]>
Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
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 | ||
b411b363 | 29 | #include <linux/module.h> |
b411b363 PR |
30 | #include <linux/drbd.h> |
31 | #include <asm/uaccess.h> | |
32 | #include <asm/types.h> | |
33 | #include <net/sock.h> | |
34 | #include <linux/ctype.h> | |
2a48fc0a | 35 | #include <linux/mutex.h> |
b411b363 PR |
36 | #include <linux/fs.h> |
37 | #include <linux/file.h> | |
38 | #include <linux/proc_fs.h> | |
39 | #include <linux/init.h> | |
40 | #include <linux/mm.h> | |
41 | #include <linux/memcontrol.h> | |
42 | #include <linux/mm_inline.h> | |
43 | #include <linux/slab.h> | |
44 | #include <linux/random.h> | |
45 | #include <linux/reboot.h> | |
46 | #include <linux/notifier.h> | |
47 | #include <linux/kthread.h> | |
113fef9e | 48 | #include <linux/workqueue.h> |
b411b363 PR |
49 | #define __KERNEL_SYSCALLS__ |
50 | #include <linux/unistd.h> | |
51 | #include <linux/vmalloc.h> | |
52 | ||
53 | #include <linux/drbd_limits.h> | |
54 | #include "drbd_int.h" | |
a3603a6e | 55 | #include "drbd_protocol.h" |
b411b363 PR |
56 | #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */ |
57 | ||
58 | #include "drbd_vli.h" | |
59 | ||
2a48fc0a | 60 | static DEFINE_MUTEX(drbd_main_mutex); |
b411b363 | 61 | int drbd_worker(struct drbd_thread *); |
b411b363 PR |
62 | |
63 | int drbd_init(void); | |
64 | static int drbd_open(struct block_device *bdev, fmode_t mode); | |
db2a144b | 65 | static void drbd_release(struct gendisk *gd, fmode_t mode); |
99920dc5 | 66 | static int w_md_sync(struct drbd_work *w, int unused); |
b411b363 | 67 | static void md_sync_timer_fn(unsigned long data); |
99920dc5 AG |
68 | static int w_bitmap_io(struct drbd_work *w, int unused); |
69 | static int w_go_diskless(struct drbd_work *w, int unused); | |
b411b363 | 70 | |
b411b363 PR |
71 | MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, " |
72 | "Lars Ellenberg <lars@linbit.com>"); | |
73 | MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION); | |
74 | MODULE_VERSION(REL_VERSION); | |
75 | MODULE_LICENSE("GPL"); | |
81a5d60e | 76 | MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices (" |
2b8a90b5 | 77 | __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")"); |
b411b363 PR |
78 | MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR); |
79 | ||
80 | #include <linux/moduleparam.h> | |
81 | /* allow_open_on_secondary */ | |
82 | MODULE_PARM_DESC(allow_oos, "DONT USE!"); | |
83 | /* thanks to these macros, if compiled into the kernel (not-module), | |
84 | * this becomes the boot parameter drbd.minor_count */ | |
85 | module_param(minor_count, uint, 0444); | |
86 | module_param(disable_sendpage, bool, 0644); | |
87 | module_param(allow_oos, bool, 0); | |
b411b363 PR |
88 | module_param(proc_details, int, 0644); |
89 | ||
90 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
91 | int enable_faults; | |
92 | int fault_rate; | |
93 | static int fault_count; | |
94 | int fault_devs; | |
95 | /* bitmap of enabled faults */ | |
96 | module_param(enable_faults, int, 0664); | |
97 | /* fault rate % value - applies to all enabled faults */ | |
98 | module_param(fault_rate, int, 0664); | |
99 | /* count of faults inserted */ | |
100 | module_param(fault_count, int, 0664); | |
101 | /* bitmap of devices to insert faults on */ | |
102 | module_param(fault_devs, int, 0644); | |
103 | #endif | |
104 | ||
105 | /* module parameter, defined */ | |
2b8a90b5 | 106 | unsigned int minor_count = DRBD_MINOR_COUNT_DEF; |
90ab5ee9 RR |
107 | bool disable_sendpage; |
108 | bool allow_oos; | |
b411b363 PR |
109 | int proc_details; /* Detail level in proc drbd*/ |
110 | ||
111 | /* Module parameter for setting the user mode helper program | |
112 | * to run. Default is /sbin/drbdadm */ | |
113 | char usermode_helper[80] = "/sbin/drbdadm"; | |
114 | ||
115 | module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644); | |
116 | ||
117 | /* in 2.6.x, our device mapping and config info contains our virtual gendisks | |
118 | * as member "struct gendisk *vdisk;" | |
119 | */ | |
05a10ec7 | 120 | struct idr drbd_devices; |
77c556f6 | 121 | struct list_head drbd_resources; |
b411b363 PR |
122 | |
123 | struct kmem_cache *drbd_request_cache; | |
6c852bec | 124 | struct kmem_cache *drbd_ee_cache; /* peer requests */ |
b411b363 PR |
125 | struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ |
126 | struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
127 | mempool_t *drbd_request_mempool; | |
128 | mempool_t *drbd_ee_mempool; | |
4281808f | 129 | mempool_t *drbd_md_io_page_pool; |
9476f39d | 130 | struct bio_set *drbd_md_io_bio_set; |
b411b363 PR |
131 | |
132 | /* I do not use a standard mempool, because: | |
133 | 1) I want to hand out the pre-allocated objects first. | |
134 | 2) I want to be able to interrupt sleeping allocation with a signal. | |
135 | Note: This is a single linked list, the next pointer is the private | |
136 | member of struct page. | |
137 | */ | |
138 | struct page *drbd_pp_pool; | |
139 | spinlock_t drbd_pp_lock; | |
140 | int drbd_pp_vacant; | |
141 | wait_queue_head_t drbd_pp_wait; | |
142 | ||
143 | DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5); | |
144 | ||
7d4e9d09 | 145 | static const struct block_device_operations drbd_ops = { |
b411b363 PR |
146 | .owner = THIS_MODULE, |
147 | .open = drbd_open, | |
148 | .release = drbd_release, | |
149 | }; | |
19f843aa | 150 | |
da4a75d2 LE |
151 | struct bio *bio_alloc_drbd(gfp_t gfp_mask) |
152 | { | |
153 | struct bio *bio; | |
19f843aa | 154 | |
da4a75d2 LE |
155 | if (!drbd_md_io_bio_set) |
156 | return bio_alloc(gfp_mask, 1); | |
19f843aa | 157 | |
da4a75d2 LE |
158 | bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set); |
159 | if (!bio) | |
160 | return NULL; | |
da4a75d2 | 161 | return bio; |
19f843aa LE |
162 | } |
163 | ||
b411b363 PR |
164 | #ifdef __CHECKER__ |
165 | /* When checking with sparse, and this is an inline function, sparse will | |
166 | give tons of false positives. When this is a real functions sparse works. | |
b411b363 | 167 | */ |
b30ab791 | 168 | int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins) |
b411b363 | 169 | { |
b411b363 | 170 | int io_allowed; |
b411b363 | 171 | |
b30ab791 AG |
172 | atomic_inc(&device->local_cnt); |
173 | io_allowed = (device->state.disk >= mins); | |
b411b363 | 174 | if (!io_allowed) { |
b30ab791 AG |
175 | if (atomic_dec_and_test(&device->local_cnt)) |
176 | wake_up(&device->misc_wait); | |
b411b363 | 177 | } |
b411b363 PR |
178 | return io_allowed; |
179 | } | |
b411b363 | 180 | |
b411b363 | 181 | #endif |
265be2d0 | 182 | |
b411b363 | 183 | /** |
b6dd1a89 | 184 | * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch |
bde89a9e | 185 | * @connection: DRBD connection. |
b411b363 PR |
186 | * @barrier_nr: Expected identifier of the DRBD write barrier packet. |
187 | * @set_size: Expected number of requests before that barrier. | |
188 | * | |
189 | * In case the passed barrier_nr or set_size does not match the oldest | |
b6dd1a89 LE |
190 | * epoch of not yet barrier-acked requests, this function will cause a |
191 | * termination of the connection. | |
b411b363 | 192 | */ |
bde89a9e | 193 | void tl_release(struct drbd_connection *connection, unsigned int barrier_nr, |
2f5cdd0b | 194 | unsigned int set_size) |
b411b363 | 195 | { |
b411b363 | 196 | struct drbd_request *r; |
b6dd1a89 LE |
197 | struct drbd_request *req = NULL; |
198 | int expect_epoch = 0; | |
199 | int expect_size = 0; | |
b411b363 | 200 | |
bde89a9e | 201 | spin_lock_irq(&connection->req_lock); |
b411b363 | 202 | |
98683650 | 203 | /* find oldest not yet barrier-acked write request, |
b6dd1a89 | 204 | * count writes in its epoch. */ |
bde89a9e | 205 | list_for_each_entry(r, &connection->transfer_log, tl_requests) { |
a0d856df | 206 | const unsigned s = r->rq_state; |
b6dd1a89 LE |
207 | if (!req) { |
208 | if (!(s & RQ_WRITE)) | |
209 | continue; | |
210 | if (!(s & RQ_NET_MASK)) | |
211 | continue; | |
212 | if (s & RQ_NET_DONE) | |
213 | continue; | |
214 | req = r; | |
215 | expect_epoch = req->epoch; | |
216 | expect_size ++; | |
217 | } else { | |
218 | if (r->epoch != expect_epoch) | |
219 | break; | |
220 | if (!(s & RQ_WRITE)) | |
221 | continue; | |
222 | /* if (s & RQ_DONE): not expected */ | |
223 | /* if (!(s & RQ_NET_MASK)): not expected */ | |
224 | expect_size++; | |
43a5182c | 225 | } |
b411b363 | 226 | } |
67098930 | 227 | |
b411b363 | 228 | /* first some paranoia code */ |
b6dd1a89 | 229 | if (req == NULL) { |
bde89a9e | 230 | conn_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n", |
2f5cdd0b | 231 | barrier_nr); |
b411b363 | 232 | goto bail; |
b411b363 | 233 | } |
b6dd1a89 | 234 | if (expect_epoch != barrier_nr) { |
bde89a9e | 235 | conn_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n", |
b6dd1a89 | 236 | barrier_nr, expect_epoch); |
b411b363 | 237 | goto bail; |
5a22db89 LE |
238 | } |
239 | ||
b6dd1a89 | 240 | if (expect_size != set_size) { |
bde89a9e | 241 | conn_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n", |
b6dd1a89 | 242 | barrier_nr, set_size, expect_size); |
b411b363 | 243 | goto bail; |
b411b363 PR |
244 | } |
245 | ||
98683650 PR |
246 | /* Clean up list of requests processed during current epoch. */ |
247 | /* this extra list walk restart is paranoia, | |
248 | * to catch requests being barrier-acked "unexpectedly". | |
249 | * It usually should find the same req again, or some READ preceding it. */ | |
bde89a9e | 250 | list_for_each_entry(req, &connection->transfer_log, tl_requests) |
98683650 PR |
251 | if (req->epoch == expect_epoch) |
252 | break; | |
bde89a9e | 253 | list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) { |
b6dd1a89 LE |
254 | if (req->epoch != expect_epoch) |
255 | break; | |
256 | _req_mod(req, BARRIER_ACKED); | |
19f843aa | 257 | } |
bde89a9e | 258 | spin_unlock_irq(&connection->req_lock); |
19f843aa | 259 | |
b411b363 | 260 | return; |
b411b363 | 261 | |
b411b363 | 262 | bail: |
bde89a9e AG |
263 | spin_unlock_irq(&connection->req_lock); |
264 | conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
b411b363 | 265 | } |
b411b363 | 266 | |
b411b363 | 267 | |
b411b363 | 268 | /** |
11b58e73 | 269 | * _tl_restart() - Walks the transfer log, and applies an action to all requests |
b30ab791 | 270 | * @device: DRBD device. |
11b58e73 | 271 | * @what: The action/event to perform with all request objects |
b411b363 | 272 | * |
8554df1c AG |
273 | * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO, |
274 | * RESTART_FROZEN_DISK_IO. | |
b411b363 | 275 | */ |
b6dd1a89 | 276 | /* must hold resource->req_lock */ |
bde89a9e | 277 | void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b411b363 | 278 | { |
b6dd1a89 | 279 | struct drbd_request *req, *r; |
b411b363 | 280 | |
bde89a9e | 281 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) |
b6dd1a89 LE |
282 | _req_mod(req, what); |
283 | } | |
738a84b2 | 284 | |
bde89a9e | 285 | void tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b6dd1a89 | 286 | { |
bde89a9e AG |
287 | spin_lock_irq(&connection->req_lock); |
288 | _tl_restart(connection, what); | |
289 | spin_unlock_irq(&connection->req_lock); | |
cdfda633 | 290 | } |
b411b363 | 291 | |
b411b363 PR |
292 | /** |
293 | * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL | |
b30ab791 | 294 | * @device: DRBD device. |
b411b363 PR |
295 | * |
296 | * This is called after the connection to the peer was lost. The storage covered | |
297 | * by the requests on the transfer gets marked as our of sync. Called from the | |
298 | * receiver thread and the worker thread. | |
299 | */ | |
bde89a9e | 300 | void tl_clear(struct drbd_connection *connection) |
b411b363 | 301 | { |
bde89a9e | 302 | tl_restart(connection, CONNECTION_LOST_WHILE_PENDING); |
b411b363 | 303 | } |
197296ff | 304 | |
cdfda633 | 305 | /** |
b30ab791 AG |
306 | * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL |
307 | * @device: DRBD device. | |
cdfda633 | 308 | */ |
b30ab791 | 309 | void tl_abort_disk_io(struct drbd_device *device) |
cdfda633 | 310 | { |
a6b32bc3 | 311 | struct drbd_connection *connection = first_peer_device(device)->connection; |
b6dd1a89 | 312 | struct drbd_request *req, *r; |
02851e9f | 313 | |
bde89a9e AG |
314 | spin_lock_irq(&connection->req_lock); |
315 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) { | |
97ddb687 LE |
316 | if (!(req->rq_state & RQ_LOCAL_PENDING)) |
317 | continue; | |
b30ab791 | 318 | if (req->w.device != device) |
b6dd1a89 LE |
319 | continue; |
320 | _req_mod(req, ABORT_DISK_IO); | |
b411b363 | 321 | } |
bde89a9e | 322 | spin_unlock_irq(&connection->req_lock); |
b411b363 PR |
323 | } |
324 | ||
b411b363 PR |
325 | static int drbd_thread_setup(void *arg) |
326 | { | |
327 | struct drbd_thread *thi = (struct drbd_thread *) arg; | |
bde89a9e | 328 | struct drbd_connection *connection = thi->connection; |
b411b363 PR |
329 | unsigned long flags; |
330 | int retval; | |
331 | ||
f1b3a6ec | 332 | snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s", |
77c556f6 AG |
333 | thi->name[0], |
334 | thi->connection->resource->name); | |
f1b3a6ec | 335 | |
b411b363 PR |
336 | restart: |
337 | retval = thi->function(thi); | |
338 | ||
339 | spin_lock_irqsave(&thi->t_lock, flags); | |
340 | ||
e77a0a5c | 341 | /* if the receiver has been "EXITING", the last thing it did |
b411b363 PR |
342 | * was set the conn state to "StandAlone", |
343 | * if now a re-connect request comes in, conn state goes C_UNCONNECTED, | |
344 | * and receiver thread will be "started". | |
e77a0a5c | 345 | * drbd_thread_start needs to set "RESTARTING" in that case. |
b411b363 | 346 | * t_state check and assignment needs to be within the same spinlock, |
e77a0a5c AG |
347 | * so either thread_start sees EXITING, and can remap to RESTARTING, |
348 | * or thread_start see NONE, and can proceed as normal. | |
b411b363 PR |
349 | */ |
350 | ||
e77a0a5c | 351 | if (thi->t_state == RESTARTING) { |
bde89a9e | 352 | conn_info(connection, "Restarting %s thread\n", thi->name); |
e77a0a5c | 353 | thi->t_state = RUNNING; |
b411b363 PR |
354 | spin_unlock_irqrestore(&thi->t_lock, flags); |
355 | goto restart; | |
356 | } | |
357 | ||
358 | thi->task = NULL; | |
e77a0a5c | 359 | thi->t_state = NONE; |
b411b363 | 360 | smp_mb(); |
992d6e91 | 361 | complete_all(&thi->stop); |
b411b363 PR |
362 | spin_unlock_irqrestore(&thi->t_lock, flags); |
363 | ||
bde89a9e | 364 | conn_info(connection, "Terminating %s\n", current->comm); |
b411b363 PR |
365 | |
366 | /* Release mod reference taken when thread was started */ | |
9dc9fbb3 | 367 | |
05a10ec7 | 368 | kref_put(&connection->kref, drbd_destroy_connection); |
b411b363 PR |
369 | module_put(THIS_MODULE); |
370 | return retval; | |
371 | } | |
372 | ||
bde89a9e | 373 | static void drbd_thread_init(struct drbd_connection *connection, struct drbd_thread *thi, |
bed879ae | 374 | int (*func) (struct drbd_thread *), char *name) |
b411b363 PR |
375 | { |
376 | spin_lock_init(&thi->t_lock); | |
377 | thi->task = NULL; | |
e77a0a5c | 378 | thi->t_state = NONE; |
b411b363 | 379 | thi->function = func; |
bde89a9e | 380 | thi->connection = connection; |
bed879ae | 381 | strncpy(thi->name, name, ARRAY_SIZE(thi->name)); |
b411b363 PR |
382 | } |
383 | ||
384 | int drbd_thread_start(struct drbd_thread *thi) | |
385 | { | |
bde89a9e | 386 | struct drbd_connection *connection = thi->connection; |
b411b363 PR |
387 | struct task_struct *nt; |
388 | unsigned long flags; | |
389 | ||
b411b363 PR |
390 | /* 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) { | |
e77a0a5c | 395 | case NONE: |
bde89a9e | 396 | conn_info(connection, "Starting %s thread (from %s [%d])\n", |
bed879ae | 397 | thi->name, current->comm, current->pid); |
b411b363 PR |
398 | |
399 | /* Get ref on module for thread - this is released when thread exits */ | |
400 | if (!try_module_get(THIS_MODULE)) { | |
bde89a9e | 401 | conn_err(connection, "Failed to get module reference in drbd_thread_start\n"); |
b411b363 | 402 | spin_unlock_irqrestore(&thi->t_lock, flags); |
81e84650 | 403 | return false; |
b411b363 PR |
404 | } |
405 | ||
bde89a9e | 406 | kref_get(&thi->connection->kref); |
9dc9fbb3 | 407 | |
b411b363 | 408 | init_completion(&thi->stop); |
b411b363 | 409 | thi->reset_cpu_mask = 1; |
e77a0a5c | 410 | thi->t_state = RUNNING; |
b411b363 PR |
411 | spin_unlock_irqrestore(&thi->t_lock, flags); |
412 | flush_signals(current); /* otherw. may get -ERESTARTNOINTR */ | |
413 | ||
414 | nt = kthread_create(drbd_thread_setup, (void *) thi, | |
77c556f6 | 415 | "drbd_%c_%s", thi->name[0], thi->connection->resource->name); |
b411b363 PR |
416 | |
417 | if (IS_ERR(nt)) { | |
bde89a9e | 418 | conn_err(connection, "Couldn't start thread\n"); |
b411b363 | 419 | |
05a10ec7 | 420 | kref_put(&connection->kref, drbd_destroy_connection); |
b411b363 | 421 | module_put(THIS_MODULE); |
81e84650 | 422 | return false; |
b411b363 PR |
423 | } |
424 | spin_lock_irqsave(&thi->t_lock, flags); | |
425 | thi->task = nt; | |
e77a0a5c | 426 | thi->t_state = RUNNING; |
b411b363 PR |
427 | spin_unlock_irqrestore(&thi->t_lock, flags); |
428 | wake_up_process(nt); | |
429 | break; | |
e77a0a5c AG |
430 | case EXITING: |
431 | thi->t_state = RESTARTING; | |
bde89a9e | 432 | conn_info(connection, "Restarting %s thread (from %s [%d])\n", |
bed879ae | 433 | thi->name, current->comm, current->pid); |
b411b363 | 434 | /* fall through */ |
e77a0a5c AG |
435 | case RUNNING: |
436 | case RESTARTING: | |
b411b363 PR |
437 | default: |
438 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
439 | break; | |
440 | } | |
441 | ||
81e84650 | 442 | return true; |
b411b363 PR |
443 | } |
444 | ||
445 | ||
446 | void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait) | |
447 | { | |
448 | unsigned long flags; | |
449 | ||
e77a0a5c | 450 | enum drbd_thread_state ns = restart ? RESTARTING : EXITING; |
b411b363 PR |
451 | |
452 | /* may be called from state engine, holding the req lock irqsave */ | |
453 | spin_lock_irqsave(&thi->t_lock, flags); | |
454 | ||
e77a0a5c | 455 | if (thi->t_state == NONE) { |
b411b363 PR |
456 | spin_unlock_irqrestore(&thi->t_lock, flags); |
457 | if (restart) | |
458 | drbd_thread_start(thi); | |
459 | return; | |
460 | } | |
461 | ||
462 | if (thi->t_state != ns) { | |
463 | if (thi->task == NULL) { | |
464 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
465 | return; | |
466 | } | |
467 | ||
468 | thi->t_state = ns; | |
469 | smp_mb(); | |
470 | init_completion(&thi->stop); | |
471 | if (thi->task != current) | |
472 | force_sig(DRBD_SIGKILL, thi->task); | |
b411b363 PR |
473 | } |
474 | ||
475 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
476 | ||
477 | if (wait) | |
478 | wait_for_completion(&thi->stop); | |
479 | } | |
480 | ||
bde89a9e | 481 | static struct drbd_thread *drbd_task_to_thread(struct drbd_connection *connection, struct task_struct *task) |
bed879ae | 482 | { |
bed879ae | 483 | struct drbd_thread *thi = |
bde89a9e AG |
484 | task == connection->receiver.task ? &connection->receiver : |
485 | task == connection->asender.task ? &connection->asender : | |
486 | task == connection->worker.task ? &connection->worker : NULL; | |
bed879ae PR |
487 | |
488 | return thi; | |
489 | } | |
490 | ||
bde89a9e | 491 | char *drbd_task_to_thread_name(struct drbd_connection *connection, struct task_struct *task) |
bed879ae | 492 | { |
bde89a9e | 493 | struct drbd_thread *thi = drbd_task_to_thread(connection, task); |
bed879ae PR |
494 | return thi ? thi->name : task->comm; |
495 | } | |
496 | ||
bde89a9e | 497 | int conn_lowest_minor(struct drbd_connection *connection) |
80822284 | 498 | { |
b30ab791 | 499 | struct drbd_device *device; |
695d08fa | 500 | int vnr = 0, m; |
774b3055 | 501 | |
695d08fa | 502 | rcu_read_lock(); |
bde89a9e | 503 | device = idr_get_next(&connection->volumes, &vnr); |
b30ab791 | 504 | m = device ? device_to_minor(device) : -1; |
695d08fa PR |
505 | rcu_read_unlock(); |
506 | ||
507 | return m; | |
80822284 | 508 | } |
774b3055 | 509 | |
b411b363 PR |
510 | #ifdef CONFIG_SMP |
511 | /** | |
512 | * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs | |
b30ab791 | 513 | * @device: DRBD device. |
b411b363 PR |
514 | * |
515 | * Forces all threads of a device onto the same CPU. This is beneficial for | |
516 | * DRBD's performance. May be overwritten by user's configuration. | |
517 | */ | |
bde89a9e | 518 | void drbd_calc_cpu_mask(struct drbd_connection *connection) |
b411b363 PR |
519 | { |
520 | int ord, cpu; | |
521 | ||
522 | /* user override. */ | |
bde89a9e | 523 | if (cpumask_weight(connection->cpu_mask)) |
b411b363 PR |
524 | return; |
525 | ||
bde89a9e | 526 | ord = conn_lowest_minor(connection) % cpumask_weight(cpu_online_mask); |
b411b363 PR |
527 | for_each_online_cpu(cpu) { |
528 | if (ord-- == 0) { | |
bde89a9e | 529 | cpumask_set_cpu(cpu, connection->cpu_mask); |
b411b363 PR |
530 | return; |
531 | } | |
532 | } | |
533 | /* should not be reached */ | |
bde89a9e | 534 | cpumask_setall(connection->cpu_mask); |
b411b363 PR |
535 | } |
536 | ||
537 | /** | |
538 | * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread | |
b30ab791 | 539 | * @device: DRBD device. |
bc31fe33 | 540 | * @thi: drbd_thread object |
b411b363 PR |
541 | * |
542 | * call in the "main loop" of _all_ threads, no need for any mutex, current won't die | |
543 | * prematurely. | |
544 | */ | |
80822284 | 545 | void drbd_thread_current_set_cpu(struct drbd_thread *thi) |
b411b363 PR |
546 | { |
547 | struct task_struct *p = current; | |
bed879ae | 548 | |
b411b363 PR |
549 | if (!thi->reset_cpu_mask) |
550 | return; | |
551 | thi->reset_cpu_mask = 0; | |
bde89a9e | 552 | set_cpus_allowed_ptr(p, thi->connection->cpu_mask); |
b411b363 PR |
553 | } |
554 | #endif | |
555 | ||
52b061a4 AG |
556 | /** |
557 | * drbd_header_size - size of a packet header | |
558 | * | |
559 | * The header size is a multiple of 8, so any payload following the header is | |
560 | * word aligned on 64-bit architectures. (The bitmap send and receive code | |
561 | * relies on this.) | |
562 | */ | |
bde89a9e | 563 | unsigned int drbd_header_size(struct drbd_connection *connection) |
b411b363 | 564 | { |
bde89a9e | 565 | if (connection->agreed_pro_version >= 100) { |
0c8e36d9 AG |
566 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8)); |
567 | return sizeof(struct p_header100); | |
568 | } else { | |
569 | BUILD_BUG_ON(sizeof(struct p_header80) != | |
570 | sizeof(struct p_header95)); | |
571 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8)); | |
572 | return sizeof(struct p_header80); | |
573 | } | |
52b061a4 | 574 | } |
b411b363 | 575 | |
e658983a | 576 | static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
577 | { |
578 | h->magic = cpu_to_be32(DRBD_MAGIC); | |
579 | h->command = cpu_to_be16(cmd); | |
580 | h->length = cpu_to_be16(size); | |
e658983a | 581 | return sizeof(struct p_header80); |
fd340c12 | 582 | } |
b411b363 | 583 | |
e658983a | 584 | static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
585 | { |
586 | h->magic = cpu_to_be16(DRBD_MAGIC_BIG); | |
b411b363 | 587 | h->command = cpu_to_be16(cmd); |
b55d84ba | 588 | h->length = cpu_to_be32(size); |
e658983a | 589 | return sizeof(struct p_header95); |
fd340c12 | 590 | } |
b411b363 | 591 | |
0c8e36d9 AG |
592 | static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd, |
593 | int size, int vnr) | |
594 | { | |
595 | h->magic = cpu_to_be32(DRBD_MAGIC_100); | |
596 | h->volume = cpu_to_be16(vnr); | |
597 | h->command = cpu_to_be16(cmd); | |
598 | h->length = cpu_to_be32(size); | |
599 | h->pad = 0; | |
600 | return sizeof(struct p_header100); | |
601 | } | |
b411b363 | 602 | |
bde89a9e | 603 | static unsigned int prepare_header(struct drbd_connection *connection, int vnr, |
0c8e36d9 | 604 | void *buffer, enum drbd_packet cmd, int size) |
d38e787e | 605 | { |
bde89a9e | 606 | if (connection->agreed_pro_version >= 100) |
0c8e36d9 | 607 | return prepare_header100(buffer, cmd, size, vnr); |
bde89a9e | 608 | else if (connection->agreed_pro_version >= 95 && |
0c8e36d9 | 609 | size > DRBD_MAX_SIZE_H80_PACKET) |
e658983a | 610 | return prepare_header95(buffer, cmd, size); |
d38e787e | 611 | else |
e658983a | 612 | return prepare_header80(buffer, cmd, size); |
b411b363 PR |
613 | } |
614 | ||
bde89a9e | 615 | static void *__conn_prepare_command(struct drbd_connection *connection, |
a7eb7bdf | 616 | struct drbd_socket *sock) |
b411b363 | 617 | { |
a7eb7bdf AG |
618 | if (!sock->socket) |
619 | return NULL; | |
bde89a9e | 620 | return sock->sbuf + drbd_header_size(connection); |
a7eb7bdf | 621 | } |
b411b363 | 622 | |
bde89a9e | 623 | void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock) |
dba58587 | 624 | { |
a7eb7bdf | 625 | void *p; |
b411b363 | 626 | |
dba58587 | 627 | mutex_lock(&sock->mutex); |
bde89a9e | 628 | p = __conn_prepare_command(connection, sock); |
a7eb7bdf | 629 | if (!p) |
dba58587 | 630 | mutex_unlock(&sock->mutex); |
b411b363 | 631 | |
a7eb7bdf | 632 | return p; |
b411b363 PR |
633 | } |
634 | ||
b30ab791 | 635 | void *drbd_prepare_command(struct drbd_device *device, struct drbd_socket *sock) |
b411b363 | 636 | { |
a6b32bc3 | 637 | return conn_prepare_command(first_peer_device(device)->connection, sock); |
dba58587 | 638 | } |
b411b363 | 639 | |
bde89a9e | 640 | static int __send_command(struct drbd_connection *connection, int vnr, |
dba58587 AG |
641 | struct drbd_socket *sock, enum drbd_packet cmd, |
642 | unsigned int header_size, void *data, | |
643 | unsigned int size) | |
644 | { | |
645 | int msg_flags; | |
646 | int err; | |
b411b363 | 647 | |
dba58587 AG |
648 | /* |
649 | * Called with @data == NULL and the size of the data blocks in @size | |
650 | * for commands that send data blocks. For those commands, omit the | |
651 | * MSG_MORE flag: this will increase the likelihood that data blocks | |
652 | * which are page aligned on the sender will end up page aligned on the | |
653 | * receiver. | |
654 | */ | |
655 | msg_flags = data ? MSG_MORE : 0; | |
656 | ||
bde89a9e | 657 | header_size += prepare_header(connection, vnr, sock->sbuf, cmd, |
e658983a | 658 | header_size + size); |
bde89a9e | 659 | err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size, |
dba58587 AG |
660 | msg_flags); |
661 | if (data && !err) | |
bde89a9e | 662 | err = drbd_send_all(connection, sock->socket, data, size, 0); |
dba58587 AG |
663 | return err; |
664 | } | |
665 | ||
bde89a9e | 666 | static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
a7eb7bdf AG |
667 | enum drbd_packet cmd, unsigned int header_size, |
668 | void *data, unsigned int size) | |
669 | { | |
bde89a9e | 670 | return __send_command(connection, 0, sock, cmd, header_size, data, size); |
a7eb7bdf AG |
671 | } |
672 | ||
bde89a9e | 673 | int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
dba58587 AG |
674 | enum drbd_packet cmd, unsigned int header_size, |
675 | void *data, unsigned int size) | |
676 | { | |
677 | int err; | |
b411b363 | 678 | |
bde89a9e | 679 | err = __conn_send_command(connection, sock, cmd, header_size, data, size); |
dba58587 AG |
680 | mutex_unlock(&sock->mutex); |
681 | return err; | |
682 | } | |
683 | ||
b30ab791 | 684 | int drbd_send_command(struct drbd_device *device, struct drbd_socket *sock, |
dba58587 AG |
685 | enum drbd_packet cmd, unsigned int header_size, |
686 | void *data, unsigned int size) | |
687 | { | |
688 | int err; | |
689 | ||
a6b32bc3 | 690 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, cmd, header_size, |
dba58587 AG |
691 | data, size); |
692 | mutex_unlock(&sock->mutex); | |
693 | return err; | |
694 | } | |
b411b363 | 695 | |
bde89a9e | 696 | int drbd_send_ping(struct drbd_connection *connection) |
e307f352 | 697 | { |
9f5bdc33 AG |
698 | struct drbd_socket *sock; |
699 | ||
bde89a9e AG |
700 | sock = &connection->meta; |
701 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 702 | return -EIO; |
bde89a9e | 703 | return conn_send_command(connection, sock, P_PING, 0, NULL, 0); |
e307f352 | 704 | } |
b411b363 | 705 | |
bde89a9e | 706 | int drbd_send_ping_ack(struct drbd_connection *connection) |
e307f352 | 707 | { |
9f5bdc33 AG |
708 | struct drbd_socket *sock; |
709 | ||
bde89a9e AG |
710 | sock = &connection->meta; |
711 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 712 | return -EIO; |
bde89a9e | 713 | return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0); |
b411b363 PR |
714 | } |
715 | ||
b30ab791 | 716 | int drbd_send_sync_param(struct drbd_device *device) |
b411b363 | 717 | { |
7c96715a | 718 | struct drbd_socket *sock; |
8e26f9cc | 719 | struct p_rs_param_95 *p; |
9f5bdc33 | 720 | int size; |
a6b32bc3 | 721 | const int apv = first_peer_device(device)->connection->agreed_pro_version; |
9f5bdc33 | 722 | enum drbd_packet cmd; |
44ed167d | 723 | struct net_conf *nc; |
daeda1cc | 724 | struct disk_conf *dc; |
9f5bdc33 | 725 | |
a6b32bc3 | 726 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 727 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
728 | if (!p) |
729 | return -EIO; | |
b411b363 | 730 | |
44ed167d | 731 | rcu_read_lock(); |
a6b32bc3 | 732 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
b411b363 PR |
733 | |
734 | size = apv <= 87 ? sizeof(struct p_rs_param) | |
735 | : apv == 88 ? sizeof(struct p_rs_param) | |
44ed167d | 736 | + strlen(nc->verify_alg) + 1 |
8e26f9cc PR |
737 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
738 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 739 | |
9f5bdc33 | 740 | cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM; |
b411b363 | 741 | |
9f5bdc33 AG |
742 | /* initialize verify_alg and csums_alg */ |
743 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
b411b363 | 744 | |
b30ab791 AG |
745 | if (get_ldev(device)) { |
746 | dc = rcu_dereference(device->ldev->disk_conf); | |
6394b935 | 747 | p->resync_rate = cpu_to_be32(dc->resync_rate); |
daeda1cc PR |
748 | p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead); |
749 | p->c_delay_target = cpu_to_be32(dc->c_delay_target); | |
750 | p->c_fill_target = cpu_to_be32(dc->c_fill_target); | |
751 | p->c_max_rate = cpu_to_be32(dc->c_max_rate); | |
b30ab791 | 752 | put_ldev(device); |
9f5bdc33 | 753 | } else { |
6394b935 | 754 | p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF); |
9f5bdc33 AG |
755 | p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF); |
756 | p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF); | |
757 | p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF); | |
758 | p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF); | |
759 | } | |
b411b363 | 760 | |
9f5bdc33 | 761 | if (apv >= 88) |
44ed167d | 762 | strcpy(p->verify_alg, nc->verify_alg); |
9f5bdc33 | 763 | if (apv >= 89) |
44ed167d PR |
764 | strcpy(p->csums_alg, nc->csums_alg); |
765 | rcu_read_unlock(); | |
b411b363 | 766 | |
b30ab791 | 767 | return drbd_send_command(device, sock, cmd, size, NULL, 0); |
b411b363 PR |
768 | } |
769 | ||
bde89a9e | 770 | int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd) |
b411b363 | 771 | { |
9f5bdc33 | 772 | struct drbd_socket *sock; |
b411b363 | 773 | struct p_protocol *p; |
44ed167d | 774 | struct net_conf *nc; |
9f5bdc33 | 775 | int size, cf; |
b411b363 | 776 | |
bde89a9e AG |
777 | sock = &connection->data; |
778 | p = __conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
779 | if (!p) |
780 | return -EIO; | |
b411b363 | 781 | |
44ed167d | 782 | rcu_read_lock(); |
bde89a9e | 783 | nc = rcu_dereference(connection->net_conf); |
b411b363 | 784 | |
bde89a9e | 785 | if (nc->tentative && connection->agreed_pro_version < 92) { |
44ed167d PR |
786 | rcu_read_unlock(); |
787 | mutex_unlock(&sock->mutex); | |
bde89a9e | 788 | conn_err(connection, "--dry-run is not supported by peer"); |
44ed167d PR |
789 | return -EOPNOTSUPP; |
790 | } | |
b411b363 | 791 | |
9f5bdc33 | 792 | size = sizeof(*p); |
bde89a9e | 793 | if (connection->agreed_pro_version >= 87) |
44ed167d | 794 | size += strlen(nc->integrity_alg) + 1; |
b411b363 | 795 | |
44ed167d PR |
796 | p->protocol = cpu_to_be32(nc->wire_protocol); |
797 | p->after_sb_0p = cpu_to_be32(nc->after_sb_0p); | |
798 | p->after_sb_1p = cpu_to_be32(nc->after_sb_1p); | |
799 | p->after_sb_2p = cpu_to_be32(nc->after_sb_2p); | |
800 | p->two_primaries = cpu_to_be32(nc->two_primaries); | |
cf14c2e9 | 801 | cf = 0; |
6139f60d AG |
802 | if (nc->discard_my_data) |
803 | cf |= CF_DISCARD_MY_DATA; | |
6dff2902 | 804 | if (nc->tentative) |
9f5bdc33 | 805 | cf |= CF_DRY_RUN; |
cf14c2e9 PR |
806 | p->conn_flags = cpu_to_be32(cf); |
807 | ||
bde89a9e | 808 | if (connection->agreed_pro_version >= 87) |
44ed167d PR |
809 | strcpy(p->integrity_alg, nc->integrity_alg); |
810 | rcu_read_unlock(); | |
b411b363 | 811 | |
bde89a9e | 812 | return __conn_send_command(connection, sock, cmd, size, NULL, 0); |
a7eb7bdf AG |
813 | } |
814 | ||
bde89a9e | 815 | int drbd_send_protocol(struct drbd_connection *connection) |
a7eb7bdf AG |
816 | { |
817 | int err; | |
818 | ||
bde89a9e AG |
819 | mutex_lock(&connection->data.mutex); |
820 | err = __drbd_send_protocol(connection, P_PROTOCOL); | |
821 | mutex_unlock(&connection->data.mutex); | |
a7eb7bdf AG |
822 | |
823 | return err; | |
b411b363 PR |
824 | } |
825 | ||
b30ab791 | 826 | static int _drbd_send_uuids(struct drbd_device *device, u64 uuid_flags) |
b411b363 | 827 | { |
9f5bdc33 AG |
828 | struct drbd_socket *sock; |
829 | struct p_uuids *p; | |
b411b363 PR |
830 | int i; |
831 | ||
b30ab791 | 832 | if (!get_ldev_if_state(device, D_NEGOTIATING)) |
2ae5f95b | 833 | return 0; |
b411b363 | 834 | |
a6b32bc3 | 835 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 836 | p = drbd_prepare_command(device, sock); |
9f5bdc33 | 837 | if (!p) { |
b30ab791 | 838 | put_ldev(device); |
9f5bdc33 AG |
839 | return -EIO; |
840 | } | |
b30ab791 | 841 | spin_lock_irq(&device->ldev->md.uuid_lock); |
b411b363 | 842 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
843 | p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
844 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 845 | |
b30ab791 AG |
846 | device->comm_bm_set = drbd_bm_total_weight(device); |
847 | p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set); | |
44ed167d | 848 | rcu_read_lock(); |
a6b32bc3 | 849 | uuid_flags |= rcu_dereference(first_peer_device(device)->connection->net_conf)->discard_my_data ? 1 : 0; |
44ed167d | 850 | rcu_read_unlock(); |
b30ab791 AG |
851 | uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0; |
852 | uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0; | |
9f5bdc33 | 853 | p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags); |
b411b363 | 854 | |
b30ab791 AG |
855 | put_ldev(device); |
856 | return drbd_send_command(device, sock, P_UUIDS, sizeof(*p), NULL, 0); | |
b411b363 PR |
857 | } |
858 | ||
b30ab791 | 859 | int drbd_send_uuids(struct drbd_device *device) |
b411b363 | 860 | { |
b30ab791 | 861 | return _drbd_send_uuids(device, 0); |
b411b363 PR |
862 | } |
863 | ||
b30ab791 | 864 | int drbd_send_uuids_skip_initial_sync(struct drbd_device *device) |
b411b363 | 865 | { |
b30ab791 | 866 | return _drbd_send_uuids(device, 8); |
b411b363 PR |
867 | } |
868 | ||
b30ab791 | 869 | void drbd_print_uuids(struct drbd_device *device, const char *text) |
62b0da3a | 870 | { |
b30ab791 AG |
871 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
872 | u64 *uuid = device->ldev->md.uuid; | |
62b0da3a LE |
873 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX\n", |
874 | text, | |
875 | (unsigned long long)uuid[UI_CURRENT], | |
876 | (unsigned long long)uuid[UI_BITMAP], | |
877 | (unsigned long long)uuid[UI_HISTORY_START], | |
878 | (unsigned long long)uuid[UI_HISTORY_END]); | |
b30ab791 | 879 | put_ldev(device); |
62b0da3a LE |
880 | } else { |
881 | dev_info(DEV, "%s effective data uuid: %016llX\n", | |
882 | text, | |
b30ab791 | 883 | (unsigned long long)device->ed_uuid); |
62b0da3a LE |
884 | } |
885 | } | |
886 | ||
b30ab791 | 887 | void drbd_gen_and_send_sync_uuid(struct drbd_device *device) |
b411b363 | 888 | { |
9f5bdc33 AG |
889 | struct drbd_socket *sock; |
890 | struct p_rs_uuid *p; | |
5a22db89 LE |
891 | u64 uuid; |
892 | ||
b30ab791 | 893 | D_ASSERT(device->state.disk == D_UP_TO_DATE); |
b411b363 | 894 | |
b30ab791 | 895 | uuid = device->ldev->md.uuid[UI_BITMAP]; |
5ba3dac5 PR |
896 | if (uuid && uuid != UUID_JUST_CREATED) |
897 | uuid = uuid + UUID_NEW_BM_OFFSET; | |
898 | else | |
899 | get_random_bytes(&uuid, sizeof(u64)); | |
b30ab791 AG |
900 | drbd_uuid_set(device, UI_BITMAP, uuid); |
901 | drbd_print_uuids(device, "updated sync UUID"); | |
902 | drbd_md_sync(device); | |
b411b363 | 903 | |
a6b32bc3 | 904 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 905 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
906 | if (p) { |
907 | p->uuid = cpu_to_be64(uuid); | |
b30ab791 | 908 | drbd_send_command(device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0); |
9f5bdc33 | 909 | } |
b411b363 PR |
910 | } |
911 | ||
b30ab791 | 912 | int drbd_send_sizes(struct drbd_device *device, int trigger_reply, enum dds_flags flags) |
b411b363 | 913 | { |
9f5bdc33 AG |
914 | struct drbd_socket *sock; |
915 | struct p_sizes *p; | |
b411b363 | 916 | sector_t d_size, u_size; |
db141b2f LE |
917 | int q_order_type; |
918 | unsigned int max_bio_size; | |
b411b363 | 919 | |
b30ab791 AG |
920 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
921 | D_ASSERT(device->ldev->backing_bdev); | |
922 | d_size = drbd_get_max_capacity(device->ldev); | |
daeda1cc | 923 | rcu_read_lock(); |
b30ab791 | 924 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
daeda1cc | 925 | rcu_read_unlock(); |
b30ab791 AG |
926 | q_order_type = drbd_queue_order_type(device); |
927 | max_bio_size = queue_max_hw_sectors(device->ldev->backing_bdev->bd_disk->queue) << 9; | |
db141b2f | 928 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE); |
b30ab791 | 929 | put_ldev(device); |
b411b363 PR |
930 | } else { |
931 | d_size = 0; | |
932 | u_size = 0; | |
933 | q_order_type = QUEUE_ORDERED_NONE; | |
99432fcc | 934 | max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */ |
b411b363 PR |
935 | } |
936 | ||
a6b32bc3 | 937 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 938 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
939 | if (!p) |
940 | return -EIO; | |
6809384c | 941 | |
a6b32bc3 | 942 | if (first_peer_device(device)->connection->agreed_pro_version <= 94) |
98683650 | 943 | max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET); |
a6b32bc3 | 944 | else if (first_peer_device(device)->connection->agreed_pro_version < 100) |
98683650 | 945 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95); |
b411b363 | 946 | |
9f5bdc33 AG |
947 | p->d_size = cpu_to_be64(d_size); |
948 | p->u_size = cpu_to_be64(u_size); | |
b30ab791 | 949 | p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev)); |
9f5bdc33 AG |
950 | p->max_bio_size = cpu_to_be32(max_bio_size); |
951 | p->queue_order_type = cpu_to_be16(q_order_type); | |
952 | p->dds_flags = cpu_to_be16(flags); | |
b30ab791 | 953 | return drbd_send_command(device, sock, P_SIZES, sizeof(*p), NULL, 0); |
b411b363 PR |
954 | } |
955 | ||
956 | /** | |
f479ea06 | 957 | * drbd_send_current_state() - Sends the drbd state to the peer |
b30ab791 | 958 | * @device: DRBD device. |
b411b363 | 959 | */ |
b30ab791 | 960 | int drbd_send_current_state(struct drbd_device *device) |
b411b363 | 961 | { |
7c96715a | 962 | struct drbd_socket *sock; |
9f5bdc33 | 963 | struct p_state *p; |
b411b363 | 964 | |
a6b32bc3 | 965 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 966 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
967 | if (!p) |
968 | return -EIO; | |
b30ab791 AG |
969 | p->state = cpu_to_be32(device->state.i); /* Within the send mutex */ |
970 | return drbd_send_command(device, sock, P_STATE, sizeof(*p), NULL, 0); | |
b411b363 PR |
971 | } |
972 | ||
f479ea06 LE |
973 | /** |
974 | * drbd_send_state() - After a state change, sends the new state to the peer | |
b30ab791 | 975 | * @device: DRBD device. |
43de7c85 | 976 | * @state: the state to send, not necessarily the current state. |
f479ea06 LE |
977 | * |
978 | * Each state change queues an "after_state_ch" work, which will eventually | |
979 | * send the resulting new state to the peer. If more state changes happen | |
980 | * between queuing and processing of the after_state_ch work, we still | |
981 | * want to send each intermediary state in the order it occurred. | |
982 | */ | |
b30ab791 | 983 | int drbd_send_state(struct drbd_device *device, union drbd_state state) |
f479ea06 | 984 | { |
43de7c85 PR |
985 | struct drbd_socket *sock; |
986 | struct p_state *p; | |
f479ea06 | 987 | |
a6b32bc3 | 988 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 989 | p = drbd_prepare_command(device, sock); |
43de7c85 PR |
990 | if (!p) |
991 | return -EIO; | |
992 | p->state = cpu_to_be32(state.i); /* Within the send mutex */ | |
b30ab791 | 993 | return drbd_send_command(device, sock, P_STATE, sizeof(*p), NULL, 0); |
43de7c85 | 994 | } |
f479ea06 | 995 | |
b30ab791 | 996 | int drbd_send_state_req(struct drbd_device *device, union drbd_state mask, union drbd_state val) |
9f5bdc33 AG |
997 | { |
998 | struct drbd_socket *sock; | |
999 | struct p_req_state *p; | |
f479ea06 | 1000 | |
a6b32bc3 | 1001 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1002 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1003 | if (!p) |
1004 | return -EIO; | |
1005 | p->mask = cpu_to_be32(mask.i); | |
1006 | p->val = cpu_to_be32(val.i); | |
b30ab791 | 1007 | return drbd_send_command(device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0); |
b411b363 | 1008 | } |
f479ea06 | 1009 | |
bde89a9e | 1010 | int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val) |
b411b363 | 1011 | { |
9f5bdc33 AG |
1012 | enum drbd_packet cmd; |
1013 | struct drbd_socket *sock; | |
1014 | struct p_req_state *p; | |
f479ea06 | 1015 | |
bde89a9e AG |
1016 | cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ; |
1017 | sock = &connection->data; | |
1018 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1019 | if (!p) |
1020 | return -EIO; | |
1021 | p->mask = cpu_to_be32(mask.i); | |
1022 | p->val = cpu_to_be32(val.i); | |
bde89a9e | 1023 | return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
f479ea06 LE |
1024 | } |
1025 | ||
b30ab791 | 1026 | void drbd_send_sr_reply(struct drbd_device *device, enum drbd_state_rv retcode) |
b411b363 | 1027 | { |
9f5bdc33 AG |
1028 | struct drbd_socket *sock; |
1029 | struct p_req_state_reply *p; | |
b411b363 | 1030 | |
a6b32bc3 | 1031 | sock = &first_peer_device(device)->connection->meta; |
b30ab791 | 1032 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1033 | if (p) { |
1034 | p->retcode = cpu_to_be32(retcode); | |
b30ab791 | 1035 | drbd_send_command(device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0); |
9f5bdc33 | 1036 | } |
b411b363 | 1037 | } |
b411b363 | 1038 | |
bde89a9e | 1039 | void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode) |
047cd4a6 | 1040 | { |
9f5bdc33 AG |
1041 | struct drbd_socket *sock; |
1042 | struct p_req_state_reply *p; | |
bde89a9e | 1043 | enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY; |
b411b363 | 1044 | |
bde89a9e AG |
1045 | sock = &connection->meta; |
1046 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1047 | if (p) { |
1048 | p->retcode = cpu_to_be32(retcode); | |
bde89a9e | 1049 | conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
9f5bdc33 | 1050 | } |
b411b363 PR |
1051 | } |
1052 | ||
a02d1240 | 1053 | static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) |
b411b363 | 1054 | { |
a02d1240 AG |
1055 | BUG_ON(code & ~0xf); |
1056 | p->encoding = (p->encoding & ~0xf) | code; | |
1057 | } | |
b411b363 | 1058 | |
a02d1240 AG |
1059 | static void dcbp_set_start(struct p_compressed_bm *p, int set) |
1060 | { | |
1061 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
1062 | } | |
b411b363 | 1063 | |
a02d1240 AG |
1064 | static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n) |
1065 | { | |
1066 | BUG_ON(n & ~0x7); | |
1067 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
b411b363 PR |
1068 | } |
1069 | ||
b30ab791 | 1070 | static int fill_bitmap_rle_bits(struct drbd_device *device, |
50d0b1ad AG |
1071 | struct p_compressed_bm *p, |
1072 | unsigned int size, | |
1073 | struct bm_xfer_ctx *c) | |
b411b363 PR |
1074 | { |
1075 | struct bitstream bs; | |
1076 | unsigned long plain_bits; | |
1077 | unsigned long tmp; | |
1078 | unsigned long rl; | |
1079 | unsigned len; | |
1080 | unsigned toggle; | |
44ed167d | 1081 | int bits, use_rle; |
b411b363 PR |
1082 | |
1083 | /* may we use this feature? */ | |
44ed167d | 1084 | rcu_read_lock(); |
a6b32bc3 | 1085 | use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle; |
44ed167d | 1086 | rcu_read_unlock(); |
a6b32bc3 | 1087 | if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90) |
44ed167d | 1088 | return 0; |
b411b363 PR |
1089 | |
1090 | if (c->bit_offset >= c->bm_bits) | |
1091 | return 0; /* nothing to do. */ | |
1092 | ||
1093 | /* use at most thus many bytes */ | |
50d0b1ad AG |
1094 | bitstream_init(&bs, p->code, size, 0); |
1095 | memset(p->code, 0, size); | |
b411b363 PR |
1096 | /* plain bits covered in this code string */ |
1097 | plain_bits = 0; | |
1098 | ||
1099 | /* p->encoding & 0x80 stores whether the first run length is set. | |
1100 | * bit offset is implicit. | |
1101 | * start with toggle == 2 to be able to tell the first iteration */ | |
1102 | toggle = 2; | |
1103 | ||
1104 | /* see how much plain bits we can stuff into one packet | |
1105 | * using RLE and VLI. */ | |
1106 | do { | |
b30ab791 AG |
1107 | tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset) |
1108 | : _drbd_bm_find_next(device, c->bit_offset); | |
b411b363 PR |
1109 | if (tmp == -1UL) |
1110 | tmp = c->bm_bits; | |
1111 | rl = tmp - c->bit_offset; | |
1112 | ||
1113 | if (toggle == 2) { /* first iteration */ | |
1114 | if (rl == 0) { | |
1115 | /* the first checked bit was set, | |
1116 | * store start value, */ | |
a02d1240 | 1117 | dcbp_set_start(p, 1); |
b411b363 PR |
1118 | /* but skip encoding of zero run length */ |
1119 | toggle = !toggle; | |
1120 | continue; | |
1121 | } | |
a02d1240 | 1122 | dcbp_set_start(p, 0); |
b411b363 PR |
1123 | } |
1124 | ||
1125 | /* paranoia: catch zero runlength. | |
1126 | * can only happen if bitmap is modified while we scan it. */ | |
1127 | if (rl == 0) { | |
1128 | dev_err(DEV, "unexpected zero runlength while encoding bitmap " | |
1129 | "t:%u bo:%lu\n", toggle, c->bit_offset); | |
1130 | return -1; | |
1131 | } | |
1132 | ||
1133 | bits = vli_encode_bits(&bs, rl); | |
1134 | if (bits == -ENOBUFS) /* buffer full */ | |
1135 | break; | |
1136 | if (bits <= 0) { | |
1137 | dev_err(DEV, "error while encoding bitmap: %d\n", bits); | |
1138 | return 0; | |
1139 | } | |
1140 | ||
1141 | toggle = !toggle; | |
1142 | plain_bits += rl; | |
1143 | c->bit_offset = tmp; | |
1144 | } while (c->bit_offset < c->bm_bits); | |
1145 | ||
1146 | len = bs.cur.b - p->code + !!bs.cur.bit; | |
1147 | ||
1148 | if (plain_bits < (len << 3)) { | |
1149 | /* incompressible with this method. | |
1150 | * we need to rewind both word and bit position. */ | |
1151 | c->bit_offset -= plain_bits; | |
1152 | bm_xfer_ctx_bit_to_word_offset(c); | |
1153 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1154 | return 0; | |
1155 | } | |
1156 | ||
1157 | /* RLE + VLI was able to compress it just fine. | |
1158 | * update c->word_offset. */ | |
1159 | bm_xfer_ctx_bit_to_word_offset(c); | |
1160 | ||
1161 | /* store pad_bits */ | |
a02d1240 | 1162 | dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7); |
b411b363 PR |
1163 | |
1164 | return len; | |
1165 | } | |
1166 | ||
f70af118 AG |
1167 | /** |
1168 | * send_bitmap_rle_or_plain | |
1169 | * | |
1170 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
1171 | * code upon failure. | |
1172 | */ | |
1173 | static int | |
b30ab791 | 1174 | send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c) |
b411b363 | 1175 | { |
a6b32bc3 AG |
1176 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
1177 | unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); | |
e658983a | 1178 | struct p_compressed_bm *p = sock->sbuf + header_size; |
a982dd57 | 1179 | int len, err; |
b411b363 | 1180 | |
b30ab791 | 1181 | len = fill_bitmap_rle_bits(device, p, |
e658983a | 1182 | DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c); |
b411b363 | 1183 | if (len < 0) |
f70af118 | 1184 | return -EIO; |
b411b363 PR |
1185 | |
1186 | if (len) { | |
a02d1240 | 1187 | dcbp_set_code(p, RLE_VLI_Bits); |
a6b32bc3 | 1188 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, |
9f5bdc33 AG |
1189 | P_COMPRESSED_BITMAP, sizeof(*p) + len, |
1190 | NULL, 0); | |
b411b363 | 1191 | c->packets[0]++; |
e658983a | 1192 | c->bytes[0] += header_size + sizeof(*p) + len; |
b411b363 PR |
1193 | |
1194 | if (c->bit_offset >= c->bm_bits) | |
1195 | len = 0; /* DONE */ | |
1196 | } else { | |
1197 | /* was not compressible. | |
1198 | * send a buffer full of plain text bits instead. */ | |
50d0b1ad AG |
1199 | unsigned int data_size; |
1200 | unsigned long num_words; | |
e658983a | 1201 | unsigned long *p = sock->sbuf + header_size; |
50d0b1ad AG |
1202 | |
1203 | data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; | |
e658983a | 1204 | num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 1205 | c->bm_words - c->word_offset); |
e658983a | 1206 | len = num_words * sizeof(*p); |
b411b363 | 1207 | if (len) |
b30ab791 | 1208 | drbd_bm_get_lel(device, c->word_offset, num_words, p); |
a6b32bc3 | 1209 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0); |
b411b363 PR |
1210 | c->word_offset += num_words; |
1211 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1212 | ||
1213 | c->packets[1]++; | |
50d0b1ad | 1214 | c->bytes[1] += header_size + len; |
b411b363 PR |
1215 | |
1216 | if (c->bit_offset > c->bm_bits) | |
1217 | c->bit_offset = c->bm_bits; | |
1218 | } | |
a982dd57 | 1219 | if (!err) { |
f70af118 | 1220 | if (len == 0) { |
b30ab791 | 1221 | INFO_bm_xfer_stats(device, "send", c); |
f70af118 AG |
1222 | return 0; |
1223 | } else | |
1224 | return 1; | |
1225 | } | |
1226 | return -EIO; | |
b411b363 PR |
1227 | } |
1228 | ||
1229 | /* See the comment at receive_bitmap() */ | |
b30ab791 | 1230 | static int _drbd_send_bitmap(struct drbd_device *device) |
b411b363 PR |
1231 | { |
1232 | struct bm_xfer_ctx c; | |
f70af118 | 1233 | int err; |
b411b363 | 1234 | |
b30ab791 | 1235 | if (!expect(device->bitmap)) |
81e84650 | 1236 | return false; |
b411b363 | 1237 | |
b30ab791 AG |
1238 | if (get_ldev(device)) { |
1239 | if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) { | |
b411b363 | 1240 | dev_info(DEV, "Writing the whole bitmap, MDF_FullSync was set.\n"); |
b30ab791 AG |
1241 | drbd_bm_set_all(device); |
1242 | if (drbd_bm_write(device)) { | |
b411b363 PR |
1243 | /* write_bm did fail! Leave full sync flag set in Meta P_DATA |
1244 | * but otherwise process as per normal - need to tell other | |
1245 | * side that a full resync is required! */ | |
1246 | dev_err(DEV, "Failed to write bitmap to disk!\n"); | |
1247 | } else { | |
b30ab791 AG |
1248 | drbd_md_clear_flag(device, MDF_FULL_SYNC); |
1249 | drbd_md_sync(device); | |
b411b363 PR |
1250 | } |
1251 | } | |
b30ab791 | 1252 | put_ldev(device); |
b411b363 PR |
1253 | } |
1254 | ||
1255 | c = (struct bm_xfer_ctx) { | |
b30ab791 AG |
1256 | .bm_bits = drbd_bm_bits(device), |
1257 | .bm_words = drbd_bm_words(device), | |
b411b363 PR |
1258 | }; |
1259 | ||
1260 | do { | |
b30ab791 | 1261 | err = send_bitmap_rle_or_plain(device, &c); |
f70af118 | 1262 | } while (err > 0); |
b411b363 | 1263 | |
f70af118 | 1264 | return err == 0; |
b411b363 PR |
1265 | } |
1266 | ||
b30ab791 | 1267 | int drbd_send_bitmap(struct drbd_device *device) |
b411b363 | 1268 | { |
a6b32bc3 | 1269 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
9f5bdc33 | 1270 | int err = -1; |
b411b363 | 1271 | |
9f5bdc33 AG |
1272 | mutex_lock(&sock->mutex); |
1273 | if (sock->socket) | |
b30ab791 | 1274 | err = !_drbd_send_bitmap(device); |
9f5bdc33 | 1275 | mutex_unlock(&sock->mutex); |
b411b363 PR |
1276 | return err; |
1277 | } | |
1278 | ||
bde89a9e | 1279 | void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size) |
b411b363 | 1280 | { |
9f5bdc33 AG |
1281 | struct drbd_socket *sock; |
1282 | struct p_barrier_ack *p; | |
b411b363 | 1283 | |
bde89a9e | 1284 | if (connection->cstate < C_WF_REPORT_PARAMS) |
9f5bdc33 | 1285 | return; |
b411b363 | 1286 | |
bde89a9e AG |
1287 | sock = &connection->meta; |
1288 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1289 | if (!p) |
1290 | return; | |
1291 | p->barrier = barrier_nr; | |
1292 | p->set_size = cpu_to_be32(set_size); | |
bde89a9e | 1293 | conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0); |
b411b363 PR |
1294 | } |
1295 | ||
1296 | /** | |
1297 | * _drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1298 | * @device: DRBD device. |
b411b363 PR |
1299 | * @cmd: Packet command code. |
1300 | * @sector: sector, needs to be in big endian byte order | |
1301 | * @blksize: size in byte, needs to be in big endian byte order | |
1302 | * @block_id: Id, big endian byte order | |
1303 | */ | |
b30ab791 | 1304 | static int _drbd_send_ack(struct drbd_device *device, enum drbd_packet cmd, |
d8763023 | 1305 | u64 sector, u32 blksize, u64 block_id) |
b411b363 | 1306 | { |
9f5bdc33 AG |
1307 | struct drbd_socket *sock; |
1308 | struct p_block_ack *p; | |
b411b363 | 1309 | |
b30ab791 | 1310 | if (device->state.conn < C_CONNECTED) |
9f5bdc33 | 1311 | return -EIO; |
b411b363 | 1312 | |
a6b32bc3 | 1313 | sock = &first_peer_device(device)->connection->meta; |
b30ab791 | 1314 | p = drbd_prepare_command(device, sock); |
9f5bdc33 | 1315 | if (!p) |
a8c32aa8 | 1316 | return -EIO; |
9f5bdc33 AG |
1317 | p->sector = sector; |
1318 | p->block_id = block_id; | |
1319 | p->blksize = blksize; | |
b30ab791 AG |
1320 | p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); |
1321 | return drbd_send_command(device, sock, cmd, sizeof(*p), NULL, 0); | |
b411b363 PR |
1322 | } |
1323 | ||
2b2bf214 LE |
1324 | /* dp->sector and dp->block_id already/still in network byte order, |
1325 | * data_size is payload size according to dp->head, | |
1326 | * and may need to be corrected for digest size. */ | |
b30ab791 | 1327 | void drbd_send_ack_dp(struct drbd_device *device, enum drbd_packet cmd, |
a9a9994d | 1328 | struct p_data *dp, int data_size) |
b411b363 | 1329 | { |
a6b32bc3 AG |
1330 | if (first_peer_device(device)->connection->peer_integrity_tfm) |
1331 | data_size -= crypto_hash_digestsize(first_peer_device(device)->connection->peer_integrity_tfm); | |
b30ab791 | 1332 | _drbd_send_ack(device, cmd, dp->sector, cpu_to_be32(data_size), |
a9a9994d | 1333 | dp->block_id); |
b411b363 PR |
1334 | } |
1335 | ||
b30ab791 | 1336 | void drbd_send_ack_rp(struct drbd_device *device, enum drbd_packet cmd, |
a9a9994d | 1337 | struct p_block_req *rp) |
b411b363 | 1338 | { |
b30ab791 | 1339 | _drbd_send_ack(device, cmd, rp->sector, rp->blksize, rp->block_id); |
b411b363 PR |
1340 | } |
1341 | ||
1342 | /** | |
1343 | * drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1344 | * @device: DRBD device |
db830c46 AG |
1345 | * @cmd: packet command code |
1346 | * @peer_req: peer request | |
b411b363 | 1347 | */ |
b30ab791 | 1348 | int drbd_send_ack(struct drbd_device *device, enum drbd_packet cmd, |
db830c46 | 1349 | struct drbd_peer_request *peer_req) |
b411b363 | 1350 | { |
b30ab791 | 1351 | return _drbd_send_ack(device, cmd, |
dd516121 AG |
1352 | cpu_to_be64(peer_req->i.sector), |
1353 | cpu_to_be32(peer_req->i.size), | |
1354 | peer_req->block_id); | |
b411b363 PR |
1355 | } |
1356 | ||
1357 | /* This function misuses the block_id field to signal if the blocks | |
1358 | * are is sync or not. */ | |
b30ab791 | 1359 | int drbd_send_ack_ex(struct drbd_device *device, enum drbd_packet cmd, |
b411b363 PR |
1360 | sector_t sector, int blksize, u64 block_id) |
1361 | { | |
b30ab791 | 1362 | return _drbd_send_ack(device, cmd, |
b411b363 PR |
1363 | cpu_to_be64(sector), |
1364 | cpu_to_be32(blksize), | |
1365 | cpu_to_be64(block_id)); | |
1366 | } | |
1367 | ||
b30ab791 | 1368 | int drbd_send_drequest(struct drbd_device *device, int cmd, |
b411b363 PR |
1369 | sector_t sector, int size, u64 block_id) |
1370 | { | |
9f5bdc33 AG |
1371 | struct drbd_socket *sock; |
1372 | struct p_block_req *p; | |
b411b363 | 1373 | |
a6b32bc3 | 1374 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1375 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1376 | if (!p) |
1377 | return -EIO; | |
1378 | p->sector = cpu_to_be64(sector); | |
1379 | p->block_id = block_id; | |
1380 | p->blksize = cpu_to_be32(size); | |
b30ab791 | 1381 | return drbd_send_command(device, sock, cmd, sizeof(*p), NULL, 0); |
b411b363 PR |
1382 | } |
1383 | ||
b30ab791 | 1384 | int drbd_send_drequest_csum(struct drbd_device *device, sector_t sector, int size, |
d8763023 | 1385 | void *digest, int digest_size, enum drbd_packet cmd) |
b411b363 | 1386 | { |
9f5bdc33 AG |
1387 | struct drbd_socket *sock; |
1388 | struct p_block_req *p; | |
b411b363 | 1389 | |
9f5bdc33 | 1390 | /* FIXME: Put the digest into the preallocated socket buffer. */ |
b411b363 | 1391 | |
a6b32bc3 | 1392 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1393 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1394 | if (!p) |
1395 | return -EIO; | |
1396 | p->sector = cpu_to_be64(sector); | |
1397 | p->block_id = ID_SYNCER /* unused */; | |
1398 | p->blksize = cpu_to_be32(size); | |
b30ab791 | 1399 | return drbd_send_command(device, sock, cmd, sizeof(*p), |
9f5bdc33 | 1400 | digest, digest_size); |
b411b363 PR |
1401 | } |
1402 | ||
b30ab791 | 1403 | int drbd_send_ov_request(struct drbd_device *device, sector_t sector, int size) |
b411b363 | 1404 | { |
9f5bdc33 AG |
1405 | struct drbd_socket *sock; |
1406 | struct p_block_req *p; | |
b411b363 | 1407 | |
a6b32bc3 | 1408 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1409 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1410 | if (!p) |
1411 | return -EIO; | |
1412 | p->sector = cpu_to_be64(sector); | |
1413 | p->block_id = ID_SYNCER /* unused */; | |
1414 | p->blksize = cpu_to_be32(size); | |
b30ab791 | 1415 | return drbd_send_command(device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0); |
b411b363 PR |
1416 | } |
1417 | ||
1418 | /* called on sndtimeo | |
81e84650 AG |
1419 | * returns false if we should retry, |
1420 | * true if we think connection is dead | |
b411b363 | 1421 | */ |
bde89a9e | 1422 | static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock) |
b411b363 PR |
1423 | { |
1424 | int drop_it; | |
b30ab791 | 1425 | /* long elapsed = (long)(jiffies - device->last_received); */ |
b411b363 | 1426 | |
bde89a9e AG |
1427 | drop_it = connection->meta.socket == sock |
1428 | || !connection->asender.task | |
1429 | || get_t_state(&connection->asender) != RUNNING | |
1430 | || connection->cstate < C_WF_REPORT_PARAMS; | |
b411b363 PR |
1431 | |
1432 | if (drop_it) | |
81e84650 | 1433 | return true; |
b411b363 | 1434 | |
bde89a9e | 1435 | drop_it = !--connection->ko_count; |
b411b363 | 1436 | if (!drop_it) { |
bde89a9e AG |
1437 | conn_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n", |
1438 | current->comm, current->pid, connection->ko_count); | |
1439 | request_ping(connection); | |
b411b363 PR |
1440 | } |
1441 | ||
b30ab791 | 1442 | return drop_it; /* && (device->state == R_PRIMARY) */; |
b411b363 PR |
1443 | } |
1444 | ||
bde89a9e | 1445 | static void drbd_update_congested(struct drbd_connection *connection) |
9e204cdd | 1446 | { |
bde89a9e | 1447 | struct sock *sk = connection->data.socket->sk; |
9e204cdd | 1448 | if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) |
bde89a9e | 1449 | set_bit(NET_CONGESTED, &connection->flags); |
9e204cdd AG |
1450 | } |
1451 | ||
b411b363 PR |
1452 | /* The idea of sendpage seems to be to put some kind of reference |
1453 | * to the page into the skb, and to hand it over to the NIC. In | |
1454 | * this process get_page() gets called. | |
1455 | * | |
1456 | * As soon as the page was really sent over the network put_page() | |
1457 | * gets called by some part of the network layer. [ NIC driver? ] | |
1458 | * | |
1459 | * [ get_page() / put_page() increment/decrement the count. If count | |
1460 | * reaches 0 the page will be freed. ] | |
1461 | * | |
1462 | * This works nicely with pages from FSs. | |
1463 | * But this means that in protocol A we might signal IO completion too early! | |
1464 | * | |
1465 | * In order not to corrupt data during a resync we must make sure | |
1466 | * that we do not reuse our own buffer pages (EEs) to early, therefore | |
1467 | * we have the net_ee list. | |
1468 | * | |
1469 | * XFS seems to have problems, still, it submits pages with page_count == 0! | |
1470 | * As a workaround, we disable sendpage on pages | |
1471 | * with page_count == 0 or PageSlab. | |
1472 | */ | |
b30ab791 | 1473 | static int _drbd_no_send_page(struct drbd_device *device, struct page *page, |
b987427b | 1474 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1475 | { |
b987427b AG |
1476 | struct socket *socket; |
1477 | void *addr; | |
1478 | int err; | |
1479 | ||
a6b32bc3 | 1480 | socket = first_peer_device(device)->connection->data.socket; |
b987427b | 1481 | addr = kmap(page) + offset; |
a6b32bc3 | 1482 | err = drbd_send_all(first_peer_device(device)->connection, socket, addr, size, msg_flags); |
b411b363 | 1483 | kunmap(page); |
b987427b | 1484 | if (!err) |
b30ab791 | 1485 | device->send_cnt += size >> 9; |
b987427b | 1486 | return err; |
b411b363 PR |
1487 | } |
1488 | ||
b30ab791 | 1489 | static int _drbd_send_page(struct drbd_device *device, struct page *page, |
ba11ad9a | 1490 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1491 | { |
a6b32bc3 | 1492 | struct socket *socket = first_peer_device(device)->connection->data.socket; |
b411b363 | 1493 | mm_segment_t oldfs = get_fs(); |
b411b363 | 1494 | int len = size; |
88b390ff | 1495 | int err = -EIO; |
b411b363 PR |
1496 | |
1497 | /* e.g. XFS meta- & log-data is in slab pages, which have a | |
1498 | * page_count of 0 and/or have PageSlab() set. | |
1499 | * we cannot use send_page for those, as that does get_page(); | |
1500 | * put_page(); and would cause either a VM_BUG directly, or | |
1501 | * __page_cache_release a page that would actually still be referenced | |
1502 | * by someone, leading to some obscure delayed Oops somewhere else. */ | |
1503 | if (disable_sendpage || (page_count(page) < 1) || PageSlab(page)) | |
b30ab791 | 1504 | return _drbd_no_send_page(device, page, offset, size, msg_flags); |
b411b363 | 1505 | |
ba11ad9a | 1506 | msg_flags |= MSG_NOSIGNAL; |
a6b32bc3 | 1507 | drbd_update_congested(first_peer_device(device)->connection); |
b411b363 PR |
1508 | set_fs(KERNEL_DS); |
1509 | do { | |
88b390ff AG |
1510 | int sent; |
1511 | ||
1512 | sent = socket->ops->sendpage(socket, page, offset, len, msg_flags); | |
b411b363 | 1513 | if (sent <= 0) { |
88b390ff | 1514 | if (sent == -EAGAIN) { |
a6b32bc3 | 1515 | if (we_should_drop_the_connection(first_peer_device(device)->connection, socket)) |
88b390ff AG |
1516 | break; |
1517 | continue; | |
1518 | } | |
b411b363 PR |
1519 | dev_warn(DEV, "%s: size=%d len=%d sent=%d\n", |
1520 | __func__, (int)size, len, sent); | |
88b390ff AG |
1521 | if (sent < 0) |
1522 | err = sent; | |
b411b363 PR |
1523 | break; |
1524 | } | |
1525 | len -= sent; | |
1526 | offset += sent; | |
b30ab791 | 1527 | } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/); |
b411b363 | 1528 | set_fs(oldfs); |
a6b32bc3 | 1529 | clear_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags); |
b411b363 | 1530 | |
88b390ff AG |
1531 | if (len == 0) { |
1532 | err = 0; | |
b30ab791 | 1533 | device->send_cnt += size >> 9; |
88b390ff AG |
1534 | } |
1535 | return err; | |
b411b363 PR |
1536 | } |
1537 | ||
b30ab791 | 1538 | static int _drbd_send_bio(struct drbd_device *device, struct bio *bio) |
b411b363 | 1539 | { |
7988613b KO |
1540 | struct bio_vec bvec; |
1541 | struct bvec_iter iter; | |
1542 | ||
ba11ad9a | 1543 | /* hint all but last page with MSG_MORE */ |
7988613b | 1544 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1545 | int err; |
1546 | ||
b30ab791 | 1547 | err = _drbd_no_send_page(device, bvec.bv_page, |
7988613b | 1548 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1549 | bio_iter_last(bvec, iter) |
7988613b | 1550 | ? 0 : MSG_MORE); |
7fae55da AG |
1551 | if (err) |
1552 | return err; | |
b411b363 | 1553 | } |
7fae55da | 1554 | return 0; |
b411b363 PR |
1555 | } |
1556 | ||
b30ab791 | 1557 | static int _drbd_send_zc_bio(struct drbd_device *device, struct bio *bio) |
b411b363 | 1558 | { |
7988613b KO |
1559 | struct bio_vec bvec; |
1560 | struct bvec_iter iter; | |
1561 | ||
ba11ad9a | 1562 | /* hint all but last page with MSG_MORE */ |
7988613b | 1563 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1564 | int err; |
1565 | ||
b30ab791 | 1566 | err = _drbd_send_page(device, bvec.bv_page, |
7988613b | 1567 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1568 | bio_iter_last(bvec, iter) ? 0 : MSG_MORE); |
7fae55da AG |
1569 | if (err) |
1570 | return err; | |
b411b363 | 1571 | } |
7fae55da | 1572 | return 0; |
b411b363 PR |
1573 | } |
1574 | ||
b30ab791 | 1575 | static int _drbd_send_zc_ee(struct drbd_device *device, |
db830c46 | 1576 | struct drbd_peer_request *peer_req) |
45bb912b | 1577 | { |
db830c46 AG |
1578 | struct page *page = peer_req->pages; |
1579 | unsigned len = peer_req->i.size; | |
9f69230c | 1580 | int err; |
db830c46 | 1581 | |
ba11ad9a | 1582 | /* hint all but last page with MSG_MORE */ |
45bb912b LE |
1583 | page_chain_for_each(page) { |
1584 | unsigned l = min_t(unsigned, len, PAGE_SIZE); | |
9f69230c | 1585 | |
b30ab791 | 1586 | err = _drbd_send_page(device, page, 0, l, |
9f69230c AG |
1587 | page_chain_next(page) ? MSG_MORE : 0); |
1588 | if (err) | |
1589 | return err; | |
45bb912b LE |
1590 | len -= l; |
1591 | } | |
9f69230c | 1592 | return 0; |
45bb912b LE |
1593 | } |
1594 | ||
b30ab791 | 1595 | static u32 bio_flags_to_wire(struct drbd_device *device, unsigned long bi_rw) |
76d2e7ec | 1596 | { |
a6b32bc3 | 1597 | if (first_peer_device(device)->connection->agreed_pro_version >= 95) |
76d2e7ec | 1598 | return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) | |
76d2e7ec PR |
1599 | (bi_rw & REQ_FUA ? DP_FUA : 0) | |
1600 | (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) | | |
1601 | (bi_rw & REQ_DISCARD ? DP_DISCARD : 0); | |
1602 | else | |
721a9602 | 1603 | return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0; |
76d2e7ec PR |
1604 | } |
1605 | ||
b411b363 PR |
1606 | /* Used to send write requests |
1607 | * R_PRIMARY -> Peer (P_DATA) | |
1608 | */ | |
b30ab791 | 1609 | int drbd_send_dblock(struct drbd_device *device, struct drbd_request *req) |
b411b363 | 1610 | { |
9f5bdc33 AG |
1611 | struct drbd_socket *sock; |
1612 | struct p_data *p; | |
b411b363 | 1613 | unsigned int dp_flags = 0; |
b411b363 | 1614 | int dgs; |
9f5bdc33 | 1615 | int err; |
b411b363 | 1616 | |
a6b32bc3 | 1617 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1618 | p = drbd_prepare_command(device, sock); |
a6b32bc3 AG |
1619 | dgs = first_peer_device(device)->connection->integrity_tfm ? |
1620 | crypto_hash_digestsize(first_peer_device(device)->connection->integrity_tfm) : 0; | |
b411b363 | 1621 | |
9f5bdc33 AG |
1622 | if (!p) |
1623 | return -EIO; | |
1624 | p->sector = cpu_to_be64(req->i.sector); | |
1625 | p->block_id = (unsigned long)req; | |
b30ab791 AG |
1626 | p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); |
1627 | dp_flags = bio_flags_to_wire(device, req->master_bio->bi_rw); | |
1628 | if (device->state.conn >= C_SYNC_SOURCE && | |
1629 | device->state.conn <= C_PAUSED_SYNC_T) | |
b411b363 | 1630 | dp_flags |= DP_MAY_SET_IN_SYNC; |
a6b32bc3 | 1631 | if (first_peer_device(device)->connection->agreed_pro_version >= 100) { |
303d1448 PR |
1632 | if (req->rq_state & RQ_EXP_RECEIVE_ACK) |
1633 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
1634 | if (req->rq_state & RQ_EXP_WRITE_ACK) | |
1635 | dp_flags |= DP_SEND_WRITE_ACK; | |
1636 | } | |
9f5bdc33 AG |
1637 | p->dp_flags = cpu_to_be32(dp_flags); |
1638 | if (dgs) | |
a6b32bc3 AG |
1639 | drbd_csum_bio(device, first_peer_device(device)->connection->integrity_tfm, req->master_bio, p + 1); |
1640 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size); | |
6bdb9b0e | 1641 | if (!err) { |
470be44a LE |
1642 | /* For protocol A, we have to memcpy the payload into |
1643 | * socket buffers, as we may complete right away | |
1644 | * as soon as we handed it over to tcp, at which point the data | |
1645 | * pages may become invalid. | |
1646 | * | |
1647 | * For data-integrity enabled, we copy it as well, so we can be | |
1648 | * sure that even if the bio pages may still be modified, it | |
1649 | * won't change the data on the wire, thus if the digest checks | |
1650 | * out ok after sending on this side, but does not fit on the | |
1651 | * receiving side, we sure have detected corruption elsewhere. | |
1652 | */ | |
303d1448 | 1653 | if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs) |
b30ab791 | 1654 | err = _drbd_send_bio(device, req->master_bio); |
b411b363 | 1655 | else |
b30ab791 | 1656 | err = _drbd_send_zc_bio(device, req->master_bio); |
470be44a LE |
1657 | |
1658 | /* double check digest, sometimes buffers have been modified in flight. */ | |
1659 | if (dgs > 0 && dgs <= 64) { | |
24c4830c | 1660 | /* 64 byte, 512 bit, is the largest digest size |
470be44a LE |
1661 | * currently supported in kernel crypto. */ |
1662 | unsigned char digest[64]; | |
a6b32bc3 | 1663 | drbd_csum_bio(device, first_peer_device(device)->connection->integrity_tfm, req->master_bio, digest); |
9f5bdc33 | 1664 | if (memcmp(p + 1, digest, dgs)) { |
470be44a LE |
1665 | dev_warn(DEV, |
1666 | "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n", | |
ace652ac | 1667 | (unsigned long long)req->i.sector, req->i.size); |
470be44a LE |
1668 | } |
1669 | } /* else if (dgs > 64) { | |
1670 | ... Be noisy about digest too large ... | |
1671 | } */ | |
b411b363 | 1672 | } |
9f5bdc33 | 1673 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
b411b363 | 1674 | |
6bdb9b0e | 1675 | return err; |
b411b363 PR |
1676 | } |
1677 | ||
1678 | /* answer packet, used to send data back for read requests: | |
1679 | * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY) | |
1680 | * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY) | |
1681 | */ | |
b30ab791 | 1682 | int drbd_send_block(struct drbd_device *device, enum drbd_packet cmd, |
db830c46 | 1683 | struct drbd_peer_request *peer_req) |
b411b363 | 1684 | { |
9f5bdc33 AG |
1685 | struct drbd_socket *sock; |
1686 | struct p_data *p; | |
7b57b89d | 1687 | int err; |
b411b363 PR |
1688 | int dgs; |
1689 | ||
a6b32bc3 | 1690 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1691 | p = drbd_prepare_command(device, sock); |
b411b363 | 1692 | |
a6b32bc3 AG |
1693 | dgs = first_peer_device(device)->connection->integrity_tfm ? |
1694 | crypto_hash_digestsize(first_peer_device(device)->connection->integrity_tfm) : 0; | |
b411b363 | 1695 | |
9f5bdc33 AG |
1696 | if (!p) |
1697 | return -EIO; | |
1698 | p->sector = cpu_to_be64(peer_req->i.sector); | |
1699 | p->block_id = peer_req->block_id; | |
1700 | p->seq_num = 0; /* unused */ | |
b17f33cb | 1701 | p->dp_flags = 0; |
9f5bdc33 | 1702 | if (dgs) |
a6b32bc3 AG |
1703 | drbd_csum_ee(device, first_peer_device(device)->connection->integrity_tfm, peer_req, p + 1); |
1704 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size); | |
7b57b89d | 1705 | if (!err) |
b30ab791 | 1706 | err = _drbd_send_zc_ee(device, peer_req); |
9f5bdc33 | 1707 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
bd26bfc5 | 1708 | |
7b57b89d | 1709 | return err; |
b411b363 PR |
1710 | } |
1711 | ||
b30ab791 | 1712 | int drbd_send_out_of_sync(struct drbd_device *device, struct drbd_request *req) |
73a01a18 | 1713 | { |
9f5bdc33 AG |
1714 | struct drbd_socket *sock; |
1715 | struct p_block_desc *p; | |
73a01a18 | 1716 | |
a6b32bc3 | 1717 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1718 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1719 | if (!p) |
1720 | return -EIO; | |
1721 | p->sector = cpu_to_be64(req->i.sector); | |
1722 | p->blksize = cpu_to_be32(req->i.size); | |
b30ab791 | 1723 | return drbd_send_command(device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0); |
73a01a18 PR |
1724 | } |
1725 | ||
b411b363 PR |
1726 | /* |
1727 | drbd_send distinguishes two cases: | |
1728 | ||
1729 | Packets sent via the data socket "sock" | |
1730 | and packets sent via the meta data socket "msock" | |
1731 | ||
1732 | sock msock | |
1733 | -----------------+-------------------------+------------------------------ | |
1734 | timeout conf.timeout / 2 conf.timeout / 2 | |
1735 | timeout action send a ping via msock Abort communication | |
1736 | and close all sockets | |
1737 | */ | |
1738 | ||
1739 | /* | |
1740 | * you must have down()ed the appropriate [m]sock_mutex elsewhere! | |
1741 | */ | |
bde89a9e | 1742 | int drbd_send(struct drbd_connection *connection, struct socket *sock, |
b411b363 PR |
1743 | void *buf, size_t size, unsigned msg_flags) |
1744 | { | |
1745 | struct kvec iov; | |
1746 | struct msghdr msg; | |
1747 | int rv, sent = 0; | |
1748 | ||
1749 | if (!sock) | |
c0d42c8e | 1750 | return -EBADR; |
b411b363 PR |
1751 | |
1752 | /* THINK if (signal_pending) return ... ? */ | |
1753 | ||
1754 | iov.iov_base = buf; | |
1755 | iov.iov_len = size; | |
1756 | ||
1757 | msg.msg_name = NULL; | |
1758 | msg.msg_namelen = 0; | |
1759 | msg.msg_control = NULL; | |
1760 | msg.msg_controllen = 0; | |
1761 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; | |
1762 | ||
bde89a9e | 1763 | if (sock == connection->data.socket) { |
44ed167d | 1764 | rcu_read_lock(); |
bde89a9e | 1765 | connection->ko_count = rcu_dereference(connection->net_conf)->ko_count; |
44ed167d | 1766 | rcu_read_unlock(); |
bde89a9e | 1767 | drbd_update_congested(connection); |
b411b363 PR |
1768 | } |
1769 | do { | |
1770 | /* STRANGE | |
1771 | * tcp_sendmsg does _not_ use its size parameter at all ? | |
1772 | * | |
1773 | * -EAGAIN on timeout, -EINTR on signal. | |
1774 | */ | |
1775 | /* THINK | |
1776 | * do we need to block DRBD_SIG if sock == &meta.socket ?? | |
1777 | * otherwise wake_asender() might interrupt some send_*Ack ! | |
1778 | */ | |
1779 | rv = kernel_sendmsg(sock, &msg, &iov, 1, size); | |
1780 | if (rv == -EAGAIN) { | |
bde89a9e | 1781 | if (we_should_drop_the_connection(connection, sock)) |
b411b363 PR |
1782 | break; |
1783 | else | |
1784 | continue; | |
1785 | } | |
b411b363 PR |
1786 | if (rv == -EINTR) { |
1787 | flush_signals(current); | |
1788 | rv = 0; | |
1789 | } | |
1790 | if (rv < 0) | |
1791 | break; | |
1792 | sent += rv; | |
1793 | iov.iov_base += rv; | |
1794 | iov.iov_len -= rv; | |
1795 | } while (sent < size); | |
1796 | ||
bde89a9e AG |
1797 | if (sock == connection->data.socket) |
1798 | clear_bit(NET_CONGESTED, &connection->flags); | |
b411b363 PR |
1799 | |
1800 | if (rv <= 0) { | |
1801 | if (rv != -EAGAIN) { | |
bde89a9e AG |
1802 | conn_err(connection, "%s_sendmsg returned %d\n", |
1803 | sock == connection->meta.socket ? "msock" : "sock", | |
bedbd2a5 | 1804 | rv); |
bde89a9e | 1805 | conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 1806 | } else |
bde89a9e | 1807 | conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); |
b411b363 PR |
1808 | } |
1809 | ||
1810 | return sent; | |
1811 | } | |
1812 | ||
fb708e40 AG |
1813 | /** |
1814 | * drbd_send_all - Send an entire buffer | |
1815 | * | |
1816 | * Returns 0 upon success and a negative error value otherwise. | |
1817 | */ | |
bde89a9e | 1818 | int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer, |
fb708e40 AG |
1819 | size_t size, unsigned msg_flags) |
1820 | { | |
1821 | int err; | |
1822 | ||
bde89a9e | 1823 | err = drbd_send(connection, sock, buffer, size, msg_flags); |
fb708e40 AG |
1824 | if (err < 0) |
1825 | return err; | |
1826 | if (err != size) | |
1827 | return -EIO; | |
1828 | return 0; | |
1829 | } | |
1830 | ||
b411b363 PR |
1831 | static int drbd_open(struct block_device *bdev, fmode_t mode) |
1832 | { | |
b30ab791 | 1833 | struct drbd_device *device = bdev->bd_disk->private_data; |
b411b363 PR |
1834 | unsigned long flags; |
1835 | int rv = 0; | |
1836 | ||
2a48fc0a | 1837 | mutex_lock(&drbd_main_mutex); |
a6b32bc3 | 1838 | spin_lock_irqsave(&first_peer_device(device)->connection->req_lock, flags); |
b30ab791 | 1839 | /* to have a stable device->state.role |
b411b363 PR |
1840 | * and no race with updating open_cnt */ |
1841 | ||
b30ab791 | 1842 | if (device->state.role != R_PRIMARY) { |
b411b363 PR |
1843 | if (mode & FMODE_WRITE) |
1844 | rv = -EROFS; | |
1845 | else if (!allow_oos) | |
1846 | rv = -EMEDIUMTYPE; | |
1847 | } | |
1848 | ||
1849 | if (!rv) | |
b30ab791 | 1850 | device->open_cnt++; |
a6b32bc3 | 1851 | spin_unlock_irqrestore(&first_peer_device(device)->connection->req_lock, flags); |
2a48fc0a | 1852 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1853 | |
1854 | return rv; | |
1855 | } | |
1856 | ||
db2a144b | 1857 | static void drbd_release(struct gendisk *gd, fmode_t mode) |
b411b363 | 1858 | { |
b30ab791 | 1859 | struct drbd_device *device = gd->private_data; |
2a48fc0a | 1860 | mutex_lock(&drbd_main_mutex); |
b30ab791 | 1861 | device->open_cnt--; |
2a48fc0a | 1862 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1863 | } |
1864 | ||
b30ab791 | 1865 | static void drbd_set_defaults(struct drbd_device *device) |
b411b363 | 1866 | { |
f399002e LE |
1867 | /* Beware! The actual layout differs |
1868 | * between big endian and little endian */ | |
b30ab791 | 1869 | device->state = (union drbd_dev_state) { |
b411b363 PR |
1870 | { .role = R_SECONDARY, |
1871 | .peer = R_UNKNOWN, | |
1872 | .conn = C_STANDALONE, | |
1873 | .disk = D_DISKLESS, | |
1874 | .pdsk = D_UNKNOWN, | |
b411b363 PR |
1875 | } }; |
1876 | } | |
1877 | ||
b30ab791 | 1878 | void drbd_init_set_defaults(struct drbd_device *device) |
b411b363 PR |
1879 | { |
1880 | /* the memset(,0,) did most of this. | |
1881 | * note: only assignments, no allocation in here */ | |
1882 | ||
b30ab791 AG |
1883 | drbd_set_defaults(device); |
1884 | ||
1885 | atomic_set(&device->ap_bio_cnt, 0); | |
1886 | atomic_set(&device->ap_pending_cnt, 0); | |
1887 | atomic_set(&device->rs_pending_cnt, 0); | |
1888 | atomic_set(&device->unacked_cnt, 0); | |
1889 | atomic_set(&device->local_cnt, 0); | |
1890 | atomic_set(&device->pp_in_use_by_net, 0); | |
1891 | atomic_set(&device->rs_sect_in, 0); | |
1892 | atomic_set(&device->rs_sect_ev, 0); | |
1893 | atomic_set(&device->ap_in_flight, 0); | |
1894 | atomic_set(&device->md_io_in_use, 0); | |
1895 | ||
1896 | mutex_init(&device->own_state_mutex); | |
1897 | device->state_mutex = &device->own_state_mutex; | |
1898 | ||
1899 | spin_lock_init(&device->al_lock); | |
1900 | spin_lock_init(&device->peer_seq_lock); | |
1901 | ||
1902 | INIT_LIST_HEAD(&device->active_ee); | |
1903 | INIT_LIST_HEAD(&device->sync_ee); | |
1904 | INIT_LIST_HEAD(&device->done_ee); | |
1905 | INIT_LIST_HEAD(&device->read_ee); | |
1906 | INIT_LIST_HEAD(&device->net_ee); | |
1907 | INIT_LIST_HEAD(&device->resync_reads); | |
1908 | INIT_LIST_HEAD(&device->resync_work.list); | |
1909 | INIT_LIST_HEAD(&device->unplug_work.list); | |
1910 | INIT_LIST_HEAD(&device->go_diskless.list); | |
1911 | INIT_LIST_HEAD(&device->md_sync_work.list); | |
1912 | INIT_LIST_HEAD(&device->start_resync_work.list); | |
1913 | INIT_LIST_HEAD(&device->bm_io_work.w.list); | |
1914 | ||
1915 | device->resync_work.cb = w_resync_timer; | |
1916 | device->unplug_work.cb = w_send_write_hint; | |
1917 | device->go_diskless.cb = w_go_diskless; | |
1918 | device->md_sync_work.cb = w_md_sync; | |
1919 | device->bm_io_work.w.cb = w_bitmap_io; | |
1920 | device->start_resync_work.cb = w_start_resync; | |
1921 | ||
1922 | device->resync_work.device = device; | |
1923 | device->unplug_work.device = device; | |
1924 | device->go_diskless.device = device; | |
1925 | device->md_sync_work.device = device; | |
1926 | device->bm_io_work.w.device = device; | |
1927 | device->start_resync_work.device = device; | |
1928 | ||
1929 | init_timer(&device->resync_timer); | |
1930 | init_timer(&device->md_sync_timer); | |
1931 | init_timer(&device->start_resync_timer); | |
1932 | init_timer(&device->request_timer); | |
1933 | device->resync_timer.function = resync_timer_fn; | |
1934 | device->resync_timer.data = (unsigned long) device; | |
1935 | device->md_sync_timer.function = md_sync_timer_fn; | |
1936 | device->md_sync_timer.data = (unsigned long) device; | |
1937 | device->start_resync_timer.function = start_resync_timer_fn; | |
1938 | device->start_resync_timer.data = (unsigned long) device; | |
1939 | device->request_timer.function = request_timer_fn; | |
1940 | device->request_timer.data = (unsigned long) device; | |
1941 | ||
1942 | init_waitqueue_head(&device->misc_wait); | |
1943 | init_waitqueue_head(&device->state_wait); | |
1944 | init_waitqueue_head(&device->ee_wait); | |
1945 | init_waitqueue_head(&device->al_wait); | |
1946 | init_waitqueue_head(&device->seq_wait); | |
1947 | ||
1948 | device->resync_wenr = LC_FREE; | |
1949 | device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1950 | device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1951 | } | |
1952 | ||
1953 | void drbd_device_cleanup(struct drbd_device *device) | |
b411b363 | 1954 | { |
1d7734a0 | 1955 | int i; |
a6b32bc3 | 1956 | if (first_peer_device(device)->connection->receiver.t_state != NONE) |
b411b363 | 1957 | dev_err(DEV, "ASSERT FAILED: receiver t_state == %d expected 0.\n", |
a6b32bc3 | 1958 | first_peer_device(device)->connection->receiver.t_state); |
b30ab791 AG |
1959 | |
1960 | device->al_writ_cnt = | |
1961 | device->bm_writ_cnt = | |
1962 | device->read_cnt = | |
1963 | device->recv_cnt = | |
1964 | device->send_cnt = | |
1965 | device->writ_cnt = | |
1966 | device->p_size = | |
1967 | device->rs_start = | |
1968 | device->rs_total = | |
1969 | device->rs_failed = 0; | |
1970 | device->rs_last_events = 0; | |
1971 | device->rs_last_sect_ev = 0; | |
1d7734a0 | 1972 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1973 | device->rs_mark_left[i] = 0; |
1974 | device->rs_mark_time[i] = 0; | |
1d7734a0 | 1975 | } |
a6b32bc3 | 1976 | D_ASSERT(first_peer_device(device)->connection->net_conf == NULL); |
b411b363 | 1977 | |
b30ab791 AG |
1978 | drbd_set_my_capacity(device, 0); |
1979 | if (device->bitmap) { | |
b411b363 | 1980 | /* maybe never allocated. */ |
b30ab791 AG |
1981 | drbd_bm_resize(device, 0, 1); |
1982 | drbd_bm_cleanup(device); | |
b411b363 PR |
1983 | } |
1984 | ||
b30ab791 AG |
1985 | drbd_free_bc(device->ldev); |
1986 | device->ldev = NULL; | |
1d041225 | 1987 | |
b30ab791 | 1988 | clear_bit(AL_SUSPENDED, &device->flags); |
b411b363 | 1989 | |
b30ab791 AG |
1990 | D_ASSERT(list_empty(&device->active_ee)); |
1991 | D_ASSERT(list_empty(&device->sync_ee)); | |
1992 | D_ASSERT(list_empty(&device->done_ee)); | |
1993 | D_ASSERT(list_empty(&device->read_ee)); | |
1994 | D_ASSERT(list_empty(&device->net_ee)); | |
1995 | D_ASSERT(list_empty(&device->resync_reads)); | |
a6b32bc3 | 1996 | D_ASSERT(list_empty(&first_peer_device(device)->connection->sender_work.q)); |
b30ab791 AG |
1997 | D_ASSERT(list_empty(&device->resync_work.list)); |
1998 | D_ASSERT(list_empty(&device->unplug_work.list)); | |
1999 | D_ASSERT(list_empty(&device->go_diskless.list)); | |
2265b473 | 2000 | |
b30ab791 | 2001 | drbd_set_defaults(device); |
b411b363 PR |
2002 | } |
2003 | ||
2004 | ||
2005 | static void drbd_destroy_mempools(void) | |
2006 | { | |
2007 | struct page *page; | |
2008 | ||
2009 | while (drbd_pp_pool) { | |
2010 | page = drbd_pp_pool; | |
2011 | drbd_pp_pool = (struct page *)page_private(page); | |
2012 | __free_page(page); | |
2013 | drbd_pp_vacant--; | |
2014 | } | |
2015 | ||
2016 | /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */ | |
2017 | ||
9476f39d LE |
2018 | if (drbd_md_io_bio_set) |
2019 | bioset_free(drbd_md_io_bio_set); | |
4281808f LE |
2020 | if (drbd_md_io_page_pool) |
2021 | mempool_destroy(drbd_md_io_page_pool); | |
b411b363 PR |
2022 | if (drbd_ee_mempool) |
2023 | mempool_destroy(drbd_ee_mempool); | |
2024 | if (drbd_request_mempool) | |
2025 | mempool_destroy(drbd_request_mempool); | |
2026 | if (drbd_ee_cache) | |
2027 | kmem_cache_destroy(drbd_ee_cache); | |
2028 | if (drbd_request_cache) | |
2029 | kmem_cache_destroy(drbd_request_cache); | |
2030 | if (drbd_bm_ext_cache) | |
2031 | kmem_cache_destroy(drbd_bm_ext_cache); | |
2032 | if (drbd_al_ext_cache) | |
2033 | kmem_cache_destroy(drbd_al_ext_cache); | |
2034 | ||
9476f39d | 2035 | drbd_md_io_bio_set = NULL; |
4281808f | 2036 | drbd_md_io_page_pool = NULL; |
b411b363 PR |
2037 | drbd_ee_mempool = NULL; |
2038 | drbd_request_mempool = NULL; | |
2039 | drbd_ee_cache = NULL; | |
2040 | drbd_request_cache = NULL; | |
2041 | drbd_bm_ext_cache = NULL; | |
2042 | drbd_al_ext_cache = NULL; | |
2043 | ||
2044 | return; | |
2045 | } | |
2046 | ||
2047 | static int drbd_create_mempools(void) | |
2048 | { | |
2049 | struct page *page; | |
1816a2b4 | 2050 | const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count; |
b411b363 PR |
2051 | int i; |
2052 | ||
2053 | /* prepare our caches and mempools */ | |
2054 | drbd_request_mempool = NULL; | |
2055 | drbd_ee_cache = NULL; | |
2056 | drbd_request_cache = NULL; | |
2057 | drbd_bm_ext_cache = NULL; | |
2058 | drbd_al_ext_cache = NULL; | |
2059 | drbd_pp_pool = NULL; | |
4281808f | 2060 | drbd_md_io_page_pool = NULL; |
9476f39d | 2061 | drbd_md_io_bio_set = NULL; |
b411b363 PR |
2062 | |
2063 | /* caches */ | |
2064 | drbd_request_cache = kmem_cache_create( | |
2065 | "drbd_req", sizeof(struct drbd_request), 0, 0, NULL); | |
2066 | if (drbd_request_cache == NULL) | |
2067 | goto Enomem; | |
2068 | ||
2069 | drbd_ee_cache = kmem_cache_create( | |
f6ffca9f | 2070 | "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL); |
b411b363 PR |
2071 | if (drbd_ee_cache == NULL) |
2072 | goto Enomem; | |
2073 | ||
2074 | drbd_bm_ext_cache = kmem_cache_create( | |
2075 | "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL); | |
2076 | if (drbd_bm_ext_cache == NULL) | |
2077 | goto Enomem; | |
2078 | ||
2079 | drbd_al_ext_cache = kmem_cache_create( | |
2080 | "drbd_al", sizeof(struct lc_element), 0, 0, NULL); | |
2081 | if (drbd_al_ext_cache == NULL) | |
2082 | goto Enomem; | |
2083 | ||
2084 | /* mempools */ | |
9476f39d LE |
2085 | drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0); |
2086 | if (drbd_md_io_bio_set == NULL) | |
2087 | goto Enomem; | |
9476f39d | 2088 | |
4281808f LE |
2089 | drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0); |
2090 | if (drbd_md_io_page_pool == NULL) | |
2091 | goto Enomem; | |
2092 | ||
b411b363 PR |
2093 | drbd_request_mempool = mempool_create(number, |
2094 | mempool_alloc_slab, mempool_free_slab, drbd_request_cache); | |
2095 | if (drbd_request_mempool == NULL) | |
2096 | goto Enomem; | |
2097 | ||
2098 | drbd_ee_mempool = mempool_create(number, | |
2099 | mempool_alloc_slab, mempool_free_slab, drbd_ee_cache); | |
2027ae1f | 2100 | if (drbd_ee_mempool == NULL) |
b411b363 PR |
2101 | goto Enomem; |
2102 | ||
2103 | /* drbd's page pool */ | |
2104 | spin_lock_init(&drbd_pp_lock); | |
2105 | ||
2106 | for (i = 0; i < number; i++) { | |
2107 | page = alloc_page(GFP_HIGHUSER); | |
2108 | if (!page) | |
2109 | goto Enomem; | |
2110 | set_page_private(page, (unsigned long)drbd_pp_pool); | |
2111 | drbd_pp_pool = page; | |
2112 | } | |
2113 | drbd_pp_vacant = number; | |
2114 | ||
2115 | return 0; | |
2116 | ||
2117 | Enomem: | |
2118 | drbd_destroy_mempools(); /* in case we allocated some */ | |
2119 | return -ENOMEM; | |
2120 | } | |
2121 | ||
2122 | static int drbd_notify_sys(struct notifier_block *this, unsigned long code, | |
2123 | void *unused) | |
2124 | { | |
2125 | /* just so we have it. you never know what interesting things we | |
2126 | * might want to do here some day... | |
2127 | */ | |
2128 | ||
2129 | return NOTIFY_DONE; | |
2130 | } | |
2131 | ||
2132 | static struct notifier_block drbd_notifier = { | |
2133 | .notifier_call = drbd_notify_sys, | |
2134 | }; | |
2135 | ||
b30ab791 | 2136 | static void drbd_release_all_peer_reqs(struct drbd_device *device) |
b411b363 PR |
2137 | { |
2138 | int rr; | |
2139 | ||
b30ab791 | 2140 | rr = drbd_free_peer_reqs(device, &device->active_ee); |
b411b363 PR |
2141 | if (rr) |
2142 | dev_err(DEV, "%d EEs in active list found!\n", rr); | |
2143 | ||
b30ab791 | 2144 | rr = drbd_free_peer_reqs(device, &device->sync_ee); |
b411b363 PR |
2145 | if (rr) |
2146 | dev_err(DEV, "%d EEs in sync list found!\n", rr); | |
2147 | ||
b30ab791 | 2148 | rr = drbd_free_peer_reqs(device, &device->read_ee); |
b411b363 PR |
2149 | if (rr) |
2150 | dev_err(DEV, "%d EEs in read list found!\n", rr); | |
2151 | ||
b30ab791 | 2152 | rr = drbd_free_peer_reqs(device, &device->done_ee); |
b411b363 PR |
2153 | if (rr) |
2154 | dev_err(DEV, "%d EEs in done list found!\n", rr); | |
2155 | ||
b30ab791 | 2156 | rr = drbd_free_peer_reqs(device, &device->net_ee); |
b411b363 PR |
2157 | if (rr) |
2158 | dev_err(DEV, "%d EEs in net list found!\n", rr); | |
2159 | } | |
2160 | ||
774b3055 | 2161 | /* caution. no locking. */ |
05a10ec7 | 2162 | void drbd_destroy_device(struct kref *kref) |
b411b363 | 2163 | { |
b30ab791 | 2164 | struct drbd_device *device = container_of(kref, struct drbd_device, kref); |
a6b32bc3 | 2165 | struct drbd_connection *connection = first_peer_device(device)->connection; |
b411b363 | 2166 | |
b30ab791 | 2167 | del_timer_sync(&device->request_timer); |
dfa8bedb | 2168 | |
b411b363 | 2169 | /* paranoia asserts */ |
b30ab791 | 2170 | D_ASSERT(device->open_cnt == 0); |
b411b363 PR |
2171 | /* end paranoia asserts */ |
2172 | ||
b411b363 PR |
2173 | /* cleanup stuff that may have been allocated during |
2174 | * device (re-)configuration or state changes */ | |
2175 | ||
b30ab791 AG |
2176 | if (device->this_bdev) |
2177 | bdput(device->this_bdev); | |
b411b363 | 2178 | |
b30ab791 AG |
2179 | drbd_free_bc(device->ldev); |
2180 | device->ldev = NULL; | |
b411b363 | 2181 | |
b30ab791 | 2182 | drbd_release_all_peer_reqs(device); |
b411b363 | 2183 | |
b30ab791 AG |
2184 | lc_destroy(device->act_log); |
2185 | lc_destroy(device->resync); | |
b411b363 | 2186 | |
b30ab791 AG |
2187 | kfree(device->p_uuid); |
2188 | /* device->p_uuid = NULL; */ | |
b411b363 | 2189 | |
b30ab791 AG |
2190 | if (device->bitmap) /* should no longer be there. */ |
2191 | drbd_bm_cleanup(device); | |
2192 | __free_page(device->md_io_page); | |
2193 | put_disk(device->vdisk); | |
2194 | blk_cleanup_queue(device->rq_queue); | |
2195 | kfree(device->rs_plan_s); | |
a6b32bc3 | 2196 | kfree(first_peer_device(device)); |
b30ab791 | 2197 | kfree(device); |
9dc9fbb3 | 2198 | |
05a10ec7 | 2199 | kref_put(&connection->kref, drbd_destroy_connection); |
b411b363 PR |
2200 | } |
2201 | ||
2312f0b3 LE |
2202 | /* One global retry thread, if we need to push back some bio and have it |
2203 | * reinserted through our make request function. | |
2204 | */ | |
2205 | static 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 | ||
2213 | static 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) { | |
b30ab791 | 2224 | struct drbd_device *device = req->w.device; |
2312f0b3 LE |
2225 | struct bio *bio = req->master_bio; |
2226 | unsigned long start_time = req->start_time; | |
9a278a79 LE |
2227 | bool expected; |
2228 | ||
2229 | expected = | |
2230 | 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) | |
2236 | dev_err(DEV, "req=%p completion_ref=%d rq_state=%x\n", | |
2237 | 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. | |
2312f0b3 | 2244 | */ |
9a278a79 | 2245 | kref_put(&req->kref, drbd_req_destroy); |
2312f0b3 LE |
2246 | |
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. */ | |
b30ab791 AG |
2260 | inc_ap_bio(device); |
2261 | __drbd_make_request(device, bio, start_time); | |
2312f0b3 LE |
2262 | } |
2263 | } | |
2264 | ||
9d05e7c4 | 2265 | void drbd_restart_request(struct drbd_request *req) |
2312f0b3 LE |
2266 | { |
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. */ | |
b30ab791 | 2275 | dec_ap_bio(req->w.device); |
b411b363 | 2276 | |
2312f0b3 | 2277 | queue_work(retry.wq, &retry.worker); |
b411b363 PR |
2278 | } |
2279 | ||
77c556f6 AG |
2280 | void drbd_destroy_resource(struct kref *kref) |
2281 | { | |
2282 | struct drbd_resource *resource = | |
2283 | container_of(kref, struct drbd_resource, kref); | |
2284 | ||
2285 | kfree(resource->name); | |
2286 | kfree(resource); | |
2287 | } | |
2288 | ||
2289 | void drbd_free_resource(struct drbd_resource *resource) | |
2290 | { | |
2291 | struct drbd_connection *connection, *tmp; | |
2292 | ||
2293 | for_each_connection_safe(connection, tmp, resource) { | |
2294 | list_del(&connection->connections); | |
2295 | kref_put(&connection->kref, drbd_destroy_connection); | |
2296 | } | |
2297 | kref_put(&resource->kref, drbd_destroy_resource); | |
2298 | } | |
2312f0b3 | 2299 | |
b411b363 PR |
2300 | static void drbd_cleanup(void) |
2301 | { | |
2302 | unsigned int i; | |
b30ab791 | 2303 | struct drbd_device *device; |
77c556f6 | 2304 | struct drbd_resource *resource, *tmp; |
b411b363 PR |
2305 | |
2306 | unregister_reboot_notifier(&drbd_notifier); | |
2307 | ||
17a93f30 LE |
2308 | /* 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 | ||
2312f0b3 LE |
2319 | if (retry.wq) |
2320 | destroy_workqueue(retry.wq); | |
b411b363 | 2321 | |
3b98c0c2 | 2322 | drbd_genl_unregister(); |
b411b363 | 2323 | |
05a10ec7 AG |
2324 | idr_for_each_entry(&drbd_devices, device, i) { |
2325 | idr_remove(&drbd_devices, device_to_minor(device)); | |
a6b32bc3 | 2326 | idr_remove(&first_peer_device(device)->connection->volumes, device->vnr); |
b30ab791 AG |
2327 | destroy_workqueue(device->submit.wq); |
2328 | del_gendisk(device->vdisk); | |
c141ebda | 2329 | /* synchronize_rcu(); No other threads running at this point */ |
05a10ec7 | 2330 | kref_put(&device->kref, drbd_destroy_device); |
b411b363 PR |
2331 | } |
2332 | ||
c141ebda | 2333 | /* not _rcu since, no other updater anymore. Genl already unregistered */ |
77c556f6 AG |
2334 | for_each_resource_safe(resource, tmp, &drbd_resources) { |
2335 | list_del(&resource->resources); | |
2336 | drbd_free_resource(resource); | |
81fa2e67 | 2337 | } |
b411b363 | 2338 | |
81a5d60e | 2339 | drbd_destroy_mempools(); |
b411b363 PR |
2340 | unregister_blkdev(DRBD_MAJOR, "drbd"); |
2341 | ||
05a10ec7 | 2342 | idr_destroy(&drbd_devices); |
81a5d60e | 2343 | |
b411b363 PR |
2344 | printk(KERN_INFO "drbd: module cleanup done.\n"); |
2345 | } | |
2346 | ||
2347 | /** | |
d97482ed | 2348 | * drbd_congested() - Callback for the flusher thread |
b411b363 | 2349 | * @congested_data: User data |
d97482ed | 2350 | * @bdi_bits: Bits the BDI flusher thread is currently interested in |
b411b363 PR |
2351 | * |
2352 | * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested. | |
2353 | */ | |
2354 | static int drbd_congested(void *congested_data, int bdi_bits) | |
2355 | { | |
b30ab791 | 2356 | struct drbd_device *device = congested_data; |
b411b363 PR |
2357 | struct request_queue *q; |
2358 | char reason = '-'; | |
2359 | int r = 0; | |
2360 | ||
b30ab791 | 2361 | if (!may_inc_ap_bio(device)) { |
b411b363 PR |
2362 | /* DRBD has frozen IO */ |
2363 | r = bdi_bits; | |
2364 | reason = 'd'; | |
2365 | goto out; | |
2366 | } | |
2367 | ||
a6b32bc3 | 2368 | if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) { |
c2ba686f LE |
2369 | r |= (1 << BDI_async_congested); |
2370 | /* Without good local data, we would need to read from remote, | |
2371 | * and that would need the worker thread as well, which is | |
2372 | * currently blocked waiting for that usermode helper to | |
2373 | * finish. | |
2374 | */ | |
b30ab791 | 2375 | if (!get_ldev_if_state(device, D_UP_TO_DATE)) |
c2ba686f LE |
2376 | r |= (1 << BDI_sync_congested); |
2377 | else | |
b30ab791 | 2378 | put_ldev(device); |
c2ba686f LE |
2379 | r &= bdi_bits; |
2380 | reason = 'c'; | |
2381 | goto out; | |
2382 | } | |
2383 | ||
b30ab791 AG |
2384 | if (get_ldev(device)) { |
2385 | q = bdev_get_queue(device->ldev->backing_bdev); | |
b411b363 | 2386 | r = bdi_congested(&q->backing_dev_info, bdi_bits); |
b30ab791 | 2387 | put_ldev(device); |
b411b363 PR |
2388 | if (r) |
2389 | reason = 'b'; | |
2390 | } | |
2391 | ||
a6b32bc3 AG |
2392 | if (bdi_bits & (1 << BDI_async_congested) && |
2393 | test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) { | |
b411b363 PR |
2394 | r |= (1 << BDI_async_congested); |
2395 | reason = reason == 'b' ? 'a' : 'n'; | |
2396 | } | |
2397 | ||
2398 | out: | |
b30ab791 | 2399 | device->congestion_reason = reason; |
b411b363 PR |
2400 | return r; |
2401 | } | |
2402 | ||
6699b655 PR |
2403 | static void drbd_init_workqueue(struct drbd_work_queue* wq) |
2404 | { | |
6699b655 PR |
2405 | spin_lock_init(&wq->q_lock); |
2406 | INIT_LIST_HEAD(&wq->q); | |
8c0785a5 | 2407 | init_waitqueue_head(&wq->q_wait); |
6699b655 PR |
2408 | } |
2409 | ||
bde89a9e | 2410 | struct drbd_connection *conn_get_by_name(const char *name) |
1aba4d7f | 2411 | { |
bde89a9e | 2412 | struct drbd_connection *connection; |
77c556f6 | 2413 | struct drbd_resource *resource; |
1aba4d7f | 2414 | |
3b98c0c2 LE |
2415 | if (!name || !name[0]) |
2416 | return NULL; | |
2417 | ||
c141ebda | 2418 | rcu_read_lock(); |
77c556f6 AG |
2419 | for_each_resource_rcu(resource, &drbd_resources) { |
2420 | if (!strcmp(resource->name, name)) { | |
2421 | connection = first_connection(resource); | |
bde89a9e | 2422 | kref_get(&connection->kref); |
1aba4d7f | 2423 | goto found; |
0ace9dfa | 2424 | } |
1aba4d7f | 2425 | } |
bde89a9e | 2426 | connection = NULL; |
1aba4d7f | 2427 | found: |
c141ebda | 2428 | rcu_read_unlock(); |
bde89a9e | 2429 | return connection; |
1aba4d7f PR |
2430 | } |
2431 | ||
bde89a9e | 2432 | struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len, |
089c075d AG |
2433 | void *peer_addr, int peer_addr_len) |
2434 | { | |
77c556f6 | 2435 | struct drbd_resource *resource; |
bde89a9e | 2436 | struct drbd_connection *connection; |
089c075d AG |
2437 | |
2438 | rcu_read_lock(); | |
77c556f6 AG |
2439 | for_each_resource_rcu(resource, &drbd_resources) { |
2440 | for_each_connection_rcu(connection, resource) { | |
2441 | if (connection->my_addr_len == my_addr_len && | |
2442 | connection->peer_addr_len == peer_addr_len && | |
2443 | !memcmp(&connection->my_addr, my_addr, my_addr_len) && | |
2444 | !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) { | |
2445 | kref_get(&connection->kref); | |
2446 | goto found; | |
2447 | } | |
089c075d AG |
2448 | } |
2449 | } | |
bde89a9e | 2450 | connection = NULL; |
089c075d AG |
2451 | found: |
2452 | rcu_read_unlock(); | |
bde89a9e | 2453 | return connection; |
089c075d AG |
2454 | } |
2455 | ||
e6ef8a5c AG |
2456 | static int drbd_alloc_socket(struct drbd_socket *socket) |
2457 | { | |
2458 | socket->rbuf = (void *) __get_free_page(GFP_KERNEL); | |
2459 | if (!socket->rbuf) | |
2460 | return -ENOMEM; | |
5a87d920 AG |
2461 | socket->sbuf = (void *) __get_free_page(GFP_KERNEL); |
2462 | if (!socket->sbuf) | |
2463 | return -ENOMEM; | |
e6ef8a5c AG |
2464 | return 0; |
2465 | } | |
2466 | ||
2467 | static void drbd_free_socket(struct drbd_socket *socket) | |
2468 | { | |
5a87d920 | 2469 | free_page((unsigned long) socket->sbuf); |
e6ef8a5c AG |
2470 | free_page((unsigned long) socket->rbuf); |
2471 | } | |
2472 | ||
bde89a9e | 2473 | void conn_free_crypto(struct drbd_connection *connection) |
91fd4dad | 2474 | { |
bde89a9e | 2475 | drbd_free_sock(connection); |
1d041225 | 2476 | |
bde89a9e AG |
2477 | crypto_free_hash(connection->csums_tfm); |
2478 | crypto_free_hash(connection->verify_tfm); | |
2479 | crypto_free_hash(connection->cram_hmac_tfm); | |
2480 | crypto_free_hash(connection->integrity_tfm); | |
2481 | crypto_free_hash(connection->peer_integrity_tfm); | |
2482 | kfree(connection->int_dig_in); | |
2483 | kfree(connection->int_dig_vv); | |
1d041225 | 2484 | |
bde89a9e AG |
2485 | connection->csums_tfm = NULL; |
2486 | connection->verify_tfm = NULL; | |
2487 | connection->cram_hmac_tfm = NULL; | |
2488 | connection->integrity_tfm = NULL; | |
2489 | connection->peer_integrity_tfm = NULL; | |
2490 | connection->int_dig_in = NULL; | |
2491 | connection->int_dig_vv = NULL; | |
91fd4dad PR |
2492 | } |
2493 | ||
bde89a9e | 2494 | int set_resource_options(struct drbd_connection *connection, struct res_opts *res_opts) |
afbbfa88 AG |
2495 | { |
2496 | cpumask_var_t new_cpu_mask; | |
2497 | int err; | |
2498 | ||
2499 | if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) | |
2500 | return -ENOMEM; | |
2501 | /* | |
2502 | retcode = ERR_NOMEM; | |
2503 | drbd_msg_put_info("unable to allocate cpumask"); | |
2504 | */ | |
2505 | ||
2506 | /* silently ignore cpu mask on UP kernel */ | |
2507 | if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) { | |
2508 | /* FIXME: Get rid of constant 32 here */ | |
c5b005ab PR |
2509 | err = bitmap_parse(res_opts->cpu_mask, 32, |
2510 | cpumask_bits(new_cpu_mask), nr_cpu_ids); | |
afbbfa88 | 2511 | if (err) { |
bde89a9e | 2512 | conn_warn(connection, "bitmap_parse() failed with %d\n", err); |
afbbfa88 AG |
2513 | /* retcode = ERR_CPU_MASK_PARSE; */ |
2514 | goto fail; | |
2515 | } | |
2516 | } | |
bde89a9e AG |
2517 | connection->res_opts = *res_opts; |
2518 | if (!cpumask_equal(connection->cpu_mask, new_cpu_mask)) { | |
2519 | cpumask_copy(connection->cpu_mask, new_cpu_mask); | |
2520 | drbd_calc_cpu_mask(connection); | |
2521 | connection->receiver.reset_cpu_mask = 1; | |
2522 | connection->asender.reset_cpu_mask = 1; | |
2523 | connection->worker.reset_cpu_mask = 1; | |
afbbfa88 AG |
2524 | } |
2525 | err = 0; | |
2526 | ||
2527 | fail: | |
2528 | free_cpumask_var(new_cpu_mask); | |
2529 | return err; | |
2530 | ||
2531 | } | |
2532 | ||
77c556f6 AG |
2533 | struct drbd_resource *drbd_create_resource(const char *name) |
2534 | { | |
2535 | struct drbd_resource *resource; | |
2536 | ||
2537 | resource = kmalloc(sizeof(struct drbd_resource), GFP_KERNEL); | |
2538 | if (!resource) | |
2539 | return NULL; | |
2540 | resource->name = kstrdup(name, GFP_KERNEL); | |
2541 | if (!resource->name) { | |
2542 | kfree(resource); | |
2543 | return NULL; | |
2544 | } | |
2545 | kref_init(&resource->kref); | |
2546 | INIT_LIST_HEAD(&resource->connections); | |
2547 | list_add_tail_rcu(&resource->resources, &drbd_resources); | |
2548 | return resource; | |
2549 | } | |
2550 | ||
ec0bddbc | 2551 | /* caller must be under genl_lock() */ |
bde89a9e | 2552 | struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts) |
2111438b | 2553 | { |
77c556f6 | 2554 | struct drbd_resource *resource; |
bde89a9e | 2555 | struct drbd_connection *connection; |
2111438b | 2556 | |
bde89a9e AG |
2557 | connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL); |
2558 | if (!connection) | |
2111438b PR |
2559 | return NULL; |
2560 | ||
bde89a9e | 2561 | if (drbd_alloc_socket(&connection->data)) |
e6ef8a5c | 2562 | goto fail; |
bde89a9e | 2563 | if (drbd_alloc_socket(&connection->meta)) |
e6ef8a5c AG |
2564 | goto fail; |
2565 | ||
bde89a9e | 2566 | if (!zalloc_cpumask_var(&connection->cpu_mask, GFP_KERNEL)) |
774b3055 PR |
2567 | goto fail; |
2568 | ||
bde89a9e | 2569 | if (set_resource_options(connection, res_opts)) |
afbbfa88 AG |
2570 | goto fail; |
2571 | ||
bde89a9e AG |
2572 | connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL); |
2573 | if (!connection->current_epoch) | |
12038a3a | 2574 | goto fail; |
b6dd1a89 | 2575 | |
bde89a9e | 2576 | INIT_LIST_HEAD(&connection->transfer_log); |
b6dd1a89 | 2577 | |
bde89a9e AG |
2578 | INIT_LIST_HEAD(&connection->current_epoch->list); |
2579 | connection->epochs = 1; | |
2580 | spin_lock_init(&connection->epoch_lock); | |
2581 | connection->write_ordering = WO_bdev_flush; | |
4b0007c0 | 2582 | |
bde89a9e AG |
2583 | connection->send.seen_any_write_yet = false; |
2584 | connection->send.current_epoch_nr = 0; | |
2585 | connection->send.current_epoch_writes = 0; | |
b6dd1a89 | 2586 | |
77c556f6 AG |
2587 | resource = drbd_create_resource(name); |
2588 | if (!resource) | |
2589 | goto fail; | |
2590 | ||
bde89a9e AG |
2591 | connection->cstate = C_STANDALONE; |
2592 | mutex_init(&connection->cstate_mutex); | |
2593 | spin_lock_init(&connection->req_lock); | |
2594 | mutex_init(&connection->conf_update); | |
2595 | init_waitqueue_head(&connection->ping_wait); | |
2596 | idr_init(&connection->volumes); | |
b2fb6dbe | 2597 | |
bde89a9e AG |
2598 | drbd_init_workqueue(&connection->sender_work); |
2599 | mutex_init(&connection->data.mutex); | |
2600 | mutex_init(&connection->meta.mutex); | |
6699b655 | 2601 | |
bde89a9e AG |
2602 | drbd_thread_init(connection, &connection->receiver, drbdd_init, "receiver"); |
2603 | drbd_thread_init(connection, &connection->worker, drbd_worker, "worker"); | |
2604 | drbd_thread_init(connection, &connection->asender, drbd_asender, "asender"); | |
392c8801 | 2605 | |
bde89a9e | 2606 | kref_init(&connection->kref); |
77c556f6 AG |
2607 | |
2608 | kref_get(&resource->kref); | |
2609 | connection->resource = resource; | |
2610 | list_add_tail_rcu(&connection->connections, &resource->connections); | |
2111438b | 2611 | |
bde89a9e | 2612 | return connection; |
2111438b PR |
2613 | |
2614 | fail: | |
bde89a9e AG |
2615 | kfree(connection->current_epoch); |
2616 | free_cpumask_var(connection->cpu_mask); | |
2617 | drbd_free_socket(&connection->meta); | |
2618 | drbd_free_socket(&connection->data); | |
bde89a9e | 2619 | kfree(connection); |
2111438b PR |
2620 | |
2621 | return NULL; | |
2622 | } | |
2623 | ||
05a10ec7 | 2624 | void drbd_destroy_connection(struct kref *kref) |
2111438b | 2625 | { |
bde89a9e | 2626 | struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref); |
77c556f6 | 2627 | struct drbd_resource *resource = connection->resource; |
9dc9fbb3 | 2628 | |
bde89a9e AG |
2629 | if (atomic_read(&connection->current_epoch->epoch_size) != 0) |
2630 | conn_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size)); | |
2631 | kfree(connection->current_epoch); | |
12038a3a | 2632 | |
bde89a9e | 2633 | idr_destroy(&connection->volumes); |
2111438b | 2634 | |
bde89a9e AG |
2635 | free_cpumask_var(connection->cpu_mask); |
2636 | drbd_free_socket(&connection->meta); | |
2637 | drbd_free_socket(&connection->data); | |
bde89a9e AG |
2638 | kfree(connection->int_dig_in); |
2639 | kfree(connection->int_dig_vv); | |
2640 | kfree(connection); | |
77c556f6 | 2641 | kref_put(&resource->kref, drbd_destroy_resource); |
2111438b PR |
2642 | } |
2643 | ||
b30ab791 | 2644 | static int init_submitter(struct drbd_device *device) |
113fef9e LE |
2645 | { |
2646 | /* opencoded create_singlethread_workqueue(), | |
2647 | * to be able to say "drbd%d", ..., minor */ | |
b30ab791 AG |
2648 | device->submit.wq = alloc_workqueue("drbd%u_submit", |
2649 | WQ_UNBOUND | WQ_MEM_RECLAIM, 1, device->minor); | |
2650 | if (!device->submit.wq) | |
113fef9e LE |
2651 | return -ENOMEM; |
2652 | ||
b30ab791 AG |
2653 | INIT_WORK(&device->submit.worker, do_submit); |
2654 | spin_lock_init(&device->submit.lock); | |
2655 | INIT_LIST_HEAD(&device->submit.writes); | |
113fef9e LE |
2656 | return 0; |
2657 | } | |
2658 | ||
a6b32bc3 | 2659 | enum drbd_ret_code drbd_create_minor(struct drbd_connection *connection, unsigned int minor, int vnr) |
b411b363 | 2660 | { |
b30ab791 | 2661 | struct drbd_device *device; |
a6b32bc3 | 2662 | struct drbd_peer_device *peer_device; |
b411b363 PR |
2663 | struct gendisk *disk; |
2664 | struct request_queue *q; | |
93e4bf7a | 2665 | int id; |
8432b314 | 2666 | enum drbd_ret_code err = ERR_NOMEM; |
774b3055 | 2667 | |
b30ab791 AG |
2668 | device = minor_to_device(minor); |
2669 | if (device) | |
774b3055 | 2670 | return ERR_MINOR_EXISTS; |
b411b363 PR |
2671 | |
2672 | /* GFP_KERNEL, we are outside of all write-out paths */ | |
b30ab791 AG |
2673 | device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL); |
2674 | if (!device) | |
774b3055 | 2675 | return ERR_NOMEM; |
a6b32bc3 AG |
2676 | peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL); |
2677 | if (!peer_device) | |
2678 | goto out_no_peer_device; | |
774b3055 | 2679 | |
a6b32bc3 AG |
2680 | INIT_LIST_HEAD(&device->peer_devices); |
2681 | list_add(&peer_device->peer_devices, &device->peer_devices); | |
bde89a9e | 2682 | kref_get(&connection->kref); |
d8628a86 | 2683 | device->resource = connection->resource; |
a6b32bc3 AG |
2684 | peer_device->connection = connection; |
2685 | peer_device->device = device; | |
b411b363 | 2686 | |
b30ab791 AG |
2687 | device->minor = minor; |
2688 | device->vnr = vnr; | |
b411b363 | 2689 | |
b30ab791 | 2690 | drbd_init_set_defaults(device); |
b411b363 PR |
2691 | |
2692 | q = blk_alloc_queue(GFP_KERNEL); | |
2693 | if (!q) | |
2694 | goto out_no_q; | |
b30ab791 AG |
2695 | device->rq_queue = q; |
2696 | q->queuedata = device; | |
b411b363 PR |
2697 | |
2698 | disk = alloc_disk(1); | |
2699 | if (!disk) | |
2700 | goto out_no_disk; | |
b30ab791 | 2701 | device->vdisk = disk; |
b411b363 | 2702 | |
81e84650 | 2703 | set_disk_ro(disk, true); |
b411b363 PR |
2704 | |
2705 | disk->queue = q; | |
2706 | disk->major = DRBD_MAJOR; | |
2707 | disk->first_minor = minor; | |
2708 | disk->fops = &drbd_ops; | |
2709 | sprintf(disk->disk_name, "drbd%d", minor); | |
b30ab791 | 2710 | disk->private_data = device; |
b411b363 | 2711 | |
b30ab791 | 2712 | device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor)); |
b411b363 | 2713 | /* we have no partitions. we contain only ourselves. */ |
b30ab791 | 2714 | device->this_bdev->bd_contains = device->this_bdev; |
b411b363 PR |
2715 | |
2716 | q->backing_dev_info.congested_fn = drbd_congested; | |
b30ab791 | 2717 | q->backing_dev_info.congested_data = device; |
b411b363 | 2718 | |
2f58dcfc | 2719 | blk_queue_make_request(q, drbd_make_request); |
a73ff323 | 2720 | blk_queue_flush(q, REQ_FLUSH | REQ_FUA); |
99432fcc PR |
2721 | /* Setting the max_hw_sectors to an odd value of 8kibyte here |
2722 | This triggers a max_bio_size message upon first attach or connect */ | |
2723 | blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8); | |
b411b363 PR |
2724 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
2725 | blk_queue_merge_bvec(q, drbd_merge_bvec); | |
a6b32bc3 | 2726 | q->queue_lock = &first_peer_device(device)->connection->req_lock; /* needed since we use */ |
b411b363 | 2727 | |
b30ab791 AG |
2728 | device->md_io_page = alloc_page(GFP_KERNEL); |
2729 | if (!device->md_io_page) | |
b411b363 PR |
2730 | goto out_no_io_page; |
2731 | ||
b30ab791 | 2732 | if (drbd_bm_init(device)) |
b411b363 | 2733 | goto out_no_bitmap; |
b30ab791 AG |
2734 | device->read_requests = RB_ROOT; |
2735 | device->write_requests = RB_ROOT; | |
b411b363 | 2736 | |
93e4bf7a AG |
2737 | id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL); |
2738 | if (id < 0) { | |
2739 | if (id == -ENOSPC) { | |
56de2102 TH |
2740 | err = ERR_MINOR_EXISTS; |
2741 | drbd_msg_put_info("requested minor exists already"); | |
2742 | } | |
8432b314 | 2743 | goto out_no_minor_idr; |
81a5d60e | 2744 | } |
8432b314 | 2745 | |
93e4bf7a AG |
2746 | id = idr_alloc(&connection->volumes, device, vnr, vnr + 1, GFP_KERNEL); |
2747 | if (id < 0) { | |
2748 | if (id == -ENOSPC) { | |
56de2102 TH |
2749 | err = ERR_INVALID_REQUEST; |
2750 | drbd_msg_put_info("requested volume exists already"); | |
2751 | } | |
8432b314 | 2752 | goto out_idr_remove_minor; |
8432b314 | 2753 | } |
56de2102 | 2754 | |
b30ab791 | 2755 | if (init_submitter(device)) { |
113fef9e LE |
2756 | err = ERR_NOMEM; |
2757 | drbd_msg_put_info("unable to create submit workqueue"); | |
2758 | goto out_idr_remove_vol; | |
2759 | } | |
2760 | ||
774b3055 | 2761 | add_disk(disk); |
b30ab791 | 2762 | kref_init(&device->kref); /* one ref for both idrs and the the add_disk */ |
774b3055 | 2763 | |
2325eb66 | 2764 | /* inherit the connection state */ |
bde89a9e | 2765 | device->state.conn = connection->cstate; |
b30ab791 AG |
2766 | if (device->state.conn == C_WF_REPORT_PARAMS) |
2767 | drbd_connected(device); | |
2325eb66 | 2768 | |
774b3055 | 2769 | return NO_ERROR; |
b411b363 | 2770 | |
113fef9e | 2771 | out_idr_remove_vol: |
93e4bf7a | 2772 | idr_remove(&connection->volumes, vnr); |
8432b314 | 2773 | out_idr_remove_minor: |
93e4bf7a | 2774 | idr_remove(&drbd_devices, minor); |
569083c0 | 2775 | synchronize_rcu(); |
8432b314 | 2776 | out_no_minor_idr: |
b30ab791 | 2777 | drbd_bm_cleanup(device); |
b411b363 | 2778 | out_no_bitmap: |
b30ab791 | 2779 | __free_page(device->md_io_page); |
b411b363 PR |
2780 | out_no_io_page: |
2781 | put_disk(disk); | |
2782 | out_no_disk: | |
2783 | blk_cleanup_queue(q); | |
2784 | out_no_q: | |
05a10ec7 | 2785 | kref_put(&connection->kref, drbd_destroy_connection); |
a6b32bc3 AG |
2786 | out_no_peer_device: |
2787 | kfree(device); | |
8432b314 | 2788 | return err; |
b411b363 PR |
2789 | } |
2790 | ||
b411b363 PR |
2791 | int __init drbd_init(void) |
2792 | { | |
2793 | int err; | |
2794 | ||
2b8a90b5 | 2795 | if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) { |
b411b363 | 2796 | printk(KERN_ERR |
81a5d60e | 2797 | "drbd: invalid minor_count (%d)\n", minor_count); |
b411b363 PR |
2798 | #ifdef MODULE |
2799 | return -EINVAL; | |
2800 | #else | |
46530e85 | 2801 | minor_count = DRBD_MINOR_COUNT_DEF; |
b411b363 PR |
2802 | #endif |
2803 | } | |
2804 | ||
b411b363 PR |
2805 | err = register_blkdev(DRBD_MAJOR, "drbd"); |
2806 | if (err) { | |
2807 | printk(KERN_ERR | |
2808 | "drbd: unable to register block device major %d\n", | |
2809 | DRBD_MAJOR); | |
2810 | return err; | |
2811 | } | |
2812 | ||
2813 | register_reboot_notifier(&drbd_notifier); | |
2814 | ||
2815 | /* | |
2816 | * allocate all necessary structs | |
2817 | */ | |
b411b363 PR |
2818 | init_waitqueue_head(&drbd_pp_wait); |
2819 | ||
2820 | drbd_proc = NULL; /* play safe for drbd_cleanup */ | |
05a10ec7 | 2821 | idr_init(&drbd_devices); |
b411b363 | 2822 | |
69babf05 | 2823 | rwlock_init(&global_state_lock); |
77c556f6 | 2824 | INIT_LIST_HEAD(&drbd_resources); |
69babf05 LE |
2825 | |
2826 | err = drbd_genl_register(); | |
2827 | if (err) { | |
2828 | printk(KERN_ERR "drbd: unable to register generic netlink family\n"); | |
2829 | goto fail; | |
2830 | } | |
2831 | ||
b411b363 PR |
2832 | err = drbd_create_mempools(); |
2833 | if (err) | |
3b98c0c2 | 2834 | goto fail; |
b411b363 | 2835 | |
6110d70b | 2836 | err = -ENOMEM; |
8c484ee4 | 2837 | drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL); |
b411b363 PR |
2838 | if (!drbd_proc) { |
2839 | printk(KERN_ERR "drbd: unable to register proc file\n"); | |
3b98c0c2 | 2840 | goto fail; |
b411b363 PR |
2841 | } |
2842 | ||
2312f0b3 LE |
2843 | retry.wq = create_singlethread_workqueue("drbd-reissue"); |
2844 | if (!retry.wq) { | |
2845 | printk(KERN_ERR "drbd: unable to create retry workqueue\n"); | |
2846 | goto fail; | |
2847 | } | |
2848 | INIT_WORK(&retry.worker, do_retry); | |
2849 | spin_lock_init(&retry.lock); | |
2850 | INIT_LIST_HEAD(&retry.writes); | |
b411b363 PR |
2851 | |
2852 | printk(KERN_INFO "drbd: initialized. " | |
2853 | "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n", | |
2854 | API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX); | |
2855 | printk(KERN_INFO "drbd: %s\n", drbd_buildtag()); | |
2856 | printk(KERN_INFO "drbd: registered as block device major %d\n", | |
2857 | DRBD_MAJOR); | |
b411b363 PR |
2858 | |
2859 | return 0; /* Success! */ | |
2860 | ||
3b98c0c2 | 2861 | fail: |
b411b363 PR |
2862 | drbd_cleanup(); |
2863 | if (err == -ENOMEM) | |
b411b363 PR |
2864 | printk(KERN_ERR "drbd: ran out of memory\n"); |
2865 | else | |
2866 | printk(KERN_ERR "drbd: initialization failure\n"); | |
2867 | return err; | |
2868 | } | |
2869 | ||
2870 | void drbd_free_bc(struct drbd_backing_dev *ldev) | |
2871 | { | |
2872 | if (ldev == NULL) | |
2873 | return; | |
2874 | ||
e525fd89 TH |
2875 | blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
2876 | blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
b411b363 | 2877 | |
94ad0a10 | 2878 | kfree(ldev->disk_conf); |
b411b363 PR |
2879 | kfree(ldev); |
2880 | } | |
2881 | ||
bde89a9e | 2882 | void drbd_free_sock(struct drbd_connection *connection) |
b411b363 | 2883 | { |
bde89a9e AG |
2884 | if (connection->data.socket) { |
2885 | mutex_lock(&connection->data.mutex); | |
2886 | kernel_sock_shutdown(connection->data.socket, SHUT_RDWR); | |
2887 | sock_release(connection->data.socket); | |
2888 | connection->data.socket = NULL; | |
2889 | mutex_unlock(&connection->data.mutex); | |
b411b363 | 2890 | } |
bde89a9e AG |
2891 | if (connection->meta.socket) { |
2892 | mutex_lock(&connection->meta.mutex); | |
2893 | kernel_sock_shutdown(connection->meta.socket, SHUT_RDWR); | |
2894 | sock_release(connection->meta.socket); | |
2895 | connection->meta.socket = NULL; | |
2896 | mutex_unlock(&connection->meta.mutex); | |
b411b363 PR |
2897 | } |
2898 | } | |
2899 | ||
b411b363 | 2900 | /* meta data management */ |
b411b363 | 2901 | |
bde89a9e | 2902 | void conn_md_sync(struct drbd_connection *connection) |
b411b363 | 2903 | { |
b30ab791 | 2904 | struct drbd_device *device; |
19fffd7b | 2905 | int vnr; |
b411b363 | 2906 | |
19fffd7b | 2907 | rcu_read_lock(); |
bde89a9e | 2908 | idr_for_each_entry(&connection->volumes, device, vnr) { |
b30ab791 | 2909 | kref_get(&device->kref); |
19fffd7b | 2910 | rcu_read_unlock(); |
b30ab791 | 2911 | drbd_md_sync(device); |
05a10ec7 | 2912 | kref_put(&device->kref, drbd_destroy_device); |
19fffd7b PR |
2913 | rcu_read_lock(); |
2914 | } | |
2915 | rcu_read_unlock(); | |
b411b363 PR |
2916 | } |
2917 | ||
ae8bf312 | 2918 | /* aligned 4kByte */ |
b411b363 | 2919 | struct meta_data_on_disk { |
cccac985 | 2920 | u64 la_size_sect; /* last agreed size. */ |
b411b363 PR |
2921 | u64 uuid[UI_SIZE]; /* UUIDs. */ |
2922 | u64 device_uuid; | |
2923 | u64 reserved_u64_1; | |
2924 | u32 flags; /* MDF */ | |
2925 | u32 magic; | |
2926 | u32 md_size_sect; | |
2927 | u32 al_offset; /* offset to this block */ | |
ae8bf312 | 2928 | u32 al_nr_extents; /* important for restoring the AL (userspace) */ |
f399002e | 2929 | /* `-- act_log->nr_elements <-- ldev->dc.al_extents */ |
b411b363 PR |
2930 | u32 bm_offset; /* offset to the bitmap, from here */ |
2931 | u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */ | |
99432fcc | 2932 | u32 la_peer_max_bio_size; /* last peer max_bio_size */ |
b411b363 | 2933 | |
3a4d4eb3 LE |
2934 | /* see al_tr_number_to_on_disk_sector() */ |
2935 | u32 al_stripes; | |
2936 | u32 al_stripe_size_4k; | |
2937 | ||
2938 | u8 reserved_u8[4096 - (7*8 + 10*4)]; | |
b411b363 PR |
2939 | } __packed; |
2940 | ||
d752b269 PR |
2941 | |
2942 | ||
b30ab791 | 2943 | void drbd_md_write(struct drbd_device *device, void *b) |
b411b363 | 2944 | { |
d752b269 | 2945 | struct meta_data_on_disk *buffer = b; |
b411b363 PR |
2946 | sector_t sector; |
2947 | int i; | |
2948 | ||
ae8bf312 | 2949 | memset(buffer, 0, sizeof(*buffer)); |
b411b363 | 2950 | |
b30ab791 | 2951 | buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev)); |
b411b363 | 2952 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
2953 | buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
2954 | buffer->flags = cpu_to_be32(device->ldev->md.flags); | |
d5d7ebd4 | 2955 | buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN); |
b411b363 | 2956 | |
b30ab791 AG |
2957 | buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect); |
2958 | buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset); | |
2959 | buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements); | |
b411b363 | 2960 | buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE); |
b30ab791 | 2961 | buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid); |
b411b363 | 2962 | |
b30ab791 AG |
2963 | buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset); |
2964 | buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size); | |
b411b363 | 2965 | |
b30ab791 AG |
2966 | buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes); |
2967 | buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k); | |
3a4d4eb3 | 2968 | |
b30ab791 AG |
2969 | D_ASSERT(drbd_md_ss(device->ldev) == device->ldev->md.md_offset); |
2970 | sector = device->ldev->md.md_offset; | |
b411b363 | 2971 | |
b30ab791 | 2972 | if (drbd_md_sync_page_io(device, device->ldev, sector, WRITE)) { |
b411b363 PR |
2973 | /* this was a try anyways ... */ |
2974 | dev_err(DEV, "meta data update failed!\n"); | |
b30ab791 | 2975 | drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR); |
b411b363 | 2976 | } |
d752b269 PR |
2977 | } |
2978 | ||
2979 | /** | |
2980 | * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set | |
b30ab791 | 2981 | * @device: DRBD device. |
d752b269 | 2982 | */ |
b30ab791 | 2983 | void drbd_md_sync(struct drbd_device *device) |
d752b269 PR |
2984 | { |
2985 | struct meta_data_on_disk *buffer; | |
2986 | ||
2987 | /* Don't accidentally change the DRBD meta data layout. */ | |
2988 | BUILD_BUG_ON(UI_SIZE != 4); | |
2989 | BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096); | |
2990 | ||
b30ab791 | 2991 | del_timer(&device->md_sync_timer); |
d752b269 | 2992 | /* timer may be rearmed by drbd_md_mark_dirty() now. */ |
b30ab791 | 2993 | if (!test_and_clear_bit(MD_DIRTY, &device->flags)) |
d752b269 PR |
2994 | return; |
2995 | ||
2996 | /* We use here D_FAILED and not D_ATTACHING because we try to write | |
2997 | * metadata even if we detach due to a disk failure! */ | |
b30ab791 | 2998 | if (!get_ldev_if_state(device, D_FAILED)) |
d752b269 PR |
2999 | return; |
3000 | ||
b30ab791 | 3001 | buffer = drbd_md_get_buffer(device); |
d752b269 PR |
3002 | if (!buffer) |
3003 | goto out; | |
3004 | ||
b30ab791 | 3005 | drbd_md_write(device, buffer); |
b411b363 | 3006 | |
b30ab791 | 3007 | /* Update device->ldev->md.la_size_sect, |
b411b363 | 3008 | * since we updated it on metadata. */ |
b30ab791 | 3009 | device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev); |
b411b363 | 3010 | |
b30ab791 | 3011 | drbd_md_put_buffer(device); |
e1711731 | 3012 | out: |
b30ab791 | 3013 | put_ldev(device); |
b411b363 PR |
3014 | } |
3015 | ||
b30ab791 | 3016 | static int check_activity_log_stripe_size(struct drbd_device *device, |
3a4d4eb3 LE |
3017 | struct meta_data_on_disk *on_disk, |
3018 | struct drbd_md *in_core) | |
3019 | { | |
3020 | u32 al_stripes = be32_to_cpu(on_disk->al_stripes); | |
3021 | u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k); | |
3022 | u64 al_size_4k; | |
3023 | ||
3024 | /* both not set: default to old fixed size activity log */ | |
3025 | if (al_stripes == 0 && al_stripe_size_4k == 0) { | |
3026 | al_stripes = 1; | |
3027 | al_stripe_size_4k = MD_32kB_SECT/8; | |
3028 | } | |
3029 | ||
3030 | /* some paranoia plausibility checks */ | |
3031 | ||
3032 | /* we need both values to be set */ | |
3033 | if (al_stripes == 0 || al_stripe_size_4k == 0) | |
3034 | goto err; | |
3035 | ||
3036 | al_size_4k = (u64)al_stripes * al_stripe_size_4k; | |
3037 | ||
3038 | /* Upper limit of activity log area, to avoid potential overflow | |
3039 | * problems in al_tr_number_to_on_disk_sector(). As right now, more | |
3040 | * than 72 * 4k blocks total only increases the amount of history, | |
3041 | * limiting this arbitrarily to 16 GB is not a real limitation ;-) */ | |
3042 | if (al_size_4k > (16 * 1024 * 1024/4)) | |
3043 | goto err; | |
3044 | ||
3045 | /* Lower limit: we need at least 8 transaction slots (32kB) | |
3046 | * to not break existing setups */ | |
3047 | if (al_size_4k < MD_32kB_SECT/8) | |
3048 | goto err; | |
3049 | ||
3050 | in_core->al_stripe_size_4k = al_stripe_size_4k; | |
3051 | in_core->al_stripes = al_stripes; | |
3052 | in_core->al_size_4k = al_size_4k; | |
3053 | ||
3054 | return 0; | |
3055 | err: | |
3056 | dev_err(DEV, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n", | |
3057 | al_stripes, al_stripe_size_4k); | |
3058 | return -EINVAL; | |
3059 | } | |
3060 | ||
b30ab791 | 3061 | static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev) |
c04ccaa6 LE |
3062 | { |
3063 | sector_t capacity = drbd_get_capacity(bdev->md_bdev); | |
3064 | struct drbd_md *in_core = &bdev->md; | |
3065 | s32 on_disk_al_sect; | |
3066 | s32 on_disk_bm_sect; | |
3067 | ||
3068 | /* The on-disk size of the activity log, calculated from offsets, and | |
3069 | * the size of the activity log calculated from the stripe settings, | |
3070 | * should match. | |
3071 | * Though we could relax this a bit: it is ok, if the striped activity log | |
3072 | * fits in the available on-disk activity log size. | |
3073 | * Right now, that would break how resize is implemented. | |
3074 | * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware | |
3075 | * of possible unused padding space in the on disk layout. */ | |
3076 | if (in_core->al_offset < 0) { | |
3077 | if (in_core->bm_offset > in_core->al_offset) | |
3078 | goto err; | |
3079 | on_disk_al_sect = -in_core->al_offset; | |
3080 | on_disk_bm_sect = in_core->al_offset - in_core->bm_offset; | |
3081 | } else { | |
3082 | if (in_core->al_offset != MD_4kB_SECT) | |
3083 | goto err; | |
3084 | if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT) | |
3085 | goto err; | |
3086 | ||
3087 | on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT; | |
3088 | on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset; | |
3089 | } | |
3090 | ||
3091 | /* old fixed size meta data is exactly that: fixed. */ | |
3092 | if (in_core->meta_dev_idx >= 0) { | |
3093 | if (in_core->md_size_sect != MD_128MB_SECT | |
3094 | || in_core->al_offset != MD_4kB_SECT | |
3095 | || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT | |
3096 | || in_core->al_stripes != 1 | |
3097 | || in_core->al_stripe_size_4k != MD_32kB_SECT/8) | |
3098 | goto err; | |
3099 | } | |
3100 | ||
3101 | if (capacity < in_core->md_size_sect) | |
3102 | goto err; | |
3103 | if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev)) | |
3104 | goto err; | |
3105 | ||
3106 | /* should be aligned, and at least 32k */ | |
3107 | if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT)) | |
3108 | goto err; | |
3109 | ||
3110 | /* should fit (for now: exactly) into the available on-disk space; | |
3111 | * overflow prevention is in check_activity_log_stripe_size() above. */ | |
3112 | if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT) | |
3113 | goto err; | |
3114 | ||
3115 | /* again, should be aligned */ | |
3116 | if (in_core->bm_offset & 7) | |
3117 | goto err; | |
3118 | ||
3119 | /* FIXME check for device grow with flex external meta data? */ | |
3120 | ||
3121 | /* can the available bitmap space cover the last agreed device size? */ | |
3122 | if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512) | |
3123 | goto err; | |
3124 | ||
3125 | return 0; | |
3126 | ||
3127 | err: | |
3128 | dev_err(DEV, "meta data offsets don't make sense: idx=%d " | |
3129 | "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, " | |
3130 | "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n", | |
3131 | in_core->meta_dev_idx, | |
3132 | in_core->al_stripes, in_core->al_stripe_size_4k, | |
3133 | in_core->al_offset, in_core->bm_offset, in_core->md_size_sect, | |
3134 | (unsigned long long)in_core->la_size_sect, | |
3135 | (unsigned long long)capacity); | |
3136 | ||
3137 | return -EINVAL; | |
3138 | } | |
3139 | ||
3140 | ||
b411b363 PR |
3141 | /** |
3142 | * drbd_md_read() - Reads in the meta data super block | |
b30ab791 | 3143 | * @device: DRBD device. |
b411b363 PR |
3144 | * @bdev: Device from which the meta data should be read in. |
3145 | * | |
3a4d4eb3 | 3146 | * Return NO_ERROR on success, and an enum drbd_ret_code in case |
d5d7ebd4 | 3147 | * something goes wrong. |
3a4d4eb3 | 3148 | * |
c04ccaa6 | 3149 | * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS, |
b30ab791 | 3150 | * even before @bdev is assigned to @device->ldev. |
b411b363 | 3151 | */ |
b30ab791 | 3152 | int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev) |
b411b363 PR |
3153 | { |
3154 | struct meta_data_on_disk *buffer; | |
d5d7ebd4 | 3155 | u32 magic, flags; |
b411b363 PR |
3156 | int i, rv = NO_ERROR; |
3157 | ||
b30ab791 | 3158 | if (device->state.disk != D_DISKLESS) |
c04ccaa6 | 3159 | return ERR_DISK_CONFIGURED; |
b411b363 | 3160 | |
b30ab791 | 3161 | buffer = drbd_md_get_buffer(device); |
e1711731 | 3162 | if (!buffer) |
c04ccaa6 | 3163 | return ERR_NOMEM; |
b411b363 | 3164 | |
c04ccaa6 LE |
3165 | /* First, figure out where our meta data superblock is located, |
3166 | * and read it. */ | |
3a4d4eb3 LE |
3167 | bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx; |
3168 | bdev->md.md_offset = drbd_md_ss(bdev); | |
b411b363 | 3169 | |
b30ab791 | 3170 | if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, READ)) { |
25985edc | 3171 | /* NOTE: can't do normal error processing here as this is |
b411b363 PR |
3172 | called BEFORE disk is attached */ |
3173 | dev_err(DEV, "Error while reading metadata.\n"); | |
3174 | rv = ERR_IO_MD_DISK; | |
3175 | goto err; | |
3176 | } | |
3177 | ||
d5d7ebd4 LE |
3178 | magic = be32_to_cpu(buffer->magic); |
3179 | flags = be32_to_cpu(buffer->flags); | |
3180 | if (magic == DRBD_MD_MAGIC_84_UNCLEAN || | |
3181 | (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) { | |
3182 | /* btw: that's Activity Log clean, not "all" clean. */ | |
3183 | dev_err(DEV, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n"); | |
3184 | rv = ERR_MD_UNCLEAN; | |
3185 | goto err; | |
3186 | } | |
3a4d4eb3 LE |
3187 | |
3188 | rv = ERR_MD_INVALID; | |
d5d7ebd4 | 3189 | if (magic != DRBD_MD_MAGIC_08) { |
43de7c85 | 3190 | if (magic == DRBD_MD_MAGIC_07) |
d5d7ebd4 LE |
3191 | dev_err(DEV, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n"); |
3192 | else | |
3193 | dev_err(DEV, "Meta data magic not found. Did you \"drbdadm create-md\"?\n"); | |
b411b363 PR |
3194 | goto err; |
3195 | } | |
3a4d4eb3 | 3196 | |
c04ccaa6 LE |
3197 | if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) { |
3198 | dev_err(DEV, "unexpected bm_bytes_per_bit: %u (expected %u)\n", | |
3199 | be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE); | |
b411b363 PR |
3200 | goto err; |
3201 | } | |
3a4d4eb3 | 3202 | |
c04ccaa6 LE |
3203 | |
3204 | /* convert to in_core endian */ | |
3205 | bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect); | |
3206 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
3207 | bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]); | |
3208 | bdev->md.flags = be32_to_cpu(buffer->flags); | |
3209 | bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid); | |
3210 | ||
3211 | bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect); | |
3212 | bdev->md.al_offset = be32_to_cpu(buffer->al_offset); | |
3213 | bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset); | |
3214 | ||
b30ab791 | 3215 | if (check_activity_log_stripe_size(device, buffer, &bdev->md)) |
b411b363 | 3216 | goto err; |
b30ab791 | 3217 | if (check_offsets_and_sizes(device, bdev)) |
c04ccaa6 LE |
3218 | goto err; |
3219 | ||
b411b363 PR |
3220 | if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) { |
3221 | dev_err(DEV, "unexpected bm_offset: %d (expected %d)\n", | |
3222 | be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset); | |
b411b363 PR |
3223 | goto err; |
3224 | } | |
3225 | if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) { | |
3226 | dev_err(DEV, "unexpected md_size: %u (expected %u)\n", | |
3227 | be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect); | |
b411b363 PR |
3228 | goto err; |
3229 | } | |
3230 | ||
3a4d4eb3 | 3231 | rv = NO_ERROR; |
b411b363 | 3232 | |
a6b32bc3 | 3233 | spin_lock_irq(&first_peer_device(device)->connection->req_lock); |
b30ab791 | 3234 | if (device->state.conn < C_CONNECTED) { |
db141b2f | 3235 | unsigned int peer; |
99432fcc | 3236 | peer = be32_to_cpu(buffer->la_peer_max_bio_size); |
db141b2f | 3237 | peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE); |
b30ab791 | 3238 | device->peer_max_bio_size = peer; |
99432fcc | 3239 | } |
a6b32bc3 | 3240 | spin_unlock_irq(&first_peer_device(device)->connection->req_lock); |
b411b363 PR |
3241 | |
3242 | err: | |
b30ab791 | 3243 | drbd_md_put_buffer(device); |
b411b363 PR |
3244 | |
3245 | return rv; | |
3246 | } | |
3247 | ||
3248 | /** | |
3249 | * drbd_md_mark_dirty() - Mark meta data super block as dirty | |
b30ab791 | 3250 | * @device: DRBD device. |
b411b363 PR |
3251 | * |
3252 | * Call this function if you change anything that should be written to | |
3253 | * the meta-data super block. This function sets MD_DIRTY, and starts a | |
3254 | * timer that ensures that within five seconds you have to call drbd_md_sync(). | |
3255 | */ | |
ca0e6098 | 3256 | #ifdef DEBUG |
b30ab791 | 3257 | void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func) |
ee15b038 | 3258 | { |
b30ab791 AG |
3259 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) { |
3260 | mod_timer(&device->md_sync_timer, jiffies + HZ); | |
3261 | device->last_md_mark_dirty.line = line; | |
3262 | device->last_md_mark_dirty.func = func; | |
ee15b038 LE |
3263 | } |
3264 | } | |
3265 | #else | |
b30ab791 | 3266 | void drbd_md_mark_dirty(struct drbd_device *device) |
b411b363 | 3267 | { |
b30ab791 AG |
3268 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) |
3269 | mod_timer(&device->md_sync_timer, jiffies + 5*HZ); | |
b411b363 | 3270 | } |
ee15b038 | 3271 | #endif |
b411b363 | 3272 | |
b30ab791 | 3273 | void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3274 | { |
3275 | int i; | |
3276 | ||
62b0da3a | 3277 | for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) |
b30ab791 | 3278 | device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i]; |
b411b363 PR |
3279 | } |
3280 | ||
b30ab791 | 3281 | void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 PR |
3282 | { |
3283 | if (idx == UI_CURRENT) { | |
b30ab791 | 3284 | if (device->state.role == R_PRIMARY) |
b411b363 PR |
3285 | val |= 1; |
3286 | else | |
3287 | val &= ~((u64)1); | |
3288 | ||
b30ab791 | 3289 | drbd_set_ed_uuid(device, val); |
b411b363 PR |
3290 | } |
3291 | ||
b30ab791 AG |
3292 | device->ldev->md.uuid[idx] = val; |
3293 | drbd_md_mark_dirty(device); | |
b411b363 PR |
3294 | } |
3295 | ||
b30ab791 | 3296 | void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
9f2247bb PR |
3297 | { |
3298 | unsigned long flags; | |
b30ab791 AG |
3299 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3300 | __drbd_uuid_set(device, idx, val); | |
3301 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
9f2247bb | 3302 | } |
b411b363 | 3303 | |
b30ab791 | 3304 | void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 | 3305 | { |
9f2247bb | 3306 | unsigned long flags; |
b30ab791 AG |
3307 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3308 | if (device->ldev->md.uuid[idx]) { | |
3309 | drbd_uuid_move_history(device); | |
3310 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx]; | |
b411b363 | 3311 | } |
b30ab791 AG |
3312 | __drbd_uuid_set(device, idx, val); |
3313 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
b411b363 PR |
3314 | } |
3315 | ||
3316 | /** | |
3317 | * drbd_uuid_new_current() - Creates a new current UUID | |
b30ab791 | 3318 | * @device: DRBD device. |
b411b363 PR |
3319 | * |
3320 | * Creates a new current UUID, and rotates the old current UUID into | |
3321 | * the bitmap slot. Causes an incremental resync upon next connect. | |
3322 | */ | |
b30ab791 | 3323 | void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3324 | { |
3325 | u64 val; | |
9f2247bb PR |
3326 | unsigned long long bm_uuid; |
3327 | ||
3328 | get_random_bytes(&val, sizeof(u64)); | |
3329 | ||
b30ab791 AG |
3330 | spin_lock_irq(&device->ldev->md.uuid_lock); |
3331 | bm_uuid = device->ldev->md.uuid[UI_BITMAP]; | |
62b0da3a LE |
3332 | |
3333 | if (bm_uuid) | |
3334 | dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid); | |
b411b363 | 3335 | |
b30ab791 AG |
3336 | device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT]; |
3337 | __drbd_uuid_set(device, UI_CURRENT, val); | |
3338 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 3339 | |
b30ab791 | 3340 | drbd_print_uuids(device, "new current UUID"); |
aaa8e2b3 | 3341 | /* get it to stable storage _now_ */ |
b30ab791 | 3342 | drbd_md_sync(device); |
b411b363 PR |
3343 | } |
3344 | ||
b30ab791 | 3345 | void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local) |
b411b363 | 3346 | { |
9f2247bb | 3347 | unsigned long flags; |
b30ab791 | 3348 | if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0) |
b411b363 PR |
3349 | return; |
3350 | ||
b30ab791 | 3351 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
b411b363 | 3352 | if (val == 0) { |
b30ab791 AG |
3353 | drbd_uuid_move_history(device); |
3354 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; | |
3355 | device->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 3356 | } else { |
b30ab791 | 3357 | unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP]; |
62b0da3a LE |
3358 | if (bm_uuid) |
3359 | dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid); | |
b411b363 | 3360 | |
b30ab791 | 3361 | device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1); |
b411b363 | 3362 | } |
b30ab791 | 3363 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); |
9f2247bb | 3364 | |
b30ab791 | 3365 | drbd_md_mark_dirty(device); |
b411b363 PR |
3366 | } |
3367 | ||
3368 | /** | |
3369 | * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3370 | * @device: DRBD device. |
b411b363 PR |
3371 | * |
3372 | * Sets all bits in the bitmap and writes the whole bitmap to stable storage. | |
3373 | */ | |
b30ab791 | 3374 | int drbd_bmio_set_n_write(struct drbd_device *device) |
b411b363 PR |
3375 | { |
3376 | int rv = -EIO; | |
3377 | ||
b30ab791 AG |
3378 | if (get_ldev_if_state(device, D_ATTACHING)) { |
3379 | drbd_md_set_flag(device, MDF_FULL_SYNC); | |
3380 | drbd_md_sync(device); | |
3381 | drbd_bm_set_all(device); | |
b411b363 | 3382 | |
b30ab791 | 3383 | rv = drbd_bm_write(device); |
b411b363 PR |
3384 | |
3385 | if (!rv) { | |
b30ab791 AG |
3386 | drbd_md_clear_flag(device, MDF_FULL_SYNC); |
3387 | drbd_md_sync(device); | |
b411b363 PR |
3388 | } |
3389 | ||
b30ab791 | 3390 | put_ldev(device); |
b411b363 PR |
3391 | } |
3392 | ||
3393 | return rv; | |
3394 | } | |
3395 | ||
3396 | /** | |
3397 | * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3398 | * @device: DRBD device. |
b411b363 PR |
3399 | * |
3400 | * Clears all bits in the bitmap and writes the whole bitmap to stable storage. | |
3401 | */ | |
b30ab791 | 3402 | int drbd_bmio_clear_n_write(struct drbd_device *device) |
b411b363 PR |
3403 | { |
3404 | int rv = -EIO; | |
3405 | ||
b30ab791 AG |
3406 | drbd_resume_al(device); |
3407 | if (get_ldev_if_state(device, D_ATTACHING)) { | |
3408 | drbd_bm_clear_all(device); | |
3409 | rv = drbd_bm_write(device); | |
3410 | put_ldev(device); | |
b411b363 PR |
3411 | } |
3412 | ||
3413 | return rv; | |
3414 | } | |
3415 | ||
99920dc5 | 3416 | static int w_bitmap_io(struct drbd_work *w, int unused) |
b411b363 PR |
3417 | { |
3418 | struct bm_io_work *work = container_of(w, struct bm_io_work, w); | |
b30ab791 | 3419 | struct drbd_device *device = w->device; |
02851e9f | 3420 | int rv = -EIO; |
b411b363 | 3421 | |
b30ab791 | 3422 | D_ASSERT(atomic_read(&device->ap_bio_cnt) == 0); |
b411b363 | 3423 | |
b30ab791 AG |
3424 | if (get_ldev(device)) { |
3425 | drbd_bm_lock(device, work->why, work->flags); | |
3426 | rv = work->io_fn(device); | |
3427 | drbd_bm_unlock(device); | |
3428 | put_ldev(device); | |
02851e9f | 3429 | } |
b411b363 | 3430 | |
b30ab791 AG |
3431 | clear_bit_unlock(BITMAP_IO, &device->flags); |
3432 | wake_up(&device->misc_wait); | |
b411b363 PR |
3433 | |
3434 | if (work->done) | |
b30ab791 | 3435 | work->done(device, rv); |
b411b363 | 3436 | |
b30ab791 | 3437 | clear_bit(BITMAP_IO_QUEUED, &device->flags); |
b411b363 | 3438 | work->why = NULL; |
20ceb2b2 | 3439 | work->flags = 0; |
b411b363 | 3440 | |
99920dc5 | 3441 | return 0; |
b411b363 PR |
3442 | } |
3443 | ||
b30ab791 | 3444 | void drbd_ldev_destroy(struct drbd_device *device) |
82f59cc6 | 3445 | { |
b30ab791 AG |
3446 | lc_destroy(device->resync); |
3447 | device->resync = NULL; | |
3448 | lc_destroy(device->act_log); | |
3449 | device->act_log = NULL; | |
82f59cc6 | 3450 | __no_warn(local, |
b30ab791 AG |
3451 | drbd_free_bc(device->ldev); |
3452 | device->ldev = NULL;); | |
82f59cc6 | 3453 | |
b30ab791 | 3454 | clear_bit(GO_DISKLESS, &device->flags); |
82f59cc6 LE |
3455 | } |
3456 | ||
99920dc5 | 3457 | static int w_go_diskless(struct drbd_work *w, int unused) |
e9e6f3ec | 3458 | { |
b30ab791 | 3459 | struct drbd_device *device = w->device; |
00d56944 | 3460 | |
b30ab791 | 3461 | D_ASSERT(device->state.disk == D_FAILED); |
9d282875 LE |
3462 | /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will |
3463 | * inc/dec it frequently. Once we are D_DISKLESS, no one will touch | |
82f59cc6 LE |
3464 | * the protected members anymore, though, so once put_ldev reaches zero |
3465 | * again, it will be safe to free them. */ | |
a2a3c74f LE |
3466 | |
3467 | /* Try to write changed bitmap pages, read errors may have just | |
3468 | * set some bits outside the area covered by the activity log. | |
3469 | * | |
3470 | * If we have an IO error during the bitmap writeout, | |
3471 | * we will want a full sync next time, just in case. | |
3472 | * (Do we want a specific meta data flag for this?) | |
3473 | * | |
3474 | * If that does not make it to stable storage either, | |
fd0017c1 PR |
3475 | * we cannot do anything about that anymore. |
3476 | * | |
3477 | * We still need to check if both bitmap and ldev are present, we may | |
3478 | * end up here after a failed attach, before ldev was even assigned. | |
3479 | */ | |
b30ab791 | 3480 | if (device->bitmap && device->ldev) { |
bb45185d PR |
3481 | /* An interrupted resync or similar is allowed to recounts bits |
3482 | * while we detach. | |
3483 | * Any modifications would not be expected anymore, though. | |
3484 | */ | |
b30ab791 | 3485 | if (drbd_bitmap_io_from_worker(device, drbd_bm_write, |
bb45185d | 3486 | "detach", BM_LOCKED_TEST_ALLOWED)) { |
b30ab791 AG |
3487 | if (test_bit(WAS_READ_ERROR, &device->flags)) { |
3488 | drbd_md_set_flag(device, MDF_FULL_SYNC); | |
3489 | drbd_md_sync(device); | |
a2a3c74f LE |
3490 | } |
3491 | } | |
3492 | } | |
3493 | ||
b30ab791 | 3494 | drbd_force_state(device, NS(disk, D_DISKLESS)); |
99920dc5 | 3495 | return 0; |
e9e6f3ec LE |
3496 | } |
3497 | ||
b411b363 PR |
3498 | /** |
3499 | * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap | |
b30ab791 | 3500 | * @device: DRBD device. |
b411b363 PR |
3501 | * @io_fn: IO callback to be called when bitmap IO is possible |
3502 | * @done: callback to be called after the bitmap IO was performed | |
3503 | * @why: Descriptive text of the reason for doing the IO | |
3504 | * | |
3505 | * While IO on the bitmap happens we freeze application IO thus we ensure | |
3506 | * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be | |
3507 | * called from worker context. It MUST NOT be used while a previous such | |
3508 | * work is still pending! | |
3509 | */ | |
b30ab791 | 3510 | void drbd_queue_bitmap_io(struct drbd_device *device, |
54761697 AG |
3511 | int (*io_fn)(struct drbd_device *), |
3512 | void (*done)(struct drbd_device *, int), | |
20ceb2b2 | 3513 | char *why, enum bm_flag flags) |
b411b363 | 3514 | { |
a6b32bc3 | 3515 | D_ASSERT(current == first_peer_device(device)->connection->worker.task); |
b411b363 | 3516 | |
b30ab791 AG |
3517 | D_ASSERT(!test_bit(BITMAP_IO_QUEUED, &device->flags)); |
3518 | D_ASSERT(!test_bit(BITMAP_IO, &device->flags)); | |
3519 | D_ASSERT(list_empty(&device->bm_io_work.w.list)); | |
3520 | if (device->bm_io_work.why) | |
b411b363 | 3521 | dev_err(DEV, "FIXME going to queue '%s' but '%s' still pending?\n", |
b30ab791 | 3522 | why, device->bm_io_work.why); |
b411b363 | 3523 | |
b30ab791 AG |
3524 | device->bm_io_work.io_fn = io_fn; |
3525 | device->bm_io_work.done = done; | |
3526 | device->bm_io_work.why = why; | |
3527 | device->bm_io_work.flags = flags; | |
b411b363 | 3528 | |
a6b32bc3 | 3529 | spin_lock_irq(&first_peer_device(device)->connection->req_lock); |
b30ab791 AG |
3530 | set_bit(BITMAP_IO, &device->flags); |
3531 | if (atomic_read(&device->ap_bio_cnt) == 0) { | |
3532 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags)) | |
a6b32bc3 | 3533 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, &device->bm_io_work.w); |
b411b363 | 3534 | } |
a6b32bc3 | 3535 | spin_unlock_irq(&first_peer_device(device)->connection->req_lock); |
b411b363 PR |
3536 | } |
3537 | ||
3538 | /** | |
3539 | * drbd_bitmap_io() - Does an IO operation on the whole bitmap | |
b30ab791 | 3540 | * @device: DRBD device. |
b411b363 PR |
3541 | * @io_fn: IO callback to be called when bitmap IO is possible |
3542 | * @why: Descriptive text of the reason for doing the IO | |
3543 | * | |
3544 | * freezes application IO while that the actual IO operations runs. This | |
3545 | * functions MAY NOT be called from worker context. | |
3546 | */ | |
b30ab791 | 3547 | int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *), |
20ceb2b2 | 3548 | char *why, enum bm_flag flags) |
b411b363 PR |
3549 | { |
3550 | int rv; | |
3551 | ||
a6b32bc3 | 3552 | D_ASSERT(current != first_peer_device(device)->connection->worker.task); |
b411b363 | 3553 | |
20ceb2b2 | 3554 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
b30ab791 | 3555 | drbd_suspend_io(device); |
b411b363 | 3556 | |
b30ab791 AG |
3557 | drbd_bm_lock(device, why, flags); |
3558 | rv = io_fn(device); | |
3559 | drbd_bm_unlock(device); | |
b411b363 | 3560 | |
20ceb2b2 | 3561 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
b30ab791 | 3562 | drbd_resume_io(device); |
b411b363 PR |
3563 | |
3564 | return rv; | |
3565 | } | |
3566 | ||
b30ab791 | 3567 | void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3568 | { |
b30ab791 AG |
3569 | if ((device->ldev->md.flags & flag) != flag) { |
3570 | drbd_md_mark_dirty(device); | |
3571 | device->ldev->md.flags |= flag; | |
b411b363 PR |
3572 | } |
3573 | } | |
3574 | ||
b30ab791 | 3575 | void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3576 | { |
b30ab791 AG |
3577 | if ((device->ldev->md.flags & flag) != 0) { |
3578 | drbd_md_mark_dirty(device); | |
3579 | device->ldev->md.flags &= ~flag; | |
b411b363 PR |
3580 | } |
3581 | } | |
3582 | int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag) | |
3583 | { | |
3584 | return (bdev->md.flags & flag) != 0; | |
3585 | } | |
3586 | ||
3587 | static void md_sync_timer_fn(unsigned long data) | |
3588 | { | |
b30ab791 | 3589 | struct drbd_device *device = (struct drbd_device *) data; |
b411b363 | 3590 | |
b792b655 | 3591 | /* must not double-queue! */ |
b30ab791 | 3592 | if (list_empty(&device->md_sync_work.list)) |
a6b32bc3 | 3593 | drbd_queue_work_front(&first_peer_device(device)->connection->sender_work, &device->md_sync_work); |
b411b363 PR |
3594 | } |
3595 | ||
99920dc5 | 3596 | static int w_md_sync(struct drbd_work *w, int unused) |
b411b363 | 3597 | { |
b30ab791 | 3598 | struct drbd_device *device = w->device; |
00d56944 | 3599 | |
b411b363 | 3600 | dev_warn(DEV, "md_sync_timer expired! Worker calls drbd_md_sync().\n"); |
ee15b038 LE |
3601 | #ifdef DEBUG |
3602 | dev_warn(DEV, "last md_mark_dirty: %s:%u\n", | |
b30ab791 | 3603 | device->last_md_mark_dirty.func, device->last_md_mark_dirty.line); |
ee15b038 | 3604 | #endif |
b30ab791 | 3605 | drbd_md_sync(device); |
99920dc5 | 3606 | return 0; |
b411b363 PR |
3607 | } |
3608 | ||
d8763023 | 3609 | const char *cmdname(enum drbd_packet cmd) |
f2ad9063 AG |
3610 | { |
3611 | /* THINK may need to become several global tables | |
3612 | * when we want to support more than | |
3613 | * one PRO_VERSION */ | |
3614 | static const char *cmdnames[] = { | |
3615 | [P_DATA] = "Data", | |
3616 | [P_DATA_REPLY] = "DataReply", | |
3617 | [P_RS_DATA_REPLY] = "RSDataReply", | |
3618 | [P_BARRIER] = "Barrier", | |
3619 | [P_BITMAP] = "ReportBitMap", | |
3620 | [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", | |
3621 | [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", | |
3622 | [P_UNPLUG_REMOTE] = "UnplugRemote", | |
3623 | [P_DATA_REQUEST] = "DataRequest", | |
3624 | [P_RS_DATA_REQUEST] = "RSDataRequest", | |
3625 | [P_SYNC_PARAM] = "SyncParam", | |
3626 | [P_SYNC_PARAM89] = "SyncParam89", | |
3627 | [P_PROTOCOL] = "ReportProtocol", | |
3628 | [P_UUIDS] = "ReportUUIDs", | |
3629 | [P_SIZES] = "ReportSizes", | |
3630 | [P_STATE] = "ReportState", | |
3631 | [P_SYNC_UUID] = "ReportSyncUUID", | |
3632 | [P_AUTH_CHALLENGE] = "AuthChallenge", | |
3633 | [P_AUTH_RESPONSE] = "AuthResponse", | |
3634 | [P_PING] = "Ping", | |
3635 | [P_PING_ACK] = "PingAck", | |
3636 | [P_RECV_ACK] = "RecvAck", | |
3637 | [P_WRITE_ACK] = "WriteAck", | |
3638 | [P_RS_WRITE_ACK] = "RSWriteAck", | |
d4dabbe2 | 3639 | [P_SUPERSEDED] = "Superseded", |
f2ad9063 AG |
3640 | [P_NEG_ACK] = "NegAck", |
3641 | [P_NEG_DREPLY] = "NegDReply", | |
3642 | [P_NEG_RS_DREPLY] = "NegRSDReply", | |
3643 | [P_BARRIER_ACK] = "BarrierAck", | |
3644 | [P_STATE_CHG_REQ] = "StateChgRequest", | |
3645 | [P_STATE_CHG_REPLY] = "StateChgReply", | |
3646 | [P_OV_REQUEST] = "OVRequest", | |
3647 | [P_OV_REPLY] = "OVReply", | |
3648 | [P_OV_RESULT] = "OVResult", | |
3649 | [P_CSUM_RS_REQUEST] = "CsumRSRequest", | |
3650 | [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", | |
3651 | [P_COMPRESSED_BITMAP] = "CBitmap", | |
3652 | [P_DELAY_PROBE] = "DelayProbe", | |
3653 | [P_OUT_OF_SYNC] = "OutOfSync", | |
7be8da07 | 3654 | [P_RETRY_WRITE] = "RetryWrite", |
ae25b336 LE |
3655 | [P_RS_CANCEL] = "RSCancel", |
3656 | [P_CONN_ST_CHG_REQ] = "conn_st_chg_req", | |
3657 | [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply", | |
036b17ea PR |
3658 | [P_RETRY_WRITE] = "retry_write", |
3659 | [P_PROTOCOL_UPDATE] = "protocol_update", | |
ae25b336 LE |
3660 | |
3661 | /* enum drbd_packet, but not commands - obsoleted flags: | |
3662 | * P_MAY_IGNORE | |
3663 | * P_MAX_OPT_CMD | |
3664 | */ | |
f2ad9063 AG |
3665 | }; |
3666 | ||
ae25b336 | 3667 | /* too big for the array: 0xfffX */ |
e5d6f33a AG |
3668 | if (cmd == P_INITIAL_META) |
3669 | return "InitialMeta"; | |
3670 | if (cmd == P_INITIAL_DATA) | |
3671 | return "InitialData"; | |
6038178e AG |
3672 | if (cmd == P_CONNECTION_FEATURES) |
3673 | return "ConnectionFeatures"; | |
6e849ce8 | 3674 | if (cmd >= ARRAY_SIZE(cmdnames)) |
f2ad9063 AG |
3675 | return "Unknown"; |
3676 | return cmdnames[cmd]; | |
3677 | } | |
3678 | ||
7be8da07 AG |
3679 | /** |
3680 | * drbd_wait_misc - wait for a request to make progress | |
b30ab791 | 3681 | * @device: device associated with the request |
7be8da07 AG |
3682 | * @i: the struct drbd_interval embedded in struct drbd_request or |
3683 | * struct drbd_peer_request | |
3684 | */ | |
b30ab791 | 3685 | int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i) |
7be8da07 | 3686 | { |
44ed167d | 3687 | struct net_conf *nc; |
7be8da07 AG |
3688 | DEFINE_WAIT(wait); |
3689 | long timeout; | |
3690 | ||
44ed167d | 3691 | rcu_read_lock(); |
a6b32bc3 | 3692 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
44ed167d PR |
3693 | if (!nc) { |
3694 | rcu_read_unlock(); | |
7be8da07 | 3695 | return -ETIMEDOUT; |
44ed167d PR |
3696 | } |
3697 | timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT; | |
3698 | rcu_read_unlock(); | |
7be8da07 | 3699 | |
b30ab791 | 3700 | /* Indicate to wake up device->misc_wait on progress. */ |
7be8da07 | 3701 | i->waiting = true; |
b30ab791 | 3702 | prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE); |
a6b32bc3 | 3703 | spin_unlock_irq(&first_peer_device(device)->connection->req_lock); |
7be8da07 | 3704 | timeout = schedule_timeout(timeout); |
b30ab791 | 3705 | finish_wait(&device->misc_wait, &wait); |
a6b32bc3 | 3706 | spin_lock_irq(&first_peer_device(device)->connection->req_lock); |
b30ab791 | 3707 | if (!timeout || device->state.conn < C_CONNECTED) |
7be8da07 AG |
3708 | return -ETIMEDOUT; |
3709 | if (signal_pending(current)) | |
3710 | return -ERESTARTSYS; | |
3711 | return 0; | |
b411b363 PR |
3712 | } |
3713 | ||
3714 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
3715 | /* Fault insertion support including random number generator shamelessly | |
3716 | * stolen from kernel/rcutorture.c */ | |
3717 | struct fault_random_state { | |
3718 | unsigned long state; | |
3719 | unsigned long count; | |
3720 | }; | |
3721 | ||
3722 | #define FAULT_RANDOM_MULT 39916801 /* prime */ | |
3723 | #define FAULT_RANDOM_ADD 479001701 /* prime */ | |
3724 | #define FAULT_RANDOM_REFRESH 10000 | |
3725 | ||
3726 | /* | |
3727 | * Crude but fast random-number generator. Uses a linear congruential | |
3728 | * generator, with occasional help from get_random_bytes(). | |
3729 | */ | |
3730 | static unsigned long | |
3731 | _drbd_fault_random(struct fault_random_state *rsp) | |
3732 | { | |
3733 | long refresh; | |
3734 | ||
49829ea7 | 3735 | if (!rsp->count--) { |
b411b363 PR |
3736 | get_random_bytes(&refresh, sizeof(refresh)); |
3737 | rsp->state += refresh; | |
3738 | rsp->count = FAULT_RANDOM_REFRESH; | |
3739 | } | |
3740 | rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD; | |
3741 | return swahw32(rsp->state); | |
3742 | } | |
3743 | ||
3744 | static char * | |
3745 | _drbd_fault_str(unsigned int type) { | |
3746 | static char *_faults[] = { | |
3747 | [DRBD_FAULT_MD_WR] = "Meta-data write", | |
3748 | [DRBD_FAULT_MD_RD] = "Meta-data read", | |
3749 | [DRBD_FAULT_RS_WR] = "Resync write", | |
3750 | [DRBD_FAULT_RS_RD] = "Resync read", | |
3751 | [DRBD_FAULT_DT_WR] = "Data write", | |
3752 | [DRBD_FAULT_DT_RD] = "Data read", | |
3753 | [DRBD_FAULT_DT_RA] = "Data read ahead", | |
3754 | [DRBD_FAULT_BM_ALLOC] = "BM allocation", | |
6b4388ac PR |
3755 | [DRBD_FAULT_AL_EE] = "EE allocation", |
3756 | [DRBD_FAULT_RECEIVE] = "receive data corruption", | |
b411b363 PR |
3757 | }; |
3758 | ||
3759 | return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**"; | |
3760 | } | |
3761 | ||
3762 | unsigned int | |
b30ab791 | 3763 | _drbd_insert_fault(struct drbd_device *device, unsigned int type) |
b411b363 PR |
3764 | { |
3765 | static struct fault_random_state rrs = {0, 0}; | |
3766 | ||
3767 | unsigned int ret = ( | |
3768 | (fault_devs == 0 || | |
b30ab791 | 3769 | ((1 << device_to_minor(device)) & fault_devs) != 0) && |
b411b363 PR |
3770 | (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate)); |
3771 | ||
3772 | if (ret) { | |
3773 | fault_count++; | |
3774 | ||
7383506c | 3775 | if (__ratelimit(&drbd_ratelimit_state)) |
b411b363 PR |
3776 | dev_warn(DEV, "***Simulating %s failure\n", |
3777 | _drbd_fault_str(type)); | |
3778 | } | |
3779 | ||
3780 | return ret; | |
3781 | } | |
3782 | #endif | |
3783 | ||
3784 | const char *drbd_buildtag(void) | |
3785 | { | |
3786 | /* DRBD built from external sources has here a reference to the | |
3787 | git hash of the source code. */ | |
3788 | ||
3789 | static char buildtag[38] = "\0uilt-in"; | |
3790 | ||
3791 | if (buildtag[0] == 0) { | |
bc4854bc CW |
3792 | #ifdef MODULE |
3793 | sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion); | |
3794 | #else | |
3795 | buildtag[0] = 'b'; | |
b411b363 | 3796 | #endif |
b411b363 PR |
3797 | } |
3798 | ||
3799 | return buildtag; | |
3800 | } | |
3801 | ||
3802 | module_init(drbd_init) | |
3803 | module_exit(drbd_cleanup) | |
3804 | ||
b411b363 PR |
3805 | EXPORT_SYMBOL(drbd_conn_str); |
3806 | EXPORT_SYMBOL(drbd_role_str); | |
3807 | EXPORT_SYMBOL(drbd_disk_str); | |
3808 | EXPORT_SYMBOL(drbd_set_st_err_str); |