blob: 686c5a59edebc84eb3da455d87c8e5c3fe291d12 [file] [log] [blame]
Philipp Reisnerb411b362009-09-25 16:07:19 -07001/*
2 drbd.c
3
4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
5
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>.
9
10 Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev
11 from Logicworks, Inc. for making SDP replication support possible.
12
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)
16 any later version.
17
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.
22
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.
26
27 */
28
Lars Ellenbergf88c5d92014-03-27 14:10:55 +010029#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
Philipp Reisnerb411b362009-09-25 16:07:19 -070031#include <linux/module.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070032#include <linux/drbd.h>
33#include <asm/uaccess.h>
34#include <asm/types.h>
35#include <net/sock.h>
36#include <linux/ctype.h>
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020037#include <linux/mutex.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070038#include <linux/fs.h>
39#include <linux/file.h>
40#include <linux/proc_fs.h>
41#include <linux/init.h>
42#include <linux/mm.h>
43#include <linux/memcontrol.h>
44#include <linux/mm_inline.h>
45#include <linux/slab.h>
46#include <linux/random.h>
47#include <linux/reboot.h>
48#include <linux/notifier.h>
49#include <linux/kthread.h>
Lars Ellenberg113fef92013-03-22 18:14:40 -060050#include <linux/workqueue.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070051#define __KERNEL_SYSCALLS__
52#include <linux/unistd.h>
53#include <linux/vmalloc.h>
54
55#include <linux/drbd_limits.h>
56#include "drbd_int.h"
Andreas Gruenbachera3603a62011-05-30 11:47:37 +020057#include "drbd_protocol.h"
Philipp Reisnerb411b362009-09-25 16:07:19 -070058#include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */
59
60#include "drbd_vli.h"
61
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020062static DEFINE_MUTEX(drbd_main_mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -070063static int drbd_open(struct block_device *bdev, fmode_t mode);
Al Virodb2a1442013-05-05 21:52:57 -040064static void drbd_release(struct gendisk *gd, fmode_t mode);
Philipp Reisnerb411b362009-09-25 16:07:19 -070065static void md_sync_timer_fn(unsigned long data);
Andreas Gruenbacher99920dc2011-03-16 15:31:39 +010066static int w_bitmap_io(struct drbd_work *w, int unused);
Philipp Reisnerb411b362009-09-25 16:07:19 -070067
Philipp Reisnerb411b362009-09-25 16:07:19 -070068MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, "
69 "Lars Ellenberg <lars@linbit.com>");
70MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION);
71MODULE_VERSION(REL_VERSION);
72MODULE_LICENSE("GPL");
Philipp Reisner81a5d602011-02-22 19:53:16 -050073MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices ("
Philipp Reisner2b8a90b2011-01-10 11:15:17 +010074 __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")");
Philipp Reisnerb411b362009-09-25 16:07:19 -070075MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR);
76
77#include <linux/moduleparam.h>
78/* allow_open_on_secondary */
79MODULE_PARM_DESC(allow_oos, "DONT USE!");
80/* thanks to these macros, if compiled into the kernel (not-module),
81 * this becomes the boot parameter drbd.minor_count */
82module_param(minor_count, uint, 0444);
83module_param(disable_sendpage, bool, 0644);
84module_param(allow_oos, bool, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -070085module_param(proc_details, int, 0644);
86
87#ifdef CONFIG_DRBD_FAULT_INJECTION
88int enable_faults;
89int fault_rate;
90static int fault_count;
91int fault_devs;
92/* bitmap of enabled faults */
93module_param(enable_faults, int, 0664);
94/* fault rate % value - applies to all enabled faults */
95module_param(fault_rate, int, 0664);
96/* count of faults inserted */
97module_param(fault_count, int, 0664);
98/* bitmap of devices to insert faults on */
99module_param(fault_devs, int, 0644);
100#endif
101
102/* module parameter, defined */
Philipp Reisner2b8a90b2011-01-10 11:15:17 +0100103unsigned int minor_count = DRBD_MINOR_COUNT_DEF;
Rusty Russell90ab5ee2012-01-13 09:32:20 +1030104bool disable_sendpage;
105bool allow_oos;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700106int proc_details; /* Detail level in proc drbd*/
107
108/* Module parameter for setting the user mode helper program
109 * to run. Default is /sbin/drbdadm */
110char usermode_helper[80] = "/sbin/drbdadm";
111
112module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644);
113
114/* in 2.6.x, our device mapping and config info contains our virtual gendisks
115 * as member "struct gendisk *vdisk;"
116 */
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +0200117struct idr drbd_devices;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +0200118struct list_head drbd_resources;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700119
120struct kmem_cache *drbd_request_cache;
Andreas Gruenbacher6c852be2011-02-04 15:38:52 +0100121struct kmem_cache *drbd_ee_cache; /* peer requests */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700122struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */
123struct kmem_cache *drbd_al_ext_cache; /* activity log extents */
124mempool_t *drbd_request_mempool;
125mempool_t *drbd_ee_mempool;
Lars Ellenberg42818082011-02-23 12:39:46 +0100126mempool_t *drbd_md_io_page_pool;
Lars Ellenberg9476f392011-02-23 17:02:01 +0100127struct bio_set *drbd_md_io_bio_set;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700128
129/* I do not use a standard mempool, because:
130 1) I want to hand out the pre-allocated objects first.
131 2) I want to be able to interrupt sleeping allocation with a signal.
132 Note: This is a single linked list, the next pointer is the private
133 member of struct page.
134 */
135struct page *drbd_pp_pool;
136spinlock_t drbd_pp_lock;
137int drbd_pp_vacant;
138wait_queue_head_t drbd_pp_wait;
139
140DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5);
141
Emese Revfy7d4e9d02009-12-14 00:59:30 +0100142static const struct block_device_operations drbd_ops = {
Philipp Reisnerb411b362009-09-25 16:07:19 -0700143 .owner = THIS_MODULE,
144 .open = drbd_open,
145 .release = drbd_release,
146};
147
Lars Ellenberg9476f392011-02-23 17:02:01 +0100148struct bio *bio_alloc_drbd(gfp_t gfp_mask)
149{
Lars Ellenbergda4a75d2011-02-23 17:02:01 +0100150 struct bio *bio;
151
Lars Ellenberg9476f392011-02-23 17:02:01 +0100152 if (!drbd_md_io_bio_set)
153 return bio_alloc(gfp_mask, 1);
154
Lars Ellenbergda4a75d2011-02-23 17:02:01 +0100155 bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
156 if (!bio)
157 return NULL;
Lars Ellenbergda4a75d2011-02-23 17:02:01 +0100158 return bio;
Lars Ellenberg9476f392011-02-23 17:02:01 +0100159}
160
Philipp Reisnerb411b362009-09-25 16:07:19 -0700161#ifdef __CHECKER__
162/* When checking with sparse, and this is an inline function, sparse will
163 give tons of false positives. When this is a real functions sparse works.
164 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200165int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700166{
167 int io_allowed;
168
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200169 atomic_inc(&device->local_cnt);
170 io_allowed = (device->state.disk >= mins);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700171 if (!io_allowed) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200172 if (atomic_dec_and_test(&device->local_cnt))
173 wake_up(&device->misc_wait);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700174 }
175 return io_allowed;
176}
177
178#endif
179
180/**
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100181 * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200182 * @connection: DRBD connection.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700183 * @barrier_nr: Expected identifier of the DRBD write barrier packet.
184 * @set_size: Expected number of requests before that barrier.
185 *
186 * In case the passed barrier_nr or set_size does not match the oldest
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100187 * epoch of not yet barrier-acked requests, this function will cause a
188 * termination of the connection.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700189 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200190void tl_release(struct drbd_connection *connection, unsigned int barrier_nr,
Philipp Reisner2f5cdd02011-02-21 14:29:27 +0100191 unsigned int set_size)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700192{
Philipp Reisnerb411b362009-09-25 16:07:19 -0700193 struct drbd_request *r;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100194 struct drbd_request *req = NULL;
195 int expect_epoch = 0;
196 int expect_size = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700197
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200198 spin_lock_irq(&connection->resource->req_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700199
Philipp Reisner98683652012-11-09 14:18:43 +0100200 /* find oldest not yet barrier-acked write request,
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100201 * count writes in its epoch. */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200202 list_for_each_entry(r, &connection->transfer_log, tl_requests) {
Lars Ellenberga0d856d2012-01-24 17:19:42 +0100203 const unsigned s = r->rq_state;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100204 if (!req) {
205 if (!(s & RQ_WRITE))
206 continue;
207 if (!(s & RQ_NET_MASK))
208 continue;
209 if (s & RQ_NET_DONE)
210 continue;
211 req = r;
212 expect_epoch = req->epoch;
213 expect_size ++;
214 } else {
215 if (r->epoch != expect_epoch)
216 break;
217 if (!(s & RQ_WRITE))
218 continue;
219 /* if (s & RQ_DONE): not expected */
220 /* if (!(s & RQ_NET_MASK)): not expected */
221 expect_size++;
222 }
223 }
Philipp Reisnerb411b362009-09-25 16:07:19 -0700224
225 /* first some paranoia code */
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100226 if (req == NULL) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +0200227 drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n",
Philipp Reisner2f5cdd02011-02-21 14:29:27 +0100228 barrier_nr);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700229 goto bail;
230 }
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100231 if (expect_epoch != barrier_nr) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +0200232 drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n",
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100233 barrier_nr, expect_epoch);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700234 goto bail;
235 }
236
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100237 if (expect_size != set_size) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +0200238 drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n",
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100239 barrier_nr, set_size, expect_size);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700240 goto bail;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700241 }
242
Philipp Reisner98683652012-11-09 14:18:43 +0100243 /* Clean up list of requests processed during current epoch. */
244 /* this extra list walk restart is paranoia,
245 * to catch requests being barrier-acked "unexpectedly".
246 * It usually should find the same req again, or some READ preceding it. */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200247 list_for_each_entry(req, &connection->transfer_log, tl_requests)
Philipp Reisner98683652012-11-09 14:18:43 +0100248 if (req->epoch == expect_epoch)
249 break;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200250 list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) {
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100251 if (req->epoch != expect_epoch)
252 break;
253 _req_mod(req, BARRIER_ACKED);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700254 }
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200255 spin_unlock_irq(&connection->resource->req_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700256
257 return;
258
259bail:
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200260 spin_unlock_irq(&connection->resource->req_lock);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200261 conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700262}
263
Philipp Reisner617049a2010-12-22 12:48:31 +0100264
Philipp Reisner11b58e72010-05-12 17:08:26 +0200265/**
266 * _tl_restart() - Walks the transfer log, and applies an action to all requests
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200267 * @device: DRBD device.
Philipp Reisner11b58e72010-05-12 17:08:26 +0200268 * @what: The action/event to perform with all request objects
269 *
Andreas Gruenbacher8554df12011-01-25 15:37:43 +0100270 * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO,
271 * RESTART_FROZEN_DISK_IO.
Philipp Reisner11b58e72010-05-12 17:08:26 +0200272 */
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100273/* must hold resource->req_lock */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200274void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
Philipp Reisner11b58e72010-05-12 17:08:26 +0200275{
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100276 struct drbd_request *req, *r;
Philipp Reisner11b58e72010-05-12 17:08:26 +0200277
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200278 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests)
Philipp Reisner509fc012012-07-31 11:22:58 +0200279 _req_mod(req, what);
Philipp Reisner11b58e72010-05-12 17:08:26 +0200280}
281
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200282void tl_restart(struct drbd_connection *connection, enum drbd_req_event what)
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100283{
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200284 spin_lock_irq(&connection->resource->req_lock);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200285 _tl_restart(connection, what);
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200286 spin_unlock_irq(&connection->resource->req_lock);
Philipp Reisnercdfda632011-07-05 15:38:59 +0200287}
Philipp Reisnerb411b362009-09-25 16:07:19 -0700288
289/**
290 * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200291 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700292 *
293 * This is called after the connection to the peer was lost. The storage covered
294 * by the requests on the transfer gets marked as our of sync. Called from the
295 * receiver thread and the worker thread.
296 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200297void tl_clear(struct drbd_connection *connection)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700298{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200299 tl_restart(connection, CONNECTION_LOST_WHILE_PENDING);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700300}
301
302/**
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200303 * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL
304 * @device: DRBD device.
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200305 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200306void tl_abort_disk_io(struct drbd_device *device)
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200307{
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +0200308 struct drbd_connection *connection = first_peer_device(device)->connection;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100309 struct drbd_request *req, *r;
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200310
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200311 spin_lock_irq(&connection->resource->req_lock);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200312 list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) {
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200313 if (!(req->rq_state & RQ_LOCAL_PENDING))
314 continue;
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +0200315 if (req->device != device)
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +0100316 continue;
317 _req_mod(req, ABORT_DISK_IO);
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200318 }
Andreas Gruenbacher05008132011-07-07 14:19:42 +0200319 spin_unlock_irq(&connection->resource->req_lock);
Philipp Reisnerfd2491f2011-07-18 16:25:15 +0200320}
321
Philipp Reisnerb411b362009-09-25 16:07:19 -0700322static int drbd_thread_setup(void *arg)
323{
324 struct drbd_thread *thi = (struct drbd_thread *) arg;
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200325 struct drbd_resource *resource = thi->resource;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700326 unsigned long flags;
327 int retval;
328
Philipp Reisnerf1b3a6e2011-02-08 15:35:58 +0100329 snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s",
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +0200330 thi->name[0],
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200331 resource->name);
Philipp Reisnerf1b3a6e2011-02-08 15:35:58 +0100332
Philipp Reisnerb411b362009-09-25 16:07:19 -0700333restart:
334 retval = thi->function(thi);
335
336 spin_lock_irqsave(&thi->t_lock, flags);
337
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100338 /* if the receiver has been "EXITING", the last thing it did
Philipp Reisnerb411b362009-09-25 16:07:19 -0700339 * was set the conn state to "StandAlone",
340 * if now a re-connect request comes in, conn state goes C_UNCONNECTED,
341 * and receiver thread will be "started".
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100342 * drbd_thread_start needs to set "RESTARTING" in that case.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700343 * t_state check and assignment needs to be within the same spinlock,
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100344 * so either thread_start sees EXITING, and can remap to RESTARTING,
345 * or thread_start see NONE, and can proceed as normal.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700346 */
347
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100348 if (thi->t_state == RESTARTING) {
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200349 drbd_info(resource, "Restarting %s thread\n", thi->name);
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100350 thi->t_state = RUNNING;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700351 spin_unlock_irqrestore(&thi->t_lock, flags);
352 goto restart;
353 }
354
355 thi->task = NULL;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100356 thi->t_state = NONE;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700357 smp_mb();
Lars Ellenberg992d6e92011-05-02 11:47:18 +0200358 complete_all(&thi->stop);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700359 spin_unlock_irqrestore(&thi->t_lock, flags);
360
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200361 drbd_info(resource, "Terminating %s\n", current->comm);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700362
363 /* Release mod reference taken when thread was started */
Philipp Reisner9dc9fbb2011-04-22 15:23:32 +0200364
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200365 if (thi->connection)
366 kref_put(&thi->connection->kref, drbd_destroy_connection);
367 kref_put(&resource->kref, drbd_destroy_resource);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700368 module_put(THIS_MODULE);
369 return retval;
370}
371
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200372static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi,
Andreas Gruenbacherc60b0252011-08-10 15:05:02 +0200373 int (*func) (struct drbd_thread *), const char *name)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700374{
375 spin_lock_init(&thi->t_lock);
376 thi->task = NULL;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100377 thi->t_state = NONE;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700378 thi->function = func;
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200379 thi->resource = resource;
380 thi->connection = NULL;
Andreas Gruenbacherc60b0252011-08-10 15:05:02 +0200381 thi->name = name;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700382}
383
384int drbd_thread_start(struct drbd_thread *thi)
385{
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200386 struct drbd_resource *resource = thi->resource;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700387 struct task_struct *nt;
388 unsigned long flags;
389
Philipp Reisnerb411b362009-09-25 16:07:19 -0700390 /* is used from state engine doing drbd_thread_stop_nowait,
391 * while holding the req lock irqsave */
392 spin_lock_irqsave(&thi->t_lock, flags);
393
394 switch (thi->t_state) {
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100395 case NONE:
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200396 drbd_info(resource, "Starting %s thread (from %s [%d])\n",
Philipp Reisnerbed879a2011-02-04 14:00:37 +0100397 thi->name, current->comm, current->pid);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700398
399 /* Get ref on module for thread - this is released when thread exits */
400 if (!try_module_get(THIS_MODULE)) {
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200401 drbd_err(resource, "Failed to get module reference in drbd_thread_start\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -0700402 spin_unlock_irqrestore(&thi->t_lock, flags);
Andreas Gruenbacher81e84652010-12-09 15:03:57 +0100403 return false;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700404 }
405
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200406 kref_get(&resource->kref);
407 if (thi->connection)
408 kref_get(&thi->connection->kref);
Philipp Reisner9dc9fbb2011-04-22 15:23:32 +0200409
Philipp Reisnerb411b362009-09-25 16:07:19 -0700410 init_completion(&thi->stop);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700411 thi->reset_cpu_mask = 1;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100412 thi->t_state = RUNNING;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700413 spin_unlock_irqrestore(&thi->t_lock, flags);
414 flush_signals(current); /* otherw. may get -ERESTARTNOINTR */
415
416 nt = kthread_create(drbd_thread_setup, (void *) thi,
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200417 "drbd_%c_%s", thi->name[0], thi->resource->name);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700418
419 if (IS_ERR(nt)) {
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200420 drbd_err(resource, "Couldn't start thread\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -0700421
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200422 if (thi->connection)
423 kref_put(&thi->connection->kref, drbd_destroy_connection);
424 kref_put(&resource->kref, drbd_destroy_resource);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700425 module_put(THIS_MODULE);
Andreas Gruenbacher81e84652010-12-09 15:03:57 +0100426 return false;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700427 }
428 spin_lock_irqsave(&thi->t_lock, flags);
429 thi->task = nt;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100430 thi->t_state = RUNNING;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700431 spin_unlock_irqrestore(&thi->t_lock, flags);
432 wake_up_process(nt);
433 break;
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100434 case EXITING:
435 thi->t_state = RESTARTING;
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200436 drbd_info(resource, "Restarting %s thread (from %s [%d])\n",
Philipp Reisnerbed879a2011-02-04 14:00:37 +0100437 thi->name, current->comm, current->pid);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700438 /* fall through */
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100439 case RUNNING:
440 case RESTARTING:
Philipp Reisnerb411b362009-09-25 16:07:19 -0700441 default:
442 spin_unlock_irqrestore(&thi->t_lock, flags);
443 break;
444 }
445
Andreas Gruenbacher81e84652010-12-09 15:03:57 +0100446 return true;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700447}
448
449
450void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait)
451{
452 unsigned long flags;
453
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100454 enum drbd_thread_state ns = restart ? RESTARTING : EXITING;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700455
456 /* may be called from state engine, holding the req lock irqsave */
457 spin_lock_irqsave(&thi->t_lock, flags);
458
Andreas Gruenbachere77a0a52011-01-25 15:43:39 +0100459 if (thi->t_state == NONE) {
Philipp Reisnerb411b362009-09-25 16:07:19 -0700460 spin_unlock_irqrestore(&thi->t_lock, flags);
461 if (restart)
462 drbd_thread_start(thi);
463 return;
464 }
465
466 if (thi->t_state != ns) {
467 if (thi->task == NULL) {
468 spin_unlock_irqrestore(&thi->t_lock, flags);
469 return;
470 }
471
472 thi->t_state = ns;
473 smp_mb();
474 init_completion(&thi->stop);
475 if (thi->task != current)
476 force_sig(DRBD_SIGKILL, thi->task);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700477 }
478
479 spin_unlock_irqrestore(&thi->t_lock, flags);
480
481 if (wait)
482 wait_for_completion(&thi->stop);
483}
484
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200485int conn_lowest_minor(struct drbd_connection *connection)
Philipp Reisner80822282011-02-08 12:46:30 +0100486{
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +0200487 struct drbd_peer_device *peer_device;
488 int vnr = 0, minor = -1;
Philipp Reisner774b3052011-02-22 02:07:03 -0500489
Philipp Reisner695d08f2011-04-11 22:53:32 -0700490 rcu_read_lock();
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +0200491 peer_device = idr_get_next(&connection->peer_devices, &vnr);
492 if (peer_device)
493 minor = device_to_minor(peer_device->device);
Philipp Reisner695d08f2011-04-11 22:53:32 -0700494 rcu_read_unlock();
495
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +0200496 return minor;
Philipp Reisner80822282011-02-08 12:46:30 +0100497}
Philipp Reisner774b3052011-02-22 02:07:03 -0500498
Philipp Reisnerb411b362009-09-25 16:07:19 -0700499#ifdef CONFIG_SMP
500/**
501 * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs
Philipp Reisnerb411b362009-09-25 16:07:19 -0700502 *
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200503 * Forces all threads of a resource onto the same CPU. This is beneficial for
Philipp Reisnerb411b362009-09-25 16:07:19 -0700504 * DRBD's performance. May be overwritten by user's configuration.
505 */
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200506static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700507{
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200508 unsigned int *resources_per_cpu, min_index = ~0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700509
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200510 resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL);
511 if (resources_per_cpu) {
512 struct drbd_resource *resource;
513 unsigned int cpu, min = ~0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700514
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200515 rcu_read_lock();
516 for_each_resource_rcu(resource, &drbd_resources) {
517 for_each_cpu(cpu, resource->cpu_mask)
518 resources_per_cpu[cpu]++;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700519 }
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200520 rcu_read_unlock();
521 for_each_online_cpu(cpu) {
522 if (resources_per_cpu[cpu] < min) {
523 min = resources_per_cpu[cpu];
524 min_index = cpu;
525 }
526 }
527 kfree(resources_per_cpu);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700528 }
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200529 if (min_index == ~0) {
530 cpumask_setall(*cpu_mask);
531 return;
532 }
533 cpumask_set_cpu(min_index, *cpu_mask);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700534}
535
536/**
537 * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200538 * @device: DRBD device.
Philipp Reisnerbc31fe32011-02-07 11:14:38 +0100539 * @thi: drbd_thread object
Philipp Reisnerb411b362009-09-25 16:07:19 -0700540 *
541 * call in the "main loop" of _all_ threads, no need for any mutex, current won't die
542 * prematurely.
543 */
Philipp Reisner80822282011-02-08 12:46:30 +0100544void drbd_thread_current_set_cpu(struct drbd_thread *thi)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700545{
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200546 struct drbd_resource *resource = thi->resource;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700547 struct task_struct *p = current;
Philipp Reisnerbed879a2011-02-04 14:00:37 +0100548
Philipp Reisnerb411b362009-09-25 16:07:19 -0700549 if (!thi->reset_cpu_mask)
550 return;
551 thi->reset_cpu_mask = 0;
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +0200552 set_cpus_allowed_ptr(p, resource->cpu_mask);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700553}
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +0200554#else
555#define drbd_calc_cpu_mask(A) ({})
Philipp Reisnerb411b362009-09-25 16:07:19 -0700556#endif
557
Andreas Gruenbacher52b061a2011-03-30 11:38:49 +0200558/**
559 * drbd_header_size - size of a packet header
560 *
561 * The header size is a multiple of 8, so any payload following the header is
562 * word aligned on 64-bit architectures. (The bitmap send and receive code
563 * relies on this.)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700564 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200565unsigned int drbd_header_size(struct drbd_connection *connection)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700566{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200567 if (connection->agreed_pro_version >= 100) {
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200568 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8));
569 return sizeof(struct p_header100);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700570 } else {
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200571 BUILD_BUG_ON(sizeof(struct p_header80) !=
572 sizeof(struct p_header95));
573 BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8));
574 return sizeof(struct p_header80);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700575 }
Andreas Gruenbacher52b061a2011-03-30 11:38:49 +0200576}
Philipp Reisnerb411b362009-09-25 16:07:19 -0700577
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200578static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size)
Philipp Reisnerfd340c12011-01-19 16:57:39 +0100579{
580 h->magic = cpu_to_be32(DRBD_MAGIC);
581 h->command = cpu_to_be16(cmd);
582 h->length = cpu_to_be16(size);
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200583 return sizeof(struct p_header80);
Philipp Reisnerfd340c12011-01-19 16:57:39 +0100584}
Philipp Reisnerb411b362009-09-25 16:07:19 -0700585
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200586static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size)
Philipp Reisnerfd340c12011-01-19 16:57:39 +0100587{
588 h->magic = cpu_to_be16(DRBD_MAGIC_BIG);
589 h->command = cpu_to_be16(cmd);
Andreas Gruenbacherb55d84b2011-03-22 13:17:47 +0100590 h->length = cpu_to_be32(size);
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200591 return sizeof(struct p_header95);
Philipp Reisnerfd340c12011-01-19 16:57:39 +0100592}
593
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200594static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd,
595 int size, int vnr)
Philipp Reisnerd38e7872011-02-07 15:32:04 +0100596{
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200597 h->magic = cpu_to_be32(DRBD_MAGIC_100);
598 h->volume = cpu_to_be16(vnr);
599 h->command = cpu_to_be16(cmd);
600 h->length = cpu_to_be32(size);
601 h->pad = 0;
602 return sizeof(struct p_header100);
603}
604
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200605static unsigned int prepare_header(struct drbd_connection *connection, int vnr,
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200606 void *buffer, enum drbd_packet cmd, int size)
607{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200608 if (connection->agreed_pro_version >= 100)
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200609 return prepare_header100(buffer, cmd, size, vnr);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200610 else if (connection->agreed_pro_version >= 95 &&
Andreas Gruenbacher0c8e36d2011-03-30 16:00:17 +0200611 size > DRBD_MAX_SIZE_H80_PACKET)
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200612 return prepare_header95(buffer, cmd, size);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700613 else
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200614 return prepare_header80(buffer, cmd, size);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700615}
616
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200617static void *__conn_prepare_command(struct drbd_connection *connection,
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200618 struct drbd_socket *sock)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700619{
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200620 if (!sock->socket)
621 return NULL;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200622 return sock->sbuf + drbd_header_size(connection);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700623}
624
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200625void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700626{
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200627 void *p;
628
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200629 mutex_lock(&sock->mutex);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200630 p = __conn_prepare_command(connection, sock);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200631 if (!p)
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200632 mutex_unlock(&sock->mutex);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200633
634 return p;
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200635}
636
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200637void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock)
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200638{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200639 return conn_prepare_command(peer_device->connection, sock);
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200640}
641
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200642static int __send_command(struct drbd_connection *connection, int vnr,
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200643 struct drbd_socket *sock, enum drbd_packet cmd,
644 unsigned int header_size, void *data,
645 unsigned int size)
646{
647 int msg_flags;
648 int err;
649
650 /*
651 * Called with @data == NULL and the size of the data blocks in @size
652 * for commands that send data blocks. For those commands, omit the
653 * MSG_MORE flag: this will increase the likelihood that data blocks
654 * which are page aligned on the sender will end up page aligned on the
655 * receiver.
656 */
657 msg_flags = data ? MSG_MORE : 0;
658
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200659 header_size += prepare_header(connection, vnr, sock->sbuf, cmd,
Andreas Gruenbachere6589832011-03-30 12:54:42 +0200660 header_size + size);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200661 err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size,
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200662 msg_flags);
663 if (data && !err)
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200664 err = drbd_send_all(connection, sock->socket, data, size, 0);
Lars Ellenberg123ff122014-02-05 06:13:53 +0100665 /* DRBD protocol "pings" are latency critical.
666 * This is supposed to trigger tcp_push_pending_frames() */
667 if (!err && (cmd == P_PING || cmd == P_PING_ACK))
668 drbd_tcp_nodelay(sock->socket);
669
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200670 return err;
671}
672
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200673static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200674 enum drbd_packet cmd, unsigned int header_size,
675 void *data, unsigned int size)
676{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200677 return __send_command(connection, 0, sock, cmd, header_size, data, size);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200678}
679
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200680int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock,
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200681 enum drbd_packet cmd, unsigned int header_size,
682 void *data, unsigned int size)
683{
684 int err;
685
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200686 err = __conn_send_command(connection, sock, cmd, header_size, data, size);
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200687 mutex_unlock(&sock->mutex);
688 return err;
689}
690
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200691int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock,
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200692 enum drbd_packet cmd, unsigned int header_size,
693 void *data, unsigned int size)
694{
695 int err;
696
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200697 err = __send_command(peer_device->connection, peer_device->device->vnr,
698 sock, cmd, header_size, data, size);
Andreas Gruenbacherdba58582011-03-29 16:55:40 +0200699 mutex_unlock(&sock->mutex);
700 return err;
701}
702
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200703int drbd_send_ping(struct drbd_connection *connection)
Andreas Gruenbachere307f352011-03-22 10:55:48 +0100704{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200705 struct drbd_socket *sock;
706
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200707 sock = &connection->meta;
708 if (!conn_prepare_command(connection, sock))
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200709 return -EIO;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200710 return conn_send_command(connection, sock, P_PING, 0, NULL, 0);
Andreas Gruenbachere307f352011-03-22 10:55:48 +0100711}
712
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200713int drbd_send_ping_ack(struct drbd_connection *connection)
Andreas Gruenbachere307f352011-03-22 10:55:48 +0100714{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200715 struct drbd_socket *sock;
716
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200717 sock = &connection->meta;
718 if (!conn_prepare_command(connection, sock))
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200719 return -EIO;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200720 return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0);
Andreas Gruenbachere307f352011-03-22 10:55:48 +0100721}
722
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200723int drbd_send_sync_param(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700724{
Andreas Gruenbacher7c967152011-03-22 00:49:36 +0100725 struct drbd_socket *sock;
Philipp Reisner8e26f9c2010-07-06 17:25:54 +0200726 struct p_rs_param_95 *p;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200727 int size;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200728 const int apv = peer_device->connection->agreed_pro_version;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200729 enum drbd_packet cmd;
Philipp Reisner44ed1672011-04-19 17:10:19 +0200730 struct net_conf *nc;
Philipp Reisnerdaeda1c2011-05-03 15:00:55 +0200731 struct disk_conf *dc;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200732
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200733 sock = &peer_device->connection->data;
734 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200735 if (!p)
736 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700737
Philipp Reisner44ed1672011-04-19 17:10:19 +0200738 rcu_read_lock();
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200739 nc = rcu_dereference(peer_device->connection->net_conf);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700740
741 size = apv <= 87 ? sizeof(struct p_rs_param)
742 : apv == 88 ? sizeof(struct p_rs_param)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200743 + strlen(nc->verify_alg) + 1
Philipp Reisner8e26f9c2010-07-06 17:25:54 +0200744 : apv <= 94 ? sizeof(struct p_rs_param_89)
745 : /* apv >= 95 */ sizeof(struct p_rs_param_95);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700746
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200747 cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700748
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200749 /* initialize verify_alg and csums_alg */
750 memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700751
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200752 if (get_ldev(peer_device->device)) {
753 dc = rcu_dereference(peer_device->device->ldev->disk_conf);
Andreas Gruenbacher6394b932011-05-11 14:29:52 +0200754 p->resync_rate = cpu_to_be32(dc->resync_rate);
Philipp Reisnerdaeda1c2011-05-03 15:00:55 +0200755 p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead);
756 p->c_delay_target = cpu_to_be32(dc->c_delay_target);
757 p->c_fill_target = cpu_to_be32(dc->c_fill_target);
758 p->c_max_rate = cpu_to_be32(dc->c_max_rate);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200759 put_ldev(peer_device->device);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200760 } else {
Andreas Gruenbacher6394b932011-05-11 14:29:52 +0200761 p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200762 p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF);
763 p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF);
764 p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF);
765 p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF);
766 }
Philipp Reisnerb411b362009-09-25 16:07:19 -0700767
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200768 if (apv >= 88)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200769 strcpy(p->verify_alg, nc->verify_alg);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200770 if (apv >= 89)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200771 strcpy(p->csums_alg, nc->csums_alg);
772 rcu_read_unlock();
Philipp Reisnerb411b362009-09-25 16:07:19 -0700773
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200774 return drbd_send_command(peer_device, sock, cmd, size, NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700775}
776
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200777int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700778{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200779 struct drbd_socket *sock;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700780 struct p_protocol *p;
Philipp Reisner44ed1672011-04-19 17:10:19 +0200781 struct net_conf *nc;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200782 int size, cf;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700783
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200784 sock = &connection->data;
785 p = __conn_prepare_command(connection, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200786 if (!p)
787 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700788
Philipp Reisner44ed1672011-04-19 17:10:19 +0200789 rcu_read_lock();
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200790 nc = rcu_dereference(connection->net_conf);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700791
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200792 if (nc->tentative && connection->agreed_pro_version < 92) {
Philipp Reisner44ed1672011-04-19 17:10:19 +0200793 rcu_read_unlock();
794 mutex_unlock(&sock->mutex);
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +0200795 drbd_err(connection, "--dry-run is not supported by peer");
Philipp Reisner44ed1672011-04-19 17:10:19 +0200796 return -EOPNOTSUPP;
Philipp Reisnercf14c2e2010-02-02 21:03:50 +0100797 }
Philipp Reisner44ed1672011-04-19 17:10:19 +0200798
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200799 size = sizeof(*p);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200800 if (connection->agreed_pro_version >= 87)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200801 size += strlen(nc->integrity_alg) + 1;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700802
Philipp Reisner44ed1672011-04-19 17:10:19 +0200803 p->protocol = cpu_to_be32(nc->wire_protocol);
804 p->after_sb_0p = cpu_to_be32(nc->after_sb_0p);
805 p->after_sb_1p = cpu_to_be32(nc->after_sb_1p);
806 p->after_sb_2p = cpu_to_be32(nc->after_sb_2p);
807 p->two_primaries = cpu_to_be32(nc->two_primaries);
Philipp Reisnercf14c2e2010-02-02 21:03:50 +0100808 cf = 0;
Andreas Gruenbacher6139f602011-05-06 20:00:02 +0200809 if (nc->discard_my_data)
810 cf |= CF_DISCARD_MY_DATA;
Andreas Gruenbacher6dff2902011-06-28 14:18:12 +0200811 if (nc->tentative)
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200812 cf |= CF_DRY_RUN;
Philipp Reisnercf14c2e2010-02-02 21:03:50 +0100813 p->conn_flags = cpu_to_be32(cf);
814
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200815 if (connection->agreed_pro_version >= 87)
Philipp Reisner44ed1672011-04-19 17:10:19 +0200816 strcpy(p->integrity_alg, nc->integrity_alg);
817 rcu_read_unlock();
Philipp Reisnerb411b362009-09-25 16:07:19 -0700818
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200819 return __conn_send_command(connection, sock, cmd, size, NULL, 0);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200820}
821
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200822int drbd_send_protocol(struct drbd_connection *connection)
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200823{
824 int err;
825
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +0200826 mutex_lock(&connection->data.mutex);
827 err = __drbd_send_protocol(connection, P_PROTOCOL);
828 mutex_unlock(&connection->data.mutex);
Andreas Gruenbachera7eb7bd2011-04-29 13:19:58 +0200829
830 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700831}
832
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200833static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700834{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200835 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200836 struct drbd_socket *sock;
837 struct p_uuids *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700838 int i;
839
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200840 if (!get_ldev_if_state(device, D_NEGOTIATING))
Andreas Gruenbacher2ae5f952011-03-16 01:07:20 +0100841 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700842
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200843 sock = &peer_device->connection->data;
844 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200845 if (!p) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200846 put_ldev(device);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200847 return -EIO;
848 }
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200849 spin_lock_irq(&device->ldev->md.uuid_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700850 for (i = UI_CURRENT; i < UI_SIZE; i++)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200851 p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
852 spin_unlock_irq(&device->ldev->md.uuid_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700853
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200854 device->comm_bm_set = drbd_bm_total_weight(device);
855 p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set);
Philipp Reisner44ed1672011-04-19 17:10:19 +0200856 rcu_read_lock();
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200857 uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0;
Philipp Reisner44ed1672011-04-19 17:10:19 +0200858 rcu_read_unlock();
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200859 uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0;
860 uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200861 p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700862
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200863 put_ldev(device);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200864 return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700865}
866
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200867int drbd_send_uuids(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700868{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200869 return _drbd_send_uuids(peer_device, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700870}
871
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200872int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700873{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200874 return _drbd_send_uuids(peer_device, 8);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700875}
876
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200877void drbd_print_uuids(struct drbd_device *device, const char *text)
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100878{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200879 if (get_ldev_if_state(device, D_NEGOTIATING)) {
880 u64 *uuid = device->ldev->md.uuid;
Andreas Gruenbacherd0180172011-07-03 17:53:52 +0200881 drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n",
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100882 text,
883 (unsigned long long)uuid[UI_CURRENT],
884 (unsigned long long)uuid[UI_BITMAP],
885 (unsigned long long)uuid[UI_HISTORY_START],
886 (unsigned long long)uuid[UI_HISTORY_END]);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200887 put_ldev(device);
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100888 } else {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +0200889 drbd_info(device, "%s effective data uuid: %016llX\n",
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100890 text,
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200891 (unsigned long long)device->ed_uuid);
Lars Ellenberg62b0da32011-01-20 13:25:21 +0100892 }
893}
894
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200895void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700896{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200897 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200898 struct drbd_socket *sock;
899 struct p_rs_uuid *p;
Lars Ellenberg5a22db82010-12-17 21:14:23 +0100900 u64 uuid;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700901
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +0200902 D_ASSERT(device, device->state.disk == D_UP_TO_DATE);
Lars Ellenberg5a22db82010-12-17 21:14:23 +0100903
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200904 uuid = device->ldev->md.uuid[UI_BITMAP];
Philipp Reisner5ba3dac2011-10-05 15:54:18 +0200905 if (uuid && uuid != UUID_JUST_CREATED)
906 uuid = uuid + UUID_NEW_BM_OFFSET;
907 else
908 get_random_bytes(&uuid, sizeof(u64));
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200909 drbd_uuid_set(device, UI_BITMAP, uuid);
910 drbd_print_uuids(device, "updated sync UUID");
911 drbd_md_sync(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700912
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200913 sock = &peer_device->connection->data;
914 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200915 if (p) {
916 p->uuid = cpu_to_be64(uuid);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200917 drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200918 }
Philipp Reisnerb411b362009-09-25 16:07:19 -0700919}
920
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200921int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700922{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200923 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200924 struct drbd_socket *sock;
925 struct p_sizes *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700926 sector_t d_size, u_size;
Lars Ellenbergdb141b22012-06-25 19:15:58 +0200927 int q_order_type;
928 unsigned int max_bio_size;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700929
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200930 if (get_ldev_if_state(device, D_NEGOTIATING)) {
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +0200931 D_ASSERT(device, device->ldev->backing_bdev);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200932 d_size = drbd_get_max_capacity(device->ldev);
Philipp Reisnerdaeda1c2011-05-03 15:00:55 +0200933 rcu_read_lock();
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200934 u_size = rcu_dereference(device->ldev->disk_conf)->disk_size;
Philipp Reisnerdaeda1c2011-05-03 15:00:55 +0200935 rcu_read_unlock();
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200936 q_order_type = drbd_queue_order_type(device);
937 max_bio_size = queue_max_hw_sectors(device->ldev->backing_bdev->bd_disk->queue) << 9;
Lars Ellenbergdb141b22012-06-25 19:15:58 +0200938 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200939 put_ldev(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700940 } else {
941 d_size = 0;
942 u_size = 0;
943 q_order_type = QUEUE_ORDERED_NONE;
Philipp Reisner99432fc2011-05-20 16:39:13 +0200944 max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700945 }
946
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200947 sock = &peer_device->connection->data;
948 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200949 if (!p)
950 return -EIO;
Philipp Reisner2ffca4f2011-06-30 15:43:06 +0200951
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200952 if (peer_device->connection->agreed_pro_version <= 94)
Lars Ellenbergdb141b22012-06-25 19:15:58 +0200953 max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200954 else if (peer_device->connection->agreed_pro_version < 100)
Philipp Reisner98683652012-11-09 14:18:43 +0100955 max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95);
Philipp Reisner68093842011-06-30 15:43:06 +0200956
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200957 p->d_size = cpu_to_be64(d_size);
958 p->u_size = cpu_to_be64(u_size);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +0200959 p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev));
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200960 p->max_bio_size = cpu_to_be32(max_bio_size);
961 p->queue_order_type = cpu_to_be16(q_order_type);
962 p->dds_flags = cpu_to_be16(flags);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200963 return drbd_send_command(peer_device, sock, P_SIZES, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700964}
965
966/**
Lars Ellenbergf479ea02011-10-27 16:52:30 +0200967 * drbd_send_current_state() - Sends the drbd state to the peer
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200968 * @peer_device: DRBD peer device.
Philipp Reisnerb411b362009-09-25 16:07:19 -0700969 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200970int drbd_send_current_state(struct drbd_peer_device *peer_device)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700971{
Andreas Gruenbacher7c967152011-03-22 00:49:36 +0100972 struct drbd_socket *sock;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200973 struct p_state *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700974
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200975 sock = &peer_device->connection->data;
976 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +0200977 if (!p)
978 return -EIO;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200979 p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */
980 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700981}
982
Lars Ellenbergf479ea02011-10-27 16:52:30 +0200983/**
984 * drbd_send_state() - After a state change, sends the new state to the peer
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200985 * @peer_device: DRBD peer device.
Philipp Reisner43de7c82011-11-10 13:16:13 +0100986 * @state: the state to send, not necessarily the current state.
Lars Ellenbergf479ea02011-10-27 16:52:30 +0200987 *
988 * Each state change queues an "after_state_ch" work, which will eventually
989 * send the resulting new state to the peer. If more state changes happen
990 * between queuing and processing of the after_state_ch work, we still
991 * want to send each intermediary state in the order it occurred.
992 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200993int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state)
Lars Ellenbergf479ea02011-10-27 16:52:30 +0200994{
Philipp Reisner43de7c82011-11-10 13:16:13 +0100995 struct drbd_socket *sock;
996 struct p_state *p;
Lars Ellenbergf479ea02011-10-27 16:52:30 +0200997
Andreas Gruenbacher69a22772011-08-09 00:47:13 +0200998 sock = &peer_device->connection->data;
999 p = drbd_prepare_command(peer_device, sock);
Philipp Reisner43de7c82011-11-10 13:16:13 +01001000 if (!p)
1001 return -EIO;
1002 p->state = cpu_to_be32(state.i); /* Within the send mutex */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001003 return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0);
Philipp Reisner43de7c82011-11-10 13:16:13 +01001004}
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001005
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001006int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001007{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001008 struct drbd_socket *sock;
1009 struct p_req_state *p;
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001010
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001011 sock = &peer_device->connection->data;
1012 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001013 if (!p)
1014 return -EIO;
1015 p->mask = cpu_to_be32(mask.i);
1016 p->val = cpu_to_be32(val.i);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001017 return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001018}
1019
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001020int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val)
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001021{
1022 enum drbd_packet cmd;
1023 struct drbd_socket *sock;
1024 struct p_req_state *p;
1025
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001026 cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ;
1027 sock = &connection->data;
1028 p = conn_prepare_command(connection, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001029 if (!p)
1030 return -EIO;
1031 p->mask = cpu_to_be32(mask.i);
1032 p->val = cpu_to_be32(val.i);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001033 return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001034}
1035
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001036void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001037{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001038 struct drbd_socket *sock;
1039 struct p_req_state_reply *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001040
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001041 sock = &peer_device->connection->meta;
1042 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001043 if (p) {
1044 p->retcode = cpu_to_be32(retcode);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001045 drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0);
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001046 }
Lars Ellenbergf479ea02011-10-27 16:52:30 +02001047}
1048
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001049void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001050{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001051 struct drbd_socket *sock;
1052 struct p_req_state_reply *p;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001053 enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001054
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001055 sock = &connection->meta;
1056 p = conn_prepare_command(connection, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001057 if (p) {
1058 p->retcode = cpu_to_be32(retcode);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001059 conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001060 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07001061}
1062
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001063static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001064{
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001065 BUG_ON(code & ~0xf);
1066 p->encoding = (p->encoding & ~0xf) | code;
1067}
Philipp Reisnerb411b362009-09-25 16:07:19 -07001068
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001069static void dcbp_set_start(struct p_compressed_bm *p, int set)
1070{
1071 p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0);
1072}
Philipp Reisnerb411b362009-09-25 16:07:19 -07001073
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001074static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n)
1075{
1076 BUG_ON(n & ~0x7);
1077 p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001078}
1079
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001080static int fill_bitmap_rle_bits(struct drbd_device *device,
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001081 struct p_compressed_bm *p,
1082 unsigned int size,
1083 struct bm_xfer_ctx *c)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001084{
1085 struct bitstream bs;
1086 unsigned long plain_bits;
1087 unsigned long tmp;
1088 unsigned long rl;
1089 unsigned len;
1090 unsigned toggle;
Philipp Reisner44ed1672011-04-19 17:10:19 +02001091 int bits, use_rle;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001092
1093 /* may we use this feature? */
Philipp Reisner44ed1672011-04-19 17:10:19 +02001094 rcu_read_lock();
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001095 use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle;
Philipp Reisner44ed1672011-04-19 17:10:19 +02001096 rcu_read_unlock();
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001097 if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90)
Philipp Reisner44ed1672011-04-19 17:10:19 +02001098 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001099
1100 if (c->bit_offset >= c->bm_bits)
1101 return 0; /* nothing to do. */
1102
1103 /* use at most thus many bytes */
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001104 bitstream_init(&bs, p->code, size, 0);
1105 memset(p->code, 0, size);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001106 /* plain bits covered in this code string */
1107 plain_bits = 0;
1108
1109 /* p->encoding & 0x80 stores whether the first run length is set.
1110 * bit offset is implicit.
1111 * start with toggle == 2 to be able to tell the first iteration */
1112 toggle = 2;
1113
1114 /* see how much plain bits we can stuff into one packet
1115 * using RLE and VLI. */
1116 do {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001117 tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset)
1118 : _drbd_bm_find_next(device, c->bit_offset);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001119 if (tmp == -1UL)
1120 tmp = c->bm_bits;
1121 rl = tmp - c->bit_offset;
1122
1123 if (toggle == 2) { /* first iteration */
1124 if (rl == 0) {
1125 /* the first checked bit was set,
1126 * store start value, */
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001127 dcbp_set_start(p, 1);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001128 /* but skip encoding of zero run length */
1129 toggle = !toggle;
1130 continue;
1131 }
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001132 dcbp_set_start(p, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001133 }
1134
1135 /* paranoia: catch zero runlength.
1136 * can only happen if bitmap is modified while we scan it. */
1137 if (rl == 0) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001138 drbd_err(device, "unexpected zero runlength while encoding bitmap "
Philipp Reisnerb411b362009-09-25 16:07:19 -07001139 "t:%u bo:%lu\n", toggle, c->bit_offset);
1140 return -1;
1141 }
1142
1143 bits = vli_encode_bits(&bs, rl);
1144 if (bits == -ENOBUFS) /* buffer full */
1145 break;
1146 if (bits <= 0) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001147 drbd_err(device, "error while encoding bitmap: %d\n", bits);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001148 return 0;
1149 }
1150
1151 toggle = !toggle;
1152 plain_bits += rl;
1153 c->bit_offset = tmp;
1154 } while (c->bit_offset < c->bm_bits);
1155
1156 len = bs.cur.b - p->code + !!bs.cur.bit;
1157
1158 if (plain_bits < (len << 3)) {
1159 /* incompressible with this method.
1160 * we need to rewind both word and bit position. */
1161 c->bit_offset -= plain_bits;
1162 bm_xfer_ctx_bit_to_word_offset(c);
1163 c->bit_offset = c->word_offset * BITS_PER_LONG;
1164 return 0;
1165 }
1166
1167 /* RLE + VLI was able to compress it just fine.
1168 * update c->word_offset. */
1169 bm_xfer_ctx_bit_to_word_offset(c);
1170
1171 /* store pad_bits */
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001172 dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001173
1174 return len;
1175}
1176
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001177/**
1178 * send_bitmap_rle_or_plain
1179 *
1180 * Return 0 when done, 1 when another iteration is needed, and a negative error
1181 * code upon failure.
1182 */
1183static int
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001184send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001185{
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001186 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
1187 unsigned int header_size = drbd_header_size(first_peer_device(device)->connection);
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001188 struct p_compressed_bm *p = sock->sbuf + header_size;
Andreas Gruenbachera982dd52010-12-10 00:45:25 +01001189 int len, err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001190
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001191 len = fill_bitmap_rle_bits(device, p,
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001192 DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001193 if (len < 0)
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001194 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001195
1196 if (len) {
Andreas Gruenbachera02d1242011-03-22 17:20:45 +01001197 dcbp_set_code(p, RLE_VLI_Bits);
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001198 err = __send_command(first_peer_device(device)->connection, device->vnr, sock,
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001199 P_COMPRESSED_BITMAP, sizeof(*p) + len,
1200 NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001201 c->packets[0]++;
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001202 c->bytes[0] += header_size + sizeof(*p) + len;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001203
1204 if (c->bit_offset >= c->bm_bits)
1205 len = 0; /* DONE */
1206 } else {
1207 /* was not compressible.
1208 * send a buffer full of plain text bits instead. */
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001209 unsigned int data_size;
1210 unsigned long num_words;
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001211 unsigned long *p = sock->sbuf + header_size;
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001212
1213 data_size = DRBD_SOCKET_BUFFER_SIZE - header_size;
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001214 num_words = min_t(size_t, data_size / sizeof(*p),
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001215 c->bm_words - c->word_offset);
Andreas Gruenbachere6589832011-03-30 12:54:42 +02001216 len = num_words * sizeof(*p);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001217 if (len)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001218 drbd_bm_get_lel(device, c->word_offset, num_words, p);
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001219 err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001220 c->word_offset += num_words;
1221 c->bit_offset = c->word_offset * BITS_PER_LONG;
1222
1223 c->packets[1]++;
Andreas Gruenbacher50d0b1a2011-03-30 11:53:51 +02001224 c->bytes[1] += header_size + len;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001225
1226 if (c->bit_offset > c->bm_bits)
1227 c->bit_offset = c->bm_bits;
1228 }
Andreas Gruenbachera982dd52010-12-10 00:45:25 +01001229 if (!err) {
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001230 if (len == 0) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001231 INFO_bm_xfer_stats(device, "send", c);
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001232 return 0;
1233 } else
1234 return 1;
1235 }
1236 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001237}
1238
1239/* See the comment at receive_bitmap() */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001240static int _drbd_send_bitmap(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001241{
1242 struct bm_xfer_ctx c;
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001243 int err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001244
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001245 if (!expect(device->bitmap))
Andreas Gruenbacher81e84652010-12-09 15:03:57 +01001246 return false;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001247
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001248 if (get_ldev(device)) {
1249 if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001250 drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n");
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001251 drbd_bm_set_all(device);
1252 if (drbd_bm_write(device)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001253 /* write_bm did fail! Leave full sync flag set in Meta P_DATA
1254 * but otherwise process as per normal - need to tell other
1255 * side that a full resync is required! */
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001256 drbd_err(device, "Failed to write bitmap to disk!\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07001257 } else {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001258 drbd_md_clear_flag(device, MDF_FULL_SYNC);
1259 drbd_md_sync(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001260 }
1261 }
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001262 put_ldev(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001263 }
1264
1265 c = (struct bm_xfer_ctx) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001266 .bm_bits = drbd_bm_bits(device),
1267 .bm_words = drbd_bm_words(device),
Philipp Reisnerb411b362009-09-25 16:07:19 -07001268 };
1269
1270 do {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001271 err = send_bitmap_rle_or_plain(device, &c);
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001272 } while (err > 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001273
Andreas Gruenbacherf70af112010-12-11 18:51:50 +01001274 return err == 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001275}
1276
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001277int drbd_send_bitmap(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001278{
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001279 struct drbd_socket *sock = &first_peer_device(device)->connection->data;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001280 int err = -1;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001281
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001282 mutex_lock(&sock->mutex);
1283 if (sock->socket)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001284 err = !_drbd_send_bitmap(device);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001285 mutex_unlock(&sock->mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001286 return err;
1287}
1288
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001289void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001290{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001291 struct drbd_socket *sock;
1292 struct p_barrier_ack *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001293
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001294 if (connection->cstate < C_WF_REPORT_PARAMS)
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001295 return;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001296
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001297 sock = &connection->meta;
1298 p = conn_prepare_command(connection, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001299 if (!p)
1300 return;
1301 p->barrier = barrier_nr;
1302 p->set_size = cpu_to_be32(set_size);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001303 conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001304}
1305
1306/**
1307 * _drbd_send_ack() - Sends an ack packet
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001308 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07001309 * @cmd: Packet command code.
1310 * @sector: sector, needs to be in big endian byte order
1311 * @blksize: size in byte, needs to be in big endian byte order
1312 * @block_id: Id, big endian byte order
1313 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001314static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbacherd8763022011-01-26 17:39:41 +01001315 u64 sector, u32 blksize, u64 block_id)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001316{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001317 struct drbd_socket *sock;
1318 struct p_block_ack *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001319
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001320 if (peer_device->device->state.conn < C_CONNECTED)
Andreas Gruenbachera8c32aa2011-03-16 01:27:22 +01001321 return -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001322
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001323 sock = &peer_device->connection->meta;
1324 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001325 if (!p)
1326 return -EIO;
1327 p->sector = sector;
1328 p->block_id = block_id;
1329 p->blksize = blksize;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001330 p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq));
1331 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001332}
1333
Lars Ellenberg2b2bf212010-10-06 11:46:55 +02001334/* dp->sector and dp->block_id already/still in network byte order,
1335 * data_size is payload size according to dp->head,
1336 * and may need to be corrected for digest size. */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001337void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbachera9a99942011-03-16 01:30:14 +01001338 struct p_data *dp, int data_size)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001339{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001340 if (peer_device->connection->peer_integrity_tfm)
1341 data_size -= crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm);
1342 _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size),
Andreas Gruenbachera9a99942011-03-16 01:30:14 +01001343 dp->block_id);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001344}
1345
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001346void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbachera9a99942011-03-16 01:30:14 +01001347 struct p_block_req *rp)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001348{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001349 _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001350}
1351
1352/**
1353 * drbd_send_ack() - Sends an ack packet
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001354 * @device: DRBD device
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001355 * @cmd: packet command code
1356 * @peer_req: peer request
Philipp Reisnerb411b362009-09-25 16:07:19 -07001357 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001358int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001359 struct drbd_peer_request *peer_req)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001360{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001361 return _drbd_send_ack(peer_device, cmd,
Andreas Gruenbacherdd516122011-03-16 15:39:08 +01001362 cpu_to_be64(peer_req->i.sector),
1363 cpu_to_be32(peer_req->i.size),
1364 peer_req->block_id);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001365}
1366
1367/* This function misuses the block_id field to signal if the blocks
1368 * are is sync or not. */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001369int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001370 sector_t sector, int blksize, u64 block_id)
1371{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001372 return _drbd_send_ack(peer_device, cmd,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001373 cpu_to_be64(sector),
1374 cpu_to_be32(blksize),
1375 cpu_to_be64(block_id));
1376}
1377
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001378int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001379 sector_t sector, int size, u64 block_id)
1380{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001381 struct drbd_socket *sock;
1382 struct p_block_req *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001383
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001384 sock = &peer_device->connection->data;
1385 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001386 if (!p)
1387 return -EIO;
1388 p->sector = cpu_to_be64(sector);
1389 p->block_id = block_id;
1390 p->blksize = cpu_to_be32(size);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001391 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001392}
1393
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001394int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size,
Andreas Gruenbacherd8763022011-01-26 17:39:41 +01001395 void *digest, int digest_size, enum drbd_packet cmd)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001396{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001397 struct drbd_socket *sock;
1398 struct p_block_req *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001399
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001400 /* FIXME: Put the digest into the preallocated socket buffer. */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001401
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001402 sock = &peer_device->connection->data;
1403 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001404 if (!p)
1405 return -EIO;
1406 p->sector = cpu_to_be64(sector);
1407 p->block_id = ID_SYNCER /* unused */;
1408 p->blksize = cpu_to_be32(size);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001409 return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001410}
1411
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001412int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001413{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001414 struct drbd_socket *sock;
1415 struct p_block_req *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001416
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001417 sock = &peer_device->connection->data;
1418 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001419 if (!p)
1420 return -EIO;
1421 p->sector = cpu_to_be64(sector);
1422 p->block_id = ID_SYNCER /* unused */;
1423 p->blksize = cpu_to_be32(size);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001424 return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001425}
1426
1427/* called on sndtimeo
Andreas Gruenbacher81e84652010-12-09 15:03:57 +01001428 * returns false if we should retry,
1429 * true if we think connection is dead
Philipp Reisnerb411b362009-09-25 16:07:19 -07001430 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001431static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001432{
1433 int drop_it;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001434 /* long elapsed = (long)(jiffies - device->last_received); */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001435
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001436 drop_it = connection->meta.socket == sock
1437 || !connection->asender.task
1438 || get_t_state(&connection->asender) != RUNNING
1439 || connection->cstate < C_WF_REPORT_PARAMS;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001440
1441 if (drop_it)
Andreas Gruenbacher81e84652010-12-09 15:03:57 +01001442 return true;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001443
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001444 drop_it = !--connection->ko_count;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001445 if (!drop_it) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +02001446 drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n",
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001447 current->comm, current->pid, connection->ko_count);
1448 request_ping(connection);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001449 }
1450
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001451 return drop_it; /* && (device->state == R_PRIMARY) */;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001452}
1453
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001454static void drbd_update_congested(struct drbd_connection *connection)
Andreas Gruenbacher9e204cd2011-01-26 18:45:11 +01001455{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001456 struct sock *sk = connection->data.socket->sk;
Andreas Gruenbacher9e204cd2011-01-26 18:45:11 +01001457 if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5)
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001458 set_bit(NET_CONGESTED, &connection->flags);
Andreas Gruenbacher9e204cd2011-01-26 18:45:11 +01001459}
1460
Philipp Reisnerb411b362009-09-25 16:07:19 -07001461/* The idea of sendpage seems to be to put some kind of reference
1462 * to the page into the skb, and to hand it over to the NIC. In
1463 * this process get_page() gets called.
1464 *
1465 * As soon as the page was really sent over the network put_page()
1466 * gets called by some part of the network layer. [ NIC driver? ]
1467 *
1468 * [ get_page() / put_page() increment/decrement the count. If count
1469 * reaches 0 the page will be freed. ]
1470 *
1471 * This works nicely with pages from FSs.
1472 * But this means that in protocol A we might signal IO completion too early!
1473 *
1474 * In order not to corrupt data during a resync we must make sure
1475 * that we do not reuse our own buffer pages (EEs) to early, therefore
1476 * we have the net_ee list.
1477 *
1478 * XFS seems to have problems, still, it submits pages with page_count == 0!
1479 * As a workaround, we disable sendpage on pages
1480 * with page_count == 0 or PageSlab.
1481 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001482static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page,
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001483 int offset, size_t size, unsigned msg_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001484{
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001485 struct socket *socket;
1486 void *addr;
1487 int err;
1488
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001489 socket = peer_device->connection->data.socket;
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001490 addr = kmap(page) + offset;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001491 err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001492 kunmap(page);
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001493 if (!err)
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001494 peer_device->device->send_cnt += size >> 9;
Andreas Gruenbacherb9874272011-03-16 09:41:10 +01001495 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001496}
1497
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001498static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page,
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001499 int offset, size_t size, unsigned msg_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001500{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001501 struct socket *socket = peer_device->connection->data.socket;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001502 mm_segment_t oldfs = get_fs();
Philipp Reisnerb411b362009-09-25 16:07:19 -07001503 int len = size;
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001504 int err = -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001505
1506 /* e.g. XFS meta- & log-data is in slab pages, which have a
1507 * page_count of 0 and/or have PageSlab() set.
1508 * we cannot use send_page for those, as that does get_page();
1509 * put_page(); and would cause either a VM_BUG directly, or
1510 * __page_cache_release a page that would actually still be referenced
1511 * by someone, leading to some obscure delayed Oops somewhere else. */
1512 if (disable_sendpage || (page_count(page) < 1) || PageSlab(page))
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001513 return _drbd_no_send_page(peer_device, page, offset, size, msg_flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001514
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001515 msg_flags |= MSG_NOSIGNAL;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001516 drbd_update_congested(peer_device->connection);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001517 set_fs(KERNEL_DS);
1518 do {
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001519 int sent;
1520
1521 sent = socket->ops->sendpage(socket, page, offset, len, msg_flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001522 if (sent <= 0) {
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001523 if (sent == -EAGAIN) {
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001524 if (we_should_drop_the_connection(peer_device->connection, socket))
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001525 break;
1526 continue;
1527 }
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001528 drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07001529 __func__, (int)size, len, sent);
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001530 if (sent < 0)
1531 err = sent;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001532 break;
1533 }
1534 len -= sent;
1535 offset += sent;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001536 } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001537 set_fs(oldfs);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001538 clear_bit(NET_CONGESTED, &peer_device->connection->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001539
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001540 if (len == 0) {
1541 err = 0;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001542 peer_device->device->send_cnt += size >> 9;
Andreas Gruenbacher88b390f2011-03-16 10:44:16 +01001543 }
1544 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001545}
1546
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001547static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001548{
Kent Overstreet79886132013-11-23 17:19:00 -08001549 struct bio_vec bvec;
1550 struct bvec_iter iter;
1551
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001552 /* hint all but last page with MSG_MORE */
Kent Overstreet79886132013-11-23 17:19:00 -08001553 bio_for_each_segment(bvec, bio, iter) {
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001554 int err;
1555
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001556 err = _drbd_no_send_page(peer_device, bvec.bv_page,
Kent Overstreet79886132013-11-23 17:19:00 -08001557 bvec.bv_offset, bvec.bv_len,
Kent Overstreet4550dd62013-08-07 14:26:21 -07001558 bio_iter_last(bvec, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08001559 ? 0 : MSG_MORE);
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001560 if (err)
1561 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001562 }
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001563 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001564}
1565
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001566static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001567{
Kent Overstreet79886132013-11-23 17:19:00 -08001568 struct bio_vec bvec;
1569 struct bvec_iter iter;
1570
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001571 /* hint all but last page with MSG_MORE */
Kent Overstreet79886132013-11-23 17:19:00 -08001572 bio_for_each_segment(bvec, bio, iter) {
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001573 int err;
1574
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001575 err = _drbd_send_page(peer_device, bvec.bv_page,
Kent Overstreet79886132013-11-23 17:19:00 -08001576 bvec.bv_offset, bvec.bv_len,
Kent Overstreet4550dd62013-08-07 14:26:21 -07001577 bio_iter_last(bvec, iter) ? 0 : MSG_MORE);
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001578 if (err)
1579 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001580 }
Andreas Gruenbacher7fae55d2011-03-16 11:46:33 +01001581 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001582}
1583
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001584static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device,
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001585 struct drbd_peer_request *peer_req)
Lars Ellenberg45bb9122010-05-14 17:10:48 +02001586{
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001587 struct page *page = peer_req->pages;
1588 unsigned len = peer_req->i.size;
Andreas Gruenbacher9f692302011-03-16 10:49:09 +01001589 int err;
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001590
Lars Ellenbergba11ad92010-05-25 16:26:16 +02001591 /* hint all but last page with MSG_MORE */
Lars Ellenberg45bb9122010-05-14 17:10:48 +02001592 page_chain_for_each(page) {
1593 unsigned l = min_t(unsigned, len, PAGE_SIZE);
Andreas Gruenbacher9f692302011-03-16 10:49:09 +01001594
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001595 err = _drbd_send_page(peer_device, page, 0, l,
Andreas Gruenbacher9f692302011-03-16 10:49:09 +01001596 page_chain_next(page) ? MSG_MORE : 0);
1597 if (err)
1598 return err;
Lars Ellenberg45bb9122010-05-14 17:10:48 +02001599 len -= l;
1600 }
Andreas Gruenbacher9f692302011-03-16 10:49:09 +01001601 return 0;
Lars Ellenberg45bb9122010-05-14 17:10:48 +02001602}
1603
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001604static u32 bio_flags_to_wire(struct drbd_connection *connection, unsigned long bi_rw)
Philipp Reisner76d2e7e2010-08-25 11:58:05 +02001605{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001606 if (connection->agreed_pro_version >= 95)
Philipp Reisner76d2e7e2010-08-25 11:58:05 +02001607 return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) |
Philipp Reisner76d2e7e2010-08-25 11:58:05 +02001608 (bi_rw & REQ_FUA ? DP_FUA : 0) |
1609 (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) |
1610 (bi_rw & REQ_DISCARD ? DP_DISCARD : 0);
1611 else
Jens Axboe721a9602011-03-09 11:56:30 +01001612 return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0;
Philipp Reisner76d2e7e2010-08-25 11:58:05 +02001613}
1614
Lars Ellenberg2f632ae2014-04-28 18:43:24 +02001615/* Used to send write or TRIM aka REQ_DISCARD requests
1616 * R_PRIMARY -> Peer (P_DATA, P_TRIM)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001617 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001618int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001619{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001620 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001621 struct drbd_socket *sock;
1622 struct p_data *p;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001623 unsigned int dp_flags = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001624 int dgs;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001625 int err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001626
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001627 sock = &peer_device->connection->data;
1628 p = drbd_prepare_command(peer_device, sock);
1629 dgs = peer_device->connection->integrity_tfm ?
1630 crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001631
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001632 if (!p)
1633 return -EIO;
1634 p->sector = cpu_to_be64(req->i.sector);
1635 p->block_id = (unsigned long)req;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001636 p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq));
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001637 dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio->bi_rw);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001638 if (device->state.conn >= C_SYNC_SOURCE &&
1639 device->state.conn <= C_PAUSED_SYNC_T)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001640 dp_flags |= DP_MAY_SET_IN_SYNC;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001641 if (peer_device->connection->agreed_pro_version >= 100) {
Philipp Reisner303d1442011-04-13 16:24:47 -07001642 if (req->rq_state & RQ_EXP_RECEIVE_ACK)
1643 dp_flags |= DP_SEND_RECEIVE_ACK;
Lars Ellenberg08d0dab2014-03-20 11:19:22 +01001644 /* During resync, request an explicit write ack,
1645 * even in protocol != C */
1646 if (req->rq_state & RQ_EXP_WRITE_ACK
1647 || (dp_flags & DP_MAY_SET_IN_SYNC))
Philipp Reisner303d1442011-04-13 16:24:47 -07001648 dp_flags |= DP_SEND_WRITE_ACK;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001649 }
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001650 p->dp_flags = cpu_to_be32(dp_flags);
Lars Ellenberg2f632ae2014-04-28 18:43:24 +02001651
1652 if (dp_flags & DP_DISCARD) {
1653 struct p_trim *t = (struct p_trim*)p;
1654 t->size = cpu_to_be32(req->i.size);
1655 err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
1656 goto out;
1657 }
1658
1659 /* our digest is still only over the payload.
1660 * TRIM does not carry any payload. */
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001661 if (dgs)
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001662 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, p + 1);
1663 err = __send_command(peer_device->connection, device->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size);
Andreas Gruenbacher6bdb9b02011-03-16 11:52:58 +01001664 if (!err) {
Lars Ellenberg470be442010-11-10 10:36:52 +01001665 /* For protocol A, we have to memcpy the payload into
1666 * socket buffers, as we may complete right away
1667 * as soon as we handed it over to tcp, at which point the data
1668 * pages may become invalid.
1669 *
1670 * For data-integrity enabled, we copy it as well, so we can be
1671 * sure that even if the bio pages may still be modified, it
1672 * won't change the data on the wire, thus if the digest checks
1673 * out ok after sending on this side, but does not fit on the
1674 * receiving side, we sure have detected corruption elsewhere.
1675 */
Philipp Reisner303d1442011-04-13 16:24:47 -07001676 if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs)
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001677 err = _drbd_send_bio(peer_device, req->master_bio);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001678 else
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001679 err = _drbd_send_zc_bio(peer_device, req->master_bio);
Lars Ellenberg470be442010-11-10 10:36:52 +01001680
1681 /* double check digest, sometimes buffers have been modified in flight. */
1682 if (dgs > 0 && dgs <= 64) {
Bart Van Assche24c48302011-05-21 18:32:29 +02001683 /* 64 byte, 512 bit, is the largest digest size
Lars Ellenberg470be442010-11-10 10:36:52 +01001684 * currently supported in kernel crypto. */
1685 unsigned char digest[64];
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001686 drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001687 if (memcmp(p + 1, digest, dgs)) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001688 drbd_warn(device,
Lars Ellenberg470be442010-11-10 10:36:52 +01001689 "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
Andreas Gruenbacherace652a2011-01-03 17:09:58 +01001690 (unsigned long long)req->i.sector, req->i.size);
Lars Ellenberg470be442010-11-10 10:36:52 +01001691 }
1692 } /* else if (dgs > 64) {
1693 ... Be noisy about digest too large ...
1694 } */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001695 }
Lars Ellenberg2f632ae2014-04-28 18:43:24 +02001696out:
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001697 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001698
Andreas Gruenbacher6bdb9b02011-03-16 11:52:58 +01001699 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001700}
1701
1702/* answer packet, used to send data back for read requests:
1703 * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY)
1704 * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY)
1705 */
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001706int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
Andreas Gruenbacherdb830c42011-02-04 15:57:48 +01001707 struct drbd_peer_request *peer_req)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001708{
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001709 struct drbd_device *device = peer_device->device;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001710 struct drbd_socket *sock;
1711 struct p_data *p;
Andreas Gruenbacher7b57b89d2011-03-16 11:35:20 +01001712 int err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001713 int dgs;
1714
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001715 sock = &peer_device->connection->data;
1716 p = drbd_prepare_command(peer_device, sock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001717
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001718 dgs = peer_device->connection->integrity_tfm ?
1719 crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001720
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001721 if (!p)
1722 return -EIO;
1723 p->sector = cpu_to_be64(peer_req->i.sector);
1724 p->block_id = peer_req->block_id;
1725 p->seq_num = 0; /* unused */
Lars Ellenbergb17f33c2012-02-08 15:32:51 +01001726 p->dp_flags = 0;
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001727 if (dgs)
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001728 drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
1729 err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size);
Andreas Gruenbacher7b57b89d2011-03-16 11:35:20 +01001730 if (!err)
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001731 err = _drbd_send_zc_ee(peer_device, peer_req);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001732 mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001733
Andreas Gruenbacher7b57b89d2011-03-16 11:35:20 +01001734 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001735}
1736
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001737int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req)
Philipp Reisner73a01a12010-10-27 14:33:00 +02001738{
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001739 struct drbd_socket *sock;
1740 struct p_block_desc *p;
Philipp Reisner73a01a12010-10-27 14:33:00 +02001741
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001742 sock = &peer_device->connection->data;
1743 p = drbd_prepare_command(peer_device, sock);
Andreas Gruenbacher9f5bdc32011-03-28 14:23:08 +02001744 if (!p)
1745 return -EIO;
1746 p->sector = cpu_to_be64(req->i.sector);
1747 p->blksize = cpu_to_be32(req->i.size);
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02001748 return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0);
Philipp Reisner73a01a12010-10-27 14:33:00 +02001749}
1750
Philipp Reisnerb411b362009-09-25 16:07:19 -07001751/*
1752 drbd_send distinguishes two cases:
1753
1754 Packets sent via the data socket "sock"
1755 and packets sent via the meta data socket "msock"
1756
1757 sock msock
1758 -----------------+-------------------------+------------------------------
1759 timeout conf.timeout / 2 conf.timeout / 2
1760 timeout action send a ping via msock Abort communication
1761 and close all sockets
1762*/
1763
1764/*
1765 * you must have down()ed the appropriate [m]sock_mutex elsewhere!
1766 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001767int drbd_send(struct drbd_connection *connection, struct socket *sock,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001768 void *buf, size_t size, unsigned msg_flags)
1769{
1770 struct kvec iov;
1771 struct msghdr msg;
1772 int rv, sent = 0;
1773
1774 if (!sock)
Andreas Gruenbacherc0d42c82010-12-09 23:52:22 +01001775 return -EBADR;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001776
1777 /* THINK if (signal_pending) return ... ? */
1778
1779 iov.iov_base = buf;
1780 iov.iov_len = size;
1781
1782 msg.msg_name = NULL;
1783 msg.msg_namelen = 0;
1784 msg.msg_control = NULL;
1785 msg.msg_controllen = 0;
1786 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
1787
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001788 if (sock == connection->data.socket) {
Philipp Reisner44ed1672011-04-19 17:10:19 +02001789 rcu_read_lock();
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001790 connection->ko_count = rcu_dereference(connection->net_conf)->ko_count;
Philipp Reisner44ed1672011-04-19 17:10:19 +02001791 rcu_read_unlock();
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001792 drbd_update_congested(connection);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001793 }
1794 do {
1795 /* STRANGE
1796 * tcp_sendmsg does _not_ use its size parameter at all ?
1797 *
1798 * -EAGAIN on timeout, -EINTR on signal.
1799 */
1800/* THINK
1801 * do we need to block DRBD_SIG if sock == &meta.socket ??
1802 * otherwise wake_asender() might interrupt some send_*Ack !
1803 */
1804 rv = kernel_sendmsg(sock, &msg, &iov, 1, size);
1805 if (rv == -EAGAIN) {
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001806 if (we_should_drop_the_connection(connection, sock))
Philipp Reisnerb411b362009-09-25 16:07:19 -07001807 break;
1808 else
1809 continue;
1810 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07001811 if (rv == -EINTR) {
1812 flush_signals(current);
1813 rv = 0;
1814 }
1815 if (rv < 0)
1816 break;
1817 sent += rv;
1818 iov.iov_base += rv;
1819 iov.iov_len -= rv;
1820 } while (sent < size);
1821
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001822 if (sock == connection->data.socket)
1823 clear_bit(NET_CONGESTED, &connection->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001824
1825 if (rv <= 0) {
1826 if (rv != -EAGAIN) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +02001827 drbd_err(connection, "%s_sendmsg returned %d\n",
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001828 sock == connection->meta.socket ? "msock" : "sock",
Philipp Reisnerbedbd2a2011-02-07 15:08:48 +01001829 rv);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001830 conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001831 } else
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001832 conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001833 }
1834
1835 return sent;
1836}
1837
Andreas Gruenbacherfb708e42010-12-15 17:04:36 +01001838/**
1839 * drbd_send_all - Send an entire buffer
1840 *
1841 * Returns 0 upon success and a negative error value otherwise.
1842 */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001843int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer,
Andreas Gruenbacherfb708e42010-12-15 17:04:36 +01001844 size_t size, unsigned msg_flags)
1845{
1846 int err;
1847
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02001848 err = drbd_send(connection, sock, buffer, size, msg_flags);
Andreas Gruenbacherfb708e42010-12-15 17:04:36 +01001849 if (err < 0)
1850 return err;
1851 if (err != size)
1852 return -EIO;
1853 return 0;
1854}
1855
Philipp Reisnerb411b362009-09-25 16:07:19 -07001856static int drbd_open(struct block_device *bdev, fmode_t mode)
1857{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001858 struct drbd_device *device = bdev->bd_disk->private_data;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001859 unsigned long flags;
1860 int rv = 0;
1861
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001862 mutex_lock(&drbd_main_mutex);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02001863 spin_lock_irqsave(&device->resource->req_lock, flags);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001864 /* to have a stable device->state.role
Philipp Reisnerb411b362009-09-25 16:07:19 -07001865 * and no race with updating open_cnt */
1866
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001867 if (device->state.role != R_PRIMARY) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001868 if (mode & FMODE_WRITE)
1869 rv = -EROFS;
1870 else if (!allow_oos)
1871 rv = -EMEDIUMTYPE;
1872 }
1873
1874 if (!rv)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001875 device->open_cnt++;
Andreas Gruenbacher05008132011-07-07 14:19:42 +02001876 spin_unlock_irqrestore(&device->resource->req_lock, flags);
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001877 mutex_unlock(&drbd_main_mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001878
1879 return rv;
1880}
1881
Al Virodb2a1442013-05-05 21:52:57 -04001882static void drbd_release(struct gendisk *gd, fmode_t mode)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001883{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001884 struct drbd_device *device = gd->private_data;
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001885 mutex_lock(&drbd_main_mutex);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001886 device->open_cnt--;
Arnd Bergmann2a48fc02010-06-02 14:28:52 +02001887 mutex_unlock(&drbd_main_mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001888}
1889
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001890static void drbd_set_defaults(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001891{
Lars Ellenbergf3990022011-03-23 14:31:09 +01001892 /* Beware! The actual layout differs
1893 * between big endian and little endian */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001894 device->state = (union drbd_dev_state) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001895 { .role = R_SECONDARY,
1896 .peer = R_UNKNOWN,
1897 .conn = C_STANDALONE,
1898 .disk = D_DISKLESS,
1899 .pdsk = D_UNKNOWN,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001900 } };
1901}
1902
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001903void drbd_init_set_defaults(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001904{
1905 /* the memset(,0,) did most of this.
1906 * note: only assignments, no allocation in here */
1907
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001908 drbd_set_defaults(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001909
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001910 atomic_set(&device->ap_bio_cnt, 0);
1911 atomic_set(&device->ap_pending_cnt, 0);
1912 atomic_set(&device->rs_pending_cnt, 0);
1913 atomic_set(&device->unacked_cnt, 0);
1914 atomic_set(&device->local_cnt, 0);
1915 atomic_set(&device->pp_in_use_by_net, 0);
1916 atomic_set(&device->rs_sect_in, 0);
1917 atomic_set(&device->rs_sect_ev, 0);
1918 atomic_set(&device->ap_in_flight, 0);
1919 atomic_set(&device->md_io_in_use, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001920
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001921 mutex_init(&device->own_state_mutex);
1922 device->state_mutex = &device->own_state_mutex;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001923
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001924 spin_lock_init(&device->al_lock);
1925 spin_lock_init(&device->peer_seq_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001926
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001927 INIT_LIST_HEAD(&device->active_ee);
1928 INIT_LIST_HEAD(&device->sync_ee);
1929 INIT_LIST_HEAD(&device->done_ee);
1930 INIT_LIST_HEAD(&device->read_ee);
1931 INIT_LIST_HEAD(&device->net_ee);
1932 INIT_LIST_HEAD(&device->resync_reads);
1933 INIT_LIST_HEAD(&device->resync_work.list);
1934 INIT_LIST_HEAD(&device->unplug_work.list);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001935 INIT_LIST_HEAD(&device->bm_io_work.w.list);
Philipp Reisner0ced55a2010-04-30 15:26:20 +02001936
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001937 device->resync_work.cb = w_resync_timer;
1938 device->unplug_work.cb = w_send_write_hint;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001939 device->bm_io_work.w.cb = w_bitmap_io;
Philipp Reisnera21e9292011-02-08 15:08:49 +01001940
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001941 init_timer(&device->resync_timer);
1942 init_timer(&device->md_sync_timer);
1943 init_timer(&device->start_resync_timer);
1944 init_timer(&device->request_timer);
1945 device->resync_timer.function = resync_timer_fn;
1946 device->resync_timer.data = (unsigned long) device;
1947 device->md_sync_timer.function = md_sync_timer_fn;
1948 device->md_sync_timer.data = (unsigned long) device;
1949 device->start_resync_timer.function = start_resync_timer_fn;
1950 device->start_resync_timer.data = (unsigned long) device;
1951 device->request_timer.function = request_timer_fn;
1952 device->request_timer.data = (unsigned long) device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001953
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001954 init_waitqueue_head(&device->misc_wait);
1955 init_waitqueue_head(&device->state_wait);
1956 init_waitqueue_head(&device->ee_wait);
1957 init_waitqueue_head(&device->al_wait);
1958 init_waitqueue_head(&device->seq_wait);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001959
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001960 device->resync_wenr = LC_FREE;
1961 device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
1962 device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001963}
1964
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001965void drbd_device_cleanup(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001966{
Lars Ellenberg1d7734a2010-08-11 21:21:50 +02001967 int i;
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001968 if (first_peer_device(device)->connection->receiver.t_state != NONE)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02001969 drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n",
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02001970 first_peer_device(device)->connection->receiver.t_state);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001971
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001972 device->al_writ_cnt =
1973 device->bm_writ_cnt =
1974 device->read_cnt =
1975 device->recv_cnt =
1976 device->send_cnt =
1977 device->writ_cnt =
1978 device->p_size =
1979 device->rs_start =
1980 device->rs_total =
1981 device->rs_failed = 0;
1982 device->rs_last_events = 0;
1983 device->rs_last_sect_ev = 0;
Lars Ellenberg1d7734a2010-08-11 21:21:50 +02001984 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001985 device->rs_mark_left[i] = 0;
1986 device->rs_mark_time[i] = 0;
Lars Ellenberg1d7734a2010-08-11 21:21:50 +02001987 }
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02001988 D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001989
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001990 drbd_set_my_capacity(device, 0);
1991 if (device->bitmap) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001992 /* maybe never allocated. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001993 drbd_bm_resize(device, 0, 1);
1994 drbd_bm_cleanup(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001995 }
1996
Philipp Reisner28995af2013-11-22 16:48:14 +01001997 drbd_free_ldev(device->ldev);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02001998 device->ldev = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001999
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002000 clear_bit(AL_SUSPENDED, &device->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002001
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02002002 D_ASSERT(device, list_empty(&device->active_ee));
2003 D_ASSERT(device, list_empty(&device->sync_ee));
2004 D_ASSERT(device, list_empty(&device->done_ee));
2005 D_ASSERT(device, list_empty(&device->read_ee));
2006 D_ASSERT(device, list_empty(&device->net_ee));
2007 D_ASSERT(device, list_empty(&device->resync_reads));
2008 D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q));
2009 D_ASSERT(device, list_empty(&device->resync_work.list));
2010 D_ASSERT(device, list_empty(&device->unplug_work.list));
Lars Ellenberg2265b472010-12-16 15:41:26 +01002011
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002012 drbd_set_defaults(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002013}
2014
2015
2016static void drbd_destroy_mempools(void)
2017{
2018 struct page *page;
2019
2020 while (drbd_pp_pool) {
2021 page = drbd_pp_pool;
2022 drbd_pp_pool = (struct page *)page_private(page);
2023 __free_page(page);
2024 drbd_pp_vacant--;
2025 }
2026
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02002027 /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */
Philipp Reisnerb411b362009-09-25 16:07:19 -07002028
Lars Ellenberg9476f392011-02-23 17:02:01 +01002029 if (drbd_md_io_bio_set)
2030 bioset_free(drbd_md_io_bio_set);
Lars Ellenberg42818082011-02-23 12:39:46 +01002031 if (drbd_md_io_page_pool)
2032 mempool_destroy(drbd_md_io_page_pool);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002033 if (drbd_ee_mempool)
2034 mempool_destroy(drbd_ee_mempool);
2035 if (drbd_request_mempool)
2036 mempool_destroy(drbd_request_mempool);
2037 if (drbd_ee_cache)
2038 kmem_cache_destroy(drbd_ee_cache);
2039 if (drbd_request_cache)
2040 kmem_cache_destroy(drbd_request_cache);
2041 if (drbd_bm_ext_cache)
2042 kmem_cache_destroy(drbd_bm_ext_cache);
2043 if (drbd_al_ext_cache)
2044 kmem_cache_destroy(drbd_al_ext_cache);
2045
Lars Ellenberg9476f392011-02-23 17:02:01 +01002046 drbd_md_io_bio_set = NULL;
Lars Ellenberg42818082011-02-23 12:39:46 +01002047 drbd_md_io_page_pool = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002048 drbd_ee_mempool = NULL;
2049 drbd_request_mempool = NULL;
2050 drbd_ee_cache = NULL;
2051 drbd_request_cache = NULL;
2052 drbd_bm_ext_cache = NULL;
2053 drbd_al_ext_cache = NULL;
2054
2055 return;
2056}
2057
2058static int drbd_create_mempools(void)
2059{
2060 struct page *page;
Lars Ellenberg1816a2b2010-11-11 15:19:07 +01002061 const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002062 int i;
2063
2064 /* prepare our caches and mempools */
2065 drbd_request_mempool = NULL;
2066 drbd_ee_cache = NULL;
2067 drbd_request_cache = NULL;
2068 drbd_bm_ext_cache = NULL;
2069 drbd_al_ext_cache = NULL;
2070 drbd_pp_pool = NULL;
Lars Ellenberg42818082011-02-23 12:39:46 +01002071 drbd_md_io_page_pool = NULL;
Lars Ellenberg9476f392011-02-23 17:02:01 +01002072 drbd_md_io_bio_set = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002073
2074 /* caches */
2075 drbd_request_cache = kmem_cache_create(
2076 "drbd_req", sizeof(struct drbd_request), 0, 0, NULL);
2077 if (drbd_request_cache == NULL)
2078 goto Enomem;
2079
2080 drbd_ee_cache = kmem_cache_create(
Andreas Gruenbacherf6ffca92011-02-04 15:30:34 +01002081 "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002082 if (drbd_ee_cache == NULL)
2083 goto Enomem;
2084
2085 drbd_bm_ext_cache = kmem_cache_create(
2086 "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL);
2087 if (drbd_bm_ext_cache == NULL)
2088 goto Enomem;
2089
2090 drbd_al_ext_cache = kmem_cache_create(
2091 "drbd_al", sizeof(struct lc_element), 0, 0, NULL);
2092 if (drbd_al_ext_cache == NULL)
2093 goto Enomem;
2094
2095 /* mempools */
Lars Ellenberg9476f392011-02-23 17:02:01 +01002096 drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0);
2097 if (drbd_md_io_bio_set == NULL)
2098 goto Enomem;
Lars Ellenberg9476f392011-02-23 17:02:01 +01002099
Lars Ellenberg42818082011-02-23 12:39:46 +01002100 drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0);
2101 if (drbd_md_io_page_pool == NULL)
2102 goto Enomem;
2103
Philipp Reisnerb411b362009-09-25 16:07:19 -07002104 drbd_request_mempool = mempool_create(number,
2105 mempool_alloc_slab, mempool_free_slab, drbd_request_cache);
2106 if (drbd_request_mempool == NULL)
2107 goto Enomem;
2108
2109 drbd_ee_mempool = mempool_create(number,
2110 mempool_alloc_slab, mempool_free_slab, drbd_ee_cache);
Nicolas Kaiser2027ae12010-10-28 06:15:26 -06002111 if (drbd_ee_mempool == NULL)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002112 goto Enomem;
2113
2114 /* drbd's page pool */
2115 spin_lock_init(&drbd_pp_lock);
2116
2117 for (i = 0; i < number; i++) {
2118 page = alloc_page(GFP_HIGHUSER);
2119 if (!page)
2120 goto Enomem;
2121 set_page_private(page, (unsigned long)drbd_pp_pool);
2122 drbd_pp_pool = page;
2123 }
2124 drbd_pp_vacant = number;
2125
2126 return 0;
2127
2128Enomem:
2129 drbd_destroy_mempools(); /* in case we allocated some */
2130 return -ENOMEM;
2131}
2132
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002133static void drbd_release_all_peer_reqs(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002134{
2135 int rr;
2136
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002137 rr = drbd_free_peer_reqs(device, &device->active_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002138 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002139 drbd_err(device, "%d EEs in active list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002140
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002141 rr = drbd_free_peer_reqs(device, &device->sync_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002142 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002143 drbd_err(device, "%d EEs in sync list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002144
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002145 rr = drbd_free_peer_reqs(device, &device->read_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002146 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002147 drbd_err(device, "%d EEs in read list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002148
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002149 rr = drbd_free_peer_reqs(device, &device->done_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002150 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002151 drbd_err(device, "%d EEs in done list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002152
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002153 rr = drbd_free_peer_reqs(device, &device->net_ee);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002154 if (rr)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002155 drbd_err(device, "%d EEs in net list found!\n", rr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002156}
2157
Philipp Reisner774b3052011-02-22 02:07:03 -05002158/* caution. no locking. */
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02002159void drbd_destroy_device(struct kref *kref)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002160{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002161 struct drbd_device *device = container_of(kref, struct drbd_device, kref);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002162 struct drbd_resource *resource = device->resource;
2163 struct drbd_connection *connection;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002164
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002165 del_timer_sync(&device->request_timer);
Philipp Reisnerdfa8bed2011-06-29 14:06:08 +02002166
Philipp Reisnerb411b362009-09-25 16:07:19 -07002167 /* paranoia asserts */
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02002168 D_ASSERT(device, device->open_cnt == 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002169 /* end paranoia asserts */
2170
Philipp Reisnerb411b362009-09-25 16:07:19 -07002171 /* cleanup stuff that may have been allocated during
2172 * device (re-)configuration or state changes */
2173
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002174 if (device->this_bdev)
2175 bdput(device->this_bdev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002176
Philipp Reisner28995af2013-11-22 16:48:14 +01002177 drbd_free_ldev(device->ldev);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002178 device->ldev = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002179
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002180 drbd_release_all_peer_reqs(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002181
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002182 lc_destroy(device->act_log);
2183 lc_destroy(device->resync);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002184
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002185 kfree(device->p_uuid);
2186 /* device->p_uuid = NULL; */
Philipp Reisnerb411b362009-09-25 16:07:19 -07002187
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002188 if (device->bitmap) /* should no longer be there. */
2189 drbd_bm_cleanup(device);
2190 __free_page(device->md_io_page);
2191 put_disk(device->vdisk);
2192 blk_cleanup_queue(device->rq_queue);
2193 kfree(device->rs_plan_s);
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02002194 kfree(first_peer_device(device));
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002195 kfree(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002196
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002197 for_each_connection(connection, resource)
2198 kref_put(&connection->kref, drbd_destroy_connection);
2199 kref_put(&resource->kref, drbd_destroy_resource);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002200}
2201
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002202/* One global retry thread, if we need to push back some bio and have it
2203 * reinserted through our make request function.
2204 */
2205static struct retry_worker {
2206 struct workqueue_struct *wq;
2207 struct work_struct worker;
2208
2209 spinlock_t lock;
2210 struct list_head writes;
2211} retry;
2212
2213static void do_retry(struct work_struct *ws)
2214{
2215 struct retry_worker *retry = container_of(ws, struct retry_worker, worker);
2216 LIST_HEAD(writes);
2217 struct drbd_request *req, *tmp;
2218
2219 spin_lock_irq(&retry->lock);
2220 list_splice_init(&retry->writes, &writes);
2221 spin_unlock_irq(&retry->lock);
2222
2223 list_for_each_entry_safe(req, tmp, &writes, tl_requests) {
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02002224 struct drbd_device *device = req->device;
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002225 struct bio *bio = req->master_bio;
2226 unsigned long start_time = req->start_time;
Lars Ellenberg9a278a72012-07-24 10:12:36 +02002227 bool expected;
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002228
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02002229 expected =
Lars Ellenberg9a278a72012-07-24 10:12:36 +02002230 expect(atomic_read(&req->completion_ref) == 0) &&
2231 expect(req->rq_state & RQ_POSTPONED) &&
2232 expect((req->rq_state & RQ_LOCAL_PENDING) == 0 ||
2233 (req->rq_state & RQ_LOCAL_ABORTED) != 0);
2234
2235 if (!expected)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02002236 drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n",
Lars Ellenberg9a278a72012-07-24 10:12:36 +02002237 req, atomic_read(&req->completion_ref),
2238 req->rq_state);
2239
2240 /* We still need to put one kref associated with the
2241 * "completion_ref" going zero in the code path that queued it
2242 * here. The request object may still be referenced by a
2243 * frozen local req->private_bio, in case we force-detached.
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002244 */
Lars Ellenberg9a278a72012-07-24 10:12:36 +02002245 kref_put(&req->kref, drbd_req_destroy);
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002246
2247 /* A single suspended or otherwise blocking device may stall
2248 * all others as well. Fortunately, this code path is to
2249 * recover from a situation that "should not happen":
2250 * concurrent writes in multi-primary setup.
2251 * In a "normal" lifecycle, this workqueue is supposed to be
2252 * destroyed without ever doing anything.
2253 * If it turns out to be an issue anyways, we can do per
2254 * resource (replication group) or per device (minor) retry
2255 * workqueues instead.
2256 */
2257
2258 /* We are not just doing generic_make_request(),
2259 * as we want to keep the start_time information. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002260 inc_ap_bio(device);
2261 __drbd_make_request(device, bio, start_time);
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002262 }
2263}
2264
Lars Ellenberg9d05e7c2012-07-17 10:05:04 +02002265void drbd_restart_request(struct drbd_request *req)
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002266{
2267 unsigned long flags;
2268 spin_lock_irqsave(&retry.lock, flags);
2269 list_move_tail(&req->tl_requests, &retry.writes);
2270 spin_unlock_irqrestore(&retry.lock, flags);
2271
2272 /* Drop the extra reference that would otherwise
2273 * have been dropped by complete_master_bio.
2274 * do_retry() needs to grab a new one. */
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02002275 dec_ap_bio(req->device);
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002276
2277 queue_work(retry.wq, &retry.worker);
2278}
2279
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002280void drbd_destroy_resource(struct kref *kref)
2281{
2282 struct drbd_resource *resource =
2283 container_of(kref, struct drbd_resource, kref);
2284
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002285 idr_destroy(&resource->devices);
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002286 free_cpumask_var(resource->cpu_mask);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002287 kfree(resource->name);
2288 kfree(resource);
2289}
2290
2291void drbd_free_resource(struct drbd_resource *resource)
2292{
2293 struct drbd_connection *connection, *tmp;
2294
2295 for_each_connection_safe(connection, tmp, resource) {
2296 list_del(&connection->connections);
2297 kref_put(&connection->kref, drbd_destroy_connection);
2298 }
2299 kref_put(&resource->kref, drbd_destroy_resource);
2300}
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002301
Philipp Reisnerb411b362009-09-25 16:07:19 -07002302static void drbd_cleanup(void)
2303{
2304 unsigned int i;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002305 struct drbd_device *device;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002306 struct drbd_resource *resource, *tmp;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002307
Lars Ellenberg17a93f302010-11-24 10:37:35 +01002308 /* first remove proc,
2309 * drbdsetup uses it's presence to detect
2310 * whether DRBD is loaded.
2311 * If we would get stuck in proc removal,
2312 * but have netlink already deregistered,
2313 * some drbdsetup commands may wait forever
2314 * for an answer.
2315 */
2316 if (drbd_proc)
2317 remove_proc_entry("drbd", NULL);
2318
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002319 if (retry.wq)
2320 destroy_workqueue(retry.wq);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002321
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01002322 drbd_genl_unregister();
Philipp Reisnerb411b362009-09-25 16:07:19 -07002323
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002324 idr_for_each_entry(&drbd_devices, device, i)
Andreas Gruenbacherf82795d2011-07-03 23:32:26 +02002325 drbd_delete_device(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002326
Philipp Reisnerc141ebd2011-05-05 16:13:10 +02002327 /* not _rcu since, no other updater anymore. Genl already unregistered */
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002328 for_each_resource_safe(resource, tmp, &drbd_resources) {
2329 list_del(&resource->resources);
2330 drbd_free_resource(resource);
Philipp Reisner81fa2e62011-05-04 15:10:30 +02002331 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07002332
Philipp Reisner81a5d602011-02-22 19:53:16 -05002333 drbd_destroy_mempools();
Philipp Reisnerb411b362009-09-25 16:07:19 -07002334 unregister_blkdev(DRBD_MAJOR, "drbd");
2335
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02002336 idr_destroy(&drbd_devices);
Philipp Reisner81a5d602011-02-22 19:53:16 -05002337
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002338 pr_info("module cleanup done.\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07002339}
2340
2341/**
Artem Bityutskiyd97482e2012-07-25 18:12:12 +03002342 * drbd_congested() - Callback for the flusher thread
Philipp Reisnerb411b362009-09-25 16:07:19 -07002343 * @congested_data: User data
Artem Bityutskiyd97482e2012-07-25 18:12:12 +03002344 * @bdi_bits: Bits the BDI flusher thread is currently interested in
Philipp Reisnerb411b362009-09-25 16:07:19 -07002345 *
2346 * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested.
2347 */
2348static int drbd_congested(void *congested_data, int bdi_bits)
2349{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002350 struct drbd_device *device = congested_data;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002351 struct request_queue *q;
2352 char reason = '-';
2353 int r = 0;
2354
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002355 if (!may_inc_ap_bio(device)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07002356 /* DRBD has frozen IO */
2357 r = bdi_bits;
2358 reason = 'd';
2359 goto out;
2360 }
2361
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02002362 if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) {
Lars Ellenbergc2ba6862012-06-14 15:14:06 +02002363 r |= (1 << BDI_async_congested);
2364 /* Without good local data, we would need to read from remote,
2365 * and that would need the worker thread as well, which is
2366 * currently blocked waiting for that usermode helper to
2367 * finish.
2368 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002369 if (!get_ldev_if_state(device, D_UP_TO_DATE))
Lars Ellenbergc2ba6862012-06-14 15:14:06 +02002370 r |= (1 << BDI_sync_congested);
2371 else
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002372 put_ldev(device);
Lars Ellenbergc2ba6862012-06-14 15:14:06 +02002373 r &= bdi_bits;
2374 reason = 'c';
2375 goto out;
2376 }
2377
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002378 if (get_ldev(device)) {
2379 q = bdev_get_queue(device->ldev->backing_bdev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002380 r = bdi_congested(&q->backing_dev_info, bdi_bits);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002381 put_ldev(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002382 if (r)
2383 reason = 'b';
2384 }
2385
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02002386 if (bdi_bits & (1 << BDI_async_congested) &&
2387 test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07002388 r |= (1 << BDI_async_congested);
2389 reason = reason == 'b' ? 'a' : 'n';
2390 }
2391
2392out:
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002393 device->congestion_reason = reason;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002394 return r;
2395}
2396
Philipp Reisner6699b652011-02-09 11:10:24 +01002397static void drbd_init_workqueue(struct drbd_work_queue* wq)
2398{
Philipp Reisner6699b652011-02-09 11:10:24 +01002399 spin_lock_init(&wq->q_lock);
2400 INIT_LIST_HEAD(&wq->q);
Lars Ellenberg8c0785a2011-10-19 11:50:57 +02002401 init_waitqueue_head(&wq->q_wait);
Philipp Reisner6699b652011-02-09 11:10:24 +01002402}
2403
Andreas Gruenbacherb5043c52011-07-28 15:56:02 +02002404struct completion_work {
2405 struct drbd_work w;
2406 struct completion done;
2407};
2408
2409static int w_complete(struct drbd_work *w, int cancel)
2410{
2411 struct completion_work *completion_work =
2412 container_of(w, struct completion_work, w);
2413
2414 complete(&completion_work->done);
2415 return 0;
2416}
2417
2418void drbd_flush_workqueue(struct drbd_work_queue *work_queue)
2419{
2420 struct completion_work completion_work;
2421
2422 completion_work.w.cb = w_complete;
2423 init_completion(&completion_work.done);
2424 drbd_queue_work(work_queue, &completion_work.w);
2425 wait_for_completion(&completion_work.done);
2426}
2427
Andreas Gruenbacher4bc76042011-06-13 14:27:45 +02002428struct drbd_resource *drbd_find_resource(const char *name)
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002429{
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002430 struct drbd_resource *resource;
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002431
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01002432 if (!name || !name[0])
2433 return NULL;
2434
Philipp Reisnerc141ebd2011-05-05 16:13:10 +02002435 rcu_read_lock();
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002436 for_each_resource_rcu(resource, &drbd_resources) {
2437 if (!strcmp(resource->name, name)) {
Andreas Gruenbacher4bc76042011-06-13 14:27:45 +02002438 kref_get(&resource->kref);
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002439 goto found;
Philipp Reisner0ace9df2011-04-24 10:53:19 +02002440 }
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002441 }
Andreas Gruenbacher4bc76042011-06-13 14:27:45 +02002442 resource = NULL;
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002443found:
Philipp Reisnerc141ebd2011-05-05 16:13:10 +02002444 rcu_read_unlock();
Andreas Gruenbacher4bc76042011-06-13 14:27:45 +02002445 return resource;
Philipp Reisner1aba4d72011-02-21 15:38:08 +01002446}
2447
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002448struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len,
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002449 void *peer_addr, int peer_addr_len)
2450{
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002451 struct drbd_resource *resource;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002452 struct drbd_connection *connection;
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002453
2454 rcu_read_lock();
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002455 for_each_resource_rcu(resource, &drbd_resources) {
2456 for_each_connection_rcu(connection, resource) {
2457 if (connection->my_addr_len == my_addr_len &&
2458 connection->peer_addr_len == peer_addr_len &&
2459 !memcmp(&connection->my_addr, my_addr, my_addr_len) &&
2460 !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) {
2461 kref_get(&connection->kref);
2462 goto found;
2463 }
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002464 }
2465 }
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002466 connection = NULL;
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002467found:
2468 rcu_read_unlock();
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002469 return connection;
Andreas Gruenbacher089c0752011-06-14 18:28:09 +02002470}
2471
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002472static int drbd_alloc_socket(struct drbd_socket *socket)
2473{
2474 socket->rbuf = (void *) __get_free_page(GFP_KERNEL);
2475 if (!socket->rbuf)
2476 return -ENOMEM;
Andreas Gruenbacher5a87d922011-03-24 21:17:52 +01002477 socket->sbuf = (void *) __get_free_page(GFP_KERNEL);
2478 if (!socket->sbuf)
2479 return -ENOMEM;
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002480 return 0;
2481}
2482
2483static void drbd_free_socket(struct drbd_socket *socket)
2484{
Andreas Gruenbacher5a87d922011-03-24 21:17:52 +01002485 free_page((unsigned long) socket->sbuf);
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002486 free_page((unsigned long) socket->rbuf);
2487}
2488
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002489void conn_free_crypto(struct drbd_connection *connection)
Philipp Reisner91fd4da2011-04-20 17:47:29 +02002490{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002491 drbd_free_sock(connection);
Philipp Reisner1d041222011-04-22 15:20:23 +02002492
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002493 crypto_free_hash(connection->csums_tfm);
2494 crypto_free_hash(connection->verify_tfm);
2495 crypto_free_hash(connection->cram_hmac_tfm);
2496 crypto_free_hash(connection->integrity_tfm);
2497 crypto_free_hash(connection->peer_integrity_tfm);
2498 kfree(connection->int_dig_in);
2499 kfree(connection->int_dig_vv);
Philipp Reisner1d041222011-04-22 15:20:23 +02002500
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002501 connection->csums_tfm = NULL;
2502 connection->verify_tfm = NULL;
2503 connection->cram_hmac_tfm = NULL;
2504 connection->integrity_tfm = NULL;
2505 connection->peer_integrity_tfm = NULL;
2506 connection->int_dig_in = NULL;
2507 connection->int_dig_vv = NULL;
Philipp Reisner91fd4da2011-04-20 17:47:29 +02002508}
2509
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002510int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts)
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002511{
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002512 struct drbd_connection *connection;
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002513 cpumask_var_t new_cpu_mask;
2514 int err;
2515
2516 if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL))
2517 return -ENOMEM;
2518 /*
2519 retcode = ERR_NOMEM;
2520 drbd_msg_put_info("unable to allocate cpumask");
2521 */
2522
2523 /* silently ignore cpu mask on UP kernel */
2524 if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) {
Andreas Gruenbacherf44d0432011-07-22 13:53:19 +02002525 err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE,
Philipp Reisnerc5b005a2012-04-30 12:53:52 +02002526 cpumask_bits(new_cpu_mask), nr_cpu_ids);
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002527 if (err) {
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +02002528 drbd_warn(resource, "bitmap_parse() failed with %d\n", err);
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002529 /* retcode = ERR_CPU_MASK_PARSE; */
2530 goto fail;
2531 }
2532 }
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002533 resource->res_opts = *res_opts;
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002534 if (cpumask_empty(new_cpu_mask))
2535 drbd_calc_cpu_mask(&new_cpu_mask);
2536 if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) {
2537 cpumask_copy(resource->cpu_mask, new_cpu_mask);
2538 for_each_connection_rcu(connection, resource) {
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002539 connection->receiver.reset_cpu_mask = 1;
2540 connection->asender.reset_cpu_mask = 1;
2541 connection->worker.reset_cpu_mask = 1;
2542 }
Andreas Gruenbacherafbbfa82011-06-16 17:58:02 +02002543 }
2544 err = 0;
2545
2546fail:
2547 free_cpumask_var(new_cpu_mask);
2548 return err;
2549
2550}
2551
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002552struct drbd_resource *drbd_create_resource(const char *name)
2553{
2554 struct drbd_resource *resource;
2555
Andreas Gruenbacher6bbf53c2011-07-08 01:19:44 +02002556 resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002557 if (!resource)
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002558 goto fail;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002559 resource->name = kstrdup(name, GFP_KERNEL);
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002560 if (!resource->name)
2561 goto fail_free_resource;
2562 if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL))
2563 goto fail_free_name;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002564 kref_init(&resource->kref);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002565 idr_init(&resource->devices);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002566 INIT_LIST_HEAD(&resource->connections);
Philipp Reisnere9526582013-11-22 15:53:41 +01002567 resource->write_ordering = WO_bdev_flush;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002568 list_add_tail_rcu(&resource->resources, &drbd_resources);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02002569 mutex_init(&resource->conf_update);
Lars Ellenberg9e276872014-04-28 18:43:22 +02002570 mutex_init(&resource->adm_mutex);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02002571 spin_lock_init(&resource->req_lock);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002572 return resource;
Andreas Gruenbacher625a6ba2011-07-22 14:29:02 +02002573
2574fail_free_name:
2575 kfree(resource->name);
2576fail_free_resource:
2577 kfree(resource);
2578fail:
2579 return NULL;
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002580}
2581
Philipp Reisnerec0bddb2011-05-04 15:47:01 +02002582/* caller must be under genl_lock() */
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002583struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts)
Philipp Reisner21114382011-01-19 12:26:59 +01002584{
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002585 struct drbd_resource *resource;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002586 struct drbd_connection *connection;
Philipp Reisner21114382011-01-19 12:26:59 +01002587
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002588 connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL);
2589 if (!connection)
Philipp Reisner21114382011-01-19 12:26:59 +01002590 return NULL;
2591
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002592 if (drbd_alloc_socket(&connection->data))
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002593 goto fail;
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002594 if (drbd_alloc_socket(&connection->meta))
Andreas Gruenbachere6ef8a52011-03-24 18:07:54 +01002595 goto fail;
2596
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002597 connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL);
2598 if (!connection->current_epoch)
Philipp Reisner12038a32011-11-09 19:18:00 +01002599 goto fail;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +01002600
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002601 INIT_LIST_HEAD(&connection->transfer_log);
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +01002602
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002603 INIT_LIST_HEAD(&connection->current_epoch->list);
2604 connection->epochs = 1;
2605 spin_lock_init(&connection->epoch_lock);
Philipp Reisner4b0007c2011-11-09 20:12:34 +01002606
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002607 connection->send.seen_any_write_yet = false;
2608 connection->send.current_epoch_nr = 0;
2609 connection->send.current_epoch_writes = 0;
Lars Ellenbergb6dd1a82011-11-28 15:04:49 +01002610
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002611 resource = drbd_create_resource(name);
2612 if (!resource)
2613 goto fail;
2614
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002615 connection->cstate = C_STANDALONE;
2616 mutex_init(&connection->cstate_mutex);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002617 init_waitqueue_head(&connection->ping_wait);
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002618 idr_init(&connection->peer_devices);
Philipp Reisnerb2fb6dbe2011-01-19 13:48:44 +01002619
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002620 drbd_init_workqueue(&connection->sender_work);
2621 mutex_init(&connection->data.mutex);
2622 mutex_init(&connection->meta.mutex);
Philipp Reisner6699b652011-02-09 11:10:24 +01002623
Andreas Gruenbacher2457b6d2011-07-21 13:45:21 +02002624 drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver");
2625 connection->receiver.connection = connection;
2626 drbd_thread_init(resource, &connection->worker, drbd_worker, "worker");
2627 connection->worker.connection = connection;
2628 drbd_thread_init(resource, &connection->asender, drbd_asender, "asender");
2629 connection->asender.connection = connection;
Philipp Reisner392c8802011-02-09 10:33:31 +01002630
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002631 kref_init(&connection->kref);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002632
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002633 connection->resource = resource;
Philipp Reisner21114382011-01-19 12:26:59 +01002634
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002635 if (set_resource_options(resource, res_opts))
2636 goto fail_resource;
2637
2638 kref_get(&resource->kref);
2639 list_add_tail_rcu(&connection->connections, &resource->connections);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002640 return connection;
Philipp Reisner21114382011-01-19 12:26:59 +01002641
Andreas Gruenbachereb6bea62011-06-21 16:11:28 +02002642fail_resource:
2643 list_del(&resource->resources);
2644 drbd_free_resource(resource);
Philipp Reisner21114382011-01-19 12:26:59 +01002645fail:
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002646 kfree(connection->current_epoch);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002647 drbd_free_socket(&connection->meta);
2648 drbd_free_socket(&connection->data);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002649 kfree(connection);
Philipp Reisner21114382011-01-19 12:26:59 +01002650 return NULL;
2651}
2652
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02002653void drbd_destroy_connection(struct kref *kref)
Philipp Reisner21114382011-01-19 12:26:59 +01002654{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002655 struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002656 struct drbd_resource *resource = connection->resource;
Philipp Reisner9dc9fbb2011-04-22 15:23:32 +02002657
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002658 if (atomic_read(&connection->current_epoch->epoch_size) != 0)
Andreas Gruenbacher1ec861e2011-07-06 11:01:44 +02002659 drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size));
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002660 kfree(connection->current_epoch);
Philipp Reisner12038a32011-11-09 19:18:00 +01002661
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002662 idr_destroy(&connection->peer_devices);
Philipp Reisner21114382011-01-19 12:26:59 +01002663
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002664 drbd_free_socket(&connection->meta);
2665 drbd_free_socket(&connection->data);
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002666 kfree(connection->int_dig_in);
2667 kfree(connection->int_dig_vv);
2668 kfree(connection);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002669 kref_put(&resource->kref, drbd_destroy_resource);
Philipp Reisner21114382011-01-19 12:26:59 +01002670}
2671
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002672static int init_submitter(struct drbd_device *device)
Lars Ellenberg113fef92013-03-22 18:14:40 -06002673{
2674 /* opencoded create_singlethread_workqueue(),
2675 * to be able to say "drbd%d", ..., minor */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002676 device->submit.wq = alloc_workqueue("drbd%u_submit",
2677 WQ_UNBOUND | WQ_MEM_RECLAIM, 1, device->minor);
2678 if (!device->submit.wq)
Lars Ellenberg113fef92013-03-22 18:14:40 -06002679 return -ENOMEM;
2680
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002681 INIT_WORK(&device->submit.worker, do_submit);
2682 spin_lock_init(&device->submit.lock);
2683 INIT_LIST_HEAD(&device->submit.writes);
Lars Ellenberg113fef92013-03-22 18:14:40 -06002684 return 0;
2685}
2686
Lars Ellenberga910b122014-04-28 18:43:21 +02002687enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002688{
Lars Ellenberga910b122014-04-28 18:43:21 +02002689 struct drbd_resource *resource = adm_ctx->resource;
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002690 struct drbd_connection *connection;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002691 struct drbd_device *device;
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002692 struct drbd_peer_device *peer_device, *tmp_peer_device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002693 struct gendisk *disk;
2694 struct request_queue *q;
Andreas Gruenbacher93e4bf72013-05-23 14:57:17 +02002695 int id;
Lars Ellenberga910b122014-04-28 18:43:21 +02002696 int vnr = adm_ctx->volume;
Lars Ellenberg8432b312011-03-08 16:11:16 +01002697 enum drbd_ret_code err = ERR_NOMEM;
Philipp Reisner774b3052011-02-22 02:07:03 -05002698
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002699 device = minor_to_device(minor);
2700 if (device)
Philipp Reisner774b3052011-02-22 02:07:03 -05002701 return ERR_MINOR_EXISTS;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002702
2703 /* GFP_KERNEL, we are outside of all write-out paths */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002704 device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL);
2705 if (!device)
Philipp Reisner774b3052011-02-22 02:07:03 -05002706 return ERR_NOMEM;
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002707 kref_init(&device->kref);
2708
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002709 kref_get(&resource->kref);
2710 device->resource = resource;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002711 device->minor = minor;
2712 device->vnr = vnr;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002713
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002714 drbd_init_set_defaults(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002715
2716 q = blk_alloc_queue(GFP_KERNEL);
2717 if (!q)
2718 goto out_no_q;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002719 device->rq_queue = q;
2720 q->queuedata = device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002721
2722 disk = alloc_disk(1);
2723 if (!disk)
2724 goto out_no_disk;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002725 device->vdisk = disk;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002726
Andreas Gruenbacher81e84652010-12-09 15:03:57 +01002727 set_disk_ro(disk, true);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002728
2729 disk->queue = q;
2730 disk->major = DRBD_MAJOR;
2731 disk->first_minor = minor;
2732 disk->fops = &drbd_ops;
2733 sprintf(disk->disk_name, "drbd%d", minor);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002734 disk->private_data = device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002735
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002736 device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor));
Philipp Reisnerb411b362009-09-25 16:07:19 -07002737 /* we have no partitions. we contain only ourselves. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002738 device->this_bdev->bd_contains = device->this_bdev;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002739
2740 q->backing_dev_info.congested_fn = drbd_congested;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002741 q->backing_dev_info.congested_data = device;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002742
Andreas Gruenbacher2f58dcf2010-12-13 17:48:19 +01002743 blk_queue_make_request(q, drbd_make_request);
Lars Ellenberga73ff322012-06-25 19:15:38 +02002744 blk_queue_flush(q, REQ_FLUSH | REQ_FUA);
Philipp Reisner99432fc2011-05-20 16:39:13 +02002745 /* Setting the max_hw_sectors to an odd value of 8kibyte here
2746 This triggers a max_bio_size message upon first attach or connect */
2747 blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002748 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
2749 blk_queue_merge_bvec(q, drbd_merge_bvec);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02002750 q->queue_lock = &resource->req_lock;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002751
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002752 device->md_io_page = alloc_page(GFP_KERNEL);
2753 if (!device->md_io_page)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002754 goto out_no_io_page;
2755
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002756 if (drbd_bm_init(device))
Philipp Reisnerb411b362009-09-25 16:07:19 -07002757 goto out_no_bitmap;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002758 device->read_requests = RB_ROOT;
2759 device->write_requests = RB_ROOT;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002760
Andreas Gruenbacher93e4bf72013-05-23 14:57:17 +02002761 id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL);
2762 if (id < 0) {
2763 if (id == -ENOSPC) {
Tejun Heo56de2102013-02-27 17:04:01 -08002764 err = ERR_MINOR_EXISTS;
Lars Ellenberga910b122014-04-28 18:43:21 +02002765 drbd_msg_put_info(adm_ctx->reply_skb, "requested minor exists already");
Tejun Heo56de2102013-02-27 17:04:01 -08002766 }
Lars Ellenberg8432b312011-03-08 16:11:16 +01002767 goto out_no_minor_idr;
Tejun Heo56de2102013-02-27 17:04:01 -08002768 }
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002769 kref_get(&device->kref);
2770
2771 id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL);
2772 if (id < 0) {
2773 if (id == -ENOSPC) {
2774 err = ERR_MINOR_EXISTS;
Lars Ellenberga910b122014-04-28 18:43:21 +02002775 drbd_msg_put_info(adm_ctx->reply_skb, "requested minor exists already");
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002776 }
2777 goto out_idr_remove_minor;
2778 }
2779 kref_get(&device->kref);
Tejun Heo56de2102013-02-27 17:04:01 -08002780
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002781 INIT_LIST_HEAD(&device->peer_devices);
2782 for_each_connection(connection, resource) {
2783 peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL);
2784 if (!peer_device)
2785 goto out_idr_remove_from_resource;
2786 peer_device->connection = connection;
2787 peer_device->device = device;
2788
2789 list_add(&peer_device->peer_devices, &device->peer_devices);
2790 kref_get(&device->kref);
2791
2792 id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL);
2793 if (id < 0) {
2794 if (id == -ENOSPC) {
2795 err = ERR_INVALID_REQUEST;
Lars Ellenberga910b122014-04-28 18:43:21 +02002796 drbd_msg_put_info(adm_ctx->reply_skb, "requested volume exists already");
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002797 }
2798 goto out_idr_remove_from_resource;
Tejun Heo56de2102013-02-27 17:04:01 -08002799 }
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002800 kref_get(&connection->kref);
Lars Ellenberg569083c2011-03-07 09:49:02 +01002801 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07002802
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002803 if (init_submitter(device)) {
Lars Ellenberg113fef92013-03-22 18:14:40 -06002804 err = ERR_NOMEM;
Lars Ellenberga910b122014-04-28 18:43:21 +02002805 drbd_msg_put_info(adm_ctx->reply_skb, "unable to create submit workqueue");
Lars Ellenberg113fef92013-03-22 18:14:40 -06002806 goto out_idr_remove_vol;
2807 }
2808
Philipp Reisner774b3052011-02-22 02:07:03 -05002809 add_disk(disk);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002810
Philipp Reisner2325eb62011-03-15 16:56:18 +01002811 /* inherit the connection state */
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002812 device->state.conn = first_connection(resource)->cstate;
Andreas Gruenbacher69a22772011-08-09 00:47:13 +02002813 if (device->state.conn == C_WF_REPORT_PARAMS) {
2814 for_each_peer_device(peer_device, device)
2815 drbd_connected(peer_device);
2816 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07002817
Philipp Reisner774b3052011-02-22 02:07:03 -05002818 return NO_ERROR;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002819
Lars Ellenberg113fef92013-03-22 18:14:40 -06002820out_idr_remove_vol:
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002821 idr_remove(&connection->peer_devices, vnr);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002822out_idr_remove_from_resource:
Andreas Gruenbacherb6f85ef2011-07-06 15:03:31 +02002823 for_each_connection(connection, resource) {
2824 peer_device = idr_find(&connection->peer_devices, vnr);
2825 if (peer_device) {
2826 idr_remove(&connection->peer_devices, vnr);
2827 kref_put(&connection->kref, drbd_destroy_connection);
2828 }
2829 }
2830 for_each_peer_device_safe(peer_device, tmp_peer_device, device) {
2831 list_del(&peer_device->peer_devices);
2832 kfree(peer_device);
2833 }
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002834 idr_remove(&resource->devices, vnr);
Lars Ellenberg8432b312011-03-08 16:11:16 +01002835out_idr_remove_minor:
Andreas Gruenbacher93e4bf72013-05-23 14:57:17 +02002836 idr_remove(&drbd_devices, minor);
Lars Ellenberg569083c2011-03-07 09:49:02 +01002837 synchronize_rcu();
Lars Ellenberg8432b312011-03-08 16:11:16 +01002838out_no_minor_idr:
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002839 drbd_bm_cleanup(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002840out_no_bitmap:
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002841 __free_page(device->md_io_page);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002842out_no_io_page:
2843 put_disk(disk);
2844out_no_disk:
2845 blk_cleanup_queue(q);
2846out_no_q:
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002847 kref_put(&resource->kref, drbd_destroy_resource);
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02002848 kfree(device);
Lars Ellenberg8432b312011-03-08 16:11:16 +01002849 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002850}
2851
Andreas Gruenbacherf82795d2011-07-03 23:32:26 +02002852void drbd_delete_device(struct drbd_device *device)
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002853{
2854 struct drbd_resource *resource = device->resource;
2855 struct drbd_connection *connection;
2856 int refs = 3;
2857
2858 for_each_connection(connection, resource) {
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002859 idr_remove(&connection->peer_devices, device->vnr);
Andreas Gruenbacher803ea132011-06-09 01:40:48 +02002860 refs++;
2861 }
2862 idr_remove(&resource->devices, device->vnr);
2863 idr_remove(&drbd_devices, device_to_minor(device));
2864 del_gendisk(device->vdisk);
2865 synchronize_rcu();
2866 kref_sub(&device->kref, refs, drbd_destroy_device);
2867}
2868
Lars Ellenberg8ce953a2014-02-27 09:46:18 +01002869static int __init drbd_init(void)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002870{
2871 int err;
2872
Philipp Reisner2b8a90b2011-01-10 11:15:17 +01002873 if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002874 pr_err("invalid minor_count (%d)\n", minor_count);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002875#ifdef MODULE
2876 return -EINVAL;
2877#else
Andreas Gruenbacher46530e82011-05-31 13:08:53 +02002878 minor_count = DRBD_MINOR_COUNT_DEF;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002879#endif
2880 }
2881
Philipp Reisnerb411b362009-09-25 16:07:19 -07002882 err = register_blkdev(DRBD_MAJOR, "drbd");
2883 if (err) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002884 pr_err("unable to register block device major %d\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07002885 DRBD_MAJOR);
2886 return err;
2887 }
2888
Philipp Reisnerb411b362009-09-25 16:07:19 -07002889 /*
2890 * allocate all necessary structs
2891 */
Philipp Reisnerb411b362009-09-25 16:07:19 -07002892 init_waitqueue_head(&drbd_pp_wait);
2893
2894 drbd_proc = NULL; /* play safe for drbd_cleanup */
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02002895 idr_init(&drbd_devices);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002896
Lars Ellenberg69babf02013-10-23 10:59:15 +02002897 rwlock_init(&global_state_lock);
Andreas Gruenbacher77c556f2011-06-08 22:17:38 +02002898 INIT_LIST_HEAD(&drbd_resources);
Lars Ellenberg69babf02013-10-23 10:59:15 +02002899
2900 err = drbd_genl_register();
2901 if (err) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002902 pr_err("unable to register generic netlink family\n");
Lars Ellenberg69babf02013-10-23 10:59:15 +02002903 goto fail;
2904 }
2905
Philipp Reisnerb411b362009-09-25 16:07:19 -07002906 err = drbd_create_mempools();
2907 if (err)
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01002908 goto fail;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002909
Wei Yongjun6110d702013-06-25 16:50:04 +02002910 err = -ENOMEM;
Lars Ellenberg8c484ee2010-03-11 16:47:58 +01002911 drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002912 if (!drbd_proc) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002913 pr_err("unable to register proc file\n");
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01002914 goto fail;
Philipp Reisnerb411b362009-09-25 16:07:19 -07002915 }
2916
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002917 retry.wq = create_singlethread_workqueue("drbd-reissue");
2918 if (!retry.wq) {
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002919 pr_err("unable to create retry workqueue\n");
Lars Ellenberg2312f0b32011-11-24 10:36:25 +01002920 goto fail;
2921 }
2922 INIT_WORK(&retry.worker, do_retry);
2923 spin_lock_init(&retry.lock);
2924 INIT_LIST_HEAD(&retry.writes);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002925
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002926 pr_info("initialized. "
Philipp Reisnerb411b362009-09-25 16:07:19 -07002927 "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n",
2928 API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX);
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002929 pr_info("%s\n", drbd_buildtag());
2930 pr_info("registered as block device major %d\n", DRBD_MAJOR);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002931 return 0; /* Success! */
2932
Lars Ellenberg3b98c0c2011-03-07 12:49:34 +01002933fail:
Philipp Reisnerb411b362009-09-25 16:07:19 -07002934 drbd_cleanup();
2935 if (err == -ENOMEM)
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002936 pr_err("ran out of memory\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07002937 else
Lars Ellenbergf88c5d92014-03-27 14:10:55 +01002938 pr_err("initialization failure\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07002939 return err;
2940}
2941
Philipp Reisner28995af2013-11-22 16:48:14 +01002942void drbd_free_ldev(struct drbd_backing_dev *ldev)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002943{
2944 if (ldev == NULL)
2945 return;
2946
Tejun Heoe525fd82010-11-13 11:55:17 +01002947 blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2948 blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002949
Lars Ellenberg94ad0a12013-03-27 14:08:42 +01002950 kfree(ldev->disk_conf);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002951 kfree(ldev);
2952}
2953
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002954void drbd_free_sock(struct drbd_connection *connection)
Philipp Reisnerb411b362009-09-25 16:07:19 -07002955{
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002956 if (connection->data.socket) {
2957 mutex_lock(&connection->data.mutex);
2958 kernel_sock_shutdown(connection->data.socket, SHUT_RDWR);
2959 sock_release(connection->data.socket);
2960 connection->data.socket = NULL;
2961 mutex_unlock(&connection->data.mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002962 }
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002963 if (connection->meta.socket) {
2964 mutex_lock(&connection->meta.mutex);
2965 kernel_sock_shutdown(connection->meta.socket, SHUT_RDWR);
2966 sock_release(connection->meta.socket);
2967 connection->meta.socket = NULL;
2968 mutex_unlock(&connection->meta.mutex);
Philipp Reisnerb411b362009-09-25 16:07:19 -07002969 }
2970}
2971
Philipp Reisnerb411b362009-09-25 16:07:19 -07002972/* meta data management */
2973
Andreas Gruenbacherbde89a92011-05-30 16:32:41 +02002974void conn_md_sync(struct drbd_connection *connection)
Philipp Reisner19fffd72012-08-28 16:48:03 +02002975{
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002976 struct drbd_peer_device *peer_device;
Philipp Reisner19fffd72012-08-28 16:48:03 +02002977 int vnr;
2978
2979 rcu_read_lock();
Andreas Gruenbacherc06ece62011-06-21 17:23:59 +02002980 idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
2981 struct drbd_device *device = peer_device->device;
2982
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002983 kref_get(&device->kref);
Philipp Reisner19fffd72012-08-28 16:48:03 +02002984 rcu_read_unlock();
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02002985 drbd_md_sync(device);
Andreas Gruenbacher05a10ec2011-06-07 22:54:17 +02002986 kref_put(&device->kref, drbd_destroy_device);
Philipp Reisner19fffd72012-08-28 16:48:03 +02002987 rcu_read_lock();
2988 }
2989 rcu_read_unlock();
2990}
2991
Lars Ellenbergae8bf312013-03-19 18:16:43 +01002992/* aligned 4kByte */
Philipp Reisnerb411b362009-09-25 16:07:19 -07002993struct meta_data_on_disk {
Lars Ellenbergcccac982013-03-19 18:16:46 +01002994 u64 la_size_sect; /* last agreed size. */
Philipp Reisnerb411b362009-09-25 16:07:19 -07002995 u64 uuid[UI_SIZE]; /* UUIDs. */
2996 u64 device_uuid;
2997 u64 reserved_u64_1;
2998 u32 flags; /* MDF */
2999 u32 magic;
3000 u32 md_size_sect;
3001 u32 al_offset; /* offset to this block */
Lars Ellenbergae8bf312013-03-19 18:16:43 +01003002 u32 al_nr_extents; /* important for restoring the AL (userspace) */
Lars Ellenbergf3990022011-03-23 14:31:09 +01003003 /* `-- act_log->nr_elements <-- ldev->dc.al_extents */
Philipp Reisnerb411b362009-09-25 16:07:19 -07003004 u32 bm_offset; /* offset to the bitmap, from here */
3005 u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */
Philipp Reisner99432fc2011-05-20 16:39:13 +02003006 u32 la_peer_max_bio_size; /* last peer max_bio_size */
Philipp Reisnerb411b362009-09-25 16:07:19 -07003007
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003008 /* see al_tr_number_to_on_disk_sector() */
3009 u32 al_stripes;
3010 u32 al_stripe_size_4k;
3011
3012 u8 reserved_u8[4096 - (7*8 + 10*4)];
Philipp Reisnerb411b362009-09-25 16:07:19 -07003013} __packed;
3014
Philipp Reisnerd752b262013-06-25 16:50:08 +02003015
3016
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003017void drbd_md_write(struct drbd_device *device, void *b)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003018{
Philipp Reisnerd752b262013-06-25 16:50:08 +02003019 struct meta_data_on_disk *buffer = b;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003020 sector_t sector;
3021 int i;
3022
Lars Ellenbergae8bf312013-03-19 18:16:43 +01003023 memset(buffer, 0, sizeof(*buffer));
Philipp Reisnerb411b362009-09-25 16:07:19 -07003024
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003025 buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev));
Philipp Reisnerb411b362009-09-25 16:07:19 -07003026 for (i = UI_CURRENT; i < UI_SIZE; i++)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003027 buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]);
3028 buffer->flags = cpu_to_be32(device->ldev->md.flags);
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003029 buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003030
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003031 buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect);
3032 buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset);
3033 buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003034 buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003035 buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003036
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003037 buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset);
3038 buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003039
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003040 buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes);
3041 buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k);
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003042
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02003043 D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003044 sector = device->ldev->md.md_offset;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003045
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003046 if (drbd_md_sync_page_io(device, device->ldev, sector, WRITE)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07003047 /* this was a try anyways ... */
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003048 drbd_err(device, "meta data update failed!\n");
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003049 drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003050 }
Philipp Reisnerd752b262013-06-25 16:50:08 +02003051}
3052
3053/**
3054 * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003055 * @device: DRBD device.
Philipp Reisnerd752b262013-06-25 16:50:08 +02003056 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003057void drbd_md_sync(struct drbd_device *device)
Philipp Reisnerd752b262013-06-25 16:50:08 +02003058{
3059 struct meta_data_on_disk *buffer;
3060
3061 /* Don't accidentally change the DRBD meta data layout. */
3062 BUILD_BUG_ON(UI_SIZE != 4);
3063 BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096);
3064
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003065 del_timer(&device->md_sync_timer);
Philipp Reisnerd752b262013-06-25 16:50:08 +02003066 /* timer may be rearmed by drbd_md_mark_dirty() now. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003067 if (!test_and_clear_bit(MD_DIRTY, &device->flags))
Philipp Reisnerd752b262013-06-25 16:50:08 +02003068 return;
3069
3070 /* We use here D_FAILED and not D_ATTACHING because we try to write
3071 * metadata even if we detach due to a disk failure! */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003072 if (!get_ldev_if_state(device, D_FAILED))
Philipp Reisnerd752b262013-06-25 16:50:08 +02003073 return;
3074
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003075 buffer = drbd_md_get_buffer(device);
Philipp Reisnerd752b262013-06-25 16:50:08 +02003076 if (!buffer)
3077 goto out;
3078
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003079 drbd_md_write(device, buffer);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003080
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003081 /* Update device->ldev->md.la_size_sect,
Philipp Reisnerb411b362009-09-25 16:07:19 -07003082 * since we updated it on metadata. */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003083 device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003084
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003085 drbd_md_put_buffer(device);
Philipp Reisnere1711732011-06-27 11:51:46 +02003086out:
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003087 put_ldev(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003088}
3089
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003090static int check_activity_log_stripe_size(struct drbd_device *device,
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003091 struct meta_data_on_disk *on_disk,
3092 struct drbd_md *in_core)
3093{
3094 u32 al_stripes = be32_to_cpu(on_disk->al_stripes);
3095 u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k);
3096 u64 al_size_4k;
3097
3098 /* both not set: default to old fixed size activity log */
3099 if (al_stripes == 0 && al_stripe_size_4k == 0) {
3100 al_stripes = 1;
3101 al_stripe_size_4k = MD_32kB_SECT/8;
3102 }
3103
3104 /* some paranoia plausibility checks */
3105
3106 /* we need both values to be set */
3107 if (al_stripes == 0 || al_stripe_size_4k == 0)
3108 goto err;
3109
3110 al_size_4k = (u64)al_stripes * al_stripe_size_4k;
3111
3112 /* Upper limit of activity log area, to avoid potential overflow
3113 * problems in al_tr_number_to_on_disk_sector(). As right now, more
3114 * than 72 * 4k blocks total only increases the amount of history,
3115 * limiting this arbitrarily to 16 GB is not a real limitation ;-) */
3116 if (al_size_4k > (16 * 1024 * 1024/4))
3117 goto err;
3118
3119 /* Lower limit: we need at least 8 transaction slots (32kB)
3120 * to not break existing setups */
3121 if (al_size_4k < MD_32kB_SECT/8)
3122 goto err;
3123
3124 in_core->al_stripe_size_4k = al_stripe_size_4k;
3125 in_core->al_stripes = al_stripes;
3126 in_core->al_size_4k = al_size_4k;
3127
3128 return 0;
3129err:
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003130 drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n",
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003131 al_stripes, al_stripe_size_4k);
3132 return -EINVAL;
3133}
3134
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003135static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev)
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003136{
3137 sector_t capacity = drbd_get_capacity(bdev->md_bdev);
3138 struct drbd_md *in_core = &bdev->md;
3139 s32 on_disk_al_sect;
3140 s32 on_disk_bm_sect;
3141
3142 /* The on-disk size of the activity log, calculated from offsets, and
3143 * the size of the activity log calculated from the stripe settings,
3144 * should match.
3145 * Though we could relax this a bit: it is ok, if the striped activity log
3146 * fits in the available on-disk activity log size.
3147 * Right now, that would break how resize is implemented.
3148 * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware
3149 * of possible unused padding space in the on disk layout. */
3150 if (in_core->al_offset < 0) {
3151 if (in_core->bm_offset > in_core->al_offset)
3152 goto err;
3153 on_disk_al_sect = -in_core->al_offset;
3154 on_disk_bm_sect = in_core->al_offset - in_core->bm_offset;
3155 } else {
3156 if (in_core->al_offset != MD_4kB_SECT)
3157 goto err;
3158 if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT)
3159 goto err;
3160
3161 on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT;
3162 on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset;
3163 }
3164
3165 /* old fixed size meta data is exactly that: fixed. */
3166 if (in_core->meta_dev_idx >= 0) {
3167 if (in_core->md_size_sect != MD_128MB_SECT
3168 || in_core->al_offset != MD_4kB_SECT
3169 || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT
3170 || in_core->al_stripes != 1
3171 || in_core->al_stripe_size_4k != MD_32kB_SECT/8)
3172 goto err;
3173 }
3174
3175 if (capacity < in_core->md_size_sect)
3176 goto err;
3177 if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev))
3178 goto err;
3179
3180 /* should be aligned, and at least 32k */
3181 if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT))
3182 goto err;
3183
3184 /* should fit (for now: exactly) into the available on-disk space;
3185 * overflow prevention is in check_activity_log_stripe_size() above. */
3186 if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT)
3187 goto err;
3188
3189 /* again, should be aligned */
3190 if (in_core->bm_offset & 7)
3191 goto err;
3192
3193 /* FIXME check for device grow with flex external meta data? */
3194
3195 /* can the available bitmap space cover the last agreed device size? */
3196 if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512)
3197 goto err;
3198
3199 return 0;
3200
3201err:
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003202 drbd_err(device, "meta data offsets don't make sense: idx=%d "
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003203 "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, "
3204 "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n",
3205 in_core->meta_dev_idx,
3206 in_core->al_stripes, in_core->al_stripe_size_4k,
3207 in_core->al_offset, in_core->bm_offset, in_core->md_size_sect,
3208 (unsigned long long)in_core->la_size_sect,
3209 (unsigned long long)capacity);
3210
3211 return -EINVAL;
3212}
3213
3214
Philipp Reisnerb411b362009-09-25 16:07:19 -07003215/**
3216 * drbd_md_read() - Reads in the meta data super block
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003217 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003218 * @bdev: Device from which the meta data should be read in.
3219 *
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003220 * Return NO_ERROR on success, and an enum drbd_ret_code in case
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003221 * something goes wrong.
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003222 *
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003223 * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS,
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003224 * even before @bdev is assigned to @device->ldev.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003225 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003226int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003227{
3228 struct meta_data_on_disk *buffer;
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003229 u32 magic, flags;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003230 int i, rv = NO_ERROR;
3231
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003232 if (device->state.disk != D_DISKLESS)
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003233 return ERR_DISK_CONFIGURED;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003234
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003235 buffer = drbd_md_get_buffer(device);
Philipp Reisnere1711732011-06-27 11:51:46 +02003236 if (!buffer)
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003237 return ERR_NOMEM;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003238
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003239 /* First, figure out where our meta data superblock is located,
3240 * and read it. */
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003241 bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx;
3242 bdev->md.md_offset = drbd_md_ss(bdev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003243
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003244 if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, READ)) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003245 /* NOTE: can't do normal error processing here as this is
Philipp Reisnerb411b362009-09-25 16:07:19 -07003246 called BEFORE disk is attached */
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003247 drbd_err(device, "Error while reading metadata.\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07003248 rv = ERR_IO_MD_DISK;
3249 goto err;
3250 }
3251
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003252 magic = be32_to_cpu(buffer->magic);
3253 flags = be32_to_cpu(buffer->flags);
3254 if (magic == DRBD_MD_MAGIC_84_UNCLEAN ||
3255 (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) {
3256 /* btw: that's Activity Log clean, not "all" clean. */
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003257 drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n");
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003258 rv = ERR_MD_UNCLEAN;
3259 goto err;
3260 }
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003261
3262 rv = ERR_MD_INVALID;
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003263 if (magic != DRBD_MD_MAGIC_08) {
Philipp Reisner43de7c82011-11-10 13:16:13 +01003264 if (magic == DRBD_MD_MAGIC_07)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003265 drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n");
Lars Ellenbergd5d7ebd2011-07-05 20:59:26 +02003266 else
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003267 drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n");
Philipp Reisnerb411b362009-09-25 16:07:19 -07003268 goto err;
3269 }
3270
3271 if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003272 drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003273 be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003274 goto err;
3275 }
3276
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003277
3278 /* convert to in_core endian */
3279 bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003280 for (i = UI_CURRENT; i < UI_SIZE; i++)
3281 bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]);
3282 bdev->md.flags = be32_to_cpu(buffer->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003283 bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid);
3284
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003285 bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect);
3286 bdev->md.al_offset = be32_to_cpu(buffer->al_offset);
3287 bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset);
3288
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003289 if (check_activity_log_stripe_size(device, buffer, &bdev->md))
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003290 goto err;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003291 if (check_offsets_and_sizes(device, bdev))
Lars Ellenbergc04ccaa2013-03-19 18:16:47 +01003292 goto err;
3293
Philipp Reisnerb411b362009-09-25 16:07:19 -07003294 if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003295 drbd_err(device, "unexpected bm_offset: %d (expected %d)\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003296 be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003297 goto err;
3298 }
3299 if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) {
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003300 drbd_err(device, "unexpected md_size: %u (expected %u)\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003301 be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003302 goto err;
3303 }
3304
Lars Ellenberg3a4d4eb2013-03-19 18:16:44 +01003305 rv = NO_ERROR;
3306
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003307 spin_lock_irq(&device->resource->req_lock);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003308 if (device->state.conn < C_CONNECTED) {
Lars Ellenbergdb141b22012-06-25 19:15:58 +02003309 unsigned int peer;
Philipp Reisner99432fc2011-05-20 16:39:13 +02003310 peer = be32_to_cpu(buffer->la_peer_max_bio_size);
Lars Ellenbergdb141b22012-06-25 19:15:58 +02003311 peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003312 device->peer_max_bio_size = peer;
Philipp Reisner99432fc2011-05-20 16:39:13 +02003313 }
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003314 spin_unlock_irq(&device->resource->req_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003315
3316 err:
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003317 drbd_md_put_buffer(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003318
3319 return rv;
3320}
3321
3322/**
3323 * drbd_md_mark_dirty() - Mark meta data super block as dirty
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003324 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003325 *
3326 * Call this function if you change anything that should be written to
3327 * the meta-data super block. This function sets MD_DIRTY, and starts a
3328 * timer that ensures that within five seconds you have to call drbd_md_sync().
3329 */
Lars Ellenbergca0e6092010-10-14 15:01:21 +02003330#ifdef DEBUG
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003331void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func)
Lars Ellenbergee15b032010-09-03 10:00:09 +02003332{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003333 if (!test_and_set_bit(MD_DIRTY, &device->flags)) {
3334 mod_timer(&device->md_sync_timer, jiffies + HZ);
3335 device->last_md_mark_dirty.line = line;
3336 device->last_md_mark_dirty.func = func;
Lars Ellenbergee15b032010-09-03 10:00:09 +02003337 }
3338}
3339#else
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003340void drbd_md_mark_dirty(struct drbd_device *device)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003341{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003342 if (!test_and_set_bit(MD_DIRTY, &device->flags))
3343 mod_timer(&device->md_sync_timer, jiffies + 5*HZ);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003344}
Lars Ellenbergee15b032010-09-03 10:00:09 +02003345#endif
Philipp Reisnerb411b362009-09-25 16:07:19 -07003346
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003347void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003348{
3349 int i;
3350
Lars Ellenberg62b0da32011-01-20 13:25:21 +01003351 for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003352 device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
Philipp Reisnerb411b362009-09-25 16:07:19 -07003353}
3354
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003355void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003356{
3357 if (idx == UI_CURRENT) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003358 if (device->state.role == R_PRIMARY)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003359 val |= 1;
3360 else
3361 val &= ~((u64)1);
3362
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003363 drbd_set_ed_uuid(device, val);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003364 }
3365
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003366 device->ldev->md.uuid[idx] = val;
3367 drbd_md_mark_dirty(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003368}
3369
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003370void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003371{
3372 unsigned long flags;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003373 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3374 __drbd_uuid_set(device, idx, val);
3375 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003376}
Philipp Reisnerb411b362009-09-25 16:07:19 -07003377
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003378void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003379{
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003380 unsigned long flags;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003381 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
3382 if (device->ldev->md.uuid[idx]) {
3383 drbd_uuid_move_history(device);
3384 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx];
Philipp Reisnerb411b362009-09-25 16:07:19 -07003385 }
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003386 __drbd_uuid_set(device, idx, val);
3387 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003388}
3389
3390/**
3391 * drbd_uuid_new_current() - Creates a new current UUID
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003392 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003393 *
3394 * Creates a new current UUID, and rotates the old current UUID into
3395 * the bitmap slot. Causes an incremental resync upon next connect.
3396 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003397void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003398{
3399 u64 val;
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003400 unsigned long long bm_uuid;
3401
3402 get_random_bytes(&val, sizeof(u64));
3403
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003404 spin_lock_irq(&device->ldev->md.uuid_lock);
3405 bm_uuid = device->ldev->md.uuid[UI_BITMAP];
Philipp Reisnerb411b362009-09-25 16:07:19 -07003406
Lars Ellenberg62b0da32011-01-20 13:25:21 +01003407 if (bm_uuid)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003408 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
Lars Ellenberg62b0da32011-01-20 13:25:21 +01003409
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003410 device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT];
3411 __drbd_uuid_set(device, UI_CURRENT, val);
3412 spin_unlock_irq(&device->ldev->md.uuid_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003413
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003414 drbd_print_uuids(device, "new current UUID");
Lars Ellenbergaaa8e2b2010-10-15 13:16:53 +02003415 /* get it to stable storage _now_ */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003416 drbd_md_sync(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003417}
3418
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003419void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003420{
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003421 unsigned long flags;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003422 if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003423 return;
3424
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003425 spin_lock_irqsave(&device->ldev->md.uuid_lock, flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003426 if (val == 0) {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003427 drbd_uuid_move_history(device);
3428 device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP];
3429 device->ldev->md.uuid[UI_BITMAP] = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003430 } else {
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003431 unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP];
Lars Ellenberg62b0da32011-01-20 13:25:21 +01003432 if (bm_uuid)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003433 drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003434
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003435 device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003436 }
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003437 spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags);
Philipp Reisner9f2247b2012-08-16 14:25:58 +02003438
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003439 drbd_md_mark_dirty(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003440}
3441
3442/**
3443 * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003444 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003445 *
3446 * Sets all bits in the bitmap and writes the whole bitmap to stable storage.
3447 */
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003448int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003449{
3450 int rv = -EIO;
3451
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003452 drbd_md_set_flag(device, MDF_FULL_SYNC);
3453 drbd_md_sync(device);
3454 drbd_bm_set_all(device);
3455
3456 rv = drbd_bm_write(device);
3457
3458 if (!rv) {
3459 drbd_md_clear_flag(device, MDF_FULL_SYNC);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003460 drbd_md_sync(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003461 }
3462
3463 return rv;
3464}
3465
3466/**
3467 * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io()
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003468 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003469 *
3470 * Clears all bits in the bitmap and writes the whole bitmap to stable storage.
3471 */
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003472int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003473{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003474 drbd_resume_al(device);
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003475 drbd_bm_clear_all(device);
3476 return drbd_bm_write(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003477}
3478
Andreas Gruenbacher99920dc2011-03-16 15:31:39 +01003479static int w_bitmap_io(struct drbd_work *w, int unused)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003480{
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02003481 struct drbd_device *device =
3482 container_of(w, struct drbd_device, bm_io_work.w);
3483 struct bm_io_work *work = &device->bm_io_work;
Lars Ellenberg02851e92010-12-16 14:47:39 +01003484 int rv = -EIO;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003485
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02003486 D_ASSERT(device, atomic_read(&device->ap_bio_cnt) == 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003487
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003488 if (get_ldev(device)) {
3489 drbd_bm_lock(device, work->why, work->flags);
3490 rv = work->io_fn(device);
3491 drbd_bm_unlock(device);
3492 put_ldev(device);
Lars Ellenberg02851e92010-12-16 14:47:39 +01003493 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07003494
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003495 clear_bit_unlock(BITMAP_IO, &device->flags);
3496 wake_up(&device->misc_wait);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003497
3498 if (work->done)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003499 work->done(device, rv);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003500
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003501 clear_bit(BITMAP_IO_QUEUED, &device->flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003502 work->why = NULL;
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01003503 work->flags = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003504
Andreas Gruenbacher99920dc2011-03-16 15:31:39 +01003505 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003506}
3507
3508/**
3509 * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003510 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003511 * @io_fn: IO callback to be called when bitmap IO is possible
3512 * @done: callback to be called after the bitmap IO was performed
3513 * @why: Descriptive text of the reason for doing the IO
3514 *
3515 * While IO on the bitmap happens we freeze application IO thus we ensure
3516 * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be
3517 * called from worker context. It MUST NOT be used while a previous such
3518 * work is still pending!
Philipp Reisner8fe39aa2013-11-22 13:22:13 +01003519 *
3520 * Its worker function encloses the call of io_fn() by get_ldev() and
3521 * put_ldev().
Philipp Reisnerb411b362009-09-25 16:07:19 -07003522 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003523void drbd_queue_bitmap_io(struct drbd_device *device,
Andreas Gruenbacher54761692011-05-30 16:15:21 +02003524 int (*io_fn)(struct drbd_device *),
3525 void (*done)(struct drbd_device *, int),
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01003526 char *why, enum bm_flag flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003527{
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02003528 D_ASSERT(device, current == first_peer_device(device)->connection->worker.task);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003529
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02003530 D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags));
3531 D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags));
3532 D_ASSERT(device, list_empty(&device->bm_io_work.w.list));
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003533 if (device->bm_io_work.why)
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003534 drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n",
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003535 why, device->bm_io_work.why);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003536
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003537 device->bm_io_work.io_fn = io_fn;
3538 device->bm_io_work.done = done;
3539 device->bm_io_work.why = why;
3540 device->bm_io_work.flags = flags;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003541
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003542 spin_lock_irq(&device->resource->req_lock);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003543 set_bit(BITMAP_IO, &device->flags);
3544 if (atomic_read(&device->ap_bio_cnt) == 0) {
3545 if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags))
Andreas Gruenbacher84b8c062011-07-28 15:27:51 +02003546 drbd_queue_work(&first_peer_device(device)->connection->sender_work,
3547 &device->bm_io_work.w);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003548 }
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003549 spin_unlock_irq(&device->resource->req_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003550}
3551
3552/**
3553 * drbd_bitmap_io() - Does an IO operation on the whole bitmap
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003554 * @device: DRBD device.
Philipp Reisnerb411b362009-09-25 16:07:19 -07003555 * @io_fn: IO callback to be called when bitmap IO is possible
3556 * @why: Descriptive text of the reason for doing the IO
3557 *
3558 * freezes application IO while that the actual IO operations runs. This
3559 * functions MAY NOT be called from worker context.
3560 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003561int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *),
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01003562 char *why, enum bm_flag flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003563{
3564 int rv;
3565
Andreas Gruenbacher0b0ba1e2011-06-27 16:23:33 +02003566 D_ASSERT(device, current != first_peer_device(device)->connection->worker.task);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003567
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01003568 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003569 drbd_suspend_io(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003570
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003571 drbd_bm_lock(device, why, flags);
3572 rv = io_fn(device);
3573 drbd_bm_unlock(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003574
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01003575 if ((flags & BM_LOCKED_SET_ALLOWED) == 0)
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003576 drbd_resume_io(device);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003577
3578 return rv;
3579}
3580
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003581void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003582{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003583 if ((device->ldev->md.flags & flag) != flag) {
3584 drbd_md_mark_dirty(device);
3585 device->ldev->md.flags |= flag;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003586 }
3587}
3588
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003589void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003590{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003591 if ((device->ldev->md.flags & flag) != 0) {
3592 drbd_md_mark_dirty(device);
3593 device->ldev->md.flags &= ~flag;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003594 }
3595}
3596int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag)
3597{
3598 return (bdev->md.flags & flag) != 0;
3599}
3600
3601static void md_sync_timer_fn(unsigned long data)
3602{
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003603 struct drbd_device *device = (struct drbd_device *) data;
Lars Ellenbergac0acb92014-02-11 09:47:58 +01003604 drbd_device_post_work(device, MD_SYNC);
Philipp Reisnerb411b362009-09-25 16:07:19 -07003605}
3606
Andreas Gruenbacherd8763022011-01-26 17:39:41 +01003607const char *cmdname(enum drbd_packet cmd)
Andreas Gruenbacherf2ad9062011-01-26 17:13:25 +01003608{
3609 /* THINK may need to become several global tables
3610 * when we want to support more than
3611 * one PRO_VERSION */
3612 static const char *cmdnames[] = {
3613 [P_DATA] = "Data",
3614 [P_DATA_REPLY] = "DataReply",
3615 [P_RS_DATA_REPLY] = "RSDataReply",
3616 [P_BARRIER] = "Barrier",
3617 [P_BITMAP] = "ReportBitMap",
3618 [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget",
3619 [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource",
3620 [P_UNPLUG_REMOTE] = "UnplugRemote",
3621 [P_DATA_REQUEST] = "DataRequest",
3622 [P_RS_DATA_REQUEST] = "RSDataRequest",
3623 [P_SYNC_PARAM] = "SyncParam",
3624 [P_SYNC_PARAM89] = "SyncParam89",
3625 [P_PROTOCOL] = "ReportProtocol",
3626 [P_UUIDS] = "ReportUUIDs",
3627 [P_SIZES] = "ReportSizes",
3628 [P_STATE] = "ReportState",
3629 [P_SYNC_UUID] = "ReportSyncUUID",
3630 [P_AUTH_CHALLENGE] = "AuthChallenge",
3631 [P_AUTH_RESPONSE] = "AuthResponse",
3632 [P_PING] = "Ping",
3633 [P_PING_ACK] = "PingAck",
3634 [P_RECV_ACK] = "RecvAck",
3635 [P_WRITE_ACK] = "WriteAck",
3636 [P_RS_WRITE_ACK] = "RSWriteAck",
Lars Ellenbergd4dabbe2012-08-01 12:33:51 +02003637 [P_SUPERSEDED] = "Superseded",
Andreas Gruenbacherf2ad9062011-01-26 17:13:25 +01003638 [P_NEG_ACK] = "NegAck",
3639 [P_NEG_DREPLY] = "NegDReply",
3640 [P_NEG_RS_DREPLY] = "NegRSDReply",
3641 [P_BARRIER_ACK] = "BarrierAck",
3642 [P_STATE_CHG_REQ] = "StateChgRequest",
3643 [P_STATE_CHG_REPLY] = "StateChgReply",
3644 [P_OV_REQUEST] = "OVRequest",
3645 [P_OV_REPLY] = "OVReply",
3646 [P_OV_RESULT] = "OVResult",
3647 [P_CSUM_RS_REQUEST] = "CsumRSRequest",
3648 [P_RS_IS_IN_SYNC] = "CsumRSIsInSync",
3649 [P_COMPRESSED_BITMAP] = "CBitmap",
3650 [P_DELAY_PROBE] = "DelayProbe",
3651 [P_OUT_OF_SYNC] = "OutOfSync",
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003652 [P_RETRY_WRITE] = "RetryWrite",
Lars Ellenbergae25b332011-04-24 00:01:16 +02003653 [P_RS_CANCEL] = "RSCancel",
3654 [P_CONN_ST_CHG_REQ] = "conn_st_chg_req",
3655 [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply",
Philipp Reisner036b17e2011-05-16 17:38:11 +02003656 [P_RETRY_WRITE] = "retry_write",
3657 [P_PROTOCOL_UPDATE] = "protocol_update",
Lars Ellenbergae25b332011-04-24 00:01:16 +02003658
3659 /* enum drbd_packet, but not commands - obsoleted flags:
3660 * P_MAY_IGNORE
3661 * P_MAX_OPT_CMD
3662 */
Andreas Gruenbacherf2ad9062011-01-26 17:13:25 +01003663 };
3664
Lars Ellenbergae25b332011-04-24 00:01:16 +02003665 /* too big for the array: 0xfffX */
Andreas Gruenbachere5d6f332011-03-28 16:44:40 +02003666 if (cmd == P_INITIAL_META)
3667 return "InitialMeta";
3668 if (cmd == P_INITIAL_DATA)
3669 return "InitialData";
Andreas Gruenbacher60381782011-03-28 17:05:50 +02003670 if (cmd == P_CONNECTION_FEATURES)
3671 return "ConnectionFeatures";
Andreas Gruenbacher6e849ce2011-03-14 17:27:45 +01003672 if (cmd >= ARRAY_SIZE(cmdnames))
Andreas Gruenbacherf2ad9062011-01-26 17:13:25 +01003673 return "Unknown";
3674 return cmdnames[cmd];
3675}
3676
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003677/**
3678 * drbd_wait_misc - wait for a request to make progress
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003679 * @device: device associated with the request
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003680 * @i: the struct drbd_interval embedded in struct drbd_request or
3681 * struct drbd_peer_request
3682 */
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003683int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i)
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003684{
Philipp Reisner44ed1672011-04-19 17:10:19 +02003685 struct net_conf *nc;
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003686 DEFINE_WAIT(wait);
3687 long timeout;
3688
Philipp Reisner44ed1672011-04-19 17:10:19 +02003689 rcu_read_lock();
Andreas Gruenbachera6b32bc2011-05-31 14:33:49 +02003690 nc = rcu_dereference(first_peer_device(device)->connection->net_conf);
Philipp Reisner44ed1672011-04-19 17:10:19 +02003691 if (!nc) {
3692 rcu_read_unlock();
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003693 return -ETIMEDOUT;
Philipp Reisner44ed1672011-04-19 17:10:19 +02003694 }
3695 timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT;
3696 rcu_read_unlock();
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003697
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003698 /* Indicate to wake up device->misc_wait on progress. */
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003699 i->waiting = true;
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003700 prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003701 spin_unlock_irq(&device->resource->req_lock);
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003702 timeout = schedule_timeout(timeout);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003703 finish_wait(&device->misc_wait, &wait);
Andreas Gruenbacher05008132011-07-07 14:19:42 +02003704 spin_lock_irq(&device->resource->req_lock);
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003705 if (!timeout || device->state.conn < C_CONNECTED)
Andreas Gruenbacher7be8da02011-02-22 02:15:32 +01003706 return -ETIMEDOUT;
3707 if (signal_pending(current))
3708 return -ERESTARTSYS;
3709 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07003710}
3711
3712#ifdef CONFIG_DRBD_FAULT_INJECTION
3713/* Fault insertion support including random number generator shamelessly
3714 * stolen from kernel/rcutorture.c */
3715struct fault_random_state {
3716 unsigned long state;
3717 unsigned long count;
3718};
3719
3720#define FAULT_RANDOM_MULT 39916801 /* prime */
3721#define FAULT_RANDOM_ADD 479001701 /* prime */
3722#define FAULT_RANDOM_REFRESH 10000
3723
3724/*
3725 * Crude but fast random-number generator. Uses a linear congruential
3726 * generator, with occasional help from get_random_bytes().
3727 */
3728static unsigned long
3729_drbd_fault_random(struct fault_random_state *rsp)
3730{
3731 long refresh;
3732
Roel Kluin49829ea2009-12-15 22:55:44 +01003733 if (!rsp->count--) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07003734 get_random_bytes(&refresh, sizeof(refresh));
3735 rsp->state += refresh;
3736 rsp->count = FAULT_RANDOM_REFRESH;
3737 }
3738 rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD;
3739 return swahw32(rsp->state);
3740}
3741
3742static char *
3743_drbd_fault_str(unsigned int type) {
3744 static char *_faults[] = {
3745 [DRBD_FAULT_MD_WR] = "Meta-data write",
3746 [DRBD_FAULT_MD_RD] = "Meta-data read",
3747 [DRBD_FAULT_RS_WR] = "Resync write",
3748 [DRBD_FAULT_RS_RD] = "Resync read",
3749 [DRBD_FAULT_DT_WR] = "Data write",
3750 [DRBD_FAULT_DT_RD] = "Data read",
3751 [DRBD_FAULT_DT_RA] = "Data read ahead",
3752 [DRBD_FAULT_BM_ALLOC] = "BM allocation",
Philipp Reisner6b4388a2010-04-26 14:11:45 +02003753 [DRBD_FAULT_AL_EE] = "EE allocation",
3754 [DRBD_FAULT_RECEIVE] = "receive data corruption",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003755 };
3756
3757 return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**";
3758}
3759
3760unsigned int
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003761_drbd_insert_fault(struct drbd_device *device, unsigned int type)
Philipp Reisnerb411b362009-09-25 16:07:19 -07003762{
3763 static struct fault_random_state rrs = {0, 0};
3764
3765 unsigned int ret = (
3766 (fault_devs == 0 ||
Andreas Gruenbacherb30ab792011-07-03 13:26:43 +02003767 ((1 << device_to_minor(device)) & fault_devs) != 0) &&
Philipp Reisnerb411b362009-09-25 16:07:19 -07003768 (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate));
3769
3770 if (ret) {
3771 fault_count++;
3772
Lars Ellenberg73835062010-05-27 11:51:56 +02003773 if (__ratelimit(&drbd_ratelimit_state))
Andreas Gruenbacherd0180172011-07-03 17:53:52 +02003774 drbd_warn(device, "***Simulating %s failure\n",
Philipp Reisnerb411b362009-09-25 16:07:19 -07003775 _drbd_fault_str(type));
3776 }
3777
3778 return ret;
3779}
3780#endif
3781
3782const char *drbd_buildtag(void)
3783{
3784 /* DRBD built from external sources has here a reference to the
3785 git hash of the source code. */
3786
3787 static char buildtag[38] = "\0uilt-in";
3788
3789 if (buildtag[0] == 0) {
Cong Wangbc4854b2012-04-03 14:13:36 +08003790#ifdef MODULE
3791 sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion);
3792#else
3793 buildtag[0] = 'b';
Philipp Reisnerb411b362009-09-25 16:07:19 -07003794#endif
Philipp Reisnerb411b362009-09-25 16:07:19 -07003795 }
3796
3797 return buildtag;
3798}
3799
3800module_init(drbd_init)
3801module_exit(drbd_cleanup)
3802
Philipp Reisnerb411b362009-09-25 16:07:19 -07003803EXPORT_SYMBOL(drbd_conn_str);
3804EXPORT_SYMBOL(drbd_role_str);
3805EXPORT_SYMBOL(drbd_disk_str);
3806EXPORT_SYMBOL(drbd_set_st_err_str);