]>
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 | ||
65d11ed6 | 63 | static int drbd_do_handshake(struct drbd_tconn *tconn); |
13e6037d | 64 | static int drbd_do_auth(struct drbd_tconn *tconn); |
360cc740 | 65 | static int drbd_disconnected(int vnr, void *p, void *data); |
b411b363 PR |
66 | |
67 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *, struct drbd_epoch *, enum epoch_event); | |
00d56944 | 68 | static int e_end_block(struct drbd_work *, int); |
b411b363 | 69 | |
b411b363 PR |
70 | |
71 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
72 | ||
45bb912b LE |
73 | /* |
74 | * some helper functions to deal with single linked page lists, | |
75 | * page->private being our "next" pointer. | |
76 | */ | |
77 | ||
78 | /* If at least n pages are linked at head, get n pages off. | |
79 | * Otherwise, don't modify head, and return NULL. | |
80 | * Locking is the responsibility of the caller. | |
81 | */ | |
82 | static struct page *page_chain_del(struct page **head, int n) | |
83 | { | |
84 | struct page *page; | |
85 | struct page *tmp; | |
86 | ||
87 | BUG_ON(!n); | |
88 | BUG_ON(!head); | |
89 | ||
90 | page = *head; | |
23ce4227 PR |
91 | |
92 | if (!page) | |
93 | return NULL; | |
94 | ||
45bb912b LE |
95 | while (page) { |
96 | tmp = page_chain_next(page); | |
97 | if (--n == 0) | |
98 | break; /* found sufficient pages */ | |
99 | if (tmp == NULL) | |
100 | /* insufficient pages, don't use any of them. */ | |
101 | return NULL; | |
102 | page = tmp; | |
103 | } | |
104 | ||
105 | /* add end of list marker for the returned list */ | |
106 | set_page_private(page, 0); | |
107 | /* actual return value, and adjustment of head */ | |
108 | page = *head; | |
109 | *head = tmp; | |
110 | return page; | |
111 | } | |
112 | ||
113 | /* may be used outside of locks to find the tail of a (usually short) | |
114 | * "private" page chain, before adding it back to a global chain head | |
115 | * with page_chain_add() under a spinlock. */ | |
116 | static struct page *page_chain_tail(struct page *page, int *len) | |
117 | { | |
118 | struct page *tmp; | |
119 | int i = 1; | |
120 | while ((tmp = page_chain_next(page))) | |
121 | ++i, page = tmp; | |
122 | if (len) | |
123 | *len = i; | |
124 | return page; | |
125 | } | |
126 | ||
127 | static int page_chain_free(struct page *page) | |
128 | { | |
129 | struct page *tmp; | |
130 | int i = 0; | |
131 | page_chain_for_each_safe(page, tmp) { | |
132 | put_page(page); | |
133 | ++i; | |
134 | } | |
135 | return i; | |
136 | } | |
137 | ||
138 | static void page_chain_add(struct page **head, | |
139 | struct page *chain_first, struct page *chain_last) | |
140 | { | |
141 | #if 1 | |
142 | struct page *tmp; | |
143 | tmp = page_chain_tail(chain_first, NULL); | |
144 | BUG_ON(tmp != chain_last); | |
145 | #endif | |
146 | ||
147 | /* add chain to head */ | |
148 | set_page_private(chain_last, (unsigned long)*head); | |
149 | *head = chain_first; | |
150 | } | |
151 | ||
152 | static struct page *drbd_pp_first_pages_or_try_alloc(struct drbd_conf *mdev, int number) | |
b411b363 PR |
153 | { |
154 | struct page *page = NULL; | |
45bb912b LE |
155 | struct page *tmp = NULL; |
156 | int i = 0; | |
b411b363 PR |
157 | |
158 | /* Yes, testing drbd_pp_vacant outside the lock is racy. | |
159 | * So what. It saves a spin_lock. */ | |
45bb912b | 160 | if (drbd_pp_vacant >= number) { |
b411b363 | 161 | spin_lock(&drbd_pp_lock); |
45bb912b LE |
162 | page = page_chain_del(&drbd_pp_pool, number); |
163 | if (page) | |
164 | drbd_pp_vacant -= number; | |
b411b363 | 165 | spin_unlock(&drbd_pp_lock); |
45bb912b LE |
166 | if (page) |
167 | return page; | |
b411b363 | 168 | } |
45bb912b | 169 | |
b411b363 PR |
170 | /* GFP_TRY, because we must not cause arbitrary write-out: in a DRBD |
171 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
172 | * which in turn might block on the other node at this very place. */ | |
45bb912b LE |
173 | for (i = 0; i < number; i++) { |
174 | tmp = alloc_page(GFP_TRY); | |
175 | if (!tmp) | |
176 | break; | |
177 | set_page_private(tmp, (unsigned long)page); | |
178 | page = tmp; | |
179 | } | |
180 | ||
181 | if (i == number) | |
182 | return page; | |
183 | ||
184 | /* Not enough pages immediately available this time. | |
185 | * No need to jump around here, drbd_pp_alloc will retry this | |
186 | * function "soon". */ | |
187 | if (page) { | |
188 | tmp = page_chain_tail(page, NULL); | |
189 | spin_lock(&drbd_pp_lock); | |
190 | page_chain_add(&drbd_pp_pool, page, tmp); | |
191 | drbd_pp_vacant += i; | |
192 | spin_unlock(&drbd_pp_lock); | |
193 | } | |
194 | return NULL; | |
b411b363 PR |
195 | } |
196 | ||
b411b363 PR |
197 | static void reclaim_net_ee(struct drbd_conf *mdev, struct list_head *to_be_freed) |
198 | { | |
db830c46 | 199 | struct drbd_peer_request *peer_req; |
b411b363 PR |
200 | struct list_head *le, *tle; |
201 | ||
202 | /* The EEs are always appended to the end of the list. Since | |
203 | they are sent in order over the wire, they have to finish | |
204 | in order. As soon as we see the first not finished we can | |
205 | stop to examine the list... */ | |
206 | ||
207 | list_for_each_safe(le, tle, &mdev->net_ee) { | |
db830c46 AG |
208 | peer_req = list_entry(le, struct drbd_peer_request, w.list); |
209 | if (drbd_ee_has_active_page(peer_req)) | |
b411b363 PR |
210 | break; |
211 | list_move(le, to_be_freed); | |
212 | } | |
213 | } | |
214 | ||
215 | static void drbd_kick_lo_and_reclaim_net(struct drbd_conf *mdev) | |
216 | { | |
217 | LIST_HEAD(reclaimed); | |
db830c46 | 218 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 219 | |
87eeee41 | 220 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 221 | reclaim_net_ee(mdev, &reclaimed); |
87eeee41 | 222 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 223 | |
db830c46 AG |
224 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
225 | drbd_free_net_ee(mdev, peer_req); | |
b411b363 PR |
226 | } |
227 | ||
228 | /** | |
45bb912b | 229 | * drbd_pp_alloc() - Returns @number pages, retries forever (or until signalled) |
b411b363 | 230 | * @mdev: DRBD device. |
45bb912b LE |
231 | * @number: number of pages requested |
232 | * @retry: whether to retry, if not enough pages are available right now | |
233 | * | |
234 | * Tries to allocate number pages, first from our own page pool, then from | |
235 | * the kernel, unless this allocation would exceed the max_buffers setting. | |
236 | * Possibly retry until DRBD frees sufficient pages somewhere else. | |
b411b363 | 237 | * |
45bb912b | 238 | * Returns a page chain linked via page->private. |
b411b363 | 239 | */ |
45bb912b | 240 | static struct page *drbd_pp_alloc(struct drbd_conf *mdev, unsigned number, bool retry) |
b411b363 PR |
241 | { |
242 | struct page *page = NULL; | |
243 | DEFINE_WAIT(wait); | |
244 | ||
45bb912b LE |
245 | /* Yes, we may run up to @number over max_buffers. If we |
246 | * follow it strictly, the admin will get it wrong anyways. */ | |
89e58e75 | 247 | if (atomic_read(&mdev->pp_in_use) < mdev->tconn->net_conf->max_buffers) |
45bb912b | 248 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 | 249 | |
45bb912b | 250 | while (page == NULL) { |
b411b363 PR |
251 | prepare_to_wait(&drbd_pp_wait, &wait, TASK_INTERRUPTIBLE); |
252 | ||
253 | drbd_kick_lo_and_reclaim_net(mdev); | |
254 | ||
89e58e75 | 255 | if (atomic_read(&mdev->pp_in_use) < mdev->tconn->net_conf->max_buffers) { |
45bb912b | 256 | page = drbd_pp_first_pages_or_try_alloc(mdev, number); |
b411b363 PR |
257 | if (page) |
258 | break; | |
259 | } | |
260 | ||
261 | if (!retry) | |
262 | break; | |
263 | ||
264 | if (signal_pending(current)) { | |
265 | dev_warn(DEV, "drbd_pp_alloc interrupted!\n"); | |
266 | break; | |
267 | } | |
268 | ||
269 | schedule(); | |
270 | } | |
271 | finish_wait(&drbd_pp_wait, &wait); | |
272 | ||
45bb912b LE |
273 | if (page) |
274 | atomic_add(number, &mdev->pp_in_use); | |
b411b363 PR |
275 | return page; |
276 | } | |
277 | ||
278 | /* Must not be used from irq, as that may deadlock: see drbd_pp_alloc. | |
87eeee41 | 279 | * Is also used from inside an other spin_lock_irq(&mdev->tconn->req_lock); |
45bb912b LE |
280 | * Either links the page chain back to the global pool, |
281 | * or returns all pages to the system. */ | |
435f0740 | 282 | static void drbd_pp_free(struct drbd_conf *mdev, struct page *page, int is_net) |
b411b363 | 283 | { |
435f0740 | 284 | atomic_t *a = is_net ? &mdev->pp_in_use_by_net : &mdev->pp_in_use; |
b411b363 | 285 | int i; |
435f0740 | 286 | |
1816a2b4 | 287 | if (drbd_pp_vacant > (DRBD_MAX_BIO_SIZE/PAGE_SIZE)*minor_count) |
45bb912b LE |
288 | i = page_chain_free(page); |
289 | else { | |
290 | struct page *tmp; | |
291 | tmp = page_chain_tail(page, &i); | |
292 | spin_lock(&drbd_pp_lock); | |
293 | page_chain_add(&drbd_pp_pool, page, tmp); | |
294 | drbd_pp_vacant += i; | |
295 | spin_unlock(&drbd_pp_lock); | |
b411b363 | 296 | } |
435f0740 | 297 | i = atomic_sub_return(i, a); |
45bb912b | 298 | if (i < 0) |
435f0740 LE |
299 | dev_warn(DEV, "ASSERTION FAILED: %s: %d < 0\n", |
300 | is_net ? "pp_in_use_by_net" : "pp_in_use", i); | |
b411b363 PR |
301 | wake_up(&drbd_pp_wait); |
302 | } | |
303 | ||
304 | /* | |
305 | You need to hold the req_lock: | |
306 | _drbd_wait_ee_list_empty() | |
307 | ||
308 | You must not have the req_lock: | |
309 | drbd_free_ee() | |
310 | drbd_alloc_ee() | |
311 | drbd_init_ee() | |
312 | drbd_release_ee() | |
313 | drbd_ee_fix_bhs() | |
314 | drbd_process_done_ee() | |
315 | drbd_clear_done_ee() | |
316 | drbd_wait_ee_list_empty() | |
317 | */ | |
318 | ||
f6ffca9f AG |
319 | struct drbd_peer_request * |
320 | drbd_alloc_ee(struct drbd_conf *mdev, u64 id, sector_t sector, | |
321 | unsigned int data_size, gfp_t gfp_mask) __must_hold(local) | |
b411b363 | 322 | { |
db830c46 | 323 | struct drbd_peer_request *peer_req; |
b411b363 | 324 | struct page *page; |
45bb912b | 325 | unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT; |
b411b363 | 326 | |
0cf9d27e | 327 | if (drbd_insert_fault(mdev, DRBD_FAULT_AL_EE)) |
b411b363 PR |
328 | return NULL; |
329 | ||
db830c46 AG |
330 | peer_req = mempool_alloc(drbd_ee_mempool, gfp_mask & ~__GFP_HIGHMEM); |
331 | if (!peer_req) { | |
b411b363 PR |
332 | if (!(gfp_mask & __GFP_NOWARN)) |
333 | dev_err(DEV, "alloc_ee: Allocation of an EE failed\n"); | |
334 | return NULL; | |
335 | } | |
336 | ||
45bb912b LE |
337 | page = drbd_pp_alloc(mdev, nr_pages, (gfp_mask & __GFP_WAIT)); |
338 | if (!page) | |
339 | goto fail; | |
b411b363 | 340 | |
db830c46 AG |
341 | drbd_clear_interval(&peer_req->i); |
342 | peer_req->i.size = data_size; | |
343 | peer_req->i.sector = sector; | |
344 | peer_req->i.local = false; | |
345 | peer_req->i.waiting = false; | |
346 | ||
347 | peer_req->epoch = NULL; | |
a21e9298 | 348 | peer_req->w.mdev = mdev; |
db830c46 AG |
349 | peer_req->pages = page; |
350 | atomic_set(&peer_req->pending_bios, 0); | |
351 | peer_req->flags = 0; | |
9a8e7753 AG |
352 | /* |
353 | * The block_id is opaque to the receiver. It is not endianness | |
354 | * converted, and sent back to the sender unchanged. | |
355 | */ | |
db830c46 | 356 | peer_req->block_id = id; |
b411b363 | 357 | |
db830c46 | 358 | return peer_req; |
b411b363 | 359 | |
45bb912b | 360 | fail: |
db830c46 | 361 | mempool_free(peer_req, drbd_ee_mempool); |
b411b363 PR |
362 | return NULL; |
363 | } | |
364 | ||
db830c46 | 365 | void drbd_free_some_ee(struct drbd_conf *mdev, struct drbd_peer_request *peer_req, |
f6ffca9f | 366 | int is_net) |
b411b363 | 367 | { |
db830c46 AG |
368 | if (peer_req->flags & EE_HAS_DIGEST) |
369 | kfree(peer_req->digest); | |
370 | drbd_pp_free(mdev, peer_req->pages, is_net); | |
371 | D_ASSERT(atomic_read(&peer_req->pending_bios) == 0); | |
372 | D_ASSERT(drbd_interval_empty(&peer_req->i)); | |
373 | mempool_free(peer_req, drbd_ee_mempool); | |
b411b363 PR |
374 | } |
375 | ||
376 | int drbd_release_ee(struct drbd_conf *mdev, struct list_head *list) | |
377 | { | |
378 | LIST_HEAD(work_list); | |
db830c46 | 379 | struct drbd_peer_request *peer_req, *t; |
b411b363 | 380 | int count = 0; |
435f0740 | 381 | int is_net = list == &mdev->net_ee; |
b411b363 | 382 | |
87eeee41 | 383 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 384 | list_splice_init(list, &work_list); |
87eeee41 | 385 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 386 | |
db830c46 AG |
387 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
388 | drbd_free_some_ee(mdev, peer_req, is_net); | |
b411b363 PR |
389 | count++; |
390 | } | |
391 | return count; | |
392 | } | |
393 | ||
394 | ||
32862ec7 | 395 | /* See also comments in _req_mod(,BARRIER_ACKED) |
b411b363 PR |
396 | * and receive_Barrier. |
397 | * | |
398 | * Move entries from net_ee to done_ee, if ready. | |
399 | * Grab done_ee, call all callbacks, free the entries. | |
400 | * The callbacks typically send out ACKs. | |
401 | */ | |
402 | static int drbd_process_done_ee(struct drbd_conf *mdev) | |
403 | { | |
404 | LIST_HEAD(work_list); | |
405 | LIST_HEAD(reclaimed); | |
db830c46 | 406 | struct drbd_peer_request *peer_req, *t; |
b411b363 PR |
407 | int ok = (mdev->state.conn >= C_WF_REPORT_PARAMS); |
408 | ||
87eeee41 | 409 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
410 | reclaim_net_ee(mdev, &reclaimed); |
411 | list_splice_init(&mdev->done_ee, &work_list); | |
87eeee41 | 412 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 413 | |
db830c46 AG |
414 | list_for_each_entry_safe(peer_req, t, &reclaimed, w.list) |
415 | drbd_free_net_ee(mdev, peer_req); | |
b411b363 PR |
416 | |
417 | /* possible callbacks here: | |
7be8da07 | 418 | * e_end_block, and e_end_resync_block, e_send_discard_write. |
b411b363 PR |
419 | * all ignore the last argument. |
420 | */ | |
db830c46 | 421 | list_for_each_entry_safe(peer_req, t, &work_list, w.list) { |
b411b363 | 422 | /* list_del not necessary, next/prev members not touched */ |
00d56944 | 423 | ok = peer_req->w.cb(&peer_req->w, !ok) && ok; |
db830c46 | 424 | drbd_free_ee(mdev, peer_req); |
b411b363 PR |
425 | } |
426 | wake_up(&mdev->ee_wait); | |
427 | ||
428 | return ok; | |
429 | } | |
430 | ||
431 | void _drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
432 | { | |
433 | DEFINE_WAIT(wait); | |
434 | ||
435 | /* avoids spin_lock/unlock | |
436 | * and calling prepare_to_wait in the fast path */ | |
437 | while (!list_empty(head)) { | |
438 | prepare_to_wait(&mdev->ee_wait, &wait, TASK_UNINTERRUPTIBLE); | |
87eeee41 | 439 | spin_unlock_irq(&mdev->tconn->req_lock); |
7eaceacc | 440 | io_schedule(); |
b411b363 | 441 | finish_wait(&mdev->ee_wait, &wait); |
87eeee41 | 442 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
443 | } |
444 | } | |
445 | ||
446 | void drbd_wait_ee_list_empty(struct drbd_conf *mdev, struct list_head *head) | |
447 | { | |
87eeee41 | 448 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 449 | _drbd_wait_ee_list_empty(mdev, head); |
87eeee41 | 450 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
451 | } |
452 | ||
453 | /* see also kernel_accept; which is only present since 2.6.18. | |
454 | * also we want to log which part of it failed, exactly */ | |
7653620d | 455 | static int drbd_accept(const char **what, struct socket *sock, struct socket **newsock) |
b411b363 PR |
456 | { |
457 | struct sock *sk = sock->sk; | |
458 | int err = 0; | |
459 | ||
460 | *what = "listen"; | |
461 | err = sock->ops->listen(sock, 5); | |
462 | if (err < 0) | |
463 | goto out; | |
464 | ||
465 | *what = "sock_create_lite"; | |
466 | err = sock_create_lite(sk->sk_family, sk->sk_type, sk->sk_protocol, | |
467 | newsock); | |
468 | if (err < 0) | |
469 | goto out; | |
470 | ||
471 | *what = "accept"; | |
472 | err = sock->ops->accept(sock, *newsock, 0); | |
473 | if (err < 0) { | |
474 | sock_release(*newsock); | |
475 | *newsock = NULL; | |
476 | goto out; | |
477 | } | |
478 | (*newsock)->ops = sock->ops; | |
479 | ||
480 | out: | |
481 | return err; | |
482 | } | |
483 | ||
dbd9eea0 | 484 | static int drbd_recv_short(struct socket *sock, void *buf, size_t size, int flags) |
b411b363 PR |
485 | { |
486 | mm_segment_t oldfs; | |
487 | struct kvec iov = { | |
488 | .iov_base = buf, | |
489 | .iov_len = size, | |
490 | }; | |
491 | struct msghdr msg = { | |
492 | .msg_iovlen = 1, | |
493 | .msg_iov = (struct iovec *)&iov, | |
494 | .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL) | |
495 | }; | |
496 | int rv; | |
497 | ||
498 | oldfs = get_fs(); | |
499 | set_fs(KERNEL_DS); | |
500 | rv = sock_recvmsg(sock, &msg, size, msg.msg_flags); | |
501 | set_fs(oldfs); | |
502 | ||
503 | return rv; | |
504 | } | |
505 | ||
de0ff338 | 506 | static int drbd_recv(struct drbd_tconn *tconn, void *buf, size_t size) |
b411b363 PR |
507 | { |
508 | mm_segment_t oldfs; | |
509 | struct kvec iov = { | |
510 | .iov_base = buf, | |
511 | .iov_len = size, | |
512 | }; | |
513 | struct msghdr msg = { | |
514 | .msg_iovlen = 1, | |
515 | .msg_iov = (struct iovec *)&iov, | |
516 | .msg_flags = MSG_WAITALL | MSG_NOSIGNAL | |
517 | }; | |
518 | int rv; | |
519 | ||
520 | oldfs = get_fs(); | |
521 | set_fs(KERNEL_DS); | |
522 | ||
523 | for (;;) { | |
de0ff338 | 524 | rv = sock_recvmsg(tconn->data.socket, &msg, size, msg.msg_flags); |
b411b363 PR |
525 | if (rv == size) |
526 | break; | |
527 | ||
528 | /* Note: | |
529 | * ECONNRESET other side closed the connection | |
530 | * ERESTARTSYS (on sock) we got a signal | |
531 | */ | |
532 | ||
533 | if (rv < 0) { | |
534 | if (rv == -ECONNRESET) | |
de0ff338 | 535 | conn_info(tconn, "sock was reset by peer\n"); |
b411b363 | 536 | else if (rv != -ERESTARTSYS) |
de0ff338 | 537 | conn_err(tconn, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
538 | break; |
539 | } else if (rv == 0) { | |
de0ff338 | 540 | conn_info(tconn, "sock was shut down by peer\n"); |
b411b363 PR |
541 | break; |
542 | } else { | |
543 | /* signal came in, or peer/link went down, | |
544 | * after we read a partial message | |
545 | */ | |
546 | /* D_ASSERT(signal_pending(current)); */ | |
547 | break; | |
548 | } | |
549 | }; | |
550 | ||
551 | set_fs(oldfs); | |
552 | ||
553 | if (rv != size) | |
bbeb641c | 554 | conn_request_state(tconn, NS(conn, C_BROKEN_PIPE), CS_HARD); |
b411b363 PR |
555 | |
556 | return rv; | |
557 | } | |
558 | ||
5dbf1673 LE |
559 | /* quoting tcp(7): |
560 | * On individual connections, the socket buffer size must be set prior to the | |
561 | * listen(2) or connect(2) calls in order to have it take effect. | |
562 | * This is our wrapper to do so. | |
563 | */ | |
564 | static void drbd_setbufsize(struct socket *sock, unsigned int snd, | |
565 | unsigned int rcv) | |
566 | { | |
567 | /* open coded SO_SNDBUF, SO_RCVBUF */ | |
568 | if (snd) { | |
569 | sock->sk->sk_sndbuf = snd; | |
570 | sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK; | |
571 | } | |
572 | if (rcv) { | |
573 | sock->sk->sk_rcvbuf = rcv; | |
574 | sock->sk->sk_userlocks |= SOCK_RCVBUF_LOCK; | |
575 | } | |
576 | } | |
577 | ||
eac3e990 | 578 | static struct socket *drbd_try_connect(struct drbd_tconn *tconn) |
b411b363 PR |
579 | { |
580 | const char *what; | |
581 | struct socket *sock; | |
582 | struct sockaddr_in6 src_in6; | |
583 | int err; | |
584 | int disconnect_on_error = 1; | |
585 | ||
eac3e990 | 586 | if (!get_net_conf(tconn)) |
b411b363 PR |
587 | return NULL; |
588 | ||
589 | what = "sock_create_kern"; | |
eac3e990 | 590 | err = sock_create_kern(((struct sockaddr *)tconn->net_conf->my_addr)->sa_family, |
b411b363 PR |
591 | SOCK_STREAM, IPPROTO_TCP, &sock); |
592 | if (err < 0) { | |
593 | sock = NULL; | |
594 | goto out; | |
595 | } | |
596 | ||
597 | sock->sk->sk_rcvtimeo = | |
eac3e990 PR |
598 | sock->sk->sk_sndtimeo = tconn->net_conf->try_connect_int*HZ; |
599 | drbd_setbufsize(sock, tconn->net_conf->sndbuf_size, | |
600 | tconn->net_conf->rcvbuf_size); | |
b411b363 PR |
601 | |
602 | /* explicitly bind to the configured IP as source IP | |
603 | * for the outgoing connections. | |
604 | * This is needed for multihomed hosts and to be | |
605 | * able to use lo: interfaces for drbd. | |
606 | * Make sure to use 0 as port number, so linux selects | |
607 | * a free one dynamically. | |
608 | */ | |
eac3e990 PR |
609 | memcpy(&src_in6, tconn->net_conf->my_addr, |
610 | min_t(int, tconn->net_conf->my_addr_len, sizeof(src_in6))); | |
611 | if (((struct sockaddr *)tconn->net_conf->my_addr)->sa_family == AF_INET6) | |
b411b363 PR |
612 | src_in6.sin6_port = 0; |
613 | else | |
614 | ((struct sockaddr_in *)&src_in6)->sin_port = 0; /* AF_INET & AF_SCI */ | |
615 | ||
616 | what = "bind before connect"; | |
617 | err = sock->ops->bind(sock, | |
618 | (struct sockaddr *) &src_in6, | |
eac3e990 | 619 | tconn->net_conf->my_addr_len); |
b411b363 PR |
620 | if (err < 0) |
621 | goto out; | |
622 | ||
623 | /* connect may fail, peer not yet available. | |
624 | * stay C_WF_CONNECTION, don't go Disconnecting! */ | |
625 | disconnect_on_error = 0; | |
626 | what = "connect"; | |
627 | err = sock->ops->connect(sock, | |
eac3e990 PR |
628 | (struct sockaddr *)tconn->net_conf->peer_addr, |
629 | tconn->net_conf->peer_addr_len, 0); | |
b411b363 PR |
630 | |
631 | out: | |
632 | if (err < 0) { | |
633 | if (sock) { | |
634 | sock_release(sock); | |
635 | sock = NULL; | |
636 | } | |
637 | switch (-err) { | |
638 | /* timeout, busy, signal pending */ | |
639 | case ETIMEDOUT: case EAGAIN: case EINPROGRESS: | |
640 | case EINTR: case ERESTARTSYS: | |
641 | /* peer not (yet) available, network problem */ | |
642 | case ECONNREFUSED: case ENETUNREACH: | |
643 | case EHOSTDOWN: case EHOSTUNREACH: | |
644 | disconnect_on_error = 0; | |
645 | break; | |
646 | default: | |
eac3e990 | 647 | conn_err(tconn, "%s failed, err = %d\n", what, err); |
b411b363 PR |
648 | } |
649 | if (disconnect_on_error) | |
bbeb641c | 650 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 651 | } |
eac3e990 | 652 | put_net_conf(tconn); |
b411b363 PR |
653 | return sock; |
654 | } | |
655 | ||
7653620d | 656 | static struct socket *drbd_wait_for_connect(struct drbd_tconn *tconn) |
b411b363 PR |
657 | { |
658 | int timeo, err; | |
659 | struct socket *s_estab = NULL, *s_listen; | |
660 | const char *what; | |
661 | ||
7653620d | 662 | if (!get_net_conf(tconn)) |
b411b363 PR |
663 | return NULL; |
664 | ||
665 | what = "sock_create_kern"; | |
7653620d | 666 | err = sock_create_kern(((struct sockaddr *)tconn->net_conf->my_addr)->sa_family, |
b411b363 PR |
667 | SOCK_STREAM, IPPROTO_TCP, &s_listen); |
668 | if (err) { | |
669 | s_listen = NULL; | |
670 | goto out; | |
671 | } | |
672 | ||
7653620d | 673 | timeo = tconn->net_conf->try_connect_int * HZ; |
b411b363 PR |
674 | timeo += (random32() & 1) ? timeo / 7 : -timeo / 7; /* 28.5% random jitter */ |
675 | ||
676 | s_listen->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
677 | s_listen->sk->sk_rcvtimeo = timeo; | |
678 | s_listen->sk->sk_sndtimeo = timeo; | |
7653620d PR |
679 | drbd_setbufsize(s_listen, tconn->net_conf->sndbuf_size, |
680 | tconn->net_conf->rcvbuf_size); | |
b411b363 PR |
681 | |
682 | what = "bind before listen"; | |
683 | err = s_listen->ops->bind(s_listen, | |
7653620d PR |
684 | (struct sockaddr *) tconn->net_conf->my_addr, |
685 | tconn->net_conf->my_addr_len); | |
b411b363 PR |
686 | if (err < 0) |
687 | goto out; | |
688 | ||
7653620d | 689 | err = drbd_accept(&what, s_listen, &s_estab); |
b411b363 PR |
690 | |
691 | out: | |
692 | if (s_listen) | |
693 | sock_release(s_listen); | |
694 | if (err < 0) { | |
695 | if (err != -EAGAIN && err != -EINTR && err != -ERESTARTSYS) { | |
7653620d | 696 | conn_err(tconn, "%s failed, err = %d\n", what, err); |
bbeb641c | 697 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
698 | } |
699 | } | |
7653620d | 700 | put_net_conf(tconn); |
b411b363 PR |
701 | |
702 | return s_estab; | |
703 | } | |
704 | ||
d38e787e | 705 | static int drbd_send_fp(struct drbd_tconn *tconn, struct socket *sock, enum drbd_packet cmd) |
b411b363 | 706 | { |
d38e787e | 707 | struct p_header *h = &tconn->data.sbuf.header; |
b411b363 | 708 | |
d38e787e | 709 | return _conn_send_cmd(tconn, 0, sock, cmd, h, sizeof(*h), 0); |
b411b363 PR |
710 | } |
711 | ||
a25b63f1 | 712 | static enum drbd_packet drbd_recv_fp(struct drbd_tconn *tconn, struct socket *sock) |
b411b363 | 713 | { |
a25b63f1 | 714 | struct p_header80 *h = &tconn->data.rbuf.header.h80; |
b411b363 PR |
715 | int rr; |
716 | ||
dbd9eea0 | 717 | rr = drbd_recv_short(sock, h, sizeof(*h), 0); |
b411b363 | 718 | |
ca9bc12b | 719 | if (rr == sizeof(*h) && h->magic == cpu_to_be32(DRBD_MAGIC)) |
b411b363 PR |
720 | return be16_to_cpu(h->command); |
721 | ||
722 | return 0xffff; | |
723 | } | |
724 | ||
725 | /** | |
726 | * drbd_socket_okay() - Free the socket if its connection is not okay | |
b411b363 PR |
727 | * @sock: pointer to the pointer to the socket. |
728 | */ | |
dbd9eea0 | 729 | static int drbd_socket_okay(struct socket **sock) |
b411b363 PR |
730 | { |
731 | int rr; | |
732 | char tb[4]; | |
733 | ||
734 | if (!*sock) | |
81e84650 | 735 | return false; |
b411b363 | 736 | |
dbd9eea0 | 737 | rr = drbd_recv_short(*sock, tb, 4, MSG_DONTWAIT | MSG_PEEK); |
b411b363 PR |
738 | |
739 | if (rr > 0 || rr == -EAGAIN) { | |
81e84650 | 740 | return true; |
b411b363 PR |
741 | } else { |
742 | sock_release(*sock); | |
743 | *sock = NULL; | |
81e84650 | 744 | return false; |
b411b363 PR |
745 | } |
746 | } | |
747 | ||
907599e0 PR |
748 | static int drbd_connected(int vnr, void *p, void *data) |
749 | { | |
750 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
751 | int ok = 1; | |
752 | ||
753 | atomic_set(&mdev->packet_seq, 0); | |
754 | mdev->peer_seq = 0; | |
755 | ||
8410da8f PR |
756 | mdev->state_mutex = mdev->tconn->agreed_pro_version < 100 ? |
757 | &mdev->tconn->cstate_mutex : | |
758 | &mdev->own_state_mutex; | |
759 | ||
907599e0 PR |
760 | ok &= drbd_send_sync_param(mdev, &mdev->sync_conf); |
761 | ok &= drbd_send_sizes(mdev, 0, 0); | |
762 | ok &= drbd_send_uuids(mdev); | |
763 | ok &= drbd_send_state(mdev); | |
764 | clear_bit(USE_DEGR_WFC_T, &mdev->flags); | |
765 | clear_bit(RESIZE_PENDING, &mdev->flags); | |
766 | ||
8410da8f | 767 | |
907599e0 PR |
768 | return !ok; |
769 | } | |
770 | ||
b411b363 PR |
771 | /* |
772 | * return values: | |
773 | * 1 yes, we have a valid connection | |
774 | * 0 oops, did not work out, please try again | |
775 | * -1 peer talks different language, | |
776 | * no point in trying again, please go standalone. | |
777 | * -2 We do not have a network config... | |
778 | */ | |
907599e0 | 779 | static int drbd_connect(struct drbd_tconn *tconn) |
b411b363 PR |
780 | { |
781 | struct socket *s, *sock, *msock; | |
782 | int try, h, ok; | |
783 | ||
bbeb641c | 784 | if (conn_request_state(tconn, NS(conn, C_WF_CONNECTION), CS_VERBOSE) < SS_SUCCESS) |
b411b363 PR |
785 | return -2; |
786 | ||
907599e0 PR |
787 | clear_bit(DISCARD_CONCURRENT, &tconn->flags); |
788 | tconn->agreed_pro_version = 99; | |
fd340c12 PR |
789 | /* agreed_pro_version must be smaller than 100 so we send the old |
790 | header (h80) in the first packet and in the handshake packet. */ | |
b411b363 PR |
791 | |
792 | sock = NULL; | |
793 | msock = NULL; | |
794 | ||
795 | do { | |
796 | for (try = 0;;) { | |
797 | /* 3 tries, this should take less than a second! */ | |
907599e0 | 798 | s = drbd_try_connect(tconn); |
b411b363 PR |
799 | if (s || ++try >= 3) |
800 | break; | |
801 | /* give the other side time to call bind() & listen() */ | |
20ee6390 | 802 | schedule_timeout_interruptible(HZ / 10); |
b411b363 PR |
803 | } |
804 | ||
805 | if (s) { | |
806 | if (!sock) { | |
907599e0 | 807 | drbd_send_fp(tconn, s, P_HAND_SHAKE_S); |
b411b363 PR |
808 | sock = s; |
809 | s = NULL; | |
810 | } else if (!msock) { | |
907599e0 | 811 | drbd_send_fp(tconn, s, P_HAND_SHAKE_M); |
b411b363 PR |
812 | msock = s; |
813 | s = NULL; | |
814 | } else { | |
907599e0 | 815 | conn_err(tconn, "Logic error in drbd_connect()\n"); |
b411b363 PR |
816 | goto out_release_sockets; |
817 | } | |
818 | } | |
819 | ||
820 | if (sock && msock) { | |
907599e0 | 821 | schedule_timeout_interruptible(tconn->net_conf->ping_timeo*HZ/10); |
dbd9eea0 PR |
822 | ok = drbd_socket_okay(&sock); |
823 | ok = drbd_socket_okay(&msock) && ok; | |
b411b363 PR |
824 | if (ok) |
825 | break; | |
826 | } | |
827 | ||
828 | retry: | |
907599e0 | 829 | s = drbd_wait_for_connect(tconn); |
b411b363 | 830 | if (s) { |
907599e0 | 831 | try = drbd_recv_fp(tconn, s); |
dbd9eea0 PR |
832 | drbd_socket_okay(&sock); |
833 | drbd_socket_okay(&msock); | |
b411b363 PR |
834 | switch (try) { |
835 | case P_HAND_SHAKE_S: | |
836 | if (sock) { | |
907599e0 | 837 | conn_warn(tconn, "initial packet S crossed\n"); |
b411b363 PR |
838 | sock_release(sock); |
839 | } | |
840 | sock = s; | |
841 | break; | |
842 | case P_HAND_SHAKE_M: | |
843 | if (msock) { | |
907599e0 | 844 | conn_warn(tconn, "initial packet M crossed\n"); |
b411b363 PR |
845 | sock_release(msock); |
846 | } | |
847 | msock = s; | |
907599e0 | 848 | set_bit(DISCARD_CONCURRENT, &tconn->flags); |
b411b363 PR |
849 | break; |
850 | default: | |
907599e0 | 851 | conn_warn(tconn, "Error receiving initial packet\n"); |
b411b363 PR |
852 | sock_release(s); |
853 | if (random32() & 1) | |
854 | goto retry; | |
855 | } | |
856 | } | |
857 | ||
bbeb641c | 858 | if (tconn->cstate <= C_DISCONNECTING) |
b411b363 PR |
859 | goto out_release_sockets; |
860 | if (signal_pending(current)) { | |
861 | flush_signals(current); | |
862 | smp_rmb(); | |
907599e0 | 863 | if (get_t_state(&tconn->receiver) == EXITING) |
b411b363 PR |
864 | goto out_release_sockets; |
865 | } | |
866 | ||
867 | if (sock && msock) { | |
dbd9eea0 PR |
868 | ok = drbd_socket_okay(&sock); |
869 | ok = drbd_socket_okay(&msock) && ok; | |
b411b363 PR |
870 | if (ok) |
871 | break; | |
872 | } | |
873 | } while (1); | |
874 | ||
875 | msock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
876 | sock->sk->sk_reuse = 1; /* SO_REUSEADDR */ | |
877 | ||
878 | sock->sk->sk_allocation = GFP_NOIO; | |
879 | msock->sk->sk_allocation = GFP_NOIO; | |
880 | ||
881 | sock->sk->sk_priority = TC_PRIO_INTERACTIVE_BULK; | |
882 | msock->sk->sk_priority = TC_PRIO_INTERACTIVE; | |
883 | ||
b411b363 | 884 | /* NOT YET ... |
907599e0 | 885 | * sock->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
b411b363 PR |
886 | * sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
887 | * first set it to the P_HAND_SHAKE timeout, | |
888 | * which we set to 4x the configured ping_timeout. */ | |
889 | sock->sk->sk_sndtimeo = | |
907599e0 | 890 | sock->sk->sk_rcvtimeo = tconn->net_conf->ping_timeo*4*HZ/10; |
b411b363 | 891 | |
907599e0 PR |
892 | msock->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
893 | msock->sk->sk_rcvtimeo = tconn->net_conf->ping_int*HZ; | |
b411b363 PR |
894 | |
895 | /* we don't want delays. | |
25985edc | 896 | * we use TCP_CORK where appropriate, though */ |
b411b363 PR |
897 | drbd_tcp_nodelay(sock); |
898 | drbd_tcp_nodelay(msock); | |
899 | ||
907599e0 PR |
900 | tconn->data.socket = sock; |
901 | tconn->meta.socket = msock; | |
902 | tconn->last_received = jiffies; | |
b411b363 | 903 | |
907599e0 | 904 | h = drbd_do_handshake(tconn); |
b411b363 PR |
905 | if (h <= 0) |
906 | return h; | |
907 | ||
907599e0 | 908 | if (tconn->cram_hmac_tfm) { |
b411b363 | 909 | /* drbd_request_state(mdev, NS(conn, WFAuth)); */ |
907599e0 | 910 | switch (drbd_do_auth(tconn)) { |
b10d96cb | 911 | case -1: |
907599e0 | 912 | conn_err(tconn, "Authentication of peer failed\n"); |
b411b363 | 913 | return -1; |
b10d96cb | 914 | case 0: |
907599e0 | 915 | conn_err(tconn, "Authentication of peer failed, trying again.\n"); |
b10d96cb | 916 | return 0; |
b411b363 PR |
917 | } |
918 | } | |
919 | ||
bbeb641c | 920 | if (conn_request_state(tconn, NS(conn, C_WF_REPORT_PARAMS), CS_VERBOSE) < SS_SUCCESS) |
b411b363 PR |
921 | return 0; |
922 | ||
907599e0 | 923 | sock->sk->sk_sndtimeo = tconn->net_conf->timeout*HZ/10; |
b411b363 PR |
924 | sock->sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; |
925 | ||
907599e0 | 926 | drbd_thread_start(&tconn->asender); |
b411b363 | 927 | |
907599e0 | 928 | if (drbd_send_protocol(tconn) == -1) |
7e2455c1 | 929 | return -1; |
b411b363 | 930 | |
907599e0 | 931 | return !idr_for_each(&tconn->volumes, drbd_connected, tconn); |
b411b363 PR |
932 | |
933 | out_release_sockets: | |
934 | if (sock) | |
935 | sock_release(sock); | |
936 | if (msock) | |
937 | sock_release(msock); | |
938 | return -1; | |
939 | } | |
940 | ||
ce243853 | 941 | static bool decode_header(struct drbd_tconn *tconn, struct p_header *h, struct packet_info *pi) |
b411b363 | 942 | { |
fd340c12 | 943 | if (h->h80.magic == cpu_to_be32(DRBD_MAGIC)) { |
77351055 PR |
944 | pi->cmd = be16_to_cpu(h->h80.command); |
945 | pi->size = be16_to_cpu(h->h80.length); | |
eefc2f7d | 946 | pi->vnr = 0; |
ca9bc12b | 947 | } else if (h->h95.magic == cpu_to_be16(DRBD_MAGIC_BIG)) { |
77351055 PR |
948 | pi->cmd = be16_to_cpu(h->h95.command); |
949 | pi->size = be32_to_cpu(h->h95.length) & 0x00ffffff; | |
950 | pi->vnr = 0; | |
02918be2 | 951 | } else { |
ce243853 | 952 | conn_err(tconn, "magic?? on data m: 0x%08x c: %d l: %d\n", |
004352fa LE |
953 | be32_to_cpu(h->h80.magic), |
954 | be16_to_cpu(h->h80.command), | |
955 | be16_to_cpu(h->h80.length)); | |
81e84650 | 956 | return false; |
b411b363 | 957 | } |
257d0af6 PR |
958 | return true; |
959 | } | |
960 | ||
9ba7aa00 | 961 | static int drbd_recv_header(struct drbd_tconn *tconn, struct packet_info *pi) |
257d0af6 | 962 | { |
9ba7aa00 | 963 | struct p_header *h = &tconn->data.rbuf.header; |
257d0af6 PR |
964 | int r; |
965 | ||
9ba7aa00 | 966 | r = drbd_recv(tconn, h, sizeof(*h)); |
257d0af6 PR |
967 | if (unlikely(r != sizeof(*h))) { |
968 | if (!signal_pending(current)) | |
9ba7aa00 | 969 | conn_warn(tconn, "short read expecting header on sock: r=%d\n", r); |
257d0af6 PR |
970 | return false; |
971 | } | |
972 | ||
9ba7aa00 PR |
973 | r = decode_header(tconn, h, pi); |
974 | tconn->last_received = jiffies; | |
b411b363 | 975 | |
257d0af6 | 976 | return r; |
b411b363 PR |
977 | } |
978 | ||
2451fc3b | 979 | static void drbd_flush(struct drbd_conf *mdev) |
b411b363 PR |
980 | { |
981 | int rv; | |
982 | ||
983 | if (mdev->write_ordering >= WO_bdev_flush && get_ldev(mdev)) { | |
fbd9b09a | 984 | rv = blkdev_issue_flush(mdev->ldev->backing_bdev, GFP_KERNEL, |
dd3932ed | 985 | NULL); |
b411b363 PR |
986 | if (rv) { |
987 | dev_err(DEV, "local disk flush failed with status %d\n", rv); | |
988 | /* would rather check on EOPNOTSUPP, but that is not reliable. | |
989 | * don't try again for ANY return value != 0 | |
990 | * if (rv == -EOPNOTSUPP) */ | |
991 | drbd_bump_write_ordering(mdev, WO_drain_io); | |
992 | } | |
993 | put_ldev(mdev); | |
994 | } | |
b411b363 PR |
995 | } |
996 | ||
997 | /** | |
998 | * drbd_may_finish_epoch() - Applies an epoch_event to the epoch's state, eventually finishes it. | |
999 | * @mdev: DRBD device. | |
1000 | * @epoch: Epoch object. | |
1001 | * @ev: Epoch event. | |
1002 | */ | |
1003 | static enum finish_epoch drbd_may_finish_epoch(struct drbd_conf *mdev, | |
1004 | struct drbd_epoch *epoch, | |
1005 | enum epoch_event ev) | |
1006 | { | |
2451fc3b | 1007 | int epoch_size; |
b411b363 | 1008 | struct drbd_epoch *next_epoch; |
b411b363 PR |
1009 | enum finish_epoch rv = FE_STILL_LIVE; |
1010 | ||
1011 | spin_lock(&mdev->epoch_lock); | |
1012 | do { | |
1013 | next_epoch = NULL; | |
b411b363 PR |
1014 | |
1015 | epoch_size = atomic_read(&epoch->epoch_size); | |
1016 | ||
1017 | switch (ev & ~EV_CLEANUP) { | |
1018 | case EV_PUT: | |
1019 | atomic_dec(&epoch->active); | |
1020 | break; | |
1021 | case EV_GOT_BARRIER_NR: | |
1022 | set_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags); | |
b411b363 PR |
1023 | break; |
1024 | case EV_BECAME_LAST: | |
1025 | /* nothing to do*/ | |
1026 | break; | |
1027 | } | |
1028 | ||
b411b363 PR |
1029 | if (epoch_size != 0 && |
1030 | atomic_read(&epoch->active) == 0 && | |
2451fc3b | 1031 | test_bit(DE_HAVE_BARRIER_NUMBER, &epoch->flags)) { |
b411b363 PR |
1032 | if (!(ev & EV_CLEANUP)) { |
1033 | spin_unlock(&mdev->epoch_lock); | |
1034 | drbd_send_b_ack(mdev, epoch->barrier_nr, epoch_size); | |
1035 | spin_lock(&mdev->epoch_lock); | |
1036 | } | |
1037 | dec_unacked(mdev); | |
1038 | ||
1039 | if (mdev->current_epoch != epoch) { | |
1040 | next_epoch = list_entry(epoch->list.next, struct drbd_epoch, list); | |
1041 | list_del(&epoch->list); | |
1042 | ev = EV_BECAME_LAST | (ev & EV_CLEANUP); | |
1043 | mdev->epochs--; | |
b411b363 PR |
1044 | kfree(epoch); |
1045 | ||
1046 | if (rv == FE_STILL_LIVE) | |
1047 | rv = FE_DESTROYED; | |
1048 | } else { | |
1049 | epoch->flags = 0; | |
1050 | atomic_set(&epoch->epoch_size, 0); | |
698f9315 | 1051 | /* atomic_set(&epoch->active, 0); is already zero */ |
b411b363 PR |
1052 | if (rv == FE_STILL_LIVE) |
1053 | rv = FE_RECYCLED; | |
2451fc3b | 1054 | wake_up(&mdev->ee_wait); |
b411b363 PR |
1055 | } |
1056 | } | |
1057 | ||
1058 | if (!next_epoch) | |
1059 | break; | |
1060 | ||
1061 | epoch = next_epoch; | |
1062 | } while (1); | |
1063 | ||
1064 | spin_unlock(&mdev->epoch_lock); | |
1065 | ||
b411b363 PR |
1066 | return rv; |
1067 | } | |
1068 | ||
1069 | /** | |
1070 | * drbd_bump_write_ordering() - Fall back to an other write ordering method | |
1071 | * @mdev: DRBD device. | |
1072 | * @wo: Write ordering method to try. | |
1073 | */ | |
1074 | void drbd_bump_write_ordering(struct drbd_conf *mdev, enum write_ordering_e wo) __must_hold(local) | |
1075 | { | |
1076 | enum write_ordering_e pwo; | |
1077 | static char *write_ordering_str[] = { | |
1078 | [WO_none] = "none", | |
1079 | [WO_drain_io] = "drain", | |
1080 | [WO_bdev_flush] = "flush", | |
b411b363 PR |
1081 | }; |
1082 | ||
1083 | pwo = mdev->write_ordering; | |
1084 | wo = min(pwo, wo); | |
b411b363 PR |
1085 | if (wo == WO_bdev_flush && mdev->ldev->dc.no_disk_flush) |
1086 | wo = WO_drain_io; | |
1087 | if (wo == WO_drain_io && mdev->ldev->dc.no_disk_drain) | |
1088 | wo = WO_none; | |
1089 | mdev->write_ordering = wo; | |
2451fc3b | 1090 | if (pwo != mdev->write_ordering || wo == WO_bdev_flush) |
b411b363 PR |
1091 | dev_info(DEV, "Method to ensure write ordering: %s\n", write_ordering_str[mdev->write_ordering]); |
1092 | } | |
1093 | ||
45bb912b | 1094 | /** |
fbe29dec | 1095 | * drbd_submit_peer_request() |
45bb912b | 1096 | * @mdev: DRBD device. |
db830c46 | 1097 | * @peer_req: peer request |
45bb912b | 1098 | * @rw: flag field, see bio->bi_rw |
10f6d992 LE |
1099 | * |
1100 | * May spread the pages to multiple bios, | |
1101 | * depending on bio_add_page restrictions. | |
1102 | * | |
1103 | * Returns 0 if all bios have been submitted, | |
1104 | * -ENOMEM if we could not allocate enough bios, | |
1105 | * -ENOSPC (any better suggestion?) if we have not been able to bio_add_page a | |
1106 | * single page to an empty bio (which should never happen and likely indicates | |
1107 | * that the lower level IO stack is in some way broken). This has been observed | |
1108 | * on certain Xen deployments. | |
45bb912b LE |
1109 | */ |
1110 | /* TODO allocate from our own bio_set. */ | |
fbe29dec AG |
1111 | int drbd_submit_peer_request(struct drbd_conf *mdev, |
1112 | struct drbd_peer_request *peer_req, | |
1113 | const unsigned rw, const int fault_type) | |
45bb912b LE |
1114 | { |
1115 | struct bio *bios = NULL; | |
1116 | struct bio *bio; | |
db830c46 AG |
1117 | struct page *page = peer_req->pages; |
1118 | sector_t sector = peer_req->i.sector; | |
1119 | unsigned ds = peer_req->i.size; | |
45bb912b LE |
1120 | unsigned n_bios = 0; |
1121 | unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT; | |
10f6d992 | 1122 | int err = -ENOMEM; |
45bb912b LE |
1123 | |
1124 | /* In most cases, we will only need one bio. But in case the lower | |
1125 | * level restrictions happen to be different at this offset on this | |
1126 | * side than those of the sending peer, we may need to submit the | |
1127 | * request in more than one bio. */ | |
1128 | next_bio: | |
1129 | bio = bio_alloc(GFP_NOIO, nr_pages); | |
1130 | if (!bio) { | |
1131 | dev_err(DEV, "submit_ee: Allocation of a bio failed\n"); | |
1132 | goto fail; | |
1133 | } | |
db830c46 | 1134 | /* > peer_req->i.sector, unless this is the first bio */ |
45bb912b LE |
1135 | bio->bi_sector = sector; |
1136 | bio->bi_bdev = mdev->ldev->backing_bdev; | |
45bb912b | 1137 | bio->bi_rw = rw; |
db830c46 | 1138 | bio->bi_private = peer_req; |
fcefa62e | 1139 | bio->bi_end_io = drbd_peer_request_endio; |
45bb912b LE |
1140 | |
1141 | bio->bi_next = bios; | |
1142 | bios = bio; | |
1143 | ++n_bios; | |
1144 | ||
1145 | page_chain_for_each(page) { | |
1146 | unsigned len = min_t(unsigned, ds, PAGE_SIZE); | |
1147 | if (!bio_add_page(bio, page, len, 0)) { | |
10f6d992 LE |
1148 | /* A single page must always be possible! |
1149 | * But in case it fails anyways, | |
1150 | * we deal with it, and complain (below). */ | |
1151 | if (bio->bi_vcnt == 0) { | |
1152 | dev_err(DEV, | |
1153 | "bio_add_page failed for len=%u, " | |
1154 | "bi_vcnt=0 (bi_sector=%llu)\n", | |
1155 | len, (unsigned long long)bio->bi_sector); | |
1156 | err = -ENOSPC; | |
1157 | goto fail; | |
1158 | } | |
45bb912b LE |
1159 | goto next_bio; |
1160 | } | |
1161 | ds -= len; | |
1162 | sector += len >> 9; | |
1163 | --nr_pages; | |
1164 | } | |
1165 | D_ASSERT(page == NULL); | |
1166 | D_ASSERT(ds == 0); | |
1167 | ||
db830c46 | 1168 | atomic_set(&peer_req->pending_bios, n_bios); |
45bb912b LE |
1169 | do { |
1170 | bio = bios; | |
1171 | bios = bios->bi_next; | |
1172 | bio->bi_next = NULL; | |
1173 | ||
45bb912b | 1174 | drbd_generic_make_request(mdev, fault_type, bio); |
45bb912b | 1175 | } while (bios); |
45bb912b LE |
1176 | return 0; |
1177 | ||
1178 | fail: | |
1179 | while (bios) { | |
1180 | bio = bios; | |
1181 | bios = bios->bi_next; | |
1182 | bio_put(bio); | |
1183 | } | |
10f6d992 | 1184 | return err; |
45bb912b LE |
1185 | } |
1186 | ||
53840641 | 1187 | static void drbd_remove_epoch_entry_interval(struct drbd_conf *mdev, |
db830c46 | 1188 | struct drbd_peer_request *peer_req) |
53840641 | 1189 | { |
db830c46 | 1190 | struct drbd_interval *i = &peer_req->i; |
53840641 AG |
1191 | |
1192 | drbd_remove_interval(&mdev->write_requests, i); | |
1193 | drbd_clear_interval(i); | |
1194 | ||
6c852bec | 1195 | /* Wake up any processes waiting for this peer request to complete. */ |
53840641 AG |
1196 | if (i->waiting) |
1197 | wake_up(&mdev->misc_wait); | |
1198 | } | |
1199 | ||
d8763023 AG |
1200 | static int receive_Barrier(struct drbd_conf *mdev, enum drbd_packet cmd, |
1201 | unsigned int data_size) | |
b411b363 | 1202 | { |
2451fc3b | 1203 | int rv; |
e42325a5 | 1204 | struct p_barrier *p = &mdev->tconn->data.rbuf.barrier; |
b411b363 PR |
1205 | struct drbd_epoch *epoch; |
1206 | ||
b411b363 PR |
1207 | inc_unacked(mdev); |
1208 | ||
b411b363 PR |
1209 | mdev->current_epoch->barrier_nr = p->barrier; |
1210 | rv = drbd_may_finish_epoch(mdev, mdev->current_epoch, EV_GOT_BARRIER_NR); | |
1211 | ||
1212 | /* P_BARRIER_ACK may imply that the corresponding extent is dropped from | |
1213 | * the activity log, which means it would not be resynced in case the | |
1214 | * R_PRIMARY crashes now. | |
1215 | * Therefore we must send the barrier_ack after the barrier request was | |
1216 | * completed. */ | |
1217 | switch (mdev->write_ordering) { | |
b411b363 PR |
1218 | case WO_none: |
1219 | if (rv == FE_RECYCLED) | |
81e84650 | 1220 | return true; |
2451fc3b PR |
1221 | |
1222 | /* receiver context, in the writeout path of the other node. | |
1223 | * avoid potential distributed deadlock */ | |
1224 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1225 | if (epoch) | |
1226 | break; | |
1227 | else | |
1228 | dev_warn(DEV, "Allocation of an epoch failed, slowing down\n"); | |
1229 | /* Fall through */ | |
b411b363 PR |
1230 | |
1231 | case WO_bdev_flush: | |
1232 | case WO_drain_io: | |
b411b363 | 1233 | drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
2451fc3b PR |
1234 | drbd_flush(mdev); |
1235 | ||
1236 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1237 | epoch = kmalloc(sizeof(struct drbd_epoch), GFP_NOIO); | |
1238 | if (epoch) | |
1239 | break; | |
b411b363 PR |
1240 | } |
1241 | ||
2451fc3b PR |
1242 | epoch = mdev->current_epoch; |
1243 | wait_event(mdev->ee_wait, atomic_read(&epoch->epoch_size) == 0); | |
1244 | ||
1245 | D_ASSERT(atomic_read(&epoch->active) == 0); | |
1246 | D_ASSERT(epoch->flags == 0); | |
b411b363 | 1247 | |
81e84650 | 1248 | return true; |
2451fc3b PR |
1249 | default: |
1250 | dev_err(DEV, "Strangeness in mdev->write_ordering %d\n", mdev->write_ordering); | |
81e84650 | 1251 | return false; |
b411b363 PR |
1252 | } |
1253 | ||
1254 | epoch->flags = 0; | |
1255 | atomic_set(&epoch->epoch_size, 0); | |
1256 | atomic_set(&epoch->active, 0); | |
1257 | ||
1258 | spin_lock(&mdev->epoch_lock); | |
1259 | if (atomic_read(&mdev->current_epoch->epoch_size)) { | |
1260 | list_add(&epoch->list, &mdev->current_epoch->list); | |
1261 | mdev->current_epoch = epoch; | |
1262 | mdev->epochs++; | |
b411b363 PR |
1263 | } else { |
1264 | /* The current_epoch got recycled while we allocated this one... */ | |
1265 | kfree(epoch); | |
1266 | } | |
1267 | spin_unlock(&mdev->epoch_lock); | |
1268 | ||
81e84650 | 1269 | return true; |
b411b363 PR |
1270 | } |
1271 | ||
1272 | /* used from receive_RSDataReply (recv_resync_read) | |
1273 | * and from receive_Data */ | |
f6ffca9f AG |
1274 | static struct drbd_peer_request * |
1275 | read_in_block(struct drbd_conf *mdev, u64 id, sector_t sector, | |
1276 | int data_size) __must_hold(local) | |
b411b363 | 1277 | { |
6666032a | 1278 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); |
db830c46 | 1279 | struct drbd_peer_request *peer_req; |
b411b363 | 1280 | struct page *page; |
45bb912b | 1281 | int dgs, ds, rr; |
a0638456 PR |
1282 | void *dig_in = mdev->tconn->int_dig_in; |
1283 | void *dig_vv = mdev->tconn->int_dig_vv; | |
6b4388ac | 1284 | unsigned long *data; |
b411b363 | 1285 | |
a0638456 PR |
1286 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ? |
1287 | crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0; | |
b411b363 PR |
1288 | |
1289 | if (dgs) { | |
de0ff338 | 1290 | rr = drbd_recv(mdev->tconn, dig_in, dgs); |
b411b363 | 1291 | if (rr != dgs) { |
0ddc5549 LE |
1292 | if (!signal_pending(current)) |
1293 | dev_warn(DEV, | |
1294 | "short read receiving data digest: read %d expected %d\n", | |
1295 | rr, dgs); | |
b411b363 PR |
1296 | return NULL; |
1297 | } | |
1298 | } | |
1299 | ||
1300 | data_size -= dgs; | |
1301 | ||
841ce241 AG |
1302 | if (!expect(data_size != 0)) |
1303 | return NULL; | |
1304 | if (!expect(IS_ALIGNED(data_size, 512))) | |
1305 | return NULL; | |
1306 | if (!expect(data_size <= DRBD_MAX_BIO_SIZE)) | |
1307 | return NULL; | |
b411b363 | 1308 | |
6666032a LE |
1309 | /* even though we trust out peer, |
1310 | * we sometimes have to double check. */ | |
1311 | if (sector + (data_size>>9) > capacity) { | |
fdda6544 LE |
1312 | dev_err(DEV, "request from peer beyond end of local disk: " |
1313 | "capacity: %llus < sector: %llus + size: %u\n", | |
6666032a LE |
1314 | (unsigned long long)capacity, |
1315 | (unsigned long long)sector, data_size); | |
1316 | return NULL; | |
1317 | } | |
1318 | ||
b411b363 PR |
1319 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD |
1320 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
1321 | * which in turn might block on the other node at this very place. */ | |
db830c46 AG |
1322 | peer_req = drbd_alloc_ee(mdev, id, sector, data_size, GFP_NOIO); |
1323 | if (!peer_req) | |
b411b363 | 1324 | return NULL; |
45bb912b | 1325 | |
b411b363 | 1326 | ds = data_size; |
db830c46 | 1327 | page = peer_req->pages; |
45bb912b LE |
1328 | page_chain_for_each(page) { |
1329 | unsigned len = min_t(int, ds, PAGE_SIZE); | |
6b4388ac | 1330 | data = kmap(page); |
de0ff338 | 1331 | rr = drbd_recv(mdev->tconn, data, len); |
0cf9d27e | 1332 | if (drbd_insert_fault(mdev, DRBD_FAULT_RECEIVE)) { |
6b4388ac PR |
1333 | dev_err(DEV, "Fault injection: Corrupting data on receive\n"); |
1334 | data[0] = data[0] ^ (unsigned long)-1; | |
1335 | } | |
b411b363 | 1336 | kunmap(page); |
45bb912b | 1337 | if (rr != len) { |
db830c46 | 1338 | drbd_free_ee(mdev, peer_req); |
0ddc5549 LE |
1339 | if (!signal_pending(current)) |
1340 | dev_warn(DEV, "short read receiving data: read %d expected %d\n", | |
1341 | rr, len); | |
b411b363 PR |
1342 | return NULL; |
1343 | } | |
1344 | ds -= rr; | |
1345 | } | |
1346 | ||
1347 | if (dgs) { | |
db830c46 | 1348 | drbd_csum_ee(mdev, mdev->tconn->integrity_r_tfm, peer_req, dig_vv); |
b411b363 | 1349 | if (memcmp(dig_in, dig_vv, dgs)) { |
470be44a LE |
1350 | dev_err(DEV, "Digest integrity check FAILED: %llus +%u\n", |
1351 | (unsigned long long)sector, data_size); | |
b411b363 | 1352 | drbd_bcast_ee(mdev, "digest failed", |
db830c46 AG |
1353 | dgs, dig_in, dig_vv, peer_req); |
1354 | drbd_free_ee(mdev, peer_req); | |
b411b363 PR |
1355 | return NULL; |
1356 | } | |
1357 | } | |
1358 | mdev->recv_cnt += data_size>>9; | |
db830c46 | 1359 | return peer_req; |
b411b363 PR |
1360 | } |
1361 | ||
1362 | /* drbd_drain_block() just takes a data block | |
1363 | * out of the socket input buffer, and discards it. | |
1364 | */ | |
1365 | static int drbd_drain_block(struct drbd_conf *mdev, int data_size) | |
1366 | { | |
1367 | struct page *page; | |
1368 | int rr, rv = 1; | |
1369 | void *data; | |
1370 | ||
c3470cde | 1371 | if (!data_size) |
81e84650 | 1372 | return true; |
c3470cde | 1373 | |
45bb912b | 1374 | page = drbd_pp_alloc(mdev, 1, 1); |
b411b363 PR |
1375 | |
1376 | data = kmap(page); | |
1377 | while (data_size) { | |
de0ff338 | 1378 | rr = drbd_recv(mdev->tconn, data, min_t(int, data_size, PAGE_SIZE)); |
b411b363 PR |
1379 | if (rr != min_t(int, data_size, PAGE_SIZE)) { |
1380 | rv = 0; | |
0ddc5549 LE |
1381 | if (!signal_pending(current)) |
1382 | dev_warn(DEV, | |
1383 | "short read receiving data: read %d expected %d\n", | |
1384 | rr, min_t(int, data_size, PAGE_SIZE)); | |
b411b363 PR |
1385 | break; |
1386 | } | |
1387 | data_size -= rr; | |
1388 | } | |
1389 | kunmap(page); | |
435f0740 | 1390 | drbd_pp_free(mdev, page, 0); |
b411b363 PR |
1391 | return rv; |
1392 | } | |
1393 | ||
1394 | static int recv_dless_read(struct drbd_conf *mdev, struct drbd_request *req, | |
1395 | sector_t sector, int data_size) | |
1396 | { | |
1397 | struct bio_vec *bvec; | |
1398 | struct bio *bio; | |
1399 | int dgs, rr, i, expect; | |
a0638456 PR |
1400 | void *dig_in = mdev->tconn->int_dig_in; |
1401 | void *dig_vv = mdev->tconn->int_dig_vv; | |
b411b363 | 1402 | |
a0638456 PR |
1403 | dgs = (mdev->tconn->agreed_pro_version >= 87 && mdev->tconn->integrity_r_tfm) ? |
1404 | crypto_hash_digestsize(mdev->tconn->integrity_r_tfm) : 0; | |
b411b363 PR |
1405 | |
1406 | if (dgs) { | |
de0ff338 | 1407 | rr = drbd_recv(mdev->tconn, dig_in, dgs); |
b411b363 | 1408 | if (rr != dgs) { |
0ddc5549 LE |
1409 | if (!signal_pending(current)) |
1410 | dev_warn(DEV, | |
1411 | "short read receiving data reply digest: read %d expected %d\n", | |
1412 | rr, dgs); | |
b411b363 PR |
1413 | return 0; |
1414 | } | |
1415 | } | |
1416 | ||
1417 | data_size -= dgs; | |
1418 | ||
1419 | /* optimistically update recv_cnt. if receiving fails below, | |
1420 | * we disconnect anyways, and counters will be reset. */ | |
1421 | mdev->recv_cnt += data_size>>9; | |
1422 | ||
1423 | bio = req->master_bio; | |
1424 | D_ASSERT(sector == bio->bi_sector); | |
1425 | ||
1426 | bio_for_each_segment(bvec, bio, i) { | |
1427 | expect = min_t(int, data_size, bvec->bv_len); | |
de0ff338 | 1428 | rr = drbd_recv(mdev->tconn, |
b411b363 PR |
1429 | kmap(bvec->bv_page)+bvec->bv_offset, |
1430 | expect); | |
1431 | kunmap(bvec->bv_page); | |
1432 | if (rr != expect) { | |
0ddc5549 LE |
1433 | if (!signal_pending(current)) |
1434 | dev_warn(DEV, "short read receiving data reply: " | |
1435 | "read %d expected %d\n", | |
1436 | rr, expect); | |
b411b363 PR |
1437 | return 0; |
1438 | } | |
1439 | data_size -= rr; | |
1440 | } | |
1441 | ||
1442 | if (dgs) { | |
a0638456 | 1443 | drbd_csum_bio(mdev, mdev->tconn->integrity_r_tfm, bio, dig_vv); |
b411b363 PR |
1444 | if (memcmp(dig_in, dig_vv, dgs)) { |
1445 | dev_err(DEV, "Digest integrity check FAILED. Broken NICs?\n"); | |
1446 | return 0; | |
1447 | } | |
1448 | } | |
1449 | ||
1450 | D_ASSERT(data_size == 0); | |
1451 | return 1; | |
1452 | } | |
1453 | ||
1454 | /* e_end_resync_block() is called via | |
1455 | * drbd_process_done_ee() by asender only */ | |
00d56944 | 1456 | static int e_end_resync_block(struct drbd_work *w, int unused) |
b411b363 | 1457 | { |
8050e6d0 AG |
1458 | struct drbd_peer_request *peer_req = |
1459 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1460 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1461 | sector_t sector = peer_req->i.sector; |
b411b363 PR |
1462 | int ok; |
1463 | ||
db830c46 | 1464 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1465 | |
db830c46 AG |
1466 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
1467 | drbd_set_in_sync(mdev, sector, peer_req->i.size); | |
1468 | ok = drbd_send_ack(mdev, P_RS_WRITE_ACK, peer_req); | |
b411b363 PR |
1469 | } else { |
1470 | /* Record failure to sync */ | |
db830c46 | 1471 | drbd_rs_failed_io(mdev, sector, peer_req->i.size); |
b411b363 | 1472 | |
db830c46 | 1473 | ok = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1474 | } |
1475 | dec_unacked(mdev); | |
1476 | ||
1477 | return ok; | |
1478 | } | |
1479 | ||
1480 | static int recv_resync_read(struct drbd_conf *mdev, sector_t sector, int data_size) __releases(local) | |
1481 | { | |
db830c46 | 1482 | struct drbd_peer_request *peer_req; |
b411b363 | 1483 | |
db830c46 AG |
1484 | peer_req = read_in_block(mdev, ID_SYNCER, sector, data_size); |
1485 | if (!peer_req) | |
45bb912b | 1486 | goto fail; |
b411b363 PR |
1487 | |
1488 | dec_rs_pending(mdev); | |
1489 | ||
b411b363 PR |
1490 | inc_unacked(mdev); |
1491 | /* corresponding dec_unacked() in e_end_resync_block() | |
1492 | * respective _drbd_clear_done_ee */ | |
1493 | ||
db830c46 | 1494 | peer_req->w.cb = e_end_resync_block; |
45bb912b | 1495 | |
87eeee41 | 1496 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1497 | list_add(&peer_req->w.list, &mdev->sync_ee); |
87eeee41 | 1498 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1499 | |
0f0601f4 | 1500 | atomic_add(data_size >> 9, &mdev->rs_sect_ev); |
fbe29dec | 1501 | if (drbd_submit_peer_request(mdev, peer_req, WRITE, DRBD_FAULT_RS_WR) == 0) |
81e84650 | 1502 | return true; |
b411b363 | 1503 | |
10f6d992 LE |
1504 | /* don't care for the reason here */ |
1505 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 1506 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 1507 | list_del(&peer_req->w.list); |
87eeee41 | 1508 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 1509 | |
db830c46 | 1510 | drbd_free_ee(mdev, peer_req); |
45bb912b LE |
1511 | fail: |
1512 | put_ldev(mdev); | |
81e84650 | 1513 | return false; |
b411b363 PR |
1514 | } |
1515 | ||
668eebc6 | 1516 | static struct drbd_request * |
bc9c5c41 AG |
1517 | find_request(struct drbd_conf *mdev, struct rb_root *root, u64 id, |
1518 | sector_t sector, bool missing_ok, const char *func) | |
51624585 | 1519 | { |
51624585 AG |
1520 | struct drbd_request *req; |
1521 | ||
bc9c5c41 AG |
1522 | /* Request object according to our peer */ |
1523 | req = (struct drbd_request *)(unsigned long)id; | |
5e472264 | 1524 | if (drbd_contains_interval(root, sector, &req->i) && req->i.local) |
668eebc6 | 1525 | return req; |
c3afd8f5 AG |
1526 | if (!missing_ok) { |
1527 | dev_err(DEV, "%s: failed to find request %lu, sector %llus\n", func, | |
1528 | (unsigned long)id, (unsigned long long)sector); | |
1529 | } | |
51624585 AG |
1530 | return NULL; |
1531 | } | |
1532 | ||
d8763023 AG |
1533 | static int receive_DataReply(struct drbd_conf *mdev, enum drbd_packet cmd, |
1534 | unsigned int data_size) | |
b411b363 PR |
1535 | { |
1536 | struct drbd_request *req; | |
1537 | sector_t sector; | |
b411b363 | 1538 | int ok; |
e42325a5 | 1539 | struct p_data *p = &mdev->tconn->data.rbuf.data; |
b411b363 PR |
1540 | |
1541 | sector = be64_to_cpu(p->sector); | |
1542 | ||
87eeee41 | 1543 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 1544 | req = find_request(mdev, &mdev->read_requests, p->block_id, sector, false, __func__); |
87eeee41 | 1545 | spin_unlock_irq(&mdev->tconn->req_lock); |
c3afd8f5 | 1546 | if (unlikely(!req)) |
81e84650 | 1547 | return false; |
b411b363 | 1548 | |
24c4830c | 1549 | /* hlist_del(&req->collision) is done in _req_may_be_done, to avoid |
b411b363 PR |
1550 | * special casing it there for the various failure cases. |
1551 | * still no race with drbd_fail_pending_reads */ | |
1552 | ok = recv_dless_read(mdev, req, sector, data_size); | |
1553 | ||
1554 | if (ok) | |
8554df1c | 1555 | req_mod(req, DATA_RECEIVED); |
b411b363 PR |
1556 | /* else: nothing. handled from drbd_disconnect... |
1557 | * I don't think we may complete this just yet | |
1558 | * in case we are "on-disconnect: freeze" */ | |
1559 | ||
1560 | return ok; | |
1561 | } | |
1562 | ||
d8763023 AG |
1563 | static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packet cmd, |
1564 | unsigned int data_size) | |
b411b363 PR |
1565 | { |
1566 | sector_t sector; | |
b411b363 | 1567 | int ok; |
e42325a5 | 1568 | struct p_data *p = &mdev->tconn->data.rbuf.data; |
b411b363 PR |
1569 | |
1570 | sector = be64_to_cpu(p->sector); | |
1571 | D_ASSERT(p->block_id == ID_SYNCER); | |
1572 | ||
1573 | if (get_ldev(mdev)) { | |
1574 | /* data is submitted to disk within recv_resync_read. | |
1575 | * corresponding put_ldev done below on error, | |
fcefa62e | 1576 | * or in drbd_peer_request_endio. */ |
b411b363 PR |
1577 | ok = recv_resync_read(mdev, sector, data_size); |
1578 | } else { | |
1579 | if (__ratelimit(&drbd_ratelimit_state)) | |
1580 | dev_err(DEV, "Can not write resync data to local disk.\n"); | |
1581 | ||
1582 | ok = drbd_drain_block(mdev, data_size); | |
1583 | ||
2b2bf214 | 1584 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 PR |
1585 | } |
1586 | ||
778f271d PR |
1587 | atomic_add(data_size >> 9, &mdev->rs_sect_in); |
1588 | ||
b411b363 PR |
1589 | return ok; |
1590 | } | |
1591 | ||
7be8da07 AG |
1592 | static int w_restart_write(struct drbd_work *w, int cancel) |
1593 | { | |
1594 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
1595 | struct drbd_conf *mdev = w->mdev; | |
1596 | struct bio *bio; | |
1597 | unsigned long start_time; | |
1598 | unsigned long flags; | |
1599 | ||
1600 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); | |
1601 | if (!expect(req->rq_state & RQ_POSTPONED)) { | |
1602 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
1603 | return 0; | |
1604 | } | |
1605 | bio = req->master_bio; | |
1606 | start_time = req->start_time; | |
1607 | /* Postponed requests will not have their master_bio completed! */ | |
1608 | __req_mod(req, DISCARD_WRITE, NULL); | |
1609 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); | |
1610 | ||
1611 | while (__drbd_make_request(mdev, bio, start_time)) | |
1612 | /* retry */ ; | |
1613 | return 1; | |
1614 | } | |
1615 | ||
1616 | static void restart_conflicting_writes(struct drbd_conf *mdev, | |
1617 | sector_t sector, int size) | |
1618 | { | |
1619 | struct drbd_interval *i; | |
1620 | struct drbd_request *req; | |
1621 | ||
1622 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1623 | if (!i->local) | |
1624 | continue; | |
1625 | req = container_of(i, struct drbd_request, i); | |
1626 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1627 | !(req->rq_state & RQ_POSTPONED)) | |
1628 | continue; | |
1629 | if (expect(list_empty(&req->w.list))) { | |
1630 | req->w.mdev = mdev; | |
1631 | req->w.cb = w_restart_write; | |
1632 | drbd_queue_work(&mdev->tconn->data.work, &req->w); | |
1633 | } | |
1634 | } | |
1635 | } | |
1636 | ||
b411b363 PR |
1637 | /* e_end_block() is called via drbd_process_done_ee(). |
1638 | * this means this function only runs in the asender thread | |
1639 | */ | |
00d56944 | 1640 | static int e_end_block(struct drbd_work *w, int cancel) |
b411b363 | 1641 | { |
8050e6d0 AG |
1642 | struct drbd_peer_request *peer_req = |
1643 | container_of(w, struct drbd_peer_request, w); | |
00d56944 | 1644 | struct drbd_conf *mdev = w->mdev; |
db830c46 | 1645 | sector_t sector = peer_req->i.sector; |
b411b363 PR |
1646 | int ok = 1, pcmd; |
1647 | ||
89e58e75 | 1648 | if (mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C) { |
db830c46 | 1649 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1650 | pcmd = (mdev->state.conn >= C_SYNC_SOURCE && |
1651 | mdev->state.conn <= C_PAUSED_SYNC_T && | |
db830c46 | 1652 | peer_req->flags & EE_MAY_SET_IN_SYNC) ? |
b411b363 | 1653 | P_RS_WRITE_ACK : P_WRITE_ACK; |
db830c46 | 1654 | ok &= drbd_send_ack(mdev, pcmd, peer_req); |
b411b363 | 1655 | if (pcmd == P_RS_WRITE_ACK) |
db830c46 | 1656 | drbd_set_in_sync(mdev, sector, peer_req->i.size); |
b411b363 | 1657 | } else { |
db830c46 | 1658 | ok = drbd_send_ack(mdev, P_NEG_ACK, peer_req); |
b411b363 PR |
1659 | /* we expect it to be marked out of sync anyways... |
1660 | * maybe assert this? */ | |
1661 | } | |
1662 | dec_unacked(mdev); | |
1663 | } | |
1664 | /* we delete from the conflict detection hash _after_ we sent out the | |
1665 | * P_WRITE_ACK / P_NEG_ACK, to get the sequence number right. */ | |
89e58e75 | 1666 | if (mdev->tconn->net_conf->two_primaries) { |
87eeee41 | 1667 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
1668 | D_ASSERT(!drbd_interval_empty(&peer_req->i)); |
1669 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
7be8da07 AG |
1670 | if (peer_req->flags & EE_RESTART_REQUESTS) |
1671 | restart_conflicting_writes(mdev, sector, peer_req->i.size); | |
87eeee41 | 1672 | spin_unlock_irq(&mdev->tconn->req_lock); |
bb3bfe96 | 1673 | } else |
db830c46 | 1674 | D_ASSERT(drbd_interval_empty(&peer_req->i)); |
b411b363 | 1675 | |
db830c46 | 1676 | drbd_may_finish_epoch(mdev, peer_req->epoch, EV_PUT + (cancel ? EV_CLEANUP : 0)); |
b411b363 PR |
1677 | |
1678 | return ok; | |
1679 | } | |
1680 | ||
7be8da07 | 1681 | static int e_send_ack(struct drbd_work *w, enum drbd_packet ack) |
b411b363 | 1682 | { |
7be8da07 | 1683 | struct drbd_conf *mdev = w->mdev; |
8050e6d0 AG |
1684 | struct drbd_peer_request *peer_req = |
1685 | container_of(w, struct drbd_peer_request, w); | |
206d3589 | 1686 | int ok; |
b411b363 | 1687 | |
7be8da07 | 1688 | ok = drbd_send_ack(mdev, ack, peer_req); |
b411b363 PR |
1689 | dec_unacked(mdev); |
1690 | ||
1691 | return ok; | |
1692 | } | |
1693 | ||
7be8da07 AG |
1694 | static int e_send_discard_write(struct drbd_work *w, int unused) |
1695 | { | |
1696 | return e_send_ack(w, P_DISCARD_WRITE); | |
1697 | } | |
1698 | ||
1699 | static int e_send_retry_write(struct drbd_work *w, int unused) | |
1700 | { | |
1701 | struct drbd_tconn *tconn = w->mdev->tconn; | |
1702 | ||
1703 | return e_send_ack(w, tconn->agreed_pro_version >= 100 ? | |
1704 | P_RETRY_WRITE : P_DISCARD_WRITE); | |
1705 | } | |
1706 | ||
3e394da1 AG |
1707 | static bool seq_greater(u32 a, u32 b) |
1708 | { | |
1709 | /* | |
1710 | * We assume 32-bit wrap-around here. | |
1711 | * For 24-bit wrap-around, we would have to shift: | |
1712 | * a <<= 8; b <<= 8; | |
1713 | */ | |
1714 | return (s32)a - (s32)b > 0; | |
1715 | } | |
1716 | ||
1717 | static u32 seq_max(u32 a, u32 b) | |
1718 | { | |
1719 | return seq_greater(a, b) ? a : b; | |
1720 | } | |
1721 | ||
7be8da07 AG |
1722 | static bool need_peer_seq(struct drbd_conf *mdev) |
1723 | { | |
1724 | struct drbd_tconn *tconn = mdev->tconn; | |
1725 | ||
1726 | /* | |
1727 | * We only need to keep track of the last packet_seq number of our peer | |
1728 | * if we are in dual-primary mode and we have the discard flag set; see | |
1729 | * handle_write_conflicts(). | |
1730 | */ | |
1731 | return tconn->net_conf->two_primaries && | |
1732 | test_bit(DISCARD_CONCURRENT, &tconn->flags); | |
1733 | } | |
1734 | ||
43ae077d | 1735 | static void update_peer_seq(struct drbd_conf *mdev, unsigned int peer_seq) |
3e394da1 | 1736 | { |
43ae077d | 1737 | unsigned int old_peer_seq; |
3e394da1 | 1738 | |
7be8da07 AG |
1739 | if (need_peer_seq(mdev)) { |
1740 | spin_lock(&mdev->peer_seq_lock); | |
1741 | old_peer_seq = mdev->peer_seq; | |
1742 | mdev->peer_seq = seq_max(mdev->peer_seq, peer_seq); | |
1743 | spin_unlock(&mdev->peer_seq_lock); | |
1744 | if (old_peer_seq != peer_seq) | |
1745 | wake_up(&mdev->seq_wait); | |
1746 | } | |
3e394da1 AG |
1747 | } |
1748 | ||
b411b363 PR |
1749 | /* Called from receive_Data. |
1750 | * Synchronize packets on sock with packets on msock. | |
1751 | * | |
1752 | * This is here so even when a P_DATA packet traveling via sock overtook an Ack | |
1753 | * packet traveling on msock, they are still processed in the order they have | |
1754 | * been sent. | |
1755 | * | |
1756 | * Note: we don't care for Ack packets overtaking P_DATA packets. | |
1757 | * | |
1758 | * In case packet_seq is larger than mdev->peer_seq number, there are | |
1759 | * outstanding packets on the msock. We wait for them to arrive. | |
1760 | * In case we are the logically next packet, we update mdev->peer_seq | |
1761 | * ourselves. Correctly handles 32bit wrap around. | |
1762 | * | |
1763 | * Assume we have a 10 GBit connection, that is about 1<<30 byte per second, | |
1764 | * about 1<<21 sectors per second. So "worst" case, we have 1<<3 == 8 seconds | |
1765 | * for the 24bit wrap (historical atomic_t guarantee on some archs), and we have | |
1766 | * 1<<9 == 512 seconds aka ages for the 32bit wrap around... | |
1767 | * | |
1768 | * returns 0 if we may process the packet, | |
1769 | * -ERESTARTSYS if we were interrupted (by disconnect signal). */ | |
7be8da07 | 1770 | static int wait_for_and_update_peer_seq(struct drbd_conf *mdev, const u32 peer_seq) |
b411b363 PR |
1771 | { |
1772 | DEFINE_WAIT(wait); | |
b411b363 | 1773 | long timeout; |
7be8da07 AG |
1774 | int ret; |
1775 | ||
1776 | if (!need_peer_seq(mdev)) | |
1777 | return 0; | |
1778 | ||
b411b363 PR |
1779 | spin_lock(&mdev->peer_seq_lock); |
1780 | for (;;) { | |
7be8da07 AG |
1781 | if (!seq_greater(peer_seq - 1, mdev->peer_seq)) { |
1782 | mdev->peer_seq = seq_max(mdev->peer_seq, peer_seq); | |
1783 | ret = 0; | |
b411b363 | 1784 | break; |
7be8da07 | 1785 | } |
b411b363 PR |
1786 | if (signal_pending(current)) { |
1787 | ret = -ERESTARTSYS; | |
1788 | break; | |
1789 | } | |
7be8da07 | 1790 | prepare_to_wait(&mdev->seq_wait, &wait, TASK_INTERRUPTIBLE); |
b411b363 | 1791 | spin_unlock(&mdev->peer_seq_lock); |
71b1c1eb AG |
1792 | timeout = mdev->tconn->net_conf->ping_timeo*HZ/10; |
1793 | timeout = schedule_timeout(timeout); | |
b411b363 | 1794 | spin_lock(&mdev->peer_seq_lock); |
7be8da07 | 1795 | if (!timeout) { |
b411b363 | 1796 | ret = -ETIMEDOUT; |
71b1c1eb | 1797 | dev_err(DEV, "Timed out waiting for missing ack packets; disconnecting\n"); |
b411b363 PR |
1798 | break; |
1799 | } | |
1800 | } | |
b411b363 | 1801 | spin_unlock(&mdev->peer_seq_lock); |
7be8da07 | 1802 | finish_wait(&mdev->seq_wait, &wait); |
b411b363 PR |
1803 | return ret; |
1804 | } | |
1805 | ||
688593c5 LE |
1806 | /* see also bio_flags_to_wire() |
1807 | * DRBD_REQ_*, because we need to semantically map the flags to data packet | |
1808 | * flags and back. We may replicate to other kernel versions. */ | |
1809 | static unsigned long wire_flags_to_bio(struct drbd_conf *mdev, u32 dpf) | |
76d2e7ec | 1810 | { |
688593c5 LE |
1811 | return (dpf & DP_RW_SYNC ? REQ_SYNC : 0) | |
1812 | (dpf & DP_FUA ? REQ_FUA : 0) | | |
1813 | (dpf & DP_FLUSH ? REQ_FLUSH : 0) | | |
1814 | (dpf & DP_DISCARD ? REQ_DISCARD : 0); | |
76d2e7ec PR |
1815 | } |
1816 | ||
7be8da07 AG |
1817 | static void fail_postponed_requests(struct drbd_conf *mdev, sector_t sector, |
1818 | unsigned int size) | |
1819 | { | |
1820 | struct drbd_interval *i; | |
1821 | ||
1822 | repeat: | |
1823 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1824 | struct drbd_request *req; | |
1825 | struct bio_and_error m; | |
1826 | ||
1827 | if (!i->local) | |
1828 | continue; | |
1829 | req = container_of(i, struct drbd_request, i); | |
1830 | if (!(req->rq_state & RQ_POSTPONED)) | |
1831 | continue; | |
1832 | req->rq_state &= ~RQ_POSTPONED; | |
1833 | __req_mod(req, NEG_ACKED, &m); | |
1834 | spin_unlock_irq(&mdev->tconn->req_lock); | |
1835 | if (m.bio) | |
1836 | complete_master_bio(mdev, &m); | |
1837 | spin_lock_irq(&mdev->tconn->req_lock); | |
1838 | goto repeat; | |
1839 | } | |
1840 | } | |
1841 | ||
1842 | static int handle_write_conflicts(struct drbd_conf *mdev, | |
1843 | struct drbd_peer_request *peer_req) | |
1844 | { | |
1845 | struct drbd_tconn *tconn = mdev->tconn; | |
1846 | bool resolve_conflicts = test_bit(DISCARD_CONCURRENT, &tconn->flags); | |
1847 | sector_t sector = peer_req->i.sector; | |
1848 | const unsigned int size = peer_req->i.size; | |
1849 | struct drbd_interval *i; | |
1850 | bool equal; | |
1851 | int err; | |
1852 | ||
1853 | /* | |
1854 | * Inserting the peer request into the write_requests tree will prevent | |
1855 | * new conflicting local requests from being added. | |
1856 | */ | |
1857 | drbd_insert_interval(&mdev->write_requests, &peer_req->i); | |
1858 | ||
1859 | repeat: | |
1860 | drbd_for_each_overlap(i, &mdev->write_requests, sector, size) { | |
1861 | if (i == &peer_req->i) | |
1862 | continue; | |
1863 | ||
1864 | if (!i->local) { | |
1865 | /* | |
1866 | * Our peer has sent a conflicting remote request; this | |
1867 | * should not happen in a two-node setup. Wait for the | |
1868 | * earlier peer request to complete. | |
1869 | */ | |
1870 | err = drbd_wait_misc(mdev, i); | |
1871 | if (err) | |
1872 | goto out; | |
1873 | goto repeat; | |
1874 | } | |
1875 | ||
1876 | equal = i->sector == sector && i->size == size; | |
1877 | if (resolve_conflicts) { | |
1878 | /* | |
1879 | * If the peer request is fully contained within the | |
1880 | * overlapping request, it can be discarded; otherwise, | |
1881 | * it will be retried once all overlapping requests | |
1882 | * have completed. | |
1883 | */ | |
1884 | bool discard = i->sector <= sector && i->sector + | |
1885 | (i->size >> 9) >= sector + (size >> 9); | |
1886 | ||
1887 | if (!equal) | |
1888 | dev_alert(DEV, "Concurrent writes detected: " | |
1889 | "local=%llus +%u, remote=%llus +%u, " | |
1890 | "assuming %s came first\n", | |
1891 | (unsigned long long)i->sector, i->size, | |
1892 | (unsigned long long)sector, size, | |
1893 | discard ? "local" : "remote"); | |
1894 | ||
1895 | inc_unacked(mdev); | |
1896 | peer_req->w.cb = discard ? e_send_discard_write : | |
1897 | e_send_retry_write; | |
1898 | list_add_tail(&peer_req->w.list, &mdev->done_ee); | |
1899 | wake_asender(mdev->tconn); | |
1900 | ||
1901 | err = -ENOENT; | |
1902 | goto out; | |
1903 | } else { | |
1904 | struct drbd_request *req = | |
1905 | container_of(i, struct drbd_request, i); | |
1906 | ||
1907 | if (!equal) | |
1908 | dev_alert(DEV, "Concurrent writes detected: " | |
1909 | "local=%llus +%u, remote=%llus +%u\n", | |
1910 | (unsigned long long)i->sector, i->size, | |
1911 | (unsigned long long)sector, size); | |
1912 | ||
1913 | if (req->rq_state & RQ_LOCAL_PENDING || | |
1914 | !(req->rq_state & RQ_POSTPONED)) { | |
1915 | /* | |
1916 | * Wait for the node with the discard flag to | |
1917 | * decide if this request will be discarded or | |
1918 | * retried. Requests that are discarded will | |
1919 | * disappear from the write_requests tree. | |
1920 | * | |
1921 | * In addition, wait for the conflicting | |
1922 | * request to finish locally before submitting | |
1923 | * the conflicting peer request. | |
1924 | */ | |
1925 | err = drbd_wait_misc(mdev, &req->i); | |
1926 | if (err) { | |
1927 | _conn_request_state(mdev->tconn, | |
1928 | NS(conn, C_TIMEOUT), | |
1929 | CS_HARD); | |
1930 | fail_postponed_requests(mdev, sector, size); | |
1931 | goto out; | |
1932 | } | |
1933 | goto repeat; | |
1934 | } | |
1935 | /* | |
1936 | * Remember to restart the conflicting requests after | |
1937 | * the new peer request has completed. | |
1938 | */ | |
1939 | peer_req->flags |= EE_RESTART_REQUESTS; | |
1940 | } | |
1941 | } | |
1942 | err = 0; | |
1943 | ||
1944 | out: | |
1945 | if (err) | |
1946 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
1947 | return err; | |
1948 | } | |
1949 | ||
b411b363 | 1950 | /* mirrored write */ |
d8763023 AG |
1951 | static int receive_Data(struct drbd_conf *mdev, enum drbd_packet cmd, |
1952 | unsigned int data_size) | |
b411b363 PR |
1953 | { |
1954 | sector_t sector; | |
db830c46 | 1955 | struct drbd_peer_request *peer_req; |
e42325a5 | 1956 | struct p_data *p = &mdev->tconn->data.rbuf.data; |
7be8da07 | 1957 | u32 peer_seq = be32_to_cpu(p->seq_num); |
b411b363 PR |
1958 | int rw = WRITE; |
1959 | u32 dp_flags; | |
7be8da07 | 1960 | int err; |
b411b363 | 1961 | |
b411b363 | 1962 | |
7be8da07 AG |
1963 | if (!get_ldev(mdev)) { |
1964 | err = wait_for_and_update_peer_seq(mdev, peer_seq); | |
2b2bf214 | 1965 | drbd_send_ack_dp(mdev, P_NEG_ACK, p, data_size); |
b411b363 | 1966 | atomic_inc(&mdev->current_epoch->epoch_size); |
7be8da07 | 1967 | return drbd_drain_block(mdev, data_size) && err == 0; |
b411b363 PR |
1968 | } |
1969 | ||
fcefa62e AG |
1970 | /* |
1971 | * Corresponding put_ldev done either below (on various errors), or in | |
1972 | * drbd_peer_request_endio, if we successfully submit the data at the | |
1973 | * end of this function. | |
1974 | */ | |
b411b363 PR |
1975 | |
1976 | sector = be64_to_cpu(p->sector); | |
db830c46 AG |
1977 | peer_req = read_in_block(mdev, p->block_id, sector, data_size); |
1978 | if (!peer_req) { | |
b411b363 | 1979 | put_ldev(mdev); |
81e84650 | 1980 | return false; |
b411b363 PR |
1981 | } |
1982 | ||
db830c46 | 1983 | peer_req->w.cb = e_end_block; |
b411b363 | 1984 | |
688593c5 LE |
1985 | dp_flags = be32_to_cpu(p->dp_flags); |
1986 | rw |= wire_flags_to_bio(mdev, dp_flags); | |
1987 | ||
1988 | if (dp_flags & DP_MAY_SET_IN_SYNC) | |
db830c46 | 1989 | peer_req->flags |= EE_MAY_SET_IN_SYNC; |
688593c5 | 1990 | |
b411b363 | 1991 | spin_lock(&mdev->epoch_lock); |
db830c46 AG |
1992 | peer_req->epoch = mdev->current_epoch; |
1993 | atomic_inc(&peer_req->epoch->epoch_size); | |
1994 | atomic_inc(&peer_req->epoch->active); | |
b411b363 PR |
1995 | spin_unlock(&mdev->epoch_lock); |
1996 | ||
7be8da07 AG |
1997 | if (mdev->tconn->net_conf->two_primaries) { |
1998 | err = wait_for_and_update_peer_seq(mdev, peer_seq); | |
1999 | if (err) | |
b411b363 | 2000 | goto out_interrupted; |
87eeee41 | 2001 | spin_lock_irq(&mdev->tconn->req_lock); |
7be8da07 AG |
2002 | err = handle_write_conflicts(mdev, peer_req); |
2003 | if (err) { | |
2004 | spin_unlock_irq(&mdev->tconn->req_lock); | |
2005 | if (err == -ENOENT) { | |
b411b363 | 2006 | put_ldev(mdev); |
81e84650 | 2007 | return true; |
b411b363 | 2008 | } |
7be8da07 | 2009 | goto out_interrupted; |
b411b363 | 2010 | } |
7be8da07 AG |
2011 | } else |
2012 | spin_lock_irq(&mdev->tconn->req_lock); | |
db830c46 | 2013 | list_add(&peer_req->w.list, &mdev->active_ee); |
87eeee41 | 2014 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2015 | |
89e58e75 | 2016 | switch (mdev->tconn->net_conf->wire_protocol) { |
b411b363 PR |
2017 | case DRBD_PROT_C: |
2018 | inc_unacked(mdev); | |
2019 | /* corresponding dec_unacked() in e_end_block() | |
2020 | * respective _drbd_clear_done_ee */ | |
2021 | break; | |
2022 | case DRBD_PROT_B: | |
2023 | /* I really don't like it that the receiver thread | |
2024 | * sends on the msock, but anyways */ | |
db830c46 | 2025 | drbd_send_ack(mdev, P_RECV_ACK, peer_req); |
b411b363 PR |
2026 | break; |
2027 | case DRBD_PROT_A: | |
2028 | /* nothing to do */ | |
2029 | break; | |
2030 | } | |
2031 | ||
6719fb03 | 2032 | if (mdev->state.pdsk < D_INCONSISTENT) { |
b411b363 | 2033 | /* In case we have the only disk of the cluster, */ |
db830c46 AG |
2034 | drbd_set_out_of_sync(mdev, peer_req->i.sector, peer_req->i.size); |
2035 | peer_req->flags |= EE_CALL_AL_COMPLETE_IO; | |
2036 | peer_req->flags &= ~EE_MAY_SET_IN_SYNC; | |
2037 | drbd_al_begin_io(mdev, peer_req->i.sector); | |
b411b363 PR |
2038 | } |
2039 | ||
fbe29dec | 2040 | if (drbd_submit_peer_request(mdev, peer_req, rw, DRBD_FAULT_DT_WR) == 0) |
81e84650 | 2041 | return true; |
b411b363 | 2042 | |
10f6d992 LE |
2043 | /* don't care for the reason here */ |
2044 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2045 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
2046 | list_del(&peer_req->w.list); |
2047 | drbd_remove_epoch_entry_interval(mdev, peer_req); | |
87eeee41 | 2048 | spin_unlock_irq(&mdev->tconn->req_lock); |
db830c46 AG |
2049 | if (peer_req->flags & EE_CALL_AL_COMPLETE_IO) |
2050 | drbd_al_complete_io(mdev, peer_req->i.sector); | |
22cc37a9 | 2051 | |
b411b363 | 2052 | out_interrupted: |
db830c46 | 2053 | drbd_may_finish_epoch(mdev, peer_req->epoch, EV_PUT + EV_CLEANUP); |
b411b363 | 2054 | put_ldev(mdev); |
db830c46 | 2055 | drbd_free_ee(mdev, peer_req); |
81e84650 | 2056 | return false; |
b411b363 PR |
2057 | } |
2058 | ||
0f0601f4 LE |
2059 | /* We may throttle resync, if the lower device seems to be busy, |
2060 | * and current sync rate is above c_min_rate. | |
2061 | * | |
2062 | * To decide whether or not the lower device is busy, we use a scheme similar | |
2063 | * to MD RAID is_mddev_idle(): if the partition stats reveal "significant" | |
2064 | * (more than 64 sectors) of activity we cannot account for with our own resync | |
2065 | * activity, it obviously is "busy". | |
2066 | * | |
2067 | * The current sync rate used here uses only the most recent two step marks, | |
2068 | * to have a short time average so we can react faster. | |
2069 | */ | |
e3555d85 | 2070 | int drbd_rs_should_slow_down(struct drbd_conf *mdev, sector_t sector) |
0f0601f4 LE |
2071 | { |
2072 | struct gendisk *disk = mdev->ldev->backing_bdev->bd_contains->bd_disk; | |
2073 | unsigned long db, dt, dbdt; | |
e3555d85 | 2074 | struct lc_element *tmp; |
0f0601f4 LE |
2075 | int curr_events; |
2076 | int throttle = 0; | |
2077 | ||
2078 | /* feature disabled? */ | |
2079 | if (mdev->sync_conf.c_min_rate == 0) | |
2080 | return 0; | |
2081 | ||
e3555d85 PR |
2082 | spin_lock_irq(&mdev->al_lock); |
2083 | tmp = lc_find(mdev->resync, BM_SECT_TO_EXT(sector)); | |
2084 | if (tmp) { | |
2085 | struct bm_extent *bm_ext = lc_entry(tmp, struct bm_extent, lce); | |
2086 | if (test_bit(BME_PRIORITY, &bm_ext->flags)) { | |
2087 | spin_unlock_irq(&mdev->al_lock); | |
2088 | return 0; | |
2089 | } | |
2090 | /* Do not slow down if app IO is already waiting for this extent */ | |
2091 | } | |
2092 | spin_unlock_irq(&mdev->al_lock); | |
2093 | ||
0f0601f4 LE |
2094 | curr_events = (int)part_stat_read(&disk->part0, sectors[0]) + |
2095 | (int)part_stat_read(&disk->part0, sectors[1]) - | |
2096 | atomic_read(&mdev->rs_sect_ev); | |
e3555d85 | 2097 | |
0f0601f4 LE |
2098 | if (!mdev->rs_last_events || curr_events - mdev->rs_last_events > 64) { |
2099 | unsigned long rs_left; | |
2100 | int i; | |
2101 | ||
2102 | mdev->rs_last_events = curr_events; | |
2103 | ||
2104 | /* sync speed average over the last 2*DRBD_SYNC_MARK_STEP, | |
2105 | * approx. */ | |
2649f080 LE |
2106 | i = (mdev->rs_last_mark + DRBD_SYNC_MARKS-1) % DRBD_SYNC_MARKS; |
2107 | ||
2108 | if (mdev->state.conn == C_VERIFY_S || mdev->state.conn == C_VERIFY_T) | |
2109 | rs_left = mdev->ov_left; | |
2110 | else | |
2111 | rs_left = drbd_bm_total_weight(mdev) - mdev->rs_failed; | |
0f0601f4 LE |
2112 | |
2113 | dt = ((long)jiffies - (long)mdev->rs_mark_time[i]) / HZ; | |
2114 | if (!dt) | |
2115 | dt++; | |
2116 | db = mdev->rs_mark_left[i] - rs_left; | |
2117 | dbdt = Bit2KB(db/dt); | |
2118 | ||
2119 | if (dbdt > mdev->sync_conf.c_min_rate) | |
2120 | throttle = 1; | |
2121 | } | |
2122 | return throttle; | |
2123 | } | |
2124 | ||
2125 | ||
d8763023 AG |
2126 | static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packet cmd, |
2127 | unsigned int digest_size) | |
b411b363 PR |
2128 | { |
2129 | sector_t sector; | |
2130 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
db830c46 | 2131 | struct drbd_peer_request *peer_req; |
b411b363 | 2132 | struct digest_info *di = NULL; |
b18b37be | 2133 | int size, verb; |
b411b363 | 2134 | unsigned int fault_type; |
e42325a5 | 2135 | struct p_block_req *p = &mdev->tconn->data.rbuf.block_req; |
b411b363 PR |
2136 | |
2137 | sector = be64_to_cpu(p->sector); | |
2138 | size = be32_to_cpu(p->blksize); | |
2139 | ||
c670a398 | 2140 | if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) { |
b411b363 PR |
2141 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, |
2142 | (unsigned long long)sector, size); | |
81e84650 | 2143 | return false; |
b411b363 PR |
2144 | } |
2145 | if (sector + (size>>9) > capacity) { | |
2146 | dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__, | |
2147 | (unsigned long long)sector, size); | |
81e84650 | 2148 | return false; |
b411b363 PR |
2149 | } |
2150 | ||
2151 | if (!get_ldev_if_state(mdev, D_UP_TO_DATE)) { | |
b18b37be PR |
2152 | verb = 1; |
2153 | switch (cmd) { | |
2154 | case P_DATA_REQUEST: | |
2155 | drbd_send_ack_rp(mdev, P_NEG_DREPLY, p); | |
2156 | break; | |
2157 | case P_RS_DATA_REQUEST: | |
2158 | case P_CSUM_RS_REQUEST: | |
2159 | case P_OV_REQUEST: | |
2160 | drbd_send_ack_rp(mdev, P_NEG_RS_DREPLY , p); | |
2161 | break; | |
2162 | case P_OV_REPLY: | |
2163 | verb = 0; | |
2164 | dec_rs_pending(mdev); | |
2165 | drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, ID_IN_SYNC); | |
2166 | break; | |
2167 | default: | |
2168 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
2169 | cmdname(cmd)); | |
2170 | } | |
2171 | if (verb && __ratelimit(&drbd_ratelimit_state)) | |
b411b363 PR |
2172 | dev_err(DEV, "Can not satisfy peer's read request, " |
2173 | "no local data.\n"); | |
b18b37be | 2174 | |
a821cc4a LE |
2175 | /* drain possibly payload */ |
2176 | return drbd_drain_block(mdev, digest_size); | |
b411b363 PR |
2177 | } |
2178 | ||
2179 | /* GFP_NOIO, because we must not cause arbitrary write-out: in a DRBD | |
2180 | * "criss-cross" setup, that might cause write-out on some other DRBD, | |
2181 | * which in turn might block on the other node at this very place. */ | |
db830c46 AG |
2182 | peer_req = drbd_alloc_ee(mdev, p->block_id, sector, size, GFP_NOIO); |
2183 | if (!peer_req) { | |
b411b363 | 2184 | put_ldev(mdev); |
81e84650 | 2185 | return false; |
b411b363 PR |
2186 | } |
2187 | ||
02918be2 | 2188 | switch (cmd) { |
b411b363 | 2189 | case P_DATA_REQUEST: |
db830c46 | 2190 | peer_req->w.cb = w_e_end_data_req; |
b411b363 | 2191 | fault_type = DRBD_FAULT_DT_RD; |
80a40e43 LE |
2192 | /* application IO, don't drbd_rs_begin_io */ |
2193 | goto submit; | |
2194 | ||
b411b363 | 2195 | case P_RS_DATA_REQUEST: |
db830c46 | 2196 | peer_req->w.cb = w_e_end_rsdata_req; |
b411b363 | 2197 | fault_type = DRBD_FAULT_RS_RD; |
5f9915bb LE |
2198 | /* used in the sector offset progress display */ |
2199 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
b411b363 PR |
2200 | break; |
2201 | ||
2202 | case P_OV_REPLY: | |
2203 | case P_CSUM_RS_REQUEST: | |
2204 | fault_type = DRBD_FAULT_RS_RD; | |
b411b363 PR |
2205 | di = kmalloc(sizeof(*di) + digest_size, GFP_NOIO); |
2206 | if (!di) | |
2207 | goto out_free_e; | |
2208 | ||
2209 | di->digest_size = digest_size; | |
2210 | di->digest = (((char *)di)+sizeof(struct digest_info)); | |
2211 | ||
db830c46 AG |
2212 | peer_req->digest = di; |
2213 | peer_req->flags |= EE_HAS_DIGEST; | |
c36c3ced | 2214 | |
de0ff338 | 2215 | if (drbd_recv(mdev->tconn, di->digest, digest_size) != digest_size) |
b411b363 PR |
2216 | goto out_free_e; |
2217 | ||
02918be2 | 2218 | if (cmd == P_CSUM_RS_REQUEST) { |
31890f4a | 2219 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
db830c46 | 2220 | peer_req->w.cb = w_e_end_csum_rs_req; |
5f9915bb LE |
2221 | /* used in the sector offset progress display */ |
2222 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
02918be2 | 2223 | } else if (cmd == P_OV_REPLY) { |
2649f080 LE |
2224 | /* track progress, we may need to throttle */ |
2225 | atomic_add(size >> 9, &mdev->rs_sect_in); | |
db830c46 | 2226 | peer_req->w.cb = w_e_end_ov_reply; |
b411b363 | 2227 | dec_rs_pending(mdev); |
0f0601f4 LE |
2228 | /* drbd_rs_begin_io done when we sent this request, |
2229 | * but accounting still needs to be done. */ | |
2230 | goto submit_for_resync; | |
b411b363 PR |
2231 | } |
2232 | break; | |
2233 | ||
2234 | case P_OV_REQUEST: | |
b411b363 | 2235 | if (mdev->ov_start_sector == ~(sector_t)0 && |
31890f4a | 2236 | mdev->tconn->agreed_pro_version >= 90) { |
de228bba LE |
2237 | unsigned long now = jiffies; |
2238 | int i; | |
b411b363 PR |
2239 | mdev->ov_start_sector = sector; |
2240 | mdev->ov_position = sector; | |
30b743a2 LE |
2241 | mdev->ov_left = drbd_bm_bits(mdev) - BM_SECT_TO_BIT(sector); |
2242 | mdev->rs_total = mdev->ov_left; | |
de228bba LE |
2243 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { |
2244 | mdev->rs_mark_left[i] = mdev->ov_left; | |
2245 | mdev->rs_mark_time[i] = now; | |
2246 | } | |
b411b363 PR |
2247 | dev_info(DEV, "Online Verify start sector: %llu\n", |
2248 | (unsigned long long)sector); | |
2249 | } | |
db830c46 | 2250 | peer_req->w.cb = w_e_end_ov_req; |
b411b363 | 2251 | fault_type = DRBD_FAULT_RS_RD; |
b411b363 PR |
2252 | break; |
2253 | ||
b411b363 PR |
2254 | default: |
2255 | dev_err(DEV, "unexpected command (%s) in receive_DataRequest\n", | |
02918be2 | 2256 | cmdname(cmd)); |
b411b363 | 2257 | fault_type = DRBD_FAULT_MAX; |
80a40e43 | 2258 | goto out_free_e; |
b411b363 PR |
2259 | } |
2260 | ||
0f0601f4 LE |
2261 | /* Throttle, drbd_rs_begin_io and submit should become asynchronous |
2262 | * wrt the receiver, but it is not as straightforward as it may seem. | |
2263 | * Various places in the resync start and stop logic assume resync | |
2264 | * requests are processed in order, requeuing this on the worker thread | |
2265 | * introduces a bunch of new code for synchronization between threads. | |
2266 | * | |
2267 | * Unlimited throttling before drbd_rs_begin_io may stall the resync | |
2268 | * "forever", throttling after drbd_rs_begin_io will lock that extent | |
2269 | * for application writes for the same time. For now, just throttle | |
2270 | * here, where the rest of the code expects the receiver to sleep for | |
2271 | * a while, anyways. | |
2272 | */ | |
2273 | ||
2274 | /* Throttle before drbd_rs_begin_io, as that locks out application IO; | |
2275 | * this defers syncer requests for some time, before letting at least | |
2276 | * on request through. The resync controller on the receiving side | |
2277 | * will adapt to the incoming rate accordingly. | |
2278 | * | |
2279 | * We cannot throttle here if remote is Primary/SyncTarget: | |
2280 | * we would also throttle its application reads. | |
2281 | * In that case, throttling is done on the SyncTarget only. | |
2282 | */ | |
e3555d85 PR |
2283 | if (mdev->state.peer != R_PRIMARY && drbd_rs_should_slow_down(mdev, sector)) |
2284 | schedule_timeout_uninterruptible(HZ/10); | |
2285 | if (drbd_rs_begin_io(mdev, sector)) | |
80a40e43 | 2286 | goto out_free_e; |
b411b363 | 2287 | |
0f0601f4 LE |
2288 | submit_for_resync: |
2289 | atomic_add(size >> 9, &mdev->rs_sect_ev); | |
2290 | ||
80a40e43 | 2291 | submit: |
b411b363 | 2292 | inc_unacked(mdev); |
87eeee41 | 2293 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2294 | list_add_tail(&peer_req->w.list, &mdev->read_ee); |
87eeee41 | 2295 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 2296 | |
fbe29dec | 2297 | if (drbd_submit_peer_request(mdev, peer_req, READ, fault_type) == 0) |
81e84650 | 2298 | return true; |
b411b363 | 2299 | |
10f6d992 LE |
2300 | /* don't care for the reason here */ |
2301 | dev_err(DEV, "submit failed, triggering re-connect\n"); | |
87eeee41 | 2302 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 2303 | list_del(&peer_req->w.list); |
87eeee41 | 2304 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 LE |
2305 | /* no drbd_rs_complete_io(), we are dropping the connection anyways */ |
2306 | ||
b411b363 | 2307 | out_free_e: |
b411b363 | 2308 | put_ldev(mdev); |
db830c46 | 2309 | drbd_free_ee(mdev, peer_req); |
81e84650 | 2310 | return false; |
b411b363 PR |
2311 | } |
2312 | ||
2313 | static int drbd_asb_recover_0p(struct drbd_conf *mdev) __must_hold(local) | |
2314 | { | |
2315 | int self, peer, rv = -100; | |
2316 | unsigned long ch_self, ch_peer; | |
2317 | ||
2318 | self = mdev->ldev->md.uuid[UI_BITMAP] & 1; | |
2319 | peer = mdev->p_uuid[UI_BITMAP] & 1; | |
2320 | ||
2321 | ch_peer = mdev->p_uuid[UI_SIZE]; | |
2322 | ch_self = mdev->comm_bm_set; | |
2323 | ||
89e58e75 | 2324 | switch (mdev->tconn->net_conf->after_sb_0p) { |
b411b363 PR |
2325 | case ASB_CONSENSUS: |
2326 | case ASB_DISCARD_SECONDARY: | |
2327 | case ASB_CALL_HELPER: | |
2328 | dev_err(DEV, "Configuration error.\n"); | |
2329 | break; | |
2330 | case ASB_DISCONNECT: | |
2331 | break; | |
2332 | case ASB_DISCARD_YOUNGER_PRI: | |
2333 | if (self == 0 && peer == 1) { | |
2334 | rv = -1; | |
2335 | break; | |
2336 | } | |
2337 | if (self == 1 && peer == 0) { | |
2338 | rv = 1; | |
2339 | break; | |
2340 | } | |
2341 | /* Else fall through to one of the other strategies... */ | |
2342 | case ASB_DISCARD_OLDER_PRI: | |
2343 | if (self == 0 && peer == 1) { | |
2344 | rv = 1; | |
2345 | break; | |
2346 | } | |
2347 | if (self == 1 && peer == 0) { | |
2348 | rv = -1; | |
2349 | break; | |
2350 | } | |
2351 | /* Else fall through to one of the other strategies... */ | |
ad19bf6e | 2352 | dev_warn(DEV, "Discard younger/older primary did not find a decision\n" |
b411b363 PR |
2353 | "Using discard-least-changes instead\n"); |
2354 | case ASB_DISCARD_ZERO_CHG: | |
2355 | if (ch_peer == 0 && ch_self == 0) { | |
25703f83 | 2356 | rv = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags) |
b411b363 PR |
2357 | ? -1 : 1; |
2358 | break; | |
2359 | } else { | |
2360 | if (ch_peer == 0) { rv = 1; break; } | |
2361 | if (ch_self == 0) { rv = -1; break; } | |
2362 | } | |
89e58e75 | 2363 | if (mdev->tconn->net_conf->after_sb_0p == ASB_DISCARD_ZERO_CHG) |
b411b363 PR |
2364 | break; |
2365 | case ASB_DISCARD_LEAST_CHG: | |
2366 | if (ch_self < ch_peer) | |
2367 | rv = -1; | |
2368 | else if (ch_self > ch_peer) | |
2369 | rv = 1; | |
2370 | else /* ( ch_self == ch_peer ) */ | |
2371 | /* Well, then use something else. */ | |
25703f83 | 2372 | rv = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags) |
b411b363 PR |
2373 | ? -1 : 1; |
2374 | break; | |
2375 | case ASB_DISCARD_LOCAL: | |
2376 | rv = -1; | |
2377 | break; | |
2378 | case ASB_DISCARD_REMOTE: | |
2379 | rv = 1; | |
2380 | } | |
2381 | ||
2382 | return rv; | |
2383 | } | |
2384 | ||
2385 | static int drbd_asb_recover_1p(struct drbd_conf *mdev) __must_hold(local) | |
2386 | { | |
6184ea21 | 2387 | int hg, rv = -100; |
b411b363 | 2388 | |
89e58e75 | 2389 | switch (mdev->tconn->net_conf->after_sb_1p) { |
b411b363 PR |
2390 | case ASB_DISCARD_YOUNGER_PRI: |
2391 | case ASB_DISCARD_OLDER_PRI: | |
2392 | case ASB_DISCARD_LEAST_CHG: | |
2393 | case ASB_DISCARD_LOCAL: | |
2394 | case ASB_DISCARD_REMOTE: | |
2395 | dev_err(DEV, "Configuration error.\n"); | |
2396 | break; | |
2397 | case ASB_DISCONNECT: | |
2398 | break; | |
2399 | case ASB_CONSENSUS: | |
2400 | hg = drbd_asb_recover_0p(mdev); | |
2401 | if (hg == -1 && mdev->state.role == R_SECONDARY) | |
2402 | rv = hg; | |
2403 | if (hg == 1 && mdev->state.role == R_PRIMARY) | |
2404 | rv = hg; | |
2405 | break; | |
2406 | case ASB_VIOLENTLY: | |
2407 | rv = drbd_asb_recover_0p(mdev); | |
2408 | break; | |
2409 | case ASB_DISCARD_SECONDARY: | |
2410 | return mdev->state.role == R_PRIMARY ? 1 : -1; | |
2411 | case ASB_CALL_HELPER: | |
2412 | hg = drbd_asb_recover_0p(mdev); | |
2413 | if (hg == -1 && mdev->state.role == R_PRIMARY) { | |
bb437946 AG |
2414 | enum drbd_state_rv rv2; |
2415 | ||
2416 | drbd_set_role(mdev, R_SECONDARY, 0); | |
b411b363 PR |
2417 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2418 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2419 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2420 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2421 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2422 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2423 | } else { | |
2424 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2425 | rv = hg; | |
2426 | } | |
2427 | } else | |
2428 | rv = hg; | |
2429 | } | |
2430 | ||
2431 | return rv; | |
2432 | } | |
2433 | ||
2434 | static int drbd_asb_recover_2p(struct drbd_conf *mdev) __must_hold(local) | |
2435 | { | |
6184ea21 | 2436 | int hg, rv = -100; |
b411b363 | 2437 | |
89e58e75 | 2438 | switch (mdev->tconn->net_conf->after_sb_2p) { |
b411b363 PR |
2439 | case ASB_DISCARD_YOUNGER_PRI: |
2440 | case ASB_DISCARD_OLDER_PRI: | |
2441 | case ASB_DISCARD_LEAST_CHG: | |
2442 | case ASB_DISCARD_LOCAL: | |
2443 | case ASB_DISCARD_REMOTE: | |
2444 | case ASB_CONSENSUS: | |
2445 | case ASB_DISCARD_SECONDARY: | |
2446 | dev_err(DEV, "Configuration error.\n"); | |
2447 | break; | |
2448 | case ASB_VIOLENTLY: | |
2449 | rv = drbd_asb_recover_0p(mdev); | |
2450 | break; | |
2451 | case ASB_DISCONNECT: | |
2452 | break; | |
2453 | case ASB_CALL_HELPER: | |
2454 | hg = drbd_asb_recover_0p(mdev); | |
2455 | if (hg == -1) { | |
bb437946 AG |
2456 | enum drbd_state_rv rv2; |
2457 | ||
b411b363 PR |
2458 | /* drbd_change_state() does not sleep while in SS_IN_TRANSIENT_STATE, |
2459 | * we might be here in C_WF_REPORT_PARAMS which is transient. | |
2460 | * we do not need to wait for the after state change work either. */ | |
bb437946 AG |
2461 | rv2 = drbd_change_state(mdev, CS_VERBOSE, NS(role, R_SECONDARY)); |
2462 | if (rv2 != SS_SUCCESS) { | |
b411b363 PR |
2463 | drbd_khelper(mdev, "pri-lost-after-sb"); |
2464 | } else { | |
2465 | dev_warn(DEV, "Successfully gave up primary role.\n"); | |
2466 | rv = hg; | |
2467 | } | |
2468 | } else | |
2469 | rv = hg; | |
2470 | } | |
2471 | ||
2472 | return rv; | |
2473 | } | |
2474 | ||
2475 | static void drbd_uuid_dump(struct drbd_conf *mdev, char *text, u64 *uuid, | |
2476 | u64 bits, u64 flags) | |
2477 | { | |
2478 | if (!uuid) { | |
2479 | dev_info(DEV, "%s uuid info vanished while I was looking!\n", text); | |
2480 | return; | |
2481 | } | |
2482 | dev_info(DEV, "%s %016llX:%016llX:%016llX:%016llX bits:%llu flags:%llX\n", | |
2483 | text, | |
2484 | (unsigned long long)uuid[UI_CURRENT], | |
2485 | (unsigned long long)uuid[UI_BITMAP], | |
2486 | (unsigned long long)uuid[UI_HISTORY_START], | |
2487 | (unsigned long long)uuid[UI_HISTORY_END], | |
2488 | (unsigned long long)bits, | |
2489 | (unsigned long long)flags); | |
2490 | } | |
2491 | ||
2492 | /* | |
2493 | 100 after split brain try auto recover | |
2494 | 2 C_SYNC_SOURCE set BitMap | |
2495 | 1 C_SYNC_SOURCE use BitMap | |
2496 | 0 no Sync | |
2497 | -1 C_SYNC_TARGET use BitMap | |
2498 | -2 C_SYNC_TARGET set BitMap | |
2499 | -100 after split brain, disconnect | |
2500 | -1000 unrelated data | |
4a23f264 PR |
2501 | -1091 requires proto 91 |
2502 | -1096 requires proto 96 | |
b411b363 PR |
2503 | */ |
2504 | static int drbd_uuid_compare(struct drbd_conf *mdev, int *rule_nr) __must_hold(local) | |
2505 | { | |
2506 | u64 self, peer; | |
2507 | int i, j; | |
2508 | ||
2509 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2510 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2511 | ||
2512 | *rule_nr = 10; | |
2513 | if (self == UUID_JUST_CREATED && peer == UUID_JUST_CREATED) | |
2514 | return 0; | |
2515 | ||
2516 | *rule_nr = 20; | |
2517 | if ((self == UUID_JUST_CREATED || self == (u64)0) && | |
2518 | peer != UUID_JUST_CREATED) | |
2519 | return -2; | |
2520 | ||
2521 | *rule_nr = 30; | |
2522 | if (self != UUID_JUST_CREATED && | |
2523 | (peer == UUID_JUST_CREATED || peer == (u64)0)) | |
2524 | return 2; | |
2525 | ||
2526 | if (self == peer) { | |
2527 | int rct, dc; /* roles at crash time */ | |
2528 | ||
2529 | if (mdev->p_uuid[UI_BITMAP] == (u64)0 && mdev->ldev->md.uuid[UI_BITMAP] != (u64)0) { | |
2530 | ||
31890f4a | 2531 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2532 | return -1091; |
b411b363 PR |
2533 | |
2534 | if ((mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) && | |
2535 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1))) { | |
2536 | dev_info(DEV, "was SyncSource, missed the resync finished event, corrected myself:\n"); | |
2537 | drbd_uuid_set_bm(mdev, 0UL); | |
2538 | ||
2539 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, | |
2540 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2541 | *rule_nr = 34; | |
2542 | } else { | |
2543 | dev_info(DEV, "was SyncSource (peer failed to write sync_uuid)\n"); | |
2544 | *rule_nr = 36; | |
2545 | } | |
2546 | ||
2547 | return 1; | |
2548 | } | |
2549 | ||
2550 | if (mdev->ldev->md.uuid[UI_BITMAP] == (u64)0 && mdev->p_uuid[UI_BITMAP] != (u64)0) { | |
2551 | ||
31890f4a | 2552 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2553 | return -1091; |
b411b363 PR |
2554 | |
2555 | if ((mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == (mdev->p_uuid[UI_BITMAP] & ~((u64)1)) && | |
2556 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1))) { | |
2557 | dev_info(DEV, "was SyncTarget, peer missed the resync finished event, corrected peer:\n"); | |
2558 | ||
2559 | mdev->p_uuid[UI_HISTORY_START + 1] = mdev->p_uuid[UI_HISTORY_START]; | |
2560 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_BITMAP]; | |
2561 | mdev->p_uuid[UI_BITMAP] = 0UL; | |
2562 | ||
2563 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2564 | *rule_nr = 35; | |
2565 | } else { | |
2566 | dev_info(DEV, "was SyncTarget (failed to write sync_uuid)\n"); | |
2567 | *rule_nr = 37; | |
2568 | } | |
2569 | ||
2570 | return -1; | |
2571 | } | |
2572 | ||
2573 | /* Common power [off|failure] */ | |
2574 | rct = (test_bit(CRASHED_PRIMARY, &mdev->flags) ? 1 : 0) + | |
2575 | (mdev->p_uuid[UI_FLAGS] & 2); | |
2576 | /* lowest bit is set when we were primary, | |
2577 | * next bit (weight 2) is set when peer was primary */ | |
2578 | *rule_nr = 40; | |
2579 | ||
2580 | switch (rct) { | |
2581 | case 0: /* !self_pri && !peer_pri */ return 0; | |
2582 | case 1: /* self_pri && !peer_pri */ return 1; | |
2583 | case 2: /* !self_pri && peer_pri */ return -1; | |
2584 | case 3: /* self_pri && peer_pri */ | |
25703f83 | 2585 | dc = test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags); |
b411b363 PR |
2586 | return dc ? -1 : 1; |
2587 | } | |
2588 | } | |
2589 | ||
2590 | *rule_nr = 50; | |
2591 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2592 | if (self == peer) | |
2593 | return -1; | |
2594 | ||
2595 | *rule_nr = 51; | |
2596 | peer = mdev->p_uuid[UI_HISTORY_START] & ~((u64)1); | |
2597 | if (self == peer) { | |
31890f4a | 2598 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2599 | (mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1)) == |
2600 | (mdev->p_uuid[UI_HISTORY_START + 1] & ~((u64)1)) : | |
2601 | peer + UUID_NEW_BM_OFFSET == (mdev->p_uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2602 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2603 | resync as sync source modifications of the peer's UUIDs. */ | |
2604 | ||
31890f4a | 2605 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2606 | return -1091; |
b411b363 PR |
2607 | |
2608 | mdev->p_uuid[UI_BITMAP] = mdev->p_uuid[UI_HISTORY_START]; | |
2609 | mdev->p_uuid[UI_HISTORY_START] = mdev->p_uuid[UI_HISTORY_START + 1]; | |
4a23f264 PR |
2610 | |
2611 | dev_info(DEV, "Did not got last syncUUID packet, corrected:\n"); | |
2612 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2613 | ||
b411b363 PR |
2614 | return -1; |
2615 | } | |
2616 | } | |
2617 | ||
2618 | *rule_nr = 60; | |
2619 | self = mdev->ldev->md.uuid[UI_CURRENT] & ~((u64)1); | |
2620 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2621 | peer = mdev->p_uuid[i] & ~((u64)1); | |
2622 | if (self == peer) | |
2623 | return -2; | |
2624 | } | |
2625 | ||
2626 | *rule_nr = 70; | |
2627 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2628 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); | |
2629 | if (self == peer) | |
2630 | return 1; | |
2631 | ||
2632 | *rule_nr = 71; | |
2633 | self = mdev->ldev->md.uuid[UI_HISTORY_START] & ~((u64)1); | |
2634 | if (self == peer) { | |
31890f4a | 2635 | if (mdev->tconn->agreed_pro_version < 96 ? |
4a23f264 PR |
2636 | (mdev->ldev->md.uuid[UI_HISTORY_START + 1] & ~((u64)1)) == |
2637 | (mdev->p_uuid[UI_HISTORY_START] & ~((u64)1)) : | |
2638 | self + UUID_NEW_BM_OFFSET == (mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1))) { | |
b411b363 PR |
2639 | /* The last P_SYNC_UUID did not get though. Undo the last start of |
2640 | resync as sync source modifications of our UUIDs. */ | |
2641 | ||
31890f4a | 2642 | if (mdev->tconn->agreed_pro_version < 91) |
4a23f264 | 2643 | return -1091; |
b411b363 PR |
2644 | |
2645 | _drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_HISTORY_START]); | |
2646 | _drbd_uuid_set(mdev, UI_HISTORY_START, mdev->ldev->md.uuid[UI_HISTORY_START + 1]); | |
2647 | ||
4a23f264 | 2648 | dev_info(DEV, "Last syncUUID did not get through, corrected:\n"); |
b411b363 PR |
2649 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, |
2650 | mdev->state.disk >= D_NEGOTIATING ? drbd_bm_total_weight(mdev) : 0, 0); | |
2651 | ||
2652 | return 1; | |
2653 | } | |
2654 | } | |
2655 | ||
2656 | ||
2657 | *rule_nr = 80; | |
d8c2a36b | 2658 | peer = mdev->p_uuid[UI_CURRENT] & ~((u64)1); |
b411b363 PR |
2659 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { |
2660 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2661 | if (self == peer) | |
2662 | return 2; | |
2663 | } | |
2664 | ||
2665 | *rule_nr = 90; | |
2666 | self = mdev->ldev->md.uuid[UI_BITMAP] & ~((u64)1); | |
2667 | peer = mdev->p_uuid[UI_BITMAP] & ~((u64)1); | |
2668 | if (self == peer && self != ((u64)0)) | |
2669 | return 100; | |
2670 | ||
2671 | *rule_nr = 100; | |
2672 | for (i = UI_HISTORY_START; i <= UI_HISTORY_END; i++) { | |
2673 | self = mdev->ldev->md.uuid[i] & ~((u64)1); | |
2674 | for (j = UI_HISTORY_START; j <= UI_HISTORY_END; j++) { | |
2675 | peer = mdev->p_uuid[j] & ~((u64)1); | |
2676 | if (self == peer) | |
2677 | return -100; | |
2678 | } | |
2679 | } | |
2680 | ||
2681 | return -1000; | |
2682 | } | |
2683 | ||
2684 | /* drbd_sync_handshake() returns the new conn state on success, or | |
2685 | CONN_MASK (-1) on failure. | |
2686 | */ | |
2687 | static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_role peer_role, | |
2688 | enum drbd_disk_state peer_disk) __must_hold(local) | |
2689 | { | |
2690 | int hg, rule_nr; | |
2691 | enum drbd_conns rv = C_MASK; | |
2692 | enum drbd_disk_state mydisk; | |
2693 | ||
2694 | mydisk = mdev->state.disk; | |
2695 | if (mydisk == D_NEGOTIATING) | |
2696 | mydisk = mdev->new_state_tmp.disk; | |
2697 | ||
2698 | dev_info(DEV, "drbd_sync_handshake:\n"); | |
2699 | drbd_uuid_dump(mdev, "self", mdev->ldev->md.uuid, mdev->comm_bm_set, 0); | |
2700 | drbd_uuid_dump(mdev, "peer", mdev->p_uuid, | |
2701 | mdev->p_uuid[UI_SIZE], mdev->p_uuid[UI_FLAGS]); | |
2702 | ||
2703 | hg = drbd_uuid_compare(mdev, &rule_nr); | |
2704 | ||
2705 | dev_info(DEV, "uuid_compare()=%d by rule %d\n", hg, rule_nr); | |
2706 | ||
2707 | if (hg == -1000) { | |
2708 | dev_alert(DEV, "Unrelated data, aborting!\n"); | |
2709 | return C_MASK; | |
2710 | } | |
4a23f264 PR |
2711 | if (hg < -1000) { |
2712 | dev_alert(DEV, "To resolve this both sides have to support at least protocol %d\n", -hg - 1000); | |
b411b363 PR |
2713 | return C_MASK; |
2714 | } | |
2715 | ||
2716 | if ((mydisk == D_INCONSISTENT && peer_disk > D_INCONSISTENT) || | |
2717 | (peer_disk == D_INCONSISTENT && mydisk > D_INCONSISTENT)) { | |
2718 | int f = (hg == -100) || abs(hg) == 2; | |
2719 | hg = mydisk > D_INCONSISTENT ? 1 : -1; | |
2720 | if (f) | |
2721 | hg = hg*2; | |
2722 | dev_info(DEV, "Becoming sync %s due to disk states.\n", | |
2723 | hg > 0 ? "source" : "target"); | |
2724 | } | |
2725 | ||
3a11a487 AG |
2726 | if (abs(hg) == 100) |
2727 | drbd_khelper(mdev, "initial-split-brain"); | |
2728 | ||
89e58e75 | 2729 | if (hg == 100 || (hg == -100 && mdev->tconn->net_conf->always_asbp)) { |
b411b363 PR |
2730 | int pcount = (mdev->state.role == R_PRIMARY) |
2731 | + (peer_role == R_PRIMARY); | |
2732 | int forced = (hg == -100); | |
2733 | ||
2734 | switch (pcount) { | |
2735 | case 0: | |
2736 | hg = drbd_asb_recover_0p(mdev); | |
2737 | break; | |
2738 | case 1: | |
2739 | hg = drbd_asb_recover_1p(mdev); | |
2740 | break; | |
2741 | case 2: | |
2742 | hg = drbd_asb_recover_2p(mdev); | |
2743 | break; | |
2744 | } | |
2745 | if (abs(hg) < 100) { | |
2746 | dev_warn(DEV, "Split-Brain detected, %d primaries, " | |
2747 | "automatically solved. Sync from %s node\n", | |
2748 | pcount, (hg < 0) ? "peer" : "this"); | |
2749 | if (forced) { | |
2750 | dev_warn(DEV, "Doing a full sync, since" | |
2751 | " UUIDs where ambiguous.\n"); | |
2752 | hg = hg*2; | |
2753 | } | |
2754 | } | |
2755 | } | |
2756 | ||
2757 | if (hg == -100) { | |
89e58e75 | 2758 | if (mdev->tconn->net_conf->want_lose && !(mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 | 2759 | hg = -1; |
89e58e75 | 2760 | if (!mdev->tconn->net_conf->want_lose && (mdev->p_uuid[UI_FLAGS]&1)) |
b411b363 PR |
2761 | hg = 1; |
2762 | ||
2763 | if (abs(hg) < 100) | |
2764 | dev_warn(DEV, "Split-Brain detected, manually solved. " | |
2765 | "Sync from %s node\n", | |
2766 | (hg < 0) ? "peer" : "this"); | |
2767 | } | |
2768 | ||
2769 | if (hg == -100) { | |
580b9767 LE |
2770 | /* FIXME this log message is not correct if we end up here |
2771 | * after an attempted attach on a diskless node. | |
2772 | * We just refuse to attach -- well, we drop the "connection" | |
2773 | * to that disk, in a way... */ | |
3a11a487 | 2774 | dev_alert(DEV, "Split-Brain detected but unresolved, dropping connection!\n"); |
b411b363 PR |
2775 | drbd_khelper(mdev, "split-brain"); |
2776 | return C_MASK; | |
2777 | } | |
2778 | ||
2779 | if (hg > 0 && mydisk <= D_INCONSISTENT) { | |
2780 | dev_err(DEV, "I shall become SyncSource, but I am inconsistent!\n"); | |
2781 | return C_MASK; | |
2782 | } | |
2783 | ||
2784 | if (hg < 0 && /* by intention we do not use mydisk here. */ | |
2785 | mdev->state.role == R_PRIMARY && mdev->state.disk >= D_CONSISTENT) { | |
89e58e75 | 2786 | switch (mdev->tconn->net_conf->rr_conflict) { |
b411b363 PR |
2787 | case ASB_CALL_HELPER: |
2788 | drbd_khelper(mdev, "pri-lost"); | |
2789 | /* fall through */ | |
2790 | case ASB_DISCONNECT: | |
2791 | dev_err(DEV, "I shall become SyncTarget, but I am primary!\n"); | |
2792 | return C_MASK; | |
2793 | case ASB_VIOLENTLY: | |
2794 | dev_warn(DEV, "Becoming SyncTarget, violating the stable-data" | |
2795 | "assumption\n"); | |
2796 | } | |
2797 | } | |
2798 | ||
89e58e75 | 2799 | if (mdev->tconn->net_conf->dry_run || test_bit(CONN_DRY_RUN, &mdev->flags)) { |
cf14c2e9 PR |
2800 | if (hg == 0) |
2801 | dev_info(DEV, "dry-run connect: No resync, would become Connected immediately.\n"); | |
2802 | else | |
2803 | dev_info(DEV, "dry-run connect: Would become %s, doing a %s resync.", | |
2804 | drbd_conn_str(hg > 0 ? C_SYNC_SOURCE : C_SYNC_TARGET), | |
2805 | abs(hg) >= 2 ? "full" : "bit-map based"); | |
2806 | return C_MASK; | |
2807 | } | |
2808 | ||
b411b363 PR |
2809 | if (abs(hg) >= 2) { |
2810 | dev_info(DEV, "Writing the whole bitmap, full sync required after drbd_sync_handshake.\n"); | |
20ceb2b2 LE |
2811 | if (drbd_bitmap_io(mdev, &drbd_bmio_set_n_write, "set_n_write from sync_handshake", |
2812 | BM_LOCKED_SET_ALLOWED)) | |
b411b363 PR |
2813 | return C_MASK; |
2814 | } | |
2815 | ||
2816 | if (hg > 0) { /* become sync source. */ | |
2817 | rv = C_WF_BITMAP_S; | |
2818 | } else if (hg < 0) { /* become sync target */ | |
2819 | rv = C_WF_BITMAP_T; | |
2820 | } else { | |
2821 | rv = C_CONNECTED; | |
2822 | if (drbd_bm_total_weight(mdev)) { | |
2823 | dev_info(DEV, "No resync, but %lu bits in bitmap!\n", | |
2824 | drbd_bm_total_weight(mdev)); | |
2825 | } | |
2826 | } | |
2827 | ||
2828 | return rv; | |
2829 | } | |
2830 | ||
2831 | /* returns 1 if invalid */ | |
2832 | static int cmp_after_sb(enum drbd_after_sb_p peer, enum drbd_after_sb_p self) | |
2833 | { | |
2834 | /* ASB_DISCARD_REMOTE - ASB_DISCARD_LOCAL is valid */ | |
2835 | if ((peer == ASB_DISCARD_REMOTE && self == ASB_DISCARD_LOCAL) || | |
2836 | (self == ASB_DISCARD_REMOTE && peer == ASB_DISCARD_LOCAL)) | |
2837 | return 0; | |
2838 | ||
2839 | /* any other things with ASB_DISCARD_REMOTE or ASB_DISCARD_LOCAL are invalid */ | |
2840 | if (peer == ASB_DISCARD_REMOTE || peer == ASB_DISCARD_LOCAL || | |
2841 | self == ASB_DISCARD_REMOTE || self == ASB_DISCARD_LOCAL) | |
2842 | return 1; | |
2843 | ||
2844 | /* everything else is valid if they are equal on both sides. */ | |
2845 | if (peer == self) | |
2846 | return 0; | |
2847 | ||
2848 | /* everything es is invalid. */ | |
2849 | return 1; | |
2850 | } | |
2851 | ||
d8763023 AG |
2852 | static int receive_protocol(struct drbd_conf *mdev, enum drbd_packet cmd, |
2853 | unsigned int data_size) | |
b411b363 | 2854 | { |
e42325a5 | 2855 | struct p_protocol *p = &mdev->tconn->data.rbuf.protocol; |
b411b363 | 2856 | int p_proto, p_after_sb_0p, p_after_sb_1p, p_after_sb_2p; |
cf14c2e9 | 2857 | int p_want_lose, p_two_primaries, cf; |
b411b363 PR |
2858 | char p_integrity_alg[SHARED_SECRET_MAX] = ""; |
2859 | ||
b411b363 PR |
2860 | p_proto = be32_to_cpu(p->protocol); |
2861 | p_after_sb_0p = be32_to_cpu(p->after_sb_0p); | |
2862 | p_after_sb_1p = be32_to_cpu(p->after_sb_1p); | |
2863 | p_after_sb_2p = be32_to_cpu(p->after_sb_2p); | |
b411b363 | 2864 | p_two_primaries = be32_to_cpu(p->two_primaries); |
cf14c2e9 PR |
2865 | cf = be32_to_cpu(p->conn_flags); |
2866 | p_want_lose = cf & CF_WANT_LOSE; | |
2867 | ||
2868 | clear_bit(CONN_DRY_RUN, &mdev->flags); | |
2869 | ||
2870 | if (cf & CF_DRY_RUN) | |
2871 | set_bit(CONN_DRY_RUN, &mdev->flags); | |
b411b363 | 2872 | |
89e58e75 | 2873 | if (p_proto != mdev->tconn->net_conf->wire_protocol) { |
b411b363 PR |
2874 | dev_err(DEV, "incompatible communication protocols\n"); |
2875 | goto disconnect; | |
2876 | } | |
2877 | ||
89e58e75 | 2878 | if (cmp_after_sb(p_after_sb_0p, mdev->tconn->net_conf->after_sb_0p)) { |
b411b363 PR |
2879 | dev_err(DEV, "incompatible after-sb-0pri settings\n"); |
2880 | goto disconnect; | |
2881 | } | |
2882 | ||
89e58e75 | 2883 | if (cmp_after_sb(p_after_sb_1p, mdev->tconn->net_conf->after_sb_1p)) { |
b411b363 PR |
2884 | dev_err(DEV, "incompatible after-sb-1pri settings\n"); |
2885 | goto disconnect; | |
2886 | } | |
2887 | ||
89e58e75 | 2888 | if (cmp_after_sb(p_after_sb_2p, mdev->tconn->net_conf->after_sb_2p)) { |
b411b363 PR |
2889 | dev_err(DEV, "incompatible after-sb-2pri settings\n"); |
2890 | goto disconnect; | |
2891 | } | |
2892 | ||
89e58e75 | 2893 | if (p_want_lose && mdev->tconn->net_conf->want_lose) { |
b411b363 PR |
2894 | dev_err(DEV, "both sides have the 'want_lose' flag set\n"); |
2895 | goto disconnect; | |
2896 | } | |
2897 | ||
89e58e75 | 2898 | if (p_two_primaries != mdev->tconn->net_conf->two_primaries) { |
b411b363 PR |
2899 | dev_err(DEV, "incompatible setting of the two-primaries options\n"); |
2900 | goto disconnect; | |
2901 | } | |
2902 | ||
31890f4a | 2903 | if (mdev->tconn->agreed_pro_version >= 87) { |
89e58e75 | 2904 | unsigned char *my_alg = mdev->tconn->net_conf->integrity_alg; |
b411b363 | 2905 | |
de0ff338 | 2906 | if (drbd_recv(mdev->tconn, p_integrity_alg, data_size) != data_size) |
81e84650 | 2907 | return false; |
b411b363 PR |
2908 | |
2909 | p_integrity_alg[SHARED_SECRET_MAX-1] = 0; | |
2910 | if (strcmp(p_integrity_alg, my_alg)) { | |
2911 | dev_err(DEV, "incompatible setting of the data-integrity-alg\n"); | |
2912 | goto disconnect; | |
2913 | } | |
2914 | dev_info(DEV, "data-integrity-alg: %s\n", | |
2915 | my_alg[0] ? my_alg : (unsigned char *)"<not-used>"); | |
2916 | } | |
2917 | ||
81e84650 | 2918 | return true; |
b411b363 PR |
2919 | |
2920 | disconnect: | |
2921 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 2922 | return false; |
b411b363 PR |
2923 | } |
2924 | ||
2925 | /* helper function | |
2926 | * input: alg name, feature name | |
2927 | * return: NULL (alg name was "") | |
2928 | * ERR_PTR(error) if something goes wrong | |
2929 | * or the crypto hash ptr, if it worked out ok. */ | |
2930 | struct crypto_hash *drbd_crypto_alloc_digest_safe(const struct drbd_conf *mdev, | |
2931 | const char *alg, const char *name) | |
2932 | { | |
2933 | struct crypto_hash *tfm; | |
2934 | ||
2935 | if (!alg[0]) | |
2936 | return NULL; | |
2937 | ||
2938 | tfm = crypto_alloc_hash(alg, 0, CRYPTO_ALG_ASYNC); | |
2939 | if (IS_ERR(tfm)) { | |
2940 | dev_err(DEV, "Can not allocate \"%s\" as %s (reason: %ld)\n", | |
2941 | alg, name, PTR_ERR(tfm)); | |
2942 | return tfm; | |
2943 | } | |
2944 | if (!drbd_crypto_is_hash(crypto_hash_tfm(tfm))) { | |
2945 | crypto_free_hash(tfm); | |
2946 | dev_err(DEV, "\"%s\" is not a digest (%s)\n", alg, name); | |
2947 | return ERR_PTR(-EINVAL); | |
2948 | } | |
2949 | return tfm; | |
2950 | } | |
2951 | ||
d8763023 AG |
2952 | static int receive_SyncParam(struct drbd_conf *mdev, enum drbd_packet cmd, |
2953 | unsigned int packet_size) | |
b411b363 | 2954 | { |
81e84650 | 2955 | int ok = true; |
e42325a5 | 2956 | struct p_rs_param_95 *p = &mdev->tconn->data.rbuf.rs_param_95; |
b411b363 PR |
2957 | unsigned int header_size, data_size, exp_max_sz; |
2958 | struct crypto_hash *verify_tfm = NULL; | |
2959 | struct crypto_hash *csums_tfm = NULL; | |
31890f4a | 2960 | const int apv = mdev->tconn->agreed_pro_version; |
778f271d PR |
2961 | int *rs_plan_s = NULL; |
2962 | int fifo_size = 0; | |
b411b363 PR |
2963 | |
2964 | exp_max_sz = apv <= 87 ? sizeof(struct p_rs_param) | |
2965 | : apv == 88 ? sizeof(struct p_rs_param) | |
2966 | + SHARED_SECRET_MAX | |
8e26f9cc PR |
2967 | : apv <= 94 ? sizeof(struct p_rs_param_89) |
2968 | : /* apv >= 95 */ sizeof(struct p_rs_param_95); | |
b411b363 | 2969 | |
02918be2 | 2970 | if (packet_size > exp_max_sz) { |
b411b363 | 2971 | dev_err(DEV, "SyncParam packet too long: received %u, expected <= %u bytes\n", |
02918be2 | 2972 | packet_size, exp_max_sz); |
81e84650 | 2973 | return false; |
b411b363 PR |
2974 | } |
2975 | ||
2976 | if (apv <= 88) { | |
257d0af6 | 2977 | header_size = sizeof(struct p_rs_param) - sizeof(struct p_header); |
02918be2 | 2978 | data_size = packet_size - header_size; |
8e26f9cc | 2979 | } else if (apv <= 94) { |
257d0af6 | 2980 | header_size = sizeof(struct p_rs_param_89) - sizeof(struct p_header); |
02918be2 | 2981 | data_size = packet_size - header_size; |
b411b363 | 2982 | D_ASSERT(data_size == 0); |
8e26f9cc | 2983 | } else { |
257d0af6 | 2984 | header_size = sizeof(struct p_rs_param_95) - sizeof(struct p_header); |
02918be2 | 2985 | data_size = packet_size - header_size; |
b411b363 PR |
2986 | D_ASSERT(data_size == 0); |
2987 | } | |
2988 | ||
2989 | /* initialize verify_alg and csums_alg */ | |
2990 | memset(p->verify_alg, 0, 2 * SHARED_SECRET_MAX); | |
2991 | ||
de0ff338 | 2992 | if (drbd_recv(mdev->tconn, &p->head.payload, header_size) != header_size) |
81e84650 | 2993 | return false; |
b411b363 PR |
2994 | |
2995 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
2996 | ||
2997 | if (apv >= 88) { | |
2998 | if (apv == 88) { | |
2999 | if (data_size > SHARED_SECRET_MAX) { | |
3000 | dev_err(DEV, "verify-alg too long, " | |
3001 | "peer wants %u, accepting only %u byte\n", | |
3002 | data_size, SHARED_SECRET_MAX); | |
81e84650 | 3003 | return false; |
b411b363 PR |
3004 | } |
3005 | ||
de0ff338 | 3006 | if (drbd_recv(mdev->tconn, p->verify_alg, data_size) != data_size) |
81e84650 | 3007 | return false; |
b411b363 PR |
3008 | |
3009 | /* we expect NUL terminated string */ | |
3010 | /* but just in case someone tries to be evil */ | |
3011 | D_ASSERT(p->verify_alg[data_size-1] == 0); | |
3012 | p->verify_alg[data_size-1] = 0; | |
3013 | ||
3014 | } else /* apv >= 89 */ { | |
3015 | /* we still expect NUL terminated strings */ | |
3016 | /* but just in case someone tries to be evil */ | |
3017 | D_ASSERT(p->verify_alg[SHARED_SECRET_MAX-1] == 0); | |
3018 | D_ASSERT(p->csums_alg[SHARED_SECRET_MAX-1] == 0); | |
3019 | p->verify_alg[SHARED_SECRET_MAX-1] = 0; | |
3020 | p->csums_alg[SHARED_SECRET_MAX-1] = 0; | |
3021 | } | |
3022 | ||
3023 | if (strcmp(mdev->sync_conf.verify_alg, p->verify_alg)) { | |
3024 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
3025 | dev_err(DEV, "Different verify-alg settings. me=\"%s\" peer=\"%s\"\n", | |
3026 | mdev->sync_conf.verify_alg, p->verify_alg); | |
3027 | goto disconnect; | |
3028 | } | |
3029 | verify_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3030 | p->verify_alg, "verify-alg"); | |
3031 | if (IS_ERR(verify_tfm)) { | |
3032 | verify_tfm = NULL; | |
3033 | goto disconnect; | |
3034 | } | |
3035 | } | |
3036 | ||
3037 | if (apv >= 89 && strcmp(mdev->sync_conf.csums_alg, p->csums_alg)) { | |
3038 | if (mdev->state.conn == C_WF_REPORT_PARAMS) { | |
3039 | dev_err(DEV, "Different csums-alg settings. me=\"%s\" peer=\"%s\"\n", | |
3040 | mdev->sync_conf.csums_alg, p->csums_alg); | |
3041 | goto disconnect; | |
3042 | } | |
3043 | csums_tfm = drbd_crypto_alloc_digest_safe(mdev, | |
3044 | p->csums_alg, "csums-alg"); | |
3045 | if (IS_ERR(csums_tfm)) { | |
3046 | csums_tfm = NULL; | |
3047 | goto disconnect; | |
3048 | } | |
3049 | } | |
3050 | ||
8e26f9cc PR |
3051 | if (apv > 94) { |
3052 | mdev->sync_conf.rate = be32_to_cpu(p->rate); | |
3053 | mdev->sync_conf.c_plan_ahead = be32_to_cpu(p->c_plan_ahead); | |
3054 | mdev->sync_conf.c_delay_target = be32_to_cpu(p->c_delay_target); | |
3055 | mdev->sync_conf.c_fill_target = be32_to_cpu(p->c_fill_target); | |
3056 | mdev->sync_conf.c_max_rate = be32_to_cpu(p->c_max_rate); | |
778f271d PR |
3057 | |
3058 | fifo_size = (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; | |
3059 | if (fifo_size != mdev->rs_plan_s.size && fifo_size > 0) { | |
3060 | rs_plan_s = kzalloc(sizeof(int) * fifo_size, GFP_KERNEL); | |
3061 | if (!rs_plan_s) { | |
3062 | dev_err(DEV, "kmalloc of fifo_buffer failed"); | |
3063 | goto disconnect; | |
3064 | } | |
3065 | } | |
8e26f9cc | 3066 | } |
b411b363 PR |
3067 | |
3068 | spin_lock(&mdev->peer_seq_lock); | |
3069 | /* lock against drbd_nl_syncer_conf() */ | |
3070 | if (verify_tfm) { | |
3071 | strcpy(mdev->sync_conf.verify_alg, p->verify_alg); | |
3072 | mdev->sync_conf.verify_alg_len = strlen(p->verify_alg) + 1; | |
3073 | crypto_free_hash(mdev->verify_tfm); | |
3074 | mdev->verify_tfm = verify_tfm; | |
3075 | dev_info(DEV, "using verify-alg: \"%s\"\n", p->verify_alg); | |
3076 | } | |
3077 | if (csums_tfm) { | |
3078 | strcpy(mdev->sync_conf.csums_alg, p->csums_alg); | |
3079 | mdev->sync_conf.csums_alg_len = strlen(p->csums_alg) + 1; | |
3080 | crypto_free_hash(mdev->csums_tfm); | |
3081 | mdev->csums_tfm = csums_tfm; | |
3082 | dev_info(DEV, "using csums-alg: \"%s\"\n", p->csums_alg); | |
3083 | } | |
778f271d PR |
3084 | if (fifo_size != mdev->rs_plan_s.size) { |
3085 | kfree(mdev->rs_plan_s.values); | |
3086 | mdev->rs_plan_s.values = rs_plan_s; | |
3087 | mdev->rs_plan_s.size = fifo_size; | |
3088 | mdev->rs_planed = 0; | |
3089 | } | |
b411b363 PR |
3090 | spin_unlock(&mdev->peer_seq_lock); |
3091 | } | |
3092 | ||
3093 | return ok; | |
3094 | disconnect: | |
3095 | /* just for completeness: actually not needed, | |
3096 | * as this is not reached if csums_tfm was ok. */ | |
3097 | crypto_free_hash(csums_tfm); | |
3098 | /* but free the verify_tfm again, if csums_tfm did not work out */ | |
3099 | crypto_free_hash(verify_tfm); | |
3100 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3101 | return false; |
b411b363 PR |
3102 | } |
3103 | ||
b411b363 PR |
3104 | /* warn if the arguments differ by more than 12.5% */ |
3105 | static void warn_if_differ_considerably(struct drbd_conf *mdev, | |
3106 | const char *s, sector_t a, sector_t b) | |
3107 | { | |
3108 | sector_t d; | |
3109 | if (a == 0 || b == 0) | |
3110 | return; | |
3111 | d = (a > b) ? (a - b) : (b - a); | |
3112 | if (d > (a>>3) || d > (b>>3)) | |
3113 | dev_warn(DEV, "Considerable difference in %s: %llus vs. %llus\n", s, | |
3114 | (unsigned long long)a, (unsigned long long)b); | |
3115 | } | |
3116 | ||
d8763023 AG |
3117 | static int receive_sizes(struct drbd_conf *mdev, enum drbd_packet cmd, |
3118 | unsigned int data_size) | |
b411b363 | 3119 | { |
e42325a5 | 3120 | struct p_sizes *p = &mdev->tconn->data.rbuf.sizes; |
b411b363 | 3121 | enum determine_dev_size dd = unchanged; |
b411b363 PR |
3122 | sector_t p_size, p_usize, my_usize; |
3123 | int ldsc = 0; /* local disk size changed */ | |
e89b591c | 3124 | enum dds_flags ddsf; |
b411b363 | 3125 | |
b411b363 PR |
3126 | p_size = be64_to_cpu(p->d_size); |
3127 | p_usize = be64_to_cpu(p->u_size); | |
3128 | ||
3129 | if (p_size == 0 && mdev->state.disk == D_DISKLESS) { | |
3130 | dev_err(DEV, "some backing storage is needed\n"); | |
3131 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3132 | return false; |
b411b363 PR |
3133 | } |
3134 | ||
3135 | /* just store the peer's disk size for now. | |
3136 | * we still need to figure out whether we accept that. */ | |
3137 | mdev->p_size = p_size; | |
3138 | ||
b411b363 PR |
3139 | if (get_ldev(mdev)) { |
3140 | warn_if_differ_considerably(mdev, "lower level device sizes", | |
3141 | p_size, drbd_get_max_capacity(mdev->ldev)); | |
3142 | warn_if_differ_considerably(mdev, "user requested size", | |
3143 | p_usize, mdev->ldev->dc.disk_size); | |
3144 | ||
3145 | /* if this is the first connect, or an otherwise expected | |
3146 | * param exchange, choose the minimum */ | |
3147 | if (mdev->state.conn == C_WF_REPORT_PARAMS) | |
3148 | p_usize = min_not_zero((sector_t)mdev->ldev->dc.disk_size, | |
3149 | p_usize); | |
3150 | ||
3151 | my_usize = mdev->ldev->dc.disk_size; | |
3152 | ||
3153 | if (mdev->ldev->dc.disk_size != p_usize) { | |
3154 | mdev->ldev->dc.disk_size = p_usize; | |
3155 | dev_info(DEV, "Peer sets u_size to %lu sectors\n", | |
3156 | (unsigned long)mdev->ldev->dc.disk_size); | |
3157 | } | |
3158 | ||
3159 | /* Never shrink a device with usable data during connect. | |
3160 | But allow online shrinking if we are connected. */ | |
a393db6f | 3161 | if (drbd_new_dev_size(mdev, mdev->ldev, 0) < |
b411b363 PR |
3162 | drbd_get_capacity(mdev->this_bdev) && |
3163 | mdev->state.disk >= D_OUTDATED && | |
3164 | mdev->state.conn < C_CONNECTED) { | |
3165 | dev_err(DEV, "The peer's disk size is too small!\n"); | |
3166 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
3167 | mdev->ldev->dc.disk_size = my_usize; | |
3168 | put_ldev(mdev); | |
81e84650 | 3169 | return false; |
b411b363 PR |
3170 | } |
3171 | put_ldev(mdev); | |
3172 | } | |
b411b363 | 3173 | |
e89b591c | 3174 | ddsf = be16_to_cpu(p->dds_flags); |
b411b363 | 3175 | if (get_ldev(mdev)) { |
24c4830c | 3176 | dd = drbd_determine_dev_size(mdev, ddsf); |
b411b363 PR |
3177 | put_ldev(mdev); |
3178 | if (dd == dev_size_error) | |
81e84650 | 3179 | return false; |
b411b363 PR |
3180 | drbd_md_sync(mdev); |
3181 | } else { | |
3182 | /* I am diskless, need to accept the peer's size. */ | |
3183 | drbd_set_my_capacity(mdev, p_size); | |
3184 | } | |
3185 | ||
99432fcc PR |
3186 | mdev->peer_max_bio_size = be32_to_cpu(p->max_bio_size); |
3187 | drbd_reconsider_max_bio_size(mdev); | |
3188 | ||
b411b363 PR |
3189 | if (get_ldev(mdev)) { |
3190 | if (mdev->ldev->known_size != drbd_get_capacity(mdev->ldev->backing_bdev)) { | |
3191 | mdev->ldev->known_size = drbd_get_capacity(mdev->ldev->backing_bdev); | |
3192 | ldsc = 1; | |
3193 | } | |
3194 | ||
b411b363 PR |
3195 | put_ldev(mdev); |
3196 | } | |
3197 | ||
3198 | if (mdev->state.conn > C_WF_REPORT_PARAMS) { | |
3199 | if (be64_to_cpu(p->c_size) != | |
3200 | drbd_get_capacity(mdev->this_bdev) || ldsc) { | |
3201 | /* we have different sizes, probably peer | |
3202 | * needs to know my new size... */ | |
e89b591c | 3203 | drbd_send_sizes(mdev, 0, ddsf); |
b411b363 PR |
3204 | } |
3205 | if (test_and_clear_bit(RESIZE_PENDING, &mdev->flags) || | |
3206 | (dd == grew && mdev->state.conn == C_CONNECTED)) { | |
3207 | if (mdev->state.pdsk >= D_INCONSISTENT && | |
e89b591c PR |
3208 | mdev->state.disk >= D_INCONSISTENT) { |
3209 | if (ddsf & DDSF_NO_RESYNC) | |
3210 | dev_info(DEV, "Resync of new storage suppressed with --assume-clean\n"); | |
3211 | else | |
3212 | resync_after_online_grow(mdev); | |
3213 | } else | |
b411b363 PR |
3214 | set_bit(RESYNC_AFTER_NEG, &mdev->flags); |
3215 | } | |
3216 | } | |
3217 | ||
81e84650 | 3218 | return true; |
b411b363 PR |
3219 | } |
3220 | ||
d8763023 AG |
3221 | static int receive_uuids(struct drbd_conf *mdev, enum drbd_packet cmd, |
3222 | unsigned int data_size) | |
b411b363 | 3223 | { |
e42325a5 | 3224 | struct p_uuids *p = &mdev->tconn->data.rbuf.uuids; |
b411b363 | 3225 | u64 *p_uuid; |
62b0da3a | 3226 | int i, updated_uuids = 0; |
b411b363 | 3227 | |
b411b363 PR |
3228 | p_uuid = kmalloc(sizeof(u64)*UI_EXTENDED_SIZE, GFP_NOIO); |
3229 | ||
3230 | for (i = UI_CURRENT; i < UI_EXTENDED_SIZE; i++) | |
3231 | p_uuid[i] = be64_to_cpu(p->uuid[i]); | |
3232 | ||
3233 | kfree(mdev->p_uuid); | |
3234 | mdev->p_uuid = p_uuid; | |
3235 | ||
3236 | if (mdev->state.conn < C_CONNECTED && | |
3237 | mdev->state.disk < D_INCONSISTENT && | |
3238 | mdev->state.role == R_PRIMARY && | |
3239 | (mdev->ed_uuid & ~((u64)1)) != (p_uuid[UI_CURRENT] & ~((u64)1))) { | |
3240 | dev_err(DEV, "Can only connect to data with current UUID=%016llX\n", | |
3241 | (unsigned long long)mdev->ed_uuid); | |
3242 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3243 | return false; |
b411b363 PR |
3244 | } |
3245 | ||
3246 | if (get_ldev(mdev)) { | |
3247 | int skip_initial_sync = | |
3248 | mdev->state.conn == C_CONNECTED && | |
31890f4a | 3249 | mdev->tconn->agreed_pro_version >= 90 && |
b411b363 PR |
3250 | mdev->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && |
3251 | (p_uuid[UI_FLAGS] & 8); | |
3252 | if (skip_initial_sync) { | |
3253 | dev_info(DEV, "Accepted new current UUID, preparing to skip initial sync\n"); | |
3254 | drbd_bitmap_io(mdev, &drbd_bmio_clear_n_write, | |
20ceb2b2 LE |
3255 | "clear_n_write from receive_uuids", |
3256 | BM_LOCKED_TEST_ALLOWED); | |
b411b363 PR |
3257 | _drbd_uuid_set(mdev, UI_CURRENT, p_uuid[UI_CURRENT]); |
3258 | _drbd_uuid_set(mdev, UI_BITMAP, 0); | |
3259 | _drbd_set_state(_NS2(mdev, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE), | |
3260 | CS_VERBOSE, NULL); | |
3261 | drbd_md_sync(mdev); | |
62b0da3a | 3262 | updated_uuids = 1; |
b411b363 PR |
3263 | } |
3264 | put_ldev(mdev); | |
18a50fa2 PR |
3265 | } else if (mdev->state.disk < D_INCONSISTENT && |
3266 | mdev->state.role == R_PRIMARY) { | |
3267 | /* I am a diskless primary, the peer just created a new current UUID | |
3268 | for me. */ | |
62b0da3a | 3269 | updated_uuids = drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
b411b363 PR |
3270 | } |
3271 | ||
3272 | /* Before we test for the disk state, we should wait until an eventually | |
3273 | ongoing cluster wide state change is finished. That is important if | |
3274 | we are primary and are detaching from our disk. We need to see the | |
3275 | new disk state... */ | |
8410da8f PR |
3276 | mutex_lock(mdev->state_mutex); |
3277 | mutex_unlock(mdev->state_mutex); | |
b411b363 | 3278 | if (mdev->state.conn >= C_CONNECTED && mdev->state.disk < D_INCONSISTENT) |
62b0da3a LE |
3279 | updated_uuids |= drbd_set_ed_uuid(mdev, p_uuid[UI_CURRENT]); |
3280 | ||
3281 | if (updated_uuids) | |
3282 | drbd_print_uuids(mdev, "receiver updated UUIDs to"); | |
b411b363 | 3283 | |
81e84650 | 3284 | return true; |
b411b363 PR |
3285 | } |
3286 | ||
3287 | /** | |
3288 | * convert_state() - Converts the peer's view of the cluster state to our point of view | |
3289 | * @ps: The state as seen by the peer. | |
3290 | */ | |
3291 | static union drbd_state convert_state(union drbd_state ps) | |
3292 | { | |
3293 | union drbd_state ms; | |
3294 | ||
3295 | static enum drbd_conns c_tab[] = { | |
3296 | [C_CONNECTED] = C_CONNECTED, | |
3297 | ||
3298 | [C_STARTING_SYNC_S] = C_STARTING_SYNC_T, | |
3299 | [C_STARTING_SYNC_T] = C_STARTING_SYNC_S, | |
3300 | [C_DISCONNECTING] = C_TEAR_DOWN, /* C_NETWORK_FAILURE, */ | |
3301 | [C_VERIFY_S] = C_VERIFY_T, | |
3302 | [C_MASK] = C_MASK, | |
3303 | }; | |
3304 | ||
3305 | ms.i = ps.i; | |
3306 | ||
3307 | ms.conn = c_tab[ps.conn]; | |
3308 | ms.peer = ps.role; | |
3309 | ms.role = ps.peer; | |
3310 | ms.pdsk = ps.disk; | |
3311 | ms.disk = ps.pdsk; | |
3312 | ms.peer_isp = (ps.aftr_isp | ps.user_isp); | |
3313 | ||
3314 | return ms; | |
3315 | } | |
3316 | ||
d8763023 AG |
3317 | static int receive_req_state(struct drbd_conf *mdev, enum drbd_packet cmd, |
3318 | unsigned int data_size) | |
b411b363 | 3319 | { |
e42325a5 | 3320 | struct p_req_state *p = &mdev->tconn->data.rbuf.req_state; |
b411b363 | 3321 | union drbd_state mask, val; |
bf885f8a | 3322 | enum drbd_state_rv rv; |
b411b363 | 3323 | |
b411b363 PR |
3324 | mask.i = be32_to_cpu(p->mask); |
3325 | val.i = be32_to_cpu(p->val); | |
3326 | ||
25703f83 | 3327 | if (test_bit(DISCARD_CONCURRENT, &mdev->tconn->flags) && |
8410da8f | 3328 | mutex_is_locked(mdev->state_mutex)) { |
b411b363 | 3329 | drbd_send_sr_reply(mdev, SS_CONCURRENT_ST_CHG); |
81e84650 | 3330 | return true; |
b411b363 PR |
3331 | } |
3332 | ||
3333 | mask = convert_state(mask); | |
3334 | val = convert_state(val); | |
3335 | ||
047cd4a6 PR |
3336 | if (cmd == P_CONN_ST_CHG_REQ) { |
3337 | rv = conn_request_state(mdev->tconn, mask, val, CS_VERBOSE | CS_LOCAL_ONLY); | |
3338 | conn_send_sr_reply(mdev->tconn, rv); | |
3339 | } else { | |
3340 | rv = drbd_change_state(mdev, CS_VERBOSE, mask, val); | |
3341 | drbd_send_sr_reply(mdev, rv); | |
3342 | } | |
b411b363 | 3343 | |
b411b363 PR |
3344 | drbd_md_sync(mdev); |
3345 | ||
81e84650 | 3346 | return true; |
b411b363 PR |
3347 | } |
3348 | ||
d8763023 AG |
3349 | static int receive_state(struct drbd_conf *mdev, enum drbd_packet cmd, |
3350 | unsigned int data_size) | |
b411b363 | 3351 | { |
e42325a5 | 3352 | struct p_state *p = &mdev->tconn->data.rbuf.state; |
4ac4aada | 3353 | union drbd_state os, ns, peer_state; |
b411b363 | 3354 | enum drbd_disk_state real_peer_disk; |
65d922c3 | 3355 | enum chg_state_flags cs_flags; |
b411b363 PR |
3356 | int rv; |
3357 | ||
b411b363 PR |
3358 | peer_state.i = be32_to_cpu(p->state); |
3359 | ||
3360 | real_peer_disk = peer_state.disk; | |
3361 | if (peer_state.disk == D_NEGOTIATING) { | |
3362 | real_peer_disk = mdev->p_uuid[UI_FLAGS] & 4 ? D_INCONSISTENT : D_CONSISTENT; | |
3363 | dev_info(DEV, "real peer disk state = %s\n", drbd_disk_str(real_peer_disk)); | |
3364 | } | |
3365 | ||
87eeee41 | 3366 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 | 3367 | retry: |
4ac4aada | 3368 | os = ns = mdev->state; |
87eeee41 | 3369 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 3370 | |
e9ef7bb6 LE |
3371 | /* peer says his disk is uptodate, while we think it is inconsistent, |
3372 | * and this happens while we think we have a sync going on. */ | |
3373 | if (os.pdsk == D_INCONSISTENT && real_peer_disk == D_UP_TO_DATE && | |
3374 | os.conn > C_CONNECTED && os.disk == D_UP_TO_DATE) { | |
3375 | /* If we are (becoming) SyncSource, but peer is still in sync | |
3376 | * preparation, ignore its uptodate-ness to avoid flapping, it | |
3377 | * will change to inconsistent once the peer reaches active | |
3378 | * syncing states. | |
3379 | * It may have changed syncer-paused flags, however, so we | |
3380 | * cannot ignore this completely. */ | |
3381 | if (peer_state.conn > C_CONNECTED && | |
3382 | peer_state.conn < C_SYNC_SOURCE) | |
3383 | real_peer_disk = D_INCONSISTENT; | |
3384 | ||
3385 | /* if peer_state changes to connected at the same time, | |
3386 | * it explicitly notifies us that it finished resync. | |
3387 | * Maybe we should finish it up, too? */ | |
3388 | else if (os.conn >= C_SYNC_SOURCE && | |
3389 | peer_state.conn == C_CONNECTED) { | |
3390 | if (drbd_bm_total_weight(mdev) <= mdev->rs_failed) | |
3391 | drbd_resync_finished(mdev); | |
81e84650 | 3392 | return true; |
e9ef7bb6 LE |
3393 | } |
3394 | } | |
3395 | ||
3396 | /* peer says his disk is inconsistent, while we think it is uptodate, | |
3397 | * and this happens while the peer still thinks we have a sync going on, | |
3398 | * but we think we are already done with the sync. | |
3399 | * We ignore this to avoid flapping pdsk. | |
3400 | * This should not happen, if the peer is a recent version of drbd. */ | |
3401 | if (os.pdsk == D_UP_TO_DATE && real_peer_disk == D_INCONSISTENT && | |
3402 | os.conn == C_CONNECTED && peer_state.conn > C_SYNC_SOURCE) | |
3403 | real_peer_disk = D_UP_TO_DATE; | |
3404 | ||
4ac4aada LE |
3405 | if (ns.conn == C_WF_REPORT_PARAMS) |
3406 | ns.conn = C_CONNECTED; | |
b411b363 | 3407 | |
67531718 PR |
3408 | if (peer_state.conn == C_AHEAD) |
3409 | ns.conn = C_BEHIND; | |
3410 | ||
b411b363 PR |
3411 | if (mdev->p_uuid && peer_state.disk >= D_NEGOTIATING && |
3412 | get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3413 | int cr; /* consider resync */ | |
3414 | ||
3415 | /* if we established a new connection */ | |
4ac4aada | 3416 | cr = (os.conn < C_CONNECTED); |
b411b363 PR |
3417 | /* if we had an established connection |
3418 | * and one of the nodes newly attaches a disk */ | |
4ac4aada | 3419 | cr |= (os.conn == C_CONNECTED && |
b411b363 | 3420 | (peer_state.disk == D_NEGOTIATING || |
4ac4aada | 3421 | os.disk == D_NEGOTIATING)); |
b411b363 PR |
3422 | /* if we have both been inconsistent, and the peer has been |
3423 | * forced to be UpToDate with --overwrite-data */ | |
3424 | cr |= test_bit(CONSIDER_RESYNC, &mdev->flags); | |
3425 | /* if we had been plain connected, and the admin requested to | |
3426 | * start a sync by "invalidate" or "invalidate-remote" */ | |
4ac4aada | 3427 | cr |= (os.conn == C_CONNECTED && |
b411b363 PR |
3428 | (peer_state.conn >= C_STARTING_SYNC_S && |
3429 | peer_state.conn <= C_WF_BITMAP_T)); | |
3430 | ||
3431 | if (cr) | |
4ac4aada | 3432 | ns.conn = drbd_sync_handshake(mdev, peer_state.role, real_peer_disk); |
b411b363 PR |
3433 | |
3434 | put_ldev(mdev); | |
4ac4aada LE |
3435 | if (ns.conn == C_MASK) { |
3436 | ns.conn = C_CONNECTED; | |
b411b363 | 3437 | if (mdev->state.disk == D_NEGOTIATING) { |
82f59cc6 | 3438 | drbd_force_state(mdev, NS(disk, D_FAILED)); |
b411b363 PR |
3439 | } else if (peer_state.disk == D_NEGOTIATING) { |
3440 | dev_err(DEV, "Disk attach process on the peer node was aborted.\n"); | |
3441 | peer_state.disk = D_DISKLESS; | |
580b9767 | 3442 | real_peer_disk = D_DISKLESS; |
b411b363 | 3443 | } else { |
cf14c2e9 | 3444 | if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags)) |
81e84650 | 3445 | return false; |
4ac4aada | 3446 | D_ASSERT(os.conn == C_WF_REPORT_PARAMS); |
b411b363 | 3447 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); |
81e84650 | 3448 | return false; |
b411b363 PR |
3449 | } |
3450 | } | |
3451 | } | |
3452 | ||
87eeee41 | 3453 | spin_lock_irq(&mdev->tconn->req_lock); |
4ac4aada | 3454 | if (mdev->state.i != os.i) |
b411b363 PR |
3455 | goto retry; |
3456 | clear_bit(CONSIDER_RESYNC, &mdev->flags); | |
b411b363 PR |
3457 | ns.peer = peer_state.role; |
3458 | ns.pdsk = real_peer_disk; | |
3459 | ns.peer_isp = (peer_state.aftr_isp | peer_state.user_isp); | |
4ac4aada | 3460 | if ((ns.conn == C_CONNECTED || ns.conn == C_WF_BITMAP_S) && ns.disk == D_NEGOTIATING) |
b411b363 | 3461 | ns.disk = mdev->new_state_tmp.disk; |
4ac4aada LE |
3462 | cs_flags = CS_VERBOSE + (os.conn < C_CONNECTED && ns.conn >= C_CONNECTED ? 0 : CS_HARD); |
3463 | if (ns.pdsk == D_CONSISTENT && is_susp(ns) && ns.conn == C_CONNECTED && os.conn < C_CONNECTED && | |
481c6f50 | 3464 | test_bit(NEW_CUR_UUID, &mdev->flags)) { |
8554df1c | 3465 | /* Do not allow tl_restart(RESEND) for a rebooted peer. We can only allow this |
481c6f50 | 3466 | for temporal network outages! */ |
87eeee41 | 3467 | spin_unlock_irq(&mdev->tconn->req_lock); |
481c6f50 PR |
3468 | dev_err(DEV, "Aborting Connect, can not thaw IO with an only Consistent peer\n"); |
3469 | tl_clear(mdev); | |
3470 | drbd_uuid_new_current(mdev); | |
3471 | clear_bit(NEW_CUR_UUID, &mdev->flags); | |
3472 | drbd_force_state(mdev, NS2(conn, C_PROTOCOL_ERROR, susp, 0)); | |
81e84650 | 3473 | return false; |
481c6f50 | 3474 | } |
65d922c3 | 3475 | rv = _drbd_set_state(mdev, ns, cs_flags, NULL); |
b411b363 | 3476 | ns = mdev->state; |
87eeee41 | 3477 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3478 | |
3479 | if (rv < SS_SUCCESS) { | |
3480 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
81e84650 | 3481 | return false; |
b411b363 PR |
3482 | } |
3483 | ||
4ac4aada LE |
3484 | if (os.conn > C_WF_REPORT_PARAMS) { |
3485 | if (ns.conn > C_CONNECTED && peer_state.conn <= C_CONNECTED && | |
b411b363 PR |
3486 | peer_state.disk != D_NEGOTIATING ) { |
3487 | /* we want resync, peer has not yet decided to sync... */ | |
3488 | /* Nowadays only used when forcing a node into primary role and | |
3489 | setting its disk to UpToDate with that */ | |
3490 | drbd_send_uuids(mdev); | |
3491 | drbd_send_state(mdev); | |
3492 | } | |
3493 | } | |
3494 | ||
89e58e75 | 3495 | mdev->tconn->net_conf->want_lose = 0; |
b411b363 PR |
3496 | |
3497 | drbd_md_sync(mdev); /* update connected indicator, la_size, ... */ | |
3498 | ||
81e84650 | 3499 | return true; |
b411b363 PR |
3500 | } |
3501 | ||
d8763023 AG |
3502 | static int receive_sync_uuid(struct drbd_conf *mdev, enum drbd_packet cmd, |
3503 | unsigned int data_size) | |
b411b363 | 3504 | { |
e42325a5 | 3505 | struct p_rs_uuid *p = &mdev->tconn->data.rbuf.rs_uuid; |
b411b363 PR |
3506 | |
3507 | wait_event(mdev->misc_wait, | |
3508 | mdev->state.conn == C_WF_SYNC_UUID || | |
c4752ef1 | 3509 | mdev->state.conn == C_BEHIND || |
b411b363 PR |
3510 | mdev->state.conn < C_CONNECTED || |
3511 | mdev->state.disk < D_NEGOTIATING); | |
3512 | ||
3513 | /* D_ASSERT( mdev->state.conn == C_WF_SYNC_UUID ); */ | |
3514 | ||
b411b363 PR |
3515 | /* Here the _drbd_uuid_ functions are right, current should |
3516 | _not_ be rotated into the history */ | |
3517 | if (get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
3518 | _drbd_uuid_set(mdev, UI_CURRENT, be64_to_cpu(p->uuid)); | |
3519 | _drbd_uuid_set(mdev, UI_BITMAP, 0UL); | |
3520 | ||
62b0da3a | 3521 | drbd_print_uuids(mdev, "updated sync uuid"); |
b411b363 PR |
3522 | drbd_start_resync(mdev, C_SYNC_TARGET); |
3523 | ||
3524 | put_ldev(mdev); | |
3525 | } else | |
3526 | dev_err(DEV, "Ignoring SyncUUID packet!\n"); | |
3527 | ||
81e84650 | 3528 | return true; |
b411b363 PR |
3529 | } |
3530 | ||
2c46407d AG |
3531 | /** |
3532 | * receive_bitmap_plain | |
3533 | * | |
3534 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3535 | * code upon failure. | |
3536 | */ | |
3537 | static int | |
02918be2 PR |
3538 | receive_bitmap_plain(struct drbd_conf *mdev, unsigned int data_size, |
3539 | unsigned long *buffer, struct bm_xfer_ctx *c) | |
b411b363 PR |
3540 | { |
3541 | unsigned num_words = min_t(size_t, BM_PACKET_WORDS, c->bm_words - c->word_offset); | |
3542 | unsigned want = num_words * sizeof(long); | |
2c46407d | 3543 | int err; |
b411b363 | 3544 | |
02918be2 PR |
3545 | if (want != data_size) { |
3546 | dev_err(DEV, "%s:want (%u) != data_size (%u)\n", __func__, want, data_size); | |
2c46407d | 3547 | return -EIO; |
b411b363 PR |
3548 | } |
3549 | if (want == 0) | |
2c46407d | 3550 | return 0; |
de0ff338 | 3551 | err = drbd_recv(mdev->tconn, buffer, want); |
2c46407d AG |
3552 | if (err != want) { |
3553 | if (err >= 0) | |
3554 | err = -EIO; | |
3555 | return err; | |
3556 | } | |
b411b363 PR |
3557 | |
3558 | drbd_bm_merge_lel(mdev, c->word_offset, num_words, buffer); | |
3559 | ||
3560 | c->word_offset += num_words; | |
3561 | c->bit_offset = c->word_offset * BITS_PER_LONG; | |
3562 | if (c->bit_offset > c->bm_bits) | |
3563 | c->bit_offset = c->bm_bits; | |
3564 | ||
2c46407d | 3565 | return 1; |
b411b363 PR |
3566 | } |
3567 | ||
2c46407d AG |
3568 | /** |
3569 | * recv_bm_rle_bits | |
3570 | * | |
3571 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3572 | * code upon failure. | |
3573 | */ | |
3574 | static int | |
b411b363 PR |
3575 | recv_bm_rle_bits(struct drbd_conf *mdev, |
3576 | struct p_compressed_bm *p, | |
c6d25cfe PR |
3577 | struct bm_xfer_ctx *c, |
3578 | unsigned int len) | |
b411b363 PR |
3579 | { |
3580 | struct bitstream bs; | |
3581 | u64 look_ahead; | |
3582 | u64 rl; | |
3583 | u64 tmp; | |
3584 | unsigned long s = c->bit_offset; | |
3585 | unsigned long e; | |
b411b363 PR |
3586 | int toggle = DCBP_get_start(p); |
3587 | int have; | |
3588 | int bits; | |
3589 | ||
3590 | bitstream_init(&bs, p->code, len, DCBP_get_pad_bits(p)); | |
3591 | ||
3592 | bits = bitstream_get_bits(&bs, &look_ahead, 64); | |
3593 | if (bits < 0) | |
2c46407d | 3594 | return -EIO; |
b411b363 PR |
3595 | |
3596 | for (have = bits; have > 0; s += rl, toggle = !toggle) { | |
3597 | bits = vli_decode_bits(&rl, look_ahead); | |
3598 | if (bits <= 0) | |
2c46407d | 3599 | return -EIO; |
b411b363 PR |
3600 | |
3601 | if (toggle) { | |
3602 | e = s + rl -1; | |
3603 | if (e >= c->bm_bits) { | |
3604 | dev_err(DEV, "bitmap overflow (e:%lu) while decoding bm RLE packet\n", e); | |
2c46407d | 3605 | return -EIO; |
b411b363 PR |
3606 | } |
3607 | _drbd_bm_set_bits(mdev, s, e); | |
3608 | } | |
3609 | ||
3610 | if (have < bits) { | |
3611 | dev_err(DEV, "bitmap decoding error: h:%d b:%d la:0x%08llx l:%u/%u\n", | |
3612 | have, bits, look_ahead, | |
3613 | (unsigned int)(bs.cur.b - p->code), | |
3614 | (unsigned int)bs.buf_len); | |
2c46407d | 3615 | return -EIO; |
b411b363 PR |
3616 | } |
3617 | look_ahead >>= bits; | |
3618 | have -= bits; | |
3619 | ||
3620 | bits = bitstream_get_bits(&bs, &tmp, 64 - have); | |
3621 | if (bits < 0) | |
2c46407d | 3622 | return -EIO; |
b411b363 PR |
3623 | look_ahead |= tmp << have; |
3624 | have += bits; | |
3625 | } | |
3626 | ||
3627 | c->bit_offset = s; | |
3628 | bm_xfer_ctx_bit_to_word_offset(c); | |
3629 | ||
2c46407d | 3630 | return (s != c->bm_bits); |
b411b363 PR |
3631 | } |
3632 | ||
2c46407d AG |
3633 | /** |
3634 | * decode_bitmap_c | |
3635 | * | |
3636 | * Return 0 when done, 1 when another iteration is needed, and a negative error | |
3637 | * code upon failure. | |
3638 | */ | |
3639 | static int | |
b411b363 PR |
3640 | decode_bitmap_c(struct drbd_conf *mdev, |
3641 | struct p_compressed_bm *p, | |
c6d25cfe PR |
3642 | struct bm_xfer_ctx *c, |
3643 | unsigned int len) | |
b411b363 PR |
3644 | { |
3645 | if (DCBP_get_code(p) == RLE_VLI_Bits) | |
c6d25cfe | 3646 | return recv_bm_rle_bits(mdev, p, c, len); |
b411b363 PR |
3647 | |
3648 | /* other variants had been implemented for evaluation, | |
3649 | * but have been dropped as this one turned out to be "best" | |
3650 | * during all our tests. */ | |
3651 | ||
3652 | dev_err(DEV, "receive_bitmap_c: unknown encoding %u\n", p->encoding); | |
3653 | drbd_force_state(mdev, NS(conn, C_PROTOCOL_ERROR)); | |
2c46407d | 3654 | return -EIO; |
b411b363 PR |
3655 | } |
3656 | ||
3657 | void INFO_bm_xfer_stats(struct drbd_conf *mdev, | |
3658 | const char *direction, struct bm_xfer_ctx *c) | |
3659 | { | |
3660 | /* what would it take to transfer it "plaintext" */ | |
c012949a | 3661 | unsigned plain = sizeof(struct p_header) * |
b411b363 PR |
3662 | ((c->bm_words+BM_PACKET_WORDS-1)/BM_PACKET_WORDS+1) |
3663 | + c->bm_words * sizeof(long); | |
3664 | unsigned total = c->bytes[0] + c->bytes[1]; | |
3665 | unsigned r; | |
3666 | ||
3667 | /* total can not be zero. but just in case: */ | |
3668 | if (total == 0) | |
3669 | return; | |
3670 | ||
3671 | /* don't report if not compressed */ | |
3672 | if (total >= plain) | |
3673 | return; | |
3674 | ||
3675 | /* total < plain. check for overflow, still */ | |
3676 | r = (total > UINT_MAX/1000) ? (total / (plain/1000)) | |
3677 | : (1000 * total / plain); | |
3678 | ||
3679 | if (r > 1000) | |
3680 | r = 1000; | |
3681 | ||
3682 | r = 1000 - r; | |
3683 | dev_info(DEV, "%s bitmap stats [Bytes(packets)]: plain %u(%u), RLE %u(%u), " | |
3684 | "total %u; compression: %u.%u%%\n", | |
3685 | direction, | |
3686 | c->bytes[1], c->packets[1], | |
3687 | c->bytes[0], c->packets[0], | |
3688 | total, r/10, r % 10); | |
3689 | } | |
3690 | ||
3691 | /* Since we are processing the bitfield from lower addresses to higher, | |
3692 | it does not matter if the process it in 32 bit chunks or 64 bit | |
3693 | chunks as long as it is little endian. (Understand it as byte stream, | |
3694 | beginning with the lowest byte...) If we would use big endian | |
3695 | we would need to process it from the highest address to the lowest, | |
3696 | in order to be agnostic to the 32 vs 64 bits issue. | |
3697 | ||
3698 | returns 0 on failure, 1 if we successfully received it. */ | |
d8763023 AG |
3699 | static int receive_bitmap(struct drbd_conf *mdev, enum drbd_packet cmd, |
3700 | unsigned int data_size) | |
b411b363 PR |
3701 | { |
3702 | struct bm_xfer_ctx c; | |
3703 | void *buffer; | |
2c46407d | 3704 | int err; |
81e84650 | 3705 | int ok = false; |
257d0af6 | 3706 | struct p_header *h = &mdev->tconn->data.rbuf.header; |
77351055 | 3707 | struct packet_info pi; |
b411b363 | 3708 | |
20ceb2b2 LE |
3709 | drbd_bm_lock(mdev, "receive bitmap", BM_LOCKED_SET_ALLOWED); |
3710 | /* you are supposed to send additional out-of-sync information | |
3711 | * if you actually set bits during this phase */ | |
b411b363 PR |
3712 | |
3713 | /* maybe we should use some per thread scratch page, | |
3714 | * and allocate that during initial device creation? */ | |
3715 | buffer = (unsigned long *) __get_free_page(GFP_NOIO); | |
3716 | if (!buffer) { | |
3717 | dev_err(DEV, "failed to allocate one page buffer in %s\n", __func__); | |
3718 | goto out; | |
3719 | } | |
3720 | ||
3721 | c = (struct bm_xfer_ctx) { | |
3722 | .bm_bits = drbd_bm_bits(mdev), | |
3723 | .bm_words = drbd_bm_words(mdev), | |
3724 | }; | |
3725 | ||
2c46407d | 3726 | for(;;) { |
02918be2 | 3727 | if (cmd == P_BITMAP) { |
2c46407d | 3728 | err = receive_bitmap_plain(mdev, data_size, buffer, &c); |
02918be2 | 3729 | } else if (cmd == P_COMPRESSED_BITMAP) { |
b411b363 PR |
3730 | /* MAYBE: sanity check that we speak proto >= 90, |
3731 | * and the feature is enabled! */ | |
3732 | struct p_compressed_bm *p; | |
3733 | ||
02918be2 | 3734 | if (data_size > BM_PACKET_PAYLOAD_BYTES) { |
b411b363 PR |
3735 | dev_err(DEV, "ReportCBitmap packet too large\n"); |
3736 | goto out; | |
3737 | } | |
3738 | /* use the page buff */ | |
3739 | p = buffer; | |
3740 | memcpy(p, h, sizeof(*h)); | |
de0ff338 | 3741 | if (drbd_recv(mdev->tconn, p->head.payload, data_size) != data_size) |
b411b363 | 3742 | goto out; |
004352fa LE |
3743 | if (data_size <= (sizeof(*p) - sizeof(p->head))) { |
3744 | dev_err(DEV, "ReportCBitmap packet too small (l:%u)\n", data_size); | |
78fcbdae | 3745 | goto out; |
b411b363 | 3746 | } |
c6d25cfe | 3747 | err = decode_bitmap_c(mdev, p, &c, data_size); |
b411b363 | 3748 | } else { |
02918be2 | 3749 | dev_warn(DEV, "receive_bitmap: cmd neither ReportBitMap nor ReportCBitMap (is 0x%x)", cmd); |
b411b363 PR |
3750 | goto out; |
3751 | } | |
3752 | ||
02918be2 | 3753 | c.packets[cmd == P_BITMAP]++; |
257d0af6 | 3754 | c.bytes[cmd == P_BITMAP] += sizeof(struct p_header) + data_size; |
b411b363 | 3755 | |
2c46407d AG |
3756 | if (err <= 0) { |
3757 | if (err < 0) | |
3758 | goto out; | |
b411b363 | 3759 | break; |
2c46407d | 3760 | } |
9ba7aa00 | 3761 | if (!drbd_recv_header(mdev->tconn, &pi)) |
b411b363 | 3762 | goto out; |
77351055 PR |
3763 | cmd = pi.cmd; |
3764 | data_size = pi.size; | |
2c46407d | 3765 | } |
b411b363 PR |
3766 | |
3767 | INFO_bm_xfer_stats(mdev, "receive", &c); | |
3768 | ||
3769 | if (mdev->state.conn == C_WF_BITMAP_T) { | |
de1f8e4a AG |
3770 | enum drbd_state_rv rv; |
3771 | ||
b411b363 PR |
3772 | ok = !drbd_send_bitmap(mdev); |
3773 | if (!ok) | |
3774 | goto out; | |
3775 | /* Omit CS_ORDERED with this state transition to avoid deadlocks. */ | |
de1f8e4a AG |
3776 | rv = _drbd_request_state(mdev, NS(conn, C_WF_SYNC_UUID), CS_VERBOSE); |
3777 | D_ASSERT(rv == SS_SUCCESS); | |
b411b363 PR |
3778 | } else if (mdev->state.conn != C_WF_BITMAP_S) { |
3779 | /* admin may have requested C_DISCONNECTING, | |
3780 | * other threads may have noticed network errors */ | |
3781 | dev_info(DEV, "unexpected cstate (%s) in receive_bitmap\n", | |
3782 | drbd_conn_str(mdev->state.conn)); | |
3783 | } | |
3784 | ||
81e84650 | 3785 | ok = true; |
b411b363 | 3786 | out: |
20ceb2b2 | 3787 | drbd_bm_unlock(mdev); |
b411b363 PR |
3788 | if (ok && mdev->state.conn == C_WF_BITMAP_S) |
3789 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
3790 | free_page((unsigned long) buffer); | |
3791 | return ok; | |
3792 | } | |
3793 | ||
d8763023 AG |
3794 | static int receive_skip(struct drbd_conf *mdev, enum drbd_packet cmd, |
3795 | unsigned int data_size) | |
b411b363 PR |
3796 | { |
3797 | /* TODO zero copy sink :) */ | |
3798 | static char sink[128]; | |
3799 | int size, want, r; | |
3800 | ||
02918be2 PR |
3801 | dev_warn(DEV, "skipping unknown optional packet type %d, l: %d!\n", |
3802 | cmd, data_size); | |
b411b363 | 3803 | |
02918be2 | 3804 | size = data_size; |
b411b363 PR |
3805 | while (size > 0) { |
3806 | want = min_t(int, size, sizeof(sink)); | |
de0ff338 | 3807 | r = drbd_recv(mdev->tconn, sink, want); |
841ce241 AG |
3808 | if (!expect(r > 0)) |
3809 | break; | |
b411b363 PR |
3810 | size -= r; |
3811 | } | |
3812 | return size == 0; | |
3813 | } | |
3814 | ||
d8763023 AG |
3815 | static int receive_UnplugRemote(struct drbd_conf *mdev, enum drbd_packet cmd, |
3816 | unsigned int data_size) | |
0ced55a3 | 3817 | { |
e7f52dfb LE |
3818 | /* Make sure we've acked all the TCP data associated |
3819 | * with the data requests being unplugged */ | |
e42325a5 | 3820 | drbd_tcp_quickack(mdev->tconn->data.socket); |
0ced55a3 | 3821 | |
81e84650 | 3822 | return true; |
0ced55a3 PR |
3823 | } |
3824 | ||
d8763023 AG |
3825 | static int receive_out_of_sync(struct drbd_conf *mdev, enum drbd_packet cmd, |
3826 | unsigned int data_size) | |
73a01a18 | 3827 | { |
e42325a5 | 3828 | struct p_block_desc *p = &mdev->tconn->data.rbuf.block_desc; |
73a01a18 | 3829 | |
f735e363 LE |
3830 | switch (mdev->state.conn) { |
3831 | case C_WF_SYNC_UUID: | |
3832 | case C_WF_BITMAP_T: | |
3833 | case C_BEHIND: | |
3834 | break; | |
3835 | default: | |
3836 | dev_err(DEV, "ASSERT FAILED cstate = %s, expected: WFSyncUUID|WFBitMapT|Behind\n", | |
3837 | drbd_conn_str(mdev->state.conn)); | |
3838 | } | |
3839 | ||
73a01a18 PR |
3840 | drbd_set_out_of_sync(mdev, be64_to_cpu(p->sector), be32_to_cpu(p->blksize)); |
3841 | ||
81e84650 | 3842 | return true; |
73a01a18 PR |
3843 | } |
3844 | ||
d8763023 AG |
3845 | typedef int (*drbd_cmd_handler_f)(struct drbd_conf *, enum drbd_packet cmd, |
3846 | unsigned int to_receive); | |
02918be2 PR |
3847 | |
3848 | struct data_cmd { | |
3849 | int expect_payload; | |
3850 | size_t pkt_size; | |
3851 | drbd_cmd_handler_f function; | |
3852 | }; | |
3853 | ||
3854 | static struct data_cmd drbd_cmd_handler[] = { | |
3855 | [P_DATA] = { 1, sizeof(struct p_data), receive_Data }, | |
3856 | [P_DATA_REPLY] = { 1, sizeof(struct p_data), receive_DataReply }, | |
3857 | [P_RS_DATA_REPLY] = { 1, sizeof(struct p_data), receive_RSDataReply } , | |
3858 | [P_BARRIER] = { 0, sizeof(struct p_barrier), receive_Barrier } , | |
257d0af6 PR |
3859 | [P_BITMAP] = { 1, sizeof(struct p_header), receive_bitmap } , |
3860 | [P_COMPRESSED_BITMAP] = { 1, sizeof(struct p_header), receive_bitmap } , | |
3861 | [P_UNPLUG_REMOTE] = { 0, sizeof(struct p_header), receive_UnplugRemote }, | |
02918be2 PR |
3862 | [P_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, |
3863 | [P_RS_DATA_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
257d0af6 PR |
3864 | [P_SYNC_PARAM] = { 1, sizeof(struct p_header), receive_SyncParam }, |
3865 | [P_SYNC_PARAM89] = { 1, sizeof(struct p_header), receive_SyncParam }, | |
02918be2 PR |
3866 | [P_PROTOCOL] = { 1, sizeof(struct p_protocol), receive_protocol }, |
3867 | [P_UUIDS] = { 0, sizeof(struct p_uuids), receive_uuids }, | |
3868 | [P_SIZES] = { 0, sizeof(struct p_sizes), receive_sizes }, | |
3869 | [P_STATE] = { 0, sizeof(struct p_state), receive_state }, | |
3870 | [P_STATE_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, | |
3871 | [P_SYNC_UUID] = { 0, sizeof(struct p_rs_uuid), receive_sync_uuid }, | |
3872 | [P_OV_REQUEST] = { 0, sizeof(struct p_block_req), receive_DataRequest }, | |
3873 | [P_OV_REPLY] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3874 | [P_CSUM_RS_REQUEST] = { 1, sizeof(struct p_block_req), receive_DataRequest }, | |
3875 | [P_DELAY_PROBE] = { 0, sizeof(struct p_delay_probe93), receive_skip }, | |
73a01a18 | 3876 | [P_OUT_OF_SYNC] = { 0, sizeof(struct p_block_desc), receive_out_of_sync }, |
047cd4a6 | 3877 | [P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_state }, |
b411b363 PR |
3878 | }; |
3879 | ||
02918be2 | 3880 | /* All handler functions that expect a sub-header get that sub-heder in |
e42325a5 | 3881 | mdev->tconn->data.rbuf.header.head.payload. |
02918be2 | 3882 | |
e42325a5 | 3883 | Usually in mdev->tconn->data.rbuf.header.head the callback can find the usual |
02918be2 PR |
3884 | p_header, but they may not rely on that. Since there is also p_header95 ! |
3885 | */ | |
b411b363 | 3886 | |
eefc2f7d | 3887 | static void drbdd(struct drbd_tconn *tconn) |
b411b363 | 3888 | { |
eefc2f7d | 3889 | struct p_header *header = &tconn->data.rbuf.header; |
77351055 | 3890 | struct packet_info pi; |
02918be2 PR |
3891 | size_t shs; /* sub header size */ |
3892 | int rv; | |
b411b363 | 3893 | |
eefc2f7d PR |
3894 | while (get_t_state(&tconn->receiver) == RUNNING) { |
3895 | drbd_thread_current_set_cpu(&tconn->receiver); | |
3896 | if (!drbd_recv_header(tconn, &pi)) | |
02918be2 | 3897 | goto err_out; |
b411b363 | 3898 | |
6e849ce8 AG |
3899 | if (unlikely(pi.cmd >= ARRAY_SIZE(drbd_cmd_handler) || |
3900 | !drbd_cmd_handler[pi.cmd].function)) { | |
eefc2f7d | 3901 | conn_err(tconn, "unknown packet type %d, l: %d!\n", pi.cmd, pi.size); |
02918be2 | 3902 | goto err_out; |
0b33a916 | 3903 | } |
b411b363 | 3904 | |
77351055 PR |
3905 | shs = drbd_cmd_handler[pi.cmd].pkt_size - sizeof(struct p_header); |
3906 | if (pi.size - shs > 0 && !drbd_cmd_handler[pi.cmd].expect_payload) { | |
eefc2f7d | 3907 | conn_err(tconn, "No payload expected %s l:%d\n", cmdname(pi.cmd), pi.size); |
02918be2 | 3908 | goto err_out; |
b411b363 | 3909 | } |
b411b363 | 3910 | |
c13f7e1a | 3911 | if (shs) { |
eefc2f7d | 3912 | rv = drbd_recv(tconn, &header->payload, shs); |
c13f7e1a | 3913 | if (unlikely(rv != shs)) { |
0ddc5549 | 3914 | if (!signal_pending(current)) |
eefc2f7d | 3915 | conn_warn(tconn, "short read while reading sub header: rv=%d\n", rv); |
c13f7e1a LE |
3916 | goto err_out; |
3917 | } | |
3918 | } | |
3919 | ||
eefc2f7d | 3920 | rv = drbd_cmd_handler[pi.cmd].function(vnr_to_mdev(tconn, pi.vnr), pi.cmd, pi.size - shs); |
b411b363 | 3921 | |
02918be2 | 3922 | if (unlikely(!rv)) { |
eefc2f7d | 3923 | conn_err(tconn, "error receiving %s, l: %d!\n", |
77351055 | 3924 | cmdname(pi.cmd), pi.size); |
02918be2 | 3925 | goto err_out; |
b411b363 PR |
3926 | } |
3927 | } | |
b411b363 | 3928 | |
02918be2 PR |
3929 | if (0) { |
3930 | err_out: | |
bbeb641c | 3931 | conn_request_state(tconn, NS(conn, C_PROTOCOL_ERROR), CS_HARD); |
02918be2 | 3932 | } |
b411b363 PR |
3933 | } |
3934 | ||
0e29d163 | 3935 | void conn_flush_workqueue(struct drbd_tconn *tconn) |
b411b363 PR |
3936 | { |
3937 | struct drbd_wq_barrier barr; | |
3938 | ||
3939 | barr.w.cb = w_prev_work_done; | |
0e29d163 | 3940 | barr.w.tconn = tconn; |
b411b363 | 3941 | init_completion(&barr.done); |
0e29d163 | 3942 | drbd_queue_work(&tconn->data.work, &barr.w); |
b411b363 PR |
3943 | wait_for_completion(&barr.done); |
3944 | } | |
3945 | ||
360cc740 | 3946 | static void drbd_disconnect(struct drbd_tconn *tconn) |
b411b363 | 3947 | { |
bbeb641c | 3948 | enum drbd_conns oc; |
b411b363 | 3949 | int rv = SS_UNKNOWN_ERROR; |
b411b363 | 3950 | |
bbeb641c | 3951 | if (tconn->cstate == C_STANDALONE) |
b411b363 | 3952 | return; |
b411b363 PR |
3953 | |
3954 | /* asender does not clean up anything. it must not interfere, either */ | |
360cc740 PR |
3955 | drbd_thread_stop(&tconn->asender); |
3956 | drbd_free_sock(tconn); | |
3957 | ||
3958 | idr_for_each(&tconn->volumes, drbd_disconnected, tconn); | |
3959 | ||
3960 | conn_info(tconn, "Connection closed\n"); | |
3961 | ||
3962 | spin_lock_irq(&tconn->req_lock); | |
bbeb641c PR |
3963 | oc = tconn->cstate; |
3964 | if (oc >= C_UNCONNECTED) | |
3965 | rv = _conn_request_state(tconn, NS(conn, C_UNCONNECTED), CS_VERBOSE); | |
3966 | ||
360cc740 PR |
3967 | spin_unlock_irq(&tconn->req_lock); |
3968 | ||
bbeb641c | 3969 | if (oc == C_DISCONNECTING) { |
360cc740 PR |
3970 | wait_event(tconn->net_cnt_wait, atomic_read(&tconn->net_cnt) == 0); |
3971 | ||
3972 | crypto_free_hash(tconn->cram_hmac_tfm); | |
3973 | tconn->cram_hmac_tfm = NULL; | |
3974 | ||
3975 | kfree(tconn->net_conf); | |
3976 | tconn->net_conf = NULL; | |
bbeb641c | 3977 | conn_request_state(tconn, NS(conn, C_STANDALONE), CS_VERBOSE); |
360cc740 PR |
3978 | } |
3979 | } | |
3980 | ||
3981 | static int drbd_disconnected(int vnr, void *p, void *data) | |
3982 | { | |
3983 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
3984 | enum drbd_fencing_p fp; | |
3985 | unsigned int i; | |
b411b363 | 3986 | |
85719573 | 3987 | /* wait for current activity to cease. */ |
87eeee41 | 3988 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3989 | _drbd_wait_ee_list_empty(mdev, &mdev->active_ee); |
3990 | _drbd_wait_ee_list_empty(mdev, &mdev->sync_ee); | |
3991 | _drbd_wait_ee_list_empty(mdev, &mdev->read_ee); | |
87eeee41 | 3992 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
3993 | |
3994 | /* We do not have data structures that would allow us to | |
3995 | * get the rs_pending_cnt down to 0 again. | |
3996 | * * On C_SYNC_TARGET we do not have any data structures describing | |
3997 | * the pending RSDataRequest's we have sent. | |
3998 | * * On C_SYNC_SOURCE there is no data structure that tracks | |
3999 | * the P_RS_DATA_REPLY blocks that we sent to the SyncTarget. | |
4000 | * And no, it is not the sum of the reference counts in the | |
4001 | * resync_LRU. The resync_LRU tracks the whole operation including | |
4002 | * the disk-IO, while the rs_pending_cnt only tracks the blocks | |
4003 | * on the fly. */ | |
4004 | drbd_rs_cancel_all(mdev); | |
4005 | mdev->rs_total = 0; | |
4006 | mdev->rs_failed = 0; | |
4007 | atomic_set(&mdev->rs_pending_cnt, 0); | |
4008 | wake_up(&mdev->misc_wait); | |
4009 | ||
7fde2be9 PR |
4010 | del_timer(&mdev->request_timer); |
4011 | ||
b411b363 | 4012 | del_timer_sync(&mdev->resync_timer); |
b411b363 PR |
4013 | resync_timer_fn((unsigned long)mdev); |
4014 | ||
b411b363 PR |
4015 | /* wait for all w_e_end_data_req, w_e_end_rsdata_req, w_send_barrier, |
4016 | * w_make_resync_request etc. which may still be on the worker queue | |
4017 | * to be "canceled" */ | |
a21e9298 | 4018 | drbd_flush_workqueue(mdev); |
b411b363 PR |
4019 | |
4020 | /* This also does reclaim_net_ee(). If we do this too early, we might | |
4021 | * miss some resync ee and pages.*/ | |
4022 | drbd_process_done_ee(mdev); | |
4023 | ||
4024 | kfree(mdev->p_uuid); | |
4025 | mdev->p_uuid = NULL; | |
4026 | ||
fb22c402 | 4027 | if (!is_susp(mdev->state)) |
b411b363 PR |
4028 | tl_clear(mdev); |
4029 | ||
b411b363 PR |
4030 | drbd_md_sync(mdev); |
4031 | ||
4032 | fp = FP_DONT_CARE; | |
4033 | if (get_ldev(mdev)) { | |
4034 | fp = mdev->ldev->dc.fencing; | |
4035 | put_ldev(mdev); | |
4036 | } | |
4037 | ||
87f7be4c PR |
4038 | if (mdev->state.role == R_PRIMARY && fp >= FP_RESOURCE && mdev->state.pdsk >= D_UNKNOWN) |
4039 | drbd_try_outdate_peer_async(mdev); | |
b411b363 | 4040 | |
20ceb2b2 LE |
4041 | /* serialize with bitmap writeout triggered by the state change, |
4042 | * if any. */ | |
4043 | wait_event(mdev->misc_wait, !test_bit(BITMAP_IO, &mdev->flags)); | |
4044 | ||
b411b363 PR |
4045 | /* tcp_close and release of sendpage pages can be deferred. I don't |
4046 | * want to use SO_LINGER, because apparently it can be deferred for | |
4047 | * more than 20 seconds (longest time I checked). | |
4048 | * | |
4049 | * Actually we don't care for exactly when the network stack does its | |
4050 | * put_page(), but release our reference on these pages right here. | |
4051 | */ | |
4052 | i = drbd_release_ee(mdev, &mdev->net_ee); | |
4053 | if (i) | |
4054 | dev_info(DEV, "net_ee not empty, killed %u entries\n", i); | |
435f0740 LE |
4055 | i = atomic_read(&mdev->pp_in_use_by_net); |
4056 | if (i) | |
4057 | dev_info(DEV, "pp_in_use_by_net = %d, expected 0\n", i); | |
b411b363 PR |
4058 | i = atomic_read(&mdev->pp_in_use); |
4059 | if (i) | |
45bb912b | 4060 | dev_info(DEV, "pp_in_use = %d, expected 0\n", i); |
b411b363 PR |
4061 | |
4062 | D_ASSERT(list_empty(&mdev->read_ee)); | |
4063 | D_ASSERT(list_empty(&mdev->active_ee)); | |
4064 | D_ASSERT(list_empty(&mdev->sync_ee)); | |
4065 | D_ASSERT(list_empty(&mdev->done_ee)); | |
4066 | ||
4067 | /* ok, no more ee's on the fly, it is safe to reset the epoch_size */ | |
4068 | atomic_set(&mdev->current_epoch->epoch_size, 0); | |
4069 | D_ASSERT(list_empty(&mdev->current_epoch->list)); | |
360cc740 PR |
4070 | |
4071 | return 0; | |
b411b363 PR |
4072 | } |
4073 | ||
4074 | /* | |
4075 | * We support PRO_VERSION_MIN to PRO_VERSION_MAX. The protocol version | |
4076 | * we can agree on is stored in agreed_pro_version. | |
4077 | * | |
4078 | * feature flags and the reserved array should be enough room for future | |
4079 | * enhancements of the handshake protocol, and possible plugins... | |
4080 | * | |
4081 | * for now, they are expected to be zero, but ignored. | |
4082 | */ | |
8a22cccc | 4083 | static int drbd_send_handshake(struct drbd_tconn *tconn) |
b411b363 | 4084 | { |
e6b3ea83 | 4085 | /* ASSERT current == mdev->tconn->receiver ... */ |
8a22cccc | 4086 | struct p_handshake *p = &tconn->data.sbuf.handshake; |
b411b363 PR |
4087 | int ok; |
4088 | ||
8a22cccc PR |
4089 | if (mutex_lock_interruptible(&tconn->data.mutex)) { |
4090 | conn_err(tconn, "interrupted during initial handshake\n"); | |
b411b363 PR |
4091 | return 0; /* interrupted. not ok. */ |
4092 | } | |
4093 | ||
8a22cccc PR |
4094 | if (tconn->data.socket == NULL) { |
4095 | mutex_unlock(&tconn->data.mutex); | |
b411b363 PR |
4096 | return 0; |
4097 | } | |
4098 | ||
4099 | memset(p, 0, sizeof(*p)); | |
4100 | p->protocol_min = cpu_to_be32(PRO_VERSION_MIN); | |
4101 | p->protocol_max = cpu_to_be32(PRO_VERSION_MAX); | |
8a22cccc PR |
4102 | ok = _conn_send_cmd(tconn, 0, tconn->data.socket, P_HAND_SHAKE, |
4103 | &p->head, sizeof(*p), 0); | |
4104 | mutex_unlock(&tconn->data.mutex); | |
b411b363 PR |
4105 | return ok; |
4106 | } | |
4107 | ||
4108 | /* | |
4109 | * return values: | |
4110 | * 1 yes, we have a valid connection | |
4111 | * 0 oops, did not work out, please try again | |
4112 | * -1 peer talks different language, | |
4113 | * no point in trying again, please go standalone. | |
4114 | */ | |
65d11ed6 | 4115 | static int drbd_do_handshake(struct drbd_tconn *tconn) |
b411b363 | 4116 | { |
65d11ed6 PR |
4117 | /* ASSERT current == tconn->receiver ... */ |
4118 | struct p_handshake *p = &tconn->data.rbuf.handshake; | |
02918be2 | 4119 | const int expect = sizeof(struct p_handshake) - sizeof(struct p_header80); |
77351055 | 4120 | struct packet_info pi; |
b411b363 PR |
4121 | int rv; |
4122 | ||
65d11ed6 | 4123 | rv = drbd_send_handshake(tconn); |
b411b363 PR |
4124 | if (!rv) |
4125 | return 0; | |
4126 | ||
65d11ed6 | 4127 | rv = drbd_recv_header(tconn, &pi); |
b411b363 PR |
4128 | if (!rv) |
4129 | return 0; | |
4130 | ||
77351055 | 4131 | if (pi.cmd != P_HAND_SHAKE) { |
65d11ed6 | 4132 | conn_err(tconn, "expected HandShake packet, received: %s (0x%04x)\n", |
77351055 | 4133 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4134 | return -1; |
4135 | } | |
4136 | ||
77351055 | 4137 | if (pi.size != expect) { |
65d11ed6 | 4138 | conn_err(tconn, "expected HandShake length: %u, received: %u\n", |
77351055 | 4139 | expect, pi.size); |
b411b363 PR |
4140 | return -1; |
4141 | } | |
4142 | ||
65d11ed6 | 4143 | rv = drbd_recv(tconn, &p->head.payload, expect); |
b411b363 PR |
4144 | |
4145 | if (rv != expect) { | |
0ddc5549 | 4146 | if (!signal_pending(current)) |
65d11ed6 | 4147 | conn_warn(tconn, "short read receiving handshake packet: l=%u\n", rv); |
b411b363 PR |
4148 | return 0; |
4149 | } | |
4150 | ||
b411b363 PR |
4151 | p->protocol_min = be32_to_cpu(p->protocol_min); |
4152 | p->protocol_max = be32_to_cpu(p->protocol_max); | |
4153 | if (p->protocol_max == 0) | |
4154 | p->protocol_max = p->protocol_min; | |
4155 | ||
4156 | if (PRO_VERSION_MAX < p->protocol_min || | |
4157 | PRO_VERSION_MIN > p->protocol_max) | |
4158 | goto incompat; | |
4159 | ||
65d11ed6 | 4160 | tconn->agreed_pro_version = min_t(int, PRO_VERSION_MAX, p->protocol_max); |
b411b363 | 4161 | |
65d11ed6 PR |
4162 | conn_info(tconn, "Handshake successful: " |
4163 | "Agreed network protocol version %d\n", tconn->agreed_pro_version); | |
b411b363 PR |
4164 | |
4165 | return 1; | |
4166 | ||
4167 | incompat: | |
65d11ed6 | 4168 | conn_err(tconn, "incompatible DRBD dialects: " |
b411b363 PR |
4169 | "I support %d-%d, peer supports %d-%d\n", |
4170 | PRO_VERSION_MIN, PRO_VERSION_MAX, | |
4171 | p->protocol_min, p->protocol_max); | |
4172 | return -1; | |
4173 | } | |
4174 | ||
4175 | #if !defined(CONFIG_CRYPTO_HMAC) && !defined(CONFIG_CRYPTO_HMAC_MODULE) | |
13e6037d | 4176 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 PR |
4177 | { |
4178 | dev_err(DEV, "This kernel was build without CONFIG_CRYPTO_HMAC.\n"); | |
4179 | dev_err(DEV, "You need to disable 'cram-hmac-alg' in drbd.conf.\n"); | |
b10d96cb | 4180 | return -1; |
b411b363 PR |
4181 | } |
4182 | #else | |
4183 | #define CHALLENGE_LEN 64 | |
b10d96cb JT |
4184 | |
4185 | /* Return value: | |
4186 | 1 - auth succeeded, | |
4187 | 0 - failed, try again (network error), | |
4188 | -1 - auth failed, don't try again. | |
4189 | */ | |
4190 | ||
13e6037d | 4191 | static int drbd_do_auth(struct drbd_tconn *tconn) |
b411b363 PR |
4192 | { |
4193 | char my_challenge[CHALLENGE_LEN]; /* 64 Bytes... */ | |
4194 | struct scatterlist sg; | |
4195 | char *response = NULL; | |
4196 | char *right_response = NULL; | |
4197 | char *peers_ch = NULL; | |
13e6037d | 4198 | unsigned int key_len = strlen(tconn->net_conf->shared_secret); |
b411b363 PR |
4199 | unsigned int resp_size; |
4200 | struct hash_desc desc; | |
77351055 | 4201 | struct packet_info pi; |
b411b363 PR |
4202 | int rv; |
4203 | ||
13e6037d | 4204 | desc.tfm = tconn->cram_hmac_tfm; |
b411b363 PR |
4205 | desc.flags = 0; |
4206 | ||
13e6037d PR |
4207 | rv = crypto_hash_setkey(tconn->cram_hmac_tfm, |
4208 | (u8 *)tconn->net_conf->shared_secret, key_len); | |
b411b363 | 4209 | if (rv) { |
13e6037d | 4210 | conn_err(tconn, "crypto_hash_setkey() failed with %d\n", rv); |
b10d96cb | 4211 | rv = -1; |
b411b363 PR |
4212 | goto fail; |
4213 | } | |
4214 | ||
4215 | get_random_bytes(my_challenge, CHALLENGE_LEN); | |
4216 | ||
13e6037d | 4217 | rv = conn_send_cmd2(tconn, P_AUTH_CHALLENGE, my_challenge, CHALLENGE_LEN); |
b411b363 PR |
4218 | if (!rv) |
4219 | goto fail; | |
4220 | ||
13e6037d | 4221 | rv = drbd_recv_header(tconn, &pi); |
b411b363 PR |
4222 | if (!rv) |
4223 | goto fail; | |
4224 | ||
77351055 | 4225 | if (pi.cmd != P_AUTH_CHALLENGE) { |
13e6037d | 4226 | conn_err(tconn, "expected AuthChallenge packet, received: %s (0x%04x)\n", |
77351055 | 4227 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4228 | rv = 0; |
4229 | goto fail; | |
4230 | } | |
4231 | ||
77351055 | 4232 | if (pi.size > CHALLENGE_LEN * 2) { |
13e6037d | 4233 | conn_err(tconn, "expected AuthChallenge payload too big.\n"); |
b10d96cb | 4234 | rv = -1; |
b411b363 PR |
4235 | goto fail; |
4236 | } | |
4237 | ||
77351055 | 4238 | peers_ch = kmalloc(pi.size, GFP_NOIO); |
b411b363 | 4239 | if (peers_ch == NULL) { |
13e6037d | 4240 | conn_err(tconn, "kmalloc of peers_ch failed\n"); |
b10d96cb | 4241 | rv = -1; |
b411b363 PR |
4242 | goto fail; |
4243 | } | |
4244 | ||
13e6037d | 4245 | rv = drbd_recv(tconn, peers_ch, pi.size); |
b411b363 | 4246 | |
77351055 | 4247 | if (rv != pi.size) { |
0ddc5549 | 4248 | if (!signal_pending(current)) |
13e6037d | 4249 | conn_warn(tconn, "short read AuthChallenge: l=%u\n", rv); |
b411b363 PR |
4250 | rv = 0; |
4251 | goto fail; | |
4252 | } | |
4253 | ||
13e6037d | 4254 | resp_size = crypto_hash_digestsize(tconn->cram_hmac_tfm); |
b411b363 PR |
4255 | response = kmalloc(resp_size, GFP_NOIO); |
4256 | if (response == NULL) { | |
13e6037d | 4257 | conn_err(tconn, "kmalloc of response failed\n"); |
b10d96cb | 4258 | rv = -1; |
b411b363 PR |
4259 | goto fail; |
4260 | } | |
4261 | ||
4262 | sg_init_table(&sg, 1); | |
77351055 | 4263 | sg_set_buf(&sg, peers_ch, pi.size); |
b411b363 PR |
4264 | |
4265 | rv = crypto_hash_digest(&desc, &sg, sg.length, response); | |
4266 | if (rv) { | |
13e6037d | 4267 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4268 | rv = -1; |
b411b363 PR |
4269 | goto fail; |
4270 | } | |
4271 | ||
13e6037d | 4272 | rv = conn_send_cmd2(tconn, P_AUTH_RESPONSE, response, resp_size); |
b411b363 PR |
4273 | if (!rv) |
4274 | goto fail; | |
4275 | ||
13e6037d | 4276 | rv = drbd_recv_header(tconn, &pi); |
b411b363 PR |
4277 | if (!rv) |
4278 | goto fail; | |
4279 | ||
77351055 | 4280 | if (pi.cmd != P_AUTH_RESPONSE) { |
13e6037d | 4281 | conn_err(tconn, "expected AuthResponse packet, received: %s (0x%04x)\n", |
77351055 | 4282 | cmdname(pi.cmd), pi.cmd); |
b411b363 PR |
4283 | rv = 0; |
4284 | goto fail; | |
4285 | } | |
4286 | ||
77351055 | 4287 | if (pi.size != resp_size) { |
13e6037d | 4288 | conn_err(tconn, "expected AuthResponse payload of wrong size\n"); |
b411b363 PR |
4289 | rv = 0; |
4290 | goto fail; | |
4291 | } | |
4292 | ||
13e6037d | 4293 | rv = drbd_recv(tconn, response , resp_size); |
b411b363 PR |
4294 | |
4295 | if (rv != resp_size) { | |
0ddc5549 | 4296 | if (!signal_pending(current)) |
13e6037d | 4297 | conn_warn(tconn, "short read receiving AuthResponse: l=%u\n", rv); |
b411b363 PR |
4298 | rv = 0; |
4299 | goto fail; | |
4300 | } | |
4301 | ||
4302 | right_response = kmalloc(resp_size, GFP_NOIO); | |
2d1ee87d | 4303 | if (right_response == NULL) { |
13e6037d | 4304 | conn_err(tconn, "kmalloc of right_response failed\n"); |
b10d96cb | 4305 | rv = -1; |
b411b363 PR |
4306 | goto fail; |
4307 | } | |
4308 | ||
4309 | sg_set_buf(&sg, my_challenge, CHALLENGE_LEN); | |
4310 | ||
4311 | rv = crypto_hash_digest(&desc, &sg, sg.length, right_response); | |
4312 | if (rv) { | |
13e6037d | 4313 | conn_err(tconn, "crypto_hash_digest() failed with %d\n", rv); |
b10d96cb | 4314 | rv = -1; |
b411b363 PR |
4315 | goto fail; |
4316 | } | |
4317 | ||
4318 | rv = !memcmp(response, right_response, resp_size); | |
4319 | ||
4320 | if (rv) | |
13e6037d PR |
4321 | conn_info(tconn, "Peer authenticated using %d bytes of '%s' HMAC\n", |
4322 | resp_size, tconn->net_conf->cram_hmac_alg); | |
b10d96cb JT |
4323 | else |
4324 | rv = -1; | |
b411b363 PR |
4325 | |
4326 | fail: | |
4327 | kfree(peers_ch); | |
4328 | kfree(response); | |
4329 | kfree(right_response); | |
4330 | ||
4331 | return rv; | |
4332 | } | |
4333 | #endif | |
4334 | ||
4335 | int drbdd_init(struct drbd_thread *thi) | |
4336 | { | |
392c8801 | 4337 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 PR |
4338 | int h; |
4339 | ||
4d641dd7 | 4340 | conn_info(tconn, "receiver (re)started\n"); |
b411b363 PR |
4341 | |
4342 | do { | |
4d641dd7 | 4343 | h = drbd_connect(tconn); |
b411b363 | 4344 | if (h == 0) { |
4d641dd7 | 4345 | drbd_disconnect(tconn); |
20ee6390 | 4346 | schedule_timeout_interruptible(HZ); |
b411b363 PR |
4347 | } |
4348 | if (h == -1) { | |
4d641dd7 | 4349 | conn_warn(tconn, "Discarding network configuration.\n"); |
bbeb641c | 4350 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 PR |
4351 | } |
4352 | } while (h == 0); | |
4353 | ||
4354 | if (h > 0) { | |
4d641dd7 PR |
4355 | if (get_net_conf(tconn)) { |
4356 | drbdd(tconn); | |
4357 | put_net_conf(tconn); | |
b411b363 PR |
4358 | } |
4359 | } | |
4360 | ||
4d641dd7 | 4361 | drbd_disconnect(tconn); |
b411b363 | 4362 | |
4d641dd7 | 4363 | conn_info(tconn, "receiver terminated\n"); |
b411b363 PR |
4364 | return 0; |
4365 | } | |
4366 | ||
4367 | /* ********* acknowledge sender ******** */ | |
4368 | ||
d8763023 | 4369 | static int got_RqSReply(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4370 | { |
257d0af6 | 4371 | struct p_req_state_reply *p = &mdev->tconn->meta.rbuf.req_state_reply; |
fc3b10a4 | 4372 | struct drbd_tconn *tconn = mdev->tconn; |
b411b363 PR |
4373 | |
4374 | int retcode = be32_to_cpu(p->retcode); | |
4375 | ||
fc3b10a4 PR |
4376 | if (cmd == P_STATE_CHG_REPLY) { |
4377 | if (retcode >= SS_SUCCESS) { | |
4378 | set_bit(CL_ST_CHG_SUCCESS, &mdev->flags); | |
4379 | } else { | |
4380 | set_bit(CL_ST_CHG_FAIL, &mdev->flags); | |
4381 | dev_err(DEV, "Requested state change failed by peer: %s (%d)\n", | |
4382 | drbd_set_st_err_str(retcode), retcode); | |
4383 | } | |
4384 | wake_up(&mdev->state_wait); | |
4385 | } else /* conn == P_CONN_ST_CHG_REPLY */ { | |
4386 | if (retcode >= SS_SUCCESS) { | |
4387 | set_bit(CONN_WD_ST_CHG_OKAY, &tconn->flags); | |
4388 | } else { | |
4389 | set_bit(CONN_WD_ST_CHG_FAIL, &tconn->flags); | |
4390 | conn_err(tconn, "Requested state change failed by peer: %s (%d)\n", | |
4391 | drbd_set_st_err_str(retcode), retcode); | |
4392 | } | |
4393 | wake_up(&tconn->ping_wait); | |
b411b363 | 4394 | } |
81e84650 | 4395 | return true; |
b411b363 PR |
4396 | } |
4397 | ||
d8763023 | 4398 | static int got_Ping(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4399 | { |
2a67d8b9 | 4400 | return drbd_send_ping_ack(mdev->tconn); |
b411b363 PR |
4401 | |
4402 | } | |
4403 | ||
d8763023 | 4404 | static int got_PingAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4405 | { |
2a67d8b9 | 4406 | struct drbd_tconn *tconn = mdev->tconn; |
b411b363 | 4407 | /* restore idle timeout */ |
2a67d8b9 PR |
4408 | tconn->meta.socket->sk->sk_rcvtimeo = tconn->net_conf->ping_int*HZ; |
4409 | if (!test_and_set_bit(GOT_PING_ACK, &tconn->flags)) | |
4410 | wake_up(&tconn->ping_wait); | |
b411b363 | 4411 | |
81e84650 | 4412 | return true; |
b411b363 PR |
4413 | } |
4414 | ||
d8763023 | 4415 | static int got_IsInSync(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4416 | { |
257d0af6 | 4417 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4418 | sector_t sector = be64_to_cpu(p->sector); |
4419 | int blksize = be32_to_cpu(p->blksize); | |
4420 | ||
31890f4a | 4421 | D_ASSERT(mdev->tconn->agreed_pro_version >= 89); |
b411b363 PR |
4422 | |
4423 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4424 | ||
1d53f09e LE |
4425 | if (get_ldev(mdev)) { |
4426 | drbd_rs_complete_io(mdev, sector); | |
4427 | drbd_set_in_sync(mdev, sector, blksize); | |
4428 | /* rs_same_csums is supposed to count in units of BM_BLOCK_SIZE */ | |
4429 | mdev->rs_same_csum += (blksize >> BM_BLOCK_SHIFT); | |
4430 | put_ldev(mdev); | |
4431 | } | |
b411b363 | 4432 | dec_rs_pending(mdev); |
778f271d | 4433 | atomic_add(blksize >> 9, &mdev->rs_sect_in); |
b411b363 | 4434 | |
81e84650 | 4435 | return true; |
b411b363 PR |
4436 | } |
4437 | ||
bc9c5c41 AG |
4438 | static int |
4439 | validate_req_change_req_state(struct drbd_conf *mdev, u64 id, sector_t sector, | |
4440 | struct rb_root *root, const char *func, | |
4441 | enum drbd_req_event what, bool missing_ok) | |
b411b363 PR |
4442 | { |
4443 | struct drbd_request *req; | |
4444 | struct bio_and_error m; | |
4445 | ||
87eeee41 | 4446 | spin_lock_irq(&mdev->tconn->req_lock); |
bc9c5c41 | 4447 | req = find_request(mdev, root, id, sector, missing_ok, func); |
b411b363 | 4448 | if (unlikely(!req)) { |
87eeee41 | 4449 | spin_unlock_irq(&mdev->tconn->req_lock); |
81e84650 | 4450 | return false; |
b411b363 PR |
4451 | } |
4452 | __req_mod(req, what, &m); | |
87eeee41 | 4453 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
4454 | |
4455 | if (m.bio) | |
4456 | complete_master_bio(mdev, &m); | |
81e84650 | 4457 | return true; |
b411b363 PR |
4458 | } |
4459 | ||
d8763023 | 4460 | static int got_BlockAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4461 | { |
257d0af6 | 4462 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4463 | sector_t sector = be64_to_cpu(p->sector); |
4464 | int blksize = be32_to_cpu(p->blksize); | |
4465 | enum drbd_req_event what; | |
4466 | ||
4467 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4468 | ||
579b57ed | 4469 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4470 | drbd_set_in_sync(mdev, sector, blksize); |
4471 | dec_rs_pending(mdev); | |
81e84650 | 4472 | return true; |
b411b363 | 4473 | } |
257d0af6 | 4474 | switch (cmd) { |
b411b363 | 4475 | case P_RS_WRITE_ACK: |
89e58e75 | 4476 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
8554df1c | 4477 | what = WRITE_ACKED_BY_PEER_AND_SIS; |
b411b363 PR |
4478 | break; |
4479 | case P_WRITE_ACK: | |
89e58e75 | 4480 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
8554df1c | 4481 | what = WRITE_ACKED_BY_PEER; |
b411b363 PR |
4482 | break; |
4483 | case P_RECV_ACK: | |
89e58e75 | 4484 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_B); |
8554df1c | 4485 | what = RECV_ACKED_BY_PEER; |
b411b363 | 4486 | break; |
7be8da07 | 4487 | case P_DISCARD_WRITE: |
89e58e75 | 4488 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); |
7be8da07 AG |
4489 | what = DISCARD_WRITE; |
4490 | break; | |
4491 | case P_RETRY_WRITE: | |
4492 | D_ASSERT(mdev->tconn->net_conf->wire_protocol == DRBD_PROT_C); | |
4493 | what = POSTPONE_WRITE; | |
b411b363 PR |
4494 | break; |
4495 | default: | |
4496 | D_ASSERT(0); | |
81e84650 | 4497 | return false; |
b411b363 PR |
4498 | } |
4499 | ||
4500 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
bc9c5c41 AG |
4501 | &mdev->write_requests, __func__, |
4502 | what, false); | |
b411b363 PR |
4503 | } |
4504 | ||
d8763023 | 4505 | static int got_NegAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4506 | { |
257d0af6 | 4507 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 | 4508 | sector_t sector = be64_to_cpu(p->sector); |
2deb8336 | 4509 | int size = be32_to_cpu(p->blksize); |
89e58e75 PR |
4510 | bool missing_ok = mdev->tconn->net_conf->wire_protocol == DRBD_PROT_A || |
4511 | mdev->tconn->net_conf->wire_protocol == DRBD_PROT_B; | |
c3afd8f5 | 4512 | bool found; |
b411b363 PR |
4513 | |
4514 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4515 | ||
579b57ed | 4516 | if (p->block_id == ID_SYNCER) { |
b411b363 PR |
4517 | dec_rs_pending(mdev); |
4518 | drbd_rs_failed_io(mdev, sector, size); | |
81e84650 | 4519 | return true; |
b411b363 | 4520 | } |
2deb8336 | 4521 | |
c3afd8f5 | 4522 | found = validate_req_change_req_state(mdev, p->block_id, sector, |
bc9c5c41 | 4523 | &mdev->write_requests, __func__, |
8554df1c | 4524 | NEG_ACKED, missing_ok); |
c3afd8f5 AG |
4525 | if (!found) { |
4526 | /* Protocol A has no P_WRITE_ACKs, but has P_NEG_ACKs. | |
4527 | The master bio might already be completed, therefore the | |
4528 | request is no longer in the collision hash. */ | |
4529 | /* In Protocol B we might already have got a P_RECV_ACK | |
4530 | but then get a P_NEG_ACK afterwards. */ | |
4531 | if (!missing_ok) | |
2deb8336 | 4532 | return false; |
c3afd8f5 | 4533 | drbd_set_out_of_sync(mdev, sector, size); |
2deb8336 | 4534 | } |
2deb8336 | 4535 | return true; |
b411b363 PR |
4536 | } |
4537 | ||
d8763023 | 4538 | static int got_NegDReply(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4539 | { |
257d0af6 | 4540 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4541 | sector_t sector = be64_to_cpu(p->sector); |
4542 | ||
4543 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
7be8da07 | 4544 | |
b411b363 PR |
4545 | dev_err(DEV, "Got NegDReply; Sector %llus, len %u; Fail original request.\n", |
4546 | (unsigned long long)sector, be32_to_cpu(p->blksize)); | |
4547 | ||
4548 | return validate_req_change_req_state(mdev, p->block_id, sector, | |
bc9c5c41 | 4549 | &mdev->read_requests, __func__, |
8554df1c | 4550 | NEG_ACKED, false); |
b411b363 PR |
4551 | } |
4552 | ||
d8763023 | 4553 | static int got_NegRSDReply(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 PR |
4554 | { |
4555 | sector_t sector; | |
4556 | int size; | |
257d0af6 | 4557 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4558 | |
4559 | sector = be64_to_cpu(p->sector); | |
4560 | size = be32_to_cpu(p->blksize); | |
b411b363 PR |
4561 | |
4562 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4563 | ||
4564 | dec_rs_pending(mdev); | |
4565 | ||
4566 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
4567 | drbd_rs_complete_io(mdev, sector); | |
257d0af6 | 4568 | switch (cmd) { |
d612d309 PR |
4569 | case P_NEG_RS_DREPLY: |
4570 | drbd_rs_failed_io(mdev, sector, size); | |
4571 | case P_RS_CANCEL: | |
4572 | break; | |
4573 | default: | |
4574 | D_ASSERT(0); | |
4575 | put_ldev(mdev); | |
4576 | return false; | |
4577 | } | |
b411b363 PR |
4578 | put_ldev(mdev); |
4579 | } | |
4580 | ||
81e84650 | 4581 | return true; |
b411b363 PR |
4582 | } |
4583 | ||
d8763023 | 4584 | static int got_BarrierAck(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4585 | { |
257d0af6 | 4586 | struct p_barrier_ack *p = &mdev->tconn->meta.rbuf.barrier_ack; |
b411b363 PR |
4587 | |
4588 | tl_release(mdev, p->barrier, be32_to_cpu(p->set_size)); | |
4589 | ||
c4752ef1 PR |
4590 | if (mdev->state.conn == C_AHEAD && |
4591 | atomic_read(&mdev->ap_in_flight) == 0 && | |
370a43e7 PR |
4592 | !test_and_set_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags)) { |
4593 | mdev->start_resync_timer.expires = jiffies + HZ; | |
4594 | add_timer(&mdev->start_resync_timer); | |
c4752ef1 PR |
4595 | } |
4596 | ||
81e84650 | 4597 | return true; |
b411b363 PR |
4598 | } |
4599 | ||
d8763023 | 4600 | static int got_OVResult(struct drbd_conf *mdev, enum drbd_packet cmd) |
b411b363 | 4601 | { |
257d0af6 | 4602 | struct p_block_ack *p = &mdev->tconn->meta.rbuf.block_ack; |
b411b363 PR |
4603 | struct drbd_work *w; |
4604 | sector_t sector; | |
4605 | int size; | |
4606 | ||
4607 | sector = be64_to_cpu(p->sector); | |
4608 | size = be32_to_cpu(p->blksize); | |
4609 | ||
4610 | update_peer_seq(mdev, be32_to_cpu(p->seq_num)); | |
4611 | ||
4612 | if (be64_to_cpu(p->block_id) == ID_OUT_OF_SYNC) | |
4613 | drbd_ov_oos_found(mdev, sector, size); | |
4614 | else | |
4615 | ov_oos_print(mdev); | |
4616 | ||
1d53f09e | 4617 | if (!get_ldev(mdev)) |
81e84650 | 4618 | return true; |
1d53f09e | 4619 | |
b411b363 PR |
4620 | drbd_rs_complete_io(mdev, sector); |
4621 | dec_rs_pending(mdev); | |
4622 | ||
ea5442af LE |
4623 | --mdev->ov_left; |
4624 | ||
4625 | /* let's advance progress step marks only for every other megabyte */ | |
4626 | if ((mdev->ov_left & 0x200) == 0x200) | |
4627 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
4628 | ||
4629 | if (mdev->ov_left == 0) { | |
b411b363 PR |
4630 | w = kmalloc(sizeof(*w), GFP_NOIO); |
4631 | if (w) { | |
4632 | w->cb = w_ov_finished; | |
a21e9298 | 4633 | w->mdev = mdev; |
e42325a5 | 4634 | drbd_queue_work_front(&mdev->tconn->data.work, w); |
b411b363 PR |
4635 | } else { |
4636 | dev_err(DEV, "kmalloc(w) failed."); | |
4637 | ov_oos_print(mdev); | |
4638 | drbd_resync_finished(mdev); | |
4639 | } | |
4640 | } | |
1d53f09e | 4641 | put_ldev(mdev); |
81e84650 | 4642 | return true; |
b411b363 PR |
4643 | } |
4644 | ||
d8763023 | 4645 | static int got_skip(struct drbd_conf *mdev, enum drbd_packet cmd) |
0ced55a3 | 4646 | { |
81e84650 | 4647 | return true; |
0ced55a3 PR |
4648 | } |
4649 | ||
32862ec7 PR |
4650 | static int _drbd_process_done_ee(int vnr, void *p, void *data) |
4651 | { | |
4652 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
4653 | return !drbd_process_done_ee(mdev); | |
4654 | } | |
4655 | ||
4656 | static int _check_ee_empty(int vnr, void *p, void *data) | |
4657 | { | |
4658 | struct drbd_conf *mdev = (struct drbd_conf *)p; | |
4659 | struct drbd_tconn *tconn = mdev->tconn; | |
4660 | int not_empty; | |
4661 | ||
4662 | spin_lock_irq(&tconn->req_lock); | |
4663 | not_empty = !list_empty(&mdev->done_ee); | |
4664 | spin_unlock_irq(&tconn->req_lock); | |
4665 | ||
4666 | return not_empty; | |
4667 | } | |
4668 | ||
4669 | static int tconn_process_done_ee(struct drbd_tconn *tconn) | |
4670 | { | |
4671 | int not_empty, err; | |
4672 | ||
4673 | do { | |
4674 | clear_bit(SIGNAL_ASENDER, &tconn->flags); | |
4675 | flush_signals(current); | |
4676 | err = idr_for_each(&tconn->volumes, _drbd_process_done_ee, NULL); | |
4677 | if (err) | |
4678 | return err; | |
4679 | set_bit(SIGNAL_ASENDER, &tconn->flags); | |
4680 | not_empty = idr_for_each(&tconn->volumes, _check_ee_empty, NULL); | |
4681 | } while (not_empty); | |
4682 | ||
4683 | return 0; | |
4684 | } | |
4685 | ||
7201b972 AG |
4686 | struct asender_cmd { |
4687 | size_t pkt_size; | |
4688 | int (*process)(struct drbd_conf *, enum drbd_packet); | |
4689 | }; | |
4690 | ||
4691 | static struct asender_cmd asender_tbl[] = { | |
4692 | [P_PING] = { sizeof(struct p_header), got_Ping }, | |
4693 | [P_PING_ACK] = { sizeof(struct p_header), got_PingAck }, | |
4694 | [P_RECV_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4695 | [P_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4696 | [P_RS_WRITE_ACK] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4697 | [P_DISCARD_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4698 | [P_NEG_ACK] = { sizeof(struct p_block_ack), got_NegAck }, | |
4699 | [P_NEG_DREPLY] = { sizeof(struct p_block_ack), got_NegDReply }, | |
4700 | [P_NEG_RS_DREPLY] = { sizeof(struct p_block_ack), got_NegRSDReply}, | |
4701 | [P_OV_RESULT] = { sizeof(struct p_block_ack), got_OVResult }, | |
4702 | [P_BARRIER_ACK] = { sizeof(struct p_barrier_ack), got_BarrierAck }, | |
4703 | [P_STATE_CHG_REPLY] = { sizeof(struct p_req_state_reply), got_RqSReply }, | |
4704 | [P_RS_IS_IN_SYNC] = { sizeof(struct p_block_ack), got_IsInSync }, | |
4705 | [P_DELAY_PROBE] = { sizeof(struct p_delay_probe93), got_skip }, | |
4706 | [P_RS_CANCEL] = { sizeof(struct p_block_ack), got_NegRSDReply}, | |
4707 | [P_CONN_ST_CHG_REPLY]={ sizeof(struct p_req_state_reply), got_RqSReply }, | |
4708 | [P_RETRY_WRITE] = { sizeof(struct p_block_ack), got_BlockAck }, | |
4709 | }; | |
4710 | ||
b411b363 PR |
4711 | int drbd_asender(struct drbd_thread *thi) |
4712 | { | |
392c8801 | 4713 | struct drbd_tconn *tconn = thi->tconn; |
32862ec7 | 4714 | struct p_header *h = &tconn->meta.rbuf.header; |
b411b363 | 4715 | struct asender_cmd *cmd = NULL; |
77351055 | 4716 | struct packet_info pi; |
257d0af6 | 4717 | int rv; |
b411b363 PR |
4718 | void *buf = h; |
4719 | int received = 0; | |
257d0af6 | 4720 | int expect = sizeof(struct p_header); |
f36af18c | 4721 | int ping_timeout_active = 0; |
b411b363 | 4722 | |
b411b363 PR |
4723 | current->policy = SCHED_RR; /* Make this a realtime task! */ |
4724 | current->rt_priority = 2; /* more important than all other tasks */ | |
4725 | ||
e77a0a5c | 4726 | while (get_t_state(thi) == RUNNING) { |
80822284 | 4727 | drbd_thread_current_set_cpu(thi); |
32862ec7 | 4728 | if (test_and_clear_bit(SEND_PING, &tconn->flags)) { |
2a67d8b9 | 4729 | if (!drbd_send_ping(tconn)) { |
32862ec7 | 4730 | conn_err(tconn, "drbd_send_ping has failed\n"); |
841ce241 AG |
4731 | goto reconnect; |
4732 | } | |
32862ec7 PR |
4733 | tconn->meta.socket->sk->sk_rcvtimeo = |
4734 | tconn->net_conf->ping_timeo*HZ/10; | |
f36af18c | 4735 | ping_timeout_active = 1; |
b411b363 PR |
4736 | } |
4737 | ||
32862ec7 PR |
4738 | /* TODO: conditionally cork; it may hurt latency if we cork without |
4739 | much to send */ | |
4740 | if (!tconn->net_conf->no_cork) | |
4741 | drbd_tcp_cork(tconn->meta.socket); | |
4742 | if (tconn_process_done_ee(tconn)) | |
4743 | goto reconnect; | |
b411b363 | 4744 | /* but unconditionally uncork unless disabled */ |
32862ec7 PR |
4745 | if (!tconn->net_conf->no_cork) |
4746 | drbd_tcp_uncork(tconn->meta.socket); | |
b411b363 PR |
4747 | |
4748 | /* short circuit, recv_msg would return EINTR anyways. */ | |
4749 | if (signal_pending(current)) | |
4750 | continue; | |
4751 | ||
32862ec7 PR |
4752 | rv = drbd_recv_short(tconn->meta.socket, buf, expect-received, 0); |
4753 | clear_bit(SIGNAL_ASENDER, &tconn->flags); | |
b411b363 PR |
4754 | |
4755 | flush_signals(current); | |
4756 | ||
4757 | /* Note: | |
4758 | * -EINTR (on meta) we got a signal | |
4759 | * -EAGAIN (on meta) rcvtimeo expired | |
4760 | * -ECONNRESET other side closed the connection | |
4761 | * -ERESTARTSYS (on data) we got a signal | |
4762 | * rv < 0 other than above: unexpected error! | |
4763 | * rv == expected: full header or command | |
4764 | * rv < expected: "woken" by signal during receive | |
4765 | * rv == 0 : "connection shut down by peer" | |
4766 | */ | |
4767 | if (likely(rv > 0)) { | |
4768 | received += rv; | |
4769 | buf += rv; | |
4770 | } else if (rv == 0) { | |
32862ec7 | 4771 | conn_err(tconn, "meta connection shut down by peer.\n"); |
b411b363 PR |
4772 | goto reconnect; |
4773 | } else if (rv == -EAGAIN) { | |
cb6518cb LE |
4774 | /* If the data socket received something meanwhile, |
4775 | * that is good enough: peer is still alive. */ | |
32862ec7 PR |
4776 | if (time_after(tconn->last_received, |
4777 | jiffies - tconn->meta.socket->sk->sk_rcvtimeo)) | |
cb6518cb | 4778 | continue; |
f36af18c | 4779 | if (ping_timeout_active) { |
32862ec7 | 4780 | conn_err(tconn, "PingAck did not arrive in time.\n"); |
b411b363 PR |
4781 | goto reconnect; |
4782 | } | |
32862ec7 | 4783 | set_bit(SEND_PING, &tconn->flags); |
b411b363 PR |
4784 | continue; |
4785 | } else if (rv == -EINTR) { | |
4786 | continue; | |
4787 | } else { | |
32862ec7 | 4788 | conn_err(tconn, "sock_recvmsg returned %d\n", rv); |
b411b363 PR |
4789 | goto reconnect; |
4790 | } | |
4791 | ||
4792 | if (received == expect && cmd == NULL) { | |
32862ec7 | 4793 | if (!decode_header(tconn, h, &pi)) |
b411b363 | 4794 | goto reconnect; |
7201b972 AG |
4795 | cmd = &asender_tbl[pi.cmd]; |
4796 | if (pi.cmd >= ARRAY_SIZE(asender_tbl) || !cmd) { | |
32862ec7 | 4797 | conn_err(tconn, "unknown command %d on meta (l: %d)\n", |
77351055 | 4798 | pi.cmd, pi.size); |
b411b363 PR |
4799 | goto disconnect; |
4800 | } | |
4801 | expect = cmd->pkt_size; | |
77351055 | 4802 | if (pi.size != expect - sizeof(struct p_header)) { |
32862ec7 | 4803 | conn_err(tconn, "Wrong packet size on meta (c: %d, l: %d)\n", |
77351055 | 4804 | pi.cmd, pi.size); |
b411b363 | 4805 | goto reconnect; |
257d0af6 | 4806 | } |
b411b363 PR |
4807 | } |
4808 | if (received == expect) { | |
32862ec7 PR |
4809 | tconn->last_received = jiffies; |
4810 | if (!cmd->process(vnr_to_mdev(tconn, pi.vnr), pi.cmd)) | |
b411b363 PR |
4811 | goto reconnect; |
4812 | ||
f36af18c LE |
4813 | /* the idle_timeout (ping-int) |
4814 | * has been restored in got_PingAck() */ | |
7201b972 | 4815 | if (cmd == &asender_tbl[P_PING_ACK]) |
f36af18c LE |
4816 | ping_timeout_active = 0; |
4817 | ||
b411b363 PR |
4818 | buf = h; |
4819 | received = 0; | |
257d0af6 | 4820 | expect = sizeof(struct p_header); |
b411b363 PR |
4821 | cmd = NULL; |
4822 | } | |
4823 | } | |
4824 | ||
4825 | if (0) { | |
4826 | reconnect: | |
bbeb641c | 4827 | conn_request_state(tconn, NS(conn, C_NETWORK_FAILURE), CS_HARD); |
b411b363 PR |
4828 | } |
4829 | if (0) { | |
4830 | disconnect: | |
bbeb641c | 4831 | conn_request_state(tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
b411b363 | 4832 | } |
32862ec7 | 4833 | clear_bit(SIGNAL_ASENDER, &tconn->flags); |
b411b363 | 4834 | |
32862ec7 | 4835 | conn_info(tconn, "asender terminated\n"); |
b411b363 PR |
4836 | |
4837 | return 0; | |
4838 | } |