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 | |
0500813f | 201 | spin_lock_irq(&connection->resource->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) { |
1ec861eb | 230 | drbd_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) { |
1ec861eb | 235 | drbd_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) { |
1ec861eb | 241 | drbd_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 | } |
0500813f | 258 | spin_unlock_irq(&connection->resource->req_lock); |
19f843aa | 259 | |
b411b363 | 260 | return; |
b411b363 | 261 | |
b411b363 | 262 | bail: |
0500813f | 263 | spin_unlock_irq(&connection->resource->req_lock); |
bde89a9e | 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 | { |
0500813f | 287 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 288 | _tl_restart(connection, what); |
0500813f | 289 | spin_unlock_irq(&connection->resource->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 | |
0500813f | 314 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 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 | } |
0500813f | 322 | spin_unlock_irq(&connection->resource->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) { |
1ec861eb | 352 | drbd_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 | ||
1ec861eb | 364 | drbd_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: |
1ec861eb | 396 | drbd_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)) { | |
1ec861eb | 401 | drbd_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)) { | |
1ec861eb | 418 | drbd_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; | |
1ec861eb | 432 | drbd_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 | { |
c06ece6b AG |
499 | struct drbd_peer_device *peer_device; |
500 | int vnr = 0, minor = -1; | |
774b3055 | 501 | |
695d08fa | 502 | rcu_read_lock(); |
c06ece6b AG |
503 | peer_device = idr_get_next(&connection->peer_devices, &vnr); |
504 | if (peer_device) | |
505 | minor = device_to_minor(peer_device->device); | |
695d08fa PR |
506 | rcu_read_unlock(); |
507 | ||
c06ece6b | 508 | return minor; |
80822284 | 509 | } |
774b3055 | 510 | |
b411b363 PR |
511 | #ifdef CONFIG_SMP |
512 | /** | |
513 | * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs | |
b411b363 | 514 | * |
625a6ba2 | 515 | * Forces all threads of a resource onto the same CPU. This is beneficial for |
b411b363 PR |
516 | * DRBD's performance. May be overwritten by user's configuration. |
517 | */ | |
625a6ba2 | 518 | static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask) |
b411b363 | 519 | { |
625a6ba2 | 520 | unsigned int *resources_per_cpu, min_index = ~0; |
b411b363 | 521 | |
625a6ba2 AG |
522 | resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL); |
523 | if (resources_per_cpu) { | |
524 | struct drbd_resource *resource; | |
525 | unsigned int cpu, min = ~0; | |
b411b363 | 526 | |
625a6ba2 AG |
527 | rcu_read_lock(); |
528 | for_each_resource_rcu(resource, &drbd_resources) { | |
529 | for_each_cpu(cpu, resource->cpu_mask) | |
530 | resources_per_cpu[cpu]++; | |
b411b363 | 531 | } |
625a6ba2 AG |
532 | rcu_read_unlock(); |
533 | for_each_online_cpu(cpu) { | |
534 | if (resources_per_cpu[cpu] < min) { | |
535 | min = resources_per_cpu[cpu]; | |
536 | min_index = cpu; | |
537 | } | |
538 | } | |
539 | kfree(resources_per_cpu); | |
540 | } | |
541 | if (min_index == ~0) { | |
542 | cpumask_setall(*cpu_mask); | |
543 | return; | |
b411b363 | 544 | } |
625a6ba2 | 545 | cpumask_set_cpu(min_index, *cpu_mask); |
b411b363 PR |
546 | } |
547 | ||
548 | /** | |
549 | * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread | |
b30ab791 | 550 | * @device: DRBD device. |
bc31fe33 | 551 | * @thi: drbd_thread object |
b411b363 PR |
552 | * |
553 | * call in the "main loop" of _all_ threads, no need for any mutex, current won't die | |
554 | * prematurely. | |
555 | */ | |
80822284 | 556 | void drbd_thread_current_set_cpu(struct drbd_thread *thi) |
b411b363 PR |
557 | { |
558 | struct task_struct *p = current; | |
bed879ae | 559 | |
b411b363 PR |
560 | if (!thi->reset_cpu_mask) |
561 | return; | |
562 | thi->reset_cpu_mask = 0; | |
625a6ba2 | 563 | set_cpus_allowed_ptr(p, thi->connection->resource->cpu_mask); |
b411b363 | 564 | } |
625a6ba2 AG |
565 | #else |
566 | #define drbd_calc_cpu_mask(A) ({}) | |
b411b363 PR |
567 | #endif |
568 | ||
52b061a4 AG |
569 | /** |
570 | * drbd_header_size - size of a packet header | |
571 | * | |
572 | * The header size is a multiple of 8, so any payload following the header is | |
573 | * word aligned on 64-bit architectures. (The bitmap send and receive code | |
574 | * relies on this.) | |
575 | */ | |
bde89a9e | 576 | unsigned int drbd_header_size(struct drbd_connection *connection) |
b411b363 | 577 | { |
bde89a9e | 578 | if (connection->agreed_pro_version >= 100) { |
0c8e36d9 AG |
579 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8)); |
580 | return sizeof(struct p_header100); | |
581 | } else { | |
582 | BUILD_BUG_ON(sizeof(struct p_header80) != | |
583 | sizeof(struct p_header95)); | |
584 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8)); | |
585 | return sizeof(struct p_header80); | |
586 | } | |
52b061a4 | 587 | } |
b411b363 | 588 | |
e658983a | 589 | static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
590 | { |
591 | h->magic = cpu_to_be32(DRBD_MAGIC); | |
592 | h->command = cpu_to_be16(cmd); | |
593 | h->length = cpu_to_be16(size); | |
e658983a | 594 | return sizeof(struct p_header80); |
fd340c12 | 595 | } |
b411b363 | 596 | |
e658983a | 597 | static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
598 | { |
599 | h->magic = cpu_to_be16(DRBD_MAGIC_BIG); | |
b411b363 | 600 | h->command = cpu_to_be16(cmd); |
b55d84ba | 601 | h->length = cpu_to_be32(size); |
e658983a | 602 | return sizeof(struct p_header95); |
fd340c12 | 603 | } |
b411b363 | 604 | |
0c8e36d9 AG |
605 | static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd, |
606 | int size, int vnr) | |
607 | { | |
608 | h->magic = cpu_to_be32(DRBD_MAGIC_100); | |
609 | h->volume = cpu_to_be16(vnr); | |
610 | h->command = cpu_to_be16(cmd); | |
611 | h->length = cpu_to_be32(size); | |
612 | h->pad = 0; | |
613 | return sizeof(struct p_header100); | |
614 | } | |
b411b363 | 615 | |
bde89a9e | 616 | static unsigned int prepare_header(struct drbd_connection *connection, int vnr, |
0c8e36d9 | 617 | void *buffer, enum drbd_packet cmd, int size) |
d38e787e | 618 | { |
bde89a9e | 619 | if (connection->agreed_pro_version >= 100) |
0c8e36d9 | 620 | return prepare_header100(buffer, cmd, size, vnr); |
bde89a9e | 621 | else if (connection->agreed_pro_version >= 95 && |
0c8e36d9 | 622 | size > DRBD_MAX_SIZE_H80_PACKET) |
e658983a | 623 | return prepare_header95(buffer, cmd, size); |
d38e787e | 624 | else |
e658983a | 625 | return prepare_header80(buffer, cmd, size); |
b411b363 PR |
626 | } |
627 | ||
bde89a9e | 628 | static void *__conn_prepare_command(struct drbd_connection *connection, |
a7eb7bdf | 629 | struct drbd_socket *sock) |
b411b363 | 630 | { |
a7eb7bdf AG |
631 | if (!sock->socket) |
632 | return NULL; | |
bde89a9e | 633 | return sock->sbuf + drbd_header_size(connection); |
a7eb7bdf | 634 | } |
b411b363 | 635 | |
bde89a9e | 636 | void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock) |
dba58587 | 637 | { |
a7eb7bdf | 638 | void *p; |
b411b363 | 639 | |
dba58587 | 640 | mutex_lock(&sock->mutex); |
bde89a9e | 641 | p = __conn_prepare_command(connection, sock); |
a7eb7bdf | 642 | if (!p) |
dba58587 | 643 | mutex_unlock(&sock->mutex); |
b411b363 | 644 | |
a7eb7bdf | 645 | return p; |
b411b363 PR |
646 | } |
647 | ||
b30ab791 | 648 | void *drbd_prepare_command(struct drbd_device *device, struct drbd_socket *sock) |
b411b363 | 649 | { |
a6b32bc3 | 650 | return conn_prepare_command(first_peer_device(device)->connection, sock); |
dba58587 | 651 | } |
b411b363 | 652 | |
bde89a9e | 653 | static int __send_command(struct drbd_connection *connection, int vnr, |
dba58587 AG |
654 | struct drbd_socket *sock, enum drbd_packet cmd, |
655 | unsigned int header_size, void *data, | |
656 | unsigned int size) | |
657 | { | |
658 | int msg_flags; | |
659 | int err; | |
b411b363 | 660 | |
dba58587 AG |
661 | /* |
662 | * Called with @data == NULL and the size of the data blocks in @size | |
663 | * for commands that send data blocks. For those commands, omit the | |
664 | * MSG_MORE flag: this will increase the likelihood that data blocks | |
665 | * which are page aligned on the sender will end up page aligned on the | |
666 | * receiver. | |
667 | */ | |
668 | msg_flags = data ? MSG_MORE : 0; | |
669 | ||
bde89a9e | 670 | header_size += prepare_header(connection, vnr, sock->sbuf, cmd, |
e658983a | 671 | header_size + size); |
bde89a9e | 672 | err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size, |
dba58587 AG |
673 | msg_flags); |
674 | if (data && !err) | |
bde89a9e | 675 | err = drbd_send_all(connection, sock->socket, data, size, 0); |
dba58587 AG |
676 | return err; |
677 | } | |
678 | ||
bde89a9e | 679 | static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
a7eb7bdf AG |
680 | enum drbd_packet cmd, unsigned int header_size, |
681 | void *data, unsigned int size) | |
682 | { | |
bde89a9e | 683 | return __send_command(connection, 0, sock, cmd, header_size, data, size); |
a7eb7bdf AG |
684 | } |
685 | ||
bde89a9e | 686 | int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
dba58587 AG |
687 | enum drbd_packet cmd, unsigned int header_size, |
688 | void *data, unsigned int size) | |
689 | { | |
690 | int err; | |
b411b363 | 691 | |
bde89a9e | 692 | err = __conn_send_command(connection, sock, cmd, header_size, data, size); |
dba58587 AG |
693 | mutex_unlock(&sock->mutex); |
694 | return err; | |
695 | } | |
696 | ||
b30ab791 | 697 | int drbd_send_command(struct drbd_device *device, struct drbd_socket *sock, |
dba58587 AG |
698 | enum drbd_packet cmd, unsigned int header_size, |
699 | void *data, unsigned int size) | |
700 | { | |
701 | int err; | |
702 | ||
a6b32bc3 | 703 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, cmd, header_size, |
dba58587 AG |
704 | data, size); |
705 | mutex_unlock(&sock->mutex); | |
706 | return err; | |
707 | } | |
b411b363 | 708 | |
bde89a9e | 709 | int drbd_send_ping(struct drbd_connection *connection) |
e307f352 | 710 | { |
9f5bdc33 AG |
711 | struct drbd_socket *sock; |
712 | ||
bde89a9e AG |
713 | sock = &connection->meta; |
714 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 715 | return -EIO; |
bde89a9e | 716 | return conn_send_command(connection, sock, P_PING, 0, NULL, 0); |
e307f352 | 717 | } |
b411b363 | 718 | |
bde89a9e | 719 | int drbd_send_ping_ack(struct drbd_connection *connection) |
e307f352 | 720 | { |
9f5bdc33 AG |
721 | struct drbd_socket *sock; |
722 | ||
bde89a9e AG |
723 | sock = &connection->meta; |
724 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 725 | return -EIO; |
bde89a9e | 726 | return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0); |
b411b363 PR |
727 | } |
728 | ||
b30ab791 | 729 | int drbd_send_sync_param(struct drbd_device *device) |
b411b363 | 730 | { |
7c96715a | 731 | struct drbd_socket *sock; |
8e26f9cc | 732 | struct p_rs_param_95 *p; |
9f5bdc33 | 733 | int size; |
a6b32bc3 | 734 | const int apv = first_peer_device(device)->connection->agreed_pro_version; |
9f5bdc33 | 735 | enum drbd_packet cmd; |
44ed167d | 736 | struct net_conf *nc; |
daeda1cc | 737 | struct disk_conf *dc; |
9f5bdc33 | 738 | |
a6b32bc3 | 739 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 740 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
741 | if (!p) |
742 | return -EIO; | |
b411b363 | 743 | |
44ed167d | 744 | rcu_read_lock(); |
a6b32bc3 | 745 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
b411b363 PR |
746 | |
747 | size = apv <= 87 ? sizeof(struct p_rs_param) | |
748 | : apv == 88 ? sizeof(struct p_rs_param) | |
44ed167d | 749 | + strlen(nc->verify_alg) + 1 |
8e26f9cc PR |
750 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
751 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 752 | |
9f5bdc33 | 753 | cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM; |
b411b363 | 754 | |
9f5bdc33 AG |
755 | /* initialize verify_alg and csums_alg */ |
756 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
b411b363 | 757 | |
b30ab791 AG |
758 | if (get_ldev(device)) { |
759 | dc = rcu_dereference(device->ldev->disk_conf); | |
6394b935 | 760 | p->resync_rate = cpu_to_be32(dc->resync_rate); |
daeda1cc PR |
761 | p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead); |
762 | p->c_delay_target = cpu_to_be32(dc->c_delay_target); | |
763 | p->c_fill_target = cpu_to_be32(dc->c_fill_target); | |
764 | p->c_max_rate = cpu_to_be32(dc->c_max_rate); | |
b30ab791 | 765 | put_ldev(device); |
9f5bdc33 | 766 | } else { |
6394b935 | 767 | p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF); |
9f5bdc33 AG |
768 | p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF); |
769 | p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF); | |
770 | p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF); | |
771 | p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF); | |
772 | } | |
b411b363 | 773 | |
9f5bdc33 | 774 | if (apv >= 88) |
44ed167d | 775 | strcpy(p->verify_alg, nc->verify_alg); |
9f5bdc33 | 776 | if (apv >= 89) |
44ed167d PR |
777 | strcpy(p->csums_alg, nc->csums_alg); |
778 | rcu_read_unlock(); | |
b411b363 | 779 | |
b30ab791 | 780 | return drbd_send_command(device, sock, cmd, size, NULL, 0); |
b411b363 PR |
781 | } |
782 | ||
bde89a9e | 783 | int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd) |
b411b363 | 784 | { |
9f5bdc33 | 785 | struct drbd_socket *sock; |
b411b363 | 786 | struct p_protocol *p; |
44ed167d | 787 | struct net_conf *nc; |
9f5bdc33 | 788 | int size, cf; |
b411b363 | 789 | |
bde89a9e AG |
790 | sock = &connection->data; |
791 | p = __conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
792 | if (!p) |
793 | return -EIO; | |
b411b363 | 794 | |
44ed167d | 795 | rcu_read_lock(); |
bde89a9e | 796 | nc = rcu_dereference(connection->net_conf); |
b411b363 | 797 | |
bde89a9e | 798 | if (nc->tentative && connection->agreed_pro_version < 92) { |
44ed167d PR |
799 | rcu_read_unlock(); |
800 | mutex_unlock(&sock->mutex); | |
1ec861eb | 801 | drbd_err(connection, "--dry-run is not supported by peer"); |
44ed167d PR |
802 | return -EOPNOTSUPP; |
803 | } | |
b411b363 | 804 | |
9f5bdc33 | 805 | size = sizeof(*p); |
bde89a9e | 806 | if (connection->agreed_pro_version >= 87) |
44ed167d | 807 | size += strlen(nc->integrity_alg) + 1; |
b411b363 | 808 | |
44ed167d PR |
809 | p->protocol = cpu_to_be32(nc->wire_protocol); |
810 | p->after_sb_0p = cpu_to_be32(nc->after_sb_0p); | |
811 | p->after_sb_1p = cpu_to_be32(nc->after_sb_1p); | |
812 | p->after_sb_2p = cpu_to_be32(nc->after_sb_2p); | |
813 | p->two_primaries = cpu_to_be32(nc->two_primaries); | |
cf14c2e9 | 814 | cf = 0; |
6139f60d AG |
815 | if (nc->discard_my_data) |
816 | cf |= CF_DISCARD_MY_DATA; | |
6dff2902 | 817 | if (nc->tentative) |
9f5bdc33 | 818 | cf |= CF_DRY_RUN; |
cf14c2e9 PR |
819 | p->conn_flags = cpu_to_be32(cf); |
820 | ||
bde89a9e | 821 | if (connection->agreed_pro_version >= 87) |
44ed167d PR |
822 | strcpy(p->integrity_alg, nc->integrity_alg); |
823 | rcu_read_unlock(); | |
b411b363 | 824 | |
bde89a9e | 825 | return __conn_send_command(connection, sock, cmd, size, NULL, 0); |
a7eb7bdf AG |
826 | } |
827 | ||
bde89a9e | 828 | int drbd_send_protocol(struct drbd_connection *connection) |
a7eb7bdf AG |
829 | { |
830 | int err; | |
831 | ||
bde89a9e AG |
832 | mutex_lock(&connection->data.mutex); |
833 | err = __drbd_send_protocol(connection, P_PROTOCOL); | |
834 | mutex_unlock(&connection->data.mutex); | |
a7eb7bdf AG |
835 | |
836 | return err; | |
b411b363 PR |
837 | } |
838 | ||
b30ab791 | 839 | static int _drbd_send_uuids(struct drbd_device *device, u64 uuid_flags) |
b411b363 | 840 | { |
9f5bdc33 AG |
841 | struct drbd_socket *sock; |
842 | struct p_uuids *p; | |
b411b363 PR |
843 | int i; |
844 | ||
b30ab791 | 845 | if (!get_ldev_if_state(device, D_NEGOTIATING)) |
2ae5f95b | 846 | return 0; |
b411b363 | 847 | |
a6b32bc3 | 848 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 849 | p = drbd_prepare_command(device, sock); |
9f5bdc33 | 850 | if (!p) { |
b30ab791 | 851 | put_ldev(device); |
9f5bdc33 AG |
852 | return -EIO; |
853 | } | |
b30ab791 | 854 | spin_lock_irq(&device->ldev->md.uuid_lock); |
b411b363 | 855 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
856 | p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
857 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 858 | |
b30ab791 AG |
859 | device->comm_bm_set = drbd_bm_total_weight(device); |
860 | p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set); | |
44ed167d | 861 | rcu_read_lock(); |
a6b32bc3 | 862 | uuid_flags |= rcu_dereference(first_peer_device(device)->connection->net_conf)->discard_my_data ? 1 : 0; |
44ed167d | 863 | rcu_read_unlock(); |
b30ab791 AG |
864 | uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0; |
865 | uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0; | |
9f5bdc33 | 866 | p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags); |
b411b363 | 867 | |
b30ab791 AG |
868 | put_ldev(device); |
869 | return drbd_send_command(device, sock, P_UUIDS, sizeof(*p), NULL, 0); | |
b411b363 PR |
870 | } |
871 | ||
b30ab791 | 872 | int drbd_send_uuids(struct drbd_device *device) |
b411b363 | 873 | { |
b30ab791 | 874 | return _drbd_send_uuids(device, 0); |
b411b363 PR |
875 | } |
876 | ||
b30ab791 | 877 | int drbd_send_uuids_skip_initial_sync(struct drbd_device *device) |
b411b363 | 878 | { |
b30ab791 | 879 | return _drbd_send_uuids(device, 8); |
b411b363 PR |
880 | } |
881 | ||
b30ab791 | 882 | void drbd_print_uuids(struct drbd_device *device, const char *text) |
62b0da3a | 883 | { |
b30ab791 AG |
884 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
885 | u64 *uuid = device->ldev->md.uuid; | |
d0180171 | 886 | drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n", |
62b0da3a LE |
887 | text, |
888 | (unsigned long long)uuid[UI_CURRENT], | |
889 | (unsigned long long)uuid[UI_BITMAP], | |
890 | (unsigned long long)uuid[UI_HISTORY_START], | |
891 | (unsigned long long)uuid[UI_HISTORY_END]); | |
b30ab791 | 892 | put_ldev(device); |
62b0da3a | 893 | } else { |
d0180171 | 894 | drbd_info(device, "%s effective data uuid: %016llX\n", |
62b0da3a | 895 | text, |
b30ab791 | 896 | (unsigned long long)device->ed_uuid); |
62b0da3a LE |
897 | } |
898 | } | |
899 | ||
b30ab791 | 900 | void drbd_gen_and_send_sync_uuid(struct drbd_device *device) |
b411b363 | 901 | { |
9f5bdc33 AG |
902 | struct drbd_socket *sock; |
903 | struct p_rs_uuid *p; | |
5a22db89 LE |
904 | u64 uuid; |
905 | ||
0b0ba1ef | 906 | D_ASSERT(device, device->state.disk == D_UP_TO_DATE); |
b411b363 | 907 | |
b30ab791 | 908 | uuid = device->ldev->md.uuid[UI_BITMAP]; |
5ba3dac5 PR |
909 | if (uuid && uuid != UUID_JUST_CREATED) |
910 | uuid = uuid + UUID_NEW_BM_OFFSET; | |
911 | else | |
912 | get_random_bytes(&uuid, sizeof(u64)); | |
b30ab791 AG |
913 | drbd_uuid_set(device, UI_BITMAP, uuid); |
914 | drbd_print_uuids(device, "updated sync UUID"); | |
915 | drbd_md_sync(device); | |
b411b363 | 916 | |
a6b32bc3 | 917 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 918 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
919 | if (p) { |
920 | p->uuid = cpu_to_be64(uuid); | |
b30ab791 | 921 | drbd_send_command(device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0); |
9f5bdc33 | 922 | } |
b411b363 PR |
923 | } |
924 | ||
b30ab791 | 925 | int drbd_send_sizes(struct drbd_device *device, int trigger_reply, enum dds_flags flags) |
b411b363 | 926 | { |
9f5bdc33 AG |
927 | struct drbd_socket *sock; |
928 | struct p_sizes *p; | |
b411b363 | 929 | sector_t d_size, u_size; |
db141b2f LE |
930 | int q_order_type; |
931 | unsigned int max_bio_size; | |
b411b363 | 932 | |
b30ab791 | 933 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
0b0ba1ef | 934 | D_ASSERT(device, device->ldev->backing_bdev); |
b30ab791 | 935 | d_size = drbd_get_max_capacity(device->ldev); |
daeda1cc | 936 | rcu_read_lock(); |
b30ab791 | 937 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
daeda1cc | 938 | rcu_read_unlock(); |
b30ab791 AG |
939 | q_order_type = drbd_queue_order_type(device); |
940 | max_bio_size = queue_max_hw_sectors(device->ldev->backing_bdev->bd_disk->queue) << 9; | |
db141b2f | 941 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE); |
b30ab791 | 942 | put_ldev(device); |
b411b363 PR |
943 | } else { |
944 | d_size = 0; | |
945 | u_size = 0; | |
946 | q_order_type = QUEUE_ORDERED_NONE; | |
99432fcc | 947 | max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */ |
b411b363 PR |
948 | } |
949 | ||
a6b32bc3 | 950 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 951 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
952 | if (!p) |
953 | return -EIO; | |
6809384c | 954 | |
a6b32bc3 | 955 | if (first_peer_device(device)->connection->agreed_pro_version <= 94) |
98683650 | 956 | max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET); |
a6b32bc3 | 957 | else if (first_peer_device(device)->connection->agreed_pro_version < 100) |
98683650 | 958 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95); |
b411b363 | 959 | |
9f5bdc33 AG |
960 | p->d_size = cpu_to_be64(d_size); |
961 | p->u_size = cpu_to_be64(u_size); | |
b30ab791 | 962 | p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev)); |
9f5bdc33 AG |
963 | p->max_bio_size = cpu_to_be32(max_bio_size); |
964 | p->queue_order_type = cpu_to_be16(q_order_type); | |
965 | p->dds_flags = cpu_to_be16(flags); | |
b30ab791 | 966 | return drbd_send_command(device, sock, P_SIZES, sizeof(*p), NULL, 0); |
b411b363 PR |
967 | } |
968 | ||
969 | /** | |
f479ea06 | 970 | * drbd_send_current_state() - Sends the drbd state to the peer |
b30ab791 | 971 | * @device: DRBD device. |
b411b363 | 972 | */ |
b30ab791 | 973 | int drbd_send_current_state(struct drbd_device *device) |
b411b363 | 974 | { |
7c96715a | 975 | struct drbd_socket *sock; |
9f5bdc33 | 976 | struct p_state *p; |
b411b363 | 977 | |
a6b32bc3 | 978 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 979 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
980 | if (!p) |
981 | return -EIO; | |
b30ab791 AG |
982 | p->state = cpu_to_be32(device->state.i); /* Within the send mutex */ |
983 | return drbd_send_command(device, sock, P_STATE, sizeof(*p), NULL, 0); | |
b411b363 PR |
984 | } |
985 | ||
f479ea06 LE |
986 | /** |
987 | * drbd_send_state() - After a state change, sends the new state to the peer | |
b30ab791 | 988 | * @device: DRBD device. |
43de7c85 | 989 | * @state: the state to send, not necessarily the current state. |
f479ea06 LE |
990 | * |
991 | * Each state change queues an "after_state_ch" work, which will eventually | |
992 | * send the resulting new state to the peer. If more state changes happen | |
993 | * between queuing and processing of the after_state_ch work, we still | |
994 | * want to send each intermediary state in the order it occurred. | |
995 | */ | |
b30ab791 | 996 | int drbd_send_state(struct drbd_device *device, union drbd_state state) |
f479ea06 | 997 | { |
43de7c85 PR |
998 | struct drbd_socket *sock; |
999 | struct p_state *p; | |
f479ea06 | 1000 | |
a6b32bc3 | 1001 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1002 | p = drbd_prepare_command(device, sock); |
43de7c85 PR |
1003 | if (!p) |
1004 | return -EIO; | |
1005 | p->state = cpu_to_be32(state.i); /* Within the send mutex */ | |
b30ab791 | 1006 | return drbd_send_command(device, sock, P_STATE, sizeof(*p), NULL, 0); |
43de7c85 | 1007 | } |
f479ea06 | 1008 | |
b30ab791 | 1009 | int drbd_send_state_req(struct drbd_device *device, union drbd_state mask, union drbd_state val) |
9f5bdc33 AG |
1010 | { |
1011 | struct drbd_socket *sock; | |
1012 | struct p_req_state *p; | |
f479ea06 | 1013 | |
a6b32bc3 | 1014 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1015 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1016 | if (!p) |
1017 | return -EIO; | |
1018 | p->mask = cpu_to_be32(mask.i); | |
1019 | p->val = cpu_to_be32(val.i); | |
b30ab791 | 1020 | return drbd_send_command(device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0); |
b411b363 | 1021 | } |
f479ea06 | 1022 | |
bde89a9e | 1023 | int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val) |
b411b363 | 1024 | { |
9f5bdc33 AG |
1025 | enum drbd_packet cmd; |
1026 | struct drbd_socket *sock; | |
1027 | struct p_req_state *p; | |
f479ea06 | 1028 | |
bde89a9e AG |
1029 | cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ; |
1030 | sock = &connection->data; | |
1031 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1032 | if (!p) |
1033 | return -EIO; | |
1034 | p->mask = cpu_to_be32(mask.i); | |
1035 | p->val = cpu_to_be32(val.i); | |
bde89a9e | 1036 | return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
f479ea06 LE |
1037 | } |
1038 | ||
b30ab791 | 1039 | void drbd_send_sr_reply(struct drbd_device *device, enum drbd_state_rv retcode) |
b411b363 | 1040 | { |
9f5bdc33 AG |
1041 | struct drbd_socket *sock; |
1042 | struct p_req_state_reply *p; | |
b411b363 | 1043 | |
a6b32bc3 | 1044 | sock = &first_peer_device(device)->connection->meta; |
b30ab791 | 1045 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1046 | if (p) { |
1047 | p->retcode = cpu_to_be32(retcode); | |
b30ab791 | 1048 | drbd_send_command(device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0); |
9f5bdc33 | 1049 | } |
b411b363 | 1050 | } |
b411b363 | 1051 | |
bde89a9e | 1052 | void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode) |
047cd4a6 | 1053 | { |
9f5bdc33 AG |
1054 | struct drbd_socket *sock; |
1055 | struct p_req_state_reply *p; | |
bde89a9e | 1056 | enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY; |
b411b363 | 1057 | |
bde89a9e AG |
1058 | sock = &connection->meta; |
1059 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1060 | if (p) { |
1061 | p->retcode = cpu_to_be32(retcode); | |
bde89a9e | 1062 | conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
9f5bdc33 | 1063 | } |
b411b363 PR |
1064 | } |
1065 | ||
a02d1240 | 1066 | static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) |
b411b363 | 1067 | { |
a02d1240 AG |
1068 | BUG_ON(code & ~0xf); |
1069 | p->encoding = (p->encoding & ~0xf) | code; | |
1070 | } | |
b411b363 | 1071 | |
a02d1240 AG |
1072 | static void dcbp_set_start(struct p_compressed_bm *p, int set) |
1073 | { | |
1074 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
1075 | } | |
b411b363 | 1076 | |
a02d1240 AG |
1077 | static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n) |
1078 | { | |
1079 | BUG_ON(n & ~0x7); | |
1080 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
b411b363 PR |
1081 | } |
1082 | ||
b30ab791 | 1083 | static int fill_bitmap_rle_bits(struct drbd_device *device, |
50d0b1ad AG |
1084 | struct p_compressed_bm *p, |
1085 | unsigned int size, | |
1086 | struct bm_xfer_ctx *c) | |
b411b363 PR |
1087 | { |
1088 | struct bitstream bs; | |
1089 | unsigned long plain_bits; | |
1090 | unsigned long tmp; | |
1091 | unsigned long rl; | |
1092 | unsigned len; | |
1093 | unsigned toggle; | |
44ed167d | 1094 | int bits, use_rle; |
b411b363 PR |
1095 | |
1096 | /* may we use this feature? */ | |
44ed167d | 1097 | rcu_read_lock(); |
a6b32bc3 | 1098 | use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle; |
44ed167d | 1099 | rcu_read_unlock(); |
a6b32bc3 | 1100 | if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90) |
44ed167d | 1101 | return 0; |
b411b363 PR |
1102 | |
1103 | if (c->bit_offset >= c->bm_bits) | |
1104 | return 0; /* nothing to do. */ | |
1105 | ||
1106 | /* use at most thus many bytes */ | |
50d0b1ad AG |
1107 | bitstream_init(&bs, p->code, size, 0); |
1108 | memset(p->code, 0, size); | |
b411b363 PR |
1109 | /* plain bits covered in this code string */ |
1110 | plain_bits = 0; | |
1111 | ||
1112 | /* p->encoding & 0x80 stores whether the first run length is set. | |
1113 | * bit offset is implicit. | |
1114 | * start with toggle == 2 to be able to tell the first iteration */ | |
1115 | toggle = 2; | |
1116 | ||
1117 | /* see how much plain bits we can stuff into one packet | |
1118 | * using RLE and VLI. */ | |
1119 | do { | |
b30ab791 AG |
1120 | tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset) |
1121 | : _drbd_bm_find_next(device, c->bit_offset); | |
b411b363 PR |
1122 | if (tmp == -1UL) |
1123 | tmp = c->bm_bits; | |
1124 | rl = tmp - c->bit_offset; | |
1125 | ||
1126 | if (toggle == 2) { /* first iteration */ | |
1127 | if (rl == 0) { | |
1128 | /* the first checked bit was set, | |
1129 | * store start value, */ | |
a02d1240 | 1130 | dcbp_set_start(p, 1); |
b411b363 PR |
1131 | /* but skip encoding of zero run length */ |
1132 | toggle = !toggle; | |
1133 | continue; | |
1134 | } | |
a02d1240 | 1135 | dcbp_set_start(p, 0); |
b411b363 PR |
1136 | } |
1137 | ||
1138 | /* paranoia: catch zero runlength. | |
1139 | * can only happen if bitmap is modified while we scan it. */ | |
1140 | if (rl == 0) { | |
d0180171 | 1141 | drbd_err(device, "unexpected zero runlength while encoding bitmap " |
b411b363 PR |
1142 | "t:%u bo:%lu\n", toggle, c->bit_offset); |
1143 | return -1; | |
1144 | } | |
1145 | ||
1146 | bits = vli_encode_bits(&bs, rl); | |
1147 | if (bits == -ENOBUFS) /* buffer full */ | |
1148 | break; | |
1149 | if (bits <= 0) { | |
d0180171 | 1150 | drbd_err(device, "error while encoding bitmap: %d\n", bits); |
b411b363 PR |
1151 | return 0; |
1152 | } | |
1153 | ||
1154 | toggle = !toggle; | |
1155 | plain_bits += rl; | |
1156 | c->bit_offset = tmp; | |
1157 | } while (c->bit_offset < c->bm_bits); | |
1158 | ||
1159 | len = bs.cur.b - p->code + !!bs.cur.bit; | |
1160 | ||
1161 | if (plain_bits < (len << 3)) { | |
1162 | /* incompressible with this method. | |
1163 | * we need to rewind both word and bit position. */ | |
1164 | c->bit_offset -= plain_bits; | |
1165 | bm_xfer_ctx_bit_to_word_offset(c); | |
1166 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1167 | return 0; | |
1168 | } | |
1169 | ||
1170 | /* RLE + VLI was able to compress it just fine. | |
1171 | * update c->word_offset. */ | |
1172 | bm_xfer_ctx_bit_to_word_offset(c); | |
1173 | ||
1174 | /* store pad_bits */ | |
a02d1240 | 1175 | dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7); |
b411b363 PR |
1176 | |
1177 | return len; | |
1178 | } | |
1179 | ||
f70af118 AG |
1180 | /** |
1181 | * send_bitmap_rle_or_plain | |
1182 | * | |
1183 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
1184 | * code upon failure. | |
1185 | */ | |
1186 | static int | |
b30ab791 | 1187 | send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c) |
b411b363 | 1188 | { |
a6b32bc3 AG |
1189 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
1190 | unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); | |
e658983a | 1191 | struct p_compressed_bm *p = sock->sbuf + header_size; |
a982dd57 | 1192 | int len, err; |
b411b363 | 1193 | |
b30ab791 | 1194 | len = fill_bitmap_rle_bits(device, p, |
e658983a | 1195 | DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c); |
b411b363 | 1196 | if (len < 0) |
f70af118 | 1197 | return -EIO; |
b411b363 PR |
1198 | |
1199 | if (len) { | |
a02d1240 | 1200 | dcbp_set_code(p, RLE_VLI_Bits); |
a6b32bc3 | 1201 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, |
9f5bdc33 AG |
1202 | P_COMPRESSED_BITMAP, sizeof(*p) + len, |
1203 | NULL, 0); | |
b411b363 | 1204 | c->packets[0]++; |
e658983a | 1205 | c->bytes[0] += header_size + sizeof(*p) + len; |
b411b363 PR |
1206 | |
1207 | if (c->bit_offset >= c->bm_bits) | |
1208 | len = 0; /* DONE */ | |
1209 | } else { | |
1210 | /* was not compressible. | |
1211 | * send a buffer full of plain text bits instead. */ | |
50d0b1ad AG |
1212 | unsigned int data_size; |
1213 | unsigned long num_words; | |
e658983a | 1214 | unsigned long *p = sock->sbuf + header_size; |
50d0b1ad AG |
1215 | |
1216 | data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; | |
e658983a | 1217 | num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 1218 | c->bm_words - c->word_offset); |
e658983a | 1219 | len = num_words * sizeof(*p); |
b411b363 | 1220 | if (len) |
b30ab791 | 1221 | drbd_bm_get_lel(device, c->word_offset, num_words, p); |
a6b32bc3 | 1222 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0); |
b411b363 PR |
1223 | c->word_offset += num_words; |
1224 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1225 | ||
1226 | c->packets[1]++; | |
50d0b1ad | 1227 | c->bytes[1] += header_size + len; |
b411b363 PR |
1228 | |
1229 | if (c->bit_offset > c->bm_bits) | |
1230 | c->bit_offset = c->bm_bits; | |
1231 | } | |
a982dd57 | 1232 | if (!err) { |
f70af118 | 1233 | if (len == 0) { |
b30ab791 | 1234 | INFO_bm_xfer_stats(device, "send", c); |
f70af118 AG |
1235 | return 0; |
1236 | } else | |
1237 | return 1; | |
1238 | } | |
1239 | return -EIO; | |
b411b363 PR |
1240 | } |
1241 | ||
1242 | /* See the comment at receive_bitmap() */ | |
b30ab791 | 1243 | static int _drbd_send_bitmap(struct drbd_device *device) |
b411b363 PR |
1244 | { |
1245 | struct bm_xfer_ctx c; | |
f70af118 | 1246 | int err; |
b411b363 | 1247 | |
b30ab791 | 1248 | if (!expect(device->bitmap)) |
81e84650 | 1249 | return false; |
b411b363 | 1250 | |
b30ab791 AG |
1251 | if (get_ldev(device)) { |
1252 | if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) { | |
d0180171 | 1253 | drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n"); |
b30ab791 AG |
1254 | drbd_bm_set_all(device); |
1255 | if (drbd_bm_write(device)) { | |
b411b363 PR |
1256 | /* write_bm did fail! Leave full sync flag set in Meta P_DATA |
1257 | * but otherwise process as per normal - need to tell other | |
1258 | * side that a full resync is required! */ | |
d0180171 | 1259 | drbd_err(device, "Failed to write bitmap to disk!\n"); |
b411b363 | 1260 | } else { |
b30ab791 AG |
1261 | drbd_md_clear_flag(device, MDF_FULL_SYNC); |
1262 | drbd_md_sync(device); | |
b411b363 PR |
1263 | } |
1264 | } | |
b30ab791 | 1265 | put_ldev(device); |
b411b363 PR |
1266 | } |
1267 | ||
1268 | c = (struct bm_xfer_ctx) { | |
b30ab791 AG |
1269 | .bm_bits = drbd_bm_bits(device), |
1270 | .bm_words = drbd_bm_words(device), | |
b411b363 PR |
1271 | }; |
1272 | ||
1273 | do { | |
b30ab791 | 1274 | err = send_bitmap_rle_or_plain(device, &c); |
f70af118 | 1275 | } while (err > 0); |
b411b363 | 1276 | |
f70af118 | 1277 | return err == 0; |
b411b363 PR |
1278 | } |
1279 | ||
b30ab791 | 1280 | int drbd_send_bitmap(struct drbd_device *device) |
b411b363 | 1281 | { |
a6b32bc3 | 1282 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
9f5bdc33 | 1283 | int err = -1; |
b411b363 | 1284 | |
9f5bdc33 AG |
1285 | mutex_lock(&sock->mutex); |
1286 | if (sock->socket) | |
b30ab791 | 1287 | err = !_drbd_send_bitmap(device); |
9f5bdc33 | 1288 | mutex_unlock(&sock->mutex); |
b411b363 PR |
1289 | return err; |
1290 | } | |
1291 | ||
bde89a9e | 1292 | void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size) |
b411b363 | 1293 | { |
9f5bdc33 AG |
1294 | struct drbd_socket *sock; |
1295 | struct p_barrier_ack *p; | |
b411b363 | 1296 | |
bde89a9e | 1297 | if (connection->cstate < C_WF_REPORT_PARAMS) |
9f5bdc33 | 1298 | return; |
b411b363 | 1299 | |
bde89a9e AG |
1300 | sock = &connection->meta; |
1301 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1302 | if (!p) |
1303 | return; | |
1304 | p->barrier = barrier_nr; | |
1305 | p->set_size = cpu_to_be32(set_size); | |
bde89a9e | 1306 | conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0); |
b411b363 PR |
1307 | } |
1308 | ||
1309 | /** | |
1310 | * _drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1311 | * @device: DRBD device. |
b411b363 PR |
1312 | * @cmd: Packet command code. |
1313 | * @sector: sector, needs to be in big endian byte order | |
1314 | * @blksize: size in byte, needs to be in big endian byte order | |
1315 | * @block_id: Id, big endian byte order | |
1316 | */ | |
b30ab791 | 1317 | static int _drbd_send_ack(struct drbd_device *device, enum drbd_packet cmd, |
d8763023 | 1318 | u64 sector, u32 blksize, u64 block_id) |
b411b363 | 1319 | { |
9f5bdc33 AG |
1320 | struct drbd_socket *sock; |
1321 | struct p_block_ack *p; | |
b411b363 | 1322 | |
b30ab791 | 1323 | if (device->state.conn < C_CONNECTED) |
9f5bdc33 | 1324 | return -EIO; |
b411b363 | 1325 | |
a6b32bc3 | 1326 | sock = &first_peer_device(device)->connection->meta; |
b30ab791 | 1327 | p = drbd_prepare_command(device, sock); |
9f5bdc33 | 1328 | if (!p) |
a8c32aa8 | 1329 | return -EIO; |
9f5bdc33 AG |
1330 | p->sector = sector; |
1331 | p->block_id = block_id; | |
1332 | p->blksize = blksize; | |
b30ab791 AG |
1333 | p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); |
1334 | return drbd_send_command(device, sock, cmd, sizeof(*p), NULL, 0); | |
b411b363 PR |
1335 | } |
1336 | ||
2b2bf214 LE |
1337 | /* dp->sector and dp->block_id already/still in network byte order, |
1338 | * data_size is payload size according to dp->head, | |
1339 | * and may need to be corrected for digest size. */ | |
b30ab791 | 1340 | void drbd_send_ack_dp(struct drbd_device *device, enum drbd_packet cmd, |
a9a9994d | 1341 | struct p_data *dp, int data_size) |
b411b363 | 1342 | { |
a6b32bc3 AG |
1343 | if (first_peer_device(device)->connection->peer_integrity_tfm) |
1344 | data_size -= crypto_hash_digestsize(first_peer_device(device)->connection->peer_integrity_tfm); | |
b30ab791 | 1345 | _drbd_send_ack(device, cmd, dp->sector, cpu_to_be32(data_size), |
a9a9994d | 1346 | dp->block_id); |
b411b363 PR |
1347 | } |
1348 | ||
b30ab791 | 1349 | void drbd_send_ack_rp(struct drbd_device *device, enum drbd_packet cmd, |
a9a9994d | 1350 | struct p_block_req *rp) |
b411b363 | 1351 | { |
b30ab791 | 1352 | _drbd_send_ack(device, cmd, rp->sector, rp->blksize, rp->block_id); |
b411b363 PR |
1353 | } |
1354 | ||
1355 | /** | |
1356 | * drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1357 | * @device: DRBD device |
db830c46 AG |
1358 | * @cmd: packet command code |
1359 | * @peer_req: peer request | |
b411b363 | 1360 | */ |
b30ab791 | 1361 | int drbd_send_ack(struct drbd_device *device, enum drbd_packet cmd, |
db830c46 | 1362 | struct drbd_peer_request *peer_req) |
b411b363 | 1363 | { |
b30ab791 | 1364 | return _drbd_send_ack(device, cmd, |
dd516121 AG |
1365 | cpu_to_be64(peer_req->i.sector), |
1366 | cpu_to_be32(peer_req->i.size), | |
1367 | peer_req->block_id); | |
b411b363 PR |
1368 | } |
1369 | ||
1370 | /* This function misuses the block_id field to signal if the blocks | |
1371 | * are is sync or not. */ | |
b30ab791 | 1372 | int drbd_send_ack_ex(struct drbd_device *device, enum drbd_packet cmd, |
b411b363 PR |
1373 | sector_t sector, int blksize, u64 block_id) |
1374 | { | |
b30ab791 | 1375 | return _drbd_send_ack(device, cmd, |
b411b363 PR |
1376 | cpu_to_be64(sector), |
1377 | cpu_to_be32(blksize), | |
1378 | cpu_to_be64(block_id)); | |
1379 | } | |
1380 | ||
b30ab791 | 1381 | int drbd_send_drequest(struct drbd_device *device, int cmd, |
b411b363 PR |
1382 | sector_t sector, int size, u64 block_id) |
1383 | { | |
9f5bdc33 AG |
1384 | struct drbd_socket *sock; |
1385 | struct p_block_req *p; | |
b411b363 | 1386 | |
a6b32bc3 | 1387 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1388 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1389 | if (!p) |
1390 | return -EIO; | |
1391 | p->sector = cpu_to_be64(sector); | |
1392 | p->block_id = block_id; | |
1393 | p->blksize = cpu_to_be32(size); | |
b30ab791 | 1394 | return drbd_send_command(device, sock, cmd, sizeof(*p), NULL, 0); |
b411b363 PR |
1395 | } |
1396 | ||
b30ab791 | 1397 | int drbd_send_drequest_csum(struct drbd_device *device, sector_t sector, int size, |
d8763023 | 1398 | void *digest, int digest_size, enum drbd_packet cmd) |
b411b363 | 1399 | { |
9f5bdc33 AG |
1400 | struct drbd_socket *sock; |
1401 | struct p_block_req *p; | |
b411b363 | 1402 | |
9f5bdc33 | 1403 | /* FIXME: Put the digest into the preallocated socket buffer. */ |
b411b363 | 1404 | |
a6b32bc3 | 1405 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1406 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1407 | if (!p) |
1408 | return -EIO; | |
1409 | p->sector = cpu_to_be64(sector); | |
1410 | p->block_id = ID_SYNCER /* unused */; | |
1411 | p->blksize = cpu_to_be32(size); | |
b30ab791 | 1412 | return drbd_send_command(device, sock, cmd, sizeof(*p), |
9f5bdc33 | 1413 | digest, digest_size); |
b411b363 PR |
1414 | } |
1415 | ||
b30ab791 | 1416 | int drbd_send_ov_request(struct drbd_device *device, sector_t sector, int size) |
b411b363 | 1417 | { |
9f5bdc33 AG |
1418 | struct drbd_socket *sock; |
1419 | struct p_block_req *p; | |
b411b363 | 1420 | |
a6b32bc3 | 1421 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1422 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1423 | if (!p) |
1424 | return -EIO; | |
1425 | p->sector = cpu_to_be64(sector); | |
1426 | p->block_id = ID_SYNCER /* unused */; | |
1427 | p->blksize = cpu_to_be32(size); | |
b30ab791 | 1428 | return drbd_send_command(device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0); |
b411b363 PR |
1429 | } |
1430 | ||
1431 | /* called on sndtimeo | |
81e84650 AG |
1432 | * returns false if we should retry, |
1433 | * true if we think connection is dead | |
b411b363 | 1434 | */ |
bde89a9e | 1435 | static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock) |
b411b363 PR |
1436 | { |
1437 | int drop_it; | |
b30ab791 | 1438 | /* long elapsed = (long)(jiffies - device->last_received); */ |
b411b363 | 1439 | |
bde89a9e AG |
1440 | drop_it = connection->meta.socket == sock |
1441 | || !connection->asender.task | |
1442 | || get_t_state(&connection->asender) != RUNNING | |
1443 | || connection->cstate < C_WF_REPORT_PARAMS; | |
b411b363 PR |
1444 | |
1445 | if (drop_it) | |
81e84650 | 1446 | return true; |
b411b363 | 1447 | |
bde89a9e | 1448 | drop_it = !--connection->ko_count; |
b411b363 | 1449 | if (!drop_it) { |
1ec861eb | 1450 | drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n", |
bde89a9e AG |
1451 | current->comm, current->pid, connection->ko_count); |
1452 | request_ping(connection); | |
b411b363 PR |
1453 | } |
1454 | ||
b30ab791 | 1455 | return drop_it; /* && (device->state == R_PRIMARY) */; |
b411b363 PR |
1456 | } |
1457 | ||
bde89a9e | 1458 | static void drbd_update_congested(struct drbd_connection *connection) |
9e204cdd | 1459 | { |
bde89a9e | 1460 | struct sock *sk = connection->data.socket->sk; |
9e204cdd | 1461 | if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) |
bde89a9e | 1462 | set_bit(NET_CONGESTED, &connection->flags); |
9e204cdd AG |
1463 | } |
1464 | ||
b411b363 PR |
1465 | /* The idea of sendpage seems to be to put some kind of reference |
1466 | * to the page into the skb, and to hand it over to the NIC. In | |
1467 | * this process get_page() gets called. | |
1468 | * | |
1469 | * As soon as the page was really sent over the network put_page() | |
1470 | * gets called by some part of the network layer. [ NIC driver? ] | |
1471 | * | |
1472 | * [ get_page() / put_page() increment/decrement the count. If count | |
1473 | * reaches 0 the page will be freed. ] | |
1474 | * | |
1475 | * This works nicely with pages from FSs. | |
1476 | * But this means that in protocol A we might signal IO completion too early! | |
1477 | * | |
1478 | * In order not to corrupt data during a resync we must make sure | |
1479 | * that we do not reuse our own buffer pages (EEs) to early, therefore | |
1480 | * we have the net_ee list. | |
1481 | * | |
1482 | * XFS seems to have problems, still, it submits pages with page_count == 0! | |
1483 | * As a workaround, we disable sendpage on pages | |
1484 | * with page_count == 0 or PageSlab. | |
1485 | */ | |
b30ab791 | 1486 | static int _drbd_no_send_page(struct drbd_device *device, struct page *page, |
b987427b | 1487 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1488 | { |
b987427b AG |
1489 | struct socket *socket; |
1490 | void *addr; | |
1491 | int err; | |
1492 | ||
a6b32bc3 | 1493 | socket = first_peer_device(device)->connection->data.socket; |
b987427b | 1494 | addr = kmap(page) + offset; |
a6b32bc3 | 1495 | err = drbd_send_all(first_peer_device(device)->connection, socket, addr, size, msg_flags); |
b411b363 | 1496 | kunmap(page); |
b987427b | 1497 | if (!err) |
b30ab791 | 1498 | device->send_cnt += size >> 9; |
b987427b | 1499 | return err; |
b411b363 PR |
1500 | } |
1501 | ||
b30ab791 | 1502 | static int _drbd_send_page(struct drbd_device *device, struct page *page, |
ba11ad9a | 1503 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1504 | { |
a6b32bc3 | 1505 | struct socket *socket = first_peer_device(device)->connection->data.socket; |
b411b363 | 1506 | mm_segment_t oldfs = get_fs(); |
b411b363 | 1507 | int len = size; |
88b390ff | 1508 | int err = -EIO; |
b411b363 PR |
1509 | |
1510 | /* e.g. XFS meta- & log-data is in slab pages, which have a | |
1511 | * page_count of 0 and/or have PageSlab() set. | |
1512 | * we cannot use send_page for those, as that does get_page(); | |
1513 | * put_page(); and would cause either a VM_BUG directly, or | |
1514 | * __page_cache_release a page that would actually still be referenced | |
1515 | * by someone, leading to some obscure delayed Oops somewhere else. */ | |
1516 | if (disable_sendpage || (page_count(page) < 1) || PageSlab(page)) | |
b30ab791 | 1517 | return _drbd_no_send_page(device, page, offset, size, msg_flags); |
b411b363 | 1518 | |
ba11ad9a | 1519 | msg_flags |= MSG_NOSIGNAL; |
a6b32bc3 | 1520 | drbd_update_congested(first_peer_device(device)->connection); |
b411b363 PR |
1521 | set_fs(KERNEL_DS); |
1522 | do { | |
88b390ff AG |
1523 | int sent; |
1524 | ||
1525 | sent = socket->ops->sendpage(socket, page, offset, len, msg_flags); | |
b411b363 | 1526 | if (sent <= 0) { |
88b390ff | 1527 | if (sent == -EAGAIN) { |
a6b32bc3 | 1528 | if (we_should_drop_the_connection(first_peer_device(device)->connection, socket)) |
88b390ff AG |
1529 | break; |
1530 | continue; | |
1531 | } | |
d0180171 | 1532 | drbd_warn(device, "%s: size=%d len=%d sent=%d\n", |
b411b363 | 1533 | __func__, (int)size, len, sent); |
88b390ff AG |
1534 | if (sent < 0) |
1535 | err = sent; | |
b411b363 PR |
1536 | break; |
1537 | } | |
1538 | len -= sent; | |
1539 | offset += sent; | |
b30ab791 | 1540 | } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/); |
b411b363 | 1541 | set_fs(oldfs); |
a6b32bc3 | 1542 | clear_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags); |
b411b363 | 1543 | |
88b390ff AG |
1544 | if (len == 0) { |
1545 | err = 0; | |
b30ab791 | 1546 | device->send_cnt += size >> 9; |
88b390ff AG |
1547 | } |
1548 | return err; | |
b411b363 PR |
1549 | } |
1550 | ||
b30ab791 | 1551 | static int _drbd_send_bio(struct drbd_device *device, struct bio *bio) |
b411b363 | 1552 | { |
7988613b KO |
1553 | struct bio_vec bvec; |
1554 | struct bvec_iter iter; | |
1555 | ||
ba11ad9a | 1556 | /* hint all but last page with MSG_MORE */ |
7988613b | 1557 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1558 | int err; |
1559 | ||
b30ab791 | 1560 | err = _drbd_no_send_page(device, bvec.bv_page, |
7988613b | 1561 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1562 | bio_iter_last(bvec, iter) |
7988613b | 1563 | ? 0 : MSG_MORE); |
7fae55da AG |
1564 | if (err) |
1565 | return err; | |
b411b363 | 1566 | } |
7fae55da | 1567 | return 0; |
b411b363 PR |
1568 | } |
1569 | ||
b30ab791 | 1570 | static int _drbd_send_zc_bio(struct drbd_device *device, struct bio *bio) |
b411b363 | 1571 | { |
7988613b KO |
1572 | struct bio_vec bvec; |
1573 | struct bvec_iter iter; | |
1574 | ||
ba11ad9a | 1575 | /* hint all but last page with MSG_MORE */ |
7988613b | 1576 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1577 | int err; |
1578 | ||
b30ab791 | 1579 | err = _drbd_send_page(device, bvec.bv_page, |
7988613b | 1580 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1581 | bio_iter_last(bvec, iter) ? 0 : MSG_MORE); |
7fae55da AG |
1582 | if (err) |
1583 | return err; | |
b411b363 | 1584 | } |
7fae55da | 1585 | return 0; |
b411b363 PR |
1586 | } |
1587 | ||
b30ab791 | 1588 | static int _drbd_send_zc_ee(struct drbd_device *device, |
db830c46 | 1589 | struct drbd_peer_request *peer_req) |
45bb912b | 1590 | { |
db830c46 AG |
1591 | struct page *page = peer_req->pages; |
1592 | unsigned len = peer_req->i.size; | |
9f69230c | 1593 | int err; |
db830c46 | 1594 | |
ba11ad9a | 1595 | /* hint all but last page with MSG_MORE */ |
45bb912b LE |
1596 | page_chain_for_each(page) { |
1597 | unsigned l = min_t(unsigned, len, PAGE_SIZE); | |
9f69230c | 1598 | |
b30ab791 | 1599 | err = _drbd_send_page(device, page, 0, l, |
9f69230c AG |
1600 | page_chain_next(page) ? MSG_MORE : 0); |
1601 | if (err) | |
1602 | return err; | |
45bb912b LE |
1603 | len -= l; |
1604 | } | |
9f69230c | 1605 | return 0; |
45bb912b LE |
1606 | } |
1607 | ||
b30ab791 | 1608 | static u32 bio_flags_to_wire(struct drbd_device *device, unsigned long bi_rw) |
76d2e7ec | 1609 | { |
a6b32bc3 | 1610 | if (first_peer_device(device)->connection->agreed_pro_version >= 95) |
76d2e7ec | 1611 | return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) | |
76d2e7ec PR |
1612 | (bi_rw & REQ_FUA ? DP_FUA : 0) | |
1613 | (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) | | |
1614 | (bi_rw & REQ_DISCARD ? DP_DISCARD : 0); | |
1615 | else | |
721a9602 | 1616 | return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0; |
76d2e7ec PR |
1617 | } |
1618 | ||
b411b363 PR |
1619 | /* Used to send write requests |
1620 | * R_PRIMARY -> Peer (P_DATA) | |
1621 | */ | |
b30ab791 | 1622 | int drbd_send_dblock(struct drbd_device *device, struct drbd_request *req) |
b411b363 | 1623 | { |
9f5bdc33 AG |
1624 | struct drbd_socket *sock; |
1625 | struct p_data *p; | |
b411b363 | 1626 | unsigned int dp_flags = 0; |
b411b363 | 1627 | int dgs; |
9f5bdc33 | 1628 | int err; |
b411b363 | 1629 | |
a6b32bc3 | 1630 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1631 | p = drbd_prepare_command(device, sock); |
a6b32bc3 AG |
1632 | dgs = first_peer_device(device)->connection->integrity_tfm ? |
1633 | crypto_hash_digestsize(first_peer_device(device)->connection->integrity_tfm) : 0; | |
b411b363 | 1634 | |
9f5bdc33 AG |
1635 | if (!p) |
1636 | return -EIO; | |
1637 | p->sector = cpu_to_be64(req->i.sector); | |
1638 | p->block_id = (unsigned long)req; | |
b30ab791 AG |
1639 | p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); |
1640 | dp_flags = bio_flags_to_wire(device, req->master_bio->bi_rw); | |
1641 | if (device->state.conn >= C_SYNC_SOURCE && | |
1642 | device->state.conn <= C_PAUSED_SYNC_T) | |
b411b363 | 1643 | dp_flags |= DP_MAY_SET_IN_SYNC; |
a6b32bc3 | 1644 | if (first_peer_device(device)->connection->agreed_pro_version >= 100) { |
303d1448 PR |
1645 | if (req->rq_state & RQ_EXP_RECEIVE_ACK) |
1646 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
1647 | if (req->rq_state & RQ_EXP_WRITE_ACK) | |
1648 | dp_flags |= DP_SEND_WRITE_ACK; | |
1649 | } | |
9f5bdc33 AG |
1650 | p->dp_flags = cpu_to_be32(dp_flags); |
1651 | if (dgs) | |
a6b32bc3 AG |
1652 | drbd_csum_bio(device, first_peer_device(device)->connection->integrity_tfm, req->master_bio, p + 1); |
1653 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size); | |
6bdb9b0e | 1654 | if (!err) { |
470be44a LE |
1655 | /* For protocol A, we have to memcpy the payload into |
1656 | * socket buffers, as we may complete right away | |
1657 | * as soon as we handed it over to tcp, at which point the data | |
1658 | * pages may become invalid. | |
1659 | * | |
1660 | * For data-integrity enabled, we copy it as well, so we can be | |
1661 | * sure that even if the bio pages may still be modified, it | |
1662 | * won't change the data on the wire, thus if the digest checks | |
1663 | * out ok after sending on this side, but does not fit on the | |
1664 | * receiving side, we sure have detected corruption elsewhere. | |
1665 | */ | |
303d1448 | 1666 | if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs) |
b30ab791 | 1667 | err = _drbd_send_bio(device, req->master_bio); |
b411b363 | 1668 | else |
b30ab791 | 1669 | err = _drbd_send_zc_bio(device, req->master_bio); |
470be44a LE |
1670 | |
1671 | /* double check digest, sometimes buffers have been modified in flight. */ | |
1672 | if (dgs > 0 && dgs <= 64) { | |
24c4830c | 1673 | /* 64 byte, 512 bit, is the largest digest size |
470be44a LE |
1674 | * currently supported in kernel crypto. */ |
1675 | unsigned char digest[64]; | |
a6b32bc3 | 1676 | drbd_csum_bio(device, first_peer_device(device)->connection->integrity_tfm, req->master_bio, digest); |
9f5bdc33 | 1677 | if (memcmp(p + 1, digest, dgs)) { |
d0180171 | 1678 | drbd_warn(device, |
470be44a | 1679 | "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n", |
ace652ac | 1680 | (unsigned long long)req->i.sector, req->i.size); |
470be44a LE |
1681 | } |
1682 | } /* else if (dgs > 64) { | |
1683 | ... Be noisy about digest too large ... | |
1684 | } */ | |
b411b363 | 1685 | } |
9f5bdc33 | 1686 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
b411b363 | 1687 | |
6bdb9b0e | 1688 | return err; |
b411b363 PR |
1689 | } |
1690 | ||
1691 | /* answer packet, used to send data back for read requests: | |
1692 | * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY) | |
1693 | * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY) | |
1694 | */ | |
b30ab791 | 1695 | int drbd_send_block(struct drbd_device *device, enum drbd_packet cmd, |
db830c46 | 1696 | struct drbd_peer_request *peer_req) |
b411b363 | 1697 | { |
9f5bdc33 AG |
1698 | struct drbd_socket *sock; |
1699 | struct p_data *p; | |
7b57b89d | 1700 | int err; |
b411b363 PR |
1701 | int dgs; |
1702 | ||
a6b32bc3 | 1703 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1704 | p = drbd_prepare_command(device, sock); |
b411b363 | 1705 | |
a6b32bc3 AG |
1706 | dgs = first_peer_device(device)->connection->integrity_tfm ? |
1707 | crypto_hash_digestsize(first_peer_device(device)->connection->integrity_tfm) : 0; | |
b411b363 | 1708 | |
9f5bdc33 AG |
1709 | if (!p) |
1710 | return -EIO; | |
1711 | p->sector = cpu_to_be64(peer_req->i.sector); | |
1712 | p->block_id = peer_req->block_id; | |
1713 | p->seq_num = 0; /* unused */ | |
b17f33cb | 1714 | p->dp_flags = 0; |
9f5bdc33 | 1715 | if (dgs) |
a6b32bc3 AG |
1716 | drbd_csum_ee(device, first_peer_device(device)->connection->integrity_tfm, peer_req, p + 1); |
1717 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size); | |
7b57b89d | 1718 | if (!err) |
b30ab791 | 1719 | err = _drbd_send_zc_ee(device, peer_req); |
9f5bdc33 | 1720 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
bd26bfc5 | 1721 | |
7b57b89d | 1722 | return err; |
b411b363 PR |
1723 | } |
1724 | ||
b30ab791 | 1725 | int drbd_send_out_of_sync(struct drbd_device *device, struct drbd_request *req) |
73a01a18 | 1726 | { |
9f5bdc33 AG |
1727 | struct drbd_socket *sock; |
1728 | struct p_block_desc *p; | |
73a01a18 | 1729 | |
a6b32bc3 | 1730 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1731 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1732 | if (!p) |
1733 | return -EIO; | |
1734 | p->sector = cpu_to_be64(req->i.sector); | |
1735 | p->blksize = cpu_to_be32(req->i.size); | |
b30ab791 | 1736 | return drbd_send_command(device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0); |
73a01a18 PR |
1737 | } |
1738 | ||
b411b363 PR |
1739 | /* |
1740 | drbd_send distinguishes two cases: | |
1741 | ||
1742 | Packets sent via the data socket "sock" | |
1743 | and packets sent via the meta data socket "msock" | |
1744 | ||
1745 | sock msock | |
1746 | -----------------+-------------------------+------------------------------ | |
1747 | timeout conf.timeout / 2 conf.timeout / 2 | |
1748 | timeout action send a ping via msock Abort communication | |
1749 | and close all sockets | |
1750 | */ | |
1751 | ||
1752 | /* | |
1753 | * you must have down()ed the appropriate [m]sock_mutex elsewhere! | |
1754 | */ | |
bde89a9e | 1755 | int drbd_send(struct drbd_connection *connection, struct socket *sock, |
b411b363 PR |
1756 | void *buf, size_t size, unsigned msg_flags) |
1757 | { | |
1758 | struct kvec iov; | |
1759 | struct msghdr msg; | |
1760 | int rv, sent = 0; | |
1761 | ||
1762 | if (!sock) | |
c0d42c8e | 1763 | return -EBADR; |
b411b363 PR |
1764 | |
1765 | /* THINK if (signal_pending) return ... ? */ | |
1766 | ||
1767 | iov.iov_base = buf; | |
1768 | iov.iov_len = size; | |
1769 | ||
1770 | msg.msg_name = NULL; | |
1771 | msg.msg_namelen = 0; | |
1772 | msg.msg_control = NULL; | |
1773 | msg.msg_controllen = 0; | |
1774 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; | |
1775 | ||
bde89a9e | 1776 | if (sock == connection->data.socket) { |
44ed167d | 1777 | rcu_read_lock(); |
bde89a9e | 1778 | connection->ko_count = rcu_dereference(connection->net_conf)->ko_count; |
44ed167d | 1779 | rcu_read_unlock(); |
bde89a9e | 1780 | drbd_update_congested(connection); |
b411b363 PR |
1781 | } |
1782 | do { | |
1783 | /* STRANGE | |
1784 | * tcp_sendmsg does _not_ use its size parameter at all ? | |
1785 | * | |
1786 | * -EAGAIN on timeout, -EINTR on signal. | |
1787 | */ | |
1788 | /* THINK | |
1789 | * do we need to block DRBD_SIG if sock == &meta.socket ?? | |
1790 | * otherwise wake_asender() might interrupt some send_*Ack ! | |
1791 | */ | |
1792 | rv = kernel_sendmsg(sock, &msg, &iov, 1, size); | |
1793 | if (rv == -EAGAIN) { | |
bde89a9e | 1794 | if (we_should_drop_the_connection(connection, sock)) |
b411b363 PR |
1795 | break; |
1796 | else | |
1797 | continue; | |
1798 | } | |
b411b363 PR |
1799 | if (rv == -EINTR) { |
1800 | flush_signals(current); | |
1801 | rv = 0; | |
1802 | } | |
1803 | if (rv < 0) | |
1804 | break; | |
1805 | sent += rv; | |
1806 | iov.iov_base += rv; | |
1807 | iov.iov_len -= rv; | |
1808 | } while (sent < size); | |
1809 | ||
bde89a9e AG |
1810 | if (sock == connection->data.socket) |
1811 | clear_bit(NET_CONGESTED, &connection->flags); | |
b411b363 PR |
1812 | |
1813 | if (rv <= 0) { | |
1814 | if (rv != -EAGAIN) { | |
1ec861eb | 1815 | drbd_err(connection, "%s_sendmsg returned %d\n", |
bde89a9e | 1816 | sock == connection->meta.socket ? "msock" : "sock", |
bedbd2a5 | 1817 | rv); |
bde89a9e | 1818 | conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 1819 | } else |
bde89a9e | 1820 | conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); |
b411b363 PR |
1821 | } |
1822 | ||
1823 | return sent; | |
1824 | } | |
1825 | ||
fb708e40 AG |
1826 | /** |
1827 | * drbd_send_all - Send an entire buffer | |
1828 | * | |
1829 | * Returns 0 upon success and a negative error value otherwise. | |
1830 | */ | |
bde89a9e | 1831 | int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer, |
fb708e40 AG |
1832 | size_t size, unsigned msg_flags) |
1833 | { | |
1834 | int err; | |
1835 | ||
bde89a9e | 1836 | err = drbd_send(connection, sock, buffer, size, msg_flags); |
fb708e40 AG |
1837 | if (err < 0) |
1838 | return err; | |
1839 | if (err != size) | |
1840 | return -EIO; | |
1841 | return 0; | |
1842 | } | |
1843 | ||
b411b363 PR |
1844 | static int drbd_open(struct block_device *bdev, fmode_t mode) |
1845 | { | |
b30ab791 | 1846 | struct drbd_device *device = bdev->bd_disk->private_data; |
b411b363 PR |
1847 | unsigned long flags; |
1848 | int rv = 0; | |
1849 | ||
2a48fc0a | 1850 | mutex_lock(&drbd_main_mutex); |
0500813f | 1851 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 1852 | /* to have a stable device->state.role |
b411b363 PR |
1853 | * and no race with updating open_cnt */ |
1854 | ||
b30ab791 | 1855 | if (device->state.role != R_PRIMARY) { |
b411b363 PR |
1856 | if (mode & FMODE_WRITE) |
1857 | rv = -EROFS; | |
1858 | else if (!allow_oos) | |
1859 | rv = -EMEDIUMTYPE; | |
1860 | } | |
1861 | ||
1862 | if (!rv) | |
b30ab791 | 1863 | device->open_cnt++; |
0500813f | 1864 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
2a48fc0a | 1865 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1866 | |
1867 | return rv; | |
1868 | } | |
1869 | ||
db2a144b | 1870 | static void drbd_release(struct gendisk *gd, fmode_t mode) |
b411b363 | 1871 | { |
b30ab791 | 1872 | struct drbd_device *device = gd->private_data; |
2a48fc0a | 1873 | mutex_lock(&drbd_main_mutex); |
b30ab791 | 1874 | device->open_cnt--; |
2a48fc0a | 1875 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1876 | } |
1877 | ||
b30ab791 | 1878 | static void drbd_set_defaults(struct drbd_device *device) |
b411b363 | 1879 | { |
f399002e LE |
1880 | /* Beware! The actual layout differs |
1881 | * between big endian and little endian */ | |
b30ab791 | 1882 | device->state = (union drbd_dev_state) { |
b411b363 PR |
1883 | { .role = R_SECONDARY, |
1884 | .peer = R_UNKNOWN, | |
1885 | .conn = C_STANDALONE, | |
1886 | .disk = D_DISKLESS, | |
1887 | .pdsk = D_UNKNOWN, | |
b411b363 PR |
1888 | } }; |
1889 | } | |
1890 | ||
b30ab791 | 1891 | void drbd_init_set_defaults(struct drbd_device *device) |
b411b363 PR |
1892 | { |
1893 | /* the memset(,0,) did most of this. | |
1894 | * note: only assignments, no allocation in here */ | |
1895 | ||
b30ab791 AG |
1896 | drbd_set_defaults(device); |
1897 | ||
1898 | atomic_set(&device->ap_bio_cnt, 0); | |
1899 | atomic_set(&device->ap_pending_cnt, 0); | |
1900 | atomic_set(&device->rs_pending_cnt, 0); | |
1901 | atomic_set(&device->unacked_cnt, 0); | |
1902 | atomic_set(&device->local_cnt, 0); | |
1903 | atomic_set(&device->pp_in_use_by_net, 0); | |
1904 | atomic_set(&device->rs_sect_in, 0); | |
1905 | atomic_set(&device->rs_sect_ev, 0); | |
1906 | atomic_set(&device->ap_in_flight, 0); | |
1907 | atomic_set(&device->md_io_in_use, 0); | |
1908 | ||
1909 | mutex_init(&device->own_state_mutex); | |
1910 | device->state_mutex = &device->own_state_mutex; | |
1911 | ||
1912 | spin_lock_init(&device->al_lock); | |
1913 | spin_lock_init(&device->peer_seq_lock); | |
1914 | ||
1915 | INIT_LIST_HEAD(&device->active_ee); | |
1916 | INIT_LIST_HEAD(&device->sync_ee); | |
1917 | INIT_LIST_HEAD(&device->done_ee); | |
1918 | INIT_LIST_HEAD(&device->read_ee); | |
1919 | INIT_LIST_HEAD(&device->net_ee); | |
1920 | INIT_LIST_HEAD(&device->resync_reads); | |
1921 | INIT_LIST_HEAD(&device->resync_work.list); | |
1922 | INIT_LIST_HEAD(&device->unplug_work.list); | |
1923 | INIT_LIST_HEAD(&device->go_diskless.list); | |
1924 | INIT_LIST_HEAD(&device->md_sync_work.list); | |
1925 | INIT_LIST_HEAD(&device->start_resync_work.list); | |
1926 | INIT_LIST_HEAD(&device->bm_io_work.w.list); | |
1927 | ||
1928 | device->resync_work.cb = w_resync_timer; | |
1929 | device->unplug_work.cb = w_send_write_hint; | |
1930 | device->go_diskless.cb = w_go_diskless; | |
1931 | device->md_sync_work.cb = w_md_sync; | |
1932 | device->bm_io_work.w.cb = w_bitmap_io; | |
1933 | device->start_resync_work.cb = w_start_resync; | |
1934 | ||
1935 | device->resync_work.device = device; | |
1936 | device->unplug_work.device = device; | |
1937 | device->go_diskless.device = device; | |
1938 | device->md_sync_work.device = device; | |
1939 | device->bm_io_work.w.device = device; | |
1940 | device->start_resync_work.device = device; | |
1941 | ||
1942 | init_timer(&device->resync_timer); | |
1943 | init_timer(&device->md_sync_timer); | |
1944 | init_timer(&device->start_resync_timer); | |
1945 | init_timer(&device->request_timer); | |
1946 | device->resync_timer.function = resync_timer_fn; | |
1947 | device->resync_timer.data = (unsigned long) device; | |
1948 | device->md_sync_timer.function = md_sync_timer_fn; | |
1949 | device->md_sync_timer.data = (unsigned long) device; | |
1950 | device->start_resync_timer.function = start_resync_timer_fn; | |
1951 | device->start_resync_timer.data = (unsigned long) device; | |
1952 | device->request_timer.function = request_timer_fn; | |
1953 | device->request_timer.data = (unsigned long) device; | |
1954 | ||
1955 | init_waitqueue_head(&device->misc_wait); | |
1956 | init_waitqueue_head(&device->state_wait); | |
1957 | init_waitqueue_head(&device->ee_wait); | |
1958 | init_waitqueue_head(&device->al_wait); | |
1959 | init_waitqueue_head(&device->seq_wait); | |
1960 | ||
1961 | device->resync_wenr = LC_FREE; | |
1962 | device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1963 | device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1964 | } | |
1965 | ||
1966 | void drbd_device_cleanup(struct drbd_device *device) | |
b411b363 | 1967 | { |
1d7734a0 | 1968 | int i; |
a6b32bc3 | 1969 | if (first_peer_device(device)->connection->receiver.t_state != NONE) |
d0180171 | 1970 | drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n", |
a6b32bc3 | 1971 | first_peer_device(device)->connection->receiver.t_state); |
b30ab791 AG |
1972 | |
1973 | device->al_writ_cnt = | |
1974 | device->bm_writ_cnt = | |
1975 | device->read_cnt = | |
1976 | device->recv_cnt = | |
1977 | device->send_cnt = | |
1978 | device->writ_cnt = | |
1979 | device->p_size = | |
1980 | device->rs_start = | |
1981 | device->rs_total = | |
1982 | device->rs_failed = 0; | |
1983 | device->rs_last_events = 0; | |
1984 | device->rs_last_sect_ev = 0; | |
1d7734a0 | 1985 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1986 | device->rs_mark_left[i] = 0; |
1987 | device->rs_mark_time[i] = 0; | |
1d7734a0 | 1988 | } |
0b0ba1ef | 1989 | D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL); |
b411b363 | 1990 | |
b30ab791 AG |
1991 | drbd_set_my_capacity(device, 0); |
1992 | if (device->bitmap) { | |
b411b363 | 1993 | /* maybe never allocated. */ |
b30ab791 AG |
1994 | drbd_bm_resize(device, 0, 1); |
1995 | drbd_bm_cleanup(device); | |
b411b363 PR |
1996 | } |
1997 | ||
b30ab791 AG |
1998 | drbd_free_bc(device->ldev); |
1999 | device->ldev = NULL; | |
1d041225 | 2000 | |
b30ab791 | 2001 | clear_bit(AL_SUSPENDED, &device->flags); |
b411b363 | 2002 | |
0b0ba1ef AG |
2003 | D_ASSERT(device, list_empty(&device->active_ee)); |
2004 | D_ASSERT(device, list_empty(&device->sync_ee)); | |
2005 | D_ASSERT(device, list_empty(&device->done_ee)); | |
2006 | D_ASSERT(device, list_empty(&device->read_ee)); | |
2007 | D_ASSERT(device, list_empty(&device->net_ee)); | |
2008 | D_ASSERT(device, list_empty(&device->resync_reads)); | |
2009 | D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q)); | |
2010 | D_ASSERT(device, list_empty(&device->resync_work.list)); | |
2011 | D_ASSERT(device, list_empty(&device->unplug_work.list)); | |
2012 | D_ASSERT(device, list_empty(&device->go_diskless.list)); | |
2265b473 | 2013 | |
b30ab791 | 2014 | drbd_set_defaults(device); |
b411b363 PR |
2015 | } |
2016 | ||
2017 | ||
2018 | static void drbd_destroy_mempools(void) | |
2019 | { | |
2020 | struct page *page; | |
2021 | ||
2022 | while (drbd_pp_pool) { | |
2023 | page = drbd_pp_pool; | |
2024 | drbd_pp_pool = (struct page *)page_private(page); | |
2025 | __free_page(page); | |
2026 | drbd_pp_vacant--; | |
2027 | } | |
2028 | ||
0b0ba1ef | 2029 | /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */ |
b411b363 | 2030 | |
9476f39d LE |
2031 | if (drbd_md_io_bio_set) |
2032 | bioset_free(drbd_md_io_bio_set); | |
4281808f LE |
2033 | if (drbd_md_io_page_pool) |
2034 | mempool_destroy(drbd_md_io_page_pool); | |
b411b363 PR |
2035 | if (drbd_ee_mempool) |
2036 | mempool_destroy(drbd_ee_mempool); | |
2037 | if (drbd_request_mempool) | |
2038 | mempool_destroy(drbd_request_mempool); | |
2039 | if (drbd_ee_cache) | |
2040 | kmem_cache_destroy(drbd_ee_cache); | |
2041 | if (drbd_request_cache) | |
2042 | kmem_cache_destroy(drbd_request_cache); | |
2043 | if (drbd_bm_ext_cache) | |
2044 | kmem_cache_destroy(drbd_bm_ext_cache); | |
2045 | if (drbd_al_ext_cache) | |
2046 | kmem_cache_destroy(drbd_al_ext_cache); | |
2047 | ||
9476f39d | 2048 | drbd_md_io_bio_set = NULL; |
4281808f | 2049 | drbd_md_io_page_pool = NULL; |
b411b363 PR |
2050 | drbd_ee_mempool = NULL; |
2051 | drbd_request_mempool = NULL; | |
2052 | drbd_ee_cache = NULL; | |
2053 | drbd_request_cache = NULL; | |
2054 | drbd_bm_ext_cache = NULL; | |
2055 | drbd_al_ext_cache = NULL; | |
2056 | ||
2057 | return; | |
2058 | } | |
2059 | ||
2060 | static int drbd_create_mempools(void) | |
2061 | { | |
2062 | struct page *page; | |
1816a2b4 | 2063 | const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count; |
b411b363 PR |
2064 | int i; |
2065 | ||
2066 | /* prepare our caches and mempools */ | |
2067 | drbd_request_mempool = NULL; | |
2068 | drbd_ee_cache = NULL; | |
2069 | drbd_request_cache = NULL; | |
2070 | drbd_bm_ext_cache = NULL; | |
2071 | drbd_al_ext_cache = NULL; | |
2072 | drbd_pp_pool = NULL; | |
4281808f | 2073 | drbd_md_io_page_pool = NULL; |
9476f39d | 2074 | drbd_md_io_bio_set = NULL; |
b411b363 PR |
2075 | |
2076 | /* caches */ | |
2077 | drbd_request_cache = kmem_cache_create( | |
2078 | "drbd_req", sizeof(struct drbd_request), 0, 0, NULL); | |
2079 | if (drbd_request_cache == NULL) | |
2080 | goto Enomem; | |
2081 | ||
2082 | drbd_ee_cache = kmem_cache_create( | |
f6ffca9f | 2083 | "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL); |
b411b363 PR |
2084 | if (drbd_ee_cache == NULL) |
2085 | goto Enomem; | |
2086 | ||
2087 | drbd_bm_ext_cache = kmem_cache_create( | |
2088 | "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL); | |
2089 | if (drbd_bm_ext_cache == NULL) | |
2090 | goto Enomem; | |
2091 | ||
2092 | drbd_al_ext_cache = kmem_cache_create( | |
2093 | "drbd_al", sizeof(struct lc_element), 0, 0, NULL); | |
2094 | if (drbd_al_ext_cache == NULL) | |
2095 | goto Enomem; | |
2096 | ||
2097 | /* mempools */ | |
9476f39d LE |
2098 | drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0); |
2099 | if (drbd_md_io_bio_set == NULL) | |
2100 | goto Enomem; | |
9476f39d | 2101 | |
4281808f LE |
2102 | drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0); |
2103 | if (drbd_md_io_page_pool == NULL) | |
2104 | goto Enomem; | |
2105 | ||
b411b363 PR |
2106 | drbd_request_mempool = mempool_create(number, |
2107 | mempool_alloc_slab, mempool_free_slab, drbd_request_cache); | |
2108 | if (drbd_request_mempool == NULL) | |
2109 | goto Enomem; | |
2110 | ||
2111 | drbd_ee_mempool = mempool_create(number, | |
2112 | mempool_alloc_slab, mempool_free_slab, drbd_ee_cache); | |
2027ae1f | 2113 | if (drbd_ee_mempool == NULL) |
b411b363 PR |
2114 | goto Enomem; |
2115 | ||
2116 | /* drbd's page pool */ | |
2117 | spin_lock_init(&drbd_pp_lock); | |
2118 | ||
2119 | for (i = 0; i < number; i++) { | |
2120 | page = alloc_page(GFP_HIGHUSER); | |
2121 | if (!page) | |
2122 | goto Enomem; | |
2123 | set_page_private(page, (unsigned long)drbd_pp_pool); | |
2124 | drbd_pp_pool = page; | |
2125 | } | |
2126 | drbd_pp_vacant = number; | |
2127 | ||
2128 | return 0; | |
2129 | ||
2130 | Enomem: | |
2131 | drbd_destroy_mempools(); /* in case we allocated some */ | |
2132 | return -ENOMEM; | |
2133 | } | |
2134 | ||
2135 | static int drbd_notify_sys(struct notifier_block *this, unsigned long code, | |
2136 | void *unused) | |
2137 | { | |
2138 | /* just so we have it. you never know what interesting things we | |
2139 | * might want to do here some day... | |
2140 | */ | |
2141 | ||
2142 | return NOTIFY_DONE; | |
2143 | } | |
2144 | ||
2145 | static struct notifier_block drbd_notifier = { | |
2146 | .notifier_call = drbd_notify_sys, | |
2147 | }; | |
2148 | ||
b30ab791 | 2149 | static void drbd_release_all_peer_reqs(struct drbd_device *device) |
b411b363 PR |
2150 | { |
2151 | int rr; | |
2152 | ||
b30ab791 | 2153 | rr = drbd_free_peer_reqs(device, &device->active_ee); |
b411b363 | 2154 | if (rr) |
d0180171 | 2155 | drbd_err(device, "%d EEs in active list found!\n", rr); |
b411b363 | 2156 | |
b30ab791 | 2157 | rr = drbd_free_peer_reqs(device, &device->sync_ee); |
b411b363 | 2158 | if (rr) |
d0180171 | 2159 | drbd_err(device, "%d EEs in sync list found!\n", rr); |
b411b363 | 2160 | |
b30ab791 | 2161 | rr = drbd_free_peer_reqs(device, &device->read_ee); |
b411b363 | 2162 | if (rr) |
d0180171 | 2163 | drbd_err(device, "%d EEs in read list found!\n", rr); |
b411b363 | 2164 | |
b30ab791 | 2165 | rr = drbd_free_peer_reqs(device, &device->done_ee); |
b411b363 | 2166 | if (rr) |
d0180171 | 2167 | drbd_err(device, "%d EEs in done list found!\n", rr); |
b411b363 | 2168 | |
b30ab791 | 2169 | rr = drbd_free_peer_reqs(device, &device->net_ee); |
b411b363 | 2170 | if (rr) |
d0180171 | 2171 | drbd_err(device, "%d EEs in net list found!\n", rr); |
b411b363 PR |
2172 | } |
2173 | ||
774b3055 | 2174 | /* caution. no locking. */ |
05a10ec7 | 2175 | void drbd_destroy_device(struct kref *kref) |
b411b363 | 2176 | { |
b30ab791 | 2177 | struct drbd_device *device = container_of(kref, struct drbd_device, kref); |
803ea134 AG |
2178 | struct drbd_resource *resource = device->resource; |
2179 | struct drbd_connection *connection; | |
b411b363 | 2180 | |
b30ab791 | 2181 | del_timer_sync(&device->request_timer); |
dfa8bedb | 2182 | |
b411b363 | 2183 | /* paranoia asserts */ |
0b0ba1ef | 2184 | D_ASSERT(device, device->open_cnt == 0); |
b411b363 PR |
2185 | /* end paranoia asserts */ |
2186 | ||
b411b363 PR |
2187 | /* cleanup stuff that may have been allocated during |
2188 | * device (re-)configuration or state changes */ | |
2189 | ||
b30ab791 AG |
2190 | if (device->this_bdev) |
2191 | bdput(device->this_bdev); | |
b411b363 | 2192 | |
b30ab791 AG |
2193 | drbd_free_bc(device->ldev); |
2194 | device->ldev = NULL; | |
b411b363 | 2195 | |
b30ab791 | 2196 | drbd_release_all_peer_reqs(device); |
b411b363 | 2197 | |
b30ab791 AG |
2198 | lc_destroy(device->act_log); |
2199 | lc_destroy(device->resync); | |
b411b363 | 2200 | |
b30ab791 AG |
2201 | kfree(device->p_uuid); |
2202 | /* device->p_uuid = NULL; */ | |
b411b363 | 2203 | |
b30ab791 AG |
2204 | if (device->bitmap) /* should no longer be there. */ |
2205 | drbd_bm_cleanup(device); | |
2206 | __free_page(device->md_io_page); | |
2207 | put_disk(device->vdisk); | |
2208 | blk_cleanup_queue(device->rq_queue); | |
2209 | kfree(device->rs_plan_s); | |
a6b32bc3 | 2210 | kfree(first_peer_device(device)); |
b30ab791 | 2211 | kfree(device); |
9dc9fbb3 | 2212 | |
803ea134 AG |
2213 | for_each_connection(connection, resource) |
2214 | kref_put(&connection->kref, drbd_destroy_connection); | |
2215 | kref_put(&resource->kref, drbd_destroy_resource); | |
b411b363 PR |
2216 | } |
2217 | ||
2312f0b3 LE |
2218 | /* One global retry thread, if we need to push back some bio and have it |
2219 | * reinserted through our make request function. | |
2220 | */ | |
2221 | static struct retry_worker { | |
2222 | struct workqueue_struct *wq; | |
2223 | struct work_struct worker; | |
2224 | ||
2225 | spinlock_t lock; | |
2226 | struct list_head writes; | |
2227 | } retry; | |
2228 | ||
2229 | static void do_retry(struct work_struct *ws) | |
2230 | { | |
2231 | struct retry_worker *retry = container_of(ws, struct retry_worker, worker); | |
2232 | LIST_HEAD(writes); | |
2233 | struct drbd_request *req, *tmp; | |
2234 | ||
2235 | spin_lock_irq(&retry->lock); | |
2236 | list_splice_init(&retry->writes, &writes); | |
2237 | spin_unlock_irq(&retry->lock); | |
2238 | ||
2239 | list_for_each_entry_safe(req, tmp, &writes, tl_requests) { | |
b30ab791 | 2240 | struct drbd_device *device = req->w.device; |
2312f0b3 LE |
2241 | struct bio *bio = req->master_bio; |
2242 | unsigned long start_time = req->start_time; | |
9a278a79 LE |
2243 | bool expected; |
2244 | ||
2245 | expected = | |
2246 | expect(atomic_read(&req->completion_ref) == 0) && | |
2247 | expect(req->rq_state & RQ_POSTPONED) && | |
2248 | expect((req->rq_state & RQ_LOCAL_PENDING) == 0 || | |
2249 | (req->rq_state & RQ_LOCAL_ABORTED) != 0); | |
2250 | ||
2251 | if (!expected) | |
d0180171 | 2252 | drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n", |
9a278a79 LE |
2253 | req, atomic_read(&req->completion_ref), |
2254 | req->rq_state); | |
2255 | ||
2256 | /* We still need to put one kref associated with the | |
2257 | * "completion_ref" going zero in the code path that queued it | |
2258 | * here. The request object may still be referenced by a | |
2259 | * frozen local req->private_bio, in case we force-detached. | |
2312f0b3 | 2260 | */ |
9a278a79 | 2261 | kref_put(&req->kref, drbd_req_destroy); |
2312f0b3 LE |
2262 | |
2263 | /* A single suspended or otherwise blocking device may stall | |
2264 | * all others as well. Fortunately, this code path is to | |
2265 | * recover from a situation that "should not happen": | |
2266 | * concurrent writes in multi-primary setup. | |
2267 | * In a "normal" lifecycle, this workqueue is supposed to be | |
2268 | * destroyed without ever doing anything. | |
2269 | * If it turns out to be an issue anyways, we can do per | |
2270 | * resource (replication group) or per device (minor) retry | |
2271 | * workqueues instead. | |
2272 | */ | |
2273 | ||
2274 | /* We are not just doing generic_make_request(), | |
2275 | * as we want to keep the start_time information. */ | |
b30ab791 AG |
2276 | inc_ap_bio(device); |
2277 | __drbd_make_request(device, bio, start_time); | |
2312f0b3 LE |
2278 | } |
2279 | } | |
2280 | ||
9d05e7c4 | 2281 | void drbd_restart_request(struct drbd_request *req) |
2312f0b3 LE |
2282 | { |
2283 | unsigned long flags; | |
2284 | spin_lock_irqsave(&retry.lock, flags); | |
2285 | list_move_tail(&req->tl_requests, &retry.writes); | |
2286 | spin_unlock_irqrestore(&retry.lock, flags); | |
2287 | ||
2288 | /* Drop the extra reference that would otherwise | |
2289 | * have been dropped by complete_master_bio. | |
2290 | * do_retry() needs to grab a new one. */ | |
b30ab791 | 2291 | dec_ap_bio(req->w.device); |
b411b363 | 2292 | |
2312f0b3 | 2293 | queue_work(retry.wq, &retry.worker); |
b411b363 PR |
2294 | } |
2295 | ||
77c556f6 AG |
2296 | void drbd_destroy_resource(struct kref *kref) |
2297 | { | |
2298 | struct drbd_resource *resource = | |
2299 | container_of(kref, struct drbd_resource, kref); | |
2300 | ||
803ea134 | 2301 | idr_destroy(&resource->devices); |
625a6ba2 | 2302 | free_cpumask_var(resource->cpu_mask); |
77c556f6 AG |
2303 | kfree(resource->name); |
2304 | kfree(resource); | |
2305 | } | |
2306 | ||
2307 | void drbd_free_resource(struct drbd_resource *resource) | |
2308 | { | |
2309 | struct drbd_connection *connection, *tmp; | |
2310 | ||
2311 | for_each_connection_safe(connection, tmp, resource) { | |
2312 | list_del(&connection->connections); | |
2313 | kref_put(&connection->kref, drbd_destroy_connection); | |
2314 | } | |
2315 | kref_put(&resource->kref, drbd_destroy_resource); | |
2316 | } | |
2312f0b3 | 2317 | |
b411b363 PR |
2318 | static void drbd_cleanup(void) |
2319 | { | |
2320 | unsigned int i; | |
b30ab791 | 2321 | struct drbd_device *device; |
77c556f6 | 2322 | struct drbd_resource *resource, *tmp; |
b411b363 PR |
2323 | |
2324 | unregister_reboot_notifier(&drbd_notifier); | |
2325 | ||
17a93f30 LE |
2326 | /* first remove proc, |
2327 | * drbdsetup uses it's presence to detect | |
2328 | * whether DRBD is loaded. | |
2329 | * If we would get stuck in proc removal, | |
2330 | * but have netlink already deregistered, | |
2331 | * some drbdsetup commands may wait forever | |
2332 | * for an answer. | |
2333 | */ | |
2334 | if (drbd_proc) | |
2335 | remove_proc_entry("drbd", NULL); | |
2336 | ||
2312f0b3 LE |
2337 | if (retry.wq) |
2338 | destroy_workqueue(retry.wq); | |
b411b363 | 2339 | |
3b98c0c2 | 2340 | drbd_genl_unregister(); |
b411b363 | 2341 | |
803ea134 | 2342 | idr_for_each_entry(&drbd_devices, device, i) |
f82795d6 | 2343 | drbd_delete_device(device); |
b411b363 | 2344 | |
c141ebda | 2345 | /* not _rcu since, no other updater anymore. Genl already unregistered */ |
77c556f6 AG |
2346 | for_each_resource_safe(resource, tmp, &drbd_resources) { |
2347 | list_del(&resource->resources); | |
2348 | drbd_free_resource(resource); | |
81fa2e67 | 2349 | } |
b411b363 | 2350 | |
81a5d60e | 2351 | drbd_destroy_mempools(); |
b411b363 PR |
2352 | unregister_blkdev(DRBD_MAJOR, "drbd"); |
2353 | ||
05a10ec7 | 2354 | idr_destroy(&drbd_devices); |
81a5d60e | 2355 | |
b411b363 PR |
2356 | printk(KERN_INFO "drbd: module cleanup done.\n"); |
2357 | } | |
2358 | ||
2359 | /** | |
d97482ed | 2360 | * drbd_congested() - Callback for the flusher thread |
b411b363 | 2361 | * @congested_data: User data |
d97482ed | 2362 | * @bdi_bits: Bits the BDI flusher thread is currently interested in |
b411b363 PR |
2363 | * |
2364 | * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested. | |
2365 | */ | |
2366 | static int drbd_congested(void *congested_data, int bdi_bits) | |
2367 | { | |
b30ab791 | 2368 | struct drbd_device *device = congested_data; |
b411b363 PR |
2369 | struct request_queue *q; |
2370 | char reason = '-'; | |
2371 | int r = 0; | |
2372 | ||
b30ab791 | 2373 | if (!may_inc_ap_bio(device)) { |
b411b363 PR |
2374 | /* DRBD has frozen IO */ |
2375 | r = bdi_bits; | |
2376 | reason = 'd'; | |
2377 | goto out; | |
2378 | } | |
2379 | ||
a6b32bc3 | 2380 | if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) { |
c2ba686f LE |
2381 | r |= (1 << BDI_async_congested); |
2382 | /* Without good local data, we would need to read from remote, | |
2383 | * and that would need the worker thread as well, which is | |
2384 | * currently blocked waiting for that usermode helper to | |
2385 | * finish. | |
2386 | */ | |
b30ab791 | 2387 | if (!get_ldev_if_state(device, D_UP_TO_DATE)) |
c2ba686f LE |
2388 | r |= (1 << BDI_sync_congested); |
2389 | else | |
b30ab791 | 2390 | put_ldev(device); |
c2ba686f LE |
2391 | r &= bdi_bits; |
2392 | reason = 'c'; | |
2393 | goto out; | |
2394 | } | |
2395 | ||
b30ab791 AG |
2396 | if (get_ldev(device)) { |
2397 | q = bdev_get_queue(device->ldev->backing_bdev); | |
b411b363 | 2398 | r = bdi_congested(&q->backing_dev_info, bdi_bits); |
b30ab791 | 2399 | put_ldev(device); |
b411b363 PR |
2400 | if (r) |
2401 | reason = 'b'; | |
2402 | } | |
2403 | ||
a6b32bc3 AG |
2404 | if (bdi_bits & (1 << BDI_async_congested) && |
2405 | test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) { | |
b411b363 PR |
2406 | r |= (1 << BDI_async_congested); |
2407 | reason = reason == 'b' ? 'a' : 'n'; | |
2408 | } | |
2409 | ||
2410 | out: | |
b30ab791 | 2411 | device->congestion_reason = reason; |
b411b363 PR |
2412 | return r; |
2413 | } | |
2414 | ||
6699b655 PR |
2415 | static void drbd_init_workqueue(struct drbd_work_queue* wq) |
2416 | { | |
6699b655 PR |
2417 | spin_lock_init(&wq->q_lock); |
2418 | INIT_LIST_HEAD(&wq->q); | |
8c0785a5 | 2419 | init_waitqueue_head(&wq->q_wait); |
6699b655 PR |
2420 | } |
2421 | ||
4bc76048 | 2422 | struct drbd_resource *drbd_find_resource(const char *name) |
1aba4d7f | 2423 | { |
77c556f6 | 2424 | struct drbd_resource *resource; |
1aba4d7f | 2425 | |
3b98c0c2 LE |
2426 | if (!name || !name[0]) |
2427 | return NULL; | |
2428 | ||
c141ebda | 2429 | rcu_read_lock(); |
77c556f6 AG |
2430 | for_each_resource_rcu(resource, &drbd_resources) { |
2431 | if (!strcmp(resource->name, name)) { | |
4bc76048 | 2432 | kref_get(&resource->kref); |
1aba4d7f | 2433 | goto found; |
0ace9dfa | 2434 | } |
1aba4d7f | 2435 | } |
4bc76048 | 2436 | resource = NULL; |
1aba4d7f | 2437 | found: |
c141ebda | 2438 | rcu_read_unlock(); |
4bc76048 | 2439 | return resource; |
1aba4d7f PR |
2440 | } |
2441 | ||
bde89a9e | 2442 | struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len, |
089c075d AG |
2443 | void *peer_addr, int peer_addr_len) |
2444 | { | |
77c556f6 | 2445 | struct drbd_resource *resource; |
bde89a9e | 2446 | struct drbd_connection *connection; |
089c075d AG |
2447 | |
2448 | rcu_read_lock(); | |
77c556f6 AG |
2449 | for_each_resource_rcu(resource, &drbd_resources) { |
2450 | for_each_connection_rcu(connection, resource) { | |
2451 | if (connection->my_addr_len == my_addr_len && | |
2452 | connection->peer_addr_len == peer_addr_len && | |
2453 | !memcmp(&connection->my_addr, my_addr, my_addr_len) && | |
2454 | !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) { | |
2455 | kref_get(&connection->kref); | |
2456 | goto found; | |
2457 | } | |
089c075d AG |
2458 | } |
2459 | } | |
bde89a9e | 2460 | connection = NULL; |
089c075d AG |
2461 | found: |
2462 | rcu_read_unlock(); | |
bde89a9e | 2463 | return connection; |
089c075d AG |
2464 | } |
2465 | ||
e6ef8a5c AG |
2466 | static int drbd_alloc_socket(struct drbd_socket *socket) |
2467 | { | |
2468 | socket->rbuf = (void *) __get_free_page(GFP_KERNEL); | |
2469 | if (!socket->rbuf) | |
2470 | return -ENOMEM; | |
5a87d920 AG |
2471 | socket->sbuf = (void *) __get_free_page(GFP_KERNEL); |
2472 | if (!socket->sbuf) | |
2473 | return -ENOMEM; | |
e6ef8a5c AG |
2474 | return 0; |
2475 | } | |
2476 | ||
2477 | static void drbd_free_socket(struct drbd_socket *socket) | |
2478 | { | |
5a87d920 | 2479 | free_page((unsigned long) socket->sbuf); |
e6ef8a5c AG |
2480 | free_page((unsigned long) socket->rbuf); |
2481 | } | |
2482 | ||
bde89a9e | 2483 | void conn_free_crypto(struct drbd_connection *connection) |
91fd4dad | 2484 | { |
bde89a9e | 2485 | drbd_free_sock(connection); |
1d041225 | 2486 | |
bde89a9e AG |
2487 | crypto_free_hash(connection->csums_tfm); |
2488 | crypto_free_hash(connection->verify_tfm); | |
2489 | crypto_free_hash(connection->cram_hmac_tfm); | |
2490 | crypto_free_hash(connection->integrity_tfm); | |
2491 | crypto_free_hash(connection->peer_integrity_tfm); | |
2492 | kfree(connection->int_dig_in); | |
2493 | kfree(connection->int_dig_vv); | |
1d041225 | 2494 | |
bde89a9e AG |
2495 | connection->csums_tfm = NULL; |
2496 | connection->verify_tfm = NULL; | |
2497 | connection->cram_hmac_tfm = NULL; | |
2498 | connection->integrity_tfm = NULL; | |
2499 | connection->peer_integrity_tfm = NULL; | |
2500 | connection->int_dig_in = NULL; | |
2501 | connection->int_dig_vv = NULL; | |
91fd4dad PR |
2502 | } |
2503 | ||
eb6bea67 | 2504 | int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts) |
afbbfa88 | 2505 | { |
eb6bea67 | 2506 | struct drbd_connection *connection; |
afbbfa88 AG |
2507 | cpumask_var_t new_cpu_mask; |
2508 | int err; | |
2509 | ||
2510 | if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) | |
2511 | return -ENOMEM; | |
2512 | /* | |
2513 | retcode = ERR_NOMEM; | |
2514 | drbd_msg_put_info("unable to allocate cpumask"); | |
2515 | */ | |
2516 | ||
2517 | /* silently ignore cpu mask on UP kernel */ | |
2518 | if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) { | |
f44d0436 | 2519 | err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE, |
c5b005ab | 2520 | cpumask_bits(new_cpu_mask), nr_cpu_ids); |
afbbfa88 | 2521 | if (err) { |
1ec861eb | 2522 | drbd_warn(resource, "bitmap_parse() failed with %d\n", err); |
afbbfa88 AG |
2523 | /* retcode = ERR_CPU_MASK_PARSE; */ |
2524 | goto fail; | |
2525 | } | |
2526 | } | |
eb6bea67 | 2527 | resource->res_opts = *res_opts; |
625a6ba2 AG |
2528 | if (cpumask_empty(new_cpu_mask)) |
2529 | drbd_calc_cpu_mask(&new_cpu_mask); | |
2530 | if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) { | |
2531 | cpumask_copy(resource->cpu_mask, new_cpu_mask); | |
2532 | for_each_connection_rcu(connection, resource) { | |
eb6bea67 AG |
2533 | connection->receiver.reset_cpu_mask = 1; |
2534 | connection->asender.reset_cpu_mask = 1; | |
2535 | connection->worker.reset_cpu_mask = 1; | |
2536 | } | |
afbbfa88 AG |
2537 | } |
2538 | err = 0; | |
2539 | ||
2540 | fail: | |
2541 | free_cpumask_var(new_cpu_mask); | |
2542 | return err; | |
2543 | ||
2544 | } | |
2545 | ||
77c556f6 AG |
2546 | struct drbd_resource *drbd_create_resource(const char *name) |
2547 | { | |
2548 | struct drbd_resource *resource; | |
2549 | ||
6bbf53ca | 2550 | resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL); |
77c556f6 | 2551 | if (!resource) |
625a6ba2 | 2552 | goto fail; |
77c556f6 | 2553 | resource->name = kstrdup(name, GFP_KERNEL); |
625a6ba2 AG |
2554 | if (!resource->name) |
2555 | goto fail_free_resource; | |
2556 | if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL)) | |
2557 | goto fail_free_name; | |
77c556f6 | 2558 | kref_init(&resource->kref); |
803ea134 | 2559 | idr_init(&resource->devices); |
77c556f6 AG |
2560 | INIT_LIST_HEAD(&resource->connections); |
2561 | list_add_tail_rcu(&resource->resources, &drbd_resources); | |
0500813f AG |
2562 | mutex_init(&resource->conf_update); |
2563 | spin_lock_init(&resource->req_lock); | |
77c556f6 | 2564 | return resource; |
625a6ba2 AG |
2565 | |
2566 | fail_free_name: | |
2567 | kfree(resource->name); | |
2568 | fail_free_resource: | |
2569 | kfree(resource); | |
2570 | fail: | |
2571 | return NULL; | |
77c556f6 AG |
2572 | } |
2573 | ||
ec0bddbc | 2574 | /* caller must be under genl_lock() */ |
bde89a9e | 2575 | struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts) |
2111438b | 2576 | { |
77c556f6 | 2577 | struct drbd_resource *resource; |
bde89a9e | 2578 | struct drbd_connection *connection; |
2111438b | 2579 | |
bde89a9e AG |
2580 | connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL); |
2581 | if (!connection) | |
2111438b PR |
2582 | return NULL; |
2583 | ||
bde89a9e | 2584 | if (drbd_alloc_socket(&connection->data)) |
e6ef8a5c | 2585 | goto fail; |
bde89a9e | 2586 | if (drbd_alloc_socket(&connection->meta)) |
e6ef8a5c AG |
2587 | goto fail; |
2588 | ||
bde89a9e AG |
2589 | connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL); |
2590 | if (!connection->current_epoch) | |
12038a3a | 2591 | goto fail; |
b6dd1a89 | 2592 | |
bde89a9e | 2593 | INIT_LIST_HEAD(&connection->transfer_log); |
b6dd1a89 | 2594 | |
bde89a9e AG |
2595 | INIT_LIST_HEAD(&connection->current_epoch->list); |
2596 | connection->epochs = 1; | |
2597 | spin_lock_init(&connection->epoch_lock); | |
2598 | connection->write_ordering = WO_bdev_flush; | |
4b0007c0 | 2599 | |
bde89a9e AG |
2600 | connection->send.seen_any_write_yet = false; |
2601 | connection->send.current_epoch_nr = 0; | |
2602 | connection->send.current_epoch_writes = 0; | |
b6dd1a89 | 2603 | |
77c556f6 AG |
2604 | resource = drbd_create_resource(name); |
2605 | if (!resource) | |
2606 | goto fail; | |
2607 | ||
bde89a9e AG |
2608 | connection->cstate = C_STANDALONE; |
2609 | mutex_init(&connection->cstate_mutex); | |
bde89a9e | 2610 | init_waitqueue_head(&connection->ping_wait); |
c06ece6b | 2611 | idr_init(&connection->peer_devices); |
b2fb6dbe | 2612 | |
bde89a9e AG |
2613 | drbd_init_workqueue(&connection->sender_work); |
2614 | mutex_init(&connection->data.mutex); | |
2615 | mutex_init(&connection->meta.mutex); | |
6699b655 | 2616 | |
bde89a9e AG |
2617 | drbd_thread_init(connection, &connection->receiver, drbdd_init, "receiver"); |
2618 | drbd_thread_init(connection, &connection->worker, drbd_worker, "worker"); | |
2619 | drbd_thread_init(connection, &connection->asender, drbd_asender, "asender"); | |
392c8801 | 2620 | |
bde89a9e | 2621 | kref_init(&connection->kref); |
77c556f6 | 2622 | |
77c556f6 | 2623 | connection->resource = resource; |
2111438b | 2624 | |
eb6bea67 AG |
2625 | if (set_resource_options(resource, res_opts)) |
2626 | goto fail_resource; | |
2627 | ||
2628 | kref_get(&resource->kref); | |
2629 | list_add_tail_rcu(&connection->connections, &resource->connections); | |
bde89a9e | 2630 | return connection; |
2111438b | 2631 | |
eb6bea67 AG |
2632 | fail_resource: |
2633 | list_del(&resource->resources); | |
2634 | drbd_free_resource(resource); | |
2111438b | 2635 | fail: |
bde89a9e | 2636 | kfree(connection->current_epoch); |
bde89a9e AG |
2637 | drbd_free_socket(&connection->meta); |
2638 | drbd_free_socket(&connection->data); | |
bde89a9e | 2639 | kfree(connection); |
2111438b PR |
2640 | return NULL; |
2641 | } | |
2642 | ||
05a10ec7 | 2643 | void drbd_destroy_connection(struct kref *kref) |
2111438b | 2644 | { |
bde89a9e | 2645 | struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref); |
77c556f6 | 2646 | struct drbd_resource *resource = connection->resource; |
9dc9fbb3 | 2647 | |
bde89a9e | 2648 | if (atomic_read(&connection->current_epoch->epoch_size) != 0) |
1ec861eb | 2649 | drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size)); |
bde89a9e | 2650 | kfree(connection->current_epoch); |
12038a3a | 2651 | |
c06ece6b | 2652 | idr_destroy(&connection->peer_devices); |
2111438b | 2653 | |
bde89a9e AG |
2654 | drbd_free_socket(&connection->meta); |
2655 | drbd_free_socket(&connection->data); | |
bde89a9e AG |
2656 | kfree(connection->int_dig_in); |
2657 | kfree(connection->int_dig_vv); | |
2658 | kfree(connection); | |
77c556f6 | 2659 | kref_put(&resource->kref, drbd_destroy_resource); |
2111438b PR |
2660 | } |
2661 | ||
b30ab791 | 2662 | static int init_submitter(struct drbd_device *device) |
113fef9e LE |
2663 | { |
2664 | /* opencoded create_singlethread_workqueue(), | |
2665 | * to be able to say "drbd%d", ..., minor */ | |
b30ab791 AG |
2666 | device->submit.wq = alloc_workqueue("drbd%u_submit", |
2667 | WQ_UNBOUND | WQ_MEM_RECLAIM, 1, device->minor); | |
2668 | if (!device->submit.wq) | |
113fef9e LE |
2669 | return -ENOMEM; |
2670 | ||
b30ab791 AG |
2671 | INIT_WORK(&device->submit.worker, do_submit); |
2672 | spin_lock_init(&device->submit.lock); | |
2673 | INIT_LIST_HEAD(&device->submit.writes); | |
113fef9e LE |
2674 | return 0; |
2675 | } | |
2676 | ||
59515a2e | 2677 | enum drbd_ret_code drbd_create_device(struct drbd_resource *resource, unsigned int minor, int vnr) |
b411b363 | 2678 | { |
b6f85ef9 | 2679 | struct drbd_connection *connection; |
b30ab791 | 2680 | struct drbd_device *device; |
b6f85ef9 | 2681 | struct drbd_peer_device *peer_device, *tmp_peer_device; |
b411b363 PR |
2682 | struct gendisk *disk; |
2683 | struct request_queue *q; | |
93e4bf7a | 2684 | int id; |
8432b314 | 2685 | enum drbd_ret_code err = ERR_NOMEM; |
774b3055 | 2686 | |
b30ab791 AG |
2687 | device = minor_to_device(minor); |
2688 | if (device) | |
774b3055 | 2689 | return ERR_MINOR_EXISTS; |
b411b363 PR |
2690 | |
2691 | /* GFP_KERNEL, we are outside of all write-out paths */ | |
b30ab791 AG |
2692 | device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL); |
2693 | if (!device) | |
774b3055 | 2694 | return ERR_NOMEM; |
803ea134 AG |
2695 | kref_init(&device->kref); |
2696 | ||
803ea134 AG |
2697 | kref_get(&resource->kref); |
2698 | device->resource = resource; | |
b30ab791 AG |
2699 | device->minor = minor; |
2700 | device->vnr = vnr; | |
b411b363 | 2701 | |
b30ab791 | 2702 | drbd_init_set_defaults(device); |
b411b363 PR |
2703 | |
2704 | q = blk_alloc_queue(GFP_KERNEL); | |
2705 | if (!q) | |
2706 | goto out_no_q; | |
b30ab791 AG |
2707 | device->rq_queue = q; |
2708 | q->queuedata = device; | |
b411b363 PR |
2709 | |
2710 | disk = alloc_disk(1); | |
2711 | if (!disk) | |
2712 | goto out_no_disk; | |
b30ab791 | 2713 | device->vdisk = disk; |
b411b363 | 2714 | |
81e84650 | 2715 | set_disk_ro(disk, true); |
b411b363 PR |
2716 | |
2717 | disk->queue = q; | |
2718 | disk->major = DRBD_MAJOR; | |
2719 | disk->first_minor = minor; | |
2720 | disk->fops = &drbd_ops; | |
2721 | sprintf(disk->disk_name, "drbd%d", minor); | |
b30ab791 | 2722 | disk->private_data = device; |
b411b363 | 2723 | |
b30ab791 | 2724 | device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor)); |
b411b363 | 2725 | /* we have no partitions. we contain only ourselves. */ |
b30ab791 | 2726 | device->this_bdev->bd_contains = device->this_bdev; |
b411b363 PR |
2727 | |
2728 | q->backing_dev_info.congested_fn = drbd_congested; | |
b30ab791 | 2729 | q->backing_dev_info.congested_data = device; |
b411b363 | 2730 | |
2f58dcfc | 2731 | blk_queue_make_request(q, drbd_make_request); |
a73ff323 | 2732 | blk_queue_flush(q, REQ_FLUSH | REQ_FUA); |
99432fcc PR |
2733 | /* Setting the max_hw_sectors to an odd value of 8kibyte here |
2734 | This triggers a max_bio_size message upon first attach or connect */ | |
2735 | blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8); | |
b411b363 PR |
2736 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
2737 | blk_queue_merge_bvec(q, drbd_merge_bvec); | |
0500813f | 2738 | q->queue_lock = &resource->req_lock; |
b411b363 | 2739 | |
b30ab791 AG |
2740 | device->md_io_page = alloc_page(GFP_KERNEL); |
2741 | if (!device->md_io_page) | |
b411b363 PR |
2742 | goto out_no_io_page; |
2743 | ||
b30ab791 | 2744 | if (drbd_bm_init(device)) |
b411b363 | 2745 | goto out_no_bitmap; |
b30ab791 AG |
2746 | device->read_requests = RB_ROOT; |
2747 | device->write_requests = RB_ROOT; | |
b411b363 | 2748 | |
93e4bf7a AG |
2749 | id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL); |
2750 | if (id < 0) { | |
2751 | if (id == -ENOSPC) { | |
56de2102 TH |
2752 | err = ERR_MINOR_EXISTS; |
2753 | drbd_msg_put_info("requested minor exists already"); | |
2754 | } | |
8432b314 | 2755 | goto out_no_minor_idr; |
81a5d60e | 2756 | } |
803ea134 AG |
2757 | kref_get(&device->kref); |
2758 | ||
2759 | id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL); | |
2760 | if (id < 0) { | |
2761 | if (id == -ENOSPC) { | |
2762 | err = ERR_MINOR_EXISTS; | |
2763 | drbd_msg_put_info("requested minor exists already"); | |
2764 | } | |
2765 | goto out_idr_remove_minor; | |
2766 | } | |
2767 | kref_get(&device->kref); | |
8432b314 | 2768 | |
b6f85ef9 AG |
2769 | INIT_LIST_HEAD(&device->peer_devices); |
2770 | for_each_connection(connection, resource) { | |
2771 | peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL); | |
2772 | if (!peer_device) | |
2773 | goto out_idr_remove_from_resource; | |
2774 | peer_device->connection = connection; | |
2775 | peer_device->device = device; | |
2776 | ||
2777 | list_add(&peer_device->peer_devices, &device->peer_devices); | |
2778 | kref_get(&device->kref); | |
2779 | ||
2780 | id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL); | |
2781 | if (id < 0) { | |
2782 | if (id == -ENOSPC) { | |
2783 | err = ERR_INVALID_REQUEST; | |
2784 | drbd_msg_put_info("requested volume exists already"); | |
2785 | } | |
2786 | goto out_idr_remove_from_resource; | |
56de2102 | 2787 | } |
b6f85ef9 | 2788 | kref_get(&connection->kref); |
8432b314 | 2789 | } |
56de2102 | 2790 | |
b30ab791 | 2791 | if (init_submitter(device)) { |
113fef9e LE |
2792 | err = ERR_NOMEM; |
2793 | drbd_msg_put_info("unable to create submit workqueue"); | |
2794 | goto out_idr_remove_vol; | |
2795 | } | |
2796 | ||
774b3055 PR |
2797 | add_disk(disk); |
2798 | ||
2325eb66 | 2799 | /* inherit the connection state */ |
b6f85ef9 | 2800 | device->state.conn = first_connection(resource)->cstate; |
b30ab791 AG |
2801 | if (device->state.conn == C_WF_REPORT_PARAMS) |
2802 | drbd_connected(device); | |
2325eb66 | 2803 | |
774b3055 | 2804 | return NO_ERROR; |
b411b363 | 2805 | |
113fef9e | 2806 | out_idr_remove_vol: |
c06ece6b | 2807 | idr_remove(&connection->peer_devices, vnr); |
803ea134 | 2808 | out_idr_remove_from_resource: |
b6f85ef9 AG |
2809 | for_each_connection(connection, resource) { |
2810 | peer_device = idr_find(&connection->peer_devices, vnr); | |
2811 | if (peer_device) { | |
2812 | idr_remove(&connection->peer_devices, vnr); | |
2813 | kref_put(&connection->kref, drbd_destroy_connection); | |
2814 | } | |
2815 | } | |
2816 | for_each_peer_device_safe(peer_device, tmp_peer_device, device) { | |
2817 | list_del(&peer_device->peer_devices); | |
2818 | kfree(peer_device); | |
2819 | } | |
803ea134 | 2820 | idr_remove(&resource->devices, vnr); |
8432b314 | 2821 | out_idr_remove_minor: |
93e4bf7a | 2822 | idr_remove(&drbd_devices, minor); |
569083c0 | 2823 | synchronize_rcu(); |
8432b314 | 2824 | out_no_minor_idr: |
b30ab791 | 2825 | drbd_bm_cleanup(device); |
b411b363 | 2826 | out_no_bitmap: |
b30ab791 | 2827 | __free_page(device->md_io_page); |
b411b363 PR |
2828 | out_no_io_page: |
2829 | put_disk(disk); | |
2830 | out_no_disk: | |
2831 | blk_cleanup_queue(q); | |
2832 | out_no_q: | |
803ea134 | 2833 | kref_put(&resource->kref, drbd_destroy_resource); |
a6b32bc3 | 2834 | kfree(device); |
8432b314 | 2835 | return err; |
b411b363 PR |
2836 | } |
2837 | ||
f82795d6 | 2838 | void drbd_delete_device(struct drbd_device *device) |
803ea134 AG |
2839 | { |
2840 | struct drbd_resource *resource = device->resource; | |
2841 | struct drbd_connection *connection; | |
2842 | int refs = 3; | |
2843 | ||
2844 | for_each_connection(connection, resource) { | |
c06ece6b | 2845 | idr_remove(&connection->peer_devices, device->vnr); |
803ea134 AG |
2846 | refs++; |
2847 | } | |
2848 | idr_remove(&resource->devices, device->vnr); | |
2849 | idr_remove(&drbd_devices, device_to_minor(device)); | |
2850 | del_gendisk(device->vdisk); | |
2851 | synchronize_rcu(); | |
2852 | kref_sub(&device->kref, refs, drbd_destroy_device); | |
2853 | } | |
2854 | ||
b411b363 PR |
2855 | int __init drbd_init(void) |
2856 | { | |
2857 | int err; | |
2858 | ||
2b8a90b5 | 2859 | if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) { |
b411b363 | 2860 | printk(KERN_ERR |
81a5d60e | 2861 | "drbd: invalid minor_count (%d)\n", minor_count); |
b411b363 PR |
2862 | #ifdef MODULE |
2863 | return -EINVAL; | |
2864 | #else | |
46530e85 | 2865 | minor_count = DRBD_MINOR_COUNT_DEF; |
b411b363 PR |
2866 | #endif |
2867 | } | |
2868 | ||
b411b363 PR |
2869 | err = register_blkdev(DRBD_MAJOR, "drbd"); |
2870 | if (err) { | |
2871 | printk(KERN_ERR | |
2872 | "drbd: unable to register block device major %d\n", | |
2873 | DRBD_MAJOR); | |
2874 | return err; | |
2875 | } | |
2876 | ||
2877 | register_reboot_notifier(&drbd_notifier); | |
2878 | ||
2879 | /* | |
2880 | * allocate all necessary structs | |
2881 | */ | |
b411b363 PR |
2882 | init_waitqueue_head(&drbd_pp_wait); |
2883 | ||
2884 | drbd_proc = NULL; /* play safe for drbd_cleanup */ | |
05a10ec7 | 2885 | idr_init(&drbd_devices); |
b411b363 | 2886 | |
69babf05 | 2887 | rwlock_init(&global_state_lock); |
77c556f6 | 2888 | INIT_LIST_HEAD(&drbd_resources); |
69babf05 LE |
2889 | |
2890 | err = drbd_genl_register(); | |
2891 | if (err) { | |
2892 | printk(KERN_ERR "drbd: unable to register generic netlink family\n"); | |
2893 | goto fail; | |
2894 | } | |
2895 | ||
b411b363 PR |
2896 | err = drbd_create_mempools(); |
2897 | if (err) | |
3b98c0c2 | 2898 | goto fail; |
b411b363 | 2899 | |
6110d70b | 2900 | err = -ENOMEM; |
8c484ee4 | 2901 | drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL); |
b411b363 PR |
2902 | if (!drbd_proc) { |
2903 | printk(KERN_ERR "drbd: unable to register proc file\n"); | |
3b98c0c2 | 2904 | goto fail; |
b411b363 PR |
2905 | } |
2906 | ||
2312f0b3 LE |
2907 | retry.wq = create_singlethread_workqueue("drbd-reissue"); |
2908 | if (!retry.wq) { | |
2909 | printk(KERN_ERR "drbd: unable to create retry workqueue\n"); | |
2910 | goto fail; | |
2911 | } | |
2912 | INIT_WORK(&retry.worker, do_retry); | |
2913 | spin_lock_init(&retry.lock); | |
2914 | INIT_LIST_HEAD(&retry.writes); | |
b411b363 PR |
2915 | |
2916 | printk(KERN_INFO "drbd: initialized. " | |
2917 | "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n", | |
2918 | API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX); | |
2919 | printk(KERN_INFO "drbd: %s\n", drbd_buildtag()); | |
2920 | printk(KERN_INFO "drbd: registered as block device major %d\n", | |
2921 | DRBD_MAJOR); | |
b411b363 PR |
2922 | |
2923 | return 0; /* Success! */ | |
2924 | ||
3b98c0c2 | 2925 | fail: |
b411b363 PR |
2926 | drbd_cleanup(); |
2927 | if (err == -ENOMEM) | |
b411b363 PR |
2928 | printk(KERN_ERR "drbd: ran out of memory\n"); |
2929 | else | |
2930 | printk(KERN_ERR "drbd: initialization failure\n"); | |
2931 | return err; | |
2932 | } | |
2933 | ||
2934 | void drbd_free_bc(struct drbd_backing_dev *ldev) | |
2935 | { | |
2936 | if (ldev == NULL) | |
2937 | return; | |
2938 | ||
e525fd89 TH |
2939 | blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
2940 | blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
b411b363 | 2941 | |
94ad0a10 | 2942 | kfree(ldev->disk_conf); |
b411b363 PR |
2943 | kfree(ldev); |
2944 | } | |
2945 | ||
bde89a9e | 2946 | void drbd_free_sock(struct drbd_connection *connection) |
b411b363 | 2947 | { |
bde89a9e AG |
2948 | if (connection->data.socket) { |
2949 | mutex_lock(&connection->data.mutex); | |
2950 | kernel_sock_shutdown(connection->data.socket, SHUT_RDWR); | |
2951 | sock_release(connection->data.socket); | |
2952 | connection->data.socket = NULL; | |
2953 | mutex_unlock(&connection->data.mutex); | |
b411b363 | 2954 | } |
bde89a9e AG |
2955 | if (connection->meta.socket) { |
2956 | mutex_lock(&connection->meta.mutex); | |
2957 | kernel_sock_shutdown(connection->meta.socket, SHUT_RDWR); | |
2958 | sock_release(connection->meta.socket); | |
2959 | connection->meta.socket = NULL; | |
2960 | mutex_unlock(&connection->meta.mutex); | |
b411b363 PR |
2961 | } |
2962 | } | |
2963 | ||
b411b363 | 2964 | /* meta data management */ |
b411b363 | 2965 | |
bde89a9e | 2966 | void conn_md_sync(struct drbd_connection *connection) |
b411b363 | 2967 | { |
c06ece6b | 2968 | struct drbd_peer_device *peer_device; |
19fffd7b | 2969 | int vnr; |
b411b363 | 2970 | |
19fffd7b | 2971 | rcu_read_lock(); |
c06ece6b AG |
2972 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
2973 | struct drbd_device *device = peer_device->device; | |
2974 | ||
b30ab791 | 2975 | kref_get(&device->kref); |
19fffd7b | 2976 | rcu_read_unlock(); |
b30ab791 | 2977 | drbd_md_sync(device); |
05a10ec7 | 2978 | kref_put(&device->kref, drbd_destroy_device); |
19fffd7b PR |
2979 | rcu_read_lock(); |
2980 | } | |
2981 | rcu_read_unlock(); | |
b411b363 PR |
2982 | } |
2983 | ||
ae8bf312 | 2984 | /* aligned 4kByte */ |
b411b363 | 2985 | struct meta_data_on_disk { |
cccac985 | 2986 | u64 la_size_sect; /* last agreed size. */ |
b411b363 PR |
2987 | u64 uuid[UI_SIZE]; /* UUIDs. */ |
2988 | u64 device_uuid; | |
2989 | u64 reserved_u64_1; | |
2990 | u32 flags; /* MDF */ | |
2991 | u32 magic; | |
2992 | u32 md_size_sect; | |
2993 | u32 al_offset; /* offset to this block */ | |
ae8bf312 | 2994 | u32 al_nr_extents; /* important for restoring the AL (userspace) */ |
f399002e | 2995 | /* `-- act_log->nr_elements <-- ldev->dc.al_extents */ |
b411b363 PR |
2996 | u32 bm_offset; /* offset to the bitmap, from here */ |
2997 | u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */ | |
99432fcc | 2998 | u32 la_peer_max_bio_size; /* last peer max_bio_size */ |
b411b363 | 2999 | |
3a4d4eb3 LE |
3000 | /* see al_tr_number_to_on_disk_sector() */ |
3001 | u32 al_stripes; | |
3002 | u32 al_stripe_size_4k; | |
3003 | ||
3004 | u8 reserved_u8[4096 - (7*8 + 10*4)]; | |
b411b363 PR |
3005 | } __packed; |
3006 | ||
d752b269 PR |
3007 | |
3008 | ||
b30ab791 | 3009 | void drbd_md_write(struct drbd_device *device, void *b) |
b411b363 | 3010 | { |
d752b269 | 3011 | struct meta_data_on_disk *buffer = b; |
b411b363 PR |
3012 | sector_t sector; |
3013 | int i; | |
3014 | ||
ae8bf312 | 3015 | memset(buffer, 0, sizeof(*buffer)); |
b411b363 | 3016 | |
b30ab791 | 3017 | buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev)); |
b411b363 | 3018 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
3019 | buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
3020 | buffer->flags = cpu_to_be32(device->ldev->md.flags); | |
d5d7ebd4 | 3021 | buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN); |
b411b363 | 3022 | |
b30ab791 AG |
3023 | buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect); |
3024 | buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset); | |
3025 | buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements); | |
b411b363 | 3026 | buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE); |
b30ab791 | 3027 | buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid); |
b411b363 | 3028 | |
b30ab791 AG |
3029 | buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset); |
3030 | buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size); | |
b411b363 | 3031 | |
b30ab791 AG |
3032 | buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes); |
3033 | buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k); | |
3a4d4eb3 | 3034 | |
0b0ba1ef | 3035 | D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset); |
b30ab791 | 3036 | sector = device->ldev->md.md_offset; |
b411b363 | 3037 | |
b30ab791 | 3038 | if (drbd_md_sync_page_io(device, device->ldev, sector, WRITE)) { |
b411b363 | 3039 | /* this was a try anyways ... */ |
d0180171 | 3040 | drbd_err(device, "meta data update failed!\n"); |
b30ab791 | 3041 | drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR); |
b411b363 | 3042 | } |
d752b269 PR |
3043 | } |
3044 | ||
3045 | /** | |
3046 | * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set | |
b30ab791 | 3047 | * @device: DRBD device. |
d752b269 | 3048 | */ |
b30ab791 | 3049 | void drbd_md_sync(struct drbd_device *device) |
d752b269 PR |
3050 | { |
3051 | struct meta_data_on_disk *buffer; | |
3052 | ||
3053 | /* Don't accidentally change the DRBD meta data layout. */ | |
3054 | BUILD_BUG_ON(UI_SIZE != 4); | |
3055 | BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096); | |
3056 | ||
b30ab791 | 3057 | del_timer(&device->md_sync_timer); |
d752b269 | 3058 | /* timer may be rearmed by drbd_md_mark_dirty() now. */ |
b30ab791 | 3059 | if (!test_and_clear_bit(MD_DIRTY, &device->flags)) |
d752b269 PR |
3060 | return; |
3061 | ||
3062 | /* We use here D_FAILED and not D_ATTACHING because we try to write | |
3063 | * metadata even if we detach due to a disk failure! */ | |
b30ab791 | 3064 | if (!get_ldev_if_state(device, D_FAILED)) |
d752b269 PR |
3065 | return; |
3066 | ||
b30ab791 | 3067 | buffer = drbd_md_get_buffer(device); |
d752b269 PR |
3068 | if (!buffer) |
3069 | goto out; | |
3070 | ||
b30ab791 | 3071 | drbd_md_write(device, buffer); |
b411b363 | 3072 | |
b30ab791 | 3073 | /* Update device->ldev->md.la_size_sect, |
b411b363 | 3074 | * since we updated it on metadata. */ |
b30ab791 | 3075 | device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev); |
b411b363 | 3076 | |
b30ab791 | 3077 | drbd_md_put_buffer(device); |
e1711731 | 3078 | out: |
b30ab791 | 3079 | put_ldev(device); |
b411b363 PR |
3080 | } |
3081 | ||
b30ab791 | 3082 | static int check_activity_log_stripe_size(struct drbd_device *device, |
3a4d4eb3 LE |
3083 | struct meta_data_on_disk *on_disk, |
3084 | struct drbd_md *in_core) | |
3085 | { | |
3086 | u32 al_stripes = be32_to_cpu(on_disk->al_stripes); | |
3087 | u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k); | |
3088 | u64 al_size_4k; | |
3089 | ||
3090 | /* both not set: default to old fixed size activity log */ | |
3091 | if (al_stripes == 0 && al_stripe_size_4k == 0) { | |
3092 | al_stripes = 1; | |
3093 | al_stripe_size_4k = MD_32kB_SECT/8; | |
3094 | } | |
3095 | ||
3096 | /* some paranoia plausibility checks */ | |
3097 | ||
3098 | /* we need both values to be set */ | |
3099 | if (al_stripes == 0 || al_stripe_size_4k == 0) | |
3100 | goto err; | |
3101 | ||
3102 | al_size_4k = (u64)al_stripes * al_stripe_size_4k; | |
3103 | ||
3104 | /* Upper limit of activity log area, to avoid potential overflow | |
3105 | * problems in al_tr_number_to_on_disk_sector(). As right now, more | |
3106 | * than 72 * 4k blocks total only increases the amount of history, | |
3107 | * limiting this arbitrarily to 16 GB is not a real limitation ;-) */ | |
3108 | if (al_size_4k > (16 * 1024 * 1024/4)) | |
3109 | goto err; | |
3110 | ||
3111 | /* Lower limit: we need at least 8 transaction slots (32kB) | |
3112 | * to not break existing setups */ | |
3113 | if (al_size_4k < MD_32kB_SECT/8) | |
3114 | goto err; | |
3115 | ||
3116 | in_core->al_stripe_size_4k = al_stripe_size_4k; | |
3117 | in_core->al_stripes = al_stripes; | |
3118 | in_core->al_size_4k = al_size_4k; | |
3119 | ||
3120 | return 0; | |
3121 | err: | |
d0180171 | 3122 | drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n", |
3a4d4eb3 LE |
3123 | al_stripes, al_stripe_size_4k); |
3124 | return -EINVAL; | |
3125 | } | |
3126 | ||
b30ab791 | 3127 | static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev) |
c04ccaa6 LE |
3128 | { |
3129 | sector_t capacity = drbd_get_capacity(bdev->md_bdev); | |
3130 | struct drbd_md *in_core = &bdev->md; | |
3131 | s32 on_disk_al_sect; | |
3132 | s32 on_disk_bm_sect; | |
3133 | ||
3134 | /* The on-disk size of the activity log, calculated from offsets, and | |
3135 | * the size of the activity log calculated from the stripe settings, | |
3136 | * should match. | |
3137 | * Though we could relax this a bit: it is ok, if the striped activity log | |
3138 | * fits in the available on-disk activity log size. | |
3139 | * Right now, that would break how resize is implemented. | |
3140 | * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware | |
3141 | * of possible unused padding space in the on disk layout. */ | |
3142 | if (in_core->al_offset < 0) { | |
3143 | if (in_core->bm_offset > in_core->al_offset) | |
3144 | goto err; | |
3145 | on_disk_al_sect = -in_core->al_offset; | |
3146 | on_disk_bm_sect = in_core->al_offset - in_core->bm_offset; | |
3147 | } else { | |
3148 | if (in_core->al_offset != MD_4kB_SECT) | |
3149 | goto err; | |
3150 | if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT) | |
3151 | goto err; | |
3152 | ||
3153 | on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT; | |
3154 | on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset; | |
3155 | } | |
3156 | ||
3157 | /* old fixed size meta data is exactly that: fixed. */ | |
3158 | if (in_core->meta_dev_idx >= 0) { | |
3159 | if (in_core->md_size_sect != MD_128MB_SECT | |
3160 | || in_core->al_offset != MD_4kB_SECT | |
3161 | || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT | |
3162 | || in_core->al_stripes != 1 | |
3163 | || in_core->al_stripe_size_4k != MD_32kB_SECT/8) | |
3164 | goto err; | |
3165 | } | |
3166 | ||
3167 | if (capacity < in_core->md_size_sect) | |
3168 | goto err; | |
3169 | if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev)) | |
3170 | goto err; | |
3171 | ||
3172 | /* should be aligned, and at least 32k */ | |
3173 | if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT)) | |
3174 | goto err; | |
3175 | ||
3176 | /* should fit (for now: exactly) into the available on-disk space; | |
3177 | * overflow prevention is in check_activity_log_stripe_size() above. */ | |
3178 | if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT) | |
3179 | goto err; | |
3180 | ||
3181 | /* again, should be aligned */ | |
3182 | if (in_core->bm_offset & 7) | |
3183 | goto err; | |
3184 | ||
3185 | /* FIXME check for device grow with flex external meta data? */ | |
3186 | ||
3187 | /* can the available bitmap space cover the last agreed device size? */ | |
3188 | if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512) | |
3189 | goto err; | |
3190 | ||
3191 | return 0; | |
3192 | ||
3193 | err: | |
d0180171 | 3194 | drbd_err(device, "meta data offsets don't make sense: idx=%d " |
c04ccaa6 LE |
3195 | "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, " |
3196 | "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n", | |
3197 | in_core->meta_dev_idx, | |
3198 | in_core->al_stripes, in_core->al_stripe_size_4k, | |
3199 | in_core->al_offset, in_core->bm_offset, in_core->md_size_sect, | |
3200 | (unsigned long long)in_core->la_size_sect, | |
3201 | (unsigned long long)capacity); | |
3202 | ||
3203 | return -EINVAL; | |
3204 | } | |
3205 | ||
3206 | ||
b411b363 PR |
3207 | /** |
3208 | * drbd_md_read() - Reads in the meta data super block | |
b30ab791 | 3209 | * @device: DRBD device. |
b411b363 PR |
3210 | * @bdev: Device from which the meta data should be read in. |
3211 | * | |
3a4d4eb3 | 3212 | * Return NO_ERROR on success, and an enum drbd_ret_code in case |
d5d7ebd4 | 3213 | * something goes wrong. |
3a4d4eb3 | 3214 | * |
c04ccaa6 | 3215 | * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS, |
b30ab791 | 3216 | * even before @bdev is assigned to @device->ldev. |
b411b363 | 3217 | */ |
b30ab791 | 3218 | int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev) |
b411b363 PR |
3219 | { |
3220 | struct meta_data_on_disk *buffer; | |
d5d7ebd4 | 3221 | u32 magic, flags; |
b411b363 PR |
3222 | int i, rv = NO_ERROR; |
3223 | ||
b30ab791 | 3224 | if (device->state.disk != D_DISKLESS) |
c04ccaa6 | 3225 | return ERR_DISK_CONFIGURED; |
b411b363 | 3226 | |
b30ab791 | 3227 | buffer = drbd_md_get_buffer(device); |
e1711731 | 3228 | if (!buffer) |
c04ccaa6 | 3229 | return ERR_NOMEM; |
b411b363 | 3230 | |
c04ccaa6 LE |
3231 | /* First, figure out where our meta data superblock is located, |
3232 | * and read it. */ | |
3a4d4eb3 LE |
3233 | bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx; |
3234 | bdev->md.md_offset = drbd_md_ss(bdev); | |
b411b363 | 3235 | |
b30ab791 | 3236 | if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, READ)) { |
25985edc | 3237 | /* NOTE: can't do normal error processing here as this is |
b411b363 | 3238 | called BEFORE disk is attached */ |
d0180171 | 3239 | drbd_err(device, "Error while reading metadata.\n"); |
b411b363 PR |
3240 | rv = ERR_IO_MD_DISK; |
3241 | goto err; | |
3242 | } | |
3243 | ||
d5d7ebd4 LE |
3244 | magic = be32_to_cpu(buffer->magic); |
3245 | flags = be32_to_cpu(buffer->flags); | |
3246 | if (magic == DRBD_MD_MAGIC_84_UNCLEAN || | |
3247 | (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) { | |
3248 | /* btw: that's Activity Log clean, not "all" clean. */ | |
d0180171 | 3249 | drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n"); |
d5d7ebd4 LE |
3250 | rv = ERR_MD_UNCLEAN; |
3251 | goto err; | |
3252 | } | |
3a4d4eb3 LE |
3253 | |
3254 | rv = ERR_MD_INVALID; | |
d5d7ebd4 | 3255 | if (magic != DRBD_MD_MAGIC_08) { |
43de7c85 | 3256 | if (magic == DRBD_MD_MAGIC_07) |
d0180171 | 3257 | drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n"); |
d5d7ebd4 | 3258 | else |
d0180171 | 3259 | drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n"); |
b411b363 PR |
3260 | goto err; |
3261 | } | |
3a4d4eb3 | 3262 | |
c04ccaa6 | 3263 | if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) { |
d0180171 | 3264 | drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n", |
c04ccaa6 | 3265 | be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE); |
b411b363 PR |
3266 | goto err; |
3267 | } | |
3a4d4eb3 | 3268 | |
c04ccaa6 LE |
3269 | |
3270 | /* convert to in_core endian */ | |
3271 | bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect); | |
3272 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
3273 | bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]); | |
3274 | bdev->md.flags = be32_to_cpu(buffer->flags); | |
3275 | bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid); | |
3276 | ||
3277 | bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect); | |
3278 | bdev->md.al_offset = be32_to_cpu(buffer->al_offset); | |
3279 | bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset); | |
3280 | ||
b30ab791 | 3281 | if (check_activity_log_stripe_size(device, buffer, &bdev->md)) |
b411b363 | 3282 | goto err; |
b30ab791 | 3283 | if (check_offsets_and_sizes(device, bdev)) |
c04ccaa6 LE |
3284 | goto err; |
3285 | ||
b411b363 | 3286 | if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) { |
d0180171 | 3287 | drbd_err(device, "unexpected bm_offset: %d (expected %d)\n", |
b411b363 | 3288 | be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset); |
b411b363 PR |
3289 | goto err; |
3290 | } | |
3291 | if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) { | |
d0180171 | 3292 | drbd_err(device, "unexpected md_size: %u (expected %u)\n", |
b411b363 | 3293 | be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect); |
b411b363 PR |
3294 | goto err; |
3295 | } | |
3296 | ||
3a4d4eb3 | 3297 | rv = NO_ERROR; |
b411b363 | 3298 | |
0500813f | 3299 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3300 | if (device->state.conn < C_CONNECTED) { |
db141b2f | 3301 | unsigned int peer; |
99432fcc | 3302 | peer = be32_to_cpu(buffer->la_peer_max_bio_size); |
db141b2f | 3303 | peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE); |
b30ab791 | 3304 | device->peer_max_bio_size = peer; |
99432fcc | 3305 | } |
0500813f | 3306 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
3307 | |
3308 | err: | |
b30ab791 | 3309 | drbd_md_put_buffer(device); |
b411b363 PR |
3310 | |
3311 | return rv; | |
3312 | } | |
3313 | ||
3314 | /** | |
3315 | * drbd_md_mark_dirty() - Mark meta data super block as dirty | |
b30ab791 | 3316 | * @device: DRBD device. |
b411b363 PR |
3317 | * |
3318 | * Call this function if you change anything that should be written to | |
3319 | * the meta-data super block. This function sets MD_DIRTY, and starts a | |
3320 | * timer that ensures that within five seconds you have to call drbd_md_sync(). | |
3321 | */ | |
ca0e6098 | 3322 | #ifdef DEBUG |
b30ab791 | 3323 | void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func) |
ee15b038 | 3324 | { |
b30ab791 AG |
3325 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) { |
3326 | mod_timer(&device->md_sync_timer, jiffies + HZ); | |
3327 | device->last_md_mark_dirty.line = line; | |
3328 | device->last_md_mark_dirty.func = func; | |
ee15b038 LE |
3329 | } |
3330 | } | |
3331 | #else | |
b30ab791 | 3332 | void drbd_md_mark_dirty(struct drbd_device *device) |
b411b363 | 3333 | { |
b30ab791 AG |
3334 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) |
3335 | mod_timer(&device->md_sync_timer, jiffies + 5*HZ); | |
b411b363 | 3336 | } |
ee15b038 | 3337 | #endif |
b411b363 | 3338 | |
b30ab791 | 3339 | void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3340 | { |
3341 | int i; | |
3342 | ||
62b0da3a | 3343 | for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) |
b30ab791 | 3344 | device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i]; |
b411b363 PR |
3345 | } |
3346 | ||
b30ab791 | 3347 | void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 PR |
3348 | { |
3349 | if (idx == UI_CURRENT) { | |
b30ab791 | 3350 | if (device->state.role == R_PRIMARY) |
b411b363 PR |
3351 | val |= 1; |
3352 | else | |
3353 | val &= ~((u64)1); | |
3354 | ||
b30ab791 | 3355 | drbd_set_ed_uuid(device, val); |
b411b363 PR |
3356 | } |
3357 | ||
b30ab791 AG |
3358 | device->ldev->md.uuid[idx] = val; |
3359 | drbd_md_mark_dirty(device); | |
b411b363 PR |
3360 | } |
3361 | ||
b30ab791 | 3362 | void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
9f2247bb PR |
3363 | { |
3364 | unsigned long flags; | |
b30ab791 AG |
3365 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3366 | __drbd_uuid_set(device, idx, val); | |
3367 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
9f2247bb | 3368 | } |
b411b363 | 3369 | |
b30ab791 | 3370 | void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 | 3371 | { |
9f2247bb | 3372 | unsigned long flags; |
b30ab791 AG |
3373 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3374 | if (device->ldev->md.uuid[idx]) { | |
3375 | drbd_uuid_move_history(device); | |
3376 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx]; | |
b411b363 | 3377 | } |
b30ab791 AG |
3378 | __drbd_uuid_set(device, idx, val); |
3379 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
b411b363 PR |
3380 | } |
3381 | ||
3382 | /** | |
3383 | * drbd_uuid_new_current() - Creates a new current UUID | |
b30ab791 | 3384 | * @device: DRBD device. |
b411b363 PR |
3385 | * |
3386 | * Creates a new current UUID, and rotates the old current UUID into | |
3387 | * the bitmap slot. Causes an incremental resync upon next connect. | |
3388 | */ | |
b30ab791 | 3389 | void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3390 | { |
3391 | u64 val; | |
9f2247bb PR |
3392 | unsigned long long bm_uuid; |
3393 | ||
3394 | get_random_bytes(&val, sizeof(u64)); | |
3395 | ||
b30ab791 AG |
3396 | spin_lock_irq(&device->ldev->md.uuid_lock); |
3397 | bm_uuid = device->ldev->md.uuid[UI_BITMAP]; | |
62b0da3a LE |
3398 | |
3399 | if (bm_uuid) | |
d0180171 | 3400 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3401 | |
b30ab791 AG |
3402 | device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT]; |
3403 | __drbd_uuid_set(device, UI_CURRENT, val); | |
3404 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 3405 | |
b30ab791 | 3406 | drbd_print_uuids(device, "new current UUID"); |
aaa8e2b3 | 3407 | /* get it to stable storage _now_ */ |
b30ab791 | 3408 | drbd_md_sync(device); |
b411b363 PR |
3409 | } |
3410 | ||
b30ab791 | 3411 | void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local) |
b411b363 | 3412 | { |
9f2247bb | 3413 | unsigned long flags; |
b30ab791 | 3414 | if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0) |
b411b363 PR |
3415 | return; |
3416 | ||
b30ab791 | 3417 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
b411b363 | 3418 | if (val == 0) { |
b30ab791 AG |
3419 | drbd_uuid_move_history(device); |
3420 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; | |
3421 | device->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 3422 | } else { |
b30ab791 | 3423 | unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP]; |
62b0da3a | 3424 | if (bm_uuid) |
d0180171 | 3425 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3426 | |
b30ab791 | 3427 | device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1); |
b411b363 | 3428 | } |
b30ab791 | 3429 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); |
9f2247bb | 3430 | |
b30ab791 | 3431 | drbd_md_mark_dirty(device); |
b411b363 PR |
3432 | } |
3433 | ||
3434 | /** | |
3435 | * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3436 | * @device: DRBD device. |
b411b363 PR |
3437 | * |
3438 | * Sets all bits in the bitmap and writes the whole bitmap to stable storage. | |
3439 | */ | |
b30ab791 | 3440 | int drbd_bmio_set_n_write(struct drbd_device *device) |
b411b363 PR |
3441 | { |
3442 | int rv = -EIO; | |
3443 | ||
b30ab791 AG |
3444 | if (get_ldev_if_state(device, D_ATTACHING)) { |
3445 | drbd_md_set_flag(device, MDF_FULL_SYNC); | |
3446 | drbd_md_sync(device); | |
3447 | drbd_bm_set_all(device); | |
b411b363 | 3448 | |
b30ab791 | 3449 | rv = drbd_bm_write(device); |
b411b363 PR |
3450 | |
3451 | if (!rv) { | |
b30ab791 AG |
3452 | drbd_md_clear_flag(device, MDF_FULL_SYNC); |
3453 | drbd_md_sync(device); | |
b411b363 PR |
3454 | } |
3455 | ||
b30ab791 | 3456 | put_ldev(device); |
b411b363 PR |
3457 | } |
3458 | ||
3459 | return rv; | |
3460 | } | |
3461 | ||
3462 | /** | |
3463 | * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3464 | * @device: DRBD device. |
b411b363 PR |
3465 | * |
3466 | * Clears all bits in the bitmap and writes the whole bitmap to stable storage. | |
3467 | */ | |
b30ab791 | 3468 | int drbd_bmio_clear_n_write(struct drbd_device *device) |
b411b363 PR |
3469 | { |
3470 | int rv = -EIO; | |
3471 | ||
b30ab791 AG |
3472 | drbd_resume_al(device); |
3473 | if (get_ldev_if_state(device, D_ATTACHING)) { | |
3474 | drbd_bm_clear_all(device); | |
3475 | rv = drbd_bm_write(device); | |
3476 | put_ldev(device); | |
b411b363 PR |
3477 | } |
3478 | ||
3479 | return rv; | |
3480 | } | |
3481 | ||
99920dc5 | 3482 | static int w_bitmap_io(struct drbd_work *w, int unused) |
b411b363 PR |
3483 | { |
3484 | struct bm_io_work *work = container_of(w, struct bm_io_work, w); | |
b30ab791 | 3485 | struct drbd_device *device = w->device; |
02851e9f | 3486 | int rv = -EIO; |
b411b363 | 3487 | |
0b0ba1ef | 3488 | D_ASSERT(device, atomic_read(&device->ap_bio_cnt) == 0); |
b411b363 | 3489 | |
b30ab791 AG |
3490 | if (get_ldev(device)) { |
3491 | drbd_bm_lock(device, work->why, work->flags); | |
3492 | rv = work->io_fn(device); | |
3493 | drbd_bm_unlock(device); | |
3494 | put_ldev(device); | |
02851e9f | 3495 | } |
b411b363 | 3496 | |
b30ab791 AG |
3497 | clear_bit_unlock(BITMAP_IO, &device->flags); |
3498 | wake_up(&device->misc_wait); | |
b411b363 PR |
3499 | |
3500 | if (work->done) | |
b30ab791 | 3501 | work->done(device, rv); |
b411b363 | 3502 | |
b30ab791 | 3503 | clear_bit(BITMAP_IO_QUEUED, &device->flags); |
b411b363 | 3504 | work->why = NULL; |
20ceb2b2 | 3505 | work->flags = 0; |
b411b363 | 3506 | |
99920dc5 | 3507 | return 0; |
b411b363 PR |
3508 | } |
3509 | ||
b30ab791 | 3510 | void drbd_ldev_destroy(struct drbd_device *device) |
82f59cc6 | 3511 | { |
b30ab791 AG |
3512 | lc_destroy(device->resync); |
3513 | device->resync = NULL; | |
3514 | lc_destroy(device->act_log); | |
3515 | device->act_log = NULL; | |
82f59cc6 | 3516 | __no_warn(local, |
b30ab791 AG |
3517 | drbd_free_bc(device->ldev); |
3518 | device->ldev = NULL;); | |
82f59cc6 | 3519 | |
b30ab791 | 3520 | clear_bit(GO_DISKLESS, &device->flags); |
82f59cc6 LE |
3521 | } |
3522 | ||
99920dc5 | 3523 | static int w_go_diskless(struct drbd_work *w, int unused) |
e9e6f3ec | 3524 | { |
b30ab791 | 3525 | struct drbd_device *device = w->device; |
00d56944 | 3526 | |
0b0ba1ef | 3527 | D_ASSERT(device, device->state.disk == D_FAILED); |
9d282875 LE |
3528 | /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will |
3529 | * inc/dec it frequently. Once we are D_DISKLESS, no one will touch | |
82f59cc6 LE |
3530 | * the protected members anymore, though, so once put_ldev reaches zero |
3531 | * again, it will be safe to free them. */ | |
a2a3c74f LE |
3532 | |
3533 | /* Try to write changed bitmap pages, read errors may have just | |
3534 | * set some bits outside the area covered by the activity log. | |
3535 | * | |
3536 | * If we have an IO error during the bitmap writeout, | |
3537 | * we will want a full sync next time, just in case. | |
3538 | * (Do we want a specific meta data flag for this?) | |
3539 | * | |
3540 | * If that does not make it to stable storage either, | |
fd0017c1 PR |
3541 | * we cannot do anything about that anymore. |
3542 | * | |
3543 | * We still need to check if both bitmap and ldev are present, we may | |
3544 | * end up here after a failed attach, before ldev was even assigned. | |
3545 | */ | |
b30ab791 | 3546 | if (device->bitmap && device->ldev) { |
bb45185d PR |
3547 | /* An interrupted resync or similar is allowed to recounts bits |
3548 | * while we detach. | |
3549 | * Any modifications would not be expected anymore, though. | |
3550 | */ | |
b30ab791 | 3551 | if (drbd_bitmap_io_from_worker(device, drbd_bm_write, |
bb45185d | 3552 | "detach", BM_LOCKED_TEST_ALLOWED)) { |
b30ab791 AG |
3553 | if (test_bit(WAS_READ_ERROR, &device->flags)) { |
3554 | drbd_md_set_flag(device, MDF_FULL_SYNC); | |
3555 | drbd_md_sync(device); | |
a2a3c74f LE |
3556 | } |
3557 | } | |
3558 | } | |
3559 | ||
b30ab791 | 3560 | drbd_force_state(device, NS(disk, D_DISKLESS)); |
99920dc5 | 3561 | return 0; |
e9e6f3ec LE |
3562 | } |
3563 | ||
b411b363 PR |
3564 | /** |
3565 | * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap | |
b30ab791 | 3566 | * @device: DRBD device. |
b411b363 PR |
3567 | * @io_fn: IO callback to be called when bitmap IO is possible |
3568 | * @done: callback to be called after the bitmap IO was performed | |
3569 | * @why: Descriptive text of the reason for doing the IO | |
3570 | * | |
3571 | * While IO on the bitmap happens we freeze application IO thus we ensure | |
3572 | * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be | |
3573 | * called from worker context. It MUST NOT be used while a previous such | |
3574 | * work is still pending! | |
3575 | */ | |
b30ab791 | 3576 | void drbd_queue_bitmap_io(struct drbd_device *device, |
54761697 AG |
3577 | int (*io_fn)(struct drbd_device *), |
3578 | void (*done)(struct drbd_device *, int), | |
20ceb2b2 | 3579 | char *why, enum bm_flag flags) |
b411b363 | 3580 | { |
0b0ba1ef | 3581 | D_ASSERT(device, current == first_peer_device(device)->connection->worker.task); |
b411b363 | 3582 | |
0b0ba1ef AG |
3583 | D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags)); |
3584 | D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags)); | |
3585 | D_ASSERT(device, list_empty(&device->bm_io_work.w.list)); | |
b30ab791 | 3586 | if (device->bm_io_work.why) |
d0180171 | 3587 | drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n", |
b30ab791 | 3588 | why, device->bm_io_work.why); |
b411b363 | 3589 | |
b30ab791 AG |
3590 | device->bm_io_work.io_fn = io_fn; |
3591 | device->bm_io_work.done = done; | |
3592 | device->bm_io_work.why = why; | |
3593 | device->bm_io_work.flags = flags; | |
b411b363 | 3594 | |
0500813f | 3595 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 AG |
3596 | set_bit(BITMAP_IO, &device->flags); |
3597 | if (atomic_read(&device->ap_bio_cnt) == 0) { | |
3598 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags)) | |
a6b32bc3 | 3599 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, &device->bm_io_work.w); |
b411b363 | 3600 | } |
0500813f | 3601 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
3602 | } |
3603 | ||
3604 | /** | |
3605 | * drbd_bitmap_io() - Does an IO operation on the whole bitmap | |
b30ab791 | 3606 | * @device: DRBD device. |
b411b363 PR |
3607 | * @io_fn: IO callback to be called when bitmap IO is possible |
3608 | * @why: Descriptive text of the reason for doing the IO | |
3609 | * | |
3610 | * freezes application IO while that the actual IO operations runs. This | |
3611 | * functions MAY NOT be called from worker context. | |
3612 | */ | |
b30ab791 | 3613 | int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *), |
20ceb2b2 | 3614 | char *why, enum bm_flag flags) |
b411b363 PR |
3615 | { |
3616 | int rv; | |
3617 | ||
0b0ba1ef | 3618 | D_ASSERT(device, current != first_peer_device(device)->connection->worker.task); |
b411b363 | 3619 | |
20ceb2b2 | 3620 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
b30ab791 | 3621 | drbd_suspend_io(device); |
b411b363 | 3622 | |
b30ab791 AG |
3623 | drbd_bm_lock(device, why, flags); |
3624 | rv = io_fn(device); | |
3625 | drbd_bm_unlock(device); | |
b411b363 | 3626 | |
20ceb2b2 | 3627 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
b30ab791 | 3628 | drbd_resume_io(device); |
b411b363 PR |
3629 | |
3630 | return rv; | |
3631 | } | |
3632 | ||
b30ab791 | 3633 | void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3634 | { |
b30ab791 AG |
3635 | if ((device->ldev->md.flags & flag) != flag) { |
3636 | drbd_md_mark_dirty(device); | |
3637 | device->ldev->md.flags |= flag; | |
b411b363 PR |
3638 | } |
3639 | } | |
3640 | ||
b30ab791 | 3641 | void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3642 | { |
b30ab791 AG |
3643 | if ((device->ldev->md.flags & flag) != 0) { |
3644 | drbd_md_mark_dirty(device); | |
3645 | device->ldev->md.flags &= ~flag; | |
b411b363 PR |
3646 | } |
3647 | } | |
3648 | int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag) | |
3649 | { | |
3650 | return (bdev->md.flags & flag) != 0; | |
3651 | } | |
3652 | ||
3653 | static void md_sync_timer_fn(unsigned long data) | |
3654 | { | |
b30ab791 | 3655 | struct drbd_device *device = (struct drbd_device *) data; |
b411b363 | 3656 | |
b792b655 | 3657 | /* must not double-queue! */ |
b30ab791 | 3658 | if (list_empty(&device->md_sync_work.list)) |
a6b32bc3 | 3659 | drbd_queue_work_front(&first_peer_device(device)->connection->sender_work, &device->md_sync_work); |
b411b363 PR |
3660 | } |
3661 | ||
99920dc5 | 3662 | static int w_md_sync(struct drbd_work *w, int unused) |
b411b363 | 3663 | { |
b30ab791 | 3664 | struct drbd_device *device = w->device; |
00d56944 | 3665 | |
d0180171 | 3666 | drbd_warn(device, "md_sync_timer expired! Worker calls drbd_md_sync().\n"); |
ee15b038 | 3667 | #ifdef DEBUG |
d0180171 | 3668 | drbd_warn(device, "last md_mark_dirty: %s:%u\n", |
b30ab791 | 3669 | device->last_md_mark_dirty.func, device->last_md_mark_dirty.line); |
ee15b038 | 3670 | #endif |
b30ab791 | 3671 | drbd_md_sync(device); |
99920dc5 | 3672 | return 0; |
b411b363 PR |
3673 | } |
3674 | ||
d8763023 | 3675 | const char *cmdname(enum drbd_packet cmd) |
f2ad9063 AG |
3676 | { |
3677 | /* THINK may need to become several global tables | |
3678 | * when we want to support more than | |
3679 | * one PRO_VERSION */ | |
3680 | static const char *cmdnames[] = { | |
3681 | [P_DATA] = "Data", | |
3682 | [P_DATA_REPLY] = "DataReply", | |
3683 | [P_RS_DATA_REPLY] = "RSDataReply", | |
3684 | [P_BARRIER] = "Barrier", | |
3685 | [P_BITMAP] = "ReportBitMap", | |
3686 | [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", | |
3687 | [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", | |
3688 | [P_UNPLUG_REMOTE] = "UnplugRemote", | |
3689 | [P_DATA_REQUEST] = "DataRequest", | |
3690 | [P_RS_DATA_REQUEST] = "RSDataRequest", | |
3691 | [P_SYNC_PARAM] = "SyncParam", | |
3692 | [P_SYNC_PARAM89] = "SyncParam89", | |
3693 | [P_PROTOCOL] = "ReportProtocol", | |
3694 | [P_UUIDS] = "ReportUUIDs", | |
3695 | [P_SIZES] = "ReportSizes", | |
3696 | [P_STATE] = "ReportState", | |
3697 | [P_SYNC_UUID] = "ReportSyncUUID", | |
3698 | [P_AUTH_CHALLENGE] = "AuthChallenge", | |
3699 | [P_AUTH_RESPONSE] = "AuthResponse", | |
3700 | [P_PING] = "Ping", | |
3701 | [P_PING_ACK] = "PingAck", | |
3702 | [P_RECV_ACK] = "RecvAck", | |
3703 | [P_WRITE_ACK] = "WriteAck", | |
3704 | [P_RS_WRITE_ACK] = "RSWriteAck", | |
d4dabbe2 | 3705 | [P_SUPERSEDED] = "Superseded", |
f2ad9063 AG |
3706 | [P_NEG_ACK] = "NegAck", |
3707 | [P_NEG_DREPLY] = "NegDReply", | |
3708 | [P_NEG_RS_DREPLY] = "NegRSDReply", | |
3709 | [P_BARRIER_ACK] = "BarrierAck", | |
3710 | [P_STATE_CHG_REQ] = "StateChgRequest", | |
3711 | [P_STATE_CHG_REPLY] = "StateChgReply", | |
3712 | [P_OV_REQUEST] = "OVRequest", | |
3713 | [P_OV_REPLY] = "OVReply", | |
3714 | [P_OV_RESULT] = "OVResult", | |
3715 | [P_CSUM_RS_REQUEST] = "CsumRSRequest", | |
3716 | [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", | |
3717 | [P_COMPRESSED_BITMAP] = "CBitmap", | |
3718 | [P_DELAY_PROBE] = "DelayProbe", | |
3719 | [P_OUT_OF_SYNC] = "OutOfSync", | |
7be8da07 | 3720 | [P_RETRY_WRITE] = "RetryWrite", |
ae25b336 LE |
3721 | [P_RS_CANCEL] = "RSCancel", |
3722 | [P_CONN_ST_CHG_REQ] = "conn_st_chg_req", | |
3723 | [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply", | |
036b17ea PR |
3724 | [P_RETRY_WRITE] = "retry_write", |
3725 | [P_PROTOCOL_UPDATE] = "protocol_update", | |
ae25b336 LE |
3726 | |
3727 | /* enum drbd_packet, but not commands - obsoleted flags: | |
3728 | * P_MAY_IGNORE | |
3729 | * P_MAX_OPT_CMD | |
3730 | */ | |
f2ad9063 AG |
3731 | }; |
3732 | ||
ae25b336 | 3733 | /* too big for the array: 0xfffX */ |
e5d6f33a AG |
3734 | if (cmd == P_INITIAL_META) |
3735 | return "InitialMeta"; | |
3736 | if (cmd == P_INITIAL_DATA) | |
3737 | return "InitialData"; | |
6038178e AG |
3738 | if (cmd == P_CONNECTION_FEATURES) |
3739 | return "ConnectionFeatures"; | |
6e849ce8 | 3740 | if (cmd >= ARRAY_SIZE(cmdnames)) |
f2ad9063 AG |
3741 | return "Unknown"; |
3742 | return cmdnames[cmd]; | |
3743 | } | |
3744 | ||
7be8da07 AG |
3745 | /** |
3746 | * drbd_wait_misc - wait for a request to make progress | |
b30ab791 | 3747 | * @device: device associated with the request |
7be8da07 AG |
3748 | * @i: the struct drbd_interval embedded in struct drbd_request or |
3749 | * struct drbd_peer_request | |
3750 | */ | |
b30ab791 | 3751 | int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i) |
7be8da07 | 3752 | { |
44ed167d | 3753 | struct net_conf *nc; |
7be8da07 AG |
3754 | DEFINE_WAIT(wait); |
3755 | long timeout; | |
3756 | ||
44ed167d | 3757 | rcu_read_lock(); |
a6b32bc3 | 3758 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
44ed167d PR |
3759 | if (!nc) { |
3760 | rcu_read_unlock(); | |
7be8da07 | 3761 | return -ETIMEDOUT; |
44ed167d PR |
3762 | } |
3763 | timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT; | |
3764 | rcu_read_unlock(); | |
7be8da07 | 3765 | |
b30ab791 | 3766 | /* Indicate to wake up device->misc_wait on progress. */ |
7be8da07 | 3767 | i->waiting = true; |
b30ab791 | 3768 | prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE); |
0500813f | 3769 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 3770 | timeout = schedule_timeout(timeout); |
b30ab791 | 3771 | finish_wait(&device->misc_wait, &wait); |
0500813f | 3772 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3773 | if (!timeout || device->state.conn < C_CONNECTED) |
7be8da07 AG |
3774 | return -ETIMEDOUT; |
3775 | if (signal_pending(current)) | |
3776 | return -ERESTARTSYS; | |
3777 | return 0; | |
b411b363 PR |
3778 | } |
3779 | ||
3780 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
3781 | /* Fault insertion support including random number generator shamelessly | |
3782 | * stolen from kernel/rcutorture.c */ | |
3783 | struct fault_random_state { | |
3784 | unsigned long state; | |
3785 | unsigned long count; | |
3786 | }; | |
3787 | ||
3788 | #define FAULT_RANDOM_MULT 39916801 /* prime */ | |
3789 | #define FAULT_RANDOM_ADD 479001701 /* prime */ | |
3790 | #define FAULT_RANDOM_REFRESH 10000 | |
3791 | ||
3792 | /* | |
3793 | * Crude but fast random-number generator. Uses a linear congruential | |
3794 | * generator, with occasional help from get_random_bytes(). | |
3795 | */ | |
3796 | static unsigned long | |
3797 | _drbd_fault_random(struct fault_random_state *rsp) | |
3798 | { | |
3799 | long refresh; | |
3800 | ||
49829ea7 | 3801 | if (!rsp->count--) { |
b411b363 PR |
3802 | get_random_bytes(&refresh, sizeof(refresh)); |
3803 | rsp->state += refresh; | |
3804 | rsp->count = FAULT_RANDOM_REFRESH; | |
3805 | } | |
3806 | rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD; | |
3807 | return swahw32(rsp->state); | |
3808 | } | |
3809 | ||
3810 | static char * | |
3811 | _drbd_fault_str(unsigned int type) { | |
3812 | static char *_faults[] = { | |
3813 | [DRBD_FAULT_MD_WR] = "Meta-data write", | |
3814 | [DRBD_FAULT_MD_RD] = "Meta-data read", | |
3815 | [DRBD_FAULT_RS_WR] = "Resync write", | |
3816 | [DRBD_FAULT_RS_RD] = "Resync read", | |
3817 | [DRBD_FAULT_DT_WR] = "Data write", | |
3818 | [DRBD_FAULT_DT_RD] = "Data read", | |
3819 | [DRBD_FAULT_DT_RA] = "Data read ahead", | |
3820 | [DRBD_FAULT_BM_ALLOC] = "BM allocation", | |
6b4388ac PR |
3821 | [DRBD_FAULT_AL_EE] = "EE allocation", |
3822 | [DRBD_FAULT_RECEIVE] = "receive data corruption", | |
b411b363 PR |
3823 | }; |
3824 | ||
3825 | return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**"; | |
3826 | } | |
3827 | ||
3828 | unsigned int | |
b30ab791 | 3829 | _drbd_insert_fault(struct drbd_device *device, unsigned int type) |
b411b363 PR |
3830 | { |
3831 | static struct fault_random_state rrs = {0, 0}; | |
3832 | ||
3833 | unsigned int ret = ( | |
3834 | (fault_devs == 0 || | |
b30ab791 | 3835 | ((1 << device_to_minor(device)) & fault_devs) != 0) && |
b411b363 PR |
3836 | (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate)); |
3837 | ||
3838 | if (ret) { | |
3839 | fault_count++; | |
3840 | ||
7383506c | 3841 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 3842 | drbd_warn(device, "***Simulating %s failure\n", |
b411b363 PR |
3843 | _drbd_fault_str(type)); |
3844 | } | |
3845 | ||
3846 | return ret; | |
3847 | } | |
3848 | #endif | |
3849 | ||
3850 | const char *drbd_buildtag(void) | |
3851 | { | |
3852 | /* DRBD built from external sources has here a reference to the | |
3853 | git hash of the source code. */ | |
3854 | ||
3855 | static char buildtag[38] = "\0uilt-in"; | |
3856 | ||
3857 | if (buildtag[0] == 0) { | |
bc4854bc CW |
3858 | #ifdef MODULE |
3859 | sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion); | |
3860 | #else | |
3861 | buildtag[0] = 'b'; | |
b411b363 | 3862 | #endif |
b411b363 PR |
3863 | } |
3864 | ||
3865 | return buildtag; | |
3866 | } | |
3867 | ||
3868 | module_init(drbd_init) | |
3869 | module_exit(drbd_cleanup) | |
3870 | ||
b411b363 PR |
3871 | EXPORT_SYMBOL(drbd_conn_str); |
3872 | EXPORT_SYMBOL(drbd_role_str); | |
3873 | EXPORT_SYMBOL(drbd_disk_str); | |
3874 | EXPORT_SYMBOL(drbd_set_st_err_str); |