4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
11 from Logicworks, Inc. for making SDP replication support possible.
13 drbd is free software; you can redistribute it and/or modify
14 it under the terms of the GNU General Public License as published by
15 the Free Software Foundation; either version 2, or (at your option)
18 drbd is distributed in the hope that it will be useful,
19 but WITHOUT ANY WARRANTY; without even the implied warranty of
20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 GNU General Public License for more details.
23 You should have received a copy of the GNU General Public License
24 along with drbd; see the file COPYING. If not, write to
25 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/module.h>
32 #include <linux/jiffies.h>
33 #include <linux/drbd.h>
34 #include <linux/uaccess.h>
35 #include <asm/types.h>
37 #include <linux/ctype.h>
38 #include <linux/mutex.h>
40 #include <linux/file.h>
41 #include <linux/proc_fs.h>
42 #include <linux/init.h>
44 #include <linux/memcontrol.h>
45 #include <linux/mm_inline.h>
46 #include <linux/slab.h>
47 #include <linux/random.h>
48 #include <linux/reboot.h>
49 #include <linux/notifier.h>
50 #include <linux/kthread.h>
51 #include <linux/workqueue.h>
52 #define __KERNEL_SYSCALLS__
53 #include <linux/unistd.h>
54 #include <linux/vmalloc.h>
56 #include <linux/drbd_limits.h>
58 #include "drbd_protocol.h"
59 #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
61 #include "drbd_debugfs.h"
63 static DEFINE_MUTEX(drbd_main_mutex);
64 static int drbd_open(struct block_device *bdev, fmode_t mode);
65 static void drbd_release(struct gendisk *gd, fmode_t mode);
66 static void md_sync_timer_fn(unsigned long data);
67 static int w_bitmap_io(struct drbd_work *w, int unused);
69 MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
70 "Lars Ellenberg <lars@linbit.com>");
71 MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
72 MODULE_VERSION(REL_VERSION);
73 MODULE_LICENSE("GPL");
74 MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
75 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
76 MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
78 #include <linux/moduleparam.h>
79 /* allow_open_on_secondary */
80 MODULE_PARM_DESC(allow_oos, "DONT USE!");
81 /* thanks to these macros, if compiled into the kernel (not-module),
82 * this becomes the boot parameter drbd.minor_count */
83 module_param(minor_count, uint, 0444);
84 module_param(disable_sendpage, bool, 0644);
85 module_param(allow_oos, bool, 0);
86 module_param(proc_details, int, 0644);
88 #ifdef CONFIG_DRBD_FAULT_INJECTION
91 static int fault_count;
93 /* bitmap of enabled faults */
94 module_param(enable_faults, int, 0664);
95 /* fault rate % value - applies to all enabled faults */
96 module_param(fault_rate, int, 0664);
97 /* count of faults inserted */
98 module_param(fault_count, int, 0664);
99 /* bitmap of devices to insert faults on */
100 module_param(fault_devs, int, 0644);
103 /* module parameter, defined */
104 unsigned int minor_count = DRBD_MINOR_COUNT_DEF;
105 bool disable_sendpage;
107 int proc_details; /* Detail level in proc drbd*/
109 /* Module parameter for setting the user mode helper program
110 * to run. Default is /sbin/drbdadm */
111 char usermode_helper[80] = "/sbin/drbdadm";
113 module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644);
115 /* in 2.6.x, our device mapping and config info contains our virtual gendisks
116 * as member "struct gendisk *vdisk;"
118 struct idr drbd_devices;
119 struct list_head drbd_resources;
120 struct mutex resources_mutex;
122 struct kmem_cache *drbd_request_cache;
123 struct kmem_cache *drbd_ee_cache; /* peer requests */
124 struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
125 struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
126 mempool_t *drbd_request_mempool;
127 mempool_t *drbd_ee_mempool;
128 mempool_t *drbd_md_io_page_pool;
129 struct bio_set *drbd_md_io_bio_set;
131 /* I do not use a standard mempool, because:
132 1) I want to hand out the pre-allocated objects first.
133 2) I want to be able to interrupt sleeping allocation with a signal.
134 Note: This is a single linked list, the next pointer is the private
135 member of struct page.
137 struct page *drbd_pp_pool;
138 spinlock_t drbd_pp_lock;
140 wait_queue_head_t drbd_pp_wait;
142 DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
144 static const struct block_device_operations drbd_ops = {
145 .owner = THIS_MODULE,
147 .release = drbd_release,
150 struct bio *bio_alloc_drbd(gfp_t gfp_mask)
154 if (!drbd_md_io_bio_set)
155 return bio_alloc(gfp_mask, 1);
157 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
164 /* When checking with sparse, and this is an inline function, sparse will
165 give tons of false positives. When this is a real functions sparse works.
167 int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
171 atomic_inc(&device->local_cnt);
172 io_allowed = (device->state.disk >= mins);
174 if (atomic_dec_and_test(&device->local_cnt))
175 wake_up(&device->misc_wait);
183 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
184 * @connection: DRBD connection.
185 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
186 * @set_size: Expected number of requests before that barrier.
188 * In case the passed barrier_nr or set_size does not match the oldest
189 * epoch of not yet barrier-acked requests, this function will cause a
190 * termination of the connection.
192 void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
193 unsigned int set_size)
195 struct drbd_request *r;
196 struct drbd_request *req = NULL;
197 int expect_epoch = 0;
200 spin_lock_irq(&connection->resource->req_lock);
202 /* find oldest not yet barrier-acked write request,
203 * count writes in its epoch. */
204 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
205 const unsigned s = r->rq_state;
209 if (!(s & RQ_NET_MASK))
214 expect_epoch = req->epoch;
217 if (r->epoch != expect_epoch)
221 /* if (s & RQ_DONE): not expected */
222 /* if (!(s & RQ_NET_MASK)): not expected */
227 /* first some paranoia code */
229 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
233 if (expect_epoch != barrier_nr) {
234 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
235 barrier_nr, expect_epoch);
239 if (expect_size != set_size) {
240 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
241 barrier_nr, set_size, expect_size);
245 /* Clean up list of requests processed during current epoch. */
246 /* this extra list walk restart is paranoia,
247 * to catch requests being barrier-acked "unexpectedly".
248 * It usually should find the same req again, or some READ preceding it. */
249 list_for_each_entry(req, &connection->transfer_log, tl_requests)
250 if (req->epoch == expect_epoch)
252 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
253 if (req->epoch != expect_epoch)
255 _req_mod(req, BARRIER_ACKED);
257 spin_unlock_irq(&connection->resource->req_lock);
262 spin_unlock_irq(&connection->resource->req_lock);
263 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
268 * _tl_restart() - Walks the transfer log, and applies an action to all requests
269 * @connection: DRBD connection to operate on.
270 * @what: The action/event to perform with all request objects
272 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
273 * RESTART_FROZEN_DISK_IO.
275 /* must hold resource->req_lock */
276 void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
278 struct drbd_request *req, *r;
280 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
284 void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
286 spin_lock_irq(&connection->resource->req_lock);
287 _tl_restart(connection, what);
288 spin_unlock_irq(&connection->resource->req_lock);
292 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
293 * @device: DRBD device.
295 * This is called after the connection to the peer was lost. The storage covered
296 * by the requests on the transfer gets marked as our of sync. Called from the
297 * receiver thread and the worker thread.
299 void tl_clear(struct drbd_connection *connection)
301 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
305 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
306 * @device: DRBD device.
308 void tl_abort_disk_io(struct drbd_device *device)
310 struct drbd_connection *connection = first_peer_device(device)->connection;
311 struct drbd_request *req, *r;
313 spin_lock_irq(&connection->resource->req_lock);
314 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
315 if (!(req->rq_state & RQ_LOCAL_PENDING))
317 if (req->device != device)
319 _req_mod(req, ABORT_DISK_IO);
321 spin_unlock_irq(&connection->resource->req_lock);
324 static int drbd_thread_setup(void *arg)
326 struct drbd_thread *thi = (struct drbd_thread *) arg;
327 struct drbd_resource *resource = thi->resource;
331 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
335 allow_kernel_signal(DRBD_SIGKILL);
336 allow_kernel_signal(SIGXCPU);
338 retval = thi->function(thi);
340 spin_lock_irqsave(&thi->t_lock, flags);
342 /* if the receiver has been "EXITING", the last thing it did
343 * was set the conn state to "StandAlone",
344 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
345 * and receiver thread will be "started".
346 * drbd_thread_start needs to set "RESTARTING" in that case.
347 * t_state check and assignment needs to be within the same spinlock,
348 * so either thread_start sees EXITING, and can remap to RESTARTING,
349 * or thread_start see NONE, and can proceed as normal.
352 if (thi->t_state == RESTARTING) {
353 drbd_info(resource, "Restarting %s thread\n", thi->name);
354 thi->t_state = RUNNING;
355 spin_unlock_irqrestore(&thi->t_lock, flags);
362 complete_all(&thi->stop);
363 spin_unlock_irqrestore(&thi->t_lock, flags);
365 drbd_info(resource, "Terminating %s\n", current->comm);
367 /* Release mod reference taken when thread was started */
370 kref_put(&thi->connection->kref, drbd_destroy_connection);
371 kref_put(&resource->kref, drbd_destroy_resource);
372 module_put(THIS_MODULE);
376 static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
377 int (*func) (struct drbd_thread *), const char *name)
379 spin_lock_init(&thi->t_lock);
382 thi->function = func;
383 thi->resource = resource;
384 thi->connection = NULL;
388 int drbd_thread_start(struct drbd_thread *thi)
390 struct drbd_resource *resource = thi->resource;
391 struct task_struct *nt;
394 /* is used from state engine doing drbd_thread_stop_nowait,
395 * while holding the req lock irqsave */
396 spin_lock_irqsave(&thi->t_lock, flags);
398 switch (thi->t_state) {
400 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
401 thi->name, current->comm, current->pid);
403 /* Get ref on module for thread - this is released when thread exits */
404 if (!try_module_get(THIS_MODULE)) {
405 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
406 spin_unlock_irqrestore(&thi->t_lock, flags);
410 kref_get(&resource->kref);
412 kref_get(&thi->connection->kref);
414 init_completion(&thi->stop);
415 thi->reset_cpu_mask = 1;
416 thi->t_state = RUNNING;
417 spin_unlock_irqrestore(&thi->t_lock, flags);
418 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
420 nt = kthread_create(drbd_thread_setup, (void *) thi,
421 "drbd_%c_%s", thi->name[0], thi->resource->name);
424 drbd_err(resource, "Couldn't start thread\n");
427 kref_put(&thi->connection->kref, drbd_destroy_connection);
428 kref_put(&resource->kref, drbd_destroy_resource);
429 module_put(THIS_MODULE);
432 spin_lock_irqsave(&thi->t_lock, flags);
434 thi->t_state = RUNNING;
435 spin_unlock_irqrestore(&thi->t_lock, flags);
439 thi->t_state = RESTARTING;
440 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
441 thi->name, current->comm, current->pid);
446 spin_unlock_irqrestore(&thi->t_lock, flags);
454 void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
458 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
460 /* may be called from state engine, holding the req lock irqsave */
461 spin_lock_irqsave(&thi->t_lock, flags);
463 if (thi->t_state == NONE) {
464 spin_unlock_irqrestore(&thi->t_lock, flags);
466 drbd_thread_start(thi);
470 if (thi->t_state != ns) {
471 if (thi->task == NULL) {
472 spin_unlock_irqrestore(&thi->t_lock, flags);
478 init_completion(&thi->stop);
479 if (thi->task != current)
480 force_sig(DRBD_SIGKILL, thi->task);
483 spin_unlock_irqrestore(&thi->t_lock, flags);
486 wait_for_completion(&thi->stop);
489 int conn_lowest_minor(struct drbd_connection *connection)
491 struct drbd_peer_device *peer_device;
492 int vnr = 0, minor = -1;
495 peer_device = idr_get_next(&connection->peer_devices, &vnr);
497 minor = device_to_minor(peer_device->device);
505 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
507 * Forces all threads of a resource onto the same CPU. This is beneficial for
508 * DRBD's performance. May be overwritten by user's configuration.
510 static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
512 unsigned int *resources_per_cpu, min_index = ~0;
514 resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL);
515 if (resources_per_cpu) {
516 struct drbd_resource *resource;
517 unsigned int cpu, min = ~0;
520 for_each_resource_rcu(resource, &drbd_resources) {
521 for_each_cpu(cpu, resource->cpu_mask)
522 resources_per_cpu[cpu]++;
525 for_each_online_cpu(cpu) {
526 if (resources_per_cpu[cpu] < min) {
527 min = resources_per_cpu[cpu];
531 kfree(resources_per_cpu);
533 if (min_index == ~0) {
534 cpumask_setall(*cpu_mask);
537 cpumask_set_cpu(min_index, *cpu_mask);
541 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
542 * @device: DRBD device.
543 * @thi: drbd_thread object
545 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
548 void drbd_thread_current_set_cpu(struct drbd_thread *thi)
550 struct drbd_resource *resource = thi->resource;
551 struct task_struct *p = current;
553 if (!thi->reset_cpu_mask)
555 thi->reset_cpu_mask = 0;
556 set_cpus_allowed_ptr(p, resource->cpu_mask);
559 #define drbd_calc_cpu_mask(A) ({})
563 * drbd_header_size - size of a packet header
565 * The header size is a multiple of 8, so any payload following the header is
566 * word aligned on 64-bit architectures. (The bitmap send and receive code
569 unsigned int drbd_header_size(struct drbd_connection *connection)
571 if (connection->agreed_pro_version >= 100) {
572 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
573 return sizeof(struct p_header100);
575 BUILD_BUG_ON(sizeof(struct p_header80) !=
576 sizeof(struct p_header95));
577 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
578 return sizeof(struct p_header80);
582 static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
584 h->magic = cpu_to_be32(DRBD_MAGIC);
585 h->command = cpu_to_be16(cmd);
586 h->length = cpu_to_be16(size);
587 return sizeof(struct p_header80);
590 static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
592 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
593 h->command = cpu_to_be16(cmd);
594 h->length = cpu_to_be32(size);
595 return sizeof(struct p_header95);
598 static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
601 h->magic = cpu_to_be32(DRBD_MAGIC_100);
602 h->volume = cpu_to_be16(vnr);
603 h->command = cpu_to_be16(cmd);
604 h->length = cpu_to_be32(size);
606 return sizeof(struct p_header100);
609 static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
610 void *buffer, enum drbd_packet cmd, int size)
612 if (connection->agreed_pro_version >= 100)
613 return prepare_header100(buffer, cmd, size, vnr);
614 else if (connection->agreed_pro_version >= 95 &&
615 size > DRBD_MAX_SIZE_H80_PACKET)
616 return prepare_header95(buffer, cmd, size);
618 return prepare_header80(buffer, cmd, size);
621 static void *__conn_prepare_command(struct drbd_connection *connection,
622 struct drbd_socket *sock)
626 return sock->sbuf + drbd_header_size(connection);
629 void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
633 mutex_lock(&sock->mutex);
634 p = __conn_prepare_command(connection, sock);
636 mutex_unlock(&sock->mutex);
641 void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
643 return conn_prepare_command(peer_device->connection, sock);
646 static int __send_command(struct drbd_connection *connection, int vnr,
647 struct drbd_socket *sock, enum drbd_packet cmd,
648 unsigned int header_size, void *data,
655 * Called with @data == NULL and the size of the data blocks in @size
656 * for commands that send data blocks. For those commands, omit the
657 * MSG_MORE flag: this will increase the likelihood that data blocks
658 * which are page aligned on the sender will end up page aligned on the
661 msg_flags = data ? MSG_MORE : 0;
663 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
665 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
668 err = drbd_send_all(connection, sock->socket, data, size, 0);
669 /* DRBD protocol "pings" are latency critical.
670 * This is supposed to trigger tcp_push_pending_frames() */
671 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
672 drbd_tcp_nodelay(sock->socket);
677 static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
678 enum drbd_packet cmd, unsigned int header_size,
679 void *data, unsigned int size)
681 return __send_command(connection, 0, sock, cmd, header_size, data, size);
684 int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
685 enum drbd_packet cmd, unsigned int header_size,
686 void *data, unsigned int size)
690 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
691 mutex_unlock(&sock->mutex);
695 int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
696 enum drbd_packet cmd, unsigned int header_size,
697 void *data, unsigned int size)
701 err = __send_command(peer_device->connection, peer_device->device->vnr,
702 sock, cmd, header_size, data, size);
703 mutex_unlock(&sock->mutex);
707 int drbd_send_ping(struct drbd_connection *connection)
709 struct drbd_socket *sock;
711 sock = &connection->meta;
712 if (!conn_prepare_command(connection, sock))
714 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
717 int drbd_send_ping_ack(struct drbd_connection *connection)
719 struct drbd_socket *sock;
721 sock = &connection->meta;
722 if (!conn_prepare_command(connection, sock))
724 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
727 int drbd_send_sync_param(struct drbd_peer_device *peer_device)
729 struct drbd_socket *sock;
730 struct p_rs_param_95 *p;
732 const int apv = peer_device->connection->agreed_pro_version;
733 enum drbd_packet cmd;
735 struct disk_conf *dc;
737 sock = &peer_device->connection->data;
738 p = drbd_prepare_command(peer_device, sock);
743 nc = rcu_dereference(peer_device->connection->net_conf);
745 size = apv <= 87 ? sizeof(struct p_rs_param)
746 : apv == 88 ? sizeof(struct p_rs_param)
747 + strlen(nc->verify_alg) + 1
748 : apv <= 94 ? sizeof(struct p_rs_param_89)
749 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
751 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
753 /* initialize verify_alg and csums_alg */
754 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
756 if (get_ldev(peer_device->device)) {
757 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
758 p->resync_rate = cpu_to_be32(dc->resync_rate);
759 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
760 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
761 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
762 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
763 put_ldev(peer_device->device);
765 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
766 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
767 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
768 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
769 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
773 strcpy(p->verify_alg, nc->verify_alg);
775 strcpy(p->csums_alg, nc->csums_alg);
778 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
781 int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
783 struct drbd_socket *sock;
784 struct p_protocol *p;
788 sock = &connection->data;
789 p = __conn_prepare_command(connection, sock);
794 nc = rcu_dereference(connection->net_conf);
796 if (nc->tentative && connection->agreed_pro_version < 92) {
798 drbd_err(connection, "--dry-run is not supported by peer");
803 if (connection->agreed_pro_version >= 87)
804 size += strlen(nc->integrity_alg) + 1;
806 p->protocol = cpu_to_be32(nc->wire_protocol);
807 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
808 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
809 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
810 p->two_primaries = cpu_to_be32(nc->two_primaries);
812 if (nc->discard_my_data)
813 cf |= CF_DISCARD_MY_DATA;
816 p->conn_flags = cpu_to_be32(cf);
818 if (connection->agreed_pro_version >= 87)
819 strcpy(p->integrity_alg, nc->integrity_alg);
822 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
825 int drbd_send_protocol(struct drbd_connection *connection)
829 mutex_lock(&connection->data.mutex);
830 err = __drbd_send_protocol(connection, P_PROTOCOL);
831 mutex_unlock(&connection->data.mutex);
836 static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
838 struct drbd_device *device = peer_device->device;
839 struct drbd_socket *sock;
843 if (!get_ldev_if_state(device, D_NEGOTIATING))
846 sock = &peer_device->connection->data;
847 p = drbd_prepare_command(peer_device, sock);
852 spin_lock_irq(&device->ldev->md.uuid_lock);
853 for (i = UI_CURRENT; i < UI_SIZE; i++)
854 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
855 spin_unlock_irq(&device->ldev->md.uuid_lock);
857 device->comm_bm_set = drbd_bm_total_weight(device);
858 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
860 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
862 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
863 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
864 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
867 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
870 int drbd_send_uuids(struct drbd_peer_device *peer_device)
872 return _drbd_send_uuids(peer_device, 0);
875 int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
877 return _drbd_send_uuids(peer_device, 8);
880 void drbd_print_uuids(struct drbd_device *device, const char *text)
882 if (get_ldev_if_state(device, D_NEGOTIATING)) {
883 u64 *uuid = device->ldev->md.uuid;
884 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
886 (unsigned long long)uuid[UI_CURRENT],
887 (unsigned long long)uuid[UI_BITMAP],
888 (unsigned long long)uuid[UI_HISTORY_START],
889 (unsigned long long)uuid[UI_HISTORY_END]);
892 drbd_info(device, "%s effective data uuid: %016llX\n",
894 (unsigned long long)device->ed_uuid);
898 void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
900 struct drbd_device *device = peer_device->device;
901 struct drbd_socket *sock;
905 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
907 uuid = device->ldev->md.uuid[UI_BITMAP];
908 if (uuid && uuid != UUID_JUST_CREATED)
909 uuid = uuid + UUID_NEW_BM_OFFSET;
911 get_random_bytes(&uuid, sizeof(u64));
912 drbd_uuid_set(device, UI_BITMAP, uuid);
913 drbd_print_uuids(device, "updated sync UUID");
914 drbd_md_sync(device);
916 sock = &peer_device->connection->data;
917 p = drbd_prepare_command(peer_device, sock);
919 p->uuid = cpu_to_be64(uuid);
920 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
924 /* communicated if (agreed_features & DRBD_FF_WSAME) */
925 void assign_p_sizes_qlim(struct drbd_device *device, struct p_sizes *p, struct request_queue *q)
928 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
929 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
930 p->qlim->alignment_offset = cpu_to_be32(queue_alignment_offset(q));
931 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
932 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
933 p->qlim->discard_enabled = blk_queue_discard(q);
934 p->qlim->discard_zeroes_data = queue_discard_zeroes_data(q);
935 p->qlim->write_same_capable = !!q->limits.max_write_same_sectors;
937 q = device->rq_queue;
938 p->qlim->physical_block_size = cpu_to_be32(queue_physical_block_size(q));
939 p->qlim->logical_block_size = cpu_to_be32(queue_logical_block_size(q));
940 p->qlim->alignment_offset = 0;
941 p->qlim->io_min = cpu_to_be32(queue_io_min(q));
942 p->qlim->io_opt = cpu_to_be32(queue_io_opt(q));
943 p->qlim->discard_enabled = 0;
944 p->qlim->discard_zeroes_data = 0;
945 p->qlim->write_same_capable = 0;
949 int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
951 struct drbd_device *device = peer_device->device;
952 struct drbd_socket *sock;
954 sector_t d_size, u_size;
956 unsigned int max_bio_size;
957 unsigned int packet_size;
959 sock = &peer_device->connection->data;
960 p = drbd_prepare_command(peer_device, sock);
964 packet_size = sizeof(*p);
965 if (peer_device->connection->agreed_features & DRBD_FF_WSAME)
966 packet_size += sizeof(p->qlim[0]);
968 memset(p, 0, packet_size);
969 if (get_ldev_if_state(device, D_NEGOTIATING)) {
970 struct request_queue *q = bdev_get_queue(device->ldev->backing_bdev);
971 d_size = drbd_get_max_capacity(device->ldev);
973 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
975 q_order_type = drbd_queue_order_type(device);
976 max_bio_size = queue_max_hw_sectors(q) << 9;
977 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
978 assign_p_sizes_qlim(device, p, q);
983 q_order_type = QUEUE_ORDERED_NONE;
984 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
985 assign_p_sizes_qlim(device, p, NULL);
988 if (peer_device->connection->agreed_pro_version <= 94)
989 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
990 else if (peer_device->connection->agreed_pro_version < 100)
991 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
993 p->d_size = cpu_to_be64(d_size);
994 p->u_size = cpu_to_be64(u_size);
995 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
996 p->max_bio_size = cpu_to_be32(max_bio_size);
997 p->queue_order_type = cpu_to_be16(q_order_type);
998 p->dds_flags = cpu_to_be16(flags);
1000 return drbd_send_command(peer_device, sock, P_SIZES, packet_size, NULL, 0);
1004 * drbd_send_current_state() - Sends the drbd state to the peer
1005 * @peer_device: DRBD peer device.
1007 int drbd_send_current_state(struct drbd_peer_device *peer_device)
1009 struct drbd_socket *sock;
1012 sock = &peer_device->connection->data;
1013 p = drbd_prepare_command(peer_device, sock);
1016 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
1017 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1021 * drbd_send_state() - After a state change, sends the new state to the peer
1022 * @peer_device: DRBD peer device.
1023 * @state: the state to send, not necessarily the current state.
1025 * Each state change queues an "after_state_ch" work, which will eventually
1026 * send the resulting new state to the peer. If more state changes happen
1027 * between queuing and processing of the after_state_ch work, we still
1028 * want to send each intermediary state in the order it occurred.
1030 int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
1032 struct drbd_socket *sock;
1035 sock = &peer_device->connection->data;
1036 p = drbd_prepare_command(peer_device, sock);
1039 p->state = cpu_to_be32(state.i); /* Within the send mutex */
1040 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
1043 int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
1045 struct drbd_socket *sock;
1046 struct p_req_state *p;
1048 sock = &peer_device->connection->data;
1049 p = drbd_prepare_command(peer_device, sock);
1052 p->mask = cpu_to_be32(mask.i);
1053 p->val = cpu_to_be32(val.i);
1054 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
1057 int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
1059 enum drbd_packet cmd;
1060 struct drbd_socket *sock;
1061 struct p_req_state *p;
1063 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1064 sock = &connection->data;
1065 p = conn_prepare_command(connection, sock);
1068 p->mask = cpu_to_be32(mask.i);
1069 p->val = cpu_to_be32(val.i);
1070 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1073 void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
1075 struct drbd_socket *sock;
1076 struct p_req_state_reply *p;
1078 sock = &peer_device->connection->meta;
1079 p = drbd_prepare_command(peer_device, sock);
1081 p->retcode = cpu_to_be32(retcode);
1082 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
1086 void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
1088 struct drbd_socket *sock;
1089 struct p_req_state_reply *p;
1090 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
1092 sock = &connection->meta;
1093 p = conn_prepare_command(connection, sock);
1095 p->retcode = cpu_to_be32(retcode);
1096 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
1100 static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
1102 BUG_ON(code & ~0xf);
1103 p->encoding = (p->encoding & ~0xf) | code;
1106 static void dcbp_set_start(struct p_compressed_bm *p, int set)
1108 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1111 static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1114 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
1117 static int fill_bitmap_rle_bits(struct drbd_device *device,
1118 struct p_compressed_bm *p,
1120 struct bm_xfer_ctx *c)
1122 struct bitstream bs;
1123 unsigned long plain_bits;
1130 /* may we use this feature? */
1132 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
1134 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
1137 if (c->bit_offset >= c->bm_bits)
1138 return 0; /* nothing to do. */
1140 /* use at most thus many bytes */
1141 bitstream_init(&bs, p->code, size, 0);
1142 memset(p->code, 0, size);
1143 /* plain bits covered in this code string */
1146 /* p->encoding & 0x80 stores whether the first run length is set.
1147 * bit offset is implicit.
1148 * start with toggle == 2 to be able to tell the first iteration */
1151 /* see how much plain bits we can stuff into one packet
1152 * using RLE and VLI. */
1154 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1155 : _drbd_bm_find_next(device, c->bit_offset);
1158 rl = tmp - c->bit_offset;
1160 if (toggle == 2) { /* first iteration */
1162 /* the first checked bit was set,
1163 * store start value, */
1164 dcbp_set_start(p, 1);
1165 /* but skip encoding of zero run length */
1169 dcbp_set_start(p, 0);
1172 /* paranoia: catch zero runlength.
1173 * can only happen if bitmap is modified while we scan it. */
1175 drbd_err(device, "unexpected zero runlength while encoding bitmap "
1176 "t:%u bo:%lu\n", toggle, c->bit_offset);
1180 bits = vli_encode_bits(&bs, rl);
1181 if (bits == -ENOBUFS) /* buffer full */
1184 drbd_err(device, "error while encoding bitmap: %d\n", bits);
1190 c->bit_offset = tmp;
1191 } while (c->bit_offset < c->bm_bits);
1193 len = bs.cur.b - p->code + !!bs.cur.bit;
1195 if (plain_bits < (len << 3)) {
1196 /* incompressible with this method.
1197 * we need to rewind both word and bit position. */
1198 c->bit_offset -= plain_bits;
1199 bm_xfer_ctx_bit_to_word_offset(c);
1200 c->bit_offset = c->word_offset * BITS_PER_LONG;
1204 /* RLE + VLI was able to compress it just fine.
1205 * update c->word_offset. */
1206 bm_xfer_ctx_bit_to_word_offset(c);
1208 /* store pad_bits */
1209 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
1215 * send_bitmap_rle_or_plain
1217 * Return 0 when done, 1 when another iteration is needed, and a negative error
1218 * code upon failure.
1221 send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
1223 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1224 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
1225 struct p_compressed_bm *p = sock->sbuf + header_size;
1228 len = fill_bitmap_rle_bits(device, p,
1229 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
1234 dcbp_set_code(p, RLE_VLI_Bits);
1235 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
1236 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1239 c->bytes[0] += header_size + sizeof(*p) + len;
1241 if (c->bit_offset >= c->bm_bits)
1244 /* was not compressible.
1245 * send a buffer full of plain text bits instead. */
1246 unsigned int data_size;
1247 unsigned long num_words;
1248 unsigned long *p = sock->sbuf + header_size;
1250 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
1251 num_words = min_t(size_t, data_size / sizeof(*p),
1252 c->bm_words - c->word_offset);
1253 len = num_words * sizeof(*p);
1255 drbd_bm_get_lel(device, c->word_offset, num_words, p);
1256 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
1257 c->word_offset += num_words;
1258 c->bit_offset = c->word_offset * BITS_PER_LONG;
1261 c->bytes[1] += header_size + len;
1263 if (c->bit_offset > c->bm_bits)
1264 c->bit_offset = c->bm_bits;
1268 INFO_bm_xfer_stats(device, "send", c);
1276 /* See the comment at receive_bitmap() */
1277 static int _drbd_send_bitmap(struct drbd_device *device)
1279 struct bm_xfer_ctx c;
1282 if (!expect(device->bitmap))
1285 if (get_ldev(device)) {
1286 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
1287 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
1288 drbd_bm_set_all(device);
1289 if (drbd_bm_write(device)) {
1290 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1291 * but otherwise process as per normal - need to tell other
1292 * side that a full resync is required! */
1293 drbd_err(device, "Failed to write bitmap to disk!\n");
1295 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1296 drbd_md_sync(device);
1302 c = (struct bm_xfer_ctx) {
1303 .bm_bits = drbd_bm_bits(device),
1304 .bm_words = drbd_bm_words(device),
1308 err = send_bitmap_rle_or_plain(device, &c);
1314 int drbd_send_bitmap(struct drbd_device *device)
1316 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1319 mutex_lock(&sock->mutex);
1321 err = !_drbd_send_bitmap(device);
1322 mutex_unlock(&sock->mutex);
1326 void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
1328 struct drbd_socket *sock;
1329 struct p_barrier_ack *p;
1331 if (connection->cstate < C_WF_REPORT_PARAMS)
1334 sock = &connection->meta;
1335 p = conn_prepare_command(connection, sock);
1338 p->barrier = barrier_nr;
1339 p->set_size = cpu_to_be32(set_size);
1340 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
1344 * _drbd_send_ack() - Sends an ack packet
1345 * @device: DRBD device.
1346 * @cmd: Packet command code.
1347 * @sector: sector, needs to be in big endian byte order
1348 * @blksize: size in byte, needs to be in big endian byte order
1349 * @block_id: Id, big endian byte order
1351 static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1352 u64 sector, u32 blksize, u64 block_id)
1354 struct drbd_socket *sock;
1355 struct p_block_ack *p;
1357 if (peer_device->device->state.conn < C_CONNECTED)
1360 sock = &peer_device->connection->meta;
1361 p = drbd_prepare_command(peer_device, sock);
1365 p->block_id = block_id;
1366 p->blksize = blksize;
1367 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1368 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1371 /* dp->sector and dp->block_id already/still in network byte order,
1372 * data_size is payload size according to dp->head,
1373 * and may need to be corrected for digest size. */
1374 void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1375 struct p_data *dp, int data_size)
1377 if (peer_device->connection->peer_integrity_tfm)
1378 data_size -= crypto_ahash_digestsize(peer_device->connection->peer_integrity_tfm);
1379 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
1383 void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1384 struct p_block_req *rp)
1386 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
1390 * drbd_send_ack() - Sends an ack packet
1391 * @device: DRBD device
1392 * @cmd: packet command code
1393 * @peer_req: peer request
1395 int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1396 struct drbd_peer_request *peer_req)
1398 return _drbd_send_ack(peer_device, cmd,
1399 cpu_to_be64(peer_req->i.sector),
1400 cpu_to_be32(peer_req->i.size),
1401 peer_req->block_id);
1404 /* This function misuses the block_id field to signal if the blocks
1405 * are is sync or not. */
1406 int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1407 sector_t sector, int blksize, u64 block_id)
1409 return _drbd_send_ack(peer_device, cmd,
1410 cpu_to_be64(sector),
1411 cpu_to_be32(blksize),
1412 cpu_to_be64(block_id));
1415 int drbd_send_rs_deallocated(struct drbd_peer_device *peer_device,
1416 struct drbd_peer_request *peer_req)
1418 struct drbd_socket *sock;
1419 struct p_block_desc *p;
1421 sock = &peer_device->connection->data;
1422 p = drbd_prepare_command(peer_device, sock);
1425 p->sector = cpu_to_be64(peer_req->i.sector);
1426 p->blksize = cpu_to_be32(peer_req->i.size);
1428 return drbd_send_command(peer_device, sock, P_RS_DEALLOCATED, sizeof(*p), NULL, 0);
1431 int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
1432 sector_t sector, int size, u64 block_id)
1434 struct drbd_socket *sock;
1435 struct p_block_req *p;
1437 sock = &peer_device->connection->data;
1438 p = drbd_prepare_command(peer_device, sock);
1441 p->sector = cpu_to_be64(sector);
1442 p->block_id = block_id;
1443 p->blksize = cpu_to_be32(size);
1444 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
1447 int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
1448 void *digest, int digest_size, enum drbd_packet cmd)
1450 struct drbd_socket *sock;
1451 struct p_block_req *p;
1453 /* FIXME: Put the digest into the preallocated socket buffer. */
1455 sock = &peer_device->connection->data;
1456 p = drbd_prepare_command(peer_device, sock);
1459 p->sector = cpu_to_be64(sector);
1460 p->block_id = ID_SYNCER /* unused */;
1461 p->blksize = cpu_to_be32(size);
1462 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
1465 int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
1467 struct drbd_socket *sock;
1468 struct p_block_req *p;
1470 sock = &peer_device->connection->data;
1471 p = drbd_prepare_command(peer_device, sock);
1474 p->sector = cpu_to_be64(sector);
1475 p->block_id = ID_SYNCER /* unused */;
1476 p->blksize = cpu_to_be32(size);
1477 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
1480 /* called on sndtimeo
1481 * returns false if we should retry,
1482 * true if we think connection is dead
1484 static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
1487 /* long elapsed = (long)(jiffies - device->last_received); */
1489 drop_it = connection->meta.socket == sock
1490 || !connection->ack_receiver.task
1491 || get_t_state(&connection->ack_receiver) != RUNNING
1492 || connection->cstate < C_WF_REPORT_PARAMS;
1497 drop_it = !--connection->ko_count;
1499 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
1500 current->comm, current->pid, connection->ko_count);
1501 request_ping(connection);
1504 return drop_it; /* && (device->state == R_PRIMARY) */;
1507 static void drbd_update_congested(struct drbd_connection *connection)
1509 struct sock *sk = connection->data.socket->sk;
1510 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
1511 set_bit(NET_CONGESTED, &connection->flags);
1514 /* The idea of sendpage seems to be to put some kind of reference
1515 * to the page into the skb, and to hand it over to the NIC. In
1516 * this process get_page() gets called.
1518 * As soon as the page was really sent over the network put_page()
1519 * gets called by some part of the network layer. [ NIC driver? ]
1521 * [ get_page() / put_page() increment/decrement the count. If count
1522 * reaches 0 the page will be freed. ]
1524 * This works nicely with pages from FSs.
1525 * But this means that in protocol A we might signal IO completion too early!
1527 * In order not to corrupt data during a resync we must make sure
1528 * that we do not reuse our own buffer pages (EEs) to early, therefore
1529 * we have the net_ee list.
1531 * XFS seems to have problems, still, it submits pages with page_count == 0!
1532 * As a workaround, we disable sendpage on pages
1533 * with page_count == 0 or PageSlab.
1535 static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
1536 int offset, size_t size, unsigned msg_flags)
1538 struct socket *socket;
1542 socket = peer_device->connection->data.socket;
1543 addr = kmap(page) + offset;
1544 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
1547 peer_device->device->send_cnt += size >> 9;
1551 static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
1552 int offset, size_t size, unsigned msg_flags)
1554 struct socket *socket = peer_device->connection->data.socket;
1555 mm_segment_t oldfs = get_fs();
1559 /* e.g. XFS meta- & log-data is in slab pages, which have a
1560 * page_count of 0 and/or have PageSlab() set.
1561 * we cannot use send_page for those, as that does get_page();
1562 * put_page(); and would cause either a VM_BUG directly, or
1563 * __page_cache_release a page that would actually still be referenced
1564 * by someone, leading to some obscure delayed Oops somewhere else. */
1565 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
1566 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
1568 msg_flags |= MSG_NOSIGNAL;
1569 drbd_update_congested(peer_device->connection);
1574 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
1576 if (sent == -EAGAIN) {
1577 if (we_should_drop_the_connection(peer_device->connection, socket))
1581 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
1582 __func__, (int)size, len, sent);
1589 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
1591 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
1595 peer_device->device->send_cnt += size >> 9;
1600 static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1602 struct bio_vec bvec;
1603 struct bvec_iter iter;
1605 /* hint all but last page with MSG_MORE */
1606 bio_for_each_segment(bvec, bio, iter) {
1609 err = _drbd_no_send_page(peer_device, bvec.bv_page,
1610 bvec.bv_offset, bvec.bv_len,
1611 bio_iter_last(bvec, iter)
1615 /* REQ_OP_WRITE_SAME has only one segment */
1616 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1622 static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
1624 struct bio_vec bvec;
1625 struct bvec_iter iter;
1627 /* hint all but last page with MSG_MORE */
1628 bio_for_each_segment(bvec, bio, iter) {
1631 err = _drbd_send_page(peer_device, bvec.bv_page,
1632 bvec.bv_offset, bvec.bv_len,
1633 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
1636 /* REQ_OP_WRITE_SAME has only one segment */
1637 if (bio_op(bio) == REQ_OP_WRITE_SAME)
1643 static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
1644 struct drbd_peer_request *peer_req)
1646 struct page *page = peer_req->pages;
1647 unsigned len = peer_req->i.size;
1650 /* hint all but last page with MSG_MORE */
1651 page_chain_for_each(page) {
1652 unsigned l = min_t(unsigned, len, PAGE_SIZE);
1654 err = _drbd_send_page(peer_device, page, 0, l,
1655 page_chain_next(page) ? MSG_MORE : 0);
1663 static u32 bio_flags_to_wire(struct drbd_connection *connection,
1666 if (connection->agreed_pro_version >= 95)
1667 return (bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0) |
1668 (bio->bi_opf & REQ_FUA ? DP_FUA : 0) |
1669 (bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
1670 (bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
1671 (bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0);
1673 return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
1676 /* Used to send write or TRIM aka REQ_DISCARD requests
1677 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
1679 int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
1681 struct drbd_device *device = peer_device->device;
1682 struct drbd_socket *sock;
1684 struct p_wsame *wsame = NULL;
1686 unsigned int dp_flags = 0;
1690 sock = &peer_device->connection->data;
1691 p = drbd_prepare_command(peer_device, sock);
1692 digest_size = peer_device->connection->integrity_tfm ?
1693 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1697 p->sector = cpu_to_be64(req->i.sector);
1698 p->block_id = (unsigned long)req;
1699 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
1700 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio);
1701 if (device->state.conn >= C_SYNC_SOURCE &&
1702 device->state.conn <= C_PAUSED_SYNC_T)
1703 dp_flags |= DP_MAY_SET_IN_SYNC;
1704 if (peer_device->connection->agreed_pro_version >= 100) {
1705 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1706 dp_flags |= DP_SEND_RECEIVE_ACK;
1707 /* During resync, request an explicit write ack,
1708 * even in protocol != C */
1709 if (req->rq_state & RQ_EXP_WRITE_ACK
1710 || (dp_flags & DP_MAY_SET_IN_SYNC))
1711 dp_flags |= DP_SEND_WRITE_ACK;
1713 p->dp_flags = cpu_to_be32(dp_flags);
1715 if (dp_flags & DP_DISCARD) {
1716 struct p_trim *t = (struct p_trim*)p;
1717 t->size = cpu_to_be32(req->i.size);
1718 err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1721 if (dp_flags & DP_WSAME) {
1722 /* this will only work if DRBD_FF_WSAME is set AND the
1723 * handshake agreed that all nodes and backend devices are
1724 * WRITE_SAME capable and agree on logical_block_size */
1725 wsame = (struct p_wsame*)p;
1726 digest_out = wsame + 1;
1727 wsame->size = cpu_to_be32(req->i.size);
1731 /* our digest is still only over the payload.
1732 * TRIM does not carry any payload. */
1734 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest_out);
1737 __send_command(peer_device->connection, device->vnr, sock, P_WSAME,
1738 sizeof(*wsame) + digest_size, NULL,
1739 bio_iovec(req->master_bio).bv_len);
1742 __send_command(peer_device->connection, device->vnr, sock, P_DATA,
1743 sizeof(*p) + digest_size, NULL, req->i.size);
1745 /* For protocol A, we have to memcpy the payload into
1746 * socket buffers, as we may complete right away
1747 * as soon as we handed it over to tcp, at which point the data
1748 * pages may become invalid.
1750 * For data-integrity enabled, we copy it as well, so we can be
1751 * sure that even if the bio pages may still be modified, it
1752 * won't change the data on the wire, thus if the digest checks
1753 * out ok after sending on this side, but does not fit on the
1754 * receiving side, we sure have detected corruption elsewhere.
1756 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
1757 err = _drbd_send_bio(peer_device, req->master_bio);
1759 err = _drbd_send_zc_bio(peer_device, req->master_bio);
1761 /* double check digest, sometimes buffers have been modified in flight. */
1762 if (digest_size > 0 && digest_size <= 64) {
1763 /* 64 byte, 512 bit, is the largest digest size
1764 * currently supported in kernel crypto. */
1765 unsigned char digest[64];
1766 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
1767 if (memcmp(p + 1, digest, digest_size)) {
1769 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
1770 (unsigned long long)req->i.sector, req->i.size);
1772 } /* else if (digest_size > 64) {
1773 ... Be noisy about digest too large ...
1777 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1782 /* answer packet, used to send data back for read requests:
1783 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1784 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1786 int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
1787 struct drbd_peer_request *peer_req)
1789 struct drbd_device *device = peer_device->device;
1790 struct drbd_socket *sock;
1795 sock = &peer_device->connection->data;
1796 p = drbd_prepare_command(peer_device, sock);
1798 digest_size = peer_device->connection->integrity_tfm ?
1799 crypto_ahash_digestsize(peer_device->connection->integrity_tfm) : 0;
1803 p->sector = cpu_to_be64(peer_req->i.sector);
1804 p->block_id = peer_req->block_id;
1805 p->seq_num = 0; /* unused */
1808 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1809 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
1811 err = _drbd_send_zc_ee(peer_device, peer_req);
1812 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
1817 int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
1819 struct drbd_socket *sock;
1820 struct p_block_desc *p;
1822 sock = &peer_device->connection->data;
1823 p = drbd_prepare_command(peer_device, sock);
1826 p->sector = cpu_to_be64(req->i.sector);
1827 p->blksize = cpu_to_be32(req->i.size);
1828 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
1832 drbd_send distinguishes two cases:
1834 Packets sent via the data socket "sock"
1835 and packets sent via the meta data socket "msock"
1838 -----------------+-------------------------+------------------------------
1839 timeout conf.timeout / 2 conf.timeout / 2
1840 timeout action send a ping via msock Abort communication
1841 and close all sockets
1845 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1847 int drbd_send(struct drbd_connection *connection, struct socket *sock,
1848 void *buf, size_t size, unsigned msg_flags)
1857 /* THINK if (signal_pending) return ... ? */
1862 msg.msg_name = NULL;
1863 msg.msg_namelen = 0;
1864 msg.msg_control = NULL;
1865 msg.msg_controllen = 0;
1866 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1868 if (sock == connection->data.socket) {
1870 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
1872 drbd_update_congested(connection);
1875 rv = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
1876 if (rv == -EAGAIN) {
1877 if (we_should_drop_the_connection(connection, sock))
1883 flush_signals(current);
1891 } while (sent < size);
1893 if (sock == connection->data.socket)
1894 clear_bit(NET_CONGESTED, &connection->flags);
1897 if (rv != -EAGAIN) {
1898 drbd_err(connection, "%s_sendmsg returned %d\n",
1899 sock == connection->meta.socket ? "msock" : "sock",
1901 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
1903 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
1910 * drbd_send_all - Send an entire buffer
1912 * Returns 0 upon success and a negative error value otherwise.
1914 int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
1915 size_t size, unsigned msg_flags)
1919 err = drbd_send(connection, sock, buffer, size, msg_flags);
1927 static int drbd_open(struct block_device *bdev, fmode_t mode)
1929 struct drbd_device *device = bdev->bd_disk->private_data;
1930 unsigned long flags;
1933 mutex_lock(&drbd_main_mutex);
1934 spin_lock_irqsave(&device->resource->req_lock, flags);
1935 /* to have a stable device->state.role
1936 * and no race with updating open_cnt */
1938 if (device->state.role != R_PRIMARY) {
1939 if (mode & FMODE_WRITE)
1941 else if (!allow_oos)
1947 spin_unlock_irqrestore(&device->resource->req_lock, flags);
1948 mutex_unlock(&drbd_main_mutex);
1953 static void drbd_release(struct gendisk *gd, fmode_t mode)
1955 struct drbd_device *device = gd->private_data;
1956 mutex_lock(&drbd_main_mutex);
1958 mutex_unlock(&drbd_main_mutex);
1961 static void drbd_set_defaults(struct drbd_device *device)
1963 /* Beware! The actual layout differs
1964 * between big endian and little endian */
1965 device->state = (union drbd_dev_state) {
1966 { .role = R_SECONDARY,
1968 .conn = C_STANDALONE,
1974 void drbd_init_set_defaults(struct drbd_device *device)
1976 /* the memset(,0,) did most of this.
1977 * note: only assignments, no allocation in here */
1979 drbd_set_defaults(device);
1981 atomic_set(&device->ap_bio_cnt, 0);
1982 atomic_set(&device->ap_actlog_cnt, 0);
1983 atomic_set(&device->ap_pending_cnt, 0);
1984 atomic_set(&device->rs_pending_cnt, 0);
1985 atomic_set(&device->unacked_cnt, 0);
1986 atomic_set(&device->local_cnt, 0);
1987 atomic_set(&device->pp_in_use_by_net, 0);
1988 atomic_set(&device->rs_sect_in, 0);
1989 atomic_set(&device->rs_sect_ev, 0);
1990 atomic_set(&device->ap_in_flight, 0);
1991 atomic_set(&device->md_io.in_use, 0);
1993 mutex_init(&device->own_state_mutex);
1994 device->state_mutex = &device->own_state_mutex;
1996 spin_lock_init(&device->al_lock);
1997 spin_lock_init(&device->peer_seq_lock);
1999 INIT_LIST_HEAD(&device->active_ee);
2000 INIT_LIST_HEAD(&device->sync_ee);
2001 INIT_LIST_HEAD(&device->done_ee);
2002 INIT_LIST_HEAD(&device->read_ee);
2003 INIT_LIST_HEAD(&device->net_ee);
2004 INIT_LIST_HEAD(&device->resync_reads);
2005 INIT_LIST_HEAD(&device->resync_work.list);
2006 INIT_LIST_HEAD(&device->unplug_work.list);
2007 INIT_LIST_HEAD(&device->bm_io_work.w.list);
2008 INIT_LIST_HEAD(&device->pending_master_completion[0]);
2009 INIT_LIST_HEAD(&device->pending_master_completion[1]);
2010 INIT_LIST_HEAD(&device->pending_completion[0]);
2011 INIT_LIST_HEAD(&device->pending_completion[1]);
2013 device->resync_work.cb = w_resync_timer;
2014 device->unplug_work.cb = w_send_write_hint;
2015 device->bm_io_work.w.cb = w_bitmap_io;
2017 init_timer(&device->resync_timer);
2018 init_timer(&device->md_sync_timer);
2019 init_timer(&device->start_resync_timer);
2020 init_timer(&device->request_timer);
2021 device->resync_timer.function = resync_timer_fn;
2022 device->resync_timer.data = (unsigned long) device;
2023 device->md_sync_timer.function = md_sync_timer_fn;
2024 device->md_sync_timer.data = (unsigned long) device;
2025 device->start_resync_timer.function = start_resync_timer_fn;
2026 device->start_resync_timer.data = (unsigned long) device;
2027 device->request_timer.function = request_timer_fn;
2028 device->request_timer.data = (unsigned long) device;
2030 init_waitqueue_head(&device->misc_wait);
2031 init_waitqueue_head(&device->state_wait);
2032 init_waitqueue_head(&device->ee_wait);
2033 init_waitqueue_head(&device->al_wait);
2034 init_waitqueue_head(&device->seq_wait);
2036 device->resync_wenr = LC_FREE;
2037 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2038 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
2041 void drbd_device_cleanup(struct drbd_device *device)
2044 if (first_peer_device(device)->connection->receiver.t_state != NONE)
2045 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
2046 first_peer_device(device)->connection->receiver.t_state);
2048 device->al_writ_cnt =
2049 device->bm_writ_cnt =
2057 device->rs_failed = 0;
2058 device->rs_last_events = 0;
2059 device->rs_last_sect_ev = 0;
2060 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
2061 device->rs_mark_left[i] = 0;
2062 device->rs_mark_time[i] = 0;
2064 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
2066 drbd_set_my_capacity(device, 0);
2067 if (device->bitmap) {
2068 /* maybe never allocated. */
2069 drbd_bm_resize(device, 0, 1);
2070 drbd_bm_cleanup(device);
2073 drbd_backing_dev_free(device, device->ldev);
2074 device->ldev = NULL;
2076 clear_bit(AL_SUSPENDED, &device->flags);
2078 D_ASSERT(device, list_empty(&device->active_ee));
2079 D_ASSERT(device, list_empty(&device->sync_ee));
2080 D_ASSERT(device, list_empty(&device->done_ee));
2081 D_ASSERT(device, list_empty(&device->read_ee));
2082 D_ASSERT(device, list_empty(&device->net_ee));
2083 D_ASSERT(device, list_empty(&device->resync_reads));
2084 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2085 D_ASSERT(device, list_empty(&device->resync_work.list));
2086 D_ASSERT(device, list_empty(&device->unplug_work.list));
2088 drbd_set_defaults(device);
2092 static void drbd_destroy_mempools(void)
2096 while (drbd_pp_pool) {
2097 page = drbd_pp_pool;
2098 drbd_pp_pool = (struct page *)page_private(page);
2103 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
2105 if (drbd_md_io_bio_set)
2106 bioset_free(drbd_md_io_bio_set);
2107 if (drbd_md_io_page_pool)
2108 mempool_destroy(drbd_md_io_page_pool);
2109 if (drbd_ee_mempool)
2110 mempool_destroy(drbd_ee_mempool);
2111 if (drbd_request_mempool)
2112 mempool_destroy(drbd_request_mempool);
2114 kmem_cache_destroy(drbd_ee_cache);
2115 if (drbd_request_cache)
2116 kmem_cache_destroy(drbd_request_cache);
2117 if (drbd_bm_ext_cache)
2118 kmem_cache_destroy(drbd_bm_ext_cache);
2119 if (drbd_al_ext_cache)
2120 kmem_cache_destroy(drbd_al_ext_cache);
2122 drbd_md_io_bio_set = NULL;
2123 drbd_md_io_page_pool = NULL;
2124 drbd_ee_mempool = NULL;
2125 drbd_request_mempool = NULL;
2126 drbd_ee_cache = NULL;
2127 drbd_request_cache = NULL;
2128 drbd_bm_ext_cache = NULL;
2129 drbd_al_ext_cache = NULL;
2134 static int drbd_create_mempools(void)
2137 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
2140 /* prepare our caches and mempools */
2141 drbd_request_mempool = NULL;
2142 drbd_ee_cache = NULL;
2143 drbd_request_cache = NULL;
2144 drbd_bm_ext_cache = NULL;
2145 drbd_al_ext_cache = NULL;
2146 drbd_pp_pool = NULL;
2147 drbd_md_io_page_pool = NULL;
2148 drbd_md_io_bio_set = NULL;
2151 drbd_request_cache = kmem_cache_create(
2152 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2153 if (drbd_request_cache == NULL)
2156 drbd_ee_cache = kmem_cache_create(
2157 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
2158 if (drbd_ee_cache == NULL)
2161 drbd_bm_ext_cache = kmem_cache_create(
2162 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2163 if (drbd_bm_ext_cache == NULL)
2166 drbd_al_ext_cache = kmem_cache_create(
2167 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2168 if (drbd_al_ext_cache == NULL)
2172 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2173 if (drbd_md_io_bio_set == NULL)
2176 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2177 if (drbd_md_io_page_pool == NULL)
2180 drbd_request_mempool = mempool_create_slab_pool(number,
2181 drbd_request_cache);
2182 if (drbd_request_mempool == NULL)
2185 drbd_ee_mempool = mempool_create_slab_pool(number, drbd_ee_cache);
2186 if (drbd_ee_mempool == NULL)
2189 /* drbd's page pool */
2190 spin_lock_init(&drbd_pp_lock);
2192 for (i = 0; i < number; i++) {
2193 page = alloc_page(GFP_HIGHUSER);
2196 set_page_private(page, (unsigned long)drbd_pp_pool);
2197 drbd_pp_pool = page;
2199 drbd_pp_vacant = number;
2204 drbd_destroy_mempools(); /* in case we allocated some */
2208 static void drbd_release_all_peer_reqs(struct drbd_device *device)
2212 rr = drbd_free_peer_reqs(device, &device->active_ee);
2214 drbd_err(device, "%d EEs in active list found!\n", rr);
2216 rr = drbd_free_peer_reqs(device, &device->sync_ee);
2218 drbd_err(device, "%d EEs in sync list found!\n", rr);
2220 rr = drbd_free_peer_reqs(device, &device->read_ee);
2222 drbd_err(device, "%d EEs in read list found!\n", rr);
2224 rr = drbd_free_peer_reqs(device, &device->done_ee);
2226 drbd_err(device, "%d EEs in done list found!\n", rr);
2228 rr = drbd_free_peer_reqs(device, &device->net_ee);
2230 drbd_err(device, "%d EEs in net list found!\n", rr);
2233 /* caution. no locking. */
2234 void drbd_destroy_device(struct kref *kref)
2236 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
2237 struct drbd_resource *resource = device->resource;
2238 struct drbd_peer_device *peer_device, *tmp_peer_device;
2240 del_timer_sync(&device->request_timer);
2242 /* paranoia asserts */
2243 D_ASSERT(device, device->open_cnt == 0);
2244 /* end paranoia asserts */
2246 /* cleanup stuff that may have been allocated during
2247 * device (re-)configuration or state changes */
2249 if (device->this_bdev)
2250 bdput(device->this_bdev);
2252 drbd_backing_dev_free(device, device->ldev);
2253 device->ldev = NULL;
2255 drbd_release_all_peer_reqs(device);
2257 lc_destroy(device->act_log);
2258 lc_destroy(device->resync);
2260 kfree(device->p_uuid);
2261 /* device->p_uuid = NULL; */
2263 if (device->bitmap) /* should no longer be there. */
2264 drbd_bm_cleanup(device);
2265 __free_page(device->md_io.page);
2266 put_disk(device->vdisk);
2267 blk_cleanup_queue(device->rq_queue);
2268 kfree(device->rs_plan_s);
2270 /* not for_each_connection(connection, resource):
2271 * those may have been cleaned up and disassociated already.
2273 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2274 kref_put(&peer_device->connection->kref, drbd_destroy_connection);
2277 memset(device, 0xfd, sizeof(*device));
2279 kref_put(&resource->kref, drbd_destroy_resource);
2282 /* One global retry thread, if we need to push back some bio and have it
2283 * reinserted through our make request function.
2285 static struct retry_worker {
2286 struct workqueue_struct *wq;
2287 struct work_struct worker;
2290 struct list_head writes;
2293 static void do_retry(struct work_struct *ws)
2295 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2297 struct drbd_request *req, *tmp;
2299 spin_lock_irq(&retry->lock);
2300 list_splice_init(&retry->writes, &writes);
2301 spin_unlock_irq(&retry->lock);
2303 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
2304 struct drbd_device *device = req->device;
2305 struct bio *bio = req->master_bio;
2306 unsigned long start_jif = req->start_jif;
2310 expect(atomic_read(&req->completion_ref) == 0) &&
2311 expect(req->rq_state & RQ_POSTPONED) &&
2312 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2313 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2316 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
2317 req, atomic_read(&req->completion_ref),
2320 /* We still need to put one kref associated with the
2321 * "completion_ref" going zero in the code path that queued it
2322 * here. The request object may still be referenced by a
2323 * frozen local req->private_bio, in case we force-detached.
2325 kref_put(&req->kref, drbd_req_destroy);
2327 /* A single suspended or otherwise blocking device may stall
2328 * all others as well. Fortunately, this code path is to
2329 * recover from a situation that "should not happen":
2330 * concurrent writes in multi-primary setup.
2331 * In a "normal" lifecycle, this workqueue is supposed to be
2332 * destroyed without ever doing anything.
2333 * If it turns out to be an issue anyways, we can do per
2334 * resource (replication group) or per device (minor) retry
2335 * workqueues instead.
2338 /* We are not just doing generic_make_request(),
2339 * as we want to keep the start_time information. */
2341 __drbd_make_request(device, bio, start_jif);
2345 /* called via drbd_req_put_completion_ref(),
2346 * holds resource->req_lock */
2347 void drbd_restart_request(struct drbd_request *req)
2349 unsigned long flags;
2350 spin_lock_irqsave(&retry.lock, flags);
2351 list_move_tail(&req->tl_requests, &retry.writes);
2352 spin_unlock_irqrestore(&retry.lock, flags);
2354 /* Drop the extra reference that would otherwise
2355 * have been dropped by complete_master_bio.
2356 * do_retry() needs to grab a new one. */
2357 dec_ap_bio(req->device);
2359 queue_work(retry.wq, &retry.worker);
2362 void drbd_destroy_resource(struct kref *kref)
2364 struct drbd_resource *resource =
2365 container_of(kref, struct drbd_resource, kref);
2367 idr_destroy(&resource->devices);
2368 free_cpumask_var(resource->cpu_mask);
2369 kfree(resource->name);
2370 memset(resource, 0xf2, sizeof(*resource));
2374 void drbd_free_resource(struct drbd_resource *resource)
2376 struct drbd_connection *connection, *tmp;
2378 for_each_connection_safe(connection, tmp, resource) {
2379 list_del(&connection->connections);
2380 drbd_debugfs_connection_cleanup(connection);
2381 kref_put(&connection->kref, drbd_destroy_connection);
2383 drbd_debugfs_resource_cleanup(resource);
2384 kref_put(&resource->kref, drbd_destroy_resource);
2387 static void drbd_cleanup(void)
2390 struct drbd_device *device;
2391 struct drbd_resource *resource, *tmp;
2393 /* first remove proc,
2394 * drbdsetup uses it's presence to detect
2395 * whether DRBD is loaded.
2396 * If we would get stuck in proc removal,
2397 * but have netlink already deregistered,
2398 * some drbdsetup commands may wait forever
2402 remove_proc_entry("drbd", NULL);
2405 destroy_workqueue(retry.wq);
2407 drbd_genl_unregister();
2408 drbd_debugfs_cleanup();
2410 idr_for_each_entry(&drbd_devices, device, i)
2411 drbd_delete_device(device);
2413 /* not _rcu since, no other updater anymore. Genl already unregistered */
2414 for_each_resource_safe(resource, tmp, &drbd_resources) {
2415 list_del(&resource->resources);
2416 drbd_free_resource(resource);
2419 drbd_destroy_mempools();
2420 unregister_blkdev(DRBD_MAJOR, "drbd");
2422 idr_destroy(&drbd_devices);
2424 pr_info("module cleanup done.\n");
2428 * drbd_congested() - Callback for the flusher thread
2429 * @congested_data: User data
2430 * @bdi_bits: Bits the BDI flusher thread is currently interested in
2432 * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested.
2434 static int drbd_congested(void *congested_data, int bdi_bits)
2436 struct drbd_device *device = congested_data;
2437 struct request_queue *q;
2441 if (!may_inc_ap_bio(device)) {
2442 /* DRBD has frozen IO */
2448 if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
2449 r |= (1 << WB_async_congested);
2450 /* Without good local data, we would need to read from remote,
2451 * and that would need the worker thread as well, which is
2452 * currently blocked waiting for that usermode helper to
2455 if (!get_ldev_if_state(device, D_UP_TO_DATE))
2456 r |= (1 << WB_sync_congested);
2464 if (get_ldev(device)) {
2465 q = bdev_get_queue(device->ldev->backing_bdev);
2466 r = bdi_congested(&q->backing_dev_info, bdi_bits);
2472 if (bdi_bits & (1 << WB_async_congested) &&
2473 test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
2474 r |= (1 << WB_async_congested);
2475 reason = reason == 'b' ? 'a' : 'n';
2479 device->congestion_reason = reason;
2483 static void drbd_init_workqueue(struct drbd_work_queue* wq)
2485 spin_lock_init(&wq->q_lock);
2486 INIT_LIST_HEAD(&wq->q);
2487 init_waitqueue_head(&wq->q_wait);
2490 struct completion_work {
2492 struct completion done;
2495 static int w_complete(struct drbd_work *w, int cancel)
2497 struct completion_work *completion_work =
2498 container_of(w, struct completion_work, w);
2500 complete(&completion_work->done);
2504 void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2506 struct completion_work completion_work;
2508 completion_work.w.cb = w_complete;
2509 init_completion(&completion_work.done);
2510 drbd_queue_work(work_queue, &completion_work.w);
2511 wait_for_completion(&completion_work.done);
2514 struct drbd_resource *drbd_find_resource(const char *name)
2516 struct drbd_resource *resource;
2518 if (!name || !name[0])
2522 for_each_resource_rcu(resource, &drbd_resources) {
2523 if (!strcmp(resource->name, name)) {
2524 kref_get(&resource->kref);
2534 struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
2535 void *peer_addr, int peer_addr_len)
2537 struct drbd_resource *resource;
2538 struct drbd_connection *connection;
2541 for_each_resource_rcu(resource, &drbd_resources) {
2542 for_each_connection_rcu(connection, resource) {
2543 if (connection->my_addr_len == my_addr_len &&
2544 connection->peer_addr_len == peer_addr_len &&
2545 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2546 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2547 kref_get(&connection->kref);
2558 static int drbd_alloc_socket(struct drbd_socket *socket)
2560 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2563 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2569 static void drbd_free_socket(struct drbd_socket *socket)
2571 free_page((unsigned long) socket->sbuf);
2572 free_page((unsigned long) socket->rbuf);
2575 void conn_free_crypto(struct drbd_connection *connection)
2577 drbd_free_sock(connection);
2579 crypto_free_ahash(connection->csums_tfm);
2580 crypto_free_ahash(connection->verify_tfm);
2581 crypto_free_shash(connection->cram_hmac_tfm);
2582 crypto_free_ahash(connection->integrity_tfm);
2583 crypto_free_ahash(connection->peer_integrity_tfm);
2584 kfree(connection->int_dig_in);
2585 kfree(connection->int_dig_vv);
2587 connection->csums_tfm = NULL;
2588 connection->verify_tfm = NULL;
2589 connection->cram_hmac_tfm = NULL;
2590 connection->integrity_tfm = NULL;
2591 connection->peer_integrity_tfm = NULL;
2592 connection->int_dig_in = NULL;
2593 connection->int_dig_vv = NULL;
2596 int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
2598 struct drbd_connection *connection;
2599 cpumask_var_t new_cpu_mask;
2602 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2605 /* silently ignore cpu mask on UP kernel */
2606 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
2607 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
2608 cpumask_bits(new_cpu_mask), nr_cpu_ids);
2609 if (err == -EOVERFLOW) {
2610 /* So what. mask it out. */
2611 cpumask_var_t tmp_cpu_mask;
2612 if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) {
2613 cpumask_setall(tmp_cpu_mask);
2614 cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask);
2615 drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n",
2617 strlen(res_opts->cpu_mask) > 12 ? "..." : "",
2619 free_cpumask_var(tmp_cpu_mask);
2624 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
2625 /* retcode = ERR_CPU_MASK_PARSE; */
2629 resource->res_opts = *res_opts;
2630 if (cpumask_empty(new_cpu_mask))
2631 drbd_calc_cpu_mask(&new_cpu_mask);
2632 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2633 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2634 for_each_connection_rcu(connection, resource) {
2635 connection->receiver.reset_cpu_mask = 1;
2636 connection->ack_receiver.reset_cpu_mask = 1;
2637 connection->worker.reset_cpu_mask = 1;
2643 free_cpumask_var(new_cpu_mask);
2648 struct drbd_resource *drbd_create_resource(const char *name)
2650 struct drbd_resource *resource;
2652 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
2655 resource->name = kstrdup(name, GFP_KERNEL);
2656 if (!resource->name)
2657 goto fail_free_resource;
2658 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2659 goto fail_free_name;
2660 kref_init(&resource->kref);
2661 idr_init(&resource->devices);
2662 INIT_LIST_HEAD(&resource->connections);
2663 resource->write_ordering = WO_BDEV_FLUSH;
2664 list_add_tail_rcu(&resource->resources, &drbd_resources);
2665 mutex_init(&resource->conf_update);
2666 mutex_init(&resource->adm_mutex);
2667 spin_lock_init(&resource->req_lock);
2668 drbd_debugfs_resource_add(resource);
2672 kfree(resource->name);
2679 /* caller must be under adm_mutex */
2680 struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
2682 struct drbd_resource *resource;
2683 struct drbd_connection *connection;
2685 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2689 if (drbd_alloc_socket(&connection->data))
2691 if (drbd_alloc_socket(&connection->meta))
2694 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2695 if (!connection->current_epoch)
2698 INIT_LIST_HEAD(&connection->transfer_log);
2700 INIT_LIST_HEAD(&connection->current_epoch->list);
2701 connection->epochs = 1;
2702 spin_lock_init(&connection->epoch_lock);
2704 connection->send.seen_any_write_yet = false;
2705 connection->send.current_epoch_nr = 0;
2706 connection->send.current_epoch_writes = 0;
2708 resource = drbd_create_resource(name);
2712 connection->cstate = C_STANDALONE;
2713 mutex_init(&connection->cstate_mutex);
2714 init_waitqueue_head(&connection->ping_wait);
2715 idr_init(&connection->peer_devices);
2717 drbd_init_workqueue(&connection->sender_work);
2718 mutex_init(&connection->data.mutex);
2719 mutex_init(&connection->meta.mutex);
2721 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2722 connection->receiver.connection = connection;
2723 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2724 connection->worker.connection = connection;
2725 drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv");
2726 connection->ack_receiver.connection = connection;
2728 kref_init(&connection->kref);
2730 connection->resource = resource;
2732 if (set_resource_options(resource, res_opts))
2735 kref_get(&resource->kref);
2736 list_add_tail_rcu(&connection->connections, &resource->connections);
2737 drbd_debugfs_connection_add(connection);
2741 list_del(&resource->resources);
2742 drbd_free_resource(resource);
2744 kfree(connection->current_epoch);
2745 drbd_free_socket(&connection->meta);
2746 drbd_free_socket(&connection->data);
2751 void drbd_destroy_connection(struct kref *kref)
2753 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
2754 struct drbd_resource *resource = connection->resource;
2756 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
2757 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
2758 kfree(connection->current_epoch);
2760 idr_destroy(&connection->peer_devices);
2762 drbd_free_socket(&connection->meta);
2763 drbd_free_socket(&connection->data);
2764 kfree(connection->int_dig_in);
2765 kfree(connection->int_dig_vv);
2766 memset(connection, 0xfc, sizeof(*connection));
2768 kref_put(&resource->kref, drbd_destroy_resource);
2771 static int init_submitter(struct drbd_device *device)
2773 /* opencoded create_singlethread_workqueue(),
2774 * to be able to say "drbd%d", ..., minor */
2776 alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor);
2777 if (!device->submit.wq)
2780 INIT_WORK(&device->submit.worker, do_submit);
2781 INIT_LIST_HEAD(&device->submit.writes);
2785 enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
2787 struct drbd_resource *resource = adm_ctx->resource;
2788 struct drbd_connection *connection;
2789 struct drbd_device *device;
2790 struct drbd_peer_device *peer_device, *tmp_peer_device;
2791 struct gendisk *disk;
2792 struct request_queue *q;
2794 int vnr = adm_ctx->volume;
2795 enum drbd_ret_code err = ERR_NOMEM;
2797 device = minor_to_device(minor);
2799 return ERR_MINOR_OR_VOLUME_EXISTS;
2801 /* GFP_KERNEL, we are outside of all write-out paths */
2802 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2805 kref_init(&device->kref);
2807 kref_get(&resource->kref);
2808 device->resource = resource;
2809 device->minor = minor;
2812 drbd_init_set_defaults(device);
2814 q = blk_alloc_queue(GFP_KERNEL);
2817 device->rq_queue = q;
2818 q->queuedata = device;
2820 disk = alloc_disk(1);
2823 device->vdisk = disk;
2825 set_disk_ro(disk, true);
2828 disk->major = DRBD_MAJOR;
2829 disk->first_minor = minor;
2830 disk->fops = &drbd_ops;
2831 sprintf(disk->disk_name, "drbd%d", minor);
2832 disk->private_data = device;
2834 device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
2835 /* we have no partitions. we contain only ourselves. */
2836 device->this_bdev->bd_contains = device->this_bdev;
2838 q->backing_dev_info.congested_fn = drbd_congested;
2839 q->backing_dev_info.congested_data = device;
2841 blk_queue_make_request(q, drbd_make_request);
2842 blk_queue_write_cache(q, true, true);
2843 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2844 This triggers a max_bio_size message upon first attach or connect */
2845 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
2846 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2847 q->queue_lock = &resource->req_lock;
2849 device->md_io.page = alloc_page(GFP_KERNEL);
2850 if (!device->md_io.page)
2851 goto out_no_io_page;
2853 if (drbd_bm_init(device))
2855 device->read_requests = RB_ROOT;
2856 device->write_requests = RB_ROOT;
2858 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2861 err = ERR_MINOR_OR_VOLUME_EXISTS;
2862 goto out_no_minor_idr;
2864 kref_get(&device->kref);
2866 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2869 err = ERR_MINOR_OR_VOLUME_EXISTS;
2870 goto out_idr_remove_minor;
2872 kref_get(&device->kref);
2874 INIT_LIST_HEAD(&device->peer_devices);
2875 INIT_LIST_HEAD(&device->pending_bitmap_io);
2876 for_each_connection(connection, resource) {
2877 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2879 goto out_idr_remove_from_resource;
2880 peer_device->connection = connection;
2881 peer_device->device = device;
2883 list_add(&peer_device->peer_devices, &device->peer_devices);
2884 kref_get(&device->kref);
2886 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2889 err = ERR_INVALID_REQUEST;
2890 goto out_idr_remove_from_resource;
2892 kref_get(&connection->kref);
2893 INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf);
2896 if (init_submitter(device)) {
2898 goto out_idr_remove_vol;
2903 /* inherit the connection state */
2904 device->state.conn = first_connection(resource)->cstate;
2905 if (device->state.conn == C_WF_REPORT_PARAMS) {
2906 for_each_peer_device(peer_device, device)
2907 drbd_connected(peer_device);
2909 /* move to create_peer_device() */
2910 for_each_peer_device(peer_device, device)
2911 drbd_debugfs_peer_device_add(peer_device);
2912 drbd_debugfs_device_add(device);
2916 idr_remove(&connection->peer_devices, vnr);
2917 out_idr_remove_from_resource:
2918 for_each_connection(connection, resource) {
2919 peer_device = idr_find(&connection->peer_devices, vnr);
2921 idr_remove(&connection->peer_devices, vnr);
2922 kref_put(&connection->kref, drbd_destroy_connection);
2925 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2926 list_del(&peer_device->peer_devices);
2929 idr_remove(&resource->devices, vnr);
2930 out_idr_remove_minor:
2931 idr_remove(&drbd_devices, minor);
2934 drbd_bm_cleanup(device);
2936 __free_page(device->md_io.page);
2940 blk_cleanup_queue(q);
2942 kref_put(&resource->kref, drbd_destroy_resource);
2947 void drbd_delete_device(struct drbd_device *device)
2949 struct drbd_resource *resource = device->resource;
2950 struct drbd_connection *connection;
2951 struct drbd_peer_device *peer_device;
2954 /* move to free_peer_device() */
2955 for_each_peer_device(peer_device, device)
2956 drbd_debugfs_peer_device_cleanup(peer_device);
2957 drbd_debugfs_device_cleanup(device);
2958 for_each_connection(connection, resource) {
2959 idr_remove(&connection->peer_devices, device->vnr);
2962 idr_remove(&resource->devices, device->vnr);
2963 idr_remove(&drbd_devices, device_to_minor(device));
2964 del_gendisk(device->vdisk);
2966 kref_sub(&device->kref, refs, drbd_destroy_device);
2969 static int __init drbd_init(void)
2973 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
2974 pr_err("invalid minor_count (%d)\n", minor_count);
2978 minor_count = DRBD_MINOR_COUNT_DEF;
2982 err = register_blkdev(DRBD_MAJOR, "drbd");
2984 pr_err("unable to register block device major %d\n",
2990 * allocate all necessary structs
2992 init_waitqueue_head(&drbd_pp_wait);
2994 drbd_proc = NULL; /* play safe for drbd_cleanup */
2995 idr_init(&drbd_devices);
2997 mutex_init(&resources_mutex);
2998 INIT_LIST_HEAD(&drbd_resources);
3000 err = drbd_genl_register();
3002 pr_err("unable to register generic netlink family\n");
3006 err = drbd_create_mempools();
3011 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
3013 pr_err("unable to register proc file\n");
3017 retry.wq = create_singlethread_workqueue("drbd-reissue");
3019 pr_err("unable to create retry workqueue\n");
3022 INIT_WORK(&retry.worker, do_retry);
3023 spin_lock_init(&retry.lock);
3024 INIT_LIST_HEAD(&retry.writes);
3026 if (drbd_debugfs_init())
3027 pr_notice("failed to initialize debugfs -- will not be available\n");
3029 pr_info("initialized. "
3030 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
3031 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
3032 pr_info("%s\n", drbd_buildtag());
3033 pr_info("registered as block device major %d\n", DRBD_MAJOR);
3034 return 0; /* Success! */
3039 pr_err("ran out of memory\n");
3041 pr_err("initialization failure\n");
3045 static void drbd_free_one_sock(struct drbd_socket *ds)
3048 mutex_lock(&ds->mutex);
3051 mutex_unlock(&ds->mutex);
3053 /* so debugfs does not need to mutex_lock() */
3055 kernel_sock_shutdown(s, SHUT_RDWR);
3060 void drbd_free_sock(struct drbd_connection *connection)
3062 if (connection->data.socket)
3063 drbd_free_one_sock(&connection->data);
3064 if (connection->meta.socket)
3065 drbd_free_one_sock(&connection->meta);
3068 /* meta data management */
3070 void conn_md_sync(struct drbd_connection *connection)
3072 struct drbd_peer_device *peer_device;
3076 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
3077 struct drbd_device *device = peer_device->device;
3079 kref_get(&device->kref);
3081 drbd_md_sync(device);
3082 kref_put(&device->kref, drbd_destroy_device);
3088 /* aligned 4kByte */
3089 struct meta_data_on_disk {
3090 u64 la_size_sect; /* last agreed size. */
3091 u64 uuid[UI_SIZE]; /* UUIDs. */
3094 u32 flags; /* MDF */
3097 u32 al_offset; /* offset to this block */
3098 u32 al_nr_extents; /* important for restoring the AL (userspace) */
3099 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
3100 u32 bm_offset; /* offset to the bitmap, from here */
3101 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
3102 u32 la_peer_max_bio_size; /* last peer max_bio_size */
3104 /* see al_tr_number_to_on_disk_sector() */
3106 u32 al_stripe_size_4k;
3108 u8 reserved_u8[4096 - (7*8 + 10*4)];
3113 void drbd_md_write(struct drbd_device *device, void *b)
3115 struct meta_data_on_disk *buffer = b;
3119 memset(buffer, 0, sizeof(*buffer));
3121 buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
3122 for (i = UI_CURRENT; i < UI_SIZE; i++)
3123 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3124 buffer->flags = cpu_to_be32(device->ldev->md.flags);
3125 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
3127 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3128 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3129 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
3130 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
3131 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
3133 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3134 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
3136 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3137 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
3139 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
3140 sector = device->ldev->md.md_offset;
3142 if (drbd_md_sync_page_io(device, device->ldev, sector, REQ_OP_WRITE)) {
3143 /* this was a try anyways ... */
3144 drbd_err(device, "meta data update failed!\n");
3145 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
3150 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
3151 * @device: DRBD device.
3153 void drbd_md_sync(struct drbd_device *device)
3155 struct meta_data_on_disk *buffer;
3157 /* Don't accidentally change the DRBD meta data layout. */
3158 BUILD_BUG_ON(UI_SIZE != 4);
3159 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3161 del_timer(&device->md_sync_timer);
3162 /* timer may be rearmed by drbd_md_mark_dirty() now. */
3163 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
3166 /* We use here D_FAILED and not D_ATTACHING because we try to write
3167 * metadata even if we detach due to a disk failure! */
3168 if (!get_ldev_if_state(device, D_FAILED))
3171 buffer = drbd_md_get_buffer(device, __func__);
3175 drbd_md_write(device, buffer);
3177 /* Update device->ldev->md.la_size_sect,
3178 * since we updated it on metadata. */
3179 device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
3181 drbd_md_put_buffer(device);
3186 static int check_activity_log_stripe_size(struct drbd_device *device,
3187 struct meta_data_on_disk *on_disk,
3188 struct drbd_md *in_core)
3190 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3191 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3194 /* both not set: default to old fixed size activity log */
3195 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3197 al_stripe_size_4k = MD_32kB_SECT/8;
3200 /* some paranoia plausibility checks */
3202 /* we need both values to be set */
3203 if (al_stripes == 0 || al_stripe_size_4k == 0)
3206 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3208 /* Upper limit of activity log area, to avoid potential overflow
3209 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3210 * than 72 * 4k blocks total only increases the amount of history,
3211 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3212 if (al_size_4k > (16 * 1024 * 1024/4))
3215 /* Lower limit: we need at least 8 transaction slots (32kB)
3216 * to not break existing setups */
3217 if (al_size_4k < MD_32kB_SECT/8)
3220 in_core->al_stripe_size_4k = al_stripe_size_4k;
3221 in_core->al_stripes = al_stripes;
3222 in_core->al_size_4k = al_size_4k;
3226 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
3227 al_stripes, al_stripe_size_4k);
3231 static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
3233 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3234 struct drbd_md *in_core = &bdev->md;
3235 s32 on_disk_al_sect;
3236 s32 on_disk_bm_sect;
3238 /* The on-disk size of the activity log, calculated from offsets, and
3239 * the size of the activity log calculated from the stripe settings,
3241 * Though we could relax this a bit: it is ok, if the striped activity log
3242 * fits in the available on-disk activity log size.
3243 * Right now, that would break how resize is implemented.
3244 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3245 * of possible unused padding space in the on disk layout. */
3246 if (in_core->al_offset < 0) {
3247 if (in_core->bm_offset > in_core->al_offset)
3249 on_disk_al_sect = -in_core->al_offset;
3250 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3252 if (in_core->al_offset != MD_4kB_SECT)
3254 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3257 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3258 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3261 /* old fixed size meta data is exactly that: fixed. */
3262 if (in_core->meta_dev_idx >= 0) {
3263 if (in_core->md_size_sect != MD_128MB_SECT
3264 || in_core->al_offset != MD_4kB_SECT
3265 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3266 || in_core->al_stripes != 1
3267 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3271 if (capacity < in_core->md_size_sect)
3273 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3276 /* should be aligned, and at least 32k */
3277 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3280 /* should fit (for now: exactly) into the available on-disk space;
3281 * overflow prevention is in check_activity_log_stripe_size() above. */
3282 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3285 /* again, should be aligned */
3286 if (in_core->bm_offset & 7)
3289 /* FIXME check for device grow with flex external meta data? */
3291 /* can the available bitmap space cover the last agreed device size? */
3292 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3298 drbd_err(device, "meta data offsets don't make sense: idx=%d "
3299 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3300 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3301 in_core->meta_dev_idx,
3302 in_core->al_stripes, in_core->al_stripe_size_4k,
3303 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3304 (unsigned long long)in_core->la_size_sect,
3305 (unsigned long long)capacity);
3312 * drbd_md_read() - Reads in the meta data super block
3313 * @device: DRBD device.
3314 * @bdev: Device from which the meta data should be read in.
3316 * Return NO_ERROR on success, and an enum drbd_ret_code in case
3317 * something goes wrong.
3319 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
3320 * even before @bdev is assigned to @device->ldev.
3322 int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
3324 struct meta_data_on_disk *buffer;
3326 int i, rv = NO_ERROR;
3328 if (device->state.disk != D_DISKLESS)
3329 return ERR_DISK_CONFIGURED;
3331 buffer = drbd_md_get_buffer(device, __func__);
3335 /* First, figure out where our meta data superblock is located,
3337 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3338 bdev->md.md_offset = drbd_md_ss(bdev);
3339 /* Even for (flexible or indexed) external meta data,
3340 * initially restrict us to the 4k superblock for now.
3341 * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */
3342 bdev->md.md_size_sect = 8;
3344 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset,
3346 /* NOTE: can't do normal error processing here as this is
3347 called BEFORE disk is attached */
3348 drbd_err(device, "Error while reading metadata.\n");
3349 rv = ERR_IO_MD_DISK;
3353 magic = be32_to_cpu(buffer->magic);
3354 flags = be32_to_cpu(buffer->flags);
3355 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3356 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3357 /* btw: that's Activity Log clean, not "all" clean. */
3358 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
3359 rv = ERR_MD_UNCLEAN;
3363 rv = ERR_MD_INVALID;
3364 if (magic != DRBD_MD_MAGIC_08) {
3365 if (magic == DRBD_MD_MAGIC_07)
3366 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
3368 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
3372 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
3373 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
3374 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
3379 /* convert to in_core endian */
3380 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
3381 for (i = UI_CURRENT; i < UI_SIZE; i++)
3382 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3383 bdev->md.flags = be32_to_cpu(buffer->flags);
3384 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3386 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3387 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3388 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3390 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
3392 if (check_offsets_and_sizes(device, bdev))
3395 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
3396 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
3397 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
3400 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
3401 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
3402 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
3408 spin_lock_irq(&device->resource->req_lock);
3409 if (device->state.conn < C_CONNECTED) {
3411 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
3412 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
3413 device->peer_max_bio_size = peer;
3415 spin_unlock_irq(&device->resource->req_lock);
3418 drbd_md_put_buffer(device);
3424 * drbd_md_mark_dirty() - Mark meta data super block as dirty
3425 * @device: DRBD device.
3427 * Call this function if you change anything that should be written to
3428 * the meta-data super block. This function sets MD_DIRTY, and starts a
3429 * timer that ensures that within five seconds you have to call drbd_md_sync().
3432 void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
3434 if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3435 mod_timer(&device->md_sync_timer, jiffies + HZ);
3436 device->last_md_mark_dirty.line = line;
3437 device->last_md_mark_dirty.func = func;
3441 void drbd_md_mark_dirty(struct drbd_device *device)
3443 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3444 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
3448 void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
3452 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
3453 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
3456 void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3458 if (idx == UI_CURRENT) {
3459 if (device->state.role == R_PRIMARY)
3464 drbd_set_ed_uuid(device, val);
3467 device->ldev->md.uuid[idx] = val;
3468 drbd_md_mark_dirty(device);
3471 void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3473 unsigned long flags;
3474 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3475 __drbd_uuid_set(device, idx, val);
3476 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3479 void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
3481 unsigned long flags;
3482 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3483 if (device->ldev->md.uuid[idx]) {
3484 drbd_uuid_move_history(device);
3485 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
3487 __drbd_uuid_set(device, idx, val);
3488 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3492 * drbd_uuid_new_current() - Creates a new current UUID
3493 * @device: DRBD device.
3495 * Creates a new current UUID, and rotates the old current UUID into
3496 * the bitmap slot. Causes an incremental resync upon next connect.
3498 void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
3501 unsigned long long bm_uuid;
3503 get_random_bytes(&val, sizeof(u64));
3505 spin_lock_irq(&device->ldev->md.uuid_lock);
3506 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3509 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3511 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3512 __drbd_uuid_set(device, UI_CURRENT, val);
3513 spin_unlock_irq(&device->ldev->md.uuid_lock);
3515 drbd_print_uuids(device, "new current UUID");
3516 /* get it to stable storage _now_ */
3517 drbd_md_sync(device);
3520 void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
3522 unsigned long flags;
3523 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
3526 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3528 drbd_uuid_move_history(device);
3529 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3530 device->ldev->md.uuid[UI_BITMAP] = 0;
3532 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
3534 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
3536 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
3538 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
3540 drbd_md_mark_dirty(device);
3544 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3545 * @device: DRBD device.
3547 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3549 int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
3553 drbd_md_set_flag(device, MDF_FULL_SYNC);
3554 drbd_md_sync(device);
3555 drbd_bm_set_all(device);
3557 rv = drbd_bm_write(device);
3560 drbd_md_clear_flag(device, MDF_FULL_SYNC);
3561 drbd_md_sync(device);
3568 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
3569 * @device: DRBD device.
3571 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3573 int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
3575 drbd_resume_al(device);
3576 drbd_bm_clear_all(device);
3577 return drbd_bm_write(device);
3580 static int w_bitmap_io(struct drbd_work *w, int unused)
3582 struct drbd_device *device =
3583 container_of(w, struct drbd_device, bm_io_work.w);
3584 struct bm_io_work *work = &device->bm_io_work;
3587 if (work->flags != BM_LOCKED_CHANGE_ALLOWED) {
3588 int cnt = atomic_read(&device->ap_bio_cnt);
3590 drbd_err(device, "FIXME: ap_bio_cnt %d, expected 0; queued for '%s'\n",
3594 if (get_ldev(device)) {
3595 drbd_bm_lock(device, work->why, work->flags);
3596 rv = work->io_fn(device);
3597 drbd_bm_unlock(device);
3601 clear_bit_unlock(BITMAP_IO, &device->flags);
3602 wake_up(&device->misc_wait);
3605 work->done(device, rv);
3607 clear_bit(BITMAP_IO_QUEUED, &device->flags);
3615 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
3616 * @device: DRBD device.
3617 * @io_fn: IO callback to be called when bitmap IO is possible
3618 * @done: callback to be called after the bitmap IO was performed
3619 * @why: Descriptive text of the reason for doing the IO
3621 * While IO on the bitmap happens we freeze application IO thus we ensure
3622 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3623 * called from worker context. It MUST NOT be used while a previous such
3624 * work is still pending!
3626 * Its worker function encloses the call of io_fn() by get_ldev() and
3629 void drbd_queue_bitmap_io(struct drbd_device *device,
3630 int (*io_fn)(struct drbd_device *),
3631 void (*done)(struct drbd_device *, int),
3632 char *why, enum bm_flag flags)
3634 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
3636 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3637 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3638 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
3639 if (device->bm_io_work.why)
3640 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
3641 why, device->bm_io_work.why);
3643 device->bm_io_work.io_fn = io_fn;
3644 device->bm_io_work.done = done;
3645 device->bm_io_work.why = why;
3646 device->bm_io_work.flags = flags;
3648 spin_lock_irq(&device->resource->req_lock);
3649 set_bit(BITMAP_IO, &device->flags);
3650 /* don't wait for pending application IO if the caller indicates that
3651 * application IO does not conflict anyways. */
3652 if (flags == BM_LOCKED_CHANGE_ALLOWED || atomic_read(&device->ap_bio_cnt) == 0) {
3653 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
3654 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3655 &device->bm_io_work.w);
3657 spin_unlock_irq(&device->resource->req_lock);
3661 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
3662 * @device: DRBD device.
3663 * @io_fn: IO callback to be called when bitmap IO is possible
3664 * @why: Descriptive text of the reason for doing the IO
3666 * freezes application IO while that the actual IO operations runs. This
3667 * functions MAY NOT be called from worker context.
3669 int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
3670 char *why, enum bm_flag flags)
3672 /* Only suspend io, if some operation is supposed to be locked out */
3673 const bool do_suspend_io = flags & (BM_DONT_CLEAR|BM_DONT_SET|BM_DONT_TEST);
3676 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
3679 drbd_suspend_io(device);
3681 drbd_bm_lock(device, why, flags);
3683 drbd_bm_unlock(device);
3686 drbd_resume_io(device);
3691 void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
3693 if ((device->ldev->md.flags & flag) != flag) {
3694 drbd_md_mark_dirty(device);
3695 device->ldev->md.flags |= flag;
3699 void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
3701 if ((device->ldev->md.flags & flag) != 0) {
3702 drbd_md_mark_dirty(device);
3703 device->ldev->md.flags &= ~flag;
3706 int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3708 return (bdev->md.flags & flag) != 0;
3711 static void md_sync_timer_fn(unsigned long data)
3713 struct drbd_device *device = (struct drbd_device *) data;
3714 drbd_device_post_work(device, MD_SYNC);
3717 const char *cmdname(enum drbd_packet cmd)
3719 /* THINK may need to become several global tables
3720 * when we want to support more than
3721 * one PRO_VERSION */
3722 static const char *cmdnames[] = {
3724 [P_WSAME] = "WriteSame",
3726 [P_DATA_REPLY] = "DataReply",
3727 [P_RS_DATA_REPLY] = "RSDataReply",
3728 [P_BARRIER] = "Barrier",
3729 [P_BITMAP] = "ReportBitMap",
3730 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3731 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3732 [P_UNPLUG_REMOTE] = "UnplugRemote",
3733 [P_DATA_REQUEST] = "DataRequest",
3734 [P_RS_DATA_REQUEST] = "RSDataRequest",
3735 [P_SYNC_PARAM] = "SyncParam",
3736 [P_SYNC_PARAM89] = "SyncParam89",
3737 [P_PROTOCOL] = "ReportProtocol",
3738 [P_UUIDS] = "ReportUUIDs",
3739 [P_SIZES] = "ReportSizes",
3740 [P_STATE] = "ReportState",
3741 [P_SYNC_UUID] = "ReportSyncUUID",
3742 [P_AUTH_CHALLENGE] = "AuthChallenge",
3743 [P_AUTH_RESPONSE] = "AuthResponse",
3745 [P_PING_ACK] = "PingAck",
3746 [P_RECV_ACK] = "RecvAck",
3747 [P_WRITE_ACK] = "WriteAck",
3748 [P_RS_WRITE_ACK] = "RSWriteAck",
3749 [P_SUPERSEDED] = "Superseded",
3750 [P_NEG_ACK] = "NegAck",
3751 [P_NEG_DREPLY] = "NegDReply",
3752 [P_NEG_RS_DREPLY] = "NegRSDReply",
3753 [P_BARRIER_ACK] = "BarrierAck",
3754 [P_STATE_CHG_REQ] = "StateChgRequest",
3755 [P_STATE_CHG_REPLY] = "StateChgReply",
3756 [P_OV_REQUEST] = "OVRequest",
3757 [P_OV_REPLY] = "OVReply",
3758 [P_OV_RESULT] = "OVResult",
3759 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3760 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3761 [P_COMPRESSED_BITMAP] = "CBitmap",
3762 [P_DELAY_PROBE] = "DelayProbe",
3763 [P_OUT_OF_SYNC] = "OutOfSync",
3764 [P_RETRY_WRITE] = "RetryWrite",
3765 [P_RS_CANCEL] = "RSCancel",
3766 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3767 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
3768 [P_RETRY_WRITE] = "retry_write",
3769 [P_PROTOCOL_UPDATE] = "protocol_update",
3770 [P_RS_THIN_REQ] = "rs_thin_req",
3771 [P_RS_DEALLOCATED] = "rs_deallocated",
3773 /* enum drbd_packet, but not commands - obsoleted flags:
3779 /* too big for the array: 0xfffX */
3780 if (cmd == P_INITIAL_META)
3781 return "InitialMeta";
3782 if (cmd == P_INITIAL_DATA)
3783 return "InitialData";
3784 if (cmd == P_CONNECTION_FEATURES)
3785 return "ConnectionFeatures";
3786 if (cmd >= ARRAY_SIZE(cmdnames))
3788 return cmdnames[cmd];
3792 * drbd_wait_misc - wait for a request to make progress
3793 * @device: device associated with the request
3794 * @i: the struct drbd_interval embedded in struct drbd_request or
3795 * struct drbd_peer_request
3797 int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
3799 struct net_conf *nc;
3804 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
3809 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3812 /* Indicate to wake up device->misc_wait on progress. */
3814 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
3815 spin_unlock_irq(&device->resource->req_lock);
3816 timeout = schedule_timeout(timeout);
3817 finish_wait(&device->misc_wait, &wait);
3818 spin_lock_irq(&device->resource->req_lock);
3819 if (!timeout || device->state.conn < C_CONNECTED)
3821 if (signal_pending(current))
3822 return -ERESTARTSYS;
3826 void lock_all_resources(void)
3828 struct drbd_resource *resource;
3829 int __maybe_unused i = 0;
3831 mutex_lock(&resources_mutex);
3832 local_irq_disable();
3833 for_each_resource(resource, &drbd_resources)
3834 spin_lock_nested(&resource->req_lock, i++);
3837 void unlock_all_resources(void)
3839 struct drbd_resource *resource;
3841 for_each_resource(resource, &drbd_resources)
3842 spin_unlock(&resource->req_lock);
3844 mutex_unlock(&resources_mutex);
3847 #ifdef CONFIG_DRBD_FAULT_INJECTION
3848 /* Fault insertion support including random number generator shamelessly
3849 * stolen from kernel/rcutorture.c */
3850 struct fault_random_state {
3851 unsigned long state;
3852 unsigned long count;
3855 #define FAULT_RANDOM_MULT 39916801 /* prime */
3856 #define FAULT_RANDOM_ADD 479001701 /* prime */
3857 #define FAULT_RANDOM_REFRESH 10000
3860 * Crude but fast random-number generator. Uses a linear congruential
3861 * generator, with occasional help from get_random_bytes().
3863 static unsigned long
3864 _drbd_fault_random(struct fault_random_state *rsp)
3868 if (!rsp->count--) {
3869 get_random_bytes(&refresh, sizeof(refresh));
3870 rsp->state += refresh;
3871 rsp->count = FAULT_RANDOM_REFRESH;
3873 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3874 return swahw32(rsp->state);
3878 _drbd_fault_str(unsigned int type) {
3879 static char *_faults[] = {
3880 [DRBD_FAULT_MD_WR] = "Meta-data write",
3881 [DRBD_FAULT_MD_RD] = "Meta-data read",
3882 [DRBD_FAULT_RS_WR] = "Resync write",
3883 [DRBD_FAULT_RS_RD] = "Resync read",
3884 [DRBD_FAULT_DT_WR] = "Data write",
3885 [DRBD_FAULT_DT_RD] = "Data read",
3886 [DRBD_FAULT_DT_RA] = "Data read ahead",
3887 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
3888 [DRBD_FAULT_AL_EE] = "EE allocation",
3889 [DRBD_FAULT_RECEIVE] = "receive data corruption",
3892 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3896 _drbd_insert_fault(struct drbd_device *device, unsigned int type)
3898 static struct fault_random_state rrs = {0, 0};
3900 unsigned int ret = (
3902 ((1 << device_to_minor(device)) & fault_devs) != 0) &&
3903 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3908 if (__ratelimit(&drbd_ratelimit_state))
3909 drbd_warn(device, "***Simulating %s failure\n",
3910 _drbd_fault_str(type));
3917 const char *drbd_buildtag(void)
3919 /* DRBD built from external sources has here a reference to the
3920 git hash of the source code. */
3922 static char buildtag[38] = "\0uilt-in";
3924 if (buildtag[0] == 0) {
3926 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3935 module_init(drbd_init)
3936 module_exit(drbd_cleanup)
3938 EXPORT_SYMBOL(drbd_conn_str);
3939 EXPORT_SYMBOL(drbd_role_str);
3940 EXPORT_SYMBOL(drbd_disk_str);
3941 EXPORT_SYMBOL(drbd_set_st_err_str);