1 // SPDX-License-Identifier: GPL-2.0-or-later
5 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
7 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
8 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
9 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
11 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
12 from Logicworks, Inc. for making SDP replication support possible.
17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 #include <linux/module.h>
20 #include <linux/jiffies.h>
21 #include <linux/drbd.h>
22 #include <linux/uaccess.h>
23 #include <asm/types.h>
25 #include <linux/ctype.h>
26 #include <linux/mutex.h>
28 #include <linux/file.h>
29 #include <linux/proc_fs.h>
30 #include <linux/init.h>
32 #include <linux/memcontrol.h>
33 #include <linux/mm_inline.h>
34 #include <linux/slab.h>
35 #include <linux/random.h>
36 #include <linux/reboot.h>
37 #include <linux/notifier.h>
38 #include <linux/kthread.h>
39 #include <linux/workqueue.h>
40 #define __KERNEL_SYSCALLS__
41 #include <linux/unistd.h>
42 #include <linux/vmalloc.h>
43 #include <linux/sched/signal.h>
45 #include <linux/drbd_limits.h>
47 #include "drbd_protocol.h"
48 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
50 #include "drbd_debugfs.h"
52 static DEFINE_MUTEX(drbd_main_mutex);
53 static int drbd_open(struct block_device *bdev, fmode_t mode);
54 static void drbd_release(struct gendisk *gd, fmode_t mode);
55 static void md_sync_timer_fn(struct timer_list *t);
56 static int w_bitmap_io(struct drbd_work *w, int unused);
58 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
59 "Lars Ellenberg <lars@linbit.com>");
60 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
61 MODULE_VERSION(REL_VERSION);
62 MODULE_LICENSE("GPL");
63 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
64 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
65 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
67 #include <linux/moduleparam.h>
68 /* thanks to these macros, if compiled into the kernel (not-module),
69 * these become boot parameters (e.g., drbd.minor_count) */
71 #ifdef CONFIG_DRBD_FAULT_INJECTION
72 int drbd_enable_faults;
74 static int drbd_fault_count;
75 static int drbd_fault_devs;
76 /* bitmap of enabled faults */
77 module_param_named(enable_faults, drbd_enable_faults, int, 0664);
78 /* fault rate % value - applies to all enabled faults */
79 module_param_named(fault_rate, drbd_fault_rate, int, 0664);
80 /* count of faults inserted */
81 module_param_named(fault_count, drbd_fault_count, int, 0664);
82 /* bitmap of devices to insert faults on */
83 module_param_named(fault_devs, drbd_fault_devs, int, 0644);
86 /* module parameters we can keep static */
87 static bool drbd_allow_oos; /* allow_open_on_secondary */
88 static bool drbd_disable_sendpage;
89 MODULE_PARM_DESC(allow_oos, "DONT USE!");
90 module_param_named(allow_oos, drbd_allow_oos, bool, 0);
91 module_param_named(disable_sendpage, drbd_disable_sendpage, bool, 0644);
93 /* module parameters we share */
94 int drbd_proc_details; /* Detail level in proc drbd*/
95 module_param_named(proc_details, drbd_proc_details, int, 0644);
96 /* module parameters shared with defaults */
97 unsigned int drbd_minor_count = DRBD_MINOR_COUNT_DEF;
98 /* Module parameter for setting the user mode helper program
99 * to run. Default is /sbin/drbdadm */
100 char drbd_usermode_helper[80] = "/sbin/drbdadm";
101 module_param_named(minor_count, drbd_minor_count, uint, 0444);
102 module_param_string(usermode_helper, drbd_usermode_helper, sizeof(drbd_usermode_helper), 0644);
104 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
105 * as member "struct gendisk *vdisk;"
107 struct idr drbd_devices;
108 struct list_head drbd_resources;
109 struct mutex resources_mutex;
111 struct kmem_cache *drbd_request_cache;
112 struct kmem_cache *drbd_ee_cache; /* peer requests */
113 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
114 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
115 mempool_t drbd_request_mempool;
116 mempool_t drbd_ee_mempool;
117 mempool_t drbd_md_io_page_pool;
118 struct bio_set drbd_md_io_bio_set;
119 struct bio_set drbd_io_bio_set;
121 /* I do not use a standard mempool, because:
122 1) I want to hand out the pre-allocated objects first.
123 2) I want to be able to interrupt sleeping allocation with a signal.
124 Note: This is a single linked list, the next pointer is the private
125 member of struct page.
127 struct page *drbd_pp_pool;
128 spinlock_t drbd_pp_lock;
130 wait_queue_head_t drbd_pp_wait;
132 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
134 static const struct block_device_operations drbd_ops = {
135 .owner = THIS_MODULE,
136 .submit_bio = drbd_submit_bio,
138 .release = drbd_release,
141 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
145 if (!bioset_initialized(&drbd_md_io_bio_set))
146 return bio_alloc(gfp_mask, 1);
148 bio = bio_alloc_bioset(gfp_mask, 1, &drbd_md_io_bio_set);
155 /* When checking with sparse, and this is an inline function, sparse will
156 give tons of false positives. When this is a real functions sparse works.
158 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
162 atomic_inc(&device->local_cnt);
163 io_allowed = (device->state.disk >= mins);
165 if (atomic_dec_and_test(&device->local_cnt))
166 wake_up(&device->misc_wait);
174 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
175 * @connection: DRBD connection.
176 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
177 * @set_size: Expected number of requests before that barrier.
179 * In case the passed barrier_nr or set_size does not match the oldest
180 * epoch of not yet barrier-acked requests, this function will cause a
181 * termination of the connection.
183 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
184 unsigned int set_size)
186 struct drbd_request *r;
187 struct drbd_request *req = NULL;
188 int expect_epoch = 0;
191 spin_lock_irq(&connection->resource->req_lock);
193 /* find oldest not yet barrier-acked write request,
194 * count writes in its epoch. */
195 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
196 const unsigned s = r->rq_state;
200 if (!(s & RQ_NET_MASK))
205 expect_epoch = req->epoch;
208 if (r->epoch != expect_epoch)
212 /* if (s & RQ_DONE): not expected */
213 /* if (!(s & RQ_NET_MASK)): not expected */
218 /* first some paranoia code */
220 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
224 if (expect_epoch != barrier_nr) {
225 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
226 barrier_nr, expect_epoch);
230 if (expect_size != set_size) {
231 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
232 barrier_nr, set_size, expect_size);
236 /* Clean up list of requests processed during current epoch. */
237 /* this extra list walk restart is paranoia,
238 * to catch requests being barrier-acked "unexpectedly".
239 * It usually should find the same req again, or some READ preceding it. */
240 list_for_each_entry(req, &connection->transfer_log, tl_requests)
241 if (req->epoch == expect_epoch)
243 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
244 if (req->epoch != expect_epoch)
246 _req_mod(req, BARRIER_ACKED);
248 spin_unlock_irq(&connection->resource->req_lock);
253 spin_unlock_irq(&connection->resource->req_lock);
254 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
259 * _tl_restart() - Walks the transfer log, and applies an action to all requests
260 * @connection: DRBD connection to operate on.
261 * @what: The action/event to perform with all request objects
263 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
264 * RESTART_FROZEN_DISK_IO.
266 /* must hold resource->req_lock */
267 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
269 struct drbd_request *req, *r;
271 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
275 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
277 spin_lock_irq(&connection->resource->req_lock);
278 _tl_restart(connection, what);
279 spin_unlock_irq(&connection->resource->req_lock);
283 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
284 * @device: DRBD device.
286 * This is called after the connection to the peer was lost. The storage covered
287 * by the requests on the transfer gets marked as our of sync. Called from the
288 * receiver thread and the worker thread.
290 void tl_clear(struct drbd_connection *connection)
292 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
296 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
297 * @device: DRBD device.
299 void tl_abort_disk_io(struct drbd_device *device)
301 struct drbd_connection *connection = first_peer_device(device)->connection;
302 struct drbd_request *req, *r;
304 spin_lock_irq(&connection->resource->req_lock);
305 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
306 if (!(req->rq_state & RQ_LOCAL_PENDING))
308 if (req->device != device)
310 _req_mod(req, ABORT_DISK_IO);
312 spin_unlock_irq(&connection->resource->req_lock);
315 static int drbd_thread_setup(void *arg)
317 struct drbd_thread *thi = (struct drbd_thread *) arg;
318 struct drbd_resource *resource = thi->resource;
322 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
326 allow_kernel_signal(DRBD_SIGKILL);
327 allow_kernel_signal(SIGXCPU);
329 retval = thi->function(thi);
331 spin_lock_irqsave(&thi->t_lock, flags);
333 /* if the receiver has been "EXITING", the last thing it did
334 * was set the conn state to "StandAlone",
335 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
336 * and receiver thread will be "started".
337 * drbd_thread_start needs to set "RESTARTING" in that case.
338 * t_state check and assignment needs to be within the same spinlock,
339 * so either thread_start sees EXITING, and can remap to RESTARTING,
340 * or thread_start see NONE, and can proceed as normal.
343 if (thi->t_state == RESTARTING) {
344 drbd_info(resource, "Restarting %s thread\n", thi->name);
345 thi->t_state = RUNNING;
346 spin_unlock_irqrestore(&thi->t_lock, flags);
353 complete_all(&thi->stop);
354 spin_unlock_irqrestore(&thi->t_lock, flags);
356 drbd_info(resource, "Terminating %s\n", current->comm);
358 /* Release mod reference taken when thread was started */
361 kref_put(&thi->connection->kref, drbd_destroy_connection);
362 kref_put(&resource->kref, drbd_destroy_resource);
363 module_put(THIS_MODULE);
367 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
368 int (*func) (struct drbd_thread *), const char *name)
370 spin_lock_init(&thi->t_lock);
373 thi->function = func;
374 thi->resource = resource;
375 thi->connection = NULL;
379 int drbd_thread_start(struct drbd_thread *thi)
381 struct drbd_resource *resource = thi->resource;
382 struct task_struct *nt;
385 /* is used from state engine doing drbd_thread_stop_nowait,
386 * while holding the req lock irqsave */
387 spin_lock_irqsave(&thi->t_lock, flags);
389 switch (thi->t_state) {
391 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
392 thi->name, current->comm, current->pid);
394 /* Get ref on module for thread - this is released when thread exits */
395 if (!try_module_get(THIS_MODULE)) {
396 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
397 spin_unlock_irqrestore(&thi->t_lock, flags);
401 kref_get(&resource->kref);
403 kref_get(&thi->connection->kref);
405 init_completion(&thi->stop);
406 thi->reset_cpu_mask = 1;
407 thi->t_state = RUNNING;
408 spin_unlock_irqrestore(&thi->t_lock, flags);
409 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
411 nt = kthread_create(drbd_thread_setup, (void *) thi,
412 "drbd_%c_%s", thi->name[0], thi->resource->name);
415 drbd_err(resource, "Couldn't start thread\n");
418 kref_put(&thi->connection->kref, drbd_destroy_connection);
419 kref_put(&resource->kref, drbd_destroy_resource);
420 module_put(THIS_MODULE);
423 spin_lock_irqsave(&thi->t_lock, flags);
425 thi->t_state = RUNNING;
426 spin_unlock_irqrestore(&thi->t_lock, flags);
430 thi->t_state = RESTARTING;
431 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
432 thi->name, current->comm, current->pid);
437 spin_unlock_irqrestore(&thi->t_lock, flags);
445 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
449 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
451 /* may be called from state engine, holding the req lock irqsave */
452 spin_lock_irqsave(&thi->t_lock, flags);
454 if (thi->t_state == NONE) {
455 spin_unlock_irqrestore(&thi->t_lock, flags);
457 drbd_thread_start(thi);
461 if (thi->t_state != ns) {
462 if (thi->task == NULL) {
463 spin_unlock_irqrestore(&thi->t_lock, flags);
469 init_completion(&thi->stop);
470 if (thi->task != current)
471 send_sig(DRBD_SIGKILL, thi->task, 1);
474 spin_unlock_irqrestore(&thi->t_lock, flags);
477 wait_for_completion(&thi->stop);
480 int conn_lowest_minor(struct drbd_connection *connection)
482 struct drbd_peer_device *peer_device;
483 int vnr = 0, minor = -1;
486 peer_device = idr_get_next(&connection->peer_devices, &vnr);
488 minor = device_to_minor(peer_device->device);
496 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
498 * Forces all threads of a resource onto the same CPU. This is beneficial for
499 * DRBD's performance. May be overwritten by user's configuration.
501 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
503 unsigned int *resources_per_cpu, min_index = ~0;
505 resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu),
507 if (resources_per_cpu) {
508 struct drbd_resource *resource;
509 unsigned int cpu, min = ~0;
512 for_each_resource_rcu(resource, &drbd_resources) {
513 for_each_cpu(cpu, resource->cpu_mask)
514 resources_per_cpu[cpu]++;
517 for_each_online_cpu(cpu) {
518 if (resources_per_cpu[cpu] < min) {
519 min = resources_per_cpu[cpu];
523 kfree(resources_per_cpu);
525 if (min_index == ~0) {
526 cpumask_setall(*cpu_mask);
529 cpumask_set_cpu(min_index, *cpu_mask);
533 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
534 * @device: DRBD device.
535 * @thi: drbd_thread object
537 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
540 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
542 struct drbd_resource *resource = thi->resource;
543 struct task_struct *p = current;
545 if (!thi->reset_cpu_mask)
547 thi->reset_cpu_mask = 0;
548 set_cpus_allowed_ptr(p, resource->cpu_mask);
551 #define drbd_calc_cpu_mask(A) ({})
555 * drbd_header_size - size of a packet header
557 * The header size is a multiple of 8, so any payload following the header is
558 * word aligned on 64-bit architectures. (The bitmap send and receive code
561 unsigned int drbd_header_size(struct drbd_connection *connection)
563 if (connection->agreed_pro_version >= 100) {
564 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
565 return sizeof(struct p_header100);
567 BUILD_BUG_ON(sizeof(struct p_header80) !=
568 sizeof(struct p_header95));
569 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
570 return sizeof(struct p_header80);
574 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
576 h->magic = cpu_to_be32(DRBD_MAGIC);
577 h->command = cpu_to_be16(cmd);
578 h->length = cpu_to_be16(size);
579 return sizeof(struct p_header80);
582 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
584 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
585 h->command = cpu_to_be16(cmd);
586 h->length = cpu_to_be32(size);
587 return sizeof(struct p_header95);
590 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
593 h->magic = cpu_to_be32(DRBD_MAGIC_100);
594 h->volume = cpu_to_be16(vnr);
595 h->command = cpu_to_be16(cmd);
596 h->length = cpu_to_be32(size);
598 return sizeof(struct p_header100);
601 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
602 void *buffer, enum drbd_packet cmd, int size)
604 if (connection->agreed_pro_version >= 100)
605 return prepare_header100(buffer, cmd, size, vnr);
606 else if (connection->agreed_pro_version >= 95 &&
607 size > DRBD_MAX_SIZE_H80_PACKET)
608 return prepare_header95(buffer, cmd, size);
610 return prepare_header80(buffer, cmd, size);
613 static void *__conn_prepare_command(struct drbd_connection *connection,
614 struct drbd_socket *sock)
618 return sock->sbuf + drbd_header_size(connection);
621 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
625 mutex_lock(&sock->mutex);
626 p = __conn_prepare_command(connection, sock);
628 mutex_unlock(&sock->mutex);
633 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
635 return conn_prepare_command(peer_device->connection, sock);
638 static int __send_command(struct drbd_connection *connection, int vnr,
639 struct drbd_socket *sock, enum drbd_packet cmd,
640 unsigned int header_size, void *data,
647 * Called with @data == NULL and the size of the data blocks in @size
648 * for commands that send data blocks. For those commands, omit the
649 * MSG_MORE flag: this will increase the likelihood that data blocks
650 * which are page aligned on the sender will end up page aligned on the
653 msg_flags = data ? MSG_MORE : 0;
655 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
657 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
660 err = drbd_send_all(connection, sock->socket, data, size, 0);
661 /* DRBD protocol "pings" are latency critical.
662 * This is supposed to trigger tcp_push_pending_frames() */
663 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
664 tcp_sock_set_nodelay(sock->socket->sk);
669 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
670 enum drbd_packet cmd, unsigned int header_size,
671 void *data, unsigned int size)
673 return __send_command(connection, 0, sock, cmd, header_size, data, size);
676 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
677 enum drbd_packet cmd, unsigned int header_size,
678 void *data, unsigned int size)
682 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
683 mutex_unlock(&sock->mutex);
687 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
688 enum drbd_packet cmd, unsigned int header_size,
689 void *data, unsigned int size)
693 err = __send_command(peer_device->connection, peer_device->device->vnr,
694 sock, cmd, header_size, data, size);
695 mutex_unlock(&sock->mutex);
699 int drbd_send_ping(struct drbd_connection *connection)
701 struct drbd_socket *sock;
703 sock = &connection->meta;
704 if (!conn_prepare_command(connection, sock))
706 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
709 int drbd_send_ping_ack(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_ACK, 0, NULL, 0);
719 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
721 struct drbd_socket *sock;
722 struct p_rs_param_95 *p;
724 const int apv = peer_device->connection->agreed_pro_version;
725 enum drbd_packet cmd;
727 struct disk_conf *dc;
729 sock = &peer_device->connection->data;
730 p = drbd_prepare_command(peer_device, sock);
735 nc = rcu_dereference(peer_device->connection->net_conf);
737 size = apv <= 87 ? sizeof(struct p_rs_param)
738 : apv == 88 ? sizeof(struct p_rs_param)
739 + strlen(nc->verify_alg) + 1
740 : apv <= 94 ? sizeof(struct p_rs_param_89)
741 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
743 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
745 /* initialize verify_alg and csums_alg */
746 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
748 if (get_ldev(peer_device->device)) {
749 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
750 p->resync_rate = cpu_to_be32(dc->resync_rate);
751 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
752 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
753 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
754 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
755 put_ldev(peer_device->device);
757 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
758 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
759 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
760 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
761 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
765 strcpy(p->verify_alg, nc->verify_alg);
767 strcpy(p->csums_alg, nc->csums_alg);
770 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
773 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
775 struct drbd_socket *sock;
776 struct p_protocol *p;
780 sock = &connection->data;
781 p = __conn_prepare_command(connection, sock);
786 nc = rcu_dereference(connection->net_conf);
788 if (nc->tentative && connection->agreed_pro_version < 92) {
790 drbd_err(connection, "--dry-run is not supported by peer");
795 if (connection->agreed_pro_version >= 87)
796 size += strlen(nc->integrity_alg) + 1;
798 p->protocol = cpu_to_be32(nc->wire_protocol);
799 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
800 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
801 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
802 p->two_primaries = cpu_to_be32(nc->two_primaries);
804 if (nc->discard_my_data)
805 cf |= CF_DISCARD_MY_DATA;
808 p->conn_flags = cpu_to_be32(cf);
810 if (connection->agreed_pro_version >= 87)
811 strcpy(p->integrity_alg, nc->integrity_alg);
814 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
817 int drbd_send_protocol(struct drbd_connection *connection)
821 mutex_lock(&connection->data.mutex);
822 err = __drbd_send_protocol(connection, P_PROTOCOL);
823 mutex_unlock(&connection->data.mutex);
828 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
830 struct drbd_device *device = peer_device->device;
831 struct drbd_socket *sock;
835 if (!get_ldev_if_state(device, D_NEGOTIATING))
838 sock = &peer_device->connection->data;
839 p = drbd_prepare_command(peer_device, sock);
844 spin_lock_irq(&device->ldev->md.uuid_lock);
845 for (i = UI_CURRENT; i < UI_SIZE; i++)
846 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
847 spin_unlock_irq(&device->ldev->md.uuid_lock);
849 device->comm_bm_set = drbd_bm_total_weight(device);
850 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
852 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
854 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
855 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
856 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
859 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
862 int drbd_send_uuids(struct drbd_peer_device *peer_device)
864 return _drbd_send_uuids(peer_device, 0);
867 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
869 return _drbd_send_uuids(peer_device, 8);
872 void drbd_print_uuids(struct drbd_device *device, const char *text)
874 if (get_ldev_if_state(device, D_NEGOTIATING)) {
875 u64 *uuid = device->ldev->md.uuid;
876 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
878 (unsigned long long)uuid[UI_CURRENT],
879 (unsigned long long)uuid[UI_BITMAP],
880 (unsigned long long)uuid[UI_HISTORY_START],
881 (unsigned long long)uuid[UI_HISTORY_END]);
884 drbd_info(device, "%s effective data uuid: %016llX\n",
886 (unsigned long long)device->ed_uuid);
890 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
892 struct drbd_device *device = peer_device->device;
893 struct drbd_socket *sock;
897 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
899 uuid = device->ldev->md.uuid[UI_BITMAP];
900 if (uuid && uuid != UUID_JUST_CREATED)
901 uuid = uuid + UUID_NEW_BM_OFFSET;
903 get_random_bytes(&uuid, sizeof(u64));
904 drbd_uuid_set(device, UI_BITMAP, uuid);
905 drbd_print_uuids(device, "updated sync UUID");
906 drbd_md_sync(device);
908 sock = &peer_device->connection->data;
909 p = drbd_prepare_command(peer_device, sock);
911 p->uuid = cpu_to_be64(uuid);
912 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
916 /* communicated if (agreed_features & DRBD_FF_WSAME) */
918 assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p,
919 struct request_queue *q)
922 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
923 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
924 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
925 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
926 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
927 p->qlim->discard_enabled = blk_queue_discard(q);
928 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
930 q = device->rq_queue;
931 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
932 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
933 p->qlim->alignment_offset = 0;
934 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
935 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
936 p->qlim->discard_enabled = 0;
937 p->qlim->write_same_capable = 0;
941 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
943 struct drbd_device *device = peer_device->device;
944 struct drbd_socket *sock;
946 sector_t d_size, u_size;
948 unsigned int max_bio_size;
949 unsigned int packet_size;
951 sock = &peer_device->connection->data;
952 p = drbd_prepare_command(peer_device, sock);
956 packet_size = sizeof(*p);
957 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
958 packet_size += sizeof(p->qlim[0]);
960 memset(p, 0, packet_size);
961 if (get_ldev_if_state(device, D_NEGOTIATING)) {
962 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
963 d_size = drbd_get_max_capacity(device->ldev);
965 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
967 q_order_type = drbd_queue_order_type(device);
968 max_bio_size = queue_max_hw_sectors(q) << 9;
969 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
970 assign_p_sizes_qlim(device, p, q);
975 q_order_type = QUEUE_ORDERED_NONE;
976 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
977 assign_p_sizes_qlim(device, p, NULL);
980 if (peer_device->connection->agreed_pro_version <= 94)
981 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
982 else if (peer_device->connection->agreed_pro_version < 100)
983 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
985 p->d_size = cpu_to_be64(d_size);
986 p->u_size = cpu_to_be64(u_size);
987 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
988 p->max_bio_size = cpu_to_be32(max_bio_size);
989 p->queue_order_type = cpu_to_be16(q_order_type);
990 p->dds_flags = cpu_to_be16(flags);
992 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
996 * drbd_send_current_state() - Sends the drbd state to the peer
997 * @peer_device: DRBD peer device.
999 int drbd_send_current_state(struct drbd_peer_device *peer_device)
1001 struct drbd_socket *sock;
1004 sock = &peer_device->connection->data;
1005 p = drbd_prepare_command(peer_device, sock);
1008 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1009 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1013 * drbd_send_state() - After a state change, sends the new state to the peer
1014 * @peer_device: DRBD peer device.
1015 * @state: the state to send, not necessarily the current state.
1017 * Each state change queues an "after_state_ch" work, which will eventually
1018 * send the resulting new state to the peer. If more state changes happen
1019 * between queuing and processing of the after_state_ch work, we still
1020 * want to send each intermediary state in the order it occurred.
1022 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1024 struct drbd_socket *sock;
1027 sock = &peer_device->connection->data;
1028 p = drbd_prepare_command(peer_device, sock);
1031 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1032 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1035 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1037 struct drbd_socket *sock;
1038 struct p_req_state *p;
1040 sock = &peer_device->connection->data;
1041 p = drbd_prepare_command(peer_device, sock);
1044 p->mask = cpu_to_be32(mask.i);
1045 p->val = cpu_to_be32(val.i);
1046 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1049 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1051 enum drbd_packet cmd;
1052 struct drbd_socket *sock;
1053 struct p_req_state *p;
1055 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1056 sock = &connection->data;
1057 p = conn_prepare_command(connection, sock);
1060 p->mask = cpu_to_be32(mask.i);
1061 p->val = cpu_to_be32(val.i);
1062 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1065 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1067 struct drbd_socket *sock;
1068 struct p_req_state_reply *p;
1070 sock = &peer_device->connection->meta;
1071 p = drbd_prepare_command(peer_device, sock);
1073 p->retcode = cpu_to_be32(retcode);
1074 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1078 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1080 struct drbd_socket *sock;
1081 struct p_req_state_reply *p;
1082 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1084 sock = &connection->meta;
1085 p = conn_prepare_command(connection, sock);
1087 p->retcode = cpu_to_be32(retcode);
1088 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1092 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1094 BUG_ON(code & ~0xf);
1095 p->encoding = (p->encoding & ~0xf) | code;
1098 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1100 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1103 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1106 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1109 static int fill_bitmap_rle_bits(struct drbd_device *device,
1110 struct p_compressed_bm *p,
1112 struct bm_xfer_ctx *c)
1114 struct bitstream bs;
1115 unsigned long plain_bits;
1122 /* may we use this feature? */
1124 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1126 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1129 if (c->bit_offset >= c->bm_bits)
1130 return 0; /* nothing to do. */
1132 /* use at most thus many bytes */
1133 bitstream_init(&bs, p->code, size, 0);
1134 memset(p->code, 0, size);
1135 /* plain bits covered in this code string */
1138 /* p->encoding & 0x80 stores whether the first run length is set.
1139 * bit offset is implicit.
1140 * start with toggle == 2 to be able to tell the first iteration */
1143 /* see how much plain bits we can stuff into one packet
1144 * using RLE and VLI. */
1146 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1147 : _drbd_bm_find_next(device, c->bit_offset);
1150 rl = tmp - c->bit_offset;
1152 if (toggle == 2) { /* first iteration */
1154 /* the first checked bit was set,
1155 * store start value, */
1156 dcbp_set_start(p, 1);
1157 /* but skip encoding of zero run length */
1161 dcbp_set_start(p, 0);
1164 /* paranoia: catch zero runlength.
1165 * can only happen if bitmap is modified while we scan it. */
1167 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1168 "t:%u bo:%lu\n", toggle, c->bit_offset);
1172 bits = vli_encode_bits(&bs, rl);
1173 if (bits == -ENOBUFS) /* buffer full */
1176 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1182 c->bit_offset = tmp;
1183 } while (c->bit_offset < c->bm_bits);
1185 len = bs.cur.b - p->code + !!bs.cur.bit;
1187 if (plain_bits < (len << 3)) {
1188 /* incompressible with this method.
1189 * we need to rewind both word and bit position. */
1190 c->bit_offset -= plain_bits;
1191 bm_xfer_ctx_bit_to_word_offset(c);
1192 c->bit_offset = c->word_offset * BITS_PER_LONG;
1196 /* RLE + VLI was able to compress it just fine.
1197 * update c->word_offset. */
1198 bm_xfer_ctx_bit_to_word_offset(c);
1200 /* store pad_bits */
1201 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1207 * send_bitmap_rle_or_plain
1209 * Return 0 when done, 1 when another iteration is needed, and a negative error
1210 * code upon failure.
1213 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1215 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1216 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1217 struct p_compressed_bm *p = sock->sbuf + header_size;
1220 len = fill_bitmap_rle_bits(device, p,
1221 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1226 dcbp_set_code(p, RLE_VLI_Bits);
1227 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1228 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1231 c->bytes[0] += header_size + sizeof(*p) + len;
1233 if (c->bit_offset >= c->bm_bits)
1236 /* was not compressible.
1237 * send a buffer full of plain text bits instead. */
1238 unsigned int data_size;
1239 unsigned long num_words;
1240 unsigned long *p = sock->sbuf + header_size;
1242 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1243 num_words = min_t(size_t, data_size / sizeof(*p),
1244 c->bm_words - c->word_offset);
1245 len = num_words * sizeof(*p);
1247 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1248 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1249 c->word_offset += num_words;
1250 c->bit_offset = c->word_offset * BITS_PER_LONG;
1253 c->bytes[1] += header_size + len;
1255 if (c->bit_offset > c->bm_bits)
1256 c->bit_offset = c->bm_bits;
1260 INFO_bm_xfer_stats(device, "send", c);
1268 /* See the comment at receive_bitmap() */
1269 static int _drbd_send_bitmap(struct drbd_device *device)
1271 struct bm_xfer_ctx c;
1274 if (!expect(device->bitmap))
1277 if (get_ldev(device)) {
1278 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1279 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1280 drbd_bm_set_all(device);
1281 if (drbd_bm_write(device)) {
1282 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1283 * but otherwise process as per normal - need to tell other
1284 * side that a full resync is required! */
1285 drbd_err(device, "Failed to write bitmap to disk!\n");
1287 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1288 drbd_md_sync(device);
1294 c = (struct bm_xfer_ctx) {
1295 .bm_bits = drbd_bm_bits(device),
1296 .bm_words = drbd_bm_words(device),
1300 err = send_bitmap_rle_or_plain(device, &c);
1306 int drbd_send_bitmap(struct drbd_device *device)
1308 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1311 mutex_lock(&sock->mutex);
1313 err = !_drbd_send_bitmap(device);
1314 mutex_unlock(&sock->mutex);
1318 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1320 struct drbd_socket *sock;
1321 struct p_barrier_ack *p;
1323 if (connection->cstate < C_WF_REPORT_PARAMS)
1326 sock = &connection->meta;
1327 p = conn_prepare_command(connection, sock);
1330 p->barrier = barrier_nr;
1331 p->set_size = cpu_to_be32(set_size);
1332 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1336 * _drbd_send_ack() - Sends an ack packet
1337 * @device: DRBD device.
1338 * @cmd: Packet command code.
1339 * @sector: sector, needs to be in big endian byte order
1340 * @blksize: size in byte, needs to be in big endian byte order
1341 * @block_id: Id, big endian byte order
1343 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1344 u64 sector, u32 blksize, u64 block_id)
1346 struct drbd_socket *sock;
1347 struct p_block_ack *p;
1349 if (peer_device->device->state.conn < C_CONNECTED)
1352 sock = &peer_device->connection->meta;
1353 p = drbd_prepare_command(peer_device, sock);
1357 p->block_id = block_id;
1358 p->blksize = blksize;
1359 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1360 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1363 /* dp->sector and dp->block_id already/still in network byte order,
1364 * data_size is payload size according to dp->head,
1365 * and may need to be corrected for digest size. */
1366 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1367 struct p_data *dp, int data_size)
1369 if (peer_device->connection->peer_integrity_tfm)
1370 data_size -= crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm);
1371 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1375 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1376 struct p_block_req *rp)
1378 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1382 * drbd_send_ack() - Sends an ack packet
1383 * @device: DRBD device
1384 * @cmd: packet command code
1385 * @peer_req: peer request
1387 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1388 struct drbd_peer_request *peer_req)
1390 return _drbd_send_ack(peer_device, cmd,
1391 cpu_to_be64(peer_req->i.sector),
1392 cpu_to_be32(peer_req->i.size),
1393 peer_req->block_id);
1396 /* This function misuses the block_id field to signal if the blocks
1397 * are is sync or not. */
1398 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1399 sector_t sector, int blksize, u64 block_id)
1401 return _drbd_send_ack(peer_device, cmd,
1402 cpu_to_be64(sector),
1403 cpu_to_be32(blksize),
1404 cpu_to_be64(block_id));
1407 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1408 struct drbd_peer_request *peer_req)
1410 struct drbd_socket *sock;
1411 struct p_block_desc *p;
1413 sock = &peer_device->connection->data;
1414 p = drbd_prepare_command(peer_device, sock);
1417 p->sector = cpu_to_be64(peer_req->i.sector);
1418 p->blksize = cpu_to_be32(peer_req->i.size);
1420 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1423 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1424 sector_t sector, int size, u64 block_id)
1426 struct drbd_socket *sock;
1427 struct p_block_req *p;
1429 sock = &peer_device->connection->data;
1430 p = drbd_prepare_command(peer_device, sock);
1433 p->sector = cpu_to_be64(sector);
1434 p->block_id = block_id;
1435 p->blksize = cpu_to_be32(size);
1436 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1439 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1440 void *digest, int digest_size, enum drbd_packet cmd)
1442 struct drbd_socket *sock;
1443 struct p_block_req *p;
1445 /* FIXME: Put the digest into the preallocated socket buffer. */
1447 sock = &peer_device->connection->data;
1448 p = drbd_prepare_command(peer_device, sock);
1451 p->sector = cpu_to_be64(sector);
1452 p->block_id = ID_SYNCER /* unused */;
1453 p->blksize = cpu_to_be32(size);
1454 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1457 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1459 struct drbd_socket *sock;
1460 struct p_block_req *p;
1462 sock = &peer_device->connection->data;
1463 p = drbd_prepare_command(peer_device, sock);
1466 p->sector = cpu_to_be64(sector);
1467 p->block_id = ID_SYNCER /* unused */;
1468 p->blksize = cpu_to_be32(size);
1469 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1472 /* called on sndtimeo
1473 * returns false if we should retry,
1474 * true if we think connection is dead
1476 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1479 /* long elapsed = (long)(jiffies - device->last_received); */
1481 drop_it = connection->meta.socket == sock
1482 || !connection->ack_receiver.task
1483 || get_t_state(&connection->ack_receiver) != RUNNING
1484 || connection->cstate < C_WF_REPORT_PARAMS;
1489 drop_it = !--connection->ko_count;
1491 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1492 current->comm, current->pid, connection->ko_count);
1493 request_ping(connection);
1496 return drop_it; /* && (device->state == R_PRIMARY) */;
1499 static void drbd_update_congested(struct drbd_connection *connection)
1501 struct sock *sk = connection->data.socket->sk;
1502 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1503 set_bit(NET_CONGESTED, &connection->flags);
1506 /* The idea of sendpage seems to be to put some kind of reference
1507 * to the page into the skb, and to hand it over to the NIC. In
1508 * this process get_page() gets called.
1510 * As soon as the page was really sent over the network put_page()
1511 * gets called by some part of the network layer. [ NIC driver? ]
1513 * [ get_page() / put_page() increment/decrement the count. If count
1514 * reaches 0 the page will be freed. ]
1516 * This works nicely with pages from FSs.
1517 * But this means that in protocol A we might signal IO completion too early!
1519 * In order not to corrupt data during a resync we must make sure
1520 * that we do not reuse our own buffer pages (EEs) to early, therefore
1521 * we have the net_ee list.
1523 * XFS seems to have problems, still, it submits pages with page_count == 0!
1524 * As a workaround, we disable sendpage on pages
1525 * with page_count == 0 or PageSlab.
1527 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1528 int offset, size_t size, unsigned msg_flags)
1530 struct socket *socket;
1534 socket = peer_device->connection->data.socket;
1535 addr = kmap(page) + offset;
1536 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1539 peer_device->device->send_cnt += size >> 9;
1543 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1544 int offset, size_t size, unsigned msg_flags)
1546 struct socket *socket = peer_device->connection->data.socket;
1550 /* e.g. XFS meta- & log-data is in slab pages, which have a
1551 * page_count of 0 and/or have PageSlab() set.
1552 * we cannot use send_page for those, as that does get_page();
1553 * put_page(); and would cause either a VM_BUG directly, or
1554 * __page_cache_release a page that would actually still be referenced
1555 * by someone, leading to some obscure delayed Oops somewhere else. */
1556 if (drbd_disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1557 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1559 msg_flags |= MSG_NOSIGNAL;
1560 drbd_update_congested(peer_device->connection);
1564 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1566 if (sent == -EAGAIN) {
1567 if (we_should_drop_the_connection(peer_device->connection, socket))
1571 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1572 __func__, (int)size, len, sent);
1579 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1580 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1584 peer_device->device->send_cnt += size >> 9;
1589 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1591 struct bio_vec bvec;
1592 struct bvec_iter iter;
1594 /* hint all but last page with MSG_MORE */
1595 bio_for_each_segment(bvec, bio, iter) {
1598 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1599 bvec.bv_offset, bvec.bv_len,
1600 bio_iter_last(bvec, iter)
1604 /* REQ_OP_WRITE_SAME has only one segment */
1605 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1611 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1613 struct bio_vec bvec;
1614 struct bvec_iter iter;
1616 /* hint all but last page with MSG_MORE */
1617 bio_for_each_segment(bvec, bio, iter) {
1620 err = _drbd_send_page(peer_device, bvec.bv_page,
1621 bvec.bv_offset, bvec.bv_len,
1622 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1625 /* REQ_OP_WRITE_SAME has only one segment */
1626 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1632 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1633 struct drbd_peer_request *peer_req)
1635 struct page *page = peer_req->pages;
1636 unsigned len = peer_req->i.size;
1639 /* hint all but last page with MSG_MORE */
1640 page_chain_for_each(page) {
1641 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1643 err = _drbd_send_page(peer_device, page, 0, l,
1644 page_chain_next(page) ? MSG_MORE : 0);
1652 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1655 if (connection->agreed_pro_version >= 95)
1656 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1657 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1658 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1659 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1660 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1661 (bio_op(bio) == REQ_OP_WRITE_ZEROES ?
1662 ((connection->agreed_features & DRBD_FF_WZEROES) ?
1663 (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0))
1667 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1670 /* Used to send write or TRIM aka REQ_OP_DISCARD requests
1671 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1673 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1675 struct drbd_device *device = peer_device->device;
1676 struct drbd_socket *sock;
1678 struct p_wsame *wsame = NULL;
1680 unsigned int dp_flags = 0;
1684 sock = &peer_device->connection->data;
1685 p = drbd_prepare_command(peer_device, sock);
1686 digest_size = peer_device->connection->integrity_tfm ?
1687 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1691 p->sector = cpu_to_be64(req->i.sector);
1692 p->block_id = (unsigned long)req;
1693 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1694 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1695 if (device->state.conn >= C_SYNC_SOURCE &&
1696 device->state.conn <= C_PAUSED_SYNC_T)
1697 dp_flags |= DP_MAY_SET_IN_SYNC;
1698 if (peer_device->connection->agreed_pro_version >= 100) {
1699 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1700 dp_flags |= DP_SEND_RECEIVE_ACK;
1701 /* During resync, request an explicit write ack,
1702 * even in protocol != C */
1703 if (req->rq_state & RQ_EXP_WRITE_ACK
1704 || (dp_flags & DP_MAY_SET_IN_SYNC))
1705 dp_flags |= DP_SEND_WRITE_ACK;
1707 p->dp_flags = cpu_to_be32(dp_flags);
1709 if (dp_flags & (DP_DISCARD|DP_ZEROES)) {
1710 enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM;
1711 struct p_trim *t = (struct p_trim*)p;
1712 t->size = cpu_to_be32(req->i.size);
1713 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0);
1716 if (dp_flags & DP_WSAME) {
1717 /* this will only work if DRBD_FF_WSAME is set AND the
1718 * handshake agreed that all nodes and backend devices are
1719 * WRITE_SAME capable and agree on logical_block_size */
1720 wsame = (struct p_wsame*)p;
1721 digest_out = wsame + 1;
1722 wsame->size = cpu_to_be32(req->i.size);
1726 /* our digest is still only over the payload.
1727 * TRIM does not carry any payload. */
1729 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1732 __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1733 sizeof(*wsame) + digest_size, NULL,
1734 bio_iovec(req->master_bio).bv_len);
1737 __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1738 sizeof(*p) + digest_size, NULL, req->i.size);
1740 /* For protocol A, we have to memcpy the payload into
1741 * socket buffers, as we may complete right away
1742 * as soon as we handed it over to tcp, at which point the data
1743 * pages may become invalid.
1745 * For data-integrity enabled, we copy it as well, so we can be
1746 * sure that even if the bio pages may still be modified, it
1747 * won't change the data on the wire, thus if the digest checks
1748 * out ok after sending on this side, but does not fit on the
1749 * receiving side, we sure have detected corruption elsewhere.
1751 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1752 err = _drbd_send_bio(peer_device, req->master_bio);
1754 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1756 /* double check digest, sometimes buffers have been modified in flight. */
1757 if (digest_size > 0 && digest_size <= 64) {
1758 /* 64 byte, 512 bit, is the largest digest size
1759 * currently supported in kernel crypto. */
1760 unsigned char digest[64];
1761 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1762 if (memcmp(p + 1, digest, digest_size)) {
1764 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1765 (unsigned long long)req->i.sector, req->i.size);
1767 } /* else if (digest_size > 64) {
1768 ... Be noisy about digest too large ...
1772 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1777 /* answer packet, used to send data back for read requests:
1778 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1779 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1781 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1782 struct drbd_peer_request *peer_req)
1784 struct drbd_device *device = peer_device->device;
1785 struct drbd_socket *sock;
1790 sock = &peer_device->connection->data;
1791 p = drbd_prepare_command(peer_device, sock);
1793 digest_size = peer_device->connection->integrity_tfm ?
1794 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1798 p->sector = cpu_to_be64(peer_req->i.sector);
1799 p->block_id = peer_req->block_id;
1800 p->seq_num = 0; /* unused */
1803 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1804 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1806 err = _drbd_send_zc_ee(peer_device, peer_req);
1807 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1812 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1814 struct drbd_socket *sock;
1815 struct p_block_desc *p;
1817 sock = &peer_device->connection->data;
1818 p = drbd_prepare_command(peer_device, sock);
1821 p->sector = cpu_to_be64(req->i.sector);
1822 p->blksize = cpu_to_be32(req->i.size);
1823 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1827 drbd_send distinguishes two cases:
1829 Packets sent via the data socket "sock"
1830 and packets sent via the meta data socket "msock"
1833 -----------------+-------------------------+------------------------------
1834 timeout conf.timeout / 2 conf.timeout / 2
1835 timeout action send a ping via msock Abort communication
1836 and close all sockets
1840 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1842 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1843 void *buf, size_t size, unsigned msg_flags)
1845 struct kvec iov = {.iov_base = buf, .iov_len = size};
1846 struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL};
1852 /* THINK if (signal_pending) return ... ? */
1854 iov_iter_kvec(&msg.msg_iter, WRITE, &iov, 1, size);
1856 if (sock == connection->data.socket) {
1858 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1860 drbd_update_congested(connection);
1863 rv = sock_sendmsg(sock, &msg);
1864 if (rv == -EAGAIN) {
1865 if (we_should_drop_the_connection(connection, sock))
1871 flush_signals(current);
1877 } while (sent < size);
1879 if (sock == connection->data.socket)
1880 clear_bit(NET_CONGESTED, &connection->flags);
1883 if (rv != -EAGAIN) {
1884 drbd_err(connection, "%s_sendmsg returned %d\n",
1885 sock == connection->meta.socket ? "msock" : "sock",
1887 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1889 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1896 * drbd_send_all - Send an entire buffer
1898 * Returns 0 upon success and a negative error value otherwise.
1900 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1901 size_t size, unsigned msg_flags)
1905 err = drbd_send(connection, sock, buffer, size, msg_flags);
1913 static int drbd_open(struct block_device *bdev, fmode_t mode)
1915 struct drbd_device *device = bdev->bd_disk->private_data;
1916 unsigned long flags;
1919 mutex_lock(&drbd_main_mutex);
1920 spin_lock_irqsave(&device->resource->req_lock, flags);
1921 /* to have a stable device->state.role
1922 * and no race with updating open_cnt */
1924 if (device->state.role != R_PRIMARY) {
1925 if (mode & FMODE_WRITE)
1927 else if (!drbd_allow_oos)
1933 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1934 mutex_unlock(&drbd_main_mutex);
1939 static void drbd_release(struct gendisk *gd, fmode_t mode)
1941 struct drbd_device *device = gd->private_data;
1942 mutex_lock(&drbd_main_mutex);
1944 mutex_unlock(&drbd_main_mutex);
1947 /* need to hold resource->req_lock */
1948 void drbd_queue_unplug(struct drbd_device *device)
1950 if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1951 D_ASSERT(device, device->state.role == R_PRIMARY);
1952 if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1953 drbd_queue_work_if_unqueued(
1954 &first_peer_device(device)->connection->sender_work,
1955 &device->unplug_work);
1960 static void drbd_set_defaults(struct drbd_device *device)
1962 /* Beware! The actual layout differs
1963 * between big endian and little endian */
1964 device->state = (union drbd_dev_state) {
1965 { .role = R_SECONDARY,
1967 .conn = C_STANDALONE,
1973 void drbd_init_set_defaults(struct drbd_device *device)
1975 /* the memset(,0,) did most of this.
1976 * note: only assignments, no allocation in here */
1978 drbd_set_defaults(device);
1980 atomic_set(&device->ap_bio_cnt, 0);
1981 atomic_set(&device->ap_actlog_cnt, 0);
1982 atomic_set(&device->ap_pending_cnt, 0);
1983 atomic_set(&device->rs_pending_cnt, 0);
1984 atomic_set(&device->unacked_cnt, 0);
1985 atomic_set(&device->local_cnt, 0);
1986 atomic_set(&device->pp_in_use_by_net, 0);
1987 atomic_set(&device->rs_sect_in, 0);
1988 atomic_set(&device->rs_sect_ev, 0);
1989 atomic_set(&device->ap_in_flight, 0);
1990 atomic_set(&device->md_io.in_use, 0);
1992 mutex_init(&device->own_state_mutex);
1993 device->state_mutex = &device->own_state_mutex;
1995 spin_lock_init(&device->al_lock);
1996 spin_lock_init(&device->peer_seq_lock);
1998 INIT_LIST_HEAD(&device->active_ee);
1999 INIT_LIST_HEAD(&device->sync_ee);
2000 INIT_LIST_HEAD(&device->done_ee);
2001 INIT_LIST_HEAD(&device->read_ee);
2002 INIT_LIST_HEAD(&device->net_ee);
2003 INIT_LIST_HEAD(&device->resync_reads);
2004 INIT_LIST_HEAD(&device->resync_work.list);
2005 INIT_LIST_HEAD(&device->unplug_work.list);
2006 INIT_LIST_HEAD(&device->bm_io_work.w.list);
2007 INIT_LIST_HEAD(&device->pending_master_completion[0]);
2008 INIT_LIST_HEAD(&device->pending_master_completion[1]);
2009 INIT_LIST_HEAD(&device->pending_completion[0]);
2010 INIT_LIST_HEAD(&device->pending_completion[1]);
2012 device->resync_work.cb = w_resync_timer;
2013 device->unplug_work.cb = w_send_write_hint;
2014 device->bm_io_work.w.cb = w_bitmap_io;
2016 timer_setup(&device->resync_timer, resync_timer_fn, 0);
2017 timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0);
2018 timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0);
2019 timer_setup(&device->request_timer, request_timer_fn, 0);
2021 init_waitqueue_head(&device->misc_wait);
2022 init_waitqueue_head(&device->state_wait);
2023 init_waitqueue_head(&device->ee_wait);
2024 init_waitqueue_head(&device->al_wait);
2025 init_waitqueue_head(&device->seq_wait);
2027 device->resync_wenr = LC_FREE;
2028 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2029 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2032 static void _drbd_set_my_capacity(struct drbd_device *device, sector_t size)
2034 /* set_capacity(device->this_bdev->bd_disk, size); */
2035 set_capacity(device->vdisk, size);
2036 device->this_bdev->bd_inode->i_size = (loff_t)size << 9;
2039 void drbd_set_my_capacity(struct drbd_device *device, sector_t size)
2042 _drbd_set_my_capacity(device, size);
2043 drbd_info(device, "size = %s (%llu KB)\n",
2044 ppsize(ppb, size>>1), (unsigned long long)size>>1);
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 bioset_exit(&drbd_io_bio_set);
2112 bioset_exit(&drbd_md_io_bio_set);
2113 mempool_exit(&drbd_md_io_page_pool);
2114 mempool_exit(&drbd_ee_mempool);
2115 mempool_exit(&drbd_request_mempool);
2116 kmem_cache_destroy(drbd_ee_cache);
2117 kmem_cache_destroy(drbd_request_cache);
2118 kmem_cache_destroy(drbd_bm_ext_cache);
2119 kmem_cache_destroy(drbd_al_ext_cache);
2121 drbd_ee_cache = NULL;
2122 drbd_request_cache = NULL;
2123 drbd_bm_ext_cache = NULL;
2124 drbd_al_ext_cache = NULL;
2129 static int drbd_create_mempools(void)
2132 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2136 drbd_request_cache = kmem_cache_create(
2137 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2138 if (drbd_request_cache == NULL)
2141 drbd_ee_cache = kmem_cache_create(
2142 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2143 if (drbd_ee_cache == NULL)
2146 drbd_bm_ext_cache = kmem_cache_create(
2147 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2148 if (drbd_bm_ext_cache == NULL)
2151 drbd_al_ext_cache = kmem_cache_create(
2152 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2153 if (drbd_al_ext_cache == NULL)
2157 ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0);
2161 ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0,
2166 ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0);
2170 ret = mempool_init_slab_pool(&drbd_request_mempool, number,
2171 drbd_request_cache);
2175 ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache);
2179 /* drbd's page pool */
2180 spin_lock_init(&drbd_pp_lock);
2182 for (i = 0; i < number; i++) {
2183 page = alloc_page(GFP_HIGHUSER);
2186 set_page_private(page, (unsigned long)drbd_pp_pool);
2187 drbd_pp_pool = page;
2189 drbd_pp_vacant = number;
2194 drbd_destroy_mempools(); /* in case we allocated some */
2198 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2202 rr = drbd_free_peer_reqs(device, &device->active_ee);
2204 drbd_err(device, "%d EEs in active list found!\n", rr);
2206 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2208 drbd_err(device, "%d EEs in sync list found!\n", rr);
2210 rr = drbd_free_peer_reqs(device, &device->read_ee);
2212 drbd_err(device, "%d EEs in read list found!\n", rr);
2214 rr = drbd_free_peer_reqs(device, &device->done_ee);
2216 drbd_err(device, "%d EEs in done list found!\n", rr);
2218 rr = drbd_free_peer_reqs(device, &device->net_ee);
2220 drbd_err(device, "%d EEs in net list found!\n", rr);
2223 /* caution. no locking. */
2224 void drbd_destroy_device(struct kref *kref)
2226 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2227 struct drbd_resource *resource = device->resource;
2228 struct drbd_peer_device *peer_device, *tmp_peer_device;
2230 del_timer_sync(&device->request_timer);
2232 /* paranoia asserts */
2233 D_ASSERT(device, device->open_cnt == 0);
2234 /* end paranoia asserts */
2236 /* cleanup stuff that may have been allocated during
2237 * device (re-)configuration or state changes */
2239 if (device->this_bdev)
2240 bdput(device->this_bdev);
2242 drbd_backing_dev_free(device, device->ldev);
2243 device->ldev = NULL;
2245 drbd_release_all_peer_reqs(device);
2247 lc_destroy(device->act_log);
2248 lc_destroy(device->resync);
2250 kfree(device->p_uuid);
2251 /* device->p_uuid = NULL; */
2253 if (device->bitmap) /* should no longer be there. */
2254 drbd_bm_cleanup(device);
2255 __free_page(device->md_io.page);
2256 put_disk(device->vdisk);
2257 blk_cleanup_queue(device->rq_queue);
2258 kfree(device->rs_plan_s);
2260 /* not for_each_connection(connection, resource):
2261 * those may have been cleaned up and disassociated already.
2263 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2264 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2267 memset(device, 0xfd, sizeof(*device));
2269 kref_put(&resource->kref, drbd_destroy_resource);
2272 /* One global retry thread, if we need to push back some bio and have it
2273 * reinserted through our make request function.
2275 static struct retry_worker {
2276 struct workqueue_struct *wq;
2277 struct work_struct worker;
2280 struct list_head writes;
2283 static void do_retry(struct work_struct *ws)
2285 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2287 struct drbd_request *req, *tmp;
2289 spin_lock_irq(&retry->lock);
2290 list_splice_init(&retry->writes, &writes);
2291 spin_unlock_irq(&retry->lock);
2293 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2294 struct drbd_device *device = req->device;
2295 struct bio *bio = req->master_bio;
2296 unsigned long start_jif = req->start_jif;
2300 expect(atomic_read(&req->completion_ref) == 0) &&
2301 expect(req->rq_state & RQ_POSTPONED) &&
2302 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2303 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2306 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2307 req, atomic_read(&req->completion_ref),
2310 /* We still need to put one kref associated with the
2311 * "completion_ref" going zero in the code path that queued it
2312 * here. The request object may still be referenced by a
2313 * frozen local req->private_bio, in case we force-detached.
2315 kref_put(&req->kref, drbd_req_destroy);
2317 /* A single suspended or otherwise blocking device may stall
2318 * all others as well. Fortunately, this code path is to
2319 * recover from a situation that "should not happen":
2320 * concurrent writes in multi-primary setup.
2321 * In a "normal" lifecycle, this workqueue is supposed to be
2322 * destroyed without ever doing anything.
2323 * If it turns out to be an issue anyways, we can do per
2324 * resource (replication group) or per device (minor) retry
2325 * workqueues instead.
2328 /* We are not just doing submit_bio_noacct(),
2329 * as we want to keep the start_time information. */
2331 __drbd_make_request(device, bio, start_jif);
2335 /* called via drbd_req_put_completion_ref(),
2336 * holds resource->req_lock */
2337 void drbd_restart_request(struct drbd_request *req)
2339 unsigned long flags;
2340 spin_lock_irqsave(&retry.lock, flags);
2341 list_move_tail(&req->tl_requests, &retry.writes);
2342 spin_unlock_irqrestore(&retry.lock, flags);
2344 /* Drop the extra reference that would otherwise
2345 * have been dropped by complete_master_bio.
2346 * do_retry() needs to grab a new one. */
2347 dec_ap_bio(req->device);
2349 queue_work(retry.wq, &retry.worker);
2352 void drbd_destroy_resource(struct kref *kref)
2354 struct drbd_resource *resource =
2355 container_of(kref, struct drbd_resource, kref);
2357 idr_destroy(&resource->devices);
2358 free_cpumask_var(resource->cpu_mask);
2359 kfree(resource->name);
2360 memset(resource, 0xf2, sizeof(*resource));
2364 void drbd_free_resource(struct drbd_resource *resource)
2366 struct drbd_connection *connection, *tmp;
2368 for_each_connection_safe(connection, tmp, resource) {
2369 list_del(&connection->connections);
2370 drbd_debugfs_connection_cleanup(connection);
2371 kref_put(&connection->kref, drbd_destroy_connection);
2373 drbd_debugfs_resource_cleanup(resource);
2374 kref_put(&resource->kref, drbd_destroy_resource);
2377 static void drbd_cleanup(void)
2380 struct drbd_device *device;
2381 struct drbd_resource *resource, *tmp;
2383 /* first remove proc,
2384 * drbdsetup uses it's presence to detect
2385 * whether DRBD is loaded.
2386 * If we would get stuck in proc removal,
2387 * but have netlink already deregistered,
2388 * some drbdsetup commands may wait forever
2392 remove_proc_entry("drbd", NULL);
2395 destroy_workqueue(retry.wq);
2397 drbd_genl_unregister();
2399 idr_for_each_entry(&drbd_devices, device, i)
2400 drbd_delete_device(device);
2402 /* not _rcu since, no other updater anymore. Genl already unregistered */
2403 for_each_resource_safe(resource, tmp, &drbd_resources) {
2404 list_del(&resource->resources);
2405 drbd_free_resource(resource);
2408 drbd_debugfs_cleanup();
2410 drbd_destroy_mempools();
2411 unregister_blkdev(DRBD_MAJOR, "drbd");
2413 idr_destroy(&drbd_devices);
2415 pr_info("module cleanup done.\n");
2418 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2420 spin_lock_init(&wq->q_lock);
2421 INIT_LIST_HEAD(&wq->q);
2422 init_waitqueue_head(&wq->q_wait);
2425 struct completion_work {
2427 struct completion done;
2430 static int w_complete(struct drbd_work *w, int cancel)
2432 struct completion_work *completion_work =
2433 container_of(w, struct completion_work, w);
2435 complete(&completion_work->done);
2439 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2441 struct completion_work completion_work;
2443 completion_work.w.cb = w_complete;
2444 init_completion(&completion_work.done);
2445 drbd_queue_work(work_queue, &completion_work.w);
2446 wait_for_completion(&completion_work.done);
2449 struct drbd_resource *drbd_find_resource(const char *name)
2451 struct drbd_resource *resource;
2453 if (!name || !name[0])
2457 for_each_resource_rcu(resource, &drbd_resources) {
2458 if (!strcmp(resource->name, name)) {
2459 kref_get(&resource->kref);
2469 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2470 void *peer_addr, int peer_addr_len)
2472 struct drbd_resource *resource;
2473 struct drbd_connection *connection;
2476 for_each_resource_rcu(resource, &drbd_resources) {
2477 for_each_connection_rcu(connection, resource) {
2478 if (connection->my_addr_len == my_addr_len &&
2479 connection->peer_addr_len == peer_addr_len &&
2480 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2481 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2482 kref_get(&connection->kref);
2493 static int drbd_alloc_socket(struct drbd_socket *socket)
2495 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2498 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2504 static void drbd_free_socket(struct drbd_socket *socket)
2506 free_page((unsigned long) socket->sbuf);
2507 free_page((unsigned long) socket->rbuf);
2510 void conn_free_crypto(struct drbd_connection *connection)
2512 drbd_free_sock(connection);
2514 crypto_free_shash(connection->csums_tfm);
2515 crypto_free_shash(connection->verify_tfm);
2516 crypto_free_shash(connection->cram_hmac_tfm);
2517 crypto_free_shash(connection->integrity_tfm);
2518 crypto_free_shash(connection->peer_integrity_tfm);
2519 kfree(connection->int_dig_in);
2520 kfree(connection->int_dig_vv);
2522 connection->csums_tfm = NULL;
2523 connection->verify_tfm = NULL;
2524 connection->cram_hmac_tfm = NULL;
2525 connection->integrity_tfm = NULL;
2526 connection->peer_integrity_tfm = NULL;
2527 connection->int_dig_in = NULL;
2528 connection->int_dig_vv = NULL;
2531 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2533 struct drbd_connection *connection;
2534 cpumask_var_t new_cpu_mask;
2537 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2540 /* silently ignore cpu mask on UP kernel */
2541 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2542 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2543 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2544 if (err == -EOVERFLOW) {
2545 /* So what. mask it out. */
2546 cpumask_var_t tmp_cpu_mask;
2547 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2548 cpumask_setall(tmp_cpu_mask);
2549 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2550 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2552 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2554 free_cpumask_var(tmp_cpu_mask);
2559 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2560 /* retcode = ERR_CPU_MASK_PARSE; */
2564 resource->res_opts = *res_opts;
2565 if (cpumask_empty(new_cpu_mask))
2566 drbd_calc_cpu_mask(&new_cpu_mask);
2567 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2568 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2569 for_each_connection_rcu(connection, resource) {
2570 connection->receiver.reset_cpu_mask = 1;
2571 connection->ack_receiver.reset_cpu_mask = 1;
2572 connection->worker.reset_cpu_mask = 1;
2578 free_cpumask_var(new_cpu_mask);
2583 struct drbd_resource *drbd_create_resource(const char *name)
2585 struct drbd_resource *resource;
2587 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2590 resource->name = kstrdup(name, GFP_KERNEL);
2591 if (!resource->name)
2592 goto fail_free_resource;
2593 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2594 goto fail_free_name;
2595 kref_init(&resource->kref);
2596 idr_init(&resource->devices);
2597 INIT_LIST_HEAD(&resource->connections);
2598 resource->write_ordering = WO_BDEV_FLUSH;
2599 list_add_tail_rcu(&resource->resources, &drbd_resources);
2600 mutex_init(&resource->conf_update);
2601 mutex_init(&resource->adm_mutex);
2602 spin_lock_init(&resource->req_lock);
2603 drbd_debugfs_resource_add(resource);
2607 kfree(resource->name);
2614 /* caller must be under adm_mutex */
2615 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2617 struct drbd_resource *resource;
2618 struct drbd_connection *connection;
2620 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2624 if (drbd_alloc_socket(&connection->data))
2626 if (drbd_alloc_socket(&connection->meta))
2629 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2630 if (!connection->current_epoch)
2633 INIT_LIST_HEAD(&connection->transfer_log);
2635 INIT_LIST_HEAD(&connection->current_epoch->list);
2636 connection->epochs = 1;
2637 spin_lock_init(&connection->epoch_lock);
2639 connection->send.seen_any_write_yet = false;
2640 connection->send.current_epoch_nr = 0;
2641 connection->send.current_epoch_writes = 0;
2643 resource = drbd_create_resource(name);
2647 connection->cstate = C_STANDALONE;
2648 mutex_init(&connection->cstate_mutex);
2649 init_waitqueue_head(&connection->ping_wait);
2650 idr_init(&connection->peer_devices);
2652 drbd_init_workqueue(&connection->sender_work);
2653 mutex_init(&connection->data.mutex);
2654 mutex_init(&connection->meta.mutex);
2656 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2657 connection->receiver.connection = connection;
2658 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2659 connection->worker.connection = connection;
2660 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2661 connection->ack_receiver.connection = connection;
2663 kref_init(&connection->kref);
2665 connection->resource = resource;
2667 if (set_resource_options(resource, res_opts))
2670 kref_get(&resource->kref);
2671 list_add_tail_rcu(&connection->connections, &resource->connections);
2672 drbd_debugfs_connection_add(connection);
2676 list_del(&resource->resources);
2677 drbd_free_resource(resource);
2679 kfree(connection->current_epoch);
2680 drbd_free_socket(&connection->meta);
2681 drbd_free_socket(&connection->data);
2686 void drbd_destroy_connection(struct kref *kref)
2688 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2689 struct drbd_resource *resource = connection->resource;
2691 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
2692 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2693 kfree(connection->current_epoch);
2695 idr_destroy(&connection->peer_devices);
2697 drbd_free_socket(&connection->meta);
2698 drbd_free_socket(&connection->data);
2699 kfree(connection->int_dig_in);
2700 kfree(connection->int_dig_vv);
2701 memset(connection, 0xfc, sizeof(*connection));
2703 kref_put(&resource->kref, drbd_destroy_resource);
2706 static int init_submitter(struct drbd_device *device)
2708 /* opencoded create_singlethread_workqueue(),
2709 * to be able to say "drbd%d", ..., minor */
2711 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2712 if (!device->submit.wq)
2715 INIT_WORK(&device->submit.worker, do_submit);
2716 INIT_LIST_HEAD(&device->submit.writes);
2720 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2722 struct drbd_resource *resource = adm_ctx->resource;
2723 struct drbd_connection *connection;
2724 struct drbd_device *device;
2725 struct drbd_peer_device *peer_device, *tmp_peer_device;
2726 struct gendisk *disk;
2727 struct request_queue *q;
2729 int vnr = adm_ctx->volume;
2730 enum drbd_ret_code err = ERR_NOMEM;
2732 device = minor_to_device(minor);
2734 return ERR_MINOR_OR_VOLUME_EXISTS;
2736 /* GFP_KERNEL, we are outside of all write-out paths */
2737 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2740 kref_init(&device->kref);
2742 kref_get(&resource->kref);
2743 device->resource = resource;
2744 device->minor = minor;
2747 drbd_init_set_defaults(device);
2749 q = blk_alloc_queue(NUMA_NO_NODE);
2752 device->rq_queue = q;
2754 disk = alloc_disk(1);
2757 device->vdisk = disk;
2759 set_disk_ro(disk, true);
2762 disk->major = DRBD_MAJOR;
2763 disk->first_minor = minor;
2764 disk->fops = &drbd_ops;
2765 sprintf(disk->disk_name, "drbd%d", minor);
2766 disk->private_data = device;
2768 device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2769 /* we have no partitions. we contain only ourselves. */
2770 device->this_bdev->bd_contains = device->this_bdev;
2772 blk_queue_write_cache(q, true, true);
2773 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2774 This triggers a max_bio_size message upon first attach or connect */
2775 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2777 device->md_io.page = alloc_page(GFP_KERNEL);
2778 if (!device->md_io.page)
2779 goto out_no_io_page;
2781 if (drbd_bm_init(device))
2783 device->read_requests = RB_ROOT;
2784 device->write_requests = RB_ROOT;
2786 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2789 err = ERR_MINOR_OR_VOLUME_EXISTS;
2790 goto out_no_minor_idr;
2792 kref_get(&device->kref);
2794 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2797 err = ERR_MINOR_OR_VOLUME_EXISTS;
2798 goto out_idr_remove_minor;
2800 kref_get(&device->kref);
2802 INIT_LIST_HEAD(&device->peer_devices);
2803 INIT_LIST_HEAD(&device->pending_bitmap_io);
2804 for_each_connection(connection, resource) {
2805 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2807 goto out_idr_remove_from_resource;
2808 peer_device->connection = connection;
2809 peer_device->device = device;
2811 list_add(&peer_device->peer_devices, &device->peer_devices);
2812 kref_get(&device->kref);
2814 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2817 err = ERR_INVALID_REQUEST;
2818 goto out_idr_remove_from_resource;
2820 kref_get(&connection->kref);
2821 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2824 if (init_submitter(device)) {
2826 goto out_idr_remove_vol;
2831 /* inherit the connection state */
2832 device->state.conn = first_connection(resource)->cstate;
2833 if (device->state.conn == C_WF_REPORT_PARAMS) {
2834 for_each_peer_device(peer_device, device)
2835 drbd_connected(peer_device);
2837 /* move to create_peer_device() */
2838 for_each_peer_device(peer_device, device)
2839 drbd_debugfs_peer_device_add(peer_device);
2840 drbd_debugfs_device_add(device);
2844 idr_remove(&connection->peer_devices, vnr);
2845 out_idr_remove_from_resource:
2846 for_each_connection(connection, resource) {
2847 peer_device = idr_remove(&connection->peer_devices, vnr);
2849 kref_put(&connection->kref, drbd_destroy_connection);
2851 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2852 list_del(&peer_device->peer_devices);
2855 idr_remove(&resource->devices, vnr);
2856 out_idr_remove_minor:
2857 idr_remove(&drbd_devices, minor);
2860 drbd_bm_cleanup(device);
2862 __free_page(device->md_io.page);
2866 blk_cleanup_queue(q);
2868 kref_put(&resource->kref, drbd_destroy_resource);
2873 void drbd_delete_device(struct drbd_device *device)
2875 struct drbd_resource *resource = device->resource;
2876 struct drbd_connection *connection;
2877 struct drbd_peer_device *peer_device;
2879 /* move to free_peer_device() */
2880 for_each_peer_device(peer_device, device)
2881 drbd_debugfs_peer_device_cleanup(peer_device);
2882 drbd_debugfs_device_cleanup(device);
2883 for_each_connection(connection, resource) {
2884 idr_remove(&connection->peer_devices, device->vnr);
2885 kref_put(&device->kref, drbd_destroy_device);
2887 idr_remove(&resource->devices, device->vnr);
2888 kref_put(&device->kref, drbd_destroy_device);
2889 idr_remove(&drbd_devices, device_to_minor(device));
2890 kref_put(&device->kref, drbd_destroy_device);
2891 del_gendisk(device->vdisk);
2893 kref_put(&device->kref, drbd_destroy_device);
2896 static int __init drbd_init(void)
2900 if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2901 pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2905 drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2909 err = register_blkdev(DRBD_MAJOR, "drbd");
2911 pr_err("unable to register block device major %d\n",
2917 * allocate all necessary structs
2919 init_waitqueue_head(&drbd_pp_wait);
2921 drbd_proc = NULL; /* play safe for drbd_cleanup */
2922 idr_init(&drbd_devices);
2924 mutex_init(&resources_mutex);
2925 INIT_LIST_HEAD(&drbd_resources);
2927 err = drbd_genl_register();
2929 pr_err("unable to register generic netlink family\n");
2933 err = drbd_create_mempools();
2938 drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show);
2940 pr_err("unable to register proc file\n");
2944 retry.wq = create_singlethread_workqueue("drbd-reissue");
2946 pr_err("unable to create retry workqueue\n");
2949 INIT_WORK(&retry.worker, do_retry);
2950 spin_lock_init(&retry.lock);
2951 INIT_LIST_HEAD(&retry.writes);
2953 drbd_debugfs_init();
2955 pr_info("initialized. "
2956 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2957 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2958 pr_info("%s\n", drbd_buildtag());
2959 pr_info("registered as block device major %d\n", DRBD_MAJOR);
2960 return 0; /* Success! */
2965 pr_err("ran out of memory\n");
2967 pr_err("initialization failure\n");
2971 static void drbd_free_one_sock(struct drbd_socket *ds)
2974 mutex_lock(&ds->mutex);
2977 mutex_unlock(&ds->mutex);
2979 /* so debugfs does not need to mutex_lock() */
2981 kernel_sock_shutdown(s, SHUT_RDWR);
2986 void drbd_free_sock(struct drbd_connection *connection)
2988 if (connection->data.socket)
2989 drbd_free_one_sock(&connection->data);
2990 if (connection->meta.socket)
2991 drbd_free_one_sock(&connection->meta);
2994 /* meta data management */
2996 void conn_md_sync(struct drbd_connection *connection)
2998 struct drbd_peer_device *peer_device;
3002 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3003 struct drbd_device *device = peer_device->device;
3005 kref_get(&device->kref);
3007 drbd_md_sync(device);
3008 kref_put(&device->kref, drbd_destroy_device);
3014 /* aligned 4kByte */
3015 struct meta_data_on_disk {
3016 u64 la_size_sect; /* last agreed size. */
3017 u64 uuid[UI_SIZE]; /* UUIDs. */
3020 u32 flags; /* MDF */
3023 u32 al_offset; /* offset to this block */
3024 u32 al_nr_extents; /* important for restoring the AL (userspace) */
3025 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3026 u32 bm_offset; /* offset to the bitmap, from here */
3027 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
3028 u32 la_peer_max_bio_size; /* last peer max_bio_size */
3030 /* see al_tr_number_to_on_disk_sector() */
3032 u32 al_stripe_size_4k;
3034 u8 reserved_u8[4096 - (7*8 + 10*4)];
3039 void drbd_md_write(struct drbd_device *device, void *b)
3041 struct meta_data_on_disk *buffer = b;
3045 memset(buffer, 0, sizeof(*buffer));
3047 buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3048 for (i = UI_CURRENT; i < UI_SIZE; i++)
3049 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3050 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3051 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3053 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3054 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3055 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3056 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3057 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3059 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3060 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3062 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3063 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3065 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3066 sector = device->ldev->md.md_offset;
3068 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3069 /* this was a try anyways ... */
3070 drbd_err(device, "meta data update failed!\n");
3071 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3076 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3077 * @device: DRBD device.
3079 void drbd_md_sync(struct drbd_device *device)
3081 struct meta_data_on_disk *buffer;
3083 /* Don't accidentally change the DRBD meta data layout. */
3084 BUILD_BUG_ON(UI_SIZE != 4);
3085 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3087 del_timer(&device->md_sync_timer);
3088 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3089 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3092 /* We use here D_FAILED and not D_ATTACHING because we try to write
3093 * metadata even if we detach due to a disk failure! */
3094 if (!get_ldev_if_state(device, D_FAILED))
3097 buffer = drbd_md_get_buffer(device, __func__);
3101 drbd_md_write(device, buffer);
3103 /* Update device->ldev->md.la_size_sect,
3104 * since we updated it on metadata. */
3105 device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3107 drbd_md_put_buffer(device);
3112 static int check_activity_log_stripe_size(struct drbd_device *device,
3113 struct meta_data_on_disk *on_disk,
3114 struct drbd_md *in_core)
3116 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3117 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3120 /* both not set: default to old fixed size activity log */
3121 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3123 al_stripe_size_4k = MD_32kB_SECT/8;
3126 /* some paranoia plausibility checks */
3128 /* we need both values to be set */
3129 if (al_stripes == 0 || al_stripe_size_4k == 0)
3132 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3134 /* Upper limit of activity log area, to avoid potential overflow
3135 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3136 * than 72 * 4k blocks total only increases the amount of history,
3137 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3138 if (al_size_4k > (16 * 1024 * 1024/4))
3141 /* Lower limit: we need at least 8 transaction slots (32kB)
3142 * to not break existing setups */
3143 if (al_size_4k < MD_32kB_SECT/8)
3146 in_core->al_stripe_size_4k = al_stripe_size_4k;
3147 in_core->al_stripes = al_stripes;
3148 in_core->al_size_4k = al_size_4k;
3152 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3153 al_stripes, al_stripe_size_4k);
3157 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3159 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3160 struct drbd_md *in_core = &bdev->md;
3161 s32 on_disk_al_sect;
3162 s32 on_disk_bm_sect;
3164 /* The on-disk size of the activity log, calculated from offsets, and
3165 * the size of the activity log calculated from the stripe settings,
3167 * Though we could relax this a bit: it is ok, if the striped activity log
3168 * fits in the available on-disk activity log size.
3169 * Right now, that would break how resize is implemented.
3170 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3171 * of possible unused padding space in the on disk layout. */
3172 if (in_core->al_offset < 0) {
3173 if (in_core->bm_offset > in_core->al_offset)
3175 on_disk_al_sect = -in_core->al_offset;
3176 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3178 if (in_core->al_offset != MD_4kB_SECT)
3180 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3183 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3184 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3187 /* old fixed size meta data is exactly that: fixed. */
3188 if (in_core->meta_dev_idx >= 0) {
3189 if (in_core->md_size_sect != MD_128MB_SECT
3190 || in_core->al_offset != MD_4kB_SECT
3191 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3192 || in_core->al_stripes != 1
3193 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3197 if (capacity < in_core->md_size_sect)
3199 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3202 /* should be aligned, and at least 32k */
3203 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3206 /* should fit (for now: exactly) into the available on-disk space;
3207 * overflow prevention is in check_activity_log_stripe_size() above. */
3208 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3211 /* again, should be aligned */
3212 if (in_core->bm_offset & 7)
3215 /* FIXME check for device grow with flex external meta data? */
3217 /* can the available bitmap space cover the last agreed device size? */
3218 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3224 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3225 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3226 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3227 in_core->meta_dev_idx,
3228 in_core->al_stripes, in_core->al_stripe_size_4k,
3229 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3230 (unsigned long long)in_core->la_size_sect,
3231 (unsigned long long)capacity);
3238 * drbd_md_read() - Reads in the meta data super block
3239 * @device: DRBD device.
3240 * @bdev: Device from which the meta data should be read in.
3242 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3243 * something goes wrong.
3245 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3246 * even before @bdev is assigned to @device->ldev.
3248 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3250 struct meta_data_on_disk *buffer;
3252 int i, rv = NO_ERROR;
3254 if (device->state.disk != D_DISKLESS)
3255 return ERR_DISK_CONFIGURED;
3257 buffer = drbd_md_get_buffer(device, __func__);
3261 /* First, figure out where our meta data superblock is located,
3263 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3264 bdev->md.md_offset = drbd_md_ss(bdev);
3265 /* Even for (flexible or indexed) external meta data,
3266 * initially restrict us to the 4k superblock for now.
3267 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3268 bdev->md.md_size_sect = 8;
3270 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3272 /* NOTE: can't do normal error processing here as this is
3273 called BEFORE disk is attached */
3274 drbd_err(device, "Error while reading metadata.\n");
3275 rv = ERR_IO_MD_DISK;
3279 magic = be32_to_cpu(buffer->magic);
3280 flags = be32_to_cpu(buffer->flags);
3281 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3282 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3283 /* btw: that's Activity Log clean, not "all" clean. */
3284 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3285 rv = ERR_MD_UNCLEAN;
3289 rv = ERR_MD_INVALID;
3290 if (magic != DRBD_MD_MAGIC_08) {
3291 if (magic == DRBD_MD_MAGIC_07)
3292 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3294 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3298 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3299 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3300 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3305 /* convert to in_core endian */
3306 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3307 for (i = UI_CURRENT; i < UI_SIZE; i++)
3308 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3309 bdev->md.flags = be32_to_cpu(buffer->flags);
3310 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3312 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3313 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3314 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3316 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3318 if (check_offsets_and_sizes(device, bdev))
3321 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3322 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3323 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3326 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3327 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3328 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3334 spin_lock_irq(&device->resource->req_lock);
3335 if (device->state.conn < C_CONNECTED) {
3337 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3338 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3339 device->peer_max_bio_size = peer;
3341 spin_unlock_irq(&device->resource->req_lock);
3344 drbd_md_put_buffer(device);
3350 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3351 * @device: DRBD device.
3353 * Call this function if you change anything that should be written to
3354 * the meta-data super block. This function sets MD_DIRTY, and starts a
3355 * timer that ensures that within five seconds you have to call drbd_md_sync().
3357 void drbd_md_mark_dirty(struct drbd_device *device)
3359 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3360 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3363 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3367 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3368 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3371 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3373 if (idx == UI_CURRENT) {
3374 if (device->state.role == R_PRIMARY)
3379 drbd_set_ed_uuid(device, val);
3382 device->ldev->md.uuid[idx] = val;
3383 drbd_md_mark_dirty(device);
3386 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3388 unsigned long flags;
3389 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3390 __drbd_uuid_set(device, idx, val);
3391 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3394 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3396 unsigned long flags;
3397 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3398 if (device->ldev->md.uuid[idx]) {
3399 drbd_uuid_move_history(device);
3400 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3402 __drbd_uuid_set(device, idx, val);
3403 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3407 * drbd_uuid_new_current() - Creates a new current UUID
3408 * @device: DRBD device.
3410 * Creates a new current UUID, and rotates the old current UUID into
3411 * the bitmap slot. Causes an incremental resync upon next connect.
3413 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3416 unsigned long long bm_uuid;
3418 get_random_bytes(&val, sizeof(u64));
3420 spin_lock_irq(&device->ldev->md.uuid_lock);
3421 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3424 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3426 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3427 __drbd_uuid_set(device, UI_CURRENT, val);
3428 spin_unlock_irq(&device->ldev->md.uuid_lock);
3430 drbd_print_uuids(device, "new current UUID");
3431 /* get it to stable storage _now_ */
3432 drbd_md_sync(device);
3435 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3437 unsigned long flags;
3438 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3441 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3443 drbd_uuid_move_history(device);
3444 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3445 device->ldev->md.uuid[UI_BITMAP] = 0;
3447 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3449 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3451 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3453 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3455 drbd_md_mark_dirty(device);
3459 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3460 * @device: DRBD device.
3462 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3464 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3468 drbd_md_set_flag(device, MDF_FULL_SYNC);
3469 drbd_md_sync(device);
3470 drbd_bm_set_all(device);
3472 rv = drbd_bm_write(device);
3475 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3476 drbd_md_sync(device);
3483 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3484 * @device: DRBD device.
3486 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3488 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3490 drbd_resume_al(device);
3491 drbd_bm_clear_all(device);
3492 return drbd_bm_write(device);
3495 static int w_bitmap_io(struct drbd_work *w, int unused)
3497 struct drbd_device *device =
3498 container_of(w, struct drbd_device, bm_io_work.w);
3499 struct bm_io_work *work = &device->bm_io_work;
3502 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3503 int cnt = atomic_read(&device->ap_bio_cnt);
3505 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3509 if (get_ldev(device)) {
3510 drbd_bm_lock(device, work->why, work->flags);
3511 rv = work->io_fn(device);
3512 drbd_bm_unlock(device);
3516 clear_bit_unlock(BITMAP_IO, &device->flags);
3517 wake_up(&device->misc_wait);
3520 work->done(device, rv);
3522 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3530 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3531 * @device: DRBD device.
3532 * @io_fn: IO callback to be called when bitmap IO is possible
3533 * @done: callback to be called after the bitmap IO was performed
3534 * @why: Descriptive text of the reason for doing the IO
3536 * While IO on the bitmap happens we freeze application IO thus we ensure
3537 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3538 * called from worker context. It MUST NOT be used while a previous such
3539 * work is still pending!
3541 * Its worker function encloses the call of io_fn() by get_ldev() and
3544 void drbd_queue_bitmap_io(struct drbd_device *device,
3545 int (*io_fn)(struct drbd_device *),
3546 void (*done)(struct drbd_device *, int),
3547 char *why, enum bm_flag flags)
3549 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3551 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3552 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3553 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3554 if (device->bm_io_work.why)
3555 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3556 why, device->bm_io_work.why);
3558 device->bm_io_work.io_fn = io_fn;
3559 device->bm_io_work.done = done;
3560 device->bm_io_work.why = why;
3561 device->bm_io_work.flags = flags;
3563 spin_lock_irq(&device->resource->req_lock);
3564 set_bit(BITMAP_IO, &device->flags);
3565 /* don't wait for pending application IO if the caller indicates that
3566 * application IO does not conflict anyways. */
3567 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3568 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3569 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3570 &device->bm_io_work.w);
3572 spin_unlock_irq(&device->resource->req_lock);
3576 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3577 * @device: DRBD device.
3578 * @io_fn: IO callback to be called when bitmap IO is possible
3579 * @why: Descriptive text of the reason for doing the IO
3581 * freezes application IO while that the actual IO operations runs. This
3582 * functions MAY NOT be called from worker context.
3584 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3585 char *why, enum bm_flag flags)
3587 /* Only suspend io, if some operation is supposed to be locked out */
3588 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3591 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3594 drbd_suspend_io(device);
3596 drbd_bm_lock(device, why, flags);
3598 drbd_bm_unlock(device);
3601 drbd_resume_io(device);
3606 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3608 if ((device->ldev->md.flags & flag) != flag) {
3609 drbd_md_mark_dirty(device);
3610 device->ldev->md.flags |= flag;
3614 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3616 if ((device->ldev->md.flags & flag) != 0) {
3617 drbd_md_mark_dirty(device);
3618 device->ldev->md.flags &= ~flag;
3621 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3623 return (bdev->md.flags & flag) != 0;
3626 static void md_sync_timer_fn(struct timer_list *t)
3628 struct drbd_device *device = from_timer(device, t, md_sync_timer);
3629 drbd_device_post_work(device, MD_SYNC);
3632 const char *cmdname(enum drbd_packet cmd)
3634 /* THINK may need to become several global tables
3635 * when we want to support more than
3636 * one PRO_VERSION */
3637 static const char *cmdnames[] = {
3639 [P_WSAME] = "WriteSame",
3641 [P_DATA_REPLY] = "DataReply",
3642 [P_RS_DATA_REPLY] = "RSDataReply",
3643 [P_BARRIER] = "Barrier",
3644 [P_BITMAP] = "ReportBitMap",
3645 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3646 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3647 [P_UNPLUG_REMOTE] = "UnplugRemote",
3648 [P_DATA_REQUEST] = "DataRequest",
3649 [P_RS_DATA_REQUEST] = "RSDataRequest",
3650 [P_SYNC_PARAM] = "SyncParam",
3651 [P_SYNC_PARAM89] = "SyncParam89",
3652 [P_PROTOCOL] = "ReportProtocol",
3653 [P_UUIDS] = "ReportUUIDs",
3654 [P_SIZES] = "ReportSizes",
3655 [P_STATE] = "ReportState",
3656 [P_SYNC_UUID] = "ReportSyncUUID",
3657 [P_AUTH_CHALLENGE] = "AuthChallenge",
3658 [P_AUTH_RESPONSE] = "AuthResponse",
3660 [P_PING_ACK] = "PingAck",
3661 [P_RECV_ACK] = "RecvAck",
3662 [P_WRITE_ACK] = "WriteAck",
3663 [P_RS_WRITE_ACK] = "RSWriteAck",
3664 [P_SUPERSEDED] = "Superseded",
3665 [P_NEG_ACK] = "NegAck",
3666 [P_NEG_DREPLY] = "NegDReply",
3667 [P_NEG_RS_DREPLY] = "NegRSDReply",
3668 [P_BARRIER_ACK] = "BarrierAck",
3669 [P_STATE_CHG_REQ] = "StateChgRequest",
3670 [P_STATE_CHG_REPLY] = "StateChgReply",
3671 [P_OV_REQUEST] = "OVRequest",
3672 [P_OV_REPLY] = "OVReply",
3673 [P_OV_RESULT] = "OVResult",
3674 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3675 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3676 [P_COMPRESSED_BITMAP] = "CBitmap",
3677 [P_DELAY_PROBE] = "DelayProbe",
3678 [P_OUT_OF_SYNC] = "OutOfSync",
3679 [P_RETRY_WRITE] = "RetryWrite",
3680 [P_RS_CANCEL] = "RSCancel",
3681 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3682 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3683 [P_RETRY_WRITE] = "retry_write",
3684 [P_PROTOCOL_UPDATE] = "protocol_update",
3685 [P_RS_THIN_REQ] = "rs_thin_req",
3686 [P_RS_DEALLOCATED] = "rs_deallocated",
3688 /* enum drbd_packet, but not commands - obsoleted flags:
3694 /* too big for the array: 0xfffX */
3695 if (cmd == P_INITIAL_META)
3696 return "InitialMeta";
3697 if (cmd == P_INITIAL_DATA)
3698 return "InitialData";
3699 if (cmd == P_CONNECTION_FEATURES)
3700 return "ConnectionFeatures";
3701 if (cmd >= ARRAY_SIZE(cmdnames))
3703 return cmdnames[cmd];
3707 * drbd_wait_misc - wait for a request to make progress
3708 * @device: device associated with the request
3709 * @i: the struct drbd_interval embedded in struct drbd_request or
3710 * struct drbd_peer_request
3712 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3714 struct net_conf *nc;
3719 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3724 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3727 /* Indicate to wake up device->misc_wait on progress. */
3729 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3730 spin_unlock_irq(&device->resource->req_lock);
3731 timeout = schedule_timeout(timeout);
3732 finish_wait(&device->misc_wait, &wait);
3733 spin_lock_irq(&device->resource->req_lock);
3734 if (!timeout || device->state.conn < C_CONNECTED)
3736 if (signal_pending(current))
3737 return -ERESTARTSYS;
3741 void lock_all_resources(void)
3743 struct drbd_resource *resource;
3744 int __maybe_unused i = 0;
3746 mutex_lock(&resources_mutex);
3747 local_irq_disable();
3748 for_each_resource(resource, &drbd_resources)
3749 spin_lock_nested(&resource->req_lock, i++);
3752 void unlock_all_resources(void)
3754 struct drbd_resource *resource;
3756 for_each_resource(resource, &drbd_resources)
3757 spin_unlock(&resource->req_lock);
3759 mutex_unlock(&resources_mutex);
3762 #ifdef CONFIG_DRBD_FAULT_INJECTION
3763 /* Fault insertion support including random number generator shamelessly
3764 * stolen from kernel/rcutorture.c */
3765 struct fault_random_state {
3766 unsigned long state;
3767 unsigned long count;
3770 #define FAULT_RANDOM_MULT 39916801 /* prime */
3771 #define FAULT_RANDOM_ADD 479001701 /* prime */
3772 #define FAULT_RANDOM_REFRESH 10000
3775 * Crude but fast random-number generator. Uses a linear congruential
3776 * generator, with occasional help from get_random_bytes().
3778 static unsigned long
3779 _drbd_fault_random(struct fault_random_state *rsp)
3783 if (!rsp->count--) {
3784 get_random_bytes(&refresh, sizeof(refresh));
3785 rsp->state += refresh;
3786 rsp->count = FAULT_RANDOM_REFRESH;
3788 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3789 return swahw32(rsp->state);
3793 _drbd_fault_str(unsigned int type) {
3794 static char *_faults[] = {
3795 [DRBD_FAULT_MD_WR] = "Meta-data write",
3796 [DRBD_FAULT_MD_RD] = "Meta-data read",
3797 [DRBD_FAULT_RS_WR] = "Resync write",
3798 [DRBD_FAULT_RS_RD] = "Resync read",
3799 [DRBD_FAULT_DT_WR] = "Data write",
3800 [DRBD_FAULT_DT_RD] = "Data read",
3801 [DRBD_FAULT_DT_RA] = "Data read ahead",
3802 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3803 [DRBD_FAULT_AL_EE] = "EE allocation",
3804 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3807 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3811 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3813 static struct fault_random_state rrs = {0, 0};
3815 unsigned int ret = (
3816 (drbd_fault_devs == 0 ||
3817 ((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3818 (((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3823 if (__ratelimit(&drbd_ratelimit_state))
3824 drbd_warn(device, "***Simulating %s failure\n",
3825 _drbd_fault_str(type));
3832 const char *drbd_buildtag(void)
3834 /* DRBD built from external sources has here a reference to the
3835 git hash of the source code. */
3837 static char buildtag[38] = "\0uilt-in";
3839 if (buildtag[0] == 0) {
3841 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3850 module_init(drbd_init)
3851 module_exit(drbd_cleanup)
3853 EXPORT_SYMBOL(drbd_conn_str);
3854 EXPORT_SYMBOL(drbd_role_str);
3855 EXPORT_SYMBOL(drbd_disk_str);
3856 EXPORT_SYMBOL(drbd_set_st_err_str);