Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev | |
11 | from Logicworks, Inc. for making SDP replication support possible. | |
12 | ||
13 | drbd is free software; you can redistribute it and/or modify | |
14 | it under the terms of the GNU General Public License as published by | |
15 | the Free Software Foundation; either version 2, or (at your option) | |
16 | any later version. | |
17 | ||
18 | drbd is distributed in the hope that it will be useful, | |
19 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | GNU General Public License for more details. | |
22 | ||
23 | You should have received a copy of the GNU General Public License | |
24 | along with drbd; see the file COPYING. If not, write to | |
25 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
26 | ||
27 | */ | |
28 | ||
b411b363 | 29 | #include <linux/module.h> |
b411b363 PR |
30 | #include <linux/drbd.h> |
31 | #include <asm/uaccess.h> | |
32 | #include <asm/types.h> | |
33 | #include <net/sock.h> | |
34 | #include <linux/ctype.h> | |
2a48fc0a | 35 | #include <linux/mutex.h> |
b411b363 PR |
36 | #include <linux/fs.h> |
37 | #include <linux/file.h> | |
38 | #include <linux/proc_fs.h> | |
39 | #include <linux/init.h> | |
40 | #include <linux/mm.h> | |
41 | #include <linux/memcontrol.h> | |
42 | #include <linux/mm_inline.h> | |
43 | #include <linux/slab.h> | |
44 | #include <linux/random.h> | |
45 | #include <linux/reboot.h> | |
46 | #include <linux/notifier.h> | |
47 | #include <linux/kthread.h> | |
113fef9e | 48 | #include <linux/workqueue.h> |
b411b363 PR |
49 | #define __KERNEL_SYSCALLS__ |
50 | #include <linux/unistd.h> | |
51 | #include <linux/vmalloc.h> | |
52 | ||
53 | #include <linux/drbd_limits.h> | |
54 | #include "drbd_int.h" | |
a3603a6e | 55 | #include "drbd_protocol.h" |
b411b363 PR |
56 | #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */ |
57 | ||
58 | #include "drbd_vli.h" | |
59 | ||
2a48fc0a | 60 | static DEFINE_MUTEX(drbd_main_mutex); |
b411b363 | 61 | int drbd_worker(struct drbd_thread *); |
b411b363 PR |
62 | |
63 | int drbd_init(void); | |
64 | static int drbd_open(struct block_device *bdev, fmode_t mode); | |
db2a144b | 65 | static void drbd_release(struct gendisk *gd, fmode_t mode); |
99920dc5 | 66 | static int w_md_sync(struct drbd_work *w, int unused); |
b411b363 | 67 | static void md_sync_timer_fn(unsigned long data); |
99920dc5 AG |
68 | static int w_bitmap_io(struct drbd_work *w, int unused); |
69 | static int w_go_diskless(struct drbd_work *w, int unused); | |
b411b363 | 70 | |
b411b363 PR |
71 | MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, " |
72 | "Lars Ellenberg <lars@linbit.com>"); | |
73 | MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION); | |
74 | MODULE_VERSION(REL_VERSION); | |
75 | MODULE_LICENSE("GPL"); | |
81a5d60e | 76 | MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices (" |
2b8a90b5 | 77 | __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")"); |
b411b363 PR |
78 | MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR); |
79 | ||
80 | #include <linux/moduleparam.h> | |
81 | /* allow_open_on_secondary */ | |
82 | MODULE_PARM_DESC(allow_oos, "DONT USE!"); | |
83 | /* thanks to these macros, if compiled into the kernel (not-module), | |
84 | * this becomes the boot parameter drbd.minor_count */ | |
85 | module_param(minor_count, uint, 0444); | |
86 | module_param(disable_sendpage, bool, 0644); | |
87 | module_param(allow_oos, bool, 0); | |
b411b363 PR |
88 | module_param(proc_details, int, 0644); |
89 | ||
90 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
91 | int enable_faults; | |
92 | int fault_rate; | |
93 | static int fault_count; | |
94 | int fault_devs; | |
95 | /* bitmap of enabled faults */ | |
96 | module_param(enable_faults, int, 0664); | |
97 | /* fault rate % value - applies to all enabled faults */ | |
98 | module_param(fault_rate, int, 0664); | |
99 | /* count of faults inserted */ | |
100 | module_param(fault_count, int, 0664); | |
101 | /* bitmap of devices to insert faults on */ | |
102 | module_param(fault_devs, int, 0644); | |
103 | #endif | |
104 | ||
105 | /* module parameter, defined */ | |
2b8a90b5 | 106 | unsigned int minor_count = DRBD_MINOR_COUNT_DEF; |
90ab5ee9 RR |
107 | bool disable_sendpage; |
108 | bool allow_oos; | |
b411b363 PR |
109 | int proc_details; /* Detail level in proc drbd*/ |
110 | ||
111 | /* Module parameter for setting the user mode helper program | |
112 | * to run. Default is /sbin/drbdadm */ | |
113 | char usermode_helper[80] = "/sbin/drbdadm"; | |
114 | ||
115 | module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644); | |
116 | ||
117 | /* in 2.6.x, our device mapping and config info contains our virtual gendisks | |
118 | * as member "struct gendisk *vdisk;" | |
119 | */ | |
05a10ec7 | 120 | struct idr drbd_devices; |
77c556f6 | 121 | struct list_head drbd_resources; |
b411b363 PR |
122 | |
123 | struct kmem_cache *drbd_request_cache; | |
6c852bec | 124 | struct kmem_cache *drbd_ee_cache; /* peer requests */ |
b411b363 PR |
125 | struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ |
126 | struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
127 | mempool_t *drbd_request_mempool; | |
128 | mempool_t *drbd_ee_mempool; | |
4281808f | 129 | mempool_t *drbd_md_io_page_pool; |
9476f39d | 130 | struct bio_set *drbd_md_io_bio_set; |
b411b363 PR |
131 | |
132 | /* I do not use a standard mempool, because: | |
133 | 1) I want to hand out the pre-allocated objects first. | |
134 | 2) I want to be able to interrupt sleeping allocation with a signal. | |
135 | Note: This is a single linked list, the next pointer is the private | |
136 | member of struct page. | |
137 | */ | |
138 | struct page *drbd_pp_pool; | |
139 | spinlock_t drbd_pp_lock; | |
140 | int drbd_pp_vacant; | |
141 | wait_queue_head_t drbd_pp_wait; | |
142 | ||
143 | DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5); | |
144 | ||
7d4e9d09 | 145 | static const struct block_device_operations drbd_ops = { |
b411b363 PR |
146 | .owner = THIS_MODULE, |
147 | .open = drbd_open, | |
148 | .release = drbd_release, | |
149 | }; | |
19f843aa | 150 | |
da4a75d2 LE |
151 | struct bio *bio_alloc_drbd(gfp_t gfp_mask) |
152 | { | |
153 | struct bio *bio; | |
19f843aa | 154 | |
da4a75d2 LE |
155 | if (!drbd_md_io_bio_set) |
156 | return bio_alloc(gfp_mask, 1); | |
19f843aa | 157 | |
da4a75d2 LE |
158 | bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set); |
159 | if (!bio) | |
160 | return NULL; | |
da4a75d2 | 161 | return bio; |
19f843aa LE |
162 | } |
163 | ||
b411b363 PR |
164 | #ifdef __CHECKER__ |
165 | /* When checking with sparse, and this is an inline function, sparse will | |
166 | give tons of false positives. When this is a real functions sparse works. | |
b411b363 | 167 | */ |
b30ab791 | 168 | int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins) |
b411b363 | 169 | { |
b411b363 | 170 | int io_allowed; |
b411b363 | 171 | |
b30ab791 AG |
172 | atomic_inc(&device->local_cnt); |
173 | io_allowed = (device->state.disk >= mins); | |
b411b363 | 174 | if (!io_allowed) { |
b30ab791 AG |
175 | if (atomic_dec_and_test(&device->local_cnt)) |
176 | wake_up(&device->misc_wait); | |
b411b363 | 177 | } |
b411b363 PR |
178 | return io_allowed; |
179 | } | |
b411b363 | 180 | |
b411b363 | 181 | #endif |
265be2d0 | 182 | |
b411b363 | 183 | /** |
b6dd1a89 | 184 | * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch |
bde89a9e | 185 | * @connection: DRBD connection. |
b411b363 PR |
186 | * @barrier_nr: Expected identifier of the DRBD write barrier packet. |
187 | * @set_size: Expected number of requests before that barrier. | |
188 | * | |
189 | * In case the passed barrier_nr or set_size does not match the oldest | |
b6dd1a89 LE |
190 | * epoch of not yet barrier-acked requests, this function will cause a |
191 | * termination of the connection. | |
b411b363 | 192 | */ |
bde89a9e | 193 | void tl_release(struct drbd_connection *connection, unsigned int barrier_nr, |
2f5cdd0b | 194 | unsigned int set_size) |
b411b363 | 195 | { |
b411b363 | 196 | struct drbd_request *r; |
b6dd1a89 LE |
197 | struct drbd_request *req = NULL; |
198 | int expect_epoch = 0; | |
199 | int expect_size = 0; | |
b411b363 | 200 | |
bde89a9e | 201 | spin_lock_irq(&connection->req_lock); |
b411b363 | 202 | |
98683650 | 203 | /* find oldest not yet barrier-acked write request, |
b6dd1a89 | 204 | * count writes in its epoch. */ |
bde89a9e | 205 | list_for_each_entry(r, &connection->transfer_log, tl_requests) { |
a0d856df | 206 | const unsigned s = r->rq_state; |
b6dd1a89 LE |
207 | if (!req) { |
208 | if (!(s & RQ_WRITE)) | |
209 | continue; | |
210 | if (!(s & RQ_NET_MASK)) | |
211 | continue; | |
212 | if (s & RQ_NET_DONE) | |
213 | continue; | |
214 | req = r; | |
215 | expect_epoch = req->epoch; | |
216 | expect_size ++; | |
217 | } else { | |
218 | if (r->epoch != expect_epoch) | |
219 | break; | |
220 | if (!(s & RQ_WRITE)) | |
221 | continue; | |
222 | /* if (s & RQ_DONE): not expected */ | |
223 | /* if (!(s & RQ_NET_MASK)): not expected */ | |
224 | expect_size++; | |
43a5182c | 225 | } |
b411b363 | 226 | } |
67098930 | 227 | |
b411b363 | 228 | /* first some paranoia code */ |
b6dd1a89 | 229 | if (req == NULL) { |
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 | } |
bde89a9e | 258 | spin_unlock_irq(&connection->req_lock); |
19f843aa | 259 | |
b411b363 | 260 | return; |
b411b363 | 261 | |
b411b363 | 262 | bail: |
bde89a9e AG |
263 | spin_unlock_irq(&connection->req_lock); |
264 | conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
b411b363 | 265 | } |
b411b363 | 266 | |
b411b363 | 267 | |
b411b363 | 268 | /** |
11b58e73 | 269 | * _tl_restart() - Walks the transfer log, and applies an action to all requests |
b30ab791 | 270 | * @device: DRBD device. |
11b58e73 | 271 | * @what: The action/event to perform with all request objects |
b411b363 | 272 | * |
8554df1c AG |
273 | * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO, |
274 | * RESTART_FROZEN_DISK_IO. | |
b411b363 | 275 | */ |
b6dd1a89 | 276 | /* must hold resource->req_lock */ |
bde89a9e | 277 | void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b411b363 | 278 | { |
b6dd1a89 | 279 | struct drbd_request *req, *r; |
b411b363 | 280 | |
bde89a9e | 281 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) |
b6dd1a89 LE |
282 | _req_mod(req, what); |
283 | } | |
738a84b2 | 284 | |
bde89a9e | 285 | void tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b6dd1a89 | 286 | { |
bde89a9e AG |
287 | spin_lock_irq(&connection->req_lock); |
288 | _tl_restart(connection, what); | |
289 | spin_unlock_irq(&connection->req_lock); | |
cdfda633 | 290 | } |
b411b363 | 291 | |
b411b363 PR |
292 | /** |
293 | * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL | |
b30ab791 | 294 | * @device: DRBD device. |
b411b363 PR |
295 | * |
296 | * This is called after the connection to the peer was lost. The storage covered | |
297 | * by the requests on the transfer gets marked as our of sync. Called from the | |
298 | * receiver thread and the worker thread. | |
299 | */ | |
bde89a9e | 300 | void tl_clear(struct drbd_connection *connection) |
b411b363 | 301 | { |
bde89a9e | 302 | tl_restart(connection, CONNECTION_LOST_WHILE_PENDING); |
b411b363 | 303 | } |
197296ff | 304 | |
cdfda633 | 305 | /** |
b30ab791 AG |
306 | * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL |
307 | * @device: DRBD device. | |
cdfda633 | 308 | */ |
b30ab791 | 309 | void tl_abort_disk_io(struct drbd_device *device) |
cdfda633 | 310 | { |
a6b32bc3 | 311 | struct drbd_connection *connection = first_peer_device(device)->connection; |
b6dd1a89 | 312 | struct drbd_request *req, *r; |
02851e9f | 313 | |
bde89a9e AG |
314 | spin_lock_irq(&connection->req_lock); |
315 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) { | |
97ddb687 LE |
316 | if (!(req->rq_state & RQ_LOCAL_PENDING)) |
317 | continue; | |
b30ab791 | 318 | if (req->w.device != device) |
b6dd1a89 LE |
319 | continue; |
320 | _req_mod(req, ABORT_DISK_IO); | |
b411b363 | 321 | } |
bde89a9e | 322 | spin_unlock_irq(&connection->req_lock); |
b411b363 PR |
323 | } |
324 | ||
b411b363 PR |
325 | static int drbd_thread_setup(void *arg) |
326 | { | |
327 | struct drbd_thread *thi = (struct drbd_thread *) arg; | |
bde89a9e | 328 | struct drbd_connection *connection = thi->connection; |
b411b363 PR |
329 | unsigned long flags; |
330 | int retval; | |
331 | ||
f1b3a6ec | 332 | snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s", |
77c556f6 AG |
333 | thi->name[0], |
334 | thi->connection->resource->name); | |
f1b3a6ec | 335 | |
b411b363 PR |
336 | restart: |
337 | retval = thi->function(thi); | |
338 | ||
339 | spin_lock_irqsave(&thi->t_lock, flags); | |
340 | ||
e77a0a5c | 341 | /* if the receiver has been "EXITING", the last thing it did |
b411b363 PR |
342 | * was set the conn state to "StandAlone", |
343 | * if now a re-connect request comes in, conn state goes C_UNCONNECTED, | |
344 | * and receiver thread will be "started". | |
e77a0a5c | 345 | * drbd_thread_start needs to set "RESTARTING" in that case. |
b411b363 | 346 | * t_state check and assignment needs to be within the same spinlock, |
e77a0a5c AG |
347 | * so either thread_start sees EXITING, and can remap to RESTARTING, |
348 | * or thread_start see NONE, and can proceed as normal. | |
b411b363 PR |
349 | */ |
350 | ||
e77a0a5c | 351 | if (thi->t_state == RESTARTING) { |
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 | |
b30ab791 | 514 | * @device: DRBD device. |
b411b363 PR |
515 | * |
516 | * Forces all threads of a device onto the same CPU. This is beneficial for | |
517 | * DRBD's performance. May be overwritten by user's configuration. | |
518 | */ | |
bde89a9e | 519 | void drbd_calc_cpu_mask(struct drbd_connection *connection) |
b411b363 PR |
520 | { |
521 | int ord, cpu; | |
522 | ||
523 | /* user override. */ | |
bde89a9e | 524 | if (cpumask_weight(connection->cpu_mask)) |
b411b363 PR |
525 | return; |
526 | ||
bde89a9e | 527 | ord = conn_lowest_minor(connection) % cpumask_weight(cpu_online_mask); |
b411b363 PR |
528 | for_each_online_cpu(cpu) { |
529 | if (ord-- == 0) { | |
bde89a9e | 530 | cpumask_set_cpu(cpu, connection->cpu_mask); |
b411b363 PR |
531 | return; |
532 | } | |
533 | } | |
534 | /* should not be reached */ | |
bde89a9e | 535 | cpumask_setall(connection->cpu_mask); |
b411b363 PR |
536 | } |
537 | ||
538 | /** | |
539 | * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread | |
b30ab791 | 540 | * @device: DRBD device. |
bc31fe33 | 541 | * @thi: drbd_thread object |
b411b363 PR |
542 | * |
543 | * call in the "main loop" of _all_ threads, no need for any mutex, current won't die | |
544 | * prematurely. | |
545 | */ | |
80822284 | 546 | void drbd_thread_current_set_cpu(struct drbd_thread *thi) |
b411b363 PR |
547 | { |
548 | struct task_struct *p = current; | |
bed879ae | 549 | |
b411b363 PR |
550 | if (!thi->reset_cpu_mask) |
551 | return; | |
552 | thi->reset_cpu_mask = 0; | |
bde89a9e | 553 | set_cpus_allowed_ptr(p, thi->connection->cpu_mask); |
b411b363 PR |
554 | } |
555 | #endif | |
556 | ||
52b061a4 AG |
557 | /** |
558 | * drbd_header_size - size of a packet header | |
559 | * | |
560 | * The header size is a multiple of 8, so any payload following the header is | |
561 | * word aligned on 64-bit architectures. (The bitmap send and receive code | |
562 | * relies on this.) | |
563 | */ | |
bde89a9e | 564 | unsigned int drbd_header_size(struct drbd_connection *connection) |
b411b363 | 565 | { |
bde89a9e | 566 | if (connection->agreed_pro_version >= 100) { |
0c8e36d9 AG |
567 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8)); |
568 | return sizeof(struct p_header100); | |
569 | } else { | |
570 | BUILD_BUG_ON(sizeof(struct p_header80) != | |
571 | sizeof(struct p_header95)); | |
572 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8)); | |
573 | return sizeof(struct p_header80); | |
574 | } | |
52b061a4 | 575 | } |
b411b363 | 576 | |
e658983a | 577 | static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
578 | { |
579 | h->magic = cpu_to_be32(DRBD_MAGIC); | |
580 | h->command = cpu_to_be16(cmd); | |
581 | h->length = cpu_to_be16(size); | |
e658983a | 582 | return sizeof(struct p_header80); |
fd340c12 | 583 | } |
b411b363 | 584 | |
e658983a | 585 | static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
586 | { |
587 | h->magic = cpu_to_be16(DRBD_MAGIC_BIG); | |
b411b363 | 588 | h->command = cpu_to_be16(cmd); |
b55d84ba | 589 | h->length = cpu_to_be32(size); |
e658983a | 590 | return sizeof(struct p_header95); |
fd340c12 | 591 | } |
b411b363 | 592 | |
0c8e36d9 AG |
593 | static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd, |
594 | int size, int vnr) | |
595 | { | |
596 | h->magic = cpu_to_be32(DRBD_MAGIC_100); | |
597 | h->volume = cpu_to_be16(vnr); | |
598 | h->command = cpu_to_be16(cmd); | |
599 | h->length = cpu_to_be32(size); | |
600 | h->pad = 0; | |
601 | return sizeof(struct p_header100); | |
602 | } | |
b411b363 | 603 | |
bde89a9e | 604 | static unsigned int prepare_header(struct drbd_connection *connection, int vnr, |
0c8e36d9 | 605 | void *buffer, enum drbd_packet cmd, int size) |
d38e787e | 606 | { |
bde89a9e | 607 | if (connection->agreed_pro_version >= 100) |
0c8e36d9 | 608 | return prepare_header100(buffer, cmd, size, vnr); |
bde89a9e | 609 | else if (connection->agreed_pro_version >= 95 && |
0c8e36d9 | 610 | size > DRBD_MAX_SIZE_H80_PACKET) |
e658983a | 611 | return prepare_header95(buffer, cmd, size); |
d38e787e | 612 | else |
e658983a | 613 | return prepare_header80(buffer, cmd, size); |
b411b363 PR |
614 | } |
615 | ||
bde89a9e | 616 | static void *__conn_prepare_command(struct drbd_connection *connection, |
a7eb7bdf | 617 | struct drbd_socket *sock) |
b411b363 | 618 | { |
a7eb7bdf AG |
619 | if (!sock->socket) |
620 | return NULL; | |
bde89a9e | 621 | return sock->sbuf + drbd_header_size(connection); |
a7eb7bdf | 622 | } |
b411b363 | 623 | |
bde89a9e | 624 | void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock) |
dba58587 | 625 | { |
a7eb7bdf | 626 | void *p; |
b411b363 | 627 | |
dba58587 | 628 | mutex_lock(&sock->mutex); |
bde89a9e | 629 | p = __conn_prepare_command(connection, sock); |
a7eb7bdf | 630 | if (!p) |
dba58587 | 631 | mutex_unlock(&sock->mutex); |
b411b363 | 632 | |
a7eb7bdf | 633 | return p; |
b411b363 PR |
634 | } |
635 | ||
b30ab791 | 636 | void *drbd_prepare_command(struct drbd_device *device, struct drbd_socket *sock) |
b411b363 | 637 | { |
a6b32bc3 | 638 | return conn_prepare_command(first_peer_device(device)->connection, sock); |
dba58587 | 639 | } |
b411b363 | 640 | |
bde89a9e | 641 | static int __send_command(struct drbd_connection *connection, int vnr, |
dba58587 AG |
642 | struct drbd_socket *sock, enum drbd_packet cmd, |
643 | unsigned int header_size, void *data, | |
644 | unsigned int size) | |
645 | { | |
646 | int msg_flags; | |
647 | int err; | |
b411b363 | 648 | |
dba58587 AG |
649 | /* |
650 | * Called with @data == NULL and the size of the data blocks in @size | |
651 | * for commands that send data blocks. For those commands, omit the | |
652 | * MSG_MORE flag: this will increase the likelihood that data blocks | |
653 | * which are page aligned on the sender will end up page aligned on the | |
654 | * receiver. | |
655 | */ | |
656 | msg_flags = data ? MSG_MORE : 0; | |
657 | ||
bde89a9e | 658 | header_size += prepare_header(connection, vnr, sock->sbuf, cmd, |
e658983a | 659 | header_size + size); |
bde89a9e | 660 | err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size, |
dba58587 AG |
661 | msg_flags); |
662 | if (data && !err) | |
bde89a9e | 663 | err = drbd_send_all(connection, sock->socket, data, size, 0); |
dba58587 AG |
664 | return err; |
665 | } | |
666 | ||
bde89a9e | 667 | static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
a7eb7bdf AG |
668 | enum drbd_packet cmd, unsigned int header_size, |
669 | void *data, unsigned int size) | |
670 | { | |
bde89a9e | 671 | return __send_command(connection, 0, sock, cmd, header_size, data, size); |
a7eb7bdf AG |
672 | } |
673 | ||
bde89a9e | 674 | int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
dba58587 AG |
675 | enum drbd_packet cmd, unsigned int header_size, |
676 | void *data, unsigned int size) | |
677 | { | |
678 | int err; | |
b411b363 | 679 | |
bde89a9e | 680 | err = __conn_send_command(connection, sock, cmd, header_size, data, size); |
dba58587 AG |
681 | mutex_unlock(&sock->mutex); |
682 | return err; | |
683 | } | |
684 | ||
b30ab791 | 685 | int drbd_send_command(struct drbd_device *device, struct drbd_socket *sock, |
dba58587 AG |
686 | enum drbd_packet cmd, unsigned int header_size, |
687 | void *data, unsigned int size) | |
688 | { | |
689 | int err; | |
690 | ||
a6b32bc3 | 691 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, cmd, header_size, |
dba58587 AG |
692 | data, size); |
693 | mutex_unlock(&sock->mutex); | |
694 | return err; | |
695 | } | |
b411b363 | 696 | |
bde89a9e | 697 | int drbd_send_ping(struct drbd_connection *connection) |
e307f352 | 698 | { |
9f5bdc33 AG |
699 | struct drbd_socket *sock; |
700 | ||
bde89a9e AG |
701 | sock = &connection->meta; |
702 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 703 | return -EIO; |
bde89a9e | 704 | return conn_send_command(connection, sock, P_PING, 0, NULL, 0); |
e307f352 | 705 | } |
b411b363 | 706 | |
bde89a9e | 707 | int drbd_send_ping_ack(struct drbd_connection *connection) |
e307f352 | 708 | { |
9f5bdc33 AG |
709 | struct drbd_socket *sock; |
710 | ||
bde89a9e AG |
711 | sock = &connection->meta; |
712 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 713 | return -EIO; |
bde89a9e | 714 | return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0); |
b411b363 PR |
715 | } |
716 | ||
b30ab791 | 717 | int drbd_send_sync_param(struct drbd_device *device) |
b411b363 | 718 | { |
7c96715a | 719 | struct drbd_socket *sock; |
8e26f9cc | 720 | struct p_rs_param_95 *p; |
9f5bdc33 | 721 | int size; |
a6b32bc3 | 722 | const int apv = first_peer_device(device)->connection->agreed_pro_version; |
9f5bdc33 | 723 | enum drbd_packet cmd; |
44ed167d | 724 | struct net_conf *nc; |
daeda1cc | 725 | struct disk_conf *dc; |
9f5bdc33 | 726 | |
a6b32bc3 | 727 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 728 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
729 | if (!p) |
730 | return -EIO; | |
b411b363 | 731 | |
44ed167d | 732 | rcu_read_lock(); |
a6b32bc3 | 733 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
b411b363 PR |
734 | |
735 | size = apv <= 87 ? sizeof(struct p_rs_param) | |
736 | : apv == 88 ? sizeof(struct p_rs_param) | |
44ed167d | 737 | + strlen(nc->verify_alg) + 1 |
8e26f9cc PR |
738 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
739 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 740 | |
9f5bdc33 | 741 | cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM; |
b411b363 | 742 | |
9f5bdc33 AG |
743 | /* initialize verify_alg and csums_alg */ |
744 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
b411b363 | 745 | |
b30ab791 AG |
746 | if (get_ldev(device)) { |
747 | dc = rcu_dereference(device->ldev->disk_conf); | |
6394b935 | 748 | p->resync_rate = cpu_to_be32(dc->resync_rate); |
daeda1cc PR |
749 | p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead); |
750 | p->c_delay_target = cpu_to_be32(dc->c_delay_target); | |
751 | p->c_fill_target = cpu_to_be32(dc->c_fill_target); | |
752 | p->c_max_rate = cpu_to_be32(dc->c_max_rate); | |
b30ab791 | 753 | put_ldev(device); |
9f5bdc33 | 754 | } else { |
6394b935 | 755 | p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF); |
9f5bdc33 AG |
756 | p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF); |
757 | p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF); | |
758 | p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF); | |
759 | p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF); | |
760 | } | |
b411b363 | 761 | |
9f5bdc33 | 762 | if (apv >= 88) |
44ed167d | 763 | strcpy(p->verify_alg, nc->verify_alg); |
9f5bdc33 | 764 | if (apv >= 89) |
44ed167d PR |
765 | strcpy(p->csums_alg, nc->csums_alg); |
766 | rcu_read_unlock(); | |
b411b363 | 767 | |
b30ab791 | 768 | return drbd_send_command(device, sock, cmd, size, NULL, 0); |
b411b363 PR |
769 | } |
770 | ||
bde89a9e | 771 | int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd) |
b411b363 | 772 | { |
9f5bdc33 | 773 | struct drbd_socket *sock; |
b411b363 | 774 | struct p_protocol *p; |
44ed167d | 775 | struct net_conf *nc; |
9f5bdc33 | 776 | int size, cf; |
b411b363 | 777 | |
bde89a9e AG |
778 | sock = &connection->data; |
779 | p = __conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
780 | if (!p) |
781 | return -EIO; | |
b411b363 | 782 | |
44ed167d | 783 | rcu_read_lock(); |
bde89a9e | 784 | nc = rcu_dereference(connection->net_conf); |
b411b363 | 785 | |
bde89a9e | 786 | if (nc->tentative && connection->agreed_pro_version < 92) { |
44ed167d PR |
787 | rcu_read_unlock(); |
788 | mutex_unlock(&sock->mutex); | |
1ec861eb | 789 | drbd_err(connection, "--dry-run is not supported by peer"); |
44ed167d PR |
790 | return -EOPNOTSUPP; |
791 | } | |
b411b363 | 792 | |
9f5bdc33 | 793 | size = sizeof(*p); |
bde89a9e | 794 | if (connection->agreed_pro_version >= 87) |
44ed167d | 795 | size += strlen(nc->integrity_alg) + 1; |
b411b363 | 796 | |
44ed167d PR |
797 | p->protocol = cpu_to_be32(nc->wire_protocol); |
798 | p->after_sb_0p = cpu_to_be32(nc->after_sb_0p); | |
799 | p->after_sb_1p = cpu_to_be32(nc->after_sb_1p); | |
800 | p->after_sb_2p = cpu_to_be32(nc->after_sb_2p); | |
801 | p->two_primaries = cpu_to_be32(nc->two_primaries); | |
cf14c2e9 | 802 | cf = 0; |
6139f60d AG |
803 | if (nc->discard_my_data) |
804 | cf |= CF_DISCARD_MY_DATA; | |
6dff2902 | 805 | if (nc->tentative) |
9f5bdc33 | 806 | cf |= CF_DRY_RUN; |
cf14c2e9 PR |
807 | p->conn_flags = cpu_to_be32(cf); |
808 | ||
bde89a9e | 809 | if (connection->agreed_pro_version >= 87) |
44ed167d PR |
810 | strcpy(p->integrity_alg, nc->integrity_alg); |
811 | rcu_read_unlock(); | |
b411b363 | 812 | |
bde89a9e | 813 | return __conn_send_command(connection, sock, cmd, size, NULL, 0); |
a7eb7bdf AG |
814 | } |
815 | ||
bde89a9e | 816 | int drbd_send_protocol(struct drbd_connection *connection) |
a7eb7bdf AG |
817 | { |
818 | int err; | |
819 | ||
bde89a9e AG |
820 | mutex_lock(&connection->data.mutex); |
821 | err = __drbd_send_protocol(connection, P_PROTOCOL); | |
822 | mutex_unlock(&connection->data.mutex); | |
a7eb7bdf AG |
823 | |
824 | return err; | |
b411b363 PR |
825 | } |
826 | ||
b30ab791 | 827 | static int _drbd_send_uuids(struct drbd_device *device, u64 uuid_flags) |
b411b363 | 828 | { |
9f5bdc33 AG |
829 | struct drbd_socket *sock; |
830 | struct p_uuids *p; | |
b411b363 PR |
831 | int i; |
832 | ||
b30ab791 | 833 | if (!get_ldev_if_state(device, D_NEGOTIATING)) |
2ae5f95b | 834 | return 0; |
b411b363 | 835 | |
a6b32bc3 | 836 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 837 | p = drbd_prepare_command(device, sock); |
9f5bdc33 | 838 | if (!p) { |
b30ab791 | 839 | put_ldev(device); |
9f5bdc33 AG |
840 | return -EIO; |
841 | } | |
b30ab791 | 842 | spin_lock_irq(&device->ldev->md.uuid_lock); |
b411b363 | 843 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
844 | p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
845 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 846 | |
b30ab791 AG |
847 | device->comm_bm_set = drbd_bm_total_weight(device); |
848 | p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set); | |
44ed167d | 849 | rcu_read_lock(); |
a6b32bc3 | 850 | uuid_flags |= rcu_dereference(first_peer_device(device)->connection->net_conf)->discard_my_data ? 1 : 0; |
44ed167d | 851 | rcu_read_unlock(); |
b30ab791 AG |
852 | uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0; |
853 | uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0; | |
9f5bdc33 | 854 | p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags); |
b411b363 | 855 | |
b30ab791 AG |
856 | put_ldev(device); |
857 | return drbd_send_command(device, sock, P_UUIDS, sizeof(*p), NULL, 0); | |
b411b363 PR |
858 | } |
859 | ||
b30ab791 | 860 | int drbd_send_uuids(struct drbd_device *device) |
b411b363 | 861 | { |
b30ab791 | 862 | return _drbd_send_uuids(device, 0); |
b411b363 PR |
863 | } |
864 | ||
b30ab791 | 865 | int drbd_send_uuids_skip_initial_sync(struct drbd_device *device) |
b411b363 | 866 | { |
b30ab791 | 867 | return _drbd_send_uuids(device, 8); |
b411b363 PR |
868 | } |
869 | ||
b30ab791 | 870 | void drbd_print_uuids(struct drbd_device *device, const char *text) |
62b0da3a | 871 | { |
b30ab791 AG |
872 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
873 | u64 *uuid = device->ldev->md.uuid; | |
d0180171 | 874 | drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n", |
62b0da3a LE |
875 | text, |
876 | (unsigned long long)uuid[UI_CURRENT], | |
877 | (unsigned long long)uuid[UI_BITMAP], | |
878 | (unsigned long long)uuid[UI_HISTORY_START], | |
879 | (unsigned long long)uuid[UI_HISTORY_END]); | |
b30ab791 | 880 | put_ldev(device); |
62b0da3a | 881 | } else { |
d0180171 | 882 | drbd_info(device, "%s effective data uuid: %016llX\n", |
62b0da3a | 883 | text, |
b30ab791 | 884 | (unsigned long long)device->ed_uuid); |
62b0da3a LE |
885 | } |
886 | } | |
887 | ||
b30ab791 | 888 | void drbd_gen_and_send_sync_uuid(struct drbd_device *device) |
b411b363 | 889 | { |
9f5bdc33 AG |
890 | struct drbd_socket *sock; |
891 | struct p_rs_uuid *p; | |
5a22db89 LE |
892 | u64 uuid; |
893 | ||
0b0ba1ef | 894 | D_ASSERT(device, device->state.disk == D_UP_TO_DATE); |
b411b363 | 895 | |
b30ab791 | 896 | uuid = device->ldev->md.uuid[UI_BITMAP]; |
5ba3dac5 PR |
897 | if (uuid && uuid != UUID_JUST_CREATED) |
898 | uuid = uuid + UUID_NEW_BM_OFFSET; | |
899 | else | |
900 | get_random_bytes(&uuid, sizeof(u64)); | |
b30ab791 AG |
901 | drbd_uuid_set(device, UI_BITMAP, uuid); |
902 | drbd_print_uuids(device, "updated sync UUID"); | |
903 | drbd_md_sync(device); | |
b411b363 | 904 | |
a6b32bc3 | 905 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 906 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
907 | if (p) { |
908 | p->uuid = cpu_to_be64(uuid); | |
b30ab791 | 909 | drbd_send_command(device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0); |
9f5bdc33 | 910 | } |
b411b363 PR |
911 | } |
912 | ||
b30ab791 | 913 | int drbd_send_sizes(struct drbd_device *device, int trigger_reply, enum dds_flags flags) |
b411b363 | 914 | { |
9f5bdc33 AG |
915 | struct drbd_socket *sock; |
916 | struct p_sizes *p; | |
b411b363 | 917 | sector_t d_size, u_size; |
db141b2f LE |
918 | int q_order_type; |
919 | unsigned int max_bio_size; | |
b411b363 | 920 | |
b30ab791 | 921 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
0b0ba1ef | 922 | D_ASSERT(device, device->ldev->backing_bdev); |
b30ab791 | 923 | d_size = drbd_get_max_capacity(device->ldev); |
daeda1cc | 924 | rcu_read_lock(); |
b30ab791 | 925 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
daeda1cc | 926 | rcu_read_unlock(); |
b30ab791 AG |
927 | q_order_type = drbd_queue_order_type(device); |
928 | max_bio_size = queue_max_hw_sectors(device->ldev->backing_bdev->bd_disk->queue) << 9; | |
db141b2f | 929 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE); |
b30ab791 | 930 | put_ldev(device); |
b411b363 PR |
931 | } else { |
932 | d_size = 0; | |
933 | u_size = 0; | |
934 | q_order_type = QUEUE_ORDERED_NONE; | |
99432fcc | 935 | max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */ |
b411b363 PR |
936 | } |
937 | ||
a6b32bc3 | 938 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 939 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
940 | if (!p) |
941 | return -EIO; | |
6809384c | 942 | |
a6b32bc3 | 943 | if (first_peer_device(device)->connection->agreed_pro_version <= 94) |
98683650 | 944 | max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET); |
a6b32bc3 | 945 | else if (first_peer_device(device)->connection->agreed_pro_version < 100) |
98683650 | 946 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95); |
b411b363 | 947 | |
9f5bdc33 AG |
948 | p->d_size = cpu_to_be64(d_size); |
949 | p->u_size = cpu_to_be64(u_size); | |
b30ab791 | 950 | p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev)); |
9f5bdc33 AG |
951 | p->max_bio_size = cpu_to_be32(max_bio_size); |
952 | p->queue_order_type = cpu_to_be16(q_order_type); | |
953 | p->dds_flags = cpu_to_be16(flags); | |
b30ab791 | 954 | return drbd_send_command(device, sock, P_SIZES, sizeof(*p), NULL, 0); |
b411b363 PR |
955 | } |
956 | ||
957 | /** | |
f479ea06 | 958 | * drbd_send_current_state() - Sends the drbd state to the peer |
b30ab791 | 959 | * @device: DRBD device. |
b411b363 | 960 | */ |
b30ab791 | 961 | int drbd_send_current_state(struct drbd_device *device) |
b411b363 | 962 | { |
7c96715a | 963 | struct drbd_socket *sock; |
9f5bdc33 | 964 | struct p_state *p; |
b411b363 | 965 | |
a6b32bc3 | 966 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 967 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
968 | if (!p) |
969 | return -EIO; | |
b30ab791 AG |
970 | p->state = cpu_to_be32(device->state.i); /* Within the send mutex */ |
971 | return drbd_send_command(device, sock, P_STATE, sizeof(*p), NULL, 0); | |
b411b363 PR |
972 | } |
973 | ||
f479ea06 LE |
974 | /** |
975 | * drbd_send_state() - After a state change, sends the new state to the peer | |
b30ab791 | 976 | * @device: DRBD device. |
43de7c85 | 977 | * @state: the state to send, not necessarily the current state. |
f479ea06 LE |
978 | * |
979 | * Each state change queues an "after_state_ch" work, which will eventually | |
980 | * send the resulting new state to the peer. If more state changes happen | |
981 | * between queuing and processing of the after_state_ch work, we still | |
982 | * want to send each intermediary state in the order it occurred. | |
983 | */ | |
b30ab791 | 984 | int drbd_send_state(struct drbd_device *device, union drbd_state state) |
f479ea06 | 985 | { |
43de7c85 PR |
986 | struct drbd_socket *sock; |
987 | struct p_state *p; | |
f479ea06 | 988 | |
a6b32bc3 | 989 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 990 | p = drbd_prepare_command(device, sock); |
43de7c85 PR |
991 | if (!p) |
992 | return -EIO; | |
993 | p->state = cpu_to_be32(state.i); /* Within the send mutex */ | |
b30ab791 | 994 | return drbd_send_command(device, sock, P_STATE, sizeof(*p), NULL, 0); |
43de7c85 | 995 | } |
f479ea06 | 996 | |
b30ab791 | 997 | int drbd_send_state_req(struct drbd_device *device, union drbd_state mask, union drbd_state val) |
9f5bdc33 AG |
998 | { |
999 | struct drbd_socket *sock; | |
1000 | struct p_req_state *p; | |
f479ea06 | 1001 | |
a6b32bc3 | 1002 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1003 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1004 | if (!p) |
1005 | return -EIO; | |
1006 | p->mask = cpu_to_be32(mask.i); | |
1007 | p->val = cpu_to_be32(val.i); | |
b30ab791 | 1008 | return drbd_send_command(device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0); |
b411b363 | 1009 | } |
f479ea06 | 1010 | |
bde89a9e | 1011 | int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val) |
b411b363 | 1012 | { |
9f5bdc33 AG |
1013 | enum drbd_packet cmd; |
1014 | struct drbd_socket *sock; | |
1015 | struct p_req_state *p; | |
f479ea06 | 1016 | |
bde89a9e AG |
1017 | cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ; |
1018 | sock = &connection->data; | |
1019 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1020 | if (!p) |
1021 | return -EIO; | |
1022 | p->mask = cpu_to_be32(mask.i); | |
1023 | p->val = cpu_to_be32(val.i); | |
bde89a9e | 1024 | return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
f479ea06 LE |
1025 | } |
1026 | ||
b30ab791 | 1027 | void drbd_send_sr_reply(struct drbd_device *device, enum drbd_state_rv retcode) |
b411b363 | 1028 | { |
9f5bdc33 AG |
1029 | struct drbd_socket *sock; |
1030 | struct p_req_state_reply *p; | |
b411b363 | 1031 | |
a6b32bc3 | 1032 | sock = &first_peer_device(device)->connection->meta; |
b30ab791 | 1033 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1034 | if (p) { |
1035 | p->retcode = cpu_to_be32(retcode); | |
b30ab791 | 1036 | drbd_send_command(device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0); |
9f5bdc33 | 1037 | } |
b411b363 | 1038 | } |
b411b363 | 1039 | |
bde89a9e | 1040 | void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode) |
047cd4a6 | 1041 | { |
9f5bdc33 AG |
1042 | struct drbd_socket *sock; |
1043 | struct p_req_state_reply *p; | |
bde89a9e | 1044 | enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY; |
b411b363 | 1045 | |
bde89a9e AG |
1046 | sock = &connection->meta; |
1047 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1048 | if (p) { |
1049 | p->retcode = cpu_to_be32(retcode); | |
bde89a9e | 1050 | conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
9f5bdc33 | 1051 | } |
b411b363 PR |
1052 | } |
1053 | ||
a02d1240 | 1054 | static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) |
b411b363 | 1055 | { |
a02d1240 AG |
1056 | BUG_ON(code & ~0xf); |
1057 | p->encoding = (p->encoding & ~0xf) | code; | |
1058 | } | |
b411b363 | 1059 | |
a02d1240 AG |
1060 | static void dcbp_set_start(struct p_compressed_bm *p, int set) |
1061 | { | |
1062 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
1063 | } | |
b411b363 | 1064 | |
a02d1240 AG |
1065 | static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n) |
1066 | { | |
1067 | BUG_ON(n & ~0x7); | |
1068 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
b411b363 PR |
1069 | } |
1070 | ||
b30ab791 | 1071 | static int fill_bitmap_rle_bits(struct drbd_device *device, |
50d0b1ad AG |
1072 | struct p_compressed_bm *p, |
1073 | unsigned int size, | |
1074 | struct bm_xfer_ctx *c) | |
b411b363 PR |
1075 | { |
1076 | struct bitstream bs; | |
1077 | unsigned long plain_bits; | |
1078 | unsigned long tmp; | |
1079 | unsigned long rl; | |
1080 | unsigned len; | |
1081 | unsigned toggle; | |
44ed167d | 1082 | int bits, use_rle; |
b411b363 PR |
1083 | |
1084 | /* may we use this feature? */ | |
44ed167d | 1085 | rcu_read_lock(); |
a6b32bc3 | 1086 | use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle; |
44ed167d | 1087 | rcu_read_unlock(); |
a6b32bc3 | 1088 | if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90) |
44ed167d | 1089 | return 0; |
b411b363 PR |
1090 | |
1091 | if (c->bit_offset >= c->bm_bits) | |
1092 | return 0; /* nothing to do. */ | |
1093 | ||
1094 | /* use at most thus many bytes */ | |
50d0b1ad AG |
1095 | bitstream_init(&bs, p->code, size, 0); |
1096 | memset(p->code, 0, size); | |
b411b363 PR |
1097 | /* plain bits covered in this code string */ |
1098 | plain_bits = 0; | |
1099 | ||
1100 | /* p->encoding & 0x80 stores whether the first run length is set. | |
1101 | * bit offset is implicit. | |
1102 | * start with toggle == 2 to be able to tell the first iteration */ | |
1103 | toggle = 2; | |
1104 | ||
1105 | /* see how much plain bits we can stuff into one packet | |
1106 | * using RLE and VLI. */ | |
1107 | do { | |
b30ab791 AG |
1108 | tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset) |
1109 | : _drbd_bm_find_next(device, c->bit_offset); | |
b411b363 PR |
1110 | if (tmp == -1UL) |
1111 | tmp = c->bm_bits; | |
1112 | rl = tmp - c->bit_offset; | |
1113 | ||
1114 | if (toggle == 2) { /* first iteration */ | |
1115 | if (rl == 0) { | |
1116 | /* the first checked bit was set, | |
1117 | * store start value, */ | |
a02d1240 | 1118 | dcbp_set_start(p, 1); |
b411b363 PR |
1119 | /* but skip encoding of zero run length */ |
1120 | toggle = !toggle; | |
1121 | continue; | |
1122 | } | |
a02d1240 | 1123 | dcbp_set_start(p, 0); |
b411b363 PR |
1124 | } |
1125 | ||
1126 | /* paranoia: catch zero runlength. | |
1127 | * can only happen if bitmap is modified while we scan it. */ | |
1128 | if (rl == 0) { | |
d0180171 | 1129 | drbd_err(device, "unexpected zero runlength while encoding bitmap " |
b411b363 PR |
1130 | "t:%u bo:%lu\n", toggle, c->bit_offset); |
1131 | return -1; | |
1132 | } | |
1133 | ||
1134 | bits = vli_encode_bits(&bs, rl); | |
1135 | if (bits == -ENOBUFS) /* buffer full */ | |
1136 | break; | |
1137 | if (bits <= 0) { | |
d0180171 | 1138 | drbd_err(device, "error while encoding bitmap: %d\n", bits); |
b411b363 PR |
1139 | return 0; |
1140 | } | |
1141 | ||
1142 | toggle = !toggle; | |
1143 | plain_bits += rl; | |
1144 | c->bit_offset = tmp; | |
1145 | } while (c->bit_offset < c->bm_bits); | |
1146 | ||
1147 | len = bs.cur.b - p->code + !!bs.cur.bit; | |
1148 | ||
1149 | if (plain_bits < (len << 3)) { | |
1150 | /* incompressible with this method. | |
1151 | * we need to rewind both word and bit position. */ | |
1152 | c->bit_offset -= plain_bits; | |
1153 | bm_xfer_ctx_bit_to_word_offset(c); | |
1154 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1155 | return 0; | |
1156 | } | |
1157 | ||
1158 | /* RLE + VLI was able to compress it just fine. | |
1159 | * update c->word_offset. */ | |
1160 | bm_xfer_ctx_bit_to_word_offset(c); | |
1161 | ||
1162 | /* store pad_bits */ | |
a02d1240 | 1163 | dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7); |
b411b363 PR |
1164 | |
1165 | return len; | |
1166 | } | |
1167 | ||
f70af118 AG |
1168 | /** |
1169 | * send_bitmap_rle_or_plain | |
1170 | * | |
1171 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
1172 | * code upon failure. | |
1173 | */ | |
1174 | static int | |
b30ab791 | 1175 | send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c) |
b411b363 | 1176 | { |
a6b32bc3 AG |
1177 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
1178 | unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); | |
e658983a | 1179 | struct p_compressed_bm *p = sock->sbuf + header_size; |
a982dd57 | 1180 | int len, err; |
b411b363 | 1181 | |
b30ab791 | 1182 | len = fill_bitmap_rle_bits(device, p, |
e658983a | 1183 | DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c); |
b411b363 | 1184 | if (len < 0) |
f70af118 | 1185 | return -EIO; |
b411b363 PR |
1186 | |
1187 | if (len) { | |
a02d1240 | 1188 | dcbp_set_code(p, RLE_VLI_Bits); |
a6b32bc3 | 1189 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, |
9f5bdc33 AG |
1190 | P_COMPRESSED_BITMAP, sizeof(*p) + len, |
1191 | NULL, 0); | |
b411b363 | 1192 | c->packets[0]++; |
e658983a | 1193 | c->bytes[0] += header_size + sizeof(*p) + len; |
b411b363 PR |
1194 | |
1195 | if (c->bit_offset >= c->bm_bits) | |
1196 | len = 0; /* DONE */ | |
1197 | } else { | |
1198 | /* was not compressible. | |
1199 | * send a buffer full of plain text bits instead. */ | |
50d0b1ad AG |
1200 | unsigned int data_size; |
1201 | unsigned long num_words; | |
e658983a | 1202 | unsigned long *p = sock->sbuf + header_size; |
50d0b1ad AG |
1203 | |
1204 | data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; | |
e658983a | 1205 | num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 1206 | c->bm_words - c->word_offset); |
e658983a | 1207 | len = num_words * sizeof(*p); |
b411b363 | 1208 | if (len) |
b30ab791 | 1209 | drbd_bm_get_lel(device, c->word_offset, num_words, p); |
a6b32bc3 | 1210 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0); |
b411b363 PR |
1211 | c->word_offset += num_words; |
1212 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1213 | ||
1214 | c->packets[1]++; | |
50d0b1ad | 1215 | c->bytes[1] += header_size + len; |
b411b363 PR |
1216 | |
1217 | if (c->bit_offset > c->bm_bits) | |
1218 | c->bit_offset = c->bm_bits; | |
1219 | } | |
a982dd57 | 1220 | if (!err) { |
f70af118 | 1221 | if (len == 0) { |
b30ab791 | 1222 | INFO_bm_xfer_stats(device, "send", c); |
f70af118 AG |
1223 | return 0; |
1224 | } else | |
1225 | return 1; | |
1226 | } | |
1227 | return -EIO; | |
b411b363 PR |
1228 | } |
1229 | ||
1230 | /* See the comment at receive_bitmap() */ | |
b30ab791 | 1231 | static int _drbd_send_bitmap(struct drbd_device *device) |
b411b363 PR |
1232 | { |
1233 | struct bm_xfer_ctx c; | |
f70af118 | 1234 | int err; |
b411b363 | 1235 | |
b30ab791 | 1236 | if (!expect(device->bitmap)) |
81e84650 | 1237 | return false; |
b411b363 | 1238 | |
b30ab791 AG |
1239 | if (get_ldev(device)) { |
1240 | if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) { | |
d0180171 | 1241 | drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n"); |
b30ab791 AG |
1242 | drbd_bm_set_all(device); |
1243 | if (drbd_bm_write(device)) { | |
b411b363 PR |
1244 | /* write_bm did fail! Leave full sync flag set in Meta P_DATA |
1245 | * but otherwise process as per normal - need to tell other | |
1246 | * side that a full resync is required! */ | |
d0180171 | 1247 | drbd_err(device, "Failed to write bitmap to disk!\n"); |
b411b363 | 1248 | } else { |
b30ab791 AG |
1249 | drbd_md_clear_flag(device, MDF_FULL_SYNC); |
1250 | drbd_md_sync(device); | |
b411b363 PR |
1251 | } |
1252 | } | |
b30ab791 | 1253 | put_ldev(device); |
b411b363 PR |
1254 | } |
1255 | ||
1256 | c = (struct bm_xfer_ctx) { | |
b30ab791 AG |
1257 | .bm_bits = drbd_bm_bits(device), |
1258 | .bm_words = drbd_bm_words(device), | |
b411b363 PR |
1259 | }; |
1260 | ||
1261 | do { | |
b30ab791 | 1262 | err = send_bitmap_rle_or_plain(device, &c); |
f70af118 | 1263 | } while (err > 0); |
b411b363 | 1264 | |
f70af118 | 1265 | return err == 0; |
b411b363 PR |
1266 | } |
1267 | ||
b30ab791 | 1268 | int drbd_send_bitmap(struct drbd_device *device) |
b411b363 | 1269 | { |
a6b32bc3 | 1270 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
9f5bdc33 | 1271 | int err = -1; |
b411b363 | 1272 | |
9f5bdc33 AG |
1273 | mutex_lock(&sock->mutex); |
1274 | if (sock->socket) | |
b30ab791 | 1275 | err = !_drbd_send_bitmap(device); |
9f5bdc33 | 1276 | mutex_unlock(&sock->mutex); |
b411b363 PR |
1277 | return err; |
1278 | } | |
1279 | ||
bde89a9e | 1280 | void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size) |
b411b363 | 1281 | { |
9f5bdc33 AG |
1282 | struct drbd_socket *sock; |
1283 | struct p_barrier_ack *p; | |
b411b363 | 1284 | |
bde89a9e | 1285 | if (connection->cstate < C_WF_REPORT_PARAMS) |
9f5bdc33 | 1286 | return; |
b411b363 | 1287 | |
bde89a9e AG |
1288 | sock = &connection->meta; |
1289 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1290 | if (!p) |
1291 | return; | |
1292 | p->barrier = barrier_nr; | |
1293 | p->set_size = cpu_to_be32(set_size); | |
bde89a9e | 1294 | conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0); |
b411b363 PR |
1295 | } |
1296 | ||
1297 | /** | |
1298 | * _drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1299 | * @device: DRBD device. |
b411b363 PR |
1300 | * @cmd: Packet command code. |
1301 | * @sector: sector, needs to be in big endian byte order | |
1302 | * @blksize: size in byte, needs to be in big endian byte order | |
1303 | * @block_id: Id, big endian byte order | |
1304 | */ | |
b30ab791 | 1305 | static int _drbd_send_ack(struct drbd_device *device, enum drbd_packet cmd, |
d8763023 | 1306 | u64 sector, u32 blksize, u64 block_id) |
b411b363 | 1307 | { |
9f5bdc33 AG |
1308 | struct drbd_socket *sock; |
1309 | struct p_block_ack *p; | |
b411b363 | 1310 | |
b30ab791 | 1311 | if (device->state.conn < C_CONNECTED) |
9f5bdc33 | 1312 | return -EIO; |
b411b363 | 1313 | |
a6b32bc3 | 1314 | sock = &first_peer_device(device)->connection->meta; |
b30ab791 | 1315 | p = drbd_prepare_command(device, sock); |
9f5bdc33 | 1316 | if (!p) |
a8c32aa8 | 1317 | return -EIO; |
9f5bdc33 AG |
1318 | p->sector = sector; |
1319 | p->block_id = block_id; | |
1320 | p->blksize = blksize; | |
b30ab791 AG |
1321 | p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); |
1322 | return drbd_send_command(device, sock, cmd, sizeof(*p), NULL, 0); | |
b411b363 PR |
1323 | } |
1324 | ||
2b2bf214 LE |
1325 | /* dp->sector and dp->block_id already/still in network byte order, |
1326 | * data_size is payload size according to dp->head, | |
1327 | * and may need to be corrected for digest size. */ | |
b30ab791 | 1328 | void drbd_send_ack_dp(struct drbd_device *device, enum drbd_packet cmd, |
a9a9994d | 1329 | struct p_data *dp, int data_size) |
b411b363 | 1330 | { |
a6b32bc3 AG |
1331 | if (first_peer_device(device)->connection->peer_integrity_tfm) |
1332 | data_size -= crypto_hash_digestsize(first_peer_device(device)->connection->peer_integrity_tfm); | |
b30ab791 | 1333 | _drbd_send_ack(device, cmd, dp->sector, cpu_to_be32(data_size), |
a9a9994d | 1334 | dp->block_id); |
b411b363 PR |
1335 | } |
1336 | ||
b30ab791 | 1337 | void drbd_send_ack_rp(struct drbd_device *device, enum drbd_packet cmd, |
a9a9994d | 1338 | struct p_block_req *rp) |
b411b363 | 1339 | { |
b30ab791 | 1340 | _drbd_send_ack(device, cmd, rp->sector, rp->blksize, rp->block_id); |
b411b363 PR |
1341 | } |
1342 | ||
1343 | /** | |
1344 | * drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1345 | * @device: DRBD device |
db830c46 AG |
1346 | * @cmd: packet command code |
1347 | * @peer_req: peer request | |
b411b363 | 1348 | */ |
b30ab791 | 1349 | int drbd_send_ack(struct drbd_device *device, enum drbd_packet cmd, |
db830c46 | 1350 | struct drbd_peer_request *peer_req) |
b411b363 | 1351 | { |
b30ab791 | 1352 | return _drbd_send_ack(device, cmd, |
dd516121 AG |
1353 | cpu_to_be64(peer_req->i.sector), |
1354 | cpu_to_be32(peer_req->i.size), | |
1355 | peer_req->block_id); | |
b411b363 PR |
1356 | } |
1357 | ||
1358 | /* This function misuses the block_id field to signal if the blocks | |
1359 | * are is sync or not. */ | |
b30ab791 | 1360 | int drbd_send_ack_ex(struct drbd_device *device, enum drbd_packet cmd, |
b411b363 PR |
1361 | sector_t sector, int blksize, u64 block_id) |
1362 | { | |
b30ab791 | 1363 | return _drbd_send_ack(device, cmd, |
b411b363 PR |
1364 | cpu_to_be64(sector), |
1365 | cpu_to_be32(blksize), | |
1366 | cpu_to_be64(block_id)); | |
1367 | } | |
1368 | ||
b30ab791 | 1369 | int drbd_send_drequest(struct drbd_device *device, int cmd, |
b411b363 PR |
1370 | sector_t sector, int size, u64 block_id) |
1371 | { | |
9f5bdc33 AG |
1372 | struct drbd_socket *sock; |
1373 | struct p_block_req *p; | |
b411b363 | 1374 | |
a6b32bc3 | 1375 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1376 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1377 | if (!p) |
1378 | return -EIO; | |
1379 | p->sector = cpu_to_be64(sector); | |
1380 | p->block_id = block_id; | |
1381 | p->blksize = cpu_to_be32(size); | |
b30ab791 | 1382 | return drbd_send_command(device, sock, cmd, sizeof(*p), NULL, 0); |
b411b363 PR |
1383 | } |
1384 | ||
b30ab791 | 1385 | int drbd_send_drequest_csum(struct drbd_device *device, sector_t sector, int size, |
d8763023 | 1386 | void *digest, int digest_size, enum drbd_packet cmd) |
b411b363 | 1387 | { |
9f5bdc33 AG |
1388 | struct drbd_socket *sock; |
1389 | struct p_block_req *p; | |
b411b363 | 1390 | |
9f5bdc33 | 1391 | /* FIXME: Put the digest into the preallocated socket buffer. */ |
b411b363 | 1392 | |
a6b32bc3 | 1393 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1394 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1395 | if (!p) |
1396 | return -EIO; | |
1397 | p->sector = cpu_to_be64(sector); | |
1398 | p->block_id = ID_SYNCER /* unused */; | |
1399 | p->blksize = cpu_to_be32(size); | |
b30ab791 | 1400 | return drbd_send_command(device, sock, cmd, sizeof(*p), |
9f5bdc33 | 1401 | digest, digest_size); |
b411b363 PR |
1402 | } |
1403 | ||
b30ab791 | 1404 | int drbd_send_ov_request(struct drbd_device *device, sector_t sector, int size) |
b411b363 | 1405 | { |
9f5bdc33 AG |
1406 | struct drbd_socket *sock; |
1407 | struct p_block_req *p; | |
b411b363 | 1408 | |
a6b32bc3 | 1409 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1410 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1411 | if (!p) |
1412 | return -EIO; | |
1413 | p->sector = cpu_to_be64(sector); | |
1414 | p->block_id = ID_SYNCER /* unused */; | |
1415 | p->blksize = cpu_to_be32(size); | |
b30ab791 | 1416 | return drbd_send_command(device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0); |
b411b363 PR |
1417 | } |
1418 | ||
1419 | /* called on sndtimeo | |
81e84650 AG |
1420 | * returns false if we should retry, |
1421 | * true if we think connection is dead | |
b411b363 | 1422 | */ |
bde89a9e | 1423 | static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock) |
b411b363 PR |
1424 | { |
1425 | int drop_it; | |
b30ab791 | 1426 | /* long elapsed = (long)(jiffies - device->last_received); */ |
b411b363 | 1427 | |
bde89a9e AG |
1428 | drop_it = connection->meta.socket == sock |
1429 | || !connection->asender.task | |
1430 | || get_t_state(&connection->asender) != RUNNING | |
1431 | || connection->cstate < C_WF_REPORT_PARAMS; | |
b411b363 PR |
1432 | |
1433 | if (drop_it) | |
81e84650 | 1434 | return true; |
b411b363 | 1435 | |
bde89a9e | 1436 | drop_it = !--connection->ko_count; |
b411b363 | 1437 | if (!drop_it) { |
1ec861eb | 1438 | drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n", |
bde89a9e AG |
1439 | current->comm, current->pid, connection->ko_count); |
1440 | request_ping(connection); | |
b411b363 PR |
1441 | } |
1442 | ||
b30ab791 | 1443 | return drop_it; /* && (device->state == R_PRIMARY) */; |
b411b363 PR |
1444 | } |
1445 | ||
bde89a9e | 1446 | static void drbd_update_congested(struct drbd_connection *connection) |
9e204cdd | 1447 | { |
bde89a9e | 1448 | struct sock *sk = connection->data.socket->sk; |
9e204cdd | 1449 | if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) |
bde89a9e | 1450 | set_bit(NET_CONGESTED, &connection->flags); |
9e204cdd AG |
1451 | } |
1452 | ||
b411b363 PR |
1453 | /* The idea of sendpage seems to be to put some kind of reference |
1454 | * to the page into the skb, and to hand it over to the NIC. In | |
1455 | * this process get_page() gets called. | |
1456 | * | |
1457 | * As soon as the page was really sent over the network put_page() | |
1458 | * gets called by some part of the network layer. [ NIC driver? ] | |
1459 | * | |
1460 | * [ get_page() / put_page() increment/decrement the count. If count | |
1461 | * reaches 0 the page will be freed. ] | |
1462 | * | |
1463 | * This works nicely with pages from FSs. | |
1464 | * But this means that in protocol A we might signal IO completion too early! | |
1465 | * | |
1466 | * In order not to corrupt data during a resync we must make sure | |
1467 | * that we do not reuse our own buffer pages (EEs) to early, therefore | |
1468 | * we have the net_ee list. | |
1469 | * | |
1470 | * XFS seems to have problems, still, it submits pages with page_count == 0! | |
1471 | * As a workaround, we disable sendpage on pages | |
1472 | * with page_count == 0 or PageSlab. | |
1473 | */ | |
b30ab791 | 1474 | static int _drbd_no_send_page(struct drbd_device *device, struct page *page, |
b987427b | 1475 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1476 | { |
b987427b AG |
1477 | struct socket *socket; |
1478 | void *addr; | |
1479 | int err; | |
1480 | ||
a6b32bc3 | 1481 | socket = first_peer_device(device)->connection->data.socket; |
b987427b | 1482 | addr = kmap(page) + offset; |
a6b32bc3 | 1483 | err = drbd_send_all(first_peer_device(device)->connection, socket, addr, size, msg_flags); |
b411b363 | 1484 | kunmap(page); |
b987427b | 1485 | if (!err) |
b30ab791 | 1486 | device->send_cnt += size >> 9; |
b987427b | 1487 | return err; |
b411b363 PR |
1488 | } |
1489 | ||
b30ab791 | 1490 | static int _drbd_send_page(struct drbd_device *device, struct page *page, |
ba11ad9a | 1491 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1492 | { |
a6b32bc3 | 1493 | struct socket *socket = first_peer_device(device)->connection->data.socket; |
b411b363 | 1494 | mm_segment_t oldfs = get_fs(); |
b411b363 | 1495 | int len = size; |
88b390ff | 1496 | int err = -EIO; |
b411b363 PR |
1497 | |
1498 | /* e.g. XFS meta- & log-data is in slab pages, which have a | |
1499 | * page_count of 0 and/or have PageSlab() set. | |
1500 | * we cannot use send_page for those, as that does get_page(); | |
1501 | * put_page(); and would cause either a VM_BUG directly, or | |
1502 | * __page_cache_release a page that would actually still be referenced | |
1503 | * by someone, leading to some obscure delayed Oops somewhere else. */ | |
1504 | if (disable_sendpage || (page_count(page) < 1) || PageSlab(page)) | |
b30ab791 | 1505 | return _drbd_no_send_page(device, page, offset, size, msg_flags); |
b411b363 | 1506 | |
ba11ad9a | 1507 | msg_flags |= MSG_NOSIGNAL; |
a6b32bc3 | 1508 | drbd_update_congested(first_peer_device(device)->connection); |
b411b363 PR |
1509 | set_fs(KERNEL_DS); |
1510 | do { | |
88b390ff AG |
1511 | int sent; |
1512 | ||
1513 | sent = socket->ops->sendpage(socket, page, offset, len, msg_flags); | |
b411b363 | 1514 | if (sent <= 0) { |
88b390ff | 1515 | if (sent == -EAGAIN) { |
a6b32bc3 | 1516 | if (we_should_drop_the_connection(first_peer_device(device)->connection, socket)) |
88b390ff AG |
1517 | break; |
1518 | continue; | |
1519 | } | |
d0180171 | 1520 | drbd_warn(device, "%s: size=%d len=%d sent=%d\n", |
b411b363 | 1521 | __func__, (int)size, len, sent); |
88b390ff AG |
1522 | if (sent < 0) |
1523 | err = sent; | |
b411b363 PR |
1524 | break; |
1525 | } | |
1526 | len -= sent; | |
1527 | offset += sent; | |
b30ab791 | 1528 | } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/); |
b411b363 | 1529 | set_fs(oldfs); |
a6b32bc3 | 1530 | clear_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags); |
b411b363 | 1531 | |
88b390ff AG |
1532 | if (len == 0) { |
1533 | err = 0; | |
b30ab791 | 1534 | device->send_cnt += size >> 9; |
88b390ff AG |
1535 | } |
1536 | return err; | |
b411b363 PR |
1537 | } |
1538 | ||
b30ab791 | 1539 | static int _drbd_send_bio(struct drbd_device *device, struct bio *bio) |
b411b363 | 1540 | { |
7988613b KO |
1541 | struct bio_vec bvec; |
1542 | struct bvec_iter iter; | |
1543 | ||
ba11ad9a | 1544 | /* hint all but last page with MSG_MORE */ |
7988613b | 1545 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1546 | int err; |
1547 | ||
b30ab791 | 1548 | err = _drbd_no_send_page(device, bvec.bv_page, |
7988613b | 1549 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1550 | bio_iter_last(bvec, iter) |
7988613b | 1551 | ? 0 : MSG_MORE); |
7fae55da AG |
1552 | if (err) |
1553 | return err; | |
b411b363 | 1554 | } |
7fae55da | 1555 | return 0; |
b411b363 PR |
1556 | } |
1557 | ||
b30ab791 | 1558 | static int _drbd_send_zc_bio(struct drbd_device *device, struct bio *bio) |
b411b363 | 1559 | { |
7988613b KO |
1560 | struct bio_vec bvec; |
1561 | struct bvec_iter iter; | |
1562 | ||
ba11ad9a | 1563 | /* hint all but last page with MSG_MORE */ |
7988613b | 1564 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1565 | int err; |
1566 | ||
b30ab791 | 1567 | err = _drbd_send_page(device, bvec.bv_page, |
7988613b | 1568 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1569 | bio_iter_last(bvec, iter) ? 0 : MSG_MORE); |
7fae55da AG |
1570 | if (err) |
1571 | return err; | |
b411b363 | 1572 | } |
7fae55da | 1573 | return 0; |
b411b363 PR |
1574 | } |
1575 | ||
b30ab791 | 1576 | static int _drbd_send_zc_ee(struct drbd_device *device, |
db830c46 | 1577 | struct drbd_peer_request *peer_req) |
45bb912b | 1578 | { |
db830c46 AG |
1579 | struct page *page = peer_req->pages; |
1580 | unsigned len = peer_req->i.size; | |
9f69230c | 1581 | int err; |
db830c46 | 1582 | |
ba11ad9a | 1583 | /* hint all but last page with MSG_MORE */ |
45bb912b LE |
1584 | page_chain_for_each(page) { |
1585 | unsigned l = min_t(unsigned, len, PAGE_SIZE); | |
9f69230c | 1586 | |
b30ab791 | 1587 | err = _drbd_send_page(device, page, 0, l, |
9f69230c AG |
1588 | page_chain_next(page) ? MSG_MORE : 0); |
1589 | if (err) | |
1590 | return err; | |
45bb912b LE |
1591 | len -= l; |
1592 | } | |
9f69230c | 1593 | return 0; |
45bb912b LE |
1594 | } |
1595 | ||
b30ab791 | 1596 | static u32 bio_flags_to_wire(struct drbd_device *device, unsigned long bi_rw) |
76d2e7ec | 1597 | { |
a6b32bc3 | 1598 | if (first_peer_device(device)->connection->agreed_pro_version >= 95) |
76d2e7ec | 1599 | return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) | |
76d2e7ec PR |
1600 | (bi_rw & REQ_FUA ? DP_FUA : 0) | |
1601 | (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) | | |
1602 | (bi_rw & REQ_DISCARD ? DP_DISCARD : 0); | |
1603 | else | |
721a9602 | 1604 | return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0; |
76d2e7ec PR |
1605 | } |
1606 | ||
b411b363 PR |
1607 | /* Used to send write requests |
1608 | * R_PRIMARY -> Peer (P_DATA) | |
1609 | */ | |
b30ab791 | 1610 | int drbd_send_dblock(struct drbd_device *device, struct drbd_request *req) |
b411b363 | 1611 | { |
9f5bdc33 AG |
1612 | struct drbd_socket *sock; |
1613 | struct p_data *p; | |
b411b363 | 1614 | unsigned int dp_flags = 0; |
b411b363 | 1615 | int dgs; |
9f5bdc33 | 1616 | int err; |
b411b363 | 1617 | |
a6b32bc3 | 1618 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1619 | p = drbd_prepare_command(device, sock); |
a6b32bc3 AG |
1620 | dgs = first_peer_device(device)->connection->integrity_tfm ? |
1621 | crypto_hash_digestsize(first_peer_device(device)->connection->integrity_tfm) : 0; | |
b411b363 | 1622 | |
9f5bdc33 AG |
1623 | if (!p) |
1624 | return -EIO; | |
1625 | p->sector = cpu_to_be64(req->i.sector); | |
1626 | p->block_id = (unsigned long)req; | |
b30ab791 AG |
1627 | p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); |
1628 | dp_flags = bio_flags_to_wire(device, req->master_bio->bi_rw); | |
1629 | if (device->state.conn >= C_SYNC_SOURCE && | |
1630 | device->state.conn <= C_PAUSED_SYNC_T) | |
b411b363 | 1631 | dp_flags |= DP_MAY_SET_IN_SYNC; |
a6b32bc3 | 1632 | if (first_peer_device(device)->connection->agreed_pro_version >= 100) { |
303d1448 PR |
1633 | if (req->rq_state & RQ_EXP_RECEIVE_ACK) |
1634 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
1635 | if (req->rq_state & RQ_EXP_WRITE_ACK) | |
1636 | dp_flags |= DP_SEND_WRITE_ACK; | |
1637 | } | |
9f5bdc33 AG |
1638 | p->dp_flags = cpu_to_be32(dp_flags); |
1639 | if (dgs) | |
a6b32bc3 AG |
1640 | drbd_csum_bio(device, first_peer_device(device)->connection->integrity_tfm, req->master_bio, p + 1); |
1641 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size); | |
6bdb9b0e | 1642 | if (!err) { |
470be44a LE |
1643 | /* For protocol A, we have to memcpy the payload into |
1644 | * socket buffers, as we may complete right away | |
1645 | * as soon as we handed it over to tcp, at which point the data | |
1646 | * pages may become invalid. | |
1647 | * | |
1648 | * For data-integrity enabled, we copy it as well, so we can be | |
1649 | * sure that even if the bio pages may still be modified, it | |
1650 | * won't change the data on the wire, thus if the digest checks | |
1651 | * out ok after sending on this side, but does not fit on the | |
1652 | * receiving side, we sure have detected corruption elsewhere. | |
1653 | */ | |
303d1448 | 1654 | if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs) |
b30ab791 | 1655 | err = _drbd_send_bio(device, req->master_bio); |
b411b363 | 1656 | else |
b30ab791 | 1657 | err = _drbd_send_zc_bio(device, req->master_bio); |
470be44a LE |
1658 | |
1659 | /* double check digest, sometimes buffers have been modified in flight. */ | |
1660 | if (dgs > 0 && dgs <= 64) { | |
24c4830c | 1661 | /* 64 byte, 512 bit, is the largest digest size |
470be44a LE |
1662 | * currently supported in kernel crypto. */ |
1663 | unsigned char digest[64]; | |
a6b32bc3 | 1664 | drbd_csum_bio(device, first_peer_device(device)->connection->integrity_tfm, req->master_bio, digest); |
9f5bdc33 | 1665 | if (memcmp(p + 1, digest, dgs)) { |
d0180171 | 1666 | drbd_warn(device, |
470be44a | 1667 | "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n", |
ace652ac | 1668 | (unsigned long long)req->i.sector, req->i.size); |
470be44a LE |
1669 | } |
1670 | } /* else if (dgs > 64) { | |
1671 | ... Be noisy about digest too large ... | |
1672 | } */ | |
b411b363 | 1673 | } |
9f5bdc33 | 1674 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
b411b363 | 1675 | |
6bdb9b0e | 1676 | return err; |
b411b363 PR |
1677 | } |
1678 | ||
1679 | /* answer packet, used to send data back for read requests: | |
1680 | * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY) | |
1681 | * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY) | |
1682 | */ | |
b30ab791 | 1683 | int drbd_send_block(struct drbd_device *device, enum drbd_packet cmd, |
db830c46 | 1684 | struct drbd_peer_request *peer_req) |
b411b363 | 1685 | { |
9f5bdc33 AG |
1686 | struct drbd_socket *sock; |
1687 | struct p_data *p; | |
7b57b89d | 1688 | int err; |
b411b363 PR |
1689 | int dgs; |
1690 | ||
a6b32bc3 | 1691 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1692 | p = drbd_prepare_command(device, sock); |
b411b363 | 1693 | |
a6b32bc3 AG |
1694 | dgs = first_peer_device(device)->connection->integrity_tfm ? |
1695 | crypto_hash_digestsize(first_peer_device(device)->connection->integrity_tfm) : 0; | |
b411b363 | 1696 | |
9f5bdc33 AG |
1697 | if (!p) |
1698 | return -EIO; | |
1699 | p->sector = cpu_to_be64(peer_req->i.sector); | |
1700 | p->block_id = peer_req->block_id; | |
1701 | p->seq_num = 0; /* unused */ | |
b17f33cb | 1702 | p->dp_flags = 0; |
9f5bdc33 | 1703 | if (dgs) |
a6b32bc3 AG |
1704 | drbd_csum_ee(device, first_peer_device(device)->connection->integrity_tfm, peer_req, p + 1); |
1705 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size); | |
7b57b89d | 1706 | if (!err) |
b30ab791 | 1707 | err = _drbd_send_zc_ee(device, peer_req); |
9f5bdc33 | 1708 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
bd26bfc5 | 1709 | |
7b57b89d | 1710 | return err; |
b411b363 PR |
1711 | } |
1712 | ||
b30ab791 | 1713 | int drbd_send_out_of_sync(struct drbd_device *device, struct drbd_request *req) |
73a01a18 | 1714 | { |
9f5bdc33 AG |
1715 | struct drbd_socket *sock; |
1716 | struct p_block_desc *p; | |
73a01a18 | 1717 | |
a6b32bc3 | 1718 | sock = &first_peer_device(device)->connection->data; |
b30ab791 | 1719 | p = drbd_prepare_command(device, sock); |
9f5bdc33 AG |
1720 | if (!p) |
1721 | return -EIO; | |
1722 | p->sector = cpu_to_be64(req->i.sector); | |
1723 | p->blksize = cpu_to_be32(req->i.size); | |
b30ab791 | 1724 | return drbd_send_command(device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0); |
73a01a18 PR |
1725 | } |
1726 | ||
b411b363 PR |
1727 | /* |
1728 | drbd_send distinguishes two cases: | |
1729 | ||
1730 | Packets sent via the data socket "sock" | |
1731 | and packets sent via the meta data socket "msock" | |
1732 | ||
1733 | sock msock | |
1734 | -----------------+-------------------------+------------------------------ | |
1735 | timeout conf.timeout / 2 conf.timeout / 2 | |
1736 | timeout action send a ping via msock Abort communication | |
1737 | and close all sockets | |
1738 | */ | |
1739 | ||
1740 | /* | |
1741 | * you must have down()ed the appropriate [m]sock_mutex elsewhere! | |
1742 | */ | |
bde89a9e | 1743 | int drbd_send(struct drbd_connection *connection, struct socket *sock, |
b411b363 PR |
1744 | void *buf, size_t size, unsigned msg_flags) |
1745 | { | |
1746 | struct kvec iov; | |
1747 | struct msghdr msg; | |
1748 | int rv, sent = 0; | |
1749 | ||
1750 | if (!sock) | |
c0d42c8e | 1751 | return -EBADR; |
b411b363 PR |
1752 | |
1753 | /* THINK if (signal_pending) return ... ? */ | |
1754 | ||
1755 | iov.iov_base = buf; | |
1756 | iov.iov_len = size; | |
1757 | ||
1758 | msg.msg_name = NULL; | |
1759 | msg.msg_namelen = 0; | |
1760 | msg.msg_control = NULL; | |
1761 | msg.msg_controllen = 0; | |
1762 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; | |
1763 | ||
bde89a9e | 1764 | if (sock == connection->data.socket) { |
44ed167d | 1765 | rcu_read_lock(); |
bde89a9e | 1766 | connection->ko_count = rcu_dereference(connection->net_conf)->ko_count; |
44ed167d | 1767 | rcu_read_unlock(); |
bde89a9e | 1768 | drbd_update_congested(connection); |
b411b363 PR |
1769 | } |
1770 | do { | |
1771 | /* STRANGE | |
1772 | * tcp_sendmsg does _not_ use its size parameter at all ? | |
1773 | * | |
1774 | * -EAGAIN on timeout, -EINTR on signal. | |
1775 | */ | |
1776 | /* THINK | |
1777 | * do we need to block DRBD_SIG if sock == &meta.socket ?? | |
1778 | * otherwise wake_asender() might interrupt some send_*Ack ! | |
1779 | */ | |
1780 | rv = kernel_sendmsg(sock, &msg, &iov, 1, size); | |
1781 | if (rv == -EAGAIN) { | |
bde89a9e | 1782 | if (we_should_drop_the_connection(connection, sock)) |
b411b363 PR |
1783 | break; |
1784 | else | |
1785 | continue; | |
1786 | } | |
b411b363 PR |
1787 | if (rv == -EINTR) { |
1788 | flush_signals(current); | |
1789 | rv = 0; | |
1790 | } | |
1791 | if (rv < 0) | |
1792 | break; | |
1793 | sent += rv; | |
1794 | iov.iov_base += rv; | |
1795 | iov.iov_len -= rv; | |
1796 | } while (sent < size); | |
1797 | ||
bde89a9e AG |
1798 | if (sock == connection->data.socket) |
1799 | clear_bit(NET_CONGESTED, &connection->flags); | |
b411b363 PR |
1800 | |
1801 | if (rv <= 0) { | |
1802 | if (rv != -EAGAIN) { | |
1ec861eb | 1803 | drbd_err(connection, "%s_sendmsg returned %d\n", |
bde89a9e | 1804 | sock == connection->meta.socket ? "msock" : "sock", |
bedbd2a5 | 1805 | rv); |
bde89a9e | 1806 | conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 1807 | } else |
bde89a9e | 1808 | conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); |
b411b363 PR |
1809 | } |
1810 | ||
1811 | return sent; | |
1812 | } | |
1813 | ||
fb708e40 AG |
1814 | /** |
1815 | * drbd_send_all - Send an entire buffer | |
1816 | * | |
1817 | * Returns 0 upon success and a negative error value otherwise. | |
1818 | */ | |
bde89a9e | 1819 | int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer, |
fb708e40 AG |
1820 | size_t size, unsigned msg_flags) |
1821 | { | |
1822 | int err; | |
1823 | ||
bde89a9e | 1824 | err = drbd_send(connection, sock, buffer, size, msg_flags); |
fb708e40 AG |
1825 | if (err < 0) |
1826 | return err; | |
1827 | if (err != size) | |
1828 | return -EIO; | |
1829 | return 0; | |
1830 | } | |
1831 | ||
b411b363 PR |
1832 | static int drbd_open(struct block_device *bdev, fmode_t mode) |
1833 | { | |
b30ab791 | 1834 | struct drbd_device *device = bdev->bd_disk->private_data; |
b411b363 PR |
1835 | unsigned long flags; |
1836 | int rv = 0; | |
1837 | ||
2a48fc0a | 1838 | mutex_lock(&drbd_main_mutex); |
a6b32bc3 | 1839 | spin_lock_irqsave(&first_peer_device(device)->connection->req_lock, flags); |
b30ab791 | 1840 | /* to have a stable device->state.role |
b411b363 PR |
1841 | * and no race with updating open_cnt */ |
1842 | ||
b30ab791 | 1843 | if (device->state.role != R_PRIMARY) { |
b411b363 PR |
1844 | if (mode & FMODE_WRITE) |
1845 | rv = -EROFS; | |
1846 | else if (!allow_oos) | |
1847 | rv = -EMEDIUMTYPE; | |
1848 | } | |
1849 | ||
1850 | if (!rv) | |
b30ab791 | 1851 | device->open_cnt++; |
a6b32bc3 | 1852 | spin_unlock_irqrestore(&first_peer_device(device)->connection->req_lock, flags); |
2a48fc0a | 1853 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1854 | |
1855 | return rv; | |
1856 | } | |
1857 | ||
db2a144b | 1858 | static void drbd_release(struct gendisk *gd, fmode_t mode) |
b411b363 | 1859 | { |
b30ab791 | 1860 | struct drbd_device *device = gd->private_data; |
2a48fc0a | 1861 | mutex_lock(&drbd_main_mutex); |
b30ab791 | 1862 | device->open_cnt--; |
2a48fc0a | 1863 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1864 | } |
1865 | ||
b30ab791 | 1866 | static void drbd_set_defaults(struct drbd_device *device) |
b411b363 | 1867 | { |
f399002e LE |
1868 | /* Beware! The actual layout differs |
1869 | * between big endian and little endian */ | |
b30ab791 | 1870 | device->state = (union drbd_dev_state) { |
b411b363 PR |
1871 | { .role = R_SECONDARY, |
1872 | .peer = R_UNKNOWN, | |
1873 | .conn = C_STANDALONE, | |
1874 | .disk = D_DISKLESS, | |
1875 | .pdsk = D_UNKNOWN, | |
b411b363 PR |
1876 | } }; |
1877 | } | |
1878 | ||
b30ab791 | 1879 | void drbd_init_set_defaults(struct drbd_device *device) |
b411b363 PR |
1880 | { |
1881 | /* the memset(,0,) did most of this. | |
1882 | * note: only assignments, no allocation in here */ | |
1883 | ||
b30ab791 AG |
1884 | drbd_set_defaults(device); |
1885 | ||
1886 | atomic_set(&device->ap_bio_cnt, 0); | |
1887 | atomic_set(&device->ap_pending_cnt, 0); | |
1888 | atomic_set(&device->rs_pending_cnt, 0); | |
1889 | atomic_set(&device->unacked_cnt, 0); | |
1890 | atomic_set(&device->local_cnt, 0); | |
1891 | atomic_set(&device->pp_in_use_by_net, 0); | |
1892 | atomic_set(&device->rs_sect_in, 0); | |
1893 | atomic_set(&device->rs_sect_ev, 0); | |
1894 | atomic_set(&device->ap_in_flight, 0); | |
1895 | atomic_set(&device->md_io_in_use, 0); | |
1896 | ||
1897 | mutex_init(&device->own_state_mutex); | |
1898 | device->state_mutex = &device->own_state_mutex; | |
1899 | ||
1900 | spin_lock_init(&device->al_lock); | |
1901 | spin_lock_init(&device->peer_seq_lock); | |
1902 | ||
1903 | INIT_LIST_HEAD(&device->active_ee); | |
1904 | INIT_LIST_HEAD(&device->sync_ee); | |
1905 | INIT_LIST_HEAD(&device->done_ee); | |
1906 | INIT_LIST_HEAD(&device->read_ee); | |
1907 | INIT_LIST_HEAD(&device->net_ee); | |
1908 | INIT_LIST_HEAD(&device->resync_reads); | |
1909 | INIT_LIST_HEAD(&device->resync_work.list); | |
1910 | INIT_LIST_HEAD(&device->unplug_work.list); | |
1911 | INIT_LIST_HEAD(&device->go_diskless.list); | |
1912 | INIT_LIST_HEAD(&device->md_sync_work.list); | |
1913 | INIT_LIST_HEAD(&device->start_resync_work.list); | |
1914 | INIT_LIST_HEAD(&device->bm_io_work.w.list); | |
1915 | ||
1916 | device->resync_work.cb = w_resync_timer; | |
1917 | device->unplug_work.cb = w_send_write_hint; | |
1918 | device->go_diskless.cb = w_go_diskless; | |
1919 | device->md_sync_work.cb = w_md_sync; | |
1920 | device->bm_io_work.w.cb = w_bitmap_io; | |
1921 | device->start_resync_work.cb = w_start_resync; | |
1922 | ||
1923 | device->resync_work.device = device; | |
1924 | device->unplug_work.device = device; | |
1925 | device->go_diskless.device = device; | |
1926 | device->md_sync_work.device = device; | |
1927 | device->bm_io_work.w.device = device; | |
1928 | device->start_resync_work.device = device; | |
1929 | ||
1930 | init_timer(&device->resync_timer); | |
1931 | init_timer(&device->md_sync_timer); | |
1932 | init_timer(&device->start_resync_timer); | |
1933 | init_timer(&device->request_timer); | |
1934 | device->resync_timer.function = resync_timer_fn; | |
1935 | device->resync_timer.data = (unsigned long) device; | |
1936 | device->md_sync_timer.function = md_sync_timer_fn; | |
1937 | device->md_sync_timer.data = (unsigned long) device; | |
1938 | device->start_resync_timer.function = start_resync_timer_fn; | |
1939 | device->start_resync_timer.data = (unsigned long) device; | |
1940 | device->request_timer.function = request_timer_fn; | |
1941 | device->request_timer.data = (unsigned long) device; | |
1942 | ||
1943 | init_waitqueue_head(&device->misc_wait); | |
1944 | init_waitqueue_head(&device->state_wait); | |
1945 | init_waitqueue_head(&device->ee_wait); | |
1946 | init_waitqueue_head(&device->al_wait); | |
1947 | init_waitqueue_head(&device->seq_wait); | |
1948 | ||
1949 | device->resync_wenr = LC_FREE; | |
1950 | device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1951 | device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1952 | } | |
1953 | ||
1954 | void drbd_device_cleanup(struct drbd_device *device) | |
b411b363 | 1955 | { |
1d7734a0 | 1956 | int i; |
a6b32bc3 | 1957 | if (first_peer_device(device)->connection->receiver.t_state != NONE) |
d0180171 | 1958 | drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n", |
a6b32bc3 | 1959 | first_peer_device(device)->connection->receiver.t_state); |
b30ab791 AG |
1960 | |
1961 | device->al_writ_cnt = | |
1962 | device->bm_writ_cnt = | |
1963 | device->read_cnt = | |
1964 | device->recv_cnt = | |
1965 | device->send_cnt = | |
1966 | device->writ_cnt = | |
1967 | device->p_size = | |
1968 | device->rs_start = | |
1969 | device->rs_total = | |
1970 | device->rs_failed = 0; | |
1971 | device->rs_last_events = 0; | |
1972 | device->rs_last_sect_ev = 0; | |
1d7734a0 | 1973 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1974 | device->rs_mark_left[i] = 0; |
1975 | device->rs_mark_time[i] = 0; | |
1d7734a0 | 1976 | } |
0b0ba1ef | 1977 | D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL); |
b411b363 | 1978 | |
b30ab791 AG |
1979 | drbd_set_my_capacity(device, 0); |
1980 | if (device->bitmap) { | |
b411b363 | 1981 | /* maybe never allocated. */ |
b30ab791 AG |
1982 | drbd_bm_resize(device, 0, 1); |
1983 | drbd_bm_cleanup(device); | |
b411b363 PR |
1984 | } |
1985 | ||
b30ab791 AG |
1986 | drbd_free_bc(device->ldev); |
1987 | device->ldev = NULL; | |
1d041225 | 1988 | |
b30ab791 | 1989 | clear_bit(AL_SUSPENDED, &device->flags); |
b411b363 | 1990 | |
0b0ba1ef AG |
1991 | D_ASSERT(device, list_empty(&device->active_ee)); |
1992 | D_ASSERT(device, list_empty(&device->sync_ee)); | |
1993 | D_ASSERT(device, list_empty(&device->done_ee)); | |
1994 | D_ASSERT(device, list_empty(&device->read_ee)); | |
1995 | D_ASSERT(device, list_empty(&device->net_ee)); | |
1996 | D_ASSERT(device, list_empty(&device->resync_reads)); | |
1997 | D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q)); | |
1998 | D_ASSERT(device, list_empty(&device->resync_work.list)); | |
1999 | D_ASSERT(device, list_empty(&device->unplug_work.list)); | |
2000 | D_ASSERT(device, list_empty(&device->go_diskless.list)); | |
2265b473 | 2001 | |
b30ab791 | 2002 | drbd_set_defaults(device); |
b411b363 PR |
2003 | } |
2004 | ||
2005 | ||
2006 | static void drbd_destroy_mempools(void) | |
2007 | { | |
2008 | struct page *page; | |
2009 | ||
2010 | while (drbd_pp_pool) { | |
2011 | page = drbd_pp_pool; | |
2012 | drbd_pp_pool = (struct page *)page_private(page); | |
2013 | __free_page(page); | |
2014 | drbd_pp_vacant--; | |
2015 | } | |
2016 | ||
0b0ba1ef | 2017 | /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */ |
b411b363 | 2018 | |
9476f39d LE |
2019 | if (drbd_md_io_bio_set) |
2020 | bioset_free(drbd_md_io_bio_set); | |
4281808f LE |
2021 | if (drbd_md_io_page_pool) |
2022 | mempool_destroy(drbd_md_io_page_pool); | |
b411b363 PR |
2023 | if (drbd_ee_mempool) |
2024 | mempool_destroy(drbd_ee_mempool); | |
2025 | if (drbd_request_mempool) | |
2026 | mempool_destroy(drbd_request_mempool); | |
2027 | if (drbd_ee_cache) | |
2028 | kmem_cache_destroy(drbd_ee_cache); | |
2029 | if (drbd_request_cache) | |
2030 | kmem_cache_destroy(drbd_request_cache); | |
2031 | if (drbd_bm_ext_cache) | |
2032 | kmem_cache_destroy(drbd_bm_ext_cache); | |
2033 | if (drbd_al_ext_cache) | |
2034 | kmem_cache_destroy(drbd_al_ext_cache); | |
2035 | ||
9476f39d | 2036 | drbd_md_io_bio_set = NULL; |
4281808f | 2037 | drbd_md_io_page_pool = NULL; |
b411b363 PR |
2038 | drbd_ee_mempool = NULL; |
2039 | drbd_request_mempool = NULL; | |
2040 | drbd_ee_cache = NULL; | |
2041 | drbd_request_cache = NULL; | |
2042 | drbd_bm_ext_cache = NULL; | |
2043 | drbd_al_ext_cache = NULL; | |
2044 | ||
2045 | return; | |
2046 | } | |
2047 | ||
2048 | static int drbd_create_mempools(void) | |
2049 | { | |
2050 | struct page *page; | |
1816a2b4 | 2051 | const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count; |
b411b363 PR |
2052 | int i; |
2053 | ||
2054 | /* prepare our caches and mempools */ | |
2055 | drbd_request_mempool = NULL; | |
2056 | drbd_ee_cache = NULL; | |
2057 | drbd_request_cache = NULL; | |
2058 | drbd_bm_ext_cache = NULL; | |
2059 | drbd_al_ext_cache = NULL; | |
2060 | drbd_pp_pool = NULL; | |
4281808f | 2061 | drbd_md_io_page_pool = NULL; |
9476f39d | 2062 | drbd_md_io_bio_set = NULL; |
b411b363 PR |
2063 | |
2064 | /* caches */ | |
2065 | drbd_request_cache = kmem_cache_create( | |
2066 | "drbd_req", sizeof(struct drbd_request), 0, 0, NULL); | |
2067 | if (drbd_request_cache == NULL) | |
2068 | goto Enomem; | |
2069 | ||
2070 | drbd_ee_cache = kmem_cache_create( | |
f6ffca9f | 2071 | "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL); |
b411b363 PR |
2072 | if (drbd_ee_cache == NULL) |
2073 | goto Enomem; | |
2074 | ||
2075 | drbd_bm_ext_cache = kmem_cache_create( | |
2076 | "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL); | |
2077 | if (drbd_bm_ext_cache == NULL) | |
2078 | goto Enomem; | |
2079 | ||
2080 | drbd_al_ext_cache = kmem_cache_create( | |
2081 | "drbd_al", sizeof(struct lc_element), 0, 0, NULL); | |
2082 | if (drbd_al_ext_cache == NULL) | |
2083 | goto Enomem; | |
2084 | ||
2085 | /* mempools */ | |
9476f39d LE |
2086 | drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0); |
2087 | if (drbd_md_io_bio_set == NULL) | |
2088 | goto Enomem; | |
9476f39d | 2089 | |
4281808f LE |
2090 | drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0); |
2091 | if (drbd_md_io_page_pool == NULL) | |
2092 | goto Enomem; | |
2093 | ||
b411b363 PR |
2094 | drbd_request_mempool = mempool_create(number, |
2095 | mempool_alloc_slab, mempool_free_slab, drbd_request_cache); | |
2096 | if (drbd_request_mempool == NULL) | |
2097 | goto Enomem; | |
2098 | ||
2099 | drbd_ee_mempool = mempool_create(number, | |
2100 | mempool_alloc_slab, mempool_free_slab, drbd_ee_cache); | |
2027ae1f | 2101 | if (drbd_ee_mempool == NULL) |
b411b363 PR |
2102 | goto Enomem; |
2103 | ||
2104 | /* drbd's page pool */ | |
2105 | spin_lock_init(&drbd_pp_lock); | |
2106 | ||
2107 | for (i = 0; i < number; i++) { | |
2108 | page = alloc_page(GFP_HIGHUSER); | |
2109 | if (!page) | |
2110 | goto Enomem; | |
2111 | set_page_private(page, (unsigned long)drbd_pp_pool); | |
2112 | drbd_pp_pool = page; | |
2113 | } | |
2114 | drbd_pp_vacant = number; | |
2115 | ||
2116 | return 0; | |
2117 | ||
2118 | Enomem: | |
2119 | drbd_destroy_mempools(); /* in case we allocated some */ | |
2120 | return -ENOMEM; | |
2121 | } | |
2122 | ||
2123 | static int drbd_notify_sys(struct notifier_block *this, unsigned long code, | |
2124 | void *unused) | |
2125 | { | |
2126 | /* just so we have it. you never know what interesting things we | |
2127 | * might want to do here some day... | |
2128 | */ | |
2129 | ||
2130 | return NOTIFY_DONE; | |
2131 | } | |
2132 | ||
2133 | static struct notifier_block drbd_notifier = { | |
2134 | .notifier_call = drbd_notify_sys, | |
2135 | }; | |
2136 | ||
b30ab791 | 2137 | static void drbd_release_all_peer_reqs(struct drbd_device *device) |
b411b363 PR |
2138 | { |
2139 | int rr; | |
2140 | ||
b30ab791 | 2141 | rr = drbd_free_peer_reqs(device, &device->active_ee); |
b411b363 | 2142 | if (rr) |
d0180171 | 2143 | drbd_err(device, "%d EEs in active list found!\n", rr); |
b411b363 | 2144 | |
b30ab791 | 2145 | rr = drbd_free_peer_reqs(device, &device->sync_ee); |
b411b363 | 2146 | if (rr) |
d0180171 | 2147 | drbd_err(device, "%d EEs in sync list found!\n", rr); |
b411b363 | 2148 | |
b30ab791 | 2149 | rr = drbd_free_peer_reqs(device, &device->read_ee); |
b411b363 | 2150 | if (rr) |
d0180171 | 2151 | drbd_err(device, "%d EEs in read list found!\n", rr); |
b411b363 | 2152 | |
b30ab791 | 2153 | rr = drbd_free_peer_reqs(device, &device->done_ee); |
b411b363 | 2154 | if (rr) |
d0180171 | 2155 | drbd_err(device, "%d EEs in done list found!\n", rr); |
b411b363 | 2156 | |
b30ab791 | 2157 | rr = drbd_free_peer_reqs(device, &device->net_ee); |
b411b363 | 2158 | if (rr) |
d0180171 | 2159 | drbd_err(device, "%d EEs in net list found!\n", rr); |
b411b363 PR |
2160 | } |
2161 | ||
774b3055 | 2162 | /* caution. no locking. */ |
05a10ec7 | 2163 | void drbd_destroy_device(struct kref *kref) |
b411b363 | 2164 | { |
b30ab791 | 2165 | struct drbd_device *device = container_of(kref, struct drbd_device, kref); |
803ea134 AG |
2166 | struct drbd_resource *resource = device->resource; |
2167 | struct drbd_connection *connection; | |
b411b363 | 2168 | |
b30ab791 | 2169 | del_timer_sync(&device->request_timer); |
dfa8bedb | 2170 | |
b411b363 | 2171 | /* paranoia asserts */ |
0b0ba1ef | 2172 | D_ASSERT(device, device->open_cnt == 0); |
b411b363 PR |
2173 | /* end paranoia asserts */ |
2174 | ||
b411b363 PR |
2175 | /* cleanup stuff that may have been allocated during |
2176 | * device (re-)configuration or state changes */ | |
2177 | ||
b30ab791 AG |
2178 | if (device->this_bdev) |
2179 | bdput(device->this_bdev); | |
b411b363 | 2180 | |
b30ab791 AG |
2181 | drbd_free_bc(device->ldev); |
2182 | device->ldev = NULL; | |
b411b363 | 2183 | |
b30ab791 | 2184 | drbd_release_all_peer_reqs(device); |
b411b363 | 2185 | |
b30ab791 AG |
2186 | lc_destroy(device->act_log); |
2187 | lc_destroy(device->resync); | |
b411b363 | 2188 | |
b30ab791 AG |
2189 | kfree(device->p_uuid); |
2190 | /* device->p_uuid = NULL; */ | |
b411b363 | 2191 | |
b30ab791 AG |
2192 | if (device->bitmap) /* should no longer be there. */ |
2193 | drbd_bm_cleanup(device); | |
2194 | __free_page(device->md_io_page); | |
2195 | put_disk(device->vdisk); | |
2196 | blk_cleanup_queue(device->rq_queue); | |
2197 | kfree(device->rs_plan_s); | |
a6b32bc3 | 2198 | kfree(first_peer_device(device)); |
b30ab791 | 2199 | kfree(device); |
9dc9fbb3 | 2200 | |
803ea134 AG |
2201 | for_each_connection(connection, resource) |
2202 | kref_put(&connection->kref, drbd_destroy_connection); | |
2203 | kref_put(&resource->kref, drbd_destroy_resource); | |
b411b363 PR |
2204 | } |
2205 | ||
2312f0b3 LE |
2206 | /* One global retry thread, if we need to push back some bio and have it |
2207 | * reinserted through our make request function. | |
2208 | */ | |
2209 | static struct retry_worker { | |
2210 | struct workqueue_struct *wq; | |
2211 | struct work_struct worker; | |
2212 | ||
2213 | spinlock_t lock; | |
2214 | struct list_head writes; | |
2215 | } retry; | |
2216 | ||
2217 | static void do_retry(struct work_struct *ws) | |
2218 | { | |
2219 | struct retry_worker *retry = container_of(ws, struct retry_worker, worker); | |
2220 | LIST_HEAD(writes); | |
2221 | struct drbd_request *req, *tmp; | |
2222 | ||
2223 | spin_lock_irq(&retry->lock); | |
2224 | list_splice_init(&retry->writes, &writes); | |
2225 | spin_unlock_irq(&retry->lock); | |
2226 | ||
2227 | list_for_each_entry_safe(req, tmp, &writes, tl_requests) { | |
b30ab791 | 2228 | struct drbd_device *device = req->w.device; |
2312f0b3 LE |
2229 | struct bio *bio = req->master_bio; |
2230 | unsigned long start_time = req->start_time; | |
9a278a79 LE |
2231 | bool expected; |
2232 | ||
2233 | expected = | |
2234 | expect(atomic_read(&req->completion_ref) == 0) && | |
2235 | expect(req->rq_state & RQ_POSTPONED) && | |
2236 | expect((req->rq_state & RQ_LOCAL_PENDING) == 0 || | |
2237 | (req->rq_state & RQ_LOCAL_ABORTED) != 0); | |
2238 | ||
2239 | if (!expected) | |
d0180171 | 2240 | drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n", |
9a278a79 LE |
2241 | req, atomic_read(&req->completion_ref), |
2242 | req->rq_state); | |
2243 | ||
2244 | /* We still need to put one kref associated with the | |
2245 | * "completion_ref" going zero in the code path that queued it | |
2246 | * here. The request object may still be referenced by a | |
2247 | * frozen local req->private_bio, in case we force-detached. | |
2312f0b3 | 2248 | */ |
9a278a79 | 2249 | kref_put(&req->kref, drbd_req_destroy); |
2312f0b3 LE |
2250 | |
2251 | /* A single suspended or otherwise blocking device may stall | |
2252 | * all others as well. Fortunately, this code path is to | |
2253 | * recover from a situation that "should not happen": | |
2254 | * concurrent writes in multi-primary setup. | |
2255 | * In a "normal" lifecycle, this workqueue is supposed to be | |
2256 | * destroyed without ever doing anything. | |
2257 | * If it turns out to be an issue anyways, we can do per | |
2258 | * resource (replication group) or per device (minor) retry | |
2259 | * workqueues instead. | |
2260 | */ | |
2261 | ||
2262 | /* We are not just doing generic_make_request(), | |
2263 | * as we want to keep the start_time information. */ | |
b30ab791 AG |
2264 | inc_ap_bio(device); |
2265 | __drbd_make_request(device, bio, start_time); | |
2312f0b3 LE |
2266 | } |
2267 | } | |
2268 | ||
9d05e7c4 | 2269 | void drbd_restart_request(struct drbd_request *req) |
2312f0b3 LE |
2270 | { |
2271 | unsigned long flags; | |
2272 | spin_lock_irqsave(&retry.lock, flags); | |
2273 | list_move_tail(&req->tl_requests, &retry.writes); | |
2274 | spin_unlock_irqrestore(&retry.lock, flags); | |
2275 | ||
2276 | /* Drop the extra reference that would otherwise | |
2277 | * have been dropped by complete_master_bio. | |
2278 | * do_retry() needs to grab a new one. */ | |
b30ab791 | 2279 | dec_ap_bio(req->w.device); |
b411b363 | 2280 | |
2312f0b3 | 2281 | queue_work(retry.wq, &retry.worker); |
b411b363 PR |
2282 | } |
2283 | ||
77c556f6 AG |
2284 | void drbd_destroy_resource(struct kref *kref) |
2285 | { | |
2286 | struct drbd_resource *resource = | |
2287 | container_of(kref, struct drbd_resource, kref); | |
2288 | ||
803ea134 | 2289 | idr_destroy(&resource->devices); |
77c556f6 AG |
2290 | kfree(resource->name); |
2291 | kfree(resource); | |
2292 | } | |
2293 | ||
2294 | void drbd_free_resource(struct drbd_resource *resource) | |
2295 | { | |
2296 | struct drbd_connection *connection, *tmp; | |
2297 | ||
2298 | for_each_connection_safe(connection, tmp, resource) { | |
2299 | list_del(&connection->connections); | |
2300 | kref_put(&connection->kref, drbd_destroy_connection); | |
2301 | } | |
2302 | kref_put(&resource->kref, drbd_destroy_resource); | |
2303 | } | |
2312f0b3 | 2304 | |
b411b363 PR |
2305 | static void drbd_cleanup(void) |
2306 | { | |
2307 | unsigned int i; | |
b30ab791 | 2308 | struct drbd_device *device; |
77c556f6 | 2309 | struct drbd_resource *resource, *tmp; |
b411b363 PR |
2310 | |
2311 | unregister_reboot_notifier(&drbd_notifier); | |
2312 | ||
17a93f30 LE |
2313 | /* first remove proc, |
2314 | * drbdsetup uses it's presence to detect | |
2315 | * whether DRBD is loaded. | |
2316 | * If we would get stuck in proc removal, | |
2317 | * but have netlink already deregistered, | |
2318 | * some drbdsetup commands may wait forever | |
2319 | * for an answer. | |
2320 | */ | |
2321 | if (drbd_proc) | |
2322 | remove_proc_entry("drbd", NULL); | |
2323 | ||
2312f0b3 LE |
2324 | if (retry.wq) |
2325 | destroy_workqueue(retry.wq); | |
b411b363 | 2326 | |
3b98c0c2 | 2327 | drbd_genl_unregister(); |
b411b363 | 2328 | |
803ea134 AG |
2329 | idr_for_each_entry(&drbd_devices, device, i) |
2330 | drbd_delete_minor(device); | |
b411b363 | 2331 | |
c141ebda | 2332 | /* not _rcu since, no other updater anymore. Genl already unregistered */ |
77c556f6 AG |
2333 | for_each_resource_safe(resource, tmp, &drbd_resources) { |
2334 | list_del(&resource->resources); | |
2335 | drbd_free_resource(resource); | |
81fa2e67 | 2336 | } |
b411b363 | 2337 | |
81a5d60e | 2338 | drbd_destroy_mempools(); |
b411b363 PR |
2339 | unregister_blkdev(DRBD_MAJOR, "drbd"); |
2340 | ||
05a10ec7 | 2341 | idr_destroy(&drbd_devices); |
81a5d60e | 2342 | |
b411b363 PR |
2343 | printk(KERN_INFO "drbd: module cleanup done.\n"); |
2344 | } | |
2345 | ||
2346 | /** | |
d97482ed | 2347 | * drbd_congested() - Callback for the flusher thread |
b411b363 | 2348 | * @congested_data: User data |
d97482ed | 2349 | * @bdi_bits: Bits the BDI flusher thread is currently interested in |
b411b363 PR |
2350 | * |
2351 | * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested. | |
2352 | */ | |
2353 | static int drbd_congested(void *congested_data, int bdi_bits) | |
2354 | { | |
b30ab791 | 2355 | struct drbd_device *device = congested_data; |
b411b363 PR |
2356 | struct request_queue *q; |
2357 | char reason = '-'; | |
2358 | int r = 0; | |
2359 | ||
b30ab791 | 2360 | if (!may_inc_ap_bio(device)) { |
b411b363 PR |
2361 | /* DRBD has frozen IO */ |
2362 | r = bdi_bits; | |
2363 | reason = 'd'; | |
2364 | goto out; | |
2365 | } | |
2366 | ||
a6b32bc3 | 2367 | if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) { |
c2ba686f LE |
2368 | r |= (1 << BDI_async_congested); |
2369 | /* Without good local data, we would need to read from remote, | |
2370 | * and that would need the worker thread as well, which is | |
2371 | * currently blocked waiting for that usermode helper to | |
2372 | * finish. | |
2373 | */ | |
b30ab791 | 2374 | if (!get_ldev_if_state(device, D_UP_TO_DATE)) |
c2ba686f LE |
2375 | r |= (1 << BDI_sync_congested); |
2376 | else | |
b30ab791 | 2377 | put_ldev(device); |
c2ba686f LE |
2378 | r &= bdi_bits; |
2379 | reason = 'c'; | |
2380 | goto out; | |
2381 | } | |
2382 | ||
b30ab791 AG |
2383 | if (get_ldev(device)) { |
2384 | q = bdev_get_queue(device->ldev->backing_bdev); | |
b411b363 | 2385 | r = bdi_congested(&q->backing_dev_info, bdi_bits); |
b30ab791 | 2386 | put_ldev(device); |
b411b363 PR |
2387 | if (r) |
2388 | reason = 'b'; | |
2389 | } | |
2390 | ||
a6b32bc3 AG |
2391 | if (bdi_bits & (1 << BDI_async_congested) && |
2392 | test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) { | |
b411b363 PR |
2393 | r |= (1 << BDI_async_congested); |
2394 | reason = reason == 'b' ? 'a' : 'n'; | |
2395 | } | |
2396 | ||
2397 | out: | |
b30ab791 | 2398 | device->congestion_reason = reason; |
b411b363 PR |
2399 | return r; |
2400 | } | |
2401 | ||
6699b655 PR |
2402 | static void drbd_init_workqueue(struct drbd_work_queue* wq) |
2403 | { | |
6699b655 PR |
2404 | spin_lock_init(&wq->q_lock); |
2405 | INIT_LIST_HEAD(&wq->q); | |
8c0785a5 | 2406 | init_waitqueue_head(&wq->q_wait); |
6699b655 PR |
2407 | } |
2408 | ||
4bc76048 | 2409 | struct drbd_resource *drbd_find_resource(const char *name) |
1aba4d7f | 2410 | { |
77c556f6 | 2411 | struct drbd_resource *resource; |
1aba4d7f | 2412 | |
3b98c0c2 LE |
2413 | if (!name || !name[0]) |
2414 | return NULL; | |
2415 | ||
c141ebda | 2416 | rcu_read_lock(); |
77c556f6 AG |
2417 | for_each_resource_rcu(resource, &drbd_resources) { |
2418 | if (!strcmp(resource->name, name)) { | |
4bc76048 | 2419 | kref_get(&resource->kref); |
1aba4d7f | 2420 | goto found; |
0ace9dfa | 2421 | } |
1aba4d7f | 2422 | } |
4bc76048 | 2423 | resource = NULL; |
1aba4d7f | 2424 | found: |
c141ebda | 2425 | rcu_read_unlock(); |
4bc76048 | 2426 | return resource; |
1aba4d7f PR |
2427 | } |
2428 | ||
bde89a9e | 2429 | struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len, |
089c075d AG |
2430 | void *peer_addr, int peer_addr_len) |
2431 | { | |
77c556f6 | 2432 | struct drbd_resource *resource; |
bde89a9e | 2433 | struct drbd_connection *connection; |
089c075d AG |
2434 | |
2435 | rcu_read_lock(); | |
77c556f6 AG |
2436 | for_each_resource_rcu(resource, &drbd_resources) { |
2437 | for_each_connection_rcu(connection, resource) { | |
2438 | if (connection->my_addr_len == my_addr_len && | |
2439 | connection->peer_addr_len == peer_addr_len && | |
2440 | !memcmp(&connection->my_addr, my_addr, my_addr_len) && | |
2441 | !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) { | |
2442 | kref_get(&connection->kref); | |
2443 | goto found; | |
2444 | } | |
089c075d AG |
2445 | } |
2446 | } | |
bde89a9e | 2447 | connection = NULL; |
089c075d AG |
2448 | found: |
2449 | rcu_read_unlock(); | |
bde89a9e | 2450 | return connection; |
089c075d AG |
2451 | } |
2452 | ||
e6ef8a5c AG |
2453 | static int drbd_alloc_socket(struct drbd_socket *socket) |
2454 | { | |
2455 | socket->rbuf = (void *) __get_free_page(GFP_KERNEL); | |
2456 | if (!socket->rbuf) | |
2457 | return -ENOMEM; | |
5a87d920 AG |
2458 | socket->sbuf = (void *) __get_free_page(GFP_KERNEL); |
2459 | if (!socket->sbuf) | |
2460 | return -ENOMEM; | |
e6ef8a5c AG |
2461 | return 0; |
2462 | } | |
2463 | ||
2464 | static void drbd_free_socket(struct drbd_socket *socket) | |
2465 | { | |
5a87d920 | 2466 | free_page((unsigned long) socket->sbuf); |
e6ef8a5c AG |
2467 | free_page((unsigned long) socket->rbuf); |
2468 | } | |
2469 | ||
bde89a9e | 2470 | void conn_free_crypto(struct drbd_connection *connection) |
91fd4dad | 2471 | { |
bde89a9e | 2472 | drbd_free_sock(connection); |
1d041225 | 2473 | |
bde89a9e AG |
2474 | crypto_free_hash(connection->csums_tfm); |
2475 | crypto_free_hash(connection->verify_tfm); | |
2476 | crypto_free_hash(connection->cram_hmac_tfm); | |
2477 | crypto_free_hash(connection->integrity_tfm); | |
2478 | crypto_free_hash(connection->peer_integrity_tfm); | |
2479 | kfree(connection->int_dig_in); | |
2480 | kfree(connection->int_dig_vv); | |
1d041225 | 2481 | |
bde89a9e AG |
2482 | connection->csums_tfm = NULL; |
2483 | connection->verify_tfm = NULL; | |
2484 | connection->cram_hmac_tfm = NULL; | |
2485 | connection->integrity_tfm = NULL; | |
2486 | connection->peer_integrity_tfm = NULL; | |
2487 | connection->int_dig_in = NULL; | |
2488 | connection->int_dig_vv = NULL; | |
91fd4dad PR |
2489 | } |
2490 | ||
eb6bea67 | 2491 | int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts) |
afbbfa88 | 2492 | { |
eb6bea67 | 2493 | struct drbd_connection *connection; |
afbbfa88 AG |
2494 | cpumask_var_t new_cpu_mask; |
2495 | int err; | |
2496 | ||
2497 | if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) | |
2498 | return -ENOMEM; | |
2499 | /* | |
2500 | retcode = ERR_NOMEM; | |
2501 | drbd_msg_put_info("unable to allocate cpumask"); | |
2502 | */ | |
2503 | ||
2504 | /* silently ignore cpu mask on UP kernel */ | |
2505 | if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) { | |
2506 | /* FIXME: Get rid of constant 32 here */ | |
c5b005ab PR |
2507 | err = bitmap_parse(res_opts->cpu_mask, 32, |
2508 | cpumask_bits(new_cpu_mask), nr_cpu_ids); | |
afbbfa88 | 2509 | if (err) { |
1ec861eb | 2510 | drbd_warn(resource, "bitmap_parse() failed with %d\n", err); |
afbbfa88 AG |
2511 | /* retcode = ERR_CPU_MASK_PARSE; */ |
2512 | goto fail; | |
2513 | } | |
2514 | } | |
eb6bea67 AG |
2515 | resource->res_opts = *res_opts; |
2516 | for_each_connection_rcu(connection, resource) { | |
2517 | if (!cpumask_equal(connection->cpu_mask, new_cpu_mask)) { | |
2518 | cpumask_copy(connection->cpu_mask, new_cpu_mask); | |
2519 | drbd_calc_cpu_mask(connection); | |
2520 | connection->receiver.reset_cpu_mask = 1; | |
2521 | connection->asender.reset_cpu_mask = 1; | |
2522 | connection->worker.reset_cpu_mask = 1; | |
2523 | } | |
afbbfa88 AG |
2524 | } |
2525 | err = 0; | |
2526 | ||
2527 | fail: | |
2528 | free_cpumask_var(new_cpu_mask); | |
2529 | return err; | |
2530 | ||
2531 | } | |
2532 | ||
77c556f6 AG |
2533 | struct drbd_resource *drbd_create_resource(const char *name) |
2534 | { | |
2535 | struct drbd_resource *resource; | |
2536 | ||
2537 | resource = kmalloc(sizeof(struct drbd_resource), GFP_KERNEL); | |
2538 | if (!resource) | |
2539 | return NULL; | |
2540 | resource->name = kstrdup(name, GFP_KERNEL); | |
2541 | if (!resource->name) { | |
2542 | kfree(resource); | |
2543 | return NULL; | |
2544 | } | |
2545 | kref_init(&resource->kref); | |
803ea134 | 2546 | idr_init(&resource->devices); |
77c556f6 AG |
2547 | INIT_LIST_HEAD(&resource->connections); |
2548 | list_add_tail_rcu(&resource->resources, &drbd_resources); | |
2549 | return resource; | |
2550 | } | |
2551 | ||
ec0bddbc | 2552 | /* caller must be under genl_lock() */ |
bde89a9e | 2553 | struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts) |
2111438b | 2554 | { |
77c556f6 | 2555 | struct drbd_resource *resource; |
bde89a9e | 2556 | struct drbd_connection *connection; |
2111438b | 2557 | |
bde89a9e AG |
2558 | connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL); |
2559 | if (!connection) | |
2111438b PR |
2560 | return NULL; |
2561 | ||
bde89a9e | 2562 | if (drbd_alloc_socket(&connection->data)) |
e6ef8a5c | 2563 | goto fail; |
bde89a9e | 2564 | if (drbd_alloc_socket(&connection->meta)) |
e6ef8a5c AG |
2565 | goto fail; |
2566 | ||
bde89a9e | 2567 | if (!zalloc_cpumask_var(&connection->cpu_mask, GFP_KERNEL)) |
774b3055 PR |
2568 | goto fail; |
2569 | ||
bde89a9e AG |
2570 | connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL); |
2571 | if (!connection->current_epoch) | |
12038a3a | 2572 | goto fail; |
b6dd1a89 | 2573 | |
bde89a9e | 2574 | INIT_LIST_HEAD(&connection->transfer_log); |
b6dd1a89 | 2575 | |
bde89a9e AG |
2576 | INIT_LIST_HEAD(&connection->current_epoch->list); |
2577 | connection->epochs = 1; | |
2578 | spin_lock_init(&connection->epoch_lock); | |
2579 | connection->write_ordering = WO_bdev_flush; | |
4b0007c0 | 2580 | |
bde89a9e AG |
2581 | connection->send.seen_any_write_yet = false; |
2582 | connection->send.current_epoch_nr = 0; | |
2583 | connection->send.current_epoch_writes = 0; | |
b6dd1a89 | 2584 | |
77c556f6 AG |
2585 | resource = drbd_create_resource(name); |
2586 | if (!resource) | |
2587 | goto fail; | |
2588 | ||
bde89a9e AG |
2589 | connection->cstate = C_STANDALONE; |
2590 | mutex_init(&connection->cstate_mutex); | |
2591 | spin_lock_init(&connection->req_lock); | |
2592 | mutex_init(&connection->conf_update); | |
2593 | init_waitqueue_head(&connection->ping_wait); | |
c06ece6b | 2594 | idr_init(&connection->peer_devices); |
b2fb6dbe | 2595 | |
bde89a9e AG |
2596 | drbd_init_workqueue(&connection->sender_work); |
2597 | mutex_init(&connection->data.mutex); | |
2598 | mutex_init(&connection->meta.mutex); | |
6699b655 | 2599 | |
bde89a9e AG |
2600 | drbd_thread_init(connection, &connection->receiver, drbdd_init, "receiver"); |
2601 | drbd_thread_init(connection, &connection->worker, drbd_worker, "worker"); | |
2602 | drbd_thread_init(connection, &connection->asender, drbd_asender, "asender"); | |
392c8801 | 2603 | |
bde89a9e | 2604 | kref_init(&connection->kref); |
77c556f6 | 2605 | |
77c556f6 | 2606 | connection->resource = resource; |
2111438b | 2607 | |
eb6bea67 AG |
2608 | if (set_resource_options(resource, res_opts)) |
2609 | goto fail_resource; | |
2610 | ||
2611 | kref_get(&resource->kref); | |
2612 | list_add_tail_rcu(&connection->connections, &resource->connections); | |
bde89a9e | 2613 | return connection; |
2111438b | 2614 | |
eb6bea67 AG |
2615 | fail_resource: |
2616 | list_del(&resource->resources); | |
2617 | drbd_free_resource(resource); | |
2111438b | 2618 | fail: |
bde89a9e AG |
2619 | kfree(connection->current_epoch); |
2620 | free_cpumask_var(connection->cpu_mask); | |
2621 | drbd_free_socket(&connection->meta); | |
2622 | drbd_free_socket(&connection->data); | |
bde89a9e | 2623 | kfree(connection); |
2111438b PR |
2624 | return NULL; |
2625 | } | |
2626 | ||
05a10ec7 | 2627 | void drbd_destroy_connection(struct kref *kref) |
2111438b | 2628 | { |
bde89a9e | 2629 | struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref); |
77c556f6 | 2630 | struct drbd_resource *resource = connection->resource; |
9dc9fbb3 | 2631 | |
bde89a9e | 2632 | if (atomic_read(&connection->current_epoch->epoch_size) != 0) |
1ec861eb | 2633 | drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size)); |
bde89a9e | 2634 | kfree(connection->current_epoch); |
12038a3a | 2635 | |
c06ece6b | 2636 | idr_destroy(&connection->peer_devices); |
2111438b | 2637 | |
bde89a9e AG |
2638 | free_cpumask_var(connection->cpu_mask); |
2639 | drbd_free_socket(&connection->meta); | |
2640 | drbd_free_socket(&connection->data); | |
bde89a9e AG |
2641 | kfree(connection->int_dig_in); |
2642 | kfree(connection->int_dig_vv); | |
2643 | kfree(connection); | |
77c556f6 | 2644 | kref_put(&resource->kref, drbd_destroy_resource); |
2111438b PR |
2645 | } |
2646 | ||
b30ab791 | 2647 | static int init_submitter(struct drbd_device *device) |
113fef9e LE |
2648 | { |
2649 | /* opencoded create_singlethread_workqueue(), | |
2650 | * to be able to say "drbd%d", ..., minor */ | |
b30ab791 AG |
2651 | device->submit.wq = alloc_workqueue("drbd%u_submit", |
2652 | WQ_UNBOUND | WQ_MEM_RECLAIM, 1, device->minor); | |
2653 | if (!device->submit.wq) | |
113fef9e LE |
2654 | return -ENOMEM; |
2655 | ||
b30ab791 AG |
2656 | INIT_WORK(&device->submit.worker, do_submit); |
2657 | spin_lock_init(&device->submit.lock); | |
2658 | INIT_LIST_HEAD(&device->submit.writes); | |
113fef9e LE |
2659 | return 0; |
2660 | } | |
2661 | ||
a6b32bc3 | 2662 | enum drbd_ret_code drbd_create_minor(struct drbd_connection *connection, unsigned int minor, int vnr) |
b411b363 | 2663 | { |
803ea134 | 2664 | struct drbd_resource *resource = connection->resource; |
b30ab791 | 2665 | struct drbd_device *device; |
a6b32bc3 | 2666 | struct drbd_peer_device *peer_device; |
b411b363 PR |
2667 | struct gendisk *disk; |
2668 | struct request_queue *q; | |
93e4bf7a | 2669 | int id; |
8432b314 | 2670 | enum drbd_ret_code err = ERR_NOMEM; |
774b3055 | 2671 | |
b30ab791 AG |
2672 | device = minor_to_device(minor); |
2673 | if (device) | |
774b3055 | 2674 | return ERR_MINOR_EXISTS; |
b411b363 PR |
2675 | |
2676 | /* GFP_KERNEL, we are outside of all write-out paths */ | |
b30ab791 AG |
2677 | device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL); |
2678 | if (!device) | |
774b3055 | 2679 | return ERR_NOMEM; |
803ea134 AG |
2680 | kref_init(&device->kref); |
2681 | ||
a6b32bc3 AG |
2682 | peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL); |
2683 | if (!peer_device) | |
2684 | goto out_no_peer_device; | |
774b3055 | 2685 | |
a6b32bc3 AG |
2686 | INIT_LIST_HEAD(&device->peer_devices); |
2687 | list_add(&peer_device->peer_devices, &device->peer_devices); | |
803ea134 AG |
2688 | kref_get(&resource->kref); |
2689 | device->resource = resource; | |
bde89a9e | 2690 | kref_get(&connection->kref); |
a6b32bc3 AG |
2691 | peer_device->connection = connection; |
2692 | peer_device->device = device; | |
b411b363 | 2693 | |
b30ab791 AG |
2694 | device->minor = minor; |
2695 | device->vnr = vnr; | |
b411b363 | 2696 | |
b30ab791 | 2697 | drbd_init_set_defaults(device); |
b411b363 PR |
2698 | |
2699 | q = blk_alloc_queue(GFP_KERNEL); | |
2700 | if (!q) | |
2701 | goto out_no_q; | |
b30ab791 AG |
2702 | device->rq_queue = q; |
2703 | q->queuedata = device; | |
b411b363 PR |
2704 | |
2705 | disk = alloc_disk(1); | |
2706 | if (!disk) | |
2707 | goto out_no_disk; | |
b30ab791 | 2708 | device->vdisk = disk; |
b411b363 | 2709 | |
81e84650 | 2710 | set_disk_ro(disk, true); |
b411b363 PR |
2711 | |
2712 | disk->queue = q; | |
2713 | disk->major = DRBD_MAJOR; | |
2714 | disk->first_minor = minor; | |
2715 | disk->fops = &drbd_ops; | |
2716 | sprintf(disk->disk_name, "drbd%d", minor); | |
b30ab791 | 2717 | disk->private_data = device; |
b411b363 | 2718 | |
b30ab791 | 2719 | device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor)); |
b411b363 | 2720 | /* we have no partitions. we contain only ourselves. */ |
b30ab791 | 2721 | device->this_bdev->bd_contains = device->this_bdev; |
b411b363 PR |
2722 | |
2723 | q->backing_dev_info.congested_fn = drbd_congested; | |
b30ab791 | 2724 | q->backing_dev_info.congested_data = device; |
b411b363 | 2725 | |
2f58dcfc | 2726 | blk_queue_make_request(q, drbd_make_request); |
a73ff323 | 2727 | blk_queue_flush(q, REQ_FLUSH | REQ_FUA); |
99432fcc PR |
2728 | /* Setting the max_hw_sectors to an odd value of 8kibyte here |
2729 | This triggers a max_bio_size message upon first attach or connect */ | |
2730 | blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8); | |
b411b363 PR |
2731 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
2732 | blk_queue_merge_bvec(q, drbd_merge_bvec); | |
803ea134 | 2733 | q->queue_lock = &connection->req_lock; |
b411b363 | 2734 | |
b30ab791 AG |
2735 | device->md_io_page = alloc_page(GFP_KERNEL); |
2736 | if (!device->md_io_page) | |
b411b363 PR |
2737 | goto out_no_io_page; |
2738 | ||
b30ab791 | 2739 | if (drbd_bm_init(device)) |
b411b363 | 2740 | goto out_no_bitmap; |
b30ab791 AG |
2741 | device->read_requests = RB_ROOT; |
2742 | device->write_requests = RB_ROOT; | |
b411b363 | 2743 | |
93e4bf7a AG |
2744 | id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL); |
2745 | if (id < 0) { | |
2746 | if (id == -ENOSPC) { | |
56de2102 TH |
2747 | err = ERR_MINOR_EXISTS; |
2748 | drbd_msg_put_info("requested minor exists already"); | |
2749 | } | |
8432b314 | 2750 | goto out_no_minor_idr; |
81a5d60e | 2751 | } |
803ea134 AG |
2752 | kref_get(&device->kref); |
2753 | ||
2754 | id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL); | |
2755 | if (id < 0) { | |
2756 | if (id == -ENOSPC) { | |
2757 | err = ERR_MINOR_EXISTS; | |
2758 | drbd_msg_put_info("requested minor exists already"); | |
2759 | } | |
2760 | goto out_idr_remove_minor; | |
2761 | } | |
2762 | kref_get(&device->kref); | |
8432b314 | 2763 | |
c06ece6b | 2764 | id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL); |
93e4bf7a AG |
2765 | if (id < 0) { |
2766 | if (id == -ENOSPC) { | |
56de2102 TH |
2767 | err = ERR_INVALID_REQUEST; |
2768 | drbd_msg_put_info("requested volume exists already"); | |
2769 | } | |
803ea134 | 2770 | goto out_idr_remove_from_resource; |
8432b314 | 2771 | } |
803ea134 | 2772 | kref_get(&device->kref); |
56de2102 | 2773 | |
b30ab791 | 2774 | if (init_submitter(device)) { |
113fef9e LE |
2775 | err = ERR_NOMEM; |
2776 | drbd_msg_put_info("unable to create submit workqueue"); | |
2777 | goto out_idr_remove_vol; | |
2778 | } | |
2779 | ||
774b3055 PR |
2780 | add_disk(disk); |
2781 | ||
2325eb66 | 2782 | /* inherit the connection state */ |
bde89a9e | 2783 | device->state.conn = connection->cstate; |
b30ab791 AG |
2784 | if (device->state.conn == C_WF_REPORT_PARAMS) |
2785 | drbd_connected(device); | |
2325eb66 | 2786 | |
774b3055 | 2787 | return NO_ERROR; |
b411b363 | 2788 | |
113fef9e | 2789 | out_idr_remove_vol: |
c06ece6b | 2790 | idr_remove(&connection->peer_devices, vnr); |
803ea134 AG |
2791 | out_idr_remove_from_resource: |
2792 | idr_remove(&resource->devices, vnr); | |
8432b314 | 2793 | out_idr_remove_minor: |
93e4bf7a | 2794 | idr_remove(&drbd_devices, minor); |
569083c0 | 2795 | synchronize_rcu(); |
8432b314 | 2796 | out_no_minor_idr: |
b30ab791 | 2797 | drbd_bm_cleanup(device); |
b411b363 | 2798 | out_no_bitmap: |
b30ab791 | 2799 | __free_page(device->md_io_page); |
b411b363 PR |
2800 | out_no_io_page: |
2801 | put_disk(disk); | |
2802 | out_no_disk: | |
2803 | blk_cleanup_queue(q); | |
2804 | out_no_q: | |
05a10ec7 | 2805 | kref_put(&connection->kref, drbd_destroy_connection); |
803ea134 | 2806 | kref_put(&resource->kref, drbd_destroy_resource); |
a6b32bc3 AG |
2807 | out_no_peer_device: |
2808 | kfree(device); | |
8432b314 | 2809 | return err; |
b411b363 PR |
2810 | } |
2811 | ||
803ea134 AG |
2812 | void drbd_delete_minor(struct drbd_device *device) |
2813 | { | |
2814 | struct drbd_resource *resource = device->resource; | |
2815 | struct drbd_connection *connection; | |
2816 | int refs = 3; | |
2817 | ||
2818 | for_each_connection(connection, resource) { | |
c06ece6b | 2819 | idr_remove(&connection->peer_devices, device->vnr); |
803ea134 AG |
2820 | refs++; |
2821 | } | |
2822 | idr_remove(&resource->devices, device->vnr); | |
2823 | idr_remove(&drbd_devices, device_to_minor(device)); | |
2824 | del_gendisk(device->vdisk); | |
2825 | synchronize_rcu(); | |
2826 | kref_sub(&device->kref, refs, drbd_destroy_device); | |
2827 | } | |
2828 | ||
b411b363 PR |
2829 | int __init drbd_init(void) |
2830 | { | |
2831 | int err; | |
2832 | ||
2b8a90b5 | 2833 | if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) { |
b411b363 | 2834 | printk(KERN_ERR |
81a5d60e | 2835 | "drbd: invalid minor_count (%d)\n", minor_count); |
b411b363 PR |
2836 | #ifdef MODULE |
2837 | return -EINVAL; | |
2838 | #else | |
46530e85 | 2839 | minor_count = DRBD_MINOR_COUNT_DEF; |
b411b363 PR |
2840 | #endif |
2841 | } | |
2842 | ||
b411b363 PR |
2843 | err = register_blkdev(DRBD_MAJOR, "drbd"); |
2844 | if (err) { | |
2845 | printk(KERN_ERR | |
2846 | "drbd: unable to register block device major %d\n", | |
2847 | DRBD_MAJOR); | |
2848 | return err; | |
2849 | } | |
2850 | ||
2851 | register_reboot_notifier(&drbd_notifier); | |
2852 | ||
2853 | /* | |
2854 | * allocate all necessary structs | |
2855 | */ | |
b411b363 PR |
2856 | init_waitqueue_head(&drbd_pp_wait); |
2857 | ||
2858 | drbd_proc = NULL; /* play safe for drbd_cleanup */ | |
05a10ec7 | 2859 | idr_init(&drbd_devices); |
b411b363 | 2860 | |
69babf05 | 2861 | rwlock_init(&global_state_lock); |
77c556f6 | 2862 | INIT_LIST_HEAD(&drbd_resources); |
69babf05 LE |
2863 | |
2864 | err = drbd_genl_register(); | |
2865 | if (err) { | |
2866 | printk(KERN_ERR "drbd: unable to register generic netlink family\n"); | |
2867 | goto fail; | |
2868 | } | |
2869 | ||
b411b363 PR |
2870 | err = drbd_create_mempools(); |
2871 | if (err) | |
3b98c0c2 | 2872 | goto fail; |
b411b363 | 2873 | |
6110d70b | 2874 | err = -ENOMEM; |
8c484ee4 | 2875 | drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL); |
b411b363 PR |
2876 | if (!drbd_proc) { |
2877 | printk(KERN_ERR "drbd: unable to register proc file\n"); | |
3b98c0c2 | 2878 | goto fail; |
b411b363 PR |
2879 | } |
2880 | ||
2312f0b3 LE |
2881 | retry.wq = create_singlethread_workqueue("drbd-reissue"); |
2882 | if (!retry.wq) { | |
2883 | printk(KERN_ERR "drbd: unable to create retry workqueue\n"); | |
2884 | goto fail; | |
2885 | } | |
2886 | INIT_WORK(&retry.worker, do_retry); | |
2887 | spin_lock_init(&retry.lock); | |
2888 | INIT_LIST_HEAD(&retry.writes); | |
b411b363 PR |
2889 | |
2890 | printk(KERN_INFO "drbd: initialized. " | |
2891 | "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n", | |
2892 | API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX); | |
2893 | printk(KERN_INFO "drbd: %s\n", drbd_buildtag()); | |
2894 | printk(KERN_INFO "drbd: registered as block device major %d\n", | |
2895 | DRBD_MAJOR); | |
b411b363 PR |
2896 | |
2897 | return 0; /* Success! */ | |
2898 | ||
3b98c0c2 | 2899 | fail: |
b411b363 PR |
2900 | drbd_cleanup(); |
2901 | if (err == -ENOMEM) | |
b411b363 PR |
2902 | printk(KERN_ERR "drbd: ran out of memory\n"); |
2903 | else | |
2904 | printk(KERN_ERR "drbd: initialization failure\n"); | |
2905 | return err; | |
2906 | } | |
2907 | ||
2908 | void drbd_free_bc(struct drbd_backing_dev *ldev) | |
2909 | { | |
2910 | if (ldev == NULL) | |
2911 | return; | |
2912 | ||
e525fd89 TH |
2913 | blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
2914 | blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
b411b363 | 2915 | |
94ad0a10 | 2916 | kfree(ldev->disk_conf); |
b411b363 PR |
2917 | kfree(ldev); |
2918 | } | |
2919 | ||
bde89a9e | 2920 | void drbd_free_sock(struct drbd_connection *connection) |
b411b363 | 2921 | { |
bde89a9e AG |
2922 | if (connection->data.socket) { |
2923 | mutex_lock(&connection->data.mutex); | |
2924 | kernel_sock_shutdown(connection->data.socket, SHUT_RDWR); | |
2925 | sock_release(connection->data.socket); | |
2926 | connection->data.socket = NULL; | |
2927 | mutex_unlock(&connection->data.mutex); | |
b411b363 | 2928 | } |
bde89a9e AG |
2929 | if (connection->meta.socket) { |
2930 | mutex_lock(&connection->meta.mutex); | |
2931 | kernel_sock_shutdown(connection->meta.socket, SHUT_RDWR); | |
2932 | sock_release(connection->meta.socket); | |
2933 | connection->meta.socket = NULL; | |
2934 | mutex_unlock(&connection->meta.mutex); | |
b411b363 PR |
2935 | } |
2936 | } | |
2937 | ||
b411b363 | 2938 | /* meta data management */ |
b411b363 | 2939 | |
bde89a9e | 2940 | void conn_md_sync(struct drbd_connection *connection) |
b411b363 | 2941 | { |
c06ece6b | 2942 | struct drbd_peer_device *peer_device; |
19fffd7b | 2943 | int vnr; |
b411b363 | 2944 | |
19fffd7b | 2945 | rcu_read_lock(); |
c06ece6b AG |
2946 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
2947 | struct drbd_device *device = peer_device->device; | |
2948 | ||
b30ab791 | 2949 | kref_get(&device->kref); |
19fffd7b | 2950 | rcu_read_unlock(); |
b30ab791 | 2951 | drbd_md_sync(device); |
05a10ec7 | 2952 | kref_put(&device->kref, drbd_destroy_device); |
19fffd7b PR |
2953 | rcu_read_lock(); |
2954 | } | |
2955 | rcu_read_unlock(); | |
b411b363 PR |
2956 | } |
2957 | ||
ae8bf312 | 2958 | /* aligned 4kByte */ |
b411b363 | 2959 | struct meta_data_on_disk { |
cccac985 | 2960 | u64 la_size_sect; /* last agreed size. */ |
b411b363 PR |
2961 | u64 uuid[UI_SIZE]; /* UUIDs. */ |
2962 | u64 device_uuid; | |
2963 | u64 reserved_u64_1; | |
2964 | u32 flags; /* MDF */ | |
2965 | u32 magic; | |
2966 | u32 md_size_sect; | |
2967 | u32 al_offset; /* offset to this block */ | |
ae8bf312 | 2968 | u32 al_nr_extents; /* important for restoring the AL (userspace) */ |
f399002e | 2969 | /* `-- act_log->nr_elements <-- ldev->dc.al_extents */ |
b411b363 PR |
2970 | u32 bm_offset; /* offset to the bitmap, from here */ |
2971 | u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */ | |
99432fcc | 2972 | u32 la_peer_max_bio_size; /* last peer max_bio_size */ |
b411b363 | 2973 | |
3a4d4eb3 LE |
2974 | /* see al_tr_number_to_on_disk_sector() */ |
2975 | u32 al_stripes; | |
2976 | u32 al_stripe_size_4k; | |
2977 | ||
2978 | u8 reserved_u8[4096 - (7*8 + 10*4)]; | |
b411b363 PR |
2979 | } __packed; |
2980 | ||
d752b269 PR |
2981 | |
2982 | ||
b30ab791 | 2983 | void drbd_md_write(struct drbd_device *device, void *b) |
b411b363 | 2984 | { |
d752b269 | 2985 | struct meta_data_on_disk *buffer = b; |
b411b363 PR |
2986 | sector_t sector; |
2987 | int i; | |
2988 | ||
ae8bf312 | 2989 | memset(buffer, 0, sizeof(*buffer)); |
b411b363 | 2990 | |
b30ab791 | 2991 | buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev)); |
b411b363 | 2992 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
2993 | buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
2994 | buffer->flags = cpu_to_be32(device->ldev->md.flags); | |
d5d7ebd4 | 2995 | buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN); |
b411b363 | 2996 | |
b30ab791 AG |
2997 | buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect); |
2998 | buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset); | |
2999 | buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements); | |
b411b363 | 3000 | buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE); |
b30ab791 | 3001 | buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid); |
b411b363 | 3002 | |
b30ab791 AG |
3003 | buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset); |
3004 | buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size); | |
b411b363 | 3005 | |
b30ab791 AG |
3006 | buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes); |
3007 | buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k); | |
3a4d4eb3 | 3008 | |
0b0ba1ef | 3009 | D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset); |
b30ab791 | 3010 | sector = device->ldev->md.md_offset; |
b411b363 | 3011 | |
b30ab791 | 3012 | if (drbd_md_sync_page_io(device, device->ldev, sector, WRITE)) { |
b411b363 | 3013 | /* this was a try anyways ... */ |
d0180171 | 3014 | drbd_err(device, "meta data update failed!\n"); |
b30ab791 | 3015 | drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR); |
b411b363 | 3016 | } |
d752b269 PR |
3017 | } |
3018 | ||
3019 | /** | |
3020 | * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set | |
b30ab791 | 3021 | * @device: DRBD device. |
d752b269 | 3022 | */ |
b30ab791 | 3023 | void drbd_md_sync(struct drbd_device *device) |
d752b269 PR |
3024 | { |
3025 | struct meta_data_on_disk *buffer; | |
3026 | ||
3027 | /* Don't accidentally change the DRBD meta data layout. */ | |
3028 | BUILD_BUG_ON(UI_SIZE != 4); | |
3029 | BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096); | |
3030 | ||
b30ab791 | 3031 | del_timer(&device->md_sync_timer); |
d752b269 | 3032 | /* timer may be rearmed by drbd_md_mark_dirty() now. */ |
b30ab791 | 3033 | if (!test_and_clear_bit(MD_DIRTY, &device->flags)) |
d752b269 PR |
3034 | return; |
3035 | ||
3036 | /* We use here D_FAILED and not D_ATTACHING because we try to write | |
3037 | * metadata even if we detach due to a disk failure! */ | |
b30ab791 | 3038 | if (!get_ldev_if_state(device, D_FAILED)) |
d752b269 PR |
3039 | return; |
3040 | ||
b30ab791 | 3041 | buffer = drbd_md_get_buffer(device); |
d752b269 PR |
3042 | if (!buffer) |
3043 | goto out; | |
3044 | ||
b30ab791 | 3045 | drbd_md_write(device, buffer); |
b411b363 | 3046 | |
b30ab791 | 3047 | /* Update device->ldev->md.la_size_sect, |
b411b363 | 3048 | * since we updated it on metadata. */ |
b30ab791 | 3049 | device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev); |
b411b363 | 3050 | |
b30ab791 | 3051 | drbd_md_put_buffer(device); |
e1711731 | 3052 | out: |
b30ab791 | 3053 | put_ldev(device); |
b411b363 PR |
3054 | } |
3055 | ||
b30ab791 | 3056 | static int check_activity_log_stripe_size(struct drbd_device *device, |
3a4d4eb3 LE |
3057 | struct meta_data_on_disk *on_disk, |
3058 | struct drbd_md *in_core) | |
3059 | { | |
3060 | u32 al_stripes = be32_to_cpu(on_disk->al_stripes); | |
3061 | u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k); | |
3062 | u64 al_size_4k; | |
3063 | ||
3064 | /* both not set: default to old fixed size activity log */ | |
3065 | if (al_stripes == 0 && al_stripe_size_4k == 0) { | |
3066 | al_stripes = 1; | |
3067 | al_stripe_size_4k = MD_32kB_SECT/8; | |
3068 | } | |
3069 | ||
3070 | /* some paranoia plausibility checks */ | |
3071 | ||
3072 | /* we need both values to be set */ | |
3073 | if (al_stripes == 0 || al_stripe_size_4k == 0) | |
3074 | goto err; | |
3075 | ||
3076 | al_size_4k = (u64)al_stripes * al_stripe_size_4k; | |
3077 | ||
3078 | /* Upper limit of activity log area, to avoid potential overflow | |
3079 | * problems in al_tr_number_to_on_disk_sector(). As right now, more | |
3080 | * than 72 * 4k blocks total only increases the amount of history, | |
3081 | * limiting this arbitrarily to 16 GB is not a real limitation ;-) */ | |
3082 | if (al_size_4k > (16 * 1024 * 1024/4)) | |
3083 | goto err; | |
3084 | ||
3085 | /* Lower limit: we need at least 8 transaction slots (32kB) | |
3086 | * to not break existing setups */ | |
3087 | if (al_size_4k < MD_32kB_SECT/8) | |
3088 | goto err; | |
3089 | ||
3090 | in_core->al_stripe_size_4k = al_stripe_size_4k; | |
3091 | in_core->al_stripes = al_stripes; | |
3092 | in_core->al_size_4k = al_size_4k; | |
3093 | ||
3094 | return 0; | |
3095 | err: | |
d0180171 | 3096 | drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n", |
3a4d4eb3 LE |
3097 | al_stripes, al_stripe_size_4k); |
3098 | return -EINVAL; | |
3099 | } | |
3100 | ||
b30ab791 | 3101 | static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev) |
c04ccaa6 LE |
3102 | { |
3103 | sector_t capacity = drbd_get_capacity(bdev->md_bdev); | |
3104 | struct drbd_md *in_core = &bdev->md; | |
3105 | s32 on_disk_al_sect; | |
3106 | s32 on_disk_bm_sect; | |
3107 | ||
3108 | /* The on-disk size of the activity log, calculated from offsets, and | |
3109 | * the size of the activity log calculated from the stripe settings, | |
3110 | * should match. | |
3111 | * Though we could relax this a bit: it is ok, if the striped activity log | |
3112 | * fits in the available on-disk activity log size. | |
3113 | * Right now, that would break how resize is implemented. | |
3114 | * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware | |
3115 | * of possible unused padding space in the on disk layout. */ | |
3116 | if (in_core->al_offset < 0) { | |
3117 | if (in_core->bm_offset > in_core->al_offset) | |
3118 | goto err; | |
3119 | on_disk_al_sect = -in_core->al_offset; | |
3120 | on_disk_bm_sect = in_core->al_offset - in_core->bm_offset; | |
3121 | } else { | |
3122 | if (in_core->al_offset != MD_4kB_SECT) | |
3123 | goto err; | |
3124 | if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT) | |
3125 | goto err; | |
3126 | ||
3127 | on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT; | |
3128 | on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset; | |
3129 | } | |
3130 | ||
3131 | /* old fixed size meta data is exactly that: fixed. */ | |
3132 | if (in_core->meta_dev_idx >= 0) { | |
3133 | if (in_core->md_size_sect != MD_128MB_SECT | |
3134 | || in_core->al_offset != MD_4kB_SECT | |
3135 | || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT | |
3136 | || in_core->al_stripes != 1 | |
3137 | || in_core->al_stripe_size_4k != MD_32kB_SECT/8) | |
3138 | goto err; | |
3139 | } | |
3140 | ||
3141 | if (capacity < in_core->md_size_sect) | |
3142 | goto err; | |
3143 | if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev)) | |
3144 | goto err; | |
3145 | ||
3146 | /* should be aligned, and at least 32k */ | |
3147 | if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT)) | |
3148 | goto err; | |
3149 | ||
3150 | /* should fit (for now: exactly) into the available on-disk space; | |
3151 | * overflow prevention is in check_activity_log_stripe_size() above. */ | |
3152 | if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT) | |
3153 | goto err; | |
3154 | ||
3155 | /* again, should be aligned */ | |
3156 | if (in_core->bm_offset & 7) | |
3157 | goto err; | |
3158 | ||
3159 | /* FIXME check for device grow with flex external meta data? */ | |
3160 | ||
3161 | /* can the available bitmap space cover the last agreed device size? */ | |
3162 | if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512) | |
3163 | goto err; | |
3164 | ||
3165 | return 0; | |
3166 | ||
3167 | err: | |
d0180171 | 3168 | drbd_err(device, "meta data offsets don't make sense: idx=%d " |
c04ccaa6 LE |
3169 | "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, " |
3170 | "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n", | |
3171 | in_core->meta_dev_idx, | |
3172 | in_core->al_stripes, in_core->al_stripe_size_4k, | |
3173 | in_core->al_offset, in_core->bm_offset, in_core->md_size_sect, | |
3174 | (unsigned long long)in_core->la_size_sect, | |
3175 | (unsigned long long)capacity); | |
3176 | ||
3177 | return -EINVAL; | |
3178 | } | |
3179 | ||
3180 | ||
b411b363 PR |
3181 | /** |
3182 | * drbd_md_read() - Reads in the meta data super block | |
b30ab791 | 3183 | * @device: DRBD device. |
b411b363 PR |
3184 | * @bdev: Device from which the meta data should be read in. |
3185 | * | |
3a4d4eb3 | 3186 | * Return NO_ERROR on success, and an enum drbd_ret_code in case |
d5d7ebd4 | 3187 | * something goes wrong. |
3a4d4eb3 | 3188 | * |
c04ccaa6 | 3189 | * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS, |
b30ab791 | 3190 | * even before @bdev is assigned to @device->ldev. |
b411b363 | 3191 | */ |
b30ab791 | 3192 | int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev) |
b411b363 PR |
3193 | { |
3194 | struct meta_data_on_disk *buffer; | |
d5d7ebd4 | 3195 | u32 magic, flags; |
b411b363 PR |
3196 | int i, rv = NO_ERROR; |
3197 | ||
b30ab791 | 3198 | if (device->state.disk != D_DISKLESS) |
c04ccaa6 | 3199 | return ERR_DISK_CONFIGURED; |
b411b363 | 3200 | |
b30ab791 | 3201 | buffer = drbd_md_get_buffer(device); |
e1711731 | 3202 | if (!buffer) |
c04ccaa6 | 3203 | return ERR_NOMEM; |
b411b363 | 3204 | |
c04ccaa6 LE |
3205 | /* First, figure out where our meta data superblock is located, |
3206 | * and read it. */ | |
3a4d4eb3 LE |
3207 | bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx; |
3208 | bdev->md.md_offset = drbd_md_ss(bdev); | |
b411b363 | 3209 | |
b30ab791 | 3210 | if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, READ)) { |
25985edc | 3211 | /* NOTE: can't do normal error processing here as this is |
b411b363 | 3212 | called BEFORE disk is attached */ |
d0180171 | 3213 | drbd_err(device, "Error while reading metadata.\n"); |
b411b363 PR |
3214 | rv = ERR_IO_MD_DISK; |
3215 | goto err; | |
3216 | } | |
3217 | ||
d5d7ebd4 LE |
3218 | magic = be32_to_cpu(buffer->magic); |
3219 | flags = be32_to_cpu(buffer->flags); | |
3220 | if (magic == DRBD_MD_MAGIC_84_UNCLEAN || | |
3221 | (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) { | |
3222 | /* btw: that's Activity Log clean, not "all" clean. */ | |
d0180171 | 3223 | drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n"); |
d5d7ebd4 LE |
3224 | rv = ERR_MD_UNCLEAN; |
3225 | goto err; | |
3226 | } | |
3a4d4eb3 LE |
3227 | |
3228 | rv = ERR_MD_INVALID; | |
d5d7ebd4 | 3229 | if (magic != DRBD_MD_MAGIC_08) { |
43de7c85 | 3230 | if (magic == DRBD_MD_MAGIC_07) |
d0180171 | 3231 | drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n"); |
d5d7ebd4 | 3232 | else |
d0180171 | 3233 | drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n"); |
b411b363 PR |
3234 | goto err; |
3235 | } | |
3a4d4eb3 | 3236 | |
c04ccaa6 | 3237 | if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) { |
d0180171 | 3238 | drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n", |
c04ccaa6 | 3239 | be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE); |
b411b363 PR |
3240 | goto err; |
3241 | } | |
3a4d4eb3 | 3242 | |
c04ccaa6 LE |
3243 | |
3244 | /* convert to in_core endian */ | |
3245 | bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect); | |
3246 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
3247 | bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]); | |
3248 | bdev->md.flags = be32_to_cpu(buffer->flags); | |
3249 | bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid); | |
3250 | ||
3251 | bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect); | |
3252 | bdev->md.al_offset = be32_to_cpu(buffer->al_offset); | |
3253 | bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset); | |
3254 | ||
b30ab791 | 3255 | if (check_activity_log_stripe_size(device, buffer, &bdev->md)) |
b411b363 | 3256 | goto err; |
b30ab791 | 3257 | if (check_offsets_and_sizes(device, bdev)) |
c04ccaa6 LE |
3258 | goto err; |
3259 | ||
b411b363 | 3260 | if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) { |
d0180171 | 3261 | drbd_err(device, "unexpected bm_offset: %d (expected %d)\n", |
b411b363 | 3262 | be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset); |
b411b363 PR |
3263 | goto err; |
3264 | } | |
3265 | if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) { | |
d0180171 | 3266 | drbd_err(device, "unexpected md_size: %u (expected %u)\n", |
b411b363 | 3267 | be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect); |
b411b363 PR |
3268 | goto err; |
3269 | } | |
3270 | ||
3a4d4eb3 | 3271 | rv = NO_ERROR; |
b411b363 | 3272 | |
a6b32bc3 | 3273 | spin_lock_irq(&first_peer_device(device)->connection->req_lock); |
b30ab791 | 3274 | if (device->state.conn < C_CONNECTED) { |
db141b2f | 3275 | unsigned int peer; |
99432fcc | 3276 | peer = be32_to_cpu(buffer->la_peer_max_bio_size); |
db141b2f | 3277 | peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE); |
b30ab791 | 3278 | device->peer_max_bio_size = peer; |
99432fcc | 3279 | } |
a6b32bc3 | 3280 | spin_unlock_irq(&first_peer_device(device)->connection->req_lock); |
b411b363 PR |
3281 | |
3282 | err: | |
b30ab791 | 3283 | drbd_md_put_buffer(device); |
b411b363 PR |
3284 | |
3285 | return rv; | |
3286 | } | |
3287 | ||
3288 | /** | |
3289 | * drbd_md_mark_dirty() - Mark meta data super block as dirty | |
b30ab791 | 3290 | * @device: DRBD device. |
b411b363 PR |
3291 | * |
3292 | * Call this function if you change anything that should be written to | |
3293 | * the meta-data super block. This function sets MD_DIRTY, and starts a | |
3294 | * timer that ensures that within five seconds you have to call drbd_md_sync(). | |
3295 | */ | |
ca0e6098 | 3296 | #ifdef DEBUG |
b30ab791 | 3297 | void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func) |
ee15b038 | 3298 | { |
b30ab791 AG |
3299 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) { |
3300 | mod_timer(&device->md_sync_timer, jiffies + HZ); | |
3301 | device->last_md_mark_dirty.line = line; | |
3302 | device->last_md_mark_dirty.func = func; | |
ee15b038 LE |
3303 | } |
3304 | } | |
3305 | #else | |
b30ab791 | 3306 | void drbd_md_mark_dirty(struct drbd_device *device) |
b411b363 | 3307 | { |
b30ab791 AG |
3308 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) |
3309 | mod_timer(&device->md_sync_timer, jiffies + 5*HZ); | |
b411b363 | 3310 | } |
ee15b038 | 3311 | #endif |
b411b363 | 3312 | |
b30ab791 | 3313 | void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3314 | { |
3315 | int i; | |
3316 | ||
62b0da3a | 3317 | for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) |
b30ab791 | 3318 | device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i]; |
b411b363 PR |
3319 | } |
3320 | ||
b30ab791 | 3321 | void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 PR |
3322 | { |
3323 | if (idx == UI_CURRENT) { | |
b30ab791 | 3324 | if (device->state.role == R_PRIMARY) |
b411b363 PR |
3325 | val |= 1; |
3326 | else | |
3327 | val &= ~((u64)1); | |
3328 | ||
b30ab791 | 3329 | drbd_set_ed_uuid(device, val); |
b411b363 PR |
3330 | } |
3331 | ||
b30ab791 AG |
3332 | device->ldev->md.uuid[idx] = val; |
3333 | drbd_md_mark_dirty(device); | |
b411b363 PR |
3334 | } |
3335 | ||
b30ab791 | 3336 | void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
9f2247bb PR |
3337 | { |
3338 | unsigned long flags; | |
b30ab791 AG |
3339 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3340 | __drbd_uuid_set(device, idx, val); | |
3341 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
9f2247bb | 3342 | } |
b411b363 | 3343 | |
b30ab791 | 3344 | void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 | 3345 | { |
9f2247bb | 3346 | unsigned long flags; |
b30ab791 AG |
3347 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3348 | if (device->ldev->md.uuid[idx]) { | |
3349 | drbd_uuid_move_history(device); | |
3350 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx]; | |
b411b363 | 3351 | } |
b30ab791 AG |
3352 | __drbd_uuid_set(device, idx, val); |
3353 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
b411b363 PR |
3354 | } |
3355 | ||
3356 | /** | |
3357 | * drbd_uuid_new_current() - Creates a new current UUID | |
b30ab791 | 3358 | * @device: DRBD device. |
b411b363 PR |
3359 | * |
3360 | * Creates a new current UUID, and rotates the old current UUID into | |
3361 | * the bitmap slot. Causes an incremental resync upon next connect. | |
3362 | */ | |
b30ab791 | 3363 | void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3364 | { |
3365 | u64 val; | |
9f2247bb PR |
3366 | unsigned long long bm_uuid; |
3367 | ||
3368 | get_random_bytes(&val, sizeof(u64)); | |
3369 | ||
b30ab791 AG |
3370 | spin_lock_irq(&device->ldev->md.uuid_lock); |
3371 | bm_uuid = device->ldev->md.uuid[UI_BITMAP]; | |
62b0da3a LE |
3372 | |
3373 | if (bm_uuid) | |
d0180171 | 3374 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3375 | |
b30ab791 AG |
3376 | device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT]; |
3377 | __drbd_uuid_set(device, UI_CURRENT, val); | |
3378 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 3379 | |
b30ab791 | 3380 | drbd_print_uuids(device, "new current UUID"); |
aaa8e2b3 | 3381 | /* get it to stable storage _now_ */ |
b30ab791 | 3382 | drbd_md_sync(device); |
b411b363 PR |
3383 | } |
3384 | ||
b30ab791 | 3385 | void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local) |
b411b363 | 3386 | { |
9f2247bb | 3387 | unsigned long flags; |
b30ab791 | 3388 | if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0) |
b411b363 PR |
3389 | return; |
3390 | ||
b30ab791 | 3391 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
b411b363 | 3392 | if (val == 0) { |
b30ab791 AG |
3393 | drbd_uuid_move_history(device); |
3394 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; | |
3395 | device->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 3396 | } else { |
b30ab791 | 3397 | unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP]; |
62b0da3a | 3398 | if (bm_uuid) |
d0180171 | 3399 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3400 | |
b30ab791 | 3401 | device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1); |
b411b363 | 3402 | } |
b30ab791 | 3403 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); |
9f2247bb | 3404 | |
b30ab791 | 3405 | drbd_md_mark_dirty(device); |
b411b363 PR |
3406 | } |
3407 | ||
3408 | /** | |
3409 | * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3410 | * @device: DRBD device. |
b411b363 PR |
3411 | * |
3412 | * Sets all bits in the bitmap and writes the whole bitmap to stable storage. | |
3413 | */ | |
b30ab791 | 3414 | int drbd_bmio_set_n_write(struct drbd_device *device) |
b411b363 PR |
3415 | { |
3416 | int rv = -EIO; | |
3417 | ||
b30ab791 AG |
3418 | if (get_ldev_if_state(device, D_ATTACHING)) { |
3419 | drbd_md_set_flag(device, MDF_FULL_SYNC); | |
3420 | drbd_md_sync(device); | |
3421 | drbd_bm_set_all(device); | |
b411b363 | 3422 | |
b30ab791 | 3423 | rv = drbd_bm_write(device); |
b411b363 PR |
3424 | |
3425 | if (!rv) { | |
b30ab791 AG |
3426 | drbd_md_clear_flag(device, MDF_FULL_SYNC); |
3427 | drbd_md_sync(device); | |
b411b363 PR |
3428 | } |
3429 | ||
b30ab791 | 3430 | put_ldev(device); |
b411b363 PR |
3431 | } |
3432 | ||
3433 | return rv; | |
3434 | } | |
3435 | ||
3436 | /** | |
3437 | * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3438 | * @device: DRBD device. |
b411b363 PR |
3439 | * |
3440 | * Clears all bits in the bitmap and writes the whole bitmap to stable storage. | |
3441 | */ | |
b30ab791 | 3442 | int drbd_bmio_clear_n_write(struct drbd_device *device) |
b411b363 PR |
3443 | { |
3444 | int rv = -EIO; | |
3445 | ||
b30ab791 AG |
3446 | drbd_resume_al(device); |
3447 | if (get_ldev_if_state(device, D_ATTACHING)) { | |
3448 | drbd_bm_clear_all(device); | |
3449 | rv = drbd_bm_write(device); | |
3450 | put_ldev(device); | |
b411b363 PR |
3451 | } |
3452 | ||
3453 | return rv; | |
3454 | } | |
3455 | ||
99920dc5 | 3456 | static int w_bitmap_io(struct drbd_work *w, int unused) |
b411b363 PR |
3457 | { |
3458 | struct bm_io_work *work = container_of(w, struct bm_io_work, w); | |
b30ab791 | 3459 | struct drbd_device *device = w->device; |
02851e9f | 3460 | int rv = -EIO; |
b411b363 | 3461 | |
0b0ba1ef | 3462 | D_ASSERT(device, atomic_read(&device->ap_bio_cnt) == 0); |
b411b363 | 3463 | |
b30ab791 AG |
3464 | if (get_ldev(device)) { |
3465 | drbd_bm_lock(device, work->why, work->flags); | |
3466 | rv = work->io_fn(device); | |
3467 | drbd_bm_unlock(device); | |
3468 | put_ldev(device); | |
02851e9f | 3469 | } |
b411b363 | 3470 | |
b30ab791 AG |
3471 | clear_bit_unlock(BITMAP_IO, &device->flags); |
3472 | wake_up(&device->misc_wait); | |
b411b363 PR |
3473 | |
3474 | if (work->done) | |
b30ab791 | 3475 | work->done(device, rv); |
b411b363 | 3476 | |
b30ab791 | 3477 | clear_bit(BITMAP_IO_QUEUED, &device->flags); |
b411b363 | 3478 | work->why = NULL; |
20ceb2b2 | 3479 | work->flags = 0; |
b411b363 | 3480 | |
99920dc5 | 3481 | return 0; |
b411b363 PR |
3482 | } |
3483 | ||
b30ab791 | 3484 | void drbd_ldev_destroy(struct drbd_device *device) |
82f59cc6 | 3485 | { |
b30ab791 AG |
3486 | lc_destroy(device->resync); |
3487 | device->resync = NULL; | |
3488 | lc_destroy(device->act_log); | |
3489 | device->act_log = NULL; | |
82f59cc6 | 3490 | __no_warn(local, |
b30ab791 AG |
3491 | drbd_free_bc(device->ldev); |
3492 | device->ldev = NULL;); | |
82f59cc6 | 3493 | |
b30ab791 | 3494 | clear_bit(GO_DISKLESS, &device->flags); |
82f59cc6 LE |
3495 | } |
3496 | ||
99920dc5 | 3497 | static int w_go_diskless(struct drbd_work *w, int unused) |
e9e6f3ec | 3498 | { |
b30ab791 | 3499 | struct drbd_device *device = w->device; |
00d56944 | 3500 | |
0b0ba1ef | 3501 | D_ASSERT(device, device->state.disk == D_FAILED); |
9d282875 LE |
3502 | /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will |
3503 | * inc/dec it frequently. Once we are D_DISKLESS, no one will touch | |
82f59cc6 LE |
3504 | * the protected members anymore, though, so once put_ldev reaches zero |
3505 | * again, it will be safe to free them. */ | |
a2a3c74f LE |
3506 | |
3507 | /* Try to write changed bitmap pages, read errors may have just | |
3508 | * set some bits outside the area covered by the activity log. | |
3509 | * | |
3510 | * If we have an IO error during the bitmap writeout, | |
3511 | * we will want a full sync next time, just in case. | |
3512 | * (Do we want a specific meta data flag for this?) | |
3513 | * | |
3514 | * If that does not make it to stable storage either, | |
fd0017c1 PR |
3515 | * we cannot do anything about that anymore. |
3516 | * | |
3517 | * We still need to check if both bitmap and ldev are present, we may | |
3518 | * end up here after a failed attach, before ldev was even assigned. | |
3519 | */ | |
b30ab791 | 3520 | if (device->bitmap && device->ldev) { |
bb45185d PR |
3521 | /* An interrupted resync or similar is allowed to recounts bits |
3522 | * while we detach. | |
3523 | * Any modifications would not be expected anymore, though. | |
3524 | */ | |
b30ab791 | 3525 | if (drbd_bitmap_io_from_worker(device, drbd_bm_write, |
bb45185d | 3526 | "detach", BM_LOCKED_TEST_ALLOWED)) { |
b30ab791 AG |
3527 | if (test_bit(WAS_READ_ERROR, &device->flags)) { |
3528 | drbd_md_set_flag(device, MDF_FULL_SYNC); | |
3529 | drbd_md_sync(device); | |
a2a3c74f LE |
3530 | } |
3531 | } | |
3532 | } | |
3533 | ||
b30ab791 | 3534 | drbd_force_state(device, NS(disk, D_DISKLESS)); |
99920dc5 | 3535 | return 0; |
e9e6f3ec LE |
3536 | } |
3537 | ||
b411b363 PR |
3538 | /** |
3539 | * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap | |
b30ab791 | 3540 | * @device: DRBD device. |
b411b363 PR |
3541 | * @io_fn: IO callback to be called when bitmap IO is possible |
3542 | * @done: callback to be called after the bitmap IO was performed | |
3543 | * @why: Descriptive text of the reason for doing the IO | |
3544 | * | |
3545 | * While IO on the bitmap happens we freeze application IO thus we ensure | |
3546 | * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be | |
3547 | * called from worker context. It MUST NOT be used while a previous such | |
3548 | * work is still pending! | |
3549 | */ | |
b30ab791 | 3550 | void drbd_queue_bitmap_io(struct drbd_device *device, |
54761697 AG |
3551 | int (*io_fn)(struct drbd_device *), |
3552 | void (*done)(struct drbd_device *, int), | |
20ceb2b2 | 3553 | char *why, enum bm_flag flags) |
b411b363 | 3554 | { |
0b0ba1ef | 3555 | D_ASSERT(device, current == first_peer_device(device)->connection->worker.task); |
b411b363 | 3556 | |
0b0ba1ef AG |
3557 | D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags)); |
3558 | D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags)); | |
3559 | D_ASSERT(device, list_empty(&device->bm_io_work.w.list)); | |
b30ab791 | 3560 | if (device->bm_io_work.why) |
d0180171 | 3561 | drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n", |
b30ab791 | 3562 | why, device->bm_io_work.why); |
b411b363 | 3563 | |
b30ab791 AG |
3564 | device->bm_io_work.io_fn = io_fn; |
3565 | device->bm_io_work.done = done; | |
3566 | device->bm_io_work.why = why; | |
3567 | device->bm_io_work.flags = flags; | |
b411b363 | 3568 | |
a6b32bc3 | 3569 | spin_lock_irq(&first_peer_device(device)->connection->req_lock); |
b30ab791 AG |
3570 | set_bit(BITMAP_IO, &device->flags); |
3571 | if (atomic_read(&device->ap_bio_cnt) == 0) { | |
3572 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags)) | |
a6b32bc3 | 3573 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, &device->bm_io_work.w); |
b411b363 | 3574 | } |
a6b32bc3 | 3575 | spin_unlock_irq(&first_peer_device(device)->connection->req_lock); |
b411b363 PR |
3576 | } |
3577 | ||
3578 | /** | |
3579 | * drbd_bitmap_io() - Does an IO operation on the whole bitmap | |
b30ab791 | 3580 | * @device: DRBD device. |
b411b363 PR |
3581 | * @io_fn: IO callback to be called when bitmap IO is possible |
3582 | * @why: Descriptive text of the reason for doing the IO | |
3583 | * | |
3584 | * freezes application IO while that the actual IO operations runs. This | |
3585 | * functions MAY NOT be called from worker context. | |
3586 | */ | |
b30ab791 | 3587 | int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *), |
20ceb2b2 | 3588 | char *why, enum bm_flag flags) |
b411b363 PR |
3589 | { |
3590 | int rv; | |
3591 | ||
0b0ba1ef | 3592 | D_ASSERT(device, current != first_peer_device(device)->connection->worker.task); |
b411b363 | 3593 | |
20ceb2b2 | 3594 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
b30ab791 | 3595 | drbd_suspend_io(device); |
b411b363 | 3596 | |
b30ab791 AG |
3597 | drbd_bm_lock(device, why, flags); |
3598 | rv = io_fn(device); | |
3599 | drbd_bm_unlock(device); | |
b411b363 | 3600 | |
20ceb2b2 | 3601 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
b30ab791 | 3602 | drbd_resume_io(device); |
b411b363 PR |
3603 | |
3604 | return rv; | |
3605 | } | |
3606 | ||
b30ab791 | 3607 | void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3608 | { |
b30ab791 AG |
3609 | if ((device->ldev->md.flags & flag) != flag) { |
3610 | drbd_md_mark_dirty(device); | |
3611 | device->ldev->md.flags |= flag; | |
b411b363 PR |
3612 | } |
3613 | } | |
3614 | ||
b30ab791 | 3615 | void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3616 | { |
b30ab791 AG |
3617 | if ((device->ldev->md.flags & flag) != 0) { |
3618 | drbd_md_mark_dirty(device); | |
3619 | device->ldev->md.flags &= ~flag; | |
b411b363 PR |
3620 | } |
3621 | } | |
3622 | int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag) | |
3623 | { | |
3624 | return (bdev->md.flags & flag) != 0; | |
3625 | } | |
3626 | ||
3627 | static void md_sync_timer_fn(unsigned long data) | |
3628 | { | |
b30ab791 | 3629 | struct drbd_device *device = (struct drbd_device *) data; |
b411b363 | 3630 | |
b792b655 | 3631 | /* must not double-queue! */ |
b30ab791 | 3632 | if (list_empty(&device->md_sync_work.list)) |
a6b32bc3 | 3633 | drbd_queue_work_front(&first_peer_device(device)->connection->sender_work, &device->md_sync_work); |
b411b363 PR |
3634 | } |
3635 | ||
99920dc5 | 3636 | static int w_md_sync(struct drbd_work *w, int unused) |
b411b363 | 3637 | { |
b30ab791 | 3638 | struct drbd_device *device = w->device; |
00d56944 | 3639 | |
d0180171 | 3640 | drbd_warn(device, "md_sync_timer expired! Worker calls drbd_md_sync().\n"); |
ee15b038 | 3641 | #ifdef DEBUG |
d0180171 | 3642 | drbd_warn(device, "last md_mark_dirty: %s:%u\n", |
b30ab791 | 3643 | device->last_md_mark_dirty.func, device->last_md_mark_dirty.line); |
ee15b038 | 3644 | #endif |
b30ab791 | 3645 | drbd_md_sync(device); |
99920dc5 | 3646 | return 0; |
b411b363 PR |
3647 | } |
3648 | ||
d8763023 | 3649 | const char *cmdname(enum drbd_packet cmd) |
f2ad9063 AG |
3650 | { |
3651 | /* THINK may need to become several global tables | |
3652 | * when we want to support more than | |
3653 | * one PRO_VERSION */ | |
3654 | static const char *cmdnames[] = { | |
3655 | [P_DATA] = "Data", | |
3656 | [P_DATA_REPLY] = "DataReply", | |
3657 | [P_RS_DATA_REPLY] = "RSDataReply", | |
3658 | [P_BARRIER] = "Barrier", | |
3659 | [P_BITMAP] = "ReportBitMap", | |
3660 | [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", | |
3661 | [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", | |
3662 | [P_UNPLUG_REMOTE] = "UnplugRemote", | |
3663 | [P_DATA_REQUEST] = "DataRequest", | |
3664 | [P_RS_DATA_REQUEST] = "RSDataRequest", | |
3665 | [P_SYNC_PARAM] = "SyncParam", | |
3666 | [P_SYNC_PARAM89] = "SyncParam89", | |
3667 | [P_PROTOCOL] = "ReportProtocol", | |
3668 | [P_UUIDS] = "ReportUUIDs", | |
3669 | [P_SIZES] = "ReportSizes", | |
3670 | [P_STATE] = "ReportState", | |
3671 | [P_SYNC_UUID] = "ReportSyncUUID", | |
3672 | [P_AUTH_CHALLENGE] = "AuthChallenge", | |
3673 | [P_AUTH_RESPONSE] = "AuthResponse", | |
3674 | [P_PING] = "Ping", | |
3675 | [P_PING_ACK] = "PingAck", | |
3676 | [P_RECV_ACK] = "RecvAck", | |
3677 | [P_WRITE_ACK] = "WriteAck", | |
3678 | [P_RS_WRITE_ACK] = "RSWriteAck", | |
d4dabbe2 | 3679 | [P_SUPERSEDED] = "Superseded", |
f2ad9063 AG |
3680 | [P_NEG_ACK] = "NegAck", |
3681 | [P_NEG_DREPLY] = "NegDReply", | |
3682 | [P_NEG_RS_DREPLY] = "NegRSDReply", | |
3683 | [P_BARRIER_ACK] = "BarrierAck", | |
3684 | [P_STATE_CHG_REQ] = "StateChgRequest", | |
3685 | [P_STATE_CHG_REPLY] = "StateChgReply", | |
3686 | [P_OV_REQUEST] = "OVRequest", | |
3687 | [P_OV_REPLY] = "OVReply", | |
3688 | [P_OV_RESULT] = "OVResult", | |
3689 | [P_CSUM_RS_REQUEST] = "CsumRSRequest", | |
3690 | [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", | |
3691 | [P_COMPRESSED_BITMAP] = "CBitmap", | |
3692 | [P_DELAY_PROBE] = "DelayProbe", | |
3693 | [P_OUT_OF_SYNC] = "OutOfSync", | |
7be8da07 | 3694 | [P_RETRY_WRITE] = "RetryWrite", |
ae25b336 LE |
3695 | [P_RS_CANCEL] = "RSCancel", |
3696 | [P_CONN_ST_CHG_REQ] = "conn_st_chg_req", | |
3697 | [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply", | |
036b17ea PR |
3698 | [P_RETRY_WRITE] = "retry_write", |
3699 | [P_PROTOCOL_UPDATE] = "protocol_update", | |
ae25b336 LE |
3700 | |
3701 | /* enum drbd_packet, but not commands - obsoleted flags: | |
3702 | * P_MAY_IGNORE | |
3703 | * P_MAX_OPT_CMD | |
3704 | */ | |
f2ad9063 AG |
3705 | }; |
3706 | ||
ae25b336 | 3707 | /* too big for the array: 0xfffX */ |
e5d6f33a AG |
3708 | if (cmd == P_INITIAL_META) |
3709 | return "InitialMeta"; | |
3710 | if (cmd == P_INITIAL_DATA) | |
3711 | return "InitialData"; | |
6038178e AG |
3712 | if (cmd == P_CONNECTION_FEATURES) |
3713 | return "ConnectionFeatures"; | |
6e849ce8 | 3714 | if (cmd >= ARRAY_SIZE(cmdnames)) |
f2ad9063 AG |
3715 | return "Unknown"; |
3716 | return cmdnames[cmd]; | |
3717 | } | |
3718 | ||
7be8da07 AG |
3719 | /** |
3720 | * drbd_wait_misc - wait for a request to make progress | |
b30ab791 | 3721 | * @device: device associated with the request |
7be8da07 AG |
3722 | * @i: the struct drbd_interval embedded in struct drbd_request or |
3723 | * struct drbd_peer_request | |
3724 | */ | |
b30ab791 | 3725 | int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i) |
7be8da07 | 3726 | { |
44ed167d | 3727 | struct net_conf *nc; |
7be8da07 AG |
3728 | DEFINE_WAIT(wait); |
3729 | long timeout; | |
3730 | ||
44ed167d | 3731 | rcu_read_lock(); |
a6b32bc3 | 3732 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
44ed167d PR |
3733 | if (!nc) { |
3734 | rcu_read_unlock(); | |
7be8da07 | 3735 | return -ETIMEDOUT; |
44ed167d PR |
3736 | } |
3737 | timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT; | |
3738 | rcu_read_unlock(); | |
7be8da07 | 3739 | |
b30ab791 | 3740 | /* Indicate to wake up device->misc_wait on progress. */ |
7be8da07 | 3741 | i->waiting = true; |
b30ab791 | 3742 | prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE); |
a6b32bc3 | 3743 | spin_unlock_irq(&first_peer_device(device)->connection->req_lock); |
7be8da07 | 3744 | timeout = schedule_timeout(timeout); |
b30ab791 | 3745 | finish_wait(&device->misc_wait, &wait); |
a6b32bc3 | 3746 | spin_lock_irq(&first_peer_device(device)->connection->req_lock); |
b30ab791 | 3747 | if (!timeout || device->state.conn < C_CONNECTED) |
7be8da07 AG |
3748 | return -ETIMEDOUT; |
3749 | if (signal_pending(current)) | |
3750 | return -ERESTARTSYS; | |
3751 | return 0; | |
b411b363 PR |
3752 | } |
3753 | ||
3754 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
3755 | /* Fault insertion support including random number generator shamelessly | |
3756 | * stolen from kernel/rcutorture.c */ | |
3757 | struct fault_random_state { | |
3758 | unsigned long state; | |
3759 | unsigned long count; | |
3760 | }; | |
3761 | ||
3762 | #define FAULT_RANDOM_MULT 39916801 /* prime */ | |
3763 | #define FAULT_RANDOM_ADD 479001701 /* prime */ | |
3764 | #define FAULT_RANDOM_REFRESH 10000 | |
3765 | ||
3766 | /* | |
3767 | * Crude but fast random-number generator. Uses a linear congruential | |
3768 | * generator, with occasional help from get_random_bytes(). | |
3769 | */ | |
3770 | static unsigned long | |
3771 | _drbd_fault_random(struct fault_random_state *rsp) | |
3772 | { | |
3773 | long refresh; | |
3774 | ||
49829ea7 | 3775 | if (!rsp->count--) { |
b411b363 PR |
3776 | get_random_bytes(&refresh, sizeof(refresh)); |
3777 | rsp->state += refresh; | |
3778 | rsp->count = FAULT_RANDOM_REFRESH; | |
3779 | } | |
3780 | rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD; | |
3781 | return swahw32(rsp->state); | |
3782 | } | |
3783 | ||
3784 | static char * | |
3785 | _drbd_fault_str(unsigned int type) { | |
3786 | static char *_faults[] = { | |
3787 | [DRBD_FAULT_MD_WR] = "Meta-data write", | |
3788 | [DRBD_FAULT_MD_RD] = "Meta-data read", | |
3789 | [DRBD_FAULT_RS_WR] = "Resync write", | |
3790 | [DRBD_FAULT_RS_RD] = "Resync read", | |
3791 | [DRBD_FAULT_DT_WR] = "Data write", | |
3792 | [DRBD_FAULT_DT_RD] = "Data read", | |
3793 | [DRBD_FAULT_DT_RA] = "Data read ahead", | |
3794 | [DRBD_FAULT_BM_ALLOC] = "BM allocation", | |
6b4388ac PR |
3795 | [DRBD_FAULT_AL_EE] = "EE allocation", |
3796 | [DRBD_FAULT_RECEIVE] = "receive data corruption", | |
b411b363 PR |
3797 | }; |
3798 | ||
3799 | return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**"; | |
3800 | } | |
3801 | ||
3802 | unsigned int | |
b30ab791 | 3803 | _drbd_insert_fault(struct drbd_device *device, unsigned int type) |
b411b363 PR |
3804 | { |
3805 | static struct fault_random_state rrs = {0, 0}; | |
3806 | ||
3807 | unsigned int ret = ( | |
3808 | (fault_devs == 0 || | |
b30ab791 | 3809 | ((1 << device_to_minor(device)) & fault_devs) != 0) && |
b411b363 PR |
3810 | (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate)); |
3811 | ||
3812 | if (ret) { | |
3813 | fault_count++; | |
3814 | ||
7383506c | 3815 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 3816 | drbd_warn(device, "***Simulating %s failure\n", |
b411b363 PR |
3817 | _drbd_fault_str(type)); |
3818 | } | |
3819 | ||
3820 | return ret; | |
3821 | } | |
3822 | #endif | |
3823 | ||
3824 | const char *drbd_buildtag(void) | |
3825 | { | |
3826 | /* DRBD built from external sources has here a reference to the | |
3827 | git hash of the source code. */ | |
3828 | ||
3829 | static char buildtag[38] = "\0uilt-in"; | |
3830 | ||
3831 | if (buildtag[0] == 0) { | |
bc4854bc CW |
3832 | #ifdef MODULE |
3833 | sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion); | |
3834 | #else | |
3835 | buildtag[0] = 'b'; | |
b411b363 | 3836 | #endif |
b411b363 PR |
3837 | } |
3838 | ||
3839 | return buildtag; | |
3840 | } | |
3841 | ||
3842 | module_init(drbd_init) | |
3843 | module_exit(drbd_cleanup) | |
3844 | ||
b411b363 PR |
3845 | EXPORT_SYMBOL(drbd_conn_str); |
3846 | EXPORT_SYMBOL(drbd_role_str); | |
3847 | EXPORT_SYMBOL(drbd_disk_str); | |
3848 | EXPORT_SYMBOL(drbd_set_st_err_str); |