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 | ||
b411b363 PR |
51 | enum finish_epoch { |
52 | FE_STILL_LIVE, | |
53 | FE_DESTROYED, | |
54 | FE_RECYCLED, | |
55 | }; | |
56 | ||
57 | static int drbd_do_handshake(struct drbd_conf *mdev); | |
58 | static int drbd_do_auth(struct drbd_conf *mdev); | |
59 | ||
60 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event); | |
61 | static int e_end_block(struct drbd_conf *, struct drbd_work *, int); | |
62 | ||
b411b363 PR |
63 | |
64 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
65 | ||
45bb912b LE |
66 | /* |
67 | * some helper functions to deal with single linked page lists, | |
68 | * page->private being our "next" pointer. | |
69 | */ | |
70 | ||
71 | /* If at least n pages are linked at head, get n pages off. | |
72 | * Otherwise, don't modify head, and return NULL. | |
73 | * Locking is the responsibility of the caller. | |
74 | */ | |
75 | static struct page *page_chain_del(struct page **head, int n) | |
76 | { | |
77 | struct page *page; | |
78 | struct page *tmp; | |
79 | ||
80 | BUG_ON(!n); | |
81 | BUG_ON(!head); | |
82 | ||
83 | page = *head; | |
23ce4227 PR |
84 | |
85 | if (!page) | |
86 | return NULL; | |
87 | ||
45bb912b LE |
88 | while (page) { |
89 | tmp = page_chain_next(page); | |
90 | if (--n == 0) | |
91 | break; /* found sufficient pages */ | |
92 | if (tmp == NULL) | |
93 | /* insufficient pages, don't use any of them. */ | |
94 | return NULL; | |
95 | page = tmp; | |
96 | } | |
97 | ||
98 | /* add end of list marker for the returned list */ | |
99 | set_page_private(page, 0); | |
100 | /* actual return value, and adjustment of head */ | |
101 | page = *head; | |
102 | *head = tmp; | |
103 | return page; | |
104 | } | |
105 | ||
106 | /* may be used outside of locks to find the tail of a (usually short) | |
107 | * "private" page chain, before adding it back to a global chain head | |
108 | * with page_chain_add() under a spinlock. */ | |
109 | static struct page *page_chain_tail(struct page *page, int *len) | |
110 | { | |
111 | struct page *tmp; | |
112 | int i = 1; | |
113 | while ((tmp = page_chain_next(page))) | |
114 | ++i, page = tmp; | |
115 | if (len) | |
116 | *len = i; | |
117 | return page; | |
118 | } | |
119 | ||
120 | static int page_chain_free(struct page *page) | |
121 | { | |
122 | struct page *tmp; | |
123 | int i = 0; | |
124 | page_chain_for_each_safe(page, tmp) { | |
125 | put_page(page); | |
126 | ++i; | |
127 | } | |
128 | return i; | |
129 | } | |
130 | ||
131 | static void page_chain_add(struct page **head, | |
132 | struct page *chain_first, struct page *chain_last) | |
133 | { | |
134 | #if 1 | |
135 | struct page *tmp; | |
136 | tmp = page_chain_tail(chain_first, NULL); | |
137 | BUG_ON(tmp != chain_last); | |
138 | #endif | |
139 | ||
140 | /* add chain to head */ | |
141 | set_page_private(chain_last, (unsigned long)*head); | |
142 | *head = chain_first; | |
143 | } | |
144 | ||
145 | static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number) | |
b411b363 PR |
146 | { |
147 | struct page *page = NULL; | |
45bb912b LE |
148 | struct page *tmp = NULL; |
149 | int i = 0; | |
b411b363 PR |
150 | |
151 | /* Yes, testing drbd_pp_vacant outside the lock is racy. | |
152 | * So what. It saves a spin_lock. */ | |
45bb912b | 153 | if (drbd_pp_vacant >= number) { |
b411b363 | 154 | spin_lock(&drbd_pp_lock); |
45bb912b LE |
155 | page = page_chain_del(&drbd_pp_pool, number); |
156 | if (page) | |
157 | drbd_pp_vacant -= number; | |
b411b363 | 158 | spin_unlock(&drbd_pp_lock); |
45bb912b LE |
159 | if (page) |
160 | return page; | |
b411b363 | 161 | } |
45bb912b | 162 | |
b411b363 PR |
163 | /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD |
164 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
165 | * which in turn might block on the other node at this very place. */ | |
45bb912b LE |
166 | for (i = 0; i < number; i++) { |
167 | tmp = alloc_page(GFP_TRY); | |
168 | if (!tmp) | |
169 | break; | |
170 | set_page_private(tmp, (unsigned long)page); | |
171 | page = tmp; | |
172 | } | |
173 | ||
174 | if (i == number) | |
175 | return page; | |
176 | ||
177 | /* Not enough pages immediately available this time. | |
178 | * No need to jump around here, drbd_pp_alloc will retry this | |
179 | * function "soon". */ | |
180 | if (page) { | |
181 | tmp = page_chain_tail(page, NULL); | |
182 | spin_lock(&drbd_pp_lock); | |
183 | page_chain_add(&drbd_pp_pool, page, tmp); | |
184 | drbd_pp_vacant += i; | |
185 | spin_unlock(&drbd_pp_lock); | |
186 | } | |
187 | return NULL; | |
b411b363 PR |
188 | } |
189 | ||
b411b363 PR |
190 | static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed) |
191 | { | |
f6ffca9f | 192 | struct drbd_peer_request *e; |
b411b363 PR |
193 | struct list_head *le, *tle; |
194 | ||
195 | /* The EEs are always appended to the end of the list. Since | |
196 | they are sent in order over the wire, they have to finish | |
197 | in order. As soon as we see the first not finished we can | |
198 | stop to examine the list... */ | |
199 | ||
200 | list_for_each_safe(le, tle, &mdev->net_ee) { | |
f6ffca9f | 201 | e = list_entry(le, struct drbd_peer_request, w.list); |
45bb912b | 202 | if (drbd_ee_has_active_page(e)) |
b411b363 PR |
203 | break; |
204 | list_move(le, to_be_freed); | |
205 | } | |
206 | } | |
207 | ||
208 | static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) | |
209 | { | |
210 | LIST_HEAD(reclaimed); | |
f6ffca9f | 211 | struct drbd_peer_request *e, *t; |
b411b363 | 212 | |
87eeee41 | 213 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 214 | reclaim_net_ee(mdev, &reclaimed); |
87eeee41 | 215 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
216 | |
217 | list_for_each_entry_safe(e, t, &reclaimed, w.list) | |
435f0740 | 218 | drbd_free_net_ee(mdev, e); |
b411b363 PR |
219 | } |
220 | ||
221 | /** | |
45bb912b | 222 | * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled) |
b411b363 | 223 | * @mdev: DRBD device. |
45bb912b LE |
224 | * @number: number of pages requested |
225 | * @retry: whether to retry, if not enough pages are available right now | |
226 | * | |
227 | * Tries to allocate number pages, first from our own page pool, then from | |
228 | * the kernel, unless this allocation would exceed the max_buffers setting. | |
229 | * Possibly retry until DRBD frees sufficient pages somewhere else. | |
b411b363 | 230 | * |
45bb912b | 231 | * Returns a page chain linked via page->private. |
b411b363 | 232 | */ |
45bb912b | 233 | static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry) |
b411b363 PR |
234 | { |
235 | struct page *page = NULL; | |
236 | DEFINE_WAIT(wait); | |
237 | ||
45bb912b LE |
238 | /* Yes, we may run up to @number over max_buffers. If we |
239 | * follow it strictly, the admin will get it wrong anyways. */ | |
89e58e75 | 240 | if (atomic_read(&mdev->pp_in_use) < mdev->tconn->net_conf->max_buffers) |
45bb912b | 241 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 | 242 | |
45bb912b | 243 | while (page == NULL) { |
b411b363 PR |
244 | prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); |
245 | ||
246 | drbd_kick_lo_and_reclaim_net(mdev); | |
247 | ||
89e58e75 | 248 | if (atomic_read(&mdev->pp_in_use) < mdev->tconn->net_conf->max_buffers) { |
45bb912b | 249 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 PR |
250 | if (page) |
251 | break; | |
252 | } | |
253 | ||
254 | if (!retry) | |
255 | break; | |
256 | ||
257 | if (signal_pending(current)) { | |
258 | dev_warn(DEV, "drbd_pp_alloc interrupted!\n"); | |
259 | break; | |
260 | } | |
261 | ||
262 | schedule(); | |
263 | } | |
264 | finish_wait(&drbd_pp_wait, &wait); | |
265 | ||
45bb912b LE |
266 | if (page) |
267 | atomic_add(number, &mdev->pp_in_use); | |
b411b363 PR |
268 | return page; |
269 | } | |
270 | ||
271 | /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc. | |
87eeee41 | 272 | * Is also used from inside an other spin_lock_irq(&mdev->tconn->req_lock); |
45bb912b LE |
273 | * Either links the page chain back to the global pool, |
274 | * or returns all pages to the system. */ | |
435f0740 | 275 | static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net) |
b411b363 | 276 | { |
435f0740 | 277 | atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use; |
b411b363 | 278 | int i; |
435f0740 | 279 | |
1816a2b4 | 280 | if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE)*minor_count) |
45bb912b LE |
281 | i = page_chain_free(page); |
282 | else { | |
283 | struct page *tmp; | |
284 | tmp = page_chain_tail(page, &i); | |
285 | spin_lock(&drbd_pp_lock); | |
286 | page_chain_add(&drbd_pp_pool, page, tmp); | |
287 | drbd_pp_vacant += i; | |
288 | spin_unlock(&drbd_pp_lock); | |
b411b363 | 289 | } |
435f0740 | 290 | i = atomic_sub_return(i, a); |
45bb912b | 291 | if (i < 0) |
435f0740 LE |
292 | dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n", |
293 | is_net ? "pp_in_use_by_net" : "pp_in_use", i); | |
b411b363 PR |
294 | wake_up(&drbd_pp_wait); |
295 | } | |
296 | ||
297 | /* | |
298 | You need to hold the req_lock: | |
299 | _drbd_wait_ee_list_empty() | |
300 | ||
301 | You must not have the req_lock: | |
302 | drbd_free_ee() | |
303 | drbd_alloc_ee() | |
304 | drbd_init_ee() | |
305 | drbd_release_ee() | |
306 | drbd_ee_fix_bhs() | |
307 | drbd_process_done_ee() | |
308 | drbd_clear_done_ee() | |
309 | drbd_wait_ee_list_empty() | |
310 | */ | |
311 | ||
f6ffca9f AG |
312 | struct drbd_peer_request * |
313 | drbd_alloc_ee(struct drbd_conf *mdev, u64 id, sector_t sector, | |
314 | unsigned int data_size, gfp_t gfp_mask) __must_hold(local) | |
b411b363 | 315 | { |
f6ffca9f | 316 | struct drbd_peer_request *e; |
b411b363 | 317 | struct page *page; |
45bb912b | 318 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b411b363 | 319 | |
0cf9d27e | 320 | if (drbd_insert_fault(mdev, DRBD_FAULT_AL_EE)) |
b411b363 PR |
321 | return NULL; |
322 | ||
323 | e = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); | |
324 | if (!e) { | |
325 | if (!(gfp_mask & __GFP_NOWARN)) | |
326 | dev_err(DEV, "alloc_ee: Allocation of an EE failed\n"); | |
327 | return NULL; | |
328 | } | |
329 | ||
45bb912b LE |
330 | page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT)); |
331 | if (!page) | |
332 | goto fail; | |
b411b363 | 333 | |
8b946255 | 334 | drbd_clear_interval(&e->i); |
53840641 AG |
335 | e->i.size = data_size; |
336 | e->i.sector = sector; | |
5e472264 | 337 | e->i.local = false; |
53840641 AG |
338 | e->i.waiting = false; |
339 | ||
b411b363 | 340 | e->epoch = NULL; |
45bb912b LE |
341 | e->mdev = mdev; |
342 | e->pages = page; | |
343 | atomic_set(&e->pending_bios, 0); | |
b411b363 | 344 | e->flags = 0; |
9a8e7753 AG |
345 | /* |
346 | * The block_id is opaque to the receiver. It is not endianness | |
347 | * converted, and sent back to the sender unchanged. | |
348 | */ | |
45bb912b | 349 | e->block_id = id; |
b411b363 | 350 | |
b411b363 PR |
351 | return e; |
352 | ||
45bb912b | 353 | fail: |
b411b363 | 354 | mempool_free(e, drbd_ee_mempool); |
b411b363 PR |
355 | return NULL; |
356 | } | |
357 | ||
f6ffca9f AG |
358 | void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_peer_request *e, |
359 | int is_net) | |
b411b363 | 360 | { |
c36c3ced LE |
361 | if (e->flags & EE_HAS_DIGEST) |
362 | kfree(e->digest); | |
435f0740 | 363 | drbd_pp_free(mdev, e->pages, is_net); |
45bb912b | 364 | D_ASSERT(atomic_read(&e->pending_bios) == 0); |
8b946255 | 365 | D_ASSERT(drbd_interval_empty(&e->i)); |
b411b363 PR |
366 | mempool_free(e, drbd_ee_mempool); |
367 | } | |
368 | ||
369 | int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) | |
370 | { | |
371 | LIST_HEAD(work_list); | |
f6ffca9f | 372 | struct drbd_peer_request *e, *t; |
b411b363 | 373 | int count = 0; |
435f0740 | 374 | int is_net = list == &mdev->net_ee; |
b411b363 | 375 | |
87eeee41 | 376 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 377 | list_splice_init(list, &work_list); |
87eeee41 | 378 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
379 | |
380 | list_for_each_entry_safe(e, t, &work_list, w.list) { | |
435f0740 | 381 | drbd_free_some_ee(mdev, e, is_net); |
b411b363 PR |
382 | count++; |
383 | } | |
384 | return count; | |
385 | } | |
386 | ||
387 | ||
388 | /* | |
389 | * This function is called from _asender only_ | |
8554df1c | 390 | * but see also comments in _req_mod(,BARRIER_ACKED) |
b411b363 PR |
391 | * and receive_Barrier. |
392 | * | |
393 | * Move entries from net_ee to done_ee, if ready. | |
394 | * Grab done_ee, call all callbacks, free the entries. | |
395 | * The callbacks typically send out ACKs. | |
396 | */ | |
397 | static int drbd_process_done_ee(struct drbd_conf *mdev) | |
398 | { | |
399 | LIST_HEAD(work_list); | |
400 | LIST_HEAD(reclaimed); | |
f6ffca9f | 401 | struct drbd_peer_request *e, *t; |
b411b363 PR |
402 | int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS); |
403 | ||
87eeee41 | 404 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
405 | reclaim_net_ee(mdev, &reclaimed); |
406 | list_splice_init(&mdev->done_ee, &work_list); | |
87eeee41 | 407 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
408 | |
409 | list_for_each_entry_safe(e, t, &reclaimed, w.list) | |
435f0740 | 410 | drbd_free_net_ee(mdev, e); |
b411b363 PR |
411 | |
412 | /* possible callbacks here: | |
413 | * e_end_block, and e_end_resync_block, e_send_discard_ack. | |
414 | * all ignore the last argument. | |
415 | */ | |
416 | list_for_each_entry_safe(e, t, &work_list, w.list) { | |
b411b363 PR |
417 | /* list_del not necessary, next/prev members not touched */ |
418 | ok = e->w.cb(mdev, &e->w, !ok) && ok; | |
419 | drbd_free_ee(mdev, e); | |
420 | } | |
421 | wake_up(&mdev->ee_wait); | |
422 | ||
423 | return ok; | |
424 | } | |
425 | ||
426 | void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
427 | { | |
428 | DEFINE_WAIT(wait); | |
429 | ||
430 | /* avoids spin_lock/unlock | |
431 | * and calling prepare_to_wait in the fast path */ | |
432 | while (!list_empty(head)) { | |
433 | prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); | |
87eeee41 | 434 | spin_unlock_irq(&mdev->tconn->req_lock); |
7eaceacc | 435 | io_schedule(); |
b411b363 | 436 | finish_wait(&mdev->ee_wait, &wait); |
87eeee41 | 437 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
438 | } |
439 | } | |
440 | ||
441 | void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
442 | { | |
87eeee41 | 443 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 444 | _drbd_wait_ee_list_empty(mdev, head); |
87eeee41 | 445 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
446 | } |
447 | ||
448 | /* see also kernel_accept; which is only present since 2.6.18. | |
449 | * also we want to log which part of it failed, exactly */ | |
450 | static int drbd_accept(struct drbd_conf *mdev, const char **what, | |
451 | struct socket *sock, struct socket **newsock) | |
452 | { | |
453 | struct sock *sk = sock->sk; | |
454 | int err = 0; | |
455 | ||
456 | *what = "listen"; | |
457 | err = sock->ops->listen(sock, 5); | |
458 | if (err < 0) | |
459 | goto out; | |
460 | ||
461 | *what = "sock_create_lite"; | |
462 | err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, | |
463 | newsock); | |
464 | if (err < 0) | |
465 | goto out; | |
466 | ||
467 | *what = "accept"; | |
468 | err = sock->ops->accept(sock, *newsock, 0); | |
469 | if (err < 0) { | |
470 | sock_release(*newsock); | |
471 | *newsock = NULL; | |
472 | goto out; | |
473 | } | |
474 | (*newsock)->ops = sock->ops; | |
475 | ||
476 | out: | |
477 | return err; | |
478 | } | |
479 | ||
480 | static int drbd_recv_short(struct drbd_conf *mdev, struct socket *sock, | |
481 | void *buf, size_t size, int flags) | |
482 | { | |
483 | mm_segment_t oldfs; | |
484 | struct kvec iov = { | |
485 | .iov_base = buf, | |
486 | .iov_len = size, | |
487 | }; | |
488 | struct msghdr msg = { | |
489 | .msg_iovlen = 1, | |
490 | .msg_iov = (struct iovec *)&iov, | |
491 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) | |
492 | }; | |
493 | int rv; | |
494 | ||
495 | oldfs = get_fs(); | |
496 | set_fs(KERNEL_DS); | |
497 | rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); | |
498 | set_fs(oldfs); | |
499 | ||
500 | return rv; | |
501 | } | |
502 | ||
503 | static int drbd_recv(struct drbd_conf *mdev, void *buf, size_t size) | |
504 | { | |
505 | mm_segment_t oldfs; | |
506 | struct kvec iov = { | |
507 | .iov_base = buf, | |
508 | .iov_len = size, | |
509 | }; | |
510 | struct msghdr msg = { | |
511 | .msg_iovlen = 1, | |
512 | .msg_iov = (struct iovec *)&iov, | |
513 | .msg_flags = MSG_WAITALL | MSG_NOSIGNAL | |
514 | }; | |
515 | int rv; | |
516 | ||
517 | oldfs = get_fs(); | |
518 | set_fs(KERNEL_DS); | |
519 | ||
520 | for (;;) { | |
e42325a5 | 521 | rv = sock_recvmsg(mdev->tconn->data.socket, &msg, size, msg.msg_flags); |
b411b363 PR |
522 | if (rv == size) |
523 | break; | |
524 | ||
525 | /* Note: | |
526 | * ECONNRESET other side closed the connection | |
527 | * ERESTARTSYS (on sock) we got a signal | |
528 | */ | |
529 | ||
530 | if (rv < 0) { | |
531 | if (rv == -ECONNRESET) | |
532 | dev_info(DEV, "sock was reset by peer\n"); | |
533 | else if (rv != -ERESTARTSYS) | |
534 | dev_err(DEV, "sock_recvmsg returned %d\n", rv); | |
535 | break; | |
536 | } else if (rv == 0) { | |
537 | dev_info(DEV, "sock was shut down by peer\n"); | |
538 | break; | |
539 | } else { | |
540 | /* signal came in, or peer/link went down, | |
541 | * after we read a partial message | |
542 | */ | |
543 | /* D_ASSERT(signal_pending(current)); */ | |
544 | break; | |
545 | } | |
546 | }; | |
547 | ||
548 | set_fs(oldfs); | |
549 | ||
550 | if (rv != size) | |
551 | drbd_force_state(mdev, NS(conn, C_BROKEN_PIPE)); | |
552 | ||
553 | return rv; | |
554 | } | |
555 | ||
5dbf1673 LE |
556 | /* quoting tcp(7): |
557 | * On individual connections, the socket buffer size must be set prior to the | |
558 | * listen(2) or connect(2) calls in order to have it take effect. | |
559 | * This is our wrapper to do so. | |
560 | */ | |
561 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
562 | unsigned int rcv) | |
563 | { | |
564 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
565 | if (snd) { | |
566 | sock->sk->sk_sndbuf = snd; | |
567 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
568 | } | |
569 | if (rcv) { | |
570 | sock->sk->sk_rcvbuf = rcv; | |
571 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
572 | } | |
573 | } | |
574 | ||
b411b363 PR |
575 | static struct socket *drbd_try_connect(struct drbd_conf *mdev) |
576 | { | |
577 | const char *what; | |
578 | struct socket *sock; | |
579 | struct sockaddr_in6 src_in6; | |
580 | int err; | |
581 | int disconnect_on_error = 1; | |
582 | ||
b2fb6dbe | 583 | if (!get_net_conf(mdev->tconn)) |
b411b363 PR |
584 | return NULL; |
585 | ||
586 | what = "sock_create_kern"; | |
89e58e75 | 587 | err = sock_create_kern(((struct sockaddr *)mdev->tconn->net_conf->my_addr)->sa_family, |
b411b363 PR |
588 | SOCK_STREAM, IPPROTO_TCP, &sock); |
589 | if (err < 0) { | |
590 | sock = NULL; | |
591 | goto out; | |
592 | } | |
593 | ||
594 | sock->sk->sk_rcvtimeo = | |
89e58e75 PR |
595 | sock->sk->sk_sndtimeo = mdev->tconn->net_conf->try_connect_int*HZ; |
596 | drbd_setbufsize(sock, mdev->tconn->net_conf->sndbuf_size, | |
597 | mdev->tconn->net_conf->rcvbuf_size); | |
b411b363 PR |
598 | |
599 | /* explicitly bind to the configured IP as source IP | |
600 | * for the outgoing connections. | |
601 | * This is needed for multihomed hosts and to be | |
602 | * able to use lo: interfaces for drbd. | |
603 | * Make sure to use 0 as port number, so linux selects | |
604 | * a free one dynamically. | |
605 | */ | |
89e58e75 PR |
606 | memcpy(&src_in6, mdev->tconn->net_conf->my_addr, |
607 | min_t(int, mdev->tconn->net_conf->my_addr_len, sizeof(src_in6))); | |
608 | if (((struct sockaddr *)mdev->tconn->net_conf->my_addr)->sa_family == AF_INET6) | |
b411b363 PR |
609 | src_in6.sin6_port = 0; |
610 | else | |
611 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
612 | ||
613 | what = "bind before connect"; | |
614 | err = sock->ops->bind(sock, | |
615 | (struct sockaddr *) &src_in6, | |
89e58e75 | 616 | mdev->tconn->net_conf->my_addr_len); |
b411b363 PR |
617 | if (err < 0) |
618 | goto out; | |
619 | ||
620 | /* connect may fail, peer not yet available. | |
621 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
622 | disconnect_on_error = 0; | |
623 | what = "connect"; | |
624 | err = sock->ops->connect(sock, | |
89e58e75 PR |
625 | (struct sockaddr *)mdev->tconn->net_conf->peer_addr, |
626 | mdev->tconn->net_conf->peer_addr_len, 0); | |
b411b363 PR |
627 | |
628 | out: | |
629 | if (err < 0) { | |
630 | if (sock) { | |
631 | sock_release(sock); | |
632 | sock = NULL; | |
633 | } | |
634 | switch (-err) { | |
635 | /* timeout, busy, signal pending */ | |
636 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
637 | case EINTR: case ERESTARTSYS: | |
638 | /* peer not (yet) available, network problem */ | |
639 | case ECONNREFUSED: case ENETUNREACH: | |
640 | case EHOSTDOWN: case EHOSTUNREACH: | |
641 | disconnect_on_error = 0; | |
642 | break; | |
643 | default: | |
644 | dev_err(DEV, "%s failed, err = %d\n", what, err); | |
645 | } | |
646 | if (disconnect_on_error) | |
647 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
648 | } | |
b2fb6dbe | 649 | put_net_conf(mdev->tconn); |
b411b363 PR |
650 | return sock; |
651 | } | |
652 | ||
653 | static struct socket *drbd_wait_for_connect(struct drbd_conf *mdev) | |
654 | { | |
655 | int timeo, err; | |
656 | struct socket *s_estab = NULL, *s_listen; | |
657 | const char *what; | |
658 | ||
b2fb6dbe | 659 | if (!get_net_conf(mdev->tconn)) |
b411b363 PR |
660 | return NULL; |
661 | ||
662 | what = "sock_create_kern"; | |
89e58e75 | 663 | err = sock_create_kern(((struct sockaddr *)mdev->tconn->net_conf->my_addr)->sa_family, |
b411b363 PR |
664 | SOCK_STREAM, IPPROTO_TCP, &s_listen); |
665 | if (err) { | |
666 | s_listen = NULL; | |
667 | goto out; | |
668 | } | |
669 | ||
89e58e75 | 670 | timeo = mdev->tconn->net_conf->try_connect_int * HZ; |
b411b363 PR |
671 | timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ |
672 | ||
673 | s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
674 | s_listen->sk->sk_rcvtimeo = timeo; | |
675 | s_listen->sk->sk_sndtimeo = timeo; | |
89e58e75 PR |
676 | drbd_setbufsize(s_listen, mdev->tconn->net_conf->sndbuf_size, |
677 | mdev->tconn->net_conf->rcvbuf_size); | |
b411b363 PR |
678 | |
679 | what = "bind before listen"; | |
680 | err = s_listen->ops->bind(s_listen, | |
89e58e75 PR |
681 | (struct sockaddr *) mdev->tconn->net_conf->my_addr, |
682 | mdev->tconn->net_conf->my_addr_len); | |
b411b363 PR |
683 | if (err < 0) |
684 | goto out; | |
685 | ||
686 | err = drbd_accept(mdev, &what, s_listen, &s_estab); | |
687 | ||
688 | out: | |
689 | if (s_listen) | |
690 | sock_release(s_listen); | |
691 | if (err < 0) { | |
692 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
693 | dev_err(DEV, "%s failed, err = %d\n", what, err); | |
694 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
695 | } | |
696 | } | |
b2fb6dbe | 697 | put_net_conf(mdev->tconn); |
b411b363 PR |
698 | |
699 | return s_estab; | |
700 | } | |
701 | ||
d8763023 AG |
702 | static int drbd_send_fp(struct drbd_conf *mdev, struct socket *sock, |
703 | enum drbd_packet cmd) | |
b411b363 | 704 | { |
c012949a | 705 | struct p_header *h = &mdev->tconn->data.sbuf.header; |
b411b363 PR |
706 | |
707 | return _drbd_send_cmd(mdev, sock, cmd, h, sizeof(*h), 0); | |
708 | } | |
709 | ||
d8763023 AG |
710 | static enum drbd_packet drbd_recv_fp(struct drbd_conf *mdev, |
711 | struct socket *sock) | |
b411b363 | 712 | { |
e42325a5 | 713 | struct p_header80 *h = &mdev->tconn->data.rbuf.header.h80; |
b411b363 PR |
714 | int rr; |
715 | ||
716 | rr = drbd_recv_short(mdev, sock, h, sizeof(*h), 0); | |
717 | ||
ca9bc12b | 718 | if (rr == sizeof(*h) && h->magic == cpu_to_be32(DRBD_MAGIC)) |
b411b363 PR |
719 | return be16_to_cpu(h->command); |
720 | ||
721 | return 0xffff; | |
722 | } | |
723 | ||
724 | /** | |
725 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
726 | * @mdev: DRBD device. | |
727 | * @sock: pointer to the pointer to the socket. | |
728 | */ | |
729 | static int drbd_socket_okay(struct drbd_conf *mdev, struct socket **sock) | |
730 | { | |
731 | int rr; | |
732 | char tb[4]; | |
733 | ||
734 | if (!*sock) | |
81e84650 | 735 | return false; |
b411b363 PR |
736 | |
737 | rr = drbd_recv_short(mdev, *sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); | |
738 | ||
739 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 740 | return true; |
b411b363 PR |
741 | } else { |
742 | sock_release(*sock); | |
743 | *sock = NULL; | |
81e84650 | 744 | return false; |
b411b363 PR |
745 | } |
746 | } | |
747 | ||
748 | /* | |
749 | * return values: | |
750 | * 1 yes, we have a valid connection | |
751 | * 0 oops, did not work out, please try again | |
752 | * -1 peer talks different language, | |
753 | * no point in trying again, please go standalone. | |
754 | * -2 We do not have a network config... | |
755 | */ | |
756 | static int drbd_connect(struct drbd_conf *mdev) | |
757 | { | |
758 | struct socket *s, *sock, *msock; | |
759 | int try, h, ok; | |
760 | ||
e42325a5 | 761 | D_ASSERT(!mdev->tconn->data.socket); |
b411b363 | 762 | |
b411b363 PR |
763 | if (drbd_request_state(mdev, NS(conn, C_WF_CONNECTION)) < SS_SUCCESS) |
764 | return -2; | |
765 | ||
766 | clear_bit(DISCARD_CONCURRENT, &mdev->flags); | |
fd340c12 PR |
767 | mdev->tconn->agreed_pro_version = 99; |
768 | /* agreed_pro_version must be smaller than 100 so we send the old | |
769 | header (h80) in the first packet and in the handshake packet. */ | |
b411b363 PR |
770 | |
771 | sock = NULL; | |
772 | msock = NULL; | |
773 | ||
774 | do { | |
775 | for (try = 0;;) { | |
776 | /* 3 tries, this should take less than a second! */ | |
777 | s = drbd_try_connect(mdev); | |
778 | if (s || ++try >= 3) | |
779 | break; | |
780 | /* give the other side time to call bind() & listen() */ | |
20ee6390 | 781 | schedule_timeout_interruptible(HZ / 10); |
b411b363 PR |
782 | } |
783 | ||
784 | if (s) { | |
785 | if (!sock) { | |
786 | drbd_send_fp(mdev, s, P_HAND_SHAKE_S); | |
787 | sock = s; | |
788 | s = NULL; | |
789 | } else if (!msock) { | |
790 | drbd_send_fp(mdev, s, P_HAND_SHAKE_M); | |
791 | msock = s; | |
792 | s = NULL; | |
793 | } else { | |
794 | dev_err(DEV, "Logic error in drbd_connect()\n"); | |
795 | goto out_release_sockets; | |
796 | } | |
797 | } | |
798 | ||
799 | if (sock && msock) { | |
89e58e75 | 800 | schedule_timeout_interruptible(mdev->tconn->net_conf->ping_timeo*HZ/10); |
b411b363 PR |
801 | ok = drbd_socket_okay(mdev, &sock); |
802 | ok = drbd_socket_okay(mdev, &msock) && ok; | |
803 | if (ok) | |
804 | break; | |
805 | } | |
806 | ||
807 | retry: | |
808 | s = drbd_wait_for_connect(mdev); | |
809 | if (s) { | |
810 | try = drbd_recv_fp(mdev, s); | |
811 | drbd_socket_okay(mdev, &sock); | |
812 | drbd_socket_okay(mdev, &msock); | |
813 | switch (try) { | |
814 | case P_HAND_SHAKE_S: | |
815 | if (sock) { | |
816 | dev_warn(DEV, "initial packet S crossed\n"); | |
817 | sock_release(sock); | |
818 | } | |
819 | sock = s; | |
820 | break; | |
821 | case P_HAND_SHAKE_M: | |
822 | if (msock) { | |
823 | dev_warn(DEV, "initial packet M crossed\n"); | |
824 | sock_release(msock); | |
825 | } | |
826 | msock = s; | |
827 | set_bit(DISCARD_CONCURRENT, &mdev->flags); | |
828 | break; | |
829 | default: | |
830 | dev_warn(DEV, "Error receiving initial packet\n"); | |
831 | sock_release(s); | |
832 | if (random32() & 1) | |
833 | goto retry; | |
834 | } | |
835 | } | |
836 | ||
837 | if (mdev->state.conn <= C_DISCONNECTING) | |
838 | goto out_release_sockets; | |
839 | if (signal_pending(current)) { | |
840 | flush_signals(current); | |
841 | smp_rmb(); | |
e6b3ea83 | 842 | if (get_t_state(&mdev->tconn->receiver) == EXITING) |
b411b363 PR |
843 | goto out_release_sockets; |
844 | } | |
845 | ||
846 | if (sock && msock) { | |
847 | ok = drbd_socket_okay(mdev, &sock); | |
848 | ok = drbd_socket_okay(mdev, &msock) && ok; | |
849 | if (ok) | |
850 | break; | |
851 | } | |
852 | } while (1); | |
853 | ||
854 | msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
855 | sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
856 | ||
857 | sock->sk->sk_allocation = GFP_NOIO; | |
858 | msock->sk->sk_allocation = GFP_NOIO; | |
859 | ||
860 | sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; | |
861 | msock->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
862 | ||
b411b363 | 863 | /* NOT YET ... |
89e58e75 | 864 | * sock->sk->sk_sndtimeo = mdev->tconn->net_conf->timeout*HZ/10; |
b411b363 PR |
865 | * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
866 | * first set it to the P_HAND_SHAKE timeout, | |
867 | * which we set to 4x the configured ping_timeout. */ | |
868 | sock->sk->sk_sndtimeo = | |
89e58e75 | 869 | sock->sk->sk_rcvtimeo = mdev->tconn->net_conf->ping_timeo*4*HZ/10; |
b411b363 | 870 | |
89e58e75 PR |
871 | msock->sk->sk_sndtimeo = mdev->tconn->net_conf->timeout*HZ/10; |
872 | msock->sk->sk_rcvtimeo = mdev->tconn->net_conf->ping_int*HZ; | |
b411b363 PR |
873 | |
874 | /* we don't want delays. | |
25985edc | 875 | * we use TCP_CORK where appropriate, though */ |
b411b363 PR |
876 | drbd_tcp_nodelay(sock); |
877 | drbd_tcp_nodelay(msock); | |
878 | ||
e42325a5 PR |
879 | mdev->tconn->data.socket = sock; |
880 | mdev->tconn->meta.socket = msock; | |
31890f4a | 881 | mdev->tconn->last_received = jiffies; |
b411b363 | 882 | |
e6b3ea83 | 883 | D_ASSERT(mdev->tconn->asender.task == NULL); |
b411b363 PR |
884 | |
885 | h = drbd_do_handshake(mdev); | |
886 | if (h <= 0) | |
887 | return h; | |
888 | ||
a0638456 | 889 | if (mdev->tconn->cram_hmac_tfm) { |
b411b363 | 890 | /* drbd_request_state(mdev, NS(conn, WFAuth)); */ |
b10d96cb JT |
891 | switch (drbd_do_auth(mdev)) { |
892 | case -1: | |
b411b363 PR |
893 | dev_err(DEV, "Authentication of peer failed\n"); |
894 | return -1; | |
b10d96cb JT |
895 | case 0: |
896 | dev_err(DEV, "Authentication of peer failed, trying again.\n"); | |
897 | return 0; | |
b411b363 PR |
898 | } |
899 | } | |
900 | ||
901 | if (drbd_request_state(mdev, NS(conn, C_WF_REPORT_PARAMS)) < SS_SUCCESS) | |
902 | return 0; | |
903 | ||
89e58e75 | 904 | sock->sk->sk_sndtimeo = mdev->tconn->net_conf->timeout*HZ/10; |
b411b363 PR |
905 | sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
906 | ||
907 | atomic_set(&mdev->packet_seq, 0); | |
908 | mdev->peer_seq = 0; | |
909 | ||
e6b3ea83 | 910 | drbd_thread_start(&mdev->tconn->asender); |
b411b363 | 911 | |
148efa16 | 912 | if (drbd_send_protocol(mdev) == -1) |
7e2455c1 | 913 | return -1; |
b411b363 | 914 | drbd_send_sync_param(mdev, &mdev->sync_conf); |
e89b591c | 915 | drbd_send_sizes(mdev, 0, 0); |
b411b363 PR |
916 | drbd_send_uuids(mdev); |
917 | drbd_send_state(mdev); | |
918 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); | |
919 | clear_bit(RESIZE_PENDING, &mdev->flags); | |
7fde2be9 | 920 | mod_timer(&mdev->request_timer, jiffies + HZ); /* just start it here. */ |
b411b363 PR |
921 | |
922 | return 1; | |
923 | ||
924 | out_release_sockets: | |
925 | if (sock) | |
926 | sock_release(sock); | |
927 | if (msock) | |
928 | sock_release(msock); | |
929 | return -1; | |
930 | } | |
931 | ||
d8763023 AG |
932 | static bool decode_header(struct drbd_conf *mdev, struct p_header *h, |
933 | enum drbd_packet *cmd, unsigned int *packet_size) | |
b411b363 | 934 | { |
fd340c12 | 935 | if (h->h80.magic == cpu_to_be32(DRBD_MAGIC)) { |
02918be2 PR |
936 | *cmd = be16_to_cpu(h->h80.command); |
937 | *packet_size = be16_to_cpu(h->h80.length); | |
ca9bc12b | 938 | } else if (h->h95.magic == cpu_to_be16(DRBD_MAGIC_BIG)) { |
02918be2 | 939 | *cmd = be16_to_cpu(h->h95.command); |
fd340c12 | 940 | *packet_size = be32_to_cpu(h->h95.length) & 0x00ffffff; |
02918be2 | 941 | } else { |
004352fa LE |
942 | dev_err(DEV, "magic?? on data m: 0x%08x c: %d l: %d\n", |
943 | be32_to_cpu(h->h80.magic), | |
944 | be16_to_cpu(h->h80.command), | |
945 | be16_to_cpu(h->h80.length)); | |
81e84650 | 946 | return false; |
b411b363 | 947 | } |
257d0af6 PR |
948 | return true; |
949 | } | |
950 | ||
d8763023 AG |
951 | static int drbd_recv_header(struct drbd_conf *mdev, enum drbd_packet *cmd, |
952 | unsigned int *packet_size) | |
257d0af6 PR |
953 | { |
954 | struct p_header *h = &mdev->tconn->data.rbuf.header; | |
955 | int r; | |
956 | ||
957 | r = drbd_recv(mdev, h, sizeof(*h)); | |
958 | if (unlikely(r != sizeof(*h))) { | |
959 | if (!signal_pending(current)) | |
960 | dev_warn(DEV, "short read expecting header on sock: r=%d\n", r); | |
961 | return false; | |
962 | } | |
963 | ||
964 | r = decode_header(mdev, h, cmd, packet_size); | |
31890f4a | 965 | mdev->tconn->last_received = jiffies; |
b411b363 | 966 | |
257d0af6 | 967 | return r; |
b411b363 PR |
968 | } |
969 | ||
2451fc3b | 970 | static void drbd_flush(struct drbd_conf *mdev) |
b411b363 PR |
971 | { |
972 | int rv; | |
973 | ||
974 | if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { | |
fbd9b09a | 975 | rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL, |
dd3932ed | 976 | NULL); |
b411b363 PR |
977 | if (rv) { |
978 | dev_err(DEV, "local disk flush failed with status %d\n", rv); | |
979 | /* would rather check on EOPNOTSUPP, but that is not reliable. | |
980 | * don't try again for ANY return value != 0 | |
981 | * if (rv == -EOPNOTSUPP) */ | |
982 | drbd_bump_write_ordering(mdev, WO_drain_io); | |
983 | } | |
984 | put_ldev(mdev); | |
985 | } | |
b411b363 PR |
986 | } |
987 | ||
988 | /** | |
989 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
990 | * @mdev: DRBD device. | |
991 | * @epoch: Epoch object. | |
992 | * @ev: Epoch event. | |
993 | */ | |
994 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, | |
995 | struct drbd_epoch *epoch, | |
996 | enum epoch_event ev) | |
997 | { | |
2451fc3b | 998 | int epoch_size; |
b411b363 | 999 | struct drbd_epoch *next_epoch; |
b411b363 PR |
1000 | enum finish_epoch rv = FE_STILL_LIVE; |
1001 | ||
1002 | spin_lock(&mdev->epoch_lock); | |
1003 | do { | |
1004 | next_epoch = NULL; | |
b411b363 PR |
1005 | |
1006 | epoch_size = atomic_read(&epoch->epoch_size); | |
1007 | ||
1008 | switch (ev & ~EV_CLEANUP) { | |
1009 | case EV_PUT: | |
1010 | atomic_dec(&epoch->active); | |
1011 | break; | |
1012 | case EV_GOT_BARRIER_NR: | |
1013 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1014 | break; |
1015 | case EV_BECAME_LAST: | |
1016 | /* nothing to do*/ | |
1017 | break; | |
1018 | } | |
1019 | ||
b411b363 PR |
1020 | if (epoch_size != 0 && |
1021 | atomic_read(&epoch->active) == 0 && | |
2451fc3b | 1022 | test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) { |
b411b363 PR |
1023 | if (!(ev & EV_CLEANUP)) { |
1024 | spin_unlock(&mdev->epoch_lock); | |
1025 | drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); | |
1026 | spin_lock(&mdev->epoch_lock); | |
1027 | } | |
1028 | dec_unacked(mdev); | |
1029 | ||
1030 | if (mdev->current_epoch != epoch) { | |
1031 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); | |
1032 | list_del(&epoch->list); | |
1033 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
1034 | mdev->epochs--; | |
b411b363 PR |
1035 | kfree(epoch); |
1036 | ||
1037 | if (rv == FE_STILL_LIVE) | |
1038 | rv = FE_DESTROYED; | |
1039 | } else { | |
1040 | epoch->flags = 0; | |
1041 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1042 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1043 | if (rv == FE_STILL_LIVE) |
1044 | rv = FE_RECYCLED; | |
2451fc3b | 1045 | wake_up(&mdev->ee_wait); |
b411b363 PR |
1046 | } |
1047 | } | |
1048 | ||
1049 | if (!next_epoch) | |
1050 | break; | |
1051 | ||
1052 | epoch = next_epoch; | |
1053 | } while (1); | |
1054 | ||
1055 | spin_unlock(&mdev->epoch_lock); | |
1056 | ||
b411b363 PR |
1057 | return rv; |
1058 | } | |
1059 | ||
1060 | /** | |
1061 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
1062 | * @mdev: DRBD device. | |
1063 | * @wo: Write ordering method to try. | |
1064 | */ | |
1065 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) | |
1066 | { | |
1067 | enum write_ordering_e pwo; | |
1068 | static char *write_ordering_str[] = { | |
1069 | [WO_none] = "none", | |
1070 | [WO_drain_io] = "drain", | |
1071 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1072 | }; |
1073 | ||
1074 | pwo = mdev->write_ordering; | |
1075 | wo = min(pwo, wo); | |
b411b363 PR |
1076 | if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) |
1077 | wo = WO_drain_io; | |
1078 | if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) | |
1079 | wo = WO_none; | |
1080 | mdev->write_ordering = wo; | |
2451fc3b | 1081 | if (pwo != mdev->write_ordering || wo == WO_bdev_flush) |
b411b363 PR |
1082 | dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); |
1083 | } | |
1084 | ||
45bb912b LE |
1085 | /** |
1086 | * drbd_submit_ee() | |
1087 | * @mdev: DRBD device. | |
1088 | * @e: epoch entry | |
1089 | * @rw: flag field, see bio->bi_rw | |
10f6d992 LE |
1090 | * |
1091 | * May spread the pages to multiple bios, | |
1092 | * depending on bio_add_page restrictions. | |
1093 | * | |
1094 | * Returns 0 if all bios have been submitted, | |
1095 | * -ENOMEM if we could not allocate enough bios, | |
1096 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1097 | * single page to an empty bio (which should never happen and likely indicates | |
1098 | * that the lower level IO stack is in some way broken). This has been observed | |
1099 | * on certain Xen deployments. | |
45bb912b LE |
1100 | */ |
1101 | /* TODO allocate from our own bio_set. */ | |
f6ffca9f AG |
1102 | int drbd_submit_ee(struct drbd_conf *mdev, struct drbd_peer_request *e, |
1103 | const unsigned rw, const int fault_type) | |
45bb912b LE |
1104 | { |
1105 | struct bio *bios = NULL; | |
1106 | struct bio *bio; | |
1107 | struct page *page = e->pages; | |
010f6e67 AG |
1108 | sector_t sector = e->i.sector; |
1109 | unsigned ds = e->i.size; | |
45bb912b LE |
1110 | unsigned n_bios = 0; |
1111 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
10f6d992 | 1112 | int err = -ENOMEM; |
45bb912b LE |
1113 | |
1114 | /* In most cases, we will only need one bio. But in case the lower | |
1115 | * level restrictions happen to be different at this offset on this | |
1116 | * side than those of the sending peer, we may need to submit the | |
1117 | * request in more than one bio. */ | |
1118 | next_bio: | |
1119 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1120 | if (!bio) { | |
1121 | dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); | |
1122 | goto fail; | |
1123 | } | |
010f6e67 | 1124 | /* > e->i.sector, unless this is the first bio */ |
45bb912b LE |
1125 | bio->bi_sector = sector; |
1126 | bio->bi_bdev = mdev->ldev->backing_bdev; | |
45bb912b LE |
1127 | bio->bi_rw = rw; |
1128 | bio->bi_private = e; | |
1129 | bio->bi_end_io = drbd_endio_sec; | |
1130 | ||
1131 | bio->bi_next = bios; | |
1132 | bios = bio; | |
1133 | ++n_bios; | |
1134 | ||
1135 | page_chain_for_each(page) { | |
1136 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1137 | if (!bio_add_page(bio, page, len, 0)) { | |
10f6d992 LE |
1138 | /* A single page must always be possible! |
1139 | * But in case it fails anyways, | |
1140 | * we deal with it, and complain (below). */ | |
1141 | if (bio->bi_vcnt == 0) { | |
1142 | dev_err(DEV, | |
1143 | "bio_add_page failed for len=%u, " | |
1144 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
1145 | len, (unsigned long long)bio->bi_sector); | |
1146 | err = -ENOSPC; | |
1147 | goto fail; | |
1148 | } | |
45bb912b LE |
1149 | goto next_bio; |
1150 | } | |
1151 | ds -= len; | |
1152 | sector += len >> 9; | |
1153 | --nr_pages; | |
1154 | } | |
1155 | D_ASSERT(page == NULL); | |
1156 | D_ASSERT(ds == 0); | |
1157 | ||
1158 | atomic_set(&e->pending_bios, n_bios); | |
1159 | do { | |
1160 | bio = bios; | |
1161 | bios = bios->bi_next; | |
1162 | bio->bi_next = NULL; | |
1163 | ||
45bb912b | 1164 | drbd_generic_make_request(mdev, fault_type, bio); |
45bb912b | 1165 | } while (bios); |
45bb912b LE |
1166 | return 0; |
1167 | ||
1168 | fail: | |
1169 | while (bios) { | |
1170 | bio = bios; | |
1171 | bios = bios->bi_next; | |
1172 | bio_put(bio); | |
1173 | } | |
10f6d992 | 1174 | return err; |
45bb912b LE |
1175 | } |
1176 | ||
53840641 | 1177 | static void drbd_remove_epoch_entry_interval(struct drbd_conf *mdev, |
f6ffca9f | 1178 | struct drbd_peer_request *e) |
53840641 AG |
1179 | { |
1180 | struct drbd_interval *i = &e->i; | |
1181 | ||
1182 | drbd_remove_interval(&mdev->write_requests, i); | |
1183 | drbd_clear_interval(i); | |
1184 | ||
1185 | /* Wake up any processes waiting for this epoch entry to complete. */ | |
1186 | if (i->waiting) | |
1187 | wake_up(&mdev->misc_wait); | |
1188 | } | |
1189 | ||
d8763023 AG |
1190 | static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packet cmd, |
1191 | unsigned int data_size) | |
b411b363 | 1192 | { |
2451fc3b | 1193 | int rv; |
e42325a5 | 1194 | struct p_barrier *p = &mdev->tconn->data.rbuf.barrier; |
b411b363 PR |
1195 | struct drbd_epoch *epoch; |
1196 | ||
b411b363 PR |
1197 | inc_unacked(mdev); |
1198 | ||
b411b363 PR |
1199 | mdev->current_epoch->barrier_nr = p->barrier; |
1200 | rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); | |
1201 | ||
1202 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1203 | * the activity log, which means it would not be resynced in case the | |
1204 | * R_PRIMARY crashes now. | |
1205 | * Therefore we must send the barrier_ack after the barrier request was | |
1206 | * completed. */ | |
1207 | switch (mdev->write_ordering) { | |
b411b363 PR |
1208 | case WO_none: |
1209 | if (rv == FE_RECYCLED) | |
81e84650 | 1210 | return true; |
2451fc3b PR |
1211 | |
1212 | /* receiver context, in the writeout path of the other node. | |
1213 | * avoid potential distributed deadlock */ | |
1214 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1215 | if (epoch) | |
1216 | break; | |
1217 | else | |
1218 | dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); | |
1219 | /* Fall through */ | |
b411b363 PR |
1220 | |
1221 | case WO_bdev_flush: | |
1222 | case WO_drain_io: | |
b411b363 | 1223 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
2451fc3b PR |
1224 | drbd_flush(mdev); |
1225 | ||
1226 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1227 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1228 | if (epoch) | |
1229 | break; | |
b411b363 PR |
1230 | } |
1231 | ||
2451fc3b PR |
1232 | epoch = mdev->current_epoch; |
1233 | wait_event(mdev->ee_wait, atomic_read(&epoch->epoch_size) == 0); | |
1234 | ||
1235 | D_ASSERT(atomic_read(&epoch->active) == 0); | |
1236 | D_ASSERT(epoch->flags == 0); | |
b411b363 | 1237 | |
81e84650 | 1238 | return true; |
2451fc3b PR |
1239 | default: |
1240 | dev_err(DEV, "Strangeness in mdev->write_ordering %d\n", mdev->write_ordering); | |
81e84650 | 1241 | return false; |
b411b363 PR |
1242 | } |
1243 | ||
1244 | epoch->flags = 0; | |
1245 | atomic_set(&epoch->epoch_size, 0); | |
1246 | atomic_set(&epoch->active, 0); | |
1247 | ||
1248 | spin_lock(&mdev->epoch_lock); | |
1249 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1250 | list_add(&epoch->list, &mdev->current_epoch->list); | |
1251 | mdev->current_epoch = epoch; | |
1252 | mdev->epochs++; | |
b411b363 PR |
1253 | } else { |
1254 | /* The current_epoch got recycled while we allocated this one... */ | |
1255 | kfree(epoch); | |
1256 | } | |
1257 | spin_unlock(&mdev->epoch_lock); | |
1258 | ||
81e84650 | 1259 | return true; |
b411b363 PR |
1260 | } |
1261 | ||
1262 | /* used from receive_RSDataReply (recv_resync_read) | |
1263 | * and from receive_Data */ | |
f6ffca9f AG |
1264 | static struct drbd_peer_request * |
1265 | read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, | |
1266 | int data_size) __must_hold(local) | |
b411b363 | 1267 | { |
6666032a | 1268 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); |
f6ffca9f | 1269 | struct drbd_peer_request *e; |
b411b363 | 1270 | struct page *page; |
45bb912b | 1271 | int dgs, ds, rr; |
a0638456 PR |
1272 | void *dig_in = mdev->tconn->int_dig_in; |
1273 | void *dig_vv = mdev->tconn->int_dig_vv; | |
6b4388ac | 1274 | unsigned long *data; |
b411b363 | 1275 | |
a0638456 PR |
1276 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ? |
1277 | crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0; | |
b411b363 PR |
1278 | |
1279 | if (dgs) { | |
1280 | rr = drbd_recv(mdev, dig_in, dgs); | |
1281 | if (rr != dgs) { | |
0ddc5549 LE |
1282 | if (!signal_pending(current)) |
1283 | dev_warn(DEV, | |
1284 | "short read receiving data digest: read %d expected %d\n", | |
1285 | rr, dgs); | |
b411b363 PR |
1286 | return NULL; |
1287 | } | |
1288 | } | |
1289 | ||
1290 | data_size -= dgs; | |
1291 | ||
841ce241 AG |
1292 | if (!expect(data_size != 0)) |
1293 | return NULL; | |
1294 | if (!expect(IS_ALIGNED(data_size, 512))) | |
1295 | return NULL; | |
1296 | if (!expect(data_size <= DRBD_MAX_BIO_SIZE)) | |
1297 | return NULL; | |
b411b363 | 1298 | |
6666032a LE |
1299 | /* even though we trust out peer, |
1300 | * we sometimes have to double check. */ | |
1301 | if (sector + (data_size>>9) > capacity) { | |
fdda6544 LE |
1302 | dev_err(DEV, "request from peer beyond end of local disk: " |
1303 | "capacity: %llus < sector: %llus + size: %u\n", | |
6666032a LE |
1304 | (unsigned long long)capacity, |
1305 | (unsigned long long)sector, data_size); | |
1306 | return NULL; | |
1307 | } | |
1308 | ||
b411b363 PR |
1309 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1310 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1311 | * which in turn might block on the other node at this very place. */ | |
1312 | e = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); | |
1313 | if (!e) | |
1314 | return NULL; | |
45bb912b | 1315 | |
b411b363 | 1316 | ds = data_size; |
45bb912b LE |
1317 | page = e->pages; |
1318 | page_chain_for_each(page) { | |
1319 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1320 | data = kmap(page); |
45bb912b | 1321 | rr = drbd_recv(mdev, data, len); |
0cf9d27e | 1322 | if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) { |
6b4388ac PR |
1323 | dev_err(DEV, "Fault injection: Corrupting data on receive\n"); |
1324 | data[0] = data[0] ^ (unsigned long)-1; | |
1325 | } | |
b411b363 | 1326 | kunmap(page); |
45bb912b | 1327 | if (rr != len) { |
b411b363 | 1328 | drbd_free_ee(mdev, e); |
0ddc5549 LE |
1329 | if (!signal_pending(current)) |
1330 | dev_warn(DEV, "short read receiving data: read %d expected %d\n", | |
1331 | rr, len); | |
b411b363 PR |
1332 | return NULL; |
1333 | } | |
1334 | ds -= rr; | |
1335 | } | |
1336 | ||
1337 | if (dgs) { | |
a0638456 | 1338 | drbd_csum_ee(mdev, mdev->tconn->integrity_r_tfm, e, dig_vv); |
b411b363 | 1339 | if (memcmp(dig_in, dig_vv, dgs)) { |
470be44a LE |
1340 | dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n", |
1341 | (unsigned long long)sector, data_size); | |
b411b363 PR |
1342 | drbd_bcast_ee(mdev, "digest failed", |
1343 | dgs, dig_in, dig_vv, e); | |
1344 | drbd_free_ee(mdev, e); | |
1345 | return NULL; | |
1346 | } | |
1347 | } | |
1348 | mdev->recv_cnt += data_size>>9; | |
1349 | return e; | |
1350 | } | |
1351 | ||
1352 | /* drbd_drain_block() just takes a data block | |
1353 | * out of the socket input buffer, and discards it. | |
1354 | */ | |
1355 | static int drbd_drain_block(struct drbd_conf *mdev, int data_size) | |
1356 | { | |
1357 | struct page *page; | |
1358 | int rr, rv = 1; | |
1359 | void *data; | |
1360 | ||
c3470cde | 1361 | if (!data_size) |
81e84650 | 1362 | return true; |
c3470cde | 1363 | |
45bb912b | 1364 | page = drbd_pp_alloc(mdev, 1, 1); |
b411b363 PR |
1365 | |
1366 | data = kmap(page); | |
1367 | while (data_size) { | |
1368 | rr = drbd_recv(mdev, data, min_t(int, data_size, PAGE_SIZE)); | |
1369 | if (rr != min_t(int, data_size, PAGE_SIZE)) { | |
1370 | rv = 0; | |
0ddc5549 LE |
1371 | if (!signal_pending(current)) |
1372 | dev_warn(DEV, | |
1373 | "short read receiving data: read %d expected %d\n", | |
1374 | rr, min_t(int, data_size, PAGE_SIZE)); | |
b411b363 PR |
1375 | break; |
1376 | } | |
1377 | data_size -= rr; | |
1378 | } | |
1379 | kunmap(page); | |
435f0740 | 1380 | drbd_pp_free(mdev, page, 0); |
b411b363 PR |
1381 | return rv; |
1382 | } | |
1383 | ||
1384 | static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, | |
1385 | sector_t sector, int data_size) | |
1386 | { | |
1387 | struct bio_vec *bvec; | |
1388 | struct bio *bio; | |
1389 | int dgs, rr, i, expect; | |
a0638456 PR |
1390 | void *dig_in = mdev->tconn->int_dig_in; |
1391 | void *dig_vv = mdev->tconn->int_dig_vv; | |
b411b363 | 1392 | |
a0638456 PR |
1393 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ? |
1394 | crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0; | |
b411b363 PR |
1395 | |
1396 | if (dgs) { | |
1397 | rr = drbd_recv(mdev, dig_in, dgs); | |
1398 | if (rr != dgs) { | |
0ddc5549 LE |
1399 | if (!signal_pending(current)) |
1400 | dev_warn(DEV, | |
1401 | "short read receiving data reply digest: read %d expected %d\n", | |
1402 | rr, dgs); | |
b411b363 PR |
1403 | return 0; |
1404 | } | |
1405 | } | |
1406 | ||
1407 | data_size -= dgs; | |
1408 | ||
1409 | /* optimistically update recv_cnt. if receiving fails below, | |
1410 | * we disconnect anyways, and counters will be reset. */ | |
1411 | mdev->recv_cnt += data_size>>9; | |
1412 | ||
1413 | bio = req->master_bio; | |
1414 | D_ASSERT(sector == bio->bi_sector); | |
1415 | ||
1416 | bio_for_each_segment(bvec, bio, i) { | |
1417 | expect = min_t(int, data_size, bvec->bv_len); | |
1418 | rr = drbd_recv(mdev, | |
1419 | kmap(bvec->bv_page)+bvec->bv_offset, | |
1420 | expect); | |
1421 | kunmap(bvec->bv_page); | |
1422 | if (rr != expect) { | |
0ddc5549 LE |
1423 | if (!signal_pending(current)) |
1424 | dev_warn(DEV, "short read receiving data reply: " | |
1425 | "read %d expected %d\n", | |
1426 | rr, expect); | |
b411b363 PR |
1427 | return 0; |
1428 | } | |
1429 | data_size -= rr; | |
1430 | } | |
1431 | ||
1432 | if (dgs) { | |
a0638456 | 1433 | drbd_csum_bio(mdev, mdev->tconn->integrity_r_tfm, bio, dig_vv); |
b411b363 PR |
1434 | if (memcmp(dig_in, dig_vv, dgs)) { |
1435 | dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); | |
1436 | return 0; | |
1437 | } | |
1438 | } | |
1439 | ||
1440 | D_ASSERT(data_size == 0); | |
1441 | return 1; | |
1442 | } | |
1443 | ||
1444 | /* e_end_resync_block() is called via | |
1445 | * drbd_process_done_ee() by asender only */ | |
1446 | static int e_end_resync_block(struct drbd_conf *mdev, struct drbd_work *w, int unused) | |
1447 | { | |
f6ffca9f | 1448 | struct drbd_peer_request *e = (struct drbd_peer_request *)w; |
010f6e67 | 1449 | sector_t sector = e->i.sector; |
b411b363 PR |
1450 | int ok; |
1451 | ||
8b946255 | 1452 | D_ASSERT(drbd_interval_empty(&e->i)); |
b411b363 | 1453 | |
45bb912b | 1454 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
010f6e67 | 1455 | drbd_set_in_sync(mdev, sector, e->i.size); |
b411b363 PR |
1456 | ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, e); |
1457 | } else { | |
1458 | /* Record failure to sync */ | |
010f6e67 | 1459 | drbd_rs_failed_io(mdev, sector, e->i.size); |
b411b363 PR |
1460 | |
1461 | ok = drbd_send_ack(mdev, P_NEG_ACK, e); | |
1462 | } | |
1463 | dec_unacked(mdev); | |
1464 | ||
1465 | return ok; | |
1466 | } | |
1467 | ||
1468 | static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) | |
1469 | { | |
f6ffca9f | 1470 | struct drbd_peer_request *e; |
b411b363 PR |
1471 | |
1472 | e = read_in_block(mdev, ID_SYNCER, sector, data_size); | |
45bb912b LE |
1473 | if (!e) |
1474 | goto fail; | |
b411b363 PR |
1475 | |
1476 | dec_rs_pending(mdev); | |
1477 | ||
b411b363 PR |
1478 | inc_unacked(mdev); |
1479 | /* corresponding dec_unacked() in e_end_resync_block() | |
1480 | * respective _drbd_clear_done_ee */ | |
1481 | ||
45bb912b LE |
1482 | e->w.cb = e_end_resync_block; |
1483 | ||
87eeee41 | 1484 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 1485 | list_add(&e->w.list, &mdev->sync_ee); |
87eeee41 | 1486 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1487 | |
0f0601f4 | 1488 | atomic_add(data_size >> 9, &mdev->rs_sect_ev); |
45bb912b | 1489 | if (drbd_submit_ee(mdev, e, WRITE, DRBD_FAULT_RS_WR) == 0) |
81e84650 | 1490 | return true; |
b411b363 | 1491 | |
10f6d992 LE |
1492 | /* don't care for the reason here */ |
1493 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 1494 | spin_lock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 1495 | list_del(&e->w.list); |
87eeee41 | 1496 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 1497 | |
45bb912b LE |
1498 | drbd_free_ee(mdev, e); |
1499 | fail: | |
1500 | put_ldev(mdev); | |
81e84650 | 1501 | return false; |
b411b363 PR |
1502 | } |
1503 | ||
668eebc6 | 1504 | static struct drbd_request * |
bc9c5c41 AG |
1505 | find_request(struct drbd_conf *mdev, struct rb_root *root, u64 id, |
1506 | sector_t sector, bool missing_ok, const char *func) | |
51624585 | 1507 | { |
51624585 AG |
1508 | struct drbd_request *req; |
1509 | ||
bc9c5c41 AG |
1510 | /* Request object according to our peer */ |
1511 | req = (struct drbd_request *)(unsigned long)id; | |
5e472264 | 1512 | if (drbd_contains_interval(root, sector, &req->i) && req->i.local) |
668eebc6 | 1513 | return req; |
c3afd8f5 AG |
1514 | if (!missing_ok) { |
1515 | dev_err(DEV, "%s: failed to find request %lu, sector %llus\n", func, | |
1516 | (unsigned long)id, (unsigned long long)sector); | |
1517 | } | |
51624585 AG |
1518 | return NULL; |
1519 | } | |
1520 | ||
d8763023 AG |
1521 | static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packet cmd, |
1522 | unsigned int data_size) | |
b411b363 PR |
1523 | { |
1524 | struct drbd_request *req; | |
1525 | sector_t sector; | |
b411b363 | 1526 | int ok; |
e42325a5 | 1527 | struct p_data *p = &mdev->tconn->data.rbuf.data; |
b411b363 PR |
1528 | |
1529 | sector = be64_to_cpu(p->sector); | |
1530 | ||
87eeee41 | 1531 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 1532 | req = find_request(mdev, &mdev->read_requests, p->block_id, sector, false, __func__); |
87eeee41 | 1533 | spin_unlock_irq(&mdev->tconn->req_lock); |
c3afd8f5 | 1534 | if (unlikely(!req)) |
81e84650 | 1535 | return false; |
b411b363 | 1536 | |
24c4830c | 1537 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1538 | * special casing it there for the various failure cases. |
1539 | * still no race with drbd_fail_pending_reads */ | |
1540 | ok = recv_dless_read(mdev, req, sector, data_size); | |
1541 | ||
1542 | if (ok) | |
8554df1c | 1543 | req_mod(req, DATA_RECEIVED); |
b411b363 PR |
1544 | /* else: nothing. handled from drbd_disconnect... |
1545 | * I don't think we may complete this just yet | |
1546 | * in case we are "on-disconnect: freeze" */ | |
1547 | ||
1548 | return ok; | |
1549 | } | |
1550 | ||
d8763023 AG |
1551 | static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packet cmd, |
1552 | unsigned int data_size) | |
b411b363 PR |
1553 | { |
1554 | sector_t sector; | |
b411b363 | 1555 | int ok; |
e42325a5 | 1556 | struct p_data *p = &mdev->tconn->data.rbuf.data; |
b411b363 PR |
1557 | |
1558 | sector = be64_to_cpu(p->sector); | |
1559 | D_ASSERT(p->block_id == ID_SYNCER); | |
1560 | ||
1561 | if (get_ldev(mdev)) { | |
1562 | /* data is submitted to disk within recv_resync_read. | |
1563 | * corresponding put_ldev done below on error, | |
9c50842a | 1564 | * or in drbd_endio_sec. */ |
b411b363 PR |
1565 | ok = recv_resync_read(mdev, sector, data_size); |
1566 | } else { | |
1567 | if (__ratelimit(&drbd_ratelimit_state)) | |
1568 | dev_err(DEV, "Can not write resync data to local disk.\n"); | |
1569 | ||
1570 | ok = drbd_drain_block(mdev, data_size); | |
1571 | ||
2b2bf214 | 1572 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 PR |
1573 | } |
1574 | ||
778f271d PR |
1575 | atomic_add(data_size >> 9, &mdev->rs_sect_in); |
1576 | ||
b411b363 PR |
1577 | return ok; |
1578 | } | |
1579 | ||
1580 | /* e_end_block() is called via drbd_process_done_ee(). | |
1581 | * this means this function only runs in the asender thread | |
1582 | */ | |
1583 | static int e_end_block(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1584 | { | |
f6ffca9f | 1585 | struct drbd_peer_request *e = (struct drbd_peer_request *)w; |
010f6e67 | 1586 | sector_t sector = e->i.sector; |
b411b363 PR |
1587 | int ok = 1, pcmd; |
1588 | ||
89e58e75 | 1589 | if (mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C) { |
45bb912b | 1590 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1591 | pcmd = (mdev->state.conn >= C_SYNC_SOURCE && |
1592 | mdev->state.conn <= C_PAUSED_SYNC_T && | |
1593 | e->flags & EE_MAY_SET_IN_SYNC) ? | |
1594 | P_RS_WRITE_ACK : P_WRITE_ACK; | |
1595 | ok &= drbd_send_ack(mdev, pcmd, e); | |
1596 | if (pcmd == P_RS_WRITE_ACK) | |
010f6e67 | 1597 | drbd_set_in_sync(mdev, sector, e->i.size); |
b411b363 PR |
1598 | } else { |
1599 | ok = drbd_send_ack(mdev, P_NEG_ACK, e); | |
1600 | /* we expect it to be marked out of sync anyways... | |
1601 | * maybe assert this? */ | |
1602 | } | |
1603 | dec_unacked(mdev); | |
1604 | } | |
1605 | /* we delete from the conflict detection hash _after_ we sent out the | |
1606 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
89e58e75 | 1607 | if (mdev->tconn->net_conf->two_primaries) { |
87eeee41 | 1608 | spin_lock_irq(&mdev->tconn->req_lock); |
8b946255 | 1609 | D_ASSERT(!drbd_interval_empty(&e->i)); |
53840641 | 1610 | drbd_remove_epoch_entry_interval(mdev, e); |
87eeee41 | 1611 | spin_unlock_irq(&mdev->tconn->req_lock); |
bb3bfe96 | 1612 | } else |
8b946255 | 1613 | D_ASSERT(drbd_interval_empty(&e->i)); |
b411b363 PR |
1614 | |
1615 | drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); | |
1616 | ||
1617 | return ok; | |
1618 | } | |
1619 | ||
1620 | static int e_send_discard_ack(struct drbd_conf *mdev, struct drbd_work *w, int unused) | |
1621 | { | |
f6ffca9f | 1622 | struct drbd_peer_request *e = (struct drbd_peer_request *)w; |
b411b363 PR |
1623 | int ok = 1; |
1624 | ||
89e58e75 | 1625 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
b411b363 PR |
1626 | ok = drbd_send_ack(mdev, P_DISCARD_ACK, e); |
1627 | ||
87eeee41 | 1628 | spin_lock_irq(&mdev->tconn->req_lock); |
8b946255 | 1629 | D_ASSERT(!drbd_interval_empty(&e->i)); |
53840641 | 1630 | drbd_remove_epoch_entry_interval(mdev, e); |
87eeee41 | 1631 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
1632 | |
1633 | dec_unacked(mdev); | |
1634 | ||
1635 | return ok; | |
1636 | } | |
1637 | ||
3e394da1 AG |
1638 | static bool seq_greater(u32 a, u32 b) |
1639 | { | |
1640 | /* | |
1641 | * We assume 32-bit wrap-around here. | |
1642 | * For 24-bit wrap-around, we would have to shift: | |
1643 | * a <<= 8; b <<= 8; | |
1644 | */ | |
1645 | return (s32)a - (s32)b > 0; | |
1646 | } | |
1647 | ||
1648 | static u32 seq_max(u32 a, u32 b) | |
1649 | { | |
1650 | return seq_greater(a, b) ? a : b; | |
1651 | } | |
1652 | ||
43ae077d | 1653 | static void update_peer_seq(struct drbd_conf *mdev, unsigned int peer_seq) |
3e394da1 | 1654 | { |
43ae077d | 1655 | unsigned int old_peer_seq; |
3e394da1 AG |
1656 | |
1657 | spin_lock(&mdev->peer_seq_lock); | |
43ae077d AG |
1658 | old_peer_seq = mdev->peer_seq; |
1659 | mdev->peer_seq = seq_max(mdev->peer_seq, peer_seq); | |
3e394da1 | 1660 | spin_unlock(&mdev->peer_seq_lock); |
43ae077d | 1661 | if (old_peer_seq != peer_seq) |
3e394da1 AG |
1662 | wake_up(&mdev->seq_wait); |
1663 | } | |
1664 | ||
b411b363 PR |
1665 | /* Called from receive_Data. |
1666 | * Synchronize packets on sock with packets on msock. | |
1667 | * | |
1668 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
1669 | * packet traveling on msock, they are still processed in the order they have | |
1670 | * been sent. | |
1671 | * | |
1672 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
1673 | * | |
1674 | * In case packet_seq is larger than mdev->peer_seq number, there are | |
1675 | * outstanding packets on the msock. We wait for them to arrive. | |
1676 | * In case we are the logically next packet, we update mdev->peer_seq | |
1677 | * ourselves. Correctly handles 32bit wrap around. | |
1678 | * | |
1679 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
1680 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
1681 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
1682 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
1683 | * | |
1684 | * returns 0 if we may process the packet, | |
1685 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
1686 | static int drbd_wait_peer_seq(struct drbd_conf *mdev, const u32 packet_seq) | |
1687 | { | |
1688 | DEFINE_WAIT(wait); | |
1689 | unsigned int p_seq; | |
1690 | long timeout; | |
1691 | int ret = 0; | |
1692 | spin_lock(&mdev->peer_seq_lock); | |
1693 | for (;;) { | |
1694 | prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); | |
3e394da1 | 1695 | if (!seq_greater(packet_seq, mdev->peer_seq + 1)) |
b411b363 PR |
1696 | break; |
1697 | if (signal_pending(current)) { | |
1698 | ret = -ERESTARTSYS; | |
1699 | break; | |
1700 | } | |
1701 | p_seq = mdev->peer_seq; | |
1702 | spin_unlock(&mdev->peer_seq_lock); | |
1703 | timeout = schedule_timeout(30*HZ); | |
1704 | spin_lock(&mdev->peer_seq_lock); | |
1705 | if (timeout == 0 && p_seq == mdev->peer_seq) { | |
1706 | ret = -ETIMEDOUT; | |
1707 | dev_err(DEV, "ASSERT FAILED waited 30 seconds for sequence update, forcing reconnect\n"); | |
1708 | break; | |
1709 | } | |
1710 | } | |
1711 | finish_wait(&mdev->seq_wait, &wait); | |
1712 | if (mdev->peer_seq+1 == packet_seq) | |
1713 | mdev->peer_seq++; | |
1714 | spin_unlock(&mdev->peer_seq_lock); | |
1715 | return ret; | |
1716 | } | |
1717 | ||
688593c5 LE |
1718 | /* see also bio_flags_to_wire() |
1719 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
1720 | * flags and back. We may replicate to other kernel versions. */ | |
1721 | static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf) | |
76d2e7ec | 1722 | { |
688593c5 LE |
1723 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
1724 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
1725 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
1726 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
1727 | } |
1728 | ||
b411b363 | 1729 | /* mirrored write */ |
d8763023 AG |
1730 | static int receive_Data(struct drbd_conf *mdev, enum drbd_packet cmd, |
1731 | unsigned int data_size) | |
b411b363 PR |
1732 | { |
1733 | sector_t sector; | |
f6ffca9f | 1734 | struct drbd_peer_request *e; |
e42325a5 | 1735 | struct p_data *p = &mdev->tconn->data.rbuf.data; |
b411b363 PR |
1736 | int rw = WRITE; |
1737 | u32 dp_flags; | |
1738 | ||
b411b363 | 1739 | if (!get_ldev(mdev)) { |
b411b363 PR |
1740 | spin_lock(&mdev->peer_seq_lock); |
1741 | if (mdev->peer_seq+1 == be32_to_cpu(p->seq_num)) | |
1742 | mdev->peer_seq++; | |
1743 | spin_unlock(&mdev->peer_seq_lock); | |
1744 | ||
2b2bf214 | 1745 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 PR |
1746 | atomic_inc(&mdev->current_epoch->epoch_size); |
1747 | return drbd_drain_block(mdev, data_size); | |
1748 | } | |
1749 | ||
1750 | /* get_ldev(mdev) successful. | |
1751 | * Corresponding put_ldev done either below (on various errors), | |
9c50842a | 1752 | * or in drbd_endio_sec, if we successfully submit the data at |
b411b363 PR |
1753 | * the end of this function. */ |
1754 | ||
1755 | sector = be64_to_cpu(p->sector); | |
1756 | e = read_in_block(mdev, p->block_id, sector, data_size); | |
1757 | if (!e) { | |
1758 | put_ldev(mdev); | |
81e84650 | 1759 | return false; |
b411b363 PR |
1760 | } |
1761 | ||
b411b363 PR |
1762 | e->w.cb = e_end_block; |
1763 | ||
688593c5 LE |
1764 | dp_flags = be32_to_cpu(p->dp_flags); |
1765 | rw |= wire_flags_to_bio(mdev, dp_flags); | |
1766 | ||
1767 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
1768 | e->flags |= EE_MAY_SET_IN_SYNC; | |
1769 | ||
b411b363 PR |
1770 | spin_lock(&mdev->epoch_lock); |
1771 | e->epoch = mdev->current_epoch; | |
1772 | atomic_inc(&e->epoch->epoch_size); | |
1773 | atomic_inc(&e->epoch->active); | |
b411b363 PR |
1774 | spin_unlock(&mdev->epoch_lock); |
1775 | ||
b411b363 | 1776 | /* I'm the receiver, I do hold a net_cnt reference. */ |
89e58e75 | 1777 | if (!mdev->tconn->net_conf->two_primaries) { |
87eeee41 | 1778 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
1779 | } else { |
1780 | /* don't get the req_lock yet, | |
1781 | * we may sleep in drbd_wait_peer_seq */ | |
010f6e67 | 1782 | const int size = e->i.size; |
b411b363 PR |
1783 | const int discard = test_bit(DISCARD_CONCURRENT, &mdev->flags); |
1784 | DEFINE_WAIT(wait); | |
b411b363 PR |
1785 | int first; |
1786 | ||
89e58e75 | 1787 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
b411b363 PR |
1788 | |
1789 | /* conflict detection and handling: | |
1790 | * 1. wait on the sequence number, | |
1791 | * in case this data packet overtook ACK packets. | |
5e472264 | 1792 | * 2. check for conflicting write requests. |
b411b363 PR |
1793 | * |
1794 | * Note: for two_primaries, we are protocol C, | |
1795 | * so there cannot be any request that is DONE | |
1796 | * but still on the transfer log. | |
1797 | * | |
b411b363 PR |
1798 | * if no conflicting request is found: |
1799 | * submit. | |
1800 | * | |
1801 | * if any conflicting request is found | |
1802 | * that has not yet been acked, | |
1803 | * AND I have the "discard concurrent writes" flag: | |
1804 | * queue (via done_ee) the P_DISCARD_ACK; OUT. | |
1805 | * | |
1806 | * if any conflicting request is found: | |
1807 | * block the receiver, waiting on misc_wait | |
1808 | * until no more conflicting requests are there, | |
1809 | * or we get interrupted (disconnect). | |
1810 | * | |
1811 | * we do not just write after local io completion of those | |
1812 | * requests, but only after req is done completely, i.e. | |
1813 | * we wait for the P_DISCARD_ACK to arrive! | |
1814 | * | |
1815 | * then proceed normally, i.e. submit. | |
1816 | */ | |
1817 | if (drbd_wait_peer_seq(mdev, be32_to_cpu(p->seq_num))) | |
1818 | goto out_interrupted; | |
1819 | ||
87eeee41 | 1820 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 1821 | |
b411b363 PR |
1822 | first = 1; |
1823 | for (;;) { | |
de696716 | 1824 | struct drbd_interval *i; |
b411b363 PR |
1825 | int have_unacked = 0; |
1826 | int have_conflict = 0; | |
1827 | prepare_to_wait(&mdev->misc_wait, &wait, | |
1828 | TASK_INTERRUPTIBLE); | |
de696716 AG |
1829 | |
1830 | i = drbd_find_overlap(&mdev->write_requests, sector, size); | |
1831 | if (i) { | |
de696716 AG |
1832 | /* only ALERT on first iteration, |
1833 | * we may be woken up early... */ | |
1834 | if (first) | |
5e472264 | 1835 | dev_alert(DEV, "%s[%u] Concurrent %s write detected!" |
de696716 AG |
1836 | " new: %llus +%u; pending: %llus +%u\n", |
1837 | current->comm, current->pid, | |
5e472264 | 1838 | i->local ? "local" : "remote", |
de696716 | 1839 | (unsigned long long)sector, size, |
5e472264 AG |
1840 | (unsigned long long)i->sector, i->size); |
1841 | ||
1842 | if (i->local) { | |
1843 | struct drbd_request *req2; | |
1844 | ||
1845 | req2 = container_of(i, struct drbd_request, i); | |
1846 | if (req2->rq_state & RQ_NET_PENDING) | |
1847 | ++have_unacked; | |
1848 | } | |
de696716 | 1849 | ++have_conflict; |
b411b363 | 1850 | } |
b411b363 PR |
1851 | if (!have_conflict) |
1852 | break; | |
1853 | ||
1854 | /* Discard Ack only for the _first_ iteration */ | |
1855 | if (first && discard && have_unacked) { | |
1856 | dev_alert(DEV, "Concurrent write! [DISCARD BY FLAG] sec=%llus\n", | |
1857 | (unsigned long long)sector); | |
1858 | inc_unacked(mdev); | |
1859 | e->w.cb = e_send_discard_ack; | |
1860 | list_add_tail(&e->w.list, &mdev->done_ee); | |
1861 | ||
87eeee41 | 1862 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
1863 | |
1864 | /* we could probably send that P_DISCARD_ACK ourselves, | |
1865 | * but I don't like the receiver using the msock */ | |
1866 | ||
1867 | put_ldev(mdev); | |
1868 | wake_asender(mdev); | |
1869 | finish_wait(&mdev->misc_wait, &wait); | |
81e84650 | 1870 | return true; |
b411b363 PR |
1871 | } |
1872 | ||
1873 | if (signal_pending(current)) { | |
87eeee41 | 1874 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
1875 | finish_wait(&mdev->misc_wait, &wait); |
1876 | goto out_interrupted; | |
1877 | } | |
1878 | ||
a500c2ef | 1879 | /* Indicate to wake up mdev->misc_wait upon completion. */ |
53840641 | 1880 | i->waiting = true; |
a500c2ef | 1881 | |
87eeee41 | 1882 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
1883 | if (first) { |
1884 | first = 0; | |
1885 | dev_alert(DEV, "Concurrent write! [W AFTERWARDS] " | |
1886 | "sec=%llus\n", (unsigned long long)sector); | |
1887 | } else if (discard) { | |
1888 | /* we had none on the first iteration. | |
1889 | * there must be none now. */ | |
1890 | D_ASSERT(have_unacked == 0); | |
1891 | } | |
1892 | schedule(); | |
87eeee41 | 1893 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
1894 | } |
1895 | finish_wait(&mdev->misc_wait, &wait); | |
5e472264 AG |
1896 | |
1897 | drbd_insert_interval(&mdev->write_requests, &e->i); | |
b411b363 PR |
1898 | } |
1899 | ||
1900 | list_add(&e->w.list, &mdev->active_ee); | |
87eeee41 | 1901 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1902 | |
89e58e75 | 1903 | switch (mdev->tconn->net_conf->wire_protocol) { |
b411b363 PR |
1904 | case DRBD_PROT_C: |
1905 | inc_unacked(mdev); | |
1906 | /* corresponding dec_unacked() in e_end_block() | |
1907 | * respective _drbd_clear_done_ee */ | |
1908 | break; | |
1909 | case DRBD_PROT_B: | |
1910 | /* I really don't like it that the receiver thread | |
1911 | * sends on the msock, but anyways */ | |
1912 | drbd_send_ack(mdev, P_RECV_ACK, e); | |
1913 | break; | |
1914 | case DRBD_PROT_A: | |
1915 | /* nothing to do */ | |
1916 | break; | |
1917 | } | |
1918 | ||
6719fb03 | 1919 | if (mdev->state.pdsk < D_INCONSISTENT) { |
b411b363 | 1920 | /* In case we have the only disk of the cluster, */ |
010f6e67 | 1921 | drbd_set_out_of_sync(mdev, e->i.sector, e->i.size); |
b411b363 | 1922 | e->flags |= EE_CALL_AL_COMPLETE_IO; |
6719fb03 | 1923 | e->flags &= ~EE_MAY_SET_IN_SYNC; |
010f6e67 | 1924 | drbd_al_begin_io(mdev, e->i.sector); |
b411b363 PR |
1925 | } |
1926 | ||
45bb912b | 1927 | if (drbd_submit_ee(mdev, e, rw, DRBD_FAULT_DT_WR) == 0) |
81e84650 | 1928 | return true; |
b411b363 | 1929 | |
10f6d992 LE |
1930 | /* don't care for the reason here */ |
1931 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 1932 | spin_lock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 1933 | list_del(&e->w.list); |
53840641 | 1934 | drbd_remove_epoch_entry_interval(mdev, e); |
87eeee41 | 1935 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 1936 | if (e->flags & EE_CALL_AL_COMPLETE_IO) |
010f6e67 | 1937 | drbd_al_complete_io(mdev, e->i.sector); |
22cc37a9 | 1938 | |
b411b363 | 1939 | out_interrupted: |
10f6d992 | 1940 | drbd_may_finish_epoch(mdev, e->epoch, EV_PUT + EV_CLEANUP); |
b411b363 PR |
1941 | put_ldev(mdev); |
1942 | drbd_free_ee(mdev, e); | |
81e84650 | 1943 | return false; |
b411b363 PR |
1944 | } |
1945 | ||
0f0601f4 LE |
1946 | /* We may throttle resync, if the lower device seems to be busy, |
1947 | * and current sync rate is above c_min_rate. | |
1948 | * | |
1949 | * To decide whether or not the lower device is busy, we use a scheme similar | |
1950 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
1951 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
1952 | * activity, it obviously is "busy". | |
1953 | * | |
1954 | * The current sync rate used here uses only the most recent two step marks, | |
1955 | * to have a short time average so we can react faster. | |
1956 | */ | |
e3555d85 | 1957 | int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector) |
0f0601f4 LE |
1958 | { |
1959 | struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; | |
1960 | unsigned long db, dt, dbdt; | |
e3555d85 | 1961 | struct lc_element *tmp; |
0f0601f4 LE |
1962 | int curr_events; |
1963 | int throttle = 0; | |
1964 | ||
1965 | /* feature disabled? */ | |
1966 | if (mdev->sync_conf.c_min_rate == 0) | |
1967 | return 0; | |
1968 | ||
e3555d85 PR |
1969 | spin_lock_irq(&mdev->al_lock); |
1970 | tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector)); | |
1971 | if (tmp) { | |
1972 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
1973 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) { | |
1974 | spin_unlock_irq(&mdev->al_lock); | |
1975 | return 0; | |
1976 | } | |
1977 | /* Do not slow down if app IO is already waiting for this extent */ | |
1978 | } | |
1979 | spin_unlock_irq(&mdev->al_lock); | |
1980 | ||
0f0601f4 LE |
1981 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
1982 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
1983 | atomic_read(&mdev->rs_sect_ev); | |
e3555d85 | 1984 | |
0f0601f4 LE |
1985 | if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { |
1986 | unsigned long rs_left; | |
1987 | int i; | |
1988 | ||
1989 | mdev->rs_last_events = curr_events; | |
1990 | ||
1991 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
1992 | * approx. */ | |
2649f080 LE |
1993 | i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
1994 | ||
1995 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) | |
1996 | rs_left = mdev->ov_left; | |
1997 | else | |
1998 | rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
0f0601f4 LE |
1999 | |
2000 | dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; | |
2001 | if (!dt) | |
2002 | dt++; | |
2003 | db = mdev->rs_mark_left[i] - rs_left; | |
2004 | dbdt = Bit2KB(db/dt); | |
2005 | ||
2006 | if (dbdt > mdev->sync_conf.c_min_rate) | |
2007 | throttle = 1; | |
2008 | } | |
2009 | return throttle; | |
2010 | } | |
2011 | ||
2012 | ||
d8763023 AG |
2013 | static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packet cmd, |
2014 | unsigned int digest_size) | |
b411b363 PR |
2015 | { |
2016 | sector_t sector; | |
2017 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
f6ffca9f | 2018 | struct drbd_peer_request *e; |
b411b363 | 2019 | struct digest_info *di = NULL; |
b18b37be | 2020 | int size, verb; |
b411b363 | 2021 | unsigned int fault_type; |
e42325a5 | 2022 | struct p_block_req *p = &mdev->tconn->data.rbuf.block_req; |
b411b363 PR |
2023 | |
2024 | sector = be64_to_cpu(p->sector); | |
2025 | size = be32_to_cpu(p->blksize); | |
2026 | ||
1816a2b4 | 2027 | if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
2028 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
2029 | (unsigned long long)sector, size); | |
81e84650 | 2030 | return false; |
b411b363 PR |
2031 | } |
2032 | if (sector + (size>>9) > capacity) { | |
2033 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, | |
2034 | (unsigned long long)sector, size); | |
81e84650 | 2035 | return false; |
b411b363 PR |
2036 | } |
2037 | ||
2038 | if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { | |
b18b37be PR |
2039 | verb = 1; |
2040 | switch (cmd) { | |
2041 | case P_DATA_REQUEST: | |
2042 | drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); | |
2043 | break; | |
2044 | case P_RS_DATA_REQUEST: | |
2045 | case P_CSUM_RS_REQUEST: | |
2046 | case P_OV_REQUEST: | |
2047 | drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); | |
2048 | break; | |
2049 | case P_OV_REPLY: | |
2050 | verb = 0; | |
2051 | dec_rs_pending(mdev); | |
2052 | drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); | |
2053 | break; | |
2054 | default: | |
2055 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
2056 | cmdname(cmd)); | |
2057 | } | |
2058 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
b411b363 PR |
2059 | dev_err(DEV, "Can not satisfy peer's read request, " |
2060 | "no local data.\n"); | |
b18b37be | 2061 | |
a821cc4a LE |
2062 | /* drain possibly payload */ |
2063 | return drbd_drain_block(mdev, digest_size); | |
b411b363 PR |
2064 | } |
2065 | ||
2066 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2067 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2068 | * which in turn might block on the other node at this very place. */ | |
2069 | e = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); | |
2070 | if (!e) { | |
2071 | put_ldev(mdev); | |
81e84650 | 2072 | return false; |
b411b363 PR |
2073 | } |
2074 | ||
02918be2 | 2075 | switch (cmd) { |
b411b363 PR |
2076 | case P_DATA_REQUEST: |
2077 | e->w.cb = w_e_end_data_req; | |
2078 | fault_type = DRBD_FAULT_DT_RD; | |
80a40e43 LE |
2079 | /* application IO, don't drbd_rs_begin_io */ |
2080 | goto submit; | |
2081 | ||
b411b363 PR |
2082 | case P_RS_DATA_REQUEST: |
2083 | e->w.cb = w_e_end_rsdata_req; | |
2084 | fault_type = DRBD_FAULT_RS_RD; | |
5f9915bb LE |
2085 | /* used in the sector offset progress display */ |
2086 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
b411b363 PR |
2087 | break; |
2088 | ||
2089 | case P_OV_REPLY: | |
2090 | case P_CSUM_RS_REQUEST: | |
2091 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2092 | di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO); |
2093 | if (!di) | |
2094 | goto out_free_e; | |
2095 | ||
2096 | di->digest_size = digest_size; | |
2097 | di->digest = (((char *)di)+sizeof(struct digest_info)); | |
2098 | ||
c36c3ced LE |
2099 | e->digest = di; |
2100 | e->flags |= EE_HAS_DIGEST; | |
2101 | ||
b411b363 PR |
2102 | if (drbd_recv(mdev, di->digest, digest_size) != digest_size) |
2103 | goto out_free_e; | |
2104 | ||
02918be2 | 2105 | if (cmd == P_CSUM_RS_REQUEST) { |
31890f4a | 2106 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
b411b363 | 2107 | e->w.cb = w_e_end_csum_rs_req; |
5f9915bb LE |
2108 | /* used in the sector offset progress display */ |
2109 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
02918be2 | 2110 | } else if (cmd == P_OV_REPLY) { |
2649f080 LE |
2111 | /* track progress, we may need to throttle */ |
2112 | atomic_add(size >> 9, &mdev->rs_sect_in); | |
b411b363 PR |
2113 | e->w.cb = w_e_end_ov_reply; |
2114 | dec_rs_pending(mdev); | |
0f0601f4 LE |
2115 | /* drbd_rs_begin_io done when we sent this request, |
2116 | * but accounting still needs to be done. */ | |
2117 | goto submit_for_resync; | |
b411b363 PR |
2118 | } |
2119 | break; | |
2120 | ||
2121 | case P_OV_REQUEST: | |
b411b363 | 2122 | if (mdev->ov_start_sector == ~(sector_t)0 && |
31890f4a | 2123 | mdev->tconn->agreed_pro_version >= 90) { |
de228bba LE |
2124 | unsigned long now = jiffies; |
2125 | int i; | |
b411b363 PR |
2126 | mdev->ov_start_sector = sector; |
2127 | mdev->ov_position = sector; | |
30b743a2 LE |
2128 | mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector); |
2129 | mdev->rs_total = mdev->ov_left; | |
de228bba LE |
2130 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
2131 | mdev->rs_mark_left[i] = mdev->ov_left; | |
2132 | mdev->rs_mark_time[i] = now; | |
2133 | } | |
b411b363 PR |
2134 | dev_info(DEV, "Online Verify start sector: %llu\n", |
2135 | (unsigned long long)sector); | |
2136 | } | |
2137 | e->w.cb = w_e_end_ov_req; | |
2138 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2139 | break; |
2140 | ||
b411b363 PR |
2141 | default: |
2142 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
02918be2 | 2143 | cmdname(cmd)); |
b411b363 | 2144 | fault_type = DRBD_FAULT_MAX; |
80a40e43 | 2145 | goto out_free_e; |
b411b363 PR |
2146 | } |
2147 | ||
0f0601f4 LE |
2148 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2149 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2150 | * Various places in the resync start and stop logic assume resync | |
2151 | * requests are processed in order, requeuing this on the worker thread | |
2152 | * introduces a bunch of new code for synchronization between threads. | |
2153 | * | |
2154 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2155 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2156 | * for application writes for the same time. For now, just throttle | |
2157 | * here, where the rest of the code expects the receiver to sleep for | |
2158 | * a while, anyways. | |
2159 | */ | |
2160 | ||
2161 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2162 | * this defers syncer requests for some time, before letting at least | |
2163 | * on request through. The resync controller on the receiving side | |
2164 | * will adapt to the incoming rate accordingly. | |
2165 | * | |
2166 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2167 | * we would also throttle its application reads. | |
2168 | * In that case, throttling is done on the SyncTarget only. | |
2169 | */ | |
e3555d85 PR |
2170 | if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector)) |
2171 | schedule_timeout_uninterruptible(HZ/10); | |
2172 | if (drbd_rs_begin_io(mdev, sector)) | |
80a40e43 | 2173 | goto out_free_e; |
b411b363 | 2174 | |
0f0601f4 LE |
2175 | submit_for_resync: |
2176 | atomic_add(size >> 9, &mdev->rs_sect_ev); | |
2177 | ||
80a40e43 | 2178 | submit: |
b411b363 | 2179 | inc_unacked(mdev); |
87eeee41 | 2180 | spin_lock_irq(&mdev->tconn->req_lock); |
80a40e43 | 2181 | list_add_tail(&e->w.list, &mdev->read_ee); |
87eeee41 | 2182 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2183 | |
45bb912b | 2184 | if (drbd_submit_ee(mdev, e, READ, fault_type) == 0) |
81e84650 | 2185 | return true; |
b411b363 | 2186 | |
10f6d992 LE |
2187 | /* don't care for the reason here */ |
2188 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2189 | spin_lock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 2190 | list_del(&e->w.list); |
87eeee41 | 2191 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 LE |
2192 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ |
2193 | ||
b411b363 | 2194 | out_free_e: |
b411b363 PR |
2195 | put_ldev(mdev); |
2196 | drbd_free_ee(mdev, e); | |
81e84650 | 2197 | return false; |
b411b363 PR |
2198 | } |
2199 | ||
2200 | static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) | |
2201 | { | |
2202 | int self, peer, rv = -100; | |
2203 | unsigned long ch_self, ch_peer; | |
2204 | ||
2205 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2206 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2207 | ||
2208 | ch_peer = mdev->p_uuid[UI_SIZE]; | |
2209 | ch_self = mdev->comm_bm_set; | |
2210 | ||
89e58e75 | 2211 | switch (mdev->tconn->net_conf->after_sb_0p) { |
b411b363 PR |
2212 | case ASB_CONSENSUS: |
2213 | case ASB_DISCARD_SECONDARY: | |
2214 | case ASB_CALL_HELPER: | |
2215 | dev_err(DEV, "Configuration error.\n"); | |
2216 | break; | |
2217 | case ASB_DISCONNECT: | |
2218 | break; | |
2219 | case ASB_DISCARD_YOUNGER_PRI: | |
2220 | if (self == 0 && peer == 1) { | |
2221 | rv = -1; | |
2222 | break; | |
2223 | } | |
2224 | if (self == 1 && peer == 0) { | |
2225 | rv = 1; | |
2226 | break; | |
2227 | } | |
2228 | /* Else fall through to one of the other strategies... */ | |
2229 | case ASB_DISCARD_OLDER_PRI: | |
2230 | if (self == 0 && peer == 1) { | |
2231 | rv = 1; | |
2232 | break; | |
2233 | } | |
2234 | if (self == 1 && peer == 0) { | |
2235 | rv = -1; | |
2236 | break; | |
2237 | } | |
2238 | /* Else fall through to one of the other strategies... */ | |
ad19bf6e | 2239 | dev_warn(DEV, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2240 | "Using discard-least-changes instead\n"); |
2241 | case ASB_DISCARD_ZERO_CHG: | |
2242 | if (ch_peer == 0 && ch_self == 0) { | |
2243 | rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) | |
2244 | ? -1 : 1; | |
2245 | break; | |
2246 | } else { | |
2247 | if (ch_peer == 0) { rv = 1; break; } | |
2248 | if (ch_self == 0) { rv = -1; break; } | |
2249 | } | |
89e58e75 | 2250 | if (mdev->tconn->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) |
b411b363 PR |
2251 | break; |
2252 | case ASB_DISCARD_LEAST_CHG: | |
2253 | if (ch_self < ch_peer) | |
2254 | rv = -1; | |
2255 | else if (ch_self > ch_peer) | |
2256 | rv = 1; | |
2257 | else /* ( ch_self == ch_peer ) */ | |
2258 | /* Well, then use something else. */ | |
2259 | rv = test_bit(DISCARD_CONCURRENT, &mdev->flags) | |
2260 | ? -1 : 1; | |
2261 | break; | |
2262 | case ASB_DISCARD_LOCAL: | |
2263 | rv = -1; | |
2264 | break; | |
2265 | case ASB_DISCARD_REMOTE: | |
2266 | rv = 1; | |
2267 | } | |
2268 | ||
2269 | return rv; | |
2270 | } | |
2271 | ||
2272 | static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) | |
2273 | { | |
6184ea21 | 2274 | int hg, rv = -100; |
b411b363 | 2275 | |
89e58e75 | 2276 | switch (mdev->tconn->net_conf->after_sb_1p) { |
b411b363 PR |
2277 | case ASB_DISCARD_YOUNGER_PRI: |
2278 | case ASB_DISCARD_OLDER_PRI: | |
2279 | case ASB_DISCARD_LEAST_CHG: | |
2280 | case ASB_DISCARD_LOCAL: | |
2281 | case ASB_DISCARD_REMOTE: | |
2282 | dev_err(DEV, "Configuration error.\n"); | |
2283 | break; | |
2284 | case ASB_DISCONNECT: | |
2285 | break; | |
2286 | case ASB_CONSENSUS: | |
2287 | hg = drbd_asb_recover_0p(mdev); | |
2288 | if (hg == -1 && mdev->state.role == R_SECONDARY) | |
2289 | rv = hg; | |
2290 | if (hg == 1 && mdev->state.role == R_PRIMARY) | |
2291 | rv = hg; | |
2292 | break; | |
2293 | case ASB_VIOLENTLY: | |
2294 | rv = drbd_asb_recover_0p(mdev); | |
2295 | break; | |
2296 | case ASB_DISCARD_SECONDARY: | |
2297 | return mdev->state.role == R_PRIMARY ? 1 : -1; | |
2298 | case ASB_CALL_HELPER: | |
2299 | hg = drbd_asb_recover_0p(mdev); | |
2300 | if (hg == -1 && mdev->state.role == R_PRIMARY) { | |
bb437946 AG |
2301 | enum drbd_state_rv rv2; |
2302 | ||
2303 | drbd_set_role(mdev, R_SECONDARY, 0); | |
b411b363 PR |
2304 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2305 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2306 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2307 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2308 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2309 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2310 | } else { | |
2311 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2312 | rv = hg; | |
2313 | } | |
2314 | } else | |
2315 | rv = hg; | |
2316 | } | |
2317 | ||
2318 | return rv; | |
2319 | } | |
2320 | ||
2321 | static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) | |
2322 | { | |
6184ea21 | 2323 | int hg, rv = -100; |
b411b363 | 2324 | |
89e58e75 | 2325 | switch (mdev->tconn->net_conf->after_sb_2p) { |
b411b363 PR |
2326 | case ASB_DISCARD_YOUNGER_PRI: |
2327 | case ASB_DISCARD_OLDER_PRI: | |
2328 | case ASB_DISCARD_LEAST_CHG: | |
2329 | case ASB_DISCARD_LOCAL: | |
2330 | case ASB_DISCARD_REMOTE: | |
2331 | case ASB_CONSENSUS: | |
2332 | case ASB_DISCARD_SECONDARY: | |
2333 | dev_err(DEV, "Configuration error.\n"); | |
2334 | break; | |
2335 | case ASB_VIOLENTLY: | |
2336 | rv = drbd_asb_recover_0p(mdev); | |
2337 | break; | |
2338 | case ASB_DISCONNECT: | |
2339 | break; | |
2340 | case ASB_CALL_HELPER: | |
2341 | hg = drbd_asb_recover_0p(mdev); | |
2342 | if (hg == -1) { | |
bb437946 AG |
2343 | enum drbd_state_rv rv2; |
2344 | ||
b411b363 PR |
2345 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2346 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2347 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2348 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2349 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2350 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2351 | } else { | |
2352 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2353 | rv = hg; | |
2354 | } | |
2355 | } else | |
2356 | rv = hg; | |
2357 | } | |
2358 | ||
2359 | return rv; | |
2360 | } | |
2361 | ||
2362 | static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, | |
2363 | u64 bits, u64 flags) | |
2364 | { | |
2365 | if (!uuid) { | |
2366 | dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); | |
2367 | return; | |
2368 | } | |
2369 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", | |
2370 | text, | |
2371 | (unsigned long long)uuid[UI_CURRENT], | |
2372 | (unsigned long long)uuid[UI_BITMAP], | |
2373 | (unsigned long long)uuid[UI_HISTORY_START], | |
2374 | (unsigned long long)uuid[UI_HISTORY_END], | |
2375 | (unsigned long long)bits, | |
2376 | (unsigned long long)flags); | |
2377 | } | |
2378 | ||
2379 | /* | |
2380 | 100 after split brain try auto recover | |
2381 | 2 C_SYNC_SOURCE set BitMap | |
2382 | 1 C_SYNC_SOURCE use BitMap | |
2383 | 0 no Sync | |
2384 | -1 C_SYNC_TARGET use BitMap | |
2385 | -2 C_SYNC_TARGET set BitMap | |
2386 | -100 after split brain, disconnect | |
2387 | -1000 unrelated data | |
4a23f264 PR |
2388 | -1091 requires proto 91 |
2389 | -1096 requires proto 96 | |
b411b363 PR |
2390 | */ |
2391 | static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) | |
2392 | { | |
2393 | u64 self, peer; | |
2394 | int i, j; | |
2395 | ||
2396 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2397 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2398 | ||
2399 | *rule_nr = 10; | |
2400 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2401 | return 0; | |
2402 | ||
2403 | *rule_nr = 20; | |
2404 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2405 | peer != UUID_JUST_CREATED) | |
2406 | return -2; | |
2407 | ||
2408 | *rule_nr = 30; | |
2409 | if (self != UUID_JUST_CREATED && | |
2410 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2411 | return 2; | |
2412 | ||
2413 | if (self == peer) { | |
2414 | int rct, dc; /* roles at crash time */ | |
2415 | ||
2416 | if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { | |
2417 | ||
31890f4a | 2418 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2419 | return -1091; |
b411b363 PR |
2420 | |
2421 | if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && | |
2422 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
2423 | dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); | |
2424 | drbd_uuid_set_bm(mdev, 0UL); | |
2425 | ||
2426 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2427 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2428 | *rule_nr = 34; | |
2429 | } else { | |
2430 | dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); | |
2431 | *rule_nr = 36; | |
2432 | } | |
2433 | ||
2434 | return 1; | |
2435 | } | |
2436 | ||
2437 | if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { | |
2438 | ||
31890f4a | 2439 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2440 | return -1091; |
b411b363 PR |
2441 | |
2442 | if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && | |
2443 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
2444 | dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); | |
2445 | ||
2446 | mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; | |
2447 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; | |
2448 | mdev->p_uuid[UI_BITMAP] = 0UL; | |
2449 | ||
2450 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2451 | *rule_nr = 35; | |
2452 | } else { | |
2453 | dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); | |
2454 | *rule_nr = 37; | |
2455 | } | |
2456 | ||
2457 | return -1; | |
2458 | } | |
2459 | ||
2460 | /* Common power [off|failure] */ | |
2461 | rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + | |
2462 | (mdev->p_uuid[UI_FLAGS] & 2); | |
2463 | /* lowest bit is set when we were primary, | |
2464 | * next bit (weight 2) is set when peer was primary */ | |
2465 | *rule_nr = 40; | |
2466 | ||
2467 | switch (rct) { | |
2468 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2469 | case 1: /* self_pri && !peer_pri */ return 1; | |
2470 | case 2: /* !self_pri && peer_pri */ return -1; | |
2471 | case 3: /* self_pri && peer_pri */ | |
2472 | dc = test_bit(DISCARD_CONCURRENT, &mdev->flags); | |
2473 | return dc ? -1 : 1; | |
2474 | } | |
2475 | } | |
2476 | ||
2477 | *rule_nr = 50; | |
2478 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2479 | if (self == peer) | |
2480 | return -1; | |
2481 | ||
2482 | *rule_nr = 51; | |
2483 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2484 | if (self == peer) { | |
31890f4a | 2485 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2486 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == |
2487 | (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
2488 | peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2489 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2490 | resync as sync source modifications of the peer's UUIDs. */ | |
2491 | ||
31890f4a | 2492 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2493 | return -1091; |
b411b363 PR |
2494 | |
2495 | mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; | |
2496 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 PR |
2497 | |
2498 | dev_info(DEV, "Did not got last syncUUID packet, corrected:\n"); | |
2499 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2500 | ||
b411b363 PR |
2501 | return -1; |
2502 | } | |
2503 | } | |
2504 | ||
2505 | *rule_nr = 60; | |
2506 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2507 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2508 | peer = mdev->p_uuid[i] & ~((u64)1); | |
2509 | if (self == peer) | |
2510 | return -2; | |
2511 | } | |
2512 | ||
2513 | *rule_nr = 70; | |
2514 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2515 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2516 | if (self == peer) | |
2517 | return 1; | |
2518 | ||
2519 | *rule_nr = 71; | |
2520 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2521 | if (self == peer) { | |
31890f4a | 2522 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2523 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == |
2524 | (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
2525 | self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2526 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2527 | resync as sync source modifications of our UUIDs. */ | |
2528 | ||
31890f4a | 2529 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2530 | return -1091; |
b411b363 PR |
2531 | |
2532 | _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); | |
2533 | _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); | |
2534 | ||
4a23f264 | 2535 | dev_info(DEV, "Last syncUUID did not get through, corrected:\n"); |
b411b363 PR |
2536 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, |
2537 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2538 | ||
2539 | return 1; | |
2540 | } | |
2541 | } | |
2542 | ||
2543 | ||
2544 | *rule_nr = 80; | |
d8c2a36b | 2545 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 PR |
2546 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
2547 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2548 | if (self == peer) | |
2549 | return 2; | |
2550 | } | |
2551 | ||
2552 | *rule_nr = 90; | |
2553 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2554 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2555 | if (self == peer && self != ((u64)0)) | |
2556 | return 100; | |
2557 | ||
2558 | *rule_nr = 100; | |
2559 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2560 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2561 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { | |
2562 | peer = mdev->p_uuid[j] & ~((u64)1); | |
2563 | if (self == peer) | |
2564 | return -100; | |
2565 | } | |
2566 | } | |
2567 | ||
2568 | return -1000; | |
2569 | } | |
2570 | ||
2571 | /* drbd_sync_handshake() returns the new conn state on success, or | |
2572 | CONN_MASK (-1) on failure. | |
2573 | */ | |
2574 | static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, | |
2575 | enum drbd_disk_state peer_disk) __must_hold(local) | |
2576 | { | |
2577 | int hg, rule_nr; | |
2578 | enum drbd_conns rv = C_MASK; | |
2579 | enum drbd_disk_state mydisk; | |
2580 | ||
2581 | mydisk = mdev->state.disk; | |
2582 | if (mydisk == D_NEGOTIATING) | |
2583 | mydisk = mdev->new_state_tmp.disk; | |
2584 | ||
2585 | dev_info(DEV, "drbd_sync_handshake:\n"); | |
2586 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); | |
2587 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, | |
2588 | mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2589 | ||
2590 | hg = drbd_uuid_compare(mdev, &rule_nr); | |
2591 | ||
2592 | dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); | |
2593 | ||
2594 | if (hg == -1000) { | |
2595 | dev_alert(DEV, "Unrelated data, aborting!\n"); | |
2596 | return C_MASK; | |
2597 | } | |
4a23f264 PR |
2598 | if (hg < -1000) { |
2599 | dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); | |
b411b363 PR |
2600 | return C_MASK; |
2601 | } | |
2602 | ||
2603 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
2604 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
2605 | int f = (hg == -100) || abs(hg) == 2; | |
2606 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
2607 | if (f) | |
2608 | hg = hg*2; | |
2609 | dev_info(DEV, "Becoming sync %s due to disk states.\n", | |
2610 | hg > 0 ? "source" : "target"); | |
2611 | } | |
2612 | ||
3a11a487 AG |
2613 | if (abs(hg) == 100) |
2614 | drbd_khelper(mdev, "initial-split-brain"); | |
2615 | ||
89e58e75 | 2616 | if (hg == 100 || (hg == -100 && mdev->tconn->net_conf->always_asbp)) { |
b411b363 PR |
2617 | int pcount = (mdev->state.role == R_PRIMARY) |
2618 | + (peer_role == R_PRIMARY); | |
2619 | int forced = (hg == -100); | |
2620 | ||
2621 | switch (pcount) { | |
2622 | case 0: | |
2623 | hg = drbd_asb_recover_0p(mdev); | |
2624 | break; | |
2625 | case 1: | |
2626 | hg = drbd_asb_recover_1p(mdev); | |
2627 | break; | |
2628 | case 2: | |
2629 | hg = drbd_asb_recover_2p(mdev); | |
2630 | break; | |
2631 | } | |
2632 | if (abs(hg) < 100) { | |
2633 | dev_warn(DEV, "Split-Brain detected, %d primaries, " | |
2634 | "automatically solved. Sync from %s node\n", | |
2635 | pcount, (hg < 0) ? "peer" : "this"); | |
2636 | if (forced) { | |
2637 | dev_warn(DEV, "Doing a full sync, since" | |
2638 | " UUIDs where ambiguous.\n"); | |
2639 | hg = hg*2; | |
2640 | } | |
2641 | } | |
2642 | } | |
2643 | ||
2644 | if (hg == -100) { | |
89e58e75 | 2645 | if (mdev->tconn->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 | 2646 | hg = -1; |
89e58e75 | 2647 | if (!mdev->tconn->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 PR |
2648 | hg = 1; |
2649 | ||
2650 | if (abs(hg) < 100) | |
2651 | dev_warn(DEV, "Split-Brain detected, manually solved. " | |
2652 | "Sync from %s node\n", | |
2653 | (hg < 0) ? "peer" : "this"); | |
2654 | } | |
2655 | ||
2656 | if (hg == -100) { | |
580b9767 LE |
2657 | /* FIXME this log message is not correct if we end up here |
2658 | * after an attempted attach on a diskless node. | |
2659 | * We just refuse to attach -- well, we drop the "connection" | |
2660 | * to that disk, in a way... */ | |
3a11a487 | 2661 | dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); |
b411b363 PR |
2662 | drbd_khelper(mdev, "split-brain"); |
2663 | return C_MASK; | |
2664 | } | |
2665 | ||
2666 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
2667 | dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); | |
2668 | return C_MASK; | |
2669 | } | |
2670 | ||
2671 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
2672 | mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { | |
89e58e75 | 2673 | switch (mdev->tconn->net_conf->rr_conflict) { |
b411b363 PR |
2674 | case ASB_CALL_HELPER: |
2675 | drbd_khelper(mdev, "pri-lost"); | |
2676 | /* fall through */ | |
2677 | case ASB_DISCONNECT: | |
2678 | dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); | |
2679 | return C_MASK; | |
2680 | case ASB_VIOLENTLY: | |
2681 | dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" | |
2682 | "assumption\n"); | |
2683 | } | |
2684 | } | |
2685 | ||
89e58e75 | 2686 | if (mdev->tconn->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) { |
cf14c2e9 PR |
2687 | if (hg == 0) |
2688 | dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); | |
2689 | else | |
2690 | dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", | |
2691 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), | |
2692 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
2693 | return C_MASK; | |
2694 | } | |
2695 | ||
b411b363 PR |
2696 | if (abs(hg) >= 2) { |
2697 | dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); | |
20ceb2b2 LE |
2698 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
2699 | BM_LOCKED_SET_ALLOWED)) | |
b411b363 PR |
2700 | return C_MASK; |
2701 | } | |
2702 | ||
2703 | if (hg > 0) { /* become sync source. */ | |
2704 | rv = C_WF_BITMAP_S; | |
2705 | } else if (hg < 0) { /* become sync target */ | |
2706 | rv = C_WF_BITMAP_T; | |
2707 | } else { | |
2708 | rv = C_CONNECTED; | |
2709 | if (drbd_bm_total_weight(mdev)) { | |
2710 | dev_info(DEV, "No resync, but %lu bits in bitmap!\n", | |
2711 | drbd_bm_total_weight(mdev)); | |
2712 | } | |
2713 | } | |
2714 | ||
2715 | return rv; | |
2716 | } | |
2717 | ||
2718 | /* returns 1 if invalid */ | |
2719 | static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) | |
2720 | { | |
2721 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
2722 | if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || | |
2723 | (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) | |
2724 | return 0; | |
2725 | ||
2726 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
2727 | if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || | |
2728 | self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) | |
2729 | return 1; | |
2730 | ||
2731 | /* everything else is valid if they are equal on both sides. */ | |
2732 | if (peer == self) | |
2733 | return 0; | |
2734 | ||
2735 | /* everything es is invalid. */ | |
2736 | return 1; | |
2737 | } | |
2738 | ||
d8763023 AG |
2739 | static int receive_protocol(struct drbd_conf *mdev, enum drbd_packet cmd, |
2740 | unsigned int data_size) | |
b411b363 | 2741 | { |
e42325a5 | 2742 | struct p_protocol *p = &mdev->tconn->data.rbuf.protocol; |
b411b363 | 2743 | int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
cf14c2e9 | 2744 | int p_want_lose, p_two_primaries, cf; |
b411b363 PR |
2745 | char p_integrity_alg[SHARED_SECRET_MAX] = ""; |
2746 | ||
b411b363 PR |
2747 | p_proto = be32_to_cpu(p->protocol); |
2748 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
2749 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
2750 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 2751 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 PR |
2752 | cf = be32_to_cpu(p->conn_flags); |
2753 | p_want_lose = cf & CF_WANT_LOSE; | |
2754 | ||
2755 | clear_bit(CONN_DRY_RUN, &mdev->flags); | |
2756 | ||
2757 | if (cf & CF_DRY_RUN) | |
2758 | set_bit(CONN_DRY_RUN, &mdev->flags); | |
b411b363 | 2759 | |
89e58e75 | 2760 | if (p_proto != mdev->tconn->net_conf->wire_protocol) { |
b411b363 PR |
2761 | dev_err(DEV, "incompatible communication protocols\n"); |
2762 | goto disconnect; | |
2763 | } | |
2764 | ||
89e58e75 | 2765 | if (cmp_after_sb(p_after_sb_0p, mdev->tconn->net_conf->after_sb_0p)) { |
b411b363 PR |
2766 | dev_err(DEV, "incompatible after-sb-0pri settings\n"); |
2767 | goto disconnect; | |
2768 | } | |
2769 | ||
89e58e75 | 2770 | if (cmp_after_sb(p_after_sb_1p, mdev->tconn->net_conf->after_sb_1p)) { |
b411b363 PR |
2771 | dev_err(DEV, "incompatible after-sb-1pri settings\n"); |
2772 | goto disconnect; | |
2773 | } | |
2774 | ||
89e58e75 | 2775 | if (cmp_after_sb(p_after_sb_2p, mdev->tconn->net_conf->after_sb_2p)) { |
b411b363 PR |
2776 | dev_err(DEV, "incompatible after-sb-2pri settings\n"); |
2777 | goto disconnect; | |
2778 | } | |
2779 | ||
89e58e75 | 2780 | if (p_want_lose && mdev->tconn->net_conf->want_lose) { |
b411b363 PR |
2781 | dev_err(DEV, "both sides have the 'want_lose' flag set\n"); |
2782 | goto disconnect; | |
2783 | } | |
2784 | ||
89e58e75 | 2785 | if (p_two_primaries != mdev->tconn->net_conf->two_primaries) { |
b411b363 PR |
2786 | dev_err(DEV, "incompatible setting of the two-primaries options\n"); |
2787 | goto disconnect; | |
2788 | } | |
2789 | ||
31890f4a | 2790 | if (mdev->tconn->agreed_pro_version >= 87) { |
89e58e75 | 2791 | unsigned char *my_alg = mdev->tconn->net_conf->integrity_alg; |
b411b363 PR |
2792 | |
2793 | if (drbd_recv(mdev, p_integrity_alg, data_size) != data_size) | |
81e84650 | 2794 | return false; |
b411b363 PR |
2795 | |
2796 | p_integrity_alg[SHARED_SECRET_MAX-1] = 0; | |
2797 | if (strcmp(p_integrity_alg, my_alg)) { | |
2798 | dev_err(DEV, "incompatible setting of the data-integrity-alg\n"); | |
2799 | goto disconnect; | |
2800 | } | |
2801 | dev_info(DEV, "data-integrity-alg: %s\n", | |
2802 | my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); | |
2803 | } | |
2804 | ||
81e84650 | 2805 | return true; |
b411b363 PR |
2806 | |
2807 | disconnect: | |
2808 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 2809 | return false; |
b411b363 PR |
2810 | } |
2811 | ||
2812 | /* helper function | |
2813 | * input: alg name, feature name | |
2814 | * return: NULL (alg name was "") | |
2815 | * ERR_PTR(error) if something goes wrong | |
2816 | * or the crypto hash ptr, if it worked out ok. */ | |
2817 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, | |
2818 | const char *alg, const char *name) | |
2819 | { | |
2820 | struct crypto_hash *tfm; | |
2821 | ||
2822 | if (!alg[0]) | |
2823 | return NULL; | |
2824 | ||
2825 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
2826 | if (IS_ERR(tfm)) { | |
2827 | dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", | |
2828 | alg, name, PTR_ERR(tfm)); | |
2829 | return tfm; | |
2830 | } | |
2831 | if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { | |
2832 | crypto_free_hash(tfm); | |
2833 | dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); | |
2834 | return ERR_PTR(-EINVAL); | |
2835 | } | |
2836 | return tfm; | |
2837 | } | |
2838 | ||
d8763023 AG |
2839 | static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packet cmd, |
2840 | unsigned int packet_size) | |
b411b363 | 2841 | { |
81e84650 | 2842 | int ok = true; |
e42325a5 | 2843 | struct p_rs_param_95 *p = &mdev->tconn->data.rbuf.rs_param_95; |
b411b363 PR |
2844 | unsigned int header_size, data_size, exp_max_sz; |
2845 | struct crypto_hash *verify_tfm = NULL; | |
2846 | struct crypto_hash *csums_tfm = NULL; | |
31890f4a | 2847 | const int apv = mdev->tconn->agreed_pro_version; |
778f271d PR |
2848 | int *rs_plan_s = NULL; |
2849 | int fifo_size = 0; | |
b411b363 PR |
2850 | |
2851 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
2852 | : apv == 88 ? sizeof(struct p_rs_param) | |
2853 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
2854 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
2855 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 2856 | |
02918be2 | 2857 | if (packet_size > exp_max_sz) { |
b411b363 | 2858 | dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
02918be2 | 2859 | packet_size, exp_max_sz); |
81e84650 | 2860 | return false; |
b411b363 PR |
2861 | } |
2862 | ||
2863 | if (apv <= 88) { | |
257d0af6 | 2864 | header_size = sizeof(struct p_rs_param) - sizeof(struct p_header); |
02918be2 | 2865 | data_size = packet_size - header_size; |
8e26f9cc | 2866 | } else if (apv <= 94) { |
257d0af6 | 2867 | header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header); |
02918be2 | 2868 | data_size = packet_size - header_size; |
b411b363 | 2869 | D_ASSERT(data_size == 0); |
8e26f9cc | 2870 | } else { |
257d0af6 | 2871 | header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header); |
02918be2 | 2872 | data_size = packet_size - header_size; |
b411b363 PR |
2873 | D_ASSERT(data_size == 0); |
2874 | } | |
2875 | ||
2876 | /* initialize verify_alg and csums_alg */ | |
2877 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
2878 | ||
02918be2 | 2879 | if (drbd_recv(mdev, &p->head.payload, header_size) != header_size) |
81e84650 | 2880 | return false; |
b411b363 PR |
2881 | |
2882 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
2883 | ||
2884 | if (apv >= 88) { | |
2885 | if (apv == 88) { | |
2886 | if (data_size > SHARED_SECRET_MAX) { | |
2887 | dev_err(DEV, "verify-alg too long, " | |
2888 | "peer wants %u, accepting only %u byte\n", | |
2889 | data_size, SHARED_SECRET_MAX); | |
81e84650 | 2890 | return false; |
b411b363 PR |
2891 | } |
2892 | ||
2893 | if (drbd_recv(mdev, p->verify_alg, data_size) != data_size) | |
81e84650 | 2894 | return false; |
b411b363 PR |
2895 | |
2896 | /* we expect NUL terminated string */ | |
2897 | /* but just in case someone tries to be evil */ | |
2898 | D_ASSERT(p->verify_alg[data_size-1] == 0); | |
2899 | p->verify_alg[data_size-1] = 0; | |
2900 | ||
2901 | } else /* apv >= 89 */ { | |
2902 | /* we still expect NUL terminated strings */ | |
2903 | /* but just in case someone tries to be evil */ | |
2904 | D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); | |
2905 | D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
2906 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; | |
2907 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
2908 | } | |
2909 | ||
2910 | if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) { | |
2911 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
2912 | dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2913 | mdev->sync_conf.verify_alg, p->verify_alg); | |
2914 | goto disconnect; | |
2915 | } | |
2916 | verify_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
2917 | p->verify_alg, "verify-alg"); | |
2918 | if (IS_ERR(verify_tfm)) { | |
2919 | verify_tfm = NULL; | |
2920 | goto disconnect; | |
2921 | } | |
2922 | } | |
2923 | ||
2924 | if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) { | |
2925 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
2926 | dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", | |
2927 | mdev->sync_conf.csums_alg, p->csums_alg); | |
2928 | goto disconnect; | |
2929 | } | |
2930 | csums_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
2931 | p->csums_alg, "csums-alg"); | |
2932 | if (IS_ERR(csums_tfm)) { | |
2933 | csums_tfm = NULL; | |
2934 | goto disconnect; | |
2935 | } | |
2936 | } | |
2937 | ||
8e26f9cc PR |
2938 | if (apv > 94) { |
2939 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
2940 | mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead); | |
2941 | mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target); | |
2942 | mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target); | |
2943 | mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d PR |
2944 | |
2945 | fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; | |
2946 | if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { | |
2947 | rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); | |
2948 | if (!rs_plan_s) { | |
2949 | dev_err(DEV, "kmalloc of fifo_buffer failed"); | |
2950 | goto disconnect; | |
2951 | } | |
2952 | } | |
8e26f9cc | 2953 | } |
b411b363 PR |
2954 | |
2955 | spin_lock(&mdev->peer_seq_lock); | |
2956 | /* lock against drbd_nl_syncer_conf() */ | |
2957 | if (verify_tfm) { | |
2958 | strcpy(mdev->sync_conf.verify_alg, p->verify_alg); | |
2959 | mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1; | |
2960 | crypto_free_hash(mdev->verify_tfm); | |
2961 | mdev->verify_tfm = verify_tfm; | |
2962 | dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); | |
2963 | } | |
2964 | if (csums_tfm) { | |
2965 | strcpy(mdev->sync_conf.csums_alg, p->csums_alg); | |
2966 | mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1; | |
2967 | crypto_free_hash(mdev->csums_tfm); | |
2968 | mdev->csums_tfm = csums_tfm; | |
2969 | dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); | |
2970 | } | |
778f271d PR |
2971 | if (fifo_size != mdev->rs_plan_s.size) { |
2972 | kfree(mdev->rs_plan_s.values); | |
2973 | mdev->rs_plan_s.values = rs_plan_s; | |
2974 | mdev->rs_plan_s.size = fifo_size; | |
2975 | mdev->rs_planed = 0; | |
2976 | } | |
b411b363 PR |
2977 | spin_unlock(&mdev->peer_seq_lock); |
2978 | } | |
2979 | ||
2980 | return ok; | |
2981 | disconnect: | |
2982 | /* just for completeness: actually not needed, | |
2983 | * as this is not reached if csums_tfm was ok. */ | |
2984 | crypto_free_hash(csums_tfm); | |
2985 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
2986 | crypto_free_hash(verify_tfm); | |
2987 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 2988 | return false; |
b411b363 PR |
2989 | } |
2990 | ||
b411b363 PR |
2991 | /* warn if the arguments differ by more than 12.5% */ |
2992 | static void warn_if_differ_considerably(struct drbd_conf *mdev, | |
2993 | const char *s, sector_t a, sector_t b) | |
2994 | { | |
2995 | sector_t d; | |
2996 | if (a == 0 || b == 0) | |
2997 | return; | |
2998 | d = (a > b) ? (a - b) : (b - a); | |
2999 | if (d > (a>>3) || d > (b>>3)) | |
3000 | dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, | |
3001 | (unsigned long long)a, (unsigned long long)b); | |
3002 | } | |
3003 | ||
d8763023 AG |
3004 | static int receive_sizes(struct drbd_conf *mdev, enum drbd_packet cmd, |
3005 | unsigned int data_size) | |
b411b363 | 3006 | { |
e42325a5 | 3007 | struct p_sizes *p = &mdev->tconn->data.rbuf.sizes; |
b411b363 | 3008 | enum determine_dev_size dd = unchanged; |
b411b363 PR |
3009 | sector_t p_size, p_usize, my_usize; |
3010 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 3011 | enum dds_flags ddsf; |
b411b363 | 3012 | |
b411b363 PR |
3013 | p_size = be64_to_cpu(p->d_size); |
3014 | p_usize = be64_to_cpu(p->u_size); | |
3015 | ||
3016 | if (p_size == 0 && mdev->state.disk == D_DISKLESS) { | |
3017 | dev_err(DEV, "some backing storage is needed\n"); | |
3018 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3019 | return false; |
b411b363 PR |
3020 | } |
3021 | ||
3022 | /* just store the peer's disk size for now. | |
3023 | * we still need to figure out whether we accept that. */ | |
3024 | mdev->p_size = p_size; | |
3025 | ||
b411b363 PR |
3026 | if (get_ldev(mdev)) { |
3027 | warn_if_differ_considerably(mdev, "lower level device sizes", | |
3028 | p_size, drbd_get_max_capacity(mdev->ldev)); | |
3029 | warn_if_differ_considerably(mdev, "user requested size", | |
3030 | p_usize, mdev->ldev->dc.disk_size); | |
3031 | ||
3032 | /* if this is the first connect, or an otherwise expected | |
3033 | * param exchange, choose the minimum */ | |
3034 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
3035 | p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, | |
3036 | p_usize); | |
3037 | ||
3038 | my_usize = mdev->ldev->dc.disk_size; | |
3039 | ||
3040 | if (mdev->ldev->dc.disk_size != p_usize) { | |
3041 | mdev->ldev->dc.disk_size = p_usize; | |
3042 | dev_info(DEV, "Peer sets u_size to %lu sectors\n", | |
3043 | (unsigned long)mdev->ldev->dc.disk_size); | |
3044 | } | |
3045 | ||
3046 | /* Never shrink a device with usable data during connect. | |
3047 | But allow online shrinking if we are connected. */ | |
a393db6f | 3048 | if (drbd_new_dev_size(mdev, mdev->ldev, 0) < |
b411b363 PR |
3049 | drbd_get_capacity(mdev->this_bdev) && |
3050 | mdev->state.disk >= D_OUTDATED && | |
3051 | mdev->state.conn < C_CONNECTED) { | |
3052 | dev_err(DEV, "The peer's disk size is too small!\n"); | |
3053 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
3054 | mdev->ldev->dc.disk_size = my_usize; | |
3055 | put_ldev(mdev); | |
81e84650 | 3056 | return false; |
b411b363 PR |
3057 | } |
3058 | put_ldev(mdev); | |
3059 | } | |
b411b363 | 3060 | |
e89b591c | 3061 | ddsf = be16_to_cpu(p->dds_flags); |
b411b363 | 3062 | if (get_ldev(mdev)) { |
24c4830c | 3063 | dd = drbd_determine_dev_size(mdev, ddsf); |
b411b363 PR |
3064 | put_ldev(mdev); |
3065 | if (dd == dev_size_error) | |
81e84650 | 3066 | return false; |
b411b363 PR |
3067 | drbd_md_sync(mdev); |
3068 | } else { | |
3069 | /* I am diskless, need to accept the peer's size. */ | |
3070 | drbd_set_my_capacity(mdev, p_size); | |
3071 | } | |
3072 | ||
99432fcc PR |
3073 | mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
3074 | drbd_reconsider_max_bio_size(mdev); | |
3075 | ||
b411b363 PR |
3076 | if (get_ldev(mdev)) { |
3077 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { | |
3078 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); | |
3079 | ldsc = 1; | |
3080 | } | |
3081 | ||
b411b363 PR |
3082 | put_ldev(mdev); |
3083 | } | |
3084 | ||
3085 | if (mdev->state.conn > C_WF_REPORT_PARAMS) { | |
3086 | if (be64_to_cpu(p->c_size) != | |
3087 | drbd_get_capacity(mdev->this_bdev) || ldsc) { | |
3088 | /* we have different sizes, probably peer | |
3089 | * needs to know my new size... */ | |
e89b591c | 3090 | drbd_send_sizes(mdev, 0, ddsf); |
b411b363 PR |
3091 | } |
3092 | if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || | |
3093 | (dd == grew && mdev->state.conn == C_CONNECTED)) { | |
3094 | if (mdev->state.pdsk >= D_INCONSISTENT && | |
e89b591c PR |
3095 | mdev->state.disk >= D_INCONSISTENT) { |
3096 | if (ddsf & DDSF_NO_RESYNC) | |
3097 | dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); | |
3098 | else | |
3099 | resync_after_online_grow(mdev); | |
3100 | } else | |
b411b363 PR |
3101 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); |
3102 | } | |
3103 | } | |
3104 | ||
81e84650 | 3105 | return true; |
b411b363 PR |
3106 | } |
3107 | ||
d8763023 AG |
3108 | static int receive_uuids(struct drbd_conf *mdev, enum drbd_packet cmd, |
3109 | unsigned int data_size) | |
b411b363 | 3110 | { |
e42325a5 | 3111 | struct p_uuids *p = &mdev->tconn->data.rbuf.uuids; |
b411b363 | 3112 | u64 *p_uuid; |
62b0da3a | 3113 | int i, updated_uuids = 0; |
b411b363 | 3114 | |
b411b363 PR |
3115 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
3116 | ||
3117 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3118 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3119 | ||
3120 | kfree(mdev->p_uuid); | |
3121 | mdev->p_uuid = p_uuid; | |
3122 | ||
3123 | if (mdev->state.conn < C_CONNECTED && | |
3124 | mdev->state.disk < D_INCONSISTENT && | |
3125 | mdev->state.role == R_PRIMARY && | |
3126 | (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
3127 | dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", | |
3128 | (unsigned long long)mdev->ed_uuid); | |
3129 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3130 | return false; |
b411b363 PR |
3131 | } |
3132 | ||
3133 | if (get_ldev(mdev)) { | |
3134 | int skip_initial_sync = | |
3135 | mdev->state.conn == C_CONNECTED && | |
31890f4a | 3136 | mdev->tconn->agreed_pro_version >= 90 && |
b411b363 PR |
3137 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && |
3138 | (p_uuid[UI_FLAGS] & 8); | |
3139 | if (skip_initial_sync) { | |
3140 | dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); | |
3141 | drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, | |
20ceb2b2 LE |
3142 | "clear_n_write from receive_uuids", |
3143 | BM_LOCKED_TEST_ALLOWED); | |
b411b363 PR |
3144 | _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); |
3145 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
3146 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
3147 | CS_VERBOSE, NULL); | |
3148 | drbd_md_sync(mdev); | |
62b0da3a | 3149 | updated_uuids = 1; |
b411b363 PR |
3150 | } |
3151 | put_ldev(mdev); | |
18a50fa2 PR |
3152 | } else if (mdev->state.disk < D_INCONSISTENT && |
3153 | mdev->state.role == R_PRIMARY) { | |
3154 | /* I am a diskless primary, the peer just created a new current UUID | |
3155 | for me. */ | |
62b0da3a | 3156 | updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
b411b363 PR |
3157 | } |
3158 | ||
3159 | /* Before we test for the disk state, we should wait until an eventually | |
3160 | ongoing cluster wide state change is finished. That is important if | |
3161 | we are primary and are detaching from our disk. We need to see the | |
3162 | new disk state... */ | |
3163 | wait_event(mdev->misc_wait, !test_bit(CLUSTER_ST_CHANGE, &mdev->flags)); | |
3164 | if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) | |
62b0da3a LE |
3165 | updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
3166 | ||
3167 | if (updated_uuids) | |
3168 | drbd_print_uuids(mdev, "receiver updated UUIDs to"); | |
b411b363 | 3169 | |
81e84650 | 3170 | return true; |
b411b363 PR |
3171 | } |
3172 | ||
3173 | /** | |
3174 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3175 | * @ps: The state as seen by the peer. | |
3176 | */ | |
3177 | static union drbd_state convert_state(union drbd_state ps) | |
3178 | { | |
3179 | union drbd_state ms; | |
3180 | ||
3181 | static enum drbd_conns c_tab[] = { | |
3182 | [C_CONNECTED] = C_CONNECTED, | |
3183 | ||
3184 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3185 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3186 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3187 | [C_VERIFY_S] = C_VERIFY_T, | |
3188 | [C_MASK] = C_MASK, | |
3189 | }; | |
3190 | ||
3191 | ms.i = ps.i; | |
3192 | ||
3193 | ms.conn = c_tab[ps.conn]; | |
3194 | ms.peer = ps.role; | |
3195 | ms.role = ps.peer; | |
3196 | ms.pdsk = ps.disk; | |
3197 | ms.disk = ps.pdsk; | |
3198 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3199 | ||
3200 | return ms; | |
3201 | } | |
3202 | ||
d8763023 AG |
3203 | static int receive_req_state(struct drbd_conf *mdev, enum drbd_packet cmd, |
3204 | unsigned int data_size) | |
b411b363 | 3205 | { |
e42325a5 | 3206 | struct p_req_state *p = &mdev->tconn->data.rbuf.req_state; |
b411b363 | 3207 | union drbd_state mask, val; |
bf885f8a | 3208 | enum drbd_state_rv rv; |
b411b363 | 3209 | |
b411b363 PR |
3210 | mask.i = be32_to_cpu(p->mask); |
3211 | val.i = be32_to_cpu(p->val); | |
3212 | ||
3213 | if (test_bit(DISCARD_CONCURRENT, &mdev->flags) && | |
3214 | test_bit(CLUSTER_ST_CHANGE, &mdev->flags)) { | |
3215 | drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); | |
81e84650 | 3216 | return true; |
b411b363 PR |
3217 | } |
3218 | ||
3219 | mask = convert_state(mask); | |
3220 | val = convert_state(val); | |
3221 | ||
3222 | rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); | |
3223 | ||
3224 | drbd_send_sr_reply(mdev, rv); | |
3225 | drbd_md_sync(mdev); | |
3226 | ||
81e84650 | 3227 | return true; |
b411b363 PR |
3228 | } |
3229 | ||
d8763023 AG |
3230 | static int receive_state(struct drbd_conf *mdev, enum drbd_packet cmd, |
3231 | unsigned int data_size) | |
b411b363 | 3232 | { |
e42325a5 | 3233 | struct p_state *p = &mdev->tconn->data.rbuf.state; |
4ac4aada | 3234 | union drbd_state os, ns, peer_state; |
b411b363 | 3235 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 3236 | enum chg_state_flags cs_flags; |
b411b363 PR |
3237 | int rv; |
3238 | ||
b411b363 PR |
3239 | peer_state.i = be32_to_cpu(p->state); |
3240 | ||
3241 | real_peer_disk = peer_state.disk; | |
3242 | if (peer_state.disk == D_NEGOTIATING) { | |
3243 | real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; | |
3244 | dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); | |
3245 | } | |
3246 | ||
87eeee41 | 3247 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 3248 | retry: |
4ac4aada | 3249 | os = ns = mdev->state; |
87eeee41 | 3250 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 3251 | |
e9ef7bb6 LE |
3252 | /* peer says his disk is uptodate, while we think it is inconsistent, |
3253 | * and this happens while we think we have a sync going on. */ | |
3254 | if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE && | |
3255 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { | |
3256 | /* If we are (becoming) SyncSource, but peer is still in sync | |
3257 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
3258 | * will change to inconsistent once the peer reaches active | |
3259 | * syncing states. | |
3260 | * It may have changed syncer-paused flags, however, so we | |
3261 | * cannot ignore this completely. */ | |
3262 | if (peer_state.conn > C_CONNECTED && | |
3263 | peer_state.conn < C_SYNC_SOURCE) | |
3264 | real_peer_disk = D_INCONSISTENT; | |
3265 | ||
3266 | /* if peer_state changes to connected at the same time, | |
3267 | * it explicitly notifies us that it finished resync. | |
3268 | * Maybe we should finish it up, too? */ | |
3269 | else if (os.conn >= C_SYNC_SOURCE && | |
3270 | peer_state.conn == C_CONNECTED) { | |
3271 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) | |
3272 | drbd_resync_finished(mdev); | |
81e84650 | 3273 | return true; |
e9ef7bb6 LE |
3274 | } |
3275 | } | |
3276 | ||
3277 | /* peer says his disk is inconsistent, while we think it is uptodate, | |
3278 | * and this happens while the peer still thinks we have a sync going on, | |
3279 | * but we think we are already done with the sync. | |
3280 | * We ignore this to avoid flapping pdsk. | |
3281 | * This should not happen, if the peer is a recent version of drbd. */ | |
3282 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
3283 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
3284 | real_peer_disk = D_UP_TO_DATE; | |
3285 | ||
4ac4aada LE |
3286 | if (ns.conn == C_WF_REPORT_PARAMS) |
3287 | ns.conn = C_CONNECTED; | |
b411b363 | 3288 | |
67531718 PR |
3289 | if (peer_state.conn == C_AHEAD) |
3290 | ns.conn = C_BEHIND; | |
3291 | ||
b411b363 PR |
3292 | if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && |
3293 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3294 | int cr; /* consider resync */ | |
3295 | ||
3296 | /* if we established a new connection */ | |
4ac4aada | 3297 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
3298 | /* if we had an established connection |
3299 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 3300 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 3301 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 3302 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
3303 | /* if we have both been inconsistent, and the peer has been |
3304 | * forced to be UpToDate with --overwrite-data */ | |
3305 | cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); | |
3306 | /* if we had been plain connected, and the admin requested to | |
3307 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 3308 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
3309 | (peer_state.conn >= C_STARTING_SYNC_S && |
3310 | peer_state.conn <= C_WF_BITMAP_T)); | |
3311 | ||
3312 | if (cr) | |
4ac4aada | 3313 | ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); |
b411b363 PR |
3314 | |
3315 | put_ldev(mdev); | |
4ac4aada LE |
3316 | if (ns.conn == C_MASK) { |
3317 | ns.conn = C_CONNECTED; | |
b411b363 | 3318 | if (mdev->state.disk == D_NEGOTIATING) { |
82f59cc6 | 3319 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
b411b363 PR |
3320 | } else if (peer_state.disk == D_NEGOTIATING) { |
3321 | dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); | |
3322 | peer_state.disk = D_DISKLESS; | |
580b9767 | 3323 | real_peer_disk = D_DISKLESS; |
b411b363 | 3324 | } else { |
cf14c2e9 | 3325 | if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags)) |
81e84650 | 3326 | return false; |
4ac4aada | 3327 | D_ASSERT(os.conn == C_WF_REPORT_PARAMS); |
b411b363 | 3328 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); |
81e84650 | 3329 | return false; |
b411b363 PR |
3330 | } |
3331 | } | |
3332 | } | |
3333 | ||
87eeee41 | 3334 | spin_lock_irq(&mdev->tconn->req_lock); |
4ac4aada | 3335 | if (mdev->state.i != os.i) |
b411b363 PR |
3336 | goto retry; |
3337 | clear_bit(CONSIDER_RESYNC, &mdev->flags); | |
b411b363 PR |
3338 | ns.peer = peer_state.role; |
3339 | ns.pdsk = real_peer_disk; | |
3340 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 3341 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b411b363 | 3342 | ns.disk = mdev->new_state_tmp.disk; |
4ac4aada LE |
3343 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
3344 | if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && | |
481c6f50 | 3345 | test_bit(NEW_CUR_UUID, &mdev->flags)) { |
8554df1c | 3346 | /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this |
481c6f50 | 3347 | for temporal network outages! */ |
87eeee41 | 3348 | spin_unlock_irq(&mdev->tconn->req_lock); |
481c6f50 PR |
3349 | dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); |
3350 | tl_clear(mdev); | |
3351 | drbd_uuid_new_current(mdev); | |
3352 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
3353 | drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0)); | |
81e84650 | 3354 | return false; |
481c6f50 | 3355 | } |
65d922c3 | 3356 | rv = _drbd_set_state(mdev, ns, cs_flags, NULL); |
b411b363 | 3357 | ns = mdev->state; |
87eeee41 | 3358 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3359 | |
3360 | if (rv < SS_SUCCESS) { | |
3361 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3362 | return false; |
b411b363 PR |
3363 | } |
3364 | ||
4ac4aada LE |
3365 | if (os.conn > C_WF_REPORT_PARAMS) { |
3366 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
3367 | peer_state.disk != D_NEGOTIATING ) { |
3368 | /* we want resync, peer has not yet decided to sync... */ | |
3369 | /* Nowadays only used when forcing a node into primary role and | |
3370 | setting its disk to UpToDate with that */ | |
3371 | drbd_send_uuids(mdev); | |
3372 | drbd_send_state(mdev); | |
3373 | } | |
3374 | } | |
3375 | ||
89e58e75 | 3376 | mdev->tconn->net_conf->want_lose = 0; |
b411b363 PR |
3377 | |
3378 | drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ | |
3379 | ||
81e84650 | 3380 | return true; |
b411b363 PR |
3381 | } |
3382 | ||
d8763023 AG |
3383 | static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packet cmd, |
3384 | unsigned int data_size) | |
b411b363 | 3385 | { |
e42325a5 | 3386 | struct p_rs_uuid *p = &mdev->tconn->data.rbuf.rs_uuid; |
b411b363 PR |
3387 | |
3388 | wait_event(mdev->misc_wait, | |
3389 | mdev->state.conn == C_WF_SYNC_UUID || | |
c4752ef1 | 3390 | mdev->state.conn == C_BEHIND || |
b411b363 PR |
3391 | mdev->state.conn < C_CONNECTED || |
3392 | mdev->state.disk < D_NEGOTIATING); | |
3393 | ||
3394 | /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ | |
3395 | ||
b411b363 PR |
3396 | /* Here the _drbd_uuid_ functions are right, current should |
3397 | _not_ be rotated into the history */ | |
3398 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3399 | _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); | |
3400 | _drbd_uuid_set(mdev, UI_BITMAP, 0UL); | |
3401 | ||
62b0da3a | 3402 | drbd_print_uuids(mdev, "updated sync uuid"); |
b411b363 PR |
3403 | drbd_start_resync(mdev, C_SYNC_TARGET); |
3404 | ||
3405 | put_ldev(mdev); | |
3406 | } else | |
3407 | dev_err(DEV, "Ignoring SyncUUID packet!\n"); | |
3408 | ||
81e84650 | 3409 | return true; |
b411b363 PR |
3410 | } |
3411 | ||
2c46407d AG |
3412 | /** |
3413 | * receive_bitmap_plain | |
3414 | * | |
3415 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3416 | * code upon failure. | |
3417 | */ | |
3418 | static int | |
02918be2 PR |
3419 | receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size, |
3420 | unsigned long *buffer, struct bm_xfer_ctx *c) | |
b411b363 PR |
3421 | { |
3422 | unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); | |
3423 | unsigned want = num_words * sizeof(long); | |
2c46407d | 3424 | int err; |
b411b363 | 3425 | |
02918be2 PR |
3426 | if (want != data_size) { |
3427 | dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size); | |
2c46407d | 3428 | return -EIO; |
b411b363 PR |
3429 | } |
3430 | if (want == 0) | |
2c46407d AG |
3431 | return 0; |
3432 | err = drbd_recv(mdev, buffer, want); | |
3433 | if (err != want) { | |
3434 | if (err >= 0) | |
3435 | err = -EIO; | |
3436 | return err; | |
3437 | } | |
b411b363 PR |
3438 | |
3439 | drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); | |
3440 | ||
3441 | c->word_offset += num_words; | |
3442 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
3443 | if (c->bit_offset > c->bm_bits) | |
3444 | c->bit_offset = c->bm_bits; | |
3445 | ||
2c46407d | 3446 | return 1; |
b411b363 PR |
3447 | } |
3448 | ||
2c46407d AG |
3449 | /** |
3450 | * recv_bm_rle_bits | |
3451 | * | |
3452 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3453 | * code upon failure. | |
3454 | */ | |
3455 | static int | |
b411b363 PR |
3456 | recv_bm_rle_bits(struct drbd_conf *mdev, |
3457 | struct p_compressed_bm *p, | |
c6d25cfe PR |
3458 | struct bm_xfer_ctx *c, |
3459 | unsigned int len) | |
b411b363 PR |
3460 | { |
3461 | struct bitstream bs; | |
3462 | u64 look_ahead; | |
3463 | u64 rl; | |
3464 | u64 tmp; | |
3465 | unsigned long s = c->bit_offset; | |
3466 | unsigned long e; | |
b411b363 PR |
3467 | int toggle = DCBP_get_start(p); |
3468 | int have; | |
3469 | int bits; | |
3470 | ||
3471 | bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p)); | |
3472 | ||
3473 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
3474 | if (bits < 0) | |
2c46407d | 3475 | return -EIO; |
b411b363 PR |
3476 | |
3477 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
3478 | bits = vli_decode_bits(&rl, look_ahead); | |
3479 | if (bits <= 0) | |
2c46407d | 3480 | return -EIO; |
b411b363 PR |
3481 | |
3482 | if (toggle) { | |
3483 | e = s + rl -1; | |
3484 | if (e >= c->bm_bits) { | |
3485 | dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); | |
2c46407d | 3486 | return -EIO; |
b411b363 PR |
3487 | } |
3488 | _drbd_bm_set_bits(mdev, s, e); | |
3489 | } | |
3490 | ||
3491 | if (have < bits) { | |
3492 | dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", | |
3493 | have, bits, look_ahead, | |
3494 | (unsigned int)(bs.cur.b - p->code), | |
3495 | (unsigned int)bs.buf_len); | |
2c46407d | 3496 | return -EIO; |
b411b363 PR |
3497 | } |
3498 | look_ahead >>= bits; | |
3499 | have -= bits; | |
3500 | ||
3501 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
3502 | if (bits < 0) | |
2c46407d | 3503 | return -EIO; |
b411b363 PR |
3504 | look_ahead |= tmp << have; |
3505 | have += bits; | |
3506 | } | |
3507 | ||
3508 | c->bit_offset = s; | |
3509 | bm_xfer_ctx_bit_to_word_offset(c); | |
3510 | ||
2c46407d | 3511 | return (s != c->bm_bits); |
b411b363 PR |
3512 | } |
3513 | ||
2c46407d AG |
3514 | /** |
3515 | * decode_bitmap_c | |
3516 | * | |
3517 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3518 | * code upon failure. | |
3519 | */ | |
3520 | static int | |
b411b363 PR |
3521 | decode_bitmap_c(struct drbd_conf *mdev, |
3522 | struct p_compressed_bm *p, | |
c6d25cfe PR |
3523 | struct bm_xfer_ctx *c, |
3524 | unsigned int len) | |
b411b363 PR |
3525 | { |
3526 | if (DCBP_get_code(p) == RLE_VLI_Bits) | |
c6d25cfe | 3527 | return recv_bm_rle_bits(mdev, p, c, len); |
b411b363 PR |
3528 | |
3529 | /* other variants had been implemented for evaluation, | |
3530 | * but have been dropped as this one turned out to be "best" | |
3531 | * during all our tests. */ | |
3532 | ||
3533 | dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); | |
3534 | drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); | |
2c46407d | 3535 | return -EIO; |
b411b363 PR |
3536 | } |
3537 | ||
3538 | void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
3539 | const char *direction, struct bm_xfer_ctx *c) | |
3540 | { | |
3541 | /* what would it take to transfer it "plaintext" */ | |
c012949a | 3542 | unsigned plain = sizeof(struct p_header) * |
b411b363 PR |
3543 | ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) |
3544 | + c->bm_words * sizeof(long); | |
3545 | unsigned total = c->bytes[0] + c->bytes[1]; | |
3546 | unsigned r; | |
3547 | ||
3548 | /* total can not be zero. but just in case: */ | |
3549 | if (total == 0) | |
3550 | return; | |
3551 | ||
3552 | /* don't report if not compressed */ | |
3553 | if (total >= plain) | |
3554 | return; | |
3555 | ||
3556 | /* total < plain. check for overflow, still */ | |
3557 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
3558 | : (1000 * total / plain); | |
3559 | ||
3560 | if (r > 1000) | |
3561 | r = 1000; | |
3562 | ||
3563 | r = 1000 - r; | |
3564 | dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " | |
3565 | "total %u; compression: %u.%u%%\n", | |
3566 | direction, | |
3567 | c->bytes[1], c->packets[1], | |
3568 | c->bytes[0], c->packets[0], | |
3569 | total, r/10, r % 10); | |
3570 | } | |
3571 | ||
3572 | /* Since we are processing the bitfield from lower addresses to higher, | |
3573 | it does not matter if the process it in 32 bit chunks or 64 bit | |
3574 | chunks as long as it is little endian. (Understand it as byte stream, | |
3575 | beginning with the lowest byte...) If we would use big endian | |
3576 | we would need to process it from the highest address to the lowest, | |
3577 | in order to be agnostic to the 32 vs 64 bits issue. | |
3578 | ||
3579 | returns 0 on failure, 1 if we successfully received it. */ | |
d8763023 AG |
3580 | static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packet cmd, |
3581 | unsigned int data_size) | |
b411b363 PR |
3582 | { |
3583 | struct bm_xfer_ctx c; | |
3584 | void *buffer; | |
2c46407d | 3585 | int err; |
81e84650 | 3586 | int ok = false; |
257d0af6 | 3587 | struct p_header *h = &mdev->tconn->data.rbuf.header; |
b411b363 | 3588 | |
20ceb2b2 LE |
3589 | drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
3590 | /* you are supposed to send additional out-of-sync information | |
3591 | * if you actually set bits during this phase */ | |
b411b363 PR |
3592 | |
3593 | /* maybe we should use some per thread scratch page, | |
3594 | * and allocate that during initial device creation? */ | |
3595 | buffer = (unsigned long *) __get_free_page(GFP_NOIO); | |
3596 | if (!buffer) { | |
3597 | dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); | |
3598 | goto out; | |
3599 | } | |
3600 | ||
3601 | c = (struct bm_xfer_ctx) { | |
3602 | .bm_bits = drbd_bm_bits(mdev), | |
3603 | .bm_words = drbd_bm_words(mdev), | |
3604 | }; | |
3605 | ||
2c46407d | 3606 | for(;;) { |
02918be2 | 3607 | if (cmd == P_BITMAP) { |
2c46407d | 3608 | err = receive_bitmap_plain(mdev, data_size, buffer, &c); |
02918be2 | 3609 | } else if (cmd == P_COMPRESSED_BITMAP) { |
b411b363 PR |
3610 | /* MAYBE: sanity check that we speak proto >= 90, |
3611 | * and the feature is enabled! */ | |
3612 | struct p_compressed_bm *p; | |
3613 | ||
02918be2 | 3614 | if (data_size > BM_PACKET_PAYLOAD_BYTES) { |
b411b363 PR |
3615 | dev_err(DEV, "ReportCBitmap packet too large\n"); |
3616 | goto out; | |
3617 | } | |
3618 | /* use the page buff */ | |
3619 | p = buffer; | |
3620 | memcpy(p, h, sizeof(*h)); | |
02918be2 | 3621 | if (drbd_recv(mdev, p->head.payload, data_size) != data_size) |
b411b363 | 3622 | goto out; |
004352fa LE |
3623 | if (data_size <= (sizeof(*p) - sizeof(p->head))) { |
3624 | dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size); | |
78fcbdae | 3625 | goto out; |
b411b363 | 3626 | } |
c6d25cfe | 3627 | err = decode_bitmap_c(mdev, p, &c, data_size); |
b411b363 | 3628 | } else { |
02918be2 | 3629 | dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd); |
b411b363 PR |
3630 | goto out; |
3631 | } | |
3632 | ||
02918be2 | 3633 | c.packets[cmd == P_BITMAP]++; |
257d0af6 | 3634 | c.bytes[cmd == P_BITMAP] += sizeof(struct p_header) + data_size; |
b411b363 | 3635 | |
2c46407d AG |
3636 | if (err <= 0) { |
3637 | if (err < 0) | |
3638 | goto out; | |
b411b363 | 3639 | break; |
2c46407d | 3640 | } |
02918be2 | 3641 | if (!drbd_recv_header(mdev, &cmd, &data_size)) |
b411b363 | 3642 | goto out; |
2c46407d | 3643 | } |
b411b363 PR |
3644 | |
3645 | INFO_bm_xfer_stats(mdev, "receive", &c); | |
3646 | ||
3647 | if (mdev->state.conn == C_WF_BITMAP_T) { | |
de1f8e4a AG |
3648 | enum drbd_state_rv rv; |
3649 | ||
b411b363 PR |
3650 | ok = !drbd_send_bitmap(mdev); |
3651 | if (!ok) | |
3652 | goto out; | |
3653 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
de1f8e4a AG |
3654 | rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
3655 | D_ASSERT(rv == SS_SUCCESS); | |
b411b363 PR |
3656 | } else if (mdev->state.conn != C_WF_BITMAP_S) { |
3657 | /* admin may have requested C_DISCONNECTING, | |
3658 | * other threads may have noticed network errors */ | |
3659 | dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", | |
3660 | drbd_conn_str(mdev->state.conn)); | |
3661 | } | |
3662 | ||
81e84650 | 3663 | ok = true; |
b411b363 | 3664 | out: |
20ceb2b2 | 3665 | drbd_bm_unlock(mdev); |
b411b363 PR |
3666 | if (ok && mdev->state.conn == C_WF_BITMAP_S) |
3667 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
3668 | free_page((unsigned long) buffer); | |
3669 | return ok; | |
3670 | } | |
3671 | ||
d8763023 AG |
3672 | static int receive_skip(struct drbd_conf *mdev, enum drbd_packet cmd, |
3673 | unsigned int data_size) | |
b411b363 PR |
3674 | { |
3675 | /* TODO zero copy sink :) */ | |
3676 | static char sink[128]; | |
3677 | int size, want, r; | |
3678 | ||
02918be2 PR |
3679 | dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n", |
3680 | cmd, data_size); | |
b411b363 | 3681 | |
02918be2 | 3682 | size = data_size; |
b411b363 PR |
3683 | while (size > 0) { |
3684 | want = min_t(int, size, sizeof(sink)); | |
3685 | r = drbd_recv(mdev, sink, want); | |
841ce241 AG |
3686 | if (!expect(r > 0)) |
3687 | break; | |
b411b363 PR |
3688 | size -= r; |
3689 | } | |
3690 | return size == 0; | |
3691 | } | |
3692 | ||
d8763023 AG |
3693 | static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packet cmd, |
3694 | unsigned int data_size) | |
0ced55a3 | 3695 | { |
e7f52dfb LE |
3696 | /* Make sure we've acked all the TCP data associated |
3697 | * with the data requests being unplugged */ | |
e42325a5 | 3698 | drbd_tcp_quickack(mdev->tconn->data.socket); |
0ced55a3 | 3699 | |
81e84650 | 3700 | return true; |
0ced55a3 PR |
3701 | } |
3702 | ||
d8763023 AG |
3703 | static int receive_out_of_sync(struct drbd_conf *mdev, enum drbd_packet cmd, |
3704 | unsigned int data_size) | |
73a01a18 | 3705 | { |
e42325a5 | 3706 | struct p_block_desc *p = &mdev->tconn->data.rbuf.block_desc; |
73a01a18 | 3707 | |
f735e363 LE |
3708 | switch (mdev->state.conn) { |
3709 | case C_WF_SYNC_UUID: | |
3710 | case C_WF_BITMAP_T: | |
3711 | case C_BEHIND: | |
3712 | break; | |
3713 | default: | |
3714 | dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", | |
3715 | drbd_conn_str(mdev->state.conn)); | |
3716 | } | |
3717 | ||
73a01a18 PR |
3718 | drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
3719 | ||
81e84650 | 3720 | return true; |
73a01a18 PR |
3721 | } |
3722 | ||
d8763023 AG |
3723 | typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packet cmd, |
3724 | unsigned int to_receive); | |
02918be2 PR |
3725 | |
3726 | struct data_cmd { | |
3727 | int expect_payload; | |
3728 | size_t pkt_size; | |
3729 | drbd_cmd_handler_f function; | |
3730 | }; | |
3731 | ||
3732 | static struct data_cmd drbd_cmd_handler[] = { | |
3733 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
3734 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
3735 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
3736 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
257d0af6 PR |
3737 | [P_BITMAP] = { 1, sizeof(struct p_header), receive_bitmap } , |
3738 | [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header), receive_bitmap } , | |
3739 | [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header), receive_UnplugRemote }, | |
02918be2 PR |
3740 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, |
3741 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
257d0af6 PR |
3742 | [P_SYNC_PARAM] = { 1, sizeof(struct p_header), receive_SyncParam }, |
3743 | [P_SYNC_PARAM89] = { 1, sizeof(struct p_header), receive_SyncParam }, | |
02918be2 PR |
3744 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, |
3745 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
3746 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
3747 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
3748 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
3749 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
3750 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3751 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3752 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3753 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
73a01a18 | 3754 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, |
b411b363 PR |
3755 | /* anything missing from this table is in |
3756 | * the asender_tbl, see get_asender_cmd */ | |
02918be2 | 3757 | [P_MAX_CMD] = { 0, 0, NULL }, |
b411b363 PR |
3758 | }; |
3759 | ||
02918be2 | 3760 | /* All handler functions that expect a sub-header get that sub-heder in |
e42325a5 | 3761 | mdev->tconn->data.rbuf.header.head.payload. |
02918be2 | 3762 | |
e42325a5 | 3763 | Usually in mdev->tconn->data.rbuf.header.head the callback can find the usual |
02918be2 PR |
3764 | p_header, but they may not rely on that. Since there is also p_header95 ! |
3765 | */ | |
b411b363 PR |
3766 | |
3767 | static void drbdd(struct drbd_conf *mdev) | |
3768 | { | |
c012949a | 3769 | struct p_header *header = &mdev->tconn->data.rbuf.header; |
02918be2 | 3770 | unsigned int packet_size; |
d8763023 | 3771 | enum drbd_packet cmd; |
02918be2 PR |
3772 | size_t shs; /* sub header size */ |
3773 | int rv; | |
b411b363 | 3774 | |
e6b3ea83 | 3775 | while (get_t_state(&mdev->tconn->receiver) == RUNNING) { |
b411b363 | 3776 | drbd_thread_current_set_cpu(mdev); |
02918be2 PR |
3777 | if (!drbd_recv_header(mdev, &cmd, &packet_size)) |
3778 | goto err_out; | |
b411b363 | 3779 | |
02918be2 PR |
3780 | if (unlikely(cmd >= P_MAX_CMD || !drbd_cmd_handler[cmd].function)) { |
3781 | dev_err(DEV, "unknown packet type %d, l: %d!\n", cmd, packet_size); | |
3782 | goto err_out; | |
0b33a916 | 3783 | } |
b411b363 | 3784 | |
c012949a | 3785 | shs = drbd_cmd_handler[cmd].pkt_size - sizeof(struct p_header); |
02918be2 PR |
3786 | if (packet_size - shs > 0 && !drbd_cmd_handler[cmd].expect_payload) { |
3787 | dev_err(DEV, "No payload expected %s l:%d\n", cmdname(cmd), packet_size); | |
3788 | goto err_out; | |
b411b363 | 3789 | } |
b411b363 | 3790 | |
c13f7e1a | 3791 | if (shs) { |
c012949a | 3792 | rv = drbd_recv(mdev, &header->payload, shs); |
c13f7e1a | 3793 | if (unlikely(rv != shs)) { |
0ddc5549 LE |
3794 | if (!signal_pending(current)) |
3795 | dev_warn(DEV, "short read while reading sub header: rv=%d\n", rv); | |
c13f7e1a LE |
3796 | goto err_out; |
3797 | } | |
3798 | } | |
3799 | ||
02918be2 | 3800 | rv = drbd_cmd_handler[cmd].function(mdev, cmd, packet_size - shs); |
b411b363 | 3801 | |
02918be2 | 3802 | if (unlikely(!rv)) { |
b411b363 | 3803 | dev_err(DEV, "error receiving %s, l: %d!\n", |
02918be2 PR |
3804 | cmdname(cmd), packet_size); |
3805 | goto err_out; | |
b411b363 PR |
3806 | } |
3807 | } | |
b411b363 | 3808 | |
02918be2 PR |
3809 | if (0) { |
3810 | err_out: | |
3811 | drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); | |
3812 | } | |
856c50c7 LE |
3813 | /* If we leave here, we probably want to update at least the |
3814 | * "Connected" indicator on stable storage. Do so explicitly here. */ | |
3815 | drbd_md_sync(mdev); | |
b411b363 PR |
3816 | } |
3817 | ||
191d3cc8 | 3818 | void drbd_flush_workqueue(struct drbd_tconn *tconn) |
b411b363 PR |
3819 | { |
3820 | struct drbd_wq_barrier barr; | |
3821 | ||
3822 | barr.w.cb = w_prev_work_done; | |
3823 | init_completion(&barr.done); | |
191d3cc8 | 3824 | drbd_queue_work(&tconn->data.work, &barr.w); |
b411b363 PR |
3825 | wait_for_completion(&barr.done); |
3826 | } | |
3827 | ||
3828 | static void drbd_disconnect(struct drbd_conf *mdev) | |
3829 | { | |
3830 | enum drbd_fencing_p fp; | |
3831 | union drbd_state os, ns; | |
3832 | int rv = SS_UNKNOWN_ERROR; | |
3833 | unsigned int i; | |
3834 | ||
3835 | if (mdev->state.conn == C_STANDALONE) | |
3836 | return; | |
b411b363 PR |
3837 | |
3838 | /* asender does not clean up anything. it must not interfere, either */ | |
e6b3ea83 | 3839 | drbd_thread_stop(&mdev->tconn->asender); |
b411b363 | 3840 | drbd_free_sock(mdev); |
b411b363 | 3841 | |
85719573 | 3842 | /* wait for current activity to cease. */ |
87eeee41 | 3843 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3844 | _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
3845 | _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); | |
3846 | _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); | |
87eeee41 | 3847 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3848 | |
3849 | /* We do not have data structures that would allow us to | |
3850 | * get the rs_pending_cnt down to 0 again. | |
3851 | * * On C_SYNC_TARGET we do not have any data structures describing | |
3852 | * the pending RSDataRequest's we have sent. | |
3853 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
3854 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
3855 | * And no, it is not the sum of the reference counts in the | |
3856 | * resync_LRU. The resync_LRU tracks the whole operation including | |
3857 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
3858 | * on the fly. */ | |
3859 | drbd_rs_cancel_all(mdev); | |
3860 | mdev->rs_total = 0; | |
3861 | mdev->rs_failed = 0; | |
3862 | atomic_set(&mdev->rs_pending_cnt, 0); | |
3863 | wake_up(&mdev->misc_wait); | |
3864 | ||
7fde2be9 PR |
3865 | del_timer(&mdev->request_timer); |
3866 | ||
b411b363 PR |
3867 | /* make sure syncer is stopped and w_resume_next_sg queued */ |
3868 | del_timer_sync(&mdev->resync_timer); | |
b411b363 PR |
3869 | resync_timer_fn((unsigned long)mdev); |
3870 | ||
b411b363 PR |
3871 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
3872 | * w_make_resync_request etc. which may still be on the worker queue | |
3873 | * to be "canceled" */ | |
191d3cc8 | 3874 | drbd_flush_workqueue(mdev->tconn); |
b411b363 PR |
3875 | |
3876 | /* This also does reclaim_net_ee(). If we do this too early, we might | |
3877 | * miss some resync ee and pages.*/ | |
3878 | drbd_process_done_ee(mdev); | |
3879 | ||
3880 | kfree(mdev->p_uuid); | |
3881 | mdev->p_uuid = NULL; | |
3882 | ||
fb22c402 | 3883 | if (!is_susp(mdev->state)) |
b411b363 PR |
3884 | tl_clear(mdev); |
3885 | ||
b411b363 PR |
3886 | dev_info(DEV, "Connection closed\n"); |
3887 | ||
3888 | drbd_md_sync(mdev); | |
3889 | ||
3890 | fp = FP_DONT_CARE; | |
3891 | if (get_ldev(mdev)) { | |
3892 | fp = mdev->ldev->dc.fencing; | |
3893 | put_ldev(mdev); | |
3894 | } | |
3895 | ||
87f7be4c PR |
3896 | if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) |
3897 | drbd_try_outdate_peer_async(mdev); | |
b411b363 | 3898 | |
87eeee41 | 3899 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3900 | os = mdev->state; |
3901 | if (os.conn >= C_UNCONNECTED) { | |
3902 | /* Do not restart in case we are C_DISCONNECTING */ | |
3903 | ns = os; | |
3904 | ns.conn = C_UNCONNECTED; | |
3905 | rv = _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
3906 | } | |
87eeee41 | 3907 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3908 | |
3909 | if (os.conn == C_DISCONNECTING) { | |
b2fb6dbe | 3910 | wait_event(mdev->tconn->net_cnt_wait, atomic_read(&mdev->tconn->net_cnt) == 0); |
b411b363 | 3911 | |
a0638456 PR |
3912 | crypto_free_hash(mdev->tconn->cram_hmac_tfm); |
3913 | mdev->tconn->cram_hmac_tfm = NULL; | |
b411b363 | 3914 | |
89e58e75 PR |
3915 | kfree(mdev->tconn->net_conf); |
3916 | mdev->tconn->net_conf = NULL; | |
b411b363 PR |
3917 | drbd_request_state(mdev, NS(conn, C_STANDALONE)); |
3918 | } | |
3919 | ||
20ceb2b2 LE |
3920 | /* serialize with bitmap writeout triggered by the state change, |
3921 | * if any. */ | |
3922 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); | |
3923 | ||
b411b363 PR |
3924 | /* tcp_close and release of sendpage pages can be deferred. I don't |
3925 | * want to use SO_LINGER, because apparently it can be deferred for | |
3926 | * more than 20 seconds (longest time I checked). | |
3927 | * | |
3928 | * Actually we don't care for exactly when the network stack does its | |
3929 | * put_page(), but release our reference on these pages right here. | |
3930 | */ | |
3931 | i = drbd_release_ee(mdev, &mdev->net_ee); | |
3932 | if (i) | |
3933 | dev_info(DEV, "net_ee not empty, killed %u entries\n", i); | |
435f0740 LE |
3934 | i = atomic_read(&mdev->pp_in_use_by_net); |
3935 | if (i) | |
3936 | dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); | |
b411b363 PR |
3937 | i = atomic_read(&mdev->pp_in_use); |
3938 | if (i) | |
45bb912b | 3939 | dev_info(DEV, "pp_in_use = %d, expected 0\n", i); |
b411b363 PR |
3940 | |
3941 | D_ASSERT(list_empty(&mdev->read_ee)); | |
3942 | D_ASSERT(list_empty(&mdev->active_ee)); | |
3943 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
3944 | D_ASSERT(list_empty(&mdev->done_ee)); | |
3945 | ||
3946 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ | |
3947 | atomic_set(&mdev->current_epoch->epoch_size, 0); | |
3948 | D_ASSERT(list_empty(&mdev->current_epoch->list)); | |
3949 | } | |
3950 | ||
3951 | /* | |
3952 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
3953 | * we can agree on is stored in agreed_pro_version. | |
3954 | * | |
3955 | * feature flags and the reserved array should be enough room for future | |
3956 | * enhancements of the handshake protocol, and possible plugins... | |
3957 | * | |
3958 | * for now, they are expected to be zero, but ignored. | |
3959 | */ | |
3960 | static int drbd_send_handshake(struct drbd_conf *mdev) | |
3961 | { | |
e6b3ea83 | 3962 | /* ASSERT current == mdev->tconn->receiver ... */ |
e42325a5 | 3963 | struct p_handshake *p = &mdev->tconn->data.sbuf.handshake; |
b411b363 PR |
3964 | int ok; |
3965 | ||
e42325a5 | 3966 | if (mutex_lock_interruptible(&mdev->tconn->data.mutex)) { |
b411b363 PR |
3967 | dev_err(DEV, "interrupted during initial handshake\n"); |
3968 | return 0; /* interrupted. not ok. */ | |
3969 | } | |
3970 | ||
e42325a5 PR |
3971 | if (mdev->tconn->data.socket == NULL) { |
3972 | mutex_unlock(&mdev->tconn->data.mutex); | |
b411b363 PR |
3973 | return 0; |
3974 | } | |
3975 | ||
3976 | memset(p, 0, sizeof(*p)); | |
3977 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
3978 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
c012949a PR |
3979 | ok = _drbd_send_cmd(mdev, mdev->tconn->data.socket, P_HAND_SHAKE, |
3980 | &p->head, sizeof(*p), 0 ); | |
e42325a5 | 3981 | mutex_unlock(&mdev->tconn->data.mutex); |
b411b363 PR |
3982 | return ok; |
3983 | } | |
3984 | ||
3985 | /* | |
3986 | * return values: | |
3987 | * 1 yes, we have a valid connection | |
3988 | * 0 oops, did not work out, please try again | |
3989 | * -1 peer talks different language, | |
3990 | * no point in trying again, please go standalone. | |
3991 | */ | |
3992 | static int drbd_do_handshake(struct drbd_conf *mdev) | |
3993 | { | |
e6b3ea83 | 3994 | /* ASSERT current == mdev->tconn->receiver ... */ |
e42325a5 | 3995 | struct p_handshake *p = &mdev->tconn->data.rbuf.handshake; |
02918be2 PR |
3996 | const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80); |
3997 | unsigned int length; | |
d8763023 | 3998 | enum drbd_packet cmd; |
b411b363 PR |
3999 | int rv; |
4000 | ||
4001 | rv = drbd_send_handshake(mdev); | |
4002 | if (!rv) | |
4003 | return 0; | |
4004 | ||
02918be2 | 4005 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
4006 | if (!rv) |
4007 | return 0; | |
4008 | ||
02918be2 | 4009 | if (cmd != P_HAND_SHAKE) { |
b411b363 | 4010 | dev_err(DEV, "expected HandShake packet, received: %s (0x%04x)\n", |
02918be2 | 4011 | cmdname(cmd), cmd); |
b411b363 PR |
4012 | return -1; |
4013 | } | |
4014 | ||
02918be2 | 4015 | if (length != expect) { |
b411b363 | 4016 | dev_err(DEV, "expected HandShake length: %u, received: %u\n", |
02918be2 | 4017 | expect, length); |
b411b363 PR |
4018 | return -1; |
4019 | } | |
4020 | ||
4021 | rv = drbd_recv(mdev, &p->head.payload, expect); | |
4022 | ||
4023 | if (rv != expect) { | |
0ddc5549 LE |
4024 | if (!signal_pending(current)) |
4025 | dev_warn(DEV, "short read receiving handshake packet: l=%u\n", rv); | |
b411b363 PR |
4026 | return 0; |
4027 | } | |
4028 | ||
b411b363 PR |
4029 | p->protocol_min = be32_to_cpu(p->protocol_min); |
4030 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
4031 | if (p->protocol_max == 0) | |
4032 | p->protocol_max = p->protocol_min; | |
4033 | ||
4034 | if (PRO_VERSION_MAX < p->protocol_min || | |
4035 | PRO_VERSION_MIN > p->protocol_max) | |
4036 | goto incompat; | |
4037 | ||
31890f4a | 4038 | mdev->tconn->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); |
b411b363 PR |
4039 | |
4040 | dev_info(DEV, "Handshake successful: " | |
31890f4a | 4041 | "Agreed network protocol version %d\n", mdev->tconn->agreed_pro_version); |
b411b363 PR |
4042 | |
4043 | return 1; | |
4044 | ||
4045 | incompat: | |
4046 | dev_err(DEV, "incompatible DRBD dialects: " | |
4047 | "I support %d-%d, peer supports %d-%d\n", | |
4048 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
4049 | p->protocol_min, p->protocol_max); | |
4050 | return -1; | |
4051 | } | |
4052 | ||
4053 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
4054 | static int drbd_do_auth(struct drbd_conf *mdev) | |
4055 | { | |
4056 | dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); | |
4057 | dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4058 | return -1; |
b411b363 PR |
4059 | } |
4060 | #else | |
4061 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4062 | |
4063 | /* Return value: | |
4064 | 1 - auth succeeded, | |
4065 | 0 - failed, try again (network error), | |
4066 | -1 - auth failed, don't try again. | |
4067 | */ | |
4068 | ||
b411b363 PR |
4069 | static int drbd_do_auth(struct drbd_conf *mdev) |
4070 | { | |
4071 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ | |
4072 | struct scatterlist sg; | |
4073 | char *response = NULL; | |
4074 | char *right_response = NULL; | |
4075 | char *peers_ch = NULL; | |
89e58e75 | 4076 | unsigned int key_len = strlen(mdev->tconn->net_conf->shared_secret); |
b411b363 PR |
4077 | unsigned int resp_size; |
4078 | struct hash_desc desc; | |
d8763023 | 4079 | enum drbd_packet cmd; |
02918be2 | 4080 | unsigned int length; |
b411b363 PR |
4081 | int rv; |
4082 | ||
a0638456 | 4083 | desc.tfm = mdev->tconn->cram_hmac_tfm; |
b411b363 PR |
4084 | desc.flags = 0; |
4085 | ||
a0638456 | 4086 | rv = crypto_hash_setkey(mdev->tconn->cram_hmac_tfm, |
89e58e75 | 4087 | (u8 *)mdev->tconn->net_conf->shared_secret, key_len); |
b411b363 PR |
4088 | if (rv) { |
4089 | dev_err(DEV, "crypto_hash_setkey() failed with %d\n", rv); | |
b10d96cb | 4090 | rv = -1; |
b411b363 PR |
4091 | goto fail; |
4092 | } | |
4093 | ||
4094 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4095 | ||
4096 | rv = drbd_send_cmd2(mdev, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); | |
4097 | if (!rv) | |
4098 | goto fail; | |
4099 | ||
02918be2 | 4100 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
4101 | if (!rv) |
4102 | goto fail; | |
4103 | ||
02918be2 | 4104 | if (cmd != P_AUTH_CHALLENGE) { |
b411b363 | 4105 | dev_err(DEV, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
02918be2 | 4106 | cmdname(cmd), cmd); |
b411b363 PR |
4107 | rv = 0; |
4108 | goto fail; | |
4109 | } | |
4110 | ||
02918be2 | 4111 | if (length > CHALLENGE_LEN * 2) { |
b411b363 | 4112 | dev_err(DEV, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4113 | rv = -1; |
b411b363 PR |
4114 | goto fail; |
4115 | } | |
4116 | ||
02918be2 | 4117 | peers_ch = kmalloc(length, GFP_NOIO); |
b411b363 PR |
4118 | if (peers_ch == NULL) { |
4119 | dev_err(DEV, "kmalloc of peers_ch failed\n"); | |
b10d96cb | 4120 | rv = -1; |
b411b363 PR |
4121 | goto fail; |
4122 | } | |
4123 | ||
02918be2 | 4124 | rv = drbd_recv(mdev, peers_ch, length); |
b411b363 | 4125 | |
02918be2 | 4126 | if (rv != length) { |
0ddc5549 LE |
4127 | if (!signal_pending(current)) |
4128 | dev_warn(DEV, "short read AuthChallenge: l=%u\n", rv); | |
b411b363 PR |
4129 | rv = 0; |
4130 | goto fail; | |
4131 | } | |
4132 | ||
a0638456 | 4133 | resp_size = crypto_hash_digestsize(mdev->tconn->cram_hmac_tfm); |
b411b363 PR |
4134 | response = kmalloc(resp_size, GFP_NOIO); |
4135 | if (response == NULL) { | |
4136 | dev_err(DEV, "kmalloc of response failed\n"); | |
b10d96cb | 4137 | rv = -1; |
b411b363 PR |
4138 | goto fail; |
4139 | } | |
4140 | ||
4141 | sg_init_table(&sg, 1); | |
02918be2 | 4142 | sg_set_buf(&sg, peers_ch, length); |
b411b363 PR |
4143 | |
4144 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4145 | if (rv) { | |
4146 | dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); | |
b10d96cb | 4147 | rv = -1; |
b411b363 PR |
4148 | goto fail; |
4149 | } | |
4150 | ||
4151 | rv = drbd_send_cmd2(mdev, P_AUTH_RESPONSE, response, resp_size); | |
4152 | if (!rv) | |
4153 | goto fail; | |
4154 | ||
02918be2 | 4155 | rv = drbd_recv_header(mdev, &cmd, &length); |
b411b363 PR |
4156 | if (!rv) |
4157 | goto fail; | |
4158 | ||
02918be2 | 4159 | if (cmd != P_AUTH_RESPONSE) { |
b411b363 | 4160 | dev_err(DEV, "expected AuthResponse packet, received: %s (0x%04x)\n", |
02918be2 | 4161 | cmdname(cmd), cmd); |
b411b363 PR |
4162 | rv = 0; |
4163 | goto fail; | |
4164 | } | |
4165 | ||
02918be2 | 4166 | if (length != resp_size) { |
b411b363 PR |
4167 | dev_err(DEV, "expected AuthResponse payload of wrong size\n"); |
4168 | rv = 0; | |
4169 | goto fail; | |
4170 | } | |
4171 | ||
4172 | rv = drbd_recv(mdev, response , resp_size); | |
4173 | ||
4174 | if (rv != resp_size) { | |
0ddc5549 LE |
4175 | if (!signal_pending(current)) |
4176 | dev_warn(DEV, "short read receiving AuthResponse: l=%u\n", rv); | |
b411b363 PR |
4177 | rv = 0; |
4178 | goto fail; | |
4179 | } | |
4180 | ||
4181 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4182 | if (right_response == NULL) { |
b411b363 | 4183 | dev_err(DEV, "kmalloc of right_response failed\n"); |
b10d96cb | 4184 | rv = -1; |
b411b363 PR |
4185 | goto fail; |
4186 | } | |
4187 | ||
4188 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4189 | ||
4190 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4191 | if (rv) { | |
4192 | dev_err(DEV, "crypto_hash_digest() failed with %d\n", rv); | |
b10d96cb | 4193 | rv = -1; |
b411b363 PR |
4194 | goto fail; |
4195 | } | |
4196 | ||
4197 | rv = !memcmp(response, right_response, resp_size); | |
4198 | ||
4199 | if (rv) | |
4200 | dev_info(DEV, "Peer authenticated using %d bytes of '%s' HMAC\n", | |
89e58e75 | 4201 | resp_size, mdev->tconn->net_conf->cram_hmac_alg); |
b10d96cb JT |
4202 | else |
4203 | rv = -1; | |
b411b363 PR |
4204 | |
4205 | fail: | |
4206 | kfree(peers_ch); | |
4207 | kfree(response); | |
4208 | kfree(right_response); | |
4209 | ||
4210 | return rv; | |
4211 | } | |
4212 | #endif | |
4213 | ||
4214 | int drbdd_init(struct drbd_thread *thi) | |
4215 | { | |
4216 | struct drbd_conf *mdev = thi->mdev; | |
4217 | unsigned int minor = mdev_to_minor(mdev); | |
4218 | int h; | |
4219 | ||
4220 | sprintf(current->comm, "drbd%d_receiver", minor); | |
4221 | ||
4222 | dev_info(DEV, "receiver (re)started\n"); | |
4223 | ||
4224 | do { | |
4225 | h = drbd_connect(mdev); | |
4226 | if (h == 0) { | |
4227 | drbd_disconnect(mdev); | |
20ee6390 | 4228 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
4229 | } |
4230 | if (h == -1) { | |
4231 | dev_warn(DEV, "Discarding network configuration.\n"); | |
4232 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
4233 | } | |
4234 | } while (h == 0); | |
4235 | ||
4236 | if (h > 0) { | |
b2fb6dbe | 4237 | if (get_net_conf(mdev->tconn)) { |
b411b363 | 4238 | drbdd(mdev); |
b2fb6dbe | 4239 | put_net_conf(mdev->tconn); |
b411b363 PR |
4240 | } |
4241 | } | |
4242 | ||
4243 | drbd_disconnect(mdev); | |
4244 | ||
4245 | dev_info(DEV, "receiver terminated\n"); | |
4246 | return 0; | |
4247 | } | |
4248 | ||
4249 | /* ********* acknowledge sender ******** */ | |
4250 | ||
d8763023 | 4251 | static int got_RqSReply(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4252 | { |
257d0af6 | 4253 | struct p_req_state_reply *p = &mdev->tconn->meta.rbuf.req_state_reply; |
b411b363 PR |
4254 | |
4255 | int retcode = be32_to_cpu(p->retcode); | |
4256 | ||
4257 | if (retcode >= SS_SUCCESS) { | |
4258 | set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); | |
4259 | } else { | |
4260 | set_bit(CL_ST_CHG_FAIL, &mdev->flags); | |
4261 | dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", | |
4262 | drbd_set_st_err_str(retcode), retcode); | |
4263 | } | |
4264 | wake_up(&mdev->state_wait); | |
4265 | ||
81e84650 | 4266 | return true; |
b411b363 PR |
4267 | } |
4268 | ||
d8763023 | 4269 | static int got_Ping(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 PR |
4270 | { |
4271 | return drbd_send_ping_ack(mdev); | |
4272 | ||
4273 | } | |
4274 | ||
d8763023 | 4275 | static int got_PingAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 PR |
4276 | { |
4277 | /* restore idle timeout */ | |
e42325a5 | 4278 | mdev->tconn->meta.socket->sk->sk_rcvtimeo = mdev->tconn->net_conf->ping_int*HZ; |
309d1608 PR |
4279 | if (!test_and_set_bit(GOT_PING_ACK, &mdev->flags)) |
4280 | wake_up(&mdev->misc_wait); | |
b411b363 | 4281 | |
81e84650 | 4282 | return true; |
b411b363 PR |
4283 | } |
4284 | ||
d8763023 | 4285 | static int got_IsInSync(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4286 | { |
257d0af6 | 4287 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4288 | sector_t sector = be64_to_cpu(p->sector); |
4289 | int blksize = be32_to_cpu(p->blksize); | |
4290 | ||
31890f4a | 4291 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
b411b363 PR |
4292 | |
4293 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4294 | ||
1d53f09e LE |
4295 | if (get_ldev(mdev)) { |
4296 | drbd_rs_complete_io(mdev, sector); | |
4297 | drbd_set_in_sync(mdev, sector, blksize); | |
4298 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ | |
4299 | mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); | |
4300 | put_ldev(mdev); | |
4301 | } | |
b411b363 | 4302 | dec_rs_pending(mdev); |
778f271d | 4303 | atomic_add(blksize >> 9, &mdev->rs_sect_in); |
b411b363 | 4304 | |
81e84650 | 4305 | return true; |
b411b363 PR |
4306 | } |
4307 | ||
bc9c5c41 AG |
4308 | static int |
4309 | validate_req_change_req_state(struct drbd_conf *mdev, u64 id, sector_t sector, | |
4310 | struct rb_root *root, const char *func, | |
4311 | enum drbd_req_event what, bool missing_ok) | |
b411b363 PR |
4312 | { |
4313 | struct drbd_request *req; | |
4314 | struct bio_and_error m; | |
4315 | ||
87eeee41 | 4316 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 4317 | req = find_request(mdev, root, id, sector, missing_ok, func); |
b411b363 | 4318 | if (unlikely(!req)) { |
87eeee41 | 4319 | spin_unlock_irq(&mdev->tconn->req_lock); |
81e84650 | 4320 | return false; |
b411b363 PR |
4321 | } |
4322 | __req_mod(req, what, &m); | |
87eeee41 | 4323 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4324 | |
4325 | if (m.bio) | |
4326 | complete_master_bio(mdev, &m); | |
81e84650 | 4327 | return true; |
b411b363 PR |
4328 | } |
4329 | ||
d8763023 | 4330 | static int got_BlockAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4331 | { |
257d0af6 | 4332 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4333 | sector_t sector = be64_to_cpu(p->sector); |
4334 | int blksize = be32_to_cpu(p->blksize); | |
4335 | enum drbd_req_event what; | |
4336 | ||
4337 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4338 | ||
579b57ed | 4339 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4340 | drbd_set_in_sync(mdev, sector, blksize); |
4341 | dec_rs_pending(mdev); | |
81e84650 | 4342 | return true; |
b411b363 | 4343 | } |
257d0af6 | 4344 | switch (cmd) { |
b411b363 | 4345 | case P_RS_WRITE_ACK: |
89e58e75 | 4346 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
8554df1c | 4347 | what = WRITE_ACKED_BY_PEER_AND_SIS; |
b411b363 PR |
4348 | break; |
4349 | case P_WRITE_ACK: | |
89e58e75 | 4350 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
8554df1c | 4351 | what = WRITE_ACKED_BY_PEER; |
b411b363 PR |
4352 | break; |
4353 | case P_RECV_ACK: | |
89e58e75 | 4354 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_B); |
8554df1c | 4355 | what = RECV_ACKED_BY_PEER; |
b411b363 PR |
4356 | break; |
4357 | case P_DISCARD_ACK: | |
89e58e75 | 4358 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
8554df1c | 4359 | what = CONFLICT_DISCARDED_BY_PEER; |
b411b363 PR |
4360 | break; |
4361 | default: | |
4362 | D_ASSERT(0); | |
81e84650 | 4363 | return false; |
b411b363 PR |
4364 | } |
4365 | ||
4366 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
bc9c5c41 AG |
4367 | &mdev->write_requests, __func__, |
4368 | what, false); | |
b411b363 PR |
4369 | } |
4370 | ||
d8763023 | 4371 | static int got_NegAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4372 | { |
257d0af6 | 4373 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 | 4374 | sector_t sector = be64_to_cpu(p->sector); |
2deb8336 | 4375 | int size = be32_to_cpu(p->blksize); |
89e58e75 PR |
4376 | bool missing_ok = mdev->tconn->net_conf->wire_protocol == DRBD_PROT_A || |
4377 | mdev->tconn->net_conf->wire_protocol == DRBD_PROT_B; | |
c3afd8f5 | 4378 | bool found; |
b411b363 PR |
4379 | |
4380 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4381 | ||
579b57ed | 4382 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4383 | dec_rs_pending(mdev); |
4384 | drbd_rs_failed_io(mdev, sector, size); | |
81e84650 | 4385 | return true; |
b411b363 | 4386 | } |
2deb8336 | 4387 | |
c3afd8f5 | 4388 | found = validate_req_change_req_state(mdev, p->block_id, sector, |
bc9c5c41 | 4389 | &mdev->write_requests, __func__, |
8554df1c | 4390 | NEG_ACKED, missing_ok); |
c3afd8f5 AG |
4391 | if (!found) { |
4392 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. | |
4393 | The master bio might already be completed, therefore the | |
4394 | request is no longer in the collision hash. */ | |
4395 | /* In Protocol B we might already have got a P_RECV_ACK | |
4396 | but then get a P_NEG_ACK afterwards. */ | |
4397 | if (!missing_ok) | |
2deb8336 | 4398 | return false; |
c3afd8f5 | 4399 | drbd_set_out_of_sync(mdev, sector, size); |
2deb8336 | 4400 | } |
2deb8336 | 4401 | return true; |
b411b363 PR |
4402 | } |
4403 | ||
d8763023 | 4404 | static int got_NegDReply(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4405 | { |
257d0af6 | 4406 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4407 | sector_t sector = be64_to_cpu(p->sector); |
4408 | ||
4409 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4410 | dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", | |
4411 | (unsigned long long)sector, be32_to_cpu(p->blksize)); | |
4412 | ||
4413 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
bc9c5c41 | 4414 | &mdev->read_requests, __func__, |
8554df1c | 4415 | NEG_ACKED, false); |
b411b363 PR |
4416 | } |
4417 | ||
d8763023 | 4418 | static int got_NegRSDReply(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 PR |
4419 | { |
4420 | sector_t sector; | |
4421 | int size; | |
257d0af6 | 4422 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4423 | |
4424 | sector = be64_to_cpu(p->sector); | |
4425 | size = be32_to_cpu(p->blksize); | |
b411b363 PR |
4426 | |
4427 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4428 | ||
4429 | dec_rs_pending(mdev); | |
4430 | ||
4431 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
4432 | drbd_rs_complete_io(mdev, sector); | |
257d0af6 | 4433 | switch (cmd) { |
d612d309 PR |
4434 | case P_NEG_RS_DREPLY: |
4435 | drbd_rs_failed_io(mdev, sector, size); | |
4436 | case P_RS_CANCEL: | |
4437 | break; | |
4438 | default: | |
4439 | D_ASSERT(0); | |
4440 | put_ldev(mdev); | |
4441 | return false; | |
4442 | } | |
b411b363 PR |
4443 | put_ldev(mdev); |
4444 | } | |
4445 | ||
81e84650 | 4446 | return true; |
b411b363 PR |
4447 | } |
4448 | ||
d8763023 | 4449 | static int got_BarrierAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4450 | { |
257d0af6 | 4451 | struct p_barrier_ack *p = &mdev->tconn->meta.rbuf.barrier_ack; |
b411b363 PR |
4452 | |
4453 | tl_release(mdev, p->barrier, be32_to_cpu(p->set_size)); | |
4454 | ||
c4752ef1 PR |
4455 | if (mdev->state.conn == C_AHEAD && |
4456 | atomic_read(&mdev->ap_in_flight) == 0 && | |
370a43e7 PR |
4457 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags)) { |
4458 | mdev->start_resync_timer.expires = jiffies + HZ; | |
4459 | add_timer(&mdev->start_resync_timer); | |
c4752ef1 PR |
4460 | } |
4461 | ||
81e84650 | 4462 | return true; |
b411b363 PR |
4463 | } |
4464 | ||
d8763023 | 4465 | static int got_OVResult(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4466 | { |
257d0af6 | 4467 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4468 | struct drbd_work *w; |
4469 | sector_t sector; | |
4470 | int size; | |
4471 | ||
4472 | sector = be64_to_cpu(p->sector); | |
4473 | size = be32_to_cpu(p->blksize); | |
4474 | ||
4475 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4476 | ||
4477 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
4478 | drbd_ov_oos_found(mdev, sector, size); | |
4479 | else | |
4480 | ov_oos_print(mdev); | |
4481 | ||
1d53f09e | 4482 | if (!get_ldev(mdev)) |
81e84650 | 4483 | return true; |
1d53f09e | 4484 | |
b411b363 PR |
4485 | drbd_rs_complete_io(mdev, sector); |
4486 | dec_rs_pending(mdev); | |
4487 | ||
ea5442af LE |
4488 | --mdev->ov_left; |
4489 | ||
4490 | /* let's advance progress step marks only for every other megabyte */ | |
4491 | if ((mdev->ov_left & 0x200) == 0x200) | |
4492 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
4493 | ||
4494 | if (mdev->ov_left == 0) { | |
b411b363 PR |
4495 | w = kmalloc(sizeof(*w), GFP_NOIO); |
4496 | if (w) { | |
4497 | w->cb = w_ov_finished; | |
e42325a5 | 4498 | drbd_queue_work_front(&mdev->tconn->data.work, w); |
b411b363 PR |
4499 | } else { |
4500 | dev_err(DEV, "kmalloc(w) failed."); | |
4501 | ov_oos_print(mdev); | |
4502 | drbd_resync_finished(mdev); | |
4503 | } | |
4504 | } | |
1d53f09e | 4505 | put_ldev(mdev); |
81e84650 | 4506 | return true; |
b411b363 PR |
4507 | } |
4508 | ||
d8763023 | 4509 | static int got_skip(struct drbd_conf *mdev, enum drbd_packet cmd) |
0ced55a3 | 4510 | { |
81e84650 | 4511 | return true; |
0ced55a3 PR |
4512 | } |
4513 | ||
b411b363 PR |
4514 | struct asender_cmd { |
4515 | size_t pkt_size; | |
d8763023 | 4516 | int (*process)(struct drbd_conf *mdev, enum drbd_packet cmd); |
b411b363 PR |
4517 | }; |
4518 | ||
4519 | static struct asender_cmd *get_asender_cmd(int cmd) | |
4520 | { | |
4521 | static struct asender_cmd asender_tbl[] = { | |
4522 | /* anything missing from this table is in | |
4523 | * the drbd_cmd_handler (drbd_default_handler) table, | |
4524 | * see the beginning of drbdd() */ | |
257d0af6 PR |
4525 | [P_PING] = { sizeof(struct p_header), got_Ping }, |
4526 | [P_PING_ACK] = { sizeof(struct p_header), got_PingAck }, | |
b411b363 PR |
4527 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, |
4528 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4529 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4530 | [P_DISCARD_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4531 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, | |
4532 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
4533 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply}, | |
4534 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, | |
4535 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
4536 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
4537 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
02918be2 | 4538 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, |
d612d309 | 4539 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply}, |
b411b363 PR |
4540 | [P_MAX_CMD] = { 0, NULL }, |
4541 | }; | |
4542 | if (cmd > P_MAX_CMD || asender_tbl[cmd].process == NULL) | |
4543 | return NULL; | |
4544 | return &asender_tbl[cmd]; | |
4545 | } | |
4546 | ||
4547 | int drbd_asender(struct drbd_thread *thi) | |
4548 | { | |
4549 | struct drbd_conf *mdev = thi->mdev; | |
257d0af6 | 4550 | struct p_header *h = &mdev->tconn->meta.rbuf.header; |
b411b363 PR |
4551 | struct asender_cmd *cmd = NULL; |
4552 | ||
257d0af6 | 4553 | int rv; |
b411b363 PR |
4554 | void *buf = h; |
4555 | int received = 0; | |
257d0af6 | 4556 | int expect = sizeof(struct p_header); |
f36af18c | 4557 | int ping_timeout_active = 0; |
257d0af6 | 4558 | int empty, pkt_size; |
d8763023 | 4559 | enum drbd_packet cmd_nr; |
b411b363 PR |
4560 | |
4561 | sprintf(current->comm, "drbd%d_asender", mdev_to_minor(mdev)); | |
4562 | ||
4563 | current->policy = SCHED_RR; /* Make this a realtime task! */ | |
4564 | current->rt_priority = 2; /* more important than all other tasks */ | |
4565 | ||
e77a0a5c | 4566 | while (get_t_state(thi) == RUNNING) { |
b411b363 PR |
4567 | drbd_thread_current_set_cpu(mdev); |
4568 | if (test_and_clear_bit(SEND_PING, &mdev->flags)) { | |
841ce241 AG |
4569 | if (!drbd_send_ping(mdev)) { |
4570 | dev_err(DEV, "drbd_send_ping has failed\n"); | |
4571 | goto reconnect; | |
4572 | } | |
e42325a5 | 4573 | mdev->tconn->meta.socket->sk->sk_rcvtimeo = |
89e58e75 | 4574 | mdev->tconn->net_conf->ping_timeo*HZ/10; |
f36af18c | 4575 | ping_timeout_active = 1; |
b411b363 PR |
4576 | } |
4577 | ||
4578 | /* conditionally cork; | |
4579 | * it may hurt latency if we cork without much to send */ | |
89e58e75 | 4580 | if (!mdev->tconn->net_conf->no_cork && |
b411b363 | 4581 | 3 < atomic_read(&mdev->unacked_cnt)) |
e42325a5 | 4582 | drbd_tcp_cork(mdev->tconn->meta.socket); |
b411b363 PR |
4583 | while (1) { |
4584 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4585 | flush_signals(current); | |
0f8488e1 | 4586 | if (!drbd_process_done_ee(mdev)) |
b411b363 | 4587 | goto reconnect; |
b411b363 PR |
4588 | /* to avoid race with newly queued ACKs */ |
4589 | set_bit(SIGNAL_ASENDER, &mdev->flags); | |
87eeee41 | 4590 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 4591 | empty = list_empty(&mdev->done_ee); |
87eeee41 | 4592 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4593 | /* new ack may have been queued right here, |
4594 | * but then there is also a signal pending, | |
4595 | * and we start over... */ | |
4596 | if (empty) | |
4597 | break; | |
4598 | } | |
4599 | /* but unconditionally uncork unless disabled */ | |
89e58e75 | 4600 | if (!mdev->tconn->net_conf->no_cork) |
e42325a5 | 4601 | drbd_tcp_uncork(mdev->tconn->meta.socket); |
b411b363 PR |
4602 | |
4603 | /* short circuit, recv_msg would return EINTR anyways. */ | |
4604 | if (signal_pending(current)) | |
4605 | continue; | |
4606 | ||
e42325a5 | 4607 | rv = drbd_recv_short(mdev, mdev->tconn->meta.socket, |
b411b363 PR |
4608 | buf, expect-received, 0); |
4609 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4610 | ||
4611 | flush_signals(current); | |
4612 | ||
4613 | /* Note: | |
4614 | * -EINTR (on meta) we got a signal | |
4615 | * -EAGAIN (on meta) rcvtimeo expired | |
4616 | * -ECONNRESET other side closed the connection | |
4617 | * -ERESTARTSYS (on data) we got a signal | |
4618 | * rv < 0 other than above: unexpected error! | |
4619 | * rv == expected: full header or command | |
4620 | * rv < expected: "woken" by signal during receive | |
4621 | * rv == 0 : "connection shut down by peer" | |
4622 | */ | |
4623 | if (likely(rv > 0)) { | |
4624 | received += rv; | |
4625 | buf += rv; | |
4626 | } else if (rv == 0) { | |
4627 | dev_err(DEV, "meta connection shut down by peer.\n"); | |
4628 | goto reconnect; | |
4629 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
4630 | /* If the data socket received something meanwhile, |
4631 | * that is good enough: peer is still alive. */ | |
31890f4a | 4632 | if (time_after(mdev->tconn->last_received, |
e42325a5 | 4633 | jiffies - mdev->tconn->meta.socket->sk->sk_rcvtimeo)) |
cb6518cb | 4634 | continue; |
f36af18c | 4635 | if (ping_timeout_active) { |
b411b363 PR |
4636 | dev_err(DEV, "PingAck did not arrive in time.\n"); |
4637 | goto reconnect; | |
4638 | } | |
4639 | set_bit(SEND_PING, &mdev->flags); | |
4640 | continue; | |
4641 | } else if (rv == -EINTR) { | |
4642 | continue; | |
4643 | } else { | |
4644 | dev_err(DEV, "sock_recvmsg returned %d\n", rv); | |
4645 | goto reconnect; | |
4646 | } | |
4647 | ||
4648 | if (received == expect && cmd == NULL) { | |
257d0af6 | 4649 | if (!decode_header(mdev, h, &cmd_nr, &pkt_size)) |
b411b363 | 4650 | goto reconnect; |
257d0af6 | 4651 | cmd = get_asender_cmd(cmd_nr); |
b411b363 | 4652 | if (unlikely(cmd == NULL)) { |
257d0af6 PR |
4653 | dev_err(DEV, "unknown command %d on meta (l: %d)\n", |
4654 | cmd_nr, pkt_size); | |
b411b363 PR |
4655 | goto disconnect; |
4656 | } | |
4657 | expect = cmd->pkt_size; | |
257d0af6 PR |
4658 | if (pkt_size != expect - sizeof(struct p_header)) { |
4659 | dev_err(DEV, "Wrong packet size on meta (c: %d, l: %d)\n", | |
4660 | cmd_nr, pkt_size); | |
b411b363 | 4661 | goto reconnect; |
257d0af6 | 4662 | } |
b411b363 PR |
4663 | } |
4664 | if (received == expect) { | |
31890f4a | 4665 | mdev->tconn->last_received = jiffies; |
b411b363 | 4666 | D_ASSERT(cmd != NULL); |
257d0af6 | 4667 | if (!cmd->process(mdev, cmd_nr)) |
b411b363 PR |
4668 | goto reconnect; |
4669 | ||
f36af18c LE |
4670 | /* the idle_timeout (ping-int) |
4671 | * has been restored in got_PingAck() */ | |
4672 | if (cmd == get_asender_cmd(P_PING_ACK)) | |
4673 | ping_timeout_active = 0; | |
4674 | ||
b411b363 PR |
4675 | buf = h; |
4676 | received = 0; | |
257d0af6 | 4677 | expect = sizeof(struct p_header); |
b411b363 PR |
4678 | cmd = NULL; |
4679 | } | |
4680 | } | |
4681 | ||
4682 | if (0) { | |
4683 | reconnect: | |
4684 | drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); | |
856c50c7 | 4685 | drbd_md_sync(mdev); |
b411b363 PR |
4686 | } |
4687 | if (0) { | |
4688 | disconnect: | |
4689 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
856c50c7 | 4690 | drbd_md_sync(mdev); |
b411b363 PR |
4691 | } |
4692 | clear_bit(SIGNAL_ASENDER, &mdev->flags); | |
4693 | ||
4694 | D_ASSERT(mdev->state.conn < C_CONNECTED); | |
4695 | dev_info(DEV, "asender terminated\n"); | |
4696 | ||
4697 | return 0; | |
4698 | } |