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, *tmp = 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) {
245 req = list_prepare_entry(tmp, &connection->transfer_log, tl_requests);
246 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
247 if (req->epoch != expect_epoch)
249 _req_mod(req, BARRIER_ACKED);
251 spin_unlock_irq(&connection->resource->req_lock);
256 spin_unlock_irq(&connection->resource->req_lock);
257 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
262 * _tl_restart() - Walks the transfer log, and applies an action to all requests
263 * @connection: DRBD connection to operate on.
264 * @what: The action/event to perform with all request objects
266 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
267 * RESTART_FROZEN_DISK_IO.
269 /* must hold resource->req_lock */
270 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
272 struct drbd_request *req, *r;
274 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
278 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
280 spin_lock_irq(&connection->resource->req_lock);
281 _tl_restart(connection, what);
282 spin_unlock_irq(&connection->resource->req_lock);
286 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
287 * @device: DRBD device.
289 * This is called after the connection to the peer was lost. The storage covered
290 * by the requests on the transfer gets marked as our of sync. Called from the
291 * receiver thread and the worker thread.
293 void tl_clear(struct drbd_connection *connection)
295 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
299 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
300 * @device: DRBD device.
302 void tl_abort_disk_io(struct drbd_device *device)
304 struct drbd_connection *connection = first_peer_device(device)->connection;
305 struct drbd_request *req, *r;
307 spin_lock_irq(&connection->resource->req_lock);
308 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
309 if (!(req->rq_state & RQ_LOCAL_PENDING))
311 if (req->device != device)
313 _req_mod(req, ABORT_DISK_IO);
315 spin_unlock_irq(&connection->resource->req_lock);
318 static int drbd_thread_setup(void *arg)
320 struct drbd_thread *thi = (struct drbd_thread *) arg;
321 struct drbd_resource *resource = thi->resource;
325 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
329 allow_kernel_signal(DRBD_SIGKILL);
330 allow_kernel_signal(SIGXCPU);
332 retval = thi->function(thi);
334 spin_lock_irqsave(&thi->t_lock, flags);
336 /* if the receiver has been "EXITING", the last thing it did
337 * was set the conn state to "StandAlone",
338 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
339 * and receiver thread will be "started".
340 * drbd_thread_start needs to set "RESTARTING" in that case.
341 * t_state check and assignment needs to be within the same spinlock,
342 * so either thread_start sees EXITING, and can remap to RESTARTING,
343 * or thread_start see NONE, and can proceed as normal.
346 if (thi->t_state == RESTARTING) {
347 drbd_info(resource, "Restarting %s thread\n", thi->name);
348 thi->t_state = RUNNING;
349 spin_unlock_irqrestore(&thi->t_lock, flags);
356 complete_all(&thi->stop);
357 spin_unlock_irqrestore(&thi->t_lock, flags);
359 drbd_info(resource, "Terminating %s\n", current->comm);
361 /* Release mod reference taken when thread was started */
364 kref_put(&thi->connection->kref, drbd_destroy_connection);
365 kref_put(&resource->kref, drbd_destroy_resource);
366 module_put(THIS_MODULE);
370 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
371 int (*func) (struct drbd_thread *), const char *name)
373 spin_lock_init(&thi->t_lock);
376 thi->function = func;
377 thi->resource = resource;
378 thi->connection = NULL;
382 int drbd_thread_start(struct drbd_thread *thi)
384 struct drbd_resource *resource = thi->resource;
385 struct task_struct *nt;
388 /* is used from state engine doing drbd_thread_stop_nowait,
389 * while holding the req lock irqsave */
390 spin_lock_irqsave(&thi->t_lock, flags);
392 switch (thi->t_state) {
394 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
395 thi->name, current->comm, current->pid);
397 /* Get ref on module for thread - this is released when thread exits */
398 if (!try_module_get(THIS_MODULE)) {
399 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
400 spin_unlock_irqrestore(&thi->t_lock, flags);
404 kref_get(&resource->kref);
406 kref_get(&thi->connection->kref);
408 init_completion(&thi->stop);
409 thi->reset_cpu_mask = 1;
410 thi->t_state = RUNNING;
411 spin_unlock_irqrestore(&thi->t_lock, flags);
412 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
414 nt = kthread_create(drbd_thread_setup, (void *) thi,
415 "drbd_%c_%s", thi->name[0], thi->resource->name);
418 drbd_err(resource, "Couldn't start thread\n");
421 kref_put(&thi->connection->kref, drbd_destroy_connection);
422 kref_put(&resource->kref, drbd_destroy_resource);
423 module_put(THIS_MODULE);
426 spin_lock_irqsave(&thi->t_lock, flags);
428 thi->t_state = RUNNING;
429 spin_unlock_irqrestore(&thi->t_lock, flags);
433 thi->t_state = RESTARTING;
434 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
435 thi->name, current->comm, current->pid);
440 spin_unlock_irqrestore(&thi->t_lock, flags);
448 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
452 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
454 /* may be called from state engine, holding the req lock irqsave */
455 spin_lock_irqsave(&thi->t_lock, flags);
457 if (thi->t_state == NONE) {
458 spin_unlock_irqrestore(&thi->t_lock, flags);
460 drbd_thread_start(thi);
464 if (thi->t_state != ns) {
465 if (thi->task == NULL) {
466 spin_unlock_irqrestore(&thi->t_lock, flags);
472 init_completion(&thi->stop);
473 if (thi->task != current)
474 send_sig(DRBD_SIGKILL, thi->task, 1);
477 spin_unlock_irqrestore(&thi->t_lock, flags);
480 wait_for_completion(&thi->stop);
483 int conn_lowest_minor(struct drbd_connection *connection)
485 struct drbd_peer_device *peer_device;
486 int vnr = 0, minor = -1;
489 peer_device = idr_get_next(&connection->peer_devices, &vnr);
491 minor = device_to_minor(peer_device->device);
499 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
501 * Forces all threads of a resource onto the same CPU. This is beneficial for
502 * DRBD's performance. May be overwritten by user's configuration.
504 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
506 unsigned int *resources_per_cpu, min_index = ~0;
508 resources_per_cpu = kcalloc(nr_cpu_ids, sizeof(*resources_per_cpu),
510 if (resources_per_cpu) {
511 struct drbd_resource *resource;
512 unsigned int cpu, min = ~0;
515 for_each_resource_rcu(resource, &drbd_resources) {
516 for_each_cpu(cpu, resource->cpu_mask)
517 resources_per_cpu[cpu]++;
520 for_each_online_cpu(cpu) {
521 if (resources_per_cpu[cpu] < min) {
522 min = resources_per_cpu[cpu];
526 kfree(resources_per_cpu);
528 if (min_index == ~0) {
529 cpumask_setall(*cpu_mask);
532 cpumask_set_cpu(min_index, *cpu_mask);
536 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
537 * @device: DRBD device.
538 * @thi: drbd_thread object
540 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
543 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
545 struct drbd_resource *resource = thi->resource;
546 struct task_struct *p = current;
548 if (!thi->reset_cpu_mask)
550 thi->reset_cpu_mask = 0;
551 set_cpus_allowed_ptr(p, resource->cpu_mask);
554 #define drbd_calc_cpu_mask(A) ({})
558 * drbd_header_size - size of a packet header
560 * The header size is a multiple of 8, so any payload following the header is
561 * word aligned on 64-bit architectures. (The bitmap send and receive code
564 unsigned int drbd_header_size(struct drbd_connection *connection)
566 if (connection->agreed_pro_version >= 100) {
567 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
568 return sizeof(struct p_header100);
570 BUILD_BUG_ON(sizeof(struct p_header80) !=
571 sizeof(struct p_header95));
572 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
573 return sizeof(struct p_header80);
577 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
579 h->magic = cpu_to_be32(DRBD_MAGIC);
580 h->command = cpu_to_be16(cmd);
581 h->length = cpu_to_be16(size);
582 return sizeof(struct p_header80);
585 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
587 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
588 h->command = cpu_to_be16(cmd);
589 h->length = cpu_to_be32(size);
590 return sizeof(struct p_header95);
593 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
596 h->magic = cpu_to_be32(DRBD_MAGIC_100);
597 h->volume = cpu_to_be16(vnr);
598 h->command = cpu_to_be16(cmd);
599 h->length = cpu_to_be32(size);
601 return sizeof(struct p_header100);
604 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
605 void *buffer, enum drbd_packet cmd, int size)
607 if (connection->agreed_pro_version >= 100)
608 return prepare_header100(buffer, cmd, size, vnr);
609 else if (connection->agreed_pro_version >= 95 &&
610 size > DRBD_MAX_SIZE_H80_PACKET)
611 return prepare_header95(buffer, cmd, size);
613 return prepare_header80(buffer, cmd, size);
616 static void *__conn_prepare_command(struct drbd_connection *connection,
617 struct drbd_socket *sock)
621 return sock->sbuf + drbd_header_size(connection);
624 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
628 mutex_lock(&sock->mutex);
629 p = __conn_prepare_command(connection, sock);
631 mutex_unlock(&sock->mutex);
636 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
638 return conn_prepare_command(peer_device->connection, sock);
641 static int __send_command(struct drbd_connection *connection, int vnr,
642 struct drbd_socket *sock, enum drbd_packet cmd,
643 unsigned int header_size, void *data,
650 * Called with @data == NULL and the size of the data blocks in @size
651 * for commands that send data blocks. For those commands, omit the
652 * MSG_MORE flag: this will increase the likelihood that data blocks
653 * which are page aligned on the sender will end up page aligned on the
656 msg_flags = data ? MSG_MORE : 0;
658 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
660 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
663 err = drbd_send_all(connection, sock->socket, data, size, 0);
664 /* DRBD protocol "pings" are latency critical.
665 * This is supposed to trigger tcp_push_pending_frames() */
666 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
667 tcp_sock_set_nodelay(sock->socket->sk);
672 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
673 enum drbd_packet cmd, unsigned int header_size,
674 void *data, unsigned int size)
676 return __send_command(connection, 0, sock, cmd, header_size, data, size);
679 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
680 enum drbd_packet cmd, unsigned int header_size,
681 void *data, unsigned int size)
685 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
686 mutex_unlock(&sock->mutex);
690 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
691 enum drbd_packet cmd, unsigned int header_size,
692 void *data, unsigned int size)
696 err = __send_command(peer_device->connection, peer_device->device->vnr,
697 sock, cmd, header_size, data, size);
698 mutex_unlock(&sock->mutex);
702 int drbd_send_ping(struct drbd_connection *connection)
704 struct drbd_socket *sock;
706 sock = &connection->meta;
707 if (!conn_prepare_command(connection, sock))
709 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
712 int drbd_send_ping_ack(struct drbd_connection *connection)
714 struct drbd_socket *sock;
716 sock = &connection->meta;
717 if (!conn_prepare_command(connection, sock))
719 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
722 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
724 struct drbd_socket *sock;
725 struct p_rs_param_95 *p;
727 const int apv = peer_device->connection->agreed_pro_version;
728 enum drbd_packet cmd;
730 struct disk_conf *dc;
732 sock = &peer_device->connection->data;
733 p = drbd_prepare_command(peer_device, sock);
738 nc = rcu_dereference(peer_device->connection->net_conf);
740 size = apv <= 87 ? sizeof(struct p_rs_param)
741 : apv == 88 ? sizeof(struct p_rs_param)
742 + strlen(nc->verify_alg) + 1
743 : apv <= 94 ? sizeof(struct p_rs_param_89)
744 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
746 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
748 /* initialize verify_alg and csums_alg */
749 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
751 if (get_ldev(peer_device->device)) {
752 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
753 p->resync_rate = cpu_to_be32(dc->resync_rate);
754 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
755 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
756 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
757 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
758 put_ldev(peer_device->device);
760 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
761 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
762 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
763 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
764 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
768 strcpy(p->verify_alg, nc->verify_alg);
770 strcpy(p->csums_alg, nc->csums_alg);
773 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
776 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
778 struct drbd_socket *sock;
779 struct p_protocol *p;
783 sock = &connection->data;
784 p = __conn_prepare_command(connection, sock);
789 nc = rcu_dereference(connection->net_conf);
791 if (nc->tentative && connection->agreed_pro_version < 92) {
793 drbd_err(connection, "--dry-run is not supported by peer");
798 if (connection->agreed_pro_version >= 87)
799 size += strlen(nc->integrity_alg) + 1;
801 p->protocol = cpu_to_be32(nc->wire_protocol);
802 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
803 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
804 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
805 p->two_primaries = cpu_to_be32(nc->two_primaries);
807 if (nc->discard_my_data)
808 cf |= CF_DISCARD_MY_DATA;
811 p->conn_flags = cpu_to_be32(cf);
813 if (connection->agreed_pro_version >= 87)
814 strcpy(p->integrity_alg, nc->integrity_alg);
817 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
820 int drbd_send_protocol(struct drbd_connection *connection)
824 mutex_lock(&connection->data.mutex);
825 err = __drbd_send_protocol(connection, P_PROTOCOL);
826 mutex_unlock(&connection->data.mutex);
831 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
833 struct drbd_device *device = peer_device->device;
834 struct drbd_socket *sock;
838 if (!get_ldev_if_state(device, D_NEGOTIATING))
841 sock = &peer_device->connection->data;
842 p = drbd_prepare_command(peer_device, sock);
847 spin_lock_irq(&device->ldev->md.uuid_lock);
848 for (i = UI_CURRENT; i < UI_SIZE; i++)
849 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
850 spin_unlock_irq(&device->ldev->md.uuid_lock);
852 device->comm_bm_set = drbd_bm_total_weight(device);
853 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
855 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
857 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
858 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
859 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
862 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
865 int drbd_send_uuids(struct drbd_peer_device *peer_device)
867 return _drbd_send_uuids(peer_device, 0);
870 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
872 return _drbd_send_uuids(peer_device, 8);
875 void drbd_print_uuids(struct drbd_device *device, const char *text)
877 if (get_ldev_if_state(device, D_NEGOTIATING)) {
878 u64 *uuid = device->ldev->md.uuid;
879 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
881 (unsigned long long)uuid[UI_CURRENT],
882 (unsigned long long)uuid[UI_BITMAP],
883 (unsigned long long)uuid[UI_HISTORY_START],
884 (unsigned long long)uuid[UI_HISTORY_END]);
887 drbd_info(device, "%s effective data uuid: %016llX\n",
889 (unsigned long long)device->ed_uuid);
893 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
895 struct drbd_device *device = peer_device->device;
896 struct drbd_socket *sock;
900 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
902 uuid = device->ldev->md.uuid[UI_BITMAP];
903 if (uuid && uuid != UUID_JUST_CREATED)
904 uuid = uuid + UUID_NEW_BM_OFFSET;
906 get_random_bytes(&uuid, sizeof(u64));
907 drbd_uuid_set(device, UI_BITMAP, uuid);
908 drbd_print_uuids(device, "updated sync UUID");
909 drbd_md_sync(device);
911 sock = &peer_device->connection->data;
912 p = drbd_prepare_command(peer_device, sock);
914 p->uuid = cpu_to_be64(uuid);
915 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
919 /* communicated if (agreed_features & DRBD_FF_WSAME) */
921 assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p,
922 struct request_queue *q)
925 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
926 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
927 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
928 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
929 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
930 p->qlim->discard_enabled = blk_queue_discard(q);
931 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
933 q = device->rq_queue;
934 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
935 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
936 p->qlim->alignment_offset = 0;
937 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
938 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
939 p->qlim->discard_enabled = 0;
940 p->qlim->write_same_capable = 0;
944 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
946 struct drbd_device *device = peer_device->device;
947 struct drbd_socket *sock;
949 sector_t d_size, u_size;
951 unsigned int max_bio_size;
952 unsigned int packet_size;
954 sock = &peer_device->connection->data;
955 p = drbd_prepare_command(peer_device, sock);
959 packet_size = sizeof(*p);
960 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
961 packet_size += sizeof(p->qlim[0]);
963 memset(p, 0, packet_size);
964 if (get_ldev_if_state(device, D_NEGOTIATING)) {
965 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
966 d_size = drbd_get_max_capacity(device->ldev);
968 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
970 q_order_type = drbd_queue_order_type(device);
971 max_bio_size = queue_max_hw_sectors(q) << 9;
972 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
973 assign_p_sizes_qlim(device, p, q);
978 q_order_type = QUEUE_ORDERED_NONE;
979 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
980 assign_p_sizes_qlim(device, p, NULL);
983 if (peer_device->connection->agreed_pro_version <= 94)
984 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
985 else if (peer_device->connection->agreed_pro_version < 100)
986 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
988 p->d_size = cpu_to_be64(d_size);
989 p->u_size = cpu_to_be64(u_size);
993 p->c_size = cpu_to_be64(get_capacity(device->vdisk));
994 p->max_bio_size = cpu_to_be32(max_bio_size);
995 p->queue_order_type = cpu_to_be16(q_order_type);
996 p->dds_flags = cpu_to_be16(flags);
998 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
1002 * drbd_send_current_state() - Sends the drbd state to the peer
1003 * @peer_device: DRBD peer device.
1005 int drbd_send_current_state(struct drbd_peer_device *peer_device)
1007 struct drbd_socket *sock;
1010 sock = &peer_device->connection->data;
1011 p = drbd_prepare_command(peer_device, sock);
1014 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1015 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1019 * drbd_send_state() - After a state change, sends the new state to the peer
1020 * @peer_device: DRBD peer device.
1021 * @state: the state to send, not necessarily the current state.
1023 * Each state change queues an "after_state_ch" work, which will eventually
1024 * send the resulting new state to the peer. If more state changes happen
1025 * between queuing and processing of the after_state_ch work, we still
1026 * want to send each intermediary state in the order it occurred.
1028 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1030 struct drbd_socket *sock;
1033 sock = &peer_device->connection->data;
1034 p = drbd_prepare_command(peer_device, sock);
1037 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1038 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1041 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1043 struct drbd_socket *sock;
1044 struct p_req_state *p;
1046 sock = &peer_device->connection->data;
1047 p = drbd_prepare_command(peer_device, sock);
1050 p->mask = cpu_to_be32(mask.i);
1051 p->val = cpu_to_be32(val.i);
1052 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1055 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1057 enum drbd_packet cmd;
1058 struct drbd_socket *sock;
1059 struct p_req_state *p;
1061 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1062 sock = &connection->data;
1063 p = conn_prepare_command(connection, sock);
1066 p->mask = cpu_to_be32(mask.i);
1067 p->val = cpu_to_be32(val.i);
1068 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1071 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1073 struct drbd_socket *sock;
1074 struct p_req_state_reply *p;
1076 sock = &peer_device->connection->meta;
1077 p = drbd_prepare_command(peer_device, sock);
1079 p->retcode = cpu_to_be32(retcode);
1080 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1084 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1086 struct drbd_socket *sock;
1087 struct p_req_state_reply *p;
1088 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1090 sock = &connection->meta;
1091 p = conn_prepare_command(connection, sock);
1093 p->retcode = cpu_to_be32(retcode);
1094 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1098 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1100 BUG_ON(code & ~0xf);
1101 p->encoding = (p->encoding & ~0xf) | code;
1104 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1106 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1109 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1112 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1115 static int fill_bitmap_rle_bits(struct drbd_device *device,
1116 struct p_compressed_bm *p,
1118 struct bm_xfer_ctx *c)
1120 struct bitstream bs;
1121 unsigned long plain_bits;
1128 /* may we use this feature? */
1130 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1132 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1135 if (c->bit_offset >= c->bm_bits)
1136 return 0; /* nothing to do. */
1138 /* use at most thus many bytes */
1139 bitstream_init(&bs, p->code, size, 0);
1140 memset(p->code, 0, size);
1141 /* plain bits covered in this code string */
1144 /* p->encoding & 0x80 stores whether the first run length is set.
1145 * bit offset is implicit.
1146 * start with toggle == 2 to be able to tell the first iteration */
1149 /* see how much plain bits we can stuff into one packet
1150 * using RLE and VLI. */
1152 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1153 : _drbd_bm_find_next(device, c->bit_offset);
1156 rl = tmp - c->bit_offset;
1158 if (toggle == 2) { /* first iteration */
1160 /* the first checked bit was set,
1161 * store start value, */
1162 dcbp_set_start(p, 1);
1163 /* but skip encoding of zero run length */
1167 dcbp_set_start(p, 0);
1170 /* paranoia: catch zero runlength.
1171 * can only happen if bitmap is modified while we scan it. */
1173 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1174 "t:%u bo:%lu\n", toggle, c->bit_offset);
1178 bits = vli_encode_bits(&bs, rl);
1179 if (bits == -ENOBUFS) /* buffer full */
1182 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1188 c->bit_offset = tmp;
1189 } while (c->bit_offset < c->bm_bits);
1191 len = bs.cur.b - p->code + !!bs.cur.bit;
1193 if (plain_bits < (len << 3)) {
1194 /* incompressible with this method.
1195 * we need to rewind both word and bit position. */
1196 c->bit_offset -= plain_bits;
1197 bm_xfer_ctx_bit_to_word_offset(c);
1198 c->bit_offset = c->word_offset * BITS_PER_LONG;
1202 /* RLE + VLI was able to compress it just fine.
1203 * update c->word_offset. */
1204 bm_xfer_ctx_bit_to_word_offset(c);
1206 /* store pad_bits */
1207 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1213 * send_bitmap_rle_or_plain
1215 * Return 0 when done, 1 when another iteration is needed, and a negative error
1216 * code upon failure.
1219 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1221 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1222 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1223 struct p_compressed_bm *p = sock->sbuf + header_size;
1226 len = fill_bitmap_rle_bits(device, p,
1227 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1232 dcbp_set_code(p, RLE_VLI_Bits);
1233 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1234 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1237 c->bytes[0] += header_size + sizeof(*p) + len;
1239 if (c->bit_offset >= c->bm_bits)
1242 /* was not compressible.
1243 * send a buffer full of plain text bits instead. */
1244 unsigned int data_size;
1245 unsigned long num_words;
1246 unsigned long *p = sock->sbuf + header_size;
1248 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1249 num_words = min_t(size_t, data_size / sizeof(*p),
1250 c->bm_words - c->word_offset);
1251 len = num_words * sizeof(*p);
1253 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1254 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1255 c->word_offset += num_words;
1256 c->bit_offset = c->word_offset * BITS_PER_LONG;
1259 c->bytes[1] += header_size + len;
1261 if (c->bit_offset > c->bm_bits)
1262 c->bit_offset = c->bm_bits;
1266 INFO_bm_xfer_stats(device, "send", c);
1274 /* See the comment at receive_bitmap() */
1275 static int _drbd_send_bitmap(struct drbd_device *device)
1277 struct bm_xfer_ctx c;
1280 if (!expect(device->bitmap))
1283 if (get_ldev(device)) {
1284 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1285 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1286 drbd_bm_set_all(device);
1287 if (drbd_bm_write(device)) {
1288 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1289 * but otherwise process as per normal - need to tell other
1290 * side that a full resync is required! */
1291 drbd_err(device, "Failed to write bitmap to disk!\n");
1293 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1294 drbd_md_sync(device);
1300 c = (struct bm_xfer_ctx) {
1301 .bm_bits = drbd_bm_bits(device),
1302 .bm_words = drbd_bm_words(device),
1306 err = send_bitmap_rle_or_plain(device, &c);
1312 int drbd_send_bitmap(struct drbd_device *device)
1314 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1317 mutex_lock(&sock->mutex);
1319 err = !_drbd_send_bitmap(device);
1320 mutex_unlock(&sock->mutex);
1324 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1326 struct drbd_socket *sock;
1327 struct p_barrier_ack *p;
1329 if (connection->cstate < C_WF_REPORT_PARAMS)
1332 sock = &connection->meta;
1333 p = conn_prepare_command(connection, sock);
1336 p->barrier = barrier_nr;
1337 p->set_size = cpu_to_be32(set_size);
1338 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1342 * _drbd_send_ack() - Sends an ack packet
1343 * @device: DRBD device.
1344 * @cmd: Packet command code.
1345 * @sector: sector, needs to be in big endian byte order
1346 * @blksize: size in byte, needs to be in big endian byte order
1347 * @block_id: Id, big endian byte order
1349 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1350 u64 sector, u32 blksize, u64 block_id)
1352 struct drbd_socket *sock;
1353 struct p_block_ack *p;
1355 if (peer_device->device->state.conn < C_CONNECTED)
1358 sock = &peer_device->connection->meta;
1359 p = drbd_prepare_command(peer_device, sock);
1363 p->block_id = block_id;
1364 p->blksize = blksize;
1365 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1366 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1369 /* dp->sector and dp->block_id already/still in network byte order,
1370 * data_size is payload size according to dp->head,
1371 * and may need to be corrected for digest size. */
1372 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1373 struct p_data *dp, int data_size)
1375 if (peer_device->connection->peer_integrity_tfm)
1376 data_size -= crypto_shash_digestsize(peer_device->connection->peer_integrity_tfm);
1377 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1381 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1382 struct p_block_req *rp)
1384 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1388 * drbd_send_ack() - Sends an ack packet
1389 * @device: DRBD device
1390 * @cmd: packet command code
1391 * @peer_req: peer request
1393 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1394 struct drbd_peer_request *peer_req)
1396 return _drbd_send_ack(peer_device, cmd,
1397 cpu_to_be64(peer_req->i.sector),
1398 cpu_to_be32(peer_req->i.size),
1399 peer_req->block_id);
1402 /* This function misuses the block_id field to signal if the blocks
1403 * are is sync or not. */
1404 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1405 sector_t sector, int blksize, u64 block_id)
1407 return _drbd_send_ack(peer_device, cmd,
1408 cpu_to_be64(sector),
1409 cpu_to_be32(blksize),
1410 cpu_to_be64(block_id));
1413 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1414 struct drbd_peer_request *peer_req)
1416 struct drbd_socket *sock;
1417 struct p_block_desc *p;
1419 sock = &peer_device->connection->data;
1420 p = drbd_prepare_command(peer_device, sock);
1423 p->sector = cpu_to_be64(peer_req->i.sector);
1424 p->blksize = cpu_to_be32(peer_req->i.size);
1426 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1429 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1430 sector_t sector, int size, u64 block_id)
1432 struct drbd_socket *sock;
1433 struct p_block_req *p;
1435 sock = &peer_device->connection->data;
1436 p = drbd_prepare_command(peer_device, sock);
1439 p->sector = cpu_to_be64(sector);
1440 p->block_id = block_id;
1441 p->blksize = cpu_to_be32(size);
1442 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1445 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1446 void *digest, int digest_size, enum drbd_packet cmd)
1448 struct drbd_socket *sock;
1449 struct p_block_req *p;
1451 /* FIXME: Put the digest into the preallocated socket buffer. */
1453 sock = &peer_device->connection->data;
1454 p = drbd_prepare_command(peer_device, sock);
1457 p->sector = cpu_to_be64(sector);
1458 p->block_id = ID_SYNCER /* unused */;
1459 p->blksize = cpu_to_be32(size);
1460 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1463 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1465 struct drbd_socket *sock;
1466 struct p_block_req *p;
1468 sock = &peer_device->connection->data;
1469 p = drbd_prepare_command(peer_device, sock);
1472 p->sector = cpu_to_be64(sector);
1473 p->block_id = ID_SYNCER /* unused */;
1474 p->blksize = cpu_to_be32(size);
1475 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1478 /* called on sndtimeo
1479 * returns false if we should retry,
1480 * true if we think connection is dead
1482 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1485 /* long elapsed = (long)(jiffies - device->last_received); */
1487 drop_it = connection->meta.socket == sock
1488 || !connection->ack_receiver.task
1489 || get_t_state(&connection->ack_receiver) != RUNNING
1490 || connection->cstate < C_WF_REPORT_PARAMS;
1495 drop_it = !--connection->ko_count;
1497 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1498 current->comm, current->pid, connection->ko_count);
1499 request_ping(connection);
1502 return drop_it; /* && (device->state == R_PRIMARY) */;
1505 static void drbd_update_congested(struct drbd_connection *connection)
1507 struct sock *sk = connection->data.socket->sk;
1508 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1509 set_bit(NET_CONGESTED, &connection->flags);
1512 /* The idea of sendpage seems to be to put some kind of reference
1513 * to the page into the skb, and to hand it over to the NIC. In
1514 * this process get_page() gets called.
1516 * As soon as the page was really sent over the network put_page()
1517 * gets called by some part of the network layer. [ NIC driver? ]
1519 * [ get_page() / put_page() increment/decrement the count. If count
1520 * reaches 0 the page will be freed. ]
1522 * This works nicely with pages from FSs.
1523 * But this means that in protocol A we might signal IO completion too early!
1525 * In order not to corrupt data during a resync we must make sure
1526 * that we do not reuse our own buffer pages (EEs) to early, therefore
1527 * we have the net_ee list.
1529 * XFS seems to have problems, still, it submits pages with page_count == 0!
1530 * As a workaround, we disable sendpage on pages
1531 * with page_count == 0 or PageSlab.
1533 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1534 int offset, size_t size, unsigned msg_flags)
1536 struct socket *socket;
1540 socket = peer_device->connection->data.socket;
1541 addr = kmap(page) + offset;
1542 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1545 peer_device->device->send_cnt += size >> 9;
1549 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1550 int offset, size_t size, unsigned msg_flags)
1552 struct socket *socket = peer_device->connection->data.socket;
1556 /* e.g. XFS meta- & log-data is in slab pages, which have a
1557 * page_count of 0 and/or have PageSlab() set.
1558 * we cannot use send_page for those, as that does get_page();
1559 * put_page(); and would cause either a VM_BUG directly, or
1560 * __page_cache_release a page that would actually still be referenced
1561 * by someone, leading to some obscure delayed Oops somewhere else. */
1562 if (drbd_disable_sendpage || !sendpage_ok(page))
1563 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1565 msg_flags |= MSG_NOSIGNAL;
1566 drbd_update_congested(peer_device->connection);
1570 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1572 if (sent == -EAGAIN) {
1573 if (we_should_drop_the_connection(peer_device->connection, socket))
1577 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1578 __func__, (int)size, len, sent);
1585 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1586 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1590 peer_device->device->send_cnt += size >> 9;
1595 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1597 struct bio_vec bvec;
1598 struct bvec_iter iter;
1600 /* hint all but last page with MSG_MORE */
1601 bio_for_each_segment(bvec, bio, iter) {
1604 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1605 bvec.bv_offset, bvec.bv_len,
1606 bio_iter_last(bvec, iter)
1610 /* REQ_OP_WRITE_SAME has only one segment */
1611 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1617 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1619 struct bio_vec bvec;
1620 struct bvec_iter iter;
1622 /* hint all but last page with MSG_MORE */
1623 bio_for_each_segment(bvec, bio, iter) {
1626 err = _drbd_send_page(peer_device, bvec.bv_page,
1627 bvec.bv_offset, bvec.bv_len,
1628 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1631 /* REQ_OP_WRITE_SAME has only one segment */
1632 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1638 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1639 struct drbd_peer_request *peer_req)
1641 struct page *page = peer_req->pages;
1642 unsigned len = peer_req->i.size;
1645 /* hint all but last page with MSG_MORE */
1646 page_chain_for_each(page) {
1647 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1649 err = _drbd_send_page(peer_device, page, 0, l,
1650 page_chain_next(page) ? MSG_MORE : 0);
1658 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1661 if (connection->agreed_pro_version >= 95)
1662 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1663 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1664 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1665 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1666 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
1667 (bio_op(bio) == REQ_OP_WRITE_ZEROES ?
1668 ((connection->agreed_features & DRBD_FF_WZEROES) ?
1669 (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0))
1673 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1676 /* Used to send write or TRIM aka REQ_OP_DISCARD requests
1677 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1679 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1681 struct drbd_device *device = peer_device->device;
1682 struct drbd_socket *sock;
1684 struct p_wsame *wsame = NULL;
1686 unsigned int dp_flags = 0;
1690 sock = &peer_device->connection->data;
1691 p = drbd_prepare_command(peer_device, sock);
1692 digest_size = peer_device->connection->integrity_tfm ?
1693 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1697 p->sector = cpu_to_be64(req->i.sector);
1698 p->block_id = (unsigned long)req;
1699 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1700 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1701 if (device->state.conn >= C_SYNC_SOURCE &&
1702 device->state.conn <= C_PAUSED_SYNC_T)
1703 dp_flags |= DP_MAY_SET_IN_SYNC;
1704 if (peer_device->connection->agreed_pro_version >= 100) {
1705 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1706 dp_flags |= DP_SEND_RECEIVE_ACK;
1707 /* During resync, request an explicit write ack,
1708 * even in protocol != C */
1709 if (req->rq_state & RQ_EXP_WRITE_ACK
1710 || (dp_flags & DP_MAY_SET_IN_SYNC))
1711 dp_flags |= DP_SEND_WRITE_ACK;
1713 p->dp_flags = cpu_to_be32(dp_flags);
1715 if (dp_flags & (DP_DISCARD|DP_ZEROES)) {
1716 enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM;
1717 struct p_trim *t = (struct p_trim*)p;
1718 t->size = cpu_to_be32(req->i.size);
1719 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0);
1722 if (dp_flags & DP_WSAME) {
1723 /* this will only work if DRBD_FF_WSAME is set AND the
1724 * handshake agreed that all nodes and backend devices are
1725 * WRITE_SAME capable and agree on logical_block_size */
1726 wsame = (struct p_wsame*)p;
1727 digest_out = wsame + 1;
1728 wsame->size = cpu_to_be32(req->i.size);
1732 /* our digest is still only over the payload.
1733 * TRIM does not carry any payload. */
1735 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1738 __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1739 sizeof(*wsame) + digest_size, NULL,
1740 bio_iovec(req->master_bio).bv_len);
1743 __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1744 sizeof(*p) + digest_size, NULL, req->i.size);
1746 /* For protocol A, we have to memcpy the payload into
1747 * socket buffers, as we may complete right away
1748 * as soon as we handed it over to tcp, at which point the data
1749 * pages may become invalid.
1751 * For data-integrity enabled, we copy it as well, so we can be
1752 * sure that even if the bio pages may still be modified, it
1753 * won't change the data on the wire, thus if the digest checks
1754 * out ok after sending on this side, but does not fit on the
1755 * receiving side, we sure have detected corruption elsewhere.
1757 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1758 err = _drbd_send_bio(peer_device, req->master_bio);
1760 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1762 /* double check digest, sometimes buffers have been modified in flight. */
1763 if (digest_size > 0 && digest_size <= 64) {
1764 /* 64 byte, 512 bit, is the largest digest size
1765 * currently supported in kernel crypto. */
1766 unsigned char digest[64];
1767 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1768 if (memcmp(p + 1, digest, digest_size)) {
1770 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1771 (unsigned long long)req->i.sector, req->i.size);
1773 } /* else if (digest_size > 64) {
1774 ... Be noisy about digest too large ...
1778 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1783 /* answer packet, used to send data back for read requests:
1784 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1785 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1787 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1788 struct drbd_peer_request *peer_req)
1790 struct drbd_device *device = peer_device->device;
1791 struct drbd_socket *sock;
1796 sock = &peer_device->connection->data;
1797 p = drbd_prepare_command(peer_device, sock);
1799 digest_size = peer_device->connection->integrity_tfm ?
1800 crypto_shash_digestsize(peer_device->connection->integrity_tfm) : 0;
1804 p->sector = cpu_to_be64(peer_req->i.sector);
1805 p->block_id = peer_req->block_id;
1806 p->seq_num = 0; /* unused */
1809 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1810 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1812 err = _drbd_send_zc_ee(peer_device, peer_req);
1813 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1818 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1820 struct drbd_socket *sock;
1821 struct p_block_desc *p;
1823 sock = &peer_device->connection->data;
1824 p = drbd_prepare_command(peer_device, sock);
1827 p->sector = cpu_to_be64(req->i.sector);
1828 p->blksize = cpu_to_be32(req->i.size);
1829 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1833 drbd_send distinguishes two cases:
1835 Packets sent via the data socket "sock"
1836 and packets sent via the meta data socket "msock"
1839 -----------------+-------------------------+------------------------------
1840 timeout conf.timeout / 2 conf.timeout / 2
1841 timeout action send a ping via msock Abort communication
1842 and close all sockets
1846 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1848 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1849 void *buf, size_t size, unsigned msg_flags)
1851 struct kvec iov = {.iov_base = buf, .iov_len = size};
1852 struct msghdr msg = {.msg_flags = msg_flags | MSG_NOSIGNAL};
1858 /* THINK if (signal_pending) return ... ? */
1860 iov_iter_kvec(&msg.msg_iter, WRITE, &iov, 1, size);
1862 if (sock == connection->data.socket) {
1864 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1866 drbd_update_congested(connection);
1869 rv = sock_sendmsg(sock, &msg);
1870 if (rv == -EAGAIN) {
1871 if (we_should_drop_the_connection(connection, sock))
1877 flush_signals(current);
1883 } while (sent < size);
1885 if (sock == connection->data.socket)
1886 clear_bit(NET_CONGESTED, &connection->flags);
1889 if (rv != -EAGAIN) {
1890 drbd_err(connection, "%s_sendmsg returned %d\n",
1891 sock == connection->meta.socket ? "msock" : "sock",
1893 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1895 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1902 * drbd_send_all - Send an entire buffer
1904 * Returns 0 upon success and a negative error value otherwise.
1906 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1907 size_t size, unsigned msg_flags)
1911 err = drbd_send(connection, sock, buffer, size, msg_flags);
1919 static int drbd_open(struct block_device *bdev, fmode_t mode)
1921 struct drbd_device *device = bdev->bd_disk->private_data;
1922 unsigned long flags;
1925 mutex_lock(&drbd_main_mutex);
1926 spin_lock_irqsave(&device->resource->req_lock, flags);
1927 /* to have a stable device->state.role
1928 * and no race with updating open_cnt */
1930 if (device->state.role != R_PRIMARY) {
1931 if (mode & FMODE_WRITE)
1933 else if (!drbd_allow_oos)
1939 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1940 mutex_unlock(&drbd_main_mutex);
1945 static void drbd_release(struct gendisk *gd, fmode_t mode)
1947 struct drbd_device *device = gd->private_data;
1948 mutex_lock(&drbd_main_mutex);
1950 mutex_unlock(&drbd_main_mutex);
1953 /* need to hold resource->req_lock */
1954 void drbd_queue_unplug(struct drbd_device *device)
1956 if (device->state.pdsk >= D_INCONSISTENT && device->state.conn >= C_CONNECTED) {
1957 D_ASSERT(device, device->state.role == R_PRIMARY);
1958 if (test_and_clear_bit(UNPLUG_REMOTE, &device->flags)) {
1959 drbd_queue_work_if_unqueued(
1960 &first_peer_device(device)->connection->sender_work,
1961 &device->unplug_work);
1966 static void drbd_set_defaults(struct drbd_device *device)
1968 /* Beware! The actual layout differs
1969 * between big endian and little endian */
1970 device->state = (union drbd_dev_state) {
1971 { .role = R_SECONDARY,
1973 .conn = C_STANDALONE,
1979 void drbd_init_set_defaults(struct drbd_device *device)
1981 /* the memset(,0,) did most of this.
1982 * note: only assignments, no allocation in here */
1984 drbd_set_defaults(device);
1986 atomic_set(&device->ap_bio_cnt, 0);
1987 atomic_set(&device->ap_actlog_cnt, 0);
1988 atomic_set(&device->ap_pending_cnt, 0);
1989 atomic_set(&device->rs_pending_cnt, 0);
1990 atomic_set(&device->unacked_cnt, 0);
1991 atomic_set(&device->local_cnt, 0);
1992 atomic_set(&device->pp_in_use_by_net, 0);
1993 atomic_set(&device->rs_sect_in, 0);
1994 atomic_set(&device->rs_sect_ev, 0);
1995 atomic_set(&device->ap_in_flight, 0);
1996 atomic_set(&device->md_io.in_use, 0);
1998 mutex_init(&device->own_state_mutex);
1999 device->state_mutex = &device->own_state_mutex;
2001 spin_lock_init(&device->al_lock);
2002 spin_lock_init(&device->peer_seq_lock);
2004 INIT_LIST_HEAD(&device->active_ee);
2005 INIT_LIST_HEAD(&device->sync_ee);
2006 INIT_LIST_HEAD(&device->done_ee);
2007 INIT_LIST_HEAD(&device->read_ee);
2008 INIT_LIST_HEAD(&device->net_ee);
2009 INIT_LIST_HEAD(&device->resync_reads);
2010 INIT_LIST_HEAD(&device->resync_work.list);
2011 INIT_LIST_HEAD(&device->unplug_work.list);
2012 INIT_LIST_HEAD(&device->bm_io_work.w.list);
2013 INIT_LIST_HEAD(&device->pending_master_completion[0]);
2014 INIT_LIST_HEAD(&device->pending_master_completion[1]);
2015 INIT_LIST_HEAD(&device->pending_completion[0]);
2016 INIT_LIST_HEAD(&device->pending_completion[1]);
2018 device->resync_work.cb = w_resync_timer;
2019 device->unplug_work.cb = w_send_write_hint;
2020 device->bm_io_work.w.cb = w_bitmap_io;
2022 timer_setup(&device->resync_timer, resync_timer_fn, 0);
2023 timer_setup(&device->md_sync_timer, md_sync_timer_fn, 0);
2024 timer_setup(&device->start_resync_timer, start_resync_timer_fn, 0);
2025 timer_setup(&device->request_timer, request_timer_fn, 0);
2027 init_waitqueue_head(&device->misc_wait);
2028 init_waitqueue_head(&device->state_wait);
2029 init_waitqueue_head(&device->ee_wait);
2030 init_waitqueue_head(&device->al_wait);
2031 init_waitqueue_head(&device->seq_wait);
2033 device->resync_wenr = LC_FREE;
2034 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2035 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2038 void drbd_set_my_capacity(struct drbd_device *device, sector_t size)
2042 set_capacity(device->vdisk, size);
2043 revalidate_disk_size(device->vdisk, false);
2045 drbd_info(device, "size = %s (%llu KB)\n",
2046 ppsize(ppb, size>>1), (unsigned long long)size>>1);
2049 void drbd_device_cleanup(struct drbd_device *device)
2052 if (first_peer_device(device)->connection->receiver.t_state != NONE)
2053 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2054 first_peer_device(device)->connection->receiver.t_state);
2056 device->al_writ_cnt =
2057 device->bm_writ_cnt =
2065 device->rs_failed = 0;
2066 device->rs_last_events = 0;
2067 device->rs_last_sect_ev = 0;
2068 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2069 device->rs_mark_left[i] = 0;
2070 device->rs_mark_time[i] = 0;
2072 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2074 set_capacity(device->vdisk, 0);
2075 revalidate_disk_size(device->vdisk, false);
2076 if (device->bitmap) {
2077 /* maybe never allocated. */
2078 drbd_bm_resize(device, 0, 1);
2079 drbd_bm_cleanup(device);
2082 drbd_backing_dev_free(device, device->ldev);
2083 device->ldev = NULL;
2085 clear_bit(AL_SUSPENDED, &device->flags);
2087 D_ASSERT(device, list_empty(&device->active_ee));
2088 D_ASSERT(device, list_empty(&device->sync_ee));
2089 D_ASSERT(device, list_empty(&device->done_ee));
2090 D_ASSERT(device, list_empty(&device->read_ee));
2091 D_ASSERT(device, list_empty(&device->net_ee));
2092 D_ASSERT(device, list_empty(&device->resync_reads));
2093 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2094 D_ASSERT(device, list_empty(&device->resync_work.list));
2095 D_ASSERT(device, list_empty(&device->unplug_work.list));
2097 drbd_set_defaults(device);
2101 static void drbd_destroy_mempools(void)
2105 while (drbd_pp_pool) {
2106 page = drbd_pp_pool;
2107 drbd_pp_pool = (struct page *)page_private(page);
2112 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2114 bioset_exit(&drbd_io_bio_set);
2115 bioset_exit(&drbd_md_io_bio_set);
2116 mempool_exit(&drbd_md_io_page_pool);
2117 mempool_exit(&drbd_ee_mempool);
2118 mempool_exit(&drbd_request_mempool);
2119 kmem_cache_destroy(drbd_ee_cache);
2120 kmem_cache_destroy(drbd_request_cache);
2121 kmem_cache_destroy(drbd_bm_ext_cache);
2122 kmem_cache_destroy(drbd_al_ext_cache);
2124 drbd_ee_cache = NULL;
2125 drbd_request_cache = NULL;
2126 drbd_bm_ext_cache = NULL;
2127 drbd_al_ext_cache = NULL;
2132 static int drbd_create_mempools(void)
2135 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * drbd_minor_count;
2139 drbd_request_cache = kmem_cache_create(
2140 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2141 if (drbd_request_cache == NULL)
2144 drbd_ee_cache = kmem_cache_create(
2145 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2146 if (drbd_ee_cache == NULL)
2149 drbd_bm_ext_cache = kmem_cache_create(
2150 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2151 if (drbd_bm_ext_cache == NULL)
2154 drbd_al_ext_cache = kmem_cache_create(
2155 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2156 if (drbd_al_ext_cache == NULL)
2160 ret = bioset_init(&drbd_io_bio_set, BIO_POOL_SIZE, 0, 0);
2164 ret = bioset_init(&drbd_md_io_bio_set, DRBD_MIN_POOL_PAGES, 0,
2169 ret = mempool_init_page_pool(&drbd_md_io_page_pool, DRBD_MIN_POOL_PAGES, 0);
2173 ret = mempool_init_slab_pool(&drbd_request_mempool, number,
2174 drbd_request_cache);
2178 ret = mempool_init_slab_pool(&drbd_ee_mempool, number, drbd_ee_cache);
2182 /* drbd's page pool */
2183 spin_lock_init(&drbd_pp_lock);
2185 for (i = 0; i < number; i++) {
2186 page = alloc_page(GFP_HIGHUSER);
2189 set_page_private(page, (unsigned long)drbd_pp_pool);
2190 drbd_pp_pool = page;
2192 drbd_pp_vacant = number;
2197 drbd_destroy_mempools(); /* in case we allocated some */
2201 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2205 rr = drbd_free_peer_reqs(device, &device->active_ee);
2207 drbd_err(device, "%d EEs in active list found!\n", rr);
2209 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2211 drbd_err(device, "%d EEs in sync list found!\n", rr);
2213 rr = drbd_free_peer_reqs(device, &device->read_ee);
2215 drbd_err(device, "%d EEs in read list found!\n", rr);
2217 rr = drbd_free_peer_reqs(device, &device->done_ee);
2219 drbd_err(device, "%d EEs in done list found!\n", rr);
2221 rr = drbd_free_peer_reqs(device, &device->net_ee);
2223 drbd_err(device, "%d EEs in net list found!\n", rr);
2226 /* caution. no locking. */
2227 void drbd_destroy_device(struct kref *kref)
2229 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2230 struct drbd_resource *resource = device->resource;
2231 struct drbd_peer_device *peer_device, *tmp_peer_device;
2233 del_timer_sync(&device->request_timer);
2235 /* paranoia asserts */
2236 D_ASSERT(device, device->open_cnt == 0);
2237 /* end paranoia asserts */
2239 /* cleanup stuff that may have been allocated during
2240 * device (re-)configuration or state changes */
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 blk_queue_write_cache(q, true, true);
2769 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2770 This triggers a max_bio_size message upon first attach or connect */
2771 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2773 device->md_io.page = alloc_page(GFP_KERNEL);
2774 if (!device->md_io.page)
2775 goto out_no_io_page;
2777 if (drbd_bm_init(device))
2779 device->read_requests = RB_ROOT;
2780 device->write_requests = RB_ROOT;
2782 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2785 err = ERR_MINOR_OR_VOLUME_EXISTS;
2786 goto out_no_minor_idr;
2788 kref_get(&device->kref);
2790 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2793 err = ERR_MINOR_OR_VOLUME_EXISTS;
2794 goto out_idr_remove_minor;
2796 kref_get(&device->kref);
2798 INIT_LIST_HEAD(&device->peer_devices);
2799 INIT_LIST_HEAD(&device->pending_bitmap_io);
2800 for_each_connection(connection, resource) {
2801 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2803 goto out_idr_remove_from_resource;
2804 peer_device->connection = connection;
2805 peer_device->device = device;
2807 list_add(&peer_device->peer_devices, &device->peer_devices);
2808 kref_get(&device->kref);
2810 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2813 err = ERR_INVALID_REQUEST;
2814 goto out_idr_remove_from_resource;
2816 kref_get(&connection->kref);
2817 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2820 if (init_submitter(device)) {
2822 goto out_idr_remove_vol;
2827 /* inherit the connection state */
2828 device->state.conn = first_connection(resource)->cstate;
2829 if (device->state.conn == C_WF_REPORT_PARAMS) {
2830 for_each_peer_device(peer_device, device)
2831 drbd_connected(peer_device);
2833 /* move to create_peer_device() */
2834 for_each_peer_device(peer_device, device)
2835 drbd_debugfs_peer_device_add(peer_device);
2836 drbd_debugfs_device_add(device);
2840 idr_remove(&connection->peer_devices, vnr);
2841 out_idr_remove_from_resource:
2842 for_each_connection(connection, resource) {
2843 peer_device = idr_remove(&connection->peer_devices, vnr);
2845 kref_put(&connection->kref, drbd_destroy_connection);
2847 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2848 list_del(&peer_device->peer_devices);
2851 idr_remove(&resource->devices, vnr);
2852 out_idr_remove_minor:
2853 idr_remove(&drbd_devices, minor);
2856 drbd_bm_cleanup(device);
2858 __free_page(device->md_io.page);
2862 blk_cleanup_queue(q);
2864 kref_put(&resource->kref, drbd_destroy_resource);
2869 void drbd_delete_device(struct drbd_device *device)
2871 struct drbd_resource *resource = device->resource;
2872 struct drbd_connection *connection;
2873 struct drbd_peer_device *peer_device;
2875 /* move to free_peer_device() */
2876 for_each_peer_device(peer_device, device)
2877 drbd_debugfs_peer_device_cleanup(peer_device);
2878 drbd_debugfs_device_cleanup(device);
2879 for_each_connection(connection, resource) {
2880 idr_remove(&connection->peer_devices, device->vnr);
2881 kref_put(&device->kref, drbd_destroy_device);
2883 idr_remove(&resource->devices, device->vnr);
2884 kref_put(&device->kref, drbd_destroy_device);
2885 idr_remove(&drbd_devices, device_to_minor(device));
2886 kref_put(&device->kref, drbd_destroy_device);
2887 del_gendisk(device->vdisk);
2889 kref_put(&device->kref, drbd_destroy_device);
2892 static int __init drbd_init(void)
2896 if (drbd_minor_count < DRBD_MINOR_COUNT_MIN || drbd_minor_count > DRBD_MINOR_COUNT_MAX) {
2897 pr_err("invalid minor_count (%d)\n", drbd_minor_count);
2901 drbd_minor_count = DRBD_MINOR_COUNT_DEF;
2905 err = register_blkdev(DRBD_MAJOR, "drbd");
2907 pr_err("unable to register block device major %d\n",
2913 * allocate all necessary structs
2915 init_waitqueue_head(&drbd_pp_wait);
2917 drbd_proc = NULL; /* play safe for drbd_cleanup */
2918 idr_init(&drbd_devices);
2920 mutex_init(&resources_mutex);
2921 INIT_LIST_HEAD(&drbd_resources);
2923 err = drbd_genl_register();
2925 pr_err("unable to register generic netlink family\n");
2929 err = drbd_create_mempools();
2934 drbd_proc = proc_create_single("drbd", S_IFREG | 0444 , NULL, drbd_seq_show);
2936 pr_err("unable to register proc file\n");
2940 retry.wq = create_singlethread_workqueue("drbd-reissue");
2942 pr_err("unable to create retry workqueue\n");
2945 INIT_WORK(&retry.worker, do_retry);
2946 spin_lock_init(&retry.lock);
2947 INIT_LIST_HEAD(&retry.writes);
2949 drbd_debugfs_init();
2951 pr_info("initialized. "
2952 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2953 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
2954 pr_info("%s\n", drbd_buildtag());
2955 pr_info("registered as block device major %d\n", DRBD_MAJOR);
2956 return 0; /* Success! */
2961 pr_err("ran out of memory\n");
2963 pr_err("initialization failure\n");
2967 static void drbd_free_one_sock(struct drbd_socket *ds)
2970 mutex_lock(&ds->mutex);
2973 mutex_unlock(&ds->mutex);
2975 /* so debugfs does not need to mutex_lock() */
2977 kernel_sock_shutdown(s, SHUT_RDWR);
2982 void drbd_free_sock(struct drbd_connection *connection)
2984 if (connection->data.socket)
2985 drbd_free_one_sock(&connection->data);
2986 if (connection->meta.socket)
2987 drbd_free_one_sock(&connection->meta);
2990 /* meta data management */
2992 void conn_md_sync(struct drbd_connection *connection)
2994 struct drbd_peer_device *peer_device;
2998 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2999 struct drbd_device *device = peer_device->device;
3001 kref_get(&device->kref);
3003 drbd_md_sync(device);
3004 kref_put(&device->kref, drbd_destroy_device);
3010 /* aligned 4kByte */
3011 struct meta_data_on_disk {
3012 u64 la_size_sect; /* last agreed size. */
3013 u64 uuid[UI_SIZE]; /* UUIDs. */
3016 u32 flags; /* MDF */
3019 u32 al_offset; /* offset to this block */
3020 u32 al_nr_extents; /* important for restoring the AL (userspace) */
3021 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3022 u32 bm_offset; /* offset to the bitmap, from here */
3023 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
3024 u32 la_peer_max_bio_size; /* last peer max_bio_size */
3026 /* see al_tr_number_to_on_disk_sector() */
3028 u32 al_stripe_size_4k;
3030 u8 reserved_u8[4096 - (7*8 + 10*4)];
3035 void drbd_md_write(struct drbd_device *device, void *b)
3037 struct meta_data_on_disk *buffer = b;
3041 memset(buffer, 0, sizeof(*buffer));
3043 buffer->la_size_sect = cpu_to_be64(get_capacity(device->vdisk));
3044 for (i = UI_CURRENT; i < UI_SIZE; i++)
3045 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3046 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3047 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3049 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3050 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3051 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3052 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3053 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3055 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3056 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3058 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3059 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3061 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3062 sector = device->ldev->md.md_offset;
3064 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3065 /* this was a try anyways ... */
3066 drbd_err(device, "meta data update failed!\n");
3067 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3072 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3073 * @device: DRBD device.
3075 void drbd_md_sync(struct drbd_device *device)
3077 struct meta_data_on_disk *buffer;
3079 /* Don't accidentally change the DRBD meta data layout. */
3080 BUILD_BUG_ON(UI_SIZE != 4);
3081 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3083 del_timer(&device->md_sync_timer);
3084 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3085 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3088 /* We use here D_FAILED and not D_ATTACHING because we try to write
3089 * metadata even if we detach due to a disk failure! */
3090 if (!get_ldev_if_state(device, D_FAILED))
3093 buffer = drbd_md_get_buffer(device, __func__);
3097 drbd_md_write(device, buffer);
3099 /* Update device->ldev->md.la_size_sect,
3100 * since we updated it on metadata. */
3101 device->ldev->md.la_size_sect = get_capacity(device->vdisk);
3103 drbd_md_put_buffer(device);
3108 static int check_activity_log_stripe_size(struct drbd_device *device,
3109 struct meta_data_on_disk *on_disk,
3110 struct drbd_md *in_core)
3112 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3113 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3116 /* both not set: default to old fixed size activity log */
3117 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3119 al_stripe_size_4k = MD_32kB_SECT/8;
3122 /* some paranoia plausibility checks */
3124 /* we need both values to be set */
3125 if (al_stripes == 0 || al_stripe_size_4k == 0)
3128 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3130 /* Upper limit of activity log area, to avoid potential overflow
3131 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3132 * than 72 * 4k blocks total only increases the amount of history,
3133 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3134 if (al_size_4k > (16 * 1024 * 1024/4))
3137 /* Lower limit: we need at least 8 transaction slots (32kB)
3138 * to not break existing setups */
3139 if (al_size_4k < MD_32kB_SECT/8)
3142 in_core->al_stripe_size_4k = al_stripe_size_4k;
3143 in_core->al_stripes = al_stripes;
3144 in_core->al_size_4k = al_size_4k;
3148 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3149 al_stripes, al_stripe_size_4k);
3153 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3155 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3156 struct drbd_md *in_core = &bdev->md;
3157 s32 on_disk_al_sect;
3158 s32 on_disk_bm_sect;
3160 /* The on-disk size of the activity log, calculated from offsets, and
3161 * the size of the activity log calculated from the stripe settings,
3163 * Though we could relax this a bit: it is ok, if the striped activity log
3164 * fits in the available on-disk activity log size.
3165 * Right now, that would break how resize is implemented.
3166 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3167 * of possible unused padding space in the on disk layout. */
3168 if (in_core->al_offset < 0) {
3169 if (in_core->bm_offset > in_core->al_offset)
3171 on_disk_al_sect = -in_core->al_offset;
3172 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3174 if (in_core->al_offset != MD_4kB_SECT)
3176 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3179 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3180 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3183 /* old fixed size meta data is exactly that: fixed. */
3184 if (in_core->meta_dev_idx >= 0) {
3185 if (in_core->md_size_sect != MD_128MB_SECT
3186 || in_core->al_offset != MD_4kB_SECT
3187 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3188 || in_core->al_stripes != 1
3189 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3193 if (capacity < in_core->md_size_sect)
3195 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3198 /* should be aligned, and at least 32k */
3199 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3202 /* should fit (for now: exactly) into the available on-disk space;
3203 * overflow prevention is in check_activity_log_stripe_size() above. */
3204 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3207 /* again, should be aligned */
3208 if (in_core->bm_offset & 7)
3211 /* FIXME check for device grow with flex external meta data? */
3213 /* can the available bitmap space cover the last agreed device size? */
3214 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3220 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3221 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3222 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3223 in_core->meta_dev_idx,
3224 in_core->al_stripes, in_core->al_stripe_size_4k,
3225 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3226 (unsigned long long)in_core->la_size_sect,
3227 (unsigned long long)capacity);
3234 * drbd_md_read() - Reads in the meta data super block
3235 * @device: DRBD device.
3236 * @bdev: Device from which the meta data should be read in.
3238 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3239 * something goes wrong.
3241 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3242 * even before @bdev is assigned to @device->ldev.
3244 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3246 struct meta_data_on_disk *buffer;
3248 int i, rv = NO_ERROR;
3250 if (device->state.disk != D_DISKLESS)
3251 return ERR_DISK_CONFIGURED;
3253 buffer = drbd_md_get_buffer(device, __func__);
3257 /* First, figure out where our meta data superblock is located,
3259 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3260 bdev->md.md_offset = drbd_md_ss(bdev);
3261 /* Even for (flexible or indexed) external meta data,
3262 * initially restrict us to the 4k superblock for now.
3263 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3264 bdev->md.md_size_sect = 8;
3266 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3268 /* NOTE: can't do normal error processing here as this is
3269 called BEFORE disk is attached */
3270 drbd_err(device, "Error while reading metadata.\n");
3271 rv = ERR_IO_MD_DISK;
3275 magic = be32_to_cpu(buffer->magic);
3276 flags = be32_to_cpu(buffer->flags);
3277 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3278 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3279 /* btw: that's Activity Log clean, not "all" clean. */
3280 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3281 rv = ERR_MD_UNCLEAN;
3285 rv = ERR_MD_INVALID;
3286 if (magic != DRBD_MD_MAGIC_08) {
3287 if (magic == DRBD_MD_MAGIC_07)
3288 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3290 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3294 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3295 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3296 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3301 /* convert to in_core endian */
3302 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3303 for (i = UI_CURRENT; i < UI_SIZE; i++)
3304 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3305 bdev->md.flags = be32_to_cpu(buffer->flags);
3306 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3308 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3309 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3310 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3312 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3314 if (check_offsets_and_sizes(device, bdev))
3317 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3318 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3319 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3322 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3323 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3324 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3330 spin_lock_irq(&device->resource->req_lock);
3331 if (device->state.conn < C_CONNECTED) {
3333 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3334 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3335 device->peer_max_bio_size = peer;
3337 spin_unlock_irq(&device->resource->req_lock);
3340 drbd_md_put_buffer(device);
3346 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3347 * @device: DRBD device.
3349 * Call this function if you change anything that should be written to
3350 * the meta-data super block. This function sets MD_DIRTY, and starts a
3351 * timer that ensures that within five seconds you have to call drbd_md_sync().
3353 void drbd_md_mark_dirty(struct drbd_device *device)
3355 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3356 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3359 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3363 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3364 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3367 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3369 if (idx == UI_CURRENT) {
3370 if (device->state.role == R_PRIMARY)
3375 drbd_set_ed_uuid(device, val);
3378 device->ldev->md.uuid[idx] = val;
3379 drbd_md_mark_dirty(device);
3382 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3384 unsigned long flags;
3385 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3386 __drbd_uuid_set(device, idx, val);
3387 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3390 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3392 unsigned long flags;
3393 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3394 if (device->ldev->md.uuid[idx]) {
3395 drbd_uuid_move_history(device);
3396 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3398 __drbd_uuid_set(device, idx, val);
3399 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3403 * drbd_uuid_new_current() - Creates a new current UUID
3404 * @device: DRBD device.
3406 * Creates a new current UUID, and rotates the old current UUID into
3407 * the bitmap slot. Causes an incremental resync upon next connect.
3409 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3412 unsigned long long bm_uuid;
3414 get_random_bytes(&val, sizeof(u64));
3416 spin_lock_irq(&device->ldev->md.uuid_lock);
3417 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3420 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3422 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3423 __drbd_uuid_set(device, UI_CURRENT, val);
3424 spin_unlock_irq(&device->ldev->md.uuid_lock);
3426 drbd_print_uuids(device, "new current UUID");
3427 /* get it to stable storage _now_ */
3428 drbd_md_sync(device);
3431 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3433 unsigned long flags;
3434 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3437 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3439 drbd_uuid_move_history(device);
3440 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3441 device->ldev->md.uuid[UI_BITMAP] = 0;
3443 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3445 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3447 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3449 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3451 drbd_md_mark_dirty(device);
3455 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3456 * @device: DRBD device.
3458 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3460 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3464 drbd_md_set_flag(device, MDF_FULL_SYNC);
3465 drbd_md_sync(device);
3466 drbd_bm_set_all(device);
3468 rv = drbd_bm_write(device);
3471 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3472 drbd_md_sync(device);
3479 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3480 * @device: DRBD device.
3482 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3484 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3486 drbd_resume_al(device);
3487 drbd_bm_clear_all(device);
3488 return drbd_bm_write(device);
3491 static int w_bitmap_io(struct drbd_work *w, int unused)
3493 struct drbd_device *device =
3494 container_of(w, struct drbd_device, bm_io_work.w);
3495 struct bm_io_work *work = &device->bm_io_work;
3498 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3499 int cnt = atomic_read(&device->ap_bio_cnt);
3501 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3505 if (get_ldev(device)) {
3506 drbd_bm_lock(device, work->why, work->flags);
3507 rv = work->io_fn(device);
3508 drbd_bm_unlock(device);
3512 clear_bit_unlock(BITMAP_IO, &device->flags);
3513 wake_up(&device->misc_wait);
3516 work->done(device, rv);
3518 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3526 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3527 * @device: DRBD device.
3528 * @io_fn: IO callback to be called when bitmap IO is possible
3529 * @done: callback to be called after the bitmap IO was performed
3530 * @why: Descriptive text of the reason for doing the IO
3532 * While IO on the bitmap happens we freeze application IO thus we ensure
3533 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3534 * called from worker context. It MUST NOT be used while a previous such
3535 * work is still pending!
3537 * Its worker function encloses the call of io_fn() by get_ldev() and
3540 void drbd_queue_bitmap_io(struct drbd_device *device,
3541 int (*io_fn)(struct drbd_device *),
3542 void (*done)(struct drbd_device *, int),
3543 char *why, enum bm_flag flags)
3545 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3547 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3548 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3549 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3550 if (device->bm_io_work.why)
3551 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3552 why, device->bm_io_work.why);
3554 device->bm_io_work.io_fn = io_fn;
3555 device->bm_io_work.done = done;
3556 device->bm_io_work.why = why;
3557 device->bm_io_work.flags = flags;
3559 spin_lock_irq(&device->resource->req_lock);
3560 set_bit(BITMAP_IO, &device->flags);
3561 /* don't wait for pending application IO if the caller indicates that
3562 * application IO does not conflict anyways. */
3563 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3564 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3565 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3566 &device->bm_io_work.w);
3568 spin_unlock_irq(&device->resource->req_lock);
3572 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3573 * @device: DRBD device.
3574 * @io_fn: IO callback to be called when bitmap IO is possible
3575 * @why: Descriptive text of the reason for doing the IO
3577 * freezes application IO while that the actual IO operations runs. This
3578 * functions MAY NOT be called from worker context.
3580 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3581 char *why, enum bm_flag flags)
3583 /* Only suspend io, if some operation is supposed to be locked out */
3584 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3587 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3590 drbd_suspend_io(device);
3592 drbd_bm_lock(device, why, flags);
3594 drbd_bm_unlock(device);
3597 drbd_resume_io(device);
3602 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3604 if ((device->ldev->md.flags & flag) != flag) {
3605 drbd_md_mark_dirty(device);
3606 device->ldev->md.flags |= flag;
3610 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3612 if ((device->ldev->md.flags & flag) != 0) {
3613 drbd_md_mark_dirty(device);
3614 device->ldev->md.flags &= ~flag;
3617 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3619 return (bdev->md.flags & flag) != 0;
3622 static void md_sync_timer_fn(struct timer_list *t)
3624 struct drbd_device *device = from_timer(device, t, md_sync_timer);
3625 drbd_device_post_work(device, MD_SYNC);
3628 const char *cmdname(enum drbd_packet cmd)
3630 /* THINK may need to become several global tables
3631 * when we want to support more than
3632 * one PRO_VERSION */
3633 static const char *cmdnames[] = {
3636 [P_DATA_REPLY] = "DataReply",
3637 [P_RS_DATA_REPLY] = "RSDataReply",
3638 [P_BARRIER] = "Barrier",
3639 [P_BITMAP] = "ReportBitMap",
3640 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3641 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3642 [P_UNPLUG_REMOTE] = "UnplugRemote",
3643 [P_DATA_REQUEST] = "DataRequest",
3644 [P_RS_DATA_REQUEST] = "RSDataRequest",
3645 [P_SYNC_PARAM] = "SyncParam",
3646 [P_PROTOCOL] = "ReportProtocol",
3647 [P_UUIDS] = "ReportUUIDs",
3648 [P_SIZES] = "ReportSizes",
3649 [P_STATE] = "ReportState",
3650 [P_SYNC_UUID] = "ReportSyncUUID",
3651 [P_AUTH_CHALLENGE] = "AuthChallenge",
3652 [P_AUTH_RESPONSE] = "AuthResponse",
3653 [P_STATE_CHG_REQ] = "StateChgRequest",
3655 [P_PING_ACK] = "PingAck",
3656 [P_RECV_ACK] = "RecvAck",
3657 [P_WRITE_ACK] = "WriteAck",
3658 [P_RS_WRITE_ACK] = "RSWriteAck",
3659 [P_SUPERSEDED] = "Superseded",
3660 [P_NEG_ACK] = "NegAck",
3661 [P_NEG_DREPLY] = "NegDReply",
3662 [P_NEG_RS_DREPLY] = "NegRSDReply",
3663 [P_BARRIER_ACK] = "BarrierAck",
3664 [P_STATE_CHG_REPLY] = "StateChgReply",
3665 [P_OV_REQUEST] = "OVRequest",
3666 [P_OV_REPLY] = "OVReply",
3667 [P_OV_RESULT] = "OVResult",
3668 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3669 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3670 [P_SYNC_PARAM89] = "SyncParam89",
3671 [P_COMPRESSED_BITMAP] = "CBitmap",
3672 [P_DELAY_PROBE] = "DelayProbe",
3673 [P_OUT_OF_SYNC] = "OutOfSync",
3674 [P_RS_CANCEL] = "RSCancel",
3675 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3676 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3677 [P_RETRY_WRITE] = "retry_write",
3678 [P_PROTOCOL_UPDATE] = "protocol_update",
3680 [P_RS_THIN_REQ] = "rs_thin_req",
3681 [P_RS_DEALLOCATED] = "rs_deallocated",
3682 [P_WSAME] = "WriteSame",
3683 [P_ZEROES] = "Zeroes",
3685 /* enum drbd_packet, but not commands - obsoleted flags:
3691 /* too big for the array: 0xfffX */
3692 if (cmd == P_INITIAL_META)
3693 return "InitialMeta";
3694 if (cmd == P_INITIAL_DATA)
3695 return "InitialData";
3696 if (cmd == P_CONNECTION_FEATURES)
3697 return "ConnectionFeatures";
3698 if (cmd >= ARRAY_SIZE(cmdnames))
3700 return cmdnames[cmd];
3704 * drbd_wait_misc - wait for a request to make progress
3705 * @device: device associated with the request
3706 * @i: the struct drbd_interval embedded in struct drbd_request or
3707 * struct drbd_peer_request
3709 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3711 struct net_conf *nc;
3716 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3721 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3724 /* Indicate to wake up device->misc_wait on progress. */
3726 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3727 spin_unlock_irq(&device->resource->req_lock);
3728 timeout = schedule_timeout(timeout);
3729 finish_wait(&device->misc_wait, &wait);
3730 spin_lock_irq(&device->resource->req_lock);
3731 if (!timeout || device->state.conn < C_CONNECTED)
3733 if (signal_pending(current))
3734 return -ERESTARTSYS;
3738 void lock_all_resources(void)
3740 struct drbd_resource *resource;
3741 int __maybe_unused i = 0;
3743 mutex_lock(&resources_mutex);
3744 local_irq_disable();
3745 for_each_resource(resource, &drbd_resources)
3746 spin_lock_nested(&resource->req_lock, i++);
3749 void unlock_all_resources(void)
3751 struct drbd_resource *resource;
3753 for_each_resource(resource, &drbd_resources)
3754 spin_unlock(&resource->req_lock);
3756 mutex_unlock(&resources_mutex);
3759 #ifdef CONFIG_DRBD_FAULT_INJECTION
3760 /* Fault insertion support including random number generator shamelessly
3761 * stolen from kernel/rcutorture.c */
3762 struct fault_random_state {
3763 unsigned long state;
3764 unsigned long count;
3767 #define FAULT_RANDOM_MULT 39916801 /* prime */
3768 #define FAULT_RANDOM_ADD 479001701 /* prime */
3769 #define FAULT_RANDOM_REFRESH 10000
3772 * Crude but fast random-number generator. Uses a linear congruential
3773 * generator, with occasional help from get_random_bytes().
3775 static unsigned long
3776 _drbd_fault_random(struct fault_random_state *rsp)
3780 if (!rsp->count--) {
3781 get_random_bytes(&refresh, sizeof(refresh));
3782 rsp->state += refresh;
3783 rsp->count = FAULT_RANDOM_REFRESH;
3785 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3786 return swahw32(rsp->state);
3790 _drbd_fault_str(unsigned int type) {
3791 static char *_faults[] = {
3792 [DRBD_FAULT_MD_WR] = "Meta-data write",
3793 [DRBD_FAULT_MD_RD] = "Meta-data read",
3794 [DRBD_FAULT_RS_WR] = "Resync write",
3795 [DRBD_FAULT_RS_RD] = "Resync read",
3796 [DRBD_FAULT_DT_WR] = "Data write",
3797 [DRBD_FAULT_DT_RD] = "Data read",
3798 [DRBD_FAULT_DT_RA] = "Data read ahead",
3799 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3800 [DRBD_FAULT_AL_EE] = "EE allocation",
3801 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3804 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3808 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3810 static struct fault_random_state rrs = {0, 0};
3812 unsigned int ret = (
3813 (drbd_fault_devs == 0 ||
3814 ((1 << device_to_minor(device)) & drbd_fault_devs) != 0) &&
3815 (((_drbd_fault_random(&rrs) % 100) + 1) <= drbd_fault_rate));
3820 if (__ratelimit(&drbd_ratelimit_state))
3821 drbd_warn(device, "***Simulating %s failure\n",
3822 _drbd_fault_str(type));
3829 const char *drbd_buildtag(void)
3831 /* DRBD built from external sources has here a reference to the
3832 git hash of the source code. */
3834 static char buildtag[38] = "\0uilt-in";
3836 if (buildtag[0] == 0) {
3838 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3847 module_init(drbd_init)
3848 module_exit(drbd_cleanup)
3850 EXPORT_SYMBOL(drbd_conn_str);
3851 EXPORT_SYMBOL(drbd_role_str);
3852 EXPORT_SYMBOL(drbd_disk_str);
3853 EXPORT_SYMBOL(drbd_set_st_err_str);