]>
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 | ||
f88c5d90 LE |
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
30 | ||
b411b363 | 31 | #include <linux/module.h> |
e5f891b2 | 32 | #include <linux/jiffies.h> |
b411b363 PR |
33 | #include <linux/drbd.h> |
34 | #include <asm/uaccess.h> | |
35 | #include <asm/types.h> | |
36 | #include <net/sock.h> | |
37 | #include <linux/ctype.h> | |
2a48fc0a | 38 | #include <linux/mutex.h> |
b411b363 PR |
39 | #include <linux/fs.h> |
40 | #include <linux/file.h> | |
41 | #include <linux/proc_fs.h> | |
42 | #include <linux/init.h> | |
43 | #include <linux/mm.h> | |
44 | #include <linux/memcontrol.h> | |
45 | #include <linux/mm_inline.h> | |
46 | #include <linux/slab.h> | |
47 | #include <linux/random.h> | |
48 | #include <linux/reboot.h> | |
49 | #include <linux/notifier.h> | |
50 | #include <linux/kthread.h> | |
113fef9e | 51 | #include <linux/workqueue.h> |
b411b363 PR |
52 | #define __KERNEL_SYSCALLS__ |
53 | #include <linux/unistd.h> | |
54 | #include <linux/vmalloc.h> | |
55 | ||
56 | #include <linux/drbd_limits.h> | |
57 | #include "drbd_int.h" | |
a3603a6e | 58 | #include "drbd_protocol.h" |
b411b363 | 59 | #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */ |
b411b363 | 60 | #include "drbd_vli.h" |
4d3d5aa8 | 61 | #include "drbd_debugfs.h" |
b411b363 | 62 | |
2a48fc0a | 63 | static DEFINE_MUTEX(drbd_main_mutex); |
b411b363 | 64 | static int drbd_open(struct block_device *bdev, fmode_t mode); |
db2a144b | 65 | static void drbd_release(struct gendisk *gd, fmode_t mode); |
b411b363 | 66 | static void md_sync_timer_fn(unsigned long data); |
99920dc5 | 67 | static int w_bitmap_io(struct drbd_work *w, int unused); |
b411b363 | 68 | |
b411b363 PR |
69 | MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, " |
70 | "Lars Ellenberg <lars@linbit.com>"); | |
71 | MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION); | |
72 | MODULE_VERSION(REL_VERSION); | |
73 | MODULE_LICENSE("GPL"); | |
81a5d60e | 74 | MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices (" |
2b8a90b5 | 75 | __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")"); |
b411b363 PR |
76 | MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR); |
77 | ||
78 | #include <linux/moduleparam.h> | |
79 | /* allow_open_on_secondary */ | |
80 | MODULE_PARM_DESC(allow_oos, "DONT USE!"); | |
81 | /* thanks to these macros, if compiled into the kernel (not-module), | |
82 | * this becomes the boot parameter drbd.minor_count */ | |
83 | module_param(minor_count, uint, 0444); | |
84 | module_param(disable_sendpage, bool, 0644); | |
85 | module_param(allow_oos, bool, 0); | |
b411b363 PR |
86 | module_param(proc_details, int, 0644); |
87 | ||
88 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
89 | int enable_faults; | |
90 | int fault_rate; | |
91 | static int fault_count; | |
92 | int fault_devs; | |
93 | /* bitmap of enabled faults */ | |
94 | module_param(enable_faults, int, 0664); | |
95 | /* fault rate % value - applies to all enabled faults */ | |
96 | module_param(fault_rate, int, 0664); | |
97 | /* count of faults inserted */ | |
98 | module_param(fault_count, int, 0664); | |
99 | /* bitmap of devices to insert faults on */ | |
100 | module_param(fault_devs, int, 0644); | |
101 | #endif | |
102 | ||
103 | /* module parameter, defined */ | |
2b8a90b5 | 104 | unsigned int minor_count = DRBD_MINOR_COUNT_DEF; |
90ab5ee9 RR |
105 | bool disable_sendpage; |
106 | bool allow_oos; | |
b411b363 PR |
107 | int proc_details; /* Detail level in proc drbd*/ |
108 | ||
109 | /* Module parameter for setting the user mode helper program | |
110 | * to run. Default is /sbin/drbdadm */ | |
111 | char usermode_helper[80] = "/sbin/drbdadm"; | |
112 | ||
113 | module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644); | |
114 | ||
115 | /* in 2.6.x, our device mapping and config info contains our virtual gendisks | |
116 | * as member "struct gendisk *vdisk;" | |
117 | */ | |
05a10ec7 | 118 | struct idr drbd_devices; |
77c556f6 | 119 | struct list_head drbd_resources; |
28bc3b8c | 120 | struct mutex resources_mutex; |
b411b363 PR |
121 | |
122 | struct kmem_cache *drbd_request_cache; | |
6c852bec | 123 | struct kmem_cache *drbd_ee_cache; /* peer requests */ |
b411b363 PR |
124 | struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ |
125 | struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
126 | mempool_t *drbd_request_mempool; | |
127 | mempool_t *drbd_ee_mempool; | |
4281808f | 128 | mempool_t *drbd_md_io_page_pool; |
9476f39d | 129 | struct bio_set *drbd_md_io_bio_set; |
b411b363 PR |
130 | |
131 | /* I do not use a standard mempool, because: | |
132 | 1) I want to hand out the pre-allocated objects first. | |
133 | 2) I want to be able to interrupt sleeping allocation with a signal. | |
134 | Note: This is a single linked list, the next pointer is the private | |
135 | member of struct page. | |
136 | */ | |
137 | struct page *drbd_pp_pool; | |
138 | spinlock_t drbd_pp_lock; | |
139 | int drbd_pp_vacant; | |
140 | wait_queue_head_t drbd_pp_wait; | |
141 | ||
142 | DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5); | |
143 | ||
7d4e9d09 | 144 | static const struct block_device_operations drbd_ops = { |
b411b363 PR |
145 | .owner = THIS_MODULE, |
146 | .open = drbd_open, | |
147 | .release = drbd_release, | |
148 | }; | |
19f843aa | 149 | |
da4a75d2 LE |
150 | struct bio *bio_alloc_drbd(gfp_t gfp_mask) |
151 | { | |
152 | struct bio *bio; | |
19f843aa | 153 | |
da4a75d2 LE |
154 | if (!drbd_md_io_bio_set) |
155 | return bio_alloc(gfp_mask, 1); | |
19f843aa | 156 | |
da4a75d2 LE |
157 | bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set); |
158 | if (!bio) | |
159 | return NULL; | |
da4a75d2 | 160 | return bio; |
19f843aa LE |
161 | } |
162 | ||
b411b363 PR |
163 | #ifdef __CHECKER__ |
164 | /* When checking with sparse, and this is an inline function, sparse will | |
165 | give tons of false positives. When this is a real functions sparse works. | |
b411b363 | 166 | */ |
b30ab791 | 167 | int _get_ldev_if_state(struct drbd_device *device, enum drbd_disk_state mins) |
b411b363 | 168 | { |
b411b363 | 169 | int io_allowed; |
b411b363 | 170 | |
b30ab791 AG |
171 | atomic_inc(&device->local_cnt); |
172 | io_allowed = (device->state.disk >= mins); | |
b411b363 | 173 | if (!io_allowed) { |
b30ab791 AG |
174 | if (atomic_dec_and_test(&device->local_cnt)) |
175 | wake_up(&device->misc_wait); | |
b411b363 | 176 | } |
b411b363 PR |
177 | return io_allowed; |
178 | } | |
b411b363 | 179 | |
b411b363 | 180 | #endif |
265be2d0 | 181 | |
b411b363 | 182 | /** |
b6dd1a89 | 183 | * tl_release() - mark as BARRIER_ACKED all requests in the corresponding transfer log epoch |
bde89a9e | 184 | * @connection: DRBD connection. |
b411b363 PR |
185 | * @barrier_nr: Expected identifier of the DRBD write barrier packet. |
186 | * @set_size: Expected number of requests before that barrier. | |
187 | * | |
188 | * In case the passed barrier_nr or set_size does not match the oldest | |
b6dd1a89 LE |
189 | * epoch of not yet barrier-acked requests, this function will cause a |
190 | * termination of the connection. | |
b411b363 | 191 | */ |
bde89a9e | 192 | void tl_release(struct drbd_connection *connection, unsigned int barrier_nr, |
2f5cdd0b | 193 | unsigned int set_size) |
b411b363 | 194 | { |
b411b363 | 195 | struct drbd_request *r; |
b6dd1a89 LE |
196 | struct drbd_request *req = NULL; |
197 | int expect_epoch = 0; | |
198 | int expect_size = 0; | |
b411b363 | 199 | |
0500813f | 200 | spin_lock_irq(&connection->resource->req_lock); |
b411b363 | 201 | |
98683650 | 202 | /* find oldest not yet barrier-acked write request, |
b6dd1a89 | 203 | * count writes in its epoch. */ |
bde89a9e | 204 | list_for_each_entry(r, &connection->transfer_log, tl_requests) { |
a0d856df | 205 | const unsigned s = r->rq_state; |
b6dd1a89 LE |
206 | if (!req) { |
207 | if (!(s & RQ_WRITE)) | |
208 | continue; | |
209 | if (!(s & RQ_NET_MASK)) | |
210 | continue; | |
211 | if (s & RQ_NET_DONE) | |
212 | continue; | |
213 | req = r; | |
214 | expect_epoch = req->epoch; | |
215 | expect_size ++; | |
216 | } else { | |
217 | if (r->epoch != expect_epoch) | |
218 | break; | |
219 | if (!(s & RQ_WRITE)) | |
220 | continue; | |
221 | /* if (s & RQ_DONE): not expected */ | |
222 | /* if (!(s & RQ_NET_MASK)): not expected */ | |
223 | expect_size++; | |
43a5182c | 224 | } |
b411b363 | 225 | } |
67098930 | 226 | |
b411b363 | 227 | /* first some paranoia code */ |
b6dd1a89 | 228 | if (req == NULL) { |
1ec861eb | 229 | drbd_err(connection, "BAD! BarrierAck #%u received, but no epoch in tl!?\n", |
2f5cdd0b | 230 | barrier_nr); |
b411b363 | 231 | goto bail; |
b411b363 | 232 | } |
b6dd1a89 | 233 | if (expect_epoch != barrier_nr) { |
1ec861eb | 234 | drbd_err(connection, "BAD! BarrierAck #%u received, expected #%u!\n", |
b6dd1a89 | 235 | barrier_nr, expect_epoch); |
b411b363 | 236 | goto bail; |
5a22db89 LE |
237 | } |
238 | ||
b6dd1a89 | 239 | if (expect_size != set_size) { |
1ec861eb | 240 | drbd_err(connection, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n", |
b6dd1a89 | 241 | barrier_nr, set_size, expect_size); |
b411b363 | 242 | goto bail; |
b411b363 PR |
243 | } |
244 | ||
98683650 PR |
245 | /* Clean up list of requests processed during current epoch. */ |
246 | /* this extra list walk restart is paranoia, | |
247 | * to catch requests being barrier-acked "unexpectedly". | |
248 | * It usually should find the same req again, or some READ preceding it. */ | |
bde89a9e | 249 | list_for_each_entry(req, &connection->transfer_log, tl_requests) |
98683650 PR |
250 | if (req->epoch == expect_epoch) |
251 | break; | |
bde89a9e | 252 | list_for_each_entry_safe_from(req, r, &connection->transfer_log, tl_requests) { |
b6dd1a89 LE |
253 | if (req->epoch != expect_epoch) |
254 | break; | |
255 | _req_mod(req, BARRIER_ACKED); | |
19f843aa | 256 | } |
0500813f | 257 | spin_unlock_irq(&connection->resource->req_lock); |
19f843aa | 258 | |
b411b363 | 259 | return; |
b411b363 | 260 | |
b411b363 | 261 | bail: |
0500813f | 262 | spin_unlock_irq(&connection->resource->req_lock); |
bde89a9e | 263 | conn_request_state(connection, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
b411b363 | 264 | } |
b411b363 | 265 | |
b411b363 | 266 | |
b411b363 | 267 | /** |
11b58e73 | 268 | * _tl_restart() - Walks the transfer log, and applies an action to all requests |
e5f891b2 | 269 | * @connection: DRBD connection to operate on. |
11b58e73 | 270 | * @what: The action/event to perform with all request objects |
b411b363 | 271 | * |
8554df1c AG |
272 | * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO, |
273 | * RESTART_FROZEN_DISK_IO. | |
b411b363 | 274 | */ |
b6dd1a89 | 275 | /* must hold resource->req_lock */ |
bde89a9e | 276 | void _tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b411b363 | 277 | { |
b6dd1a89 | 278 | struct drbd_request *req, *r; |
b411b363 | 279 | |
bde89a9e | 280 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) |
b6dd1a89 LE |
281 | _req_mod(req, what); |
282 | } | |
738a84b2 | 283 | |
bde89a9e | 284 | void tl_restart(struct drbd_connection *connection, enum drbd_req_event what) |
b6dd1a89 | 285 | { |
0500813f | 286 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 287 | _tl_restart(connection, what); |
0500813f | 288 | spin_unlock_irq(&connection->resource->req_lock); |
cdfda633 | 289 | } |
b411b363 | 290 | |
b411b363 PR |
291 | /** |
292 | * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL | |
b30ab791 | 293 | * @device: DRBD device. |
b411b363 PR |
294 | * |
295 | * This is called after the connection to the peer was lost. The storage covered | |
296 | * by the requests on the transfer gets marked as our of sync. Called from the | |
297 | * receiver thread and the worker thread. | |
298 | */ | |
bde89a9e | 299 | void tl_clear(struct drbd_connection *connection) |
b411b363 | 300 | { |
bde89a9e | 301 | tl_restart(connection, CONNECTION_LOST_WHILE_PENDING); |
b411b363 | 302 | } |
197296ff | 303 | |
cdfda633 | 304 | /** |
b30ab791 AG |
305 | * tl_abort_disk_io() - Abort disk I/O for all requests for a certain device in the TL |
306 | * @device: DRBD device. | |
cdfda633 | 307 | */ |
b30ab791 | 308 | void tl_abort_disk_io(struct drbd_device *device) |
cdfda633 | 309 | { |
a6b32bc3 | 310 | struct drbd_connection *connection = first_peer_device(device)->connection; |
b6dd1a89 | 311 | struct drbd_request *req, *r; |
02851e9f | 312 | |
0500813f | 313 | spin_lock_irq(&connection->resource->req_lock); |
bde89a9e | 314 | list_for_each_entry_safe(req, r, &connection->transfer_log, tl_requests) { |
97ddb687 LE |
315 | if (!(req->rq_state & RQ_LOCAL_PENDING)) |
316 | continue; | |
84b8c06b | 317 | if (req->device != device) |
b6dd1a89 LE |
318 | continue; |
319 | _req_mod(req, ABORT_DISK_IO); | |
b411b363 | 320 | } |
0500813f | 321 | spin_unlock_irq(&connection->resource->req_lock); |
b411b363 PR |
322 | } |
323 | ||
b411b363 PR |
324 | static int drbd_thread_setup(void *arg) |
325 | { | |
326 | struct drbd_thread *thi = (struct drbd_thread *) arg; | |
2457b6d5 | 327 | struct drbd_resource *resource = thi->resource; |
b411b363 PR |
328 | unsigned long flags; |
329 | int retval; | |
330 | ||
f1b3a6ec | 331 | snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s", |
77c556f6 | 332 | thi->name[0], |
2457b6d5 | 333 | resource->name); |
f1b3a6ec | 334 | |
b411b363 PR |
335 | restart: |
336 | retval = thi->function(thi); | |
337 | ||
338 | spin_lock_irqsave(&thi->t_lock, flags); | |
339 | ||
e77a0a5c | 340 | /* if the receiver has been "EXITING", the last thing it did |
b411b363 PR |
341 | * was set the conn state to "StandAlone", |
342 | * if now a re-connect request comes in, conn state goes C_UNCONNECTED, | |
343 | * and receiver thread will be "started". | |
e77a0a5c | 344 | * drbd_thread_start needs to set "RESTARTING" in that case. |
b411b363 | 345 | * t_state check and assignment needs to be within the same spinlock, |
e77a0a5c AG |
346 | * so either thread_start sees EXITING, and can remap to RESTARTING, |
347 | * or thread_start see NONE, and can proceed as normal. | |
b411b363 PR |
348 | */ |
349 | ||
e77a0a5c | 350 | if (thi->t_state == RESTARTING) { |
2457b6d5 | 351 | drbd_info(resource, "Restarting %s thread\n", thi->name); |
e77a0a5c | 352 | thi->t_state = RUNNING; |
b411b363 PR |
353 | spin_unlock_irqrestore(&thi->t_lock, flags); |
354 | goto restart; | |
355 | } | |
356 | ||
357 | thi->task = NULL; | |
e77a0a5c | 358 | thi->t_state = NONE; |
b411b363 | 359 | smp_mb(); |
992d6e91 | 360 | complete_all(&thi->stop); |
b411b363 PR |
361 | spin_unlock_irqrestore(&thi->t_lock, flags); |
362 | ||
2457b6d5 | 363 | drbd_info(resource, "Terminating %s\n", current->comm); |
b411b363 PR |
364 | |
365 | /* Release mod reference taken when thread was started */ | |
9dc9fbb3 | 366 | |
2457b6d5 AG |
367 | if (thi->connection) |
368 | kref_put(&thi->connection->kref, drbd_destroy_connection); | |
369 | kref_put(&resource->kref, drbd_destroy_resource); | |
b411b363 PR |
370 | module_put(THIS_MODULE); |
371 | return retval; | |
372 | } | |
373 | ||
2457b6d5 | 374 | static void drbd_thread_init(struct drbd_resource *resource, struct drbd_thread *thi, |
c60b0251 | 375 | int (*func) (struct drbd_thread *), const char *name) |
b411b363 PR |
376 | { |
377 | spin_lock_init(&thi->t_lock); | |
378 | thi->task = NULL; | |
e77a0a5c | 379 | thi->t_state = NONE; |
b411b363 | 380 | thi->function = func; |
2457b6d5 AG |
381 | thi->resource = resource; |
382 | thi->connection = NULL; | |
c60b0251 | 383 | thi->name = name; |
b411b363 PR |
384 | } |
385 | ||
386 | int drbd_thread_start(struct drbd_thread *thi) | |
387 | { | |
2457b6d5 | 388 | struct drbd_resource *resource = thi->resource; |
b411b363 PR |
389 | struct task_struct *nt; |
390 | unsigned long flags; | |
391 | ||
b411b363 PR |
392 | /* is used from state engine doing drbd_thread_stop_nowait, |
393 | * while holding the req lock irqsave */ | |
394 | spin_lock_irqsave(&thi->t_lock, flags); | |
395 | ||
396 | switch (thi->t_state) { | |
e77a0a5c | 397 | case NONE: |
2457b6d5 | 398 | drbd_info(resource, "Starting %s thread (from %s [%d])\n", |
bed879ae | 399 | thi->name, current->comm, current->pid); |
b411b363 PR |
400 | |
401 | /* Get ref on module for thread - this is released when thread exits */ | |
402 | if (!try_module_get(THIS_MODULE)) { | |
2457b6d5 | 403 | drbd_err(resource, "Failed to get module reference in drbd_thread_start\n"); |
b411b363 | 404 | spin_unlock_irqrestore(&thi->t_lock, flags); |
81e84650 | 405 | return false; |
b411b363 PR |
406 | } |
407 | ||
2457b6d5 AG |
408 | kref_get(&resource->kref); |
409 | if (thi->connection) | |
410 | kref_get(&thi->connection->kref); | |
9dc9fbb3 | 411 | |
b411b363 | 412 | init_completion(&thi->stop); |
b411b363 | 413 | thi->reset_cpu_mask = 1; |
e77a0a5c | 414 | thi->t_state = RUNNING; |
b411b363 PR |
415 | spin_unlock_irqrestore(&thi->t_lock, flags); |
416 | flush_signals(current); /* otherw. may get -ERESTARTNOINTR */ | |
417 | ||
418 | nt = kthread_create(drbd_thread_setup, (void *) thi, | |
2457b6d5 | 419 | "drbd_%c_%s", thi->name[0], thi->resource->name); |
b411b363 PR |
420 | |
421 | if (IS_ERR(nt)) { | |
2457b6d5 | 422 | drbd_err(resource, "Couldn't start thread\n"); |
b411b363 | 423 | |
2457b6d5 AG |
424 | if (thi->connection) |
425 | kref_put(&thi->connection->kref, drbd_destroy_connection); | |
426 | kref_put(&resource->kref, drbd_destroy_resource); | |
b411b363 | 427 | module_put(THIS_MODULE); |
81e84650 | 428 | return false; |
b411b363 PR |
429 | } |
430 | spin_lock_irqsave(&thi->t_lock, flags); | |
431 | thi->task = nt; | |
e77a0a5c | 432 | thi->t_state = RUNNING; |
b411b363 PR |
433 | spin_unlock_irqrestore(&thi->t_lock, flags); |
434 | wake_up_process(nt); | |
435 | break; | |
e77a0a5c AG |
436 | case EXITING: |
437 | thi->t_state = RESTARTING; | |
2457b6d5 | 438 | drbd_info(resource, "Restarting %s thread (from %s [%d])\n", |
bed879ae | 439 | thi->name, current->comm, current->pid); |
b411b363 | 440 | /* fall through */ |
e77a0a5c AG |
441 | case RUNNING: |
442 | case RESTARTING: | |
b411b363 PR |
443 | default: |
444 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
445 | break; | |
446 | } | |
447 | ||
81e84650 | 448 | return true; |
b411b363 PR |
449 | } |
450 | ||
451 | ||
452 | void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait) | |
453 | { | |
454 | unsigned long flags; | |
455 | ||
e77a0a5c | 456 | enum drbd_thread_state ns = restart ? RESTARTING : EXITING; |
b411b363 PR |
457 | |
458 | /* may be called from state engine, holding the req lock irqsave */ | |
459 | spin_lock_irqsave(&thi->t_lock, flags); | |
460 | ||
e77a0a5c | 461 | if (thi->t_state == NONE) { |
b411b363 PR |
462 | spin_unlock_irqrestore(&thi->t_lock, flags); |
463 | if (restart) | |
464 | drbd_thread_start(thi); | |
465 | return; | |
466 | } | |
467 | ||
468 | if (thi->t_state != ns) { | |
469 | if (thi->task == NULL) { | |
470 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
471 | return; | |
472 | } | |
473 | ||
474 | thi->t_state = ns; | |
475 | smp_mb(); | |
476 | init_completion(&thi->stop); | |
477 | if (thi->task != current) | |
478 | force_sig(DRBD_SIGKILL, thi->task); | |
b411b363 PR |
479 | } |
480 | ||
481 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
482 | ||
483 | if (wait) | |
484 | wait_for_completion(&thi->stop); | |
485 | } | |
486 | ||
bde89a9e | 487 | int conn_lowest_minor(struct drbd_connection *connection) |
80822284 | 488 | { |
c06ece6b AG |
489 | struct drbd_peer_device *peer_device; |
490 | int vnr = 0, minor = -1; | |
774b3055 | 491 | |
695d08fa | 492 | rcu_read_lock(); |
c06ece6b AG |
493 | peer_device = idr_get_next(&connection->peer_devices, &vnr); |
494 | if (peer_device) | |
495 | minor = device_to_minor(peer_device->device); | |
695d08fa PR |
496 | rcu_read_unlock(); |
497 | ||
c06ece6b | 498 | return minor; |
80822284 | 499 | } |
774b3055 | 500 | |
b411b363 PR |
501 | #ifdef CONFIG_SMP |
502 | /** | |
503 | * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs | |
b411b363 | 504 | * |
625a6ba2 | 505 | * Forces all threads of a resource onto the same CPU. This is beneficial for |
b411b363 PR |
506 | * DRBD's performance. May be overwritten by user's configuration. |
507 | */ | |
625a6ba2 | 508 | static void drbd_calc_cpu_mask(cpumask_var_t *cpu_mask) |
b411b363 | 509 | { |
625a6ba2 | 510 | unsigned int *resources_per_cpu, min_index = ~0; |
b411b363 | 511 | |
625a6ba2 AG |
512 | resources_per_cpu = kzalloc(nr_cpu_ids * sizeof(*resources_per_cpu), GFP_KERNEL); |
513 | if (resources_per_cpu) { | |
514 | struct drbd_resource *resource; | |
515 | unsigned int cpu, min = ~0; | |
b411b363 | 516 | |
625a6ba2 AG |
517 | rcu_read_lock(); |
518 | for_each_resource_rcu(resource, &drbd_resources) { | |
519 | for_each_cpu(cpu, resource->cpu_mask) | |
520 | resources_per_cpu[cpu]++; | |
b411b363 | 521 | } |
625a6ba2 AG |
522 | rcu_read_unlock(); |
523 | for_each_online_cpu(cpu) { | |
524 | if (resources_per_cpu[cpu] < min) { | |
525 | min = resources_per_cpu[cpu]; | |
526 | min_index = cpu; | |
527 | } | |
528 | } | |
529 | kfree(resources_per_cpu); | |
530 | } | |
531 | if (min_index == ~0) { | |
532 | cpumask_setall(*cpu_mask); | |
533 | return; | |
b411b363 | 534 | } |
625a6ba2 | 535 | cpumask_set_cpu(min_index, *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 | 547 | { |
2457b6d5 | 548 | struct drbd_resource *resource = thi->resource; |
b411b363 | 549 | struct task_struct *p = current; |
bed879ae | 550 | |
b411b363 PR |
551 | if (!thi->reset_cpu_mask) |
552 | return; | |
553 | thi->reset_cpu_mask = 0; | |
2457b6d5 | 554 | set_cpus_allowed_ptr(p, resource->cpu_mask); |
b411b363 | 555 | } |
625a6ba2 AG |
556 | #else |
557 | #define drbd_calc_cpu_mask(A) ({}) | |
b411b363 PR |
558 | #endif |
559 | ||
52b061a4 AG |
560 | /** |
561 | * drbd_header_size - size of a packet header | |
562 | * | |
563 | * The header size is a multiple of 8, so any payload following the header is | |
564 | * word aligned on 64-bit architectures. (The bitmap send and receive code | |
565 | * relies on this.) | |
566 | */ | |
bde89a9e | 567 | unsigned int drbd_header_size(struct drbd_connection *connection) |
b411b363 | 568 | { |
bde89a9e | 569 | if (connection->agreed_pro_version >= 100) { |
0c8e36d9 AG |
570 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header100), 8)); |
571 | return sizeof(struct p_header100); | |
572 | } else { | |
573 | BUILD_BUG_ON(sizeof(struct p_header80) != | |
574 | sizeof(struct p_header95)); | |
575 | BUILD_BUG_ON(!IS_ALIGNED(sizeof(struct p_header80), 8)); | |
576 | return sizeof(struct p_header80); | |
577 | } | |
52b061a4 | 578 | } |
b411b363 | 579 | |
e658983a | 580 | static unsigned int prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
581 | { |
582 | h->magic = cpu_to_be32(DRBD_MAGIC); | |
583 | h->command = cpu_to_be16(cmd); | |
584 | h->length = cpu_to_be16(size); | |
e658983a | 585 | return sizeof(struct p_header80); |
fd340c12 | 586 | } |
b411b363 | 587 | |
e658983a | 588 | static unsigned int prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
589 | { |
590 | h->magic = cpu_to_be16(DRBD_MAGIC_BIG); | |
b411b363 | 591 | h->command = cpu_to_be16(cmd); |
b55d84ba | 592 | h->length = cpu_to_be32(size); |
e658983a | 593 | return sizeof(struct p_header95); |
fd340c12 | 594 | } |
b411b363 | 595 | |
0c8e36d9 AG |
596 | static unsigned int prepare_header100(struct p_header100 *h, enum drbd_packet cmd, |
597 | int size, int vnr) | |
598 | { | |
599 | h->magic = cpu_to_be32(DRBD_MAGIC_100); | |
600 | h->volume = cpu_to_be16(vnr); | |
601 | h->command = cpu_to_be16(cmd); | |
602 | h->length = cpu_to_be32(size); | |
603 | h->pad = 0; | |
604 | return sizeof(struct p_header100); | |
605 | } | |
b411b363 | 606 | |
bde89a9e | 607 | static unsigned int prepare_header(struct drbd_connection *connection, int vnr, |
0c8e36d9 | 608 | void *buffer, enum drbd_packet cmd, int size) |
d38e787e | 609 | { |
bde89a9e | 610 | if (connection->agreed_pro_version >= 100) |
0c8e36d9 | 611 | return prepare_header100(buffer, cmd, size, vnr); |
bde89a9e | 612 | else if (connection->agreed_pro_version >= 95 && |
0c8e36d9 | 613 | size > DRBD_MAX_SIZE_H80_PACKET) |
e658983a | 614 | return prepare_header95(buffer, cmd, size); |
d38e787e | 615 | else |
e658983a | 616 | return prepare_header80(buffer, cmd, size); |
b411b363 PR |
617 | } |
618 | ||
bde89a9e | 619 | static void *__conn_prepare_command(struct drbd_connection *connection, |
a7eb7bdf | 620 | struct drbd_socket *sock) |
b411b363 | 621 | { |
a7eb7bdf AG |
622 | if (!sock->socket) |
623 | return NULL; | |
bde89a9e | 624 | return sock->sbuf + drbd_header_size(connection); |
a7eb7bdf | 625 | } |
b411b363 | 626 | |
bde89a9e | 627 | void *conn_prepare_command(struct drbd_connection *connection, struct drbd_socket *sock) |
dba58587 | 628 | { |
a7eb7bdf | 629 | void *p; |
b411b363 | 630 | |
dba58587 | 631 | mutex_lock(&sock->mutex); |
bde89a9e | 632 | p = __conn_prepare_command(connection, sock); |
a7eb7bdf | 633 | if (!p) |
dba58587 | 634 | mutex_unlock(&sock->mutex); |
b411b363 | 635 | |
a7eb7bdf | 636 | return p; |
b411b363 PR |
637 | } |
638 | ||
69a22773 | 639 | void *drbd_prepare_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock) |
b411b363 | 640 | { |
69a22773 | 641 | return conn_prepare_command(peer_device->connection, sock); |
dba58587 | 642 | } |
b411b363 | 643 | |
bde89a9e | 644 | static int __send_command(struct drbd_connection *connection, int vnr, |
dba58587 AG |
645 | struct drbd_socket *sock, enum drbd_packet cmd, |
646 | unsigned int header_size, void *data, | |
647 | unsigned int size) | |
648 | { | |
649 | int msg_flags; | |
650 | int err; | |
b411b363 | 651 | |
dba58587 AG |
652 | /* |
653 | * Called with @data == NULL and the size of the data blocks in @size | |
654 | * for commands that send data blocks. For those commands, omit the | |
655 | * MSG_MORE flag: this will increase the likelihood that data blocks | |
656 | * which are page aligned on the sender will end up page aligned on the | |
657 | * receiver. | |
658 | */ | |
659 | msg_flags = data ? MSG_MORE : 0; | |
660 | ||
bde89a9e | 661 | header_size += prepare_header(connection, vnr, sock->sbuf, cmd, |
e658983a | 662 | header_size + size); |
bde89a9e | 663 | err = drbd_send_all(connection, sock->socket, sock->sbuf, header_size, |
dba58587 AG |
664 | msg_flags); |
665 | if (data && !err) | |
bde89a9e | 666 | err = drbd_send_all(connection, sock->socket, data, size, 0); |
123ff122 LE |
667 | /* DRBD protocol "pings" are latency critical. |
668 | * This is supposed to trigger tcp_push_pending_frames() */ | |
669 | if (!err && (cmd == P_PING || cmd == P_PING_ACK)) | |
670 | drbd_tcp_nodelay(sock->socket); | |
671 | ||
dba58587 AG |
672 | return err; |
673 | } | |
674 | ||
bde89a9e | 675 | static int __conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
a7eb7bdf AG |
676 | enum drbd_packet cmd, unsigned int header_size, |
677 | void *data, unsigned int size) | |
678 | { | |
bde89a9e | 679 | return __send_command(connection, 0, sock, cmd, header_size, data, size); |
a7eb7bdf AG |
680 | } |
681 | ||
bde89a9e | 682 | int conn_send_command(struct drbd_connection *connection, struct drbd_socket *sock, |
dba58587 AG |
683 | enum drbd_packet cmd, unsigned int header_size, |
684 | void *data, unsigned int size) | |
685 | { | |
686 | int err; | |
b411b363 | 687 | |
bde89a9e | 688 | err = __conn_send_command(connection, sock, cmd, header_size, data, size); |
dba58587 AG |
689 | mutex_unlock(&sock->mutex); |
690 | return err; | |
691 | } | |
692 | ||
69a22773 | 693 | int drbd_send_command(struct drbd_peer_device *peer_device, struct drbd_socket *sock, |
dba58587 AG |
694 | enum drbd_packet cmd, unsigned int header_size, |
695 | void *data, unsigned int size) | |
696 | { | |
697 | int err; | |
698 | ||
69a22773 AG |
699 | err = __send_command(peer_device->connection, peer_device->device->vnr, |
700 | sock, cmd, header_size, data, size); | |
dba58587 AG |
701 | mutex_unlock(&sock->mutex); |
702 | return err; | |
703 | } | |
b411b363 | 704 | |
bde89a9e | 705 | int drbd_send_ping(struct drbd_connection *connection) |
e307f352 | 706 | { |
9f5bdc33 AG |
707 | struct drbd_socket *sock; |
708 | ||
bde89a9e AG |
709 | sock = &connection->meta; |
710 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 711 | return -EIO; |
bde89a9e | 712 | return conn_send_command(connection, sock, P_PING, 0, NULL, 0); |
e307f352 | 713 | } |
b411b363 | 714 | |
bde89a9e | 715 | int drbd_send_ping_ack(struct drbd_connection *connection) |
e307f352 | 716 | { |
9f5bdc33 AG |
717 | struct drbd_socket *sock; |
718 | ||
bde89a9e AG |
719 | sock = &connection->meta; |
720 | if (!conn_prepare_command(connection, sock)) | |
9f5bdc33 | 721 | return -EIO; |
bde89a9e | 722 | return conn_send_command(connection, sock, P_PING_ACK, 0, NULL, 0); |
b411b363 PR |
723 | } |
724 | ||
69a22773 | 725 | int drbd_send_sync_param(struct drbd_peer_device *peer_device) |
b411b363 | 726 | { |
7c96715a | 727 | struct drbd_socket *sock; |
8e26f9cc | 728 | struct p_rs_param_95 *p; |
9f5bdc33 | 729 | int size; |
69a22773 | 730 | const int apv = peer_device->connection->agreed_pro_version; |
9f5bdc33 | 731 | enum drbd_packet cmd; |
44ed167d | 732 | struct net_conf *nc; |
daeda1cc | 733 | struct disk_conf *dc; |
9f5bdc33 | 734 | |
69a22773 AG |
735 | sock = &peer_device->connection->data; |
736 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
737 | if (!p) |
738 | return -EIO; | |
b411b363 | 739 | |
44ed167d | 740 | rcu_read_lock(); |
69a22773 | 741 | nc = rcu_dereference(peer_device->connection->net_conf); |
b411b363 PR |
742 | |
743 | size = apv <= 87 ? sizeof(struct p_rs_param) | |
744 | : apv == 88 ? sizeof(struct p_rs_param) | |
44ed167d | 745 | + strlen(nc->verify_alg) + 1 |
8e26f9cc PR |
746 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
747 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 748 | |
9f5bdc33 | 749 | cmd = apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM; |
b411b363 | 750 | |
9f5bdc33 AG |
751 | /* initialize verify_alg and csums_alg */ |
752 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
b411b363 | 753 | |
69a22773 AG |
754 | if (get_ldev(peer_device->device)) { |
755 | dc = rcu_dereference(peer_device->device->ldev->disk_conf); | |
6394b935 | 756 | p->resync_rate = cpu_to_be32(dc->resync_rate); |
daeda1cc PR |
757 | p->c_plan_ahead = cpu_to_be32(dc->c_plan_ahead); |
758 | p->c_delay_target = cpu_to_be32(dc->c_delay_target); | |
759 | p->c_fill_target = cpu_to_be32(dc->c_fill_target); | |
760 | p->c_max_rate = cpu_to_be32(dc->c_max_rate); | |
69a22773 | 761 | put_ldev(peer_device->device); |
9f5bdc33 | 762 | } else { |
6394b935 | 763 | p->resync_rate = cpu_to_be32(DRBD_RESYNC_RATE_DEF); |
9f5bdc33 AG |
764 | p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF); |
765 | p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF); | |
766 | p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF); | |
767 | p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF); | |
768 | } | |
b411b363 | 769 | |
9f5bdc33 | 770 | if (apv >= 88) |
44ed167d | 771 | strcpy(p->verify_alg, nc->verify_alg); |
9f5bdc33 | 772 | if (apv >= 89) |
44ed167d PR |
773 | strcpy(p->csums_alg, nc->csums_alg); |
774 | rcu_read_unlock(); | |
b411b363 | 775 | |
69a22773 | 776 | return drbd_send_command(peer_device, sock, cmd, size, NULL, 0); |
b411b363 PR |
777 | } |
778 | ||
bde89a9e | 779 | int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_packet cmd) |
b411b363 | 780 | { |
9f5bdc33 | 781 | struct drbd_socket *sock; |
b411b363 | 782 | struct p_protocol *p; |
44ed167d | 783 | struct net_conf *nc; |
9f5bdc33 | 784 | int size, cf; |
b411b363 | 785 | |
bde89a9e AG |
786 | sock = &connection->data; |
787 | p = __conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
788 | if (!p) |
789 | return -EIO; | |
b411b363 | 790 | |
44ed167d | 791 | rcu_read_lock(); |
bde89a9e | 792 | nc = rcu_dereference(connection->net_conf); |
b411b363 | 793 | |
bde89a9e | 794 | if (nc->tentative && connection->agreed_pro_version < 92) { |
44ed167d PR |
795 | rcu_read_unlock(); |
796 | mutex_unlock(&sock->mutex); | |
1ec861eb | 797 | drbd_err(connection, "--dry-run is not supported by peer"); |
44ed167d PR |
798 | return -EOPNOTSUPP; |
799 | } | |
b411b363 | 800 | |
9f5bdc33 | 801 | size = sizeof(*p); |
bde89a9e | 802 | if (connection->agreed_pro_version >= 87) |
44ed167d | 803 | size += strlen(nc->integrity_alg) + 1; |
b411b363 | 804 | |
44ed167d PR |
805 | p->protocol = cpu_to_be32(nc->wire_protocol); |
806 | p->after_sb_0p = cpu_to_be32(nc->after_sb_0p); | |
807 | p->after_sb_1p = cpu_to_be32(nc->after_sb_1p); | |
808 | p->after_sb_2p = cpu_to_be32(nc->after_sb_2p); | |
809 | p->two_primaries = cpu_to_be32(nc->two_primaries); | |
cf14c2e9 | 810 | cf = 0; |
6139f60d AG |
811 | if (nc->discard_my_data) |
812 | cf |= CF_DISCARD_MY_DATA; | |
6dff2902 | 813 | if (nc->tentative) |
9f5bdc33 | 814 | cf |= CF_DRY_RUN; |
cf14c2e9 PR |
815 | p->conn_flags = cpu_to_be32(cf); |
816 | ||
bde89a9e | 817 | if (connection->agreed_pro_version >= 87) |
44ed167d PR |
818 | strcpy(p->integrity_alg, nc->integrity_alg); |
819 | rcu_read_unlock(); | |
b411b363 | 820 | |
bde89a9e | 821 | return __conn_send_command(connection, sock, cmd, size, NULL, 0); |
a7eb7bdf AG |
822 | } |
823 | ||
bde89a9e | 824 | int drbd_send_protocol(struct drbd_connection *connection) |
a7eb7bdf AG |
825 | { |
826 | int err; | |
827 | ||
bde89a9e AG |
828 | mutex_lock(&connection->data.mutex); |
829 | err = __drbd_send_protocol(connection, P_PROTOCOL); | |
830 | mutex_unlock(&connection->data.mutex); | |
a7eb7bdf AG |
831 | |
832 | return err; | |
b411b363 PR |
833 | } |
834 | ||
69a22773 | 835 | static int _drbd_send_uuids(struct drbd_peer_device *peer_device, u64 uuid_flags) |
b411b363 | 836 | { |
69a22773 | 837 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
838 | struct drbd_socket *sock; |
839 | struct p_uuids *p; | |
b411b363 PR |
840 | int i; |
841 | ||
b30ab791 | 842 | if (!get_ldev_if_state(device, D_NEGOTIATING)) |
2ae5f95b | 843 | return 0; |
b411b363 | 844 | |
69a22773 AG |
845 | sock = &peer_device->connection->data; |
846 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 | 847 | if (!p) { |
b30ab791 | 848 | put_ldev(device); |
9f5bdc33 AG |
849 | return -EIO; |
850 | } | |
b30ab791 | 851 | spin_lock_irq(&device->ldev->md.uuid_lock); |
b411b363 | 852 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
853 | p->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
854 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 855 | |
b30ab791 AG |
856 | device->comm_bm_set = drbd_bm_total_weight(device); |
857 | p->uuid[UI_SIZE] = cpu_to_be64(device->comm_bm_set); | |
44ed167d | 858 | rcu_read_lock(); |
69a22773 | 859 | uuid_flags |= rcu_dereference(peer_device->connection->net_conf)->discard_my_data ? 1 : 0; |
44ed167d | 860 | rcu_read_unlock(); |
b30ab791 AG |
861 | uuid_flags |= test_bit(CRASHED_PRIMARY, &device->flags) ? 2 : 0; |
862 | uuid_flags |= device->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0; | |
9f5bdc33 | 863 | p->uuid[UI_FLAGS] = cpu_to_be64(uuid_flags); |
b411b363 | 864 | |
b30ab791 | 865 | put_ldev(device); |
69a22773 | 866 | return drbd_send_command(peer_device, sock, P_UUIDS, sizeof(*p), NULL, 0); |
b411b363 PR |
867 | } |
868 | ||
69a22773 | 869 | int drbd_send_uuids(struct drbd_peer_device *peer_device) |
b411b363 | 870 | { |
69a22773 | 871 | return _drbd_send_uuids(peer_device, 0); |
b411b363 PR |
872 | } |
873 | ||
69a22773 | 874 | int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device) |
b411b363 | 875 | { |
69a22773 | 876 | return _drbd_send_uuids(peer_device, 8); |
b411b363 PR |
877 | } |
878 | ||
b30ab791 | 879 | void drbd_print_uuids(struct drbd_device *device, const char *text) |
62b0da3a | 880 | { |
b30ab791 AG |
881 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
882 | u64 *uuid = device->ldev->md.uuid; | |
d0180171 | 883 | drbd_info(device, "%s %016llX:%016llX:%016llX:%016llX\n", |
62b0da3a LE |
884 | text, |
885 | (unsigned long long)uuid[UI_CURRENT], | |
886 | (unsigned long long)uuid[UI_BITMAP], | |
887 | (unsigned long long)uuid[UI_HISTORY_START], | |
888 | (unsigned long long)uuid[UI_HISTORY_END]); | |
b30ab791 | 889 | put_ldev(device); |
62b0da3a | 890 | } else { |
d0180171 | 891 | drbd_info(device, "%s effective data uuid: %016llX\n", |
62b0da3a | 892 | text, |
b30ab791 | 893 | (unsigned long long)device->ed_uuid); |
62b0da3a LE |
894 | } |
895 | } | |
896 | ||
69a22773 | 897 | void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *peer_device) |
b411b363 | 898 | { |
69a22773 | 899 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
900 | struct drbd_socket *sock; |
901 | struct p_rs_uuid *p; | |
5a22db89 LE |
902 | u64 uuid; |
903 | ||
0b0ba1ef | 904 | D_ASSERT(device, device->state.disk == D_UP_TO_DATE); |
b411b363 | 905 | |
b30ab791 | 906 | uuid = device->ldev->md.uuid[UI_BITMAP]; |
5ba3dac5 PR |
907 | if (uuid && uuid != UUID_JUST_CREATED) |
908 | uuid = uuid + UUID_NEW_BM_OFFSET; | |
909 | else | |
910 | get_random_bytes(&uuid, sizeof(u64)); | |
b30ab791 AG |
911 | drbd_uuid_set(device, UI_BITMAP, uuid); |
912 | drbd_print_uuids(device, "updated sync UUID"); | |
913 | drbd_md_sync(device); | |
b411b363 | 914 | |
69a22773 AG |
915 | sock = &peer_device->connection->data; |
916 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
917 | if (p) { |
918 | p->uuid = cpu_to_be64(uuid); | |
69a22773 | 919 | drbd_send_command(peer_device, sock, P_SYNC_UUID, sizeof(*p), NULL, 0); |
9f5bdc33 | 920 | } |
b411b363 PR |
921 | } |
922 | ||
69a22773 | 923 | int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags) |
b411b363 | 924 | { |
69a22773 | 925 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
926 | struct drbd_socket *sock; |
927 | struct p_sizes *p; | |
b411b363 | 928 | sector_t d_size, u_size; |
db141b2f LE |
929 | int q_order_type; |
930 | unsigned int max_bio_size; | |
b411b363 | 931 | |
b30ab791 | 932 | if (get_ldev_if_state(device, D_NEGOTIATING)) { |
0b0ba1ef | 933 | D_ASSERT(device, device->ldev->backing_bdev); |
b30ab791 | 934 | d_size = drbd_get_max_capacity(device->ldev); |
daeda1cc | 935 | rcu_read_lock(); |
b30ab791 | 936 | u_size = rcu_dereference(device->ldev->disk_conf)->disk_size; |
daeda1cc | 937 | rcu_read_unlock(); |
b30ab791 AG |
938 | q_order_type = drbd_queue_order_type(device); |
939 | max_bio_size = queue_max_hw_sectors(device->ldev->backing_bdev->bd_disk->queue) << 9; | |
db141b2f | 940 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE); |
b30ab791 | 941 | put_ldev(device); |
b411b363 PR |
942 | } else { |
943 | d_size = 0; | |
944 | u_size = 0; | |
945 | q_order_type = QUEUE_ORDERED_NONE; | |
99432fcc | 946 | max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */ |
b411b363 PR |
947 | } |
948 | ||
69a22773 AG |
949 | sock = &peer_device->connection->data; |
950 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
951 | if (!p) |
952 | return -EIO; | |
6809384c | 953 | |
69a22773 | 954 | if (peer_device->connection->agreed_pro_version <= 94) |
98683650 | 955 | max_bio_size = min(max_bio_size, DRBD_MAX_SIZE_H80_PACKET); |
69a22773 | 956 | else if (peer_device->connection->agreed_pro_version < 100) |
98683650 | 957 | max_bio_size = min(max_bio_size, DRBD_MAX_BIO_SIZE_P95); |
b411b363 | 958 | |
9f5bdc33 AG |
959 | p->d_size = cpu_to_be64(d_size); |
960 | p->u_size = cpu_to_be64(u_size); | |
b30ab791 | 961 | p->c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(device->this_bdev)); |
9f5bdc33 AG |
962 | p->max_bio_size = cpu_to_be32(max_bio_size); |
963 | p->queue_order_type = cpu_to_be16(q_order_type); | |
964 | p->dds_flags = cpu_to_be16(flags); | |
69a22773 | 965 | return drbd_send_command(peer_device, sock, P_SIZES, sizeof(*p), NULL, 0); |
b411b363 PR |
966 | } |
967 | ||
968 | /** | |
f479ea06 | 969 | * drbd_send_current_state() - Sends the drbd state to the peer |
69a22773 | 970 | * @peer_device: DRBD peer device. |
b411b363 | 971 | */ |
69a22773 | 972 | int drbd_send_current_state(struct drbd_peer_device *peer_device) |
b411b363 | 973 | { |
7c96715a | 974 | struct drbd_socket *sock; |
9f5bdc33 | 975 | struct p_state *p; |
b411b363 | 976 | |
69a22773 AG |
977 | sock = &peer_device->connection->data; |
978 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
979 | if (!p) |
980 | return -EIO; | |
69a22773 AG |
981 | p->state = cpu_to_be32(peer_device->device->state.i); /* Within the send mutex */ |
982 | return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0); | |
b411b363 PR |
983 | } |
984 | ||
f479ea06 LE |
985 | /** |
986 | * drbd_send_state() - After a state change, sends the new state to the peer | |
69a22773 | 987 | * @peer_device: DRBD peer device. |
43de7c85 | 988 | * @state: the state to send, not necessarily the current state. |
f479ea06 LE |
989 | * |
990 | * Each state change queues an "after_state_ch" work, which will eventually | |
991 | * send the resulting new state to the peer. If more state changes happen | |
992 | * between queuing and processing of the after_state_ch work, we still | |
993 | * want to send each intermediary state in the order it occurred. | |
994 | */ | |
69a22773 | 995 | int drbd_send_state(struct drbd_peer_device *peer_device, union drbd_state state) |
f479ea06 | 996 | { |
43de7c85 PR |
997 | struct drbd_socket *sock; |
998 | struct p_state *p; | |
f479ea06 | 999 | |
69a22773 AG |
1000 | sock = &peer_device->connection->data; |
1001 | p = drbd_prepare_command(peer_device, sock); | |
43de7c85 PR |
1002 | if (!p) |
1003 | return -EIO; | |
1004 | p->state = cpu_to_be32(state.i); /* Within the send mutex */ | |
69a22773 | 1005 | return drbd_send_command(peer_device, sock, P_STATE, sizeof(*p), NULL, 0); |
43de7c85 | 1006 | } |
f479ea06 | 1007 | |
69a22773 | 1008 | int drbd_send_state_req(struct drbd_peer_device *peer_device, union drbd_state mask, union drbd_state val) |
9f5bdc33 AG |
1009 | { |
1010 | struct drbd_socket *sock; | |
1011 | struct p_req_state *p; | |
f479ea06 | 1012 | |
69a22773 AG |
1013 | sock = &peer_device->connection->data; |
1014 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1015 | if (!p) |
1016 | return -EIO; | |
1017 | p->mask = cpu_to_be32(mask.i); | |
1018 | p->val = cpu_to_be32(val.i); | |
69a22773 | 1019 | return drbd_send_command(peer_device, sock, P_STATE_CHG_REQ, sizeof(*p), NULL, 0); |
b411b363 | 1020 | } |
f479ea06 | 1021 | |
bde89a9e | 1022 | int conn_send_state_req(struct drbd_connection *connection, union drbd_state mask, union drbd_state val) |
b411b363 | 1023 | { |
9f5bdc33 AG |
1024 | enum drbd_packet cmd; |
1025 | struct drbd_socket *sock; | |
1026 | struct p_req_state *p; | |
f479ea06 | 1027 | |
bde89a9e AG |
1028 | cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REQ : P_CONN_ST_CHG_REQ; |
1029 | sock = &connection->data; | |
1030 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1031 | if (!p) |
1032 | return -EIO; | |
1033 | p->mask = cpu_to_be32(mask.i); | |
1034 | p->val = cpu_to_be32(val.i); | |
bde89a9e | 1035 | return conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
f479ea06 LE |
1036 | } |
1037 | ||
69a22773 | 1038 | void drbd_send_sr_reply(struct drbd_peer_device *peer_device, enum drbd_state_rv retcode) |
b411b363 | 1039 | { |
9f5bdc33 AG |
1040 | struct drbd_socket *sock; |
1041 | struct p_req_state_reply *p; | |
b411b363 | 1042 | |
69a22773 AG |
1043 | sock = &peer_device->connection->meta; |
1044 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1045 | if (p) { |
1046 | p->retcode = cpu_to_be32(retcode); | |
69a22773 | 1047 | drbd_send_command(peer_device, sock, P_STATE_CHG_REPLY, sizeof(*p), NULL, 0); |
9f5bdc33 | 1048 | } |
b411b363 | 1049 | } |
b411b363 | 1050 | |
bde89a9e | 1051 | void conn_send_sr_reply(struct drbd_connection *connection, enum drbd_state_rv retcode) |
047cd4a6 | 1052 | { |
9f5bdc33 AG |
1053 | struct drbd_socket *sock; |
1054 | struct p_req_state_reply *p; | |
bde89a9e | 1055 | enum drbd_packet cmd = connection->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY; |
b411b363 | 1056 | |
bde89a9e AG |
1057 | sock = &connection->meta; |
1058 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1059 | if (p) { |
1060 | p->retcode = cpu_to_be32(retcode); | |
bde89a9e | 1061 | conn_send_command(connection, sock, cmd, sizeof(*p), NULL, 0); |
9f5bdc33 | 1062 | } |
b411b363 PR |
1063 | } |
1064 | ||
a02d1240 | 1065 | static void dcbp_set_code(struct p_compressed_bm *p, enum drbd_bitmap_code code) |
b411b363 | 1066 | { |
a02d1240 AG |
1067 | BUG_ON(code & ~0xf); |
1068 | p->encoding = (p->encoding & ~0xf) | code; | |
1069 | } | |
b411b363 | 1070 | |
a02d1240 AG |
1071 | static void dcbp_set_start(struct p_compressed_bm *p, int set) |
1072 | { | |
1073 | p->encoding = (p->encoding & ~0x80) | (set ? 0x80 : 0); | |
1074 | } | |
b411b363 | 1075 | |
a02d1240 AG |
1076 | static void dcbp_set_pad_bits(struct p_compressed_bm *p, int n) |
1077 | { | |
1078 | BUG_ON(n & ~0x7); | |
1079 | p->encoding = (p->encoding & (~0x7 << 4)) | (n << 4); | |
b411b363 PR |
1080 | } |
1081 | ||
b30ab791 | 1082 | static int fill_bitmap_rle_bits(struct drbd_device *device, |
50d0b1ad AG |
1083 | struct p_compressed_bm *p, |
1084 | unsigned int size, | |
1085 | struct bm_xfer_ctx *c) | |
b411b363 PR |
1086 | { |
1087 | struct bitstream bs; | |
1088 | unsigned long plain_bits; | |
1089 | unsigned long tmp; | |
1090 | unsigned long rl; | |
1091 | unsigned len; | |
1092 | unsigned toggle; | |
44ed167d | 1093 | int bits, use_rle; |
b411b363 PR |
1094 | |
1095 | /* may we use this feature? */ | |
44ed167d | 1096 | rcu_read_lock(); |
a6b32bc3 | 1097 | use_rle = rcu_dereference(first_peer_device(device)->connection->net_conf)->use_rle; |
44ed167d | 1098 | rcu_read_unlock(); |
a6b32bc3 | 1099 | if (!use_rle || first_peer_device(device)->connection->agreed_pro_version < 90) |
44ed167d | 1100 | return 0; |
b411b363 PR |
1101 | |
1102 | if (c->bit_offset >= c->bm_bits) | |
1103 | return 0; /* nothing to do. */ | |
1104 | ||
1105 | /* use at most thus many bytes */ | |
50d0b1ad AG |
1106 | bitstream_init(&bs, p->code, size, 0); |
1107 | memset(p->code, 0, size); | |
b411b363 PR |
1108 | /* plain bits covered in this code string */ |
1109 | plain_bits = 0; | |
1110 | ||
1111 | /* p->encoding & 0x80 stores whether the first run length is set. | |
1112 | * bit offset is implicit. | |
1113 | * start with toggle == 2 to be able to tell the first iteration */ | |
1114 | toggle = 2; | |
1115 | ||
1116 | /* see how much plain bits we can stuff into one packet | |
1117 | * using RLE and VLI. */ | |
1118 | do { | |
b30ab791 AG |
1119 | tmp = (toggle == 0) ? _drbd_bm_find_next_zero(device, c->bit_offset) |
1120 | : _drbd_bm_find_next(device, c->bit_offset); | |
b411b363 PR |
1121 | if (tmp == -1UL) |
1122 | tmp = c->bm_bits; | |
1123 | rl = tmp - c->bit_offset; | |
1124 | ||
1125 | if (toggle == 2) { /* first iteration */ | |
1126 | if (rl == 0) { | |
1127 | /* the first checked bit was set, | |
1128 | * store start value, */ | |
a02d1240 | 1129 | dcbp_set_start(p, 1); |
b411b363 PR |
1130 | /* but skip encoding of zero run length */ |
1131 | toggle = !toggle; | |
1132 | continue; | |
1133 | } | |
a02d1240 | 1134 | dcbp_set_start(p, 0); |
b411b363 PR |
1135 | } |
1136 | ||
1137 | /* paranoia: catch zero runlength. | |
1138 | * can only happen if bitmap is modified while we scan it. */ | |
1139 | if (rl == 0) { | |
d0180171 | 1140 | drbd_err(device, "unexpected zero runlength while encoding bitmap " |
b411b363 PR |
1141 | "t:%u bo:%lu\n", toggle, c->bit_offset); |
1142 | return -1; | |
1143 | } | |
1144 | ||
1145 | bits = vli_encode_bits(&bs, rl); | |
1146 | if (bits == -ENOBUFS) /* buffer full */ | |
1147 | break; | |
1148 | if (bits <= 0) { | |
d0180171 | 1149 | drbd_err(device, "error while encoding bitmap: %d\n", bits); |
b411b363 PR |
1150 | return 0; |
1151 | } | |
1152 | ||
1153 | toggle = !toggle; | |
1154 | plain_bits += rl; | |
1155 | c->bit_offset = tmp; | |
1156 | } while (c->bit_offset < c->bm_bits); | |
1157 | ||
1158 | len = bs.cur.b - p->code + !!bs.cur.bit; | |
1159 | ||
1160 | if (plain_bits < (len << 3)) { | |
1161 | /* incompressible with this method. | |
1162 | * we need to rewind both word and bit position. */ | |
1163 | c->bit_offset -= plain_bits; | |
1164 | bm_xfer_ctx_bit_to_word_offset(c); | |
1165 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1166 | return 0; | |
1167 | } | |
1168 | ||
1169 | /* RLE + VLI was able to compress it just fine. | |
1170 | * update c->word_offset. */ | |
1171 | bm_xfer_ctx_bit_to_word_offset(c); | |
1172 | ||
1173 | /* store pad_bits */ | |
a02d1240 | 1174 | dcbp_set_pad_bits(p, (8 - bs.cur.bit) & 0x7); |
b411b363 PR |
1175 | |
1176 | return len; | |
1177 | } | |
1178 | ||
f70af118 AG |
1179 | /** |
1180 | * send_bitmap_rle_or_plain | |
1181 | * | |
1182 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
1183 | * code upon failure. | |
1184 | */ | |
1185 | static int | |
b30ab791 | 1186 | send_bitmap_rle_or_plain(struct drbd_device *device, struct bm_xfer_ctx *c) |
b411b363 | 1187 | { |
a6b32bc3 AG |
1188 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
1189 | unsigned int header_size = drbd_header_size(first_peer_device(device)->connection); | |
e658983a | 1190 | struct p_compressed_bm *p = sock->sbuf + header_size; |
a982dd57 | 1191 | int len, err; |
b411b363 | 1192 | |
b30ab791 | 1193 | len = fill_bitmap_rle_bits(device, p, |
e658983a | 1194 | DRBD_SOCKET_BUFFER_SIZE - header_size - sizeof(*p), c); |
b411b363 | 1195 | if (len < 0) |
f70af118 | 1196 | return -EIO; |
b411b363 PR |
1197 | |
1198 | if (len) { | |
a02d1240 | 1199 | dcbp_set_code(p, RLE_VLI_Bits); |
a6b32bc3 | 1200 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, |
9f5bdc33 AG |
1201 | P_COMPRESSED_BITMAP, sizeof(*p) + len, |
1202 | NULL, 0); | |
b411b363 | 1203 | c->packets[0]++; |
e658983a | 1204 | c->bytes[0] += header_size + sizeof(*p) + len; |
b411b363 PR |
1205 | |
1206 | if (c->bit_offset >= c->bm_bits) | |
1207 | len = 0; /* DONE */ | |
1208 | } else { | |
1209 | /* was not compressible. | |
1210 | * send a buffer full of plain text bits instead. */ | |
50d0b1ad AG |
1211 | unsigned int data_size; |
1212 | unsigned long num_words; | |
e658983a | 1213 | unsigned long *p = sock->sbuf + header_size; |
50d0b1ad AG |
1214 | |
1215 | data_size = DRBD_SOCKET_BUFFER_SIZE - header_size; | |
e658983a | 1216 | num_words = min_t(size_t, data_size / sizeof(*p), |
50d0b1ad | 1217 | c->bm_words - c->word_offset); |
e658983a | 1218 | len = num_words * sizeof(*p); |
b411b363 | 1219 | if (len) |
b30ab791 | 1220 | drbd_bm_get_lel(device, c->word_offset, num_words, p); |
a6b32bc3 | 1221 | err = __send_command(first_peer_device(device)->connection, device->vnr, sock, P_BITMAP, len, NULL, 0); |
b411b363 PR |
1222 | c->word_offset += num_words; |
1223 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1224 | ||
1225 | c->packets[1]++; | |
50d0b1ad | 1226 | c->bytes[1] += header_size + len; |
b411b363 PR |
1227 | |
1228 | if (c->bit_offset > c->bm_bits) | |
1229 | c->bit_offset = c->bm_bits; | |
1230 | } | |
a982dd57 | 1231 | if (!err) { |
f70af118 | 1232 | if (len == 0) { |
b30ab791 | 1233 | INFO_bm_xfer_stats(device, "send", c); |
f70af118 AG |
1234 | return 0; |
1235 | } else | |
1236 | return 1; | |
1237 | } | |
1238 | return -EIO; | |
b411b363 PR |
1239 | } |
1240 | ||
1241 | /* See the comment at receive_bitmap() */ | |
b30ab791 | 1242 | static int _drbd_send_bitmap(struct drbd_device *device) |
b411b363 PR |
1243 | { |
1244 | struct bm_xfer_ctx c; | |
f70af118 | 1245 | int err; |
b411b363 | 1246 | |
b30ab791 | 1247 | if (!expect(device->bitmap)) |
81e84650 | 1248 | return false; |
b411b363 | 1249 | |
b30ab791 AG |
1250 | if (get_ldev(device)) { |
1251 | if (drbd_md_test_flag(device->ldev, MDF_FULL_SYNC)) { | |
d0180171 | 1252 | drbd_info(device, "Writing the whole bitmap, MDF_FullSync was set.\n"); |
b30ab791 AG |
1253 | drbd_bm_set_all(device); |
1254 | if (drbd_bm_write(device)) { | |
b411b363 PR |
1255 | /* write_bm did fail! Leave full sync flag set in Meta P_DATA |
1256 | * but otherwise process as per normal - need to tell other | |
1257 | * side that a full resync is required! */ | |
d0180171 | 1258 | drbd_err(device, "Failed to write bitmap to disk!\n"); |
b411b363 | 1259 | } else { |
b30ab791 AG |
1260 | drbd_md_clear_flag(device, MDF_FULL_SYNC); |
1261 | drbd_md_sync(device); | |
b411b363 PR |
1262 | } |
1263 | } | |
b30ab791 | 1264 | put_ldev(device); |
b411b363 PR |
1265 | } |
1266 | ||
1267 | c = (struct bm_xfer_ctx) { | |
b30ab791 AG |
1268 | .bm_bits = drbd_bm_bits(device), |
1269 | .bm_words = drbd_bm_words(device), | |
b411b363 PR |
1270 | }; |
1271 | ||
1272 | do { | |
b30ab791 | 1273 | err = send_bitmap_rle_or_plain(device, &c); |
f70af118 | 1274 | } while (err > 0); |
b411b363 | 1275 | |
f70af118 | 1276 | return err == 0; |
b411b363 PR |
1277 | } |
1278 | ||
b30ab791 | 1279 | int drbd_send_bitmap(struct drbd_device *device) |
b411b363 | 1280 | { |
a6b32bc3 | 1281 | struct drbd_socket *sock = &first_peer_device(device)->connection->data; |
9f5bdc33 | 1282 | int err = -1; |
b411b363 | 1283 | |
9f5bdc33 AG |
1284 | mutex_lock(&sock->mutex); |
1285 | if (sock->socket) | |
b30ab791 | 1286 | err = !_drbd_send_bitmap(device); |
9f5bdc33 | 1287 | mutex_unlock(&sock->mutex); |
b411b363 PR |
1288 | return err; |
1289 | } | |
1290 | ||
bde89a9e | 1291 | void drbd_send_b_ack(struct drbd_connection *connection, u32 barrier_nr, u32 set_size) |
b411b363 | 1292 | { |
9f5bdc33 AG |
1293 | struct drbd_socket *sock; |
1294 | struct p_barrier_ack *p; | |
b411b363 | 1295 | |
bde89a9e | 1296 | if (connection->cstate < C_WF_REPORT_PARAMS) |
9f5bdc33 | 1297 | return; |
b411b363 | 1298 | |
bde89a9e AG |
1299 | sock = &connection->meta; |
1300 | p = conn_prepare_command(connection, sock); | |
9f5bdc33 AG |
1301 | if (!p) |
1302 | return; | |
1303 | p->barrier = barrier_nr; | |
1304 | p->set_size = cpu_to_be32(set_size); | |
bde89a9e | 1305 | conn_send_command(connection, sock, P_BARRIER_ACK, sizeof(*p), NULL, 0); |
b411b363 PR |
1306 | } |
1307 | ||
1308 | /** | |
1309 | * _drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1310 | * @device: DRBD device. |
b411b363 PR |
1311 | * @cmd: Packet command code. |
1312 | * @sector: sector, needs to be in big endian byte order | |
1313 | * @blksize: size in byte, needs to be in big endian byte order | |
1314 | * @block_id: Id, big endian byte order | |
1315 | */ | |
69a22773 | 1316 | static int _drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
d8763023 | 1317 | u64 sector, u32 blksize, u64 block_id) |
b411b363 | 1318 | { |
9f5bdc33 AG |
1319 | struct drbd_socket *sock; |
1320 | struct p_block_ack *p; | |
b411b363 | 1321 | |
69a22773 | 1322 | if (peer_device->device->state.conn < C_CONNECTED) |
9f5bdc33 | 1323 | return -EIO; |
b411b363 | 1324 | |
69a22773 AG |
1325 | sock = &peer_device->connection->meta; |
1326 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 | 1327 | if (!p) |
a8c32aa8 | 1328 | return -EIO; |
9f5bdc33 AG |
1329 | p->sector = sector; |
1330 | p->block_id = block_id; | |
1331 | p->blksize = blksize; | |
69a22773 AG |
1332 | p->seq_num = cpu_to_be32(atomic_inc_return(&peer_device->device->packet_seq)); |
1333 | return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0); | |
b411b363 PR |
1334 | } |
1335 | ||
2b2bf214 LE |
1336 | /* dp->sector and dp->block_id already/still in network byte order, |
1337 | * data_size is payload size according to dp->head, | |
1338 | * and may need to be corrected for digest size. */ | |
69a22773 | 1339 | void drbd_send_ack_dp(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
a9a9994d | 1340 | struct p_data *dp, int data_size) |
b411b363 | 1341 | { |
69a22773 AG |
1342 | if (peer_device->connection->peer_integrity_tfm) |
1343 | data_size -= crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm); | |
1344 | _drbd_send_ack(peer_device, cmd, dp->sector, cpu_to_be32(data_size), | |
a9a9994d | 1345 | dp->block_id); |
b411b363 PR |
1346 | } |
1347 | ||
69a22773 | 1348 | void drbd_send_ack_rp(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
a9a9994d | 1349 | struct p_block_req *rp) |
b411b363 | 1350 | { |
69a22773 | 1351 | _drbd_send_ack(peer_device, cmd, rp->sector, rp->blksize, rp->block_id); |
b411b363 PR |
1352 | } |
1353 | ||
1354 | /** | |
1355 | * drbd_send_ack() - Sends an ack packet | |
b30ab791 | 1356 | * @device: DRBD device |
db830c46 AG |
1357 | * @cmd: packet command code |
1358 | * @peer_req: peer request | |
b411b363 | 1359 | */ |
69a22773 | 1360 | int drbd_send_ack(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
db830c46 | 1361 | struct drbd_peer_request *peer_req) |
b411b363 | 1362 | { |
69a22773 | 1363 | return _drbd_send_ack(peer_device, cmd, |
dd516121 AG |
1364 | cpu_to_be64(peer_req->i.sector), |
1365 | cpu_to_be32(peer_req->i.size), | |
1366 | peer_req->block_id); | |
b411b363 PR |
1367 | } |
1368 | ||
1369 | /* This function misuses the block_id field to signal if the blocks | |
1370 | * are is sync or not. */ | |
69a22773 | 1371 | int drbd_send_ack_ex(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
b411b363 PR |
1372 | sector_t sector, int blksize, u64 block_id) |
1373 | { | |
69a22773 | 1374 | return _drbd_send_ack(peer_device, cmd, |
b411b363 PR |
1375 | cpu_to_be64(sector), |
1376 | cpu_to_be32(blksize), | |
1377 | cpu_to_be64(block_id)); | |
1378 | } | |
1379 | ||
69a22773 | 1380 | int drbd_send_drequest(struct drbd_peer_device *peer_device, int cmd, |
b411b363 PR |
1381 | sector_t sector, int size, u64 block_id) |
1382 | { | |
9f5bdc33 AG |
1383 | struct drbd_socket *sock; |
1384 | struct p_block_req *p; | |
b411b363 | 1385 | |
69a22773 AG |
1386 | sock = &peer_device->connection->data; |
1387 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1388 | if (!p) |
1389 | return -EIO; | |
1390 | p->sector = cpu_to_be64(sector); | |
1391 | p->block_id = block_id; | |
1392 | p->blksize = cpu_to_be32(size); | |
69a22773 | 1393 | return drbd_send_command(peer_device, sock, cmd, sizeof(*p), NULL, 0); |
b411b363 PR |
1394 | } |
1395 | ||
69a22773 | 1396 | int drbd_send_drequest_csum(struct drbd_peer_device *peer_device, sector_t sector, int size, |
d8763023 | 1397 | void *digest, int digest_size, enum drbd_packet cmd) |
b411b363 | 1398 | { |
9f5bdc33 AG |
1399 | struct drbd_socket *sock; |
1400 | struct p_block_req *p; | |
b411b363 | 1401 | |
9f5bdc33 | 1402 | /* FIXME: Put the digest into the preallocated socket buffer. */ |
b411b363 | 1403 | |
69a22773 AG |
1404 | sock = &peer_device->connection->data; |
1405 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1406 | if (!p) |
1407 | return -EIO; | |
1408 | p->sector = cpu_to_be64(sector); | |
1409 | p->block_id = ID_SYNCER /* unused */; | |
1410 | p->blksize = cpu_to_be32(size); | |
69a22773 | 1411 | return drbd_send_command(peer_device, sock, cmd, sizeof(*p), digest, digest_size); |
b411b363 PR |
1412 | } |
1413 | ||
69a22773 | 1414 | int drbd_send_ov_request(struct drbd_peer_device *peer_device, sector_t sector, int size) |
b411b363 | 1415 | { |
9f5bdc33 AG |
1416 | struct drbd_socket *sock; |
1417 | struct p_block_req *p; | |
b411b363 | 1418 | |
69a22773 AG |
1419 | sock = &peer_device->connection->data; |
1420 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1421 | if (!p) |
1422 | return -EIO; | |
1423 | p->sector = cpu_to_be64(sector); | |
1424 | p->block_id = ID_SYNCER /* unused */; | |
1425 | p->blksize = cpu_to_be32(size); | |
69a22773 | 1426 | return drbd_send_command(peer_device, sock, P_OV_REQUEST, sizeof(*p), NULL, 0); |
b411b363 PR |
1427 | } |
1428 | ||
1429 | /* called on sndtimeo | |
81e84650 AG |
1430 | * returns false if we should retry, |
1431 | * true if we think connection is dead | |
b411b363 | 1432 | */ |
bde89a9e | 1433 | static int we_should_drop_the_connection(struct drbd_connection *connection, struct socket *sock) |
b411b363 PR |
1434 | { |
1435 | int drop_it; | |
b30ab791 | 1436 | /* long elapsed = (long)(jiffies - device->last_received); */ |
b411b363 | 1437 | |
bde89a9e | 1438 | drop_it = connection->meta.socket == sock |
1c03e520 PR |
1439 | || !connection->ack_receiver.task |
1440 | || get_t_state(&connection->ack_receiver) != RUNNING | |
bde89a9e | 1441 | || connection->cstate < C_WF_REPORT_PARAMS; |
b411b363 PR |
1442 | |
1443 | if (drop_it) | |
81e84650 | 1444 | return true; |
b411b363 | 1445 | |
bde89a9e | 1446 | drop_it = !--connection->ko_count; |
b411b363 | 1447 | if (!drop_it) { |
1ec861eb | 1448 | drbd_err(connection, "[%s/%d] sock_sendmsg time expired, ko = %u\n", |
bde89a9e AG |
1449 | current->comm, current->pid, connection->ko_count); |
1450 | request_ping(connection); | |
b411b363 PR |
1451 | } |
1452 | ||
b30ab791 | 1453 | return drop_it; /* && (device->state == R_PRIMARY) */; |
b411b363 PR |
1454 | } |
1455 | ||
bde89a9e | 1456 | static void drbd_update_congested(struct drbd_connection *connection) |
9e204cdd | 1457 | { |
bde89a9e | 1458 | struct sock *sk = connection->data.socket->sk; |
9e204cdd | 1459 | if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) |
bde89a9e | 1460 | set_bit(NET_CONGESTED, &connection->flags); |
9e204cdd AG |
1461 | } |
1462 | ||
b411b363 PR |
1463 | /* The idea of sendpage seems to be to put some kind of reference |
1464 | * to the page into the skb, and to hand it over to the NIC. In | |
1465 | * this process get_page() gets called. | |
1466 | * | |
1467 | * As soon as the page was really sent over the network put_page() | |
1468 | * gets called by some part of the network layer. [ NIC driver? ] | |
1469 | * | |
1470 | * [ get_page() / put_page() increment/decrement the count. If count | |
1471 | * reaches 0 the page will be freed. ] | |
1472 | * | |
1473 | * This works nicely with pages from FSs. | |
1474 | * But this means that in protocol A we might signal IO completion too early! | |
1475 | * | |
1476 | * In order not to corrupt data during a resync we must make sure | |
1477 | * that we do not reuse our own buffer pages (EEs) to early, therefore | |
1478 | * we have the net_ee list. | |
1479 | * | |
1480 | * XFS seems to have problems, still, it submits pages with page_count == 0! | |
1481 | * As a workaround, we disable sendpage on pages | |
1482 | * with page_count == 0 or PageSlab. | |
1483 | */ | |
69a22773 | 1484 | static int _drbd_no_send_page(struct drbd_peer_device *peer_device, struct page *page, |
b987427b | 1485 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1486 | { |
b987427b AG |
1487 | struct socket *socket; |
1488 | void *addr; | |
1489 | int err; | |
1490 | ||
69a22773 | 1491 | socket = peer_device->connection->data.socket; |
b987427b | 1492 | addr = kmap(page) + offset; |
69a22773 | 1493 | err = drbd_send_all(peer_device->connection, socket, addr, size, msg_flags); |
b411b363 | 1494 | kunmap(page); |
b987427b | 1495 | if (!err) |
69a22773 | 1496 | peer_device->device->send_cnt += size >> 9; |
b987427b | 1497 | return err; |
b411b363 PR |
1498 | } |
1499 | ||
69a22773 | 1500 | static int _drbd_send_page(struct drbd_peer_device *peer_device, struct page *page, |
ba11ad9a | 1501 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1502 | { |
69a22773 | 1503 | struct socket *socket = peer_device->connection->data.socket; |
b411b363 | 1504 | mm_segment_t oldfs = get_fs(); |
b411b363 | 1505 | int len = size; |
88b390ff | 1506 | int err = -EIO; |
b411b363 PR |
1507 | |
1508 | /* e.g. XFS meta- & log-data is in slab pages, which have a | |
1509 | * page_count of 0 and/or have PageSlab() set. | |
1510 | * we cannot use send_page for those, as that does get_page(); | |
1511 | * put_page(); and would cause either a VM_BUG directly, or | |
1512 | * __page_cache_release a page that would actually still be referenced | |
1513 | * by someone, leading to some obscure delayed Oops somewhere else. */ | |
1514 | if (disable_sendpage || (page_count(page) < 1) || PageSlab(page)) | |
69a22773 | 1515 | return _drbd_no_send_page(peer_device, page, offset, size, msg_flags); |
b411b363 | 1516 | |
ba11ad9a | 1517 | msg_flags |= MSG_NOSIGNAL; |
69a22773 | 1518 | drbd_update_congested(peer_device->connection); |
b411b363 PR |
1519 | set_fs(KERNEL_DS); |
1520 | do { | |
88b390ff AG |
1521 | int sent; |
1522 | ||
1523 | sent = socket->ops->sendpage(socket, page, offset, len, msg_flags); | |
b411b363 | 1524 | if (sent <= 0) { |
88b390ff | 1525 | if (sent == -EAGAIN) { |
69a22773 | 1526 | if (we_should_drop_the_connection(peer_device->connection, socket)) |
88b390ff AG |
1527 | break; |
1528 | continue; | |
1529 | } | |
69a22773 | 1530 | drbd_warn(peer_device->device, "%s: size=%d len=%d sent=%d\n", |
b411b363 | 1531 | __func__, (int)size, len, sent); |
88b390ff AG |
1532 | if (sent < 0) |
1533 | err = sent; | |
b411b363 PR |
1534 | break; |
1535 | } | |
1536 | len -= sent; | |
1537 | offset += sent; | |
b30ab791 | 1538 | } while (len > 0 /* THINK && device->cstate >= C_CONNECTED*/); |
b411b363 | 1539 | set_fs(oldfs); |
69a22773 | 1540 | clear_bit(NET_CONGESTED, &peer_device->connection->flags); |
b411b363 | 1541 | |
88b390ff AG |
1542 | if (len == 0) { |
1543 | err = 0; | |
69a22773 | 1544 | peer_device->device->send_cnt += size >> 9; |
88b390ff AG |
1545 | } |
1546 | return err; | |
b411b363 PR |
1547 | } |
1548 | ||
69a22773 | 1549 | static int _drbd_send_bio(struct drbd_peer_device *peer_device, struct bio *bio) |
b411b363 | 1550 | { |
7988613b KO |
1551 | struct bio_vec bvec; |
1552 | struct bvec_iter iter; | |
1553 | ||
ba11ad9a | 1554 | /* hint all but last page with MSG_MORE */ |
7988613b | 1555 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1556 | int err; |
1557 | ||
69a22773 | 1558 | err = _drbd_no_send_page(peer_device, bvec.bv_page, |
7988613b | 1559 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1560 | bio_iter_last(bvec, iter) |
7988613b | 1561 | ? 0 : MSG_MORE); |
7fae55da AG |
1562 | if (err) |
1563 | return err; | |
b411b363 | 1564 | } |
7fae55da | 1565 | return 0; |
b411b363 PR |
1566 | } |
1567 | ||
69a22773 | 1568 | static int _drbd_send_zc_bio(struct drbd_peer_device *peer_device, struct bio *bio) |
b411b363 | 1569 | { |
7988613b KO |
1570 | struct bio_vec bvec; |
1571 | struct bvec_iter iter; | |
1572 | ||
ba11ad9a | 1573 | /* hint all but last page with MSG_MORE */ |
7988613b | 1574 | bio_for_each_segment(bvec, bio, iter) { |
7fae55da AG |
1575 | int err; |
1576 | ||
69a22773 | 1577 | err = _drbd_send_page(peer_device, bvec.bv_page, |
7988613b | 1578 | bvec.bv_offset, bvec.bv_len, |
4550dd6c | 1579 | bio_iter_last(bvec, iter) ? 0 : MSG_MORE); |
7fae55da AG |
1580 | if (err) |
1581 | return err; | |
b411b363 | 1582 | } |
7fae55da | 1583 | return 0; |
b411b363 PR |
1584 | } |
1585 | ||
69a22773 | 1586 | static int _drbd_send_zc_ee(struct drbd_peer_device *peer_device, |
db830c46 | 1587 | struct drbd_peer_request *peer_req) |
45bb912b | 1588 | { |
db830c46 AG |
1589 | struct page *page = peer_req->pages; |
1590 | unsigned len = peer_req->i.size; | |
9f69230c | 1591 | int err; |
db830c46 | 1592 | |
ba11ad9a | 1593 | /* hint all but last page with MSG_MORE */ |
45bb912b LE |
1594 | page_chain_for_each(page) { |
1595 | unsigned l = min_t(unsigned, len, PAGE_SIZE); | |
9f69230c | 1596 | |
69a22773 | 1597 | err = _drbd_send_page(peer_device, page, 0, l, |
9f69230c AG |
1598 | page_chain_next(page) ? MSG_MORE : 0); |
1599 | if (err) | |
1600 | return err; | |
45bb912b LE |
1601 | len -= l; |
1602 | } | |
9f69230c | 1603 | return 0; |
45bb912b LE |
1604 | } |
1605 | ||
69a22773 | 1606 | static u32 bio_flags_to_wire(struct drbd_connection *connection, unsigned long bi_rw) |
76d2e7ec | 1607 | { |
69a22773 | 1608 | if (connection->agreed_pro_version >= 95) |
76d2e7ec | 1609 | return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) | |
76d2e7ec PR |
1610 | (bi_rw & REQ_FUA ? DP_FUA : 0) | |
1611 | (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) | | |
1612 | (bi_rw & REQ_DISCARD ? DP_DISCARD : 0); | |
1613 | else | |
721a9602 | 1614 | return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0; |
76d2e7ec PR |
1615 | } |
1616 | ||
2f632aeb LE |
1617 | /* Used to send write or TRIM aka REQ_DISCARD requests |
1618 | * R_PRIMARY -> Peer (P_DATA, P_TRIM) | |
b411b363 | 1619 | */ |
69a22773 | 1620 | int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *req) |
b411b363 | 1621 | { |
69a22773 | 1622 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
1623 | struct drbd_socket *sock; |
1624 | struct p_data *p; | |
b411b363 | 1625 | unsigned int dp_flags = 0; |
11f8b2b6 | 1626 | int digest_size; |
9f5bdc33 | 1627 | int err; |
b411b363 | 1628 | |
69a22773 AG |
1629 | sock = &peer_device->connection->data; |
1630 | p = drbd_prepare_command(peer_device, sock); | |
11f8b2b6 AG |
1631 | digest_size = peer_device->connection->integrity_tfm ? |
1632 | crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0; | |
b411b363 | 1633 | |
9f5bdc33 AG |
1634 | if (!p) |
1635 | return -EIO; | |
1636 | p->sector = cpu_to_be64(req->i.sector); | |
1637 | p->block_id = (unsigned long)req; | |
b30ab791 | 1638 | p->seq_num = cpu_to_be32(atomic_inc_return(&device->packet_seq)); |
69a22773 | 1639 | dp_flags = bio_flags_to_wire(peer_device->connection, req->master_bio->bi_rw); |
b30ab791 AG |
1640 | if (device->state.conn >= C_SYNC_SOURCE && |
1641 | device->state.conn <= C_PAUSED_SYNC_T) | |
b411b363 | 1642 | dp_flags |= DP_MAY_SET_IN_SYNC; |
69a22773 | 1643 | if (peer_device->connection->agreed_pro_version >= 100) { |
303d1448 PR |
1644 | if (req->rq_state & RQ_EXP_RECEIVE_ACK) |
1645 | dp_flags |= DP_SEND_RECEIVE_ACK; | |
08d0dabf LE |
1646 | /* During resync, request an explicit write ack, |
1647 | * even in protocol != C */ | |
1648 | if (req->rq_state & RQ_EXP_WRITE_ACK | |
1649 | || (dp_flags & DP_MAY_SET_IN_SYNC)) | |
303d1448 PR |
1650 | dp_flags |= DP_SEND_WRITE_ACK; |
1651 | } | |
9f5bdc33 | 1652 | p->dp_flags = cpu_to_be32(dp_flags); |
2f632aeb LE |
1653 | |
1654 | if (dp_flags & DP_DISCARD) { | |
1655 | struct p_trim *t = (struct p_trim*)p; | |
1656 | t->size = cpu_to_be32(req->i.size); | |
1657 | err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0); | |
1658 | goto out; | |
1659 | } | |
1660 | ||
1661 | /* our digest is still only over the payload. | |
1662 | * TRIM does not carry any payload. */ | |
11f8b2b6 | 1663 | if (digest_size) |
69a22773 | 1664 | drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, p + 1); |
11f8b2b6 | 1665 | err = __send_command(peer_device->connection, device->vnr, sock, P_DATA, sizeof(*p) + digest_size, NULL, req->i.size); |
6bdb9b0e | 1666 | if (!err) { |
470be44a LE |
1667 | /* For protocol A, we have to memcpy the payload into |
1668 | * socket buffers, as we may complete right away | |
1669 | * as soon as we handed it over to tcp, at which point the data | |
1670 | * pages may become invalid. | |
1671 | * | |
1672 | * For data-integrity enabled, we copy it as well, so we can be | |
1673 | * sure that even if the bio pages may still be modified, it | |
1674 | * won't change the data on the wire, thus if the digest checks | |
1675 | * out ok after sending on this side, but does not fit on the | |
1676 | * receiving side, we sure have detected corruption elsewhere. | |
1677 | */ | |
11f8b2b6 | 1678 | if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size) |
69a22773 | 1679 | err = _drbd_send_bio(peer_device, req->master_bio); |
b411b363 | 1680 | else |
69a22773 | 1681 | err = _drbd_send_zc_bio(peer_device, req->master_bio); |
470be44a LE |
1682 | |
1683 | /* double check digest, sometimes buffers have been modified in flight. */ | |
11f8b2b6 | 1684 | if (digest_size > 0 && digest_size <= 64) { |
24c4830c | 1685 | /* 64 byte, 512 bit, is the largest digest size |
470be44a LE |
1686 | * currently supported in kernel crypto. */ |
1687 | unsigned char digest[64]; | |
69a22773 | 1688 | drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest); |
11f8b2b6 | 1689 | if (memcmp(p + 1, digest, digest_size)) { |
d0180171 | 1690 | drbd_warn(device, |
470be44a | 1691 | "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n", |
ace652ac | 1692 | (unsigned long long)req->i.sector, req->i.size); |
470be44a | 1693 | } |
11f8b2b6 | 1694 | } /* else if (digest_size > 64) { |
470be44a LE |
1695 | ... Be noisy about digest too large ... |
1696 | } */ | |
b411b363 | 1697 | } |
2f632aeb | 1698 | out: |
9f5bdc33 | 1699 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
b411b363 | 1700 | |
6bdb9b0e | 1701 | return err; |
b411b363 PR |
1702 | } |
1703 | ||
1704 | /* answer packet, used to send data back for read requests: | |
1705 | * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY) | |
1706 | * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY) | |
1707 | */ | |
69a22773 | 1708 | int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd, |
db830c46 | 1709 | struct drbd_peer_request *peer_req) |
b411b363 | 1710 | { |
69a22773 | 1711 | struct drbd_device *device = peer_device->device; |
9f5bdc33 AG |
1712 | struct drbd_socket *sock; |
1713 | struct p_data *p; | |
7b57b89d | 1714 | int err; |
11f8b2b6 | 1715 | int digest_size; |
b411b363 | 1716 | |
69a22773 AG |
1717 | sock = &peer_device->connection->data; |
1718 | p = drbd_prepare_command(peer_device, sock); | |
b411b363 | 1719 | |
11f8b2b6 AG |
1720 | digest_size = peer_device->connection->integrity_tfm ? |
1721 | crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0; | |
b411b363 | 1722 | |
9f5bdc33 AG |
1723 | if (!p) |
1724 | return -EIO; | |
1725 | p->sector = cpu_to_be64(peer_req->i.sector); | |
1726 | p->block_id = peer_req->block_id; | |
1727 | p->seq_num = 0; /* unused */ | |
b17f33cb | 1728 | p->dp_flags = 0; |
11f8b2b6 | 1729 | if (digest_size) |
69a22773 | 1730 | drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1); |
11f8b2b6 | 1731 | err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size); |
7b57b89d | 1732 | if (!err) |
69a22773 | 1733 | err = _drbd_send_zc_ee(peer_device, peer_req); |
9f5bdc33 | 1734 | mutex_unlock(&sock->mutex); /* locked by drbd_prepare_command() */ |
bd26bfc5 | 1735 | |
7b57b89d | 1736 | return err; |
b411b363 PR |
1737 | } |
1738 | ||
69a22773 | 1739 | int drbd_send_out_of_sync(struct drbd_peer_device *peer_device, struct drbd_request *req) |
73a01a18 | 1740 | { |
9f5bdc33 AG |
1741 | struct drbd_socket *sock; |
1742 | struct p_block_desc *p; | |
73a01a18 | 1743 | |
69a22773 AG |
1744 | sock = &peer_device->connection->data; |
1745 | p = drbd_prepare_command(peer_device, sock); | |
9f5bdc33 AG |
1746 | if (!p) |
1747 | return -EIO; | |
1748 | p->sector = cpu_to_be64(req->i.sector); | |
1749 | p->blksize = cpu_to_be32(req->i.size); | |
69a22773 | 1750 | return drbd_send_command(peer_device, sock, P_OUT_OF_SYNC, sizeof(*p), NULL, 0); |
73a01a18 PR |
1751 | } |
1752 | ||
b411b363 PR |
1753 | /* |
1754 | drbd_send distinguishes two cases: | |
1755 | ||
1756 | Packets sent via the data socket "sock" | |
1757 | and packets sent via the meta data socket "msock" | |
1758 | ||
1759 | sock msock | |
1760 | -----------------+-------------------------+------------------------------ | |
1761 | timeout conf.timeout / 2 conf.timeout / 2 | |
1762 | timeout action send a ping via msock Abort communication | |
1763 | and close all sockets | |
1764 | */ | |
1765 | ||
1766 | /* | |
1767 | * you must have down()ed the appropriate [m]sock_mutex elsewhere! | |
1768 | */ | |
bde89a9e | 1769 | int drbd_send(struct drbd_connection *connection, struct socket *sock, |
b411b363 PR |
1770 | void *buf, size_t size, unsigned msg_flags) |
1771 | { | |
1772 | struct kvec iov; | |
1773 | struct msghdr msg; | |
1774 | int rv, sent = 0; | |
1775 | ||
1776 | if (!sock) | |
c0d42c8e | 1777 | return -EBADR; |
b411b363 PR |
1778 | |
1779 | /* THINK if (signal_pending) return ... ? */ | |
1780 | ||
1781 | iov.iov_base = buf; | |
1782 | iov.iov_len = size; | |
1783 | ||
1784 | msg.msg_name = NULL; | |
1785 | msg.msg_namelen = 0; | |
1786 | msg.msg_control = NULL; | |
1787 | msg.msg_controllen = 0; | |
1788 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; | |
1789 | ||
bde89a9e | 1790 | if (sock == connection->data.socket) { |
44ed167d | 1791 | rcu_read_lock(); |
bde89a9e | 1792 | connection->ko_count = rcu_dereference(connection->net_conf)->ko_count; |
44ed167d | 1793 | rcu_read_unlock(); |
bde89a9e | 1794 | drbd_update_congested(connection); |
b411b363 PR |
1795 | } |
1796 | do { | |
b411b363 PR |
1797 | rv = kernel_sendmsg(sock, &msg, &iov, 1, size); |
1798 | if (rv == -EAGAIN) { | |
bde89a9e | 1799 | if (we_should_drop_the_connection(connection, sock)) |
b411b363 PR |
1800 | break; |
1801 | else | |
1802 | continue; | |
1803 | } | |
b411b363 PR |
1804 | if (rv == -EINTR) { |
1805 | flush_signals(current); | |
1806 | rv = 0; | |
1807 | } | |
1808 | if (rv < 0) | |
1809 | break; | |
1810 | sent += rv; | |
1811 | iov.iov_base += rv; | |
1812 | iov.iov_len -= rv; | |
1813 | } while (sent < size); | |
1814 | ||
bde89a9e AG |
1815 | if (sock == connection->data.socket) |
1816 | clear_bit(NET_CONGESTED, &connection->flags); | |
b411b363 PR |
1817 | |
1818 | if (rv <= 0) { | |
1819 | if (rv != -EAGAIN) { | |
1ec861eb | 1820 | drbd_err(connection, "%s_sendmsg returned %d\n", |
bde89a9e | 1821 | sock == connection->meta.socket ? "msock" : "sock", |
bedbd2a5 | 1822 | rv); |
bde89a9e | 1823 | conn_request_state(connection, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 1824 | } else |
bde89a9e | 1825 | conn_request_state(connection, NS(conn, C_TIMEOUT), CS_HARD); |
b411b363 PR |
1826 | } |
1827 | ||
1828 | return sent; | |
1829 | } | |
1830 | ||
fb708e40 AG |
1831 | /** |
1832 | * drbd_send_all - Send an entire buffer | |
1833 | * | |
1834 | * Returns 0 upon success and a negative error value otherwise. | |
1835 | */ | |
bde89a9e | 1836 | int drbd_send_all(struct drbd_connection *connection, struct socket *sock, void *buffer, |
fb708e40 AG |
1837 | size_t size, unsigned msg_flags) |
1838 | { | |
1839 | int err; | |
1840 | ||
bde89a9e | 1841 | err = drbd_send(connection, sock, buffer, size, msg_flags); |
fb708e40 AG |
1842 | if (err < 0) |
1843 | return err; | |
1844 | if (err != size) | |
1845 | return -EIO; | |
1846 | return 0; | |
1847 | } | |
1848 | ||
b411b363 PR |
1849 | static int drbd_open(struct block_device *bdev, fmode_t mode) |
1850 | { | |
b30ab791 | 1851 | struct drbd_device *device = bdev->bd_disk->private_data; |
b411b363 PR |
1852 | unsigned long flags; |
1853 | int rv = 0; | |
1854 | ||
2a48fc0a | 1855 | mutex_lock(&drbd_main_mutex); |
0500813f | 1856 | spin_lock_irqsave(&device->resource->req_lock, flags); |
b30ab791 | 1857 | /* to have a stable device->state.role |
b411b363 PR |
1858 | * and no race with updating open_cnt */ |
1859 | ||
b30ab791 | 1860 | if (device->state.role != R_PRIMARY) { |
b411b363 PR |
1861 | if (mode & FMODE_WRITE) |
1862 | rv = -EROFS; | |
1863 | else if (!allow_oos) | |
1864 | rv = -EMEDIUMTYPE; | |
1865 | } | |
1866 | ||
1867 | if (!rv) | |
b30ab791 | 1868 | device->open_cnt++; |
0500813f | 1869 | spin_unlock_irqrestore(&device->resource->req_lock, flags); |
2a48fc0a | 1870 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1871 | |
1872 | return rv; | |
1873 | } | |
1874 | ||
db2a144b | 1875 | static void drbd_release(struct gendisk *gd, fmode_t mode) |
b411b363 | 1876 | { |
b30ab791 | 1877 | struct drbd_device *device = gd->private_data; |
2a48fc0a | 1878 | mutex_lock(&drbd_main_mutex); |
b30ab791 | 1879 | device->open_cnt--; |
2a48fc0a | 1880 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1881 | } |
1882 | ||
b30ab791 | 1883 | static void drbd_set_defaults(struct drbd_device *device) |
b411b363 | 1884 | { |
f399002e LE |
1885 | /* Beware! The actual layout differs |
1886 | * between big endian and little endian */ | |
b30ab791 | 1887 | device->state = (union drbd_dev_state) { |
b411b363 PR |
1888 | { .role = R_SECONDARY, |
1889 | .peer = R_UNKNOWN, | |
1890 | .conn = C_STANDALONE, | |
1891 | .disk = D_DISKLESS, | |
1892 | .pdsk = D_UNKNOWN, | |
b411b363 PR |
1893 | } }; |
1894 | } | |
1895 | ||
b30ab791 | 1896 | void drbd_init_set_defaults(struct drbd_device *device) |
b411b363 PR |
1897 | { |
1898 | /* the memset(,0,) did most of this. | |
1899 | * note: only assignments, no allocation in here */ | |
1900 | ||
b30ab791 AG |
1901 | drbd_set_defaults(device); |
1902 | ||
1903 | atomic_set(&device->ap_bio_cnt, 0); | |
ad3fee79 | 1904 | atomic_set(&device->ap_actlog_cnt, 0); |
b30ab791 AG |
1905 | atomic_set(&device->ap_pending_cnt, 0); |
1906 | atomic_set(&device->rs_pending_cnt, 0); | |
1907 | atomic_set(&device->unacked_cnt, 0); | |
1908 | atomic_set(&device->local_cnt, 0); | |
1909 | atomic_set(&device->pp_in_use_by_net, 0); | |
1910 | atomic_set(&device->rs_sect_in, 0); | |
1911 | atomic_set(&device->rs_sect_ev, 0); | |
1912 | atomic_set(&device->ap_in_flight, 0); | |
e37d2438 | 1913 | atomic_set(&device->md_io.in_use, 0); |
b30ab791 AG |
1914 | |
1915 | mutex_init(&device->own_state_mutex); | |
1916 | device->state_mutex = &device->own_state_mutex; | |
1917 | ||
1918 | spin_lock_init(&device->al_lock); | |
1919 | spin_lock_init(&device->peer_seq_lock); | |
1920 | ||
1921 | INIT_LIST_HEAD(&device->active_ee); | |
1922 | INIT_LIST_HEAD(&device->sync_ee); | |
1923 | INIT_LIST_HEAD(&device->done_ee); | |
1924 | INIT_LIST_HEAD(&device->read_ee); | |
1925 | INIT_LIST_HEAD(&device->net_ee); | |
1926 | INIT_LIST_HEAD(&device->resync_reads); | |
1927 | INIT_LIST_HEAD(&device->resync_work.list); | |
1928 | INIT_LIST_HEAD(&device->unplug_work.list); | |
b30ab791 | 1929 | INIT_LIST_HEAD(&device->bm_io_work.w.list); |
844a6ae7 LE |
1930 | INIT_LIST_HEAD(&device->pending_master_completion[0]); |
1931 | INIT_LIST_HEAD(&device->pending_master_completion[1]); | |
1932 | INIT_LIST_HEAD(&device->pending_completion[0]); | |
1933 | INIT_LIST_HEAD(&device->pending_completion[1]); | |
b30ab791 AG |
1934 | |
1935 | device->resync_work.cb = w_resync_timer; | |
1936 | device->unplug_work.cb = w_send_write_hint; | |
b30ab791 | 1937 | device->bm_io_work.w.cb = w_bitmap_io; |
b30ab791 | 1938 | |
b30ab791 AG |
1939 | init_timer(&device->resync_timer); |
1940 | init_timer(&device->md_sync_timer); | |
1941 | init_timer(&device->start_resync_timer); | |
1942 | init_timer(&device->request_timer); | |
1943 | device->resync_timer.function = resync_timer_fn; | |
1944 | device->resync_timer.data = (unsigned long) device; | |
1945 | device->md_sync_timer.function = md_sync_timer_fn; | |
1946 | device->md_sync_timer.data = (unsigned long) device; | |
1947 | device->start_resync_timer.function = start_resync_timer_fn; | |
1948 | device->start_resync_timer.data = (unsigned long) device; | |
1949 | device->request_timer.function = request_timer_fn; | |
1950 | device->request_timer.data = (unsigned long) device; | |
1951 | ||
1952 | init_waitqueue_head(&device->misc_wait); | |
1953 | init_waitqueue_head(&device->state_wait); | |
1954 | init_waitqueue_head(&device->ee_wait); | |
1955 | init_waitqueue_head(&device->al_wait); | |
1956 | init_waitqueue_head(&device->seq_wait); | |
1957 | ||
1958 | device->resync_wenr = LC_FREE; | |
1959 | device->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1960 | device->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
1961 | } | |
1962 | ||
1963 | void drbd_device_cleanup(struct drbd_device *device) | |
b411b363 | 1964 | { |
1d7734a0 | 1965 | int i; |
a6b32bc3 | 1966 | if (first_peer_device(device)->connection->receiver.t_state != NONE) |
d0180171 | 1967 | drbd_err(device, "ASSERT FAILED: receiver t_state == %d expected 0.\n", |
a6b32bc3 | 1968 | first_peer_device(device)->connection->receiver.t_state); |
b30ab791 AG |
1969 | |
1970 | device->al_writ_cnt = | |
1971 | device->bm_writ_cnt = | |
1972 | device->read_cnt = | |
1973 | device->recv_cnt = | |
1974 | device->send_cnt = | |
1975 | device->writ_cnt = | |
1976 | device->p_size = | |
1977 | device->rs_start = | |
1978 | device->rs_total = | |
1979 | device->rs_failed = 0; | |
1980 | device->rs_last_events = 0; | |
1981 | device->rs_last_sect_ev = 0; | |
1d7734a0 | 1982 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
b30ab791 AG |
1983 | device->rs_mark_left[i] = 0; |
1984 | device->rs_mark_time[i] = 0; | |
1d7734a0 | 1985 | } |
0b0ba1ef | 1986 | D_ASSERT(device, first_peer_device(device)->connection->net_conf == NULL); |
b411b363 | 1987 | |
b30ab791 AG |
1988 | drbd_set_my_capacity(device, 0); |
1989 | if (device->bitmap) { | |
b411b363 | 1990 | /* maybe never allocated. */ |
b30ab791 AG |
1991 | drbd_bm_resize(device, 0, 1); |
1992 | drbd_bm_cleanup(device); | |
b411b363 PR |
1993 | } |
1994 | ||
63a7c8ad | 1995 | drbd_backing_dev_free(device, device->ldev); |
b30ab791 | 1996 | device->ldev = NULL; |
1d041225 | 1997 | |
b30ab791 | 1998 | clear_bit(AL_SUSPENDED, &device->flags); |
b411b363 | 1999 | |
0b0ba1ef AG |
2000 | D_ASSERT(device, list_empty(&device->active_ee)); |
2001 | D_ASSERT(device, list_empty(&device->sync_ee)); | |
2002 | D_ASSERT(device, list_empty(&device->done_ee)); | |
2003 | D_ASSERT(device, list_empty(&device->read_ee)); | |
2004 | D_ASSERT(device, list_empty(&device->net_ee)); | |
2005 | D_ASSERT(device, list_empty(&device->resync_reads)); | |
2006 | D_ASSERT(device, list_empty(&first_peer_device(device)->connection->sender_work.q)); | |
2007 | D_ASSERT(device, list_empty(&device->resync_work.list)); | |
2008 | D_ASSERT(device, list_empty(&device->unplug_work.list)); | |
2265b473 | 2009 | |
b30ab791 | 2010 | drbd_set_defaults(device); |
b411b363 PR |
2011 | } |
2012 | ||
2013 | ||
2014 | static void drbd_destroy_mempools(void) | |
2015 | { | |
2016 | struct page *page; | |
2017 | ||
2018 | while (drbd_pp_pool) { | |
2019 | page = drbd_pp_pool; | |
2020 | drbd_pp_pool = (struct page *)page_private(page); | |
2021 | __free_page(page); | |
2022 | drbd_pp_vacant--; | |
2023 | } | |
2024 | ||
0b0ba1ef | 2025 | /* D_ASSERT(device, atomic_read(&drbd_pp_vacant)==0); */ |
b411b363 | 2026 | |
9476f39d LE |
2027 | if (drbd_md_io_bio_set) |
2028 | bioset_free(drbd_md_io_bio_set); | |
4281808f LE |
2029 | if (drbd_md_io_page_pool) |
2030 | mempool_destroy(drbd_md_io_page_pool); | |
b411b363 PR |
2031 | if (drbd_ee_mempool) |
2032 | mempool_destroy(drbd_ee_mempool); | |
2033 | if (drbd_request_mempool) | |
2034 | mempool_destroy(drbd_request_mempool); | |
2035 | if (drbd_ee_cache) | |
2036 | kmem_cache_destroy(drbd_ee_cache); | |
2037 | if (drbd_request_cache) | |
2038 | kmem_cache_destroy(drbd_request_cache); | |
2039 | if (drbd_bm_ext_cache) | |
2040 | kmem_cache_destroy(drbd_bm_ext_cache); | |
2041 | if (drbd_al_ext_cache) | |
2042 | kmem_cache_destroy(drbd_al_ext_cache); | |
2043 | ||
9476f39d | 2044 | drbd_md_io_bio_set = NULL; |
4281808f | 2045 | drbd_md_io_page_pool = NULL; |
b411b363 PR |
2046 | drbd_ee_mempool = NULL; |
2047 | drbd_request_mempool = NULL; | |
2048 | drbd_ee_cache = NULL; | |
2049 | drbd_request_cache = NULL; | |
2050 | drbd_bm_ext_cache = NULL; | |
2051 | drbd_al_ext_cache = NULL; | |
2052 | ||
2053 | return; | |
2054 | } | |
2055 | ||
2056 | static int drbd_create_mempools(void) | |
2057 | { | |
2058 | struct page *page; | |
1816a2b4 | 2059 | const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count; |
b411b363 PR |
2060 | int i; |
2061 | ||
2062 | /* prepare our caches and mempools */ | |
2063 | drbd_request_mempool = NULL; | |
2064 | drbd_ee_cache = NULL; | |
2065 | drbd_request_cache = NULL; | |
2066 | drbd_bm_ext_cache = NULL; | |
2067 | drbd_al_ext_cache = NULL; | |
2068 | drbd_pp_pool = NULL; | |
4281808f | 2069 | drbd_md_io_page_pool = NULL; |
9476f39d | 2070 | drbd_md_io_bio_set = NULL; |
b411b363 PR |
2071 | |
2072 | /* caches */ | |
2073 | drbd_request_cache = kmem_cache_create( | |
2074 | "drbd_req", sizeof(struct drbd_request), 0, 0, NULL); | |
2075 | if (drbd_request_cache == NULL) | |
2076 | goto Enomem; | |
2077 | ||
2078 | drbd_ee_cache = kmem_cache_create( | |
f6ffca9f | 2079 | "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL); |
b411b363 PR |
2080 | if (drbd_ee_cache == NULL) |
2081 | goto Enomem; | |
2082 | ||
2083 | drbd_bm_ext_cache = kmem_cache_create( | |
2084 | "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL); | |
2085 | if (drbd_bm_ext_cache == NULL) | |
2086 | goto Enomem; | |
2087 | ||
2088 | drbd_al_ext_cache = kmem_cache_create( | |
2089 | "drbd_al", sizeof(struct lc_element), 0, 0, NULL); | |
2090 | if (drbd_al_ext_cache == NULL) | |
2091 | goto Enomem; | |
2092 | ||
2093 | /* mempools */ | |
9476f39d LE |
2094 | drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0); |
2095 | if (drbd_md_io_bio_set == NULL) | |
2096 | goto Enomem; | |
9476f39d | 2097 | |
4281808f LE |
2098 | drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0); |
2099 | if (drbd_md_io_page_pool == NULL) | |
2100 | goto Enomem; | |
2101 | ||
cbc4ffdb DR |
2102 | drbd_request_mempool = mempool_create_slab_pool(number, |
2103 | drbd_request_cache); | |
b411b363 PR |
2104 | if (drbd_request_mempool == NULL) |
2105 | goto Enomem; | |
2106 | ||
cbc4ffdb | 2107 | drbd_ee_mempool = mempool_create_slab_pool(number, drbd_ee_cache); |
2027ae1f | 2108 | if (drbd_ee_mempool == NULL) |
b411b363 PR |
2109 | goto Enomem; |
2110 | ||
2111 | /* drbd's page pool */ | |
2112 | spin_lock_init(&drbd_pp_lock); | |
2113 | ||
2114 | for (i = 0; i < number; i++) { | |
2115 | page = alloc_page(GFP_HIGHUSER); | |
2116 | if (!page) | |
2117 | goto Enomem; | |
2118 | set_page_private(page, (unsigned long)drbd_pp_pool); | |
2119 | drbd_pp_pool = page; | |
2120 | } | |
2121 | drbd_pp_vacant = number; | |
2122 | ||
2123 | return 0; | |
2124 | ||
2125 | Enomem: | |
2126 | drbd_destroy_mempools(); /* in case we allocated some */ | |
2127 | return -ENOMEM; | |
2128 | } | |
2129 | ||
b30ab791 | 2130 | static void drbd_release_all_peer_reqs(struct drbd_device *device) |
b411b363 PR |
2131 | { |
2132 | int rr; | |
2133 | ||
b30ab791 | 2134 | rr = drbd_free_peer_reqs(device, &device->active_ee); |
b411b363 | 2135 | if (rr) |
d0180171 | 2136 | drbd_err(device, "%d EEs in active list found!\n", rr); |
b411b363 | 2137 | |
b30ab791 | 2138 | rr = drbd_free_peer_reqs(device, &device->sync_ee); |
b411b363 | 2139 | if (rr) |
d0180171 | 2140 | drbd_err(device, "%d EEs in sync list found!\n", rr); |
b411b363 | 2141 | |
b30ab791 | 2142 | rr = drbd_free_peer_reqs(device, &device->read_ee); |
b411b363 | 2143 | if (rr) |
d0180171 | 2144 | drbd_err(device, "%d EEs in read list found!\n", rr); |
b411b363 | 2145 | |
b30ab791 | 2146 | rr = drbd_free_peer_reqs(device, &device->done_ee); |
b411b363 | 2147 | if (rr) |
d0180171 | 2148 | drbd_err(device, "%d EEs in done list found!\n", rr); |
b411b363 | 2149 | |
b30ab791 | 2150 | rr = drbd_free_peer_reqs(device, &device->net_ee); |
b411b363 | 2151 | if (rr) |
d0180171 | 2152 | drbd_err(device, "%d EEs in net list found!\n", rr); |
b411b363 PR |
2153 | } |
2154 | ||
774b3055 | 2155 | /* caution. no locking. */ |
05a10ec7 | 2156 | void drbd_destroy_device(struct kref *kref) |
b411b363 | 2157 | { |
b30ab791 | 2158 | struct drbd_device *device = container_of(kref, struct drbd_device, kref); |
803ea134 | 2159 | struct drbd_resource *resource = device->resource; |
a8ba0d60 | 2160 | struct drbd_peer_device *peer_device, *tmp_peer_device; |
b411b363 | 2161 | |
b30ab791 | 2162 | del_timer_sync(&device->request_timer); |
dfa8bedb | 2163 | |
b411b363 | 2164 | /* paranoia asserts */ |
0b0ba1ef | 2165 | D_ASSERT(device, device->open_cnt == 0); |
b411b363 PR |
2166 | /* end paranoia asserts */ |
2167 | ||
b411b363 PR |
2168 | /* cleanup stuff that may have been allocated during |
2169 | * device (re-)configuration or state changes */ | |
2170 | ||
b30ab791 AG |
2171 | if (device->this_bdev) |
2172 | bdput(device->this_bdev); | |
b411b363 | 2173 | |
63a7c8ad | 2174 | drbd_backing_dev_free(device, device->ldev); |
b30ab791 | 2175 | device->ldev = NULL; |
b411b363 | 2176 | |
b30ab791 | 2177 | drbd_release_all_peer_reqs(device); |
b411b363 | 2178 | |
b30ab791 AG |
2179 | lc_destroy(device->act_log); |
2180 | lc_destroy(device->resync); | |
b411b363 | 2181 | |
b30ab791 AG |
2182 | kfree(device->p_uuid); |
2183 | /* device->p_uuid = NULL; */ | |
b411b363 | 2184 | |
b30ab791 AG |
2185 | if (device->bitmap) /* should no longer be there. */ |
2186 | drbd_bm_cleanup(device); | |
e37d2438 | 2187 | __free_page(device->md_io.page); |
b30ab791 AG |
2188 | put_disk(device->vdisk); |
2189 | blk_cleanup_queue(device->rq_queue); | |
2190 | kfree(device->rs_plan_s); | |
a8ba0d60 LE |
2191 | |
2192 | /* not for_each_connection(connection, resource): | |
2193 | * those may have been cleaned up and disassociated already. | |
2194 | */ | |
2195 | for_each_peer_device_safe(peer_device, tmp_peer_device, device) { | |
2196 | kref_put(&peer_device->connection->kref, drbd_destroy_connection); | |
2197 | kfree(peer_device); | |
2198 | } | |
c2258ffc | 2199 | memset(device, 0xfd, sizeof(*device)); |
b30ab791 | 2200 | kfree(device); |
803ea134 | 2201 | kref_put(&resource->kref, drbd_destroy_resource); |
b411b363 PR |
2202 | } |
2203 | ||
2312f0b3 LE |
2204 | /* One global retry thread, if we need to push back some bio and have it |
2205 | * reinserted through our make request function. | |
2206 | */ | |
2207 | static struct retry_worker { | |
2208 | struct workqueue_struct *wq; | |
2209 | struct work_struct worker; | |
2210 | ||
2211 | spinlock_t lock; | |
2212 | struct list_head writes; | |
2213 | } retry; | |
2214 | ||
2215 | static void do_retry(struct work_struct *ws) | |
2216 | { | |
2217 | struct retry_worker *retry = container_of(ws, struct retry_worker, worker); | |
2218 | LIST_HEAD(writes); | |
2219 | struct drbd_request *req, *tmp; | |
2220 | ||
2221 | spin_lock_irq(&retry->lock); | |
2222 | list_splice_init(&retry->writes, &writes); | |
2223 | spin_unlock_irq(&retry->lock); | |
2224 | ||
2225 | list_for_each_entry_safe(req, tmp, &writes, tl_requests) { | |
84b8c06b | 2226 | struct drbd_device *device = req->device; |
2312f0b3 | 2227 | struct bio *bio = req->master_bio; |
e5f891b2 | 2228 | unsigned long start_jif = req->start_jif; |
9a278a79 LE |
2229 | bool expected; |
2230 | ||
84b8c06b | 2231 | expected = |
9a278a79 LE |
2232 | expect(atomic_read(&req->completion_ref) == 0) && |
2233 | expect(req->rq_state & RQ_POSTPONED) && | |
2234 | expect((req->rq_state & RQ_LOCAL_PENDING) == 0 || | |
2235 | (req->rq_state & RQ_LOCAL_ABORTED) != 0); | |
2236 | ||
2237 | if (!expected) | |
d0180171 | 2238 | drbd_err(device, "req=%p completion_ref=%d rq_state=%x\n", |
9a278a79 LE |
2239 | req, atomic_read(&req->completion_ref), |
2240 | req->rq_state); | |
2241 | ||
2242 | /* We still need to put one kref associated with the | |
2243 | * "completion_ref" going zero in the code path that queued it | |
2244 | * here. The request object may still be referenced by a | |
2245 | * frozen local req->private_bio, in case we force-detached. | |
2312f0b3 | 2246 | */ |
9a278a79 | 2247 | kref_put(&req->kref, drbd_req_destroy); |
2312f0b3 LE |
2248 | |
2249 | /* A single suspended or otherwise blocking device may stall | |
2250 | * all others as well. Fortunately, this code path is to | |
2251 | * recover from a situation that "should not happen": | |
2252 | * concurrent writes in multi-primary setup. | |
2253 | * In a "normal" lifecycle, this workqueue is supposed to be | |
2254 | * destroyed without ever doing anything. | |
2255 | * If it turns out to be an issue anyways, we can do per | |
2256 | * resource (replication group) or per device (minor) retry | |
2257 | * workqueues instead. | |
2258 | */ | |
2259 | ||
2260 | /* We are not just doing generic_make_request(), | |
2261 | * as we want to keep the start_time information. */ | |
b30ab791 | 2262 | inc_ap_bio(device); |
e5f891b2 | 2263 | __drbd_make_request(device, bio, start_jif); |
2312f0b3 LE |
2264 | } |
2265 | } | |
2266 | ||
844a6ae7 LE |
2267 | /* called via drbd_req_put_completion_ref(), |
2268 | * holds resource->req_lock */ | |
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. */ | |
84b8c06b | 2279 | dec_ap_bio(req->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); |
625a6ba2 | 2290 | free_cpumask_var(resource->cpu_mask); |
77c556f6 | 2291 | kfree(resource->name); |
c2258ffc | 2292 | memset(resource, 0xf2, sizeof(*resource)); |
77c556f6 AG |
2293 | kfree(resource); |
2294 | } | |
2295 | ||
2296 | void drbd_free_resource(struct drbd_resource *resource) | |
2297 | { | |
2298 | struct drbd_connection *connection, *tmp; | |
2299 | ||
2300 | for_each_connection_safe(connection, tmp, resource) { | |
2301 | list_del(&connection->connections); | |
4d3d5aa8 | 2302 | drbd_debugfs_connection_cleanup(connection); |
77c556f6 AG |
2303 | kref_put(&connection->kref, drbd_destroy_connection); |
2304 | } | |
4d3d5aa8 | 2305 | drbd_debugfs_resource_cleanup(resource); |
77c556f6 AG |
2306 | kref_put(&resource->kref, drbd_destroy_resource); |
2307 | } | |
2312f0b3 | 2308 | |
b411b363 PR |
2309 | static void drbd_cleanup(void) |
2310 | { | |
2311 | unsigned int i; | |
b30ab791 | 2312 | struct drbd_device *device; |
77c556f6 | 2313 | struct drbd_resource *resource, *tmp; |
b411b363 | 2314 | |
17a93f30 LE |
2315 | /* first remove proc, |
2316 | * drbdsetup uses it's presence to detect | |
2317 | * whether DRBD is loaded. | |
2318 | * If we would get stuck in proc removal, | |
2319 | * but have netlink already deregistered, | |
2320 | * some drbdsetup commands may wait forever | |
2321 | * for an answer. | |
2322 | */ | |
2323 | if (drbd_proc) | |
2324 | remove_proc_entry("drbd", NULL); | |
2325 | ||
2312f0b3 LE |
2326 | if (retry.wq) |
2327 | destroy_workqueue(retry.wq); | |
b411b363 | 2328 | |
3b98c0c2 | 2329 | drbd_genl_unregister(); |
4d3d5aa8 | 2330 | drbd_debugfs_cleanup(); |
b411b363 | 2331 | |
803ea134 | 2332 | idr_for_each_entry(&drbd_devices, device, i) |
f82795d6 | 2333 | drbd_delete_device(device); |
b411b363 | 2334 | |
c141ebda | 2335 | /* not _rcu since, no other updater anymore. Genl already unregistered */ |
77c556f6 AG |
2336 | for_each_resource_safe(resource, tmp, &drbd_resources) { |
2337 | list_del(&resource->resources); | |
2338 | drbd_free_resource(resource); | |
81fa2e67 | 2339 | } |
b411b363 | 2340 | |
81a5d60e | 2341 | drbd_destroy_mempools(); |
b411b363 PR |
2342 | unregister_blkdev(DRBD_MAJOR, "drbd"); |
2343 | ||
05a10ec7 | 2344 | idr_destroy(&drbd_devices); |
81a5d60e | 2345 | |
f88c5d90 | 2346 | pr_info("module cleanup done.\n"); |
b411b363 PR |
2347 | } |
2348 | ||
2349 | /** | |
d97482ed | 2350 | * drbd_congested() - Callback for the flusher thread |
b411b363 | 2351 | * @congested_data: User data |
d97482ed | 2352 | * @bdi_bits: Bits the BDI flusher thread is currently interested in |
b411b363 | 2353 | * |
4452226e | 2354 | * Returns 1<<WB_async_congested and/or 1<<WB_sync_congested if we are congested. |
b411b363 PR |
2355 | */ |
2356 | static int drbd_congested(void *congested_data, int bdi_bits) | |
2357 | { | |
b30ab791 | 2358 | struct drbd_device *device = congested_data; |
b411b363 PR |
2359 | struct request_queue *q; |
2360 | char reason = '-'; | |
2361 | int r = 0; | |
2362 | ||
b30ab791 | 2363 | if (!may_inc_ap_bio(device)) { |
b411b363 PR |
2364 | /* DRBD has frozen IO */ |
2365 | r = bdi_bits; | |
2366 | reason = 'd'; | |
2367 | goto out; | |
2368 | } | |
2369 | ||
a6b32bc3 | 2370 | if (test_bit(CALLBACK_PENDING, &first_peer_device(device)->connection->flags)) { |
4452226e | 2371 | r |= (1 << WB_async_congested); |
c2ba686f LE |
2372 | /* Without good local data, we would need to read from remote, |
2373 | * and that would need the worker thread as well, which is | |
2374 | * currently blocked waiting for that usermode helper to | |
2375 | * finish. | |
2376 | */ | |
b30ab791 | 2377 | if (!get_ldev_if_state(device, D_UP_TO_DATE)) |
4452226e | 2378 | r |= (1 << WB_sync_congested); |
c2ba686f | 2379 | else |
b30ab791 | 2380 | put_ldev(device); |
c2ba686f LE |
2381 | r &= bdi_bits; |
2382 | reason = 'c'; | |
2383 | goto out; | |
2384 | } | |
2385 | ||
b30ab791 AG |
2386 | if (get_ldev(device)) { |
2387 | q = bdev_get_queue(device->ldev->backing_bdev); | |
b411b363 | 2388 | r = bdi_congested(&q->backing_dev_info, bdi_bits); |
b30ab791 | 2389 | put_ldev(device); |
b411b363 PR |
2390 | if (r) |
2391 | reason = 'b'; | |
2392 | } | |
2393 | ||
4452226e | 2394 | if (bdi_bits & (1 << WB_async_congested) && |
a6b32bc3 | 2395 | test_bit(NET_CONGESTED, &first_peer_device(device)->connection->flags)) { |
4452226e | 2396 | r |= (1 << WB_async_congested); |
b411b363 PR |
2397 | reason = reason == 'b' ? 'a' : 'n'; |
2398 | } | |
2399 | ||
2400 | out: | |
b30ab791 | 2401 | device->congestion_reason = reason; |
b411b363 PR |
2402 | return r; |
2403 | } | |
2404 | ||
6699b655 PR |
2405 | static void drbd_init_workqueue(struct drbd_work_queue* wq) |
2406 | { | |
6699b655 PR |
2407 | spin_lock_init(&wq->q_lock); |
2408 | INIT_LIST_HEAD(&wq->q); | |
8c0785a5 | 2409 | init_waitqueue_head(&wq->q_wait); |
6699b655 PR |
2410 | } |
2411 | ||
b5043c5e AG |
2412 | struct completion_work { |
2413 | struct drbd_work w; | |
2414 | struct completion done; | |
2415 | }; | |
2416 | ||
2417 | static int w_complete(struct drbd_work *w, int cancel) | |
2418 | { | |
2419 | struct completion_work *completion_work = | |
2420 | container_of(w, struct completion_work, w); | |
2421 | ||
2422 | complete(&completion_work->done); | |
2423 | return 0; | |
2424 | } | |
2425 | ||
2426 | void drbd_flush_workqueue(struct drbd_work_queue *work_queue) | |
2427 | { | |
2428 | struct completion_work completion_work; | |
2429 | ||
2430 | completion_work.w.cb = w_complete; | |
2431 | init_completion(&completion_work.done); | |
2432 | drbd_queue_work(work_queue, &completion_work.w); | |
2433 | wait_for_completion(&completion_work.done); | |
2434 | } | |
2435 | ||
4bc76048 | 2436 | struct drbd_resource *drbd_find_resource(const char *name) |
1aba4d7f | 2437 | { |
77c556f6 | 2438 | struct drbd_resource *resource; |
1aba4d7f | 2439 | |
3b98c0c2 LE |
2440 | if (!name || !name[0]) |
2441 | return NULL; | |
2442 | ||
c141ebda | 2443 | rcu_read_lock(); |
77c556f6 AG |
2444 | for_each_resource_rcu(resource, &drbd_resources) { |
2445 | if (!strcmp(resource->name, name)) { | |
4bc76048 | 2446 | kref_get(&resource->kref); |
1aba4d7f | 2447 | goto found; |
0ace9dfa | 2448 | } |
1aba4d7f | 2449 | } |
4bc76048 | 2450 | resource = NULL; |
1aba4d7f | 2451 | found: |
c141ebda | 2452 | rcu_read_unlock(); |
4bc76048 | 2453 | return resource; |
1aba4d7f PR |
2454 | } |
2455 | ||
bde89a9e | 2456 | struct drbd_connection *conn_get_by_addrs(void *my_addr, int my_addr_len, |
089c075d AG |
2457 | void *peer_addr, int peer_addr_len) |
2458 | { | |
77c556f6 | 2459 | struct drbd_resource *resource; |
bde89a9e | 2460 | struct drbd_connection *connection; |
089c075d AG |
2461 | |
2462 | rcu_read_lock(); | |
77c556f6 AG |
2463 | for_each_resource_rcu(resource, &drbd_resources) { |
2464 | for_each_connection_rcu(connection, resource) { | |
2465 | if (connection->my_addr_len == my_addr_len && | |
2466 | connection->peer_addr_len == peer_addr_len && | |
2467 | !memcmp(&connection->my_addr, my_addr, my_addr_len) && | |
2468 | !memcmp(&connection->peer_addr, peer_addr, peer_addr_len)) { | |
2469 | kref_get(&connection->kref); | |
2470 | goto found; | |
2471 | } | |
089c075d AG |
2472 | } |
2473 | } | |
bde89a9e | 2474 | connection = NULL; |
089c075d AG |
2475 | found: |
2476 | rcu_read_unlock(); | |
bde89a9e | 2477 | return connection; |
089c075d AG |
2478 | } |
2479 | ||
e6ef8a5c AG |
2480 | static int drbd_alloc_socket(struct drbd_socket *socket) |
2481 | { | |
2482 | socket->rbuf = (void *) __get_free_page(GFP_KERNEL); | |
2483 | if (!socket->rbuf) | |
2484 | return -ENOMEM; | |
5a87d920 AG |
2485 | socket->sbuf = (void *) __get_free_page(GFP_KERNEL); |
2486 | if (!socket->sbuf) | |
2487 | return -ENOMEM; | |
e6ef8a5c AG |
2488 | return 0; |
2489 | } | |
2490 | ||
2491 | static void drbd_free_socket(struct drbd_socket *socket) | |
2492 | { | |
5a87d920 | 2493 | free_page((unsigned long) socket->sbuf); |
e6ef8a5c AG |
2494 | free_page((unsigned long) socket->rbuf); |
2495 | } | |
2496 | ||
bde89a9e | 2497 | void conn_free_crypto(struct drbd_connection *connection) |
91fd4dad | 2498 | { |
bde89a9e | 2499 | drbd_free_sock(connection); |
1d041225 | 2500 | |
bde89a9e AG |
2501 | crypto_free_hash(connection->csums_tfm); |
2502 | crypto_free_hash(connection->verify_tfm); | |
2503 | crypto_free_hash(connection->cram_hmac_tfm); | |
2504 | crypto_free_hash(connection->integrity_tfm); | |
2505 | crypto_free_hash(connection->peer_integrity_tfm); | |
2506 | kfree(connection->int_dig_in); | |
2507 | kfree(connection->int_dig_vv); | |
1d041225 | 2508 | |
bde89a9e AG |
2509 | connection->csums_tfm = NULL; |
2510 | connection->verify_tfm = NULL; | |
2511 | connection->cram_hmac_tfm = NULL; | |
2512 | connection->integrity_tfm = NULL; | |
2513 | connection->peer_integrity_tfm = NULL; | |
2514 | connection->int_dig_in = NULL; | |
2515 | connection->int_dig_vv = NULL; | |
91fd4dad PR |
2516 | } |
2517 | ||
eb6bea67 | 2518 | int set_resource_options(struct drbd_resource *resource, struct res_opts *res_opts) |
afbbfa88 | 2519 | { |
eb6bea67 | 2520 | struct drbd_connection *connection; |
afbbfa88 AG |
2521 | cpumask_var_t new_cpu_mask; |
2522 | int err; | |
2523 | ||
2524 | if (!zalloc_cpumask_var(&new_cpu_mask, GFP_KERNEL)) | |
2525 | return -ENOMEM; | |
afbbfa88 AG |
2526 | |
2527 | /* silently ignore cpu mask on UP kernel */ | |
2528 | if (nr_cpu_ids > 1 && res_opts->cpu_mask[0] != 0) { | |
f44d0436 | 2529 | err = bitmap_parse(res_opts->cpu_mask, DRBD_CPU_MASK_SIZE, |
c5b005ab | 2530 | cpumask_bits(new_cpu_mask), nr_cpu_ids); |
1e39152f LE |
2531 | if (err == -EOVERFLOW) { |
2532 | /* So what. mask it out. */ | |
2533 | cpumask_var_t tmp_cpu_mask; | |
2534 | if (zalloc_cpumask_var(&tmp_cpu_mask, GFP_KERNEL)) { | |
2535 | cpumask_setall(tmp_cpu_mask); | |
2536 | cpumask_and(new_cpu_mask, new_cpu_mask, tmp_cpu_mask); | |
2537 | drbd_warn(resource, "Overflow in bitmap_parse(%.12s%s), truncating to %u bits\n", | |
2538 | res_opts->cpu_mask, | |
2539 | strlen(res_opts->cpu_mask) > 12 ? "..." : "", | |
2540 | nr_cpu_ids); | |
2541 | free_cpumask_var(tmp_cpu_mask); | |
2542 | err = 0; | |
2543 | } | |
2544 | } | |
afbbfa88 | 2545 | if (err) { |
1ec861eb | 2546 | drbd_warn(resource, "bitmap_parse() failed with %d\n", err); |
afbbfa88 AG |
2547 | /* retcode = ERR_CPU_MASK_PARSE; */ |
2548 | goto fail; | |
2549 | } | |
2550 | } | |
eb6bea67 | 2551 | resource->res_opts = *res_opts; |
625a6ba2 AG |
2552 | if (cpumask_empty(new_cpu_mask)) |
2553 | drbd_calc_cpu_mask(&new_cpu_mask); | |
2554 | if (!cpumask_equal(resource->cpu_mask, new_cpu_mask)) { | |
2555 | cpumask_copy(resource->cpu_mask, new_cpu_mask); | |
2556 | for_each_connection_rcu(connection, resource) { | |
eb6bea67 | 2557 | connection->receiver.reset_cpu_mask = 1; |
1c03e520 | 2558 | connection->ack_receiver.reset_cpu_mask = 1; |
eb6bea67 AG |
2559 | connection->worker.reset_cpu_mask = 1; |
2560 | } | |
afbbfa88 AG |
2561 | } |
2562 | err = 0; | |
2563 | ||
2564 | fail: | |
2565 | free_cpumask_var(new_cpu_mask); | |
2566 | return err; | |
2567 | ||
2568 | } | |
2569 | ||
77c556f6 AG |
2570 | struct drbd_resource *drbd_create_resource(const char *name) |
2571 | { | |
2572 | struct drbd_resource *resource; | |
2573 | ||
6bbf53ca | 2574 | resource = kzalloc(sizeof(struct drbd_resource), GFP_KERNEL); |
77c556f6 | 2575 | if (!resource) |
625a6ba2 | 2576 | goto fail; |
77c556f6 | 2577 | resource->name = kstrdup(name, GFP_KERNEL); |
625a6ba2 AG |
2578 | if (!resource->name) |
2579 | goto fail_free_resource; | |
2580 | if (!zalloc_cpumask_var(&resource->cpu_mask, GFP_KERNEL)) | |
2581 | goto fail_free_name; | |
77c556f6 | 2582 | kref_init(&resource->kref); |
803ea134 | 2583 | idr_init(&resource->devices); |
77c556f6 | 2584 | INIT_LIST_HEAD(&resource->connections); |
f6ba8636 | 2585 | resource->write_ordering = WO_BDEV_FLUSH; |
77c556f6 | 2586 | list_add_tail_rcu(&resource->resources, &drbd_resources); |
0500813f | 2587 | mutex_init(&resource->conf_update); |
9e276872 | 2588 | mutex_init(&resource->adm_mutex); |
0500813f | 2589 | spin_lock_init(&resource->req_lock); |
4d3d5aa8 | 2590 | drbd_debugfs_resource_add(resource); |
77c556f6 | 2591 | return resource; |
625a6ba2 AG |
2592 | |
2593 | fail_free_name: | |
2594 | kfree(resource->name); | |
2595 | fail_free_resource: | |
2596 | kfree(resource); | |
2597 | fail: | |
2598 | return NULL; | |
77c556f6 AG |
2599 | } |
2600 | ||
4d3d5aa8 | 2601 | /* caller must be under adm_mutex */ |
bde89a9e | 2602 | struct drbd_connection *conn_create(const char *name, struct res_opts *res_opts) |
2111438b | 2603 | { |
77c556f6 | 2604 | struct drbd_resource *resource; |
bde89a9e | 2605 | struct drbd_connection *connection; |
2111438b | 2606 | |
bde89a9e AG |
2607 | connection = kzalloc(sizeof(struct drbd_connection), GFP_KERNEL); |
2608 | if (!connection) | |
2111438b PR |
2609 | return NULL; |
2610 | ||
bde89a9e | 2611 | if (drbd_alloc_socket(&connection->data)) |
e6ef8a5c | 2612 | goto fail; |
bde89a9e | 2613 | if (drbd_alloc_socket(&connection->meta)) |
e6ef8a5c AG |
2614 | goto fail; |
2615 | ||
bde89a9e AG |
2616 | connection->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL); |
2617 | if (!connection->current_epoch) | |
12038a3a | 2618 | goto fail; |
b6dd1a89 | 2619 | |
bde89a9e | 2620 | INIT_LIST_HEAD(&connection->transfer_log); |
b6dd1a89 | 2621 | |
bde89a9e AG |
2622 | INIT_LIST_HEAD(&connection->current_epoch->list); |
2623 | connection->epochs = 1; | |
2624 | spin_lock_init(&connection->epoch_lock); | |
4b0007c0 | 2625 | |
bde89a9e AG |
2626 | connection->send.seen_any_write_yet = false; |
2627 | connection->send.current_epoch_nr = 0; | |
2628 | connection->send.current_epoch_writes = 0; | |
b6dd1a89 | 2629 | |
77c556f6 AG |
2630 | resource = drbd_create_resource(name); |
2631 | if (!resource) | |
2632 | goto fail; | |
2633 | ||
bde89a9e AG |
2634 | connection->cstate = C_STANDALONE; |
2635 | mutex_init(&connection->cstate_mutex); | |
bde89a9e | 2636 | init_waitqueue_head(&connection->ping_wait); |
c06ece6b | 2637 | idr_init(&connection->peer_devices); |
b2fb6dbe | 2638 | |
bde89a9e AG |
2639 | drbd_init_workqueue(&connection->sender_work); |
2640 | mutex_init(&connection->data.mutex); | |
2641 | mutex_init(&connection->meta.mutex); | |
6699b655 | 2642 | |
2457b6d5 AG |
2643 | drbd_thread_init(resource, &connection->receiver, drbd_receiver, "receiver"); |
2644 | connection->receiver.connection = connection; | |
2645 | drbd_thread_init(resource, &connection->worker, drbd_worker, "worker"); | |
2646 | connection->worker.connection = connection; | |
1c03e520 PR |
2647 | drbd_thread_init(resource, &connection->ack_receiver, drbd_ack_receiver, "ack_recv"); |
2648 | connection->ack_receiver.connection = connection; | |
392c8801 | 2649 | |
bde89a9e | 2650 | kref_init(&connection->kref); |
77c556f6 | 2651 | |
77c556f6 | 2652 | connection->resource = resource; |
2111438b | 2653 | |
eb6bea67 AG |
2654 | if (set_resource_options(resource, res_opts)) |
2655 | goto fail_resource; | |
2656 | ||
2657 | kref_get(&resource->kref); | |
2658 | list_add_tail_rcu(&connection->connections, &resource->connections); | |
4d3d5aa8 | 2659 | drbd_debugfs_connection_add(connection); |
bde89a9e | 2660 | return connection; |
2111438b | 2661 | |
eb6bea67 AG |
2662 | fail_resource: |
2663 | list_del(&resource->resources); | |
2664 | drbd_free_resource(resource); | |
2111438b | 2665 | fail: |
bde89a9e | 2666 | kfree(connection->current_epoch); |
bde89a9e AG |
2667 | drbd_free_socket(&connection->meta); |
2668 | drbd_free_socket(&connection->data); | |
bde89a9e | 2669 | kfree(connection); |
2111438b PR |
2670 | return NULL; |
2671 | } | |
2672 | ||
05a10ec7 | 2673 | void drbd_destroy_connection(struct kref *kref) |
2111438b | 2674 | { |
bde89a9e | 2675 | struct drbd_connection *connection = container_of(kref, struct drbd_connection, kref); |
77c556f6 | 2676 | struct drbd_resource *resource = connection->resource; |
9dc9fbb3 | 2677 | |
bde89a9e | 2678 | if (atomic_read(&connection->current_epoch->epoch_size) != 0) |
1ec861eb | 2679 | drbd_err(connection, "epoch_size:%d\n", atomic_read(&connection->current_epoch->epoch_size)); |
bde89a9e | 2680 | kfree(connection->current_epoch); |
12038a3a | 2681 | |
c06ece6b | 2682 | idr_destroy(&connection->peer_devices); |
2111438b | 2683 | |
bde89a9e AG |
2684 | drbd_free_socket(&connection->meta); |
2685 | drbd_free_socket(&connection->data); | |
bde89a9e AG |
2686 | kfree(connection->int_dig_in); |
2687 | kfree(connection->int_dig_vv); | |
c2258ffc | 2688 | memset(connection, 0xfc, sizeof(*connection)); |
bde89a9e | 2689 | kfree(connection); |
77c556f6 | 2690 | kref_put(&resource->kref, drbd_destroy_resource); |
2111438b PR |
2691 | } |
2692 | ||
b30ab791 | 2693 | static int init_submitter(struct drbd_device *device) |
113fef9e LE |
2694 | { |
2695 | /* opencoded create_singlethread_workqueue(), | |
2696 | * to be able to say "drbd%d", ..., minor */ | |
39e91a60 LE |
2697 | device->submit.wq = |
2698 | alloc_ordered_workqueue("drbd%u_submit", WQ_MEM_RECLAIM, device->minor); | |
b30ab791 | 2699 | if (!device->submit.wq) |
113fef9e LE |
2700 | return -ENOMEM; |
2701 | ||
b30ab791 | 2702 | INIT_WORK(&device->submit.worker, do_submit); |
b30ab791 | 2703 | INIT_LIST_HEAD(&device->submit.writes); |
113fef9e LE |
2704 | return 0; |
2705 | } | |
2706 | ||
a910b123 | 2707 | enum drbd_ret_code drbd_create_device(struct drbd_config_context *adm_ctx, unsigned int minor) |
b411b363 | 2708 | { |
a910b123 | 2709 | struct drbd_resource *resource = adm_ctx->resource; |
b6f85ef9 | 2710 | struct drbd_connection *connection; |
b30ab791 | 2711 | struct drbd_device *device; |
b6f85ef9 | 2712 | struct drbd_peer_device *peer_device, *tmp_peer_device; |
b411b363 PR |
2713 | struct gendisk *disk; |
2714 | struct request_queue *q; | |
93e4bf7a | 2715 | int id; |
a910b123 | 2716 | int vnr = adm_ctx->volume; |
8432b314 | 2717 | enum drbd_ret_code err = ERR_NOMEM; |
774b3055 | 2718 | |
b30ab791 AG |
2719 | device = minor_to_device(minor); |
2720 | if (device) | |
179e20b8 | 2721 | return ERR_MINOR_OR_VOLUME_EXISTS; |
b411b363 PR |
2722 | |
2723 | /* GFP_KERNEL, we are outside of all write-out paths */ | |
b30ab791 AG |
2724 | device = kzalloc(sizeof(struct drbd_device), GFP_KERNEL); |
2725 | if (!device) | |
774b3055 | 2726 | return ERR_NOMEM; |
803ea134 AG |
2727 | kref_init(&device->kref); |
2728 | ||
803ea134 AG |
2729 | kref_get(&resource->kref); |
2730 | device->resource = resource; | |
b30ab791 AG |
2731 | device->minor = minor; |
2732 | device->vnr = vnr; | |
b411b363 | 2733 | |
b30ab791 | 2734 | drbd_init_set_defaults(device); |
b411b363 PR |
2735 | |
2736 | q = blk_alloc_queue(GFP_KERNEL); | |
2737 | if (!q) | |
2738 | goto out_no_q; | |
b30ab791 AG |
2739 | device->rq_queue = q; |
2740 | q->queuedata = device; | |
b411b363 PR |
2741 | |
2742 | disk = alloc_disk(1); | |
2743 | if (!disk) | |
2744 | goto out_no_disk; | |
b30ab791 | 2745 | device->vdisk = disk; |
b411b363 | 2746 | |
81e84650 | 2747 | set_disk_ro(disk, true); |
b411b363 PR |
2748 | |
2749 | disk->queue = q; | |
2750 | disk->major = DRBD_MAJOR; | |
2751 | disk->first_minor = minor; | |
2752 | disk->fops = &drbd_ops; | |
2753 | sprintf(disk->disk_name, "drbd%d", minor); | |
b30ab791 | 2754 | disk->private_data = device; |
b411b363 | 2755 | |
b30ab791 | 2756 | device->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor)); |
b411b363 | 2757 | /* we have no partitions. we contain only ourselves. */ |
b30ab791 | 2758 | device->this_bdev->bd_contains = device->this_bdev; |
b411b363 PR |
2759 | |
2760 | q->backing_dev_info.congested_fn = drbd_congested; | |
b30ab791 | 2761 | q->backing_dev_info.congested_data = device; |
b411b363 | 2762 | |
2f58dcfc | 2763 | blk_queue_make_request(q, drbd_make_request); |
a73ff323 | 2764 | blk_queue_flush(q, REQ_FLUSH | REQ_FUA); |
99432fcc PR |
2765 | /* Setting the max_hw_sectors to an odd value of 8kibyte here |
2766 | This triggers a max_bio_size message upon first attach or connect */ | |
2767 | blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8); | |
b411b363 | 2768 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
0500813f | 2769 | q->queue_lock = &resource->req_lock; |
b411b363 | 2770 | |
e37d2438 LE |
2771 | device->md_io.page = alloc_page(GFP_KERNEL); |
2772 | if (!device->md_io.page) | |
b411b363 PR |
2773 | goto out_no_io_page; |
2774 | ||
b30ab791 | 2775 | if (drbd_bm_init(device)) |
b411b363 | 2776 | goto out_no_bitmap; |
b30ab791 AG |
2777 | device->read_requests = RB_ROOT; |
2778 | device->write_requests = RB_ROOT; | |
b411b363 | 2779 | |
93e4bf7a AG |
2780 | id = idr_alloc(&drbd_devices, device, minor, minor + 1, GFP_KERNEL); |
2781 | if (id < 0) { | |
f221f4bc | 2782 | if (id == -ENOSPC) |
179e20b8 | 2783 | err = ERR_MINOR_OR_VOLUME_EXISTS; |
8432b314 | 2784 | goto out_no_minor_idr; |
81a5d60e | 2785 | } |
803ea134 AG |
2786 | kref_get(&device->kref); |
2787 | ||
2788 | id = idr_alloc(&resource->devices, device, vnr, vnr + 1, GFP_KERNEL); | |
2789 | if (id < 0) { | |
f221f4bc | 2790 | if (id == -ENOSPC) |
179e20b8 | 2791 | err = ERR_MINOR_OR_VOLUME_EXISTS; |
803ea134 AG |
2792 | goto out_idr_remove_minor; |
2793 | } | |
2794 | kref_get(&device->kref); | |
8432b314 | 2795 | |
b6f85ef9 | 2796 | INIT_LIST_HEAD(&device->peer_devices); |
4ce49266 | 2797 | INIT_LIST_HEAD(&device->pending_bitmap_io); |
b6f85ef9 AG |
2798 | for_each_connection(connection, resource) { |
2799 | peer_device = kzalloc(sizeof(struct drbd_peer_device), GFP_KERNEL); | |
2800 | if (!peer_device) | |
2801 | goto out_idr_remove_from_resource; | |
2802 | peer_device->connection = connection; | |
2803 | peer_device->device = device; | |
2804 | ||
2805 | list_add(&peer_device->peer_devices, &device->peer_devices); | |
2806 | kref_get(&device->kref); | |
2807 | ||
2808 | id = idr_alloc(&connection->peer_devices, peer_device, vnr, vnr + 1, GFP_KERNEL); | |
2809 | if (id < 0) { | |
f221f4bc | 2810 | if (id == -ENOSPC) |
b6f85ef9 | 2811 | err = ERR_INVALID_REQUEST; |
b6f85ef9 | 2812 | goto out_idr_remove_from_resource; |
56de2102 | 2813 | } |
b6f85ef9 | 2814 | kref_get(&connection->kref); |
668700b4 | 2815 | INIT_WORK(&peer_device->send_acks_work, drbd_send_acks_wf); |
8432b314 | 2816 | } |
56de2102 | 2817 | |
b30ab791 | 2818 | if (init_submitter(device)) { |
113fef9e | 2819 | err = ERR_NOMEM; |
113fef9e LE |
2820 | goto out_idr_remove_vol; |
2821 | } | |
2822 | ||
774b3055 PR |
2823 | add_disk(disk); |
2824 | ||
2325eb66 | 2825 | /* inherit the connection state */ |
b6f85ef9 | 2826 | device->state.conn = first_connection(resource)->cstate; |
69a22773 AG |
2827 | if (device->state.conn == C_WF_REPORT_PARAMS) { |
2828 | for_each_peer_device(peer_device, device) | |
2829 | drbd_connected(peer_device); | |
2830 | } | |
4d3d5aa8 LE |
2831 | /* move to create_peer_device() */ |
2832 | for_each_peer_device(peer_device, device) | |
2833 | drbd_debugfs_peer_device_add(peer_device); | |
2834 | drbd_debugfs_device_add(device); | |
774b3055 | 2835 | return NO_ERROR; |
b411b363 | 2836 | |
113fef9e | 2837 | out_idr_remove_vol: |
c06ece6b | 2838 | idr_remove(&connection->peer_devices, vnr); |
803ea134 | 2839 | out_idr_remove_from_resource: |
b6f85ef9 AG |
2840 | for_each_connection(connection, resource) { |
2841 | peer_device = idr_find(&connection->peer_devices, vnr); | |
2842 | if (peer_device) { | |
2843 | idr_remove(&connection->peer_devices, vnr); | |
2844 | kref_put(&connection->kref, drbd_destroy_connection); | |
2845 | } | |
2846 | } | |
2847 | for_each_peer_device_safe(peer_device, tmp_peer_device, device) { | |
2848 | list_del(&peer_device->peer_devices); | |
2849 | kfree(peer_device); | |
2850 | } | |
803ea134 | 2851 | idr_remove(&resource->devices, vnr); |
8432b314 | 2852 | out_idr_remove_minor: |
93e4bf7a | 2853 | idr_remove(&drbd_devices, minor); |
569083c0 | 2854 | synchronize_rcu(); |
8432b314 | 2855 | out_no_minor_idr: |
b30ab791 | 2856 | drbd_bm_cleanup(device); |
b411b363 | 2857 | out_no_bitmap: |
e37d2438 | 2858 | __free_page(device->md_io.page); |
b411b363 PR |
2859 | out_no_io_page: |
2860 | put_disk(disk); | |
2861 | out_no_disk: | |
2862 | blk_cleanup_queue(q); | |
2863 | out_no_q: | |
803ea134 | 2864 | kref_put(&resource->kref, drbd_destroy_resource); |
a6b32bc3 | 2865 | kfree(device); |
8432b314 | 2866 | return err; |
b411b363 PR |
2867 | } |
2868 | ||
f82795d6 | 2869 | void drbd_delete_device(struct drbd_device *device) |
803ea134 AG |
2870 | { |
2871 | struct drbd_resource *resource = device->resource; | |
2872 | struct drbd_connection *connection; | |
4d3d5aa8 | 2873 | struct drbd_peer_device *peer_device; |
803ea134 AG |
2874 | int refs = 3; |
2875 | ||
4d3d5aa8 LE |
2876 | /* move to free_peer_device() */ |
2877 | for_each_peer_device(peer_device, device) | |
2878 | drbd_debugfs_peer_device_cleanup(peer_device); | |
2879 | drbd_debugfs_device_cleanup(device); | |
803ea134 | 2880 | for_each_connection(connection, resource) { |
c06ece6b | 2881 | idr_remove(&connection->peer_devices, device->vnr); |
803ea134 AG |
2882 | refs++; |
2883 | } | |
2884 | idr_remove(&resource->devices, device->vnr); | |
2885 | idr_remove(&drbd_devices, device_to_minor(device)); | |
2886 | del_gendisk(device->vdisk); | |
2887 | synchronize_rcu(); | |
2888 | kref_sub(&device->kref, refs, drbd_destroy_device); | |
2889 | } | |
2890 | ||
8ce953aa | 2891 | static int __init drbd_init(void) |
b411b363 PR |
2892 | { |
2893 | int err; | |
2894 | ||
2b8a90b5 | 2895 | if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) { |
f88c5d90 | 2896 | pr_err("invalid minor_count (%d)\n", minor_count); |
b411b363 PR |
2897 | #ifdef MODULE |
2898 | return -EINVAL; | |
2899 | #else | |
46530e85 | 2900 | minor_count = DRBD_MINOR_COUNT_DEF; |
b411b363 PR |
2901 | #endif |
2902 | } | |
2903 | ||
b411b363 PR |
2904 | err = register_blkdev(DRBD_MAJOR, "drbd"); |
2905 | if (err) { | |
f88c5d90 | 2906 | pr_err("unable to register block device major %d\n", |
b411b363 PR |
2907 | DRBD_MAJOR); |
2908 | return err; | |
2909 | } | |
2910 | ||
b411b363 PR |
2911 | /* |
2912 | * allocate all necessary structs | |
2913 | */ | |
b411b363 PR |
2914 | init_waitqueue_head(&drbd_pp_wait); |
2915 | ||
2916 | drbd_proc = NULL; /* play safe for drbd_cleanup */ | |
05a10ec7 | 2917 | idr_init(&drbd_devices); |
b411b363 | 2918 | |
28bc3b8c | 2919 | mutex_init(&resources_mutex); |
77c556f6 | 2920 | INIT_LIST_HEAD(&drbd_resources); |
69babf05 LE |
2921 | |
2922 | err = drbd_genl_register(); | |
2923 | if (err) { | |
f88c5d90 | 2924 | pr_err("unable to register generic netlink family\n"); |
69babf05 LE |
2925 | goto fail; |
2926 | } | |
2927 | ||
b411b363 PR |
2928 | err = drbd_create_mempools(); |
2929 | if (err) | |
3b98c0c2 | 2930 | goto fail; |
b411b363 | 2931 | |
6110d70b | 2932 | err = -ENOMEM; |
8c484ee4 | 2933 | drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL); |
b411b363 | 2934 | if (!drbd_proc) { |
f88c5d90 | 2935 | pr_err("unable to register proc file\n"); |
3b98c0c2 | 2936 | goto fail; |
b411b363 PR |
2937 | } |
2938 | ||
2312f0b3 LE |
2939 | retry.wq = create_singlethread_workqueue("drbd-reissue"); |
2940 | if (!retry.wq) { | |
f88c5d90 | 2941 | pr_err("unable to create retry workqueue\n"); |
2312f0b3 LE |
2942 | goto fail; |
2943 | } | |
2944 | INIT_WORK(&retry.worker, do_retry); | |
2945 | spin_lock_init(&retry.lock); | |
2946 | INIT_LIST_HEAD(&retry.writes); | |
b411b363 | 2947 | |
4d3d5aa8 LE |
2948 | if (drbd_debugfs_init()) |
2949 | pr_notice("failed to initialize debugfs -- will not be available\n"); | |
2950 | ||
f88c5d90 | 2951 | pr_info("initialized. " |
b411b363 PR |
2952 | "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n", |
2953 | API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX); | |
f88c5d90 LE |
2954 | pr_info("%s\n", drbd_buildtag()); |
2955 | pr_info("registered as block device major %d\n", DRBD_MAJOR); | |
b411b363 PR |
2956 | return 0; /* Success! */ |
2957 | ||
3b98c0c2 | 2958 | fail: |
b411b363 PR |
2959 | drbd_cleanup(); |
2960 | if (err == -ENOMEM) | |
f88c5d90 | 2961 | pr_err("ran out of memory\n"); |
b411b363 | 2962 | else |
f88c5d90 | 2963 | pr_err("initialization failure\n"); |
b411b363 PR |
2964 | return err; |
2965 | } | |
2966 | ||
f418815f | 2967 | static void drbd_free_one_sock(struct drbd_socket *ds) |
b411b363 | 2968 | { |
f418815f LE |
2969 | struct socket *s; |
2970 | mutex_lock(&ds->mutex); | |
2971 | s = ds->socket; | |
2972 | ds->socket = NULL; | |
2973 | mutex_unlock(&ds->mutex); | |
2974 | if (s) { | |
2975 | /* so debugfs does not need to mutex_lock() */ | |
2976 | synchronize_rcu(); | |
2977 | kernel_sock_shutdown(s, SHUT_RDWR); | |
2978 | sock_release(s); | |
b411b363 PR |
2979 | } |
2980 | } | |
2981 | ||
f418815f LE |
2982 | void drbd_free_sock(struct drbd_connection *connection) |
2983 | { | |
2984 | if (connection->data.socket) | |
2985 | drbd_free_one_sock(&connection->data); | |
2986 | if (connection->meta.socket) | |
2987 | drbd_free_one_sock(&connection->meta); | |
2988 | } | |
2989 | ||
b411b363 | 2990 | /* meta data management */ |
b411b363 | 2991 | |
bde89a9e | 2992 | void conn_md_sync(struct drbd_connection *connection) |
b411b363 | 2993 | { |
c06ece6b | 2994 | struct drbd_peer_device *peer_device; |
19fffd7b | 2995 | int vnr; |
b411b363 | 2996 | |
19fffd7b | 2997 | rcu_read_lock(); |
c06ece6b AG |
2998 | idr_for_each_entry(&connection->peer_devices, peer_device, vnr) { |
2999 | struct drbd_device *device = peer_device->device; | |
3000 | ||
b30ab791 | 3001 | kref_get(&device->kref); |
19fffd7b | 3002 | rcu_read_unlock(); |
b30ab791 | 3003 | drbd_md_sync(device); |
05a10ec7 | 3004 | kref_put(&device->kref, drbd_destroy_device); |
19fffd7b PR |
3005 | rcu_read_lock(); |
3006 | } | |
3007 | rcu_read_unlock(); | |
b411b363 PR |
3008 | } |
3009 | ||
ae8bf312 | 3010 | /* aligned 4kByte */ |
b411b363 | 3011 | struct meta_data_on_disk { |
cccac985 | 3012 | u64 la_size_sect; /* last agreed size. */ |
b411b363 PR |
3013 | u64 uuid[UI_SIZE]; /* UUIDs. */ |
3014 | u64 device_uuid; | |
3015 | u64 reserved_u64_1; | |
3016 | u32 flags; /* MDF */ | |
3017 | u32 magic; | |
3018 | u32 md_size_sect; | |
3019 | u32 al_offset; /* offset to this block */ | |
ae8bf312 | 3020 | u32 al_nr_extents; /* important for restoring the AL (userspace) */ |
f399002e | 3021 | /* `-- act_log->nr_elements <-- ldev->dc.al_extents */ |
b411b363 PR |
3022 | u32 bm_offset; /* offset to the bitmap, from here */ |
3023 | u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */ | |
99432fcc | 3024 | u32 la_peer_max_bio_size; /* last peer max_bio_size */ |
b411b363 | 3025 | |
3a4d4eb3 LE |
3026 | /* see al_tr_number_to_on_disk_sector() */ |
3027 | u32 al_stripes; | |
3028 | u32 al_stripe_size_4k; | |
3029 | ||
3030 | u8 reserved_u8[4096 - (7*8 + 10*4)]; | |
b411b363 PR |
3031 | } __packed; |
3032 | ||
d752b269 PR |
3033 | |
3034 | ||
b30ab791 | 3035 | void drbd_md_write(struct drbd_device *device, void *b) |
b411b363 | 3036 | { |
d752b269 | 3037 | struct meta_data_on_disk *buffer = b; |
b411b363 PR |
3038 | sector_t sector; |
3039 | int i; | |
3040 | ||
ae8bf312 | 3041 | memset(buffer, 0, sizeof(*buffer)); |
b411b363 | 3042 | |
b30ab791 | 3043 | buffer->la_size_sect = cpu_to_be64(drbd_get_capacity(device->this_bdev)); |
b411b363 | 3044 | for (i = UI_CURRENT; i < UI_SIZE; i++) |
b30ab791 AG |
3045 | buffer->uuid[i] = cpu_to_be64(device->ldev->md.uuid[i]); |
3046 | buffer->flags = cpu_to_be32(device->ldev->md.flags); | |
d5d7ebd4 | 3047 | buffer->magic = cpu_to_be32(DRBD_MD_MAGIC_84_UNCLEAN); |
b411b363 | 3048 | |
b30ab791 AG |
3049 | buffer->md_size_sect = cpu_to_be32(device->ldev->md.md_size_sect); |
3050 | buffer->al_offset = cpu_to_be32(device->ldev->md.al_offset); | |
3051 | buffer->al_nr_extents = cpu_to_be32(device->act_log->nr_elements); | |
b411b363 | 3052 | buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE); |
b30ab791 | 3053 | buffer->device_uuid = cpu_to_be64(device->ldev->md.device_uuid); |
b411b363 | 3054 | |
b30ab791 AG |
3055 | buffer->bm_offset = cpu_to_be32(device->ldev->md.bm_offset); |
3056 | buffer->la_peer_max_bio_size = cpu_to_be32(device->peer_max_bio_size); | |
b411b363 | 3057 | |
b30ab791 AG |
3058 | buffer->al_stripes = cpu_to_be32(device->ldev->md.al_stripes); |
3059 | buffer->al_stripe_size_4k = cpu_to_be32(device->ldev->md.al_stripe_size_4k); | |
3a4d4eb3 | 3060 | |
0b0ba1ef | 3061 | D_ASSERT(device, drbd_md_ss(device->ldev) == device->ldev->md.md_offset); |
b30ab791 | 3062 | sector = device->ldev->md.md_offset; |
b411b363 | 3063 | |
b30ab791 | 3064 | if (drbd_md_sync_page_io(device, device->ldev, sector, WRITE)) { |
b411b363 | 3065 | /* this was a try anyways ... */ |
d0180171 | 3066 | drbd_err(device, "meta data update failed!\n"); |
b30ab791 | 3067 | drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR); |
b411b363 | 3068 | } |
d752b269 PR |
3069 | } |
3070 | ||
3071 | /** | |
3072 | * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set | |
b30ab791 | 3073 | * @device: DRBD device. |
d752b269 | 3074 | */ |
b30ab791 | 3075 | void drbd_md_sync(struct drbd_device *device) |
d752b269 PR |
3076 | { |
3077 | struct meta_data_on_disk *buffer; | |
3078 | ||
3079 | /* Don't accidentally change the DRBD meta data layout. */ | |
3080 | BUILD_BUG_ON(UI_SIZE != 4); | |
3081 | BUILD_BUG_ON(sizeof(struct meta_data_on_disk) != 4096); | |
3082 | ||
b30ab791 | 3083 | del_timer(&device->md_sync_timer); |
d752b269 | 3084 | /* timer may be rearmed by drbd_md_mark_dirty() now. */ |
b30ab791 | 3085 | if (!test_and_clear_bit(MD_DIRTY, &device->flags)) |
d752b269 PR |
3086 | return; |
3087 | ||
3088 | /* We use here D_FAILED and not D_ATTACHING because we try to write | |
3089 | * metadata even if we detach due to a disk failure! */ | |
b30ab791 | 3090 | if (!get_ldev_if_state(device, D_FAILED)) |
d752b269 PR |
3091 | return; |
3092 | ||
e37d2438 | 3093 | buffer = drbd_md_get_buffer(device, __func__); |
d752b269 PR |
3094 | if (!buffer) |
3095 | goto out; | |
3096 | ||
b30ab791 | 3097 | drbd_md_write(device, buffer); |
b411b363 | 3098 | |
b30ab791 | 3099 | /* Update device->ldev->md.la_size_sect, |
b411b363 | 3100 | * since we updated it on metadata. */ |
b30ab791 | 3101 | device->ldev->md.la_size_sect = drbd_get_capacity(device->this_bdev); |
b411b363 | 3102 | |
b30ab791 | 3103 | drbd_md_put_buffer(device); |
e1711731 | 3104 | out: |
b30ab791 | 3105 | put_ldev(device); |
b411b363 PR |
3106 | } |
3107 | ||
b30ab791 | 3108 | static int check_activity_log_stripe_size(struct drbd_device *device, |
3a4d4eb3 LE |
3109 | struct meta_data_on_disk *on_disk, |
3110 | struct drbd_md *in_core) | |
3111 | { | |
3112 | u32 al_stripes = be32_to_cpu(on_disk->al_stripes); | |
3113 | u32 al_stripe_size_4k = be32_to_cpu(on_disk->al_stripe_size_4k); | |
3114 | u64 al_size_4k; | |
3115 | ||
3116 | /* both not set: default to old fixed size activity log */ | |
3117 | if (al_stripes == 0 && al_stripe_size_4k == 0) { | |
3118 | al_stripes = 1; | |
3119 | al_stripe_size_4k = MD_32kB_SECT/8; | |
3120 | } | |
3121 | ||
3122 | /* some paranoia plausibility checks */ | |
3123 | ||
3124 | /* we need both values to be set */ | |
3125 | if (al_stripes == 0 || al_stripe_size_4k == 0) | |
3126 | goto err; | |
3127 | ||
3128 | al_size_4k = (u64)al_stripes * al_stripe_size_4k; | |
3129 | ||
3130 | /* Upper limit of activity log area, to avoid potential overflow | |
3131 | * problems in al_tr_number_to_on_disk_sector(). As right now, more | |
3132 | * than 72 * 4k blocks total only increases the amount of history, | |
3133 | * limiting this arbitrarily to 16 GB is not a real limitation ;-) */ | |
3134 | if (al_size_4k > (16 * 1024 * 1024/4)) | |
3135 | goto err; | |
3136 | ||
3137 | /* Lower limit: we need at least 8 transaction slots (32kB) | |
3138 | * to not break existing setups */ | |
3139 | if (al_size_4k < MD_32kB_SECT/8) | |
3140 | goto err; | |
3141 | ||
3142 | in_core->al_stripe_size_4k = al_stripe_size_4k; | |
3143 | in_core->al_stripes = al_stripes; | |
3144 | in_core->al_size_4k = al_size_4k; | |
3145 | ||
3146 | return 0; | |
3147 | err: | |
d0180171 | 3148 | drbd_err(device, "invalid activity log striping: al_stripes=%u, al_stripe_size_4k=%u\n", |
3a4d4eb3 LE |
3149 | al_stripes, al_stripe_size_4k); |
3150 | return -EINVAL; | |
3151 | } | |
3152 | ||
b30ab791 | 3153 | static int check_offsets_and_sizes(struct drbd_device *device, struct drbd_backing_dev *bdev) |
c04ccaa6 LE |
3154 | { |
3155 | sector_t capacity = drbd_get_capacity(bdev->md_bdev); | |
3156 | struct drbd_md *in_core = &bdev->md; | |
3157 | s32 on_disk_al_sect; | |
3158 | s32 on_disk_bm_sect; | |
3159 | ||
3160 | /* The on-disk size of the activity log, calculated from offsets, and | |
3161 | * the size of the activity log calculated from the stripe settings, | |
3162 | * should match. | |
3163 | * Though we could relax this a bit: it is ok, if the striped activity log | |
3164 | * fits in the available on-disk activity log size. | |
3165 | * Right now, that would break how resize is implemented. | |
3166 | * TODO: make drbd_determine_dev_size() (and the drbdmeta tool) aware | |
3167 | * of possible unused padding space in the on disk layout. */ | |
3168 | if (in_core->al_offset < 0) { | |
3169 | if (in_core->bm_offset > in_core->al_offset) | |
3170 | goto err; | |
3171 | on_disk_al_sect = -in_core->al_offset; | |
3172 | on_disk_bm_sect = in_core->al_offset - in_core->bm_offset; | |
3173 | } else { | |
3174 | if (in_core->al_offset != MD_4kB_SECT) | |
3175 | goto err; | |
3176 | if (in_core->bm_offset < in_core->al_offset + in_core->al_size_4k * MD_4kB_SECT) | |
3177 | goto err; | |
3178 | ||
3179 | on_disk_al_sect = in_core->bm_offset - MD_4kB_SECT; | |
3180 | on_disk_bm_sect = in_core->md_size_sect - in_core->bm_offset; | |
3181 | } | |
3182 | ||
3183 | /* old fixed size meta data is exactly that: fixed. */ | |
3184 | if (in_core->meta_dev_idx >= 0) { | |
3185 | if (in_core->md_size_sect != MD_128MB_SECT | |
3186 | || in_core->al_offset != MD_4kB_SECT | |
3187 | || in_core->bm_offset != MD_4kB_SECT + MD_32kB_SECT | |
3188 | || in_core->al_stripes != 1 | |
3189 | || in_core->al_stripe_size_4k != MD_32kB_SECT/8) | |
3190 | goto err; | |
3191 | } | |
3192 | ||
3193 | if (capacity < in_core->md_size_sect) | |
3194 | goto err; | |
3195 | if (capacity - in_core->md_size_sect < drbd_md_first_sector(bdev)) | |
3196 | goto err; | |
3197 | ||
3198 | /* should be aligned, and at least 32k */ | |
3199 | if ((on_disk_al_sect & 7) || (on_disk_al_sect < MD_32kB_SECT)) | |
3200 | goto err; | |
3201 | ||
3202 | /* should fit (for now: exactly) into the available on-disk space; | |
3203 | * overflow prevention is in check_activity_log_stripe_size() above. */ | |
3204 | if (on_disk_al_sect != in_core->al_size_4k * MD_4kB_SECT) | |
3205 | goto err; | |
3206 | ||
3207 | /* again, should be aligned */ | |
3208 | if (in_core->bm_offset & 7) | |
3209 | goto err; | |
3210 | ||
3211 | /* FIXME check for device grow with flex external meta data? */ | |
3212 | ||
3213 | /* can the available bitmap space cover the last agreed device size? */ | |
3214 | if (on_disk_bm_sect < (in_core->la_size_sect+7)/MD_4kB_SECT/8/512) | |
3215 | goto err; | |
3216 | ||
3217 | return 0; | |
3218 | ||
3219 | err: | |
d0180171 | 3220 | drbd_err(device, "meta data offsets don't make sense: idx=%d " |
c04ccaa6 LE |
3221 | "al_s=%u, al_sz4k=%u, al_offset=%d, bm_offset=%d, " |
3222 | "md_size_sect=%u, la_size=%llu, md_capacity=%llu\n", | |
3223 | in_core->meta_dev_idx, | |
3224 | in_core->al_stripes, in_core->al_stripe_size_4k, | |
3225 | in_core->al_offset, in_core->bm_offset, in_core->md_size_sect, | |
3226 | (unsigned long long)in_core->la_size_sect, | |
3227 | (unsigned long long)capacity); | |
3228 | ||
3229 | return -EINVAL; | |
3230 | } | |
3231 | ||
3232 | ||
b411b363 PR |
3233 | /** |
3234 | * drbd_md_read() - Reads in the meta data super block | |
b30ab791 | 3235 | * @device: DRBD device. |
b411b363 PR |
3236 | * @bdev: Device from which the meta data should be read in. |
3237 | * | |
3a4d4eb3 | 3238 | * Return NO_ERROR on success, and an enum drbd_ret_code in case |
d5d7ebd4 | 3239 | * something goes wrong. |
3a4d4eb3 | 3240 | * |
c04ccaa6 | 3241 | * Called exactly once during drbd_adm_attach(), while still being D_DISKLESS, |
b30ab791 | 3242 | * even before @bdev is assigned to @device->ldev. |
b411b363 | 3243 | */ |
b30ab791 | 3244 | int drbd_md_read(struct drbd_device *device, struct drbd_backing_dev *bdev) |
b411b363 PR |
3245 | { |
3246 | struct meta_data_on_disk *buffer; | |
d5d7ebd4 | 3247 | u32 magic, flags; |
b411b363 PR |
3248 | int i, rv = NO_ERROR; |
3249 | ||
b30ab791 | 3250 | if (device->state.disk != D_DISKLESS) |
c04ccaa6 | 3251 | return ERR_DISK_CONFIGURED; |
b411b363 | 3252 | |
e37d2438 | 3253 | buffer = drbd_md_get_buffer(device, __func__); |
e1711731 | 3254 | if (!buffer) |
c04ccaa6 | 3255 | return ERR_NOMEM; |
b411b363 | 3256 | |
c04ccaa6 LE |
3257 | /* First, figure out where our meta data superblock is located, |
3258 | * and read it. */ | |
3a4d4eb3 LE |
3259 | bdev->md.meta_dev_idx = bdev->disk_conf->meta_dev_idx; |
3260 | bdev->md.md_offset = drbd_md_ss(bdev); | |
edb5e5f6 LE |
3261 | /* Even for (flexible or indexed) external meta data, |
3262 | * initially restrict us to the 4k superblock for now. | |
3263 | * Affects the paranoia out-of-range access check in drbd_md_sync_page_io(). */ | |
3264 | bdev->md.md_size_sect = 8; | |
b411b363 | 3265 | |
b30ab791 | 3266 | if (drbd_md_sync_page_io(device, bdev, bdev->md.md_offset, READ)) { |
25985edc | 3267 | /* NOTE: can't do normal error processing here as this is |
b411b363 | 3268 | called BEFORE disk is attached */ |
d0180171 | 3269 | drbd_err(device, "Error while reading metadata.\n"); |
b411b363 PR |
3270 | rv = ERR_IO_MD_DISK; |
3271 | goto err; | |
3272 | } | |
3273 | ||
d5d7ebd4 LE |
3274 | magic = be32_to_cpu(buffer->magic); |
3275 | flags = be32_to_cpu(buffer->flags); | |
3276 | if (magic == DRBD_MD_MAGIC_84_UNCLEAN || | |
3277 | (magic == DRBD_MD_MAGIC_08 && !(flags & MDF_AL_CLEAN))) { | |
3278 | /* btw: that's Activity Log clean, not "all" clean. */ | |
d0180171 | 3279 | drbd_err(device, "Found unclean meta data. Did you \"drbdadm apply-al\"?\n"); |
d5d7ebd4 LE |
3280 | rv = ERR_MD_UNCLEAN; |
3281 | goto err; | |
3282 | } | |
3a4d4eb3 LE |
3283 | |
3284 | rv = ERR_MD_INVALID; | |
d5d7ebd4 | 3285 | if (magic != DRBD_MD_MAGIC_08) { |
43de7c85 | 3286 | if (magic == DRBD_MD_MAGIC_07) |
d0180171 | 3287 | drbd_err(device, "Found old (0.7) meta data magic. Did you \"drbdadm create-md\"?\n"); |
d5d7ebd4 | 3288 | else |
d0180171 | 3289 | drbd_err(device, "Meta data magic not found. Did you \"drbdadm create-md\"?\n"); |
b411b363 PR |
3290 | goto err; |
3291 | } | |
3a4d4eb3 | 3292 | |
c04ccaa6 | 3293 | if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) { |
d0180171 | 3294 | drbd_err(device, "unexpected bm_bytes_per_bit: %u (expected %u)\n", |
c04ccaa6 | 3295 | be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE); |
b411b363 PR |
3296 | goto err; |
3297 | } | |
3a4d4eb3 | 3298 | |
c04ccaa6 LE |
3299 | |
3300 | /* convert to in_core endian */ | |
3301 | bdev->md.la_size_sect = be64_to_cpu(buffer->la_size_sect); | |
3302 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
3303 | bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]); | |
3304 | bdev->md.flags = be32_to_cpu(buffer->flags); | |
3305 | bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid); | |
3306 | ||
3307 | bdev->md.md_size_sect = be32_to_cpu(buffer->md_size_sect); | |
3308 | bdev->md.al_offset = be32_to_cpu(buffer->al_offset); | |
3309 | bdev->md.bm_offset = be32_to_cpu(buffer->bm_offset); | |
3310 | ||
b30ab791 | 3311 | if (check_activity_log_stripe_size(device, buffer, &bdev->md)) |
b411b363 | 3312 | goto err; |
b30ab791 | 3313 | if (check_offsets_and_sizes(device, bdev)) |
c04ccaa6 LE |
3314 | goto err; |
3315 | ||
b411b363 | 3316 | if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) { |
d0180171 | 3317 | drbd_err(device, "unexpected bm_offset: %d (expected %d)\n", |
b411b363 | 3318 | be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset); |
b411b363 PR |
3319 | goto err; |
3320 | } | |
3321 | if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) { | |
d0180171 | 3322 | drbd_err(device, "unexpected md_size: %u (expected %u)\n", |
b411b363 | 3323 | be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect); |
b411b363 PR |
3324 | goto err; |
3325 | } | |
3326 | ||
3a4d4eb3 | 3327 | rv = NO_ERROR; |
b411b363 | 3328 | |
0500813f | 3329 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3330 | if (device->state.conn < C_CONNECTED) { |
db141b2f | 3331 | unsigned int peer; |
99432fcc | 3332 | peer = be32_to_cpu(buffer->la_peer_max_bio_size); |
db141b2f | 3333 | peer = max(peer, DRBD_MAX_BIO_SIZE_SAFE); |
b30ab791 | 3334 | device->peer_max_bio_size = peer; |
99432fcc | 3335 | } |
0500813f | 3336 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
3337 | |
3338 | err: | |
b30ab791 | 3339 | drbd_md_put_buffer(device); |
b411b363 PR |
3340 | |
3341 | return rv; | |
3342 | } | |
3343 | ||
3344 | /** | |
3345 | * drbd_md_mark_dirty() - Mark meta data super block as dirty | |
b30ab791 | 3346 | * @device: DRBD device. |
b411b363 PR |
3347 | * |
3348 | * Call this function if you change anything that should be written to | |
3349 | * the meta-data super block. This function sets MD_DIRTY, and starts a | |
3350 | * timer that ensures that within five seconds you have to call drbd_md_sync(). | |
3351 | */ | |
ca0e6098 | 3352 | #ifdef DEBUG |
b30ab791 | 3353 | void drbd_md_mark_dirty_(struct drbd_device *device, unsigned int line, const char *func) |
ee15b038 | 3354 | { |
b30ab791 AG |
3355 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) { |
3356 | mod_timer(&device->md_sync_timer, jiffies + HZ); | |
3357 | device->last_md_mark_dirty.line = line; | |
3358 | device->last_md_mark_dirty.func = func; | |
ee15b038 LE |
3359 | } |
3360 | } | |
3361 | #else | |
b30ab791 | 3362 | void drbd_md_mark_dirty(struct drbd_device *device) |
b411b363 | 3363 | { |
b30ab791 AG |
3364 | if (!test_and_set_bit(MD_DIRTY, &device->flags)) |
3365 | mod_timer(&device->md_sync_timer, jiffies + 5*HZ); | |
b411b363 | 3366 | } |
ee15b038 | 3367 | #endif |
b411b363 | 3368 | |
b30ab791 | 3369 | void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3370 | { |
3371 | int i; | |
3372 | ||
62b0da3a | 3373 | for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) |
b30ab791 | 3374 | device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i]; |
b411b363 PR |
3375 | } |
3376 | ||
b30ab791 | 3377 | void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 PR |
3378 | { |
3379 | if (idx == UI_CURRENT) { | |
b30ab791 | 3380 | if (device->state.role == R_PRIMARY) |
b411b363 PR |
3381 | val |= 1; |
3382 | else | |
3383 | val &= ~((u64)1); | |
3384 | ||
b30ab791 | 3385 | drbd_set_ed_uuid(device, val); |
b411b363 PR |
3386 | } |
3387 | ||
b30ab791 AG |
3388 | device->ldev->md.uuid[idx] = val; |
3389 | drbd_md_mark_dirty(device); | |
b411b363 PR |
3390 | } |
3391 | ||
b30ab791 | 3392 | void _drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
9f2247bb PR |
3393 | { |
3394 | unsigned long flags; | |
b30ab791 AG |
3395 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3396 | __drbd_uuid_set(device, idx, val); | |
3397 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
9f2247bb | 3398 | } |
b411b363 | 3399 | |
b30ab791 | 3400 | void drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local) |
b411b363 | 3401 | { |
9f2247bb | 3402 | unsigned long flags; |
b30ab791 AG |
3403 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
3404 | if (device->ldev->md.uuid[idx]) { | |
3405 | drbd_uuid_move_history(device); | |
3406 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[idx]; | |
b411b363 | 3407 | } |
b30ab791 AG |
3408 | __drbd_uuid_set(device, idx, val); |
3409 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); | |
b411b363 PR |
3410 | } |
3411 | ||
3412 | /** | |
3413 | * drbd_uuid_new_current() - Creates a new current UUID | |
b30ab791 | 3414 | * @device: DRBD device. |
b411b363 PR |
3415 | * |
3416 | * Creates a new current UUID, and rotates the old current UUID into | |
3417 | * the bitmap slot. Causes an incremental resync upon next connect. | |
3418 | */ | |
b30ab791 | 3419 | void drbd_uuid_new_current(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3420 | { |
3421 | u64 val; | |
9f2247bb PR |
3422 | unsigned long long bm_uuid; |
3423 | ||
3424 | get_random_bytes(&val, sizeof(u64)); | |
3425 | ||
b30ab791 AG |
3426 | spin_lock_irq(&device->ldev->md.uuid_lock); |
3427 | bm_uuid = device->ldev->md.uuid[UI_BITMAP]; | |
62b0da3a LE |
3428 | |
3429 | if (bm_uuid) | |
d0180171 | 3430 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3431 | |
b30ab791 AG |
3432 | device->ldev->md.uuid[UI_BITMAP] = device->ldev->md.uuid[UI_CURRENT]; |
3433 | __drbd_uuid_set(device, UI_CURRENT, val); | |
3434 | spin_unlock_irq(&device->ldev->md.uuid_lock); | |
b411b363 | 3435 | |
b30ab791 | 3436 | drbd_print_uuids(device, "new current UUID"); |
aaa8e2b3 | 3437 | /* get it to stable storage _now_ */ |
b30ab791 | 3438 | drbd_md_sync(device); |
b411b363 PR |
3439 | } |
3440 | ||
b30ab791 | 3441 | void drbd_uuid_set_bm(struct drbd_device *device, u64 val) __must_hold(local) |
b411b363 | 3442 | { |
9f2247bb | 3443 | unsigned long flags; |
b30ab791 | 3444 | if (device->ldev->md.uuid[UI_BITMAP] == 0 && val == 0) |
b411b363 PR |
3445 | return; |
3446 | ||
b30ab791 | 3447 | spin_lock_irqsave(&device->ldev->md.uuid_lock, flags); |
b411b363 | 3448 | if (val == 0) { |
b30ab791 AG |
3449 | drbd_uuid_move_history(device); |
3450 | device->ldev->md.uuid[UI_HISTORY_START] = device->ldev->md.uuid[UI_BITMAP]; | |
3451 | device->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 3452 | } else { |
b30ab791 | 3453 | unsigned long long bm_uuid = device->ldev->md.uuid[UI_BITMAP]; |
62b0da3a | 3454 | if (bm_uuid) |
d0180171 | 3455 | drbd_warn(device, "bm UUID was already set: %llX\n", bm_uuid); |
b411b363 | 3456 | |
b30ab791 | 3457 | device->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1); |
b411b363 | 3458 | } |
b30ab791 | 3459 | spin_unlock_irqrestore(&device->ldev->md.uuid_lock, flags); |
9f2247bb | 3460 | |
b30ab791 | 3461 | drbd_md_mark_dirty(device); |
b411b363 PR |
3462 | } |
3463 | ||
3464 | /** | |
3465 | * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3466 | * @device: DRBD device. |
b411b363 PR |
3467 | * |
3468 | * Sets all bits in the bitmap and writes the whole bitmap to stable storage. | |
3469 | */ | |
8fe39aac | 3470 | int drbd_bmio_set_n_write(struct drbd_device *device) __must_hold(local) |
b411b363 PR |
3471 | { |
3472 | int rv = -EIO; | |
3473 | ||
8fe39aac PR |
3474 | drbd_md_set_flag(device, MDF_FULL_SYNC); |
3475 | drbd_md_sync(device); | |
3476 | drbd_bm_set_all(device); | |
b411b363 | 3477 | |
8fe39aac | 3478 | rv = drbd_bm_write(device); |
b411b363 | 3479 | |
8fe39aac PR |
3480 | if (!rv) { |
3481 | drbd_md_clear_flag(device, MDF_FULL_SYNC); | |
3482 | drbd_md_sync(device); | |
b411b363 PR |
3483 | } |
3484 | ||
3485 | return rv; | |
3486 | } | |
3487 | ||
3488 | /** | |
3489 | * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
b30ab791 | 3490 | * @device: DRBD device. |
b411b363 PR |
3491 | * |
3492 | * Clears all bits in the bitmap and writes the whole bitmap to stable storage. | |
3493 | */ | |
8fe39aac | 3494 | int drbd_bmio_clear_n_write(struct drbd_device *device) __must_hold(local) |
b411b363 | 3495 | { |
b30ab791 | 3496 | drbd_resume_al(device); |
8fe39aac PR |
3497 | drbd_bm_clear_all(device); |
3498 | return drbd_bm_write(device); | |
b411b363 PR |
3499 | } |
3500 | ||
99920dc5 | 3501 | static int w_bitmap_io(struct drbd_work *w, int unused) |
b411b363 | 3502 | { |
84b8c06b AG |
3503 | struct drbd_device *device = |
3504 | container_of(w, struct drbd_device, bm_io_work.w); | |
3505 | struct bm_io_work *work = &device->bm_io_work; | |
02851e9f | 3506 | int rv = -EIO; |
b411b363 | 3507 | |
0b0ba1ef | 3508 | D_ASSERT(device, atomic_read(&device->ap_bio_cnt) == 0); |
b411b363 | 3509 | |
b30ab791 AG |
3510 | if (get_ldev(device)) { |
3511 | drbd_bm_lock(device, work->why, work->flags); | |
3512 | rv = work->io_fn(device); | |
3513 | drbd_bm_unlock(device); | |
3514 | put_ldev(device); | |
02851e9f | 3515 | } |
b411b363 | 3516 | |
b30ab791 AG |
3517 | clear_bit_unlock(BITMAP_IO, &device->flags); |
3518 | wake_up(&device->misc_wait); | |
b411b363 PR |
3519 | |
3520 | if (work->done) | |
b30ab791 | 3521 | work->done(device, rv); |
b411b363 | 3522 | |
b30ab791 | 3523 | clear_bit(BITMAP_IO_QUEUED, &device->flags); |
b411b363 | 3524 | work->why = NULL; |
20ceb2b2 | 3525 | work->flags = 0; |
b411b363 | 3526 | |
99920dc5 | 3527 | return 0; |
b411b363 PR |
3528 | } |
3529 | ||
3530 | /** | |
3531 | * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap | |
b30ab791 | 3532 | * @device: DRBD device. |
b411b363 PR |
3533 | * @io_fn: IO callback to be called when bitmap IO is possible |
3534 | * @done: callback to be called after the bitmap IO was performed | |
3535 | * @why: Descriptive text of the reason for doing the IO | |
3536 | * | |
3537 | * While IO on the bitmap happens we freeze application IO thus we ensure | |
3538 | * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be | |
3539 | * called from worker context. It MUST NOT be used while a previous such | |
3540 | * work is still pending! | |
8fe39aac PR |
3541 | * |
3542 | * Its worker function encloses the call of io_fn() by get_ldev() and | |
3543 | * put_ldev(). | |
b411b363 | 3544 | */ |
b30ab791 | 3545 | void drbd_queue_bitmap_io(struct drbd_device *device, |
54761697 AG |
3546 | int (*io_fn)(struct drbd_device *), |
3547 | void (*done)(struct drbd_device *, int), | |
20ceb2b2 | 3548 | char *why, enum bm_flag flags) |
b411b363 | 3549 | { |
0b0ba1ef | 3550 | D_ASSERT(device, current == first_peer_device(device)->connection->worker.task); |
b411b363 | 3551 | |
0b0ba1ef AG |
3552 | D_ASSERT(device, !test_bit(BITMAP_IO_QUEUED, &device->flags)); |
3553 | D_ASSERT(device, !test_bit(BITMAP_IO, &device->flags)); | |
3554 | D_ASSERT(device, list_empty(&device->bm_io_work.w.list)); | |
b30ab791 | 3555 | if (device->bm_io_work.why) |
d0180171 | 3556 | drbd_err(device, "FIXME going to queue '%s' but '%s' still pending?\n", |
b30ab791 | 3557 | why, device->bm_io_work.why); |
b411b363 | 3558 | |
b30ab791 AG |
3559 | device->bm_io_work.io_fn = io_fn; |
3560 | device->bm_io_work.done = done; | |
3561 | device->bm_io_work.why = why; | |
3562 | device->bm_io_work.flags = flags; | |
b411b363 | 3563 | |
0500813f | 3564 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 AG |
3565 | set_bit(BITMAP_IO, &device->flags); |
3566 | if (atomic_read(&device->ap_bio_cnt) == 0) { | |
3567 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &device->flags)) | |
84b8c06b AG |
3568 | drbd_queue_work(&first_peer_device(device)->connection->sender_work, |
3569 | &device->bm_io_work.w); | |
b411b363 | 3570 | } |
0500813f | 3571 | spin_unlock_irq(&device->resource->req_lock); |
b411b363 PR |
3572 | } |
3573 | ||
3574 | /** | |
3575 | * drbd_bitmap_io() - Does an IO operation on the whole bitmap | |
b30ab791 | 3576 | * @device: DRBD device. |
b411b363 PR |
3577 | * @io_fn: IO callback to be called when bitmap IO is possible |
3578 | * @why: Descriptive text of the reason for doing the IO | |
3579 | * | |
3580 | * freezes application IO while that the actual IO operations runs. This | |
3581 | * functions MAY NOT be called from worker context. | |
3582 | */ | |
b30ab791 | 3583 | int drbd_bitmap_io(struct drbd_device *device, int (*io_fn)(struct drbd_device *), |
20ceb2b2 | 3584 | char *why, enum bm_flag flags) |
b411b363 PR |
3585 | { |
3586 | int rv; | |
3587 | ||
0b0ba1ef | 3588 | D_ASSERT(device, current != first_peer_device(device)->connection->worker.task); |
b411b363 | 3589 | |
20ceb2b2 | 3590 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
b30ab791 | 3591 | drbd_suspend_io(device); |
b411b363 | 3592 | |
b30ab791 AG |
3593 | drbd_bm_lock(device, why, flags); |
3594 | rv = io_fn(device); | |
3595 | drbd_bm_unlock(device); | |
b411b363 | 3596 | |
20ceb2b2 | 3597 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
b30ab791 | 3598 | drbd_resume_io(device); |
b411b363 PR |
3599 | |
3600 | return rv; | |
3601 | } | |
3602 | ||
b30ab791 | 3603 | void drbd_md_set_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3604 | { |
b30ab791 AG |
3605 | if ((device->ldev->md.flags & flag) != flag) { |
3606 | drbd_md_mark_dirty(device); | |
3607 | device->ldev->md.flags |= flag; | |
b411b363 PR |
3608 | } |
3609 | } | |
3610 | ||
b30ab791 | 3611 | void drbd_md_clear_flag(struct drbd_device *device, int flag) __must_hold(local) |
b411b363 | 3612 | { |
b30ab791 AG |
3613 | if ((device->ldev->md.flags & flag) != 0) { |
3614 | drbd_md_mark_dirty(device); | |
3615 | device->ldev->md.flags &= ~flag; | |
b411b363 PR |
3616 | } |
3617 | } | |
3618 | int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag) | |
3619 | { | |
3620 | return (bdev->md.flags & flag) != 0; | |
3621 | } | |
3622 | ||
3623 | static void md_sync_timer_fn(unsigned long data) | |
3624 | { | |
b30ab791 | 3625 | struct drbd_device *device = (struct drbd_device *) data; |
ac0acb9e | 3626 | drbd_device_post_work(device, MD_SYNC); |
b411b363 PR |
3627 | } |
3628 | ||
d8763023 | 3629 | const char *cmdname(enum drbd_packet cmd) |
f2ad9063 AG |
3630 | { |
3631 | /* THINK may need to become several global tables | |
3632 | * when we want to support more than | |
3633 | * one PRO_VERSION */ | |
3634 | static const char *cmdnames[] = { | |
3635 | [P_DATA] = "Data", | |
3636 | [P_DATA_REPLY] = "DataReply", | |
3637 | [P_RS_DATA_REPLY] = "RSDataReply", | |
3638 | [P_BARRIER] = "Barrier", | |
3639 | [P_BITMAP] = "ReportBitMap", | |
3640 | [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", | |
3641 | [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", | |
3642 | [P_UNPLUG_REMOTE] = "UnplugRemote", | |
3643 | [P_DATA_REQUEST] = "DataRequest", | |
3644 | [P_RS_DATA_REQUEST] = "RSDataRequest", | |
3645 | [P_SYNC_PARAM] = "SyncParam", | |
3646 | [P_SYNC_PARAM89] = "SyncParam89", | |
3647 | [P_PROTOCOL] = "ReportProtocol", | |
3648 | [P_UUIDS] = "ReportUUIDs", | |
3649 | [P_SIZES] = "ReportSizes", | |
3650 | [P_STATE] = "ReportState", | |
3651 | [P_SYNC_UUID] = "ReportSyncUUID", | |
3652 | [P_AUTH_CHALLENGE] = "AuthChallenge", | |
3653 | [P_AUTH_RESPONSE] = "AuthResponse", | |
3654 | [P_PING] = "Ping", | |
3655 | [P_PING_ACK] = "PingAck", | |
3656 | [P_RECV_ACK] = "RecvAck", | |
3657 | [P_WRITE_ACK] = "WriteAck", | |
3658 | [P_RS_WRITE_ACK] = "RSWriteAck", | |
d4dabbe2 | 3659 | [P_SUPERSEDED] = "Superseded", |
f2ad9063 AG |
3660 | [P_NEG_ACK] = "NegAck", |
3661 | [P_NEG_DREPLY] = "NegDReply", | |
3662 | [P_NEG_RS_DREPLY] = "NegRSDReply", | |
3663 | [P_BARRIER_ACK] = "BarrierAck", | |
3664 | [P_STATE_CHG_REQ] = "StateChgRequest", | |
3665 | [P_STATE_CHG_REPLY] = "StateChgReply", | |
3666 | [P_OV_REQUEST] = "OVRequest", | |
3667 | [P_OV_REPLY] = "OVReply", | |
3668 | [P_OV_RESULT] = "OVResult", | |
3669 | [P_CSUM_RS_REQUEST] = "CsumRSRequest", | |
3670 | [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", | |
3671 | [P_COMPRESSED_BITMAP] = "CBitmap", | |
3672 | [P_DELAY_PROBE] = "DelayProbe", | |
3673 | [P_OUT_OF_SYNC] = "OutOfSync", | |
7be8da07 | 3674 | [P_RETRY_WRITE] = "RetryWrite", |
ae25b336 LE |
3675 | [P_RS_CANCEL] = "RSCancel", |
3676 | [P_CONN_ST_CHG_REQ] = "conn_st_chg_req", | |
3677 | [P_CONN_ST_CHG_REPLY] = "conn_st_chg_reply", | |
036b17ea PR |
3678 | [P_RETRY_WRITE] = "retry_write", |
3679 | [P_PROTOCOL_UPDATE] = "protocol_update", | |
ae25b336 LE |
3680 | |
3681 | /* enum drbd_packet, but not commands - obsoleted flags: | |
3682 | * P_MAY_IGNORE | |
3683 | * P_MAX_OPT_CMD | |
3684 | */ | |
f2ad9063 AG |
3685 | }; |
3686 | ||
ae25b336 | 3687 | /* too big for the array: 0xfffX */ |
e5d6f33a AG |
3688 | if (cmd == P_INITIAL_META) |
3689 | return "InitialMeta"; | |
3690 | if (cmd == P_INITIAL_DATA) | |
3691 | return "InitialData"; | |
6038178e AG |
3692 | if (cmd == P_CONNECTION_FEATURES) |
3693 | return "ConnectionFeatures"; | |
6e849ce8 | 3694 | if (cmd >= ARRAY_SIZE(cmdnames)) |
f2ad9063 AG |
3695 | return "Unknown"; |
3696 | return cmdnames[cmd]; | |
3697 | } | |
3698 | ||
7be8da07 AG |
3699 | /** |
3700 | * drbd_wait_misc - wait for a request to make progress | |
b30ab791 | 3701 | * @device: device associated with the request |
7be8da07 AG |
3702 | * @i: the struct drbd_interval embedded in struct drbd_request or |
3703 | * struct drbd_peer_request | |
3704 | */ | |
b30ab791 | 3705 | int drbd_wait_misc(struct drbd_device *device, struct drbd_interval *i) |
7be8da07 | 3706 | { |
44ed167d | 3707 | struct net_conf *nc; |
7be8da07 AG |
3708 | DEFINE_WAIT(wait); |
3709 | long timeout; | |
3710 | ||
44ed167d | 3711 | rcu_read_lock(); |
a6b32bc3 | 3712 | nc = rcu_dereference(first_peer_device(device)->connection->net_conf); |
44ed167d PR |
3713 | if (!nc) { |
3714 | rcu_read_unlock(); | |
7be8da07 | 3715 | return -ETIMEDOUT; |
44ed167d PR |
3716 | } |
3717 | timeout = nc->ko_count ? nc->timeout * HZ / 10 * nc->ko_count : MAX_SCHEDULE_TIMEOUT; | |
3718 | rcu_read_unlock(); | |
7be8da07 | 3719 | |
b30ab791 | 3720 | /* Indicate to wake up device->misc_wait on progress. */ |
7be8da07 | 3721 | i->waiting = true; |
b30ab791 | 3722 | prepare_to_wait(&device->misc_wait, &wait, TASK_INTERRUPTIBLE); |
0500813f | 3723 | spin_unlock_irq(&device->resource->req_lock); |
7be8da07 | 3724 | timeout = schedule_timeout(timeout); |
b30ab791 | 3725 | finish_wait(&device->misc_wait, &wait); |
0500813f | 3726 | spin_lock_irq(&device->resource->req_lock); |
b30ab791 | 3727 | if (!timeout || device->state.conn < C_CONNECTED) |
7be8da07 AG |
3728 | return -ETIMEDOUT; |
3729 | if (signal_pending(current)) | |
3730 | return -ERESTARTSYS; | |
3731 | return 0; | |
b411b363 PR |
3732 | } |
3733 | ||
28bc3b8c AG |
3734 | void lock_all_resources(void) |
3735 | { | |
3736 | struct drbd_resource *resource; | |
3737 | int __maybe_unused i = 0; | |
3738 | ||
3739 | mutex_lock(&resources_mutex); | |
3740 | local_irq_disable(); | |
3741 | for_each_resource(resource, &drbd_resources) | |
3742 | spin_lock_nested(&resource->req_lock, i++); | |
3743 | } | |
3744 | ||
3745 | void unlock_all_resources(void) | |
3746 | { | |
3747 | struct drbd_resource *resource; | |
3748 | ||
3749 | for_each_resource(resource, &drbd_resources) | |
3750 | spin_unlock(&resource->req_lock); | |
3751 | local_irq_enable(); | |
3752 | mutex_unlock(&resources_mutex); | |
3753 | } | |
3754 | ||
b411b363 PR |
3755 | #ifdef CONFIG_DRBD_FAULT_INJECTION |
3756 | /* Fault insertion support including random number generator shamelessly | |
3757 | * stolen from kernel/rcutorture.c */ | |
3758 | struct fault_random_state { | |
3759 | unsigned long state; | |
3760 | unsigned long count; | |
3761 | }; | |
3762 | ||
3763 | #define FAULT_RANDOM_MULT 39916801 /* prime */ | |
3764 | #define FAULT_RANDOM_ADD 479001701 /* prime */ | |
3765 | #define FAULT_RANDOM_REFRESH 10000 | |
3766 | ||
3767 | /* | |
3768 | * Crude but fast random-number generator. Uses a linear congruential | |
3769 | * generator, with occasional help from get_random_bytes(). | |
3770 | */ | |
3771 | static unsigned long | |
3772 | _drbd_fault_random(struct fault_random_state *rsp) | |
3773 | { | |
3774 | long refresh; | |
3775 | ||
49829ea7 | 3776 | if (!rsp->count--) { |
b411b363 PR |
3777 | get_random_bytes(&refresh, sizeof(refresh)); |
3778 | rsp->state += refresh; | |
3779 | rsp->count = FAULT_RANDOM_REFRESH; | |
3780 | } | |
3781 | rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD; | |
3782 | return swahw32(rsp->state); | |
3783 | } | |
3784 | ||
3785 | static char * | |
3786 | _drbd_fault_str(unsigned int type) { | |
3787 | static char *_faults[] = { | |
3788 | [DRBD_FAULT_MD_WR] = "Meta-data write", | |
3789 | [DRBD_FAULT_MD_RD] = "Meta-data read", | |
3790 | [DRBD_FAULT_RS_WR] = "Resync write", | |
3791 | [DRBD_FAULT_RS_RD] = "Resync read", | |
3792 | [DRBD_FAULT_DT_WR] = "Data write", | |
3793 | [DRBD_FAULT_DT_RD] = "Data read", | |
3794 | [DRBD_FAULT_DT_RA] = "Data read ahead", | |
3795 | [DRBD_FAULT_BM_ALLOC] = "BM allocation", | |
6b4388ac PR |
3796 | [DRBD_FAULT_AL_EE] = "EE allocation", |
3797 | [DRBD_FAULT_RECEIVE] = "receive data corruption", | |
b411b363 PR |
3798 | }; |
3799 | ||
3800 | return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**"; | |
3801 | } | |
3802 | ||
3803 | unsigned int | |
b30ab791 | 3804 | _drbd_insert_fault(struct drbd_device *device, unsigned int type) |
b411b363 PR |
3805 | { |
3806 | static struct fault_random_state rrs = {0, 0}; | |
3807 | ||
3808 | unsigned int ret = ( | |
3809 | (fault_devs == 0 || | |
b30ab791 | 3810 | ((1 << device_to_minor(device)) & fault_devs) != 0) && |
b411b363 PR |
3811 | (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate)); |
3812 | ||
3813 | if (ret) { | |
3814 | fault_count++; | |
3815 | ||
7383506c | 3816 | if (__ratelimit(&drbd_ratelimit_state)) |
d0180171 | 3817 | drbd_warn(device, "***Simulating %s failure\n", |
b411b363 PR |
3818 | _drbd_fault_str(type)); |
3819 | } | |
3820 | ||
3821 | return ret; | |
3822 | } | |
3823 | #endif | |
3824 | ||
3825 | const char *drbd_buildtag(void) | |
3826 | { | |
3827 | /* DRBD built from external sources has here a reference to the | |
3828 | git hash of the source code. */ | |
3829 | ||
3830 | static char buildtag[38] = "\0uilt-in"; | |
3831 | ||
3832 | if (buildtag[0] == 0) { | |
bc4854bc CW |
3833 | #ifdef MODULE |
3834 | sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion); | |
3835 | #else | |
3836 | buildtag[0] = 'b'; | |
b411b363 | 3837 | #endif |
b411b363 PR |
3838 | } |
3839 | ||
3840 | return buildtag; | |
3841 | } | |
3842 | ||
3843 | module_init(drbd_init) | |
3844 | module_exit(drbd_cleanup) | |
3845 | ||
b411b363 PR |
3846 | EXPORT_SYMBOL(drbd_conn_str); |
3847 | EXPORT_SYMBOL(drbd_role_str); | |
3848 | EXPORT_SYMBOL(drbd_disk_str); | |
3849 | EXPORT_SYMBOL(drbd_set_st_err_str); |