Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_receiver.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | */ | |
24 | ||
25 | ||
b411b363 PR |
26 | #include <linux/module.h> |
27 | ||
28 | #include <asm/uaccess.h> | |
29 | #include <net/sock.h> | |
30 | ||
b411b363 PR |
31 | #include <linux/drbd.h> |
32 | #include <linux/fs.h> | |
33 | #include <linux/file.h> | |
34 | #include <linux/in.h> | |
35 | #include <linux/mm.h> | |
36 | #include <linux/memcontrol.h> | |
37 | #include <linux/mm_inline.h> | |
38 | #include <linux/slab.h> | |
b411b363 PR |
39 | #include <linux/pkt_sched.h> |
40 | #define __KERNEL_SYSCALLS__ | |
41 | #include <linux/unistd.h> | |
42 | #include <linux/vmalloc.h> | |
43 | #include <linux/random.h> | |
b411b363 PR |
44 | #include <linux/string.h> |
45 | #include <linux/scatterlist.h> | |
46 | #include "drbd_int.h" | |
b411b363 PR |
47 | #include "drbd_req.h" |
48 | ||
49 | #include "drbd_vli.h" | |
50 | ||
77351055 PR |
51 | struct packet_info { |
52 | enum drbd_packet cmd; | |
53 | int size; | |
54 | int vnr; | |
55 | }; | |
56 | ||
b411b363 PR |
57 | enum finish_epoch { |
58 | FE_STILL_LIVE, | |
59 | FE_DESTROYED, | |
60 | FE_RECYCLED, | |
61 | }; | |
62 | ||
a4fbda8e PR |
63 | enum mdev_or_conn { |
64 | MDEV, | |
65 | CONN, | |
66 | }; | |
67 | ||
65d11ed6 | 68 | static int drbd_do_handshake(struct drbd_tconn *tconn); |
13e6037d | 69 | static int drbd_do_auth(struct drbd_tconn *tconn); |
360cc740 | 70 | static int drbd_disconnected(int vnr, void *p, void *data); |
b411b363 PR |
71 | |
72 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event); | |
99920dc5 | 73 | static int e_end_block(struct drbd_work *, int); |
b411b363 | 74 | |
b411b363 PR |
75 | |
76 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
77 | ||
45bb912b LE |
78 | /* |
79 | * some helper functions to deal with single linked page lists, | |
80 | * page->private being our "next" pointer. | |
81 | */ | |
82 | ||
83 | /* If at least n pages are linked at head, get n pages off. | |
84 | * Otherwise, don't modify head, and return NULL. | |
85 | * Locking is the responsibility of the caller. | |
86 | */ | |
87 | static struct page *page_chain_del(struct page **head, int n) | |
88 | { | |
89 | struct page *page; | |
90 | struct page *tmp; | |
91 | ||
92 | BUG_ON(!n); | |
93 | BUG_ON(!head); | |
94 | ||
95 | page = *head; | |
23ce4227 PR |
96 | |
97 | if (!page) | |
98 | return NULL; | |
99 | ||
45bb912b LE |
100 | while (page) { |
101 | tmp = page_chain_next(page); | |
102 | if (--n == 0) | |
103 | break; /* found sufficient pages */ | |
104 | if (tmp == NULL) | |
105 | /* insufficient pages, don't use any of them. */ | |
106 | return NULL; | |
107 | page = tmp; | |
108 | } | |
109 | ||
110 | /* add end of list marker for the returned list */ | |
111 | set_page_private(page, 0); | |
112 | /* actual return value, and adjustment of head */ | |
113 | page = *head; | |
114 | *head = tmp; | |
115 | return page; | |
116 | } | |
117 | ||
118 | /* may be used outside of locks to find the tail of a (usually short) | |
119 | * "private" page chain, before adding it back to a global chain head | |
120 | * with page_chain_add() under a spinlock. */ | |
121 | static struct page *page_chain_tail(struct page *page, int *len) | |
122 | { | |
123 | struct page *tmp; | |
124 | int i = 1; | |
125 | while ((tmp = page_chain_next(page))) | |
126 | ++i, page = tmp; | |
127 | if (len) | |
128 | *len = i; | |
129 | return page; | |
130 | } | |
131 | ||
132 | static int page_chain_free(struct page *page) | |
133 | { | |
134 | struct page *tmp; | |
135 | int i = 0; | |
136 | page_chain_for_each_safe(page, tmp) { | |
137 | put_page(page); | |
138 | ++i; | |
139 | } | |
140 | return i; | |
141 | } | |
142 | ||
143 | static void page_chain_add(struct page **head, | |
144 | struct page *chain_first, struct page *chain_last) | |
145 | { | |
146 | #if 1 | |
147 | struct page *tmp; | |
148 | tmp = page_chain_tail(chain_first, NULL); | |
149 | BUG_ON(tmp != chain_last); | |
150 | #endif | |
151 | ||
152 | /* add chain to head */ | |
153 | set_page_private(chain_last, (unsigned long)*head); | |
154 | *head = chain_first; | |
155 | } | |
156 | ||
157 | static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number) | |
b411b363 PR |
158 | { |
159 | struct page *page = NULL; | |
45bb912b LE |
160 | struct page *tmp = NULL; |
161 | int i = 0; | |
b411b363 PR |
162 | |
163 | /* Yes, testing drbd_pp_vacant outside the lock is racy. | |
164 | * So what. It saves a spin_lock. */ | |
45bb912b | 165 | if (drbd_pp_vacant >= number) { |
b411b363 | 166 | spin_lock(&drbd_pp_lock); |
45bb912b LE |
167 | page = page_chain_del(&drbd_pp_pool, number); |
168 | if (page) | |
169 | drbd_pp_vacant -= number; | |
b411b363 | 170 | spin_unlock(&drbd_pp_lock); |
45bb912b LE |
171 | if (page) |
172 | return page; | |
b411b363 | 173 | } |
45bb912b | 174 | |
b411b363 PR |
175 | /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD |
176 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
177 | * which in turn might block on the other node at this very place. */ | |
45bb912b LE |
178 | for (i = 0; i < number; i++) { |
179 | tmp = alloc_page(GFP_TRY); | |
180 | if (!tmp) | |
181 | break; | |
182 | set_page_private(tmp, (unsigned long)page); | |
183 | page = tmp; | |
184 | } | |
185 | ||
186 | if (i == number) | |
187 | return page; | |
188 | ||
189 | /* Not enough pages immediately available this time. | |
190 | * No need to jump around here, drbd_pp_alloc will retry this | |
191 | * function "soon". */ | |
192 | if (page) { | |
193 | tmp = page_chain_tail(page, NULL); | |
194 | spin_lock(&drbd_pp_lock); | |
195 | page_chain_add(&drbd_pp_pool, page, tmp); | |
196 | drbd_pp_vacant += i; | |
197 | spin_unlock(&drbd_pp_lock); | |
198 | } | |
199 | return NULL; | |
b411b363 PR |
200 | } |
201 | ||
b411b363 PR |
202 | static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed) |
203 | { | |
db830c46 | 204 | struct drbd_peer_request *peer_req; |
b411b363 PR |
205 | struct list_head *le, *tle; |
206 | ||
207 | /* The EEs are always appended to the end of the list. Since | |
208 | they are sent in order over the wire, they have to finish | |
209 | in order. As soon as we see the first not finished we can | |
210 | stop to examine the list... */ | |
211 | ||
212 | list_for_each_safe(le, tle, &mdev->net_ee) { | |
db830c46 AG |
213 | peer_req = list_entry(le, struct drbd_peer_request, w.list); |
214 | if (drbd_ee_has_active_page(peer_req)) | |
b411b363 PR |
215 | break; |
216 | list_move(le, to_be_freed); | |
217 | } | |
218 | } | |
219 | ||
220 | static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) | |
221 | { | |
222 | LIST_HEAD(reclaimed); | |
db830c46 | 223 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 224 | |
87eeee41 | 225 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 226 | reclaim_net_ee(mdev, &reclaimed); |
87eeee41 | 227 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 228 | |
db830c46 AG |
229 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
230 | drbd_free_net_ee(mdev, peer_req); | |
b411b363 PR |
231 | } |
232 | ||
233 | /** | |
45bb912b | 234 | * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled) |
b411b363 | 235 | * @mdev: DRBD device. |
45bb912b LE |
236 | * @number: number of pages requested |
237 | * @retry: whether to retry, if not enough pages are available right now | |
238 | * | |
239 | * Tries to allocate number pages, first from our own page pool, then from | |
240 | * the kernel, unless this allocation would exceed the max_buffers setting. | |
241 | * Possibly retry until DRBD frees sufficient pages somewhere else. | |
b411b363 | 242 | * |
45bb912b | 243 | * Returns a page chain linked via page->private. |
b411b363 | 244 | */ |
45bb912b | 245 | static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry) |
b411b363 PR |
246 | { |
247 | struct page *page = NULL; | |
248 | DEFINE_WAIT(wait); | |
249 | ||
45bb912b LE |
250 | /* Yes, we may run up to @number over max_buffers. If we |
251 | * follow it strictly, the admin will get it wrong anyways. */ | |
89e58e75 | 252 | if (atomic_read(&mdev->pp_in_use) < mdev->tconn->net_conf->max_buffers) |
45bb912b | 253 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 | 254 | |
45bb912b | 255 | while (page == NULL) { |
b411b363 PR |
256 | prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); |
257 | ||
258 | drbd_kick_lo_and_reclaim_net(mdev); | |
259 | ||
89e58e75 | 260 | if (atomic_read(&mdev->pp_in_use) < mdev->tconn->net_conf->max_buffers) { |
45bb912b | 261 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 PR |
262 | if (page) |
263 | break; | |
264 | } | |
265 | ||
266 | if (!retry) | |
267 | break; | |
268 | ||
269 | if (signal_pending(current)) { | |
270 | dev_warn(DEV, "drbd_pp_alloc interrupted!\n"); | |
271 | break; | |
272 | } | |
273 | ||
274 | schedule(); | |
275 | } | |
276 | finish_wait(&drbd_pp_wait, &wait); | |
277 | ||
45bb912b LE |
278 | if (page) |
279 | atomic_add(number, &mdev->pp_in_use); | |
b411b363 PR |
280 | return page; |
281 | } | |
282 | ||
283 | /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc. | |
87eeee41 | 284 | * Is also used from inside an other spin_lock_irq(&mdev->tconn->req_lock); |
45bb912b LE |
285 | * Either links the page chain back to the global pool, |
286 | * or returns all pages to the system. */ | |
435f0740 | 287 | static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net) |
b411b363 | 288 | { |
435f0740 | 289 | atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use; |
b411b363 | 290 | int i; |
435f0740 | 291 | |
81a5d60e | 292 | if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE) * minor_count) |
45bb912b LE |
293 | i = page_chain_free(page); |
294 | else { | |
295 | struct page *tmp; | |
296 | tmp = page_chain_tail(page, &i); | |
297 | spin_lock(&drbd_pp_lock); | |
298 | page_chain_add(&drbd_pp_pool, page, tmp); | |
299 | drbd_pp_vacant += i; | |
300 | spin_unlock(&drbd_pp_lock); | |
b411b363 | 301 | } |
435f0740 | 302 | i = atomic_sub_return(i, a); |
45bb912b | 303 | if (i < 0) |
435f0740 LE |
304 | dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n", |
305 | is_net ? "pp_in_use_by_net" : "pp_in_use", i); | |
b411b363 PR |
306 | wake_up(&drbd_pp_wait); |
307 | } | |
308 | ||
309 | /* | |
310 | You need to hold the req_lock: | |
311 | _drbd_wait_ee_list_empty() | |
312 | ||
313 | You must not have the req_lock: | |
314 | drbd_free_ee() | |
315 | drbd_alloc_ee() | |
316 | drbd_init_ee() | |
317 | drbd_release_ee() | |
318 | drbd_ee_fix_bhs() | |
319 | drbd_process_done_ee() | |
320 | drbd_clear_done_ee() | |
321 | drbd_wait_ee_list_empty() | |
322 | */ | |
323 | ||
f6ffca9f AG |
324 | struct drbd_peer_request * |
325 | drbd_alloc_ee(struct drbd_conf *mdev, u64 id, sector_t sector, | |
326 | unsigned int data_size, gfp_t gfp_mask) __must_hold(local) | |
b411b363 | 327 | { |
db830c46 | 328 | struct drbd_peer_request *peer_req; |
b411b363 | 329 | struct page *page; |
45bb912b | 330 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b411b363 | 331 | |
0cf9d27e | 332 | if (drbd_insert_fault(mdev, DRBD_FAULT_AL_EE)) |
b411b363 PR |
333 | return NULL; |
334 | ||
db830c46 AG |
335 | peer_req = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); |
336 | if (!peer_req) { | |
b411b363 PR |
337 | if (!(gfp_mask & __GFP_NOWARN)) |
338 | dev_err(DEV, "alloc_ee: Allocation of an EE failed\n"); | |
339 | return NULL; | |
340 | } | |
341 | ||
45bb912b LE |
342 | page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT)); |
343 | if (!page) | |
344 | goto fail; | |
b411b363 | 345 | |
db830c46 AG |
346 | drbd_clear_interval(&peer_req->i); |
347 | peer_req->i.size = data_size; | |
348 | peer_req->i.sector = sector; | |
349 | peer_req->i.local = false; | |
350 | peer_req->i.waiting = false; | |
351 | ||
352 | peer_req->epoch = NULL; | |
a21e9298 | 353 | peer_req->w.mdev = mdev; |
db830c46 AG |
354 | peer_req->pages = page; |
355 | atomic_set(&peer_req->pending_bios, 0); | |
356 | peer_req->flags = 0; | |
9a8e7753 AG |
357 | /* |
358 | * The block_id is opaque to the receiver. It is not endianness | |
359 | * converted, and sent back to the sender unchanged. | |
360 | */ | |
db830c46 | 361 | peer_req->block_id = id; |
b411b363 | 362 | |
db830c46 | 363 | return peer_req; |
b411b363 | 364 | |
45bb912b | 365 | fail: |
db830c46 | 366 | mempool_free(peer_req, drbd_ee_mempool); |
b411b363 PR |
367 | return NULL; |
368 | } | |
369 | ||
db830c46 | 370 | void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_peer_request *peer_req, |
f6ffca9f | 371 | int is_net) |
b411b363 | 372 | { |
db830c46 AG |
373 | if (peer_req->flags & EE_HAS_DIGEST) |
374 | kfree(peer_req->digest); | |
375 | drbd_pp_free(mdev, peer_req->pages, is_net); | |
376 | D_ASSERT(atomic_read(&peer_req->pending_bios) == 0); | |
377 | D_ASSERT(drbd_interval_empty(&peer_req->i)); | |
378 | mempool_free(peer_req, drbd_ee_mempool); | |
b411b363 PR |
379 | } |
380 | ||
381 | int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) | |
382 | { | |
383 | LIST_HEAD(work_list); | |
db830c46 | 384 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 385 | int count = 0; |
435f0740 | 386 | int is_net = list == &mdev->net_ee; |
b411b363 | 387 | |
87eeee41 | 388 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 389 | list_splice_init(list, &work_list); |
87eeee41 | 390 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 391 | |
db830c46 AG |
392 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
393 | drbd_free_some_ee(mdev, peer_req, is_net); | |
b411b363 PR |
394 | count++; |
395 | } | |
396 | return count; | |
397 | } | |
398 | ||
399 | ||
32862ec7 | 400 | /* See also comments in _req_mod(,BARRIER_ACKED) |
b411b363 PR |
401 | * and receive_Barrier. |
402 | * | |
403 | * Move entries from net_ee to done_ee, if ready. | |
404 | * Grab done_ee, call all callbacks, free the entries. | |
405 | * The callbacks typically send out ACKs. | |
406 | */ | |
407 | static int drbd_process_done_ee(struct drbd_conf *mdev) | |
408 | { | |
409 | LIST_HEAD(work_list); | |
410 | LIST_HEAD(reclaimed); | |
db830c46 | 411 | struct drbd_peer_request *peer_req, *t; |
e2b3032b | 412 | int err = 0; |
b411b363 | 413 | |
87eeee41 | 414 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
415 | reclaim_net_ee(mdev, &reclaimed); |
416 | list_splice_init(&mdev->done_ee, &work_list); | |
87eeee41 | 417 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 418 | |
db830c46 AG |
419 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
420 | drbd_free_net_ee(mdev, peer_req); | |
b411b363 PR |
421 | |
422 | /* possible callbacks here: | |
7be8da07 | 423 | * e_end_block, and e_end_resync_block, e_send_discard_write. |
b411b363 PR |
424 | * all ignore the last argument. |
425 | */ | |
db830c46 | 426 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
e2b3032b AG |
427 | int err2; |
428 | ||
b411b363 | 429 | /* list_del not necessary, next/prev members not touched */ |
e2b3032b AG |
430 | err2 = peer_req->w.cb(&peer_req->w, !!err); |
431 | if (!err) | |
432 | err = err2; | |
db830c46 | 433 | drbd_free_ee(mdev, peer_req); |
b411b363 PR |
434 | } |
435 | wake_up(&mdev->ee_wait); | |
436 | ||
e2b3032b | 437 | return err; |
b411b363 PR |
438 | } |
439 | ||
440 | void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
441 | { | |
442 | DEFINE_WAIT(wait); | |
443 | ||
444 | /* avoids spin_lock/unlock | |
445 | * and calling prepare_to_wait in the fast path */ | |
446 | while (!list_empty(head)) { | |
447 | prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); | |
87eeee41 | 448 | spin_unlock_irq(&mdev->tconn->req_lock); |
7eaceacc | 449 | io_schedule(); |
b411b363 | 450 | finish_wait(&mdev->ee_wait, &wait); |
87eeee41 | 451 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
452 | } |
453 | } | |
454 | ||
455 | void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
456 | { | |
87eeee41 | 457 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 458 | _drbd_wait_ee_list_empty(mdev, head); |
87eeee41 | 459 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
460 | } |
461 | ||
462 | /* see also kernel_accept; which is only present since 2.6.18. | |
463 | * also we want to log which part of it failed, exactly */ | |
7653620d | 464 | static int drbd_accept(const char **what, struct socket *sock, struct socket **newsock) |
b411b363 PR |
465 | { |
466 | struct sock *sk = sock->sk; | |
467 | int err = 0; | |
468 | ||
469 | *what = "listen"; | |
470 | err = sock->ops->listen(sock, 5); | |
471 | if (err < 0) | |
472 | goto out; | |
473 | ||
474 | *what = "sock_create_lite"; | |
475 | err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, | |
476 | newsock); | |
477 | if (err < 0) | |
478 | goto out; | |
479 | ||
480 | *what = "accept"; | |
481 | err = sock->ops->accept(sock, *newsock, 0); | |
482 | if (err < 0) { | |
483 | sock_release(*newsock); | |
484 | *newsock = NULL; | |
485 | goto out; | |
486 | } | |
487 | (*newsock)->ops = sock->ops; | |
488 | ||
489 | out: | |
490 | return err; | |
491 | } | |
492 | ||
dbd9eea0 | 493 | static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags) |
b411b363 PR |
494 | { |
495 | mm_segment_t oldfs; | |
496 | struct kvec iov = { | |
497 | .iov_base = buf, | |
498 | .iov_len = size, | |
499 | }; | |
500 | struct msghdr msg = { | |
501 | .msg_iovlen = 1, | |
502 | .msg_iov = (struct iovec *)&iov, | |
503 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) | |
504 | }; | |
505 | int rv; | |
506 | ||
507 | oldfs = get_fs(); | |
508 | set_fs(KERNEL_DS); | |
509 | rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); | |
510 | set_fs(oldfs); | |
511 | ||
512 | return rv; | |
513 | } | |
514 | ||
de0ff338 | 515 | static int drbd_recv(struct drbd_tconn *tconn, void *buf, size_t size) |
b411b363 PR |
516 | { |
517 | mm_segment_t oldfs; | |
518 | struct kvec iov = { | |
519 | .iov_base = buf, | |
520 | .iov_len = size, | |
521 | }; | |
522 | struct msghdr msg = { | |
523 | .msg_iovlen = 1, | |
524 | .msg_iov = (struct iovec *)&iov, | |
525 | .msg_flags = MSG_WAITALL | MSG_NOSIGNAL | |
526 | }; | |
527 | int rv; | |
528 | ||
529 | oldfs = get_fs(); | |
530 | set_fs(KERNEL_DS); | |
531 | ||
532 | for (;;) { | |
de0ff338 | 533 | rv = sock_recvmsg(tconn->data.socket, &msg, size, msg.msg_flags); |
b411b363 PR |
534 | if (rv == size) |
535 | break; | |
536 | ||
537 | /* Note: | |
538 | * ECONNRESET other side closed the connection | |
539 | * ERESTARTSYS (on sock) we got a signal | |
540 | */ | |
541 | ||
542 | if (rv < 0) { | |
543 | if (rv == -ECONNRESET) | |
de0ff338 | 544 | conn_info(tconn, "sock was reset by peer\n"); |
b411b363 | 545 | else if (rv != -ERESTARTSYS) |
de0ff338 | 546 | conn_err(tconn, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
547 | break; |
548 | } else if (rv == 0) { | |
de0ff338 | 549 | conn_info(tconn, "sock was shut down by peer\n"); |
b411b363 PR |
550 | break; |
551 | } else { | |
552 | /* signal came in, or peer/link went down, | |
553 | * after we read a partial message | |
554 | */ | |
555 | /* D_ASSERT(signal_pending(current)); */ | |
556 | break; | |
557 | } | |
558 | }; | |
559 | ||
560 | set_fs(oldfs); | |
561 | ||
562 | if (rv != size) | |
bbeb641c | 563 | conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 PR |
564 | |
565 | return rv; | |
566 | } | |
567 | ||
c6967746 AG |
568 | static int drbd_recv_all(struct drbd_tconn *tconn, void *buf, size_t size) |
569 | { | |
570 | int err; | |
571 | ||
572 | err = drbd_recv(tconn, buf, size); | |
573 | if (err != size) { | |
574 | if (err >= 0) | |
575 | err = -EIO; | |
576 | } else | |
577 | err = 0; | |
578 | return err; | |
579 | } | |
580 | ||
a5c31904 AG |
581 | static int drbd_recv_all_warn(struct drbd_tconn *tconn, void *buf, size_t size) |
582 | { | |
583 | int err; | |
584 | ||
585 | err = drbd_recv_all(tconn, buf, size); | |
586 | if (err && !signal_pending(current)) | |
587 | conn_warn(tconn, "short read (expected size %d)\n", (int)size); | |
588 | return err; | |
589 | } | |
590 | ||
5dbf1673 LE |
591 | /* quoting tcp(7): |
592 | * On individual connections, the socket buffer size must be set prior to the | |
593 | * listen(2) or connect(2) calls in order to have it take effect. | |
594 | * This is our wrapper to do so. | |
595 | */ | |
596 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
597 | unsigned int rcv) | |
598 | { | |
599 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
600 | if (snd) { | |
601 | sock->sk->sk_sndbuf = snd; | |
602 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
603 | } | |
604 | if (rcv) { | |
605 | sock->sk->sk_rcvbuf = rcv; | |
606 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
607 | } | |
608 | } | |
609 | ||
eac3e990 | 610 | static struct socket *drbd_try_connect(struct drbd_tconn *tconn) |
b411b363 PR |
611 | { |
612 | const char *what; | |
613 | struct socket *sock; | |
614 | struct sockaddr_in6 src_in6; | |
615 | int err; | |
616 | int disconnect_on_error = 1; | |
617 | ||
eac3e990 | 618 | if (!get_net_conf(tconn)) |
b411b363 PR |
619 | return NULL; |
620 | ||
621 | what = "sock_create_kern"; | |
eac3e990 | 622 | err = sock_create_kern(((struct sockaddr *)tconn->net_conf->my_addr)->sa_family, |
b411b363 PR |
623 | SOCK_STREAM, IPPROTO_TCP, &sock); |
624 | if (err < 0) { | |
625 | sock = NULL; | |
626 | goto out; | |
627 | } | |
628 | ||
629 | sock->sk->sk_rcvtimeo = | |
eac3e990 PR |
630 | sock->sk->sk_sndtimeo = tconn->net_conf->try_connect_int*HZ; |
631 | drbd_setbufsize(sock, tconn->net_conf->sndbuf_size, | |
632 | tconn->net_conf->rcvbuf_size); | |
b411b363 PR |
633 | |
634 | /* explicitly bind to the configured IP as source IP | |
635 | * for the outgoing connections. | |
636 | * This is needed for multihomed hosts and to be | |
637 | * able to use lo: interfaces for drbd. | |
638 | * Make sure to use 0 as port number, so linux selects | |
639 | * a free one dynamically. | |
640 | */ | |
eac3e990 PR |
641 | memcpy(&src_in6, tconn->net_conf->my_addr, |
642 | min_t(int, tconn->net_conf->my_addr_len, sizeof(src_in6))); | |
643 | if (((struct sockaddr *)tconn->net_conf->my_addr)->sa_family == AF_INET6) | |
b411b363 PR |
644 | src_in6.sin6_port = 0; |
645 | else | |
646 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
647 | ||
648 | what = "bind before connect"; | |
649 | err = sock->ops->bind(sock, | |
650 | (struct sockaddr *) &src_in6, | |
eac3e990 | 651 | tconn->net_conf->my_addr_len); |
b411b363 PR |
652 | if (err < 0) |
653 | goto out; | |
654 | ||
655 | /* connect may fail, peer not yet available. | |
656 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
657 | disconnect_on_error = 0; | |
658 | what = "connect"; | |
659 | err = sock->ops->connect(sock, | |
eac3e990 PR |
660 | (struct sockaddr *)tconn->net_conf->peer_addr, |
661 | tconn->net_conf->peer_addr_len, 0); | |
b411b363 PR |
662 | |
663 | out: | |
664 | if (err < 0) { | |
665 | if (sock) { | |
666 | sock_release(sock); | |
667 | sock = NULL; | |
668 | } | |
669 | switch (-err) { | |
670 | /* timeout, busy, signal pending */ | |
671 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
672 | case EINTR: case ERESTARTSYS: | |
673 | /* peer not (yet) available, network problem */ | |
674 | case ECONNREFUSED: case ENETUNREACH: | |
675 | case EHOSTDOWN: case EHOSTUNREACH: | |
676 | disconnect_on_error = 0; | |
677 | break; | |
678 | default: | |
eac3e990 | 679 | conn_err(tconn, "%s failed, err = %d\n", what, err); |
b411b363 PR |
680 | } |
681 | if (disconnect_on_error) | |
bbeb641c | 682 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 683 | } |
eac3e990 | 684 | put_net_conf(tconn); |
b411b363 PR |
685 | return sock; |
686 | } | |
687 | ||
7653620d | 688 | static struct socket *drbd_wait_for_connect(struct drbd_tconn *tconn) |
b411b363 PR |
689 | { |
690 | int timeo, err; | |
691 | struct socket *s_estab = NULL, *s_listen; | |
692 | const char *what; | |
693 | ||
7653620d | 694 | if (!get_net_conf(tconn)) |
b411b363 PR |
695 | return NULL; |
696 | ||
697 | what = "sock_create_kern"; | |
7653620d | 698 | err = sock_create_kern(((struct sockaddr *)tconn->net_conf->my_addr)->sa_family, |
b411b363 PR |
699 | SOCK_STREAM, IPPROTO_TCP, &s_listen); |
700 | if (err) { | |
701 | s_listen = NULL; | |
702 | goto out; | |
703 | } | |
704 | ||
7653620d | 705 | timeo = tconn->net_conf->try_connect_int * HZ; |
b411b363 PR |
706 | timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ |
707 | ||
708 | s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
709 | s_listen->sk->sk_rcvtimeo = timeo; | |
710 | s_listen->sk->sk_sndtimeo = timeo; | |
7653620d PR |
711 | drbd_setbufsize(s_listen, tconn->net_conf->sndbuf_size, |
712 | tconn->net_conf->rcvbuf_size); | |
b411b363 PR |
713 | |
714 | what = "bind before listen"; | |
715 | err = s_listen->ops->bind(s_listen, | |
7653620d PR |
716 | (struct sockaddr *) tconn->net_conf->my_addr, |
717 | tconn->net_conf->my_addr_len); | |
b411b363 PR |
718 | if (err < 0) |
719 | goto out; | |
720 | ||
7653620d | 721 | err = drbd_accept(&what, s_listen, &s_estab); |
b411b363 PR |
722 | |
723 | out: | |
724 | if (s_listen) | |
725 | sock_release(s_listen); | |
726 | if (err < 0) { | |
727 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
7653620d | 728 | conn_err(tconn, "%s failed, err = %d\n", what, err); |
bbeb641c | 729 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
730 | } |
731 | } | |
7653620d | 732 | put_net_conf(tconn); |
b411b363 PR |
733 | |
734 | return s_estab; | |
735 | } | |
736 | ||
d38e787e | 737 | static int drbd_send_fp(struct drbd_tconn *tconn, struct socket *sock, enum drbd_packet cmd) |
b411b363 | 738 | { |
d38e787e | 739 | struct p_header *h = &tconn->data.sbuf.header; |
b411b363 | 740 | |
ecf2363c | 741 | return !_conn_send_cmd(tconn, 0, sock, cmd, h, sizeof(*h), 0); |
b411b363 PR |
742 | } |
743 | ||
a25b63f1 | 744 | static enum drbd_packet drbd_recv_fp(struct drbd_tconn *tconn, struct socket *sock) |
b411b363 | 745 | { |
a25b63f1 | 746 | struct p_header80 *h = &tconn->data.rbuf.header.h80; |
b411b363 PR |
747 | int rr; |
748 | ||
dbd9eea0 | 749 | rr = drbd_recv_short(sock, h, sizeof(*h), 0); |
b411b363 | 750 | |
ca9bc12b | 751 | if (rr == sizeof(*h) && h->magic == cpu_to_be32(DRBD_MAGIC)) |
b411b363 PR |
752 | return be16_to_cpu(h->command); |
753 | ||
754 | return 0xffff; | |
755 | } | |
756 | ||
757 | /** | |
758 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
b411b363 PR |
759 | * @sock: pointer to the pointer to the socket. |
760 | */ | |
dbd9eea0 | 761 | static int drbd_socket_okay(struct socket **sock) |
b411b363 PR |
762 | { |
763 | int rr; | |
764 | char tb[4]; | |
765 | ||
766 | if (!*sock) | |
81e84650 | 767 | return false; |
b411b363 | 768 | |
dbd9eea0 | 769 | rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); |
b411b363 PR |
770 | |
771 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 772 | return true; |
b411b363 PR |
773 | } else { |
774 | sock_release(*sock); | |
775 | *sock = NULL; | |
81e84650 | 776 | return false; |
b411b363 PR |
777 | } |
778 | } | |
2325eb66 PR |
779 | /* Gets called if a connection is established, or if a new minor gets created |
780 | in a connection */ | |
781 | int drbd_connected(int vnr, void *p, void *data) | |
907599e0 PR |
782 | { |
783 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
0829f5ed | 784 | int err; |
907599e0 PR |
785 | |
786 | atomic_set(&mdev->packet_seq, 0); | |
787 | mdev->peer_seq = 0; | |
788 | ||
8410da8f PR |
789 | mdev->state_mutex = mdev->tconn->agreed_pro_version < 100 ? |
790 | &mdev->tconn->cstate_mutex : | |
791 | &mdev->own_state_mutex; | |
792 | ||
0829f5ed AG |
793 | err = drbd_send_sync_param(mdev); |
794 | if (!err) | |
795 | err = drbd_send_sizes(mdev, 0, 0); | |
796 | if (!err) | |
797 | err = drbd_send_uuids(mdev); | |
798 | if (!err) | |
799 | err = drbd_send_state(mdev); | |
907599e0 PR |
800 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); |
801 | clear_bit(RESIZE_PENDING, &mdev->flags); | |
0829f5ed | 802 | return err; |
907599e0 PR |
803 | } |
804 | ||
b411b363 PR |
805 | /* |
806 | * return values: | |
807 | * 1 yes, we have a valid connection | |
808 | * 0 oops, did not work out, please try again | |
809 | * -1 peer talks different language, | |
810 | * no point in trying again, please go standalone. | |
811 | * -2 We do not have a network config... | |
812 | */ | |
907599e0 | 813 | static int drbd_connect(struct drbd_tconn *tconn) |
b411b363 PR |
814 | { |
815 | struct socket *s, *sock, *msock; | |
816 | int try, h, ok; | |
817 | ||
bbeb641c | 818 | if (conn_request_state(tconn, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) |
b411b363 PR |
819 | return -2; |
820 | ||
907599e0 PR |
821 | clear_bit(DISCARD_CONCURRENT, &tconn->flags); |
822 | tconn->agreed_pro_version = 99; | |
fd340c12 PR |
823 | /* agreed_pro_version must be smaller than 100 so we send the old |
824 | header (h80) in the first packet and in the handshake packet. */ | |
b411b363 PR |
825 | |
826 | sock = NULL; | |
827 | msock = NULL; | |
828 | ||
829 | do { | |
830 | for (try = 0;;) { | |
831 | /* 3 tries, this should take less than a second! */ | |
907599e0 | 832 | s = drbd_try_connect(tconn); |
b411b363 PR |
833 | if (s || ++try >= 3) |
834 | break; | |
835 | /* give the other side time to call bind() & listen() */ | |
20ee6390 | 836 | schedule_timeout_interruptible(HZ / 10); |
b411b363 PR |
837 | } |
838 | ||
839 | if (s) { | |
840 | if (!sock) { | |
907599e0 | 841 | drbd_send_fp(tconn, s, P_HAND_SHAKE_S); |
b411b363 PR |
842 | sock = s; |
843 | s = NULL; | |
844 | } else if (!msock) { | |
907599e0 | 845 | drbd_send_fp(tconn, s, P_HAND_SHAKE_M); |
b411b363 PR |
846 | msock = s; |
847 | s = NULL; | |
848 | } else { | |
907599e0 | 849 | conn_err(tconn, "Logic error in drbd_connect()\n"); |
b411b363 PR |
850 | goto out_release_sockets; |
851 | } | |
852 | } | |
853 | ||
854 | if (sock && msock) { | |
907599e0 | 855 | schedule_timeout_interruptible(tconn->net_conf->ping_timeo*HZ/10); |
dbd9eea0 PR |
856 | ok = drbd_socket_okay(&sock); |
857 | ok = drbd_socket_okay(&msock) && ok; | |
b411b363 PR |
858 | if (ok) |
859 | break; | |
860 | } | |
861 | ||
862 | retry: | |
907599e0 | 863 | s = drbd_wait_for_connect(tconn); |
b411b363 | 864 | if (s) { |
907599e0 | 865 | try = drbd_recv_fp(tconn, s); |
dbd9eea0 PR |
866 | drbd_socket_okay(&sock); |
867 | drbd_socket_okay(&msock); | |
b411b363 PR |
868 | switch (try) { |
869 | case P_HAND_SHAKE_S: | |
870 | if (sock) { | |
907599e0 | 871 | conn_warn(tconn, "initial packet S crossed\n"); |
b411b363 PR |
872 | sock_release(sock); |
873 | } | |
874 | sock = s; | |
875 | break; | |
876 | case P_HAND_SHAKE_M: | |
877 | if (msock) { | |
907599e0 | 878 | conn_warn(tconn, "initial packet M crossed\n"); |
b411b363 PR |
879 | sock_release(msock); |
880 | } | |
881 | msock = s; | |
907599e0 | 882 | set_bit(DISCARD_CONCURRENT, &tconn->flags); |
b411b363 PR |
883 | break; |
884 | default: | |
907599e0 | 885 | conn_warn(tconn, "Error receiving initial packet\n"); |
b411b363 PR |
886 | sock_release(s); |
887 | if (random32() & 1) | |
888 | goto retry; | |
889 | } | |
890 | } | |
891 | ||
bbeb641c | 892 | if (tconn->cstate <= C_DISCONNECTING) |
b411b363 PR |
893 | goto out_release_sockets; |
894 | if (signal_pending(current)) { | |
895 | flush_signals(current); | |
896 | smp_rmb(); | |
907599e0 | 897 | if (get_t_state(&tconn->receiver) == EXITING) |
b411b363 PR |
898 | goto out_release_sockets; |
899 | } | |
900 | ||
901 | if (sock && msock) { | |
dbd9eea0 PR |
902 | ok = drbd_socket_okay(&sock); |
903 | ok = drbd_socket_okay(&msock) && ok; | |
b411b363 PR |
904 | if (ok) |
905 | break; | |
906 | } | |
907 | } while (1); | |
908 | ||
909 | msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
910 | sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
911 | ||
912 | sock->sk->sk_allocation = GFP_NOIO; | |
913 | msock->sk->sk_allocation = GFP_NOIO; | |
914 | ||
915 | sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; | |
916 | msock->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
917 | ||
b411b363 | 918 | /* NOT YET ... |
907599e0 | 919 | * sock->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
b411b363 PR |
920 | * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
921 | * first set it to the P_HAND_SHAKE timeout, | |
922 | * which we set to 4x the configured ping_timeout. */ | |
923 | sock->sk->sk_sndtimeo = | |
907599e0 | 924 | sock->sk->sk_rcvtimeo = tconn->net_conf->ping_timeo*4*HZ/10; |
b411b363 | 925 | |
907599e0 PR |
926 | msock->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
927 | msock->sk->sk_rcvtimeo = tconn->net_conf->ping_int*HZ; | |
b411b363 PR |
928 | |
929 | /* we don't want delays. | |
25985edc | 930 | * we use TCP_CORK where appropriate, though */ |
b411b363 PR |
931 | drbd_tcp_nodelay(sock); |
932 | drbd_tcp_nodelay(msock); | |
933 | ||
907599e0 PR |
934 | tconn->data.socket = sock; |
935 | tconn->meta.socket = msock; | |
936 | tconn->last_received = jiffies; | |
b411b363 | 937 | |
907599e0 | 938 | h = drbd_do_handshake(tconn); |
b411b363 PR |
939 | if (h <= 0) |
940 | return h; | |
941 | ||
907599e0 | 942 | if (tconn->cram_hmac_tfm) { |
b411b363 | 943 | /* drbd_request_state(mdev, NS(conn, WFAuth)); */ |
907599e0 | 944 | switch (drbd_do_auth(tconn)) { |
b10d96cb | 945 | case -1: |
907599e0 | 946 | conn_err(tconn, "Authentication of peer failed\n"); |
b411b363 | 947 | return -1; |
b10d96cb | 948 | case 0: |
907599e0 | 949 | conn_err(tconn, "Authentication of peer failed, trying again.\n"); |
b10d96cb | 950 | return 0; |
b411b363 PR |
951 | } |
952 | } | |
953 | ||
bbeb641c | 954 | if (conn_request_state(tconn, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE) < SS_SUCCESS) |
b411b363 PR |
955 | return 0; |
956 | ||
907599e0 | 957 | sock->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
b411b363 PR |
958 | sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
959 | ||
907599e0 | 960 | drbd_thread_start(&tconn->asender); |
b411b363 | 961 | |
387eb308 | 962 | if (drbd_send_protocol(tconn) == -EOPNOTSUPP) |
7e2455c1 | 963 | return -1; |
b411b363 | 964 | |
907599e0 | 965 | return !idr_for_each(&tconn->volumes, drbd_connected, tconn); |
b411b363 PR |
966 | |
967 | out_release_sockets: | |
968 | if (sock) | |
969 | sock_release(sock); | |
970 | if (msock) | |
971 | sock_release(msock); | |
972 | return -1; | |
973 | } | |
974 | ||
8172f3e9 | 975 | static int decode_header(struct drbd_tconn *tconn, struct p_header *h, struct packet_info *pi) |
b411b363 | 976 | { |
fd340c12 | 977 | if (h->h80.magic == cpu_to_be32(DRBD_MAGIC)) { |
77351055 PR |
978 | pi->cmd = be16_to_cpu(h->h80.command); |
979 | pi->size = be16_to_cpu(h->h80.length); | |
eefc2f7d | 980 | pi->vnr = 0; |
ca9bc12b | 981 | } else if (h->h95.magic == cpu_to_be16(DRBD_MAGIC_BIG)) { |
77351055 PR |
982 | pi->cmd = be16_to_cpu(h->h95.command); |
983 | pi->size = be32_to_cpu(h->h95.length) & 0x00ffffff; | |
984 | pi->vnr = 0; | |
02918be2 | 985 | } else { |
ce243853 | 986 | conn_err(tconn, "magic?? on data m: 0x%08x c: %d l: %d\n", |
004352fa LE |
987 | be32_to_cpu(h->h80.magic), |
988 | be16_to_cpu(h->h80.command), | |
989 | be16_to_cpu(h->h80.length)); | |
8172f3e9 | 990 | return -EINVAL; |
b411b363 | 991 | } |
8172f3e9 | 992 | return 0; |
257d0af6 PR |
993 | } |
994 | ||
9ba7aa00 | 995 | static int drbd_recv_header(struct drbd_tconn *tconn, struct packet_info *pi) |
257d0af6 | 996 | { |
9ba7aa00 | 997 | struct p_header *h = &tconn->data.rbuf.header; |
69bc7bc3 | 998 | int err; |
257d0af6 | 999 | |
a5c31904 AG |
1000 | err = drbd_recv_all_warn(tconn, h, sizeof(*h)); |
1001 | if (err) | |
69bc7bc3 | 1002 | return err; |
257d0af6 | 1003 | |
69bc7bc3 | 1004 | err = decode_header(tconn, h, pi); |
9ba7aa00 | 1005 | tconn->last_received = jiffies; |
b411b363 | 1006 | |
69bc7bc3 | 1007 | return err; |
b411b363 PR |
1008 | } |
1009 | ||
2451fc3b | 1010 | static void drbd_flush(struct drbd_conf *mdev) |
b411b363 PR |
1011 | { |
1012 | int rv; | |
1013 | ||
1014 | if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { | |
fbd9b09a | 1015 | rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL, |
dd3932ed | 1016 | NULL); |
b411b363 PR |
1017 | if (rv) { |
1018 | dev_err(DEV, "local disk flush failed with status %d\n", rv); | |
1019 | /* would rather check on EOPNOTSUPP, but that is not reliable. | |
1020 | * don't try again for ANY return value != 0 | |
1021 | * if (rv == -EOPNOTSUPP) */ | |
1022 | drbd_bump_write_ordering(mdev, WO_drain_io); | |
1023 | } | |
1024 | put_ldev(mdev); | |
1025 | } | |
b411b363 PR |
1026 | } |
1027 | ||
1028 | /** | |
1029 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
1030 | * @mdev: DRBD device. | |
1031 | * @epoch: Epoch object. | |
1032 | * @ev: Epoch event. | |
1033 | */ | |
1034 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, | |
1035 | struct drbd_epoch *epoch, | |
1036 | enum epoch_event ev) | |
1037 | { | |
2451fc3b | 1038 | int epoch_size; |
b411b363 | 1039 | struct drbd_epoch *next_epoch; |
b411b363 PR |
1040 | enum finish_epoch rv = FE_STILL_LIVE; |
1041 | ||
1042 | spin_lock(&mdev->epoch_lock); | |
1043 | do { | |
1044 | next_epoch = NULL; | |
b411b363 PR |
1045 | |
1046 | epoch_size = atomic_read(&epoch->epoch_size); | |
1047 | ||
1048 | switch (ev & ~EV_CLEANUP) { | |
1049 | case EV_PUT: | |
1050 | atomic_dec(&epoch->active); | |
1051 | break; | |
1052 | case EV_GOT_BARRIER_NR: | |
1053 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1054 | break; |
1055 | case EV_BECAME_LAST: | |
1056 | /* nothing to do*/ | |
1057 | break; | |
1058 | } | |
1059 | ||
b411b363 PR |
1060 | if (epoch_size != 0 && |
1061 | atomic_read(&epoch->active) == 0 && | |
2451fc3b | 1062 | test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) { |
b411b363 PR |
1063 | if (!(ev & EV_CLEANUP)) { |
1064 | spin_unlock(&mdev->epoch_lock); | |
1065 | drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); | |
1066 | spin_lock(&mdev->epoch_lock); | |
1067 | } | |
1068 | dec_unacked(mdev); | |
1069 | ||
1070 | if (mdev->current_epoch != epoch) { | |
1071 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); | |
1072 | list_del(&epoch->list); | |
1073 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
1074 | mdev->epochs--; | |
b411b363 PR |
1075 | kfree(epoch); |
1076 | ||
1077 | if (rv == FE_STILL_LIVE) | |
1078 | rv = FE_DESTROYED; | |
1079 | } else { | |
1080 | epoch->flags = 0; | |
1081 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1082 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1083 | if (rv == FE_STILL_LIVE) |
1084 | rv = FE_RECYCLED; | |
2451fc3b | 1085 | wake_up(&mdev->ee_wait); |
b411b363 PR |
1086 | } |
1087 | } | |
1088 | ||
1089 | if (!next_epoch) | |
1090 | break; | |
1091 | ||
1092 | epoch = next_epoch; | |
1093 | } while (1); | |
1094 | ||
1095 | spin_unlock(&mdev->epoch_lock); | |
1096 | ||
b411b363 PR |
1097 | return rv; |
1098 | } | |
1099 | ||
1100 | /** | |
1101 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
1102 | * @mdev: DRBD device. | |
1103 | * @wo: Write ordering method to try. | |
1104 | */ | |
1105 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) | |
1106 | { | |
1107 | enum write_ordering_e pwo; | |
1108 | static char *write_ordering_str[] = { | |
1109 | [WO_none] = "none", | |
1110 | [WO_drain_io] = "drain", | |
1111 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1112 | }; |
1113 | ||
1114 | pwo = mdev->write_ordering; | |
1115 | wo = min(pwo, wo); | |
b411b363 PR |
1116 | if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) |
1117 | wo = WO_drain_io; | |
1118 | if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) | |
1119 | wo = WO_none; | |
1120 | mdev->write_ordering = wo; | |
2451fc3b | 1121 | if (pwo != mdev->write_ordering || wo == WO_bdev_flush) |
b411b363 PR |
1122 | dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); |
1123 | } | |
1124 | ||
45bb912b | 1125 | /** |
fbe29dec | 1126 | * drbd_submit_peer_request() |
45bb912b | 1127 | * @mdev: DRBD device. |
db830c46 | 1128 | * @peer_req: peer request |
45bb912b | 1129 | * @rw: flag field, see bio->bi_rw |
10f6d992 LE |
1130 | * |
1131 | * May spread the pages to multiple bios, | |
1132 | * depending on bio_add_page restrictions. | |
1133 | * | |
1134 | * Returns 0 if all bios have been submitted, | |
1135 | * -ENOMEM if we could not allocate enough bios, | |
1136 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1137 | * single page to an empty bio (which should never happen and likely indicates | |
1138 | * that the lower level IO stack is in some way broken). This has been observed | |
1139 | * on certain Xen deployments. | |
45bb912b LE |
1140 | */ |
1141 | /* TODO allocate from our own bio_set. */ | |
fbe29dec AG |
1142 | int drbd_submit_peer_request(struct drbd_conf *mdev, |
1143 | struct drbd_peer_request *peer_req, | |
1144 | const unsigned rw, const int fault_type) | |
45bb912b LE |
1145 | { |
1146 | struct bio *bios = NULL; | |
1147 | struct bio *bio; | |
db830c46 AG |
1148 | struct page *page = peer_req->pages; |
1149 | sector_t sector = peer_req->i.sector; | |
1150 | unsigned ds = peer_req->i.size; | |
45bb912b LE |
1151 | unsigned n_bios = 0; |
1152 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
10f6d992 | 1153 | int err = -ENOMEM; |
45bb912b LE |
1154 | |
1155 | /* In most cases, we will only need one bio. But in case the lower | |
1156 | * level restrictions happen to be different at this offset on this | |
1157 | * side than those of the sending peer, we may need to submit the | |
da4a75d2 LE |
1158 | * request in more than one bio. |
1159 | * | |
1160 | * Plain bio_alloc is good enough here, this is no DRBD internally | |
1161 | * generated bio, but a bio allocated on behalf of the peer. | |
1162 | */ | |
45bb912b LE |
1163 | next_bio: |
1164 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1165 | if (!bio) { | |
1166 | dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); | |
1167 | goto fail; | |
1168 | } | |
db830c46 | 1169 | /* > peer_req->i.sector, unless this is the first bio */ |
45bb912b LE |
1170 | bio->bi_sector = sector; |
1171 | bio->bi_bdev = mdev->ldev->backing_bdev; | |
45bb912b | 1172 | bio->bi_rw = rw; |
db830c46 | 1173 | bio->bi_private = peer_req; |
fcefa62e | 1174 | bio->bi_end_io = drbd_peer_request_endio; |
45bb912b LE |
1175 | |
1176 | bio->bi_next = bios; | |
1177 | bios = bio; | |
1178 | ++n_bios; | |
1179 | ||
1180 | page_chain_for_each(page) { | |
1181 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1182 | if (!bio_add_page(bio, page, len, 0)) { | |
10f6d992 LE |
1183 | /* A single page must always be possible! |
1184 | * But in case it fails anyways, | |
1185 | * we deal with it, and complain (below). */ | |
1186 | if (bio->bi_vcnt == 0) { | |
1187 | dev_err(DEV, | |
1188 | "bio_add_page failed for len=%u, " | |
1189 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
1190 | len, (unsigned long long)bio->bi_sector); | |
1191 | err = -ENOSPC; | |
1192 | goto fail; | |
1193 | } | |
45bb912b LE |
1194 | goto next_bio; |
1195 | } | |
1196 | ds -= len; | |
1197 | sector += len >> 9; | |
1198 | --nr_pages; | |
1199 | } | |
1200 | D_ASSERT(page == NULL); | |
1201 | D_ASSERT(ds == 0); | |
1202 | ||
db830c46 | 1203 | atomic_set(&peer_req->pending_bios, n_bios); |
45bb912b LE |
1204 | do { |
1205 | bio = bios; | |
1206 | bios = bios->bi_next; | |
1207 | bio->bi_next = NULL; | |
1208 | ||
45bb912b | 1209 | drbd_generic_make_request(mdev, fault_type, bio); |
45bb912b | 1210 | } while (bios); |
45bb912b LE |
1211 | return 0; |
1212 | ||
1213 | fail: | |
1214 | while (bios) { | |
1215 | bio = bios; | |
1216 | bios = bios->bi_next; | |
1217 | bio_put(bio); | |
1218 | } | |
10f6d992 | 1219 | return err; |
45bb912b LE |
1220 | } |
1221 | ||
53840641 | 1222 | static void drbd_remove_epoch_entry_interval(struct drbd_conf *mdev, |
db830c46 | 1223 | struct drbd_peer_request *peer_req) |
53840641 | 1224 | { |
db830c46 | 1225 | struct drbd_interval *i = &peer_req->i; |
53840641 AG |
1226 | |
1227 | drbd_remove_interval(&mdev->write_requests, i); | |
1228 | drbd_clear_interval(i); | |
1229 | ||
6c852bec | 1230 | /* Wake up any processes waiting for this peer request to complete. */ |
53840641 AG |
1231 | if (i->waiting) |
1232 | wake_up(&mdev->misc_wait); | |
1233 | } | |
1234 | ||
d8763023 AG |
1235 | static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packet cmd, |
1236 | unsigned int data_size) | |
b411b363 | 1237 | { |
2451fc3b | 1238 | int rv; |
e42325a5 | 1239 | struct p_barrier *p = &mdev->tconn->data.rbuf.barrier; |
b411b363 PR |
1240 | struct drbd_epoch *epoch; |
1241 | ||
b411b363 PR |
1242 | inc_unacked(mdev); |
1243 | ||
b411b363 PR |
1244 | mdev->current_epoch->barrier_nr = p->barrier; |
1245 | rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); | |
1246 | ||
1247 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1248 | * the activity log, which means it would not be resynced in case the | |
1249 | * R_PRIMARY crashes now. | |
1250 | * Therefore we must send the barrier_ack after the barrier request was | |
1251 | * completed. */ | |
1252 | switch (mdev->write_ordering) { | |
b411b363 PR |
1253 | case WO_none: |
1254 | if (rv == FE_RECYCLED) | |
82bc0194 | 1255 | return 0; |
2451fc3b PR |
1256 | |
1257 | /* receiver context, in the writeout path of the other node. | |
1258 | * avoid potential distributed deadlock */ | |
1259 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1260 | if (epoch) | |
1261 | break; | |
1262 | else | |
1263 | dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); | |
1264 | /* Fall through */ | |
b411b363 PR |
1265 | |
1266 | case WO_bdev_flush: | |
1267 | case WO_drain_io: | |
b411b363 | 1268 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
2451fc3b PR |
1269 | drbd_flush(mdev); |
1270 | ||
1271 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1272 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1273 | if (epoch) | |
1274 | break; | |
b411b363 PR |
1275 | } |
1276 | ||
2451fc3b PR |
1277 | epoch = mdev->current_epoch; |
1278 | wait_event(mdev->ee_wait, atomic_read(&epoch->epoch_size) == 0); | |
1279 | ||
1280 | D_ASSERT(atomic_read(&epoch->active) == 0); | |
1281 | D_ASSERT(epoch->flags == 0); | |
b411b363 | 1282 | |
82bc0194 | 1283 | return 0; |
2451fc3b PR |
1284 | default: |
1285 | dev_err(DEV, "Strangeness in mdev->write_ordering %d\n", mdev->write_ordering); | |
82bc0194 | 1286 | return -EIO; |
b411b363 PR |
1287 | } |
1288 | ||
1289 | epoch->flags = 0; | |
1290 | atomic_set(&epoch->epoch_size, 0); | |
1291 | atomic_set(&epoch->active, 0); | |
1292 | ||
1293 | spin_lock(&mdev->epoch_lock); | |
1294 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1295 | list_add(&epoch->list, &mdev->current_epoch->list); | |
1296 | mdev->current_epoch = epoch; | |
1297 | mdev->epochs++; | |
b411b363 PR |
1298 | } else { |
1299 | /* The current_epoch got recycled while we allocated this one... */ | |
1300 | kfree(epoch); | |
1301 | } | |
1302 | spin_unlock(&mdev->epoch_lock); | |
1303 | ||
82bc0194 | 1304 | return 0; |
b411b363 PR |
1305 | } |
1306 | ||
1307 | /* used from receive_RSDataReply (recv_resync_read) | |
1308 | * and from receive_Data */ | |
f6ffca9f AG |
1309 | static struct drbd_peer_request * |
1310 | read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, | |
1311 | int data_size) __must_hold(local) | |
b411b363 | 1312 | { |
6666032a | 1313 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); |
db830c46 | 1314 | struct drbd_peer_request *peer_req; |
b411b363 | 1315 | struct page *page; |
a5c31904 | 1316 | int dgs, ds, err; |
a0638456 PR |
1317 | void *dig_in = mdev->tconn->int_dig_in; |
1318 | void *dig_vv = mdev->tconn->int_dig_vv; | |
6b4388ac | 1319 | unsigned long *data; |
b411b363 | 1320 | |
a0638456 PR |
1321 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ? |
1322 | crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0; | |
b411b363 PR |
1323 | |
1324 | if (dgs) { | |
a5c31904 AG |
1325 | err = drbd_recv_all_warn(mdev->tconn, dig_in, dgs); |
1326 | if (err) | |
b411b363 | 1327 | return NULL; |
b411b363 PR |
1328 | } |
1329 | ||
1330 | data_size -= dgs; | |
1331 | ||
841ce241 AG |
1332 | if (!expect(data_size != 0)) |
1333 | return NULL; | |
1334 | if (!expect(IS_ALIGNED(data_size, 512))) | |
1335 | return NULL; | |
1336 | if (!expect(data_size <= DRBD_MAX_BIO_SIZE)) | |
1337 | return NULL; | |
b411b363 | 1338 | |
6666032a LE |
1339 | /* even though we trust out peer, |
1340 | * we sometimes have to double check. */ | |
1341 | if (sector + (data_size>>9) > capacity) { | |
fdda6544 LE |
1342 | dev_err(DEV, "request from peer beyond end of local disk: " |
1343 | "capacity: %llus < sector: %llus + size: %u\n", | |
6666032a LE |
1344 | (unsigned long long)capacity, |
1345 | (unsigned long long)sector, data_size); | |
1346 | return NULL; | |
1347 | } | |
1348 | ||
b411b363 PR |
1349 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1350 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1351 | * which in turn might block on the other node at this very place. */ | |
db830c46 AG |
1352 | peer_req = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); |
1353 | if (!peer_req) | |
b411b363 | 1354 | return NULL; |
45bb912b | 1355 | |
b411b363 | 1356 | ds = data_size; |
db830c46 | 1357 | page = peer_req->pages; |
45bb912b LE |
1358 | page_chain_for_each(page) { |
1359 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1360 | data = kmap(page); |
a5c31904 | 1361 | err = drbd_recv_all_warn(mdev->tconn, data, len); |
0cf9d27e | 1362 | if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) { |
6b4388ac PR |
1363 | dev_err(DEV, "Fault injection: Corrupting data on receive\n"); |
1364 | data[0] = data[0] ^ (unsigned long)-1; | |
1365 | } | |
b411b363 | 1366 | kunmap(page); |
a5c31904 | 1367 | if (err) { |
db830c46 | 1368 | drbd_free_ee(mdev, peer_req); |
b411b363 PR |
1369 | return NULL; |
1370 | } | |
a5c31904 | 1371 | ds -= len; |
b411b363 PR |
1372 | } |
1373 | ||
1374 | if (dgs) { | |
db830c46 | 1375 | drbd_csum_ee(mdev, mdev->tconn->integrity_r_tfm, peer_req, dig_vv); |
b411b363 | 1376 | if (memcmp(dig_in, dig_vv, dgs)) { |
470be44a LE |
1377 | dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n", |
1378 | (unsigned long long)sector, data_size); | |
db830c46 | 1379 | drbd_free_ee(mdev, peer_req); |
b411b363 PR |
1380 | return NULL; |
1381 | } | |
1382 | } | |
1383 | mdev->recv_cnt += data_size>>9; | |
db830c46 | 1384 | return peer_req; |
b411b363 PR |
1385 | } |
1386 | ||
1387 | /* drbd_drain_block() just takes a data block | |
1388 | * out of the socket input buffer, and discards it. | |
1389 | */ | |
1390 | static int drbd_drain_block(struct drbd_conf *mdev, int data_size) | |
1391 | { | |
1392 | struct page *page; | |
a5c31904 | 1393 | int err = 0; |
b411b363 PR |
1394 | void *data; |
1395 | ||
c3470cde | 1396 | if (!data_size) |
fc5be839 | 1397 | return 0; |
c3470cde | 1398 | |
45bb912b | 1399 | page = drbd_pp_alloc(mdev, 1, 1); |
b411b363 PR |
1400 | |
1401 | data = kmap(page); | |
1402 | while (data_size) { | |
fc5be839 AG |
1403 | unsigned int len = min_t(int, data_size, PAGE_SIZE); |
1404 | ||
a5c31904 AG |
1405 | err = drbd_recv_all_warn(mdev->tconn, data, len); |
1406 | if (err) | |
b411b363 | 1407 | break; |
a5c31904 | 1408 | data_size -= len; |
b411b363 PR |
1409 | } |
1410 | kunmap(page); | |
435f0740 | 1411 | drbd_pp_free(mdev, page, 0); |
fc5be839 | 1412 | return err; |
b411b363 PR |
1413 | } |
1414 | ||
1415 | static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, | |
1416 | sector_t sector, int data_size) | |
1417 | { | |
1418 | struct bio_vec *bvec; | |
1419 | struct bio *bio; | |
a5c31904 | 1420 | int dgs, err, i, expect; |
a0638456 PR |
1421 | void *dig_in = mdev->tconn->int_dig_in; |
1422 | void *dig_vv = mdev->tconn->int_dig_vv; | |
b411b363 | 1423 | |
a0638456 PR |
1424 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ? |
1425 | crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0; | |
b411b363 PR |
1426 | |
1427 | if (dgs) { | |
a5c31904 AG |
1428 | err = drbd_recv_all_warn(mdev->tconn, dig_in, dgs); |
1429 | if (err) | |
1430 | return err; | |
b411b363 PR |
1431 | } |
1432 | ||
1433 | data_size -= dgs; | |
1434 | ||
1435 | /* optimistically update recv_cnt. if receiving fails below, | |
1436 | * we disconnect anyways, and counters will be reset. */ | |
1437 | mdev->recv_cnt += data_size>>9; | |
1438 | ||
1439 | bio = req->master_bio; | |
1440 | D_ASSERT(sector == bio->bi_sector); | |
1441 | ||
1442 | bio_for_each_segment(bvec, bio, i) { | |
a5c31904 | 1443 | void *mapped = kmap(bvec->bv_page) + bvec->bv_offset; |
b411b363 | 1444 | expect = min_t(int, data_size, bvec->bv_len); |
a5c31904 | 1445 | err = drbd_recv_all_warn(mdev->tconn, mapped, expect); |
b411b363 | 1446 | kunmap(bvec->bv_page); |
a5c31904 AG |
1447 | if (err) |
1448 | return err; | |
1449 | data_size -= expect; | |
b411b363 PR |
1450 | } |
1451 | ||
1452 | if (dgs) { | |
a0638456 | 1453 | drbd_csum_bio(mdev, mdev->tconn->integrity_r_tfm, bio, dig_vv); |
b411b363 PR |
1454 | if (memcmp(dig_in, dig_vv, dgs)) { |
1455 | dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); | |
28284cef | 1456 | return -EINVAL; |
b411b363 PR |
1457 | } |
1458 | } | |
1459 | ||
1460 | D_ASSERT(data_size == 0); | |
28284cef | 1461 | return 0; |
b411b363 PR |
1462 | } |
1463 | ||
1464 | /* e_end_resync_block() is called via | |
1465 | * drbd_process_done_ee() by asender only */ | |
99920dc5 | 1466 | static int e_end_resync_block(struct drbd_work *w, int unused) |
b411b363 | 1467 | { |
8050e6d0 AG |
1468 | struct drbd_peer_request *peer_req = |
1469 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1470 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1471 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1472 | int err; |
b411b363 | 1473 | |
db830c46 | 1474 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1475 | |
db830c46 AG |
1476 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
1477 | drbd_set_in_sync(mdev, sector, peer_req->i.size); | |
99920dc5 | 1478 | err = drbd_send_ack(mdev, P_RS_WRITE_ACK, peer_req); |
b411b363 PR |
1479 | } else { |
1480 | /* Record failure to sync */ | |
db830c46 | 1481 | drbd_rs_failed_io(mdev, sector, peer_req->i.size); |
b411b363 | 1482 | |
99920dc5 | 1483 | err = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1484 | } |
1485 | dec_unacked(mdev); | |
1486 | ||
99920dc5 | 1487 | return err; |
b411b363 PR |
1488 | } |
1489 | ||
1490 | static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) | |
1491 | { | |
db830c46 | 1492 | struct drbd_peer_request *peer_req; |
b411b363 | 1493 | |
db830c46 AG |
1494 | peer_req = read_in_block(mdev, ID_SYNCER, sector, data_size); |
1495 | if (!peer_req) | |
45bb912b | 1496 | goto fail; |
b411b363 PR |
1497 | |
1498 | dec_rs_pending(mdev); | |
1499 | ||
b411b363 PR |
1500 | inc_unacked(mdev); |
1501 | /* corresponding dec_unacked() in e_end_resync_block() | |
1502 | * respective _drbd_clear_done_ee */ | |
1503 | ||
db830c46 | 1504 | peer_req->w.cb = e_end_resync_block; |
45bb912b | 1505 | |
87eeee41 | 1506 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1507 | list_add(&peer_req->w.list, &mdev->sync_ee); |
87eeee41 | 1508 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1509 | |
0f0601f4 | 1510 | atomic_add(data_size >> 9, &mdev->rs_sect_ev); |
fbe29dec | 1511 | if (drbd_submit_peer_request(mdev, peer_req, WRITE, DRBD_FAULT_RS_WR) == 0) |
e1c1b0fc | 1512 | return 0; |
b411b363 | 1513 | |
10f6d992 LE |
1514 | /* don't care for the reason here */ |
1515 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 1516 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1517 | list_del(&peer_req->w.list); |
87eeee41 | 1518 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 1519 | |
db830c46 | 1520 | drbd_free_ee(mdev, peer_req); |
45bb912b LE |
1521 | fail: |
1522 | put_ldev(mdev); | |
e1c1b0fc | 1523 | return -EIO; |
b411b363 PR |
1524 | } |
1525 | ||
668eebc6 | 1526 | static struct drbd_request * |
bc9c5c41 AG |
1527 | find_request(struct drbd_conf *mdev, struct rb_root *root, u64 id, |
1528 | sector_t sector, bool missing_ok, const char *func) | |
51624585 | 1529 | { |
51624585 AG |
1530 | struct drbd_request *req; |
1531 | ||
bc9c5c41 AG |
1532 | /* Request object according to our peer */ |
1533 | req = (struct drbd_request *)(unsigned long)id; | |
5e472264 | 1534 | if (drbd_contains_interval(root, sector, &req->i) && req->i.local) |
668eebc6 | 1535 | return req; |
c3afd8f5 AG |
1536 | if (!missing_ok) { |
1537 | dev_err(DEV, "%s: failed to find request %lu, sector %llus\n", func, | |
1538 | (unsigned long)id, (unsigned long long)sector); | |
1539 | } | |
51624585 AG |
1540 | return NULL; |
1541 | } | |
1542 | ||
d8763023 AG |
1543 | static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packet cmd, |
1544 | unsigned int data_size) | |
b411b363 PR |
1545 | { |
1546 | struct drbd_request *req; | |
1547 | sector_t sector; | |
82bc0194 | 1548 | int err; |
e42325a5 | 1549 | struct p_data *p = &mdev->tconn->data.rbuf.data; |
b411b363 PR |
1550 | |
1551 | sector = be64_to_cpu(p->sector); | |
1552 | ||
87eeee41 | 1553 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 1554 | req = find_request(mdev, &mdev->read_requests, p->block_id, sector, false, __func__); |
87eeee41 | 1555 | spin_unlock_irq(&mdev->tconn->req_lock); |
c3afd8f5 | 1556 | if (unlikely(!req)) |
82bc0194 | 1557 | return -EIO; |
b411b363 | 1558 | |
24c4830c | 1559 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1560 | * special casing it there for the various failure cases. |
1561 | * still no race with drbd_fail_pending_reads */ | |
82bc0194 AG |
1562 | err = recv_dless_read(mdev, req, sector, data_size); |
1563 | if (!err) | |
8554df1c | 1564 | req_mod(req, DATA_RECEIVED); |
b411b363 PR |
1565 | /* else: nothing. handled from drbd_disconnect... |
1566 | * I don't think we may complete this just yet | |
1567 | * in case we are "on-disconnect: freeze" */ | |
1568 | ||
82bc0194 | 1569 | return err; |
b411b363 PR |
1570 | } |
1571 | ||
d8763023 AG |
1572 | static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packet cmd, |
1573 | unsigned int data_size) | |
b411b363 PR |
1574 | { |
1575 | sector_t sector; | |
82bc0194 | 1576 | int err; |
e42325a5 | 1577 | struct p_data *p = &mdev->tconn->data.rbuf.data; |
b411b363 PR |
1578 | |
1579 | sector = be64_to_cpu(p->sector); | |
1580 | D_ASSERT(p->block_id == ID_SYNCER); | |
1581 | ||
1582 | if (get_ldev(mdev)) { | |
1583 | /* data is submitted to disk within recv_resync_read. | |
1584 | * corresponding put_ldev done below on error, | |
fcefa62e | 1585 | * or in drbd_peer_request_endio. */ |
82bc0194 | 1586 | err = recv_resync_read(mdev, sector, data_size); |
b411b363 PR |
1587 | } else { |
1588 | if (__ratelimit(&drbd_ratelimit_state)) | |
1589 | dev_err(DEV, "Can not write resync data to local disk.\n"); | |
1590 | ||
82bc0194 | 1591 | err = drbd_drain_block(mdev, data_size); |
b411b363 | 1592 | |
2b2bf214 | 1593 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 PR |
1594 | } |
1595 | ||
778f271d PR |
1596 | atomic_add(data_size >> 9, &mdev->rs_sect_in); |
1597 | ||
82bc0194 | 1598 | return err; |
b411b363 PR |
1599 | } |
1600 | ||
99920dc5 | 1601 | static int w_restart_write(struct drbd_work *w, int cancel) |
7be8da07 AG |
1602 | { |
1603 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
1604 | struct drbd_conf *mdev = w->mdev; | |
1605 | struct bio *bio; | |
1606 | unsigned long start_time; | |
1607 | unsigned long flags; | |
1608 | ||
1609 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
1610 | if (!expect(req->rq_state & RQ_POSTPONED)) { | |
1611 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
99920dc5 | 1612 | return -EIO; |
7be8da07 AG |
1613 | } |
1614 | bio = req->master_bio; | |
1615 | start_time = req->start_time; | |
1616 | /* Postponed requests will not have their master_bio completed! */ | |
1617 | __req_mod(req, DISCARD_WRITE, NULL); | |
1618 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
1619 | ||
1620 | while (__drbd_make_request(mdev, bio, start_time)) | |
1621 | /* retry */ ; | |
99920dc5 | 1622 | return 0; |
7be8da07 AG |
1623 | } |
1624 | ||
1625 | static void restart_conflicting_writes(struct drbd_conf *mdev, | |
1626 | sector_t sector, int size) | |
1627 | { | |
1628 | struct drbd_interval *i; | |
1629 | struct drbd_request *req; | |
1630 | ||
1631 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1632 | if (!i->local) | |
1633 | continue; | |
1634 | req = container_of(i, struct drbd_request, i); | |
1635 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1636 | !(req->rq_state & RQ_POSTPONED)) | |
1637 | continue; | |
1638 | if (expect(list_empty(&req->w.list))) { | |
1639 | req->w.mdev = mdev; | |
1640 | req->w.cb = w_restart_write; | |
1641 | drbd_queue_work(&mdev->tconn->data.work, &req->w); | |
1642 | } | |
1643 | } | |
1644 | } | |
1645 | ||
b411b363 PR |
1646 | /* e_end_block() is called via drbd_process_done_ee(). |
1647 | * this means this function only runs in the asender thread | |
1648 | */ | |
99920dc5 | 1649 | static int e_end_block(struct drbd_work *w, int cancel) |
b411b363 | 1650 | { |
8050e6d0 AG |
1651 | struct drbd_peer_request *peer_req = |
1652 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1653 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1654 | sector_t sector = peer_req->i.sector; |
99920dc5 | 1655 | int err = 0, pcmd; |
b411b363 | 1656 | |
89e58e75 | 1657 | if (mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C) { |
db830c46 | 1658 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1659 | pcmd = (mdev->state.conn >= C_SYNC_SOURCE && |
1660 | mdev->state.conn <= C_PAUSED_SYNC_T && | |
db830c46 | 1661 | peer_req->flags & EE_MAY_SET_IN_SYNC) ? |
b411b363 | 1662 | P_RS_WRITE_ACK : P_WRITE_ACK; |
99920dc5 | 1663 | err = drbd_send_ack(mdev, pcmd, peer_req); |
b411b363 | 1664 | if (pcmd == P_RS_WRITE_ACK) |
db830c46 | 1665 | drbd_set_in_sync(mdev, sector, peer_req->i.size); |
b411b363 | 1666 | } else { |
99920dc5 | 1667 | err = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1668 | /* we expect it to be marked out of sync anyways... |
1669 | * maybe assert this? */ | |
1670 | } | |
1671 | dec_unacked(mdev); | |
1672 | } | |
1673 | /* we delete from the conflict detection hash _after_ we sent out the | |
1674 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
89e58e75 | 1675 | if (mdev->tconn->net_conf->two_primaries) { |
87eeee41 | 1676 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
1677 | D_ASSERT(!drbd_interval_empty(&peer_req->i)); |
1678 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
7be8da07 AG |
1679 | if (peer_req->flags & EE_RESTART_REQUESTS) |
1680 | restart_conflicting_writes(mdev, sector, peer_req->i.size); | |
87eeee41 | 1681 | spin_unlock_irq(&mdev->tconn->req_lock); |
bb3bfe96 | 1682 | } else |
db830c46 | 1683 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1684 | |
db830c46 | 1685 | drbd_may_finish_epoch(mdev, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); |
b411b363 | 1686 | |
99920dc5 | 1687 | return err; |
b411b363 PR |
1688 | } |
1689 | ||
7be8da07 | 1690 | static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) |
b411b363 | 1691 | { |
7be8da07 | 1692 | struct drbd_conf *mdev = w->mdev; |
8050e6d0 AG |
1693 | struct drbd_peer_request *peer_req = |
1694 | container_of(w, struct drbd_peer_request, w); | |
99920dc5 | 1695 | int err; |
b411b363 | 1696 | |
99920dc5 | 1697 | err = drbd_send_ack(mdev, ack, peer_req); |
b411b363 PR |
1698 | dec_unacked(mdev); |
1699 | ||
99920dc5 | 1700 | return err; |
b411b363 PR |
1701 | } |
1702 | ||
99920dc5 | 1703 | static int e_send_discard_write(struct drbd_work *w, int unused) |
7be8da07 AG |
1704 | { |
1705 | return e_send_ack(w, P_DISCARD_WRITE); | |
1706 | } | |
1707 | ||
99920dc5 | 1708 | static int e_send_retry_write(struct drbd_work *w, int unused) |
7be8da07 AG |
1709 | { |
1710 | struct drbd_tconn *tconn = w->mdev->tconn; | |
1711 | ||
1712 | return e_send_ack(w, tconn->agreed_pro_version >= 100 ? | |
1713 | P_RETRY_WRITE : P_DISCARD_WRITE); | |
1714 | } | |
1715 | ||
3e394da1 AG |
1716 | static bool seq_greater(u32 a, u32 b) |
1717 | { | |
1718 | /* | |
1719 | * We assume 32-bit wrap-around here. | |
1720 | * For 24-bit wrap-around, we would have to shift: | |
1721 | * a <<= 8; b <<= 8; | |
1722 | */ | |
1723 | return (s32)a - (s32)b > 0; | |
1724 | } | |
1725 | ||
1726 | static u32 seq_max(u32 a, u32 b) | |
1727 | { | |
1728 | return seq_greater(a, b) ? a : b; | |
1729 | } | |
1730 | ||
7be8da07 AG |
1731 | static bool need_peer_seq(struct drbd_conf *mdev) |
1732 | { | |
1733 | struct drbd_tconn *tconn = mdev->tconn; | |
1734 | ||
1735 | /* | |
1736 | * We only need to keep track of the last packet_seq number of our peer | |
1737 | * if we are in dual-primary mode and we have the discard flag set; see | |
1738 | * handle_write_conflicts(). | |
1739 | */ | |
1740 | return tconn->net_conf->two_primaries && | |
1741 | test_bit(DISCARD_CONCURRENT, &tconn->flags); | |
1742 | } | |
1743 | ||
43ae077d | 1744 | static void update_peer_seq(struct drbd_conf *mdev, unsigned int peer_seq) |
3e394da1 | 1745 | { |
3c13b680 | 1746 | unsigned int newest_peer_seq; |
3e394da1 | 1747 | |
7be8da07 AG |
1748 | if (need_peer_seq(mdev)) { |
1749 | spin_lock(&mdev->peer_seq_lock); | |
3c13b680 LE |
1750 | newest_peer_seq = seq_max(mdev->peer_seq, peer_seq); |
1751 | mdev->peer_seq = newest_peer_seq; | |
7be8da07 | 1752 | spin_unlock(&mdev->peer_seq_lock); |
3c13b680 LE |
1753 | /* wake up only if we actually changed mdev->peer_seq */ |
1754 | if (peer_seq == newest_peer_seq) | |
7be8da07 AG |
1755 | wake_up(&mdev->seq_wait); |
1756 | } | |
3e394da1 AG |
1757 | } |
1758 | ||
b411b363 PR |
1759 | /* Called from receive_Data. |
1760 | * Synchronize packets on sock with packets on msock. | |
1761 | * | |
1762 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
1763 | * packet traveling on msock, they are still processed in the order they have | |
1764 | * been sent. | |
1765 | * | |
1766 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
1767 | * | |
1768 | * In case packet_seq is larger than mdev->peer_seq number, there are | |
1769 | * outstanding packets on the msock. We wait for them to arrive. | |
1770 | * In case we are the logically next packet, we update mdev->peer_seq | |
1771 | * ourselves. Correctly handles 32bit wrap around. | |
1772 | * | |
1773 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
1774 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
1775 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
1776 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
1777 | * | |
1778 | * returns 0 if we may process the packet, | |
1779 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
7be8da07 | 1780 | static int wait_for_and_update_peer_seq(struct drbd_conf *mdev, const u32 peer_seq) |
b411b363 PR |
1781 | { |
1782 | DEFINE_WAIT(wait); | |
b411b363 | 1783 | long timeout; |
7be8da07 AG |
1784 | int ret; |
1785 | ||
1786 | if (!need_peer_seq(mdev)) | |
1787 | return 0; | |
1788 | ||
b411b363 PR |
1789 | spin_lock(&mdev->peer_seq_lock); |
1790 | for (;;) { | |
7be8da07 AG |
1791 | if (!seq_greater(peer_seq - 1, mdev->peer_seq)) { |
1792 | mdev->peer_seq = seq_max(mdev->peer_seq, peer_seq); | |
1793 | ret = 0; | |
b411b363 | 1794 | break; |
7be8da07 | 1795 | } |
b411b363 PR |
1796 | if (signal_pending(current)) { |
1797 | ret = -ERESTARTSYS; | |
1798 | break; | |
1799 | } | |
7be8da07 | 1800 | prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); |
b411b363 | 1801 | spin_unlock(&mdev->peer_seq_lock); |
71b1c1eb AG |
1802 | timeout = mdev->tconn->net_conf->ping_timeo*HZ/10; |
1803 | timeout = schedule_timeout(timeout); | |
b411b363 | 1804 | spin_lock(&mdev->peer_seq_lock); |
7be8da07 | 1805 | if (!timeout) { |
b411b363 | 1806 | ret = -ETIMEDOUT; |
71b1c1eb | 1807 | dev_err(DEV, "Timed out waiting for missing ack packets; disconnecting\n"); |
b411b363 PR |
1808 | break; |
1809 | } | |
1810 | } | |
b411b363 | 1811 | spin_unlock(&mdev->peer_seq_lock); |
7be8da07 | 1812 | finish_wait(&mdev->seq_wait, &wait); |
b411b363 PR |
1813 | return ret; |
1814 | } | |
1815 | ||
688593c5 LE |
1816 | /* see also bio_flags_to_wire() |
1817 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
1818 | * flags and back. We may replicate to other kernel versions. */ | |
1819 | static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf) | |
76d2e7ec | 1820 | { |
688593c5 LE |
1821 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
1822 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
1823 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
1824 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
1825 | } |
1826 | ||
7be8da07 AG |
1827 | static void fail_postponed_requests(struct drbd_conf *mdev, sector_t sector, |
1828 | unsigned int size) | |
1829 | { | |
1830 | struct drbd_interval *i; | |
1831 | ||
1832 | repeat: | |
1833 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1834 | struct drbd_request *req; | |
1835 | struct bio_and_error m; | |
1836 | ||
1837 | if (!i->local) | |
1838 | continue; | |
1839 | req = container_of(i, struct drbd_request, i); | |
1840 | if (!(req->rq_state & RQ_POSTPONED)) | |
1841 | continue; | |
1842 | req->rq_state &= ~RQ_POSTPONED; | |
1843 | __req_mod(req, NEG_ACKED, &m); | |
1844 | spin_unlock_irq(&mdev->tconn->req_lock); | |
1845 | if (m.bio) | |
1846 | complete_master_bio(mdev, &m); | |
1847 | spin_lock_irq(&mdev->tconn->req_lock); | |
1848 | goto repeat; | |
1849 | } | |
1850 | } | |
1851 | ||
1852 | static int handle_write_conflicts(struct drbd_conf *mdev, | |
1853 | struct drbd_peer_request *peer_req) | |
1854 | { | |
1855 | struct drbd_tconn *tconn = mdev->tconn; | |
1856 | bool resolve_conflicts = test_bit(DISCARD_CONCURRENT, &tconn->flags); | |
1857 | sector_t sector = peer_req->i.sector; | |
1858 | const unsigned int size = peer_req->i.size; | |
1859 | struct drbd_interval *i; | |
1860 | bool equal; | |
1861 | int err; | |
1862 | ||
1863 | /* | |
1864 | * Inserting the peer request into the write_requests tree will prevent | |
1865 | * new conflicting local requests from being added. | |
1866 | */ | |
1867 | drbd_insert_interval(&mdev->write_requests, &peer_req->i); | |
1868 | ||
1869 | repeat: | |
1870 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1871 | if (i == &peer_req->i) | |
1872 | continue; | |
1873 | ||
1874 | if (!i->local) { | |
1875 | /* | |
1876 | * Our peer has sent a conflicting remote request; this | |
1877 | * should not happen in a two-node setup. Wait for the | |
1878 | * earlier peer request to complete. | |
1879 | */ | |
1880 | err = drbd_wait_misc(mdev, i); | |
1881 | if (err) | |
1882 | goto out; | |
1883 | goto repeat; | |
1884 | } | |
1885 | ||
1886 | equal = i->sector == sector && i->size == size; | |
1887 | if (resolve_conflicts) { | |
1888 | /* | |
1889 | * If the peer request is fully contained within the | |
1890 | * overlapping request, it can be discarded; otherwise, | |
1891 | * it will be retried once all overlapping requests | |
1892 | * have completed. | |
1893 | */ | |
1894 | bool discard = i->sector <= sector && i->sector + | |
1895 | (i->size >> 9) >= sector + (size >> 9); | |
1896 | ||
1897 | if (!equal) | |
1898 | dev_alert(DEV, "Concurrent writes detected: " | |
1899 | "local=%llus +%u, remote=%llus +%u, " | |
1900 | "assuming %s came first\n", | |
1901 | (unsigned long long)i->sector, i->size, | |
1902 | (unsigned long long)sector, size, | |
1903 | discard ? "local" : "remote"); | |
1904 | ||
1905 | inc_unacked(mdev); | |
1906 | peer_req->w.cb = discard ? e_send_discard_write : | |
1907 | e_send_retry_write; | |
1908 | list_add_tail(&peer_req->w.list, &mdev->done_ee); | |
1909 | wake_asender(mdev->tconn); | |
1910 | ||
1911 | err = -ENOENT; | |
1912 | goto out; | |
1913 | } else { | |
1914 | struct drbd_request *req = | |
1915 | container_of(i, struct drbd_request, i); | |
1916 | ||
1917 | if (!equal) | |
1918 | dev_alert(DEV, "Concurrent writes detected: " | |
1919 | "local=%llus +%u, remote=%llus +%u\n", | |
1920 | (unsigned long long)i->sector, i->size, | |
1921 | (unsigned long long)sector, size); | |
1922 | ||
1923 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1924 | !(req->rq_state & RQ_POSTPONED)) { | |
1925 | /* | |
1926 | * Wait for the node with the discard flag to | |
1927 | * decide if this request will be discarded or | |
1928 | * retried. Requests that are discarded will | |
1929 | * disappear from the write_requests tree. | |
1930 | * | |
1931 | * In addition, wait for the conflicting | |
1932 | * request to finish locally before submitting | |
1933 | * the conflicting peer request. | |
1934 | */ | |
1935 | err = drbd_wait_misc(mdev, &req->i); | |
1936 | if (err) { | |
1937 | _conn_request_state(mdev->tconn, | |
1938 | NS(conn, C_TIMEOUT), | |
1939 | CS_HARD); | |
1940 | fail_postponed_requests(mdev, sector, size); | |
1941 | goto out; | |
1942 | } | |
1943 | goto repeat; | |
1944 | } | |
1945 | /* | |
1946 | * Remember to restart the conflicting requests after | |
1947 | * the new peer request has completed. | |
1948 | */ | |
1949 | peer_req->flags |= EE_RESTART_REQUESTS; | |
1950 | } | |
1951 | } | |
1952 | err = 0; | |
1953 | ||
1954 | out: | |
1955 | if (err) | |
1956 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
1957 | return err; | |
1958 | } | |
1959 | ||
b411b363 | 1960 | /* mirrored write */ |
d8763023 AG |
1961 | static int receive_Data(struct drbd_conf *mdev, enum drbd_packet cmd, |
1962 | unsigned int data_size) | |
b411b363 PR |
1963 | { |
1964 | sector_t sector; | |
db830c46 | 1965 | struct drbd_peer_request *peer_req; |
e42325a5 | 1966 | struct p_data *p = &mdev->tconn->data.rbuf.data; |
7be8da07 | 1967 | u32 peer_seq = be32_to_cpu(p->seq_num); |
b411b363 PR |
1968 | int rw = WRITE; |
1969 | u32 dp_flags; | |
7be8da07 | 1970 | int err; |
b411b363 | 1971 | |
7be8da07 | 1972 | if (!get_ldev(mdev)) { |
82bc0194 AG |
1973 | int err2; |
1974 | ||
7be8da07 | 1975 | err = wait_for_and_update_peer_seq(mdev, peer_seq); |
2b2bf214 | 1976 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 | 1977 | atomic_inc(&mdev->current_epoch->epoch_size); |
82bc0194 AG |
1978 | err2 = drbd_drain_block(mdev, data_size); |
1979 | if (!err) | |
1980 | err = err2; | |
1981 | return err; | |
b411b363 PR |
1982 | } |
1983 | ||
fcefa62e AG |
1984 | /* |
1985 | * Corresponding put_ldev done either below (on various errors), or in | |
1986 | * drbd_peer_request_endio, if we successfully submit the data at the | |
1987 | * end of this function. | |
1988 | */ | |
b411b363 PR |
1989 | |
1990 | sector = be64_to_cpu(p->sector); | |
db830c46 AG |
1991 | peer_req = read_in_block(mdev, p->block_id, sector, data_size); |
1992 | if (!peer_req) { | |
b411b363 | 1993 | put_ldev(mdev); |
82bc0194 | 1994 | return -EIO; |
b411b363 PR |
1995 | } |
1996 | ||
db830c46 | 1997 | peer_req->w.cb = e_end_block; |
b411b363 | 1998 | |
688593c5 LE |
1999 | dp_flags = be32_to_cpu(p->dp_flags); |
2000 | rw |= wire_flags_to_bio(mdev, dp_flags); | |
2001 | ||
2002 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
db830c46 | 2003 | peer_req->flags |= EE_MAY_SET_IN_SYNC; |
688593c5 | 2004 | |
b411b363 | 2005 | spin_lock(&mdev->epoch_lock); |
db830c46 AG |
2006 | peer_req->epoch = mdev->current_epoch; |
2007 | atomic_inc(&peer_req->epoch->epoch_size); | |
2008 | atomic_inc(&peer_req->epoch->active); | |
b411b363 PR |
2009 | spin_unlock(&mdev->epoch_lock); |
2010 | ||
7be8da07 AG |
2011 | if (mdev->tconn->net_conf->two_primaries) { |
2012 | err = wait_for_and_update_peer_seq(mdev, peer_seq); | |
2013 | if (err) | |
b411b363 | 2014 | goto out_interrupted; |
87eeee41 | 2015 | spin_lock_irq(&mdev->tconn->req_lock); |
7be8da07 AG |
2016 | err = handle_write_conflicts(mdev, peer_req); |
2017 | if (err) { | |
2018 | spin_unlock_irq(&mdev->tconn->req_lock); | |
2019 | if (err == -ENOENT) { | |
b411b363 | 2020 | put_ldev(mdev); |
82bc0194 | 2021 | return 0; |
b411b363 | 2022 | } |
7be8da07 | 2023 | goto out_interrupted; |
b411b363 | 2024 | } |
7be8da07 AG |
2025 | } else |
2026 | spin_lock_irq(&mdev->tconn->req_lock); | |
db830c46 | 2027 | list_add(&peer_req->w.list, &mdev->active_ee); |
87eeee41 | 2028 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2029 | |
89e58e75 | 2030 | switch (mdev->tconn->net_conf->wire_protocol) { |
b411b363 PR |
2031 | case DRBD_PROT_C: |
2032 | inc_unacked(mdev); | |
2033 | /* corresponding dec_unacked() in e_end_block() | |
2034 | * respective _drbd_clear_done_ee */ | |
2035 | break; | |
2036 | case DRBD_PROT_B: | |
2037 | /* I really don't like it that the receiver thread | |
2038 | * sends on the msock, but anyways */ | |
db830c46 | 2039 | drbd_send_ack(mdev, P_RECV_ACK, peer_req); |
b411b363 PR |
2040 | break; |
2041 | case DRBD_PROT_A: | |
2042 | /* nothing to do */ | |
2043 | break; | |
2044 | } | |
2045 | ||
6719fb03 | 2046 | if (mdev->state.pdsk < D_INCONSISTENT) { |
b411b363 | 2047 | /* In case we have the only disk of the cluster, */ |
db830c46 AG |
2048 | drbd_set_out_of_sync(mdev, peer_req->i.sector, peer_req->i.size); |
2049 | peer_req->flags |= EE_CALL_AL_COMPLETE_IO; | |
2050 | peer_req->flags &= ~EE_MAY_SET_IN_SYNC; | |
2051 | drbd_al_begin_io(mdev, peer_req->i.sector); | |
b411b363 PR |
2052 | } |
2053 | ||
82bc0194 AG |
2054 | err = drbd_submit_peer_request(mdev, peer_req, rw, DRBD_FAULT_DT_WR); |
2055 | if (!err) | |
2056 | return 0; | |
b411b363 | 2057 | |
10f6d992 LE |
2058 | /* don't care for the reason here */ |
2059 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2060 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
2061 | list_del(&peer_req->w.list); |
2062 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
87eeee41 | 2063 | spin_unlock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
2064 | if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) |
2065 | drbd_al_complete_io(mdev, peer_req->i.sector); | |
22cc37a9 | 2066 | |
b411b363 | 2067 | out_interrupted: |
db830c46 | 2068 | drbd_may_finish_epoch(mdev, peer_req->epoch, EV_PUT + EV_CLEANUP); |
b411b363 | 2069 | put_ldev(mdev); |
db830c46 | 2070 | drbd_free_ee(mdev, peer_req); |
82bc0194 | 2071 | return err; |
b411b363 PR |
2072 | } |
2073 | ||
0f0601f4 LE |
2074 | /* We may throttle resync, if the lower device seems to be busy, |
2075 | * and current sync rate is above c_min_rate. | |
2076 | * | |
2077 | * To decide whether or not the lower device is busy, we use a scheme similar | |
2078 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
2079 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
2080 | * activity, it obviously is "busy". | |
2081 | * | |
2082 | * The current sync rate used here uses only the most recent two step marks, | |
2083 | * to have a short time average so we can react faster. | |
2084 | */ | |
e3555d85 | 2085 | int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector) |
0f0601f4 LE |
2086 | { |
2087 | struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; | |
2088 | unsigned long db, dt, dbdt; | |
e3555d85 | 2089 | struct lc_element *tmp; |
0f0601f4 LE |
2090 | int curr_events; |
2091 | int throttle = 0; | |
2092 | ||
2093 | /* feature disabled? */ | |
f399002e | 2094 | if (mdev->ldev->dc.c_min_rate == 0) |
0f0601f4 LE |
2095 | return 0; |
2096 | ||
e3555d85 PR |
2097 | spin_lock_irq(&mdev->al_lock); |
2098 | tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector)); | |
2099 | if (tmp) { | |
2100 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
2101 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) { | |
2102 | spin_unlock_irq(&mdev->al_lock); | |
2103 | return 0; | |
2104 | } | |
2105 | /* Do not slow down if app IO is already waiting for this extent */ | |
2106 | } | |
2107 | spin_unlock_irq(&mdev->al_lock); | |
2108 | ||
0f0601f4 LE |
2109 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
2110 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
2111 | atomic_read(&mdev->rs_sect_ev); | |
e3555d85 | 2112 | |
0f0601f4 LE |
2113 | if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { |
2114 | unsigned long rs_left; | |
2115 | int i; | |
2116 | ||
2117 | mdev->rs_last_events = curr_events; | |
2118 | ||
2119 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
2120 | * approx. */ | |
2649f080 LE |
2121 | i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
2122 | ||
2123 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) | |
2124 | rs_left = mdev->ov_left; | |
2125 | else | |
2126 | rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
0f0601f4 LE |
2127 | |
2128 | dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; | |
2129 | if (!dt) | |
2130 | dt++; | |
2131 | db = mdev->rs_mark_left[i] - rs_left; | |
2132 | dbdt = Bit2KB(db/dt); | |
2133 | ||
f399002e | 2134 | if (dbdt > mdev->ldev->dc.c_min_rate) |
0f0601f4 LE |
2135 | throttle = 1; |
2136 | } | |
2137 | return throttle; | |
2138 | } | |
2139 | ||
2140 | ||
d8763023 AG |
2141 | static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packet cmd, |
2142 | unsigned int digest_size) | |
b411b363 PR |
2143 | { |
2144 | sector_t sector; | |
2145 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
db830c46 | 2146 | struct drbd_peer_request *peer_req; |
b411b363 | 2147 | struct digest_info *di = NULL; |
b18b37be | 2148 | int size, verb; |
b411b363 | 2149 | unsigned int fault_type; |
e42325a5 | 2150 | struct p_block_req *p = &mdev->tconn->data.rbuf.block_req; |
b411b363 PR |
2151 | |
2152 | sector = be64_to_cpu(p->sector); | |
2153 | size = be32_to_cpu(p->blksize); | |
2154 | ||
c670a398 | 2155 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
2156 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
2157 | (unsigned long long)sector, size); | |
82bc0194 | 2158 | return -EINVAL; |
b411b363 PR |
2159 | } |
2160 | if (sector + (size>>9) > capacity) { | |
2161 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, | |
2162 | (unsigned long long)sector, size); | |
82bc0194 | 2163 | return -EINVAL; |
b411b363 PR |
2164 | } |
2165 | ||
2166 | if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { | |
b18b37be PR |
2167 | verb = 1; |
2168 | switch (cmd) { | |
2169 | case P_DATA_REQUEST: | |
2170 | drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); | |
2171 | break; | |
2172 | case P_RS_DATA_REQUEST: | |
2173 | case P_CSUM_RS_REQUEST: | |
2174 | case P_OV_REQUEST: | |
2175 | drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); | |
2176 | break; | |
2177 | case P_OV_REPLY: | |
2178 | verb = 0; | |
2179 | dec_rs_pending(mdev); | |
2180 | drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); | |
2181 | break; | |
2182 | default: | |
2183 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
2184 | cmdname(cmd)); | |
2185 | } | |
2186 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
b411b363 PR |
2187 | dev_err(DEV, "Can not satisfy peer's read request, " |
2188 | "no local data.\n"); | |
b18b37be | 2189 | |
a821cc4a | 2190 | /* drain possibly payload */ |
82bc0194 | 2191 | return drbd_drain_block(mdev, digest_size); |
b411b363 PR |
2192 | } |
2193 | ||
2194 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2195 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2196 | * which in turn might block on the other node at this very place. */ | |
db830c46 AG |
2197 | peer_req = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); |
2198 | if (!peer_req) { | |
b411b363 | 2199 | put_ldev(mdev); |
82bc0194 | 2200 | return -ENOMEM; |
b411b363 PR |
2201 | } |
2202 | ||
02918be2 | 2203 | switch (cmd) { |
b411b363 | 2204 | case P_DATA_REQUEST: |
db830c46 | 2205 | peer_req->w.cb = w_e_end_data_req; |
b411b363 | 2206 | fault_type = DRBD_FAULT_DT_RD; |
80a40e43 LE |
2207 | /* application IO, don't drbd_rs_begin_io */ |
2208 | goto submit; | |
2209 | ||
b411b363 | 2210 | case P_RS_DATA_REQUEST: |
db830c46 | 2211 | peer_req->w.cb = w_e_end_rsdata_req; |
b411b363 | 2212 | fault_type = DRBD_FAULT_RS_RD; |
5f9915bb LE |
2213 | /* used in the sector offset progress display */ |
2214 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
b411b363 PR |
2215 | break; |
2216 | ||
2217 | case P_OV_REPLY: | |
2218 | case P_CSUM_RS_REQUEST: | |
2219 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2220 | di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO); |
2221 | if (!di) | |
2222 | goto out_free_e; | |
2223 | ||
2224 | di->digest_size = digest_size; | |
2225 | di->digest = (((char *)di)+sizeof(struct digest_info)); | |
2226 | ||
db830c46 AG |
2227 | peer_req->digest = di; |
2228 | peer_req->flags |= EE_HAS_DIGEST; | |
c36c3ced | 2229 | |
de0ff338 | 2230 | if (drbd_recv(mdev->tconn, di->digest, digest_size) != digest_size) |
b411b363 PR |
2231 | goto out_free_e; |
2232 | ||
02918be2 | 2233 | if (cmd == P_CSUM_RS_REQUEST) { |
31890f4a | 2234 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
db830c46 | 2235 | peer_req->w.cb = w_e_end_csum_rs_req; |
5f9915bb LE |
2236 | /* used in the sector offset progress display */ |
2237 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
02918be2 | 2238 | } else if (cmd == P_OV_REPLY) { |
2649f080 LE |
2239 | /* track progress, we may need to throttle */ |
2240 | atomic_add(size >> 9, &mdev->rs_sect_in); | |
db830c46 | 2241 | peer_req->w.cb = w_e_end_ov_reply; |
b411b363 | 2242 | dec_rs_pending(mdev); |
0f0601f4 LE |
2243 | /* drbd_rs_begin_io done when we sent this request, |
2244 | * but accounting still needs to be done. */ | |
2245 | goto submit_for_resync; | |
b411b363 PR |
2246 | } |
2247 | break; | |
2248 | ||
2249 | case P_OV_REQUEST: | |
b411b363 | 2250 | if (mdev->ov_start_sector == ~(sector_t)0 && |
31890f4a | 2251 | mdev->tconn->agreed_pro_version >= 90) { |
de228bba LE |
2252 | unsigned long now = jiffies; |
2253 | int i; | |
b411b363 PR |
2254 | mdev->ov_start_sector = sector; |
2255 | mdev->ov_position = sector; | |
30b743a2 LE |
2256 | mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector); |
2257 | mdev->rs_total = mdev->ov_left; | |
de228bba LE |
2258 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
2259 | mdev->rs_mark_left[i] = mdev->ov_left; | |
2260 | mdev->rs_mark_time[i] = now; | |
2261 | } | |
b411b363 PR |
2262 | dev_info(DEV, "Online Verify start sector: %llu\n", |
2263 | (unsigned long long)sector); | |
2264 | } | |
db830c46 | 2265 | peer_req->w.cb = w_e_end_ov_req; |
b411b363 | 2266 | fault_type = DRBD_FAULT_RS_RD; |
b411b363 PR |
2267 | break; |
2268 | ||
b411b363 PR |
2269 | default: |
2270 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
02918be2 | 2271 | cmdname(cmd)); |
b411b363 | 2272 | fault_type = DRBD_FAULT_MAX; |
80a40e43 | 2273 | goto out_free_e; |
b411b363 PR |
2274 | } |
2275 | ||
0f0601f4 LE |
2276 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2277 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2278 | * Various places in the resync start and stop logic assume resync | |
2279 | * requests are processed in order, requeuing this on the worker thread | |
2280 | * introduces a bunch of new code for synchronization between threads. | |
2281 | * | |
2282 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2283 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2284 | * for application writes for the same time. For now, just throttle | |
2285 | * here, where the rest of the code expects the receiver to sleep for | |
2286 | * a while, anyways. | |
2287 | */ | |
2288 | ||
2289 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2290 | * this defers syncer requests for some time, before letting at least | |
2291 | * on request through. The resync controller on the receiving side | |
2292 | * will adapt to the incoming rate accordingly. | |
2293 | * | |
2294 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2295 | * we would also throttle its application reads. | |
2296 | * In that case, throttling is done on the SyncTarget only. | |
2297 | */ | |
e3555d85 PR |
2298 | if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector)) |
2299 | schedule_timeout_uninterruptible(HZ/10); | |
2300 | if (drbd_rs_begin_io(mdev, sector)) | |
80a40e43 | 2301 | goto out_free_e; |
b411b363 | 2302 | |
0f0601f4 LE |
2303 | submit_for_resync: |
2304 | atomic_add(size >> 9, &mdev->rs_sect_ev); | |
2305 | ||
80a40e43 | 2306 | submit: |
b411b363 | 2307 | inc_unacked(mdev); |
87eeee41 | 2308 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2309 | list_add_tail(&peer_req->w.list, &mdev->read_ee); |
87eeee41 | 2310 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2311 | |
fbe29dec | 2312 | if (drbd_submit_peer_request(mdev, peer_req, READ, fault_type) == 0) |
82bc0194 | 2313 | return 0; |
b411b363 | 2314 | |
10f6d992 LE |
2315 | /* don't care for the reason here */ |
2316 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2317 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2318 | list_del(&peer_req->w.list); |
87eeee41 | 2319 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 LE |
2320 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ |
2321 | ||
b411b363 | 2322 | out_free_e: |
b411b363 | 2323 | put_ldev(mdev); |
db830c46 | 2324 | drbd_free_ee(mdev, peer_req); |
82bc0194 | 2325 | return -EIO; |
b411b363 PR |
2326 | } |
2327 | ||
2328 | static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) | |
2329 | { | |
2330 | int self, peer, rv = -100; | |
2331 | unsigned long ch_self, ch_peer; | |
2332 | ||
2333 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2334 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2335 | ||
2336 | ch_peer = mdev->p_uuid[UI_SIZE]; | |
2337 | ch_self = mdev->comm_bm_set; | |
2338 | ||
89e58e75 | 2339 | switch (mdev->tconn->net_conf->after_sb_0p) { |
b411b363 PR |
2340 | case ASB_CONSENSUS: |
2341 | case ASB_DISCARD_SECONDARY: | |
2342 | case ASB_CALL_HELPER: | |
2343 | dev_err(DEV, "Configuration error.\n"); | |
2344 | break; | |
2345 | case ASB_DISCONNECT: | |
2346 | break; | |
2347 | case ASB_DISCARD_YOUNGER_PRI: | |
2348 | if (self == 0 && peer == 1) { | |
2349 | rv = -1; | |
2350 | break; | |
2351 | } | |
2352 | if (self == 1 && peer == 0) { | |
2353 | rv = 1; | |
2354 | break; | |
2355 | } | |
2356 | /* Else fall through to one of the other strategies... */ | |
2357 | case ASB_DISCARD_OLDER_PRI: | |
2358 | if (self == 0 && peer == 1) { | |
2359 | rv = 1; | |
2360 | break; | |
2361 | } | |
2362 | if (self == 1 && peer == 0) { | |
2363 | rv = -1; | |
2364 | break; | |
2365 | } | |
2366 | /* Else fall through to one of the other strategies... */ | |
ad19bf6e | 2367 | dev_warn(DEV, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2368 | "Using discard-least-changes instead\n"); |
2369 | case ASB_DISCARD_ZERO_CHG: | |
2370 | if (ch_peer == 0 && ch_self == 0) { | |
25703f83 | 2371 | rv = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags) |
b411b363 PR |
2372 | ? -1 : 1; |
2373 | break; | |
2374 | } else { | |
2375 | if (ch_peer == 0) { rv = 1; break; } | |
2376 | if (ch_self == 0) { rv = -1; break; } | |
2377 | } | |
89e58e75 | 2378 | if (mdev->tconn->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) |
b411b363 PR |
2379 | break; |
2380 | case ASB_DISCARD_LEAST_CHG: | |
2381 | if (ch_self < ch_peer) | |
2382 | rv = -1; | |
2383 | else if (ch_self > ch_peer) | |
2384 | rv = 1; | |
2385 | else /* ( ch_self == ch_peer ) */ | |
2386 | /* Well, then use something else. */ | |
25703f83 | 2387 | rv = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags) |
b411b363 PR |
2388 | ? -1 : 1; |
2389 | break; | |
2390 | case ASB_DISCARD_LOCAL: | |
2391 | rv = -1; | |
2392 | break; | |
2393 | case ASB_DISCARD_REMOTE: | |
2394 | rv = 1; | |
2395 | } | |
2396 | ||
2397 | return rv; | |
2398 | } | |
2399 | ||
2400 | static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) | |
2401 | { | |
6184ea21 | 2402 | int hg, rv = -100; |
b411b363 | 2403 | |
89e58e75 | 2404 | switch (mdev->tconn->net_conf->after_sb_1p) { |
b411b363 PR |
2405 | case ASB_DISCARD_YOUNGER_PRI: |
2406 | case ASB_DISCARD_OLDER_PRI: | |
2407 | case ASB_DISCARD_LEAST_CHG: | |
2408 | case ASB_DISCARD_LOCAL: | |
2409 | case ASB_DISCARD_REMOTE: | |
2410 | dev_err(DEV, "Configuration error.\n"); | |
2411 | break; | |
2412 | case ASB_DISCONNECT: | |
2413 | break; | |
2414 | case ASB_CONSENSUS: | |
2415 | hg = drbd_asb_recover_0p(mdev); | |
2416 | if (hg == -1 && mdev->state.role == R_SECONDARY) | |
2417 | rv = hg; | |
2418 | if (hg == 1 && mdev->state.role == R_PRIMARY) | |
2419 | rv = hg; | |
2420 | break; | |
2421 | case ASB_VIOLENTLY: | |
2422 | rv = drbd_asb_recover_0p(mdev); | |
2423 | break; | |
2424 | case ASB_DISCARD_SECONDARY: | |
2425 | return mdev->state.role == R_PRIMARY ? 1 : -1; | |
2426 | case ASB_CALL_HELPER: | |
2427 | hg = drbd_asb_recover_0p(mdev); | |
2428 | if (hg == -1 && mdev->state.role == R_PRIMARY) { | |
bb437946 AG |
2429 | enum drbd_state_rv rv2; |
2430 | ||
2431 | drbd_set_role(mdev, R_SECONDARY, 0); | |
b411b363 PR |
2432 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2433 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2434 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2435 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2436 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2437 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2438 | } else { | |
2439 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2440 | rv = hg; | |
2441 | } | |
2442 | } else | |
2443 | rv = hg; | |
2444 | } | |
2445 | ||
2446 | return rv; | |
2447 | } | |
2448 | ||
2449 | static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) | |
2450 | { | |
6184ea21 | 2451 | int hg, rv = -100; |
b411b363 | 2452 | |
89e58e75 | 2453 | switch (mdev->tconn->net_conf->after_sb_2p) { |
b411b363 PR |
2454 | case ASB_DISCARD_YOUNGER_PRI: |
2455 | case ASB_DISCARD_OLDER_PRI: | |
2456 | case ASB_DISCARD_LEAST_CHG: | |
2457 | case ASB_DISCARD_LOCAL: | |
2458 | case ASB_DISCARD_REMOTE: | |
2459 | case ASB_CONSENSUS: | |
2460 | case ASB_DISCARD_SECONDARY: | |
2461 | dev_err(DEV, "Configuration error.\n"); | |
2462 | break; | |
2463 | case ASB_VIOLENTLY: | |
2464 | rv = drbd_asb_recover_0p(mdev); | |
2465 | break; | |
2466 | case ASB_DISCONNECT: | |
2467 | break; | |
2468 | case ASB_CALL_HELPER: | |
2469 | hg = drbd_asb_recover_0p(mdev); | |
2470 | if (hg == -1) { | |
bb437946 AG |
2471 | enum drbd_state_rv rv2; |
2472 | ||
b411b363 PR |
2473 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2474 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2475 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2476 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2477 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2478 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2479 | } else { | |
2480 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2481 | rv = hg; | |
2482 | } | |
2483 | } else | |
2484 | rv = hg; | |
2485 | } | |
2486 | ||
2487 | return rv; | |
2488 | } | |
2489 | ||
2490 | static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, | |
2491 | u64 bits, u64 flags) | |
2492 | { | |
2493 | if (!uuid) { | |
2494 | dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); | |
2495 | return; | |
2496 | } | |
2497 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", | |
2498 | text, | |
2499 | (unsigned long long)uuid[UI_CURRENT], | |
2500 | (unsigned long long)uuid[UI_BITMAP], | |
2501 | (unsigned long long)uuid[UI_HISTORY_START], | |
2502 | (unsigned long long)uuid[UI_HISTORY_END], | |
2503 | (unsigned long long)bits, | |
2504 | (unsigned long long)flags); | |
2505 | } | |
2506 | ||
2507 | /* | |
2508 | 100 after split brain try auto recover | |
2509 | 2 C_SYNC_SOURCE set BitMap | |
2510 | 1 C_SYNC_SOURCE use BitMap | |
2511 | 0 no Sync | |
2512 | -1 C_SYNC_TARGET use BitMap | |
2513 | -2 C_SYNC_TARGET set BitMap | |
2514 | -100 after split brain, disconnect | |
2515 | -1000 unrelated data | |
4a23f264 PR |
2516 | -1091 requires proto 91 |
2517 | -1096 requires proto 96 | |
b411b363 PR |
2518 | */ |
2519 | static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) | |
2520 | { | |
2521 | u64 self, peer; | |
2522 | int i, j; | |
2523 | ||
2524 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2525 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2526 | ||
2527 | *rule_nr = 10; | |
2528 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2529 | return 0; | |
2530 | ||
2531 | *rule_nr = 20; | |
2532 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2533 | peer != UUID_JUST_CREATED) | |
2534 | return -2; | |
2535 | ||
2536 | *rule_nr = 30; | |
2537 | if (self != UUID_JUST_CREATED && | |
2538 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2539 | return 2; | |
2540 | ||
2541 | if (self == peer) { | |
2542 | int rct, dc; /* roles at crash time */ | |
2543 | ||
2544 | if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { | |
2545 | ||
31890f4a | 2546 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2547 | return -1091; |
b411b363 PR |
2548 | |
2549 | if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && | |
2550 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
2551 | dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); | |
2552 | drbd_uuid_set_bm(mdev, 0UL); | |
2553 | ||
2554 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2555 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2556 | *rule_nr = 34; | |
2557 | } else { | |
2558 | dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); | |
2559 | *rule_nr = 36; | |
2560 | } | |
2561 | ||
2562 | return 1; | |
2563 | } | |
2564 | ||
2565 | if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { | |
2566 | ||
31890f4a | 2567 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2568 | return -1091; |
b411b363 PR |
2569 | |
2570 | if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && | |
2571 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
2572 | dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); | |
2573 | ||
2574 | mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; | |
2575 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; | |
2576 | mdev->p_uuid[UI_BITMAP] = 0UL; | |
2577 | ||
2578 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2579 | *rule_nr = 35; | |
2580 | } else { | |
2581 | dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); | |
2582 | *rule_nr = 37; | |
2583 | } | |
2584 | ||
2585 | return -1; | |
2586 | } | |
2587 | ||
2588 | /* Common power [off|failure] */ | |
2589 | rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + | |
2590 | (mdev->p_uuid[UI_FLAGS] & 2); | |
2591 | /* lowest bit is set when we were primary, | |
2592 | * next bit (weight 2) is set when peer was primary */ | |
2593 | *rule_nr = 40; | |
2594 | ||
2595 | switch (rct) { | |
2596 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2597 | case 1: /* self_pri && !peer_pri */ return 1; | |
2598 | case 2: /* !self_pri && peer_pri */ return -1; | |
2599 | case 3: /* self_pri && peer_pri */ | |
25703f83 | 2600 | dc = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags); |
b411b363 PR |
2601 | return dc ? -1 : 1; |
2602 | } | |
2603 | } | |
2604 | ||
2605 | *rule_nr = 50; | |
2606 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2607 | if (self == peer) | |
2608 | return -1; | |
2609 | ||
2610 | *rule_nr = 51; | |
2611 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2612 | if (self == peer) { | |
31890f4a | 2613 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2614 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == |
2615 | (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
2616 | peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2617 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2618 | resync as sync source modifications of the peer's UUIDs. */ | |
2619 | ||
31890f4a | 2620 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2621 | return -1091; |
b411b363 PR |
2622 | |
2623 | mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; | |
2624 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 PR |
2625 | |
2626 | dev_info(DEV, "Did not got last syncUUID packet, corrected:\n"); | |
2627 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2628 | ||
b411b363 PR |
2629 | return -1; |
2630 | } | |
2631 | } | |
2632 | ||
2633 | *rule_nr = 60; | |
2634 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2635 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2636 | peer = mdev->p_uuid[i] & ~((u64)1); | |
2637 | if (self == peer) | |
2638 | return -2; | |
2639 | } | |
2640 | ||
2641 | *rule_nr = 70; | |
2642 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2643 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2644 | if (self == peer) | |
2645 | return 1; | |
2646 | ||
2647 | *rule_nr = 71; | |
2648 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2649 | if (self == peer) { | |
31890f4a | 2650 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2651 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == |
2652 | (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
2653 | self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2654 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2655 | resync as sync source modifications of our UUIDs. */ | |
2656 | ||
31890f4a | 2657 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2658 | return -1091; |
b411b363 PR |
2659 | |
2660 | _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); | |
2661 | _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); | |
2662 | ||
4a23f264 | 2663 | dev_info(DEV, "Last syncUUID did not get through, corrected:\n"); |
b411b363 PR |
2664 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, |
2665 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2666 | ||
2667 | return 1; | |
2668 | } | |
2669 | } | |
2670 | ||
2671 | ||
2672 | *rule_nr = 80; | |
d8c2a36b | 2673 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 PR |
2674 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
2675 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2676 | if (self == peer) | |
2677 | return 2; | |
2678 | } | |
2679 | ||
2680 | *rule_nr = 90; | |
2681 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2682 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2683 | if (self == peer && self != ((u64)0)) | |
2684 | return 100; | |
2685 | ||
2686 | *rule_nr = 100; | |
2687 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2688 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2689 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { | |
2690 | peer = mdev->p_uuid[j] & ~((u64)1); | |
2691 | if (self == peer) | |
2692 | return -100; | |
2693 | } | |
2694 | } | |
2695 | ||
2696 | return -1000; | |
2697 | } | |
2698 | ||
2699 | /* drbd_sync_handshake() returns the new conn state on success, or | |
2700 | CONN_MASK (-1) on failure. | |
2701 | */ | |
2702 | static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, | |
2703 | enum drbd_disk_state peer_disk) __must_hold(local) | |
2704 | { | |
2705 | int hg, rule_nr; | |
2706 | enum drbd_conns rv = C_MASK; | |
2707 | enum drbd_disk_state mydisk; | |
2708 | ||
2709 | mydisk = mdev->state.disk; | |
2710 | if (mydisk == D_NEGOTIATING) | |
2711 | mydisk = mdev->new_state_tmp.disk; | |
2712 | ||
2713 | dev_info(DEV, "drbd_sync_handshake:\n"); | |
2714 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); | |
2715 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, | |
2716 | mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2717 | ||
2718 | hg = drbd_uuid_compare(mdev, &rule_nr); | |
2719 | ||
2720 | dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); | |
2721 | ||
2722 | if (hg == -1000) { | |
2723 | dev_alert(DEV, "Unrelated data, aborting!\n"); | |
2724 | return C_MASK; | |
2725 | } | |
4a23f264 PR |
2726 | if (hg < -1000) { |
2727 | dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); | |
b411b363 PR |
2728 | return C_MASK; |
2729 | } | |
2730 | ||
2731 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
2732 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
2733 | int f = (hg == -100) || abs(hg) == 2; | |
2734 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
2735 | if (f) | |
2736 | hg = hg*2; | |
2737 | dev_info(DEV, "Becoming sync %s due to disk states.\n", | |
2738 | hg > 0 ? "source" : "target"); | |
2739 | } | |
2740 | ||
3a11a487 AG |
2741 | if (abs(hg) == 100) |
2742 | drbd_khelper(mdev, "initial-split-brain"); | |
2743 | ||
89e58e75 | 2744 | if (hg == 100 || (hg == -100 && mdev->tconn->net_conf->always_asbp)) { |
b411b363 PR |
2745 | int pcount = (mdev->state.role == R_PRIMARY) |
2746 | + (peer_role == R_PRIMARY); | |
2747 | int forced = (hg == -100); | |
2748 | ||
2749 | switch (pcount) { | |
2750 | case 0: | |
2751 | hg = drbd_asb_recover_0p(mdev); | |
2752 | break; | |
2753 | case 1: | |
2754 | hg = drbd_asb_recover_1p(mdev); | |
2755 | break; | |
2756 | case 2: | |
2757 | hg = drbd_asb_recover_2p(mdev); | |
2758 | break; | |
2759 | } | |
2760 | if (abs(hg) < 100) { | |
2761 | dev_warn(DEV, "Split-Brain detected, %d primaries, " | |
2762 | "automatically solved. Sync from %s node\n", | |
2763 | pcount, (hg < 0) ? "peer" : "this"); | |
2764 | if (forced) { | |
2765 | dev_warn(DEV, "Doing a full sync, since" | |
2766 | " UUIDs where ambiguous.\n"); | |
2767 | hg = hg*2; | |
2768 | } | |
2769 | } | |
2770 | } | |
2771 | ||
2772 | if (hg == -100) { | |
89e58e75 | 2773 | if (mdev->tconn->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 | 2774 | hg = -1; |
89e58e75 | 2775 | if (!mdev->tconn->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 PR |
2776 | hg = 1; |
2777 | ||
2778 | if (abs(hg) < 100) | |
2779 | dev_warn(DEV, "Split-Brain detected, manually solved. " | |
2780 | "Sync from %s node\n", | |
2781 | (hg < 0) ? "peer" : "this"); | |
2782 | } | |
2783 | ||
2784 | if (hg == -100) { | |
580b9767 LE |
2785 | /* FIXME this log message is not correct if we end up here |
2786 | * after an attempted attach on a diskless node. | |
2787 | * We just refuse to attach -- well, we drop the "connection" | |
2788 | * to that disk, in a way... */ | |
3a11a487 | 2789 | dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); |
b411b363 PR |
2790 | drbd_khelper(mdev, "split-brain"); |
2791 | return C_MASK; | |
2792 | } | |
2793 | ||
2794 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
2795 | dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); | |
2796 | return C_MASK; | |
2797 | } | |
2798 | ||
2799 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
2800 | mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { | |
89e58e75 | 2801 | switch (mdev->tconn->net_conf->rr_conflict) { |
b411b363 PR |
2802 | case ASB_CALL_HELPER: |
2803 | drbd_khelper(mdev, "pri-lost"); | |
2804 | /* fall through */ | |
2805 | case ASB_DISCONNECT: | |
2806 | dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); | |
2807 | return C_MASK; | |
2808 | case ASB_VIOLENTLY: | |
2809 | dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" | |
2810 | "assumption\n"); | |
2811 | } | |
2812 | } | |
2813 | ||
8169e41b | 2814 | if (mdev->tconn->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->tconn->flags)) { |
cf14c2e9 PR |
2815 | if (hg == 0) |
2816 | dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); | |
2817 | else | |
2818 | dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", | |
2819 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), | |
2820 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
2821 | return C_MASK; | |
2822 | } | |
2823 | ||
b411b363 PR |
2824 | if (abs(hg) >= 2) { |
2825 | dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); | |
20ceb2b2 LE |
2826 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
2827 | BM_LOCKED_SET_ALLOWED)) | |
b411b363 PR |
2828 | return C_MASK; |
2829 | } | |
2830 | ||
2831 | if (hg > 0) { /* become sync source. */ | |
2832 | rv = C_WF_BITMAP_S; | |
2833 | } else if (hg < 0) { /* become sync target */ | |
2834 | rv = C_WF_BITMAP_T; | |
2835 | } else { | |
2836 | rv = C_CONNECTED; | |
2837 | if (drbd_bm_total_weight(mdev)) { | |
2838 | dev_info(DEV, "No resync, but %lu bits in bitmap!\n", | |
2839 | drbd_bm_total_weight(mdev)); | |
2840 | } | |
2841 | } | |
2842 | ||
2843 | return rv; | |
2844 | } | |
2845 | ||
2846 | /* returns 1 if invalid */ | |
2847 | static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) | |
2848 | { | |
2849 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
2850 | if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || | |
2851 | (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) | |
2852 | return 0; | |
2853 | ||
2854 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
2855 | if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || | |
2856 | self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) | |
2857 | return 1; | |
2858 | ||
2859 | /* everything else is valid if they are equal on both sides. */ | |
2860 | if (peer == self) | |
2861 | return 0; | |
2862 | ||
2863 | /* everything es is invalid. */ | |
2864 | return 1; | |
2865 | } | |
2866 | ||
7204624c | 2867 | static int receive_protocol(struct drbd_tconn *tconn, enum drbd_packet cmd, |
d8763023 | 2868 | unsigned int data_size) |
b411b363 | 2869 | { |
7204624c | 2870 | struct p_protocol *p = &tconn->data.rbuf.protocol; |
b411b363 | 2871 | int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
cf14c2e9 | 2872 | int p_want_lose, p_two_primaries, cf; |
b411b363 PR |
2873 | char p_integrity_alg[SHARED_SECRET_MAX] = ""; |
2874 | ||
b411b363 PR |
2875 | p_proto = be32_to_cpu(p->protocol); |
2876 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
2877 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
2878 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 2879 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 PR |
2880 | cf = be32_to_cpu(p->conn_flags); |
2881 | p_want_lose = cf & CF_WANT_LOSE; | |
2882 | ||
7204624c | 2883 | clear_bit(CONN_DRY_RUN, &tconn->flags); |
cf14c2e9 PR |
2884 | |
2885 | if (cf & CF_DRY_RUN) | |
7204624c | 2886 | set_bit(CONN_DRY_RUN, &tconn->flags); |
b411b363 | 2887 | |
7204624c PR |
2888 | if (p_proto != tconn->net_conf->wire_protocol) { |
2889 | conn_err(tconn, "incompatible communication protocols\n"); | |
b411b363 PR |
2890 | goto disconnect; |
2891 | } | |
2892 | ||
7204624c PR |
2893 | if (cmp_after_sb(p_after_sb_0p, tconn->net_conf->after_sb_0p)) { |
2894 | conn_err(tconn, "incompatible after-sb-0pri settings\n"); | |
b411b363 PR |
2895 | goto disconnect; |
2896 | } | |
2897 | ||
7204624c PR |
2898 | if (cmp_after_sb(p_after_sb_1p, tconn->net_conf->after_sb_1p)) { |
2899 | conn_err(tconn, "incompatible after-sb-1pri settings\n"); | |
b411b363 PR |
2900 | goto disconnect; |
2901 | } | |
2902 | ||
7204624c PR |
2903 | if (cmp_after_sb(p_after_sb_2p, tconn->net_conf->after_sb_2p)) { |
2904 | conn_err(tconn, "incompatible after-sb-2pri settings\n"); | |
b411b363 PR |
2905 | goto disconnect; |
2906 | } | |
2907 | ||
7204624c PR |
2908 | if (p_want_lose && tconn->net_conf->want_lose) { |
2909 | conn_err(tconn, "both sides have the 'want_lose' flag set\n"); | |
b411b363 PR |
2910 | goto disconnect; |
2911 | } | |
2912 | ||
7204624c PR |
2913 | if (p_two_primaries != tconn->net_conf->two_primaries) { |
2914 | conn_err(tconn, "incompatible setting of the two-primaries options\n"); | |
b411b363 PR |
2915 | goto disconnect; |
2916 | } | |
2917 | ||
7204624c PR |
2918 | if (tconn->agreed_pro_version >= 87) { |
2919 | unsigned char *my_alg = tconn->net_conf->integrity_alg; | |
82bc0194 | 2920 | int err; |
b411b363 | 2921 | |
82bc0194 AG |
2922 | err = drbd_recv_all(tconn, p_integrity_alg, data_size); |
2923 | if (err) | |
2924 | return err; | |
b411b363 PR |
2925 | |
2926 | p_integrity_alg[SHARED_SECRET_MAX-1] = 0; | |
2927 | if (strcmp(p_integrity_alg, my_alg)) { | |
7204624c | 2928 | conn_err(tconn, "incompatible setting of the data-integrity-alg\n"); |
b411b363 PR |
2929 | goto disconnect; |
2930 | } | |
7204624c | 2931 | conn_info(tconn, "data-integrity-alg: %s\n", |
b411b363 PR |
2932 | my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); |
2933 | } | |
2934 | ||
82bc0194 | 2935 | return 0; |
b411b363 PR |
2936 | |
2937 | disconnect: | |
7204624c | 2938 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 2939 | return -EIO; |
b411b363 PR |
2940 | } |
2941 | ||
2942 | /* helper function | |
2943 | * input: alg name, feature name | |
2944 | * return: NULL (alg name was "") | |
2945 | * ERR_PTR(error) if something goes wrong | |
2946 | * or the crypto hash ptr, if it worked out ok. */ | |
2947 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, | |
2948 | const char *alg, const char *name) | |
2949 | { | |
2950 | struct crypto_hash *tfm; | |
2951 | ||
2952 | if (!alg[0]) | |
2953 | return NULL; | |
2954 | ||
2955 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
2956 | if (IS_ERR(tfm)) { | |
2957 | dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", | |
2958 | alg, name, PTR_ERR(tfm)); | |
2959 | return tfm; | |
2960 | } | |
2961 | if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { | |
2962 | crypto_free_hash(tfm); | |
2963 | dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); | |
2964 | return ERR_PTR(-EINVAL); | |
2965 | } | |
2966 | return tfm; | |
2967 | } | |
2968 | ||
d8763023 AG |
2969 | static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packet cmd, |
2970 | unsigned int packet_size) | |
b411b363 | 2971 | { |
e42325a5 | 2972 | struct p_rs_param_95 *p = &mdev->tconn->data.rbuf.rs_param_95; |
b411b363 PR |
2973 | unsigned int header_size, data_size, exp_max_sz; |
2974 | struct crypto_hash *verify_tfm = NULL; | |
2975 | struct crypto_hash *csums_tfm = NULL; | |
31890f4a | 2976 | const int apv = mdev->tconn->agreed_pro_version; |
778f271d PR |
2977 | int *rs_plan_s = NULL; |
2978 | int fifo_size = 0; | |
82bc0194 | 2979 | int err; |
b411b363 PR |
2980 | |
2981 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
2982 | : apv == 88 ? sizeof(struct p_rs_param) | |
2983 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
2984 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
2985 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 2986 | |
02918be2 | 2987 | if (packet_size > exp_max_sz) { |
b411b363 | 2988 | dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
02918be2 | 2989 | packet_size, exp_max_sz); |
82bc0194 | 2990 | return -EIO; |
b411b363 PR |
2991 | } |
2992 | ||
2993 | if (apv <= 88) { | |
257d0af6 | 2994 | header_size = sizeof(struct p_rs_param) - sizeof(struct p_header); |
02918be2 | 2995 | data_size = packet_size - header_size; |
8e26f9cc | 2996 | } else if (apv <= 94) { |
257d0af6 | 2997 | header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header); |
02918be2 | 2998 | data_size = packet_size - header_size; |
b411b363 | 2999 | D_ASSERT(data_size == 0); |
8e26f9cc | 3000 | } else { |
257d0af6 | 3001 | header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header); |
02918be2 | 3002 | data_size = packet_size - header_size; |
b411b363 PR |
3003 | D_ASSERT(data_size == 0); |
3004 | } | |
3005 | ||
3006 | /* initialize verify_alg and csums_alg */ | |
3007 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
3008 | ||
82bc0194 AG |
3009 | err = drbd_recv_all(mdev->tconn, &p->head.payload, header_size); |
3010 | if (err) | |
3011 | return err; | |
b411b363 | 3012 | |
f399002e LE |
3013 | if (get_ldev(mdev)) { |
3014 | mdev->ldev->dc.resync_rate = be32_to_cpu(p->rate); | |
3015 | put_ldev(mdev); | |
3016 | } | |
b411b363 PR |
3017 | |
3018 | if (apv >= 88) { | |
3019 | if (apv == 88) { | |
3020 | if (data_size > SHARED_SECRET_MAX) { | |
3021 | dev_err(DEV, "verify-alg too long, " | |
3022 | "peer wants %u, accepting only %u byte\n", | |
3023 | data_size, SHARED_SECRET_MAX); | |
82bc0194 | 3024 | return -EIO; |
b411b363 PR |
3025 | } |
3026 | ||
82bc0194 AG |
3027 | err = drbd_recv_all(mdev->tconn, p->verify_alg, data_size); |
3028 | if (err) | |
3029 | return err; | |
b411b363 PR |
3030 | |
3031 | /* we expect NUL terminated string */ | |
3032 | /* but just in case someone tries to be evil */ | |
3033 | D_ASSERT(p->verify_alg[data_size-1] == 0); | |
3034 | p->verify_alg[data_size-1] = 0; | |
3035 | ||
3036 | } else /* apv >= 89 */ { | |
3037 | /* we still expect NUL terminated strings */ | |
3038 | /* but just in case someone tries to be evil */ | |
3039 | D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); | |
3040 | D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
3041 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; | |
3042 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
3043 | } | |
3044 | ||
f399002e | 3045 | if (strcmp(mdev->tconn->net_conf->verify_alg, p->verify_alg)) { |
b411b363 PR |
3046 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { |
3047 | dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", | |
f399002e | 3048 | mdev->tconn->net_conf->verify_alg, p->verify_alg); |
b411b363 PR |
3049 | goto disconnect; |
3050 | } | |
3051 | verify_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3052 | p->verify_alg, "verify-alg"); | |
3053 | if (IS_ERR(verify_tfm)) { | |
3054 | verify_tfm = NULL; | |
3055 | goto disconnect; | |
3056 | } | |
3057 | } | |
3058 | ||
f399002e | 3059 | if (apv >= 89 && strcmp(mdev->tconn->net_conf->csums_alg, p->csums_alg)) { |
b411b363 PR |
3060 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { |
3061 | dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", | |
f399002e | 3062 | mdev->tconn->net_conf->csums_alg, p->csums_alg); |
b411b363 PR |
3063 | goto disconnect; |
3064 | } | |
3065 | csums_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3066 | p->csums_alg, "csums-alg"); | |
3067 | if (IS_ERR(csums_tfm)) { | |
3068 | csums_tfm = NULL; | |
3069 | goto disconnect; | |
3070 | } | |
3071 | } | |
3072 | ||
f399002e LE |
3073 | if (apv > 94 && get_ldev(mdev)) { |
3074 | mdev->ldev->dc.resync_rate = be32_to_cpu(p->rate); | |
3075 | mdev->ldev->dc.c_plan_ahead = be32_to_cpu(p->c_plan_ahead); | |
3076 | mdev->ldev->dc.c_delay_target = be32_to_cpu(p->c_delay_target); | |
3077 | mdev->ldev->dc.c_fill_target = be32_to_cpu(p->c_fill_target); | |
3078 | mdev->ldev->dc.c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d | 3079 | |
f399002e | 3080 | fifo_size = (mdev->ldev->dc.c_plan_ahead * 10 * SLEEP_TIME) / HZ; |
778f271d PR |
3081 | if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { |
3082 | rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); | |
3083 | if (!rs_plan_s) { | |
3084 | dev_err(DEV, "kmalloc of fifo_buffer failed"); | |
f399002e | 3085 | put_ldev(mdev); |
778f271d PR |
3086 | goto disconnect; |
3087 | } | |
3088 | } | |
f399002e | 3089 | put_ldev(mdev); |
8e26f9cc | 3090 | } |
b411b363 PR |
3091 | |
3092 | spin_lock(&mdev->peer_seq_lock); | |
3093 | /* lock against drbd_nl_syncer_conf() */ | |
3094 | if (verify_tfm) { | |
f399002e LE |
3095 | strcpy(mdev->tconn->net_conf->verify_alg, p->verify_alg); |
3096 | mdev->tconn->net_conf->verify_alg_len = strlen(p->verify_alg) + 1; | |
3097 | crypto_free_hash(mdev->tconn->verify_tfm); | |
3098 | mdev->tconn->verify_tfm = verify_tfm; | |
b411b363 PR |
3099 | dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); |
3100 | } | |
3101 | if (csums_tfm) { | |
f399002e LE |
3102 | strcpy(mdev->tconn->net_conf->csums_alg, p->csums_alg); |
3103 | mdev->tconn->net_conf->csums_alg_len = strlen(p->csums_alg) + 1; | |
3104 | crypto_free_hash(mdev->tconn->csums_tfm); | |
3105 | mdev->tconn->csums_tfm = csums_tfm; | |
b411b363 PR |
3106 | dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); |
3107 | } | |
778f271d PR |
3108 | if (fifo_size != mdev->rs_plan_s.size) { |
3109 | kfree(mdev->rs_plan_s.values); | |
3110 | mdev->rs_plan_s.values = rs_plan_s; | |
3111 | mdev->rs_plan_s.size = fifo_size; | |
3112 | mdev->rs_planed = 0; | |
3113 | } | |
b411b363 PR |
3114 | spin_unlock(&mdev->peer_seq_lock); |
3115 | } | |
82bc0194 | 3116 | return 0; |
b411b363 | 3117 | |
b411b363 PR |
3118 | disconnect: |
3119 | /* just for completeness: actually not needed, | |
3120 | * as this is not reached if csums_tfm was ok. */ | |
3121 | crypto_free_hash(csums_tfm); | |
3122 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
3123 | crypto_free_hash(verify_tfm); | |
38fa9988 | 3124 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3125 | return -EIO; |
b411b363 PR |
3126 | } |
3127 | ||
b411b363 PR |
3128 | /* warn if the arguments differ by more than 12.5% */ |
3129 | static void warn_if_differ_considerably(struct drbd_conf *mdev, | |
3130 | const char *s, sector_t a, sector_t b) | |
3131 | { | |
3132 | sector_t d; | |
3133 | if (a == 0 || b == 0) | |
3134 | return; | |
3135 | d = (a > b) ? (a - b) : (b - a); | |
3136 | if (d > (a>>3) || d > (b>>3)) | |
3137 | dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, | |
3138 | (unsigned long long)a, (unsigned long long)b); | |
3139 | } | |
3140 | ||
d8763023 AG |
3141 | static int receive_sizes(struct drbd_conf *mdev, enum drbd_packet cmd, |
3142 | unsigned int data_size) | |
b411b363 | 3143 | { |
e42325a5 | 3144 | struct p_sizes *p = &mdev->tconn->data.rbuf.sizes; |
b411b363 | 3145 | enum determine_dev_size dd = unchanged; |
b411b363 PR |
3146 | sector_t p_size, p_usize, my_usize; |
3147 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 3148 | enum dds_flags ddsf; |
b411b363 | 3149 | |
b411b363 PR |
3150 | p_size = be64_to_cpu(p->d_size); |
3151 | p_usize = be64_to_cpu(p->u_size); | |
3152 | ||
b411b363 PR |
3153 | /* just store the peer's disk size for now. |
3154 | * we still need to figure out whether we accept that. */ | |
3155 | mdev->p_size = p_size; | |
3156 | ||
b411b363 PR |
3157 | if (get_ldev(mdev)) { |
3158 | warn_if_differ_considerably(mdev, "lower level device sizes", | |
3159 | p_size, drbd_get_max_capacity(mdev->ldev)); | |
3160 | warn_if_differ_considerably(mdev, "user requested size", | |
3161 | p_usize, mdev->ldev->dc.disk_size); | |
3162 | ||
3163 | /* if this is the first connect, or an otherwise expected | |
3164 | * param exchange, choose the minimum */ | |
3165 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
3166 | p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, | |
3167 | p_usize); | |
3168 | ||
3169 | my_usize = mdev->ldev->dc.disk_size; | |
3170 | ||
3171 | if (mdev->ldev->dc.disk_size != p_usize) { | |
3172 | mdev->ldev->dc.disk_size = p_usize; | |
3173 | dev_info(DEV, "Peer sets u_size to %lu sectors\n", | |
3174 | (unsigned long)mdev->ldev->dc.disk_size); | |
3175 | } | |
3176 | ||
3177 | /* Never shrink a device with usable data during connect. | |
3178 | But allow online shrinking if we are connected. */ | |
a393db6f | 3179 | if (drbd_new_dev_size(mdev, mdev->ldev, 0) < |
b411b363 PR |
3180 | drbd_get_capacity(mdev->this_bdev) && |
3181 | mdev->state.disk >= D_OUTDATED && | |
3182 | mdev->state.conn < C_CONNECTED) { | |
3183 | dev_err(DEV, "The peer's disk size is too small!\n"); | |
38fa9988 | 3184 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
3185 | mdev->ldev->dc.disk_size = my_usize; |
3186 | put_ldev(mdev); | |
82bc0194 | 3187 | return -EIO; |
b411b363 PR |
3188 | } |
3189 | put_ldev(mdev); | |
3190 | } | |
b411b363 | 3191 | |
e89b591c | 3192 | ddsf = be16_to_cpu(p->dds_flags); |
b411b363 | 3193 | if (get_ldev(mdev)) { |
24c4830c | 3194 | dd = drbd_determine_dev_size(mdev, ddsf); |
b411b363 PR |
3195 | put_ldev(mdev); |
3196 | if (dd == dev_size_error) | |
82bc0194 | 3197 | return -EIO; |
b411b363 PR |
3198 | drbd_md_sync(mdev); |
3199 | } else { | |
3200 | /* I am diskless, need to accept the peer's size. */ | |
3201 | drbd_set_my_capacity(mdev, p_size); | |
3202 | } | |
3203 | ||
99432fcc PR |
3204 | mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
3205 | drbd_reconsider_max_bio_size(mdev); | |
3206 | ||
b411b363 PR |
3207 | if (get_ldev(mdev)) { |
3208 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { | |
3209 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); | |
3210 | ldsc = 1; | |
3211 | } | |
3212 | ||
b411b363 PR |
3213 | put_ldev(mdev); |
3214 | } | |
3215 | ||
3216 | if (mdev->state.conn > C_WF_REPORT_PARAMS) { | |
3217 | if (be64_to_cpu(p->c_size) != | |
3218 | drbd_get_capacity(mdev->this_bdev) || ldsc) { | |
3219 | /* we have different sizes, probably peer | |
3220 | * needs to know my new size... */ | |
e89b591c | 3221 | drbd_send_sizes(mdev, 0, ddsf); |
b411b363 PR |
3222 | } |
3223 | if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || | |
3224 | (dd == grew && mdev->state.conn == C_CONNECTED)) { | |
3225 | if (mdev->state.pdsk >= D_INCONSISTENT && | |
e89b591c PR |
3226 | mdev->state.disk >= D_INCONSISTENT) { |
3227 | if (ddsf & DDSF_NO_RESYNC) | |
3228 | dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); | |
3229 | else | |
3230 | resync_after_online_grow(mdev); | |
3231 | } else | |
b411b363 PR |
3232 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); |
3233 | } | |
3234 | } | |
3235 | ||
82bc0194 | 3236 | return 0; |
b411b363 PR |
3237 | } |
3238 | ||
d8763023 AG |
3239 | static int receive_uuids(struct drbd_conf *mdev, enum drbd_packet cmd, |
3240 | unsigned int data_size) | |
b411b363 | 3241 | { |
e42325a5 | 3242 | struct p_uuids *p = &mdev->tconn->data.rbuf.uuids; |
b411b363 | 3243 | u64 *p_uuid; |
62b0da3a | 3244 | int i, updated_uuids = 0; |
b411b363 | 3245 | |
b411b363 PR |
3246 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
3247 | ||
3248 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3249 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3250 | ||
3251 | kfree(mdev->p_uuid); | |
3252 | mdev->p_uuid = p_uuid; | |
3253 | ||
3254 | if (mdev->state.conn < C_CONNECTED && | |
3255 | mdev->state.disk < D_INCONSISTENT && | |
3256 | mdev->state.role == R_PRIMARY && | |
3257 | (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
3258 | dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", | |
3259 | (unsigned long long)mdev->ed_uuid); | |
38fa9988 | 3260 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3261 | return -EIO; |
b411b363 PR |
3262 | } |
3263 | ||
3264 | if (get_ldev(mdev)) { | |
3265 | int skip_initial_sync = | |
3266 | mdev->state.conn == C_CONNECTED && | |
31890f4a | 3267 | mdev->tconn->agreed_pro_version >= 90 && |
b411b363 PR |
3268 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && |
3269 | (p_uuid[UI_FLAGS] & 8); | |
3270 | if (skip_initial_sync) { | |
3271 | dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); | |
3272 | drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, | |
20ceb2b2 LE |
3273 | "clear_n_write from receive_uuids", |
3274 | BM_LOCKED_TEST_ALLOWED); | |
b411b363 PR |
3275 | _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); |
3276 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
3277 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
3278 | CS_VERBOSE, NULL); | |
3279 | drbd_md_sync(mdev); | |
62b0da3a | 3280 | updated_uuids = 1; |
b411b363 PR |
3281 | } |
3282 | put_ldev(mdev); | |
18a50fa2 PR |
3283 | } else if (mdev->state.disk < D_INCONSISTENT && |
3284 | mdev->state.role == R_PRIMARY) { | |
3285 | /* I am a diskless primary, the peer just created a new current UUID | |
3286 | for me. */ | |
62b0da3a | 3287 | updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
b411b363 PR |
3288 | } |
3289 | ||
3290 | /* Before we test for the disk state, we should wait until an eventually | |
3291 | ongoing cluster wide state change is finished. That is important if | |
3292 | we are primary and are detaching from our disk. We need to see the | |
3293 | new disk state... */ | |
8410da8f PR |
3294 | mutex_lock(mdev->state_mutex); |
3295 | mutex_unlock(mdev->state_mutex); | |
b411b363 | 3296 | if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) |
62b0da3a LE |
3297 | updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
3298 | ||
3299 | if (updated_uuids) | |
3300 | drbd_print_uuids(mdev, "receiver updated UUIDs to"); | |
b411b363 | 3301 | |
82bc0194 | 3302 | return 0; |
b411b363 PR |
3303 | } |
3304 | ||
3305 | /** | |
3306 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3307 | * @ps: The state as seen by the peer. | |
3308 | */ | |
3309 | static union drbd_state convert_state(union drbd_state ps) | |
3310 | { | |
3311 | union drbd_state ms; | |
3312 | ||
3313 | static enum drbd_conns c_tab[] = { | |
3314 | [C_CONNECTED] = C_CONNECTED, | |
3315 | ||
3316 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3317 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3318 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3319 | [C_VERIFY_S] = C_VERIFY_T, | |
3320 | [C_MASK] = C_MASK, | |
3321 | }; | |
3322 | ||
3323 | ms.i = ps.i; | |
3324 | ||
3325 | ms.conn = c_tab[ps.conn]; | |
3326 | ms.peer = ps.role; | |
3327 | ms.role = ps.peer; | |
3328 | ms.pdsk = ps.disk; | |
3329 | ms.disk = ps.pdsk; | |
3330 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3331 | ||
3332 | return ms; | |
3333 | } | |
3334 | ||
d8763023 AG |
3335 | static int receive_req_state(struct drbd_conf *mdev, enum drbd_packet cmd, |
3336 | unsigned int data_size) | |
b411b363 | 3337 | { |
e42325a5 | 3338 | struct p_req_state *p = &mdev->tconn->data.rbuf.req_state; |
b411b363 | 3339 | union drbd_state mask, val; |
bf885f8a | 3340 | enum drbd_state_rv rv; |
b411b363 | 3341 | |
b411b363 PR |
3342 | mask.i = be32_to_cpu(p->mask); |
3343 | val.i = be32_to_cpu(p->val); | |
3344 | ||
25703f83 | 3345 | if (test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags) && |
8410da8f | 3346 | mutex_is_locked(mdev->state_mutex)) { |
b411b363 | 3347 | drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); |
82bc0194 | 3348 | return 0; |
b411b363 PR |
3349 | } |
3350 | ||
3351 | mask = convert_state(mask); | |
3352 | val = convert_state(val); | |
3353 | ||
dfafcc8a PR |
3354 | rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); |
3355 | drbd_send_sr_reply(mdev, rv); | |
b411b363 | 3356 | |
b411b363 PR |
3357 | drbd_md_sync(mdev); |
3358 | ||
82bc0194 | 3359 | return 0; |
b411b363 PR |
3360 | } |
3361 | ||
dfafcc8a PR |
3362 | static int receive_req_conn_state(struct drbd_tconn *tconn, enum drbd_packet cmd, |
3363 | unsigned int data_size) | |
3364 | { | |
3365 | struct p_req_state *p = &tconn->data.rbuf.req_state; | |
3366 | union drbd_state mask, val; | |
3367 | enum drbd_state_rv rv; | |
3368 | ||
3369 | mask.i = be32_to_cpu(p->mask); | |
3370 | val.i = be32_to_cpu(p->val); | |
3371 | ||
3372 | if (test_bit(DISCARD_CONCURRENT, &tconn->flags) && | |
3373 | mutex_is_locked(&tconn->cstate_mutex)) { | |
3374 | conn_send_sr_reply(tconn, SS_CONCURRENT_ST_CHG); | |
82bc0194 | 3375 | return 0; |
dfafcc8a PR |
3376 | } |
3377 | ||
3378 | mask = convert_state(mask); | |
3379 | val = convert_state(val); | |
3380 | ||
3381 | rv = conn_request_state(tconn, mask, val, CS_VERBOSE | CS_LOCAL_ONLY); | |
3382 | conn_send_sr_reply(tconn, rv); | |
3383 | ||
82bc0194 | 3384 | return 0; |
dfafcc8a PR |
3385 | } |
3386 | ||
d8763023 AG |
3387 | static int receive_state(struct drbd_conf *mdev, enum drbd_packet cmd, |
3388 | unsigned int data_size) | |
b411b363 | 3389 | { |
e42325a5 | 3390 | struct p_state *p = &mdev->tconn->data.rbuf.state; |
4ac4aada | 3391 | union drbd_state os, ns, peer_state; |
b411b363 | 3392 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 3393 | enum chg_state_flags cs_flags; |
b411b363 PR |
3394 | int rv; |
3395 | ||
b411b363 PR |
3396 | peer_state.i = be32_to_cpu(p->state); |
3397 | ||
3398 | real_peer_disk = peer_state.disk; | |
3399 | if (peer_state.disk == D_NEGOTIATING) { | |
3400 | real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; | |
3401 | dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); | |
3402 | } | |
3403 | ||
87eeee41 | 3404 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 3405 | retry: |
4ac4aada | 3406 | os = ns = mdev->state; |
87eeee41 | 3407 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 3408 | |
e9ef7bb6 LE |
3409 | /* peer says his disk is uptodate, while we think it is inconsistent, |
3410 | * and this happens while we think we have a sync going on. */ | |
3411 | if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE && | |
3412 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { | |
3413 | /* If we are (becoming) SyncSource, but peer is still in sync | |
3414 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
3415 | * will change to inconsistent once the peer reaches active | |
3416 | * syncing states. | |
3417 | * It may have changed syncer-paused flags, however, so we | |
3418 | * cannot ignore this completely. */ | |
3419 | if (peer_state.conn > C_CONNECTED && | |
3420 | peer_state.conn < C_SYNC_SOURCE) | |
3421 | real_peer_disk = D_INCONSISTENT; | |
3422 | ||
3423 | /* if peer_state changes to connected at the same time, | |
3424 | * it explicitly notifies us that it finished resync. | |
3425 | * Maybe we should finish it up, too? */ | |
3426 | else if (os.conn >= C_SYNC_SOURCE && | |
3427 | peer_state.conn == C_CONNECTED) { | |
3428 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) | |
3429 | drbd_resync_finished(mdev); | |
82bc0194 | 3430 | return 0; |
e9ef7bb6 LE |
3431 | } |
3432 | } | |
3433 | ||
3434 | /* peer says his disk is inconsistent, while we think it is uptodate, | |
3435 | * and this happens while the peer still thinks we have a sync going on, | |
3436 | * but we think we are already done with the sync. | |
3437 | * We ignore this to avoid flapping pdsk. | |
3438 | * This should not happen, if the peer is a recent version of drbd. */ | |
3439 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
3440 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
3441 | real_peer_disk = D_UP_TO_DATE; | |
3442 | ||
4ac4aada LE |
3443 | if (ns.conn == C_WF_REPORT_PARAMS) |
3444 | ns.conn = C_CONNECTED; | |
b411b363 | 3445 | |
67531718 PR |
3446 | if (peer_state.conn == C_AHEAD) |
3447 | ns.conn = C_BEHIND; | |
3448 | ||
b411b363 PR |
3449 | if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && |
3450 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3451 | int cr; /* consider resync */ | |
3452 | ||
3453 | /* if we established a new connection */ | |
4ac4aada | 3454 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
3455 | /* if we had an established connection |
3456 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 3457 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 3458 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 3459 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
3460 | /* if we have both been inconsistent, and the peer has been |
3461 | * forced to be UpToDate with --overwrite-data */ | |
3462 | cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); | |
3463 | /* if we had been plain connected, and the admin requested to | |
3464 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 3465 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
3466 | (peer_state.conn >= C_STARTING_SYNC_S && |
3467 | peer_state.conn <= C_WF_BITMAP_T)); | |
3468 | ||
3469 | if (cr) | |
4ac4aada | 3470 | ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); |
b411b363 PR |
3471 | |
3472 | put_ldev(mdev); | |
4ac4aada LE |
3473 | if (ns.conn == C_MASK) { |
3474 | ns.conn = C_CONNECTED; | |
b411b363 | 3475 | if (mdev->state.disk == D_NEGOTIATING) { |
82f59cc6 | 3476 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
b411b363 PR |
3477 | } else if (peer_state.disk == D_NEGOTIATING) { |
3478 | dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); | |
3479 | peer_state.disk = D_DISKLESS; | |
580b9767 | 3480 | real_peer_disk = D_DISKLESS; |
b411b363 | 3481 | } else { |
8169e41b | 3482 | if (test_and_clear_bit(CONN_DRY_RUN, &mdev->tconn->flags)) |
82bc0194 | 3483 | return -EIO; |
4ac4aada | 3484 | D_ASSERT(os.conn == C_WF_REPORT_PARAMS); |
38fa9988 | 3485 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3486 | return -EIO; |
b411b363 PR |
3487 | } |
3488 | } | |
3489 | } | |
3490 | ||
87eeee41 | 3491 | spin_lock_irq(&mdev->tconn->req_lock); |
4ac4aada | 3492 | if (mdev->state.i != os.i) |
b411b363 PR |
3493 | goto retry; |
3494 | clear_bit(CONSIDER_RESYNC, &mdev->flags); | |
b411b363 PR |
3495 | ns.peer = peer_state.role; |
3496 | ns.pdsk = real_peer_disk; | |
3497 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 3498 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b411b363 | 3499 | ns.disk = mdev->new_state_tmp.disk; |
4ac4aada LE |
3500 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
3501 | if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && | |
481c6f50 | 3502 | test_bit(NEW_CUR_UUID, &mdev->flags)) { |
8554df1c | 3503 | /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this |
481c6f50 | 3504 | for temporal network outages! */ |
87eeee41 | 3505 | spin_unlock_irq(&mdev->tconn->req_lock); |
481c6f50 | 3506 | dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); |
2f5cdd0b | 3507 | tl_clear(mdev->tconn); |
481c6f50 PR |
3508 | drbd_uuid_new_current(mdev); |
3509 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
38fa9988 | 3510 | conn_request_state(mdev->tconn, NS2(conn, C_PROTOCOL_ERROR, susp, 0), CS_HARD); |
82bc0194 | 3511 | return -EIO; |
481c6f50 | 3512 | } |
65d922c3 | 3513 | rv = _drbd_set_state(mdev, ns, cs_flags, NULL); |
b411b363 | 3514 | ns = mdev->state; |
87eeee41 | 3515 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3516 | |
3517 | if (rv < SS_SUCCESS) { | |
38fa9988 | 3518 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
82bc0194 | 3519 | return -EIO; |
b411b363 PR |
3520 | } |
3521 | ||
4ac4aada LE |
3522 | if (os.conn > C_WF_REPORT_PARAMS) { |
3523 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
3524 | peer_state.disk != D_NEGOTIATING ) { |
3525 | /* we want resync, peer has not yet decided to sync... */ | |
3526 | /* Nowadays only used when forcing a node into primary role and | |
3527 | setting its disk to UpToDate with that */ | |
3528 | drbd_send_uuids(mdev); | |
3529 | drbd_send_state(mdev); | |
3530 | } | |
3531 | } | |
3532 | ||
89e58e75 | 3533 | mdev->tconn->net_conf->want_lose = 0; |
b411b363 PR |
3534 | |
3535 | drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ | |
3536 | ||
82bc0194 | 3537 | return 0; |
b411b363 PR |
3538 | } |
3539 | ||
d8763023 AG |
3540 | static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packet cmd, |
3541 | unsigned int data_size) | |
b411b363 | 3542 | { |
e42325a5 | 3543 | struct p_rs_uuid *p = &mdev->tconn->data.rbuf.rs_uuid; |
b411b363 PR |
3544 | |
3545 | wait_event(mdev->misc_wait, | |
3546 | mdev->state.conn == C_WF_SYNC_UUID || | |
c4752ef1 | 3547 | mdev->state.conn == C_BEHIND || |
b411b363 PR |
3548 | mdev->state.conn < C_CONNECTED || |
3549 | mdev->state.disk < D_NEGOTIATING); | |
3550 | ||
3551 | /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ | |
3552 | ||
b411b363 PR |
3553 | /* Here the _drbd_uuid_ functions are right, current should |
3554 | _not_ be rotated into the history */ | |
3555 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3556 | _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); | |
3557 | _drbd_uuid_set(mdev, UI_BITMAP, 0UL); | |
3558 | ||
62b0da3a | 3559 | drbd_print_uuids(mdev, "updated sync uuid"); |
b411b363 PR |
3560 | drbd_start_resync(mdev, C_SYNC_TARGET); |
3561 | ||
3562 | put_ldev(mdev); | |
3563 | } else | |
3564 | dev_err(DEV, "Ignoring SyncUUID packet!\n"); | |
3565 | ||
82bc0194 | 3566 | return 0; |
b411b363 PR |
3567 | } |
3568 | ||
2c46407d AG |
3569 | /** |
3570 | * receive_bitmap_plain | |
3571 | * | |
3572 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3573 | * code upon failure. | |
3574 | */ | |
3575 | static int | |
02918be2 PR |
3576 | receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size, |
3577 | unsigned long *buffer, struct bm_xfer_ctx *c) | |
b411b363 PR |
3578 | { |
3579 | unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); | |
3580 | unsigned want = num_words * sizeof(long); | |
2c46407d | 3581 | int err; |
b411b363 | 3582 | |
02918be2 PR |
3583 | if (want != data_size) { |
3584 | dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size); | |
2c46407d | 3585 | return -EIO; |
b411b363 PR |
3586 | } |
3587 | if (want == 0) | |
2c46407d | 3588 | return 0; |
82bc0194 AG |
3589 | err = drbd_recv_all(mdev->tconn, buffer, want); |
3590 | if (err) | |
2c46407d | 3591 | return err; |
b411b363 PR |
3592 | |
3593 | drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); | |
3594 | ||
3595 | c->word_offset += num_words; | |
3596 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
3597 | if (c->bit_offset > c->bm_bits) | |
3598 | c->bit_offset = c->bm_bits; | |
3599 | ||
2c46407d | 3600 | return 1; |
b411b363 PR |
3601 | } |
3602 | ||
2c46407d AG |
3603 | /** |
3604 | * recv_bm_rle_bits | |
3605 | * | |
3606 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3607 | * code upon failure. | |
3608 | */ | |
3609 | static int | |
b411b363 PR |
3610 | recv_bm_rle_bits(struct drbd_conf *mdev, |
3611 | struct p_compressed_bm *p, | |
c6d25cfe PR |
3612 | struct bm_xfer_ctx *c, |
3613 | unsigned int len) | |
b411b363 PR |
3614 | { |
3615 | struct bitstream bs; | |
3616 | u64 look_ahead; | |
3617 | u64 rl; | |
3618 | u64 tmp; | |
3619 | unsigned long s = c->bit_offset; | |
3620 | unsigned long e; | |
b411b363 PR |
3621 | int toggle = DCBP_get_start(p); |
3622 | int have; | |
3623 | int bits; | |
3624 | ||
3625 | bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p)); | |
3626 | ||
3627 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
3628 | if (bits < 0) | |
2c46407d | 3629 | return -EIO; |
b411b363 PR |
3630 | |
3631 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
3632 | bits = vli_decode_bits(&rl, look_ahead); | |
3633 | if (bits <= 0) | |
2c46407d | 3634 | return -EIO; |
b411b363 PR |
3635 | |
3636 | if (toggle) { | |
3637 | e = s + rl -1; | |
3638 | if (e >= c->bm_bits) { | |
3639 | dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); | |
2c46407d | 3640 | return -EIO; |
b411b363 PR |
3641 | } |
3642 | _drbd_bm_set_bits(mdev, s, e); | |
3643 | } | |
3644 | ||
3645 | if (have < bits) { | |
3646 | dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", | |
3647 | have, bits, look_ahead, | |
3648 | (unsigned int)(bs.cur.b - p->code), | |
3649 | (unsigned int)bs.buf_len); | |
2c46407d | 3650 | return -EIO; |
b411b363 PR |
3651 | } |
3652 | look_ahead >>= bits; | |
3653 | have -= bits; | |
3654 | ||
3655 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
3656 | if (bits < 0) | |
2c46407d | 3657 | return -EIO; |
b411b363 PR |
3658 | look_ahead |= tmp << have; |
3659 | have += bits; | |
3660 | } | |
3661 | ||
3662 | c->bit_offset = s; | |
3663 | bm_xfer_ctx_bit_to_word_offset(c); | |
3664 | ||
2c46407d | 3665 | return (s != c->bm_bits); |
b411b363 PR |
3666 | } |
3667 | ||
2c46407d AG |
3668 | /** |
3669 | * decode_bitmap_c | |
3670 | * | |
3671 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3672 | * code upon failure. | |
3673 | */ | |
3674 | static int | |
b411b363 PR |
3675 | decode_bitmap_c(struct drbd_conf *mdev, |
3676 | struct p_compressed_bm *p, | |
c6d25cfe PR |
3677 | struct bm_xfer_ctx *c, |
3678 | unsigned int len) | |
b411b363 PR |
3679 | { |
3680 | if (DCBP_get_code(p) == RLE_VLI_Bits) | |
c6d25cfe | 3681 | return recv_bm_rle_bits(mdev, p, c, len); |
b411b363 PR |
3682 | |
3683 | /* other variants had been implemented for evaluation, | |
3684 | * but have been dropped as this one turned out to be "best" | |
3685 | * during all our tests. */ | |
3686 | ||
3687 | dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); | |
38fa9988 | 3688 | conn_request_state(mdev->tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
2c46407d | 3689 | return -EIO; |
b411b363 PR |
3690 | } |
3691 | ||
3692 | void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
3693 | const char *direction, struct bm_xfer_ctx *c) | |
3694 | { | |
3695 | /* what would it take to transfer it "plaintext" */ | |
c012949a | 3696 | unsigned plain = sizeof(struct p_header) * |
b411b363 PR |
3697 | ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) |
3698 | + c->bm_words * sizeof(long); | |
3699 | unsigned total = c->bytes[0] + c->bytes[1]; | |
3700 | unsigned r; | |
3701 | ||
3702 | /* total can not be zero. but just in case: */ | |
3703 | if (total == 0) | |
3704 | return; | |
3705 | ||
3706 | /* don't report if not compressed */ | |
3707 | if (total >= plain) | |
3708 | return; | |
3709 | ||
3710 | /* total < plain. check for overflow, still */ | |
3711 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
3712 | : (1000 * total / plain); | |
3713 | ||
3714 | if (r > 1000) | |
3715 | r = 1000; | |
3716 | ||
3717 | r = 1000 - r; | |
3718 | dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " | |
3719 | "total %u; compression: %u.%u%%\n", | |
3720 | direction, | |
3721 | c->bytes[1], c->packets[1], | |
3722 | c->bytes[0], c->packets[0], | |
3723 | total, r/10, r % 10); | |
3724 | } | |
3725 | ||
3726 | /* Since we are processing the bitfield from lower addresses to higher, | |
3727 | it does not matter if the process it in 32 bit chunks or 64 bit | |
3728 | chunks as long as it is little endian. (Understand it as byte stream, | |
3729 | beginning with the lowest byte...) If we would use big endian | |
3730 | we would need to process it from the highest address to the lowest, | |
3731 | in order to be agnostic to the 32 vs 64 bits issue. | |
3732 | ||
3733 | returns 0 on failure, 1 if we successfully received it. */ | |
d8763023 AG |
3734 | static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packet cmd, |
3735 | unsigned int data_size) | |
b411b363 PR |
3736 | { |
3737 | struct bm_xfer_ctx c; | |
3738 | void *buffer; | |
2c46407d | 3739 | int err; |
257d0af6 | 3740 | struct p_header *h = &mdev->tconn->data.rbuf.header; |
77351055 | 3741 | struct packet_info pi; |
b411b363 | 3742 | |
20ceb2b2 LE |
3743 | drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
3744 | /* you are supposed to send additional out-of-sync information | |
3745 | * if you actually set bits during this phase */ | |
b411b363 PR |
3746 | |
3747 | /* maybe we should use some per thread scratch page, | |
3748 | * and allocate that during initial device creation? */ | |
3749 | buffer = (unsigned long *) __get_free_page(GFP_NOIO); | |
3750 | if (!buffer) { | |
3751 | dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); | |
82bc0194 | 3752 | err = -ENOMEM; |
b411b363 PR |
3753 | goto out; |
3754 | } | |
3755 | ||
3756 | c = (struct bm_xfer_ctx) { | |
3757 | .bm_bits = drbd_bm_bits(mdev), | |
3758 | .bm_words = drbd_bm_words(mdev), | |
3759 | }; | |
3760 | ||
2c46407d | 3761 | for(;;) { |
02918be2 | 3762 | if (cmd == P_BITMAP) { |
2c46407d | 3763 | err = receive_bitmap_plain(mdev, data_size, buffer, &c); |
02918be2 | 3764 | } else if (cmd == P_COMPRESSED_BITMAP) { |
b411b363 PR |
3765 | /* MAYBE: sanity check that we speak proto >= 90, |
3766 | * and the feature is enabled! */ | |
3767 | struct p_compressed_bm *p; | |
3768 | ||
02918be2 | 3769 | if (data_size > BM_PACKET_PAYLOAD_BYTES) { |
b411b363 | 3770 | dev_err(DEV, "ReportCBitmap packet too large\n"); |
82bc0194 | 3771 | err = -EIO; |
b411b363 PR |
3772 | goto out; |
3773 | } | |
3774 | /* use the page buff */ | |
3775 | p = buffer; | |
3776 | memcpy(p, h, sizeof(*h)); | |
82bc0194 AG |
3777 | err = drbd_recv_all(mdev->tconn, p->head.payload, data_size); |
3778 | if (err) | |
3779 | goto out; | |
004352fa LE |
3780 | if (data_size <= (sizeof(*p) - sizeof(p->head))) { |
3781 | dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size); | |
82bc0194 | 3782 | err = -EIO; |
78fcbdae | 3783 | goto out; |
b411b363 | 3784 | } |
c6d25cfe | 3785 | err = decode_bitmap_c(mdev, p, &c, data_size); |
b411b363 | 3786 | } else { |
02918be2 | 3787 | dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd); |
82bc0194 | 3788 | err = -EIO; |
b411b363 PR |
3789 | goto out; |
3790 | } | |
3791 | ||
02918be2 | 3792 | c.packets[cmd == P_BITMAP]++; |
257d0af6 | 3793 | c.bytes[cmd == P_BITMAP] += sizeof(struct p_header) + data_size; |
b411b363 | 3794 | |
2c46407d AG |
3795 | if (err <= 0) { |
3796 | if (err < 0) | |
3797 | goto out; | |
b411b363 | 3798 | break; |
2c46407d | 3799 | } |
82bc0194 AG |
3800 | err = drbd_recv_header(mdev->tconn, &pi); |
3801 | if (err) | |
b411b363 | 3802 | goto out; |
77351055 PR |
3803 | cmd = pi.cmd; |
3804 | data_size = pi.size; | |
2c46407d | 3805 | } |
b411b363 PR |
3806 | |
3807 | INFO_bm_xfer_stats(mdev, "receive", &c); | |
3808 | ||
3809 | if (mdev->state.conn == C_WF_BITMAP_T) { | |
de1f8e4a AG |
3810 | enum drbd_state_rv rv; |
3811 | ||
82bc0194 AG |
3812 | err = drbd_send_bitmap(mdev); |
3813 | if (err) | |
b411b363 PR |
3814 | goto out; |
3815 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
de1f8e4a AG |
3816 | rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
3817 | D_ASSERT(rv == SS_SUCCESS); | |
b411b363 PR |
3818 | } else if (mdev->state.conn != C_WF_BITMAP_S) { |
3819 | /* admin may have requested C_DISCONNECTING, | |
3820 | * other threads may have noticed network errors */ | |
3821 | dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", | |
3822 | drbd_conn_str(mdev->state.conn)); | |
3823 | } | |
82bc0194 | 3824 | err = 0; |
b411b363 | 3825 | |
b411b363 | 3826 | out: |
20ceb2b2 | 3827 | drbd_bm_unlock(mdev); |
82bc0194 | 3828 | if (!err && mdev->state.conn == C_WF_BITMAP_S) |
b411b363 PR |
3829 | drbd_start_resync(mdev, C_SYNC_SOURCE); |
3830 | free_page((unsigned long) buffer); | |
82bc0194 | 3831 | return err; |
b411b363 PR |
3832 | } |
3833 | ||
2de876ef | 3834 | static int _tconn_receive_skip(struct drbd_tconn *tconn, unsigned int data_size) |
b411b363 PR |
3835 | { |
3836 | /* TODO zero copy sink :) */ | |
3837 | static char sink[128]; | |
3838 | int size, want, r; | |
3839 | ||
02918be2 | 3840 | size = data_size; |
b411b363 PR |
3841 | while (size > 0) { |
3842 | want = min_t(int, size, sizeof(sink)); | |
2de876ef PR |
3843 | r = drbd_recv(tconn, sink, want); |
3844 | if (r <= 0) | |
841ce241 | 3845 | break; |
b411b363 PR |
3846 | size -= r; |
3847 | } | |
82bc0194 | 3848 | return size ? -EIO : 0; |
b411b363 PR |
3849 | } |
3850 | ||
2de876ef PR |
3851 | static int receive_skip(struct drbd_conf *mdev, enum drbd_packet cmd, |
3852 | unsigned int data_size) | |
3853 | { | |
3854 | dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n", | |
3855 | cmd, data_size); | |
3856 | ||
3857 | return _tconn_receive_skip(mdev->tconn, data_size); | |
3858 | } | |
3859 | ||
3860 | static int tconn_receive_skip(struct drbd_tconn *tconn, enum drbd_packet cmd, unsigned int data_size) | |
3861 | { | |
3862 | conn_warn(tconn, "skipping packet for non existing volume type %d, l: %d!\n", | |
3863 | cmd, data_size); | |
3864 | ||
3865 | return _tconn_receive_skip(tconn, data_size); | |
3866 | } | |
3867 | ||
d8763023 AG |
3868 | static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packet cmd, |
3869 | unsigned int data_size) | |
0ced55a3 | 3870 | { |
e7f52dfb LE |
3871 | /* Make sure we've acked all the TCP data associated |
3872 | * with the data requests being unplugged */ | |
e42325a5 | 3873 | drbd_tcp_quickack(mdev->tconn->data.socket); |
0ced55a3 | 3874 | |
82bc0194 | 3875 | return 0; |
0ced55a3 PR |
3876 | } |
3877 | ||
d8763023 AG |
3878 | static int receive_out_of_sync(struct drbd_conf *mdev, enum drbd_packet cmd, |
3879 | unsigned int data_size) | |
73a01a18 | 3880 | { |
e42325a5 | 3881 | struct p_block_desc *p = &mdev->tconn->data.rbuf.block_desc; |
73a01a18 | 3882 | |
f735e363 LE |
3883 | switch (mdev->state.conn) { |
3884 | case C_WF_SYNC_UUID: | |
3885 | case C_WF_BITMAP_T: | |
3886 | case C_BEHIND: | |
3887 | break; | |
3888 | default: | |
3889 | dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", | |
3890 | drbd_conn_str(mdev->state.conn)); | |
3891 | } | |
3892 | ||
73a01a18 PR |
3893 | drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
3894 | ||
82bc0194 | 3895 | return 0; |
73a01a18 PR |
3896 | } |
3897 | ||
02918be2 PR |
3898 | struct data_cmd { |
3899 | int expect_payload; | |
3900 | size_t pkt_size; | |
a4fbda8e | 3901 | enum mdev_or_conn fa_type; /* first argument's type */ |
d9ae84e7 PR |
3902 | union { |
3903 | int (*mdev_fn)(struct drbd_conf *, enum drbd_packet cmd, | |
3904 | unsigned int to_receive); | |
3905 | int (*conn_fn)(struct drbd_tconn *, enum drbd_packet cmd, | |
3906 | unsigned int to_receive); | |
3907 | }; | |
02918be2 PR |
3908 | }; |
3909 | ||
3910 | static struct data_cmd drbd_cmd_handler[] = { | |
d9ae84e7 PR |
3911 | [P_DATA] = { 1, sizeof(struct p_data), MDEV, { receive_Data } }, |
3912 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), MDEV, { receive_DataReply } }, | |
3913 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), MDEV, { receive_RSDataReply } } , | |
3914 | [P_BARRIER] = { 0, sizeof(struct p_barrier), MDEV, { receive_Barrier } } , | |
3915 | [P_BITMAP] = { 1, sizeof(struct p_header), MDEV, { receive_bitmap } } , | |
3916 | [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header), MDEV, { receive_bitmap } } , | |
3917 | [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header), MDEV, { receive_UnplugRemote } }, | |
3918 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), MDEV, { receive_DataRequest } }, | |
3919 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), MDEV, { receive_DataRequest } }, | |
3920 | [P_SYNC_PARAM] = { 1, sizeof(struct p_header), MDEV, { receive_SyncParam } }, | |
3921 | [P_SYNC_PARAM89] = { 1, sizeof(struct p_header), MDEV, { receive_SyncParam } }, | |
7204624c | 3922 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), CONN, { .conn_fn = receive_protocol } }, |
d9ae84e7 PR |
3923 | [P_UUIDS] = { 0, sizeof(struct p_uuids), MDEV, { receive_uuids } }, |
3924 | [P_SIZES] = { 0, sizeof(struct p_sizes), MDEV, { receive_sizes } }, | |
3925 | [P_STATE] = { 0, sizeof(struct p_state), MDEV, { receive_state } }, | |
3926 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), MDEV, { receive_req_state } }, | |
3927 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), MDEV, { receive_sync_uuid } }, | |
3928 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), MDEV, { receive_DataRequest } }, | |
3929 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), MDEV, { receive_DataRequest } }, | |
3930 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), MDEV, { receive_DataRequest } }, | |
3931 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), MDEV, { receive_skip } }, | |
3932 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), MDEV, { receive_out_of_sync } }, | |
dfafcc8a | 3933 | [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), CONN, { .conn_fn = receive_req_conn_state } }, |
b411b363 PR |
3934 | }; |
3935 | ||
02918be2 | 3936 | /* All handler functions that expect a sub-header get that sub-heder in |
e42325a5 | 3937 | mdev->tconn->data.rbuf.header.head.payload. |
02918be2 | 3938 | |
e42325a5 | 3939 | Usually in mdev->tconn->data.rbuf.header.head the callback can find the usual |
02918be2 PR |
3940 | p_header, but they may not rely on that. Since there is also p_header95 ! |
3941 | */ | |
b411b363 | 3942 | |
eefc2f7d | 3943 | static void drbdd(struct drbd_tconn *tconn) |
b411b363 | 3944 | { |
eefc2f7d | 3945 | struct p_header *header = &tconn->data.rbuf.header; |
77351055 | 3946 | struct packet_info pi; |
02918be2 | 3947 | size_t shs; /* sub header size */ |
82bc0194 | 3948 | int err; |
b411b363 | 3949 | |
eefc2f7d PR |
3950 | while (get_t_state(&tconn->receiver) == RUNNING) { |
3951 | drbd_thread_current_set_cpu(&tconn->receiver); | |
69bc7bc3 | 3952 | if (drbd_recv_header(tconn, &pi)) |
02918be2 | 3953 | goto err_out; |
b411b363 | 3954 | |
6e849ce8 | 3955 | if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || |
d9ae84e7 | 3956 | !drbd_cmd_handler[pi.cmd].mdev_fn)) { |
eefc2f7d | 3957 | conn_err(tconn, "unknown packet type %d, l: %d!\n", pi.cmd, pi.size); |
02918be2 | 3958 | goto err_out; |
0b33a916 | 3959 | } |
b411b363 | 3960 | |
77351055 PR |
3961 | shs = drbd_cmd_handler[pi.cmd].pkt_size - sizeof(struct p_header); |
3962 | if (pi.size - shs > 0 && !drbd_cmd_handler[pi.cmd].expect_payload) { | |
eefc2f7d | 3963 | conn_err(tconn, "No payload expected %s l:%d\n", cmdname(pi.cmd), pi.size); |
02918be2 | 3964 | goto err_out; |
b411b363 | 3965 | } |
b411b363 | 3966 | |
c13f7e1a | 3967 | if (shs) { |
a5c31904 AG |
3968 | err = drbd_recv_all_warn(tconn, &header->payload, shs); |
3969 | if (err) | |
c13f7e1a | 3970 | goto err_out; |
c13f7e1a LE |
3971 | } |
3972 | ||
a4fbda8e | 3973 | if (drbd_cmd_handler[pi.cmd].fa_type == CONN) { |
82bc0194 | 3974 | err = drbd_cmd_handler[pi.cmd].conn_fn(tconn, pi.cmd, pi.size - shs); |
d9ae84e7 PR |
3975 | } else { |
3976 | struct drbd_conf *mdev = vnr_to_mdev(tconn, pi.vnr); | |
82bc0194 | 3977 | err = mdev ? |
d9ae84e7 PR |
3978 | drbd_cmd_handler[pi.cmd].mdev_fn(mdev, pi.cmd, pi.size - shs) : |
3979 | tconn_receive_skip(tconn, pi.cmd, pi.size - shs); | |
3980 | } | |
b411b363 | 3981 | |
82bc0194 | 3982 | if (unlikely(err)) { |
eefc2f7d | 3983 | conn_err(tconn, "error receiving %s, l: %d!\n", |
77351055 | 3984 | cmdname(pi.cmd), pi.size); |
02918be2 | 3985 | goto err_out; |
b411b363 PR |
3986 | } |
3987 | } | |
82bc0194 | 3988 | return; |
b411b363 | 3989 | |
82bc0194 AG |
3990 | err_out: |
3991 | conn_request_state(tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); | |
b411b363 PR |
3992 | } |
3993 | ||
0e29d163 | 3994 | void conn_flush_workqueue(struct drbd_tconn *tconn) |
b411b363 PR |
3995 | { |
3996 | struct drbd_wq_barrier barr; | |
3997 | ||
3998 | barr.w.cb = w_prev_work_done; | |
0e29d163 | 3999 | barr.w.tconn = tconn; |
b411b363 | 4000 | init_completion(&barr.done); |
0e29d163 | 4001 | drbd_queue_work(&tconn->data.work, &barr.w); |
b411b363 PR |
4002 | wait_for_completion(&barr.done); |
4003 | } | |
4004 | ||
360cc740 | 4005 | static void drbd_disconnect(struct drbd_tconn *tconn) |
b411b363 | 4006 | { |
bbeb641c | 4007 | enum drbd_conns oc; |
b411b363 | 4008 | int rv = SS_UNKNOWN_ERROR; |
b411b363 | 4009 | |
bbeb641c | 4010 | if (tconn->cstate == C_STANDALONE) |
b411b363 | 4011 | return; |
b411b363 PR |
4012 | |
4013 | /* asender does not clean up anything. it must not interfere, either */ | |
360cc740 PR |
4014 | drbd_thread_stop(&tconn->asender); |
4015 | drbd_free_sock(tconn); | |
4016 | ||
4017 | idr_for_each(&tconn->volumes, drbd_disconnected, tconn); | |
4018 | ||
4019 | conn_info(tconn, "Connection closed\n"); | |
4020 | ||
4021 | spin_lock_irq(&tconn->req_lock); | |
bbeb641c PR |
4022 | oc = tconn->cstate; |
4023 | if (oc >= C_UNCONNECTED) | |
4024 | rv = _conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE); | |
4025 | ||
360cc740 PR |
4026 | spin_unlock_irq(&tconn->req_lock); |
4027 | ||
bbeb641c | 4028 | if (oc == C_DISCONNECTING) { |
360cc740 PR |
4029 | wait_event(tconn->net_cnt_wait, atomic_read(&tconn->net_cnt) == 0); |
4030 | ||
4031 | crypto_free_hash(tconn->cram_hmac_tfm); | |
4032 | tconn->cram_hmac_tfm = NULL; | |
4033 | ||
4034 | kfree(tconn->net_conf); | |
4035 | tconn->net_conf = NULL; | |
bbeb641c | 4036 | conn_request_state(tconn, NS(conn, C_STANDALONE), CS_VERBOSE); |
360cc740 PR |
4037 | } |
4038 | } | |
4039 | ||
4040 | static int drbd_disconnected(int vnr, void *p, void *data) | |
4041 | { | |
4042 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
4043 | enum drbd_fencing_p fp; | |
4044 | unsigned int i; | |
b411b363 | 4045 | |
85719573 | 4046 | /* wait for current activity to cease. */ |
87eeee41 | 4047 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4048 | _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
4049 | _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); | |
4050 | _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); | |
87eeee41 | 4051 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4052 | |
4053 | /* We do not have data structures that would allow us to | |
4054 | * get the rs_pending_cnt down to 0 again. | |
4055 | * * On C_SYNC_TARGET we do not have any data structures describing | |
4056 | * the pending RSDataRequest's we have sent. | |
4057 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
4058 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
4059 | * And no, it is not the sum of the reference counts in the | |
4060 | * resync_LRU. The resync_LRU tracks the whole operation including | |
4061 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
4062 | * on the fly. */ | |
4063 | drbd_rs_cancel_all(mdev); | |
4064 | mdev->rs_total = 0; | |
4065 | mdev->rs_failed = 0; | |
4066 | atomic_set(&mdev->rs_pending_cnt, 0); | |
4067 | wake_up(&mdev->misc_wait); | |
4068 | ||
7fde2be9 PR |
4069 | del_timer(&mdev->request_timer); |
4070 | ||
b411b363 | 4071 | del_timer_sync(&mdev->resync_timer); |
b411b363 PR |
4072 | resync_timer_fn((unsigned long)mdev); |
4073 | ||
b411b363 PR |
4074 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
4075 | * w_make_resync_request etc. which may still be on the worker queue | |
4076 | * to be "canceled" */ | |
a21e9298 | 4077 | drbd_flush_workqueue(mdev); |
b411b363 PR |
4078 | |
4079 | /* This also does reclaim_net_ee(). If we do this too early, we might | |
4080 | * miss some resync ee and pages.*/ | |
4081 | drbd_process_done_ee(mdev); | |
4082 | ||
4083 | kfree(mdev->p_uuid); | |
4084 | mdev->p_uuid = NULL; | |
4085 | ||
fb22c402 | 4086 | if (!is_susp(mdev->state)) |
2f5cdd0b | 4087 | tl_clear(mdev->tconn); |
b411b363 | 4088 | |
b411b363 PR |
4089 | drbd_md_sync(mdev); |
4090 | ||
4091 | fp = FP_DONT_CARE; | |
4092 | if (get_ldev(mdev)) { | |
4093 | fp = mdev->ldev->dc.fencing; | |
4094 | put_ldev(mdev); | |
4095 | } | |
4096 | ||
87f7be4c PR |
4097 | if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) |
4098 | drbd_try_outdate_peer_async(mdev); | |
b411b363 | 4099 | |
20ceb2b2 LE |
4100 | /* serialize with bitmap writeout triggered by the state change, |
4101 | * if any. */ | |
4102 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); | |
4103 | ||
b411b363 PR |
4104 | /* tcp_close and release of sendpage pages can be deferred. I don't |
4105 | * want to use SO_LINGER, because apparently it can be deferred for | |
4106 | * more than 20 seconds (longest time I checked). | |
4107 | * | |
4108 | * Actually we don't care for exactly when the network stack does its | |
4109 | * put_page(), but release our reference on these pages right here. | |
4110 | */ | |
4111 | i = drbd_release_ee(mdev, &mdev->net_ee); | |
4112 | if (i) | |
4113 | dev_info(DEV, "net_ee not empty, killed %u entries\n", i); | |
435f0740 LE |
4114 | i = atomic_read(&mdev->pp_in_use_by_net); |
4115 | if (i) | |
4116 | dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); | |
b411b363 PR |
4117 | i = atomic_read(&mdev->pp_in_use); |
4118 | if (i) | |
45bb912b | 4119 | dev_info(DEV, "pp_in_use = %d, expected 0\n", i); |
b411b363 PR |
4120 | |
4121 | D_ASSERT(list_empty(&mdev->read_ee)); | |
4122 | D_ASSERT(list_empty(&mdev->active_ee)); | |
4123 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
4124 | D_ASSERT(list_empty(&mdev->done_ee)); | |
4125 | ||
4126 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ | |
4127 | atomic_set(&mdev->current_epoch->epoch_size, 0); | |
4128 | D_ASSERT(list_empty(&mdev->current_epoch->list)); | |
360cc740 PR |
4129 | |
4130 | return 0; | |
b411b363 PR |
4131 | } |
4132 | ||
4133 | /* | |
4134 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
4135 | * we can agree on is stored in agreed_pro_version. | |
4136 | * | |
4137 | * feature flags and the reserved array should be enough room for future | |
4138 | * enhancements of the handshake protocol, and possible plugins... | |
4139 | * | |
4140 | * for now, they are expected to be zero, but ignored. | |
4141 | */ | |
8a22cccc | 4142 | static int drbd_send_handshake(struct drbd_tconn *tconn) |
b411b363 | 4143 | { |
e6b3ea83 | 4144 | /* ASSERT current == mdev->tconn->receiver ... */ |
8a22cccc | 4145 | struct p_handshake *p = &tconn->data.sbuf.handshake; |
e8d17b01 | 4146 | int err; |
b411b363 | 4147 | |
8a22cccc PR |
4148 | if (mutex_lock_interruptible(&tconn->data.mutex)) { |
4149 | conn_err(tconn, "interrupted during initial handshake\n"); | |
e8d17b01 | 4150 | return -EINTR; |
b411b363 PR |
4151 | } |
4152 | ||
8a22cccc PR |
4153 | if (tconn->data.socket == NULL) { |
4154 | mutex_unlock(&tconn->data.mutex); | |
e8d17b01 | 4155 | return -EIO; |
b411b363 PR |
4156 | } |
4157 | ||
4158 | memset(p, 0, sizeof(*p)); | |
4159 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
4160 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
e8d17b01 | 4161 | err = _conn_send_cmd(tconn, 0, tconn->data.socket, P_HAND_SHAKE, |
ecf2363c | 4162 | &p->head, sizeof(*p), 0); |
8a22cccc | 4163 | mutex_unlock(&tconn->data.mutex); |
e8d17b01 | 4164 | return err; |
b411b363 PR |
4165 | } |
4166 | ||
4167 | /* | |
4168 | * return values: | |
4169 | * 1 yes, we have a valid connection | |
4170 | * 0 oops, did not work out, please try again | |
4171 | * -1 peer talks different language, | |
4172 | * no point in trying again, please go standalone. | |
4173 | */ | |
65d11ed6 | 4174 | static int drbd_do_handshake(struct drbd_tconn *tconn) |
b411b363 | 4175 | { |
65d11ed6 PR |
4176 | /* ASSERT current == tconn->receiver ... */ |
4177 | struct p_handshake *p = &tconn->data.rbuf.handshake; | |
02918be2 | 4178 | const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80); |
77351055 | 4179 | struct packet_info pi; |
a5c31904 | 4180 | int err; |
b411b363 | 4181 | |
e8d17b01 AG |
4182 | err = drbd_send_handshake(tconn); |
4183 | if (err) | |
b411b363 PR |
4184 | return 0; |
4185 | ||
69bc7bc3 AG |
4186 | err = drbd_recv_header(tconn, &pi); |
4187 | if (err) | |
b411b363 PR |
4188 | return 0; |
4189 | ||
77351055 | 4190 | if (pi.cmd != P_HAND_SHAKE) { |
65d11ed6 | 4191 | conn_err(tconn, "expected HandShake packet, received: %s (0x%04x)\n", |
77351055 | 4192 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4193 | return -1; |
4194 | } | |
4195 | ||
77351055 | 4196 | if (pi.size != expect) { |
65d11ed6 | 4197 | conn_err(tconn, "expected HandShake length: %u, received: %u\n", |
77351055 | 4198 | expect, pi.size); |
b411b363 PR |
4199 | return -1; |
4200 | } | |
4201 | ||
a5c31904 AG |
4202 | err = drbd_recv_all_warn(tconn, &p->head.payload, expect); |
4203 | if (err) | |
b411b363 | 4204 | return 0; |
b411b363 | 4205 | |
b411b363 PR |
4206 | p->protocol_min = be32_to_cpu(p->protocol_min); |
4207 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
4208 | if (p->protocol_max == 0) | |
4209 | p->protocol_max = p->protocol_min; | |
4210 | ||
4211 | if (PRO_VERSION_MAX < p->protocol_min || | |
4212 | PRO_VERSION_MIN > p->protocol_max) | |
4213 | goto incompat; | |
4214 | ||
65d11ed6 | 4215 | tconn->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); |
b411b363 | 4216 | |
65d11ed6 PR |
4217 | conn_info(tconn, "Handshake successful: " |
4218 | "Agreed network protocol version %d\n", tconn->agreed_pro_version); | |
b411b363 PR |
4219 | |
4220 | return 1; | |
4221 | ||
4222 | incompat: | |
65d11ed6 | 4223 | conn_err(tconn, "incompatible DRBD dialects: " |
b411b363 PR |
4224 | "I support %d-%d, peer supports %d-%d\n", |
4225 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
4226 | p->protocol_min, p->protocol_max); | |
4227 | return -1; | |
4228 | } | |
4229 | ||
4230 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
13e6037d | 4231 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 PR |
4232 | { |
4233 | dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); | |
4234 | dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4235 | return -1; |
b411b363 PR |
4236 | } |
4237 | #else | |
4238 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4239 | |
4240 | /* Return value: | |
4241 | 1 - auth succeeded, | |
4242 | 0 - failed, try again (network error), | |
4243 | -1 - auth failed, don't try again. | |
4244 | */ | |
4245 | ||
13e6037d | 4246 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 PR |
4247 | { |
4248 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ | |
4249 | struct scatterlist sg; | |
4250 | char *response = NULL; | |
4251 | char *right_response = NULL; | |
4252 | char *peers_ch = NULL; | |
13e6037d | 4253 | unsigned int key_len = strlen(tconn->net_conf->shared_secret); |
b411b363 PR |
4254 | unsigned int resp_size; |
4255 | struct hash_desc desc; | |
77351055 | 4256 | struct packet_info pi; |
69bc7bc3 | 4257 | int err, rv; |
b411b363 | 4258 | |
13e6037d | 4259 | desc.tfm = tconn->cram_hmac_tfm; |
b411b363 PR |
4260 | desc.flags = 0; |
4261 | ||
13e6037d PR |
4262 | rv = crypto_hash_setkey(tconn->cram_hmac_tfm, |
4263 | (u8 *)tconn->net_conf->shared_secret, key_len); | |
b411b363 | 4264 | if (rv) { |
13e6037d | 4265 | conn_err(tconn, "crypto_hash_setkey() failed with %d\n", rv); |
b10d96cb | 4266 | rv = -1; |
b411b363 PR |
4267 | goto fail; |
4268 | } | |
4269 | ||
4270 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4271 | ||
ce9879cb | 4272 | rv = !conn_send_cmd2(tconn, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); |
b411b363 PR |
4273 | if (!rv) |
4274 | goto fail; | |
4275 | ||
69bc7bc3 AG |
4276 | err = drbd_recv_header(tconn, &pi); |
4277 | if (err) { | |
4278 | rv = 0; | |
b411b363 | 4279 | goto fail; |
69bc7bc3 | 4280 | } |
b411b363 | 4281 | |
77351055 | 4282 | if (pi.cmd != P_AUTH_CHALLENGE) { |
13e6037d | 4283 | conn_err(tconn, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
77351055 | 4284 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4285 | rv = 0; |
4286 | goto fail; | |
4287 | } | |
4288 | ||
77351055 | 4289 | if (pi.size > CHALLENGE_LEN * 2) { |
13e6037d | 4290 | conn_err(tconn, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4291 | rv = -1; |
b411b363 PR |
4292 | goto fail; |
4293 | } | |
4294 | ||
77351055 | 4295 | peers_ch = kmalloc(pi.size, GFP_NOIO); |
b411b363 | 4296 | if (peers_ch == NULL) { |
13e6037d | 4297 | conn_err(tconn, "kmalloc of peers_ch failed\n"); |
b10d96cb | 4298 | rv = -1; |
b411b363 PR |
4299 | goto fail; |
4300 | } | |
4301 | ||
a5c31904 AG |
4302 | err = drbd_recv_all_warn(tconn, peers_ch, pi.size); |
4303 | if (err) { | |
b411b363 PR |
4304 | rv = 0; |
4305 | goto fail; | |
4306 | } | |
4307 | ||
13e6037d | 4308 | resp_size = crypto_hash_digestsize(tconn->cram_hmac_tfm); |
b411b363 PR |
4309 | response = kmalloc(resp_size, GFP_NOIO); |
4310 | if (response == NULL) { | |
13e6037d | 4311 | conn_err(tconn, "kmalloc of response failed\n"); |
b10d96cb | 4312 | rv = -1; |
b411b363 PR |
4313 | goto fail; |
4314 | } | |
4315 | ||
4316 | sg_init_table(&sg, 1); | |
77351055 | 4317 | sg_set_buf(&sg, peers_ch, pi.size); |
b411b363 PR |
4318 | |
4319 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4320 | if (rv) { | |
13e6037d | 4321 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4322 | rv = -1; |
b411b363 PR |
4323 | goto fail; |
4324 | } | |
4325 | ||
ce9879cb | 4326 | rv = !conn_send_cmd2(tconn, P_AUTH_RESPONSE, response, resp_size); |
b411b363 PR |
4327 | if (!rv) |
4328 | goto fail; | |
4329 | ||
69bc7bc3 AG |
4330 | err = drbd_recv_header(tconn, &pi); |
4331 | if (err) { | |
4332 | rv = 0; | |
b411b363 | 4333 | goto fail; |
69bc7bc3 | 4334 | } |
b411b363 | 4335 | |
77351055 | 4336 | if (pi.cmd != P_AUTH_RESPONSE) { |
13e6037d | 4337 | conn_err(tconn, "expected AuthResponse packet, received: %s (0x%04x)\n", |
77351055 | 4338 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4339 | rv = 0; |
4340 | goto fail; | |
4341 | } | |
4342 | ||
77351055 | 4343 | if (pi.size != resp_size) { |
13e6037d | 4344 | conn_err(tconn, "expected AuthResponse payload of wrong size\n"); |
b411b363 PR |
4345 | rv = 0; |
4346 | goto fail; | |
4347 | } | |
4348 | ||
a5c31904 AG |
4349 | err = drbd_recv_all_warn(tconn, response , resp_size); |
4350 | if (err) { | |
b411b363 PR |
4351 | rv = 0; |
4352 | goto fail; | |
4353 | } | |
4354 | ||
4355 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4356 | if (right_response == NULL) { |
13e6037d | 4357 | conn_err(tconn, "kmalloc of right_response failed\n"); |
b10d96cb | 4358 | rv = -1; |
b411b363 PR |
4359 | goto fail; |
4360 | } | |
4361 | ||
4362 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4363 | ||
4364 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4365 | if (rv) { | |
13e6037d | 4366 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4367 | rv = -1; |
b411b363 PR |
4368 | goto fail; |
4369 | } | |
4370 | ||
4371 | rv = !memcmp(response, right_response, resp_size); | |
4372 | ||
4373 | if (rv) | |
13e6037d PR |
4374 | conn_info(tconn, "Peer authenticated using %d bytes of '%s' HMAC\n", |
4375 | resp_size, tconn->net_conf->cram_hmac_alg); | |
b10d96cb JT |
4376 | else |
4377 | rv = -1; | |
b411b363 PR |
4378 | |
4379 | fail: | |
4380 | kfree(peers_ch); | |
4381 | kfree(response); | |
4382 | kfree(right_response); | |
4383 | ||
4384 | return rv; | |
4385 | } | |
4386 | #endif | |
4387 | ||
4388 | int drbdd_init(struct drbd_thread *thi) | |
4389 | { | |
392c8801 | 4390 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 PR |
4391 | int h; |
4392 | ||
4d641dd7 | 4393 | conn_info(tconn, "receiver (re)started\n"); |
b411b363 PR |
4394 | |
4395 | do { | |
4d641dd7 | 4396 | h = drbd_connect(tconn); |
b411b363 | 4397 | if (h == 0) { |
4d641dd7 | 4398 | drbd_disconnect(tconn); |
20ee6390 | 4399 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
4400 | } |
4401 | if (h == -1) { | |
4d641dd7 | 4402 | conn_warn(tconn, "Discarding network configuration.\n"); |
bbeb641c | 4403 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
4404 | } |
4405 | } while (h == 0); | |
4406 | ||
4407 | if (h > 0) { | |
4d641dd7 PR |
4408 | if (get_net_conf(tconn)) { |
4409 | drbdd(tconn); | |
4410 | put_net_conf(tconn); | |
b411b363 PR |
4411 | } |
4412 | } | |
4413 | ||
4d641dd7 | 4414 | drbd_disconnect(tconn); |
b411b363 | 4415 | |
4d641dd7 | 4416 | conn_info(tconn, "receiver terminated\n"); |
b411b363 PR |
4417 | return 0; |
4418 | } | |
4419 | ||
4420 | /* ********* acknowledge sender ******** */ | |
4421 | ||
e4f78ede PR |
4422 | static int got_conn_RqSReply(struct drbd_tconn *tconn, enum drbd_packet cmd) |
4423 | { | |
4424 | struct p_req_state_reply *p = &tconn->meta.rbuf.req_state_reply; | |
4425 | int retcode = be32_to_cpu(p->retcode); | |
4426 | ||
4427 | if (retcode >= SS_SUCCESS) { | |
4428 | set_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags); | |
4429 | } else { | |
4430 | set_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags); | |
4431 | conn_err(tconn, "Requested state change failed by peer: %s (%d)\n", | |
4432 | drbd_set_st_err_str(retcode), retcode); | |
4433 | } | |
4434 | wake_up(&tconn->ping_wait); | |
4435 | ||
4436 | return true; | |
4437 | } | |
4438 | ||
d8763023 | 4439 | static int got_RqSReply(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4440 | { |
257d0af6 | 4441 | struct p_req_state_reply *p = &mdev->tconn->meta.rbuf.req_state_reply; |
b411b363 PR |
4442 | int retcode = be32_to_cpu(p->retcode); |
4443 | ||
e4f78ede PR |
4444 | if (retcode >= SS_SUCCESS) { |
4445 | set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); | |
4446 | } else { | |
4447 | set_bit(CL_ST_CHG_FAIL, &mdev->flags); | |
4448 | dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", | |
4449 | drbd_set_st_err_str(retcode), retcode); | |
b411b363 | 4450 | } |
e4f78ede PR |
4451 | wake_up(&mdev->state_wait); |
4452 | ||
81e84650 | 4453 | return true; |
b411b363 PR |
4454 | } |
4455 | ||
f19e4f8b | 4456 | static int got_Ping(struct drbd_tconn *tconn, enum drbd_packet cmd) |
b411b363 | 4457 | { |
f19e4f8b | 4458 | return drbd_send_ping_ack(tconn); |
b411b363 PR |
4459 | |
4460 | } | |
4461 | ||
f19e4f8b | 4462 | static int got_PingAck(struct drbd_tconn *tconn, enum drbd_packet cmd) |
b411b363 PR |
4463 | { |
4464 | /* restore idle timeout */ | |
2a67d8b9 PR |
4465 | tconn->meta.socket->sk->sk_rcvtimeo = tconn->net_conf->ping_int*HZ; |
4466 | if (!test_and_set_bit(GOT_PING_ACK, &tconn->flags)) | |
4467 | wake_up(&tconn->ping_wait); | |
b411b363 | 4468 | |
81e84650 | 4469 | return true; |
b411b363 PR |
4470 | } |
4471 | ||
d8763023 | 4472 | static int got_IsInSync(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4473 | { |
257d0af6 | 4474 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4475 | sector_t sector = be64_to_cpu(p->sector); |
4476 | int blksize = be32_to_cpu(p->blksize); | |
4477 | ||
31890f4a | 4478 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
b411b363 PR |
4479 | |
4480 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4481 | ||
1d53f09e LE |
4482 | if (get_ldev(mdev)) { |
4483 | drbd_rs_complete_io(mdev, sector); | |
4484 | drbd_set_in_sync(mdev, sector, blksize); | |
4485 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ | |
4486 | mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); | |
4487 | put_ldev(mdev); | |
4488 | } | |
b411b363 | 4489 | dec_rs_pending(mdev); |
778f271d | 4490 | atomic_add(blksize >> 9, &mdev->rs_sect_in); |
b411b363 | 4491 | |
81e84650 | 4492 | return true; |
b411b363 PR |
4493 | } |
4494 | ||
bc9c5c41 AG |
4495 | static int |
4496 | validate_req_change_req_state(struct drbd_conf *mdev, u64 id, sector_t sector, | |
4497 | struct rb_root *root, const char *func, | |
4498 | enum drbd_req_event what, bool missing_ok) | |
b411b363 PR |
4499 | { |
4500 | struct drbd_request *req; | |
4501 | struct bio_and_error m; | |
4502 | ||
87eeee41 | 4503 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 4504 | req = find_request(mdev, root, id, sector, missing_ok, func); |
b411b363 | 4505 | if (unlikely(!req)) { |
87eeee41 | 4506 | spin_unlock_irq(&mdev->tconn->req_lock); |
81e84650 | 4507 | return false; |
b411b363 PR |
4508 | } |
4509 | __req_mod(req, what, &m); | |
87eeee41 | 4510 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4511 | |
4512 | if (m.bio) | |
4513 | complete_master_bio(mdev, &m); | |
81e84650 | 4514 | return true; |
b411b363 PR |
4515 | } |
4516 | ||
d8763023 | 4517 | static int got_BlockAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4518 | { |
257d0af6 | 4519 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4520 | sector_t sector = be64_to_cpu(p->sector); |
4521 | int blksize = be32_to_cpu(p->blksize); | |
4522 | enum drbd_req_event what; | |
4523 | ||
4524 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4525 | ||
579b57ed | 4526 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4527 | drbd_set_in_sync(mdev, sector, blksize); |
4528 | dec_rs_pending(mdev); | |
81e84650 | 4529 | return true; |
b411b363 | 4530 | } |
257d0af6 | 4531 | switch (cmd) { |
b411b363 | 4532 | case P_RS_WRITE_ACK: |
89e58e75 | 4533 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
8554df1c | 4534 | what = WRITE_ACKED_BY_PEER_AND_SIS; |
b411b363 PR |
4535 | break; |
4536 | case P_WRITE_ACK: | |
89e58e75 | 4537 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
8554df1c | 4538 | what = WRITE_ACKED_BY_PEER; |
b411b363 PR |
4539 | break; |
4540 | case P_RECV_ACK: | |
89e58e75 | 4541 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_B); |
8554df1c | 4542 | what = RECV_ACKED_BY_PEER; |
b411b363 | 4543 | break; |
7be8da07 | 4544 | case P_DISCARD_WRITE: |
89e58e75 | 4545 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
7be8da07 AG |
4546 | what = DISCARD_WRITE; |
4547 | break; | |
4548 | case P_RETRY_WRITE: | |
4549 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); | |
4550 | what = POSTPONE_WRITE; | |
b411b363 PR |
4551 | break; |
4552 | default: | |
4553 | D_ASSERT(0); | |
81e84650 | 4554 | return false; |
b411b363 PR |
4555 | } |
4556 | ||
4557 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
bc9c5c41 AG |
4558 | &mdev->write_requests, __func__, |
4559 | what, false); | |
b411b363 PR |
4560 | } |
4561 | ||
d8763023 | 4562 | static int got_NegAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4563 | { |
257d0af6 | 4564 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 | 4565 | sector_t sector = be64_to_cpu(p->sector); |
2deb8336 | 4566 | int size = be32_to_cpu(p->blksize); |
89e58e75 PR |
4567 | bool missing_ok = mdev->tconn->net_conf->wire_protocol == DRBD_PROT_A || |
4568 | mdev->tconn->net_conf->wire_protocol == DRBD_PROT_B; | |
c3afd8f5 | 4569 | bool found; |
b411b363 PR |
4570 | |
4571 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4572 | ||
579b57ed | 4573 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4574 | dec_rs_pending(mdev); |
4575 | drbd_rs_failed_io(mdev, sector, size); | |
81e84650 | 4576 | return true; |
b411b363 | 4577 | } |
2deb8336 | 4578 | |
c3afd8f5 | 4579 | found = validate_req_change_req_state(mdev, p->block_id, sector, |
bc9c5c41 | 4580 | &mdev->write_requests, __func__, |
8554df1c | 4581 | NEG_ACKED, missing_ok); |
c3afd8f5 AG |
4582 | if (!found) { |
4583 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. | |
4584 | The master bio might already be completed, therefore the | |
4585 | request is no longer in the collision hash. */ | |
4586 | /* In Protocol B we might already have got a P_RECV_ACK | |
4587 | but then get a P_NEG_ACK afterwards. */ | |
4588 | if (!missing_ok) | |
2deb8336 | 4589 | return false; |
c3afd8f5 | 4590 | drbd_set_out_of_sync(mdev, sector, size); |
2deb8336 | 4591 | } |
2deb8336 | 4592 | return true; |
b411b363 PR |
4593 | } |
4594 | ||
d8763023 | 4595 | static int got_NegDReply(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4596 | { |
257d0af6 | 4597 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4598 | sector_t sector = be64_to_cpu(p->sector); |
4599 | ||
4600 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
7be8da07 | 4601 | |
b411b363 PR |
4602 | dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", |
4603 | (unsigned long long)sector, be32_to_cpu(p->blksize)); | |
4604 | ||
4605 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
bc9c5c41 | 4606 | &mdev->read_requests, __func__, |
8554df1c | 4607 | NEG_ACKED, false); |
b411b363 PR |
4608 | } |
4609 | ||
d8763023 | 4610 | static int got_NegRSDReply(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 PR |
4611 | { |
4612 | sector_t sector; | |
4613 | int size; | |
257d0af6 | 4614 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4615 | |
4616 | sector = be64_to_cpu(p->sector); | |
4617 | size = be32_to_cpu(p->blksize); | |
b411b363 PR |
4618 | |
4619 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4620 | ||
4621 | dec_rs_pending(mdev); | |
4622 | ||
4623 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
4624 | drbd_rs_complete_io(mdev, sector); | |
257d0af6 | 4625 | switch (cmd) { |
d612d309 PR |
4626 | case P_NEG_RS_DREPLY: |
4627 | drbd_rs_failed_io(mdev, sector, size); | |
4628 | case P_RS_CANCEL: | |
4629 | break; | |
4630 | default: | |
4631 | D_ASSERT(0); | |
4632 | put_ldev(mdev); | |
4633 | return false; | |
4634 | } | |
b411b363 PR |
4635 | put_ldev(mdev); |
4636 | } | |
4637 | ||
81e84650 | 4638 | return true; |
b411b363 PR |
4639 | } |
4640 | ||
d8763023 | 4641 | static int got_BarrierAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4642 | { |
257d0af6 | 4643 | struct p_barrier_ack *p = &mdev->tconn->meta.rbuf.barrier_ack; |
b411b363 | 4644 | |
2f5cdd0b | 4645 | tl_release(mdev->tconn, p->barrier, be32_to_cpu(p->set_size)); |
b411b363 | 4646 | |
c4752ef1 PR |
4647 | if (mdev->state.conn == C_AHEAD && |
4648 | atomic_read(&mdev->ap_in_flight) == 0 && | |
370a43e7 PR |
4649 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags)) { |
4650 | mdev->start_resync_timer.expires = jiffies + HZ; | |
4651 | add_timer(&mdev->start_resync_timer); | |
c4752ef1 PR |
4652 | } |
4653 | ||
81e84650 | 4654 | return true; |
b411b363 PR |
4655 | } |
4656 | ||
d8763023 | 4657 | static int got_OVResult(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4658 | { |
257d0af6 | 4659 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4660 | struct drbd_work *w; |
4661 | sector_t sector; | |
4662 | int size; | |
4663 | ||
4664 | sector = be64_to_cpu(p->sector); | |
4665 | size = be32_to_cpu(p->blksize); | |
4666 | ||
4667 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4668 | ||
4669 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
8f7bed77 | 4670 | drbd_ov_out_of_sync_found(mdev, sector, size); |
b411b363 | 4671 | else |
8f7bed77 | 4672 | ov_out_of_sync_print(mdev); |
b411b363 | 4673 | |
1d53f09e | 4674 | if (!get_ldev(mdev)) |
81e84650 | 4675 | return true; |
1d53f09e | 4676 | |
b411b363 PR |
4677 | drbd_rs_complete_io(mdev, sector); |
4678 | dec_rs_pending(mdev); | |
4679 | ||
ea5442af LE |
4680 | --mdev->ov_left; |
4681 | ||
4682 | /* let's advance progress step marks only for every other megabyte */ | |
4683 | if ((mdev->ov_left & 0x200) == 0x200) | |
4684 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
4685 | ||
4686 | if (mdev->ov_left == 0) { | |
b411b363 PR |
4687 | w = kmalloc(sizeof(*w), GFP_NOIO); |
4688 | if (w) { | |
4689 | w->cb = w_ov_finished; | |
a21e9298 | 4690 | w->mdev = mdev; |
e42325a5 | 4691 | drbd_queue_work_front(&mdev->tconn->data.work, w); |
b411b363 PR |
4692 | } else { |
4693 | dev_err(DEV, "kmalloc(w) failed."); | |
8f7bed77 | 4694 | ov_out_of_sync_print(mdev); |
b411b363 PR |
4695 | drbd_resync_finished(mdev); |
4696 | } | |
4697 | } | |
1d53f09e | 4698 | put_ldev(mdev); |
81e84650 | 4699 | return true; |
b411b363 PR |
4700 | } |
4701 | ||
d8763023 | 4702 | static int got_skip(struct drbd_conf *mdev, enum drbd_packet cmd) |
0ced55a3 | 4703 | { |
81e84650 | 4704 | return true; |
0ced55a3 PR |
4705 | } |
4706 | ||
32862ec7 PR |
4707 | static int tconn_process_done_ee(struct drbd_tconn *tconn) |
4708 | { | |
082a3439 PR |
4709 | struct drbd_conf *mdev; |
4710 | int i, not_empty = 0; | |
32862ec7 PR |
4711 | |
4712 | do { | |
4713 | clear_bit(SIGNAL_ASENDER, &tconn->flags); | |
4714 | flush_signals(current); | |
082a3439 | 4715 | idr_for_each_entry(&tconn->volumes, mdev, i) { |
e2b3032b | 4716 | if (drbd_process_done_ee(mdev)) |
082a3439 PR |
4717 | return 1; /* error */ |
4718 | } | |
32862ec7 | 4719 | set_bit(SIGNAL_ASENDER, &tconn->flags); |
082a3439 PR |
4720 | |
4721 | spin_lock_irq(&tconn->req_lock); | |
4722 | idr_for_each_entry(&tconn->volumes, mdev, i) { | |
4723 | not_empty = !list_empty(&mdev->done_ee); | |
4724 | if (not_empty) | |
4725 | break; | |
4726 | } | |
4727 | spin_unlock_irq(&tconn->req_lock); | |
32862ec7 PR |
4728 | } while (not_empty); |
4729 | ||
4730 | return 0; | |
4731 | } | |
4732 | ||
7201b972 AG |
4733 | struct asender_cmd { |
4734 | size_t pkt_size; | |
a4fbda8e PR |
4735 | enum mdev_or_conn fa_type; /* first argument's type */ |
4736 | union { | |
4737 | int (*mdev_fn)(struct drbd_conf *mdev, enum drbd_packet cmd); | |
4738 | int (*conn_fn)(struct drbd_tconn *tconn, enum drbd_packet cmd); | |
4739 | }; | |
7201b972 AG |
4740 | }; |
4741 | ||
4742 | static struct asender_cmd asender_tbl[] = { | |
f19e4f8b PR |
4743 | [P_PING] = { sizeof(struct p_header), CONN, { .conn_fn = got_Ping } }, |
4744 | [P_PING_ACK] = { sizeof(struct p_header), CONN, { .conn_fn = got_PingAck } }, | |
a4fbda8e PR |
4745 | [P_RECV_ACK] = { sizeof(struct p_block_ack), MDEV, { got_BlockAck } }, |
4746 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), MDEV, { got_BlockAck } }, | |
4747 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), MDEV, { got_BlockAck } }, | |
4748 | [P_DISCARD_WRITE] = { sizeof(struct p_block_ack), MDEV, { got_BlockAck } }, | |
4749 | [P_NEG_ACK] = { sizeof(struct p_block_ack), MDEV, { got_NegAck } }, | |
4750 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), MDEV, { got_NegDReply } }, | |
4751 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), MDEV, { got_NegRSDReply } }, | |
4752 | [P_OV_RESULT] = { sizeof(struct p_block_ack), MDEV, { got_OVResult } }, | |
4753 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), MDEV, { got_BarrierAck } }, | |
4754 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), MDEV, { got_RqSReply } }, | |
4755 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), MDEV, { got_IsInSync } }, | |
4756 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), MDEV, { got_skip } }, | |
4757 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), MDEV, { got_NegRSDReply } }, | |
e4f78ede | 4758 | [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), CONN, {.conn_fn = got_conn_RqSReply}}, |
a4fbda8e | 4759 | [P_RETRY_WRITE] = { sizeof(struct p_block_ack), MDEV, { got_BlockAck } }, |
7201b972 AG |
4760 | }; |
4761 | ||
b411b363 PR |
4762 | int drbd_asender(struct drbd_thread *thi) |
4763 | { | |
392c8801 | 4764 | struct drbd_tconn *tconn = thi->tconn; |
32862ec7 | 4765 | struct p_header *h = &tconn->meta.rbuf.header; |
b411b363 | 4766 | struct asender_cmd *cmd = NULL; |
77351055 | 4767 | struct packet_info pi; |
257d0af6 | 4768 | int rv; |
b411b363 PR |
4769 | void *buf = h; |
4770 | int received = 0; | |
257d0af6 | 4771 | int expect = sizeof(struct p_header); |
f36af18c | 4772 | int ping_timeout_active = 0; |
b411b363 | 4773 | |
b411b363 PR |
4774 | current->policy = SCHED_RR; /* Make this a realtime task! */ |
4775 | current->rt_priority = 2; /* more important than all other tasks */ | |
4776 | ||
e77a0a5c | 4777 | while (get_t_state(thi) == RUNNING) { |
80822284 | 4778 | drbd_thread_current_set_cpu(thi); |
32862ec7 | 4779 | if (test_and_clear_bit(SEND_PING, &tconn->flags)) { |
2a67d8b9 | 4780 | if (!drbd_send_ping(tconn)) { |
32862ec7 | 4781 | conn_err(tconn, "drbd_send_ping has failed\n"); |
841ce241 AG |
4782 | goto reconnect; |
4783 | } | |
32862ec7 PR |
4784 | tconn->meta.socket->sk->sk_rcvtimeo = |
4785 | tconn->net_conf->ping_timeo*HZ/10; | |
f36af18c | 4786 | ping_timeout_active = 1; |
b411b363 PR |
4787 | } |
4788 | ||
32862ec7 PR |
4789 | /* TODO: conditionally cork; it may hurt latency if we cork without |
4790 | much to send */ | |
4791 | if (!tconn->net_conf->no_cork) | |
4792 | drbd_tcp_cork(tconn->meta.socket); | |
082a3439 PR |
4793 | if (tconn_process_done_ee(tconn)) { |
4794 | conn_err(tconn, "tconn_process_done_ee() failed\n"); | |
32862ec7 | 4795 | goto reconnect; |
082a3439 | 4796 | } |
b411b363 | 4797 | /* but unconditionally uncork unless disabled */ |
32862ec7 PR |
4798 | if (!tconn->net_conf->no_cork) |
4799 | drbd_tcp_uncork(tconn->meta.socket); | |
b411b363 PR |
4800 | |
4801 | /* short circuit, recv_msg would return EINTR anyways. */ | |
4802 | if (signal_pending(current)) | |
4803 | continue; | |
4804 | ||
32862ec7 PR |
4805 | rv = drbd_recv_short(tconn->meta.socket, buf, expect-received, 0); |
4806 | clear_bit(SIGNAL_ASENDER, &tconn->flags); | |
b411b363 PR |
4807 | |
4808 | flush_signals(current); | |
4809 | ||
4810 | /* Note: | |
4811 | * -EINTR (on meta) we got a signal | |
4812 | * -EAGAIN (on meta) rcvtimeo expired | |
4813 | * -ECONNRESET other side closed the connection | |
4814 | * -ERESTARTSYS (on data) we got a signal | |
4815 | * rv < 0 other than above: unexpected error! | |
4816 | * rv == expected: full header or command | |
4817 | * rv < expected: "woken" by signal during receive | |
4818 | * rv == 0 : "connection shut down by peer" | |
4819 | */ | |
4820 | if (likely(rv > 0)) { | |
4821 | received += rv; | |
4822 | buf += rv; | |
4823 | } else if (rv == 0) { | |
32862ec7 | 4824 | conn_err(tconn, "meta connection shut down by peer.\n"); |
b411b363 PR |
4825 | goto reconnect; |
4826 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
4827 | /* If the data socket received something meanwhile, |
4828 | * that is good enough: peer is still alive. */ | |
32862ec7 PR |
4829 | if (time_after(tconn->last_received, |
4830 | jiffies - tconn->meta.socket->sk->sk_rcvtimeo)) | |
cb6518cb | 4831 | continue; |
f36af18c | 4832 | if (ping_timeout_active) { |
32862ec7 | 4833 | conn_err(tconn, "PingAck did not arrive in time.\n"); |
b411b363 PR |
4834 | goto reconnect; |
4835 | } | |
32862ec7 | 4836 | set_bit(SEND_PING, &tconn->flags); |
b411b363 PR |
4837 | continue; |
4838 | } else if (rv == -EINTR) { | |
4839 | continue; | |
4840 | } else { | |
32862ec7 | 4841 | conn_err(tconn, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
4842 | goto reconnect; |
4843 | } | |
4844 | ||
4845 | if (received == expect && cmd == NULL) { | |
8172f3e9 | 4846 | if (decode_header(tconn, h, &pi)) |
b411b363 | 4847 | goto reconnect; |
7201b972 AG |
4848 | cmd = &asender_tbl[pi.cmd]; |
4849 | if (pi.cmd >= ARRAY_SIZE(asender_tbl) || !cmd) { | |
32862ec7 | 4850 | conn_err(tconn, "unknown command %d on meta (l: %d)\n", |
77351055 | 4851 | pi.cmd, pi.size); |
b411b363 PR |
4852 | goto disconnect; |
4853 | } | |
4854 | expect = cmd->pkt_size; | |
77351055 | 4855 | if (pi.size != expect - sizeof(struct p_header)) { |
32862ec7 | 4856 | conn_err(tconn, "Wrong packet size on meta (c: %d, l: %d)\n", |
77351055 | 4857 | pi.cmd, pi.size); |
b411b363 | 4858 | goto reconnect; |
257d0af6 | 4859 | } |
b411b363 PR |
4860 | } |
4861 | if (received == expect) { | |
a4fbda8e PR |
4862 | bool rv; |
4863 | ||
4864 | if (cmd->fa_type == CONN) { | |
4865 | rv = cmd->conn_fn(tconn, pi.cmd); | |
4866 | } else { | |
4867 | struct drbd_conf *mdev = vnr_to_mdev(tconn, pi.vnr); | |
4868 | rv = cmd->mdev_fn(mdev, pi.cmd); | |
4869 | } | |
4870 | ||
4871 | if (!rv) | |
b411b363 PR |
4872 | goto reconnect; |
4873 | ||
a4fbda8e PR |
4874 | tconn->last_received = jiffies; |
4875 | ||
f36af18c LE |
4876 | /* the idle_timeout (ping-int) |
4877 | * has been restored in got_PingAck() */ | |
7201b972 | 4878 | if (cmd == &asender_tbl[P_PING_ACK]) |
f36af18c LE |
4879 | ping_timeout_active = 0; |
4880 | ||
b411b363 PR |
4881 | buf = h; |
4882 | received = 0; | |
257d0af6 | 4883 | expect = sizeof(struct p_header); |
b411b363 PR |
4884 | cmd = NULL; |
4885 | } | |
4886 | } | |
4887 | ||
4888 | if (0) { | |
4889 | reconnect: | |
bbeb641c | 4890 | conn_request_state(tconn, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
b411b363 PR |
4891 | } |
4892 | if (0) { | |
4893 | disconnect: | |
bbeb641c | 4894 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 4895 | } |
32862ec7 | 4896 | clear_bit(SIGNAL_ASENDER, &tconn->flags); |
b411b363 | 4897 | |
32862ec7 | 4898 | conn_info(tconn, "asender terminated\n"); |
b411b363 PR |
4899 | |
4900 | return 0; | |
4901 | } |