4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
11 from Logicworks, Inc. for making SDP replication support possible.
13 drbd is free software; you can redistribute it and/or modify
14 it under the terms of the GNU General Public License as published by
15 the Free Software Foundation; either version 2, or (at your option)
18 drbd is distributed in the hope that it will be useful,
19 but WITHOUT ANY WARRANTY; without even the implied warranty of
20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 GNU General Public License for more details.
23 You should have received a copy of the GNU General Public License
24 along with drbd; see the file COPYING. If not, write to
25 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/module.h>
32 #include <linux/jiffies.h>
33 #include <linux/drbd.h>
34 #include <linux/uaccess.h>
35 #include <asm/types.h>
37 #include <linux/ctype.h>
38 #include <linux/mutex.h>
40 #include <linux/file.h>
41 #include <linux/proc_fs.h>
42 #include <linux/init.h>
44 #include <linux/memcontrol.h>
45 #include <linux/mm_inline.h>
46 #include <linux/slab.h>
47 #include <linux/random.h>
48 #include <linux/reboot.h>
49 #include <linux/notifier.h>
50 #include <linux/kthread.h>
51 #include <linux/workqueue.h>
52 #define __KERNEL_SYSCALLS__
53 #include <linux/unistd.h>
54 #include <linux/vmalloc.h>
55 #include <linux/sched/signal.h>
57 #include <linux/drbd_limits.h>
59 #include "drbd_protocol.h"
60 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
62 #include "drbd_debugfs.h"
64 static DEFINE_MUTEX(drbd_main_mutex);
65 static int drbd_open(struct block_device *bdev, fmode_t mode);
66 static void drbd_release(struct gendisk *gd, fmode_t mode);
67 static void md_sync_timer_fn(unsigned long data);
68 static int w_bitmap_io(struct drbd_work *w, int unused);
70 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
71 "Lars Ellenberg <lars@linbit.com>");
72 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
73 MODULE_VERSION(REL_VERSION);
74 MODULE_LICENSE("GPL");
75 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
76 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
77 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
79 #include <linux/moduleparam.h>
80 /* thanks to these macros, if compiled into the kernel (not-module),
81 * these become boot parameters (e.g., drbd.minor_count) */
83 #ifdef CONFIG_DRBD_FAULT_INJECTION
84 int drbd_enable_faults;
86 static int drbd_fault_count;
87 static int drbd_fault_devs;
88 /* bitmap of enabled faults */
89 module_param_named(enable_faults, drbd_enable_faults, int, 0664);
90 /* fault rate % value - applies to all enabled faults */
91 module_param_named(fault_rate, drbd_fault_rate, int, 0664);
92 /* count of faults inserted */
93 module_param_named(fault_count, drbd_fault_count, int, 0664);
94 /* bitmap of devices to insert faults on */
95 module_param_named(fault_devs, drbd_fault_devs, int, 0644);
98 /* module parameters we can keep static */
99 static bool drbd_allow_oos; /* allow_open_on_secondary */
100 static bool drbd_disable_sendpage;
101 MODULE_PARM_DESC(allow_oos, "DONT USE!");
102 module_param_named(allow_oos, drbd_allow_oos, bool, 0);
103 module_param_named(disable_sendpage, drbd_disable_sendpage, bool, 0644);
105 /* module parameters we share */
106 int drbd_proc_details; /* Detail level in proc drbd*/
107 module_param_named(proc_details, drbd_proc_details, int, 0644);
108 /* module parameters shared with defaults */
109 unsigned int drbd_minor_count = DRBD_MINOR_COUNT_DEF;
110 /* Module parameter for setting the user mode helper program
111 * to run. Default is /sbin/drbdadm */
112 char drbd_usermode_helper[80] = "/sbin/drbdadm";
113 module_param_named(minor_count, drbd_minor_count, uint, 0444);
114 module_param_string(usermode_helper, drbd_usermode_helper, sizeof(drbd_usermode_helper), 0644);
116 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
117 * as member "struct gendisk *vdisk;"
119 struct idr drbd_devices;
120 struct list_head drbd_resources;
121 struct mutex resources_mutex;
123 struct kmem_cache *drbd_request_cache;
124 struct kmem_cache *drbd_ee_cache; /* peer requests */
125 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
126 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
127 mempool_t *drbd_request_mempool;
128 mempool_t *drbd_ee_mempool;
129 mempool_t *drbd_md_io_page_pool;
130 struct bio_set *drbd_md_io_bio_set;
131 struct bio_set *drbd_io_bio_set;
133 /* I do not use a standard mempool, because:
134 1) I want to hand out the pre-allocated objects first.
135 2) I want to be able to interrupt sleeping allocation with a signal.
136 Note: This is a single linked list, the next pointer is the private
137 member of struct page.
139 struct page *drbd_pp_pool;
140 spinlock_t drbd_pp_lock;
142 wait_queue_head_t drbd_pp_wait;
144 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
146 static const struct block_device_operations drbd_ops = {
147 .owner = THIS_MODULE,
149 .release = drbd_release,
152 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
156 if (!drbd_md_io_bio_set)
157 return bio_alloc(gfp_mask, 1);
159 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
166 /* When checking with sparse, and this is an inline function, sparse will
167 give tons of false positives. When this is a real functions sparse works.
169 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
173 atomic_inc(&device->local_cnt);
174 io_allowed = (device->state.disk >= mins);
176 if (atomic_dec_and_test(&device->local_cnt))
177 wake_up(&device->misc_wait);
185 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
186 * @connection: DRBD connection.
187 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
188 * @set_size: Expected number of requests before that barrier.
190 * In case the passed barrier_nr or set_size does not match the oldest
191 * epoch of not yet barrier-acked requests, this function will cause a
192 * termination of the connection.
194 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
195 unsigned int set_size)
197 struct drbd_request *r;
198 struct drbd_request *req = NULL;
199 int expect_epoch = 0;
202 spin_lock_irq(&connection->resource->req_lock);
204 /* find oldest not yet barrier-acked write request,
205 * count writes in its epoch. */
206 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
207 const unsigned s = r->rq_state;
211 if (!(s & RQ_NET_MASK))
216 expect_epoch = req->epoch;
219 if (r->epoch != expect_epoch)
223 /* if (s & RQ_DONE): not expected */
224 /* if (!(s & RQ_NET_MASK)): not expected */
229 /* first some paranoia code */
231 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
235 if (expect_epoch != barrier_nr) {
236 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
237 barrier_nr, expect_epoch);
241 if (expect_size != set_size) {
242 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
243 barrier_nr, set_size, expect_size);
247 /* Clean up list of requests processed during current epoch. */
248 /* this extra list walk restart is paranoia,
249 * to catch requests being barrier-acked "unexpectedly".
250 * It usually should find the same req again, or some READ preceding it. */
251 list_for_each_entry(req, &connection->transfer_log, tl_requests)
252 if (req->epoch == expect_epoch)
254 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
255 if (req->epoch != expect_epoch)
257 _req_mod(req, BARRIER_ACKED);
259 spin_unlock_irq(&connection->resource->req_lock);
264 spin_unlock_irq(&connection->resource->req_lock);
265 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
270 * _tl_restart() - Walks the transfer log, and applies an action to all requests
271 * @connection: DRBD connection to operate on.
272 * @what: The action/event to perform with all request objects
274 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
275 * RESTART_FROZEN_DISK_IO.
277 /* must hold resource->req_lock */
278 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
280 struct drbd_request *req, *r;
282 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
286 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
288 spin_lock_irq(&connection->resource->req_lock);
289 _tl_restart(connection, what);
290 spin_unlock_irq(&connection->resource->req_lock);
294 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
295 * @device: DRBD device.
297 * This is called after the connection to the peer was lost. The storage covered
298 * by the requests on the transfer gets marked as our of sync. Called from the
299 * receiver thread and the worker thread.
301 void tl_clear(struct drbd_connection *connection)
303 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
307 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
308 * @device: DRBD device.
310 void tl_abort_disk_io(struct drbd_device *device)
312 struct drbd_connection *connection = first_peer_device(device)->connection;
313 struct drbd_request *req, *r;
315 spin_lock_irq(&connection->resource->req_lock);
316 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
317 if (!(req->rq_state & RQ_LOCAL_PENDING))
319 if (req->device != device)
321 _req_mod(req, ABORT_DISK_IO);
323 spin_unlock_irq(&connection->resource->req_lock);
326 static int drbd_thread_setup(void *arg)
328 struct drbd_thread *thi = (struct drbd_thread *) arg;
329 struct drbd_resource *resource = thi->resource;
333 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
337 allow_kernel_signal(DRBD_SIGKILL);
338 allow_kernel_signal(SIGXCPU);
340 retval = thi->function(thi);
342 spin_lock_irqsave(&thi->t_lock, flags);
344 /* if the receiver has been "EXITING", the last thing it did
345 * was set the conn state to "StandAlone",
346 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
347 * and receiver thread will be "started".
348 * drbd_thread_start needs to set "RESTARTING" in that case.
349 * t_state check and assignment needs to be within the same spinlock,
350 * so either thread_start sees EXITING, and can remap to RESTARTING,
351 * or thread_start see NONE, and can proceed as normal.
354 if (thi->t_state == RESTARTING) {
355 drbd_info(resource, "Restarting %s thread\n", thi->name);
356 thi->t_state = RUNNING;
357 spin_unlock_irqrestore(&thi->t_lock, flags);
364 complete_all(&thi->stop);
365 spin_unlock_irqrestore(&thi->t_lock, flags);
367 drbd_info(resource, "Terminating %s\n", current->comm);
369 /* Release mod reference taken when thread was started */
372 kref_put(&thi->connection->kref, drbd_destroy_connection);
373 kref_put(&resource->kref, drbd_destroy_resource);
374 module_put(THIS_MODULE);
378 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
379 int (*func) (struct drbd_thread *), const char *name)
381 spin_lock_init(&thi->t_lock);
384 thi->function = func;
385 thi->resource = resource;
386 thi->connection = NULL;
390 int drbd_thread_start(struct drbd_thread *thi)
392 struct drbd_resource *resource = thi->resource;
393 struct task_struct *nt;
396 /* is used from state engine doing drbd_thread_stop_nowait,
397 * while holding the req lock irqsave */
398 spin_lock_irqsave(&thi->t_lock, flags);
400 switch (thi->t_state) {
402 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
403 thi->name, current->comm, current->pid);
405 /* Get ref on module for thread - this is released when thread exits */
406 if (!try_module_get(THIS_MODULE)) {
407 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
408 spin_unlock_irqrestore(&thi->t_lock, flags);
412 kref_get(&resource->kref);
414 kref_get(&thi->connection->kref);
416 init_completion(&thi->stop);
417 thi->reset_cpu_mask = 1;
418 thi->t_state = RUNNING;
419 spin_unlock_irqrestore(&thi->t_lock, flags);
420 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
422 nt = kthread_create(drbd_thread_setup, (void *) thi,
423 "drbd_%c_%s", thi->name[0], thi->resource->name);
426 drbd_err(resource, "Couldn't start thread\n");
429 kref_put(&thi->connection->kref, drbd_destroy_connection);
430 kref_put(&resource->kref, drbd_destroy_resource);
431 module_put(THIS_MODULE);
434 spin_lock_irqsave(&thi->t_lock, flags);
436 thi->t_state = RUNNING;
437 spin_unlock_irqrestore(&thi->t_lock, flags);
441 thi->t_state = RESTARTING;
442 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
443 thi->name, current->comm, current->pid);
448 spin_unlock_irqrestore(&thi->t_lock, flags);
456 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
460 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
462 /* may be called from state engine, holding the req lock irqsave */
463 spin_lock_irqsave(&thi->t_lock, flags);
465 if (thi->t_state == NONE) {
466 spin_unlock_irqrestore(&thi->t_lock, flags);
468 drbd_thread_start(thi);
472 if (thi->t_state != ns) {
473 if (thi->task == NULL) {
474 spin_unlock_irqrestore(&thi->t_lock, flags);
480 init_completion(&thi->stop);
481 if (thi->task != current)
482 force_sig(DRBD_SIGKILL, thi->task);
485 spin_unlock_irqrestore(&thi->t_lock, flags);
488 wait_for_completion(&thi->stop);
491 int conn_lowest_minor(struct drbd_connection *connection)
493 struct drbd_peer_device *peer_device;
494 int vnr = 0, minor = -1;
497 peer_device = idr_get_next(&connection->peer_devices, &vnr);
499 minor = device_to_minor(peer_device->device);
507 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
509 * Forces all threads of a resource onto the same CPU. This is beneficial for
510 * DRBD's performance. May be overwritten by user's configuration.
512 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
514 unsigned int *resources_per_cpu, min_index = ~0;
516 resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL);
517 if (resources_per_cpu) {
518 struct drbd_resource *resource;
519 unsigned int cpu, min = ~0;
522 for_each_resource_rcu(resource, &drbd_resources) {
523 for_each_cpu(cpu, resource->cpu_mask)
524 resources_per_cpu[cpu]++;
527 for_each_online_cpu(cpu) {
528 if (resources_per_cpu[cpu] < min) {
529 min = resources_per_cpu[cpu];
533 kfree(resources_per_cpu);
535 if (min_index == ~0) {
536 cpumask_setall(*cpu_mask);
539 cpumask_set_cpu(min_index, *cpu_mask);
543 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
544 * @device: DRBD device.
545 * @thi: drbd_thread object
547 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
550 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
552 struct drbd_resource *resource = thi->resource;
553 struct task_struct *p = current;
555 if (!thi->reset_cpu_mask)
557 thi->reset_cpu_mask = 0;
558 set_cpus_allowed_ptr(p, resource->cpu_mask);
561 #define drbd_calc_cpu_mask(A) ({})
565 * drbd_header_size - size of a packet header
567 * The header size is a multiple of 8, so any payload following the header is
568 * word aligned on 64-bit architectures. (The bitmap send and receive code
571 unsigned int drbd_header_size(struct drbd_connection *connection)
573 if (connection->agreed_pro_version >= 100) {
574 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
575 return sizeof(struct p_header100);
577 BUILD_BUG_ON(sizeof(struct p_header80) !=
578 sizeof(struct p_header95));
579 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
580 return sizeof(struct p_header80);
584 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
586 h->magic = cpu_to_be32(DRBD_MAGIC);
587 h->command = cpu_to_be16(cmd);
588 h->length = cpu_to_be16(size);
589 return sizeof(struct p_header80);
592 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
594 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
595 h->command = cpu_to_be16(cmd);
596 h->length = cpu_to_be32(size);
597 return sizeof(struct p_header95);
600 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
603 h->magic = cpu_to_be32(DRBD_MAGIC_100);
604 h->volume = cpu_to_be16(vnr);
605 h->command = cpu_to_be16(cmd);
606 h->length = cpu_to_be32(size);
608 return sizeof(struct p_header100);
611 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
612 void *buffer, enum drbd_packet cmd, int size)
614 if (connection->agreed_pro_version >= 100)
615 return prepare_header100(buffer, cmd, size, vnr);
616 else if (connection->agreed_pro_version >= 95 &&
617 size > DRBD_MAX_SIZE_H80_PACKET)
618 return prepare_header95(buffer, cmd, size);
620 return prepare_header80(buffer, cmd, size);
623 static void *__conn_prepare_command(struct drbd_connection *connection,
624 struct drbd_socket *sock)
628 return sock->sbuf + drbd_header_size(connection);
631 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
635 mutex_lock(&sock->mutex);
636 p = __conn_prepare_command(connection, sock);
638 mutex_unlock(&sock->mutex);
643 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
645 return conn_prepare_command(peer_device->connection, sock);
648 static int __send_command(struct drbd_connection *connection, int vnr,
649 struct drbd_socket *sock, enum drbd_packet cmd,
650 unsigned int header_size, void *data,
657 * Called with @data == NULL and the size of the data blocks in @size
658 * for commands that send data blocks. For those commands, omit the
659 * MSG_MORE flag: this will increase the likelihood that data blocks
660 * which are page aligned on the sender will end up page aligned on the
663 msg_flags = data ? MSG_MORE : 0;
665 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
667 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
670 err = drbd_send_all(connection, sock->socket, data, size, 0);
671 /* DRBD protocol "pings" are latency critical.
672 * This is supposed to trigger tcp_push_pending_frames() */
673 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
674 drbd_tcp_nodelay(sock->socket);
679 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
680 enum drbd_packet cmd, unsigned int header_size,
681 void *data, unsigned int size)
683 return __send_command(connection, 0, sock, cmd, header_size, data, size);
686 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
687 enum drbd_packet cmd, unsigned int header_size,
688 void *data, unsigned int size)
692 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
693 mutex_unlock(&sock->mutex);
697 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
698 enum drbd_packet cmd, unsigned int header_size,
699 void *data, unsigned int size)
703 err = __send_command(peer_device->connection, peer_device->device->vnr,
704 sock, cmd, header_size, data, size);
705 mutex_unlock(&sock->mutex);
709 int drbd_send_ping(struct drbd_connection *connection)
711 struct drbd_socket *sock;
713 sock = &connection->meta;
714 if (!conn_prepare_command(connection, sock))
716 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
719 int drbd_send_ping_ack(struct drbd_connection *connection)
721 struct drbd_socket *sock;
723 sock = &connection->meta;
724 if (!conn_prepare_command(connection, sock))
726 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
729 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
731 struct drbd_socket *sock;
732 struct p_rs_param_95 *p;
734 const int apv = peer_device->connection->agreed_pro_version;
735 enum drbd_packet cmd;
737 struct disk_conf *dc;
739 sock = &peer_device->connection->data;
740 p = drbd_prepare_command(peer_device, sock);
745 nc = rcu_dereference(peer_device->connection->net_conf);
747 size = apv <= 87 ? sizeof(struct p_rs_param)
748 : apv == 88 ? sizeof(struct p_rs_param)
749 + strlen(nc->verify_alg) + 1
750 : apv <= 94 ? sizeof(struct p_rs_param_89)
751 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
753 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
755 /* initialize verify_alg and csums_alg */
756 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
758 if (get_ldev(peer_device->device)) {
759 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
760 p->resync_rate = cpu_to_be32(dc->resync_rate);
761 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
762 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
763 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
764 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
765 put_ldev(peer_device->device);
767 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
768 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
769 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
770 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
771 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
775 strcpy(p->verify_alg, nc->verify_alg);
777 strcpy(p->csums_alg, nc->csums_alg);
780 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
783 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
785 struct drbd_socket *sock;
786 struct p_protocol *p;
790 sock = &connection->data;
791 p = __conn_prepare_command(connection, sock);
796 nc = rcu_dereference(connection->net_conf);
798 if (nc->tentative && connection->agreed_pro_version < 92) {
800 drbd_err(connection, "--dry-run is not supported by peer");
805 if (connection->agreed_pro_version >= 87)
806 size += strlen(nc->integrity_alg) + 1;
808 p->protocol = cpu_to_be32(nc->wire_protocol);
809 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
810 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
811 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
812 p->two_primaries = cpu_to_be32(nc->two_primaries);
814 if (nc->discard_my_data)
815 cf |= CF_DISCARD_MY_DATA;
818 p->conn_flags = cpu_to_be32(cf);
820 if (connection->agreed_pro_version >= 87)
821 strcpy(p->integrity_alg, nc->integrity_alg);
824 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
827 int drbd_send_protocol(struct drbd_connection *connection)
831 mutex_lock(&connection->data.mutex);
832 err = __drbd_send_protocol(connection, P_PROTOCOL);
833 mutex_unlock(&connection->data.mutex);
838 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
840 struct drbd_device *device = peer_device->device;
841 struct drbd_socket *sock;
845 if (!get_ldev_if_state(device, D_NEGOTIATING))
848 sock = &peer_device->connection->data;
849 p = drbd_prepare_command(peer_device, sock);
854 spin_lock_irq(&device->ldev->md.uuid_lock);
855 for (i = UI_CURRENT; i < UI_SIZE; i++)
856 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
857 spin_unlock_irq(&device->ldev->md.uuid_lock);
859 device->comm_bm_set = drbd_bm_total_weight(device);
860 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
862 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
864 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
865 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
866 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
869 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
872 int drbd_send_uuids(struct drbd_peer_device *peer_device)
874 return _drbd_send_uuids(peer_device, 0);
877 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
879 return _drbd_send_uuids(peer_device, 8);
882 void drbd_print_uuids(struct drbd_device *device, const char *text)
884 if (get_ldev_if_state(device, D_NEGOTIATING)) {
885 u64 *uuid = device->ldev->md.uuid;
886 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
888 (unsigned long long)uuid[UI_CURRENT],
889 (unsigned long long)uuid[UI_BITMAP],
890 (unsigned long long)uuid[UI_HISTORY_START],
891 (unsigned long long)uuid[UI_HISTORY_END]);
894 drbd_info(device, "%s effective data uuid: %016llX\n",
896 (unsigned long long)device->ed_uuid);
900 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
902 struct drbd_device *device = peer_device->device;
903 struct drbd_socket *sock;
907 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
909 uuid = device->ldev->md.uuid[UI_BITMAP];
910 if (uuid && uuid != UUID_JUST_CREATED)
911 uuid = uuid + UUID_NEW_BM_OFFSET;
913 get_random_bytes(&uuid, sizeof(u64));
914 drbd_uuid_set(device, UI_BITMAP, uuid);
915 drbd_print_uuids(device, "updated sync UUID");
916 drbd_md_sync(device);
918 sock = &peer_device->connection->data;
919 p = drbd_prepare_command(peer_device, sock);
921 p->uuid = cpu_to_be64(uuid);
922 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
926 /* communicated if (agreed_features & DRBD_FF_WSAME) */
928 assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p,
929 struct request_queue *q)
932 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
933 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
934 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
935 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
936 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
937 p->qlim->discard_enabled = blk_queue_discard(q);
938 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
940 q = device->rq_queue;
941 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
942 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
943 p->qlim->alignment_offset = 0;
944 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
945 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
946 p->qlim->discard_enabled = 0;
947 p->qlim->write_same_capable = 0;
951 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
953 struct drbd_device *device = peer_device->device;
954 struct drbd_socket *sock;
956 sector_t d_size, u_size;
958 unsigned int max_bio_size;
959 unsigned int packet_size;
961 sock = &peer_device->connection->data;
962 p = drbd_prepare_command(peer_device, sock);
966 packet_size = sizeof(*p);
967 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
968 packet_size += sizeof(p->qlim[0]);
970 memset(p, 0, packet_size);
971 if (get_ldev_if_state(device, D_NEGOTIATING)) {
972 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
973 d_size = drbd_get_max_capacity(device->ldev);
975 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
977 q_order_type = drbd_queue_order_type(device);
978 max_bio_size = queue_max_hw_sectors(q) << 9;
979 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
980 assign_p_sizes_qlim(device, p, q);
985 q_order_type = QUEUE_ORDERED_NONE;
986 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
987 assign_p_sizes_qlim(device, p, NULL);
990 if (peer_device->connection->agreed_pro_version <= 94)
991 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
992 else if (peer_device->connection->agreed_pro_version < 100)
993 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
995 p->d_size = cpu_to_be64(d_size);
996 p->u_size = cpu_to_be64(u_size);
997 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
998 p->max_bio_size = cpu_to_be32(max_bio_size);
999 p->queue_order_type = cpu_to_be16(q_order_type);
1000 p->dds_flags = cpu_to_be16(flags);
1002 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
1006 * drbd_send_current_state() - Sends the drbd state to the peer
1007 * @peer_device: DRBD peer device.
1009 int drbd_send_current_state(struct drbd_peer_device *peer_device)
1011 struct drbd_socket *sock;
1014 sock = &peer_device->connection->data;
1015 p = drbd_prepare_command(peer_device, sock);
1018 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1019 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1023 * drbd_send_state() - After a state change, sends the new state to the peer
1024 * @peer_device: DRBD peer device.
1025 * @state: the state to send, not necessarily the current state.
1027 * Each state change queues an "after_state_ch" work, which will eventually
1028 * send the resulting new state to the peer. If more state changes happen
1029 * between queuing and processing of the after_state_ch work, we still
1030 * want to send each intermediary state in the order it occurred.
1032 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1034 struct drbd_socket *sock;
1037 sock = &peer_device->connection->data;
1038 p = drbd_prepare_command(peer_device, sock);
1041 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1042 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1045 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1047 struct drbd_socket *sock;
1048 struct p_req_state *p;
1050 sock = &peer_device->connection->data;
1051 p = drbd_prepare_command(peer_device, sock);
1054 p->mask = cpu_to_be32(mask.i);
1055 p->val = cpu_to_be32(val.i);
1056 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1059 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1061 enum drbd_packet cmd;
1062 struct drbd_socket *sock;
1063 struct p_req_state *p;
1065 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1066 sock = &connection->data;
1067 p = conn_prepare_command(connection, sock);
1070 p->mask = cpu_to_be32(mask.i);
1071 p->val = cpu_to_be32(val.i);
1072 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1075 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1077 struct drbd_socket *sock;
1078 struct p_req_state_reply *p;
1080 sock = &peer_device->connection->meta;
1081 p = drbd_prepare_command(peer_device, sock);
1083 p->retcode = cpu_to_be32(retcode);
1084 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1088 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1090 struct drbd_socket *sock;
1091 struct p_req_state_reply *p;
1092 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1094 sock = &connection->meta;
1095 p = conn_prepare_command(connection, sock);
1097 p->retcode = cpu_to_be32(retcode);
1098 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1102 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1104 BUG_ON(code & ~0xf);
1105 p->encoding = (p->encoding & ~0xf) | code;
1108 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1110 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1113 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1116 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1119 static int fill_bitmap_rle_bits(struct drbd_device *device,
1120 struct p_compressed_bm *p,
1122 struct bm_xfer_ctx *c)
1124 struct bitstream bs;
1125 unsigned long plain_bits;
1132 /* may we use this feature? */
1134 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1136 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1139 if (c->bit_offset >= c->bm_bits)
1140 return 0; /* nothing to do. */
1142 /* use at most thus many bytes */
1143 bitstream_init(&bs, p->code, size, 0);
1144 memset(p->code, 0, size);
1145 /* plain bits covered in this code string */
1148 /* p->encoding & 0x80 stores whether the first run length is set.
1149 * bit offset is implicit.
1150 * start with toggle == 2 to be able to tell the first iteration */
1153 /* see how much plain bits we can stuff into one packet
1154 * using RLE and VLI. */
1156 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1157 : _drbd_bm_find_next(device, c->bit_offset);
1160 rl = tmp - c->bit_offset;
1162 if (toggle == 2) { /* first iteration */
1164 /* the first checked bit was set,
1165 * store start value, */
1166 dcbp_set_start(p, 1);
1167 /* but skip encoding of zero run length */
1171 dcbp_set_start(p, 0);
1174 /* paranoia: catch zero runlength.
1175 * can only happen if bitmap is modified while we scan it. */
1177 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1178 "t:%u bo:%lu\n", toggle, c->bit_offset);
1182 bits = vli_encode_bits(&bs, rl);
1183 if (bits == -ENOBUFS) /* buffer full */
1186 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1192 c->bit_offset = tmp;
1193 } while (c->bit_offset < c->bm_bits);
1195 len = bs.cur.b - p->code + !!bs.cur.bit;
1197 if (plain_bits < (len << 3)) {
1198 /* incompressible with this method.
1199 * we need to rewind both word and bit position. */
1200 c->bit_offset -= plain_bits;
1201 bm_xfer_ctx_bit_to_word_offset(c);
1202 c->bit_offset = c->word_offset * BITS_PER_LONG;
1206 /* RLE + VLI was able to compress it just fine.
1207 * update c->word_offset. */
1208 bm_xfer_ctx_bit_to_word_offset(c);
1210 /* store pad_bits */
1211 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1217 * send_bitmap_rle_or_plain
1219 * Return 0 when done, 1 when another iteration is needed, and a negative error
1220 * code upon failure.
1223 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1225 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1226 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1227 struct p_compressed_bm *p = sock->sbuf + header_size;
1230 len = fill_bitmap_rle_bits(device, p,
1231 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1236 dcbp_set_code(p, RLE_VLI_Bits);
1237 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1238 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1241 c->bytes[0] += header_size + sizeof(*p) + len;
1243 if (c->bit_offset >= c->bm_bits)
1246 /* was not compressible.
1247 * send a buffer full of plain text bits instead. */
1248 unsigned int data_size;
1249 unsigned long num_words;
1250 unsigned long *p = sock->sbuf + header_size;
1252 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1253 num_words = min_t(size_t, data_size / sizeof(*p),
1254 c->bm_words - c->word_offset);
1255 len = num_words * sizeof(*p);
1257 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1258 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1259 c->word_offset += num_words;
1260 c->bit_offset = c->word_offset * BITS_PER_LONG;
1263 c->bytes[1] += header_size + len;
1265 if (c->bit_offset > c->bm_bits)
1266 c->bit_offset = c->bm_bits;
1270 INFO_bm_xfer_stats(device, "send", c);
1278 /* See the comment at receive_bitmap() */
1279 static int _drbd_send_bitmap(struct drbd_device *device)
1281 struct bm_xfer_ctx c;
1284 if (!expect(device->bitmap))
1287 if (get_ldev(device)) {
1288 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1289 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1290 drbd_bm_set_all(device);
1291 if (drbd_bm_write(device)) {
1292 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1293 * but otherwise process as per normal - need to tell other
1294 * side that a full resync is required! */
1295 drbd_err(device, "Failed to write bitmap to disk!\n");
1297 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1298 drbd_md_sync(device);
1304 c = (struct bm_xfer_ctx) {
1305 .bm_bits = drbd_bm_bits(device),
1306 .bm_words = drbd_bm_words(device),
1310 err = send_bitmap_rle_or_plain(device, &c);
1316 int drbd_send_bitmap(struct drbd_device *device)
1318 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1321 mutex_lock(&sock->mutex);
1323 err = !_drbd_send_bitmap(device);
1324 mutex_unlock(&sock->mutex);
1328 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1330 struct drbd_socket *sock;
1331 struct p_barrier_ack *p;
1333 if (connection->cstate < C_WF_REPORT_PARAMS)
1336 sock = &connection->meta;
1337 p = conn_prepare_command(connection, sock);
1340 p->barrier = barrier_nr;
1341 p->set_size = cpu_to_be32(set_size);
1342 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1346 * _drbd_send_ack() - Sends an ack packet
1347 * @device: DRBD device.
1348 * @cmd: Packet command code.
1349 * @sector: sector, needs to be in big endian byte order
1350 * @blksize: size in byte, needs to be in big endian byte order
1351 * @block_id: Id, big endian byte order
1353 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1354 u64 sector, u32 blksize, u64 block_id)
1356 struct drbd_socket *sock;
1357 struct p_block_ack *p;
1359 if (peer_device->device->state.conn < C_CONNECTED)
1362 sock = &peer_device->connection->meta;
1363 p = drbd_prepare_command(peer_device, sock);
1367 p->block_id = block_id;
1368 p->blksize = blksize;
1369 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1370 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1373 /* dp->sector and dp->block_id already/still in network byte order,
1374 * data_size is payload size according to dp->head,
1375 * and may need to be corrected for digest size. */
1376 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1377 struct p_data *dp, int data_size)
1379 if (peer_device->connection->peer_integrity_tfm)
1380 data_size -= crypto_ahash_digestsize(peer_device->connection->peer_integrity_tfm);
1381 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1385 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1386 struct p_block_req *rp)
1388 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1392 * drbd_send_ack() - Sends an ack packet
1393 * @device: DRBD device
1394 * @cmd: packet command code
1395 * @peer_req: peer request
1397 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1398 struct drbd_peer_request *peer_req)
1400 return _drbd_send_ack(peer_device, cmd,
1401 cpu_to_be64(peer_req->i.sector),
1402 cpu_to_be32(peer_req->i.size),
1403 peer_req->block_id);
1406 /* This function misuses the block_id field to signal if the blocks
1407 * are is sync or not. */
1408 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1409 sector_t sector, int blksize, u64 block_id)
1411 return _drbd_send_ack(peer_device, cmd,
1412 cpu_to_be64(sector),
1413 cpu_to_be32(blksize),
1414 cpu_to_be64(block_id));
1417 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1418 struct drbd_peer_request *peer_req)
1420 struct drbd_socket *sock;
1421 struct p_block_desc *p;
1423 sock = &peer_device->connection->data;
1424 p = drbd_prepare_command(peer_device, sock);
1427 p->sector = cpu_to_be64(peer_req->i.sector);
1428 p->blksize = cpu_to_be32(peer_req->i.size);
1430 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1433 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1434 sector_t sector, int size, u64 block_id)
1436 struct drbd_socket *sock;
1437 struct p_block_req *p;
1439 sock = &peer_device->connection->data;
1440 p = drbd_prepare_command(peer_device, sock);
1443 p->sector = cpu_to_be64(sector);
1444 p->block_id = block_id;
1445 p->blksize = cpu_to_be32(size);
1446 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1449 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1450 void *digest, int digest_size, enum drbd_packet cmd)
1452 struct drbd_socket *sock;
1453 struct p_block_req *p;
1455 /* FIXME: Put the digest into the preallocated socket buffer. */
1457 sock = &peer_device->connection->data;
1458 p = drbd_prepare_command(peer_device, sock);
1461 p->sector = cpu_to_be64(sector);
1462 p->block_id = ID_SYNCER /* unused */;
1463 p->blksize = cpu_to_be32(size);
1464 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1467 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1469 struct drbd_socket *sock;
1470 struct p_block_req *p;
1472 sock = &peer_device->connection->data;
1473 p = drbd_prepare_command(peer_device, sock);
1476 p->sector = cpu_to_be64(sector);
1477 p->block_id = ID_SYNCER /* unused */;
1478 p->blksize = cpu_to_be32(size);
1479 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1482 /* called on sndtimeo
1483 * returns false if we should retry,
1484 * true if we think connection is dead
1486 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1489 /* long elapsed = (long)(jiffies - device->last_received); */
1491 drop_it = connection->meta.socket == sock
1492 || !connection->ack_receiver.task
1493 || get_t_state(&connection->ack_receiver) != RUNNING
1494 || connection->cstate < C_WF_REPORT_PARAMS;
1499 drop_it = !--connection->ko_count;
1501 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1502 current->comm, current->pid, connection->ko_count);
1503 request_ping(connection);
1506 return drop_it; /* && (device->state == R_PRIMARY) */;
1509 static void drbd_update_congested(struct drbd_connection *connection)
1511 struct sock *sk = connection->data.socket->sk;
1512 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1513 set_bit(NET_CONGESTED, &connection->flags);
1516 /* The idea of sendpage seems to be to put some kind of reference
1517 * to the page into the skb, and to hand it over to the NIC. In
1518 * this process get_page() gets called.
1520 * As soon as the page was really sent over the network put_page()
1521 * gets called by some part of the network layer. [ NIC driver? ]
1523 * [ get_page() / put_page() increment/decrement the count. If count
1524 * reaches 0 the page will be freed. ]
1526 * This works nicely with pages from FSs.
1527 * But this means that in protocol A we might signal IO completion too early!
1529 * In order not to corrupt data during a resync we must make sure
1530 * that we do not reuse our own buffer pages (EEs) to early, therefore
1531 * we have the net_ee list.
1533 * XFS seems to have problems, still, it submits pages with page_count == 0!
1534 * As a workaround, we disable sendpage on pages
1535 * with page_count == 0 or PageSlab.
1537 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1538 int offset, size_t size, unsigned msg_flags)
1540 struct socket *socket;
1544 socket = peer_device->connection->data.socket;
1545 addr = kmap(page) + offset;
1546 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1549 peer_device->device->send_cnt += size >> 9;
1553 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1554 int offset, size_t size, unsigned msg_flags)
1556 struct socket *socket = peer_device->connection->data.socket;
1560 /* e.g. XFS meta- & log-data is in slab pages, which have a
1561 * page_count of 0 and/or have PageSlab() set.
1562 * we cannot use send_page for those, as that does get_page();
1563 * put_page(); and would cause either a VM_BUG directly, or
1564 * __page_cache_release a page that would actually still be referenced
1565 * by someone, leading to some obscure delayed Oops somewhere else. */
1566 if (drbd_disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1567 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1569 msg_flags |= MSG_NOSIGNAL;
1570 drbd_update_congested(peer_device->connection);
1574 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1576 if (sent == -EAGAIN) {
1577 if (we_should_drop_the_connection(peer_device->connection, socket))
1581 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1582 __func__, (int)size, len, sent);
1589 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1590 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1594 peer_device->device->send_cnt += size >> 9;
1599 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1601 struct bio_vec bvec;
1602 struct bvec_iter iter;
1604 /* hint all but last page with MSG_MORE */
1605 bio_for_each_segment(bvec, bio, iter) {
1608 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1609 bvec.bv_offset, bvec.bv_len,
1610 bio_iter_last(bvec, iter)
1614 /* REQ_OP_WRITE_SAME has only one segment */
1615 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1621 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1623 struct bio_vec bvec;
1624 struct bvec_iter iter;
1626 /* hint all but last page with MSG_MORE */
1627 bio_for_each_segment(bvec, bio, iter) {
1630 err = _drbd_send_page(peer_device, bvec.bv_page,
1631 bvec.bv_offset, bvec.bv_len,
1632 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1635 /* REQ_OP_WRITE_SAME has only one segment */
1636 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1642 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1643 struct drbd_peer_request *peer_req)
1645 struct page *page = peer_req->pages;
1646 unsigned len = peer_req->i.size;
1649 /* hint all but last page with MSG_MORE */
1650 page_chain_for_each(page) {
1651 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1653 err = _drbd_send_page(peer_device, page, 0, l,
1654 page_chain_next(page) ? MSG_MORE : 0);
1662 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1665 if (connection->agreed_pro_version >= 95)
1666 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1667 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1668 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1669 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1670 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1671 (bio_op(bio) == REQ_OP_WRITE_ZEROES ? DP_DISCARD : 0);
1673 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1676 /* Used to send write or TRIM aka REQ_DISCARD requests
1677 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1679 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1681 struct drbd_device *device = peer_device->device;
1682 struct drbd_socket *sock;
1684 struct p_wsame *wsame = NULL;
1686 unsigned int dp_flags = 0;
1690 sock = &peer_device->connection->data;
1691 p = drbd_prepare_command(peer_device, sock);
1692 digest_size = peer_device->connection->integrity_tfm ?
1693 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1697 p->sector = cpu_to_be64(req->i.sector);
1698 p->block_id = (unsigned long)req;
1699 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1700 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1701 if (device->state.conn >= C_SYNC_SOURCE &&
1702 device->state.conn <= C_PAUSED_SYNC_T)
1703 dp_flags |= DP_MAY_SET_IN_SYNC;
1704 if (peer_device->connection->agreed_pro_version >= 100) {
1705 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1706 dp_flags |= DP_SEND_RECEIVE_ACK;
1707 /* During resync, request an explicit write ack,
1708 * even in protocol != C */
1709 if (req->rq_state & RQ_EXP_WRITE_ACK
1710 || (dp_flags & DP_MAY_SET_IN_SYNC))
1711 dp_flags |= DP_SEND_WRITE_ACK;
1713 p->dp_flags = cpu_to_be32(dp_flags);
1715 if (dp_flags & DP_DISCARD) {
1716 struct p_trim *t = (struct p_trim*)p;
1717 t->size = cpu_to_be32(req->i.size);
1718 err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1721 if (dp_flags & DP_WSAME) {
1722 /* this will only work if DRBD_FF_WSAME is set AND the
1723 * handshake agreed that all nodes and backend devices are
1724 * WRITE_SAME capable and agree on logical_block_size */
1725 wsame = (struct p_wsame*)p;
1726 digest_out = wsame + 1;
1727 wsame->size = cpu_to_be32(req->i.size);
1731 /* our digest is still only over the payload.
1732 * TRIM does not carry any payload. */
1734 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1737 __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1738 sizeof(*wsame) + digest_size, NULL,
1739 bio_iovec(req->master_bio).bv_len);
1742 __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1743 sizeof(*p) + digest_size, NULL, req->i.size);
1745 /* For protocol A, we have to memcpy the payload into
1746 * socket buffers, as we may complete right away
1747 * as soon as we handed it over to tcp, at which point the data
1748 * pages may become invalid.
1750 * For data-integrity enabled, we copy it as well, so we can be
1751 * sure that even if the bio pages may still be modified, it
1752 * won't change the data on the wire, thus if the digest checks
1753 * out ok after sending on this side, but does not fit on the
1754 * receiving side, we sure have detected corruption elsewhere.
1756 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1757 err = _drbd_send_bio(peer_device, req->master_bio);
1759 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1761 /* double check digest, sometimes buffers have been modified in flight. */
1762 if (digest_size > 0 && digest_size <= 64) {
1763 /* 64 byte, 512 bit, is the largest digest size
1764 * currently supported in kernel crypto. */
1765 unsigned char digest[64];
1766 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1767 if (memcmp(p + 1, digest, digest_size)) {
1769 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1770 (unsigned long long)req->i.sector, req->i.size);
1772 } /* else if (digest_size > 64) {
1773 ... Be noisy about digest too large ...
1777 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1782 /* answer packet, used to send data back for read requests:
1783 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1784 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1786 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1787 struct drbd_peer_request *peer_req)
1789 struct drbd_device *device = peer_device->device;
1790 struct drbd_socket *sock;
1795 sock = &peer_device->connection->data;
1796 p = drbd_prepare_command(peer_device, sock);
1798 digest_size = peer_device->connection->integrity_tfm ?
1799 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1803 p->sector = cpu_to_be64(peer_req->i.sector);
1804 p->block_id = peer_req->block_id;
1805 p->seq_num = 0; /* unused */
1808 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1809 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1811 err = _drbd_send_zc_ee(peer_device, peer_req);
1812 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1817 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1819 struct drbd_socket *sock;
1820 struct p_block_desc *p;
1822 sock = &peer_device->connection->data;
1823 p = drbd_prepare_command(peer_device, sock);
1826 p->sector = cpu_to_be64(req->i.sector);
1827 p->blksize = cpu_to_be32(req->i.size);
1828 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1832 drbd_send distinguishes two cases:
1834 Packets sent via the data socket "sock"
1835 and packets sent via the meta data socket "msock"
1838 -----------------+-------------------------+------------------------------
1839 timeout conf.timeout / 2 conf.timeout / 2
1840 timeout action send a ping via msock Abort communication
1841 and close all sockets
1845 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1847 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1848 void *buf, size_t size, unsigned msg_flags)
1850 struct kvec iov = {.iov_base = buf, .iov_len = size};
1857 /* THINK if (signal_pending) return ... ? */
1859 msg.msg_name = NULL;
1860 msg.msg_namelen = 0;
1861 msg.msg_control = NULL;
1862 msg.msg_controllen = 0;
1863 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1865 iov_iter_kvec(&msg.msg_iter, WRITE | ITER_KVEC, &iov, 1, size);
1867 if (sock == connection->data.socket) {
1869 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1871 drbd_update_congested(connection);
1874 rv = sock_sendmsg(sock, &msg);
1875 if (rv == -EAGAIN) {
1876 if (we_should_drop_the_connection(connection, sock))
1882 flush_signals(current);
1888 } while (sent < size);
1890 if (sock == connection->data.socket)
1891 clear_bit(NET_CONGESTED, &connection->flags);
1894 if (rv != -EAGAIN) {
1895 drbd_err(connection, "%s_sendmsg returned %d\n",
1896 sock == connection->meta.socket ? "msock" : "sock",
1898 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1900 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1907 * drbd_send_all - Send an entire buffer
1909 * Returns 0 upon success and a negative error value otherwise.
1911 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1912 size_t size, unsigned msg_flags)
1916 err = drbd_send(connection, sock, buffer, size, msg_flags);
1924 static int drbd_open(struct block_device *bdev, fmode_t mode)
1926 struct drbd_device *device = bdev->bd_disk->private_data;
1927 unsigned long flags;
1930 mutex_lock(&drbd_main_mutex);
1931 spin_lock_irqsave(&device->resource->req_lock, flags);
1932 /* to have a stable device->state.role
1933 * and no race with updating open_cnt */
1935 if (device->state.role != R_PRIMARY) {
1936 if (mode & FMODE_WRITE)
1938 else if (!drbd_allow_oos)
1944 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1945 mutex_unlock(&drbd_main_mutex);
1950 static void drbd_release(struct gendisk *gd, fmode_t mode)
1952 struct drbd_device *device = gd->private_data;
1953 mutex_lock(&drbd_main_mutex);
1955 mutex_unlock(&drbd_main_mutex);
1958 /* need to hold resource->req_lock */
1959 void drbd_queue_unplug(struct drbd_device *device)
1961 if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1962 D_ASSERT(device, device->state.role == R_PRIMARY);
1963 if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1964 drbd_queue_work_if_unqueued(
1965 &first_peer_device(device)->connection->sender_work,
1966 &device->unplug_work);
1971 static void drbd_set_defaults(struct drbd_device *device)
1973 /* Beware! The actual layout differs
1974 * between big endian and little endian */
1975 device->state = (union drbd_dev_state) {
1976 { .role = R_SECONDARY,
1978 .conn = C_STANDALONE,
1984 void drbd_init_set_defaults(struct drbd_device *device)
1986 /* the memset(,0,) did most of this.
1987 * note: only assignments, no allocation in here */
1989 drbd_set_defaults(device);
1991 atomic_set(&device->ap_bio_cnt, 0);
1992 atomic_set(&device->ap_actlog_cnt, 0);
1993 atomic_set(&device->ap_pending_cnt, 0);
1994 atomic_set(&device->rs_pending_cnt, 0);
1995 atomic_set(&device->unacked_cnt, 0);
1996 atomic_set(&device->local_cnt, 0);
1997 atomic_set(&device->pp_in_use_by_net, 0);
1998 atomic_set(&device->rs_sect_in, 0);
1999 atomic_set(&device->rs_sect_ev, 0);
2000 atomic_set(&device->ap_in_flight, 0);
2001 atomic_set(&device->md_io.in_use, 0);
2003 mutex_init(&device->own_state_mutex);
2004 device->state_mutex = &device->own_state_mutex;
2006 spin_lock_init(&device->al_lock);
2007 spin_lock_init(&device->peer_seq_lock);
2009 INIT_LIST_HEAD(&device->active_ee);
2010 INIT_LIST_HEAD(&device->sync_ee);
2011 INIT_LIST_HEAD(&device->done_ee);
2012 INIT_LIST_HEAD(&device->read_ee);
2013 INIT_LIST_HEAD(&device->net_ee);
2014 INIT_LIST_HEAD(&device->resync_reads);
2015 INIT_LIST_HEAD(&device->resync_work.list);
2016 INIT_LIST_HEAD(&device->unplug_work.list);
2017 INIT_LIST_HEAD(&device->bm_io_work.w.list);
2018 INIT_LIST_HEAD(&device->pending_master_completion[0]);
2019 INIT_LIST_HEAD(&device->pending_master_completion[1]);
2020 INIT_LIST_HEAD(&device->pending_completion[0]);
2021 INIT_LIST_HEAD(&device->pending_completion[1]);
2023 device->resync_work.cb = w_resync_timer;
2024 device->unplug_work.cb = w_send_write_hint;
2025 device->bm_io_work.w.cb = w_bitmap_io;
2027 setup_timer(&device->resync_timer, resync_timer_fn,
2028 (unsigned long)device);
2029 setup_timer(&device->md_sync_timer, md_sync_timer_fn,
2030 (unsigned long)device);
2031 setup_timer(&device->start_resync_timer, start_resync_timer_fn,
2032 (unsigned long)device);
2033 setup_timer(&device->request_timer, request_timer_fn,
2034 (unsigned long)device);
2036 init_waitqueue_head(&device->misc_wait);
2037 init_waitqueue_head(&device->state_wait);
2038 init_waitqueue_head(&device->ee_wait);
2039 init_waitqueue_head(&device->al_wait);
2040 init_waitqueue_head(&device->seq_wait);
2042 device->resync_wenr = LC_FREE;
2043 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2044 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2047 void drbd_device_cleanup(struct drbd_device *device)
2050 if (first_peer_device(device)->connection->receiver.t_state != NONE)
2051 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2052 first_peer_device(device)->connection->receiver.t_state);
2054 device->al_writ_cnt =
2055 device->bm_writ_cnt =
2063 device->rs_failed = 0;
2064 device->rs_last_events = 0;
2065 device->rs_last_sect_ev = 0;
2066 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2067 device->rs_mark_left[i] = 0;
2068 device->rs_mark_time[i] = 0;
2070 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2072 drbd_set_my_capacity(device, 0);
2073 if (device->bitmap) {
2074 /* maybe never allocated. */
2075 drbd_bm_resize(device, 0, 1);
2076 drbd_bm_cleanup(device);
2079 drbd_backing_dev_free(device, device->ldev);
2080 device->ldev = NULL;
2082 clear_bit(AL_SUSPENDED, &device->flags);
2084 D_ASSERT(device, list_empty(&device->active_ee));
2085 D_ASSERT(device, list_empty(&device->sync_ee));
2086 D_ASSERT(device, list_empty(&device->done_ee));
2087 D_ASSERT(device, list_empty(&device->read_ee));
2088 D_ASSERT(device, list_empty(&device->net_ee));
2089 D_ASSERT(device, list_empty(&device->resync_reads));
2090 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2091 D_ASSERT(device, list_empty(&device->resync_work.list));
2092 D_ASSERT(device, list_empty(&device->unplug_work.list));
2094 drbd_set_defaults(device);
2098 static void drbd_destroy_mempools(void)
2102 while (drbd_pp_pool) {
2103 page = drbd_pp_pool;
2104 drbd_pp_pool = (struct page *)page_private(page);
2109 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2111 if (drbd_io_bio_set)
2112 bioset_free(drbd_io_bio_set);
2113 if (drbd_md_io_bio_set)
2114 bioset_free(drbd_md_io_bio_set);
2115 if (drbd_md_io_page_pool)
2116 mempool_destroy(drbd_md_io_page_pool);
2117 if (drbd_ee_mempool)
2118 mempool_destroy(drbd_ee_mempool);
2119 if (drbd_request_mempool)
2120 mempool_destroy(drbd_request_mempool);
2122 kmem_cache_destroy(drbd_ee_cache);
2123 if (drbd_request_cache)
2124 kmem_cache_destroy(drbd_request_cache);
2125 if (drbd_bm_ext_cache)
2126 kmem_cache_destroy(drbd_bm_ext_cache);
2127 if (drbd_al_ext_cache)
2128 kmem_cache_destroy(drbd_al_ext_cache);
2130 drbd_io_bio_set = NULL;
2131 drbd_md_io_bio_set = NULL;
2132 drbd_md_io_page_pool = NULL;
2133 drbd_ee_mempool = NULL;
2134 drbd_request_mempool = NULL;
2135 drbd_ee_cache = NULL;
2136 drbd_request_cache = NULL;
2137 drbd_bm_ext_cache = NULL;
2138 drbd_al_ext_cache = NULL;
2143 static int drbd_create_mempools(void)
2146 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2149 /* prepare our caches and mempools */
2150 drbd_request_mempool = NULL;
2151 drbd_ee_cache = NULL;
2152 drbd_request_cache = NULL;
2153 drbd_bm_ext_cache = NULL;
2154 drbd_al_ext_cache = NULL;
2155 drbd_pp_pool = NULL;
2156 drbd_md_io_page_pool = NULL;
2157 drbd_md_io_bio_set = NULL;
2158 drbd_io_bio_set = NULL;
2161 drbd_request_cache = kmem_cache_create(
2162 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2163 if (drbd_request_cache == NULL)
2166 drbd_ee_cache = kmem_cache_create(
2167 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2168 if (drbd_ee_cache == NULL)
2171 drbd_bm_ext_cache = kmem_cache_create(
2172 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2173 if (drbd_bm_ext_cache == NULL)
2176 drbd_al_ext_cache = kmem_cache_create(
2177 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2178 if (drbd_al_ext_cache == NULL)
2182 drbd_io_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
2183 if (drbd_io_bio_set == NULL)
2186 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0,
2188 if (drbd_md_io_bio_set == NULL)
2191 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2192 if (drbd_md_io_page_pool == NULL)
2195 drbd_request_mempool = mempool_create_slab_pool(number,
2196 drbd_request_cache);
2197 if (drbd_request_mempool == NULL)
2200 drbd_ee_mempool = mempool_create_slab_pool(number, drbd_ee_cache);
2201 if (drbd_ee_mempool == NULL)
2204 /* drbd's page pool */
2205 spin_lock_init(&drbd_pp_lock);
2207 for (i = 0; i < number; i++) {
2208 page = alloc_page(GFP_HIGHUSER);
2211 set_page_private(page, (unsigned long)drbd_pp_pool);
2212 drbd_pp_pool = page;
2214 drbd_pp_vacant = number;
2219 drbd_destroy_mempools(); /* in case we allocated some */
2223 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2227 rr = drbd_free_peer_reqs(device, &device->active_ee);
2229 drbd_err(device, "%d EEs in active list found!\n", rr);
2231 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2233 drbd_err(device, "%d EEs in sync list found!\n", rr);
2235 rr = drbd_free_peer_reqs(device, &device->read_ee);
2237 drbd_err(device, "%d EEs in read list found!\n", rr);
2239 rr = drbd_free_peer_reqs(device, &device->done_ee);
2241 drbd_err(device, "%d EEs in done list found!\n", rr);
2243 rr = drbd_free_peer_reqs(device, &device->net_ee);
2245 drbd_err(device, "%d EEs in net list found!\n", rr);
2248 /* caution. no locking. */
2249 void drbd_destroy_device(struct kref *kref)
2251 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2252 struct drbd_resource *resource = device->resource;
2253 struct drbd_peer_device *peer_device, *tmp_peer_device;
2255 del_timer_sync(&device->request_timer);
2257 /* paranoia asserts */
2258 D_ASSERT(device, device->open_cnt == 0);
2259 /* end paranoia asserts */
2261 /* cleanup stuff that may have been allocated during
2262 * device (re-)configuration or state changes */
2264 if (device->this_bdev)
2265 bdput(device->this_bdev);
2267 drbd_backing_dev_free(device, device->ldev);
2268 device->ldev = NULL;
2270 drbd_release_all_peer_reqs(device);
2272 lc_destroy(device->act_log);
2273 lc_destroy(device->resync);
2275 kfree(device->p_uuid);
2276 /* device->p_uuid = NULL; */
2278 if (device->bitmap) /* should no longer be there. */
2279 drbd_bm_cleanup(device);
2280 __free_page(device->md_io.page);
2281 put_disk(device->vdisk);
2282 blk_cleanup_queue(device->rq_queue);
2283 kfree(device->rs_plan_s);
2285 /* not for_each_connection(connection, resource):
2286 * those may have been cleaned up and disassociated already.
2288 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2289 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2292 memset(device, 0xfd, sizeof(*device));
2294 kref_put(&resource->kref, drbd_destroy_resource);
2297 /* One global retry thread, if we need to push back some bio and have it
2298 * reinserted through our make request function.
2300 static struct retry_worker {
2301 struct workqueue_struct *wq;
2302 struct work_struct worker;
2305 struct list_head writes;
2308 static void do_retry(struct work_struct *ws)
2310 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2312 struct drbd_request *req, *tmp;
2314 spin_lock_irq(&retry->lock);
2315 list_splice_init(&retry->writes, &writes);
2316 spin_unlock_irq(&retry->lock);
2318 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2319 struct drbd_device *device = req->device;
2320 struct bio *bio = req->master_bio;
2321 unsigned long start_jif = req->start_jif;
2325 expect(atomic_read(&req->completion_ref) == 0) &&
2326 expect(req->rq_state & RQ_POSTPONED) &&
2327 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2328 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2331 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2332 req, atomic_read(&req->completion_ref),
2335 /* We still need to put one kref associated with the
2336 * "completion_ref" going zero in the code path that queued it
2337 * here. The request object may still be referenced by a
2338 * frozen local req->private_bio, in case we force-detached.
2340 kref_put(&req->kref, drbd_req_destroy);
2342 /* A single suspended or otherwise blocking device may stall
2343 * all others as well. Fortunately, this code path is to
2344 * recover from a situation that "should not happen":
2345 * concurrent writes in multi-primary setup.
2346 * In a "normal" lifecycle, this workqueue is supposed to be
2347 * destroyed without ever doing anything.
2348 * If it turns out to be an issue anyways, we can do per
2349 * resource (replication group) or per device (minor) retry
2350 * workqueues instead.
2353 /* We are not just doing generic_make_request(),
2354 * as we want to keep the start_time information. */
2356 __drbd_make_request(device, bio, start_jif);
2360 /* called via drbd_req_put_completion_ref(),
2361 * holds resource->req_lock */
2362 void drbd_restart_request(struct drbd_request *req)
2364 unsigned long flags;
2365 spin_lock_irqsave(&retry.lock, flags);
2366 list_move_tail(&req->tl_requests, &retry.writes);
2367 spin_unlock_irqrestore(&retry.lock, flags);
2369 /* Drop the extra reference that would otherwise
2370 * have been dropped by complete_master_bio.
2371 * do_retry() needs to grab a new one. */
2372 dec_ap_bio(req->device);
2374 queue_work(retry.wq, &retry.worker);
2377 void drbd_destroy_resource(struct kref *kref)
2379 struct drbd_resource *resource =
2380 container_of(kref, struct drbd_resource, kref);
2382 idr_destroy(&resource->devices);
2383 free_cpumask_var(resource->cpu_mask);
2384 kfree(resource->name);
2385 memset(resource, 0xf2, sizeof(*resource));
2389 void drbd_free_resource(struct drbd_resource *resource)
2391 struct drbd_connection *connection, *tmp;
2393 for_each_connection_safe(connection, tmp, resource) {
2394 list_del(&connection->connections);
2395 drbd_debugfs_connection_cleanup(connection);
2396 kref_put(&connection->kref, drbd_destroy_connection);
2398 drbd_debugfs_resource_cleanup(resource);
2399 kref_put(&resource->kref, drbd_destroy_resource);
2402 static void drbd_cleanup(void)
2405 struct drbd_device *device;
2406 struct drbd_resource *resource, *tmp;
2408 /* first remove proc,
2409 * drbdsetup uses it's presence to detect
2410 * whether DRBD is loaded.
2411 * If we would get stuck in proc removal,
2412 * but have netlink already deregistered,
2413 * some drbdsetup commands may wait forever
2417 remove_proc_entry("drbd", NULL);
2420 destroy_workqueue(retry.wq);
2422 drbd_genl_unregister();
2424 idr_for_each_entry(&drbd_devices, device, i)
2425 drbd_delete_device(device);
2427 /* not _rcu since, no other updater anymore. Genl already unregistered */
2428 for_each_resource_safe(resource, tmp, &drbd_resources) {
2429 list_del(&resource->resources);
2430 drbd_free_resource(resource);
2433 drbd_debugfs_cleanup();
2435 drbd_destroy_mempools();
2436 unregister_blkdev(DRBD_MAJOR, "drbd");
2438 idr_destroy(&drbd_devices);
2440 pr_info("module cleanup done.\n");
2444 * drbd_congested() - Callback for the flusher thread
2445 * @congested_data: User data
2446 * @bdi_bits: Bits the BDI flusher thread is currently interested in
2448 * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested.
2450 static int drbd_congested(void *congested_data, int bdi_bits)
2452 struct drbd_device *device = congested_data;
2453 struct request_queue *q;
2457 if (!may_inc_ap_bio(device)) {
2458 /* DRBD has frozen IO */
2464 if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
2465 r |= (1 << WB_async_congested);
2466 /* Without good local data, we would need to read from remote,
2467 * and that would need the worker thread as well, which is
2468 * currently blocked waiting for that usermode helper to
2471 if (!get_ldev_if_state(device, D_UP_TO_DATE))
2472 r |= (1 << WB_sync_congested);
2480 if (get_ldev(device)) {
2481 q = bdev_get_queue(device->ldev->backing_bdev);
2482 r = bdi_congested(q->backing_dev_info, bdi_bits);
2488 if (bdi_bits & (1 << WB_async_congested) &&
2489 test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
2490 r |= (1 << WB_async_congested);
2491 reason = reason == 'b' ? 'a' : 'n';
2495 device->congestion_reason = reason;
2499 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2501 spin_lock_init(&wq->q_lock);
2502 INIT_LIST_HEAD(&wq->q);
2503 init_waitqueue_head(&wq->q_wait);
2506 struct completion_work {
2508 struct completion done;
2511 static int w_complete(struct drbd_work *w, int cancel)
2513 struct completion_work *completion_work =
2514 container_of(w, struct completion_work, w);
2516 complete(&completion_work->done);
2520 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2522 struct completion_work completion_work;
2524 completion_work.w.cb = w_complete;
2525 init_completion(&completion_work.done);
2526 drbd_queue_work(work_queue, &completion_work.w);
2527 wait_for_completion(&completion_work.done);
2530 struct drbd_resource *drbd_find_resource(const char *name)
2532 struct drbd_resource *resource;
2534 if (!name || !name[0])
2538 for_each_resource_rcu(resource, &drbd_resources) {
2539 if (!strcmp(resource->name, name)) {
2540 kref_get(&resource->kref);
2550 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2551 void *peer_addr, int peer_addr_len)
2553 struct drbd_resource *resource;
2554 struct drbd_connection *connection;
2557 for_each_resource_rcu(resource, &drbd_resources) {
2558 for_each_connection_rcu(connection, resource) {
2559 if (connection->my_addr_len == my_addr_len &&
2560 connection->peer_addr_len == peer_addr_len &&
2561 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2562 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2563 kref_get(&connection->kref);
2574 static int drbd_alloc_socket(struct drbd_socket *socket)
2576 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2579 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2585 static void drbd_free_socket(struct drbd_socket *socket)
2587 free_page((unsigned long) socket->sbuf);
2588 free_page((unsigned long) socket->rbuf);
2591 void conn_free_crypto(struct drbd_connection *connection)
2593 drbd_free_sock(connection);
2595 crypto_free_ahash(connection->csums_tfm);
2596 crypto_free_ahash(connection->verify_tfm);
2597 crypto_free_shash(connection->cram_hmac_tfm);
2598 crypto_free_ahash(connection->integrity_tfm);
2599 crypto_free_ahash(connection->peer_integrity_tfm);
2600 kfree(connection->int_dig_in);
2601 kfree(connection->int_dig_vv);
2603 connection->csums_tfm = NULL;
2604 connection->verify_tfm = NULL;
2605 connection->cram_hmac_tfm = NULL;
2606 connection->integrity_tfm = NULL;
2607 connection->peer_integrity_tfm = NULL;
2608 connection->int_dig_in = NULL;
2609 connection->int_dig_vv = NULL;
2612 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2614 struct drbd_connection *connection;
2615 cpumask_var_t new_cpu_mask;
2618 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2621 /* silently ignore cpu mask on UP kernel */
2622 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2623 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2624 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2625 if (err == -EOVERFLOW) {
2626 /* So what. mask it out. */
2627 cpumask_var_t tmp_cpu_mask;
2628 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2629 cpumask_setall(tmp_cpu_mask);
2630 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2631 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2633 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2635 free_cpumask_var(tmp_cpu_mask);
2640 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2641 /* retcode = ERR_CPU_MASK_PARSE; */
2645 resource->res_opts = *res_opts;
2646 if (cpumask_empty(new_cpu_mask))
2647 drbd_calc_cpu_mask(&new_cpu_mask);
2648 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2649 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2650 for_each_connection_rcu(connection, resource) {
2651 connection->receiver.reset_cpu_mask = 1;
2652 connection->ack_receiver.reset_cpu_mask = 1;
2653 connection->worker.reset_cpu_mask = 1;
2659 free_cpumask_var(new_cpu_mask);
2664 struct drbd_resource *drbd_create_resource(const char *name)
2666 struct drbd_resource *resource;
2668 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2671 resource->name = kstrdup(name, GFP_KERNEL);
2672 if (!resource->name)
2673 goto fail_free_resource;
2674 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2675 goto fail_free_name;
2676 kref_init(&resource->kref);
2677 idr_init(&resource->devices);
2678 INIT_LIST_HEAD(&resource->connections);
2679 resource->write_ordering = WO_BDEV_FLUSH;
2680 list_add_tail_rcu(&resource->resources, &drbd_resources);
2681 mutex_init(&resource->conf_update);
2682 mutex_init(&resource->adm_mutex);
2683 spin_lock_init(&resource->req_lock);
2684 drbd_debugfs_resource_add(resource);
2688 kfree(resource->name);
2695 /* caller must be under adm_mutex */
2696 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2698 struct drbd_resource *resource;
2699 struct drbd_connection *connection;
2701 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2705 if (drbd_alloc_socket(&connection->data))
2707 if (drbd_alloc_socket(&connection->meta))
2710 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2711 if (!connection->current_epoch)
2714 INIT_LIST_HEAD(&connection->transfer_log);
2716 INIT_LIST_HEAD(&connection->current_epoch->list);
2717 connection->epochs = 1;
2718 spin_lock_init(&connection->epoch_lock);
2720 connection->send.seen_any_write_yet = false;
2721 connection->send.current_epoch_nr = 0;
2722 connection->send.current_epoch_writes = 0;
2724 resource = drbd_create_resource(name);
2728 connection->cstate = C_STANDALONE;
2729 mutex_init(&connection->cstate_mutex);
2730 init_waitqueue_head(&connection->ping_wait);
2731 idr_init(&connection->peer_devices);
2733 drbd_init_workqueue(&connection->sender_work);
2734 mutex_init(&connection->data.mutex);
2735 mutex_init(&connection->meta.mutex);
2737 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2738 connection->receiver.connection = connection;
2739 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2740 connection->worker.connection = connection;
2741 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2742 connection->ack_receiver.connection = connection;
2744 kref_init(&connection->kref);
2746 connection->resource = resource;
2748 if (set_resource_options(resource, res_opts))
2751 kref_get(&resource->kref);
2752 list_add_tail_rcu(&connection->connections, &resource->connections);
2753 drbd_debugfs_connection_add(connection);
2757 list_del(&resource->resources);
2758 drbd_free_resource(resource);
2760 kfree(connection->current_epoch);
2761 drbd_free_socket(&connection->meta);
2762 drbd_free_socket(&connection->data);
2767 void drbd_destroy_connection(struct kref *kref)
2769 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2770 struct drbd_resource *resource = connection->resource;
2772 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
2773 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2774 kfree(connection->current_epoch);
2776 idr_destroy(&connection->peer_devices);
2778 drbd_free_socket(&connection->meta);
2779 drbd_free_socket(&connection->data);
2780 kfree(connection->int_dig_in);
2781 kfree(connection->int_dig_vv);
2782 memset(connection, 0xfc, sizeof(*connection));
2784 kref_put(&resource->kref, drbd_destroy_resource);
2787 static int init_submitter(struct drbd_device *device)
2789 /* opencoded create_singlethread_workqueue(),
2790 * to be able to say "drbd%d", ..., minor */
2792 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2793 if (!device->submit.wq)
2796 INIT_WORK(&device->submit.worker, do_submit);
2797 INIT_LIST_HEAD(&device->submit.writes);
2801 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2803 struct drbd_resource *resource = adm_ctx->resource;
2804 struct drbd_connection *connection;
2805 struct drbd_device *device;
2806 struct drbd_peer_device *peer_device, *tmp_peer_device;
2807 struct gendisk *disk;
2808 struct request_queue *q;
2810 int vnr = adm_ctx->volume;
2811 enum drbd_ret_code err = ERR_NOMEM;
2813 device = minor_to_device(minor);
2815 return ERR_MINOR_OR_VOLUME_EXISTS;
2817 /* GFP_KERNEL, we are outside of all write-out paths */
2818 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2821 kref_init(&device->kref);
2823 kref_get(&resource->kref);
2824 device->resource = resource;
2825 device->minor = minor;
2828 drbd_init_set_defaults(device);
2830 q = blk_alloc_queue(GFP_KERNEL);
2833 device->rq_queue = q;
2834 q->queuedata = device;
2836 disk = alloc_disk(1);
2839 device->vdisk = disk;
2841 set_disk_ro(disk, true);
2844 disk->major = DRBD_MAJOR;
2845 disk->first_minor = minor;
2846 disk->fops = &drbd_ops;
2847 sprintf(disk->disk_name, "drbd%d", minor);
2848 disk->private_data = device;
2850 device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2851 /* we have no partitions. we contain only ourselves. */
2852 device->this_bdev->bd_contains = device->this_bdev;
2854 q->backing_dev_info->congested_fn = drbd_congested;
2855 q->backing_dev_info->congested_data = device;
2857 blk_queue_make_request(q, drbd_make_request);
2858 blk_queue_write_cache(q, true, true);
2859 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2860 This triggers a max_bio_size message upon first attach or connect */
2861 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2862 q->queue_lock = &resource->req_lock;
2864 device->md_io.page = alloc_page(GFP_KERNEL);
2865 if (!device->md_io.page)
2866 goto out_no_io_page;
2868 if (drbd_bm_init(device))
2870 device->read_requests = RB_ROOT;
2871 device->write_requests = RB_ROOT;
2873 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2876 err = ERR_MINOR_OR_VOLUME_EXISTS;
2877 goto out_no_minor_idr;
2879 kref_get(&device->kref);
2881 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2884 err = ERR_MINOR_OR_VOLUME_EXISTS;
2885 goto out_idr_remove_minor;
2887 kref_get(&device->kref);
2889 INIT_LIST_HEAD(&device->peer_devices);
2890 INIT_LIST_HEAD(&device->pending_bitmap_io);
2891 for_each_connection(connection, resource) {
2892 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2894 goto out_idr_remove_from_resource;
2895 peer_device->connection = connection;
2896 peer_device->device = device;
2898 list_add(&peer_device->peer_devices, &device->peer_devices);
2899 kref_get(&device->kref);
2901 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2904 err = ERR_INVALID_REQUEST;
2905 goto out_idr_remove_from_resource;
2907 kref_get(&connection->kref);
2908 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2911 if (init_submitter(device)) {
2913 goto out_idr_remove_vol;
2918 /* inherit the connection state */
2919 device->state.conn = first_connection(resource)->cstate;
2920 if (device->state.conn == C_WF_REPORT_PARAMS) {
2921 for_each_peer_device(peer_device, device)
2922 drbd_connected(peer_device);
2924 /* move to create_peer_device() */
2925 for_each_peer_device(peer_device, device)
2926 drbd_debugfs_peer_device_add(peer_device);
2927 drbd_debugfs_device_add(device);
2931 idr_remove(&connection->peer_devices, vnr);
2932 out_idr_remove_from_resource:
2933 for_each_connection(connection, resource) {
2934 peer_device = idr_remove(&connection->peer_devices, vnr);
2936 kref_put(&connection->kref, drbd_destroy_connection);
2938 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2939 list_del(&peer_device->peer_devices);
2942 idr_remove(&resource->devices, vnr);
2943 out_idr_remove_minor:
2944 idr_remove(&drbd_devices, minor);
2947 drbd_bm_cleanup(device);
2949 __free_page(device->md_io.page);
2953 blk_cleanup_queue(q);
2955 kref_put(&resource->kref, drbd_destroy_resource);
2960 void drbd_delete_device(struct drbd_device *device)
2962 struct drbd_resource *resource = device->resource;
2963 struct drbd_connection *connection;
2964 struct drbd_peer_device *peer_device;
2966 /* move to free_peer_device() */
2967 for_each_peer_device(peer_device, device)
2968 drbd_debugfs_peer_device_cleanup(peer_device);
2969 drbd_debugfs_device_cleanup(device);
2970 for_each_connection(connection, resource) {
2971 idr_remove(&connection->peer_devices, device->vnr);
2972 kref_put(&device->kref, drbd_destroy_device);
2974 idr_remove(&resource->devices, device->vnr);
2975 kref_put(&device->kref, drbd_destroy_device);
2976 idr_remove(&drbd_devices, device_to_minor(device));
2977 kref_put(&device->kref, drbd_destroy_device);
2978 del_gendisk(device->vdisk);
2980 kref_put(&device->kref, drbd_destroy_device);
2983 static int __init drbd_init(void)
2987 if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2988 pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2992 drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2996 err = register_blkdev(DRBD_MAJOR, "drbd");
2998 pr_err("unable to register block device major %d\n",
3004 * allocate all necessary structs
3006 init_waitqueue_head(&drbd_pp_wait);
3008 drbd_proc = NULL; /* play safe for drbd_cleanup */
3009 idr_init(&drbd_devices);
3011 mutex_init(&resources_mutex);
3012 INIT_LIST_HEAD(&drbd_resources);
3014 err = drbd_genl_register();
3016 pr_err("unable to register generic netlink family\n");
3020 err = drbd_create_mempools();
3025 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
3027 pr_err("unable to register proc file\n");
3031 retry.wq = create_singlethread_workqueue("drbd-reissue");
3033 pr_err("unable to create retry workqueue\n");
3036 INIT_WORK(&retry.worker, do_retry);
3037 spin_lock_init(&retry.lock);
3038 INIT_LIST_HEAD(&retry.writes);
3040 if (drbd_debugfs_init())
3041 pr_notice("failed to initialize debugfs -- will not be available\n");
3043 pr_info("initialized. "
3044 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
3045 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
3046 pr_info("%s\n", drbd_buildtag());
3047 pr_info("registered as block device major %d\n", DRBD_MAJOR);
3048 return 0; /* Success! */
3053 pr_err("ran out of memory\n");
3055 pr_err("initialization failure\n");
3059 static void drbd_free_one_sock(struct drbd_socket *ds)
3062 mutex_lock(&ds->mutex);
3065 mutex_unlock(&ds->mutex);
3067 /* so debugfs does not need to mutex_lock() */
3069 kernel_sock_shutdown(s, SHUT_RDWR);
3074 void drbd_free_sock(struct drbd_connection *connection)
3076 if (connection->data.socket)
3077 drbd_free_one_sock(&connection->data);
3078 if (connection->meta.socket)
3079 drbd_free_one_sock(&connection->meta);
3082 /* meta data management */
3084 void conn_md_sync(struct drbd_connection *connection)
3086 struct drbd_peer_device *peer_device;
3090 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3091 struct drbd_device *device = peer_device->device;
3093 kref_get(&device->kref);
3095 drbd_md_sync(device);
3096 kref_put(&device->kref, drbd_destroy_device);
3102 /* aligned 4kByte */
3103 struct meta_data_on_disk {
3104 u64 la_size_sect; /* last agreed size. */
3105 u64 uuid[UI_SIZE]; /* UUIDs. */
3108 u32 flags; /* MDF */
3111 u32 al_offset; /* offset to this block */
3112 u32 al_nr_extents; /* important for restoring the AL (userspace) */
3113 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3114 u32 bm_offset; /* offset to the bitmap, from here */
3115 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
3116 u32 la_peer_max_bio_size; /* last peer max_bio_size */
3118 /* see al_tr_number_to_on_disk_sector() */
3120 u32 al_stripe_size_4k;
3122 u8 reserved_u8[4096 - (7*8 + 10*4)];
3127 void drbd_md_write(struct drbd_device *device, void *b)
3129 struct meta_data_on_disk *buffer = b;
3133 memset(buffer, 0, sizeof(*buffer));
3135 buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3136 for (i = UI_CURRENT; i < UI_SIZE; i++)
3137 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3138 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3139 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3141 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3142 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3143 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3144 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3145 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3147 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3148 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3150 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3151 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3153 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3154 sector = device->ldev->md.md_offset;
3156 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3157 /* this was a try anyways ... */
3158 drbd_err(device, "meta data update failed!\n");
3159 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3164 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3165 * @device: DRBD device.
3167 void drbd_md_sync(struct drbd_device *device)
3169 struct meta_data_on_disk *buffer;
3171 /* Don't accidentally change the DRBD meta data layout. */
3172 BUILD_BUG_ON(UI_SIZE != 4);
3173 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3175 del_timer(&device->md_sync_timer);
3176 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3177 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3180 /* We use here D_FAILED and not D_ATTACHING because we try to write
3181 * metadata even if we detach due to a disk failure! */
3182 if (!get_ldev_if_state(device, D_FAILED))
3185 buffer = drbd_md_get_buffer(device, __func__);
3189 drbd_md_write(device, buffer);
3191 /* Update device->ldev->md.la_size_sect,
3192 * since we updated it on metadata. */
3193 device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3195 drbd_md_put_buffer(device);
3200 static int check_activity_log_stripe_size(struct drbd_device *device,
3201 struct meta_data_on_disk *on_disk,
3202 struct drbd_md *in_core)
3204 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3205 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3208 /* both not set: default to old fixed size activity log */
3209 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3211 al_stripe_size_4k = MD_32kB_SECT/8;
3214 /* some paranoia plausibility checks */
3216 /* we need both values to be set */
3217 if (al_stripes == 0 || al_stripe_size_4k == 0)
3220 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3222 /* Upper limit of activity log area, to avoid potential overflow
3223 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3224 * than 72 * 4k blocks total only increases the amount of history,
3225 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3226 if (al_size_4k > (16 * 1024 * 1024/4))
3229 /* Lower limit: we need at least 8 transaction slots (32kB)
3230 * to not break existing setups */
3231 if (al_size_4k < MD_32kB_SECT/8)
3234 in_core->al_stripe_size_4k = al_stripe_size_4k;
3235 in_core->al_stripes = al_stripes;
3236 in_core->al_size_4k = al_size_4k;
3240 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3241 al_stripes, al_stripe_size_4k);
3245 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3247 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3248 struct drbd_md *in_core = &bdev->md;
3249 s32 on_disk_al_sect;
3250 s32 on_disk_bm_sect;
3252 /* The on-disk size of the activity log, calculated from offsets, and
3253 * the size of the activity log calculated from the stripe settings,
3255 * Though we could relax this a bit: it is ok, if the striped activity log
3256 * fits in the available on-disk activity log size.
3257 * Right now, that would break how resize is implemented.
3258 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3259 * of possible unused padding space in the on disk layout. */
3260 if (in_core->al_offset < 0) {
3261 if (in_core->bm_offset > in_core->al_offset)
3263 on_disk_al_sect = -in_core->al_offset;
3264 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3266 if (in_core->al_offset != MD_4kB_SECT)
3268 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3271 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3272 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3275 /* old fixed size meta data is exactly that: fixed. */
3276 if (in_core->meta_dev_idx >= 0) {
3277 if (in_core->md_size_sect != MD_128MB_SECT
3278 || in_core->al_offset != MD_4kB_SECT
3279 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3280 || in_core->al_stripes != 1
3281 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3285 if (capacity < in_core->md_size_sect)
3287 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3290 /* should be aligned, and at least 32k */
3291 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3294 /* should fit (for now: exactly) into the available on-disk space;
3295 * overflow prevention is in check_activity_log_stripe_size() above. */
3296 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3299 /* again, should be aligned */
3300 if (in_core->bm_offset & 7)
3303 /* FIXME check for device grow with flex external meta data? */
3305 /* can the available bitmap space cover the last agreed device size? */
3306 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3312 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3313 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3314 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3315 in_core->meta_dev_idx,
3316 in_core->al_stripes, in_core->al_stripe_size_4k,
3317 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3318 (unsigned long long)in_core->la_size_sect,
3319 (unsigned long long)capacity);
3326 * drbd_md_read() - Reads in the meta data super block
3327 * @device: DRBD device.
3328 * @bdev: Device from which the meta data should be read in.
3330 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3331 * something goes wrong.
3333 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3334 * even before @bdev is assigned to @device->ldev.
3336 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3338 struct meta_data_on_disk *buffer;
3340 int i, rv = NO_ERROR;
3342 if (device->state.disk != D_DISKLESS)
3343 return ERR_DISK_CONFIGURED;
3345 buffer = drbd_md_get_buffer(device, __func__);
3349 /* First, figure out where our meta data superblock is located,
3351 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3352 bdev->md.md_offset = drbd_md_ss(bdev);
3353 /* Even for (flexible or indexed) external meta data,
3354 * initially restrict us to the 4k superblock for now.
3355 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3356 bdev->md.md_size_sect = 8;
3358 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3360 /* NOTE: can't do normal error processing here as this is
3361 called BEFORE disk is attached */
3362 drbd_err(device, "Error while reading metadata.\n");
3363 rv = ERR_IO_MD_DISK;
3367 magic = be32_to_cpu(buffer->magic);
3368 flags = be32_to_cpu(buffer->flags);
3369 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3370 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3371 /* btw: that's Activity Log clean, not "all" clean. */
3372 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3373 rv = ERR_MD_UNCLEAN;
3377 rv = ERR_MD_INVALID;
3378 if (magic != DRBD_MD_MAGIC_08) {
3379 if (magic == DRBD_MD_MAGIC_07)
3380 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3382 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3386 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3387 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3388 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3393 /* convert to in_core endian */
3394 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3395 for (i = UI_CURRENT; i < UI_SIZE; i++)
3396 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3397 bdev->md.flags = be32_to_cpu(buffer->flags);
3398 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3400 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3401 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3402 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3404 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3406 if (check_offsets_and_sizes(device, bdev))
3409 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3410 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3411 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3414 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3415 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3416 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3422 spin_lock_irq(&device->resource->req_lock);
3423 if (device->state.conn < C_CONNECTED) {
3425 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3426 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3427 device->peer_max_bio_size = peer;
3429 spin_unlock_irq(&device->resource->req_lock);
3432 drbd_md_put_buffer(device);
3438 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3439 * @device: DRBD device.
3441 * Call this function if you change anything that should be written to
3442 * the meta-data super block. This function sets MD_DIRTY, and starts a
3443 * timer that ensures that within five seconds you have to call drbd_md_sync().
3446 void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
3448 if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3449 mod_timer(&device->md_sync_timer, jiffies + HZ);
3450 device->last_md_mark_dirty.line = line;
3451 device->last_md_mark_dirty.func = func;
3455 void drbd_md_mark_dirty(struct drbd_device *device)
3457 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3458 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3462 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3466 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3467 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3470 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3472 if (idx == UI_CURRENT) {
3473 if (device->state.role == R_PRIMARY)
3478 drbd_set_ed_uuid(device, val);
3481 device->ldev->md.uuid[idx] = val;
3482 drbd_md_mark_dirty(device);
3485 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3487 unsigned long flags;
3488 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3489 __drbd_uuid_set(device, idx, val);
3490 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3493 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3495 unsigned long flags;
3496 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3497 if (device->ldev->md.uuid[idx]) {
3498 drbd_uuid_move_history(device);
3499 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3501 __drbd_uuid_set(device, idx, val);
3502 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3506 * drbd_uuid_new_current() - Creates a new current UUID
3507 * @device: DRBD device.
3509 * Creates a new current UUID, and rotates the old current UUID into
3510 * the bitmap slot. Causes an incremental resync upon next connect.
3512 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3515 unsigned long long bm_uuid;
3517 get_random_bytes(&val, sizeof(u64));
3519 spin_lock_irq(&device->ldev->md.uuid_lock);
3520 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3523 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3525 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3526 __drbd_uuid_set(device, UI_CURRENT, val);
3527 spin_unlock_irq(&device->ldev->md.uuid_lock);
3529 drbd_print_uuids(device, "new current UUID");
3530 /* get it to stable storage _now_ */
3531 drbd_md_sync(device);
3534 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3536 unsigned long flags;
3537 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3540 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3542 drbd_uuid_move_history(device);
3543 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3544 device->ldev->md.uuid[UI_BITMAP] = 0;
3546 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3548 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3550 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3552 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3554 drbd_md_mark_dirty(device);
3558 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3559 * @device: DRBD device.
3561 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3563 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3567 drbd_md_set_flag(device, MDF_FULL_SYNC);
3568 drbd_md_sync(device);
3569 drbd_bm_set_all(device);
3571 rv = drbd_bm_write(device);
3574 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3575 drbd_md_sync(device);
3582 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3583 * @device: DRBD device.
3585 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3587 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3589 drbd_resume_al(device);
3590 drbd_bm_clear_all(device);
3591 return drbd_bm_write(device);
3594 static int w_bitmap_io(struct drbd_work *w, int unused)
3596 struct drbd_device *device =
3597 container_of(w, struct drbd_device, bm_io_work.w);
3598 struct bm_io_work *work = &device->bm_io_work;
3601 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3602 int cnt = atomic_read(&device->ap_bio_cnt);
3604 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3608 if (get_ldev(device)) {
3609 drbd_bm_lock(device, work->why, work->flags);
3610 rv = work->io_fn(device);
3611 drbd_bm_unlock(device);
3615 clear_bit_unlock(BITMAP_IO, &device->flags);
3616 wake_up(&device->misc_wait);
3619 work->done(device, rv);
3621 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3629 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3630 * @device: DRBD device.
3631 * @io_fn: IO callback to be called when bitmap IO is possible
3632 * @done: callback to be called after the bitmap IO was performed
3633 * @why: Descriptive text of the reason for doing the IO
3635 * While IO on the bitmap happens we freeze application IO thus we ensure
3636 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3637 * called from worker context. It MUST NOT be used while a previous such
3638 * work is still pending!
3640 * Its worker function encloses the call of io_fn() by get_ldev() and
3643 void drbd_queue_bitmap_io(struct drbd_device *device,
3644 int (*io_fn)(struct drbd_device *),
3645 void (*done)(struct drbd_device *, int),
3646 char *why, enum bm_flag flags)
3648 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3650 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3651 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3652 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3653 if (device->bm_io_work.why)
3654 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3655 why, device->bm_io_work.why);
3657 device->bm_io_work.io_fn = io_fn;
3658 device->bm_io_work.done = done;
3659 device->bm_io_work.why = why;
3660 device->bm_io_work.flags = flags;
3662 spin_lock_irq(&device->resource->req_lock);
3663 set_bit(BITMAP_IO, &device->flags);
3664 /* don't wait for pending application IO if the caller indicates that
3665 * application IO does not conflict anyways. */
3666 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3667 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3668 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3669 &device->bm_io_work.w);
3671 spin_unlock_irq(&device->resource->req_lock);
3675 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3676 * @device: DRBD device.
3677 * @io_fn: IO callback to be called when bitmap IO is possible
3678 * @why: Descriptive text of the reason for doing the IO
3680 * freezes application IO while that the actual IO operations runs. This
3681 * functions MAY NOT be called from worker context.
3683 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3684 char *why, enum bm_flag flags)
3686 /* Only suspend io, if some operation is supposed to be locked out */
3687 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3690 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3693 drbd_suspend_io(device);
3695 drbd_bm_lock(device, why, flags);
3697 drbd_bm_unlock(device);
3700 drbd_resume_io(device);
3705 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3707 if ((device->ldev->md.flags & flag) != flag) {
3708 drbd_md_mark_dirty(device);
3709 device->ldev->md.flags |= flag;
3713 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3715 if ((device->ldev->md.flags & flag) != 0) {
3716 drbd_md_mark_dirty(device);
3717 device->ldev->md.flags &= ~flag;
3720 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3722 return (bdev->md.flags & flag) != 0;
3725 static void md_sync_timer_fn(unsigned long data)
3727 struct drbd_device *device = (struct drbd_device *) data;
3728 drbd_device_post_work(device, MD_SYNC);
3731 const char *cmdname(enum drbd_packet cmd)
3733 /* THINK may need to become several global tables
3734 * when we want to support more than
3735 * one PRO_VERSION */
3736 static const char *cmdnames[] = {
3738 [P_WSAME] = "WriteSame",
3740 [P_DATA_REPLY] = "DataReply",
3741 [P_RS_DATA_REPLY] = "RSDataReply",
3742 [P_BARRIER] = "Barrier",
3743 [P_BITMAP] = "ReportBitMap",
3744 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3745 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3746 [P_UNPLUG_REMOTE] = "UnplugRemote",
3747 [P_DATA_REQUEST] = "DataRequest",
3748 [P_RS_DATA_REQUEST] = "RSDataRequest",
3749 [P_SYNC_PARAM] = "SyncParam",
3750 [P_SYNC_PARAM89] = "SyncParam89",
3751 [P_PROTOCOL] = "ReportProtocol",
3752 [P_UUIDS] = "ReportUUIDs",
3753 [P_SIZES] = "ReportSizes",
3754 [P_STATE] = "ReportState",
3755 [P_SYNC_UUID] = "ReportSyncUUID",
3756 [P_AUTH_CHALLENGE] = "AuthChallenge",
3757 [P_AUTH_RESPONSE] = "AuthResponse",
3759 [P_PING_ACK] = "PingAck",
3760 [P_RECV_ACK] = "RecvAck",
3761 [P_WRITE_ACK] = "WriteAck",
3762 [P_RS_WRITE_ACK] = "RSWriteAck",
3763 [P_SUPERSEDED] = "Superseded",
3764 [P_NEG_ACK] = "NegAck",
3765 [P_NEG_DREPLY] = "NegDReply",
3766 [P_NEG_RS_DREPLY] = "NegRSDReply",
3767 [P_BARRIER_ACK] = "BarrierAck",
3768 [P_STATE_CHG_REQ] = "StateChgRequest",
3769 [P_STATE_CHG_REPLY] = "StateChgReply",
3770 [P_OV_REQUEST] = "OVRequest",
3771 [P_OV_REPLY] = "OVReply",
3772 [P_OV_RESULT] = "OVResult",
3773 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3774 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3775 [P_COMPRESSED_BITMAP] = "CBitmap",
3776 [P_DELAY_PROBE] = "DelayProbe",
3777 [P_OUT_OF_SYNC] = "OutOfSync",
3778 [P_RETRY_WRITE] = "RetryWrite",
3779 [P_RS_CANCEL] = "RSCancel",
3780 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3781 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3782 [P_RETRY_WRITE] = "retry_write",
3783 [P_PROTOCOL_UPDATE] = "protocol_update",
3784 [P_RS_THIN_REQ] = "rs_thin_req",
3785 [P_RS_DEALLOCATED] = "rs_deallocated",
3787 /* enum drbd_packet, but not commands - obsoleted flags:
3793 /* too big for the array: 0xfffX */
3794 if (cmd == P_INITIAL_META)
3795 return "InitialMeta";
3796 if (cmd == P_INITIAL_DATA)
3797 return "InitialData";
3798 if (cmd == P_CONNECTION_FEATURES)
3799 return "ConnectionFeatures";
3800 if (cmd >= ARRAY_SIZE(cmdnames))
3802 return cmdnames[cmd];
3806 * drbd_wait_misc - wait for a request to make progress
3807 * @device: device associated with the request
3808 * @i: the struct drbd_interval embedded in struct drbd_request or
3809 * struct drbd_peer_request
3811 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3813 struct net_conf *nc;
3818 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3823 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3826 /* Indicate to wake up device->misc_wait on progress. */
3828 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3829 spin_unlock_irq(&device->resource->req_lock);
3830 timeout = schedule_timeout(timeout);
3831 finish_wait(&device->misc_wait, &wait);
3832 spin_lock_irq(&device->resource->req_lock);
3833 if (!timeout || device->state.conn < C_CONNECTED)
3835 if (signal_pending(current))
3836 return -ERESTARTSYS;
3840 void lock_all_resources(void)
3842 struct drbd_resource *resource;
3843 int __maybe_unused i = 0;
3845 mutex_lock(&resources_mutex);
3846 local_irq_disable();
3847 for_each_resource(resource, &drbd_resources)
3848 spin_lock_nested(&resource->req_lock, i++);
3851 void unlock_all_resources(void)
3853 struct drbd_resource *resource;
3855 for_each_resource(resource, &drbd_resources)
3856 spin_unlock(&resource->req_lock);
3858 mutex_unlock(&resources_mutex);
3861 #ifdef CONFIG_DRBD_FAULT_INJECTION
3862 /* Fault insertion support including random number generator shamelessly
3863 * stolen from kernel/rcutorture.c */
3864 struct fault_random_state {
3865 unsigned long state;
3866 unsigned long count;
3869 #define FAULT_RANDOM_MULT 39916801 /* prime */
3870 #define FAULT_RANDOM_ADD 479001701 /* prime */
3871 #define FAULT_RANDOM_REFRESH 10000
3874 * Crude but fast random-number generator. Uses a linear congruential
3875 * generator, with occasional help from get_random_bytes().
3877 static unsigned long
3878 _drbd_fault_random(struct fault_random_state *rsp)
3882 if (!rsp->count--) {
3883 get_random_bytes(&refresh, sizeof(refresh));
3884 rsp->state += refresh;
3885 rsp->count = FAULT_RANDOM_REFRESH;
3887 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3888 return swahw32(rsp->state);
3892 _drbd_fault_str(unsigned int type) {
3893 static char *_faults[] = {
3894 [DRBD_FAULT_MD_WR] = "Meta-data write",
3895 [DRBD_FAULT_MD_RD] = "Meta-data read",
3896 [DRBD_FAULT_RS_WR] = "Resync write",
3897 [DRBD_FAULT_RS_RD] = "Resync read",
3898 [DRBD_FAULT_DT_WR] = "Data write",
3899 [DRBD_FAULT_DT_RD] = "Data read",
3900 [DRBD_FAULT_DT_RA] = "Data read ahead",
3901 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3902 [DRBD_FAULT_AL_EE] = "EE allocation",
3903 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3906 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3910 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3912 static struct fault_random_state rrs = {0, 0};
3914 unsigned int ret = (
3915 (drbd_fault_devs == 0 ||
3916 ((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3917 (((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3922 if (__ratelimit(&drbd_ratelimit_state))
3923 drbd_warn(device, "***Simulating %s failure\n",
3924 _drbd_fault_str(type));
3931 const char *drbd_buildtag(void)
3933 /* DRBD built from external sources has here a reference to the
3934 git hash of the source code. */
3936 static char buildtag[38] = "\0uilt-in";
3938 if (buildtag[0] == 0) {
3940 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3949 module_init(drbd_init)
3950 module_exit(drbd_cleanup)
3952 EXPORT_SYMBOL(drbd_conn_str);
3953 EXPORT_SYMBOL(drbd_role_str);
3954 EXPORT_SYMBOL(drbd_disk_str);
3955 EXPORT_SYMBOL(drbd_set_st_err_str);