empty(&device->done_ee));
2011 D_ASSERT(device, list_empty(&device->read_ee));
2012 D_ASSERT(device, list_empty(&device->net_ee));
2013 D_ASSERT(device, list_empty(&device->resync_reads));
2014 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2015 D_ASSERT(device, list_empty(&device->resync_work.list));
2016 D_ASSERT(device, list_empty(&device->unplug_work.list));
2018 drbd_set_defaults(device);
2022 static void drbd_destroy_mempools(void)
2026 while (drbd_pp_pool) {
2027 page = drbd_pp_pool;
2028 drbd_pp_pool = (struct page *)page_private(page);
2033 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2035 if (drbd_md_io_bio_set)
2036 bioset_free(drbd_md_io_bio_set);
2037 if (drbd_md_io_page_pool)
2038 mempool_destroy(drbd_md_io_page_pool);
2039 if (drbd_ee_mempool)
2040 mempool_destroy(drbd_ee_mempool);
2041 if (drbd_request_mempool)
2042 mempool_destroy(drbd_request_mempool);
2044 kmem_cache_destroy(drbd_ee_cache);
2045 if (drbd_request_cache)
2046 kmem_cache_destroy(drbd_request_cache);
2047 if (drbd_bm_ext_cache)
2048 kmem_cache_destroy(drbd_bm_ext_cache);
2049 if (drbd_al_ext_cache)
2050 kmem_cache_destroy(drbd_al_ext_cache);
2052 drbd_md_io_bio_set = NULL;
2053 drbd_md_io_page_pool = NULL;
2054 drbd_ee_mempool = NULL;
2055 drbd_request_mempool = NULL;
2056 drbd_ee_cache = NULL;
2057 drbd_request_cache = NULL;
2058 drbd_bm_ext_cache = NULL;
2059 drbd_al_ext_cache = NULL;
2064 static int drbd_create_mempools(void)
2067 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
2070 /* prepare our caches and mempools */
2071 drbd_request_mempool = NULL;
2072 drbd_ee_cache = NULL;
2073 drbd_request_cache = NULL;
2074 drbd_bm_ext_cache = NULL;
2075 drbd_al_ext_cache = NULL;
2076 drbd_pp_pool = NULL;
2077 drbd_md_io_page_pool = NULL;
2078 drbd_md_io_bio_set = NULL;
2081 drbd_request_cache = kmem_cache_create(
2082 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2083 if (drbd_request_cache == NULL)
2086 drbd_ee_cache = kmem_cache_create(
2087 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2088 if (drbd_ee_cache == NULL)
2091 drbd_bm_ext_cache = kmem_cache_create(
2092 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2093 if (drbd_bm_ext_cache == NULL)
2096 drbd_al_ext_cache = kmem_cache_create(
2097 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2098 if (drbd_al_ext_cache == NULL)
2102 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2103 if (drbd_md_io_bio_set == NULL)
2106 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2107 if (drbd_md_io_page_pool == NULL)
2110 drbd_request_mempool = mempool_create(number,
2111 mempool_alloc_slab, mempool_free_slab, drbd_request_cache);
2112 if (drbd_request_mempool == NULL)
2115 drbd_ee_mempool = mempool_create(number,
2116 mempool_alloc_slab, mempool_free_slab, drbd_ee_cache);
2117 if (drbd_ee_mempool == NULL)
2120 /* drbd's page pool */
2121 spin_lock_init(&drbd_pp_lock);
2123 for (i = 0; i < number; i++) {
2124 page = alloc_page(GFP_HIGHUSER);
2127 set_page_private(page, (unsigned long)drbd_pp_pool);
2128 drbd_pp_pool = page;
2130 drbd_pp_vacant = number;
2135 drbd_destroy_mempools(); /* in case we allocated some */
2139 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2143 rr = drbd_free_peer_reqs(device, &device->active_ee);
2145 drbd_err(device, "%d EEs in active list found!\n", rr);
2147 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2149 drbd_err(device, "%d EEs in sync list found!\n", rr);
2151 rr = drbd_free_peer_reqs(device, &device->read_ee);
2153 drbd_err(device, "%d EEs in read list found!\n", rr);
2155 rr = drbd_free_peer_reqs(device, &device->done_ee);
2157 drbd_err(device, "%d EEs in done list found!\n", rr);
2159 rr = drbd_free_peer_reqs(device, &device->net_ee);
2161 drbd_err(device, "%d EEs in net list found!\n", rr);
2164 /* caution. no locking. */
2165 void drbd_destroy_device(struct kref *kref)
2167 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2168 struct drbd_resource *resource = device->resource;
2169 struct drbd_peer_device *peer_device, *tmp_peer_device;
2171 del_timer_sync(&device->request_timer);
2173 /* paranoia asserts */
2174 D_ASSERT(device, device->open_cnt == 0);
2175 /* end paranoia asserts */
2177 /* cleanup stuff that may have been allocated during
2178 * device (re-)configuration or state changes */
2180 if (device->this_bdev)
2181 bdput(device->this_bdev);
2183 drbd_free_ldev(device->ldev);
2184 device->ldev = NULL;
2186 drbd_release_all_peer_reqs(device);
2188 lc_destroy(device->act_log);
2189 lc_destroy(device->resync);
2191 kfree(device->p_uuid);
2192 /* device->p_uuid = NULL; */
2194 if (device->bitmap) /* should no longer be there. */
2195 drbd_bm_cleanup(device);
2196 __free_page(device->md_io.page);
2197 put_disk(device->vdisk);
2198 blk_cleanup_queue(device->rq_queue);
2199 kfree(device->rs_plan_s);
2201 /* not for_each_connection(connection, resource):
2202 * those may have been cleaned up and disassociated already.
2204 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2205 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2208 memset(device, 0xfd, sizeof(*device));
2210 kref_put(&resource->kref, drbd_destroy_resource);
2213 /* One global retry thread, if we need to push back some bio and have it
2214 * reinserted through our make request function.
2216 static struct retry_worker {
2217 struct workqueue_struct *wq;
2218 struct work_struct worker;
2221 struct list_head writes;
2224 static void do_retry(struct work_struct *ws)
2226 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2228 struct drbd_request *req, *tmp;
2230 spin_lock_irq(&retry->lock);
2231 list_splice_init(&retry->writes, &writes);
2232 spin_unlock_irq(&retry->lock);
2234 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2235 struct drbd_device *device = req->device;
2236 struct bio *bio = req->master_bio;
2237 unsigned long start_jif = req->start_jif;
2241 expect(atomic_read(&req->completion_ref) == 0) &&
2242 expect(req->rq_state & RQ_POSTPONED) &&
2243 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2244 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2247 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2248 req, atomic_read(&req->completion_ref),
2251 /* We still need to put one kref associated with the
2252 * "completion_ref" going zero in the code path that queued it
2253 * here. The request object may still be referenced by a
2254 * frozen local req->private_bio, in case we force-detached.
2256 kref_put(&req->kref, drbd_req_destroy);
2258 /* A single suspended or otherwise blocking device may stall
2259 * all others as well. Fortunately, this code path is to
2260 * recover from a situation that "should not happen":
2261 * concurrent writes in multi-primary setup.
2262 * In a "normal" lifecycle, this workqueue is supposed to be
2263 * destroyed without ever doing anything.
2264 * If it turns out to be an issue anyways, we can do per
2265 * resource (replication group) or per device (minor) retry
2266 * workqueues instead.
2269 /* We are not just doing generic_make_request(),
2270 * as we want to keep the start_time information. */
2272 __drbd_make_request(device, bio, start_jif);
2276 /* called via drbd_req_put_completion_ref(),
2277 * holds resource->req_lock */
2278 void drbd_restart_request(struct drbd_request *req)
2280 unsigned long flags;
2281 spin_lock_irqsave(&retry.lock, flags);
2282 list_move_tail(&req->tl_requests, &retry.writes);
2283 spin_unlock_irqrestore(&retry.lock, flags);
2285 /* Drop the extra reference that would otherwise
2286 * have been dropped by complete_master_bio.
2287 * do_retry() needs to grab a new one. */
2288 dec_ap_bio(req->device);
2290 queue_work(retry.wq, &retry.worker);
2293 void drbd_destroy_resource(struct kref *kref)
2295 struct drbd_resource *resource =
2296 container_of(kref, struct drbd_resource, kref);
2298 idr_destroy(&resource->devices);
2299 free_cpumask_var(resource->cpu_mask);
2300 kfree(resource->name);
2301 memset(resource, 0xf2, sizeof(*resource));
2305 void drbd_free_resource(struct drbd_resource *resource)
2307 struct drbd_connection *connection, *tmp;
2309 for_each_connection_safe(connection, tmp, resource) {
2310 list_del(&connection->connections);
2311 drbd_debugfs_connection_cleanup(connection);
2312 kref_put(&connection->kref, drbd_destroy_connection);
2314 drbd_debugfs_resource_cleanup(resource);
2315 kref_put(&resource->kref, drbd_destroy_resource);
2318 static void drbd_cleanup(void)
2321 struct drbd_device *device;
2322 struct drbd_resource *resource, *tmp;
2324 /* first remove proc,
2325 * drbdsetup uses it's presence to detect
2326 * whether DRBD is loaded.
2327 * If we would get stuck in proc removal,
2328 * but have netlink already deregistered,
2329 * some drbdsetup commands may wait forever
2333 remove_proc_entry("drbd", NULL);
2336 destroy_workqueue(retry.wq);
2338 drbd_genl_unregister();
2339 drbd_debugfs_cleanup();
2341 idr_for_each_entry(&drbd_devices, device, i)
2342 drbd_delete_device(device);
2344 /* not _rcu since, no other updater anymore. Genl already unregistered */
2345 for_each_resource_safe(resource, tmp, &drbd_resources) {
2346 list_del(&resource->resources);
2347 drbd_free_resource(resource);
2350 drbd_destroy_mempools();
2351 unregister_blkdev(DRBD_MAJOR, "drbd");
2353 idr_destroy(&drbd_devices);
2355 pr_info("module cleanup done.\n");
2359 * drbd_congested() - Callback for the flusher thread
2360 * @congested_data: User data
2361 * @bdi_bits: Bits the BDI flusher thread is currently interested in
2363 * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2365 static int drbd_congested(void *congested_data, int bdi_bits)
2367 struct drbd_device *device = congested_data;
2368 struct request_queue *q;
2372 if (!may_inc_ap_bio(device)) {
2373 /* DRBD has frozen IO */
2379 if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
2380 r |= (1 << BDI_async_congested);
2381 /* Without good local data, we would need to read from remote,
2382 * and that would need the worker thread as well, which is
2383 * currently blocked waiting for that usermode helper to
2386 if (!get_ldev_if_state(device, D_UP_TO_DATE))
2387 r |= (1 << BDI_sync_congested);
2395 if (get_ldev(device)) {
2396 q = bdev_get_queue(device->ldev->backing_bdev);
2397 r = bdi_congested(&q->backing_dev_info, bdi_bits);
2403 if (bdi_bits & (1 << BDI_async_congested) &&
2404 test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
2405 r |= (1 << BDI_async_congested);
2406 reason = reason == 'b' ? 'a' : 'n';
2410 device->congestion_reason = reason;
2414 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2416 spin_lock_init(&wq->q_lock);
2417 INIT_LIST_HEAD(&wq->q);
2418 init_waitqueue_head(&wq->q_wait);
2421 struct completion_work {
2423 struct completion done;
2426 static int w_complete(struct drbd_work *w, int cancel)
2428 struct completion_work *completion_work =
2429 container_of(w, struct completion_work, w);
2431 complete(&completion_work->done);
2435 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2437 struct completion_work completion_work;
2439 completion_work.w.cb = w_complete;
2440 init_completion(&completion_work.done);
2441 drbd_queue_work(work_queue, &completion_work.w);
2442 wait_for_completion(&completion_work.done);
2445 struct drbd_resource *drbd_find_resource(const char *name)
2447 struct drbd_resource *resource;
2449 if (!name || !name[0])
2453 for_each_resource_rcu(resource, &drbd_resources) {
2454 if (!strcmp(resource->name, name)) {
2455 kref_get(&resource->kref);
2465 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2466 void *peer_addr, int peer_addr_len)
2468 struct drbd_resource *resource;
2469 struct drbd_connection *connection;
2472 for_each_resource_rcu(resource, &drbd_resources) {
2473 for_each_connection_rcu(connection, resource) {
2474 if (connection->my_addr_len == my_addr_len &&
2475 connection->peer_addr_len == peer_addr_len &&
2476 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2477 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2478 kref_get(&connection->kref);
2489 static int drbd_alloc_socket(struct drbd_socket *socket)
2491 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2494 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2500 static void drbd_free_socket(struct drbd_socket *socket)
2502 free_page((unsigned long) socket->sbuf);
2503 free_page((unsigned long) socket->rbuf);
2506 void conn_free_crypto(struct drbd_connection *connection)
2508 drbd_free_sock(connection);
2510 crypto_free_hash(connection->csums_tfm);
2511 crypto_free_hash(connection->verify_tfm);
2512 crypto_free_hash(connection->cram_hmac_tfm);
2513 crypto_free_hash(connection->integrity_tfm);
2514 crypto_free_hash(connection->peer_integrity_tfm);
2515 kfree(connection->int_dig_in);
2516 kfree(connection->int_dig_vv);
2518 connection->csums_tfm = NULL;
2519 connection->verify_tfm = NULL;
2520 connection->cram_hmac_tfm = NULL;
2521 connection->integrity_tfm = NULL;
2522 connection->peer_integrity_tfm = NULL;
2523 connection->int_dig_in = NULL;
2524 connection->int_dig_vv = NULL;
2527 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2529 struct drbd_connection *connection;
2530 cpumask_var_t new_cpu_mask;
2533 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2536 retcode = ERR_NOMEM;
2537 drbd_msg_put_info("unable to allocate cpumask");
2540 /* silently ignore cpu mask on UP kernel */
2541 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2542 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2543 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2544 if (err == -EOVERFLOW) {
2545 /* So what. mask it out. */
2546 cpumask_var_t tmp_cpu_mask;
2547 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2548 cpumask_setall(tmp_cpu_mask);
2549 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2550 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2552 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2554 free_cpumask_var(tmp_cpu_mask);
2559 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2560 /* retcode = ERR_CPU_MASK_PARSE; */
2564 resource->res_opts = *res_opts;
2565 if (cpumask_empty(new_cpu_mask))
2566 drbd_calc_cpu_mask(&new_cpu_mask);
2567 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2568 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2569 for_each_connection_rcu(connection, resource) {
2570 connection->receiver.reset_cpu_mask = 1;
2571 connection->asender.reset_cpu_mask = 1;
2572 connection->worker.reset_cpu_mask = 1;
2578 free_cpumask_var(new_cpu_mask);
2583 struct drbd_resource *drbd_create_resource(const char *name)
2585 struct drbd_resource *resource;
2587 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2590 resource->name = kstrdup(name, GFP_KERNEL);
2591 if (!resource->name)
2592 goto fail_free_resource;
2593 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2594 goto fail_free_name;
2595 kref_init(&resource->kref);
2596 idr_init(&resource->devices);
2597 INIT_LIST_HEAD(&resource->connections);
2598 resource->write_ordering = WO_bdev_flush;
2599 list_add_tail_rcu(&resource->resources, &drbd_resources);
2600 mutex_init(&resource->conf_update);
2601 mutex_init(&resource->adm_mutex);
2602 spin_lock_init(&resource->req_lock);
2603 drbd_debugfs_resource_add(resource);
2607 kfree(resource->name);
2614 /* caller must be under adm_mutex */
2615 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2617 struct drbd_resource *resource;
2618 struct drbd_connection *connection;
2620 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2624 if (drbd_alloc_socket(&connection->data))
2626 if (drbd_alloc_socket(&connection->meta))
2629 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2630 if (!connection->current_epoch)
2633 INIT_LIST_HEAD(&connection->transfer_log);
2635 INIT_LIST_HEAD(&connection->current_epoch->list);
2636 connection->epochs = 1;
2637 spin_lock_init(&connection->epoch_lock);
2639 connection->send.seen_any_write_yet = false;
2640 connection->send.current_epoch_nr = 0;
2641 connection->send.current_epoch_writes = 0;
2643 resource = drbd_create_resource(name);
2647 connection->cstate = C_STANDALONE;
2648 mutex_init(&connection->cstate_mutex);
2649 init_waitqueue_head(&connection->ping_wait);
2650 idr_init(&connection->peer_devices);
2652 drbd_init_workqueue(&connection->sender_work);
2653 mutex_init(&connection->data.mutex);
2654 mutex_init(&connection->meta.mutex);
2656 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2657 connection->receiver.connection = connection;
2658 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2659 connection->worker.connection = connection;
2660 drbd_thread_init(resource, &connection->asender, drbd_asender, "asender");
2661 connection->asender.connection = connection;
2663 kref_init(&connection->kref);
2665 connection->resource = resource;
2667 if (set_resource_options(resource, res_opts))
2670 kref_get(&resource->kref);
2671 list_add_tail_rcu(&connection->connections, &resource->connections);
2672 drbd_debugfs_connection_add(connection);
2676 list_del(&resource->resources);
2677 drbd_free_resource(resource);
2679 kfree(connection->current_epoch);
2680 drbd_free_socket(&connection->meta);
2681 drbd_free_socket(&connection->data);
2686 void drbd_destroy_connection(struct kref *kref)
2688 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2689 struct drbd_resource *resource = connection->resource;
2691 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
2692 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2693 kfree(connection->current_epoch);
2695 idr_destroy(&connection->peer_devices);
2697 drbd_free_socket(&connection->meta);
2698 drbd_free_socket(&connection->data);
2699 kfree(connection->int_dig_in);
2700 kfree(connection->int_dig_vv);
2701 memset(connection, 0xfc, sizeof(*connection));
2703 kref_put(&resource->kref, drbd_destroy_resource);
2706 static int init_submitter(struct drbd_device *device)
2708 /* opencoded create_singlethread_workqueue(),
2709 * to be able to say "drbd%d", ..., minor */
2710 device->submit.wq = alloc_workqueue("drbd%u_submit",
2711 WQ_UNBOUND | WQ_MEM_RECLAIM, 1, device->minor);
2712 if (!device->submit.wq)
2715 INIT_WORK(&device->submit.worker, do_submit);
2716 INIT_LIST_HEAD(&device->submit.writes);
2720 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2722 struct drbd_resource *resource = adm_ctx->resource;
2723 struct drbd_connection *connection;
2724 struct drbd_device *device;
2725 struct drbd_peer_device *peer_device, *tmp_peer_device;
2726 struct gendisk *disk;
2727 struct request_queue *q;
2729 int vnr = adm_ctx->volume;
2730 enum drbd_ret_code err = ERR_NOMEM;
2732 device = minor_to_device(minor);
2734 return ERR_MINOR_EXISTS;
2736 /* GFP_KERNEL, we are outside of all write-out paths */
2737 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2740 kref_init(&device->kref);
2742 kref_get(&resource->kref);
2743 device->resource = resource;
2744 device->minor = minor;
2747 drbd_init_set_defaults(device);
2749 q = blk_alloc_queue(GFP_KERNEL);
2752 device->rq_queue = q;
2753 q->queuedata = device;
2755 disk = alloc_disk(1);
2758 device->vdisk = disk;
2760 set_disk_ro(disk, true);
2763 disk->major = DRBD_MAJOR;
2764 disk->first_minor = minor;
2765 disk->fops = &drbd_ops;
2766 sprintf(disk->disk_name, "drbd%d", minor);
2767 disk->private_data = device;
2769 device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2770 /* we have no partitions. we contain only ourselves. */
2771 device->this_bdev->bd_contains = device->this_bdev;
2773 q->backing_dev_info.congested_fn = drbd_congested;
2774 q->backing_dev_info.congested_data = device;
2776 blk_queue_make_request(q, drbd_make_request);
2777 blk_queue_flush(q, REQ_FLUSH | REQ_FUA);
2778 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2779 This triggers a max_bio_size message upon first attach or connect */
2780 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2781 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2782 blk_queue_merge_bvec(q, drbd_merge_bvec);
2783 q->queue_lock = &resource->req_lock;
2785 device->md_io.page = alloc_page(GFP_KERNEL);
2786 if (!device->md_io.page)
2787 goto out_no_io_page;
2789 if (drbd_bm_init(device))
2791 device->read_requests = RB_ROOT;
2792 device->write_requests = RB_ROOT;
2794 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2796 if (id == -ENOSPC) {
2797 err = ERR_MINOR_EXISTS;
2798 drbd_msg_put_info(adm_ctx->reply_skb, "requested minor exists already");
2800 goto out_no_minor_idr;
2802 kref_get(&device->kref);
2804 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2806 if (id == -ENOSPC) {
2807 err = ERR_MINOR_EXISTS;
2808 drbd_msg_put_info(adm_ctx->reply_skb, "requested minor exists already");
2810 goto out_idr_remove_minor;
2812 kref_get(&device->kref);
2814 INIT_LIST_HEAD(&device->peer_devices);
2815 INIT_LIST_HEAD(&device->pending_bitmap_io);
2816 for_each_connection(connection, resource) {
2817 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2819 goto out_idr_remove_from_resource;
2820 peer_device->connection = connection;
2821 peer_device->device = device;
2823 list_add(&peer_device->peer_devices, &device->peer_devices);
2824 kref_get(&device->kref);
2826 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2828 if (id == -ENOSPC) {
2829 err = ERR_INVALID_REQUEST;
2830 drbd_msg_put_info(adm_ctx->reply_skb, "requested volume exists already");
2832 goto out_idr_remove_from_resource;
2834 kref_get(&connection->kref);
2837 if (init_submitter(device)) {
2839 drbd_msg_put_info(adm_ctx->reply_skb, "unable to create submit workqueue");
2840 goto out_idr_remove_vol;
2845 /* inherit the connection state */
2846 device->state.conn = first_connection(resource)->cstate;
2847 if (device->state.conn == C_WF_REPORT_PARAMS) {
2848 for_each_peer_device(peer_device, device)
2849 drbd_connected(peer_device);
2851 /* move to create_peer_device() */
2852 for_each_peer_device(peer_device, device)
2853 drbd_debugfs_peer_device_add(peer_device);
2854 drbd_debugfs_device_add(device);
2858 idr_remove(&connection->peer_devices, vnr);
2859 out_idr_remove_from_resource:
2860 for_each_connection(connection, resource) {
2861 peer_device = idr_find(&connection->peer_devices, vnr);
2863 idr_remove(&connection->peer_devices, vnr);
2864 kref_put(&connection->kref, drbd_destroy_connection);
2867 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2868 list_del(&peer_device->peer_devices);
2871 idr_remove(&resource->devices, vnr);
2872 out_idr_remove_minor:
2873 idr_remove(&drbd_devices, minor);
2876 drbd_bm_cleanup(device);
2878 __free_page(device->md_io.page);
2882 blk_cleanup_queue(q);
2884 kref_put(&resource->kref, drbd_destroy_resource);
2889 void drbd_delete_device(struct drbd_device *device)
2891 struct drbd_resource *resource = device->resource;
2892 struct drbd_connection *connection;
2893 struct drbd_peer_device *peer_device;
2896 /* move to free_peer_device() */
2897 for_each_peer_device(peer_device, device)
2898 drbd_debugfs_peer_device_cleanup(peer_device);
2899 drbd_debugfs_device_cleanup(device);
2900 for_each_connection(connection, resource) {
2901 idr_remove(&connection->peer_devices, device->vnr);
2904 idr_remove(&resource->devices, device->vnr);
2905 idr_remove(&drbd_devices, device_to_minor(device));
2906 del_gendisk(device->vdisk);
2908 kref_sub(&device->kref, refs, drbd_destroy_device);
2911 static int __init drbd_init(void)
2915 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2916 pr_err("invalid minor_count (%d)\n", minor_count);
2920 minor_count = DRBD_MINOR_COUNT_DEF;
2924 err = register_blkdev(DRBD_MAJOR, "drbd");
2926 pr_err("unable to register block device major %d\n",
2932 * allocate all necessary structs
2934 init_waitqueue_head(&drbd_pp_wait);
2936 drbd_proc = NULL; /* play safe for drbd_cleanup */
2937 idr_init(&drbd_devices);
2939 rwlock_init(&global_state_lock);
2940 INIT_LIST_HEAD(&drbd_resources);
2942 err = drbd_genl_register();
2944 pr_err("unable to register generic netlink family\n");
2948 err = drbd_create_mempools();
2953 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
2955 pr_err("unable to register proc file\n");
2959 retry.wq = create_singlethread_workqueue("drbd-reissue");
2961 pr_err("unable to create retry workqueue\n");
2964 INIT_WORK(&retry.worker, do_retry);
2965 spin_lock_init(&retry.lock);
2966 INIT_LIST_HEAD(&retry.writes);
2968 if (drbd_debugfs_init())
2969 pr_notice("failed to initialize debugfs -- will not be available\n");
2971 pr_info("initialized. "
2972 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2973 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2974 pr_info("%s\n", drbd_buildtag());
2975 pr_info("registered as block device major %d\n", DRBD_MAJOR);
2976 return 0; /* Success! */
2981 pr_err("ran out of memory\n");
2983 pr_err("initialization failure\n");
2987 void drbd_free_ldev(struct drbd_backing_dev *ldev)
2992 blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2993 blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2995 kfree(ldev->disk_conf);
2999 static void drbd_free_one_sock(struct drbd_socket *ds)
3002 mutex_lock(&ds->mutex);
3005 mutex_unlock(&ds->mutex);
3007 /* so debugfs does not need to mutex_lock() */
3009 kernel_sock_shutdown(s, SHUT_RDWR);
3014 void drbd_free_sock(struct drbd_connection *connection)
3016 if (connection->data.socket)
3017 drbd_free_one_sock(&connection->data);
3018 if (connection->meta.socket)
3019 drbd_free_one_sock(&connection->meta);
3022 /* meta data management */
3024 void conn_md_sync(struct drbd_connection *connection)
3026 struct drbd_peer_device *peer_device;
3030 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3031 struct drbd_device *device = peer_device->device;
3033 kref_get(&device->kref);
3035 drbd_md_sync(device);
3036 kref_put(&device->kref, drbd_destroy_device);
3042 /* aligned 4kByte */
3043 struct meta_data_on_disk {
3044 u64 la_size_sect; /* last agreed size. */
3045 u64 uuid[UI_SIZE]; /* UUIDs. */
3048 u32 flags; /* MDF */
3051 u32 al_offset; /* offset to this block */
3052 u32 al_nr_extents; /* important for restoring the AL (userspace) */
3053 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3054 u32 bm_offset; /* offset to the bitmap, from here */
3055 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
3056 u32 la_peer_max_bio_size; /* last peer max_bio_size */
3058 /* see al_tr_number_to_on_disk_sector() */
3060 u32 al_stripe_size_4k;
3062 u8 reserved_u8[4096 - (7*8 + 10*4)];
3067 void drbd_md_write(struct drbd_device *device, void *b)
3069 struct meta_data_on_disk *buffer = b;
3073 memset(buffer, 0, sizeof(*buffer));
3075 buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3076 for (i = UI_CURRENT; i < UI_SIZE; i++)
3077 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3078 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3079 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3081 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3082 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3083 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3084 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3085 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3087 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3088 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3090 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3091 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3093 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3094 sector = device->ldev->md.md_offset;
3096 if (drbd_md_sync_page_io(device, device->ldev, sector, WRITE)) {
3097 /* this was a try anyways ... */
3098 drbd_err(device, "meta data update failed!\n");
3099 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3104 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3105 * @device: DRBD device.
3107 void drbd_md_sync(struct drbd_device *device)
3109 struct meta_data_on_disk *buffer;
3111 /* Don't accidentally change the DRBD meta data layout. */
3112 BUILD_BUG_ON(UI_SIZE != 4);
3113 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3115 del_timer(&device->md_sync_timer);
3116 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3117 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3120 /* We use here D_FAILED and not D_ATTACHING because we try to write
3121 * metadata even if we detach due to a disk failure! */
3122 if (!get_ldev_if_state(device, D_FAILED))
3125 buffer = drbd_md_get_buffer(device, __func__);
3129 drbd_md_write(device, buffer);
3131 /* Update device->ldev->md.la_size_sect,
3132 * since we updated it on metadata. */
3133 device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3135 drbd_md_put_buffer(device);
3140 static int check_activity_log_stripe_size(struct drbd_device *device,
3141 struct meta_data_on_disk *on_disk,
3142 struct drbd_md *in_core)
3144 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3145 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3148 /* both not set: default to old fixed size activity log */
3149 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3151 al_stripe_size_4k = MD_32kB_SECT/8;
3154 /* some paranoia plausibility checks */
3156 /* we need both values to be set */
3157 if (al_stripes == 0 || al_stripe_size_4k == 0)
3160 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3162 /* Upper limit of activity log area, to avoid potential overflow
3163 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3164 * than 72 * 4k blocks total only increases the amount of history,
3165 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3166 if (al_size_4k > (16 * 1024 * 1024/4))
3169 /* Lower limit: we need at least 8 transaction slots (32kB)
3170 * to not break existing setups */
3171 if (al_size_4k < MD_32kB_SECT/8)
3174 in_core->al_stripe_size_4k = al_stripe_size_4k;
3175 in_core->al_stripes = al_stripes;
3176 in_core->al_size_4k = al_size_4k;
3180 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3181 al_stripes, al_stripe_size_4k);
3185 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3187 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3188 struct drbd_md *in_core = &bdev->md;
3189 s32 on_disk_al_sect;
3190 s32 on_disk_bm_sect;
3192 /* The on-disk size of the activity log, calculated from offsets, and
3193 * the size of the activity log calculated from the stripe settings,
3195 * Though we could relax this a bit: it is ok, if the striped activity log
3196 * fits in the available on-disk activity log size.
3197 * Right now, that would break how resize is implemented.
3198 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3199 * of possible unused padding space in the on disk layout. */
3200 if (in_core->al_offset < 0) {
3201 if (in_core->bm_offset > in_core->al_offset)
3203 on_disk_al_sect = -in_core->al_offset;
3204 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3206 if (in_core->al_offset != MD_4kB_SECT)
3208 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3211 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3212 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3215 /* old fixed size meta data is exactly that: fixed. */
3216 if (in_core->meta_dev_idx >= 0) {
3217 if (in_core->md_size_sect != MD_128MB_SECT
3218 || in_core->al_offset != MD_4kB_SECT
3219 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3220 || in_core->al_stripes != 1
3221 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3225 if (capacity < in_core->md_size_sect)
3227 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3230 /* should be aligned, and at least 32k */
3231 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3234 /* should fit (for now: exactly) into the available on-disk space;
3235 * overflow prevention is in check_activity_log_stripe_size() above. */
3236 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3239 /* again, should be aligned */
3240 if (in_core->bm_offset & 7)
3243 /* FIXME check for device grow with flex external meta data? */
3245 /* can the available bitmap space cover the last agreed device size? */
3246 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3252 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3253 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3254 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3255 in_core->meta_dev_idx,
3256 in_core->al_stripes, in_core->al_stripe_size_4k,
3257 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3258 (unsigned long long)in_core->la_size_sect,
3259 (unsigned long long)capacity);
3266 * drbd_md_read() - Reads in the meta data super block
3267 * @device: DRBD device.
3268 * @bdev: Device from which the meta data should be read in.
3270 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3271 * something goes wrong.
3273 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3274 * even before @bdev is assigned to @device->ldev.
3276 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3278 struct meta_data_on_disk *buffer;
3280 int i, rv = NO_ERROR;
3282 if (device->state.disk != D_DISKLESS)
3283 return ERR_DISK_CONFIGURED;
3285 buffer = drbd_md_get_buffer(device, __func__);
3289 /* First, figure out where our meta data superblock is located,
3291 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3292 bdev->md.md_offset = drbd_md_ss(bdev);
3294 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, READ)) {
3295 /* NOTE: can't do normal error processing here as this is
3296 called BEFORE disk is attached */
3297 drbd_err(device, "Error while reading metadata.\n");
3298 rv = ERR_IO_MD_DISK;
3302 magic = be32_to_cpu(buffer->magic);
3303 flags = be32_to_cpu(buffer->flags);
3304 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3305 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3306 /* btw: that's Activity Log clean, not "all" clean. */
3307 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3308 rv = ERR_MD_UNCLEAN;
3312 rv = ERR_MD_INVALID;
3313 if (magic != DRBD_MD_MAGIC_08) {
3314 if (magic == DRBD_MD_MAGIC_07)
3315 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3317 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3321 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3322 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3323 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3328 /* convert to in_core endian */
3329 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3330 for (i = UI_CURRENT; i < UI_SIZE; i++)
3331 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3332 bdev->md.flags = be32_to_cpu(buffer->flags);
3333 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3335 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3336 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3337 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3339 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3341 if (check_offsets_and_sizes(device, bdev))
3344 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3345 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3346 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3349 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3350 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3351 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3357 spin_lock_irq(&device->resource->req_lock);
3358 if (device->state.conn < C_CONNECTED) {
3360 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3361 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3362 device->peer_max_bio_size = peer;
3364 spin_unlock_irq(&device->resource->req_lock);
3367 drbd_md_put_buffer(device);
3373 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3374 * @device: DRBD device.
3376 * Call this function if you change anything that should be written to
3377 * the meta-data super block. This function sets MD_DIRTY, and starts a
3378 * timer that ensures that within five seconds you have to call drbd_md_sync().
3381 void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
3383 if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3384 mod_timer(&device->md_sync_timer, jiffies + HZ);
3385 device->last_md_mark_dirty.line = line;
3386 device->last_md_mark_dirty.func = func;
3390 void drbd_md_mark_dirty(struct drbd_device *device)
3392 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3393 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3397 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3401 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3402 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3405 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3407 if (idx == UI_CURRENT) {
3408 if (device->state.role == R_PRIMARY)
3413 drbd_set_ed_uuid(device, val);
3416 device->ldev->md.uuid[idx] = val;
3417 drbd_md_mark_dirty(device);
3420 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3422 unsigned long flags;
3423 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3424 __drbd_uuid_set(device, idx, val);
3425 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3428 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3430 unsigned long flags;
3431 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3432 if (device->ldev->md.uuid[idx]) {
3433 drbd_uuid_move_history(device);
3434 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3436 __drbd_uuid_set(device, idx, val);
3437 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3441 * drbd_uuid_new_current() - Creates a new current UUID
3442 * @device: DRBD device.
3444 * Creates a new current UUID, and rotates the old current UUID into
3445 * the bitmap slot. Causes an incremental resync upon next connect.
3447 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3450 unsigned long long bm_uuid;
3452 get_random_bytes(&val, sizeof(u64));
3454 spin_lock_irq(&device->ldev->md.uuid_lock);
3455 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3458 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3460 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3461 __drbd_uuid_set(device, UI_CURRENT, val);
3462 spin_unlock_irq(&device->ldev->md.uuid_lock);
3464 drbd_print_uuids(device, "new current UUID");
3465 /* get it to stable storage _now_ */
3466 drbd_md_sync(device);
3469 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3471 unsigned long flags;
3472 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3475 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3477 drbd_uuid_move_history(device);
3478 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3479 device->ldev->md.uuid[UI_BITMAP] = 0;
3481 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3483 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3485 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3487 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3489 drbd_md_mark_dirty(device);
3493 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3494 * @device: DRBD device.
3496 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3498 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3502 drbd_md_set_flag(device, MDF_FULL_SYNC);
3503 drbd_md_sync(device);
3504 drbd_bm_set_all(device);
3506 rv = drbd_bm_write(device);
3509 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3510 drbd_md_sync(device);
3517 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3518 * @device: DRBD device.
3520 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3522 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3524 drbd_resume_al(device);
3525 drbd_bm_clear_all(device);
3526 return drbd_bm_write(device);
3529 static int w_bitmap_io(struct drbd_work *w, int unused)
3531 struct drbd_device *device =
3532 container_of(w, struct drbd_device, bm_io_work.w);
3533 struct bm_io_work *work = &device->bm_io_work;
3536 D_ASSERT(device, atomic_read(&device->ap_bio_cnt) == 0);
3538 if (get_ldev(device)) {
3539 drbd_bm_lock(device, work->why, work->flags);
3540 rv = work->io_fn(device);
3541 drbd_bm_unlock(device);
3545 clear_bit_unlock(BITMAP_IO, &device->flags);
3546 wake_up(&device->misc_wait);
3549 work->done(device, rv);
3551 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3559 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3560 * @device: DRBD device.
3561 * @io_fn: IO callback to be called when bitmap IO is possible
3562 * @done: callback to be called after the bitmap IO was performed
3563 * @why: Descriptive text of the reason for doing the IO
3565 * While IO on the bitmap happens we freeze application IO thus we ensure
3566 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3567 * called from worker context. It MUST NOT be used while a previous such
3568 * work is still pending!
3570 * Its worker function encloses the call of io_fn() by get_ldev() and
3573 void drbd_queue_bitmap_io(struct drbd_device *device,
3574 int (*io_fn)(struct drbd_device *),
3575 void (*done)(struct drbd_device *, int),
3576 char *why, enum bm_flag flags)
3578 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3580 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3581 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3582 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3583 if (device->bm_io_work.why)
3584 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3585 why, device->bm_io_work.why);
3587 device->bm_io_work.io_fn = io_fn;
3588 device->bm_io_work.done = done;
3589 device->bm_io_work.why = why;
3590 device->bm_io_work.flags = flags;
3592 spin_lock_irq(&device->resource->req_lock);
3593 set_bit(BITMAP_IO, &device->flags);
3594 if (atomic_read(&device->ap_bio_cnt) == 0) {
3595 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3596 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3597 &device->bm_io_work.w);
3599 spin_unlock_irq(&device->resource->req_lock);
3603 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3604 * @device: DRBD device.
3605 * @io_fn: IO callback to be called when bitmap IO is possible
3606 * @why: Descriptive text of the reason for doing the IO
3608 * freezes application IO while that the actual IO operations runs. This
3609 * functions MAY NOT be called from worker context.
3611 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3612 char *why, enum bm_flag flags)
3616 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3618 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3619 drbd_suspend_io(device);
3621 drbd_bm_lock(device, why, flags);
3623 drbd_bm_unlock(device);
3625 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
3626 drbd_resume_io(device);
3631 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3633 if ((device->ldev->md.flags & flag) != flag) {
3634 drbd_md_mark_dirty(device);
3635 device->ldev->md.flags |= flag;
3639 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3641 if ((device->ldev->md.flags & flag) != 0) {
3642 drbd_md_mark_dirty(device);
3643 device->ldev->md.flags &= ~flag;
3646 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3648 return (bdev->md.flags & flag) != 0;
3651 static void md_sync_timer_fn(unsigned long data)
3653 struct drbd_device *device = (struct drbd_device *) data;
3654 drbd_device_post_work(device, MD_SYNC);
3657 const char *cmdname(enum drbd_packet cmd)
3659 /* THINK may need to become several global tables
3660 * when we want to support more than
3661 * one PRO_VERSION */
3662 static const char *cmdnames[] = {
3664 [P_DATA_REPLY] = "DataReply",
3665 [P_RS_DATA_REPLY] = "RSDataReply",
3666 [P_BARRIER] = "Barrier",
3667 [P_BITMAP] = "ReportBitMap",
3668 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3669 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3670 [P_UNPLUG_REMOTE] = "UnplugRemote",
3671 [P_DATA_REQUEST] = "DataRequest",
3672 [P_RS_DATA_REQUEST] = "RSDataRequest",
3673 [P_SYNC_PARAM] = "SyncParam",
3674 [P_SYNC_PARAM89] = "SyncParam89",
3675 [P_PROTOCOL] = "ReportProtocol",
3676 [P_UUIDS] = "ReportUUIDs",
3677 [P_SIZES] = "ReportSizes",
3678 [P_STATE] = "ReportState",
3679 [P_SYNC_UUID] = "ReportSyncUUID",
3680 [P_AUTH_CHALLENGE] = "AuthChallenge",
3681 [P_AUTH_RESPONSE] = "AuthResponse",
3683 [P_PING_ACK] = "PingAck",
3684 [P_RECV_ACK] = "RecvAck",
3685 [P_WRITE_ACK] = "WriteAck",
3686 [P_RS_WRITE_ACK] = "RSWriteAck",
3687 [P_SUPERSEDED] = "Superseded",
3688 [P_NEG_ACK] = "NegAck",
3689 [P_NEG_DREPLY] = "NegDReply",
3690 [P_NEG_RS_DREPLY] = "NegRSDReply",
3691 [P_BARRIER_ACK] = "BarrierAck",
3692 [P_STATE_CHG_REQ] = "StateChgRequest",
3693 [P_STATE_CHG_REPLY] = "StateChgReply",
3694 [P_OV_REQUEST] = "OVRequest",
3695 [P_OV_REPLY] = "OVReply",
3696 [P_OV_RESULT] = "OVResult",
3697 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3698 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3699 [P_COMPRESSED_BITMAP] = "CBitmap",
3700 [P_DELAY_PROBE] = "DelayProbe",
3701 [P_OUT_OF_SYNC] = "OutOfSync",
3702 [P_RETRY_WRITE] = "RetryWrite",
3703 [P_RS_CANCEL] = "RSCancel",
3704 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3705 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3706 [P_RETRY_WRITE] = "retry_write",
3707 [P_PROTOCOL_UPDATE] = "protocol_update",
3709 /* enum drbd_packet, but not commands - obsoleted flags:
3715 /* too big for the array: 0xfffX */
3716 if (cmd == P_INITIAL_META)
3717 return "InitialMeta";
3718 if (cmd == P_INITIAL_DATA)
3719 return "InitialData";
3720 if (cmd == P_CONNECTION_FEATURES)
3721 return "ConnectionFeatures";
3722 if (cmd >= ARRAY_SIZE(cmdnames))
3724 return cmdnames[cmd];
3728 * drbd_wait_misc - wait for a request to make progress
3729 * @device: device associated with the request
3730 * @i: the struct drbd_interval embedded in struct drbd_request or
3731 * struct drbd_peer_request
3733 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3735 struct net_conf *nc;
3740 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3745 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3748 /* Indicate to wake up device->misc_wait on progress. */
3750 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3751 spin_unlock_irq(&device->resource->req_lock);
3752 timeout = schedule_timeout(timeout);
3753 finish_wait(&device->misc_wait, &wait);
3754 spin_lock_irq(&device->resource->req_lock);
3755 if (!timeout || device->state.conn < C_CONNECTED)
3757 if (signal_pending(current))
3758 return -ERESTARTSYS;
3762 #ifdef CONFIG_DRBD_FAULT_INJECTION
3763 /* Fault insertion support including random number generator shamelessly
3764 * stolen from kernel/rcutorture.c */
3765 struct fault_random_state {
3766 unsigned long state;
3767 unsigned long count;
3770 #define FAULT_RANDOM_MULT 39916801 /* prime */
3771 #define FAULT_RANDOM_ADD 479001701 /* prime */
3772 #define FAULT_RANDOM_REFRESH 10000
3775 * Crude but fast random-number generator. Uses a linear congruential
3776 * generator, with occasional help from get_random_bytes().
3778 static unsigned long
3779 _drbd_fault_random(struct fault_random_state *rsp)
3783 if (!rsp->count--) {
3784 get_random_bytes(&refresh, sizeof(refresh));
3785 rsp->state += refresh;
3786 rsp->count = FAULT_RANDOM_REFRESH;
3788 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3789 return swahw32(rsp->state);
3793 _drbd_fault_str(unsigned int type) {
3794 static char *_faults[] = {
3795 [DRBD_FAULT_MD_WR] = "Meta-data write",
3796 [DRBD_FAULT_MD_RD] = "Meta-data read",
3797 [DRBD_FAULT_RS_WR] = "Resync write",
3798 [DRBD_FAULT_RS_RD] = "Resync read",
3799 [DRBD_FAULT_DT_WR] = "Data write",
3800 [DRBD_FAULT_DT_RD] = "Data read",
3801 [DRBD_FAULT_DT_RA] = "Data read ahead",
3802 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3803 [DRBD_FAULT_AL_EE] = "EE allocation",
3804 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3807 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3811 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3813 static struct fault_random_state rrs = {0, 0};
3815 unsigned int ret = (
3817 ((1 << device_to_minor(device)) & fault_devs) != 0) &&
3818 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3823 if (__ratelimit(&drbd_ratelimit_state))
3824 drbd_warn(device, "***Simulating %s failure\n",
3825 _drbd_fault_str(type));
3832 const char *drbd_buildtag(void)
3834 /* DRBD built from external sources has here a reference to the
3835 git hash of the source code. */
3837 static char buildtag[38] = "\0uilt-in";
3839 if (buildtag[0] == 0) {
3841 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3850 module_init(drbd_init)
3851 module_exit(drbd_cleanup)
3853 EXPORT_SYMBOL(drbd_conn_str);
3854 EXPORT_SYMBOL(drbd_role_str);
3855 EXPORT_SYMBOL(drbd_disk_str);
3856 EXPORT_SYMBOL(drbd_set_st_err_str);