Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | Thanks to Carter Burden, Bart Grantham and Gennadiy Nerubayev | |
11 | from Logicworks, Inc. for making SDP replication support possible. | |
12 | ||
13 | drbd is free software; you can redistribute it and/or modify | |
14 | it under the terms of the GNU General Public License as published by | |
15 | the Free Software Foundation; either version 2, or (at your option) | |
16 | any later version. | |
17 | ||
18 | drbd is distributed in the hope that it will be useful, | |
19 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | GNU General Public License for more details. | |
22 | ||
23 | You should have received a copy of the GNU General Public License | |
24 | along with drbd; see the file COPYING. If not, write to | |
25 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
26 | ||
27 | */ | |
28 | ||
b411b363 | 29 | #include <linux/module.h> |
b411b363 PR |
30 | #include <linux/drbd.h> |
31 | #include <asm/uaccess.h> | |
32 | #include <asm/types.h> | |
33 | #include <net/sock.h> | |
34 | #include <linux/ctype.h> | |
2a48fc0a | 35 | #include <linux/mutex.h> |
b411b363 PR |
36 | #include <linux/fs.h> |
37 | #include <linux/file.h> | |
38 | #include <linux/proc_fs.h> | |
39 | #include <linux/init.h> | |
40 | #include <linux/mm.h> | |
41 | #include <linux/memcontrol.h> | |
42 | #include <linux/mm_inline.h> | |
43 | #include <linux/slab.h> | |
44 | #include <linux/random.h> | |
45 | #include <linux/reboot.h> | |
46 | #include <linux/notifier.h> | |
47 | #include <linux/kthread.h> | |
48 | ||
49 | #define __KERNEL_SYSCALLS__ | |
50 | #include <linux/unistd.h> | |
51 | #include <linux/vmalloc.h> | |
52 | ||
53 | #include <linux/drbd_limits.h> | |
54 | #include "drbd_int.h" | |
b411b363 PR |
55 | #include "drbd_req.h" /* only for _req_mod in tl_release and tl_clear */ |
56 | ||
57 | #include "drbd_vli.h" | |
58 | ||
2a48fc0a | 59 | static DEFINE_MUTEX(drbd_main_mutex); |
b411b363 PR |
60 | int drbdd_init(struct drbd_thread *); |
61 | int drbd_worker(struct drbd_thread *); | |
62 | int drbd_asender(struct drbd_thread *); | |
63 | ||
64 | int drbd_init(void); | |
65 | static int drbd_open(struct block_device *bdev, fmode_t mode); | |
66 | static int drbd_release(struct gendisk *gd, fmode_t mode); | |
00d56944 | 67 | static int w_md_sync(struct drbd_work *w, int unused); |
b411b363 | 68 | static void md_sync_timer_fn(unsigned long data); |
00d56944 PR |
69 | static int w_bitmap_io(struct drbd_work *w, int unused); |
70 | static int w_go_diskless(struct drbd_work *w, int unused); | |
b411b363 | 71 | |
b411b363 PR |
72 | MODULE_AUTHOR("Philipp Reisner <phil@linbit.com>, " |
73 | "Lars Ellenberg <lars@linbit.com>"); | |
74 | MODULE_DESCRIPTION("drbd - Distributed Replicated Block Device v" REL_VERSION); | |
75 | MODULE_VERSION(REL_VERSION); | |
76 | MODULE_LICENSE("GPL"); | |
81a5d60e | 77 | MODULE_PARM_DESC(minor_count, "Approximate number of drbd devices (" |
2b8a90b5 | 78 | __stringify(DRBD_MINOR_COUNT_MIN) "-" __stringify(DRBD_MINOR_COUNT_MAX) ")"); |
b411b363 PR |
79 | MODULE_ALIAS_BLOCKDEV_MAJOR(DRBD_MAJOR); |
80 | ||
81 | #include <linux/moduleparam.h> | |
82 | /* allow_open_on_secondary */ | |
83 | MODULE_PARM_DESC(allow_oos, "DONT USE!"); | |
84 | /* thanks to these macros, if compiled into the kernel (not-module), | |
85 | * this becomes the boot parameter drbd.minor_count */ | |
86 | module_param(minor_count, uint, 0444); | |
87 | module_param(disable_sendpage, bool, 0644); | |
88 | module_param(allow_oos, bool, 0); | |
b411b363 PR |
89 | module_param(proc_details, int, 0644); |
90 | ||
91 | #ifdef CONFIG_DRBD_FAULT_INJECTION | |
92 | int enable_faults; | |
93 | int fault_rate; | |
94 | static int fault_count; | |
95 | int fault_devs; | |
96 | /* bitmap of enabled faults */ | |
97 | module_param(enable_faults, int, 0664); | |
98 | /* fault rate % value - applies to all enabled faults */ | |
99 | module_param(fault_rate, int, 0664); | |
100 | /* count of faults inserted */ | |
101 | module_param(fault_count, int, 0664); | |
102 | /* bitmap of devices to insert faults on */ | |
103 | module_param(fault_devs, int, 0644); | |
104 | #endif | |
105 | ||
106 | /* module parameter, defined */ | |
2b8a90b5 | 107 | unsigned int minor_count = DRBD_MINOR_COUNT_DEF; |
b411b363 PR |
108 | int disable_sendpage; |
109 | int allow_oos; | |
b411b363 PR |
110 | int proc_details; /* Detail level in proc drbd*/ |
111 | ||
112 | /* Module parameter for setting the user mode helper program | |
113 | * to run. Default is /sbin/drbdadm */ | |
114 | char usermode_helper[80] = "/sbin/drbdadm"; | |
115 | ||
116 | module_param_string(usermode_helper, usermode_helper, sizeof(usermode_helper), 0644); | |
117 | ||
118 | /* in 2.6.x, our device mapping and config info contains our virtual gendisks | |
119 | * as member "struct gendisk *vdisk;" | |
120 | */ | |
81a5d60e | 121 | struct idr minors; |
2111438b | 122 | struct list_head drbd_tconns; /* list of struct drbd_tconn */ |
543cc10b | 123 | DEFINE_MUTEX(drbd_cfg_mutex); |
b411b363 PR |
124 | |
125 | struct kmem_cache *drbd_request_cache; | |
6c852bec | 126 | struct kmem_cache *drbd_ee_cache; /* peer requests */ |
b411b363 PR |
127 | struct kmem_cache *drbd_bm_ext_cache; /* bitmap extents */ |
128 | struct kmem_cache *drbd_al_ext_cache; /* activity log extents */ | |
129 | mempool_t *drbd_request_mempool; | |
130 | mempool_t *drbd_ee_mempool; | |
35abf594 | 131 | mempool_t *drbd_md_io_page_pool; |
da4a75d2 | 132 | struct bio_set *drbd_md_io_bio_set; |
b411b363 PR |
133 | |
134 | /* I do not use a standard mempool, because: | |
135 | 1) I want to hand out the pre-allocated objects first. | |
136 | 2) I want to be able to interrupt sleeping allocation with a signal. | |
137 | Note: This is a single linked list, the next pointer is the private | |
138 | member of struct page. | |
139 | */ | |
140 | struct page *drbd_pp_pool; | |
141 | spinlock_t drbd_pp_lock; | |
142 | int drbd_pp_vacant; | |
143 | wait_queue_head_t drbd_pp_wait; | |
144 | ||
145 | DEFINE_RATELIMIT_STATE(drbd_ratelimit_state, 5 * HZ, 5); | |
146 | ||
7d4e9d09 | 147 | static const struct block_device_operations drbd_ops = { |
b411b363 PR |
148 | .owner = THIS_MODULE, |
149 | .open = drbd_open, | |
150 | .release = drbd_release, | |
151 | }; | |
152 | ||
da4a75d2 LE |
153 | static void bio_destructor_drbd(struct bio *bio) |
154 | { | |
155 | bio_free(bio, drbd_md_io_bio_set); | |
156 | } | |
157 | ||
158 | struct bio *bio_alloc_drbd(gfp_t gfp_mask) | |
159 | { | |
160 | struct bio *bio; | |
161 | ||
162 | if (!drbd_md_io_bio_set) | |
163 | return bio_alloc(gfp_mask, 1); | |
164 | ||
165 | bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set); | |
166 | if (!bio) | |
167 | return NULL; | |
168 | bio->bi_destructor = bio_destructor_drbd; | |
169 | return bio; | |
170 | } | |
171 | ||
b411b363 PR |
172 | #ifdef __CHECKER__ |
173 | /* When checking with sparse, and this is an inline function, sparse will | |
174 | give tons of false positives. When this is a real functions sparse works. | |
175 | */ | |
176 | int _get_ldev_if_state(struct drbd_conf *mdev, enum drbd_disk_state mins) | |
177 | { | |
178 | int io_allowed; | |
179 | ||
180 | atomic_inc(&mdev->local_cnt); | |
181 | io_allowed = (mdev->state.disk >= mins); | |
182 | if (!io_allowed) { | |
183 | if (atomic_dec_and_test(&mdev->local_cnt)) | |
184 | wake_up(&mdev->misc_wait); | |
185 | } | |
186 | return io_allowed; | |
187 | } | |
188 | ||
189 | #endif | |
190 | ||
191 | /** | |
192 | * DOC: The transfer log | |
193 | * | |
194 | * The transfer log is a single linked list of &struct drbd_tl_epoch objects. | |
87eeee41 | 195 | * mdev->tconn->newest_tle points to the head, mdev->tconn->oldest_tle points to the tail |
b411b363 PR |
196 | * of the list. There is always at least one &struct drbd_tl_epoch object. |
197 | * | |
198 | * Each &struct drbd_tl_epoch has a circular double linked list of requests | |
199 | * attached. | |
200 | */ | |
2f5cdd0b | 201 | static int tl_init(struct drbd_tconn *tconn) |
b411b363 PR |
202 | { |
203 | struct drbd_tl_epoch *b; | |
204 | ||
205 | /* during device minor initialization, we may well use GFP_KERNEL */ | |
206 | b = kmalloc(sizeof(struct drbd_tl_epoch), GFP_KERNEL); | |
207 | if (!b) | |
208 | return 0; | |
209 | INIT_LIST_HEAD(&b->requests); | |
210 | INIT_LIST_HEAD(&b->w.list); | |
211 | b->next = NULL; | |
212 | b->br_number = 4711; | |
7e602c0a | 213 | b->n_writes = 0; |
b411b363 PR |
214 | b->w.cb = NULL; /* if this is != NULL, we need to dec_ap_pending in tl_clear */ |
215 | ||
2f5cdd0b PR |
216 | tconn->oldest_tle = b; |
217 | tconn->newest_tle = b; | |
218 | INIT_LIST_HEAD(&tconn->out_of_sequence_requests); | |
b411b363 | 219 | |
b411b363 PR |
220 | return 1; |
221 | } | |
222 | ||
2f5cdd0b | 223 | static void tl_cleanup(struct drbd_tconn *tconn) |
b411b363 | 224 | { |
2f5cdd0b PR |
225 | if (tconn->oldest_tle != tconn->newest_tle) |
226 | conn_err(tconn, "ASSERT FAILED: oldest_tle == newest_tle\n"); | |
227 | if (!list_empty(&tconn->out_of_sequence_requests)) | |
228 | conn_err(tconn, "ASSERT FAILED: list_empty(out_of_sequence_requests)\n"); | |
229 | kfree(tconn->oldest_tle); | |
230 | tconn->oldest_tle = NULL; | |
231 | kfree(tconn->unused_spare_tle); | |
232 | tconn->unused_spare_tle = NULL; | |
d628769b AG |
233 | } |
234 | ||
b411b363 PR |
235 | /** |
236 | * _tl_add_barrier() - Adds a barrier to the transfer log | |
237 | * @mdev: DRBD device. | |
238 | * @new: Barrier to be added before the current head of the TL. | |
239 | * | |
240 | * The caller must hold the req_lock. | |
241 | */ | |
2f5cdd0b | 242 | void _tl_add_barrier(struct drbd_tconn *tconn, struct drbd_tl_epoch *new) |
b411b363 PR |
243 | { |
244 | struct drbd_tl_epoch *newest_before; | |
245 | ||
246 | INIT_LIST_HEAD(&new->requests); | |
247 | INIT_LIST_HEAD(&new->w.list); | |
248 | new->w.cb = NULL; /* if this is != NULL, we need to dec_ap_pending in tl_clear */ | |
249 | new->next = NULL; | |
7e602c0a | 250 | new->n_writes = 0; |
b411b363 | 251 | |
2f5cdd0b | 252 | newest_before = tconn->newest_tle; |
b411b363 PR |
253 | /* never send a barrier number == 0, because that is special-cased |
254 | * when using TCQ for our write ordering code */ | |
255 | new->br_number = (newest_before->br_number+1) ?: 1; | |
2f5cdd0b PR |
256 | if (tconn->newest_tle != new) { |
257 | tconn->newest_tle->next = new; | |
258 | tconn->newest_tle = new; | |
b411b363 PR |
259 | } |
260 | } | |
261 | ||
262 | /** | |
263 | * tl_release() - Free or recycle the oldest &struct drbd_tl_epoch object of the TL | |
264 | * @mdev: DRBD device. | |
265 | * @barrier_nr: Expected identifier of the DRBD write barrier packet. | |
266 | * @set_size: Expected number of requests before that barrier. | |
267 | * | |
268 | * In case the passed barrier_nr or set_size does not match the oldest | |
269 | * &struct drbd_tl_epoch objects this function will cause a termination | |
270 | * of the connection. | |
271 | */ | |
2f5cdd0b PR |
272 | void tl_release(struct drbd_tconn *tconn, unsigned int barrier_nr, |
273 | unsigned int set_size) | |
b411b363 | 274 | { |
2f5cdd0b | 275 | struct drbd_conf *mdev; |
b411b363 PR |
276 | struct drbd_tl_epoch *b, *nob; /* next old barrier */ |
277 | struct list_head *le, *tle; | |
278 | struct drbd_request *r; | |
279 | ||
2f5cdd0b | 280 | spin_lock_irq(&tconn->req_lock); |
b411b363 | 281 | |
2f5cdd0b | 282 | b = tconn->oldest_tle; |
b411b363 PR |
283 | |
284 | /* first some paranoia code */ | |
285 | if (b == NULL) { | |
2f5cdd0b PR |
286 | conn_err(tconn, "BAD! BarrierAck #%u received, but no epoch in tl!?\n", |
287 | barrier_nr); | |
b411b363 PR |
288 | goto bail; |
289 | } | |
290 | if (b->br_number != barrier_nr) { | |
2f5cdd0b PR |
291 | conn_err(tconn, "BAD! BarrierAck #%u received, expected #%u!\n", |
292 | barrier_nr, b->br_number); | |
b411b363 PR |
293 | goto bail; |
294 | } | |
7e602c0a | 295 | if (b->n_writes != set_size) { |
2f5cdd0b PR |
296 | conn_err(tconn, "BAD! BarrierAck #%u received with n_writes=%u, expected n_writes=%u!\n", |
297 | barrier_nr, set_size, b->n_writes); | |
b411b363 PR |
298 | goto bail; |
299 | } | |
300 | ||
301 | /* Clean up list of requests processed during current epoch */ | |
302 | list_for_each_safe(le, tle, &b->requests) { | |
303 | r = list_entry(le, struct drbd_request, tl_requests); | |
8554df1c | 304 | _req_mod(r, BARRIER_ACKED); |
b411b363 PR |
305 | } |
306 | /* There could be requests on the list waiting for completion | |
307 | of the write to the local disk. To avoid corruptions of | |
308 | slab's data structures we have to remove the lists head. | |
309 | ||
310 | Also there could have been a barrier ack out of sequence, overtaking | |
311 | the write acks - which would be a bug and violating write ordering. | |
312 | To not deadlock in case we lose connection while such requests are | |
313 | still pending, we need some way to find them for the | |
8554df1c | 314 | _req_mode(CONNECTION_LOST_WHILE_PENDING). |
b411b363 PR |
315 | |
316 | These have been list_move'd to the out_of_sequence_requests list in | |
8554df1c | 317 | _req_mod(, BARRIER_ACKED) above. |
b411b363 PR |
318 | */ |
319 | list_del_init(&b->requests); | |
2f5cdd0b | 320 | mdev = b->w.mdev; |
b411b363 PR |
321 | |
322 | nob = b->next; | |
323 | if (test_and_clear_bit(CREATE_BARRIER, &mdev->flags)) { | |
2f5cdd0b | 324 | _tl_add_barrier(tconn, b); |
b411b363 | 325 | if (nob) |
2f5cdd0b | 326 | tconn->oldest_tle = nob; |
b411b363 | 327 | /* if nob == NULL b was the only barrier, and becomes the new |
2f5cdd0b | 328 | barrier. Therefore tconn->oldest_tle points already to b */ |
b411b363 PR |
329 | } else { |
330 | D_ASSERT(nob != NULL); | |
2f5cdd0b | 331 | tconn->oldest_tle = nob; |
b411b363 PR |
332 | kfree(b); |
333 | } | |
334 | ||
2f5cdd0b | 335 | spin_unlock_irq(&tconn->req_lock); |
b411b363 PR |
336 | dec_ap_pending(mdev); |
337 | ||
338 | return; | |
339 | ||
340 | bail: | |
2f5cdd0b PR |
341 | spin_unlock_irq(&tconn->req_lock); |
342 | conn_request_state(tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
b411b363 PR |
343 | } |
344 | ||
617049aa | 345 | |
b411b363 | 346 | /** |
11b58e73 | 347 | * _tl_restart() - Walks the transfer log, and applies an action to all requests |
b411b363 | 348 | * @mdev: DRBD device. |
11b58e73 | 349 | * @what: The action/event to perform with all request objects |
b411b363 | 350 | * |
8554df1c AG |
351 | * @what might be one of CONNECTION_LOST_WHILE_PENDING, RESEND, FAIL_FROZEN_DISK_IO, |
352 | * RESTART_FROZEN_DISK_IO. | |
b411b363 | 353 | */ |
2f5cdd0b | 354 | void _tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what) |
b411b363 | 355 | { |
11b58e73 | 356 | struct drbd_tl_epoch *b, *tmp, **pn; |
b9b98716 | 357 | struct list_head *le, *tle, carry_reads; |
11b58e73 PR |
358 | struct drbd_request *req; |
359 | int rv, n_writes, n_reads; | |
b411b363 | 360 | |
2f5cdd0b PR |
361 | b = tconn->oldest_tle; |
362 | pn = &tconn->oldest_tle; | |
b411b363 | 363 | while (b) { |
11b58e73 PR |
364 | n_writes = 0; |
365 | n_reads = 0; | |
b9b98716 | 366 | INIT_LIST_HEAD(&carry_reads); |
b411b363 | 367 | list_for_each_safe(le, tle, &b->requests) { |
11b58e73 PR |
368 | req = list_entry(le, struct drbd_request, tl_requests); |
369 | rv = _req_mod(req, what); | |
370 | ||
371 | n_writes += (rv & MR_WRITE) >> MR_WRITE_SHIFT; | |
372 | n_reads += (rv & MR_READ) >> MR_READ_SHIFT; | |
b411b363 PR |
373 | } |
374 | tmp = b->next; | |
375 | ||
b9b98716 | 376 | if (n_writes) { |
8554df1c | 377 | if (what == RESEND) { |
11b58e73 PR |
378 | b->n_writes = n_writes; |
379 | if (b->w.cb == NULL) { | |
380 | b->w.cb = w_send_barrier; | |
2f5cdd0b PR |
381 | inc_ap_pending(b->w.mdev); |
382 | set_bit(CREATE_BARRIER, &b->w.mdev->flags); | |
11b58e73 PR |
383 | } |
384 | ||
2f5cdd0b | 385 | drbd_queue_work(&tconn->data.work, &b->w); |
11b58e73 PR |
386 | } |
387 | pn = &b->next; | |
388 | } else { | |
b9b98716 PR |
389 | if (n_reads) |
390 | list_add(&carry_reads, &b->requests); | |
11b58e73 PR |
391 | /* there could still be requests on that ring list, |
392 | * in case local io is still pending */ | |
393 | list_del(&b->requests); | |
394 | ||
395 | /* dec_ap_pending corresponding to queue_barrier. | |
396 | * the newest barrier may not have been queued yet, | |
397 | * in which case w.cb is still NULL. */ | |
398 | if (b->w.cb != NULL) | |
2f5cdd0b | 399 | dec_ap_pending(b->w.mdev); |
11b58e73 | 400 | |
2f5cdd0b | 401 | if (b == tconn->newest_tle) { |
11b58e73 | 402 | /* recycle, but reinit! */ |
2f5cdd0b PR |
403 | if (tmp != NULL) |
404 | conn_err(tconn, "ASSERT FAILED tmp == NULL"); | |
11b58e73 | 405 | INIT_LIST_HEAD(&b->requests); |
b9b98716 | 406 | list_splice(&carry_reads, &b->requests); |
11b58e73 PR |
407 | INIT_LIST_HEAD(&b->w.list); |
408 | b->w.cb = NULL; | |
409 | b->br_number = net_random(); | |
410 | b->n_writes = 0; | |
411 | ||
412 | *pn = b; | |
413 | break; | |
414 | } | |
415 | *pn = tmp; | |
416 | kfree(b); | |
b411b363 | 417 | } |
b411b363 | 418 | b = tmp; |
b9b98716 | 419 | list_splice(&carry_reads, &b->requests); |
b411b363 | 420 | } |
11b58e73 PR |
421 | } |
422 | ||
b411b363 PR |
423 | |
424 | /** | |
425 | * tl_clear() - Clears all requests and &struct drbd_tl_epoch objects out of the TL | |
426 | * @mdev: DRBD device. | |
427 | * | |
428 | * This is called after the connection to the peer was lost. The storage covered | |
429 | * by the requests on the transfer gets marked as our of sync. Called from the | |
430 | * receiver thread and the worker thread. | |
431 | */ | |
2f5cdd0b | 432 | void tl_clear(struct drbd_tconn *tconn) |
b411b363 | 433 | { |
2f5cdd0b | 434 | struct drbd_conf *mdev; |
b411b363 PR |
435 | struct list_head *le, *tle; |
436 | struct drbd_request *r; | |
e90285e0 | 437 | int vnr; |
b411b363 | 438 | |
2f5cdd0b | 439 | spin_lock_irq(&tconn->req_lock); |
b411b363 | 440 | |
2f5cdd0b | 441 | _tl_restart(tconn, CONNECTION_LOST_WHILE_PENDING); |
b411b363 PR |
442 | |
443 | /* we expect this list to be empty. */ | |
2f5cdd0b PR |
444 | if (!list_empty(&tconn->out_of_sequence_requests)) |
445 | conn_err(tconn, "ASSERT FAILED list_empty(&out_of_sequence_requests)\n"); | |
b411b363 PR |
446 | |
447 | /* but just in case, clean it up anyways! */ | |
2f5cdd0b | 448 | list_for_each_safe(le, tle, &tconn->out_of_sequence_requests) { |
b411b363 PR |
449 | r = list_entry(le, struct drbd_request, tl_requests); |
450 | /* It would be nice to complete outside of spinlock. | |
451 | * But this is easier for now. */ | |
8554df1c | 452 | _req_mod(r, CONNECTION_LOST_WHILE_PENDING); |
b411b363 PR |
453 | } |
454 | ||
455 | /* ensure bit indicating barrier is required is clear */ | |
e90285e0 | 456 | idr_for_each_entry(&tconn->volumes, mdev, vnr) |
2f5cdd0b | 457 | clear_bit(CREATE_BARRIER, &mdev->flags); |
b411b363 | 458 | |
2f5cdd0b | 459 | spin_unlock_irq(&tconn->req_lock); |
b411b363 PR |
460 | } |
461 | ||
2f5cdd0b | 462 | void tl_restart(struct drbd_tconn *tconn, enum drbd_req_event what) |
11b58e73 | 463 | { |
2f5cdd0b PR |
464 | spin_lock_irq(&tconn->req_lock); |
465 | _tl_restart(tconn, what); | |
466 | spin_unlock_irq(&tconn->req_lock); | |
b411b363 PR |
467 | } |
468 | ||
b411b363 PR |
469 | static int drbd_thread_setup(void *arg) |
470 | { | |
471 | struct drbd_thread *thi = (struct drbd_thread *) arg; | |
392c8801 | 472 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 PR |
473 | unsigned long flags; |
474 | int retval; | |
475 | ||
f1b3a6ec | 476 | snprintf(current->comm, sizeof(current->comm), "drbd_%c_%s", |
392c8801 | 477 | thi->name[0], thi->tconn->name); |
f1b3a6ec | 478 | |
b411b363 PR |
479 | restart: |
480 | retval = thi->function(thi); | |
481 | ||
482 | spin_lock_irqsave(&thi->t_lock, flags); | |
483 | ||
e77a0a5c | 484 | /* if the receiver has been "EXITING", the last thing it did |
b411b363 PR |
485 | * was set the conn state to "StandAlone", |
486 | * if now a re-connect request comes in, conn state goes C_UNCONNECTED, | |
487 | * and receiver thread will be "started". | |
e77a0a5c | 488 | * drbd_thread_start needs to set "RESTARTING" in that case. |
b411b363 | 489 | * t_state check and assignment needs to be within the same spinlock, |
e77a0a5c AG |
490 | * so either thread_start sees EXITING, and can remap to RESTARTING, |
491 | * or thread_start see NONE, and can proceed as normal. | |
b411b363 PR |
492 | */ |
493 | ||
e77a0a5c | 494 | if (thi->t_state == RESTARTING) { |
392c8801 | 495 | conn_info(tconn, "Restarting %s thread\n", thi->name); |
e77a0a5c | 496 | thi->t_state = RUNNING; |
b411b363 PR |
497 | spin_unlock_irqrestore(&thi->t_lock, flags); |
498 | goto restart; | |
499 | } | |
500 | ||
501 | thi->task = NULL; | |
e77a0a5c | 502 | thi->t_state = NONE; |
b411b363 PR |
503 | smp_mb(); |
504 | complete(&thi->stop); | |
505 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
506 | ||
392c8801 | 507 | conn_info(tconn, "Terminating %s\n", current->comm); |
b411b363 PR |
508 | |
509 | /* Release mod reference taken when thread was started */ | |
510 | module_put(THIS_MODULE); | |
511 | return retval; | |
512 | } | |
513 | ||
392c8801 | 514 | static void drbd_thread_init(struct drbd_tconn *tconn, struct drbd_thread *thi, |
bed879ae | 515 | int (*func) (struct drbd_thread *), char *name) |
b411b363 PR |
516 | { |
517 | spin_lock_init(&thi->t_lock); | |
518 | thi->task = NULL; | |
e77a0a5c | 519 | thi->t_state = NONE; |
b411b363 | 520 | thi->function = func; |
392c8801 | 521 | thi->tconn = tconn; |
bed879ae | 522 | strncpy(thi->name, name, ARRAY_SIZE(thi->name)); |
b411b363 PR |
523 | } |
524 | ||
525 | int drbd_thread_start(struct drbd_thread *thi) | |
526 | { | |
392c8801 | 527 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 PR |
528 | struct task_struct *nt; |
529 | unsigned long flags; | |
530 | ||
b411b363 PR |
531 | /* is used from state engine doing drbd_thread_stop_nowait, |
532 | * while holding the req lock irqsave */ | |
533 | spin_lock_irqsave(&thi->t_lock, flags); | |
534 | ||
535 | switch (thi->t_state) { | |
e77a0a5c | 536 | case NONE: |
392c8801 | 537 | conn_info(tconn, "Starting %s thread (from %s [%d])\n", |
bed879ae | 538 | thi->name, current->comm, current->pid); |
b411b363 PR |
539 | |
540 | /* Get ref on module for thread - this is released when thread exits */ | |
541 | if (!try_module_get(THIS_MODULE)) { | |
392c8801 | 542 | conn_err(tconn, "Failed to get module reference in drbd_thread_start\n"); |
b411b363 | 543 | spin_unlock_irqrestore(&thi->t_lock, flags); |
81e84650 | 544 | return false; |
b411b363 PR |
545 | } |
546 | ||
547 | init_completion(&thi->stop); | |
b411b363 | 548 | thi->reset_cpu_mask = 1; |
e77a0a5c | 549 | thi->t_state = RUNNING; |
b411b363 PR |
550 | spin_unlock_irqrestore(&thi->t_lock, flags); |
551 | flush_signals(current); /* otherw. may get -ERESTARTNOINTR */ | |
552 | ||
553 | nt = kthread_create(drbd_thread_setup, (void *) thi, | |
392c8801 | 554 | "drbd_%c_%s", thi->name[0], thi->tconn->name); |
b411b363 PR |
555 | |
556 | if (IS_ERR(nt)) { | |
392c8801 | 557 | conn_err(tconn, "Couldn't start thread\n"); |
b411b363 PR |
558 | |
559 | module_put(THIS_MODULE); | |
81e84650 | 560 | return false; |
b411b363 PR |
561 | } |
562 | spin_lock_irqsave(&thi->t_lock, flags); | |
563 | thi->task = nt; | |
e77a0a5c | 564 | thi->t_state = RUNNING; |
b411b363 PR |
565 | spin_unlock_irqrestore(&thi->t_lock, flags); |
566 | wake_up_process(nt); | |
567 | break; | |
e77a0a5c AG |
568 | case EXITING: |
569 | thi->t_state = RESTARTING; | |
392c8801 | 570 | conn_info(tconn, "Restarting %s thread (from %s [%d])\n", |
bed879ae | 571 | thi->name, current->comm, current->pid); |
b411b363 | 572 | /* fall through */ |
e77a0a5c AG |
573 | case RUNNING: |
574 | case RESTARTING: | |
b411b363 PR |
575 | default: |
576 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
577 | break; | |
578 | } | |
579 | ||
81e84650 | 580 | return true; |
b411b363 PR |
581 | } |
582 | ||
583 | ||
584 | void _drbd_thread_stop(struct drbd_thread *thi, int restart, int wait) | |
585 | { | |
586 | unsigned long flags; | |
587 | ||
e77a0a5c | 588 | enum drbd_thread_state ns = restart ? RESTARTING : EXITING; |
b411b363 PR |
589 | |
590 | /* may be called from state engine, holding the req lock irqsave */ | |
591 | spin_lock_irqsave(&thi->t_lock, flags); | |
592 | ||
e77a0a5c | 593 | if (thi->t_state == NONE) { |
b411b363 PR |
594 | spin_unlock_irqrestore(&thi->t_lock, flags); |
595 | if (restart) | |
596 | drbd_thread_start(thi); | |
597 | return; | |
598 | } | |
599 | ||
600 | if (thi->t_state != ns) { | |
601 | if (thi->task == NULL) { | |
602 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
603 | return; | |
604 | } | |
605 | ||
606 | thi->t_state = ns; | |
607 | smp_mb(); | |
608 | init_completion(&thi->stop); | |
609 | if (thi->task != current) | |
610 | force_sig(DRBD_SIGKILL, thi->task); | |
b411b363 PR |
611 | } |
612 | ||
613 | spin_unlock_irqrestore(&thi->t_lock, flags); | |
614 | ||
615 | if (wait) | |
616 | wait_for_completion(&thi->stop); | |
617 | } | |
618 | ||
392c8801 | 619 | static struct drbd_thread *drbd_task_to_thread(struct drbd_tconn *tconn, struct task_struct *task) |
bed879ae | 620 | { |
bed879ae PR |
621 | struct drbd_thread *thi = |
622 | task == tconn->receiver.task ? &tconn->receiver : | |
623 | task == tconn->asender.task ? &tconn->asender : | |
624 | task == tconn->worker.task ? &tconn->worker : NULL; | |
625 | ||
626 | return thi; | |
627 | } | |
628 | ||
392c8801 | 629 | char *drbd_task_to_thread_name(struct drbd_tconn *tconn, struct task_struct *task) |
bed879ae | 630 | { |
392c8801 | 631 | struct drbd_thread *thi = drbd_task_to_thread(tconn, task); |
bed879ae PR |
632 | return thi ? thi->name : task->comm; |
633 | } | |
634 | ||
80883197 | 635 | int conn_lowest_minor(struct drbd_tconn *tconn) |
80822284 | 636 | { |
e90285e0 PR |
637 | int vnr = 0; |
638 | struct drbd_conf *mdev; | |
774b3055 | 639 | |
e90285e0 PR |
640 | mdev = idr_get_next(&tconn->volumes, &vnr); |
641 | if (!mdev) | |
774b3055 | 642 | return -1; |
e90285e0 | 643 | return mdev_to_minor(mdev); |
80822284 | 644 | } |
774b3055 PR |
645 | |
646 | #ifdef CONFIG_SMP | |
b411b363 PR |
647 | /** |
648 | * drbd_calc_cpu_mask() - Generate CPU masks, spread over all CPUs | |
649 | * @mdev: DRBD device. | |
650 | * | |
651 | * Forces all threads of a device onto the same CPU. This is beneficial for | |
652 | * DRBD's performance. May be overwritten by user's configuration. | |
653 | */ | |
80822284 | 654 | void drbd_calc_cpu_mask(struct drbd_tconn *tconn) |
b411b363 PR |
655 | { |
656 | int ord, cpu; | |
657 | ||
658 | /* user override. */ | |
80822284 | 659 | if (cpumask_weight(tconn->cpu_mask)) |
b411b363 PR |
660 | return; |
661 | ||
80822284 | 662 | ord = conn_lowest_minor(tconn) % cpumask_weight(cpu_online_mask); |
b411b363 PR |
663 | for_each_online_cpu(cpu) { |
664 | if (ord-- == 0) { | |
80822284 | 665 | cpumask_set_cpu(cpu, tconn->cpu_mask); |
b411b363 PR |
666 | return; |
667 | } | |
668 | } | |
669 | /* should not be reached */ | |
80822284 | 670 | cpumask_setall(tconn->cpu_mask); |
b411b363 PR |
671 | } |
672 | ||
673 | /** | |
674 | * drbd_thread_current_set_cpu() - modifies the cpu mask of the _current_ thread | |
675 | * @mdev: DRBD device. | |
bc31fe33 | 676 | * @thi: drbd_thread object |
b411b363 PR |
677 | * |
678 | * call in the "main loop" of _all_ threads, no need for any mutex, current won't die | |
679 | * prematurely. | |
680 | */ | |
80822284 | 681 | void drbd_thread_current_set_cpu(struct drbd_thread *thi) |
b411b363 PR |
682 | { |
683 | struct task_struct *p = current; | |
bed879ae | 684 | |
b411b363 PR |
685 | if (!thi->reset_cpu_mask) |
686 | return; | |
687 | thi->reset_cpu_mask = 0; | |
392c8801 | 688 | set_cpus_allowed_ptr(p, thi->tconn->cpu_mask); |
b411b363 PR |
689 | } |
690 | #endif | |
691 | ||
d38e787e | 692 | static void prepare_header80(struct p_header80 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
693 | { |
694 | h->magic = cpu_to_be32(DRBD_MAGIC); | |
695 | h->command = cpu_to_be16(cmd); | |
696 | h->length = cpu_to_be16(size); | |
697 | } | |
698 | ||
d38e787e | 699 | static void prepare_header95(struct p_header95 *h, enum drbd_packet cmd, int size) |
fd340c12 PR |
700 | { |
701 | h->magic = cpu_to_be16(DRBD_MAGIC_BIG); | |
702 | h->command = cpu_to_be16(cmd); | |
703 | h->length = cpu_to_be32(size); | |
704 | } | |
705 | ||
d38e787e PR |
706 | static void _prepare_header(struct drbd_tconn *tconn, int vnr, struct p_header *h, |
707 | enum drbd_packet cmd, int size) | |
708 | { | |
709 | if (tconn->agreed_pro_version >= 100 || size > DRBD_MAX_SIZE_H80_PACKET) | |
710 | prepare_header95(&h->h95, cmd, size); | |
711 | else | |
712 | prepare_header80(&h->h80, cmd, size); | |
713 | } | |
714 | ||
fd340c12 | 715 | static void prepare_header(struct drbd_conf *mdev, struct p_header *h, |
d8763023 | 716 | enum drbd_packet cmd, int size) |
fd340c12 | 717 | { |
d38e787e | 718 | _prepare_header(mdev->tconn, mdev->vnr, h, cmd, size); |
fd340c12 PR |
719 | } |
720 | ||
b411b363 | 721 | /* the appropriate socket mutex must be held already */ |
d38e787e | 722 | int _conn_send_cmd(struct drbd_tconn *tconn, int vnr, struct socket *sock, |
d8763023 AG |
723 | enum drbd_packet cmd, struct p_header *h, size_t size, |
724 | unsigned msg_flags) | |
b411b363 PR |
725 | { |
726 | int sent, ok; | |
727 | ||
d38e787e | 728 | _prepare_header(tconn, vnr, h, cmd, size - sizeof(struct p_header)); |
b411b363 | 729 | |
d38e787e | 730 | sent = drbd_send(tconn, sock, h, size, msg_flags); |
b411b363 PR |
731 | |
732 | ok = (sent == size); | |
0ddc5549 | 733 | if (!ok && !signal_pending(current)) |
d38e787e PR |
734 | conn_warn(tconn, "short sent %s size=%d sent=%d\n", |
735 | cmdname(cmd), (int)size, sent); | |
b411b363 PR |
736 | return ok; |
737 | } | |
738 | ||
739 | /* don't pass the socket. we may only look at it | |
740 | * when we hold the appropriate socket mutex. | |
741 | */ | |
2a67d8b9 | 742 | int conn_send_cmd(struct drbd_tconn *tconn, int vnr, int use_data_socket, |
d8763023 | 743 | enum drbd_packet cmd, struct p_header *h, size_t size) |
b411b363 PR |
744 | { |
745 | int ok = 0; | |
746 | struct socket *sock; | |
747 | ||
748 | if (use_data_socket) { | |
2a67d8b9 PR |
749 | mutex_lock(&tconn->data.mutex); |
750 | sock = tconn->data.socket; | |
b411b363 | 751 | } else { |
2a67d8b9 PR |
752 | mutex_lock(&tconn->meta.mutex); |
753 | sock = tconn->meta.socket; | |
b411b363 PR |
754 | } |
755 | ||
756 | /* drbd_disconnect() could have called drbd_free_sock() | |
757 | * while we were waiting in down()... */ | |
758 | if (likely(sock != NULL)) | |
2a67d8b9 | 759 | ok = _conn_send_cmd(tconn, vnr, sock, cmd, h, size, 0); |
b411b363 PR |
760 | |
761 | if (use_data_socket) | |
2a67d8b9 | 762 | mutex_unlock(&tconn->data.mutex); |
b411b363 | 763 | else |
2a67d8b9 | 764 | mutex_unlock(&tconn->meta.mutex); |
b411b363 PR |
765 | return ok; |
766 | } | |
767 | ||
61120870 | 768 | int conn_send_cmd2(struct drbd_tconn *tconn, enum drbd_packet cmd, char *data, |
b411b363 PR |
769 | size_t size) |
770 | { | |
61120870 | 771 | struct p_header80 h; |
b411b363 PR |
772 | int ok; |
773 | ||
61120870 | 774 | prepare_header80(&h, cmd, size); |
b411b363 | 775 | |
11b0be28 | 776 | if (drbd_get_data_sock(tconn)) |
b411b363 PR |
777 | return 0; |
778 | ||
b411b363 | 779 | ok = (sizeof(h) == |
61120870 | 780 | drbd_send(tconn, tconn->data.socket, &h, sizeof(h), 0)); |
b411b363 | 781 | ok = ok && (size == |
61120870 | 782 | drbd_send(tconn, tconn->data.socket, data, size, 0)); |
b411b363 | 783 | |
61120870 | 784 | drbd_put_data_sock(tconn); |
b411b363 PR |
785 | |
786 | return ok; | |
787 | } | |
788 | ||
f399002e | 789 | int drbd_send_sync_param(struct drbd_conf *mdev) |
b411b363 | 790 | { |
8e26f9cc | 791 | struct p_rs_param_95 *p; |
b411b363 PR |
792 | struct socket *sock; |
793 | int size, rv; | |
31890f4a | 794 | const int apv = mdev->tconn->agreed_pro_version; |
b411b363 PR |
795 | |
796 | size = apv <= 87 ? sizeof(struct p_rs_param) | |
797 | : apv == 88 ? sizeof(struct p_rs_param) | |
f399002e | 798 | + strlen(mdev->tconn->net_conf->verify_alg) + 1 |
8e26f9cc PR |
799 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
800 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 PR |
801 | |
802 | /* used from admin command context and receiver/worker context. | |
803 | * to avoid kmalloc, grab the socket right here, | |
804 | * then use the pre-allocated sbuf there */ | |
e42325a5 PR |
805 | mutex_lock(&mdev->tconn->data.mutex); |
806 | sock = mdev->tconn->data.socket; | |
b411b363 PR |
807 | |
808 | if (likely(sock != NULL)) { | |
d8763023 AG |
809 | enum drbd_packet cmd = |
810 | apv >= 89 ? P_SYNC_PARAM89 : P_SYNC_PARAM; | |
b411b363 | 811 | |
e42325a5 | 812 | p = &mdev->tconn->data.sbuf.rs_param_95; |
b411b363 PR |
813 | |
814 | /* initialize verify_alg and csums_alg */ | |
815 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
816 | ||
f399002e LE |
817 | if (get_ldev(mdev)) { |
818 | p->rate = cpu_to_be32(mdev->ldev->dc.resync_rate); | |
819 | p->c_plan_ahead = cpu_to_be32(mdev->ldev->dc.c_plan_ahead); | |
820 | p->c_delay_target = cpu_to_be32(mdev->ldev->dc.c_delay_target); | |
821 | p->c_fill_target = cpu_to_be32(mdev->ldev->dc.c_fill_target); | |
822 | p->c_max_rate = cpu_to_be32(mdev->ldev->dc.c_max_rate); | |
823 | put_ldev(mdev); | |
824 | } else { | |
825 | p->rate = cpu_to_be32(DRBD_RATE_DEF); | |
826 | p->c_plan_ahead = cpu_to_be32(DRBD_C_PLAN_AHEAD_DEF); | |
827 | p->c_delay_target = cpu_to_be32(DRBD_C_DELAY_TARGET_DEF); | |
828 | p->c_fill_target = cpu_to_be32(DRBD_C_FILL_TARGET_DEF); | |
829 | p->c_max_rate = cpu_to_be32(DRBD_C_MAX_RATE_DEF); | |
830 | } | |
b411b363 PR |
831 | |
832 | if (apv >= 88) | |
f399002e | 833 | strcpy(p->verify_alg, mdev->tconn->net_conf->verify_alg); |
b411b363 | 834 | if (apv >= 89) |
f399002e | 835 | strcpy(p->csums_alg, mdev->tconn->net_conf->csums_alg); |
b411b363 PR |
836 | |
837 | rv = _drbd_send_cmd(mdev, sock, cmd, &p->head, size, 0); | |
838 | } else | |
839 | rv = 0; /* not ok */ | |
840 | ||
e42325a5 | 841 | mutex_unlock(&mdev->tconn->data.mutex); |
b411b363 PR |
842 | |
843 | return rv; | |
844 | } | |
845 | ||
dc8228d1 | 846 | int drbd_send_protocol(struct drbd_tconn *tconn) |
b411b363 PR |
847 | { |
848 | struct p_protocol *p; | |
cf14c2e9 | 849 | int size, cf, rv; |
b411b363 PR |
850 | |
851 | size = sizeof(struct p_protocol); | |
852 | ||
dc8228d1 PR |
853 | if (tconn->agreed_pro_version >= 87) |
854 | size += strlen(tconn->net_conf->integrity_alg) + 1; | |
b411b363 PR |
855 | |
856 | /* we must not recurse into our own queue, | |
857 | * as that is blocked during handshake */ | |
858 | p = kmalloc(size, GFP_NOIO); | |
859 | if (p == NULL) | |
860 | return 0; | |
861 | ||
dc8228d1 PR |
862 | p->protocol = cpu_to_be32(tconn->net_conf->wire_protocol); |
863 | p->after_sb_0p = cpu_to_be32(tconn->net_conf->after_sb_0p); | |
864 | p->after_sb_1p = cpu_to_be32(tconn->net_conf->after_sb_1p); | |
865 | p->after_sb_2p = cpu_to_be32(tconn->net_conf->after_sb_2p); | |
866 | p->two_primaries = cpu_to_be32(tconn->net_conf->two_primaries); | |
b411b363 | 867 | |
cf14c2e9 | 868 | cf = 0; |
dc8228d1 | 869 | if (tconn->net_conf->want_lose) |
cf14c2e9 | 870 | cf |= CF_WANT_LOSE; |
dc8228d1 PR |
871 | if (tconn->net_conf->dry_run) { |
872 | if (tconn->agreed_pro_version >= 92) | |
cf14c2e9 PR |
873 | cf |= CF_DRY_RUN; |
874 | else { | |
dc8228d1 | 875 | conn_err(tconn, "--dry-run is not supported by peer"); |
7ac314c8 | 876 | kfree(p); |
148efa16 | 877 | return -1; |
cf14c2e9 PR |
878 | } |
879 | } | |
880 | p->conn_flags = cpu_to_be32(cf); | |
881 | ||
dc8228d1 PR |
882 | if (tconn->agreed_pro_version >= 87) |
883 | strcpy(p->integrity_alg, tconn->net_conf->integrity_alg); | |
b411b363 | 884 | |
dc8228d1 | 885 | rv = conn_send_cmd2(tconn, P_PROTOCOL, p->head.payload, size - sizeof(struct p_header)); |
b411b363 PR |
886 | kfree(p); |
887 | return rv; | |
888 | } | |
889 | ||
890 | int _drbd_send_uuids(struct drbd_conf *mdev, u64 uuid_flags) | |
891 | { | |
892 | struct p_uuids p; | |
893 | int i; | |
894 | ||
895 | if (!get_ldev_if_state(mdev, D_NEGOTIATING)) | |
896 | return 1; | |
897 | ||
898 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
899 | p.uuid[i] = mdev->ldev ? cpu_to_be64(mdev->ldev->md.uuid[i]) : 0; | |
900 | ||
901 | mdev->comm_bm_set = drbd_bm_total_weight(mdev); | |
902 | p.uuid[UI_SIZE] = cpu_to_be64(mdev->comm_bm_set); | |
89e58e75 | 903 | uuid_flags |= mdev->tconn->net_conf->want_lose ? 1 : 0; |
b411b363 PR |
904 | uuid_flags |= test_bit(CRASHED_PRIMARY, &mdev->flags) ? 2 : 0; |
905 | uuid_flags |= mdev->new_state_tmp.disk == D_INCONSISTENT ? 4 : 0; | |
906 | p.uuid[UI_FLAGS] = cpu_to_be64(uuid_flags); | |
907 | ||
908 | put_ldev(mdev); | |
909 | ||
c012949a | 910 | return drbd_send_cmd(mdev, USE_DATA_SOCKET, P_UUIDS, &p.head, sizeof(p)); |
b411b363 PR |
911 | } |
912 | ||
913 | int drbd_send_uuids(struct drbd_conf *mdev) | |
914 | { | |
915 | return _drbd_send_uuids(mdev, 0); | |
916 | } | |
917 | ||
918 | int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev) | |
919 | { | |
920 | return _drbd_send_uuids(mdev, 8); | |
921 | } | |
922 | ||
62b0da3a LE |
923 | void drbd_print_uuids(struct drbd_conf *mdev, const char *text) |
924 | { | |
925 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
926 | u64 *uuid = mdev->ldev->md.uuid; | |
927 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX\n", | |
928 | text, | |
929 | (unsigned long long)uuid[UI_CURRENT], | |
930 | (unsigned long long)uuid[UI_BITMAP], | |
931 | (unsigned long long)uuid[UI_HISTORY_START], | |
932 | (unsigned long long)uuid[UI_HISTORY_END]); | |
933 | put_ldev(mdev); | |
934 | } else { | |
935 | dev_info(DEV, "%s effective data uuid: %016llX\n", | |
936 | text, | |
937 | (unsigned long long)mdev->ed_uuid); | |
938 | } | |
939 | } | |
940 | ||
5a22db89 | 941 | int drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev) |
b411b363 PR |
942 | { |
943 | struct p_rs_uuid p; | |
5a22db89 LE |
944 | u64 uuid; |
945 | ||
946 | D_ASSERT(mdev->state.disk == D_UP_TO_DATE); | |
b411b363 | 947 | |
4a23f264 | 948 | uuid = mdev->ldev->md.uuid[UI_BITMAP] + UUID_NEW_BM_OFFSET; |
5a22db89 | 949 | drbd_uuid_set(mdev, UI_BITMAP, uuid); |
62b0da3a | 950 | drbd_print_uuids(mdev, "updated sync UUID"); |
5a22db89 LE |
951 | drbd_md_sync(mdev); |
952 | p.uuid = cpu_to_be64(uuid); | |
b411b363 | 953 | |
c012949a | 954 | return drbd_send_cmd(mdev, USE_DATA_SOCKET, P_SYNC_UUID, &p.head, sizeof(p)); |
b411b363 PR |
955 | } |
956 | ||
e89b591c | 957 | int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags) |
b411b363 PR |
958 | { |
959 | struct p_sizes p; | |
960 | sector_t d_size, u_size; | |
99432fcc | 961 | int q_order_type, max_bio_size; |
b411b363 PR |
962 | int ok; |
963 | ||
964 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
965 | D_ASSERT(mdev->ldev->backing_bdev); | |
966 | d_size = drbd_get_max_capacity(mdev->ldev); | |
967 | u_size = mdev->ldev->dc.disk_size; | |
968 | q_order_type = drbd_queue_order_type(mdev); | |
99432fcc PR |
969 | max_bio_size = queue_max_hw_sectors(mdev->ldev->backing_bdev->bd_disk->queue) << 9; |
970 | max_bio_size = min_t(int, max_bio_size, DRBD_MAX_BIO_SIZE); | |
b411b363 PR |
971 | put_ldev(mdev); |
972 | } else { | |
973 | d_size = 0; | |
974 | u_size = 0; | |
975 | q_order_type = QUEUE_ORDERED_NONE; | |
99432fcc | 976 | max_bio_size = DRBD_MAX_BIO_SIZE; /* ... multiple BIOs per peer_request */ |
b411b363 PR |
977 | } |
978 | ||
979 | p.d_size = cpu_to_be64(d_size); | |
980 | p.u_size = cpu_to_be64(u_size); | |
981 | p.c_size = cpu_to_be64(trigger_reply ? 0 : drbd_get_capacity(mdev->this_bdev)); | |
99432fcc | 982 | p.max_bio_size = cpu_to_be32(max_bio_size); |
e89b591c PR |
983 | p.queue_order_type = cpu_to_be16(q_order_type); |
984 | p.dds_flags = cpu_to_be16(flags); | |
b411b363 | 985 | |
c012949a | 986 | ok = drbd_send_cmd(mdev, USE_DATA_SOCKET, P_SIZES, &p.head, sizeof(p)); |
b411b363 PR |
987 | return ok; |
988 | } | |
989 | ||
990 | /** | |
991 | * drbd_send_state() - Sends the drbd state to the peer | |
992 | * @mdev: DRBD device. | |
993 | */ | |
994 | int drbd_send_state(struct drbd_conf *mdev) | |
995 | { | |
996 | struct socket *sock; | |
997 | struct p_state p; | |
998 | int ok = 0; | |
999 | ||
e42325a5 | 1000 | mutex_lock(&mdev->tconn->data.mutex); |
b411b363 PR |
1001 | |
1002 | p.state = cpu_to_be32(mdev->state.i); /* Within the send mutex */ | |
e42325a5 | 1003 | sock = mdev->tconn->data.socket; |
b411b363 PR |
1004 | |
1005 | if (likely(sock != NULL)) { | |
c012949a | 1006 | ok = _drbd_send_cmd(mdev, sock, P_STATE, &p.head, sizeof(p), 0); |
b411b363 PR |
1007 | } |
1008 | ||
e42325a5 | 1009 | mutex_unlock(&mdev->tconn->data.mutex); |
b411b363 | 1010 | |
b411b363 PR |
1011 | return ok; |
1012 | } | |
1013 | ||
cf29c9d8 PR |
1014 | int _conn_send_state_req(struct drbd_tconn *tconn, int vnr, enum drbd_packet cmd, |
1015 | union drbd_state mask, union drbd_state val) | |
b411b363 PR |
1016 | { |
1017 | struct p_req_state p; | |
1018 | ||
1019 | p.mask = cpu_to_be32(mask.i); | |
1020 | p.val = cpu_to_be32(val.i); | |
1021 | ||
cf29c9d8 | 1022 | return conn_send_cmd(tconn, vnr, USE_DATA_SOCKET, cmd, &p.head, sizeof(p)); |
b411b363 PR |
1023 | } |
1024 | ||
bf885f8a | 1025 | int drbd_send_sr_reply(struct drbd_conf *mdev, enum drbd_state_rv retcode) |
b411b363 PR |
1026 | { |
1027 | struct p_req_state_reply p; | |
1028 | ||
1029 | p.retcode = cpu_to_be32(retcode); | |
1030 | ||
c012949a | 1031 | return drbd_send_cmd(mdev, USE_META_SOCKET, P_STATE_CHG_REPLY, &p.head, sizeof(p)); |
b411b363 PR |
1032 | } |
1033 | ||
047cd4a6 PR |
1034 | int conn_send_sr_reply(struct drbd_tconn *tconn, enum drbd_state_rv retcode) |
1035 | { | |
1036 | struct p_req_state_reply p; | |
1037 | enum drbd_packet cmd = tconn->agreed_pro_version < 100 ? P_STATE_CHG_REPLY : P_CONN_ST_CHG_REPLY; | |
1038 | ||
1039 | p.retcode = cpu_to_be32(retcode); | |
1040 | ||
1041 | return conn_send_cmd(tconn, 0, USE_META_SOCKET, cmd, &p.head, sizeof(p)); | |
1042 | } | |
1043 | ||
b411b363 PR |
1044 | int fill_bitmap_rle_bits(struct drbd_conf *mdev, |
1045 | struct p_compressed_bm *p, | |
1046 | struct bm_xfer_ctx *c) | |
1047 | { | |
1048 | struct bitstream bs; | |
1049 | unsigned long plain_bits; | |
1050 | unsigned long tmp; | |
1051 | unsigned long rl; | |
1052 | unsigned len; | |
1053 | unsigned toggle; | |
1054 | int bits; | |
1055 | ||
1056 | /* may we use this feature? */ | |
f399002e | 1057 | if ((mdev->tconn->net_conf->use_rle == 0) || |
31890f4a | 1058 | (mdev->tconn->agreed_pro_version < 90)) |
b411b363 PR |
1059 | return 0; |
1060 | ||
1061 | if (c->bit_offset >= c->bm_bits) | |
1062 | return 0; /* nothing to do. */ | |
1063 | ||
1064 | /* use at most thus many bytes */ | |
1065 | bitstream_init(&bs, p->code, BM_PACKET_VLI_BYTES_MAX, 0); | |
1066 | memset(p->code, 0, BM_PACKET_VLI_BYTES_MAX); | |
1067 | /* plain bits covered in this code string */ | |
1068 | plain_bits = 0; | |
1069 | ||
1070 | /* p->encoding & 0x80 stores whether the first run length is set. | |
1071 | * bit offset is implicit. | |
1072 | * start with toggle == 2 to be able to tell the first iteration */ | |
1073 | toggle = 2; | |
1074 | ||
1075 | /* see how much plain bits we can stuff into one packet | |
1076 | * using RLE and VLI. */ | |
1077 | do { | |
1078 | tmp = (toggle == 0) ? _drbd_bm_find_next_zero(mdev, c->bit_offset) | |
1079 | : _drbd_bm_find_next(mdev, c->bit_offset); | |
1080 | if (tmp == -1UL) | |
1081 | tmp = c->bm_bits; | |
1082 | rl = tmp - c->bit_offset; | |
1083 | ||
1084 | if (toggle == 2) { /* first iteration */ | |
1085 | if (rl == 0) { | |
1086 | /* the first checked bit was set, | |
1087 | * store start value, */ | |
1088 | DCBP_set_start(p, 1); | |
1089 | /* but skip encoding of zero run length */ | |
1090 | toggle = !toggle; | |
1091 | continue; | |
1092 | } | |
1093 | DCBP_set_start(p, 0); | |
1094 | } | |
1095 | ||
1096 | /* paranoia: catch zero runlength. | |
1097 | * can only happen if bitmap is modified while we scan it. */ | |
1098 | if (rl == 0) { | |
1099 | dev_err(DEV, "unexpected zero runlength while encoding bitmap " | |
1100 | "t:%u bo:%lu\n", toggle, c->bit_offset); | |
1101 | return -1; | |
1102 | } | |
1103 | ||
1104 | bits = vli_encode_bits(&bs, rl); | |
1105 | if (bits == -ENOBUFS) /* buffer full */ | |
1106 | break; | |
1107 | if (bits <= 0) { | |
1108 | dev_err(DEV, "error while encoding bitmap: %d\n", bits); | |
1109 | return 0; | |
1110 | } | |
1111 | ||
1112 | toggle = !toggle; | |
1113 | plain_bits += rl; | |
1114 | c->bit_offset = tmp; | |
1115 | } while (c->bit_offset < c->bm_bits); | |
1116 | ||
1117 | len = bs.cur.b - p->code + !!bs.cur.bit; | |
1118 | ||
1119 | if (plain_bits < (len << 3)) { | |
1120 | /* incompressible with this method. | |
1121 | * we need to rewind both word and bit position. */ | |
1122 | c->bit_offset -= plain_bits; | |
1123 | bm_xfer_ctx_bit_to_word_offset(c); | |
1124 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1125 | return 0; | |
1126 | } | |
1127 | ||
1128 | /* RLE + VLI was able to compress it just fine. | |
1129 | * update c->word_offset. */ | |
1130 | bm_xfer_ctx_bit_to_word_offset(c); | |
1131 | ||
1132 | /* store pad_bits */ | |
1133 | DCBP_set_pad_bits(p, (8 - bs.cur.bit) & 0x7); | |
1134 | ||
1135 | return len; | |
1136 | } | |
1137 | ||
f70af118 AG |
1138 | /** |
1139 | * send_bitmap_rle_or_plain | |
1140 | * | |
1141 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
1142 | * code upon failure. | |
1143 | */ | |
1144 | static int | |
b411b363 | 1145 | send_bitmap_rle_or_plain(struct drbd_conf *mdev, |
c012949a | 1146 | struct p_header *h, struct bm_xfer_ctx *c) |
b411b363 PR |
1147 | { |
1148 | struct p_compressed_bm *p = (void*)h; | |
1149 | unsigned long num_words; | |
1150 | int len; | |
1151 | int ok; | |
1152 | ||
1153 | len = fill_bitmap_rle_bits(mdev, p, c); | |
1154 | ||
1155 | if (len < 0) | |
f70af118 | 1156 | return -EIO; |
b411b363 PR |
1157 | |
1158 | if (len) { | |
1159 | DCBP_set_code(p, RLE_VLI_Bits); | |
e42325a5 | 1160 | ok = _drbd_send_cmd(mdev, mdev->tconn->data.socket, P_COMPRESSED_BITMAP, h, |
b411b363 PR |
1161 | sizeof(*p) + len, 0); |
1162 | ||
1163 | c->packets[0]++; | |
1164 | c->bytes[0] += sizeof(*p) + len; | |
1165 | ||
1166 | if (c->bit_offset >= c->bm_bits) | |
1167 | len = 0; /* DONE */ | |
1168 | } else { | |
1169 | /* was not compressible. | |
1170 | * send a buffer full of plain text bits instead. */ | |
1171 | num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); | |
1172 | len = num_words * sizeof(long); | |
1173 | if (len) | |
1174 | drbd_bm_get_lel(mdev, c->word_offset, num_words, (unsigned long*)h->payload); | |
e42325a5 | 1175 | ok = _drbd_send_cmd(mdev, mdev->tconn->data.socket, P_BITMAP, |
0b70a13d | 1176 | h, sizeof(struct p_header80) + len, 0); |
b411b363 PR |
1177 | c->word_offset += num_words; |
1178 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
1179 | ||
1180 | c->packets[1]++; | |
0b70a13d | 1181 | c->bytes[1] += sizeof(struct p_header80) + len; |
b411b363 PR |
1182 | |
1183 | if (c->bit_offset > c->bm_bits) | |
1184 | c->bit_offset = c->bm_bits; | |
1185 | } | |
f70af118 AG |
1186 | if (ok) { |
1187 | if (len == 0) { | |
1188 | INFO_bm_xfer_stats(mdev, "send", c); | |
1189 | return 0; | |
1190 | } else | |
1191 | return 1; | |
1192 | } | |
1193 | return -EIO; | |
b411b363 PR |
1194 | } |
1195 | ||
1196 | /* See the comment at receive_bitmap() */ | |
1197 | int _drbd_send_bitmap(struct drbd_conf *mdev) | |
1198 | { | |
1199 | struct bm_xfer_ctx c; | |
c012949a | 1200 | struct p_header *p; |
f70af118 | 1201 | int err; |
b411b363 | 1202 | |
841ce241 AG |
1203 | if (!expect(mdev->bitmap)) |
1204 | return false; | |
b411b363 PR |
1205 | |
1206 | /* maybe we should use some per thread scratch page, | |
1207 | * and allocate that during initial device creation? */ | |
c012949a | 1208 | p = (struct p_header *) __get_free_page(GFP_NOIO); |
b411b363 PR |
1209 | if (!p) { |
1210 | dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); | |
81e84650 | 1211 | return false; |
b411b363 PR |
1212 | } |
1213 | ||
1214 | if (get_ldev(mdev)) { | |
1215 | if (drbd_md_test_flag(mdev->ldev, MDF_FULL_SYNC)) { | |
1216 | dev_info(DEV, "Writing the whole bitmap, MDF_FullSync was set.\n"); | |
1217 | drbd_bm_set_all(mdev); | |
1218 | if (drbd_bm_write(mdev)) { | |
1219 | /* write_bm did fail! Leave full sync flag set in Meta P_DATA | |
1220 | * but otherwise process as per normal - need to tell other | |
1221 | * side that a full resync is required! */ | |
1222 | dev_err(DEV, "Failed to write bitmap to disk!\n"); | |
1223 | } else { | |
1224 | drbd_md_clear_flag(mdev, MDF_FULL_SYNC); | |
1225 | drbd_md_sync(mdev); | |
1226 | } | |
1227 | } | |
1228 | put_ldev(mdev); | |
1229 | } | |
1230 | ||
1231 | c = (struct bm_xfer_ctx) { | |
1232 | .bm_bits = drbd_bm_bits(mdev), | |
1233 | .bm_words = drbd_bm_words(mdev), | |
1234 | }; | |
1235 | ||
1236 | do { | |
f70af118 AG |
1237 | err = send_bitmap_rle_or_plain(mdev, p, &c); |
1238 | } while (err > 0); | |
b411b363 PR |
1239 | |
1240 | free_page((unsigned long) p); | |
f70af118 | 1241 | return err == 0; |
b411b363 PR |
1242 | } |
1243 | ||
1244 | int drbd_send_bitmap(struct drbd_conf *mdev) | |
1245 | { | |
1246 | int err; | |
1247 | ||
11b0be28 | 1248 | if (drbd_get_data_sock(mdev->tconn)) |
b411b363 PR |
1249 | return -1; |
1250 | err = !_drbd_send_bitmap(mdev); | |
61120870 | 1251 | drbd_put_data_sock(mdev->tconn); |
b411b363 PR |
1252 | return err; |
1253 | } | |
1254 | ||
1255 | int drbd_send_b_ack(struct drbd_conf *mdev, u32 barrier_nr, u32 set_size) | |
1256 | { | |
1257 | int ok; | |
1258 | struct p_barrier_ack p; | |
1259 | ||
1260 | p.barrier = barrier_nr; | |
1261 | p.set_size = cpu_to_be32(set_size); | |
1262 | ||
1263 | if (mdev->state.conn < C_CONNECTED) | |
81e84650 | 1264 | return false; |
c012949a | 1265 | ok = drbd_send_cmd(mdev, USE_META_SOCKET, P_BARRIER_ACK, &p.head, sizeof(p)); |
b411b363 PR |
1266 | return ok; |
1267 | } | |
1268 | ||
1269 | /** | |
1270 | * _drbd_send_ack() - Sends an ack packet | |
1271 | * @mdev: DRBD device. | |
1272 | * @cmd: Packet command code. | |
1273 | * @sector: sector, needs to be in big endian byte order | |
1274 | * @blksize: size in byte, needs to be in big endian byte order | |
1275 | * @block_id: Id, big endian byte order | |
1276 | */ | |
d8763023 AG |
1277 | static int _drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd, |
1278 | u64 sector, u32 blksize, u64 block_id) | |
b411b363 PR |
1279 | { |
1280 | int ok; | |
1281 | struct p_block_ack p; | |
1282 | ||
1283 | p.sector = sector; | |
1284 | p.block_id = block_id; | |
1285 | p.blksize = blksize; | |
8ccf218e | 1286 | p.seq_num = cpu_to_be32(atomic_inc_return(&mdev->packet_seq)); |
b411b363 | 1287 | |
e42325a5 | 1288 | if (!mdev->tconn->meta.socket || mdev->state.conn < C_CONNECTED) |
81e84650 | 1289 | return false; |
c012949a | 1290 | ok = drbd_send_cmd(mdev, USE_META_SOCKET, cmd, &p.head, sizeof(p)); |
b411b363 PR |
1291 | return ok; |
1292 | } | |
1293 | ||
2b2bf214 LE |
1294 | /* dp->sector and dp->block_id already/still in network byte order, |
1295 | * data_size is payload size according to dp->head, | |
1296 | * and may need to be corrected for digest size. */ | |
d8763023 | 1297 | int drbd_send_ack_dp(struct drbd_conf *mdev, enum drbd_packet cmd, |
2b2bf214 | 1298 | struct p_data *dp, int data_size) |
b411b363 | 1299 | { |
a0638456 PR |
1300 | data_size -= (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ? |
1301 | crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0; | |
b411b363 PR |
1302 | return _drbd_send_ack(mdev, cmd, dp->sector, cpu_to_be32(data_size), |
1303 | dp->block_id); | |
1304 | } | |
1305 | ||
d8763023 | 1306 | int drbd_send_ack_rp(struct drbd_conf *mdev, enum drbd_packet cmd, |
b411b363 PR |
1307 | struct p_block_req *rp) |
1308 | { | |
1309 | return _drbd_send_ack(mdev, cmd, rp->sector, rp->blksize, rp->block_id); | |
1310 | } | |
1311 | ||
1312 | /** | |
1313 | * drbd_send_ack() - Sends an ack packet | |
db830c46 AG |
1314 | * @mdev: DRBD device |
1315 | * @cmd: packet command code | |
1316 | * @peer_req: peer request | |
b411b363 | 1317 | */ |
d8763023 | 1318 | int drbd_send_ack(struct drbd_conf *mdev, enum drbd_packet cmd, |
db830c46 | 1319 | struct drbd_peer_request *peer_req) |
b411b363 PR |
1320 | { |
1321 | return _drbd_send_ack(mdev, cmd, | |
db830c46 AG |
1322 | cpu_to_be64(peer_req->i.sector), |
1323 | cpu_to_be32(peer_req->i.size), | |
1324 | peer_req->block_id); | |
b411b363 PR |
1325 | } |
1326 | ||
1327 | /* This function misuses the block_id field to signal if the blocks | |
1328 | * are is sync or not. */ | |
d8763023 | 1329 | int drbd_send_ack_ex(struct drbd_conf *mdev, enum drbd_packet cmd, |
b411b363 PR |
1330 | sector_t sector, int blksize, u64 block_id) |
1331 | { | |
1332 | return _drbd_send_ack(mdev, cmd, | |
1333 | cpu_to_be64(sector), | |
1334 | cpu_to_be32(blksize), | |
1335 | cpu_to_be64(block_id)); | |
1336 | } | |
1337 | ||
1338 | int drbd_send_drequest(struct drbd_conf *mdev, int cmd, | |
1339 | sector_t sector, int size, u64 block_id) | |
1340 | { | |
1341 | int ok; | |
1342 | struct p_block_req p; | |
1343 | ||
1344 | p.sector = cpu_to_be64(sector); | |
1345 | p.block_id = block_id; | |
1346 | p.blksize = cpu_to_be32(size); | |
1347 | ||
c012949a | 1348 | ok = drbd_send_cmd(mdev, USE_DATA_SOCKET, cmd, &p.head, sizeof(p)); |
b411b363 PR |
1349 | return ok; |
1350 | } | |
1351 | ||
d8763023 AG |
1352 | int drbd_send_drequest_csum(struct drbd_conf *mdev, sector_t sector, int size, |
1353 | void *digest, int digest_size, enum drbd_packet cmd) | |
b411b363 PR |
1354 | { |
1355 | int ok; | |
1356 | struct p_block_req p; | |
1357 | ||
fd340c12 | 1358 | prepare_header(mdev, &p.head, cmd, sizeof(p) - sizeof(struct p_header) + digest_size); |
b411b363 | 1359 | p.sector = cpu_to_be64(sector); |
9a8e7753 | 1360 | p.block_id = ID_SYNCER /* unused */; |
b411b363 PR |
1361 | p.blksize = cpu_to_be32(size); |
1362 | ||
e42325a5 | 1363 | mutex_lock(&mdev->tconn->data.mutex); |
b411b363 | 1364 | |
bedbd2a5 PR |
1365 | ok = (sizeof(p) == drbd_send(mdev->tconn, mdev->tconn->data.socket, &p, sizeof(p), 0)); |
1366 | ok = ok && (digest_size == drbd_send(mdev->tconn, mdev->tconn->data.socket, digest, digest_size, 0)); | |
b411b363 | 1367 | |
e42325a5 | 1368 | mutex_unlock(&mdev->tconn->data.mutex); |
b411b363 PR |
1369 | |
1370 | return ok; | |
1371 | } | |
1372 | ||
1373 | int drbd_send_ov_request(struct drbd_conf *mdev, sector_t sector, int size) | |
1374 | { | |
1375 | int ok; | |
1376 | struct p_block_req p; | |
1377 | ||
1378 | p.sector = cpu_to_be64(sector); | |
9a8e7753 | 1379 | p.block_id = ID_SYNCER /* unused */; |
b411b363 PR |
1380 | p.blksize = cpu_to_be32(size); |
1381 | ||
c012949a | 1382 | ok = drbd_send_cmd(mdev, USE_DATA_SOCKET, P_OV_REQUEST, &p.head, sizeof(p)); |
b411b363 PR |
1383 | return ok; |
1384 | } | |
1385 | ||
1386 | /* called on sndtimeo | |
81e84650 AG |
1387 | * returns false if we should retry, |
1388 | * true if we think connection is dead | |
b411b363 | 1389 | */ |
1a7ba646 | 1390 | static int we_should_drop_the_connection(struct drbd_tconn *tconn, struct socket *sock) |
b411b363 PR |
1391 | { |
1392 | int drop_it; | |
1393 | /* long elapsed = (long)(jiffies - mdev->last_received); */ | |
1394 | ||
1a7ba646 PR |
1395 | drop_it = tconn->meta.socket == sock |
1396 | || !tconn->asender.task | |
1397 | || get_t_state(&tconn->asender) != RUNNING | |
bbeb641c | 1398 | || tconn->cstate < C_WF_REPORT_PARAMS; |
b411b363 PR |
1399 | |
1400 | if (drop_it) | |
81e84650 | 1401 | return true; |
b411b363 | 1402 | |
1a7ba646 | 1403 | drop_it = !--tconn->ko_count; |
b411b363 | 1404 | if (!drop_it) { |
1a7ba646 PR |
1405 | conn_err(tconn, "[%s/%d] sock_sendmsg time expired, ko = %u\n", |
1406 | current->comm, current->pid, tconn->ko_count); | |
1407 | request_ping(tconn); | |
b411b363 PR |
1408 | } |
1409 | ||
1410 | return drop_it; /* && (mdev->state == R_PRIMARY) */; | |
1411 | } | |
1412 | ||
1a7ba646 | 1413 | static void drbd_update_congested(struct drbd_tconn *tconn) |
9e204cdd | 1414 | { |
1a7ba646 | 1415 | struct sock *sk = tconn->data.socket->sk; |
9e204cdd | 1416 | if (sk->sk_wmem_queued > sk->sk_sndbuf * 4 / 5) |
1a7ba646 | 1417 | set_bit(NET_CONGESTED, &tconn->flags); |
9e204cdd AG |
1418 | } |
1419 | ||
b411b363 PR |
1420 | /* The idea of sendpage seems to be to put some kind of reference |
1421 | * to the page into the skb, and to hand it over to the NIC. In | |
1422 | * this process get_page() gets called. | |
1423 | * | |
1424 | * As soon as the page was really sent over the network put_page() | |
1425 | * gets called by some part of the network layer. [ NIC driver? ] | |
1426 | * | |
1427 | * [ get_page() / put_page() increment/decrement the count. If count | |
1428 | * reaches 0 the page will be freed. ] | |
1429 | * | |
1430 | * This works nicely with pages from FSs. | |
1431 | * But this means that in protocol A we might signal IO completion too early! | |
1432 | * | |
1433 | * In order not to corrupt data during a resync we must make sure | |
1434 | * that we do not reuse our own buffer pages (EEs) to early, therefore | |
1435 | * we have the net_ee list. | |
1436 | * | |
1437 | * XFS seems to have problems, still, it submits pages with page_count == 0! | |
1438 | * As a workaround, we disable sendpage on pages | |
1439 | * with page_count == 0 or PageSlab. | |
1440 | */ | |
1441 | static int _drbd_no_send_page(struct drbd_conf *mdev, struct page *page, | |
ba11ad9a | 1442 | int offset, size_t size, unsigned msg_flags) |
b411b363 | 1443 | { |
bedbd2a5 | 1444 | int sent = drbd_send(mdev->tconn, mdev->tconn->data.socket, kmap(page) + offset, size, msg_flags); |
b411b363 PR |
1445 | kunmap(page); |
1446 | if (sent == size) | |
1447 | mdev->send_cnt += size>>9; | |
1448 | return sent == size; | |
1449 | } | |
1450 | ||
1451 | static int _drbd_send_page(struct drbd_conf *mdev, struct page *page, | |
ba11ad9a | 1452 | int offset, size_t size, unsigned msg_flags) |
b411b363 PR |
1453 | { |
1454 | mm_segment_t oldfs = get_fs(); | |
1455 | int sent, ok; | |
1456 | int len = size; | |
1457 | ||
1458 | /* e.g. XFS meta- & log-data is in slab pages, which have a | |
1459 | * page_count of 0 and/or have PageSlab() set. | |
1460 | * we cannot use send_page for those, as that does get_page(); | |
1461 | * put_page(); and would cause either a VM_BUG directly, or | |
1462 | * __page_cache_release a page that would actually still be referenced | |
1463 | * by someone, leading to some obscure delayed Oops somewhere else. */ | |
1464 | if (disable_sendpage || (page_count(page) < 1) || PageSlab(page)) | |
ba11ad9a | 1465 | return _drbd_no_send_page(mdev, page, offset, size, msg_flags); |
b411b363 | 1466 | |
ba11ad9a | 1467 | msg_flags |= MSG_NOSIGNAL; |
1a7ba646 | 1468 | drbd_update_congested(mdev->tconn); |
b411b363 PR |
1469 | set_fs(KERNEL_DS); |
1470 | do { | |
e42325a5 | 1471 | sent = mdev->tconn->data.socket->ops->sendpage(mdev->tconn->data.socket, page, |
b411b363 | 1472 | offset, len, |
ba11ad9a | 1473 | msg_flags); |
b411b363 | 1474 | if (sent == -EAGAIN) { |
1a7ba646 | 1475 | if (we_should_drop_the_connection(mdev->tconn, |
e42325a5 | 1476 | mdev->tconn->data.socket)) |
b411b363 PR |
1477 | break; |
1478 | else | |
1479 | continue; | |
1480 | } | |
1481 | if (sent <= 0) { | |
1482 | dev_warn(DEV, "%s: size=%d len=%d sent=%d\n", | |
1483 | __func__, (int)size, len, sent); | |
1484 | break; | |
1485 | } | |
1486 | len -= sent; | |
1487 | offset += sent; | |
1488 | } while (len > 0 /* THINK && mdev->cstate >= C_CONNECTED*/); | |
1489 | set_fs(oldfs); | |
01a311a5 | 1490 | clear_bit(NET_CONGESTED, &mdev->tconn->flags); |
b411b363 PR |
1491 | |
1492 | ok = (len == 0); | |
1493 | if (likely(ok)) | |
1494 | mdev->send_cnt += size>>9; | |
1495 | return ok; | |
1496 | } | |
1497 | ||
1498 | static int _drbd_send_bio(struct drbd_conf *mdev, struct bio *bio) | |
1499 | { | |
1500 | struct bio_vec *bvec; | |
1501 | int i; | |
ba11ad9a | 1502 | /* hint all but last page with MSG_MORE */ |
b411b363 PR |
1503 | __bio_for_each_segment(bvec, bio, i, 0) { |
1504 | if (!_drbd_no_send_page(mdev, bvec->bv_page, | |
ba11ad9a LE |
1505 | bvec->bv_offset, bvec->bv_len, |
1506 | i == bio->bi_vcnt -1 ? 0 : MSG_MORE)) | |
b411b363 PR |
1507 | return 0; |
1508 | } | |
1509 | return 1; | |
1510 | } | |
1511 | ||
1512 | static int _drbd_send_zc_bio(struct drbd_conf *mdev, struct bio *bio) | |
1513 | { | |
1514 | struct bio_vec *bvec; | |
1515 | int i; | |
ba11ad9a | 1516 | /* hint all but last page with MSG_MORE */ |
b411b363 PR |
1517 | __bio_for_each_segment(bvec, bio, i, 0) { |
1518 | if (!_drbd_send_page(mdev, bvec->bv_page, | |
ba11ad9a LE |
1519 | bvec->bv_offset, bvec->bv_len, |
1520 | i == bio->bi_vcnt -1 ? 0 : MSG_MORE)) | |
b411b363 PR |
1521 | return 0; |
1522 | } | |
b411b363 PR |
1523 | return 1; |
1524 | } | |
1525 | ||
db830c46 AG |
1526 | static int _drbd_send_zc_ee(struct drbd_conf *mdev, |
1527 | struct drbd_peer_request *peer_req) | |
45bb912b | 1528 | { |
db830c46 AG |
1529 | struct page *page = peer_req->pages; |
1530 | unsigned len = peer_req->i.size; | |
1531 | ||
ba11ad9a | 1532 | /* hint all but last page with MSG_MORE */ |
45bb912b LE |
1533 | page_chain_for_each(page) { |
1534 | unsigned l = min_t(unsigned, len, PAGE_SIZE); | |
ba11ad9a LE |
1535 | if (!_drbd_send_page(mdev, page, 0, l, |
1536 | page_chain_next(page) ? MSG_MORE : 0)) | |
45bb912b LE |
1537 | return 0; |
1538 | len -= l; | |
1539 | } | |
1540 | return 1; | |
1541 | } | |
1542 | ||
76d2e7ec PR |
1543 | static u32 bio_flags_to_wire(struct drbd_conf *mdev, unsigned long bi_rw) |
1544 | { | |
31890f4a | 1545 | if (mdev->tconn->agreed_pro_version >= 95) |
76d2e7ec | 1546 | return (bi_rw & REQ_SYNC ? DP_RW_SYNC : 0) | |
76d2e7ec PR |
1547 | (bi_rw & REQ_FUA ? DP_FUA : 0) | |
1548 | (bi_rw & REQ_FLUSH ? DP_FLUSH : 0) | | |
1549 | (bi_rw & REQ_DISCARD ? DP_DISCARD : 0); | |
1550 | else | |
721a9602 | 1551 | return bi_rw & REQ_SYNC ? DP_RW_SYNC : 0; |
76d2e7ec PR |
1552 | } |
1553 | ||
b411b363 PR |
1554 | /* Used to send write requests |
1555 | * R_PRIMARY -> Peer (P_DATA) | |
1556 | */ | |
1557 | int drbd_send_dblock(struct drbd_conf *mdev, struct drbd_request *req) | |
1558 | { | |
1559 | int ok = 1; | |
1560 | struct p_data p; | |
1561 | unsigned int dp_flags = 0; | |
1562 | void *dgb; | |
1563 | int dgs; | |
1564 | ||
11b0be28 | 1565 | if (drbd_get_data_sock(mdev->tconn)) |
b411b363 PR |
1566 | return 0; |
1567 | ||
a0638456 PR |
1568 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_w_tfm) ? |
1569 | crypto_hash_digestsize(mdev->tconn->integrity_w_tfm) : 0; | |
b411b363 | 1570 | |
fd340c12 | 1571 | prepare_header(mdev, &p.head, P_DATA, sizeof(p) - sizeof(struct p_header) + dgs + req->i.size); |
ace652ac | 1572 | p.sector = cpu_to_be64(req->i.sector); |
b411b363 | 1573 | p.block_id = (unsigned long)req; |
8ccf218e | 1574 | p.seq_num = cpu_to_be32(req->seq_num = atomic_inc_return(&mdev->packet_seq)); |
b411b363 | 1575 | |
76d2e7ec PR |
1576 | dp_flags = bio_flags_to_wire(mdev, req->master_bio->bi_rw); |
1577 | ||
b411b363 PR |
1578 | if (mdev->state.conn >= C_SYNC_SOURCE && |
1579 | mdev->state.conn <= C_PAUSED_SYNC_T) | |
1580 | dp_flags |= DP_MAY_SET_IN_SYNC; | |
1581 | ||
1582 | p.dp_flags = cpu_to_be32(dp_flags); | |
b411b363 PR |
1583 | set_bit(UNPLUG_REMOTE, &mdev->flags); |
1584 | ok = (sizeof(p) == | |
bedbd2a5 | 1585 | drbd_send(mdev->tconn, mdev->tconn->data.socket, &p, sizeof(p), dgs ? MSG_MORE : 0)); |
b411b363 | 1586 | if (ok && dgs) { |
a0638456 PR |
1587 | dgb = mdev->tconn->int_dig_out; |
1588 | drbd_csum_bio(mdev, mdev->tconn->integrity_w_tfm, req->master_bio, dgb); | |
bedbd2a5 | 1589 | ok = dgs == drbd_send(mdev->tconn, mdev->tconn->data.socket, dgb, dgs, 0); |
b411b363 PR |
1590 | } |
1591 | if (ok) { | |
470be44a LE |
1592 | /* For protocol A, we have to memcpy the payload into |
1593 | * socket buffers, as we may complete right away | |
1594 | * as soon as we handed it over to tcp, at which point the data | |
1595 | * pages may become invalid. | |
1596 | * | |
1597 | * For data-integrity enabled, we copy it as well, so we can be | |
1598 | * sure that even if the bio pages may still be modified, it | |
1599 | * won't change the data on the wire, thus if the digest checks | |
1600 | * out ok after sending on this side, but does not fit on the | |
1601 | * receiving side, we sure have detected corruption elsewhere. | |
1602 | */ | |
89e58e75 | 1603 | if (mdev->tconn->net_conf->wire_protocol == DRBD_PROT_A || dgs) |
b411b363 PR |
1604 | ok = _drbd_send_bio(mdev, req->master_bio); |
1605 | else | |
1606 | ok = _drbd_send_zc_bio(mdev, req->master_bio); | |
470be44a LE |
1607 | |
1608 | /* double check digest, sometimes buffers have been modified in flight. */ | |
1609 | if (dgs > 0 && dgs <= 64) { | |
24c4830c | 1610 | /* 64 byte, 512 bit, is the largest digest size |
470be44a LE |
1611 | * currently supported in kernel crypto. */ |
1612 | unsigned char digest[64]; | |
a0638456 PR |
1613 | drbd_csum_bio(mdev, mdev->tconn->integrity_w_tfm, req->master_bio, digest); |
1614 | if (memcmp(mdev->tconn->int_dig_out, digest, dgs)) { | |
470be44a LE |
1615 | dev_warn(DEV, |
1616 | "Digest mismatch, buffer modified by upper layers during write: %llus +%u\n", | |
ace652ac | 1617 | (unsigned long long)req->i.sector, req->i.size); |
470be44a LE |
1618 | } |
1619 | } /* else if (dgs > 64) { | |
1620 | ... Be noisy about digest too large ... | |
1621 | } */ | |
b411b363 PR |
1622 | } |
1623 | ||
61120870 | 1624 | drbd_put_data_sock(mdev->tconn); |
bd26bfc5 | 1625 | |
b411b363 PR |
1626 | return ok; |
1627 | } | |
1628 | ||
1629 | /* answer packet, used to send data back for read requests: | |
1630 | * Peer -> (diskless) R_PRIMARY (P_DATA_REPLY) | |
1631 | * C_SYNC_SOURCE -> C_SYNC_TARGET (P_RS_DATA_REPLY) | |
1632 | */ | |
d8763023 | 1633 | int drbd_send_block(struct drbd_conf *mdev, enum drbd_packet cmd, |
db830c46 | 1634 | struct drbd_peer_request *peer_req) |
b411b363 PR |
1635 | { |
1636 | int ok; | |
1637 | struct p_data p; | |
1638 | void *dgb; | |
1639 | int dgs; | |
1640 | ||
a0638456 PR |
1641 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_w_tfm) ? |
1642 | crypto_hash_digestsize(mdev->tconn->integrity_w_tfm) : 0; | |
b411b363 | 1643 | |
db830c46 AG |
1644 | prepare_header(mdev, &p.head, cmd, sizeof(p) - |
1645 | sizeof(struct p_header80) + | |
1646 | dgs + peer_req->i.size); | |
1647 | p.sector = cpu_to_be64(peer_req->i.sector); | |
1648 | p.block_id = peer_req->block_id; | |
cc378270 | 1649 | p.seq_num = 0; /* unused */ |
b411b363 PR |
1650 | |
1651 | /* Only called by our kernel thread. | |
1652 | * This one may be interrupted by DRBD_SIG and/or DRBD_SIGKILL | |
1653 | * in response to admin command or module unload. | |
1654 | */ | |
11b0be28 | 1655 | if (drbd_get_data_sock(mdev->tconn)) |
b411b363 PR |
1656 | return 0; |
1657 | ||
bedbd2a5 | 1658 | ok = sizeof(p) == drbd_send(mdev->tconn, mdev->tconn->data.socket, &p, sizeof(p), dgs ? MSG_MORE : 0); |
b411b363 | 1659 | if (ok && dgs) { |
a0638456 | 1660 | dgb = mdev->tconn->int_dig_out; |
db830c46 | 1661 | drbd_csum_ee(mdev, mdev->tconn->integrity_w_tfm, peer_req, dgb); |
bedbd2a5 | 1662 | ok = dgs == drbd_send(mdev->tconn, mdev->tconn->data.socket, dgb, dgs, 0); |
b411b363 PR |
1663 | } |
1664 | if (ok) | |
db830c46 | 1665 | ok = _drbd_send_zc_ee(mdev, peer_req); |
b411b363 | 1666 | |
61120870 | 1667 | drbd_put_data_sock(mdev->tconn); |
bd26bfc5 | 1668 | |
b411b363 PR |
1669 | return ok; |
1670 | } | |
1671 | ||
73a01a18 PR |
1672 | int drbd_send_oos(struct drbd_conf *mdev, struct drbd_request *req) |
1673 | { | |
1674 | struct p_block_desc p; | |
1675 | ||
ace652ac AG |
1676 | p.sector = cpu_to_be64(req->i.sector); |
1677 | p.blksize = cpu_to_be32(req->i.size); | |
73a01a18 PR |
1678 | |
1679 | return drbd_send_cmd(mdev, USE_DATA_SOCKET, P_OUT_OF_SYNC, &p.head, sizeof(p)); | |
1680 | } | |
1681 | ||
b411b363 PR |
1682 | /* |
1683 | drbd_send distinguishes two cases: | |
1684 | ||
1685 | Packets sent via the data socket "sock" | |
1686 | and packets sent via the meta data socket "msock" | |
1687 | ||
1688 | sock msock | |
1689 | -----------------+-------------------------+------------------------------ | |
1690 | timeout conf.timeout / 2 conf.timeout / 2 | |
1691 | timeout action send a ping via msock Abort communication | |
1692 | and close all sockets | |
1693 | */ | |
1694 | ||
1695 | /* | |
1696 | * you must have down()ed the appropriate [m]sock_mutex elsewhere! | |
1697 | */ | |
bedbd2a5 | 1698 | int drbd_send(struct drbd_tconn *tconn, struct socket *sock, |
b411b363 PR |
1699 | void *buf, size_t size, unsigned msg_flags) |
1700 | { | |
1701 | struct kvec iov; | |
1702 | struct msghdr msg; | |
1703 | int rv, sent = 0; | |
1704 | ||
1705 | if (!sock) | |
c0d42c8e | 1706 | return -EBADR; |
b411b363 PR |
1707 | |
1708 | /* THINK if (signal_pending) return ... ? */ | |
1709 | ||
1710 | iov.iov_base = buf; | |
1711 | iov.iov_len = size; | |
1712 | ||
1713 | msg.msg_name = NULL; | |
1714 | msg.msg_namelen = 0; | |
1715 | msg.msg_control = NULL; | |
1716 | msg.msg_controllen = 0; | |
1717 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; | |
1718 | ||
bedbd2a5 PR |
1719 | if (sock == tconn->data.socket) { |
1720 | tconn->ko_count = tconn->net_conf->ko_count; | |
1721 | drbd_update_congested(tconn); | |
b411b363 PR |
1722 | } |
1723 | do { | |
1724 | /* STRANGE | |
1725 | * tcp_sendmsg does _not_ use its size parameter at all ? | |
1726 | * | |
1727 | * -EAGAIN on timeout, -EINTR on signal. | |
1728 | */ | |
1729 | /* THINK | |
1730 | * do we need to block DRBD_SIG if sock == &meta.socket ?? | |
1731 | * otherwise wake_asender() might interrupt some send_*Ack ! | |
1732 | */ | |
1733 | rv = kernel_sendmsg(sock, &msg, &iov, 1, size); | |
1734 | if (rv == -EAGAIN) { | |
bedbd2a5 | 1735 | if (we_should_drop_the_connection(tconn, sock)) |
b411b363 PR |
1736 | break; |
1737 | else | |
1738 | continue; | |
1739 | } | |
b411b363 PR |
1740 | if (rv == -EINTR) { |
1741 | flush_signals(current); | |
1742 | rv = 0; | |
1743 | } | |
1744 | if (rv < 0) | |
1745 | break; | |
1746 | sent += rv; | |
1747 | iov.iov_base += rv; | |
1748 | iov.iov_len -= rv; | |
1749 | } while (sent < size); | |
1750 | ||
bedbd2a5 PR |
1751 | if (sock == tconn->data.socket) |
1752 | clear_bit(NET_CONGESTED, &tconn->flags); | |
b411b363 PR |
1753 | |
1754 | if (rv <= 0) { | |
1755 | if (rv != -EAGAIN) { | |
bedbd2a5 PR |
1756 | conn_err(tconn, "%s_sendmsg returned %d\n", |
1757 | sock == tconn->meta.socket ? "msock" : "sock", | |
1758 | rv); | |
bbeb641c | 1759 | conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 | 1760 | } else |
bbeb641c | 1761 | conn_request_state(tconn, NS(conn, C_TIMEOUT), CS_HARD); |
b411b363 PR |
1762 | } |
1763 | ||
1764 | return sent; | |
1765 | } | |
1766 | ||
1767 | static int drbd_open(struct block_device *bdev, fmode_t mode) | |
1768 | { | |
1769 | struct drbd_conf *mdev = bdev->bd_disk->private_data; | |
1770 | unsigned long flags; | |
1771 | int rv = 0; | |
1772 | ||
2a48fc0a | 1773 | mutex_lock(&drbd_main_mutex); |
87eeee41 | 1774 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
b411b363 PR |
1775 | /* to have a stable mdev->state.role |
1776 | * and no race with updating open_cnt */ | |
1777 | ||
1778 | if (mdev->state.role != R_PRIMARY) { | |
1779 | if (mode & FMODE_WRITE) | |
1780 | rv = -EROFS; | |
1781 | else if (!allow_oos) | |
1782 | rv = -EMEDIUMTYPE; | |
1783 | } | |
1784 | ||
1785 | if (!rv) | |
1786 | mdev->open_cnt++; | |
87eeee41 | 1787 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
2a48fc0a | 1788 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1789 | |
1790 | return rv; | |
1791 | } | |
1792 | ||
1793 | static int drbd_release(struct gendisk *gd, fmode_t mode) | |
1794 | { | |
1795 | struct drbd_conf *mdev = gd->private_data; | |
2a48fc0a | 1796 | mutex_lock(&drbd_main_mutex); |
b411b363 | 1797 | mdev->open_cnt--; |
2a48fc0a | 1798 | mutex_unlock(&drbd_main_mutex); |
b411b363 PR |
1799 | return 0; |
1800 | } | |
1801 | ||
b411b363 PR |
1802 | static void drbd_set_defaults(struct drbd_conf *mdev) |
1803 | { | |
f399002e LE |
1804 | /* Beware! The actual layout differs |
1805 | * between big endian and little endian */ | |
b411b363 PR |
1806 | mdev->state = (union drbd_state) { |
1807 | { .role = R_SECONDARY, | |
1808 | .peer = R_UNKNOWN, | |
1809 | .conn = C_STANDALONE, | |
1810 | .disk = D_DISKLESS, | |
1811 | .pdsk = D_UNKNOWN, | |
fb22c402 PR |
1812 | .susp = 0, |
1813 | .susp_nod = 0, | |
1814 | .susp_fen = 0 | |
b411b363 PR |
1815 | } }; |
1816 | } | |
1817 | ||
1818 | void drbd_init_set_defaults(struct drbd_conf *mdev) | |
1819 | { | |
1820 | /* the memset(,0,) did most of this. | |
1821 | * note: only assignments, no allocation in here */ | |
1822 | ||
1823 | drbd_set_defaults(mdev); | |
1824 | ||
b411b363 PR |
1825 | atomic_set(&mdev->ap_bio_cnt, 0); |
1826 | atomic_set(&mdev->ap_pending_cnt, 0); | |
1827 | atomic_set(&mdev->rs_pending_cnt, 0); | |
1828 | atomic_set(&mdev->unacked_cnt, 0); | |
1829 | atomic_set(&mdev->local_cnt, 0); | |
b411b363 | 1830 | atomic_set(&mdev->pp_in_use, 0); |
435f0740 | 1831 | atomic_set(&mdev->pp_in_use_by_net, 0); |
778f271d | 1832 | atomic_set(&mdev->rs_sect_in, 0); |
0f0601f4 | 1833 | atomic_set(&mdev->rs_sect_ev, 0); |
759fbdfb | 1834 | atomic_set(&mdev->ap_in_flight, 0); |
b411b363 PR |
1835 | |
1836 | mutex_init(&mdev->md_io_mutex); | |
8410da8f PR |
1837 | mutex_init(&mdev->own_state_mutex); |
1838 | mdev->state_mutex = &mdev->own_state_mutex; | |
b411b363 | 1839 | |
b411b363 | 1840 | spin_lock_init(&mdev->al_lock); |
b411b363 PR |
1841 | spin_lock_init(&mdev->peer_seq_lock); |
1842 | spin_lock_init(&mdev->epoch_lock); | |
1843 | ||
1844 | INIT_LIST_HEAD(&mdev->active_ee); | |
1845 | INIT_LIST_HEAD(&mdev->sync_ee); | |
1846 | INIT_LIST_HEAD(&mdev->done_ee); | |
1847 | INIT_LIST_HEAD(&mdev->read_ee); | |
1848 | INIT_LIST_HEAD(&mdev->net_ee); | |
1849 | INIT_LIST_HEAD(&mdev->resync_reads); | |
b411b363 PR |
1850 | INIT_LIST_HEAD(&mdev->resync_work.list); |
1851 | INIT_LIST_HEAD(&mdev->unplug_work.list); | |
e9e6f3ec | 1852 | INIT_LIST_HEAD(&mdev->go_diskless.list); |
b411b363 | 1853 | INIT_LIST_HEAD(&mdev->md_sync_work.list); |
c4752ef1 | 1854 | INIT_LIST_HEAD(&mdev->start_resync_work.list); |
b411b363 | 1855 | INIT_LIST_HEAD(&mdev->bm_io_work.w.list); |
0ced55a3 | 1856 | |
794abb75 | 1857 | mdev->resync_work.cb = w_resync_timer; |
b411b363 | 1858 | mdev->unplug_work.cb = w_send_write_hint; |
e9e6f3ec | 1859 | mdev->go_diskless.cb = w_go_diskless; |
b411b363 PR |
1860 | mdev->md_sync_work.cb = w_md_sync; |
1861 | mdev->bm_io_work.w.cb = w_bitmap_io; | |
370a43e7 | 1862 | mdev->start_resync_work.cb = w_start_resync; |
a21e9298 PR |
1863 | |
1864 | mdev->resync_work.mdev = mdev; | |
1865 | mdev->unplug_work.mdev = mdev; | |
1866 | mdev->go_diskless.mdev = mdev; | |
1867 | mdev->md_sync_work.mdev = mdev; | |
1868 | mdev->bm_io_work.w.mdev = mdev; | |
1869 | mdev->start_resync_work.mdev = mdev; | |
1870 | ||
b411b363 PR |
1871 | init_timer(&mdev->resync_timer); |
1872 | init_timer(&mdev->md_sync_timer); | |
370a43e7 | 1873 | init_timer(&mdev->start_resync_timer); |
7fde2be9 | 1874 | init_timer(&mdev->request_timer); |
b411b363 PR |
1875 | mdev->resync_timer.function = resync_timer_fn; |
1876 | mdev->resync_timer.data = (unsigned long) mdev; | |
1877 | mdev->md_sync_timer.function = md_sync_timer_fn; | |
1878 | mdev->md_sync_timer.data = (unsigned long) mdev; | |
370a43e7 PR |
1879 | mdev->start_resync_timer.function = start_resync_timer_fn; |
1880 | mdev->start_resync_timer.data = (unsigned long) mdev; | |
7fde2be9 PR |
1881 | mdev->request_timer.function = request_timer_fn; |
1882 | mdev->request_timer.data = (unsigned long) mdev; | |
b411b363 PR |
1883 | |
1884 | init_waitqueue_head(&mdev->misc_wait); | |
1885 | init_waitqueue_head(&mdev->state_wait); | |
1886 | init_waitqueue_head(&mdev->ee_wait); | |
1887 | init_waitqueue_head(&mdev->al_wait); | |
1888 | init_waitqueue_head(&mdev->seq_wait); | |
1889 | ||
fd340c12 | 1890 | /* mdev->tconn->agreed_pro_version gets initialized in drbd_connect() */ |
2451fc3b | 1891 | mdev->write_ordering = WO_bdev_flush; |
b411b363 | 1892 | mdev->resync_wenr = LC_FREE; |
99432fcc PR |
1893 | mdev->peer_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; |
1894 | mdev->local_max_bio_size = DRBD_MAX_BIO_SIZE_SAFE; | |
b411b363 PR |
1895 | } |
1896 | ||
1897 | void drbd_mdev_cleanup(struct drbd_conf *mdev) | |
1898 | { | |
1d7734a0 | 1899 | int i; |
e6b3ea83 | 1900 | if (mdev->tconn->receiver.t_state != NONE) |
b411b363 | 1901 | dev_err(DEV, "ASSERT FAILED: receiver t_state == %d expected 0.\n", |
e6b3ea83 | 1902 | mdev->tconn->receiver.t_state); |
b411b363 PR |
1903 | |
1904 | /* no need to lock it, I'm the only thread alive */ | |
1905 | if (atomic_read(&mdev->current_epoch->epoch_size) != 0) | |
1906 | dev_err(DEV, "epoch_size:%d\n", atomic_read(&mdev->current_epoch->epoch_size)); | |
1907 | mdev->al_writ_cnt = | |
1908 | mdev->bm_writ_cnt = | |
1909 | mdev->read_cnt = | |
1910 | mdev->recv_cnt = | |
1911 | mdev->send_cnt = | |
1912 | mdev->writ_cnt = | |
1913 | mdev->p_size = | |
1914 | mdev->rs_start = | |
1915 | mdev->rs_total = | |
1d7734a0 LE |
1916 | mdev->rs_failed = 0; |
1917 | mdev->rs_last_events = 0; | |
0f0601f4 | 1918 | mdev->rs_last_sect_ev = 0; |
1d7734a0 LE |
1919 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
1920 | mdev->rs_mark_left[i] = 0; | |
1921 | mdev->rs_mark_time[i] = 0; | |
1922 | } | |
89e58e75 | 1923 | D_ASSERT(mdev->tconn->net_conf == NULL); |
b411b363 PR |
1924 | |
1925 | drbd_set_my_capacity(mdev, 0); | |
1926 | if (mdev->bitmap) { | |
1927 | /* maybe never allocated. */ | |
02d9a94b | 1928 | drbd_bm_resize(mdev, 0, 1); |
b411b363 PR |
1929 | drbd_bm_cleanup(mdev); |
1930 | } | |
1931 | ||
1932 | drbd_free_resources(mdev); | |
0778286a | 1933 | clear_bit(AL_SUSPENDED, &mdev->flags); |
b411b363 PR |
1934 | |
1935 | /* | |
1936 | * currently we drbd_init_ee only on module load, so | |
1937 | * we may do drbd_release_ee only on module unload! | |
1938 | */ | |
1939 | D_ASSERT(list_empty(&mdev->active_ee)); | |
1940 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
1941 | D_ASSERT(list_empty(&mdev->done_ee)); | |
1942 | D_ASSERT(list_empty(&mdev->read_ee)); | |
1943 | D_ASSERT(list_empty(&mdev->net_ee)); | |
1944 | D_ASSERT(list_empty(&mdev->resync_reads)); | |
e42325a5 PR |
1945 | D_ASSERT(list_empty(&mdev->tconn->data.work.q)); |
1946 | D_ASSERT(list_empty(&mdev->tconn->meta.work.q)); | |
b411b363 PR |
1947 | D_ASSERT(list_empty(&mdev->resync_work.list)); |
1948 | D_ASSERT(list_empty(&mdev->unplug_work.list)); | |
e9e6f3ec | 1949 | D_ASSERT(list_empty(&mdev->go_diskless.list)); |
2265b473 LE |
1950 | |
1951 | drbd_set_defaults(mdev); | |
b411b363 PR |
1952 | } |
1953 | ||
1954 | ||
1955 | static void drbd_destroy_mempools(void) | |
1956 | { | |
1957 | struct page *page; | |
1958 | ||
1959 | while (drbd_pp_pool) { | |
1960 | page = drbd_pp_pool; | |
1961 | drbd_pp_pool = (struct page *)page_private(page); | |
1962 | __free_page(page); | |
1963 | drbd_pp_vacant--; | |
1964 | } | |
1965 | ||
1966 | /* D_ASSERT(atomic_read(&drbd_pp_vacant)==0); */ | |
1967 | ||
da4a75d2 LE |
1968 | if (drbd_md_io_bio_set) |
1969 | bioset_free(drbd_md_io_bio_set); | |
35abf594 LE |
1970 | if (drbd_md_io_page_pool) |
1971 | mempool_destroy(drbd_md_io_page_pool); | |
b411b363 PR |
1972 | if (drbd_ee_mempool) |
1973 | mempool_destroy(drbd_ee_mempool); | |
1974 | if (drbd_request_mempool) | |
1975 | mempool_destroy(drbd_request_mempool); | |
1976 | if (drbd_ee_cache) | |
1977 | kmem_cache_destroy(drbd_ee_cache); | |
1978 | if (drbd_request_cache) | |
1979 | kmem_cache_destroy(drbd_request_cache); | |
1980 | if (drbd_bm_ext_cache) | |
1981 | kmem_cache_destroy(drbd_bm_ext_cache); | |
1982 | if (drbd_al_ext_cache) | |
1983 | kmem_cache_destroy(drbd_al_ext_cache); | |
1984 | ||
da4a75d2 | 1985 | drbd_md_io_bio_set = NULL; |
35abf594 | 1986 | drbd_md_io_page_pool = NULL; |
b411b363 PR |
1987 | drbd_ee_mempool = NULL; |
1988 | drbd_request_mempool = NULL; | |
1989 | drbd_ee_cache = NULL; | |
1990 | drbd_request_cache = NULL; | |
1991 | drbd_bm_ext_cache = NULL; | |
1992 | drbd_al_ext_cache = NULL; | |
1993 | ||
1994 | return; | |
1995 | } | |
1996 | ||
1997 | static int drbd_create_mempools(void) | |
1998 | { | |
1999 | struct page *page; | |
1816a2b4 | 2000 | const int number = (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count; |
b411b363 PR |
2001 | int i; |
2002 | ||
2003 | /* prepare our caches and mempools */ | |
2004 | drbd_request_mempool = NULL; | |
2005 | drbd_ee_cache = NULL; | |
2006 | drbd_request_cache = NULL; | |
2007 | drbd_bm_ext_cache = NULL; | |
2008 | drbd_al_ext_cache = NULL; | |
2009 | drbd_pp_pool = NULL; | |
35abf594 | 2010 | drbd_md_io_page_pool = NULL; |
da4a75d2 | 2011 | drbd_md_io_bio_set = NULL; |
b411b363 PR |
2012 | |
2013 | /* caches */ | |
2014 | drbd_request_cache = kmem_cache_create( | |
2015 | "drbd_req", sizeof(struct drbd_request), 0, 0, NULL); | |
2016 | if (drbd_request_cache == NULL) | |
2017 | goto Enomem; | |
2018 | ||
2019 | drbd_ee_cache = kmem_cache_create( | |
f6ffca9f | 2020 | "drbd_ee", sizeof(struct drbd_peer_request), 0, 0, NULL); |
b411b363 PR |
2021 | if (drbd_ee_cache == NULL) |
2022 | goto Enomem; | |
2023 | ||
2024 | drbd_bm_ext_cache = kmem_cache_create( | |
2025 | "drbd_bm", sizeof(struct bm_extent), 0, 0, NULL); | |
2026 | if (drbd_bm_ext_cache == NULL) | |
2027 | goto Enomem; | |
2028 | ||
2029 | drbd_al_ext_cache = kmem_cache_create( | |
2030 | "drbd_al", sizeof(struct lc_element), 0, 0, NULL); | |
2031 | if (drbd_al_ext_cache == NULL) | |
2032 | goto Enomem; | |
2033 | ||
2034 | /* mempools */ | |
da4a75d2 LE |
2035 | drbd_md_io_bio_set = bioset_create(DRBD_MIN_POOL_PAGES, 0); |
2036 | if (drbd_md_io_bio_set == NULL) | |
2037 | goto Enomem; | |
2038 | ||
35abf594 LE |
2039 | drbd_md_io_page_pool = mempool_create_page_pool(DRBD_MIN_POOL_PAGES, 0); |
2040 | if (drbd_md_io_page_pool == NULL) | |
2041 | goto Enomem; | |
2042 | ||
b411b363 PR |
2043 | drbd_request_mempool = mempool_create(number, |
2044 | mempool_alloc_slab, mempool_free_slab, drbd_request_cache); | |
2045 | if (drbd_request_mempool == NULL) | |
2046 | goto Enomem; | |
2047 | ||
2048 | drbd_ee_mempool = mempool_create(number, | |
2049 | mempool_alloc_slab, mempool_free_slab, drbd_ee_cache); | |
2027ae1f | 2050 | if (drbd_ee_mempool == NULL) |
b411b363 PR |
2051 | goto Enomem; |
2052 | ||
2053 | /* drbd's page pool */ | |
2054 | spin_lock_init(&drbd_pp_lock); | |
2055 | ||
2056 | for (i = 0; i < number; i++) { | |
2057 | page = alloc_page(GFP_HIGHUSER); | |
2058 | if (!page) | |
2059 | goto Enomem; | |
2060 | set_page_private(page, (unsigned long)drbd_pp_pool); | |
2061 | drbd_pp_pool = page; | |
2062 | } | |
2063 | drbd_pp_vacant = number; | |
2064 | ||
2065 | return 0; | |
2066 | ||
2067 | Enomem: | |
2068 | drbd_destroy_mempools(); /* in case we allocated some */ | |
2069 | return -ENOMEM; | |
2070 | } | |
2071 | ||
2072 | static int drbd_notify_sys(struct notifier_block *this, unsigned long code, | |
2073 | void *unused) | |
2074 | { | |
2075 | /* just so we have it. you never know what interesting things we | |
2076 | * might want to do here some day... | |
2077 | */ | |
2078 | ||
2079 | return NOTIFY_DONE; | |
2080 | } | |
2081 | ||
2082 | static struct notifier_block drbd_notifier = { | |
2083 | .notifier_call = drbd_notify_sys, | |
2084 | }; | |
2085 | ||
2086 | static void drbd_release_ee_lists(struct drbd_conf *mdev) | |
2087 | { | |
2088 | int rr; | |
2089 | ||
2090 | rr = drbd_release_ee(mdev, &mdev->active_ee); | |
2091 | if (rr) | |
2092 | dev_err(DEV, "%d EEs in active list found!\n", rr); | |
2093 | ||
2094 | rr = drbd_release_ee(mdev, &mdev->sync_ee); | |
2095 | if (rr) | |
2096 | dev_err(DEV, "%d EEs in sync list found!\n", rr); | |
2097 | ||
2098 | rr = drbd_release_ee(mdev, &mdev->read_ee); | |
2099 | if (rr) | |
2100 | dev_err(DEV, "%d EEs in read list found!\n", rr); | |
2101 | ||
2102 | rr = drbd_release_ee(mdev, &mdev->done_ee); | |
2103 | if (rr) | |
2104 | dev_err(DEV, "%d EEs in done list found!\n", rr); | |
2105 | ||
2106 | rr = drbd_release_ee(mdev, &mdev->net_ee); | |
2107 | if (rr) | |
2108 | dev_err(DEV, "%d EEs in net list found!\n", rr); | |
2109 | } | |
2110 | ||
774b3055 PR |
2111 | /* caution. no locking. */ |
2112 | void drbd_delete_device(unsigned int minor) | |
b411b363 PR |
2113 | { |
2114 | struct drbd_conf *mdev = minor_to_mdev(minor); | |
2115 | ||
2116 | if (!mdev) | |
2117 | return; | |
2118 | ||
569083c0 LE |
2119 | idr_remove(&mdev->tconn->volumes, mdev->vnr); |
2120 | idr_remove(&minors, minor); | |
2121 | synchronize_rcu(); | |
774b3055 | 2122 | |
b411b363 | 2123 | /* paranoia asserts */ |
70dc65e1 | 2124 | D_ASSERT(mdev->open_cnt == 0); |
e42325a5 | 2125 | D_ASSERT(list_empty(&mdev->tconn->data.work.q)); |
b411b363 PR |
2126 | /* end paranoia asserts */ |
2127 | ||
2128 | del_gendisk(mdev->vdisk); | |
2129 | ||
2130 | /* cleanup stuff that may have been allocated during | |
2131 | * device (re-)configuration or state changes */ | |
2132 | ||
2133 | if (mdev->this_bdev) | |
2134 | bdput(mdev->this_bdev); | |
2135 | ||
2136 | drbd_free_resources(mdev); | |
2137 | ||
2138 | drbd_release_ee_lists(mdev); | |
2139 | ||
b411b363 PR |
2140 | lc_destroy(mdev->act_log); |
2141 | lc_destroy(mdev->resync); | |
2142 | ||
2143 | kfree(mdev->p_uuid); | |
2144 | /* mdev->p_uuid = NULL; */ | |
2145 | ||
b411b363 PR |
2146 | /* cleanup the rest that has been |
2147 | * allocated from drbd_new_device | |
2148 | * and actually free the mdev itself */ | |
2149 | drbd_free_mdev(mdev); | |
2150 | } | |
2151 | ||
2152 | static void drbd_cleanup(void) | |
2153 | { | |
2154 | unsigned int i; | |
81a5d60e | 2155 | struct drbd_conf *mdev; |
b411b363 PR |
2156 | |
2157 | unregister_reboot_notifier(&drbd_notifier); | |
2158 | ||
17a93f30 LE |
2159 | /* first remove proc, |
2160 | * drbdsetup uses it's presence to detect | |
2161 | * whether DRBD is loaded. | |
2162 | * If we would get stuck in proc removal, | |
2163 | * but have netlink already deregistered, | |
2164 | * some drbdsetup commands may wait forever | |
2165 | * for an answer. | |
2166 | */ | |
2167 | if (drbd_proc) | |
2168 | remove_proc_entry("drbd", NULL); | |
2169 | ||
3b98c0c2 | 2170 | drbd_genl_unregister(); |
b411b363 | 2171 | |
81a5d60e PR |
2172 | idr_for_each_entry(&minors, mdev, i) |
2173 | drbd_delete_device(i); | |
2174 | drbd_destroy_mempools(); | |
b411b363 PR |
2175 | unregister_blkdev(DRBD_MAJOR, "drbd"); |
2176 | ||
81a5d60e PR |
2177 | idr_destroy(&minors); |
2178 | ||
b411b363 PR |
2179 | printk(KERN_INFO "drbd: module cleanup done.\n"); |
2180 | } | |
2181 | ||
2182 | /** | |
2183 | * drbd_congested() - Callback for pdflush | |
2184 | * @congested_data: User data | |
2185 | * @bdi_bits: Bits pdflush is currently interested in | |
2186 | * | |
2187 | * Returns 1<<BDI_async_congested and/or 1<<BDI_sync_congested if we are congested. | |
2188 | */ | |
2189 | static int drbd_congested(void *congested_data, int bdi_bits) | |
2190 | { | |
2191 | struct drbd_conf *mdev = congested_data; | |
2192 | struct request_queue *q; | |
2193 | char reason = '-'; | |
2194 | int r = 0; | |
2195 | ||
1b881ef7 | 2196 | if (!may_inc_ap_bio(mdev)) { |
b411b363 PR |
2197 | /* DRBD has frozen IO */ |
2198 | r = bdi_bits; | |
2199 | reason = 'd'; | |
2200 | goto out; | |
2201 | } | |
2202 | ||
2203 | if (get_ldev(mdev)) { | |
2204 | q = bdev_get_queue(mdev->ldev->backing_bdev); | |
2205 | r = bdi_congested(&q->backing_dev_info, bdi_bits); | |
2206 | put_ldev(mdev); | |
2207 | if (r) | |
2208 | reason = 'b'; | |
2209 | } | |
2210 | ||
01a311a5 | 2211 | if (bdi_bits & (1 << BDI_async_congested) && test_bit(NET_CONGESTED, &mdev->tconn->flags)) { |
b411b363 PR |
2212 | r |= (1 << BDI_async_congested); |
2213 | reason = reason == 'b' ? 'a' : 'n'; | |
2214 | } | |
2215 | ||
2216 | out: | |
2217 | mdev->congestion_reason = reason; | |
2218 | return r; | |
2219 | } | |
2220 | ||
6699b655 PR |
2221 | static void drbd_init_workqueue(struct drbd_work_queue* wq) |
2222 | { | |
2223 | sema_init(&wq->s, 0); | |
2224 | spin_lock_init(&wq->q_lock); | |
2225 | INIT_LIST_HEAD(&wq->q); | |
2226 | } | |
2227 | ||
1aba4d7f PR |
2228 | struct drbd_tconn *conn_by_name(const char *name) |
2229 | { | |
2230 | struct drbd_tconn *tconn; | |
2231 | ||
3b98c0c2 LE |
2232 | if (!name || !name[0]) |
2233 | return NULL; | |
2234 | ||
543cc10b | 2235 | mutex_lock(&drbd_cfg_mutex); |
1aba4d7f PR |
2236 | list_for_each_entry(tconn, &drbd_tconns, all_tconn) { |
2237 | if (!strcmp(tconn->name, name)) | |
2238 | goto found; | |
2239 | } | |
2240 | tconn = NULL; | |
2241 | found: | |
543cc10b | 2242 | mutex_unlock(&drbd_cfg_mutex); |
1aba4d7f PR |
2243 | return tconn; |
2244 | } | |
2245 | ||
3b98c0c2 | 2246 | struct drbd_tconn *drbd_new_tconn(const char *name) |
2111438b PR |
2247 | { |
2248 | struct drbd_tconn *tconn; | |
2249 | ||
2250 | tconn = kzalloc(sizeof(struct drbd_tconn), GFP_KERNEL); | |
2251 | if (!tconn) | |
2252 | return NULL; | |
2253 | ||
2254 | tconn->name = kstrdup(name, GFP_KERNEL); | |
2255 | if (!tconn->name) | |
2256 | goto fail; | |
2257 | ||
774b3055 PR |
2258 | if (!zalloc_cpumask_var(&tconn->cpu_mask, GFP_KERNEL)) |
2259 | goto fail; | |
2260 | ||
2f5cdd0b PR |
2261 | if (!tl_init(tconn)) |
2262 | goto fail; | |
2263 | ||
bbeb641c | 2264 | tconn->cstate = C_STANDALONE; |
8410da8f | 2265 | mutex_init(&tconn->cstate_mutex); |
6699b655 | 2266 | spin_lock_init(&tconn->req_lock); |
b2fb6dbe PR |
2267 | atomic_set(&tconn->net_cnt, 0); |
2268 | init_waitqueue_head(&tconn->net_cnt_wait); | |
2a67d8b9 | 2269 | init_waitqueue_head(&tconn->ping_wait); |
062e879c | 2270 | idr_init(&tconn->volumes); |
b2fb6dbe | 2271 | |
6699b655 PR |
2272 | drbd_init_workqueue(&tconn->data.work); |
2273 | mutex_init(&tconn->data.mutex); | |
2274 | ||
2275 | drbd_init_workqueue(&tconn->meta.work); | |
2276 | mutex_init(&tconn->meta.mutex); | |
2277 | ||
392c8801 PR |
2278 | drbd_thread_init(tconn, &tconn->receiver, drbdd_init, "receiver"); |
2279 | drbd_thread_init(tconn, &tconn->worker, drbd_worker, "worker"); | |
2280 | drbd_thread_init(tconn, &tconn->asender, drbd_asender, "asender"); | |
2281 | ||
f399002e LE |
2282 | tconn->res_opts = (struct res_opts) { |
2283 | {}, 0, /* cpu_mask */ | |
2284 | DRBD_ON_NO_DATA_DEF, /* on_no_data */ | |
2285 | }; | |
2286 | ||
543cc10b LE |
2287 | mutex_lock(&drbd_cfg_mutex); |
2288 | list_add_tail(&tconn->all_tconn, &drbd_tconns); | |
2289 | mutex_unlock(&drbd_cfg_mutex); | |
2111438b PR |
2290 | |
2291 | return tconn; | |
2292 | ||
2293 | fail: | |
2f5cdd0b | 2294 | tl_cleanup(tconn); |
774b3055 | 2295 | free_cpumask_var(tconn->cpu_mask); |
2111438b PR |
2296 | kfree(tconn->name); |
2297 | kfree(tconn); | |
2298 | ||
2299 | return NULL; | |
2300 | } | |
2301 | ||
2302 | void drbd_free_tconn(struct drbd_tconn *tconn) | |
2303 | { | |
2111438b | 2304 | list_del(&tconn->all_tconn); |
062e879c | 2305 | idr_destroy(&tconn->volumes); |
2111438b | 2306 | |
774b3055 | 2307 | free_cpumask_var(tconn->cpu_mask); |
2111438b | 2308 | kfree(tconn->name); |
b42a70ad PR |
2309 | kfree(tconn->int_dig_out); |
2310 | kfree(tconn->int_dig_in); | |
2311 | kfree(tconn->int_dig_vv); | |
2111438b PR |
2312 | kfree(tconn); |
2313 | } | |
2314 | ||
774b3055 | 2315 | enum drbd_ret_code conn_new_minor(struct drbd_tconn *tconn, unsigned int minor, int vnr) |
b411b363 PR |
2316 | { |
2317 | struct drbd_conf *mdev; | |
2318 | struct gendisk *disk; | |
2319 | struct request_queue *q; | |
774b3055 | 2320 | int vnr_got = vnr; |
81a5d60e | 2321 | int minor_got = minor; |
8432b314 | 2322 | enum drbd_ret_code err = ERR_NOMEM; |
774b3055 PR |
2323 | |
2324 | mdev = minor_to_mdev(minor); | |
2325 | if (mdev) | |
2326 | return ERR_MINOR_EXISTS; | |
b411b363 PR |
2327 | |
2328 | /* GFP_KERNEL, we are outside of all write-out paths */ | |
2329 | mdev = kzalloc(sizeof(struct drbd_conf), GFP_KERNEL); | |
2330 | if (!mdev) | |
774b3055 PR |
2331 | return ERR_NOMEM; |
2332 | ||
2333 | mdev->tconn = tconn; | |
b411b363 | 2334 | mdev->minor = minor; |
3b98c0c2 | 2335 | mdev->vnr = vnr; |
b411b363 PR |
2336 | |
2337 | drbd_init_set_defaults(mdev); | |
2338 | ||
2339 | q = blk_alloc_queue(GFP_KERNEL); | |
2340 | if (!q) | |
2341 | goto out_no_q; | |
2342 | mdev->rq_queue = q; | |
2343 | q->queuedata = mdev; | |
b411b363 PR |
2344 | |
2345 | disk = alloc_disk(1); | |
2346 | if (!disk) | |
2347 | goto out_no_disk; | |
2348 | mdev->vdisk = disk; | |
2349 | ||
81e84650 | 2350 | set_disk_ro(disk, true); |
b411b363 PR |
2351 | |
2352 | disk->queue = q; | |
2353 | disk->major = DRBD_MAJOR; | |
2354 | disk->first_minor = minor; | |
2355 | disk->fops = &drbd_ops; | |
2356 | sprintf(disk->disk_name, "drbd%d", minor); | |
2357 | disk->private_data = mdev; | |
2358 | ||
2359 | mdev->this_bdev = bdget(MKDEV(DRBD_MAJOR, minor)); | |
2360 | /* we have no partitions. we contain only ourselves. */ | |
2361 | mdev->this_bdev->bd_contains = mdev->this_bdev; | |
2362 | ||
2363 | q->backing_dev_info.congested_fn = drbd_congested; | |
2364 | q->backing_dev_info.congested_data = mdev; | |
2365 | ||
2f58dcfc | 2366 | blk_queue_make_request(q, drbd_make_request); |
99432fcc PR |
2367 | /* Setting the max_hw_sectors to an odd value of 8kibyte here |
2368 | This triggers a max_bio_size message upon first attach or connect */ | |
2369 | blk_queue_max_hw_sectors(q, DRBD_MAX_BIO_SIZE_SAFE >> 8); | |
b411b363 PR |
2370 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
2371 | blk_queue_merge_bvec(q, drbd_merge_bvec); | |
87eeee41 | 2372 | q->queue_lock = &mdev->tconn->req_lock; /* needed since we use */ |
b411b363 PR |
2373 | |
2374 | mdev->md_io_page = alloc_page(GFP_KERNEL); | |
2375 | if (!mdev->md_io_page) | |
2376 | goto out_no_io_page; | |
2377 | ||
2378 | if (drbd_bm_init(mdev)) | |
2379 | goto out_no_bitmap; | |
dac1389c | 2380 | mdev->read_requests = RB_ROOT; |
de696716 | 2381 | mdev->write_requests = RB_ROOT; |
b411b363 | 2382 | |
b411b363 PR |
2383 | mdev->current_epoch = kzalloc(sizeof(struct drbd_epoch), GFP_KERNEL); |
2384 | if (!mdev->current_epoch) | |
2385 | goto out_no_epoch; | |
2386 | ||
2387 | INIT_LIST_HEAD(&mdev->current_epoch->list); | |
2388 | mdev->epochs = 1; | |
2389 | ||
81a5d60e | 2390 | if (!idr_pre_get(&minors, GFP_KERNEL)) |
8432b314 LE |
2391 | goto out_no_minor_idr; |
2392 | if (idr_get_new_above(&minors, mdev, minor, &minor_got)) | |
2393 | goto out_no_minor_idr; | |
81a5d60e | 2394 | if (minor_got != minor) { |
8432b314 LE |
2395 | err = ERR_MINOR_EXISTS; |
2396 | drbd_msg_put_info("requested minor exists already"); | |
569083c0 | 2397 | goto out_idr_remove_minor; |
81a5d60e | 2398 | } |
8432b314 LE |
2399 | |
2400 | if (!idr_pre_get(&tconn->volumes, GFP_KERNEL)) | |
2401 | goto out_idr_remove_minor; | |
2402 | if (idr_get_new_above(&tconn->volumes, mdev, vnr, &vnr_got)) | |
2403 | goto out_idr_remove_minor; | |
2404 | if (vnr_got != vnr) { | |
2405 | err = ERR_INVALID_REQUEST; | |
2406 | drbd_msg_put_info("requested volume exists already"); | |
2407 | goto out_idr_remove_vol; | |
2408 | } | |
774b3055 PR |
2409 | add_disk(disk); |
2410 | ||
2325eb66 PR |
2411 | /* inherit the connection state */ |
2412 | mdev->state.conn = tconn->cstate; | |
2413 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
2414 | drbd_connected(vnr, mdev, tconn); | |
2415 | ||
774b3055 | 2416 | return NO_ERROR; |
b411b363 | 2417 | |
569083c0 LE |
2418 | out_idr_remove_vol: |
2419 | idr_remove(&tconn->volumes, vnr_got); | |
8432b314 LE |
2420 | out_idr_remove_minor: |
2421 | idr_remove(&minors, minor_got); | |
569083c0 | 2422 | synchronize_rcu(); |
8432b314 | 2423 | out_no_minor_idr: |
81a5d60e | 2424 | kfree(mdev->current_epoch); |
b411b363 | 2425 | out_no_epoch: |
b411b363 PR |
2426 | drbd_bm_cleanup(mdev); |
2427 | out_no_bitmap: | |
2428 | __free_page(mdev->md_io_page); | |
2429 | out_no_io_page: | |
2430 | put_disk(disk); | |
2431 | out_no_disk: | |
2432 | blk_cleanup_queue(q); | |
2433 | out_no_q: | |
b411b363 | 2434 | kfree(mdev); |
8432b314 | 2435 | return err; |
b411b363 PR |
2436 | } |
2437 | ||
2438 | /* counterpart of drbd_new_device. | |
2439 | * last part of drbd_delete_device. */ | |
2440 | void drbd_free_mdev(struct drbd_conf *mdev) | |
2441 | { | |
2442 | kfree(mdev->current_epoch); | |
b411b363 PR |
2443 | if (mdev->bitmap) /* should no longer be there. */ |
2444 | drbd_bm_cleanup(mdev); | |
2445 | __free_page(mdev->md_io_page); | |
2446 | put_disk(mdev->vdisk); | |
2447 | blk_cleanup_queue(mdev->rq_queue); | |
b411b363 PR |
2448 | kfree(mdev); |
2449 | } | |
2450 | ||
2451 | ||
2452 | int __init drbd_init(void) | |
2453 | { | |
2454 | int err; | |
2455 | ||
fd340c12 PR |
2456 | BUILD_BUG_ON(sizeof(struct p_header80) != sizeof(struct p_header95)); |
2457 | BUILD_BUG_ON(sizeof(struct p_handshake) != 80); | |
b411b363 | 2458 | |
2b8a90b5 | 2459 | if (minor_count < DRBD_MINOR_COUNT_MIN || minor_count > DRBD_MINOR_COUNT_MAX) { |
b411b363 | 2460 | printk(KERN_ERR |
81a5d60e | 2461 | "drbd: invalid minor_count (%d)\n", minor_count); |
b411b363 PR |
2462 | #ifdef MODULE |
2463 | return -EINVAL; | |
2464 | #else | |
2465 | minor_count = 8; | |
2466 | #endif | |
2467 | } | |
2468 | ||
b411b363 PR |
2469 | err = register_blkdev(DRBD_MAJOR, "drbd"); |
2470 | if (err) { | |
2471 | printk(KERN_ERR | |
2472 | "drbd: unable to register block device major %d\n", | |
2473 | DRBD_MAJOR); | |
2474 | return err; | |
2475 | } | |
2476 | ||
3b98c0c2 LE |
2477 | err = drbd_genl_register(); |
2478 | if (err) { | |
2479 | printk(KERN_ERR "drbd: unable to register generic netlink family\n"); | |
2480 | goto fail; | |
2481 | } | |
2482 | ||
2483 | ||
b411b363 PR |
2484 | register_reboot_notifier(&drbd_notifier); |
2485 | ||
2486 | /* | |
2487 | * allocate all necessary structs | |
2488 | */ | |
2489 | err = -ENOMEM; | |
2490 | ||
2491 | init_waitqueue_head(&drbd_pp_wait); | |
2492 | ||
2493 | drbd_proc = NULL; /* play safe for drbd_cleanup */ | |
81a5d60e | 2494 | idr_init(&minors); |
b411b363 PR |
2495 | |
2496 | err = drbd_create_mempools(); | |
2497 | if (err) | |
3b98c0c2 | 2498 | goto fail; |
b411b363 | 2499 | |
8c484ee4 | 2500 | drbd_proc = proc_create_data("drbd", S_IFREG | S_IRUGO , NULL, &drbd_proc_fops, NULL); |
b411b363 PR |
2501 | if (!drbd_proc) { |
2502 | printk(KERN_ERR "drbd: unable to register proc file\n"); | |
3b98c0c2 | 2503 | goto fail; |
b411b363 PR |
2504 | } |
2505 | ||
2506 | rwlock_init(&global_state_lock); | |
2111438b | 2507 | INIT_LIST_HEAD(&drbd_tconns); |
b411b363 PR |
2508 | |
2509 | printk(KERN_INFO "drbd: initialized. " | |
2510 | "Version: " REL_VERSION " (api:%d/proto:%d-%d)\n", | |
2511 | API_VERSION, PRO_VERSION_MIN, PRO_VERSION_MAX); | |
2512 | printk(KERN_INFO "drbd: %s\n", drbd_buildtag()); | |
2513 | printk(KERN_INFO "drbd: registered as block device major %d\n", | |
2514 | DRBD_MAJOR); | |
b411b363 PR |
2515 | |
2516 | return 0; /* Success! */ | |
2517 | ||
3b98c0c2 | 2518 | fail: |
b411b363 PR |
2519 | drbd_cleanup(); |
2520 | if (err == -ENOMEM) | |
2521 | /* currently always the case */ | |
2522 | printk(KERN_ERR "drbd: ran out of memory\n"); | |
2523 | else | |
2524 | printk(KERN_ERR "drbd: initialization failure\n"); | |
2525 | return err; | |
2526 | } | |
2527 | ||
2528 | void drbd_free_bc(struct drbd_backing_dev *ldev) | |
2529 | { | |
2530 | if (ldev == NULL) | |
2531 | return; | |
2532 | ||
e525fd89 TH |
2533 | blkdev_put(ldev->backing_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
2534 | blkdev_put(ldev->md_bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); | |
b411b363 PR |
2535 | |
2536 | kfree(ldev); | |
2537 | } | |
2538 | ||
360cc740 PR |
2539 | void drbd_free_sock(struct drbd_tconn *tconn) |
2540 | { | |
2541 | if (tconn->data.socket) { | |
2542 | mutex_lock(&tconn->data.mutex); | |
2543 | kernel_sock_shutdown(tconn->data.socket, SHUT_RDWR); | |
2544 | sock_release(tconn->data.socket); | |
2545 | tconn->data.socket = NULL; | |
2546 | mutex_unlock(&tconn->data.mutex); | |
b411b363 | 2547 | } |
360cc740 PR |
2548 | if (tconn->meta.socket) { |
2549 | mutex_lock(&tconn->meta.mutex); | |
2550 | kernel_sock_shutdown(tconn->meta.socket, SHUT_RDWR); | |
2551 | sock_release(tconn->meta.socket); | |
2552 | tconn->meta.socket = NULL; | |
2553 | mutex_unlock(&tconn->meta.mutex); | |
b411b363 PR |
2554 | } |
2555 | } | |
2556 | ||
2557 | ||
2558 | void drbd_free_resources(struct drbd_conf *mdev) | |
2559 | { | |
f399002e LE |
2560 | crypto_free_hash(mdev->tconn->csums_tfm); |
2561 | mdev->tconn->csums_tfm = NULL; | |
2562 | crypto_free_hash(mdev->tconn->verify_tfm); | |
2563 | mdev->tconn->verify_tfm = NULL; | |
a0638456 PR |
2564 | crypto_free_hash(mdev->tconn->cram_hmac_tfm); |
2565 | mdev->tconn->cram_hmac_tfm = NULL; | |
2566 | crypto_free_hash(mdev->tconn->integrity_w_tfm); | |
2567 | mdev->tconn->integrity_w_tfm = NULL; | |
2568 | crypto_free_hash(mdev->tconn->integrity_r_tfm); | |
2569 | mdev->tconn->integrity_r_tfm = NULL; | |
b411b363 | 2570 | |
360cc740 | 2571 | drbd_free_sock(mdev->tconn); |
b411b363 PR |
2572 | |
2573 | __no_warn(local, | |
2574 | drbd_free_bc(mdev->ldev); | |
2575 | mdev->ldev = NULL;); | |
2576 | } | |
2577 | ||
2578 | /* meta data management */ | |
2579 | ||
2580 | struct meta_data_on_disk { | |
2581 | u64 la_size; /* last agreed size. */ | |
2582 | u64 uuid[UI_SIZE]; /* UUIDs. */ | |
2583 | u64 device_uuid; | |
2584 | u64 reserved_u64_1; | |
2585 | u32 flags; /* MDF */ | |
2586 | u32 magic; | |
2587 | u32 md_size_sect; | |
2588 | u32 al_offset; /* offset to this block */ | |
2589 | u32 al_nr_extents; /* important for restoring the AL */ | |
f399002e | 2590 | /* `-- act_log->nr_elements <-- ldev->dc.al_extents */ |
b411b363 PR |
2591 | u32 bm_offset; /* offset to the bitmap, from here */ |
2592 | u32 bm_bytes_per_bit; /* BM_BLOCK_SIZE */ | |
99432fcc PR |
2593 | u32 la_peer_max_bio_size; /* last peer max_bio_size */ |
2594 | u32 reserved_u32[3]; | |
b411b363 PR |
2595 | |
2596 | } __packed; | |
2597 | ||
2598 | /** | |
2599 | * drbd_md_sync() - Writes the meta data super block if the MD_DIRTY flag bit is set | |
2600 | * @mdev: DRBD device. | |
2601 | */ | |
2602 | void drbd_md_sync(struct drbd_conf *mdev) | |
2603 | { | |
2604 | struct meta_data_on_disk *buffer; | |
2605 | sector_t sector; | |
2606 | int i; | |
2607 | ||
ee15b038 LE |
2608 | del_timer(&mdev->md_sync_timer); |
2609 | /* timer may be rearmed by drbd_md_mark_dirty() now. */ | |
b411b363 PR |
2610 | if (!test_and_clear_bit(MD_DIRTY, &mdev->flags)) |
2611 | return; | |
b411b363 PR |
2612 | |
2613 | /* We use here D_FAILED and not D_ATTACHING because we try to write | |
2614 | * metadata even if we detach due to a disk failure! */ | |
2615 | if (!get_ldev_if_state(mdev, D_FAILED)) | |
2616 | return; | |
2617 | ||
b411b363 PR |
2618 | mutex_lock(&mdev->md_io_mutex); |
2619 | buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page); | |
2620 | memset(buffer, 0, 512); | |
2621 | ||
2622 | buffer->la_size = cpu_to_be64(drbd_get_capacity(mdev->this_bdev)); | |
2623 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
2624 | buffer->uuid[i] = cpu_to_be64(mdev->ldev->md.uuid[i]); | |
2625 | buffer->flags = cpu_to_be32(mdev->ldev->md.flags); | |
2626 | buffer->magic = cpu_to_be32(DRBD_MD_MAGIC); | |
2627 | ||
2628 | buffer->md_size_sect = cpu_to_be32(mdev->ldev->md.md_size_sect); | |
2629 | buffer->al_offset = cpu_to_be32(mdev->ldev->md.al_offset); | |
2630 | buffer->al_nr_extents = cpu_to_be32(mdev->act_log->nr_elements); | |
2631 | buffer->bm_bytes_per_bit = cpu_to_be32(BM_BLOCK_SIZE); | |
2632 | buffer->device_uuid = cpu_to_be64(mdev->ldev->md.device_uuid); | |
2633 | ||
2634 | buffer->bm_offset = cpu_to_be32(mdev->ldev->md.bm_offset); | |
99432fcc | 2635 | buffer->la_peer_max_bio_size = cpu_to_be32(mdev->peer_max_bio_size); |
b411b363 PR |
2636 | |
2637 | D_ASSERT(drbd_md_ss__(mdev, mdev->ldev) == mdev->ldev->md.md_offset); | |
2638 | sector = mdev->ldev->md.md_offset; | |
2639 | ||
3f3a9b84 | 2640 | if (!drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) { |
b411b363 PR |
2641 | /* this was a try anyways ... */ |
2642 | dev_err(DEV, "meta data update failed!\n"); | |
81e84650 | 2643 | drbd_chk_io_error(mdev, 1, true); |
b411b363 PR |
2644 | } |
2645 | ||
2646 | /* Update mdev->ldev->md.la_size_sect, | |
2647 | * since we updated it on metadata. */ | |
2648 | mdev->ldev->md.la_size_sect = drbd_get_capacity(mdev->this_bdev); | |
2649 | ||
2650 | mutex_unlock(&mdev->md_io_mutex); | |
2651 | put_ldev(mdev); | |
2652 | } | |
2653 | ||
2654 | /** | |
2655 | * drbd_md_read() - Reads in the meta data super block | |
2656 | * @mdev: DRBD device. | |
2657 | * @bdev: Device from which the meta data should be read in. | |
2658 | * | |
116676ca | 2659 | * Return 0 (NO_ERROR) on success, and an enum drbd_ret_code in case |
b411b363 PR |
2660 | * something goes wrong. Currently only: ERR_IO_MD_DISK, ERR_MD_INVALID. |
2661 | */ | |
2662 | int drbd_md_read(struct drbd_conf *mdev, struct drbd_backing_dev *bdev) | |
2663 | { | |
2664 | struct meta_data_on_disk *buffer; | |
2665 | int i, rv = NO_ERROR; | |
2666 | ||
2667 | if (!get_ldev_if_state(mdev, D_ATTACHING)) | |
2668 | return ERR_IO_MD_DISK; | |
2669 | ||
b411b363 PR |
2670 | mutex_lock(&mdev->md_io_mutex); |
2671 | buffer = (struct meta_data_on_disk *)page_address(mdev->md_io_page); | |
2672 | ||
2673 | if (!drbd_md_sync_page_io(mdev, bdev, bdev->md.md_offset, READ)) { | |
25985edc | 2674 | /* NOTE: can't do normal error processing here as this is |
b411b363 PR |
2675 | called BEFORE disk is attached */ |
2676 | dev_err(DEV, "Error while reading metadata.\n"); | |
2677 | rv = ERR_IO_MD_DISK; | |
2678 | goto err; | |
2679 | } | |
2680 | ||
e7fad8af | 2681 | if (buffer->magic != cpu_to_be32(DRBD_MD_MAGIC)) { |
b411b363 PR |
2682 | dev_err(DEV, "Error while reading metadata, magic not found.\n"); |
2683 | rv = ERR_MD_INVALID; | |
2684 | goto err; | |
2685 | } | |
2686 | if (be32_to_cpu(buffer->al_offset) != bdev->md.al_offset) { | |
2687 | dev_err(DEV, "unexpected al_offset: %d (expected %d)\n", | |
2688 | be32_to_cpu(buffer->al_offset), bdev->md.al_offset); | |
2689 | rv = ERR_MD_INVALID; | |
2690 | goto err; | |
2691 | } | |
2692 | if (be32_to_cpu(buffer->bm_offset) != bdev->md.bm_offset) { | |
2693 | dev_err(DEV, "unexpected bm_offset: %d (expected %d)\n", | |
2694 | be32_to_cpu(buffer->bm_offset), bdev->md.bm_offset); | |
2695 | rv = ERR_MD_INVALID; | |
2696 | goto err; | |
2697 | } | |
2698 | if (be32_to_cpu(buffer->md_size_sect) != bdev->md.md_size_sect) { | |
2699 | dev_err(DEV, "unexpected md_size: %u (expected %u)\n", | |
2700 | be32_to_cpu(buffer->md_size_sect), bdev->md.md_size_sect); | |
2701 | rv = ERR_MD_INVALID; | |
2702 | goto err; | |
2703 | } | |
2704 | ||
2705 | if (be32_to_cpu(buffer->bm_bytes_per_bit) != BM_BLOCK_SIZE) { | |
2706 | dev_err(DEV, "unexpected bm_bytes_per_bit: %u (expected %u)\n", | |
2707 | be32_to_cpu(buffer->bm_bytes_per_bit), BM_BLOCK_SIZE); | |
2708 | rv = ERR_MD_INVALID; | |
2709 | goto err; | |
2710 | } | |
2711 | ||
2712 | bdev->md.la_size_sect = be64_to_cpu(buffer->la_size); | |
2713 | for (i = UI_CURRENT; i < UI_SIZE; i++) | |
2714 | bdev->md.uuid[i] = be64_to_cpu(buffer->uuid[i]); | |
2715 | bdev->md.flags = be32_to_cpu(buffer->flags); | |
f399002e | 2716 | bdev->dc.al_extents = be32_to_cpu(buffer->al_nr_extents); |
b411b363 PR |
2717 | bdev->md.device_uuid = be64_to_cpu(buffer->device_uuid); |
2718 | ||
87eeee41 | 2719 | spin_lock_irq(&mdev->tconn->req_lock); |
99432fcc PR |
2720 | if (mdev->state.conn < C_CONNECTED) { |
2721 | int peer; | |
2722 | peer = be32_to_cpu(buffer->la_peer_max_bio_size); | |
2723 | peer = max_t(int, peer, DRBD_MAX_BIO_SIZE_SAFE); | |
2724 | mdev->peer_max_bio_size = peer; | |
2725 | } | |
87eeee41 | 2726 | spin_unlock_irq(&mdev->tconn->req_lock); |
99432fcc | 2727 | |
f399002e LE |
2728 | if (bdev->dc.al_extents < 7) |
2729 | bdev->dc.al_extents = 127; | |
b411b363 PR |
2730 | |
2731 | err: | |
2732 | mutex_unlock(&mdev->md_io_mutex); | |
2733 | put_ldev(mdev); | |
2734 | ||
2735 | return rv; | |
2736 | } | |
2737 | ||
2738 | /** | |
2739 | * drbd_md_mark_dirty() - Mark meta data super block as dirty | |
2740 | * @mdev: DRBD device. | |
2741 | * | |
2742 | * Call this function if you change anything that should be written to | |
2743 | * the meta-data super block. This function sets MD_DIRTY, and starts a | |
2744 | * timer that ensures that within five seconds you have to call drbd_md_sync(). | |
2745 | */ | |
ca0e6098 | 2746 | #ifdef DEBUG |
ee15b038 LE |
2747 | void drbd_md_mark_dirty_(struct drbd_conf *mdev, unsigned int line, const char *func) |
2748 | { | |
2749 | if (!test_and_set_bit(MD_DIRTY, &mdev->flags)) { | |
2750 | mod_timer(&mdev->md_sync_timer, jiffies + HZ); | |
2751 | mdev->last_md_mark_dirty.line = line; | |
2752 | mdev->last_md_mark_dirty.func = func; | |
2753 | } | |
2754 | } | |
2755 | #else | |
b411b363 PR |
2756 | void drbd_md_mark_dirty(struct drbd_conf *mdev) |
2757 | { | |
ee15b038 | 2758 | if (!test_and_set_bit(MD_DIRTY, &mdev->flags)) |
ca0e6098 | 2759 | mod_timer(&mdev->md_sync_timer, jiffies + 5*HZ); |
b411b363 | 2760 | } |
ee15b038 | 2761 | #endif |
b411b363 PR |
2762 | |
2763 | static void drbd_uuid_move_history(struct drbd_conf *mdev) __must_hold(local) | |
2764 | { | |
2765 | int i; | |
2766 | ||
62b0da3a | 2767 | for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++) |
b411b363 | 2768 | mdev->ldev->md.uuid[i+1] = mdev->ldev->md.uuid[i]; |
b411b363 PR |
2769 | } |
2770 | ||
2771 | void _drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local) | |
2772 | { | |
2773 | if (idx == UI_CURRENT) { | |
2774 | if (mdev->state.role == R_PRIMARY) | |
2775 | val |= 1; | |
2776 | else | |
2777 | val &= ~((u64)1); | |
2778 | ||
2779 | drbd_set_ed_uuid(mdev, val); | |
2780 | } | |
2781 | ||
2782 | mdev->ldev->md.uuid[idx] = val; | |
b411b363 PR |
2783 | drbd_md_mark_dirty(mdev); |
2784 | } | |
2785 | ||
2786 | ||
2787 | void drbd_uuid_set(struct drbd_conf *mdev, int idx, u64 val) __must_hold(local) | |
2788 | { | |
2789 | if (mdev->ldev->md.uuid[idx]) { | |
2790 | drbd_uuid_move_history(mdev); | |
2791 | mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[idx]; | |
b411b363 PR |
2792 | } |
2793 | _drbd_uuid_set(mdev, idx, val); | |
2794 | } | |
2795 | ||
2796 | /** | |
2797 | * drbd_uuid_new_current() - Creates a new current UUID | |
2798 | * @mdev: DRBD device. | |
2799 | * | |
2800 | * Creates a new current UUID, and rotates the old current UUID into | |
2801 | * the bitmap slot. Causes an incremental resync upon next connect. | |
2802 | */ | |
2803 | void drbd_uuid_new_current(struct drbd_conf *mdev) __must_hold(local) | |
2804 | { | |
2805 | u64 val; | |
62b0da3a LE |
2806 | unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP]; |
2807 | ||
2808 | if (bm_uuid) | |
2809 | dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid); | |
b411b363 | 2810 | |
b411b363 | 2811 | mdev->ldev->md.uuid[UI_BITMAP] = mdev->ldev->md.uuid[UI_CURRENT]; |
b411b363 PR |
2812 | |
2813 | get_random_bytes(&val, sizeof(u64)); | |
2814 | _drbd_uuid_set(mdev, UI_CURRENT, val); | |
62b0da3a | 2815 | drbd_print_uuids(mdev, "new current UUID"); |
aaa8e2b3 LE |
2816 | /* get it to stable storage _now_ */ |
2817 | drbd_md_sync(mdev); | |
b411b363 PR |
2818 | } |
2819 | ||
2820 | void drbd_uuid_set_bm(struct drbd_conf *mdev, u64 val) __must_hold(local) | |
2821 | { | |
2822 | if (mdev->ldev->md.uuid[UI_BITMAP] == 0 && val == 0) | |
2823 | return; | |
2824 | ||
2825 | if (val == 0) { | |
2826 | drbd_uuid_move_history(mdev); | |
2827 | mdev->ldev->md.uuid[UI_HISTORY_START] = mdev->ldev->md.uuid[UI_BITMAP]; | |
2828 | mdev->ldev->md.uuid[UI_BITMAP] = 0; | |
b411b363 | 2829 | } else { |
62b0da3a LE |
2830 | unsigned long long bm_uuid = mdev->ldev->md.uuid[UI_BITMAP]; |
2831 | if (bm_uuid) | |
2832 | dev_warn(DEV, "bm UUID was already set: %llX\n", bm_uuid); | |
b411b363 | 2833 | |
62b0da3a | 2834 | mdev->ldev->md.uuid[UI_BITMAP] = val & ~((u64)1); |
b411b363 PR |
2835 | } |
2836 | drbd_md_mark_dirty(mdev); | |
2837 | } | |
2838 | ||
2839 | /** | |
2840 | * drbd_bmio_set_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
2841 | * @mdev: DRBD device. | |
2842 | * | |
2843 | * Sets all bits in the bitmap and writes the whole bitmap to stable storage. | |
2844 | */ | |
2845 | int drbd_bmio_set_n_write(struct drbd_conf *mdev) | |
2846 | { | |
2847 | int rv = -EIO; | |
2848 | ||
2849 | if (get_ldev_if_state(mdev, D_ATTACHING)) { | |
2850 | drbd_md_set_flag(mdev, MDF_FULL_SYNC); | |
2851 | drbd_md_sync(mdev); | |
2852 | drbd_bm_set_all(mdev); | |
2853 | ||
2854 | rv = drbd_bm_write(mdev); | |
2855 | ||
2856 | if (!rv) { | |
2857 | drbd_md_clear_flag(mdev, MDF_FULL_SYNC); | |
2858 | drbd_md_sync(mdev); | |
2859 | } | |
2860 | ||
2861 | put_ldev(mdev); | |
2862 | } | |
2863 | ||
2864 | return rv; | |
2865 | } | |
2866 | ||
2867 | /** | |
2868 | * drbd_bmio_clear_n_write() - io_fn for drbd_queue_bitmap_io() or drbd_bitmap_io() | |
2869 | * @mdev: DRBD device. | |
2870 | * | |
2871 | * Clears all bits in the bitmap and writes the whole bitmap to stable storage. | |
2872 | */ | |
2873 | int drbd_bmio_clear_n_write(struct drbd_conf *mdev) | |
2874 | { | |
2875 | int rv = -EIO; | |
2876 | ||
0778286a | 2877 | drbd_resume_al(mdev); |
b411b363 PR |
2878 | if (get_ldev_if_state(mdev, D_ATTACHING)) { |
2879 | drbd_bm_clear_all(mdev); | |
2880 | rv = drbd_bm_write(mdev); | |
2881 | put_ldev(mdev); | |
2882 | } | |
2883 | ||
2884 | return rv; | |
2885 | } | |
2886 | ||
00d56944 | 2887 | static int w_bitmap_io(struct drbd_work *w, int unused) |
b411b363 PR |
2888 | { |
2889 | struct bm_io_work *work = container_of(w, struct bm_io_work, w); | |
00d56944 | 2890 | struct drbd_conf *mdev = w->mdev; |
02851e9f | 2891 | int rv = -EIO; |
b411b363 PR |
2892 | |
2893 | D_ASSERT(atomic_read(&mdev->ap_bio_cnt) == 0); | |
2894 | ||
02851e9f | 2895 | if (get_ldev(mdev)) { |
20ceb2b2 | 2896 | drbd_bm_lock(mdev, work->why, work->flags); |
02851e9f LE |
2897 | rv = work->io_fn(mdev); |
2898 | drbd_bm_unlock(mdev); | |
2899 | put_ldev(mdev); | |
2900 | } | |
b411b363 | 2901 | |
4738fa16 | 2902 | clear_bit_unlock(BITMAP_IO, &mdev->flags); |
b411b363 PR |
2903 | wake_up(&mdev->misc_wait); |
2904 | ||
2905 | if (work->done) | |
2906 | work->done(mdev, rv); | |
2907 | ||
2908 | clear_bit(BITMAP_IO_QUEUED, &mdev->flags); | |
2909 | work->why = NULL; | |
20ceb2b2 | 2910 | work->flags = 0; |
b411b363 PR |
2911 | |
2912 | return 1; | |
2913 | } | |
2914 | ||
82f59cc6 LE |
2915 | void drbd_ldev_destroy(struct drbd_conf *mdev) |
2916 | { | |
2917 | lc_destroy(mdev->resync); | |
2918 | mdev->resync = NULL; | |
2919 | lc_destroy(mdev->act_log); | |
2920 | mdev->act_log = NULL; | |
2921 | __no_warn(local, | |
2922 | drbd_free_bc(mdev->ldev); | |
2923 | mdev->ldev = NULL;); | |
2924 | ||
82f59cc6 LE |
2925 | clear_bit(GO_DISKLESS, &mdev->flags); |
2926 | } | |
2927 | ||
00d56944 | 2928 | static int w_go_diskless(struct drbd_work *w, int unused) |
e9e6f3ec | 2929 | { |
00d56944 PR |
2930 | struct drbd_conf *mdev = w->mdev; |
2931 | ||
e9e6f3ec | 2932 | D_ASSERT(mdev->state.disk == D_FAILED); |
9d282875 LE |
2933 | /* we cannot assert local_cnt == 0 here, as get_ldev_if_state will |
2934 | * inc/dec it frequently. Once we are D_DISKLESS, no one will touch | |
82f59cc6 LE |
2935 | * the protected members anymore, though, so once put_ldev reaches zero |
2936 | * again, it will be safe to free them. */ | |
e9e6f3ec | 2937 | drbd_force_state(mdev, NS(disk, D_DISKLESS)); |
e9e6f3ec LE |
2938 | return 1; |
2939 | } | |
2940 | ||
2941 | void drbd_go_diskless(struct drbd_conf *mdev) | |
2942 | { | |
2943 | D_ASSERT(mdev->state.disk == D_FAILED); | |
2944 | if (!test_and_set_bit(GO_DISKLESS, &mdev->flags)) | |
e42325a5 | 2945 | drbd_queue_work(&mdev->tconn->data.work, &mdev->go_diskless); |
e9e6f3ec LE |
2946 | } |
2947 | ||
b411b363 PR |
2948 | /** |
2949 | * drbd_queue_bitmap_io() - Queues an IO operation on the whole bitmap | |
2950 | * @mdev: DRBD device. | |
2951 | * @io_fn: IO callback to be called when bitmap IO is possible | |
2952 | * @done: callback to be called after the bitmap IO was performed | |
2953 | * @why: Descriptive text of the reason for doing the IO | |
2954 | * | |
2955 | * While IO on the bitmap happens we freeze application IO thus we ensure | |
2956 | * that drbd_set_out_of_sync() can not be called. This function MAY ONLY be | |
2957 | * called from worker context. It MUST NOT be used while a previous such | |
2958 | * work is still pending! | |
2959 | */ | |
2960 | void drbd_queue_bitmap_io(struct drbd_conf *mdev, | |
2961 | int (*io_fn)(struct drbd_conf *), | |
2962 | void (*done)(struct drbd_conf *, int), | |
20ceb2b2 | 2963 | char *why, enum bm_flag flags) |
b411b363 | 2964 | { |
e6b3ea83 | 2965 | D_ASSERT(current == mdev->tconn->worker.task); |
b411b363 PR |
2966 | |
2967 | D_ASSERT(!test_bit(BITMAP_IO_QUEUED, &mdev->flags)); | |
2968 | D_ASSERT(!test_bit(BITMAP_IO, &mdev->flags)); | |
2969 | D_ASSERT(list_empty(&mdev->bm_io_work.w.list)); | |
2970 | if (mdev->bm_io_work.why) | |
2971 | dev_err(DEV, "FIXME going to queue '%s' but '%s' still pending?\n", | |
2972 | why, mdev->bm_io_work.why); | |
2973 | ||
2974 | mdev->bm_io_work.io_fn = io_fn; | |
2975 | mdev->bm_io_work.done = done; | |
2976 | mdev->bm_io_work.why = why; | |
20ceb2b2 | 2977 | mdev->bm_io_work.flags = flags; |
b411b363 | 2978 | |
87eeee41 | 2979 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
2980 | set_bit(BITMAP_IO, &mdev->flags); |
2981 | if (atomic_read(&mdev->ap_bio_cnt) == 0) { | |
127b3178 | 2982 | if (!test_and_set_bit(BITMAP_IO_QUEUED, &mdev->flags)) |
e42325a5 | 2983 | drbd_queue_work(&mdev->tconn->data.work, &mdev->bm_io_work.w); |
b411b363 | 2984 | } |
87eeee41 | 2985 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
2986 | } |
2987 | ||
2988 | /** | |
2989 | * drbd_bitmap_io() - Does an IO operation on the whole bitmap | |
2990 | * @mdev: DRBD device. | |
2991 | * @io_fn: IO callback to be called when bitmap IO is possible | |
2992 | * @why: Descriptive text of the reason for doing the IO | |
2993 | * | |
2994 | * freezes application IO while that the actual IO operations runs. This | |
2995 | * functions MAY NOT be called from worker context. | |
2996 | */ | |
20ceb2b2 LE |
2997 | int drbd_bitmap_io(struct drbd_conf *mdev, int (*io_fn)(struct drbd_conf *), |
2998 | char *why, enum bm_flag flags) | |
b411b363 PR |
2999 | { |
3000 | int rv; | |
3001 | ||
e6b3ea83 | 3002 | D_ASSERT(current != mdev->tconn->worker.task); |
b411b363 | 3003 | |
20ceb2b2 LE |
3004 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
3005 | drbd_suspend_io(mdev); | |
b411b363 | 3006 | |
20ceb2b2 | 3007 | drbd_bm_lock(mdev, why, flags); |
b411b363 PR |
3008 | rv = io_fn(mdev); |
3009 | drbd_bm_unlock(mdev); | |
3010 | ||
20ceb2b2 LE |
3011 | if ((flags & BM_LOCKED_SET_ALLOWED) == 0) |
3012 | drbd_resume_io(mdev); | |
b411b363 PR |
3013 | |
3014 | return rv; | |
3015 | } | |
3016 | ||
3017 | void drbd_md_set_flag(struct drbd_conf *mdev, int flag) __must_hold(local) | |
3018 | { | |
3019 | if ((mdev->ldev->md.flags & flag) != flag) { | |
3020 | drbd_md_mark_dirty(mdev); | |
3021 | mdev->ldev->md.flags |= flag; | |
3022 | } | |
3023 | } | |
3024 | ||
3025 | void drbd_md_clear_flag(struct drbd_conf *mdev, int flag) __must_hold(local) | |
3026 | { | |
3027 | if ((mdev->ldev->md.flags & flag) != 0) { | |
3028 | drbd_md_mark_dirty(mdev); | |
3029 | mdev->ldev->md.flags &= ~flag; | |
3030 | } | |
3031 | } | |
3032 | int drbd_md_test_flag(struct drbd_backing_dev *bdev, int flag) | |
3033 | { | |
3034 | return (bdev->md.flags & flag) != 0; | |
3035 | } | |
3036 | ||
3037 | static void md_sync_timer_fn(unsigned long data) | |
3038 | { | |
3039 | struct drbd_conf *mdev = (struct drbd_conf *) data; | |
3040 | ||
e42325a5 | 3041 | drbd_queue_work_front(&mdev->tconn->data.work, &mdev->md_sync_work); |
b411b363 PR |
3042 | } |
3043 | ||
00d56944 | 3044 | static int w_md_sync(struct drbd_work *w, int unused) |
b411b363 | 3045 | { |
00d56944 PR |
3046 | struct drbd_conf *mdev = w->mdev; |
3047 | ||
b411b363 | 3048 | dev_warn(DEV, "md_sync_timer expired! Worker calls drbd_md_sync().\n"); |
ee15b038 LE |
3049 | #ifdef DEBUG |
3050 | dev_warn(DEV, "last md_mark_dirty: %s:%u\n", | |
3051 | mdev->last_md_mark_dirty.func, mdev->last_md_mark_dirty.line); | |
3052 | #endif | |
b411b363 | 3053 | drbd_md_sync(mdev); |
b411b363 PR |
3054 | return 1; |
3055 | } | |
3056 | ||
d8763023 | 3057 | const char *cmdname(enum drbd_packet cmd) |
f2ad9063 AG |
3058 | { |
3059 | /* THINK may need to become several global tables | |
3060 | * when we want to support more than | |
3061 | * one PRO_VERSION */ | |
3062 | static const char *cmdnames[] = { | |
3063 | [P_DATA] = "Data", | |
3064 | [P_DATA_REPLY] = "DataReply", | |
3065 | [P_RS_DATA_REPLY] = "RSDataReply", | |
3066 | [P_BARRIER] = "Barrier", | |
3067 | [P_BITMAP] = "ReportBitMap", | |
3068 | [P_BECOME_SYNC_TARGET] = "BecomeSyncTarget", | |
3069 | [P_BECOME_SYNC_SOURCE] = "BecomeSyncSource", | |
3070 | [P_UNPLUG_REMOTE] = "UnplugRemote", | |
3071 | [P_DATA_REQUEST] = "DataRequest", | |
3072 | [P_RS_DATA_REQUEST] = "RSDataRequest", | |
3073 | [P_SYNC_PARAM] = "SyncParam", | |
3074 | [P_SYNC_PARAM89] = "SyncParam89", | |
3075 | [P_PROTOCOL] = "ReportProtocol", | |
3076 | [P_UUIDS] = "ReportUUIDs", | |
3077 | [P_SIZES] = "ReportSizes", | |
3078 | [P_STATE] = "ReportState", | |
3079 | [P_SYNC_UUID] = "ReportSyncUUID", | |
3080 | [P_AUTH_CHALLENGE] = "AuthChallenge", | |
3081 | [P_AUTH_RESPONSE] = "AuthResponse", | |
3082 | [P_PING] = "Ping", | |
3083 | [P_PING_ACK] = "PingAck", | |
3084 | [P_RECV_ACK] = "RecvAck", | |
3085 | [P_WRITE_ACK] = "WriteAck", | |
3086 | [P_RS_WRITE_ACK] = "RSWriteAck", | |
7be8da07 | 3087 | [P_DISCARD_WRITE] = "DiscardWrite", |
f2ad9063 AG |
3088 | [P_NEG_ACK] = "NegAck", |
3089 | [P_NEG_DREPLY] = "NegDReply", | |
3090 | [P_NEG_RS_DREPLY] = "NegRSDReply", | |
3091 | [P_BARRIER_ACK] = "BarrierAck", | |
3092 | [P_STATE_CHG_REQ] = "StateChgRequest", | |
3093 | [P_STATE_CHG_REPLY] = "StateChgReply", | |
3094 | [P_OV_REQUEST] = "OVRequest", | |
3095 | [P_OV_REPLY] = "OVReply", | |
3096 | [P_OV_RESULT] = "OVResult", | |
3097 | [P_CSUM_RS_REQUEST] = "CsumRSRequest", | |
3098 | [P_RS_IS_IN_SYNC] = "CsumRSIsInSync", | |
3099 | [P_COMPRESSED_BITMAP] = "CBitmap", | |
3100 | [P_DELAY_PROBE] = "DelayProbe", | |
3101 | [P_OUT_OF_SYNC] = "OutOfSync", | |
7be8da07 | 3102 | [P_RETRY_WRITE] = "RetryWrite", |
f2ad9063 AG |
3103 | }; |
3104 | ||
3105 | if (cmd == P_HAND_SHAKE_M) | |
3106 | return "HandShakeM"; | |
3107 | if (cmd == P_HAND_SHAKE_S) | |
3108 | return "HandShakeS"; | |
3109 | if (cmd == P_HAND_SHAKE) | |
3110 | return "HandShake"; | |
6e849ce8 | 3111 | if (cmd >= ARRAY_SIZE(cmdnames)) |
f2ad9063 AG |
3112 | return "Unknown"; |
3113 | return cmdnames[cmd]; | |
3114 | } | |
3115 | ||
7be8da07 AG |
3116 | /** |
3117 | * drbd_wait_misc - wait for a request to make progress | |
3118 | * @mdev: device associated with the request | |
3119 | * @i: the struct drbd_interval embedded in struct drbd_request or | |
3120 | * struct drbd_peer_request | |
3121 | */ | |
3122 | int drbd_wait_misc(struct drbd_conf *mdev, struct drbd_interval *i) | |
3123 | { | |
3124 | struct net_conf *net_conf = mdev->tconn->net_conf; | |
3125 | DEFINE_WAIT(wait); | |
3126 | long timeout; | |
3127 | ||
3128 | if (!net_conf) | |
3129 | return -ETIMEDOUT; | |
3130 | timeout = MAX_SCHEDULE_TIMEOUT; | |
3131 | if (net_conf->ko_count) | |
3132 | timeout = net_conf->timeout * HZ / 10 * net_conf->ko_count; | |
3133 | ||
3134 | /* Indicate to wake up mdev->misc_wait on progress. */ | |
3135 | i->waiting = true; | |
3136 | prepare_to_wait(&mdev->misc_wait, &wait, TASK_INTERRUPTIBLE); | |
3137 | spin_unlock_irq(&mdev->tconn->req_lock); | |
3138 | timeout = schedule_timeout(timeout); | |
3139 | finish_wait(&mdev->misc_wait, &wait); | |
3140 | spin_lock_irq(&mdev->tconn->req_lock); | |
3141 | if (!timeout || mdev->state.conn < C_CONNECTED) | |
3142 | return -ETIMEDOUT; | |
3143 | if (signal_pending(current)) | |
3144 | return -ERESTARTSYS; | |
3145 | return 0; | |
3146 | } | |
3147 | ||
b411b363 PR |
3148 | #ifdef CONFIG_DRBD_FAULT_INJECTION |
3149 | /* Fault insertion support including random number generator shamelessly | |
3150 | * stolen from kernel/rcutorture.c */ | |
3151 | struct fault_random_state { | |
3152 | unsigned long state; | |
3153 | unsigned long count; | |
3154 | }; | |
3155 | ||
3156 | #define FAULT_RANDOM_MULT 39916801 /* prime */ | |
3157 | #define FAULT_RANDOM_ADD 479001701 /* prime */ | |
3158 | #define FAULT_RANDOM_REFRESH 10000 | |
3159 | ||
3160 | /* | |
3161 | * Crude but fast random-number generator. Uses a linear congruential | |
3162 | * generator, with occasional help from get_random_bytes(). | |
3163 | */ | |
3164 | static unsigned long | |
3165 | _drbd_fault_random(struct fault_random_state *rsp) | |
3166 | { | |
3167 | long refresh; | |
3168 | ||
49829ea7 | 3169 | if (!rsp->count--) { |
b411b363 PR |
3170 | get_random_bytes(&refresh, sizeof(refresh)); |
3171 | rsp->state += refresh; | |
3172 | rsp->count = FAULT_RANDOM_REFRESH; | |
3173 | } | |
3174 | rsp->state = rsp->state * FAULT_RANDOM_MULT + FAULT_RANDOM_ADD; | |
3175 | return swahw32(rsp->state); | |
3176 | } | |
3177 | ||
3178 | static char * | |
3179 | _drbd_fault_str(unsigned int type) { | |
3180 | static char *_faults[] = { | |
3181 | [DRBD_FAULT_MD_WR] = "Meta-data write", | |
3182 | [DRBD_FAULT_MD_RD] = "Meta-data read", | |
3183 | [DRBD_FAULT_RS_WR] = "Resync write", | |
3184 | [DRBD_FAULT_RS_RD] = "Resync read", | |
3185 | [DRBD_FAULT_DT_WR] = "Data write", | |
3186 | [DRBD_FAULT_DT_RD] = "Data read", | |
3187 | [DRBD_FAULT_DT_RA] = "Data read ahead", | |
3188 | [DRBD_FAULT_BM_ALLOC] = "BM allocation", | |
6b4388ac PR |
3189 | [DRBD_FAULT_AL_EE] = "EE allocation", |
3190 | [DRBD_FAULT_RECEIVE] = "receive data corruption", | |
b411b363 PR |
3191 | }; |
3192 | ||
3193 | return (type < DRBD_FAULT_MAX) ? _faults[type] : "**Unknown**"; | |
3194 | } | |
3195 | ||
3196 | unsigned int | |
3197 | _drbd_insert_fault(struct drbd_conf *mdev, unsigned int type) | |
3198 | { | |
3199 | static struct fault_random_state rrs = {0, 0}; | |
3200 | ||
3201 | unsigned int ret = ( | |
3202 | (fault_devs == 0 || | |
3203 | ((1 << mdev_to_minor(mdev)) & fault_devs) != 0) && | |
3204 | (((_drbd_fault_random(&rrs) % 100) + 1) <= fault_rate)); | |
3205 | ||
3206 | if (ret) { | |
3207 | fault_count++; | |
3208 | ||
7383506c | 3209 | if (__ratelimit(&drbd_ratelimit_state)) |
b411b363 PR |
3210 | dev_warn(DEV, "***Simulating %s failure\n", |
3211 | _drbd_fault_str(type)); | |
3212 | } | |
3213 | ||
3214 | return ret; | |
3215 | } | |
3216 | #endif | |
3217 | ||
3218 | const char *drbd_buildtag(void) | |
3219 | { | |
3220 | /* DRBD built from external sources has here a reference to the | |
3221 | git hash of the source code. */ | |
3222 | ||
3223 | static char buildtag[38] = "\0uilt-in"; | |
3224 | ||
3225 | if (buildtag[0] == 0) { | |
3226 | #ifdef CONFIG_MODULES | |
3227 | if (THIS_MODULE != NULL) | |
3228 | sprintf(buildtag, "srcversion: %-24s", THIS_MODULE->srcversion); | |
3229 | else | |
3230 | #endif | |
3231 | buildtag[0] = 'b'; | |
3232 | } | |
3233 | ||
3234 | return buildtag; | |
3235 | } | |
3236 | ||
3237 | module_init(drbd_init) | |
3238 | module_exit(drbd_cleanup) | |
3239 | ||
b411b363 PR |
3240 | EXPORT_SYMBOL(drbd_conn_str); |
3241 | EXPORT_SYMBOL(drbd_role_str); | |
3242 | EXPORT_SYMBOL(drbd_disk_str); | |
3243 | EXPORT_SYMBOL(drbd_set_st_err_str); |