Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_receiver.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 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | */ | |
24 | ||
25 | ||
b411b363 PR |
26 | #include <linux/module.h> |
27 | ||
28 | #include <asm/uaccess.h> | |
29 | #include <net/sock.h> | |
30 | ||
b411b363 PR |
31 | #include <linux/drbd.h> |
32 | #include <linux/fs.h> | |
33 | #include <linux/file.h> | |
34 | #include <linux/in.h> | |
35 | #include <linux/mm.h> | |
36 | #include <linux/memcontrol.h> | |
37 | #include <linux/mm_inline.h> | |
38 | #include <linux/slab.h> | |
b411b363 PR |
39 | #include <linux/pkt_sched.h> |
40 | #define __KERNEL_SYSCALLS__ | |
41 | #include <linux/unistd.h> | |
42 | #include <linux/vmalloc.h> | |
43 | #include <linux/random.h> | |
b411b363 PR |
44 | #include <linux/string.h> |
45 | #include <linux/scatterlist.h> | |
46 | #include "drbd_int.h" | |
b411b363 PR |
47 | #include "drbd_req.h" |
48 | ||
49 | #include "drbd_vli.h" | |
50 | ||
77351055 PR |
51 | struct packet_info { |
52 | enum drbd_packet cmd; | |
e2857216 AG |
53 | unsigned int size; |
54 | unsigned int vnr; | |
77351055 PR |
55 | }; |
56 | ||
b411b363 PR |
57 | enum finish_epoch { |
58 | FE_STILL_LIVE, | |
59 | FE_DESTROYED, | |
60 | FE_RECYCLED, | |
61 | }; | |
62 | ||
6038178e | 63 | static int drbd_do_features(struct drbd_tconn *tconn); |
13e6037d | 64 | static int drbd_do_auth(struct drbd_tconn *tconn); |
360cc740 | 65 | static int drbd_disconnected(int vnr, void *p, void *data); |
b411b363 PR |
66 | |
67 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event); | |
99920dc5 | 68 | static int e_end_block(struct drbd_work *, int); |
b411b363 | 69 | |
b411b363 PR |
70 | |
71 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
72 | ||
45bb912b LE |
73 | /* |
74 | * some helper functions to deal with single linked page lists, | |
75 | * page->private being our "next" pointer. | |
76 | */ | |
77 | ||
78 | /* If at least n pages are linked at head, get n pages off. | |
79 | * Otherwise, don't modify head, and return NULL. | |
80 | * Locking is the responsibility of the caller. | |
81 | */ | |
82 | static struct page *page_chain_del(struct page **head, int n) | |
83 | { | |
84 | struct page *page; | |
85 | struct page *tmp; | |
86 | ||
87 | BUG_ON(!n); | |
88 | BUG_ON(!head); | |
89 | ||
90 | page = *head; | |
23ce4227 PR |
91 | |
92 | if (!page) | |
93 | return NULL; | |
94 | ||
45bb912b LE |
95 | while (page) { |
96 | tmp = page_chain_next(page); | |
97 | if (--n == 0) | |
98 | break; /* found sufficient pages */ | |
99 | if (tmp == NULL) | |
100 | /* insufficient pages, don't use any of them. */ | |
101 | return NULL; | |
102 | page = tmp; | |
103 | } | |
104 | ||
105 | /* add end of list marker for the returned list */ | |
106 | set_page_private(page, 0); | |
107 | /* actual return value, and adjustment of head */ | |
108 | page = *head; | |
109 | *head = tmp; | |
110 | return page; | |
111 | } | |
112 | ||
113 | /* may be used outside of locks to find the tail of a (usually short) | |
114 | * "private" page chain, before adding it back to a global chain head | |
115 | * with page_chain_add() under a spinlock. */ | |
116 | static struct page *page_chain_tail(struct page *page, int *len) | |
117 | { | |
118 | struct page *tmp; | |
119 | int i = 1; | |
120 | while ((tmp = page_chain_next(page))) | |
121 | ++i, page = tmp; | |
122 | if (len) | |
123 | *len = i; | |
124 | return page; | |
125 | } | |
126 | ||
127 | static int page_chain_free(struct page *page) | |
128 | { | |
129 | struct page *tmp; | |
130 | int i = 0; | |
131 | page_chain_for_each_safe(page, tmp) { | |
132 | put_page(page); | |
133 | ++i; | |
134 | } | |
135 | return i; | |
136 | } | |
137 | ||
138 | static void page_chain_add(struct page **head, | |
139 | struct page *chain_first, struct page *chain_last) | |
140 | { | |
141 | #if 1 | |
142 | struct page *tmp; | |
143 | tmp = page_chain_tail(chain_first, NULL); | |
144 | BUG_ON(tmp != chain_last); | |
145 | #endif | |
146 | ||
147 | /* add chain to head */ | |
148 | set_page_private(chain_last, (unsigned long)*head); | |
149 | *head = chain_first; | |
150 | } | |
151 | ||
152 | static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number) | |
b411b363 PR |
153 | { |
154 | struct page *page = NULL; | |
45bb912b LE |
155 | struct page *tmp = NULL; |
156 | int i = 0; | |
b411b363 PR |
157 | |
158 | /* Yes, testing drbd_pp_vacant outside the lock is racy. | |
159 | * So what. It saves a spin_lock. */ | |
45bb912b | 160 | if (drbd_pp_vacant >= number) { |
b411b363 | 161 | spin_lock(&drbd_pp_lock); |
45bb912b LE |
162 | page = page_chain_del(&drbd_pp_pool, number); |
163 | if (page) | |
164 | drbd_pp_vacant -= number; | |
b411b363 | 165 | spin_unlock(&drbd_pp_lock); |
45bb912b LE |
166 | if (page) |
167 | return page; | |
b411b363 | 168 | } |
45bb912b | 169 | |
b411b363 PR |
170 | /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD |
171 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
172 | * which in turn might block on the other node at this very place. */ | |
45bb912b LE |
173 | for (i = 0; i < number; i++) { |
174 | tmp = alloc_page(GFP_TRY); | |
175 | if (!tmp) | |
176 | break; | |
177 | set_page_private(tmp, (unsigned long)page); | |
178 | page = tmp; | |
179 | } | |
180 | ||
181 | if (i == number) | |
182 | return page; | |
183 | ||
184 | /* Not enough pages immediately available this time. | |
185 | * No need to jump around here, drbd_pp_alloc will retry this | |
186 | * function "soon". */ | |
187 | if (page) { | |
188 | tmp = page_chain_tail(page, NULL); | |
189 | spin_lock(&drbd_pp_lock); | |
190 | page_chain_add(&drbd_pp_pool, page, tmp); | |
191 | drbd_pp_vacant += i; | |
192 | spin_unlock(&drbd_pp_lock); | |
193 | } | |
194 | return NULL; | |
b411b363 PR |
195 | } |
196 | ||
b411b363 PR |
197 | static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed) |
198 | { | |
db830c46 | 199 | struct drbd_peer_request *peer_req; |
b411b363 PR |
200 | struct list_head *le, *tle; |
201 | ||
202 | /* The EEs are always appended to the end of the list. Since | |
203 | they are sent in order over the wire, they have to finish | |
204 | in order. As soon as we see the first not finished we can | |
205 | stop to examine the list... */ | |
206 | ||
207 | list_for_each_safe(le, tle, &mdev->net_ee) { | |
db830c46 AG |
208 | peer_req = list_entry(le, struct drbd_peer_request, w.list); |
209 | if (drbd_ee_has_active_page(peer_req)) | |
b411b363 PR |
210 | break; |
211 | list_move(le, to_be_freed); | |
212 | } | |
213 | } | |
214 | ||
215 | static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) | |
216 | { | |
217 | LIST_HEAD(reclaimed); | |
db830c46 | 218 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 219 | |
87eeee41 | 220 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 221 | reclaim_net_ee(mdev, &reclaimed); |
87eeee41 | 222 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 223 | |
db830c46 AG |
224 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
225 | drbd_free_net_ee(mdev, peer_req); | |
b411b363 PR |
226 | } |
227 | ||
228 | /** | |
45bb912b | 229 | * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled) |
b411b363 | 230 | * @mdev: DRBD device. |
45bb912b LE |
231 | * @number: number of pages requested |
232 | * @retry: whether to retry, if not enough pages are available right now | |
233 | * | |
234 | * Tries to allocate number pages, first from our own page pool, then from | |
235 | * the kernel, unless this allocation would exceed the max_buffers setting. | |
236 | * Possibly retry until DRBD frees sufficient pages somewhere else. | |
b411b363 | 237 | * |
45bb912b | 238 | * Returns a page chain linked via page->private. |
b411b363 | 239 | */ |
45bb912b | 240 | static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry) |
b411b363 PR |
241 | { |
242 | struct page *page = NULL; | |
243 | DEFINE_WAIT(wait); | |
244 | ||
45bb912b LE |
245 | /* Yes, we may run up to @number over max_buffers. If we |
246 | * follow it strictly, the admin will get it wrong anyways. */ | |
89e58e75 | 247 | if (atomic_read(&mdev->pp_in_use) < mdev->tconn->net_conf->max_buffers) |
45bb912b | 248 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 | 249 | |
45bb912b | 250 | while (page == NULL) { |
b411b363 PR |
251 | prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); |
252 | ||
253 | drbd_kick_lo_and_reclaim_net(mdev); | |
254 | ||
89e58e75 | 255 | if (atomic_read(&mdev->pp_in_use) < mdev->tconn->net_conf->max_buffers) { |
45bb912b | 256 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 PR |
257 | if (page) |
258 | break; | |
259 | } | |
260 | ||
261 | if (!retry) | |
262 | break; | |
263 | ||
264 | if (signal_pending(current)) { | |
265 | dev_warn(DEV, "drbd_pp_alloc interrupted!\n"); | |
266 | break; | |
267 | } | |
268 | ||
269 | schedule(); | |
270 | } | |
271 | finish_wait(&drbd_pp_wait, &wait); | |
272 | ||
45bb912b LE |
273 | if (page) |
274 | atomic_add(number, &mdev->pp_in_use); | |
b411b363 PR |
275 | return page; |
276 | } | |
277 | ||
278 | /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc. | |
87eeee41 | 279 | * Is also used from inside an other spin_lock_irq(&mdev->tconn->req_lock); |
45bb912b LE |
280 | * Either links the page chain back to the global pool, |
281 | * or returns all pages to the system. */ | |
435f0740 | 282 | static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net) |
b411b363 | 283 | { |
435f0740 | 284 | atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use; |
b411b363 | 285 | int i; |
435f0740 | 286 | |
81a5d60e | 287 | if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count) |
45bb912b LE |
288 | i = page_chain_free(page); |
289 | else { | |
290 | struct page *tmp; | |
291 | tmp = page_chain_tail(page, &i); | |
292 | spin_lock(&drbd_pp_lock); | |
293 | page_chain_add(&drbd_pp_pool, page, tmp); | |
294 | drbd_pp_vacant += i; | |
295 | spin_unlock(&drbd_pp_lock); | |
b411b363 | 296 | } |
435f0740 | 297 | i = atomic_sub_return(i, a); |
45bb912b | 298 | if (i < 0) |
435f0740 LE |
299 | dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n", |
300 | is_net ? "pp_in_use_by_net" : "pp_in_use", i); | |
b411b363 PR |
301 | wake_up(&drbd_pp_wait); |
302 | } | |
303 | ||
304 | /* | |
305 | You need to hold the req_lock: | |
306 | _drbd_wait_ee_list_empty() | |
307 | ||
308 | You must not have the req_lock: | |
309 | drbd_free_ee() | |
310 | drbd_alloc_ee() | |
311 | drbd_init_ee() | |
312 | drbd_release_ee() | |
313 | drbd_ee_fix_bhs() | |
314 | drbd_process_done_ee() | |
315 | drbd_clear_done_ee() | |
316 | drbd_wait_ee_list_empty() | |
317 | */ | |
318 | ||
f6ffca9f AG |
319 | struct drbd_peer_request * |
320 | drbd_alloc_ee(struct drbd_conf *mdev, u64 id, sector_t sector, | |
321 | unsigned int data_size, gfp_t gfp_mask) __must_hold(local) | |
b411b363 | 322 | { |
db830c46 | 323 | struct drbd_peer_request *peer_req; |
b411b363 | 324 | struct page *page; |
45bb912b | 325 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b411b363 | 326 | |
0cf9d27e | 327 | if (drbd_insert_fault(mdev, DRBD_FAULT_AL_EE)) |
b411b363 PR |
328 | return NULL; |
329 | ||
db830c46 AG |
330 | peer_req = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); |
331 | if (!peer_req) { | |
b411b363 PR |
332 | if (!(gfp_mask & __GFP_NOWARN)) |
333 | dev_err(DEV, "alloc_ee: Allocation of an EE failed\n"); | |
334 | return NULL; | |
335 | } | |
336 | ||
45bb912b LE |
337 | page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT)); |
338 | if (!page) | |
339 | goto fail; | |
b411b363 | 340 | |
db830c46 AG |
341 | drbd_clear_interval(&peer_req->i); |
342 | peer_req->i.size = data_size; | |
343 | peer_req->i.sector = sector; | |
344 | peer_req->i.local = false; | |
345 | peer_req->i.waiting = false; | |
346 | ||
347 | peer_req->epoch = NULL; | |
a21e9298 | 348 | peer_req->w.mdev = mdev; |
db830c46 AG |
349 | peer_req->pages = page; |
350 | atomic_set(&peer_req->pending_bios, 0); | |
351 | peer_req->flags = 0; | |
9a8e7753 AG |
352 | /* |
353 | * The block_id is opaque to the receiver. It is not endianness | |
354 | * converted, and sent back to the sender unchanged. | |
355 | */ | |
db830c46 | 356 | peer_req->block_id = id; |
b411b363 | 357 | |
db830c46 | 358 | return peer_req; |
b411b363 | 359 | |
45bb912b | 360 | fail: |
db830c46 | 361 | mempool_free(peer_req, drbd_ee_mempool); |
b411b363 PR |
362 | return NULL; |
363 | } | |
364 | ||
db830c46 | 365 | void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_peer_request *peer_req, |
f6ffca9f | 366 | int is_net) |
b411b363 | 367 | { |
db830c46 AG |
368 | if (peer_req->flags & EE_HAS_DIGEST) |
369 | kfree(peer_req->digest); | |
370 | drbd_pp_free(mdev, peer_req->pages, is_net); | |
371 | D_ASSERT(atomic_read(&peer_req->pending_bios) == 0); | |
372 | D_ASSERT(drbd_interval_empty(&peer_req->i)); | |
373 | mempool_free(peer_req, drbd_ee_mempool); | |
b411b363 PR |
374 | } |
375 | ||
376 | int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) | |
377 | { | |
378 | LIST_HEAD(work_list); | |
db830c46 | 379 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 380 | int count = 0; |
435f0740 | 381 | int is_net = list == &mdev->net_ee; |
b411b363 | 382 | |
87eeee41 | 383 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 384 | list_splice_init(list, &work_list); |
87eeee41 | 385 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 386 | |
db830c46 AG |
387 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
388 | drbd_free_some_ee(mdev, peer_req, is_net); | |
b411b363 PR |
389 | count++; |
390 | } | |
391 | return count; | |
392 | } | |
393 | ||
394 | ||
32862ec7 | 395 | /* See also comments in _req_mod(,BARRIER_ACKED) |
b411b363 PR |
396 | * and receive_Barrier. |
397 | * | |
398 | * Move entries from net_ee to done_ee, if ready. | |
399 | * Grab done_ee, call all callbacks, free the entries. | |
400 | * The callbacks typically send out ACKs. | |
401 | */ | |
402 | static int drbd_process_done_ee(struct drbd_conf *mdev) | |
403 | { | |
404 | LIST_HEAD(work_list); | |
405 | LIST_HEAD(reclaimed); | |
db830c46 | 406 | struct drbd_peer_request *peer_req, *t; |
e2b3032b | 407 | int err = 0; |
b411b363 | 408 | |
87eeee41 | 409 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
410 | reclaim_net_ee(mdev, &reclaimed); |
411 | list_splice_init(&mdev->done_ee, &work_list); | |
87eeee41 | 412 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 413 | |
db830c46 AG |
414 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
415 | drbd_free_net_ee(mdev, peer_req); | |
b411b363 PR |
416 | |
417 | /* possible callbacks here: | |
7be8da07 | 418 | * e_end_block, and e_end_resync_block, e_send_discard_write. |
b411b363 PR |
419 | * all ignore the last argument. |
420 | */ | |
db830c46 | 421 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
e2b3032b AG |
422 | int err2; |
423 | ||
b411b363 | 424 | /* list_del not necessary, next/prev members not touched */ |
e2b3032b AG |
425 | err2 = peer_req->w.cb(&peer_req->w, !!err); |
426 | if (!err) | |
427 | err = err2; | |
db830c46 | 428 | drbd_free_ee(mdev, peer_req); |
b411b363 PR |
429 | } |
430 | wake_up(&mdev->ee_wait); | |
431 | ||
e2b3032b | 432 | return err; |
b411b363 PR |
433 | } |
434 | ||
435 | void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
436 | { | |
437 | DEFINE_WAIT(wait); | |
438 | ||
439 | /* avoids spin_lock/unlock | |
440 | * and calling prepare_to_wait in the fast path */ | |
441 | while (!list_empty(head)) { | |
442 | prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); | |
87eeee41 | 443 | spin_unlock_irq(&mdev->tconn->req_lock); |
7eaceacc | 444 | io_schedule(); |
b411b363 | 445 | finish_wait(&mdev->ee_wait, &wait); |
87eeee41 | 446 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
447 | } |
448 | } | |
449 | ||
450 | void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
451 | { | |
87eeee41 | 452 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 453 | _drbd_wait_ee_list_empty(mdev, head); |
87eeee41 | 454 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
455 | } |
456 | ||
457 | /* see also kernel_accept; which is only present since 2.6.18. | |
458 | * also we want to log which part of it failed, exactly */ | |
7653620d | 459 | static int drbd_accept(const char **what, struct socket *sock, struct socket **newsock) |
b411b363 PR |
460 | { |
461 | struct sock *sk = sock->sk; | |
462 | int err = 0; | |
463 | ||
464 | *what = "listen"; | |
465 | err = sock->ops->listen(sock, 5); | |
466 | if (err < 0) | |
467 | goto out; | |
468 | ||
469 | *what = "sock_create_lite"; | |
470 | err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, | |
471 | newsock); | |
472 | if (err < 0) | |
473 | goto out; | |
474 | ||
475 | *what = "accept"; | |
476 | err = sock->ops->accept(sock, *newsock, 0); | |
477 | if (err < 0) { | |
478 | sock_release(*newsock); | |
479 | *newsock = NULL; | |
480 | goto out; | |
481 | } | |
482 | (*newsock)->ops = sock->ops; | |
483 | ||
484 | out: | |
485 | return err; | |
486 | } | |
487 | ||
dbd9eea0 | 488 | static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags) |
b411b363 PR |
489 | { |
490 | mm_segment_t oldfs; | |
491 | struct kvec iov = { | |
492 | .iov_base = buf, | |
493 | .iov_len = size, | |
494 | }; | |
495 | struct msghdr msg = { | |
496 | .msg_iovlen = 1, | |
497 | .msg_iov = (struct iovec *)&iov, | |
498 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) | |
499 | }; | |
500 | int rv; | |
501 | ||
502 | oldfs = get_fs(); | |
503 | set_fs(KERNEL_DS); | |
504 | rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); | |
505 | set_fs(oldfs); | |
506 | ||
507 | return rv; | |
508 | } | |
509 | ||
de0ff338 | 510 | static int drbd_recv(struct drbd_tconn *tconn, void *buf, size_t size) |
b411b363 PR |
511 | { |
512 | mm_segment_t oldfs; | |
513 | struct kvec iov = { | |
514 | .iov_base = buf, | |
515 | .iov_len = size, | |
516 | }; | |
517 | struct msghdr msg = { | |
518 | .msg_iovlen = 1, | |
519 | .msg_iov = (struct iovec *)&iov, | |
520 | .msg_flags = MSG_WAITALL | MSG_NOSIGNAL | |
521 | }; | |
522 | int rv; | |
523 | ||
524 | oldfs = get_fs(); | |
525 | set_fs(KERNEL_DS); | |
526 | ||
527 | for (;;) { | |
de0ff338 | 528 | rv = sock_recvmsg(tconn->data.socket, &msg, size, msg.msg_flags); |
b411b363 PR |
529 | if (rv == size) |
530 | break; | |
531 | ||
532 | /* Note: | |
533 | * ECONNRESET other side closed the connection | |
534 | * ERESTARTSYS (on sock) we got a signal | |
535 | */ | |
536 | ||
537 | if (rv < 0) { | |
538 | if (rv == -ECONNRESET) | |
de0ff338 | 539 | conn_info(tconn, "sock was reset by peer\n"); |
b411b363 | 540 | else if (rv != -ERESTARTSYS) |
de0ff338 | 541 | conn_err(tconn, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
542 | break; |
543 | } else if (rv == 0) { | |
de0ff338 | 544 | conn_info(tconn, "sock was shut down by peer\n"); |
b411b363 PR |
545 | break; |
546 | } else { | |
547 | /* signal came in, or peer/link went down, | |
548 | * after we read a partial message | |
549 | */ | |
550 | /* D_ASSERT(signal_pending(current)); */ | |
551 | break; | |
552 | } | |
553 | }; | |
554 | ||
555 | set_fs(oldfs); | |
556 | ||
557 | if (rv != size) | |
bbeb641c | 558 | conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 PR |
559 | |
560 | return rv; | |
561 | } | |
562 | ||
c6967746 AG |
563 | static int drbd_recv_all(struct drbd_tconn *tconn, void *buf, size_t size) |
564 | { | |
565 | int err; | |
566 | ||
567 | err = drbd_recv(tconn, buf, size); | |
568 | if (err != size) { | |
569 | if (err >= 0) | |
570 | err = -EIO; | |
571 | } else | |
572 | err = 0; | |
573 | return err; | |
574 | } | |
575 | ||
a5c31904 AG |
576 | static int drbd_recv_all_warn(struct drbd_tconn *tconn, void *buf, size_t size) |
577 | { | |
578 | int err; | |
579 | ||
580 | err = drbd_recv_all(tconn, buf, size); | |
581 | if (err && !signal_pending(current)) | |
582 | conn_warn(tconn, "short read (expected size %d)\n", (int)size); | |
583 | return err; | |
584 | } | |
585 | ||
5dbf1673 LE |
586 | /* quoting tcp(7): |
587 | * On individual connections, the socket buffer size must be set prior to the | |
588 | * listen(2) or connect(2) calls in order to have it take effect. | |
589 | * This is our wrapper to do so. | |
590 | */ | |
591 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
592 | unsigned int rcv) | |
593 | { | |
594 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
595 | if (snd) { | |
596 | sock->sk->sk_sndbuf = snd; | |
597 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
598 | } | |
599 | if (rcv) { | |
600 | sock->sk->sk_rcvbuf = rcv; | |
601 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
602 | } | |
603 | } | |
604 | ||
eac3e990 | 605 | static struct socket *drbd_try_connect(struct drbd_tconn *tconn) |
b411b363 PR |
606 | { |
607 | const char *what; | |
608 | struct socket *sock; | |
609 | struct sockaddr_in6 src_in6; | |
610 | int err; | |
611 | int disconnect_on_error = 1; | |
612 | ||
eac3e990 | 613 | if (!get_net_conf(tconn)) |
b411b363 PR |
614 | return NULL; |
615 | ||
616 | what = "sock_create_kern"; | |
eac3e990 | 617 | err = sock_create_kern(((struct sockaddr *)tconn->net_conf->my_addr)->sa_family, |
b411b363 PR |
618 | SOCK_STREAM, IPPROTO_TCP, &sock); |
619 | if (err < 0) { | |
620 | sock = NULL; | |
621 | goto out; | |
622 | } | |
623 | ||
624 | sock->sk->sk_rcvtimeo = | |
eac3e990 PR |
625 | sock->sk->sk_sndtimeo = tconn->net_conf->try_connect_int*HZ; |
626 | drbd_setbufsize(sock, tconn->net_conf->sndbuf_size, | |
627 | tconn->net_conf->rcvbuf_size); | |
b411b363 PR |
628 | |
629 | /* explicitly bind to the configured IP as source IP | |
630 | * for the outgoing connections. | |
631 | * This is needed for multihomed hosts and to be | |
632 | * able to use lo: interfaces for drbd. | |
633 | * Make sure to use 0 as port number, so linux selects | |
634 | * a free one dynamically. | |
635 | */ | |
eac3e990 PR |
636 | memcpy(&src_in6, tconn->net_conf->my_addr, |
637 | min_t(int, tconn->net_conf->my_addr_len, sizeof(src_in6))); | |
638 | if (((struct sockaddr *)tconn->net_conf->my_addr)->sa_family == AF_INET6) | |
b411b363 PR |
639 | src_in6.sin6_port = 0; |
640 | else | |
641 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
642 | ||
643 | what = "bind before connect"; | |
644 | err = sock->ops->bind(sock, | |
645 | (struct sockaddr *) &src_in6, | |
eac3e990 | 646 | tconn->net_conf->my_addr_len); |
b411b363 PR |
647 | if (err < 0) |
648 | goto out; | |
649 | ||
650 | /* connect may fail, peer not yet available. | |
651 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
652 | disconnect_on_error = 0; | |
653 | what = "connect"; | |
654 | err = sock->ops->connect(sock, | |
eac3e990 PR |
655 | (struct sockaddr *)tconn->net_conf->peer_addr, |
656 | tconn->net_conf->peer_addr_len, 0); | |
b411b363 PR |
657 | |
658 | out: | |
659 | if (err < 0) { | |
660 | if (sock) { | |
661 | sock_release(sock); | |
662 | sock = NULL; | |
663 | } | |
664 | switch (-err) { | |
665 | /* timeout, busy, signal pending */ | |
666 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
667 | case EINTR: case ERESTARTSYS: | |
668 | /* peer not (yet) available, network problem */ | |
669 | case ECONNREFUSED: case ENETUNREACH: | |
670 | case EHOSTDOWN: case EHOSTUNREACH: | |
671 | disconnect_on_error = 0; | |
672 | break; | |
673 | default: | |
eac3e990 | 674 | conn_err(tconn, "%s failed, err = %d\n", what, err); |
b411b363 PR |
675 | } |
676 | if (disconnect_on_error) | |
bbeb641c | 677 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 678 | } |
eac3e990 | 679 | put_net_conf(tconn); |
b411b363 PR |
680 | return sock; |
681 | } | |
682 | ||
7653620d | 683 | static struct socket *drbd_wait_for_connect(struct drbd_tconn *tconn) |
b411b363 PR |
684 | { |
685 | int timeo, err; | |
686 | struct socket *s_estab = NULL, *s_listen; | |
687 | const char *what; | |
688 | ||
7653620d | 689 | if (!get_net_conf(tconn)) |
b411b363 PR |
690 | return NULL; |
691 | ||
692 | what = "sock_create_kern"; | |
7653620d | 693 | err = sock_create_kern(((struct sockaddr *)tconn->net_conf->my_addr)->sa_family, |
b411b363 PR |
694 | SOCK_STREAM, IPPROTO_TCP, &s_listen); |
695 | if (err) { | |
696 | s_listen = NULL; | |
697 | goto out; | |
698 | } | |
699 | ||
7653620d | 700 | timeo = tconn->net_conf->try_connect_int * HZ; |
b411b363 PR |
701 | timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ |
702 | ||
703 | s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
704 | s_listen->sk->sk_rcvtimeo = timeo; | |
705 | s_listen->sk->sk_sndtimeo = timeo; | |
7653620d PR |
706 | drbd_setbufsize(s_listen, tconn->net_conf->sndbuf_size, |
707 | tconn->net_conf->rcvbuf_size); | |
b411b363 PR |
708 | |
709 | what = "bind before listen"; | |
710 | err = s_listen->ops->bind(s_listen, | |
7653620d PR |
711 | (struct sockaddr *) tconn->net_conf->my_addr, |
712 | tconn->net_conf->my_addr_len); | |
b411b363 PR |
713 | if (err < 0) |
714 | goto out; | |
715 | ||
7653620d | 716 | err = drbd_accept(&what, s_listen, &s_estab); |
b411b363 PR |
717 | |
718 | out: | |
719 | if (s_listen) | |
720 | sock_release(s_listen); | |
721 | if (err < 0) { | |
722 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
7653620d | 723 | conn_err(tconn, "%s failed, err = %d\n", what, err); |
bbeb641c | 724 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
725 | } |
726 | } | |
7653620d | 727 | put_net_conf(tconn); |
b411b363 PR |
728 | |
729 | return s_estab; | |
730 | } | |
731 | ||
7c96715a | 732 | static int drbd_send_fp(struct drbd_tconn *tconn, struct drbd_socket *sock, enum drbd_packet cmd) |
b411b363 | 733 | { |
5a87d920 | 734 | struct p_header *h = tconn->data.sbuf; |
b411b363 | 735 | |
ecf2363c | 736 | return !_conn_send_cmd(tconn, 0, sock, cmd, h, sizeof(*h), 0); |
b411b363 PR |
737 | } |
738 | ||
a25b63f1 | 739 | static enum drbd_packet drbd_recv_fp(struct drbd_tconn *tconn, struct socket *sock) |
b411b363 | 740 | { |
7c96715a | 741 | struct p_header80 h; |
b411b363 PR |
742 | int rr; |
743 | ||
7c96715a | 744 | rr = drbd_recv_short(sock, &h, sizeof(h), 0); |
b411b363 | 745 | |
7c96715a AG |
746 | if (rr == sizeof(h) && h.magic == cpu_to_be32(DRBD_MAGIC)) |
747 | return be16_to_cpu(h.command); | |
b411b363 PR |
748 | |
749 | return 0xffff; | |
750 | } | |
751 | ||
752 | /** | |
753 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
b411b363 PR |
754 | * @sock: pointer to the pointer to the socket. |
755 | */ | |
dbd9eea0 | 756 | static int drbd_socket_okay(struct socket **sock) |
b411b363 PR |
757 | { |
758 | int rr; | |
759 | char tb[4]; | |
760 | ||
761 | if (!*sock) | |
81e84650 | 762 | return false; |
b411b363 | 763 | |
dbd9eea0 | 764 | rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); |
b411b363 PR |
765 | |
766 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 767 | return true; |
b411b363 PR |
768 | } else { |
769 | sock_release(*sock); | |
770 | *sock = NULL; | |
81e84650 | 771 | return false; |
b411b363 PR |
772 | } |
773 | } | |
2325eb66 PR |
774 | /* Gets called if a connection is established, or if a new minor gets created |
775 | in a connection */ | |
776 | int drbd_connected(int vnr, void *p, void *data) | |
907599e0 PR |
777 | { |
778 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
0829f5ed | 779 | int err; |
907599e0 PR |
780 | |
781 | atomic_set(&mdev->packet_seq, 0); | |
782 | mdev->peer_seq = 0; | |
783 | ||
8410da8f PR |
784 | mdev->state_mutex = mdev->tconn->agreed_pro_version < 100 ? |
785 | &mdev->tconn->cstate_mutex : | |
786 | &mdev->own_state_mutex; | |
787 | ||
0829f5ed AG |
788 | err = drbd_send_sync_param(mdev); |
789 | if (!err) | |
790 | err = drbd_send_sizes(mdev, 0, 0); | |
791 | if (!err) | |
792 | err = drbd_send_uuids(mdev); | |
793 | if (!err) | |
794 | err = drbd_send_state(mdev); | |
907599e0 PR |
795 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); |
796 | clear_bit(RESIZE_PENDING, &mdev->flags); | |
8b924f1d | 797 | mod_timer(&mdev->request_timer, jiffies + HZ); /* just start it here. */ |
0829f5ed | 798 | return err; |
907599e0 PR |
799 | } |
800 | ||
b411b363 PR |
801 | /* |
802 | * return values: | |
803 | * 1 yes, we have a valid connection | |
804 | * 0 oops, did not work out, please try again | |
805 | * -1 peer talks different language, | |
806 | * no point in trying again, please go standalone. | |
807 | * -2 We do not have a network config... | |
808 | */ | |
907599e0 | 809 | static int drbd_connect(struct drbd_tconn *tconn) |
b411b363 | 810 | { |
2bf89621 | 811 | struct socket *sock, *msock; |
b411b363 PR |
812 | int try, h, ok; |
813 | ||
bbeb641c | 814 | if (conn_request_state(tconn, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) |
b411b363 PR |
815 | return -2; |
816 | ||
907599e0 | 817 | clear_bit(DISCARD_CONCURRENT, &tconn->flags); |
0916e0e3 AG |
818 | |
819 | /* Assume that the peer only understands protocol 80 until we know better. */ | |
820 | tconn->agreed_pro_version = 80; | |
b411b363 | 821 | |
b411b363 | 822 | do { |
2bf89621 AG |
823 | struct socket *s; |
824 | ||
b411b363 PR |
825 | for (try = 0;;) { |
826 | /* 3 tries, this should take less than a second! */ | |
907599e0 | 827 | s = drbd_try_connect(tconn); |
b411b363 PR |
828 | if (s || ++try >= 3) |
829 | break; | |
830 | /* give the other side time to call bind() & listen() */ | |
20ee6390 | 831 | schedule_timeout_interruptible(HZ / 10); |
b411b363 PR |
832 | } |
833 | ||
834 | if (s) { | |
2bf89621 AG |
835 | if (!tconn->data.socket) { |
836 | tconn->data.socket = s; | |
e5d6f33a | 837 | drbd_send_fp(tconn, &tconn->data, P_INITIAL_DATA); |
2bf89621 AG |
838 | } else if (!tconn->meta.socket) { |
839 | tconn->meta.socket = s; | |
e5d6f33a | 840 | drbd_send_fp(tconn, &tconn->meta, P_INITIAL_META); |
b411b363 | 841 | } else { |
907599e0 | 842 | conn_err(tconn, "Logic error in drbd_connect()\n"); |
b411b363 PR |
843 | goto out_release_sockets; |
844 | } | |
845 | } | |
846 | ||
2bf89621 | 847 | if (tconn->data.socket && tconn->meta.socket) { |
907599e0 | 848 | schedule_timeout_interruptible(tconn->net_conf->ping_timeo*HZ/10); |
2bf89621 AG |
849 | ok = drbd_socket_okay(&tconn->data.socket); |
850 | ok = drbd_socket_okay(&tconn->meta.socket) && ok; | |
b411b363 PR |
851 | if (ok) |
852 | break; | |
853 | } | |
854 | ||
855 | retry: | |
907599e0 | 856 | s = drbd_wait_for_connect(tconn); |
b411b363 | 857 | if (s) { |
907599e0 | 858 | try = drbd_recv_fp(tconn, s); |
2bf89621 AG |
859 | drbd_socket_okay(&tconn->data.socket); |
860 | drbd_socket_okay(&tconn->meta.socket); | |
b411b363 | 861 | switch (try) { |
e5d6f33a | 862 | case P_INITIAL_DATA: |
2bf89621 | 863 | if (tconn->data.socket) { |
907599e0 | 864 | conn_warn(tconn, "initial packet S crossed\n"); |
2bf89621 | 865 | sock_release(tconn->data.socket); |
b411b363 | 866 | } |
2bf89621 | 867 | tconn->data.socket = s; |
b411b363 | 868 | break; |
e5d6f33a | 869 | case P_INITIAL_META: |
2bf89621 | 870 | if (tconn->meta.socket) { |
907599e0 | 871 | conn_warn(tconn, "initial packet M crossed\n"); |
2bf89621 | 872 | sock_release(tconn->meta.socket); |
b411b363 | 873 | } |
2bf89621 | 874 | tconn->meta.socket = s; |
907599e0 | 875 | set_bit(DISCARD_CONCURRENT, &tconn->flags); |
b411b363 PR |
876 | break; |
877 | default: | |
907599e0 | 878 | conn_warn(tconn, "Error receiving initial packet\n"); |
b411b363 PR |
879 | sock_release(s); |
880 | if (random32() & 1) | |
881 | goto retry; | |
882 | } | |
883 | } | |
884 | ||
bbeb641c | 885 | if (tconn->cstate <= C_DISCONNECTING) |
b411b363 PR |
886 | goto out_release_sockets; |
887 | if (signal_pending(current)) { | |
888 | flush_signals(current); | |
889 | smp_rmb(); | |
907599e0 | 890 | if (get_t_state(&tconn->receiver) == EXITING) |
b411b363 PR |
891 | goto out_release_sockets; |
892 | } | |
893 | ||
2bf89621 AG |
894 | if (tconn->data.socket && &tconn->meta.socket) { |
895 | ok = drbd_socket_okay(&tconn->data.socket); | |
896 | ok = drbd_socket_okay(&tconn->meta.socket) && ok; | |
b411b363 PR |
897 | if (ok) |
898 | break; | |
899 | } | |
900 | } while (1); | |
901 | ||
2bf89621 AG |
902 | sock = tconn->data.socket; |
903 | msock = tconn->meta.socket; | |
904 | ||
b411b363 PR |
905 | msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ |
906 | sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
907 | ||
908 | sock->sk->sk_allocation = GFP_NOIO; | |
909 | msock->sk->sk_allocation = GFP_NOIO; | |
910 | ||
911 | sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; | |
912 | msock->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
913 | ||
b411b363 | 914 | /* NOT YET ... |
907599e0 | 915 | * sock->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
b411b363 | 916 | * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
6038178e | 917 | * first set it to the P_CONNECTION_FEATURES timeout, |
b411b363 PR |
918 | * which we set to 4x the configured ping_timeout. */ |
919 | sock->sk->sk_sndtimeo = | |
907599e0 | 920 | sock->sk->sk_rcvtimeo = tconn->net_conf->ping_timeo*4*HZ/10; |
b411b363 | 921 | |
907599e0 PR |
922 | msock->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
923 | msock->sk->sk_rcvtimeo = tconn->net_conf->ping_int*HZ; | |
b411b363 PR |
924 | |
925 | /* we don't want delays. | |
25985edc | 926 | * we use TCP_CORK where appropriate, though */ |
b411b363 PR |
927 | drbd_tcp_nodelay(sock); |
928 | drbd_tcp_nodelay(msock); | |
929 | ||
907599e0 | 930 | tconn->last_received = jiffies; |
b411b363 | 931 | |
6038178e | 932 | h = drbd_do_features(tconn); |
b411b363 PR |
933 | if (h <= 0) |
934 | return h; | |
935 | ||
907599e0 | 936 | if (tconn->cram_hmac_tfm) { |
b411b363 | 937 | /* drbd_request_state(mdev, NS(conn, WFAuth)); */ |
907599e0 | 938 | switch (drbd_do_auth(tconn)) { |
b10d96cb | 939 | case -1: |
907599e0 | 940 | conn_err(tconn, "Authentication of peer failed\n"); |
b411b363 | 941 | return -1; |
b10d96cb | 942 | case 0: |
907599e0 | 943 | conn_err(tconn, "Authentication of peer failed, trying again.\n"); |
b10d96cb | 944 | return 0; |
b411b363 PR |
945 | } |
946 | } | |
947 | ||
bbeb641c | 948 | if (conn_request_state(tconn, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE) < SS_SUCCESS) |
b411b363 PR |
949 | return 0; |
950 | ||
907599e0 | 951 | sock->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
b411b363 PR |
952 | sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
953 | ||
907599e0 | 954 | drbd_thread_start(&tconn->asender); |
b411b363 | 955 | |
387eb308 | 956 | if (drbd_send_protocol(tconn) == -EOPNOTSUPP) |
7e2455c1 | 957 | return -1; |
b411b363 | 958 | |
907599e0 | 959 | return !idr_for_each(&tconn->volumes, drbd_connected, tconn); |
b411b363 PR |
960 | |
961 | out_release_sockets: | |
2bf89621 AG |
962 | if (tconn->data.socket) { |
963 | sock_release(tconn->data.socket); | |
964 | tconn->data.socket = NULL; | |
965 | } | |
966 | if (tconn->meta.socket) { | |
967 | sock_release(tconn->meta.socket); | |
968 | tconn->meta.socket = NULL; | |
969 | } | |
b411b363 PR |
970 | return -1; |
971 | } | |
972 | ||
8172f3e9 | 973 | static int decode_header(struct drbd_tconn *tconn, struct p_header *h, struct packet_info *pi) |
b411b363 | 974 | { |
fd340c12 | 975 | if (h->h80.magic == cpu_to_be32(DRBD_MAGIC)) { |
77351055 PR |
976 | pi->cmd = be16_to_cpu(h->h80.command); |
977 | pi->size = be16_to_cpu(h->h80.length); | |
eefc2f7d | 978 | pi->vnr = 0; |
ca9bc12b | 979 | } else if (h->h95.magic == cpu_to_be16(DRBD_MAGIC_BIG)) { |
77351055 PR |
980 | pi->cmd = be16_to_cpu(h->h95.command); |
981 | pi->size = be32_to_cpu(h->h95.length) & 0x00ffffff; | |
982 | pi->vnr = 0; | |
02918be2 | 983 | } else { |
ce243853 | 984 | conn_err(tconn, "magic?? on data m: 0x%08x c: %d l: %d\n", |
004352fa LE |
985 | be32_to_cpu(h->h80.magic), |
986 | be16_to_cpu(h->h80.command), | |
987 | be16_to_cpu(h->h80.length)); | |
8172f3e9 | 988 | return -EINVAL; |
b411b363 | 989 | } |
8172f3e9 | 990 | return 0; |
257d0af6 PR |
991 | } |
992 | ||
9ba7aa00 | 993 | static int drbd_recv_header(struct drbd_tconn *tconn, struct packet_info *pi) |
257d0af6 | 994 | { |
e6ef8a5c | 995 | struct p_header *h = tconn->data.rbuf; |
69bc7bc3 | 996 | int err; |
257d0af6 | 997 | |
a5c31904 AG |
998 | err = drbd_recv_all_warn(tconn, h, sizeof(*h)); |
999 | if (err) | |
69bc7bc3 | 1000 | return err; |
257d0af6 | 1001 | |
69bc7bc3 | 1002 | err = decode_header(tconn, h, pi); |
9ba7aa00 | 1003 | tconn->last_received = jiffies; |
b411b363 | 1004 | |
69bc7bc3 | 1005 | return err; |
b411b363 PR |
1006 | } |
1007 | ||
2451fc3b | 1008 | static void drbd_flush(struct drbd_conf *mdev) |
b411b363 PR |
1009 | { |
1010 | int rv; | |
1011 | ||
1012 | if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { | |
fbd9b09a | 1013 | rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL, |
dd3932ed | 1014 | NULL); |
b411b363 PR |
1015 | if (rv) { |
1016 | dev_err(DEV, "local disk flush failed with status %d\n", rv); | |
1017 | /* would rather check on EOPNOTSUPP, but that is not reliable. | |
1018 | * don't try again for ANY return value != 0 | |
1019 | * if (rv == -EOPNOTSUPP) */ | |
1020 | drbd_bump_write_ordering(mdev, WO_drain_io); | |
1021 | } | |
1022 | put_ldev(mdev); | |
1023 | } | |
b411b363 PR |
1024 | } |
1025 | ||
1026 | /** | |
1027 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
1028 | * @mdev: DRBD device. | |
1029 | * @epoch: Epoch object. | |
1030 | * @ev: Epoch event. | |
1031 | */ | |
1032 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, | |
1033 | struct drbd_epoch *epoch, | |
1034 | enum epoch_event ev) | |
1035 | { | |
2451fc3b | 1036 | int epoch_size; |
b411b363 | 1037 | struct drbd_epoch *next_epoch; |
b411b363 PR |
1038 | enum finish_epoch rv = FE_STILL_LIVE; |
1039 | ||
1040 | spin_lock(&mdev->epoch_lock); | |
1041 | do { | |
1042 | next_epoch = NULL; | |
b411b363 PR |
1043 | |
1044 | epoch_size = atomic_read(&epoch->epoch_size); | |
1045 | ||
1046 | switch (ev & ~EV_CLEANUP) { | |
1047 | case EV_PUT: | |
1048 | atomic_dec(&epoch->active); | |
1049 | break; | |
1050 | case EV_GOT_BARRIER_NR: | |
1051 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1052 | break; |
1053 | case EV_BECAME_LAST: | |
1054 | /* nothing to do*/ | |
1055 | break; | |
1056 | } | |
1057 | ||
b411b363 PR |
1058 | if (epoch_size != 0 && |
1059 | atomic_read(&epoch->active) == 0 && | |
2451fc3b | 1060 | test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) { |
b411b363 PR |
1061 | if (!(ev & EV_CLEANUP)) { |
1062 | spin_unlock(&mdev->epoch_lock); | |
1063 | drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); | |
1064 | spin_lock(&mdev->epoch_lock); | |
1065 | } | |
1066 | dec_unacked(mdev); | |
1067 | ||
1068 | if (mdev->current_epoch != epoch) { | |
1069 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); | |
1070 | list_del(&epoch->list); | |
1071 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
1072 | mdev->epochs--; | |
b411b363 PR |
1073 | kfree(epoch); |
1074 | ||
1075 | if (rv == FE_STILL_LIVE) | |
1076 | rv = FE_DESTROYED; | |
1077 | } else { | |
1078 | epoch->flags = 0; | |
1079 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1080 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1081 | if (rv == FE_STILL_LIVE) |
1082 | rv = FE_RECYCLED; | |
2451fc3b | 1083 | wake_up(&mdev->ee_wait); |
b411b363 PR |
1084 | } |
1085 | } | |
1086 | ||
1087 | if (!next_epoch) | |
1088 | break; | |
1089 | ||
1090 | epoch = next_epoch; | |
1091 | } while (1); | |
1092 | ||
1093 | spin_unlock(&mdev->epoch_lock); | |
1094 | ||
b411b363 PR |
1095 | return rv; |
1096 | } | |
1097 | ||
1098 | /** | |
1099 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
1100 | * @mdev: DRBD device. | |
1101 | * @wo: Write ordering method to try. | |
1102 | */ | |
1103 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) | |
1104 | { | |
1105 | enum write_ordering_e pwo; | |
1106 | static char *write_ordering_str[] = { | |
1107 | [WO_none] = "none", | |
1108 | [WO_drain_io] = "drain", | |
1109 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1110 | }; |
1111 | ||
1112 | pwo = mdev->write_ordering; | |
1113 | wo = min(pwo, wo); | |
b411b363 PR |
1114 | if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) |
1115 | wo = WO_drain_io; | |
1116 | if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) | |
1117 | wo = WO_none; | |
1118 | mdev->write_ordering = wo; | |
2451fc3b | 1119 | if (pwo != mdev->write_ordering || wo == WO_bdev_flush) |
b411b363 PR |
1120 | dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); |
1121 | } | |
1122 | ||
45bb912b | 1123 | /** |
fbe29dec | 1124 | * drbd_submit_peer_request() |
45bb912b | 1125 | * @mdev: DRBD device. |
db830c46 | 1126 | * @peer_req: peer request |
45bb912b | 1127 | * @rw: flag field, see bio->bi_rw |
10f6d992 LE |
1128 | * |
1129 | * May spread the pages to multiple bios, | |
1130 | * depending on bio_add_page restrictions. | |
1131 | * | |
1132 | * Returns 0 if all bios have been submitted, | |
1133 | * -ENOMEM if we could not allocate enough bios, | |
1134 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1135 | * single page to an empty bio (which should never happen and likely indicates | |
1136 | * that the lower level IO stack is in some way broken). This has been observed | |
1137 | * on certain Xen deployments. | |
45bb912b LE |
1138 | */ |
1139 | /* TODO allocate from our own bio_set. */ | |
fbe29dec AG |
1140 | int drbd_submit_peer_request(struct drbd_conf *mdev, |
1141 | struct drbd_peer_request *peer_req, | |
1142 | const unsigned rw, const int fault_type) | |
45bb912b LE |
1143 | { |
1144 | struct bio *bios = NULL; | |
1145 | struct bio *bio; | |
db830c46 AG |
1146 | struct page *page = peer_req->pages; |
1147 | sector_t sector = peer_req->i.sector; | |
1148 | unsigned ds = peer_req->i.size; | |
45bb912b LE |
1149 | unsigned n_bios = 0; |
1150 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
10f6d992 | 1151 | int err = -ENOMEM; |
45bb912b LE |
1152 | |
1153 | /* In most cases, we will only need one bio. But in case the lower | |
1154 | * level restrictions happen to be different at this offset on this | |
1155 | * side than those of the sending peer, we may need to submit the | |
da4a75d2 LE |
1156 | * request in more than one bio. |
1157 | * | |
1158 | * Plain bio_alloc is good enough here, this is no DRBD internally | |
1159 | * generated bio, but a bio allocated on behalf of the peer. | |
1160 | */ | |
45bb912b LE |
1161 | next_bio: |
1162 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1163 | if (!bio) { | |
1164 | dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); | |
1165 | goto fail; | |
1166 | } | |
db830c46 | 1167 | /* > peer_req->i.sector, unless this is the first bio */ |
45bb912b LE |
1168 | bio->bi_sector = sector; |
1169 | bio->bi_bdev = mdev->ldev->backing_bdev; | |
45bb912b | 1170 | bio->bi_rw = rw; |
db830c46 | 1171 | bio->bi_private = peer_req; |
fcefa62e | 1172 | bio->bi_end_io = drbd_peer_request_endio; |
45bb912b LE |
1173 | |
1174 | bio->bi_next = bios; | |
1175 | bios = bio; | |
1176 | ++n_bios; | |
1177 | ||
1178 | page_chain_for_each(page) { | |
1179 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1180 | if (!bio_add_page(bio, page, len, 0)) { | |
10f6d992 LE |
1181 | /* A single page must always be possible! |
1182 | * But in case it fails anyways, | |
1183 | * we deal with it, and complain (below). */ | |
1184 | if (bio->bi_vcnt == 0) { | |
1185 | dev_err(DEV, | |
1186 | "bio_add_page failed for len=%u, " | |
1187 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
1188 | len, (unsigned long long)bio->bi_sector); | |
1189 | err = -ENOSPC; | |
1190 | goto fail; | |
1191 | } | |
45bb912b LE |
1192 | goto next_bio; |
1193 | } | |
1194 | ds -= len; | |
1195 | sector += len >> 9; | |
1196 | --nr_pages; | |
1197 | } | |
1198 | D_ASSERT(page == NULL); | |
1199 | D_ASSERT(ds == 0); | |
1200 | ||
db830c46 | 1201 | atomic_set(&peer_req->pending_bios, n_bios); |
45bb912b LE |
1202 | do { |
1203 | bio = bios; | |
1204 | bios = bios->bi_next; | |
1205 | bio->bi_next = NULL; | |
1206 | ||
45bb912b | 1207 | drbd_generic_make_request(mdev, fault_type, bio); |
45bb912b | 1208 | } while (bios); |
45bb912b LE |
1209 | return 0; |
1210 | ||
1211 | fail: | |
1212 | while (bios) { | |
1213 | bio = bios; | |
1214 | bios = bios->bi_next; | |
1215 | bio_put(bio); | |
1216 | } | |
10f6d992 | 1217 | return err; |
45bb912b LE |
1218 | } |
1219 | ||
53840641 | 1220 | static void drbd_remove_epoch_entry_interval(struct drbd_conf *mdev, |
db830c46 | 1221 | struct drbd_peer_request *peer_req) |
53840641 | 1222 | { |
db830c46 | 1223 | struct drbd_interval *i = &peer_req->i; |
53840641 AG |
1224 | |
1225 | drbd_remove_interval(&mdev->write_requests, i); | |
1226 | drbd_clear_interval(i); | |
1227 | ||
6c852bec | 1228 | /* Wake up any processes waiting for this peer request to complete. */ |
53840641 AG |
1229 | if (i->waiting) |
1230 | wake_up(&mdev->misc_wait); | |
1231 | } | |
1232 | ||
4a76b161 | 1233 | static int receive_Barrier(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1234 | { |
4a76b161 | 1235 | struct drbd_conf *mdev; |
2451fc3b | 1236 | int rv; |
4a76b161 | 1237 | struct p_barrier *p = tconn->data.rbuf; |
b411b363 PR |
1238 | struct drbd_epoch *epoch; |
1239 | ||
4a76b161 AG |
1240 | mdev = vnr_to_mdev(tconn, pi->vnr); |
1241 | if (!mdev) | |
1242 | return -EIO; | |
1243 | ||
b411b363 PR |
1244 | inc_unacked(mdev); |
1245 | ||
b411b363 PR |
1246 | mdev->current_epoch->barrier_nr = p->barrier; |
1247 | rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); | |
1248 | ||
1249 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1250 | * the activity log, which means it would not be resynced in case the | |
1251 | * R_PRIMARY crashes now. | |
1252 | * Therefore we must send the barrier_ack after the barrier request was | |
1253 | * completed. */ | |
1254 | switch (mdev->write_ordering) { | |
b411b363 PR |
1255 | case WO_none: |
1256 | if (rv == FE_RECYCLED) | |
82bc0194 | 1257 | return 0; |
2451fc3b PR |
1258 | |
1259 | /* receiver context, in the writeout path of the other node. | |
1260 | * avoid potential distributed deadlock */ | |
1261 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1262 | if (epoch) | |
1263 | break; | |
1264 | else | |
1265 | dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); | |
1266 | /* Fall through */ | |
b411b363 PR |
1267 | |
1268 | case WO_bdev_flush: | |
1269 | case WO_drain_io: | |
b411b363 | 1270 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
2451fc3b PR |
1271 | drbd_flush(mdev); |
1272 | ||
1273 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1274 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1275 | if (epoch) | |
1276 | break; | |
b411b363 PR |
1277 | } |
1278 | ||
2451fc3b PR |
1279 | epoch = mdev->current_epoch; |
1280 | wait_event(mdev->ee_wait, atomic_read(&epoch->epoch_size) == 0); | |
1281 | ||
1282 | D_ASSERT(atomic_read(&epoch->active) == 0); | |
1283 | D_ASSERT(epoch->flags == 0); | |
b411b363 | 1284 | |
82bc0194 | 1285 | return 0; |
2451fc3b PR |
1286 | default: |
1287 | dev_err(DEV, "Strangeness in mdev->write_ordering %d\n", mdev->write_ordering); | |
82bc0194 | 1288 | return -EIO; |
b411b363 PR |
1289 | } |
1290 | ||
1291 | epoch->flags = 0; | |
1292 | atomic_set(&epoch->epoch_size, 0); | |
1293 | atomic_set(&epoch->active, 0); | |
1294 | ||
1295 | spin_lock(&mdev->epoch_lock); | |
1296 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1297 | list_add(&epoch->list, &mdev->current_epoch->list); | |
1298 | mdev->current_epoch = epoch; | |
1299 | mdev->epochs++; | |
b411b363 PR |
1300 | } else { |
1301 | /* The current_epoch got recycled while we allocated this one... */ | |
1302 | kfree(epoch); | |
1303 | } | |
1304 | spin_unlock(&mdev->epoch_lock); | |
1305 | ||
82bc0194 | 1306 | return 0; |
b411b363 PR |
1307 | } |
1308 | ||
1309 | /* used from receive_RSDataReply (recv_resync_read) | |
1310 | * and from receive_Data */ | |
f6ffca9f AG |
1311 | static struct drbd_peer_request * |
1312 | read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, | |
1313 | int data_size) __must_hold(local) | |
b411b363 | 1314 | { |
6666032a | 1315 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); |
db830c46 | 1316 | struct drbd_peer_request *peer_req; |
b411b363 | 1317 | struct page *page; |
a5c31904 | 1318 | int dgs, ds, err; |
a0638456 PR |
1319 | void *dig_in = mdev->tconn->int_dig_in; |
1320 | void *dig_vv = mdev->tconn->int_dig_vv; | |
6b4388ac | 1321 | unsigned long *data; |
b411b363 | 1322 | |
a0638456 PR |
1323 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ? |
1324 | crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0; | |
b411b363 PR |
1325 | |
1326 | if (dgs) { | |
a5c31904 AG |
1327 | err = drbd_recv_all_warn(mdev->tconn, dig_in, dgs); |
1328 | if (err) | |
b411b363 | 1329 | return NULL; |
b411b363 PR |
1330 | } |
1331 | ||
1332 | data_size -= dgs; | |
1333 | ||
841ce241 AG |
1334 | if (!expect(data_size != 0)) |
1335 | return NULL; | |
1336 | if (!expect(IS_ALIGNED(data_size, 512))) | |
1337 | return NULL; | |
1338 | if (!expect(data_size <= DRBD_MAX_BIO_SIZE)) | |
1339 | return NULL; | |
b411b363 | 1340 | |
6666032a LE |
1341 | /* even though we trust out peer, |
1342 | * we sometimes have to double check. */ | |
1343 | if (sector + (data_size>>9) > capacity) { | |
fdda6544 LE |
1344 | dev_err(DEV, "request from peer beyond end of local disk: " |
1345 | "capacity: %llus < sector: %llus + size: %u\n", | |
6666032a LE |
1346 | (unsigned long long)capacity, |
1347 | (unsigned long long)sector, data_size); | |
1348 | return NULL; | |
1349 | } | |
1350 | ||
b411b363 PR |
1351 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1352 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1353 | * which in turn might block on the other node at this very place. */ | |
db830c46 AG |
1354 | peer_req = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); |
1355 | if (!peer_req) | |
b411b363 | 1356 | return NULL; |
45bb912b | 1357 | |
b411b363 | 1358 | ds = data_size; |
db830c46 | 1359 | page = peer_req->pages; |
45bb912b LE |
1360 | page_chain_for_each(page) { |
1361 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1362 | data = kmap(page); |
a5c31904 | 1363 | err = drbd_recv_all_warn(mdev->tconn, data, len); |
0cf9d27e | 1364 | if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) { |
6b4388ac PR |
1365 | dev_err(DEV, "Fault injection: Corrupting data on receive\n"); |
1366 | data[0] = data[0] ^ (unsigned long)-1; | |
1367 | } | |
b411b363 | 1368 | kunmap(page); |
a5c31904 | 1369 | if (err) { |
db830c46 | 1370 | drbd_free_ee(mdev, peer_req); |
b411b363 PR |
1371 | return NULL; |
1372 | } | |
a5c31904 | 1373 | ds -= len; |
b411b363 PR |
1374 | } |
1375 | ||
1376 | if (dgs) { | |
db830c46 | 1377 | drbd_csum_ee(mdev, mdev->tconn->integrity_r_tfm, peer_req, dig_vv); |
b411b363 | 1378 | if (memcmp(dig_in, dig_vv, dgs)) { |
470be44a LE |
1379 | dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n", |
1380 | (unsigned long long)sector, data_size); | |
db830c46 | 1381 | drbd_free_ee(mdev, peer_req); |
b411b363 PR |
1382 | return NULL; |
1383 | } | |
1384 | } | |
1385 | mdev->recv_cnt += data_size>>9; | |
db830c46 | 1386 | return peer_req; |
b411b363 PR |
1387 | } |
1388 | ||
1389 | /* drbd_drain_block() just takes a data block | |
1390 | * out of the socket input buffer, and discards it. | |
1391 | */ | |
1392 | static int drbd_drain_block(struct drbd_conf *mdev, int data_size) | |
1393 | { | |
1394 | struct page *page; | |
a5c31904 | 1395 | int err = 0; |
b411b363 PR |
1396 | void *data; |
1397 | ||
c3470cde | 1398 | if (!data_size) |
fc5be839 | 1399 | return 0; |
c3470cde | 1400 | |
45bb912b | 1401 | page = drbd_pp_alloc(mdev, 1, 1); |
b411b363 PR |
1402 | |
1403 | data = kmap(page); | |
1404 | while (data_size) { | |
fc5be839 AG |
1405 | unsigned int len = min_t(int, data_size, PAGE_SIZE); |
1406 | ||
a5c31904 AG |
1407 | err = drbd_recv_all_warn(mdev->tconn, data, len); |
1408 | if (err) | |
b411b363 | 1409 | break; |
a5c31904 | 1410 | data_size -= len; |
b411b363 PR |
1411 | } |
1412 | kunmap(page); | |
435f0740 | 1413 | drbd_pp_free(mdev, page, 0); |
fc5be839 | 1414 | return err; |
b411b363 PR |
1415 | } |
1416 | ||
1417 | static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, | |
1418 | sector_t sector, int data_size) | |
1419 | { | |
1420 | struct bio_vec *bvec; | |
1421 | struct bio *bio; | |
a5c31904 | 1422 | int dgs, err, i, expect; |
a0638456 PR |
1423 | void *dig_in = mdev->tconn->int_dig_in; |
1424 | void *dig_vv = mdev->tconn->int_dig_vv; | |
b411b363 | 1425 | |
a0638456 PR |
1426 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ? |
1427 | crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0; | |
b411b363 PR |
1428 | |
1429 | if (dgs) { | |
a5c31904 AG |
1430 | err = drbd_recv_all_warn(mdev->tconn, dig_in, dgs); |
1431 | if (err) | |
1432 | return err; | |
b411b363 PR |
1433 | } |
1434 | ||
1435 | data_size -= dgs; | |
1436 | ||
1437 | /* optimistically update recv_cnt. if receiving fails below, | |
1438 | * we disconnect anyways, and counters will be reset. */ | |
1439 | mdev->recv_cnt += data_size>>9; | |
1440 | ||
1441 | bio = req->master_bio; | |
1442 | D_ASSERT(sector == bio->bi_sector); | |
1443 | ||
1444 | bio_for_each_segment(bvec, bio, i) { | |
a5c31904 | 1445 | void *mapped = kmap(bvec->bv_page) + bvec->bv_offset; |
b411b363 | 1446 | expect = min_t(int, data_size, bvec->bv_len); |
a5c31904 | 1447 | err = drbd_recv_all_warn(mdev->tconn, mapped, expect); |
b411b363 | 1448 | kunmap(bvec->bv_page); |
a5c31904 AG |
1449 | if (err) |
1450 | return err; | |
1451 | data_size -= expect; | |
b411b363 PR |
1452 | } |
1453 | ||
1454 | if (dgs) { | |
a0638456 | 1455 | drbd_csum_bio(mdev, mdev->tconn->integrity_r_tfm, bio, dig_vv); |
b411b363 PR |
1456 | if (memcmp(dig_in, dig_vv, dgs)) { |
1457 | dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); | |
28284cef | 1458 | return -EINVAL; |
b411b363 PR |
1459 | } |
1460 | } | |
1461 | ||
1462 | D_ASSERT(data_size == 0); | |
28284cef | 1463 | return 0; |
b411b363 PR |
1464 | } |
1465 | ||
1466 | /* e_end_resync_block() is called via | |
1467 | * drbd_process_done_ee() by asender only */ | |
99920dc5 | 1468 | static int e_end_resync_block(struct drbd_work *w, int unused) |
b411b363 | 1469 | { |
8050e6d0 AG |
1470 | struct drbd_peer_request *peer_req = |
1471 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1472 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1473 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1474 | int err; |
b411b363 | 1475 | |
db830c46 | 1476 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1477 | |
db830c46 AG |
1478 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
1479 | drbd_set_in_sync(mdev, sector, peer_req->i.size); | |
99920dc5 | 1480 | err = drbd_send_ack(mdev, P_RS_WRITE_ACK, peer_req); |
b411b363 PR |
1481 | } else { |
1482 | /* Record failure to sync */ | |
db830c46 | 1483 | drbd_rs_failed_io(mdev, sector, peer_req->i.size); |
b411b363 | 1484 | |
99920dc5 | 1485 | err = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1486 | } |
1487 | dec_unacked(mdev); | |
1488 | ||
99920dc5 | 1489 | return err; |
b411b363 PR |
1490 | } |
1491 | ||
1492 | static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) | |
1493 | { | |
db830c46 | 1494 | struct drbd_peer_request *peer_req; |
b411b363 | 1495 | |
db830c46 AG |
1496 | peer_req = read_in_block(mdev, ID_SYNCER, sector, data_size); |
1497 | if (!peer_req) | |
45bb912b | 1498 | goto fail; |
b411b363 PR |
1499 | |
1500 | dec_rs_pending(mdev); | |
1501 | ||
b411b363 PR |
1502 | inc_unacked(mdev); |
1503 | /* corresponding dec_unacked() in e_end_resync_block() | |
1504 | * respective _drbd_clear_done_ee */ | |
1505 | ||
db830c46 | 1506 | peer_req->w.cb = e_end_resync_block; |
45bb912b | 1507 | |
87eeee41 | 1508 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1509 | list_add(&peer_req->w.list, &mdev->sync_ee); |
87eeee41 | 1510 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1511 | |
0f0601f4 | 1512 | atomic_add(data_size >> 9, &mdev->rs_sect_ev); |
fbe29dec | 1513 | if (drbd_submit_peer_request(mdev, peer_req, WRITE, DRBD_FAULT_RS_WR) == 0) |
e1c1b0fc | 1514 | return 0; |
b411b363 | 1515 | |
10f6d992 LE |
1516 | /* don't care for the reason here */ |
1517 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 1518 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1519 | list_del(&peer_req->w.list); |
87eeee41 | 1520 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 1521 | |
db830c46 | 1522 | drbd_free_ee(mdev, peer_req); |
45bb912b LE |
1523 | fail: |
1524 | put_ldev(mdev); | |
e1c1b0fc | 1525 | return -EIO; |
b411b363 PR |
1526 | } |
1527 | ||
668eebc6 | 1528 | static struct drbd_request * |
bc9c5c41 AG |
1529 | find_request(struct drbd_conf *mdev, struct rb_root *root, u64 id, |
1530 | sector_t sector, bool missing_ok, const char *func) | |
51624585 | 1531 | { |
51624585 AG |
1532 | struct drbd_request *req; |
1533 | ||
bc9c5c41 AG |
1534 | /* Request object according to our peer */ |
1535 | req = (struct drbd_request *)(unsigned long)id; | |
5e472264 | 1536 | if (drbd_contains_interval(root, sector, &req->i) && req->i.local) |
668eebc6 | 1537 | return req; |
c3afd8f5 AG |
1538 | if (!missing_ok) { |
1539 | dev_err(DEV, "%s: failed to find request %lu, sector %llus\n", func, | |
1540 | (unsigned long)id, (unsigned long long)sector); | |
1541 | } | |
51624585 AG |
1542 | return NULL; |
1543 | } | |
1544 | ||
4a76b161 | 1545 | static int receive_DataReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1546 | { |
4a76b161 | 1547 | struct drbd_conf *mdev; |
b411b363 PR |
1548 | struct drbd_request *req; |
1549 | sector_t sector; | |
82bc0194 | 1550 | int err; |
4a76b161 AG |
1551 | struct p_data *p = tconn->data.rbuf; |
1552 | ||
1553 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
1554 | if (!mdev) | |
1555 | return -EIO; | |
b411b363 PR |
1556 | |
1557 | sector = be64_to_cpu(p->sector); | |
1558 | ||
87eeee41 | 1559 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 1560 | req = find_request(mdev, &mdev->read_requests, p->block_id, sector, false, __func__); |
87eeee41 | 1561 | spin_unlock_irq(&mdev->tconn->req_lock); |
c3afd8f5 | 1562 | if (unlikely(!req)) |
82bc0194 | 1563 | return -EIO; |
b411b363 | 1564 | |
24c4830c | 1565 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1566 | * special casing it there for the various failure cases. |
1567 | * still no race with drbd_fail_pending_reads */ | |
e2857216 | 1568 | err = recv_dless_read(mdev, req, sector, pi->size); |
82bc0194 | 1569 | if (!err) |
8554df1c | 1570 | req_mod(req, DATA_RECEIVED); |
b411b363 PR |
1571 | /* else: nothing. handled from drbd_disconnect... |
1572 | * I don't think we may complete this just yet | |
1573 | * in case we are "on-disconnect: freeze" */ | |
1574 | ||
82bc0194 | 1575 | return err; |
b411b363 PR |
1576 | } |
1577 | ||
4a76b161 | 1578 | static int receive_RSDataReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1579 | { |
4a76b161 | 1580 | struct drbd_conf *mdev; |
b411b363 | 1581 | sector_t sector; |
82bc0194 | 1582 | int err; |
4a76b161 AG |
1583 | struct p_data *p = tconn->data.rbuf; |
1584 | ||
1585 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
1586 | if (!mdev) | |
1587 | return -EIO; | |
b411b363 PR |
1588 | |
1589 | sector = be64_to_cpu(p->sector); | |
1590 | D_ASSERT(p->block_id == ID_SYNCER); | |
1591 | ||
1592 | if (get_ldev(mdev)) { | |
1593 | /* data is submitted to disk within recv_resync_read. | |
1594 | * corresponding put_ldev done below on error, | |
fcefa62e | 1595 | * or in drbd_peer_request_endio. */ |
e2857216 | 1596 | err = recv_resync_read(mdev, sector, pi->size); |
b411b363 PR |
1597 | } else { |
1598 | if (__ratelimit(&drbd_ratelimit_state)) | |
1599 | dev_err(DEV, "Can not write resync data to local disk.\n"); | |
1600 | ||
e2857216 | 1601 | err = drbd_drain_block(mdev, pi->size); |
b411b363 | 1602 | |
e2857216 | 1603 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, pi->size); |
b411b363 PR |
1604 | } |
1605 | ||
e2857216 | 1606 | atomic_add(pi->size >> 9, &mdev->rs_sect_in); |
778f271d | 1607 | |
82bc0194 | 1608 | return err; |
b411b363 PR |
1609 | } |
1610 | ||
99920dc5 | 1611 | static int w_restart_write(struct drbd_work *w, int cancel) |
7be8da07 AG |
1612 | { |
1613 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
1614 | struct drbd_conf *mdev = w->mdev; | |
1615 | struct bio *bio; | |
1616 | unsigned long start_time; | |
1617 | unsigned long flags; | |
1618 | ||
1619 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
1620 | if (!expect(req->rq_state & RQ_POSTPONED)) { | |
1621 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
99920dc5 | 1622 | return -EIO; |
7be8da07 AG |
1623 | } |
1624 | bio = req->master_bio; | |
1625 | start_time = req->start_time; | |
1626 | /* Postponed requests will not have their master_bio completed! */ | |
1627 | __req_mod(req, DISCARD_WRITE, NULL); | |
1628 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
1629 | ||
1630 | while (__drbd_make_request(mdev, bio, start_time)) | |
1631 | /* retry */ ; | |
99920dc5 | 1632 | return 0; |
7be8da07 AG |
1633 | } |
1634 | ||
1635 | static void restart_conflicting_writes(struct drbd_conf *mdev, | |
1636 | sector_t sector, int size) | |
1637 | { | |
1638 | struct drbd_interval *i; | |
1639 | struct drbd_request *req; | |
1640 | ||
1641 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1642 | if (!i->local) | |
1643 | continue; | |
1644 | req = container_of(i, struct drbd_request, i); | |
1645 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1646 | !(req->rq_state & RQ_POSTPONED)) | |
1647 | continue; | |
1648 | if (expect(list_empty(&req->w.list))) { | |
1649 | req->w.mdev = mdev; | |
1650 | req->w.cb = w_restart_write; | |
1651 | drbd_queue_work(&mdev->tconn->data.work, &req->w); | |
1652 | } | |
1653 | } | |
1654 | } | |
1655 | ||
b411b363 PR |
1656 | /* e_end_block() is called via drbd_process_done_ee(). |
1657 | * this means this function only runs in the asender thread | |
1658 | */ | |
99920dc5 | 1659 | static int e_end_block(struct drbd_work *w, int cancel) |
b411b363 | 1660 | { |
8050e6d0 AG |
1661 | struct drbd_peer_request *peer_req = |
1662 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1663 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1664 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1665 | int err = 0, pcmd; |
b411b363 | 1666 | |
89e58e75 | 1667 | if (mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C) { |
db830c46 | 1668 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1669 | pcmd = (mdev->state.conn >= C_SYNC_SOURCE && |
1670 | mdev->state.conn <= C_PAUSED_SYNC_T && | |
db830c46 | 1671 | peer_req->flags & EE_MAY_SET_IN_SYNC) ? |
b411b363 | 1672 | P_RS_WRITE_ACK : P_WRITE_ACK; |
99920dc5 | 1673 | err = drbd_send_ack(mdev, pcmd, peer_req); |
b411b363 | 1674 | if (pcmd == P_RS_WRITE_ACK) |
db830c46 | 1675 | drbd_set_in_sync(mdev, sector, peer_req->i.size); |
b411b363 | 1676 | } else { |
99920dc5 | 1677 | err = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1678 | /* we expect it to be marked out of sync anyways... |
1679 | * maybe assert this? */ | |
1680 | } | |
1681 | dec_unacked(mdev); | |
1682 | } | |
1683 | /* we delete from the conflict detection hash _after_ we sent out the | |
1684 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
89e58e75 | 1685 | if (mdev->tconn->net_conf->two_primaries) { |
87eeee41 | 1686 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
1687 | D_ASSERT(!drbd_interval_empty(&peer_req->i)); |
1688 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
7be8da07 AG |
1689 | if (peer_req->flags & EE_RESTART_REQUESTS) |
1690 | restart_conflicting_writes(mdev, sector, peer_req->i.size); | |
87eeee41 | 1691 | spin_unlock_irq(&mdev->tconn->req_lock); |
bb3bfe96 | 1692 | } else |
db830c46 | 1693 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1694 | |
db830c46 | 1695 | drbd_may_finish_epoch(mdev, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); |
b411b363 | 1696 | |
99920dc5 | 1697 | return err; |
b411b363 PR |
1698 | } |
1699 | ||
7be8da07 | 1700 | static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) |
b411b363 | 1701 | { |
7be8da07 | 1702 | struct drbd_conf *mdev = w->mdev; |
8050e6d0 AG |
1703 | struct drbd_peer_request *peer_req = |
1704 | container_of(w, struct drbd_peer_request, w); | |
99920dc5 | 1705 | int err; |
b411b363 | 1706 | |
99920dc5 | 1707 | err = drbd_send_ack(mdev, ack, peer_req); |
b411b363 PR |
1708 | dec_unacked(mdev); |
1709 | ||
99920dc5 | 1710 | return err; |
b411b363 PR |
1711 | } |
1712 | ||
99920dc5 | 1713 | static int e_send_discard_write(struct drbd_work *w, int unused) |
7be8da07 AG |
1714 | { |
1715 | return e_send_ack(w, P_DISCARD_WRITE); | |
1716 | } | |
1717 | ||
99920dc5 | 1718 | static int e_send_retry_write(struct drbd_work *w, int unused) |
7be8da07 AG |
1719 | { |
1720 | struct drbd_tconn *tconn = w->mdev->tconn; | |
1721 | ||
1722 | return e_send_ack(w, tconn->agreed_pro_version >= 100 ? | |
1723 | P_RETRY_WRITE : P_DISCARD_WRITE); | |
1724 | } | |
1725 | ||
3e394da1 AG |
1726 | static bool seq_greater(u32 a, u32 b) |
1727 | { | |
1728 | /* | |
1729 | * We assume 32-bit wrap-around here. | |
1730 | * For 24-bit wrap-around, we would have to shift: | |
1731 | * a <<= 8; b <<= 8; | |
1732 | */ | |
1733 | return (s32)a - (s32)b > 0; | |
1734 | } | |
1735 | ||
1736 | static u32 seq_max(u32 a, u32 b) | |
1737 | { | |
1738 | return seq_greater(a, b) ? a : b; | |
1739 | } | |
1740 | ||
7be8da07 AG |
1741 | static bool need_peer_seq(struct drbd_conf *mdev) |
1742 | { | |
1743 | struct drbd_tconn *tconn = mdev->tconn; | |
1744 | ||
1745 | /* | |
1746 | * We only need to keep track of the last packet_seq number of our peer | |
1747 | * if we are in dual-primary mode and we have the discard flag set; see | |
1748 | * handle_write_conflicts(). | |
1749 | */ | |
1750 | return tconn->net_conf->two_primaries && | |
1751 | test_bit(DISCARD_CONCURRENT, &tconn->flags); | |
1752 | } | |
1753 | ||
43ae077d | 1754 | static void update_peer_seq(struct drbd_conf *mdev, unsigned int peer_seq) |
3e394da1 | 1755 | { |
3c13b680 | 1756 | unsigned int newest_peer_seq; |
3e394da1 | 1757 | |
7be8da07 AG |
1758 | if (need_peer_seq(mdev)) { |
1759 | spin_lock(&mdev->peer_seq_lock); | |
3c13b680 LE |
1760 | newest_peer_seq = seq_max(mdev->peer_seq, peer_seq); |
1761 | mdev->peer_seq = newest_peer_seq; | |
7be8da07 | 1762 | spin_unlock(&mdev->peer_seq_lock); |
3c13b680 LE |
1763 | /* wake up only if we actually changed mdev->peer_seq */ |
1764 | if (peer_seq == newest_peer_seq) | |
7be8da07 AG |
1765 | wake_up(&mdev->seq_wait); |
1766 | } | |
3e394da1 AG |
1767 | } |
1768 | ||
b411b363 PR |
1769 | /* Called from receive_Data. |
1770 | * Synchronize packets on sock with packets on msock. | |
1771 | * | |
1772 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
1773 | * packet traveling on msock, they are still processed in the order they have | |
1774 | * been sent. | |
1775 | * | |
1776 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
1777 | * | |
1778 | * In case packet_seq is larger than mdev->peer_seq number, there are | |
1779 | * outstanding packets on the msock. We wait for them to arrive. | |
1780 | * In case we are the logically next packet, we update mdev->peer_seq | |
1781 | * ourselves. Correctly handles 32bit wrap around. | |
1782 | * | |
1783 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
1784 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
1785 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
1786 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
1787 | * | |
1788 | * returns 0 if we may process the packet, | |
1789 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
7be8da07 | 1790 | static int wait_for_and_update_peer_seq(struct drbd_conf *mdev, const u32 peer_seq) |
b411b363 PR |
1791 | { |
1792 | DEFINE_WAIT(wait); | |
b411b363 | 1793 | long timeout; |
7be8da07 AG |
1794 | int ret; |
1795 | ||
1796 | if (!need_peer_seq(mdev)) | |
1797 | return 0; | |
1798 | ||
b411b363 PR |
1799 | spin_lock(&mdev->peer_seq_lock); |
1800 | for (;;) { | |
7be8da07 AG |
1801 | if (!seq_greater(peer_seq - 1, mdev->peer_seq)) { |
1802 | mdev->peer_seq = seq_max(mdev->peer_seq, peer_seq); | |
1803 | ret = 0; | |
b411b363 | 1804 | break; |
7be8da07 | 1805 | } |
b411b363 PR |
1806 | if (signal_pending(current)) { |
1807 | ret = -ERESTARTSYS; | |
1808 | break; | |
1809 | } | |
7be8da07 | 1810 | prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); |
b411b363 | 1811 | spin_unlock(&mdev->peer_seq_lock); |
71b1c1eb AG |
1812 | timeout = mdev->tconn->net_conf->ping_timeo*HZ/10; |
1813 | timeout = schedule_timeout(timeout); | |
b411b363 | 1814 | spin_lock(&mdev->peer_seq_lock); |
7be8da07 | 1815 | if (!timeout) { |
b411b363 | 1816 | ret = -ETIMEDOUT; |
71b1c1eb | 1817 | dev_err(DEV, "Timed out waiting for missing ack packets; disconnecting\n"); |
b411b363 PR |
1818 | break; |
1819 | } | |
1820 | } | |
b411b363 | 1821 | spin_unlock(&mdev->peer_seq_lock); |
7be8da07 | 1822 | finish_wait(&mdev->seq_wait, &wait); |
b411b363 PR |
1823 | return ret; |
1824 | } | |
1825 | ||
688593c5 LE |
1826 | /* see also bio_flags_to_wire() |
1827 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
1828 | * flags and back. We may replicate to other kernel versions. */ | |
1829 | static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf) | |
76d2e7ec | 1830 | { |
688593c5 LE |
1831 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
1832 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
1833 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
1834 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
1835 | } |
1836 | ||
7be8da07 AG |
1837 | static void fail_postponed_requests(struct drbd_conf *mdev, sector_t sector, |
1838 | unsigned int size) | |
1839 | { | |
1840 | struct drbd_interval *i; | |
1841 | ||
1842 | repeat: | |
1843 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1844 | struct drbd_request *req; | |
1845 | struct bio_and_error m; | |
1846 | ||
1847 | if (!i->local) | |
1848 | continue; | |
1849 | req = container_of(i, struct drbd_request, i); | |
1850 | if (!(req->rq_state & RQ_POSTPONED)) | |
1851 | continue; | |
1852 | req->rq_state &= ~RQ_POSTPONED; | |
1853 | __req_mod(req, NEG_ACKED, &m); | |
1854 | spin_unlock_irq(&mdev->tconn->req_lock); | |
1855 | if (m.bio) | |
1856 | complete_master_bio(mdev, &m); | |
1857 | spin_lock_irq(&mdev->tconn->req_lock); | |
1858 | goto repeat; | |
1859 | } | |
1860 | } | |
1861 | ||
1862 | static int handle_write_conflicts(struct drbd_conf *mdev, | |
1863 | struct drbd_peer_request *peer_req) | |
1864 | { | |
1865 | struct drbd_tconn *tconn = mdev->tconn; | |
1866 | bool resolve_conflicts = test_bit(DISCARD_CONCURRENT, &tconn->flags); | |
1867 | sector_t sector = peer_req->i.sector; | |
1868 | const unsigned int size = peer_req->i.size; | |
1869 | struct drbd_interval *i; | |
1870 | bool equal; | |
1871 | int err; | |
1872 | ||
1873 | /* | |
1874 | * Inserting the peer request into the write_requests tree will prevent | |
1875 | * new conflicting local requests from being added. | |
1876 | */ | |
1877 | drbd_insert_interval(&mdev->write_requests, &peer_req->i); | |
1878 | ||
1879 | repeat: | |
1880 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1881 | if (i == &peer_req->i) | |
1882 | continue; | |
1883 | ||
1884 | if (!i->local) { | |
1885 | /* | |
1886 | * Our peer has sent a conflicting remote request; this | |
1887 | * should not happen in a two-node setup. Wait for the | |
1888 | * earlier peer request to complete. | |
1889 | */ | |
1890 | err = drbd_wait_misc(mdev, i); | |
1891 | if (err) | |
1892 | goto out; | |
1893 | goto repeat; | |
1894 | } | |
1895 | ||
1896 | equal = i->sector == sector && i->size == size; | |
1897 | if (resolve_conflicts) { | |
1898 | /* | |
1899 | * If the peer request is fully contained within the | |
1900 | * overlapping request, it can be discarded; otherwise, | |
1901 | * it will be retried once all overlapping requests | |
1902 | * have completed. | |
1903 | */ | |
1904 | bool discard = i->sector <= sector && i->sector + | |
1905 | (i->size >> 9) >= sector + (size >> 9); | |
1906 | ||
1907 | if (!equal) | |
1908 | dev_alert(DEV, "Concurrent writes detected: " | |
1909 | "local=%llus +%u, remote=%llus +%u, " | |
1910 | "assuming %s came first\n", | |
1911 | (unsigned long long)i->sector, i->size, | |
1912 | (unsigned long long)sector, size, | |
1913 | discard ? "local" : "remote"); | |
1914 | ||
1915 | inc_unacked(mdev); | |
1916 | peer_req->w.cb = discard ? e_send_discard_write : | |
1917 | e_send_retry_write; | |
1918 | list_add_tail(&peer_req->w.list, &mdev->done_ee); | |
1919 | wake_asender(mdev->tconn); | |
1920 | ||
1921 | err = -ENOENT; | |
1922 | goto out; | |
1923 | } else { | |
1924 | struct drbd_request *req = | |
1925 | container_of(i, struct drbd_request, i); | |
1926 | ||
1927 | if (!equal) | |
1928 | dev_alert(DEV, "Concurrent writes detected: " | |
1929 | "local=%llus +%u, remote=%llus +%u\n", | |
1930 | (unsigned long long)i->sector, i->size, | |
1931 | (unsigned long long)sector, size); | |
1932 | ||
1933 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1934 | !(req->rq_state & RQ_POSTPONED)) { | |
1935 | /* | |
1936 | * Wait for the node with the discard flag to | |
1937 | * decide if this request will be discarded or | |
1938 | * retried. Requests that are discarded will | |
1939 | * disappear from the write_requests tree. | |
1940 | * | |
1941 | * In addition, wait for the conflicting | |
1942 | * request to finish locally before submitting | |
1943 | * the conflicting peer request. | |
1944 | */ | |
1945 | err = drbd_wait_misc(mdev, &req->i); | |
1946 | if (err) { | |
1947 | _conn_request_state(mdev->tconn, | |
1948 | NS(conn, C_TIMEOUT), | |
1949 | CS_HARD); | |
1950 | fail_postponed_requests(mdev, sector, size); | |
1951 | goto out; | |
1952 | } | |
1953 | goto repeat; | |
1954 | } | |
1955 | /* | |
1956 | * Remember to restart the conflicting requests after | |
1957 | * the new peer request has completed. | |
1958 | */ | |
1959 | peer_req->flags |= EE_RESTART_REQUESTS; | |
1960 | } | |
1961 | } | |
1962 | err = 0; | |
1963 | ||
1964 | out: | |
1965 | if (err) | |
1966 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
1967 | return err; | |
1968 | } | |
1969 | ||
b411b363 | 1970 | /* mirrored write */ |
4a76b161 | 1971 | static int receive_Data(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 1972 | { |
4a76b161 | 1973 | struct drbd_conf *mdev; |
b411b363 | 1974 | sector_t sector; |
db830c46 | 1975 | struct drbd_peer_request *peer_req; |
4a76b161 | 1976 | struct p_data *p = tconn->data.rbuf; |
7be8da07 | 1977 | u32 peer_seq = be32_to_cpu(p->seq_num); |
b411b363 PR |
1978 | int rw = WRITE; |
1979 | u32 dp_flags; | |
7be8da07 | 1980 | int err; |
b411b363 | 1981 | |
4a76b161 AG |
1982 | mdev = vnr_to_mdev(tconn, pi->vnr); |
1983 | if (!mdev) | |
1984 | return -EIO; | |
1985 | ||
7be8da07 | 1986 | if (!get_ldev(mdev)) { |
82bc0194 AG |
1987 | int err2; |
1988 | ||
7be8da07 | 1989 | err = wait_for_and_update_peer_seq(mdev, peer_seq); |
e2857216 | 1990 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, pi->size); |
b411b363 | 1991 | atomic_inc(&mdev->current_epoch->epoch_size); |
e2857216 | 1992 | err2 = drbd_drain_block(mdev, pi->size); |
82bc0194 AG |
1993 | if (!err) |
1994 | err = err2; | |
1995 | return err; | |
b411b363 PR |
1996 | } |
1997 | ||
fcefa62e AG |
1998 | /* |
1999 | * Corresponding put_ldev done either below (on various errors), or in | |
2000 | * drbd_peer_request_endio, if we successfully submit the data at the | |
2001 | * end of this function. | |
2002 | */ | |
b411b363 PR |
2003 | |
2004 | sector = be64_to_cpu(p->sector); | |
e2857216 | 2005 | peer_req = read_in_block(mdev, p->block_id, sector, pi->size); |
db830c46 | 2006 | if (!peer_req) { |
b411b363 | 2007 | put_ldev(mdev); |
82bc0194 | 2008 | return -EIO; |
b411b363 PR |
2009 | } |
2010 | ||
db830c46 | 2011 | peer_req->w.cb = e_end_block; |
b411b363 | 2012 | |
688593c5 LE |
2013 | dp_flags = be32_to_cpu(p->dp_flags); |
2014 | rw |= wire_flags_to_bio(mdev, dp_flags); | |
2015 | ||
2016 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
db830c46 | 2017 | peer_req->flags |= EE_MAY_SET_IN_SYNC; |
688593c5 | 2018 | |
b411b363 | 2019 | spin_lock(&mdev->epoch_lock); |
db830c46 AG |
2020 | peer_req->epoch = mdev->current_epoch; |
2021 | atomic_inc(&peer_req->epoch->epoch_size); | |
2022 | atomic_inc(&peer_req->epoch->active); | |
b411b363 PR |
2023 | spin_unlock(&mdev->epoch_lock); |
2024 | ||
7be8da07 AG |
2025 | if (mdev->tconn->net_conf->two_primaries) { |
2026 | err = wait_for_and_update_peer_seq(mdev, peer_seq); | |
2027 | if (err) | |
b411b363 | 2028 | goto out_interrupted; |
87eeee41 | 2029 | spin_lock_irq(&mdev->tconn->req_lock); |
7be8da07 AG |
2030 | err = handle_write_conflicts(mdev, peer_req); |
2031 | if (err) { | |
2032 | spin_unlock_irq(&mdev->tconn->req_lock); | |
2033 | if (err == -ENOENT) { | |
b411b363 | 2034 | put_ldev(mdev); |
82bc0194 | 2035 | return 0; |
b411b363 | 2036 | } |
7be8da07 | 2037 | goto out_interrupted; |
b411b363 | 2038 | } |
7be8da07 AG |
2039 | } else |
2040 | spin_lock_irq(&mdev->tconn->req_lock); | |
db830c46 | 2041 | list_add(&peer_req->w.list, &mdev->active_ee); |
87eeee41 | 2042 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2043 | |
89e58e75 | 2044 | switch (mdev->tconn->net_conf->wire_protocol) { |
b411b363 PR |
2045 | case DRBD_PROT_C: |
2046 | inc_unacked(mdev); | |
2047 | /* corresponding dec_unacked() in e_end_block() | |
2048 | * respective _drbd_clear_done_ee */ | |
2049 | break; | |
2050 | case DRBD_PROT_B: | |
2051 | /* I really don't like it that the receiver thread | |
2052 | * sends on the msock, but anyways */ | |
db830c46 | 2053 | drbd_send_ack(mdev, P_RECV_ACK, peer_req); |
b411b363 PR |
2054 | break; |
2055 | case DRBD_PROT_A: | |
2056 | /* nothing to do */ | |
2057 | break; | |
2058 | } | |
2059 | ||
6719fb03 | 2060 | if (mdev->state.pdsk < D_INCONSISTENT) { |
b411b363 | 2061 | /* In case we have the only disk of the cluster, */ |
db830c46 AG |
2062 | drbd_set_out_of_sync(mdev, peer_req->i.sector, peer_req->i.size); |
2063 | peer_req->flags |= EE_CALL_AL_COMPLETE_IO; | |
2064 | peer_req->flags &= ~EE_MAY_SET_IN_SYNC; | |
181286ad | 2065 | drbd_al_begin_io(mdev, &peer_req->i); |
b411b363 PR |
2066 | } |
2067 | ||
82bc0194 AG |
2068 | err = drbd_submit_peer_request(mdev, peer_req, rw, DRBD_FAULT_DT_WR); |
2069 | if (!err) | |
2070 | return 0; | |
b411b363 | 2071 | |
10f6d992 LE |
2072 | /* don't care for the reason here */ |
2073 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2074 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
2075 | list_del(&peer_req->w.list); |
2076 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
87eeee41 | 2077 | spin_unlock_irq(&mdev->tconn->req_lock); |
db830c46 | 2078 | if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) |
181286ad | 2079 | drbd_al_complete_io(mdev, &peer_req->i); |
22cc37a9 | 2080 | |
b411b363 | 2081 | out_interrupted: |
db830c46 | 2082 | drbd_may_finish_epoch(mdev, peer_req->epoch, EV_PUT + EV_CLEANUP); |
b411b363 | 2083 | put_ldev(mdev); |
db830c46 | 2084 | drbd_free_ee(mdev, peer_req); |
82bc0194 | 2085 | return err; |
b411b363 PR |
2086 | } |
2087 | ||
0f0601f4 LE |
2088 | /* We may throttle resync, if the lower device seems to be busy, |
2089 | * and current sync rate is above c_min_rate. | |
2090 | * | |
2091 | * To decide whether or not the lower device is busy, we use a scheme similar | |
2092 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
2093 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
2094 | * activity, it obviously is "busy". | |
2095 | * | |
2096 | * The current sync rate used here uses only the most recent two step marks, | |
2097 | * to have a short time average so we can react faster. | |
2098 | */ | |
e3555d85 | 2099 | int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector) |
0f0601f4 LE |
2100 | { |
2101 | struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; | |
2102 | unsigned long db, dt, dbdt; | |
e3555d85 | 2103 | struct lc_element *tmp; |
0f0601f4 LE |
2104 | int curr_events; |
2105 | int throttle = 0; | |
2106 | ||
2107 | /* feature disabled? */ | |
f399002e | 2108 | if (mdev->ldev->dc.c_min_rate == 0) |
0f0601f4 LE |
2109 | return 0; |
2110 | ||
e3555d85 PR |
2111 | spin_lock_irq(&mdev->al_lock); |
2112 | tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector)); | |
2113 | if (tmp) { | |
2114 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
2115 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) { | |
2116 | spin_unlock_irq(&mdev->al_lock); | |
2117 | return 0; | |
2118 | } | |
2119 | /* Do not slow down if app IO is already waiting for this extent */ | |
2120 | } | |
2121 | spin_unlock_irq(&mdev->al_lock); | |
2122 | ||
0f0601f4 LE |
2123 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
2124 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
2125 | atomic_read(&mdev->rs_sect_ev); | |
e3555d85 | 2126 | |
0f0601f4 LE |
2127 | if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { |
2128 | unsigned long rs_left; | |
2129 | int i; | |
2130 | ||
2131 | mdev->rs_last_events = curr_events; | |
2132 | ||
2133 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
2134 | * approx. */ | |
2649f080 LE |
2135 | i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
2136 | ||
2137 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) | |
2138 | rs_left = mdev->ov_left; | |
2139 | else | |
2140 | rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
0f0601f4 LE |
2141 | |
2142 | dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; | |
2143 | if (!dt) | |
2144 | dt++; | |
2145 | db = mdev->rs_mark_left[i] - rs_left; | |
2146 | dbdt = Bit2KB(db/dt); | |
2147 | ||
f399002e | 2148 | if (dbdt > mdev->ldev->dc.c_min_rate) |
0f0601f4 LE |
2149 | throttle = 1; |
2150 | } | |
2151 | return throttle; | |
2152 | } | |
2153 | ||
2154 | ||
4a76b161 | 2155 | static int receive_DataRequest(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 2156 | { |
4a76b161 | 2157 | struct drbd_conf *mdev; |
b411b363 | 2158 | sector_t sector; |
4a76b161 | 2159 | sector_t capacity; |
db830c46 | 2160 | struct drbd_peer_request *peer_req; |
b411b363 | 2161 | struct digest_info *di = NULL; |
b18b37be | 2162 | int size, verb; |
b411b363 | 2163 | unsigned int fault_type; |
4a76b161 AG |
2164 | struct p_block_req *p = tconn->data.rbuf; |
2165 | ||
2166 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
2167 | if (!mdev) | |
2168 | return -EIO; | |
2169 | capacity = drbd_get_capacity(mdev->this_bdev); | |
b411b363 PR |
2170 | |
2171 | sector = be64_to_cpu(p->sector); | |
2172 | size = be32_to_cpu(p->blksize); | |
2173 | ||
c670a398 | 2174 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
2175 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
2176 | (unsigned long long)sector, size); | |
82bc0194 | 2177 | return -EINVAL; |
b411b363 PR |
2178 | } |
2179 | if (sector + (size>>9) > capacity) { | |
2180 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, | |
2181 | (unsigned long long)sector, size); | |
82bc0194 | 2182 | return -EINVAL; |
b411b363 PR |
2183 | } |
2184 | ||
2185 | if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { | |
b18b37be | 2186 | verb = 1; |
e2857216 | 2187 | switch (pi->cmd) { |
b18b37be PR |
2188 | case P_DATA_REQUEST: |
2189 | drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); | |
2190 | break; | |
2191 | case P_RS_DATA_REQUEST: | |
2192 | case P_CSUM_RS_REQUEST: | |
2193 | case P_OV_REQUEST: | |
2194 | drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); | |
2195 | break; | |
2196 | case P_OV_REPLY: | |
2197 | verb = 0; | |
2198 | dec_rs_pending(mdev); | |
2199 | drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); | |
2200 | break; | |
2201 | default: | |
49ba9b1b | 2202 | BUG(); |
b18b37be PR |
2203 | } |
2204 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
b411b363 PR |
2205 | dev_err(DEV, "Can not satisfy peer's read request, " |
2206 | "no local data.\n"); | |
b18b37be | 2207 | |
a821cc4a | 2208 | /* drain possibly payload */ |
e2857216 | 2209 | return drbd_drain_block(mdev, pi->size); |
b411b363 PR |
2210 | } |
2211 | ||
2212 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2213 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2214 | * which in turn might block on the other node at this very place. */ | |
db830c46 AG |
2215 | peer_req = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); |
2216 | if (!peer_req) { | |
b411b363 | 2217 | put_ldev(mdev); |
82bc0194 | 2218 | return -ENOMEM; |
b411b363 PR |
2219 | } |
2220 | ||
e2857216 | 2221 | switch (pi->cmd) { |
b411b363 | 2222 | case P_DATA_REQUEST: |
db830c46 | 2223 | peer_req->w.cb = w_e_end_data_req; |
b411b363 | 2224 | fault_type = DRBD_FAULT_DT_RD; |
80a40e43 LE |
2225 | /* application IO, don't drbd_rs_begin_io */ |
2226 | goto submit; | |
2227 | ||
b411b363 | 2228 | case P_RS_DATA_REQUEST: |
db830c46 | 2229 | peer_req->w.cb = w_e_end_rsdata_req; |
b411b363 | 2230 | fault_type = DRBD_FAULT_RS_RD; |
5f9915bb LE |
2231 | /* used in the sector offset progress display */ |
2232 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
b411b363 PR |
2233 | break; |
2234 | ||
2235 | case P_OV_REPLY: | |
2236 | case P_CSUM_RS_REQUEST: | |
2237 | fault_type = DRBD_FAULT_RS_RD; | |
e2857216 | 2238 | di = kmalloc(sizeof(*di) + pi->size, GFP_NOIO); |
b411b363 PR |
2239 | if (!di) |
2240 | goto out_free_e; | |
2241 | ||
e2857216 | 2242 | di->digest_size = pi->size; |
b411b363 PR |
2243 | di->digest = (((char *)di)+sizeof(struct digest_info)); |
2244 | ||
db830c46 AG |
2245 | peer_req->digest = di; |
2246 | peer_req->flags |= EE_HAS_DIGEST; | |
c36c3ced | 2247 | |
e2857216 | 2248 | if (drbd_recv_all(mdev->tconn, di->digest, pi->size)) |
b411b363 PR |
2249 | goto out_free_e; |
2250 | ||
e2857216 | 2251 | if (pi->cmd == P_CSUM_RS_REQUEST) { |
31890f4a | 2252 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
db830c46 | 2253 | peer_req->w.cb = w_e_end_csum_rs_req; |
5f9915bb LE |
2254 | /* used in the sector offset progress display */ |
2255 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
e2857216 | 2256 | } else if (pi->cmd == P_OV_REPLY) { |
2649f080 LE |
2257 | /* track progress, we may need to throttle */ |
2258 | atomic_add(size >> 9, &mdev->rs_sect_in); | |
db830c46 | 2259 | peer_req->w.cb = w_e_end_ov_reply; |
b411b363 | 2260 | dec_rs_pending(mdev); |
0f0601f4 LE |
2261 | /* drbd_rs_begin_io done when we sent this request, |
2262 | * but accounting still needs to be done. */ | |
2263 | goto submit_for_resync; | |
b411b363 PR |
2264 | } |
2265 | break; | |
2266 | ||
2267 | case P_OV_REQUEST: | |
b411b363 | 2268 | if (mdev->ov_start_sector == ~(sector_t)0 && |
31890f4a | 2269 | mdev->tconn->agreed_pro_version >= 90) { |
de228bba LE |
2270 | unsigned long now = jiffies; |
2271 | int i; | |
b411b363 PR |
2272 | mdev->ov_start_sector = sector; |
2273 | mdev->ov_position = sector; | |
30b743a2 LE |
2274 | mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector); |
2275 | mdev->rs_total = mdev->ov_left; | |
de228bba LE |
2276 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
2277 | mdev->rs_mark_left[i] = mdev->ov_left; | |
2278 | mdev->rs_mark_time[i] = now; | |
2279 | } | |
b411b363 PR |
2280 | dev_info(DEV, "Online Verify start sector: %llu\n", |
2281 | (unsigned long long)sector); | |
2282 | } | |
db830c46 | 2283 | peer_req->w.cb = w_e_end_ov_req; |
b411b363 | 2284 | fault_type = DRBD_FAULT_RS_RD; |
b411b363 PR |
2285 | break; |
2286 | ||
b411b363 | 2287 | default: |
49ba9b1b | 2288 | BUG(); |
b411b363 PR |
2289 | } |
2290 | ||
0f0601f4 LE |
2291 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2292 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2293 | * Various places in the resync start and stop logic assume resync | |
2294 | * requests are processed in order, requeuing this on the worker thread | |
2295 | * introduces a bunch of new code for synchronization between threads. | |
2296 | * | |
2297 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2298 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2299 | * for application writes for the same time. For now, just throttle | |
2300 | * here, where the rest of the code expects the receiver to sleep for | |
2301 | * a while, anyways. | |
2302 | */ | |
2303 | ||
2304 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2305 | * this defers syncer requests for some time, before letting at least | |
2306 | * on request through. The resync controller on the receiving side | |
2307 | * will adapt to the incoming rate accordingly. | |
2308 | * | |
2309 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2310 | * we would also throttle its application reads. | |
2311 | * In that case, throttling is done on the SyncTarget only. | |
2312 | */ | |
e3555d85 PR |
2313 | if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector)) |
2314 | schedule_timeout_uninterruptible(HZ/10); | |
2315 | if (drbd_rs_begin_io(mdev, sector)) | |
80a40e43 | 2316 | goto out_free_e; |
b411b363 | 2317 | |
0f0601f4 LE |
2318 | submit_for_resync: |
2319 | atomic_add(size >> 9, &mdev->rs_sect_ev); | |
2320 | ||
80a40e43 | 2321 | submit: |
b411b363 | 2322 | inc_unacked(mdev); |
87eeee41 | 2323 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2324 | list_add_tail(&peer_req->w.list, &mdev->read_ee); |
87eeee41 | 2325 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2326 | |
fbe29dec | 2327 | if (drbd_submit_peer_request(mdev, peer_req, READ, fault_type) == 0) |
82bc0194 | 2328 | return 0; |
b411b363 | 2329 | |
10f6d992 LE |
2330 | /* don't care for the reason here */ |
2331 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2332 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2333 | list_del(&peer_req->w.list); |
87eeee41 | 2334 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 LE |
2335 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ |
2336 | ||
b411b363 | 2337 | out_free_e: |
b411b363 | 2338 | put_ldev(mdev); |
db830c46 | 2339 | drbd_free_ee(mdev, peer_req); |
82bc0194 | 2340 | return -EIO; |
b411b363 PR |
2341 | } |
2342 | ||
2343 | static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) | |
2344 | { | |
2345 | int self, peer, rv = -100; | |
2346 | unsigned long ch_self, ch_peer; | |
2347 | ||
2348 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2349 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2350 | ||
2351 | ch_peer = mdev->p_uuid[UI_SIZE]; | |
2352 | ch_self = mdev->comm_bm_set; | |
2353 | ||
89e58e75 | 2354 | switch (mdev->tconn->net_conf->after_sb_0p) { |
b411b363 PR |
2355 | case ASB_CONSENSUS: |
2356 | case ASB_DISCARD_SECONDARY: | |
2357 | case ASB_CALL_HELPER: | |
2358 | dev_err(DEV, "Configuration error.\n"); | |
2359 | break; | |
2360 | case ASB_DISCONNECT: | |
2361 | break; | |
2362 | case ASB_DISCARD_YOUNGER_PRI: | |
2363 | if (self == 0 && peer == 1) { | |
2364 | rv = -1; | |
2365 | break; | |
2366 | } | |
2367 | if (self == 1 && peer == 0) { | |
2368 | rv = 1; | |
2369 | break; | |
2370 | } | |
2371 | /* Else fall through to one of the other strategies... */ | |
2372 | case ASB_DISCARD_OLDER_PRI: | |
2373 | if (self == 0 && peer == 1) { | |
2374 | rv = 1; | |
2375 | break; | |
2376 | } | |
2377 | if (self == 1 && peer == 0) { | |
2378 | rv = -1; | |
2379 | break; | |
2380 | } | |
2381 | /* Else fall through to one of the other strategies... */ | |
ad19bf6e | 2382 | dev_warn(DEV, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2383 | "Using discard-least-changes instead\n"); |
2384 | case ASB_DISCARD_ZERO_CHG: | |
2385 | if (ch_peer == 0 && ch_self == 0) { | |
25703f83 | 2386 | rv = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags) |
b411b363 PR |
2387 | ? -1 : 1; |
2388 | break; | |
2389 | } else { | |
2390 | if (ch_peer == 0) { rv = 1; break; } | |
2391 | if (ch_self == 0) { rv = -1; break; } | |
2392 | } | |
89e58e75 | 2393 | if (mdev->tconn->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) |
b411b363 PR |
2394 | break; |
2395 | case ASB_DISCARD_LEAST_CHG: | |
2396 | if (ch_self < ch_peer) | |
2397 | rv = -1; | |
2398 | else if (ch_self > ch_peer) | |
2399 | rv = 1; | |
2400 | else /* ( ch_self == ch_peer ) */ | |
2401 | /* Well, then use something else. */ | |
25703f83 | 2402 | rv = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags) |
b411b363 PR |
2403 | ? -1 : 1; |
2404 | break; | |
2405 | case ASB_DISCARD_LOCAL: | |
2406 | rv = -1; | |
2407 | break; | |
2408 | case ASB_DISCARD_REMOTE: | |
2409 | rv = 1; | |
2410 | } | |
2411 | ||
2412 | return rv; | |
2413 | } | |
2414 | ||
2415 | static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) | |
2416 | { | |
6184ea21 | 2417 | int hg, rv = -100; |
b411b363 | 2418 | |
89e58e75 | 2419 | switch (mdev->tconn->net_conf->after_sb_1p) { |
b411b363 PR |
2420 | case ASB_DISCARD_YOUNGER_PRI: |
2421 | case ASB_DISCARD_OLDER_PRI: | |
2422 | case ASB_DISCARD_LEAST_CHG: | |
2423 | case ASB_DISCARD_LOCAL: | |
2424 | case ASB_DISCARD_REMOTE: | |
2425 | dev_err(DEV, "Configuration error.\n"); | |
2426 | break; | |
2427 | case ASB_DISCONNECT: | |
2428 | break; | |
2429 | case ASB_CONSENSUS: | |
2430 | hg = drbd_asb_recover_0p(mdev); | |
2431 | if (hg == -1 && mdev->state.role == R_SECONDARY) | |
2432 | rv = hg; | |
2433 | if (hg == 1 && mdev->state.role == R_PRIMARY) | |
2434 | rv = hg; | |
2435 | break; | |
2436 | case ASB_VIOLENTLY: | |
2437 | rv = drbd_asb_recover_0p(mdev); | |
2438 | break; | |
2439 | case ASB_DISCARD_SECONDARY: | |
2440 | return mdev->state.role == R_PRIMARY ? 1 : -1; | |
2441 | case ASB_CALL_HELPER: | |
2442 | hg = drbd_asb_recover_0p(mdev); | |
2443 | if (hg == -1 && mdev->state.role == R_PRIMARY) { | |
bb437946 AG |
2444 | enum drbd_state_rv rv2; |
2445 | ||
2446 | drbd_set_role(mdev, R_SECONDARY, 0); | |
b411b363 PR |
2447 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2448 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2449 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2450 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2451 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2452 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2453 | } else { | |
2454 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2455 | rv = hg; | |
2456 | } | |
2457 | } else | |
2458 | rv = hg; | |
2459 | } | |
2460 | ||
2461 | return rv; | |
2462 | } | |
2463 | ||
2464 | static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) | |
2465 | { | |
6184ea21 | 2466 | int hg, rv = -100; |
b411b363 | 2467 | |
89e58e75 | 2468 | switch (mdev->tconn->net_conf->after_sb_2p) { |
b411b363 PR |
2469 | case ASB_DISCARD_YOUNGER_PRI: |
2470 | case ASB_DISCARD_OLDER_PRI: | |
2471 | case ASB_DISCARD_LEAST_CHG: | |
2472 | case ASB_DISCARD_LOCAL: | |
2473 | case ASB_DISCARD_REMOTE: | |
2474 | case ASB_CONSENSUS: | |
2475 | case ASB_DISCARD_SECONDARY: | |
2476 | dev_err(DEV, "Configuration error.\n"); | |
2477 | break; | |
2478 | case ASB_VIOLENTLY: | |
2479 | rv = drbd_asb_recover_0p(mdev); | |
2480 | break; | |
2481 | case ASB_DISCONNECT: | |
2482 | break; | |
2483 | case ASB_CALL_HELPER: | |
2484 | hg = drbd_asb_recover_0p(mdev); | |
2485 | if (hg == -1) { | |
bb437946 AG |
2486 | enum drbd_state_rv rv2; |
2487 | ||
b411b363 PR |
2488 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2489 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2490 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2491 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2492 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2493 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2494 | } else { | |
2495 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2496 | rv = hg; | |
2497 | } | |
2498 | } else | |
2499 | rv = hg; | |
2500 | } | |
2501 | ||
2502 | return rv; | |
2503 | } | |
2504 | ||
2505 | static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, | |
2506 | u64 bits, u64 flags) | |
2507 | { | |
2508 | if (!uuid) { | |
2509 | dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); | |
2510 | return; | |
2511 | } | |
2512 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", | |
2513 | text, | |
2514 | (unsigned long long)uuid[UI_CURRENT], | |
2515 | (unsigned long long)uuid[UI_BITMAP], | |
2516 | (unsigned long long)uuid[UI_HISTORY_START], | |
2517 | (unsigned long long)uuid[UI_HISTORY_END], | |
2518 | (unsigned long long)bits, | |
2519 | (unsigned long long)flags); | |
2520 | } | |
2521 | ||
2522 | /* | |
2523 | 100 after split brain try auto recover | |
2524 | 2 C_SYNC_SOURCE set BitMap | |
2525 | 1 C_SYNC_SOURCE use BitMap | |
2526 | 0 no Sync | |
2527 | -1 C_SYNC_TARGET use BitMap | |
2528 | -2 C_SYNC_TARGET set BitMap | |
2529 | -100 after split brain, disconnect | |
2530 | -1000 unrelated data | |
4a23f264 PR |
2531 | -1091 requires proto 91 |
2532 | -1096 requires proto 96 | |
b411b363 PR |
2533 | */ |
2534 | static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) | |
2535 | { | |
2536 | u64 self, peer; | |
2537 | int i, j; | |
2538 | ||
2539 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2540 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2541 | ||
2542 | *rule_nr = 10; | |
2543 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2544 | return 0; | |
2545 | ||
2546 | *rule_nr = 20; | |
2547 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2548 | peer != UUID_JUST_CREATED) | |
2549 | return -2; | |
2550 | ||
2551 | *rule_nr = 30; | |
2552 | if (self != UUID_JUST_CREATED && | |
2553 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2554 | return 2; | |
2555 | ||
2556 | if (self == peer) { | |
2557 | int rct, dc; /* roles at crash time */ | |
2558 | ||
2559 | if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { | |
2560 | ||
31890f4a | 2561 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2562 | return -1091; |
b411b363 PR |
2563 | |
2564 | if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && | |
2565 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
2566 | dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); | |
2567 | drbd_uuid_set_bm(mdev, 0UL); | |
2568 | ||
2569 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2570 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2571 | *rule_nr = 34; | |
2572 | } else { | |
2573 | dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); | |
2574 | *rule_nr = 36; | |
2575 | } | |
2576 | ||
2577 | return 1; | |
2578 | } | |
2579 | ||
2580 | if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { | |
2581 | ||
31890f4a | 2582 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2583 | return -1091; |
b411b363 PR |
2584 | |
2585 | if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && | |
2586 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
2587 | dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); | |
2588 | ||
2589 | mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; | |
2590 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; | |
2591 | mdev->p_uuid[UI_BITMAP] = 0UL; | |
2592 | ||
2593 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2594 | *rule_nr = 35; | |
2595 | } else { | |
2596 | dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); | |
2597 | *rule_nr = 37; | |
2598 | } | |
2599 | ||
2600 | return -1; | |
2601 | } | |
2602 | ||
2603 | /* Common power [off|failure] */ | |
2604 | rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + | |
2605 | (mdev->p_uuid[UI_FLAGS] & 2); | |
2606 | /* lowest bit is set when we were primary, | |
2607 | * next bit (weight 2) is set when peer was primary */ | |
2608 | *rule_nr = 40; | |
2609 | ||
2610 | switch (rct) { | |
2611 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2612 | case 1: /* self_pri && !peer_pri */ return 1; | |
2613 | case 2: /* !self_pri && peer_pri */ return -1; | |
2614 | case 3: /* self_pri && peer_pri */ | |
25703f83 | 2615 | dc = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags); |
b411b363 PR |
2616 | return dc ? -1 : 1; |
2617 | } | |
2618 | } | |
2619 | ||
2620 | *rule_nr = 50; | |
2621 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2622 | if (self == peer) | |
2623 | return -1; | |
2624 | ||
2625 | *rule_nr = 51; | |
2626 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2627 | if (self == peer) { | |
31890f4a | 2628 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2629 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == |
2630 | (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
2631 | peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2632 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2633 | resync as sync source modifications of the peer's UUIDs. */ | |
2634 | ||
31890f4a | 2635 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2636 | return -1091; |
b411b363 PR |
2637 | |
2638 | mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; | |
2639 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 PR |
2640 | |
2641 | dev_info(DEV, "Did not got last syncUUID packet, corrected:\n"); | |
2642 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2643 | ||
b411b363 PR |
2644 | return -1; |
2645 | } | |
2646 | } | |
2647 | ||
2648 | *rule_nr = 60; | |
2649 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2650 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2651 | peer = mdev->p_uuid[i] & ~((u64)1); | |
2652 | if (self == peer) | |
2653 | return -2; | |
2654 | } | |
2655 | ||
2656 | *rule_nr = 70; | |
2657 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2658 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2659 | if (self == peer) | |
2660 | return 1; | |
2661 | ||
2662 | *rule_nr = 71; | |
2663 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2664 | if (self == peer) { | |
31890f4a | 2665 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2666 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == |
2667 | (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
2668 | self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2669 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2670 | resync as sync source modifications of our UUIDs. */ | |
2671 | ||
31890f4a | 2672 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2673 | return -1091; |
b411b363 PR |
2674 | |
2675 | _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); | |
2676 | _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); | |
2677 | ||
4a23f264 | 2678 | dev_info(DEV, "Last syncUUID did not get through, corrected:\n"); |
b411b363 PR |
2679 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, |
2680 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2681 | ||
2682 | return 1; | |
2683 | } | |
2684 | } | |
2685 | ||
2686 | ||
2687 | *rule_nr = 80; | |
d8c2a36b | 2688 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 PR |
2689 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
2690 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2691 | if (self == peer) | |
2692 | return 2; | |
2693 | } | |
2694 | ||
2695 | *rule_nr = 90; | |
2696 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2697 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2698 | if (self == peer && self != ((u64)0)) | |
2699 | return 100; | |
2700 | ||
2701 | *rule_nr = 100; | |
2702 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2703 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2704 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { | |
2705 | peer = mdev->p_uuid[j] & ~((u64)1); | |
2706 | if (self == peer) | |
2707 | return -100; | |
2708 | } | |
2709 | } | |
2710 | ||
2711 | return -1000; | |
2712 | } | |
2713 | ||
2714 | /* drbd_sync_handshake() returns the new conn state on success, or | |
2715 | CONN_MASK (-1) on failure. | |
2716 | */ | |
2717 | static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, | |
2718 | enum drbd_disk_state peer_disk) __must_hold(local) | |
2719 | { | |
2720 | int hg, rule_nr; | |
2721 | enum drbd_conns rv = C_MASK; | |
2722 | enum drbd_disk_state mydisk; | |
2723 | ||
2724 | mydisk = mdev->state.disk; | |
2725 | if (mydisk == D_NEGOTIATING) | |
2726 | mydisk = mdev->new_state_tmp.disk; | |
2727 | ||
2728 | dev_info(DEV, "drbd_sync_handshake:\n"); | |
2729 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); | |
2730 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, | |
2731 | mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2732 | ||
2733 | hg = drbd_uuid_compare(mdev, &rule_nr); | |
2734 | ||
2735 | dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); | |
2736 | ||
2737 | if (hg == -1000) { | |
2738 | dev_alert(DEV, "Unrelated data, aborting!\n"); | |
2739 | return C_MASK; | |
2740 | } | |
4a23f264 PR |
2741 | if (hg < -1000) { |
2742 | dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); | |
b411b363 PR |
2743 | return C_MASK; |
2744 | } | |
2745 | ||
2746 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
2747 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
2748 | int f = (hg == -100) || abs(hg) == 2; | |
2749 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
2750 | if (f) | |
2751 | hg = hg*2; | |
2752 | dev_info(DEV, "Becoming sync %s due to disk states.\n", | |
2753 | hg > 0 ? "source" : "target"); | |
2754 | } | |
2755 | ||
3a11a487 AG |
2756 | if (abs(hg) == 100) |
2757 | drbd_khelper(mdev, "initial-split-brain"); | |
2758 | ||
89e58e75 | 2759 | if (hg == 100 || (hg == -100 && mdev->tconn->net_conf->always_asbp)) { |
b411b363 PR |
2760 | int pcount = (mdev->state.role == R_PRIMARY) |
2761 | + (peer_role == R_PRIMARY); | |
2762 | int forced = (hg == -100); | |
2763 | ||
2764 | switch (pcount) { | |
2765 | case 0: | |
2766 | hg = drbd_asb_recover_0p(mdev); | |
2767 | break; | |
2768 | case 1: | |
2769 | hg = drbd_asb_recover_1p(mdev); | |
2770 | break; | |
2771 | case 2: | |
2772 | hg = drbd_asb_recover_2p(mdev); | |
2773 | break; | |
2774 | } | |
2775 | if (abs(hg) < 100) { | |
2776 | dev_warn(DEV, "Split-Brain detected, %d primaries, " | |
2777 | "automatically solved. Sync from %s node\n", | |
2778 | pcount, (hg < 0) ? "peer" : "this"); | |
2779 | if (forced) { | |
2780 | dev_warn(DEV, "Doing a full sync, since" | |
2781 | " UUIDs where ambiguous.\n"); | |
2782 | hg = hg*2; | |
2783 | } | |
2784 | } | |
2785 | } | |
2786 | ||
2787 | if (hg == -100) { | |
89e58e75 | 2788 | if (mdev->tconn->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 | 2789 | hg = -1; |
89e58e75 | 2790 | if (!mdev->tconn->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 PR |
2791 | hg = 1; |
2792 | ||
2793 | if (abs(hg) < 100) | |
2794 | dev_warn(DEV, "Split-Brain detected, manually solved. " | |
2795 | "Sync from %s node\n", | |
2796 | (hg < 0) ? "peer" : "this"); | |
2797 | } | |
2798 | ||
2799 | if (hg == -100) { | |
580b9767 LE |
2800 | /* FIXME this log message is not correct if we end up here |
2801 | * after an attempted attach on a diskless node. | |
2802 | * We just refuse to attach -- well, we drop the "connection" | |
2803 | * to that disk, in a way... */ | |
3a11a487 | 2804 | dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); |
b411b363 PR |
2805 | drbd_khelper(mdev, "split-brain"); |
2806 | return C_MASK; | |
2807 | } | |
2808 | ||
2809 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
2810 | dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); | |
2811 | return C_MASK; | |
2812 | } | |
2813 | ||
2814 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
2815 | mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { | |
89e58e75 | 2816 | switch (mdev->tconn->net_conf->rr_conflict) { |
b411b363 PR |
2817 | case ASB_CALL_HELPER: |
2818 | drbd_khelper(mdev, "pri-lost"); | |
2819 | /* fall through */ | |
2820 | case ASB_DISCONNECT: | |
2821 | dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); | |
2822 | return C_MASK; | |
2823 | case ASB_VIOLENTLY: | |
2824 | dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" | |
2825 | "assumption\n"); | |
2826 | } | |
2827 | } | |
2828 | ||
8169e41b | 2829 | if (mdev->tconn->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->tconn->flags)) { |
cf14c2e9 PR |
2830 | if (hg == 0) |
2831 | dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); | |
2832 | else | |
2833 | dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", | |
2834 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), | |
2835 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
2836 | return C_MASK; | |
2837 | } | |
2838 | ||
b411b363 PR |
2839 | if (abs(hg) >= 2) { |
2840 | dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); | |
20ceb2b2 LE |
2841 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
2842 | BM_LOCKED_SET_ALLOWED)) | |
b411b363 PR |
2843 | return C_MASK; |
2844 | } | |
2845 | ||
2846 | if (hg > 0) { /* become sync source. */ | |
2847 | rv = C_WF_BITMAP_S; | |
2848 | } else if (hg < 0) { /* become sync target */ | |
2849 | rv = C_WF_BITMAP_T; | |
2850 | } else { | |
2851 | rv = C_CONNECTED; | |
2852 | if (drbd_bm_total_weight(mdev)) { | |
2853 | dev_info(DEV, "No resync, but %lu bits in bitmap!\n", | |
2854 | drbd_bm_total_weight(mdev)); | |
2855 | } | |
2856 | } | |
2857 | ||
2858 | return rv; | |
2859 | } | |
2860 | ||
2861 | /* returns 1 if invalid */ | |
2862 | static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) | |
2863 | { | |
2864 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
2865 | if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || | |
2866 | (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) | |
2867 | return 0; | |
2868 | ||
2869 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
2870 | if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || | |
2871 | self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) | |
2872 | return 1; | |
2873 | ||
2874 | /* everything else is valid if they are equal on both sides. */ | |
2875 | if (peer == self) | |
2876 | return 0; | |
2877 | ||
2878 | /* everything es is invalid. */ | |
2879 | return 1; | |
2880 | } | |
2881 | ||
e2857216 | 2882 | static int receive_protocol(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 2883 | { |
e6ef8a5c | 2884 | struct p_protocol *p = tconn->data.rbuf; |
b411b363 | 2885 | int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
cf14c2e9 | 2886 | int p_want_lose, p_two_primaries, cf; |
b411b363 PR |
2887 | char p_integrity_alg[SHARED_SECRET_MAX] = ""; |
2888 | ||
b411b363 PR |
2889 | p_proto = be32_to_cpu(p->protocol); |
2890 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
2891 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
2892 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 2893 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 PR |
2894 | cf = be32_to_cpu(p->conn_flags); |
2895 | p_want_lose = cf & CF_WANT_LOSE; | |
2896 | ||
7204624c | 2897 | clear_bit(CONN_DRY_RUN, &tconn->flags); |
cf14c2e9 PR |
2898 | |
2899 | if (cf & CF_DRY_RUN) | |
7204624c | 2900 | set_bit(CONN_DRY_RUN, &tconn->flags); |
b411b363 | 2901 | |
7204624c PR |
2902 | if (p_proto != tconn->net_conf->wire_protocol) { |
2903 | conn_err(tconn, "incompatible communication protocols\n"); | |
b411b363 PR |
2904 | goto disconnect; |
2905 | } | |
2906 | ||
7204624c PR |
2907 | if (cmp_after_sb(p_after_sb_0p, tconn->net_conf->after_sb_0p)) { |
2908 | conn_err(tconn, "incompatible after-sb-0pri settings\n"); | |
b411b363 PR |
2909 | goto disconnect; |
2910 | } | |
2911 | ||
7204624c PR |
2912 | if (cmp_after_sb(p_after_sb_1p, tconn->net_conf->after_sb_1p)) { |
2913 | conn_err(tconn, "incompatible after-sb-1pri settings\n"); | |
b411b363 PR |
2914 | goto disconnect; |
2915 | } | |
2916 | ||
7204624c PR |
2917 | if (cmp_after_sb(p_after_sb_2p, tconn->net_conf->after_sb_2p)) { |
2918 | conn_err(tconn, "incompatible after-sb-2pri settings\n"); | |
b411b363 PR |
2919 | goto disconnect; |
2920 | } | |
2921 | ||
7204624c PR |
2922 | if (p_want_lose && tconn->net_conf->want_lose) { |
2923 | conn_err(tconn, "both sides have the 'want_lose' flag set\n"); | |
b411b363 PR |
2924 | goto disconnect; |
2925 | } | |
2926 | ||
7204624c PR |
2927 | if (p_two_primaries != tconn->net_conf->two_primaries) { |
2928 | conn_err(tconn, "incompatible setting of the two-primaries options\n"); | |
b411b363 PR |
2929 | goto disconnect; |
2930 | } | |
2931 | ||
7204624c PR |
2932 | if (tconn->agreed_pro_version >= 87) { |
2933 | unsigned char *my_alg = tconn->net_conf->integrity_alg; | |
82bc0194 | 2934 | int err; |
b411b363 | 2935 | |
e2857216 | 2936 | err = drbd_recv_all(tconn, p_integrity_alg, pi->size); |
82bc0194 AG |
2937 | if (err) |
2938 | return err; | |
b411b363 PR |
2939 | |
2940 | p_integrity_alg[SHARED_SECRET_MAX-1] = 0; | |
2941 | if (strcmp(p_integrity_alg, my_alg)) { | |
7204624c | 2942 | conn_err(tconn, "incompatible setting of the data-integrity-alg\n"); |
b411b363 PR |
2943 | goto disconnect; |
2944 | } | |
7204624c | 2945 | conn_info(tconn, "data-integrity-alg: %s\n", |
b411b363 PR |
2946 | my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); |
2947 | } | |
2948 | ||
82bc0194 | 2949 | return 0; |
b411b363 PR |
2950 | |
2951 | disconnect: | |
7204624c | 2952 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 2953 | return -EIO; |
b411b363 PR |
2954 | } |
2955 | ||
2956 | /* helper function | |
2957 | * input: alg name, feature name | |
2958 | * return: NULL (alg name was "") | |
2959 | * ERR_PTR(error) if something goes wrong | |
2960 | * or the crypto hash ptr, if it worked out ok. */ | |
2961 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, | |
2962 | const char *alg, const char *name) | |
2963 | { | |
2964 | struct crypto_hash *tfm; | |
2965 | ||
2966 | if (!alg[0]) | |
2967 | return NULL; | |
2968 | ||
2969 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
2970 | if (IS_ERR(tfm)) { | |
2971 | dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", | |
2972 | alg, name, PTR_ERR(tfm)); | |
2973 | return tfm; | |
2974 | } | |
2975 | if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { | |
2976 | crypto_free_hash(tfm); | |
2977 | dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); | |
2978 | return ERR_PTR(-EINVAL); | |
2979 | } | |
2980 | return tfm; | |
2981 | } | |
2982 | ||
4a76b161 AG |
2983 | static int ignore_remaining_packet(struct drbd_tconn *tconn, struct packet_info *pi) |
2984 | { | |
2985 | void *buffer = tconn->data.rbuf; | |
2986 | int size = pi->size; | |
2987 | ||
2988 | while (size) { | |
2989 | int s = min_t(int, size, DRBD_SOCKET_BUFFER_SIZE); | |
2990 | s = drbd_recv(tconn, buffer, s); | |
2991 | if (s <= 0) { | |
2992 | if (s < 0) | |
2993 | return s; | |
2994 | break; | |
2995 | } | |
2996 | size -= s; | |
2997 | } | |
2998 | if (size) | |
2999 | return -EIO; | |
3000 | return 0; | |
3001 | } | |
3002 | ||
3003 | /* | |
3004 | * config_unknown_volume - device configuration command for unknown volume | |
3005 | * | |
3006 | * When a device is added to an existing connection, the node on which the | |
3007 | * device is added first will send configuration commands to its peer but the | |
3008 | * peer will not know about the device yet. It will warn and ignore these | |
3009 | * commands. Once the device is added on the second node, the second node will | |
3010 | * send the same device configuration commands, but in the other direction. | |
3011 | * | |
3012 | * (We can also end up here if drbd is misconfigured.) | |
3013 | */ | |
3014 | static int config_unknown_volume(struct drbd_tconn *tconn, struct packet_info *pi) | |
3015 | { | |
3016 | conn_warn(tconn, "Volume %u unknown; ignoring %s packet\n", | |
3017 | pi->vnr, cmdname(pi->cmd)); | |
3018 | return ignore_remaining_packet(tconn, pi); | |
3019 | } | |
3020 | ||
3021 | static int receive_SyncParam(struct drbd_tconn *tconn, struct packet_info *pi) | |
b411b363 | 3022 | { |
4a76b161 AG |
3023 | struct drbd_conf *mdev; |
3024 | struct p_rs_param_95 *p = tconn->data.rbuf; | |
b411b363 PR |
3025 | unsigned int header_size, data_size, exp_max_sz; |
3026 | struct crypto_hash *verify_tfm = NULL; | |
3027 | struct crypto_hash *csums_tfm = NULL; | |
4a76b161 | 3028 | const int apv = tconn->agreed_pro_version; |
778f271d PR |
3029 | int *rs_plan_s = NULL; |
3030 | int fifo_size = 0; | |
82bc0194 | 3031 | int err; |
b411b363 | 3032 | |
4a76b161 AG |
3033 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3034 | if (!mdev) | |
3035 | return config_unknown_volume(tconn, pi); | |
3036 | ||
b411b363 PR |
3037 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) |
3038 | : apv == 88 ? sizeof(struct p_rs_param) | |
3039 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
3040 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
3041 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 3042 | |
e2857216 | 3043 | if (pi->size > exp_max_sz) { |
b411b363 | 3044 | dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
e2857216 | 3045 | pi->size, exp_max_sz); |
82bc0194 | 3046 | return -EIO; |
b411b363 PR |
3047 | } |
3048 | ||
3049 | if (apv <= 88) { | |
257d0af6 | 3050 | header_size = sizeof(struct p_rs_param) - sizeof(struct p_header); |
e2857216 | 3051 | data_size = pi->size - header_size; |
8e26f9cc | 3052 | } else if (apv <= 94) { |
257d0af6 | 3053 | header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header); |
e2857216 | 3054 | data_size = pi->size - header_size; |
b411b363 | 3055 | D_ASSERT(data_size == 0); |
8e26f9cc | 3056 | } else { |
257d0af6 | 3057 | header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header); |
e2857216 | 3058 | data_size = pi->size - header_size; |
b411b363 PR |
3059 | D_ASSERT(data_size == 0); |
3060 | } | |
3061 | ||
3062 | /* initialize verify_alg and csums_alg */ | |
3063 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
3064 | ||
82bc0194 AG |
3065 | err = drbd_recv_all(mdev->tconn, &p->head.payload, header_size); |
3066 | if (err) | |
3067 | return err; | |
b411b363 | 3068 | |
f399002e LE |
3069 | if (get_ldev(mdev)) { |
3070 | mdev->ldev->dc.resync_rate = be32_to_cpu(p->rate); | |
3071 | put_ldev(mdev); | |
3072 | } | |
b411b363 PR |
3073 | |
3074 | if (apv >= 88) { | |
3075 | if (apv == 88) { | |
3076 | if (data_size > SHARED_SECRET_MAX) { | |
3077 | dev_err(DEV, "verify-alg too long, " | |
3078 | "peer wants %u, accepting only %u byte\n", | |
3079 | data_size, SHARED_SECRET_MAX); | |
82bc0194 | 3080 | return -EIO; |
b411b363 PR |
3081 | } |
3082 | ||
82bc0194 AG |
3083 | err = drbd_recv_all(mdev->tconn, p->verify_alg, data_size); |
3084 | if (err) | |
3085 | return err; | |
b411b363 PR |
3086 | |
3087 | /* we expect NUL terminated string */ | |
3088 | /* but just in case someone tries to be evil */ | |
3089 | D_ASSERT(p->verify_alg[data_size-1] == 0); | |
3090 | p->verify_alg[data_size-1] = 0; | |
3091 | ||
3092 | } else /* apv >= 89 */ { | |
3093 | /* we still expect NUL terminated strings */ | |
3094 | /* but just in case someone tries to be evil */ | |
3095 | D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); | |
3096 | D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
3097 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; | |
3098 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
3099 | } | |
3100 | ||
f399002e | 3101 | if (strcmp(mdev->tconn->net_conf->verify_alg, p->verify_alg)) { |
b411b363 PR |
3102 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { |
3103 | dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", | |
f399002e | 3104 | mdev->tconn->net_conf->verify_alg, p->verify_alg); |
b411b363 PR |
3105 | goto disconnect; |
3106 | } | |
3107 | verify_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3108 | p->verify_alg, "verify-alg"); | |
3109 | if (IS_ERR(verify_tfm)) { | |
3110 | verify_tfm = NULL; | |
3111 | goto disconnect; | |
3112 | } | |
3113 | } | |
3114 | ||
f399002e | 3115 | if (apv >= 89 && strcmp(mdev->tconn->net_conf->csums_alg, p->csums_alg)) { |
b411b363 PR |
3116 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { |
3117 | dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", | |
f399002e | 3118 | mdev->tconn->net_conf->csums_alg, p->csums_alg); |
b411b363 PR |
3119 | goto disconnect; |
3120 | } | |
3121 | csums_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3122 | p->csums_alg, "csums-alg"); | |
3123 | if (IS_ERR(csums_tfm)) { | |
3124 | csums_tfm = NULL; | |
3125 | goto disconnect; | |
3126 | } | |
3127 | } | |
3128 | ||
f399002e LE |
3129 | if (apv > 94 && get_ldev(mdev)) { |
3130 | mdev->ldev->dc.resync_rate = be32_to_cpu(p->rate); | |
3131 | mdev->ldev->dc.c_plan_ahead = be32_to_cpu(p->c_plan_ahead); | |
3132 | mdev->ldev->dc.c_delay_target = be32_to_cpu(p->c_delay_target); | |
3133 | mdev->ldev->dc.c_fill_target = be32_to_cpu(p->c_fill_target); | |
3134 | mdev->ldev->dc.c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d | 3135 | |
f399002e | 3136 | fifo_size = (mdev->ldev->dc.c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
778f271d PR |
3137 | if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { |
3138 | rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); | |
3139 | if (!rs_plan_s) { | |
3140 | dev_err(DEV, "kmalloc of fifo_buffer failed"); | |
f399002e | 3141 | put_ldev(mdev); |
778f271d PR |
3142 | goto disconnect; |
3143 | } | |
3144 | } | |
f399002e | 3145 | put_ldev(mdev); |
8e26f9cc | 3146 | } |
b411b363 PR |
3147 | |
3148 | spin_lock(&mdev->peer_seq_lock); | |
3149 | /* lock against drbd_nl_syncer_conf() */ | |
3150 | if (verify_tfm) { | |
f399002e LE |
3151 | strcpy(mdev->tconn->net_conf->verify_alg, p->verify_alg); |
3152 | mdev->tconn->net_conf->verify_alg_len = strlen(p->verify_alg) + 1; | |
3153 | crypto_free_hash(mdev->tconn->verify_tfm); | |
3154 | mdev->tconn->verify_tfm = verify_tfm; | |
b411b363 PR |
3155 | dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); |
3156 | } | |
3157 | if (csums_tfm) { | |
f399002e LE |
3158 | strcpy(mdev->tconn->net_conf->csums_alg, p->csums_alg); |
3159 | mdev->tconn->net_conf->csums_alg_len = strlen(p->csums_alg) + 1; | |
3160 | crypto_free_hash(mdev->tconn->csums_tfm); | |
3161 | mdev->tconn->csums_tfm = csums_tfm; | |
b411b363 PR |
3162 | dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); |
3163 | } | |
778f271d PR |
3164 | if (fifo_size != mdev->rs_plan_s.size) { |
3165 | kfree(mdev->rs_plan_s.values); | |
3166 | mdev->rs_plan_s.values = rs_plan_s; | |
3167 | mdev->rs_plan_s.size = fifo_size; | |
3168 | mdev->rs_planed = 0; | |
3169 | } | |
b411b363 PR |
3170 | spin_unlock(&mdev->peer_seq_lock); |
3171 | } | |
82bc0194 | 3172 | return 0; |
b411b363 | 3173 | |
b411b363 PR |
3174 | disconnect: |
3175 | /* just for completeness: actually not needed, | |
3176 | * as this is not reached if csums_tfm was ok. */ | |
3177 | crypto_free_hash(csums_tfm); | |
3178 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
3179 | crypto_free_hash(verify_tfm); | |
38fa9988 | 3180 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3181 | return -EIO; |
b411b363 PR |
3182 | } |
3183 | ||
b411b363 PR |
3184 | /* warn if the arguments differ by more than 12.5% */ |
3185 | static void warn_if_differ_considerably(struct drbd_conf *mdev, | |
3186 | const char *s, sector_t a, sector_t b) | |
3187 | { | |
3188 | sector_t d; | |
3189 | if (a == 0 || b == 0) | |
3190 | return; | |
3191 | d = (a > b) ? (a - b) : (b - a); | |
3192 | if (d > (a>>3) || d > (b>>3)) | |
3193 | dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, | |
3194 | (unsigned long long)a, (unsigned long long)b); | |
3195 | } | |
3196 | ||
4a76b161 | 3197 | static int receive_sizes(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3198 | { |
4a76b161 AG |
3199 | struct drbd_conf *mdev; |
3200 | struct p_sizes *p = tconn->data.rbuf; | |
b411b363 | 3201 | enum determine_dev_size dd = unchanged; |
b411b363 PR |
3202 | sector_t p_size, p_usize, my_usize; |
3203 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 3204 | enum dds_flags ddsf; |
b411b363 | 3205 | |
4a76b161 AG |
3206 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3207 | if (!mdev) | |
3208 | return config_unknown_volume(tconn, pi); | |
3209 | ||
b411b363 PR |
3210 | p_size = be64_to_cpu(p->d_size); |
3211 | p_usize = be64_to_cpu(p->u_size); | |
3212 | ||
b411b363 PR |
3213 | /* just store the peer's disk size for now. |
3214 | * we still need to figure out whether we accept that. */ | |
3215 | mdev->p_size = p_size; | |
3216 | ||
b411b363 PR |
3217 | if (get_ldev(mdev)) { |
3218 | warn_if_differ_considerably(mdev, "lower level device sizes", | |
3219 | p_size, drbd_get_max_capacity(mdev->ldev)); | |
3220 | warn_if_differ_considerably(mdev, "user requested size", | |
3221 | p_usize, mdev->ldev->dc.disk_size); | |
3222 | ||
3223 | /* if this is the first connect, or an otherwise expected | |
3224 | * param exchange, choose the minimum */ | |
3225 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
3226 | p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, | |
3227 | p_usize); | |
3228 | ||
3229 | my_usize = mdev->ldev->dc.disk_size; | |
3230 | ||
3231 | if (mdev->ldev->dc.disk_size != p_usize) { | |
3232 | mdev->ldev->dc.disk_size = p_usize; | |
3233 | dev_info(DEV, "Peer sets u_size to %lu sectors\n", | |
3234 | (unsigned long)mdev->ldev->dc.disk_size); | |
3235 | } | |
3236 | ||
3237 | /* Never shrink a device with usable data during connect. | |
3238 | But allow online shrinking if we are connected. */ | |
a393db6f | 3239 | if (drbd_new_dev_size(mdev, mdev->ldev, 0) < |
b411b363 PR |
3240 | drbd_get_capacity(mdev->this_bdev) && |
3241 | mdev->state.disk >= D_OUTDATED && | |
3242 | mdev->state.conn < C_CONNECTED) { | |
3243 | dev_err(DEV, "The peer's disk size is too small!\n"); | |
38fa9988 | 3244 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
3245 | mdev->ldev->dc.disk_size = my_usize; |
3246 | put_ldev(mdev); | |
82bc0194 | 3247 | return -EIO; |
b411b363 PR |
3248 | } |
3249 | put_ldev(mdev); | |
3250 | } | |
b411b363 | 3251 | |
e89b591c | 3252 | ddsf = be16_to_cpu(p->dds_flags); |
b411b363 | 3253 | if (get_ldev(mdev)) { |
24c4830c | 3254 | dd = drbd_determine_dev_size(mdev, ddsf); |
b411b363 PR |
3255 | put_ldev(mdev); |
3256 | if (dd == dev_size_error) | |
82bc0194 | 3257 | return -EIO; |
b411b363 PR |
3258 | drbd_md_sync(mdev); |
3259 | } else { | |
3260 | /* I am diskless, need to accept the peer's size. */ | |
3261 | drbd_set_my_capacity(mdev, p_size); | |
3262 | } | |
3263 | ||
99432fcc PR |
3264 | mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
3265 | drbd_reconsider_max_bio_size(mdev); | |
3266 | ||
b411b363 PR |
3267 | if (get_ldev(mdev)) { |
3268 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { | |
3269 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); | |
3270 | ldsc = 1; | |
3271 | } | |
3272 | ||
b411b363 PR |
3273 | put_ldev(mdev); |
3274 | } | |
3275 | ||
3276 | if (mdev->state.conn > C_WF_REPORT_PARAMS) { | |
3277 | if (be64_to_cpu(p->c_size) != | |
3278 | drbd_get_capacity(mdev->this_bdev) || ldsc) { | |
3279 | /* we have different sizes, probably peer | |
3280 | * needs to know my new size... */ | |
e89b591c | 3281 | drbd_send_sizes(mdev, 0, ddsf); |
b411b363 PR |
3282 | } |
3283 | if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || | |
3284 | (dd == grew && mdev->state.conn == C_CONNECTED)) { | |
3285 | if (mdev->state.pdsk >= D_INCONSISTENT && | |
e89b591c PR |
3286 | mdev->state.disk >= D_INCONSISTENT) { |
3287 | if (ddsf & DDSF_NO_RESYNC) | |
3288 | dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); | |
3289 | else | |
3290 | resync_after_online_grow(mdev); | |
3291 | } else | |
b411b363 PR |
3292 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); |
3293 | } | |
3294 | } | |
3295 | ||
82bc0194 | 3296 | return 0; |
b411b363 PR |
3297 | } |
3298 | ||
4a76b161 | 3299 | static int receive_uuids(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3300 | { |
4a76b161 AG |
3301 | struct drbd_conf *mdev; |
3302 | struct p_uuids *p = tconn->data.rbuf; | |
b411b363 | 3303 | u64 *p_uuid; |
62b0da3a | 3304 | int i, updated_uuids = 0; |
b411b363 | 3305 | |
4a76b161 AG |
3306 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3307 | if (!mdev) | |
3308 | return config_unknown_volume(tconn, pi); | |
3309 | ||
b411b363 PR |
3310 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
3311 | ||
3312 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3313 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3314 | ||
3315 | kfree(mdev->p_uuid); | |
3316 | mdev->p_uuid = p_uuid; | |
3317 | ||
3318 | if (mdev->state.conn < C_CONNECTED && | |
3319 | mdev->state.disk < D_INCONSISTENT && | |
3320 | mdev->state.role == R_PRIMARY && | |
3321 | (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
3322 | dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", | |
3323 | (unsigned long long)mdev->ed_uuid); | |
38fa9988 | 3324 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3325 | return -EIO; |
b411b363 PR |
3326 | } |
3327 | ||
3328 | if (get_ldev(mdev)) { | |
3329 | int skip_initial_sync = | |
3330 | mdev->state.conn == C_CONNECTED && | |
31890f4a | 3331 | mdev->tconn->agreed_pro_version >= 90 && |
b411b363 PR |
3332 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && |
3333 | (p_uuid[UI_FLAGS] & 8); | |
3334 | if (skip_initial_sync) { | |
3335 | dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); | |
3336 | drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, | |
20ceb2b2 LE |
3337 | "clear_n_write from receive_uuids", |
3338 | BM_LOCKED_TEST_ALLOWED); | |
b411b363 PR |
3339 | _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); |
3340 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
3341 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
3342 | CS_VERBOSE, NULL); | |
3343 | drbd_md_sync(mdev); | |
62b0da3a | 3344 | updated_uuids = 1; |
b411b363 PR |
3345 | } |
3346 | put_ldev(mdev); | |
18a50fa2 PR |
3347 | } else if (mdev->state.disk < D_INCONSISTENT && |
3348 | mdev->state.role == R_PRIMARY) { | |
3349 | /* I am a diskless primary, the peer just created a new current UUID | |
3350 | for me. */ | |
62b0da3a | 3351 | updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
b411b363 PR |
3352 | } |
3353 | ||
3354 | /* Before we test for the disk state, we should wait until an eventually | |
3355 | ongoing cluster wide state change is finished. That is important if | |
3356 | we are primary and are detaching from our disk. We need to see the | |
3357 | new disk state... */ | |
8410da8f PR |
3358 | mutex_lock(mdev->state_mutex); |
3359 | mutex_unlock(mdev->state_mutex); | |
b411b363 | 3360 | if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) |
62b0da3a LE |
3361 | updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
3362 | ||
3363 | if (updated_uuids) | |
3364 | drbd_print_uuids(mdev, "receiver updated UUIDs to"); | |
b411b363 | 3365 | |
82bc0194 | 3366 | return 0; |
b411b363 PR |
3367 | } |
3368 | ||
3369 | /** | |
3370 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3371 | * @ps: The state as seen by the peer. | |
3372 | */ | |
3373 | static union drbd_state convert_state(union drbd_state ps) | |
3374 | { | |
3375 | union drbd_state ms; | |
3376 | ||
3377 | static enum drbd_conns c_tab[] = { | |
3378 | [C_CONNECTED] = C_CONNECTED, | |
3379 | ||
3380 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3381 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3382 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3383 | [C_VERIFY_S] = C_VERIFY_T, | |
3384 | [C_MASK] = C_MASK, | |
3385 | }; | |
3386 | ||
3387 | ms.i = ps.i; | |
3388 | ||
3389 | ms.conn = c_tab[ps.conn]; | |
3390 | ms.peer = ps.role; | |
3391 | ms.role = ps.peer; | |
3392 | ms.pdsk = ps.disk; | |
3393 | ms.disk = ps.pdsk; | |
3394 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3395 | ||
3396 | return ms; | |
3397 | } | |
3398 | ||
4a76b161 | 3399 | static int receive_req_state(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3400 | { |
4a76b161 AG |
3401 | struct drbd_conf *mdev; |
3402 | struct p_req_state *p = tconn->data.rbuf; | |
b411b363 | 3403 | union drbd_state mask, val; |
bf885f8a | 3404 | enum drbd_state_rv rv; |
b411b363 | 3405 | |
4a76b161 AG |
3406 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3407 | if (!mdev) | |
3408 | return -EIO; | |
3409 | ||
b411b363 PR |
3410 | mask.i = be32_to_cpu(p->mask); |
3411 | val.i = be32_to_cpu(p->val); | |
3412 | ||
25703f83 | 3413 | if (test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags) && |
8410da8f | 3414 | mutex_is_locked(mdev->state_mutex)) { |
b411b363 | 3415 | drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); |
82bc0194 | 3416 | return 0; |
b411b363 PR |
3417 | } |
3418 | ||
3419 | mask = convert_state(mask); | |
3420 | val = convert_state(val); | |
3421 | ||
dfafcc8a PR |
3422 | rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); |
3423 | drbd_send_sr_reply(mdev, rv); | |
b411b363 | 3424 | |
b411b363 PR |
3425 | drbd_md_sync(mdev); |
3426 | ||
82bc0194 | 3427 | return 0; |
b411b363 PR |
3428 | } |
3429 | ||
e2857216 | 3430 | static int receive_req_conn_state(struct drbd_tconn *tconn, struct packet_info *pi) |
dfafcc8a | 3431 | { |
e6ef8a5c | 3432 | struct p_req_state *p = tconn->data.rbuf; |
dfafcc8a PR |
3433 | union drbd_state mask, val; |
3434 | enum drbd_state_rv rv; | |
3435 | ||
3436 | mask.i = be32_to_cpu(p->mask); | |
3437 | val.i = be32_to_cpu(p->val); | |
3438 | ||
3439 | if (test_bit(DISCARD_CONCURRENT, &tconn->flags) && | |
3440 | mutex_is_locked(&tconn->cstate_mutex)) { | |
3441 | conn_send_sr_reply(tconn, SS_CONCURRENT_ST_CHG); | |
82bc0194 | 3442 | return 0; |
dfafcc8a PR |
3443 | } |
3444 | ||
3445 | mask = convert_state(mask); | |
3446 | val = convert_state(val); | |
3447 | ||
778bcf2e | 3448 | rv = conn_request_state(tconn, mask, val, CS_VERBOSE | CS_LOCAL_ONLY | CS_IGN_OUTD_FAIL); |
dfafcc8a PR |
3449 | conn_send_sr_reply(tconn, rv); |
3450 | ||
82bc0194 | 3451 | return 0; |
dfafcc8a PR |
3452 | } |
3453 | ||
4a76b161 | 3454 | static int receive_state(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3455 | { |
4a76b161 AG |
3456 | struct drbd_conf *mdev; |
3457 | struct p_state *p = tconn->data.rbuf; | |
4ac4aada | 3458 | union drbd_state os, ns, peer_state; |
b411b363 | 3459 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 3460 | enum chg_state_flags cs_flags; |
b411b363 PR |
3461 | int rv; |
3462 | ||
4a76b161 AG |
3463 | mdev = vnr_to_mdev(tconn, pi->vnr); |
3464 | if (!mdev) | |
3465 | return config_unknown_volume(tconn, pi); | |
3466 | ||
b411b363 PR |
3467 | peer_state.i = be32_to_cpu(p->state); |
3468 | ||
3469 | real_peer_disk = peer_state.disk; | |
3470 | if (peer_state.disk == D_NEGOTIATING) { | |
3471 | real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; | |
3472 | dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); | |
3473 | } | |
3474 | ||
87eeee41 | 3475 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 3476 | retry: |
78bae59b | 3477 | os = ns = drbd_read_state(mdev); |
87eeee41 | 3478 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 3479 | |
e9ef7bb6 LE |
3480 | /* peer says his disk is uptodate, while we think it is inconsistent, |
3481 | * and this happens while we think we have a sync going on. */ | |
3482 | if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE && | |
3483 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { | |
3484 | /* If we are (becoming) SyncSource, but peer is still in sync | |
3485 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
3486 | * will change to inconsistent once the peer reaches active | |
3487 | * syncing states. | |
3488 | * It may have changed syncer-paused flags, however, so we | |
3489 | * cannot ignore this completely. */ | |
3490 | if (peer_state.conn > C_CONNECTED && | |
3491 | peer_state.conn < C_SYNC_SOURCE) | |
3492 | real_peer_disk = D_INCONSISTENT; | |
3493 | ||
3494 | /* if peer_state changes to connected at the same time, | |
3495 | * it explicitly notifies us that it finished resync. | |
3496 | * Maybe we should finish it up, too? */ | |
3497 | else if (os.conn >= C_SYNC_SOURCE && | |
3498 | peer_state.conn == C_CONNECTED) { | |
3499 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) | |
3500 | drbd_resync_finished(mdev); | |
82bc0194 | 3501 | return 0; |
e9ef7bb6 LE |
3502 | } |
3503 | } | |
3504 | ||
3505 | /* peer says his disk is inconsistent, while we think it is uptodate, | |
3506 | * and this happens while the peer still thinks we have a sync going on, | |
3507 | * but we think we are already done with the sync. | |
3508 | * We ignore this to avoid flapping pdsk. | |
3509 | * This should not happen, if the peer is a recent version of drbd. */ | |
3510 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
3511 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
3512 | real_peer_disk = D_UP_TO_DATE; | |
3513 | ||
4ac4aada LE |
3514 | if (ns.conn == C_WF_REPORT_PARAMS) |
3515 | ns.conn = C_CONNECTED; | |
b411b363 | 3516 | |
67531718 PR |
3517 | if (peer_state.conn == C_AHEAD) |
3518 | ns.conn = C_BEHIND; | |
3519 | ||
b411b363 PR |
3520 | if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && |
3521 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3522 | int cr; /* consider resync */ | |
3523 | ||
3524 | /* if we established a new connection */ | |
4ac4aada | 3525 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
3526 | /* if we had an established connection |
3527 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 3528 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 3529 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 3530 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
3531 | /* if we have both been inconsistent, and the peer has been |
3532 | * forced to be UpToDate with --overwrite-data */ | |
3533 | cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); | |
3534 | /* if we had been plain connected, and the admin requested to | |
3535 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 3536 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
3537 | (peer_state.conn >= C_STARTING_SYNC_S && |
3538 | peer_state.conn <= C_WF_BITMAP_T)); | |
3539 | ||
3540 | if (cr) | |
4ac4aada | 3541 | ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); |
b411b363 PR |
3542 | |
3543 | put_ldev(mdev); | |
4ac4aada LE |
3544 | if (ns.conn == C_MASK) { |
3545 | ns.conn = C_CONNECTED; | |
b411b363 | 3546 | if (mdev->state.disk == D_NEGOTIATING) { |
82f59cc6 | 3547 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
b411b363 PR |
3548 | } else if (peer_state.disk == D_NEGOTIATING) { |
3549 | dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); | |
3550 | peer_state.disk = D_DISKLESS; | |
580b9767 | 3551 | real_peer_disk = D_DISKLESS; |
b411b363 | 3552 | } else { |
8169e41b | 3553 | if (test_and_clear_bit(CONN_DRY_RUN, &mdev->tconn->flags)) |
82bc0194 | 3554 | return -EIO; |
4ac4aada | 3555 | D_ASSERT(os.conn == C_WF_REPORT_PARAMS); |
38fa9988 | 3556 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3557 | return -EIO; |
b411b363 PR |
3558 | } |
3559 | } | |
3560 | } | |
3561 | ||
87eeee41 | 3562 | spin_lock_irq(&mdev->tconn->req_lock); |
78bae59b | 3563 | if (os.i != drbd_read_state(mdev).i) |
b411b363 PR |
3564 | goto retry; |
3565 | clear_bit(CONSIDER_RESYNC, &mdev->flags); | |
b411b363 PR |
3566 | ns.peer = peer_state.role; |
3567 | ns.pdsk = real_peer_disk; | |
3568 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 3569 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b411b363 | 3570 | ns.disk = mdev->new_state_tmp.disk; |
4ac4aada | 3571 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
2aebfabb | 3572 | if (ns.pdsk == D_CONSISTENT && drbd_suspended(mdev) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && |
481c6f50 | 3573 | test_bit(NEW_CUR_UUID, &mdev->flags)) { |
8554df1c | 3574 | /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this |
481c6f50 | 3575 | for temporal network outages! */ |
87eeee41 | 3576 | spin_unlock_irq(&mdev->tconn->req_lock); |
481c6f50 | 3577 | dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); |
2f5cdd0b | 3578 | tl_clear(mdev->tconn); |
481c6f50 PR |
3579 | drbd_uuid_new_current(mdev); |
3580 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
38fa9988 | 3581 | conn_request_state(mdev->tconn, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD); |
82bc0194 | 3582 | return -EIO; |
481c6f50 | 3583 | } |
65d922c3 | 3584 | rv = _drbd_set_state(mdev, ns, cs_flags, NULL); |
78bae59b | 3585 | ns = drbd_read_state(mdev); |
87eeee41 | 3586 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3587 | |
3588 | if (rv < SS_SUCCESS) { | |
38fa9988 | 3589 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3590 | return -EIO; |
b411b363 PR |
3591 | } |
3592 | ||
4ac4aada LE |
3593 | if (os.conn > C_WF_REPORT_PARAMS) { |
3594 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
3595 | peer_state.disk != D_NEGOTIATING ) { |
3596 | /* we want resync, peer has not yet decided to sync... */ | |
3597 | /* Nowadays only used when forcing a node into primary role and | |
3598 | setting its disk to UpToDate with that */ | |
3599 | drbd_send_uuids(mdev); | |
3600 | drbd_send_state(mdev); | |
3601 | } | |
3602 | } | |
3603 | ||
89e58e75 | 3604 | mdev->tconn->net_conf->want_lose = 0; |
b411b363 PR |
3605 | |
3606 | drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ | |
3607 | ||
82bc0194 | 3608 | return 0; |
b411b363 PR |
3609 | } |
3610 | ||
4a76b161 | 3611 | static int receive_sync_uuid(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3612 | { |
4a76b161 AG |
3613 | struct drbd_conf *mdev; |
3614 | struct p_rs_uuid *p = tconn->data.rbuf; | |
3615 | ||
3616 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
3617 | if (!mdev) | |
3618 | return -EIO; | |
b411b363 PR |
3619 | |
3620 | wait_event(mdev->misc_wait, | |
3621 | mdev->state.conn == C_WF_SYNC_UUID || | |
c4752ef1 | 3622 | mdev->state.conn == C_BEHIND || |
b411b363 PR |
3623 | mdev->state.conn < C_CONNECTED || |
3624 | mdev->state.disk < D_NEGOTIATING); | |
3625 | ||
3626 | /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ | |
3627 | ||
b411b363 PR |
3628 | /* Here the _drbd_uuid_ functions are right, current should |
3629 | _not_ be rotated into the history */ | |
3630 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3631 | _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); | |
3632 | _drbd_uuid_set(mdev, UI_BITMAP, 0UL); | |
3633 | ||
62b0da3a | 3634 | drbd_print_uuids(mdev, "updated sync uuid"); |
b411b363 PR |
3635 | drbd_start_resync(mdev, C_SYNC_TARGET); |
3636 | ||
3637 | put_ldev(mdev); | |
3638 | } else | |
3639 | dev_err(DEV, "Ignoring SyncUUID packet!\n"); | |
3640 | ||
82bc0194 | 3641 | return 0; |
b411b363 PR |
3642 | } |
3643 | ||
2c46407d AG |
3644 | /** |
3645 | * receive_bitmap_plain | |
3646 | * | |
3647 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3648 | * code upon failure. | |
3649 | */ | |
3650 | static int | |
02918be2 | 3651 | receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size, |
fc56815c | 3652 | struct p_header *h, struct bm_xfer_ctx *c) |
b411b363 | 3653 | { |
fc56815c | 3654 | unsigned long *buffer = (unsigned long *)h->payload; |
b411b363 PR |
3655 | unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); |
3656 | unsigned want = num_words * sizeof(long); | |
2c46407d | 3657 | int err; |
b411b363 | 3658 | |
02918be2 PR |
3659 | if (want != data_size) { |
3660 | dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size); | |
2c46407d | 3661 | return -EIO; |
b411b363 PR |
3662 | } |
3663 | if (want == 0) | |
2c46407d | 3664 | return 0; |
82bc0194 AG |
3665 | err = drbd_recv_all(mdev->tconn, buffer, want); |
3666 | if (err) | |
2c46407d | 3667 | return err; |
b411b363 PR |
3668 | |
3669 | drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); | |
3670 | ||
3671 | c->word_offset += num_words; | |
3672 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
3673 | if (c->bit_offset > c->bm_bits) | |
3674 | c->bit_offset = c->bm_bits; | |
3675 | ||
2c46407d | 3676 | return 1; |
b411b363 PR |
3677 | } |
3678 | ||
a02d1240 AG |
3679 | static enum drbd_bitmap_code dcbp_get_code(struct p_compressed_bm *p) |
3680 | { | |
3681 | return (enum drbd_bitmap_code)(p->encoding & 0x0f); | |
3682 | } | |
3683 | ||
3684 | static int dcbp_get_start(struct p_compressed_bm *p) | |
3685 | { | |
3686 | return (p->encoding & 0x80) != 0; | |
3687 | } | |
3688 | ||
3689 | static int dcbp_get_pad_bits(struct p_compressed_bm *p) | |
3690 | { | |
3691 | return (p->encoding >> 4) & 0x7; | |
3692 | } | |
3693 | ||
2c46407d AG |
3694 | /** |
3695 | * recv_bm_rle_bits | |
3696 | * | |
3697 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3698 | * code upon failure. | |
3699 | */ | |
3700 | static int | |
b411b363 PR |
3701 | recv_bm_rle_bits(struct drbd_conf *mdev, |
3702 | struct p_compressed_bm *p, | |
c6d25cfe PR |
3703 | struct bm_xfer_ctx *c, |
3704 | unsigned int len) | |
b411b363 PR |
3705 | { |
3706 | struct bitstream bs; | |
3707 | u64 look_ahead; | |
3708 | u64 rl; | |
3709 | u64 tmp; | |
3710 | unsigned long s = c->bit_offset; | |
3711 | unsigned long e; | |
a02d1240 | 3712 | int toggle = dcbp_get_start(p); |
b411b363 PR |
3713 | int have; |
3714 | int bits; | |
3715 | ||
a02d1240 | 3716 | bitstream_init(&bs, p->code, len, dcbp_get_pad_bits(p)); |
b411b363 PR |
3717 | |
3718 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
3719 | if (bits < 0) | |
2c46407d | 3720 | return -EIO; |
b411b363 PR |
3721 | |
3722 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
3723 | bits = vli_decode_bits(&rl, look_ahead); | |
3724 | if (bits <= 0) | |
2c46407d | 3725 | return -EIO; |
b411b363 PR |
3726 | |
3727 | if (toggle) { | |
3728 | e = s + rl -1; | |
3729 | if (e >= c->bm_bits) { | |
3730 | dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); | |
2c46407d | 3731 | return -EIO; |
b411b363 PR |
3732 | } |
3733 | _drbd_bm_set_bits(mdev, s, e); | |
3734 | } | |
3735 | ||
3736 | if (have < bits) { | |
3737 | dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", | |
3738 | have, bits, look_ahead, | |
3739 | (unsigned int)(bs.cur.b - p->code), | |
3740 | (unsigned int)bs.buf_len); | |
2c46407d | 3741 | return -EIO; |
b411b363 PR |
3742 | } |
3743 | look_ahead >>= bits; | |
3744 | have -= bits; | |
3745 | ||
3746 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
3747 | if (bits < 0) | |
2c46407d | 3748 | return -EIO; |
b411b363 PR |
3749 | look_ahead |= tmp << have; |
3750 | have += bits; | |
3751 | } | |
3752 | ||
3753 | c->bit_offset = s; | |
3754 | bm_xfer_ctx_bit_to_word_offset(c); | |
3755 | ||
2c46407d | 3756 | return (s != c->bm_bits); |
b411b363 PR |
3757 | } |
3758 | ||
2c46407d AG |
3759 | /** |
3760 | * decode_bitmap_c | |
3761 | * | |
3762 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3763 | * code upon failure. | |
3764 | */ | |
3765 | static int | |
b411b363 PR |
3766 | decode_bitmap_c(struct drbd_conf *mdev, |
3767 | struct p_compressed_bm *p, | |
c6d25cfe PR |
3768 | struct bm_xfer_ctx *c, |
3769 | unsigned int len) | |
b411b363 | 3770 | { |
a02d1240 | 3771 | if (dcbp_get_code(p) == RLE_VLI_Bits) |
c6d25cfe | 3772 | return recv_bm_rle_bits(mdev, p, c, len); |
b411b363 PR |
3773 | |
3774 | /* other variants had been implemented for evaluation, | |
3775 | * but have been dropped as this one turned out to be "best" | |
3776 | * during all our tests. */ | |
3777 | ||
3778 | dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); | |
38fa9988 | 3779 | conn_request_state(mdev->tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
2c46407d | 3780 | return -EIO; |
b411b363 PR |
3781 | } |
3782 | ||
3783 | void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
3784 | const char *direction, struct bm_xfer_ctx *c) | |
3785 | { | |
3786 | /* what would it take to transfer it "plaintext" */ | |
c012949a | 3787 | unsigned plain = sizeof(struct p_header) * |
b411b363 PR |
3788 | ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) |
3789 | + c->bm_words * sizeof(long); | |
3790 | unsigned total = c->bytes[0] + c->bytes[1]; | |
3791 | unsigned r; | |
3792 | ||
3793 | /* total can not be zero. but just in case: */ | |
3794 | if (total == 0) | |
3795 | return; | |
3796 | ||
3797 | /* don't report if not compressed */ | |
3798 | if (total >= plain) | |
3799 | return; | |
3800 | ||
3801 | /* total < plain. check for overflow, still */ | |
3802 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
3803 | : (1000 * total / plain); | |
3804 | ||
3805 | if (r > 1000) | |
3806 | r = 1000; | |
3807 | ||
3808 | r = 1000 - r; | |
3809 | dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " | |
3810 | "total %u; compression: %u.%u%%\n", | |
3811 | direction, | |
3812 | c->bytes[1], c->packets[1], | |
3813 | c->bytes[0], c->packets[0], | |
3814 | total, r/10, r % 10); | |
3815 | } | |
3816 | ||
3817 | /* Since we are processing the bitfield from lower addresses to higher, | |
3818 | it does not matter if the process it in 32 bit chunks or 64 bit | |
3819 | chunks as long as it is little endian. (Understand it as byte stream, | |
3820 | beginning with the lowest byte...) If we would use big endian | |
3821 | we would need to process it from the highest address to the lowest, | |
3822 | in order to be agnostic to the 32 vs 64 bits issue. | |
3823 | ||
3824 | returns 0 on failure, 1 if we successfully received it. */ | |
4a76b161 | 3825 | static int receive_bitmap(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3826 | { |
4a76b161 | 3827 | struct drbd_conf *mdev; |
b411b363 | 3828 | struct bm_xfer_ctx c; |
2c46407d | 3829 | int err; |
4a76b161 AG |
3830 | struct p_header *h = tconn->data.rbuf; |
3831 | ||
3832 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
3833 | if (!mdev) | |
3834 | return -EIO; | |
b411b363 | 3835 | |
20ceb2b2 LE |
3836 | drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
3837 | /* you are supposed to send additional out-of-sync information | |
3838 | * if you actually set bits during this phase */ | |
b411b363 | 3839 | |
b411b363 PR |
3840 | c = (struct bm_xfer_ctx) { |
3841 | .bm_bits = drbd_bm_bits(mdev), | |
3842 | .bm_words = drbd_bm_words(mdev), | |
3843 | }; | |
3844 | ||
2c46407d | 3845 | for(;;) { |
e2857216 AG |
3846 | if (pi->cmd == P_BITMAP) { |
3847 | err = receive_bitmap_plain(mdev, pi->size, h, &c); | |
3848 | } else if (pi->cmd == P_COMPRESSED_BITMAP) { | |
b411b363 PR |
3849 | /* MAYBE: sanity check that we speak proto >= 90, |
3850 | * and the feature is enabled! */ | |
3851 | struct p_compressed_bm *p; | |
3852 | ||
e2857216 | 3853 | if (pi->size > BM_PACKET_PAYLOAD_BYTES) { |
b411b363 | 3854 | dev_err(DEV, "ReportCBitmap packet too large\n"); |
82bc0194 | 3855 | err = -EIO; |
b411b363 PR |
3856 | goto out; |
3857 | } | |
fc56815c AG |
3858 | |
3859 | p = mdev->tconn->data.rbuf; | |
e2857216 | 3860 | err = drbd_recv_all(mdev->tconn, p->head.payload, pi->size); |
82bc0194 AG |
3861 | if (err) |
3862 | goto out; | |
e2857216 AG |
3863 | if (pi->size <= (sizeof(*p) - sizeof(p->head))) { |
3864 | dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", pi->size); | |
82bc0194 | 3865 | err = -EIO; |
78fcbdae | 3866 | goto out; |
b411b363 | 3867 | } |
e2857216 | 3868 | err = decode_bitmap_c(mdev, p, &c, pi->size); |
b411b363 | 3869 | } else { |
e2857216 | 3870 | dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", pi->cmd); |
82bc0194 | 3871 | err = -EIO; |
b411b363 PR |
3872 | goto out; |
3873 | } | |
3874 | ||
e2857216 AG |
3875 | c.packets[pi->cmd == P_BITMAP]++; |
3876 | c.bytes[pi->cmd == P_BITMAP] += sizeof(struct p_header) + pi->size; | |
b411b363 | 3877 | |
2c46407d AG |
3878 | if (err <= 0) { |
3879 | if (err < 0) | |
3880 | goto out; | |
b411b363 | 3881 | break; |
2c46407d | 3882 | } |
e2857216 | 3883 | err = drbd_recv_header(mdev->tconn, pi); |
82bc0194 | 3884 | if (err) |
b411b363 | 3885 | goto out; |
2c46407d | 3886 | } |
b411b363 PR |
3887 | |
3888 | INFO_bm_xfer_stats(mdev, "receive", &c); | |
3889 | ||
3890 | if (mdev->state.conn == C_WF_BITMAP_T) { | |
de1f8e4a AG |
3891 | enum drbd_state_rv rv; |
3892 | ||
82bc0194 AG |
3893 | err = drbd_send_bitmap(mdev); |
3894 | if (err) | |
b411b363 PR |
3895 | goto out; |
3896 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
de1f8e4a AG |
3897 | rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
3898 | D_ASSERT(rv == SS_SUCCESS); | |
b411b363 PR |
3899 | } else if (mdev->state.conn != C_WF_BITMAP_S) { |
3900 | /* admin may have requested C_DISCONNECTING, | |
3901 | * other threads may have noticed network errors */ | |
3902 | dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", | |
3903 | drbd_conn_str(mdev->state.conn)); | |
3904 | } | |
82bc0194 | 3905 | err = 0; |
b411b363 | 3906 | |
b411b363 | 3907 | out: |
20ceb2b2 | 3908 | drbd_bm_unlock(mdev); |
82bc0194 | 3909 | if (!err && mdev->state.conn == C_WF_BITMAP_S) |
b411b363 | 3910 | drbd_start_resync(mdev, C_SYNC_SOURCE); |
82bc0194 | 3911 | return err; |
b411b363 PR |
3912 | } |
3913 | ||
4a76b161 | 3914 | static int receive_skip(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 3915 | { |
4a76b161 | 3916 | conn_warn(tconn, "skipping unknown optional packet type %d, l: %d!\n", |
e2857216 | 3917 | pi->cmd, pi->size); |
2de876ef | 3918 | |
4a76b161 | 3919 | return ignore_remaining_packet(tconn, pi); |
2de876ef PR |
3920 | } |
3921 | ||
4a76b161 | 3922 | static int receive_UnplugRemote(struct drbd_tconn *tconn, struct packet_info *pi) |
0ced55a3 | 3923 | { |
e7f52dfb LE |
3924 | /* Make sure we've acked all the TCP data associated |
3925 | * with the data requests being unplugged */ | |
4a76b161 | 3926 | drbd_tcp_quickack(tconn->data.socket); |
0ced55a3 | 3927 | |
82bc0194 | 3928 | return 0; |
0ced55a3 PR |
3929 | } |
3930 | ||
4a76b161 | 3931 | static int receive_out_of_sync(struct drbd_tconn *tconn, struct packet_info *pi) |
73a01a18 | 3932 | { |
4a76b161 AG |
3933 | struct drbd_conf *mdev; |
3934 | struct p_block_desc *p = tconn->data.rbuf; | |
3935 | ||
3936 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
3937 | if (!mdev) | |
3938 | return -EIO; | |
73a01a18 | 3939 | |
f735e363 LE |
3940 | switch (mdev->state.conn) { |
3941 | case C_WF_SYNC_UUID: | |
3942 | case C_WF_BITMAP_T: | |
3943 | case C_BEHIND: | |
3944 | break; | |
3945 | default: | |
3946 | dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", | |
3947 | drbd_conn_str(mdev->state.conn)); | |
3948 | } | |
3949 | ||
73a01a18 PR |
3950 | drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
3951 | ||
82bc0194 | 3952 | return 0; |
73a01a18 PR |
3953 | } |
3954 | ||
02918be2 PR |
3955 | struct data_cmd { |
3956 | int expect_payload; | |
3957 | size_t pkt_size; | |
4a76b161 | 3958 | int (*fn)(struct drbd_tconn *, struct packet_info *); |
02918be2 PR |
3959 | }; |
3960 | ||
3961 | static struct data_cmd drbd_cmd_handler[] = { | |
4a76b161 AG |
3962 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, |
3963 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
3964 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
3965 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
3966 | [P_BITMAP] = { 1, sizeof(struct p_header), receive_bitmap } , | |
3967 | [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header), receive_bitmap } , | |
3968 | [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header), receive_UnplugRemote }, | |
3969 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3970 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3971 | [P_SYNC_PARAM] = { 1, sizeof(struct p_header), receive_SyncParam }, | |
3972 | [P_SYNC_PARAM89] = { 1, sizeof(struct p_header), receive_SyncParam }, | |
3973 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, | |
3974 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
3975 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
3976 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
3977 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
3978 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
3979 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3980 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3981 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3982 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
3983 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, | |
3984 | [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state }, | |
b411b363 PR |
3985 | }; |
3986 | ||
eefc2f7d | 3987 | static void drbdd(struct drbd_tconn *tconn) |
b411b363 | 3988 | { |
e6ef8a5c | 3989 | struct p_header *header = tconn->data.rbuf; |
77351055 | 3990 | struct packet_info pi; |
02918be2 | 3991 | size_t shs; /* sub header size */ |
82bc0194 | 3992 | int err; |
b411b363 | 3993 | |
eefc2f7d | 3994 | while (get_t_state(&tconn->receiver) == RUNNING) { |
deebe195 AG |
3995 | struct data_cmd *cmd; |
3996 | ||
eefc2f7d | 3997 | drbd_thread_current_set_cpu(&tconn->receiver); |
69bc7bc3 | 3998 | if (drbd_recv_header(tconn, &pi)) |
02918be2 | 3999 | goto err_out; |
b411b363 | 4000 | |
deebe195 | 4001 | cmd = &drbd_cmd_handler[pi.cmd]; |
4a76b161 | 4002 | if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || !cmd->fn)) { |
eefc2f7d | 4003 | conn_err(tconn, "unknown packet type %d, l: %d!\n", pi.cmd, pi.size); |
02918be2 | 4004 | goto err_out; |
0b33a916 | 4005 | } |
b411b363 | 4006 | |
deebe195 AG |
4007 | shs = cmd->pkt_size - sizeof(struct p_header); |
4008 | if (pi.size - shs > 0 && !cmd->expect_payload) { | |
eefc2f7d | 4009 | conn_err(tconn, "No payload expected %s l:%d\n", cmdname(pi.cmd), pi.size); |
02918be2 | 4010 | goto err_out; |
b411b363 | 4011 | } |
b411b363 | 4012 | |
c13f7e1a | 4013 | if (shs) { |
a5c31904 AG |
4014 | err = drbd_recv_all_warn(tconn, &header->payload, shs); |
4015 | if (err) | |
c13f7e1a | 4016 | goto err_out; |
e2857216 | 4017 | pi.size -= shs; |
c13f7e1a LE |
4018 | } |
4019 | ||
4a76b161 AG |
4020 | err = cmd->fn(tconn, &pi); |
4021 | if (err) { | |
eefc2f7d | 4022 | conn_err(tconn, "error receiving %s, l: %d!\n", |
77351055 | 4023 | cmdname(pi.cmd), pi.size); |
02918be2 | 4024 | goto err_out; |
b411b363 PR |
4025 | } |
4026 | } | |
82bc0194 | 4027 | return; |
b411b363 | 4028 | |
82bc0194 AG |
4029 | err_out: |
4030 | conn_request_state(tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
b411b363 PR |
4031 | } |
4032 | ||
0e29d163 | 4033 | void conn_flush_workqueue(struct drbd_tconn *tconn) |
b411b363 PR |
4034 | { |
4035 | struct drbd_wq_barrier barr; | |
4036 | ||
4037 | barr.w.cb = w_prev_work_done; | |
0e29d163 | 4038 | barr.w.tconn = tconn; |
b411b363 | 4039 | init_completion(&barr.done); |
0e29d163 | 4040 | drbd_queue_work(&tconn->data.work, &barr.w); |
b411b363 PR |
4041 | wait_for_completion(&barr.done); |
4042 | } | |
4043 | ||
360cc740 | 4044 | static void drbd_disconnect(struct drbd_tconn *tconn) |
b411b363 | 4045 | { |
bbeb641c | 4046 | enum drbd_conns oc; |
b411b363 | 4047 | int rv = SS_UNKNOWN_ERROR; |
b411b363 | 4048 | |
bbeb641c | 4049 | if (tconn->cstate == C_STANDALONE) |
b411b363 | 4050 | return; |
b411b363 PR |
4051 | |
4052 | /* asender does not clean up anything. it must not interfere, either */ | |
360cc740 PR |
4053 | drbd_thread_stop(&tconn->asender); |
4054 | drbd_free_sock(tconn); | |
4055 | ||
4056 | idr_for_each(&tconn->volumes, drbd_disconnected, tconn); | |
360cc740 PR |
4057 | conn_info(tconn, "Connection closed\n"); |
4058 | ||
cb703454 PR |
4059 | if (conn_highest_role(tconn) == R_PRIMARY && conn_highest_pdsk(tconn) >= D_UNKNOWN) |
4060 | conn_try_outdate_peer_async(tconn); | |
4061 | ||
360cc740 | 4062 | spin_lock_irq(&tconn->req_lock); |
bbeb641c PR |
4063 | oc = tconn->cstate; |
4064 | if (oc >= C_UNCONNECTED) | |
4065 | rv = _conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE); | |
4066 | ||
360cc740 PR |
4067 | spin_unlock_irq(&tconn->req_lock); |
4068 | ||
bbeb641c | 4069 | if (oc == C_DISCONNECTING) { |
360cc740 PR |
4070 | wait_event(tconn->net_cnt_wait, atomic_read(&tconn->net_cnt) == 0); |
4071 | ||
4072 | crypto_free_hash(tconn->cram_hmac_tfm); | |
4073 | tconn->cram_hmac_tfm = NULL; | |
4074 | ||
4075 | kfree(tconn->net_conf); | |
4076 | tconn->net_conf = NULL; | |
bbeb641c | 4077 | conn_request_state(tconn, NS(conn, C_STANDALONE), CS_VERBOSE); |
360cc740 PR |
4078 | } |
4079 | } | |
4080 | ||
4081 | static int drbd_disconnected(int vnr, void *p, void *data) | |
4082 | { | |
4083 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
4084 | enum drbd_fencing_p fp; | |
4085 | unsigned int i; | |
b411b363 | 4086 | |
85719573 | 4087 | /* wait for current activity to cease. */ |
87eeee41 | 4088 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4089 | _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
4090 | _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); | |
4091 | _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); | |
87eeee41 | 4092 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4093 | |
4094 | /* We do not have data structures that would allow us to | |
4095 | * get the rs_pending_cnt down to 0 again. | |
4096 | * * On C_SYNC_TARGET we do not have any data structures describing | |
4097 | * the pending RSDataRequest's we have sent. | |
4098 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
4099 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
4100 | * And no, it is not the sum of the reference counts in the | |
4101 | * resync_LRU. The resync_LRU tracks the whole operation including | |
4102 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
4103 | * on the fly. */ | |
4104 | drbd_rs_cancel_all(mdev); | |
4105 | mdev->rs_total = 0; | |
4106 | mdev->rs_failed = 0; | |
4107 | atomic_set(&mdev->rs_pending_cnt, 0); | |
4108 | wake_up(&mdev->misc_wait); | |
4109 | ||
7fde2be9 PR |
4110 | del_timer(&mdev->request_timer); |
4111 | ||
b411b363 | 4112 | del_timer_sync(&mdev->resync_timer); |
b411b363 PR |
4113 | resync_timer_fn((unsigned long)mdev); |
4114 | ||
b411b363 PR |
4115 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
4116 | * w_make_resync_request etc. which may still be on the worker queue | |
4117 | * to be "canceled" */ | |
a21e9298 | 4118 | drbd_flush_workqueue(mdev); |
b411b363 PR |
4119 | |
4120 | /* This also does reclaim_net_ee(). If we do this too early, we might | |
4121 | * miss some resync ee and pages.*/ | |
4122 | drbd_process_done_ee(mdev); | |
4123 | ||
4124 | kfree(mdev->p_uuid); | |
4125 | mdev->p_uuid = NULL; | |
4126 | ||
2aebfabb | 4127 | if (!drbd_suspended(mdev)) |
2f5cdd0b | 4128 | tl_clear(mdev->tconn); |
b411b363 | 4129 | |
b411b363 PR |
4130 | drbd_md_sync(mdev); |
4131 | ||
4132 | fp = FP_DONT_CARE; | |
4133 | if (get_ldev(mdev)) { | |
4134 | fp = mdev->ldev->dc.fencing; | |
4135 | put_ldev(mdev); | |
4136 | } | |
4137 | ||
20ceb2b2 LE |
4138 | /* serialize with bitmap writeout triggered by the state change, |
4139 | * if any. */ | |
4140 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); | |
4141 | ||
b411b363 PR |
4142 | /* tcp_close and release of sendpage pages can be deferred. I don't |
4143 | * want to use SO_LINGER, because apparently it can be deferred for | |
4144 | * more than 20 seconds (longest time I checked). | |
4145 | * | |
4146 | * Actually we don't care for exactly when the network stack does its | |
4147 | * put_page(), but release our reference on these pages right here. | |
4148 | */ | |
4149 | i = drbd_release_ee(mdev, &mdev->net_ee); | |
4150 | if (i) | |
4151 | dev_info(DEV, "net_ee not empty, killed %u entries\n", i); | |
435f0740 LE |
4152 | i = atomic_read(&mdev->pp_in_use_by_net); |
4153 | if (i) | |
4154 | dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); | |
b411b363 PR |
4155 | i = atomic_read(&mdev->pp_in_use); |
4156 | if (i) | |
45bb912b | 4157 | dev_info(DEV, "pp_in_use = %d, expected 0\n", i); |
b411b363 PR |
4158 | |
4159 | D_ASSERT(list_empty(&mdev->read_ee)); | |
4160 | D_ASSERT(list_empty(&mdev->active_ee)); | |
4161 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
4162 | D_ASSERT(list_empty(&mdev->done_ee)); | |
4163 | ||
4164 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ | |
4165 | atomic_set(&mdev->current_epoch->epoch_size, 0); | |
4166 | D_ASSERT(list_empty(&mdev->current_epoch->list)); | |
360cc740 PR |
4167 | |
4168 | return 0; | |
b411b363 PR |
4169 | } |
4170 | ||
4171 | /* | |
4172 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
4173 | * we can agree on is stored in agreed_pro_version. | |
4174 | * | |
4175 | * feature flags and the reserved array should be enough room for future | |
4176 | * enhancements of the handshake protocol, and possible plugins... | |
4177 | * | |
4178 | * for now, they are expected to be zero, but ignored. | |
4179 | */ | |
6038178e | 4180 | static int drbd_send_features(struct drbd_tconn *tconn) |
b411b363 | 4181 | { |
e6b3ea83 | 4182 | /* ASSERT current == mdev->tconn->receiver ... */ |
6038178e | 4183 | struct p_connection_features *p = tconn->data.sbuf; |
e8d17b01 | 4184 | int err; |
b411b363 | 4185 | |
8a22cccc PR |
4186 | if (mutex_lock_interruptible(&tconn->data.mutex)) { |
4187 | conn_err(tconn, "interrupted during initial handshake\n"); | |
e8d17b01 | 4188 | return -EINTR; |
b411b363 PR |
4189 | } |
4190 | ||
8a22cccc PR |
4191 | if (tconn->data.socket == NULL) { |
4192 | mutex_unlock(&tconn->data.mutex); | |
e8d17b01 | 4193 | return -EIO; |
b411b363 PR |
4194 | } |
4195 | ||
4196 | memset(p, 0, sizeof(*p)); | |
4197 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
4198 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
6038178e | 4199 | err = _conn_send_cmd(tconn, 0, &tconn->data, P_CONNECTION_FEATURES, |
ecf2363c | 4200 | &p->head, sizeof(*p), 0); |
8a22cccc | 4201 | mutex_unlock(&tconn->data.mutex); |
e8d17b01 | 4202 | return err; |
b411b363 PR |
4203 | } |
4204 | ||
4205 | /* | |
4206 | * return values: | |
4207 | * 1 yes, we have a valid connection | |
4208 | * 0 oops, did not work out, please try again | |
4209 | * -1 peer talks different language, | |
4210 | * no point in trying again, please go standalone. | |
4211 | */ | |
6038178e | 4212 | static int drbd_do_features(struct drbd_tconn *tconn) |
b411b363 | 4213 | { |
65d11ed6 | 4214 | /* ASSERT current == tconn->receiver ... */ |
6038178e AG |
4215 | struct p_connection_features *p = tconn->data.rbuf; |
4216 | const int expect = sizeof(struct p_connection_features) - sizeof(struct p_header80); | |
77351055 | 4217 | struct packet_info pi; |
a5c31904 | 4218 | int err; |
b411b363 | 4219 | |
6038178e | 4220 | err = drbd_send_features(tconn); |
e8d17b01 | 4221 | if (err) |
b411b363 PR |
4222 | return 0; |
4223 | ||
69bc7bc3 AG |
4224 | err = drbd_recv_header(tconn, &pi); |
4225 | if (err) | |
b411b363 PR |
4226 | return 0; |
4227 | ||
6038178e AG |
4228 | if (pi.cmd != P_CONNECTION_FEATURES) { |
4229 | conn_err(tconn, "expected ConnectionFeatures packet, received: %s (0x%04x)\n", | |
77351055 | 4230 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4231 | return -1; |
4232 | } | |
4233 | ||
77351055 | 4234 | if (pi.size != expect) { |
6038178e | 4235 | conn_err(tconn, "expected ConnectionFeatures length: %u, received: %u\n", |
77351055 | 4236 | expect, pi.size); |
b411b363 PR |
4237 | return -1; |
4238 | } | |
4239 | ||
a5c31904 AG |
4240 | err = drbd_recv_all_warn(tconn, &p->head.payload, expect); |
4241 | if (err) | |
b411b363 | 4242 | return 0; |
b411b363 | 4243 | |
b411b363 PR |
4244 | p->protocol_min = be32_to_cpu(p->protocol_min); |
4245 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
4246 | if (p->protocol_max == 0) | |
4247 | p->protocol_max = p->protocol_min; | |
4248 | ||
4249 | if (PRO_VERSION_MAX < p->protocol_min || | |
4250 | PRO_VERSION_MIN > p->protocol_max) | |
4251 | goto incompat; | |
4252 | ||
65d11ed6 | 4253 | tconn->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); |
b411b363 | 4254 | |
65d11ed6 PR |
4255 | conn_info(tconn, "Handshake successful: " |
4256 | "Agreed network protocol version %d\n", tconn->agreed_pro_version); | |
b411b363 PR |
4257 | |
4258 | return 1; | |
4259 | ||
4260 | incompat: | |
65d11ed6 | 4261 | conn_err(tconn, "incompatible DRBD dialects: " |
b411b363 PR |
4262 | "I support %d-%d, peer supports %d-%d\n", |
4263 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
4264 | p->protocol_min, p->protocol_max); | |
4265 | return -1; | |
4266 | } | |
4267 | ||
4268 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
13e6037d | 4269 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 PR |
4270 | { |
4271 | dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); | |
4272 | dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4273 | return -1; |
b411b363 PR |
4274 | } |
4275 | #else | |
4276 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4277 | |
4278 | /* Return value: | |
4279 | 1 - auth succeeded, | |
4280 | 0 - failed, try again (network error), | |
4281 | -1 - auth failed, don't try again. | |
4282 | */ | |
4283 | ||
13e6037d | 4284 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 PR |
4285 | { |
4286 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ | |
4287 | struct scatterlist sg; | |
4288 | char *response = NULL; | |
4289 | char *right_response = NULL; | |
4290 | char *peers_ch = NULL; | |
13e6037d | 4291 | unsigned int key_len = strlen(tconn->net_conf->shared_secret); |
b411b363 PR |
4292 | unsigned int resp_size; |
4293 | struct hash_desc desc; | |
77351055 | 4294 | struct packet_info pi; |
69bc7bc3 | 4295 | int err, rv; |
b411b363 | 4296 | |
13e6037d | 4297 | desc.tfm = tconn->cram_hmac_tfm; |
b411b363 PR |
4298 | desc.flags = 0; |
4299 | ||
13e6037d PR |
4300 | rv = crypto_hash_setkey(tconn->cram_hmac_tfm, |
4301 | (u8 *)tconn->net_conf->shared_secret, key_len); | |
b411b363 | 4302 | if (rv) { |
13e6037d | 4303 | conn_err(tconn, "crypto_hash_setkey() failed with %d\n", rv); |
b10d96cb | 4304 | rv = -1; |
b411b363 PR |
4305 | goto fail; |
4306 | } | |
4307 | ||
4308 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4309 | ||
ce9879cb | 4310 | rv = !conn_send_cmd2(tconn, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); |
b411b363 PR |
4311 | if (!rv) |
4312 | goto fail; | |
4313 | ||
69bc7bc3 AG |
4314 | err = drbd_recv_header(tconn, &pi); |
4315 | if (err) { | |
4316 | rv = 0; | |
b411b363 | 4317 | goto fail; |
69bc7bc3 | 4318 | } |
b411b363 | 4319 | |
77351055 | 4320 | if (pi.cmd != P_AUTH_CHALLENGE) { |
13e6037d | 4321 | conn_err(tconn, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
77351055 | 4322 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4323 | rv = 0; |
4324 | goto fail; | |
4325 | } | |
4326 | ||
77351055 | 4327 | if (pi.size > CHALLENGE_LEN * 2) { |
13e6037d | 4328 | conn_err(tconn, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4329 | rv = -1; |
b411b363 PR |
4330 | goto fail; |
4331 | } | |
4332 | ||
77351055 | 4333 | peers_ch = kmalloc(pi.size, GFP_NOIO); |
b411b363 | 4334 | if (peers_ch == NULL) { |
13e6037d | 4335 | conn_err(tconn, "kmalloc of peers_ch failed\n"); |
b10d96cb | 4336 | rv = -1; |
b411b363 PR |
4337 | goto fail; |
4338 | } | |
4339 | ||
a5c31904 AG |
4340 | err = drbd_recv_all_warn(tconn, peers_ch, pi.size); |
4341 | if (err) { | |
b411b363 PR |
4342 | rv = 0; |
4343 | goto fail; | |
4344 | } | |
4345 | ||
13e6037d | 4346 | resp_size = crypto_hash_digestsize(tconn->cram_hmac_tfm); |
b411b363 PR |
4347 | response = kmalloc(resp_size, GFP_NOIO); |
4348 | if (response == NULL) { | |
13e6037d | 4349 | conn_err(tconn, "kmalloc of response failed\n"); |
b10d96cb | 4350 | rv = -1; |
b411b363 PR |
4351 | goto fail; |
4352 | } | |
4353 | ||
4354 | sg_init_table(&sg, 1); | |
77351055 | 4355 | sg_set_buf(&sg, peers_ch, pi.size); |
b411b363 PR |
4356 | |
4357 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4358 | if (rv) { | |
13e6037d | 4359 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4360 | rv = -1; |
b411b363 PR |
4361 | goto fail; |
4362 | } | |
4363 | ||
ce9879cb | 4364 | rv = !conn_send_cmd2(tconn, P_AUTH_RESPONSE, response, resp_size); |
b411b363 PR |
4365 | if (!rv) |
4366 | goto fail; | |
4367 | ||
69bc7bc3 AG |
4368 | err = drbd_recv_header(tconn, &pi); |
4369 | if (err) { | |
4370 | rv = 0; | |
b411b363 | 4371 | goto fail; |
69bc7bc3 | 4372 | } |
b411b363 | 4373 | |
77351055 | 4374 | if (pi.cmd != P_AUTH_RESPONSE) { |
13e6037d | 4375 | conn_err(tconn, "expected AuthResponse packet, received: %s (0x%04x)\n", |
77351055 | 4376 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4377 | rv = 0; |
4378 | goto fail; | |
4379 | } | |
4380 | ||
77351055 | 4381 | if (pi.size != resp_size) { |
13e6037d | 4382 | conn_err(tconn, "expected AuthResponse payload of wrong size\n"); |
b411b363 PR |
4383 | rv = 0; |
4384 | goto fail; | |
4385 | } | |
4386 | ||
a5c31904 AG |
4387 | err = drbd_recv_all_warn(tconn, response , resp_size); |
4388 | if (err) { | |
b411b363 PR |
4389 | rv = 0; |
4390 | goto fail; | |
4391 | } | |
4392 | ||
4393 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4394 | if (right_response == NULL) { |
13e6037d | 4395 | conn_err(tconn, "kmalloc of right_response failed\n"); |
b10d96cb | 4396 | rv = -1; |
b411b363 PR |
4397 | goto fail; |
4398 | } | |
4399 | ||
4400 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4401 | ||
4402 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4403 | if (rv) { | |
13e6037d | 4404 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4405 | rv = -1; |
b411b363 PR |
4406 | goto fail; |
4407 | } | |
4408 | ||
4409 | rv = !memcmp(response, right_response, resp_size); | |
4410 | ||
4411 | if (rv) | |
13e6037d PR |
4412 | conn_info(tconn, "Peer authenticated using %d bytes of '%s' HMAC\n", |
4413 | resp_size, tconn->net_conf->cram_hmac_alg); | |
b10d96cb JT |
4414 | else |
4415 | rv = -1; | |
b411b363 PR |
4416 | |
4417 | fail: | |
4418 | kfree(peers_ch); | |
4419 | kfree(response); | |
4420 | kfree(right_response); | |
4421 | ||
4422 | return rv; | |
4423 | } | |
4424 | #endif | |
4425 | ||
4426 | int drbdd_init(struct drbd_thread *thi) | |
4427 | { | |
392c8801 | 4428 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 PR |
4429 | int h; |
4430 | ||
4d641dd7 | 4431 | conn_info(tconn, "receiver (re)started\n"); |
b411b363 PR |
4432 | |
4433 | do { | |
4d641dd7 | 4434 | h = drbd_connect(tconn); |
b411b363 | 4435 | if (h == 0) { |
4d641dd7 | 4436 | drbd_disconnect(tconn); |
20ee6390 | 4437 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
4438 | } |
4439 | if (h == -1) { | |
4d641dd7 | 4440 | conn_warn(tconn, "Discarding network configuration.\n"); |
bbeb641c | 4441 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
4442 | } |
4443 | } while (h == 0); | |
4444 | ||
4445 | if (h > 0) { | |
4d641dd7 PR |
4446 | if (get_net_conf(tconn)) { |
4447 | drbdd(tconn); | |
4448 | put_net_conf(tconn); | |
b411b363 PR |
4449 | } |
4450 | } | |
4451 | ||
4d641dd7 | 4452 | drbd_disconnect(tconn); |
b411b363 | 4453 | |
4d641dd7 | 4454 | conn_info(tconn, "receiver terminated\n"); |
b411b363 PR |
4455 | return 0; |
4456 | } | |
4457 | ||
4458 | /* ********* acknowledge sender ******** */ | |
4459 | ||
e05e1e59 | 4460 | static int got_conn_RqSReply(struct drbd_tconn *tconn, struct packet_info *pi) |
e4f78ede | 4461 | { |
e6ef8a5c | 4462 | struct p_req_state_reply *p = tconn->meta.rbuf; |
e4f78ede PR |
4463 | int retcode = be32_to_cpu(p->retcode); |
4464 | ||
4465 | if (retcode >= SS_SUCCESS) { | |
4466 | set_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags); | |
4467 | } else { | |
4468 | set_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags); | |
4469 | conn_err(tconn, "Requested state change failed by peer: %s (%d)\n", | |
4470 | drbd_set_st_err_str(retcode), retcode); | |
4471 | } | |
4472 | wake_up(&tconn->ping_wait); | |
4473 | ||
4474 | return true; | |
4475 | } | |
4476 | ||
1952e916 | 4477 | static int got_RqSReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4478 | { |
1952e916 AG |
4479 | struct drbd_conf *mdev; |
4480 | struct p_req_state_reply *p = tconn->meta.rbuf; | |
b411b363 PR |
4481 | int retcode = be32_to_cpu(p->retcode); |
4482 | ||
1952e916 AG |
4483 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4484 | if (!mdev) | |
4485 | return false; | |
4486 | ||
e4f78ede PR |
4487 | if (retcode >= SS_SUCCESS) { |
4488 | set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); | |
4489 | } else { | |
4490 | set_bit(CL_ST_CHG_FAIL, &mdev->flags); | |
4491 | dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", | |
4492 | drbd_set_st_err_str(retcode), retcode); | |
b411b363 | 4493 | } |
e4f78ede PR |
4494 | wake_up(&mdev->state_wait); |
4495 | ||
81e84650 | 4496 | return true; |
b411b363 PR |
4497 | } |
4498 | ||
e05e1e59 | 4499 | static int got_Ping(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4500 | { |
a17647aa | 4501 | return !drbd_send_ping_ack(tconn); |
b411b363 PR |
4502 | |
4503 | } | |
4504 | ||
e05e1e59 | 4505 | static int got_PingAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 PR |
4506 | { |
4507 | /* restore idle timeout */ | |
2a67d8b9 PR |
4508 | tconn->meta.socket->sk->sk_rcvtimeo = tconn->net_conf->ping_int*HZ; |
4509 | if (!test_and_set_bit(GOT_PING_ACK, &tconn->flags)) | |
4510 | wake_up(&tconn->ping_wait); | |
b411b363 | 4511 | |
81e84650 | 4512 | return true; |
b411b363 PR |
4513 | } |
4514 | ||
1952e916 | 4515 | static int got_IsInSync(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4516 | { |
1952e916 AG |
4517 | struct drbd_conf *mdev; |
4518 | struct p_block_ack *p = tconn->meta.rbuf; | |
b411b363 PR |
4519 | sector_t sector = be64_to_cpu(p->sector); |
4520 | int blksize = be32_to_cpu(p->blksize); | |
4521 | ||
1952e916 AG |
4522 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4523 | if (!mdev) | |
4524 | return false; | |
4525 | ||
31890f4a | 4526 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
b411b363 PR |
4527 | |
4528 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4529 | ||
1d53f09e LE |
4530 | if (get_ldev(mdev)) { |
4531 | drbd_rs_complete_io(mdev, sector); | |
4532 | drbd_set_in_sync(mdev, sector, blksize); | |
4533 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ | |
4534 | mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); | |
4535 | put_ldev(mdev); | |
4536 | } | |
b411b363 | 4537 | dec_rs_pending(mdev); |
778f271d | 4538 | atomic_add(blksize >> 9, &mdev->rs_sect_in); |
b411b363 | 4539 | |
81e84650 | 4540 | return true; |
b411b363 PR |
4541 | } |
4542 | ||
bc9c5c41 AG |
4543 | static int |
4544 | validate_req_change_req_state(struct drbd_conf *mdev, u64 id, sector_t sector, | |
4545 | struct rb_root *root, const char *func, | |
4546 | enum drbd_req_event what, bool missing_ok) | |
b411b363 PR |
4547 | { |
4548 | struct drbd_request *req; | |
4549 | struct bio_and_error m; | |
4550 | ||
87eeee41 | 4551 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 4552 | req = find_request(mdev, root, id, sector, missing_ok, func); |
b411b363 | 4553 | if (unlikely(!req)) { |
87eeee41 | 4554 | spin_unlock_irq(&mdev->tconn->req_lock); |
81e84650 | 4555 | return false; |
b411b363 PR |
4556 | } |
4557 | __req_mod(req, what, &m); | |
87eeee41 | 4558 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4559 | |
4560 | if (m.bio) | |
4561 | complete_master_bio(mdev, &m); | |
81e84650 | 4562 | return true; |
b411b363 PR |
4563 | } |
4564 | ||
1952e916 | 4565 | static int got_BlockAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4566 | { |
1952e916 AG |
4567 | struct drbd_conf *mdev; |
4568 | struct p_block_ack *p = tconn->meta.rbuf; | |
b411b363 PR |
4569 | sector_t sector = be64_to_cpu(p->sector); |
4570 | int blksize = be32_to_cpu(p->blksize); | |
4571 | enum drbd_req_event what; | |
4572 | ||
1952e916 AG |
4573 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4574 | if (!mdev) | |
4575 | return false; | |
4576 | ||
b411b363 PR |
4577 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); |
4578 | ||
579b57ed | 4579 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4580 | drbd_set_in_sync(mdev, sector, blksize); |
4581 | dec_rs_pending(mdev); | |
81e84650 | 4582 | return true; |
b411b363 | 4583 | } |
e05e1e59 | 4584 | switch (pi->cmd) { |
b411b363 | 4585 | case P_RS_WRITE_ACK: |
89e58e75 | 4586 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
8554df1c | 4587 | what = WRITE_ACKED_BY_PEER_AND_SIS; |
b411b363 PR |
4588 | break; |
4589 | case P_WRITE_ACK: | |
89e58e75 | 4590 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
8554df1c | 4591 | what = WRITE_ACKED_BY_PEER; |
b411b363 PR |
4592 | break; |
4593 | case P_RECV_ACK: | |
89e58e75 | 4594 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_B); |
8554df1c | 4595 | what = RECV_ACKED_BY_PEER; |
b411b363 | 4596 | break; |
7be8da07 | 4597 | case P_DISCARD_WRITE: |
89e58e75 | 4598 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
7be8da07 AG |
4599 | what = DISCARD_WRITE; |
4600 | break; | |
4601 | case P_RETRY_WRITE: | |
4602 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); | |
4603 | what = POSTPONE_WRITE; | |
b411b363 PR |
4604 | break; |
4605 | default: | |
4606 | D_ASSERT(0); | |
81e84650 | 4607 | return false; |
b411b363 PR |
4608 | } |
4609 | ||
4610 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
bc9c5c41 AG |
4611 | &mdev->write_requests, __func__, |
4612 | what, false); | |
b411b363 PR |
4613 | } |
4614 | ||
1952e916 | 4615 | static int got_NegAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4616 | { |
1952e916 AG |
4617 | struct drbd_conf *mdev; |
4618 | struct p_block_ack *p = tconn->meta.rbuf; | |
b411b363 | 4619 | sector_t sector = be64_to_cpu(p->sector); |
2deb8336 | 4620 | int size = be32_to_cpu(p->blksize); |
1952e916 AG |
4621 | bool missing_ok = tconn->net_conf->wire_protocol == DRBD_PROT_A || |
4622 | tconn->net_conf->wire_protocol == DRBD_PROT_B; | |
c3afd8f5 | 4623 | bool found; |
b411b363 | 4624 | |
1952e916 AG |
4625 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4626 | if (!mdev) | |
4627 | return false; | |
4628 | ||
b411b363 PR |
4629 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); |
4630 | ||
579b57ed | 4631 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4632 | dec_rs_pending(mdev); |
4633 | drbd_rs_failed_io(mdev, sector, size); | |
81e84650 | 4634 | return true; |
b411b363 | 4635 | } |
2deb8336 | 4636 | |
c3afd8f5 | 4637 | found = validate_req_change_req_state(mdev, p->block_id, sector, |
bc9c5c41 | 4638 | &mdev->write_requests, __func__, |
8554df1c | 4639 | NEG_ACKED, missing_ok); |
c3afd8f5 AG |
4640 | if (!found) { |
4641 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. | |
4642 | The master bio might already be completed, therefore the | |
4643 | request is no longer in the collision hash. */ | |
4644 | /* In Protocol B we might already have got a P_RECV_ACK | |
4645 | but then get a P_NEG_ACK afterwards. */ | |
4646 | if (!missing_ok) | |
2deb8336 | 4647 | return false; |
c3afd8f5 | 4648 | drbd_set_out_of_sync(mdev, sector, size); |
2deb8336 | 4649 | } |
2deb8336 | 4650 | return true; |
b411b363 PR |
4651 | } |
4652 | ||
1952e916 | 4653 | static int got_NegDReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4654 | { |
1952e916 AG |
4655 | struct drbd_conf *mdev; |
4656 | struct p_block_ack *p = tconn->meta.rbuf; | |
b411b363 PR |
4657 | sector_t sector = be64_to_cpu(p->sector); |
4658 | ||
1952e916 AG |
4659 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4660 | if (!mdev) | |
4661 | return false; | |
4662 | ||
b411b363 | 4663 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); |
7be8da07 | 4664 | |
b411b363 PR |
4665 | dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", |
4666 | (unsigned long long)sector, be32_to_cpu(p->blksize)); | |
4667 | ||
4668 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
bc9c5c41 | 4669 | &mdev->read_requests, __func__, |
8554df1c | 4670 | NEG_ACKED, false); |
b411b363 PR |
4671 | } |
4672 | ||
1952e916 | 4673 | static int got_NegRSDReply(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4674 | { |
1952e916 | 4675 | struct drbd_conf *mdev; |
b411b363 PR |
4676 | sector_t sector; |
4677 | int size; | |
1952e916 AG |
4678 | struct p_block_ack *p = tconn->meta.rbuf; |
4679 | ||
4680 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
4681 | if (!mdev) | |
4682 | return false; | |
b411b363 PR |
4683 | |
4684 | sector = be64_to_cpu(p->sector); | |
4685 | size = be32_to_cpu(p->blksize); | |
b411b363 PR |
4686 | |
4687 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4688 | ||
4689 | dec_rs_pending(mdev); | |
4690 | ||
4691 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
4692 | drbd_rs_complete_io(mdev, sector); | |
e05e1e59 | 4693 | switch (pi->cmd) { |
d612d309 PR |
4694 | case P_NEG_RS_DREPLY: |
4695 | drbd_rs_failed_io(mdev, sector, size); | |
4696 | case P_RS_CANCEL: | |
4697 | break; | |
4698 | default: | |
4699 | D_ASSERT(0); | |
4700 | put_ldev(mdev); | |
4701 | return false; | |
4702 | } | |
b411b363 PR |
4703 | put_ldev(mdev); |
4704 | } | |
4705 | ||
81e84650 | 4706 | return true; |
b411b363 PR |
4707 | } |
4708 | ||
1952e916 | 4709 | static int got_BarrierAck(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4710 | { |
1952e916 AG |
4711 | struct drbd_conf *mdev; |
4712 | struct p_barrier_ack *p = tconn->meta.rbuf; | |
4713 | ||
4714 | mdev = vnr_to_mdev(tconn, pi->vnr); | |
4715 | if (!mdev) | |
4716 | return false; | |
b411b363 | 4717 | |
2f5cdd0b | 4718 | tl_release(mdev->tconn, p->barrier, be32_to_cpu(p->set_size)); |
b411b363 | 4719 | |
c4752ef1 PR |
4720 | if (mdev->state.conn == C_AHEAD && |
4721 | atomic_read(&mdev->ap_in_flight) == 0 && | |
370a43e7 PR |
4722 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags)) { |
4723 | mdev->start_resync_timer.expires = jiffies + HZ; | |
4724 | add_timer(&mdev->start_resync_timer); | |
c4752ef1 PR |
4725 | } |
4726 | ||
81e84650 | 4727 | return true; |
b411b363 PR |
4728 | } |
4729 | ||
1952e916 | 4730 | static int got_OVResult(struct drbd_tconn *tconn, struct packet_info *pi) |
b411b363 | 4731 | { |
1952e916 AG |
4732 | struct drbd_conf *mdev; |
4733 | struct p_block_ack *p = tconn->meta.rbuf; | |
b411b363 PR |
4734 | struct drbd_work *w; |
4735 | sector_t sector; | |
4736 | int size; | |
4737 | ||
1952e916 AG |
4738 | mdev = vnr_to_mdev(tconn, pi->vnr); |
4739 | if (!mdev) | |
4740 | return false; | |
4741 | ||
b411b363 PR |
4742 | sector = be64_to_cpu(p->sector); |
4743 | size = be32_to_cpu(p->blksize); | |
4744 | ||
4745 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4746 | ||
4747 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
8f7bed77 | 4748 | drbd_ov_out_of_sync_found(mdev, sector, size); |
b411b363 | 4749 | else |
8f7bed77 | 4750 | ov_out_of_sync_print(mdev); |
b411b363 | 4751 | |
1d53f09e | 4752 | if (!get_ldev(mdev)) |
81e84650 | 4753 | return true; |
1d53f09e | 4754 | |
b411b363 PR |
4755 | drbd_rs_complete_io(mdev, sector); |
4756 | dec_rs_pending(mdev); | |
4757 | ||
ea5442af LE |
4758 | --mdev->ov_left; |
4759 | ||
4760 | /* let's advance progress step marks only for every other megabyte */ | |
4761 | if ((mdev->ov_left & 0x200) == 0x200) | |
4762 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
4763 | ||
4764 | if (mdev->ov_left == 0) { | |
b411b363 PR |
4765 | w = kmalloc(sizeof(*w), GFP_NOIO); |
4766 | if (w) { | |
4767 | w->cb = w_ov_finished; | |
a21e9298 | 4768 | w->mdev = mdev; |
e42325a5 | 4769 | drbd_queue_work_front(&mdev->tconn->data.work, w); |
b411b363 PR |
4770 | } else { |
4771 | dev_err(DEV, "kmalloc(w) failed."); | |
8f7bed77 | 4772 | ov_out_of_sync_print(mdev); |
b411b363 PR |
4773 | drbd_resync_finished(mdev); |
4774 | } | |
4775 | } | |
1d53f09e | 4776 | put_ldev(mdev); |
81e84650 | 4777 | return true; |
b411b363 PR |
4778 | } |
4779 | ||
1952e916 | 4780 | static int got_skip(struct drbd_tconn *tconn, struct packet_info *pi) |
0ced55a3 | 4781 | { |
81e84650 | 4782 | return true; |
0ced55a3 PR |
4783 | } |
4784 | ||
32862ec7 PR |
4785 | static int tconn_process_done_ee(struct drbd_tconn *tconn) |
4786 | { | |
082a3439 PR |
4787 | struct drbd_conf *mdev; |
4788 | int i, not_empty = 0; | |
32862ec7 PR |
4789 | |
4790 | do { | |
4791 | clear_bit(SIGNAL_ASENDER, &tconn->flags); | |
4792 | flush_signals(current); | |
082a3439 | 4793 | idr_for_each_entry(&tconn->volumes, mdev, i) { |
e2b3032b | 4794 | if (drbd_process_done_ee(mdev)) |
082a3439 PR |
4795 | return 1; /* error */ |
4796 | } | |
32862ec7 | 4797 | set_bit(SIGNAL_ASENDER, &tconn->flags); |
082a3439 PR |
4798 | |
4799 | spin_lock_irq(&tconn->req_lock); | |
4800 | idr_for_each_entry(&tconn->volumes, mdev, i) { | |
4801 | not_empty = !list_empty(&mdev->done_ee); | |
4802 | if (not_empty) | |
4803 | break; | |
4804 | } | |
4805 | spin_unlock_irq(&tconn->req_lock); | |
32862ec7 PR |
4806 | } while (not_empty); |
4807 | ||
4808 | return 0; | |
4809 | } | |
4810 | ||
7201b972 AG |
4811 | struct asender_cmd { |
4812 | size_t pkt_size; | |
1952e916 | 4813 | int (*fn)(struct drbd_tconn *tconn, struct packet_info *); |
7201b972 AG |
4814 | }; |
4815 | ||
4816 | static struct asender_cmd asender_tbl[] = { | |
1952e916 AG |
4817 | [P_PING] = { sizeof(struct p_header), got_Ping }, |
4818 | [P_PING_ACK] = { sizeof(struct p_header), got_PingAck }, | |
4819 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4820 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4821 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4822 | [P_DISCARD_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4823 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, | |
4824 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
4825 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply }, | |
4826 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, | |
4827 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
4828 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
4829 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
4830 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, | |
4831 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply }, | |
4832 | [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_conn_RqSReply }, | |
4833 | [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, | |
7201b972 AG |
4834 | }; |
4835 | ||
b411b363 PR |
4836 | int drbd_asender(struct drbd_thread *thi) |
4837 | { | |
392c8801 | 4838 | struct drbd_tconn *tconn = thi->tconn; |
e6ef8a5c | 4839 | struct p_header *h = tconn->meta.rbuf; |
b411b363 | 4840 | struct asender_cmd *cmd = NULL; |
77351055 | 4841 | struct packet_info pi; |
257d0af6 | 4842 | int rv; |
b411b363 PR |
4843 | void *buf = h; |
4844 | int received = 0; | |
257d0af6 | 4845 | int expect = sizeof(struct p_header); |
f36af18c | 4846 | int ping_timeout_active = 0; |
b411b363 | 4847 | |
b411b363 PR |
4848 | current->policy = SCHED_RR; /* Make this a realtime task! */ |
4849 | current->rt_priority = 2; /* more important than all other tasks */ | |
4850 | ||
e77a0a5c | 4851 | while (get_t_state(thi) == RUNNING) { |
80822284 | 4852 | drbd_thread_current_set_cpu(thi); |
32862ec7 | 4853 | if (test_and_clear_bit(SEND_PING, &tconn->flags)) { |
a17647aa | 4854 | if (drbd_send_ping(tconn)) { |
32862ec7 | 4855 | conn_err(tconn, "drbd_send_ping has failed\n"); |
841ce241 AG |
4856 | goto reconnect; |
4857 | } | |
32862ec7 PR |
4858 | tconn->meta.socket->sk->sk_rcvtimeo = |
4859 | tconn->net_conf->ping_timeo*HZ/10; | |
f36af18c | 4860 | ping_timeout_active = 1; |
b411b363 PR |
4861 | } |
4862 | ||
32862ec7 PR |
4863 | /* TODO: conditionally cork; it may hurt latency if we cork without |
4864 | much to send */ | |
4865 | if (!tconn->net_conf->no_cork) | |
4866 | drbd_tcp_cork(tconn->meta.socket); | |
082a3439 PR |
4867 | if (tconn_process_done_ee(tconn)) { |
4868 | conn_err(tconn, "tconn_process_done_ee() failed\n"); | |
32862ec7 | 4869 | goto reconnect; |
082a3439 | 4870 | } |
b411b363 | 4871 | /* but unconditionally uncork unless disabled */ |
32862ec7 PR |
4872 | if (!tconn->net_conf->no_cork) |
4873 | drbd_tcp_uncork(tconn->meta.socket); | |
b411b363 PR |
4874 | |
4875 | /* short circuit, recv_msg would return EINTR anyways. */ | |
4876 | if (signal_pending(current)) | |
4877 | continue; | |
4878 | ||
32862ec7 PR |
4879 | rv = drbd_recv_short(tconn->meta.socket, buf, expect-received, 0); |
4880 | clear_bit(SIGNAL_ASENDER, &tconn->flags); | |
b411b363 PR |
4881 | |
4882 | flush_signals(current); | |
4883 | ||
4884 | /* Note: | |
4885 | * -EINTR (on meta) we got a signal | |
4886 | * -EAGAIN (on meta) rcvtimeo expired | |
4887 | * -ECONNRESET other side closed the connection | |
4888 | * -ERESTARTSYS (on data) we got a signal | |
4889 | * rv < 0 other than above: unexpected error! | |
4890 | * rv == expected: full header or command | |
4891 | * rv < expected: "woken" by signal during receive | |
4892 | * rv == 0 : "connection shut down by peer" | |
4893 | */ | |
4894 | if (likely(rv > 0)) { | |
4895 | received += rv; | |
4896 | buf += rv; | |
4897 | } else if (rv == 0) { | |
32862ec7 | 4898 | conn_err(tconn, "meta connection shut down by peer.\n"); |
b411b363 PR |
4899 | goto reconnect; |
4900 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
4901 | /* If the data socket received something meanwhile, |
4902 | * that is good enough: peer is still alive. */ | |
32862ec7 PR |
4903 | if (time_after(tconn->last_received, |
4904 | jiffies - tconn->meta.socket->sk->sk_rcvtimeo)) | |
cb6518cb | 4905 | continue; |
f36af18c | 4906 | if (ping_timeout_active) { |
32862ec7 | 4907 | conn_err(tconn, "PingAck did not arrive in time.\n"); |
b411b363 PR |
4908 | goto reconnect; |
4909 | } | |
32862ec7 | 4910 | set_bit(SEND_PING, &tconn->flags); |
b411b363 PR |
4911 | continue; |
4912 | } else if (rv == -EINTR) { | |
4913 | continue; | |
4914 | } else { | |
32862ec7 | 4915 | conn_err(tconn, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
4916 | goto reconnect; |
4917 | } | |
4918 | ||
4919 | if (received == expect && cmd == NULL) { | |
8172f3e9 | 4920 | if (decode_header(tconn, h, &pi)) |
b411b363 | 4921 | goto reconnect; |
7201b972 | 4922 | cmd = &asender_tbl[pi.cmd]; |
1952e916 | 4923 | if (pi.cmd >= ARRAY_SIZE(asender_tbl) || !cmd->fn) { |
32862ec7 | 4924 | conn_err(tconn, "unknown command %d on meta (l: %d)\n", |
77351055 | 4925 | pi.cmd, pi.size); |
b411b363 PR |
4926 | goto disconnect; |
4927 | } | |
4928 | expect = cmd->pkt_size; | |
77351055 | 4929 | if (pi.size != expect - sizeof(struct p_header)) { |
32862ec7 | 4930 | conn_err(tconn, "Wrong packet size on meta (c: %d, l: %d)\n", |
77351055 | 4931 | pi.cmd, pi.size); |
b411b363 | 4932 | goto reconnect; |
257d0af6 | 4933 | } |
b411b363 PR |
4934 | } |
4935 | if (received == expect) { | |
a4fbda8e PR |
4936 | bool rv; |
4937 | ||
1952e916 AG |
4938 | rv = cmd->fn(tconn, &pi); |
4939 | if (!rv) { | |
4940 | conn_err(tconn, "%pf failed\n", cmd->fn); | |
b411b363 | 4941 | goto reconnect; |
1952e916 | 4942 | } |
b411b363 | 4943 | |
a4fbda8e PR |
4944 | tconn->last_received = jiffies; |
4945 | ||
f36af18c LE |
4946 | /* the idle_timeout (ping-int) |
4947 | * has been restored in got_PingAck() */ | |
7201b972 | 4948 | if (cmd == &asender_tbl[P_PING_ACK]) |
f36af18c LE |
4949 | ping_timeout_active = 0; |
4950 | ||
b411b363 PR |
4951 | buf = h; |
4952 | received = 0; | |
257d0af6 | 4953 | expect = sizeof(struct p_header); |
b411b363 PR |
4954 | cmd = NULL; |
4955 | } | |
4956 | } | |
4957 | ||
4958 | if (0) { | |
4959 | reconnect: | |
bbeb641c | 4960 | conn_request_state(tconn, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
b411b363 PR |
4961 | } |
4962 | if (0) { | |
4963 | disconnect: | |
bbeb641c | 4964 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 4965 | } |
32862ec7 | 4966 | clear_bit(SIGNAL_ASENDER, &tconn->flags); |
b411b363 | 4967 | |
32862ec7 | 4968 | conn_info(tconn, "asender terminated\n"); |
b411b363 PR |
4969 | |
4970 | return 0; | |
4971 | } |